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
1708.02219	1	1708	2017-08-07T17:36:03	Superconducting Caps for Quantum Integrated Circuits	[ "physics.app-ph", "quant-ph" ]	We report on the fabrication and metrology of superconducting caps for qubit circuits. As part of a 3D quantum integrated circuit architecture, a cap chip forms the upper half of an enclosure that provides isolation, increases vacuum participation ratio, and improves performance of individual resonant elements. Here, we demonstrate that such caps can be reliably fabricated, placed on a circuit chip, and form superconducting connections to the circuit.	physics.app-ph	physics	Superconducting Caps for Quantum Integrated Circuits William O'Brien, Mehrnoosh Vahidpour, Jon Tyler Whyland, Joel Angeles, Jayss Marshall, Diego Scarabelli, Genya Crossman, Kamal Yadav, Yuvraj Mohan, Catvu Bui, Vijay Rawat, Russ Renzas, Nagesh Vodrahalli, Andrew Bestwick, Chad Rigetti Rigetti Computing, 775 Heinz Avenue, Berkeley, CA 94710 (Dated: August 8, 2017) We report on the fabrication and metrology of superconducting caps for qubit circuits. As part of a 3D quantum integrated circuit architecture, a cap chip forms the upper half of an enclosure that provides isolation, increases vacuum participation ratio, and improves performance of individual resonant elements. Here, we demonstrate that such caps can be reliably fabricated, placed on a circuit chip, and form superconducting connections to the circuit. MOTIVATION Superconducting qubits are highly sensitive detectors of material quality. The coupling of qubits' electromag- netic modes with lossy, defective materials can limit their coherence time. In a 2D circuit geometry the field lines are largely in the plane, as in Fig. 1A, where the mode interacts strongly with interface and substrate defects. One way to avoid this is to place qubits inside a 3D cav- ity, which increases the modes' participation in lossless vacuum, thereby reducing microwave loss and increasing qubit lifetime [1]. However, this approach sacrifices the scalability and density of 2D circuits. To confer some benefits 3D cavities onto a 2D circuit, we propose a geometry as illustrated in Fig. 1B in which a superconducting cap is placed on top of the chip. Here, field lines terminate preferentially on the walls of the en- closure, which increases the spatial overlap of the mode with the lossless medium of free space. Microwave simu- lations of superconducting resonators underneath a cap, as in Fig. 1C., indicate a strong response of the electro- magnetic vacuum participation ratio to the cap height. In addition, when employing a non-planar quantum in- tegrated circuit architecture including superconducting vias, we can engineer separate shielded enclosures to im- prove the electromagnetic isolation and reduce crosstalk between circuit elements. Simulations indicate that such enclosures reduce crosstalk between neighboring resonant elements by 15 dB. In summary, we expect that bonding caps with a super- conducting liner over qubit circuits will confer improved coherence times, reduced crosstalk, and better immunity to environmental noise. FIG. 1. Schematic illustration of cap effect on mode profile. Arrows indicate how the electric field lines, extending from source (S) to ground (G), change in a A) 2D and B) 3D circuit geometry. It can be seen that the 3D geometry pulls more of the field lines into free space, decreasing dielectric losses. C) Modeling results show the expected vacuum participation enhancement with decreasing cap height. Pair and Psubstrate are the fraction of electric fields in the vacuum and substrate, respectively. FABRICATION The cap is composed of pockets etched in Si, coated with superconducting metal, and patterned with indium bumps as a superconducting adhesive for bonding to the circuit chip. Caps are fabricated from silicon-on-insulator wafers, in which a 24 µm Si device layer is sandwiched on both sides by 1 µm of oxide, with thick Si underneath for structural support. The top side of the wafer is pat- terned with photoresist, first to mask an inductively cou- pled plasma etch that selectively removes the top oxide (Fig. 2A) to pattern bumps that limit the gap between cap and circuit, and then a second time to mask a deep FIG. 3. Images from cap fabrication. A) and B) are SEM images after sputter deposition, displaying smooth, vertical sidewalls and a continuous superconducting liner. Shown in C) is a confocal microscope image of the profile after liftoff of the indium bumps. the proximity effect. To form this electrical and mechanical connection be- tween circuit and cap, evaporated indium bumps (with Ti adhesion layers) are patterned on the cap. Electrically, indium serves as an ideal Type-I superconducting bond to ground the cap, as it has a Tc above 3 K. Several groups have successfully used superconducting indium bumps to bond chips with qubits to another chip with readout and control signals with high qubit coherence (>20 µs) [3– 5], which is a more technically demanding application. Electron-beam evaporation is chosen for indium deposi- tion, since it is well known to produce high quality indium films, and is amenable for liftoff (Fig. 2D). The patterned bumps are found to be well defined, as seen in the con- focal microscope image of Fig. 3C. The circuit chip is produced in two liftoff steps. First, we pattern bond pads for contacting the indium, deposit 85 nm Al, 5 nm Ti, and 60 nm Pd, then perform liftoff. The palladium is a noble metal that does not oxidize, and is commonly used for bonding purposes. Although it does not superconduct, it will experience the supercon- ducting proximity effect by being sandwiched between Al and In. It is in this way possible to form a continuous superconducting path from the circuit to cap chip despite the use of normal metals, albeit it with a weak link, so long as those normal layers are thin enough. The sec- ond step of the circuit chip fabrication is to pattern, de- posit, and liftoff superconducting circuit features (ground plane, resonators, etc.) with 160 nm of Al, allowing for some overlap on the bonding pads to ensure a good electrical connection (Fig. 2E). This is performed in a Plassys e-beam evaporation system, designed for Joseph- son Junction (JJ) fabrication. To form the 3D structure, we use a flip-chip bonder to affix the cap chip to the circuit chip (Fig. 2F). The two chips are precisely aligned and pressed together under up 2 FIG. 2. Process flow diagram for cap fabrication. A) Oxide bumps patterned and etched, B) DRIE pocket etched using Bosch process through photoresist mask down to oxide stop, C) sputter deposition of Al/Mo, D) indium bumps with tita- nium adhesion layer deposited by e-beam evaporation, after liftoff process, E) circuit chip fabrication, involving Al/Ti/Pd pads for bonding to indium, F) final bonded structure. reactive ion etch (DRIE) down to the buried oxide. The DRIE follows the Bosch process [2], producing a square pocket with vertical sidewalls 24 µm deep (Fig. 2B). The DRIE step employs a LF substrate bias to avoid footing near the buried oxide. The surface is then conformally coated with 1 µm sputtered Al to form a continuous su- perconducting shield. The resulting profile is visualized in Fig. 3A and Fig. 3B, confirming a continuous sput- tered film and smooth, vertical sidewalls from the Bosch process. In order to form high quality interfaces, the terminat- ing metal layers on the cap and circuit (which will medi- ate the indium connection) are selected to avoid or mit- igate native oxides. On the cap side, termination with an Al native oxide is avoided by capping the Al in-situ (without breaking vacuum) with a thin, 200 nm Mo film (Fig. 2C). Mo is a refractory metal that is superconduct- ing near 1 K. Similarly, Pd is chosen to terminate bond pads on the circuit side. Although Pd is not supercon- ducting at our operating temperature, it is designed to be thin enough to still behave as a superconductor through c(cid:13) Copyright 2017 Rigetti & Co, Inc. FIG. 4. Schematic of test structures for indium bonding, in profile. A) Regular cells compose nearly 90% of the sites, and include structures to measure the roundtrip resistance from circuit to cap and back, including the electrical contact in the full bond stack Pd-In-Ti-Mo-Al. B) Indium shorts out adja- cent pads, which isolates the impact of the In-Pd interface. C) Trenches are etched around adjacent indium bumps, which could potentially sever continuity if the Al film is not entirely conformal. to 45 kg of weight at around 70◦ C. Note that the 1 µm oxide bumps are patterned to place a limit on how close the cap and circuit chip come in contact during bonding. Though in practice, we do not need to apply the required force to press the indium flush with the oxide bumps. TEST STRUCTURES To assess the quality of the bond across the die, test structures shown in Fig. 4 were designed to measure series resistance across indium bumps. The conductive path is designed such that current must traverse from the circuit metal to the cap through an indium bump, then back to the circuit along an adjacent indium bump. Some special test structures were also implemented, to test the confor- mality of the sputtered Al, and others to test the intrin- sic contact resistance between indium and palladium. To test for Al conformality, trenches with the same depth as the pocket are etched into the cap between adjacent indium bumps. If the Al layer is not continuous, current injected through one bump is not able to find a return path to the circuit. To test the contact resistance of just the indium-palladium interface, two adjacent pads are shorted with indium, such that current does not need to flow through the cap for continuity. The test structures were fabricated on a 31 mm cir- cuit chip and a corresponding 29 mm cap chip, arranged in a 9 × 9 grid with signal lines designed so that all 81 structures could be measured from the circuit chip edges (Fig. 5A). This arrangement allows us to probe sites dis- tributed across a large chip and test planarity of the bond. Most of the test structures were standard cells, with 6 In-connected cells and 4 with trenches (Fig. 5B). c(cid:13) Copyright 2017 Rigetti & Co, Inc. FIG. 5. A) Arrangement of 81 test structures on a 31 mm cir- cuit chip and 29 mm cap chip. Bond pads are at the perimeter of the circuit chip to allow for probing after bonding. B) Dis- tribution of test structure types across chip. FIG. 6. Cryogenic DC resistance with labeled superconduct- ing transitions associated with Mo/Al and In. The instru- mentation used for this measurement sets a lower bound for the critical current of 36 µA. RESULTS At room temperature, series resistances below 5 Ω were obtained at 80 out of 81 test sites, demonstrating excel- lent continuity and high process fidelity. From the lowest resistance values measured, the room temperature con- tact resistance between indium and molybdenum is less than 0.2 Ω. No difference was measured between the dif- ferent test structure types, indicating that the Al layer is indeed continuous across the pockets. Furthermore no major asymmetries in resistance were observed between the left and right sides or top and bottom of the chip, indicating that bond parallelism across a large 31 mm chip is acceptable. We also performed cryogenic measurements to assess the superconducting properties. An example is shown in Fig. 6. As we cool the sample down, we see sharp resistance drops at 2.9-3.1 K and 0.8-0.9 K, which cor- respond approximately to the superconducting transition 3 temperatures of In and Mo/Al, respectively. The absence of separate Al and Mo transitions could be explained by a variety of superconducting-normal interface effects. In any case, under 0.8 K we observe zero resistance, below the sensitivity of our instruments, indicative of a superconducting path from the circuit chip to the cap chip and back. Our instruments set the lower bound of the critical current of this path as 36 µA. CONCLUSIONS The application of a Mo capping layer to the Al cap has proven to be an effective means of mitigating the deleteri- ous effects of native oxide on the electrical and mechan- ical contact of indium bumps between circuit and cap die. In addition, the consistently low room-temperature resistances and clear low-temperature superconductivity demonstrate that the cap can be kept grounded by robust supercurrents through the bonds, at locations across the die. With successful basic metrology of this technology es- tablished, we can move onto integration with supercon- ducting quantum circuits, including qubits. Further work is needed to systematically establish enhancements to qubit performance. [1] Hanhee Paik, et al., Observation of High Coherence in Josephson Junction Qubits Measured in a Three- Dimensional Circuit QED Architecture, Phys. Rev. Lett. 107, 240501 (2011). [2] F. Marty, L. Rousseau, B. Saadany, B. Mercier, O. Fran- cais, Y. Mita, T. Bourouina, Advanced etching of silicon based on deep reactive ion etching for silicon high aspect ratio microstructures and three-dimensional micro- and nanostructures, Microelectronics Journal 36, 7 (2005). [3] D. Rosenberg, et al., 3D integrated superconducting qubits, arXiv:1706.04116 (2017). [4] M. Mohseni, et. al., Commercialize quantum technologies in five years, Nature 543, 7644 (2017). [5] J. Mutus, et. al., H46.00006: 3D integration of supercon- ducting qubits with bump bonds: Part 1, J. Kelly, et. al., H46.00007: 3D integration of superconducting qubits with bump bonds: Part 2, E. Lucero, et. al., H46.00008: 3D integration of superconducting qubits with bump bonds: Part 3, APS March Meeting 2017 c(cid:13) Copyright 2017 Rigetti & Co, Inc. 4
1811.12487	1	1811	2018-11-29T21:04:32	Smart Table Based on Metasurface for Wireless Power Transfer	[ "physics.app-ph" ]	Metasurfaces have been investigated and its numerous exotic functionalities and the potentials to arbitrarily control of the electromagnetic fields have been extensively explored. However, only limited types of metasurface have finally entered into real products. Here, we introduce a concept of a metasurface-based smart table for wirelessly charging portable devices and report its first prototype. The proposed metasurface can efficiently transform evanescent fields into propagating waves which significantly improves the near field coupling to charge a receiving device arbitrarily placed on its surface wirelessly through magnetic resonance coupling. In this way, power transfer efficiency of 80$\%$ is experimentally obtained when the receiver is placed at any distances from the transmitter. The proposed concept enables a variety of important applications in the fields of consumer electronics, electric automobiles, implanted medical devices, etc. The further developed metasurface-based smart table may serve as an ultimate 2-dimensional platform and support charging multiple receivers.	physics.app-ph	physics	Smart Table Based on Metasurface for Wireless Power Transfer Mingzhao Song,1 Kseniia Baryshnikova,1 Aleksandr Markvart,1 Pavel Belov,1 Elizaveta Nenasheva,2 Constantin Simovski,1, 3 and Polina Kapitanova1 1)Department of Nanophotonics and Metamaterials, ITMO University, 197101 Saint Petersburg, Russia 2)Giricond Research Institute, Ceramics Co., Ltd., Saint Petersburg 194223, Russia 3)School of Electrical Engineering, Department of Electronics and Nanoengineering, Aalto University, P.O. Box 15500, 00076 Aalto, Finland (Dated: December 3, 2018) Metasurfaces have been investigated and its numerous exotic functionalities and the po- tentials to arbitrarily control of the electromagnetic fields have been extensively explored. However, only limited types of metasurface have finally entered into real products. Here, we introduce a concept of a metasurface-based smart table for wirelessly charging portable devices and report its first prototype. The proposed metasurface can efficiently transform evanescent fields into propagating waves which significantly improves the near field coupling to charge a receiving device arbitrarily placed on its surface wirelessly through magnetic reso- nance coupling. In this way, power transfer efficiency of 80% is experimentally obtained when the receiver is placed at any distances from the transmitter. The proposed concept enables a variety of important applications in the fields of consumer electronics, electric automobiles, implanted medical devices, etc. The further developed metasurface-based smart table may serve as an ultimate 2-dimensional platform and support charging multiple receivers. Keywords: wireless power transfer, magnetic resonance, magnetic near field, dielectric, meta- surface I. INTRODUCTION Arbitrary tailor, mold and manipulation of electromag- netic fields is the ultimate goal in electromagnetic re- search from radio-frequencies to optics. Recently, meta- surfaces have attracted great interests due to its poten- tials to provide a profound control over electromagnetic fields.1 -- 3 A variety of functionalities are demonstrated for a broad frequency band ranging from microwave to visible light. Initially, the high impedance surfaces were proposed in radio frequency and microwave regimes to reduce the antenna profile and improve the radiation patterns4. Recently these 2D artificial structures were dubbed as metasurfaces1 and their applications have soon spread to more specific areas such as enhancement of magnetic resonance imaging5. In optical regime, meta- surfaces have paved the way for flat optics and photon- ics6,7. They can be designed to possess the required properties to replace bulky optical components. Differ- ent realizations are demonstrated for specific purposes, for instance, beam focusing lens 8, tunable lens9, perfect absorbers10, wavefront shapers11, polarizers12 etc. How- ever, the applications of metasurfaces for wireless power transfer (WPT) are yet to be explored. With an increasing demands for conveniently charging electronic devices, WPT is considered as a promising so- lution. Much effort has been made to develop different types of WPT systems,13 -- 18 among which the magnetic resonant coupling has become a hot research topic since it was proposed for the first time in 2007 especially due to its potentials for safe and mid-range charging.19 -- 21 In such systems two or more resonators with the same resonance frequency are coupled by magnetic evanescent fields. The typical charging distance is 2-10 times of the resonator characteristic dimension. The charging dis- tance is limited primarily due to the physical principle of evanescent field decay rather than engineering restric- tions. Thus there is little room to further extend the operation distance of a magnetic resonant WPT system in 3D space. One of the ways is to increase the cou- pling coefficient between two distant resonators with the use of metamaterials. Different metamaterials designs for WPT have been recently introduced and intensively stud- ied.22 -- 26 Usually, metamaterials operate as a super-lens focusing near magnetic fields of a transmitter and redi- recting it to a receiver.25 It helps to improve the WPT efficiency up to 30% at the fixed distance of the trans- mitter and receiver placed at the optimized position (or focal points) from two opposite sides of the metamate- rial.25 However, the whole WPT system with bulk meta- material is very cumbersome which reduces its chances for practical applications. Unlike the bulk metamaterials, thin planar metasur- faces possess low intrinsic losses and at the same time provide a desired manipulation of the electromagnetic fields. The goal of this paper is to introduce a novel concept of a metasurface-based smart table for WPT ap- plications. In our vision the smart table with the meta- surface embedded or placed underneath the desktop al- lows powering multiple devices simultaneously, regard- less of how they are located on the table and oriented relative to each other (see Fig. 1(a)). In terms of the operational principle we consider different scenarios of the metasurface designs. First, the metasurface plays a role of an intermediary between the transmitter and the receiver enhancing the WPT performances, as shown in Fig. 1(b). The metasurface comprises multiple power transfer channels. When a receiving resonator is placed on the metasurface and detected, the channel underneath is switched on by activating the corresponding transmit- 2 Figure 1. (a) Artist's view of multiple electronic devices being powered by a wireless charging table. (b) Metasurface as an intermediary for enhancing WPT performance. (c) Metasurface as a transmitting resonator of WPT system. ter driven by the power divider. At the same time, the inactivated channel keeps off. Thus the major part of the power will flow only through the activated channel in the form of current I0, whereas the leakage current in other channels is negligible (I0 >> I1). The advantage of this type of design is that the power can be fully used with minimum losses when the power transfer channel is well designed. But it requires load detection elements which increases the engineering complexity. In the other design, the metasurface itself behaves as a transmitting resonator providing a desired magnetic field profile, for instance, a uniform field distribution or hot spots, as shown in Fig. 1(c). In this case, the metasurface operates on a certain mode which requires a simple excitation. How- ever, part of the energy will be dissipated in the form of radiation. Here we report the first scenario and demon- strate the metasurface design which serves as a substrate to transmitting and receiving resonators allowing an ef- ficient WPT up to the distances exceeding the resonator size by an order of magnitude. To demonstrate the func- tion of the metasurface and its possibilities for long-range power transfer, the WPT efficiency from one resonator to another was investigated both numerically and exper- imentally. The calculations of specific absorption rate (SAR) which is a critical factor in terms of the safety issues was performed. Our design being affordable and compact paves the way to the realization of ubiquitous wireless charging. II. DESIGN OF WPT SYSTEM Recently, wire media, arrays of parallel metal wires fre- are studied in microwave, terahertz and optical quency regimes revealing unique electromagnetic prop- erties.27 One of these properties is the efficient conver- sion of evanescent waves into waves propagating in the wire medium.28 -- 30 This property is a key prerequisite for the subwavelength imaging in wire-medium endo- scopes27,31 -- 33. And it is also the prerequisite of an ef- ficient WPT between two resonators. In our case, the resonant coupling is the coupling of near fields which are packages of evanescent waves. For a disk resonator with colossal permittivity operating at a magnetic dipole mode the electric field is mostly concen- trated inside while the magnetic field is maximal near the surface of the disk.34 The evanescent tails of the magnetic field offer the WPT with 50% efficiency at the distances twofold of the resonator diameter.34 The further distance increase results in an efficiency decay. However, if a wire medium is located at a distance smaller that the res- onator size the resonator may be strongly coupled to the wire medium. Its near magnetic field is converted into propagating modes of the wire medium and the power 3 Figure 2. Geometry of proposed WPT system. can be transported from one resonator to another along the wires. Such mechanism still belongs to WPT be- cause two resonators are not electrically connected. In the present work, we suggest a more practical way than the use of bulk wire medium. A metasurface of parallel wires, a planar analogue of the wire medium, is also ca- pable to convert the evanescent waves into propagating modes and vice versa. It can nicely serve as an interme- diary for distant coupling between two disk resonators. The WPT system under study consisted of a trans- mitter and a receiver placed above the metasurface as shown in Fig.2. The transmitter comprised a dielectric disk resonator and a transmitting loop on top of it at the distance s1. The dielectric disk had a diameter of D and a thickness of h. The loop had a diameter of D0 and was made of a conducting wire with diameter of w. The receiver consisted of an identical dielectric resonator and a receiving loop. The receiver was located at a varying distance d from the transmitter. Both transmitter and receiver were placed at height s2 above the metasurface. The last one was a regular array of parallel copper wires with period a satisfying the criterion w (cid:28) a (cid:28) λ, where λ was the wavelength in free space. Thus this parameter was chosen as a=10 mm. There was no limitation for the total number of wires in the metasurface, but it de- pended on the practical needs. Here we used 21 wires to cover 20 cm interval which was the width of our metasur- face. The length of the metasurface L was also selected from practical requirements. Two cases of L=65 cm and L=120 cm were studied. III. RESULTS AND DISCUSSION A. Operational modes of WPT system Eigenmode analysis of a single dielectric resonator was conducted, which revealed that all the magnetic modes were identified at the frequencies above 230 MHz, i.e.magnetic dipole (MD), quadrupole (MQ) and oc- tupole (MO) mode hold at 232 MHz, 296 MHz and 347 MHz, respectively, whereas all the electric modes were observed at the frequencies higher than 720 MHz, i.e. electric dipole (ED) and quadrupole (EQ) mode is at 725 MHz and 730 MHz, respectively. Such a wide range of magnetic and electric eigenmodes separation is resulted from the geometric parameters of the dielectric disk res- onator. Therefore we mainly target to use the MD mode of the disk which can be excited by a current loop34 be- ing aware of the electric type of modes located far away from magnetic types. In the proposed WPT system the transmitting res- onator was coupled to the metasurface and, via the meta- surface, to another disk. Here the frequencies of all modes were shifted with respect to those of an individual disk and could hybridize if the coupling with the meta- surface was too strong. For two cases of L = 65 cm and L = 120 cm, we performed numerical simulations of the whole WPT system using the frequency solver in CST Microwave Studio Suite varying d and keeping the same value s2 which determined the coupling with the meta- surface, see Fig.3 for d = 25 cm. For the WPT system with L=120 cm in Fig.3 one can see two maxima in the transmission coefficient spectrum -- at frequency nearly equal to 200 MHz and at nearly 270 MHz. For the design with L = 65 cm two peaks are also observed. The WPT efficiency was estimated from the S-parameters using the equation η = S212. It was found that for L = 120 cm the efficiencies of η=49% at 200 MHz and η=81% at 270 MHz are reached. In the case of metasurface length of L = 65 cm the efficiencies of η=49% at 170 MHz and η=70% at 270 MHz are obtained. Thus, for given meta- surfaces we found two frequencies for an efficient power transfer for the given distance between resonators. How- ever, the first transmission resonance shifted to 170 MHz, more than by 15% compared to the case of L=120 cm, whereas the second peak of transmission remained at the same frequency. Thus, this resonance was not a dimen- sional one of the metasurface and was not affected by its finite size. The main question then arose: which of these two transmission peaks corresponded to the MD mode? To clarify the nature of both resonances we analyzed the electromagnetic field distributions for the case L = 120 cm. The electric and magnetic field distributions across the metasurface and across the disk at two fre- quencies are shown in Fig.4. As one can see in Fig.4(e) that at 200 MHz the electric field concentrates at the edges and in the middle of the metasurface forming a typ- ical standing wave pattern. The symmetry of the stand- ing wave was broken due to the asymmetric placement of the disks. The magnetic field distribution confirmed that the pattern of the electromagnetic field corresponded to the standing wave at 200 MHz where a resonance of the disk and that of a metasurface overlapped. It was clear that the coupling at this frequency must be sensitive to the distance d. At 270 MHz the electric field kept uni- form between the resonators. The minima of E at the disks axes was a hint that 270 MHz was the disk reso- nance of the magnetic type. Also, the maximal amplitude of the electric field at 270 MHz was lower than that at 200 MHz, which was good from safety reasons and was another advantage of this operation frequency. At 270 MHz the features of the standing wave regime were also present in the electromagnetic field distribution but are very weak, and the role of the disk resonance was more important. And in the color map of the magnetic field at 270 MHz the axial field concentration was observed Figure 3. Comparison between the simulated S-parameter spectra of the structures with the length of metasurface (a) L = 120 cm and (b) L = 65 cm. 4 Figure 4. Simulated electric field at 270 MHz of (a) absolute value on the cross section cutting through two disk resonators and (b) vector form on the cross section half way between two resonators. Simulated magnetic field at 270 MHz of (c) absolute value and (d) vector form. (d) Comparison between electric and magnetic fields at 200 MHz and 270 MHz. which was a typical feature of MD resonance. In the steady regime the reactive part of the mode energy was stored in the disk and did not play any role, whereas the active part -- with a nonzero Poynting vector -- coupled to the metasurface. Fig. 4(b) and (d) demonstrate that this active power excited a TEM mode in the metasurface and transmitted along it as in a multi-wire transmission line. At a distance d this guided mode excited the same MD mode in the receiving disk that was similarly cou- pled to the metasurface. Finally, the active energy of the transmitting disk was received by the wire loop applied to the receiving disk resonator. To finally verify the mode type we performed the multipole decomposition of the induced displacement currents inside the transmitting resonator of the WPT 00.20.40.60.81160180200220240260280300Frequency, MHz00.20.40.60.81S11S21S11S21L = 120 cmL = 65 cmS-parametersS-parametersab0510H, A/m270 MHz0400800E, V/mabcde020406080100120x, cm270 MHz200 MHz270 MHz200 MHz0400800E, V/m0510H, A/m270 MHzsystem. Four terms were taken into consideration, namely, electric dipole (ED), magnetic dipole (MD), elec- tric quadrupole (EQ) and magnetic quadrupole (MQ) modes.35,36 The calculated multipoles are shown in Fig.5. A strong ED mode at 200 MHz frequency and the MD mode at 270 MHz were observed, whereas both electric and magnetic quadrupole modes do not resonate below 300 MHz and are negligibly small. The electric dipole in the disk was induced by the external electric field of the finite metasurface experiencing the local enhancement at the Fabry-Perot resonance. It was so because the po- larized disk induced a dipole moment at the edge of the metasurface where charges were strongly accumulated on the ends of the wires. However, for our purposes this ef- fect was not important. We confirmed that the frequency of MD mode was 270 MHz which we would explore fur- ther. B. WPT efficiency Now let's characterize the WPT system in terms of the efficiency depending on the distance d. We performed nu- merical simulations to investigate the S-parameter spec- tra as a function of d ranging from 0 cm to 100 cm. The calculated WPT efficiency η as a function of distance d between the transmitter and the receiver is plotted in Fig. 6(c). It has a periodical tendency which can be explained by the presence of the standing wave features in the electric and magnetic field profiles. The maximal values of WPT efficiency as high as 80% were obtained at the distances of d=25 cm and d=80 cm where the magnetic field is maximal and the electric field is mini- mal at the center of the receiving disk. Domains where the efficiency was below 40% cover less than one half of the interval of possible distances. Moreover, in these two intervals where η was low the magnetic field kept suffi- ciently high for an efficient WPT. At d = 0 cm and d = 60 cm the ports in the transmitting and receiving loops were 5 mismatched due to the over coupling through the meta- surface, thereby leading to deep minima in the efficiency. The well-known method to partially cure these minima of efficiency is to introduce matching networks.37 -- 39 Then we proceeded to the experimental studies. The first prototype of the smart table was fabricated (see Fig. 6(a)) consisting of two identical disk resonators with D = 84 mm and h = 6.6 mm made from microwave ce- ramics based on Ba, Sr, T iO3 solution doped with Mg.40 To excite the resonator a non-resonant Faraday shielded loop with a diameter of D0 = 72 mm was fabricated using a segment of a coaxial cable. The end of the Faraday loop was connected to the Agilent PNA E8362C Vector Net- work Analyzer (VNA). The Faraday loop was co-axially placed near the resonator at the distance of s1 = 1 mm same as in the simulations. Both transmitter and re- ceiver were placed at the distance s2 = 1 mm above the metasurface which was fabricated as an array of copper wires with parameters as performed in the simulations. The measured S-parameters as a function of frequency for different d ranging from 0 to 100 cm are shown in Fig.6(b). For any d the maximum of S21 occurred around 270 MHz, indicating that the MD mode we theo- retically studied above was indeed excited in the system. By contrast, the first resonance frequency kept changing as d increases which was a clear evidence of its nature as a dimensional resonance of the metasurface. Next, the WPT efficiency on the MD mode was obtained from the measured S-parameters and depicted as red dots in Fig.6(c). The maximal efficiency of 83% was obtained at d = 25 cm. In this work, we employed a simple but effective method to match the whole system to reveal the maximal possible transfer efficiency. To match both ports we did not insert any lumped matching circuit. Instead, at each d we mechanically tuned the distance s1 and s2 so that S21 is maximal and both S11 and S22 are close to zero. This method is proved effective in Ref. 34. The measurements of the S-parameters with this matching procedure was done for d ranging from 0 to 100 cm with 5 cm step and the retrieved efficiency is depicted as green stars in Fig.6(d). The WPT efficiency was almost stable at 80±3% for all investigated distances d, which provided the possibility to place the receiver on the metasurface at any distances from the transmit- ter and it would be efficiently coupled to it. Moreover, the metasurface could have different dimensions, e.g. the metasurface length L could be further increased to cover more area or decreased for customized needs. The only restriction of such WPT system was the misalignment be- tween the transmitter and the receiver -- the power chan- nel was stretched along the metasurface. But it could be solved by employing multiple transmitting resonators under the control of the power divider, as demonstrated in Fig. 1(b). C. Safety Issue Figure 5. Multipole decomposition of the transmitting disk resonator. One of the target applications of such the WPT system is an ubiquitous wireless charging. Thus safety issues related to human exposure under electromagnetic fields 160180200220240260280300Frequency, MHz00.10.20.30.40.50.60.70.80.91Normalized Multipoles, a.u.EDMDEQMQ6 Figure 6. (a) Photo of the experiment setup. (b) Measured transmission coefficient spectra for d ranging from 0 to 100 cm. (c) Comparison between measured and simulated efficiency (d)Measured WPT efficiency when both ports are matched. (EMF) and specific absorption rate (SAR) must be taken into consideration. Here we considered SAR - a more rigorous safety metric for a measure of how much power is absorbed by biologic tissues. If the proposed metasurface is used in a wireless charging system it can be embedded in a plastic or wooden desk. Thus we considered a typical scenario that a human arm was placed on top of the metasurface between the transmitter and the receiver. We investigated how much power can be absorbed by the human body by means of calculating the SAR. mode) and found that it was 0.327 which is twice higher than the MD mode. There are no nonlinear effects in the WPT system, and maximum of SAR for different in- put power values could be obtained by scaling up these results. Thus, according to International Electrotechni- cal Commission standard41, where the SAR limit is 2.0 W/kg averaged over 10 g of tissue absorbing the most signal, the maximal permitted input power of 6 W was allowed at the frequency of the MD mode and only 3 W for the ED mode. To perform the SAR analysis we used CST Microwave Studio and a CAD model of the front part of a human arm (see Fig.7(a)). The human arm model comprised the details of the main biological tissues of the arm (skin, fat, muscle, bone, blood, etc) which are characterized by corresponding electromagnetic properties. The dis- tance between the transmitter and the receiver was fixed to provide the maximum of the WPT efficiency without matching. The arm placed in middle was hanging at the height 2 mm from its bottom edge to the metasur- face. Under a 0.5 W input power, the maximal SAR was 0.165 at the frequency of 270 MHz which corresponds to the MD mode (see Fig.7(b)). For comparison we also simulated the SAR at the frequency of 200 MHz (ED IV. CONCLUSION In this paper, we discussed two typical scenarios where metasurfaces could be applied for wireless power trans- fer system. Especially, we proposed a first design of an cost-effective metasurface-based smart table as an inter- mediate between WPT transmitter and receiver. Both numerical and experimental studies verified the efficient and distance-independent WPT (efficiency higher than 80%) for the case when the transmitting and receiv- ing devices were both placed on the table. Though the metasurface operated like an effective transmission line 7 Figure 7. (a) Simulation model of the WPT system with an arm located between the resonators 2 mm above the metasurface. Simulated SAR values at the frequency of (b) ED mode and (c) MD mode. our system was not similar to the well-known WPT sys- tems with a capacitive coupling. The key difference was a uniquely efficient magnetic coupling of our transmit- ting/receiving device with proposed metasurface. This coupling was granted by the magnetic Mie resonance of a dielectric disk with colossal permittivity. This physical mechanism allowed us to successfully perform the reso- nant WPT along the metasurface. The only restriction of such WPT system was the misalignment between the transmitter and the receiver -- the power channel was stretched along the metasurface, which was mainly due to the extreme anisotropy of the metasurface and can be improved by an isotropic metasurface design. V. ACKNOWLEDGMENT The authors are thankful for Georgiy Solomakha and Stanislav Glybovski for useful discussions. The calcula- tions of multipole decomposition have been supported by the Russian Science Foundation (Project No. 17- 72-10230). The numerical simulation and experimen- tal investigation of the wireless power transfer system have been supported by the Russian Science Foundation (Project No. 17-79-20379). M.S. acknowledges the sup- port from the China Scholarship Council (201508090124). REFERENCES 1S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S. Kivshar, and C. R. Simovski, "Metasurfaces: From microwaves to visible," Physics Reports, vol. 634, pp. 1 -- 72, 2016. Metasurfaces: From microwaves to visible. 2H.-T. Chen, A. J. Taylor, and N. Yu, "A review of metasur- faces: physics and applications," Reports on Progress in Physics, vol. 79, no. 7, p. 076401, 2016. 3F. Ding, A. Pors, and S. I. Bozhevolnyi, "Gradient metasurfaces: a review of fundamentals and applications," Reports on Progress in Physics, vol. 81, no. 2, p. 026401, 2018. 4D. Sievenpiper, L. Zhang, R. F. J. Broas, N. G. Alexopolous, and E. Yablonovitch, "High-impedance electromagnetic surfaces with a forbidden frequency band," IEEE Transactions on Microwave Theory and Techniques, vol. 47, pp. 2059 -- 2074, Nov 1999. 5A. P. Slobozhanyuk, A. N. Poddubny, A. J. Raaijmakers, C. A. van Den Berg, A. V. Kozachenko, I. V. I. A. Dubrovina, Melchakova, Y. S. Kivshar, and P. A. Belov, "Enhancement of magnetic resonance imaging with metasurfaces," Advanced ma- terials, vol. 28, no. 9, pp. 1832 -- 1838, 2016. 6N. Yu and F. Capasso, "Flat optics with designer metasurfaces," Nature materials, vol. 13, no. 2, p. 139, 2014. 7A. V. Kildishev, A. Boltasseva, and V. M. Shalaev, "Pla- nar photonics with metasurfaces," Science, vol. 339, no. 6125, p. 1232009, 2013. 8M. Khorasaninejad, W. T. Chen, R. C. Devlin, J. Oh, A. Y. Zhu, and F. Capasso, "Metalenses at visible wavelengths: Diffraction- limited focusing and subwavelength resolution imaging," Science, vol. 352, no. 6290, pp. 1190 -- 1194, 2016. 9E. Arbabi, A. Arbabi, S. M. Kamali, Y. Horie, M. Faraji-Dana, and A. Faraon, "Mems-tunable dielectric metasurface lens," Na- ture communications, vol. 9, no. 1, p. 812, 2018. 10Y. Ra'Di, C. Simovski, and S. Tretyakov, "Thin perfect ab- sorbers for electromagnetic waves: theory, design, and realiza- tions," Physical Review Applied, vol. 3, no. 3, p. 037001, 2015. 11N. Yu, P. Genevet, M. A. Kats, F. Aieta, J.-P. Tetienne, F. Ca- passo, and Z. Gaburro, "Light propagation with phase discon- tinuities: generalized laws of reflection and refraction," science, p. 1210713, 2011. 12Y. Zhao, M. Belkin, and A. Al`u, "Twisted optical metamaterials for planarized ultrathin broadband circular polarizers," Nature communications, vol. 3, p. 870, 2012. 13S. Y. R. Hui, W. Zhong, and C. K. Lee, "A critical review of re- cent progress in mid-range wireless power transfer," IEEE Trans- actions on Power Electronics, vol. 29, pp. 4500 -- 4511, Sept 2014. 14F. Lu, H. Zhang, and C. Mi, "A review on the recent develop- ment of capacitive wireless power transfer technology," Energies, vol. 10, no. 11, p. 1752, 2017. 15M. Song, P. Belov, and P. Kapitanova, "Wireless power trans- fer inspired by the modern trends in electromagnetics," Applied Physics Reviews, vol. 4, no. 2, p. 021102, 2017. 16S. Assawaworrarit, X. Yu, and S. Fan, "Robust wireless power transfer using a nonlinear parity -- time-symmetric circuit," Na- ture, vol. 546, no. 7658, p. 387, 2017. 17A. Krasnok, D. G. Baranov, A. Generalov, S. Li, and A. Al`u, "Coherently enhanced wireless power transfer," Physical review letters, vol. 120, no. 14, p. 143901, 2018. 18M. Song, I. Iorsh, P. Kapitanova, E. Nenasheva, and P. Belov, "Wireless power transfer based on magnetic quadrupole coupling in dielectric resonators," Applied Physics Letters, vol. 108, no. 2, p. 023902, 2016. 19A. Costanzo, M. Dionigi, D. Masotti, M. Mongiardo, G. Monti, L. Tarricone, and R. Sorrentino, "Electromagnetic energy har- vesting and wireless power transmission: A unified approach," Proceedings of the IEEE, vol. 102, no. 11, pp. 1692 -- 1711, 2014. 20A. Kurs, A. Karalis, R. Moffatt, J. D. Joannopoulos, P. Fisher, and M. Soljaci´c, "Wireless power transfer via strongly coupled magnetic resonances," science, vol. 317, no. 5834, pp. 83 -- 86, 2007. 21A. Karalis, J. D. Joannopoulos, and M. Soljaci´c, "Efficient wire- less non-radiative mid-range energy transfer," Annals of physics, vol. 323, no. 1, pp. 34 -- 48, 2008. 22Y. Urzhumov and D. R. Smith, "Metamaterial-enhanced cou- pling between magnetic dipoles for efficient wireless power trans- fer," Physical Review B, vol. 83, no. 20, p. 205114, 2011. 23W.-C. Chen, C. M. Bingham, K. M. Mak, N. W. Caira, and W. J. Padilla, "Extremely subwavelength planar magnetic metamate- rials," Physical Review B, vol. 85, no. 20, p. 201104, 2012. 24Z. Dong, F. Yang, and J. S. Ho, "Enhanced electromagnetic en- ergy harvesting with subwavelength chiral structures," Physical Review Applied, vol. 8, no. 4, p. 044026, 2017. 25B. Wang, K. H. Teo, T. Nishino, W. Yerazunis, J. Barnwell, and J. Zhang, "Experiments on wireless power transfer with meta- materials," Applied Physics Letters, vol. 98, no. 25, p. 254101, 2011. 26J. S. Ho, B. Qiu, Y. Tanabe, A. J. Yeh, S. Fan, and A. S. Poon, "Planar immersion lens with metasurfaces," Physical Review B, vol. 91, no. 12, p. 125145, 2015. 27C. R. Simovski, P. A. Belov, A. V. Atrashchenko, and Y. S. Kivshar, "Wire metamaterials: physics and applications," Ad- vanced Materials, vol. 24, no. 31, pp. 4229 -- 4248, 2012. 28A. Rahman, P. A. Belov, M. G. Silveirinha, C. R. Simovski, Y. Hao, and C. Parini, "The importance of fabry -- perot resonance and the role of shielding in subwavelength imaging performance of multiwire endoscopes," Applied Physics Letters, vol. 94, no. 3, p. 031104, 2009. 29A. Rahman, P. A. Belov, Y. Hao, and C. Parini, "Periscope-like endoscope for transmission of a near field in the infrared range," Optics letters, vol. 35, no. 2, pp. 142 -- 144, 2010. 30P. A. Belov, G. K. Palikaras, Y. Zhao, A. Rahman, C. R. Simovski, Y. Hao, and C. Parini, "Experimental demonstra- tion of multiwire endoscopes capable of manipulating near-fields with subwavelength resolution," Applied Physics Letters, vol. 97, no. 19, p. 191905, 2010. 31X. Radu, D. Garray, and C. Craeye, "Toward a wire medium endoscope for mri imaging," Metamaterials, vol. 3, no. 2, pp. 90 -- 99, 2009. 32A. Tuniz, K. J. Kaltenecker, B. M. Fischer, M. Walther, S. C. Fleming, A. Argyros, and B. T. Kuhlmey, "Metamaterial fibres 8 for subdiffraction imaging and focusing at terahertz frequencies over optically long distances," Nature communications, vol. 4, p. 2706, 2013. 33A. Slobozhanyuk, I. Melchakova, A. Kozachenko, D. Filonov, C. Simovski, and P. Belov, "An endoscope based on extremely anisotropic metamaterials for applications in magnetic resonance imaging," Journal of Communications Technology and Electron- ics, vol. 59, no. 6, pp. 562 -- 570, 2014. 34M. Song, P. Belov, and P. Kapitanova, "Wireless power transfer based on dielectric resonators with colossal permittivity," Applied Physics Letters, vol. 109, no. 22, p. 223902, 2016. 35A. B. Evlyukhin, T. Fischer, C. Reinhardt, and B. N. Chichkov, "Optical theorem and multipole scattering of light by arbitrar- ily shaped nanoparticles," Physical Review B, vol. 94, no. 20, p. 205434, 2016. 36P. D. Terekhov, K. V. Baryshnikova, Y. A. Artemyev, A. Karabchevsky, A. S. Shalin, and A. B. Evlyukhin, "Multi- polar response of nonspherical silicon nanoparticles in the visible and near-infrared spectral ranges," Physical Review B, vol. 96, no. 3, p. 035443, 2017. 37J. O. Sophocles, Electromagnetic waves and antennas. 2003. 38C. J. Stevens, "Magnetoinductive waves and wireless power transfer," IEEE Transactions on Power Electronics, vol. 30, no. 11, pp. 6182 -- 6190, 2015. 39N. Inagaki, "Theory of image impedance matching for induc- tively coupled power transfer systems," IEEE Transactions on Microwave Theory and Techniques, vol. 62, no. 4, pp. 901 -- 908, 2014. 40E. Nenasheva, N. Kartenko, I. Gaidamaka, O. Trubitsyna, S. Re- dozubov, A. Dedyk, and A. Kanareykin, "Low loss microwave ferroelectric ceramics for high power tunable devices," Journal of the European Ceramic Society, vol. 30, no. 2, pp. 395 -- 400, 2010. 41"Determination of rf field strength, power density and sar in the vicinity of radiocommunication base stations for the purpose of evaluating human exposure," tech. rep., IEC, 2017.
1910.01838	1	1910	2019-10-04T08:51:42	Active Metasurfaces as a Platform for Capacitive Wireless Power Transfer Supporting Multiple Receivers	[ "physics.app-ph" ]	As wireless power transfer (WPT) repeaters, metasurfaces can enhance field coupling, improving the WPT operation. In this paper, we show that metasurfaces can also be used as transmitters of capacitive WPT systems. Such a metasurface-based WPT system can feed multiple receivers and provide robust operation against load or position variations. We formulate an analytical model of such WPT systems. We also discuss the exact solution of a particular example with $N$ identical receivers.	physics.app-ph	physics	Active Metasurfaces as a Platform for Capacitive Wireless Power Transfer Supporting Multiple Receivers Fu Liu, Prasad Jayathurathnage and Sergei A. Tretyakov Department of Electronics and Nanoengineering, Aalto University, P.O. Box 15500, Espoo, Finland [email protected] Abstract -- As wireless power transfer (WPT) repeaters, metasurfaces can enhance field cou- pling, improving the WPT operation. In this paper, we show that metasurfaces can also be used as transmitters of capacitive WPT systems. Such a metasurface-based WPT system can feed multiple receivers and provide robust operation against load or position variations. We formu- late an analytical model of such WPT systems. We also discuss the exact solution of a particular example with N identical receivers. I. INTRODUCTION Integrated with lumped elements or mixed-signal integrated circuits, tunable metasurfaces have shown great capability to manipulate electromagnetic waves and a vast spectrum of applications have been achieved, such as polarization conversion, wavefront shaping, holography, and tunable perfect anomalous reflections [1, 2]. On the other hand, wireless power transfer (WPT) is an emerging and fast growing research topic and metasurfaces (and metamaterials) can be introduced to WPT systems to enhance the system efficiency by engineering local evanescent fields [3]. For example, a wire medium slab can convert evanescent waves to propagating modes and therefore it can be used to increase magnetic coupling between two distant coils. Based on this idea, a smart table, which can support multiple receivers, has been constructed by using a planar version of wire media [4]. In this paper, rather than using magnetic coupling, we show that a tunable metasurface can work as a smart table which can wirelessly transfer power to multiple receivers with capacitive coupling. Moreover, the proposed solution ensures robust WPT operations subjected to changes of receiver position and load resistance. Recent studies have shown that robust WPT operations can be achieved by using a capacitive self-adaptive WPT system, in which the operational amplifier (op-amp) works as a switch. Robustness is ensured by a special circuit topology where the coupling capacitors and the load are parts of the feedback loop [5, 6]. Therefore, when the receiver position (equivalently, the capacitive coupling strength) or load resistance change, the system can automatically adjust itself to the optimal working condition. Metasurfaces formed by patch arrays can be naturally used to form capacitors, thus, they can be used as transmitters for a capacitive WPT system. Moreover, many patches of the metasurface can support multiple receivers. Such a working arrangement is schematically shown in Fig. 1, where different rows of patches on the metasurface are connected to terminal 1 (the output terminal of op- amp) or terminal 2 (the inverting terminal of the op-amp) of a self-adaptive WPT system, see Fig. 2(a). Whenever a receiver is present and the self-oscillating condition is satisfied, a feedback loop through that receiver is formed and power is wirelessly transferred to the load through the capacitive link. 9 1 0 2 t c O 4 ] h p - p p a . s c i s y h p [ 1 v 8 3 8 1 0 . 0 1 9 1 : v i X r a Fig. 1: Schematic of the metasurface-based WPT system supporting multiple receivers. Receiver 1Receiver 2Receiver 3Fig. 2: (a) Circuit diagram of the metasurface-based capacitive WPT system utilizing the self-oscillating approach. The notations of the voltages and currents are also shown. (b) The oscillation period for different numbers of identical receivers. The solid line (symbols) are from the analytical formula Eq. 7 (LTSpice simulation). II. ANALYTICAL MODEL OF THE METASURFACE BASED WPT SYSTEM The circuit diagram of the metasurface-based capacitive WPT system supporting multiple receivers is shown in Fig. 2(a). For simplicity, we assume that the op-amp is ideal, i.e., lossless with infinite slew rate and infinite input impedance. We also assume that the coupling capacitance of the two capacitors for each receiver, formed by the patches on the metasurface and the patches on the receiver, are the same and denote them as Cp,n for the nth receiver, as shown in Fig. 2(a). When the system operates, capacitors C0 and Cp,n(n = 1, 2, ...) undergo charging and discharging processes due to non-zero and different voltages Vo and V1. By comparing V1 to the reference voltages Vr, the op-amp will switch the output Vo between VCC and VEE (= −VCC in this work for symmetric operation). Therefore voltage oscillations are formed and power is wirelessly generated at each load position [6]. In such a multiple receiver WPT system, the master equations can be formulated from the circuit theory. First of all, for each receiver, the voltage drops across the two coupling capacitors are equal because the capacitances are equal, i.e., Vo − V3,n = V2,n − V1, which gives V2,n = Vo + V1 − V3,n. Secondly, as the current in through the two coupling capacitors and the load RL,n are identical, we obtain 2V3,n − Vo − V1 d(Vo − V3,n) in = Cp,n = dt RL,n , (n = 1, 2, ..., N ) (1) for each receiver, where N is the total number of receivers. Finally, from Kirchhoff's current law, the total current through the capacitor C0 is the sum of the currents through all the N receivers (note that zero current flows into the op-amp due to its infinite input impedance), giving N(cid:88) C0 dV1 dt = 2V3,n − Vo − V1 n=1 RL,n . (2) Equations (1) and (2), in total N +1 equations, are the master equations of the WPT system with multiple receivers shown in Fig. 2(a). For a particular configuration, C0, Cp,n, RL,n, Vo, and reference voltage Vr = βVo, where β = R1/(R1 + R2) is given by the voltage divider resistors, are known and there are N + 1 unknown variables, i.e., V1 and V3,n. Therefore, the system can be solved with proper initial conditions. In the next section, we give one example for many identical receivers, providing an analytical solution. III. EXACT SOLUTION FOR N IDENTICAL RECEIVERS When all N receivers are identical, i.e., same coupling Cp and load RL, the total current i will be equally distributed to the loads, and the master equations (1) and (2) can be simplified into −Cp C0 dV3 dt dV1 dt 2V3 − Vo − V1 , 2V3 − Vo − V1 RL = = N RL . (3) (4) On the other hand, the initial conditions can be set when the output voltage switches from VCC to VEE, which gives Vot=0 = −VCC, V1t=0 = βVCC, and V3t=0 = −(1 + βC0/(N Cp))VCC. Therefore, the solution for this WPT −+(cid:1844)(cid:2869)(cid:1844)(cid:2870)op-amp(cid:1829)(cid:2868)⋯12(cid:1866)⋯(cid:1848)(cid:2928)(cid:1848)(cid:2925)(cid:1848)(cid:2869)(cid:1848)(cid:2870),(cid:3041)(cid:1848)(cid:2871),(cid:3041)(cid:1829)(cid:2926),(cid:3041)(cid:1829)(cid:2926),(cid:3041)(cid:1844)(cid:2896),(cid:3041)(cid:1861)(cid:3041)(cid:1861)(cid:1848)(cid:2887)(cid:2887)(cid:1848)(cid:2889)(cid:2889)(cid:1844)(cid:2896),(cid:2870)(cid:1844)(cid:2896),(cid:2869)Terminal 1Terminal 2246810020406080100 Analytic SimulationPeriod T (ms)N(a)(b)system with N identical receivers is analytically obtained as (cid:18) VCC (cid:19) , − 2C0+N Cp C0 CpRL t − N Cp (5) V1(t) = V3(t) = −C0/(N Cp)V1(t) − VCC. 2C0 + N Cp (N Cp + 2βC0 + N βCp)e (6) Similarly to [6], we can find the oscillation period (from V1(T /2) = −βVCC), the oscillation condition (from V1(t → ∞) < −βVCC), and the averaged transferred power to each load (from 1/T(cid:82) T 0 (V3 − V2)2/RLdt) as T = 2 C0CpRL 2C0 + N Cp ln (cid:20) N Cp + (2C0 + N Cp)β N Cp − (2C0 + N Cp)β (cid:21) , β < Pavg = N Cp , 2C0 + N Cp V 2 CC RL N Cp 2(2C0 + N Cp)β (cid:30) (cid:20) N Cp + (2C0 + N Cp)β N Cp − (2C0 + N Cp)β (cid:21) . ln (7) (8) (9) Here, we note that these results simplify to those in [6] when N = 1. The results show that the WPT operation is robust against the load and coupling (receiver position) variations within the working range, similarly to [6]. To verify the analytical results, we have performed LTSpice simulations for different numbers of identical receivers. The system is configured with lumped elements R1 = 1 MΩ, R2 = 3.9 MΩ, C0 = 1 nF, Cp = 3 nF, RL = 100 kHz, and VCC/EE = ±10 V. The simulated periods for N = 2, 3, ..., 9 identical receivers are plotted as symbols in Fig. 2(b), and they agree very well with the analytic formula Eq. 7 (solid line). In the presentation, we will also discuss the dynamic adaptation of the system with many different receivers at variable positions. IV. CONCLUSION The possibility of using active and tunable metasurfaces for multiple-receiver WPT systems has been discussed. The master equations of such a WPT system have been formulated and analytical results obtained for particular examples of multiple identical receivers. The working properties of this WPT system still holds when multiple receivers are different. However, analytical solutions may be difficult to find as the initial conditions of V3,n involve time integration of all the currents. We hope that these results will encourage future implementations of such self-adaptive smart tables for WPT applications. This work was supported by the European Union's Horizon 2020 Future Emerging Technologies call (FETOPEN- RIA) under Grant Agreement No. 736876 (project VISORSURF) ACKNOWLEDGEMENT REFERENCES [1] F. Liu, A. Pitilakis, M. S. Mirmoosa, O. Tsilipakos, X. Wang, A. C. Tasolamprou, S. Abadal, A. Cabellos-Aparicio, E. Alarcn, C. Liaskos, N. V. Kantartzis, M. Kafesaki, E. N. Economou, C. M. Soukoulis, and S. A. Tretyakov, Programmable metasurfaces: state of the art and prospects, 2018 IEEE International Symposium on Circuits and Systems (ISCAS), 2018 [2] F. Liu, O. Tsilipakos, A. Pitilakis, A. C. Tasolamprou, M. S. Mirmoosa, N. V. Kantartzis, D.-H. Kwon, M. Kafesaki, C. M. Soukoulis, S. A. Tretyakov, "Intelligent metasurfaces with continuously tunable local surface impedance for multiple reconfigurable functions," Physical Review Applied, vol. 11, p. 044024, 2019. [3] M. Song, P. Belov, P. Kapitanova, "Wireless power transfer inspired by the modern trends in electromagnetics," Applied Physics Reviews, vol. 4, p. 021102, 2017. [4] M. Song, K. Baryshnikova, A. Markvart, P. Belov, E. Nenasheva, C. Simovski, and P. Kapitanova, "Smart table based on a metasurface for wireless power transfer," Physical Review Applied, vol. 11, p. 054046, 2019. [5] Y. Ra'di, B. Chowkwale, C. Valagiannopoulos, F. Liu, A. Al`u, C. R. Simovski, S. A. Tretyakov, "On-site wireless power generation," IEEE Transacions on Antennas and Propagation, vol. 66, pp. 4260 -- 4268, 2018. [6] F. Liu, B. Chowkwale, P. Jayathurathnage, S. Tretyakov, "Robust Wireless Power Transfer: A Self-Adaptive Approach," arXiv preprint:1806.09438, 2018.
1908.10444	1	1908	2019-07-30T20:22:52	Mathematical Modelling and Comparative Study of Elliptical and Circular Capacitive Pressure Microsensor	[ "physics.app-ph" ]	A lot of work on clamped circular, square and rectangular shaped capacitive pressure sensor has been carried out. However, to the best of our knowledge, no elaborate work has been performed on mathematical formulation and simulation of clamped elliptical shaped capacitive pressure sensor in literature. This paper describes the mathematical modelling and simulation of normal mode clamped elliptical shaped capacitive pressure sensor. The study of various performance parameters like maximum diaphragm deflection, capacitance variation, mechanical sensitivity, capacitive sensitivity and non-linearity is also carried out for operating pressure range of 0 kPa to 18 kPa. For circular capacitive pressure sensor, the operating pressure range is modified according to physical dimensions i.e. thickness and radius of diaphragm, separation gap between plates and maximum deflection. For same overlapping area between plates (10000*pi mm2), the comparison of elliptical shapes of different eccentricities with circular shape diaphragm is carried out. The thickness of diaphragm is taken as 2 um and separation gap is 1 um in all the designs which are used in this work. In this comparative study, it is observed that elliptical shaped capacitive pressure sensors have better linearity than circular diaphragm pressure sensor.	physics.app-ph	physics	Mathematical Modelling and Comparative Study of Elliptical and Circular Capacitive Pressure Microsensor Rishabh Bhooshan Mishra, S. Santosh Kumar* Process Technologies Group, Smart Sensor Area, CSIR -- Central Electronics Engineering Research Institute (CEERI), Pilani, Rajasthan, India -- 333031 *Email: [email protected] Abstract. A lot of work on clamped circular, square and rectangular shaped capacitive pressure sensor has been carried out. However, to the best of our knowledge, no elaborate work has been performed on mathematical formulation and simulation of clamped elliptical shaped capacitive pressure sensor in literature. This paper describes the mathematical modelling and simulation of normal mode clamped elliptical shaped capacitive pressure sensor. The study of various performance parameters like maximum diaphragm deflection, capacitance variation, mechanical sensitivity, capacitive sensitivity and non-linearity is also carried out for operating pressure range of 0kPa -- 18 kPa. For circular capacitive pressure sensor, the operating pressure range is modified according to physical dimensions i.e. thickness and radius of diaphragm, separation gap between plates and maximum deflection. For same overlapping area between plates (10000 mm2), the comparison of elliptical shapes of different eccentricities with circular shape diaphragm is carried out. The thickness of diaphragm is taken as2 µm and separation gap is 1 µm in all the designs which are used in this work. In this comparative study, it is observed that elliptical shaped capacitive pressure sensors have better linearity than circular diaphragm pressure sensor. 1. Introduction Micro electromechanical system (MEMS) is popular and developed technology, adopted from IC- processing, for fabricating the micro-sensors for various process variable measurements like pressure, temperature, distance, acceleration, fluid flow, rotation and angular velocity etc. In general, MEMS devices consist of moveable micromechanical components like bridges, beams, gears, diaphragms, channels, micro-pumps, valves, cantilevers which are integrated or interfaced with electronic components or devices. The Silicon based devices/structures are micro-fabricated by a particular process flow sequence of photolithography, etching and deposition to fabricate desired components/structure. The integration of microelectronic and mechanical components in a single chip makes the MEMS devices more versatile than conventional devices. To apply conventional solid mechanics theories and practices to the MEMS based systems are interesting and challenging task. For measuring the absolute, gauge or differential pressure using piezo-resistive and capacitive pressure sensors, the MEMS based sensors utilize movement in diaphragm due to pressure application [1-4]. In piezo-resistive pressure sensor, four piezo-resistors are mounted on the diaphragm and connected in whetstone bridge fashion. After pressure application, there is deflection in the diaphragm on which piezo-resistors are mounted. The change in resistance of implanted piezo-resistors unbalances the Wheatstone bridge which provides the output voltage according to the applied pressure [5-6]. However, due to several advantages of capacitive pressure sensor over piezo-resistive pressure sensor like long term stability, less power consumption and temperature drift etc., the capacitive pressure sensors are used in several applications like biomedical, aerospace, consumer electronics and automobile. Achieving linearity characteristic and sealing vacuum cavity is a major tough task in fabrication of the capacitive pressure sensor. Since there is no need to mount any piezo-resistor in the capacitive sensor diaphragm, so scaling the dimension of the diaphragm is bit easier. However, nonlinearity between applied pressure vs. capacitance change, large impedance of output signal of sensor, small change in capacitance and parasitic capacitance between output of device and capacitance measurement circuitry are some of the major disadvantage of capacitive pressure sensor. Some issues must be addressed by circuitry which is used to measure the change in capacitance. As sealing the vacuum cavity is complex task, so proper care must be been taken at the time of fabrication and packaging the absolute capacitive pressure sensor [7-9]. Bio-MEMS is, one of revolutionized area of MEMS application, utilizing VLSI or micro- fabrication technology for implantable/physiological devices because of several advantages. In living beings, several implantable devices are based on MEMS technology like Biosensors for in-vivo sensing, immune-isolation devices, drug delivery systems (micro-reservoir, micro-particle) and some inject able devices like micro-needle, micro-module. Subsequently, the data obtained from the Bio- MEMS devices is transmitted using wireless communication technology for study of various body- parts or continuous monitoring purpose. In medical sector, if any part of living-being dysfunctions then several steps can be taken to save the life [10-11]. The general capacitance measurement circuitry for measuring capacitive pressure sensor is astable mutivibrator which converts the capacitance variation (due to applied pressure) into frequency. The astable mutivibrator contains a LF351 Op-amp and three external resistances. And MS3110 is the commercially available Integrated Circuitry for capacitance variation measurement which converts capacitance variation into voltage [12]. The algometric capacitance to digital converters can also be utilized for measuring the base capacitance and variation in capacitance after pressure application [13]. 2. Paper Structure The sufficient amount of work like design, modelling, simulation and fabrication of square and circular shaped capacitive pressure sensors have been performed earlier. However, as per best of our literature review, no mathematical modelling and numerical simulation of clamped elliptical shaped capacitive pressure sensor is capacitive pressure sensor. The simulation and fabrication of L-shaped clamped elliptical capacitive pressure sensor, with signal conditioning circuitry, have been reported [14]. The finite element analysis (FEA) of elliptical shaped capacitive pressure sensor, with temperature variation, has been reported [15]. However the mathematical analysis and study of various performance characteristics with numerical analysis is unavailable. In the presented work, elliptical shaped diaphragms of different eccentricities are chosen and compared with the circular shaped diaphragm. The overlapping area between plates is kept constant for all sensor designs. In all, six different designs consisting of one circular shaped and five elliptical shaped capacitive pressure sensors of 2 µm thicknesses and 1 µm separation gap are chosen for numerical simulation. Overlapping area of different designs are same i.e. 10000π µm2. The specifications of six different designs are listed in Table 1. Where, a and b are semi-major and semi- minor axes of ellipse and L is radius of circle. Table 1. Dimensions of Specified Sensors. S. N. Diaphragm Shape Diaphragm dimensions (µm) 1. Circular L = 100 Elliptical a = 125, b = 80 2. 3. Elliptical a = 200, b = 50 4. Elliptical a = 250, b = 40 5. Elliptical a = 500, b = 20 6. Elliptical a = 1000, b = 10 3. Deflection Theory and Kirchhoff's assumption The deflection theory is based on small and large deflection in diaphragm/plate. To apply small deflection theory, the diaphragm deflection must be less than 1/5th of diaphragm thickness. In case of large deflection theory, the diaphragm deflection must be five times of diaphragm thickness or more. The small deflection theory of plate is utilized for normal mode capacitive pressure sensor and large deflection theory can be utilized for touch or double touch mode capacitive pressure sensor [4, 7-8]. The Small deflection theory is based on certain Kirchhoff's assumptions, which are:      The first assumption is based on material property of diaphragm. The diaphragm must be made of elastic, homogeneous and isotropic material. The second assumption deals with the dimension or geometry of diaphragm. The diaphragm must be thin as well as flat. The dimension of diaphragm must be larger than ten times than diaphragm thickness. The third assumption states about the maximum diaphragm deflection, which must not be larger than 1/5th of diaphragm thickness. The fourth assumption is known as 'hypothesis of straight normal' which states that the middle plane of diaphragm does not have normal stress and should not be strained. The strain lines should remain straight and perpendicular to the middle plane. The middle plane of plate remains unstrained after bending. 4. Analytical Modeling 4.1 Deflection in Clamped Elliptical Diaphragm The boundary equation of classical elliptical shaped plate, shown in Figure 1, can be given by: (cid:4672) (cid:2870) (cid:4673) 𝑥 𝑎 + (cid:4672) 𝑦 𝑏 (cid:2870) (cid:4673) = 1 The eccentricity, 𝑒, of the elliptical boundary can be given by (if b  a): (1) (2) b 𝑒 = (cid:3496)1 − (cid:3436) (cid:2870) (cid:3440) 𝑏 𝑎 Y Clamped (x,y) (0,0) X a 1. The basic Figure capacitive pressure sensor consists two parallel plates which are separated by a media. In which one plate will be fixed and another is pressure moveable application. This shown figure top view of capacitive pressure sensor which elliptical shaped diaphragm. after is consist If uniform pressure P is applied, in normal direction, to clamped elliptical shaped Kirchhoff's plate of thickness, Young's modulus of elasticity and Poisson's Ratio t, E and ѵ, respectively, then plate equation is expressed by a fourth order partial differential equation which can be given by [16-17]: ∆∆ [𝑤(𝑥, 𝑦)] = 𝑃 𝐷 (3) here, flexural rigidity of plate is D = Et(cid:2870)/(12 − ѵ(cid:2870)), w(x, y) is deflection at any point (x, y) on plate and ∆ is Laplace operator or Laplacian. The deflection in plate is perpendicular to 2D -- plane. In 2D co-ordinate ∆ can be given by: 𝜕(cid:2870) 𝜕𝑦(cid:2870) 𝜕(cid:2870) 𝜕𝑥(cid:2870) + The applied boundary conditions to solve the eq. (3) are: ∆ ≡ ∇(cid:2870) = and, 𝑤 = 0 𝜕𝑤 𝜕𝑛 = 0 (4) (5) (6) here, the normal to elliptical plate edge is 𝑛. Using both the boundary conditions, deflection in clamped elliptical plate can be given by [16-17]: 𝑥 𝑤(cid:3032)(𝑥, 𝑦) = 𝑤(cid:3032),(cid:2868) (cid:3428)1 − (cid:4672) 𝑎 (cid:2870) (cid:4673) − (cid:4672) 𝑦 𝑏 (cid:2870) (cid:4673) (cid:3432) (cid:2870) (7) The maximum deflection due to pressure application is found at the centre of plate and this can be obtained after putting x = 0 and y = 0 in eq. (7). Therefore, maximum deflection in the clamped elliptical shaped diaphragm for 𝑎 > 𝑏 can be given by [16-17]: 𝑤(cid:3032),(cid:2868) = 𝑃 (cid:2871) (cid:3029)(cid:3120) + (cid:2870) (cid:3028)(cid:3118)(cid:3029)(cid:3118)(cid:4673) 8𝐷 (cid:4672) (cid:2871) (cid:3028)(cid:3120) + (8) In case of zero eccentricity i.e. 𝑎 = 𝑏 = 𝐿, the given formula of deflection in elliptical diaphragm can be modified for clamped circular shaped diaphragm which can be given by [4,7-9,12, 16-17]: 𝑤(cid:3030) = 𝑤(cid:3030),(cid:2868) (cid:3428)1 − (cid:4672) (cid:2870) (cid:4673) (cid:2870) (cid:3432) 𝑟 𝐿 here, 𝑤(cid:3030),(cid:2868) is the maximum deflection in circular shaped plate which can be given by: 𝑤(cid:3030),(cid:2868) = 𝑃𝐿(cid:2872) 64𝐷 (9) (10) From eq. (8) and (10), it is clear that the maximum deflection, in clamped elliptical and circular both, is the function of applied pressure, dimensions of plate, material properties and plate thickness. 4.2 Base Capacitance of Sensor If permittivity of medium and separation gap between plates is ε and d, respectively, then base capacitance of sensor can be given by []7-9, 12: 𝐶(cid:3029) = 𝜀 × overlapping area between plates separation gap (11) The overlapping area of elliptical shaped plate can be given by: 𝐴(cid:3032) = 𝜋𝑎𝑏 And overlapping area of circular shaped plate can be given by: 𝐴(cid:3030) = 𝜋𝐿(cid:2870) (12) (13) 4.3 Capacitance variation in sensor The capacitance in terms of applied pressure P on elliptical plate can be given by double integral [4, 7]: 𝐶(cid:3050),(cid:3032) = (cid:3509) (cid:3008) 𝜀 𝑑𝑥 𝑑𝑦 𝑑 − 𝑤(𝑥, 𝑦) (14) here, region 𝐺 is: (cid:4672) Since the term which depends on applied pressure i.e. d − w(x, y), is in denominator of capacitance variation, so the capacitance increases as diaphragm deflection increases. = 1. (cid:2870) (cid:2934) (cid:4673) (cid:2911) (cid:2870) (cid:2935) + (cid:4672) (cid:4673) (cid:2912) From Eq. (7), (8) and (14): 𝐶(cid:3050),(cid:3032) = (cid:3509) (cid:3008) 𝜀 𝑑𝑥 𝑑𝑦 𝑑 − 𝑤(cid:3032),(cid:2868) (cid:3428)1 − (cid:4672) (cid:2870) (cid:4673) (cid:3051) (cid:3028) (cid:3052) − (cid:4672) (cid:3029) (cid:2870) (cid:4673) (cid:3432) (cid:2870) After applying transformation: x = au and y = bv [18]: 𝐶(cid:3050),(cid:3032) = 𝜀𝑎𝑏 (cid:3509) (cid:3009) 𝑑𝑢 𝑑𝑣 𝑑 − 𝑤(cid:3032),(cid:2868)[1 − 𝑢(cid:2870) − 𝑣(cid:2870)](cid:2870) (15) (16) here, region H is: u(cid:2870) + v(cid:2870) = 1. After applying transformation theory of cylindrical coordinate, u = q cosφ, v = qsinφ and z= z, we get: 𝐶(cid:3050),(cid:3032) = 𝜀𝑎𝑏 (cid:3509) (cid:3010) 𝑞𝑑𝜑𝑑𝑞 𝑑 − 𝑤(cid:3032),(cid:2868)[1 − 𝑞(cid:2870)](cid:2870) here, region 𝐼 is: 𝑞(cid:2870) = 1. After solving this double integral, we get: 𝐶(cid:3050),(cid:3032) = 𝜀𝐴(cid:3032) 2(cid:3493)𝑑𝑤(cid:3032),(cid:2868) ln (cid:3629) √𝑑 + (cid:3493)𝑤(cid:3032),(cid:2868) √𝑑 − (cid:3493)𝑤(cid:3032),(cid:2868) (cid:3629) (17) (18) In case of zero eccentricity i.e. for the circular diaphragm (when a = b = L), the capacitance after pressure application can be given by [4, 7-9]: 𝐶(cid:3050),(cid:3030) = 𝜀𝐴(cid:3030) 2(cid:3493)𝑑𝑤(cid:3030),(cid:2868) ln (cid:3629) √𝑑 + (cid:3493)𝑤(cid:3030),(cid:2868) √𝑑 − (cid:3493)𝑤(cid:3030),(cid:2868) (cid:3629) (19) 4.4 Sensitivity of Sensor The mechanical sensitivity of capacitive pressure sensor can be obtained by differentiating deflection at the centre i.e. maximum deflection w.r.t. pressure which plays very important role in optimization of sensor designs, if the maximum deflection for few sensor designs is approximately same. Then mechanical sensitivity of elliptical shaped capacitive pressure sensor can be given by: 𝑆(cid:3032),(cid:3040)(cid:3032)(cid:3030)(cid:3035) = 𝜕𝑤(cid:3032),(cid:2868) 𝜕𝑃 After differentiating the Eq. (8), we get: 𝑆(cid:3032),(cid:3040)(cid:3032)(cid:3030)(cid:3035) = 1 (cid:2871) (cid:3029)(cid:3120) + (cid:2870) (cid:3028)(cid:3118)(cid:3029)(cid:3118)(cid:4673) 8𝐷 (cid:4672) (cid:2871) (cid:3028)(cid:3120) + (20) (21) The mechanical sensitivity of capacitive pressure sensor is function of dimension and flexural rigidity of plate. The capacitive sensitivity of capacitive pressure sensor is obtained by differentiating capacitance change w.r.t. pressure range. Then capacitive sensitivity of elliptical shaped capacitive pressure sensor can be given by [8-9]: 𝑆(cid:3032),(cid:3030)(cid:3028)(cid:3043) = 𝜕𝐶(cid:3050),(cid:3032) 𝜕𝑃 After performing differentiation, we get: 𝑆(cid:3032),(cid:3030)(cid:3028)(cid:3043) = 𝑆(cid:3032),(cid:3040)(cid:3032)(cid:3030)(cid:3035) 2𝑤(cid:3032),(cid:2868) (cid:4678) 𝜀𝐴(cid:3032) 𝑑 − 𝑤(cid:3032),(cid:2868) − 𝐶(cid:3050),(cid:3032)(cid:4679) (22) (23) 4.5 Non-linearity in Sensor The non-linearity of a sensor is defined as the deviation of ideal curve with obtained output curve of sensor. This ideal curve is obtained using end point straight line or least point fit curve. For each point on ideal curve there will be particular deviation in the output curve. And the maximum deviation of output curve w.r.t. ideal curve (in percentage) is called non-linearity. The non-linearity, at any specific point, of elliptical shaped capacitive pressure sensor is given by [5,9]: 𝑁𝐿(cid:3036)(%) = 𝐶(cid:3050),(cid:3032)(𝑃(cid:3036)) − 𝐶(cid:3050),(cid:3032)(𝑃(cid:3040)) × 𝐶(cid:3050),(cid:3032)(𝑃(cid:3040)) (cid:3017)(cid:3284) (cid:3017)(cid:3288) × 100 (24) here, Pi is the pressure at ideal curve, Pm is the maximum applied pressure, Cw,e(Pi) is the output capacitance at pressure Pi on ideal curve and Cw,e(Pm) is output capacitance at pressure Pm on ideal curve. 5. Simulation Results and Discussion The maximum deflection in circular and elliptical shaped diaphragm at different pressures is shown in Figure 2. The diaphragm deflection is kept less than 1/5thof diaphragm thickness (i.e. 0.4 µm) and 1/4thof separation gap (i.e. 0.25 µm) to follow small deflection theory of plates and avoid pull-in phenomena, respectively. The operating pressure range is 0kPa to 18 kPa. Silicon is used as the diaphragm material; it has a Young's Modulus of elasticity of 169.8 GPa and Poisson ratio of 0.066. The deflection is found to be more in the case of circular diaphragm than in the elliptical shaped diaphragms. All the designs have linear defection curve in applied pressure range which follows the Hook's Law. However, as the value of semi-major axis increases and semi-minor axis of elliptical diaphragm decreases, keeping constant the overlapping area between plates, the deflection in plate decreases. Figure 3 represents 2-D plot of deflection in elliptical shaped clamped diaphragm of designed sensor. In this 2-D plot, the deflection is maximum at centre of diaphragm and then decreases as we move towards the clamped edges. 250 ) 200 L=100µm a=125µm, b=80µm a=200µm, b=50µm a=250µm, b=40µm a=500µm, b=20µm a=1000µm, b=10µm m n ( n o i t c e l f e D x a M 150 100 50 0 0 2 4 8 6 12 Pressure (kPa) 10 14 16 18 Figure 2. Deflection in various diaphragms of thickness 2 µm. Figure 3. Deflection plot in elliptical diaphragm of a = 125 µm and b = 80 µm of 2 µm thickness at 18 kPa Pressure. The overlapping area of all the proposed designs are kept same, so base capacitance (capacitance without bending of diaphragm due to applied pressure) of all the designs are 0.27816 pF. The capacitance variation in different sensor designs, w.r.t. applied pressure range of 0 kPa -- 18 kPa, is given in Figure 4. The capacitance change in circular capacitive pressure sensor is found maximum. The capacitance change decreases by decreasing semi-minor axis and increasing semi-major axis, keeping the overlapping area between plates constant. L=100µm a=125µm, b=80µm a=200µm, b=50µm a=250µm, b=40µm a=500µm, b=20µm a=1000µm, b=10µm 0.305 0.300 0.295 0.290 0.285 0.280 ) F p ( e c n a t i c a p a C 0 2 4 8 6 12 Pressure (kPa) 10 ) % ( y t i r a e n L - n o N i 0 -1 -2 -3 -4 -5 14 16 18 0 2 4 L = 100 µm a=125µm, b=80µm a=200µm, b=50µm a=250µm, b=40µm a=500µm, b=20µm a=1000µm, b=10µm 8 6 12 Pressure (kPa) 10 14 16 18 Figure 4. Capacitance variation in sensors of specified dimensions. Figure 5. Non-linearity in sensors of specified dimensions. The mechanical and capacitive sensitivity of different diaphragm dimensions are shown in Table 2. Mechanical sensitivity can be important factor, if the deflection due applied pressure is different in case of different design. The mechanical sensitivity of circular shaped capacitive pressure sensor is largest among all designs and decreases with increasing the value of semi-major axis and decreasing the value of semi-minor axis in case of elliptical capacitive pressure sensor. The capacitive sensitivity of circular plate is highest amongst all proposed designs. The capacitive sensitivity decreases in case of elliptical capacitive pressure sensor as value of semi-major axis increases and semi-minor axis decreases, keeping the overlapping area between plates constant. Figure 5 shows the non-linearity of different designs. Non-linearity is one of important parameter for pressure sensors. The problem of the non-linearity can be controlled by reducing large deflection of diaphragm centre. The non-linearity in circular shaped capacitive pressure sensor is maximum. In case of elliptical shaped capacitive pressure sensor, non-linearity reduces as the semi-major axis increases and correspondingly semi-minor axis decreases. Table 2. Mechanical and Capacitive Sensitivity of Specified Sensors. S. N. Diaphragm Dimension (µm) Mech. Sensitivity (µm/kPa) Cap. Sensitivity (fF/kPa) 1. L = 100 1.3743×10-2 a = 125, b = 80 1.1042×10-2 2. 2.1906×10-3 3. a = 200, b = 50 9.2184×10-3 4. a = 250, b = 40 5. a = 500, b = 20 5.8574×10-5 a = 1000, b = 10 3.6645×10-6 6. 6. Conclusion In this work, the mathematical modelling and simulation for elliptical shaped capacitive pressure sensor is carried out and the various parameters for pressure sensors are compared with circular shape sensor, while the overlapping area between plates is kept same. To the best of our knowledge, there is no elaborate work in literature about elliptical shaped capacitive pressure sensor. The various performance parameters have been derived step by step using mathematical modelling and MATLAB® is utilized for validating the mathematical modelling. In performing simulations, the overlapping area between plates is kept constants for all specified dimensions of elliptical and circular shape. 1.5044 1.0916 2.0778×10-1 8.611×10-2 5.4343×10-3 3.3978×10-5 The circular diaphragm has more deflection, higher change in capacitance, better mechanical and capacitive sensitivities in comparison to other elliptical diaphragms even though all of them have same overlapping area between plates. However, it also has the highest non-linearity of 4.22%. The elliptical diaphragm of dimension a = 125 µm and b = 80 µm has maximum deflection of 0.24737 µm at maximum applied pressure is 18 kPa, change in capacitance at full scale value of pressure is 21.49 fF, mechanical sensitivity is 1.10418×10-2 µm/kPa, capacitive sensitivity is 1.09167 fF/kPa and maximum non-linearity is -3.089% which is 3/4th of circular shaped capacitive pressure sensor. Therefore, it is concluded that elliptical shaped diaphragms are better suited when non-linearity is an important consideration. Otherwise, the circular shaped diaphragm is more suitable in terms of all other parameters. Acknowledgement Authors acknowledge the Director, CSIR -- CEERI, Pilani, for providing generous support. This presented analysis is also supported by Dr. Ankush Jain (Scientist, CSIR-CEERI). The authors wish to acknowledge all the scientific and technical staff of Process Technologies Group, Smart Sensor Area, CSIR-CEERI. References [1] [2] [3] [4] Lin L, Yun W 1998 Design, optimisation and fabrication of surface micromachined pressure sensors Mechatronics 8 505-519 Kota S, Ananthasuresh G K, Crary S B and Wise K D 1994 Design and Fabrication of Microelectromechanical Systems J. of Mechanical Design 116 1081-1088 Ananthasuresh G K, Howell L L 2005 Mechanical design of compliant microsystems -- a perspective and prospects J. of Mechanical Design 127 736-738. Mishra R B, Santosh Kumar S and Mukhiya R 2018 Analytical Modelling and FEM Simulation of Capacitive Pressure Sensor for Intraocular Pressure Sensing IOP Conf. Ser.: Mater. Sci. Eng. 404 012026 [5] [6] [7] [8] [9] Santosh Kumar S and Pant B D 2014 Design principles and considerations for the 'ideal' silicon piezoresistive pressure sensor: a focused review Microsystem Technologies 20 1213-1247 San H, Li Y, Song Z, Yu Y and Chen X 2013 Self-Packaging Fabrication of Silicon-Glass-Based Piezoresistive Pressure Sensor IEEE Electron Device Letters 34(6) 789-791 Mishra R B, Santosh Kumar S and Mukhiya R 2019 Design and Simulation of Capacitive Pressure Sensor for Blood Pressure Sensing Application In Recent Trends in Communication, Computing and Electronics Springer 524 441-449 Mishra R B, Santosh Kumar S 2018 Pre-stressed Diaphragm based Capacitive Pressure Sensor for Blood Pressure Sensing Application Proc. of IEEE Int. Conf. on Advances in Computing, Control and Communication Technology ( Allahabad) Mishra R B, Santosh Kumar and Mukhiya R 2018 Modeling and FEM-based Simulations of Composite Membrane based Circular Capacitive Pressure Sensor Proc. of Int. Conf. VLSI, Communication and Signal Processing (Allahabad) [10] Grayson A C R, Shawgo R S, Johnson A M, Flynn N I, Li Y, Cima M J and Langer R 2014 A Bio-MEMS Review: MEMS [11] [12] [13] [14] technology for phisiologically integrated devices Proc. of the IEEE 92 6 -- 21 Chiang C-C, Lin C-C K and Ju M-S 2007 An implantable capacitive pressure sensor for biomedical applications Sensors and Actuators A: Physical 134(2) 382-388 Chang S P and Allen M G 2004 Capacitive pressure sensors with stainless steel diaphragm and substrate J. of Micromechanics and Microengineering 14 612 -- 618 Rajpal B, Santosh M, Paliwal R and Bose S C 2018 Algorithmic ADC for Capacitive and Piezoelectric MEMS Sensor IEEE SENSORS 1-4. Sundararajan A D and Hasan S M R 2015 Elliptical Diaphragm Capacitive Pressure Sensor and Signal Conditioning Circuit Fabricated in SiGe CMOS Integrated MEMS IEEE Sensors Journal 15(3) 1825-1837 [15] Kubba A E, Hasson A, Kubba A I and Hall G 2016 A micro-capacitive pressure sensor design and modelling Journal of Sensors and Sensor Systems 5 95-112 Timoshenko S and Woinowsky-Krigger S 1959 Theory of Plates and Shells (New York: Mc. Graw-Hill) Birman V 2011 Plate structures Springer Science and Business 178 138 -- 141 [16] [17] [18] Kuipers L and Timman R 1969 Handbook of Mathematics (London: Pergamon Press) p 174 -- 184
1710.10485	2	1710	2017-12-01T17:56:31	Delayed pull-in transitions in overdamped MEMS devices	[ "physics.app-ph" ]	We consider the dynamics of overdamped MEMS devices undergoing the pull-in instability. Numerous previous experiments and numerical simulations have shown a significant increase in the pull-in time under DC voltages close to the pull-in voltage. Here the transient dynamics slow down as the device passes through a meta-stable or bottleneck phase, but this slowing down is not well understood quantitatively. Using a lumped parallel-plate model, we perform a detailed analysis of the pull-in dynamics in this regime. We show that the bottleneck phenomenon is a type of critical slowing down arising from the pull-in transition. This allows us to show that the pull-in time obeys an inverse square-root scaling law as the transition is approached; moreover we determine an analytical expression for this pull-in time. We then compare our prediction to a wide range of pull-in time data reported in the literature, showing that the observed slowing down is well captured by our scaling law, which appears to be generic for overdamped pull-in under DC loads. This realization provides a useful design rule with which to tune dynamic response in applications, including state-of-the-art accelerometers and pressure sensors that use pull-in time as a sensing mechanism. We also propose a method to estimate the pull-in voltage based only on data of the pull-in times.	physics.app-ph	physics	Delayed pull-in transitions in overdamped MEMS devices ‡ Michael Gomez, Derek E. Moulton, Dominic Vella Mathematical Institute, University of Oxford, Woodstock Rd, Oxford, OX2 6GG, UK E-mail: [email protected] Abstract. We consider the dynamics of overdamped MEMS devices undergoing the pull-in instability. Numerous previous experiments and numerical simulations have shown a significant increase in the pull-in time under DC voltages close to the pull-in voltage. Here the transient dynamics slow down as the device passes through a meta-stable or bottleneck phase, but this slowing down is not well understood quantitatively. Using a lumped parallel-plate model, we perform a detailed analysis of the pull-in dynamics in this regime. We show that the bottleneck phenomenon is a type of critical slowing down arising from the pull- in transition. This allows us to show that the pull-in time obeys an inverse square-root scaling law as the transition is approached; moreover we determine an analytical expression for this pull-in time. We then compare our prediction to a wide range of pull-in time data reported in the literature, showing that the observed slowing down is well captured by our scaling law, which appears to be generic for overdamped pull-in under DC loads. This realization provides a useful design rule with which to tune dynamic response in applications, including state- of-the-art accelerometers and pressure sensors that use pull-in time as a sensing mechanism. We also propose a method to estimate the pull-in voltage based only on data of the pull-in times. ‡ This is an author-created, un-copyedited version of an article accepted for publication/published in the Journal of Micromechanics and Microengineering. IOP Publishing Ltd is not responsible for any errors or omissions in this version of the manuscript or any version derived from it. The Version of Record is available online at https://doi.org/10.1088/1361-6439/aa9a70. Delayed pull-in transitions in overdamped MEMS devices 2 1. Introduction Electrostatic actuation is the most common actu- ation mechanism in microelectromechanical systems (MEMS): it offers rapid response times, low power consumption and compatibility with existing circuit technology [1, 2]. However, the operation of electro- static devices is limited by the 'pull-in' instability [3] in which an elastic structure suddenly collapses towards a nearby electrode when a critical voltage is exceeded. Pull-in can result in failure via short circuit or stiction between components. For this reason, studies have tra- ditionally focussed on the stability of devices under a combination of electrostatic and mechanical restoring forces, with a view to developing methods that extend the operating range of a device prior to pull-in [4]. More recently, pull-in has been identified as a useful instability for smart applications. For example, the critical voltage required to pull-in is commonly used in mass sensing applications [5] and to estimate material parameters such as the elastic modulus [6]. The dynamics of the pull-in transition is also becoming the basis of many MEMS devices. In these scenarios pull-in is allowed to proceed safely (e.g. by limiting the displacement of the structure to prevent contact between components), enabling fast motions and large relative displacements to be generated in a reproducible way. For example, microvalves make use of the collapsed state to block off fluid flow in microchannels [7], and microswitches harness pull-in to rapidly switch between two remote configurations, corresponding to distinct 'off' and 'on' states [8, 9, 10]. In these applications, an understanding of the pull-in dynamics is essential since it determines the switching time of the device. The time taken to pull-in can itself be used as a sensing mechanism: the relationship between the pull-in time of a microbeam and the ambient air pressure has been proposed as a pressure sensor [11], while high-resolution accelerometers make use of the sensitivity of parallel-plate actuators to external acceleration [12, 13, 14]. In these applications, unlike microswitches and other actuators, it is not desirable to simply minimize the pull-in time. Instead, the device is operated at voltages very close to the pull-in transition, where the transient dynamics are observed to slow this slowing down down considerably. is highly sensitive to ambient conditions, including external forces, and so has widespread potential to Crucially, enable high-resolution, low-noise measurements to be made. Using pull-in time as a sensing mechanism also offers the advantage that the device may be integrated in standard circuit technology, so that commercially available micromachining processes can be used [13]. The slowing down observed in parallel-plate actuators has been attributed to a 'bottleneck' or 'meta-stable' phase that dominates the dynamics during pull-in, characterized by a temporary balance between electrostatic and mechanical restoring forces [12]: as the net force on the structure is very small, it evolves slowly and the pull-in time is large. However, a quantitative understanding of this bottleneck phenomenon is still lacking, despite the obvious importance of this regime in the operation of many MEMS devices. it is not clear how the duration of the bottleneck (and hence the pull-in time) scales with the applied voltage, the external acceleration, and the material parameters of the system. In particular, 1.1. Models of pull-in dynamics It is well known that pull-in is initiated by a saddle- node (fold) bifurcation: the equilibrium state away from collapse ceases to exist above a critical voltage (without first becoming unstable), so the system must pull-in to remain in equilibrium. This means that a standard linear stability analysis cannot be used to study the transient dynamics - there is no unstable equilibrium base state from which the system evolves. Most studies therefore adopt a purely numerical or experimental approach. For devices operating in atmospheric conditions, fluid damping (arising in the squeeze film when the air gap between components becomes very small) has been identified as playing a dominant role [15]. A large number of studies have therefore focussed on generat- ing macromodels, i.e. reduced-order models that cou- ple deformations of the structure to realistic models of the squeeze film damping, including compressibil- ity and rarefaction effects. These macromodels are then used to reduce the computational cost of simu- lating MEMS devices during pull-in (see [1] and [16] and references therein). Bottleneck phenomena have also been described in a number of macromodel simu- lations of microbeam actuators [17, 18, 19, 20]; slowing down appears to be a generic feature of the dynamics when the system is operated near the pull-in transi- Delayed pull-in transitions in overdamped MEMS devices 3 tion, though this has not been explored further. More recent studies instead address the dynamic stability of MEMS resonators under a combination of AC and DC loads [20, 21], the effects of geometric nonlinearities due to large displacements [22], contact bouncing [9, 23], and modelling structures that possess natural curva- ture where snap-through buckling can occur alongside pull-in [2, 24, 25]; for a review see [4]. Few analytical results concerning pull-in dynamics are available. While general bounds on the pull-in time have been obtained [26], these bounds are not very tight and do not give insight into possible slowing down behaviour close to pull-in. In the case of underdamped, inertia-driven systems, some progress has been made. For devices operating at very low ambient pressures, inertial effects can cause the critical voltage at pull- in (the dynamic pull-in voltage) to decrease compared to that obtained when the voltage is quasi-statically varied (the static pull-in voltage) [27]. Using energy methods, scaling laws for the pull-in time have been derived for parallel-plate actuators [28] and extracted for more complex devices such as microbeams using lumped-parameter models [29]. The key result is that the pull-in time, tPI, scales logarithmically with the difference between the applied voltage and the pull-in voltage, ∆V > 0: we have that tPI ∝ log(1/∆V ), (1) as ∆V → 0. As ∆V decreases, the pull-in time therefore increases rapidly, until eventually mechanical noise limits the response. Due to its simplicity, the scaling law (1) offers a useful design rule to tune dynamic response in applications: only a small number of runs are needed to extract the appropriate pre-factor in the scaling law to make further predictions. The need to perform parameter sweeps that at each stage involve detailed simulations can then be eliminated. However, no corresponding scaling law has been found for overdamped systems, despite the fact that (i) many MEMS devices operate in this regime [12] and (ii) there is a clear need for such a design rule as devices continue to scale down and grow in complexity [30]. Obtaining such a scaling law analytically is the primary objective of this paper. 1.2. A scaling law for overdamped pull-in To explore the possibility of a scaling law analogous to (1) but applicable to overdamped devices, we have assembled a large range of data for pull-in times reported in the literature. We focus on results for devices operating at (or near) atmospheric pressure only; we do not consider data for pull-in times in vacuum where inertial effects are important. We consider parallel-plate and microbeam devices, from both experiments and incorporating results This includes data where dynamic simulations. the actuation voltage is varying while the external acceleration is zero, as well as data from pull-in time accelerometers where the actuation voltage is fixed but the external acceleration is varied. A summary of the conditions for each data set are provided later in tables 1–2. In all cases examined, the pull-in times are measured from the point of application of a step DC voltage (stepped from zero). Where data is only available graphically, we have extracted the values using the WebPlotDigitizer (arohatgi.info/WebPlotDigitizer). For each measurement, we use the reported values of the pull-in voltage to compute the normalized distance to the pull-in transition, which we denote by . In particular, in the case of zero external acceleration = (V /VSPI)2 − 1 with V the applied voltage and VSPI the static pull-in voltage. The results are shown on logarithmic axes in figure 1, where different symbols are used to indicate different data sets (i.e. where the properties of the actuator are varied), and data from different references are distinguished using different colours. We observe that the pull-in time increases as decreases in a systematic way. Very close to the pull- in transition, the dynamics become highly sensitive to the precise value of : the pull-in time may increase by over an order of magnitude within a very narrow range of . This is the bottleneck regime in which the dynamics of pull-in are dramatically slowed down. systems, This delay behaviour is reminiscent of the critical slowing down observed near saddle-node bifurcations in a range of physical such as elastic snap-through [31], phase transitions [32], and the switching of charge density waves [33]. In these systems, the remnant or 'ghost' of the saddle-node bifurcation continues to attract trajectories that are nearby in parameter space, producing a bottleneck whose duration generically increases with decreasing distance from the bifurcation [34]. The detailed scaling of the duration of this bottleneck phase depends on the importance of inertia: a scaling ∝ −1/2 is characteristic of overdamped motion [34] while the scaling ∝ −1/4 is characteristic of underdamped motion [31]. The key observation here is that most of the data in figure 1 appears to be consistent with the same scaling law, namely tPI ∝ −1/2 as → 0. More precisely, we have fitted each data set (using least- squares) to a power law of the form tPI = α−β where β > 0; over all 27 data sets considered in figure 1, we find a mean value β ≈ 0.56 with a standard deviation of 0.14 in the fitted values. While an −1/2 scaling law has been identified as the source of slow dynamics in microbeam resonators [21], we believe this has not Delayed pull-in transitions in overdamped MEMS devices 4 Figure 1. Pull-in times of parallel-plate and microbeam devices under step DC loads reported in the literature. In total, 27 sets of data from 9 different references are included, indicated by different symbol shape and colour (details and legend are provided in tables 1–2). yet been properly appreciated as a generic feature of overdamped pull-in under DC loads. The slowing down observed in figure 1 motivates a more careful analysis of the dynamics of overdamped pull-in. In this paper we focus on the simplest possible electrostatic device: a parallel-plate actuator under a DC load. This single degree-of-freedom structure captures the balance between electrostatic and mechanical restoring forces that underlies the pull-in instability, without requiring details of the geometry of the device. It has been successfully used as a lumped-parameter model for more complex structures such as microbeams and microplates [35]. Our analysis of the parallel-plate actuator therefore allows us to consider a generic MEMS device, upon taking appropriate values of the lumped parameters. Our central result is that the bottleneck behaviour observed near the pull-in transition is a type of saddle- node ghost, and so inherits the expected scaling law [34], with the pull-in time tPI ∝ −1/2 as → 0. While some data sets in figure 1 do not appear to follow this scaling, we suggest that the discrepancy is due to sensitivity to the precise value of the reported pull- in voltage, and propose a method to obtain a more accurate value based only on measured pull-in times. The remainder of this paper is organized as follows. We begin in §2 by describing the equations governing the motion of the parallel-plate actuator. In §3, we solve the equations numerically when the system is perturbed just beyond the static pull-in transition. In the overdamped limit, we recover the bottleneck phenomenon reported previously [12]. We then perform a detailed asymptotic analysis of the solution structure in this regime, allowing us to derive an approximate expression for the pull-in time. In §4, we compare our asymptotic prediction to the experimental and numerical data given in figure 1. We show that the observed slowing down is well explained by our scaling law, and use our theory to collapse the data presented in figure 1 onto a master curve (see figure 6). Finally, in §5, we summarize and conclude our findings. 2. Theoretical formulation 2.1. Governing equations We wish to understand the bottleneck dynamics of a generic MEMS device, when the voltage is near the static pull-in transition. As the bottleneck is characterized by slow motions, and occurs well before the device comes into close contact with the actuating electrode [12], we neglect compressibility and rarefaction effects in the squeeze film - the fluid damping is assumed to be purely viscous [36]. This is justified by numerical simulations [15] that show compressibility has very little effect on the pull-in time very close to the transition. Moreover, we assume a constant damping coefficient, denoted by b, taken to be the effective value of the damping coefficient in the 10-410-310-210-110010110-610-510-410-310-210-1Delayed pull-in transitions in overdamped MEMS devices 5 whole device, which we assume is constant. This approximation is valid provided that aext varies over a timescale much longer than the timescale of pull-in (typically 1 − 10 ms). Under these assumptions, the displacement of the moving plate, x(t), obeys the equation of motion m d2x dt2 + b dx dt + kx = 1 2 0AV 2 (d0 − x)2 + maext. (2) Here the first term on the right-hand side is the electrostatic force in the parallel-plate approximation (0 is the permittivity of air), neglecting corrections due to fringing fields [3]. As initial conditions, we consider the case of a suddenly applied (step function) voltage with the plate initially at rest at the zero voltage state, i.e. x(0) = maext/k and x(0) = 0 (here and throughout denotes d/dt). These initial conditions are commonly used in applications of pull- in time in pressure sensors and accelerometers [37, 12, 13, 14]. 2.2. Non-dimensionalization To make the problem dimensionless, we note that a balance between viscous and spring forces in equation (2) leads to the timescale [t] = b/k. It is natural to scale the displacement away from the zero voltage state with the maximum allowed displacement before contact occurs. This motivates introducing the dimensionless variables T = t [t] , X = x − maext/k d0 − maext/k , Aext = maext kd0 . Equation (2) can then be written as Q2 d2X dT 2 + where Q = introduced the normalized voltage dX dT √ + X = λ (1 − X)2 , mk/b is the quality factor, and we have (3) λ = 1 2 0AV 2 0(1 − Aext)3 . kd3 The initial conditions become X(0) = X(0) = 0, (4) (5) and we note that, with this non-dimensionalization, contact between the electrodes occurs at X = 1, with physical solutions requiring X < 1. The dimensionless parameter λ is the key control parameter and may be interpreted as the ratio of the typical electrostatic force (∼ 0AV 2/[2d2 0]) to the spring force (∼ kd0) [3], together with an additional factor that depends on the acceleration of the device. For realistic MEMS devices we have Aext (cid:28) 1, owing mainly to the small value of the mass m; for example, in the accelerometer designed by [12], the range of accelerations encountered is aext ≤ Figure 2. Schematic of the mass-spring parallel-plate capacitor. Fluid between the plates is represented by a linear dashpot of damping coefficient b. bottleneck. Here b is regarded as a lumped parameter that characterizes the properties of the squeeze film, including the thickness of the air gap, the ambient pressure, the fluid viscosity, and finite-width (border) effects, as well as any additional material damping that may be present. As the geometry of the device is also slowly varying when its motions are slow, the elastic restoring forces can be approximated to leading order as a linear spring with constant stiffness [9]. We denote the effective spring constant by k, which combines properties of the mechanical restoring force, such as the dimensions of the actuator, the material stiffness and any residual stress built into the elastic structure. In addition, we use a parallel-plate approximation of the electrostatic force; this is valid provided the aspect ratio of the air gap and the slopes of deformation are small [1]. In this way, our model becomes a single- degree-of-freedom mass-spring model [3]. Physically, it is equivalent to a parallel-plate capacitor, in which one plate is fixed while the other is attached to a linear spring and damper; see figure 2. The assumptions made above are not valid outside of the bottleneck phase, where the speed of the device is increased and the details of its geometry may become important. However, by choosing suitable values for the lumped parameters b and k, we expect to correctly account for the length of the slow phase and hence approximate the total pull-in time, which is dominated by the time spent passing through the bottleneck. In particular, we discuss how a variable damping coefficient should be accounted for at the end of §3 and in Appendix A. As shown in figure 2, the properties of the moving plate are its mass m, area A and displacement x. The applied DC voltage is V , and d0 is the gap thickness in the absence of any displacement (x = 0). We also account for an external acceleration aext of the Delayed pull-in transitions in overdamped MEMS devices 80 mg, corresponding to Aext = O(10−3) for their experimental parameters. We therefore consider only the case λ ≥ 0 here. 2.3. Steady solutions The behaviour of the steady-state solutions of equation (3) is well-known (see [3], for example). Here we merely summarize the main results. For 0 ≤ λ < λfold = 4/27, there are two real solutions with 0 < X < 1, one of which is linearly stable and the other linearly unstable. At λ = λfold, these two solutions coincide and disappear at a saddle-node (fold) bifurcation with X = Xfold = 1/3. For λ > λfold, no real solutions exist. This is illustrated by the response diagram shown in figure 3a. Under quasi-static conditions, the fold point cor- responds to where pull-in is observed experimentally (no equilibrium solution away from contact exists for λ > λfold), giving the static pull-in voltage and pull-in displacement in terms of the external acceleration as VPI = 8kd3 0(1 − Aext)3 270A , xPI = d0 3 (1 + 2Aext) . (cid:115) In the case of zero external acceleration, this reduces to the classic (static) pull-in voltage and pull-in displacement of a parallel-plate capacitor, widely reported in the literature [4]; we label these as VSPI and xSPI respectively. 3. Pull-in dynamics We now consider the case when the system is perturbed just beyond the static pull-in transition, i.e. we set λ = λfold(1 + ), where 0 < (cid:28) 1 is a small parameter capturing the distance beyond the pull-in transition ( is shown schematically in figure 3a). Using the expression (4) and the fact that 0AV 2 kd3 0 λfold = 1 2 SPI , we may write as = − 1 = λ λfold (V /VSPI)2 (1 − Aext)3 − 1. (6) We see that for a fixed actuation voltage V > VSPI, the external acceleration changes the effective perturbation , with increasing as Aext increases. The result will be an associated change in the pull-in time. This is the basis on which pull-in time accelerometers operate: by repeatedly inducing pull-in and measuring the resulting pull-in times, the external acceleration can be determined after a suitable calibration is performed [13]. In practice, pull-in times can be measured 6 extremely accurately and with low noise by sensing large changes in capacitance using a high frequency clock. The sensitivity of the pull-in time to changes in is therefore the primary factor that limits the sensitivity of the accelerometer. The key observation, first confirmed in figure 3b, which displays reported by [12], is that for quality factors Q smaller than unity (i.e. overdamped devices) the motion of the plate is slowed in a bottleneck as it passes the static pull-in displacement, Xfold = 1/3. This behaviour is the dimensionless trajectories X(T ) during pull-in for different values of Q. We have obtained these trajectories by integrating equation (3) numerically with initial conditions (5) in matlab. As the ODE is singular in the limit Q → 0 (we lose the second-order derivative needed to satisfy the initial conditions), we use the matlab routine ode15s, which employs a stiff solver to capture transients in which the inertia of the plate cannot be neglected. We see from figure 3b that for Q (cid:28) 1, the bottleneck phase dominates the transient dynamics, and hence the total time taken to pull-in. The phase becomes highly dependent on the damping as Q is increased past unity, with virtually no bottleneck present for Q ≥ 2. The duration of the bottleneck is also sensitive to the perturbation , and appears to increase without bound as → 0. We now perform a detailed asymptotic analysis of equation (3) in the limit Q (cid:28) 1, showing that the bottleneck phenomenon is an instance of a saddle-node ghost [34] whose duration scales as −1/2 as → 0. 3.1. Solution structure for Q (cid:28) 1 We begin by considering the different leading order balances the solution passes through during pull- in. This analysis will confirm that the bottleneck phase does indeed dominate the pull-in dynamics, as expected from figure 3b: the bottleneck duration is much longer than any other timescale in the problem, including any intervals for which plate inertia is important. This will enable us to approximate the total pull-in time based on the duration of the bottleneck alone. Early times: At early times, 3.1.1. the initial conditions (5) imply that the displacement X is small. Linearizing equation (3) then gives Q2 d2X + X ∼ λ(1 + 2X). dT 2 + The solution satisfying X(0) = X(0) = 0 is X = λ 1 − 2λ 1 + α− α+ − α− eα+T − α+ α+ − α− eα−T dX dT (cid:18) (cid:19) , (7) Delayed pull-in transitions in overdamped MEMS devices 7 Figure 3. (a) Response diagram for the steady-state solutions of (3) (blue curves), which satisfy X = λ/(1−X)2, as the dimensionless voltage λ varies. At λ = 4/27 the stable equilibrium away from pull-in (lower solid curve) intersects an unstable solution (dashed curve) and disappears at a fold: a saddle-node bifurcation. A typical trajectory at fixed λ beyond the pull-in transition is also shown (red arrow). (b) Dimensionless trajectories X(T ) satisfying (3) and (5) for = 10−3 and different quality factor Q (coloured curves; see legend). For later comparison, the asymptotic trajectory predicted by (13) is also shown (black dotted curve). where −1 ±(cid:112)1 − 4Q2(1 − 2λ) 2Q2 α± = . When Q (cid:28) 1, we expand to find α+ = −(1 − 2λ) + O(Q2), α− = − 1 Q2 + O(1). This shows that inertia may only be neglected for T (cid:29) Q2, when eα−T is exponentially small and the leading order terms in (7) become independent of Q. In this case the solution simplifies to (cid:104) 1 − e−(1−2λ)T(cid:105) . X = λ 1 − 2λ It follows that the terms we neglected in linearizing equation (3), of size O(X 2), only remain small provided T (cid:28) 1 (as λ ≈ λfold = 4/27 is order unity). As T reaches O(1), this solution therefore breaks down and a different leading order balance emerges. Later times, T (cid:38) 1: Using the previous 3.1.2. solution to evaluate the size of terms for T = O(1) yields the updated balance dX dT (1 − X)2 , + X ∼ (8) λ with inertia now negligible. This equation can be solved to give the displacement implicitly in terms of time (e.g. [11]): (cid:90) X (1 − ξ)2 0 λ − ξ(1 − ξ)2 dξ T = (9) (Here matching into T (cid:28) 1 requires the constant of integration to be zero.) This solution is not uniformly valid during pull-in: close to contact the electrostatic force will grow very large, leading to fast motions where inertia becomes important again. We can use equation (8) directly to determine when this first occurs. Differentiating, we obtain d2X dT 2 + The ratio of the neglected inertia term to the damping term can then be evaluated as Q2 X X (1 − X)3 − 1 (1 − X)3 ∼ Q2 dX dT dX dT (cid:20) (cid:21) ∼ . . 2λ 2λ Away from X = 1, the term in square brackets is O(1) and so inertia is unimportant when Q (cid:28) 1. This first breaks down when X = 1−O(Q2/3), at which point we have X = O(Q−4/3) (using (8)) and Q2 X = O(Q−4/3). Note that these updated scalings must hold close to the pull-in time, which we denote T = TPI, as X is close to 1. Setting T = TPI − O(Qγ) and seeking a balance i.e. these between these terms shows that γ = 2, scalings hold inside the interval T = TPI − O(Q2). In summary, for Q (cid:28) 1 we have shown that inertia of the plate remains negligible for Q2 (cid:28) T (cid:28) TPI − O(Q2), Q2 (cid:28) X (cid:28) 1 − O(Q2/3). In particular, we conclude that the dynamics are first order when X passes the static pull-in displacement Xfold = 1/3. Because λ is close to its value at the fold, where the spring force exactly balances the electrostatic force, it follows that the difference between these two forces will be very small around Xfold. This explains the previous observation that 00.20.40.60.8102040608010000.20.40.60.8100.10.20.3Delayed pull-in transitions in overdamped MEMS devices 8 the bottleneck is a type of meta-stable interval characterized by a balance of forces [12]. In fact, when X = Xfold we have dX dT (cid:21)(cid:12)(cid:12)(cid:12)(cid:12)X=Xfold (cid:12)(cid:12)(cid:12)(cid:12)X=Xfold (1 − X)2 − X (cid:20) = 3 . = λ As the velocity is very small but non-zero in the bottleneck, the system appears to 'feel' the attraction of the equilibrium point at the pull-in transition; however, this delay is purely a remnant of the saddle- node bifurcation, and relies on no extra physics in the system. Note that for larger quality factors, Q = O(1), this conclusion is not valid: as the dynamics are no longer first order, a small net force does not imply slow dynamics. The inertia of the plate 'carries' it through the bottleneck without significant slow down, as is evident from the trajectories in figure 3b for Q ≥ 2. It can also be observed that the pull-in time does not simply decrease monotonically in this regime as Q is increased (e.g. the pull-in for Q = 2 is faster than that for Q = 5 in figure 3b): while high inertia carries the plate quickly through the bottleneck, it also slows down the initial dynamics, as the plate must be accelerated from its rest position. Bottleneck analysis: We now consider the 3.1.3. solution inside the bottleneck phase. While we can make progress using the implicit solution (9), we instead analyse equation (3) directly. The method we present is more general as it can be applied to systems for which no analytical solution is available. When the solution is close to the static pull-in displacement we have 1 + X(T ) , X = Xfold where X (cid:28) 1. Using λ = λfold(1+), the electrostatic force can then be expanded as (cid:104) (cid:105) (cid:20) (cid:21) X 2 3 4 1 + + X + + O( X, X 3). (8) and neglecting terms of λ (1 − X)2 = Xfold Substituting into O( X, X 3), we obtain d X dT 3 4 X 2. ∼ + (10) Equation (10) is valid in the regime (cid:28) X (cid:28) 1, i.e. the neglected terms of O( X, X 3) are smaller than the retained terms. In particular, we note the importance of retaining the quadratic term. This term is neglected in the approach taken by [38]; an analysis of their solution shows that it incorrectly predicts the pull-in time scales as −1 as → 0. pull-in transition: the first term on the right-hand side is the normalized perturbation to the bifurcation parameter (either due to a change in voltage or external acceleration), and the quadratic term is the nonlinearity that characterizes the bifurcation as being of saddle-node type (locally parabolic near the fold). In this way, equation (10) is generic for the dynamics of overdamped MEMS devices close to the pull-in transition. Similar evolution equations have been obtained using a single degree-of-freedom approximation for a microbeam [39], and in a MEMS resonator modelled as a Duffing-like oscillator [21]. However, our approach here offers new insight into why this equation should apply more generally. The solution of (10) is √ X ∼ 2 3 3 tan (cid:20) 1 2 √ 3(T − T0) , (11) (cid:21) for some constant T0. In the immediate vicinity of the static pull-in displacement, where X (cid:28) 1/2, the term in square brackets in (11) is much smaller than unity. Here the solution simplifies to X ∼ (T − T0), so that the displacement evolves linearly in time in the middle of the bottleneck. Outside of this interval, the tangent function captures how the plate begins to accelerate away from the static pull-in displacement. We note that as this linear behaviour is precisely the solution of equation (10) upon neglecting the quadratic term in favour of the term in , we deduce that (11) is asymptotically valid for all X (cid:28) 1 (rather than just (cid:28) X (cid:28) 1). The solution (11) appears to undergo finite-time blow-up as the term in square brackets approaches ±π/2. However, as soon as X grows comparable to O(1), our original assumption X (cid:28) 1 is no longer valid and the solution breaks down. In terms of the diagram in figure 3a, this means that the displacement has left the vicinity of the fold point and a local analysis can no longer be applied. Upon making use of the expansion tan x ∼ ±(π/2 ∓ x)−1 as x → ±π/2, it follows that the solution accelerates according to the power law √ X ∼ π/ ±4/3 3 ∓ (T − T0) . Here the minus sign corresponds to initially entering the bottleneck ( X < 0), while the plus sign corresponds to leaving the bottleneck towards pull-in ( X > 0). We deduce that X = O(1) when T − T0 ∼ ± π√ 3 + O(1). Up to numerical pre-factors, equation (10) is the normal form for a saddle-node bifurcation [34]. This reflects the bifurcation structure underlying the The duration of the bottleneck, denoted Tbot, is simply √ the difference between these two values and so we have Tbot = 2π/ 3 + O(1). Delayed pull-in transitions in overdamped MEMS devices 9 The bottleneck dominates the time spent in the regime where the dynamics are first order. Moreover, we showed that inertia is only important in intervals of duration O(Q2) ((cid:28) 1) around T = 0 and T = TPI. It follows that the total pull-in time is equal to the bottleneck time to leading-order: TPI = 2π√ 3 + O(1). (12) To validate the prediction (12), we numerically determine the pull-in time by integrating the full ODE (3) with initial conditions (5). As equation (3) is singular at X = 1, we use event location to stop integration as soon as (1 − X) < tol for some tolerance tol, and the corresponding time at this point then gives the pull-in time. The accuracy of this method can be justified by analysing the behaviour of (3) very close to pull-in, where a power law solution can be extracted; we use tol = 10−5, which guarantees an accuracy of O(10−6) in the computed pull-in time when we restrict to Q ≤ 10. In figure 4a we plot the computed times as a function of the normalized perturbation . We conclude that the asymptotic prediction (12) approximates the pull- in time extremely well for moderately small quality factors Q (cid:46) 1 and perturbations (cid:46) 10−1. Figure 4b shows a surface plot of the computed pull-in times for a range of values of and Q. As well as showing that the dynamics become very slow as → 0 with Q fixed, we observe that, with fixed, the dependence of the pull-in time on the quality factor Q is non-monotonic. In particular, when we fix (cid:46) 10−2, is obtained at Q ≈ 2; within a minimum in TPI a narrow range of Q close to this value, TPI varies significantly. While this minimum may seem surprising at first, it is the result of inertia being small enough for the plate to be rapidly accelerated from its rest position but large enough that it passes the pull-in displacement without significant slowing down in a bottleneck. If we imagine fixing the actuation voltage V near VSPI and varying the plate mass m, so that Q is varied while all other parameters are fixed, then this corresponds to a value of m that minimizes the pull-in time. This may be relevant to switching applications where the pull- in time needs to be minimized without increasing the voltage significantly [35] (since increasing the voltage would increase the total energy consumed). Currently, the constant T0 appearing in the bottleneck solution (11) remains undetermined. This corresponds to the time at which X = 0 (when the displacement is equal to the static pull-in displacement Xfold). However, we can find the value of T0 by a symmetry argument. From the solution (11), we see that the displacement about the static pull-in displacement is antisymmetric, i.e. we have X → − X as (T − T0) → −(T − T0). (This is a consequence of the dynamics being first order, and the symmetry of the quadratic nonlinearity in equation (10).) As the bottleneck phase dominates the entire motion in the limit (cid:28) 1, it follows that, to leading order in √ , the value of T0 is simply half of the bottleneck time: T0 ∼ π/ 3. The rescaled displacement in the bottleneck, (11), can then be written as √ X ∼ 2 (cid:34)√ (cid:35) tan . 3 3 3 2 T − π 2 (cid:34)√ (cid:35) The unscaled displacement, X, then becomes X ∼ 1 3 T − π 2 √ 2 3 9 3 2 tan + . (13) integration of the full system; to the trajectories obtained This compares well by numerical see figure 3b, where the analytical prediction is almost indistinguishable from numerical results with Q (cid:28) 1. As the motions are so fast outside the bottleneck, we see that (13) also provides a good description of the global dynamics (restricting X to the interval [0, 1]), despite the fact that the assumptions made in deriving (13) are only strictly valid in the bottleneck phase. We note that some caution is needed when using a constant damping coefficient, as in our approach here: in reality the damping coefficient may itself depend on the current gap thickness. Indeed, simulations that use a constant damping coefficient corresponding to the initial gap thickness have been shown to give large errors [12]. However, using the damping coefficient appropriate in the bottleneck phase of the motion correctly accounts for the duration of the bottleneck, and hence provides a good approximation of the total time taken to pull-in (see Appendix A). = where (V /VSPI)2 (1 − Aext)3 − 1. 4. Data comparison In §3 we derived a scaling law for the slowing down of a parallel-plate actuator close to the pull-in transition. In dimensional form, this predicts that the pull-in time increases as 2π√ tPI ∼ b (14) k 3 This result is valid for 0 < (cid:28) 1 and small quality factor, Q (cid:28) 1. As discussed at the start of §2, we expect that this result also describes the dynamics of a generic MEMS device operating in overdamped conditions; here we regard the damping coefficient b and spring constant k as lumped parameters that encapsulate the properties of the squeeze film and the mechanical restoring force during the bottleneck phase, respectively. We now compare our prediction to pull-in data reported in the literature, both from experiments and Delayed pull-in transitions in overdamped MEMS devices 10 √ Figure 4. Pull-in times TPI determined from the numerical solution of (3) with initial conditions (5). (a) Numerical results for fixed Q and variable (symbols, see legend), together with the asymptotic prediction TPI ∼ 2π/ 3 valid for (cid:28) 1 and Q (cid:28) 1 (dotted line), and the prediction TPI ∼ 9/(4) valid for (cid:29) 1 and Q (cid:28) 1 (dashed-dotted line) [11]. (b) Surface plot of TPI as a function of and Q. Also shown are slices through the surface at values ∈ {10−3, 5 × 10−3, 10−2, 5 × 10−2, 10−1} (red dotted curves). numerical simulations. The details of each data set are summarized in table 1 for parallel-plate devices, and in table 2 for microbeam devices. These provide the relevant parameter values in each study, and the type of model used (for numerical simulations). We have separated the data so that only the actuation voltage or the acceleration is varying within each data set, corresponding to a particular row in the tables. Where the properties of the actuator or the squeeze film have changed within a single reference, the data have therefore been separated into different rows of the tables. For data on parallel-plate actuators (table 1), the relevant parameters are the ratio of the actuation voltage to the pull-in voltage V /VSPI, external acceleration aext, pull-in voltage VSPI, initial gap thickness d0, plate mass m, and the damping coefficient in the bottleneck phase, b. For the data on microbeams (table 2), the parameters are the pull-in voltage VSPI, initial gap thickness d0, beam length L, beam thickness h, beam width w, and Young's modulus E. (In both tables, blank entries indicate that no value is provided in the reference.) A wide range of values are exhibited in these parameters across the studies. We also report any additional effects that may be present in experiments and simulations; these include residual stress, rarefaction effects, partial field screening, varying ambient pressures, and different boundary conditions for microbeams. The reported pull-in times were shown on logarithmic axes in figure 1 (as a function of the corresponding values of ). As well as generally confirming the expected scaling law that tPI ∝ −1/2 as → 0, we see that this rescaling collapses data from accelerometers, where the acceleration is variable and the actuation voltage varies between each data set (all other parameters fixed); see the data of [12], orange symbols. We also see a collapse in data over a single experimental system where the actuation mechanism changes, between varying the voltage (with zero acceleration) or varying the acceleration (with fixed voltage); see the data of [13], magenta symbols. This verifies that is the correct dimensionless parameter to capture both variations in the voltage and external acceleration near the pull-in transition. Some data sets plotted in figure 1 do not appear to follow the expected −1/2 scaling, curving downward slightly for small . These include the experiments/simulations of [17] (blue symbols) and the experiments of [14] (cyan symbols). However, we propose that this is due to sensitivity to the reported value of the pull-in voltage: a small error introduces shifts in the computed values of , which can cause large variations when plotted on logarithmic axes. Another way to test the scaling law, which eliminates this sensitivity, is to plot t−2 PI as a function of voltage/external acceleration on linear axes. This is shown in figure 5a for the data of [17] (blue symbols), and in figure 5b for all accelerometer data. (Due to the large range of pull-in times under varying voltage, figure 5a shows only a subset of data, for clarity.) In all cases a linear relationship is observed close to the pull- PI → 0. A linear relationship in transition, i.e. as t−2 here implies the expected −1/2 scaling, because is linear in the voltage/acceleration close to the pull-in transition. For example, in the case of zero external 10-310-210-110010110210310-210-1100101102Delayed pull-in transitions in overdamped MEMS devices 11 d n e g e L Q d e t t i F b d e t t i F ) 1 − s g ( 3 1 8 0 . 0 8 6 6 0 . 0 6 9 2 . 0 8 0 4 . 0 8 0 4 . 0 8 0 4 . 0 8 0 4 . 0 8 0 4 . 0 8 0 4 . 0 1 5 6 . 0 1 5 6 . 0 1 5 6 . 0 1 5 6 . 0 5 7 3 . 0 5 7 3 . 0 0 3 3 . 0 6 9 2 . 0 4 . 1 2 2 8 1 . 0 2 8 1 . 0 2 8 1 . 0 2 8 1 . 0 2 8 1 . 0 2 8 1 . 0 0 4 . 1 0 4 . 1 0 4 . 1 0 4 . 1 1 9 . 2 1 9 . 2 l a n o i t i d d A s t c e ff e l e d o M b C B S D F Q S C , S M D L , S M D L , S M C B S C B S C B S D F Q S C , S M D F Q S C , S M D F Q S C , S M A / N A / N E B , C B S D F Q S C , S M E B , C B S D F Q S C , S M A / N A / N A / N A / N A / N d e t r o p e R ) 1 − s g ( 5 3 . 0 5 3 . 0 2 9 1 . 0 2 9 1 . 0 2 9 1 . 0 2 9 1 . 0 2 9 1 . 0 2 9 1 . 0 ∗ 8 8 . 2 ∗ 8 8 . 2 8 8 . 2 8 8 . 2 0 8 . 4 0 8 . 4 I P S V ) V ( ) 1 − m N ( k 6 7 . 8 4 5 . 5 3 . 2 7 1 3 9 . 2 1 3 9 . 2 6 1 9 . 2 6 1 9 . 2 4 9 . 2 4 9 . 2 3 3 . 6 4 9 . 2 0 4 0 3 9 2 . 1 0 3 9 2 . 1 0 3 9 2 . 1 0 3 9 2 . 1 0 3 9 2 . 1 0 3 9 2 . 1 ∗ 3 3 . 3 ∗ 3 3 . 3 3 3 . 3 3 3 . 3 0 4 . 2 0 4 . 2 ) g µ ( m 4 1 1 . 0 2 3 1 . 0 0 0 0 1 7 2 . 4 7 2 . 4 7 2 . 4 7 2 . 4 7 2 . 4 7 2 . 4 ∗ 9 4 2 ∗ 9 4 2 9 4 2 9 4 2 ∗ ∗ 7 9 4 7 9 4 ) m µ ( 0 d 7 0 . 2 7 0 . 2 5 2 . e r u t a r e t i l e h t n i d e t r o p e r s r o t a u t c a e t a l p - l e l l a r a p f o s e m i t n i - l l u p e h t f o y r a m m u S . 1 e l b a T t x e a I P S V / V e p y t a t a D . f e R 5 2 . 2 e l b a i r a V 3 0 0 0 . 1 t n e m i r e p x E 5 2 . 2 e l b a i r a V 5 0 0 0 . 1 t n e m i r e p x E ] 2 1 [ 5 2 . 2 e l b a i r a V 9 0 0 0 . 1 t n e m i r e p x E 5 2 . 2 e l b a i r a V 3 0 0 0 . 1 n o i t a l u m S i 5 2 . 2 e l b a i r a V 5 0 0 0 . 1 n o i t a l u m S i ∗ 5 2 . 2 e l b a i r a V 1 0 . 1 t n e m i r e p x E 5 2 . 2 e l b a i r a V 9 0 0 0 . 1 n o i t a l u m S i ∗ 5 2 . 2 0 e l b a i r a V t n e m i r e p x E 5 2 . 2 e l b a i r a V 1 0 . 1 n o i t a l u m S i 5 2 . 2 0 e l b a i r a V n o i t a l u m S i 0 0 . 3 e l b a i r a V 5 0 . 1 t n e m i r e p x E ] 2 1 [ ] 2 1 [ ] 2 1 [ ] 2 1 [ ] 3 1 [ ] 3 1 [ ] 3 1 [ ] 3 1 [ ] 4 1 [ 0 0 . 3 0 e l b a i r a V t n e m i r e p x E ] 4 1 [ 0 0 0 e l b a i r a V n o i t a l u m S i e l b a i r a V n o i t a l u m S i e l b a i r a V n o i t a l u m S i ] 7 3 [ † ] 7 3 [ ] 0 4 [ ] 2 1 [ s n o i t c e r r o c , C B S ; ) n o i t a u q e s d l o n y e R ( i g n p m a d m fi l e z e e u q s e l b i s s e r p m o c , D F Q S C ; t n e i c ffi e o c t n a t s n o c h t i w g n p m a d i r a e n i l , D L ; l e d o m g n i r p s - s s a m m o d e e r f - f o - e e r g e d e l g n i s , S M e l b a c i l p p a t o n , A / N ; s t c e ff e i ) h t d w - e t i n fi ( r e d r o b , E B ; s s e r t s l a u d i s e r , S R ; ) s t c e ff e n o i t c a f e r a r ( s n o i t i d n o c y r a d n u o b p i l s o t e u d y g o l a n a c i t e n g a m o r t c e l e - c i t a t s o r t c e l e n i e g a t l o v C D t n e l a v i u q e o t d e t r e v n o c n e e b s a h a t a d ; e c r o f c i t e n g a m o r t c e l E d e t a m i t s e / g n i l l e d o m y b d e n i a t b o g n i e b s a d e t r o p e r r e t e m a r a p l a t n e m i r e p x E ∗ † Delayed pull-in transitions in overdamped MEMS devices 12 t n a t s n o c h t i w g n p m a d i r a e n i l , D L ; s n o i t i d n o c y r a d n u o b p u - d e p p e t s , E S ; s n o i t i d n o c y r a d n u o b d e p m a l c - d e p m a l c , C C ; n o i t a u q e e t a l p ) D 2 ( c i m a n y d , E P ; n o i t a u q e m a e b c i m a n y d , E B l a i t r a p , S F ; s s e r t s l a u d i s e r , S R ; ) s t c e ff e n o i t c a f e r a r ( s n o i t i d n o c y r a d n u o b p i l s o t e u d s n o i t c e r r o c , C B S ; ) n o i t a u q e s d l o n y e R ( i g n p m a d m fi l e z e e u q s e l b i s s e r p m o c , D F Q S C ; t n e i c ffi e o c d e t a m i t s e / g n i l l e d o m y b d e n i a t b o g n i e b s a d e t r o p e r r e t e m a r a p l a t n e m i r e p x E e r u s s e r p t n e i b m A ∗ ‡ e l b a c i l p p a t o n , A / N i ; g n n e ff i t s - n i a r t s , S S i ; g n n e e r c s d l e fi d n e g e L ] t [ d e t t i F s t c e ff e l a n o i t i d d A . e r u t a r e t i l e h t n i d e t r o p e r s r o t a u t c a m a e b o r c i m f o s e m i t n i - l l u p e h t f o y r a m m u S . 2 e l b a T l e d o M I P S V ) V ( ) a P G ( ) m µ ( ) m µ ( ) m µ ( E w h L ) m µ ( 0 d e p y t a t a D . f e R ) s µ ( 1 . 2 5 1 0 1 1 . 2 5 1 0 1 4 4 4 . 0 4 4 4 . 0 4 4 4 . 0 8 . 6 3 2 . 1 4 8 3 . 9 2 . 4 3 9 . 1 2 S R , C B S , ‡ r a b 3 1 0 1 . 0 S R , C B S , ‡ r a b 3 1 0 . 1 S R , C B S , ‡ r a b 3 1 0 1 . 0 D F Q S C , C C , E B D F Q S C , C C , E B D F Q S C , C C , E B S F , S R S F , S R D F Q S C , C C , E B D F Q S C , E S , E B S R , C B S D F Q S C , C C , E B S S D L , C C , E B S R S R S F D F Q S C , C C , E P D F Q S C , C C , E P A / N A / N A / N 6 7 . 8 4 5 . 5 6 7 . 8 4 5 . 5 7 . 8 4 7 . 8 4 5 . 8 4 6 7 . 8 6 7 . 8 6 7 . 8 6 7 . 8 6 7 . 8 4 6 1 4 6 1 4 6 1 ∗ 0 0 2 4 6 1 0 0 2 0 0 2 9 4 1 6 6 1 9 4 1 9 4 1 9 4 1 0 4 0 4 0 4 0 4 0 4 0 4 0 4 0 4 0 7 2 1 . 2 2 1 . 2 2 1 . 2 2 1 . 2 5 1 0 . 2 2 . 2 2 . 2 2 . 2 2 . 2 0 1 6 0 1 7 0 1 6 0 1 7 0 0 3 0 0 3 0 0 3 0 1 6 0 1 6 0 1 6 0 1 6 0 1 6 7 0 . 2 t n e m i r e p x E 7 0 . 2 t n e m i r e p x E 7 0 . 2 7 0 . 2 n o i t a l u m S i n o i t a l u m S i t n e m i r e p x E 3 . 2 3 . 2 3 . 2 3 . 2 3 . 2 n o i t a l u m S i n o i t a l u m S i n o i t a l u m S i n o i t a l u m S i n o i t a l u m S i n o i t a l u m S i n o i t a l u m S i ] 7 3 [ ] 7 3 [ ] 7 3 [ ] 7 3 [ ] 7 1 [ ] 7 1 [ ] 7 1 [ ] 8 1 [ ] 9 1 [ ] 5 1 [ ] 5 1 [ ] 5 1 [ Delayed pull-in transitions in overdamped MEMS devices 13 Figure 5. Re-scaling the pull-in times according to the scaling law tPI = O(−1/2) as → 0 for (a) the data of [17] and (b) all accelerometer data. acceleration, we have from (14): (cid:18) V (cid:19)2 − 1 ≈ 2 (V − VSPI) , VSPI = when V ≈ VSPI, and similarly in the case when the acceleration is varied. VSPI 4.1. Estimating the pull-in voltage The above analysis highlights the sensitivity of pull- in time to the actual pull-in voltage, which can be quite difficult to measure precisely - for instance, quasi-statically increasing the voltage until pull-in occurs is subject to mechanical noise as well as imprecision in voltage measurements. There may also be rounding error in the reported pull-in voltage. Hence, in what follows we suggest an alternative approach. For the data of [17], we have determined the best-fit (least-squares) line over the five data points that are closest to the pull-in transition (dotted lines on figure 5a). By finding the intercept of each best-fit line with the horizontal axis, we are able to compute 'corrected' values of the pull-in voltage. These are {48.615, 48.697, 48.477} V, which are in good agreement with the reported values of {48.7, 48.7, 48.5} V respectively. This procedure may be applied more generally as a way to estimate the pull- in voltage based only on data for the pull-in times, rather than using the static behaviour of the system before pull-in occurs. 4.2. Estimating the pre-factor Finally, we show that it is possible to obtain good quantitative agreement with the predicted pre-factor in the scaling of (14), when we use realistic values of the lumped parameters b and k. We make the pull-in times shown in figure 1 dimensionless using the timescale [t] = b/k. We can then use [t] as a √ single fitting parameter to fit each data set to the dimensionless prediction TPI ∼ 2π/ 3. This is consistent with the way we have separated each data set: as the properties of the squeeze film (e.g. plate area, ambient pressure) and the mechanical restoring force (e.g material stiffness, beam length) entering b and k do not vary in each data set, the timescale [t] is fixed. Many of the references in tables 1–2 have multiple data sets with the same timescale [t]; for example, when the actuation mechanism changes over a single experimental system (e.g. the data of [13], magenta symbols), and when numerical simulations use the same parameter values as experiments (e.g. the data of [12], orange symbols). In these cases we fit [t] to only one set of experimental data and use this value to non-dimensionalize all of the data sets. The best-fit (least-squares) timescales [t] are given in table 2 for the data on microbeams. For the data on parallel-plate actuators, the spring constant k is usually a known design parameter, and so we use the reported values of k to give the corresponding best- fit damping coefficient b. These compare reasonably well to approximate values obtained in numerical simulations of squeeze film damping (table 1). The mk/b are all corresponding quality factors Q = smaller than unity, so that the fitting performed here is consistent with our assumption that the devices are overdamped. √ With the fitted values of [t], we obtain excellent collapse over all data sets considered, up to the sensitivity to the value of the pull-in voltage used; see the main panel of figure 6. In the inset of figure 6 4850525402468101010-40-200204060800.511.52104Delayed pull-in transitions in overdamped MEMS devices 14 √ Figure 6. Main plot: Dimensionless pull-in times obtained by fitting the overdamped timescale [t] = b/k. Plotted for comparison is the prediction TPI ∼ 2π/ 3 valid for (cid:28) 1 (black dotted line), as well as the large prediction TPI ∼ 9/(4) (black dashed-dotted line) [11]. Inset: The same data, re-scaled according to the scaling law TPI = O(−1/2) as → 0. we plot T −2 PI as a function of on linear axes, which demonstrates the collapse for small without this sensitivity. 5. Conclusions In this paper we have considered the pull-in dynamics of overdamped MEMS devices. When the system is near the static pull-in voltage/acceleration, the motion is known to slow down considerably during a meta-stable or bottleneck phase. By considering a lumped-parameter parallel-plate model, we have shown that the bottleneck behaviour is an instance of a saddle-node ghost [34]; the duration of the bottleneck increases ∝ −1/2, where is the normalized distance of the system beyond the pull-in transition. A detailed asymptotic analysis then allowed us to evaluate the prefactor in this scaling law. The result √ is a simple analytical prediction for the total pull-in time: tPI ∼ (b/k)2π/ 3, in which b is the effective damping coefficient and k is the lumped mechanical stiffness applicable to the bottleneck phase. This result complements previous studies that have calculated a similar asymptotic pull-in time for underdamped devices [28, 29]. The −1/2 scaling law explains the high sensitivity of the pull-in time observed in previous experiments and numerical simulations. Moreover, because the bottleneck phase dominates the dynamics during pull- in, the resulting pull-in time is relatively insensitive to what happens outside of the bottleneck region; this includes the precise geometry of the device, the effects of compressibility and air rarefaction, and the way in which stoppers limit the displacement before contact occurs. The implication is that a simple parallel- plate model, using lumped values for the damping coefficient and spring constant, is an effective means of capturing the behaviour of a complex MEMS device. Indeed, the wide range of available data collapsed onto a single master curve (figure 6), despite the number of additional effects that are present in the range of experiments and simulations analysed (summarized in tables 1–2). Moreover, while the assumption of a constant damping coefficient is often stated to give large errors [18, 12], we have shown that, in the bottleneck regime, this assumption is sufficient to correctly predict the pull-in time. The sensitivity of the bottleneck to external perturbations is the basis of using pull-in time as a sensing mechanism, as in some pressure sensors [37, 11] and accelerometers [12, 13, 14]. Currently, the lack of linearity in the response is considered to be the main disadvantage of these devices, and it has been suggested that the pull-in time curve might 10-410-310-210-110010110-110010110200.20.40.600.020.040.06Delayed pull-in transitions in overdamped MEMS devices 15 be linearized by the introduction of extra forces [13]. Our expression for the pull-in time partly resolves the issue, as it provides a simple power law that can be used to calibrate a device. In addition, our introduction of the dimensionless parameter , equation (6), captures both variations in the voltage and external acceleration near the pull-in transition. When plotted in terms of this parameter, we observe a collapse of data over experiments where either the voltage or the acceleration was varied. Finally, we discuss the conditions under which our analysis holds. We have considered only devices with low quality factors, so that inertia of the moving electrode can be neglected during the bottleneck phase. We also focussed on DC loads that are stepped from zero, since this loading type is commonly used in applications of the pull-in time. Nevertheless, our analysis may be adapted to other types of loading, provided the behaviour before pull-in remains quasi- static; for example, if the voltage is instead stepped from a positive value. However, in the case of a voltage sweep (e.g. triangular wave), the quasi-static condition is not met and the −1/2 scaling law will not apply. Similarly, extremely close to the pull-in transition, mechanical noise will eventually become important and limit the system response. Nevertheless, we hope that the unifying perspective we have presented here will lead to new insights in the application of dynamic pull- in instabilities. research leading to these Acknowledgments The results has received funding from the European Research Council under the European Union's Horizon 2020 Programme/ERC Grant No. 637334 (D.V.) and the EPSRC Grant No. EP/ M50659X/1 (M.G.). The data that supports the plots within this paper and other findings of this study are available from http://dx.doi.org/10.5287/bodleian:7J84mXZ78. Appendix A: Assumption of a constant damping coefficient The assumption of a constant damping coefficient has often been reported to give large errors compared to simulations that incorporate a variable damping coefficient [12, 18]; based on this, it is argued that a variable damping coefficient should always be used when predicting the pull-in time for MEMS applications. For example, ref. [12] consider the pull- in dynamics of a parallel-plate actuator, showing that a constant damping coefficient approximation leads to errors of up to 40%. However, [12] use the value of the damping coefficient when the plate is in the zero voltage state, binit. This damping is much smaller than the value when the plate is near the static pull- in displacement, bPI (where the thickness of the air gap is around 2/3 of the zero voltage thickness). Because the pull-in timescale [t] depends linearly on the damping coefficient (for overdamped devices), and the system spends most of its time close to the pull-in displacement during the bottleneck phase, using binit will lead to a significant under-prediction of the pull- in time. Here, we show that using bPI (our approach in the main text) is sufficient to correctly predict the pull-in time. We modify our spring-mass model to consider a variable damping coefficient b(x): 0AV 2 (d0 − x)2 + maext. d2x dt2 + b(x) + kx = dx dt 1 2 m (15) Ignoring compressibility and rarefaction effects, the incompressible Reynolds equation may be solved approximately in the parallel-plate geometry to give [36] b(x) = µC (d0 − x)3 , where µ is the air viscosity and C is a constant that depends on the dimensions of the moving plate. The damping coefficient corresponding to the zero voltage state, x = maext/k, is then binit = µC 0(1 − Aext)3 . d3 Since a variable damping coefficient does not change the steady solutions, the static pull-in displacement is xPI = (d0/3) (1 + 2Aext), as before. The damping coefficient during the bottleneck phase is then bPI = 27 8 µC 0(1 − Aext)3 , d3 so that binit bPI 8 27 . = We non-dimensionalize in a similar way to §2.2, though now we set T = t binit/k , Q = √ mk binit . Equation (15) then becomes Q2 d2X 1 + X = dT 2 + (1 − X)3 dX dT λ (1 − X)2 , (16) and the initial conditions remain X(0) = X(0) = 0. We may then perform a local analysis of equation (16) when the solution is in the bottleneck phase, along similar lines to §3.1.3 (where now we Taylor expand the (1 − X)3 term about X = Xfold = 1/3). This shows that the dimensional pull-in time, tPI, is given by tPI ∼ bPI k 2π√ 3 , Delayed pull-in transitions in overdamped MEMS devices 16 [4] W.-M. Zhang, H. Yan, Z.-K. Peng, and G. Meng. Electrostatic pull-in instability in MEMS/NEMS: A review. Sens. Actuators A Phys, 214:187–218, 2014. [5] M. I. Younis and F. Alsaleem. Exploration of new concepts for mass detection in electrostatically-actuated structures based on nonlinear phenomena. J. Comput. Nonlin. Dyn., 4(2):021010, 2009. [6] P. M. Osterberg and S. D. Senturia. M-TEST: a test chip for MEMS material property measurement using electrostatically actuated test structures. J. Microelectromech. Syst., 6(2):107–118, 1997. [7] A. V. Desai, J. D. Tice, C. A. Apblett, and P. J. A. Kenis. Design considerations for electrostatic microvalves with applications in poly (dimethylsiloxane)-based microflu- idics. Lab Chip, 12(6):1078–1088, 2012. [8] C. T.-C. Nguyen, L. P. B. Katehi, and G. M. Rebeiz. Micromachined devices for wireless communications. Proc. IEEE, 86(8):1756–1768, 1998. [9] R. P. LaRose and K. D. Murphy. Impact dynamics of MEMS switches. Nonlinear Dyn., 60(3):327–339, 2010. [10] P. Deng, N. Wang, F. Cai, and L. Chen. A high-force and high isolation metal-contact RF MEMS switch. Microsyst. Technol., 23(10):4699–4708, 2017. [11] R. K. Gupta and S. D. Senturia. Pull-in time dynamics as a measure of absolute pressure. In Proc. IEEE. Int. Workshop on MEMS (Nagoya, Jan. 1997), pages 290– 294. IEEE, 1997. [12] L. A. Rocha, E. Cretu, and R. F. Wolffenbuttel. Behavioural analysis of the pull-in dynamic transition. J. Micromech. Microeng., 14(9):S37, 2004. [13] R. A. Dias, E. Cretu, R. Wolffenbuttel, and L. A. Rocha. Pull-in-based µg-resolution accelerometer: Characteri- zation and noise analysis. Sens. Actuators A Phys, 172(1):47–53, 2011. [14] R. A. Dias, F. S. Alves, M. Costa, H. Fonseca, J. Cabral, J. Gaspar, and L. A. Rocha. Real-time operation and characterization of a high-performance time-based accelerometer. J. Microelectromech. Syst., 24(6):1703– 1711, 2015. [15] A. Missoffe, J. Juillard, and D. Aubry. A reduced- order model of squeeze-film damping for deformable micromechanical structures including large displacement effects. J. Micromech. Microeng., 18(3):035042, 2008. [16] A. H. Nayfeh, M. and E. M. Abdel- Rahman. Reduced-order models for MEMS applications. Nonlinear Dyn., 41(1):211–236, 2005. I. Younis, [17] M.-A. Gr´etillat, Y.-J. Yang, E. S. Hung, V. Rabinovich, G. K. Ananthasuresh, N. F. De Rooij, and S. D. Senturia. Nonlinear electromechanical behaviour of an electrostatic microrelay. In Proc. Int. Conf. on Solid State Sensors and Actuators (Transducers 1997, Chicago, IL), volume 2, pages 1141–1144. IEEE, 1997. [18] E. S. Hung and S. D. Senturia. Generating efficient dynam- ical models for microelectromechanical systems from a few finite-element simulation runs. J. Microelectromech. Syst., 8(3):280–289, 1999. [19] M. I. Younis, E. M. Abdel-Rahman, and A. Nayfeh. A reduced-order model actuated microbeam-based MEMS. J. Microelectromech. Syst., 12(5):672–680, 2003. electrically for Figure 7. Trajectories obtained by numerical integration of equation (15) with different damping models b(x) (see legend). Here dimensionless quantities are defined as in §2.2, but now we set T = t/(binit/k) and Q = mk/binit. √ valid for 0 < (cid:28) 1. We conclude that setting b = bPI in our constant damping model (as done in the main text) yields the correct asymptotic expression for the pull-in time (see equation 14), while setting b = binit will lead to a prediction of tPI that is a fraction 8/27 (approximately 30%) of the true value. This is illustrated in figure 7, which compares the numerical solution of the full equation (16) with two approximate approaches: (i) the solution in which we instead assume a constant damping coefficient b(x) = bPI (corresponding to setting X = Xfold in the (1−X)3 term) and (ii) the solution with a constant coefficient b(x) = binit (setting X = 0 in the (1 − X)3 term). We see that the constant coefficient bPI successfully captures the duration of the bottleneck phase, and hence the total time taken to pull-in, while using binit leads to large errors. In ref. [12] it is reported that using a constant coefficient binit yields a pull-in time that is 60% of that obtained using a variable damping coefficient. This is larger than the ≈ 30% that we predict here. However, the simulations reported in [12] also incorporate compressibility and rarefaction effects in the squeeze film, which may account for this discrepancy. References [1] R. C. Batra, M. Porfiri, and D. Spinello. Review of mod- eling electrostatically actuated microelectromechanical systems. Smart Mater. Struct., 16(6):R23–R31, 2007. [2] S. Krylov, B. R. Ilic, D. Schreiber, S. Seretensky, and H. Craighead. The pull-in behavior of electrostatically actuated bistable microstructures. J. Micromech. Microeng., 18(5):055026, 2008. [3] J. A. Pelesko and D. H. Bernstein. Modeling MEMS and NEMS. CRC press, 2002. [20] A. H. Nayfeh, M. I. Younis, and E. M. Abdel-Rahman. Dynamic pull-in phenomenon in MEMS resonators. Nonlinear Dyn., 48(1-2):153–163, 2007. [21] S. Zaitsev, O. Shtempluck, E. Buks, and O. Gottlieb. Nonlinear damping in a micromechanical oscillator. Nonlinear Dyn., 67(1):859–883, 2012. [22] S. Chaterjee and G. Pohit. A large deflection model for the pull-in analysis of electrostatically actuated microcantilever beams. J. Sound Vib., 322(4):969–986, 2009. 010020030040050000.20.40.60.81Delayed pull-in transitions in overdamped MEMS devices 17 [23] B. McCarthy, G. G. Adams, N. E. McGruer, and D. Potter. including contact bounce, of an electrostatically actuated microswitch. J. Microelectromech. Syst., 11(3):276–283, 2002. A dynamic model, [24] K. Das and R. C. Batra. Pull-in and snap-through instabil- ities in transient deformations of microelectromechanical systems. J. Micromech. Microeng., 19(3):035008, 2009. [25] S. Krylov and N. Dick. Dynamic stability of electrostat- ically actuated initially curved shallow micro beams. Continuum Mech. Thermodyn., 22(6):445–468, 2010. [26] G. Flores, G. A. Mercado, and J. A. Pelesko. Dynamics and touchdown in electrostatic MEMS. In Proc. ASME Design Engineering Technical Conf. and Computers and Information in Engineering Conf. and 19th Biennial Conf. on Mechanical Vibration and Noise (Chicago, IL, 2–6 September 2003), volume 5, pages 1807–1814. American Society of Mechanical Engineers, 2003. [27] G. N. Nielson and G. Barbastathis. Dynamic pull- in of parallel-plate and torsional electrostatic MEMS actuators. J. Microelectromech. Syst., 15(4):811–821, 2006. [28] V. Leus and D. Elata. On the dynamic response of J. Microelectromech. electrostatic MEMS switches. Syst., 17(1):236–243, 2008. [29] M. M. Joglekar and D. N. Pawaskar. Estimation of oscillation period/switching time for electrostatically actuated microbeam type switches. Int. J. Mech. Sci., 53(2):116–125, 2011. [30] O. Y. Loh and H. D. Espinosa. Nanoelectromechanical contact switches. Nat. Nanotechnol., 7(5):283–295, 2012. [31] M. Gomez, D. E. Moulton, and D. Vella. Critical slowing instabilities. down in purely elastic 'snap-through' Nature Phys., 13:142–145, 2017. [32] P. M. Chaikin and T. C. Lubensky. Principles of condensed matter physics. Cambridge University Press, 1995. [33] S. H. Strogatz and R. M. Westervelt. Predicted power laws for delayed switching of charge-density waves. Phys. Rev. B, 40(15):10501, 1989. [34] S. H. Strogatz. Nonlinear Dynamics and Chaos. Westview Press, 2014. [35] L. M. Castaner and S. D. Senturia. Speed-energy optimization of electrostatic actuators based on pull-in. J. Microelectomech. Syst., 8(3):290–298, 1999. [36] T. Veijola, H. Kuisma, J. Lahdenpera, and T. Ryhanen. Equivalent-circuit model of the squeezed gas film in a silicon accelerometer. Sens. Actuators A Phys, 48(3):239–248, 1995. [37] R. K. Gupta, E. S. Hung, Y.-J. Yang, G. K. Ananthasuresh, and S. D. Senturia. Pull-in dynamics of electrostatically- actuated beams. In Proc. Solid State Sensor and Actuator Workshop, pages 1–2, 1996. [38] L. A. Rocha, E. Cretu, and R. F. Wolffenbuttel. analysis and modeling of the Pull-in dynamics: transitional regime. In Proc. MEMS'04 (Maastricht, The Netherlands, 25–29 January 2004), pages 249–252. IEEE, 2004. [39] S. Krylov and R. Maimon. Pull-in dynamics of an elastic beam actuated by continuously distributed electrostatic force. J. Vib. Acoust., 126(3):332–342, 2004. [40] M. H. H. Nijhuis, T. G. H. Basten, Y. H. Wijnant, H. Tijdeman, and H. A. C. Tilmans. Transient non- linear response of including squeeze film effects. Proc. Eurosensors XIII (The Hague, The Netherlands, 1999), pages 729–732, 1999. 'pull-in MEMS devices'
1810.03111	1	1810	2018-10-07T09:15:44	A Broadband Cavity-Backed Slot Radiating Element in Transmission Configuration	[ "physics.app-ph" ]	A planar technology stripline-fed slot radiating element in transmission configuration is proposed. Its main advantages are the broad impedance bandwidth achieved and the property that it only radiates into half-space, which are obtained, respectively, with the use of its complementary strip element and using a cavity-backed slot. A lattice network circuit model is proposed both to explain the behavior of the structure and to establish a design methodology. Its capabilities are shown through simulation and demonstrated in a proof of concept prototype. Measurement results show a unidirectional broadside radiation pattern and a fractional bandwidth of 48%, significantly superior to other slot-based radiating elements found in the literature. The element has the ideal characteristics for building series-fed reconfigurable arrays for wide-band applications.	physics.app-ph	physics	IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION 1 A Broadband Cavity-Backed Slot Radiating Element in Transmission Configuration Alberto Hern´andez-Escobar, Elena Abdo-S´anchez, Member, IEEE, and Carlos Camacho-Penalosa, Senior Member, IEEE (DOI: 10.1109/TAP.2018.2874069) © 2018 IEEE* 8 1 0 2 t c O 7 ] h p - p p a . s c i s y h p [ 1 v 1 1 1 3 0 . 0 1 8 1 : v i X r a Abstract -- A planar technology stripline-fed slot radiating element in transmission configuration is proposed. Its main advantages are the broad impedance bandwidth achieved and the property that it only radiates into half-space, which are obtained, respectively, with the use of its complementary strip element and using a cavity-backed slot. A lattice network circuit model is proposed both to explain the behavior of the structure and to establish a design methodology. Its capabilities are shown through simulation and demonstrated in a proof of concept prototype. Measurement results show a unidirectional broadside radiation pattern and a fractional bandwidth of 48%, significantly superior to other slot- based radiating elements found in the literature. The element has the ideal characteristics for building series-fed reconfigurable arrays for wide-band applications. Index Terms -- Broadband radiating element, cavity-backed slot, lattice network, leaky-wave antennas, stripline. I. INTRODUCTION Society's increasing needs for wide-bandwidth wireless commu- nications demands broadband, directive and reconfigurable antennas to be easily integrated into small terminals. Slot-like antennas have been studied and used intensively in recent decades [1] due to their many advantages, which include low cost, low profile, durability, easy manufacture and integration in the casing of almost any device. Therefore they are strong candidates for building arrays that can comply with the aforementioned requirements. Nevertheless, these antennas usually radiate towards the entire space, thus exhibiting bilateral radiation. This feature makes them poorly suited for their use in conventional directive arrays since the radiation pattern will always have at least two main lobes. In order to prevent radiation towards one of the half-spaces, a cavity made of a conductive material can be placed behind the slot, forming a so-called Cavity-Backed Slot (CBS). This structure was studied and used during the second half of the last century [2], [3] founding that the radiation properties are determined by the dimensions of the cavity. Later, these antennas were left out due to their bulkiness and complex manufacture using planar technology. The development of Substrate Integrated Waveguide (SIW) technology opened up a way for a compact and easy implementation of CBS antennas [4], [5] by using metallic via holes to build the cavity. In addition to the aforementioned drawbacks, slot and CBS anten- nas have a narrow impedance bandwidth due to the resonant nature * Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. Manuscript received July 27, 2017. This work was supported in part by the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 706334, in part by the Spanish Ministerio de Educaci´on, Cultura y Deporte (Programa para la Formaci´on del Profesorado Universitario) under Grant FPU15/06457 and in other part by the Spanish Ministerio de Econom´ıa y Competitividad, under the project ADDMATE TEC2016-76070-CR3-3-R. The authors are with the Departamento de Ingenier´ıa de Comunicaciones, Escuela T´ecnica Superior de Ingenier´ıa de Telecomunicaci´on, Universidad de M´alaga, Andaluc´ıa Tech, 29010 M´alaga, Spain (e-mail: [email protected]). of the element. A recent work [6] uses the SIW technique and a bow-tie slot antenna to increase its impedance bandwidth, but still only 9.4% is achieved. A solution to the narrow bandwidth which uses transmission configuration and offers an ultra broad bandwidth was proposed in [7] for the case of microstrip-fed slots. Very broad bandwidth is achieved by using a stub complementary to the slot that matches the impedance of the structure, resulting in an all-pass section. The two-port configuration of this radiating element enables the design of tunable series-fed arrays, as shown in [8]. However, the bilateral radiation problem still exists for the so-called complementary strip-slot. In an attempt to address this issue, a reflector was placed behind the array in [8] and [9] with the drawbacks of narrowing the working band of the antenna and considerably increasing its size. A novel compact stripline-fed CBS radiating element using planar technology is proposed in this communication. In a similar way as in [7], its complementary stub is placed under the slot to enhance the impedance bandwidth. The two-port transmission configuration is also adopted here and the cavity is implemented using SIW technology. The present work can be seen as a transformation of the complementary strip-slot to obtain unidirectional radiation. This is achieved by replacing the microstrip line by stripline, preventing the structure from radiating towards one half-space. These modifications in the geometry entail several challenges: the CBS behaves differently than that of a conventional slot, the stripline makes it more difficult to obtain impedance matching, and controlling the modes supported by the structure becomes a fundamental part of the design. Stripline-fed CBSs were also researched in the last century [10] -- [12] and were proposed [11] as a candidate for series-fed arrays. More recently, the stripline-feeding approach has been used as an alternative to cavity feeding [13]. Furthermore, they have previously been used in conjunction with SIW technology [14]. The impedance bandwidth of the CBS antennas found in the literature is usually very narrow, less than 5%, for both one-port and two-port (transmission) configurations. The design presented in this manuscript shows a very broad fractional impedance bandwidth (48%). Thus, to the authors' knowledge, the proposed radiating element exhibits the highest impedance bandwidth from among the other unidirectional radiating narrow slots cited in the bibliography. With these characteristics, the element can be used to make wide-band series-fed antenna arrays as in [8], but featuring unidirectional radiation. The manuscript is structured as follows: Section II introduces the antenna geometry and the modes supported by the structure. Section III extracts an equivalent circuit based on the lattice network. Section IV describes the design methodology used. Section V presents a design in the 5 GHz band and provides simulation and measurement results. Section VI discusses several aspects about the use of the element in series-fed arrays. Finally, the conclusions are summarized in Section VII. II. STRUCTURE AND SUPPORTED MODES The proposed structure consists of a CBS excited in transmission configuration by an asymmetric stripline where a stub, complemen- 2 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION Feeding Stripline Slot Stub Cavity Conductive Walls z y x Fig. 1. Geometry of the proposed structure. h Ɛr h1 h2 wslot wstub wcavity z x Fig. 2. Cross-section of the simulated transmission structure used to compute the modes supported by the structure. tary to the slot, is placed beneath it. The slot is etched on the upper ground plane and the stub is aligned with it but placed on the stripline layer. Lateral metallic walls are used to keep the structure closed, building the cavity, which is filled with dielectric material. Fig. 1 shows the geometry of the proposed radiating element. The behavior of the structure is determined by the modes supported by the section of the structure (shown in Fig. 2) along the direction transverse to the feeding line (y-axis in Fig. 1). The fundamental mode of this structure is a quasi-TEM mode with no cut-off, sup- ported by the two conductors in the structure (strip and cavity walls), in a similar way to a classic stripline TEM mode. Due to the slot, the structure is not completely closed as is a classic stripline. However, the field leaked outside the structure by the slot is not significant and, thus, it is possible to assume that the fundamental mode is almost a TEM mode. Another mode supported by the structure is the slot mode of a CBS. Although there are some similarities between the field distribution of this mode and that of the slot mode of a classic slotline, the dimensions of the cavity play a fundamental role in its behavior. Furthermore, as this mode is not affected by the strip, it is supported by a single-conductor structure (the cavity walls) and, thus, it has a cut-off frequency. Furthermore, higher-order modes can also appear. The first higher order mode supported by this structure is the TE10 mode of the rectangular waveguide made up of the metallic walls and the upper and the lower ground planes of the cavity. As will be justified in Section III, in order to cancel out the is resonant behavior of the slot and achieve broad matching, necessary for both the slot and the quasi-TEM modes to be excited. If the dimensions of the cavity are chosen properly, the element can successfully operate in a wide frequency band. The structure's lowest working frequency is limited by the cut-off frequency of the slot mode of the CBS, because its propagation is necessary for the structure to radiate. The TE10 mode is unwanted and its appearance it Fig. 3. Simulated phase constants of the three first modes propagating through the transmission structure for four different cavity cross-sections. will limit the working frequency band at higher frequencies. The cut- off frequency of the unwanted TE10 mode could limit the highest working frequency. Thus, its cut-off frequency should be as high as possible. However, since the other modes inside the structure barely excite the TE10 mode, the presence of this mode may go unnoticed even above its cut-off frequency. Furthermore, since the transmission system is short-circuited at its ends (by the metallic walls that close the structure), a cavity is formed. The TE10 mode will resonate inside the cavity at the resonance frequency of the resonant mode TE101, as explained in [15]. The effect of this higher order mode will become noticeable at this resonant frequency. A simulation of the transmission system made up of the cavity, shown in Fig. 2, has been carried out for different dimensions using the ANSYS HFSS commercial electromagnetic simulator. Fig. 3 displays the phase constants of the first three modes of this transmission system to illustrate how these modes are supported by the structure and how their cut-off frequencies change as the cavity dimensions are modified. The dielectric used in this simulation is air and the width of the slot is 0.3 mm. To increase the cut-off frequency of the TE10 mode, a narrow cavity, with a low wcavity, is preferred. However, if the cavity width, wcavity, is reduced, the cut-off frequency of the slot mode will increase. To broaden the operating frequency band, the cut-off frequency of the slot mode can be lowered by increasing the height of the stripline and thus the cavity, h. However, the radiating element will be thicker, as extracted from the study of Fig. 3. The width of the slot, wslot, also changes the cut-off frequency of its mode. Lower values of the width reduce the frequency and thus are desirable. However, very low values of wslot could be difficult to implement accurately and could reduce the radiation of the element significantly. Furthermore, to increase the bandwidth where the TE10 mode is supported but not excited, a high resonance frequency of the resonant mode TE101 is needed. To do this, the cavity must be as short as possible, with a low lcavity. However, the length of the cavity is limited by the length of the slot. In conclusion, the dimensions of the cavity must be chosen carefully due to the trade-off between the operating bandwidth of the element and its thickness. Results from Fig. 3 show that, when wcavity is 24 mm and h is 3.9 mm, there is a frequency band between 3.8 GHz and 6.2 GHz where only the desired modes, stripline and slot, are supported. If the TE10 mode is not excited beyond 6.2 GHz and if the length of the cavity is 35 mm, the working frequency band would extend up to 7.5 GHz, which corresponds to the resonance frequency of the resonant TE101 mode in this case. IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION 3 Za Z0 Zb Zb Z0 Za Fig. 4. Equivalent lattice network of the complementary structure. ZTEM,β TEM stubl Z /4slot ,β slot slotl Zb (a) Za (b) Fig. 5. Circuit model of the lattice network impedances using transmission lines. (a) Zb. (b) Za. III. LATTICE-NETWORK-BASED EQUIVALENT CIRCUIT In an analogous way as in [7], a lattice network ([16], Fig. 4) is proposed to model the element. The field distribution of a symmetric structure can be separated into two contributions: the field distribution of the even mode and the field distribution of the odd mode. Thus, if the reference planes are chosen appropriately (coincident at the symmetry plane, the lattice network separates the contribution of both the even and odd modes in each of the branches, allowing the creation a model of the structure from independent models of the even and odd modes. These two modes correspond to the TEM mode of the stub and the slot mode of the CBS as described in the previous section. in this case), Fig. 5 shows the circuit model of the impedances of each branch. The major differences with the model proposed in [7] are that, in the case of Zb, the impedance of a stripline, ZT EM , (which can be extracted from [17]) and the propagation constant of a pure TEM mode, βT EM , are used instead. In the case of Za, its modeling is more challenging than in [7] since analytical expressions to obtain the impedance of the slot mode, Zslot, and its phase constant, βslot, have not been found in the literature. For this reason, the propagation constant of a classic slotline and the simulation results of the impedance of the slot are used as an approximation. To improve the accuracy of the model, a lossy transmission line which would take into account the radiation of the slot, can be used. The attenuation constant of the line, αslot, has to be obtained from the simulation of the complete structure. This is due to the fact that the radiation of the slot heavily depends on the coupling between the feeding stripline and the slot. Therefore, both the width of the feeding line and its distance to the slot play a fundamental role. IV. DESIGN Firstly, the dimensions of the cross-section of the cavity (wcavity, h, εr, and wslot in Fig. 2) must be chosen according to the following two criteria: first, obtaining a cut-off frequency of the slot mode lower than the design frequency and, second, achieving a cut-off frequency of the TE10 as high as possible. The effect of these dimensions on the cut-off frequencies has already been discussed in Section II. is chosen so that the resonance of the slot coincides with the design frequency. The slot resonance is modeled with a transmission line as shown in Fig. 5(b). As the propagation constant, βslot, is already determined in Secondly, the length of the slot, lslot, the previous step, the length of the slot, lslot will only modify the electrical length of the transmission line. This length will also determine the minimum length of the cavity, lcavity, which sets the resonance of the resonant mode TE101. Lastly, the other parameters, wstub, lstub and h1 must be chosen in order to obtain broad impedance matching. To do this, first, let us express the image impedance of the equivalent circuit of the complementary structure shown in Fig. 4 as a function of Za and Zb, as in [16]: Zim(ω) = pZa(ω)Zb(ω). (1) The image impedance of the structure must be constant and with the same value as the characteristic impedance of the feeding line, Z0, over a wide range of frequencies. As modeled in Fig. 5, the impedances Za and Zb have poles and zeros at the corresponding resonance frequencies. For Zim to be approximately constant, these poles and zeros must coincide in frequency so they properly cancel each other out and the impedance level does not change over frequency. For this to happen, the electrical length of the transmission lines modeling Za and Zb must be the same. These conditions can be expressed as follows: 1 2 √ZT EM Zslot = Z0 βT EM lstub = βslotlslot. (2a) (2b) Although it is possible to modify ZT EM using the width of the stub, wstub, this may not be enough to fulfill condition (2a), since stripline structures present lower impedances than their microstrip counterpart. Increasing h allows a higher stripline impedance, and thus, a higher ZT EM to be achieved; however, this reduces the impedance of the slot, Zslot. In this case, the asymmetric stripline can be used to increase Zslot, as proposed in [18]. Decreasing the distance between the strip and the slot, h1, while keeping constant h (increasing h2) leads to higher values of Zslot. However, if h1 is too low, the strip will interfere with the field distribution of the slot mode and the lattice network will not be able to separate the TEM and slot modes properly. Finally, condition (2b) can be fulfilled by adjusting the length of the strip. This length only modifies the electrical length of the transmission line in the model shown in Fig. 5(a), that is, the position of the zero of Zb. V. IMPLEMENTATION AND RESULTS In order to make the overall size of the structure larger, easing the manufacturing requirements, a low permittivity substrate was chosen to implement a proof of concept structure. Thus, a suspended stripline configuration has been chosen, which also allows for a higher flexibility in the implementation of the heights h1 and h2. Three metallic layers have been printed on commercial substrates: lower ground plane, upper ground plane with the slot and the stripline layer. To suspend them in the air, nylon washers have been used around steel screws which pierce through the three layers as metallic posts to make the lateral walls of the cavity. Fig. 6 shows the geometry of the proof of concept structure. The substrate used to manufacture the three metallic layers is Rogers RO4350B with thickness of 0.25 mm and εr of 3.66. The width of the feeding stripline, wf eed, was chosen to have a characteristic impedance, Z0, of 50 Ω. The width and height of the cavity, wcavity and h, and the width of the slot, wslot, were chosen to ensure that the slot mode propagates from 3.5 GHz and the cut-off of the TE10 mode is 6 GHz. The width of the stub, wstub, together with the asymmetry of the stripline (h1 and h2) were selected to achieve condition (2b). With the current implementation possibilities, 4 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION wcavity y x y t i v a c l d e e wf dSIW 1 h 2 h b u t s l t o l s l wslot wstub sSIW h (a) (b) Fig. 6. Simulated proof of concept structure. (a) Top view. (b) Section of the cavity. The stub and slot sections have been enlarged to improve the visibility of the elements. (a) (b) Fig. 7. Manufactured prototype. (a) Top view (wf eed = 2.8 mm, wcavity = 24 mm, lcavity = 35 mm, wstub = 0.3 mm, lstub = 11.1 mm, wslot = 0.3 mm, lslot = 12.15 mm, h = 3.912 mm, h1 = 0.787 mm, h2 = 3.125 mm, dSIW = 2 mm and sSIW = 6.1 mm). (b) Lateral view. it would not have been possible to satisfy this condition using a higher effective dielectric constant. The lengths of the complementary Fig. 8. Magnitude of the S-parameters of the simulated and measured structure. elements, lslot and lstub, were designed to ensure complementarity, fulfilling condition (2a). The pole of Za and the zero of Zb are placed at 5.2 GHz. The length of the cavity, lcavity is chosen to be high enough not to affect the behavior of the strip and slot ends. The chosen value results in an approximate resonance of the TE101 resonant mode at 6.7 GHz. The analysis in previous sections did not take into account either the multiple layer structure or the presence of the metallic posts and nylon washers, so some tuning of the dimensions using HFSS was needed to obtain the final design. The prototype was manufactured and assembled. 50 Ω standard stripline connectors were used. Fig. 7 shows the result of the manufacturing process and its dimensions. In order to place the reference planes at the center of the element (coincident), a TRL calibration kit was designed. Fig. 8 shows the magnitude of the S11 and S21 parameters for both the simulated and measured cases. Using the criterion of -10 dB for the S11 to determine the impedance bandwidth, impedance matching of up to 6.7 GHz is foundin the measurements, versus 6.8 GHz in the simulation. The CBS does not exhibit significant radiation below 4 GHz (radiated power of less than 5% of the input power), which limits the use of this radiating element at lower frequencies and, thus, a fractional bandwidth of about 48% is obtained (50% in the simulation). The spurious ripple at 6.8 GHz can be explained due to the appearance of the TE101 resonant mode, as was expected from the analysis of Section II. It can be shown analytically that a small difference in either the lengths of the stubs or the effective dielectric constants can produce a kind of ripple at the frequency corresponding to the zero of Zb. Therefore, the small ripple found in measurements around 5.3 GHz is attributed to a difference in the electrical length of both stubs, due to manufacturing errors. To explain the differences between the model, simulation and measurements in Fig. 8, the values of the impedances of the lattice network, Za and Zb, are extracted from the S-parameters and compared with those of the transmission line model of Fig. 5. The results are shown in Fig. 9. Good agreement between the model and simulation is found, which proves the validity of the independent designs of the stub and CBS elements. It can be seen that the pole of Za and the zero of Zb appear around the design frequency, 5.2 GHz, resulting in the complementarity of the structure. The frequency shift in the pole of Za in the measured case explains the difference between the simulation results and the measurements in Fig. 8. Due to limitations of the manufacturing technology available, the thicknesses of the substrates are thinner than expected and they bend a little across their surfaces, especially on the slot, where there is almost no dielectric left. The absence of dielectric near the slot reduces the effective εr of the slot mode, leading to a frequency shift. In IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION 5 ) m h O ( ) a Z ( l a e R 500 400 300 200 100 0 3 Model Simulation 1 Measurement Simulation 2 ) m h O ( ) a Z ( g a m I 300 200 100 0 -100 -200 -300 Model Simulation 1 Measurement Simulation 2 4 6 Frequency (GHz) 5 7 3 4 6 Frequency (GHz) 5 7 (a) Model Simulation 1 Measurement Simulation 2 80 60 40 20 0 ) m h O ( ) b Z ( l a e R (b) Model Simulation 1 Measurement Simulation 2 150 100 50 0 -50 -100 ) m h O ( ) b Z ( g a m I 0 -5 -10 ) i B d ( n i a G -15 3.5 Measured Simulation 1 Simulation 2 4 4.5 5 5.5 6 6.5 7 Frequency (GHz) Fig. 11. Measured and simulated gain of the radiating element sampled in broadside direction. 3 4 6 Frequency (GHz) 5 (c) -150 3 7 4 6 Frequency (GHz) 5 7 (d) Fig. 9. Lattice network impedances of the radiating element: model, design simulation (with nominal dimensions), prototype measurement and the pro- totype simulation (with actual manufactured prototype dimensions). (a) Real part of Za. (b) Imaginary part of Za. (c) Real part of Zb. (d) Imaginary part of Zb. (a) (b) Fig. 10. Proof of concept structure built and measured. Radiation mea- surements made in the anechoic chamber of the Laboratorio de Ensayos y Homologaci´on de Antenas, Universidad Polit´ecnica de Madrid, Madrid (Spain). order to verify this effect, additional simulation results using the actual dimensions of the prototype have been included in Fig. 9 as Simulation 2. The missing dielectric in the center of the slot has been simulated with a length of 10 mm, and the thicknesses of the top, middle and bottom layers, instead of being the target 0.25 mm, are around 0.15 mm, 0.2 mm and 0.24 mm respectively. These changes explain the frequency shift. Given the sensitivity of the element to the thicknesses of the substrates and the uncertainties introduced by the manufacturing process, the discrepancies in the impedance level are reasonable. The effect of the bending and reduced thicknesses of the substrates is also present in the manufactured TRL calibration kit and may have introduced additional errors in the measurements. Finally, the radiation properties of the prototype have been mea- sured, as shown in Fig. 10. Fig. 11 represents the gain over fre- quency for both the simulated cases, and the measurement in the broadside direction (+Z-axis in Fig. 10). Given the sensitivity of the manufacturing process, reasonable agreement is found between the measurements and the simulations, showing an increasing gain up to the frequency of 5.5 GHz, from which it slowly decreases. (c) (d) Fig. 12. Radiation gain patterns of the simulated and measured structure. (a) 5.2 GHz, XZ plane. (b) 6 GHz, XZ plane. (c) 5.2 GHz, YZ plane. (d) 6 GHz, YZ plane. Some discrepancies were expected as there are significant differences between the simulated gains in the structure with the target design dimensions and the structure with the manufactured actual dimen- sions. The low values of gain are expected, since the power radiated by this two-port element is only about 20% or less of the input power throughout the working band. This is due to power leaking to the second port as desired, since the structure is proposed as a radiating element for series-fed arrays. Fig. 12 shows the radiation gain patterns of the simulation, using HFSS, and manufactured structure at 5.2 GHz and 6 GHz. It can be seen, as expected, that the structure exhibits broadside radiation in only one half-space. Measurement and simulation differences may be due to the presence of the connectors and the heads of the screws, not considered in the simulation. Nevertheless, good agreement between them is found. The polarization obtained is linear in the principal planes with very good cross-polarization discrimination. The cross-polarization simulation results were omitted since they were less than -50 dB. 6 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION r e w o P t u p n I / r e w o P d e t a i d a R 0.3 0.25 0.2 0.15 0.1 0.05 0 3 Previous Design Wider Slot Design 4 5 6 7 Frequency (GHz) Fig. 13. Simulation results of fraction of radiated power of input power of the previous design and a design featuring a wider slot. was designed and manufactured, achieving satisfactory simulation and measurement results: S11 shows a wide impedance matching, a measured fractional bandwidth of 48%, and the element exhibits unidirectional radiation in the working bandwidth with a very pure linear polarization. The proposed structure can be used to build reconfigurable wideband, series-fed antenna arrays to be used in bandwidth-intensive wireless communications systems. Future work will focus on finding a more robust implementation of the structure. ACKNOWLEDGMENT The authors would like to thank the anonymous reviewers for their valuable comments and suggestions to improve the quality of the paper. They are also grateful to Prof. J. Esteban, from Universidad Polit´ecnica de Madrid (Spain), for greatly helping with the understanding of the structure. VI. USE IN SERIES-FED ARRAYS REFERENCES The transmission configuration of this element makes it especially suitable for traveling-wave, series-fed arrays. Its broad impedance matching allows frequency reconfigurable arrays to be built without requiring the modification of the geometry of the element (as needed when resonant elements are used), simplifying the design, as done in [8]. In these arrays, the control of the power radiated by the elements is very important. The power radiated by the proposed element can be controlled by changing the width of the slot. However, as stated in Section II, an increase in the width of the slot leads to a higher cut- off frequency in the CBS mode. To solve this problem, increasing the height of the cavity can help to maintain a wide, working bandwidth. In return, the impedance of the slot mode will change and, in order to satisfy condition (2b), the width of the stub should be readjusted. To illustrate the control of the radiation properties by changing the width of the slot, another design has been simulated. In this case, the width (wslot) and length (lslot) of the slot are set to 1.5 mm and 13.2 mm respectively. The width (wstub) and length (lstub) of the stub are 0.8 mm and 10.7 mm respectively. The distance between the strip and the bottom ground plane, h2, has been increased to 5.5 mm. The other parameters have not been modified. Fig. 13 shows the simulated fraction of the input power that is radiated by this element compared to that of the previous design. A significant increase in the percentage of the radiation power can be observed throughout the working bandwidth. This means that a series-fed array with 8 elements radiating each 25% of their input power would radiate around 90% of the input power of the array, ignoring losses. This way, by controlling the radiated power of the radiating element, it is possible to choose the size of the array and, thus, the directivity. The spacing between the elements of the array is also another important parameter to take into account. The minimum distance between elements is limited by the width of the cavity, wcavity, since the same row of metallic posts can be used to make the vertical walls of the cavity of two adjacent elements. This space should be enough for most arrays but, if this were not the case, a material with a higher εr could be used if the implementation technology allows it. VII. CONCLUSION A broadband CBS radiating element fed by stripline with a trans- mission configuration is proposed. Its main novelty is the enhanced impedance bandwidth obtained when a complementary stub is placed beneath the slot. Modes propagating through the cavity play a fun- damental role in its behavior. In order to understand the performance of the structure and simplify its design, a lattice network-based transmission line model was proposed. This methodology greatly avoids the use of a parametric analysis. A proof of concept structure [1] D. M. Pozar, "Reciprocity method of analysis for printed slot and slot- coupled microstrip antennas," IEEE Trans. Antennas Propagat., vol. 34, pp. 1439 -- 1446, Dec. 1986. [2] J. Galejs, "Admittance of rectangular slot which is backed by a rectan- gular cavity," IEEE Trans. Antennas Propagat., vol. AP-11, no. 2, pp. 119 -- 126, Mar. 1963. [3] S. A. Long, "Experimental Study of the Impedance of Cavity-Backed Slot Antennas," IEEE Trans. Antennas Propagat., vol. AP-23, no. 1, pp. 1 -- 7, Jan. 1975. [4] G. Q. Luo, Z. F. Hu, L. X. Dong, and L. L. Sun, "Planar slot antenna backed by substrate integrated waveguide cavity," IEEE Antennas Wire- less Propag. Lett., vol. 7, pp. 235 -- 239, 2008. [5] L. Ge, Y. Li, J. Wang, and C. Sim, "A low-profile reconfigurable cavity- backed slot antenna with frequency, polarization, and radiation pattern agility," IEEE Trans. Antennas Propagat., vol. 65, no. 5, pp. 2182 -- 2189, May 2017. [6] S. Mukherjee, A. Biswas, and K. V. Srivastava, "Broadband substrate in- tegrated waveguide cavity-backed bow-tie slot antenna," IEEE Antennas Wireless Propag. Lett., vol. 13, pp. 1152 -- 1155, 2014. [7] E. Abdo-S´anchez, J.E. Page, T. Mart´ın-Guerrero, J. Esteban and C. Camacho-Penalosa, "Planar Broadband Slot Radiating Element Based on Microstrip-Slot Coupling for Series-fed Arrays," IEEE Trans. Antennas Propagat., vol. 60, no. 2 Dec. 2012, pp. 6037 -- 6042. [8] E. Abdo-S´anchez, D. Palacios-Campos, C. Fr´ıas-Heras, F. Y. Ng-Molina, T. Mart´ın-Guerrero, and C. Camacho-Penalosa, "Electronically steerable and fixed-beam frequency-tunable planar traveling-wave antenna," IEEE Trans. Antennas Propagat., vol. 64, no. 4, pp. 1298 -- 1306, Apr. 2016. [9] M. van Rooyen, J. W. Odendaal, J. Joubert, "High-Gain Directional An- tenna for WLAN and WiMAX Applications," IEEE Antennas Wireless Propag. Lett., vol. 16, pp. 286 -- 289, 2017. [10] D. J. Sommers, "Slot array employing photoetched tri-plate transmission lines," IRE Trans. Microw. Theory Techn., vol. 3, no. 2, pp. 157 -- 162, Mar. 1955 [11] R. Shavit and R. Elliott, "Design of transverse slot arrays fed by a boxed stripline," IEEE Trans. Antennas Propagat., vol. AP-31, no. 4, pp. 545 -- 552, Jul. 1983. [12] S. Hashemi-Yeganeh and C. Birtcher, "Theoretical and experimental studies of cavity-backed slot antenna excited by narrow strip," IEEE Trans. Antennas Propagat., vol. 41, pp. 236 -- 241, Feb. 1993. [13] C. L ocker, T. Vaupel and T. F. Eibert, "Radiation efficient unidirectional low profile slot antenna elements for X-band application," IEEE Trans. Antennas Propagat., vol. 53, no. 8, pp. 2765 -- 2768, Aug. 2005. [14] Y. P. Huang and X. Z. Zhang, "A low cross-polarization stacked slot antenna backed by substrate integrated cavity," 2012 15th International Symposium on Antenna Technology and Applied Electromagnetics, AN- TEM 2012. [15] P. A. Rizzi, Microwave Engineering Passive Circuits, Ed. New Jersey: Prentice-Hall, 1988, pp. 428-430. [16] M. E. Van Valkenburg, Introduction to Modern Network Synthesis, Ed. New York: John Wiley & Sons Inc., 1960. [17] B. Bhat and S. K. Koul, Stripline-like Transmission Lines for Microwave Integrated Circuits, Ed. New Delhi: Wiley Eastern Limited, 1989, ch. 5. [18] P. Brachat and J. M. Baracco, "Dual-plarization slot-coupled printed antennas fed by stripline," IEEE Trans. Antennas Propagat., vol. 43, pp. 738 -- 742, Jul. 1995.
1908.02541	1	1908	2019-08-07T11:50:15	Performance Study of Strongly Coupled Magnetic Resonance	[ "physics.app-ph" ]	Strongly Coupled Magnetic Resonance (SCMR) uses electromagnetic resonance in order to efficiently transfer power wirelessly over mid-range distances. Since the energy exchange capability of resonant objects higher than non-resonant objects, strongly coupled systems are able to achieve more efficient energy transfer than other wireless power transfer systems. The paper presents detailed experimental and simulated analysis of the performance of the SCMR system. A prototype of the SCMR system was implemented and experiments were conducted to analyze the performance of the system. Finally, the resonant frequency of the system was experimentally verified and the factors influencing the wireless power transfer were also studied	physics.app-ph	physics	Performance Study of Strongly Coupled Magnetic Resonance 9 1 0 2 g u A 7 ] h p - p p a . s c i s y h p [ 1 v 1 4 5 2 0 . 8 0 9 1 : v i X r a Chathuranga M. Wijerathna Basnayaka∗, Dushantha Nalin K. Jayakody∗† , Abhishek Sharma‡ Hwang-Cheng Wang§, P Muthuchidambaranathan¶ and Kuljeet Kaur(cid:107) ∗School of Postgraduate Studies and Research, SRI LANKA Technological Campus, Sri Lanka †School of Computer Science and Robotics, National Research Tomsk Polytechnic University, RUSSIA ‡Dept. of Electronics and Communication Engineering, The LNM Institute of Information and Technology, Jaipur, INDIA ¶Dept. of Electronics and Communication Engineering, National Institute of Technology Tiruchirappalli, INDIA §Department of Electronic Engineering, National Ilan University, Ilan, TAIWAN (cid:107) Ecole de Technologie Superieure, Universite du Quebec, Montreal, CANADA Email: ∗†{chathurangab;dushanthaj}@sltc.ac.lk ‡[email protected],§[email protected] and [email protected] Abstract -- Strongly Coupled Magnetic Resonance (SCMR) uses electromagnetic resonance in order to efficiently transfer power wirelessly over mid-range distances. Since the energy exchange capability of resonant objects higher than non-resonant objects, strongly coupled systems are able to achieve more efficient energy transfer than other wireless power transfer systems. The paper presents detailed experimental and simulated analysis of the performance of the SCMR system. A prototype of the SCMR system was implemented and experiments were conducted to analyze the performance of the system. Finally, the resonant frequency of the system was experimentally verified and the factors influencing the wireless power transfer were also studied. Keywords: Wireless Power Transfer (WPT) and Inductive Cou- pling, Strongly Coupled Magnetic Resonance (SCMR) I. INTRODUCTION The concept of wireless power transfer (WPT) was invented to the world by Sir Nikola Tesla in 1899 [1],[2]. After that, many scientists have conducted various research towards the devel- opment of schemes for transferring energy over long distances without any carrier medium. As a result, a few methods were proposed to achieve WPT such as inductive coupling, resonant inductive coupling, Microwave and laser power transmission [3]-[5]. However, conventional WPT methods were unable to achieve higher efficiencies at modest distances [6]. The latest WPT technique called as Strongly Coupled Mag- netic Resonance (SCMR) was introduced to the world in the recent past [6]-[9]. This new scheme is capable of transferring energy efficiently over mid-range distance based on electro- magnetic resonance. Strongly coupled systems are able to achieve efficient energy transfer because energy the exchange capability of resonant objects surpass non-resonant objects [10], [11]. Since the four- coil system of SCMR can achieve higher efficiencies and higher quality factor compared with the traditional two coil system usually used for the inductive coupling, the detrimental effects of low coupling coefficients can be counteracted. In [7], it was possible to experimentally demonstrate efficient non-radiative power transfer over distances up to 8 times the radius of the coils. The researchers were able to transfer 60 watts with 40% efficiency over distances in excess of 2 meters. However, the magnitudes of electrical and magnetic fields of this system were above thresholds specified by general safety regulations. Kim Ean and others investigated the performance of the magnetic resonance power transfer scheme using band- pass filter models in [12]. In this paper, the authors proposed a new approach for a multi- receiver power transfer system using the band-pass filter model and the impedance matching technique to power multiple loads. The physical limitations and the radiation effect of Near- Field Coupling wireless power transfer systems were studied in [13]. Optimizing the load impedance and the distance between the two devices to improve power transfer efficiency is also addressed in this work. However, most of the research on strongly coupled magnetic resonance is still at the simulation stage. As a result, there are only a few experimental results available on the performance of the SCMR system. Then, we conducted this research in order to analyze the performance of the SCMR system. First, we simulated SCMR system model using the MATLAB simulation software to get the optimal loop and helix structure. Then we conducted an experiment to verify the resonance frequency of the system. The next experiment was conducted to obtain the optimal distance between transmitter and receiver at the resonant frequency. Finally, the factors influencing WPT were studied. In this paper, background and problem statement is described in section II. Wireless power transfer using strongly coupled magnetic resonance is described in Section III. The experi- mental setup is described in section IV. Experimental results and analysis are included in section V. Finally, our conclusions are drawn in Section VI. II. THEORY AND DESIGN OF THE STRONGLY COUPLED MAGNETIC RESONANCE A. Theoretical Aspects of the SCMR System Two resonant objects with the same resonant frequency tend to efficiently exchange power while dissipating relatively small amount of energy in outward off-resonant objects [9]. Nor- mally,there is a general strongly coupled regime of operation in a coupled resonance system. In such systems, the energy transfer can be maximized by operating that system in the strongly coupled regime. The most significant interaction for energy transfer mainly occurs between the transmitting and receiving coils. So, the efficiency of the system is determined by the separation between these two coils. The input impedance looking into the coupled coils Zin is a function of the mutual coupling between transmitting and receiving coils where the output impedance Zout is a function of the transferred power. For analyzing SCMR, both coupled mode theory (CMT) [14]-[16] and circuit theory (CT) [17], [18] can be applied. However, the coupled mode theory is highly complex and it requires long theoretical analyzes. The most of the research are conducted using the circuit theory because it is comparatively simpler than the coupled mode theory. So, we also used the circuit theory in our analysis. In order to electrify the power coil, the voltage source VS with an internal resistance of RS is used. An oscillating magnetic field is resulted by the I1 current flowing through the L1 inductor. A portion of these magnetic field lines passing through the other coils to produce flux linkages. According to the Neumann formula, the mutual inductance (Mij) between the coil i and j may be expressed as below, Mij ≈ πµNiNjri 2 + rj 2[dij 2 2rj 3 2 2] here i,j = 1,..,4 (1) Here, N is the number of turns in a coil, d is coupling distance between the coils and r is the radius of the coil . Since, all the coils in the system are mutually coupled to each other as shown in Fig. 2,the fraction of flux linking a coil due to the flow of current in nearby coils can be determined in term of coupling factor. Coupling factor is generally denoted by letter k and can be calculated from the equation 2. Mij(cid:112)LiLj kij = The resonant circuit is made up of lumped elements. The com- pensation capacitance at the power coil is made independent of the load by this structure and it helps to retain the resonance during fluctuation of load. The cross-coupling parameters k13, k24 and k14 are omitted to make the analysis straightforward. In order to establish strong power transmission with high Q- factor, L1 << L2, L4 << L3, k23 << k12 and k23 << k34 conditions should be met. The following matrix can be derived by applying Kirchhoff's circuit laws for each and every loop.  =   VS 0 0 0 Z1 jω0M12 0 0 jω0M12 −jω0M23 Z2 0 0 −jω0M23 0 0 Z3 jω0M34 jω0M34 Z4 (2)   I1 I2 I3 I4  Fig. 1. Schematic Diagram of SCMR System Basically, SCMR approach consists of four elements, namely the source element, transmitting and receiving elements and the load element as shown in Fig. 1. The source element is connected to the power source of the system and it is inductively coupled with the Transmitting element. Since both Transmitting and Receiving loops of the system are designed to resonate at the resonance frequency of the system, the quality factor (Q) between the two loops is maximized.This guarantee that the energy exchange between these loops is maximal while the loops are connected together via electro- magnetic resonance coupling. The system can be represented by an equivalent circuit diagram as shown in Fig. 2. Fig. 2. The Equivalent Circuit of SCMR System If Qi = ω0Li/Ri, using the above matrix, voltage as shown in equation 3 at the bottom of function (VL/VS) transfer can the system obtained be the page. S21 ≈ 1 + k12 2Q1Q2 + k23 2k12k23k34Q2Q3 2Q2Q3 + k34 2Q3Q4 + k12 √ Q1Q4 2k2 34Q1Q2Q3Q4 (3) The transmission efficiency of the wireless power transfer link can be shown as, η = Pout Pin = 2(cid:16) RS 2RL I4 4 I1 (cid:17) = S212 (4) Above equations show that to maximize the efficiency of the SCMR system, the Q-factor should be maximized. B. Transceiver Design Power transfer efficiency is the critical factor determine the performance of the resonant power transfer link. It is deter- mined by the self-inductance of the coil, quality factor (Q- factor) and the resonance frequency. Here, we present the fundamental theory for designing a SCMR system that uses loops and helices as the transmitting and receiving elements. In coupling based wireless power transfer systems the, helices are usually flavored as transceiver elements due to self-self- tuning capabilities of these elements. Because, helices consist of distributed inductance and capacitance and therefore they avoid the use of external capacitors to get a desired resonant frequency (fr). Normally, we can represent transmitting and receiving ele- ments of the WPT system using a series RLC circuit and according to the equivalent circuit resonance frequency of the SCR system can be expressed by using Equation 5. fr = √ 1 LC 2π (5) Here, resonators have a resonant frequency fr, and it is also the operational frequency of the SCMR wireless power transfer system.L and C are the inductance and the capacitance of the circuit respectively. The quality factor (Q-factor) of an ideal series RLC circuit is given by equation 6. Q = ωrL R = 2πfrL R (6) Consequently, the Q-factor of a resonator of SCMR system (i.e., self-resonant) can be expressed as, Q = 2πfrL Rohm + Rrad (7) Here L, Rrad and Rohm are the self-inductance, radiation resistance and ohmic resistance of the helix resonator respec- tively and which can be given by, L = µ0rN 2(cid:104) ln rc (cid:17) − 2 (cid:16) 8r (cid:105) η0N 2(cid:16) 2πfrr (cid:17)4 (cid:16)(cid:112)µ0ρπfr (cid:17) Nr c π 6 rc Rrad = Rohm = (8) (9) (10) Where µ0 is the permeability of free space, ρ is the material resistivity of the helix coil, r is the radius of the helix, rc is the cross-sectional wire radius, N is the number of turns, f is the frequency, η0 is the impedance of free space and c is the speed of light. Finally, expression for the Q-factor of a resonant helix can found by inserting the expressions for the inductance and resistances of the helix (Equation 8-10) into the equation 7 and the resulting equation can be written as, 2πfrµ0rN 2(cid:104) (cid:16) µ0ρπfrr2N (cid:17) 1 2 (cid:16) 8r rc ln + 20π2N 2 r2 c (cid:17) − 2 (cid:105) (cid:16) 2πfrr c (cid:17)4 (11) Q(fr, r, rc, N ) = III. EXPERIMENTAL SETUP The experimental setup of the WPT system is illustrated in Fig. 3. The system consists of four major components. Those are, signal generator circuit, amplifier module, transducers and the class E rectifier circuit. There is an AD9850 DDS signal generator module which is operated through Arduino. In order to increase the efficiency of the WPT system, amplifier module and a class E rectifier circuit are used. There are four coils implemented in the transducer system. Those are source coil, load coil, Tx coil and Rx coil. The Tx and Rx coils have outer radius, r of 16 cm and eight turns each. The cross-section coil radius, rc of all four types of coils is 3 mm.Our experimental setup was designed using the key parameters given in Table I. Fig. 3. The Experimental Setup of SCMR System PARAMETERS OF THE EXPERIMENTAL SETUP TABLE I system significantly reduces when the operating frequency of the system vary from the resonance frequency . Symbol L1, L4 R1, R4 L2, L3 C2, C3 R2, R3 fr Note Source / Load Coil Inductance Source / Load Coil Self Resistance Resonant Coil Inductance Resonant Coil Capacitance Resonant Coil Self Resistance Resonant Frequency Value 1 µH 0.02 Ω 24 µH 1048 µF 0.04 Ω 12 M Hz IV. EXPERIMENTAL RESULTS AND ANALYSIS A. Optimal Loop and Helix Structure Simulation for the SCMR In order to show that the impact of the design of the optimal loop on the efficiency of the SCMR system, we utilized simulations using MATLAB software. We were able to study that the Q-factor increases when the number of turns are increased. But, the frequency which gives the maximum Q-factor does not depend on it. We were also able to find out that the cross-sectional wire radius also affects the Q-factor in the same manner as number of turns. We studied that the resonant frequency which gives the maximum Q-factor is highly dependent on the radius of the coil as shown in Fig. 4. Fig. 5. Output Voltage Vs Frequency for SCMR System C. Optimal Distance at the Resonant Frequency In inductive coupling voltage gets reduced when the separation between the resonant loops is increased. But, in SCMR system, the voltage does not only depend on only on the distance. [26]. According to experimental results, the optimal distance at the resonant frequency is 36 cm. Fig. 4. The Q-Factor Vs Frequency for Difference Radius of the Loop Fig. 6. Output Voltage Vs Distance for SCMR System B. Experimentally Verifying the Resonant Frequency of the SCMR System D. Factors Influencing the WPT In order to experimentally verify the resonant frequency of the SCMR system, we observed the variation of the output voltage with the frequency as shown in Fig. 5. According to our experimental results, the resonant frequency of our system is approximately 12 MHz. Output voltage vs frequency curve of the resonant network has two hills and one valley. This phenomena is known as frequency splitting and it is further discussed under concerns of WPT in this paper. It can be clearly observed that the efficiency of the SCMR There are some problems present WPT and it is an interesting subject for research. Even though the operating principle of the WPT is well known, there are some aspects which are not fully explained and some components of the system for which improved solutions should be found out [19]. 1) Influence of Harmonics: Here, we study the influence of harmonics introduced by the source on transfer performance. First, we analyze the output voltage of the amplifier and its harmonic content. Fig. 7 shows the harmonic spectrum of the output voltage of the amplifier. One can notice that the fundamental harmonic has the highest value, but both odd and even order harmonics bring an important contribution to the output voltage of the amplifier. Fig. 7. Harmonic Spectrum of the Output Voltage of Amplifier As a result of the band-pass characteristic present in the resonant network, SCMR attenuates the other frequency com- ponents while passing through fundamental frequency com- ponents [20]. This operation is illustrated in Fig. 8. For our experimental setup the resonance frequency is selected as 12 MHz.With the aim of studying the effects of the harmonics and to verify the transducer setup, we transmitted a signal having half of the resonant frequency, i.e. 6 MHz. In this scenario, we observed that the second harmonic component of the transmitted signal, i.e. 12 MHz is predominant. Then, we can conclude that the transducers are adjusted to resonate at 12 MHz. of the WPT. In the splitting region, the resonant frequency separates into two different resonant frequencies. The cir- cuit model can be used to comparatively analyze frequency splitting in magnetically coupled wireless transfer systems. According to research, it was found out that frequency splitting occurs in the voltage gain and in the output power [21]. In [22]-[24], the factors such as the internal resistance of the source and mutual inductance between coils which are related to the frequency-splitting phenomenon were studied through theoretical calculations and experiments .It was found that reducing the source internal resistance, and increasing the mutual inductance of the source coil and the sending coil, as well as the mutual inductance of the load coil and the receiving coil, all help relieve the frequency-splitting phenomenon and improve the efficiency. There are two types of solutions available for the frequency-splitting phenomenon. Those are, tuning the frequency at close distances and adding an impedance matching network [6], [25]. V. CONCLUSION The detailed simulated and experimental study of SCMR in this paper demonstrated that by carefully designing the geometry, distance, size and properties of the transceiver loops, efficiency and distance of transmission in a WPT system can be enhanced. We used simulated mathematical results in order to optimize these parameters. The simulation results showed that that the resonant frequency which gives the maximum Q- factor is highly dependent on the radius of the coil. In this paper, the harmonics introduced by the transmitter elements were studied and we concluded that the harmonics get attenu- ated as a result of SCMR system acting as a band-pass filter. Frequency splitting phenomena were experimentally observed. In practice, a certain rated power is required, but some of the electronic components may struggle with the operation in MHz resonant frequency, for example, semiconductor devices and so on. Finally, we concluded that an efficient wireless power transfer system can be successfully implemented using opti- mum parameters of the strongly coupled magnetic resonance. REFERENCES [1] N. Tesla, Apparatus for transmitting electrical energy, US patent number 1,119,732, issued in December 1914. [2] Tesla, Nikola.,"The transmission of electrical energy without wires as a means for furthering peace", Electrical World and Engineer 1 (1905): 21-21. [3] H. Hu, K. Bao, J. Gibson, and S. V. Georgakopoulos, Printable and con- formal strongly coupled magnetic resonant systems for wireless powering, in WAMICON 2014. IEEE, 2014, pp. 14. Fig. 8. Harmonic Spectrum of the Output Voltage at the Receiving End 2) Frequency Splitting Analysis: Frequency splitting has been widely observed in SCMR which is also a key characteristic [4] A. Rajaram, D. N. K. Jayakody, K. Srinivasan, B. Chen, and V. Sharma,Opportunistic-harvesting: Rf wireless power transfer scheme for multipleaccess relays system,IEEE Access, vol. 5, pp. 1608416099, 2017. [5] D. N. K. Jayakody, J. Thompson, S. Chatzinotas, and S. Durrani,Wireless Information and Power Transfer: A New Paradigm for Green Communi- cations. Springer, 2017. Resonance Coupling Wireless Power Transmission", 2017. [12] Koh Kim Ean, Beh Teck Chuan and T. Imura, Y. Hori, "Novel band- pass filter model for multi-receiver wireless power transfer via magnetic resonance coupling and power division, 2012 IEEE 13th Annual Wireless and Microwave Technology Conference (WAMICON), vol., no., pp.1-6, 15-17 April 2012. [13] Lee Jaechun and Nam Sangwook,"Fundamental Aspects of Near-Field Coupling Small Antennas for Wireless Power Transfer, IEEE Transactions on Antennas and Propagation, vol.58, no.11, pp.3442-3449, Nov. 2010. [14] R. Sedwick, "Long range inductive power transfer with superconducting oscillators", Annals of Physics, vol. 325, no. 2, pp. 287-299, 2010. [6] A. P. Sample, D. T. Meyer, and J. R. Smith, "Analysis, experimental results, and range adaptation of magnetically coupled resonators for wireless power transfer, IEEE Trans. Ind. Electron., vol. 317, no. 5834, pp. 83-86, July 2007. [7] A. Karalis, J. D. Joannopoulos, and M. Soljacic, "Efficient wireless non- radiative mid-range energy transfer, Ann. Phys., vol. 323, pp. 3448, Jan. 2008. [8] A. Rajaram, Rabia Khan, T. Selvakumar , Dushantha Nalin K. Jayakody, R. Dinis and S. Panic "Novel SWIPT Schemes for 5G Wireless Networks, MDPI Sensors, 19(5), 1169, 2019. [9] A. Kurs, A. Karalis, R. Moffatt, J. Joannopoulos, P. Fisher, and M. Sol- jacic, "Wireless power transfer via strongly coupled magnetic resonances, Science, vol. 317, no. 5834, pp. 83-86, July 2007. [10] Dumitriu, D. Niculae, M. Iordache, L. Mandache, and G. Zainea, Onwireless power transfer, in2012 International Conference on Applied andTheoretical Electricity (ICATE). IEEE, 2012, pp. 16 [11] C. Li, H. Zhang and X. Jiang, "Parameters Optimization for Magnetic [15] N. Yin, G. Xu, Q. Yang, J. Zhao, X. Yang, J. Jin, W. Fu and M. Sun, "Analysis of Wireless Energy Transmission for Implantable Device Based on Coupled Magnetic Resonance", IEEE Transactions on Magnetics, vol. 48, no. 2, pp. 723-726, 2012. [16] F. Zhang, S. Hackworth, W. Fu, C. Li, Z. Mao and M. Sun, "Relay Effect of Wireless Power Transfer Using Strongly Coupled Magnetic Resonances", IEEE Transactions on Magnetics, vol. 47, no. 5, pp. 1478- 1481, 2011. [17] S. Barman, A. Reza, N. Kumar and T. Anowar, "Two-side Impedance Matching for Maximum Wireless Power Transmission", IETE Journal of Research, vol. 62, no. 4, pp. 532-539, 2015. [18] M. Kiani, Uei-Ming Jow and M. Ghovanloo, "Design and Optimization of a 3-Coil Inductive Link for Efficient Wireless Power Transmission", IEEE Transactions on Biomedical Circuits and Systems, vol. 5, no. 6, pp. 579-591, 2011. [19] I. Sirbu, "The influence of the frequency on the efficiency and on the power quality of a contactless power transfer system", 2015 IEEE 15th International Conference on Environment and Electrical Engineering (EEEIC), 2015. [20] H. Zeng, S. Yang and F. Peng, "Design Consideration and Comparison of Wireless Power Transfer via Harmonic Current for PHEV and EV Wireless Charging", IEEE Transactions on Power Electronics, vol. 32, no. 8, pp. 5943-5952, 2017. [21] L. Jianyu, T. Houjun and G. Xin, "Frequency Splitting Analysis of Wireless Power Transfer System Based on T-type Transformer Model", Electronics and Electrical Engineering, vol. 19, no. 10, 2013. [22] Y. Zhang, Z. Zhao and K. Chen, "Frequency-Splitting Analysis of Four- Coil Resonant Wireless Power Transfer", IEEE Transactions on Industry Applications, vol. 50, no. 4, pp. 2436-2445, 2014. [23] L. Jianyu, T. Houjun and G. Xin, "Frequency Splitting Analysis of Wireless Power Transfer System Based on T-type Transformer Model", Electronics and Electrical Engineering, vol. 19, no. 10, 2013. [24] X. Zhang, C. Xue and J. Lin, ""Distance-Insensitive Wireless Power Transfer Using Mixed Electric and Magnetic Coupling for Frequency Splitting Suppression", IEEE Transactions on Microwave Theory and Techniques, pp. 1-10, 2017. [25] C. B. Teck, T. Imura, M. Kato, and Y. Hori,"Basic study of improving efficiency of wireless power transfer via magnetic resonance coupling based on impedance matching, in Proc. IEEE Int. Symp. Ind. Electron., pp. 20112016, 2010. [26] Jonah, Olutola, "Optimization of Wireless Power Transfer via Magnetic Resonance in Different Media" (2013), FIU Electronic Thesesand Dis- sertations, 876.
1902.00328	1	1902	2019-02-01T13:56:42	High piezoelectric sensitivity and hydrostatic figures of merit in unidirectional porous ferroelectric ceramics fabricated by freeze casting	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	High performance lead zirconate titanate (PZT) ceramics with aligned porosity for sensing applications were fabricated by an ice-templating method. To demonstrate the enhanced properties of these materials and their potential for sensor and hydrophone applications, the piezoelectric voltage constants hydrostatic parameters and AC conductivity as a function of the porosity in directions both parallel and perpendicular to the freezing temperature gradient were studied. As the porosity level was increased, PZT poled parallel to the freezing direction exhibited the highest coefficients, and hydrostatic figures of merit compared to the dense and PZT poled perpendicular to the freezing direction. This work demonstrates that piezoelectric ceramics produced with aligned pores by freeze casting are a promising candidate for a range of sensor applications and the polarisation orientation relative to the freezing direction can be used to tailor the microstructure and optimise sensitivity for sensor and hydrostatic transducer applications.	physics.app-ph	physics	High piezoelectric sensitivity and hydrostatic figures of merit in unidirectional porous ferroelectric ceramics fabricated by freeze casting Yan Zhang, James Roscow, Mengying Xie, Chris Bowen Department of Mechanical Engineering, University of Bath, BA2 7AY, United Kingdom High performance lead zirconate titanate (PZT) ceramics with aligned porosity for sensing applications were fabricated by an ice-templating method. To demonstrate the enhanced properties of these materials and their potential for sensor and hydrophone applications, the piezoelectric voltage constants (g33 and g31), hydrostatic parameters (dh, gh, -d33/d31, dh·gh and dh·gh/tanδ) and AC conductivity as a function of the porosity in directions both parallel and perpendicular to the freezing temperature gradient were studied. As the porosity level was increased, PZT poled parallel to the freezing direction exhibited the highest dh, -d33/d31 and figures of merit dh·gh, dh·gh/tanδ compared to the dense and PZT poled perpendicular to the freezing direction. The gh, g33 and g31 coefficients were highest for the PZT poled perpendicular to the freezing direction; the gh was 150% to 850% times higher than dense PZT, and was attributed to the high piezoelectric activity and reduced permittivity in this orientation. This work demonstrates that piezoelectric ceramics produced with aligned pores by freeze casting are a promising candidate for a range of sensor applications and the polarisation orientation relative to the freezing direction can be used to tailor the microstructure and optimise sensitivity for sensor and hydrostatic transducer applications. Introduction Piezoelectric materials represent a popular class of active materials used in many areas[1-3], such as SONAR applications, vibration energy harvesting, structural health monitoring and non-destructive evaluation. For uniaxial sensing applications, the piezoelectric voltage constants g33 and g31 are important parameters since they represent the electric field produced per unit stress, and are of interest for accelerometers, force, pressure and acoustic sensors. Hydrophones that operate under hydrostatic conditions are also an important category of piezoelectric transducers, which are employed to detect acoustic signals in water by converting the mechanical vibrations of low frequency acoustic waves into an electrical signal[4]. For such applications, the important parameters are the hydrostatic charge (dh) coefficient, voltage (gh) coefficient, and hydrostatic figure of merit (FoM1=dh·gh), which define the actuating (transmit) capability of the material, sensitivity of the hydrophone, and the suitability for underwater sonar applications, respectively[5, 6]. At frequencies far below the resonance frequency, 1 energy dissipation is mainly dominated by the dielectric loss (tan δ), thus an alternative figure of merit of FoM2=dh·gh/tanδ has also been proposed[7, 8]. The hydrostatic figures of merit can be calculated from measured piezoelectric and dielectric properties as follows: dh=d33+2d31, gh=dh /𝜀!!!𝜀!, where, d33 and d31 are the longitudinal and transverse piezoelectric charge coefficients, 𝜀!!! is the relative permittivity at constant stress and 𝜀! is the hydrostatic performance are a high hydrostatic charge coefficient (dh) and low permittivity (𝜀!!!). For permittivity of the free space. These equations indicate that the important requirements for improved many dense ferroelectric ceramic materials d33 ≈ -2d31 which leads to a low dh, and when combined with the high permittivity of dense ferroelectrics this leads to dense materials exhibiting a low dh, gh, and dh·gh, thereby limiting their performance as transducers under hydrostatic conditions. For uniaxial sensor applications, the combination of a high piezoelectric charge coefficient and low permittivity is also beneficial since g33=d33 /𝜀!!!𝜀! and g31=d31 /𝜀!!!𝜀!. To date, a number of researchers have made significant efforts to reduce the permittivity of the sensor material by introducing porosity into the dense material[9-12]. However, the compromise between the mechanical strength and the volume fraction and type of porosity remains a limiting issue, especially for those porous ceramics with randomly distributed porosity, which is typically achieved by a traditional processing route of adding a polymeric pore-forming agent that burns out during solid-state sintering. Furthermore, the production of aligned piezoelectric structures has recently attracted considerable interest due to their ability to reduce the resistance of piezoelectric domain motion under an applied the electric field, where alignment has been explored using nanowires[13, 14], nanofibers[15], and nanopores[16]. In this paper, we exploit the inspiration drawn from the high strength of natural nacre with a layered microstructure[17], and the ability of freeze casting, also called ice-templating[18-20] as an effective way to mimic the structure of nacre by building an oriented ceramic structure within a unidirectional temperature gradient, with the realisation of anisotropic properties in directions parallel and perpendicular to the temperature gradient/pore channel. To date, there have been a number of studies on freeze-cast piezoelectric ceramics that have utilised camphene-based [21-23], tert-butyl alcohol (TBA)-based [24-34] suspensions to achieve 3-1, 3-2 and 3-3 connectivity piezoceramic-based composites, and water-based [34-40] suspensions for 2-2 connectivity composites. The majority of the investigations above were focused on the properties of the freeze-cast parallel to the temperature gradient/pore channel, and there is little work on comparing the piezoelectric properties of freeze-cast ceramics both parallel and perpendicular to the temperature gradient/pore channel. TBA-based 0.94Bi0.5Na0.5TiO3-0.06BaTiO3 [30] and K0.5Na0.5NbO3 [31] suspensions were used to achieve an aligned porous ceramic by freeze casting, whose piezoelectric constants (d33 and g33) and strain were examined, with differences between parallel and perpendicular orientations while the hydrostatic properties were evaluated in freeze casting camphene-based lead zirconate titanate-lead zinc niobate suspension[22]. However, camphene is potentially flammable and has been demonstrated 2 to exhibit higher toxicity than water [41], while TBA is not only a flammable, toxic and carcinogenic substance[42, 43], but is also an emerging environmental contaminant[44]. Moreover, laminated 2-2 connectivity based piezoelectric and pyroelectric composites are of interest due to their simple architecture and superior actuation and sensing abilities[45-47]. Therefore, using water as the solvent in the freeze casting process would be a more environmentally friendly choice with lower processing cost. In addition, our previous work[48, 49] has demonstrated that porous ceramics with an aligned porous structure formed by freeze casting water-based suspension exhibited an improved mechanical strength compared to traditional porous ceramic and also lead to a significantly reduced permittivity and heat capacity compared to that of the dense material. Since the aligned structure maintained high pyroelectric properties parallel to the freezing direction a higher charging voltage and energy was achieved when charging a capacitor via the pyroelectric effect for energy harvesting applications. However, apart from our previous research on energy harvesting applications [48], there have been no reports on the piezoelectric properties by freeze casting water-based piezoelectric suspensions and their assessment in both parallel and perpendicular directions for sensor and hydrostatic applications. Therefore, this paper provides a first insight into the properties of porous piezoelectric ceramic with aligned porosity by water based freeze casting for sensor and hydrostatic applications. The piezoelectric voltage constants (g33 and g31), hydrostatic parameters (dh, gh, -d33/d31, dh·gh and dh·gh/tanδ) and AC conductivity are assessed when the material was poled parallel and perpendicular directions to the freezing temperature gradient using a simple water-based suspension. Significant benefits will be demonstrated for many of the properties compared to the monolithic (dense) material, depending on the polarisation direction. Methods The raw materials used for water-based freeze casting process to fabricate the aligned porous PZT were reported in our previous work on energy harvesting[48]. Figure 1 shows images of the PZT powders before and after ball milling used in this work. It can be seen that the original as-received PZT powders exhibited a sphere-like morphology due to the spray drying processing, whose agglomerate particle diameter was approximately 40 µm, as shown in Fig. 1(A). According to our previous research[50, 51], the ceramic particle size has a strong influence on the rheological properties of the suspension, which is crucial not only for the stability of the suspension for pore preparation by freeze casting, but also the pore size of the porous channel and the final properties. Normally, the particle size of the ceramic[52] in the suspension should range from submicron to less than 3 µm in order to exhibit suitable stability and viscosity for freeze casting[53-56]. Therefore, in order to improve the rheological properties of the suspension for achieving the defect-free ceramic layer by freeze casting, a high-speed ball milling treatment in ethanol was utilised, and the particle 3 size of the PZT powders with uniform shape were reduced to approximately 1 µm on average in Fig. 1(B). Figure 1 SEM images of PZT powders (A) without ball milling ('as-received'), (B) after ball milling for 48h. A schematic of the freeze casting process is shown in Fig. 2. The PZT suspensions with the solid load levels of 67, 58, 48.5, 35.5, 22.5 vol.% were ball-milled for 24 h in zirconia media to generate homogeneous and fine suspensions. The prepared suspensions were de-aired (Fig. 2 (A)) before casting into a cylindrical polydimethylsiloxane (PDMS) mould with four cylinder shaped chambers (shown in Fig. S1) which was transported to a conducting cold plate in a liquid nitrogen container for the freeze casting process (Fig. 2 (B)); in this figure the freezing direction is from the base and upwards which leads to the structure shown in the image. After freezing, the frozen bodies were demoulded and freeze-dried in a vacuum at -50 °C to sublimate the ice crystal (Fig. 2 (C)) to form a 2- 2-type connectivity with the PZT material aligned in the freezing direction, but more randomly orientated perpendicular to the freezing direction. To compare with the freeze cast ceramics, dense PZT pellets were formed with the initial diameter of 10 mm and thickness of 1.2 mm by uniaxial hydraulic pressing. Finally, the porous (freeze cast) and dense (pressed) green bodies were sintered at 1250 °C for 2 h under a PbO-rich after organic additives burnt out at 600°C for 3 h. Each porous PZT cylinder was poled parallel and perpendicular to the freezing direction, denoted as PZT║ and PZT┴, respectively for the following microstructure and assessment of piezoelectric properties. A schematic showing the freezing, cutting, poling directions and SEM viewing orientation are also summarised in Fig. S2. The microstructures of the powders and sliced PZT ceramics were examined by a scanning electron microscopy (SEM, JSM6480LV, Tokyo, Japan). The apparent porosity was derived from the density data obtained by the Archimedes method with the error of ±1.5 vol.%. Each porosity was labelled with the integer value. The remnant polarization and coercive field of the ceramics were measured using a Radiant RT66B-HVi Ferroelectric Test System on initially unpoled materials. The longitudinal piezoelectric strain coefficient (d33) and the transverse piezoelectric strain coefficient (d31) were measured using a Berlincourt Piezometer (PM25, Take Control, UK) after corona poling by applying a DC voltage of 14 kV for 15 min at 120 °C. The AC conductivity, σ, of the sintered 4 ceramics were carried out from 0.1Hz to 1MHz at room temperature using an impedance analyzer (Solartron 1260, Hampshire, UK) and calculated from equation (1)[57], σ = !!"!!"!∙!! (1) !! where 𝑍! and 𝑍"are the real and imaginary parts of the impedance, A is the area of the sample and t is the sample thickness. Figure 2 Schematic of porous PZT preparation by freeze casting. (A) water-based PZT suspension in each cylindrical chamber of the mould, (B) freezing of the water from the cold base, (C) freeze drying leading to a layered PZT structure. The freezing direction is indicated (form bottom to top) along with the polarisation direction for samples poled along freezing direction (PZT║) and perpendicular to freezing direction (PZT┴). Results and discussion Figure 3(A) and (B) show SEM micrographs of the porous PZT poled parallel and perpendicular to the freezing direction, respectively. A dense lamellar ceramic wall can be seen in Fig. S3, which is desirable for high piezoelectric activity. From the macro-scale point of view, the microstructure on the top face as shown in Fig. 3(A) is equivalent to looking from above of Fig. 1(C), compared to the side face in Fig. 3(B) which is equivalent to looking from the side of Fig. 1(C). Due to the random nature of ice nucleation in the single vertical temperature gradient condition [58], there were multiple orientations on the top face of PZT║ as shown in the Fig. 3(A), while clear alignment of the PZT material with good parallelism of the lamellar ceramic layer can be readily observed in PZT┴, although part of the lamellar layers were covered by their adjacent layers in the PZT┴, as shown in the Fig. 3(B). 5 Figure 3 SEM micrographs of porous (A) PZT║ and (B) PZT┴. Figure 4 (A-E) and (F-J) show SEM micrographs of the microstructure parallel (PZT║) and perpendicular (PZT┴) to the freezing direction as the porosity fraction is increased from 20-60 vol.%, respectively. The specific porosities of 20-60 vol.% were achieved based on the results from the initial experiments that employed a ranges of solid load levels, shown in Fig. S4. With an increase of porosity, the lamellar pore size decreased while the number of the dendrites decreased accordingly in both the PZT║ and PZT┴ materials. In the 20 vol.% PZT, a large quantity of the dendrites on the lamellar surface can be found, and most of the dendrites were connected with the adjacent ceramic layer, as shown in Fig. 4 (A) and (F). These additional dendrites are beneficial not only for piezoelectric phase connection, but also for the improvement of mechanical strength. It can be seen that when the porosity reached 60 vol.%, the surface of the lamellar layers became relatively smooth with ceramic links on the edge of the layers, as shown in Fig. 4(E) and (J). In the freeze casting process, a low freezing rate can provide a longer period for ice growth, which is the replica of the lamellar pore in the SEM images, while a low solid loading can provide a low viscosity which facilitates ceramic particle rearrangement. Both of the above conditions can lead to a large lamellar pore size and the high porosity materials exhibiting a smaller number of dendrites. 6 7 Figure 4 SEM images of porous PZT║ with the porosity of (A) 20, (B) 30, (C) 40, (D) 50, and (E) 60 vol.% and PZT┴ with the porosity of (F) 20, (G) 30, (H) 40, (I) 50, and (J) 60 vol.%. Figure 5 shows the remnant polarisation (Pr) and coercive field (Ec) of the initially unpoled porous PZT as a function of the porosity ranging from 20 to 60 vol.%, respectively. It can be observed that the Pr of both PZT║ and PZT┴ decreased with an increase of porosity, which were 2.1-5.5 and 3.1-10.0 times lower than the dense PZT, respectively. The dense material is characterised as having 4 vol.% porosity, with the density of 7.2 g/cm3 based on the theoretical density of 7.5 g/cm3 from the datasheet provided by the supplier. The reduction in Pr is likely to be due to the reduced amount of polarised material and the inhomogeneous electric field distribution in the porous materials as a result of the contrast in permittivity between the high-permittivity PZT and low-permittivity air. The decrease of Pr is also associated with its connectivity (Fig. 3A), and leads to a decrease of piezoelectric properties[48], such as d33 and d31. The decrease in Pr with porosity is more rapid than predicted by 𝑃!!"#"$%=𝑃!!"#$"Í(1-p) where p is porosity [59]; this is due to the porosity also restricting polarisation of the ceramic since the electric field concentrates in the lower permittivity pore space. The PZT║ material exhibited a 1.5-1.8 times higher Pr compared to the PZT┴ due to the improved connectivity of piezo-active material along the freezing direction and therefore the lower fraction of unpoled areas in PZT║[48, 60]. The Ec values of both PZT║ and PZT┴ increased as the porosity increased, see Fig. 5B, and the PZT┴ exhibited the highest Ec values in all ranges of porosity. The increase in Ec with porosity is due to the concentration of the applied electric field in the low permittivity pore space, leading to higher applied electric field being required to achieve domain switching in the higher permittivity ferroelectric phase. For the same reason, a higher Ec is observed for the PZT┴ material as there is a reduced connection of ferroelectric along this poling direction as a result of the overlapped lamellar layers, see Fig. 3(D), resulting in a reduced piezoelectric response[61] and a higher Ec in PZT┴. The inhomogneous electric field due to the presence of pores is also reflected in the reduced rectangularity (Premnant/Psaturation) of the materials as the porosity level increases, as seen in Fig. 5(C). 8 Interestingly, the PZT║ with the porosity of 20 vol.% exhibited the lowest Ec of 7.7 kV/cm compared to both the dense (8.7 kV/cm), and all the PZT┴ materials, demonstrating easier switching of ferroelectric domains with applied electric field. The presence of a small amount of porosity (~20 vol.%) may initially provide a state of reduced internal stress or restriction of domain motion, while at higher porosity levels the applied electric field concentrates in the pore volume and leads to higher applied electric fields being required to provide domain switching. Therefore, although the existence of porosity can facilitate the switching of the domain walls to some extent[62, 63], the increased electric field concentration[60] in the low permittivity pore space leads to a higher Ec which reached to 8.9 µC/cm2 compared to the dense with the Ec of 8.7 µC/cm2 when the porosity was higher than 50 vol.%. In addition, due to randomness of the lamellar pores orientations perpendicular to the freezing direction in Fig. 3 (A) and Fig. 2 (C), the angle of the pore in PZT┴ ranged from >0to 90, while PZT║ had an angle of ~0 along the poling direction; see Fig. S5. This leads to a better level of poling of PZT║ [64] compared to PZT┴, leading to a higher polarisation for PZT║ [48, 64], as seen in Fig. 5(A). Figure 5 (A) Remnant polarisation (Pr), along with estimation based on 𝑃!!"#"$%=𝑃!!"#$"Í(1-p) , (B) coercive field (Ec) of the porous freeze-cast PZT, and (C) rectangularity ratio of with Pr / Ps various porosities. The dense material is also shown. 9 piezoelectric voltage coefficients (g33 and g31), relative permittivity (𝜀!!! ) and figures of merit (dhgh Figure 6 (A-F) show the anisotropy factor[65] -d33/d31, hydrostatic charge (dh), voltage coefficient (gh), and dhgh/tanδ) of the porous PZT as a function of the porosity ranging from 20 to 60 vol.% and dense PZTs. It can be seen from Fig. 6(A) that PZT┴ exhibited a lower -d33/d31 than the dense PZT at all porosities, while porous PZT║ exhibited a higher -d33/d31 and therefore higher anisotropy than the dense PZT. The -d33/d31 PZT║ increased with increasing porosity and was 1.2-2.0 times higher than that of PZT┴. The increase in piezoelectric anisotropy for PZT║ is advantageous since it leads to higher dh values that were determined from the d33 and d31, as shown in Fig. 6(B). It can be seen the PZT┴ exhibits a gradual reduction in dh with increasing porosity, due to a reduced -d33/d31 while the PZT║ exhibits an increase in dh with increasing porosity. It should be also noted in Fig. 6(B) that the dh of the PZT║ was higher than the dense PZT when the porosity exceeded 40 vol.%. The increase in dh for the porous material compared to the dense material is relatively modest, this is due to the fact that the dense material already has a relatively high degree of anisotropy with a -d33/d31 of ~3 (see Figure 6A); typically -d33/d31 is close to 2 for dense PZT based materials. While the PZT┴ materials exhibited relatively poor dh values it exhibits advantageous gh values that are 1.2 to 2.1 times and 1.5-8.5 higher than PZT║ and monolithic dense PZT, respectively; see Fig. 6(B), i.e. 40.1×10-3 Vm/N for PZT║, and 83.5 ×10-3 Vm/N for PZT┴ both at 60 vol.% porosity compared to that of dense PZT (9.0×10-3 Vm/N). This was due to the reduced relative permittivity, shown in Fig. 6(C) and (D), of the PZT┴ (𝜀!!! ~ 380 to 16) compared to PZT║ (𝜀!!! ~ 1407 to 581) and the dense material (𝜀!!! = 2158); at 1 kHz from the inset of Fig. 6(C) and (D). The gh value of the PZT║ was also 1.3-4.1 times higher than that of the dense PZT (see Fig. 6(B)) since the dense material exhibited a much higher permittivity, Fig. 6(C). A similar trend is observed on examination of the piezoelectric voltage coefficients (g33 and g31) in Fig. 6 (E) where both g33 and g31 of PZT┴ were 1.8-5.2 and 2.0-10.0 times higher than PZT║, and also 2.3-12.5 and 2.5-14.7 times higher than dense PZT. This indicates that PZT┴ is attractive as a piezoelectric force/pressure sensor. Fig. 6(F) shows that both the dh·gh and dh·gh/tanδ figures of merit for the PZT║ increased with increasing porosity, and were much larger than both the dense material and PZT┴. Although a reduced inter-connection of the piezoelectric phase was observed with an increase of porosity (see Fig. 4), the high degree of alignment of the pore channel along the poling direction can compensate for the reduction in interconnectivity, especially in PZT║. This is due to a combination of, relatively high piezoelectric activity (Fig. 5), high piezoelectric coefficients, high anisotropy and low permittivity achieved through the introduction of the porosity. These results compare favourably with previous analysis[48] that demonstrated the higher d33· g33 piezoelectric and pyroelectric harvesting figures of merit ((pyroelectric coefficient)2/ permittivity·heat capacity) in PZT║. The highest values of hydrostatic figures of merit were achieved for PZT║ when the porosity increased to the maximum value of 60 vol.% in this work; this corresponded to hydrostatic figures of merit that 10 were 2.7-10.2 and 2.0-10.9 times higher than the dense materials and PZT┴, respectively, as shown in Fig. 6(D). The 60 vol.% was chosen as a maximum since for higher porosity levels the material will exhibit reduced mechanical properties, and an even higher coercive field; leading to a low d33 and therefore a low dh. In addition to the advantages of high mechanical strength reported previously[48, 49], the freeze-cast samples exhibited comparable or even higher hydrostatic figure-of-merit than most of other processing methods, especially PZT║ which exhibit a higher piezoelectric coefficient and lower permittivity, as shown in Table 1. It can be seen that freeze casting generally leads to high hydrostatic figures of merit compared to other methods. Very high figures of merit are reported in Table 1 for PZT-lead zirconate niobate (PZN) [22], which is due to the high porosity levels (90vol.%); although mechanical properties can be a concern at such low ceramic volume fractions. permittivity (𝜀!!! ), (E) piezoelectric voltage coefficients (g33 and g31), and (F) hydrostatic figure of merits (dhgh and Figure 6 (A) anisotropy factor of -d33/d31, (B) hydrostatic charge (dh) and voltage coefficient (gh), (C) (D) relative dhgh/tanδ) of the porous freeze-cast PZT with various porosities. The dense material is also shown. 11 Table 1 Comparison of hydrostatic parameters with different processing methods for lead zirconate titanate (PZT) or PZT- based composite. PZN = lead zircontate niobate. Connectivity Production method Composite Freeze casting BURPS (Burnout of Polymer Spheres) Ionotropic gelation process Pore-forming agent Polymer injection (PZT rods embedded in polymer) Direct-write Dice-and-fill PZT-air (water-based) PZT/PZN-air (camphene-based) Parallel Perpendicular Parallel Pore orientation of 45 to poling direction Perpendicular PZT/PZN-air (camphene-based) Parallel PZT-air (TBA-based) PZT-air (TBA-based) PZT-air (TBA-based) Parallel Parallel Parallel PZT-air (alginate/water-based) Parallel PZT-air PZT-air PZT-epoxy PZT-polymer PZT-polymer PZT-polymer PZT-cement PZT dh Volume % (pC/N) 40 ~10 ~48 31.3 ~60 41.4 43.22 54.5 59 ~60 68 206 12 406 ~350 216 ~250 - - 200 ~223 ~35 - 102 - gh (10-3 Vm/N) 40.1 83.5 396 ~330 241 ~32 - - - - ~16.4 ~42 72 ~48 dhgh (10-12 m2/N) 8.26 1 161.01 ~115.5 53.13 ~8 9.32 7.6 10.11 ~5.7 ~0.57 ~5 7.3 5 Ref. This work [22] [23] [25] [33] [32] [37] [66] [67] [10] [68] ~44 ~222 - ~5.6 [37] 45-60 60 69 ~4.01 [69] 40 ~95 ~18 ~1.71 [70, 71] ~30 40 40 <190 ~135 <100 <0.38 ~37 <0.2 <0.72 ~5 <0.02 [72] [73] [74] 2-2 3-3 3-3 1-3 3-3 3-3 3-1 3-0 3-1 3-3 3-0 3-1 3-1 1-3 2-2 1-3 2-2 Figure 7 (A) and (B) show the AC conductivity (σ) of the PZT║ and PZT┴ at frequencies ranging from 0.1 to 106 Hz at room temperature respectively. It is clear from these two figures that the AC conductivity decreased with an increase of the porosity in both PZTs and were lower than that of the dense PZT at all the porosities and the frequencies; this includes the lowest frequencies where the conductivity is becoming less frequency dependent and is approaching the DC conductivity. The PZT║ possessed a higher AC conductivity than the PZT┴ at the same porosity and frequency, e.g. 1.3- 12 1.5 times at the frequency of 1 kHz, as shown in the insets in Fig. 7 (A) and (B). This is likely to be due to the high permittivity of the PZT║ resulting from the contribution of the dielectric phase to the overall conductivity (ωεrε0)[75]. Generally, there are two well-known models for interpreting the effects of porosity on electrical conductivity, where the solid phase has a small, but finite conductivity, and the pores have a very small (almost negligible) conductivity. For a composite consisting of both PZT and air, the low frequency conductivity parallel to the poling direction σ║ (parallel connected) and PZT perpendicular to the freezing direction (series-connected) σ┴, which can be calculated as σ║ = vPZT·σpzt + vair·σair and σ┴ = !!"#·!!"# !!"#·!!"#!!!"#·!!"#, where vPZT and vair are the volume fractions of PZT and air, σpzt and σair are the electric conductivity of the dense PZT and air[76]. Although σpzt » σair, no linear relation was found between conductivity σ and the porosity, as shown in the insets in Fig. 7 (A) and (B), owing to the existence of the dendritic ceramic link between the adjacent ceramic layers shown in Fig. 4, which means the pore regions were a mixture of the PZT ceramic links and the air. Similar situations can be also found in the piezoelectric and pyroelectric properties[48]. Furthermore, along the electric field direction during the impedance measurement, much more effective interface areas between the active phase PZT and the passive phase air were formed in the porous PZT║, therefore, the better ability of the domain movement and carrier mobility were the main reasons for the higher conductivity in the porous PZT║, making it be a more suitable candidate for hydrostatic sensor applications. Figure 7 AC conductivity (σ) of the (A) PZT║ and (B) PZT┴ with various porosities. Insets in (A) and (B) are σ values of the PZT║ and PZT┴ at 1 kHz as a function of porosity, respectively. The dense material is also shown. Conclusions Freeze casting using a water based suspension has been utilised and demonstrated to be an effective method to prepare high performance porous PZT for sensor applications with unidirectional pore channels over a range of 20-60 vol. % porosity. The hydrostatic and sensor properties of PZTs poled parallel (PZT║) and perpendicular (PZT┴) to the freezing direction were examined in detail. 13 Significant improvements in the hydrostatic figures of merit were observed compared to the dense monolithic material. In terms of dh·gh and dh·gh/tanδ, the PZT║ and PZT┴ were 2.7-10.2 and 2.0-10.9 times higher than that of the dense material. These highly attractive properties are due to their relatively high d33, reduced d31 and significantly reduced permittivity. While the PZT║ exhibited the best dh, dh·gh and dh·gh/tanδ, the PZT┴ exhibited the highest gh, g33 and g31 coefficients which was attributed to the lower permittivity of the material in this orientation. In addition, the PZT║ showed 1.3-1.5 times higher AC conductivity than the PZT┴ at the frequency of 1 kHz. In addition, it is shown that the properties are improved compared to piezoelectric composite materials manufactured by more complex methods. The PZT┴ exhibited the highest gh, g33 and g31 coefficients; up to 1.5-8.5, 2.3-12.5 and 2.5-14.7 times higher compared to the dense materials; this was attributed to the low permittivity of the material in this orientation. The coercive field increased with an increase in porosity for the materials, and this was attributed to the concentration of electric field concentration in the lower permittivity pore space. This work demonstrates water-based freeze casting provides an environmentally friendly and promising route to fabrication porous piezoelectric for both uniaxial and hydrostatic sensing applications with potential for control of the coercive field of ferroelectric materials. It is also demonstrated that changing of the poling condition relative to the freezing direction can enable control and optimisation of the relevant performance figures of merit for specific applications. In addition to varying the solid load levels, the ability of the freeze casting method to adjust the pore size, the thickness of the ceramic wall and the density of the ceramic wall in the aligned pore structures by tailoring the freezing rate, freezing temperature and temperature gradient provides significant versatility for this promising processing route to create new sensor materials. Acknowledgment Dr. Y. Zhang would like to acknowledge the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant, Agreement No. 703950 (H2020-MSCA-IF- 2015-EF-703950-HEAPPs). Prof. C. R. Bowen, Dr. M. Xie and Mr J. Roscow would like to acknowledge the funding from the European Research Council under the European Union's Seventh Framework Programme (FP/2007 -- 2013)/ERC Grant Agreement No. 320963 on Novel Energy Materials, Engineering Science and Integrated Systems (NEMESIS). References [1] R. Xie, X. Wang, G. Peng, T. Xie, D. Zhang, K. Zhou, G. Huang, G. Wang, H. Wang, A modified gelcasting approach to fabricate microscale randomized 1 -- 3 piezoelectric arrays, Ceramics International 43(1) (2017) 144-148. [2] C.R. Bowen, M. Lopez-Prieto, S. Mahon, F. Lowrie, Impedance spectroscopy of piezoelectric actuators, Scripta Materialia 42(8) (2000) 813-818. 14 [3] C.R. Bowen, A. Perry, A.C.F. Lewis, H. Kara, Processing and properties of porous piezoelectric materials with high hydrostatic figures of merit, Journal of the European Ceramic Society 24(2) (2004) 541-545. [4] R.Y. Ting, A review on the development of piezoelectric composites for underwater acoustic transducer applications, Conference Record. IEEE Instrumentation and Measurement Technology Conference, 1991, pp. 410-413. [5] H. Kara, R. Ramesh, R. Stevens, C.R. Bowen, Porous PZT ceramics for receiving transducers, IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 50(3) (2003) 289-296. [6] A. Safari, Y.H. Lee, A. Halliyal, N.R. E., 0-3 piezoelectric composites prepared by coprecipitated PbTiO3 powder, American Ceramic Society Bulletin 66 (1987) 668-670. [7] H. Li, D. Deng, Carlson Thomas J, Piezoelectric materials used in underwater acoustic transducers, Sensor Letters 10(3-4) (2012) 679-697. [8] P. S., Criterion for material selection in design of bulk piezoelectric energy harvesters, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control 57(12) (2010) 2610-2012. [9] I. Katsuyoshi, M. Toshimasa, O. Seiji, N. Kunihiro, Hydrophone sensitivity of porous Pb( Zr, Ti)O3 ceramics, Japanese Journal of Applied Physics 33(9S) (1994) 5381. [10] Y.-C. Chen, S. Wu, Piezoelectric composites with 3-3 connectivity by injecting polymer for hydrostatic sensors, Ceramics International 30(1) (2004) 69-74. [11] C.N. Della, D. Shu, The performance of 1 -- 3 piezoelectric composites with a porous non- piezoelectric matrix, Acta Materialia 56(4) (2008) 754-761. [12] W.R. McCall, K. Kim, C. Heath, G. La Pierre, D.J. Sirbuly, Piezoelectric Nanoparticle -- Polymer Composite Foams, ACS Applied Materials & Interfaces 6(22) (2014) 19504-19509. [13] C. Ou, P.E. Sanchez-Jimenez, A. Datta, F.L. Boughey, R.A. Whiter, S.-L. Sahonta, S. Kar- Narayan, Template-assisted hydrothermal growth of aligned zinc oxide nanowires for piezoelectric energy harvesting applications, ACS Applied Materials & Interfaces 8(22) (2016) 13678-13683. [14] Z. Zhou, H. Tang, H.A. Sodano, Vertically Aligned Arrays of BaTiO3 Nanowires, ACS Applied Materials & Interfaces 5(22) (2013) 11894-11899. [15] J. Yan, Y.G. Jeong, High performance flexible piezoelectric nanogenerators based on BaTiO3 nanofibers in different alignment modes, ACS Applied Materials & Interfaces 8(24) (2016) 15700- 15709. [16] A. Castro, P. Ferreira, P.M. Vilarinho, Block copolymer-assisted nanopatterning of porous lead titanate thin films for advanced electronics, The Journal of Physical Chemistry C 120(20) (2016) 10961-10967. [17] A. P. Jackson, J. F. V. Vincent, R. M. Turner, The Mechanical Design of Nacre, Proceedings of the Royal Society of London. Series B. Biological Sciences 234(1277) (1988) 415-440. 15 [18] S. Flauder, R. Sajzew, F.A. Müller, Mechanical properties of porous β-Tricalcium phosphate composites prepared by ice-templating and poly(ε-caprolactone) impregnation, ACS Applied Materials & Interfaces 7(1) (2015) 845-851. [19] F. Bouville, E. Portuguez, Y. Chang, G.L. Messing, A.J. Stevenson, E. Maire, L. Courtois, S. Deville, Templated Grain Growth in Macroporous Materials, Journal of the American Ceramic Society 97(6) (2014) 1736-1742. [20] C. Stolze, T. Janoschka, S. Flauder, F.A. Müller, M.D. Hager, U.S. Schubert, Investigation of Ice-Templated Porous Electrodes for Application in Organic Batteries, ACS Applied Materials & Interfaces 8(36) (2016) 23614-23623. [21] M.-J.P. E.P.Gorzkowski, and B. A. Bender, Prototype capacitor produced by freeze tape-casting, International Symposium on Applications of Ferroelectrics, July 2011, pp.1-3, 2011. [22] S.H. Lee, S.H. Jun, H.E. Kim, Y.H. Koh, Piezoelectric properties of PZT-based ceramic with highly aligned pores, Journal of the American Ceramic Society 91(6) (2008) 1912-1915. [23] S.-H.J. S.-H. Lee, H.-E. Kim, Y.-H. Koh, Fabrication of porous PZT -- PZN piezoelectric ceramics with high hydrostatic figure of merits using camphene-based freeze casting, Journal of the American Ceramic Society 90 (2007) 2807-2813. [24] R. Guo, C.-A. Wang, A. Yang, Effects of pore size and orientation on dielectric and piezoelectric properties of 1-3 type porous PZT ceramics, Journal of the European Ceramic Society 31(4) (2011) 605-609. [25] C.-A.W. Rui Guo, and Ankun Yang, Piezoelectric properties of the 1-3 type porous lead zirconate titanate ceramics, Journal of the American Ceramic Society 94(6) (2011) 1794-1799. [26] T. Xu, C.-A. Wang, Grain Orientation and Domain Configuration in 3-1 Type Porous PZT Ceramics with Ultrahigh Piezoelectric Properties, Journal of the American Ceramic Society 98(9) (2015) 2700-2702. [27] Z. Wang, X. Shen, N.M. Han, X. Liu, Y. Wu, W. Ye, J.-K. Kim, Ultralow Electrical Percolation in Graphene Aerogel/Epoxy Composites, Chemistry of Materials 28(18) (2016) 6731-6741. [28] T. Xu, C.-A. Wang, Effect of two-step sintering on micro-honeycomb BaTiO3 ceramics prepared by freeze-casting process, Journal of the European Ceramic Society 36(10) (2016) 2647-2652. [29] C.-A.W. Tingting Xu, Cheng Wang, Synthesis and magnetoelectric effect of composites with CoFe2O4-epoxy embedded in 3 -- 1 type porous PZT ceramics, Ceramics International 41(9) (2015) 11080-11085. [30] S. Zhu, L. Cao, Z. Xiong, C. Lu, Z. Gao, Enhanced piezoelectric properties of 3-1 type porous 0.94Bi0.5Na0.5TiO3-0.06BaTiO3 ferroelectric ceramics, Journal of the European Ceramic Society 38(4) (2017) 2251-2255. [31] Manabu Fukushima, Takahiro Fujiwara, Tobias Fey, K.-i. Kakimoto, One- or two-dimensional channel structures and properties of piezoelectric composites via freeze-casting, Journal of the American Ceramic Society 100 (2017) 5400-5408. 16 [32] A. Yang, C.A. Wang, R. Guo, Y. Huang, C.W. Nan, Porous PZT ceramics with high hydrostatic figure of merit and low acoustic impedance by TBA-based gel-casting process, Journal of the American Ceramic Society 93(5) (2010) 1427-1431. [33] A. Yang, C.A. Wang, R. Guo, Y. Huang, C.W. Nan, Effects of sintering behavior on microstructure and piezoelectric properties of porous PZT ceramics, Ceramics International 36(2) (2010) 549-554. [34] Chunyu Jiang, Xiaoxiao Tian, G. Shi, K0.5Na0.5NbO3 piezoelectric ceramics and its composites fabricated from hydrothermal powders, Advances in Intelligent Systems Research, 4th International Conference on Sensors, Mechatronics and Automation (ICSMA 2016), 2016. [35] E.P. Gorzkowski, M.J. Pan, Barium titanate-polymer composites produced via directional freezing, IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 56(8) (2009) 1613- 1616. [36] Q.F. Zhou, B.P. Zhu, C.H. Hu, K.K. Shung, E.P. Gorzkowski, M.J. Pan, Novel piezoelectric ceramic-polymer aligned composites via the freeze casting method for high frequency transducer applications, Proceedings - IEEE Ultrasonics Symposium, no. 5441756 2009. [37] W. Liu, L. Lv, Y. Li, Y. Wang, J. Wang, C. Xue, Y. Dong, J. Yang, Effects of slurry composition on the properties of 3-1 type porous PZT ceramics prepared by ionotropic gelation, Ceramics International 43(8) (2017) 6542-6547. [38] G. Liu, T.W. Button, D. Zhang, Lamellar BaTiO3 and its composites fabricated by the freeze casting technique, Journal of the European Ceramic Society 34(15) (2014) 4083-4088. [39] Y.P. Pin Li, Zijing Dong, Pan Gao, Kaolinite as a suspending agent for preparation of porous BaTiO3 ceramics via freeze casting, Journal of Electronic Materials 43(2) (2014) 459 -- 464. [40] C.B. Dong Seok Kim, Ho Jin Ma, Do Kyung Kim, Enhanced dielectric permittivity of BaTiO3/epoxy resin composites by particle alignment, Ceramics International 42(6) (2016) 7141- 7147. [41] G. Benelli, M. Govindarajan, M.S. AlSalhi, S. Devanesan, F. Maggi, High toxicity of camphene and γ-elemene from Wedelia prostrata essential oil against larvae of Spodoptera litura (Lepidoptera: Noctuidae), Environmental Science and Pollution Research (2017). [42] A.E.H.K.R. J.D. Cirvello, D.R. Farnell and C. Lindamood, Toxicity and carcinogenicity of t- butyl alcohol in rats and mice following chronic exposure in drinking water, Toxicology and Industrial Health 11 (1995) 151-166. [43] M.S. T.C. Schmidt, H. Weiss and S.B. Haderlein, Microbial degradation of methyl tert-butyl ether and tert-butyl alcohol in the subsurface, Journal of Contaminant Hydrology 70 (2004) 173-203. [44] I.I. A Sgambato, B De Paola, G Bianchino, A Boninsegna, A Bergamaschi, A Pietroiusti, and A Cittadini Differential toxic effects of methyl tertiary butyl ether and tert-butanol on rat fibroblasts in vitro Toxicol Ind Health March 25 (2009) 141-151. 17 [45] H.H.S. Chang, Z. Huang, Laminate composites with enhanced pyroelectric effects for energy harvesting, Smart Materials and Structures 19(6) (2010) 065018. [46] V.Y. Topolov, A.V. Krivoruchko, C.R. Bowen, A.A. Panich, Hydrostatic Piezoelectric Coefficients of the 2 -- 2 Composite Based on [011]-poled 0.71Pb(Mg1/3Nb2/3)O3-0.29PbTiO3 Single Crystal, Ferroelectrics 400(1) (2010) 410-416. [47] J. I. Roscow, Y. Zhang, J. T. Taylor, C. R. Bowen, Porous ferroelectrics for energy harvesting applications, The European Physical Journal Special Topics. 224 (2015) 2949-2966. [48] Y. Zhang, M. Xie, J. Roscow, Y. Bao, K. Zhou, D. Zhang, C.R. Bowen, Enhanced pyroelectric and piezoelectric properties of PZT with aligned porosity for energy harvesting applications, Journal of Materials Chemistry A 5(14) (2017) 6569-6580. [49] Y. Zhang, L. Li, B. Shen, J. Zhai, Effect of orthorhombic-tetragonal phase transition on structure and piezoelectric properties of KNN-based lead-free ceramics, Dalton Transactions 44(17) (2015) 7797-7802. [50] K. Zhou, Y. Zhang, D. Zhang, X. Zhang, Z. Li, G. Liu, W.B. Tim, Porous hydroxyapatite ceramics fabricated by an ice-templating method, Scripta Materialia 64(5) (2011) 426-429. [51] Y. Zhang, K. Zhou, Y. Bao, D. Zhang, Effects of rheological properties on ice-templated porous hydroxyapatite ceramics, Materials Science and Engineering: C 33(11) (2013) 340-346. [52] A. Lasalle, C. Guizard, E. Maire, J. Adrien, S. Deville, Particle redistribution and structural defect development during ice templating, Acta Materialia 60(11) (2012) 4594-4603. [53] T. Moritz, H.-J. Richter, Ice-mould freeze casting of porous ceramic components, Journal of the European Ceramic Society 27(16) (2007) 4595-4601. [54] S. Deville, E. Saiz, A.P. Tomsia, Ice-templated porous alumina structures, Acta Materialia 55(6) (2007) 1965-1974. [55] S. Deville, Freeze-casting of Porous Biomaterials: Structure, Properties and Opportunities, Materials 3(3) (2010) 1913-1927. [56] J. Zou, Y. Zhang, R. Li, Effect of Suspension State on the Pore Structure of Freeze-Cast Ceramics, International Journal of Applied Ceramic Technology 8(2) (2011) 482-489. [57] C.R. Bowen, S. Buschhorn, V. Adamaki, Manufacture and characterization of conductor- insulator composites based on carbon nanotubes and thermally reduced graphene oxide, Pure & Applied Chemistr 86(5) (2014) 765-774. [58] H. Bai, Y. Chen, B. Delattre, A.P. Tomsia, R.O. Ritchie, Bioinspired large-scale aligned porous materials assembled with dual temperature gradients, Science Advances 1(11) (2015) e1500849. [59] K. Nagata, Effects of porosity and grain size on hysteresis loops of piezoelectric ceramics (Pb-La) (Zr-Ti)O3, Electrical Engineering in Japan 100(4) (1980) 1-8. [60] J.I. Roscow, R.W.C. Lewis, J. Taylor, C.R. Bowen, Modelling and fabrication of porous sandwich layer barium titanate with improved piezoelectric energy harvesting figures of merit, Acta Materialia 128 (2017) 207-217. 18 [61] C.R. Bowen, H. Kara, Pore anisotropy in 3 -- 3 piezoelectric composites, Materials Chemistry and Physics 75(1 -- 3) (2002) 45-49. [62] A.N. Rybyanets, Porous piezoceramics: theory, technology, and properties, IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 58(7) (2011) 1492-1507. [63] J. Tan, Z. Li, Microstructures, dielectric and piezoelectric properties of unannealed and annealed porous 0.36BiScO3-0.64PbTiO3 ceramics, Journal of Materials Science 51(11) (2016) 5092-5103. [64] J. I. Roscow, Y. Zhang, M. J. Kraśny, R. W. C. Lewis, J. T. Taylor, C. R. Bowen, Freeze cast porous barium titanate for enhanced piezoelectric energy harvesting, Journal of Physics D: Applied Physics (2018). [65] C.R. Bowen, V.Y. Topolov, Piezoelectric sensitivity of PbTiO3-based ceramic/polymer composites with 0 -- 3 and 3 -- 3 connectivity, Acta Materialia 51(17) (2003) 4965-4976. [66] T. Zeng, X.L. Dong, H. Chen, Y.L. Wang, The effects of sintering behavior on piezoelectric properties of porous PZT ceramics for hydrophone application, Materials Science and Engineering: B 131(1 -- 3) (2006) 181-185. [67] T. Zeng, X. Dong, S. Chen, H. Yang, Processing and piezoelectric properties of porous PZT ceramics, Ceramics International 33(3) (2007) 395-399. [68] H. Chen, X. Dong, T. Zeng, Z. Zhou, H. Yang, The mechanical and electric properties of infiltrated PZT/polymer composites, Ceramics International 33(7) (2007) 1369-1374. [69] K.A. Klicker, W.A. Schulze, J.V. Biggers, Piezoelectric Composites with 3 -- 1 Connectivity and a Foamed Polyurethane Matrix, Journal of the American Ceramic Society 65(12) (1982) C-208-C-210. [70] R.E. Newnham, L.J. Bowen, K.A. Klicker, L.E. Cross, Composite piezoelectric transducers, Materials & Design 2(2) (1980) 93-106. [71] K.A. Klicker, J.V. Biggers, R.E. Newnham, Composites of PZT and Epoxy for Hydrostatic Transducer Applications, Journal of the American Ceramic Society 64(1) (1981) 5-9. [72] J. Sun, E.K. Akdoğan, M. Vittadello, A. Safari, Development of novel (2-2) piezoelectric composites by direct-write technique, 2008 17th IEEE International Symposium on the Applications of Ferroelectrics, Santa Re, NM, USA, 2008, pp. 1-2. [73] M.S. Mirza, T. Yasin, M. Ikram, S. Khan, M.N. Khan, Dielectric and piezoelectric properties of piezoceramic/polymer 1 -- 3 composites fabricated by a modified align-and-fill technique, Materials Chemistry and Physics 149(Supplement C) (2015) 670-677. [74] A. Chaipanich, R. Rianyoi, R. Potong, P. Penpokai, P. Chindaprasirt, Dielectric and Piezoelectric Properties of 2 -- 2 PZT-Portland Cement Composites, Integrated Ferroelectrics 149(1) (2013) 89-94. [75] C.R. Bowen, D.P. Almond, Modelling the 'universal' dielectric response in heterogeneous materials using microstructural electrical networks, Materials Science and Technology 22(6) (2006) 719-724. [76] H.K.B.D.R.U. William David. Kingery, Introduction to Ceramics, New York ; London : Wiley- Interscience 1976. 19
1804.02893	2	1804	2018-08-15T14:59:15	Nanoscale spin manipulation with pulsed magnetic gradient fields from a hard disc drive writer	[ "physics.app-ph", "cond-mat.mes-hall" ]	The individual and coherent control of solid-state based electron spins is important covering fields from quantum information processing and quantum metrology to material research and medical imaging. Especially for the control of individual spins in nanoscale networks, the generation of strong, fast and localized magnetic fields is crucial. Highly-engineered devices that demonstrate most of the desired features are found in nanometer size magnetic writers of hard disk drives (HDD). Currently, however, their nanoscale operation, in particular, comes at the cost of excessive magnetic noise. Here, we present HDD writers as a tool for the efficient manipulation of single as well as multiple spins. We show that their tunable gradients of up to 100 {\mu}T/nm can be used to spectrally address individual spins on the nanoscale. Their GHz Bandwidth allows to switch control fields within nanoseconds, faster than characteristic timescales such as Rabi and Larmor periods, spin-spin couplings or optical transitions, thus extending the set of feasible spin manipulations. We used the fields to drive spin transitions through non-adiabatic fast passages or enable the optical readout of spin states in strong misaligned fields. Building on these techniques, we further apply the large magnetic field gradients for microwave selective addressing of single spins and show its use for the nanoscale optical colocalization of two emitters.	physics.app-ph	physics	Nanoscale spin manipulation with pulsed magnetic gradient fields from a hard disc drive writer S. Bodenstedt1, I. Jakobi1,, J. Michl1, I. Gerhardt1,2, P. Neumann1 and J.Wrachtrup1 13. Physikalisches Institut, Universität Stuttgart and Institute for Integrated Quantum Science and Technology IQST, Pfaffen- waldring 57, D-70569 Stuttgart, Germany 2Max Planck Institute for Solid State Research, Heisenbergstrasse 1, D-70569 Stuttgart, Germany ABSTRACT: The individual and coherent control of solid-state based electron spins is important covering fields from quantum in- formation processing and quantum metrology to material research and medical imaging. Especially for the control of individual spins in nanoscale networks, the generation of strong, fast and localized magnetic fields is crucial. Highly-engineered devices that demon- strate most of the desired features are found in nanometer size mag- netic writers of hard disk drives (HDD). Currently, however, their nanoscale operation, in particular, comes at the cost of excessive magnetic noise. Here, we present HDD writers as a tool for the ef- ficient manipulation of single as well as multiple spins. We show that their tunable gradients of up to 100 µT/nm can be used to spec- trally address individual spins on the nanoscale. Their GHz Band- width allows to switch control fields within nanoseconds, faster than characteristic timescales such as Rabi and Larmor periods, spin-spin couplings or optical transitions, thus extending the set of feasible spin manipulations. We used the fields to drive spin tran- sitions through non-adiabatic fast passages or enable the optical readout of spin states in strong misaligned fields. Building on these techniques, we further apply the large magnetic field gradients for microwave selective addressing of single spins and show its use for the nanoscale optical colocalization of two emitters. Networks of interacting nitrogen-vacancy defect (NV) spins in diamond have a wealth of potential applications1, ranging from quantum information processing2 to sensing arrays3 -- 5. Spins in a coherently coupled network may exchange quantum information and form a quantum register. To fully control a network containing many NVs, the selective manipulation of individual spins has to be possible. However, when multiple NVs are located within the same diffraction limited spot, i.e. 𝑑 < 200 nm, they cannot be easily con- trolled individually if they have parallel axis orientations. One needs to resort to super-resolution techniques like STED to indi- vidually address the defects which often have a negative impact on neighboring spins6. Magnetic field sources exhibiting strong gradi- ent and a high bandwidth, i.e. fast switching times, offer an ade- quate solution. Gradient fields can be used to encode the location of NVs onto its spin properties such as its Larmor precession7. Hence, they separate these properties of otherwise indistinguisha- ble spins and can make them individually addressable. Coherent dipolar coupling among neighboring spins requires distances below some tens of nanometers8. Detuning such spins necessitates strong gradients7 on the order of 10-100 µT/nm. Fast switching times are required as the gradient field might interfere with the dynamics of the network, so that it can be exclusively applied for local manipu- lations of single spins. At the same time, fast changing control fields can also affect the spin dynamics, e.g. through non-adiabatic fast passages which have to be controlled or avoided. As a draw- back, sources able to generate strong fields at a high bandwidth may also introduce strong magnetic noise. This effect needs to be taken into account, as it further limits the distance between coherently interacting spins. We find the desired features for a gradient field source readily available in magnetic writers of commercial hard disk drives9 (HDD). Modern HDDs can produce magnetic fields on the order of 1 T with gradients of up to 10 mT/nm and pulse them well within a nanosecond10. In order to flip the magnetization, i.e. write a bit, HDDs use a microscale lithographed electromagnet, the writer, as shown in Figure 1a. The demand for higher storage capacities and faster operation has prompted the miniaturization of the device and decades worth of engineering have developed HDD writers to a so- phisticated tool. HDD writers consist of a nickel-iron alloy core with a high per- meability (µ > 10,000) that is shaped in three poles perpendicular to the air-bearing surface (ABS) of the head and a connecting yoke on top. The central pole, the write pole, is tapered to a 100 nm wide tip on the ABS while the two outer (return) poles end in broad shielding brackets close to the write pole. The front return pole closes on the write pole with a small gap of about 20 nm, while the back return-pole ends in a few micrometer distance. A pair of coils is photolithographically designed around the yoke such that the write pole is jointly magnetized in one direction when a current of up to 30 mA is applied. Once magnetized, the magnetic field is emitted perpendicular to the ABS from the write pole and curls around to the nearby front return pole. This geometry creates a strong magnetic field gradient where only a 20 nm sized area is strong enough to flip the magnetization of the recording medium underneath the ABS. In order to write data quickly the writer is designed to have a GHz bandwidth. It was previously reported that NV spins are suitable nanoscale magnetometers for the characteri- zation of such fields10. Here we reverse the situation and show their versatility in the control of single as well as multiple NV spins and demonstrate their potential use for the selective addressing of NV spins. The NV is a point defect where one carbon atom is missing from the diamond lattice and a nitrogen atom substitutes a neighboring site (see Figure 1b). In its negative charge state the NV has triplet (𝑆 = 1) ground (3A) and excited (3E) states and additional metastable singlet (𝑆 = 0) states. The Hamiltonian 𝐻 describing the spin of the triplet ground state is in its simplest form11 𝐻 = 𝐷𝑆𝑧 2 + 𝛾 2𝜋 (𝐵𝑥𝑆𝑥 + 𝐵𝑦𝑆𝑦 + 𝐵𝑧𝑆𝑧) . (1) It exhibits a zero-field splitting (ZFS) of 𝐷 = 2.87 GHz and an isotropic Zeeman interaction with a magnetic field 𝐵⃗ and a gyromagnetic ratio of 𝛾/2𝜋= 28.02 GHz/T. Figure 1c shows the 1/2 spin energy levels for axial 𝐵𝑧 and radial 𝐵⊥ = (𝐵𝑥 fields. At moderate magnetic fields 𝐵⃗ ≪ 2𝜋𝐷/𝛾 ≈ 100 mT the ZFS is the dominant term and defines the quantization axis along the symmetry axis 𝑧 of the defect. The contribution of axial fields 𝐵𝑧𝑆𝑧 leaves the quantization axis unchanged and only shifts the eigenenergies resulting in a linear Zeeman effect that splits the magnetic states 𝑚𝑆 = ±1 by a factor 𝛾/𝜋 ⋅ 𝐵𝑧. Radial contributions 𝐵𝑥𝑆𝑥, 𝐵𝑦𝑆𝑦, on the other hand, mainly mix magnetic states 𝑚𝑠 = 2 + 𝐵𝑦 2) 1 ±1 and result in quadratic Zeeman interaction. With the knowledge of two transition frequencies two components of the magnetic field 𝐵𝑧 and 𝐵⊥ and consequently the polar angle tan 𝜃 = 𝐵⊥/𝐵𝑧 can be the azimuthal angle tan 𝜙 = 𝐵𝑦/𝐵𝑥 remains drived while ambiguous12 -- 15. Figure 1. Hard disk drives (HDD) and nitrogen vacancy (NV) defects in diamond. (a) Schematic overview of a hard disk write head. In a HDD a head incorporating a writer is suspended over the disk. The writer is a microstructured electromagnet with a central pole emitting a field able to flip the magnetization of a 20 nm wide sector on the recording medium. (b) Diamond unit cell with NV defect showing the orientation of magnetic fields with respect to the defect symmetry axis. (c) Spin energy levels in the NV- triplet ground state 3A for axial (top) and radial (bottom) magnetic fields. A zero-field-splitting of 2.87 GHz acting along the symmetry axis causes an anisotropic interaction with magnetic fields. (d) Electronic level scheme with optical cycle (green, red arrows) between triplet states 3A and 3E. Intersystem-crossing (ISC, yellow) allows a non-radiative decay to the singlet states (not depicted) and back to 3A resulting in a fluoresence contrast between spin states and spin polarization. (e) Experimental setup. A HDD head is mounted on a piezo stage and rests in flat contact on a diamond membrane (1). A confocal microscope (2) has access from below to monitor NV defects close to the writer (inset). The writer is controlled by a signal generator (3). A wire antenna and a permanent magnet are used for additional control fields (4). Figure 1d depicts the electronic level scheme of the NV. A spin-conserving optical transition can be driven to excite the NV electron system into the 3E state. The spin state can be read out via (𝑧) = 0 state in the fluorescence intensity of the defect. Here the 𝑚𝑆 the optically excited state mostly decays through fluorescence (𝑧) = ±1 states have a significant rate of non- while the 𝑚𝑆 radiatively decaying through intersystem crossing (ISC)16. This leads to a fluorescence contrast between these spin states and allows for optically detected magnetic resonance (ODMR). Typically spin states are manipulated by resonant microwave radiation17. However, in radial fields the spin basis is mixed and all states have significant ISC rates. Therefore, ODMR only yields a sufficient fluorescence contrast under the constraint that radial fields are smaller than ~20 mT18,19. A schematic of the experimental setup is depicted in Figure 1e. We connect a generic perpendicular magnetic recording head to a frequency generator. In order to have the magnetic field of the writer interact with NV spins we place it with the ABS in flat con- tact to the surface of a diamond substrate. As the field strength of the writer decays strongly over distance we use NV defects 5- 20 nm below the diamond surface. Using diamond membranes of 30-50 µm thickness, we have optical access from the back surface of the diamond. The membranes are mounted on an objective-scan- ning confocal microscope (60x/1.35 NA oil objective) to excite the electronic transitions and to collect the photoluminescence (PL). The HDD head together with its assembly is mounted on a separate piezo stage. The assembly's flexibility helps to stabilize the ABS's position with respect to the diamond surface as the head will mostly be kept in place by static friction. In order to actually move the head, the tension in the assembly has to overcome the static friction. At this point the head will start to slide over the diamond surface and can follow steps with a lateral positioning accuracy on the order of 10 nm. In order to control individual spins on the nanoscale using the HDD's gradient field, the writer needs to be placed close to the NVs. However, while its ferromagnetic core enables the HDD's field amplitude and gradient, it comes at the cost of large magnetic noise even when the write current is off. Many applications of the NV rely on spin polarization or coherent phase evolution where the relaxation times 𝑇1 and 𝑇2 are the limiting factors, respectively. Hence, the impact of this noise source on the spin relaxation times needs to be characterized for such applications. To probe the noise spectrum in different frequency domains we employ relaxometry20 -- 24. By measuring the longitudinal relaxation time 𝑇1 = 1/𝛤1 we gain insight into the noise spectral density at the zero field Larmor frequency of about 3 GHz of noise components perpendicular to 𝑧. Measurements of the transversal relaxation time 𝑇2 = 1/𝛤2 probe noise components at lower frequencies25 parallel to 𝑧. The corre- sponding total relaxation rates 𝛤tot = 𝛤int + 𝛤ext can be separated into an intrinsic contribution 𝛤int caused by a variety of sources within the diamond and an external part 𝛤ext caused by the writer. For magnetic noise the external rate can be written as25 𝛤ext = 𝛾2 ⋅ 〈𝐵2〉 ⋅ ∫ 𝑆(𝜈) ⋅ 𝐹𝑖(𝜈) d𝜈 , (2) where 𝛾 is the gyromagnetic ratio, √〈𝐵2〉 the effective magnetic field at the position of the NV, 𝑆(𝜈) the normalized power spectral density (PSD) depending on the frequency 𝜈. The filter function 𝐹𝑖(𝜈) describes the spectral sensitivity depending on the measure- ment sequence. For the scan in Figure 2a the microscope objective is aligned to a single NV while for different writer positions the spin relaxation is measured to determine 𝑇1. During the measurement no electric current is applied to the writer. With an increasing distance to the write and return poles the longitudinal spin relaxation time reaches several hundred microseconds (bottom part of Figure 2a and line plot in Figure 2b). Below the write and return pole the magnetic field noise reduces 𝑇1 to values around 100 µs (upper part). Close to the expected position of the write pole (center in Figure 2a and 2 b) the longitudinal relaxation times are further reduced to only tens of microseconds. These two sets of data with and without influence of the writer allow us to estimate the magnetic noise. reduced from 17 nm to 8 nm in the presence of the writer8. Never- theless, while we find applications based on the spins' phase evo- lution to be challenging they are in principle possible. The filter function 𝐹𝑖(𝜈) of a 𝑇1 relaxometry measurement is only significant around the NV's ESR frequency around 3 GHz25. Assuming a constant PSD over this sensitivity range the spectral magnetic (or √〈𝐵2〉 ⋅ 𝑆(3 GHz) = 2.2 nT/√Hz) can be calculated at the position closest to the write pole. 〈𝐵2〉 ⋅ 𝑆(3 GHz) = 4.9 nT²/Hz field noise In a similar scenario where two parallel NV are separated by 10 nm, a gradient magnetic field on the order of 100 µT/nm is re- quired to separate the individual resonance lines by Δ𝜈 = Δ𝑟 𝛾/2𝜋 ⋅ 𝑑𝐵/𝑑𝑟 ≈ 30 MHz which allows for fast, high-fidelity and selective manipulations. The writer can easily produce such gradients, however, only in conjunction with strong, arbitrarily ori- ented fields that generally prevent ODMR. Hence, before selective microwave pulses can be applied, we must confirm that the gradient field can be switched off for initialization and read-out and that spin manipulations work in arbitrary fields. The large bandwidth of the writer allows to switch the magnetic field on fast timescales. While this is generally desired, fast magnetic field ramps can induce ad- ditional dynamics on the NV spin. One particular behavior is the non-adiabatic fast passage. If the field is misaligned from the NV symmetry axis and the Zeeman interaction exceeds the ZFS, i.e. for field strength beyond 100 mT, the eigenstates change substantially and the corresponding energies are swept over a level anti-crossing (LAC). In this case the spin state gradually follows the change of the eigenbasis in an adiabatic process (if the sweep is slow com- pared to the energy gap 𝛥 at the LAC) or to a new state in a non- adiabatic process (if the sweep is fast). The transition probability 𝑃 is well described by the Landau-Zener formula26 Figure 2. Passive effects. (a) 𝑇1 relaxometry lateral scan of the writer. The NV's spin relaxation is measured in respect to the relative position between writerp and NV (objective fixed on NV, HDD scanned, writer off) to determine 𝑇1. The estimated position of the write pole is depicted by the blue arrow. (b) Line scans along the lines (blue and green) in (a). The gray areas show the estimated position of the write (wp) and return pole (rp). (c) Comparison of the transversal relaxation time 𝑇2 of three NVs (indicated as green, orange and blue) with (saturated) and without (pale) write head on top. All lines were measured with a similar overall measurement time. Therefore, lines with longer echo times appear noisier compared to lines with shorter ones. Figure 2c shows 𝑇2 measurements for three different NVs with (saturated) and without (pale) write head. We often observe trans- verse relaxation times close to the writer are reduced to an order of a few hundred nanoseconds (NV3, green: 510 ns), compared to a few µs without writer. In some circumstances, however, we do find relaxation times on the order of a few µs (NV1, blue: 4.6 µs and NV2, orange: 2.7 µs) with the engaged writer. When a similar anal- ysis as for 𝑇1 is applied, we can determine the magnetic noise for the second frequency domain. The filter function of a spin echo measurement depends on the total evolution time. For 10-100 µs the dominant sensitivity range is in the sub MHz regime25. In this range we find noise ranging from 〈𝐵2〉 ⋅ 𝑆(1 MHz) = 13 nT²/Hz (3.6 nT/√Hz) up to 96 nT²/Hz (9.8 nT/√Hz) (NV 1 and 3 respec- tively). We find that the writer introduces large magnetic noise to NV spins in its vicinity. This may impede interferometric measure- ments, like nuclear spin detection, or even disrupt potentially cou- pled networks of spins. For instance, the maximum distance at which NV1 could coherently couple to a neighboring NV would be 𝑃 = 𝑒− 𝜋2Δ2 𝑣 , (3) where 𝜈 is the rate of change of the eigenenergies during the sweep, also known as the Landau-Zener velocity. In the case of the NV the gap energy is given by the radial field strength 𝛥 = √2 ⋅ 𝛾/2𝜋 ⋅ 𝐵⊥ at the avoided crossing and can be as large as the ZFS, i.e. 2.87 GHz. The Landau-Zener velocity 𝑣 = 𝛾/2𝜋 ⋅ 𝜕𝐵𝑧 𝜕𝑡⁄ de- pends on the speed of the magnetic field sweep. The writer can in principle sweep the eigenenergies with velocities of GHz/ns and consequently realistically drive non-adiabatic fast passages, where 𝑣 ≫ 𝛥2. Figure 3. Non-adiabatic fast passage. (a) Level scheme for slightly misaligned magnetic fields. At about 100 mT the two lower electron spin states form a level anti-crossing (LAC). (b) A ramped magnetic field greater than 100 mT can lead to an adiabatic or non- adiabatic evolution depending on the ramp time. (c) For ramp times longer than 100 ns (green) the system evolves adiabatically, whereas values shorter than 50 ns (blue) lead to a non-adiabatic evolution. (d) Pulse scheme for the measurement in (c). For 3 reference the red curve corresponds to a reduced ramp amplitude that prevents passing the LAC. For an experiment we move the writer close to a NV so that its write field has a small angle 𝜃 < 5° to the NV axis and a field strength of 𝐵 ≈ 100 mT (see Figure 3a). This setting is close to the LAC and introduces a mixing of the 𝑚𝑆 = 0 and 𝑚𝑆 = −1 states. The gap energy 𝛥 in this case is smaller than 350 MHz. We then apply a pulse sequence (see Figure 3b and d) where we polar- ize the NV spin into 𝑚𝑆 = 0 while the writer is off. Subsequently, we ramp the current beyond the avoided crossing on a varying time- scale between 5 and 200 ns. Finally, we slowly ramp it back down over a period of 1 µs and read out the spin state. Assuming the fall- ing ramp to be slow enough to change the spin state adiabatically, we expect fast passages during the rising edge to manifest as a re- duction in fluorescence. Figure 3c shows the result of the measure- ment. As expected we find non-adiabatic fast-passages for short rising edges and an exponential behavior towards an adiabatic sweep. As a reference we also recorded sweeps at lower current amplitudes that do not drive the system beyond the LAC. In this case we do not observe any population transfer, which is consistent with the theory. Assuming a linear increase of the magnetic field strength dur- ing the ramp, i.e. 𝐵𝑧(𝑡) = 𝐵max ⋅ 𝑡/𝑡ramp, the Landau-Zener for- mula and retrieve the spin state optically. In an experiment we bring the write pole close to an NV defect. We apply a pulse scheme as shown in Figure 4c, where a laser pulse initializes the spin before the current on the writer is ramped up to 10-20 mA within 100 ns. The magnetic field is kept static for the duration of a probing mi- crowave pulse before it is ramped down again for the optical read- out. Figure 4b shows a spectrum recorded with this method. From the resonances we derive a magnetic field of 𝐵 = 75.4 mT and a polar angle 𝜃 = 52° during the write pulse. For larger fields the spin transitions between the highest and lowest states may become forbidden. In such a case additional microwave pulses are required to observe both transitions. Expanding on this method, we move the writer within its accu- racy by 25 nm steps to map out the magnetic field around the write pole. Figure 4d shows the magnetic field strength and the axial and radial components that we derive from the measurement. As ex- pected we see the magnetic field to vary strongly over the range of a few hundred nanometers ranging from 15 to 28 mT. Furthermore, we find magnetic field gradients on the order of 100 µT/nm, which is comparable to the gradient of magnetized atomic force micro- scope tips or nanostructures for magnetic resonance force micro- scopes27,28. 𝑃 = 𝑒− 𝜋2𝛥2 𝑣 = 𝑒 − 𝛥2 𝜋2 ⋅ γ/2π 𝐵max ⋅𝑡ramp = 𝑒− 𝑡ramp 𝜏 (4) is expressed in terms of the maximum field strength 𝐵max and the ramp time 𝑡ramp. This formula is used to fit the measurement data in Figure 3c by introducing and adjusting the fit parameter 𝜏 = 𝛾/2𝜋 𝐵max 𝛥2 = 48 ns which depends on both 𝐵max and 𝛥. 𝜋2 ⋅ The exact value of the energy splitting Δ during the sweep is unknown, as we can only characterize static fields for correspond- ing DC currents. The dynamic field however may not necessarily have the same orientation as the settled system. The same argument holds for the maximum magnetic field strength 𝐵max. As the mag- netic field has to exceed at least the LAC at the end of the sweep, a realistic lower bound for 𝐵max can be estimated. By assuming a 𝐵max = 100 mT and taking the fit parameter 𝜏 = 48 ns into account also a lower bound for the energy splitting of 𝛥 = 77 MHz can be calculated. The corresponding maximum magnetic field rate 𝐵𝑧 = 𝜕𝐵𝑧/𝜕𝑡 for 𝑡ramp = 5 ns is 𝐵𝑧 = 20 mT/ns. For a realistic value the maximum magnetic field strength of about 𝐵max = 200 mT meas- ured with an NV for hard disk recording heads10 can be considered. This leads to a splitting of 𝛥 = 109 MHz and a magnetic field rate of 𝐵𝑧 = 40 mT/ns (𝑡ramp = 5 ns). In principle Landau-Zener transitions are coherent processes. By finding the proper ramp times 𝑡ramp, gates similar to 𝜋- or, in particular, 𝜋/2-pulses are conceivable. In our experiments we were, however, not able to observe coherent state evolutions. Small changes in the magnetic field settings immediately translate into vastly different evolution speeds. We suspect that we could not drive the magnetic field accurately enough to a consistent setting and hence blurred out any coherent evolution under a large inho- mogeneous broadening. With the knowledge of the spin's behavior during magnetic field ramps, we can now construct a scheme to manipulate and re- trieve the spin state in arbitrary fields and characterize the gradient field, i.e. search for appropriate resonances for selective address- ing. As long as the magnetic field is ramped slowly enough for the spin to transition adiabatically from one quantization basis to the other, i.e. 𝛾/2𝜋 ⋅ 𝐵 ≪ 𝛥2, the spin states populations are mapped unambiguously from basis to basis. As shown in the schematic in Figure 4a, this allows to probe for spin resonances in the radial field Figure 4. Pulsed magnetic write fields. (a) Pulsed field ODMR. Due to the large bandwidth the write field can be applied only for a 𝜋-pulse while it is off for initialization and read-out. That way spin resonances can be probed at magnetic field strengths much greater than for DC magnetic fields. (b) Spectrum with two resonances corresponding to a magnetic field strength of 75.4 mT with a misalignment of 52°. The coloring indicates the path corresponding to the scheme in (a) in case the microwave is resonant (green) or off-resonant (blue). (c) Pulse scheme for the measurements in (b) and (d). (d) Magnetic field lateral scan around the write pole obtained by pulsed field ODMR. The magnetic field strengths range from 15 mT to 28 mT leading to magnetic field gradients on the order of 100 µT/nm. The spin transitions can be observed and controlled in arbitraryly oriented magnetic fields. Finally, this new ODMR method allows using the gradient field to address individual parallel NVs on the nanoscale7,27,29. Here we move the writer over a detected NV fluorescence spot that contains 4 such a pair of NVs (see Figure 5a and b) and apply the same pulse sequence as before (see Figure 4) with moderate current ampli- tudes. At low currents we first find two resonance lines correspond- ing to the two spin transitions of both defects (Figure 5c). However, when the current is increased the lines split into two pairs of lines corresponding to the individual transitions of each defect. In this case each NV spin state can be manipulated individually with a fi- delity of 99% (calculated by the spectral overlap) and 10 MHz Rabi frequencies for the left transition pair in Figure 5c. By reducing the power broadening or applying higher write currents (corresponding to higher magnetic field gradients) the minimal distance of spec- trally addressable parallel NV pairs can be even further reduced. Figure 5. Selective addressing of a pair of NVs using tunable magnetic field gradients. (a) Schematic picture of two parallel NVs in the gradient field of the writer. (b) A confocal scan shows NVs as diffraction limited spots. The central spot contains a parallel pair. (c) ODMR spectra of the NV pair. At low currents the spectral lines overlap. With increasing current first two resonance lines appear corresponding to the two possible transitions of both NVs. If the current is further increased the lines split up into two pairs of lines corresponding to the individual transitions of each defect. The fidelity, calculated by the spectral overlap, is 99% for the left transition pair. (d) Optically reconstructed relative (lateral) position of the parallel NVs in the centered confocal spot. The figure shows a histogram (1 nm bin width) of 286 individual DESM scans. The crosshair marks the origin point, i.e. the position of NV1. The relative distance to NV2 of 44.5 nm is determined with a resolution (FWHM) of 13.2 nm. The distance between the addressed NVs is an important meas- ure for the applicability on spin networks. We determine their rel- ative position with an optical reconstruction method, called deter- ministic emitter switch microscopy (DESM)30 and at the same time demonstrate the usefulness of our addressing method. This super- resolution method requires selective microwave pulses to switch individual defects to its dark spin state so that its image can be sep- arated from a reference image where all emitters are bright. Figure 5d shows a histogram of 286 relative positions of the pair extracted from DESM scans. We find a relative distance of 44.5 nm with a resolution (FWHM) of 13.2 nm. That implies that spectral address- ing is possible on length scales close to typical dipolar coupling ranges. As we did not have to use the full strength of the gradient the approach would likewise work on defects as close as 10 nm. 5 Our experiments show that HDD writers are a valuable tool for the generation of control fields in spin experiments. We are able to use various new degrees of freedom which are enabled by their re- markable features. The GHz bandwidth allows to rapidly pulse con- trol fields, modulate the spin dynamics in run time and drive non- adiabatic fast passages. Pulsed magnetic fields allow measurements in strong off-axial fields that would otherwise not yield sufficient fluorescence contrast. This enables the use of the magnetic field gradient to address individual NV spins on the nanoscale. The method could easily be expanded to address larger clusters of NVs. However, the introduction of magnetic noise from the writer puts additional limitations on the interaction range of a coupled net- work. Yet, as the writer can produce gradients on the order of mT/nm, spins within a few nanometer distance can be addressed. Hence, HDD writers are suitable for applications controlling large spin arrays as for example in proposed diamond-based quantum processor2,7,8. Ultimately nano-fabricated writers can serve as a template for micro- and nanostructures specifically tailored to the requirements of single spin research and devices. AUTHOR INFORMATION Corresponding Author E-mail: [email protected]. Author Contributions S.B., I.J., J.M., I.G., P.N. and J.W. conceived the experiments. S.B. and I.J. performed the experiments. S.B. and I.J. analyzed the data. S.B., I.J. and J.W. wrote the manuscript. Notes The authors declare no competing financial interests. ACKNOWLEDGMENT We acknowledge financial support by the German Science Foun- dation (SPP1601 and FOR1493), the EU (ERC grant SMeL), the Volkswagen Foundation, the Humboldt Foundation, the Baden Wuerttemberg Foundation and the MPG. Furthermore, we thank Fadi El Hallak of Seagate Technology for providing hard disk head samples and technical assistance and Rob- ert McMichael of NIST CNST for fruitful discussions. REFERENCES (1) Childress, L.; Walsworth, R.; Lukin, M. Atom-like crystal defects: From quantum computers to biological sensors. Physics Today 2014, 67, 38 -- 43. (2) Dolde, F.; Jakobi, I.; Naydenov, B.; Zhao, N.; Pezzagna, S.; Tra- utmann, C.; Meijer, J.; Neumann, P.; Jelezko, F.; Wrachtrup, J. Room- temperature entanglement between single defect spins in diamond. Nature Phys 2013, 9, 139 -- 143. (3) Cappellaro, P.; Lukin, M. D. Quantum correlation in disordered spin systems: Applications to magnetic sensing. Phys. Rev. A 2009, 80, 155. (4) Raghunandan, M.; Wrachtrup, J.; Weimer, H. High-density quantum sensing with dissipative first order transitions. arXiv:1703.07358v1 2017. (5) Pfender, M.; Aslam, N.; Waldherr, G.; Neumann, P.; Wrachtrup, J. Single-spin stochastic optical reconstruction microscopy. Proceedings of the National Academy of Sciences of the United States of America 2014, 111, 14669 -- 14674. (6) Wildanger, D.; Patton, B. R.; Schill, H.; Marseglia, L.; Hadden, J. P.; Knauer, S.; Schönle, A.; Rarity, J. G.; O'Brien, J. L.; Hell, S. W. et al. Solid immersion facilitates fluorescence microscopy with nanometer reso- lution and sub-ångström emitter localization. Advanced materials (Deer- field Beach, Fla.) 2012, 24, OP309-13. (7) Zhang, H.; Arai, K.; Belthangady, C.; Jaskula, J.-C.; Walsworth, R. L. Selective addressing of solid-state spins at the nanoscale via magnetic res- onance frequency encoding. npj Quantum Inf 2017, 3, 38. (8) Jakobi, I.; Momenzadeh, S. A.; Oliveira, F. F. de; Michl, J.; Ziem, F.; Schreck, M.; Neumann, P.; Denisenko, A.; Wrachtrup, J. Efficient crea- tion of dipolar coupled nitrogen-vacancy spin qubits in diamond. J. Phys.: Conf. Ser. 2016, 752, 12001. (9) Tao, Y.; Eichler, A.; Holzherr, T.; Degen, C. L. Ultrasensitive me- chanical detection of magnetic moment using a commercial disk drive write head. Nature communications 2016, 7, 12714. (10) Jakobi, I.; Neumann, P.; Wang, Y.; Dasari, D. B. R.; El Hallak, F.; Bashir, M. A.; Markham, M.; Edmonds, A.; Twitchen, D.; Wrachtrup, J. Measuring broadband magnetic fields on the nanoscale using a hybrid quantum register. Nature nanotechnology 2017, 12, 67 -- 72. (11) Doherty, M. W.; Manson, N. B.; Delaney, P.; Jelezko, F.; Wrachtrup, J.; Hollenberg, L. C.L. The nitrogen-vacancy colour centre in diamond. Physics Reports 2013, 528, 1 -- 45. (12) Balasubramanian, G.; Chan, I. Y.; Kolesov, R.; Al-Hmoud, M.; Tisler, J.; Shin, C.; Kim, C.; Wojcik, A.; Hemmer, P. R.; Krueger, A. et al. Nanoscale imaging magnetometry with diamond spins under ambient con- ditions. Nature 2008, 455, 648 -- 651. (13) Maletinsky, P.; Hong, S.; Grinolds, M. S.; Hausmann, B.; Lukin, M. D.; Walsworth, R. L.; Loncar, M.; Yacoby, A. A robust scanning diamond sensor for nanoscale imaging with single nitrogen-vacancy centres. Nature nanotechnology 2012, 7, 320 -- 324. (14) Clevenson, H.; Trusheim, M. E.; Teale, C.; Schröder, T.; Braje, D.; Englund, D. Broadband magnetometry and temperature sensing with a light-trapping diamond waveguide. Nature Phys 2015, 11, 393 -- 397. (15) Nusran, N. M.; Momeen, M. U.; Dutt, M. V. G. High-dynamic-range magnetometry with a single electronic spin in diamond. Nature nanotech- nology 2011, 7, 109 -- 113. (16) Manson, N. B.; Harrison, J. P.; Sellars, M. J. Nitrogen-vacancy cen- ter in diamond: Model of the electronic structure and associated dynamics. Phys. Rev. B 2006, 74, 664. (17) Appel, P.; Ganzhorn, M.; Neu, E.; Maletinsky, P. Nanoscale micro- wave imaging with a single electron spin in diamond. New J. Phys. 2015, 17, 112001. (18) Tetienne, J.-P.; Rondin, L.; Spinicelli, P.; Chipaux, M.; Debu- isschert, T.; Roch, J.-F.; Jacques, V. Magnetic-field-dependent photody- namics of single NV defects in diamond: An application to qualitative all- optical magnetic imaging. New J. Phys. 2012, 14, 103033. (19) Epstein, R. J.; Mendoza, F. M.; Kato, Y. K.; Awschalom, D. D. Ani- sotropic interactions of a single spin and dark-spin spectroscopy in dia- mond. Nature Phys 2005, 1, 94 -- 98. (20) van der Sar, T.; Casola, F.; Walsworth, R.; Yacoby, A. Nanometre- scale probing of spin waves using single-electron spins. Nature communi- cations 2015, 6, 7886. (21) Tetienne, J.-P.; Hingant, T.; Rondin, L.; Cavaillès, A.; Mayer, L.; Dantelle, G.; Gacoin, T.; Wrachtrup, J.; Roch, J.-F.; Jacques, V. Spin re- laxometry of single nitrogen-vacancy defects in diamond nanocrystals for magnetic noise sensing. Phys. Rev. B 2013, 87, DOI: 10.1103/PhysRevB.87.235436. (22) Schmid-Lorch, D.; Häberle, T.; Reinhard, F.; Zappe, A.; Slota, M.; Bogani, L.; Finkler, A.; Wrachtrup, J. Relaxometry and Dephasing Imag- ing of Superparamagnetic Magnetite Nanoparticles Using a Single Qubit. Nano letters 2015, 15, 4942 -- 4947. (23) Steinert, S.; Ziem, F.; Hall, L. T.; Zappe, A.; Schweikert, M.; Götz, N.; Aird, A.; Balasubramanian, G.; Hollenberg, L.; Wrachtrup, J. Mag- netic spin imaging under ambient conditions with sub-cellular resolution. Nature communications 2013, 4, 1607. (24) Pelliccione, M.; Jenkins, A.; Ovartchaiyapong, P.; Reetz, C.; Em- manouilidou, E.; Ni, N.; Bleszynski Jayich, A. C. Scanned probe imaging of nanoscale magnetism at cryogenic temperatures with a single-spin quantum sensor. Nature nanotechnology 2016, 11, 700 -- 705. (25) Schäfer-Nolte, E.; Schlipf, L.; Ternes, M.; Reinhard, F.; Kern, K.; Wrachtrup, J. Tracking temperature-dependent relaxation times of ferritin nanomagnets with a wideband quantum spectrometer. Physical review let- ters 2014, 113, 217204. (26) Zener, C. Non-Adiabatic Crossing of Energy Levels. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 1932, 137, 696 -- 702. (27) Grinolds, M. S.; Warner, M.; Greve, K. de; Dovzhenko, Y.; Thiel, L.; Walsworth, R. L.; Hong, S.; Maletinsky, P.; Yacoby, A. Subnanometre resolution in three-dimensional magnetic resonance imaging of individual dark spins. Nature nanotechnology 2014, 9, 279 -- 284. (28) Degen, C. L.; Poggio, M.; Mamin, H. J.; Rettner, C. T.; Rugar, D. Nanoscale magnetic resonance imaging. Proceedings of the National Academy of Sciences of the United States of America 2009, 106, 1313 -- 1317. (29) Arai, K.; Belthangady, C.; Zhang, H.; Bar-Gill, N.; DeVience, S. J.; Cappellaro, P.; Yacoby, A.; Walsworth, R. L. Fourier magnetic imaging with nanoscale resolution and compressed sensing speed-up using elec- tronic spins in diamond. Nature nanotechnology 2015, 10, 859 -- 864. (30) Chen, E. H.; Gaathon, O.; Trusheim, M. E.; Englund, D. Wide-field multispectral super-resolution imaging using spin-dependent fluorescence in nanodiamonds. Nano letters 2013, 13, 2073 -- 2077. 6
1711.11373	1	1711	2017-11-30T13:17:53	Demonstration of a 2x2 programmable phase plate for electrons	[ "physics.app-ph" ]	First results on the experimental realisation of a 2x2 programmable phase plate for electrons are presented. The design consists of an array of electrostatic einzel lenses that influence the phase of electron waves passing through 4 separately controllable aperture holes. This functionality is demonstrated in a conventional transmission electron microscope operating at 300~kV and results are in very close agreement with theoretical predictions. The dynamic creation of a set of electron probes with different phase symmetry is demonstrated, thereby bringing adaptive optics in TEM one step closer to reality. The limitations of the current design and how to overcome these in the future are discussed. Simulations show how further evolved versions of the current proof of concept might open new and exciting application prospects for beam shaping and aberration correction.	physics.app-ph	physics	Demonstration of a 2x2 programmable phase plate for electrons Jo Verbeeck a Armand B´ech´e a Knut Muller-Caspary a Giulio Guzzinati a Minh Anh Luong b Martien Den Hertog c aEMAT, University of Antwerp, Groenenborgerlaan 171, 2020 Antwerp, Belgium. bUniv. Grenoble Alpes, CEA, INAC, MEM, LEMMA Group, F-38000 Grenoble, France cInstitut Neel, 25 avenue des Martyrs BP 166, 38042 Grenoble Cedex 9, France. Abstract First results on the experimental realisation of a 2x2 programmable phase plate for electrons are presented. The design consists of an array of electrostatic einzel lenses that influence the phase of electron waves passing through 4 separately controllable aperture holes. This functionality is demonstrated in a conventional transmission electron microscope operating at 300 kV and results are in very close agreement with theoretical predictions. The dynamic creation of a set of electron probes with different phase symmetry is demonstrated, thereby bringing adaptive optics in TEM one step closer to reality. The limitations of the current design and how to overcome these in the future are discussed. Simulations show how further evolved versions of the current proof of concept might open new and exciting application prospects for beam shaping and aberration correction. Key words: electron optics, phase plate, beam forming, electrostatic lens, adaptive optics 1 Introduction Adaptive optics, the technology to dynamically change the phase transfer of optical elements has its roots in astrophysics, where dynamically changing telescope mirrors can compensate for time varying atmospheric induced aber- rations for optimised observation from earth [16,14,66,41], as well as in space Email address: [email protected] (Jo Verbeeck). Preprint submitted to Elsevier Science 1 December 2017 [9]. The technology has greatly improved since and has sparked an avalanche of innovative uses in many different areas where dynamic control over optical elements is wanted. Examples are endoscopic imaging through a mulltimode optical fiber [37], second harmonic imaging as e.g. used in stimulated emis- sion depletion microscopy (STED) [28], focusing inside thick semitransparent objects and tissue [6], optical quantum encryption [8], laser welding [43] and many more. All these breakthroughs were made possible by the existence or development of so-called spatial light modulators, devices that allow for a pro- grammable change of the phase of optical waves when passing through or being reflected from the modulator. Different designs exist based on liquid crystals [56,32], piezo controlled mirror segments [57], electrostatic or magnetic influ- encing of coated flexible membranes [62], shape changing flexible refractive elements as in mamal eyes [30] and many more. Each technology has its own advantages and disadvantages in terms of insertion losses, speed, constraints to what phase functions are allowed, pixel count, power handling of intense light beams, precision and so-on. In TEM, on the other hand, adaptive optical elements are commonplace. Indeed, even the simplest round magnetic or elec- trostatic lens, is tunable by either changing the current through the coils or by changing the potential difference over electrodes [46,27]. Adaptive optics is readily available on modern aberration corrected (S)TEM instruments in the form of complicated assemblies of multipolar lenses. They can be adaptively optimised on test samples to obtain the vastly improved resolution and cur- rent density that has formed the basis of most of the successes in experimental electron microscopy over the last decade [25,58,1,45,44,36,26]. Yet, aberration correctors in their current state don't allow for full flexibility in the phase they imprint on the electron wave. This is very apparent from the many attempts that have been, and are being made, towards a so-called Zernike phase plate for optimising the contrast in (weak) phase objects [67]. Indeed, if an aberra- tion corrector would be able to change the phase on the optical axis by a given amount while leaving the rest of the wave unaffected there would be no need for the veritable zoo of different phase plate designs that exist. The magnetic vector potential in a magnetic multipole corrector is determined by the indi- vidual poles that act as boundary conditions to the free space in which the electrons travel. The vector potential component in the direction of electron motion obeys a Poisson equation: ∇2Az − 1 c2 ∂2Az ∂2t = −µ0Jz, (1) assuming straight trajectories along z-direction through a thin region of space where the vector potential is non-zero. The electron, due to the Aharanov- Bohm effect, gets phase shifted by the projected magnetic vector potential Az along its path as φ(x, y) = q −∞ Azdz. Assuming time independent solutions in free space (Jz = 0), we note that the obtainable phase plates are limited to a (small) subset of all possible phase plates constrained by a Laplace equation ¯h R ∞ 2 in 2D: ∇2 x,yφ = − q ¯h ∂Az ∂z (cid:12) (cid:12) (cid:12) (cid:12) ∞ −∞ = 0. (2) For thicker lenses (all realistic cases), the trajectory can deviate from the z-direction and more complicated arguments are needed, but in practice, for reasonable boundary conditions or multipole orders, the available phase plates tend to be smooth. Fortunately, the available phase maps advantageously to the Zernike polynomials which are closely linked to the typical low order aber- rations that are dominant in the electron microscope. However, a sharp change in phase in the center of the field as required for e.g. a Zernike phase plate would require prohibitively large magnetic multipole orders and is impractical for the foreseeable future. To overcome this limitation, the assumption of free space has to be dropped (or alternatively dynamic fields are needed) and all current Zernike phase plates indeed consist of some form of material that is in contact with the electron wave. Some of the more promising designs create a miniaturised elec- trostatic einzel lens or the Zach variant in order to obtain a tuneable local phase shift. This comes at the expense of admitting the material lens in the beam path resulting in a partial loss of electrons, decoherence, inelastic losses and charging issues [29,51,17]. A commercially available alternative, the Volta phase plate, uses local charge buildup on an insulating electron transparent film; but also here the free space requirement is dropped resulting in possi- ble artefacts like drift, (in)elastic scattering, decoherence, uncontrolled charge and discharge [11,12,13,10]. So far, most phase plates have focused on phase contrast improvement and typically consist of a single region in space that is shifted in phase with respect to the rest of the wave that is left mostly un- altered. Increased flexibility in phase manipulation of electron waves recently gained considerable attention, showing exotic phase profiles as in electron vor- tices [5,40,59,64,2,15], Airy waves [65,34,24], helicon beams [52] or even the creation of an institute logo by modulating the phase profile of the wave [53]. Nearly all of these experiments make use of either holographic reconstruction using specifically crafted phase [18,20,47] and or amplitude gratings [64,40,63] or refractive elements crafted with specific height profiles [59,18,3,54,22]. Even though these methods provide unprecedented flexibility in phase program- ming, offering a rich resource for the exploration of e.g. non-diffractive optical modes, they suffer all from being static techniques. As such, a considerable time is required to create such phase plates and swap them in the aperture position of existing TEM microscopes which needs to be repeated with any new phase plate that is needed. In this paper we expand on these ideas by creating a rudimentary proof of concept of a programmable phase plate consisting of an array of 2x2 einzel lenses, offering full dynamic control over the 4 individual phase plate elements. Even though the design has clear drawbacks in terms of total transmissivity, 3 and limited pixel count, it nevertheless constitutes an important step towards a more generic programmable phase plate for TEM. In the remainder, we will give details on the design and fabrication of this phase plate followed by the first experimental results proving that a programmable multi-element phase plate for electrons is feasible. We continue by discussing the uses of such a limited phase plate and speculate about how much upscaling would be required to enable novel and exciting methods in the TEM [24]. It has to be noted here, that several alternative physical principles exist that can be exploited to create similar or even better devices, but we focus here on the description of what we believe to be the first truly versatile programmable phase array for electrons. 2 Experiment In order to prepare an array of electrostatic einzel lenses as sketched in fig. 1.a, a set of cylindrical electrodes needs to be created, sandwiched between two ground planes. Here we simplified the setup considerably by working with a single large ground plane on which four independent cylinder electrodes are placed. The omission of the top ground plane will lead to a minor leaking of the potential of one cylinder into the space above a neighbouring cylin- der electrode. This effect results in a small amount of crosstalk between the electrodes which can be partially compensated by altering the individual po- tentials to obtain the desired phase profile. Note that the name einzel lens can be confusing here, as the lensing function of these lenses is deliberately kept so low that the dominant effect is a pure phase shift depending on the total projected potential. This also means that at the low voltages that are being used, possible aberrations caused by the lenses are negligible compared to the phase shifting effect . A numerical solution of the Poisson equation for the half-einzel lens setup is presented in fig.2 making use of a Liebmann algortihm and shows the four elec- trodes at different potential in fig.2a. A cross section through 2 neighbouring electrodes in fig2b shows the fringing fields above the electrodes which lead to some cross talk. The projected potential is shown in fig.2c expressed in units of phase assuming 300 kV electrons and with a target phase of 0, 1/2π, π, 3/2π for each electrode respectively. The relative deviation from the intended flat phase in each electrode is plotted in fig.2d showing the cross talk effect leading to phase errors of less than 0.07π in this case. Future geometrical optimisation could reduce this effect further or alternatively a top ground electrode would make this effect negligible at the expense of a more complicated setup. The device fabrication starts from a commercial Si3N4 TEM sample grid with a 200 nm 250×250 µm2 square membrane. The grid was coated on both sides 4 with about 500 nm of gold, making use of a Balzers SCD 004 sputter coater. Masks were used to make sure the top and bottom layer are not electrically contacted. In the next step, four platinum cylinders of 1.4 µm height were deposited on the top gold surface by focused ion beam induced deposition (FIBID) using an FEI Helios Nanolab. The cylinders are placed on a square grid with a distance of approximately 2.5 µm from center to center. The four cylinders were then electrically insulated from each other by FIB milling the top layer of gold until reaching the underlying SiN membrane. Thanks to the large atomic number difference between the gold and the SiN membrane,it was possible to mill proper electrodes without milling through the SiN membrane, ensuring good insulation between the bottom ground electrode and the four top electrodes. As a final step, 1 µm circular holes were drilled in the Pt cylinders all the way through the stacked layers to let the electron beam pass. The 1 µm diameter was chosen in order to have an aspect ratio between diameter and height of the cylindrical electrodes of approximately 1. This provides a rather homogeneous potential profile inside the tubular electrode as can be judged from the electrostatic simulation in fig.2. The device was placed in a Dens Solutions Wildfire in-situ heating chip holder on top of a sacrificial heating chip, only used for contacting (heater window mechanically removed). The final contacts from the heating chip to the SiN grid are made with silver paint. Then, conductivity between the electrodes or between the electrodes and ground were carefully checked with a Keithley 2400 Source Meter and possible short circuits are corrected by further ion milling if necessary. The device was finally placed under vacuum conditions to allow for outgassing prior to insertion in the transmission electron microscope. Once the contact holder was inserted in an aberration corrected FEI Titan3 microscope (operating at 300 kV), the assembly was connected to a set of tune- able voltage sources. The microscope was operated in Lorentz mode with a strongly defocused monochromator in order to obtain sufficient spatial coher- ence in the sample plane to cover the four einzel lens elements. The apertures were then illuminated with a very good approximation of an electron plane wave. By going in diffraction mode, it was possible to observe the far field pattern of the four apertures. The diffraction pattern was recorded both in focus and defocused by 500 nm to better visualize the interference formed by the four individual electron wave patches that travel through the four aperture holes. When all electrodes are grounded, the pattern displayed in fig.3 is obtained. One can observe a clear constructive interference between the four patches in the defocused case. Another pattern was obtained by applying a voltage to the two right electrodes until destructive interference between the left/right patch occurs. This requires approximately 300 mV with some deviation over the different electrodes due to remaining leakage current issues between the 5 Fig. 1. (a) sketch of an array of weakly excited electrostatic einzel lenses individ- ually controlling the phase of 4 electron beams which recombine in the far field to form a programmable interference pattern. (b) SEM image of a proof of concept implementation of a 2x2 phase plate prepared as described in this paper. (c) TEM image of the aperture showing the geometry of the 4 aperture holes as well as the limited fill factor. electrodes and the height of each cylinder not being exactly the same. This demonstrates that indeed the pixel elements influence the phase of the electron wave by π. It is then possible to estimate the sensitivity of the phase to the potential with a back-of the-envelope calculation making use of the interaction constant of σ = 6.5 V −1µm−1 for 300 kV electrons and a tube length of 1.4 µm. For a π phase shift, this would require 350 mV which is in approximate agreement with the experimental values. A more accurate estimate can be made with the potential simulation presented in fig.2 and leads to an estimated 205 mV needed to obtain pi phase shift, more accurately taking into acount potential inhomogeneities and fringe fields. The result however depends in an sensitive manner on the exact geometry which only approximately matches the actual shape of the device. In a similar manner, a voltage difference between up and down electrodes can be applied. As expected, the interference pattern has now up down symmetry with a destructive interference between the up down patches. A quadrupolar pattern and vortex pattern can also be generated, proving that a functional 2x2 programmable phase plate has been created. The obtained interference patterns very closely match the simulated patterns given in fig.4, assuming a true flat programmable phase plate device. 3 Discussion The above experimental results demonstrate the feasibility of a multi-element programmable phase plate for use in electron microscopy. This holds clear promise for future work but several shortcomings of the current implementa- 6 Fig. 2. Simulated potential of the setup (a) showing the four electrodes at different potential of 0, 103, 206, 309 mV in order to obtain 0, 1/2π, π, 3/2π phase shift. A slice through 2 neighbouring electrodes (b) shows the potential homogeneity inside the cylinders with fringe fields extending above the electrodes due to the missing top ground plane. The projected potential is shown in (c) in units of phase shift assuming 300 kV electrons. The difference with the intended phase plate of 0, 1/2π, π, 3/2π is shown in (d) with a maximum deviation that remains below 0.07π as a result from the cross talk between the neighbouring pixels due to the lack of a top ground plane. Fig. 3. (top row) Set of diffraction pattern for 5 different applied potential config- urations leading to an approximate realisation of the phase symmetry as indicated on each panel. (bottom row) Defocused diffraction patterns for the same potential configuration showing more directly the constructive and destructive interference features between the 4 patches as indicated by arrows. tion will need to be addressed. The most important shortcoming lies in the inherent material making up the pixel element electrodes, blocking part of the electron beam. In the current demonstration this allows only about 12% of the incoming beam to pass through the 4 individual lenses. Undoubtedly this so-called fill-factor can be increased substantially in later designs, but it is hard to imagine a design which would allow passing substantially more than 50%. The fill-factor will of-course depend heavily on the micro machining or lithographic capabilities that will be used in further iterations of the design. As long as the phase plate is used in setups that shape the electron beam before the sample, this does not have to be a significant drawback, as modern instruments often provide more current or electron dose than the sample can handle, and losing a fraction of this current would not limit the usability of 7 Fig. 4. Simulated diffraction patterns assuming an ideal programmable phase shift- ing device with dimensions matching the experiment. Note the strong similarity in the intensity patterns for both in focus (top row) and defocused diffraction pattern series. the device. Changing the phase behind the sample, as typical in e.g. setups for Zernike weak phase imaging, would suffer significantly from a less than unity transmitivity and it seems rather unfavourable to block electrons that carry precious information on beam sensitive materials. If however, this loss of electrons is compensated by a greater gain in contrast, it could still turn out beneficial. The blocking nature of the einzel lens structure shares however the significant drawback with other electrostatic Zernike phase plate designs that it throws away important low frequency information when used in the back focal plane, after the sample [50,51]. In order to upscale the device to a higher pixel count, lithographic techniques will be required and interconnect density may quickly put a limit to the maximum attainable number of pixels that each need to be individually contacted to a programmable voltage source. Matrix adressing methods such as those used in dynamic random acces mem- ories could solve this, but require a nonlinear switching element, such as e.g. a transistor, integrated in the vicinity of each pixel element. Implementing such schemes should allow for pixel arrays of thousands to millions of pixel elements. One could wonder how many elements would be needed to provide ultimate flexibility in phase shaping the electron beam. As a demonstration, we compare the usefulness of a modest 55 and more elaborate 2205 pixel phase plate to act as a probe-Cs corrector device in fig 5. We assumed a spherical aberration coefficient of Cs = 1 mm at 300 keV and compare to an uncor- rected system at Scherzer defocus and to a fully corrected system, fixing the opening (half) angle at 20 mrad. From the simulation we see that the 55 pixel variant shows some benefit in terms of reduced tails on the probe intensity (b,f) which results in a sharper intensity radial profile in fig.5.i as compared to the uncorrected situation. Increasing the pixel count to 2205 results in a much sharper probe with a very simmilar profile (c,g) as the diffraction lim- ited profile of the fully corrected situation (d,h). This demonstrates that such phase plates could eventually start replacing magnetic multipole aberration 8 correctors in the probe-forming system of TEM microscopes, but clearly a substantial amount of pixels would be needed. A full exploration of the ideal position, shape and number of pixels for aberration correction with a pro- grammable phase plate will provide a rich arrea of further research, but is beyond the scope of the current discussion. Nevertheless, aberration correc- tion was not the first goal of the current implementation and low pixel element devices like the one presented here, or slightly upscaled can serve as a very useful experimenting platform for beam shaping and exploring other appli- cations that focus more on tuning the phase of the electron wave to bring out specific contrast. Examples include the study of non-diffracting electron beams[4,21,35,42,39,55,65], symmetry mapping of plasmonic excitations[23], mapping of magnetic fields [64,7,33,19,48,49] or edge contrast enhancement [33]. In optics, very useful spatial light modulators exist consisting of only 37 elements [38], even though they often provide a linear phase ramp in each pixel element which can also be imagined for electrons with a slight compli- cation of the design where each einzel lens electrode is split in 3 separate planar electrodes providing a linear potential profile in triangular or hexag- onal pixel elements. The optimised pattern of pixel elements depends on the intended use, but in low pixel count designs, it is likely that some form of radial pattern will be the most efficient. In order to demonstrate some of the capabilities that e.g. a radially oriented 55 pixel phase plate would bring, we show in fig. 6 the capability of such a simple phase plate to create approxi- mations to some well-known Hermite Gaussian and Laguerre Gaussian modes that are abundant in optics research. This capability would open up the field of beam shaping TEM providing a very desirable flexibility in the quantum state of the electron probe, much like what current spatial light modulators offer in optics. The fact that such a phase plate could rapidly switch between such modes could allow for differential measurements bringing out the specific contrast that comes with a certain probe symmetry and directly compare it to a near-simultaneous experiment done with another probe symmetry. As such, this development could bring the field of beam shaping to a significantly higher level. Given the above demonstration that an upscaled version of the presented 2x2 device could have multiple uses, we can look into possible obstacles that could hinder progress. Indeed, the presence of the pixel electrodes can have unwanted effects, such as charging or decoherence due to thermal current flowing in the electrode material [61,60,31]. The current implementation did not show either of these effects to be observable, but increasing the pixel density could eventually make these effects more prominent. In this respect, the short length (1.4 µm) over which the electrons interact with the pixel electrodes, helps to limit the decoherence effect substantially as they are expected to scale with interaction length and inversely with the square of tube radius [60]. Cooling could further help to keep decoherence low, but is rather unattractive from a practical point of view. Contamination, could be another limitation of the 9 Fig. 5. Simulated performance of a programmable phase plate as a Cs corrector assuming 300 keV and Cs=1 mm. Two phase plates are simulated, a modest 55 pixel plate and a more demanding 2205 pixel phase plate. The defocus of the lens system was optimised to -75 nm and -250 nm respectively to reach the optimum probe shape. The result is compared to an ideal aberration free system with the same round aperture with opening angle 20 mrad. The resulting probe intensities are shown respectively in (e,f,g,h), the images have dimension 1x1 nm. The azimuthally averaged probe intensity is shown in (i). Note that the low pixel count phase plate in (b) does not significantly improve the probe size but does improve the tails. Introducing significantly more pixels (c) approaches the ideal case quite closely. Note that even though the probe appears similar to the ideal diffraction limited case (d), a significant amount of the probe intensity is scattered to higher angles due to the narrow pixel shape function which would lead to an increased background when using this probe for imaging. The colorwheel shown as inset in (a) shows the colorscale used in (a,b,c,d) to depict both amplitude as intensity and phase as hue. 10 Fig. 6. A series of exotic electron beams prepared with a modest programmable phase plate (55 pixels, 41% fill-factor) approximating (from left to right) HG0,1, HG1,0, HG1,1, LG0,1, LG0,2. This shows that even a modest programmable phase plate can be very useful for producing exotic beam types that where hithertho difficult to produce in a TEM requiring time consuming and static (holographic) phase plates. design as small amounts of contamination could block pixels and render the whole array useless. This could be prevented or overcome by either including a heating element and or by allowing for easy replacement of a mass produced phase plate chip whenever this situation occurs. An important advantage of the presented device is the relative insensitivity of the performance to the quality of the voltage sources driving the pixel electrodes. Indeed in the current dimensions, a voltage of the order of a few 100 mV suffices to create a 2π phase shift. As phase is defined modulo 2π and because neighbouring patches of electron waves are divided by opaque walls of the pixel elements, there is no need to apply for more phase shift than this, similar to an optical Fresnel lens. This leads to a very relaxed requirement on the quality and noise performance of the voltage source. Indeed, even an idealised 8 bit digital to analog converter could provide a more than suffi- cient 2π/256 phase resolution. This relative insensitivity of the phase to the potential also allows for very fast settling times in combination with a very low capacitance of the pixel elements with respect to each other and to the ground plane. As long as all pixels are driven from individual voltage sources (DA converters), a response speed in the range of micro to nanoseconds seems entirely feasible. Matrix addressing might limit this speed considerably de- pending on the design. Such rapidly changing phase plates could allow for autotuning and iterative measurement schemes which profit from the total absence of hysteresis effects that hamper all ferromagnetic core based optical elements in conventional TEM instruments. 11 4 Conclusion We have demonstrated a proof of concept device that allows to dynamically control the phase of 2x2 segments of a coherent electron beam in a transmis- sion electron microscope. The experimental implementation and first results demonstrate that such a device holds promise for upscaling towards a more useful higher number of pixels. Several design considerations and directions for further research are discussed. These make plausible that the presented proof of concept marks just the beginning of an exciting development that could alter the way how one thinks about electron optics, providing vastly increased flexibility, speed, repeatability and offering novel iterative measurement pro- tocols that are difficult if not impossible to implement with current electron optical technology. 5 acknowledgments J.V. and A.B. acknowledge funding from the Fund for Scientific Research Flan- ders FWO project G093417N and the European Research Council under the 7th Framework Program (FP7), ERC Starting Grant 278510 VORTEX. The Qu-Ant-EM microscope used in this work was partly funded by the Hercules fund from the Flemish Government. MdH acknowledges financial support from the ANR-COSMOS (ANR-12-JS10-0002). MdH and ML acknowledge funding from the Laboratoire d'excellence LANEF in Grenoble (ANR-10-LABX-51- 01). References [1] P E Batson, N Dellby, and O L Krivanek. Sub-Angstrom Resolution Using Aberration Corrected Electron Optics. 418(6898):617–620, aug 2002. [2] A. B´ech´e, R. Juchtmans, and J. Verbeeck. Efficient creation of electron vortex beams for high resolution STEM imaging. Ultramicroscopy, pages 1–8, may 2016. [3] A. B´ech´e, R Winkler, H Plank, F Hofer, and J Verbeeck. Focused electron beam induced deposition as a tool to create electron vortices. Micron, 80:34–38, jan 2016. [4] M V Berry and N L Balazs. Nonspreading Wave Packets. American Journal of Physics, 47(3):264–267, mar 1979. 12 [5] K.Y. Bliokh, I.P. Ivanov, G. Guzzinati, L. Clark, R. Van Boxem, A. B´ech´e, R. Juchtmans, M.A. Alonso, P. Schattschneider, F. Nori, and J. Verbeeck. Theory and applications of free-electron vortex states. Physics Reports, 690:1– 70, may 2017. [6] Martin J Booth. Adaptive Optical Microscopy: The Ongoing Quest for a Perfect Image. Light: Science & Applications, 3(4):e165, apr 2014. [7] L Clark, A. B´ech´e, G Guzzinati, and J Verbeeck. Quantitative measurement of orbital angular momentum in electron microscopy. Physical Review A, 89(5):053818, may 2014. [8] Pere Clemente, Vicente Dur´an, V´ıctor Torres-Company, Enrique Tajahuerce, and Jes´us Lancis. Optical encryption based on computational ghost imaging. Optics letters, 35(14):2391–3, jul 2010. [9] Daniel Clery. Building {{James Webb}}: The Biggest, Boldest, Riskiest Space Telescope. feb 2016. [10] R Danev, B Buijsse, M Khoshouei, J M Plitzko, and W Baumeister. Volta Potential Phase Plate for In-Focus Phase Contrast Transmission Electron Microscopy. Proceedings of the National Academy of Sciences, 111(44):15635– 15640, nov 2014. [11] Radostin Danev and Wolfgang Baumeister. Cryo-{{EM}} Single Particle Analysis with the {{Volta}} Phase Plate. 5, mar 2016. [12] Radostin Danev, Robert M Glaeser, and Kuniaki Nagayama. Practical Factors Affecting the Performance of a Thin-Film Phase Plate for Transmission Electron Microscopy. 109(4):312–325, mar 2009. [13] Radostin Danev and Kuniaki Nagayama. Single Particle Analysis Based on {{Zernike}} Phase Contrast Transmission Electron Microscopy. Journal of Structural Biology, 161(2):211–218, feb 2008. [14] Richard Davies and Markus Kasper. Adaptive {{Optics}} for {{Astronomy}}. Annual Review of Astronomy and Astrophysics, 50(1):305–351, sep 2012. [15] Christopher J. Edgcombe. Phase Plates for Transmission Electron Microscopy. In Advances in Imaging and Electron Physics, chapter 2, pages 61–102. 2017. [16] D L Fried. Optical {{Resolution Through}} a {{Randomly Inhomogeneous Medium}} for {{Very Long}} and {{Very Short Exposures}}. Journal of the Optical Society of America, 56(10):1372, oct 1966. [17] Nicole Frindt, Marco Oster, Simon Hettler, Bjorn Gamm, Levin Dieterle, In- Wolfgang Kowalsky, Dagmar Gerthsen, and Rasmus R Schroder. {{Focus Electrostatic Zach Phase Plate Imaging}} for {{Transmission Electron Microscopy}} with {{Tunable Phase Contrast}} of {{Frozen Hydrated Biological Samples}}. Microscopy and Microanalysis, 20(01):175–183, feb 2014. 13 [18] Vincenzo Grillo, Gian Carlo Gazzadi, Ebrahim Karimi, Erfan Mafakheri, Robert W. Boyd, and Stefano Frabboni. Highly efficient electron vortex beams generated by nanofabricated phase holograms. Applied Physics Letters, 104(4):1–5, 2014. [19] Vincenzo Grillo, Tyler R. Harvey, Federico Venturi, Jordan S. Pierce, Roberto Balboni, Fr´ed´eric Bouchard, Gian Carlo Gazzadi, Stefano Frabboni, Amir H. Tavabi, Zi-An Li, Rafal E. Dunin-Borkowski, Robert W. Boyd, Benjamin J. McMorran, and Ebrahim Karimi. Observation of nanoscale magnetic fields using twisted electron beams. Nature Communications, 8(1):689, dec 2017. [20] Vincenzo Grillo, Ebrahim Karimi, Gian Carlo Gazzadi, Stefano Frabboni, Mark R Dennis, and Robert W Boyd. Generation of {{Nondiffracting Electron Bessel Beams}}. Physical Review X, 4(1), jan 2014. [21] Vincenzo Grillo, Ebrahim Karimi, Gian Carlo Gazzadi, Stefano Frabboni, Mark R Dennis, and Robert W Boyd. Generation of {{Nondiffracting Electron Bessel Beams}}. Physical Review X, 4(1), jan 2014. [22] Vincenzo Grillo, Amir H. Tavabi, Emrah Yucelen, Peng-Han Lu, Federico Venturi, Hugo Larocque, Lei Jin, Aleksei Savenko, Gian Carlo Gazzadi, Roberto Balboni, Stefano Frabboni, Peter Tiemeijer, Rafal E. Dunin-Borkowski, and Ebrahim Karimi. Towards a holographic approach to spherical aberration correction in scanning transmission electron microscopy. Optics Express, 25(18):21851, sep 2017. [23] Giulio Guzzinati, Armand B´ech´e, Hugo Louren¸co-Martins, J´erome Martin, Mathieu Kociak, and Jo Verbeeck. Probing the Symmetry of the Potential of Localized Surface Plasmon Resonances with Phase-Shaped Electron Beams. Nature Communications, 8:14999, apr 2017. [24] Giulio Guzzinati, Laura Clark, Armand B´ech´e, Roeland Juchtmans, Ruben Van Boxem, Michael Mazilu, and Jo Verbeeck. Prospects for versatile phase manipulation in the TEM: Beyond aberration correction. Ultramicroscopy, 151:85–93, apr 2015. [25] M Haider, P Hartel, H Muller, S Uhlemann, and J Zach. Current and Future Aberration Correctors for the Improvement of Resolution in Electron Microscopy. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 367(1903):3665–3682, sep 2009. [26] M Haider, S Uhlemann, E Schwan, H Rose, B Kabius, and K Urban. Electron Microscopy Image Enhanced. 392(6678):768–769, 1998. [27] P.W. Hawkes and E. Kasper. Front Matter. Elsevier, 1994. [28] Stefan W Hell and Jan Wichmann. Breaking the Diffraction Resolution Limit by Stimulated Emission: Stimulated-Emission-Depletion Fluorescence Microscopy. Optics Letters, 19(11):780, jun 1994. [29] S Hettler, M Dries, J Zeelen, M Oster, R R Schroder, and D Gerthsen. High- Resolution Transmission Electron Microscopy with an Electrostatic {{Zach}} Phase Plate. New Journal of Physics, 18(5):53005, may 2016. 14 [30] Kuang-Sheng Hong, Jing Wang, Alexey Sharonov, Dinesh Chandra, Joanna Aizenberg, and Shu Yang. Tunable microfluidic optical devices with an integrated microlens array. Journal of Micromechanics and Microengineering, 16(8):1660–1666, 2006. [31] Archie Howie. Addressing {{Coulomb}}'s Singularity, Nanoparticle Recoil and {{Johnson}}'s Noise. Journal of Physics: Conference Series, 522:12001, jun 2014. [32] Yasunori Igasaki, Fanghong Li, Narihiro Yoshida, Haruyoshi Toyoda, Takashi Inoue, Naohisa Mukohzaka, Yuji Kobayashi, and Tsutomu Hara. High Efficiency Electrically-Addressable Phase-Only Spatial Light Modulator. Optical Review, 6(4):339–344, 1999. [33] Roeland Juchtmans, Laura Clark, Axel Lubk, and Jo Verbeeck. Spiral phase plate contrast in optical and electron microscopy. Physical Review A, 94(2):023838, aug 2016. [34] Ido Kaminer, Jonathan Nemirovsky, Mikael Rechtsman, Rivka Bekenstein, and Mordechai Segev. Self-accelerating Dirac particles and prolonging the lifetime of relativistic fermions. Nature Physics, 11(January):261–267, 2015. [35] M R Lapointe. Review of Non-Diffracting {{Bessel}} Beam Experiments. Optics & Laser Technology, 24(6):315–321, dec 1992. [36] Martin Linck, Peter Hartel, Stephan Uhlemann, Frank Kahl, Heiko Muller, Joachim Zach, Max. Haider, Marcel Niestadt, Maarten Bischoff, Johannes Biskupek, Zhongbo Lee, Tibor Lehnert, Felix Borrnert, Harald Rose, and Ute Kaiser. Chromatic {{Aberration Correction}} for {{Atomic Resolution TEM Imaging}} from 20 to 80 {{kV}}. Physical Review Letters, 117(7), aug 2016. [37] Reza Nasiri Mahalati, Ruo Yu Gu, and Joseph M Kahn. Resolution Limits for Imaging through Multi-Mode Fiber. Optics Express, 21(2):1656, jan 2013. [38] Silvestre Manzanera, Michael A. Helmbrecht, Carl and Austin Roorda. MEMS segmented-based adaptive optics scanning laser ophthalmoscope. Biomedical Optics Express, 2(5):1204, may 2011. J. Kempf, [39] D McGloin and K Dholakia. Bessel Beams: {{Diffraction}} in a New Light. Contemporary Physics, 46(1):15–28, jan 2005. [40] Benjamin J McMorran, Amit Agrawal, Ian M Anderson, Andrew a Herzing, Henri J Lezec, Jabez J McClelland, and John Unguris. Electron vortex beams with high quanta of orbital angular momentum. Science (New York, N.Y.), 331(6014):192–195, 2011. [41] F Merkle, P Kern, P Lena, F Rigaut, J C Fontanella, G Rousset, C Boyer, J P Gaffard, and P Sagourel. Successful Tests of Adaptive Optics. The Messenger, 58:1–4, 1989. [42] J E Morris, T Cizm´ar, H I C Dalgarno, R F Marchington, F J Gunn-Moore, and K Dholakia. Realization of Curved {{Bessel}} Beams: Propagation around Obstructions. Journal of Optics, 12(12):124002, dec 2010. 15 [43] Libor Mrna, Martin Sarbort, Simon Rerucha, and Peter Jedlicka. Adaptive optics for control of the laser welding process. EPJ Web of Conferences 48, 17:1–7, 2013. [44] H Rose. Phase-{{Contrast}} in {{Scanning}}-{{Transmission Electron}}- {{Microscopy}}. 39(4):416–436, 1974. [45] H Rose. Prospects for Aberration-Free Electron Microscopy. 103(1):1–6, apr 2005. [46] Harald Rose. Geometrical Charged-Particle Optics, volume 142 of Springer Series in Optical Sciences. Springer Berlin Heidelberg, Berlin, Heidelberg, 2009. [47] Halina Rubinsztein-Dunlop, Andrew Forbes, M V Berry, M R Dennis, David L Andrews, Masud Mansuripur, Cornelia Denz, Christina Alpmann, Peter Banzer, Thomas Bauer, Ebrahim Karimi, Lorenzo Marrucci, Miles Padgett, Monika Ritsch-Marte, Natalia M Litchinitser, Nicholas P Bigelow, C Rosales-Guzm´an, A Belmonte, J P Torres, Tyler W Neely, Mark Baker, Reuven Gordon, Alexander B Stilgoe, Jacquiline Romero, Andrew G White, Robert Fickler, Alan E Willner, Guodong Xie, Benjamin McMorran, and Andrew M Weiner. Roadmap on Structured Light. Journal of Optics, 19(1):13001, jan 2017. [48] T. Schachinger, S. Loffler, A. Steiger-Thirsfeld, M. Stoger-Pollach, S. Schneider, D. Pohl, B. Rellinghaus, and P. Schattschneider. EMCD with an electron vortex filter: Limitations and possibilities. Ultramicroscopy, 179:15–23, aug 2017. [49] P Schattschneider, S Loffler, M Stoger-Pollach, and J Verbeeck. Is magnetic chiral dichroism feasible with electron vortices? Ultramicroscopy, 136:81–5, jan 2014. [50] K Schultheiss, F P´erez-Willard, B Barton, D Gerthsen, and R R Schroder. Fabrication of a {{Boersch}} Phase Plate for Phase Contrast Imaging in a Transmission Electron Microscope. Review of Scientific Instruments, 77(3):33701, mar 2006. [51] Katrin Schultheiss, Joachim Zach, Bjoern Gamm, Manuel Dries, Nicole Frindt, Rasmus R Schroder, and Dagmar Gerthsen. New {{Electrostatic Phase Plate}} for {{Phase}}-{{Contrast Transmission Electron Microscopy}} and {{Its Application}} for {{Wave}}-{{Function Reconstruction}}. Microscopy and Microanalysis, 16(06):785–794, dec 2010. [52] Roy Shiloh and Ady Arie. {{3D}} Shaping of Electron Beams Using Amplitude Masks. 177:30–35, jun 2017. [53] Roy Shiloh, Yossi Lereah, Yigal Lilach, and Ady Arie. Sculpturing the Electron Wave Function Using Nanoscale Phase Masks. 144:26–31, sep 2014. [54] Roy Shiloh, Roei Remez, Peng-Han Lu, Lei Jin, Yossi Lereah, Amir H Tavabi, Rafal E Dunin-Borkowski, and Ady Arie. Spherical Aberration Correction in a Scanning Transmission Electron Microscope Using a Sculpted Foil. arXiv:1705.05232 [physics], may 2017. 16 [55] G A Siviloglou, J Broky, A Dogariu, and D N Christodoulides. Observation of {{Accelerating Airy Beams}}. Physical Review Letters, 99(21), nov 2007. [56] C Slinger, C Cameron, and M Stanley. Computer-Generated Holography as a Generic Display Technology. 38(8):46–53, aug 2005. [57] Ehud Steinhaus and S G Lipson. Bimorph piezoelectric flexible mirror. Journal of the Optical Society of America, 69(3):478–481, 1979. [58] R M Tromp, J B Hannon, A W Ellis, W Wan, A Berghaus, and O Schaff. A New Aberration-Corrected, Energy-Filtered {{LEEM}}/{{PEEM}} Instrument. {{I}}. {{Principles}} and Design. 110(7):852–861, jun 2010. [59] Masaya Uchida and Akira Tonomura. Generation of electron beams carrying orbital angular momentum. Nature, 464(7289):737–9, apr 2010. [60] Stephan Uhlemann, Heiko Muller, Peter Hartel, Joachim Zach, and Max. Haider. Thermal {{Magnetic Field Noise Limits Resolution}} in {{Transmission Electron Microscopy}}. Physical Review Letters, 111(4), jul 2013. [61] Stephan Uhlemann, Heiko Muller, Joachim Zach, and Max. Haider. Thermal magnetic field noise: Electron optics and decoherence. Ultramicroscopy, 151:199–210, apr 2015. [62] Gleb Vdovin and P. M. Sarro. Flexible mirror micromachined in silicon. Applied Optics, 34(16):2968, jun 1995. [63] J. Verbeeck, P. Schattschneider, S. Lazar, M. Stoger-Pollach, S. Loffler, A. Steiger-Thirsfeld, and G. Van Tendeloo. Atomic scale electron vortices for nanoresearch. Applied Physics Letters, 99(20):203109, 2011. [64] J Verbeeck, H Tian, and P Schattschneider. Production and application of electron vortex beams. Nature, 467(7313):301–304, 2010. [65] Noa Voloch-Bloch, Yossi Lereah, Yigal Lilach, Avraham Gover, and Ady Arie. Generation of Electron {{Airy}} Beams. 494(7437):331–335, feb 2013. [66] J Y Wang. Optical Resolution through a Turbulent Medium with Adaptive Phase Compensations*. Journal of the Optical Society of America, 67(3):383, mar 1977. [67] F Zernike. Phase Contrast, a New Method for the Microscopic Observation of Transparent Objects. 9(7):686–698, jul 1942. 17
1909.03318	1	1909	2019-09-07T18:32:52	Bright and Efficient Perovskite Light Emitting Electrochemical Cells Leveraging Ionic Additives	[ "physics.app-ph", "physics.optics" ]	Perovskite light-emitting diodes (PeLEDs) have drawn considerable attention for their favorable optoelectronic properties. Perovskite light-emitting electrochemical cells (PeLECs) _ devices that utilize mobile ions _ have recently been reported but have yet to reach the performance of the best PeLEDs. We leveraged a poly(ethylene oxide) electrolyte and lithium dopant in CsPbBr3 thin films to produce PeLECs of improved brightness and efficiency. In particular, we found that a single layer PeLEC from CsPbBr3:PEO:LiPF6 with 0.5% wt. LiPF6 produced highly efficient (22 cd/A) and bright (~15000 cd/m2) electroluminescence. To understand this improved performance among PeLECs, we characterized these perovskite thin films with photoluminescence (PL) spectroscopy, scanning electron microscopy (SEM), atomic force microscopy (AFM), X-ray photoelectron spectroscopy (XPS), and X-ray diffraction (XRD). These studies revealed that this optimal LiPF6 concentration improves electrical double layer formation, reduces the occurrence of voids, charge traps, and pinholes, and increases grain size and packing density.	physics.app-ph	physics	Bright and Efficient Perovskite Light Emitting Electrochemical Cells Leveraging Ionic Additives Masoud Alahbakhshi1, Aditya Mishra2, Ross Haroldson3, Arthur Ishteev6, Jiyoung Moon2, Qing Gu1, Jason D. Slinker2,3* and Anvar A. Zakhidov3,4,5* 1Department of Electrical and Computer Engineering, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 2Department of Materials Science and Engineering, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 3Department of Physics, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 4NanoTech Institute, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 5Department of Nanophotonics and Metamaterials, ITMO University, St. Petersburg, Moscow, Russia. 6Laboratory of Advanced Solar Energy, NUST MISiS, Moscow, 119049, Russia AUTHOR INFORMATION Corresponding Author [email protected], [email protected] ABSTRACT: Perovskite light emitting diodes (PeLEDs) have drawn considerable attention for their favorable optoelectronic properties. Perovskite light emitting electrochemical cells (PeLECs) -- devices that utilize mobile ions -- have recently been reported but have yet to reach the performance of the best PeLEDs. We leveraged a poly(ethylene oxide) electrolyte and lithium dopant in CsPbBr3 thin films to produce PeLECs of improved brightness and efficiency. In particular, we found that a single layer PeLEC from CsPbBr3:PEO:LiPF6 with 0.5% wt. LiPF6 produced highly efficient (22 cd/A) and bright (~15000 cd/m2) electroluminescence. To understand this improved performance among PeLECs, we characterized these perovskite thin films with photoluminescence (PL) spectroscopy, scanning electron microscopy (SEM), atomic force microscopy (AFM), X-ray photoelectron spectroscopy (XPS), and X-ray diffraction (XRD). These studies revealed that this optimal LiPF6 concentration improves electrical double layer formation, reduces the occurrence of voids, charge traps, and pinholes, and increases grain size and packing density. TOC GRAPHICS Perovskite light-emitting diodes (PeLEDs) based on inorgano−organometallic halide perovskites, such as CH3NH3PbX3 and CsPbX3 (X = Cl, Br, or I), have attracted much attention due to their low-temperature solution processability, high color purity with narrow spectral width (FWHM of 20 nm), band gap tunability and large charge carrier mobility.1-4 To date, devices based on these perovskites have achieved high luminance in excess of 10000 cd/m2 with high efficiencies (EQE ~10%), comparable to organic LEDs and quantum dot (QD) LEDs.1-7 Interestingly, effects such as hysteresis and high capacitance in perovskite semiconductor devices suggest that ion motion can largely influence device operation. In this vein, researchers have recently been investigating perovskite materials in light-emitting electrochemical cell (LEC) architectures instead of traditional LEDs.8-11 These LEC devices (PeLEC leverage ion redistribution to achieve balanced and high carrier injection, resulting in high electroluminescence efficiency. Due to this mechanism, LEC devices can be prepared from a simple architecture consisting of a single semiconducting composite layer sandwiched between two electrodes. In addition, they can operate at low voltages below the bandgap, yielding highly efficient devices. Recently, perovskite LECs (PeLECs) have been reported and show promise as electroluminescent devices.8-11 However, these PeLECs are generally limited to luminance maxima of 1000 cd/m2 or lower, below what has been typically observed in PeLEDs. This disparity suggests that further understanding and refinement of PeLEC materials and devices could produce significant improvements of brightness, efficiency, and other performance metrics. To this end, we fabricated a highly efficient (22 cd/A) and bright (~15000 cd/m2) single layer LEC based on a cesium lead halide perovskite, CsPbBr3. To achieve high performance, a polyelectrolyte and additive mobile ions were carefully selected to achieve optimal ionic redistribution and doping effects. To understand the nature of this performance and its correlation with materials properties, we characterize these perovskite thin films and devices with photoluminescence spectroscopy, scanning electron microscopy (SEM), atomic force microscopy (AFM), X-ray photoelectron spectroscopy (XPS), and X-ray diffraction (XRD). Figure 1. Illustration of PeLEC device operation and device characteristics. a) Initial LEC state showing ions uniformly distributed throughout the active layer. b) Intermediate LEC state showing cations drifting toward the cathode and anions toward the anode. c) Steady-state LEC operation with ions accumulated at the electrodes and light emission upon current injection. d) Current density and luminance versus voltage for In-Ga/CsPbBr3:PEO:LiPF6 LEC/ITO devices. The inset shows the layering of In-Ga/CsPbBr3:PEO:LiPF6/ITO. e) Maximum luminance and current efficiency as a function of LiPF6 concentration. We first consider the distinct stages of LEC device operation, which are illustrated in Figures 1a-c. Initially, ions are uniformly distributed in the film (Figure 1a). In response to an applied bias, cations drift toward and accumulate near the cathode, and anions likewise move toward and pack near the anode (Figure 1b). This leads to an electric double layer (EDL) formation at each electrode that induces higher electric fields, decreased width of the potential barriers (doping), and enhanced injection of electrons and holes that is insensitive to the workfunction of the electrodes (Figure 1c).12-15 These injected carriers are transported through the bulk and radiatively recombine in the center of the device. The key features needed for successful LEC operation are therefore: 1) A sufficient concentration of mobile anions and cations; 2) Efficient transport of ions through the bulk for balanced EDL formation at the anode and cathode, leading to efficient charge injection; 3) Facile transport of electrons and holes through the semiconductor (which, for our system, requires a percolating network of the perovskite); 4) Efficient light emission upon recombination of the electrons and holes in the bulk, typically supported by a high quantum yield of the film. In our specific case, high luminescence efficiency is supported by the spectral properties of the CsPbBr3 perovskite. To satisfy the other requirements, we introduce LiPF6 salt, a salt that we have previously used to attain high performance in LECs utilizing ionic transition metal complexes.16- 18 We prepared films with an optimal concentration of the polymer electrolyte poly(ethylene oxide) (PEO) and systematically studied the effect of LiPF6 salt addition. The LECs were constructed with a single layer of spin cast perovskite film using an ITO anode, a CsPbBr3:PEO:LiPF6 perovskite composite (1:0.8 weight ratio CsPbBr3:PEO, various Li weight fractions), and an In-Ga eutectic cathode, as illustrated in the inset of Figure 1d (see Supporting Information Figure S1 and text for details of fabrication and testing). Also in Figure 1d, the current density versus voltage (J vs V) and luminance versus voltage (L vs V) graph of our devices with different ratios of LiPF6 is shown. Figure 1e presents the maximum luminance and current efficiency from the data of Figure 1d. All of the devices showed green electroluminescence centered near 528 nm with a FWHM of 20 nm (See Supporting information Figure S2). The reference device (CsPbBr3:PEO with no LiPF6) shows a turn-on voltage (Von) of 2.5 V and a substantial maximum luminance of 8175 cd/m2 at 5.5 V, with a maximum current efficiency of 9.0 cd/A. This substantial luminance for a PeLEC is accounted for the optimized PEO electrolyte concentration within the film. As LiPF6 is added into the CsPbBr3:PEO film, the luminance and efficiency maxima increase gradually, peak at an optimal concentration, and then decrease. The best performance was achieved with a 0.5% LiPF6 weight percentage, which showed a Von of 1.9 V (below the bandgap), a maximum luminance of 14730 cd/m2 at 5.4 V, and a maximum current efficiency of 22.4 cd/A. Notably, the current efficiency (Figure 1e) was enhanced ~2.5X compared to the pristine CsPbBr3:PEO LEC. To the best of our knowledge, the highest values of luminance and current efficiency that we achieved are the best reported values so far for perovskite LECs (as distinct from LEDs). Full performance metrics for all device formulations are noted in Supporting Information Table S1, and comparisons to other literature efforts are reported in Supporting Information Table S3. Further investigation of the devices reveals other key features. Notably, all concentrations of LiPF6 enhanced current density at low operating voltage (Figure 1d), indicating LiPF6 initially assists with monopolar charge injection presumably through EDL formation and p- and n-doping effects. As voltage is increased (Figure 1d), turn on voltage is generally lowered by LiPF6 addition, denoting its ability to assist in bipolar injection. Luminance and current efficiency are enhanced by adding 0.2% or 0.5% LiPF6, but diminished by higher concentrations (Figure 1d), suggesting competing processes affect device performance. We also observe that the optimal 0.5% LiPF6 concentration considerably reduces the hysteresis in current density curves from voltage cycling (Supporting Information Figure S3). We also measured the stability of devices under constant voltage operation and found that 0.5% LiPF6 can improve the stability threefold in comparison with control devices (Supporting Information Figure S4), a favorable result in view of previous reports.19 We performed additional study to gain further mechanistic and phenomenological understanding of these observations. Figure 2. Photoluminescence properties of thin films of CsPbBr3 and CsPbBr3:PEO:LiPF6. a) Photoluminescence intensity as a function of time (λex= 405 nm). b) Photoluminescence quantum efficiency (PLQE) versus LiPF6 concentration for CsPbBr3:PEO:LiPF6 films. c) Photoluminescence spectra of CsPbBr3 and CsPbBr3:PEO:LiPF6. For further investigations of the optical, electronic, and morphological states of our perovskite films, we prepared CsPbBr3:PEO:LiPF6 thin films through an optimized one stage spin-coating process and vacuum treatment before annealing (Supporting Information Figure S5 and text). We subsequently measured the photoluminescence (PL) intensity versus time of these films (Figure 2a). For CsPbBr3 and CsPbBr3:PEO films, the PL intensity dramatically increases with time. As LiPF6 is added to CsPbBr3:PEO films, the PL dynamic trends towards constant intensity. This PL trend with lithium doping can be understood from reduced trapping.20 In the absence of lithium dopants, unfilled trap states are present, potentially due to grain boundaries, vacancies, and other imperfections that create nonradiative decay states in the middle of the optical gap. These trap states must first be filled with electronic carriers before steady PL can be achieved. Lithium salt addition produces filled trap states, consistent with doping concepts (see Supporting Information Figure S6), since Li+ promotes n-doping by electrons that fill the trap. The morphological studies described below can further clarify the nature of trap suppression. To understand the influence of LiPF6 doping on steady-state quantum yield, we measured the absolute photoluminescence quantum efficiency (PLQE) for CsPbBr3:PEO:LiPF6 films with an integrating sphere according to the methods described by de Mello et al. and Porrès et al.21-22 Figure 2b reveals that the PLQE increases from 29% for a CsPbBr3:PEO film to a maximum of 49% for a 0.5% LiPF6 film of CsPbBr3:PEO:LiPF6, and then is lowered as the LiPF6 concentration is further increased. This PLQE concentration trend of CsPbBr3:PEO:LiPF6 films qualitatively follows the concentration dependences of luminance and current efficiency for PeLEC devices (Figure 1e). This reinforces our assertion that LiPF6 suppresses nonradiative trapping states to improve emission yield in thin films and devices. PL spectra were also measured to ascertain the impact of lithium doping on optical emission from CsPbBr3:PEO:LiPF6 thin films (Figure 2c). The PL spectra show a gradual blue shift from λmax = 522 nm for pure CsPbBr3 thin film to λmax = 516 nm for the CsPbBr3:PEO:LiPF6 (5%) thin film. This blue shift may be attributed to the Burstein-Moss effect (Supporting Information Figure S6c). According to this theory, electrons populate states in the conduction band due to doping, promoting the Fermi level higher in the conduction band. Electrons from the top of the valence band can only be excited into empty conduction states above the Fermi level, consistent with the Pauli exclusion principle.23-24 Similar results have been reported for Sb and Bi doping in perovskites.25-26 Figure 3. Morphological study of CsPbBr3:PEO:LiPF6 thin films by SEM and AFM. a) SEM for thin films of I) CsPbBr3, II) CsPbBr3:PEO, III) CsPbBr3:PEO:LiPF6 (0.2%), IV) CsPbBr3:PEO:LiPF6 (0.5%), V) CsPbBr3:PEO:LiPF6 (1%), and VI) CsPbBr3:PEO:LiPF6 (2%). b) Surface roughness of CsPbBr3:PEO:LiPF6 thin films versus LiPF6 concentration as measured by AFM. The horizontal dotted line denotes the surface roughness of the pristine CsPbBr3 film. To further understand the morphological influence of the beneficial electrical and optical properties afforded by electrolyte and lithium addition, we studied thin perovskite films by scanning electron microscopy (SEM) and atomic force microscopy (AFM). The SEM image of the pristine CsPbBr3 film (Figure 3a,I) reveals randomly oriented grains and substantial pinholes in the film. Addition of PEO (Figure 3a,II) reduces the average grain and pinhole sizes, but significant pinholes still remain. Subsequent addition of low concentrations (≤ 2%) of LiPF6 in CsPbBr3:PEO increases the average grain size (Supporting Information Table S2) and decreases the number and size of pinholes (Figure 3a, III-VI, Supporting Information Figure S7). The largest and most monodispersed grain sizes are observed for 0.5% and 1.0% LiPF6 concentrations. Also, the 0.5% LiPF6 dopant film (Figure 3a, IV) provides the most continuous network of perovskite crystals with the lowest amount of pinholes. This correlates well with the high luminance and efficiency of the 0.5% LiPF6 doped LEC device (Figure 1e): reducing the number of pinholes limits detrimental leakage current, and the interpenetrating network of perovskite crystals supports facile transport of electrons and holes for efficient electroluminescence. On the other hand, excess LiPF6 (5%) was detrimental to the quality of the thin film, leading to discontinuous films (Supporting Information Figure S8). This is likely due to the formation of lithium dendrites and subsequent phase separation.27 In Figure 3b, we relate the average surface roughness of the CsPbBr3:PEO:LiPF6 films with various percentages of LiPF6 as measured by AFM (Supporting Information Figure S9). We observe a reduction in the root-mean-square roughness from 15.7 nm for the pristine CsPbBr3:PEO film to 11.9 nm for the 0.5% LiPF6 doped film, and an increase in surface roughness for higher concentrations. Again, this correlates with the optimal LiPF6 device concentration, as low surface roughness improves the spatial uniformity of the electroluminescence and limits pinhole formation. Hence, the overall interpretation from SEM and AFM analysis indicates that 0.5% LiPF6 corresponds to the optimal concentration for smooth films with large, percolating perovskite grains and minimal pinholes, all beneficial for superior device performance. Figure 4. High resolution X-ray photoelectron spectroscopy (XPS) and X-ray diffraction (XRD) of CsPbBr3:PEO:LiPF6 films, as well as a conceptual illustration of Li+-induced performance enhancement. High-resolution XPS spectral regions for a) Pb (4f5/2, 4f7/2) and b) Br (3d3/2, 3d5/2) peaks from CsPbBr3 and CsPbBr3:PEO:LiPF6 thin films of various LiPF6 weight ratios. c) XRD pattern of CsPbBr3 and CsPbBr3:PEO:LiPF6 thin films of various LiPF6 weight ratios. d) Conceptual illustration of the benefits afforded by LiPF6 doping in PeLECs. X-ray photoelectron spectroscopy (XPS) analysis of the high resolution spectra involving Pb 4f, Br 3d, C 1s, and F 1s (Figure 4a-b, and Supporting Information Figure S10) was conducted to further understand the effect of the LiPF6 dopant in CsPbBr3:PEO films. As shown in Figure 4a, the Pb 4f spectrum for the pristine CsPbBr3 film has two peaks associated with 4f5/2 and 4f7/2 orbitals located at 138.40 and 143.20 eV, respectively, which correspond to Pb2+ cations. The two Br 3d signals, corresponding to 3d3/2 and 3d5/2 peaks, are also clearly seen at 68.4 and 69.2 eV. Addition of PEO to the CsPbBr3 film shifts all of these peaks to higher binding energies, which is attributed to the interaction between PEO and metallic/ion Pb reported previously.28 Furthermore, we also observe that adding LiPF6 dopant to the CsPbBr3:PEO film further shifts these peaks. These spectral shifts are a clear indication of a lattice contraction of perovskite after Li+ incorporation, shorter Pb-Br bonds and higher binding energy for Pb 4f and Br 3d.20, 29 The presence of PEO and LiPF6 in films has been further confirmed from the C 1s and F 1s XPS spectra (Supporting Information Figure S10a-b). The adventitious C 1s signal is obvious at 285.2 eV for the pristine CsPbBr3 thin film. However, after PEO addition, a new C 1s peak appears at 287 eV, which corresponds to the C-O-C groups of the PEO.30 An F 1s spectral peak at 686.6 eV appears after adding LiPF6 to CsPbBr3:PEO, confirming the presence of the fluoride ion in the film. Also, we note that the photoemission of Li 1s could not be detected due to the very low sensitivity factor of Li. The detailed chemical bonding of the composite perovskite films was analyzed by deconvoluting the XPS peaks (Supporting Information Figure S10). Figure 4c illustrates the X-ray diffraction (XRD) spectra of CsPbBr3, CsPbBr3:PEO, and CsPbBr3:PEO:LiPF6. The thin film XRD spectra indicate the primary diffraction peaks at 15.21°, 21.46°, and 30.70° that corresponds to diffraction planes of (110), (112) and (220), respectively, in agreement with previous reports for an orthorhombic (Pnma) crystal structure.31-32 The peaks at 12.7° and 22.41° indicate the presence of a trigonal phase of Cs4PbBr6 with the diffraction planes of (102) and (213), respectively, as seen in literature, indicating a mixed phase of CsPbBr3 and Cs4PbBr6.33-35 In addition, we analyzed the peak widths and positions, finding successive contraction of the perovskite crystal lattice parameters with increasing amounts of LiPF6 (see Table S2), suggesting Li substitution in the perovskite lattice. Overall, these observations suggest several benefits of lithium salt addition that led to high performance (14730 cd/m2, 22.4 cd/A) among single layer PeLECs that are comparable to the most efficient PeLEDs, which are typically multilayer devices (Supporting Information Table S3). These benefits are summarized in Figure 4d. First, an overall consideration of the system and a detailed view of the device performance suggests that an optimal concentration of LiPF6 facilitates double layer formation, thus improving charge carrier injection. Second, SEM and AFM studies demonstrated that an optimal lithium concentration reduces the spacing between grain boundaries, leading to smooth, pinhole free films. This limits leakage current, enables high electron and hole conductivities for enhanced carrier transport. PL study correlated with XPS and XRD studies show that lithium reduces the crystal lattice parameters through filling traps and voids, reducing sources of nonradiative losses for highly efficient light emission in thin films and devices. A detailed balance of all of these features thus leads to high performance LECs of simple, single-layer architectures for next generation optoelectronic applications. ASSOCIATED CONTENT Supporting information available: experimental methods, electroluminescence spectra, luminance versus voltage and time with associated data, illustrations of PL mechanisms, SEM and AFM images, XPS spectra, and XRD data. ACKNOWLEDGMENT J.D.S. acknowledges support from the National Science Foundation (ECCS 1906505). Q.G. acknowledges support from the Welch Foundation (AT-1992-20190330). A.Z. acknowledges support from the Welch Foundation (AT-1617) and from the Ministry of Education and Science of the Russian Federation (14.Y26.31.0010). References 1. Song, J. Z.; Xu, L. M.; Li, J. H.; Xue, J.; Dong, Y. H.; Li, X. M.; Zeng, H. B., Monolayer and Few- Layer All-Inorganic Perovskites as a New Family of Two-Dimensional Semiconductors for Printable Optoelectronic Devices. Adv. Mater. 2016, 28 (24), 4861-4869. 2. Dong, Y. H.; Gu, Y.; Zou, Y. S.; Song, J. Z.; Xu, L. M.; Li, J. H.; Xue, F.; Li, X. M.; Zeng, H. B., Improving All-Inorganic Perovskite Photodetectors by Preferred Orientation and Plasmonic Effect. Small 2016, 12 (40), 5622-5632. 3. Ling, Y. C.; Tian, Y.; Wang, X.; Wang, J. C.; Knox, J. M.; Perez-Orive, F.; Du, Y. J.; Tan, L.; Hanson, K.; Ma, B. W.; Gao, H. W., Enhanced Optical and Electrical Properties of Polymer-Assisted All- Inorganic Perovskites for Light-Emitting Diodes. Adv. Mater. 2016, 28 (40), 8983-8989. 4. Kim, Y. H.; Lee, G. H.; Kim, Y. T.; Wolf, C.; Yun, H. J.; Kwon, W.; Park, C. G.; Lee, T. W., High Efficiency Perovskite Light-Emitting Diodes of Ligand-Engineered Colloidal Formamidinium Lead Bromide Nanoparticles. Nano Energy 2017, 38, 51-58. 5. Zhang, L. Q.; Yang, X. L.; Jiang, Q.; Wang, P. Y.; Yin, Z. G.; Zhang, X. W.; Tan, H. R.; Yang, Y.; Wei, M. Y.; Sutherland, B. R.; Sargent, E. H.; You, J. B., Ultra-bright and Highly Efficient Inorganic Based Perovskite Light-Emitting Diodes. Nat. Commun. 2017, 8, 15640. 6. Xiao, Z. G.; Kerner, R. A.; Zhao, L. F.; Tran, N. L.; Lee, K. M.; Koh, T. W.; Scholes, G. D.; Rand, B. P., Efficient Perovskite Light-Emitting Diodes Featuring Nanometre-Sized Crystallites. Nat. Photonics 2017, 11 (2), 108-115. 7. Kim, Y. H.; Wof, C.; Kim, Y. T.; Cho, H.; Kwon, W.; Do, S.; Sadhanala, A.; Park, C. G.; Rhee, S. W.; Im, S. H.; Friend, R. H.; Leet, T. W., Highly Efficient Light-Emitting Diodes of Colloidal Metal-Halide Perovskite Nanocrystals beyond Quantum Size. ACS Nano 2017, 11 (7), 6586-6593. 8. Ayguler, M. F.; Weber, M. D.; Puscher, B. M. D.; Medina, D. D.; Docampo, P.; Costa, R. D., Light- Emitting Electrochemical Cells Based on Hybrid Lead Halide Perovskite Nanoparticles. J. Phys. Chem. C 2015, 119 (21), 12047-12054. 9. Zhang, H. M.; Lin, H.; Liang, C. J.; Liu, H.; Liang, J. J.; Zhao, Y.; Zhang, W. G.; Sun, M. J.; Xiao, W. K.; Li, H.; Polizzi, S.; Li, D.; Zhang, F. J.; He, Z. Q.; Choy, W. C. H., Organic-Inorganic Perovskite Light-Emitting Electrochemical Cells with a Large Capacitance. Adv. Funct. Mater. 2015, 25 (46), 7226- 7232. 10. Puscher, B. M. D.; Ayguler, M. F.; Docampo, P.; Costa, R. D., Unveiling the Dynamic Processes in Hybrid Lead Bromide Perovskite Nanoparticle Thin Film Devices. Adv. Energy Mater. 2017, 7 (15), 1602283. 11. Ayguler, M. F.; Puscher, B. M. D.; Tong, Y.; Bein, T.; Urban, A. S.; Costa, R. D.; Docampo, P., Light-Emitting Electrochemical Cells Based on Inorganic Metal Halide Perovskite Nanocrystals. J. Phys. D-Appl. Phys. 2018, 51 (33), 334001. 12. Pei, Q. B.; Yu, G.; Zhang, C.; Yang, Y.; Heeger, A. J., Polymer Light-Emitting Electrochemical Cells. Science 1995, 269 (5227), 1086-1088. 13. de Mello, J. C., Interfacial Feedback Dynamics in Polymer Light-Emitting Electrochemical Cells. Phys. Rev. B 2002, 66 (23), 235210. 14. Costa, R. D.; Orti, E.; Bolink, H. J.; Monti, F.; Accorsi, G.; Armaroli, N., Luminescent Ionic Transition-Metal Complexes for Light-Emitting Electrochemical Cells. Angew. Chem.-Int. Edit. 2012, 51 (33), 8178-8211. 15. Slinker, J. D.; DeFranco, J. A.; Jaquith, M. J.; Silveira, W. R.; Zhong, Y.-W.; Moran-Mirabal, J. M.; Craighead, H. G.; Abruna, H. D.; Marohn, J. A.; Malliaras, G. G., Direct Measurement of the Electric- Field Distribution in a Light-Emitting Electrochemical Cell. Nat. Mater. 2007, 6 (11), 894-899. 16. Shen, Y.; Kuddes, D. D.; Naquin, C. A.; Hesterberg, T. W.; Kusmierz, C.; Holliday, B. J.; Slinker, J. D., Improving Light-Emitting Electrochemical Cells with Ionic Additives. Appl. Phys. Lett. 2013, 102 (20), 203305. 17. Bastatas, L. D.; Lin, K. Y.; Moore, M. D.; Suhr, K. J.; Bowler, M. H.; Shen, Y. L.; Holliday, B. J.; Slinker, J. D., Discerning the Impact of a Lithium Salt Additive in Thin-Film Light-Emitting Electrochemical Cells with Electrochemical Impedance Spectroscopy. Langmuir 2016, 32 (37), 9468-9474. 18. Suhr, K. J.; Bastatas, L. D.; Shen, Y. L.; Mitchell, L. A.; Holliday, B. J.; Slinker, J. D., Enhanced Luminance of Electrochemical Cells with a Rationally Designed Ionic Iridium Complex and an Ionic Additive. ACS Appl. Mater. Interfaces 2016, 8 (14), 8888-8892. 19. Li, J. Q.; Shan, X.; Bade, S. G. R.; Geske, T.; Jiang, Q. L.; Yang, X.; Yu, Z. B., Single-Layer Halide Perovskite Light-Emitting Diodes with Sub-Band Gap Turn-On Voltage and High Brightness. J. Phys. Chem. Lett. 2016, 7 (20), 4059-4066. 20. Fang, Z. S.; He, H. P.; Gan, L.; Li, J.; Ye, Z. Z., Understanding the Role of Lithium Doping in Reducing Nonradiative Loss in Lead Halide Perovskites. Adv. Sci. 2018, 5 (12), 1800736. 21. de Mello, J. C.; Wittmann, H. F.; Friend, R. H., An Improved Experimental Determination of External Photoluminescence Quantum Efficiency. Adv. Mater. 1997, 9 (3), 230-232. 22. Porres, L.; Holland, A.; Palsson, L. O.; Monkman, A. P.; Kemp, C.; Beeby, A., Absolute Measurements of Photoluminescence Quantum Yields of Solutions Using an Integrating Sphere. J. Fluoresc. 2006, 16 (2), 267-272. 23. Jiang, Q. L.; Chen, M. M.; Li, J. Q.; Wang, M. C.; Zeng, X. Q.; Besara, T.; Lu, J.; Xin, Y.; Shan, X.; Pan, B. C.; Wang, C. C.; Lin, S. C.; Siegrist, T.; Xiao, Q. F.; Yu, Z. B., Electrochemical Doping of Halide Perovskites with Ion Intercalation. ACS Nano 2017, 11 (1), 1073-1079. 24. Zha, F. X.; Shao, J.; Jiang, J.; Yang, W. Y., "Blueshift" in Photoluminescence and Photovoltaic Spectroscopy of the Ion-Milling Formed n-on-p HgCdTe Photodiodes. Appl. Phys. Lett. 2007, 90 (20), 201112. 25. Zhang, J.; Shang, M. H.; Wang, P.; Huang, X. K.; Xu, J.; Hu, Z. Y.; Zhu, Y. J.; Han, L. Y., n-Type Doping and Energy States Tuning in CH3NH3Pb1-xSb2x/3I3 Perovskite Solar Cells. ACS Energy Lett. 2016, 1 (3), 535-541. 26. Begum, R.; Parida, M. R.; Abdelhady, A. L.; Murali, B.; Alyami, N. M.; Ahmed, G. H.; Hedhili, M. N.; Bakr, O. M.; Mohammed, O. F., Engineering Interfacial Charge Transfer in CsPbBr3 Perovskite Nanocrystals by Heterovalent Doping. J. Am. Chem. Soc. 2017, 139 (2), 731-737. 27. Niu, S. A.; Cao, Z.; Li, S.; Yan, T. Y., Structure and Transport Properties of the LiPF6 Doped 1- Ethyl-2,3-dimethyl-imidazolium Hexaftuorophosphate Ionic Liquids: A Molecular Dynamics Study. J. Phys. Chem. B 2010, 114 (2), 877-881. 28. Yu, H. T.; Lu, Y.; Feng, Z. Q.; Wu, Y. A.; Liu, Z. W.; Xia, P. F.; Qian, J.; Chen, Y. F.; Liu, L. H.; Cao, K.; Chen, S. F.; Huang, W., A MAPbBr(3):poly(ethylene oxide) composite perovskite quantum dot emission layer: enhanced film stability, coverage and device performance. Nanoscale 2019, 11 (18), 9103- 9114. 29. Liu, M.; Zhong, G. H.; Yin, Y. M.; Miao, J. S.; Li, K.; Wang, C. Q.; Xu, X. R.; Shen, C.; Meng, H., Aluminum-Doped Cesium Lead Bromide Perovskite Nanocrystals with Stable Blue Photoluminescence Used for Display Backlight. Adv. Sci. 2017, 4 (11), 1700335. 30. Chan, C. M.; Weng, L. T., Surface Characterization of Polymer Blends by XPS and ToF-SIMS. Materials 2016, 9 (8), 655. 31. Yantara, N.; Bhaumik, S.; Yan, F.; Sabba, D.; Dewi, H. A.; Mathews, N.; Boix, P. P.; Demir, H. V.; Mhaisalkar, S., Inorganic Halide Perovskites for Efficient Light-Emitting Diodes. J. Phys. Chem. Lett. 2015, 6 (21), 4360-4364. 32. Zhang, M. Z.; Zheng, Z. P.; Fu, Q. Y.; Chen, Z.; He, J. L.; Zhang, S.; Yan, L.; Hu, Y. X.; Luo, W., Growth and Characterization of All-Inorganic Lead Halide Perovskite Semiconductor CsPbBr3 Single Crystals. CrystEngComm 2017, 19 (45), 6797-6803. 33. Zhang, H. H.; Liao, Q.; Wu, Y. S.; Chen, J. W.; Gao, Q. G.; Fu, H. B., Pure Zero-Dimensional Cs4PbBr6 Single Crystal Rhombohedral Microdisks with High Luminescence and Stability. Phys. Chem. Chem. Phys. 2017, 19 (43), 29092-29098. 34. Lou, S. Q.; Zhou, Z.; Xuan, T. T.; Li, H. L.; Jiao, J.; Zhang, H. W.; Gautier, R.; Wang, J., Chemical Transformation of Lead Halide Perovskite into Insoluble, Less Cytotoxic, and Brightly Luminescent CsPbBr3/CsPb2Br5 Composite Nanocrystals for Cell Imaging. ACS Appl. Mater. Interfaces 2019, 11 (27), 24241-24246. 35. Chen, Y. M.; Zhou, Y.; Zhao, Q.; Zhang, J. Y.; Ma, J. P.; Xuan, T. T.; Guo, S. Q.; Yong, Z. J.; Wang, J.; Kuroiwa, Y.; Moriyoshi, C.; Sun, H. T., Cs4PbBr6/CsPbBr3 Perovskite Composites with Near- Unity Luminescence Quantum Yield: Large-Scale Synthesis, Luminescence and Formation Mechanism, and White Light-Emitting Diode Application. ACS Appl. Mater. Interfaces 2018, 10 (18), 15905-15912. Supporting Information Bright and Efficient Perovskite Light Emitting Electrochemical Cells Leveraging Ionic Additives Masoud Alahbakhshi1, Aditya Mishra2, Ross Haroldson3, Arthur Ishteev6, Jiyoung Moon2, Qing Gu1, Jason D. Slinker2,3* and Anvar A. Zakhidov3,4,5* 1Department of Electrical and Computer Engineering, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 2Department of Materials Science and Engineering, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 3Department of Physics, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 4NanoTech Institute, The University of Texas at Dallas, 800 West Campbell Road, Richardson, Texas 75080-3021, United States. 5Department of Nanophotonics and Metamaterials, ITMO University, St. Petersburg, Moscow, Russia. 6Laboratory of Advanced Solar Energy, NUST MISiS, Moscow, 119049, Russia Experimental Methods Materials .......................................................................................................................... S-3 Perovskite Solution Preparation ....................................................................................... S-3 Device Fabrication ........................................................................................................... S-3 Electroluminescence Measurements ................................................................................ S-4 Photoluminescence vs. Time Measurements ................................................................... S-4 Photoluminescence Quantum Efficiency Measurements ................................................. S-4 Scanning Electron Microscopy (SEM) ............................................................................ S-4 Atomic Force Microscopy (AFM) ................................................................................... S-4 X-Ray Photoelectron Spectroscopy (XPS) ...................................................................... S-5 X-Ray Diffraction (XRD) ................................................................................................ S-5 Figures and Tables Figure S1. Experimental Setup for Electroluminescence ................................................ S-6 Figure S2. Electroluminescence Spectra .......................................................................... S-7 Table S1. PeLEC Performance Summary ........................................................................ S-8 Figure S3. PeLEC Current Density vs Voltage ................................................................ S-9 Figure S4. PeLEC Luminance vs Time.......................................................................... S-10 Figure S5. Fabrication of Thin Perovskite Films ........................................................... S-11 Figure S6. Doping Effects on Photoluminescence ........................................................ S-12 Table S2. SEM and XRD Perovskite Film Data ............................................................ S-13 Figure S7. Average Perovskite Grain Size from SEM .................................................. S-14 Figure S8. SEM and AFM of 5% LiPF6 Perovskite Films ............................................ S-15 Figure S9. 3D AFM Topography Images of Perovskite Films ...................................... S-16 Figure S10. XPS Spectra for Perovskite Films .............................................................. S-17 Table S3. Comparison of PeLED and PeLEC Performance .......................................... S-19 References .................................................................................................................................. S-20 Experimental Methods Materials: Lead (II) bromide (PbBr2; 99.99% trace metal basis), Cesium bromide (CsBr; 99.99%) and Polyethylene Oxide (PEO; M.W. > 5,000,000) were all purchased from Alfa Aesar. Lithium Hexafluorophosphate (LiPF6; 99.99%) and Dimethyl Sulfoxide (DMSO; anhydrous > 99.9 % ) were purchased from Sigma Aldrich. Gallium Indium eutectic (GaIn; 99.99%) was purchased from Beantown Chemical. Perovskite Solution Preparation: The CsPbBr3-based precursor solution was prepared by dissolving PbBr2 and CsBr in a 1:1.5 molar ratio with PEO (10 mg/ml) in anhydrous DMSO solution. Then the perovskite precursor solution was stirred at 60 °C for dissolution overnight. When all solutions were dissolved, an empty vial was weighed and the desirable amount of PEO was added, then the weight difference before and after the PEO addition was measured to get an accurate weight of the viscous solution. The weight ratio of CsPbBr3 to PEO was 100:80. Finally, these solutions were blended with 4 mg/ml DMSO solutions of LiPF6 to prepare mixtures of five different weight ratios (0.2%, 0.5%, 1%, 2% and 5%) of lithium salt with the perovskite-polymer composition. Device Fabrication: The ITO/glass substrates (Liasion Quartz (Lianyungang Jiangsu China), sheet resistance ~ 15 Ω sq-1) were cleaned sequentially with detergent solution, deionized water, acetone, Toluene and 2-propanol in an ultra-sonication bath for 15 mins. Subsequently, the substrates were dried with nitrogen and treated for 20 min with UV-ozone. To obtain CsPbBr3:PEO:LiPF6 thin films, precursor solutions were spin-coated onto ITO substrates at 1200 rpm for 45 min. Then, the thin film was put under vacuum for 1 minute to have a uniform and pinhole-free thin film. Finally, the CsPbBr3:PEO:LiPF6 film was annealed at 150 °C for 15 seconds to remove the residual solvent. Electroluminescence Measurements: All measurements were conducted using a mechanical probe-station under high vacuum <10mTorr. Current density-voltage (J-V) and luminance-voltage (L-V) characteristics were measured using a Keithley 2400 source meter and a Photo Research PR-650 spectroradiometer in the range of 0V to 6V with a 0.3V increment. Photoluminescence vs Time Measurements: Measurements were taken using an Ocean Optics QE65000 spectrometer coupled with a fiber optic cable pointed at the thin films through a 450nm longpass dielectric filter to block out the 405nm CW laser diode excitation signal. Photoluminescence Quantum Efficiency Measurements: Thin film samples deposited on glass were attached to a custom holder then placed inside a Spectral Physics integrating sphere. Samples were excited at a 15 degree angle incidence to avoid back reflecting excitation light out of the integrating sphere. Using the method developed by de Mello et al., we took measurements of the excited thin films both in and out of the excitation beam path [1]. Fluctuations in excitation power were monitored by splitting the beam with a beam splitter and placing a photodiode power meter hooked up to a Thorlabs PDA200C photodiode amplifier. The collimated beam profile was shaped with a precision cut 1000 micron diameter circular aperture. The power density was then measured to be 561±4 mW/cm2. A fiber optic cable was mounted to one of the integrating sphere ports and was coupled to the Ocean Optics QE65000 spectrometer where spectrum measurements were taken. Scanning Electron Microscopy (SEM): Secondary electron SEM images were taken with a Zeiss Supra-40 SEM using an in-lens detector at an accelerating voltage of 10kV. Atomic Force Microscopy (AFM): The AFM images were performed using a Veeco Model 3100 Dimension V to scrutinize the morphology of thin films. The thin films were scanned for 5µm × 5µm area at 0.8Hz rate using an OTESPA-R3 AFM tip from Bruker. Tapping mode AFM was used for this characterization. X-ray Photoelectron Spectroscopy (XPS): XPS measurements were conducted on perovskite thin films using a Versa Probe II at an ultrahigh vacuum of 10-8 Pa. The X-ray source was an Al Kα which has 1486.6eV photon energy, 50W gun power, 15kV operating voltage, 200um X-ray spot size, and 59° angle between the X-ray source and detector. The calibration using internal standard Au, Cu, and Ag samples was performed before obtaining the XPS data. Data was collected without injecting a flux of low energy electrons. X-Ray Diffraction (XRD): XRD measurements were collected using a Rigaku SmartLab X-ray Cu target (Ka1=1.54059 Å) and a HyPix 3000 detector. The 2-theta/omega scan was consistently performed in the 2-theta range of 10° to 55° with a 0.01° step and a ~1°/min scan speed. Supporting Information Figure S1 Figure S1. The experimental setup for capturing the electroluminescence performance of In- Ga/CsPbBr3:PEO:LiPF6/ITO LEC devices. Supporting Information Figure S2 Figure S2. Electroluminescence spectra of In-Ga/CsPbBr3:PEO:LiPF6/ITO light emitting electrochemical cells with various weight ratios of LiPF6. The inset shows the electroluminescence of a 0.5% LiPF6 PeLEC operating at 5 volts. Supporting Information Table S1 Devices Turn-on Voltage CsPbBr3:PEO (0%) CsPbBr3:PEO:LiPF6 (0.2%) CsPbBr3:PEO:LiPF6 (0.5%) CsPbBr3:PEO:LiPF6 (1%) CsPbBr3:PEO:LiPF6 (2%) CsPbBr3:PEO:LiPF6 (5%) (V) 2.5 2.2 1.9 2.4 2.4 2.7 Max CE (cd/A) Max Lum (cd/m2) Max Current Density (mA/cm2) 9.0 17.2 22.4 4.5 4.1 1.2 8175 11800 14730 5675 2619 517 2390 2952 3301 2150 1486 749 Table S1. Summrized performance of In-Ga/CsPbBr3:PEO:LiPF6/ITO LEC devices with different ratios of LiPF6. Supporting Information Figure S3 Figure S3. Current density versus voltage of In-Ga/CsPbBr3:PEO:LiPF6/ITO LEC devices with different ratios of LiPF6. The graph clearly shows that in the optimized ratio of Li (0.5%), the hysteresis state decreases significantly. Supporting Information Figure S4 Figure S4. Luminance In- Ga/CsPbBr3:PEO:LiPF6/ITO LEC devices with different ratios of LiPF6, demonstrating increased stability with optimal LiPF6 concentration. operation versus constant voltage time for of Supporting Information Figure S5 Figure S5. Illustration of the fabrication of thin films of CsPbBr3:PEO:LiPF6. Initially, components are spin cast at 1200 rpm for 45 seconds, followed by rapid vacuum treatment for 1 minute and annealing for 15 seconds at 115 °C. Supporting Information Figure S6 c) Figure S6. The mechanism of the recombination of electrons and holes for undoped thin films (a) for doped thin films (b) by Li. The doping of Li produces free electrons to fill and passivate the trap states, therefore the radiative recombination will improve. c) Illustration of the Burstein -- Moss effect causing a PL blue shift. Supporting Information Table S2 Sample CsPbBr3 CsPbBr3:PEO (0%) CsPbBr3:PEO:LiPF6 (0.2%) CsPbBr3:PEO:LiPF6 (0.5%) CsPbBr3:PEO:LiPF6 (1%) CsPbBr3:PEO:LiPF6 (2%) CsPbBr3:PEO:LiPF6 (5%) Average grain size (nm) FWHM (degree) 2θ (degree) 145 79 105 163 166 150 161 0.51 0.50 0.52 0.53 0.51 0.54 0.53 0.57 0.58 0.58 0.61 0.63 0.65 0.65 21.53 30.52 21.5 30.55 21.49 30.59 21.49 30.57 21.52 30.58 21.55 30.65 21.6 30.7 Crystallite size (nm) 16.49 17.04 16.39 16.15 16.47 16.07 16.04 15.15 14.56 14.79 13.91 13.71 13.04 13.20 Table S2. Average grain size as measured from SEM and XRD spectral data for CsPbBr3, CsPbBr3:PEO, and CsPbBr3:PEO:LiPF6 thin films. Supporting Information Figure S7 Figure S7. Histograms of CsPbBr3, CsPbBr3:PEO, and CsPbBr3:PEO:LiPF6 thin film grain size as measured from SEM imaging. Supporting Information Figure S8 Figure S8. Morphological impact of high LiPF6 concentrations in CsPbBr3:PEO:LiPF6 thin films. a) SEM and b) AFM images of CsPbBr3:PEO:LiPF6 (5%). Supporting Information Figure S9 Figure S9. 3D AFM topography images of 1) CsPbBr3, 2) CsPbBr3:PEO (0%), 3) CsPbBr3:PEO:LiPF6 (0.2%), 4) CsPbBr3:PEO:LiPF6 (0.5%), 5) CsPbBr3:PEO:LiPF6 (1%), 6) CsPbBr3:PEO:LiPF6 (2%). Supporting Information Figure S10 Figure 10. X-Ray photoelectron spectroscopy (XPS) spectra for CsPbBr3, CsPbBr3:PEO, CsPbBr3:PEO:LiPF6. CsPbBr3 CsPbBr3:PEO CsPbBr3:PEO:LiPF6 (0.5%) CsPbBr3:PEO:LiPF6 (2%) Supporting Information Table S3 Device structure ITO / PEDOT:PSS /PKNP:TMPE:LiCF3SO3/Al ITO / Pero-PEO-composite / In (Ga,Au) ITO / Pero-PEO-PVP-composite / InGa ITO / PEDOT:PSS /Perovskite / MoO3/Au ITO / Pero-PEO composite / Ag NW ITO / PEDOT:PSS /MAPbBr3 / SPB- 02T /LiF/ Ag ITO/PEDOT:PSS/CsPbBr3:PEO/TPBI/ LiF/Al ITO/PEDOT:PSS/CsPbBr3:PEG/TPBI/ LiF/Al ITO / PEDOT:PSS / CsPbBr1.25I1.75 NC: KCF3SO3/Al VACNT / SC-MAPbBr3 / VACNT Our Work Single or multilayer Light Emitting Turn on Mechanism voltage Luminescence Max (cd/m2) M S S M S M M M M S S LEC LED LED LEC LED LED LED LED LEC LEC LEC 11 3 1.9 2.4 2.6 2 2.6 2.5 4 26 1.9 1.8 4064 593178 157 (0.23 W/ Sr. m2) 21014 3490 51890 36600 8 1000 14730 Current Efficiency Ref (cd/A) 0.013 0.74 21.5 - 4.9 0.43 21.38 19 - - [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] 22.4 Table S3. Comparative performances of the best-in-class perovskite based LEDs and LECs. References [1] J. C. De Mello, H. F. Wittmann and R. H. Friend, "An improved experimental determination of external photoluminescence quantum efficiency," Advanced Materials, vol. 9, no. 3, pp. 230-232, 1997. [2] M. F. Ayguler, M. D. Weber, B. M. D. Puscher, D. D. Medina, P. Docampo and R. D. Costa,,, "Light-Emitting Electrochemical Cells Based on Hybrid Lead Halide Perovskite Nanoparticles," J. Phys. Chem. C, vol. 119, p. 12047 -- 12054, 2015. [3] Junqiang Li, Sri Ganesh R. Bade, Xin Shan and Zhibin Yu, "Single‐Layer Light‐Emitting Diodes Using Organometal Halide Perovskite/Poly(ethylene oxide) Composite Thin Films," Adv.Mater., vol. 27, p. 5196 -- 5202, 2015. [4] Junqiang Li, Xin Shan, Sri Ganesh R. Bade, Thomas Geske, Qinglong Jiang, Xin Yang and Zhibin Yu, "Single-Layer Halide Perovskite Light-Emitting Diodes with Sub-Band Gap Turn-On Voltage and High Brightness," The Journal of Physical Chemistry Letters , vol. 7, no. 20, pp. 4059-4066, 2016. [5] Huimin Zhang , Hong Lin , Chunjun Liang , Hong Liu , Jingjing Liang , Yong Zhao , Wenguan Zhang , Mengjie Sun , Weikang Xiao , Han Li , Stefano Polizzi , Dan Li ,Fujun Zhang , Zhiqun He , and Wallace C. H. Choy , "Organic -- Inorganic Perovskite Light‐Emitting Electrochemical Cells with a Large Capacitance," Adv. Funct. Mater. , vol. 25, p. 7226 -- 7232, 2015. [6] Sri Ganesh R. Bade, Junqiang Li, Xin Shan, Yichuan Ling, Yu Tian, Tristan Dilbeck, Tiglet Besara, Thomas Geske, Hanwei Gao, Biwu Ma, Kenneth Hanson, Theo Siegrist, Chengying Xu and Zhibin Yu, "Fully Printed Halide Perovskite Light-Emitting Diodes with Silver Nanowire Electrodes," ACS Nano, vol. 10, no. 2, pp. 1795-1801, 2016. [7] Jae Choul Yu, Da Bin Kim, Eui Dae Jung, Bo Ram Lee and Myoung Hoon Song, "High- performance perovskite light-emitting diodes via morphological control of perovskite films," Nanoscale, vol. 8, pp. 7036-7042 , 2016. [8] Chen Wu, Yatao Zou, Tian Wu, Muyang Ban, Vincenzo Pecunia, Yujie Han, Qipeng Liu, Tao Song, Steffen Duhm and Baoquan Sun, "Improved Performance and Stability of All‐Inorganic Perovskite Light‐Emitting Diodes by Antisolvent Vapor Treatment," Adv. Funct. Mater., vol. 27, p. 1700338, 2017. [9] Li Song, Xiaoyang Guo, Yongsheng Hu, Ying Lv, Jie Lin, Zheqin Liu, Yi Fan and Xingyuan Liu, "Efficient Inorganic Perovskite Light-Emitting Diodes with Polyethylene Glycol Passivated Ultrathin CsPbBr3 Films," The Journal of Physical Chemistry Letters, vol. 8 , no. 17, pp. 4148- 4154, 2017. [10] Meltem F Aygüler, Bianka M D Puscher, Yu Tong, Thomas Bein, Alexander S Urban, Rubén D Costa and Pablo Docampo, "Light-emitting electrochemical cells based on inorganic metal halide perovskite nanocrystals," J. Phys. D: Appl. Phys, vol. 51, 2018. [11] Pavao Andričević, Xavier Mettan, Márton Kollár, Bálint Náfrádi, Andrzej Sienkiewicz, Tonko Garma, Lidia Rossi, László Forró, and Endre Horváth, "Light-Emitting Electrochemical Cells of Single Crystal Hybrid Halide Perovskite with Vertically Aligned Carbon Nanotubes Contacts," ACS Photonics, vol. 6 , no. 4, pp. 967-975, 2019.
1804.09622	2	1804	2018-04-27T13:24:06	Negative Capacitance as Digital and Analog Performance Booster for Complementary MOS Transistors	[ "physics.app-ph" ]	Boltzmann tyranny poses a fundamental limit to lowering the energy dissipation of conventional MOS devices, a minimum increase of the gate voltage, i.e. 60 mV, is required for a 10-fold increase in drain-to-source current at 300 K. Negative Capacitance (NC) in ferroelectric materials is proposed in order to address this physical limitation of CMOS technology. A polarization destabilization in ferroelectrics causes an effective negative permittivity, resulting in a differential voltage amplification and a reduced subthreshold swing when integrated into the gate stack of a transistor. Recent demonstrations of negative capacitance concerned mainly n-type MOSFETs and their subthreshold slope. An effective technology booster should be capable of improving the performance of both n- and p-type transistors. In this work, we report a significant enhancement in both digital (subthreshold swing, on-current over off-current ratio, and overdrive) and analog (transconductance and current efficiency factor) FoM of commercial 28nm CMOS process by exploiting a PZT capacitor as the negative capacitance booster. Accordingly, a sub-thermal swing down to 10 mV/decade together with an enhanced current efficiency factor up to 10$^5$ V$^{-1}$ is obtained in both n- and p-type MOSFETs at room temperature. The overdrive voltage is enhanced up to 0.45 V, leading to a supply voltage reduction of 50\%.	physics.app-ph	physics	Negative Capacitance as Digital and Analog Performance Booster for Complementary MOS Transistors A. Saeidi,∗ F. Jazaeri, I. Stolichnov, Christian C. Enz, and Adrian M. Ionescu Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland E-mail: [email protected] Abstract Boltzmann tyranny poses a fundamental limit to lowering the energy dissipation of conventional MOS devices, a minimum increase of the gate voltage, i.e. 60 mV, is required for a 10-fold increase in drain-to-source current at 300 K. Negative Ca- pacitance (NC) in ferroelectric materials is proposed in order to address this physical limitation of CMOS technology. A polarization destabilization in ferroelectrics causes an effective negative permittivity, resulting in a differential voltage amplification and a reduced subthreshold swing when integrated into the gate stack of a transistor. Recent demonstrations of negative capacitance concerned mainly n-type MOSFETs and their subthreshold slope. An effective technology booster should be capable of improving the performance of both n- and p-type transistors. In this work, we report a significant enhancement in both digital (subthreshold swing, on-current over off-current ratio, and overdrive) and analog (transconductance and current efficiency factor) FoM of commer- cial 28nm CMOS process by exploiting a PZT capacitor as the negative capacitance booster. Accordingly, a sub-thermal swing down to 10 mV/decade together with an enhanced current efficiency factor up to 105 V−1 is obtained in both n- and p-type 1 MOSFETs at room temperature. The overdrive voltage is enhanced up to 0.45 V, leading to a supply voltage reduction of 50%. Complementary Metal-Oxide-Semiconductor (CMOS) scaling will be eventually limited by the inability to remove the heat generated in the switching process.1 The origin of this issue can be traced back to the operation principle of the silicon CMOS devices governs by the non-scalability of thermal voltage (Boltzmann's tyranny). This results in preventing these devices to achieve a sub-60 mV/decade subthreshold slope (SS) at room temperature. The SS of a MOSFET is obtained by SS = ∂Vg ∂(logId) = ∂Vg ∂ψs × ∂ψs ∂(logId) , (1) where ψs corresponds to the surface potential of the silicon channel. MOSFET, the lower limit of the second term in RHS of (1) is (kBT /q)Ln(10) and cannot be any lower than 60 mV/decade at 300 K. Since Vg is linked to ψs through a capacitive voltage divider, the first term that is known as the body factor, m, is obtained as In a conventional ∂Vg ∂ψs = 1 + Cs CM OS , (2) exceeds one, thus limits the SS to 60 mV/decade at T=300 K.2,3 A sub-thermal swing can be achieved using the proposed negative capacitance (NC) of ferroelectric materials.4,5 Negative capacitance in ferroelectrics arises from the imperfect screening of the spontaneous polariza- tion. Imperfect screening is intrinsic to semiconductor-ferroelectric and metal-ferroelectric interfaces due to their screening lengths. The physical separation of ferroelectric bound charges from the metallic screening charges creates a depolarizing field inside the ferroelec- tric and destabilizes the polarization.6 Hence, intentionally destabilizing this polarization 2 causes an effective NC that has been proposed as a way of overcoming the fundamental limitation on the power consumption of MOSFETs.7–9 The negative capacitance, originat- ing from the dynamics of the stored energy in a phase transition of ferroelectric materials, results in an internal voltage amplification in an MOS device when integrated into the gate stack. The concept of NC can be understood by considering the free energy of ferroelectrics. A ferroelectric material is traditionally modeled using a double well energy landscape. The energy characteristic of a ferroelectric capacitor, depicted in Figure 1a, is calculated by UF E = αP 2 + βP 4 + γP 6 + Eext.P , where P is the polarization, Eext is the externally ap- plied electric field, and α, β, and γ are material dependent parameters.5 In equilibrium, the ferroelectric resides in one of the wells, providing spontaneous polarization. The capacitance of a ferroelectric material can be determined by (cid:21)−1 , (3) CF E = (cid:20) d2UF E dQ2 F E which is positive at the wells considering the curvature of UF E vs. QF E characteristic (Figure 1a). Nevertheless, the curvature is negative around the origin (QF E = 0). More specifically, a ferroelectric material shows an effective NC while switching from one stable polarization state to the other one.10 It should be remarked that NC refers to negative differential capac- itance due to the small signal concept of the capacitance and relation between CF E and UF E (equation 3). The NC has been proven elusive for ferroelectrics in isolation and cannot be observed in experiments, exhibiting hysteretic jumps in the polarization (Figure 1b). How- ever, it has been confirmed that if the ferroelectric placed in-series with a positive capacitor of proper value, the NC segment can be stabilized.11,12 This NC region can be modeled by the state-of-the-art approach for modeling the dynamics of ferroelectric capacitors relying on Landau-Khalatnikov (LK) equation, ρ(dP/dt) + ∇pUF E = 0. Figure 1b compares the experimentally measured polarization vs. electric field of a PZT capacitor with the fitting result of the LK equation. 3 Figure 1: Free energy landscape and polarization characteristic of a ferroelectric. (a) Energy density function of a ferroelectric capacitor in equilibrium, showing an effective NC while switching from one stable polarization state to the other one. (b) Measured polarization vs. electric field for a PZT film (experimental) and the fitting results of LK equation (dashed curve). A ferroelectric capacitor interconnecting with the gate stack of a MOS transistor cre- ates a series connection between CF E and CM OS (Figure 2a). The ferroelectric capacitor can increase the total capacitance of the gate (C−1 M OS) while it is sta- bilized in the NC region.13,14 Specifically, the series structure brings an abrupt increase total = C−1 F E + C−1 in the differential charge in the internal node (Vint) by changing the gate voltage, thus providing a step-up voltage transformer.15,16 The internal gain of NC can be defined as β = ∂Vint/∂Vg = CF E/(CF E + CM OS). Accordingly, an NC booster can provide an internal voltage amplification (β > 1) which results in a body factor reduction, i.e. 1/β, leading to the improvement of both analog and digital parameters of the transistor: (cid:18) ∂logId (cid:19)−1 SS = ∂Vg = ∂Vint ∂logId × ∂Vg ∂Vint = 1 β × SSref , (4) (5) gm = ∂Id ∂Vg = ∂Id ∂Vint × ∂Vint ∂Vg = β × gm−ref . In order for NC to occur, the charge line of the baseline transistor is acquired to have an intersection with the negative slope of the polarization.10 Otherwise, the device charac- teristic shows a hysteresis, corresponding to the coercive fields of the ferroelectric without performance boosting.17 Additionally, to bring about the maximum enhancement in the 4 (a)(b)Electric Field, EFEPolarization, PFECFE < 0C α dP/dECFE > 0CFE > 0CFE < 0Polarization, PFEEnergy, UFEExperimentalLK equationFigure 2: Negative capacitance MOSFET and stability constraints. (a) Schematic diagram of the experimental configuration of an n-type NC-FET including the capacitance model of the structure. (b) Ferroelectric's NC is unstable by itself (A), but it can be partially (B) or fully stabilized (C) by placing in-series with a positive capacitor. The NC region of ferroelectric energy landscape is shown inside the dotted rectangular box. non-hysteretic operation of an NC-FET, the negative value of CF E should be well-matched with CM OS (CF E = CM OS while Ctotal > 0 in the whole range of operation).16,18 Generally, considering that the SS can be expressed as SS = (60mV /decade).(1+CM OS/CF E), the tran- sistor transfer characteristic becomes steep as CF E gets close to CM OS. However, a value of CF E too close to CM OS gives rise to the hysteretic behavior due to the instability of NC in the strong inversion regime.19 Additionally, both CM OS and CF E are voltage-dependent, making it extremely challenging to fully satisfy the matching condition. Therefore, the ferro- electric's NC commonly partially gets stabilized, proposing a trade-off between the hysteretic behavior and the performance-boosting due to the NC effect (Figure 2b). With the validity of NC concept being experimentally established,12,20–24 it is now of paramount importance to understand the challenges involved in the design of NC-FETs, so that the steepness and hysteresis of the device characteristic can be optimized in both n- and p-type MOSFETs. In this respect, a polycrystalline PZT capacitor is fabricated for thoroughly understanding the negative capacitance concept. It is then connected to various commercial MOSFETs, fabricated in 28 nm CMOS technology node. The hysteretic behavior of both n- and p-type NC-FETs is tuned imposing the proposed matching condition. Afterward, the impact of NC on the performance of conventional MOSFETs is investigated by measuring and analyzing the internal node voltage. Sub-thermal swing down to 10 mV/decade is observed in n- and 5 Vgp+n+n+VdVsub=Vs=0Metal GateMetal GateHigh-KVintPolarization, PFEEnergy, UFE(a)(b)+-VintVgCFECoxCsFerroelectricstable NCunstable NCABCp-type hysteretic NC-FETs. The paper reports and discusses the trade-off between the steep- ness of the subthreshold slope and the hysteresis, degrading the performance by reducing the hysteresis. The strong dependence of the NC effect on the source to drain electric field is also evidenced, reducing the impact by increasing the absolute value of Vds. Moreover, it is experimentally validated that a poly-domain ferroelectric capacitor in steady states cannot have more than one stable NC domain at the time, showing a different polarization characteristic from the expected S-shape of a single-domain ferroelectric. Experimental results and discussion As schematically shown in Figure 2a, the experimental results are obtained by connecting an external PZT capacitor to the gate of a MOSFET. This external connection offers a flexibility of testing different series combinations and tuning the hysteretic behavior. High-performance commercial n- and p-type MOSFETs are employed as the baseline transistors. An MIM structure with 45nm of polycrystalline Pb(Zr43,Ti57)O3 (PZT) is fabricated.25,26 High-quality epitaxial ferroelectric layers are commonly considered suitable for NC devices due to the formation of a mono-domain state characterized by a single coercive field.6,27,28 However, the typical behavior of poly-domain ferroelectrics can change dramatically by applying a repetitive voltage stress known as the training procedure of ferroelectrics.29 This behavior suggests that a poled ferroelectric layer can show a mono-domain like characteristic (see supplementary materials). n-type negative capacitance MOSFETs Figure 3a illustrates the input transfer characteristic of an n-type NC-FET where the gate of the baseline FET (W = 200 nm, L = 1 µm) is loaded with a PZT capacitor having an area of 30×30 µm2. The gate voltage is swept from −3 V to +3 V and back to −3 V while the drain voltage is set to 0.1 V. It should be noted that the curves are plotted with respect to the 6 Figure 3: Hysteretic n-type NC-FET. (a) Transfer characteristic of the device shows a super steep transition of 10 mV/decade together with a hysteresis of 4.5 V (Vds = 100 mV). (b) A remarkable amplification (defined as dVint/dVg) up to 20 V/V is achieved in the regions corresponding to the negative slope of the polarization (c). Extracted current efficiency factor of the device represents a significant boosting, up to 105 V−1, in the subthreshold region (d). 7 -0.4-0.20.00.20.401020304050gm/Id (V-1)Vgs_eff (V)10-1100101102103104105106gm/Id (V-1)-5-4-3-2-10110-1310-1210-1110-1010-910-810-710-610-5SS=10mV/decL=1mId (A)Vgs_eff (V) NC-FET Baseline FETW=200nm-4-3-2-1012-101234C dP/dEPFE (C/cm2)VFE (V)NC regions-4-20-0.40.00.40.81.2Vint (V)Vgs_eff (V)0510152025dVint/dVg (V/V)-0.4-0.20.00.20.410-1310-1210-1110-1010-910-810-710-610-5SS=10mV/decIoffIon60mV/decId (A)Vgs_eff (V) Baseline FET NC-FET(d)(c)(a)(b)effective gate voltage (Vgs_ef f = Vgs-Vth), which makes the results to be directly comparable and removes the effect of the two different threshold voltages. With the aid of an internal electrode, a step-up conversion of the internal voltage is explicitly observed as a result of the ionic movement in PZT. In order to qualitatively determine the voltage gain, dVint/dVg vs. Vg is plotted, representing a significant amplification up to 20 V/V (Figure 3b). This internal voltage increase allows the surface potential to be higher than the gate voltage, leading to a body factor below 1. Therefore, an SS of 10 mV/decade is observed over six orders of magnitude of the drain current. Moreover, the overdrive voltage is improved by a value of 0.45 V. Using the internal electrode and imposing the displacement vector continuity, a negative slope of the polarization is extracted in a certain range of the polarization, corresponding to the subthreshold region where a significant boosting of performance is reached (Figure 3c). It should be noted that due to the charge balance conditions, only a small fraction of the polarization get switched.30 A remarkable enhancement of the current efficiency factor, gm/Id, with a peak of 105 V−1 is demonstrated when the device is operating in the weak- inversion regime (Figure 3d). A significant improvement is achieved in both digital and analog FoM of the baseline MOSFET. The measured input transfer characteristic of the NC-FET shows a hysteresis of 4.5 V caused by the poor matching of the ferroelectric NC and MOS capacitance.18,19 The undesirable hysteretic operation of NC-FETs can be alleviated with a better match- ing of the ferroelectric and MOS capacitances which ensures the Ctotal > 0 stability condition in a wide range of the applied gate voltage.31 Considering C−1 si , where Cox and Csi correspond to the gate linear dielectric and silicon capacitances, the stability total = C−1 F E + C−1 ox + C−1 condition can be written as(cid:18) Sgate (cid:19) SF E (cid:18) 1 (cid:19)(cid:20) dox dF E SiO2 (cid:21) . + dsi si < 5γ 4(3β2 − 5αγ) (6) In equation (6), d, S, and are the thickness, area, and the permittivity of the corre- 8 Figure 4: n-type NC-FET with a reduced hysteresis. (a) Performance of an n-type NC-FET with a small hysteresis of 150 mV and a swing below 30 mV/decade while Vds is set to 100 mV (b). A steep off -to-on transition is realized in both positive and negative going branches of the drain current. (c) Internal voltage measurement shows a voltage gain of up to 10 V/V. (d) The extracted P-E curve of the ferroelectric shows a clear S-shape in a wide range of operation with a negligible hysteresis. (e) gm/Id is also boosted and reached a factor of 600 V−1. 9 -0.4-0.20.00.20.4-0.50.00.51.01.5Vint (V)Vgs_eff (V)0102030dVint/dvg (V/V)-0.4-0.20.00.20.401020304050gm/Id (V-1)Vgs_eff (V)0100200300400500600BaselineMOSFETgm/Id (V-1)NC-FET3.03.54.04.5-20246C dP/dENC regionsPFe (C/cm2)VFe (V)10-1210-1110-1010-910-810-7060120180 Baseline FET NC-FETSS (mV/dec)Id (A)-0.4-0.20.00.20.410-1310-1210-1110-1010-910-810-7Ion60mV/dec NC-FET Baseline FETId (A)Vgs_eff (V)L=1mW=100nm(b)(a)(c)(d)(e)sponding layer, respectively. Hence, in consideration of (6), another NC-FET with a different baseline FET (W = 100 nm, L = 1 µm) and a PZT capacitor of the same thickness and an area of 20×20 µm2 with a better matching of capacitances and a smaller hysteresis is depicted in Figure 4a. A reduced hysteresis of 150 mV is demonstrated while the transis- tor is operating at a constant drain voltage i.e. 0.1 V. An SS below 30 mV/decade at 300 K is reliably achieved in both positive and negative going branches of the input transfer characteristic (see Figure 4b) where the transistor charge line and the negative slope of the ferroelectric polarization have an intersection.32 As a result of an average swing well below the thermal limit of MOSFETs, the effective gate voltage can be reduced by 50%, maintain- ing the same level of the output current. Figure 4c depicts the internal voltage and internal gain plots (Vint vs Vg and dVint/dVg vs Vg). Figure 4d depicts the extracted polarization characteristic of the series connected PZT capacitor, showing an effective NC similar to the ideal expectation of NC by LK equation. The current efficiency factor is also enhanced and reached a maximum value of about 600 V−1 (Figure 4e). p-type negative capacitance MOSFETs The impact of the same NC booster on p-type commercial MOSFETs is also reported and discussed. The drain-to-source voltage was set at −0.9 V in all measurements performed in this part, otherwise mentioned. Figure 5 depicts the performance improvement that is obtained in a p-type NC-FET (W = 1 µm, L = 90 nm) using a PZT capacitor (40×40 µm2) as an NC booster. The gate voltage swept from +3 V to −3 V and returns back to the initial bias by reverse sweep. Using the NC booster, similar to n-type NC-FETs, the internal voltage is enhanced and reached values greater than the applied gate voltage, so that a steep off -to-on transition of 10 mV/decade is realized over at least 4 orders of magnitude of the drain current (Figure 5a). The NC condition is fulfilled in both forward and reverse sweeps so that a sub-thermal swing is demonstrated in both branches.32 Due to the poor matching of capacitances, a large hysteresis of 3.5 V is observed. Analyzing the internal electrode 10 voltage (Figure 5b) shows a considerable internal voltage amplification in the regions where the ferroelectric capacitor provides a clear S-shape negative slope of the polarization (Figure 5c). The gm/Id FoM is also significantly enhanced, reaching a peak of 105 V−1 (Figure 5d). Figure 5: Hysteretic p-type NC-FET. (a) Transfer characteristic of a p-type NC-FET with a large hysteresis of 3 V (Vds = 900 mV) and a swing of 15 mV/dec over five decades of current. (b) An internal voltage gain greater than one is measured in both positive and negative going branches (c). Current efficiency factor is also enhanced, reaching a factor of 105 V−1. In a different structure, a p-type NC-FET with a small hysteresis is presented (Figure 6a). A PZT capacitor with an area of 10×10 µm2 is connected to the gate of a p-MOSFET (W = 3 µm, L = 1 µm). A reduced hysteresis of 200 mV is achieved due to the proper matching of capacitances, regarding equation (6). Figure 5b reports the SS vs. Vg plot which is well below the thermal limit of MOSFETs (down to 20 mV/decade) at 300 K. The internal node measurement confirms a voltage gain greater than 1 while having a peak of 10 V/V (Figure 5c). The polarization vs. voltage plot of the PZT capacitor indicates a clear S-like 11 -0.4-0.20.00.20.410-1110-1010-910-810-710-610-5SS=10mV/dec60mV/dec NC-FET Baseline FETId (A)Vgs_eff (V)0246810-2-1012310-1110-1010-910-810-710-610-5 NC-FET Baseline FETSS=10mV/decL=90nmId (A)Vgs_eff (V)W=1m-0.4-0.20.00.20.4010203040gm/Id (V-1)Vgs_eff (V)10-1100101102103104gm/Id (V-1)BaselineMOSFETNC-FET02460246C dP/dEPFE (C/cm2)VFE (V)NC regions-20240.00.40.81.21.62.0Vint (V)Vgs_eff (V)02468dVint/dVg (V/V)(a)(b)(c)(d)Figure 6: p-type NC-FET with a reduced hysteresis. (a) Input transfer characteristic of an NC-FET with a small hysteresis of 200 mV at Vds = 900 mV. (b) A sub-thermal swing well below 60 mV/dec is obtained. (c) Measurement of the internal node shows a significant voltage gain, having a peak of 10 V/V. (d) Polarization characteristic of the ferroelectric capacitor shows an effective NC in both branches. Two discrete NC regions are observable in the reverse sweep of the gate voltage due to the polycrystallinity of the ferroelectric film. (e) gm/Id is considerably enhanced and reached a value of 400 V−1. (f) shows the impact of the drain-to-source electric field on the steepness and hysteresis of the NC-FET. 12 -1.0-0.50.00.51.010-1310-1110-910-710-510-3SS=5mV/decVDS=-0.5VVDS=-0.9VId (A)Vgs_eff (V)-0.4-0.20.00.20.4010203040NC-FETgm/Id (V-1)Vgs_eff (V)BaselineMOSFET0100200300400gm/Id (V-1)-1012-0.40.00.40.81.21.6Vint (V)Vgs_eff (V)0246810dVint/dVg (V/V)-0.4-0.20.00.20.410-1210-1110-1010-910-810-710-610-510-410-3IonSS=60mV/dec NC-FET Baseline FETL=1mId (A)Vgs_eff (V)W=3m10-1110-1010-910-810-710-610-5060120180 NC-FET Baseline FETSS (mV/dec)Vgs_eff (V)01230246C dP/dEPFE (C/cm2)VFE (V)NC regions(c)(a)(d)(e)(f)(b)curve in the positive going branch while it shows a different behavior in the reverse sweep. The ferroelectric performs two separate NC regions, demonstrating a zig-zag polarization characteristic. This mainly happens due to the following reasons; (i) the polycrystalline PZT is showing two main polarization domains despite the applied training procedure29 and (ii) a multi-domain ferroelectric can hold one negative capacitance domain at a time.33 As a result, the manifested polarization characteristic of the multi-domain ferroelectric is different from the S-shaped curve which is expected for a single-domain ferroelectric. Therefore, each domain shows a separate NC region independent of the other one (see Figure 6c), which was also expected from dVint/dVg vs. Vg curve where two individual peaks of the voltage amplification were clearly observed. Figure 5d illustrates the current efficiency enhancement with a maximum value of 400 V−1. Figure 5f investigates the impact of the drain-to-source voltage, Vsd, on the input transfer characteristic of the same NC-FET. Besides the common effect of Vsd on the level of the drain current, it is evidenced that the NC-FET under lower lateral electric field provides a more effective NC effect and hence, a steeper off -to-on transition. This is due to the fact that Vds affects the accessible region of the ferroelectric S-curve polarization during the NC-FET operation. An SS of 5 mV/decade is achieved at a Vds of 0.5 V. The off -to-on transition of NC-FETs with reduced hysteresis (both n- and p-type devices) is not as steep as one of the large hysteresis devices, confirming the proposed theory that a trade-off is needed between the steepness and hysteretic behavior.31 A ferroelectric capacitor that implies a too effective NC results in a large hysteresis together with a sharp transition. Although a super steep switching device is compelling, however, it is not appealing since the reduction of SS is accompanied with a remarkable hysteresis. Conclusion Energy efficient logic devices are required for the recent advancement of the Internet of Things (IoT) technology. A steep slope field effect transistor with a sub-thermal swing is 13 expected as a key, enabling technology for IoT applications by operating at supply voltages below 0.5 V. Here, it has been shown that the negative capacitance effect can be effectively applied to enhance both digital and analog FoM of advanced CMOS. The measured input transfer characteristics of n- and p-type MOSFETs using PZT as the NC booster shows a steep subthreshold swing down to 10 mV/decade together with an enhanced efficiency factor up to 105 V−1. The on-current over off-current ration is improved and the overdrive is boosted up to 0.45 V. Therefore, the supply voltage can be reduced by 50%, maintaining the same performance. This is due to the fact that with the aid of a series connected negative capacitor (i.e., with the internal voltage amplification provided by the NC component of the PZT capacitor) the surface potential in MOS devices is increased beyond the applied gate voltage. It has been also demonstrated that the hysteretic behavior of NC-FETs can be tuned considering the proposed stability condition. Both n- and p-type NC-FETs with large (3-4.5 V) and reduced hysteresis (150-200 mV) are implemented, arguing that a trade- off is required between the steepness and hysteretic behavior of an NC-FET. The impact of the drain-to-source electric field on the boosting effect of NC is also demonstrated and discussed, indicating that a lower lateral electric field in the channel results in a steeper off -to-on transition. It is also experimentally evidenced that a poly-domain ferroelectric cannot have more than one NC domain at a time and shows a zig-zag characteristic. Acknowledgement The authors acknowledge the ERC Advanced Grant Milli-Tech (Grant NO. 695459) for providing the financial support of this research. The authors also greatly appreciate the contributions of Mr. Robin Nigon and Prof. Paul Muralt in the fabrication of the PZT thin film. 14 References (1) Takagi, S. et al. Carrier-transport-enhanced channel CMOS for improved power con- sumption and performance. IEEE Transactions on Electron Devices 55, 21–39 (2008). (2) Khandelwal, S. et al. Circuit performance analysis of negative capacitance FinFETs. In VLSI Technology, 2016 IEEE Symposium on, 1–2 (IEEE, 2016). (3) Ionescu, A. M. et al. Ultra low power: Emerging devices and their benefits for integrated circuits. In Electron Devices Meeting (IEDM), 2011 IEEE International, 16–1 (IEEE, 2011). (4) Salahuddin, S. & Datta, S. Can the subthreshold swing in a classical FET be low- ered below 60 mv/decade? In Electron Devices Meeting, 2008. IEDM 2008. IEEE International, 1–4 (IEEE, 2008). (5) Salahuddin, S. & Datta, S. Use of negative capacitance to provide voltage amplification for low power nanoscale devices. Nano letters 8, 405–410 (2008). (6) Zubko, P. et al. Negative capacitance in multidomain ferroelectric superlattices. Nature 534, 524–528 (2016). (7) Jo, J. et al. Negative capacitance in organic/ferroelectric capacitor to implement steep switching MOS devices. Nano letters 15, 4553–4556 (2015). (8) Gao, W. et al. Room-temperature negative capacitance in a ferroelectric–dielectric superlattice heterostructure. Nano letters 14, 5814–5819 (2014). (9) Ionescu, A. M. Negative capacitance gives a positive boost. Nature nanotechnology 13, 7 (2018). (10) Khan, A. I. et al. Negative capacitance in a ferroelectric capacitor. Nature materials 14, 182 (2015). 15 (11) Appleby, D. J. et al. Experimental observation of negative capacitance in ferroelectrics at room temperature. Nano letters 14, 3864–3868 (2014). (12) Yeung, C. W., Khan, A. I., Sarker, A., Salahuddin, S. & Hu, C. Low power negative ca- pacitance fets for future quantum-well body technology. In VLSI Technology, Systems, and Applications (VLSI-TSA), 2013 International Symposium on, 1–2 (IEEE, 2013). (13) Jain, A. & Alam, M. A. Stability constraints define the minimum subthreshold swing of a negative capacitance field-effect transistor. IEEE Transactions on Electron Devices 61, 2235–2242 (2014). (14) Khan, A. I., Yeung, C. W., Hu, C. & Salahuddin, S. Ferroelectric negative capaci- tance MOSFET: Capacitance tuning & antiferroelectric operation. In Electron Devices Meeting (IEDM), 2011 IEEE International, 11–3 (IEEE, 2011). (15) Rusu, A. & Ionescu, A. M. Analytical model for predicting subthreshold slope improve- ment versus negative swing of S-shape polarization in a ferroelectric FET. In Mixed Design of Integrated Circuits and Systems (MIXDES), 2012 Proceedings of the 19th International Conference, 55–59 (IEEE, 2012). (16) Saeidi, A. et al. Negative capacitance field effect transistors; capacitance matching and non-hysteretic operation. In Solid-State Device Research Conference (ESSDERC), 2017 47th European, 78–81 (IEEE, 2017). (17) Saeidi, A., Biswas, A. & Ionescu, A. M. Modeling and simulation of low power ferro- electric non-volatile memory tunnel field effect transistors using silicon-doped hafnium oxide as gate dielectric. Solid-State Electronics 124, 16–23 (2016). (18) Saeidi, A. et al. Negative capacitance as performance booster for Tunnel FETs and MOSFETs: an experimental study. IEEE Electron Device Letters (2017). 16 (19) Saeidi, A., Jazaeri, F., Stolichnov, I. & Ionescu, A. M. Double-gate negative-capacitance MOSFET with PZT gate-stack on ultra thin body SOI: an experimentally calibrated simulation study of device performance. IEEE Transactions on Electron Devices 63, 4678–4684 (2016). (20) Karda, K., Mouli, C. & Alam, M. Switching dynamics and hot atom damage in landau switches. IEEE Electron Device Letters 37, 801–804 (2016). (21) Li, K.-S. et al. Sub-60mv-swing negative-capacitance FinFET without hysteresis. In Electron Devices Meeting (IEDM), 2015 IEEE International, 22–6 (IEEE, 2015). (22) Jimenez, D., Miranda, E. & Godoy, A. Analytic model for the surface potential and drain current in negative capacitance field-effect transistors. IEEE Transactions on Electron Devices 57, 2405–2409 (2010). (23) Aziz, A., Ghosh, S., Datta, S. & Gupta, S. K. Physics-based circuit-compatible spice model for ferroelectric transistors. IEEE Electron Device Letters 37, 805–808 (2016). (24) Hoffmann, M., Pešić, M., Slesazeck, S., Schroeder, U. & Mikolajick, T. Modeling and design considerations for negative capacitance field-effect transistors. In Ultimate In- tegration on Silicon (EUROSOI-ULIS), 2017 Joint International EUROSOI Workshop and International Conference on, 1–4 (IEEE, 2017). (25) Kidoh, H., Ogawa, T., Morimoto, A. & Shimizu, T. Ferroelectric properties of lead- zirconate-titanate films prepared by laser ablation. Applied physics letters 58, 2910– 2912 (1991). (26) Nakamura, T., Nakao, Y., Kamisawa, A. & Takasu, H. Preparation of Pb (Zr, Ti) O3 thin films on electrodes including IrO2. Applied physics letters 65, 1522–1524 (1994). (27) Khan, A. I., Radhakrishna, U., Chatterjee, K., Salahuddin, S. & Antoniadis, D. A. 17 Negative capacitance behavior in a leaky ferroelectric. IEEE Transactions on Electron Devices 63, 4416–4422 (2016). (28) Lin, C.-I., Khan, A. I., Salahuddin, S. & Hu, C. Effects of the variation of ferroelectric properties on negative capacitance FET characteristics. IEEE Transactions on Electron Devices 63, 2197–2199 (2016). (29) Saeidi, A. et al. Effect of hysteretic and non-hysteretic negative capacitance on tunnel FETs DC performance. Tech. Rep., Institute of Physics (2018). (30) Khan, A. I. et al. Negative capacitance in short-channel finfets externally connected to an epitaxial ferroelectric capacitor. IEEE Electron Device Letters 37, 111–114 (2016). (31) Jo, J. & Shin, C. Negative capacitance field effect transistor with hysteresis-free sub- 60-mv/decade switching. IEEE Electron Device Letters 37, 245–248 (2016). (32) Rusu, A., Saeidi, A. & Ionescu, A. M. Condition for the negative capacitance ef- fect in metal–ferroelectric–insulator–semiconductor devices. Nanotechnology 27, 115201 (2016). (33) Zhu, Z. et al. Negative-capacitance characteristics in a steady-state ferroelectric capac- itor made of parallel domains. IEEE Electron Device Letters 38, 1176–1179 (2017). 18
1808.09787	2	1808	2018-09-03T17:19:09	Study of the Fracturing Behavior of Thermoset Polymer Nanocomposites via Cohesive Zone Modeling	[ "physics.app-ph", "cond-mat.soft" ]	This work proposes an investigation of the fracturing behavior of polymer nanocomposites. Towards this end, the study leverages the analysis of a large bulk of fracture tests from the literature with the goal of critically investigating the effects of the nonlinear Fracture Process Zone (FPZ). It is shown that for most of the fracture tests the effects of the nonlinear FPZ are not negligible, leading to significant deviations from Linear Elastic Fracture Mechanics (LEFM) sometimes exceeding 150% depending on the specimen size and nanofiller content. To get a deeper understanding of the characteristics of the FPZ, fracture tests on geometrically-scaled Single Edge Notch Bending (SENB) specimens are analyzed leveraging a cohesive zone model. It is found that the FPZ cannot be neglected and a bi-linear cohesive crack law generally provides the best match of experimental data.	physics.app-ph	physics	A&A Program in Structures William E. Boeing Department of Aeronautics and Astronautics University of Washington Seattle, Washington 98195, USA Study of the Fracturing Behavior of Thermoset Polymer Nanocomposites via Cohesive Zone Modeling Yao Qiao, Marco Salviato INTERNAL REPORT No. 18-07/02E Submitted to Composite Structures July 2018 Study of the Fracturing Behavior of Thermoset Polymer Nanocomposites via Cohesive Zone Modeling Yao Qiaoa, Marco Salviatoa,∗ aWilliam E. Boeing Department of Aeronautics and Astronautics, University of Washington, Seattle, Washington 98195, USA Abstract This work proposes an investigation of the fracturing behavior of polymer nanocom- posites. Towards this end, the study leverages the analysis of a large bulk of fracture tests from the literature with the goal of critically investigating the effects of the non- linear Fracture Process Zone (FPZ). It is shown that for most of the fracture tests the effects of the nonlinear FPZ are not negligible, leading to significant deviations from Linear Elastic Fracture Mechanics (LEFM) sometimes exceeding 150% depending on the specimen size and nanofiller content. To get a deeper understanding of the characteristics of the FPZ, fracture tests on geometrically-scaled Single Edge Notch Bending (SENB) specimens are analyzed leveraging a cohesive zone model. It is found that the FPZ cannot be neglected and a bi-linear cohesive crack law generally provides the best match of experimental data. Keywords: Fracture, Size effect, Nanocomposites, Crack, Cohesive zone models 1. Introduction The outstanding advances in polymer nanocomposites have paved the way for their broad use in engineering. Potential applications of these materials include microelec- tronics [1], energy storage [2] and harvesting [3], soft robotics [4], and bioengineering [5]. One of the reasons of this success is that, along with remarkable enhancements of phys- ical properties such as e.g. electric and thermal conductivity [6, 7], nanomodification offers significant improvements of stiffness [8, 9], strength [10] and toughness [11 -- 16]. These aspects make it an excellent technology to enhance the mechanical behavior of ∗Corresponding Author, Email address: [email protected] (Marco Salviato) Preprint submitted to Composite Structures September 5, 2018 polymers [17 -- 23, 26 -- 32] or to improve the weak matrix-dominated properties of fiber composites [33, 34]. While a large bulk of data on the mechanical properties of polymer nanocomposites is available already, an aspect often overlooked is the effect on the fracturing behavior of the region close to the crack tip featuring most of energy dissipation, the Fracture Process Zone (FPZ). This is an important aspect since, due to the complex mesostruc- ture characterizing nanocomposites, the size of the non-linear FPZ occurring in the presence of a large stress-free crack is usually not negligible [35 -- 39]. The stress field along the FPZ is nonuniform and decreases with crack opening, due to a number of damage mechanisms such as e.g. discontinuous cracking, micro-crack deflection, plastic yielding of nanovoids, shear banding and micro-crack pinning [11 -- 13, 26, 41 -- 46]. As a consequence, the fracturing behavior and, most importantly, the energetic size effect associated with the given structural geometry, cannot be described by means of clas- sical Linear Elastic Fracture Mechanics (LEFM) which assumes the effects of the FPZ to be negligible. To seize the effects of a finite, non-negligible FPZ, the introduction of a characteristic (finite) length scale related to the fracture energy and the strength of the material is necessary [35 -- 39]. This work proposes an investigation on the fracturing behavior of thermoset polymer nanocomposites with the goal of critically investigating the effects of the nonlinear Fracture Process Zone (FPZ). By employing Size Effect Law (SEL), a formulation endowed with a characteristic length inherently related to the FPZ size, and assuming a linear cohesive behavior [40], a large bulk of literature data is analyzed. It is shown that for most of the fracture tests, the nonlinear behavior of the FPZ is not negligible, leading to significant deviations from LEFM. As the data indicate, this aspect needs to be taken into serious consideration since the use of LEFM to estimate mode I fracture energy can lead to an error as high as 157% depending on the specimen size and nanofiller content. A cohesive zone model featuring a Linear Cohesive Law (LCL) is used to further understand the fracturing behavior of polymer nanocomposites. It is shown that while 2 the LCL with corrected fracture energy by SEL is capable of capturing experimental data, this is not the case for the LCL with the fracture energy calculated by LEFM. This is the confirmation that Size Effect Law (SEL) can be adapted to re-analyze the fracture tests available in the literature for the first approximation. Taking advantage of size effect tests on thermoset-based graphene nanocomposites by Mefford et al. [13], it is also found that these materials are better described by a bi-linear cohesive law. As the results show, while the use of a linear cohesive law provides a good approximation, a bi-linear cohesive law provides a superior description of the fracturing behavior for different sizes. 2. Quasi-brittle Fracture of Nanocomposites 2.1. Size effect law for nanocomposites The fracture process in nanocomposites can be analyzed leveraging an equivalent linear elastic fracture mechanics approach to account for the presence of a FPZ of finite size as shown in Figure 1. To this end, an effective crack length a = a0 + cf with a0 = initial crack length and cf = effective FPZ length is considered. Following LEFM, the energy release rate can be written as follows: G (α) = σ2 N D E∗ g(α) (1) where α = a/D = normalized effective crack length, E∗ = E for plane stress and E∗ = E/ (1 − ν2) for plane strain, g (α) = dimensionless energy release rate and, D is represented in Figure 2 for Single Edge Notch Bending (SENB) and Compact Tension (CT) specimens respectively. σN represents the nominal stress defined as e.g. σN = 3P L/ (2tD2) for SENB specimens or σN = P/ (tD) for CT specimens where, following Figure 2, P is the applied load, t is the thickness and L is the span between the two supports for a SENB specimen as defined in ASTM D5045-99 [47]. At incipient crack onset, the energy release rate ought to be equal to the fracture energy of the material. Accordingly, the failure condition can now be written as: G (α0 + cf /D) = σ2 N cD E∗ g (α0 + cf /D) = Gf (2) 3 where Gf is the mode I fracture energy of the material and cf is the effective FPZ length, both assumed to be material properties. It should be remarked that this equation characterizes the peak load conditions if g(cid:48)(α) > 0, i.e. only if the structure has positive geometry [38]. By approximating g (α) with its Taylor series expansion at α0 and retaining only up to the linear term of the expansion, one obtains: (3) (4) Gf = σ2 N cD E∗ g(α0) + g(cid:48)(α0) cf D (cid:105) (cid:104) (cid:115) E∗Gf Dg(α0) + cf g(cid:48)(α0) which can be rearranged as follows [38]: σN c = where g(cid:48) (α0) = dg (α0) /dα. This equation, known as Bazant's Size Effect Law (SEL) [35, 36, 38, 39], relates the nominal strength to mode I fracture energy, a characteristic size of the structure, D, and to a characteristic length of the material, cf , and it can be rewritten in the σ0(cid:112)1 + D/D0 following form: with σ0 =(cid:112)E∗Gf /cf g(cid:48)(α0) and D0 = cf g(cid:48)(α0)/g(α0) = constant, depending on both σN c = (5) FPZ size and specimen geometry. Contrary to classical LEFM, Eq. (5) is endowed with a characteristic length scale D0. This is key to describe the transition from ductile to brittle behavior with increasing structure size. 2.2. Calculation of g (α) and g(cid:48) (α) 2.2.1. Single Edge Notch Bending (SENB) specimens The calculation of g(α) and g(cid:48)(α) for SENB specimens can be done according to the procedure described in [13]. This leads to the following polynomial expressions: g(α) = 1155.4α5 − 1896.7α4 + 1238.2α3 − 383.04α2 + 58.55α − 3.0796 g(cid:48)(α) = 18909α5 − 31733α4 + 20788α3 − 6461.5α2 + 955.06α − 50.88 (6) (7) 4 2.2.2. Compact Tension (CT) specimens In the case of CT specimens, the values for g(α) and g(cid:48)(α) can be determined leveraging the equations provided by ASTM D5045-99 [47]. Following the standard, the mode I Stress Intensity Factor (SIF), KI, can be written as: KI = √ P t D f (α) (8) where α = a/D and D is the distance between the center of hole to the end of the specimen as defined in ASTM D5045-99 [47] (see Figure 2b). The nominal stress σN for CT specimens can be defined as: σN = P tD (9) The mode I Stress Intensity Factor can be rewritten as follows by combining Eq. (8) and Eq. (9): √ DσN f (α) KI = (10) By considering the relationship between energy release rate and stress intensity factor for a plane strain condition, the mode I energy release rate results into the following expression: Dσ2 N E g(α) GI = (11) where g(α) = f 2(α)(1 − υ2), and f (α) is a dimensionless function accounting for geo- metrical effects and the finiteness of the structure (see e.g. [47]). Once g(α) is derived, the expression of g(cid:48)(α) can be obtained by differentiation leading to the following poly- nomial expressions for g(α) and g(cid:48)(α) respectively: g(α) = 33325α5 − 52330α4 + 32016α3 − 9019.1α2 + 1230.1α − 51.944 g(cid:48)(α) = 555868α5 − 895197α4 + 554047α3 − 159153α2 + 21035α − 917.3 (12) (13) 3. Fracture behavior of thermoset nanocomposites: analysis and discussion In the following sections, a large bulk of data on the fracturing behavior of nanocom- posites is critically analyzed employing the expressions derived in Section 2. First, 5 fracture tests data on the thermoset polymer reinforced by different nanoparticles are analyzed to investigate how the FPZ affects the failure behavior. Then, leveraging SEL and assuming a linear cohesive behavior, a large bulk of data from the literature originally elaborated by LEFM is re-analyzed to include the effects of the FPZ. 3.1. Fracture Scaling of Nanocomposites To investigate the effects of the non-linear FPZ, the fracture tests on the ther- moset polymer reinforced by nanoparticles in the literature are analyzed and discussed. Figure 3 shows the normalized structural strength σN c/σ0 of the literature data as a function of the normalized structure size D/D0 in double logarithmic scale. The solid line represents the fitting by SEL. In such a graph, the structural scaling predicted by LEFM is represented by a dashed line of slope −1/2 whereas the case of no scal- ing, as predicted by stress-based failure criteria, is represented by a horizontal line. The intersection between the LEFM asymptote, typical of brittle behavior, and the pseudo-plastic asymptote, typical of ductile behavior, corresponds to D = D0, called the transitional size [38]. As can be noted from Figure 3, the experimental data are in excellent agreement with SEL, which inherently captures the transition from strength-dominated to toughness- dominated fracture. More importantly, the figure shows that although some fracture tests reported in the literature were conducted under LEFM conditions (assumed by ASTM D5045-99 [47]), most of the data are located in the transitional region. Accordingly, the experimental data show that LEFM does not always provide an accurate method to extrapolate the structural strength of larger structures from lab tests on small-scale specimens, especially if the size of the specimens belonged to the transitional zone. In fact, the use of LEFM in such cases may lead to a significant underestimation of structural strength, thus hindering the full exploitation of graphene nanocomposite fracture properties. This is a severe limitation in several engineering applications such as e.g. aerospace or aeronautics for which structural performance optimization is of utmost importance. On the other hand, LEFM always overestimates 6 significantly the strength when used to predict the structural performance at smaller length-scales. This is a serious issue for the design of e.g. graphene-based MEMS and small electronic components or nanomodified carbon fiber composites in which the inter- fiber distance occupied by the resin is only a few micrometers and it is comparable to the FPZ size. In such cases, SEL or other material models characterized by a characteristic length scale ought to be used. 3.2. Effects of a finite FPZ on the calculation of Mode I fracture energy Notwithstanding the importance of understanding the scaling of the fracturing be- havior, the tests conducted by Mefford et al. [13] represent, to the best of the authors' knowledge, the only comprehensive investigation on the size effect in nanocomposites available to date. All the fracture tests reported in the literature were conducted on one size and analyzed by means of LEFM. Considering the remarkable effects of the nonlinear FPZ on the fracturing behavior documented in the foregoing section, it is interesting to critically re-analyze the fracture tests available in the literature by means of SEL. This formulation is endowed with a characteristic length related to the FPZ size and, different from LEFM, it has been shown to accurately capture the transition from brittle to quasi-ductile behavior of nanocomposites. 3.2.1. Application of SEL to thermoset polymer nanocomposites To understand if the quasi-brittle behavior reported in previous tests [13] is a salient feature of graphene nanocomposites only or if it characterizes other nanocomposites, a large bulk of literature data are re-analyzed by SEL using Eq.(3) in order to study the effects of the FPZ. In this analysis, in the absence of data on the effective FPZ length, cf , from the literature, it is assumed that cf = 0.44lch which, according to Cusatis et [40], corresponds to the assumption of a linear cohesive law. In this expression, is Irwin's characteristic length which depends on Young's modulus E∗, al. lch = E∗Gf /f 2 the mode I fracture energy Gf and the ultimate strength of the material ft. Substituting t this expression into Eq. (3) and rearranging one gets the following expression which relates the fracture energy calculated according to SEL to the fracture energy calculated 7 by LEFM: Gf,SEL = Gf,LEF M 1 − 0.44E∗g(cid:48)(α0)Gf,LEF M Df 2 t g(α0) (14) In this equation, Gf,LEF M = σ2 N cDg(α0)/E∗ represents the fracture energy which can be estimated by analyzing the fracture tests by LEFM. It can be observed from Eq.(14) that the correct fracture energy in the literature can be calculated by knowing three key parameters provided that g(α0) and g(cid:48)(α0) are known: (1) the fracture energy through the use of LEFM; (2) the Young's modulus of the specimens at different nanoparticle concentrations; and (3) the ultimate strength of the specimens at different nanoparticle concentrations. For cases in which those parameters are not provided by the authors, the ultimate strength, Young's modulus, and Poisson's ratio of nanocomposites are reasonably assumed to be 50 MPa, 3000 MPa, and 0.35 respectively. 3.2.2. Mode I fracture energy of thermoset polymer nanocomposites Several types of nanofillers are investigated in this re-analysis including carbon-based nano-fillers (such as carbon black, graphene oxide, graphene nanoplatelets, and multi- wall carbon nanotubes), rubber and silica nanoparticles, and nanoclay. The fracture energy estimated from LEFM compared to the calculation through SEL, Eq. (14), for nanomodified SENB and CT specimens are plotted in Figures 4-8 along with the highest difference. Figure 4 shows data elaborated from Carolan et al. [17] who conducted fracture tests on SENB specimens nano-modified by six different combinations of nanofillers. As can be noted, while for the pristine polymer the difference between LEFM and SEL is negligible, this is not the case for the nanomodified polymers, the difference increasing with increasing nanofiller content. The difference varies based on the type of nanofiller used, with the greatest value being 42.6% for the addition of 8 wt% core shell rubber mixed with 25% diluent and 8% silica. This confirms that for the SENB specimens tested in [17] the nonlinear behavior of the FPZ is not negligible, leading to a more ductile behavior compared to the pristine polymer. 8 Similar conclusions can be drawn based on Figures 5a-f which report the analysis of fracture tests conducted by Zamanian et al. [18] and Jiang et al. [8] on polymers reinforced by silica nanoparticles and silica nanoparticle+graphene oxide respectively. For the data in [18], the greatest percent difference of fracture energy between LEFM and SEL decreased as the size of silica nanoparticle increased, with the greatest dif- ference being 28% for the addition of 6 wt% 12 nm silica nanoparticles. For all the systems investigated, the maximum deviation from LEFM is for the largest amount of nanofiller, confirming that nanomodification lead to larger FPZ sizes and more pro- nounced ductility. On the other hand, the data by Jiang et al. [8] exhibit an even larger effect of the FPZ with the greatest difference in fracture energy between LEFM and SEL reaching up to 51.8% for silica nanoparticle attached to graphene oxide. A milder effect of the FPZ can be inferred from the data by Chandrasekaran et al. [19] who investigated three types of carbon-based nano-fillers (Figure 6): (1) ther- mally reduced graphene oxide; (2) graphene nanoplatelets; and (3) multi-wall carbon nanotubes. In these cases, the difference between SEL and LEFM ranges from 4.9% to 8.8%, the lowest difference among all the data analyzed in this study. For these systems, the specimen size compared to the size of the nonlinear FPZ is large enough to justify the use of LEFM which provided accurate and objective results. On the other hand, a more significant effect of the FPZ can be inferred from the data reported by Konnola et al. [10] who studied three different types of functionalized and nonfunc- tionalized nano-fillers. In this case, the greatest difference in fracture energy ranges between 15.2% to 20.3%. SENB specimens nano-modified by nanoclay and carbon black respectively were tested by Kim et al. [20]. As Figure 7 shows, in this case, the specimen size is enough to justify the use of LEFM as confirmed by the low difference with SEL (11.2% for nanoclay and 7.3% for carbon black). Similar conclusions can be drawn on the silica nanoparticles investigated by Vaziri et al. [21]. However, for the three different sizes of silica nanoparticles investigated by Dittanet et al. [22], a significant difference between LEFM and SEL is observed, confirming that these specimens belonged to the transition 9 zone between ductile and brittle behavior where the effects of the nonlinear FPZ cannot be neglected. Figure 8 shows a re-analysis of the data reported by Liu et al. [23] who tested CT specimens nano-modified by four different combinations of silica nanoparticle and rubber. As can be noted, in this case, the FPZ indeed affects the fracturing behavior significantly. Adopting LEFM, which assumes the size of the FPZ to be negligible, for the estimation of Gf would lead to an underestimation of up to 156.8% for the case of polymer reinforced by 15 wt% rubber only. This tremendous difference, the largest found in the present study, gives a tangible idea on the importance of accounting for the nonlinear damage phenomena occurring in nanocomposites which can lead to a significant deviation from the typical brittle behavior of thermoset polymers. 4. Cohesive Zone Modeling of Thermoset Nanocomposites To have a deeper understanding of the fracturing behavior of nanocomposites, a computational investigation is conducted leveraging a cohesive zone model featuring a Linear Cohesive Law (LCL) in ABAQUS Explicit 2017. To this end, as illustrated in Figure 9, a Single Edge Notch Bending (SENB) specimen is simulated by 4-node two- dimensional cohesive elements (COH2D4) with a traction-separation law to model the crack and 4-node bi-linear plain strain quadrilateral elements (CPE4R) with a linear elastic isotropic behavior to model the rest of the specimen. The width of crack is modeled as 4 µm based on the image obtained from Scanning Electron Microscopy (SEM). 4.1. Linear Cohesive Crack Law To corroborate the results discussed in the forgoing sections, the analysis by means of a cohesive zone model is carried out using the fracture energy estimated via LEFM, GLEF M f , and the one calculated through Eq.(14), GSEL f . As can be noted from Figures 10-13, the cohesive zone model using GSEL f as input shows an excellent agreement with the experimental data in the literature. However, this is not the case if Gf by LEFM is 10 used. In fact, this is a further confirmation that Size Effect Law (SEL) can be adapted to re-analyze the fracture tests available in the literature. By leveraging a linear cohesive crack modeling with the corrected Gf , the fracturing behavior on the scaling of nanocomposites can be predicted without additional tests in the lab. As Figures 14-17 show, experimental data in the literature and simulation results using a linear cohesive crack law are plotted along with the analytical expression for Cohesive Size Effect Curves (CSEC) purposed by Cusatis et al. [40]. In these Figures, it can be noted that, for large specimen sizes, the prediction on the peak load of investigated nanocomposites by using LEFM Gf leads to a significant underestimation. It is worth mentioning here that this analysis is on the assumption that nanocomposites in the literature follow a linear cohesive law. 4.2. Bi-linear Cohesive Crack Law Thanks to the comprehensive investigation on the size effect in graphene nanocom- posites by Mefford et al [13], the characteristics of the cohesive crack law can be further studied. To this end, both linear and bi-linear cohesive laws with the same fracture energy are used to match load-displacement curves obtained from experimental frac- ture tests on geometrically scaled Single Edge Notch Bending (SENB) specimens with varying contents of graphene. As illustrated in Figure 18, a linear cohesive law can be described through two parameters: (a) tensile strength, ft and (b) fracture energy, Gf , which represents the area under the linear cohesive law. On the other hand, a bilinear cohesive law requires four parameters: (a) tensile strength ft, (b) initial fracture en- ergy, Gf , which represents the area under the initial segment of the bi-linear cohesive law; (b) total fracture energy, GF , which is the total area under the bi-linear cohesive law; (d) change-of-slope stress, σk, which is the value of stress at the intersection of the initial and tail segments. It is worth mentioning here that, for the bi-linear cohe- sive law, different intersection points are investigated in order to match experimental load-displacement curves. Figures 19-20 show a comparison between the experimental load defection curves 11 and simulation through a Cohesive Zone Model (CZM) featuring a linear and bi-linear cohesive law respectively. It can be noted that, while the bi-linear cohesive law suc- cessfully matches experimental curves of specimens with different sizes and graphene concentrations, this is not the case for the linear cohesive law, with a significant un- derestimation of the experimental curves. In fact, the bi-linear cohesive law provides a very accurate description of fracture tests with errors on structural strength less than 7% whereas the linear cohesive law shows a maximum deviation from the tests of 30%. This result suggests that a bi-linear cohesive law may be better suited for the descrip- tion of the cohesive fracture behavior of nanocomposites, although a linear cohesive law may still provide reasonable results and can be used for a preliminary, course, approx- imation. Further size effect studies on different material systems will shed more light on this important aspect. A comparison between the calibrated linear and bi-linear cohesive laws for the various graphene contents is shown in Figures 21a-d. Figure 22 shows the initial, Gf , and total, GF , fracture energy as a function of graphene platelet content. It is interesting to note that the initial fracture energy does not increase significantly as a function of graphene content. The increasing total frac- ture energy for higher graphene contents can all be ascribed to the change in slope of the second part of the curve. This is the indication that, for crack opening displacements lower than about 20µm, which is e.g. the case of a crack propagating between microm- eter fibers in a unidirectional composite (Figure 23), the effects of nanomodification may be negligible. In fact, as schematically explained in inserts (b) and (c) of Figure 23), in such a case only the initial part of the bi-linear cohesive law is developed and drives the fracturing behavior. This may explain why the use of nanomodification to improve the fracturing behavior of the polymer matrix in fiber composites has met with changing fortunes. Even if increments of the total fracture energy by nanomodification can be observed from tests on laboratory-scale specimens, this does not guarantee that the initial portion of the cohesive curve, which drives the microcracking in composites, has improved. This latter aspect can be clarified only by size effect testing and cohesive zone modeling, as clearly shown in this work. 12 5. Conclusions Leveraging a large bulk of literature data, this paper investigated the effects of the Fracture Process Zone (FPZ) on the fracturing behavior of thermoset polymer nanocomposites, an aspect of utmost importance for structural design but so far over- looked. Based on the results obtained in this study, the following conclusions can be elaborated: 1. The double logarithmic plots of the normalized structural strength as a function of the normalized characteristic size of geometrically-scaled SENB specimens show that the experimental data on nanocomposites available in the literature are in excellent agreement with Size Effect Law (SEL). Most of nanocomposites are located in the transitional range in which the fracturing behavior cannot be characterized by Linear Elastic Fracture Mechanics (LEFM); 2. Size Effect Law and Cohesive Zone Modeling show that for most of the fracture tests on polymer nanocomposites investigated in this work, the effects of the nonlinear FPZ are not negligible, leading to significant deviations from LEFM. As the data indi- cate, this aspect needs to be taken into serious consideration since the use of LEFM to estimate mode I fracture energy can lead to an error as high as 156% depending on the specimen size and nanofiller content; 3. The deviation from LEFM reported in the re-analyzed results is related to the size of the Fracture Process Zone (FPZ) for increasing contents of nanofiller. In the pristine polymer the damage/fracture zone close to the crack tip, characterized by significant non-linearity due to subcritical damaging, is generally very small compared to the specimen sizes investigated. This is in agreement with the inherent assumption of LEFM of negligible non-linear effects during the fracturing process. However, the addition of nano-fillers results in larger and larger FPZs. For sufficiently small specimens, the size of the highly non-linear FPZ is not negligible compared to the specimen characteristic size thus highly affecting the fracturing behavior, this resulting into a significant deviation from LEFM; 13 4. To get a deeper understanding of the cohesive behavior of nanocomposites, re- cent fracture tests on thermosets reinforced by graphene nanoplatelets were re-analyzed via a cohesive model featuring a bi-linear law for all the sizes and graphene contents considered. It is concluded that, in general, a bi-linear cohesive law provides a very accurate description of fracture tests with errors on structural strength less than 7%. A reasonable agreement is also found leveraging a linear cohesive law with errors on structural strength no larger than 30%. 5. The analysis via the bi-linear cohesive law provided unprecedented insights on the influence of graphene nanoplatelets on the cohesive stresses. It is found that, for the size range investigated, the initial part of the cohesive law is unaffected by nanomodification. The increasing total fracture energy for higher graphene contents can all be ascribed to the change in slope of the second part of the cohesive law. Two main considerations can be made from this result: (a) the toughening by graphene nanoplatelets requires sufficiently large crack opening displacement (larger than about 20µm for the system investigated in this work), confirming that mechanisms such as crack deflection and splitting are the main sources of energy dissipation; (b) for very small crack opening displacements, such as for the case of a crack propagating between micrometer fibers in a composite, the effect of nanomodification may be negligible, since no change in the cohesive behavior is induced by graphene nanoplatelets in that regime. Of course, different nanoparticles and manufacturing processes may affect the initial portion of the cohesive law differently. Future work will focus on understanding the physical relation between the characteristics of the cohesive law and the nano/microstructure of the material. Acknowledgments Marco Salviato acknowledges the financial support from the Haythornthwaite Foun- dation through the ASME Haythornthwaite Young Investigator Award and from the University of Washington Royalty Research Fund. This work was also partially sup- ported by the William E. Boeing Department of Aeronautics and Astronautics as well 14 as the College of Engineering at the University of Washington through Salviato's start up package. References References [1] Rogers JA. Electronic materials: making graphene for microelectronics. Nat Nan- otechnol 2008;3:254-55. [2] Yoo EJ, Kim J, Hosono E, Zhou HS, Kudo T, Honma I. Large Reversible Li Storage of Graphene Nanosheet Families for Use in Rechargeable Lithium Ion Batteries. Nano Lett 2008;8:2277-82 [3] Chien CT, Hiralal P, Wang DY, Huang IS, Chen CC, Chen CW, Amaratunga GA.J. Graphene-Based Integrated Photovoltaic Energy Harvesting/Storage De- vice. Small 2015;11:2929-37. [4] Zhang JF, Ryu S, Pugno N, Wang QM, Tu Qing, Buehler MJ, Zhao XH. Multi- functionality and control of the crumpling and unfolding of large-area graphene. Nat Mater 2013;12:321. [5] Kuila T, Bose S, Khanra P, Mishra AK, Kim NH, Lee JH. Recent advances in graphene-based biosensors. Biosens Bioelectron 2011;26:4637-48. [6] Balandin AA, Ghosh S, Bao WZ, Calizo I, Teweldebrhan D, Miao F, Lau CN. Superior thermal conductivity of single-layer graphene. Nano Lett 2008;8:902-7. [7] Ramirez C, Figueiredo FM, Miranzo P, Poza P, Osendi MI. Graphene nanoplatelet/silicon nitride composites with high electrical conductivity. Carbon 2012;50:3607-15. [8] Jiang T, Kuila T, Kim NH, Ku BC, Lee JH. Enhanced mechanical properties of silanized silica nanoparticle attached graphene oxide/epoxy composites. Compos Sci Technol 2013;79:115-25. [9] King JA, Klimek DR, Miskioglu I, Odegard GM. Mechanical properties of graphene nanoplatelet/epoxy composites. J Appl Polym Sci 2013;128:4217-23. [10] Konnola R, Nair CPR, Joseph K. High strength toughened epoxy nanocomposite based on poly(ether sulfone)-grafted multi-walled carbon nanotube. Polym Advan Technol 2015;27:82-89. [11] Zappalorto M, Salviato M, Quaresimin M. Mixed mode (I+II) fracture toughness of polymer nanoclay nanocomposites. Eng Fract Mech 2013;111:50-64. [12] Zappalorto M, Salviato M, Pontefisso A, Quaresimin M. Notch effect in clay- modified epoxy: a new perspective on nanocomposite properties. Compos Inter- faces 2013;20:405-19. 15 [13] Mefford C, Qiao Y, Salviato M. Failure behavior and scaling of graphene nanocom- posites, Compos Struct 2017;176:961-72. [14] Ashrafi B, Guan JW, Mirjalili V, Zhang YF, Chun L, Hubert P, Simard B, Kingston CT, Orson B, Johnston A. Enhancement of mechanical performance of epoxy/carbon fiber laminate composites using single-walled carbon nanotubes. Compos Sci Technol 2011;71:1569-78. [15] Abhishek K, Samit R. Characterization of mixed mode fracture properties of nanographene reinforced epoxy and Mode I delamination of its carbon fiber com- posite. Compos Part B-Eng 2018;134:98-105. [16] Abhishek K, Shibo Li, Samit Roy, Julia A.K, Odegard GM. Fracture properties of nanographene reinforced EPON 862 thermoset polymer system. Compos Sci Technol 2015;114:87-93. [17] Carolan D, Ivankovic A, Kinloch AJ, Sprenger S, Taylor AC. Toughening of epoxy- based hybrid nanocomposites. Polymer 2016;97:179-190. [18] Zamanian M, Mortezaei M, Salehnia B, Jam JE. Fracture toughness of epoxy polymer modified with nanosilica particles: particle size effect. Eng Fract Mech 2013;97:193-206. [19] Chandrasekaran S, Sato N, Tolle F, Mulhaupt R, Fiedler B, Schulte K. Fracture toughness and failure mechanism of graphene based epoxy composites. Compos Sci Technol 2014;97:90-99. [20] Kim BC, Park SW, Lee DG. Fracture toughness of the nano-particle reinforced epoxy composite. Compos Struct 2008;86:69-77. [21] Vaziri HS, Abadyan M, Nouri M, Omaraei IA, Sadredini Z, Ebrahimnia M. Inves- tigation of the fracture mechanism and mechanical properties of polystyrene/silica nanocomposite in various silica contents. J Mater Sci 2011;46:5628-38. [22] Dittanet P, Pearson R. Effect of silica nanoparticle size on toughening mechanisms of filled epoxy Polymer 2012;53:1890-1905. [23] Liu HY, Wang GT, Mai YW, Zeng Y. On fracture toughness of nano-particle modified epoxy. Compos Part B-Eng 2011;42:2170-75. [24] Suman C, Nitai CA, Pranab S, Naresh CM, Tapas K. Functionalized reduced graphene oxide/epoxy composites with enhanced mechanical properties and ther- mal stability. Polym Test 2017;61:1-11 [25] Mirjalili V, Ramachandramoorthy R, Hubert P. Enhancement of fracture toughness of carbon fiber laminated composites using multi wall carbon nanotubes. Carbon 2014;79:413-423. [26] Zhang H, Tang LC, Zhang Z, Friedrich K, Sprenger S. Fracture behaviors of in situ silica nanoparticle-filled epoxy at different temperatures. Polymer 2008;49:3816-25. [27] Johnsen BB, Kinloch AJ, Mohammed RD, Taylor AC, Sprenger S. Toughening mechanisms of nanoparticle-modified epoxy polymers. Polymer 2007;48:530-41. 16 [28] Wang X, Jin J, Song M. An investigation of the mechanism of graphene toughening epoxy. Carbon 2013;65:324-33. [29] Chandrasekaran S, Seidel C, Schulte K. Preparation and characterization of graphite nano-platelet (GNP)/epoxy nano-composite: mechanical, electrical and thermal properties. Eur Polym J 2013;49:3878-88. [30] Naous W, Yu XY, Zhang QX, Naito K, Kagawa y. Morphology, tensile properties and fracture toughness of epoxy/Al2O3 nanocomposites. J Polym Sci Pol Phys 2006;44:1466-73. [31] Wetzel B, Rosso P, Haupert F, Friedrich K. Epoxy nanocomposites-fracture and toughening mechanisms. Eng Fract Mech 2006;73:2375-98. [32] Quaresimin M, Salviato M, Zappalorto M. Toughening mechanisms in nanoparticle polymer composites: experimental evidences and modeling." In book: Toughening Mechanisms in Composite Materials 2015; [33] Quaresimin M, Salviato M, Zappalorto M. Fracture and interlaminar properties of clay-modified epoxies and their glass reinforced laminates. J Eng Mech 2012; 81: 80-93. [34] Pathak AK, Borah M, Gupta A, Yokozeki T, Dhakate SR, Improved mechanical properties of carbon fiber/graphene oxide-epoxy hybrid composites. Compos Sci Technol 2016;135:28-38. [35] Bazant ZP. Size effect in blunt fracture: concrete, rock, metal. J Eng Mech-ASCE 1984;110:518-35. [36] Bazant ZP, Kazemi MT. Determination of fracture energy, process zone length and brittleness number from size effect, with application to rock and concrete. Int J Fracture 1990;44:111-31. [37] Bazant ZP, Daniel IM, Li Z. Size effect and fracture characteristics of composite laminates. J Eng Mater-T ASME 1996;118:317-24. [38] Bazant ZP, Planas J. Fracture and size effect in concrete and other quasi-brittle materials. Boca Raton:CRC Press;1998. [39] Salviato M, Kirane K, Ashari SE, Bazant ZP, Cusatis G. Experimental and nu- merical investigation of intra-laminar energy dissipation and size effect in two- dimensional textile composites. Compos Sci Technol 2016;135:67-75. [40] Cusatis G, Schauffert EA, Cohesive crack analysis of size effect, Eng Fract Mech 2009; 76:2163-73. [41] Salviato M, Zappalorto M, Quaresimin M. Plastic yielding around nanovoids, Pro- cedia Engineer 2011;10:3316-21. [42] Chandrasekaran S, Sato N, Tlle F, Mlhaupt R, Fiedler B, Schulte K. Fracture toughness and failure mechanism of graphene based epoxy composites. Compos Sci Technol 2014;97:909. 17 [43] Salviato M, Zappalorto M, Quaresimin M. Plastic shear bands and fracture tough- ness improvements of nanoparticle filled polymers: a multiscale analytical model. Compos Part A - Appl S 2013;48:144-52. [44] Salviato M, Zappalorto M, Quaresimin M. Nanoparticle debonding strength: a comprehensive study on interfacial effects. Int J Solids Struct 2013;50:3225-32. [45] Quaresimin M, Salviato M, Zappalorto M. A multi-scale and multi-mechanism ap- proach for the fracture toughness assessment of polymer nanocomposites. Compos Sci Technol 2014;91:16-21. [46] Quaresimin M., Schulte K., Zappalorto M., Chandrasekaran S., Toughening mech- anisms in polymer nanocomposites: From experiments to modelling, Compos Sci Tech 2016;123:187-204. [47] ASTM D5045-99 - Standard Test Methods for Plane-Strain Fracture Toughness and Stain Energy Release Rate of Plastic Materials 1999. 18 Figures Figure 1: Fracture Process Zone (FPZ) for thermoset polymer nanocomposites. Figure 2: Schematic representation of the SENB and CT specimens considered in this work. 19 tLDDtPPPa)b)Figure 3: Size effect curves in polymer nanocomposites: data taken from the literature [8, 10, 17 -- 25]. 20 -1.2-1-0.8-0.6-0.4-0.200.2-1-0.500.511.52SELLEFMCarolanLiuZemanianJiangKonnolaChandrasekaranKimVaziriDittanetLiuTapasHubertlog(Τ𝐷𝐷𝑜)log(Τ𝜎𝑁𝑐𝜎𝑜)et. alet. alet. alet. alet. alet. alet. alet. alet. alet. alet. alet. al21Strength CriterionFigure 4: Mode I fracture energy estimated by Linear Elastic Fracture Mechanics (LEFM) and Size Effect Law (SEL), Eq. (14). The latter formulation accounts for the finite size of the nonlinear Fracture Process Zone (FPZ) in thermoset nanocomposites. Data re-analyzed from [17]. 21 00.050.10.150.20.250.30.350.40369121518212400.20.40.60.811.21.41.61.80369121518Fracture Energy, [N/mm]66012CSR+ 25% Diluent28.4%[Unit: mm]SELLEFM0.10.40.711.31.6024681066012 CSR+ 25% Diluent+4% Silica34.3%[Unit: mm]SELLEFM0.10.40.711.31.61.9024681066012Core Shell Rubber [wt%]CSR+ 25% Diluent+8% Silica42.6%[Unit: mm]SELLEFM00.20.40.60.811.21.40369121518212466012CSR21.2%[Unit: mm]SELLEFM00.050.10.150.20.250.30.350.40369121518212466012 Silica11.1%[Unit: mm]SELLEFM66012 Silica+ 25% Diluent8.8%[Unit: mm]SELLEFMSilica Nanoparticles [wt%]Carolan et al. Carolan et al. Carolan et al. Carolan et al. Carolan et al. Carolan et al. a)b)c)d)e)f)Figure 5: Mode I fracture energy estimated by Linear Elastic Fracture Mechanics (LEFM) and Size Effect Law (SEL), Eq. (14). The latter formulation accounts for the finite size of the nonlinear Fracture Process Zone (FPZ) in thermoset nanocomposites. Data re-analyzed from [8] and [18]. 22 0.20.30.40.50.60.70.80.901234567Fracture Energy, [N/mm]652.812Silica Nanoparticle Size 12 [nm]28%[Unit: mm]SELLEFM00.10.20.30.40.50.60.70.8012345678910652.812Silica Nanoparticle Size 20 [nm]22.4%[Unit: mm]LEFMSEL0.20.30.40.50.6024681012Nanoparticles [wt%]652.812Silica Nanoparticle Size 40 [nm]19.5%[Unit: mm]SELLEFMZamanianet al. Zamanianet al. Zamanianet al. 0.10.30.50.70.91.11.31.51.700.20.40.60.811.2Graphene Oxide1015020 [Unit: mm]SELLEFM21.4%0.00.51.01.52.02.53.03.500.511.522.533.551.8%LEFMSEL1015020 [Unit: mm]0.00.51.01.52.02.500.511.522.533.536.5%LEFMSEL1015020 [Unit: mm]Jiang et al. Jiang et al. Silica Nanoparticle Attached Graphene OxideJiang et al. APTES Functionalized SiO2c)d)e)f)a)b)Figure 6: Mode I fracture energy estimated by Linear Elastic Fracture Mechanics (LEFM) and Size Effect Law (SEL), Eq. (14). The latter formulation accounts for the finite size of the nonlinear Fracture Process Zone (FPZ) in thermoset nanocomposites. Data re-analyzed from [19] and [10]. 23 0.00.10.20.30.400.20.40.6Fracture Energy, [N/mm]8.3%LEFMSEL4408 [Unit: mm]0.00.10.20.30.400.40.81.21.622.48.8%LEFMSEL4408 [Unit: mm]0.00.10.20.300.20.40.64.9%LEFMSEL4408 [Unit: mm]Chandrasekaran et al. Chandrasekaran et al. Chandrasekaran et al. 00.20.40.60.8100.10.20.30.40.53466[Unit: mm]SELLEFM00.10.20.30.40.50.60.700.10.20.30.40.5346615.2%[Unit: mm]SELLEFM0.00.20.40.60.81.01.200.10.20.30.40.5346620.3%[Unit: mm]SELLEFMKonnolaet al. Konnolaet al. Konnolaet al. Thermally Reduced Graphene OxideGraphite NanoplateletsMulti Wall Carbon NanotubeFunctionalized (PES) Multi Wall Carbon NanotubeMulti Wall Carbon NanotubeFunctionalized (COOH) Multi Wall Carbon NanotubeNanoparticles [wt%]16%c)e)d)a)f)b)Figure 7: Mode I fracture energy estimated by Linear Elastic Fracture Mechanics (LEFM) and Size Effect Law (SEL), Eq. (14). The latter formulation accounts for the finite size of the nonlinear Fracture Process Zone (FPZ) in thermoset nanocomposites. Data re-analyzed from [20], [21] and [22]. 24 0.00.10.20.302468101200.20.40.60.811.21.405101520253035404550556.3675.612.7Silica Size 23 [nm]21.3%Fracture Energy, [N/mm][Unit: mm]SELLEFM00.20.40.60.811.21.41.61.805101520253035404550556.3675.612.7Silica Size 74 [nm]34.3%[Unit: mm]SELLEFM00.20.40.60.811.21.41.60510152025303540455055Nanoparticles [wt%]6.3675.612.7Silica size 170 [nm]24.3%[Unit: mm]LEFMSELLEFMSEL3.17555.8812.78 [Unit: mm]Dittanetet al. Dittanetet al. Varziriet al. 11%Silica Nanoparticles0.00.10.20.30.40.50.60123411.2%LEFMSEL652.812 [Unit: mm]0.00.10.20.30.40.50.6012347.3%LEFMSEL652.812 [Unit: mm]Kim et al. Kim et al. NanoclayCarbon Black Dittanetet al. d)c)e)a)f)b)Figure 8: Mode I fracture energy estimated by Linear Elastic Fracture Mechanics (LEFM) and Size Effect Law (SEL), Eq. (14). The latter formulation accounts for the finite size of the nonlinear Fracture Process Zone (FPZ) in thermoset nanocomposites. Data re-analyzed from [23]. 25 0.10.30.50.70.91.11.31.50510152025Fracture Energy, [N/mm]3630 14.2%Only Silica6LEFMSEL0.01.02.03.04.05.06.07.08.09.005101520253630 156.8%Only Rubber6LEFMSEL0.00.51.01.52.02.53.00246810123630 Silica Nanoparticle [wt%]27.7%6 % Rubber+ Silica6LEFMSEL0.01.02.03.04.05.06.00246810123630 Rubber Nanoparticle [wt%]103.9%10 % Silica+ Rubber6LEFMSELLiu et al. Liu et al. Liu et al. Liu et al. [Unit: mm][Unit: mm][Unit: mm][Unit: mm]c)d)a)b)Figure 9: (a) Schematic geometry with cohesive crack modeling; (b) Cohesive crack modeling and stress profile in front of crack tip. 26 𝑓𝑡0Cohesive Crack Law(b)𝜎𝑦𝑦[Mpa]Position [mm]01202040Cohesive ElementsDisplacement, u(a)46.67-75.47-55.11-34.76-14.405.9626.32𝜎𝑦𝑦[Mpa]𝛿𝑓𝐺𝑓Figure 10: Load-displacement curves obtained by using a linear cohesive crack law with both LEFM and corrected Gf on the re-analysis of data from [17]. 27 02040608010012000.511.502040608010012000.511.5Experimental DataCarolan et al.CSR+25%DiluentExperimental DataCarolan et al.CSR+25%Diluent+4%Silica02040608010012014000.511.5Carolan et al.CSR+25%Diluent+8%SilicaExperimental Data02040608010000.511.5Carolan et al.CSRLCL (𝐺𝑓,𝑆𝐸𝐿)Experimental DataDisplacement [mm]Load [N]02040608000.20.40.60.8Experimental DataCarolan et al.Silica020406000.20.40.60.8Experimental DataCarolan et al.Silica+25%DiluentLCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)Figure 11: Load-displacement curves obtained by using a linear cohesive crack law with both LEFM and corrected Gf on the re-analysis of data from [8], [18]. 28 030609012000.51Experimental DataZamanianet al.Silica Nanoparticle Size [12nm]LCL (𝐺𝑓,𝑆𝐸𝐿)02040608010000.51Zamanianet al.Silica Nanoparticle Size [20nm]Experimental Data02040608010000.51Experimental DataZamanianet al.Silica Nanoparticle Size [40nm]0100200300400012Experimental DataJiang et al.Silica Nanoparticle Attached Graphene Oxide05010015020025030000.511.52Experimental DataJiang et al.Graphene Oxide050100150200250300350012Experimental DataJiang et al.APTES Functionalized SiO2Displacement [mm]Load [N]LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)Figure 12: Load-displacement curves obtained by using a linear cohesive crack law with both LEFM and corrected Gf on the re-analysis of data from [19 -- 22]. 29 01020304000.20.4Experimental DataChandrasekaranet al.Graphene NanoplateletsLCL (𝐺𝑓,𝐿𝐸𝐹𝑀or 𝐺𝑓,𝑆𝐸𝐿)Overlapped02040608000.20.40.6Kim et al.NanoclayExperimental Data010203000.51Experimental DataVaziriet al.Silica Nanoparticle05010015020000.51Dittanetet al.Silica Size 23 [nm]Experimental Data05010015020000.51Dittanetet al.Silica Size 74 [nm]Experimental Data05010015020000.51Dittanetet al.Silica Size 170 [nm]Experimental DataDisplacement [mm]Load [N]LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)Figure 13: Load-displacement curves obtained by using a linear cohesive crack law with both LEFM and corrected Gf on the re-analysis of data from [10], [23]. 30 01020304000.20.4Experimental DataKonnolaet al.Multi-wall Carbon Nanotube0102030405000.20.40.60.8Experimental DataKonnolaet al.PES Multi-wall Carbon Nanotube020040060000.30.6Liu et al.SilicaExperimental Data0200400600800012Experimental DataLiu et al.Rubber020040060000.51Liu et al.6 % Rubber+ SilicaLCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)020040060080000.511.52Liu et al.10 % Silica + RubberLCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)Experimental DataExperimental DataDisplacement [mm]Load [N]LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)LCL (𝐺𝑓,𝑆𝐸𝐿)LCL (𝐺𝑓,𝐿𝐸𝐹𝑀)Figure 14: Comparison between LCL results and experimental data from [17]. 31 012345601230123456012301234560123Carolan et al.CSRCSEC (𝐺𝑓,𝑆𝐸𝐿)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)0123456012Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Carolan et al.CSR + 25% DiluentCarolan et al.CSR + 25% Diluent + 4% SilicaCarolan et al.CSR + 25% Diluent + 8% Silica0123456024Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Carolan et al.Silica012345670246Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Carolan et al.Silica + 25% Diluent𝑓𝑡2𝑔′(𝛼𝑜)𝜎𝑁𝑐2−1𝐷𝑔(𝛼𝑜)𝑙𝑐ℎ𝑔′(𝛼𝑜)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)Figure 15: Comparison between LCL results and experimental data from [8], [17], [18]. 32 012345601234CSEC (𝐺𝑓.𝐿𝐸𝐹𝑀)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓.𝐿𝐸𝐹𝑀)Zemanianet al.Silica Size 12 [nm]012345601234CSEC (𝐺𝑓.𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓.𝐿𝐸𝐹𝑀)Zemanianet al.Silica Size 20 [nm]012345601234Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓.𝐿𝐸𝐹𝑀)Zemanianet al.Silica Size 40 [nm]0123456780246Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓.𝐿𝐸𝐹𝑀)Jianget al.Graphene Oxide0123450123Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓.𝐿𝐸𝐹𝑀)Jiang et al.Silica Nanoparticle Attached Graphene Oxide0123456024Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓.𝐿𝐸𝐹𝑀)Jianget al.APTES Functionalized SiO2𝑓𝑡2𝑔′(𝛼𝑜)𝜎𝑁𝑐2−1𝐷𝑔(𝛼𝑜)𝑙𝑐ℎ𝑔′(𝛼𝑜)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓.𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓.𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓.𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓.𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)Figure 16: Comparison between LCL results and experimental data from [19 -- 22]. 33 0246810120369Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Chandrasekeranet al.Graphene Nanoplatelet02468100246Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Kim et al.Nanoclay0123456024Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Vaziriet al.Silica Nanoparticle012345601234Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Dittanetet al.Silica Size 23 [nm]01234560123Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Dittanetet al.Silica Size 74 [nm]012345601234Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Dittanetet al.Silica Size 170 [nm]𝑓𝑡2𝑔′(𝛼𝑜)𝜎𝑁𝑐2−1𝐷𝑔(𝛼𝑜)𝑙𝑐ℎ𝑔′(𝛼𝑜)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)Figure 17: Comparison between LCL results and experimental data from [10], [23]. 34 012345678024CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Konnolaet al.Multi-wall Carbon Nanotube02468024CSEC (Corrected 𝐺𝑓)CSEC (𝐺𝑓,𝑆𝐸𝐿)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Konnolaet al.PES Multi-wall Carbon Nanotube0123400.511.501234500.511.502468024CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Liu et al.Silica𝑓𝑡2𝑔′(𝛼𝑜)𝜎𝑁𝑐2−1𝐷𝑔(𝛼𝑜)𝑙𝑐ℎ𝑔′(𝛼𝑜)0123450123Experimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Liu et al.6% Rubber + SilicaExperimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)Liu et al.10% Silica + RubberLiu et al.RubberExperimental DataSimulation Data (𝐺𝑓,𝑆𝐸𝐿)Simulation Data (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)CSEC (𝐺𝑓,𝐿𝐸𝐹𝑀)CSEC (𝐺𝑓,𝑆𝐸𝐿)Figure 18: Cohesive laws used in this study: (a) Linear Cohesive Law (LCL); (b) Bi-linear Cohesive Law (BCL). 35 𝑓𝑡0𝜎𝛿𝛿𝑓𝑓𝑡0𝜎𝛿𝛿𝑓𝜎𝑘𝐺𝑓𝐺𝐹Linear Cohesive LawBi-linear Cohesive Law𝐺𝑓=𝐺𝐹(b)(a)Figure 19: Load-displacement curves vs. cohesive zone model featuring a Linear Cohesive Law (LCL) for different graphene contents and specimen sizes. 36 010020030040050060070080000.20.40.60.811.6 wt% Graphene ConcentrationD=10 mmD=20 mmD=40 mm010020030040050060000.20.40.60.80.9 wt% Graphene ConcentrationD=10 mmD=20 mmD=40 mm010020030040050060000.20.40.60.80.3 wt% Graphene ConcentrationD=10 mmD=20 mmD=40 mm010020030040050000.20.40.60.8D=10 mmD=20 mmD=40 mmPure EpoxyDisplacement, u[mm]Load, P[N]DDDDLinear Cohesive LawExperimental ResultLinear Cohesive LawExperimental ResultLinear Cohesive LawExperimental ResultLinear Cohesive LawExperimental ResultFigure 20: Load-displacement curves vs. cohesive zone model featuring a Bi-linear Cohesive Law (BCL) for different graphene contents and specimen sizes. 37 010020030040050000.20.40.60.8DPure Epoxy010020030040050060000.20.40.60.8D0.3 wt% Graphene Concentration010020030040050000.20.40.60.8D0.9 wt% Graphene Concentration010020030040050060070080000.20.40.60.81D1.6 wt% Graphene ConcentrationDisplacement, u[mm]Load, P[N]D=10 mmD=20 mmD=40 mmD=10 mmD=20 mmD=40 mmD=10 mmD=20 mmD=40 mmD=10 mmD=20 mmD=40 mmBi-linear Cohesive LawExperimental ResultBi-linear Cohesive LawExperimental ResultBi-linear Cohesive LawExperimental ResultBi-linear Cohesive LawExperimental ResultFigure 21: Calibrated bi-linear cohesive law vs. linear cohesive law with the same total fracture energy. 38 01530456000.030.060.09Pure Epoxy𝐺F=0.78 N/mm𝐺initial =0.62 N/mm01530456000.030.060.090.3 wt% Graphene Concentration𝐺F=0.81 N/mm01530456000.030.060.090.12𝐺F= 0.95 N/mm01530456000.030.060.090.120.9 wt% Graphene Concentration𝐺F=1.42 N/mm1.6 wt% Graphene ConcentrationCrack Opening, 𝛿[mm]Stress, 𝜎[MPa]𝐺initial =0.6 N/mm𝐺initial =0.67 N/mm𝐺initial =0.69 N/mmLinear Cohesive LawBi-linear Cohesive LawLinear Cohesive LawBi-linear Cohesive LawLinear Cohesive LawBi-linear Cohesive LawLinear Cohesive LawBi-linear Cohesive LawFigure 22: Comparison between initial and total fracture energy of the calibrated bi-linear cohesive law. 39 00.40.81.21.6200.511.52Total Fracture EnergyInitial Fracture EnergyGraphene Content [%]Fracture Energy [N/mm]Figure 23: Schematic representation of a micro-crack propagating in a composite: (a) and (b) Cohe- sive crack formation; (c) cohesive stresses bridging the crack faces; (d) distribution of crack opening displacement and (e) corresponding stresses and displacements in the cohesive law. Note that, for a micro-crack, the cohesive stresses do not enter the second arm of the bi-linear cohesive law (σmin > σk) 40 MatrixFiber𝑓𝑡𝜎(𝑥)𝛿(𝑥)𝜎0𝛿0𝛿𝑓𝜎0𝑓𝑡Cohesive Stresses𝜎(𝑥)𝛿(𝑥)Crack𝛿0𝑥Opening displacementsBi-Linear Cohesive Law𝜎𝑘
1905.04639	1	1905	2019-05-12T03:11:16	Waveguide-integrated, plasmonic enhanced graphene photodetectors	[ "physics.app-ph", "cond-mat.mes-hall" ]	We present a micrometer scale, on-chip integrated, plasmonic enhanced graphene photodetector (GPD) for telecom wavelengths operating at zero dark current. The GPD is designed and optimized to directly generate a photovoltage and has an external responsivity~12.2V/W with a 3dB bandwidth~42GHz. We utilize Au split-gates with a$\sim$100nm gap to electrostatically create a p-n-junction and simultaneously guide a surface plasmon polariton gap-mode. This increases light-graphene interaction and optical absorption and results in an increased electronic temperature and steeper temperature gradient across the GPD channel. This paves the way to compact, on-chip integrated, power-efficient graphene based photodetectors for receivers in tele and datacom modules	physics.app-ph	physics	Waveguide-integrated, plasmonic enhanced graphene photodetectors J. E. Muench1, A. Ruocco1, M. A. Giambra2, V. Miseikis2,3,4, D. Zhang1, J. Wang1, H. F. Y. Watson1, G. C. Park1, S. Akhavan1, V. Sorianello2, M. Midrio5, A. Tomadin6, C. Coletti3,4, M. Romagnoli2, A. C. Ferrari1∗ and I. Goykhman7 1Cambridge Graphene Centre, University of Cambridge, Cambridge CB3 0FA, UK 2Consorzio Nazionale per le Telecomunicazioni, Photonic Networks and Technologies National Laboratory, 56124 Pisa, Italy 3Center for Nanotechnology Innovation @ NEST, Istituto Italiano di Tecnologia, 56127 Pisa, Italy 4Graphene Labs, Istituto Italiano di Tecnologia, 16163 Genova, Italy 5Consorzio Nazionale per le Telecomunicazioni, University of Udine, 33100 Udine, Italy 6Dipartimento di Fisica, Universit`a di Pisa, Largo Bruno Pontecorvo 3, 56127 Pisa, Italy and 7Micro- and Nanoelectronics Research Center, Technion, Haifa 320000, Israel We present a micrometer scale, on-chip integrated, plasmonic enhanced graphene photodetec- tor (GPD) for telecom wavelengths operating at zero dark current. The GPD is designed and optimized to directly generate a photovoltage and has an external responsivity∼12.2V/W with a 3dB bandwidth∼42GHz. We utilize Au split-gates with a∼100nm gap to electrostatically create a p-n-junction and simultaneously guide a surface plasmon polariton gap-mode. This increases light-graphene interaction and optical absorption and results in an increased electronic temperature and steeper temperature gradient across the GPD channel. This paves the way to compact, on-chip integrated, power-efficient graphene based photodetectors for receivers in tele and datacom modules. The ever-growing demand for global data traffic[1] is driving the development of next generation communica- tion standards[2, 3]. The increasing numbers of con- nected devices[4], the need for new functionalities, and the development of high-performance computing[5, 6] require optical communication systems performing at higher speeds, with improved energy-efficiency, whilst maintaining scalability and cost-effective manufacturing. Si photonics[7 -- 9] offers the prospect of dense (nanoscale) integration[10] relying on mature, low-cost (based on complementary metal-oxide-semiconductor (CMOS) fab- rication processes) manufacturing[8, 9], making it one of the key technologies for short-reach (<10km) opti- cal interconnects[11] beyond currently employed lithium niobate[12] and indium phosphate[13]. A variety of functionalities have been developed and demonstrated in Si photonics for local optical interconnects[11]. Electro-optic modulators based on carrier-depletion (phase-modulation) in Si[14, 15] or the Franz-Keldysh effect[16] (amplitude-modulation) in strained Si-Ge[17, 18] encode information into optical sig- nals at telecom wavelengths (λ =1.3-1.6µm). On the re- ceiver side, Ge[19] or bonded III-V[20, 21] photodetectors (PD) are needed for optical-to-electrical signal conver- sion, since the telecom photon energies are not sufficient for direct (band-to-band) photodetection in Si[22]. On-chip integrated Ge PDs[23 -- 27] are standard com- ponents in Si photonics foundries[8, 9, 22]. Their ex- ternal responsivities (in A/W), RI = Iph/Pin, where Iph is the photocurrent and Pin is the incident optical power, can exceed 1A/W[8, 23] and their bandwidth can reach 60GHz[25 -- 27]. Following the development of high temperature (> 600◦C)[19] heterogeneous inte- gration of Ge-on-Si using epitaxial growth and cyclic thermal annealing[19, 28, 29], the concentration of de- fects and threading dislocations in Ge epilayers and at Si/Ge interfaces can be reduced[19], resulting in low (<10nA[9, 27]) dark current in waveguide integrated Ge p-i-n photodiodes[24, 27]. However, Ge-on-Si integration is a complex process[19, 22, 29], as the lattice mismatch between Si and Ge[19], ion implantation[23, 25], ther- mal budget (i.e. thermal energy transfer to the wafer) management[22], and the non-planarity of Ge layers[29] require dedicated solutions during device fabrication[9]. The charge carrier mobility µ in Si and the disloca- tions and defects in grown[19] or evaporated[30] Ge lay- ers set intrinsic limitations that prevent further improve- ments to the operation speed of Ge PDs without com- promising RI [9, 26]. These shortcomings, together with the spectrally limited operation regime (band edge in Ge∼ 1.57µm[22], which can be extended to∼ 1.62µm[31] at the expense of RI ), and the incompatibility of Ge epi- taxy for monolithic integration with other material plat- forms, such as SiN, are amongst the main limitations for Ge PDs[8]. Thus, novel solutions for PDs, integrated with Si photonics, at telecom bands are needed. Graphene is a promising candidate for on-chip inte- grated photonics[32 -- 52]. The advantages of single-layer graphene (SLG) for photonics stem from its superior optoelectronic properties[53]. These include high-speed (>200GHz[54]) operation[55], broadband (ultraviolet to far-infrared) absorption[56 -- 58], efficient optical modu- lation (electro-optical index change ∆neff > 10−3)[32 -- 38], CMOS compatibility[41, 59] and integrability[32, 60, 61] with different on-chip photonics platforms, such as silicon-on-insulator (SOI)[33] and SiN[36]. In the case of waveguide-integrated graphene PDs (GPDs)[40 -- 52], high speeds up to 128GHz[49], wafer-scale integration[48] and RI ∼0.4-0.5A/W[43, 47, 50, 51] were reported. GPDs can offer broadband detection across multi- ple telecommunication channels (O-band∼1.31µm to U- band∼1.65µm)[41], bias-free operation[62], and direct generation of photovoltage[45, 62]. The latter opens up the possibility of building GPDs without the noise contri- bution of dark current[32, 46] and eliminates the need of noise-prone trans-impedance amplifier (TIA) to convert current-to-voltage in the read-out electronics[32]. 63, 64], can be built photo-voltaic exploiting different mech- GPDs (PV)[55, photo- anisms: thermoelectric (PTE)[64 -- 66], photo-gating[67], plasma- wave assisted[68] and photo-bolometric (PB)[69, 70]. The dominating effect for a given GPD depends on device configuration, design geometry, and way of operation[64, 71]. For telecom applications, where high-speed (tens GHz) operation is one of the key requirements[8, 32], PV, PTE and PB are typically considered for waveguide- integrated GPDs[32], taking advantage of the ultra-fast (∼fs-ps) carrier dynamics in SLG[72, 73]. PTE is ideal for PD operation in a voltage mode. In optically illuminated SLG, electron-electron scattering drives the formation of a 'hot' (optically excited)-carrier distribution, described by the Fermi-Dirac function[74], within <50fs[72]. This can remain at elevated tempera- tures Te, well above the lattice temperature Tl, over∼2- 4ps time scales[72], before reaching thermal equilibrium via phonons interaction[73, 75, 76]. In this hyperther- mal state, a photovoltage Vph is generated by a thermo- electric current as for the Seebeck effect[66], if both a temperature and chemical potential gradient are present in the SLG channel. The sign and magnitude of Vph de- pend on the Seebeck coefficient (S), i.e. the proportion- ality constant between temperature change and induced photovoltage[77], and Te profile in the SLG channel[66]: Vph = Z S(x) · ∇Te(x) dx (1) where x is the coordinate along the channel from drain to source, and S is given by Mott's formula[64 -- 66, 77]: S(x) = − π2k2 BTe 3e 1 σ(x) dσ(x) dµc (2) with σ(x) the conductivity, kB the Boltzmann constant, e the electron charge and µc the chemical potential (µc = EF at Te = 0[74], with EF the Fermi energy). have been in reported PTE-GPDs vertically- illuminated[66, 78 -- 81] and waveguide-integrated[45 -- 47]configurations. The latter used SLG flakes prepared by micromechanical cleavage (MC) of graphite[82], with typical device length of tens of µm[45 -- 47]. They have external voltage responsivities, defined as RV = Vph/Pin, up to∼4.7V/W[46] (at zero bias) with speeds up to 65GHz[45]. Depending on PTE-GPD design configu- ration and the requirements of the read-out electronics 2 (i.e. output photo-signal to be measured as current or voltage), the responsivity can be characterized in terms of RI or RV . The photovoltage generated by the Seebeck effect is associated with a thermoelectric current across the PD by a Ohmic relation[45, 46, 62, 79] Iph = Vph/R, with R the resistance. To increase RV for PTE-GPDs, Eq.1 suggests two strategies: 1) maximize S; 2) maximize the Te gra- dient profile in the SLG channel. The former is re- lated to µ via Eq.2 and the Drude conductivity[77], σ = eµn, where n is the charge carrier concentra- tion. Thus, S can be improved by using high-mobility SLG, e.g. encapsulating SLG in hBN[83 -- 85], using single-crystals[85, 86], or large (tens µm) domain-size[87], in combination with a transfer processes that avoid contamination[84, 88], strain[88], and cracks[89]. Ref.[32] suggested that µ > 104cm2V−1s−1 could enable RV > 100V/W. The Te gradient can be increased by creat- ing a spatially confined, localized, heat source[45] gen- erated by enhanced optical absorption in SLG over com- pact (< 10µm) device lengths[50, 51]. This could be achieved by integrating plasmonic nanostructures[90 -- 95]. Sub-wavelength plasmonic confinement and asso- ciated enhancement of near-field light-matter interac- tion were used to boost RI in Si-plasmonic PDs[90, 91], plasmonic-Ge PDs[92], plasmonic decorated GPDs for free-space[93 -- 95] and waveguide-integrated[43, 50 -- 52, 96] configurations. Refs.[50 -- 52] reported plasmonic en- hanced on-chip GPDs based on PV[50, 52] and PB[51, 52] with RI ∼ 0.5A/W and bandwidth∼ 110GHz at 1.55µm for source-drain bias<1V. Here, we report compact (∼0.5-4µm), PTE-based, waveguide-integrated, plasmonic-enhanced GPDs for telecom wavelengths with RV ∼ 12.2V/W at zero source- drain bias and zero dark current, with a 3dB cutoff frequency∼ 42GHz. To the best of our knowledge, this is the largest RV to date for waveguide-integrated GPDs operating in voltage mode. We use SLG grown by chemi- cal vapor deposition (CVD) and transferred onto low-loss (∼ 1dB/cm) planarized (i.e. fully-embedded in polished cladding[7]) SiN waveguides with a semi-dry (i.e. com- bining wet de-lamination from the growth substrate with dry lamination onto the target substrate) transfer[86], unlike previous PTE GPDs exploiting non-scalable MC SLG[45, 46]. Our design relies on Au split-gates to elec- trostatically create a p-n junction in the SLG channel, as well as to guide a confined SPP waveguide mode. By leveraging optical field enhancement and plasmonic confinement in the gap, we increase light-SLG interac- tion and optical absorption in the p-n junction region, resulting in a confined electrons heat source, compact device length, and increased RV . This combines high- performance (large RV , high-speed, bias-free, compact, direct Vph read-out) PTE GPDs in the telecom range with scalable fabrication, paving the way for graphene integrated receivers for next-generation transceivers. (a) (b) Al2O3 Ni/Au Cr/Au wgap Source Gate 2 SLG Si3N4 Si Gate 1 Drain (c) Al2O3 SLG channel SiN Waveguide Au Au SLG Si3N4 3 tgate tox SiO2 Air SiO2 FIG. 1. a) Schematic view of our GPD: SLG on SiN waveguide (brown) with split-gates, acting as plasmonic slot waveguide, to create a p-n junction in the channel (as depicted by the Dirac cones above the gates). The green arrow indicates the light propagation direction. b) Schematic cross-section of the GPD active region. c) Simulated electric field (Ex, in-plane) distribution of the fundamental SPP waveguide mode. For clarity, only the field component parallel to the SLG channel is shown. The vertical and horizontal scale bars are 100 and 250nm The design of our GPD is schematically shown in Fig.1a,b. It comprises a SLG channel on a SiN waveg- uide supporting a transverse-electric (TE, in-plane) po- larized fundamental waveguide mode. Two Au gates are placed above the channel, separated from the SLG by an Al2O3 dielectric spacer and centrally aligned with re- spect to the waveguide. When this split-gate structure is DC (direct current) biased, it forms a p-n junction, Fig.1a, and creates a S profile in the SLG channel, as for Eq.2. When an on-chip guided light signal reaches the PD area, it is evanescently coupled from the SiN waveguide to the split-gate, which acts as SPP waveg- uide, Fig.1c. The plasmonic guiding with light confine- ment in the gap (width wgap ∼100nm) leads to enhanced optical absorption and a localized hot electron distribu- tion with a Te gradient in the p-n junction (gap) region. The coupling efficiency, Pout/Pin, where Pout is the power transferred between two optical components, from pho- tonic to plasmonic waveguide mode can be optimized by tailoring wgap and dielectric spacer thickness (tox). To optimize the cross section parameters at λ=1.55µm, we perform optical simulations using a commercial finite difference solver tool (Lumerical MODE). After selecting the fundamental gap plasmon mode for a given design and λ, we extract the optical electric field distribution in the SLG channel to model the absorbed power den- sity that generates the hot carrier distribution as time- averaged electric power dissipation density[97, 98], which we refer to as Joule heat source (J) hereafter. After nor- malization to an input power of 1µW, this is used in the heat equation[47, 65, 78]: − κe(x)(cid:20) d2 dx2 ∆Te(x) − 1 ξ2 ∆Te(x)(cid:21) = J (x) (3) where ∆Te(x) = Te(x) − Tl is the local temperature fluc- tuation, ξ is the cooling length (see Methods) and κe(x) is the electronic thermal conductivity (see Methods). Eq.3 gives the Te(x) profile along the SLG channel. The S(x) profile from Eq.2 is used in Eq.1 to obtain Vph. The device parameters are chosen to maximize Vph. A second aspect of device design concerns the coupling between the dielectric and plasmonic waveguides, as well as the positioning and width of the SLG channel along the split-gate. Taper-assisted butt-coupling (end-to-end alignment) was reported to yield the lowest insertion loss (< 0.6dB)[99] for the transition from optical to plasmonic modes. However, since evanescent coupling (lateral or vertical alignment) provides simpler fabrication[100] and greater flexibility for the placement of devices on top of integrated optical circuits[101], we use this here, Fig.1a- c. To obtain the largest RV , the electric field distribution along the propagation directions needs to be considered. Light absorption in SLG or in the plasmonic structure along the device leads to an exponential decay of optical power[60]. Thus, the increase in Te follows the same de- cay. The resulting photovoltage drop at different points along the device results in an averaged potential differ- ence between source and drain contacts. To optimize RV , a compact (<10µm) device with optimized peak absorp- tion and minimal drop-off is preferable. We thus per- (a) Si3N4 (b) SLG (c) Ni/Au 4 SiO2 (d) Cr/Au (e) (f) Al2O3 Al2O3 FIG. 2. a) Planarized SiN waveguide. b) SLG transfer and shaping by oxygen plasma etch. c) Ni/Au contacts to SLG channel. d) Al2O3 gate oxide deposition. e) Cr/Au evaporation of split-gate structure. f) Al2O3 encapsulation and contact pads opening FIG. 3. a) Optical image of a GPD. Scale bar: 20µm. b) Scanning electron micrograph of split-gate. False colors: brown, Ni/Au contacts; yellow, Cr/Au gates; green, planarised SiN waveguide; white dashed line, SLG channel. Scale bar: 2µm form finite-difference time domain (FDTD) simulations in Lumerical FDTD (see Methods). The co-existence of plasmonic and dielectric waveguide leads to oscillating power exchange between both structures[100 -- 102]. For the highest coupling efficiency, the vertical distance be- tween these waveguides is typically> 150nm[100, 101], exploiting interference between quasi-even and quasi-odd eigenmodes[100 -- 102]. In our design, we keep this sepa- ration small (tens nm) to ensure overlap between SLG and gap plasmon mode, to avoid a long (> 5µm) cou- pling length[100], and to create a sharper concentration of power close to the front of the plasmonic structure. The SLG channel is placed accordingly, after a short (∼1µm) taper at the front of the SPP waveguide to re- duce mode mismatch. Fig.2 summarizes the fabrication process of our GPDs. Planarized SiN waveguides, Fig.2a, (260nm high, width 0.8-1.5µm) on 15µm SiO2 are fabricated as follows. The SiN layer is first deposited by low-pressure (LP) CVD. The SiN photonic waveguides are then defined by elec- tron beam lithography (EBL) and reactive ion etch- ing. For surface planarization, a 1.6 µm thick boron- phosphorus tetraethyl orthosilicate (BPTEOS) layer is deposited as top cladding and subsequently etched to a final thickness∼ 20nm on top of the SiN waveguides, avoiding chemical mechanical polishing. SLG is grown on pre-patterned, electropolished Cu with Cr nucleation sites as for Ref.[86]. After an initial annealing in argon (10mins), SLG growth is initiated at 25mbar with argon, hydrogen, and methane flowing at 900, 100, and 1 stan- dard cubic centimeters per minute (sccm), respectively. After growth, SLG single crystals are placed onto the nm G On Cu 2D After transfer After fabrication . ) . u a ( y t i s n e n t I 1400 1600 2600 Raman Shift (cm -1) 2800 FIG. 4. Raman spectra at 514.5 nm for SLG as grown on Cu (black), after transfer onto the SiN waveguide (red), and after device fabrication (blue). All spectra are normalised to the intensity of the G peak, I(G), and are shown after subtraction of the substrate signals. photonic chips by semi-dry transfer[86], comprising the spin-coating of a poly(methyl methacrylate) (PMMA) support layer, the attachment of a Kapton frame for handling, electrochemical delamination of SLG in sodium hydroxide, and the lamination onto the target substrate with the help of a micro-manipulator to align the crys- tals with the photonic structures. A PMMA etch mask is then used to shape the device channel and remove excess SLG over grating coupler and waveguides, defined using EBL. This is followed by oxygen plasma etching at 3W, Fig.2b. Next, contacts are defined by another EBL step. Metallization (15nm Ni/40nm Au) is done by sputter- ing, thermal evaporation and lift-off in acetone, Fig.2c. 30nm Al2O3 is used as gate oxide, via atomic layer de- position (ALD), Fig.2d. An additional EBL step and electron beam evaporation are used to fabricate the plas- monic split gates, Fig.2e. To encapsulate the device and prevent air breakdown in the gap between gate contacts when∼10V is applied, we deposit another 40nm Al2O3 by ALD. A laser writer is used to define an etch mask to open access to all contacts, Fig.2f. 5 The quality of SLG is monitored by Raman spec- troscopy at all critical points during the fabrication process, using a Renishaw InVia equipped with a 50x objective (numerical aperture NA=0.75) at 514.5nm with power below 0.5mW to exclude heating ef- fects and risk of damage. Representative spectra of SLG on Cu (after removal of Cu background photoluminescence[103]), after transfer onto the waveg- uide, and after complete device fabrication, are shown in Fig.4. The absence of a D peak confirms negligible defects are introduced during fabrication. The 2D peaks are single-Lorentzian, confirming the presence of SLG[104, 105]. On Cu, the position and full width at half-maximum of the G peak are Pos(G)∼1595cm−1 and FWHM(G)∼ 8cm−1. The position of the 2D peak, is∼2721cm−1 with FWHM(2D)∼27cm−1. Pos(2D), The 2D to G peak intensity, and area, A(2D)/A(G), ratios are∼ 1 and∼ 3.2. After transfer, Pos(G)∼ 1590cm−1, FWHM(G)∼10cm−1, FWHM(2D)∼28cm−1, Pos(2D)∼ I(2D)/I(G) ∼ 3.2 and A(2D)/A(G) ∼ 8.6. This corresponds to∼ 250meV doping[106, 107] and a carrier concentration∼4×1012cm−2. After the final encap- FWHM(G)∼9cm−1, sulation, Pos(2D)∼2689cm−1, FWHM(2D)∼30cm−1, I(2D)/I(G) ∼ 2.2 and A(2D)/A(G) ∼ 7.6, indicating∼350meV (n ∼7×1012cm−2) doping. Pos(G)∼1590cm−1, I(2D)/I(G), 2690cm−1, To determine the DC operating point, we perform elec- trical characterizations by sweeping the split-gate volt- ages (VGate 1, VGate 2) while measuring the device cur- rent IDS under a constant source-drain bias VDS=1mV, using DC probes on micromanipulators and two source measure units. To record the static photoresponse, we add two fibre probes and couple continuous wave (CW) transverse-electric (TE) polarized light at 1.50-1.58µm from a tunable laser into the SiN waveguide via an optical fibre and a grating coupler (GC). While Vph is recorded across the unbiased (VDS=0V) channel as a function of VGate 1 and VGate 2, using a lock-in amplifier under in- ternal modulation (square wave, ON-OFF) of the laser with 200Hz, we monitor the transmission with a second fibre positioned over the output GC and connected to an external InGaAs power meter to ensure constant Pin. Fig.5a plots the R map of a typical device as a func- tion of VGate 1, VGate 2. This shows a four-fold pattern, corresponding to the four doping constellations (p-n, n-p, n-n, p-p) in the SLG channel for different combinations of gate voltages. The map is symmetric with a maxi- mum R ∼9kΩ at the crossing of the charge neutrality point (CNP), between -4 and -5V. This corresponds to n- doping of the unbiased SLG channel with n∼7×1012cm−2 (∼350meV). R has contributions from channel (Rch) and contact (Rc) resistances. Rch includes a fixed contri- bution from ungated SLG regions and a gate-dependent contribution from channel segments underneath the split- 6 (a) ) V ( 2 e t a G V -2 -4 -6 -8 p-n n-n p-p n-p R(kΩ) (b) ) S m ( y t i v i t c u d n o C 1.6 1.2 0.8 0.4 9.5 8.0 6.5 5.0 3.5 2.0 -8 -6 -4 -2 -12 -9 -6 -3 0 VGate 1 (V) VGate (V) FIG. 5. (a) GPD channel resistance as function of split-gate voltages. (b) Conductivity vs. gate voltage from a 4-terminal measurement on test Hall bars gates. The gate-dependent variation in R in Fig.5a sug- gests Rch as the dominant factor. This is consistent with our contact resistivity (<1kΩµm) for CVD SLG and the calculated R based on channel geometry and sheet resis- tance obtained from independent four-terminal measure- ments on reference Hall bars. From these we also extract an average µ ∼2000cm2V−1s−1 from linear fits of the conductivity via[108] µ = dσ/dVGate/Cox, where Cox is the top gate capacitance. Fig.5b is a bi-directional gate sweep of the conductivity, indicating low hysteresis and charge-trapping in the Au/Al2O3/SLG gate capacitor. Fig.6a is a Vph map of a typical device at Pin ∼ 100µW inside the GPD. The plot exhibits a six-fold pattern with higher response for bipolar (p-n, n-p) junctions and a weaker one with sign-crossing along the diagonal (VGate 1 = VGate 2) for the unipolar (n-n, p-p) junctions. When the GPD is operated at zero VDS, this indicates a PTE-dominated photodetection as the two sign changes in Vph along a single-gate sweep line (e.g VGate 2 =const.) reflect the two sign changes of the S gradient across the junction, arising from the non-monotonic dependence of S on µc[65]. The measured photoresponse is in good agreement with the calculated one in Fig.6b. We ob- served a similar behavior on> 12 devices of different sizes across 5 chips, the shortest being 500nm in the light prop- agation direction for a footprint∼3µm2. For all devices, we got a maximum Vph close to the CNP where S is largest, with a gradual drop-off at higher doping. To calculate RV , we first estimate the optical power in- side our GPDs by taking into account: a) combined loss (∼9.6dB at peak transmission) of waveguide propagation and fiber-to-waveguide coupling (wavelength-dependent, following the response envelope of the GC); b) 3dB power reduction from the input laser modulation (square wave, ON-OFF) with a 50% duty cycle; c) 3dB power split- ting in the Y-branches and their∼0.2dB losses. We get RV ∼ 12.2V/W, the largest reported so far for waveguide-integrated GPDs in voltage mode at zero-bias. Fig.6c plots the RV wavelength dependence, showing a broadband (1.50-1.58µm) photoresponse covering the entire C-band (1.53-1.565µm[109]) and beyond. The er- ror bars indicate variations in the wavelength-dependent coupling loss (thus Pin), estimated as standard deviation from transmission measurements on>10 reference waveg- uides. We attribute the gradual increase in RV with increasing wavelengths to improved coupling efficiency from dielectric to plasmonic waveguide. Fig.6d is the Vph power dependence at 1.575µm for optical power levels comparable to those required by re- ceivers used in 100GBs−1 links[32]. The linear response indicates a power-independent RV . To highlight our GPDs' behavior as voltage sources, when a signal is generated, we place two devices back to back on the same waveguide and connect them in series. This modified design, Fig.7a, consists of two SLG chan- nels gated from the same split-gate/SPP waveguide. By connecting the drain pad of one channel to the source of the other through a metal lead crossing the waveguide behind the active region of the devices, Fig.7b, we mea- sure both GPDs individually, as well as combined. Fig.7c is a false color SEM of the active region of both detec- tors. Since each GPD is designed to maximally absorb over the device length, the power rapidly decays along the propagation direction after the first GPD. We thus Vph (mV) 1.0 0.50 0.0 -0.50 -1.0 (b) ) V ( 2 e t a G V 2 1 0 -1 -2 7 RV (norm.) 1.0 0.50 0.0 -0.50 -1.0 -8 -6 -4 -2 -2 -1 0 1 2 VGate 1 (V) VGate 1 (V) ( 1575 nm 1000 ) V µ ( e g a t l o v o t o h P 100 (a) -2 -4 -6 -8 ) V ( 2 e t a G V (c) ) W V / i ( y t i v s n o p s e R 15 12 9 6 3 1510 1520 1530 1540 1550 1560 1570 1580 Wavelength (nm) 0.01 0.1 Power at Device (mW) FIG. 6. a) Experimental photovoltage map for zero bias. b) Simulated responsivity. c) Wavelength dependence of responsivity. d) Power dependence of responsivity at 1575 nm. place the second device∼ 1µm from the first. Figs.7d,e plot the photovoltage maps for both GPDs at Pin ∼ 70µW. The GPD closer to the input GC (A) absorbs most of the light and has the six-fold pattern typical of PTE, Fig.7d. The photoresponse of the sec- ond GPD (B) is weaker, due to light absorption in SLG and metal, Fig.10d,e. A six-fold pattern is not ob- served, due to photocurrents generated in the junctions between gated and ungated sections at either end of the SLG channel. Figs.7f,g are photovoltage maps of the combination of both GPDs. The response in series is in Fig.7f, while the sum of the individual responses in Fig.7g. The two plots are good agreement, confirming that Vph,A+B = Vph,A + Vph,B. Thus, in order to increase RV in long (tens µm) PTE-GPDs with absorption only in the SLG channel, one could avoid the reduction in pho- tosignal due to the decrease in ∆Te along the device and instead add the voltages generated in different sections by subdividing the channel into several shorter devices and connecting them as cascaded GPDs. To minimize the length of metal leads for contacting and connecting individual devices, this configuration could comprise in- dividually gated devices with alternating p-n junctions to form a meandered structure. This would ideally be implemented with transparent gates, such as indium tin oxide, or a second SLG at a distance far enough from the channel, to avoid additional losses. To evaluate the frequency response we use the opti- cal heterodyne set-up in Fig.8a, combing optical signals at different frequencies. The channel is contacted with an RF probe in G-S configuration. The output of our tunable laser source is combined with that of a fixed- wavelength laser diode (Thorlabs SFL1550P) and the GPDs' response to the amplitude beating at the differ- ence frequency is monitored with an electrical spectrum analyzer (ESA, Agilent PSX N9030A). The two source 8 FIG. 7. a) Schematic of test structure to measure two GPDs in series. b) Optical image of contact pad short to connect both GPDs in series. Scale bar: 80µm. c) SEM image of active region of test structure. False colors: brown, Ni/Au contacts; yellow, Cr/Au split gate; green, planarized SiN waveguide. The SLG channels are indicated by white dashed lines. Scale bar: 5µm. d) Photovoltage map of device A at the start of the SPP waveguide. e) Photovoltage map of device B at the end of the SPP waveguide. f) Photovoltage map of devices A and B in series. g) Sum of the photovoltage maps for devices A and B outputs are combined in a 50:50 fibre coupler. We moni- tor the signal stability (i.e. output power and the position of the difference frequency) using a reference PD and an oscilloscope. Prior to coupling the combined signal into the SiN waveguide, we use an erbium-doped fibre ampli- fier (EDFA, Keyopsys CEFA-C-HG) to raise the optical power to 15dBm (∼30mW) to increase the output signal detected at the ESA. To overcome the signal reduction due to impedance mismatch between device R and the 50Ω of the measurement equipment, we use an additional low noise amplifier (LNA) between GPD and ESA. In or- der to distinguish between the frequency response of our GPDs and that of the LNA and the remaining measure- ment equipment, the response of the latter two to a low power (-80dBm) input from a 50GHz signal generator is recorded for calibration. Fig.8b plots the calibrated response (black squares) to the beating signal at different frequencies, while the split- gate is biased to set an operating point in the p-n junction regime resulting in the largest photoresponse under CW illumination in Fig.6. The response stays within 3dB of the low-frequency (1GHz) reference power until 40GHz, the limit of our measurement set-up. To determine the cut-off, we therefore modify the set- up (Fig.8c) to perform impulse response measurements, where the response to ultra-short (∼150fs) optical pulses is monitored with an oscilloscope. For excitation, we use the idler of an optical parametric oscillator, pumped by a Ti:Sa mode-locked laser at 1.55µm, attenuated in free- space prior to coupling into the optical fibre. Fig.8d is the measured impulse response at the same operating point as for the heterodyne measurements. We obtain a pulse duration, assuming a Gaussian pulse shape, ∆t ∼ 11ps. For Gaussian-shaped pulses, the time-bandwidth product is∼0.44[110]. From this we estimate a f3dB ∼ 0.44/∆t ∼ 40GHz. The fast Fourier transform of the pulse is in Fig.8b (red circles) after calibration. The trace is in good agreement with the heterodyne response and drops below -3dB at∼ 42GHz, showing high-speed operation on par with current Ge PDs, consistent with other reports of high-speed MC-SLG-based PTE GPDs[45 -- 47]. However, this bandwidth is the highest reported so far for PTE- based on-chip GPDs made from CVD SLG. In summary, we an enhanced GPDs with reported waveguide-integrated plasmonic external responsivity∼12.2V/W, a -3dB cut-off∼42GHz, and small (∼3-20µm2) device footprints, using CVD SLG on SiN. We exploited the integration of an SPP waveguide (a) (c) f 3dB ~ 42 GHz 9 0 A m p -10 l i t u d e ( d B ) -20 100 Heterodyne Impulse-response (FFT) 10 Frequency (GHz) (b) 0 -10 ) B d ( e s n o p s e R . m r o N -20 1 0.6 0.4 0.2 0.0 ) V m ( e d u t i l p m A 0 50 100 150 200 250 300 Time (ps) FIG. 8. a) Schematic heterodyne set-up. b) Frequency response at zero VDS from (black) direct heterodyne measurement and (red) fast Fourier transform of impulse response. c) Schematic impulse-response measurement set-up. d) Impulse-response for∼150fs optical pulses at zero VDS and p-n gate bias. with a SLG p-n junction to enhance light-SLG interac- tion and create a confined electron heat-source to obtain a strong, PTE-dominated photoresponse. This paves the way to power-efficient receivers for optoelectronic links. We acknowledge funding from EU Graphene Flagship, ERC Grant Hetero2D, EPSRC Grants EP/K01711X/1, EP/K017144/1, EP/N010345/1, and EP/L016087/1. METHODS Device design and modeling Plasmonic gap width, SLG placement, and thickness of metal and oxide structures are the key parameters to be optimized to achieve maximum photovoltage. To do so, we build a device model in the layout environment of a commercial eigenmode solver (Lumerical MODE Solu- tions). In order to model SLG in the optical solver and subsequent calculations consistently, we use a volumetric permittivity material model, in which SLG is described as cuboid with finite thickness t = 0.34nm and in-plane (εk) and out-of plane (ε⊥) permittivity are defined as independent tensor elements. To calculate the in-plane relative permittivity for SLG, we use[111]: ε(ω) = ε′(ω) + iε′′(ω) = εr + iσ(ω) ε0ωt (4) where ε′(ω) and ε′′(ω) are the real and imaginary part of the relative permittivity ε(ω), σ(ω) is the SLG op- tical conductivity, ω is the angular frequency, εr is the background relative permittivity (whose frequency- dependence is ignored in the small (λ =1.5-1.6µm) wave- length range under consideration), and ε0 is the per- mittivity of free space. σ(ω) is obtained by linear- response[112] in the random-phase approximation[111], and contains terms describing intraband and inter- band transitions. While the former can be evaluated analytically[112], the interband term requires a numer- ical solution[111, 112]. The out-of-plane component of the dielectric tensor matches εr. We then run the eigenmode solver, using the relative permittivity function for SLG, and select the fundamen- tal (gap plasmon) mode for further processing. We ex- port the simulation mesh grid positions, electric E and magnetic H data, effective refractive index (neff = β/k0, where β is the propagation constant of the mode and k0 is the free space wavevector[113]), and all relevant geo- metric parameters such as gap and gate width and gate oxide thickness to complete the rest of our modeling. A crucial intermediate step requires the determina- tion of the Te(x) profile in the SLG channel. The first step establishes the operating regime. As discussed in Refs.[114, 115], the energy delivered to the electronic car- rier distribution by pumping SLG with a pulsed laser can (a) ) . u . a ( t Plasmonic (tox = 10 nm) Plasmonic (tox = 30 nm) Plasmonic (tox = 50 nm) Transparent a e H e u o J l 5x -1 -3 -2 0 1 2 3 Position in channel (μm) Plasmonic (tox = 10 nm) Plasmonic (tox = 30 nm) Plasmonic (tox = 50 nm) Transparent -3 -2 -1 0 1 2 3 Position in channel (μm) Plasmonic (tox = 10 nm) Plasmonic (tox = 30 nm) Plasmonic (tox = 50 nm) Transparent (b) ) . u . a ( e T Δ (c) ) . . u a ( x d / e T d -3 -2 -1 0 1 2 3 Position in channel (μm) FIG. 9. a) Joule heat source, b) Te profile, c) dTe dx along SLG channel for a representative GPD with plasmonic gates at different heights above the channel compared to gates trans- parent at 1.55µm be sufficiently high to result in Te(x) > 1000K. When modeling photoexcited SLG under these conditions, one has to take into account the Te(x) dependence[114, 115] of all thermodynamic and transport parameters in Eqs.1- 3, i.e. µc, σ(x),κe(x), resulting in a nonlinear system of coupled equations. The contrasting case, the "weak heating" regime, is characterized by ∆Te(x) ≪ Tl[62, 81]. Under this con- 10 dition, Eqs.1-3 can be solved to linear order in the lo- cal Te(x) fluctuation, evaluating all thermodynamic and transport parameters at Te(x) = Tl. In particular, a single µc is established, following photoexcitation, by electron-electron interactions across the valence and con- duction bands, and the thermal conductivity, calculated BTeσ/(3e2)[74], from the Wiedemann-Franz law κe = π2k2 is uniform in space. The intended operation of our GPD is under CW or, during data reception, quasi-CW (i.e. the pulse duration exceeds the cooling time of the hot-carrier distribution) illumination with low Pin < 0.1mW at in-plane incidence. Furthermore, a linear dependence of Vph on Pin, such as in Fig.6d, is observed[62]. We thus conclude that our device operates in the "weak heating" regime. In order to evaluate Eq.3, we need to specify the cool- ing length ξ and the light absorption heat source. In principle, ξ, which describes the energy transfer from the electronic system to the optical phonons[65], depends on Te(x) and n. However, under the "weak heating" regime, the Te(x) dependence can be neglected, and n is uniform in the device (with opposite sign in the two regions of the p-n junction) when the photoresponse is maximal. For these reasons, in our calculations, we as- sume a constant ξ = 1µm, consistent with experimental values[47, 78]. J(x) is calculated from the simulated elec- tric field data as the time average of the electric power dissipation density[97, 98]: J(x) = 1 2 ωǫ′′E2 = 1 2 σ t E · E∗ (5) In order to relate J(x) to physically meaningful quan- tities, we integrate the normal component of the time- averaged Poynting vector over the simulation region and normalize it to a given input power. After solving Eq.3 for the Te(x) fluctuation profile ∆Te(x), we take its derivative with respect to x and ob- tain the second factor of the integrand in Eq.1. Figs.9a-c compare the absorption heat source, the resulting Te(x) profile, and its derivative for a representative GPD in presence of a plasmonic split-gate at different heights over the SLG channel, to an unperturbed fundamental dielectric waveguide mode, where the p-n junction is gen- erated by a transparent (at the chosen λ) gate, such as a split-gate made from a second SLG at a separation large enough to avoid additional optical losses. The beneficial role of plasmonic enhancement, with all other parame- ters fixed, for a sharper increase in Te(x) translates to larger Vph if the SLG channel is kept close to the SPP waveguide (<50nm). To model S along the channel, we assume that the structure is gated to achieve the maximum S below the gates, as for Eq.2 (opposite in sign but equal in magni- tude for the p-n case) and approximate the gap region with a linear interpolation between the two. Combining both factors in Eq.1 and computing the integral yields 11 FIG. 10. Field distribution a) before the GPD, b) during transition, and c) start of the split-gate/SPP waveguide. Scale bar: 1µm. (d,e) E2 along the center (red box) of dielectric waveguide and GPD at d)1.5µm and e) = 1.6µm. Scale bar: 3µm Vph or RV (if divided by Pin) as figure of merit to assess different cross section designs: RV = Vph Pin = Z S(x) dTe(x) dx 1 Pin(cid:12)(cid:12)(cid:12)(cid:12) (6) dx(cid:12)(cid:12)(cid:12)(cid:12) To calculate RV or photovoltage maps as in Fig.6b, we first generate the S profile for all voltage combinations and then proceed via Eqs.6 or 1. We then perform a sweep of gap width, SLG position, gate oxide thickness, gate contact height as a function of µ in SLG and εr. As for Fig.9, shorter distances be- tween SLG channel and SPP waveguide yield larger sig- nals. Furthermore, the SPP waveguide width affects the expected photovoltage in two ways. 1) narrower confine- ment improves the field strength at the SLG channel and RV at the cost of higher propagation losses; 2) the un- gated SLG at the center of the device with not maximum S shrinks with smaller gaps. The device parameters are then chosen based on these trends, taking into account fabrication complexity, robustness to processing-induced deviations (e.g. suppression of fundamental gap plasmon mode below a certain gap width), reliability (e.g. ox- ide break-down) and outcomes of FDTD simulations on coupling and propagation. Extracting E and H of the fundamental mode in the hybrid region is sufficient for the design of the device cross-section. However, it does not capture the field dis- tribution along the device, since optical losses, transi- tion from dielectric to plasmonic waveguide, and power exchange between different modes that co-exist in the hybrid region remain unaccounted for. To find a good combination of SLG position and width within the con- straints of the cross-section design, targeting a Te dis- tribution with maximal derivative across the PD, but minimal along the propagation direction over the device length, we perform FDTD of the transition between SiN waveguide and GPD. We construct a device model in the same way as for the eigenmode analysis and adjust the source settings to 1.5-1.6µm. We launch the funda- mental quasi-TE mode of the SiN waveguide towards the GPD and use frequency domain field monitors (FDFM) with various orientations (parallel and perpendicular to the propagation direction) to track the field and power profiles at different points. Figs.10a-c plot the field at 3 cross-sections of a representative GPD. We see transi- tions from injected mode in dielectric waveguide to field distribution resembling the fundamental gap plasmon. Figs.10d,e display the electric field intensity from a ver- tical FDMD monitor along the center line of the GPD, at two wavelengths. As expected for this non-adiabatic transition with fast decrease (<1µm) of the taper cross- section down to the target gap size, scattering and re- flections at the start of the hybrid region reduce the power at the GPD, but the desired sharp intensity profile over length scales that match the fabricated SLG chan- nel widths is achieved. Consequently, we place the SLG channel 100nm after the SPP structure has reached its fi- nal gap width. The comparison of field intensities for 1.5 and 1.6µm on the same color scale in Figs.10d,e reveals a larger peak intensity and a longer interaction length for the latter, which indicates improved coupling efficiency at larger λ, as for the wavelength-depended RV in Fig.6c. REFERENCES ∗ E-mail: [email protected] [1] https://www.cisco.com/c/en/us/solutions/collateral/service- provider/visual-networking-index-vni/complete-white- paper-c11-481360.pdf [2] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. K. Soong, and J. C. Zhang, IEEE J. Sel. Areas Commun. 32, 1065 (2014). [3] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia, O. Queseth, M. Schellmann, H. Schotten, H. Taoka et al., IEEE Commun. Mag. 52, 26 (2014). [4] https://community.arm.com/cfs-file/key/telligent- evolution-components-attachments/01-1996-00-00-00- 01-30-09/Arm-2D00-The-route-to-a-trillion-devices- 2D00-June-2017.pdf [5] S. Rumley, D. Nikolova, R. Hendry, Q. Li, D. Calhoun, and K. Bergman, J. Light. Technol. 33, 547 (2015). [6] Z. Zhou, R. Chen, X. Li, and T. Li, Opt. Fiber Technol. 44, 13 (2018). [7] G. T. Reed, Silicon Photonics: The State of the Art (John Wiley & Sons, Chichester, 2008). [8] D. Thomson, A. Zilkie, J. E. Bowers, T. Komljenovic, G. T. Reed, L. Vivien, D. Marris-Morini, E. Cassan, L. Virot, J. F´ed´eli et al., J. Opt. 18, 073003 (2016). [9] P. P. Absil, P. Verheyen, P. De Heyn, M. Pantouvaki, G. Lepage, J. De Coster, and J. Van Campenhout, Opt. Express 23, 9369 (2015). [10] A. H. Atabaki, S. Moazeni, F. Pavanello, H. Gevor- gyan, J. Notaros, L. Alloatti, M. T. Wade, C. Sun, S. A. Kruger, H. Meng et al., Nature 556, 349 (2018). [11] A. Biberman, and K. Bergman, Rep. Prog. Phys. 75, 046402 (2012). [12] E. L. Wooten, K. M. Kissa, A. Yi-Yan, E. J. Murphy, D. A. Lafaw, P. F. Hallemeier, D. Maack, D. V. Attanasio, D. J. Fritz, G. J. McBrien et al., IEEE J. Sel. Topics Quantum Electron. 6, 69 (2000). [13] R. Nagarajan, M. Kato, J. Pleumeekers, P. Evans, S. Corzine, S. Hurtt, A. Dentai, S. Murthy, M. Missey, R. Muthiah et al., IEEE J. Sel. Topics Quantum Electron. 16, 1113 (2010). [14] G. T. Reed, G. Mashanovich, F. Y. Gardes, and D. J. Thomson, Nat. Photonics 4, 518 (2010). [15] G. T. Reed, G. Z. Mashanovich, F. Y. Gardes, M. Nedeljkovic, Y. Hu, D. J. Thomson, K. Li, P. R. Wilson, S. Chen, and S. S. Hsu, Nanophotonics 3, 229 (2014). [16] B. Seraphin, and N. Bottka, Phys. Rev. 139, A560 (1965). [17] J. Liu, M. Beals, A. Pomerene, S. Bernardis, R. Sun, J. Cheng, L. C. Kimerling, and J. Michel, Nat. Photonics 2, 433 (2008). [18] S. A. Srinivasan, M. Pantouvaki, S. Gupta, H. T. Chen, P. Verheyen, G. Lepage, G. Roelkens, K. Saraswat, D. Van Thourhout, P. Absil et al., J. Light. Technol. 34, 419 (2016). [19] J. Michel, J. Liu, and L. C. Kimerling, Nat. Photonics 4, 527 (2010). [20] A. R. Hawkins, W. Wu, P. Abraham, K. Streubel, and 12 J. E. Bowers, Appl. Phys. Lett. 70, 303 (1997). [21] H. Chang, Y. Kuo, R. Jones, A. Barkai, and J. E. Bow- ers, Opt. Express 18, 23891 (2010). [22] L. Chrostowski, and M. Hochberg, Silicon Photonics Design: From Devices to Systems (Cambridge Univer- sity Press, Glasgow, 2015). [23] L. Vivien, J. Osmond, J. F´ed´eli, D. Marris-Morini, P. Crozat, J. Damlencourt, E. Cassan, Y. Lecunff, and S. Laval, Opt. Express 17, 6252 (2009). [24] C. T. DeRose, D. C. Trotter, W. A. Zortman, A. L. Starbuck, M. Fisher, M. R. Watts, and P. S. Davids, Opt. Express 19, 24897 (2011). [25] L. Vivien, A. Polzer, D. Marris-Morini, J. Osmond, J. M. Hartmann, P. Crozat, E. Cassan, C. Kopp, H. Zim- mermann, and J. M. F´ed´eli, Opt. Express 20, 1096 (2012). [26] A. Novack, M. Gould, Y. Yang, Z. Xuan, M. Streshin- sky, Y. Liu, G. Capellini, A. E. Lim, G. Lo, T. Baehr- Jones et al., Opt. Express 21, 28387 (2013). [27] H. Chen, P. Verheyen, P. De Heyn, G. Lepage, J. De Coster, S. Balakrishnan, W. Yao, L. Shen, G. Roelkens, and J. Van Campenhout, Opt. Express 24, 4622 (2016). [28] J. Wang, and S. Lee, Sensors 11, 696 (2011). [29] H. Ye, and J. Yu, Sci. Technol. Adv. Mater. 15, 024601 (2014). [30] V. Sorianello, A. De Iacovo, L. Colace, A. Fabbri, L. Tortora, E. Buffagni, and G. Assanto, Appl. Phys. Lett. 101, 081101 (2012). [31] J. Liu, D. D. Cannon, K. Wada, Y. Ishikawa, S. Jongth- ammanurak, D. T. Danielson, J. Michel, and L. C. Kimerling, Appl. Phys. Lett. 87, 011110 (2005). [32] M. Romagnoli, V. Sorianello, M. Midrio, F. H. L. Koppens, C. Huyghebaert, D. Neumaier, P. Galli, W. Templ, A. D'Errico, and A. C. Ferrari, Nat. Rev. Mater. 3, 392414 (2018). [33] M. Liu, X. Yin, E. Ulin-Avila, B. Geng, T. Zentgraf, L. Ju, F. Wang, and X. Zhang, Nature 474, 64 (2011). [34] M. Liu, X. Yin, and X. Zhang, Nano Lett. 12, 1482 (2012). [35] Y. Hu, M. Pantouvaki, J. Van Campenhout, S. Brems, I. Asselberghs, C. Huyghebaert, P. Absil, and D. Van Thourhout, Laser Photon. Rev. 10, 307 (2016). [36] C. T. Phare, Y. Daniel Lee, J. Cardenas, and M. Lipson, Nat. Photonics 9, 511 (2015). [37] V. Sorianello, M. Midrio, and M. Romagnoli, Opt. Ex- press 23, 6478 (2015). [38] V. Sorianello, M. Midrio, G. Contestabile, I. As- selberghs, J. Van Campenhout, C. Huyghebaert, I. Goykhman, A. K. Ott, A. C. Ferrari, and M. Romag- noli, Nat. Photonics 12, 40 (2018). [39] Z. Sun, A. Martinez, and F. Wang, Nat. Photonics 10, 227 (2016). [40] X. Gan, R. Shiue, Y. Gao, I. Meric, T. F. Heinz, K. Shepard, J. Hone, S. Assefa, and D. Englund, Nat. Pho- tonics 7, 883 (2013). [41] A. Pospischil, M. Humer, M. M. Furchi, D. Bachmann, R. Guider, T. Fromherz, and T. Mueller, Nat. Photonics 7, 892 (2013). [42] X. Wang, Z. Cheng, K. Xu, H. K. Tsang, and J. Xu, Nat. Photonics 7, 888 (2013). [43] I. Goykhman, U. Sassi, B. Desiatov, N. Mazurski, S. Milana, D. De Fazio, A. Eiden, J. Khurgin, J. Shappir, U. Levy et al., Nano Lett. 16, 3005 (2016). [44] D. Schall, D. Neumaier, M. Mohsin, B. Chmielak, J. Bolten, C. Porschatis, A. Prinzen, C. Matheisen, W. Kuebart, B. Junginger et al., ACS Photonics 1, 781 (2014). [45] S. Schuler, D. Schall, D. Neumaier, L. Dobusch, O. Bethge, B. Schwarz, M. Krall, and T. Mueller, Nano Lett. 16, 7107 (2016). [46] S. Schuler, D. Schall, D. Neumaier, B. Schwarz, K. Watanabe, T. Taniguchi, and T. Mueller, ACS Photon- ics 5, 4758 (2018). [47] R. J. Shiue, Y. Gao, Y. Wang, C. Peng, A. D. Robert- son, D. K. Efetov, S. Assefa, F. H. L. Koppens, J. Hone, and D. Englund, Nano Lett. 15, 7288 (2015). [48] D. Schall, C. Porschatis, M. Otto, and D. Neumaier, J. Phys. D: Appl. Phys. 50, 124004 (2017). [49] D. Schall, E. Pallecchi, G. Ducournau, V. Avramovic, M. Otto, and D. Neumaier, in Optical Fiber Communi- cation Conference, M2I.4, (Optical Society of America, 2018). [50] Y. Ding, Z. Cheng, X. Zhu, K. Yvind, J. Dong, M. Galili, H. Hu, N. A. Mortensen, S. Xiao, and L. K. Ox- enløwe, arXiv : 1808.04815 (2018). [51] P. Ma, Y. Salamin, B. Baeuerle, A. Josten, W. Heni, A. Emboras, and J. Leuthold, ACS Photonics 6, 154 (2019). [52] Z. Ma, K. Kikunage, H. Wang, S. Sun, R. Amin, M. Tahersima, R. Maiti, M. Miscuglio, H. Dalir, and V. J. Sorger, arXiv : 1812.00894 (2018). [53] F. Bonaccorso, Z. Sun, T. Hasan, and A. C. Ferrari, Nat. Photonics 4, 611 (2010). 13 M. Bernechea, F. P. G. de Arquer, F. Gatti, and F. H. L. Koppens, Nat. Nanotechnol. 7, 363 (2012). [68] L. Vicarelli, M. S. Vitiello, D. Coquillat, A. Lombardo, A. C. Ferrari, W. Knap, M. Polini, V. Pellegrini, and A. Tredicucci, Nat. Mater. 11, 865 (2012). [69] M. Freitag, T. Low, F. Xia, and P. Avouris, Nat. Pho- tonics 7, 53 (2013). [70] U. Sassi, R. Parret, S. Nanot, M. Bruna, S. Borini, D. De Fazio, Z. Zhao, E. Lidorikis, F. H. L. Koppens, A. C. Ferrari et al., Nat. Commun. 8, 14311 (2017). [71] N. Huo, and G. Konstantatos, Adv. Mater. 30, 1801164 (2018). [72] D. Brida, A. Tomadin, C. Manzoni, Y. J. Kim, A. Lom- bardo, S. Milana, R. R. Nair, K. S. Novoselov, A. C. Ferrari, G. Cerullo et al., Nat. Commun. 4, 1987 (2013). [73] A. Tomadin, D. Brida, G. Cerullo, A. C. Ferrari, and M. Polini, Phys. Rev. B 88, 035430 (2013). [74] C. Kittel, Introduction to solid state physics (Wiley New York, New York, 1996). [75] N. Bonini, M. Lazzeri, N. Marzari, and F. Mauri, Phys. Rev. Lett. 99, 176802 (2007). [76] M. Lazzeri, S. Piscanec, F. Mauri, A. Ferrari, and J. Robertson, Phys. Rev. Lett. 95, 236802 (2005). [77] N. Ashcroft, and N. Mermin, Solid State Physics (Har- court College Publishers, Fort Worth, 1976). [78] Q. Ma, N. M. Gabor, T. I. Andersen, N. L. Nair, K. Watanabe, T. Taniguchi, and P. Jarillo-Herrero, Phys. Rev. Lett. 112, 247401 (2014). [79] M. Freitag, T. Low, and P. Avouris, Nano Lett. 13, 1644 [54] A. Urich, K. Unterrainer, and T. Mueller, Nano Lett. (2013). 11, 2804 (2011). [55] F. Xia, T. Mueller, Y. Lin, A. Valdes-Garcia, and P. Avouris, Nat. Nanotechnol. 4, 839 (2009). [56] R. R. Nair, P. Blake, A. N. Grigorenko, K. S. Novoselov, T. J. Booth, T. Stauber, N. M. R. Peres, and A. K. Geim, Science 320, 1308 (2008). [57] J. M. Dawlaty, S. Shivaraman, J. Strait, P. George, M. Chandrashekhar, F. Rana, M. G. Spencer, D. Veksler, and Y. Chen, Appl. Phys. Lett. 93, 131905 (2008). [58] F. H. Koppens, T. Mueller, P. Avouris, A. C. Ferrari, M. S. Vitiello, and M. Polini, Nat. Nanotechnol. 9, 780 (2014). [59] S. Goossens, G. Navickaite, C. Monasterio, S. Gupta, J. J. Piqueras, R. P´erez, G. Burwell, I. Nikitskiy, T. Lasanta, T. Gal´an et al., Nat. Photonics 11, 366 (2017). [60] N. Youngblood, and M. Li, Nanophotonics 6, 1205 (2017). [61] M. Liu, and X. Zhang, Nat. Photonics 7, 851 (2013). [62] K. Tielrooij, L. Piatkowski, M. Massicotte, A. Woess- ner, Q. Ma, Y. Lee, K. S. Myhro, C. N. Lau, P. Jarillo- Herrero, N. F. van Hulst et al., Nat. Nanotechnol. 10, 437 (2015). [63] T. Mueller, F. Xia, and P. Avouris, Nat. Photonics 4, 297 (2010). [64] T. J. Echtermeyer, P. S. Nene, M. Trushin, R. V. Gor- bachev, A. L. Eiden, S. Milana, Z. Sun, J. Schliemann, E. Lidorikis, K. S. Novoselov et al., Nano Lett. 14, 3733 (2014). [65] J. C. W. Song, M. S. Rudner, C. M. Marcus, and L. S. Levitov, Nano Lett. 11, 4688 (2011). [66] N. M. Gabor, J. C. Song, Q. Ma, N. L. Nair, T. Tay- chatanapat, K. Watanabe, T. Taniguchi, L. S. Levitov, and P. Jarillo-Herrero, Science 334, 648 (2011). [80] P. K. Herring, A. L. Hsu, N. M. Gabor, Y. C. Shin, J. Kong, T. Palacios, and P. Jarillo-Herrero, Nano Lett. 14, 901 (2014). [81] S. Castilla, B. Terres, M. Autore, L. Viti, J. Li, A. Nikitin, I. Vangelidis, K. Watanabe, T. Taniguchi, E. Lidorikis et al., Nano Lett. , (2019). [82] K. S. Novoselov, D. Jiang, F. Schedin, T. J. Booth, V. V. Khotkevich, S. V. Morozov, and A. K. Geim, Proc. Natl. Acad. Sci. U.S.A. 102, 10451 (2005). [83] L. Wang, I. Meric, P. Y. Huang, Q. Gao, Y. Gao, H. Tran, T. Taniguchi, K. Watanabe, L. M. Campos, D. A. Muller et al., Science 342, 614 (2013). [84] D. G. Purdie, N. M. Pugno, T. Taniguchi, K. Watanabe, A. C. Ferrari, and A. Lombardo, Nat. Commun. 9, 5387 (2018). [85] D. De Fazio, D. G. Purdie, A. K. Ott, P. Braeuninger- Weimer, T. Khodkov, S. Goossens, T. Taniguchi, K. Watanabe, P. Livreri, F. H. L. Koppens et al., arXiv : 1904.01405 (2019). [86] V. Miseikis, F. Bianco, J. David, M. Gemmi, V. Pel- legrini, M. Romagnoli, and C. Coletti, 2D Mater. 4, 021004 (2017). [87] X. Li, C. W. Magnuson, A. Venugopal, J. An, J. W. Suk, B. Han, M. Borysiak, W. Cai, A. Velamakanni, Y. Zhu et al., Nano Lett. 10, 4328 (2010). [88] B. Wang, M. Huang, L. Tao, S. H. Lee, A. Jang, B. Li, H. S. Shin, D. Akinwande, and R. S. Ruoff, ACS Nano 10, 1404 (2016). [89] J. W. Suk, A. Kitt, C. W. Magnuson, Y. Hao, S. Ahmed, J. An, A. K. Swan, B. B. Goldberg, and R. S. Ruoff, ACS Nano 5, 6916 (2011). [90] I. Goykhman, B. Desiatov, J. Khurgin, J. Shappir, and U. Levy, Nano Lett. 11, 2219 (2011). [67] G. Konstantatos, M. Badioli, L. Gaudreau, J. Osmond, [91] I. Goykhman, B. Desiatov, J. Khurgin, J. Shappir, and 14 U. Levy, Opt. Express 20, 28594 (2012). [92] Y. Salamin, P. Ma, B. Baeuerle, A. Emboras, Y. Fedo- ryshyn, W. Heni, B. Cheng, A. Josten, and J. Leuthold, ACS Photonics 5, 3291 (2018). [93] T. J. Echtermeyer, L. Britnell, P. K. Jasnos, A. Lom- bardo, R. V. Gorbachev, A. N. Grigorenko, A. K. Geim, A. C. Ferrari, and K. S. Novoselov, Nat. Commun. 2, 455 (2011). [94] Z. Fang, Z. Liu, Y. Wang, P. M. Ajayan, P. Nordlander, and N. J. Halas, Nano Lett. 12, 3808 (2012). [95] T. J. Echtermeyer, S. Milana, U. Sassi, A. Eiden, M. Wu, E. Lidorikis, and A. C. Ferrari, Nano Lett. 16, 8 (2016). [96] C. Chen, N. Youngblood, R. Peng, D. Yoo, D. A. Mohr, T. W. Johnson, S. Oh, and M. Li, Nano Lett. 17, 985 (2017). [97] B. Desiatov, I. Goykhman, and U. Levy, Nano Lett. 14, 648 (2014). [104] A. C. Ferrari, J. C. Meyer, V. Scardaci, C. Casiraghi, M. Lazzeri, F. Mauri, S. Piscanec, D. Jiang, K. S. Novoselov, S. Roth et al., Phys. Rev. Lett. 97, 187401 (2006). [105] A. C. Ferrari, and D. M. Basko, Nat. Nanotechnol. 8, 235 (2013). [106] A. Das, S. Pisana, B. Chakraborty, S. Piscanec, S. K. Saha, U. V. Waghmare, K. S. Novoselov, H. R. Krish- namurthy, A. K. Geim, A. C. Ferrari et al., Nat. Nan- otechnol. 3, 210 (2008). [107] D. M. Basko, S. Piscanec, and A. C. Ferrari, Phys. Rev. B 80, 165413 (2009). [108] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). [109] J. M. Senior, and M. Y. Jamro, Optical fiber commu- nications: principles and practice (Pearson Education, Harlow, 2009). [98] W. Shin, A. Raman, and S. Fan, J. Opt. Soc. Am. B [110] J. Diels, and W. Rudolph, Ultrashort laser pulse phe- 29, 1048 (2012). nomena (Elsevier, London, 2006). [99] J. Tian, S. Yu, W. Yan, and M. Qiu, Appl. Phys. Lett. [111] N. K. Emani, A. V. Kildishev, V. M. Shalaev, and A. 95, 013504 (2009). [100] G. Dabos, D. Ketzaki, A. Manolis, E. Chatzianagnos- tou, L. Markey, J. Weeber, A. Dereux, A. L. Giesecke, C. Porschatis, B. Chmielak et al., IEEE Photon. J. 10, 2700308 (2018). [101] Q. Li, and M. Qiu, Opt. Express 18, 15531 (2010). [102] C. Delacour, S. Blaize, P. Grosse, J. M. Fedeli, A. Bruyant, R. Salas-Montiel, G. Lerondel, and A. Chel- nokov, Nano Lett. 10, 2922 (2010). [103] A. Lagatsky, Z. Sun, T. Kulmala, R. Sundaram, S. Mi- lana, F. Torrisi, O. Antipov, Y. Lee, J. Ahn, C. Brown et al., Appl. Phys. Lett. 102, 013113 (2013). Boltasseva, Nanophotonics 4, 214 (2015). [112] G. W. Hanson, J. Appl. Phys. 103, 064302 (2008); ibid. 113, 029902 (2013). [113] G. T. Reed, and A. P. Knights, Silicon Photonics: An Introduction (John Wiley & Sons, Chichester, 2004). [114] G. Soavi, G. Wang, H. Rostami, D. G. Purdie, D. De Fazio, T. Ma, B. Luo, J. Wang, A. K. Ott, D. Yoon et al., Nat. Nanotechnol. 13, 583 (2018). [115] G. Soavi, G. Wang, H. Rostami, A. Tomadin, O. Balci, I. Paradeisanos, E. Pogna, G. Cerullo, E. Lidorikis, M. Polini et al., arXiv : 1903.00989 (2019).
1806.11513	1	1806	2018-06-29T16:15:12	Silicon Nitride MOMS Oscillator for Room Temperature Quantum Optomechanics	[ "physics.app-ph", "physics.optics", "quant-ph" ]	Optomechanical SiN nano-oscillators in high-finesse Fabry-Perot cavities can be used to investigate the interaction between mechanical and optical degree of freedom for ultra-sensitive metrology and fundamental quantum mechanical studies. In this work we present a nano-oscillator made of a high-stress round-shaped SiN membrane with an integrated on-chip 3D seismic filter properly designed to reduce mechanical losses. This oscillator works in the 200 kHz - 5 MHz range and features a mechanical quality factor of $Q\simeq10^7$ and a Q-frequency product in excess of $6.2 \times 10^{12}$ Hz at room temperature, fulfilling the minimum requirement for quantum ground-state cooling of the oscillator in an optomechanical cavity. The device is obtained by MEMS DRIE bulk micromachining with a two-side silicon processing on a Silicon-On-Insulator (SOI) wafer. The microfabrication process is quite flexible and additional layers could be deposited over the SiN membrane before the DRIE steps, if required for a sensing application. Therefore, such oscillator is a promising candidate for quantum sensing applications in the context of the emerging field of Quantum Technologies.	physics.app-ph	physics	Silicon Nitride MOMS Oscillator for Room Temperature Quantum Optomechanics Enrico Serra, Bruno Morana, Antonio Borrielli, Francesco Marin, Gregory Pandraud, Antonio Pontin, Giovanni Andrea Prodi, Pasqualina M. Sarro, and Michele Bonaldi ∗ July 2, 2018 Abstract Optomechanical SiN nano-oscillators in high-finesse Fabry-Perot cavities can be used to investigate the interaction between mechanical and optical degree of freedom for ultra-sensitive metrology and fundamental quantum mechanical studies. In this work we present a nano-oscillator made of a high- stress round-shaped SiN membrane with an integrated on-chip 3D seismic filter properly designed to reduce mechanical losses. This oscillator works in the 200 kHz - 5 MHz range and features a mechanical quality factor of Q ≃ 107 and a Q-frequency product in excess of 6.2 × 1012 Hz at room temperature, fulfilling the minimum requirement for quantum ground-state cooling of the oscillator in an optomechanical cavity. The device is obtained by MEMS DRIE bulk micromachining with a two-side silicon processing on a Silicon-On-Insulator (SOI) wafer. The microfabrication process is quite flexible and additional layers could be deposited over the SiN membrane before the DRIE steps, if required for a sensing application. Therefore, such oscillator is a promising candidate for quantum sensing applications in the context of the emerging field of Quantum Technologies. 1 Introduction Light facilitates exploration of quantum phenomena and enables radical new technologies. Coupling light with resonant mechanical systems in an optical cavity (Fabry-Perot), known as cavity optomechanics [1,2], serves as a basis for fundamental studies and is a flexible tool for a wide range of scientific and technological goals. We mention for instance the sensing of forces or fields at the ultimate limits imposed by quantum mechanics [3, 4], and experiments testing the foundation of physics, i.e., fixing significant constraints for the development of quantum gravity theories [5, 6]. Effects related to the quantum fluctuations of the radiation pressure are not easily detected because of classical noise sources, like the Brownian thermal noise and spurious coupling of the optomechanical system to the environment. As the number of quantum-coherent oscillations in the presence of thermal decoherence scales with the Q×f product (mechanical quality factor × oscillator frequency), this parameter is commonly used as a figure of merit for evaluating the oscillator's performance in optomechanics [2]. For this reason Silicon nitride (SiN) micro- and nano-mechanical membrane resonators have attracted a lot of attention due to their exceptionally high-Q factors [7, 8]. Systems based on a membrane oscillator have shown for the first time the mechanical effect of the quantum noise in the light [9] and achieved one of the first observations of pondero-motive light squeezing [10]. Moreover, high-Q SiN membranes are ∗E.Serra is with Istituto Nazionale di Fisica Nucleare, TIFPA, 38123 Povo (TN), Italy (e-mail:[email protected]); B. Morana, G.Pandraud and P.M. Sarro are with ECTM-EKL, Delft University of Technology, 2628 Delft, The Nether- lands (e-mail: [email protected]; [email protected]; [email protected]); A. Borrielli, M. Bonaldi are with in- stitute of Materials for Electronics and Magnetism, Nanoscience-Trento-FBK Division, 38123 Povo (TN), Italy (e-mail: [email protected]; [email protected]); F. Marin and A. Pontin are with Dipartimento di Fisica e Astronomia and LENS, Universit`a di Firenze, and INFN Sezione di Firenze, Via Sansone 1, 50019 Sesto Fiorentino (FI), Italy and with CNR- INO, L.go Enrico Fermi 6, 50125 Firenze, Italy (email: [email protected]; [email protected]); G.A. Prodi is with Dipartimento di Fisica, Universit`a di Trento, 38123 Povo (TN), Italy (e-mail: [email protected]). The authors gratefully acknowledge the EKL IC process group for their technological support. Project QuaSeRT has received funding from the QuantERA ERA-NET Cofund in Quantum Technologies implemented within the European Union's Horizon 2020 Programme. The research has been partially supported by INFN (HUMOR project). 1 interesting because they can be used in collective optotomechanics, for the realisation of entangled states between mechanical modes [11], or in systems working in the anti-squashing regime of positive feedback for enhanced sideband cooling [12]. In many cases the oscillators consist of a commercial free-standing high-stress silicon nitride (SiN) mem- brane supported by a silicon frame, where mechanical quality factors up to many millions can be in principle obtained [13] thanks to the large tensile stress (of the order of GPa). In these systems the me- chanical loss is strongly dependent on the mounting, especially for the low frequency modes, and values randomly scattered from 104 to some 106 are usually observed [14]. Different approaches can be used to isolate the SiN membrane from its support. For instance in [11] a phononic bandgaps allowed the obser- vation of ponderomotive squeezing at moderate cryogenic temperatures, while in [15–17] the resonating part is decoupled by its substrate by high-aspect-ratio SiN trampolines. These achievements motivate the research on customised SiN resonators for optomechanics, using MEMS wafer-scale approaches and with a precise control of the overall dissipation mechanisms [18]. In fact, the mechanical dissipation in an oscillator determines the thermal fluctuation noise, and quantum behavior can emerge only if the thermal noise force is lower than noise of quantum origin, such as radiation pressure shot-noise. The Power Spectral Density (PSD) of the thermal noise force is ST h = 4kBT mω Q , where m is the mass of the oscillator, ω its angular frequency, and T the temperature. The quality factor Q is simply related to the fraction of mechanical energy loss per oscillation cycle, called φ, as: Q−1 = φ ≡ ∆Wt , with Wt the energy stored in the resonant mode and ∆Wt the energy loss per cycle. 2πWt As said above, the figure of merit which is the product of mechanical quality factor Q and frequency f is used as object function in the device optimization process. Indeed, this figure of merit determines the number of coherent oscillations in the presence of thermal fluctuations, and the minimum requirement for room-temperature quantum optomechanics is Q × f > 6.2 THz [2]. We have recently proposed a novel coupled oscillators model for the mechanical losses in a membrane oscillator [19], considering the mutual interaction between the membrane and the frame in a recoil losses analysis. We were able to design an effective shield for the losses for all mechanical modes also in the low frequency range. In this work, we present the whole microfabrication process of a MOMS oscillator built following these design rules, where a circular SiN nanomembrane [20] is integrated on-chip with a silicon loss-shield. The shield works as isolation stage and protects the oscillator from the mechanical decoherence induced by the thermal bath. The device is fully functional from room to cryogenic temperatures, and reaches high f × Q values for all resonant modes in the measured bandwidth [0.2-5] MHz. The micro- fabbrication combines silicon double-side deep-RIE etching on a Silicon-On-Insulator (SOI) wafer with controlled wet-etch HF steps for removing sacrificial oxides and controlling the release of the nanomem- brane. Additional layers can be deposited onto the SiN layer before fabrication, for sensing applications or to tailor the membrane reflectivity, without increasing the process complexity. 2 RESONATOR DESIGN A round-shaped high-stress non-stoichiometric SiN membrane (with tensile intrinsic stress of 0.830 GPa), deposited by Low-Pressure Chemical Vapour Deposition (LPCVD), of thickness around 100 nm is analysed. This membrane interacts by the action of the radiation pressure, with a 1064 nm laser beam. To improve the quality factor and to reject vibrational noise from the environment, a mechanical filtering stage is integrated on-chip by exploiting double side Deep Reactive-Ion Etching (DRIE). A detail view of the device is shown in Fig. 1. The design of the SiN MOMS oscillator comprises of three main parts: (1) the resonating thin silicon nitride (SiN) nanomembrane, (2) the loss-shield working as intermediate filter (made of a clamping hollow cylinder and flexural/torsional springs), (3) the outer silicon frame that connects the oscillator to the sample holder. Fig. 1b shows a detailed view of the cross section of the central part of the system, highlighting the thin-film layers and the SOI wafer structure. In the subsections 2.1, 2.2, 2.3 we focus our discuss on how to minimise the thermal noise effects by reducing the mechanical dissipations of the substrate with a loss shield. In subsection 2.4 we discuss how to limit the detrimental effects on Q-fact due to the oxide layer even at liquid-He (4 K) temperatures. Figure 1: (a) View of the SiN circular membrane oscillator. The main oscillator (SiN membrane) is in the center and surrounded by the loss-shield structure (filter). (b) Detailed view of the central part of the oscillator showing the SOI wafer and thin film layers (not to scale). 2.1 Total losses The energy loss in a SiN membrane is dominated by the intrinsic loss, that is the imaginary component in the stress-strain relation. This loss sets the ultimate limit Qint for the mode's Q-factors, but when the membrane is part of a complex device, different loss channels add up according to this simple relation: Q−1 T OT = φT OT = φint + φclam + φgas + φT ED (1) where: φclam is due to the clamping loss, φgas describes viscous damping arising from gas surrounding the membrane, φT ED is the thermoelastic damping. We point out that, in a highly stressed membrane oscillator, the total Q-factor is also proportional to the internal stress. In fact when the internal stress grows, the energy dissipated in a time unit remains the same, while the oscillation frequency is much higher. The energy lost per cycle φT OT , thanks to this dilution effect, is reduced and we observe a corresponding improvement of QT OT [13, 21]. Given that the internal stress cannot be much larger than 1 GPa, the technological efforts are currently directed to reduce contributions from each loss channel. In our MOMS device we can neglect the gas damping φgas because the oscillator operate in a High Vacuum setup, and the thermoelastic damping of the SiN layer is orders of magnitude lower than the other losses mechanisms. Therefore the most relevant contribution is the clamping loss φclam, that depends on the detail of the mechanical impedance of the support system. In fact, as observed for instance by Morse [22] for the case of a string supported by a sounding board, the damping depends on the mechanical impedance of the support. Only in the special case of infinitely rigid and massive support, the damping of the oscillating part is determined only by its intrinsic dissipation. In case of highly stressed membranes, the typical resonant frequency is around 1 MHz. In this frequency range, the silicon die cannot be considered as a rigid body but its full modal response must be considered in evaluating its admittance, because a support resonating at the same frequency of the membrane can be very effective in absorbing mechanical energy. For this reason we have developed a theoretical model [19] where we evaluate the loss when the membrane and support are fully-coupled, i.e. allowing for the transfer of energy in both ways. In the next section we evaluate the clamping loss when the membrane and the silicon frame are coupled structures, and we show how the loss can be reduced by using mechanical insulation stages (i.e. a loss-shield structure). Figure 2: (a) Model used for in the FEM analysis for the calculation of the thermal noise and the evaluation of the Q-factor. (b) Lumped-model for the three resonators (membrane-loss shield-wafer) for the physical interpretation of the recoil losses and the evaluation of the thermal noise at the membrane surface and of the Q-factor 2.2 Coupled system losses For a precise evaluation of total loss, we consider two-way energy transfer between membrane and its support. The coupling is originated from the recoil forces or torques of the structure surrounding the vibrating SiN membrane. The damping originating from recoil forces has been first evaluated by Saulson [23] in the analysis of two cascaded pendula used as mechanical filters in Gravitational Wave detectors. If we model the membrane of mass Mm and the wafer of mass Mw as two coupled harmonic oscillator, with ωm and ωw the resonant frequencies of the the uncoupled parts, the effective loss angle φm of the membrane in the coupled system is: φm(ωm) = φm + φw Mm Mw wω2 ω2 m w − ω2 (ω2 m)2 , ωw 6= ωm (2) where φm, φw are the intrinsic loss factor of the membrane and the wafer, respectively. This equation shows that in the case of a massive, rigid support with a small intrinsic dissipation (φw ≃ 0), the dissipation of the coupled system is essentially determined by the intrinsic loss in the membrane. But if the two parts are at resonance, i.e. ωw ≃ ωm, an approximated solution, valid for the case φm ≪ 1, 102 < φ−1 w < 105 and Mw/Mm ≫ 1 φ2 , gives: w φm ≃ φm + 1 φ2 w M 2 m M 2 w , ωw ≃ ωm (3) and we see that the loss of the coupled membrane will grow as some mechanical energy leaks toward the supporting wafer. 2.3 Shielding recoil losses To overcome the loss due to the coupling between the wafer and the resonating membrane, we intro- duce a third body with a known dynamical impedance (loss-shield stage) that can effectively decouple the membrane from the support. This intermediate stage is also effective in filtering-out low frequency displacement noise. This approach will make the Q-factor independent from the internal resonant modes of the system setup. In contrast, commercial TMAH/KOH wet-etched SiN membranes have a Q-factor dependent on the specific clamping implemented in the setup. Given that a dissipative system does not admit a rigorous description in terms of normal modes [24], we evaluate by a numerical analysis the mechanical susceptibility and the associated loss of the motion of the membrane as seen from the readout port. In our system the optical readout samples the displacement of the surface on a circular area with a gaussian intensity profile centered on the membrane, as shown in Fig. 2a. We point out that the results of the analysis are only slightly dependent on the implementation details, and the optimization remains valid if the beam shape is changed or if a radio frequency readout is used. In Fig. 2 we show a CAD image of the 3D real system, a practical implementation of a loss-shield made of flexural and torsional joints and the corresponding three-mode lumped resonating system. In both cases we indicate the application point of the laser beam used for the detection of the displacement. As a first step, we calculated the mechanical response of the lumped tree-mode oscillator, driven by a harmonic force. The effective loss (and therefore the quality factor) is evaluated from the imaginary part of the complex susceptibility. The parameters of the model are then optimized to obtain Q-factors of the order of 107. As a second step, we study the 3D model by Finite Element and tune its geometry to match the dynamic admittance of the optimal three-oscillator model. Referring to Fig. 2b, in the frequency domain the displacement array is related to the force array by a 3×3 mechanical impedance matrix as: [ Zij(ω)] u = Fd, where u = [u1, u2, u3]T is the displacement vector and Fd is the driving force vector. The tridiagonal matrix [ Zij(ω)] can be easily obtained as a function of the complex spring constants kn = kn(1 + iφn) (with kn the spring constant and φn the associated loss factor): [ Zij(ω)] =   −k1 0 −M1ω2 + k1 −k1 −M2ω2 + k1 + k2 −k2 −M3ω2 + k2 + k3 0 −k2   (4) Here (M1, k1) are the effective mass and the complex spring constant of the support (wafer + OFHC copper support), (M2, k2) are the effective mass and the complex spring constant of the loss-shield structure and (M3, k3) are the effective mass and the complex spring constant of the membrane. The formal solution of the dynamical equation is u = [ Yij (ω)] Fd, where [ Yij (ω)] = [ Zij(ω)]−1 is the admittance matrix of the system. This equation could be solved to evaluate the resonant frequencies and the complex normal modes of the system, but on a practical level we need the response of the membrane M3 to a single external force F3, as the membrane's position is the only measured variable in the system. In agreement with the fluctuation-dissipation theorem, this is equivalent to the study of the thermal noise power spectrum of the oscillator, as seen from the readout port [25]: S(ω) = − 4kBT ω ℑn[ Yij(ω)] [0, 0, F3]To. (5) This curve has a peak at the resonant frequency of the membrane oscillator, and the full width at half maximum of this peak gives a measurement of the quality factor. We point out that this approach estimates the quality factor measured by the dynamical response of the "coupled membrane" (i.e. the membrane coupled to its support system), that is the experimentally accessible quantity. Given that the membrane is only weakly coupled with its support, its admittance will remain similar in shape to the uncoupled oscillator case: ℑ( Y c(ω)) ≃ − 1 ¯Mm ¯φm ω2 m m)2 + ¯φ4 (ω2 − ¯ω2 m ¯ω4 m . (6) with some proper values of effective mass ¯Mm, effective frequency ¯ωm and effective loss ¯φm. These effective values depend on the geometrical parameters of the loss shield and wafer, (M1, k1) and (M2, k2). Numerical Table 1: Parameters of the uncoupled oscillators used in the 3-oscillator model. 3-oscillator model M1 ω1/2π 5 × 10−5 Kg 227 kHz k1 = M1ω2 1 1.01 × 108 N/m φ1 M2 ω2/2π 10−3 1 × 10−5 Kg 30 kHz k2 = M2ω2 2 3.55 × 105 N/m φ2 M3 ω3/2π 10−3 1.5 × 10−10 Kg 230 kHz k3 = M3ω2 3 3.13 × 102 N/m φ3 10−7 evaluation show that the corrections to the membrane mass and resonant frequency are negligible ( ¯Mm ≃ M3 and ¯ωm ≃ ω3), while the quality factor ¯Qm = φ−1 m can become smaller than the uncoupled Q3 if the frequency ω1 of the wafer approaches ωm. In this case the loss shield turns out to be very effective in protecting the quality factor of the membrane from excessive reduction. The optimal parameters of the loss shield, shown in Table 1 have been found with a numerical approach based on the model of Fig. 2b [19]. The actual geometry consistent with these parameters has been developed with the help of FEM modelling. 2.4 Intrinsic losses A critical issue concerning the design of MOMS oscillator is the thin oxide layer that is used as etch-stop layer for the Deep-RIE on the back side. The thin TEOS SiO2 layer is shown in the design concept of the oscillator Figure 1. This layer is completely removed from the membrane in a release step following the Deep-Rie etching of the substrate. This layer, if not removed properly at the membrane's edge, can originate an additional loss depending on its thickness. In particular, the dissipation behaviours of this layer can be critical at cryogenic temperatures, commonly required in quantum optomechanics, as SiO2 has a dissipation peak at 50 K [26] that affects its behaviour down to Liquid Helium temperatures. The intrinsic quality factor (Qint = φ−1 can be calculated, for a square-shaped membrane, with the approach described in [21] : int ) of a free-standing membrane dominated by intrinsic stress σ0 Qint(n1, n2, σ0) = 1 λ ERe EIm (1 + λ 1 + n2 n2 2 4 )−1 (7) where (n1, n2) are the mode indexes, λ = p2F/σ0hl2 with F the flexural rigidity, h, l the thickness and the side length of the square-shaped SiN layer. ERe and EIm are the real and imaginary parts of the complex Young's modulus, that are related to the loss factor according to the following relation: E = ERe + iEIm ≡ E(1 + iφ). We take 7 as an estimate for the quality factor in the case of a circular membrane with similar frequency. In 7, the physical meaning of λ is the ratio between the bending energy to the elongation energy for the fundamental mode (in our case λ = 1.08×10−3). The loss at the edge is described by the term independent from the mode indexes (n1, n2), that is: Qedge(σ0) = 1 λ ERe EIm (8) Table 2: Material parameters used for evaluating dissipation of the SiO2 layer at low and room tempera- tures (CL - clamping limit - TL thermoelastic limit) Low temperature Room temperature (at LHe) (at RT) ESiO2 (GPa) ESiN (GPa) ρSiO2 (Kg/m3) ρSiN (Kg/m3) σ0 (MPa) 77 260 2800 2100 800 77 260 2800 2100 800 φSiO2 φSiN φSi φwafer 1 × 10−3 2 × 10−5 1 × 10−6 1 × 10−3(CL) 5 × 10−4 2 × 10−5 3 × 10−4 (TL) 1 × 10−3(CL) To evaluate the lower limit of the quality factor due to the TEOS oxide layer, we estimate the components of the complex Young modulus by considering the effective thin film layer: ERe = ESiNfvSiN + ESiO2fvSiO2 = ESiNhSiN + ESiO2hSiO2 hSiN + hSiO2 EIm = ESiNφSiNfvSiN + ESiO2 φSiO2fvSiO2 = ESiNφSiNhSiN + ESiO2φSiO2 hSiO2 hSiN + hSiO2 (9) (10) where ERe and EIm are the effective real and imaginary Young modulus, fvSiN , fvSiO2 are the volume fractions while ESiN, ESiO2 are the Young modulus and the loss angle φSiN, φSiO2 , as reported in Table 2. These data are used to simulate the total membrane loss by FEM, showing that for a system without oxide the Q-factor can potentially reach 5 × 107 at 4 K, while a reduction of the overall Q-factor is expected in the bilayer sistem SiN/SiO2 (100/290) nm. At room temperature the extra dissipation due to the SiO2 layer is negligible. Results at 4 K and 300 K are reported in Figure 3. Figure 3: Results of FEM simulations. Q-factor at room temperature (300 K): (red-filled squares) 100 nm thick SiN membrane; (red-filled circles) bilayer SiN/SiO2 (100/290) nm. Q-factor at Liquid Helium (4 K): (blue squares) 100 nm thick SiN membrane; (blue circles) bilayer SiN/SiO2 (100/290) nm. 3 MICROFABRICATION Resonators were fabricated exploiting MEMS bulk-micromachining by Deep-Reaction Ion Etching (DRIE) and through wafer two-side processing, a process which has already demonstrated the capability of pro- ducing low-loss micro-mechanical systems [27, 28]. Resonators were fabricated from 4-inches Silicon-On- Figure 4: (1) SOI wafer; (2) thin film depositions: LPCVD SiN, PECVD SiO2, Al; (3) 1st DRIE etching on the back-side (BS) with a resist mask with a nested oxide mask; (4) photoresist removal and 2nd back-side (BS) etching using the nested oxide mask; (5) HF wet etching of the buried oxide; (6) 3nd back-side (BS) etching; (7) front-side (FS) resist AZ9260 and Al patterning followed by 4thfront-side (FS) DRIE etching; (8) membrane release: Al stripping by PES chemical bath and oxide HF wet etching. Insulator (SOI) Floating Zone wafer (Icemos Technology Ltd.) of total thickness 1000 ± 5 µm (device layer 250 µm), orientation < 100 > with resistivity > 1 kΩ cm (Fig. 4 step 1). We patterned the handle and device layers by multiple DRIE etching steps, using the 2 µm buried oxide as etch stop layer. Firstly, on the device layer we deposited a multilayer thin-film stack composed of a low pressure chemical vapor deposition (LPCVD) tetraethyl orthosilicate (TEOS) oxide film with thickness 290 nm (on the bottom), the 100-nm-thick LPCVD SiN membrane (in the middle) and a 1 µm thick RF-sputtered pure Al layer (on the top) as it is shown in Fig. 4 step 2. The TEOS oxide works as etch stop layer during DRIE etching steps while the Al layer is used as front side mask and as protection layer for the wet/dry etching processing phases. The SiN membrane is deposited at about 800◦C by tuning the stress to about 0.8 GPa (measured on a Si wafer by curvature methods TENCOR Flexus FLX-2908). Before RF-sputtering of the Al layer, a plasma enhanced chemical vapor deposition (PECVD) oxide mask of thickness 6 µm was deposited on the back side at 400 ◦C (Fig. 4 step 2). Silicon etching has been realized by means of DRIE two pulse BOSCH process in Omega i2L Rapier at 20 ◦C. A multi-level etching processing was performed to obtain the membrane's hole and the loss-shield masses, using resist AZ9260 and the nested PECVD oxide mask. With reference to Fig. 4 step 3-4 we first etch 200 µm and after stripping the resist we performed a second etching step of 550 µm landing on the buried-oxide. Figure 5: SEM image of the front side of the circular SiN membrane oscillator with the loss-shield. Figure 6: SEM image of the back side of the membrane oscillator showing the loss-shield masses. The estimated average etching rates were about 1.38 µm/cycle for Si and 5 nm/cycle for the PECVD oxide. The gas-flow rate and number of cycles were optimised to control the aspect ratio dependant etching and notching. Afterwards, we removed the buried oxide by means of an HF-based solution (Fig. 4 step 5) that fully etches the PECVD oxide mask. The SiN layer on the back side works as etch stop layer for HF and protects the thermal oxide underneath. To release the thin-film stack we performed the third DRIE etching step (250 µm) exploiting the 2 µm thermal oxide as mask (Fig. 4 step 6). We divided the etching step in two sub-steps in order to prevent membranes failure. We saw re-entrant sidewalls due to aspect Figure 7: SEM image of the front side showing the detail of the torsional-flexural joints. Figure 8: SEM image of the cross-section of the device showing torsional-flexural joints of the shield-loss stage. ratio dependant etching rates. By a last DRIE etching step (Fig. 4 step 7) we obtained the front-side suspending structure in the 250 µm device layer; in this case we used a 8 µm photoresists AZ9260 (to avoid micro masking of the Al layer). Front and back side of the whole oscillator structure are shown in Fig. 5 and Fig. 6 respectively, detailed views of the torsional-flexural joints are shown in Fig. 7 and Fig. 8. The last step is the SiN membrane release. First we stripped the Al layer using a PES solution, then by means of an HF-based solution we etched the TEOS oxide (Fig. 4 step 8). A HF over-etch is needed in this phase in order to completely release the SiN layer, as any remnant TEOS oxide layer could spoil the mechanical Q-factor of the membrane. The appearance of the undercut as it is shown in Fig. 9 (b) represents a good way to set a proper over-etching time. The process ends with dicing and cleaning from resist residues by oxygen plasma (TEPLA) and by HNO3 solution 99%. 4 EXPERIMENTAL CHARACTERISATION The experimental set-up used for the measurement of the quality factor is shown in Fig. 10. A beam of 3 mW reflected by the PBS1 from our light source, a Nd:YAG at 1064 nm, is aligned in a Michelson interferometer followed by a balanced homodyne detection. In details, a polarizing beam-splitter (PBS2) divides the beam into two parts, orthogonally polarized, forming the Michelson interferometer arms. Figure 9: (Top) Optical image of the front side of the SiN circular membrane with its integrated loss- shield. (Bottom) Detailed view of the (a) and (b) areas showing the undercut of SiO2 by HF wet etching for the membrane realease. At the end of the first one (reference arm) an electromagnetically-driven mirror M1 is used for phase- locking the interferometer in the condition of maximum displacement sensitivity. A double pass through a quarter-wave plate rotates by 90◦ the polarization of this beam, which is then transmitted by PBS2. The second beam (sensing arm) is focused on the membrane oscillator firmly fixed inside the vacuum chamber, and after reflection and double pass through the quarter-wave plate is reflected by the PBS2, where it overlaps with the reference beam reflected by M1. The overlapped beams are then monitored by Figure 10: Optical scheme of the Michelson interferometer apparatus for the measurement of the mechani- cal Q-factor of the modes of the SiN membrane. The main beam is split in PBS2 (polarizing beam-splitter) and then recombines in PBS2-PBS3 after reflection from mirror M1 (reference arm) and the SiN membrane (sensing arm). a homodyne detection, consisting of a half-wave plate, rotating the polarizations by 45◦, and a polarizing beamsplitter (PBS3) that divides the radiation into two equal parts sent to the photodiodes PD1 and PD2, whose outputs are subtracted. The signal obtained is a null-average, sinusoidal function of the path difference in the interferometer. Such a scheme is barely sensitive to laser power fluctuations. The difference signal is used as error signal in the locking servo-loop (the locking bandwidth is about 1 kHz) and also sent to the acquisition instruments for sensing the displacement of the membrane. The quality factor of the resonant modes are evaluated by the measurement of the free decay time after resonant excitation with a piezoelectric crystal. When the drive signal is removed, the mechanical vibration follows an exponentially damped decay whose envelope amplitude varies according to: u(t) = u0 exp(−t/τn0n1) where τnm = Q/πνn0n1 is the decay time of the mode and νn0n1 is the resonance frequency of mode with indexes (n0, n1), as explained in the next Section. In Fig. 11(a) we show the customized clamping systems for a 14 × 14 mm2 chip and the vacuum chamber Fig. 11(b). The clamping system is isolated from vibrations with a spring-mass system linked to the optical table from the bottom and the pressure in the vacuum chamber is about 1 × 10−6 mbar to prevent gas damping effects. Figure 11: (a) Clamping frame made of OFHC copper with the oscillator housed in the middle; (b) the system is mounted on top of a seismic filter and housed inside a vacuum chamber. 5 RESULTS AND DISCUSSION In Fig. 12 we report the quality factor of all of the frequency modes (up to 2 MHz) of a reference device (wf1726-14). In comparison with commercial devices, measured for instance in [21], these results highlight the effectiveness of the integrated loss-shield, as almost all modes achieve a high Q value, in good agreement with what is expected by (7) for square-shaped membranes. Figure 12: The Q-factor at room temperature of the resonant modes of a refernce device (wf1726-14). We show as a continuous line the least-square regression line of these data, that is in good agreement with the results expected for a square-shaped membrane of the same size [21]. In general, microfabricated systems should ensure good reproducibility of the operating parameters. To evaluate the robustness of our microfabrication process we compare the Q-factor measurements for five devices obtained in five different runs. Relevant geometrical parameters related to the thin film oxide and nitride layers are reported in Table 3. In Table 4 we report the intrinsic stress σ0, the measured diameter D and the Q-factor measurements for three vibrational modes. From these data we estimate the sensitivity of the Q-factor to the variations of the intrinsic stress ∆σ0 and undercut ∆Uox. The analysis refer to the Table 3: Ellipsometry measurement of the thickness of the TEOS oxide and the SiN and the average undercut after HF etching. DEVICE tox ± 0.5% tSiN ± 0.5% u ± 5% w1722-15 w1726-14 w1733-13 w1736-12 w1752-14 [nm] 298 298 302 302 298 [nm] 108 100 109 109 100 [µm] 5.48 2.35 6.37 2.60 2.21 three low-frequency oscillations with mode indexes (n0, n1) = (01)/(11)/(21), with modal shape shown in Figure 13. We point out that for these low frequencies modes a high quality factor can be obtained only thanks to the loss-shield stage. Figure 13: Modal shapes the circular membrane with mode indexes (n0, n1). The color scale, from light gray to red, is proportional to the absolute displacement from the equilibrium position. Stress variation are originated by pressure fluctuations and small variations of the gas ratio NH3/SiH4 during the LPCVD deposition. A precise estimate of the intrinsic stress σ0 of the membrane, reported in Table 4, was derived from the interferometric measurement of the frequency ν0 (Fig. 10). In fact the stress and the resonant frequency are connected by the relation: νn0,n1 = ν0αn0,n1 , where αn0,n1 is the n1-th root of the Bessel polynomial of order n0 and ν0 = 1 ρ . Here ρ = 2800 Kg/m3 is the density of the SiN film, and the diameter D0 was directly measured on the device by Leica DM750M optical microscope with magnification of 50×. According to 7, the mechanical Q-factor in a square membrane is directly related to the intrinsic stress σ0 by the dilution effect of the mechanical losses in SiN films. For this reason it is not surprising that higher quality factors are generally observed when the stress is higher. In Fig. 14 we show the relative variation of ∆Q/Q as a function of the relative variation of stress, taking as reference value the average of the measured stress over the five different devices. πD0q σ0 Figure 14: Sensitivity of the relative variations of the Q-factor, ∆Q/Q, with respect to the relative variation of the stress ∆σ0/σ0 for the first three modes: (white) mode indexes (01); (light-blue) mode indexes (11); (black) mode indexes (21). The reference values are the average of the measurements over the 5 devices: 753 MPa for the stress and (9.6 × 106, 11.06 × 106, 9.60 × 106) for quality factors of the three modes. Figure 15: Sensitivity of the relative variations of the Q-factor, ∆Q/Q, with respect to the relative variation of the undercut ∆Uox/Uox for the first three modes: (white) mode indexes (01); (light-blue) mode indexes (11); (black) mode indexes (21). The relative variations are deduced from the values measured in the reference device w1726-14. Table 4: Experimental measurements of device parameters and mechanical quality factor at room tem- perature (300 K). DEVICE σ0 D [MPa] [mm] w1722-15 703.07 1.55 w1726-14 712.22 1.65 w1733-13 862.90 1.55 w1736-12 818.48 1.50 w1752-14 671.33 1.55 νmn [kHz] 247.47 374.38 568.29 233.98 372.96 499.95 274.16 438.82 718.95 275.91 439.49 589.91 241.82 384.88 516.77 Q [106] 8.5 8.4 7.8 8.4 12.5 11.1 11.7 14.3 9.8 10.9 11.1 10.6 8.7 9.1 8.7 n0 n1 (01) (11) (21) (01) (11) (21) (01) (11) (21) (01) (11) (21) (01) (11) (21) f × Q [THz] 2.09 3.14 4.42 1.96 4.65 5.56 3.22 6.27 7.06 3.00 4.87 6.25 2.09 3.49 4.50 The undercut Uox is roughly proportional to the etching time chosen for the release of the SiN membrane (Fig. 4 step 8). Actually it is used to check if the SiN membrane is fully released from the SiO2 layer underneath: no undercut means no free-standing membrane. Starting form the reference process, corrisponding to device w1726-14, different etching time were tested to check possible effects on the quality factor. For each wafer the undercut, measured with Leica DM750M using a magnification of 50×, is reported in Table 3. As shown in Fig. 15, where no clear correlation can be seen, the undercut does not influence the Q-factor of the modes. Here the relative variations are evaluated from the reference device. 6 Conclusion This work presents the microfabrication process for the production of high-Q SiN membranes with an integrated on-chip 3D loss-shield. Within its operating bandwidth (0-5 MHz), the device preserves the intrinsic Q-factor of the pretensioned round-shaped nanomembrane, regardless of the modal order or resonant frequency. In particular, Q-factors at room temperature are about 107, achieving the minimum requirement for room-temperature quantum optomechanics Q × f > 6.2 THz for all modes with frequency higher than 600 kHz. The device is obtained by MEMS bulk-micromachining via Deep-Reaction Ion Etching (DRIE), combined with a two-side processing on a thick SOI wafer. The microfabrication process is quite flexible and additional layers could be deposited over the SiN membrane before the DRIE steps, if required for a sensing application. Therefore, such oscillator is a promising candidate for quantum sensing applications in the context of the emerging field of Quantum Technologies. References [1] I. Favero, K. Karrai, "Optomechanics of deformable optical cavities," Nat. Photonics, vol. 3, pp. 201-205, 2009, doi:10.1038/nphoton.2009.42. [2] M. Aspelmeyer, T. J. Kippenberg, F. Marquardt, "Cavity optomechanics," Rev. Mod. Phys., vol. 86, art. no. 1391, 2014, doi: 10.1103/RevModPhys.86.1391. [3] M. Metcalfe, "Applications of cavity optomechanics," Appl. Phys. Rev., vol. 1, art. 031105, 2014, doi: 10.1063/1.4896029. [4] A. Pontin, M. Bonaldi, A. Borrielli, L. Marconi, F. Marino, G. Pandraud, G. A. Prodi, P. M. Sarro, E. Serra, F. Marin, "Quantum nondemolition measurement of optical field fluctuations by optome- chanical interaction," Phys. Rev. A, vol. 97, art. no. 033833, 2018. [5] I. Pikovski, M.R. Vanner, M. Aspelmeyer, M. S. Kim, Caslav Brukner, "Probing Planck-scale physics with quantum optics," Nature Physics, vol. 8, pp. 393-397, 2012, doi: 10.1038/nphys2262. [6] M. Bawaj, C. Biancofiore, M. Bonaldi, F. Bonfigli, A. Borrielli, G. Di Giuseppe et al., "Probing deformed commutators with macroscopic harmonic oscillators," Nat. Communications, vol. 6, art. no. 7503, 2015, doi: 10.1038/ncomms8503. [7] J. D. Thompson, B. M. Zwickl, A. M. Jayich, F. Marquardt, S. M. Girvin, J. G. E. Harris, "Strong dispersive coupling of a high-finesse cavity to a micromechanical membrane," Nature, vol. 452, pp. 72-75, 2008, doi: 10.1038/nature06715 [8] Y. Tsaturyan, A. Barg, E. S. Polzik and A. Schliesser, "Ultracoherent nanomechanical resonators via soft clamping and dissipation dilution," Nature Nanotechnology, vol. 12, pp. 776783, doi: 10.1038/nnano.2017.101 [9] T. P. Purdy, R. W. Peterson, C. A. Regal, "Observation of Radiation Pressure Shot Noise on a Macroscopic Object," Science, vol. 339, n. 6121, pp. 801-804, 2013, doi: 10.1126/science.1231282. [10] T. P. Purdy, P.L. Yu, R. W. Peterson, N. S. Kampel, C. A. Regal, "Strong Optomechanical Squeezing of Light," Phys. Rev. X, vol. 3, art. no. 031012, 2013, doi: 10.1103/PhysRevX.3.031012. [11] W. H. P. Nielsen, Y. Tsaturyan, C.B. Møller, E.S. Polzik, A. Schliesser, "Multimode optomechanical system in the quantum regime," PNAS, vol. 114, n. 1, pp. 62-66, 2017, doi: 10.1073/pnas.1608412114. [12] M. Rossi, N. Kralj, S. Zippilli, R. Natali, A. Borrielli, G. Pandraud, E. Serra, G. Di Giuseppe and D. Vitali, "Enhancing sideband cooling by feedback controlled light," Phys. Rev. Lett., vol. 119, art. no. 123603, 2017, doi: 10.1103/PhysRevLett.119.123603. [13] Q. P. Unterreithmeier, T. Faust, J. P. Kotthaus, "Damping of nanomechanical resonators," Phys. Rev. Lett., vol. 105, art.no 027205, 2010, doi: 10.1103/PhysRevLett.105.027205. [14] D. J. Wilson, C. A. Regal, S. B. Papp, and H. J. Kimble, "Cavity Optomechanics with Stoichiometric SiN Films," Phys. Rev. Lett., vol 103, art. no. 207204, 2009, doi: 10.1103/PhysRevLett.103.207204. [15] R. A. Norte, J. P. Moura, S. Groblacher, "Mechanical Resonators for Quantum Optomechanics Exper- iments at Room Temperature," Phys. Rev. Lett., vol. 116, art. no. 147202, 2014, doi: 10.1103/Phys- RevLett.116.147202. [16] C. Reinhardt, T. Muller, A. Bourassa, and J. C. Sankey, "Ultralow-Noise SiN Trampoline Resonators for Sensing and Optomechanics," Phys. Rev. X, vol. 6, art. no. 021001, 2016, doi: 10.1103/Phys- RevX.6.021001. [17] M. J. Weaver, B. Pepper, F. Luna, F. M. Buters, H. J. Eerkens, G. Welker, B. Perock, K. Heeck, S. de Man, D. Bouwmeester, " Nested trampoline resonators for optomechanics" Appl. Phys. Lett. 108, 033501 (2016); https://doi.org/10.1063/1.4939828 [18] L.G. Villanueva, S. Schmid, "Evidence of Surface Loss as Ubiquitous Limiting Damping Mechanism in SiN Micro- and Nanomechanical Resonators," Phys. Rev. Lett., vol. 113, art. no. 227201, 2014, doi: 10.1103/PhysRevLett.113.227201. [19] A. Borrielli, L. Marconi, F. Marin, F. Marino, B. Morana, G. Pandraud, et al., "Control of recoil losses in nanomechanical SiN membrane resonators," Phys. Rev. B, vol. 94, art. no. 121403(R), 2016, doi: 10.1103/PhysRevB.94.12140. [20] E. Serra, M. Bawaj, A. Borrielli, G. Di Giuseppe, S. Forte, N. Kralj, "Microfabrication of large-area circular high-stress silicon nitride membranes for optomechanical applications," AIP Advances, vol. 6, art. no. 065004, 2016, doi: 10.1063/1.4953805. [21] P.-L. Yu, T. P. Purdy, C. A. Regal, "Control of Material Damping in High-Q Membrane Microres- onators," Phys. Rev. Lett., vol. 108, art. no. 083603, 2012, doi: 10.1103/PhysRevLett.108.083603 [22] Philip M. Morse, "Vibration and Sound," McGraw-Hill (New York, 1948), pp. 133-147. [23] P.R. Saulson, "Thermal Noise in Mechanical Experiments," Phys. Rev. D, vol. 42, p. 2437, 1990, doi: 10.1103/PhysRevD.42.2437. [24] L. Meirovitch, Dynamics and Control of Structures (John Wiley & Sons, 1990). [25] Yu. Levin, "Internal thermal noise in the LIGO test masses: A direct approach," Phys. Rev. D, vol. 57, p. 659, 1998, doi: 10.1103/PhysRevD.57.659. [26] X. Liu, T. H. Metcalf, Q. Wang and D. M. Photiadis, "Elastic Properties of Several Silicon Nitride Films," Mater. Res. Soc. Symp. Proc. vol. 989, p. A22-01, 2007, doi: 10.1557/PROC-0989-A22-01. [27] E. Serra, A. Borrielli, F.S. Cataliotti, F. Marin, F. Marino, A. Pontin, G.A. Prodi, and M. Bonaldi, "A "low-deformation mirror" micro-oscillator with ultra-low optical and mechanical losses," Appl. Phys. Lett., vol. 101, art. no. 071101, 2012, doi: 10.1063/1.4745510. [28] A. Borrielli, A. Pontin, F. S. Cataliotti, L. Marconi, F. Marin, F. Marino, G. Pandraud, G. A. Prodi, E. Serra, and M. Bonaldi, "Low-loss optomechanical oscillator for quantum optics experiments," Phys. Rev. Applied, vol. 3, art. no. 054009, 2015, doi: 10.1103/PhysRevApplied.3.054009.
1905.02060	1	1905	2019-05-06T14:31:00	Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry	[ "physics.app-ph" ]	Here we introduce a versatile stereolithographic route to produce three different kinds of Si-containing thermosets that yield high performance ceramics upon thermal treatment. Our approach is based on a fast and inexpensive thiol-ene free radical addition that can be applied for different classes of preceramic polymers with carbon-carbon double bonds. Due to the rapidity and efficiency of the thiol-ene click reactions, this additive manufacturing process can be effectively carried out using conventional light sources on benchtop printers. Through light initiated cross-linking, the liquid preceramic polymers transform into stable infusible thermosets that preserve their shape during the polymer-to-ceramic transformation. Through pyrolysis the thermosets transform into glassy ceramics with uniform shrinkage and high density. The obtained ceramic structures are nearly fully dense, have smooth surfaces, and are free from macroscopic voids and defects. A fabricated SiOC honeycomb was shown to exhibit a significantly higher compressive strength to weight ratio in comparison to other porous ceramics.	physics.app-ph	physics	Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. Additive Manufacturing of Ceramics from Preceramic Polymers: A Versatile Stereolithographic Approach Assisted by Thiol-Ene Click Chemistry Xifan Wang1, Franziska Schmidt1, Dorian Hanaor1, Paul H. Kamm2, Shuang Li3, and Aleksander Gurlo1 correspondence to: [email protected] 1 Fachgebiet Keramische Werkstoffe / Chair of Advanced Ceramic Materials, Technische Universität Berlin, Hardenbergstr. 40 BA3, Berlin, 10623, Germany. 2 Helmholtz-Zentrum Berlin für Materialien und Energie, Institut für Angewandte Materialforschung / Institute of Applied Materials, Hahn-Meitner-Platz 1, Berlin, 14109, Germany 3Functional Materials, Department of Chemistry, Technische Universität Berlin, Hardenbergstr. 40 BA2, Berlin, 10623, Germany Abstract Here we introduce a versatile stereolithographic route to produce three different kinds of Si-containing thermosets that yield high performance ceramics upon thermal treatment. Our approach is based on a fast and inexpensive thiol-ene free radical addition that can be applied for different classes of preceramic polymers with carbon-carbon double bonds. Due to the rapidity and efficiency of the thiol-ene click reactions, this additive manufacturing process can be effectively carried out using conventional light sources on benchtop printers. Through light initiated cross-linking, the liquid preceramic polymers transform into stable infusible thermosets that preserve their shape during the polymer-to-ceramic transformation. Through pyrolysis the thermosets transform into glassy ceramics with uniform shrinkage and high density. The obtained ceramic structures are nearly fully dense, have smooth surfaces, and are free from macroscopic voids and defects. A fabricated SiOC honeycomb was shown to exhibit a significantly higher compressive strength to weight ratio in comparison to other porous ceramics. Keywords Additive manufacturing; stereolithography; polymer derived ceramics; silicon oxycarbide; compressive strength 1. Introduction and various In industrial applications, ceramics are most commonly formed by slip casting, injection molding pressure-assisted techniques. As all of these methods rely on the usage of molds, the shape of ceramic parts is restricted to relatively simple geometries. The machining of complex shapes is further hindered by the high hardness and brittleness of ceramic materials. Additive manufacturing (AM) heralds a new era in the fabrication of ceramics. As AM technologies are based on the layer by layer consolidation of ceramic powders, they neither require molds nor are they limited by the hardness and brittleness of ceramics. This in turn allows for the fabrication of complex parts that utilize the exceptional properties of ceramics such as high strength, thermal stability and chemical resistance [1,2]. Traditional AM approaches towards ceramics rely on powder- and slurry-based technologies to task due including stereolithography of ceramic slurries containing UV-curable photopolymers [3,4], laser sintering of ceramic powder beds [5,6] and binder jetting, where a liquid organic binder is injected into a ceramic powder bed [7,8]. Forming dense objects by consolidating ceramic powders without pressure is a challenging the unavoidable presence of pores and cracks that impair the mechanical performance. For the fabrication of dense ceramics, formulations with high solid loadings are required, which consequently restrict the suitability of slurries for the manufacturing of fine structures due to their decreased flow ability, increased viscosity and inhomogeneity. When shaping ceramic slurries with light, a significant mismatch between the refractive index of the ceramic powder and that of the photocurable resin strongly reduces the curing depth and additionally causes a coarsening of the curing resolution due to the 1 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. side [9]. Consequently, it is far more challenging to 3D print 'dark' ceramic materials (Si3N4, SiC) relative to alumina, silica or zirconia. backbone. Despite scattering of incident light attaching functional acrylate groups to the silicone significant developments in the field, the full potential of PDC stereolithography has not yet been realized. The range of readily photocurable preceramic polymers remains quite confined and has thus far been limited to polysiloxanes, consequently only SiOC ceramics have been additively manufactured scale [2,17,18,20 -- 22]. A recent study examined 3D printing of SiC by blending acrylate based photopolymers with SiC precursors [23]. However, oxygen contamination from the photopolymers is unavoidable and is likely to largely degrade the product material's thermomechanical Another disadvantage of this route is the extremely low ceramic yield the mixed photopolymers are completely eliminated during pyrolysis. in macro properties. [20,23], as such ceramic compositions Polymer-derived-ceramics (PDCs) define a class of ceramic materials that are synthesized by thermal treatment (usually pyrolysis) of ceramic precursors (so-called preceramic polymers) under an inert or reacting atmosphere [10]. By using suitable preceramic polymers, various as amorphous silicon carbide (SiC), silicon oxycarbide (SiOC), and silicon carbonitride (SiCN), can be obtained after heat treatment at 1000-1100 °C in an inert atmosphere (argon or nitrogen). Because there is no sintering step, PDC parts can be formed without pressure at lower traditional ceramic technologies. Preceramic polymers can be processed using existing technologies suitable for polymers in general. Due to their distinctive nanostructure of carbon-rich and free carbon domains, PDCs show exceptional stability against oxidation, crystallization, phase separation and creep even up to 1500 °C [10]. temperatures relative shaping powder to light, which makes Currently, PDCs have been successfully utilized for the fabrication of ceramic fibers, ceramic matrix composites (CMCs) and microstructures, e.g. microelectromechanical systems (MEMS). Since some preceramic polymers contain silyl and vinyl groups, they can act as negative photoresists, exploiting a UV activated hydrosilylation reaction. For this reason, photolithography is frequently used to achieve fine two-dimensional patterning in PDC MEMS [11 -- 16]. However, UV activated hydrosilylation requires adding large amounts of photoinitiators (PI) and long exposures to high energy UV its utilization unrealistic in additive manufacturing processes. To overcome this limitation certain polysiloxanes can be modified by the attachment of curable functional groups to the silicone backbone in order to render them UV active and facilitate rapid photopolymerization. The additive manufacturing of macro-sized complex SiOC parts with excellent mechanical strength and stability against high temperature oxidation by stereolithography of preceramic polymers was first reported in [17,18]. In the first approach, two silicones terminated with vinyl and mercapto functional groups were blended trigger cross-linking [18,19]. In the second approach [17], to obtain a for stereolithography, a commercial polysiloxane (Silres ® MK Powder) was modified by photocurable polymer together suitable to Classical radical photopolymerization pathways rely mainly on the addition reaction of carbon double bonds in acrylate or methacrylate, and they have been widely used in photolithography and stereolithography processes. Acrylate or methacrylate based photopolymerization routes suffer from fundamental limitations, including high levels of polymerization-induced volume shrinkage and consequent stress development, and oxygen inhibition of the cross-linking reaction. An alternative pathway involves thiol- ene free radical addition, where SH groups undergo addition reactions with carbon-carbon double bonds. This polymerization route is considered as a "click reaction" and therefore exhibits the following characteristics [24]: (a) a high reaction yield (b) fast reaction rate, (c) solvent parameters insensitivity (d) negligible volumetric shrinkage, and (e) insensitivity towards oxygen and water. These thiol-ene click overall properties chemistry a promising approach towards photopolymerization in general and additive manufacturing in particular. It has been shown that the preceramic polymer polycarbosilazane, can be photopolymerized with thiol-ene reaction in 2D geometries and converted to silicon carbonitride based material upon pyrolysis [13]. render to using this work, we introduce a versatile In route, based on VAT stereolithographic photopolymerization, and inexpensive thiol-ene click reactions to 3D print Si-containing are subsequently pyrolyzed to obtain high performance Si-containing ceramics. Our approach is applied for different classes of pr- ceramic polymers containing carbon-carbon thermosets. These fast 2 bonds, including Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. double polysiloxane, polycarbosilane and polycarbosilazane with side vinyl groups. Due to the rapidity and efficiency of thiol-ene click chemistry reactions used here, the AM process can be effectively performed with conventional light sources (such as projectors) on benchtop Digital Light Processing (DLP) printers. Due to their high optical transparency, which minimizes the scattering effects, resin mixtures are further applicable for more sophisticated techniques such as two-photon polymerization [25,26] and microstereolithography [27]. (2,4,6-trimethylbenzoyl) oxide (BAPOs) as photoinitiator (PI), Sudan Orange G as photoabsorber and hydroquinone as free radical scavenger. In order to homogenize the resin, it was treated in an ultrasonic bath for 2 hours, followed by degassing under vacuum. After that, 1,6-hexanedithiol (2T), which provides the thiol groups to react with the alkene functionalities of preceramic polymers, was dissolved in the resin mixture via mild stirring for 1 hour. The prepared resin was subsequently stored in a brown glass bottle to avoid exposure to light. phosphine 2. Materials and methods 2.1 Materials: SPR-212) The preceramic polymers used in this work are liquid methylvinylhydrogen polysiloxane a (polyramic and (StarPCSTM allylhydrydopolycarbosilane SMP10), bought from Starfire Systems (USA), as precursors for SiOC and SiC ceramics respectively. liquid For polycarbosilazane methylvinylhydrogen (Durazane 1800) bought from Merck, Germany) was selected. 1,6-hexanedithiol, phenylbis(2,4,6-trimethylbenzoyl)phosphine oxide (BAPOs), Sudan Orange G, and hydroquinone were purchased from Sigma Aldrich , Germany. All chemicals were used without further purification. SiCN, a Formulation of photopolymerizable preceramic resins for stereolithography was conducted as follows: The preceramic resin was prepared by mixing preceramic polymers with phenylbis spectrum of relevant working Upon exposure to light with wavelengths below 450 nm, which are included in the emitted white light from the projector, the PI undergoes a cleavage process generating free radicals (see Fig. S1 in SI), which then initiate the thiol-ene click reaction as seen in Fig. 1. Sudan Orange G, which exhibits suitable optical absorbance in the the photoinitiator, was utilized to modify the sensitivity of prepared resin and therefore limit the penetration depth of the emitted light. As a result, the z-axis resolution can be controlled. Detailed information of resin sensitivity and critical exposure time can be found in Fig. S2 of the provided information. Hydroquinone acts as free radical scavenger to avoid any unwanted photopolymerization induced by background light or scattering, and was found to significantly prolong the shelf life of the resin. supplementary Fig. 1. Schematics of the photoinduced thiol-ene click reactions (cross-linking) of vinyl-containing Si- based preceramic polymers with 1,6-hexanedithiol via an anti-Markovnikov addition. 3 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. 2.2 Stereolithography and thermal pyrolysis: Stereolithography of preceramic polymer resins was conducted in an Open 3D Resin Printer LittleRP2 (LittleRP, USA) using an Acer X152H projector as a light source. The lens system of the projector was modified for short distance resolution, namely the layer thickness of each print, was set at 100 µm to enable rapid printing. As a result, objects with 2 cm height could be fabricated in approximately one hour. The printing parameters including the exposure time are focusing. The z-axis those fabricated anhydrous reported in Table S1. After printing, the parts fabricated from polysiloxane were washed with isopropanol, while from polycarbosilane and polycarbosilazane were washed with cyclohexane. Subsequently parts were post cured under exposure to UV irradiation between 350 and 400 nm for 30 mins. To transform these parts into ceramics they were pyrolyzed in a tubular furnace at 1100 °C for 2 hours under flowing nitrogen. The heating/cooling rate was set to 40 °C /h. A schematic illustration of the complete printing process is shown in Fig. 2. Fig. 2. Additive manufacturing of ceramics from preceramic polymers: schematic representation of thiol-ene click chemistry based stereolithography. (A) The preceramic photopolymerizable resin was prepared by mixing different preceramic polymers (polysiloxanes, polycarbosilane, polycarbosilazane) with 1,6-hexanedithiol dithiol along with phenylbis (2,4,6-trimethylbenzoyl) phosphine oxide (BAPOs) as photoinitiator (PI), Sudan Orange G as photoabsorber and hydroquinone as free radical scavenger (for more details, see Materials and Methods). (B) Photopolymerization of preceramic resins in a benchtop DLP printer utilizing a commercially available projector as a light source. (C) A representative printed cross-linked thermoset and the resultant glassy ceramic formed upon pyrolysis at 1100 °C under N2 atmosphere. 2.3 Characterization: Fourier transform infrared spectroscopy (FTIR) in Attenuated Total Reflection (ATR) mode carried out in an Equinox 55 (Bruker, Germany) in the range of 500 -- 4000 cm-1 was used to characterize the free radical initiated thiol-ene polymerization as well as the polymer to ceramic transformation. In-situ Differential Scanning Calorimetry (DSC) under exposure to UV light in the range between 350 and 600 nm (100 mW/s) was utilized in a DSC Q2000 (TA Instruments) to characterize the heat flow and reaction kinetics during photopolymerization. X-ray photoelectron spectroscopy (K-Alpha TM, Thermo Scientific) was used to determine the elemental composition of printed parts. The polymer to ceramic conversion was investigated in nitrogen with a heating rate of 10 °C min-1 with thermal gravimetry/differential thermal analysis (TG/DTA) on STA 409 PC LUXX (Netzsch, Germany) coupled with a mass spectrometer OMNi Star GSD 320 (Pfeiffer Vacuum, Germany). Solid-state 29Si NMR spectra was recorded with a Avance 400 MHz spectrometer (Bruker, Germany) operating at 79.44 MHz the chemical to characterize using performed ceramics. Transmission environment of silicon in polysiloxane derived SiOC electron microscopy (TEM) was used to prove the amorphous nature of SiOC. X-ray imaging (radiography) was a microfocus X-ray source and a flat panel detector (Hamamatsu, Japan) with an area of ~ 120 x 120 mm² and a pixel size of 50 µm as reported in [28]. Sample magnification was set between 6 and 10, depending on the size of the sample, resulting in an acquired projection with a pixel size between 5 and 10 µm. For the tomographic acquisition 1000 projections were taken over a sample rotation of 360°. The printing quality and microstructure of 3D printed SiOC ceramics was evaluated by Scanning Electron Microscopy in Leo Gemini 1530 (Zeiss, Germany). 2.4 Mechanical properties: Mechanical tests were conducted on SiOC honeycombs additively manufactured from polysiloxane SPR212 and pyrolyzed at 1100 °C for 2 hours under flowing nitrogen. Nanoindentation of SiOC honeycomb that had been embedded in epoxy resin and polished to a 1 µm finish, was conducted using a TI 950 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. TriboIndenter (Hysitron Inc., USA) with a tetrahedral Berkovich diamond tip at a load of 10 mN and a load rate of 50 µN/s. The elastic modulus and hardness are calculated by evaluating the slope of the curve dP/dh upon the elastic unloading with 200 indents. The compressive strength of SiOC honeycombs was determined using a universal test machine RetroLine (Zwick/Roell, Germany) with a cross-head speed of 0.1 mm/min. For this test, samples were first ground to achieve relatively parallel surfaces and a thickness of 2.78 mm. These were then sandwiched between two steel face sheets with epoxy glue to prevent horizontal movement. The solid cell wall density (ρs) of SiOC honeycombs was determined via Archimedes' method using distilled water, while the cellular density of this honeycomb (ρ) was calculated according to [29] by dividing the foam mass by its volume, which is the product of its length (a), width (b) and height (c). To demonstrate the versatility of our approach we applied three commercially available Si- containing polymers from three different classes as precursors for three different ceramic compositions, i.e. polysiloxane SPR212 as precursor for SiOC, polycarbosilane SMP10 as precursor polysilazane Durazane1800 as precursor for SiCN. The thiol- ene click reaction assisted AM process is schematically illustrated in Fig. 2. Preceramic polymers mixed with 1,6-hexanedithiol and BAPOs (PI) are cross-linked in less than 7 seconds using a commercial digital light processing (DLP) based projector. initiate SiC and for the that Upon exposure to the white light, the PI undergoes a cleavage process generating free radicals thiol-ene polymerization reactions shown in Fig. 1 in which thiol groups react with the alkene groups of preceramic polymers leading to a thioether linkage. Because of this propagating cross- linking, liquid preceramic polymers transform into robust infusible thermosets that are essential for the preservation of 3D printed geometries during subsequent processing including washing and pyrolysis. Insufficient cross-linking would lead to soft structures that would deform prior the ceramization process. to or during the 3. Results and discussion 3.1 Thiol-ene click chemistry assisted stereolithography Fig. 3. (A-C) ATR-FTIR spectra of preceramic resins before (black) and after photoinduced curing (red): (A) with polysiloxane SPR212, (B) polycarbosilane SMP10 and (C) polycarbosilazane Durazane1800. The thiol-ene click polymerization is confirmed by the strong reduction of the absorption bands of the vinyl-groups. (D) In-situ differential scanning calorimetric characterization of preceramic resins exposed to light demonstrating that the preceramic resins are effectively cured within 15 secs, which makes them suitable for stereolithographic additive manufacturing: (black) the curing reaction heat and (blue) the conversion rate (the integral curve of heat flow after normalization) . 5 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. The evolution of the thiol-ene click reaction can be observed using FTIR spectroscopy by monitoring the intensity of bands associated with vinyl groups at 1407 cm-1 (=CH2 scissoring) and 1597 cm-1 (C=C stretch) and with thiol groups at 2570 cm-1. Fig. 3 A-C display the FTIR spectra of SPR212, SMP10 and Durazane1800 polymers before and after the thiol-ene assisted photopolymerization. Because of the low concentration of thiols and relatively weak absorption associated with the S-H groups, the thiol peaks are not observed in the UV the FTIR spectra even before irradiation. Nonetheless, significant decrease of the intensity of the vinyl absorption bands in the FTIR spectra as well as the elimination of the strong thiol odor indicate a successful in all polymers exposed to light. The appearance of the C=O absorption band in SMP10 based resin is induced from the addition of BAPOs. However, calorimetric characterization under UV irradiation reveal significant differences between the polymers studied in this work. thiol-ene click reaction real-time in-situ the As shown by the black lines in Fig. 3D, the polysiloxane SPR212 as well as polysilazane Durazane1800 polymers exhibit strong and sharp exothermic peaks immediately after UV exposure, which indicate very high curing efficiencies for these polymers. The normalized integral of these exothermic peaks shows that more than 50% of the resin mixture can be cross-linked within 5 secs, which is an ideal time frame for the AM process, where fast and efficient reaction kinetics are required. In contrast, the polymerization of the SMP10 polymer was significantly slower and was accompanied by smaller heat evolution. This is because SMP10 has far fewer vinyl groups compared to SPR212 and Durazane1800, which leads to a lower degree of photoinduced cross- linking. Furthermore, the SMP10 polymer absorbs light in the blue region of the visible spectrum, as evident by its intrinsic orange color, resulting in a slower photopolymerization process. This difference has implications for the ceramization process as discussed below. Nonetheless, all preceramic polymers studied here can be applied in the stereolithographic fabrication of complex thermoset parts as demonstrated in Fig. 4. Fig. 4. Representative examples of thermoset parts fabricated from three different polymer classes: (A) Kelvin cell structure printed with polysiloxane SPR212, (B) cog wheel printed with polycarbosilane SMP10 and (C) turbine structure printed with polycarbosilazane Durazane1800. 3.2 Polymer-to-ceramic transformation and the additively manufactured ceramic components microstructure of they were subjected To transform the cross-linked thermosets into ceramics, to further processing steps. First, they were rinsed with organic solvents (isopropanol for polysiloxane and anhydrous cyclohexane for polycarbosilane and polycarbosilazane), the remaining non-polymerized resin, and post cured under UVA light between 350 -- 400 nm to enhance their mechanical stability. In the next step they were transferred to a tube furnace and pyrolyzed in N2 at 1100 °C. During pyrolysis the polymers into amorphous Si-containing decompose cross-linked preceramic remove to ceramics; this transformation is accompanied by the release of hydrogen, methane and other carbon- and hydrogen-containing species. This process entails mass loss, isotropic volume shrinkage and densification of the material. Upon the pyrolytic conversion of cross-linked polymers to the corresponding ceramics, a linear shrinkage of roughly 30 % is observed (see Fig. 2C). Since the shrinkage is isotropic, the thermoset is transformed into a ceramic without any noticeable distortion. The preservation of geometry demonstrates that thiol-ene click reactions are an effective and easily accessible pathway towards the AM fabrication of ceramic parts with excellent resolution and smooth surfaces from preceramic Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. polymers. To avoid cracking or explosion during pyrolysis, the geometry of printed preceramic polymers is designed with a feature size smaller than 2 mm in at least one direction. This reduces the diffusion path of evolving gas and facilitates the shrinking process. indicates the polymer flaws were not observed ranging from cellular lattices to bulk structures - fabricated by our approach. Substantial internal in corresponding radiography images (Fig. 5 E - H), which validates our approach towards AM of high performance ceramics. The SiOC and SiCN(O) parts were successfully fabricated with excellent resolution and very smooth, pore free surfaces. Surprisingly, SiC parts fabricated from the polycarbosilane SMP10 cracked and exploded during pyrolysis when the heating temperature reached 600 °C. At temperatures between 550 and 800 °C, SMP 10 transforms into an inorganic SiC based material accompanied by release of hydrogen (H2) and methane (CH4), as well as volumetric shrinkage, which to ceramic conversion [30 -- 32]. As discussed before, relative to SPR212 and Durazane 1800, the SMP10 polymer was cross-linked at a lower degree and more importantly its polymer structure contains greater amounts of silyl (Si- H) and carbosilane (Si-CH2-Si) groups, leading to greater hydrogen gas evolution during pyrolysis. Therefore, the significant hydrogen and methane gas egression during the ceramization process of SMP10, along with shrinkage stresses, resulted in the formation and rapid propagation of cracks manifesting in the explosion of samples (i) an Using the procedures developed in the present work, further ceramic compositions beyond the SiOC, SiC and SiCN materials studied here can be obtained by means of suitable preceramic polymers. General guidelines for the selection of appropriate ceramic precursors for use with thiol-ene click reactions can be stated as follows: inorganic-organic hybrid polymer or monomer, where the inorganic core comprises species (such as Al, B, Si, Ti or Zr) that form stable oxides, carbides, nitrides or oxycarbides upon pyrolysis; (ii) the presence of unsaturated carbon double or triple bonds, such as those in vinyl groups; (iii) the propensity to form thiyl radicals through photoinitiation. Furthermore, the methodology developed here can be easily adapted to produce composite ceramics containing fillers to impart catalytic, photocatalytic or magnetic functionalities. In this work we were able to fabricate complete black glassy ceramics composed of silicon oxycarbide and silicon carbonitride. Fig. 5 A-D displays several examples of ceramic parts - . Fig. 5. Examples of ceramic components fabricated by pyrolyzing the cross-linked thermosets showing (A,B) Kelvin cell structure and turbine impeller of SiOC and (C,D) cog wheel and turbine structure of SiCN(O). Corresponding radiography images (E-H) demonstrate the fabricated ceramic components are free from macro sized voids and defects. 7 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. (m/z= 34) due to the decomposition of thioether linkages during the major mass loss region and (B) TG curves indicating the overall weight loss. resins larger is significantly The polymer to ceramic transformation was examined by thermal gravimetry coupled with mass-spectrometry, with results shown in Fig. 6. The overall weight loss of dithiol modified SPR212 in comparison to the weight loss of the unmodified preceramic polymer, while the differences in SMP10 and Durazane 1800 resins are less significant [33,34]. The additional weight loss is attributed to the evaporation and breakdown of thioether linkages during pyrolysis. As seen in Fig. 6 A, a substantial release of sulfur - containing low molecular weight species is observed between 400 and 600 °C. In contrast, in SMP10 and Durazane 1800 resins, oxidation in air leads to the formation of Si-O and Si-OH groups, which transform into thermally stable Si-O-Si bonds. The resultant ceramic yield is thus not strongly affected by the presence of dithiol compound. Fig. 6. STA characterization of the polymer-to- ceramic transformation of the cross-linked thermosets fabricated from SPR212, SMP10 and Durazane 1800: (A) selected MS data for SPR212 indicating the release of SO2 (m/z= 64), CH3SH (m/z= 48), C4H8 (m/z= 56) and H2S Fig. 7. Characterization of additively manufactured polymer-derived ceramic materials: (A) ATR-FTIR and (B) X-ray photoelectron spectra of SiOC (red), SiC(O) (green) and SiCN(O) (blue line) synthesized at 1100°C from SPR212, SMP10 and Durazane 1800, respectively. Analysis of SPR212-derived SiOC ceramic pyrolyzed at 1100 °C: (C) TEM image, with selected area electron diffraction (SAED) pattern shown in inset indicating a homogeneous amorphous structure (D) solid state 29Si NMR spectra, revealing mixed bonding between Si, O and C. Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. Table 1. Elemental composition of ceramics pyrolyzed at 1100 °C in flowing nitrogen Preceramic polymer Polysiloxane SPR212 Polycarbosilane SMP10 Polycarbosilazane Durazane 1800 Fabricated ceramic SiC1.39O1.53S0.01 Elemental composition, at. % Si 25.48 O 38.88 C 35.47 SiC1.22(O0.78)S0.04 33.00 25.77 40.04 N 0 0 S 0.17 1.19 SiC1.02N0.59(O1.12) S0.04 28.52 28.92 26.22 15.27 1.07 As verified by FTIR and X-ray photoelectron spectroscopy and TEM characterization (Fig. 7), parts pyrolyzed at 1100 °C in flowing nitrogen are amorphous silicon-containing ceramics with a general formula of SiCxOyNz with small amount of residual sulfur. Their the chemical preceramic the manufacturing (Table 1). composition depends on for polymer applied From XPS data, it is clear that ceramics with targeted compositions of SiC and SiCN possess significant oxygen content and these materials further demonstrate broad absorption bands around 1100 cm-1 in the FTIR spectra attributed to Si-O-Si stretching vibrations. Since the additive manufacturing process was performed in ambient air, the silane (-Si-H) and silazane (Si-N-Si) bonds contained in the preceramic polymers reacted with atmospheric oxygen and moisture rapidly leading to an increased oxygen content. Consequently, "reacted" SMP10 and Durazane 1800 polymers are thermally converted into SiC(O) and SiCN(O) ceramics with deviated properties and microstructures compared to those in the intrinsic oxygen free SiC and SiCN ceramics. To minimize oxidation in stoichiometric SiC and SiCN ceramics, the entire fabrication process would have to be conducted under an inert atmosphere. We thus focus on the representative SiOC ceramics fabricated from polysiloxane SPR212. As described above and as verified by TEM, the SiOC formed by pyrolysis at 1100 °C in flowing nitrogen is amorphous. Further insight into the SPR212-derived SiOC microstructure is 29Si-NMR provided solid-state by that characterization (Fig. 7 D). The characteristic peaks at chemical shifts of -114 and -14 ppm correspond to SiO4 and SiC4 units and the peaks at -82 and -39 ppm can be assigned to mixed- bond SiO3C and SiO2C2 units [35]. The fabricated SiOC parts exhibited a measured Archimedes density of 2.10 g/cm3, which approaches 97 % of the calculated theoretical density of. 2.17 g/cm3 assuming densities of 2.2, 3.1 and 1.45 g/cm3 for amorphous SiO2, carbon, amorphous SiC and pyrolytic respectively [36]. This confirms the fabricated SiOC parts are nearly fully dense with negligible pore volume. This is essential for ceramic materials, since cracks and pores act as stress concentrators under load and initiate failures. As can be seen in the radiography images (Fig. 5), SiOC components are free from observable internal cracks and closed pores. 3- D printed cellular cubes of SiOC are completely dense and free of macro sized defects as confirmed by the homogeneous absorption of X-rays in the tomographic images of the diamond lattice (Fig. 8 A) and Kelvin cell (Fig. 8 E) structures. The corresponding SEM images (Fig. 8 B-D and F-H) demonstrate again the delicate microstructures and fine features which are achievable only in additive manufacturing. Along the printing direction, step-like surfaces are clearly visible which is related to the layer-by-layer printing process. In this work the layer thickness during printing was set to be 100 µm to yield rapid fabrication. At higher magnification, glassy smooth surfaces are observed, exhibiting no porosity. This is consistent with the radiography images, indicating the high quality fabrication of ceramic components. 9 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. Fig. 8. Morphology and microstructure of additively manufactured SPR212-dervied SiOC: (A and E) tomographic images of the diamond lattice (A) and Kelvin cell (E) structures. The delicate design is perfectly replicated without any detectable flaws. (B-D, F-H) Corresponding SEM images at higher magnification showing the step-like surfaces as well as the crack free struts. 3.3 Mechanical properties The hardness and elastic modulus of the SiOC ceramics were evaluated in nanoindentation experiments (Fig. 9). Fig. 9A shows the 3D printed honeycomb used for measurements along with an AFM image of an indent. A typical load-hold-unload curve is shown in Fig. 9B. 200 measurements were conducted to evaluate the reduced elastic modulus and hardness, as given in Fig. 9 C and D. Over a range of penetration depths, the mean values of the reduced Young's modulus and hardness are found to be 106±6 and 12±1 GPa, respectively. An earlier study showed that SiOC ceramics with lower O/C ratios behave similarly to amorphous-SiC materials, while the mechanical properties of SiOC with higher O/C ratio are similar to those in amorphous SiO2[37]. In the present work, the hardness and elastic modulus of AM fabricated SiOC materials lie between the predicted limit of amorphous SiO2 [37] and amorphous SiC [38] and are further in good agreement with reported values for silicon oxycarbide glasses with variable C/O ratio [37,39]. Fig. 9. Results of nanoindentation tests performed on 2D honeycombs of additively manufactured SPR212-derived SiOC. (A) A polished SiOC honeycomb imbedded in epoxy resin. Inset shows the AFM image of one indent. (B) Load-Hold-Unload curve of one indent. Elastic modulus is determined via the elastic unloading process. (C, D) Reduced modulus and hardness as a function of the penetration depth. 200 measurement results are shown here. 10 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. It is worth emphasizing that our fabrication method resulted in SiOC ceramics with comparable or even slightly better mechanical properties than those of SiOC glasses fabricated either via hot-pressing or via radio frequency (RF)-magnetron sputtering. Both of these methods yield high surface quality SiOC components films) with minimized defects and pore volume. (bar and thin strength Since our manufacturing methods allow for the fabrication of 2D and 3D periodic cellular structures, in the following step we evaluate the out-of-plane compression strength of 2D SiOC honeycombs as a representative example of a material with high to density characteristics, and apply Gibson and Ashby's model [29] to analyze these results. The honeycomb geometry is particularly strong when loaded along the main axis of the hexagonal prism (out-of-plane, X3) compared to the in-plane directions (X1 and X2), as thin cell walls are much stiffer under extension or compression in bending. Therefore, honeycombs are mostly used as cores in sandwich panels to provide high strength to weight ratios. Likewise, honeycombs made from ceramic materials are frequently used as catalyst supports, filtration membranes and components in heat exchangers due to their superior chemical and mechanical stability in harsh environments. than Fig. S3 shows the corresponding stress-strain curves. 2D SiOC honeycomb structures fabricated in this work were found to exhibit a foam density of 0.61 g/cm3 and a compressive strength of 216 MPa as seen in Table S2 and Fig. S3. The SiOC honeycomb demonstrates linear-elastic behavior as well as catastrophic failure, which is accompanied by a sudden drop in the measured stress. The plateau presented in the stress-strain curves before failure is resulted from slight misalignments of load bearing faces in the fabricated lattices, which in an ideal structure should be perfectly parallel, as well as an edge effect that, cell walls from internal to external can't fail simultaneously. Therefore, in our experiments the compressive strength is taken as the value at which this plateau occurs rather than the maximum crushing stress. relates Gibson and Ashby's model the mechanical properties of a cellular solid to its relative density ρrelative (the density ρ of the foam divided by the density ρs of the solid which the foam is made of) and its solid strength/bulk modulus (σs ,Es). The strength (σ) as well as the modulus (E) of the cellular material can then be expressed as follows: = 𝐶 ( 𝑚 ) (1) 𝜎 𝜎𝑠 𝐸 𝐸𝑠 𝜌∗ 𝜌𝑠 𝜌∗ 𝜌𝑠 𝑚 ) = 𝐶 ( (2) where C is a dimensional constant and the exponent m depends on the cell morphology. In the case of a 2D honeycomb carrying loads on the faces normal to X3, equation (1) and (2) can further be simplified to (3) [40]: 𝐸 𝐸𝑠 = 𝜌∗ 𝜌𝑠 𝑎𝑛𝑑 𝜎 𝜎𝑠 = 𝜌∗ 𝜌𝑠 (3) In this case the Young's modulus of the 2D honeycomb simply reflects the solid modulus scaled by the load-bearing area. Using equation (3) with calculated relative density of 0.29 and the Young's modulus of 106±6 GPa from the nanoindentation experiments, we obtain 31±2 GPa for the theoretical honeycomb stiffness. Because of the measurement error, which occurred when evaluating the true strain of SiOC honeycomb under compression, we are unable to compare the measured Young's modulus with this calculated theoretical value here. More detailed explanation can be found in SI with Table S2 and Fig S4. Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. Fig. 10. Strength-density Ashby chart for designing light strong structures. In this chart our additively manufactured SiOC honeycombs are compared to other ceramic (i.e. SiOC, Al2O3, SiC) honeycombs as well as to other natural and technical materials. Figure and material properties are generated by Nicoguaro using (https://commons.wikimedia.org/wiki/File:Material- comparison--strength-vs-density_plain.svg) except the data points of different ceramic honeycombs, which were inserted manually from the corresponding references. JavaScript and D3.js In Fig. 10, SiOC honeycombs printed in this work are compared, in terms of density and compressive strength, to natural and engineered cellular solids as well as other honeycombs made from SiOC [18], SiOC nanocomposites [41], Al2O3 [42 -- 44], and SiC [45]. The silicon oxycarbide honeycombs fabricated in this work display an outstanding strength to density ratio that is significantly higher than other materials of similar density. Its absolute strength is also comparable with fiber-reinforced polymers and metal alloys. The compressive strength of 216 MPa is comparable to that exhibited by α-Al2O3 honeycombs of approximately twice of the density and is around one order of magnitude higher than commercial SiC honeycombs. The silicon oxycarbide honeycombs fabricated in further demonstrate a higher this work compressive strength compared to SiOC honeycombs with the densities ranged from 0.5 to 0.8 g/cm3 reported in previous works (see e.g. [18] [41]). The exceptional performance of the additively manufactured SiOC honeycombs is the result of their dense and defect-free structure, which was observed by radiography and SEM characterization. As shown in equation (3), the strength of a cellular solid, whose cells walls display a stretch-dominated behavior, scales linearly with its relative density the [46]. This behavior is represented as a line with a slope of 1 in the double-logarithmic strength- density plot. Accordingly, theoretical compressive stress of cellular polymer derived SiOC ceramics can be defined by these lines passing through bulk silicon carbide, silicon nitride and silica, which thus delineate the maximum obtainable performance for PDCs [47]. As can be seen, SiOC honeycombs in this work with a density of 0.6 g/cm3 reach the theoretical compressive strength of silica based materials while approaching the predicted theoretical value of silicon carbide and silicon nitride. Shifting the ceramics' compositions towards higher C to O ratios by choosing polymers of higher carbon content and avoiding oxidation can further boost its specific strength towards the theoretical limit of SiC. By utilizing polysilazane in an oxygen free environment, silicon carbonitride ceramics can be obtained, offering towards high mechanical performance of printed cellular structures. avenues further 4. Conclusions click We have closely examined the utility of thiol- ene additive manufacturing using preceramic polymers. We demonstrated a versatile stereolithographic chemistry towards 12 [2] A. Zocca, P. Colombo, C.M. Gomes, J. Günster, D.J. Green, Additive Manufacturing of Ceramics, J. Am. Ceram. Soc. 98 (2015) 1983 -- 2001, 10.1111/jace.13700. J.W. Halloran, Freeform Fabrication of Ceramics via Stereolithography, Journal of the American Ceramic Society 79 (1996) 10.1111/j.1151- 2916.1996.tb09022.x. 2601 -- 2608, References [1] N. Travitzky, A. Bonet, B. Dermeik, T. Fey, I. Filbert‐Demut, L. Schlier, T. Schlordt, P. Greil, Additive Manufacturing of Ceramic‐Based Materials, Advanced Engineering Materials 16 (2014) 729 -- 754, 10.1002/adem.201400097. bonds and carbon in precursors [3] M.L. Griffith, Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. manufacturing route whereby high performance polymer-derived ceramics are obtained with well controlled geometry. The developed approach can be applied for the entire range of known PDC systems. For precursors to be used in thiol-ene click assisted stereolithography they should exhibit an inorganic backbone, unsaturated the photoinitiation of thiyl radicals. High levels of hydrogenated groups are detrimental as the evolution of a significant volume of hydrogen gas during pyrolysis can cause failure under certain firing conditions. The photocurable preceramic polymers of this method are easily prepared and can be applied in any SLA/DLP printer, as well as techniques like microstereolithography and two-photon polymerization to produce microstructures beyond the resolution limit of DLP techniques. The additively manufactured polysiloxane- derived SiOC components fabricated here are nearly fully dense, achieving 97 % of theoretical density. 2D SiOC honeycombs with a cellular density of 0.61 g/cm3 exhibit a compressive strength of 216 MPa, the performance of comparable porous ceramics and stretching the boundaries of material property space in terms of strength to weight ratio under compression. The presently developed methods represent a flexible and high-performance rapid additively manufactured polymer-derived ceramics that can find broad application in various engineering fields, particularly for load bearing materials applied in harsh environments and high temperatures. With appropriate AM techniques, ceramic components can be produced with the developed methods across a range of length-scales with high precision, representing a valuable new capability for industries such as aerospace, automotive, energy conversion and chemical engineering. surpassing towards route [8] B. Nan, X. Yin, L. Zhang, L. Cheng, Three‐ Dimensional Printing of Ti3SiC2‐Based Ceramics, Journal of the American Ceramic Society 94 (2011) 969 -- 972, 10.1111/j.1551- 2916.2010.04257.x. [4] T. Chartier, C. Chaput, F. Doreau, M. Loiseau, Stereolithography of structural complex ceramic parts, Journal of Materials Science 37 (2002) 3141 -- 3147, 10.1023/A:1016102210277. [5] P. Bertrand, F. Bayle, C. Combe, P. Goeuriot, I. Smurov, Ceramic components manufacturing by selective laser sintering, Applied Surface Science 989 -- 992, 10.1016/j.apsusc.2007.08.085. (2007) 254 [6] K. Shahzad, J. Deckers, Z. Zhang, J.-P. Kruth, J. Vleugels, Additive manufacturing of zirconia parts by indirect selective laser sintering, Journal of the European Ceramic Society 34 (2014) 81 -- 89, 10.1016/j.jeurceramsoc.2013.07.023. [7] H. Seitz, W. Rieder, S. Irsen, B. Leukers, C. Tille, Three‐dimensional printing of porous ceramic scaffolds for bone tissue engineering, Journal of Biomedical Materials Research Part B: Applied Biomaterials 74B (2005) 782 -- 788, 10.1002/jbm.b.30291. Acknowledgements: The research was supported by the Deutsche Forschungsgemeinschaft under the grant "GU 992/17-1". We thank Jörg Bauer for helping with UV-DSC, Anke Maerten for conducting the nanoindentation measurements and Jan Epping for conducting the solid-state NMR experiments, respectively. We also thank Gaofeng Shao and Wuqi Guo for technical support in obtaining optical images. [9] J.W. Halloran, Ceramic Stereolithography: Additive Manufacturing for Ceramics by Photopolymerization, Annu. Rev. Mater. Res. 46 (2016) 19 -- 40, 10.1146/annurev-matsci- 070115-031841. [10] P. Colombo, G. Mera, R. Riedel, G.D. Sorarù, Polymer‐Derived Ceramics: 40 Years of Research and Innovation in Advanced Ceramics, Journal of the American Ceramic Society 93 (2010) 10.1111/j.1551- 2916.2010.03876.x. 1805 -- 1837, [11] T. Hanemann, M. Ade, M. Börner, G. Motz, M. Schulz, J. Hausselt, Microstructuring of Preceramic Polymers, Adv. Eng. Mater. 869 -- 873. dx.doi.org/10.1002/1527- 2648(20021105)4:1. [12]S. Martínez-Crespiera, E. Ionescu, M. Schlosser, K. Flittner, G. Mistura, R. Riedel, H.F. Schlaak, Fabrication oxycarbide-based microcomponents via photolithographic and soft silicon of 13 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. lithography approaches, Sensors and Actuators A: Physical 242 -- 249. Journal of the European Ceramic Society (2017), 10.1016/j.jeurceramsoc.2017.03.021. [13] S.K. Reddy, N.B. Cramer, T. Cross, R. Raj, C.N. Bowman, Polymer-Derived Ceramic Materials from Thiol-ene Photopolymerizations, Chem. Mater. 4257 -- 4261, 10.1021/cm034291x. (2003) 15 [14] L.-A. Liew, Y. Liu, R. Luo, T. Cross, L. An, V.M. Bright, M.L. Dunn, J.W. Daily, R. Raj, Fabrication by photopolymerization of pre-ceramic polymer, Sensors and Actuators A: Physical 95 (2002) 120 -- 134, 10.1016/S0924-4247(01)00723-3. SiCN MEMS of of [15] T.A. Pham, D.‐P. Kim, T.‐W. Lim, S.‐H. Park, D.‐Y. Yang, K.‐S. Lee, Three‐Dimensional SiCN Ceramic Microstructures via Nano‐ Stereolithography Inorganic Polymer Photoresists, Advanced Functional Materials 16 (2006) 1235 -- 1241, 10.1002/adfm.200600009. [16] S. Park, D.-H. Lee, H.-I. Ryoo, T.-W. Lim, D.- Y. Yang, D.-P. Kim, Fabrication of three- dimensional SiC ceramic microstructures with near-zero shrinkage via dual crosslinking induced stereolithography, Chem. Commun. 0 (2009) 4880 -- 4882, 10.1039/B907923H. [17] E. Zanchetta, M. Cattaldo, G. Franchin, M. Schwentenwein, J. Homa, G. Brusatin, P. Colombo, Stereolithography of SiOC Ceramic Microcomponents, Advanced Materials 28 (2016) 370 -- 376, 10.1002/adma.201503470. [18] Z.C. Eckel, C. Zhou, J.H. Martin, A.J. Jacobsen, W.B. Carter, T.A. Schaedler, Additive manufacturing of polymer-derived ceramics, Science 58 -- 62, 10.1126/science.aad2688. (2016) 351 [19] J.M. Hundley, Z.C. Eckel, E. Schueller, K. Cante, S.M. Biesboer, B.D. Yahata, T.A. Schaedler, Geometric characterization of additively manufactured polymer derived ceramics, Additive Manufacturing 18 (2017) 95 -- 102, 10.1016/j.addma.2017.08.009. [20] J. Schmidt, P. Colombo, Digital light processing of ceramic components from polysiloxanes, Journal of the European Ceramic Society 38 (2018) 57 -- 66, 10.1016/j.jeurceramsoc.2017.07.033. [21] H. Cui, R. Hensleigh, H. Chen, X. Zheng, Additive Manufacturing and size-dependent mechanical properties of three-dimensional microarchitected, high-temperature ceramic metamaterials, Journal of Materials Research 33 (2018) 360 -- 371, 10.1557/jmr.2018.11. [22] Y. Fu, G. Xu, Z. Chen, C. liu, D. Wang, C. Lao, Multiple metals doped polymer-derived SiOC ceramics for 3D printing, Ceramics International (2018), 10.1016/j.ceramint.2018.03.075. [23] Y. de Hazan, D. Penner, SiC and SiOC ceramic articles produced by stereolithography of acrylate modified polycarbosilane systems, [24] C.E. Hoyle, C.N. Bowman, Thiol -- Ene Click Chemistry, Angewandte Chemie International Edition 49 (2010), 10.1002/anie.200903924. [25] A. Ovsianikov, J. Viertl, B. Chichkov, M. Oubaha, B. MacCraith, I. Sakellari, A. Giakoumaki, D. Gray, M. Vamvakaki, M. Farsari, C. Fotakis, Ultra-low shrinkage hybrid photosensitive material two-photon polymerization microfabrication, ACS Nano 2 (2008) 2257 -- 2262, 10.1021/nn800451w. for Portela, [26] A. Vyatskikh, S. Delalande, A. Kudo, X. Zhang, C.M. J.R. Greer, Additive manufacturing of 3D nano-architected metals, Nat Commun 9 (2018) 593, 10.1038/s41467- 018-03071-9. [27]X. Zheng, H. Lee, T.H. Weisgraber, M. Shusteff, J. DeOtte, E.B. Duoss, J.D. Kuntz, M.M. Biener, Q. Ge, J.A. Jackson, S.O. Kucheyev, N.X. Fang, C.M. ultrastiff mechanical metamaterials, Science 344 (2014) 1373 -- 1377, 10.1126/science.1252291. Spadaccini, Ultralight, [28] F. García Moreno, M. Fromme, J. Banhart, Real- time X-ray Radioscopy on Metallic Foams Using a Compact Micro-Focus Source, Adv. Eng. Mater. 416 -- 420, 10.1002/adem.200405143. (2004) 6 [29] L.J. Gibson, M.F. Ashby, Cellular Solids: Structure and Properties, Cambridge University Press, 2014, 10.1017/CBO9781139878326. [30] G.D. Soraru, F. Babonneau, J.D. Mackenzie, Structural evolutions from polycarbosilane to SiC ceramic, Journal of Materials Science 25 (1990) 3886 -- 3893, 10.1007/BF00582455. [31] H.Q. Ly, R. Taylor, R.J. Day, F. Heatley, Conversion of polycarbosilane (PCS) to SiC- based ceramic Part 1. Characterisation of PCS and curing products, Journal of Materials Science 4037 -- 4043, 10.1023/A:1017942826657. (2001) 36 [32] E. Ionescu, Polymer-Derived Ceramics, in: R. Riedel, I.-W. Chen (Eds.), Ceramics science and technology, Wiley; Wiley-VCH, Hoboken, NJ, Weinheim, 2012, pp. 457 -- 500. [33] S. Kaur, R. Riedel, E. Ionescu, Pressureless fabrication of dense monolithic SiC ceramics from a polycarbosilane, Journal of the European Ceramic Society 34 (2014) 3571 -- 3578, 10.1016/j.jeurceramsoc.2014.05.002. [34] K. Tangermann-Gerk, G. Barroso, B. Weisenseel, P. Greil, T. Fey, M. Schmidt, G. Motz, Laser pyrolysis of an organosilazane- based glass/ZrO 2 composite coating system, Materials & Design 109 (2016) 644 -- 651, 10.1016/j.matdes.2016.07.102. [35] Gabriela Mera, Alexandra Navrotsky, Sabyasachi Sen, Hans-Joachim Kleebe, Ralf Riedel, Polymer -derived SiCN and SiOC ceramics -- structure and energetics at the 14 Wang, X., Schmidt, F., Hanaor, D., Kamm, P.H., Li, S. and Gurlo, A., 2019. Additive manufacturing of ceramics from preceramic polymers: A versatile stereolithographic approach assisted by thiol-ene click chemistry. Additive Manufacturing, 27, pp.80-90. [47] H. Awaji, T. Watanabe, Y. Nagano, Compressive testing of ceramics, Ceramics International 20 (1994) 159 -- 167, 10.1016/0272- 8842(94)90034-5. nanoscale, Journal of Materials Chemistry A 1 (2013) 3826 -- 3836, 10.1039/C2TA00727D. [36] S. Martínez-Crespiera, E. Ionescu, H.-J. Kleebe, R. Riedel, Pressureless synthesis of fully dense and crack-free SiOC bulk ceramics via photo- crosslinking and pyrolysis of a polysiloxane, Journal of the European Ceramic Society 31 (2011) 913 -- 919, 10.1016/j.jeurceramsoc.2010.11.019. [37] J.V. Ryan, P. Colombo, J.A. Howell, C.G. Pantano, Tribology-Structure Relationships in Silicon Oxycarbide Thin Films, International Journal of Applied Ceramic Technology 7 (2010) 675 -- 686, 10.1111/j.1744-7402.2009.02374.x. [38] M.A. El Khakani, M. Chaker, A. Jean, S. Boily, J.C. Kieffer, M.E. O'Hern, M.F. Ravet, F. Rousseaux, Hardness and Young's modulus of amorphous a-SiC thin films determined by nanoindentation and bulge tests, J. Mater. Res. 9 (1994) 96 -- 103, 10.1557/JMR.1994.0096. [39] G.M. Renlund, S. Prochazka, R.H. Doremus, Silicon oxycarbide glasses, J. Mater. Res. 6 (1991) 2716 -- 2722, 10.1557/JMR.1991.2716. [40] J. Zhang, M.F. Ashby, The out-of-plane properties of honeycombs, International Journal of Mechanical Sciences 34 (1992) 475 -- 489, 10.1016/0020-7403(92)90013-7. [41] G. Liu, Y. Zhao, G. Wu, J. Lu, Origami and 4D printing of elastomer-derived ceramic structures, Science Advances 4 eaat0641, 10.1126/sciadv.aat0641. (2018) [42] P. Ramavath, P. Biswas, N. Ravi, R. Johnson, Prediction and validation of buckling stress (σcrt) of the ceramic honeycomb cell walls under quasi-static compression, Cogent Engineering 3 (2016) 474, 10.1080/23311916.2016.1168068. [43] R.K. Bird, T.S. Lapointe, Evaluation of Ceramic Honeycomb Core Compression Behavior at Room Temperature. [44] J.T. Muth, P.G. Dixon, L. Woish, L.J. Gibson, J.A. Lewis, Architected cellular ceramics with tailored stiffness via direct foam writing, Proc Natl Acad Sci USA 114 (2017) 1832 -- 1837, 10.1073/pnas.1616769114. [45] C. AGRAFIOTIS, ROMERO, STOBBE, M. I. MAVROIDIS, A. KONSTANDOPOULOS, B. HOFFSCHMIDT, P. V. FERNANDEZQUERO, Evaluation of porous silicon carbide monolithic honeycombs as volumetric receivers/collectors of concentrated solar radiation, Solar Energy Materials and Solar Cells 474 -- 488, (2007) 10.1016/j.solmat.2006.10.021. 91 [46] N.A. Fleck, V.S. Deshpande, M.F. Ashby, Micro-architectured materials: past, present and future, Proceedings of the Royal Society A: and Engineering Mathematical, Physical Sciences 2495 -- 2516, 10.1098/rspa.2010.0215. (2010) 466 15
1710.03700	1	1710	2017-10-08T12:15:14	Hall effect spintronics for gas detection	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	We present the concept of magnetic gas detection by the Extraordinary Hall effect (EHE). The technique is compatible with the existing conductometric gas detection technologies and allows simultaneous measurement of two independent parameters: resistivity and magnetization affected by the target gas. Feasibility of the approach is demonstrated by detecting low concentration hydrogen using thin CoPd films as the sensor material. The Hall effect sensitivity of the optimized samples exceeds 240% per 104 ppm at hydrogen concentrations below 0.5% in the hydrogen/nitrogen atmosphere, which is more than two orders of magnitude higher than the sensitivity of the conductance detection.	physics.app-ph	physics	Hall effect spintronics for gas detection. A. Gerber, G. Kopnov and M. Karpovski Raymond and Beverly Sackler Faculty of Exact Sciences, School of Physics and Astronomy Tel Aviv University Ramat Aviv 69978 Tel Aviv, Israel We present the concept of magnetic gas detection by the Extraordinary Hall effect (EHE). The technique is compatible with the existing conductometric gas detection technologies and allows simultaneous measurement of two independent parameters: resistivity and magnetization affected by the target gas. Feasibility of the approach is demonstrated by detecting low concentration hydrogen using thin CoPd films as the sensor material. The Hall effect sensitivity of the optimized samples exceeds 240% per 104 ppm at hydrogen concentrations below 0.5% in the hydrogen/nitrogen atmosphere, which is more than two orders of magnitude higher than the sensitivity of the conductance detection. 1 Reliable detection of hazardous, harmful, or toxic gases has become a major issue due to more stringent environmental and safety regulations worldwide. Solid state conductometric gas sensors present a high potential for applications where the use of conventional analytical systems such as gas chromatography or optical detection is prohibitively expensive or impossible. Operation of these sensors is based on a change of electric conductivity when exposed to an atmosphere containing specific reagents, usually caused by charge transfer between the sensor material and the adsorbed species. Two significant disadvantages of such sensors are the lack of chemical selectivity and sensitivity to humidity. The materials are normally sensitive to more than one chemical species and show cross-sensitivity when different reactive gases are present simultaneously in the atmosphere. When the only parameter measured by the sensor is the change of resistance one can only record the overall electrical effect of quite complex surface reactions. In other words, by only measuring the resistance change one does not have the needed discrimination for the correlation between specific surface species and their electrical effect. In principle, the needed discrimination can be provided by the results obtained by applying additional spectroscopic techniques. Unfortunately, most of the standard spectroscopic investigations are performed in conditions far away from the ones normally encountered in real applications: in ultra- high vacuum, at low temperatures, exposed to high concentrations of reactive gases, etc [1, 2]. The concept of magnetic gas detection has been promoted by a number of researchers that discovered that magnetic properties of several materials are modified when exposed to certain gases, for example to hydrogen. Interaction of hydrogen with ferromagnetic structures containing Pd was shown to change their structural, electronic, optical, and magnetic properties [3]. Significant modifications in susceptibility, magnetization, magnetic anisotropy and ferromagnetic resonance were found in Co/Pd multilayers [4, 5], Pd/Co/Pd tri-layers [6,7], Pd/Fe, Pd/Co and Pd/Ni bilayers [8, 9] and in Pd-rich CoPd alloy films [10, 11]. Additional materials, like Fe/Nb [12] and Fe/V superlattices [13, 14] and SnFeO2 ferrites [15] also demonstrated systematic changes in magnetic properties and exchange coupling when loaded with hydrogen. All effects mentioned above were detected by using laboratory magnetometric techniques and equipment like polarized neutron reflectivity, X-ray resonant magnetic scattering interference devices (SQUID), vibrating (XRMS), superconducting quantum 2 magnetometers, optical Kerr effect and ferromagnetic resonance setups. Adaptation of these techniques to field conditions present a formidable challenge. Therefore, although the idea of gas sensing using the magnetic properties of ferromagnetic materials has been formulated, its realization in practical devices was not implemented so far. In this Letter, we present the concept of magnetic gas detection using the extraordinary Hall effect [16, 17]. The essence of the effect is the following: electric current flowing along magnetic film generates voltage in direction perpendicular to the current direction, given by: V H  IR H  I t  R OHE RB  M 0 EHE (1) where RH is the Hall resistance, I - current, t - thickness of the film, B, and M are components of the magnetic induction and magnetization normal to the film plane. OHER is the ordinary Hall effect coefficient related to the Lorentz force acting on moving charge carriers. , the extraordinary Hall effect coefficient, is associated with a break of the right-left symmetry at spin-orbit scattering in magnetic materials. The EHE contribution can exceed significantly the ordinary Hall effect term in the relevant low field range, and the total Hall resistance RH can be approximated as: EHER (cid:1844)(cid:3009)(cid:3404)(cid:1848)(cid:3009) (cid:1835)⁄ (cid:3404) 0 tMREHE / (2) Thus, the Hall signal is directly proportional to magnetization. The effect is used as a highly sensitive tool in studies of magnetic properties of ultra-thin magnetic films and nano-structures [18]. Prospects of the Hall effect-based spintronics for magnetic sensors, memories and logic devices [19, 20] were boosted recently by discovery of a huge EHE in amorphous CoFeB oxides, that exhibit the magnetic field sensitivity three orders of magnitude higher than the best achieved in semiconducting materials [21]. Here, we suggest to use the extraordinary Hall effect as a tool for monitoring changes in magnetic properties of the sensor material exposed to specific gaseous elements. The measurement procedure is technically similar to the four-probe measurement of resistance with two modifications: 1) Hall voltage is measured in direction perpendicular to electric current flow, and 2) the measurement is generally done under a bias magnetic field. In the future sensor we envisage, two independent properties: 3 resistance and EHE will be measured simultaneously in the same magnetotransport setup. To estimate feasibility of the EHE gas detection we studied the EHE response to hydrogen using thin CoPd alloy films. Hydrogen is highly soluble in palladium, making palladium the metal of choice in hydrogen sensors. The palladium lattice expands significantly with absorption of hydrogen (0.15% in the α- phase and 3.4% in the β- phase), and resistivity of Pd increases with conversion into palladium hydride [22]. Similar response is also observed in Pd-based alloys [23]. Our earlier studies of Co-Pd alloys and multilayers revealed a strong sensitivity of the magnitude and polarity of the EHE signal on the relative content of the system, in particular for Co volume concentrations in the range 10% - 30% [24]. We started this study by assuming that absorption of hydrogen by palladium will modify the structure and electronic state of the system and thus affect the EHE signal. Polycrystalline CoxPd1-x films with Co atomic concentration x in the range 0 (cid:3409)(cid:1876)(cid:3409) 0.4 were deposited by e-beam co-evaporation from two separate targets on room temperature GaAs substrates. Co and Pd are completely soluble and form an equilibrium fcc solid solution phase at all compositions [25]. Film thickness varied between 5 nm and 20 nm. No post-deposition alloying was used. Several samples were deposited on silicon and glass substrates and demonstrated the response similar to those deposited on GaAs. Fig.1 presents the Hall resistivity (cid:2025)(cid:3009)(cid:3404)(cid:1844)(cid:3009)(cid:1872) as a function of magnetic field for four to the EHE. Polarity of the effect, defined as (cid:1856)(cid:2025)(cid:3009) (cid:1856)(cid:1828)⁄ CoxPd1-x samples with x = 0.08, 0.15, 0.2 and 0.25 (atomic concentration) and thickness 15nm, 18nm, 15nm and 14nm respectively measured in ambient air at room temperature. The ordinary Hall effect, corresponding to the high field linear slope beyond magnetization saturation, is negligible, and the observed signal is mainly due , indicates the dominance of the right-hand versus left-hand spin-orbital scattering. The polarity reverses between x = 0.15 and x = 0.2. Samples richer in Co exhibit a positive polarity, while samples richer in Pd have a negative one. The out-of-plane magnetic anisotropy with a significant hysteresis is developed in samples in a vicinity of the EHE sign reversal point. Development of the perpendicular anisotropy has been attributed to a strained state of thin CoPd films [26], that are known to have a very large magnetostriction reaching its 4 maximum in the same concentration range [27, 28]. Both the EHE and magnetostriction are results of the spin-orbit interactions in ferromagnetic materials, however the correlation between the two phenomena is not explained. Replacement of the ambient air by the pure nitrogen or by the pure carbon monoxide CO atmospheres does not affect the EHE loops. However, the response is significant when hydrogen is added. The four samples shown in Fig.1 were measured in hydrogen/nitrogen H2/N2 mixture with 4% of hydrogen, and the results are presented in Fig.2. The magnitude of the saturated EHE signal is reduced by 2% to 15% in all hydrogenated samples. The most pronounced changes are observed in the hysteresis loops of the samples with the out-of-plane anisotropy. Width of the quadratic hysteresis loop shrinks in the Co0.15Pd0.85 sample (Fig.2b). The Co0.2Pd0.8 sample also demonstrates a reduction of the coercive field together with the zero field remanence signal reduced to about a half (Fig.2c). Reduction of the coercive field and the remanence indicate the decreasing perpendicular magnetic anisotropy with hydrogen absorption. Fig.3 presents the field dependent hysteresis loops measured in 5 nm thick Co0.17Pd0.83 sample in H2/N2 atmosphere at different hydrogen concentrations between 0 and 4%. Thinner films seem to be attractive for sensing purposes due to a higher surface to volume ratio, and since the absolute value of the measured signal (Eqs.1 and 2) increases both by the reducing thickness t and by enhancing the EHE coefficient boosted by the spin-orbit surface scattering [29]. After the initial measurement in EHER N2 (99.998%) at atmospheric pressure, the sample chamber was filled with H2 4% H2/N2 mixture. The following sequence of measurements at reduced hydrogen concentrations was done after pumping the chamber to half of atmospheric pressure and refilling the chamber by nitrogen. After completing the sequence, the sample was re-measured in N2. The hysteresis loops are fully reproducible when the sequence is repeated. As seen, the saturated magnitude of the signal at high field, the remanence at zero field and the width of the hysteresis loop decrease with increasing hydrogen concentration. The quantitative data are shown in Fig.4. Fig.4a presents the coercive field (cid:1834)(cid:3030) as a dependence on the hydrogen concentration as: (cid:1834)(cid:3030)(cid:4666)(cid:1877)(cid:4667)(cid:3404)(cid:1834)(cid:3030)(cid:4666)0(cid:4667)(cid:1877)(cid:2879)(cid:3082) with (cid:1834)(cid:3030)(cid:4666)0(cid:4667)(cid:3406)7 (cid:1865)(cid:1846) function of hydrogen concentration y. Hc can be well presented by the power law 5 towards 4%. Fig. 4b presents the normalized change of the EHE signal measured at several fixed fields within the hysteresis loop (H = 0, 1.5 mT and 4 mT). The normalized and (cid:2011)  0.3, i.e. it varies significantly at low hydrogen concentrations and saturates EHE change is defined as: ∆(cid:1844)(cid:3009),(cid:3041)(cid:3042)(cid:3045)(cid:3040)(cid:4666)(cid:1877)(cid:4667)(cid:3404)∆(cid:3019)(cid:3257)(cid:4666)(cid:3052)(cid:4667) , where (cid:1877) is the (cid:4666)(cid:1845)(cid:3404)(cid:1856)∆(cid:1844)(cid:3009),(cid:3041)(cid:3042)(cid:3045)(cid:3040) (cid:1856)(cid:1877)⁄ (cid:4667) exceeds 240%/104 H2 ppm at hydrogen concentrations below hydrogen concentration. The signal varies strongly at low H2 concentrations and saturates by approaching 4%. The rate of the signal variation and the range of the linear the sensitivity response depend on (cid:3019)(cid:3257)(cid:4666)(cid:2868)(cid:4667)(cid:3404)(cid:3019)(cid:3257)(cid:4666)(cid:3052)(cid:4667)(cid:2879)(cid:3019)(cid:3257)(cid:4666)(cid:2868)(cid:4667) (cid:3019)(cid:3257)(cid:4666)(cid:2868)(cid:4667) the bias field. At 4 mT bias field 0.5%. At 1.5 mT the sensitive range extends up to 2% of hydrogen with sensitivity about 30%/104 ppm. Variation of the remnant EHE signal at zero bias field reaches 30% at 4% hydrogen. The response is not linear over a wider concentration range, which should be taken into account in calibration of the future sensors. Resistance response to hydrogen measured simultaneously with the EHE is shown in Fig.4c. The data taken at zero field (○) and in the magnetically saturated state under 0.1T bias field (×) are presented in the form of the normalized resistance change, defined as: ∆(cid:1844)(cid:3041)(cid:3042)(cid:3045)(cid:3040)(cid:3404)(cid:3019)(cid:4666)(cid:3052)(cid:4667)(cid:2879)(cid:3019)(cid:4666)(cid:2868)(cid:4667) . Both at zero and under 0.1T field resistance increases (cid:3019)(cid:4666)(cid:2868)(cid:4667) hydrogen concentration, defined as: (cid:1856)∆(cid:1844)(cid:3041)(cid:3042)(cid:3045)(cid:3040) (cid:1856)(cid:1877)⁄ , is about 0.8%/104 ppm. depend on the bias field. Fig.4c also presents the normalized EHE response ∆(cid:1844)(cid:3009),(cid:3041)(cid:3042)(cid:3045)(cid:3040) Magnetoresistance of the sample is small, negative and independent on hydrogen absorption. Therefore, the resistance changes caused by hydrogen adsorption don't about linearly with hydrogen concentration up to 4%. The resistance sensitivity to measured in the magnetically saturated state at a fixed field 0.1T. The EHE and the resistivity responses to hydrogen absorption are independent of each other. The magnetic EHE response is negative, reaches 12% at low hydrogen presence and saturates towards 4% concentration. Resistivity increases in the measured H2 concentration range with no signs of saturation. Following Eq.2, the EHE signal scales with depends on the EHE coefficient 2EHER resistivity as: , following the intrinsic Berry phase mechanism or the extrinsic mechanism of side jump scattering [30]. Changes of the saturated magnetization and of the field dependent EHER due to the skew scattering mechanism or as and magnetization M. EHER EHER 6 hysteresis loop due to gas absorption are uncorrelated with resistivity, which makes the EHE and resistivity responses independent. Reduction of the saturated EHE signal with increasing hydrogen concentration is consistent with the generally observed decrease of the total magnetization [5, 10] in hydrogenated Co/Pd systems, the effect attributed to modification of the electronic structure of the material. On the other hand, the effect of hydrogen absorption on the perpendicular anisotropy is ambivalent. Enhancement of the perpendicular magnetic anisotropy was found in hydrogenated Pd/Co/Pd trilayers [6], associated by the authors with improvements of Pd (1,1,1) orientation, and in Pd-rich alloy film [10], attributed to the development of a long-range magnetic order. The coercive field and the perpendicular magnetic anisotropy of our samples decrease with hydrogen absorption, similar to Co/Pd multilayers reported in Ref. [5]. We tentatively suggest that changes in magnetic anisotropy depend strongly on magnetostriction and strain of the material, similar to the concentration dependence of non-hydrogenated CoPd films (Fig.1). More studies are needed to clarify this point. the range of To summarize, one can expect that selectivity of solid state gas sensors will be improved by extending independent measurable parameters complementing the conductometric sensing. We argue that the extraordinary Hall effect (EHE), sensitive to variations of magnetic properties of ferromagnetic materials, can serve as such complementary magnetotransport parameter. Possibility to apply the technique for gas sensing was demonstrated by detecting low concentration hydrogen using thin CoPd films. Sensitivity of the EHE response in the optimized samples exceeds 240% per 104 ppm at hydrogen concentrations below 0.5% in the hydrogen/nitrogen atmosphere, which is more than two orders of magnitude higher than the sensitivity of the conductance detection. The research was supported by the State of Israel Ministry of Science, Technology and Space grant No.53453. 7 References 1. S. Capone, A. Forleo, L. Francioso, R. Rella, P. Siciliano, J. Spadavecchia, D.S. Presicce, A.M. Taurino, J. Optoelectron. Adv. Mat. 5, 1335 (2003); 2. N. Barsan, D. Koziej and U. Weimar, Sensors and Actuators B 121, 18 (2007); 3. K. Munbodh, F.A. Perez, and D. Lederman, J. Appl. Phys. 111, 123919 (2012); 4. W. Lin, C. Tsai, X. Liu, and A.O. Adeyeye, J. Appl. Phys. 116, 073904 (2014). 5. K. Munbodh, F. A. Perez, C. Keenan, D. Lederman, M. Zhernenkov, and M. R. Fitzsimmons, Phys. Rev. B 83, 094432 (2011); 6. S. Okamoto, O. Kitakami, and Y. Shimada, J. Magn. Magn. Mat. 239, 313 (2002); 7. W. Lin, C. Tsai, B. Wang, C. Kao, and W. Pong, Appl. Phys. Lett. 102, 252404 (2013); 8. W. Lin, C. Chi, T. Ho, C. Tsai, Thin Solid Films 531, 487 (2013); 9. C.S. Chang, M. Kotylev, and E.Ivanov, Appl. Phys. Lett. 102, 142405 (2013); 10. W. Lin, C. Tsai, H. Huang, B. Wang, V. R. Mudinepalli, and H. Chiu, Appl. Phys. Lett. 106, 012404 (2015); 11. S. Akamaru, T. Matsumoto, M. Murai, K. Nishimura. M. Hara, and M. Matsuyama, J. Alloys Compd 645, S213 (2015); 12. F. Klose, Ch. Rehm, D. Nagengast, H. Maletta, and A. Weidinger, Phys. Rev. Lett. 78, 1150 (1997); 13. D. Labergerie, C. Sutter, H. Zabel, B. Hjorvarsson, J. Magn. Magn. Mater. 192, 238 (1999); 14. A. Remhof, G. Nowak, H. Zabel, M. Bjorck, M. Parnaste, B. Hjorvarsson and V. Uzdin, Europhys.Lett. 79, 37003 (2007); 15. A. Punnoose,_ K. M. Reddy, J. Hays, A. Thurber, and M. H. Engelhard, Appl. Phys. Lett. 89, 112509 (2006); 16. E.M. Pugh, Phys. Rev. 36, 1503 (1930); 8 17. C.M. Hurd, The Hall Effect in Metals and Alloys, Plenum Press, New York, 1972; 18. A. Gerber, A. Milner, M. Karpovsky, B. Lemke, H.-U. Habermeier, J. Tuaillon- Combes, M. Négrier, O. Boisron, P. Mélinon, and A. Perez, J. Magn. Magn. Mater. 242, 90 (2002); 19. A. Gerber, J. Magn. Magn. Mater.. 310, 2749 (2007); 20. J. Moritz, B. Rodmacq, S. Auffret, and B. Dieny, J. Phys. D: Appl. Phys. 41, 135001 (2008); 21. G. Kopnov and A. Gerber, Appl. Phys. Lett. 109, 022404 (2016); 22. T. B. Flanagan and F. A. Lewis, Trans. Faraday Soc. 55, 1400 (1959); 23. K. Baba, U. Miyagawa, K. Watanabe, Y. Sakamoto, and T.B. Flanagan, J. Mater. Sci. 25, 3910 (1990); 24. G. Winer, A. Segal, M. Karpovski, V. Shelukhin, and A. Gerber, J. Appl. Phys. 118, 173901 (2015); 25. K. Ishida and T. Nishizawa, J. Phase Equilib. 12, 83 (1991); 26. J. I. Hong, S. Sankar, A. E. Berkowitz, and W. F. Egelhoff, Jr., J. Magn. Magn. Mater. 285, 359 (2005), 27. S. Hashimoto, Y. Ochiai, and K. Aso, Jpn. J. Appl. Phys. 28, 1596 (1989); 28. H. Takahashi, S. Tsunashima, S. Iwata, S. Uchiyama, Jpn. J. Appl. Phys. 32, L1328 (1993); 29. A. Gerber, A. Milner, L. Goldsmith, M. Karpovski, B. Lemke, H.-U. Habermeier, and A. Sulpice, Phys. Rev. B 65, 054426 (2002); 30. N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Rev. Mod. Phys. 82, 1539 (2010). 9 Figure captions. Fig.1. Field dependence of the Hall resistivity of CoxPd1-x films with x = 0.08, 0.15, 0.2 and 0.25 (atomic concentrations) and respective thickness 15 nm, 18 nm, 15 nm and 14 nm measured in ambient air at room temperature. Fig.2. Hall effect resistance as a function of magnetic field of CoxPd1-x films measured in air (open circles) and in hydrogen/nitrogen H2/N2 mixture with 4% of hydrogen (solid circles): (a) Co0.08Pd0.92; (b) Co0.15Pd0.85; (c) Co0.2Pd0.8; (d) Co0.25Pd0.75. Fig.3. EHE resistance hysteresis loops measured in 5 nm thick Co0.17Pd0.83 film in H2/N2 atmosphere with different H2 concentrations (y = 0%, 0.125%, 0.25%, 0.5%, 1%, 2% and 4%). Fig.4. Hydrogen concentration dependence of: (a) the coercive field; (b) the normalized EHE change ∆(cid:1844)(cid:3009),(cid:3041)(cid:3042)(cid:3045)(cid:3040) under bias fields 0 mT, 1.5 mT and 4 mT within the hysteresis loop; (c) the normalized resistance change ∆(cid:1844)(cid:3041)(cid:3042)(cid:3045)(cid:3040) at zero field (○) and under 0.1T bias field (×) – right vertical axis, and the normalized EHE change ∆(cid:1844)(cid:3009),(cid:3041)(cid:3042)(cid:3045)(cid:3040) under 0.1T bias field – left vertical axis, measured in 5 nm thick Co0.17Pd0.83 film. 10 0.25 0.2 0.08 0.15 0.2 0.0 ) m c   ( H  -0.2 -1.0 -0.5 0.0 B (T) 0.5 1.0 Fig.1 11 )  m ( H R )  m ( H R Co0.08Pd0.92 air (H2)4(N2)96 75 50 25 0 -25 -50 -75 -100 (a) -1.5 -1.0 -0.5 0.5 1.0 1.5 0.0 B (T) Fig. 2a Co0.15Pd0.85 air (H2)4(N2)96 0.02 0.04 80 60 40 20 0 -20 -40 -60 -80 (b) -0.04 -0.02 0.00 B (T) Fig. 2b 12 (c) 30 20 10 0 -10 -20 )  m ( H R -30 -0.10 -0.05 )  ( H R (d) 0.4 0.2 0.0 -0.2 -0.4 -1.5 -1.0 -0.5 Co0.2Pd0.8 air (H2)4(N2)96 0.05 0.10 Co0.25Pd0.75 air (H2)4(N2)96 0.5 1.0 1.5 0.00 B (T) Fig. 2c 0.0 B (T) Fig. 2d 13 )  ( H R 0.2 0.1 0.0 (H2)y(N2)100-y -0.1 -0.2 0 % 4% 2% 1% 0.5% 0.25% 0.125% -0.01 Co0.17Pd0.83 0.00 B (T) Fig. 3 0.01 14 7 6 5 4 3 ) T m C H ( (a) 0.1 1 H2 concentration (%) Fig. 4a 0 -20 -40 -60 -80 -100 -120 -140 -160 (b) 0 1 2 3 H2 concentration (%) Fig. 4b 15 B = 0 B = 1.5 mT B = 4 mT 4 )  ( , m r o n H R  (c) 0 -2 -4 -6 -8 -10 -12 ) % ( m r o n , H R  0 1 2 3 H2 concentration (%) 4 Fig. 4c 16 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 ) % ( m r o n R  0.2 ) m c ( H 0.0 -0.2 0.25 0.2 0.08 0.15 -1.0 -0.5 0.0 B (T) 0.5 1.0 Co0.08Pd0.92 air (H2)4(N2)96 ) m ( H R 75 50 25 0 -25 -50 -75 -100 (a) -1.5 -1.0 -0.5 0.5 1.0 1.5 0.0 B (T) Co0.15Pd0.85 air (H2)4(N2)96 80 60 40 20 0 -20 -40 -60 -80 ) m ( H R (b) -0.04 -0.02 0.00 B (T) 0.02 0.04 (c) 30 20 10 0 -10 -20 ) m ( H R Co0.2Pd0.8 air (H2)4(N2)96 -30 -0.10 -0.05 0.00 B (T) 0.05 0.10 (d) 0.4 0.2 ) ( H R 0.0 -0.2 -0.4 Co0.25Pd0.75 air (H2)4(N2)96 -1.5 -1.0 -0.5 0.5 1.0 1.5 0.0 B (T) 0.2 0.1 ) ( H R 0.0 (H2)y(N2)100-y -0.1 -0.2 0 % 4% 2% 1% 0.5% 0.25% 0.125% -0.01 Co0.17Pd0.83 0.00 B (T) 0.01 7 6 5 4 3 ) T m C H ( (a) 0.1 1 H2 concentration (%) ) ( m r o n H R , 0 -20 -40 -60 -80 -100 -120 -140 -160 B = 0 B = 1.5 mT B = 4 mT 4 (b) 0 1 2 3 H2 concentration (%) ) % ( m r o n H R , (c) 0 -2 -4 -6 -8 -10 -12 0 1 2 3 H2 concentration (%) 4 3.5 3.0 2.5 2.0 1.5 ) % ( m r o n R 1.0 0.5 0.0
1707.03673	1	1707	2017-07-12T12:24:48	Twist Coupled Kirigami Cellular Metamaterials and Mechanisms	[ "physics.app-ph" ]	Manipulation of thin sheets by folding and cutting offers opportunity to engineer structures with novel mechanical properties, and to prescribe complex force-displacement relationships via material elasticity in combination with the trajectory imposed by the fold topology. We study the mechanics of cellular Kirigami that rotates upon compression, which we call Flexigami; the addition of diagonal cuts to an equivalent closed cell permits its reversible collapse without incurring significant tensile strains in its panels. Using finite-element modeling and experiment we show how the mechanics of flexigami is governed by the coupled rigidity of the panels and hinges and we design flexigami to achieve reversible force response ranging from smooth mono-stability to sharp bi-stability. We then demonstrate the use of flexigami to construct laminates with multi-stable behavior, a rotary-linear boom actuator, and self-deploying cells with activated hinges. Advanced digital fabrication methods can enable the practical use of flexigami and other metamaterials that share its underlying principles, for applications such as morphing structures, soft robotics and medical devices.	physics.app-ph	physics	Twist-Coupled Kirigami Cellular Metamaterials and Mechanisms Nigamaa Nayakantia, Sameh H. Tawficka,b, A. John Hart∗a aDepartment of Mechanical Engineering, Masssachusetts Institute of Technology, Cambridge, MA,02139 bDepartment of Mechanical Science and Engineering, University of Illinois - Urbana Champaign, IL Abstract Manipulation of thin sheets by folding and cutting offers opportunity to engineer struc- tures with novel mechanical properties, and to prescribe complex force-displacement relation- ships via material elasticity in combination with the trajectory imposed by the fold topology. We study the mechanics of cellular Kirigami that rotates upon compression, which we call Flexigami; the addition of diagonal cuts to an equivalent closed cell permits its reversible col- lapse without incurring significant tensile strains in its panels. Using finite-element modeling and experiment we show how the mechanics of flexigami is governed by the coupled rigidity of the panels and hinges and we design flexigami to achieve reversible force response ranging from smooth mono-stability to sharp bi-stability. We then demonstrate the use of flexigami to construct laminates with multi-stable behavior, a rotary-linear boom actuator, and self- deploying cells with activated hinges. Advanced digital fabrication methods can enable the practical use of flexigami and other metamaterials that share its underlying principles, for applications such as morphing structures, soft robotics and medical devices. Introduction Origami and Kirigami provide a means of transforming thin sheets into forms with unique mechanical properties, including extreme stretchability and multi-stability ([38],[27],[32],[29]). These materials and structures have promising applications in robotics ([7], [20], [4], [22]), space structures ([37],[30],[31]), soft actuators ([28],[40],[16],[25],[33]), photovoltaics ([15]), and other domains. Traditional origami assumes that creases behave as perfect hinges with zero rotational stiffness, and that the panels and connecting hinges are perfectly rigid and have zero thickness. Though these assumptions have proven useful to represent the motion of many folded systems, the flexibility of sheets enables the transformation of folded shapes Preprint submitted to Extreme Mechanics Letters July 13, 2017 that, from a mathematical point of view, are rigidly non-foldable. For example, using the principle of virtual folds, Silverberg et.al. [34] concluded that the flexibility of the panel faces give rise to the multi-stable trajectory of the square-twist Origami tessellation . This is one of many examples, including the Miura-Ori, and Resch patterns, whereby folding of a polygonal arrangement can enable complex geometric and mechanical transformations ([19]). In addition to folded sheets with repeating unit cells, there have been many studies of folded cylinders ([24],[36],[5],[3],[35],[8]) and tubes, which may be used as deployable booms and lightweight structural members. These include tessellations of identical triangular panels arranged on helical strips ([9],[10]), as well as variations of Miura-Ori sheets wrapped into tubes and assembled into cellular structures with perpendicular load bearing capabilities ([8]). Another cylindrical topology, the Kresling pattern, has a series of parallel diagonal creases approximating the pattern that arises during twist-buckling of a thin-walled cylinder ([21],[12],[13]). However, for some geometries compression of these seamless folded cylinders results in permanent structural failure due to formation of kinks and creases in the triangular panels due to tensile strains that develop in the initial stages of compression. We study the mechanics of cut-relieved folded cylindrical cells which we call flexigami. The addition of diagonal cuts imparts flat foldability of cylindrical topologies without incur- ring kinking and also mechanical behavior that spans from smooth mono-stability to sharp bi-stability. We show that the mechanics of flexigami is governed by the interplay between its rigid kinematics and the elasticity of panels and folds. Moreover, the cut-relieved de- sign permits accurate and efficient modeling of the energetics using geometrical mechanics as well as finite-element modeling, contrasting previous studies which used simplified tri- angular truss-like structures to approximate the mechanics. We then demonstrate the use of flexigami cells to build multi-stable structures, compact rotary-linear mechanisms, and collapsible lightweight lattices. Geometry and construction of flexigami cells A flexigami cell has a N-sided polygonal base prism, surrounded by diagonally creased parallelograms (Figures 1a). The cell geometry is defined by the geometric parameters of the regular polygon prescribed by such an arrangement: the number of sides of the polygon 2 N, the side length of the polygon L, and a planar angular fraction λ. Each parallelogram is divided into a pair of triangles by a diagonal crease. The creases are perforated, and then the planar pattern is folded and glued to form the enclosed cell. When folded, the unit cell geometry is similar to a unit of the Kresling pattern, yet has cuts between each adjacent pair of triangles around the outer surface of its enclosed volume. These cuts enable strain relief and reversible collapse of the cell, along with relative rotation of the top and bottom surfaces. In the folded yet fully open, stress-free configuration, three key geometric constraints can be imposed: (1) the N-sided faces remain planar and are only permitted to rotate about the vertical (Z) axis; (2) diagonal creases folded to valleys remain straight and their lengths are preserved; and (3) the free edges of the triangular panels (AA(cid:48), BB(cid:48)) can assume any three-dimensional form, subject to the geometrical constraint of preserving their free length because their surfaces are considered to be developable. As such, the circum-radius (R), diagonal crease length (ξ), and area of the triangle (ζ) formed by joining the three points AA(cid:48)B(cid:48) or ABB(cid:48) are given by R = 2cos L 2N ξ = 2R cos (cid:105) (cid:104) π(N−2) (cid:20) (cid:20) π(N − 2) (1 − λ) 2N (cid:21) (cid:20) π(N − 2) 2N (cid:21) cos ζ = 4R2 cos (1 − λ) (1) (cid:21) (cid:20) sin λ (cid:21) π(N − 2) 2N π(N − 2) 2N Upon compression of the cell, the triangular panels must bend to accommodate the change in ζ while following the geometric constraints described above. The deviation (γ) of the panel geometry from a planar stress free configuration is expressed as γ = ζ0 − ζi ζ0 ; (2) where, ζi : instantaneous area of the triangle AA(cid:48)B(cid:48) or ABB(cid:48) and ζ0 : initial area of the 3 triangle AA(cid:48)B(cid:48) or ABB(cid:48). Figure 1c depicts this variation for various values of λ, when N = 7, L = 30. Here we observe that the change in γ for all values of λ is symmetric with respect to the relative height of a cell during compression. For λ > 0.5, we see that γ exhibits a clear global maximum, which increases monotonically with λ. Therefore, upon compression of the cell, the fractional change in ζ is accommodated by the out of plane bending of the panels. Hence, for a given (N, L), the force required to compress the unit cell, and therefore its stiffness and peak force when bistable should increase with λ. Because of the geometrical definition of the cell, its kinematic constraints are satisfied in exactly two configurations corresponding to completely open and closed states. The existence of any intermediate configurations requires material deformations as well as opening and closing of the creases, suggesting that the mechanical response of the cell is governed by the panel and hinge stiffness. Mechanical behavior of flexigami cells Displacement-controlled cyclic compression-tension tests (see b) on paper cells reveal the geometric tunability of mechanical behavior (Figure 2). As shown in Figure 2a, for λ = 0.5 the cell opens and closes smoothly; upon compression, the force increases gradually past relative displacement of H/2 and stiffenss sharply only when the folded panels contact one another. For λ > 0.5, the cell exhibits a snap-through behavior where it jumps from one stable equilibrium position to another. In this case, the force first increases linearly until reaching a peak value; at the instant of the peak force (FP ), the cell snaps, taking a negative stiffness. The force drops then to a local minimum value, and then the cell strengthens with continued compression due to monotonically increasing force required to fold the hinges and panel-panel contacts. For all the unit cells that exhibit snap-through behavior, their initial stiffness and peak force increase with λ for a constant N (Figure 2a). This agrees with the kinematic analysis showing that higher λ results in more significant out-of-plane deformations of the panels (γ), while the hinges sweep through the same net fold angle (Figure 1c). Similarly, for cells with λ = constant (here 0.8), snap through behavior is always observed and the peak force increases with N (Fig S2a). 4 When testing the flexigami cells, the top and bottom faces are held parallel, and only the bottom face is permitted to rotate. The diagonal cuts are essential to the smooth compression and bistability of flexigami cells. By contrast, seamless cells, become crumpled upon initial compression resulting in irreversible damage. This difference can be attributed to panel bending (Figure 2c, Video S1) and its influence on the bistability of cut-relieved cells. Without the cuts, the panels undergo combined stretching and bending; for instance, in Figure SI2 we shows the force-displacement response of a seamless cell (N = 7, λ = 0.8). The closed cell exhibits higher initial peak force than the equivalent cut-relieved cell, but the sidewalls become wrinkled leading to collapse. In the subsequent cycles there is a loss of peak force and the force-displacement curves do not overlap as we would see in the case of cut-relieved cells. In tests of other geometries, we observe instability in the response after the peak force for higher values of λ = 0.8 & λ = 0.9 where significant panel bending is otherwise common in cut-relieved cells. These observations suggest the influence of panel bending and irreversible nature of structural failure due to crumpling of the triangular panels. A finite element model of a flexigami cell (N = 7, λ = 0.8) with a single panel is shown in Figure 2d (Video-S2). Consistent with tensile tests on paper strips, the paper is modeled as an orthotropic material with E1 = 6.8GPa and E2= 3.1GPa (Fig. SI3). Creases are modeled as torsional springs with constant stiffness. The simulated response is compared with the experimental data for the same geometry in Figure 2b); it quite accurately predicts the initial stiffness, peak force, and negative force after snap-through, yet does not capture the stiffening upon panel-panel contact. The finite element model also lets us visualize how, during compression, the strains in the creases increase monotonically, while the strains in the panels reach a maximum and subsequently decrease as the cell is compressed fully. At the instant of maximum bending, the stresses developed in the panels increase in the direction away from the diagonal hinge. The paper cells also exhibit significant hysteresis in their compression-tension behavior. Sequentially removing panels (i.e., single triangle pairs AA(cid:48)BB(cid:48)) from a cell predictably decreased the peak force, yet decreased the hysteresis even more-so (Fig. SI2c). Therefore, we conclude that friction between adjacent panels contributes significantly to the energy dissipation upon cyclic compression-tension . Also, the mechanical behavior is insensitive to 5 loading rate within bounds tested (25-200 mm/min), and unless otherwise noted tests were performed at 25mm/min (Fig. SI2b). Interplay of panel and crease mechanics The interplay of panel bending and crease folding in twist-coupled Kirigami cells can be further understood using a geometric mechanics approach. At each instantaneous height of the compressed cell, the total energy of the system (ET ), is the summation of the bending energy of panels (EB) and energy stored in the creases (EC). The triangular panels are modeled as developable surfaces [23], and the governing pair of space curves are the free edge of the triangular panel (RAA(cid:48)(s)) and the diagonal crease (RAB(cid:48)(s)). In the deformed state, the diagonal crease remains straight (RAB(cid:48)(s)), and the free edge of each triangular panel (RAA(cid:48)(s)) is represented as RAA(cid:48)(s) = a1sin(πs) + a2sin(2πs) + a3sin(3πs) (3) Assuming isometric deformations of the triangular panels, the panels are thus developable surfaces. The surface of each panel is then generated by joining the corresponding points on the two space curves RAA(cid:48)(s) and RAB(cid:48)(s) and is mathematically represented as R(s, v) = (1 − v)RAA(cid:48)(s) + vRAB(cid:48)(s). (4) The bending energy of a developable surface is proportional to the surface integral of the squared mean curvature ([6],[18]). Thus for each side of the cell, (cid:90) (cid:90) EB = 2KB H 2dS. (5) Here, KB is the bending rigidity of the panels which is a function of Young's modulus (E), Poisson's ratio (ν) and material thickness (t). During compression, the instantaneous fold angle of the creases (θh) is calculated and the energy stored in the creases is proportional to the square of the deviation in their angle from the rest position (θh − θ0)2. We therefore consider the creases to be linear elastic torsional springs (Figures 3a)([17]), whose spring stiffness is given by KC. Each side of the N sided 6 cell has three creases, two mountain folds and one diagonal valley fold. So, the crease energy associated with a single side is EC = KC 2 (cid:20)(cid:90) 3(cid:88) i=1 (cid:21) , [(θhi − θ0i)]2 dl (6) where θ0i is the rest angle of the crease i, and θhi is the instantaneous angle of the crease i. By minimizing ET subject to prescribed kinematic and displacement boundary condi- tions, we solve for the equilibrium shape of the triangular panels at each instantaneous height during compression. From this, we learn that for lower ratios of KR = KB/KC, thus higher crease stiffnesses, the energy during compression increases monotonically (Figure 3b) implying mono-stability of the system. As KR increases further, an energy barrier develops, representing bi-stability. By plotting the energy trajectories of panel bending and crease folding (Figure 3c), we see that the kinematic path of the cell causes the panels to bend through an energy maximum, while the crease energy increases smoothly throughout the compression stroke. For each configuration (N , λ), we can therefore demarcate how λ and KR determine whether the cell is monostable or bistable (Figure 3d). Our geometric mechanics approach, wherein the panels are considered to be developable surfaces, is importantly different than prior efforts to explain the mechanics of either multi- stable origami (e.g., the square-twist, [34]) using the principle of virtual folds or the equiv- alent seamless (e.g., Kresling) cylindrical shell using a simplified truss. In the virtual folds approach, the localized facet bending is approximated as a virtual fold and the panels are treated as rigid; in the truss approach, the fold pattern is simply analyzed as a network of linear elastic beams, and the panels are not considered. As a result, these approaches predict either a transition from mono-stability to bi-stability or obtain a condition on geometrical parameters for the structure to exhibit bi-stability as the fold pattern is varied. They donot accurately represent the scaling of energetics nor the coupling between crease folding and panel bending. Our approach involves direct definition of the load-free geometry of the structure along with the material properties of the panels and creases. Minimization of the total energy allows us to determine the equilibrium shapes of the panels and creases at the state of deformation, and therefore understand how the elasticity and kinematics together 7 govern the structural response. Thus, this approach, along with appropriate formulation and computational solution methods, can be applied to any Origami or Kirigami structure. Flexigami-based mechanisms and laminates The above understanding of the geometric mechanics of flexigami cells enables the design of assemblies, laminates, and mechanisms having novel, nonlinear mechanical behavior and actuation response. First, compression of two stacked cells of opposite chirality (Figure 4a), having at least one of its end surfaces constrained against rotation, results in sequential snapping combined with net rotation. When both the top and bottom surfaces of the two- cell stack are constrained fully, the cells snap simultaneously in order to satisfy the constraint of zero net rotation. Both assemblies exhibit the same initial stiffness and peak force and can be compressed fully, yet the unconstrained assembly has three stable states and much greater hysteresis (Figure 4b). It follows that single-chirality stacks of cells with identical N and varied λ have a number of stable states equal to the number of cells (Video-S3), yet their force-displacement curve is independent of the stacking order (Figure 4c). Upon compression, the cells collapse in the sequential order of their peak force, thus a consecutively greater force is required to switch the stack to the next stable state. Also, the negative force exerted by each snapping cell is compensated internally within the stack due to elasticity of the other cells, fully isolating the end constraints from the dynamics of snapping. It can therefore be appreciated that the engineerable mechanical response of twist-coupled flexigami cells, along with their large stroke, can enable the design of collapsible and de- ployable lightweight materials. Practically, such will be limited by the formation of folded structures from structural materials, and in particular the construction of durable hinges that can reversibly endure large deformations. To demonstrate this possibility as well, we show an array of small unit cells made from a carbon fiber fabric (Carbitex). This assem- bly (Figure 4d, Video S4; N = 6, L = 15mm, λ = 0.8) is mono-stable as compared to a paper unit cell of identical geometry, as shown by the representative curve in Figure 4d. The structure springs back upon reduction of the applied force because the crease energy overcomes the panel energy beyond the peak force, once again demonstrating the interplay 8 of their relative contributions to the mechanical response. The coupling between linear extension and rotation of flexigami cells enables their con- figuration as linear-rotary actuators; for example, a stack of end-constrained cells us used to drive a simple linear stage (Figure 5, Video S5). Rotation of the base of the flexigami stack results in rapid deployment of the boom assembly, due to sequential opening of the internal cells. In the specific case shown, we achieve a mean rotary to linear motion conversion of 10.8 mm/rad. The maximum deployed length depends on (N ,L,λ) and increases with each individually; variation in the (θ-H) relation with λ is shown in Figure 1d. Contrasting cylin- drical origami structures that offer continuous reversible deployment [8], flexigami actuators, depending on their cell geometry and stacking, can be deployed either continuously or in multiple stages where each stage presents a stable equilibrium position. An alternative means of deploying flexigami cells is by direct actuation of the hinges, which is achieved herein using a shape memory alloy (SMA) foil (Nitinol, 0.127 mm thickness (product # 045514), ThermoFisher Scientific). Strips of SMA foil were first folded in half and trained to lay flat when heated past their transition temperature of 80◦C. Three such strips were then glued to three alternate creases at the base of a six-sided cell, as shown in Figure 5c. The cell, compressed in its stable closed position, was then placed on a hot plate pre-heated to 200◦C. Within 10 seconds, the cell snaps to its stable extended position (Video S6), due to the force exerted by the foil hinges upon their phase transition. This simple prototype shows the possibility to use twist-coupled cells in self-deploying structures, and to pursue further concepts for reversible actuation such as using antagonistic actuators at opposite ends of the cell, or using shape-memory hinges with two-way behavior. The unique kinematics and properties of flexigami cells, including their extreme reversible strain, can also be considered in comparison to bulk cellular materials. As such, we assess how the effective mechanical properties, namely the elastic modulus (Fig. 5d) and peak stress (Fig. 5e), scale with the relative density which is determined by the cell geometry, material density, and thickness. The modulus and strength (i.e., the peak stress upon collapse) both scale with λ, and based on measurements of paper cells the properties are mutually maximized at the lowest N (4) and highest λ (0.9). The design parameters (λ, N ) govern the trajectory of panel bending and therefore scale the modulus which is related to the initial 9 loading of the panels and the peak stress which is determined by the maximum bending deformation. For the fabricated paper cells, the contribution of crease stiffness is negligible compared to panel bending; in the carbon fiber cells, similar scaling is observed but the crease stiffness is much larger leading to self-recovery after snap-through as shown earlier. Considered as a material, the modulus of flexigami cells exhibits near-linear (E ρ1.3) scal- ing with density, compared to conventional bend-dominated materials that exhibit quadratic scaling (E ρ2). The modulus and relative density of paper cells are in fact comparable to those of previously studied ultralight materials, including silica aerogels ([14],[39],[2]) and graphene elastomers ([26]), yet the sub-quadratic scaling is attractive for exploring the lower density regime. Using a finite element model, we also predict the properties of cells with metal panels with identical geometry and thickness; for example, cells with aluminum and stainless steel panels are predicted to have modulus and strength 2 and 3 orders of magnitude greater than paper cells, respectively, at comparable density. Fabrication of monolithic twist- coupled cellular materials using additive manufacturing techniques, both at microscopic and macroscopic scales ([1], [11]) , also warrants future work. Figures 10 Figure 1: Design of a twist-coupled kirigami (flexigami) cell and its kinematic analysis. a: Unfolded unit cell pattern with edges marked in black, red lines indicating valley folds, and blue lines indicating mountain folds. Insets: Photographs of unfolded cell and folded paper cells with N = 6, λ = 0.8 b: Matrix of cells with varying N, λ. c: Variation in γ as versus λ and compression stroke. d: Relative rotation of cell versus λ and compression stroke. 11 Figure 2: Tension-compression testing and finite element modeling of flexigami cells. a: Force- displacement reponse of paper cell (N = 7, L = 30mm) for λ = 0.5, 0.6, 0.7, 0.8, 0.9. b: Force-displacement response of corresponding finite element model, compared to experimental data. c: Photographs of paper cell during quasi-static compression, using fixture where bottom plate rotates freely and top plate rotation is fixed. Green lines are marked along the edges of triangular panels highlight the bending deformations and red lines marking the creases remain straight. d: Corresponding frames from finite element simulation; color scale indicates relative displacement magnitude. 12 Figure 3: Geometric mechanics analysis of flexigami unit cell. a: Geometric model with constraints and displacement boundary conditions indicated; panels are flexible and creases are modeled as torsional springs. b: Predicted energetics of the cell as the ratio of panel to crease rigidity is varied, revealing a transition from monostability to bistability. c: Contributions of panel (bending) and crease (folding) energies to the total energy for three values of KB/KC. d: Phase diagram indicating the transition between monostable to bistable as a function of KB/KC. 13 Figure 4: Mechanics of stacked flexigami cells according to their configuration and twist con- straint. a: Photographs of cell pairs glued back-to-back, with and without relative rotation permitted upon compression. b: Force-displacement curves corresponding to the images in (a) showing the same peak force in both cases, but simultaneous versus sequential snapping determined by the end constraint. c Identical responses of four-cell stacks with arbitrary arrangements of cells with (N = 7 and λ = 0.6, 0.7, 0.8, 0.9). d A small multi-layer array of cells fabricated from flexible carbon fiber sheets, with corresponding force- displacement curve of a monostable carbon fiber cell. 14 Figure 5: Mechanisms built from twist-coupled flexigami cells, and scaling of mechanical behav- ior. a,b: A linear axis motion system actuated by a stack of identical fleixgami cells constrained at both the ends. c A self-deploying cell using small ribbons of shape memory alloy attached to three of its bottom creases; the cell snaps open within 10 seconds after placement on a hot plate. d Scaling of elastic modulus with mass density, comparing flexigami cells to low-density materials from literature, and extrapolating via FEM to potential fabrication of cells using structural metals. e Relationships between relative density and peak stress at which the cell snaps to the collapsed state. 15 Acknowledgements Funding was provided by the National Science Foundation (EFRI-1240264) and by the U.S. Army Research Office through the Institute for Soldier Nanotechnologies under contract W911NF-13-D-0001. We thank Matt Shlian for providing paper samples and initial models of a stack of collapsible folded cells, Sterling Watson for assistance in prototyping, Megan Roberts for previous work on fabrication and testing, and Sanha Kim, Abhinav Rao, and Justin Beroz for insightful discussions and advice. Methods The CAD drawings of the 2D cut pattern of individual flexigami unit cells were prepared using AutoCAD . The patterns were cut from paper (Daler-Rowney canford; 150gm - 90lb, using a laser cutter (epilog mini 24), at 2% power, 10% frequency and 25% speed of the laser cutter. To complete the unit cell, the pattern is folded sequentially along the creases, and the tabs on the triangular panels are glued using staples dot roller to one of the polygonal surfaces to obtain a closed flexigami unit cell. A Zwick tensile testing machine with a 10kN load cell was used to carry out all the tensile tests being reported in this article. Figure. SI1 shows the experimental set-up. Here we see that the top platen is in series with the load cell. The bottom platen is fixed to a custom jig using c-clamps as shown. The custom made jig allows us to have rotation of the bottom surface of the unit cell with minimum inertia and has very high resistance against tilting because of any off centered loading on the rotating plate. This lets us record the force response of the unit cells accurately. Thumb screws located on the housing when tight- ened prevents the rotation of the plates and allow us to dynamically change the boundary conditions applied to unit cells with one single set-up. 16 Supplementary Information Figure SI1: Zwick mechanical testing machine used to perform all the tests reported in this articleTest set-up highlighting different components of the jigs used to which flexigami structures were attached. Inset shows the zoom-in view of the top and bottom plates to which flexigami structures are attached at four points. 17 Compression of seamless Kresling cylinders Seamless cells with different geometric parameters (N and λ) are subjected to uni-axial compression and tension tests. In Figure SI2 we show the force-displacement response of unit cells with N = 7 and λ = 0.8. Here we observe the response of the structure to compression in the first cycle is drastically different from its response to compression during subsequent cycles. Disturbances observed in the first compression cycle correspond to formation of creases or kinks on the triangular panels to relieve the strains developed in the process of compression. The peak force and stiffness of the structure drops substantially in subsequent cycles of compression and tension. Unit cells of varying λ for N = 7 are subjected to the same compression tension tests. Permanent deformation is less pronounced in the cells with lower λ in comparison with higher values which is evident from the tests (Figure SI2b). This confirms that required deformation to accommodate the compression increases with increasing λ and provision of cyclic cuts relieves the strains which would otherwise lead to formation of severe crumpling. Compression of Flexigami Cells Effect of loading rate and panel contact In order to confirm that the force response of the structures being tested is not affected by either the test speed or by the inertia of the rotating bottom plate, we carried out a series of tests on a unit cell with N = 4, L = 60mm, λ = 0.9. Figure. SI3 shows these results. Test speed is varied from 25mm/min to 400mm/min. All the responses are overlapped on each other and form a very tight band (Figure. SI3b). This confirms that the effect of test speed and inertia of the bottom rotating plate on the response of a unit cell is negligible. All the force displacement results are reported at 25mm/min test speed. Next to understand the role played by panel contact on the observed hysteresis, we again took a unit cell of N = 4, L = 60mm, λ = 0.9. First with all the sides intact, we subjected it to displacement controlled uni-axial tension compression tests at varied speeds. After testing the specimen under different test speeds, we removed one of the sides and repeated the process. Next we removed the second side such that the remaining two sides are alternate with no interactions between them as the unit cell is being compressed. Next we brought 18 Figure SI2: Displacement controlled uni-axial compression-tension test on seamless cells. a: Thee cycle test of a unit cell with N = 7 and λ = 0.8. b: Tensile force required to compress a unit cell with N = 7 and varying λ. c: Comparison between force-displacement response of open and seamless cells with N = 7 and λ = 0.8. Insets show the damage caused in seamless cylinders due to compression in a single cycle. 19 Figure SI3: Experimental data of displacement controlled uni-axial compressed tests on a unit cell. a: Tensile force required to compressed and expand unit cells with λ = 0.8&L = 30mm. b: Tensile force required to compress and expand a unit cell with N = 4, L = 60mm, λ = 0.9 under different test speeds. We see that when all the responses are overlaid, they form a very tight band and confirm that response of a unit cell is not significantly affected by the test speed. c: Force response of a unit cell with N = 4, L = 60mm, λ = 0.9 with 4,3,2,1 sides respectively. After each cycle of compression and tension, a side is removed from the unit cell and subjected to similar test conditions. Dwindling hysteresis and peak force is observed with reducing number of sides confirming the role of panel interactions and intrinsic hysteresis of the paper for being the reason for huge observed hysteresis. 20 it down to one side. All these tests are carried out under exactly same test parameters. Figure. SI3c shows the responses of these individual cases at 25mm/min. Here we observe that as the number of sides reduces, peak force required to compress the unit cell and the amount of hysteresis in a cycle reduces significantly. Especially when only one side is left, response of the unit cell in tension almost overlaps with the response in the compression part of the same cycle. So, we can affirmatively say that the large hysteresis observed in a unit cell is inherent to the structural components and the behavior of the paper itself but is not significantly affected by the hysteresis of testing machine. Tensile properties of paper Papers are made from cellulose fiber and during the process of manufacturing, the axes of the fibers tend to be aligned parallel to the paper flow through the paper machine. This phenomenon leads to anisotropy in the mechanical properties of paper and is generally considered to be orthotropic material. The thickness of the paper is much smaller compared to the other two in-plane directions. So, we can consider this as a case of plane stress. Under plane stress condition only the values of (E1,E2,ν12,G12,G13 and G23) are required to define an orthotropic material. The Poisson's ratio ν21 is implicitly given as ν21 = (E2/E1)ν12. The stress-strain relations for the in-plane components of the stress and strain are of the form  ε1 ε2 γ12  =  1/E1 −ν12/E1 0 −ν12/E1 1/E2 0 0 0 1/G12   σ11 σ22 τ12  (7) Specimens conforming to ISO standards were cut in three different orientations as shown in Figure. SI4a using a desktop mini Epilog laser cutter. 10 samples are cut in each direction with the specified dimensions and are speckled with silver sharpie to be able to use Digital Image Correlation (DIC) technique for the estimation of in-plane strains. In order to determine the in plane modulus and Poisson's ratio, we should have informa- tion about strains developed in the specimen in the longitudinal and transverse directions. Since, attaching a strain gauge would significantly affect the properties of paper and would result in only one data point, we used the Digital Image Correlation (DIC) which is a non- contact optical strain measurement technique. 21 Figure SI4: a: Representative figure of the directions along which the samples are cut from a Daler-Rowney Canford paper and the dimensions of the sample conforming to ISO standards for paper testing. b: Speckled sample before the test and after breakage along with representative set of axial strain developed at the maximum extension. c: Stress-Strain response of the paper strips in the three measured directions. 22 Tests were conducted under displacement controlled conditions where the specimen was pulled at a constant velocity of 20mm/min with a 10KN load cell. A series of images of the specimen being deformed are captured and analyzed with Vic-2D. This provides us with the data of in-plane strains (Figure. SI4b). All the samples that break within 10mm of the clamping distance are rejected to meet with the ISO standards. In-plane stress-strain curves of the Daler-Rowney canford (150gsm) paper are represented in Figure. SI4c . Table 1 lists the average Young's modulus along the two major directions (Machine direction and Cross direction) as well as shear modulus (G12) and measured Pois- son's ration ν12 Modulus (GPa) MD (E1) 6.83 CD (E2) 3.11 G12 ν12 2.17 0.23 Table 1: Mean values of the modulus and Poisson's ratio Finite-element simulations In the finite element model presented in the manuscript, the triangular panels are modeled as thin shell structures whose behavior is defined by its shell thickness, material density, and the in plane material properties as previously discussed. These properties also determine the individual bending rigidity (Kb) of the triangular panels. Both mountain and valley folds are modeled as linear elastic torsional springs whose behavior is determined by the torsional spring constant (Kc). Element type of STRI3 in Abaqus/Standard is used are used to mesh the triangular panels along with top and bottom polygonal surfaces. The element has three nodes, each with six degrees of freedom. The strains are based on thin plate theory, 23 using small-strain approximation. Total of 13,960 STRI3 elements are used to mesh top and bottom plates; while, 6568 STRI3 elements are used to mesh each of the triangular panels. To model creases we use special purpose spring elements whose associated action is defined by the specified degrees of freedom involved. Type I and Type II creases are discretized into 9 and 17 points where these spring elements are acting. Specimen is then subjected to vertical displacement while fixing bottom plane (Fig- ure. SI1) and imposing penalty to avoid node penetration between the two triangular panels which would come in contact in the process of compression. External work done on the entire specimen and reaction forces at each and every node (from individual frames) on top and bottom panels of the specimen are obtained as output from the FE model. Vector summation of these reaction force components results in the force-displacement curve. Geometric mechanics model Below, we detail the procedure of computing bending energy of panels which are consid- ered to be developable as well as energy stored in the creases which are modeled as torsional springs. Bending Energy At any instantaneous height, we have two triangular panels and three creases correspond- ing to each of the N sides. We need to parameterize the equations of these space curves to find the energy stored in each of them. Let the co-ordinates of each of the points be defined as follows • Point A: [Ax, Ay, Az]; Point B: [Bx, By, Bz]; • Point A' :(cid:2)A(cid:48) x, A(cid:48) y, A(cid:48) z (cid:3); Point B' :(cid:2)B(cid:48) x, B(cid:48) (cid:3) (cid:104)−−→ AA(cid:48)s, (cid:80)3 y, B(cid:48) z (cid:105) Space curve RAA(cid:48)(s) is parameterized as: system of ( (cid:126)e(cid:48) 1, (cid:126)e(cid:48) 2, (cid:126)e(cid:48) 3) n=1 ansin(nπs), 0 in the co-ordinate By co-ordinate transformation, representation of RAA(cid:48)(s) in the co-ordinate system of ( (cid:126)e1, (cid:126)e2, (cid:126)e3) is : 24 Q31 Q32 Q33 Q21 Q22 Q23 Q11 Q12 Q13 Q11−−→ Q21−−→ Q31−−→ AA(cid:48) s + Q12 AA(cid:48) s + Q22 AA(cid:48) s + Q32   −−→ AA(cid:48)t λsin(nπt) 0  +  A(cid:48) A(cid:48) A(cid:48) x y z (cid:80)3 (cid:80)3 (cid:80)3 n=1 ansin(nπs) + A(cid:48) n=1 ansin(nπs) + A(cid:48) n=1 ansin(nπs) + A(cid:48) y z x  RF E = = where components of the transformation matrix Q are given by : Qij = ei · e(cid:48) j Similarly, space curve RAB(cid:48)(s) in the co-ordinate system ( (cid:126)e1, (cid:126)e2, (cid:126)e3) is given by  RAB(cid:48)(s) = B(cid:48) x + s(Ax − B(cid:48) x) y + s(Ay − B(cid:48) B(cid:48) y) z + s(Az − B(cid:48) B(cid:48) z) (8) (9) Parametric equation of the triangular panel formed AA(cid:48)B(cid:48) is given by (10) Similarly for the triangular panel formed by the three points BB(cid:48)A, it is parametrically RAA(cid:48)B(cid:48)(s, v) = (1 − v)RAA(cid:48)(s) + vRAB(cid:48)(s) written as RBB(cid:48)A(s, v) = (1 − v)RBB(cid:48)(s) + vRAB(cid:48)(s) (11) Let HAA(cid:48)B(cid:48) and HBB(cid:48)A(cid:48) represent the mean curvature of the two surfaces, the total bending energy of them is: Bending Energy = (cid:90) (cid:90) (cid:2)H 2 KB AA(cid:48)B(cid:48)(s, v) + H 2 BB(cid:48)A(cid:48)(s, v)(cid:3) ds dv (12) 25 Crease Energy For each of the N sides, we have three creases: two creases correspond to mountain folds joining each of the triangular panels to either of the polygonal surfaces; and one Valley fold which is the diagonal crease joining the two triangular panels. Creases are modeled as torsional springs with non-zero rest state which corresponds to zero crease energy. This non-zero rest state is the state in which a unit cell is in a fully open configuration. At any instantaneous height h, cosine of the angle at any particular point on the crease is defined as dot product of the normals drawn on the two surfaces joining at that point. Following equations explain the procedure • Gradient of RAA(cid:48)B(cid:48) with respect ot v is (RAA(cid:48)B(cid:48))v = −RAA(cid:48)(s) + RAB(cid:48)(s) − (RAA(cid:48)B(cid:48))v (t, v) = Q11−−→ B(cid:48) AA(cid:48) t + Q12 Q21−−→ AA(cid:48) t + Q22 Q31−−→ AA(cid:48) t + Q32 x + s(Ax − B(cid:48) x) y + s(Ay − B(cid:48) B(cid:48) y) z + s(Az − B(cid:48) B(cid:48) z) • Gradient of RAA(cid:48)B(cid:48) with respect ot s is + (cid:80)3 (cid:80)3 (cid:80)3  n=1 ansin(nπs) + A(cid:48) n=1 ansin(nπs) + A(cid:48) n=1 ansin(nπs) + A(cid:48) x (cid:80)3 (cid:80)3 (cid:80)3 AA(cid:48) + Q12 AA(cid:48) + Q22 AA(cid:48) + Q32 n=1 nanπcos(nπs) n=1 nanπcos(nπs) n=1 nanπcos(nπs) Rs(s, v) = (1 − v) Q21−−→ Q31−−→ Q11−−→  Ax − B(cid:48) Ay − B(cid:48) Az − B(cid:48) x y z + v 26  y z  (13) (14) • Equation of normal (N) to the tangent plane at a point (s, v) on the surface (RAA(cid:48)B(cid:48)) is given by NAA(cid:48)B(cid:48) = (RAA(cid:48)B(cid:48))s × (RAA(cid:48)B(cid:48))t (RAA(cid:48)B(cid:48))s × (RAA(cid:48)B(cid:48))t (15) • Similarly, equation of the normal (N) to the tangent plane at a point (s, v) on the surface (RBB(cid:48)A) is: NBB(cid:48)A = (RBB(cid:48)A)s × (RBB(cid:48)A)t (RBB(cid:48)A)s × (RAA(cid:48)B(cid:48))t (16) To compute crease energy of a crease, we need to track the change in the angle of the crease along its length and integrate it. Following equations give the instantaneous angle at each of the three creases and energy stored in them • Energy stored in the crease A(cid:48)B(cid:48) at height h CreaseEnergyA(cid:48)B(cid:48) = [(θA(cid:48)B(cid:48))h(0, v) − (θA(cid:48)B(cid:48))0(0, v)]2 dv f racKC2 where (θA(cid:48)B(cid:48))h(0, v) = acos ((NAA(cid:48)B(cid:48))h(0, v) · (cid:126)e3) • Energy stored in the crease AB(cid:48) at height h (cid:90) B(cid:48) A(cid:48) (cid:90) B(cid:48) A Crease EnergyAB(cid:48) = [(θAB(cid:48))h(s, 1) − (θAB(cid:48))0(s, 1)]2 ds KC 2 where (θAB(cid:48))h(t, 1) = acos ((NAA(cid:48)B(cid:48))h(s, 1) · (NBB(cid:48)A)h(s, 1)) • Energy stored int he crease AB at height h (cid:90) B CreaseEnergyAB = KC 2 A [(θAB)h(0, v) − (θAB)0(0, v)]2 dv where (θAB)h(0, v) = acos ((NBB(cid:48)A)h(0, v) · (cid:126)e3) 27 (17) (18) (19) Crease Energy = CreaseEnergyA(cid:48)B(cid:48) + Crease EnergyAB(cid:48) + CreaseEnergyAB So, for an N sided unit cell, total energy of the system ET otal = N [Bending Energy + Crease Energy] ET otal =N AA(cid:48)B(cid:48)(s, v) + H 2 (cid:90) (cid:90) (cid:2)H 2 (cid:90) B(cid:48) (cid:90) B(cid:48) (cid:90) B KC 2 KC 2 KC 2 A(cid:48) A A + N + N + N BB(cid:48)A(cid:48)(s, v)(cid:3) ds dv [(θA(cid:48)B(cid:48))h(0, v) − (θA(cid:48)B(cid:48))0(0, v)]2 dv [(θAB(cid:48))h(s, 1) − (θAB(cid:48))0st, 1)]2 ds [(θAB)h(0, v) − (θAB)0(0, v)]2 dv (20) (21) Minimization of ET otal at each height h provides us with the optimum values of (ai; i = 1, 2, 3) which are then used to computed RAA(cid:48), RBB(cid:48), RAA(cid:48)B(cid:48) and RBB(cid:48)A at that particular h. 28 References [1] Bauer, J., Hengsbach, S., Tesari, I., Schwaiger, R., Kraft, O., 2014. High-strength cellular ceramic composites with 3d microarchitecture. Proceedings of the National Academy of Sciences 111, 2453–2458. [2] Biener, M.M., Ye, J., Baumann, T.F., Wang, Y.M., Shin, S.J., Biener, J., Hamza, A.V., 2014. Ultra-strong and low-density nanotubular bulk materials with tunable feature sizes. Advanced Materials 26, 4808–4813. [3] Cai, J., Deng, X., Feng, J., Zhou, Y., . Geometric design and mechanical behavior of a deployable cylinder with Miura origami. Smart Materials and Structures 24, 125031. [4] Cheng, N.G., Gopinath, A., Wang, L., Iagnemma, K., Hosoi, A.E., 2014. Thermally tunable, self-healing composites for soft robotic applications. Macromolecular Materials and Engineering 299, 1279–1284. [5] Cheung, K.C., Tachi, T., Calisch, S., Miura, K., 2014. Origami interleaved tube cellular materials. Smart Materials and Structures 23, 094012. [6] Dias, M.A., 2012. Swelling and folding as mechanisms of 3D shape formation in thin elastic sheets. Dissertations Paper 637. [7] Felton, S., Tolley, M., Demaine, E., Rus, D., Wood, R., 2014. A method for building self- folding machines. Science 345, 644–646. http://science.sciencemag.org/content/ 345/6197/644.full.pdf. [8] Filipov, E.T., Tachi, T., Paulino, G.H., 2015. Origami tubes assembled into stiff, yet reconfigurable structures and metamaterials. Proceedings of the National Academy of Sciences 112, 12321–12326. http://www.pnas.org/content/112/40/12321.full.pdf. [9] Guest, S.D., Pellegrino, S., 1994a. The Folding of Triangulated Cylinders, Part I: Geometric Considerations. ASME Journal of Applied Mechanics 61, 773 – 777. [10] Guest, S.D., Pellegrino, S., 1994b. The Folding of Triangulated Cylinders, Part II: The Folding Process. ASME Journal of Applied Mechanics 61, 778–783. 29 [11] Horn, J.T., Harrysson, L.O., Marcellin-Little, J.D., West, A.H., Lascelles, X.D., Aman, R., 2014. Flexural properties of ti6aal4v rhombic dodecahedran open cellular structures fabricated with electron beam melting. Additive Manufacturing 1, 2–11. [12] Hunt, G.W., Ario, I., 2005. Twist buckling and the foldable cylinder: An exercise in origami. International Journal of Non-Linear Mechanics 40, 833–843. [13] Jianguo, C.X., Ya, Z.D., Jian, F., Yongming, T., 2015. Bistable Behavior of the Cylin- drical Origami Structure With Kresling Pattern. ASME Journal of Mechanical Design 137, 061406–8. [14] Kucheyev, S.O., Stadermann, M., Shin, S.J., Satcher, J.H., Gammon, S.A., Letts, S.A., van Buuren, T., Hamza, A.V., 2012. Super-compressibility of ultralow-density nanoporous silica. Advanced Materials 24, 776–780. [15] Lamoureux, A., Lee, K., Shlian, M., Forrest, S.R., Shtein, M., 2015. Dynamic kirigami structures for integrated solar tracking. Nature Communications 6. [16] Lazarus, A., Reis, P.M., 2015. Soft actuation of structured cylinders through auxetic behavior. Advanced Engineering Materials 17, 815–820. [17] Lechenault, F., Thiria, B., Adda-Bedia, M., 2014. Mechanical response of a creased sheet. Phys. Rev. Lett. 112, 244301. [18] Love, A.E.F., 1994. A Treatise on the Mathematical Theory of Elasticity. Dover Pub- lications, New York. [19] Lv, C., Krishnaraju, D., Konjevod, G., Yu, H., Jiang, H., 2014. Origami based Mechan- ical Metamaterials. Scientific reports 4, 5979. [20] Martinez, R.V., Branch, J.L., Fish, C.R., Jin, L., Shepherd, R.F., Nunes, R.M.D., Suo, Z., Whitesides, G.M., 2013. Robotic tentacles with three-dimensional mobility based on flexible elastomers. Advanced Materials 25, 205–212. 30 [21] Martinez, R.V., Fish, C.R., Chen, X., Whitesides, G.M., 2012. Elastomeric origami: Programmable paper-elastomer composites as pneumatic actuators. Advanced Func- tional Materials 22, 1376–1384. [22] Na, J.H., Evans, A.A., Bae, J., Chiappelli, M.C., Santangelo, C.D., Lang, R.J., Hull, T.C., Hayward, R.C., 2015. Programming reversibly self-folding origami with micropat- terned photo-crosslinkable polymer trilayers. Advanced Materials 27, 79–85. [23] Nicholas, P., Takashi, M., 2010. Shape Interrogation for computer aided design and manufacturing. Springer. [24] Nojima, T., 2002. Modelling of Folding Patterns in Flat Membranes and Cylinders by Using Origami. JSME International Journal Series C 45, 364–370. [25] Overvelde, J.T.B., Kloek, T., Dhaen, J.J.a., Bertoldi, K., 2015. Amplifying the response of soft actuators by harnessing snap-through instabilities. Proceedings of the National Academy of Sciences 112, 10863–10868. [26] Qiu, L., Liu, J.Z., Chang, S.L.Y., Wu, Y., Li, D., 2012. Biomimetic superelastic graphene-based cellular monoliths. Nature Communications 3, 1241. [27] Rafsanjani, A., Akbarzadeh, A., Pasini, D., 2015. Snapping mechanical metamaterials under tension. Advanced Materials 27, 5931–5935. [28] Rus, D., Tolley, M.T., 2015. Design fabrication and control of soft robots. Nature 521, 467–475. [29] Schenk, M., Guest, S.D., 2013. Geometry of Miura-folded metamaterials. Proceedings of the National Academy of Sciences of the United States of America 110, 3276–81. [30] Schenk, M., Kerr, S.G., Smyth, a.M., Guest, S.D., 2013. Inflatable Cylinders for De- ployable Space Structures. Proceedings of the First Conference Transformables . [31] Schenk, M., Viquerat, A.D., Seffen, K.a., Guest, S.D., 2014. Review of Inflatable Booms for Deployable Space Structures: Packing and Rigidization. Journal of Spacecraft and Rockets 51, 762–778. 31 [32] Shan, S., Kang, S.H., Raney, J.R., Wang, P., Fang, L., Candido, F., Lewis, J.A., Bertoldi, K., 2015. Multistable architected materials for trapping elastic strain energy. Advanced Materials 27, 4296–4301. [33] Shim, J., Perdigou, C., Chen, E.R., Bertoldi, K., Reis, P.M., 2012. Buckling-induced encapsulation of structured elastic shells under pressure. Proceedings of the National Academy of Sciences 109, 5978–5983. [34] Silverberg, J.L., Na, J.h., Evans, A.A., Liu, B., Hull, T.C., Santangelo, C.D., Lang, R.J., Hayward, R.C., Cohen, I., 2015. Origami structures with a critical transition to bistability arising from hidden degrees of freedom. Nature Materials 14, 389–393. [35] Tachi, T., 2009. One-DOF cylindrical deployable structures with rigid quadrilateral panels. Proceedings of the International Association for Shell and Spatial Structures Symposium . [36] Tachi, T., Miura, K., 2012. Rigid foldable cylinders and cells. Journal of the Interna- tional Association for Shell and Spatial Structures 53, 217–226. [37] Viquerat, A., Schenk, M., Sanders, B., Lappas, V., 2014. Inflatable Rigidisable Mast For End-Of-Life Deorbiting System. European Conference on Spacecraft Structures, Materials and Environmental Testing (SSMET) , 1–10. [38] Waitukaitis, S., Menaut, R., Chen, B.G.g., van Hecke, M., 2015. Origami multistability: From single vertices to metasheets. Phys. Rev. Lett. 114, 055503. [39] Worsley, A.M., Kucheyev, O.S., Satcher, H.J., Hamza, V.A., Baumann, F.T., 2009. Me- chanically robust and electrically conductive carbon nanotube foams. Applied Physics Letters 94, 073115. [40] Yang, D., Mosadegh, B., Ainla, A., Lee, B., Khashai, F., Suo, Z., Bertoldi, K., White- sides, G.M., 2015. Buckling of elastomeric beams enables actuation of soft machines. Advanced Materials 27, 6323–6327. 32
1809.09862	1	1809	2018-09-26T09:22:36	Nanostructured p-p(v) junctions obtained by G-doping	[ "physics.app-ph", "cond-mat.mes-hall" ]	Recently, geometry-induced quantum effects in periodic Si nanostructures were introduced and observed. Nanograting has been shown to originate geometry induced doping or G-doping. G-doping is based on reduction of density of quantum states and is n-type. Here, fabrication and characterization of G-doping based compensated p-p(v) junctions is reported. The p-p(v) abbreviation is introduced to emphasize voltage dependence of G-doping level. The p-type Si wafer is used for sample fabrication. First, two square islands are shaped at the surface of the wafer. Next, entire Si surface is oxidized to grow thin SiO2 insulating layer. Two windows are opened in SiO2 and nanograting is fabricated inside one of the windows using laser interference lithography followed by reactive ion etching. G-doping p-p(v) junction between p-type substrate and the nanograting layer is formed. Next, metal contacts are deposited on both islands to measure electrical characteristics of p-p(v) junction. To exclude contact resistance input Van der Pauw method is used. Obtained I-V curves are diode type with extremely low voltage drop in forward direction and reduced reverse current. Experimental I-V curves are fitted to the Shockley equation and ideality factor is found to be in the range of 0.2-0.14. Such a low values confirm that p-p(v) junction is fundamentally different from the conventional p-n junction. This difference is ascribed to G-doping which, unlike conventional doping, is external voltage dependent.	physics.app-ph	physics	Nanostructured p-p(v) junctions obtained by G-doping A. Tavkhelidze 1 , L. Jangidze 2 and G. Skhiladze 2 I Ilia State University, Cholokashvili Ave. 3-5, 0162 Tbilisi, Georgia 2 Institute of Micro and Nano Electronics, Chavchavadze Ave. 13, 0179 Tbilisi, Georgia E-mail: [email protected] Abstract Recently, geometry-induced quantum effects in periodic Si nanostructures were introduced and observed. Nanograting has been shown to originate geometry induced doping or G-doping. G-doping is based on reduction of density of quantum states and is n-type. Here, fabrication and characterization of G-doping based compensated p-p(v) junctions is reported. The p-p(v) abbreviation is introduced to emphasize voltage dependence of G-doping level. The p-type Si wafer is used for sample fabrication. First, two square islands are shaped at the surface of the wafer. Next, entire Si surface is oxidized to grow thin SiO2 insulating layer. Two windows are opened in SiO2 and nanograting is fabricated inside one of the windows using laser interference lithography followed by reactive ion etching. G- doping p-p(v) junction between p-type substrate and the nanograting layer is formed. Next, metal contacts are deposited on both islands to measure electrical characteristics of p-p(v) junction. To exclude contact resistance input Van der Pauw method is used. Obtained I-V curves are diode type with extremely low voltage drop in forward direction and reduced reverse current. Experimental I-V curves are fitted to the Shockley equation and ideality factor is found to be in the range of 0.2-0.14. Such a low values confirm that p-p(v) junction is fundamentally different from the conventional p-n junction. This difference is ascribed to G-doping which, unlike conventional doping, is external voltage dependent. Keywords: nanostructuring, p-n junction, semiconductor, doping 1. Introduction in interference Present developments lithography enabled fabrication of periodic nanostructures [1, 2]. Semiconductor nanograting (NG) layers have been introduced and fabricated [3,4]. The periodic structure, have been shown to dramatically change the electronic [4], thermoelectric [5], optical [6] and electron emission [7] properties when the nanograting dimensions are low enough to approach the de Broglie wavelength of electrons. This is due to the extraordinary boundary conditions imposed by a NG on the electron wave function. They make forbidden definite quantum states [8], and density of quantum states reduce. Electrons rejected from the valence band occupy empty quantum states in the conduction band. The electron concentration in the conduction band enhance, which is named as geometry-induced electron doping or G-doping [3]. It is equal to donor doping from the point of view of the raise in n and Fermi energy. However, there are no ionized impurities. 1 The NG geometry belongs to a class of nonintegrable quantum systems (quantum billiards) which are widely investigated [9, 10]. doping. induced There are some experimental results demonstrating nanostructure disordered nanostructures obtained by wet etching of p-Si [11] both n-type and p-type doping was observed. Periodic nanostructures made by laser radiation interaction with surfaces of Si, Ge and SiGe crystals demonstrate n-type doping [12]. In mesoporous p-Si charge carriers disappear [13]. G-doping mechanism can be employed to explain these results. In Thin Si nanograting layers have been studied experimentally. It was found, that resistivity and Hall voltage have metal type temperature dependences [4]. Dielectric function determined by ellipsometry show metallization Intensive photoluminescence was observed in Si in spite of indirect band gap [14]. layer [6]. the of fabrication and In this work, we report on investigation of G-doped p-p(v) junctions. 2. Sample Preparation and Characterization Silicon wafers p-type were used sample preparation. Wafers were p-type (Boron) with resistivity 1-10 Ω x cm and thickness 200 microns. Samples were 10mm x20mm chips with NG and plain islands and corresponding metal contacts. On the reference plain side NG was not formed. Figure 1 shows simplified process flow (only the NG side is shown). for Figure 1. The schematically the process flow for G-doped solar cell preparation: following steps: The G-doped solar cells were prepared in the 1. Back contact p+ type was made by Boron diffusion at 1050 0C -- 10 minutes. Diffusion depth was 0.7-0.8 µm. Next, square island of 2.2mm x 2.2mm and height of 0.2-0.4 µm was formed on the front side of the wafer using photolithography. Next, SiO2 insulator layer was grown using wet thermal oxidation 1000 0C during 10 minutes. Thickness of SiO2 layer was 150-170 nm. Next, 0.6 µm Al contact was deposited on the back side and backed at 540 0C during 12 minutes. 2. The 2mmx2mm window was opened in SiO2 layer using photolithography and wet etching of SiO2. the window using 3. The NG with line width of 150 nm was fabricated inside interference lithography [4] and subsequent reactive ion etching. First, negative photo resist ma-N 2401 was applied to the sample. Next, laser interference laser 2 lithography was done using Blue-Violet (375 nm) semiconductor laser as a coherent light source. The 25-35 nm depth NG was made using reactive ion etching in CF4. 4. Contacts were made by Al thermal evaporation and Ti\Ag magnetron sputtering. Conventional lift-off technology was used in both cases. The Ti\Ag films were grown using DC magnetron sputtering during the single vacuum process at a substrate temperature of 250 °C [4]. Ti\Ag contacts were not backed. Current-voltage (I-V) curves were recorded using both the 4-probe Van der Pauw method and 2-probe method. Using the 2-probe method was necessary, as the 4-probe method alone does not provide information about contacts. A Keysight multimeters 34410A and 34410 were used to record 4-probe and 2-probe voltages, and Keysight E3647A current source was used to apply current. To exclude input of thermo powers voltage shifts (typically <10-4 V) were compensated in the process of I- V curve building. 3. Results and discussion Typical I-V curve of the p-p(v) junction (blue) is shown in Fig. 2. In the same figure I-V curve of reference plain junction is plotted. It is evident that NG produce diode type I-V curve and reference plain is close to ohmic contact. Forward voltage ] A m [ T N E R R U C Reference plain 10 8 6 4 2 0 Nanograting -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 VOLTAGE [V] Figure 2. Typical I-V characteristics of NG and reference plain (sample M154). is very low with respect to Si diode and forward currents are quite high. The p-p(v) junction is formed thanks to reduction of quantum state density near the NG. Effective density of states ( CN ) in both valence and conduction bands reduce G times, where G is geometry VN and factor [3]. Consequently, internal carrier concentration near the NG also reduces NG n i = Gn i / (1) in ). However, mass (follows from the definition of action law rewritten as Forward 10000 1000 100 np NG NG = 2 / Gn i 2 (2) 0 J / J p should be satisfied. This leads to reduction of hole concentration near the nanograting pNG < . In our opinion, such a reduction of hole concentration is the reason for p-p(v) junction formation. The same I-V curve (nanograting in Fig.2), normalized and plotted in semilog scale is shown in Fig. 3. In both forward and reverse directions I-V curve is diode type. In forward direction current increases very rapidly following Shockley equation = exp( eV / η KT 1) − 0 (3) JJ / 14.0=η . Here, e is electron charge; with ideality factor V is applied voltage K is Boltzman constant; T is absolute temperature. Fig. 3 clearly indicates that p-p(v) junction is fundamentally different from p-n junction (where ). Forward current increases much faster than it 21−=η would be even for the ideal p-n junction. This can be . Such explained by voltage dependence of dependence follows from the voltage dependence of the p(v) layer thickness (layer in close proximity to the NG with reduced quantum state density). (VGG = ) 10 1 0.1 exp(7eV/KT)-1 Reverse 0 1 2 3 eV/KT 4 5 Figure 3. The I-V characteristics of p-p(v) junction (sample was used for normalization. M154). The value 14.0=η Physical mechanism responsible for such dependence is unclear yet. One of the explanations is charge accumulation layer thickness dependence on applied voltage. Forward voltages are in the range 20-60 mV and are much lower than for silicon p-n junctions and even considerably lower than for Si Schottky junctions [17]. The reverse current increases with voltage faster than it would be junction, but considerably slower than it would be for the real silicon p- n junction [16, p. 97]. the case of ideal p-n Saturation current density, forward and reverse in currents for 9 samples are given it Table 1. Table 1. Parameters of p-p(v) junctions. Last column indicates how much p-p(v) junction reverse current is reduced with respect to silicon p-n junction. Value of reverse ( JJ / ) =−np 0 1000 for eV / =KT 30 is taken from [16, p. 97]. Sample # η 0J [µA/cm2] M153 M154 M157 M159 M160 M164 M166 M168 M174 eV / KT 0.2 0.14 0.2 0.17 0.2 0.18 0.17 0.17 0.17 15 77 26 7.5 27 8.2 17 6.9 23 ≡α Forward / JJ at 1* =α 0 100 1000 250 900 100 200 300 400 2000 Reverse / JJ 1=α at 0 Reverse / JJ 10=α 0 at Reverse / JJ 30=α 0 at ( 15 7 20 20 30 15 7 15 10 35 15 40 - 70 - 10 20 20 2 2 4 5 5 2 2 7 3 3 0 ) vpp )( − ) Reverse JJ / 0 JJ / ( np − 30=α at 0.035 0.015 0.04 - - 0.07 0.01 0.02 0.02 0 Ideality factor is quite low and keeps inside the relatively narrow range η= 0.14-0.2. Saturation current 0J varies from sample to sample. The same is density true for forward current, while reverse currents do not vary much. Variation of saturation current can be ascribed to the variation of carrier concentration in p-Si (1-10 Ω x cm) wafers. Large variation of normalized forward current can be ascribed to the exponential dependence / JJ equation (3). Relative stability of reverse currents can be ascribed to saturation which also follows from equation (3). Consequently, we find good overall matching of experimental data to the equation (3). However, ideality factor is less than 1 and is very low (in the range η= 0.14-0.2). Our explanation is dependence of G-doping level on applied voltage. Saturation current density 0J is of order of 10-5 A cm- 2 which is many orders of magnitude higher with respect to silicon p-n junction (typically 10-10 A cm-2 ). Our 0J is explanation is that in p-p(v) type junction determined by majority carrier density instead of minority carrier density (as for p-n junction). The reverse current voltage dependence for two samples is show in Fig. 4. Reverse currents increase 100 M160 0 J / J E S R E V E R 10 1 10000 1000 0 J / J D R A W R O F 100 10 1 0.1 0.1 0 20 40 M157 M157 M160 eV/KT 1 2 100 120 140 160 80 eV/KT 0 60 Figure 4. Reverse current voltage dependences (red) for samples M157 and M160. Corresponding forward current dependences (green) are given in the insert. ) 0 ) ) 0 << for junction, resulting to silicon p-n JJ / np − (Table for smoothly at least until 4V reverse bias and are much less in with respects value JJ /( / ( vpp 0 )( − reverse eV =KT 30 / is found in [16, p. JJ ( / 1000 =−np 97]. Our explanation is following. In the case of G- equation (1). doping, we have reduced value of However, diffusion current which is defined by majority carrier concentration (instead of minority for p-n junction) 1 1). Value =KT / the of NG in eV 30 4 still dominates over generation current. It happens despite Si material and relatively low temperature (T=300K). Consequently, reverse current is less dependent on carrier generation and normalized reverse current is much less with respect to silicon p-n junction. Further experiments are required to improve ohmic contacts, record C-V curves and measure high frequency characteristics. Conclusions We fabricate and characterize G-doping based p-p(v) junctions. The p-type Si wafer (1-10 Ω x cm) was used for sample fabrication. G-doped p-p(v) junction has been formed between the p-type substrate and nasnograting with line width of 80-150 nm and depth of 25-35 nm. Measured I-V curves were diode type with extremely low voltage drop in forward direction and reduced reverse currents. Experimental I-V curves were fitted to the Shockley equation and ideality factor was found to be in the range of 0.2-0.14. Such a low values of ideality factor confirm that p-p(v) junction is fundamentally different from the conventional p-n junction. We attribute this difference to G-doping which, unlike conventional doping, is voltage dependent. Data collected from 9 samples indicate that saturation currents are several magnitudes higher with respect to silicon p-n junction. This is explained by key contribution of majority carriers (holes) is saturation current. Reverse currents are 1-2 order of magnitude lower with respect to silicon p-n junction which is explained by leading role of diffusion current. Forward voltage drops are 30-60 mV and are considerably lower with respect to Schottky diode. Normalized reverse currents are much less with respect to p-n junction (at least for reverse bias up to 4V). The p- p(v) diode has certain advantages over p-n and Schottky diodes and can find wide applications in power electronics and ultra high frequency electronics as well as in conventional electronics and solar cells. Acknowledgments We thank E. A. Katz, N. Gorj, I. Shah and Z. Taliashvili, for discussion and support. The authors thank the SRNSF (MTCU/91/3-250/15) and STCU (project 6191) for providing funding. References [1] Chang E-C, Mikolas D, Lin P-L, Schenk T, Wu C-L, Sung C-K and Fu C-C, 2013 Nanotechnology 24 455301 [2] Zuppella P, Luciani D, Tucceri P, DeMarco P, Gaudieri A, Kaiser J, Ottaviano L, Santucci S and Reale A Nanotechnology 2009 20 115303 [3] Tavkhelidze A 2014 Physica E 60 4 [4] Tavkhelidze A. Jangidze L, Mebonia M. Piotrowski K, Więckowski J, Taliashvili Z, Skhiladze G and Nadaraia L 2017 Physica Staus Solidi A 214 1700334 [5] Tavkhelidze A 2009 Nanotechnology 20 405401 [6] Mamedov N, Tavkhelidze A, Bayramov A, Akhmedova K, Aliyeva Y, Eyyubov G, Jangidze L, and Skhiladze G 2017 Phys. Status Solidi C 1700092 [7] Tavkhelidze A 2010 J. Appl. Phys.108 044313 [8] Kakulia D, Tavkhelidze A, Gogoberidze V, Mebonia M [9] H.-J. Stockmann, Quantum Chaos (Cambridge University 2016 Physica E 78 49 Press, UK, 2000). 94 30004 [12] Medvid A, Onufrijevs P, Jarimaviciute-Gudaitiene R, Dauksta E. andProsycevas I, 2013 Nanoscale Research Letters 8 264 [13] Polisski G, Kovalev D, Dollinger G, Sulima T, Koch F 1999 Physica B 273-274 951-954 [14] Tavkhelidze A, Bayramov A, Aliyeva Y, Jangidze L, Skhiladze G, Asadullayeva S, Alekperov O, Mamedov N, 2017 Phys. Status Solidi C, 1700093 [15] Tavkhelidze A, Jangidze L, Mebonia M, Skhiladze G, Ursutiu D, Samoila C, Taliashvili Z and Nadaraia L 2016 Energy Procedia 92 896 [16] Sze S M and Ng K K, Physics of Semiconductor Devices [17] Jie Xing, Kuijuan Jin, Meng He, Huibin Lu, Guozhen Liu and Guozhen Yang, J. 2008 Phys. D: Appl. Phys. 41 195103 [10] Backer A, Ketzmerick R, Lock S andSchanz H 2011 EPL (New Jersey: Wiley-Interscience, 2007). [11] Luchenko A I, Melnichenko N N, Svezhentsova K V, proceedins of Intermatic 2012 conference, 3-7 Dec. 2012 Moscow, part 1, p. 104 5
1908.09272	1	1908	2019-08-25T08:15:54	Mask-less Patterning of Gallium-irradiated Superconducting Silicon Using Focused Ion Beam	[ "physics.app-ph", "cond-mat.supr-con" ]	A direct patterning technique of gallium-irradiated superconducting silicon has been established by focused gallium-ion beam without any mask-based lithography process. The electrical transport measurements for line and square shaped patterns of gallium-irradiated silicon were carried out under self-field and magnetic field up to 7 T. Sharp superconducting transitions were observed in both patterns at temperature of 7 K. The line pattern exhibited a signature of higher onset temperature above 10 K. A critical dose amount to obtain the superconducting gallium-irradiated silicon was investigated by the fabrication of various samples with different doses. This technique can be used as a simple fabrication method for superconducting device.	physics.app-ph	physics	Mask-less Patterning of Gallium-irradiated Superconducting Silicon Using Focused Ion Beam Ryo Matsumotoa,b, Shintaro Adachia, El Hadi S. Sadkic, Sayaka Yamamotoa,b, Hiromi Tanakad, Hiroyuki Takeyaa, and Yoshihiko Takanoa,b aNational Institute for Materials Science, 1-2-1 Sengen, Tsukuba, Ibaraki 305-0047, Japan bGraduate School of Pure and Applied Sciences, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, cPhysics Department, College of Science,United Arab Emirates University, Al Ain UAE dNational Institute of Technology, Yonago College, 4448 Hikona, Yonago, Tottori 683-8502, Japan Ibaraki 305-8577, Japan Abstract A direct patterning technique of gallium-irradiated superconducting silicon has been established by focused gallium-ion beam without any mask-based lithography process. The electrical transport measurements for line and square shaped patterns of gallium-irradiated silicon were carried out under self-field and magnetic field up to 7 T. Sharp superconducting transitions were observed in both patterns at temperature of 7 K. The line pattern exhibited a signature of higher onset temperature above 10 K. A critical dose amount to obtain the superconducting gallium-irradiated silicon was investigated by the fabrication of various samples with different doses. This technique can be used as a simple fabrication method for superconducting device. 1 1. Introduction It has been recently discovered that heavily hole-doped group-IV semiconductors like diamond [1-3], silicon [4], and germanium [5] exhibit superconductivity. According to a McMillan relation, superconducting transition temperature Tc is proportional to the Debye temperature θ [6]. The group-IV semiconductors, especially, diamond and silicon are candidates for high-Tc superconductors because they show remarkable high Debye temperature [7]. Various methods to induce the high hole-carrier concentration in these materials have been studied. For example, high-pressure synthesis [8], chemical vapor deposition [9], electric field effects [10-12], and so on [13]. Ion implantation is a strong tool to induce hole carrier in these materials [14-16]. Recently, an appearance of superconductivity with Tc at around 7 K was reported in gallium-doped silicon via the ion implantation process [17-20]. Although the discovery with relatively high Tc above liquid helium temperature in silicon attracts wide attention, it is necessary to use a resist-based conventional lithographic process to obtain a desired pattern of a superconducting circuit. Such a patterning is crucial for the application of superconducting devices, for example, the superconducting quantum interference device (SQUID) magnetometer. The focused ion beam (FIB) is one of the most popular nanofabrication techniques for semiconducting devices [21], ultra-thin films [22], and superconducting Josephson devises [23,24]. FIB nanofabrication provides direct etching and deposition for desired shapes without resist-based process. In general, the desired region of sample can be milled by scanning the ion-beam over it in the FIB fabrication. The etching rate is determined by the dose amount of gallium ions which are tuned by the condenser lens, aperture size and dose time. If the dose amount of FIB beam achieves a certain criterion for the emergence of superconductivity in silicon, it will be possible to obtain desired patterns of superconducting silicon directly without resist-based lithographic process. In this study, we investigated the irradiation effect of accelerated gallium-ions on silicon substrate by FIB. The surface states of irradiated region were analyzed by a core-level X-ray photoelectron spectroscopy (XPS). Two kinds of line and square shaped gallium-irradiated silicon with different dose amount were evaluated using an electrical resistance measurement. 2. Experimental procedures The gallium irradiations for silicon substrate were carried out using a SMI9800SE FIB machine (Hitachi High-Technologies), equipped with Ga+ beam. The acceleration voltage of the ion beam, FIB current, chamber pressure were 30 keV, 3.6 μA with an aperture size of 2 mm, and 3×10-5 Pa, respectively. The dose amount of gallium ions, which is determined from the formula IT/S where I is FIB current, T is dose time, and S is dose area, was mainly adjusted by the dose time. The surface states of gallium-irradiated region on silicon was analyzed by the core-level XPS (AXIS-ULTRA DLD, Shimadzu/Kratos) with AlKα X-ray radiation (hν = 1486.6 eV), operating under a pressure of the order of 10-9 Torr. The background signals were subtracted by using an active Shirley method on COMPRO software [25]. The photoelectron peaks were analyzed by the pseudo-Voigt functions peak fitting. The gallium-irradiated region on silicon was evaluated from the temperature dependence of resistance via a standard four probe method using physical property measurement system (PPMS, 2 Quantum Design) with 7 T superconducting magnet. The electrodes were made by a silver-paste painting and gold wires on the irradiated region. The temperature dependence of upper critical field (Hc2) of the silicon was determined from onset Tc value. 3. Results and discussion Figure 1 (a) shows a temperature dependence of resistance in the gallium-irradiated silicon of line pattern (1×1000×2 μm) with dose amount of 5.7×1018 C/cm2. The resistance sharply dropped to zero of onset) of 7 K and zero resistance temperature Tc zero with superconducting onset temperature (Tc 6 K in agreement with the previous report of the gallium-doped silicon induced by the ion-implantation and a rapid thermal treatment [18]. For the practical applications, the anisotropy of superconductivity was evaluated through a measurement for an angle dependence of the resistance under magnetic fields at 5 K as shown in Fig. 1 (b). The superconductivity was sensitively suppressed around 0 degree, namely under magnetic field perpendicular to the substrate. In contrast, the superconductivity was robust against the magnetic field parallel to the substrate. The temperature dependence of resistance was investigated under various magnetic fields (c) parallel and (d) perpendicular to the substrate. The insets show temperature dependence of Hc2. The critical fields follow a typical parabolic behavior, which is consistent with the previous report for the gallium-doped silicon [18]. Here, a maximum critical fields at zero temperature Hc2//(0) and Hc2⊥(0) corresponding to Hc2(0) under a magnetic field which is parallel to the substrate and perpendicular to the substrate, respectively. The Hc2//(0) of 14.8 T and Hc2⊥(0) of 10.7 T were estimated by the parabolic fit. The anisotropic parameter γ = Hc2//(0) / Hc2⊥(0) was 1.4. The coherence length at zero temperature ξ//(0) and ξ ⊥ (0) were estimated as 4.7 nm and 5.5 nm, respectively, from the Ginzburg -- Landau (GL) formula Hc2(0) = Φ0/2πξ(0)2, where the Φ0 is the flux quantum. To confirm a flexibility for fabrication of superconducting silicon, we also prepared a square pattern (200×200×0.3 μm) with dose amount of 8.5×1017 C/cm2 and evaluated in Fig. 1 (e-h) using same method with that in the line pattern. The square pattern exhibited almost same Tc with that of onset is maybe caused by lower dose amount. On the other hand, more line pattern. Slightly small Tc emphasized antistrophic properties were observed as shown in Fig. 1 (f). The Hc2//(0) of 18.8 T and Hc2⊥(0) of 2.4 T were estimated by the parabolic fit. The anisotropic parameter γ = Hc2//(0) / Hc2⊥(0) was 7.8, which is worthily higher than that of the line pattern, maybe reflecting its thin-film properties [18]. The coherence length at zero temperature ξ//(0) and ξ⊥(0) were estimated as 4.2 nm and 11.7 nm, respectively. There are two possibilities for the origin of the superconductivity that is the aforementioned gallium-doped silicon and elemental β-gallium [26-28]. Here, the previously reported the critical field and the coherence length in the gallium-doped silicon and β-gallium are compared to those parallel to the substrate in gallium-irradiated silicon. According to the literatures [18, 28], the critical field and the coherence length were 9.4 T and 6 nm in the gallium-doped silicon, 57 mT and 76 nm in the β-gallium, respectively. Since the superconducting parameters from our product is similar to those from former, we conclude that the observed superconductivity in this study is originated from the gallium-doped silicon. 3 Figure 1. Superconducting properties in the gallium irradiated silicon of (a-d) line pattern (1×1000×2 μm) with dose amount of 5.7×1018 C/cm2 and (e-h) square pattern (200×200×0.3 μm) with dose amount of 8.5×1017 C/cm2. (a,e) Temperature dependence of resistance from 300 K to 2 K in square pattern. (b,f) Angle dependence of resistance at 5 K under various magnetic fields, (c,g) Enlargement around superconducting transition under various magnetic fields parallel to the line (or plane) and (d,h) perpendicular to the line (or plane). Here, we note a specific temperature dependence of resistance under magnetic field in the gallium-irradiated silicon similar to a high-quality boron-doped superconducting diamond [29]. Figure 2 shows an enlargement of the temperature dependence of resistance in the gallium irradiated silicon of line pattern under magnetic fields. The separation between the resistances above 10 K and estimated the onset of transition to be at a value of ~12 K. The onset is gradually shifted to lower temperatures with increasing magnetic field. To clear the onset of transition, differential curves of resistance for temperature dR/dT under 0 T and 7 T were shown in the inset of Fig. 3. We can see a clear separation of differential curves around 12 K under 0 T and 7 T. The signature of higher Tc maybe attributed only to better crystallinity or a combination of better crystallinity and partially larger carrier density [29]. We can expect a bulk superconductivity above 10 K if the dose condition of gallium ion beam is optimized. 4 Figure 2. Enlargement of temperature dependence of resistance in the gallium irradiated silicon of line pattern under magnetic fields. The inset is differential curve of resistance for temperature around Tc onset. To confirm a substrate dependence of gallium-irradiation effect, we fabricated line-shaped pattern on various substrates with same dose amount of 5.7×1018 C/cm2 and measured their resistance-temperature (R-T) properties as shown in Fig. 3. The pattern dimensions were 1000 μm in length, 1 μm in width, 2 μm in depth, and the irradiation time is 1 hour. The gallium-irradiated silicon exhibited a semiconducting-like behavior and a sudden drop of resistance corresponding to superconductivity. To clarify the origin of superconductivity, we investigated the gallium-irradiation effects for various substrates. Although the gallium irradiated diamond substrate showed lower resistance than that of general undoped diamond, the drop of resistance was not observed in the R-T measurement. The gallium-irradiation effects for the conductive boron-doped diamond substrate and ITO (indium tin oxide) glass substrate were also investigated to exclude a charge-up effect during the gallium radiation as seen in high resistance substrates such an undoped diamond. As a result of the R-T measurements, both conductive substrates showed no superconductivity. The irradiated region on the insulating substrate of SiO2 glass exhibited quite high resistance above 40 MΩ. In conclusion, the gallium-irradiated silicon is the only substrate that showed superconductivity, indicating that the origin of the superconductivity could be considered as a gallium-doped silicon as reported in the literature [17-20]. If superconducting gallium is deposited on the surface of the substrates, all substrates should show the superconductivity. 5 Figure 3. Optical microscope images and resistance-temperature (R-T) properties of various gallium-irradiated substrates of silicon, undoped diamond, boron-doped diamond, ITO (indium tin oxide) glass, and SiO2 glass. The gallium-irradiated silicon of square pattern was used for chemical state analysis using XPS. Figure 4 (a) shows the depth profile of core-level Ga 2p XPS spectra in the gallium-irradiated region of silicon substrate. The etching treatment was performed by an Ar gas cluster ion beam (GCIB) with 10 keV beam energy. The mean size of one cluster was approximately 1000 atoms, the scanning area of the GCIB was about 2 mm2, and the beam current was about 5 nA. The GCIB mills the sample surface very slowly without a change of the intrinsic chemical state during the irradiation [30]. According to the surface spectrum, we can see the two individual peaks around 1119.1 eV and 1116.6 eV, corresponding to the pure gallium peak [31] and the most stable oxide (Ga2O3) peak [32], respectively. The oxide peak was gradually decreased by the GCIB etching and completely disappeared at the etching depth of 4.8 nm. The literature regarding to the gallium implantation for 6 n-type silicon substrate with 30 nm thick SiO2 reported that the surface spectrum showed no signal of Ga 2p [18]. When the 14 nm depth was milled, the Ga2O3 peak appeared, and pure gallium peaks was observed from 18 nm depth, according to the previous report of depth profile [18]. Because the acceleration voltage for the gallium implantation is quite lower in our FIB process than 80 keV of the previous study [18], it could be considered that the irradiated gallium stayed at shallow region around surface with high concentration. Figure 4 (b) shows the core-level Si 2p XPS spectra. The upper spectrum was acquired from the gallium-irradiated region and lower one was from the other region on the same silicon substrate. The Si 2p photoemission is split into two peaks, one at high energy side attributed to Si 2p1/2 and the lower one attributed to Si 2p3/2. The shape of Si 2p peak exhibits a broadening feature after the gallium irradiation from a comparison between two spectra. The full width at half maximum (FWHM) of 0.57 eV in the as-prepared silicon was change to 0.68 eV in the gallium-irradiated region. The peak broadening is known as a signal from an amorphization which generated by the ion scattering around the surface [33]. These peak changes indicate that there are various states of Si-Si bonding distance and bonding angle around the surface, and maybe it affects the superconducting properties. Figure 4. (a) Depth profile of core-level Ga 2p XPS spectra in the gallium-irradiated region of silicon substrate. (b) Core-level Si 2p XPS spectra in the gallium-irradiated region and as-prepared silicon substrate. The pattern is square (200×200×0.3 μm) with dose amount of 8.5×1017 C/cm2. It is important to determine the critical dose amount for the superconductivity in the gallium-irradiated silicon for the practical application. The gallium irradiations with same dimension (200×200 μm square) by dose amounts of 37×1015 C/cm2, 227×1015 C/cm2, and 850×1015 C/cm2, were performed to examine the transport properties. Figure 5 shows the dose amount dependence of the resistance at 300 K in the square pattern of gallium-irradiated silicon. The lowest dose sample showed quite high resistance of 105 Ω order. The resistance dramatically decreased less than 103 Ω 7 order with increase of the dose amount. The decrease of resistance tended to saturate below 103 Ω. It means the gallium implantation rate is first proportional to dose amount. When the dose amount achieves a certain criterion, the implantation rate and etching rate become comparable, and then the implantation saturates. The gallium temperature dependence of each gallium-irradiated samples. The resistance in the lowest dose sample (37×1015 C/cm2) drastically inset shows the increased as a function of temperature, and it showed no sign of superconductivity at least 2 K. On the other hand, although the middle dose sample (227×1015 C/cm2) indicated clear superconducting onset ~5 K, the zero resistance was not observed. The highest dose sample (850×1015 transition with Tc C/cm2) which is same as the fig.2 (e) showed clear zero-resistance. These results indicated that the critical dose amount to obtain the zero-resistance is between 227×1015 C/cm2 and 885×1015 C/cm2. It is expected that the device fabrication, such a SQUID magnetometer, by using this mask-less patterning technique of superconducting circuit will be highly anticipated. Figure 5. Dose amount dependence of the resistance at 300 K in the square pattern of gallium-irradiated silicon. The inset is the temperature dependence of each gallium-irradiated samples of the square pattern (200×200×0.3 μm) with different dose amount. 4. Conclusion In this study, a nanofabrication technique for mask-less patterning of superconducting region by gallium-irradiated silicon on the substrate using FIB was introduced. In various substrates of silicon, diamond, boron-doped diamond, ITO glass, and SiO2 glass, only silicon substrate showed of 7 K after gallium irradiation. The line and square shapes of superconductivity with onset Tc 8 gallium irradiated silicon were fabricated to confirm the versatility of patterning. Although both patterns exhibited superconductivity, very large antistrophic behavior against applied magnetic field onset above 10 was observed in square pattern. In the line pattern, we observed a signature of higher Tc K maybe due to an inhomogeneity of dose amount. The depth profile of XPS spectrum revealed that the irradiated gallium was shallowly distributed in the silicon surface. The surface of irradiated silicon changed to the amorphous like state from the peak broadening of Si 2p XPS spectra. The critical dose amount for superconductivity is between 227×1015 C/cm2 and 885×1015 C/cm2. This direct patterning technique of superconducting circuit on silicon substrate without any masks significantly contributes to the application of superconducting devices, such as SQUID magnetometer. Acknowledgment The authors thank Prof. Dr. T. Yamaguchi and Mr. Sasama (NIMS) for the early experiments, and also thank Mr. J. Aoto (NIT, Yonago College) for the supports regarding to the sample preparations and measurements. This work was partly supported by JST CREST Grant No. JPMJCR16Q6, JST-Mirai Program Grant Number JPMJMI17A2, JSPS KAKENHI Grant Number JP17J05926 and 19H02177. References [1] E. Ekimov, V. Sidorov, E. Bauer, N. Mel'Nik, N. Curro, J. Thompson, and S. Stishov, Nature 428, 542 (2004). [2] T. Yokoya, T. Nakamura, T. Matsushita, T. Muro, Y. Takano, M. Nagao, T. Takenouchi, H. Kawarada, and T. Oguchi, Nature 438, 647 (2005). [3] Y. Takano, M. Nagao, I. Sakaguchi, M. Tachiki, T. Hatano, K. Kobayashi, H. Umezawa, and H. Kawarada, Appl. Phys. Lett. 85, 2851-2853 (2004). [4] E. Bustarret, C. Marcenat, P. Achatz, J. Kacmarcik, F. Levy, A. Huxley, L. Ortega, E. Bourgeois, X. Blase, D. Debarre, and J. Boulmer, Nature 444, 465 (2006). [5] T. Herrmannsdörfer, V. Heera, O. Ignatchik, M. Uhlarz, A. Mücklich, M. Posselt, H. Reuther, B. Schmidt, K.-H. Heinig, W. Skorupa, M. Voelskow, C. Wündisch, R. Skrotzki, M. Helm, and J. Wosnitza, Phys. Rev. Lett. 102, 217003 (2009). [6] W. L. Mcmillan, Phys. Rev. 167, 331 (1968). [7] D. F. Aminev, A. Yu. Klokov, T. I. Galkina, A. I. Sharkov, and V. G. Ral'chenko, Bull. Lebedev. Phys. Inst. 37, 152-156 (2010). [8] N. Dubrovinskaia, R. Wirth, J. Wosnitza, T. Papageorgiou, H. F. Braun, N. Miyajima, and L. Dubrovinsky, Proc. Nat. Acad. Sci. USA 105, 11619 (2008). [9] T. Kageura, M. Hideko, I. Tsuyuzaki, S. Amano, A. Morishita, T. Yamaguchi, Y. Takano, and H. Kawarada, Diamond and Relat. Mater. 90, 181-187 (2018). [10] Y. Takahide, H. Okazaki, K. Deguchi, S. Uji, H. Takeya, Y. Takano, H. Tsuboi, and H. Kawarada, Phys. Rev. B 89, 235304 (2014). [11] Y. Sasama, T. Yamaguchi, M. Tanaka, H. Takeya, and Y. Takano, J. Phys. Soc. Jpn. 86, 114703 (2017). 9 [12] Y. Sasama, T. Yamaguchi, M. Tanaka, H. Takeya, and Y. Takano, J. Phys. Soc. Jpn. 86, 014703 (2017). [13] D. Cammilleri, F. Fossard, D. D´ebarre, C. Tran Manh, C. Dubois, E. Bustarret, C. Marcenat, P. Achatz, D. Bouchier, and J. Boulmer, Thin Solid Films 517, 75 (2008). [14] T. Herrmannsdorfer, R. Skrotzki,V. Heera, O. Ignatchik, M. Uhlarz, A. Mucklich, M. Posselt, B. Schmidt, K-H Heinig, W. Skorupa, M. Voelskow, C.Wundisch, M. Helm, and J.Wosnitza, Supercond. Sci. Technol. 23, 034007 (2010). [15] S. Mirabella, D. De Salvador, E. Napolitani, E. Bruno, and F. Priolo, J. Appl. Phys. 113, 031101 (2013). [16] B. Ittermann, K. Bharuth-Ram, H. Metzner, M. Füllgrabe, M. Heemeier, F. Kroll, F. Mai, K. Marbach, P. Meier, D. Peters, H. Thiess, H. Ackermann, H.-J. Stöckmann, and J. P. F. Sellschop, Appl. Phys. Lett. 71, 3658 (1997). [17] R. Skrotzki, J. Fiedler, T. Herrmannsdorfer, V. Heera, M. Voelskow, A. Mucklich, B. Schmidt, W. Skorupa, G. Gobsch, M. Helm, and J. Wosnitza, Appl. Phys. Lett. 97, 192505 (2010). [18] J. Fiedler, V. Heera, R. Skrotzki, T. Herrmannsd€orfer, M. Voelskow, A. M€ucklich, S. Oswald, B. Schmidt, W. Skorupa, G. Gobsch, J. Wosnitza, and M. Helm, Phys. Rev. B 83, 214504 (2011). [19] V. Heera, J. Fiedler, M. Voelskow, A. Mücklich, R. Skrotzki, T. Herrmannsdörfer, and W. Skorupa, Appl. Phys. Lett. 100, 262602 (2012). [20] V. Heera, J. Fiedler, R. H€ubner, B. Schmidt, M. Voelskow, W. Skorupa, R. Skrotzki, T. Herrmannsd€orfer, J. Wosnitza, and M. Helm, New J. Phys. 15, 083022 (2013). [21] J. F. Einsle, J. Bouillard, W. Dickson, and A. V. Zayats, Nano. Res. Lett. 6, 572 (2011). [22] M. Tanaka, H. Takeya, and Y. Takano, Appl. Phys. Express 10, 023101 (2017). [23] S. Kim, Y. I. Latyshev, T. Yamashita, and S. Kishida, IEEE Trans. Appl. Supercon. 11, 948-951 (2001). [24] E. S. Sadki, S. Ooi, and K. Hirata, Appl. Phys. Lett. 85, 6206-6208 (2004). [25] R. Matsumoto, Y. Nishizawa, N. Kataoka, H. Tanaka, H. Yoshikawa, S. Tanuma, and K. Yoshihara, J. Electron Spectrosc. Relat. Phenom. 207, 55 (2016). [26] H. Parr and J. Feder, Phys. Rev. B 7, 166 (1973). [27] H. Parr, Phys. Rev. B 10, 4572 (1974). [28] D. Campanini, Z. Diao, and A. Rydh, Phys. Rev. B 97, 184517 (2018). [29] H. Okazaki, T. Wakita, T. Muro, T. Nakamura, Y. Muraoka, T. Yokoya, S. Kurihara, H. Kawarada, T. Oguchi, and Y. Takano, Appl. Phys. Lett. 106, 052601 (2015). [30] T. Miyayama, N. Sanada, M. Suzuki, J. S. Hammond, S.-Q. D. Si, and A. Takahara, J. Vac. Sci. Technol. A 28, L1 (2010). [31] K. Yokota, S. Tamura, S. Ishihara, and I. Kimura, Jpn. J. Appl. Phys. 24, 62-67 (1985). [32] C. V. Ramana, E. J. Rubio, C. D. Barraza, A. M. Gallardo, S. McPeak, S. Kotru, and J. T. Grant, J. Appl, Phys. 115, 043508 (2014). [33] Z. H. Lu, D. F. Mitchell, and M. J. Graham, Appl. Phys. Lett. 65, 552 (1994). 10
1810.05265	1	1810	2018-10-11T21:42:20	Anti-Stokes excitation of solid-state quantum emitters for nanoscale thermometry	[ "physics.app-ph", "physics.optics", "quant-ph" ]	Color centers in solids are the fundamental constituents of a plethora of applications such as lasers, light emitting diodes and sensors, as well as the foundation of advanced quantum information and communication technologies. Their photoluminescence properties are usually studied under Stokes excitation, in which the emitted photons are at a lower energy than the excitation ones. In this work, we explore the opposite Anti-Stokes process, where excitation is performed with lower energy photons. We report that the process is sufficiently efficient to excite even a single quantum system, namely the germanium-vacancy center in diamond. Consequently, we leverage the temperature-dependent, phonon-assisted mechanism to realize an all-optical nanoscale thermometry scheme that outperforms any homologous optical method employed to date. Our results frame a promising approach for exploring fundamental light-matter interactions in isolated quantum systems, and harness it towards the realization of practical nanoscale thermometry and sensing.	physics.app-ph	physics	Anti-Stokes excitation of solid-state quantum emitters for nanoscale thermometry Toan Trong Tran,1,* Blake Regan,1 Evgeny A. Ekimov,2 Zhao Mu,3 Zhou Yu,3 Weibo Gao,3 Prineha Narang,4 Alexander S. Solntsev,1 Milos Toth,1 Igor Aharonovich1 and Carlo Bradac1,* 1School of Mathematical and Physical Sciences, University of Technology Sydney, Ultimo, NSW, 2007, Australia 2Physics, RAS Kaluzhskoe Road 14, Troitsk, 142190, Russia 3Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University, 637371, Singapore 4John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA Corresponding authors: [email protected]; [email protected] Abstract Color centers in solids are the fundamental constituents of a plethora of applications such as lasers, light emitting diodes and sensors, as well as the foundation of advanced quantum information and communication technologies. Their photoluminescence properties are usually studied under Stokes excitation, in which the emitted photons are at a lower energy than the excitation ones. In this work, we explore the opposite Anti-Stokes process, where excitation is performed with lower energy photons. We report that the process is sufficiently efficient to excite even a single quantum system -- namely the germanium-vacancy center in diamond. Consequently, we leverage the temperature- dependent, phonon-assisted mechanism to realize an all-optical nanoscale thermometry scheme that outperforms any homologous optical method employed to date. Our results frame a promising approach for exploring fundamental light-matter interactions in isolated quantum systems, and harness it towards the realization of practical nanoscale thermometry and sensing. Stokes and Anti-Stokes emission are fundamental phenomena widely used to study the physico- chemical and optical properties of materials. Stokes (Anti-Stokes) photoluminescence occurs when the energy of the emitted photons is lower (higher) than that of the absorbed ones (1). In the Anti- Stokes case, the extra energy that causes upconversion of the photons can be acquired through a variety of mechanisms, ranging from multi-photon absorption (2) to Auger recombination (3) and phonon absorption (4). The latter, relevant to this work, is illustrated in Figure 1A, B. A photon with energy ℎ𝜈𝑒𝑥𝑐 at the long-wavelength tail of the absorption spectrum excites an electron from a thermally-populated first vibronic state (𝑛0 = 1) of the electronic ground state 𝐸0, to the bottom manifold (𝑛1 = 0) of an excited electronic state 𝐸1 [red arrow]. The system then returns to the ground state via spontaneous emission of an upconverted photon with a mean energy ℎ𝜈𝑠𝑒 > ℎ𝜈𝑒𝑥𝑐 [yellow arrow]. This phonon-assisted Anti-Stokes excitation process scales exponentially with temperature and is the bedrock of a variety of fundamental studies (e.g. cavity quantum electrodynamics (5)), as well as practical applications such as optical cryocooling (6, 7), bioimaging (8) and Raman spectroscopy (9). However, Anti-Stokes photoluminescence (PL) is inherently inefficient, and all work done to date on solid-state defects has been focused on ensembles (10-12) rather than individual point defects. Here, we demonstrate that Anti-Stokes PL can be used to study isolated quantum systems -- specifically atom-like color centers in diamond, over a large range of temperatures. We explore the mechanism for some of the most studied diamond defects, the nitrogen-vacancy (NV) (13) the silicon- vacancy (SiV) (14, 15) and the germanium-vacancy (GeV) (16) center. We show that Anti-Stokes excitation of selected diamond color center is an efficient process, detectable by standard photoluminescence spectroscopy and leverage this finding to demonstrate upconversion PL from a single, isolated GeV defect. We show that the Anti-Stokes excitation process is thermally-activated and proceeds through a phonon-photon absorption pathway rather than through multi-photon absorption. We exploit the high Anti-Stokes excitation efficiency to introduce an innovative approach for all-optical nanoscale thermometry based on the temperature-dependence of the Anti-Stokes to Stokes PL intensity ratio. Our technique outperforms all other previously reported all-optical nanothermometry methods. To frame the scope of the Anti-Stokes process for quantum emitters and its capacity for developing nanoscale sensing applications, we characterized diamond samples (cf. Methods) containing germanium-vacancy (GeV), silicon-vacancy (SiV) and nitrogen-vacancy (NV) centers. A schematic illustration of a diamond defect in the split-vacancy configuration (i.e. GeV, SiV) is shown in Figure 1B. For each of the diamond defects, we selected a specific pair of excitation lasers (cf. Methods and SI, Fig. SI1) with energies above (Stokes) and below (Anti-Stokes) the zero-phonon line (ZPL) of each emitter. Figures 1C -- E show room temperature Stokes [blue] and Anti-Stokes [ocher] PL spectra for ensembles of GeV, SiV and NV centers, respectively. Note that the sharp edges of the emission peaks are due to band-pass filters used to suppress the excitation lasers. The insets show the complete Stokes PL spectra of each color center. At room temperature, all color centers show Anti-Stokes PL. To confirm that the upconversion was not caused by multi-photon absorption, we measured photoluminescence intensity vs excitation power and concluded that the scaling does obey one-photon rather than two-photon absorption dynamics (cf. Supplementary Information, Fig. SI2). Next, we established a direct, quantitative comparison amongst the Anti-Stokes to Stokes PL ratios of the studied centers. Normalizing the Anti-Stokes intensity makes the comparison independent of the density of defects amongst the different samples. The comparison does, nonetheless, issue some caveats. The first regards the selection of the Stokes and Anti-Stokes laser excitation wavelengths. Our hypothesis is that the Anti-Stokes excitation process involves vibronic states of the defects which are populated via the absorption of phonons by ground-state electrons. It therefore follows that the process is proportional to the phonon density of states, making Anti-Stokes absorption ideally the most efficient for excitation wavelengths matching the density maximum -- and desirably not too narrow, spectrally. Simultaneously, for the comparison to be meaningful the difference between Anti- Stokes and Stokes excitation energies should be similar for the different color centers. Further, for practical sensing realizations, one must ultimately consider the number of color centers per unit volume of diamond realistically achievable for each type of defect -- as this affects the signal-to-noise ratio and thus the resolution of the sensor. Bearing these caveats in mind, we find that SiV and GeV centers outperform NV centers under our experimental conditions -- their Anti-Stokes emission efficiency is higher as is their attainable density of defects per nanodiamond (17, 18). The measured Anti-Stokes to Stokes PL intensity ratios are similar for GeV and SiV centers, and approximately three orders of magnitude higher than that for NV centers: 𝐼𝐴𝑆/𝐼𝑆𝐺𝑒𝑉 = (8.4±3.3)×10-2, 𝐼𝐴𝑆/𝐼𝑆𝑆𝑖𝑉 = (13.2±1.1)×10-2 and 𝐼𝐴𝑆/𝐼𝑆𝑁𝑉 = (11.9±2.8)×10- 5. The lower efficiency of the Anti-Stokes process for the NV center is somewhat counterintuitive. The NV center displays a large phononic sideband, which trivially suggests more efficient coupling to the lattice and a more efficient Anti-Stokes excitation process compared to that of the spectrally- narrower SiV and GeV centers. The much lower value of the ratio IAS/IS for the NV in our experiment is mainly due to the Anti-Stokes excitation laser being quite far below the NV ZPL energy (224.40 meV), at the long-wavelength tail of the phonon side band. Additional contributing factors to the low PL Anti-Stokes emission are the NV photo-ionization process (19) and the recently-proposed NV -- N tunneling mechanism in nitrogen-rich diamond samples (20) -- hinted by the difference between the Stokes and Anti-Stokes PL spectra seen in Figure 1E. The SiV and GeV centers have similar Anti-Stokes emission efficiencies (normalized to their respective Stokes ones), making them both good candidates for Anti-Stokes quantum measurements and potential nanothermometry applications (cf. Supplementary Information, Fig. SI3). However, owing to the fact that its excited state decay is highly nonradiative (21), the SiV center possesses a lower luminescence quantum efficiency than the GeV (16). A high quantum efficiency is desirable for it maximizes the PL signal-to-noise ratio, which ultimately determines the temperature and spatial resolution in nanothermometry. We, therefore, selected the GeV center as our primary candidate for the remainder of this work. We start by demonstrating that Anti-Stokes PL measurements are feasible down to a single quantum emitter (i.e. a single atom-like defect). Figure 2 shows the systematic analysis for the GeV center. Figure 2A is a 25×25 μm2 confocal PL scan of a single crystal diamond where the bright spots are areas that have been implanted to induce the inclusion of GeV centers (cf. Methods). In the surveyed confocal scan, we isolated single GeV centers -- like the one highlighted by a dashed red circle in Figure 2A. Figure 2B shows the PL measurement of the defect. The ZPL is clearly visible at 602 nm. The quantum nature of the emitter is shown by the second-order autocorrelation function 𝑔(2)(𝜏) which has a zero-delay-time value 𝑔(2)(𝜏 = 0) < 0.5 (not background corrected) -- considered indicative of a single-photon emitter (Fig. 2C). Only the ZPL signal (shaded in blue in Fig. 2B) was used for the antibunching measurement. Next, we carried out Anti-Stokes excitation of the identified GeV center. Remarkably, the process is efficient enough that Anti-Stokes emission from a single GeV defect can be detected in a standard PL measurement. Figure 2D shows the Anti-Stokes signal from the single GeV center from Figure 2A under laser excitation at a wavelength of 637 nm, 38 mW of power and a total acquisition time of 12 minutes. This result is notable on its own: it demonstrates, for the first time, Anti-Stokes PL from a single solid-state defect. The high efficiency of the GeV Anti-Stokes PL process makes it a compelling candidate for all- optical nanothermometry (22, 23). To quantify the sensitivity, resolution and range of a potential nanothermometer, we characterized the Stokes and Anti-Stokes PL signals from a nanodiamond (~400 nm) hosting an ensemble of GeV centers (cf. Supplementary Information, Fig. SI4), as a function of temperature. The nanodiamond GeV ensemble had a room-temperature PL intensity of ~106 counts/s, measured under 532-nm (Stokes) laser excitation at 500 µW, after a 595 -- 615-nm bandpass filter. Figure 3A shows the results for the Anti-Stokes excitation analysis (also cf. Supplementary Information, Fig. SI5, SI6). The intensity of the Anti-Stokes emission exhibits Arrhenius-type exponential scaling with temperature. The data fits very well the equation 𝐴𝑒−(𝐸𝑎/𝑘𝐵𝑇), with 𝑘𝐵 being the Boltzmann constant and 𝐸𝑎 the value for the activation energy fixed at 102.96 meV -- which is the difference in energy between the Anti-Stokes excitation laser and the GeV ZPL. The Arrhenius-type dependence shows that the Anti-Stokes excitation process is thermally activated, supporting our hypothesis that the Anti-Stokes excitation of diamond color centers involves the absorption of phonons from the lattice. Notably, the existence of an exponential dependence between Anti-Stokes PL intensity and temperature makes the mechanism ideal for high-sensitivity nanothermometry. For the purpose of realizing a practical sensor, we use the ratio between Anti-Stokes and Stokes PL as the experimental the discussion, we benchmark observable. The normalization makes the sensor independent of experimental specificities (e.g. loss of detected photons due to absorption or scattering in certain environments like living cells, or samples that change phase during a heating/cooling measurement). Figure 3B displays the Anti-Stokes to Stokes photoluminescence intensity ratio as a function of temperature, measured over the range 110 -- 330 K. Over this range, the 𝐼𝐴𝑆/𝐼𝑆 ratio fits well the exponential function 𝑎 + 𝑏𝑒−[𝑐/(𝑇−𝑇0)]. The strong dependence on temperature is highly advantageous, as it translates to extremely high sensitivity -- based on the standard definition of sensitivity, as an absolute quantity which describes the smallest amount of detectable change in a measurement. In fact, the thermometer sensitivity matches (or far exceed) that of any other all-optical method (Figure 3C), including techniques based on Raman spectroscopy which boast high sensitivity over a broad temperature range (24), but are not suitable for nanoscale thermometry because they suffer from limited spatial resolution. In terms of temperature resolution, the performance of the nanothermometer we investigated is comparable with the current best all-optical-based methods (25-28) with a noise-floor temperature resolution of 455 mK‧Hz-½, at room temperature. Note that due to the exponential dependence of the 𝐼𝐴𝑆/𝐼𝑆 ratio with temperature the resolution worsens at low temperatures yet improves rapidly at high temperatures. Specifically, in the range 110 -- 330 K, the measured temperature resolution is 2.494 -- 0.420 K‧Hz-½. Unlike sensitivity, the resolution is a relative quantity and can be improved, for instance by selecting nanodiamond hosts with a higher density of color centers or by reducing the measurement bandwidth, i.e. increasing the integration time for the PL signal. To complete the characteristics and performance of our nanothermometer against those of the current field's bests. The first factor is utility. Our approach is an all-optical, microwave-free nanothermonetry technique based on diamond color centers. Nanothermometers of this type (25-28) are broadly appealing because of their high spatial resolution, low noise floor (i.e. high temperature resolution), wide temperature range and broad applicability. The second metric is sensitivity, where all-optical nanothermometers often do not perform as well, for many rely on measuring the temperature-dependence on observables such as ZPL frequency (25, 27, 28) or amplitude (29) which vary weakly compared to the, demonstrated herein, Anti-Stokes to Stokes emission intensity ratio. We also note that techniques based on measuring PL intensity amplitude (rather than ratio), such as that of the NV center ZPL (29), have limited applicability because they suffer from a range of artifacts such as changes in photon scattering and absorption caused by changes in temperature of the measured sample. Our approach is not compromised by any of these shortcomings. The Anti-Stokes to Stokes PL ratio in diamond color centers reaches temperature sensitivities that match those of Raman-based sensors, while retaining the methods based on photoluminescent nanodiamonds, including a ~few-nm spatial resolution, as it works on single color centers that are stable in sub-10 nm nanodiamonds (17, 18). Note also that the exponential scaling with temperature of the ratio 𝐼𝐴𝑆/𝐼𝑆 makes the resolution of our method increase rapidly at high temperature. This makes it desirable, for instance, for temperature sensing in high-power electronics (30) -- in virtue as well of diamond color centers being able to withstand extremely high temperatures (>1000 K). Figure 3C visually captures the superior performance of our approach against other nanothermometry schemes. The graph shows an absolute comparison by plotting the relative sensitivity of each technique as a function of temperature. We define the relative sensitivity as (𝜕𝑂/𝜕𝑇)/𝑂 where 𝑂 is the measured observable (e.g. ZPL frequency, ZPL amplitude, etc.) The graph shows the relative sensitivity based on: i) our Anti-Stokes to Stokes PL intensity ratio, ii) the frequency shift of the GeV ZPL in our Stokes PL spectra, equivalent to iii) the same measurement reported in the literature (27), iv) the ZPL wavelength shift of the SnV (28) and v) SiV (25) diamond color centers and vi) the intensity change of the NV ZPL in diamond (29). The sensitivity of our technique is superior to that of the desirable utility features of any of these competitive methods; it matches (or slightly outperforms) the relative sensitivity benchmark of vii) the Anti-Stokes to Stokes emission intensity ratio of a sensor based on Raman spectroscopy (24). For reference, there is an entire family of nanothermometers (31-33) based on the temperature-varying properties of quantum dots (QDs) -- Figure 3C shows an example based on viii) spectral shift (31) -- yet these are often limited to a narrow temperature range (32, 33). The nanothermometry landscape also includes upconversion nanoparticles (UCNPs) (34-37). In some cases (34) they can reach sensitivities comparable to that of our approach, but they usually suffer from limited range of operative temperatures and/or by low quantum efficiency (i.e. low resolution). In conclusion, we have demonstrated Anti-Stokes photoluminescence from a single atom-like defect in diamond, and leveraged the process to demonstrate a new variant of all-optical nanothermometry with unprecedented performance. Our approach forms a basis for fundamental studies of solid-state quantum systems via Anti-Stokes processes, and for novel non-invasive sensing technologies. Methods Samples. The NV-sample consisted in synthetic type Ib ND powders (MSY ≤0.1 μm; Microdiamant) purified by nitration in concentrated sulphuric and nitric acid (H2SO4-HNO3), rinsed in deionized water, irradiated by a 3-MeV proton beam at a dose of (1 × 106 ions per cm2 and annealed in vacuum at 700 °C for 2 h to induce the formation of NV centers (Academia Sinica, Taipei Taiwan) (38) The measured NDs average size is (150.5 ± 23.3) nm. The SiV-sample consisted in NDs synthesized using a microwave plasma chemical vapor deposition (MPCVD) system from detonation ND seeds (size 4-6 nm). The growth was carried out for 30 minutes in a gas mix of hydrogen:methane 100:1, at 900 W microwave power and 60 Torr pressure. The synthesized NDs had size ∼ 0.3 -- 1 μm. For GeV centers we looked at different samples. The first consisted of GeV centers synthesized using a MPCVD method, whereby the germanium was introduced externally as a solid or vapor source. The sample for the single GeV color centers is a high-purity single crystal diamond from Element Six [N] <1 ppb implanted with germanium ions at 35 keV using a nanoFIB system (ionLINE, RAITH Nanofabrication) and an implantation dose of 100 Ge+ ions per spot. The sample was subsequently annealed at 1000 °C for 30 minutes in high vacuum. The second consisted in diamond nanoparticles hosting GeV color centers synthesized from mixtures of Adamantane, C10H14 (Sigma Aldrich, purity > 99%) with small amount of Tetraphenylgermanium C24H20Ge (Sigma Aldrich, purity > 95.5 %) at 9 GPa and 1500-1700 K, as described elsewhere (39, 40). The concentration of Ge in the growth system was about 0.4% calculated relative to the carbon- germanium mixture, Ge/(Ge+C). The third sample was a diamond membrane embedded with GeV color centers and was prepared as followings. The GeO2 covered membrane was placed in a MPCVD chamber, along with a ~1 × 1 mm2 piece of metallic germanium ~1 cm away. The conditions were: hydrogen/methane ratio of 100:1 at 60 Torr, microwave power of 900W for 10 minutes to fabricate a ~400-nm intrinsic diamond layer that contains GeV color centers. The diamond membranes were then flipped 180° and thinned by an inductive coupled plasma reactive ion etching (ICP-RIE) with argon, oxygen and SF6 etch (2:3:1), at a pressure of 45 mTorr, with a forward power of 500 W and 100 W, for the ICP and RIE respectively. Optical Characterization. The samples were mounted on a three-dimensional piezo-stage (ANPx series, Attocube Inc.) in a lab-built, open-loop cryostat (adapted from a ST500 cryostat; Janis) with flowing liquid nitrogen. The temperature at the sample was controlled via a cryogenic temperature controller (335; Lakeshore). Optical access to the sample is through a thin quartz window; the lasers are focused via a high numerical-aperture objective (NA 0.9, 100×, TU Plan Fluor; Nikon), back- collected, spectrally filtered and sent to either a spectrometer (SR303I, mounted with a Newton DU920P CCD Camera; Andor) or a pair of avalanche photodiodes (SPCM-AQR-14; Perkin Elmer) in a Hanbury-Brown and Twiss interferometer configuration (41). Stokes/Anti-Stokes excitation was carried out with the following lasers. GeV sample. Stokes excitation was carried out with a CW diode-pumped solid-state DPSS laser (SDL-532-200T, DreamLasers) at 532 nm. Anti-Stokes excitation was carried out with a TO-Can laser diode (HL63142DG, Thorlabs) at 637 nm. For the Anti-Stokes excitation on a single GeV defect (Figure 3D), a short-pass dichroic was used to increase the excitation and collection efficiency. SiV sample. Stokes excitation was carried out with a TO-Can laser diode (HL63142DG, Thorlabs) at 637 nm. Anti-Stokes excitation was carried out with a CW Titanium:Sapphire (Ti:Sap) laser (SolsTis, M2 Inc.) at 770 nm. NV sample. Stokes excitation was carried out with a CW diode-pumped solid-state DPSS laser (SDL- 532-200T, DreamLasers) at 532 nm. Anti-Stokes excitation was carried out with a picosecond gain switched laser diode (PiL607X; PILAS) operating in CW at 675 nm and with a CW Titanium:Sapphire (Ti:Sap) laser (SolsTis, M2 Inc.) at 720 nm. Thermometry. We measured the ZPL photoluminescence intensity under Stokes and Anti-Stokes excitation, and determined the relative Anti-Stokes PL efficiency for each defect, which ultimately limits the sensitivity of the nanothermometer. Acknowledgements Financial support from the Australian Research council (via DP180100077, DE180100810 and DE180100070), the Asian Office of Aerospace Research and Development grant FA2386-17-1-4064, the Office of Naval Research Global under grant number N62909-18-1-2025 are gratefully acknowledged. E. A. Ekimov is grateful to RFBR for the financial support under Grant No. 17-52- 50075. The authors thank Marcus W. Doherty, Neil B. Manson and Jeff Reimers for useful discussions. Author contributions C.B., T.T.T., I.A. and M.T. conceived the project. B.R., E.A.E., Z.M., Z.Y., and W.G. fabricated the samples. T.T.T. designed and conducted the measurements with the assistance from C.B. Data analysis was conducted by T.T.T. and C.B. All authors discussed the results and co-wrote the manuscript. Figures Figure 1. Stokes and Anti-Stokes luminescence processes for color centers in diamond. A) Energy diagram of representative electronic and vibrational energy levels for a diamond color center. The arrows show the lower (higher) energy of the Stokes (Anti-Stokes) photons with respect to the ZPL energy. In the Anti-Stokes case, the additional energy is acquired via phonon(s) absorption. B) artistic representation of the Anti-Stokes mechanism for a diamond color center which absorbs a lower- energy photon [wavy line, red] and emits a higher-energy one [wavy line, ocher] upon absorption of a phonon [wavy line, purple]. C -- E) Photoluminescence spectra of the ZPL for nanodiamond GeV (C), SiV (D) and NV (E) centers under Stokes [blue] and Anti-Stokes [ocher] excitation (the full PL spectrum under Stokes excitation is shown in the relative inset). The ZPLs (605 nm for GeV, 739 nm for SiV and 639 nm for NV) are spectrally filtered by means of bandpass filters [semitransparent rectangular boxes]. For each measurement in (C), (D) and (E) the powers of the Stokes and Anti- Stokes excitation lasers are the same: PL intensities are normalized to unity for display purposes: the measured values for Anti-Stokes to Stokes PL intensity ratios for GeV, SiV and NV centers are 𝐼𝐴𝑆/𝐼𝑆𝐺𝑒𝑉 = (8.4±3.3)×10-2, 𝐼𝐴𝑆/𝐼𝑆𝑆𝑖𝑉 = (13.2±1.1)×10-2 and 𝐼𝐴𝑆/𝐼𝑆𝑁𝑉 = (11.9±2.8)×10-5. (The line at ~770 nm in (D) is the Anti-Stokes excitation laser). Figure 2. Characterization of Anti-Stokes emission from GeV color centers. A) Confocal image of a 25×25 μm2 bulk diamond sample showing emission from GeV color centers. Each spot has a different density of GeVs. The spot indicated by the [red circle] is a single-photon GeV center as per analysis in (C). B) Photoluminescence spectrum acquired for the single center identified in (A). C) Second- order autocorrelation function 𝑔(2)(𝜏) showing the sub-Poissonian statistic, at zero-delay time, indicative of a single photon source, 𝑔(2)(0) < 0.5 (the value for 𝑔(2)(0) is not background- corrected). D) Anti-Stokes PL spectrum acquired from the single GeV center in (A). The acquisition was carried out for 12 minutes, with laser excitation at 637 nm and 38 mW of power. A bandpass filter (represented as a semi-transparent box around the ZPL of the spectrum) was used to acquire measurements in (C) and (D). Figure 3. Characterization of the Anti-Stokes GeV-based nanothermometer. A) Temperature dependence of the PL intensity signal upon Anti-Stokes excitation (637-nm wavelength). The PL intensity was measured by monitoring the GeV's ZPL (605 nm) isolated with a bandpass filter. The data fit well the Arrhenius-type equation 𝐴𝑒−(𝐸𝑎/𝑘𝐵𝑇), where the activation energy 𝐸𝑎 = 102.96 meV is fixed to coincide with the difference in energy between the excitation laser and the germanium- vacancy's ZPL. B) Plot of the Anti-Stokes to Stokes PL ratio as a function of temperature. The ratio fits an exponential curve: 𝑎 + 𝑏𝑒−[𝑐/(𝑇−𝑇0)], granting the method an extremely high sensitivity. The error bars of plots in (A) and (B) are represented as vertical, blue bars, and are mostly equivalent to or smaller than the size of the data points. C) Relative sensitivity plotted vs temperature for several different systems: our 𝐼𝐴𝑆/𝐼𝑆𝐺𝑒𝑉 measurement (i), the frequency shift of the GeV ZPL in our Stokes PL spectra (ii), and the equivalent measurement from the literature (iii), the ZPL wavelength shift of the SnV (iv) and of the SiV center (v), the intensity of the NV ZPL (vi), the Raman 𝐼𝐴𝑆/𝐼𝑆 ratio achieved for a bulk thermometer (vii) and the spectral shift of quantum dots (viii). The literature data are plotted over the entire temperature range demonstrated in each paper. References 1. G. G. Stokes, On the Change of Refrangibility of Light. Philosophical Transactions of the Royal Society of London 142, 463-562 (1852). V. Nathan, A. H. Guenther, S. S. Mitra, Review of multiphoton absorption in crystalline solids. Journal of the Optical Society of America B 2, 294-316 (1985). H. Drexler, D. Leonard, W. Hansen, J. P. Kotthaus, P. M. Petroff, Spectroscopy of Quantum Levels in Charge-Tunable InGaAs Quantum Dots. Phys. Rev. Lett. 73, 2252-2255 (1994). D. W. Shipp, F. Sinjab, I. Notingher, Raman spectroscopy: techniques and applications in the life sciences. Adv. Opt. Photon. 9, 315-428 (2017). T. M. Sweeney et al., Cavity-stimulated Raman emission from a single quantum dot spin. Nat. Photon. 8, 442 (2014). R. I. Epstein, M. I. Buchwald, B. C. Edwards, T. R. Gosnell, C. E. Mungan, Observation of laser-induced fluorescent cooling of a solid. Nature 377, 500 (1995). J. Zhang, D. Li, R. Chen, Q. Xiong, Laser cooling of a semiconductor by 40 kelvin. Nature 493, 504 (2013). C. L. Evans et al., Chemical imaging of tissue <em>in vivo</em> with video-rate coherent anti-Stokes Raman scattering microscopy. Proc. Natl. Acad. Sci. 102, 16807-16812 (2005). C. H. Camp Jr, M. T. Cicerone, Chemically sensitive bioimaging with coherent Raman scattering. Nat. Photon. 9, 295 (2015). Y. Xu et al., Strong anti-Stokes luminescence from H+-irradiated diamond. Appl. Phys. Lett. 83, 1968-1970 (2003). 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. Q. Wang et al., Photoluminescence Upconversion by Defects in Hexagonal Boron Nitride. Nano Lett., (2018). D. V. Seletskiy et al., Laser cooling of solids to cryogenic temperatures. Nat. Photon. 4, 161 (2010). 13. M. W. Doherty et al., The nitrogen-vacancy colour centre in diamond. Phys. Rep. 528, 1-45 14. 15. (2013). E. Neu et al., Single photon emission from silicon-vacancy colour centres in chemical vapour deposition nano-diamonds on iridium. New J. Phys. 13, (2011). A. Sipahigil et al., An integrated diamond nanophotonics platform for quantum-optical networks. Science 354, 847-850 (2016). 16. M. K. Bhaskar et al., Quantum Nonlinear Optics with a Germanium-Vacancy Color Center in 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. a Nanoscale Diamond Waveguide. Phys. Rev. Lett. 118, 223603 (2017). C. Bradac et al., Observation and control of blinking nitrogen-vacancy centres in discrete nanodiamonds. Nat. Nanotechnol. 5, 345-349 (2010). I. I. Vlasov et al., Molecular-sized fluorescent nanodiamonds. Nat. Nanotechnol. 9, 54-58 (2014). N. Aslam, G. Waldherr, P. Neumann, F. Jelezko, J. Wrachtrup, Photo-induced ionization dynamics of the nitrogen vacancy defect in diamond investigated by single-shot charge state detection. New J. Phys. 15, 013064 (2013). N. B. Manson et al., NV- - N+ pair centre in 1b diamond. arXiv:1807.08889 [cond-mat.mtrl- sci], (2018). E. Neu et al., Low-temperature investigations of single silicon vacancy colour centres in diamond. New J. Phys. 15, 043005 (2013). F. Vetrone et al., Temperature Sensing Using Fluorescent Nanothermometers. ACS Nano 4, 3254-3258 (2010). C. D. S. Brites et al., Thermometry at the nanoscale. Nanoscale 4, 4799-4829 (2012). R. C. Maher, L. F. Cohen, J. C. Gallop, E. C. Le Ru, P. G. Etchegoin, Temperature- Dependent Anti-Stokes/Stokes Ratios under Surface-Enhanced Raman Scattering Conditions. The Journal of Physical Chemistry B 110, 6797-6803 (2006). C. T. Nguyen et al., All-optical nanoscale thermometry with silicon-vacancy centers in diamond. Appl. Phys. Lett. 112, 203102 (2018). P.-C. Tsai et al., Measuring Nanoscale Thermostability of Cell Membranes with Single Gold -- Diamond Nanohybrids. Angewandte Chemie International Edition 56, 3025-3030 (2017). J.-W. Fan et al., Germanium-Vacancy Color Center in Diamond as a Temperature Sensor. ACS Photonics 5, 765-770 (2018). 28. M. Alkahtani et al., Tin-vacancy in diamonds for luminescent thermometry. Appl. Phys. Lett. 29. 112, 241902 (2018). T. Plakhotnik, H. Aman, H.-C. Chang, All-optical single-nanoparticle ratiometric thermometry with a noise floor of 0.3 K Hz −1/2. Nanotechnology 26, 245501 (2015). 30. M. Mecklenburg et al., Nanoscale temperature mapping in operating microelectronic devices. 31. 32. 33. 34. Science 347, 629-632 (2015). S. Li, K. Zhang, J.-M. Yang, L. Lin, H. Yang, Single Quantum Dots as Local Temperature Markers. Nano Lett. 7, 3102-3105 (2007). H. Liu et al., Intracellular Temperature Sensing: An Ultra-bright Luminescent Nanothermometer with Non-sensitivity to pH and Ionic Strength. Scientific reports 5, 14879 (2015). S. Kalytchuk et al., Carbon Dot Nanothermometry: Intracellular Photoluminescence Lifetime Thermal Sensing. ACS Nano 11, 1432-1442 (2017). Y. Tian et al., Size-dependent upconversion luminescence and temperature sensing behavior of spherical Gd2O3:Yb3+/Er3+ phosphor. RSC Advances 5, 14123-14128 (2015). 35. M. H. Alkahtani, C. L. Gomes, P. R. Hemmer, Engineering water-tolerant core/shell 36. upconversion nanoparticles for optical temperature sensing. Opt. Lett. 42, 2451-2454 (2017). J.-C. Boyer, F. C. J. M. van Veggel, Absolute quantum yield measurements of colloidal NaYF4: Er3+, Yb3+ upconverting nanoparticles. Nanoscale 2, 1417-1419 (2010). G. Jiang et al., 794 nm excited core -- shell upconversion nanoparticles for optical temperature sensing. RSC Advances 6, 11795-11801 (2016). C. C. Fu et al., Characterization and application of single fluorescent nanodiamonds as cellular biomarkers. Proc. Natl. Acad. Sci. 104, 727-732 (2007). E. A. Ekimov, O. S. Kudryavtsev, N. E. Mordvinova, O. I. Lebedev, I. I. Vlasov, High- Pressure Synthesis of Nanodiamonds from Adamantane: Myth or Reality? ChemNanoMat 4, 269-273 (2018). K. M. Kondrina, O. S. Kudryavtsev, I. I. Vlasov, R. A. Khmelnitskiy, E. A. Ekimov, High- pressure synthesis of microdiamonds from polyethylene terephthalate. Diamond Relat. Mater. 83, 190-195 (2018). T. T. Tran et al., Nanodiamonds with photostable, sub-gigahertz linewidth quantum emitters. APL Photonics 2, 116103 (2017). 37. 38. 39. 40. 41.
1809.05090	1	1809	2018-09-12T04:48:52	Stimulated Raman with Broadband LED Stokes Source for Analysis of Glucose	[ "physics.app-ph", "physics.optics" ]	We demonstrate stimulated Raman gain using a broadband LED Stokes source to measure vibrational spectra of aqueous glucose solutions. This versatile and cost-effective method increases Raman signal for a variety of applications. We measured both stimulated Raman and spontaneous Raman spectra of glucose solutions with concentrations up to 10 mM with a photon counter and lock-in amplifier. We built partial least squares regression models based on both stimulated Raman and spontaneous Raman spectral data measured with each instrument for predicting concentrations of the glucose solutions. The stimulated Raman spectra measured with the lock-in amplifier based model had the strongest predictive power and predicted the concentrations of the test set of glucose solutions with a mean squared error value an order of magnitude lower than those of the spontaneous Raman based model.	physics.app-ph	physics	Stimulated Raman with Broadband LED Stokes Source for Analysis of Glucose Peter Bullen 1, Ioannis Kymissis 1, Adler Perotte 2 1 Columbia University, 500 W 120th St, New York, NY 10027 2 Columbia University Medical Center, 617 West 168th St., New York, NY 10032 [email protected] We demonstrate stimulated Raman gain using a broadband LED Stokes source to measure vibrational spectra of aqueous glucose solutions. This versatile and for a variety of applications. We cost-effective method increases Raman signal measured both stimulated Raman and spontaneous Raman spectra of glucose solutions with concentrations up to 10 mM with a photon counter and lock-in amplifier. We built partial least squares regression models based on both stimulated Raman and spontaneous Raman spectral data measured with each instrument for predicting concentrations of the glucose solutions. The stimulated Raman spectra measured with the lock-in amplifier based model had the strongest predictive power and predicted the concentrations of the test set of glucose solutions with a mean squared error value an order of magnitude lower than those of the spontaneous Raman based model. Introduction Stimulated Raman Scattering (SRS) spectroscopy is a powerful technology for high sensitivity quantitative analysis of molecular vibrations. Raman spectra are determined by the fundamental vibrational modes of molecules, and therefore are highly specific compared spectra generated by other spectroscopic techniques such as NIR absorption spectroscopy.1 Furthermore, SRS provides a significant signal enhancement compared to the relatively weak spontaneous Raman signal, enabling detection of molecules in highly dilute solutions and a variety of other high sensitivity applications.2 SRS involves two light sources: pump and Stokes. When the frequency difference between the pump and Stokes sources corresponds to one of the Raman vibrational modes of the molecules in the sample, the molecular transition probability greatly increases.3 This technique results in intensity gain at the Stokes frequency, Stimulated Raman Gain (SRG), and intensity loss at the pump frequency, Stimulated Raman Loss (SRL).4 Both the SRG and SRL signals can be orders of magnitude higher than the spontaneous Raman signal, and either may be used for molecular detection or imaging.5,6 The increased stimulated Raman signal enables high sensitivity detection and high contrast imaging with fast integration times and low laser excitation powers, all of which are critical to biomedical applications. Due to the combination of these beneficial attributes, SRS microscopy has been used to great effect for imaging biological issue in vivo at video-rate speeds.7 In recent years, researchers have developed a wide variety of powerful SRS-based biomedical technologies including three-dimensional spectral imaging of proteins,8 in vivo brain tumor imaging,9 and a handheld in vivo SRS microscope.10 SRS also holds several advantages compared to Coherent Anti-Stokes Scattering (CARS) spectroscopy, another coherent Raman technology often used for sensitive high-speed imaging. While CARS can be obscured by non-resonant background and autofluorescence, SRS is unaffected by these phenomena.11,12,13 Depending on the modulation method used, SRS can reach higher levels of sensitivity, limited only by shot noise.14 Finally, SRS spectra are identical to spontaneous Raman spectra, which enables easier analysis and comparison to measured spontaneous Raman spectra in the literature.15 One limitation of the standard SRS system is that it can only measure the SRG or SRL signal corresponding to a single Raman-active molecular vibration at a time: that with a resonant frequency equal to difference between the pump and Stokes frequencies. This presents a challenge for applications requiring a broader Raman spectrum. Quantitative analysis of complex multi-component samples often relies on a specific vibrational mode, separate from the mode of interest, to serve as a reference measurement or internal standard. For example, in vivo blood glucose level monitoring has been demonstrated via spontaneous Raman spectroscopy with the characteristic glucose mode at 1125 cm-1 being the vibrational mode of interest and the hemoglobin mode at 1549 cm-1 serving as the internal standard.16 In order to perform similar types of full spectrum measurements with the benefit of SRS enhanced signal, researchers have developed innovative SRS systems based on tunable pump or Stokes sources which scan across the Raman spectrum,17,18 tunable optical bandpass filters which filter continuum laser sources down to scanning narrowband pump or Stokes beams,19-21 and femtosecond broadband stimulated Raman spectroscopy (FSRS).22-24 FSRS in particular has shown great promise for high speed spectroscopic applications such as monitoring chemical reactions in real time due to its high SNR, temporal resolution, and bandwidth. To further reduce noise generated by jitter between the pump and Stokes beam, an FSRS system in which both the narrowband Stokes and continuum pump originate from the same titanium-sapphire laser oscillator has been demonstrated.25 One drawback of FSRS and many SRS systems in general is the high cost and lack of portability of major components such as femtosecond lasers. Some spectroscopic applications, such as the non-invasive blood monitoring mentioned previously, would benefit from the broadband signal enhancement of FSRS, but must also be cost-effective and adaptable to suit real-world medical environments. For applications in which only a moderate Raman signal enhancement is required, a cost-effective broadband SRS solution is highly desirable. In this article, we demonstrate a broadband SRS system which uses a high-power LED as the continuum Stokes source and a cw laser pump. Due to the lower peak power of LEDs compared to pulsed laser sources, this system enhances the Raman signal with lower efficiency than the traditional SRS system involving two laser sources. However, this prototype system allows for modest SRG of all Raman modes over the spectral width of the LED, while reducing cost and complexity. We test the capability and limitations of this SRS system by measuring Raman spectra of aqueous glucose solutions. We chose to measure glucose solutions as our test samples due to the recent interest in developing non-invasive blood glucose monitoring.16,26,27 In order to take advantage of the full spectrum data and modest enhancement, we use the measured spectra to build partial least squares regression (PLS) models able to predict the glucose concentration of a solution for its Raman spectrum. Even the modest signal enhancement from our SRS system is valuable for building predictive statistical models such as PLS because data across the entire spectrum contributes to the model, leading to more accurate predictions.28,29 We build PLS models from both spontaneous and stimulated Raman spectral data and compare their predictive power for glucose solution concentration. Experimental Setup In order to make a fair comparison between broadband stimulated Raman with LED Stokes source and standard spontaneous Raman, we built an optical system capable of employing either spectroscopic technology without any modifications to the optical components. The only difference between the two techniques is that stimulated Raman involves two light sources: both the LED and laser. Regardless of the spectroscopic method, the optical system directs light to the liquid sample, collects and filters the scattered light, and detects the Raman scattered light with a photomultiplier tube (Fig. 1). The photomultiplier tube is connected to one of two instruments for the final signal measurement: either a photon counter or lock-in amplifier, and measurements were taken with both instruments for both spontaneous and stimulated Raman for comparison. Optical fibers are used to connect some parts of the optical setup to each other in order to simplify system optimization as adjustments to one portion of the system do not cause cause misalignment in other parts connected via optical fibers. For both spontaneous and stimulated Raman spectroscopy, the 532 nm pump (excitation) laser is split by a 10:90 beam splitter, and the smaller portion is reflected towards a photodiode and used as an intensity reference to compensate for laser drift. The transmitted portion of the beam is directed through an aperture stop and laser line filter to clean up the spatial profile of the beam and attenuate any spectral energy apart from the 532 nm laser line. The laser power is 200 mW at the source and is about 150 mW upon reaching the sample. A longpass dichroic beam splitter, oriented 45° with respect to the incident beam, is used to reflect the pump beam to the sample while permitting the LED Stokes beam to be transmitted through towards the same point in the sample. This is possible because the dichroic reflects over 94% of incident light at or below 532 nm, including the pump beam, and transmits over 93% of light above 541.6 nm, which corresponds to most of the Stokes Beam so that it may stimulate Raman modes with frequency shift of 333 cm-1 or greater. The pump and Stokes beams are focused into the volume of the liquid sample by an aspheric lens. Scattered light is collected in the forward direction by the a low f-number aspheric lens, and the low f-number of 1 and high numerical aperture of 0.5 ensures high collection efficiency. The collected light is collimated and coupled into an optical fiber via coupling lens, further filtered by a dielectric longpass filter, and directed into the monochromator. The longpass filter and monochromator both attenuate the intense Rayleigh scattered light from the pump beam to prevent stray light from affecting the final spectra as much as possible. The entrance and exit slits of the monochromator are set at 100 microns to insure enough light passes through while maintaining sufficient spectral resolution. The monochromator must be adjusted manually to measure the Raman signal at each wavelength. Figure 1: Stimulated Raman Spectroscopy system diagram. The beams from the excitation laser and LED are combined via dichroic beam splitter and focused into the sample. Scattered light is ultimately detected by Photomultiplier Tube (PMT) and measured using either a photon counter or lock-in amplifier. The optical chopper is only used in conjunction with the lock-in amplifier and provides the reference frequency. PD = photodiode, BS = beam splitter, MR = mirror, AS = aperture stop, LF = line filter, DC = dichroic beam splitter, LN = lens, SA = sample, CL = collimating/coupling lens, LP = longpass filter, PL = polarizer, MT = optical fiber mount Light passing through the monochromator is detected by the photomultiplier tube (Hamamatsu R4220P) and the signal is measured by either the photon counter (Stanford Research Systems SR400) or Lock-in Amplifier (Stanford Research Systems SR810). Our photomultiplier tube is optimized for visible light measurements, which is one of the reasons we chose to use visible pump and Stokes sources. The photomultiplier tube is reverse biased at 1.1 kV, a value chosen to optimize sensitivity without introducing excess noise. The photomultiplier tube is wrapped in layers of aluminum foil to block out stray light from the room. Measurements are taken with the room lights out to further minimize stray light, and the room is cooled to about 15 °C, resulting in a low dark current of 0.48 counts per second measured by the photon counter. Both the photon counter and lock-in amplifier data collection is automated with Labview software. For stimulated Raman measurements, the Stokes beam is generated by seven high powered LEDs centered at 567 nm. The LEDs are wired together in series and soldered to a single aluminum PCB base with a heat sink for temperature stabilization. The optical power from the LEDs is about 550 mW and is collected by seven ends of a multi-furcated optical fiber connected to a 3D-printed optical fiber mount. The high power is necessary for efficient SRS, especially due to the significant loss inherent to coupling incoherent light into optical fibers. The Stokes beam enters the main portion of the optical system via the multi-furcated optical fiber from the direct opposite side of the dichroic from the sample. The beam is collimated and vertically polarized to match the polarization of the pump beam for optimal stimulated Raman gain. The collimated and polarized beam is transmitted through the dichroic and focused into the sample through the same lens as the pump beam in order to increase spatial overlap between the focal volumes of the two light sources. At the position of the sample, the Stokes beam power is about 10 mW. A small portion of the beam is reflected by the dichroic beam splitter towards a photodiode and used as an intensity reference. A small portion of the pump beam is also transmitted through the dichroic and is filtered out by a longpass filter so the photodiode monitors only the Stokes beam. By monitoring a reference intensity for both pump and Stokes sources, we found that after turning the light sources on, the intensity decreased over the course of several hours by about 5% and 15% respectively before stabilizing. To further mitigate the effect of intensity drift, we took all measurements a minimum of 24 hours after turning on both light sources. The aqueous glucose solutions are contained in transparent cuvettes (Eppendorf UVette). The path length through the cuvette is 1 cm, and volume of the solutions is 2 mL. The glucose solutions are composed of DI water and pharmaceutical grade D-glucose (Sigma Aldrich). The cuvettes are mounted in a 3D-printed mount to insure precise and consistent positioning between the two lenses to minimize any systematic error between different samples due to cuvette position. Stimulated Raman Measurements When taking stimulated Raman measurements with the photon counter, two measurements are required for each data point: a combined signal (Icom) in which both the pump and Stokes beams are present and the Stokes signal (IStokes) in which the pump is blocked. The Stokes measurements were taken immediately after the combined measurement by blocking the pump laser with a manual shutter. The Raman gain, defined as the quotient of the combined and Stokes signals, is exponentially proportional to the concentration of the Raman-active molecule and the pump signal,22 as shown in equation (1), (1) where a is a proportionality constant, σR is the Raman cross section, c is the concentration, and z is the focal depth in which the beams spatially overlap. We tested our stimulated Raman system on methanol solutions for this relationship between pump intensity and Raman gain. We choose methanol for our test solution because it has greater Raman response than the glucose solutions and has a prominent Raman mode at 565 nm (1062 cm-1), which is close to one of the main glucose Raman modes at 567 nm (1120 cm-1). We increased the pump power from 200 mW to 1 W and found that the Raman gain increased exponentially with the pump power (Fig. 2). (a) (b) Figure 2: (a) Natural logarithm of Raman gain as a function of pump power for 1062 cm-1 mode in methanol. The natural logarithm follows a linear trend with respect to pump power with R2 value of 0.995, indicating that the variation between Raman gain and pump power may be modeled by exponential fit. (b) However, the Raman gain itself also follows a linear trend with respect to pump power with R2 value of .993, indicating that the Raman gain may appropriately be modeled by its first-order expansion in equation (3) at these pump power levels. However, in most stimulated Raman experiments, including ours, which use low to moderate power levels for the pump beam, the gain is relatively small, and equation (2) is satisfied so we can approximate the exponential as its first-order expansion (3). (2) (3) To make the significance of our results more apparent, we define the difference between the combined and Stokes measurements to be the stimulated Raman gain signal (ΔISRG) (4). (4) Then, substituting equations (3) and (4) into equation (1), we arrive at equation (5): which simplifies to (6) showing that the SRG signal is proportional to both the pump and Stokes intensities in our small signal approximation. (5) All stimulated Raman spectra measured using the photon counter were calculated as a difference between the combined and Stokes signals for a particular glucose solution sample (Fig. 3) according to equation (4) instead of as the quotient in equation (1) in this report. This approximation is made in order to make comparing spectra more straightforward as both spontaneous and SRG difference spectra share the same units of counts while the Raman Gain quotient is a unitless quantity. Furthermore, the measurements taken by the lock-in amplifier are essentially difference measurements as the lock-in signal is proportional to the amplitude of the component of the input signal varying at the frequency of the optical chopper. Although the SRG difference signal is proportional to the Stokes signal which varies with frequency shift, this did not significantly affect the predictive power of our partial least squares regression models. (a) (b) Figure 3: (a) Combined (pump laser and Stokes LED) and Stokes (only Stokes LED) signals for 10 mM glucose sample measured with the photon counter. (b) Raw Stimulated Raman Gain signal is approximated as the difference between the combined and Stokes signals. In order to build robust numerical models that account for intensity drift of the pump laser and Stokes LED throughout the data collection process, intensity references were taken for both light sources with photodiodes for each data point. Since the stimulated Raman gain is proportional to both the pump and Stokes intensity, each stimulated Raman gain measurement was divided by both the pump and Stokes reference taken for that point before the spectra were used by the partial least squares model. In order to compare these processed normalized Raman spectra to the spontaneous Raman spectra, each spontaneous measurement was divided by the pump reference for that point and the average of the Stokes references for the stimulated measurements since the spontaneous measurements do not involve the Stokes LED but must still be divided by an average Stokes reference for fair comparison to the normalized stimulated spectra. This normalization process improved the predictive power of our partial least squares models overall with the normalization to the pump reference being more significant than that of the Stokes reference. Experimental Process We measured the spontaneous and stimulated Raman spectra of aqueous glucose solutions with our optical setup. The solutions were contained in cuvettes and glucose concentrations ranged from 10 mM to 0 mM (only DI water). This range is useful for study because normal human blood glucose concentration averages at about 5.5 mM. Using both the photon counter and lock-in amplifier, we collected both spontaneous and stimulated Raman spectra for three separate glucose solutions at each concentration level. This would allow us to conveniently divide the data into training, validation, and test sets later when analyzing the data. We collected data from 540 to 570 nm on our monochromator for each sample, which resulted in Raman spectra from 289 to 1262 cm-1 (conversion to wavenumbers includes an instrument-specific calibration offset). Measurements were taken one data point at a time every 0.5 nm from 540 to 570 nm, which accounts for most of the significant spectral features of glucose. The integration time for each measurement was 5 seconds using the photon counter and the time constant for the lock-in amplifier was 3 seconds. The average dark current of 2.4 counts per 5 seconds was subtracted from each photon counter measurement. Results The glucose solution spectra measured via spontaneous and stimulated Raman spectroscopy methods show similar spectral features corresponding to aqueous glucose Raman modes investigated in the literature.30,31 Due to using a monochromator and taking one data point rather than a full spectrum at a time, our spectra have relatively low resolution, but be can identify groups of Raman modes that likely correspond to a particular spectral feature. The spectral feature around 1092-1139 cm-1 likely corresponds to the COH bending mode characteristic of glucose, that between 853-918 cm-1 is the combination of C-H and C-C bending and stretching modes of both alpha and beta glucose anomers, that around 675-724 cm-1 is composed of deformations in the ring, and the large group of features between 424-575 cm-1 is likely the combination of C-C-O deformation modes and endocyclic vibrational modes of the ring. The normalized Raman intensity is generally higher for the stimulated Raman measurements compared to the spontaneous Raman measurements throughout the spectra for both measurements taken with the photon counter and with the lock-in amplifier (Fig. 4). This suggests that the the pump laser is interacting with the Stokes LED and causing stimulated Raman gain in the glucose solutions. It is likely that our stimulated data are result of both the spontaneous and stimulated Raman effects scattering light simultaneously, but here we are concerned only with the total signal measured using both the pump and Stokes LED and not what portion of this signal is generated by the stimulated Raman effect. (a) (b) Figure 4: Comparison of stimulated and spontaneous Raman spectra of a 10 mM glucose solution measured with (a) photon counter and (b) lock-in amplifier. The enhancement between stimulated and spontaneous spectra is greater with the photon counter spectra, but the stimulated photon counter spectra also has significantly higher noise. The relative intensity increase between stimulated and spontaneous Raman was greater for for the measurements taken with the photon counter than those taken with the lock-in amplifier. For the photon counter, the stimulated measurements were 1.5-3 times higher than the spontaneous measurements across the spectra for a given glucose solution. However, this enhancement factor was not strongly correlated with the Raman modes, so the relative increase between stimulated and spontaneous measurements was similar for both Raman-active and non-Raman-active regions of the spectra. In contract, the stimulated measurements taken with the lock-in amplifier were 1.05-1.3 times only higher than the spontaneous measurements, but this enhancement factor was strongly correlated with the Raman modes. The enhancement factor was 1.2-1.3 around the Raman-active regions while only 1.05-1.15 for the non-Raman-active regions. The relative difference between the intensity of Raman peaks and the non-Raman-active background is generally more critical than absolute intensity for numerical analysis, so the stimulated lock-in measurements are at an advantage here. Another weakness of our photon counter stimulated Raman measurements is noise. Across all spectra, the average coefficient of variation, a measure of relative noise and defined as the quotient of the standard deviation and mean, is 16% for photon counter stimulated, 0.38% for photon counter spontaneous, 0.29% for lock-in stimulated, and 0.31% for lock-in spontaneous measurements. The photon counter stimulated Raman measurement noise is almost 2 orders of magnitude higher than all other measurement techniques, which can be confirmed qualitatively by the noiser photon counter stimulated Raman spectra (Fig. 5). While the lock-in amplifier did reduce the noise slightly from 0.38% to 0.31% for the spontaneous measurements, it is clearly most useful in reducing the stimulated measurement noise from 16% to 0.29%. While monitoring the reference intensities of both the pump laser and Stokes LED compensated for intensity drift to some extent, the Stokes signal alone measured by the photon counter was almost 2 orders of magnitude across the spectrum, so any fluctuations in the LED would have a relatively large effect on the stimulated Raman signal. Overall, the additional noise introduced by the LED is a greater detriment to the stimulated photon counter data than any signal enhancement for numerical analysis. (a) (c) (b) (d) Figure 5: (a) Photon counter spontaneous, (b) photon counter stimulated, (c) lock-in spontaneous, (d) lock-in stimulated Raman spectra of glucose solutions with concentrations from 0-10 mM. Both sets of photon counter spectra are normalized to the highest point of the photon counter stimulated data, and both sets of lock-in spectra are similarly normalized to the highest point of the lock-in stimulated data. Differentiation between the glucose solutions of different concentration can be seen for all measurement techniques. The relative Raman intensity is generally higher for the stimulated spectra compared to the respective spontaneous spectra, but the enhancement factor between stimulated and spontaneous measurements is higher for the photon counter measurements overall. The photon counter stimulated spectra are significantly noisier than the other spectra. The stimulated lock-in spectra show slightly narrower spectral features than the corresponding spontaneous spectra, and Raman modes can be resolved to a slightly higher degree in the 424-575 cm-1 region. Both spontaneous and stimulated Raman spectra of glucose solutions measured with either the photon counter or lock-in amplifier show a significant difference in Raman intensity between the varying glucose solutions (0, 2, 4, 6, 8,10 mM), especially at the group of Raman features between 424-575 cm-1 (Fig. 5). This enables any of the four sets of data to be used as a training set for a statistical model, such as partial least squares, which would predict the glucose concentration of an unknown solution given its Raman spectrum. However, the stimulated Raman spectra measured with the lock-in amplifier is the best set of data for building the predictive model because of its enhanced signal and low noise. Compared to the spontaneous lock-in spectra, the stimulated lock-in spectra are slightly enhanced, with the regions immediately surrounding Raman modes showing the greatest signal increase. The enhancement specifically at the glucose Raman modes creates greater differentiation between spectra of different glucose solutions at the regions most strongly correlated with glucose concentration. Although the stimulated photon counter spectra show even greater overall enhancement than the stimulated lock-in spectra, the enhancement factor is less correlated with Raman modes, and more importantly, these spectra have significantly greater noise, making them a weaker data set for building a predictive model. Partial Least Squares Analysis The partial least squares (PLS) regression is a simple statistical model which can be used for predicting concentrations based on spectral data. In this study, we use PLS to test and compare the predictive power of our spontaneous and stimulated Raman spectral data for predicting concentrations of glucose solutions. We programmed our PLS model in Python using the scikit-learn package [32]. We analyzed our Raman data over the full collected spectral range of 289-1262 cm-1, and two smaller portions of the spectrum: 458-625 cm-1 and 642-1129 cm-1. Since we measured three separate glucose solutions for each concentration value, the first set of measurements for all concentrations 0-10 mM was the training set, the second was the validation set, and the third was the test set. In order to find the optimal number of PLS components for the model, we trained and validated our the model using a progressively higher number of PLS components until the mean squared error of the prediction increased, signifying overfitting. Therefore, the previous number of components was optimal. The optimal number of components for the PLS model based on photon counter stimulated spectra was 2 for each spectral range, likely due the greater noise of these spectra. The number of components for the PLS model based on photon counter spontaneous spectra was 3 or 4 depending on spectral range, and that for the lock-in spontaneous and stimulated spectra was 4 or 5 depending on spectral range. Generally, the greater number of components the PLS model can use without overfitting, the stronger the predictive power of the model. Then, the trained and validated model predicted the glucose concentrations based on the test data set, and these predictions were compared to the actual glucose concentration values (Fig. 6). The mean squared error between the predicted and actual glucose concentration measured the predictive power of each model (Table 1), with lower error signifying higher predictive power of the model. (a) (b) (c) (d) Figure 6: PLS model predicted glucose concentration with respect to actual concentration using the full spectrum of data from each of the four measurement methods: (a) Photon counter spontaneous (R2 = 0.9995), (b) photon counter stimulated (R2 = 0.987), (c) lock-in spontaneous (R2 = 0.9998), (d) lock-in stimulated (R2 = 0.9998). The greater the strength of the linear regression fit, the lower the mean squared error, and the stronger the predictive power of the PLS model. Photon Counter Lock-in Amplifier Table 1: Mean squared error between predicted and actual glucose concentrations for each measurement method over three different spectral ranges. Lower mean squared error values indicate stronger predictive power of the model, so the PLS model based on lock-in stimulated data was consistently the strongest predictive model. The mean squared error between predicted and actual glucose concentrations for the PLS model based on the full spectral range of lock-in stimulated Raman data is as low as 9.96x10-4 mM, which is the lowest mean squared error from any of the models based on data from any of the measurements methods. The mean squared error values from the lock-in stimulated Raman data are about an order of magnitude smaller than those from either of the spontaneous Raman data and over 2 orders of magnitude smaller than the mean squared error values from the photon counter stimulated Raman data. When using the lock-in amplifier to reduce error from the Stokes LED, the PLS models based on stimulated Raman data are consistently the strongest glucose concentration predicting models. Interestingly, the lock-in stimulated data is only slightly increased compared to the spontaneous data, but the PLS model based on stimulated data is significantly stronger, likely due in part to the slightly narrower Raman modes of the stimulated spectra. The stimulated model's higher R-squared value of the linear regression between predicted and actual concentration further corroborate this result. Furthermore, when the training, validation, and test sets are switched around; for example, the second set is training, third is validation, and first is test; the results are similar, showing that the lock-in stimulated model is robust and consistent. Discussion This study demonstrates the viability broadband stimulated Raman with an LED Stokes source and shows the stronger predictive power of a PLS model based on this stimulated Raman data compared to spontaneous Raman when using a lock-in amplifier to reduce noise. With further developments, this technology may be useful for improved cost-effective spectroscopic molecular identification and concentration prediction in a variety of fields. Prediction of glucose concentration has particularly useful biomedical applications in non-invasive blood glucose level monitoring. However, it is important to note that this technology has several limitations, and the enhancement of Raman modes achieved by stimulated Raman gain this way is relatively small compared to traditional SRS, broadband FSRS, and other enhancement technologies, such as SERS. While broadband LED stimulated Raman is intended to be a cost-effective solution and does not necessarily need to match these powerful Raman enhancement technologies, our experiment demonstrates stimulated Raman having only a small enhancement over spontaneous Raman spectroscopy, so further improvements are necessary to be truly useful. The detection system, specifically the monochromator and photomultiplier tube, could be improved to make taking spectral measurements more practical while removing sources of signal loss. We decided to use the photomultiplier because of its high sensitivity and dynamic range. This was necessary because the unfiltered light from the Stokes LED would saturate a detector with lower dynamic range, and we needed to detect the relatively small stimulated Raman signal on top of the LED signal. Due to using a photomultiplier tube, a monochromator was necessary to isolate a single wavelength, so measuring a full spectrum was accomplished one point at a time. This made taking measurements a time intensive process, which from a practical standpoint, offset the benefit of stimulated Raman gain. Furthermore, our signal experienced significant loss through the monochromator. Replacing the monochromator and photomultiplier with a high dynamic range spectrometer would dramatically speed up the measurement process and potentially increase overall signal to noise ratio. The LED introduces several challenges to the Raman system. As an incoherent light source, the LED light diffracts quickly, and loses a significant amount of signal when coupled into the multi-furcated fiber. This results in a Stokes source orders of magnitude lower in intensity than the pump at the sample, which leads to low stimulated Raman gain. The LED also adds noise though fluctuations and intensity drift, which may counteract the benefit of creating stimulated Raman. Since the intensity of the Stokes LED is almost 2 orders of magnitude higher than that of the stimulated Raman signal, even the Poisson noise of the LED is significant relative of the stimulated signal. This effect of noise from both the LED and laser was greatly mitigated by using an optical chopper to modulate the pump beam and detecting the stimulated Raman gain signal with a lock-in amplifier synchronized to the modulation of the pump. However, using a lock-in amplifier adds a significant cost to the system, especially if a multi-channel lock-in amplifier would be used to measure a full spectrum all at once. Given cost-effectiveness is one of the main goals of this LED-based stimulated Raman system, the high cost of multi-channel lock-in amplifiers is currently a significant barrier to the practical application of this technology, but the development of cost-effective digital lock-in amplifiers is an active area of research.33,34 It may also be possible to use the relatively noisy data collected without the lock-in amplifier, using a photon counter in our case or potentially a spectrometer, if the statistical models used were more robust to noise than our partial least squares model. This could be done, in part, by designing a model that penalizes overfitting. We designed our stimulated Raman system for monitoring glucose concentrations rather than imaging, but broadband LED stimulated Raman also has potential imaging applications. However, aberrations of the LED light would have to be considered, and the imaging resolution would be limited by LED spot size, which is slightly larger than the laser spot size in our setup. Biomedical imaging would require further modifications including lowering the laser power to prevent burning of biological tissue and using NIR wavelengths instead of visible for greater depth of penetration. Supplementary Experiment Temperature changes in the glucose solution sample could potentially alter the Raman spectra, so we conducted a series of experiments to test whether or not the long exposure of the focused laser and LED beams had any effect on the resulting Raman measurements, either spontaneous or stimulated. Using the 10 mM glucose sample and the same integration time of 10 seconds as in the other experiments, we first continuously measured the spontaneous Raman intensity at a single frequency shift value for 10 minutes, for a total of 120 individual measurements. Then, we measured the stimulated Raman intensity for 10 minutes at the same frequency shift value by subtracting the alternating combined and Stokes measurements as usual. Given each stimulated Raman measurement is the difference of two consecutive measurements, this resulted in 60 stimulated Raman measurements total. This process was performed at 592, 886, 1124, 2101, and 3400 cm-1. The Raman modes at 592 and 1124 cm-1 are two of the most prominent in the spectra we collected, with that at 592 cm-1 being the global maximum and that at 1124 cm-1 being the mode which experienced the greatest increase between stimulated Raman and spontaneous Raman. We decided to also take measurements at 886 and 2101 cm-1 in order to test if there was a temperature effect on the baseline or off-peak regions, and the highest point of the strong group of water modes, 3400 cm-1, was chosen to see if the spectral features related to water would change. Since we were testing the possibility of the laser heating the samples and effecting the Raman measurements, there was a 10 minute gap between rounds of measurements so the sample could cool if needed. Three rounds of measurements were taken at each of the five frequency shift values. Our experiments showed no significant time-dependent Raman intensity change in any of the tests, suggesting that whatever heating effect the beams may have had on the samples was not significant enough to alter the Raman spectra. This is important because it rules out heating as a potential systematic source of error for our spectral measurements and allows us to make fair comparisons between our spontaneous and stimulated Raman measurements. References [1] N. C. Dingari, I. Barman, G. P. Singh, J. W. Kang, R. R. Dasari, and M. S. Feld, "Investigation of the specificity of Raman spectroscopy in non-invasive blood glucose measurements," Anal. Bioanal. Chem., vol. 400, no. 9, pp. 2871 -- 2880, Jul. 2011. [2] P. Nandakumar, A. Kovalev, and A. Volkmer, "Vibrational imaging based on stimulated Raman scattering microscopy," New J. Phys., vol. 11, no. 3, p. 33026, 2009. [3] J.-X. Cheng and X. S. Xie, "Vibrational spectroscopic imaging of living systems: An emerging platform for biology and medicine," Science (80-. )., vol. 350, no. 6264, pp. aaa8870-aaa8870, 2015. [4] C.-S. Liao and J.-X. Cheng, "In Situ and In Vivo Molecular Analysis by Coherent Raman Scattering Microscopy," Annu. Rev. Anal. Chem., vol. 9, no. 1, pp. 69 -- 93, 2016. [5] E. R. Andresen, P. Berto, and H. Rigneault, "Stimulated Raman scattering microscopy by spectral focusing and fiber-generated soliton as Stokes pulse," Opt. Lett., vol. 36, no. 13, p. 2387, 2011. [6] D. Zhang, M. N. Slipchenko, and J. X. Cheng, "Highly sensitive vibrational imaging by femtosecond pulse stimulated raman loss," J. Phys. Chem. Lett., vol. 2, no. 11, pp. 1248 -- 1253, 2011. [7] B. G. Saar, C. W. Freudiger, J. Reichman, C. M. Stanley, G. R. Holtom, and X. S. Xie, "Video-Rate Molecular Imaging in Vivo with Stimulated Raman Scattering.," Sci. (Washington, DC, United States), vol. 330, no. 6009, pp. 1368 -- 1370, 2010. [8] C. W. Freudiger, W. Min, G. R. Holtom, B. Xu, M. Dantus, and X. S. Xie, "Highly specific label-free molecular imaging with spectrally tailored excitation-stimulated Raman scattering (STE-SRS) microscopy," Nat. Photonics, vol. 5, no. 2, pp. 103 -- 109, 2011. [9] M. Ji et al., "Detection of human brain tumor infiltration with quantitative stimulated Raman scattering microscopy," Sci. Transl. Med., vol. 7, no. 309, p. 309ra163-309ra163, 2015. [10] C. S. Liao et al., "In Vivo and in Situ Spectroscopic Imaging by a Handheld Stimulated Raman Scattering Microscope," ACS Photonics, vol. 5, no. 3, pp. 947 -- 954, 2018. [11] T. W. Kee and M. T. Cicerone, "Simple approach to one-laser, broadband coherent anti-Stokes Raman scattering microscopy," Opt. Lett., vol. 29, no. 23, pp. 2701 -- 2703, 2004. [12] H. T. Beier, G. D. Noojin, and B. A. Rockwell, "Stimulated Raman scattering using a single femtosecond oscillator with flexibility for imaging and spectral applications," Opt. Express, vol. 19, no. 20, p. 18885, 2011. [13] K. Yen, T. T. Le, A. Bansal, S. D. Narasimhan, J. X. Cheng, and H. A. Tissenbaum, "A comparative study of fat storage quantitation in nematode Caenorhabditis elegans using label and label-free methods," PLoS One, vol. 5, no. 9, pp. 1 -- 10, 2010. [14] Y. Ozeki, F. Dake, S. Kajiyama, K. Fukui, and K. Itoh, "Analysis and experimental assessment of the sensitivity of stimulated Raman scattering microscopy," Opt. Express, vol. 17, no. 5, p. 3651, 2009. [15] W. Min, C. W. Freudiger, S. Lu, and X. S. Xie, "Coherent Nonlinear Optical Imaging: Beyond Fluorescence Microscopy," Annu. Rev. Phys. Chem., vol. 62, no. 1, pp. 507 -- 530, 2011. [16] J. Shao et al., "In Vivo Blood Glucose Quantification Using Raman Spectroscopy," PLoS One, 2012. [17] P. N. Malevich et al., "Ultrafast-laser-induced backward stimulated Raman scattering for tracing atmospheric gases," Opt. Express, vol. 20, no. 17, p. 18784, 2012. [18] T. Steinle et al., "Synchronization-free all-solid-state laser system for stimulated Raman scattering microscopy," Light Sci. Appl., vol. 5, no. 10, pp. 1 -- 6, 2016. [19] Y. Ozeki, W. Umemura, K. Sumimura, N. Nishizawa, K. Fukui, and K. Itoh, "Stimulated Raman hyperspectral imaging based on spectral filtering of broadband fiber laser pulses," Opt. Lett., vol. 37, no. 3, p. 431, 2012. [20] D. Fu et al., "Quantitative Chemical Imaging with Multiplex Stimulated Raman Scattering Microscopy," J. Am. Chem. Soc., vol. 134, no. 8, pp. 3623 -- 3626, 2012. [21] Z. Meng, G. I. Petrov, and V. V. Yakovlev, "Pure electrical, highly-efficient and sidelobe free coherent Raman spectroscopy using acousto-optics tunable filter (AOTF)," Sci. Rep., vol. 6, no. February, pp. 1 -- 7, 2016. [22] D. W. McCamant, P. Kukura, S. Yoon, and R. A. Mathies, "Femtosecond broadband stimulated Raman spectroscopy: Apparatus and methods," Rev. Sci. Instrum., vol. 75, no. 11, pp. 4971 -- 4980, 2004. [23] P. Kukura, D. W. McCamant, and R. A. Mathies, "Femtosecond Stimulated Raman Spectroscopy," Annu. Rev. Phys. Chem., vol. 58, no. 1, pp. 461 -- 488, 2007. [24] E. Ploetz, S. Laimgruber, S. Berner, W. Zinth, and P. Gilch, "Femtosecond stimulated Raman microscopy," Appl. Phys. B Lasers Opt., vol. 87, no. 3, pp. 389 -- 393, 2007. [25] K. Seto, Y. Okuda, E. Tokunaga, and T. Kobayashi, "Development of a multiplex stimulated Raman microscope for spectral imaging through multi-channel lock-in detection," Rev. Sci. Instrum., vol. 84, no. 8, 2013. [26] S. K. Vashist, "Non-invasive glucose monitoring technology in diabetes management: A review," Anal. Chim. Acta, vol. 750, pp. 16 -- 27, 2012. [27] M. S. Wróbel, "Non-invasive blood glucose monitoring with Raman spectroscopy: Prospects for device miniaturization," IOP Conf. Ser. Mater. Sci. Eng., vol. 104, no. 1, 2016. [28] D. M. Haaland and E. V. Thomas, "Partial Least-Squares Methods for Spectral Analyses. 1. Relation to Other Quantitative Calibration Methods and the Extraction of Qualitative Information," Anal. Chem., vol. 60, no. 11, pp. 1193 -- 1202, 1988. [29] K. P. J. Williams and N. J. Everall, "Use of micro Raman spectroscopy for the quantitative determination of polyethylene density using partial least-squares calibration," J. Raman Spectrosc., vol. 26, no. 6, pp. 426 -- 433, 1995. [30] M. Mathlouthi and D. Vinh Luu, "Laser-raman spectra of d-glucose and sucrose in aqueous solution," Carbohydr. Res., vol. 81, no. 2, pp. 203 -- 212, 1980. [31] P. D. Vasko, J. Blackwell, and J. L. Koenig, "Infrared and raman spectroscopy of carbohydrates.," Carbohydr. Res., vol. 23, no. 3, pp. 407 -- 416, 1972. [32] Scikit-learn: Machine Learning in Python, Pedregosa et al., JMLR 12, pp. 2825-2830, 2011. [33] A. Kar and M. Chandra, "A Low-Cost , Portable Alternative for a Digital Lock- In Amplifier Using TMS320C5535 DSP," 2014. [34] J. Lu, D. Pan, and L. Qiao, "The principle of a virtual multi-channel lock-in amplifier and its application to magnetoelectric measurement system," Meas. Sci. Technol., vol. 045702, no. iii, p. 11, 2007.
1811.02705	3	1811	2019-11-19T22:40:36	Heuristic Recurrent Algorithms for Photonic Ising Machines	[ "physics.app-ph", "cs.ET", "physics.optics" ]	The inability of conventional electronic architectures to efficiently solve large combinatorial problems motivates the development of novel computational hardware. There has been much effort recently toward developing novel, application-specific hardware, across many different fields of engineering, such as integrated circuits, memristors, and photonics. However, unleashing the true potential of such novel architectures requires the development of featured algorithms which optimally exploit their fundamental properties. We here present the Photonic Recurrent Ising Sampler (PRIS), a heuristic method tailored for parallel architectures that allows for fast and efficient sampling from distributions of combinatorially hard Ising problems. Since the PRIS relies essentially on vector-to-fixed matrix multiplications, we suggest the implementation of the PRIS in photonic parallel networks, which realize these operations at an unprecedented speed. The PRIS provides sample solutions to the ground state of arbitrary Ising models, by converging in probability to their associated Gibbs distribution. By running the PRIS at various noise levels, we probe the critical behavior of universality classes and their critical exponents. In addition to the attractive features of photonic networks, the PRIS relies on intrinsic dynamic noise and eigenvalue dropout to find ground states more efficiently. Our work suggests speedups in heuristic methods via photonic implementations of the PRIS. We also hint at a broader class of (meta)heuristic algorithms derived from the PRIS, such as combined simulated annealing on the noise and eigenvalue dropout levels. Our algorithm can also be implemented in a competitive manner on fast parallel electronic hardware, such as FPGAs and ASICs.	physics.app-ph	physics	Heuristic Recurrent Algorithms for Photonic Ising Machines Charles Roques-Carmes,1, 2 Yichen Shen,3 Cristian Zanoci,3 Mihika Prabhu,1, 2 Fadi Atieh,2, 3 Li Jing,3 Tena Dubcek,3 Chenkai Mao,2, 3 Miles R. Johnson,4 Vladimir Ceperi´c,3 John D. Joannopoulos,3, 5 Dirk Englund,1, 2 and Marin Soljaci´c1, 3 1Research Laboratory of Electronics, Massachusetts Institute of Technology, 50 Vassar Street, Cambridge MA 02139, USA Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA∗ 2Department of Electrical Engineering and Computer Science, 3Department of Physics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA 4Department of Mathematics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA 5Institute for Soldier Nanotechnologies, 500 Technology Square, Cambridge, MA 02139, USA The inability of conventional electronic architectures to efficiently solve large combinatorial problems motivates the development of novel computational hardware. There has been much effort recently to- ward developing novel, application-specific hardware, across many different fields of engineering, such as integrated circuits, memristors, and photonics. However, unleashing the true potential of such novel architectures requires the development of featured algorithms which optimally exploit their fundamental properties. We here present the Photonic Recurrent Ising Sampler (PRIS), a heuristic method tailored for parallel architectures that allows for fast and efficient sampling from distributions of combinatorially hard Ising problems. Since the PRIS relies essentially on vector-to-fixed matrix multiplications, we sug- gest the implementation of the PRIS in photonic parallel networks, which realize these operations at an unprecedented speed. The PRIS provides sample solutions to the ground state of arbitrary Ising models, by converging in probability to their associated Gibbs distribution. By running the PRIS at various noise levels, we probe the critical behavior of universality classes and their critical exponents. In addition to the attractive features of photonic networks, the PRIS relies on intrinsic dynamic noise and eigenvalue dropout to find ground states more efficiently. Our work suggests speedups in heuristic methods via pho- tonic implementations of the PRIS. We also hint at a broader class of (meta)heuristic algorithms derived from the PRIS, such as combined simulated annealing on the noise and eigenvalue dropout levels. Our algorithm can also be implemented in a competitive manner on fast parallel electronic hardware, such as FPGAs and ASICs. INTRODUCTION Heuristic methods -- probabilistic algorithms with stochas- tic components -- are a cornerstone of both numerical meth- ods in statistical physics1 and NP-Hard optimization2. Broad classes of problems in statistical physics, such as growth patterns in clusters3, percolation4, heterogeneity in lipid membranes5, and complex networks6, can be described by heuristic methods. These methods have proven instrumen- tal for predicting phase transitions and the critical exponents of various universality classes -- families of physical sys- tems exhibiting similar scaling properties near their critical temperature1. These heuristic algorithms have become pop- ular, as they typically outperform exact algorithms at solv- ing real-world problems7. Heuristic methods are usually tai- lored for conventional electronic hardware; however, a num- ber of optical machines have recently been shown to solve the well-known Ising8,9 and Traveling Salesman problems10,11. For computationally demanding problems, these methods can benefit from parallelization speedups1,12, but the determina- tion of an efficient parallelization approach is highly problem- specific1. Half a century before the contemporary Machine Learning Renaissance13, the Little14 and then the Hopfield15,16 networks were considered as early architectures of recurrent neural net- works (RNN). The latter was suggested as an algorithm to solve combinatorially hard problems, as it was shown to de- terministically converge to local minima of arbitrary quadratic Hamiltonians of the form (cid:88) 1≤i,j≤N H (K) = − 1 2 σiKijσj, (1) which is the most general form of an Ising Hamiltonian in the absence of an external magnetic field17. In Equation (1), we equivalently denote the set of spins as σ ∈ {−1, 1}N or S ∈ {0, 1}N (with σ = 2S − 1), and K is a N × N real symmetric matrix. In the context of physics, Ising models describe the in- teraction of many particles in terms of the coupling matrix K. These systems are observed in a particular spin config- uration σ with a probability given by the Gibbs distribution p(σ) ∝ exp(−βH (K)(σ)), where β = 1/(kBT ), with kB the Boltzmann constant and T the temperature. At low tem- perature, when β → ∞, the Gibbs probability of observing the system in its ground state approaches 1, thus naturally minimizing the quadratic function in Equation (1). As sim- ilar optimization problems are often encountered in computer science2,7, a natural idea is to engineer physical systems with dynamics governed by an equivalent Hamiltonian. Then, by sampling the physical system, one can generate candidate so- lutions to the optimization problem. This analogy between statistical physics and computer science has nurtured a great variety of concepts in both fields18, for instance, the analogy between neural networks and spin glasses15,19. Many complex systems can be formulated using the Ising model20 -- such as ferromagnets17,21, liquid-vapor transitions22, lipid membranes5, brain functions23, random photonics24, and strongly-interacting systems in quantum chromodynamics25. From the perspective of optimization, finding the spin distribution minimizing H (K) for an arbitrary matrix K belongs to the class of NP-hard problems26. Hopfield networks deterministically converge to a local minimum, thus making it impossible to scale such networks to deterministically find the global minimum27 -- thus jeop- ardizing any electronic16 or optical28 implementation of these algorithms. As a result, these early RNN architectures were soon superseded by heuristic (such as Metropolis-Hastings (MH)) and metaheuristic methods (such as simulated anneal- ing (SA)29, parallel tempering30, genetic algorithms31, Tabu search32 and local-search-based algorithms33), usually tai- lored for conventional electronic hardware. Even still, heuris- tic methods struggle to solve large problems, and could ben- efit from nanophotonic hardware demonstrating parallel, low- energy, and high-speed computations34 -- 36. In this Letter, we propose a fast and efficient heuristic method for photonic analog computing platforms, relying es- sentially on iterative matrix multiplications. Our heuristic ap- proach also takes advantage of optical passivity and dynamic noise to find ground states of arbitrary Ising problems and probe their critical behaviors, yielding accurate predictions of critical exponents of the universality classes of conven- tional Ising models. Our algorithm presents attractive scal- ing properties when benchmarked against conventional algo- rithms, such as MH. Our findings suggest a novel approach to heuristic methods for efficient optimization and sampling by leveraging the potential of matrix-to-vector accelerators, such as parallel photonic networks.34. Here, we propose a photonic implementation of a passive RNN, which models the arbitrary Ising-type Hamiltonian in Equation (1). RESULTS The proposed architecture of our photonic network is shown in Figure 1. This photonic network can map arbitrary Ising Hamiltonians described by Equation (1), with Kii = 0 (as diagonal terms only contribute to a global offset of the Hamiltonian, see Supplementary Note 1). In the follow- ing, we will refer to the eigenvalue decomposition of K as K = U DU†, where U is a unitary matrix, U† its transpose conjugate, and D a real-valued diagonal matrix. The spin state at time step t, encoded in the phase and amplitude of N parallel photonic signals S(t) ∈ {0, 1}N , first goes through a linear symmetric transformation decomposed in its eigen- value form 2J = USqα(D)U†, where Sqα(D) is a diago- nal matrix derived from D, whose design will be discussed in the next paragraphs. The signal is then fed into nonlinear optoelectronic domain, where it is perturbed by a Gaussian 2 FIG. 1. Operation principle of the PRIS. A photonic analog signal, encoding the current spin state S(t), goes through transformations in linear photonic and nonlinear optoelectronic domains. The result of this transformation S(t+1) is recurrently fed back to the input of this passive photonic system. distribution of standard deviation φ (simulating noise present in the photonic implementation) and is imparted a nonlinear threshold function Thθ (Thθ(x) = 1 if x > θ, 0 otherwise). The signal is then recurrently fed back to the linear photonic domain, and the process repeats. The static unit transforma- tion between two time steps t and t + 1 of this RNN can be summarized as X (t) ∼ N (2JS(t)φ), S(t+1) = Thθ(X (t)) (2) where N (xφ) denotes a Gaussian distribution of mean x and standard deviation φ. We call this algorithm, which is tailored for a photonic implementation, the Photonic Recurrent Ising Sampler (PRIS). The detailed choice of algorithm parameters is described in the Supplementary Note 2. This simple recurrent loop can be readily implemented in the photonic domain. For example, the linear photonic in- terference unit can be realized with MZI networks34,37 -- 39, diffractive optics40,41, ring resonator filter banks42 -- 44, and free space lens-SLM-lens systems45,46; the diagonal matrix multiplication Sqα(D) can be implemented with an electro- optical absorber, a modulator or a single MZI34,47,48; the non- linear optoelectronic unit can be implemented with an opti- cal nonlinearity47 -- 51, or analog/digital electronics52 -- 55, for in- stance by converting the optical output to an analog electronic signal, and using this electronic signal to modulate the input56. The implementation of the PRIS on several photonic architec- tures and the influence of heterogeneities, phase bit precision, and signal to noise ratio on scaling properties are discussed in the Supplementary Note 5. In the following, we will describe the properties of an ideal PRIS and how design imperfections may affect its performance. General theory of the PRIS dynamics The long-time dynamics of the PRIS is described by an effective Hamiltonian HL (see Refs.19,58 and Supplementary Note 2). This effective Hamiltonian can be computed by per- forming the following steps. First, calculate the transition linear photonic domainnonlinear opto-electronic domainoptoelectronic recurrent feedback3 FIG. 2. Scaling performance of the PRIS. (a, top) Ground state energy versus graph order of random spin glasses. A sample graph is shown as an inset in (a, bottom): a fully-connected spin glass with uniformly-distributed continuous couplings in [−1, 1]. Niter,99% versus graph size for spin glasses (a, bottom) and MAX-CUT graphs (b). (c) Niter,99% versus graph density for MAX-CUT graphs and N = 75. The graph density is defined as d = 2E/(N (N − 1)), E being the number of undirected edges. RCG denotes Random Cubic Graphs, for which E = 3N/2. Ground states are determined with the exact solver BiqMac57 (see Methods section). In this analysis, we set α = 0, and for each set of density and graph order we ran 10 graph instances 1,000 times. The number of iterations to find the ground state is measured for each run and Niter,q is defined as the q-th quantile of the measured distribution. probability of a single spin from Equation (2). Then, the tran- sition probability from an initial spin state S(t) to the next step S(t+1) can be written as W (0)(cid:16) S(t+1)S(t)(cid:17) (cid:80) e−βH 0(S(t+1)S(t)) H 0 (SS(cid:48)) = − (cid:88) S e−βH 0(SS(t)) = Jijσi (S) σj (S(cid:48)) , (4) , (3) 1≤i,j≤N where S, S(cid:48) denote arbitrary spin configurations. Let us emphasize that, unlike H (K)(S), the transition Hamiltonian H (0) (SS(cid:48)) is a function of two spin distributions S and S(cid:48). Here, β = 1/(kφ) is analogous to the inverse temperature from statistical mechanics, where k is a constant, only de- pending on the noise distribution (see Supplementary Table 1). To obtain equations (3, 4), we approximated the single spin transition probability by a rescaled sigmoid function and j Jij. In the Supplemen- tary Note 2, we investigate the more general case of arbitrary threshold vectors θi and discuss the influence of the noise dis- tribution. have enforced the condition θi =(cid:80) One can easily verify that this transition probability obeys the triangular condition (or detailed balance condition) if J is symmetric Jij = Jji. From there, an effective Hamilto- nian HL can be deduced following the procedure described by Peretto58 for distributions verifying the detailed balance condition. The effective Hamiltonian HL can be expanded, in the large noise approximation (φ (cid:29) 1, β (cid:28) 1), into H2: (cid:1), Jijσj (cid:88) j (cid:88) log cosh(cid:0)β (cid:88) i 1≤i,j≤N σi[J 2]ijσj. HL = − 1 β H2 = − β 2 H2 = βH (J 2). We set the PRIS matrix J to be a modified square-root of the Ising matrix K by imposing the following condition on the PRIS √ Sqα(D) = 2Re( D + α∆). (7) diagonal term of the Ising coupling matrix ∆ii =(cid:80) We add a diagonal offset term α∆ to the eigenvalue matrix D, in order to parametrize the number of eigenvalues remaining after taking the real part of the square root. Since lower eigen- values tend to increase the energy, they can be dropped out so that the algorithm spans the eigenspace associated with higher eigenvalues. We chose to parametrize this offset as follows: α ∈ R is called the eigenvalue dropout level, a hyperparam- eter to select the number of eigenvalues remaining from the original coupling matrix K, and ∆ > 0 is a diagonal offset matrix. For instance, ∆ can be defined as the sum of the off- j(cid:54)=i Kij. The addition of ∆ only results in a global offset on the Hamil- tonian. The purpose of the ∆ offset is to make the matrix in the square root diagonally dominant, thus symmetric positive definite, when α is large and positive. Thus, other definitions of the diagonal offset could be proposed. When α → 0, some lower eigenvalues are dropped out by taking the real part of the square root (see Supplementary Note 3); we show below that this improves the performance of the PRIS. We will spec- ify which definition of ∆ is used in our study when α (cid:54)= 0. When choosing this definition of Sqα(D) and operating the PRIS in the large noise limit, we can implement any general Ising model (Equation (1)) on the PRIS (Equation (6)). It has been noted that by setting Sqα(D) = D (i.e. the linear photonic domain matrix amounts to the Ising coupling matrix 2J = K), the free energy of the system equals the Ising free energy at any finite temperature (up to a factor of 2, thus exhibiting the same ground states) in the particular case of associative memory couplings19 with finite number of patterns and in the thermodynamic limit, thus drastically (5) (6) Examining Equation (6), we can deduce a mapping of the PRIS to the general Ising model shown in Equation (1) since abcSpin glasses GraphsGround state energy-11Niter, 99%Niter, 99%= 0.7= 0.8= 0.9opt.Niter, 99%d = 0.1d = 0.5d = 0.9RCGMAX-CUT GraphsMAX-CUT Graphsconstraining the number of degrees of freedom on the cou- plings. This regime of operation is a direct modification of the Hopfield network, an energy-based model where the cou- plings between neurons is equal to the Ising coupling between spins. The essential difference between the PRIS in the con- figuration Sqα(D) = D and a Hopfield network is that the former relies on synchronous spin updates (all spins are up- dated at every step, in this so-called Little network14) while the latter relies on sequential spin updates (a single randomly picked spin is updated at every step). The former is better suited for a photonic implementation with parallel photonic networks. In this regime of operation, the PRIS can also benefit from computational speed-ups, if implemented on a conventional architecture, for instance if the coupling matrix is sparse. However, as has been pointed out in theory19 and by our sim- ulations (see Supplementary Note 4, Figure S7), some addi- tional considerations should be taken into account in order to eliminate non-ergodic behaviors in this system. As the regime of operation described by Equation (59) is general to any cou- pling, we will use it in the following demonstrations. Finding the ground state of Ising models with the PRIS We investigate the performance of the PRIS on finding the ground state of general Ising problems (Equation (1)) with two types of Ising models: MAX-CUT graphs, which can be mapped to an instance of the unweighted MAX-CUT problem9 and all-to-all spin glasses, whose connections are uniformly distributed in [−1, 1] (an example illustration of the latter is shown as an inset in Figure 2(a)). Both families of models are computationally NP-hard problems26, thus their computational complexity grows exponentially with the graph order N. The number of steps necessary to find the ground state with 99% probability, Niter,99% is shown in Figure 2(a-b) for these two types of graphs (see definition in Supplementary Note 4 and in the Methods section). As the PRIS can be implemented with high-speed parallel photonic networks, the on-chip real time of a unit step can be less than a nanosecond34,59 (and the initial setup time for a given Ising model is typically of the order of microseconds with thermal phase shifters60). In such architectures, the PRIS would thus find ground states of arbitrary Ising problems with graph orders N ∼ 100 within less than a millisecond. We also show that the PRIS can be used as a heuristic ground state search algorithm in regimes where exact solvers typically fail (N ∼ 1, 000) and bench- mark its performance against MH and conventional meta- heuristics (SA) (see Supplementary Note 6). Interestingly, both classical and quantum optical Ising machines have ex- hibited limitations in their performance related to the graph density9,61. We observe that the PRIS is roughly insensitive to the graph density, when optimizing the noise level φ (see Figure 2(c), shaded green area). A more comprehensive com- parison should take into account the static fabrication error in 4 integrated photonic networks34 (see also Supplementary Note 5), even though careful calibration of their control electronics can significantly reduce its impact on the computation62,63. Influence of the noise and eigenvalue dropout levels For a given Ising problem, there remain two degrees of free- dom in the execution of the PRIS: the noise and eigenvalue dropout levels. The noise level φ determines the level of en- tropy in the Gibbs distribution probed by the PRIS p(E) ∝ exp(−β(E − φS(E))), where S(E) is the Boltzmann en- tropy associated with the energy level E. On the one hand, increasing φ will result in an exponential decay of the prob- ability of finding the ground state p(Hmin, φ). On the other hand, too small a noise level will not satisfy the large noise approximation HL ∼ H2 and result in large autocorrelation times (as the spin state could get stuck in a local minimum of the Hamiltonian). Figure 3(e) demonstrates the existence of an optimal noise level φ, minimizing the number of iterations required to find the ground state of a given Ising problem, for various graph sizes, densities, and eigenvalue dropout levels. This optimal noise value can be approximated upon evalua- tion of the probability of finding the ground state p(Hmin, φ) auto, as the minimum of and the energy autocorrelation time τ E the following heuristic Niter,q ∼ τ E eq (φ) + τ E auto(φ) log(1 − q) log(1 − p(Hmin, φ)) , (8) which approximates the number of iterations required to find the ground state with probability q (see Figure 3(a-e)). In this expression, τ E eq (φ) is the energy equilibrium (or burn-in) time. As can be seen in Figure 3(e), decreasing α (and thus dropping more eigenvalues, with the lowest eigenvalues being dropped out first) will result in a smaller optimal noise level φ. Com- paring the energy landscape for various eigenvalue dropout levels (Figure 3(h)) confirms this statement: as α is reduced, the energy landscape is perturbed. However, for the random spin glass studied in Figure 3(f-g), the ground state remains the same down to α = 0. This hints at a general observa- tion: as lower eigenvalues tend to increase the energy, the Ising ground state will in general be contained in the span of eigenvectors associated with higher eigenvalues (see discus- sion in the Supplementary Note 3). Nonetheless, the global picture is more complex, as the solution of this optimization problem should also enforce the constraint σ ∈ {−1, 1}N . We observe in our simulations that α = 0 yields a higher ground state probability and lower autocorrelation times than α > 0 for all the Ising problems we used in our benchmark. In some sparse models, the optimal value can even be α < 0 (see Fig- ure S3 in the Supplementary Note 4). The eigenvalue dropout is thus a parameter that constrains the dimensionality of the ground state search. The influence of eigenvalue dropout can also be seen from the perspective of the transition matrix. Figure 3(f-g) shows 5 FIG. 3. Influence of noise and eigenvalue dropout levels. (a)-(d): Probability of finding the ground state, and the inverse of the autocorre- lation time as a function of noise level φ for a sample Random Cubic Graph9 (N = 100, (50/100) eigenvalues (a), (99/100) eigenvalues (b) and a sample spin glass (N = 50, (37/100) eigenvalues (c), (26/100) eigenvalues (d)). The arrows indicate the estimated optimal noise level, from Equation (8), taking τ E eq to be constant. For this study we averaged the results of 100 runs of the PRIS with random initial states with j Kij. (e): Niter,90% versus noise level φ for these same graphs and eigenvalue dropout levels. (f)-(g): Eigenvalues of the transition matrix of a sample spin glass (N = 8) at φ = 0.5 ((f)) and φ = 2 ((g)). (h): The corresponding energy is plotted for various eigenvalue dropout levels α, corresponding to less than N eigenvalues kept from the original matrix. The inset is a schematic of the relative position of the global minimum when α = 1 (with (8/8) eigenvalues) with respect error bars representing ±σ from the mean over the 100 runs. We assumed ∆ii = (cid:80) to nearby local minima when α < 1. For this study we assumed ∆ii =(cid:80) j Kij. the eigenvalue distribution of the transition matrix for vari- ous noise and eigenvalue dropout levels. As the PRIS matrix eigenvalues are dropped out, the transition matrix eigenval- ues become more nonuniform, as in the case of large noise (Figure 3(g)). Overall, the eigenvalue dropout can be under- stood as a means of pushing the PRIS to operate in the large noise approximation, without perturbing the Hamiltonian in such a way that would prevent it from finding the ground state. The improved performance of the PRIS with α ∼ 0 hints at the following interpretation: the perturbation of the energy landscape (which affects p(Hmin)) is counterbalanced by the reduction of the energy autocorrelation time induced by the eigenvalue dropout. The existence of these two degrees of freedom suggests a realm of algorithmic techniques to opti- mize the PRIS operation. One could suggest, for instance, setting α ≈ 0, and then performing an inverse simulated an- nealing of the eigenvalue dropout level to increase the dimen- sionality of the ground state search. This class of algorithms could rely on the development of high-speed, low-loss inte- grated modulators59,64 -- 66. Random Cubic Graph N = 100Random Spin Glass N = 500.4103000.10.2000.10.20.50.30.050.11050.60.81.01.21.4Random Spin Glass N = 50Random Spin Glass N = 50Random Cubic Graph N = 100Random Cubic Graph N = 100abcde0255Energy (Little Hamiltonian)Linearized phase space000.51-5-10-15 = 0.5= 2fhgIn!uence of eigenvalue dropoutIn!uence of noise level (n.l.)Transition matrix eigenvalueEigenvalue indexEigenvalue index(6/8)(5/8)(4/8)α = 0(3/8)(8/8)(6/8)(5/8)(4/8)(3/8)Noise level Estimated optimal n.l.p(Hmin)1/τautoEstimated optimal n.l.p(Hmin)1/τautoEstimated optimal n.l.p(Hmin)1/τautoEstimated optimal n.l.p(Hmin)1/τauto(50/100)(50/100)(99/100)(99/100)(37/50)(37/50)(26/50)(26/50)Decreasing eigenvalue dropout level α(8/8)Niter, 90%DISCUSSION Detecting and characterizing critical behaviors with the PRIS 6 The existence of an effective Hamiltonian describing the PRIS dynamics (Equation (6)) further suggests the ability to generate samples of the associated Gibbs distribution at any finite temperature. This is particularly interesting considering the various ways in which noise can be added in integrated photonic circuits by tuning the operating temperature, laser power, photodiode regimes of operation, etc.52,67. This alludes to the possibility of detecting phase transitions and charac- terizing critical exponents of universality classes, leveraging the high speed at which photonic systems can generate uncor- related heuristic samples of the Gibbs distribution associated with Equations (5,6). In this part, we operate the PRIS in the regime where the linear photonic matrix is equal to the Ising coupling matrix (Sqα(D) = D)19. This allows us to speedup the computation on a CPU by leveraging symmetry and spar- sity of the coupling matrix K. We show that the regime of operation described by Equation (59) also probes the expected phase transition (see Supplementary Note 4). 1 − (cid:104)m4(cid:105)/(3(cid:104)m2(cid:105)2), where m = (cid:80)N A standard way of locating the critical temperature of a system is through the use of the Binder cumulant1 U4(L) = i=1 σi/N is the magne- tization and (cid:104).(cid:105) denotes the ensemble average. As shown in Figure 4(a), the Binder cumulants intersect for various graph sizes L2 = N at the critical temperature of TC = 2.241 (com- pared to the theoretical value of 2.269 for the two-dimensional Ferromagnetic Ising model, i.e. within 1.3%). The heuris- tic samples generated by the PRIS can be used to compute physical observables of the modeled system, which exhibit the emblematic order-disorder phase transition of the two- dimensional Ising model1,21 (Figure 4(b)). In addition, criti- cal parameters describing the scaling of the magnetization and susceptibility at the critical temperature can be extracted from the PRIS to within 10% of the theoretical value (see Supple- mentary Note 4). FIG. 4. Detecting and characterizing critical behaviors. a: Binder cumulants U4(L) for various graph sizes L on the 2D Ferromagnetic Ising model. Their intersection determines the critical temperature of the model TC (denoted by a dotted line). b: Magnetization estimated from the PRIS for various L. c: Scaling of the PRIS magnetization autocorrelation time for various Ising models, benchmarked versus the Metropolis-Hastings algorithm (MH). The complexity of a single time step scales like N 2 = L4 for MH on a CPU and like N = L2 for the PRIS on a photonic platform. For readability, error bars in (b) are not shown (see Supplementary Note 4). O(N )34,38,39. Thus, the computational complexity of gen- erating an uncorrelated sample scales like O(N 1+zPRIS/2) for the PRIS on a parallel architecture, while it scales like O(N 2+zMH/2) for a sequential implementation of MH, on a CPU for instance. Implementing the PRIS on a photonic par- allel architecture also ensures that the prefactor in this order of magnitude estimate is small (and only limited by the clock rate of a single recurrent step of this high-speed network). Thus, as long as zPRIS < zMH + 2, the PRIS exhibits a clear advantage over MH implemented on a sequential architecture. Conclusion In Figure 4(c), we benchmark the performance of the the well-known Metropolis-Hastings (MH) PRIS against algorithm1,68,69. In the context of heuristic methods, one should compare the autocorrelation time of a given observ- able. The scaling of the magnetization autocorrelation time auto = O(Lz) = O(N z/2) at the critical temperature is τ m shown in Figure 4(c) for two analytically-solvable models: the two-dimensional ferromagnetic and the infinite-range Ising models. Both algorithms yield autocorrelation time critical exponents close to the theoretical value (z ∼ 2.1)1 for the two-dimensional Ising model. However, the PRIS seems to perform better on denser models such as the infinite-range Ising model, where it yields a smaller autocorrelation time critical exponent. More significantly, the advantage of the PRIS resides in its possible implementation with any matrix- to-vector accelerator, such as parallel photonic networks, so that the computational (time) complexity of a single step is To conclude, we presented in this Letter the PRIS, a photonic-based heuristic algorithm able to probe arbitrary Ising Gibbs distributions at various temperature levels. At low temperatures, the PRIS can find ground states of arbi- trary Ising models with high probability. Our approach es- sentially relies on the use of matrix-to-vector product ac- celerators, such as photonic networks34,67, free-space optical processors28, FPGAs70, and ASICs71 (see comparison of time estimates in the Supplementary Note 5). We also perform a proof-of-concept experiment on a Xilinx Zynq UltraScale+ multiprocessor system-on-chip (MPSoC) ZCU104, an elec- tronic board containing a parallel programmable logic unit (FPGA Field Programmable Gate Arrays). We run the PRIS on large random spin glasses N = 100 and achieve algo- rithm time steps of 63 ns. This brings us closer to photonic clocks (cid:46) 1 ns, thus demonstrating that (1) the PRIS can lever- age parallel architectures of various natures, electronics and acbL = 10L = 20L = 40L = 50TemperatureMagnetizationL = 50L = 1002D FerromagneticMH, z = 2.068 PRIS, z = 0.886MH, z = 0.924LL2D FerromagneticLLIn!nite RangeL22246810345PRIS, z = 2.023 2.22.32.400.50.20.40.60.52.232.250.6log Llog τmU4 (L)autoTC2.24photonics; (2) the potential of hybrid parallel opto-electronic implementations. Details of the FPGA implementation and numerical experiments are given in Supplementary Note 7. Moreover, our system requires some amount of noise to perform better, which is an unusual behavior only observed in very few physical systems. For instance, neuroscientists have conjectured that this could be a feature of the brain and spiking neural networks72,73. The PRIS also performs a static transformation (and the state evolves to find the ground state). This kind of computation can rely on a fundamental property of photonics -- passivity -- and thus reach even higher ef- ficiencies. Non-volatile phase-change materials integrated in silicon photonic networks could be leveraged to implement the PRIS with minimal energy costs74. We also suggested a broader family of photonic metaheuris- tic algorithms which could achieve even better performance on larger graphs (see Supplementary Note 6). For instance, one could simulate annealing with photonics by reducing the system noise level (this could be achieved by leveraging quan- tum photodetection noise67, see discussion in Supplementary Notes 5 and 6). We believe that this class of algorithms that can be implemented on photonic networks is broader than the metaheuristics derived from MH, since one could also simu- late annealing on the eigenvalue dropout level α. The ability of the PRIS to detect phase transitions and probe critical exponents is particularly promising for the study of universality classes, as numerical simulations suffer from crit- ical slowing down: the autocorrelation time grows exponen- tially at the critical point, thus making most samples too corre- lated to yield accurate estimates of physical observables. Our study suggests that this fundamental issue could be bypassed with the PRIS, which can generate a very large number of samples per unit time -- only limited by the bandwidth of ac- tive silicon photonics components. The experimental realization of the PRIS on a photonic platform would require additional work compared to the demonstration of deep learning with nanophotonic circuits34. The noise level can be dynamically induced by several well- known sources of noise in photonic and electronic systems52. However, attaining a low enough noise due to heterogeneities in a static architecture, and characterizing the noise level are two experimental challenges. Moreover, the PRIS requires an additional homodyne detection unit, in order to detect both the amplitude and the phase of the output signal from the linear photonic domain. Nonetheless, these experimental challenges do not impact the promising scaling properties of the PRIS, since various photonic architectures have recently been proposed34,40,45,67,75, giving a new momentum to pho- tonic computing. 7 METHODS Numerical simulations to evaluate performance on finding ground states of Ising models To evaluate the performance of the algorithm on several Ising problems, we simulate the execution of an ideal pho- tonic system, performing computations without static error. The noise is artificially added after the matrix multiplication unit and follows a Gaussian distribution, as discussed above. This results in an algorithm similar to the one described in the section II of this work. In the main text, we present the scaling performance of the PRIS as a function of the graph order. For each graph order and density, we generate 10 random samples with these prop- erties. We then optimize the noise level (minimizing Niter,99%) on a random sample graph and generate a total of 10 samples for each pair of graph order/density. The optimal value of φ is shown in Figure S2 in Supplementary Note 4. For each randomly generated graph, we first compute its ground state with the online platform BiqMac57. We then make 100 measurements of the number of steps required (with a random initial state) to get to this ground state. From these 1000 runs, we define the estimate of finding the ground state of the problem with q percent probability Niter,q% as the q-th quantile. Also in the main text, we study the influence of eigenvalue dropout and of the noise level on the PRIS performance. We show that the optimal level of eigenvalue dropout is usually α < 1, and around α = 0. In some cases, it can even be α < 0 as we show in Figure S3 in Supplementary Note 4 where the optimal (α, φ) = (−0.15, 0.55) for a random cu- bic graph with N = 52. In addition to Figure 3(f-h) from the main text showing the influence of eigenvalue dropout on a random spin glass, the influence of dropout on a random cubic graph is shown in Figure S4 in Supplementary Note 4. Similar observations can be made, but random cubic graphs, which show highly degenerated hamiltonian landscapes, are more robust to eigenvalue dropout. Even with α = −0.8, in the case shown in Figure S4 in Supplementary Note 4 the ground state remains unaffected. Others Further details on generalization of the theory of the PRIS dynamics, construction of the weight matrix J, numerical simulations, scaling performance of the PRIS, and compari- son of the PRIS to other (meta)heuristics algorithms can be found in the Supplementary Information. ACKNOWLEDGEMENTS The authors would like to acknowledge Aram Harrow, Mehran Kardar, Ido Kaminer, Miriam Farber, Theodor Misi- akiewicz, Manan Raval, Nicholas Rivera, Nicolas Romeo, Jamison Sloan, Can Knaut, Joe Steinmeyer, and Gim P. Hom for helpful discussions. The authors would also like to thank Angelika Wiegele (Alpen-Adria-Universitt Klagenfurt) for providing solutions of the Ising models considered in this work with N ≥ 50 (computed with BiqMac57). This work was supported in part by the Semiconductor Research Corpo- ration (SRC) under SRC contract #2016-EP-2693-B (Energy Efficient Computing with Chip-Based Photonics MIT). This work was supported in part by the National Science Founda- tion (NSF) with NSF Award #CCF-1640012 (E2DCA: Type I: Collaborative Research: Energy Efficient Computing with Chip-Based Photonics). This material is based upon work supported in part by the U. S. Army Research Laboratory and the U.S. Army Research Office through the Institute for Soldier Nanotechnologies, under contract number W911NF- 18-2-0048. C. Z. was financially supported by the Whiteman Fellowship. M. P. was financially supported by NSF Graduate Research Fellowship grant number 1122374. AUTHOR CONTRIBUTIONS C. R.-.C., Y. S. and M.S. conceived the project. C. R.-C. and Y. S. developed the analytical models and numerical cal- culations, with contributions from C. Z., M. P., L. J. and T. D.; C. R.-C. and C. Z. performed the benchmarking of the PRIS on analytically solvable Ising models and large spin glasses. C. R.-C. and F.A. developed the analytics for various noise distributions. C. M., M. R. J., and C. R.-C. implemented the PRIS on FPGA. Y. S., J. D. J., D. E. and M.S. supervised the project. C. R.-C. wrote the manuscript with input from all authors. ADDITIONAL INFORMATION Correspondence and requests for materials should be ad- dressed to C. R.-C. COMPETING FINANCIAL INTERESTS The authors declare the following patent application: U.S. Patent Application No.: 16/032,737. ∗ Corresponding [email protected] authors e-mail: [email protected], yc- [1] D. P. Landau and K. Binder, A Guide to Monte Carlo Simula- tions in Statistical Physics. 2009. [2] J. Hromkovic, Algorithmics for Hard Problems: Introduction to Combinatorial Optimization, Randomization, Approximation, and Heuristics. Springer Berlin Heidelberg, 2013. 8 [3] M. Kardar, G. Parisi, and Y.-C. Zhang, "Dynamic Scaling of Growing Interfaces," Physical Review Letters, vol. 56, pp. 889 -- 892, mar 1986. [4] M. B. Isichenko, "Percolation, statistical topography, and trans- port in random media," Reviews of Modern Physics, vol. 64, pp. 961 -- 1043, oct 1992. [5] A. R. Honerkamp-Smith, S. L. Veatch, and S. L. Keller, "An introduction to critical points for biophysicists; observations of compositional heterogeneity in lipid membranes," Biochimica et Biophysica Acta (BBA) - Biomembranes, vol. 1788, pp. 53 -- 63, jan 2009. [6] R. Albert and A.-L. Barab´asi, "Statistical mechanics of com- plex networks," Reviews of Modern Physics, vol. 74, pp. 47 -- 97, jan 2002. [7] F. Glover and G. Kochenberger, Handbook of Metaheuristics. Springer, 2006. [8] Z. Wang, A. Marandi, K. Wen, R. L. Byer, and Y. Yamamoto, "Coherent Ising machine based on degenerate optical paramet- ric oscillators," Physical Review A, vol. 88, p. 063853, dec 2013. [9] P. L. McMahon, A. Marandi, Y. Haribara, R. Hamerly, C. Lan- grock, S. Tamate, T. Inagaki, H. Takesue, S. Utsunomiya, K. Ai- hara, R. L. Byer, M. M. Fejer, H. Mabuchi, and Y. Yamamoto, "A fully programmable 100-spin coherent Ising machine with all-to-all connections.," Science (New York, N.Y.), vol. 354, pp. 614 -- 617, nov 2016. [10] K. Wu, J. Garc´ıa de Abajo, C. Soci, P. Ping Shum, and N. I. Zhe- ludev, "An optical fiber network oracle for NP-complete prob- lems," Light: Science & Applications, vol. 3, pp. e147 -- e147, feb 2014. [11] M. R. V´azquez, V. Bharadwaj, B. Sotillo, S.-Z. A. Lo, R. Ram- poni, N. I. Zheludev, G. Lanzani, S. M. Eaton, and C. Soci, "Optical NP problem solver on laser-written waveguide plat- form," Optics Express, vol. 26, p. 702, jan 2018. [12] W. M. Macready, A. G. Siapas, and S. A. Kauffman, "Criti- cality and Parallelism in Combinatorial Optimization," Science, vol. 271, pp. 56 -- 59, jan 1996. [13] Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," Nature, vol. 521, pp. 436 -- 444, may 2015. [14] W. A. Little, "The existence of persistent states in the brain," Mathematical Biosciences, vol. 19, no. 1-2, pp. 101 -- 120, 1974. [15] J. J. Hopfield, "Neural networks and physical systems with emergent collective computational abilities.," Proceedings of the National Academy of Sciences of the United States of Amer- ica, vol. 79, pp. 2554 -- 8, apr 1982. [16] J. J. Hopfield and D. W. Tank, "Neural computation of decisions in optimization problems," Biological Cybernetics, vol. 52, no. 3, pp. 141 -- 152. [17] E. Ising, "Beitrag zur Theorie des Ferromagnetismus," Z. Phys., [18] M. M´ezard and A. Montanari, Information, Physics, and Com- 1925. putation. 2009. [19] D. J. Amit, H. Gutfreund, and H. Sompolinsky, "Spin-glass models of neural networks," Physical Review A, vol. 32, pp. 1007 -- 1018, aug 1985. [20] A. Pelissetto and E. Vicari, "Critical phenomena and renormalization-group theory," Physics Reports, vol. 368, pp. 549 -- 727, oct 2002. [21] L. Onsager, "Crystal Statistics. I. A Two-Dimensional Model with an Order-Disorder Transition," Physical Review, vol. 65, pp. 117 -- 149, feb 1944. [22] N. V. Brilliantov, "Effective magnetic Hamiltonian and Ginzburg criterion for fluids," Physical Review E, vol. 58, pp. 2628 -- 2631, aug 1998. [23] D. J. Amit, Modeling brain function : the world of attractor vol. 32, no. 21, pp. 3427 -- 3439, 2014. neural networks. Cambridge University Press, 1989. [24] N. Ghofraniha, I. Viola, F. Di Maria, G. Barbarella, G. Gigli, L. Leuzzi, and C. Conti, "Experimental evidence of replica symmetry breaking in random lasers," Nature Communications, vol. 6, p. 6058, dec 2015. [25] M. A. Halasz, A. D. Jackson, R. E. Shrock, M. A. Stephanov, and J. J. M. Verbaarschot, "Phase diagram of QCD," Physical Review D, vol. 58, p. 096007, sep 1998. [26] F. Barahona, "On the computational complexity of Ising spin glass models," Journal of Physics A: Mathematical and Gen- eral, vol. 15, pp. 3241 -- 3253, oct 1982. [27] J. Bruck and J. W. Goodman, "On the power of neural networks for solving hard problems," Journal of Complexity, vol. 6, pp. 129 -- 135, jun 1990. [28] N. H. Farhat, D. Psaltis, A. Prata, and E. Paek, "Optical im- plementation of the Hopfield model," Applied Optics, vol. 24, p. 1469, may 1985. [29] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, "Optimization by simulated annealing.," Science (New York, N.Y.), vol. 220, pp. 671 -- 80, may 1983. [30] D. J. Earl and M. W. Deem, "Parallel tempering: Theory, appli- cations, and new perspectives," Physical Chemistry Chemical Physics, vol. 7, p. 3910, nov 2005. [31] L. D. Davis and M. Mitchell, Handbook of Genetic Algorithms. 1991. [32] F. Glover and M. Laguna, "Tabu Search," in Handbook of Com- binatorial Optimization, pp. 2093 -- 2229, Boston, MA: Springer US, 1998. [33] E. Boros, P. L. Hammer, and G. Tavares, "Local search heuristics for Quadratic Unconstrained Binary Optimization (QUBO)," Journal of Heuristics, vol. 13, pp. 99 -- 132, feb 2007. [34] Y. Shen, N. C. Harris, S. Skirlo, M. Prabhu, T. Baehr-Jones, M. Hochberg, X. Sun, S. Zhao, H. Larochelle, D. Englund, and M. Soljaci´c, "Deep learning with coherent nanophotonic cir- cuits," Nature Photonics, vol. 11, pp. 441 -- 446, jun 2017. [35] A. Silva, F. Monticone, G. Castaldi, V. Galdi, A. Al`u, and N. Engheta, "Performing mathematical operations with meta- materials," Science, vol. 343, no. 6167, pp. 160 -- 163, 2014. [36] A. F. Koenderink, A. Al`u, and A. Polman, "Nanophotonics: Shrinking light-based technology," may 2015. [37] J. Carolan, C. Harrold, C. Sparrow, E. Mart´ın-L´opez, N. J. Rus- sell, J. W. Silverstone, P. J. Shadbolt, N. Matsuda, M. Oguma, M. Itoh, G. D. Marshall, M. G. Thompson, J. C. Matthews, T. Hashimoto, J. L. O'Brien, and A. Laing, "Universal linear optics," Science, vol. 349, pp. 711 -- 716, aug 2015. [38] M. Reck, A. Zeilinger, H. J. Bernstein, and P. Bertani, "Exper- imental realization of any discrete unitary operator," Physical Review Letters, vol. 73, pp. 58 -- 61, jul 1994. [39] W. R. Clements, P. C. Humphreys, B. J. Metcalf, W. S. Kolthammer, and I. A. Walsmley, "Optimal design for universal multiport interferometers," Optica, vol. 3, p. 1460, dec 2016. [40] X. Lin, Y. Rivenson, N. T. Yardimci, M. Veli, M. Jarrahi, and A. Ozcan, "All-Optical Machine Learning Using Diffractive Deep Neural Networks," apr 2018. [41] M. Gruber, J. Jahns, and S. Sinzinger, "Planar-integrated optical vector-matrix multiplier," Applied Optics, vol. 39, p. 5367, oct 2000. [42] A. N. Tait, M. A. Nahmias, Y. Tian, B. J. Shastri, and P. R. Prucnal, "Photonic Neuromorphic Signal Processing andCom- puting," pp. 183 -- 222, Springer, Berlin, Heidelberg, 2014. [43] A. N. Tait, M. A. Nahmias, B. J. Shastri, and P. R. Prucnal, "Broadcast and weight: An integrated network for scalable photonic spike processing," Journal of Lightwave Technology, 9 [44] K. Vandoorne, P. Mechet, T. Van Vaerenbergh, M. Fiers, G. Morthier, D. Verstraeten, B. Schrauwen, J. Dambre, and P. Bienstman, "Experimental demonstration of reservoir com- puting on a silicon photonics chip," Nature Communications, vol. 5, p. 3541, dec 2014. [45] A. Saade, F. Caltagirone, I. Carron, L. Daudet, A. Dremeau, S. Gigan, and F. Krzakala, "Random projections through multi- ple optical scattering: Approximating Kernels at the speed of light," in 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 6215 -- 6219, IEEE, mar 2016. [46] D. Pierangeli, V. Palmieri, G. Marcucci, C. Moriconi, G. Perini, M. De Spirito, M. Papi, and C. Conti, "Deep optical neural net- work by living tumour brain cells," dec 2018. [47] Z. Cheng, H. K. Tsang, X. Wang, K. Xu, and J.-B. Xu, "In-Plane Optical Absorption and Free Carrier Absorption in Graphene-on-Silicon Waveguides," IEEE Journal of Selected Topics in Quantum Electronics, vol. 20, pp. 43 -- 48, jan 2014. [48] Q. Bao, H. Zhang, Z. Ni, Y. Wang, L. Polavarapu, Z. Shen, Q.-H. Xu, D. Tang, and K. P. Loh, "Monolayer graphene as a saturable absorber in a mode-locked laser," Nano Research, vol. 4, pp. 297 -- 307, mar 2011. [49] A. C. Selden, "Pulse transmission through a saturable ab- sorber," British Journal of Applied Physics, vol. 18, pp. 743 -- 748, jun 1967. [50] M. Soljaci´c, M. Ibanescu, S. G. Johnson, Y. Fink, and J. D. Joannopoulos, "Optimal bistable switching in nonlinear pho- tonic crystals," Physical Review E, vol. 66, p. 055601, nov 2002. [51] R. W. Schirmer and A. L. Gaeta, "Nonlinear mirror based on two-photon absorption," Journal of the Optical Society of Amer- ica B, vol. 14, p. 2865, nov 1997. [52] P. Horowitz and H. Winfield, "The Art of Electronics," Ameri- can Journal of Physics, 1990. [53] B. Boser, E. Sackinger, J. Bromley, Y. Le Cun, and L. Jackel, "An analog neural network processor with programmable topology," IEEE Journal of Solid-State Circuits, vol. 26, no. 12, pp. 2017 -- 2025, 1991. [54] J. Misra and I. Saha, "Artificial neural networks in hardware: A survey of two decades of progress," Neurocomputing, vol. 74, pp. 239 -- 255, dec 2010. [55] D. Vrtaric, V. Ceperic, and A. Baric, "Area-efficient differen- tial Gaussian circuit for dedicated hardware implementations of Gaussian function based machine learning algorithms," Neu- rocomputing, vol. 118, pp. 329 -- 333, oct 2013. [56] I. A. D. Williamson, T. W. Hughes, M. Minkov, B. Bartlett, S. Pai, and S. Fan, "Reprogrammable Electro-Optic Nonlin- ear Activation Functions for Optical Neural Networks," arXiv preprints, arxiv.org:1903.04579, mar 2019. [57] F. Rendl, G. Rinaldi, and A. Wiegele, "Solving Max-Cut to optimality by intersecting semidefinite and polyhedral relax- ations," Mathematical Programming, vol. 121, pp. 307 -- 335, feb 2010. [58] P. Peretto, "Collective properties of neural networks: A statisti- cal physics approach," Biological Cybernetics, vol. 50, pp. 51 -- 62, feb 1984. [59] M. Lipson, "Guiding, Modulating, and Emitting Light on Silicon-Challenges and Opportunities," Journal of Lightwave Technology, Vol. 23, Issue 12, pp. 4222-, vol. 23, p. 4222, dec 2005. [60] N. C. Harris, Y. Ma, J. Mower, T. Baehr-Jones, D. Englund, M. Hochberg, and C. Galland, "Efficient, compact and low loss thermo-optic phase shifter in silicon," Optics Express, vol. 22, [61] R. Hamerly, T. Inagaki, P. L. McMahon, D. Venturelli, A. Marandi, T. Onodera, E. Ng, C. Langrock, K. Inaba, T. Honjo, K. Enbutsu, T. Umeki, R. Kasahara, S. Utsunomiya, S. Kako, K. I. Kawarabayashi, R. L. Byer, M. M. Fejer, H. Mabuchi, D. Englund, E. Rieffel, H. Takesue, and Y. Ya- mamoto, "Experimental investigation of performance differ- ences between coherent Ising machines and a quantum an- nealer," Science Advances, 2019. [62] D. A. B. Miller, "Perfect optics with imperfect components," Optica, vol. 2, p. 747, aug 2015. p. 10487, may 2014. [63] R. Burgwal, W. R. Clements, D. H. Smith, J. C. Gates, W. S. Kolthammer, J. J. Renema, and I. A. Walmsley, "Using an im- perfect photonic network to implement random unitaries," Op- tics Express, vol. 25, p. 28236, nov 2017. [64] V. R. Almeida, C. A. Barrios, R. R. Panepucci, and M. Lipson, "All-optical control of light on a silicon chip," Nature, vol. 431, pp. 1081 -- 1084, oct 2004. [65] C. T. Phare, Y. H. Daniel Lee, J. Cardenas, and M. Lipson, "Graphene electro-optic modulator with 30 GHz bandwidth," Nature Photonics, 2015. [66] C. Haffner, D. Chelladurai, Y. Fedoryshyn, A. Josten, B. Baeuerle, W. Heni, T. Watanabe, T. Cui, B. Cheng, S. Saha, D. L. Elder, L. R. Dalton, A. Boltasseva, V. M. Shalaev, N. Kin- sey, and J. Leuthold, "Low-loss plasmon-assisted electro-optic modulator," Nature, 2018. [67] R. Hamerly, L. Bernstein, A. Sludds, M. Soljaci´c, and D. En- glund, "Large-Scale Optical Neural Networks Based on Photo- electric Multiplication," Physical Review X, vol. 9, nov 2019. [68] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, "Equation of state calculations by fast computing machines," The Journal of Chemical Physics, 1953. [69] W. K. Hastings, "Monte carlo sampling methods using Markov chains and their applications," Biometrika, 1970. [70] J. Dean, D. Patterson, and C. Young, "A New Golden Age in Computer Architecture: Empowering the Machine-Learning Revolution," IEEE Micro, vol. 38, pp. 21 -- 29, mar 2018. [71] Y. Dou, S. Vassiliadis, G. K. Kuzmanov, and G. N. Gaydadjiev, "64-bit floating-point FPGA matrix multiplication," in Pro- ceedings of the 2005 ACM/SIGDA 13th international sympo- sium on Field-programmable gate arrays - FPGA '05, (New 10 York, New York, USA), p. 86, ACM Press, 2005. [72] D. C. Knill and A. Pouget, "The Bayesian brain: the role of uncertainty in neural coding and computation," Trends in Neu- rosciences, vol. 27, pp. 712 -- 719, dec 2004. [73] W. Maass, "Noise as a resource for computation and learn- ing in networks of spiking neurons," Proceedings of the IEEE, vol. 102, pp. 860 -- 880, may 2014. [74] Q. Wang, E. T. F. Rogers, B. Gholipour, C.-M. Wang, G. Yuan, J. Teng, and N. I. Zheludev, "Optically reconfigurable metasur- faces and photonic devices based on phase change materials," Nature Photonics, vol. 10, pp. 60 -- 65, jan 2016. [75] A. N. Tait, T. F. de Lima, E. Zhou, A. X. Wu, M. A. Nahmias, B. J. Shastri, and P. R. Prucnal, "Neuromorphic photonic net- works using silicon photonic weight banks," Scientific Reports, vol. 7, p. 7430, dec 2017. [76] R. M. Karp, "Reducibility among Combinatorial Problems," in Complexity of Computer Computations, pp. 85 -- 103, Boston, MA: Springer US, 1972. [77] C.-K. Looi, "Neural network methods in combinatorial op- timization," Computers & Operations Research, vol. 19, pp. 191 -- 208, apr 1992. [78] A. Azzalini and A. D. VALLE, "The multivariate skew-normal distribution," Biometrika, vol. 83, pp. 715 -- 726, dec 1996. [79] S. Friedli and Y. Velenik, Statistical Mechanics of Lattice Sys- tems. Cambridge University Press, nov 2017. [80] N. C. Harris, G. R. Steinbrecher, M. Prabhu, Y. Lahini, J. Mower, D. Bunandar, C. Chen, F. N. C. Wong, T. Baehr- Jones, M. Hochberg, S. Lloyd, and D. Englund, "Quantum transport simulations in a programmable nanophotonic proces- sor," Nature Photonics, vol. 11, pp. 447 -- 452, jun 2017. [81] D. Pierangeli, G. Marcucci, and C. Conti, "Large-Scale Pho- tonic Ising Machine by Spatial Light Modulation," Physical Re- view Letters, 2019. [82] Y. Nourani and B. Andresen, "A comparison of simulated an- nealing cooling strategies," Journal of Physics A: Mathematical and General, vol. 31, pp. 8373 -- 8385, oct 1998. [83] H. Cohn and M. Fielding, "Simulated Annealing: Searching for an Optimal Temperature Schedule," SIAM Journal on Opti- mization, vol. 9, pp. 779 -- 802, jan 1999. HEURISTIC RECURRENT ALGORITHMS FOR PHOTONIC ISING MACHINES SUPPLEMENTARY INFORMATION 11 CONTENTS Introduction Results General theory of the PRIS dynamics Finding the ground state of Ising models with the PRIS Influence of the noise and eigenvalue dropout levels Discussion Detecting and characterizing critical behaviors with the PRIS Conclusion Methods Numerical simulations to evaluate performance on finding ground states of Ising models Others Acknowledgements Author contributions Additional information Competing financial interests References Heuristic Recurrent Algorithms for Photonic Ising Machines SUPPLEMENTARY INFORMATION Supplementary Note 1: Theoretical preliminaries Definition of an Ising problem Mapping to the MAX-CUT problem Supplementary Note 2: General theory of Photonic Recurrent Ising Sampler Optical Parallel Recurrent MCMC algorithm Determination of the Temperature factor Transition probability Expansion of the transition probability in terms of Detailed balance condition Effective Gibbs distribution Small noise approximation Large noise approximation Discussion on h0 i Inequalities between Hamiltonians Suggested algorithms Supplementary Note 3: Construction of the weight matrix Discussion on the diagonal offset Tuning the search dimensionality with eigenvalue dropout Modified algorithm with tunable offset Supplementary Note 4: Numerical simulations Evaluating sampling performance and critical parameters on two-dimensional ferromagnetic and infinite-range Ising models Supplementary Note 5: Scaling of the PRIS performance on several photonic architectures Cascaded arrays of programmable Mach-Zehnder interferometers (MZI) 12 1 2 2 4 4 6 6 6 7 7 7 7 8 8 8 8 11 14 14 14 15 15 15 16 17 18 18 18 18 19 20 20 22 22 23 24 25 25 31 31 Optical Neural Networks based on Photoelectric Multiplication Free space optical Neural Networks Comparison table of various heuristic algorithms and architectures Supplementary Note 6: Comparison of the PRIS to several (meta)heuristics Benchmarking versus Metropolis-Hastings on large spin glasses N ∼ 1000 PRIS-A: a proposed metaheuristic variation of the PRIS Supplementary Note 7: Implementation of the PRIS on FPGA Architecture Complexity Experiment References 13 32 32 32 34 34 34 37 37 39 39 42 In this work, we first provide a comprehensive theoretical analysis of the dynamics of the Photonic Recurrent Ising Sampler (PRIS), with an emphasis on how the weight matrix in the algorithm should be chosen. We then detail the parameters of our numerical simulations carried to estimate the performance of the PRIS on finding the ground state of arbitrary Ising problems and to sample critical behaviors. 14 SUPPLEMENTARY NOTE 1: THEORETICAL PRELIMINARIES Definition of an Ising problem The general definition of an Ising problem is the following: given a matrix (Kij)(i,j)∈I 2, and a vector {bi}i∈I where I is a finite set with cardinality I = N, we want to find the spin distribution {σi}i∈I ∈ {−1, 1}N minimizing the following Hamiltonian function: HK,b({σi}) = − (cid:88) (cid:88) 1≤i<j≤N = − 1 2 1≤i,j≤N Kijσiσj −(cid:88) Kijσiσj −(cid:88) i biσi biσi. i (9) (10) In statistical mechanics, this model represents the energy of an interacting set of spins. The coupling term Kij represents the coupling between spins i and j. In the general definition of the Ising model, the coupling matrix K is assumed to be symmetric. We could also assume that it has a null diagonal, as it would only add a constant offset to the Hamiltonian. We are not making this extra assumption, as it will prove later to be useful. An external magnetic field bi can be applied, which breaks the symmetry of the problem. This general class of Ising problems is NP-hard76 and many subclasses of this problem exhibit a similar complexity. In this work, we will characterize our optical algorithms on two subclasses of problems: (cid:46) General antiferromagnets, for which the coupling K can only take two discrete values Kij ∈ {0,−1}, and bi = 0 for all i. As every problem in this subclass is equivalent to an unweighted MAX-CUT problem, and that MAX-CUT is known to be NP-hard76, this subclass is already NP-hard. (cid:46) Spin glasses, for which the coupling K can take on continuous values in the range [−1, 1], and bi = 0. However, our approach can be easily extended to any coupling matrix K. The possible extension to non-zero external magnetic fields will also be discussed. A complete study of the hardness of the various subclasses of Ising problems can be found in Ref.26. Mapping to the MAX-CUT problem The MAX-CUT problem of a weighted undirected graph can be phrased in terms of its adjacency matrix A: (cid:88) ij Find argmaxσ 1 4 Aij(1 − σiσj) := argmaxσCA(σ), (11) where σ ∈ {−1, 1}N is a spin vector. The value of this MAX-CUT problem is called Cmax. More intuitively, this problem can be interpreted as finding a subset of vertices of the graph, such that the number of edges connecting this subset to its complementary is maximized. The vertices of this subset (resp. of its complementary set) will have spin up (resp. spin down). We can map the general Ising problem (spin glass) (Equation (10)) to any weighted MAX-CUT problem. This mapping is useful because the publicly-available solver we use to find the Ising ground state works in terms of the MAX-CUT problem57. The energy of the Ising ground state and the MAX-CUT are related as follows: Kij − 2Cmax, (cid:88) (12) Hmin = − 1 2 ij where K = −A. From this linear relation, we also deduce that the solution of the weighted MAX-CUT problem and of the general Ising model (spin glass) are the same: argmaxσCA(σ) = argminσH (K)(σ). (13) 15 SUPPLEMENTARY NOTE 2: GENERAL THEORY OF PHOTONIC RECURRENT ISING SAMPLER Optical Parallel Recurrent MCMC algorithm The Little and Hopfield networks14,15 are two early forms of recurrent neural networks, originally designed to understand and model associative memory processes. In their general acceptance, the Little network can be understood as a synchronous version of the Hopfield network. Using the latter to solve the NP-hard Ising problem was previously investigated16,77. The behavior of both networks was later studied from a statistical mechanical perspective by P. Peretto58 and Daniel J. Amit and colleagues19. We here generalize their study to the case of noisy, synchronous Hopfield network, or noisy Little network58 and investigate the possibility of sampling hard Ising problems with a parallel, recurrent optical system. In the following, we equivalently use the term "neuron" and "spin" in the framework of spin glass models of neural networks. Our machine is applying the following algorithm: (cid:46) Initialize assembly of neurons {σi} with random values. The signal is coded into optical domain with the reduced spins Si = (σi + 1)/2 ∈ {0, 1}. (cid:46) At each step of the algorithm, each neuron is applied a potential vi (random variable) whose expected value is given by a matrix multiplication with C = 2J: (cid:88) j ¯vi = CijSj = N(cid:88) j=1 Jijσj + N(cid:88) j=1 Cij/2 and whose probability density is given by the function fφ with mean ¯vi and standard deviation φ: (14) (15) (16) (17) (cid:90) (cid:40) (cid:90) fφ(x) x dx = ¯vi, fφ(x)(x − ¯vi)2 dx = φ2. Thθi(Si) = if vi < θi, 0, 1, otherwise. (cid:46) A non-linear threshold Thθ is then applied to ¯vi to yield the neural state at the next time step: To summarize, the sequential transformation of the neuron state is: (18) (19) where N (xφ) is a gaussian distribution with mean x and standard deviation φ. The algorithm is fully determined by the following set of transformation variables: (C, fϕ, θ). S(t+1) = Thθ(X (t)), X (t) ∼ N (CStφ), Determination of the Temperature factor In this section, we compute the stationary distribution of the spin variable St for the general case with variables (C, fϕ, θ). We first determine the probability of a single spin-flip, knowing that the current spin state is J: W (σi(I)J) = G(hi(J)σi(I)), hi(J) = − N(cid:88) (cid:88) j=1 h0 i = j Jijσj(J) − h0 i , Cij/2 − θi, (20) (21) (22) Noise distribution Logistic Gaussian Cauchy∗ Laplace Uniform 1 1 1 3φ 2φ 1+exp (cid:1) G(x) (cid:0) πx√ 2 (1 − erf( x√ (cid:40) 1 )) φ ) + 1 π arctan( x 2 exp(− √ φ ) x ≤ 0  0 2x √ 1 − 1 φ ) x ≥ 0 2 exp( √ √ √ x+ x ≤ −√ x ∈ [−√ x ≥ √ 3φ 3φ, 3φ 1 3φ 3φ 2 2x 2 3φ[ 16 k 2 √ 3 π 0.5877 1.16 0.4735 0 ≡ maxX∈R γopt (X) 0 0.0095 0.0495 0.0199 0.6136 0.0561 TABLE I. Summary of temperature factors for various noise distributions. Each noise distribution is defined so that its standard deviation, when applicable, is equal to φ. (∗the standard deviation of the Cauchy distribution is not defined, the linear dependence is measured as a function of the scaling parameter φ in this particular case.) where we define G as a rescaled version of the noise cumulative density function: (cid:90) x −∞ F (x) = fφ(u) du, F0(x) = F (x + ¯vi), G(x) = 1 − F0(x). (23) (24) (25) In the following, we only assume that fφ is symmetric and has a standard deviation φ. The symmetry assumption is not constraining as most noise distributions found in nature have a symmetric (and, in many cases, gaussian) distribution52. Our analysis can also be extended to noise distributions whose variance is infinite by parameterizing the distribution with another parameter, usually referred to as a "scaling" parameter (see Table I). The case of fφ being a gaussian distribution is discussed in Ref.58, where the function G can be approximated by a sigmoid function when rescaling it by the adequate factor. In the general case, in order to minimize this approximation error, we choose the following rescaling factor Gγ(x) = G(γx), 1 − s(x) = 1 1 + ex , γopt = argminγ(cid:48) max where s(x) = 1/(1 + e−x) is the sigmoid function. numerically observe that this dependence is linear X∈R Gγ(cid:48)(X) − (1 − s(X)) = argminγ(cid:48) max X∈R γ(X), Naturally, the optimal scaling factor will depend on the standard deviation φ of the original probability distribution f. We Optimal scaling factors and corresponding errors are reported in Table I. Identifying the transition probability to a sigmoid function is a necessary step to compute the effective Hamiltonian of the spin distribution58. γopt = k 2 φ. (26) Transition probability By choosing adequately the temperature factor k 2 , we minimize the error when approximating the transition probability by a sigmoid function (in the following, hi = hi(J) and σi = σi(I), for readability): W (σiJ) = Gγopt (cid:19) (cid:18) hiσi (cid:19) (cid:18) hiσi (cid:18) hiσi (cid:19) γopt γopt . γopt + γopt = s ≈ s (cid:18) hiσi (cid:19) γopt (27) (28) (29) W (IJ) = = (cid:89) (cid:89) i i (cid:18) s W (σiJ) (cid:19) (cid:18) hiσi N(cid:88) kφ/2 k=1 = W 0(IJ) + W k(IJ), (cid:18) hiσi (cid:19)(cid:19) kφ/2 + (cid:80) H 0(IJ) = −(cid:88) e−βH 0(IJ) K e−βH 0(KJ) Jijσi(I)σj(J) −(cid:88) , ij 1 kφ . β = i h0 i σi(I), W 0(IJ) = e−βH 0(IJ) i 2 cosh(βhi(J)) . Another way to write this zero-th order term58 will prove to be useful later in our derivations: The general expression for the k−th order term can be derived: W k(IJ) = W 0(IJ) ... (cid:89) (cid:88) j1 The N−th order term scales, in the worst case scenario, like N 0 : jk(cid:54)∈{j1, ..., jk−1} l (2βhjl σjl ) (1 + exp (2βhjlσjl )) . W N (IJ) = (2βhjl σjl ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ≤ N 0 . (cid:81) (cid:88) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:89) k (30) (31) (32) (34) (35) (36) (37) (38) In the following, we will drop the subscript in γopt and will assume that its value is known (it can be determined numerically, given the noise distribution). The error function can be used as an expansion parameter, in order to estimate the accuracy of the approximation performed when neglecting this term (as is done in Ref.58). 17 Expansion of the transition probability in terms of The transition probability can be derived by multiplying the probabilities of single spin-flips, these events being independent: where W k(IJ) is the k-th order term in the expansion, assuming the error function ε is small (ε (cid:28) s). The zero-th order term of this expansion gives us back the result from58: W 0(IJ) = (33) For k < N, analyzing the scaling of the k−th order term requires more assumptions. Let's look at the case k = 1, to verify that we can safely neglect higher-order terms in this expansion: (2βhjσj) (1 + exp (2βhjσj)) . (39) (cid:12)(cid:12)(cid:12)(cid:12) W 1(IJ) W 0(IJ) (cid:12)(cid:12)(cid:12)(cid:12) ≤(cid:88) j This ratio scales exponentially with βhjσj/2. However, in this case, the error function also goes to zero. In addition, increasing the value of βhjσj/2 also increases the value of the Hamiltonian H 0(IJ), and thus reduces the likelihood of the transition W (IJ). Thus, the larger the ratio W 1(IJ) W 0(IJ), the smaller the transition probability. In the following, we will neglect all high order terms k ≥ 1. We do not extend our derivations to a general non-symmetric noise distribution in this work. We still suggest the following ideas to treat this extension: (cid:46) We could first treat the non-symmetric case as a perturbation of the symmetric case and thus derive the error on the effective Hamiltonian of the spin distribution. (cid:46) In this view, we suggest the parametrization of the skewness using the skew normal distribution78, which is a natural extension of the toy-model symmetric noise distribution analyzed in Ref.58. Detailed balance condition 18 To determine the stationary distribution of the spin state, we first need to verify that the transition probability satisfies the triangular equality58, equivalent to the detailed balance condition in the context of Markov Chains1: W (IK)W (KJ)W (JI) = W (JK)W (KI)W (IJ). This is equivalent to: H(IK) + H(KJ) + H(JI) = H(JK) + H(KI) + H(IJ), (41) i in the Hamiltonian). We can then deduce the which is satisfied if J is symmetric (there is no condition on the linear term h0 effective Hamiltonian by decoupling the ratio of transition probabilities into the ratio of two terms depending on distributions I and J (by using Eq. (36))58 and HL(I) = − 1 β ln F (I) gives the effective Hamiltonian describing the dynamics of the spin distribution: = = exp(β(cid:80) exp(β(cid:80) F (I) F (J) , (cid:88) W 0(IJ) W 0(JI) HL(I) = − 1 β (cid:81) i cosh(β((cid:80) (cid:81) i cosh(β((cid:80) i h0 i h0 i σi(I)) i σi(J)) j Jijσj(I) + h0 i ) j Jijσj(J) + h0 i ) (cid:88) i )) −(cid:88) log cosh(β( Jijσj(I) + h0 h0 i σi(I). i j i Effective Gibbs distribution (40) (42) (43) (44) (45) (46) (47) (48) (49) (50) Assuming J is symmetric, the stationary distribution of the spin state is a Gibbs distribution given by the effective Hamiltonian HL(I)19,58: In the following, we define .k as the usual k-norm (where k is an integer). P ({σi}) ∝ exp (−βHL({σi})). Small noise approximation In the small noise approximation (β (cid:29) 1), the effective Hamiltonian can be simplified using the following Taylor expansion : log cosh x ≈ x − log 2: HL({σi}, β) ≈ −(cid:88) (cid:12)(cid:12)(cid:88) i j Jijσj + h0 i + N β log 2 −(cid:88) h0 i σi i In the large noise approximation (β (cid:28) 1), the effective Hamiltonian can be simplified using the following Taylor expansion : log cosh x ≈ log(1 + x2/2) ≈ x2/2, = −J(cid:126)σ + (cid:126)h01 + N β log 2 − (cid:126)h0 · (cid:126)σ := H1({σi}, β). Large noise approximation HL({σi}, β) ≈ − β 2 Kijσiσj (cid:124) − ij (cid:88) (cid:123)(cid:122) (cid:88)(cid:0)h0 (cid:123)(cid:122) Quadratic term β 2 (cid:124) j (cid:33) (cid:125) −(cid:88) (cid:124) i (cid:32) (cid:88) (cid:123)(cid:122) k k + h0 i β σi Linear term Jkih0 (cid:125) (cid:1)2 (cid:125) 2 − (cid:126)h0 · (cid:126)σ := H2({σi}, β), Constant (independent of σ) = − β 2 J(cid:126)σ + (cid:126)h02 with K = J 2. Discussion on h0 i 19 Identifying the Hamiltonian in Eq. (50) to the general Ising Hamiltonian in Eq. (10) requires solving the following system of equations: (cid:88) (cid:88) k  Kij + δij∆i = β bi = β j Jij = Jji, JikJkj Jkih0 k + h0 i (51) (52) (53) with unknown variables (h0 i , J). δij is the Kronecker symbol and ∆i represents a degree of freedom on the diagonal of the Hamiltonian (which results in an additional constant term). Studying the set of general Ising Hamiltonians (K, b) (Eq. (10)) which can be mapped to an Ising network (J, h0) in the large noise approximation (Eq. (50)) would require an extensive study that we do not carry in this work. However, one can notice that: (cid:46) In the case where the external field satisfies J · h0 = 0, the system has a trivial solution h0 conditions for which a square root of K can be found are discussed in the next section). √ i = bi and J = K (the (cid:46) In the more general case, one is given N "free" degrees of freedoms ∆i in finding solutions to Eq. (51), while there are N constraints to solve in Eq. (52). Thus, it seems likely that, in non-degenerate cases, this system will have a non-trivial solution. However, one should make sure that while tuning the degrees of freedom ∆i, the matrix J remains a real square root of the matrix K (see discussion in the next section on finding a square root). If not, this algorithm should be generalized to using complex-valued matrices J. We do not discuss this generalization of the algorithm in this work. However, considering that arbitrary (complex) unitary transformations can be implemented with our photonic architecture34, this would be an interesting extension to this work. In the following, for the sake of simplicity, we will assume h0 gives: i ≡ 0. In this case, the large noise approximation Hamiltonian (cid:88) (54) HL({σi}, β) ≈ − β 2 Kijσiσj, ij with K = J 2. Thus, the implementation of the Little network to model a general Ising Hamiltonian (Eq. (10)) strongly relies on the possibility of finding a symmetric, real square root to the matrix K. Let us remind the reader of the assumptions we have made so far: (cid:46) The model is synchronous, or as stated in Little's original paper14 "we shall suppose that the neurons are not permitted to fire at any random time but rather that they are synchronized such that they can only fire at some integral multiple of a period τ". (cid:46) The neurons have no memory of states older than their previous state (Markov process). (cid:46) The connections do not evolve in time (there is no learning involved). (cid:46) J is symmetric and h0 i = 0, which results in (cid:88) j θi = Cij/2, from Eq. (22). (55) 20 FIG. 1. Transitioning between Hamiltonians. a: To approximate a given Ising model in the large noise limit, one should take into account all eigenvalues. b: Effective Hamiltonian and its large/low-noise limit approximation when α = 1 (with (16/16) eigenvalues). c: However, to reduce the noise threshold of the large noise expansion, one should drop out some eigenvalues. d: Effective Hamiltonian and its large/low- noise limit approximation when α = 0 (with (7/16) eigenvalues, all negative eigenvalues being dropped out). In this Figure, we study a random spin glass with N = 16 and choose ∆ii = (cid:80) j Kij. Inequalities between Hamiltonians Using basic algebra and functional analysis, we can show that log cosh x ≤ x2/2 and that log cosh x ≥ x − log 2, which results in the following inequality on the Hamiltonians H2({σi}, β) ≤ HL({σi}, β) ≤ H1({σi}, β). (56) By summing over spin configurations, we get the reversed inequality for the partition functions (57) Another inequality can be derived by using the equivalence of norms .1 and .2 in finite dimensions (.2 ≤ .1 ≤ N.2): √ Z2(β) ≥ ZL(β) ≥ Z1(β). 0 ≤ H2({σi}, β) ≤ −H1({σi}, β) + − 2 β log 2 ≤ N β H2({σi}). (58) (cid:114) (cid:115)−2N β These Hamiltonians are plotted for a sample random Ising problem in Figure 1. Suggested algorithms As suggested above, from the large noise expansion of the effective Hamiltonian, a natural way to probe the Gibbs distribution of some Ising model defined by coupling matrix K (Eq. (10) is to set the matrix of the recurrent loop C to be a modified square root of K: √ Sqα(J) = Re( K + α∆), (59) as the sum of the off-diagonal term of the Ising coupling matrix ∆ii = (cid:80) 21 where α ∈ [−1, 1] is the eigenvalue dropout level, a parameter we will discuss below, and ∆ is a diagonal offset matrix defined j(cid:54)=i Kij. An alternative solution would be to set C = K. This solution has been studied in the particular case of associative memory couplings19, where it was shown that the PRIS in this configuration would be described by a free energy function equal to the Ising free energy (up to a factor of 2). We will discuss the pros and cons of both algorithms in section IV. SUPPLEMENTARY NOTE 3: CONSTRUCTION OF THE WEIGHT MATRIX 22 We here propose a technique to build a square root of K in order to find an approximate solution to any Ising problem. We start with the general K Ising weight matrix K from Eq. (10). We notice that as K is symmetric and Kii = 0, K will never obey the condition of Lemma 1. Aii > 0, there exists B ∈ SN (R) such that B2 = A. Proof 1 Let X be a vector of size N and norm 1 such that AX = λX with λ < 0. We assume i0 is the coordinate of X with Lemma 1 (Diagonal dominance.) If a A ∈ SN (R) is a real symmetric matrix such that for all i, Aii ≥ (cid:80) maximum absolute value (and xi0 > 0): xi0 = x∞. We have(cid:80) = (cid:88) ≤ (cid:88) x∞Ai0i0 < Ai0i0 − λx∞ j Ai0jxj = λxi0. We thus get : j(cid:54)=i Aij and Ai0jxj j(cid:54)=i0 Ai0jx∞. By simplifying this inequality, we get Ai0i0 <(cid:80) j(cid:54)=i0 which means that A is positive semidefinite. Ai0j which contradicts our assumption. Thus, for all X, X T AX ≥ 0, j(cid:54)=i0 We can thus write A = U DU T where U is unitary and D = Diag(λ1,··· , λN ) with λi ≥ 0. The result is given by √ √ √ B = U DU T ∈ SN (R) with √ D = Diag( λ1,··· , λN ) To be able to apply Lemma 1, we need to add diagonal terms to the Ising matrix K from Eq. 10 by defining K = K + ∆ where ∆ is a diagonal matrix such that K verifies the assumptions of theorem 1. From there, we can define J = large noise approximation, we thus get the following Hamiltonian describing the PRIS: Kijσiσj − β 2 HL({σi}, β) ≈ − β 2 Kijσiσj = −β (cid:88) Kiiσ2 i ij (cid:112) K. In the (cid:88) (cid:88) 1≤i<j≤N 1≤i<j≤N (cid:88) (cid:88) 1≤i≤N = −β Kσiσj − β 2 ∆ii. 1≤i≤N Since the ∆ matrix is fixed, the second term in this equation is a constant (independent of the spin distribution, given an Ising problem to minimize). The Gibbs distribution is thus P ({σi}) ∝ exp (−βHL({σi})) ≈ Cte exp (−βHK,0) where HK,0 is the Ising Hamiltonian defined in Eq. (10). verify Lemma 1, we can choose ∆ii = (cid:80) The diagonal offset that is added to the original Ising matrix can be considered as a supplementary degree of freedom. To j(cid:54)=i Kij, to make sure K is positive definite and has a real symmetric square root Discussion on the diagonal offset J 2 = K. However, there is no reciprocal to Lemma 1. We here show that even some partial or weak reciprocals are usually wrong: (cid:46) The direct reciprocal of Lemma 1 is generally wrong. We consider the following counter-example: (cid:19) (cid:18) 1 −2 −2 8 A = . A is symmetric, has positive diagonal elements but is not diagonally dominant (the inequality does not hold for only one diagonal element). However, A is symmetric positive definite: its determinant is 4 and its trace 9, so the sum and product of eigenvalues is positive, which means both are positive. (cid:46) Even a weak reciprocal of Lemma 1 is wrong: If A ∈ SN (R), A has positive diagonal elements, such that for all i, Aii <(cid:80) eigenvalue. j(cid:54)=i Aij, then A has at least one negative For example, one can consider :  . A is symmetric, has positive diagonal elements, verifies: ∀i, Aii <(cid:80)  1 1 1 1 1 1 1 1 1 A = j(cid:54)=i Aij, but still, its eigenvalues are 0, 0, and 3. (cid:46) One (very) weak reciprocal of Lemma 1 is the following Lemma 2. It means that any Ising matrix (that has a zero diagonal) will never have a real symmetric square root, unless we add an offset to its diagonal. 23 Lemma 2 If A ∈ SN (R) (real symmetric matrix) such that for all i, Aii = 0, then A has at least one negative (and one positive) eigenvalue. Proof 2 We have(cid:80) i Aii =(cid:80) i λi = 0 so there is at least one negative and one positive eigenvalue. (cid:46) Lemma 3 (Weak reciprocal: unweighted MAX-CUT case.) If K is equal to minus the adjacency matrix of a graph, then the matrix K = K + α∆ is symmetric positive semi-definite if and only if α = 1. Proof 3 Let's first notice that this case still holds for a NP-hard subclass of Ising problems. In this case, K's elements are −1 if there is a spin-spin connection and 0 otherwise. Thus, K = K + ∆ has a zero eigenvalue: KX = 0 with X = (1, 1, ..., 1)T (because the sum of each row is equal to 0). We notice that for α < 1: (60) As seen before, K + ∆ has a zero eigenvalue, and (α − 1)∆ only has negative eigenvalues. Using Weyl's inequalities, we can deduce that K + α∆ has at least one strictly negative eigenvalue. K + α∆ = (K + ∆) + (α − 1)∆. Studying the reciprocal of Lemma 1 gives us a better insight on what is the optimal diagonal offset ∆ to add to the original Ising matrix. We can consider ∆ as a supplementary degree of freedom in our algorithm. The choice of ∆ suggested by this Lemma on diagonal dominance is particularly adapted for α > 0 and Ising problems with couplings that all have the same sign. j(cid:54)=i Kij will be much greater than Kii. Thus, α ∼ −1/N is the lower limit for non-zero J matrices. The domain of definition of α over which the number of eigenvalues actually varies is then j(cid:54)=i Kij for large spin glasses. In this study, we will specify which However, we notice that for large spin glasses, typically(cid:80) asymmetric. For this reason, we typically choose ∆ii = (cid:80) definition of ∆ is chosen for each Figure and study. Tuning the search dimensionality with eigenvalue dropout Tuning the eigenvalue dropout gives us a way of tuning the search dimensionality. If K = U DU† where U is real unitary and D = Diag(λ1, ...λN ), we can parametrize the phase space in terms of the eigenvectors of K, which we denote as ej (associated to eigenvalue λj): (cid:88) j (cid:126)σ = µjej, We can rephrase the optimization problem as follows: H((cid:126)σ) = − 1 2 = − 1 2 j µ2 σT Kσ (cid:88) (cid:80) i = ±1 2 (cid:40) (cid:80) argminµ − 1 j µjej j λj. j µ2 j λj Find As only the positive eigenvalues can decrease the energy, we would like to conclude that only the eigenvectors associated with positive eigenvalues will contribute to minimizing the energy. However, one should make sure that the hard constraint in Eq. (64) remains satisfied. We can only conclude on the heuristic result that the ground state of the optimization probably will prefer having components in eigenspaces associated with positive eigenvalues. As reducing the eigenvalue dropout level results in dropping out the negative eigenvalues, this explains why we usually observe a better performance for a certain level around α = 0. (61) (62) (63) (64) Modified algorithm with tunable offset 24 We define ∆0 as the minimum offset to verify the assumption of Lemma 1 and make sure the modified Ising matrix is symmetric positive semi-definite. ∆0 is defined as: (cid:88) Aij, ∆0 ii = ij = 0 for i (cid:54)= j. ∆0 j(cid:54)=i We define the modified algorithm as follows: (cid:46) K = K + α∆0 where α ∈ [0, 1] √ (cid:46) J = Re We define λi = i(cid:112)λi if λi < 0. Then K, defined as follows: as K is symmetric, there is a unitary (and real) U such that K = UDiag(λ1, ..., λN )U†. √ (cid:16) (cid:112) (cid:112) (cid:17) U†. J = U Diag( λ1, ..., λN ) This modified algorithm only corresponds to a particular parametrization of the weight matrix and can thus be implemented with the PRIS. The writing of this problem as a function of the coupling matrix eigenvalues (see above) proves that, as long as α ≥ 0 in the modified algorithm, the ground state of the Ising problem will likely remain unchanged (see Figure 4). 25 FIG. 2. Optimized noise level used in Figure 2 (main text). a: for MAX-CUT graphs (Figure 2c in the main text). b: for spin glasses (Figure 2b in the main text). FIG. 3. Conjugated influence of eigenvalue dropout and noise levels for finding the ground state. For each tuple (α, φ), the PRIS is run 100 times and Niter,90% is plotted (the ground state is pre-computed with BiqMac57). The colorbar is saturated at 105 iterations in order to show the valley of optimal (α, φ) values. In this Figure, we study a MAX-CUT graph with density d = 0.5 and ∆ii =(cid:80) j(cid:54)=i Kij. SUPPLEMENTARY NOTE 4: NUMERICAL SIMULATIONS Evaluating sampling performance and critical parameters on two-dimensional ferromagnetic and infinite-range Ising models In the main text, we evaluate the performance of the PRIS on sampling the Gibbs distribution of analytically solvable Ising models and estimating their critical exponents. We chose the following two problems: (cid:46) Two-dimensional ferromagnetic Ising model on a square lattice with periodic boundary conditions. This model was first analytically solved by Onsager21. Its energy is defined as (cid:88) (cid:104)ij(cid:105) H = − 1 2 σiσj, (65) where (cid:104)ij(cid:105) refers to nearest neighbors (i, j) under periodic boundary conditions (see Figure 5). (cid:46) Infinite-range Ising model, where each node is connected with a positive coupling of 1/N to all its neighbors, including 26 FIG. 4. Influence of dropout (a-c) for a random spin glass and (d-f) for a sample random cubic graph with N = 8 and φ = 0.5. (a, b, d, e): Eigenvalue distribution for various noise levels. (c, f): Energy landscape (Hamiltonian) plotted over linearized phase space for various eigenvalue dropout levels α. For this Figure, we set ∆ii =(cid:80) j Kij. FIG. 5. Examples of analytically solvable Ising models studied in our benchmark. (a) Two-dimensional ferromagnetic Ising model and (b) infinite-range (mean-field) Ising model. itself. This model can be analytically solved by mean field theory79. Its energy is defined as (cid:88) ij (cid:88) i H = − 1 2N σiσj = − 1 2N ( σi)2. (66) In order to measure the critical exponents, we need to make sure the free energy at all temperatures is equivalent to that of the corresponding Ising model at all temperatures. It has been shown that the free energy of a Little network with coupling matrix defbRANDOM CUBIC GRAPHSPIN GLASS(2/8)(1/8)(4/8)(8/8)Decreasing eigenvalue dropout level αα = 00255Energy (Little Hamiltonian)Linearized phase space0255Linearized phase space000.51-5-10-15 = 0.5 = 0.5= 2= 2acbTransition matrix eigenvalueEigenvalue indexEigenvalue index(6/8)(5/8)(4/8)α = 0(3/8)(8/8)(6/8)(5/8)(4/8)(3/8)(8/8)Eigenvalue indexEigenvalue index(2/8)(1/8)(4/8)(8/8)00-0.05-5-10-1500.5127 FIG. 6. Physical observables of the two-dimensional Ising model obtained with L = 36 square-rooted PRIS. K equals that of an Ising model defined by the same coupling matrix K, when the coupling weights are configured to learn a set of stable configurations (associative memory)19. However, this network has been discarded because of the possible existence of loops between some of its degenerate states, which can result in odd dynamics of the system, if no additional precautions are taken (see, for instance, Figure 7, where this simple algorithm actually converges to the maximal energy). We observe that adding a diagonal offset to the coupling, as we do in the square-root version of this algorithm, suppresses these odd dynamics (Figure 7(b)). The algorithm described earlier to find the ground state of Ising problems could also be used to measure critical exponents, as can be seen in Figure 6. However, there are some complications that arise: (cid:46) Taking the square root of the coupling matrix prevents the use of symmetry and sparsity to reduce the algorithm complexity. Thus, for large graphs, the time needed to make a single matrix multiplication becomes quite large on a CPU. (cid:46) Taking the square root also modifies the coupling amplitude. It is now unclear how the temperature of the system should be defined (which is a problem if we want to estimate the critical temperature with the PRIS). From the large-noise expansion of the Hamiltonian, we should define the effective temperature as T = k2φ2. However, we observe this definition does not match the theoretical value of the critical temperature of the 2D Ferromagnetic Ising model. For both Ising problems studied, we perform the following analysis: (cid:46) We first estimate the Binder cumulant U4 (see definition in the main text) as a function of the system temperature, for various graph sizes N = L2. The cumulants U4 plotted for different L intersect at the critical temperature1 (see Figure 4(a) in the main text). (cid:46) We then estimate the dependence on the linear dimension L of various observables at the critical temperature and deduce the corresponding critical exponent. This is enabled by the scaling law of observables at the critical temperature1: m ∼ L−βC /νC , χ ∼ LγC /νC , auto ∼ LzE τ E C , auto ∼ Lzm τ m C , (67) (68) (69) (70) auto, and τ m where m, χ, τ E auto are respectively the magnetization, the magnetic susceptibility, the energy autocorrelation time and the magnetization autocorrelation time1. To estimate the critical exponents, we run the algorithm 120 times for 105 iterations, with random initial conditions, at the critical temperature for each L and fit these observables with a power law in L. 28 FIG. 7. Comparison of different versions of the algorithms on the 2D Ferromagnetic Ising Model. (a) With no offset, the no square root algorithm (defined by the following regime of operation of the PRIS: C = K) results in convergence to the state with maximal energy, for some runs (with random initial conditions). (b) Adding an offset to the same regime of operation cancels this behavior and the algorithm always converges to the ground state after some time. Algorithm βC R2 γC R2 zE C R2 zm C R2 2D ferromagnetic Ising model 0.1257 0.992 1.735 0.998 1.393 0.993 2.068 0.998 0.1194 0.972 1.867 0.990 1.860 0.977 2.023 0.994 Infinite-range Ising model 0.5245 0.997 1.011 0.999 0.920 0.999 0.924 0.999 0.5650 0.999 0.922 0.999 0.914 0.999 0.886 0.999 MH PRIS MH PRIS TABLE II. Summary of critical exponents measured with the PRIS and MH. R2 is the coefficient of determination of each power law fitting. auto is the energy autocorrelation The estimates of all observables are obtained by taking every τ E time, and dropping the first 10% of the data (arbitrary burn-in or equilibrium time). Our findings are summarized in Table II, Figures 8 and 9, and are benchmarked versus the Metropolis-Hastings algorithm, which is summarized in Refs.1,68,69. auto generated samples, where τ E 29 FIG. 8. Probing the critical exponents of the 2D Ferromagnetic Ising model. Fits are shown with the resulting critical exponent for the PRIS (a, c, e, g) and the MH (b, d, f, h) algorithms for the susceptibility (a-b), energy autocorrelation time (c-d), magnetization (e-f), and magnetization autocorrelation time (g-h). 30 FIG. 9. Probing the critical exponents of the infinite-range Ising model. Fits are shown with the resulting critical exponent for the PRIS (a, c, e, g) and the MH (b, d, f, h) algorithms for the susceptibility (a-b), energy autocorrelation time (c-d), magnetization (e-f), and magnetization autocorrelation time (g-h). 31 FIG. 10. Simulated scaling of PRIS on a cascaded array of programmable MZIs. Number of algorithm steps to reach a ground state with a probability of 90% is plotted as a function of the graph order for various graph topologies: MAX-CUT graphs with densities d = 0.1 (a), d = 0.5 (b), and d = 0.9 (c), Random Cubic Graphs (RCG), and spin glasses (d). SUPPLEMENTARY NOTE 5: SCALING OF THE PRIS PERFORMANCE ON SEVERAL PHOTONIC ARCHITECTURES Cascaded arrays of programmable Mach-Zehnder interferometers (MZI) In Figure 10, we examine the influence of the bit accuracy of the setting of the phases on the performance of the machine. This can describe any system where the matrix is encoded by a cascaded array of programmable MZIs such as34,80. We assume the phase can be set with b-bit accuracy, which means that when setting phase θm (resp. φm) on the PNP, we actually draw from a uniform distribution θ ∈ [θm − 2π (2π)b ; φm + 2π (2π)b ]). (2π)b ; θm + 2π Another way to describe the bit accuracy of the PNP is through the bit-accuracy of the voltage setting34. The phase-voltage relation can be approximated by a quadratic dependence: θ = 2π where V2π is the voltage setting to achieve 2π modulation60. If we assume the voltage is set with bV accuracy, then the actual voltage is drawn from a (uniform) distribution over [V − V2π 2bV ]. This translates to the phase accuracy as: 2bV ; V + V2π (cid:17)2 (2π)b ] (resp. φ ∈ [φm − 2π (cid:16) V (cid:19)2 (cid:18) V2π θ ± ∆θ = 2π /V 2 2π (71) V ± V2π 2bV √ 2bV −1 + 2πθ 2π 4bV = θ ± (72) We can safely neglect the last term and we notice that the worst case scenario corresponds to θ ∼ 2π, for which bV − 1 = bθ. A rule of thumb results: a bV -bit accuracy of the PNP on its voltage setting corresponds to a (bV − 1)-bit accuracy of the PNP on its phase setting. Inversely, a bθ-bit accuracy of the PNP on its phase setting corresponds to a (bθ + 1)-bit accuracy of the PNP on its voltage setting. 10-910010710-210-9Graph order0100200b = 8b = 1610-910-1108100(s)badc(s)(s)(s)Graph order0100200Graph order0100200Graph order0100200b = 8b = 16b = 8b = 1610-910010810-1b = 8b = 16d = 0.1d = 0.5MAX-CUTd = 0.9MAX-CUTspin glassRandom Cubic GraphMAX-CUT10010910032 Static sources of noise could be a significant bottleneck in scaling the PRIS to large N ∼ 100 graph orders. For instance, a static noise on the phase setting of an array of MZI will result in a static error on the effective coupling between spins, thus reshaping the Hamiltonian landscapes, which could impact the algorithm efficiency. We simulate the algorithm performance as a function of the graph order N for phase resolutions of bθ = 8 and 16 bits. The resulting time on a GHz photonic architecture to find the ground state with 90% chance, Titer,90%, is also shown in Figure 10. While a 16-bit phase resolution does not impact the algorithm performance (with scaling results comparable to an ideal photonic network, see main text), an 8-bit phase resolution may increase the required number of algorithm steps by one to two orders of magnitude, depending on the graph order and topology (while still outperforming other photonic systems on a GHz architecture, such as9,61). Thus, the reduction of static noise is of paramount importance in the realization of the PRIS on large-scale static photonic networks. Optical Neural Networks based on Photoelectric Multiplication For larger graphs N ∼ 103 − 106, one could resort to recently-proposed large-scale optical neural networks based on photo- electric multiplication67. By encoding both matrix weights Cij and input signals S(t) into optical (time) domain, the measured output is added a Gaussian noise term with amplitude defined by the number of photons per Multiply And Accumulate (MAC): i S(t+1) i = Thθ CijS(t) j + w(t) i CS(t) N 3/2 √ N√ nmac (cid:88) j  . (73) For the various problems we study in this paper, the working standard deviation is usually ∼ 1. This corresponds to a working nmac of nmac ∼ N C2S(t)2 N 3 (74) We can evaluate the corresponding total energy consumption per matrix multiplication for 10 random spin glasses. We get nmac ∼ 4 (resp. ∼ 15) for N = 100 (resp. N = 1, 000). Smaller working nmac will be required for sparser graphs, since C is smaller. The corresponding SNR scales as ∼ nmac and the total energy is N 2nmac ∼ 6.2 ± 0.35 fJ/matrix multiplication (resp. 1.9 ± 0.5 pJ/matrix multiplication). There are many attractive features of these networks for the implementation of large-scale PRIS: (cid:46) This architecture naturally leverages quantum noise which perturbs the output as is required for the good execution of the PRIS. The noise level can be tuned by changing the number of photons per MAC which is proportional to the SNR. (cid:46) The non-linear function is executed in electronic domain, which allows a lot of flexibility on its implementation and re- configuration (the threshold function required for the good operation of the PRIS would be straightforward to implement), while optical nonlinearities working at low-power have not been demonstrated so far. (cid:46) This architecture is in principle scalable to very large number of spins N ∼ 106. Free space optical Neural Networks Since the PRIS relies on a static transformation, the use of free-space optical neural networks28,40 with 3D printed masks or reconfigurable SLMs is another option to achieve PRIS with N ∼ 106 neurons. The analysis we made on the influence of heterogeneities (see Figure 10) remains relevant here. A set of lens - mask - lens would realize the matrix multiplication on the signal S(t) encoded in optical domains, while the coupling matrix C is encoded in the mask transmission. The speed of such a free-space architecture would only be limited by the photodetector (typically ∼ 10 THz) and modulation (typically ∼ 1 kHz for SMLs, (cid:39) 1 GHz ) bandwidths. Comparison table of various heuristic algorithms and architectures Since the PRIS essentially relies on fast vector-to-fixed matrix multiplication, it can be implemented efficiently on various photonic and parallel electronic hardware architectures, such as FPGAs. We summarize in Table III the projected performance of various heuristic algorithms (MH and PRIS) running on several hardwares (both electronics and photonics). We observe that, Algorithm Algorithm step Architecture MH Sequential PRIS Parallelizable CPU CPU FPGA MZI network34,80 Free space optics61,81 Algorithm step time estimate with N = 100 modulation bandwidth limited (SLM ∼ 1ms, electro-optic modulators ∼ 0.1 − 1ns) ∼ 1 ms∗ ∼ 10 − 100µs∗ ∼ 5 − 100ns∗∗ ∼ 0.1 − 1ns 33 Time Complexity O(N 2) O(N 2) O(N 2/M ) O(N ) O(1) TABLE III. Comparative table of projected performance of various heuristic algorithms implemented on various architectures. Time complexity is given as a function of the total number of spins N. Regarding clock estimates: ∗ Estimate run on a 2.7 GHz Intel Core i5 with Matlab, ∗∗ Estimate run on a Xilinx Zynq UltraScale+ MPSoC ZCU104 with a systolic array architecture. See Supplementary Note 7 for more details. while photonics potentially allows the fastest clock for such algorithms and a competitive scaling factor, FPGAs can achieve similar clocks and require much less engineering (they can be bought off the shelf and are easily reconfigurable). To demonstrate our point, we perform a proof-of-concept experiment on a FPGA board, whose results are shown in Supplementary Note 7. If photonic architectures can significantly reduce the time complexity of the algorithm step by performing massive multiplexing81, it must be noted that we are neglecting time overhead such as fabrication, etc. Also, some free-space archi- tectures, such as SLM, are typically slow to reconfigure (on the order of 1ms), thus only being relevant for multiplying very large matrices. 34 FIG. 11. Benchmarking the PRIS versus MH on large spin glasses N = 1000. For both PRIS and MH, the algorithm is ran 100 times for 10,000 iterations at each temperature / noise level. The line shows the average ground state energy over 100 runs, and the shaded area corresponds to ± the standard deviation. PRIS is ran for various dropout levels corresponding to (Neig/N ) eigenvalues (see legend). Results for MH and PRIS at the smallest eigenvalue dropout level (311/1000) are also averaged over the 10 random spin glasses. Results for the PRIS at all others eigenvalue dropout levels are only averaged over spin glass 1 (see discussion). For this study, we choose ∆ii = (cid:80) j Kij. SUPPLEMENTARY NOTE 6: COMPARISON OF THE PRIS TO SEVERAL (META)HEURISTICS Benchmarking versus Metropolis-Hastings on large spin glasses N ∼ 1000 In order to evaluate the performance of the PRIS on larger graphs, for which exact solvers typically fail, we benchmark the PRIS against MH for a set of 10 random spin glasses (whose couplings are randomly chosen from a uniform distribution in the interval [−1, 1]). First, we notice that the behavior probed by the PRIS and MH is very similar over the 10 spin glasses. The temperature/noise-dependent mean ground state energy found by MH and the PRIS is shown in Figure 11. In particular, we see that the PRIS achieves mean ground state energies similar to MH. As a reminder, a heuristic mapping between the temperature from statistical physics and the effective temperature (noise level) of the PRIS is T ∼ (kφ)2. These graphs orders are large enough, so that standard exact solvers are taking a long time to find the ground state. For instance, we ran BiqMac on spin glass 1 on their online submission server57 for three hours. The algorithm could only find an approximate ground state which was outperformed by MH, PRIS-A, and SA. absolute lowest ground state energy is obtained for α = −0.2, i.e. optimizing the hyperparameter α around α ∼ 0 when running the PRIS on large graphs. We also observe that the PRIS shows an α-dependent optimal noise level, which increases with α. For spin glass 1, the (450/1000) eigenvalues. This hints at the necessity of PRIS-A: a proposed metaheuristic variation of the PRIS A route to achieving systematically low energy states is to design metaheuristics, i.e. master strategies guiding the search for an optimal ground state. Simulated Annealing (SA)29 is one of such algorithms, derived from Metropolis-Hastings. Let us reminder the reader of one possible implementation of this algorithm with the widely used geometric schedule for the temperature29,32,82,83: Start from random initial state Choose initial temperature Ti and geometric factor λ < 1 T ← Ti for all i ∈ {1, ..., Nalg, iter} T ← λT for all j ∈ {1, ..., Niter per temp.} Algorithm 1: Simulated annealing algorithm with a geometric schedule. Update state according to MH acceptance rule at temperature T end for end for (311/1000)(347/1000)(384/1000)(450/1000)(501/1000)(550/1000)α = 0PRIS0246810Temperature-1.4-1.3Ground state energy104xMHbaα-dependent optimal noise level35 PRIS Instance MH PRIS-A SA BiqMac spin glass 1 -13,716 -13,796 -13,795 -13,814 -13,746 spin glass 2 -13,632 -13,736 -13,716 -13,754 N.C. spin glass 3 -13,717 -13,777 -13,760 -13,796 N.C. spin glass 4 -13,678 -13,768 -13,781 -13,798 N.C. spin glass 5 -13,732 -13,755 -13,767 -13,782 N.C. spin glass 6 -13, 752 -13,828 -13,804 -13,832 N.C. spin glass 7 -13,723 -13,792 -13,758 -13,800 N.C. spin glass 8 -13,731 -13,769 -13,781 -13,783 N.C. spin glass 9 -13,711 -13,810 -13,798 -13,817 N.C. spin glass 10 -13,754 -13,846 -13,822 -13,855 N.C. TABLE IV. Summary of benchmarking PRIS and PRIS-A against MH and SA. For both PRIS and MH, the algorithm is ran 100 times for 10,000 iterations at each temperature / noise level. The table shows the absolute lowest ground state energy recorded. For PRIS-A and SA, the algorithm is ran 100 times with Nalg, iter temperature increments (as given by Eq.(75)) and Niter per temp. = 100. For PRIS-A (resp. SA), the initial noise level (resp. temperature) is φi = 50 (resp. Ti = 5, 000), the final noise level is φf = 0.1 (resp. Tf = 0.01) and the 0.991 (resp. λ = 0.991). For PRIS-A, the eigenvalue dropout level is taken to be α = 0, corresponding temperature geometric factor is λ = to (501/1000) eigenvalues. For BiqMac, the algorithm ran for three hours (time limit on the BiqMac online job submission platform). N.C. = Non Computed. √ This naturally inspires a metaheuristic based on the PRIS, which we call Photonic Recurrent Ising Simulated Annealing (PRIS-A): Start from random initial state Choose initial noise level φi and geometric factor λ < 1 φ ← φi for all i ∈ {1, ..., Nalg, iter} φ ← λφ for all j ∈ {1, ..., Niter per temp.} Update state according to PRIS acceptance rule at noise level φ end for end for Algorithm 2: Photonic Recurrent Ising Simulated Annealing (PRIS-A) algorithm. Let us note a couple of peculiarities: (cid:46) The noise level from the PRIS is related to an effective temperature via T ∼ φ2. Thus, one should compare SA ran with a √ geometric factor λ to PRIS-A with a geometric factor λ. (cid:46) In the PRIS-A algorithm, the eigenvalue dropout level α is also a degree of freedom. One could thus, in principle, also simulate the annealing of the eigenvalue dropout level, thus affecting the ground state search dimensionality. A comprehensive study of this new class of algorithms goes beyond the scope of this work. (cid:46) For given initial and final noise levels/temperatures and geometric factors, one can determine the number of temperature increments Nalg, iter with the formula: Nalg, iter = log φf − log φi log λ . (75) (cid:46) λ is typically chosen to be smaller but close to 1, in order to mimic adiabatic temperature variations. We verify that both for SA and PRIS-A, λ > 0.98 yields consistently low energy ground states, with no particular amelioration when increasing λ (and scaling the number of temperature increments Nalg, iter). The initial and final noise levels chosen for PRIS-A are φi = 50 and φf = 0.1. The initial and final temperatures chosen for SA are Ti = (2 ∗ 0.5877 ∗ φi)2 ∼ 3454 and Tf = (2 ∗ 0.5877 ∗ φf )2 ∼ 0.01. We observed no significant variation on the minimum ground state energy found for λ > 0.99 and ran each algorithms with a rate of λ = 0.991. The performance of the various algorithms we implement is shown in Table IV. MH yields lower ground state energies than PRIS on average of 0.53%. However, PRIS-A can outperform PRIS by a similar amount, lowering energies on average of 0.46%. This is a larger performance enhancement than SA to MH (0.11% decrease of ground state energy). Then, on average, SA outperforms PRIS-A by 0.18%. We expect optimization of the eigenvalue dropout level α (and simulated annealing on this parameter) to further enhance the performance of PRIS and PRIS-A. 36 37 FIG. 12. High level architecture of the design. SUPPLEMENTARY NOTE 7: IMPLEMENTATION OF THE PRIS ON FPGA Architecture The PRIS algorithm is primarily designed for future photonic implementations. However, both photonic chips and FPGAs (Field Programmable Gate Arrays) share parallel processing capability. Thus, it is worthwhile to first demonstrate the perfor- mance of the algorithm on an FPGA board. We have implemented the algorithm on Xilinx Zynq UltraScale+ multiprocessor system-on-chip (MPSoC) ZCU104 evaluation board based on the Pynq framework. The high-level synthesis, place, and route have been performed by Xilinx Vivado 2018.3 design suite. Zynq and Zynq Ultrascale+ devices integrate a multi-core processor (ARM Cortex-A9) and programmable logic (FPGA) in the same circuitry. PYNQ is an open-source project released by Xilinx that enables Python Productivity for Zynq devices. It incorporates the open-source Jupyter notebook infrastructure to run an Interactive Python (IPython) kernel and a web server directly on the ARM processor of the Zynq device. It also provides extensive hardware libraries (overlays) and APIs which enable easier and faster programming of FPGA. For our application, the preprocessing and preparation of data is performed in a Jupyter notebook using Python. Once ready, we transmit the data and write to memories on FPGA through AXI interface. After running the recurrent algorithm for a number of cycles on FPGA, the final state vector is transmitted back for data analysis like energy calculation and correctness verification. In this manner, the same hardware configuration could be easily reconfigured and run different problems efficiently. Noise generation and post-processing (energy calculation, and more generally extracting observable) could also be run directly on the FPGA in future versions of this implementation. The overall high level architecture is shown in Figure 12. We make use of the address-mapped AXI interface to encode different operations (e.g. Block RAM (BRAM) addressing, loop setting, result selection) into different addresses. AXI controller loads matrix, noise and threshold data into BRAMs, then sets up the initial state vector and starts the computation. The loop module is a finite-state-machine which reads the data from BRAM, adds noise, compares with threshold and then updates the state vector at every loop step. The final state vector and clock information is transmitted back through AXI. We emphasize the following considerations regarding our implementation: (cid:46) The bottleneck of the performance implementation is to read input matrix data from memory. The ZCU104 board has 312 BRAM blocks, with each block transmitting a maximum of 72 bits per clock cycle when configured under true-dual-port mode. It also has 94 UltraRAM (URAM) blocks with each block transmitting a maximum of 144 bits per clock cycle. Thus the total maximum data rate it could read from BRAM and URAM is 36288 bits per clock cycle. For our current implementation (N < 100) we could assume we have enough memory, but as the problem size increases this becomes a major limitation. (cid:46) We multiply an N-by-N b bit matrix (b being the bits we choose to represent the Ising problem data) with an N-by-1 1-bit state vector, so each data in the result is a conditional sum of one row of matrix C based on the value of the state vector. We choose to implement a binary tree for speeding up the sum as shown in Figure 13. At each time step, a certain number of rows of the matrix is loaded onto the leaves of the tree, and then gets propagated to next level based on the value of the state vector, and adds all up thereafter. The result of the sum should propagate to the root after a delay of clock cycles equal to the height of the tree. The loop is carefully pipelined, in which reading the matrix takes N 2b clock cycles, multiplication (binary tree addition) takes 38 FIG. 13. Binary tree architecture for matrix-to-vector multiplication. A row of matrix C is read from memory to the top leaves of the binary tree. The branch connecting the top leaves to the next tree layer are either active or not, depending on the value of the current state vector (0 or 1). FIG. 14. Performance comparison for various rounding values. log2 N cycles, while noise injection and thresholding are done in a single clock cycle. Details of the complexity is discussed in the next session. We are only running integer arithmetic on the FPGA, so to adapt the original algorithm which runs on float points, we need to scale the parameters (matrix, noise, and threshold) and then round them to integers. The scaling factor and bit length to represent each data needs to be carefully designed. Note that the only difference between our implemented algorithm on the FPGA and the ideal algorithm is that we round our values to a specific bit depth. Specifically, we store the matrix values, noise, and threshold values only to a specific bit precision. This originates from memory constraints on the FPGA, but it is reasonable to suspect that this rounding affects the performance of the algorithm. To test the effects of this change, we implemented both our original algorithm and the version with rounding on MATLAB, testing various bit depth with varying input, as shown in Figure 14. We found that, for a matrix of size N = 100 by 100, rounding after multiplying by 64 or even 32 performed just as well as using arbitrary bit precision (Matlab's default double-precision floating point). Specifically, for a variety of matrix inputs, on average using 64 and 32 roundings reached states of essentially equivalent energy to the arbitrary precision. Over 1000 trials each with a different input matrix, 1000 iterations each (i.e. 1000 matrix multiplications and noise additions and thresholding for each trial), on average our normal algorithm reached a minimum energy of -408.63, whereas the 32 rounding reached -409.08, and the 32 bit reached -409.17. Complexity 39 In discussing the complexity of our implementation, we must note the dependence on the particular resources of a given FPGA. Here we will discuss the complexity of a single step of our algorithm, consisting of a single matrix multiplication followed by noise addition and threshold comparison. Since the clock speed of any FPGA is variable, we give the complexity in terms of clock cycles. We focus on computing the complexity of the matrix multiplication, since it dominates the three steps. To perform this matrix multiplication, we must read the entire matrix from memory and feed it to a system of binary trees. Suppose we store each matrix entry using b bits, and that we are working with an N × N size matrix. In our FPGA, we can read 64 bits from each BRAM block per clock cycle, so one limiting factor is the total number k of BRAM blocks we can use concurrently. Note that we are also limited by the maximum number R of bits that can be held in registers, or the working memory, of our FPGA. This allows us to compute a complexity bound. Assuming the FPGA can read M bits from each BRAM block per clock cycle, it will take the FPGA at least N 2b M cycles to read the entire matrix. According to the above considerations, M = 64k + 144u, where k (resp. u) is the number of used BRAM (resp. URAM). In our current implementation, we only used BRAM memory so u = 0. Alternatively, we are also limited by the amount of data we can work with at a time, R. To complete this matrix multiplication, we must store a binary tree corresponding to each row. The result of these binary trees is a vector of N values to be sent to the noise addition step. Thus the total amount of data we will need to work with in this computation is 2N 2b bits, which will take at least 2N 2b R cycles to pass through the registers of the FPGA. To finish the analysis, note that we may feed new data into the binary trees before the original data has fully propagated through. Thus the only computation time outside of that required to load the matrix is the time for the final data to propagate through the binary trees. The size of our binary trees can be determined by considering the minimum of the amount of data we can load at a time, M = 64k, and the amount of data (in bits) we can work with at a time, R. Note that we never need a binary tree larger than the size of a row of the matrix, and that the binary tree takes up twice the space of the data inputted into it. Then the size of our binary tree is B := min( 64k 2b , N ), and it takes on the order of log2 B cycles to propagate through. Then, incorporating the time to load the matrix, the total number of cycles for a single algorithm step is on the order of b , R (cid:18) 2N 2b (cid:19) number of clock cycles per PRIS algorithm iteration = max , N 2b 64k R + log2 B. Lastly, considering noise addition and thresholding, note that this only involves reading 2N b more bits from bram. The actions of noise addition and thresholding themselves only take two clock cycles, and the time and space necessary to deal with these extra 2N b bit is negligible in complexity compared to that necessary to deal with the N 2b bits from the matrix. So these two steps do not change the overall complexity. We also need to load values into the BRAM in the first place, but we do not include this process in the analysis. Thus the overall complexity of a single iteration of our algorithm is O . The discussion above is under the assumption that for large size problems (N > 100), board registers are not enough for holding all the data in the matrix. For small size problem, we could alternatively use registers and do the multiplication in one clock cycle, then the major time consumption would be the binary tree addition, which took log2 B clock cycles. 64k ) + log2 B For problems with larger size, we store matrix in BRAM as mentioned above. We could plot the complexity (Figure 15) discussed above under different assumptions: (1) Using all of the memory blocks on the FPGA (Figures Figure 15(a) and (b)), which is optimal but takes time to implement for every problem size. (2) For most reasonably-sized problems, we can assume a linear scaling of memory blocks, i.e. M = rN b, which is sub-optimal for a given N but easier to reconfigure for various N allowing this approximation (we can switch between different problem sizes N by changing the width of BRAM IPs in our design). We also assumed that R is large, thus not being the limiting factor of the computation. (cid:16) max( 2N 2b R , N 2b (cid:17) Experiment In this experiment, we tested the correctness and runtime for a problem size of N = 100 under a 300 MHz clock. We used 16 bits to encode each data in matrix, noise and threshold. A total of k = 5 BRAM IPs are instantiated in the design, each with data length of 1600 bits, corresponding to one row of the matrix. (In this design we are using 71% of BRAM blocks on our board.) The experiment runs as follows: (cid:46) Original Test: First we generate a test case running on float numbers in Jupyter notebook using Python. We run the test for 200 clock cycles and record the state vector and energy of each iteration. 40 FIG. 15. Complexity estimate of a PRIS single algorithm iteration, implemented on FPGA. a: Graph showing the estimated time consumption on the current ZCU104 board, given that were using all of the memory blocks and under a clock of 300MHz and 16 bit encoding. b: Comparison of three different FPGA boards with different memory resources. c: Assume linear scaling of memory, i.e. M = rN b, here we let r = 10 (read 10 rows of matrix at a time). The vertical lines indicate the limit of each FPGA. For our ZCU104 board, the maximum problem size under the linear scaling assumption is N = 212. FIG. 16. Comparison of FPGA and simulated (Python) outputs for a given random spin graph with N = 100. (cid:46) Scale the problem: We scale up the matrix, noise and threshold by a factor of 128 and round to integers. We run the scaled test for 200 clock cycles and plot the energy array together with the original test to check the accuracy of the algorithm with rounded values. Several scaling factor and energy arrays are shown in Figure 14, in which we could see that scaling under 64 shows some deviation to the unrounded algorithm, and scaling above 1024 results in exactly the same sequence of state vectors. (cid:46) Correctness verification: we load the scaled test onto FPGA, and run 200 clock cycles and output state vector at each time step, with which we calculate the energy in Python. The FPGA-simulated energy and Python calculated energy is exactly the same, confirming the correctness of the implementation. (cid:46) Timing and complexity analysis: for this part we only output the state vector and clock count at the very end of 200 clock cycles. For our problem of size 100 by 100, a loop of 200 iterations cost 3803 cycles. On average each time step takes 19 clock cycles (the remaining 3 clock cycles are due to some overhead at the beginning of the run), which is in accordance with our analysis before. The measured clock cycle per algorithm step is 19 cycles, which is exactly as expected: 19 = 10 + 8 + 1 in which 10 is the cycles it takes to read the whole matrix (we are reading r = 10 rows at each clock cycle), 8 is the time for the binary addition tree, and an additional clock is required for noise injection and thresholding. In conclusion, we have implemented the PRIS for problem size N = 100 on an FPGA platform with a time per algorithm step of approximately 63 ns. 41 42 ∗ Corresponding authors e-mail: [email protected], [email protected] [1] D. P. Landau and K. Binder, A Guide to Monte Carlo Simulations in Statistical Physics. 2009. [2] J. Hromkovic, Algorithmics for Hard Problems: Introduction to Combinatorial Optimization, Randomization, Approximation, and Heuristics. Springer Berlin Heidelberg, 2013. [3] M. Kardar, G. Parisi, and Y.-C. Zhang, "Dynamic Scaling of Growing Interfaces," Physical Review Letters, vol. 56, pp. 889 -- 892, mar [4] M. B. Isichenko, "Percolation, statistical topography, and transport in random media," Reviews of Modern Physics, vol. 64, pp. 961 -- 1043, 1986. oct 1992. [5] A. R. Honerkamp-Smith, S. L. Veatch, and S. L. Keller, "An introduction to critical points for biophysicists; observations of compositional heterogeneity in lipid membranes," Biochimica et Biophysica Acta (BBA) - Biomembranes, vol. 1788, pp. 53 -- 63, jan 2009. [6] R. Albert and A.-L. Barab´asi, "Statistical mechanics of complex networks," Reviews of Modern Physics, vol. 74, pp. 47 -- 97, jan 2002. [7] F. Glover and G. Kochenberger, Handbook of Metaheuristics. Springer, 2006. [8] Z. Wang, A. Marandi, K. Wen, R. L. Byer, and Y. Yamamoto, "Coherent Ising machine based on degenerate optical parametric oscillators," Physical Review A, vol. 88, p. 063853, dec 2013. [9] P. L. McMahon, A. Marandi, Y. Haribara, R. Hamerly, C. Langrock, S. Tamate, T. Inagaki, H. Takesue, S. Utsunomiya, K. Aihara, R. L. Byer, M. M. Fejer, H. Mabuchi, and Y. Yamamoto, "A fully programmable 100-spin coherent Ising machine with all-to-all connections.," Science (New York, N.Y.), vol. 354, pp. 614 -- 617, nov 2016. [10] K. Wu, J. Garc´ıa de Abajo, C. Soci, P. Ping Shum, and N. I. Zheludev, "An optical fiber network oracle for NP-complete problems," Light: Science & Applications, vol. 3, pp. e147 -- e147, feb 2014. [11] M. R. V´azquez, V. Bharadwaj, B. Sotillo, S.-Z. A. Lo, R. Ramponi, N. I. Zheludev, G. Lanzani, S. M. Eaton, and C. Soci, "Optical NP problem solver on laser-written waveguide platform," Optics Express, vol. 26, p. 702, jan 2018. [12] W. M. Macready, A. G. Siapas, and S. A. Kauffman, "Criticality and Parallelism in Combinatorial Optimization," Science, vol. 271, pp. 56 -- 59, jan 1996. [13] Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," Nature, vol. 521, pp. 436 -- 444, may 2015. [14] W. A. Little, "The existence of persistent states in the brain," Mathematical Biosciences, vol. 19, no. 1-2, pp. 101 -- 120, 1974. [15] J. J. Hopfield, "Neural networks and physical systems with emergent collective computational abilities.," Proceedings of the National Academy of Sciences of the United States of America, vol. 79, pp. 2554 -- 8, apr 1982. [16] J. J. Hopfield and D. W. Tank, "Neural computation of decisions in optimization problems," Biological Cybernetics, vol. 52, no. 3, [17] E. Ising, "Beitrag zur Theorie des Ferromagnetismus," Z. Phys., 1925. [18] M. M´ezard and A. Montanari, Information, Physics, and Computation. 2009. [19] D. J. Amit, H. Gutfreund, and H. Sompolinsky, "Spin-glass models of neural networks," Physical Review A, vol. 32, pp. 1007 -- 1018, aug [20] A. Pelissetto and E. Vicari, "Critical phenomena and renormalization-group theory," Physics Reports, vol. 368, pp. 549 -- 727, oct 2002. [21] L. Onsager, "Crystal Statistics. I. A Two-Dimensional Model with an Order-Disorder Transition," Physical Review, vol. 65, pp. 117 -- 149, [22] N. V. Brilliantov, "Effective magnetic Hamiltonian and Ginzburg criterion for fluids," Physical Review E, vol. 58, pp. 2628 -- 2631, aug [23] D. J. Amit, Modeling brain function : the world of attractor neural networks. Cambridge University Press, 1989. [24] N. Ghofraniha, I. Viola, F. Di Maria, G. Barbarella, G. Gigli, L. Leuzzi, and C. Conti, "Experimental evidence of replica symmetry breaking in random lasers," Nature Communications, vol. 6, p. 6058, dec 2015. [25] M. A. Halasz, A. D. Jackson, R. E. Shrock, M. A. Stephanov, and J. J. M. Verbaarschot, "Phase diagram of QCD," Physical Review D, pp. 141 -- 152. 1985. feb 1944. 1998. vol. 58, p. 096007, sep 1998. pp. 3241 -- 3253, oct 1982. jun 1990. 1985. may 1983. vol. 7, p. 3910, nov 2005. [26] F. Barahona, "On the computational complexity of Ising spin glass models," Journal of Physics A: Mathematical and General, vol. 15, [27] J. Bruck and J. W. Goodman, "On the power of neural networks for solving hard problems," Journal of Complexity, vol. 6, pp. 129 -- 135, [28] N. H. Farhat, D. Psaltis, A. Prata, and E. Paek, "Optical implementation of the Hopfield model," Applied Optics, vol. 24, p. 1469, may [29] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, "Optimization by simulated annealing.," Science (New York, N.Y.), vol. 220, pp. 671 -- 80, [30] D. J. Earl and M. W. Deem, "Parallel tempering: Theory, applications, and new perspectives," Physical Chemistry Chemical Physics, [31] L. D. Davis and M. Mitchell, Handbook of Genetic Algorithms. 1991. [32] F. Glover and M. Laguna, "Tabu Search," in Handbook of Combinatorial Optimization, pp. 2093 -- 2229, Boston, MA: Springer US, 1998. [33] E. Boros, P. L. Hammer, and G. Tavares, "Local search heuristics for Quadratic Unconstrained Binary Optimization (QUBO)," Journal of Heuristics, vol. 13, pp. 99 -- 132, feb 2007. [34] Y. Shen, N. C. Harris, S. Skirlo, M. Prabhu, T. Baehr-Jones, M. Hochberg, X. Sun, S. Zhao, H. Larochelle, D. Englund, and M. Soljaci´c, "Deep learning with coherent nanophotonic circuits," Nature Photonics, vol. 11, pp. 441 -- 446, jun 2017. [35] A. Silva, F. Monticone, G. Castaldi, V. Galdi, A. Al`u, and N. Engheta, "Performing mathematical operations with metamaterials," Science, vol. 343, no. 6167, pp. 160 -- 163, 2014. Letters, vol. 73, pp. 58 -- 61, jul 1994. interferometers," Optica, vol. 3, p. 1460, dec 2016. Networks," apr 2018. [39] W. R. Clements, P. C. Humphreys, B. J. Metcalf, W. S. Kolthammer, and I. A. Walsmley, "Optimal design for universal multiport [40] X. Lin, Y. Rivenson, N. T. Yardimci, M. Veli, M. Jarrahi, and A. Ozcan, "All-Optical Machine Learning Using Diffractive Deep Neural [41] M. Gruber, J. Jahns, and S. Sinzinger, "Planar-integrated optical vector-matrix multiplier," Applied Optics, vol. 39, p. 5367, oct 2000. [42] A. N. Tait, M. A. Nahmias, Y. Tian, B. J. Shastri, and P. R. Prucnal, "Photonic Neuromorphic Signal Processing andComputing," pp. 183 -- 222, Springer, Berlin, Heidelberg, 2014. [43] A. N. Tait, M. A. Nahmias, B. J. Shastri, and P. R. Prucnal, "Broadcast and weight: An integrated network for scalable photonic spike processing," Journal of Lightwave Technology, vol. 32, no. 21, pp. 3427 -- 3439, 2014. [44] K. Vandoorne, P. Mechet, T. Van Vaerenbergh, M. Fiers, G. Morthier, D. Verstraeten, B. Schrauwen, J. Dambre, and P. Bienstman, "Experimental demonstration of reservoir computing on a silicon photonics chip," Nature Communications, vol. 5, p. 3541, dec 2014. [45] A. Saade, F. Caltagirone, I. Carron, L. Daudet, A. Dremeau, S. Gigan, and F. Krzakala, "Random projections through multiple optical scattering: Approximating Kernels at the speed of light," in 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 6215 -- 6219, IEEE, mar 2016. [46] D. Pierangeli, V. Palmieri, G. Marcucci, C. Moriconi, G. Perini, M. De Spirito, M. Papi, and C. Conti, "Deep optical neural network by living tumour brain cells," dec 2018. [47] Z. Cheng, H. K. Tsang, X. Wang, K. Xu, and J.-B. Xu, "In-Plane Optical Absorption and Free Carrier Absorption in Graphene-on-Silicon Waveguides," IEEE Journal of Selected Topics in Quantum Electronics, vol. 20, pp. 43 -- 48, jan 2014. [48] Q. Bao, H. Zhang, Z. Ni, Y. Wang, L. Polavarapu, Z. Shen, Q.-H. Xu, D. Tang, and K. P. Loh, "Monolayer graphene as a saturable absorber in a mode-locked laser," Nano Research, vol. 4, pp. 297 -- 307, mar 2011. [49] A. C. Selden, "Pulse transmission through a saturable absorber," British Journal of Applied Physics, vol. 18, pp. 743 -- 748, jun 1967. [50] M. Soljaci´c, M. Ibanescu, S. G. Johnson, Y. Fink, and J. D. Joannopoulos, "Optimal bistable switching in nonlinear photonic crystals," Physical Review E, vol. 66, p. 055601, nov 2002. [51] R. W. Schirmer and A. L. Gaeta, "Nonlinear mirror based on two-photon absorption," Journal of the Optical Society of America B, vol. 14, [52] P. Horowitz and H. Winfield, "The Art of Electronics," American Journal of Physics, 1990. [53] B. Boser, E. Sackinger, J. Bromley, Y. Le Cun, and L. Jackel, "An analog neural network processor with programmable topology," IEEE Journal of Solid-State Circuits, vol. 26, no. 12, pp. 2017 -- 2025, 1991. [54] J. Misra and I. Saha, "Artificial neural networks in hardware: A survey of two decades of progress," Neurocomputing, vol. 74, pp. 239 -- p. 2865, nov 1997. 255, dec 2010. [36] A. F. Koenderink, A. Al`u, and A. Polman, "Nanophotonics: Shrinking light-based technology," may 2015. [37] J. Carolan, C. Harrold, C. Sparrow, E. Mart´ın-L´opez, N. J. Russell, J. W. Silverstone, P. J. Shadbolt, N. Matsuda, M. Oguma, M. Itoh, G. D. Marshall, M. G. Thompson, J. C. Matthews, T. Hashimoto, J. L. O'Brien, and A. Laing, "Universal linear optics," Science, vol. 349, pp. 711 -- 716, aug 2015. [38] M. Reck, A. Zeilinger, H. J. Bernstein, and P. Bertani, "Experimental realization of any discrete unitary operator," Physical Review 43 [55] D. Vrtaric, V. Ceperic, and A. Baric, "Area-efficient differential Gaussian circuit for dedicated hardware implementations of Gaussian function based machine learning algorithms," Neurocomputing, vol. 118, pp. 329 -- 333, oct 2013. [56] I. A. D. Williamson, T. W. Hughes, M. Minkov, B. Bartlett, S. Pai, and S. Fan, "Reprogrammable Electro-Optic Nonlinear Activation Functions for Optical Neural Networks," arXiv preprints, arxiv.org:1903.04579, mar 2019. [57] F. Rendl, G. Rinaldi, and A. Wiegele, "Solving Max-Cut to optimality by intersecting semidefinite and polyhedral relaxations," Mathe- matical Programming, vol. 121, pp. 307 -- 335, feb 2010. [58] P. Peretto, "Collective properties of neural networks: A statistical physics approach," Biological Cybernetics, vol. 50, pp. 51 -- 62, feb 1984. [59] M. Lipson, "Guiding, Modulating, and Emitting Light on Silicon-Challenges and Opportunities," Journal of Lightwave Technology, Vol. 23, Issue 12, pp. 4222-, vol. 23, p. 4222, dec 2005. [60] N. C. Harris, Y. Ma, J. Mower, T. Baehr-Jones, D. Englund, M. Hochberg, and C. Galland, "Efficient, compact and low loss thermo-optic phase shifter in silicon," Optics Express, vol. 22, p. 10487, may 2014. [61] R. Hamerly, T. Inagaki, P. L. McMahon, D. Venturelli, A. Marandi, T. Onodera, E. Ng, C. Langrock, K. Inaba, T. Honjo, K. Enbutsu, T. Umeki, R. Kasahara, S. Utsunomiya, S. Kako, K. I. Kawarabayashi, R. L. Byer, M. M. Fejer, H. Mabuchi, D. Englund, E. Rieffel, H. Takesue, and Y. Yamamoto, "Experimental investigation of performance differences between coherent Ising machines and a quantum annealer," Science Advances, 2019. [62] D. A. B. Miller, "Perfect optics with imperfect components," Optica, vol. 2, p. 747, aug 2015. [63] R. Burgwal, W. R. Clements, D. H. Smith, J. C. Gates, W. S. Kolthammer, J. J. Renema, and I. A. Walmsley, "Using an imperfect photonic network to implement random unitaries," Optics Express, vol. 25, p. 28236, nov 2017. [64] V. R. Almeida, C. A. Barrios, R. R. Panepucci, and M. Lipson, "All-optical control of light on a silicon chip," Nature, vol. 431, pp. 1081 -- [65] C. T. Phare, Y. H. Daniel Lee, J. Cardenas, and M. Lipson, "Graphene electro-optic modulator with 30 GHz bandwidth," Nature Photonics, 1084, oct 2004. 2015. [66] C. Haffner, D. Chelladurai, Y. Fedoryshyn, A. Josten, B. Baeuerle, W. Heni, T. Watanabe, T. Cui, B. Cheng, S. Saha, D. L. Elder, L. R. Dalton, A. Boltasseva, V. M. Shalaev, N. Kinsey, and J. Leuthold, "Low-loss plasmon-assisted electro-optic modulator," Nature, 2018. [67] R. Hamerly, L. Bernstein, A. Sludds, M. Soljaci´c, and D. Englund, "Large-Scale Optical Neural Networks Based on Photoelectric Multiplication," Physical Review X, vol. 9, nov 2019. [68] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, "Equation of state calculations by fast computing 44 machines," The Journal of Chemical Physics, 1953. [69] W. K. Hastings, "Monte carlo sampling methods using Markov chains and their applications," Biometrika, 1970. [70] J. Dean, D. Patterson, and C. Young, "A New Golden Age in Computer Architecture: Empowering the Machine-Learning Revolution," IEEE Micro, vol. 38, pp. 21 -- 29, mar 2018. [71] Y. Dou, S. Vassiliadis, G. K. Kuzmanov, and G. N. Gaydadjiev, "64-bit floating-point FPGA matrix multiplication," in Proceedings of the 2005 ACM/SIGDA 13th international symposium on Field-programmable gate arrays - FPGA '05, (New York, New York, USA), p. 86, ACM Press, 2005. [72] D. C. Knill and A. Pouget, "The Bayesian brain: the role of uncertainty in neural coding and computation," Trends in Neurosciences, [73] W. Maass, "Noise as a resource for computation and learning in networks of spiking neurons," Proceedings of the IEEE, vol. 102, vol. 27, pp. 712 -- 719, dec 2004. pp. 860 -- 880, may 2014. [74] Q. Wang, E. T. F. Rogers, B. Gholipour, C.-M. Wang, G. Yuan, J. Teng, and N. I. Zheludev, "Optically reconfigurable metasurfaces and photonic devices based on phase change materials," Nature Photonics, vol. 10, pp. 60 -- 65, jan 2016. [75] A. N. Tait, T. F. de Lima, E. Zhou, A. X. Wu, M. A. Nahmias, B. J. Shastri, and P. R. Prucnal, "Neuromorphic photonic networks using silicon photonic weight banks," Scientific Reports, vol. 7, p. 7430, dec 2017. [76] R. M. Karp, "Reducibility among Combinatorial Problems," in Complexity of Computer Computations, pp. 85 -- 103, Boston, MA: Springer US, 1972. 1992. [77] C.-K. Looi, "Neural network methods in combinatorial optimization," Computers & Operations Research, vol. 19, pp. 191 -- 208, apr [78] A. Azzalini and A. D. VALLE, "The multivariate skew-normal distribution," Biometrika, vol. 83, pp. 715 -- 726, dec 1996. [79] S. Friedli and Y. Velenik, Statistical Mechanics of Lattice Systems. Cambridge University Press, nov 2017. [80] N. C. Harris, G. R. Steinbrecher, M. Prabhu, Y. Lahini, J. Mower, D. Bunandar, C. Chen, F. N. C. Wong, T. Baehr-Jones, M. Hochberg, S. Lloyd, and D. Englund, "Quantum transport simulations in a programmable nanophotonic processor," Nature Photonics, vol. 11, pp. 447 -- 452, jun 2017. [81] D. Pierangeli, G. Marcucci, and C. Conti, "Large-Scale Photonic Ising Machine by Spatial Light Modulation," Physical Review Letters, [82] Y. Nourani and B. Andresen, "A comparison of simulated annealing cooling strategies," Journal of Physics A: Mathematical and General, [83] H. Cohn and M. Fielding, "Simulated Annealing: Searching for an Optimal Temperature Schedule," SIAM Journal on Optimization, 2019. vol. 31, pp. 8373 -- 8385, oct 1998. vol. 9, pp. 779 -- 802, jan 1999.
1707.08894	3	1707	2018-02-13T22:00:35	Optimized Spintronic Terahertz Emitters Based on Epitaxial Grown Fe/Pt Layer Structures	[ "physics.app-ph", "cond-mat.mes-hall" ]	We report on generation of pulsed broadband terahertz radiation utilizing the inverse spin Hall effect in Fe/Pt bilayers on MgO and sapphire substrates. The emitter was optimized with respect to layer thickness, growth parameters, substrates and geometrical arrangement. The experimentally determined optimum layer thicknesses were in qualitative agreement with simulations of the spin current induced in the ferromagnetic layer. Our model takes into account generation of spin polarization, spin diffusion and accumulation in Fe and Pt and electrical as well as optical properties of the bilayer samples. Using the device in a counterintuitive orientation a Si lens was attached to increase the collection efficiency of the emitter. The optimized emitter provided a bandwidth of up to 8 THz which was mainly limited by the low-temperature-grown GaAs (LT-GaAS) photoconductive antenna used as detector and the pulse length of the pump laser. The THz pulse length was as short as 220 fs for a sub 100 fs pulse length of the 800 nm pump laser. Average pump powers as low as 25 mW (at a repetition rate of 75 MHz) have been used for terahertz generation. This and the general performance make the spintronic terahertz emitter compatible with established emitters based on nonlinear generation methods.	physics.app-ph	physics	OPEN Received: 19 July 2017 Accepted: 2 January 2018 Published: xx xx xxxx Optimized Spintronic Terahertz Emitters Based on Epitaxial Grown Fe/Pt Layer Structures Garik Torosyan1, Sascha Keller2, Laura Scheuer2, René Beigang2,3 & Evangelos Th. Papaioannou2,3 We report on generation of pulsed broadband terahertz radiation utilizing the inverse spin hall effect in Fe/Pt bilayers on MgO and sapphire substrates. The emitter was optimized with respect to layer thickness, growth parameters, substrates and geometrical arrangement. The experimentally determined optimum layer thicknesses were in qualitative agreement with simulations of the spin current induced in the ferromagnetic layer. Our model takes into account generation of spin polarization, spin diffusion and accumulation in Fe and Pt and electrical as well as optical properties of the bilayer samples. Using the device in a counterintuitive orientation a Si lens was attached to increase the collection efficiency of the emitter. The optimized emitter provided a bandwidth of up to 8 THz which was mainly limited by the low-temperature-grown GaAs (LT-GaAS) photoconductive antenna used as detector and the pulse length of the pump laser. The THz pulse length was as short as 220 fs for a sub 100 fs pulse length of the 800 nm pump laser. Average pump powers as low as 25 mW (at a repetition rate of 75 MHz) have been used for terahertz generation. This and the general performance make the spintronic terahertz emitter compatible with established emitters based on optical rectification in nonlinear crystals. The use of spin Hall effects to generate and manipulate spin currents has provided a large thrust in the research field of spintronics the last decade. Spin Hall effect and its reciprocal, the inverse spin Hall effect (ISHE), provide the means for conversion between spin and charge currents1,2. In particular, the ISHE transforms a pure spin current into a transverse charge current due to spin-orbit coupling. Extensive studies based on the ISHE have been performed since ISHE was first experimentally demonstrated by using non local detection techniques3 and by detecting electrically pure spin currents generated in spin pumping experiments4. Spin pumping effect refers to the generation of a spin current from a precessing magnetization in a magnetic layer (FM). The spin current appears at the interface of the magnetic layer with a non magnetic metallic layer (NM) and it propagates in the NM layer which has to exhibit a large spin-orbit coupling like for example Pt. The ISHE in a non magnetic layer then acts as an electrical detector of the spin currents. The utilisation of ISHE in sensing spin currents has been also applied in measuring spin caloritronic effects5,6. Recently, the decisive role of the ISHE effect on extending the field of spintronics into the terahertz (THz) regime was revealed7,8. THz spintronics has the potential of application in ultra-fast current and computer tech- nologies9. In the particular case of THz emission induced by the ISHE in FM/NM layers, a femtosecond laser pulse pumps a FM/NM heterostructure and generates non equilibrium spin polarized electrons in the FM layer. Subsequently, these electrons are diffused in the non-magnetic layer through a super diffusive process10,11. The spin current is then converted into a transient transverse charge current due to the ISHE in the NM layer. This transient current generates a short terahertz pulse that propagates perpendicular to the electrical current. The experimental demonstration of THz emission from FM/NM heterostructures due to ISHE has been beautifully shown recently7,8,12,13. However, the thickness dependence of the efficiency of the spintronic emitters, in particu- lar for very small layer thicknesses has not been well understood. Seifert et al.7, have shown that the THz signal exhibits a maximum for a specific total thickness of the metallic bilayer, although, according to their equation of 1Photonic Center Kaiserslautern, Kaiserslautern, 67663, Germany. 2University of Kaiserslautern, Department of Physics, Kaiserslautern, 67663, Germany. 3University of Kaiserslautern, Research Center Optimas, Kaiserslautern, 67663, Germany. Garik Torosyan, Sascha Keller, Laura Scheuer, René Beigang and Evangelos Th. Papaioannou contributed equally to this work. Correspondence and requests for materials should be addressed to R.B. (email: [email protected]) 1 Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9www.nature.com/scientificreportsFigure 1. Geometrical arrangement of the samples for the generation of THz radiation. In order to collimate the strongly divergent THz beam, a hyperhemispherical silicon lens is attached directly to the substrate without damaging the delicate Pt surface. the terahertz field amplitude7, the signal should scale with the metal thickness. They have attributed this behavior to a resonant enhancement in a kind of a Fabry-Pérot cavity formed by the bilayer. In contrast, Yang et al.13, have found no maximum in the signal by varying the Pt thickness in Fe (1.4 nm)/Pt (0–5 nm) bilayers. Instead, the signal was saturated above 3 nm Pt, possibly due to the spin diffusion length in Pt13. Furthermore, no pronounced maximum was observed in the Fe-thickness dependence curve for the Fe (0–5 nm)/Pt (3 nm) bilayer. Likewise, Wu et al.12, reported an increase and a saturation of the peak amplitude of THz signals for a NM/Co (4 nm)/SiO2 system as the thickness of the nonmagnetic layer (Pt or W) increased. By varying the thickness of the Co layer (0–10 nm) the THz peak reached a maximum value and then a gradual decrease. In this work, we present the effect of THz radiation from fully epitaxial Fe/Pt systems. We show the influence of the thickness of the individual layers on the THz emission amplitude and we discuss the role of the substrate. We demonstrate that for both Fe- and Pt-thickness dependencies the THz signal exhibits a maximum and a fast decrease after that. We quantify this behavior by using a model that takes into account generation of spin polarization starting at a minimum layer thickness of the magnetic material, spin diffusion, spin accumulation, and the electrical as well as optical properties of the bilayer samples. Results Geometrical arrangement of the samples. The Fe/Pt layer structures were epitaxially grown on 0.5 mm thick MgO or Al2O3 substrates with Pt as the outer layer. In principle, the layer structures can be pumped from both sides, the substrate side or the side with the Pt layer. In our experiments, the layer structures were pumped from the metal side and THz emission was detected from the substrate surface. As the Pt layer thickness is in the order of 3 nm for optimized emitters most of incident pump radiation is transmitted into the magnetic layer. For experiments with thicker Pt layers the additional absorption and reflection losses have to be taken into account when comparing the experimental results with simulations. We have therefore measured directly the total absorbed pump power for the metal layer structure. Only a certain fraction of this absorbed power will be con- verted into spin polarized electrons in a small layer close to the Fe/Pt boundary. This has to be taken into account in our model presented below. The counterintuitive geometry has the following advantages: In order to collimate the strongly divergent THz beam a hyperhemispherical silicon lens was attached directly to the substrate with the Fe/Pt plane in the focal point of the lens without damaging the delicate Pt surface. A schematic of the structures is shown in Fig. 1. Pumping the structure from the substrate side will always cause additional reflections from the substrate-air interface which in turn lead to strong oscillations in the corresponding THz spectra. In our geometry, the metal surface acts as an anti-reflection coating for the THz beam14 suppressing any reflections from the substrate sur- face. This is illustrated in Fig. 2 for a Fe/Pt sample on a MgO substrate without a silicon lens attached. When pumped from the substrate side (upper part in Fig. 2a) a second reflected pulse from the air-substrate interface propagates in the direction of the main pulse. Due to the high index of refraction of MgO at THz frequencies (n = 3) about 50% of the amplitude of the backward pulse is reflected. It can clearly be seen that there are no addi- tional reflections causing oscillations in the THz spectra when pumping the structure from the metal side (see the lower part of Fig. 2b). All further experiments were done in this counterintuitive geometry with the silicon lens attached if not stated otherwise. Because of nearly index matching between MgO and silicon there is almost no loss at the MgO silicon interface. With this set-up the samples were easily and reproducibly changed without destroying the alignment of the system allowing for a quantitative comparison between different samples (see Methods section). Pulse length and bandwidth. The pulse length and bandwidth of the THz pulse strongly depend on the rise and fall time of the transient electrical current in the nonmagnetic layer induced via the ISHE. Whereas the rise time is mainly determined by the pump pulse length and the diffusion properties of the spin current, the fall time is limited by the relaxation time of the electrons in the conducting material. For the relaxation time we have considered a superdiffusive transport process according to studies by M. Battiato15 resulting in a relaxation time 2 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9Figure 2. (a) Orientation of the sample with respect to the pump beam and (b) the generated THz pulses. In the orientation with the metal side towards the pump beam the reflection from the MgO-air interface is suppressed. Figure 3. (a) Typical THz pulse obtained from a Fe(2 nm)/Pt(3 nm) sample. (b) Spectra of the generated THz pulses for samples with two different substrates, MgO (red) and Al2O3 (blue). in the order of 5 fs in Pt16. Such a relaxation time supports bandwidths up to several tens of THz and does not limit the bandwidth we have observed in our experiments. In addition, absorption in the layer structures and the substrate influence both pulse length and bandwidth. A typical pulse and the corresponding spectra are shown in Fig. 3 for a 2 nm/3 nm Fe/Pt layer structure on MgO and Al2O3, respectively. The minimum pulse width was measured to be below 250 fs for both substrates with a ten- dency to shorter pulses for the Al2O3 substrate (see Fig. 3a). Spectra for bilayers on both substrates are shown in Fig. 3b) extending to approximately 8 THz. All spectra are shown as measured and are not corrected for the detec- tor response. Above 3 THz the well known strong THz absorption of MgO is visible for the MgO substrate. The dynamic range is well above 60 dB with a maximum around 1.5 THz. This is comparable with spectra obtained from photoconductive emitters. The maximum frequency measured in these experiments was finally determined by the frequency response of the dipole antenna of our photoconductive switch which was used as the detec- tor (see Methods section). The pulse length of 50 fs of our pump source would allow for even broader spectra. Furthermore, the absorption around 8 THz in GaAs which is used as the semiconductor material for the detector antenna limits the bandwidth. Having the choice between MgO and Al2O3 substrates MgO provides the higher dynamic range at frequencies below 3 THz. This is probably caused by the fact that there is an almost perfect epi- taxial growth of Fe on MgO whereas on Al2O3 it is not the case, resulting in a smaller dynamic range. This finding is supported by results of experiments with polycrystalline Pt layers on Fe/MgO resulting in considerably weaker THz signals. Together with a different detector (e. g. using electro-optical sampling in GaP) and even shorter pump pulses a much broader bandwidth can be obtained. This has been shown recently by Kampfrath et al.7. Thickness dependence. In order to optimize our THz emitter we have performed a systematic study of the dependence of the THz amplitude on the thickness of the Pt and Fe layers for samples epitaxially grown on 0.5 mm thick MgO substrates. In a first set of experiments the Fe layer thickness was kept constant to 12 nm whereas the Pt layer thickness was varied from 0.25 nm to 12 nm. In the second set the Pt layer thickness was fixed at 3 nm while the Fe layer thickness was changed from 1 nm to 12 nm. The results are shown in Fig. 4. In Fig. 4a) 3 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9Figure 4. (a) Pt thickness dependence of the THz field amplitude for a constant Fe thickness of 12 nm. (b) Fe thickness dependence of the THz field amplitude for a constant Pt thickness of 3 nm that has provided the highest signal in (a). the Fe (12 nm)/Pt (x nm) series is shown. As a measure of the strength of the THz signal, the peak to peak ampli- tude of the first two oscillations of the THz pulse was determined. The variation in pump beam absorption through Pt layers with changing thicknesses has been taken into account for the comparison. The theoretical fitting curve in Fig. 4a) and b) is based on the following equation 1: ⋅ tanh    d −  d  0 λ  pol Fe 2 ⋅ n air + n MgO + Z 0 1 ⋅ ( d σ Fe Fe + d σ Pt Pt ) ⋅ tanh    d Pt 2 λ Pt    P abs + d Fe ( ⋅ − e d E THz ∝ d Pt d + )/ s Pt (1) where nair, nMgO and Z0 are the index of refraction of air, the index of refraction of MgO at THz frequencies and the impedance of vacuum, respectively. THz Fe Equation 1 corresponds to a model that describes the dependence of the generated THz signal on layer thick- nesses of Fe and Pt layers. In particular, it describes the dependence on the individual layer thicknesses and not only on the total layer thickness of Fe and Pt layers. The model explains the onset of THz generation above a cer- tain Fe layer thickness and the maximum at relatively small layer thicknesses. It takes into account all successive effects after the laser pulse impinges on the bilayer, including the absorption of the fs pulse in the Pt and Fe layers, the generation and diffusion of spin currents in Fe and Pt as well as the generation of THz radiation and its atten- uation in the metal layers. In detail, the first term accounts for the absorption of the femtosecond laser pulse in the metal layers. As only spin polarized electrons within a certain distance from the boundary between Fe and Pt will reach the Pt layer only a fraction of the measured total absorbed power contributes to the generated THz signal. This fraction scales with the inverse of the total metal layer thickness 1/(dFe + dPt). The second term describes generation and diffusion of the generated spin current flowing in Fe towards the interface with Pt; to elaborate the possibility that extra thin Fe layers can loose their ferromagnetic properties below a certain thickness we intro- duce the term d0 in equation 1. Below this critical thickness we consider that the flow (if any) of spin current in Fe does not reach the Pt layer. Above this critical thickness the generated spin polarization saturates with a charac- teristic constant λpol. Magneto-optical Kerr effect measurements confirm this assumption (see Methods section). The third and forth term, the tangent hyberbolic function divided by the total impedance, refers to the spin accumulation in Pt which is responsible for the strength of the THz radiation and it depends on the finite dif- fusion length λPt of the spin current in Pt. The symbols σFe and σPt are the electrical conductivities of the two materials, respectively. This part of the equation is according to the theory of spin pumping effect17,18 and includes the shunting effect of the parallel connection of the resistances of the individual Fe and Pt layers. Although the electrical conductivity depends on layer thickness for very thin layers we have used a constant value for the layer thicknesses. This value is considerably smaller than the bulk value as in the thickness ranges in our experiments the bulk value has not been reached19. The last term describes the attenuation of the THz radiation during propagation through the metal layers (with sTHz as an effective inverse attenuation coefficient of THz radiation in the two metal layers). In the case of small losses the attenuation can be taken into account by this single exponential factor and the third factor which accounts for the multiple reflections of the THz pulse at the metal/dielectric interfaces (see Methods section). All terms together describe the layer thickness dependence of the measured THz amplitudes. The influence of the silicon lens on the thickness dependence of the generation process can be neglected as the distance between lens and metallic layers is five orders of magnitude larger than the layer thicknesses itself. Experiments with and without silicon lens resulted in the same thickness dependence. In Fig. 4a there is rather good qualitative agreement between our experimental data and the theoretical expec- tation. For the simulations we have assumed a spin current diffusion length in Pt of λPt = 1.40 nm. This is in good agreement with direct measurements of the diffusion length using microwave techniques20. The other con- stants used for this fit are nair = 1, nMgO = 3, Z0 = 377 Ω, σFe = 6·106 Ω−1m−1, σPt = 2·106 Ω−1m−1, d0 = 0.9 nm, 4 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9Figure 5. (a) Experimental set-up for THz generation. (b) THz pulses for two opposite directions of the magnetic field. λpol = 0.3 nm, and sTHz = 22 nm. The optimum Pt layer thickness for maximum THz generation seems to be between 2 and 3 nm. We have used this optimum Pt layer thickness for the determination of the optimum Fe layer thickness. With a fixed Pt layer thickness of 3 nm the Fe layer thickness was changed from 1 nm to 12 nm. The results are shown in Fig. 4b again together with the simulation using the same parameters as for the previous measurements. For the comparison the variation in pump light absorption in the magnetic layer has been accounted for. There is a steady increase in THz power with decreasing Fe layer thickness following the theoretical simulations. However, for a Fe layer thickness below 2 nm there is a considerable decrease in THz amplitude which cannot be explained alone by the decreasing pump light absorption. We attribute this decrease in signal to the onset of loss of magnetic proper- ties of the Fe layer resulting in a critical thickness d0 = 0.9 nm. As a consequence, the optimum layer structure for maximum THz generation is 3 nm of Pt on 2 nm of Fe. Discussion We have shown that Fe/Pt epitaxial bilayer systems can be considered as a competitive THz radiation source of spintronic nature. The spintronic THz emitter based on the ISHE presented here has a number of advantages compared to other optical and electro-optical THz emitters: The alignment of the emitter with respect to the pump beam is simplified as one has a free choice choosing the focusing spot within the sample surface. A col- limating Si lens can be attached directly to the substrate of the bilayer sample collecting most of the strongly divergent THz beam. Thus, changing the NM/FM bilayer samples becomes possible without loss of alignment of the total THz system. Up to pump fluences of 5 mJ/cm2 we did not observe any damage of our samples (see Methods section). There are no electrical connections required for the operation of the emitter. The polarization of the generated THz radiation can easily be changed by changing the weak magnetic field perpendicular to the direction of the pump beam. The average pump power levels required to generate useful THz signals is compa- rable to power levels used in optical rectification methods in nonlinear crystals. As the excitation of electrons in the Fe layer is independent of wavelength other pulsed laser sources can be used which are easier to operate like e.g. fs fiber lasers at 1.5 μm. The THz pulse length and bandwidth is, in principle, only limited by the relaxation time of hot electrons in the Pt material. The best configuration of the THz emitter was revealed by varying Fe and Pt layer thicknesses. Samples with 2 nm Fe and 3 nm Pt gave the maximum THz amplitude. To quantify the thickness dependence of the THz amplitude we have successfully developed a model that takes into account the optical absorption in the metal layers, the generation and diffusion of hot carries in Fe, the shunting effect, spin accumulation in Pt and the THz absorption in Fe. The optimization and the modeling of the radiation with the materials properties is necessary for future applications of the effect. Methods THz time domain set-up. A standard terahertz time-domain spectroscopy (THz-TDS) system, described in detail elsewhere21, has been used for generation and measurements of THz waveforms from different spin- tronic emitters (see Fig. 5). The system is driven by a femtosecond Ti:Sa laser delivering sub-100 fs optical pulses at a repetition rate of 75 MHz with an average output power of typically 600 mW. The laser beam is split into a pump and probe beam by a 90:10 beam-splitter. The stronger part is led through a mechanical computer-con- trolled delay line to pump the THz emitter, and the weaker part is used to gate the detector photoconductive switch with a 20 μm dipole antenna. In a classical (standard) THz-setup both the emitter and the detector operate with photoconductive antennas (PCA) whereas in the present work the PCA emitter is substituted by a spintronic (ST) Fe/Pt bilayer sample which is placed in a weak magnetic (20 mT) field perpendicular to the direction of the pump beam and in the direction of the easy axis of Fe. The direction of the magnetic field determines the polar- ization of the THz field which is perpendicular to the direction of the magnetic field. Changing the direction of magnetic field into the opposite direction changes the phase of the detected THz waveforms by 180° (Fig. 5b). In this way, by changing the orientation of the magnetic field the polarization of the generated THz radiation can be changed easily. The optical pump beam that is sharply focused onto the sample at normal incidence by an aspherical short-focus lens, excites spin polarized electrons in the magnetic layer (Fe) which give rise to a spin-current, which in turn excites a transverse transient electric current in the Pt layer. The latter results in THz pulse gener- ation of sub-picosecond duration being emitted forward and backward into free space in the form of a strongly 5 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9Figure 6. Maximum E-field amplitude as a function of pump fluence. divergent beam. The central wavelength within the beam is about 200 µm and is much longer than the diameter of its source which is smaller than 10 µm. That is why the emitted THz field would fill the half-space behind the sample. A hyperhemispherical Si-lens being attached to the back side of the sample collects it into a divergent beam in the form of a cone and directs it further. With the lens attached an enhancement factor of up to 30 in electric field amplitude has been observed. After the lens the beam is collimated with an off-axis parabolic mirror and sent to an identical parabolic mirror in the reversed configuration. The latter focuses the beam to the second Si-lens which finally focuses the beam through the GaAs substrate of the detector PCA onto the dipole gap for detection. In this way the THz optical system consisting of the two Si-lenses and the two parabolic mirrors images the point source of THz wave on the emitter surface onto the gap of the detector PCA and ensures an efficient transfer of the emitted THz emission from its source to the detector. In the present paper a 4-mirror THz-optics is used making it possible -besides two collimated beam paths- also to have an intermediate focus of the THz beam aimed at imaging applications (Fig. 5a)). In our set-up the THz beam path is determined by the silicon lens on the emitter, the parabolic mirrors and the silicon lens on the photoconducting antenna of the detector. The alignment of these components is not changed during an exchange of the spintronic emitter. If in addition the position of the pump beam focus remains constant the spintronic emitter can easily be exchanged without changing the beam path as the lateral position of the focus on the emitter is not critical assuming a homogeneous lateral layer struc- ture. The frequency response of the photoconductive dipole antenna with 20 μm dipole length limits the observ- able bandwidth. With this dipole length a maximum in the detector response around 1 THz can be expected with a reduction to 50% at 330 GHz and 2 THz. The 10% values are at 100 GHz and 3 THz. This estimation is based on calculations by Jepsen et al.22. Above 3 THz the frequency response of the detector is very flat and at 8 THz a strong phonon resonance in GaAs which is used as substrate material for the photoconductive antenna causes strong absorption of the THz radiation. Above 8 THz no THz radiation was detected. The delay line provides for the synchronous arrival of the weaker part of the optical pulse and that of the THz pulse at the detector antenna gap from either side, as well as for scanning in time the "open" state of the gap along the THz pulse duration. In each position of the delay line the transient current induced in the detector by the elec- tric field of the THz sub-picosecond pulse is proportional to its instantaneous electrical field value. It is summed up from many laser pulses which reach the detector during the single step of the delay line and integrated within the "open" state time window. It is measured as one single point of the THz wave-form and is in the order of several nano amperes at the maximum of the THz pulse. Hence, for its measurement lock-in technique has to be used. For that purpose the pump beam is mechanically chopped at 1.35 kHz frequency. By scanning the "open" state of the detector in time the THz pulse shape can be sampled. Utilizing the magnetic field dependence of the THz polarization the THz beam can, in principle, be chopped electrically by means of an alternating magnetic field around the emitter. In order to compare our results with results from other THz emitters we have characterized the generated THz radiation using the maximum dynamic range which is the maximum spectral amplitude above noise level and the maximum frequency above noise level. The last quantity, of course, grows with measurement time up to a maximum measurement time limited by the stability of the whole system. For our measurements we have used a scan range of 33 ps with a step width of 0.4 μm of our delay line, a scan speed of 30 μm/s and an integration time of 100 ms. We have measured the THz amplitude as a function of pump power for a fixed spot size of the pump beam. With a spot size of 5 μm radius, a repetition rate of 75 MHz of our pump laser and a maximum average output power of 350 mW we changed the pump fluence on the sample from 0.5 mJ/cm2 to approximately 5 mJ/cm2 with- out any damage of the sample. The slight onset of saturation of the THz signal may be caused by a temperature increase of the sample (see Fig. 6). Sample preparation-Epitaxial growth. Fe thin films were grown epitaxially on the 0.5 mm-thick MgO (100) and Al2O3 (0001) substrates by molecular beam epitaxy (MBE) technique in an ultrahigh vacuum (UHV) 6 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9Figure 7. Transmittance, reflectance (measured) and absorptance (calculated) of the Fe/Pt emitters grown on MgO substrates. chamber with a base pressure of 3 × 10−10 mbar. The growth rate was R = 0.05 Å/s controlled by a quartz crys- tal during the deposition procedure. The incident Fe beam was perpendicular to the substrates. The cleaning protocol of the MgO (001) 1 × 1 cm2 substrates involved heating at 600 °C for 1 hour. The deposition of Fe was performed at 300 °C temperature. At a next growth stage, on top of the Fe layer, a Pt layer was deposited at the same temperature. The thickness of the individual layers has been varied for Fe (1–12 nm) and for Pt (1–18 nm) as it was monitored in-situ by a calibrated quartz crystal oscillator and confirmed ex-situ by X-ray reflectivity (XRR) measurements. X-rays diffraction (XRD) measurements have shown for the thicker samples (Fe and Pt thicknesses greater than 2 nm) the presence of Fe (002) peaks revealing the epitaxy of Fe layer on the MgO sub- strate, with Fe(001)//MgO (011) for all samples. Diffraction peaks of Pt (200) and Pt (400) arising from the Pt top layer parallel to the Fe (200) planes were observed revealing the epitaxial relation of Pt on Fe. The growth of the fcc Pt layer on bcc Fe along the [100] plane direction is correlated to the Bain epitaxial orientation23,24 when the Pt cell is 45 degrees rotated with respect to the Fe lattice. The structural quality of the Fe/Pt bilayers was further investigated by measuring in-plane XRD polar plots and selectively transmission electron microscopy (TEM). Both methods confirmed the epitaxial quality of the bilayers. The growth of Fe on sapphire (0001) is not perfect epitaxial because of a larger mismatch in lattice constants. This may also influence the THz generation. In our experiments we have observed a weaker spectral amplitude at the maximum frequency whereas there is no obvi- ous absorption at higher frequencies. Theoretical model. The charge current jc in Pt responsible for THz generation depends on the generated spin polarization in Fe, the diffusion of spin polarized electrons in Fe and the diffusion of spin polarized electrons in Pt which are converted into a charge current via the inverse spin Hall effect. The first term in equation 1 takes into account the absorption in the metal layers when the sample is pumped from the metal layer side which reduces the power available for generation of spin polarized electrons in Fe. As only spin polarized electrons within a certain distance from the boundary between Fe and Pt will reach the Pt layer, only a fraction of the measured total absorbed power contributes to the generated THz signal. This fraction scales with the inverse of the total metal layer thickness 1/(dFe + dPt) assuming a constant diameter of the pump beam and an almost linear dependence of the absorbed power on layer thickness. ∝ j c P abs + d d (2) Both assumptions are valid in our case for layer thicknesses investigated in our experiments. Typical absorp- Fe Pt tion measurements are shown in Fig. 7 using P , in P trans = T / A = R = P abs / P in P refl / P , in R T (3) Where Pin is the incoming laser power focused on to the emitter, Ptrans is the laser power transmitted by the emit- ter and Pabs the laser power absorbed by the emitter. To determine the absorptance A we have measured R and T and use A 1= − − . It can be seen that for small metal thicknesses the absorptance dominates whereas for larger metal layer thicknesses the reflection is dominant. The very steep onset of THz radiation at very thin Fe layer thicknesses above a certain minimum Fe thickness is described by the second term in equation 1. It describes the development of spin polarization starting at a layer thickness above a certain "dead layer" d0 and increases with a certain constant λpol which is characteristic for the saturation of spin polarization with layer thickness (see e.g. Allenspach et al.)25. In order to support this assumption we have studied the magnetic properties of our sample series with the help of longitudinal magneto-optic magnetometry. The results are illustrated in Fig. 8. The variation of hysteresis loop characteristics as a function of Fe layer thickness demonstrates that the samples with thickness larger than 1.5 nm have the easy axis of the magnetization in-plane. The sample with 0.8 nm (2.8 monolayers (ML)) has a very 7 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9Figure 8. Magneto-optical Kerr effect measurements of the Fe/Pt emitters with Fe thicknesses ranging from 0.8 to 5.0 nm and constant Pt thickness of 3 nm. small signal and exhibits a loop that resembles a superparamagnetic behaviour of Fe indicative that the sample could be close to its Curie point. The latter is in agreement with other references about the Curie temperature of 2–3 ML of Fe layer thickness on different substrates26. In such case the "dead" magnetic layer can be estimated to be below 0.8 nm. This is also in agreement with results obtained by Allensbach et al.25 where for thicknesses up to 2 ML of Fe on Cu (001) no spin polarization was detected. For the case of a few ML of Fe there is also the possibility that the spin polarization has a perpendicular component26. Unfortunately, we have no experimental access to measure the out-of-plane component of the magnetization. However, the presence or non-presence of perpendicular magnetization is not going to change our results since our THz signal is sensitive to the in-plane component of the magnetization. Regardless of the reason (either close to Tc or no in-plane component due to the perpendicular orientation of the magnetization) the MOKE data justify our assumption of a "dead" magnetic layer for the THz emission below 0.8 nm of Fe. The third and forth term, the tangent hyberbolic function divided by the total impedance, refers to the spin accumulation in Pt which is responsible for the strength of the THz radiation and it depends on the finite diffu- sion length (λPt) of the spin current in Pt. This part of the equation is according to the theory of generation of THz radiation from a current source based on the inverse spin Hall effect8,17,18. by a complex wave vector k The last term accounts for losses in the metal layers. It can be described by a single exponential attenuation factor under the following assumptions: Deriving the generated THz signal via the ISHE using Green's function only negligible losses were assumed and a real wave vector k was used for the THz propagation in the metal layers (see e. g. Seifert et al.)8. Taking into account small losses for the generated THz radiation and replacing the real wave vector = ω⋅ results in an additional attenuation k c z factor − ⋅ where z is the propagation distance in metal, κ the imaginary part of the index of refraction, ω the angular frequency, and c the vacuum speed of light, respectively. Assuming only small losses the attenuation fac- tor is independent from z and the E field is constant across the metal layer if the thickness is small compared to typical attenuation lengths. Following the derivation of Seifert et al8. this factor becomes a constant factor which is independent from other parameters and only depends on the total thickness of the metal layers. The smaller the layer thicknesses the better is this approximation. Due to multiple reflections in the metal (accounted for by the third term in equation 1) the effective propagation length in metal is larger than the sum of the layer thicknesses of Pt and Fe and depends on the finesse of the metal Fabry-Perot structure. As this value is not known exactly an effective inverse absorption constant sTHz was used for the fit and the propagation length was set to the sum of Pt and Fe layer thicknesses. This value is smaller than the real inverse absorption constant in metal as experimentally determined by Yasuda and Hosako27 (typically sTHz = 150 nm for gold averaged over the spectral range from 0.5 THz to 3 THz). To get the real attenuation length our fitted effective inverse absorption constant has to be multiplied by the finesse. i − ⋅ ⁎ n ω n ω ⋅ c ⁎ = κ ω ⋅ c κ ω⋅ c e n ⋅ c = 8 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9The influence of the substrate is neglected in our model as the attenuation of the THz radiation in the substrate is constant for different layer thicknesses and we are discussing only the dependence on Fe/Pt layer thicknesses. The influence of the substrate can be seen in the spectral amplitude. Above 3 THz there is a strong THz absorption which reduces the THz bandwidth. There is of course a different THz absorption in different substrates which has to be taken into account when comparing different substrates. References 1. Hoffmann, A. Spin hall effects in metals. IEEE Transactions on Magn. 49, 5172–5193 (2013). 2. Sinova, J., Valenzuela, S. O., Wunderlich, J., Back, C. H. & Jungwirth, T. Spin hall effects. Rev. Mod. Phys. 87, 1213–1260 (2015). 3. Valenzuela, S. O. & Tinkham, M. Direct electronic measurement of the spin hall effect. Nat. 442, 176–9 (2006). 4. Saitoh, E., Ueda, M., Miyajima, H. & Tatara, G. Conversion of spin current into charge current at room temperature: Inverse spin- hall effect. Appl. Phys. Lett. 88, 182509 (2006). 5. Jaworski, C. M. et al. Observation of the spin-seebeck effect in a ferromagnetic semiconductor. Nat Mater. 9, 898–903 (2010). 6. Bauer, G. E. W., Saitoh, E. & van Wees, B. J. Spin caloritronics. Nat Mater. 11, 391–399 (2012). 7. Seifert, T. et al. Efficient metallic spintronic emitters of ultrabroadband terahertz radiation. Nat Photon 10, 483–488 (2016). 8. Kampfrath, T. et al. Terahertz spin current pulses controlled by magnetic heterostructures. Nat Nano 8, 256–260 (2013). 9. Walowski, J. & Münzenberg, M. Perspective: Ultrafast magnetism and thz spintronics. J. Appl. Phys. 120, 140901 (2016). 10. Battiato, M., Carva, K. & Oppeneer, P. M. Superdiffusive spin transport as a mechanism of ultrafast demagnetization. Phys. Rev. Lett. 105, 027203 (2010). (2011). 11. Melnikov, A. et al. Ultrafast transport of laser-excited spin-polarized carriers in Au/Fe/MgO(001). Phys. Rev. Lett. 107, 076601 12. Wu, Y. et al. High-performance thz emitters based on ferromagnetic/nonmagnetic heterostructures. Adv. Mater. 29, 1603031 (2017). 13. Yang, D. et al. Powerful and tunable thz emitters based on the fe/pt magnetic heterostructure. Adv. Opt. Mater. 4, 1944–1949 (2016). 14. Kroell, J., Darmo, J. & Unterrainer, K. Metallic wave-impedance matching layers for broaddband terahertz optical systems. Opt. Express. 15, 6552–6560 (2007). Phys. Rev. B 86, 024404 (2012). 15. Battiato, M. Superdiffusive spin transport and ultrafast magnetization dynamics. Ph.D Thesis, Uppsala Univ. Swed. (2013). 16. Battiato, M., Carva, K. & Oppeneer, P. M. Theory of laser-induced ultrafast superdiffusive spin transport in layered heterostructures. 17. Saitoh, E. Measurement of spin pumping voltage separated from extrinsic microwave effects. J. Phys. Soc. Jpn. 86, 011003 (2017). 18. Ando, K. et al. Inverse spin-hall effect induced by spin pumping in metallic system. J. Appl. Phys. 109, 103913 (2011). 19. Keller, S. et al. Relative weight of the inverse spin-hall and spin-rectification effects for metallic polycrystalline py/pt, epitaxial fe/pt, and insulating yig/pt bilayers: Angular dependent spin pumping measurements. Phys. Rev. B 96, 024437 (2017). 20. Conca, A. et al. Study of fully epitaxial Fe/Pt bilayers for spin pumping by ferromagnetic resonance spectroscopy. Phys. Rev. B 21. Fattinger, C. & Grischkowsky, D. Terahertz beams. Appl. Phys. Lett. 54, 490–492 (1989). 22. Jepsen, P. U., Jacobsen, R. H. & Keiding, S. R. Generation and detection of terahertz pulses from biased semiconductor antennas. J. 134405, 1–6 (2016). Opt. Soc. Am. B. 13, 2424–2436 (1996). Phys. Lett. 103, 162401 (2013). 23. Papaioannou, E. T. et al. Optimizing the spin-pumping induced inverse spin hall voltage by crystal growth in fe/pt bilayers. Appl. 24. Daniel, B. J., Nix, W. D. & Clemens, B. M. Enhanced mechanical hardness in compositionally modulated fe(001)pt(001) and fe(001) cr(001) epitaxial thin films. Thin Solid Films 253, 218–222 (1994). 25. Allenspach, R. & Bischof, A. Magnetization direction switching in fe/cu (100) epitaxial films: temperature and thickness dependence. 26. Vaz, C. A. F., Bland, J. A. C. & Lauhoff, G. Magnetism in ultrathin film structures. Reports on Prog. Phys. 71, 056501 (2008). 27. Yasuda, H. & Hosako, I. Measurement of terahertz refractive index of metal with terahertz time-domain spectroscopy. Jpn. J. Appl. Phys. Rev. Lett. 69, 3385 (1992). Phys. 47, 1632 (2008). Acknowledgements We thank Burkard Hillebrands for his scientific support. E.Th.P. and S.K. acknowledge support from the Deutsche Forschungsgemeinschaft (DFG) through the collaborative research center SFB TRR 173: SPIN+X Project B07 and the Carl Zeiss Foundation. We thank Jörg Lösch from the Institute of surface-analytics (IFOS-Kaiserslautern) for his support with the XRR-XRD measurements. We thank Marco Battiato and Hans Christian Schneider for helpful discussions. Author Contributions G.T. and L.S. conducted the experiments, S.K. and L.S. grew the samples, R.B. and E.Th.P. conceived the experiments, R.B., G.T., S.K. and E.Th.P. analyzed the results. All authors reviewed the manuscript. Additional Information Competing Interests: The authors declare that they have no competing interests. Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Cre- ative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not per- mitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. © The Author(s) 2018 9 www.nature.com/scientificreports/Scientific REPORTS (2018) 8:1311 DOI:10.1038/s41598-018-19432-9
1803.08886	2	1803	2018-03-26T20:34:49	Intercell Wireless Communication in Software-defined Metasurfaces	[ "physics.app-ph" ]	Tunable metasurfaces are ultra-thin, artificial electromagnetic components that provide engineered and externally adjustable functionalities. The programmable metasurface, the HyperSurFace, concept consists in integrating controllers within the metasurface that interact locally and communicate globally to obtain a given electromagnetic behaviour. Here, we address the design constraints introduced by both functions accommodated by the programmable metasurface, i.e., the desired metasurface operation and the unit cells wireless communication enabling such programmable functionality. The design process for meeting both sets of specifications is thoroughly discussed. Two scenarios for wireless intercell communication are proposed. The first exploits the metasurface layer itself, while the second employs a dedicated communication layer beneath the metasurface backplane. Complexity and performance trade-offs are highlighted.	physics.app-ph	physics	Intercell Wireless Communication in Software-defined Metasurfaces 8 1 0 2 r a M 6 2 ] h p - p p a . s c i s y h p [ 2 v 6 8 8 8 0 . 3 0 8 1 : v i X r a Anna C. Tasolamprou∗, Mohammad Sajjad Mirmoosa†, Odysseas Tsilipakos∗, Alexandros Pitilakis∗‡, Fu Liu†, Sergi Abadal§, Albert Cabellos-Aparicio§, Eduard Alarc ´on§, Christos Liaskos∗, Nikolaos V. Kantartzis∗‡, Sergei Tretyakov†, Maria Kafesaki∗¶, Eleftherios N. Economou∗, Costas M. Soukoulis∗k ∗Foundation for Research and Technology Hellas, 71110, Heraklion, Crete, Greece †Department of Electronics and Nanoengineering, Aalto University, P.O. Box 15500, Espoo, Finland ‡Department of Electrical and Computer Engineering, Aristotle University of Thessaloniki, Thessaloniki, Greece §NaNoNetworking Center in Catalonia (N3Cat), Universitat Polit`ecnica de Catalunya, Barcelona, Spain ¶ Department of Materials Science and Technology, University of Crete, 71003, Heraklion, Crete, Greece k Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, Iowa 50011, USA Email: [email protected] Abstract-Tunable metasurfaces are ultra-thin, artificial elec- tromagnetic components that provide engineered and externally adjustable functionalities. The programmable metasurface, the HyperSurFace, concept consists in integrating controllers within the metasurface that interact locally and communicate globally to obtain a given electromagnetic behaviour. Here, we address the design constraints introduced by both functions accommodated by the programmable metasurface, i.e., the desired metasurface operation and the unit cells wireless communication enabling such programmable functionality. The design process for meeting both sets of specifications is thoroughly discussed. Two scenarios for wireless intercell communication are proposed. The first exploits the metasurface layer itself, while the second employs a dedicated communication layer beneath the metasurface back- plane. Complexity and performance trade-offs are highlighted. I. INTRODUCTION Metasurfaces are planar artificial structures which have recently enabled the realization of novel, ultra-thin electro- magnetic (EM) components with engineered response [1], [2]. An abundance of functionalities has been demonstrated [3], [4], including perfect absorption or wavefront manipulation. Obviously, tunability or reconfigurability are highly desirable in this context. Initial studies revolved around achieving global tunability by means of external stimuli (heat, voltage, light) [3]. To add reconfigurability and the ability to host multiple functionalities, recent works have integrated biased diodes within each unit cell so that the response of each unit cell can be tuned locally [5], [6]. A step further towards the compelling vision of intercon- nectable, fully adaptive metasurfaces with multiple concurrent functionalities is the concept of HyperSurFace (HSF) [7]. The HSF paradigm builds upon the description of EM function- alities in reusable software modules. Such software-defined approach allows authorized users to easily change the behavior of the metasurface by sending preset commands. To dissem- inate, interpret, and apply those commands, a HSF requires the integration of a network of miniaturized controllers within the metamaterial structure. This poses several implementation 10 r e v i e c s n a r T ) 2 m m ( a e r A 1 0.1 1 1 Gb s/mmp 2 10 Gb s/mmp 2 100 Gb s/mmp 2 WPAN Chip-scale 10 Data Rate (Gb s)p 100 Fig. 1. Area as a function of the data rate for state-of-the-art transceivers for Wireless Personal Area Networks (WPAN) and chip-scale applications. Data extracted from [15]–[17], [19]–[38] and references therein. and co-integration challenges [8], among which we highlight and focus on the interconnection of the internal controllers. Communication among the controllers of a HSF can be either wired or wireless. A priori, wired means are preferable as the interconnect will be most likely co-integrated with the controllers within the same chip [8] and because knowledge from similar scenarios like low-power embedded systems can be reused [9], [10]. However, issues may appear when scaling the HSF in size or in controller density: in the former case, HSFs will contain multiple chips leading to complex layout issues related to combining on-chip and off-chip interconnects; in the latter case, HSFs will integrate very dense networks leading to higher latency and power consumption if conven- tional NoC topologies are used [11]. Wireless intercell communication becomes a compelling alternative in either large or dense HSFs. The use of a shared medium allows to reduce the latency and power of collective and long-range communications used during command dis- semination. Also, the lack of wiring between nodes facilitates off-chip and even off-HSF communication. This approach is possible due to recent advances in on-chip antennas in mmWave and THz bands [12]–[14], as well as the constant miniaturization of RF transceivers for short-range applications. As shown in Fig. 1, transceivers with multi-Gbps speeds and footprints as small as 0.1 mm2 have been demonstrated. Before assessing the potential applicability of existing transceivers, it is crucial to understand the EM propagation within this new enclosed and monolithic scenario. Some works have studied propagation in applications with metallic enclo- sures, but provided little room for co-design [39]–[41]. Others have investigated propagation within a computing package [42]–[44], but the structure differed considerably from HSFs. This paper performs, for the first time, a study towards the characterization of the wireless channel within a software- defined HSF. To this end, we describe two possible EM propagation paths in Sect. II, namely, through the metasurface layer or in a dedicated waveguide. We then analyze the field distribution and coupling between mmWave antennas for both cases in Sect. III and IV. Finally, Sect. V concludes the paper. II. STRUCTURE, ENVIRONMENT DESCRIPTION AND ELECTROMAGNETIC OPERATIONS As a case study we consider the software-defined HSF depicted in Fig. 2. The metasurface (MS) part consists of an array of electromagnetically thin metallic patches placed over a dielectric substrate back-plated by a metallic layer. To enable the software-based MS control, the patches are connected to a group of controller chips that lie below the metallic back plane through vertical vias. The controllers adjust the electro- magnetic behaviour of the metasurface fabric by attributing additional local resistance and reactance at will [5], [6]. The controller plane is decoupled from the MS thanks to the back plane that separates the patches from the chips. We assume at this point that each chip serves four metallic patches. Our case study MS is designed for perfect absorption and anomalous reflection operation in the microwave regime. For operation in the microwave regime, the size of the metasurface is required to be in the order of millimetres. Specifically the reference MS structure under consideration is designed to operate at f = 5 GHz (λ0 = 60 mm). It consists of periodically arranged, four- patch unit cells with xy size D × D = 12 mm × 12 mm, as seen in Fig. 2. The size of each patch is w × w = 4.2 mm × 4.2 mm. The thickness of the substrate is h = 1.575 mm and it is made of Rogers RT/Duroid 5880 with permittivity ǫr = 2.2 and loss tangent tan δ = 9 × 10−4. The physical landscape of the software-defined HSF offers several opportunities for the propagation of RF signals within the structure for wireless connectivity between the different controllers. The actual implementation depends on the tile lateral dimensions and the targeted wavelength. In this work we consider two distinct communication channels, seen in Fig. 2(d,e). The first channel is the space between the MS patches and the back plane, called MS layer (scenario A). A blind via fed form the chip serves as the antenna, while the metallic patches and the metallic back plane acts as a waveguide. The second channel is a dedicated communication plane formed by adding extra metallic plates below the chip (scenario B). Monopoles fed from the chip are inserted in the parallel-plate waveguide and excite waves that propagate within this obstacle-free environment. Note, that in both sce- narios the wave propagates in a restricted waveguide which could be considered a wire and not a free-space environment. However, features such as the probes omnidirectional radia- tion, the gap leakage in scenario A and the multi-scattering in scenario B are closely related to free-space wave propagation; hence we adopt the wireless term. The selected communication path may give rise to some undesired phenomena, such as ra- diation losses or interference, but, on the other hand, provides enhanced design opportunities and functionalities. To ensure that the electromagnetic response of the MS and the wireless communication operation are decoupled, we choose the communication frequency to be greater than 25 GHz. This decoupling is especially important in scenario A where the metasurface layer hosts both the electromagnetic waves for the MS operation as well as the communication signals. Therefore, overall, we investigate the channel commu- nication in the range f = [25 GHz, 200 GHz]. The distance between two neighbouring nodes equals D and is in the order of 5λ to 40λ, respectively; this means that the communication takes place in the near and intermediate field regime. Thus, unable to resort to simplified farfield manipulation, we use full wave electromagnetic analysis for the numerical investigation. i.e., for frequencies f > 1 THz For higher frequencies, (D > 200λ) the full wave analysis becomes cumbersome and we need to turn to simplified schemes such as ray tracing [45]. It is stressed that even though we perform the analysis for the reference case dimensions, a direct scaling of the structure along with the wavelengths of operation is possible as long as the properties of the materials involved remain the same. (a) w D (b) (c) h Copper Rogers 5880 Chip θ kinc (d) Com Scenario A (e) Com Scenario B Fig. 2. HSF unit cell: (a) Top-view and geometric parameters, (b) bottom-view with chip for the programmable operation. (c) MS operating at 5 GHz under oblique incidence. (d,e) Unit-cell side-view illustrating the two communication channels. (d) Scenario A: communication in the metasurface substrate (e) Scenario B: communication in a dedicated parallel-plate waveguide. III. CELL TO CELL COMMUNICATION IN THE METASURFACE LAYER The MS layer communication channel of the software- defined HSF is shown in Fig. 2(d). For efficient communica- tion, the electromagnetic energy should be confined between the periodic copper patches and the ground; the waves should not leak to the free-space above. This leakage is a path loss for the communication channel and should be minimized. Our study will be focused on two neighbouring unit cells with respect to the maximum power which can be transmitted from one cell to the other one. The antenna is connected to the chip and located under the center of one of the patches through a cylindrical hole that isolates it form the back plane (ground). The height of the probe antenna is L = 1.4 mm. Due to the presence of the ground, this probe may operate as a quarter- wavelength monopole antenna; however, the complex environ- ment of the MS is expected to affect the antenna operation. The corresponding frequency is f0 = c0/(4L√εr) ≈ 36 GHz. Ideally, i.e., in the absence of the copper patches, an antenna resonance (zero reactance) is expected at this frequency. The waveguide port feeding this antenna was designed to match the theoretical λ/4 monopole input impedance, without any additional optimization for the actual structure. We evaluate the neighbouring nodes communication by calculating the corresponding scattering matrix. To ensure that the MS and the communication operations are electromag- netically decoupled, we assume that the frequency is greater than 30 GHz (the MS operational frequency is 5 GHz). To minimize the free-space leakage, the gap between the patches, wgap, should be electromagnetically small. In our case study the gap is equal to wgap = 1.8 mm, therefore the absolute upper bound in the studied frequency range should not exceed 100 GHz (λ0 =3 mm). Hence, we simulate our structure in the frequency range f =[30 GHz,100 GHz]. We employ ANSYS HFSS, a commercial, 3D full-wave simulator based on the finite element method (FEM). To evaluate the communication between the antennas we calculate the transmission coefficient S21 (dB) which corresponds to the power fraction collected by the receiver. In addition, we calculate the reflection S11 coef- ficient which reveals the power fraction reflected back to the emitter. For optimum operation the magnitude of S11 should be low, meaning negligible reflection, and the magnitude of S21 should be high. S11 can be improved by employing an external matching circuit so we focus here on S21. (a) S11(dB) 0 -5 -10 -15 0 -10 -20 -30 (b) 100 zGH Receiver Transmitter (c) 38 zGH + x z _ Re{ }Ez S21(dB) -20 30 40 50 60 70 80 90 100 Frequency ( z)GH -40 Receiver 12 mm Transmitter (a) Scattering components S11 and S21. Dashed and solid curves Fig. 3. correspond to the initial and optimized structure, respectively. Ez component at (b) f = 100 GHz and (c) f = 38 GHz for the optimized structure. Figure 3(a) presents the transmission and reflection coef- ficients for the present structure under study (dashed lines). As observed, S21 is smaller than -20 dB after 45 GHz. is larger However, around the frequency of 40 GHz, than -20 dB which is acceptable from the point of view of communication. S11 has local minima at approximately 40 GHz, 75 GHz and 95 GHz. However, the transmission it coefficient S21 is maximum at 40 GHz. Thus, we adopt this frequency for wireless intercell communication. Since the low reflection coefficient does not necessarily correspond to a high transmission coefficient, we focus on the environment effect and the free-space leakage. This can be seen in Fig. 3 by comparing the S-parameters at 40 GHz and, at 75 GHz and 95 GHz. At the high frequencies, 75 GHz and 95 GHz, the low-reflected wave radiates into the free space rather than coupling to the receiving antenna. To improve the com- munication between the transmitting and receiving antennas, we optimize the geometry parameters of the structure. We keep in mind that any geometrical modification is going to affect the MS operation, shifting the resonance frequency at higher or lower values. However, if the modifications are moderate, we can readjust the MS resonance at 5 GHz by tuning the resistance and reactance values of the chip. The way to minimize the free-space leakage is by decreasing the gap between the patches wgap. Additionally we can increase the substrate thickness. Finally we select the modified parameters that optimize the communication operation; the optimum patch gap is wopt gap = 1 mm and the optimum thickness is hopt = 2.6 mm. The corresponding S-parameters are shown as solid lines in Fig. 3(a). As can be seen, S21 is significantly improved in the range f =[30 GHz, 40 GHz] (the local maximum is now -15 dB). Notice that at the same frequency range the reflection coefficient is also improved compared to the initial structure. Above 40 GHz the communication efficiency decreases, similarly to the initial structure, but remains, on average, higher than before. The distribution of the electric field Ez is shown in Fig. 3(b) and Fig. 3(c) at frequencies f = 100 GHz and f = 38 GHz, respectively. At f = 100 GHz there is significant leakage whereas at f = 38 GHz the field is confined within the MS layer. This agrees with the increased S21 coefficient at f = 38 GHz. IV. COMMUNICATION IN A DEDICATED PARALLEL PLATE WAVEGUIDE In this scenario we consider that the communication in the software-defined HSF is enabled by an additional channel, dedicated solely to transferring the signals between the com- munication nodes, Fig. 2(e). The channel is created by intro- ducing an additional metallic plate behind the chip backplane at a distance that, as explained, is specified by the desired frequency of operation. We assume that the space between the two metallic plates is empty (air). The two metallic plates and the uniform dielectric space between them, form a parallel- plate waveguide. Each node consists of a probe antenna connected to the chip through a vertical small hole in the ground plane, as seen in Fig. 2(e). The communication channel is totally electromagnetically isolated form the MS layer, thus all coupling is excluded. Moreover, the parallel plates create a closed space where no energy leakage is allowed (the holes are electromagnetically small). For these reasons, this option offers robustness and design flexibility. The parallel-plate waveguide sustains the propagation of TEM (Transverse ElectroMagnetic) waves in which both the electric and magnetic fields are perpendicular to the propa- gation direction. The TEM mode can be excited from zero frequency (DC) and is the only propagation mode supported by the waveguide up to the cut-off frequency of the first higher-order mode: f < c0/(2d). The probe acts as an omni- directional antenna that transmits or receives electromagnetic energy omnidirectionally in the horizontal plane xy. In the vertical plane, the radiation is confined by the metallic plates. (a) T 1 6 11 16 21 2 7 12 17 22 3 8 13 R 18 23 (b) f=25GHz f=60GHz f=180GHz 4 9 5 10 f=25GHz f=60GHz f=180GHz 14 15 19 20 (c) . 24 25 (d) ) B d ( N M S n o i s s i m s n a r T −5 i −15 −25 25 −5 iii −15 −25 25 −5 v −15 node 1 to 21 E z 2 max −5 ii 0 node 1 to 6 50 100 150 200 −15 −25 25 50 100 150 200 node 13 to 1 −5 iv node 1 to 25 50 100 150 200 node 13 to 23 −15 −25 25 50 100 150 200 −5 vi −15 node 7 to 17 −25 25 50 100 −25 25 150 200 50 Frequency (GHz) 100 150 200 Fig. 4. (a) Schematic of the TEM parallel waveguide 2D approximation, node no.1 radiates and node no.13 receives. (b) and (c) Electromagnetic energy distribution at f = 25GHz, f = 60 GHz and f = 180 GHz when the emitter is no.1 and no.13 respectively. (d) Power received at the node M when node N radiates, SM N , over the frequency range f =[25 GHz, 200 GHz]. Six cases of M N node pairs are schematically depicted in the insets. Since the EM energy is carried by the single TEM mode, the waveguide is naturally impedance matched with free-space; this allows the following approximation: We consider that the propagation in the 3D waveguide can be approximated by a 2D analogue where the monopoles are replaced by finite-size conducting scatterers, placed at the vertical positions of the antenna probes. Each scatterer radiates 2D cylindrical waves in the surrounding space and diffracts the energy coming from the environment. The field radiated from the emitter and the diffracted field from the scatterers interfere creating destructive or constructive patterns in the waveguide. By performing a full-wave numerical analysis via the commercial software COMSOL Multiphysics [46], we calculated the total field in each position and frequency. The 2D approximation allows us to solve for large areas and frequency spans in a relatively short time and provides us with a qualitative evaluation of the propagation properties in a multiscattering environment. A priori, we assume that the antennas are impedance matched in all the spectrum of interest and that only the TEM mode is excited, both effectively controlled by the height of the structure. We investigate the system of 25×25 nodes depicted in Fig. 4(a). Each antenna (scatterer) is a finite size copper cylinder of radius R = 0.12 mm. In this approximation we do not take into account the impedance characteristics of the antennas. The emitter is simulated as a field source that radiates omnidirectional electromagnetic waves. All the surrounding scatterers reflect the incoming wave. In this way we estimate the energy profile of the propagating waves in the presence of the reflecting obstacles. Fig. 4(b,c) present the profile of the total energy at frequency f = 25 GHz, f = 60 GHz and f = 180 GHz when the emitter is no.1 and no.13, respectively. Evidently, the electromagnetic waves interfere either destructively or constructively producing patterns of high or low energy corresponding to the dark and bright spots. In the position of the receiver we also estimate the power captured by the multipath propagation coming from all directions. The total power accumulated in the position of the receiver M when N emits, PM N , is normalized by the total radiated power from the emitter P0. The system is reciprocal, that is, SM N = SN M . Fig. 4(d) presents the power received in the position M transmitted from emitter N over the frequency range of f =[25 GHz, 200 GHz] for node pairs schematically depicted in the insets. As observed in all cases, the received power remains on average the same for each pair in the entire frequency span. However, for nearly all cases, there are some frequency points where the received power drops. For example, for the case of the pair no7-no.17 (panel vi) there appear three dips in the received power at around f = 45 GHz, f = 80 GHz and f = 115 GHz. These points correspond to destructive wave interference. Moreover we can observe the general tendency of the decreased received power with respect to the node-pair distance,i.e., for the pair no.1-no.21 (panel i) the average received power is -15 dB whereas for the pair no.1-no.6 (panel ii), the received power is on average -8 dB. Using this 2D qualitative analysis as a guideline, we can select the operation frequency for the actual 3D implementation of the wireless communication channel in the software-defined HSF. V. CONCLUSION In conclusion, we have addressed the issue of intercell wireless communication in the complex environment of a functional, software-defined metasurface. We have focused on two different scenarios with the communication taking place either in the metasurface plane or inside a dedicated channel. In both cases, we have assessed the performance by evaluating the electromagnetic field in the structure and calculating the scattering parameters between transmitting and receiving antennas. After careful design, we have obtained a transmission efficiency of -15dB and -8dB for scenarios A and B, respectively. We have thus demonstrated efficient wireless intercell connectivity without interfering with the metasurface operation taking an essential step towards realizing adaptive hypersurfaces with fully reconfigurable functionalities. ACKNOWLEDGMENT This work was supported by the European Unions Horizon 2020 research and innovation programme-Future Emerging Topics (FETOPEN) under grant agreement No 736876. REFERENCES [1] C. M. Soukoulis and M. Wegener, "Past achievements and future chal- lenges in the development of three-dimensional photonic metamaterials," Nat. Photonics, vol. 5, no. 9, pp. 523–530, 2011. [2] S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S. Kivshar, and C. R. Simovski, "Metasurfaces From microwaves to visible," Physics Reports, vol. 634, pp. 1–72, 2016. [3] H.-T. Chen, A. J. Taylor, and N. Yu, "A review of metasurfaces physics and applications," Reports on Progress in Physics, vol. 79, no. 7, p. 076401, 2016. [4] H.-H. Hsiao, C. H. Chu, and D. P. Tsai, "Fundamentals and applications of metasurfaces," Small Methods, vol. 1, p. 1600064, 2017. [5] T. J. Cui, M. Q. Qi, X. Wan, et al., "Coding Metamaterials, Digital Metamaterials and Programming Metamaterials," Light Science & Ap- plications, vol. 3, 2014. [6] H. Yang, X. Cao, F. Yang, J. Gao, et al., "A programmable metasurface with dynamic polarization, scattering and focusing control," Sci Rep, vol. 6, p. 35692, 2016. [7] C. Liaskos, A. Tsioliaridou, A. Pitsillides, et al., "Design and Devel- opment of Software Defined Metamaterials for Nanonetworks," IEEE Circuits Syst. Mag., vol. 15, no. 4, pp. 12–25, 2015. [8] S. Abadal, C. Liaskos, A. Tsioliaridou, et al., "Computing and Commu- nications for the Software-Defined Metamaterial Paradigm A Context Analysis," IEEE Access, vol. 5, pp. 6225–6235, 2017. [9] T. Bjerregaard and S. Mahadevan, "A survey of research and practices of Network-on-chip," ACM Computing Surveys, vol. 38, no. 1, pp. 1–51, 2006. [10] D. Bertozzi, G. Dimitrakopoulos, J. Flich, and S. Sonntag, "The fast evolving landscape of on-chip communication," Design Automation for Embedded Systems, vol. 19, no. 1, pp. 59–76, 2015. [11] J. Balfour and W. J. Dally, "Design tradeoffs for tiled CMP on-chip networks," in Proceedings of the ICS '06, 2006, p. 187. [12] O. Markish, B. Sheinman, O. Katz, et al., "On-chip mmWave Antennas and Transceivers," in Proceedings of the NoCS '15, 2015, p. Art. 11. [13] F. Gutierrez, S. Agarwal, K. Parrish, and T. S. Rappaport, "On-chip integrated antenna structures in CMOS for 60 GHz WPAN systems," IEEE J. Sel. Areas Commun., vol. 27, no. 8, pp. 1367–1378, 2009. [14] H. M. Cheema and A. Shamim, "The last barrier On-chip antennas," IEEE Microw. Mag., vol. 14, no. 1, pp. 79–91, 2013. [15] S. Abadal, M. Iannazzo, M. Nemirovsky, et al., "On the Area and Energy Scalability of Wireless Network-on-Chip A Model-based Benchmarked Design Space Exploration," IEEEACM Trans. Netw., vol. 23, no. 5, pp. 1501–13, 2015. [16] S. Geng, D. Liu, Y. Li, et al., "A 13.3 mW 500 Mbs IR-UWB transceiver with link margin enhancement technique for meter-range communications," IEEE J. Solid-State Circuits, vol. 50, no. 3, pp. 669– 678, 2015. [17] J. Pang, S. Maki, S. Kawai, et al., "A 128-QAM 60GHz CMOS transceiver for IEEE802.11ay with calibration of LO feedthrough and IQ imbalance," Proceedings of the ISSCC '17, vol. 60, pp. 424–425, 2017. [18] S. Kawai, R. Minami, Y. Tsukui, et al., "A Digitally-Calibrated 20-Gbs 60-GHz Direct-Conversion Transceiver in 65-nm CMOS," in Proceed- ings of the RFIC '13, 2013, pp. 137–140. [19] K. K. Tokgoz, S. Maki, S. Kawai, et al., "A 56Gbs W-band CMOS wireless transceiver," Proceedings of the ISSCC '16, vol. 59, pp. 242– 243, 2016. [20] R. Wu, S. Kawai, Y. Seo, et al., "A 42Gbs 60GHz CMOS transceiver for IEEE 802.11ay," Proceedings of the ISSCC '16, vol. 59, pp. 248–249, 2016. [21] K. Okada, R. Minami, Y. Tsukui, et al., "A 64-QAM 60GHz CMOS transceiver with 4-channel bonding," Proceedings of the ISSCC '14, vol. 57, pp. 346–347, 2014. [22] N. Dolatsha, B. Grave, M. Sawaby, et al., "A compact 130GHz fully packaged point-to-point wireless system with 3D-printed 26dBi lens antenna achieving 12.5Gbs at 1.55pJbm," Proceedings of the ISSCC '17, vol. 60, pp. 306–307, 2017. [23] A. Siligaris, O. Richard, B. Martineau, et al., "A 65-nm CMOS fully integrated transceiver module for 60-GHz wireless HD applications," IEEE J. Solid-State Circuits, vol. 46, no. 12, pp. 3005–3017, 2011. [24] C. W. Byeon, C. H. Yoon, and C. S. Park, "A 67-mW 10.7-Gbs 60- GHz OOK CMOS transceiver for short-range wireless communications," IEEE Trans. Microw. Theory Tech., vol. 61, no. 9, pp. 3391–3401, 2013. [25] H. J. Lee, J. G. Lee, C. J. Lee, et al., "High-speed and low-power OOK CMOS transmitter and receiver for wireless chip-to-chip communica- tion," Proceedings of the MTT-S IMWS-AMP '15, pp. 1–3, 2015. [26] M. Fujishima, M. Motoyoshi, K. Katayama, et al., "98 mW 10 Gbps wireless transceiver chipset with d-band COMS circuits," IEEE J. Solid- State Circuits, vol. 48, no. 10, pp. 2273–2284, 2013. [27] S. Foulon, S. Pruvost, D. Pache, et al., "A 140 GHz multi-gigabits self- heterodyne transceiver for chip-to-chip communications," Proceedings of the EuMC '14, pp. 901–904, 2014. [28] D. Fritsche, P. Starke, C. Carta, and F. Ellinger, "A Low-Power SiGe BiCMOS 190-GHz Transceiver Chipset With Demonstrated Data Rates up to 50 Gbits Using On-Chip Antennas," IEEE Trans. Microw. Theory Tech., vol. 65, no. 9, pp. 3312–3323, 2017. [29] S. Moghadami, F. Hajilou, P. Agrawal, and S. Ardalan, "A 210 GHz Fully-Integrated OOK Transceiver for Short-Range Wireless Chip-to- Chip Communication in 40 nm CMOS Technology," IEEE Transactions on Terahertz Science and Technology, vol. 5, no. 5, pp. 737–741, 2015. [30] N. Sarmah, J. Grzyb, K. Statnikov, S. Malz, et al., "A fully integrated 240-GHz direct-conversion quadrature transmitter and receiver chipset in SiGe technology," IEEE Trans. Microw. Theory Tech., vol. 64, no. 2, pp. 562–574, 2016. [31] S. Kang, S. V. Thyagarajan, and A. M. Niknejad, "A 240 GHz Fully Integrated Wideband QPSK Transmitter in 65 nm CMOS," IEEE J. Solid-State Circuits, vol. 50, no. 10, pp. 2256–2267, 2015. [32] S. V. Thyagarajan, S. Kang, and A. M. Niknejad, "A 240 GHz Fully Integrated Wideband QPSK Receiver in 65 nm CMOS," IEEE J. Solid- State Circuits, vol. 50, no. 10, pp. 2268–2280, 2015. [33] K. Nakajima, A. Maruyama, M. Kohtani, et al., "23Gbps 9.4pJbit 80100GHz band CMOS transceiver with on-board antenna for short- range communication," in Proceedings of the A-SSCC '14, 2014, pp. 173–176. [34] Z. Wang, P. Y. Chiang, P. Nazari, et al., "A CMOS 210-GHz fundamental transceiver with OOK modulation," IEEE J. Solid-State Circuits, vol. 49, no. 3, pp. 564–580, 2014. [35] X. Yu, J. Baylon, P. Wettin, et al., "Architecture and Design of Multi- Channel Millimeter-Wave Wireless Network-on-Chip," IEEE Des. Test. Comput., vol. 31, no. 6, pp. 19–28, 2014. [36] N. Weissman, S. Jameson, and E. Socher, "An F-band dual band 12+12Gbps packaged CMOS bi-directional transceiver," Proceedings of the IMS '16, vol. 2016-Augus, pp. 4–7, 2016. [37] Y. Yang, S. Zihir, H. Lin, O. Inac, et al., "A 155 GHz 20 Gbits QPSK transceiver in 45nm CMOS," Proceedings of the RFIC '14, pp. 365–368, 2014. [38] Y. Kim, S. W. Tam, et al., "A 60-GHz CMOS Transceiver with On-Chip Antenna and Periodic Near Field Directors for Multi-Gbs Contactless Connector," IEEE Microw. Wireless Compon. Lett., vol. 27, no. 4, pp. 404–406, 2017. [39] E. Genender, C. Holloway, K. Remley, et al., "Simulating the Multipath Channel With a Reverberation Chamber Application to Bit Error Rate Measurements," IEEE Trans. Electromagn. Compat., vol. 52, no. 4, pp. 766–777, 2010. [40] M. Ohira, T. Umaba, S. Kitazawa, et al., "Experimental characterization of microwave radio propagation in ICT equipment for wireless harness communications," IEEE Trans. Antennas Propag., vol. 59, no. 12, pp. 4757–4765, 2011. [41] S. Khademi, S. Prabhakar Chepuri, Z. Irahhauten, et al., "Channel Measurements and Modeling for a 60 GHz Wireless Link Within a Metal Cabinet," IEEE Trans. Wireless Commun., vol. 14, no. 9, pp. 5098–5110, 2015. [42] J. Branch, X. Guo, L. Gao, et al., "Wireless communication in a flip-chip package using integrated antennas on silicon substrates," IEEE Electron Device Lett., vol. 26, no. 2, pp. 115–117, 2005. [43] Y. P. Zhang, Z. M. Chen, and M. Sun, "Propagation Mechanisms of Radio Waves Over Intra-Chip Channels With Integrated Antennas Frequency-Domain Measurements and Time-Domain Analysis," IEEE Trans. Antennas Propag., vol. 55, no. 10, pp. 2900–2906, 2007. [44] S. Kim and A. Zajic, "Characterization of 300 GHz Wireless Channel on a Computer Motherboard," IEEE Trans. Antennas Propag., vol. 64, no. 12, pp. 5411–5423, 2016. [45] K. Kantelis, S. Amanatiadis, C. Liaskos, et al., "On the use of fdtd and ray-tracing schemes in the nanonetwork environment," IEEE Commun. Lett., vol. 18, no. 10, pp. 1823–1826, 2014. [46] www.comsol.com
1911.10315	1	1911	2019-11-23T05:18:50	Surface-plasmon-enhanced near-field radiative heat transfer between planar surfaces with a thin-film plasmonic coupler	[ "physics.app-ph" ]	In last decade, there have been enormous efforts to experimentally show the near-field enhancement of radiative heat transfer between planar structures. Several recent experiments also have striven to achieve further enhanced heat transfer with the excitation of coupled surface polaritons by introducing nanostructures on both emitter and the receiver; however, these symmetric structures are hardly employed in real-world applications. Here, we demonstrate substantially increased near-field radiative heat transfer between asymmetric structures (i.e., doped Si and SiO$_2$) by using a thin Ti film as a plasmonic coupler. The measured near-field enhancement at vacuum gap of 380 nm is found to be 3.5 times greater than that for the case without the coupler. The enhancement mechanism is thoroughly elucidated for both polarizations and a dimensionless parameter, which can quantify the coupling strength of the surface polaritons at vacuum, is suggested. As a thin film can be readily used in many engineering applications, this study will facilitate the development of the high-performance engineering devices exploiting the near-field thermal radiation.	physics.app-ph	physics	Surface-plasmon-enhanced near-field radiative heat transfer between planar surfaces with a thin-film plasmonic coupler Mikyung Lim1,†, Jaeman Song2,3,†, Seung S. Lee2, Jungchul Lee2,3, Bong Jae Lee2,3∗ 1. Korea Institute of Machinery and Materials, Daejeon 34103, South Korea 2. Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology, Daejeon 34141, South Korea 3. Center for Extreme Thermal Physics and Manufacturing, Korea Advanced Institute of Science and Technology, Daejeon 34141, South Korea †These authors contributed equally to this work. (Dated: November 26, 2019) Abstract In last decade, there have been enormous efforts to experimentally show the near-field enhance- ment of radiative heat transfer between planar structures. Several recent experiments also have striven to achieve further enhanced heat transfer with the excitation of coupled surface polaritons by introducing nanostructures on both emitter and the receiver; however, these symmetric struc- tures are hardly employed in real-world applications. Here, we demonstrate substantially increased near-field radiative heat transfer between asymmetric structures (i.e., doped Si and SiO2) by using a thin Ti film as a plasmonic coupler. The measured near-field enhancement at vacuum gap of 380 nm is found to be 3.5 times greater than that for the case without the coupler. The enhancement mechanism is thoroughly elucidated for both polarizations and a dimensionless parameter, which can quantify the coupling strength of the surface polaritons at vacuum, is suggested. As a thin film can be readily used in many engineering applications, this study will facilitate the development of the high-performance engineering devices exploiting the near-field thermal radiation. ∗ [email protected] 1 It is well known that radiative heat transfer between two spatially close media can ex- ceed the blackbody limit via tunneling of evanescent waves, which exist exclusively near surfaces [1 -- 4]. This phenomenon, so called near-field radiative heat transfer, has recently drawn enormous attention because of its tunability using nanostructures [5] and its potential applications in thermophotovoltaics (TPV) [6 -- 12], photonic cooling [13], and thermal diode [14, 15]. To develop a high-performance device for those emerging engineering applications, large planar structures separated by sub-micron gap with substantial temperature differ- ence are required. Accordingly, continuous efforts have been made towards measurement of a remarkable heat transfer between planar structures [16 -- 28], mostly with the homogeneous bulk media where radiative heat transfer is readily determined by dielectric function of the medium. Recently, a few groups have successfully demonstrated 'tunable' near-field radiative heat transfer by introducing a monolayer graphene [29 -- 31] and metallo-dielectric multilayers [32] on both the emitter and the receiver surfaces. These planar nanostructures are compatible with the existing experimental platforms [18 -- 28] and are known to change the condition of surface plasmon polaritons supported at the vacuum/emitter and the vacuum/receiver inter- faces [33 -- 38]. Considering that the tunneling of evanescent waves can be notably enhanced by the coupling of surface polaritons at the vacuum/emitter and the vacuum/receiver in- terfaces [35 -- 37], both intensity and spectral distribution of the near-field radiation can be tuned by introducing nanostructures on each surface [35 -- 37, 39]. For example, it was shown that by modifying the configuration of metallo-dielectric multilayers (e.g., thickness of each layer or number of unit cells), the surface plasmon polariton (SPP) conditions of the in- terfaces near vacuum can be tuned, which in turn can lead to the enhanced total radiative heat flux [32]. On the other hand, the near-field radiative heat transfer can also be greatly increased by placing graphene on the surfaces of intrinsic silicons and consequently mak- ing both the emitter and the receiver (i.e., graphene-coated silicons) to support SPPs [29]. Because SPPs generated at the graphene layer can be tuned by its chemical potential, a demonstration of an electronic modulation of near-field radiative heat transfer was also re- ported by applying electronic bias on the graphene layer [30]. Further, the SPPs of graphene can be coupled with the surface phonon polaritons (SPhPs) of SiO2 substrates, such that the colossal enhancement (i.e., ∼ 65 times) over the blackbody limit was achieved with a pair of graphene-coated SiO2 structures [31] by coupling of SPP-SPhPs supported at vac- 2 uum/emitter and vacuum/receiver interfaces. The authors also noted that this significant near-field enhancement over the blackbody radiation is greatly suppressed if the symmetry between the emitter and the receiver is broken. In fact, graphene/SiO2-to-SiO2 structure shows smaller heat transfer than that between SiO2 structures or that between graphene- coated SiO2 structures, because the resonant conditions (i.e., surface polariton conditions) of vacuum/emitter and vacuum/receiver interfaces become hardly matched with the asym- metric structures [31]. Despite all those great advances, researches on tuning of the near-field radiative heat transfer has focused mainly on obtaining a large heat flux by introducing the same materials on both the emitter and the receiver sides (i.e., symmetric configuration) [29 -- 32], which is, however, hard to be achieved in real-world applications [6 -- 8, 10 -- 15]. For example, the near- field TPV system (one of the most promising near-field radiative heat transfer applications) requires a TPV-cell receiver and a selective emitter for a high performance [6, 8, 11, 39]. Even if the same materials are chosen for both sides, temperature-dependent optical (and/or thermophysical) properties of the materials make the overall system to be asymmetric [40]. In this work, we experimentally demonstrate significantly enhanced near-field radiative heat transfer between asymmetric emitter and receiver by introducing a thin metal film as a plasmonic coupler. As shown in Fig. 1(a), doped Si and SiO2 are used as an emitter and a receiver, respectively. Although doped Si and SiO2 are well-known to support SPPs and SPhPs, respectively, their resonant frequencies do not overlap, which makes the coupling of surface polaritons of the emitter and the receiver occurring in the limited frequency range. Thus, there would be little synergetic effect between SPPs (associated with doped Si) and SPhPs (associated with SiO2) for enhancing the near-field heat transfer rate. If a thin Ti film is deposited on the SiO2 side [see Fig. 1(b)], the SPPs generated at vacuum/Ti/SiO2 interfaces can be effectively coupled to SPPs of vacuum/doped Si interface in a wider fre- quency range, and thus, enhances the net radiative heat transfer. The measured heat flux for these two configurations (i.e., with or without Ti-film plasmonic coupler) are well-matched with the theoretical predictions. Further, for the analysis of enhanced heat transfer via p-polarization through coupling of surface polaritons, we suggest a simple dimensionless parameter that is related to the field distribution at the vacuum to quantify the coupling strength of SPPs of the emitter and the receiver. To measure the near-field radiation between two media depicted in Figs. 1(a)-1(b), 3 (a) (c) Doped Si Thin Ti SiO2 Doped Si SiO2 Doped Si Ti SiO2 (b) x z Heater x z (d) Heater Emitter part Receiver part Doped Si Thin Ti SiO2 FIG. 1. Schematics of configurations of (a) doped-Si emitter and bare-SiO2 receiver (i.e., without plasmonic coupler) and (b) doped-Si emitter and 10-nm-Ti-film-coated SiO2 receiver (i.e., with plasmonic coupler). The conceptual field profiles for SPPs (SPhPs) supported at the interfaces are depicted. Schematics for the emitter part and the receiver part of the microdevices for measuring near-field radiation (c) between doped-Si emitter and bare-SiO2 receiver and (d) between doped-Si emitter and Ti-film-coated-SiO2 receiver. MEMS-fabricated microdevices and a custom-built three-axis nanopositioner, which was introduced in the previous work [32], are employed. As described in Figs. 1(c)-1(d), the emitter part of the microdevice is composed of the 800-nm-thick doped-Si layer deposited on one side of a fused-silica substrate and the resistive heater on the opposite side. For the receiver part of the microdevice, two different configurations are employed for comparison: one is bare SiO2 [see Fig. 1(c)] and the other is 10-nm-thick-Ti-film-coated SiO2 [see Fig. 1(d)]. The detailed information on the fabrication process, the schematics, and photographs of the emitter and the receiver parts of the microdevice are provided in Supplemental ma- terial. To precisely estimate the vacuum gap between the emitter and the receiver, four 10-nm- thick Ti electrodes with small area (i.e., 2.6% of total receiver surface for each electrode) 4 are deposited on the bare SiO2 receiver surface in case of the bare-SiO2 receiver [see Fig. 1(c)], while four-segmented 10-nm-thick-Ti films themselves are used as electrodes in case of the 10-nm-thick-Ti-film-coated-SiO2 receiver, as can be seen in Fig. 1(d). Following the previous work [32], local vacuum gaps between doped-Si-electrode of the emitter and each of four thin-Ti-film electrodes of the receiver are estimated by sequentially measuring the electrical capacitances between them. Thus, we can fully quantify the parallelism between the emitter and the receiver, as well as take into consideration the effects of bowing and tilting on the radiative heat flux by using Derjaguin-approximated average vacuum gap [41]. Given that the experimental setup is placed in a high vacuum chamber (< 1 × 10−3 Pa), conduction and convection heat transfers by air can be safely neglected. Figure 2(a) shows a cross-sectional view of the experimental setup and describes the heat flow within the system. During the experiment, the feedback control of the input power (denoted as Qin) to the heater can maintain the temperature of the emitter at the desired value. Qin is then divided into Qe-r (i.e., the radiative heat transfer between the emitter and the receiver) and Qloss that includes the background radiation as well as the parasitic conduction from the heater to the vacuum chamber through the three-axis nanopositioner. Considering that Qe-r is the summation of far-field (Qe-r,far) and near-field (Qe-r,near) contributions, Qin can be expressed as Qin = (Qe-r,near + Qe-r,far) + Qloss. If the temperature of the vacuum chamber is maintained at a constant temperature (e.g., room temperature), Qref = Qe-r,far + Qloss can also be considered as a constant while reducing the gap between the emitter and the receiver. In our experimental condition, it was confirmed that Qref can be regarded as constant within 5 minutes (see Supplementary Fig. S7) and the standard deviation of the Qref is considered in error estimation of the obtained data. Therefore, one cycle of the data acquisition from the vacuum gap of 2200 nm to the vacuum gap where the first local contact between the emitter and the receiver is detected was performed within 5 minutes. To measure the radiative heat flux between the emitter and the receiver with respect to the vacuum gap, we firstly measured Qin ≈ Qref at the vacuum gap of 2200 nm where Qe-r,near is negligible compared to Qe-r, near at the vacuum gap of 380 nm (i.e., the smallest vacuum gap achieved). As the vacuum gap is decreasing, Qin required to maintain the emitter temperature constant is increased due to the contribution of the evanescent mode, i.e., Qin(d) = Qe-r,near(d) + Qref. In this way, we could estimate Qe-r by adding a calculated Qe-r,far to measured Qe-r,near. In the upper panel of Fig. 2(b), it can be clearly seen that the estimated Qe-r increases while the temperature of 5 Picomotor actuator Qloss Heater Qin SiO2 Doped Si Qe-r d Ti SiO2 Heat sink Vacuum chamber (a) (b) d1 d3 d2 d4 FIG. 2. (a) Schematic of the cross sectional view of the microdevice attached to the custom- built nanopositioner. The heat flow from the input power of the heater, Qin into the radiative heat transfer from emitter to receiver, Qe-r and Qloss is depicted. (b) Measured change of Qin to maintain the heater temperature constant while reducing the gap (upper panel) and four measured capacitances and corresponding four local gaps while conducting one cycle of measurement (lower panel). the heater is maintained within ±0.1 K of the designated value. The corresponding vacuum gaps derived from the measured capacitances between capacitor electrodes are shown in the lower panel of Fig. 2(b) and the achieved parallelism (defined as the difference between 6 (a) (b) Symbol: Experiment Lines: Theory 10-nm-Ti-film-coated SiO2 receiver Bare-SiO2 receiver Solid line: 10-nm-Ti-film-coated SiO2 receiver Dashed line: Bare-SiO2 receiver Total p-polarization s-polarization FIG. 3. (a) Measured near-field radiative heat flux between doped-Si emitters and SiO2 with and without 10-nm Ti film. (b) Theoretically predicted value of near-field radiative heat flux for both configurations. Contributions of p- and s- polarization are also plotted. The emitter temperature is set as 430 K while maintaining receiver temperature at 300 K. d1 and d4) is 31.7 nm when the average vacuum gap is 380 nm, which corresponds to the tilting angle of 2.3 × 10−6 rad. Note that compared to the previous work [32], this differential-input-power method is much more straightforward than the heat flux estimation from the temperature differences between two thermistors based on the calibration result. Furthermore, the fabrication process for the receiver part of the microdevice is significantly simplified, because there is no need to integrate thermistors and a calibration heater. 7 The measured radiative heat flux is plotted with respect to the vacuum gap (d) in Fig. 3(a). The doped-Si emitters for both configurations are set to 430 K while the receivers, which are the SiO2 substrates with or without 10-nm Ti film overlaid, are maintained at 300 K. The data plotted in Fig. 3(a) are the values averaged from 10 independent experiments for each case. Although the minimum vacuum gap achieved for the case of bare SiO2 is slightly larger than that for the case with 10-nm-thick Ti film, it can be seen that the measured values excellently agree with the theoretical predictions for both configurations. At d = 380 nm, the measured radiative heat flux for the case with 10-nm-thick Ti film is around 2190 W/m2, which is greater by 1860 W/m2 than that at d = 2200 nm, i.e., Qe-r(d = 380 nm) − Qe-r(d = 2200 nm) = 1860 W/m2. The value of [Qe-r(d = 380 nm)−Qe-r(d = 2200 nm)] achieved with 10-nm Ti film is almost 3.5 times greater than the predicted enhancement for the case with the bare SiO2 receiver. Such a considerable near-field enhancement is attributed to stronger coupling of SPPs confined at vacuum/doped-Si and vacuum/Ti-film/SiO2 interfaces. As can be noted in Fig. 3(b), however, this enhancement not only results from the increase of heat flux via p-polarization (i.e., plasmonic contribution), but also that via s-polarization. Thus, more detailed analysis is needed. The net near-field radiation between the emitter and the receiver can be expressed as [1 -- 4]: q = Z ∞ 0 dω (qp ω+qs ω) = Z ∞ 0 dω Z ∞ 0 Θ(ω, T1) − Θ(ω, T2) π2 β,ω(β, ω) + Z s ×(cid:2)Z p β,ω(β, ω)(cid:3) dβ, (1) where ω stands for the angular frequency and β is the parallel component of the wavevector. For this study, the temperature of the emitter, T1 is 430 K and that of the receiver, T2 is 300 K. Also, Θ(ω, Ti) = exp{ω/(kB Ti)}−1 is the mean energy of the Planck oscillator, where is the Planck constant divided by 2π and kB is the Boltzmann constant. In order to elucidate ω the enhancement mechanism through coupling of surface polaritons, the effect of the Planck distribution Θ(ω, Ti) is sometimes excluded [4, 29, 31, 37], such that the analysis could be conducted based solely on the exchange function Zβ,ω(β, ω). The exchange function can be expressed for both cases with or without 10-nm-thick Ti film as [4]: Z p,s β,ω,prop(β, ω) = Z p,s β,ω,evan(β, ω) = β(1 − rp,s 41 − rp,s 01 2)(1 − rp,s 01 rp,s 02 ei2k0z d2 02 2) βIm(rp,s 01 )Im(rp,s 01 rp,s 02 )e−2Im(k0z)d 02 ei2k0z d2 1 − rp,s , (2) 8 where the expression for propagating and evanescent waves can be used when β < ω/c0 and β > ω/c0, respectively. In above equations, k0z is the normal component of wavevector in vacuum and Im() takes the imaginary part of a complex value. rp,s coefficient at the vacuum/doped-Si interface and rp,s 02 can be the reflection coefficient at the 01 stands for the reflection vacuum/SiO2 interface or the modified reflection coefficient for the vacuum/Ti-film/SiO2 multilayered structure, obtained using Airy's formula [33, 42]. The optical property of Ti film was obtained from [43] including the electron-boundary scattering effect [44, 45] and that of doped Si was taken from [4] by assuming complete ionization at high temperature. The dielectric function of SiO2 reported in [46] was employed. (a) (b) FIG. 4. The spectral heat flux, qγ ω for (a) s-polarization and (b) p-polarization at d = 400 nm. The emitter is set as doped-Si at 430 K while three different structures (i.e., 10-nm-Ti-film-coated SiO2, bare SiO2, and bulk Ti) are employed as a receiver at 300 K. 9 In Figs. 4(a)-4(b), the calculated spectral radiative heat flux for each polarization is depicted when d = 400 nm. For both polarizations, it can be clearly seen that 10-nm- thick-Ti-film-coated SiO2 leads to the greatest heat transfer rate among three cases. In other words, we can see so-called 'thin-film effect' [34, 47] in both polarizations for the configuration with 10-nm-thick-Ti-film-coated SiO2. Let us discuss the enhancement of heat flux via s-polarization first. It is well-known that the heat transfer via s-polarization is dominant for the near-field radiation between metals [48], while that through p-polarization is dominant in heat transfer between a metal and a polar material or between polar materials [33, 49]. Due to high doping concentration of doped Si, the near-field radiative heat transfer between semi-infinite doped Si and semi-infinite (i.e., bulk) Ti is dominated by s-polarization. Thus, by placing thin Ti film, heat transfer through s-polarization can be increased as can be seen in Fig. 4(a). Comparing to the bulk Ti case, the peak in spectral heat flux for the case of 10-nm-thick Ti film is broadened and shifted to a higher frequency due to the increased electron-boundary scattering [32]. Because the spectral near-field energy density above the thin-metal-film-coated surface is maximized at smaller thickness with increasing frequency [47], 10-nm-thick Ti film can even show larger spectral heat flux than bulk Ti at higher frequency, leading to an increase in total heat transfer. Consequently, the Ti-film-coated surface can result in a significant near-field heat transfer in s-polarization compared to that between bulk doped Si and bulk SiO2, and it can even exceed that between bulk doped Si and bulk Ti (e.g., when d > 180 nm, for the configuration defined here). For the near-field radiative heat flux via p-polarization, the quasi-monochromatic spectral radiative heat flux is observed for the bare-SiO2 receiver [see Fig. 4(b)]. This is because the coupling of SPPs supported at the vacuum/doped Si interface and SPhPs supported at the vacuum/SiO2 occurs only in a narrow frequency range. On the other hand, by coating the SiO2 surface with a 10-nm-thick Ti film, a broad spectral enhancement can be achieved, leading to increase in a total amount of heat transfer by p-polarization. This mechanism of heat transfer enhancement is clearly revealed with the exchange function Z p β,ω plotted with respect to the normalized parallel wavevector and the angular frequency [refer to Figs. 5(a)-5(b)]. The dispersion curves for the surface waves bounded in the entire structure can be obtained by neglecting losses of materials [50 -- 52] and shown as green-colored-curves in the figure. 10 FIG. 5. Contour plot of exchange function Z p β,ω with respect to the normalized parallel wavevector and the angular frequency. The green-curves depicting SPP dispersion relations are overlaid. (a) Receiver of bare SiO2. (b) Receiver of 10-nm-Ti-film-coated SiO2. (c) Exchange function Z p β,ω plotted with respect to the longitudinal electric field ratio (i.e., Ex1/Ex2). (d) Transmissivity with respect to the loss rate (i.e., ΓSi/ΓTi) in coupled mode theory. (e), (f ), and (g) The longitudinal electric field profiles at the points D, E, and F. In Fig. 5(a), Z p β,ω has greater values at the frequencies ∼ 2.1×1014 rad/s and ∼ 9.1×1013, which corresponds to the SPhP resonant frequencies of vacuum/SiO2 interface [18, 23]. This 11 increase in Z p β,ω results from the coupling between SPhPs of vacuum/SiO2 interface and photon-like SPP mode of vacuum/doped-Si interface (i.e., near vacuum light line). Although a dispersion curve still appears near the SPP resonant frequency for the vacuum/doped Si interface ( ∼5.4 × 1014), it cannot lead to an enhancement in Z p β,ω, given that SPhP resonant mode for vacuum/SiO2 interface does not exist in such a high frequency. On the other hand, when a thin Ti film is coated on the surface of SiO2, the SPP condition of vacuum/receiver interface is modified such that it can be excited over broad spectral range. As a result, the coupled SPP-SPhPs existing at the vacuum/Ti-film/SiO2 structure can now be interacted with the SPPs of vacuum/doped Si interface [see Fig. 4(b)]. Further, SPP dispersion curves are splited when two SPP dispersion curves of vacuum/doped Si and vacuum/receiver crosses each other near points B and E. This split of SPP dispersion curves has been reported in several previous studies [37, 38, 53] and is responsible for strong enhancement of the heat transfer as also can be seen in Fig. 5(b). To quantify the coupling strength of SPPs, we propose a dimensionless parameter based on the electric field amplitude at the vacuum interfaces. For detailed analysis, six points, indicated as A-F in Fig. 5(b), are selected along the polariton dispersion curves. When the longitudinal electric field component, Ex is considered, the ratio of its amplitude at the vacuum-emitter interface (i.e., Ex1) and the vacuum-receiver interface (i.e., Ex2) can be cal- culated along the surface-wave dispersion curves. In Fig. 5(c), Z p β,ω is plotted with respect to the longitudinal electric field ratio (i.e., Ex1/Ex2) along the dispersion curves. It can be readily seen that Ex1/Ex2 curves are divided into two branches, which corresponds to each of SPP dispersion curves in Fig. 5(b). The sign of Ex1/Ex2 reveals that the SPP dispersion curve including points A-B-C corresponds to the antisymmetric mode of excitation, whereas the branch including points D-E-F is the symmetric mode. More importantly, the magnitude of Ex1/Ex2 can provide additional information about how effectively the surface wave at the emitter-vacuum interface and that at the receiver-vacuum interface are coupled together. If the surface wave at the one of the vacuum interfaces dominates, then the resulting Ex1/Ex2 value will tend toward either Ex1/Ex2 ≪ 1 or Ex1/Ex2 ≫ 1. In Fig. 5(e), for instance, the Ex field distribution at point D is depicted where the field is mostly bounded at the vacuum/receiver interface, resulting in Ex1/Ex2 ≈ 0.16. In contrast, for point F, the field is mainly bounded at the vacuum/emitter interface [see Fig. 5(g)], and the corresponding Ex1/Ex2 value is 3.0. Comparison of Figs. 5(a) and 5(b) reveals that the dispersion curve 12 where point F locates in Fig. 5(b) is actually originated from the SPP dispersion of the vacuum/doped-Si interface, while that of point D is from the SPP dispersion of the vac- uum/receiver interface. Thus, it can be inferred that too low or too high values of Ex1/Ex2 indicate that the SPPs of the vacuum/emitter and the vacuum/receiver interfaces are un- balanced (i.e., weakly coupled). At point E where the maximum Z p β,ω occurs, however, the SPPs from each interface are strongly coupled, such that magnitude of evanescent waves at the emitter and the receiver are in similar range (i.e., Ex1/Ex2 = 1.3 ≈ 1). For the antisymmetric branch [i.e., points A-B-C in Fig. 5(c)], the similar behavior of Ex1/Ex2 can be observed, but the maximum Z p β,ω occurs when Ex1/Ex2 = 0.22. In the previous studies [37, 54, 55], the maximum thermal transmissivity, ξp β,ω (= Z p β,ω/β) is often estimated from the impedance matching condition derived from coupled mode theory. Figure 5(d) shows the transmissivity, ξp β,ω with respect to the ratio of the loss rates (i.e., ΓSi and ΓTi; refer to [54, 56] for details) obtained by activating loss of each layer. As predicted, ξp β,ω increases as two loss rates become closer (i.e., ΓSi/ΓTi approaches to 1). Because the vacuum gap is 400 nm, the maximum transmissivity is bound to 0.12, which can be obtained at the points B and E [see Fig. 5(d)]. Although it is not shown here, the maximum transmissivity can reach 1.0 when d = 100 nm, at which ΓSi/ΓTi ≈ 1. Similar to Ex1/Ex2, the values of ΓSi/ΓTi are divided into two branches. Because the dispersion curve including points B, C, and D originates from the SPP dispersion of the vacuum/receiver interface, ΓTi (i.e., loss to the Ti film) is larger than ΓSi (i.e., loss to doped Si). Similarly, for dispersion curves including points E, F, and A, ΓSi has larger value than ΓTi, because the dispersion curve including points E,F, and A stems from the SPP dispersion curves for vacuum/doped-Si interface. Interestingly, the six points A-F are divided differently for Ex1/Ex2 and ΓSi/ΓTi, providing complemental information. As a matter of fact, Ex1/Ex2 can provide the information on the mode profile of plasmonic resonances, but ΓSi/ΓTi can give the information on the decay rate to each layer for a given resonant mode. These two parameters can be used together to predict the coupling strength of the SPPs and to analyze the optimal configuration maximizing the heat transfer. Finally, it is worthwhile to mention that although strong coupling is observed at points B and E, because of Planck distribution, it cannot result in prominent enhancement in the spectral heat flux shown in Fig. 4(b). Nevertheless, the relationship between Z p β,ω and the field ratio along the dispersion curve is still valid at a lower frequency regime and significant enhancement is expected when the 13 doping concentration of the emitter becomes low. In summary, we have suggested a means to strongly increase the near-field radiative heat transfer between doped Si and SiO2 media by coating SiO2 substrate (i.e., polar material) with a thin Ti film and successfully demonstrated increased near-field heat transfer using a custom-built MEMS-integrated platform. While decreasing vacuum gap from 2200 nm to 380 nm, the enhancement in near-field radiative heat transfer between doped-Si emitter and 10-nm-thick Ti-film-coated SiO2 receiver is measured to be 1860 W/m2 which is 3.5 times larger value than that for the case of doped-Si emitter and bare SiO2 receiver. It was revealed that this thin-metal film can enhance the heat transfer both via p- and s- polarizations. In particular, the heat transfer enhanced via p-polarization results from the coupling of SPPs from vacuum-emitter and vacuum-receiver interfaces in a broad spectral range. This enhancement is predicted with the proposed dimensionless parameter, which is the ratio of the longitudinal electric field at vacuum/emitter and vacuum/receiver interfaces. Considering that thin metal film is compatible with the engineering applications of near- field radiative heat transfer such as a Schottky-junction based near-field TPV system [9, 11], the results obtained in this study will guide the future development of the high-throughput near-field devices. ACKNOWLEDGMENTS This research was supported by the Basic Science Research Program (NRF-2017R1A2B2011192, NRF-2018R1A6A3A01012563 and NRF-2019R1A2C2003605) through the National Re- search Foundation of Korea (NRF) funded by Ministry of Science and ICT. [1] D. Polder and M. Van Hove, Phys. Rev. B 4, 3303 (1971). [2] J.-P. Mulet, K. Joulain, R. Carminati, and J.-J. Greffet, Microscale Thermophys. Eng. 6, 209 (2002). [3] K. Joulain, J.-P. Mulet, F. Marquier, R. Carminati, and J.-J. Greffet, Surf. Sci. Rep. 57, 59 (2005). [4] S. Basu, B. J. Lee, and Z. M. Zhang, J. Heat Transfer 132, 023302 (2010). 14 [5] Y. Tian, A. Ghanekar, M. Ricci, M. Hyde, O. Gregory, and Y. Zheng, Materials 11, 862 (2018). [6] M. Laroche, R. Carminati, and J.-J. Greffet, J. Appl. Phys. 100, 063704 (2006). [7] S. Basu, Y.-B. Chen, and Z. Zhang, Int. J. Energy Res. 31, 689 (2007). [8] K. Park, S. Basu, W. P. King, and Z. M. Zhang, J. Quant. Spectrosc. Radiat. Transfer 109, 305 (2008). [9] R. St-Gelais, G. R. Bhatt, L. Zhu, S. Fan, and M. Lipson, ACS Nano 11, 3001 (2017). [10] A. Fiorino, L. Zhu, D. Thompson, R. Mittapally, P. Reddy, and E. Meyhofer, Nat. Nanotech. 13, 806 (2018). [11] N. Vongsoasup and K. Hanamura, Therm. Sci. Eng. 26, 29 (2018). [12] T. Inoue, T. Koyama, D. D. Kang, K. Ikeda, T. Asano, and S. Noda, Nano Lett. 19, 3948 (2019). [13] L. Zhu, A. Fiorino, D. Thompson, R. Mittapally, E. Meyhofer, and P. Reddy, Nature 566, 239 (2019). [14] A. Fiorino, D. Thompson, L. Zhu, R. Mittapally, S.-A. Biehs, O. Bezencenet, N. El-Bondry, S. Bansropun, P. Ben-Abdallah, E. Meyhofer, et al., ACS nano 12, 5774 (2018). [15] K. Ito, K. Nishikawa, A. Miura, H. Toshiyoshi, and H. Iizuka, Nano Lett. 17, 4347 (2017). [16] C. M. Hargreaves, Phys. Lett. A 30, 491 (1969). [17] G. A. Domoto, R. F. Boehm, and C. L. Tien, J. Heat Transfer 92, 412 (1970). [18] L. Hu, A. Narayanaswamy, X. Chen, and G. Chen, Appl. Phys. Lett. 92, 133106 (2008). [19] R. S. Ottens, V. Quetschke, S. Wise, A. A. Alemi, R. Lundock, G. Mueller, D. H. Reitze, D. B. Tanner, and B. F. Whiting, Phys. Rev. Lett. 107, 014301 (2011). [20] T. Kralik, P. Hanzelka, M. Zobac, V. Musilova, T. Fort, and M. Horak, Phys. Rev. Lett. 109, 224302 (2012). [21] K. Ito, A. Miura, H. Iizuka, and H. Toshiyoshi, Appl. Phys. Lett. 106, 083504 (2015). [22] M. Lim, S. S. Lee, and B. J. Lee, Phys. Rev. B 91, 195136 (2015). [23] B. Song, D. Thompson, A. Fiorino, Y. Ganjeh, P. Reddy, and E. Meyhofer, Nat. Nanotech. 11, 509 (2016). [24] J. I. Watjen, B. Zhao, and Z. M. Zhang, Appl. Phys. Lett. 109, 203112 (2016). [25] M. P. Bernardi, D. Milovich, and M. Francoeur, Nat. Commun. 7, 12900 (2016). [26] M. Ghashami, H. Geng, T. Kim, N. Iacopino, S. K. Cho, and K. Park, Phys. Rev. Lett. 120, 15 175901 (2018). [27] A. Fiorino, D. Thompson, L. Zhu, B. Song, P. Reddy, and E. Meyhofer, Nano Lett. 18, 3711 (2018). [28] J. DeSutter, L. Tang, and M. Francoeur, Nat. Nanotech. 14, 751 (2019). [29] J. Yang, W. Du, Y. Su, Y. Fu, S. Gong, S. He, and Y. Ma, Nat. Commun. 9, 4033 (2018). [30] N. Thomas, M. C. Sherrott, J. J. Brouillet, H. A. Atwater, and A. J. Minnich, Nano Lett. 19, 3898 (2019). [31] K. Shi, Y. Sun, Z. Chen, N. He, F. Bao, J. S. Evans, and S. He, Nano Lett. 19, 8082 (2019). [32] M. Lim, J. Song, S. S. Lee, and B. J. Lee, Nat. Commun. 9, 4302 (2018). [33] S.-A. Biehs, Eur. Phys. J. B 58, 423 (2007). [34] M. Francoeur, M. P. Mengu¸c, and R. Vaillon, Appl. Phys. Lett. 93, 043109 (2008). [35] R. Messina and P. Ben-Abdallah, Sci. Rep. 3, 1383 (2013). [36] M. Lim, S. S. Lee, and B. J. Lee, Opt. Express 21, 22173 (2013). [37] A. Karalis and J. D. Joannopoulos, Sci. Rep. 6, 28472 (2016). [38] H. Iizuka and S. Fan, Phys. Rev. Lett. 120, 063901 (2018). [39] M. Lim, J. Song, J. Kim, S. S. Lee, I. Lee, and B. J. Lee, J. Quant. Spectrosc. Radiat. Transfer 210, 35 (2018). [40] C. J. Fu and Z. M. Zhang, Int. J. Heat Mass Transfer 49, 1703 (2006). [41] B. V. Derjaguin, I. I. Abrikosova, and E. M. Lifshitz, Quart. Rev. Chem. Soc. 10, 295 (1956). [42] P. Yeh, Optical Waves in Layered Media (Wiley New York, 1988) Chap. 4. [43] M. A. Ordal, R. J. Bell, R. W. Alexander, L. A. Newquist, and M. R. Querry, Appl. Opt. 27, 1203 (1988). [44] I. ur Rahman and K. Ahmed, J. Appl. Phys. 49, 3625 (1978). [45] G. Ding, C. Clavero, D. Schweigert, and M. Le, AIP Adv. 5, 117234 (2015). [46] E. D. Palik, Handbook of Optical Constants of Solids, Vol. 1 (Academic press, 1985). [47] S.-A. Biehs, D. Reddig, and M. Holthaus, Eur. Phys. J. B 55, 237 (2007). [48] P.-O. Chapuis, S. Volz, C. Henkel, K. Joulain, and J.-J. Greffet, Phys. Rev. B 77, 035431 (2008). [49] C. J. Fu and W. C. Tan, J. Quant. Spectrosc. Radiat. Transfer 110, 1027 (2009). [50] C. Yeh and F. I. Shimabukuro, The Essence of Dielectric Waveguides (Springer, 2008). [51] J. Ordonez-Miranda, L. Tranchant, T. Tokunaga, B. Kim, B. Palpant, Y. Chalopin, T. Antoni, 16 and S. Volz, J. Appl. Phys. 113, 084311 (2013). [52] J. Ordonez-Miranda, L. Tranchant, Y. Chalopin, T. Antoni, and S. Volz, J. Appl. Phys. 115, 054311 (2014). [53] M. Francoeur, M. P. Mengu¸c, and R. Vaillon, J. Appl. Phys. 107, 034313 (2010). [54] A. Karalis and J. D. Joannopoulos, Appl. Phys. Lett. 107, 141108 (2015). [55] C. Lin, B. Wang, K. H. Teo, and P. Bandaru, Phys. Rev. Appl. 7, 034003 (2017). [56] C. Lin, B. Wang, and K. H. Teo, J. Appl. Phys. 121, 183101 (2017). 17
1907.09998	1	1907	2019-05-02T12:23:10	Stoichiometric Lithium Niobate Crystals: Towards Identifiable Wireless Surface Acoustic Wave Sensors Operable up to 600$^\circ$C	[ "physics.app-ph", "cond-mat.mtrl-sci", "physics.class-ph" ]	Wireless surface acoustic wave (SAW) sensors constitute a promising solution to some unsolved industrial sensing issues taking place at high temperatures. Currently, this technology enables wireless measurements up to 600-700$^\circ$C at best. However, the applicability of such sensors remains incomplete since they do not allow identification above 400$^\circ$C. The latter would require the use of a piezoelectric substrate providing a large electromechanical coupling coefficient K 2 , while being stable at high temperature. In this letter, we investigate the potentiality of stoichiometric lithium niobate (sLN) crystals for such purpose. Raman spectroscopy and X-ray diffraction attest that sLN crystals withstand high temperatures up to 800$^\circ$C, at least for several days. In situ measurements of sLN-based SAW resonators conducted up to 600$^\circ$C show that the K 2 of these crystals remains high and stable throughout the whole experiment, which is very promising for the future achievement of identifiable wireless high-temperature SAW sensors.	physics.app-ph	physics	Stoichiometric lithium niobate crystals: towards identifiable wireless surface acoustic wave sensors operable up to 600°C Jérémy Streque3, Thierry Aubert1,2, Ninel Kokanyan1,2, Florian Bartoli1,2,3, Amine Taguett1,2, Vincent Polewczyk3, Edvard Kokanyan4, Sami Hage-Ali3, Pascal Boulet3 and Omar Elmazria3* 1 Laboratoire Matériaux Optiques, Photonique et Systèmes (LMOPS), CentraleSupélec, Université Paris-Saclay, Metz, F-57070, France 2 Université de Lorraine, Matériaux Optiques, Photonique et Systemes (LMOPS), Metz, F-57070, France 3 Institut Jean Lamour, UMR 7198 Université de Lorraine -- CNRS, 2 Allée André Guinier, site Artem, 54000 Nancy, France 4 Institute for Physical Research, National Academy of Sciences of Armenia, 0203 Ashtarak-2, Armenia and the Armenian State Pedagogical University after Kh. Abovyan, 0010 Yerevan, T. Mets av. 17, Armenia. * Member, IEEE Received ..., revised …, accepted …, published …, current version ... (Dates will be inserted by IEEE; "published" is the date the accepted preprint is posted on IEEE Xplore®; "current version" is the date the typeset version is posted on Xplore®). Abstract -- Wireless surface acoustic wave (SAW) sensors constitute a promising solution to some unsolved industrial sensing issues taking place at high temperatures. Currently, this technology enables wireless measurements up to 600-700°C at best. However, the applicability of such sensors remains incomplete since they do not allow identification above 400°C. The latter would require the use of a piezoelectric substrate providing a large electromechanical coupling coefficient K2, while being stable at high temperature. In this letter, we investigate the potentiality of stoichiometric lithium niobate (sLN) crystals for such purpose. Raman spectroscopy and X-ray diffraction attest that sLN crystals withstand high temperatures up to 800°C, at least for several days. In situ measurements of sLN-based SAW resonators conducted up to 600°C show that the K2 of these crystals remains high and stable throughout the whole experiment, which is very promising for the future achievement of identifiable wireless high-temperature SAW sensors. Index Terms -- High-temperature sensors, Microsensors, Stoichiometric lithium niobate, Surface acoustic waves I. INTRODUCTION SAW devices are key components of mobile phone industry and other telecommunication systems, but are also very promising for sensing various physical quantities. They offer exciting perspectives for remote monitoring and control of moving parts, especially in harsh environments. Indeed, SAW sensors are passive devices, and just re-radiate a small part of the energy received from the RF interrogation signal [1, 2]. Wireless SAW sensors capable of measuring temperatures up to 600-700°C already exist [3-5] and are based on langasite crystals (La3Ga5SiO14, often called LGS). The surface of these crystals shows an outstanding stability at high temperatures up to 1000°C in air atmosphere [6]. Due to relatively low electromechanical coupling coefficient K2 values (typically 0.4%), LGS-based SAW sensors are configured as resonators, whose identification can only be performed by frequency segmentation, and thus for a few sensor units simultaneously. This limitation can be overtaken by the use of the reflective delay line (R-DL) configuration, making the sensors identifiable through their coded time- resolved response [7]. High-performance R-DL sensors require time pulses as short as possible, and thus the largest Corresponding author: T. Aubert ([email protected]). possible device bandwidth. In this respect, with typical K2 values up to 5% and a Curie temperature above 1000°C, ferroelectric lithium niobate (LiNbO3, often called LN) crystals are naturally strong candidates to achieve high temperature R-DL SAW sensors. LN crystals span from quasi-stoichiometric compositions to lithium-poor grades as low as approximately 45 mol% Li2O [8]. Most of commercially available high-quality LN single crystals are produced by Czochralski technique and have a composition near the congruently melting value of roughly 48.4 mol% Li2O. Single crystals of stoichiometric composition, grown either by the top-seeded solution growth (TSSG) method from potassium-containing flux [9] or from lithium-rich melts by double crucible method [10], are also commercially available, but in smaller quantities. SAW devices based on congruent LN (cLN) were studied at high temperatures over a decade ago. They are limited to temperatures below 400°C for at least two different reasons. First, cLN crystals are not thermodynamically stable below 900°C. The segregation of a niobium-rich phase LiNb3O8 occurs in such conditions [11-12]. The kinetics of this process becomes really problematic above 400°C [13]. Moreover, the time pulses attenuation of cLN-based R-DL increases drastically with temperature, prohibiting such devices to be remotely interrogated above 400°C [14]. Such effect could be related to the relatively low electrical resistivity of cLN crystals at high temperature due to Li vacancies [15-16]. 1949-307X © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information. (Inserted by IEEE) On the contrary, stoichiometric lithium niobate (sLN) crystals are expected to be thermodynamically stable up to their melting temperature at 1170°C [11], while keeping a larger electrical resistivity than cLN crystals by one order of magnitude at any temperature [17]. Promising bulk acoustic wave (BAW) signals were observed up to 880°C on sLN- based BAW resonators, with a stable quality factor Q up to 700°C [18-19]. The purpose of the present letter is to explore the potential of sLN crystals for high temperature SAW applications, in view of the future achievement of identifiable wireless R-DL sensors able to operate in the intermediate temperature range, i.e. from 400 to 600°C at least. II. EXPERIMENTAL SETUPS in order to assess by Raman characterizations were Piezoelectric substrates need to fulfill various requirements for high temperature SAW sensing compatibility: while structural and chemical stability are mostly desirable for recovering from multiple temperature variations over the lifetime of the devices, the electroacoustic properties must be high enough over the whole operating temperature range. Consequently, material firstly performed the sLN stability with temperature, in terms of structure and composition. Then the electroacoustic properties were examined through the study of a first series of SAW devices fabricated to this purpose. The investigated samples consist in Z-cut near-stoichiometric (49.8 mol% Li2O) LN substrates (MTI Corp., Richmond, CA). Some sLN crystals were placed in a furnace to be heated for 48h at 800°C in air. The evolution of the crystallographic structure was analyzed by X-ray diffraction (XRD), in Bragg- Brentano geometry (PANalytical X'Pert Pro MRD - CuKα1: λ = 1.54056 Å). Additional information regarding this evolution was spectroscopy measurements (Horiba Jobin Yvon ARAMIS). A He-Ne laser with 632.8 nm of wavelength was operated. The penetration depth was to XRD measurements, Raman spectroscopy gives the possibility to reveal the formation of a thin amorphous surface layer. It also gives access to the stoichiometry of the crystal surface. To do so, the spectra of Z(XY)Z backscattering configuration were measured in order to obtain the E[TO1] Raman mode, corresponding to the Nb/O vibration in the (X,Y) plane [20], which provides a possibility to calculate the composition of the crystal from the linewidth of this mode [21]. The Raman and XRD results were compared with those obtained from similar experiments conducted on Z-cut cLN crystals (48.4 mol% Li2O) purchased at the same supplier. The remaining samples were some electromechanical properties of sLN crystals at high temperature, in particular K2 as it is an essential parameter for the achievement of R-DL sensors. These electromechanical parameters were accessed by a classical method, based on the measurement of SAW resonators. It is worth noting that the investigated resonators are not envisioned as sensors but only as a probe to measure these parameters. Regarding the fabrication of the resonators, 150 nm-thick aluminium films were deposited by DC magnetron sputtering on the sLN substrates for the patterning of the electrodes. about 1 μm. Complementary to investigate provided then employed Aluminum is not the most suitable material to achieve electrodes able to sustain high-temperature conditions. However, it can withstand temperatures up to 600°C for short time characterizations like those conducted in this study (typically three hours for each heating-cooling cycle) and its remarkable properties in term of low density and electrical conductivity makes it particularly convenient for the design of efficient SAW resonators. Synchronous single-port resonators with a wavelength of λ = 6.5 μm were fabricated by conventional photolithography and wet etching. They were equipped with 200 reflectors on each side of their InterDigitated Transducer (IDT), which was constituted by 100 finger pairs, with an aperture of 40·λ. The metallization ratio of the electrodes was set to 40%. The direction of wave propagation was along the Y-direction of in situ electrically the crystals. These devices were characterized under air atmosphere, using a network analyzer and an RF prober station (S-1160, Signatone Corp., Gilroy, CA) equipped with a thermal probing system that enables to control the device temperature up to 600°C (S-1060, Signatone). III. RESULTS AND DISCUSSION the near-stoichiometric composition of Raman measurements conducted before annealing exactly confirmed the investigated crystals, with a measured lithium concentration of 49.85±0.05 mol% Li2O, in perfect agreement with the manufacturer data. No measurable change in the whole Raman spectra and in the E[TO1] Raman mode in particular could be observed after the 48h-long annealing treatment at 800°C (Fig. 1). These results indicate that the crystal surface does not undergo any the stoichiometric composition and the initial microstructure. losses, keeping lithium Fig. 1. E[TO1] Raman mode of sLN crystals before and after a 48h- long annealing process at 800°C. The same mode obtained for cLN crystal was added for comparison. In the inset can be seen the whole Raman spectra of the sLN crystal. All spectra were carried out in Z(XY)Z configuration. XRD measurements confirm that the lattice structure of the sLN crystals is unchanged after this annealing process (Fig. 2(a)), in contrast with cLN crystals. In that case, the (006) LiNbO3 reflex is shifted towards low angles by 0.07°, Page 2 of 4 while its intensity is divided by a factor 3 after the heating process. Moreover, a new broad peak appears at 2θ = 37.85°, which can be identified as the (60-2) LiNb3O8 reflex (Fig. 2(b)), confirming that cLN crystals segregate in such conditions contrariwise to sLN crystals. Fig. 3. Admittance magnitude of an sLN-based resonator measured at different temperatures during the heating phase. Fig. 4. Frequency-temperature law of the sLN-based SAW resonators measured between the ambient and 600°C, during the heating and the cooling phases. Effective electromechanical coupling coefficient values have then been derived from the admittance measurements, through the determination of the parallel and series resonance frequencies, and [23]: remains stable, with values close to 1.4% ± 0.2%, in the whole investigated temperature range (Fig. 5). The small observed variations of are mainly attributed to the uncertainties related to the reading of the parallel and series resonance frequencies on the admittance curves, due to the presence of small spurious modes, as these variations are not completely repeatable from a device to the other. Beyond these small variations, no significant drop of K2 can be observed over the whole temperature range of interest, which is very promising for the future achievement of identifiable high-temperature R-DL SAW sensors. Page 3 of 4 several days. Their ability Fig. 2. θ-2θ XRD patterns of reference and 48h-annealed at 800°C sLN (a) and cLN (b) crystals. Based on these structural characterizations, it can be assumed that sLN crystals can be employed as piezoelectric substrates in SAW sensors operating up to 600°C for long periods of at least to preserve good electroacoustic properties have then been verified through the RF characterization of the fabricated devices between room temperature and 600°C, as well as during the cooling phase. The measured admittance spectra are shown in Figure 3. The operating frequency is close to 603 MHz at room temperature, which corresponds to a SAW velocity of 3920 m/s, in very good agreement with the literature regarding the propagation of a Rayleigh wave on ZY-cut LN crystals [22]. The sLN crystals deliver a linear response to temperature variations, and the frequency-temperature law is reproducible which supports the fact that the aluminium electrodes withstand (Fig. 4). The sensitivity of the crystal to temperature variations is high with a temperature coefficient of frequency (TCF) of -84 ppm/K, which is appropriate for the achievement of temperature sensors. this characterization process (a) sLN (b) cLN (cid:9)57058059060061050100150200 Y11 (mS)Frequency (MHz)Heating phase 25°C 150°C 250°C 350°C 450°C 550°C [6] Aubert T and Elmazria O (2012), "Stability of langasite regarding SAW applications above 800°C in air atmosphere," Proc. IEEE Ultrasonics Intern. Symp., pp. 2098-2101, doi: 10.1109/ULTSYM.2012.0524. [7] Reindl L and Ruile W (1993), "Programmable reflectors for SAW-ID-tags," Proc. IEEE Ultrasonics Intern. Symp., pp. 125-130, doi: 10.1109/ULTSYM.1993.339692. [8] Bordui P F, Norwood R G, Jundt D H, and Fejer M M (1992), "Preparation and characterization of of congruent lithium niobate crystals," J. Appl. Phys., vol. 71, pp. 875-879, doi: 10.1063/1.351308. [9] Malovichko G I, Grachev V G, Kokanyan E P, Schirmer O F, Betzler K, Gather B, Jermann F, Klauer S, Schlarb U, and Wöhlecke M (1993), "Characterization of stoichiometric LiNbO3 grown from melts containing K2O," Applied Physics A. Solids and Surfaces, vol. 56, pp. 103-108, doi: 10.1007/BF00517674. [10] Kitamura K, Yamamoto J K, Iyi N, Kimura S, and Hayashi T (1992), "Stoichiometric LiNbO3 single crystal growth by double crucible Czochralski method using automatic powder supply system," J. Crystal Growth, vol. 116, pp. 327-332, doi: 10.1016/0022-0248(92)90640-5. [11] Svaasand L O, Eriksrud M, Nakken G, and Grande A P (1974), "Solid-solution range of LiNbO3," J. Crystal Growth, vol. 22, pp. 230-232, doi: 10.1016/0022- 0248(74)90099-2. [12] Born E, Hornsteiner J, Metzger T, and Riha E (2000), "Diffusion of Niobium in Congruent Lithium Niobate," Phys. Stat. Sol. (a), vol. 177, pp. 393-400, doi: 10.1002/(SICI)1521-396X(200002)177:2<393::AID-PSSA393>3.0.CO;2-F. [13] Hauser R, Reindl L, and Biniasch J (2003), "High-temperature stability of LiNbO3 based SAW devices," Proc. IEEE Ultrasonics Intern. Symp., pp. 192-195, doi: 10.1109/ULTSYM.2003.1293386. [14] Fachberger R, Bruckner G, Knoll G, Hauser R, Biniasch J, and Reindl L (2004), "Applicability of LiNbO3, langasite and GaPO4 in high temperature SAW sensors operating at radio frequencies," IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 51, no. 11, pp. 1427-1431, doi: 10.1109/TUFFC.2004.1367482. [15] Donnerberg H., Tomlinson S. M., Catlow C.R.A. and Schirmer O. F. (1989), "Computer-simulation studies of intrinsic defects in LiNbO3 crystals", Phys. Rev. B, vol. 40(17), pp. 11909-11916, 1989. doi: 10.1103/PhysRevB.40.11909. [16] Iyi N., Kitamura K., Izumi F., Yamamoto J.K., Hayashi T., Asano H. and Kimura S. (1992), "Comparative study of defect structures in lithium niobate with different compositions", J. Solid State Chem., vo. 101, pp. 340-352, doi: 10.1016/0022- 4596(92)90189-3 [17] Li Y Q, Zheng YQ, Tu X N, Xiong K N, Lin Q M, and Shi E W (2014), "The high temperature resistivity of lithium niobate and related crystals," Proc. IEEE SPAWDA, pp. 283-286, doi: 10.1109/SPAWDA.2014.6998581. [18] Weindenfelder A, Schulz M, Fielitz P, Shi J, Borchardt G, Becker K D, and Fritze H (2013), "Electronic and ionic transport mechanisms of stoichiometric lithium niobate at high-temperatures," Proc. Mater. Res. Soc. Symp., vol. 1519, doi: 10.1557/opl.2012.1760. [19] Weindenfelder A, Shi J, Fielitz P, Borchardt G, Becker K D, and Fritze H (2012), "Electrical and electromechanical properties of stoichiometric lithium niobate at high- temperatures," Solid State Ionics, vol. 225, pp. 26-29, doi: 10.1016/j.ssi.2012.02.026. [20] Caciuc V, Postnikov A V, and Borstel G (2000), "Ab initio structure and zone- in LiNbO3," Phys. Rev. B, vol. 61, pp. 8806-8813, doi: center phonons 10.1103/PhysRevB.61.8806. [21] Zhang Y, Guilbert L, Bourson P, Polgar K, and Fontana M D (2006), "Characterization of short-range heterogeneities in sub-congruent lithium niobate by micro-Raman spectroscopy," J. Phys.: Condens. Matter, vol. 18, pp. 957-963, doi: 10.1088/0953-8984/18/3/013. [22] D. Royer and E. Dieulesaint, Elastic Waves in Solids, vol. 1, Berlin, Germany: Springer, 2010. [23] Catherinot L, Giraud S, Chatras M, Bila S, Cros D, Baron T, Ballandras S, Estagerie L, and Monfraix P (2011), "A General Procedure for the Design of Bulk Acoustic Wave Filters," International Journal of RF and Microwave Computer-Aided Engineering, vol. 21, pp. 458-465, doi: 10.1002/mmce.20550. Fig. 5. Stability in temperature of the effective K2 coefficient calculated from admittance measurements on sLN-based resonators. IV. CONCLUSION for high-temperature SAW applications sLN crystals have been investigated in the purpose of being used in air atmosphere. Raman and XRD measurements attest that the crystals are not affected by a 48h-long heating process at 800°C in air, in contrast with LN crystals of congruent composition. In situ SAW measurements conducted on sLN- based resonators between room temperature and 600°C confirm the suitability of sLN crystals for high-temperature SAW applications, in particular for sensing temperature as the sensitivity of these crystals to this parameter is high and linear. The electromechanical coupling coefficient retrieved from RF measurements is very stable, in the order of 1.5% in the whole investigated temperature range. These promising results pave the way to the realization of high-temperature R-DL sensors. However, many challenges will have to be faced before this goal can be achieved. In particular, it will be necessary to study the behavior of SAW propagation losses on sLN crystals at high temperatures. Moreover, it will be mandatory to introduce low-density and low -- resistivity electrodes offering good thermal stability, allowing the generation of strong SAW pulses in the 2.45 GHz frequency band, whose large bandwidth makes it the most suited ISM band for the design of R-DL sensors. REFERENCES [1] A. Pohl A (2000), "A review of wireless SAW sensors," IEEE Trans. on Ultrason., Ferroelectr. Freq. Control, vol. 47, no. 2, pp. 317-332, doi: 10.1109/58.827416. [2] Rashvand H F, Abedi A, Alcaraz-Calero J M, Mitchell P D, and Mukhopadhyay S C (2014), "Wireless Sensor Systems for Space and Extreme Environments: A Review," IEEE Sensors J., vol. 14, no. 11, pp. 3955-3970, doi: 10.1109/JSEN.2014.2357030. [3] Pereira da Cunha M (2013), "Wireless Sensing in Hostile Environments," Proc. IEEE Ultrasonics Intern. Symp., pp. 1337-1346, doi: 10.1109/ULTSYM.2013.0342. [4] François B, Richter D, Fritze H, Davis Z J, Droit C, Guichardaz B, Petrini V, Martin G, Friedt J M, and Ballandras S (2015), "Wireless and passive sensors for high temperature measurements," Proc. AMA Sensor Conf., pp. 390-395, doi: 10.1016/j.proeng.2012.09.374. [5] Wall B, Gruenwald R, Klein M, and Bruckner G (2015), "A 600°C wireless and passive temperature sensor based on langasite SAW-resonators," Proc. AMA Sensor Conf., pp. 390-395, doi: 10.5162/sensor2015/C3.3. Page 4 of 4 01002003004005006000.00.51.01.5Electromechanical coupling coefficient K2eff (%)Temperature (°C) Heating Cooling
1903.07535	1	1903	2019-03-18T16:23:22	Green InGaN/GaN LEDs: High luminance and blue shift	[ "physics.app-ph" ]	We report in this paper electro-optical results on InGaN/GaN based green micro light-emitting diodes ({\mu}LEDs). Current light-voltage measurements reveal that the external quantum efficiency (EQE) behavior versus charge injection does not follow the ABC model prediction. Light-emission homogeneity investigation, carried out by photoluminescence mapping, shows that the Quantum Confinement Starck Effect (QCSE) is less significant at the edges of {\mu}LEDs. Electroluminescence shows a subsequent color green-to-blue deviation at high carrier injection levels. The extracted spectra at different current injection levels tend to show the appearance of discrete wavelength emissions. These observations may enhance the hypothesis that higher-energy excited-levels in InGaN quantum wells may also contribute to the blue shift, solely attributed to QCSE lessening under intense electric field magnitudes. We hereby present first results dealing with green {\mu}LEDs electro-optical performances with regards to their size.	physics.app-ph	physics	Green InGaN/GaN LEDs: High luminance and blue shift Anis Daamia, François Oliviera, Ludovic Dupréa, Christophe Licitraa, Franck Henrya, François Templiera, Stéphanie Le Calveza aUniversité Grenoble Alpes, CEA-LETI, Minatec Campus, III-V Lab, Grenoble, France ABSTRACT We report in this paper electro-optical results on InGaN/GaN based green micro light-emitting diodes (µLEDs). Current- light-voltage measurements reveal that the external quantum efficiency (EQE) behavior versus charge injection does not follow the ABC model prediction. Light-emission homogeneity investigation, carried out by photoluminescence mapping, shows that the Quantum Confinement Starck Effect (QCSE) is less significant at the edges of µLEDs. Electroluminescence shows a subsequent color green-to-blue deviation at high carrier injection levels. The extracted spectra at different current injection levels tend to show the appearance of discrete wavelength emissions. These observations may enhance the hypothesis that higher-energy excited-levels in InGaN quantum wells may also contribute to the blue shift, solely attributed to QCSE lessening under intense electric field magnitudes. We hereby present first results dealing with green µLEDs electro-optical performances with regards to their size. Keywords: InGaN/GaN, green µLED, blue shift, QCSE 1. INTRODUCTION Micro light-emitting diodes (µLEDs) based on InGaN/GaN quantum well structures have seen a significant amount of progress in the last two decades1, 2, 3. New emerging applications such as augmented/mixed or virtual reality are looking forward using micro-displays (µ-displays) based on these still progressing technologies. However, a majority of studies in literature are mostly dealing with blue emitting µLEDs, with fewer investigation papers on green light emitting ones. Actually, the high-content indium incorporation in InGaN alloys, to adjust the emission wavelength to green color, is not an easy process track to deal with. Nonetheless, green µLEDs are well positioned to cover high luminance needing components. Indeed, the eye sensitivity is at its maximum in this range of emitted wavelengths. Furthermore, for specific applications such as see-through glasses, a high brightness level of 5000 cd/m² or more is a mandatory keystone to achieve. Nevertheless, quantum efficiency of µLEDs is yet a challenging subject when dealing with high luminance levels. Besides, µLED size dependence has been thoroughly studied and reported in literature, but mostly for blue color emitting ones4, 5. A lack of comprehension of size effect is then still to be filled for these green light-emitting diodes dedicated to specific µ-displays applications. This paper deals with an electro-optical study on different sized green µLEDs. First, quantum efficiency performance is analyzed versus µLED size. In a second part, we focus respectively, on µLED emission homogeneity and spectral response behaviors versus electrical injection. 2.1 Current-Voltage characteristics 2. ELECTRO-OPTICAL INVESTIGATION Current density versus voltage characteristics of the different measured devices are shown on figure 1. At first glance, the current density seems independent of the µLED size on a large bias range. Larger devices (500µm and 200µm) show a slightly lower current density at high voltage values that is attributed to a probable deviation of the series resistance from an ideal geometrical law. We have shown in a recent study6 that a series resistance variation can have a major impact on the µLED current at high voltage. On the other hand, at a bias value lower than the µLED threshold voltage, the perceived current density difference is mainly related to the limitation of our apparatus sensitivity. Despite these slight observed differences, we demonstrate once again the robustness of our µLED fabrication process7. [email protected] Figure 1. Current density versus voltage characteristics of different square µLEDs. Current density is constant on a large bias range whatever the µLED size is. The legend denotes the side width of each µLED. 2.2 Optical power and quantum efficiency As regards to optical properties of our devices, figure 2(a) shows the measured optical output power versus applied voltage of the different µLEDs. The corresponding external quantum efficiency (EQE) of each device are plotted versus current density on figure 2(b). In contrary to what is reported4 on our blue LEDs, the optical threshold voltage (VTO) of green µLEDs does not seem to vary much in a certain range of geometry (width ≥ 50µm). The extracted value of VTO in this range of geometry is around 2.1 ± 0.05V. For smaller µLEDs, VTO begins to shift, attaining 2.5 V for a 10µm µLED. Moreover, the optical output power of large devices, shows two distinct regimes. At low voltage, a first rapid increase is observed. Then, this fast upturn slows down, showing a 'kink' like effect. In a second time, the power rises again with bias until a saturation regime due to droop is attained. The echo of these two regimes is evident on the EQE curves of large devices. Indeed a first increase is perceived, then the EQE reaches a plateau (EQE ~ 2%), corresponding to the optical power output 'kink' region. Afterwards, the EQE restarts to increase reaching a maximum EQEmax = 5%, before sinking down due to droop effect. It is worth pointing out that the first rapid regime seems to disappear when the size of the µLED shrinks. This is evident on EQE curves where the plateau regime tends to vanish for small geometries. Besides, the EQE threshold current density moves to higher values for the small devices (width ≤ 20 um). We hereafter, emit the hypothesis that the fading of the rapid optical output regime and the shift of the optical threshold voltage VTO are correlated. Figure 2. (a) Optical power versus voltage and (b) external quantum efficiency versus current density characteristics of different square µLEDs. Two different optical regimes are depicted on large scale µLEDs. The 1st regime tends to disappear for smaller µLEDs. It is worth noting that the 1st regime vanishing is correlated to the shift of the optical emission threshold also observed on smaller geometry µLEDs. 2.3 Light emission homogeneity Many studies in literature point out the importance of the µLED periphery on its optical efficiency8, 9. Downsizing the device geometry has usually come out with lower optical efficacies. One crucial adopted reason is the perimeter taking over the surface, when the µLED size diminishes. Numerous investigations in this direction, have shown the increase of SRH non-radiative recombination. This is primarily related to defects and a bad passivation of µLED periphery after etching steps. We have carried out photoluminescence (PL) mapping on a 3 by 3 array of green µLEDs (7x7 µm²) to check the quality of our etching process and light emission homogeneity. Figure 3 shows respectively, the obtained PL intensity and wavelength mappings. Oddly, we observe a higher signal at the edges of the µLEDs. Moreover the emitted wavelength is to some extent lower at these boundaries. This observation, presumably points out a less predominant QCSE at the periphery of our devices. Indeed, we suppose a strong relaxation of the lattice constraint at the edges of the µLED. This constraint reduction gives rise to a reduced spontaneous polarization in the quantum wells. Hence, the locally emitted light is shifted to a higher energy. Moreover, the reduced band curvature would allow a better recovery of hole and electron wave-functions, explaining the stronger emission intensity. A similar observation has been reported10 by Xie et al. using the cathodoluminescence technique. Figure 3. Photoluminescence intensity and wavelength mappings on a 3x3 7µm squared µLEDs array. A higher intensity correlated to a shorter wavelength emission is observed on the edges of µLEDs. We suppose a probable lattice relaxation at the edges of µLEDs, reducing the QCSE effect behind this observation. Another important observation shown in figure 4 reveals that light emission homogeneity is bias dependent on larger scale µLEDs (100x100 µm² LED example shown). Indeed we can see that at low voltage range, the emission is rather speckled on the surface of the device. This spattered pattern is mainly related to a non-homogeneous indium incorporation in quantum wells. Consequently, this indium spread results in a local optical threshold variation through the device surface itself. Higher content indium spots will have a lower optical threshold voltage, hence emitting at lower bias values. It is regularly reported in literature that high content indium amalgamation with GaN is a harsh path to consider11, 12. Another reason to this light speckle pattern may also come from a local variation of contact resistivity. This electrical deviation can be related to the indium spread, stated above. It can also be due to a non-homogeneous P doping, locally degrading the contact/semiconductor interface13, 14. Figure 4. Optical photos of light emission homogeneity on a 100x100 µm² green LED at different bias values: (a) V = 2.2V, (b) V = 3.5V and (c) V = 4.5V. Note the high luminance spread at low voltage that disappears at higher bias values. Regarding smaller area µLEDs, usually dedicated to micro-displays, this singularity has to be banned. Indeed luminance homogeneity is one important key factor of micro-displays. This observed spread in luminance tends to disappear when injection bias is driven upwards. Nevertheless, this voltage increase induces the appearance of another oddity. Actually, the emitted-light wavelength shifts towards lower values. More details on this shift are discussed in the next section. 3.1 Electroluminescence and color shift 3. THE BLUE SHIFT DILEMMA We carried out electroluminescence measurements on small sized unitary µLEDs (6µm diameter) in order to limit the surface-spread light-emission observed at low injection. Optical photos taken at different applied voltages, presented in figure 5, perceptibly show an intense wavelength shift of the emitted light when injection bias is increased. We clearly see a green to blue transformation of the emitted color. Figure 5. Optical photos of light emission on a 6µm diameter 'green' µLED at different bias values. The emitted wavelength is very dependent on applied voltage across the µLED. Green color turns to blue when increasing bias from 2.8V to7.5V. This important evolution has been quantified, and the color point plotted on a CIE chromaticity 1931 diagram, displayed in figure 6. Each measured point corresponds to an increase of 0.1V of the µLED bias, varying from 2.5V to 7.5V. There is no doubt that the emitted wavelength covers a large range on the CIE diagram varying from green (=540nm) to an almost true blue (=475nm). Figure 6. Color point evolution of the emitted light from a green µLED plotted on CIE chromaticity 1931 diagram. Bias varies from 2.5V to7.5V. Each point corresponds to 100mV increase in voltage. The measured wavelength varies from 540nm (green) to 475nm (blue). The wavelength blue shift is also observed in blue emitting µLEDs but at a lesser extent. It is usually attributed to the fading, under high voltage, of the Quantum Confined Starck Effect (QCSE). This later is explained by the default presence of a local electric field in the quantum wells. Its magnitude is directly depending on both the spontaneous and piezo-electric polarizations existing in InGaN/GaN light emitting devices15. In the case of green µLEDs, the high indium content induces a huge lattice mismatch, at the barrier/quantum well interface between InGaN and GaN materials. Subsequently, a high charge density appears at the quantum well interfaces. Consequently, a high magnitude piezo-electric field, related to these pseudo charge-sheets, heavily twists the energy bands in the quantum well and the QCSE is more pronounced. This energy band bending induces a red shift in optical transitions between confined energy levels in quantum wells. When the injection voltage level is increased through the µLED, the piezo-electric field is slowly screened and the energy bands are less warped. The red shift linked to QCSE is then recovered and slowly vanishes turning into a blue shift. Nevertheless, blue shift elucidations are still debated in literature. 3.2 Quantum well excited-levels filling hypothesis We present on figure 7, normalized intensity electroluminescence spectra at four different voltage values, recorded on the 6µm µLED. At low bias, close to VTO, one wavelength peak centered at 520 nm is observable (red curve). This peak corresponds to the mean indium content incorporated in the InGaN quantum well. When the voltage increases, we observe the formation of new emission peaks alongside the first one. At 3.5V, beside the 520 nm peak, we perceive the appearance of a higher intensity one, centered at 508 nm, and a slight hump at 480 nm (green curve). After increasing the bias by 2V, (blue curve) the 480 nm intensity takes over the two precedent higher wavelength peaks, and a hump at 460 nm begins to appear. This later shows the same luminescence intensity as the 480 nm peak when voltage is raised to 7.5V (purple curve). At this bias level, both precedent peaks (520 nm and 508 nm) appear as small bulges compared to the high energy peaks (460 nm and 480 nm). The discrete wavelength peaks tend to demonstrate that high energy excited levels inside quantum wells are progressively populated at high injection levels. This permits the appearance of higher energy, hence shorter wavelength optical transitions. This wavelength and peak intensity ballet versus injection bias is a plausible elucidation of the green to blue shift in our µLEDs. Figure 7. Normalized electroluminescence spectra versus bias carried out on a 6 µm diameter µLED. In contrary to what is usually believed, the blue shift in InGaN/GaN µLEDs is not solely due to the QCSE. The appearance of discrete emission peaks with variable intensities depending on injection bias level tends to corroborate the filling of high excited levels in the quantum wells explaining the color changing behavior in green µLEDs. 4. CONCLUSION InGaN/GaN based green µLEDs are less debated in literature compared to blue emitting ones. One main reason is the difficulty to incorporate high indium content in InGaN alloys. Yet, they are good candidates for simple see-through applications as the eye sensitivity to green is at its maximum. We have exposed through an electro-optical thorough analysis of different sized green µLEDs that their quantum efficiency behavior is far from being easily described by an easy ABC model. In addition, they show a less effective QCSE at their edges. The light emission homogeneity study also reveals an indium content dispersion inside quantum wells. The electroluminescence measurements divulges a drastic green to blue color shift. Beside QCSE reduction in µLEDs under bias, we suspect the filling of high energy levels in quantum wells as a likely explanation of this shift. Finally, this paper aims to a better understanding of InGaN/GaN based green µLEDs features, by a very first investigation of size effect on their electro-optical behavior. 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). ACKNOWLEDGEMENTS REFERENCES [1] 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). [2] 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). [3] 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). [4] 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). [5] 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). [6] Daami, A., Olivier, F., Sarrasin, D., Dupré, L., Templier, F., "InGaN/GaN μLEDS for display applications optical and electrical characteristics spread comprehension", Proc. Int. Conf. SSDM, Sendai, Japan, (2017). [7] Dupré, L., Marra, M., Verney, V., Aventurier, B., Henry, F., Olivier, F., Tirano, S., Daami, A., Templier, F., "Processing and characterization of high resolution GaN/InGaN LED arrays at 10 micron pitch for micro display applications," Proc. SPIE, Vol. 10104, 1010422, (2017). [8] Terano, A., Imadate, H., Shiojima, K., "Mapping etching induced damages on GaN surfaces using scanning internal photoemission microscopy", Mater. Sci. Semicond. Process., 70, Supp. C, 92-98, (2017). [9] Choi, K. J., Jang, H. W., Lee, J.-L., "Observation of inductively coupled-plasma-induced damage on n-type GaN using deep-level transient spectroscopy", Appl. Phys. Lett., 82, 1233-1235, (2003). [10] Xie, E. Y., Chen, Z. Z., Edwards, P. R., Gong, Z., Liu, N. Y., Tao, Y. B., Zhang, Y. F., Chen, Y. J., Watson, I. M., Gu, E., Martin, R. W., Zhang, G. Y., Dawson, M. D., "Strain relaxation in InGaN/GaN micro-pillars evidenced by high resolution cathodoluminescence hyperspectral imaging", J. Appl. Phys. 112, 013107, (2012). [11] Hangleiter, A., Hitzel, F., Netzel, C., Fuhrmann, D., Rossow, U., Ade, G., Hinze, P., "Suppression of Nonradiative Recombination by V-Shaped Pits in GaInN/GaN Quantum Wells Produces a Large Increase in the Light Emission Efficiency", Phys. Rev. Lett. 95, 127402, (2005). [12] Vaitkevicius, A., Mickevicius, J., Dobrovolskas, D., Tuna O., Giesen, C., Heuken, M., Tamulaitis, G., "Influence of quantum-confined Stark effect on optical properties within trench defects in InGaN quantum wells with different indium content", J. Appl. Phys. 115, 213512, (2014). [13] Liu, Q. Z., Lau, S. S., "A review of the metal -- GaN contact technology" Solid-State Electron., 42(5), 677-691, (1998). [14] Ho, J.-K., Jong, C.-S., Chiu, C. C., Huang, C.-N., Chen, C.-Y., Shih, K.-K., "Low-resistance ohmic contacts to p-type GaN", Appl. Phys. Lett., 74(9), 1275-1277, (1999). [15] 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(12), 8849-558, (1999).
1908.05213	1	1908	2019-08-11T19:04:02	Role of electrostatic potential energy in carbon nanotube augmented cement paste matrix	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	The empirical data in conjunction with the quantum mechanical calculations show that the strength enhancement in the cement-carbon nanotubes (CNTs) composites is the courtesy of electrostatic potential energy. This is contrary to the general belief that the CNTs form bridges between the adjacent grains to slow down the breaking process. The yield point for the cement paste is improved up to 25% when prepared with 0.2 % by weight of various types of CNTs. A significant strength enhancement is observed with carboxyl functionalized (COOH) CNTs compared to other types. Further, an increase in the concentration of CNTs up to 0.4 wt% has a negative effect on the strength of the matrix. The electrostatic potential energy is mapped by using density functional theory (DFT) with {\omega}B97X-D functional. At lower concentration of CNTs, ion-dipole interaction in the cement paste and the CNTs creates a very strong long range intermolecular force. Due to the increased entropy resulting from the exothermic hydration process, these forces augment the strength of the cement paste.	physics.app-ph	physics	Manuscript Role of electrostatic potential energy in carbon nanotube augmented cement paste matrix Muhammad Azeema , Muhammad Azhar Saleem b,c aDepartment of APplied Physics and Astronomy, University of Sharjah, University City, Sharjah, 27272, United Arab Emirates. b Department of Civil Engineering, University of Engineering and Technology, Lahore, Pakistan. c American University of Sharjah, Sharjah, United Arab Emirates. Abstract The empirical data in conjunction with the quantum mechanical calculations show that the strength enhancement in the cement-carbon nanotubes (CNTs) composites is the courtesy of electrostatic potential energy. This is contrary to the general belief that the CNTs form bridges between the adjacent grains to slow down the breaking process. The yield point for the cement paste is improved up to 25% when prepared with 0.2 % by weight of various types of CNTs. A significant strength enhancement is observed with carboxyl functionalized (COOH) CNTs compared to other types. Further, an increase in the concentration of CNTs up to 0.4 wt% has a negative effect on the strength of the matrix. The electrostatic potential energy is mapped by using density functional theory (DFT) with ωB97X-D functional. At lower concentration of CNTs, ion-dipole interaction in the cement paste and the CNTs creates a very strong long range intermolecular force. Due to the increased entropy resulting from the exothermic hydration process, these forces augment the strength of the cement paste.  Corresponding author. Tel.: +97165166764; fax: +0-000-000-0000 . E-mail address: [email protected] 1 1. Introduction The demand of cement is projected to rise at an unequivocal rate of 4.5% per year and has already reached 5.2 billion metric tons in 2019[1], making cement as one of the most consumable material by the humans. With the production of several billions of tons each year, the cement industry is one of the major contributor to the global CO2 emissions[2]. However if we are able to substitute the part of the cement by a supplementary cementitious materials then its environmental impact can be controlled. A variety of the materials offer this possibility, nano- dimensional materials being one of them. The theoretically postulated fascinating quantum mechanical phenomena associated with the nano-structured materials are now being observed experimentally. Nanoparticles and nanotubes are highly active and have greater surface to volume ratio therefore more recently they have been pursued to give strength to the construction materials. The results are promising and show that the mixing of the nanomaterials in the cement paste makes it less prone to failure. In general it is perceived that the nanomaterials occupy the pores in the hydrated cement paste and therefore act as binders between the grains[3,4]. There are, however, studies to suggest that the nanoparticles (particularly 3d metals) might occupy interstitial positions in the crystal structure of clinker[5,6] and thus accelerate the hydration process. Since most of the composites present in the hydrated cement are based on calcium and silicon, changing the ratio between two ions has significant effect on the physical properties of the cement composite[7]. The metallic nanoparticles may be used to replace the silicon atoms which then increases the ratio of calcium to silicon as well as changing their crystal structure. The same objective can also be achieved by using nanolimestone/nanoCaCO3[6,8] and nanosilica[9]. There is also a speculation that the mixing of the carbon nanotubes and nanofibers may change the bonding order of the cement matrix[10]. An efficient acceleration in the hydration process of the cement paste is observed by the addition of nanoalumina and graphene oxide[6] and graphene[11,12] improving its compressive and flexural strength. Carbon nanotube reinforced cementitious composites have attracted more attention because of their elastic modulus as high as 400 GPa [10] and tensile strength [0.4-5 GPa]. The Young's modulus for CNTs is around 1 TPa[13] making them superior in strength compared to steel. The fracture strain for CNTs is as high as 280, thanks to the carbon-carbon sp2 bonding 2 whereas fracture strain for steel is only around 20%. However notable is the fact that weak shear interaction between adjacent nanotubes may reduce their tensile strength. Therefore twisted CNT fibers in the form of strands experience significant reduction in their tensile strength. Further the defects introduced during the fabrication process also lead to reduced strength of CNTs. Aggregation of the CNTs permits ring opening mechanism and thus nucleation of the defects resulting in the concentration of the stress and strain at the site of defects. Therefore it is preferred that the CNTs are dispersed uniformly in the composite matrix allowing disentanglement of the tubes. The mechanical properties of CNT doped cementitious materials depend on the several factors such as dispersion, proportion and length of CNTs. Various dispersion methods resulted in 14% to 71% increase in the strength of the cement mixtures[3,4]. Dispersion of longer CNTs is challenging therefore short length CNTs are preferred. Low concentration of well dispersed CNTs have produced effective enhancement of the mechanical properties of the cement. Functional groups of CNTs (carboxyl -- COOH and hydroxyl -- OH) have also been found to affect the physical properties of the cement[4]. It is inferred from the experimental data that the hydrophilic carboxyl CNTs might initiate a chemical reaction between the carboxyl groups and the composites present in the cement. The cement paste with carboxyl functionalized tubes leads to lower concentration of tobermorite decreasing its strength considerably[14]. Porosity of the cement mixture is reduced with the increase in the CNTs concentration however, as discussed above and will be shown in the present work, an increase in the concentration does not necessarily mean an improved strength. On the other hand, the mechanism of strength enhancement however is poorly emphasized and mostly inferences are made without any concrete proof. Crack bridging and the pore filling is the main reason identified for the cement matrix toughness. Therefore the present work highlights the underlying physics of the strengthening mechanism at the electronic level. It is shown that the modification of the cementitious properties strongly depends on the concentration of the foreign elements. It is, as a matter of fact, the electrostatic interaction that the augmented strength is obtained only for a certain concentration of the dopants. A change in the concentration of the added impurities affects the order of the electrostatic interaction with the hydrated cement paste therefore despite reduced porosity[10], the strength decreases with the 3 increase in the concentration. The results are reported for the cement paste prepared with various concentrations (0.2 wt% and 0.4 wt%) of pristine, hydroxyl and carboxyl functionalized CNTs. Their compressive strength is tested after the gap of 7 and 28 days. The increase in the compressive strength is the highest for the samples with carboxyl functionalized nanotubes but the lowest for the hydroxyl functionalized. Further, a theoretical model is constructed to show that there is a strong electrostatic interaction between the CNTs and the Ca(OH)2, the major component in the hydrated paste. It is illustrated that while the dipole moment of the CNTs is almost zero, the strain applied by the neighbouring crystallites distorts the geometry of CNTs and therefore introduce a change in the density of electronic charge on the surface of CNTs. A charge imbalance creates a strong dipole moment initiating the ion-CNTs dipole interaction. This is the origin of the electrostatic interaction between different composites in the cement paste and CNTs. The exothermic hydration process increases the entropy of the system and therefore the electrostatic interaction is long range, greatly improving the mechanical strength of the cement paste. 2. The Experiment 2.1 Sample Preparation The cement used in preparation of the specimens was commonly available Type I ordinary Portland cement (OPC). The carbon nanotubes (CNTs) were obtained from Sisco Research Laboratories (SRL) and were of three types: pristine, hydroxyl functionalized (-OH) and carboxy (-COOH) functionalized. The average lengths of all three types of the CNTs was around 10-30 μm. The diameter of the pristine CNTs was around 8 nm whereas functionalized CNTs were of the diameter approximately 30-50 nm. The samples were tested for their tensile strength. Tap water was used for preparing the cement paste with a constant water-cement ratio of 0.4, for the entire stock of specimens. The equipment and the different stages of the sample preparation are exhibited in the Fig. 1(a-g). 4 (a) Weighing of cement (b) Bench mixer (c) Empty Moulds Placed on Table Vibrator (d) Cement Paste being Poured in Moulds (e) Specimens after 24 hrs (f) Specimens after Demolding (g) Specimens Placed in Water for Curing Fig. 1 Various stages of sample preparation 5 Before using in the paste, cement was weighed, Fig. 1a, and sieved to get rid of any lumps. Mixing was carried out in a bench mixer, Fig. 1b. The CNTs and water were first added to the bowl and the mixer was switched on. Mixer was allowed to run for 2 mins to ensure that CNTs are uniformly dispersed in water. Afterwards, cement was added gradually while the mixing continued. Mixing was carried out for another 15 mins from the starting time of addition of cement. Gradual addition of cement avoided formation of cement lumps and resulted in consistent paste. After completion of mixing, cement paste was poured into the steel moulds of size 50 mm x 50 mm x 50mm, Fig. 1c & 1d, which were oiled to ensure easy removal. Table type external vibrator was used for compaction. Specimens were demolded after 24 hrs, Fig. 1(e, f), and then dipped in water until the day of testing, Fig.1 (g). In total, 48 specimens were prepared and for each unique type three companions were cast to get an average compressive strength. Each specimen was a 50 mm cube, which is typically used to measure the compressive strength of cement. Four batches of cement paste were prepared. These batches comprised of the a batch of control specimens without CNTs, a batch each having 0.2% and 0.4% by dry weight of cement pristine CNTs, -OH and -- COOH functionalized CNTs. The summary of the specimens prepared is presented in the Table 1. The specimens were labeled to indicate their type and composition. For example, letter C represents the control specimens, P-2 represents the specimens with pristine CNTs having 0.2 wt% and P-4 refers to the specimens with pristine CNTs having 0.4 wt%. Specimens with OH-CNT and COOH-CNTs were labeled accordingly. Type of Specimens Control Pristine CNT OH-CNT COOH-CNT ID C P-2 P-4 OH-2 OH-4 COOH-2 COOH-4 Table 1. Test Matrix No. of Day of % Weight of Water/Cement Specimens Testing CNT Ratio 12 3, 7, 14, 28 6 6 6 6 6 6 7, 28 7, 28 7, 28 7, 28 7, 28 7, 28 0 0.2 0.4 0.2 0.4 0.2 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 6 Specimens were tested in load-control mode. Great care was taken to ensure that loading platens remain horizontal so that load gets uniformly distributed on the entire surface area of cubes. Testing of the specimens is exhibited in the Fig. 2(a, b). (a) Cube Specimen Ready for Testing (b) Specimens after Failure Fig. 2 Specimen testing 3. Results and discussions 3.1 Compressive Strength The testing results on the control cement paste samples and the CNT mixed samples are listed in the Table 2 and the Table 3 respectively. The average compressive strength of all the batches of specimens was done on 7th and 28th day. Control specimens were, however, tested on days 3, 7, 14 and 28. Testing on two days i.e. 7th and 28th provide reasonable idea about the rate of gain of strength, one being at early age and the other at maturity. Traditionally, the 28 days compressive strength is used in the construction industry for the design of concrete structures. Average compressive strength of control specimens (Table 2) on 7th day was about 30 MPa which doubled on the 28th day to almost 60.4 MPa. This indicates that the hydration of the cement phases is a slower process. While the cement paste stiffens within few hours, it takes one year for all of the cement to hydrate. However, more than 70% of the cement reacts with water in 28 days[15] therefore a significant development in the compressive strength occurs in this period of time. 7 Table 2 Testing results on the control samples Type of Specimen Age Specimen ID (Days) Failure Compressive Average Load Strength, FC FC (kN) (MPa) (MPa) C-a-3 54.3 21.72 C-b-3 3 57.5 23.00 21.69 C-c-3 C-a-7 50.9 20.36 76.9 30.76 C-b-7 7 75.8 30.32 29.96 Control C-c-7 72 28.8 C-a-14 95.4 38.16 C-b-14 14 88.6 35.44 37.65 C-c-14 C-a-28 98.4 39.36 153 61.20 C-b-28 28 121.1 48.44 60.40 C-c-28 178.9 71.56 The compressive strength of the CNT-mixed cement paste specimens are shown in the Table 3. The specimens with 0.2 wt% pristine CNT exhibited an average compressive strength of 43.4 MPa which increased by 66%, up to a value of 72.2 MPa on 28th day. However, this increase was only 46% in the case of 0.4 wt% pristine CNT specimens. It is worth mentioning that the pristine CNT specimens with 0.2 wt% and 0.4 wt% CNTs had similar strength on 7th day, around 43 MPa. Unexpectedly, 28th day compressive strength of the specimens having 0.4 wt% CNT was found to be 13% less than the specimens with 0.2 wt% CNTs. This may be 8 attributed to excessive replacement of cement in 0.4 wt% CNTs specimens as will be discussed later in detail. The CNTs have no cementitious properties therefore excessive replacement of cement must have led to reduction in compressive strength. In comparison to control specimens, 28th day compressive strength of 0.2 wt% regular-CNT specimens increased by 19.5% but for 0.4 wt% specimens this increase was only 5%. The results are even more interesting for the OH and COOH functionalized CNT mixed cement paste specimens. The 7 day compressive strengths of OH-CNT and COOH-CNT specimens were almost the same as the regular-CNT specimens, i.e. 43 MPa on average but 45% higher than the control specimens. The 28 day compressive strengths of 0.2 wt% OH-CNT and COOH-CNT specimens were 67.7 MPa and 75.1 MPa, an increase of 12% and 24%, respectively. Similar to regular-CNT specimens, reduction in compressive strengths of 0.4 wt% OH-CNT and COOH-CNT specimens was observed. The 0.4 wt% OH-CNT specimen, on day 28, failed at a stress of 53 MPa which is even less than the control specimen by almost 9%. The strength of 0.4 wt% COOH-CNT was no different. It failed at a stress of 57.7 MPa, a value 4.5% less than the control specimen. The absolute maximum compressive strength on day 28 was observed for 0.2 wt% COOH-CNT specimens, which failed at 75.1 MPa, an increase of 24% with respect to the control specimen. 9 Table 3 Testing results on the CNT mixed cement paste samples Type of Specimen Age Specimen ID (Days) CNT Failure Compressive concentration Load Strength, FC (wt%) (kN) (Mpa) Average FC (Mpa) P-2-a P-2-b 7 P-2-c P-4-a P-4-b 7 Pristine CNT P-4-c P-2-a P-2-b 28 P-2-c P-4-a P-4-b 28 P-4-c OH-2-a OH-2-b 7 OH-CNT OH-2-c OH-4-a OH-4-b 7 2 4 2 4 2 4 107.5 43 104.2 41.68 43.40 113.8 45.52 115.7 46.28 101.5 40.6 43.28 107.4 42.96 179.8 71.92 170.4 68.16 72.21 191.4 76.56 163.3 65.32 156.5 62.6 63.39 155.6 62.24 117.3 46.92 112.8 45.12 45.37 110.2 44.08 112.9 45.16 115.7 46.28 46.52 10 OH-4-c OH-2-a OH-2-b 28 OH-2-c OH-4-a OH-4-b 28 OH-4-c COOH-2-a COOH-2-b 7 COOH-2-c COOH-4-a COOH-4-b 7 COOH- CNT COOH-4-c COOH-2-a COOH-2-b 28 COOH-2-c COOH-4-a COOH-4-b 28 2 4 2 4 2 4 120.3 48.12 170.4 68.16 164.5 65.8 67.69 172.8 69.12 134.6 53.84 127.7 51.08 53.03 135.4 54.16 105.3 42.12 102.5 41 42.44 110.5 44.2 109.4 43.76 119.9 47.96 45.71 113.5 45.4 191 76.4 187.3 74.92 75.13 185.2 74.08 157.3 62.92 142.6 57.04 57.73 COOH-4-c 133.1 53.24 11 The summary of the compressive strength tests is illustrated in the Fig. 3. The trend clearly shows that although the CNTs doping for three types does enhance the strength of the hydrated cement paste in all the cases however it is true for only certain value (0.2 wt%) of concentration. Increasing the CNTs concentration, indeed has a negative effect on the strength. Fig. 3 Comparison of Compressive Strengths for all the Specimens 3.2 Carbon nanotubes - Scanning Electron Micrographs and Raman Spectra The micrographs were obtained from Tescan VEGA XM variable pressure scanning electron microscope. The three types of CNTs are shown in the Fig. 4(a-c). Apparently the CNTs are twisted and kinked in the form of bundles due to strong van der Waal forces. Disentanglement of the CNTs and their uniform dispersion in the cement paste requires a particular attention. In addition, there is a high probability of the presence of the structural and morphological defects in the CNT bundles, also confirmed by Raman spectroscopy. 12 Fig. 4 Scanning electron microscopy of CNTs (a) pristine (b) OH functionalized (c) COOH functionalized 13 The Raman spectra was collated at different sites of a single bundle and then the average was calculated for all the spectra. The spectra for the three types of CNTs is shown in the Fig. 5. The absence of the radial breathing modes (RBM), the first order Raman scattering process, at lower frequencies (less than 500 cm-1) and broader peaks indicate that the CNTs are multilayered. The RBM signals are usually associated with single walled nanotubes (SWNT) of small diameter. Multi-walled nanotubes (MWNT) on the other hand are several SWNTs of variable diameters and chirality wrapped on top of each other seamlessly. The RBM signals therefore are weaker due to incoherent out of phase stretching of bonds. The most prominent peak at 1582 cm-1 is the so called G-band, another first order Raman process which is a finger print feature shared by the graphite, SWNT and MWNTs. However there are subtle difference in the shape of peaks for the each type. The G-band is weakly asymmetric for MWNTs compared to a clear Lorentzian centered at the at a value of Raman shift 1582 cm-1 for 2D graphite. Furthermore, the D and D overtone bands are also visible at 1353 cm-1 and the 2708 cm-1 respectively. These bands are associated with the defects and the doping respectively and therefore can be used to characterize and monitor the structural changes in the CNTs[16]. For example for the spectra shown in the Fig. 5, the D/G ratio for the COOH and OH functionalized is almost 0.70 whereas for the pristine CNTs, the ratio is approximately 0.50 showing that the order of defects is higher in the functionalized CNTs. These defects play a pivotal role in developing a long range electrostatic and crystal order in the cement paste. 14 Fig. 5 Raman spectra from carbon nanotubes. 15 3.3 Cement, cement paste and cement paste mixed with CNTs -- Scanning Electron Micrographs, FTIR spectra and XRD patterns Representative SEM micrographs for the control specimens of the cement paste and the cement paste mixed with the OH and COOH functionalized CNTs are shown in the Figs. 6-9. It should be noted that the physical appearance of the control samples was of gray color whereas the CNT doped specimens appeared to be of dark gray color. Particularly the samples with higher concentration of CNTs were clearly distinguishable from the low CNTs concentration cement pastes and control samples. Based on their characteristic physical appearance, therefore, samples could be easily identified. It also goes on to show that CNTs were uniformly dispersed in the cement paste. The Fig. 6a shows the hydrated segment of a 7 days old control specimen. A deep and wide cleavage is clearly visible in the specimen. As discussed above, a 7 days old sample is not completely hydrated as shown in the Fig. 6b and therefore materials is weak. The Fig. 6d shows a micrograph from the 28 days old control sample where lumps of the hydrated cement form voids and pores despite achieving a reasonable hydration of the cement paste. The hydrated and un-hydrated phases form small clusters therefore the cracks are uneven and in random directions. Fragments of the micro-sized particles are compacted together loosely in the form of lumps. Since there is still 20%-30% un-hydrated cement present in the paste so it is highly likely that the breaking process also involves slipping of the adjacent grains as seen in the Fig. 6d. These features indicate a weak composite structure. 16 Fig. 6 SEM micrographs for control specimens (a, b) 7 days old and (c, d) are 28 17 days old. The cement paste specimen mixed with the pristine CNTs, on the other hand, are remarkably distinct. The 7 and 28 days old specimens prepared with 0.2 wt.% CNTs are shown in the Fig. 7a & 7b. The Fig. 7a clearly indicates the presence of the large number of the small hydrated crystallites for a 7 days old specimen although it is not completely hydrated. Upon achieving a complete hydration, the crystallites grow bigger as shown in the Fig. 7b. The voids and the pores are almost filled due to the formation of the large clusters of the hydrated crystals. The cracks are narrowed down and shallow. The formation of the large hydrated crystals is assisted by the presence of the pristine CNTs which provide a platform for the growth of the crystals. The specimens prepared with the 0.4 wt% of pristine CNTs offer an interesting perspective though. The hydrated crystallites still grow on the CNT sites as shown in the Fig. 7c & 7d (7 and 28 days old respectively) but the size of the crystallites is not big. The increased density of the CNTs in the cement paste has provided more sites for the crystallites to grow but for the same cement-to-water ratio, therefore the crystallite size is reduced. This trend is shared by all the specimens prepared at the CNTs concentration of 0.4 wt% as will be shown later. Smaller crystallites would naturally leave voids making the structure weaker. This is why the specimens with the higher concentrations of CNTs conceded to a significantly lower force. 18 Fig. 7 SEM micrographs for cement paste mixed with the pristine CNTs. Specimens with the 0.2 wt% of CNTs (a) 7 days old, (b) 28 days old. Specimens with 0.4 wt% CNTs (c) 7 days and (d) 28 days old. 19 The behavior of the concrete specimens mixed with the hydroxyl (OH) functionalized CNTs is very interesting. The 7 and 28 days old specimens of 0.2 wt% CNTs mixed with the cement paste, shown in the Fig. 8a and 8b, were similar to the other specimens (see Fig. 7a and 7b for example) therefore, as expected, their response was the same. However, the cement paste specimens mixed with 0.4 wt% of CNTs exhibited an anomaly. The micrographs of such samples are shown in the Fig. 8c and 8d where small sized crystallites of the hydrated products are visible, however, in contrast to the rest of the specimens, these crystals do not grow further. Strongly bonded OH groups prohibit the further growth of the crystals. The adsorption of the cationic and anionic charges on the hydroxyl functionalized CNTs is not effective resulting into poor exothermic process which inhibits the growth of the crystals. As a matter of principle, this would produce a weaker structure. The yield point for these specimens (Table 3) appears at even lower point on the compressive strength scale making them the weakest structures in our experiments. 20 Fig. 8 SEM micrographs for cement paste mixed with the hydroxyl (OH) fucntionalized CNTs. Specimens with the 0.2 wt% of CNTs (a) 7 days old, (b) 28 days old. Specimens with 0.4 wt% CNTs (c) 7 days and (d) 28 days old. 21 The concrete specimens mixed with the carboxyl (COOH) functionalized CNTs have produced unique results. The addition of the 0.2 wt% of COOH functionalized CNTs increases the compressive strength of the cement paste substantially. These are the strongest specimens in the series of the experiments. The weakly bonded H atoms in the COOH group are easily stripped off leaving the strong negative -COO- charge on the surface of the CNTs. Crystallization of the ionic compounds present in the cement paste is thus facilitated by highly negatively charged CNTs surface providing a long range order to the crystal growth. This is exactly what we see the SEM micrographs in the Fig. 9a and 9b. The size of the hydrated crystallites is large and continuous. In some cases the cleavages appeared to be covered by the crystallites which could be misinterpreted as the bridging between the grains. However, following the trend, these specimens also give up early when the concentrations of COOH-CNTs is increased up to 0.4 wt%. Therefore the bridge formation and pore filling could not be a likely explanation. The Fig. 9c and 9d shows that for a higher COOH-CNT concentrations, the crystallite size is reduced for these specimens as well. As a matter of fact, the breaking point for the specimens mixed with 0.4 wt% of OH-CNTs and COOH-CNTs lie at the same value of average force within 5% experimental uncertainty. Therefore for a concentration of 0.2 wt%, COOH-CNTs mixed specimens are the strongest but as the concentration increases to 0.4 wt%, the strength reduces even below to that of control specimens. 22 Fig. 9 SEM micrographs for the cement paste mixed with the carboxyl (COOH) functionalized CNTs. Specimens with the 0.2 wt% of CNTs (at) 7 days old, (b) 28 days old. Specimens with 0.4 wt% CNTs (c) 7 days and (d) 28 days old. 23 The XRD pattern for the cement powder, shown in the Fig. 10, illustrates that the most of the cement is composed of tricalcium silicate (3CaO·SiO2), otherwise known as alite. This is one of the major phase responsible for the strength of the cement on the larger part. Phases of calcium carbonate (CaCO3) and silicon dioxide (SiO2) can also be identified as one of the major, though not very dominating quantitatively. The other unidentified but weaker phases are the oxides of Al, Fe and K etc. The average crystallite size of the powder is almost 40 nm. The random orientation of the crystallite shows that the grains are mostly rough and angular. Fig. 10 The XRD pattern for the powder cement The XRD pattern for the cement paste, and the paste mixed 0.2 wt% and 0.4 wt% of pristine CNTs is shown in the Fig. 11. The specimens mixed with hydroxyl (-OH) and carboxylic (-COOH) carbon nanotubes are not shown here as they were identical to the pristine CNTs. The top panel of the Fig. 11 clearly shows that the powdered cement has reacted with the water to form hydrates. The one of the main hydrate phases formed is the portlandite with a chemical formula of Ca(OH)2. In addition, the other major phases visible are larnite (Ca2O4Si) and ettringite (Al2Ca6H64O50S3) whereas the ferrite and aluminate phases overlap and are indistinguishable at this level of the resolution. Apparently the features of the top panel are 24 shared by the middle and bottom panel both qualitatively and quantitatively, indicating that there has been no chemical reactions between the phases of the cement and the CNTs. The enhanced strength, thus, has not come from any new chemical compound formed within the cement paste. Fig. 11 XRD patterns for the hydrated cement (top panel), cement paste mixed with pristine CNTs at a concentrations of 0.2 wt% (middle panel) and of 0.4 wt% bottom panel. The above discussion makes it imperative to look for an alternative explanation to understand the role of the CNTs in the strengthening mechanism of the cement paste. The clue appears from the FITR spectra of the cement paste samples as shown in the Fig. 12. In the fingerprint region (below the wave number 1500 cm-1), the FTIR signals have a reasonable agreement with the literature[17 -- 19] and have been discussed at length. Therefore no further interpretation of the signals is required in this region. However the diagnostic region (above 1500 cm-1) requires some careful attention. Only the stretching vibrations that would produce a change in the dipole moment of a molecule will be registered as IR signals in the spectrum. Therefore an IR signal from a symmetrical molecule would not appear in the spectrum. The 25 signal associated at the frequency of 3418 cm-1 in the Fig. 12 is considered to be the contribution of hydrogen bonds, particularly -OH bonds of Ca(OH)2 where -OH ions are symmetrically located around the Ca+2 ion (Fig. 13). Yet a broad signal of significant intensity appears in the IR spectra of almost all the specimens. It appears to be particularly stronger for the control specimens and for the specimens prepared with 0.2 wt% of CNTs but very weak for higher CNTs concentrations and is absent from the cement powder specimen for the obvious reasons. This is a clear indication that the Ca(OH)2 is asymmetrical in the cement paste. The symmetry is affected by the presence and concentration of the CNTs. An asymmetric molecule has a non-zero dipole moment and thus have an electrostatic interaction with the molecules and nano-objects adjacent to it. Fig. 12 FTIR spectra for the cement and cement paste specimens mixed with CNTs 26 4. The Theory The electrostatic interaction between the different compounds and molecules can be mapped through the electrostatic potential. In analogy to a single charge, the electrostatic potential of a molecule sets up an electric field in which it interacts with the other particles/molecules through the Coulomb force. Moreover, since there is a charge accumulation in the ionic as well as covalently bonded molecules, the dipole moment arises which is measured by the magnitude of the charge times the separation between the charges. Higher is the electronegativity of an atom, more it has the ability to keep the electrons closer to itself resulting in electron density build-up at the specific regions around a molecule leaving other regions partially positively charged. Since the Ca(OH)2 is the major compound found in the cement paste, therefore its interaction with a pristine CNT was modeled. The equilibrium geometry, electrostatic potential and dipole moment were calculated by using density functional theory (DFT) with the functional ωB97X-D and the polarization basis set 6-31G. The DFT assumes that the exact energy of the electronic states may be described as a function of electron density. The ωB97X-D is a range- separated hybrid (RSH)-generalized gradient approximation (GGA) functional that is not separated into exchange and correlation parts. The functional can be used for the long range non- bonded interactions. The polarization basis set is 6-31G* allows the displacement of electron distributions away from the nuclear positions thus polarizing the elements. This allows us to construct the electrostatic potential surface and dipole moments for the molecules. In bulk materials systems, where there are large numbers of dipolar ions and molecules interacting with each other, the probability of the random orientation of dipoles is small. The dipoles tend to align themselves and interact with their nearest neighbors in such a way that the overall potential energy of the material system is minimized. It is consistent with the fact that the hydration process is strongly exothermic bringing the hardened material in thermodynamic equilibrium with the environment, and thus increasing its entropy. The calculated electrostatic potential surface of portlandite, Ca(OH)2, is shown in the Fig. 13. The calcium is doubly positive ion (Ca+2) which donates is two electrons to two hydroxide (OH-) ions leaving them negatively charged. Note that the both OH- ions are on the opposite sides of Ca+2 therefore the net dipole moment for the Ca(OH)2 is zero. 27 Fig. 13 The arrangement of ions (left) and the electrostatic potential surface of (right) for portlandite Further, the electrostatic potential surface for a segment of a CNT, shown in the Fig. 14, illustrates the electronic charge density is higher where CNT is terminated. The effective charge on a CNT segment is almost zero and the dipole moment is very weak. However, in an inhomogeneous materials, CNTs are far from their perfect geometry. Instead, they are likely to be under stress from the adjoining crystallites as illustrated in the Fig. 15. Figure14 The arrangement of the atoms in a CNT and the electrostatic potential surface. 28 Fig. 15 Model of stressed segment of a CNT and the resulting electrostatic potential surface Note that in this configuration there is a charge imbalance and the dipole moment is significantly large, nearly 8 D, indicating that the electrostatic force applied by the molecule is strong. The Fig. 16 shows the implications of the electrostatic interaction between a segment of a pristine CNT and a molecule of Ca(OH)2. As discussed above, the Ca(OH)2 and CNT have symmetrical molecular structures however due to Coulombic interaction between the molecules, the molecular symmetry is broken for Ca(OH)2 and the dipole moment of the Ca(OH)2-CNT composite system becomes strong. It is well understood that the ion-dipole interaction is stronger than the dipole-dipole interaction. Therefore the increase in the concentration of CNTs in the cement paste will actually enhance the dipole-dipole interaction between the closest CNT molecules therefore the ion-ion and ion-dipole interaction density will reduce. This is the reason that the increased concentration of CNTs in the cement paste has a negative effect on its. Due to strongly exothermic hydration process and high entropy of the crystallites, the electrostatic interaction is a long range thus increasing the overall strength of the material. 29 Fig. 16 Interaction of Ca(OH)2-CNT electrostatic interaction. 30 5. Conclusions The specimens of the cement paste mixed with various proportions and types of CNTs were tested for their compressive strength. The specimens prepared with the 0.2 wt% of CNTs came out to be stronger compared to the ones with no CNTs. However, the specimens became weak by increasing the proportion of CNTs to 0.4 wt%. The SEM micrographs show that the hydrated crystals in the cement paste grow at the CNTs sites. Further, the order of the crystal growth is long range in the specimens with 0.2 wt% CNTs which enhances the strength of the cement paste. On the other hand, increased concentration of CNTs in a specimen provides more sites for the crystal growth inhibiting the development of the long range. Moreover, small crystal growth was observed in the specimens mixed with the hydroxyl (-OH) functionalized CNTs because of strong OH bonds present on the surface of CNTs surface. The concrete samples with the CNTs- COOH functional groups showed growth of large crystallites. Removal of H-bonds from the COOH groups leaves strongly negative surface charge on the CNTs on which ionic crystals develop. The SEM micrographs show no clear evidence of bridging between the adjacent grains in the cement paste specimens. The XRD patterns confirmed that the control and CNTs-mixed specimen have similar chemical composition therefore the source of the enhanced strength must be due to electrostatic interaction between the cement paste and CNTs. The presence of a broad signal in the diagnostic region of FTIR spectrum further confirms the observation that the molecules of hydrated crystals are asymmetric because of the electrostatic interaction with the disordered surfaces of the CNTs. Finally the modelled electrostatic interaction between a hydrated molecule and a CNTs does indeed show a strong ion-dipole interaction. The energy release during the hydration increases the entropy of the composite material system which necessarily reinforces the electrostatic interaction. 31 References [1] CEMBUREAU, Activity Report 2017 of The European Cement Association, 2017. www.cembureau.eu. [2] NRMCA, Concrete CO 2 Fact Sheet, 2008. doi:10.1016/j.ejso.2010.10.007. [3] A. Tamimi, N.M. Hassan, K. Fattah, A. Talachi, Performance of cementitious materials produced by incorporating surface treated multiwall carbon nanotubes and silica fume, Constr. Build. Mater. 114 (2016) 934 -- 945. doi:10.1016/j.conbuildmat.2016.03.216. [4] N.M. Hassan, K.P. Fattah, A.K. Tamimi, Modelling mechanical behavior of cementitious material incorporating CNTs using design of experiments, Constr. Build. Mater. 154 (2017) 763 -- 770. doi:10.1016/j.conbuildmat.2017.07.218. [5] Y. Tao, W. Zhang, D. Shang, Z. Xia, N. Li, W.Y. Ching, F. Wang, S. Hu, Comprehending the occupying preference of manganese substitution in crystalline cement clinker phases: A theoretical study, Cem. Concr. Res. 109 (2018) 19 -- 29. doi:10.1016/j.cemconres.2018.04.003. [6] W. Li, W.H. Duan, X. Li, S.J. Chen, G. Long, Y.M. Liu, Effects of Nanoalumina and Graphene Oxide on Early-Age Hydration and Mechanical Properties of Cement Paste, J. Mater. Civ. Eng. 29 (2017) 04017087. doi:10.1061/(asce)mt.1943-5533.0001926. [7] B. Lothenbach, G. Le Saout, E. Gallucci, K. Scrivener, Influence of limestone on the hydration of Portland cements, Cem. Concr. Res. 38 (2008) 848 -- 860. doi:10.1016/j.cemconres.2008.01.002. [8] W. Li, M. Asce, Z. Huang, T. Zu, C. Shi, W.H. Duan, S.P. Shah, M. Asce, Influence of Nanolimestone on the Hydration , Mechanical Strength , and Autogenous Shrinkage of Ultrahigh-Performance Concrete, 28 (2016) 1 -- 9. doi:10.1061/(ASCE)MT.1943- 5533.0001327. [9] S.S. Kutanaei, D. Ph, A.J. Choobbasti, Effects of Nanosilica Particles and Randomly Distributed Fibers on the Ultrasonic Pulse Velocity and Mechanical Properties of Cemented Sand, 29 (2017) 1 -- 9. doi:10.1061/(ASCE)MT.1943-5533.0001761. 32 [10] G.Y. Li, P.M. Wang, X. Zhao, Mechanical behavior and microstructure of cement composites incorporating surface-treated multi-walled carbon nanotubes, Carbon N. Y. 43 (2005) 1239 -- 1245. doi:10.1016/J.CARBON.2004.12.017. [11] A. Sedaghat, M.K. Ram, A. Zayed, R. Kamal, N. Shanahan, Investigation of Physical Properties of Graphene-Cement Composite for Structural Applications, (2014). doi:10.4236/ojcm.2014.41002. [12] K. Gong, S.M. Asce, Z. Pan, A.H. Korayem, D. Ph, L. Qiu, D. Li, F. Collins, C.M. Wang, W.H. Duan, A.M. Asce, Reinforcing Effects of Graphene Oxide on Portland Cement Paste, 27 (2015) 1 -- 6. doi:10.1061/(ASCE)MT.1943-5533.0001125. [13] M.-F. Yu, O. Lourie, M.J. Dyer, K. Moloni, T.F. Kelly, R.S. Ruoff, Strength and Breaking Mechanism of Multiwalled Carbon Nanotubes Under Tensile Load, Science (80-. ). 287 (2000) 637 LP -- 640. doi:10.1126/science.287.5453.637. [14] M. Szel, Mechano-Physical Properties and Microstructure of Carbon Nanotube Reinforced Cement Paste after Thermal Load, (2017) 1 -- 22. doi:10.3390/nano7090267. [15] H.F.W. Taylor, Cement Chemistry, 2nd ed., Thomas Telford, London, 1997. doi:http://dx.doi.org/10.1680/cc.25929. [16] M.S. Dresselhaus, G. Dresselhaus, R. Saito, A. Jorio, Raman spectroscopy of carbon nanotubes, Phys. Rep. 409 (2005) 47 -- 99. doi:10.1016/j.physrep.2004.10.006. [17] M.W. Grutzeck, F. Puertas, M.T. Blanco-Varela, M.L. Granizo, T. Vazquez, A. Palomo, Chemical stability of cementitious materials based on metakaolin, Cem. Concr. Res. 29 (2002) 997 -- 1004. doi:10.1016/s0008-8846(99)00074-5. [18] O.. Omotoso, D.. Ivey, R. Mikula, Containment mechanism of trivalent chromium in tricalcium silicate, J. Hazard. Mater. 60 (1998) 1 -- 28. doi:10.1016/S0304-3894(97)00037- X. [19] M. Yousuf, A. Mollah, T.R. Hess, Y.-N. Tsai, D.L. Cocke, An FTIR and XPS investigations of the effects of carbonation on the solidification/stabilization of cement based systems-Portland type V with zinc, Cem. Concr. Res. 23 (1993) 773 -- 784. 33 doi:10.1016/0008-8846(93)90031-4. 34
1906.09033	1	1906	2019-06-21T09:55:58	XPS evidence of degradation mechanism in hybrid halide perovskites	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	The paper presents the results of measurements of XPS valence band spectra of SiO2/MAPbI3 hybrid perovskites subjected to irradiation with visible light and annealing at an exposure of 0-1000 hours. It is found from XPS survey spectra that in both cases (irradiation and annealing) a decrease in the I:Pb ratio is observed with aging time, which unambiguously indicates PbI2 phase separation as a photo and thermal product of degradation. The comparison of the XPS valence band spectra of irradiated and annealed perovskites with density functional theory calculations of the MAPbI3 and PbI2 compounds have shown a systematic decrease in the contribution of I 5p-states and allowed us to determine the threshold for degradation, which is 500 hours for light irradiation and 200 hours for annealing.	physics.app-ph	physics	XPS evidence of degradation mechanism in hybrid halide perovskites I.S. Zhidkov1, A.I. Poteryaev2, A. Kukharenko1, L.D. Finkelstein2, S.O. Cholakh, A.F. Akbulatov3, P.A. Troshin3,4, C.-C. Chueh5 and E.Z. Kurmaev1,2 1Institute of Physics and Technology, Ural Federal University, Mira 9 str., 620002 Yekaterinburg, Russia 2M.N.Mikheev Institute of Metal Physics of Ural Branch of Russian Academy of Sciences, S.Kovalevskoi 18 str., 620108 Yekaterinburg, Russia 3Institute for Problems of Chemical Physics of the Russian Academy of Sciences (ICP RAS), Semenov prospect 1, Chernogolovka, 142432, Russia 4Skolkovo Institute of Science and Technology, Nobel street 3, Moscow, 143026, Russia 5Advanced Research Center for Green Materials Science and Technology, National Taiwan University Taipei 10617, Taiwan Abstract. The paper presents the results of measurements of XPS valence band spectra of SiO2/MAPbI3 hybrid perovskites subjected to irradiation with visible light and annealing at an exposure of 0-1000 hours. It is found from XPS survey spectra that in both cases (irradiation and annealing) a decrease in the I:Pb ratio is observed with aging time, which unambiguously indicates PbI2 phase separation as a photo and thermal product of degradation. The comparison of the XPS valence band spectra of irradiated and annealed perovskites with density functional theory calculations of the MAPbI3 and PbI2 compounds have showed a systematic decrease in the contribution of I 5p-states and allowed us to determine the threshold for degradation, which is 500 hours for light irradiation and 200 hours for annealing. Introduction 1. Hybrid halide perovskites MAPbX3 (MA=CH3NH3, X=Cl, Br, I) made a real revolution in photovoltaics and have now achieved outstanding efficiency comparable to and even greater than that in silicon [1]. The main problem now is to overcome their limited service life under operating conditions and improve their photo and thermal stability [2-4]. This is due to the fact that a fundamental understanding of the physicochemical processes that cause degradation is necessary for the development of perovskite solar cells with a high resource of their practical use. It is currently well established that external factors such as light, temperature, humidity and O2 can contribute to degradation processes [5-7]. One way of protecting against exposure to humidity and oxygen of ambient atmosphere is encapsulation of hybrid perovskites, which has already shown its effectiveness [8]. However, the problem of degradation caused by exposure to visible light and temperature, that are factors present in actual use of photovoltaic devices, still remains. To investigate the mechanisms of degradation in hybrid perovskites, the various methods were used including x-ray diffraction [9-10], electron microscopy [11], thermo 1 gravimetry analysis [9], photoluminescence [12], and first-principle calculations [13] of the formation energies of various structural defects. In this article, the mechanism of photo and thermal degradation of methylammonium lead iodide - MAPbI3 is investigated using x-ray photoelectron valence band spectra which are found to be an effective tool to probe the chemical bonding and electronic structure of solids and their comparison with specially performed density functional theory (DFT) calculations. 2. Experimental and Calculation Details Glass substrates (5 Ω, Luminescence Technology Corp.) were sequentially cleaned with toluene and acetone and sonicated in deionized water, acetone and isopropanol. The MAPbX3 precursor solutions in DMF (~ 45 w. %) were spin-coated at 5000 rpm inside a nitrogen glove box. The toluene (200 μL) was dropped on the film 4-5 s after initiation of the spin-coating, inducing the film crystallization. Spinning was continued for 45 s and then the deposited films were annealed at 100C for 15 min on a hotplate installed inside the glove box. The thermal aging experiments were conducted inside a nitrogen MBraun glovebox with O2<0.1 ppm and H2O<0.1 ppm using a calibrated hot plate as a heat source. The samples were placed on the hotplate at 90 oC and covered with a non-transparent lid to avoid exposure to the ambient light. The photochemical aging experiments were performed using specially designed setups integrated with the dedicated MBraun glove box using LG sulfur plasma lamp as a standard light source, which is known to provide a good approximation of the solar AM1.5G spectrum. The light power at the samples was ~70 mW/cm2, while the temperature was 45±2 oC (provided by intense fan cooling of the sample stage). X-ray photoelectron spectroscopy (XPS) was used to measure valence band spectra (VB) with help of a PHI XPS 5000 Versaprobe-spectrometer (ULVAC-Physical Electronics, USA) with a spherical quartz monochromator and an energy analyzer working in the range of binding energies (BE) from 0 to 1500 eV. The energy resolution was E ≤ 0.5 eV. The samples were kept in the vacuum chamber for 24 h prior to the experiments and were measured at a pressure of 10-7 Pa. Electronic structure calculations were performed using VASP code[14] with the standard frozen-core projector augmented-wave (PAW) method. The cut-off energy for basis functions was set to 500 eV. Perdew, Burke, Ernzerhof's generalized gradient approximation was used for exchange-correlation [15]. Atomic positions are relaxed since all the forces on atoms are below 0.05eV/A. The spin-orbital coupling was switched on due to strong relativistic effect of Pb. 76 2 and 196 k-points were utilized in the irreducible part of the first Brilluoin zone for CH3NH3PbI3 and PbI2, respectively. 3. Results and Discussion As shown by X-ray diffraction (XRD) measurements [10], the decomposition of MAPbI3 revealed the disappearance of diffraction peaks, which are characteristics of MAPbI3, and the appearance of PbI2 peaks, which suggests the following reaction: CH3NH3PbI3 → PbI2 solid + CH3NH3I → PbI2 solid + CH3I gas +NH3 gas To study the mechanism of photo and thermal degradation of perovskite, we first measured the XPS survey spectra of irradiated and annealed MAPbI3 samples with aging time of 0-1000 hours in the energy range of 0-600 eV. These data are presented in Fig. 1, and the surface compositions determined from these spectra (a quantitative estimate of the relative concentration of elemental species is obtained from the measured spectroscopic intensities, corrected by the elemental and orbital specific photoemission cross sections) are shown in the Table (in at.%). ] s t i n U y r a r t i b r A [ y t i s n e t n I XPS Survey SiO2/MAPbI3 O1s Pb4d as prepared photo-200 hrs photo-500 hrs photo-1000 hrs Pb4f C1s I4s (a) as prepared thermo-200 hrs thermo-500 hrs thermo-1000 hrs (b) Pb4f I4d I4p Pb4d O1s I4s C1s I4d I4p N1s N1s 600 500 400 300 200 0 600 100 500 Binding Energy [eV] 400 300 200 100 0 Fig. 1. XPS survey spectra of as-prepared, light irradiated (a) and annealed (b) SiO2/MAPbI3. Table. Surface composition of as-prepared, light irradiated and annealed SiO2/MAPbI3 (in at.%) SiO2/MAPbI3 As prepared Photo 200 h Photo 500 h Photo 1000 h Thermo 200 h Thermo 500 h Thermo 1000 h C 44,9 39,5 37 50,8 51 37,7 40 O 6,7 6,1 15,1 20,2 9,4 10,4 13,3 I 27,5 32,6 25,6 11,5 21,4 27,3 24 N 8,3 5,5 - - - - - I:Pb 2,66 2.37 1,72 1.29 1,65 1,67 1,67 Si 2,1 - 3,2 6,6 2,4 4 4,5 Pb 10,3 13,7 14,8 8,9 12,9 16,3 14,3 3 As it follows from this data, only elements belonging to perovskite (C, N, Pb and I) and the substrate (Si and O) were found on the surface of the studied samples. To test the proposed mechanism for the degradation of hybrid halide perovskite involving the precipitation of a PbI2 phase, it is important to determine the I:Pb ratio and its evolution with the time of irradiation and annealing. These ratios presented in the Table show their consistent decrease when increasing the dose of radiation with visible light and annealing from 0 to 1000 hours. However, if the general trend of a change of I:Pb ratio in attitude is preserved over aging time for both external influences, then the threshold values of degradation differ significantly. So for irradiation with visible light, the threshold degradation begins with a time of 500 hours (I:Pb=1.72) whereas for annealing already at 200 hours (I:Pb=1.65). ] s t i n U y r a r t i b r A [ y t i s n e t n I XPS VB SiO2/MAPbI3 Pb 5d (a) Pb 5d as prepared photo-200 hrs photo-500 hrs photo-1000 hrs PbI2 I 5p I 5s Pb 6s (b) I 5p as prepared thermo-200 hrs thermo-500 hrs thermo-1000 hrs PbI2 I 5s Pb 6s 24 20 16 12 8 4 0 24 20 16 12 8 4 0 Binding Energy [eV] Fig. 2. XPS Pb 5d+VB spectra of as-prepared, light irradiated (a) and annealed (b) SiO2/MAPbI3. Now let's see how the change in the I:Pb ratio affects the XPS valence band spectra. Since the degradation mechanism described above assumes the unchanged lead content in perovskite, the data shown in Figure 2 (a-b) were normalized to the intensity of the XPS Pb 5d- spectra. As follows from these spectra, an increase in the exposure time actually leads to a decrease in the relative intensity of the main band located at 0-7 eV reflecting the distribution of I 5p-states, which actually means a decrease in the contribution of these states with aging time. Here, the reference is the XPS VB spectrum of the PbI2 compound which is the final product of the photo and thermal decomposition of the MAPbI3 perovskite, and as a result we see that at the limiting aging time (1000 hours), the XPS VB spectra of the irradiated and annealed samples actually coincide with those of the PbI2 compound. It is curious that the threshold values of degradation expressed in reducing the contribution of I 5p-states exactly coincide with those 4 determined from the I:Pb ratios (see Table) and correspond to 500 hours for irradiation with visible light and 200 hours for annealing. (a) XPS VB SiO2/MAPbI3 as prepared photo-200 h photo-500 h photo-1000 h PbI2 as prepared thermo-200 h thermo-500 h thermo-1000 h PbI2 (b) Fig. 3. XPS Pb 5d ] s t i n U y r a r t i b r A [ y t i s n e t n I 24 22 20 18 16 24 22 20 18 16 Binding Energy [eV] spectra of as-prepared, light irradiated (a) and annealed (b) SiO2/MAPbI3. The next confirmation of the established patterns of XPS VB spectra behavior depending on the exposure time during photo irradiation and annealing are the data on the energy position of the XPS spectra of the Pb 5d-core levels shown in the high-energy part of Figure 2 (in the energy range of 17-24.5 eV) and on a larger scale in Fig. 3. As follows from these spectra, the binding energy of the XPS Pb 5d-core levels practically coincides for light irradiated at 200 hours and the initial sample, while a further increase in exposure time up to 500 hours leads to a high-energy shift, and continues at a dose of 1000 hours until full coincidence with the binding energy of the reference PbI2 compound (Fig. 3a). The similar trends are observed for the annealed samples (Fig. 3b) only the threshold values of the onset of degradation already appear at a dose of 200 hours. 5 XPS VB SiO2/MAPbI3 as prepared photo-200 hrs photo-500 hrs photo-1000 hrs PbI2 I 5s Pb 6s ] s t i n U y r a r t i b r A [ y t i s n e t n I (a) I 5p as prepared thermo-200 hrs thermo-500 hrs thermo-1000 hrs PbI2 (b) I 5p I 5s Pb 6s 14 12 10 8 6 4 2 0 14 12 10 8 6 4 2 0 Binding Energy [eV] Fig. 4. XPS VB spectra of as-prepared, light irradiated (a) and annealed (b) SiO2/MAPbI3. Fig. 5 Comparison of XPS and DFT valence band spectra for MAPbI3 (black) and PbI2 (orange). The experimental and theoretical data are shown by dots and lines, respectively. Analysis of the elemental and orbital compositions of the molecular levels of the [CH3NH3]- organic cation showed that in the sequence of its levels shown in Fig. 5: -7, -9.1 - 11.3, -14.7 eV, the first displays the energy position of the CH3 group (main contributions from H 1s and C 2p), the second - the binding energy of N-C bonds (N 2p and C 2p), and the third group - the binding energy of NH3-group (N 2p and H 1s). Further at the larger negative energies 6 (-14.7 eV) the levels with participation of C 2s and N 2s are located.. The main valence band (from -2.2 to -5.5 eV) is formed mainly by I 5p electrons. It is located closest to the CH3 group (- 7 eV) and has a small overlap with it due to the electrostatic interaction of a positively charged electric dipole with a negative charged PbI6 octahedra. The comparison of the experimental XPS valence band spectra of MAPbI3 and PbI2 and theoretical DFT calculations of these compounds presented in Fig. 5. shows overall good agreement. The XPS valence bands at -5.5÷-2.2 eV match well with its theoretical counterparts displaying a proper reduction of the I 5p intensities going from MAPbI3 to PbI2. I 5s narrow bands for both compounds are located about -14 eV and demonstrate the same change in relative intensities. Pb 6s bands lie at -9 eV and coincide with the experimental data. As stated before the methylammonium peaks are located at -14.7, -11.3, -9.1 and -7 eV. Due to weak hybridization of the CH3NH3 cation with the PbI6 octahedron these peaks are very narrow, less than 0.2 eV. One should remind here about the experimental energy resolution of 0.5 eV, that leads to the absence of the methylammonium peaks in the experimental picture. Summing up the research it should be noted that our measurements of XPS survey, 5d and valence band spectra showed that irradiation with visible light and heating (annealing) of MAPbI3 perovskite lead to a systematic decrease in the I:Pb ratio, high-energy shift of XPS Pb 5d-lines and decrease of the contribution of I 5p-states. In all the cases listed above, the spectral characteristics obtained for the limiting aging time (1000 hours) are close to the PbI2 compound that is the supposed product of decomposition upon irradiation and annealing. The decrease in the contribution of I 5p-states observed in the experiment up to coincidence with that in the PbI2 compound is confirmed by DFT-calculations of the electronic structure, which also show a decrease in the density of I 5p-states during the transition from compound MAPbI3 to PbI2. Thus, the proposed mechanism of photo- and thermal degradation associated with the decomposition of the hybrid perovskite MAPbI3 with PbI2 phase separation was confirmed by the measurements of x-ray photoelectron spectra and DFT-calculations. Conclusion The main experimental result of this work is the establishment of the fact that the heating of the hybrid perovskite is a stronger degrading factor than exposure to the visible light. We assume. that this is due to the physical nature of the organic cation [CH3NH3]- which is an electrical dipole participating in an electrostatic dipole-dipole interaction with neighbor dipoles and a negatively charged framework of octahedra with I- ions. According to [14], the energies of these 7 interactions are close to the energy of thermal phonons, therefore the probability of their absorption by the system can resonantly exceed the absorption of visible light, whose frequency is much higher than the frequency of phonons. Acknowledgements This work is supported by Russian Science Foundation (project 19-73-30020). The DFT calculations were supported by Ministry of Education and Science of Russia (task 3.7270.2017 / 8.9) and Theme "Electron" № АААА-А18-118020190098-5. References: [1] N. J. Jeon, H. Na, E. H. Jung, T. Y. Yang, Y. G. Lee, G. J. Kim, H. W. Shin, S. I. Seok, J. M. Lee, J. W. Seo, A fluorene-terminated hole-transporting material for highly efficient and stable perovskite solar cells, Nature Energy 3 (2018) 682. [2] Nasir Alia, Sajid Raufb, Weiguang Kongc, Shahid Alid, Xiaoyu Wanga, Amir Khesroe, Chang Ping Yangb, Bin Zhub,f, Huizhen Wu, An overview of the decompositions in organo- metal halide perovskites and shielding with 2-dimensional perovskites, Renewable and Sustainable Energy Reviews 109 (2019) 160 -- 186. [3] Caleb C. Boyd, Rongrong Cheacharoen, Tomas Leijtens, and Michael D. McGehee, Understanding Degradation Mechanisms and Improving Stability of Perovskite Photovoltaics, Chem. Rev. 119 (2019) 3418-3451. [4] Emilio J. Juarez-Perez, Luis K. Ono, Iciar Uriarte, Emilio J. Cocinero, and Yabing Qi, Degradation Mechanism and Relative Stability of Methylammonium Halide Based Perovskites Analyzed on the Basis of Acid−Base Theory, ACS Appl. Mater. Interfaces 2019, 11, 12586−12593. [5] Maryline Ralaiarisoa, Ingo Salzmann, Feng-Shuo Zu, and Norbert Koch, Effect of Water, Oxygen, and Air Exposure on CH3NH3PbI3 -- xClx Perovskite Surface Electronic Properties, Adv. Electronics Materials 1800307 -- 1800314 (2018). [6] Youzhen Li, Xuemei Xu, Congcong Wang, Ben Ecker, Junliang Yang, Jinsong Huang, and Yongli Gao, Light-Induced Degradation of CH3NH3PbI3 Hybrid Perovskite Thin Film, J. Phys. Chem. C 2017, 121, 3904−3910. [7] Tanzila Tasnim Ava, Abdullah Al Mamun, Sylvain Marsillac and Gon Namkoong, A Review: Thermal Stability of Methylammonium Lead Halide Based Perovskite Solar Cells, Appl. Sci. 2019, 9, 188-212. [8] Ji-Hyun Cha, Kyungkyou Noh, Wenping Yin, Yongju Lee, Yongmin Park, Tae Kyu Ahn, Alvaro Mayoral, Jaheon Kim, Duk-Young Jung, and Osamu Terasaki, Fast Formation and 8 Encapsulation of All-inorganic Lead Halide Perovskites at Room Temperature in Metal-Organic Frameworks, J. Phys. Chem. Lett. 10 (2019) 2270-2277. [9] Nam-Koo [9] Num-Koo Kim, Young Hwan Min, Seokhwan Noh, Eunkyung Cho, Gitaeg Jeong, Minho Joo, Seh-Won Ahn, Jeong Soo Lee, Seongtak Kim, Kyuwook Ihm, Hyungju Ahn, Yoonmook Kang, Hae-Seok Lee, Donghwan Kim, Investigation of Thermally Induced Degradation in CH3NH3PbI3 Perovskite Solar Cells using In-situ Synchrotron Radiation Analysis, Scientific Reports 7 (2017) 4645-4652. [10] E.Tenuta, C. Zheng & O. Rubel, Thermodynamic origin of instability in hybrid halide perovskites, Sci. Rep. 6 (2016) 37654-37661. [11] G. Divitini, S. Cacovich, F. Matteocci, L. Cinà, A. Di Carlo & C. Ducati, In situ observation of heat-induced degradation of perovskite solar cells, Nature Energy, 1 (2016) 15012-15017. [12] Khaoula Jemli, Hiba Diab, Ferdinand Lédée, Gaelle Trippé-Allard, Damien Garrot, Bernard Geffroy, Jean-Sébastien Lauret, Pierre Audebert and Emmanuelle Deleporte, Using Low Temperature Photoluminescence Spectroscopy to Investigate CH3NH3PbI3 Hybrid Perovskite Degradation, Molecules 21 (2016) 885-897. [13] Yuanbin Xue, Yueyue Shan, and Hu Xu, First-principles study on the initial decomposition process of CH3NH3PbI3, J. Phys. Chem. 147 (2017) 124702-124706. [14] Aron Walsh, Principles of Chemical Bonding and Band Gap Engineering in HybridOrganic−Inorganic Halide Perovskites, J. Phys. Chem. C 119 (2015) 5755-5760. 9
1912.11345	1	1912	2019-12-24T13:36:12	Development of High-Speed Ion Conductance Microscopy	[ "physics.app-ph" ]	Scanning ion conductance microscopy (SICM) can image the surface topography of specimens in ionic solutions without mechanical probe--sample contact. This unique capability is advantageous for imaging fragile biological samples but its highest possible imaging rate is far lower than the level desired in biological studies. Here, we present the development of high-speed SICM. The fast imaging capability is attained by a fast Z-scanner with active vibration control and pipette probes with enhanced ion conductance. By the former, the delay of probe Z-positioning is minimized to sub-10 us, while its maximum stroke is secured at 6 um. The enhanced ion conductance lowers a noise floor in ion current detection, increasing the detection bandwidth up to 100 kHz. Thus, temporal resolution 100-fold higher than that of conventional systems is achieved, together with spatial resolution around 20 nm.	physics.app-ph	physics	Development of High-Speed Ion Conductance Microscopy Shinji Watanabe,1, ∗ Satoko Kitazawa,2 Linhao Sun,1 Noriyuki Kodera,1 and Toshio Ando1, † 1WPI Nano Life Science Institute (WPI-NanoLSI), Kanazawa University, Kakuma-machi, Kanazawa 920-1192, Japan 2Department of Physics, Institute of Science and Engineering, Kanazawa University, Kakuma-machi, Kanazawa 920-1192, Japan (Dated: December 25, 2019) Scanning ion conductance microscopy (SICM) can image the surface topography of specimens in ionic solutions without mechanical probe -- sample contact. This unique capability is advantageous for imaging fragile biological samples but its highest possible imaging rate is far lower than the level desired in biological studies. Here, we present the development of high-speed SICM. The fast imaging capability is attained by a fast Z-scanner with active vibration control and pipette probes with enhanced ion conductance. By the former, the delay of probe Z-positioning is minimized to sub-10 µs, while its maximum stroke is secured at 6 µm. The enhanced ion conductance lowers a noise floor in ion current detection, increasing the detection bandwidth up to 100 kHz. Thus, temporal resolution 100-fold higher than that of conventional systems is achieved, together with spatial resolution around 20 nm. I. INTRODUCTION Tapping mode atomic force microscopy (AFM) [1] has been widely used to visualize biological samples in aqueous solution with high spatial resolution. However, when the sample is very soft, like eukaryotic cell surfaces, the intermittent tip-sample contact significantly deforms the sample and hence blurs its image [2 -- 4]. Moreover, when the sample is extremely fragile, it is often seriously damaged [4, 5]. SICM was invented to overcome this problem [6]. SICM uses as a probe an electrolyte-filled pipette having a nanopore at the tip end, and measures an ion current that flows between an electrode inside the pipette and another electrode in the external bath solution. The ionic current resistance between the pipette tip and sample surface (referred to as the access resistance) increases when the tip approaches the sample. This sensitivity of access resistance to the tip -- sample distance enables imaging of the sample surface without mechanical tip-sample contact [7, 8] (Fig. 1). To improve fundamental performances of SICM, several devices have recently been introduced, including a technique to control the pore size of pipettes [9 -- 11] and a feedback control technique based on tip -- sample distance modulation [12 -- 14]. Moreover, the SICM nanopipette has recently been used to measure surface charge density [15 -- 21] and electrochemical activity [22, 23] as well as to deliver species [24 -- 27]. Thus, SICM is now becoming a useful tool in biological studies, especially for characterizing single cells with very soft and fragile surfaces [28]. However, the imaging speed of SICM is low; it takes from a few minutes to a few tens of minutes to capture an SICM image, which is in striking contrast to AFM. High-speed AFM is already established [29] ∗ [email protected]; † [email protected]; and has been used to observe a variety of proteins molecules and organelles in dynamic action [30]. The slow performance of SICM is due mainly to a low signal-to-noise ratio (SNR) of sensing, resulting in its low detection bandwidth (and hence low feedback bandwidth). Moreover, the low resonant frequency of the Z-scanner also limits the feedback bandwidth. ion current When the vertical scan of the pipette towards the sample is performed with velocity vz, the time delay of feedback control tdelay causes an overshoot for the vertical scan by tdelay × vz. This overshoot distance should be smaller than the closest tip -- sample distance (dc) to be maintained during imaging (see the approach curve in Fig. 1). That is, vz ≤ dc tdelay . (1) An appropriate size of dc is related to the pipette geometry, such as the tip aperture radius ra, the cone angle θc, and the outer radius of the tip ro [7, 31, 32], but dc ≈ 2ra is typically used to achieve highest possible resolution. The size of tdelay can be roughly estimated from the resonant frequency of the Z-scanner fz and the bandwidth of ion current detection Bid, as tdelay ≈ In typical SICM setups, the values of 1/fz + 1/Bid. these parameters are ra ≈ 20 nm, fz ≈1 kHz and Bid ≈ 1 kHz, yielding vz < 20 µm/s. In the representative SICM imaging mode referred to as the hopping mode [33], the tip-approach and retract cycle is repeated for a distance (hopping amplitude) of slightly larger than the sample height, hs. For example, when vz = 20 µm/s is used for the sample with hs ≈ 1 µm, it takes at least > 50 ms for pixel acquisition, which depends on the retraction speed. This pixel acquisition time corresponds to an imaging acquisition time longer than 8.3 min for 100 × 100 pixel resolution [34]. When the pipette retraction speed can be set at much larger than the approach speed vz, the imaging acquisition time can be improved but not much. Several groups have attempted to increase vz [13, 34 -- 38]. One of approaches used is to mount a shear piezoactuator with a high resonant frequency (but with a small stroke length) on the Z-scanner and this fast piezoactuator is used as a 'brake booster' [34, 35]; that is, this piezoactuator is activated only in the initial retraction phase where the tip is in close proximity to the surface. This method could cancel an overshooting displacement and therefore increase vz by 10-fold. Another approach is to increase Bid by the improvement of the SNR of current signal detection with the use of a current-source amplification scheme [37] or by the use of AC bias voltage between the electrodes (the AC current in phase with the AC bias voltage is used as an input for feedback control) [13, 38]. This bias voltage modulation method is further improved by capacitance compensation [14]. The improvement of SICM speed performance by these methods is however limited to a few times at most. Very recently, two studies demonstrated fast imaging of live cells with the use of their high-speed SICM (HS-SICM) systems [39, 40]. However, one of these studies used temporal tip -- sample contact to alter hopping amplitude [39], while the other used pipettes with ra = 80 -- 100 nm and abandoned optical observation of the sample [40]. Note that in SICM the temporal resolution has a trade-off relationship with the spatial resolution, as in the cases of other measurement techniques. Thus far, no attempts have been made to increase both fz and Bid extensively, without compromise of the spatial resolution and non-contact imaging capability of SICM. Here, we report the development of HS-SICM and demonstrate its high-speed and high resolution imaging capability. The image rate was improved by a factor of ∼100 or slightly more. This remarkable enhancement in speed was achieved by two improved performances:(i) fast pipette positioning achieved with the developed fast scanner and vibration suppression techniques and (ii) an enhanced SNR of current detection by reduction of the ionic resistance arising from the inside of the pipette (referred to as the pipette resistance). The improved fz of the Z-scanner resulted in a mechanical response time of ∼3 µs, corresponding to a ∼100-fold improvement over conventional SICM systems. The SNR of current detection was improved by a factor of ∼8, enhancing Bid from 1 to 100 kHz or slightly higher. The HS-SICM system was demonstrated to be able to capture topographic images of low-height biological samples at 0.9 -- 5 s/frame and live cells at 20 -- 28 s/frame. These high imaging rate performances are compatible with spatial resolution of 15 -- 23 nm. 2 FIG. 1. Working principle of SICM. The electrolyte-filled pipette with a nanopore at its end (see the transmission electron micrograph in the left panel) is mounted on the scanner. The ion current through the nanopore generated by the application of bias voltage between the two Ag/AgCl electrodes is measured by the ion current detector. The measured ion current, which is dependent on the tip -- surface separation d, is used as a pipette Z-position control signal. II. RESULTS AND DISCUSSION A. Strategy towards HS-SICM The speed of pipette approach towards the sample (vz) is limited, as expressed by Eq. (1). As vz depends on fz and Bid, the improvement on both fz and Bid is required to achieve HS-SICM. To increase fz, we need a fast Z-scanner for displacing the pipette along its length. Note that all commercially available Z-scanners for SICM have fz < 10 kHz. Besides, we need to establish a method to mount the pipette (∼15 mm in length) to the Z-scanner in order to minimize the generation of undesirable vibrations of the pipette. We previously developed a fast XYZ scanner with fz ≈ 100 kHz, a resonant frequency of ∼2.3 kHz in the XY directions, and stroke distances of ∼6 µm and ∼34 µm for Z and XY, respectively [41]. In this study, we further improved the dynamic response of this fast scanner. Considering this high fz with improved dynamic response, we need to increase Bid to the level of ∼100 kHz. As the ion current change caused by an altered tip-sample distance is generally small (∼1 pA), the current signal noise largely limits Bid. At a high frequency regime (> 10 kHz), the dominant noise source is the interaction between the amplifier's current noise and the total capacitance at the input [42 -- 44]. Therefore, we have to lower the total capacitance and increase the current signal to achieve Bid ≈ 100 kHz, without increasing the pipette pore size. B. High-speed Z-scanner The structure of our fast scanner developed is shown in Fig. 2a-c. A key mechanism for minimizing unwanted Z-scanner vibrations is momentum cancellation; the hollow Z-piezoactuator is sandwiched with a pair of X Y Z (cid:3) Approach curveScannerAZDistanceIon currentZsetpointContactNon-contact∆~1%dSampleIon current detector30 nmPipetted3 FIG. 2. XYZ-scanner used in this study. (a) Assembly drawing of the scanner. For the lateral scan, the displacements of X- and Y-piezoactuators are magnified via a lever arm amplification scheme. (b) Exploded view of the Z-scanner. (c) Assembly drawing showing the inertia balance design used for the Z-scanner. (d) Schematic of FF and FB damping control methods. (e) Transfer functions (frequency response) of Z-scanner. The blue and red lines indicate the gain signals measured without and with the damping control methods, respectively. The green broken and solid lines indicate the phase signals measured without and with the damping control methods, respectively. (f ) Time domain responses of the Z-scanner in the application of a square-like waveform voltage to the input (black line). The blue and red lines show the measured Z-scanner displacements without and with the use of the damping control methods, respectively. identical diaphragm-like flexures, so that the center of mass of the Z-piezoactuator hardly changes during its fast displacement (Supplementary material, SI 1). The pipette is mechanically connected only with the top flexure through being glued to the top clamp. Thanks to these designs, no noticeable resonance peaks are induced except at the resonant frequency of the Z-piezoactuator (Fig. 2e). We achieved a product value of 6 µm (maximum displacement) × 100 kHz (resonant frequency) in this Z-scanner, which exceeds more than 10-fold the value of conventional designs of SICM Z-scanner, 25 µm × 1 -- 2 kHz. In the present study, we further improved the dynamic response of Z-scanner. The sharp resonant peak shown in Fig. 2e (blue line) induces unwanted vibrations. In fact, the application of a square-like-waveform voltage to the Z-scanner (black line in Fig. 2f) generated an undesirable ringing displacement of the Z-scanner (blue line in Fig. 2f). To damp this ringing, we developed feedforward (FF) and feedback (FB) control methods (Fig. 2d). The FF control system was implemented in field-programmable-gate-array (FPGA). The gain controlled output signal from a mock Z-scanner (an electric circuit) with a transfer function similar to that of the real Z-scanner was first differentiated and then subtracted from the signal input to the Z-piezodriver [45]. Although this method was effective in reducing the Q-factor of the Z-scanner, the drift behavior of the transfer function of real Z-scanner would affect the reduced Q-factor during long-term scanning. To suppress the drift effect, the FB control implemented in an analog circuit was added as follows. The velocity of Z-scanner displacement was measured using the transimpedance amplifier via a small capacitor of ∼1 pF positioned near the Z-scanner. The gain-controlled velocity signal was subtracted from the output of the FF controller. In this way, too fast movement of the Z-scanner was prevented [46], resulting in nearly complete damping of unwanted vibrations, as shown with the red lines of Figs. 2e and 2f. Thus, the open-loop response time of Z-scanner, Q/πfz, was improved from ∼18.5 µs to 1.8 µs (critical damping). Note that measured Z-scanner ElongatingZ-displacementZ-supportFlexureGlueShrinkingTipZ-piezoCenterholezyxzyxScanner (XYZ)Y-piezoX-piezoTipZ-piezoDiaphragmflexureTip holderZ-supportPiezoDriverZd/dtΣ-Σ-++InputDigitalFF dampingI/VMock-ZFB dampingAnalogdefGain: w/o Ctrl.Gain: w/ Ctrl.100 µs40 nmw/ Ctrl.w/o ctrl.Drivesignalabc103104105Frequency (Hz)5dB360(cid:707)Phase: w/o Ctrl.Phase: w/ Ctrl.displacements (blue nad red lines) include the latency of the laser vibrometer used (2 µs) (Supplementary material, SI 1, Fig. S2). The FF/FB damping control was also applied to the XY scanners to improve their dynamic response (Supplementary material, SI 1, Fig. S3). C. Enhancement of SNR with Salt Concentration Gradient We describe here a method to improve Bid by increasing the SNR of current signal sensing. In the frequency region > 10 kHz, the dominant noise source of the ion current detector is the total capacitance at the transimpedance amplifier input, ΣC [42, 43]: IRMS ∝ B3/2 id Σ C, (2) where IRMS represents a root-mean-square current noise. The electrode-wiring and the pipette capacitance Cp dominate the total capacitance. As the Cp derives from the part of the pipette immersed in solution, thicker wall pipettes are useful in reducing Cp. We used quartz capillaries with a wall thickness of 0.5 -- 0.7 mm. The total capacitance in our setup was estimated to be ∼5 pF (Supplementary material, SI 2), yielding IRMS ∼ 8 pA at Bid = 100 kHz (although IRMS was ∼1.2 pA at Bid = 10 kHz), which was still too large. Then, we decided to increase the ion current to improve the SNR further. Since the bias voltage (Vb) larger than a typical value of ± 0.5 V induces an unstable ion current [47], we need to reduce the pipette resistance (Rp). The ion current Ii through the pipette opening is approximately described as Ii(d) = Vb Ra(d) + Rp , (3) In Eq. where d is the tip-surface distance and Ra is the access resistance that depends on d [48]. 3, the surface charge-dependent ion current rectification in the pipette is not considered [49]. Rp is usually ∼100-times larger than Ra even at d ≈ dc, and therefore, the reduction of Rp directly increases Ii. To reduce Rp, we examined the ion concentration gradient (ICG) method; a pipette back-filled with a high salt solution is immersed in a low salt solution. Since the pipette opening is very small, a concentration gradient is expected to be formed only in the close vicinity of the pipette opening. Although several studies have been performed on ICG from the viewpoint of its effect on the ion current rectification in nanopores [50 -- 52], it is unclear whether or not the ICG method is really useful and applicable to SICM, as the physiological salt concentration used in the external bath solutions is relatively high. To check this issue, we first performed a finite element method (FEM) simulation using the coupled Poisson -- Nernst -- Planck 4 FIG. 3. Spatial distribution of average concentration of K+ and Cl− obtained by FEM simulation. The surface charge density of the tip was set at −10 mC/m2. (a) Average ion concentration profile (red) and its derivative (blue) along the white arrow shown in (b). cin (KCl) = 4 M, cout (KCl) = 0.15 M and Vb = 0.01 V were used. The vertical axis represents the Z-distance from the tip aperture normalized with the tip aperture diameter; the tip aperture position is zero as indicated by the broken line. (b) Spatial distribution of average ion concentration under the same conditions as (a). (c) Enhancement of tip conductivity by ICG. The vertical axis represents the enhancement factor of the tip conductivity with respect to the ion conductivity at cin = 0.15 M. (d, e) Spatial distributions of average ion concentration at Vb = −0.1 V(d) and Vb = 0.1 V(e) for cin (KCl) = 4 M and cout (KCl) = 0.15 M. (PNP) equations that have been widely adopted to study the transport behavior of charged species [20, 53, 54]. Full details of our PNP simulation setup are described in Supplementary material, SI 3 and Methods. Figures 3a,b show a FEM simulation result obtained for the spatial profile of total ion concentration (cK+ + cCl− )/2, when 4 M KCl and physiological 0.15 M KCl solutions were used for the inside and outside of the pipette, respectively. As seen there, the region of ICG is confined 024Concentration (M)Derivative (a.u.)Normalized distance0-20-4020400.150.3Concentration (M)0246810ExperimentSimulationNormalizedconductivity Concentration inside the pipette, cin (M)0.11abcA0.150.3Concentration (M)Vb = -0.10 VdVb = 0.01 VePositive0.150.3Concentration (M)NegativeVb = +0.10 Vin a small volume around the pipette opening, while the outside salt concentration is maintained at ∼0.23 M and < 0.17 M in the regions distant from the opening by ∼2ra and > 4ra, respectively. Figure 3c (square plots) shows a simulation result for changes of ion conductance 1/Rp when the KCl concentration inside the pipette (cin) was altered, while the outside bulk KCl concentration (cout) was kept at 0.15 M. This result was very consistent with that obtained experimentally (red plots in Fig. 3c). The value of 1/Rp at cin = 4 M was ∼8-times larger than that at cin = 0.15 M. We also confirmed that the conditions of cin = 4 M and cout = 0.15 M generate a steady current with Vb < 0.5 V, and hence, allow stable SICM measurements for 7.5 nm ≤ ra ≤ 25 nm. Note that the high KCl concentration region can be confined to a smaller space when a negative bias voltage is used because of an ion current rectification effect of the negatively charged pipette (Fig. 3d, e). To confirm the SNR enhancement of Ii by ICG formed by the use of cin = 4 M and cout = 0.15 M, we measured the dynamic responses of Ii to quick change of d under Vb = 0.5 V, in the presence and absence of IGC. To measure the responses, the pipette with ra = 10 nm was initially positioned at a Z-point showing 5% reduction of Ii (see Fig. 4a). Then, the pipette was quickly retracted by 10 nm within 14 µs, and after a while quickly approached by 10 nm within 14 µs(Fig. 4b, Top), by the application of a driving signal with a rectangle-like waveform (Fig. 4b, Bottom) to the developed Z-scanner. The ion current responses measured using the transimpedance amplifier with Bid = 400 kHz are shown in Fig. 4b (Middle). With ICG, a clear response was observed (blue line), whereas without ICG no clear response was observed (red line) due to a large noise floor at this high bandwidth. With ICG, the SNR of detected current response increased linearly with increasing Vb (Fig. 4c, blue plots; Supplementary material, SI 4, Fig. S8), although the instability of detected Ii was confirmed at Vb > 0.5 V (not shown). Thus, the SNR of current detection was ∼8 times improved by the ICG method (Fig. 4c). The rising and falling times of the measured current changes with ICG were indistinguishable from those of the piezodriver voltage (Fig. 4b, Middle and Bottom), indicating no noticeable delay (< ∼2 µs) in the measured current response. Note that the physically occurring (not measured) response of current change (or the rearrangement of ion distribution) must be much faster than the response of measured current changes, because the actual response is governed by the local mass transport time in the nanospace around the pipette opening. The response time is roughly estimated to be 133 ns by adopting the diffusion time (τ ) required for ion transport by a distance of 2ra = 20 nm: 2ra = 2Dτ , where D is the nearly identical diffusion coefficient of K+ and Cl− in water (∼1.5 × 10−9 m2/s) [55]. √ Contrary to our expectation, the normalized approach curve (Ii vs d) was nearly identical between the pipettes 5 FIG. 4. Enhancement of ion current response by ICG. (a) Approach curves with (blue) and without (red) ICG method. (b) Dynamic response of ion current at Vb = 0.5 V when the tip is vertically moved (shown in green) in close proximity to the glass surface, and its dependence on the use (shown in blue) and non-use (shown in red) of ICG method. (c) Increase of SNR of ion current measurement with increasing Vb and its dependence on the use (blue) and non-use (red) of ICG method. with and without ICG (Fig. 4a) although their Rp values were largely different. This indicates a nearly identical Ra/Rp ratio between the two cases. This result was confirmed by FEM simulations performed by the use of various surface charge densities of pipette and substrate in a range of 0 -- 20 mC/m2 (Supplementary material, SI 5). In the final part of this subsection, we considered how SICM measurements with ICG would affect the membrane potential of live cells in a physiological solution. The ICG modulates local ion concentrations around the pipette tip end, which might induce a change in the local membrane potential only when the tip is in the close vicinity to the cell surface. However, it is difficult to perform experimental measurements of such a transient change of the local membrane potential. Here we estimated this change for nonexcitable HeLa cells used in this study and typical excitable cells, using the Goldman-Hodgkin-Katz voltage equation [56, 57]. In this estimation, extracellular ion concentrations around the tip end were obtained by FEM simulations. Full details of this analysis are described in Supplementary material, SI 3. For both cell types, we found that their local membrane potentials were changed by the pipette tip with ICG to various extents depending on the value of Vb (see Supplementary material SI 3, Tab. S3) . However, for nonexcitable HeLa cell, we expect that the net contribution of an ICG-induced local membrane potential change is negligible as the tip pore size is very small. It may not be however negligible for excitable cells, because a local membrane potential change would trigger the opening of voltage-gated sodium ion channels and thus generate action potential, which would propagate over the cell membrane. A quantitative estimation 0.90.940.981.02w/o ICGw/ ICGNomalized currentDistance20 nmaPiezo drive10 nmw/o ICGw/ ICGIon current1 nA100 µsGlass substrate10 nmTimec05101500.20.40.60.81w/o ICGw/ ICGSNRVbb6 are shown with the green and purple lines in Fig. 5, respectively), while the ion current was measured using the transimpedance amplifier with Bid = 400 kHz. For the initiation of ∼1.3 µm retraction of the pipette by feedback control, the set point of ion current was set at 98% of the reference ion current (i.e., 2% reduction). In the approaching regime, the ion current decreased as the tip got close proximity to the surface (red line in Fig. 5). However, in the retraction regime following the deceleration phase, the detected ion current behaved strangely; Ii initially decreased rather than increased and then reversed the changing direction, similar to the behavior of pipette Z-velocity (Supplementary material, SI 6, Fig. S10). We confirmed that this abnormal response of Ii was due to a leakage current caused by a capacitive coupling between the Z-piezoactuator and the signal line detecting Ii. We could mitigate this adverse effect by subtracting the gain-controlled Z-velocity signal from the measured Ii (Fig. S10). Although this abnormal response could not be completely cancelled as shown with the pink line in the shaded region of Fig. 5, it affected neither the feedback control nor SICM imaging. This is because the Ii signal in the retraction regime is not used in the operation of SICM. In the repeated approach and retraction experiments with the use of ICG method (Fig. 5), we achieved vz = 7.3 µm/ms for ra = 25 nm, corresponding to 63% of the value estimated as 2 × 25 nm/[1.8 µs + (400 kHz)−1] = 11.6 µm/ms. This vz value attained here is more than 300 times improvement over the vz value used in a recent SICM imaging study on biological samples (50 nm/ms for ra = 50 nm) with a conventional design of SICM Z-scanner [34]. Very recently, the Schaffer group successfully increased vz up to 4.8 µm/ms for ra = 80 -- 100 nm using their sample stage scanner and a step retraction sequence called 'turn step' [40]. Our vz value achieved for even 3 -- 4 times smaller ra still surpasses their result. We emphasize that the leakage of high salt to the outside of the tip is too small to change the bulk concentrations of ions outside and inside the tip. In addition, the region of ICG is confined to the vicinity of the tip for Vb values (Vb ≤ 0.1 V) typically used in SICM measurements (Figs. 3d, e), and therefore, the sample remains in the bath salt condition most of time as the time when the sample stays within a distance of ∼2 × ra from the tip opening is very short (∼10 µs). FIG. 5. Evaluation of improved approach velocity. The pipette tip was periodically moved in the z-direction, in close proximity to the glass substrate (right panel). (Left panel) The green line indicates the time course of tip displacement estimated from the Z-scanner's drive voltage. The red line indicates the detected ion current signal. The purple line indicates the velocity of tip displacement estimated from the output current of the Z-piezodriver. (Bottom panel) An enlarged view showing these three quantities. The ion current signal in the shaded region (shown in pink) is a false one (mostly leakage current) caused by a capacitive coupling between the Z-piezoactuator and the signal line of ion current detection. A set point value of 2% was used. for this possibility is beyond the scope of the present study. Nevertheless, our FEM simulations indicate that the ICG-induced local membrane potential change can be attenuated by the use of different Vb values and/or high concentration of NaCl solution instead of 4 M KCl solution (Supplementary material, SI 3, Tab. S4). D. Evaluation of Improved vz Here we describe a quantitative evaluation of how significantly the pipette approach velocity is improved by the ICG method and the developed Z-scanner. Besides, we describe a problem we have encountered during this evaluation. For this evaluation, the pipette filled with 4 M KCl was vertically moved above the glass substrate in 0.15 M KCl solution (the Z-displacement and its velocity E. High-Speed Imaging of Grating Patterns We evaluated the performance of our HS-SICM system by capturing topographic images of a sample made of polydimethylsiloxane that had a periodic 5 × 5 µm2 checkerboard pattern with a height step of 100 nm (Grating 1). For this imaging in hopping mode, we used ra = 5 -- 7.5 nm and Bid = 100 kHz, values smaller than those used in the above evaluation test. Therefore, we reduced vz to 150 -- 550 nm/ms. Other imaging conditions 250 µs1 µmPiezo driveIon currentZ-velocitySubstrateZIi2 %SetpointApproachRetracta.u.200 nm10 µsSetpointApproachRetractDecelerationEnlarged view2 %Z-velocityminimumIon currentMinimumPiezo driveMinimum7 FIG. 6. HS-SICM images captured for grating samples. (a, b) Areas of 25 × 25 µm2 were imaged at ∼8 s/frame with 100 × 100 pixels for Grating 1 (a) and Grating 2 with a rougher surface (b). (c) Height profiles of Grating 1 and Grating 2 along the arrows shown in (a) and (b). (d -- f ) Images of the region shown with the small rectangle in (b) captured at ∼3.5 (d), ∼8.5 (e) and ∼12 s/frame (f ). A set point value of 2% was used to capture these images. The fast scan direction is from bottom to top in these imaging experiments. FIG. 7. Topographic images of edge region of HeLa cell on glass substrate successively captured at 20 -- 28 s/frame with 200 × 100 pixels, under a pixel rate of 650 -- 1000 Hz, hopping amplitude of 300 -- 500 nm and Vb = −0.1 V. A set point value of 1.5% was used. The fast and slow scanning directions are from bottom to top and from left to right, respectively. are Vb = −0.3 V and number of pixels = 100 -- 400 × 100. Figure 6a shows a topographic image of Grating 1 captured at 8 s/frame over a 25 × 25 µm2 area with 100 × 100 pixels. Figure 6b shows a topographic image of a rougher surface area of another grating sample (Grating 2) containing an object with a height of ∼500 nm (Fig. 6c). Even for this rougher surface, its imaging was possible at 8 s/frame. Figures 6d -- f show images of a narrower area of Grating 2 marked with the small rectangle in Fig. 6b, captured at 3.5, 9 and 12 s/frame, respectively. Although fine structures were more visible in the images captured at 9 and 12 s/frame, this difference was not due to the lower imaging rates but due to the larger number of pixels. Averaged pixel rates were 2.85, 4.44 and 3.33 kHz for Figs. 6d, e and f, respectively, demonstrating high temporal and spatial 500 nm500 nm500 nmde~ 3.5 s/frame~ 9 s/framef~ 12 s/frame300 nm5000 nm5000 nmab~ 8 s/frame~ 8 s/frame160 nm1000 nmGrating 2Grating 1cGrating 2Grating 1100 X 100 pixels100 X 100 pixels100 X 100 pixels400 X 100 pixels400 X 100 pixels200 nm5 µmHeight profile2.5 µm250 nmt = 0st = 28st = 56st = 83st = 110st = 130s8 FIG. 8. Topographic images of microvilli dynamics on HeLa cell captured with HS-SICM. These images were successively captured at ∼22 s/frame with 100 × 100 pixels, under a pixel rate of 455 Hz, hopping amplitude of 500 -- 600 nm and Vb = −0.2 V. The numbers shown in each frame are the frame number. The red arrows indicate a microvillus undergoing growth and disappearance. The bottom right figure indicates the height profiles along the blue (1) and green (2) arrows shown in the frame 15. A set point value of 1% was used to capture these HS-SICM images. The fast and slow scanning directions are from left to right and from bottom to top, respectively. resolution of our HS-SICM. Note that the imaging rate depends not only on vz and a number of pixels but also on the hopping amplitude, hopping rate and the performance of the lateral movement of our scanner. In Supplementary material, Fig. S11, we show examples of images captured at higher rates (∼4 and ∼0.3 s/frame). F. High-Speed Imaging of Biological Samples Next, we examined the applicability of our HS-SICM system to biological samples. The first test sample is a live HeLa human cervical cancer cell. The imaging was carried out in hopping mode using a pipette with ra = 5 -- 7.5 nm, Bid = 100 kHz and Vb = −0.1 V. Figure 7 shows topographic images of a peripheral edge region of a (cid:24)(cid:19)(cid:19)(cid:3)(cid:81)(cid:80)(cid:24)(cid:19)(cid:19)(cid:3)(cid:81)(cid:80)(cid:20)(cid:21)(cid:22)(cid:23)(cid:24)(cid:25)(cid:26)(cid:27)(cid:28)(cid:20)(cid:19)(cid:20)(cid:20)(cid:20)(cid:21)(cid:20)(cid:22)(cid:20)(cid:23)(cid:20)(cid:24)(cid:20)(cid:25)(cid:20)(cid:26)(cid:20)(cid:27)(cid:20)(cid:28)(cid:21)(cid:19)(cid:21)(cid:20)(cid:21)(cid:21)(cid:21)(cid:22)(cid:21)(cid:23)(cid:21)(cid:24)(cid:21)(cid:25)(cid:21)(cid:26)50 nm20 nmHeight profile (frame 15)1212HeLa cell locomoting on a glass substrate in a phosphate buffer saline, captured at 20 -- 28 s/frame with 200 × 100 pixels for a scan area of 12 × 12 µm2. During the overall locomotion downwards until the cell disappearance from the imaging area within 2 min, the sheet-like structures (lamellipodia) with ∼100 nm height were observed to In this imaging, the pixel rate was grow and retract. 670 -- 1000 Hz, making a large contrast with the pixel rate of ∼70 Hz used in previous hopping-mode-SICM imaging of live cells without significant surface roughness [34, 35]. Additional HS-SICM images capturing the movement of a HeLa cell in a peripheral edge region are provided in Figs. S12 and S13 (Supplementary material, SI 7). Bright-field optical microscope images before and after these SICM measurements are also provided in Fig. S14. In these imaging experiments, ra = 2 -- 3 nm, Bid = 20 kHz, Vb = 70 mV, pixel rate = 250 Hz, and hopping amplitude = 350 nm are used. cells with significant To demonstrate the applicability of our HS-SICM surface system also to live roughness, we next imaged a central region (2 × 2 µm2) of a HeLa cell at 22 s/frame, using pipettes with ra = 5 -- 7.5 nm, Bid = 100 kHz, Vb = −0.2 V, and hopping amplitude of 600 nm (Fig. 8 and Supplementary Movie 1). The captured images show moving, growing and retracting microvilli with straight-shaped and ridge-like structures [5, 39, 58]. The red arrows on frames 11 -- 21 in Fig. 8 indicate the formation and disappearance of a single microvillus. During this dynamic process captured with a pixel rate of 0.56 kHz, the full width at half maximum (FWHM) of the microvillus was less than 50 nm (lower right in Fig. 8), when it was analyzed for the frame 15 (FWHM values of trace 1 and 2 are 38.8 ± 10.2 nm and 42.3 ± 10.2 nm, respectively). These values are less than the single pixel resolution in previous SICM measurements (100 -- 200 nm) [5, 39]. As demonstrated here, we can get SICM images with higher spatial resolution without scarifying the temporal resolution, as shown in Table 1 (compare the values of pixel rate and ra among the three studies with respective HS-SICM systems developed). In Figs. 7 and 8, slight discontinuities of images appear as noises between adjacent fast scan lines. In contrast, there are no such discontinuities in the images of stationary grating samples (Fig. 6a) captured with an even higher imaging rate than those used in Figs. 7 and 8. Moreover, during these imaging experiments, significant ion current reductions that could be caused by tip -- sample contact [39] were not observed. Therefore, we speculate that the image discontinuities appeared in the images of live HeLa cells are due to autonomous movement of the cells and the small pipette aperture; small movements of HeLa cells that cannot be detected with a large aperture pipette (ra ∼ 100 nm) appear in our high resolution images. Next, we performed HS-SICM imaging of actin filaments of ∼7 nm in diameter, under the conditions of ra = 5 -- 7.5 nm, Bid = 100 kHz, Vb = −0.2 V, 9 pixel rate of 556 -- 2000 Hz, and hopping amplitude of 100 -- 250 nm. Figure 9a shows a topographic image captured at ∼5 s/frame of an actin filament specimen placed on a glass substrate coated with positively charged aminopropyl-triethoxysilane. Figure 9b shows its enlarged image for the area shown with the red rectangle in Fig. 9a. The image exhibited a height variation along the filament, as indicated in Figs. 9c and d. The measured height obtained from the arrow position 3 is ∼7 nm. However, the measured heights obtained from the arrow positions 1, 2, 4, 5, and 6 were around 14 nm. These results may indicate that the specimen partly contains vertically stacked two actin filaments. The measured FWHM was 38.1 ± 4.2 nm for the arrow position 3. This value is 5 times larger than the diameter of an actin filament. This result can be explained by the side wall effect [60]; the tip wall thickness, i.e., ro − ri, would expand the diameter of a small object measured with SICM [59]. The side wall effect also explains that the measured FWHMs for the arrow positions 1 (66.0 ± 2.6 nm) and 2 (50.6 ± 3.7 nm) were larger than that for the arrow position 3. Despite the pixel size of 8 × 8 nm2, the 36 nm crossover repeat of the two-stranded actin helix could not be resolved. This is not due to insufficient vertical resolution but due to insufficient lateral resolution of the pipette used. Next, we used mica-supported neutral lipid bilayers containing biotin-lipid, instead of using the amino silane-coated glass substrate to avoid possible bundling of actin filaments on the positively charged surface. Figures 9e, f show images captured at 18 s/frame for partially biotinylated actin filaments immobilized on the lipid bilayers through streptavidin molecules with a low surface density. Measured heights of these filaments were ∼6 -- 7 nm (Figs. 9g and h). However, the measured height of the lipid bilayer was ∼13 nm from the mica surface (Fig. 9), much larger than the bilayer thickness of ∼5 nm [61]. This large measured thickness is possibly due to a sensitivity of Ii to negative charges on the mica surface [15, 19 -- 21]. The surface charges of objects can change dc even at constant Vb and constant set point [20], which can provide measured height largely deviating from real one. In the high resolution image of immobilized filaments (Fig. 9f), the measured value of FWHM was 33.3 ± 6.0 nm (Fig. 9h). As demonstrated here, our HS-SICM enables fast imaging of molecules without sacrificing the pixel resolution, unlike previous works [34, 35]. When the number of pixels was reduced to 50 × 50, we could achieve sub-second imaging for a low-height sample. Supplementary Movie 2 captured at 0.9 s/frame with 50 × 50 pixels over a 0.8 × 0.8 µm2 area shows high fluidity-driven morphological changes of polymers formed from a silane coupling agent placed on mica. TABLE 1. Comparison between three HS-SICM systems used for live cell imaging. The pixel rate is calculated by dividing the imaging rate by the number of pixels. ra indicates the aperture radii of the pipettes used to obtain SICM images. The attainable spatial resolution of SICM can be roughly estimated as 3 × ra [59]. PRDA is defined as pixel rate divided by ra. PRDA (s−1nm−1) Reference 10 Observation Endocytosis Exocytosis Peripheral edge Peripheral edge Microvilli Microvilli Microvilli Image rate (s/frame) 6 0.6 20 -- 28 18 1.4 20 Pixel rate (s−1) ra (nm) 68 1707 714 -- 1000 228 2926 455 50 1.4 Shevchuk et al. [35] 80 -- 100 5 -- 7.5 50 -- 100 80 -- 100 5 -- 7.5 17 -- 21 95 -- 200 2.3 -- 4.6 29.3 -- 36.6 60.7 -- 91.0 Simeonov and Schaffer [40] This work Ida et al. [39] Simeonov and Schaffer [40] This work FIG. 9. HS-SICM images of actin filaments. (a -- d) Actin filaments were attached onto a glass surface coated with positively charged amino-propyltriethoxysilane. (e -- h) Partially biotinylated actin filaments were attached onto a mica-supported lipid bilayer containing a biotin-lipid through streptavidin molecules with a low surface density. (a) HS-SICM image captured at ∼5 s/frame with 100 × 100 pixels, a pixel rate of 2000 Hz, hopping amplitude of 100 nm and Vb = −0.2 V. (b) A magnified view of the rectangle area shown in (a). (c) Height profiles along the arrows shown in (a). (d) Height profiles along the arrows shown in (b). (e, f ) HS-SICM images captured at ∼18 s/frame with 100 × 100 pixels, a pixel rate of 556 Hz, hopping amplitude of 250 nm and Vb = −0.2 V. (g) Height profile along the arrow shown in (e). (h) Height profiles along the arrows shown in (f ). A set point value of 1.5% was used to capture these HS-SICM images. G. Outlook for Higher Spatiotemporal Resolution Finally, we discuss further possible improvements of HS-SICM towards higher spatiotemporal resolution. The speed performance of SICM can be represented by the value of pixel rate divided by ra (we abbreviate this quantity as PRDA), because of a trade-off relationship between temporal resolution and spatial resolution. In Table 1, PRDA values of our HS-SICM imaging are shown, together with those of HS-SICM imaging in other labs. Although PRDA depends on sample height, our HS-SICM system set a highest record, PRDA = 95 -- 200 Hz/nm, in the imaging of a peripheral edge region of a HeLa cell. This record was attained by two means: the ICG method granting the high SNR of current detection and the high resonance frequency of the Z-scanner. Since we have not yet introduced other devices proposed previously for increasing the temporal resolution, there are still room for further speed enhancement. One of candidates to be added is (i) the 'turn step' procedure (applying a step function to the Z-piezodriver) developed by Simeonov and Schaffer for rapid pipette retraction [40]. Other candidates 400 nm~ 5 s/frame123a200 nm~ 5 s/frame465b100 nm5 nm123cd50 nm5 nm456500 nm~ 18 s/framee500 nm5 nmgh50 nm5 nm789~ 18 s/frame200 nmf78920 nm20 nm30 nm30 nmwould be (ii) further current noise reduction of the transimpedance amplifier in a high frequency region, and (iii) lock-in detection of AC current produced by modulation of the pipette Z-position with small amplitude [12]. Since our Z-scanner has a much higher resonance frequency than ever before, we will be able to use high-frequency modulation (∼100 kHz) to achieve faster lock-in detection of AC current. For higher spatial resolution, we need to explore methods to fabricate a pipette with smaller ra and ro, without significantly increasing Rp. One of possibilities would be the use of a short carbon nanotube inserted to the nanopore of a glass pipette with low Rp. III. CONCLUSIONS HS-SICM has been desired to be established not only to improve the time efficiency of imaging but also to make it possible to visualize dynamic biological processes occurring in very soft, fragile or suspended (not on a substrate) samples that cannot be imaged with HS-AFM. As demonstrated in this study, the fast imaging capability of SICM can be achieved by the improvement of speed performances of pipette Z-positioning and ion current detection. The former was attained by the new Z-scanner and implementation of vibration damping techniques to the Z-scanner. The latter was attained by the minimization of the total capacitance at the amplifier input and by the reduction of Rp achieved with the ICG method, resulting in an increased SNR of ion current detection. The resulting vz reached 0.55 µm/ms for ra = 5 -- 7.5 nm and 7.3 µm/ms for ra = 25 nm. The value of 7.3 µm/ms is larger than the recent fastest record achieved by Simeonov and Schaffer: 4.8 µm/ms for ra = 80 -- 100 nm. Consequently, the highest possible imaging rate was enhanced by ∼100-times, compared to conventional SICM systems. Even sub-second imaging is now possible for a scan area of 0.8 × 0.8 µm2 with 50 × 50 pixels, without compromise of spatial resolution. The achieved speed performance will contribute to the significant extension of SICM application in biological studies. IV. METHODS A. Fabrication of Nanopipettes We prepared pipettes by pulling laser-heated quartz glass capillaries, QF100-70-7.5 (outer diameter, 1.0 mm; inner diameter, 0.50 mm; with filament) and Q100-30-15 (outer diameter, 1.0 mm; inner diameter, 0.30 mm; without filament) from Sutter Instrument, using a laser puller (Sutter Instrument, P-2000). Just before pulling, we softly plasma-etched for 5 min at 20 W under oxygen gas flow (120 mTorr), using a plasma etcher (South Bay Technology, PE2000) to remove unwanted 11 contamination inside the pipette. The size and cone angle of each pipette tip were estimated from its scanning electron micrographs (Zeiss, SUPRA 40VP), trans- mission electron micrographs (JOEL, JEM-2000EX), and measured electrical resistance. Pipettes prepared from QF100-70-7.5 were used for the conductance measurement shown in Fig. 3c. B. Measurements of Z-scanner Transfer Function and Time Domain Response The Z-scanner displacement was measured with a laser vibrometer (Polytech, NLV-2500 or Iwatsu, ST-3761). The transfer function characterizing the Z-scanner re- sponse was obtained using an network analyzer (Agilent Technology, E5106B). A square-like-waveform voltage generated with a function generator (NF Corp., WF1948) and then amplified with a piezodriver (MESTECK, M-2141; gain, ×15; bandwidth, 1 MHz) was used for the measurement of time domain response of the Z-scanner. C. Measurement of Pipette Conductance with and without ICG To evaluate the enhancement of pipette conductance (1/Rp) by ICG, we prepared a pair of pipettes simultaneously produced from one pulled capillary, which exhibited a difference in ra less than ±10%, as confirmed by electrical conductance measurements under an identical condition. One of the pair of pipettes was applied to a conductance measurement under ICG, while the other to a conductance measurement without ICG. To obtain each plot in Fig. 3c, we measured (1/Rp) for more than 5 sets of pipettes. To avoid the non-linear current-potential problem arising from an ion current rectification effect, we used Vb ranging between −10 mV and 10 mV. D. Measurements of Approach Curves and Response of Ion Current To obtain the approach curves (Ii vs d) shown in Fig. 4a, a digital 6th-order low-pass filter was used with a cutoff frequency of 10 kHz. After the measurement of each curve under Vb = 0.1 V, the pipette was moved to a Z-position where the ion current reduction by 5% had been detected in the approach curve just obtained. Next, the cut-off frequency of the low-pass filter was increased to 400 kHz for the measurement of fast ion current response and Vb was set at a measurement value. Then, the experiment shown in Fig. 4b was performed. This series of measurements were repeated under different values of Vb and with/without ICG. The identical pipette was used through the experiments to remove variations that would arise from varied pipette shapes. After completing the experiments without ICG, the KCl solution inside the pipette was replaced with a 4 M KCl solution by being immersed the pipette in a 4 M KCl solution for a sufficiently long time (> 60 min). The Rp value of the pipetted with ICG prepared in this way was confirmed to be nearly identical to that of a similar pipette filled with a 4 M KCl solution from the beginning. performed again. Through all the HS-SICM experiments, the pipette resistance did not show a significant change, indicating no break of the pipette tip during scanning. The value of FWHM was calculated from five height profiles of each HS-SICM image. The error of FWHM was estimated from the standard deviation. E. HS-SICM Apparatus G. Sample Preparation 12 as 2a. the Z and XY directions shown in Fig. The HS-SICM apparatus used in this study was controlled with home-written software built with Lab- view 2015 (National Instruments), which was also used for data acquisition and analysis. The HS-SICM imaging head includes the XYZ-scanner composed of AE0505D08D-H0F and AE0505D08DF piezoactuators (both NEC/tokin), for respectively, The Z- and XY-piezoactuators were driven using M-2141 and M-26110-2-K piezodrivers (both MESTEK), respectively. The overall control of the imaging head was performed with a home-written FPGA-based system (NI-5782 and NI-5781 with NI PXI-7954R for the Z- and XY-position control, respectively; all National Instruments). For coarse Z-positioning, the imaging head was vertically moved with an MTS25-Z8 stepping-motor-based linear stage (travel range, 25 mm; THORLABS). For the FF control, FPGA-integrated circuits were used, while homemade analog circuities were used for FB and noise filtering. The sample was placed onto the home-built XY-coarse positioner with a travel range of 20 mm, which was placed onto an ECLIPSE Ti-U inverted optical microscope (Nikon). The ion current through the tip nanopore was detected via transimpedance amplifiers CA656F2 (bandwidth, 100 kHz; NF) and LCA-400K-10M (bandwidth, 400 kHz; FEMTO). A WF1948 function generator (NF) was used for the application of tip bias potential. F. HS-SICM Imaging in Hopping Mode The tip was approached to the surface with vz (0.05 -- 7 µm/ms) until the ion current reached a set point value. The set point was set at 1 -- 2% reduction of the "reference current" flowing when the tip was well far away from the surface (10 -- 100 pA when the ICG method was adopted). The voltage applied to the Z-scanner yielding the set point current was recorded as the sample height at the corresponding pixel position. Here, the output from the transimpedance amplifier was high-pass filtered at 5 -- 1000 Hz to suppress the effect of current drift on SICM imaging. Then, the pipette was retracted by a hopping distance (20 nm -- 50 µm within 20 -- 50 µs, depending on the hopping amplitude, during which the pipette was moved laterally towards the next pixel position. After full retraction, the tip approaching was (1) Glass substrate Cover slips (Matsunami Glass, C024321) cleaned with a piranha solution for 60 min at 70 ◦C were used as a glass substrate. (2) HeLa cells on glass HeLa cells were cultured in Dulbecco's Modified Eagle's Medium (Gibco) supplemented with 10% fetal bovine serum. The cells were deposited on a MAS-coated (Matsunami Glass, glass and maintained S9441) incubator at 37 ◦C until in a humidified 5% CO2 observation. Then, the culture medium was changed to phosphate buffer saline (Gibco, PBS). Then, HS-SICM measurements were performed at room temperature. (3) HeLa cells on plastic dish HeLa cells were seeded on plastic dishes (AS ONE, 1-8549-01) in Dulbecco's Modified Eagle's Medium (Gibco) supplemented with 10% fetal bovine serum. The cells were incubated at 37 ◦C with 5% CO2 and measured by HS-SICM 3 -- 4 days after seeding. Before the culture medium was HS-SICM measurements, changed to phosphate buffer saline (Gibco, PBS). HS-SICM measurements were performed at room temperature. (4) Actin filaments on glass substrate The glass surface was first coated with (3-aminopropyl) triethoxysilane (APTES; Sigma Aldrich). Then, a drop (12 µL) of actin filaments prepared according to the method [62] and diluted to 3 µM in Buffer A containing 25 mM KCl, 2 mM MgCl2, 1 mM EGTA, 20 mM imidazole-HCl (pH7.6) was deposited to the glass surface and incubated for 15 min. Unattached actin filaments were washed out with Buffer A. (5) Actin filaments on lipid bilayer The mica surface was coated with lipids containing 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) and 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine- N-(cap biotinyl) (biotin-cap-DPPE) in a weight ratio of 0.99:0.01, according to the method [63]. Partially biotinylated actin filaments in Buffer A prepared according to the method [64] were immobilized on the lipid bilayer surface through streptavidin with a low surface density. Unattached actin filaments were washed out with Buffer A. H. FEM Simulations We employed three-dimensional FEM simulations to study electrostatics and ionic mass transport processes in the pipette tip with ICG. In the simulation, we used the rotational symmetry along the pipette axis to reduce the simulation time. The full detail is described in SI 3. Briefly, the following set of equations were solved numerically: ∇2V = − F ε0ε Σ2 j=1Zjcj Jj = −Dj(c)∇cj − F ZjcjDj(c) ∇ · Jj = 0 n · ∇φ = RT −σ ε0ε . ∇V, (4) (5) (6) (4) describes The Poisson equation Eq. the electrostatic potential V and electric field with a spatial charge distribution in a continuous medium of permittivity ε containing the ions j of concentration cj and charge Zj. F and ε0 are the Faraday constant and the vacuum permittivity, respectively. We assume that the movement of the tip is sufficiently slow not to agitate the solution, and thus, the time-independent Nernst -- Planck equation, Eq. (5), holds, where Jj, Dj(c), R and T are the ion flux concentration of j, concentration-dependent diffusion constant of j, gas constant, and temperature in kelvin, respectively. This equation describes the diffusion and migration of the ions. The boundary condition for Eq. (5) is determined so that a zero flux or constant concentration condition is satisfied. On the other hand, the boundary conditions of 13 Eq. (4) are given so that a fixed potential or the spatial distribution of the surface charge σ, as described in Eq. (6), holds. In Eq. (6), n represents the surface normal vector. SUPPLEMENTARY MATERIALS The following data are available as Supplementary material: performance of XYZ-Scanner, current noise in our SICM system, finite-element simulation, dynamic response of measured ion current with and without use of ICG method, simulated approach curves obtained with and without the use of ICG method, current noise caused by capacitive couplings between Z-scanner and signal line of current detection, high-speed SICM imagings of grating samples and peripheral edge of HeLa cells, HS-SICM images of microvilli dynamics of HeLa cell (Movie 1), and HS-SICM images of polymers (Movie 2). ACKNOWLEDGMENTS This work was supported by a grant of JST SENTAN (JPMJSN16B4 to S.W.), Grant for Young Scientists from Hokuriku Bank (to S.W.), JSPS Grant-in-Aid for Young Scientists (B) (JP26790048 to S.W.), JSPS Grant-in-Aid for Young Scientists (A) (JP17H04818 to S.W.), JSPS Grant-in-Aid for Scientific Research on Innovative Areas (JP16H00799 to S.W.) and JSPS Grant-in-Aid for Challenging Exploratory Research (JP18K19018 to S.W.) and JSPS Grant-in-Aid for Scientific Research (S) (JP17H06121 and JP24227005 to T.A.). This work was also supported by a Kanazawa University CHOZEN project and World Premier International Research Center Initiative (WPI), MEXT , Japan. [1] P. Hansma, J. Cleveland, M. Radmacher, D. Walters, P. Hillner, M. Bezanilla, M. Fritz, D. Vie, H. Hansma, C. Prater, J. Massie, L. Fukunaga, J. Gurley, and V. Elings, Tapping mode atomic force microscopy in liquids, Applied Physics Letters 64, 1738 (1994). [2] S. Zhang, S.-J. Cho, K. Busuttil, C. Wang, F. Be- senbacher, and M. Dong, Scanning ion conductance microscopy studies of amyloid fibrils at nanoscale, Nanoscale 4, 3105 (2012). [3] T. Ushiki, M. Nakajima, M. Choi, S.-J. Cho, and F. Iwata, Scanning ion conductance microscopy for imaging biological samples in liquid: A comparative study with atomic force microscopy and scanning electron microscopy, Micron 43, 1390 (2012). [4] T. Ando, High-speed atomic force microscopy and its future prospects, Biophysical reviews 10, 285 (2018). [5] J. Seifert, J. Rheinlaender, P. Novak, Y. E. Korchev, and T. E. Schaffer, Comparison of atomic force microscopy and scanning ion conductance microscopy for live cell imaging, Langmuir 31, 6807 (2015). [6] P. Hansma, B. Drake, O. Marti, S. Gould, and C. Prater, The scanning ion-conductance microscope, Science 243, 641 (1989). [7] S. Del Linz, E. Willman, M. Caldwell, D. Klenerman, A. Fern´andez, and G. Moss, Contact-free scanning and imaging with the scanning ion conductance microscope, Analytical chemistry 86, 2353 (2014). [8] D. Thatenhorst, J. Rheinlaender, T. E. Schaffer, I. D. Dietzel, and P. Happel, Effect of sample slope on image formation in scanning ion conductance microscopy, Analytical chemistry 86, 9838 (2014). [9] L. Steinbock, J. Steinbock, and A. Radenovic, Control- lable shrinking and shaping of glass nanocapillaries under electron irradiation, Nano letters 13, 1717 (2013). [10] X. Xu, C. Li, Y. Zhou, and Y. Jin, Controllable shrinking of glass capillary nanopores down to sub-10 nm by wet-chemical silanization for signal-enhanced dna translocation, ACS sensors 2, 1452 (2017). [11] J. Y. Sze, S. Kumar, A. P. Ivanov, S.-H. Oh, and J. B. Edel, Fine tuning of nanopipettes using atomic layer deposition for single molecule sensing, Analyst 140, 4828 (2015). [12] D. Pastr´e, H. Iwamoto, J. Liu, G. Szabo, and Z. Shao, Characterization of ac mode scanning ion-conductance microscopy, Ultramicroscopy 90, 13 (2001). [13] P. Li, L. Liu, Y. Wang, Y. Yang, C. Zhang, and G. Li, Phase modulation mode of scanning ion conductance microscopy, Applied Physics Letters 105, 053113 (2014). [14] P. Li, L. Liu, Y. Yang, L. Zhou, D. Wang, Y. Wang, and G. Li, Amplitude modulation mode of scanning ion conductance microscopy, Journal of laboratory automation 20, 457 (2015). [15] K. McKelvey, S. L. Kinnear, D. Perry, D. Momotenko, and P. R. Unwin, Surface charge mapping with a nanopipette, Journal of the American Chemical Society 136, 13735 (2014). [16] K. McKelvey, D. Perry, J. C. Byers, A. W. Colburn, and P. R. Unwin, Bias modulated scanning ion conductance microscopy, Analytical chemistry 86, 3639 (2014). [17] A. Page, D. Perry, P. Young, D. Mitchell, B. G. Frenguelli, and P. R. Unwin, Fast nanoscale surface charge mapping with pulsed-potential scanning ion conductance microscopy, Analytical Chemistry 88, 10854 (2016). [18] D. Perry, R. Al Botros, D. Momotenko, S. L. Kinnear, and P. R. Unwin, Simultaneous nanoscale surface charge and topographical mapping, ACS nano 9, 7266 (2015). [19] D. Perry, B. Paulose Nadappuram, D. Momotenko, P. D. Voyias, A. Page, G. Tripathi, B. G. Frenguelli, and P. R. Unwin, Surface charge visualization at viable living cells, Journal of the American Chemical Society 138, 3152 (2016). [20] L. H. Klausen, T. Fuhs, and M. Dong, Mapping surface charge density of lipid bilayers by quantitative surface conductivity microscopy, Nature Communications 7, 12447 (2016). [21] T. Fuhs, L. H. Klausen, S. M. Sønderskov, X. Han, and M. Dong, Direct measurement of surface charge distribution in phase separating supported lipid bilayers, Nanoscale 10, 4538 (2018). [22] M. Kang, D. Perry, C. L. Bentley, G. West, A. Page, and P. R. Unwin, Simultaneous topography and reaction flux mapping at and around electrocatalytic nanoparticles, ACS nano 11, 9525 (2017). [23] Y. Takahashi, A. I. Shevchuk, P. Novak, B. Babakinejad, J. Macpherson, P. R. Unwin, H. Shiku, J. Gorelik, D. Klenerman, Y. E. Korchev, et al., Topographical and electrochemical nanoscale imaging of living cells using voltage-switching mode scanning electrochemical microscopy, Proceedings of the National Academy of Sciences 109, 11540 (2012). [24] A. Bruckbauer, L. Ying, A. M. Rothery, D. Zhou, A. I. Shevchuk, C. Abell, Y. E. Korchev, and D. Klenerman, Writing with dna and protein using a nanopipet for controlled delivery, Journal of the American Chemical Society 124, 8810 (2002). [25] B. Babakinejad, P. Jonsson, A. Lo`ppez Co`rdoba, P. Actis, P. Novak, Y. Takahashi, A. Shevchuk, U. Anand, P. Anand, A. Drews, A. Ferrer-Montiel, D. Klenerman, and Y. E. Korchev, Local delivery of molecules from a nanopipette for quantitative receptor mapping on live cells, Analytical chemistry 85, 9333 (2013). 14 [26] A. Page, M. Kang, A. Armitstead, D. Perry, and P. R. Unwin, Quantitative visualization of molecular delivery and uptake at living cells with self-referencing scanning ion conductance microscopy-scanning electrochemical microscopy, Analytical chemistry 89, 3021 (2017). [27] Y. Takahashi, A. I. Shevchuk, P. Novak, Y. Zhang, N. Ebejer, J. V. Macpherson, P. R. Unwin, A. J. Pollard, D. Roy, C. A. Clifford, H. Shiku, T. Matsue, D. Klenerman, and Y. E. Korchev, Multifunctional nanoprobes for nanoscale chemical imaging and localized chemical delivery at surfaces and interfaces, Angewandte Chemie International Edition 50, 9638 (2011). [28] A. Page, D. Perry, and P. R. Unwin, Multifunctional scanning ion conductance microscopy, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 473, 20160889 (2017). [29] T. Ando, T. Uchihashi, and T. Fukuma, High-speed atomic force microscopy for nano-visualization of dynamic biomolecular processes, Progress in Surface Science 83, 337 (2008). [30] T. Ando, T. Uchihashi, and S. Scheuring, Filming biomolecular processes by high-speed atomic force microscopy, Chemical reviews 114, 3120 (2014). [31] J. Rheinlaender and T. E. Schaffer, Image formation, resolution, and height measurement in scanning ion conductance microscopy, Journal of Applied Physics 105, 094905 (2009). [32] Y. Korchev, M. Milovanovic, C. Bashford, D. Bennett, E. Sviderskaya, I. Vodyanoy, and M. Lab, Specialized scanning ion-conductance microscope for imaging of living cells, Journal of microscopy 188, 17 (1997). [33] P. Novak, C. Li, A. I. Shevchuk, R. Stepanyan, M. Caldwell, S. Hughes, T. G. Smart, J. Gorelik, V. P. Ostanin, M. J. Lab, G. W. J. Moss, G. I. Frolenkov, D. Klenerman, and Y. E. Korchev, Nanoscale live-cell imaging using hopping probe ion conductance microscopy, Nature methods 6, 279 (2009). [34] P. Novak, A. Shevchuk, P. Ruenraroengsak, M. Miragoli, A. J. Thorley, D. Klenerman, M. J. Lab, T. D. Tetley, J. Gorelik, and Y. E. Korchev, Imaging single nanoparticle interactions with human lung cells using fast ion conductance microscopy, Nano letters 14, 1202 (2014). [35] A. I. Shevchuk, P. Novak, M. Taylor, I. A. Diakonov, A. Ziyadeh-Isleem, M. Bitoun, P. Guicheney, J. Gorelik, C. J. Merrifield, D. Klenerman, and Y. E. Korchev, An alternative mechanism of clathrin-coated pit closure revealed by ion conductance microscopy, The Journal of cell biology 197, 499 (2012). [36] G.-E. Jung, H. Noh, Y. K. Shin, S.-J. Kahng, K. Y. Baik, H.-B. Kim, N.-J. Cho, and S.-J. Cho, Closed-loop ars mode for scanning ion conductance microscopy with improved speed and stability for live cell imaging applications, Nanoscale 7, 10989 (2015). [37] J. Kim, S.-O. Kim, and N.-J. Cho, Alternative configuration scheme for signal amplification with scan- ning ion conductance microscopy, Review of Scientific Instruments 86, 023706 (2015). [38] P. Li, L. Liu, Y. Yang, Y. Wang, and G. Li, In-phase bias modulation mode of scanning ion conductance microscopy with capacitance compensation, IEEE Trans- actions on Industrial Electronics 62, 6508 (2015). [39] H. Ida, Y. Takahashi, A. Kumatani, H. Shiku, and T. Matsue, High speed scanning ion conductance mi- croscopy for quantitative analysis of nanoscale dynamics of microvilli, Analytical chemistry 89, 6015 (2017). [40] S. Simeonov and T. E. Schaffer, High-speed scanning ion conductance microscopy for sub-second topography imaging of live cells, Nanoscale 11, 8579 (2019). and for T. scanning [41] S. Watanabe High-speed xyz-nanopositioner conductance microscopy, Applied Physics Letters 111, 113106 (2017). [42] J. K. Rosenstein, M. Wanunu, C. A. Merchant, M. Drndic, and K. L. Shepard, Integrated nanopore sens- ing platform with sub-microsecond temporal resolution, Nature methods 9, 487 (2012). Ando, ion [43] R. A. Levis and J. L. Rae, The use of quartz patch pipettes for low noise single channel recording., Biophysical journal 65, 1666 (1993). [44] J. K. Rosenstein, S. Ramakrishnan, J. Roseman, and K. L. Shepard, Single ion channel recordings with cmos-anchored lipid membranes, Nano letters 13, 2682 (2013). [45] N. Kodera, H. Yamashita, and T. Ando, Active damping of the scanner for high-speed atomic force microscopy, Review of scientific instruments 76, 053708 (2005). [46] M. Kageshima, S. Togo, Y. J. Li, Y. Naitoh, and Y. Sugawara, Wideband and hysteresis-free regulation of piezoelectric actuator based on induced current for high-speed scanning probe microscopy, Review of scientific instruments 77, 103701 (2006). [47] R. W. Clarke, A. Zhukov, O. Richards, N. Johnson, V. Ostanin, and D. Klenerman, Pipette -- surface interac- tion: Current enhancement and intrinsic force, Journal of the American Chemical Society 135, 322 (2012). [48] M. A. Edwards, C. G. Williams, A. L. Whitworth, and P. R. Unwin, Scanning ion conductance microscopy: a model for experimentally realistic conditions and image interpretation, Analytical chemistry 81, 4482 (2009). [49] C. Wei, A. J. Bard, and S. W. Feldberg, Current rectification at quartz nanopipet electrodes, Analytical Chemistry 69, 4627 (1997). [50] L. Cao, W. Guo, Y. Wang, and L. Jiang, Concentration- gradient-dependent ion current rectification in charged conical nanopores, Langmuir 28, 2194 (2011). [51] X. L. Deng, T. Takami, J. W. Son, E. J. Kang, T. Kawai, and B. H. Park, Effect of concentration gradient on ionic current rectification in polyethyleneimine modified glass nano-pipettes, Scientific reports 4 (2014). [52] L.-H. Yeh, C. Hughes, Z. Zeng, and S. Qian, Tuning ion transport and selectivity by a salt gradient in a charged nanopore, Analytical chemistry 86, 2681 (2014). 15 [53] M. Z. Bazant, M. S. Kilic, B. D. Storey, and A. Ajdari, Towards an understanding of induced-charge electrokinetics at large applied voltages in concentrated solutions, Advances in colloid and interface science 152, 48 (2009). [54] D. Perry, D. Momotenko, R. A. Lazenby, M. Kang, and P. R. Unwin, Characterization of nanopipettes, Analytical chemistry 88, 5523 (2016). [55] R. A. Robinson and R. H. Stokes, Electrolyte solutions (Courier Corporation, 2002). [56] D. E. Goldman, Potential, impedance, and rectification in membranes, The Journal of general physiology 27, 37 (1943). [57] A. L. Hodgkin and B. Katz, The effect of sodium ions on the electrical activity of the giant axon of the squid, The Journal of physiology 108, 37 (1949). [58] J. Gorelik, A. I. Shevchuk, G. I. Frolenkov, I. A. Diakonov, C. J. Kros, G. P. Richardson, I. Vodyanoy, C. R. Edwards, D. Klenerman, and Y. E. Korchev, Dynamic assembly of surface structures in living cells, Proceedings of the National Academy of Sciences 100, 5819 (2003). [59] J. Rheinlaender and T. E. Schaffer, Lateral resolution and image formation in scanning ion conductance microscopy, Analytical chemistry 87, 7117 (2015). [60] L. Dorwling-Carter, M. Aramesh, C. Forr´o, R. F. Tiefenauer, I. Shorubalko, J. Voros, and T. Zambelli, Simultaneous scanning ion conductance and atomic force microscopy with a nanopore: Effect of the aperture edge on the ion current images, Journal of Applied Physics 124, 174902 (2018). [61] Z. Leonenko, E. Finot, H. Ma, T. Dahms, and D. Cramb, Investigation of temperature-induced phase transitions in dopc and dppc phospholipid bilayers using temperature- controlled scanning force microscopy, Biophysical journal 86, 3783 (2004). [62] T. Sakamoto, I. Amitani, E. Yokota, and T. Ando, Direct observation of processive movement by individual myosin v molecules, Biochemical and biophysical research communications 272, 586 (2000). [63] D. Yamamoto, T. Uchihashi, N. Kodera, H. Yamashita, S. Nishikori, T. Ogura, M. Shibata, and T. Ando, High-speed atomic force microscopy techniques for observing dynamic biomolecular processes, in Methods in enzymology, Vol. 475 (Elsevier, 2010) pp. 541 -- 564. [64] N. Kodera, D. Yamamoto, R. Ishikawa, and T. Ando, Video imaging of walking myosin v by high-speed atomic force microscopy, Nature 468, 72 (2010).
1707.06280	1	1707	2017-07-19T20:14:36	Toward Microphononic Circuits on Chip: An Evaluation of Components based on High-Contrast Evanescent Confinement of Acoustic Waves	[ "physics.app-ph" ]	We investigate the prospects for micron-scale acoustic wave components and circuits on chip in solid planar structures that do not require suspension. We leverage evanescent guiding of acoustic waves by high slowness contrast materials readily available in silicon complementary metal-oxide semiconductor (CMOS) processes. High slowness contrast provides strong confinement of GHz frequency acoustic fields in micron-scale structures. We address the fundamental implications of intrinsic material and radiation losses on operating frequency, bandwidth, device size and as a result practicality of multi-element microphononic circuits based on solid embedded waveguides. We show that a family of acoustic components based on evanescently guided acoustic waves, including waveguide bends, evanescent couplers, Y-splitters, and acoustic-wave microring resonators, can be realized in compact, micron-scale structures, and provide basic scaling and performance arguments for these components based on material properties and simulations. We further find that wave propagation losses are expected to permit high quality factor (Q), narrowband resonators and propagation lengths allowing delay lines and the coupling or cascading of multiple components to form functional circuits, of potential utility in guided acoustic signal processing on chip. We also address and simulate bends and radiation loss, providing insight into routing and resonators. Such circuits could be monolithically integrated with electronic and photonic circuits on a single chip with expanded capabilities.	physics.app-ph	physics	Toward Microphononic Circuits on Chip: An Evaluation of Components based on High-Contrast Evanescent Confinement of Acoustic Waves Yangyang Liu,1, a) Nathan Dostart,1 and Milos A. Popovi´c2 1)Department of Electrical, Computer, and Energy Engineering, University of Colorado, Boulder, CO 80309, USA 2)Department of Electrical and Computer Engineering, Boston University, Boston, MA 02215, USA (Dated: 16 May 2018) We investigate the prospects for micron-scale acoustic wave components and circuits on chip in solid planar structures that do not require suspension. We leverage evanescent guiding of acoustic waves by high slowness contrast materials readily available in silicon complementary metal-oxide semiconductor (CMOS) processes. High slowness contrast provides strong confinement of GHz frequency acoustic fields in micron-scale structures. We address the fundamental implications of intrinsic material and radiation losses on operating frequency, bandwidth, device size and as a result practicality of multi-element microphononic circuits based on solid embedded waveguides. We show that a family of acoustic components based on evanescently guided acoustic waves, including waveguide bends, evanescent couplers, Y-splitters, and acoustic-wave microring resonators, can be realized in compact, micron-scale structures, and provide basic scaling and performance arguments for these components based on material properties and simulations. We further find that wave propagation losses are expected to permit high quality factor (Q), narrowband resonators and propagation lengths allowing delay lines and the coupling or cascading of multiple components to form functional circuits, of potential utility in guided acoustic signal processing on chip. We also address and simulate bends and radiation loss, providing insight into routing and resonators. Such circuits could be monolithically integrated with electronic and photonic circuits on a single chip with expanded capabilities. I. INTRODUCTION With the tremendous advances in modern lithography and high-resolution nanofabrication that were driven by the electronic integrated circuit (IC) industry, micron- scale photonic circuits, including silicon photonic cir- cuits, have emerged over the past five decades, as a new chip technology showing substantial promise to enable many applications and provide performance superior to electronics (particularly for communication), as well as to enhance CMOS technology itself1,2. High optical re- fractive index contrast between core and cladding ma- terials in planar photonic waveguide geometries has al- lowed small, wavelength-scale microphotonic and opto- electronic components to be integrated in complex cir- cuits and systems on a chip, providing capabilities for a wide range of compact high performance applications in communication, sensing and information processing3. Since acoustic and optical (electromagnetic) waves share many mathematical similarities4, an analogous high slow- ness contrast acoustic wave confinement scheme could bring the same possibilities to acoustics to enable mi- crophononic circuits, where the manipulation of signals in the acoustic domain can maximally benefit from the advancing integration technology and wavelength scale confinement. A key benefit is that evanescent confine- ment provides for efficient coupling between circuit ele- ments such as waveguides and resonators. The viability a)Electronic mail: [email protected] travel at much slower of such a scheme will be determined by the achievable device size scale and propagation losses – which will de- termine the complexity (how far waves can propagate, i.e. through how many components), and the bandwidth (i.e. how long acoustic waves can spend in the circuit without dissipating). Coherent phonons speeds than photons and can directly interact with radiofre- quency or optical signals5, making chip-scale acous- tics or microphononics an interesting domain for re- searchers in phononics, optomechanics and micro-/nano- electro-mechanical systems (MEMS/NEMS). A wide range of systems have recently been explored, where acoustic waves interact with electronics and photon- ics in chip-scale integrated platforms to enable opti- cal delay lines5,6, narrow-linewidth RF photonic filters7, photonic-phononic memory8 , frequency locking of mi- cromechanical oscillators9, systems in the quantum me- chanical ground state10, and on-chip optomechanical sig- nal detection11. The ability to confine and guide acoustic waves not only enables the construction of individual de- vices, but may also provide a means to connect them to form a complete on-chip acoustic circuitry. Early investigation into acoustic waveguides proposed various geometries including flat overlay waveguides with thin metal strips deposited on a substrate, topologi- cal waveguides that confine acoustic fields to a ridge or wedge in a homogeneous material, and in-diffused waveg- uides that are locally implanted with metals to create a weak impedance contrast12. Later developments in surface acoustic wave (SAW) and bulk acoustic wave (BAW) technologies make use of acoustic fields verti- 7 1 0 2 l u J 9 1 ] h p - p p a . s c i s y h p [ 1 v 0 8 2 6 0 . 7 0 7 1 : v i X r a cally bound to a free material surface or between air and an acoustic reflector, and have enabled commercial applications in radar and communication systems, ra- dio/intermediate frequency signal processing, and chem- ical/biological sensing13,14, despite lack of lateral con- finement that allows sharp bends. Acoustic confinement in off-chip cylindrical waveguides has also been inves- tigated, e.g. in the context of Brillouin scattering in low-contrast silica optical fibers15 and analytically in GaAs-AlAs quantum wires16. Modern fabrication tech- nologies have enabled the construction of planar struc- tures with an inhomogeneous distribution of materials in the cross-section. Chip-scale acoustic devices so far still heavily rely upon 1) air suspension, where discon- tinuation of solid material reflects acoustic waves at free boundaries17, and 2) phononic crystal structures, where engineered slowness mismatch in a periodic composite material provides wave confinement18, or a combination of these two mechanisms19. Recently, resurgent interest in optomechanics and stimulated brillouin scattering in on-chip devices has led to investigations of novel acoustic waveguiding geometries for strong light-sound coupling, including geometrical20 and material slowness contrast based (evanescent)21,22 confinement. In this paper, we investigate the prospects of evanes- cent guiding for ultracompact acoustic components on chip in silicon-based CMOS materials. We look at waveg- uides and resonators, and investigate the limitations ex- pected to be imposed by intrinsic material loss on device lengths and resonator Q's, and show in simulation designs of several evanescent-confinement based components, in- cluding a directional coupler, Y-splitter and acoustic mi- croring resonator. Evanescent confinement means that the structures need not be air-suspended, making them suitable beyond MEMS-like processes and potentially in- tegrable with electronics and photonics in mainstream advanced CMOS2. These geometries could also be read- ily employed to add lateral confinement to conventional SAW devices, in MEMS circuits to provide evanescent coupling as an alternative to coupling springs, and in op- tomechanics to permit simultaneous and controlled guid- ing of both acoustic and optical waves. In the remainder of the paper, we first study the guid- ance properties and confinement strength of evanscently- confined waveguides in a high slowness contrast scenario. Second, we evaluate the effect of material losses in typical silicon CMOS materials, as well as the effect of radiation loss in curved waveguides where the phase velocity ex- ceeds the local speed of sound. We provide some scaled plots that provide an idea of the parameter space where useful waveguides and resonators can be designed. Last, we simulate a few components including an evanescent directional power coupler, an acoustic Y-splitter, and a microring resonator, showing 10µm scale components in all cases. Our study suggests that evanescent con- finement can support complex multi-element phononic guided-wave and resonator circuits on chip, at frequen- cies from below 1 GHz to 10's of GHz where they may 2 have the potential to offer better performance than pho- tonic circuits. II. ACOUSTIC WAVE EQUATION AND SLOWNESS CURVES Guiding of acoustic waves has marked similarities to that of optical waves. At a key time in the development for acoustic devices in the 1970's, Auld reformulated and mapped the acoustic equations and conservation laws by following the example of Maxwell's equations4. In re- fraction of both electromagnetic and acoustic waves at a material interface, the tangential component of the wave vector is conserved, as a result of continuity of tangen- tial electric field and continuity of particle velocity, re- spectively. Evanescent guiding between two materials requires existence of a critical angle, beyond which prop- agating waves are confined in the material with slower wave speed. In optics, refractive index is used to de- scribe the ratio between speed of light in vacuum and in the material. For acoustic waves, since there is not a common reference, slowness, defined to be the inverse of wave speed in the material, is used to the same effect as optical index of refraction. In this section, we derive the slowness curves for the propagation of plane waves in relevant core and cladding materials and find the conditions for evanescently- confined guided waves. At the interface between two materials with different stiffnesses and densities, acoustic waves propagating in one material can, under certain con- ditions, refract into only evanescent waves in the other material. This acoustic version of total internal reflec- tion can confine acoustic fields in a slow (analogous to a high refractive index for light) core material surrounded by a fast cladding. In addition to waveguides, traveling and standing wave resonators and other functional com- ponents can be designed using the same configuration. Propagation of acoustic waves in solids is governed by the acoustic wave equation ∇ · c : ∇sv = ρ ∂2 t v − ∂tF (1) where v is the velocity field, ∇s is the symmetric gra- dient operator, c and ρ are the material stiffness tensor and mass density, and F is external driving force den- sity. In a source free case (F = 0), eigenstate waves are supported with displacement field proportional to u ∼ exp[i(ωt− kl· r)] and propagate along l = xlx + yly +zlz. Equation 1 can be rearranged based on an assumed plane- wave solution to derive the Christoffel equation4 (cid:2)k2Γij − ρω2δij (cid:3) [vj] = 0 (2) where Γij = liKcKLlLj is the Christoffel matrix, contain- ing elements of the ∇ · c : ∇s operator applied to the assumed plane wave solution. Setting the determinant of Eq. 2 to zero, Ω(ω, kx, ky, kz) =(cid:12)(cid:12)k2Γij(lx, ly, lz) − ρω2δij (cid:12)(cid:12) = 0 (3) a dispersion relation in the form of slowness surfaces 1/vp ≡ k/ω can be fully solved for plane wave propa- gation in a homogeneous material. Figure 1(a) shows the slowness curves for plane waves propagating along crystal faces in three materials commonly used in a sili- con platform, Si, SiO2 and SiN. Silicon, like other crys- talline materials in general, is acoustically anisotropic and tri-refringent, yielding three distinct slowness curves as solutions to Eq. 3. SiO2 and SiN are amorphous and isotropic. In this case, the two shear polarizations be- come degenerate. III. GUIDING ACOUSTIC WAVES IN SOLIDS In an inhomogenous material geometry, scattering at a material boundary requires that the particle displace- ment and normal stress fields be continuous. For ex- ample, at a planar interface at z = 0 (normal position r⊥ ≡ zz = 0), the tangential wave vector k(cid:107) ≡ xkx + yky must be equal in both materials to give the same spatial field dependence exp(ik(cid:107) · r(cid:107)) that avoids discontinuities. In the slowness picture, when a wave is incident on the interface from one side, this means that the tangential component of 1/vp = k/ω must be conserved across the interface as a corresponding wave is excited in transmis- sion on the other side of the interface. If the interface un- dulation is slower than the wave in the second, cladding material, then the "transmitted" wave is evanescent with an imaginary k⊥, as in the familiar case of optical waveg- uides. vr However, since materials are in general trirefringent for acoustic propagation, with three slowness curves, for one longitudinal and two shear/transverse polarizations (amorphous materials are isotropic with degenerate shear modes), acoustic confinement is more complex to achieve than optical confinement. Depending on the relation be- tween wave speeds in materials, up to five critical angles can exist, θcr = sin−1( vi ), where vi and vr are the ve- locities of the incident and refracted waves and vi < vr. These critical angles are associated with reflected waves of other polarizations in the incident material, and re- fracted waves in the other material. Beyond all possible critical angles the incident wave undergoes total inter- nal reflection. For a cladded structure to be evanes- cently confining, the slowness of its waveguide modes must fully enclose the slowness curves of all polarizations in the cladding, to prevent scattering into propagating bulk modes. In a rectangular waveguide with Si cladding and SiO2 core, sidewalls couple shear and longitudinal plane waves into hybrid modes, among which the ones with primarily shear components have slower speeds than the fastest polarization in the cladding and can thus be fully confined in the SiO2 core. Figure 1(b) illustrates four possible configurations for high-contrast dielectric acoustic waveguides utilizing evanescent confinement be- tween two materials and free air interfaces that can po- tentially suit different applications and fabrication plat- 3 FIG. 1. (a) Slowness curves of acoustic plane waves propa- gation along crystal faces in bulk Si, SiO2 and SiN. (b) Pos- sible geometries for high-contrast dielectric acoustic waveg- uides: [i] fully-cladded [ii] half-cladded [iii] half-cladded, em- bedded [iv] side-cladded, membrane. (c) Total displacement u (blue/red represents min/max displacement), and horizon- tal ux, vertical uz and longitudinal uy components (blue/red represents negative/positive displacement) of the first three horizontally (H) and vertically (V) polarized eigenmodes in a fully cladded rectangular waveguide with a 1µm×2µm SiO2 core surrounded by Si cladding. (d) Slowness (1/vp) v.s. waveguide core width for fixed waveguide height of 0.5µm and wavelength λ = 1.5µm. SiO2purelongitudinalSiO2pureshearSipureshearSiquasishearSiquasilongitudinalSi3N4 purelongitudinalSi3N4 pureshear(a)12345678x10-41.61.822.22.42.62.8Waveguide width (um)Slowness (s/m)(d)H1H2H3H4V1V2V31/vS,Si1/vS,SiO21/vL,SiO2H1H2H3H4V1V2V3zxySiSiO2(b)[i][iv][iii][ii][100][001]x10-4kx/ωx10-4ky/ω123123(c)xzyuforms. The fully-cladded waveguide (Fig. 1(b)[i]) consists of a slow core surrounded by a fast cladding, drawing a direct analogy to cladded rectangular optical waveg- uides. Figure 1(b)[ii] and [iii] are half-cladded waveg- uides with the top surface of the core exposed to air. The embedded version (Fig. 1(b)[iii]) has a planar top surface that could further interface to transducers, such as patterned metal electrodes. Side-cladded waveguides on an air-suspended membrane (Fig. 1(b)[iv]) provide an alternative to phononic crystal waveguides for systems without a compatible layer stack to evanescently confine acoustic waves in the vertical direction. As a simple example to illustrate the behavior of evanescent mode acoustic waveguides, Fig. 1(c) shows the mode field distributions of a fully-cladded rectan- gular waveguide with SiO2 core and Si cladding simu- lated using COMSOL Multiphysics23. A 3D section of the waveguide was simulated using the Solid Mechanics module, with Perfectly Matched Layer (PML) bound- ary conditions applied on the lateral faces and Floquet boundary condition on the longitudinal faces for a given propagation constant. Guided modes split into two cat- egories, horizontal (H) modes with a strong field compo- nent in the transverse plane along the horizontal edges of the core cross-section and vertical (V) modes along the vertical edges. Figure 1(d) shows the slowness curves of the first few horizontal and vertical modes versus waveg- uide width for a fixed height of 0.5µm. All guided modes are below the shear wave slowness in the SiO2 core, but above that in the cladding, at which point waves in the core start refracting into the cladding, losing evanescent confinement at the interface. The width of an acoustic waveguide can be designed through numerical simulation to support only the lowest order horizontal and vertical modes, which approach degeneracy at small waveguide widths, and thus there is no apparent cutoff for the fun- damental vertically polarized (V1) mode. Combining evanescent confinement and free air inter- faces, the embedded half-cladded geometry, Fig. 1(b)[iii], can be a more practical configuration in terms of fabrica- tion, and it provides better access to the mode fields that are tightly concentrated on the exposed top surface of the core and leverage surface wave components. Figure 2(a) shows the field components of the first three vertically (SV) and horizontally polarized (SH) surface modes. The SV modes have a dominant vertical shear component (uz), which spreads across the core with field maxima on the top surface; the shear horizontal (ux) and lon- gitudinal (uy) components are evanescently confined to the top surface, resembling Love waves in a half-cladded slab4. The SH waves have a dominant shear horizontal (ux) component. The half-cladded acoustic waveguides could be fabricated directly into a silicon substrate us- ing localized thermal oxidation defined by a hardmask. Alternative fabrication methods include chemical vapor deposition of SiO2 on a pre-patterned silicon trench fol- lowed by chemical mechanical planarization (CMP), or a different fast material such as SiN can be used as a sub- 4 (a): Total displacement u (blue/red represents FIG. 2. min/max displacement), and horizontal ux, vertical uz and longitudinal uy components of the first three vertically (SV) and horizontally (SH) polarized surface modes (blue/red represents negative/positive displacement) of a half-cladded waveguide with a 0.5µm×1µm Si core embedded in SiO2 cladding. (2) Slowness (1/vp) v.s. waveguide core width for fixed waveguide height of 0.5 µm and wavelength λ= 0.75 µm. strate underneath the Si/SiO2 layer to provide a bottom cladding, and the oxide waveguide can be patterned from a device layer using regular scanning electron beam or optical lithography. This waveguide geometry has tight lateral confinement that allows for sharp bends with low radiation loss, enabling ultra-compact components, and high concentration of fields on the top surface, making it a versatile candidate to allow interfacing acoustic waves to other systems with different signal carrying physics, such as electronic circuits or guided light waves. Our ultimate goal in considering high slowness con- trast evanescent confinement is to construct chip-scale phononic components and coupled-element circuits that can enable a richer signal processing capability. High slowness contrast enables compact components on the order of 10 µm, which will be discussed in more detail in Sec. V. However, the practicability of a phononic (a)SV1SV2SH1Waveguide width (um)00.511.52Frequency (Hz)x1094.64.855.25.45.65.866.26.46.6SiO2SizxySV1SV2SV3SV4SH1SH2SH3(b)SV1SV2SV3SH1SH2uuuxzyucircuitry such as that described here requires that sig- nal propagation lengths in waveguides are large enough to traverse a few components, and that excitation life- times in resonators are long enough to process (e.g. fil- ter or delay) relevant bandwidth signals and couple the energy faster to the next circuit element than to the radiation loss mechanisms. Therefore, in the follow- ing, we first consider acoustic loss mechanisms and the bounds they place on performance of waveguides and res- onators in order to evaluate the viability of high-contrast microphononic circuits. We consider intrinsic material losses in the next section, and then radiation loss – a generalized anchor loss mechanism – in the following sec- tion on device design. IV. IMPACT OF MATERIAL INTRINSIC LOSSES A key consideration in understanding the utility, and range of applicability, of these acoustic waveguides is loss. In this paper we consider two mechanisms: ma- terial intrinsic loss, dealt with in this section, and acous- tic radiation loss, a fundamental loss mechanism which occurs due to bending of otherwise lossless waveguides, discussed in the next section because it is associated with device design. For conventional on-chip acoustic devices, energy dissipation is mainly caused by air damping, an- chor loss and intrinsic material losses24,25. Air damping includes a few different mechanisms and is often domi- nated by squeeze-film damping for suspended structures with small air gaps between a vibrating film and a sta- tionary substrate25,26. Anchor loss is caused by acoustic radiation into the substrate through attachments such as pedestals and spokes27,28 that provide mechanical sup- port for suspension. Having acoustic waves fully confined in solids evanescently (without air suspension) exempts this type of devices from squeeze-film damping and from conventional anchor loss (radiation into the cladding, in straight sections), but intrinsic material losses still im- pose a limit on the frequency range where low loss waveg- uides and resonant cavities with high quality factors can be achieved. Hence they are addressed first in the context of their impact on wavelength scale devices and circuits. The two main intrinsic loss mechanisms for acoustic devices are thermoelastic dissipation (TED) and phonon- phonon interaction associated dissipation (PPD), which both result from coupling between the acoustic field and thermal phonons in solids, at different time and length scales29–33. The total material intrinsic loss limited qual- ity factor Qintrinsic is given by Q−1 PPD, Using material property values listed in Table I, Fig. 3(a) plots the upper bounds, due to each of TED and PPD, to the quality factors (Q) achievable in Si, SiO2 and SiN, at room temperature (T = 300K). The upper bound Q is a loss Q of a resonator implemented in the respective material, and a specific acoustic resonator design based on evanescent confinement using multiple materials will see a average of these losses with the mode intrinsic = Q−1 TED + Q−1 5 where QTED and QPPD are Q limits due to TED and PPD respectively. TED describes the energy dissipation associated with coupling between a strain field and a temperature gra- dient through irreversible heat flow (absent in volume preserving pure shear waves). TED is dependent on the specific geometry of an acoustic device, and can be mini- mized in design by reducing the overlap of the strain field induced by the acoustic wave and the heat diffusion eigen- modes of the system32, but the bulk limit still provides a general measure of feasibility of materials for acoustic wave guiding. Landau and Lifshitz calculated the atten- uation coefficient for longitudinal waves in amorphous (isotropic) solids, which also gives an order of magnitude estimate for anisotropic crystals34. An equivalent expres- sion for QTED at angular frequency ω and temperature T is29,32,33 QTED = 9C 2 v ωκT β2ρ (4) where Cv, κ, β and ρ are the volumetric heat capacity, thermal conductivity, thermal expansion coefficient and mass density. PPD describes interactions between incident acoustic waves and thermally excited phonons that happen at shorter time and length scales than TED. Akhieser con- sidered this loss mechanism in the low frequency regime ωτ (cid:28) 1 (Akhieser regime), where τ is the lifetime of the thermal phonons, and used the Boltzmann equa- tion to derive the attenuation by calculating the increase in entropy due to collisions between thermal phonons35. Woodruff and Ehrenreich and others further developed this theory for ωτ > 1 in the regime ω (cid:28) kBT /36. A simplified expression for QPPD in the extended frequency range ω (cid:28) kBT / is31,37 (cid:0)(cid:104)γ2(cid:105) − (cid:104)γ(cid:105)2(cid:1)−1 (5) QPPD = 1 + ω2τ 2 ωτ ρv2 g CvT 3 τ CvV 2 D and 3V −3 where vg is the group velocity of the acoustic wave, re- lated to the mode (or phase) velocity vp by 1/vg ≡ ∂ ∂ω (ω/vp(ω)), and γ is the phonon mode Gruneisen pa- rameter. Here, neglecting dispersion, vg is taken to be approximately equal to the phase velocity of the acous- tic wave vp = ω/k, and τ is estimated using the kinetic relation κ = 1 , where VD, Vl and Vt are the Debye, longitudinal and transverse wave velocities29,31. (cid:104)(cid:105) indicates averaging over interact- ing thermal phonon modes37–39. field as weighting function, in addition to possible inter- face losses, but these values provide bounds. Since there is very limited data in the literature on (cid:104)γ(cid:105) and (cid:104)γ2(cid:105) for materials other than Si, a slightly different form of the QPPD in the low frequency Akhieser regime (ωτ (cid:28) 1) for longitudinal waves was plotted in addition to Eq. 5, D = V −3 l + 2V −3 t Parameter density (kg/m3) Symbol ρ thermal heat capacity conductivity expansion thermal (W/m·K) wave speed long. (×10−6K−1) (m/s) (m/s) Vt Vl β specific (J/kg·K) Cs = Cv/ρ Debye Gruneisen parameters trans. temperature long. trans. (K) θ 640 29043 29046 γL γT 0.65 0.75 −2.144 −1.7544 0.447† 0.4† long. trans. (cid:104)γ2(cid:105) − (cid:104)γ(cid:105)2 (cid:104)γ2(cid:105)†† 0.0339 0.3839 0.4639 0.8839 – – 2.640 0.542 0.842 7470 5860 574841 333341 878845 505345 Si30 SiO2 SiN 2330 230041 250045 713 100041 17045 κ 14540 1.141 1942 6 phonon lifetime (×10−12s) τ 6.7‡ 0.13‡ 4.2‡ TABLE I. Material properties used for calculating Q limits due to thermoelastic and phonon-phonon dissipations at T = 300K. † Bulk value for β-Si3N4 used as estimate. ‡ Calculated using τ = 3κ ρCsV 2 D 29,31. ††(cid:104)γ(cid:105) vanishes for transverse waves48,49. tive Gruneisen parameter γeff = (γL + 2γT )/3, where γL and γT are the longitudinal and transverse Gruneisen parameters38,44. QTED drops in inverse proportion to frequency, as does QPPD in the Akhieser regime. As ωτ exceeds 1, QPPD turns around and improves as frequency increases. For all three materials (and as a more general rule of thumb), PPD is the more limiting loss mechanism. A key interest in phononics is as a competing or complementary and potentially synergistic technology to photonics in signal processing, where relevant context is provided by assessing signal processing bandwidths, or potential interaction with optical modes via radiation pressure, photoelasticity or electrostriction. Although the Q limits in Fig. 3(a) for acoustic resonators may appear low compared to optical devices (Q ≡ ωoτr/2, with τr the lifetime), the critical metric is the reso- nance lifetime τr and associated linewidth 2/τr, and the much lower center frequency of acoustic waves makes it possible to design resonant devices with much narrower linewidths than typical optical devices on chip that are compatible with silicon processing and usually support no narrower than several GHz of bandwidth. The dashed lines in Fig. 3(b) are calculated resonance linewidths that correspond to the material loss limited Q in Fig. 3(a). Pure shear wave wavelength in bulk Si is plotted ver- sus frequency in solid line to provide a characteristic length scale [shear wave slowness in Si cladding cuts off guided modes in the all cladded waveguide geome- tries, c.f. Fig. 1(d)]. For modern fabrication processes, a relevant range of device dimensions is between 100nm and 10µm, corresponding to operating frequencies in the 1− 100 GHz range and material intrinsic linewidths from 1 kHz to 10 MHz in SiO2. In terms of spatial propagation, these linewidths correspond to a loss Q of about 3,000 to 300,000, which can be converted to an equivalent waveg- uide propagation loss attenuation constant in dB/mm by the relation αdB/mm = 0.01 . The approximate at- ln(10) tenuations corresponding to the gray region in Fig. 3(b) are 0.004 to 90 dB/mm (assuming vg ≈ 3, 000 m/s, i.e. SiO2 core). With compact circuits down 10−100 µm this is still sufficient for useful functions even at the higher end of the frequency range. These linewidths and length- scales suggests a potentially viable technology for dis- criminating signals of such or larger bandwidths and/or producing delays (and associated propagation distances) about the inverse of these bandwidths with reasonable 2Qvg ω FIG. 3. (a) Q limits due to thermoelastic (solid lines) and phonon-phonon dissipations (dashed lines) for SiO2, Si and SiN. (b) Intrinsic linewidths corresponding to the Q lim- its (dashed/dotted lines) and characteristic wavelength (pure shear wave in bulk Si, solid line) scaling versus frequency. following derivation from Duwel et al.30 (cid:34) (cid:18) Vl (cid:19)3(cid:90) θ/T Vt 0 x4exdx (ex − 1)2 (cid:35)−1 γ−2 eff QAKE,L = 3 ρV 5 l k4 BT 4 3ωτ 2π2 (6) where θ is the Debye temperature. Eqn. 6 is evalu- ated in addition to Eqn. 5 for comparison, using effec- Frequency (Hz)1061081010Intrinsic Q10010210410610810101011109107Frequency (Hz)Linewidth (Hz)10-410-2100102104106108-910-810-710-610-510-410-310(b)10610810101011109107Si, ƊνPPD,LSiO2, ƊνPPD,TSiO2, ƊνPPD,LSi, ƊνPPD,TSiN, ƊνAKE,LSi, QTEDSi, QPPD,LSi, QAKE,LSiO2, QPPD,TSiO2, QAKE,LSiO2, QTEDSiO2, QPPD,LSi, QPPD,TSiN, QAKE,LSiN, QTED(a)loss. 7 jkl u∗ (cid:82) d2r(cid:80) Alternatively to Eqns. 5 and 6 and similar expressions where acoustic attenuation due to PPD is evaluated us- ing Gruneisen parameters, the effective phonon viscosity method can be used to describe material loss from the same mechanism37,50–52. The phonon viscosity tensor is a representation that relates stress T to strain S as T = c : S + η : (∂tS). It is analogous to the imaginary part of complex refractive index (i.e. permittivity ten- sor) for optical waves, which describes material absorp- tion. The effective attenuation constant α of a particular mode of an acoustic waveguide can be calculated by an overlap integral of the displacement field u with the mate- rial viscosity tensor η, α = ω2 i ∂jηijkl∂kul, PB where PB is the power of the acoustic mode52. While an analogue to optical refractive index provides an intuitive formalism and straightforward accounting of mode field distribution in overall loss via overlap integrals, there are discrepancies between the experimental values of the vis- cosity tensor of Si as measured in the few available pre- viously published experiments37. Further, the viscosity tensors of SiO2 and SiN do not appear to be available in existing literature52. For these reasons the Akhiezer model with Gruneisen parameters is used in this pa- per to provide a rough estimate of limitations on the acoustic loss Q due to phonon-phonon interactions, which is sufficient for evaluating the promise of wavelength- scale phononic components and multi-element "circuits". A more accurate characterization of the bulk material losses, as well as of material interface losses which we do not address here, could become important in the detailed design of devices but are beyond scope for our discussion. V. RADIATION LOSS AND DESIGN OF MICROPHONONIC CIRCUIT COMPONENTS In this section, we discuss the design of components. The intrinsic material losses set the upper limits for prop- agation length and time delay, and complex circuits are only possible if useful functions can be accomplished in smaller length and time scales, thereby allowing several components to be traversed before the signal is lost. We show here that high slowness contrast allows compact enough components and strong enough confinement to enable practical microphononic circuits, including cou- pling and routing of signals between elements. A second key loss mechanism which is critical to determining the compactness of components (and hence viability of the circuits) is radiation loss, a generalized form of anchor loss that we address first. The evanescent confinement between fast and slow ma- terials enables a family of guided, traveling acoustic wave components that can be designed using techniques sim- ilar to those developed for photonic components. Fig. 4 shows simulations of example components we designed: an evanescent directional power coupler, a 1×2 3 dB wave power splitter, and an acoustic microring resonator. FIG. 4. (a) Radiation loss Q and (b) free spectral range versus ring center radius of half-cladded rings with 0.5µm×0.5µm SiO2 core and fully-cladded rings with 1µm×2µm SiO2 core from finite element simulations. (c) Simulated field profiles of acoustic directional coupler, Y-splitter and coupled ring resonator using an embedded waveguide cross-section with 0.5µm×0.5µm SiO2 core half-cladded with Si. Ring center radius (um)051015202530Radiation Q101102103104105106107108(a)Ring center radius (um)051015202530FSR (MHz)101102103half clad 0.5um x 0.5umfull clad 2um x 1um@ 5.3GHz@ 3.0GHz(b)0.5-0.550-550-5-4045015100.5-0.50.5-0.550-5-202(c)half clad 0.5um x 0.5umfull clad 2um x 1um@ 5.3GHz@ 3.0GHzThey are based on a 0.5×0.5 µm silica core embedded into a silicon substrate, similar to Fig. 2. We simulated guided acoustic wave propagation through these devices, and find that compact components on the order of 10 µm in dimension can achieve full power transfer, power split- ting, or provide a functioning high-Q resonator. We next discuss design in more detail. Radiation loss occurs when a confined mode has an accessible radiation channel, and loses energy (referred to as a leaky mode). Straight embedded acoustic waveg- uides can be designed to be fully confining, with no ra- diation loss. Curved acoustic waveguides formed in a solid, however, have a fundamental radiation loss mech- anism, analogous to that in optical waveguides. Because there is an evanescent tail of the acoustic wave extend- ing into the cladding material (orthogonal to the mode propagation direction), and the phase fronts circulate az- imuthally around a circular waveguide bend, there is a radius at which the guided wave phase fronts exceed the local speed of sound. This radius defines the acoustic radiation caustic, and results in radiation. The radia- tion loss increases exponentially with smaller radius of curvature of the waveguide. This loss mechanism lim- its how small a radius can be used to form a waveguide bend to route an acoustic signal between components, or an acoustic microring resonator. Fig. 4(a) plots the res- onator Q due to bending-induced radiation loss for two designs of acoustic ring resonator, as a function of the ring resonator radius. One can also obtain from these curves the single-pass signal attenuation in dB per 90- degree turn due to bending, relevant to compact routing of signals between components, as LdB90 = 5 2πRω vgQ . In Fig. 4(a), one ring design has a half-cladded cross- section (see Fig. 1(b)[iii], dimensions in caption of Fig. 4), and is designed for a resonant frequency of 5.3 GHz. As the radius is varied, different azimuthal mode orders (number of wavelengths around the ring) correspond to the 5.3 GHz resonance frequency. A very small increase in radius (to a few microns) is needed to make the ra- diation loss negligible, and the total Q to be limited by another (e.g. intrinsic material loss) process. A second ring design uses a fully-cladded cross-section (Fig. 1(b)[i], dimensions in caption of Fig. 4), and is designed for a 3.0 GHz resonant frequency at various radii. The first design is shown in Fig. 4(c)[iii], for a radius (to center of ring waveguide) of 6 µm, coupled to a bus waveguide. The bus waveguide excited the ring resonator via evanes- cent coupling, i.e. same power transfer mechanism seen in the directional coupler in Figure 4(c)[i], discussed next. This provides a 2-port (notch, or all-pass) filter function. Both ring resonator geometries allow for tight micron scale bends that are suitable for integrated systems on chip. 2 ln 10 Because each azimuthal order yields a resonance, the ring resonator has a period pattern of resonances with a spacing called the free spectral range (FSR). Smaller- radius resonators have fewer wavelengths around, so they need a larger increase in frequency to add a full extra 8 wavelength and reach the next resonance condition – that is, the spacing between azimuthal mode frequencies is larger. The FSR is given by the inverse of the round trip travel time (group delay) around the ring, i.e. FSR= vg/Lrt, where Lrt is the cavity round trip length. Figure 4(b) shows the free spectral range (FSR) of the two specific designs showing that several MHz to several tens of MHz FSRs are supported. The FSR can accommodate a number of frequency channels and serve as a multiplexer if the FSR is much larger than a passband of one filter. Assuming radii are chosen large enough, bending loss Q's can exceed 104 to 105, so that intrinsic linewidths are 0.05 MHz to 0.5 MHz. The other loss mechanism considered here was material in- trinsic loss, and Fig. 3 shows that at 3 − 5.3 GHz fre- quencies the intrinsic linewidth is limited to 20− 70 kHz, corresponding to Q's of 75,000 to 150,000. Total loss Q is due to the summation of inverse Q's, 1/Qtotal = 1/Qintrinsic + 1/Qbending. Thus, total loss Q's in the 104 to 105 range might be expected, giving intrinsic cavity linewidths of 50 kHz to 500 kHz. A resonant filter of the kind shown in Fig. 4(c)[iii], configured with a waveguide- ring gap spacing to produce critical coupling results in a notch filter with twice the bandwidth, 100 kHz to 1 MHz. Thus, with several MHz FSR, a frequency demultiplexer comprising several channels to tens of channels could be designed. In general, since radiation loss is subject to design (e.g. choice of ring radius) while material losses are more constraining , a designer will usually aim for a radiation Q that considerably exceeds the intrinsic Q so as to not further degrade the loss Q and linewidth. This analysis shows that compact, few-micron-scale bends and resonators are realizable in an embedded acoustic waveg- uide platform with high slowness contrast. Beyond ring resonators, other acoustic components can be designed as well, by analogy with their optical counterparts. Having covered the viability of waveguides and res- onators, the two basic elements of wave systems in that one carries power and the other stores energy, we next turn to the fundamental elements needed to intercon- nect them. Our goal is again to evaluate the scaling of these components in the high slowness contrast regime. Fig 4(c) shows field profiles from frequency domain sim- ulations of an acoustic directional coupler [left] and a Y-splitter [middle], using an embedded waveguide cross- section with 0.5µm×0.5µm SiO2 core half-cladded with Si (c.f. Fig. 1b [iii]). The directional coupler (DC) is a fundamental compo- nent, having two input ports and two output ports, that enables a designed splitting ratio of the power in a wave at one input port into the two output ports. DCs enable the connection of resonators to ports as in the exam- ple given above, and the construction of interferometers. The DC shown in Fig 4(c)[left] employs evanescent cou- pling whereby modes of two waveguides that are in closer transverse proximity than the extent of their evanescent fields outside the core will interact and exchange signif- icant power if they are synchronous, i.e. their propa- gation constants are matched, which automatically oc- curs with two identical guides. A few comments can be made with respect to the scaling of directional couplers. The basic figure of merit for a directional coupler is the length for a given fraction of power transferred – here we choose the full-power transfer length as a baseline. In the high slowness contrast regime, the power transfer can be rapid and full-transfer lengths very short – order of 10 µm in the case of Fig. 4(c)[left]. Since power transfer from one waveguide to the other after length l of coupling is t212 = sin(κl)2, where κ is the coupling strength (in rad/m) which falls exponentially with the gap between the waveguides due to the exponential evanescent field. The full power transfer length, or beat length, is given by lfull = π/(2κ), and κ can be related to the propagation constants of the symmetric and antisymmetric guided su- permodes of the guided pair, κ = βs − βa – the stronger the coupling, the higher the splitting of βs and βa. Since the phase velocities of the supermodes vp,s and vp,a typ- ically fall between the core and cladding wave speeds for bulk dominated modes, the guidance condition and ma- terial slowness contrast imposes an upper limit on the coupling strength, κ < ω/vp,core − ω/vp,cladding. Clearly, high slowness contrast provides a larger upper bound to the coupling strength, and hence shorter coupling length. For our case using SiO2 core and Si cladding, and as- suming shear wave dominance in the modes, we can ap- proximately use 3,000 and 6,000 m/s as the respective velocities. For frequencies in the gray region in Fig. 3(b) (about 1 to 100 GHz), this lower bound on coupler length is 5 µm or shorter depending on frequency. Note that all of these estimates are independent of particular geome- try. In reality confinement and design for radiation loss will produce designs with longer lengths – consistent with our 10 µm length example in Fig 4(c)[left]. Directional couplers can be straight, and do not incur limitations of bending losses, although bringing isolated waveguides to them does, so the total length of a coupler including con- nections might be 2-3 times the size given our estimates of bend radii. Another fundamental component of wave circuits (such as integrated optical circuits, or microwave circuits) are Y-branch 3 dB splitters. Directional couplers provide ar- bitrary splitting ratios, but broadband designs are dif- ficult to achieve even in integrated optics, where 1% bandwidth is considered broadband (i.e. 2 THz of a 200 THz carrier). In acoustic circuits, one may desire 10% or higher bandwidths, requiring wideband design more akin to microwave engineering than integrated pho- tonics. DCs are furthermore sensitive to fabrication vari- ations. Hence, the Y-branch splitter is a device that guar- antees 3 dB (50:50) splitting, an important ratio for inter- ferometers and power splitter trees, by symmetry. The Y-branch splitter involves a splitting region and branch arms to separate the ports. The splitting region can be very compact – a few microns in both in-plane dimen- sions following similar slowness contrast arguments to the DC length. Based on bend radii of order 10 µm, 9 Fig 4(c)[middle] shows a 3 dB power splitter that us- ing S-bends that is about 20 µm long and 10 µm wide (including port separation) for 3.35 GHz operating fre- quency. Smaller structures are possible using multi-mode interference or more advanced taper concepts and more aggressive bend design. VI. CONCLUSION Microphononic circuits can realize both coupled- element circuits via evanescent coupling based on em- bedded waveguides (analogously to dielectric integrated photonic circuits), or via physically interconnected sus- pended structures (closer to microwave circuits based on metal-walled microwave cavities, where the free bound- ary in acoustics corresponds to the perfect electric con- ductor wall of a microwave cavity). In this paper, we investigated the former and found that the confinement and losses are consistent with enabling micron-scale de- vices and circuits that operate on 1-100 GHz bandwidth signals, where the low end of frequencies is likely to be limited by device size and the high end by losses and lithographic resolution. High-slowness-contrast embedded acoustic waveguides allow complete, radiation-free guided wave confinement in 1 µm scale cross-sections. Intrinsic material losses per- mit operation in the 1-100 GHz frequency range with Q's in the thousands and linewidths that permit efficient sig- nal processing – and tens of microns to millimeters or cen- timeters of low loss propagation depending on frequency. Basic building blocks based on evanescent confinement and coupling of embedded structures include waveguide bends, ring resonators, directional couplers and Y split- ters that all benefit from the high slowness contrast to provide 10 µm scale structures at few-GHz frequencies, thus enabling complex multi-element circuits. We believe this kind of microphononic circuit platform, which does not require suspended components, warrants further in- vestigation, and could provide valuable components inte- grable directly with microelectronics and microphotonics in CMOS as well as specialized custom chip technology. Evanescently confined microphononic circuits could find applications in sensing, communication and RF sig- nal processing, interfaced either to electronics or pho- tonics. With additional attention paid to simultaneous confinement of optical and acoustic waves, applications in optomechanics, signal domain transduction, and mi- crowave photonic signal processing could benefit. The presented study of guidance and loss supports the fea- sibility of such structures in dimension and frequency ranges of interest, and suggests the next steps of exper- imentally demonstrating these geometries. The possibil- ity of microphononic circuits being incorporated within planar CMOS technology, could enable complex systems- on-chip that can benefit from the narrowband signal pro- cessing capabilities and long time delays enabled by con- fined sound waves. Investigation of some of these possi- bilities is a worthy subject for future work. ACKNOWLEDGMENTS This work was supported by a 2012 Packard Fellowship for Science and Engineering (Grant #2012-38222). 1I. P. Kaminow, "Optical integrated circuits: a personal perspec- tive," Journal of Lightwave Technology 26, 994–1004 (2008). 2C. Sun, M. T. Wade, Y. Lee, J. S. Orcutt, L. Alloatti, M. S. Geor- gas, A. S. Waterman, J. M. Shainline, R. R. Avizienis, S. Lin, et al., "Single-chip microprocessor that communicates directly using light," Nature 528, 534–538 (2015). 3M. Popovi´c, Theory and design of high-index-contrast micropho- tonic circuits, Ph.D. thesis, Massachusetts Institute of Technol- ogy (2008). 4B. A. Auld, Acoustic fields and waves in solids, Vol. 2 (RE Krieger, 1990). 5K. Fang, M. H. Matheny, X. Luan, and O. Painter, "Opti- cal transduction and routing of microwave phonons in cavity- optomechanical circuits," Nature Photonics (2016). 6A. H. Safavi-Naeini, T. M. Alegre, J. Chan, M. Eichenfield, M. Winger, Q. Lin, J. T. Hill, D. Chang, and O. Painter, "Elec- tromagnetically induced transparency and slow light with op- tomechanics," Nature 472, 69–73 (2011). 7H. Shin, J. A. Cox, R. Jarecki, A. Starbuck, Z. Wang, and P. T. Rakich, "Control of coherent information via on-chip photonic– phononic emitter–receivers," Nature communications 6 (2015). 8M. Merklein, B. Stiller, K. Vu, S. J. Madden, and B. J. Eggleton, "A chip-integrated coherent photonic-phononic memory," arXiv preprint arXiv:1608.08767 (2016). 9S. Y. Shah, M. Zhang, R. Rand, and M. Lipson, "Master-slave locking of optomechanical oscillators over a long distance," Phys- ical review letters 114, 113602 (2015). 10J. Chan, T. M. Alegre, A. H. Safavi-Naeini, J. T. Hill, A. Krause, S. Groblacher, M. Aspelmeyer, and O. Painter, "Laser cooling of a nanomechanical oscillator into its quantum ground state," Nature 478, 89–92 (2011). 11X. Sun, K. Xu, and H. X. Tang, "Monolithically integrated, ultrahigh-frequency cavity nano-optoelectromechanical system with on-chip germanium waveguide photodetector," Optics let- ters 39, 2514–2517 (2014). 12A. Oliner, "Waveguides for acoustic surface waves: A review," Proceedings of the IEEE 64, 615–627 (1976). 13C. C. Ruppel and T. A. Fjeldly, Advances in surface acoustic wave technology, systems and applications, Vol. 1 (World Scien- tific, 2000). 14C. K. Campbell, "Applications of surface acoustic and shallow bulk acoustic wave devices," Proceedings of the IEEE 77, 1453– 1484 (1989). 15R. Shelby, M. Levenson, and P. Bayer, "Guided acoustic-wave brillouin scattering," Physical Review B 31, 5244 (1985). 16N. Nishiguchi, "Guided acoustic phonons in quantum wires: The- ory of phonon fiber," Japanese journal of applied physics 33, 2852 (1994). 17H. Shin, W. Qiu, R. Jarecki, J. A. Cox, R. H. Olsson III, A. Star- buck, Z. Wang, and P. T. Rakich, "Tailorable stimulated bril- louin scattering in nanoscale silicon waveguides," Nature com- munications 4 (2013). 18R. Marathe, B. Bahr, W. Wang, Z. Mahmood, L. Daniel, and D. Weinstein, "Resonant body transistors in IBM's 32 nm SOI CMOS technology," Microelectromechanical Systems, Journal of 23, 636–650 (2014). 19D. Hatanaka, I. Mahboob, K. Onomitsu, and H. Yamaguchi, "Phonon waveguides for electromechanical circuits," Nature nan- otechnology 9, 520–524 (2014). 20C. J. Sarabalis, J. T. Hill, and A. H. Safavi-Naeini, "Guided acoustic and optical waves in silicon-on-insulator for brillouin 10 scattering and optomechanics," APL Photonics 1, 071301 (2016), http://dx.doi.org/10.1063/1.4955002. 21C. G. Poulton, R. Pant, A. Byrnes, S. Fan, M. Steel, and B. J. Eggleton, "Design for broadband on-chip isolator using stimu- lated brillouin scattering in dispersion-engineered chalcogenide waveguides," Optics express 20, 21235–21246 (2012). 22C. G. Poulton, R. Pant, and B. J. Eggleton, "Acoustic confine- ment and stimulated brillouin scattering in integrated optical waveguides," JOSA B 30, 2657–2664 (2013). 23C. Multiphysics, "Comsol multiphysics user guide (version 4.3 a)," COMSOL, AB , 39–40 (2012). 24N. Lobontiu, Dynamics of microelectromechanical systems, Vol. 17 (Springer Science & Business Media, 2014). 25G. Wu, D. Xu, B. Xiong, and Y. Wang, "Effect of air damping on quality factor of bulk mode microresonators," Microelectronic Engineering 103, 86–91 (2013). 26J. Brotz, "Damping in CMOS-MEMS resonators," Masters Project Report in ECE. Carnegie Mellon University (2004). 27Y. Liu and M. A. Popovi´c, "High-Q contacted ring microcavities with scatterer-avoiding "wiggler" bloch wave supermode fields," Applied Physics Letters 104, 201102 (2014). 28Y. Liu, J. M. Shainline, X. Zeng, and M. A. Popovi´c, "Ultra- low-loss CMOS-compatible waveguide crossing arrays based on multimode bloch waves and imaginary coupling," Optics letters 39, 335–338 (2014). 29V. B. Braginsky, V. Mitrofanov, V. I. Panov, and C. Eller, Sys- tems with small dissipation (University of Chicago Press, 1985). 30A. Duwel, J. Lozow, C. J. Fisher, T. Phillips, R. H. Olsson, and M. Weinberg, "Thermal energy loss mechanisms in micro- to nano-scale devices," SPIE Defense, Security, and Sensing , 80311C–80311C (2011). 31H. J. Maris, "Interaction of sound waves with thermal phonons in dielectric crystals," Physical Acoustics 8, 279–345 (2012). 32S. Chandorkar, M. Agarwal, R. Melamud, R. Candler, K. Good- son, and T. Kenny, "Limits of quality factor in bulk-mode mi- cromechanical resonators," MEMS 2008. IEEE 21st International Conference on , 74–77 (2008). 33F. Ayazi, L. Sorenson, and R. Tabrizian, "Energy dissipation in micromechanical resonators," SPIE Defense, Security, and Sens- ing , 803119 (2011). 34L. D. Landau and E. M. Lifshitz, Course of Theoretical Physics Vol 7: Theory and Elasticity (Pergamon press, 1959). 35A. Akhiezer, "On the sound absorption in solids," Journal of Physics (USSR) 1, 277 (1939). 36T. Woodruff and H. Ehrenreich, "Absorption of sound in insula- tors," Physical Review 123, 1553 (1961). 37D. Li and D. G. Cahill, "Attenuation of 7 ghz surface acoustic waves on silicon," Physical Review B 94, 104306 (2016). 38O. L. Anderson, "The gruneisen ratio for the last 30 years," Geo- physical Journal International 143, 279–294 (2000). 39R. Nava, M. Vecchi, J. Romero, and B. Fernandez, "Akhiezer damping and the thermal conductivity of pure and impure di- electrics," Physical Review B 14, 800 (1976). 40A. S. Grove, Physics and technology of semiconductor devices (John Wiley & Sons Inc, 1967). 41C. Livermore and J. Voldman, "6.777j/2.751j material prop- erties database," http://web.mit.edu/~6.777/www/matprops/ pecvd_sio2.htm (), accessed Dec 19, 2016. 42M. Madou, "Fundamentals of microfabrication. 1997," CRC Press LLC. Cap´ıtulo 1 (2003). 43M. Kaviany, Principles of heat transfer (John Wiley & Sons, 2002). 44R. J. Wang, W. H. Wang, F. Y. Li, L. M. Wang, Y. Zhang, P. Wen, and J. F. Wang, "The gruneisen parameter for bulk amorphous materials," Journal of Physics: Condensed Matter 15, 603 (2003). 45C. Livermore and J. Voldman, "6.777j/2.751j material prop- erties database," http://www.mit.edu/~6.777/matprops/pecvd_ sin.htm (), accessed Dec 19, 2016. 46D. Morelli and J. Heremans, "Thermal conductivity of germa- nium, silicon, and carbon nitrides," Applied physics letters 81, 5126–5128 (2002). 47R. Bruls, H. Hintzen, G. de With, R. Metselaar, and J. van Miltenburg, "The temperature dependence of the Gruneisen pa- rameters of MgSiN2, AlN and β-Si3N4," Journal of Physics and Chemistry of Solids 62, 783–792 (2001). 48M. Lewis, "Attenuation of high-frequency elastic waves in quartz and fused silica," The Journal of the Acoustical Society of Amer- ica 44, 713–716 (1968). 49W. P. Mason, Crystal physics of interaction processes, Vol. 23 (Academic Press, 1966). 50J. Lamb, M. Redwood, and Z. Shteinshleifer, "Absorption of compressional waves in solids from 100 to 1000 mc/sec," Physical Review Letters 3, 28 (1959). 51B. Helme and P. King, "The phonon viscosity tensor of Si, Ge, GaAs, and InSb," physica status solidi (a) 45 (1978). 52C. Wolff, R. Soref, C. Poulton, and B. Eggleton, "Germanium as a material for stimulated brillouin scattering in the mid- infrared," Optics express 22, 30735–30747 (2014). 11
1812.03806	1	1812	2018-10-31T15:50:58	Broadband, Temperature Tolerant and Passively Biased Resonantly Enhanced Mach-Zehnder Modulators	[ "physics.app-ph", "physics.optics" ]	We describe a resonantly enhanced Mach-Zehnder modulator (MZM) that can be operated over a wide temperature range of 55C without being actively biased, while providing a significant resonant enhancement of 6.8 at the nominal wavelength / temperature compared to a linear MZM driven with a distributed driver. More importantly, it enables a ~20X improvement in power consumption compared to a 50 {\Omega} matched linear traveling wave modulator with comparable phase shifter technology, drive voltage and output optical modulation amplitude. Passive biasing of the Mach-Zehnder interferometer is further implemented by replacing a splitter element in the MZM with a novel device combining splitting and fiber coupling functionalities in a single, multi-modal structure, that converts permanent fiber placement into a phase correction. Both concepts are combined in a single modulator device, removing the need for any type of active control in a wide temperature operation range.	physics.app-ph	physics	Broadband, Temperature Tolerant and Passively Biased Resonantly Enhanced Mach-Zehnder Modulators S. Romero-García, A. Moscoso-Mártir, J. Nojic, S. Sharif-Azadeh, J. Müller, B. Shen, F. Merget, J. Witzens Institute of Integrated Photonics RWTH Aachen University Aachen, Germany importantly, Abstract -- We describe a resonantly enhanced Mach-Zehnder modulator (MZM) that can be operated over a wide temperature range of 55oC without being actively biased, while providing a significant resonant enhancement of 6.8 at the nominal wavelength / temperature compared to a linear MZM driven with a distributed driver. More it enables a ~20X improvement in power consumption compared to a 50 W matched linear travelling wave modulator with comparable phase shifter technology, drive voltage and output optical modulation amplitude. Passive biasing of the Mach-Zehnder interferometer is further implemented by replacing a splitter element in the MZM with a novel device combining splitting and fiber coupling functionalities in a single, multi-modal structure, that converts permanent fiber placement into a phase correction. Both concepts are combined in a single modulator device, removing the need for any type of active control in a wide temperature operation range. Keywords -- Electro-optic modulators; Datacom; Resonant ring modulators; Electro-optic transceivers. I. INTRODUCTION Reducing the power consumption of electro-optic transceivers in data centers has become a key objective on par with the reduction of manufacturing cost: Since interconnects have been estimated to consume 23% of the total datacenter energy intake [1], their power consumption is a burden on both operating expenses (OpEx) and the required cooling infrastructure. reducing the power consumption associated Travelling wave (TW) modulators can make up a substantial portion of an electro-optic transmitter's power consumption, which has motivated the investigation of alternative modulator and driver topologies enabling a significant reduction of the required power. Distributed drivers are one such alternative [2] that has already found its way into industrial practice [3]. This allows to dissipation of the high-speed signal delivered to the modulator (the RF power consumption) by a factor 6 to 8 [4]. Another approach consists in shrinking the size of the modulator to a fraction of the radio-frequency (RF) wavelength, to enable driving it as an electrical lumped element with a nearby low output impedance driver. In order to overcome the reduction of modulation efficiency resulting from shrinking of the phase shifter length, resonantly enhanced devices such as resonant ring modulators (RRMs) have attracted considerable attention in silicon photonics [5]. Discussions in the literature typically focus on the resonant enhancement itself, i.e., the increase in modulation efficiency as compared to a linear, non-resonant modulator with an equally long embedded phase shifter length. Equivalently, this essentially also corresponds to the power consumption reduction as compared to a linear modulator with a phase shifter length chosen to result in the same modulation contrast, assuming it can be driven as a lumped element. Another important improvement, however, arises from the much-reduced dimensions of resonantly enhanced modulators, which allow them to be driven as a lumped element in the first place. While longer linear devices can be driven with a similar RF power consumption, as associated to the signal delivered to the modulator, as a lumped element modulator by means of a distributed driver, a single lumped element load allows achieving improvement without incurring the additional integration or packaging complexity associated to distributed drivers, i.e., the monolithic integration of a distributed driver via CMOS photonic process integration [3] or the more conventional route of co-packaging of the distributed driver via flip-chipping and a micro-bump array. A lumped element modulator, on the other hand, can be simply integrated with its driver via a pair of wire-bonds, provided the capacitive load of the modulator remains sufficiently small to tolerate the parasitic inductance of the wire-bonds [6]. While the RF power consumption inside the modulator itself only constitutes a portion of the power drawn by the modulator driver, at least the output stage of the driver needs to be sized according to the load, so that it remains an important quantity. this power consumption In addition to the RF power consumption, additional power is associated to the stabilization of the modulators against temperature changes and process bias: Silicon possesses a relatively high thermo-optic coefficient of 1.87e-4 RIU/K at room temperature [7] and the effective index of silicon/SiO2 single mode waveguides is particularly prone to roughness induced variations due to the high index contrast, so that index variations in interferometric and resonant devices have to be compensated for. This is a tractable problem in linear modulators, that are nominally balanced so that variations induced by temperature swings are relatively modest and can be straightforwardly compensated for. Self-referenced resonant devices on the other hand are very sensitive to temperature changes and process biases: The typical resonance wavelength non-uniformity across dies due to fabrication and layer thickness is on the order of ±1 nm [8]. Although advanced schemes have been developed to reduce phase tuning requirements, e.g. in multi-channel WDM transceivers [9], in a typical system configuration wavelength tuning by a full free spectral range For (FSR) is required at start-up, followed by additional tuning during operation to compensate for environmental temperature swings. While thermally isolating resonant devices has resulted in very substantial improvements of the power efficiency of thermal tuners [10], this only partially alleviates the problem, as on the one hand process complexity is increased and on the other hand limitations to the maximum allowable device temperature as constrained by device reliability remain. Finally, athermal resonant devices have been realized by means of cladding materials, thermooptic coefficients [11]. Here too, process complexity is increased and long term reliability associated to the introduction of polymers needs to be ascertained. typically polymers, with opposite the aforementioned reasons, a lumped element, resonantly enhanced device that does not require dynamic tuning and that can be fabricated in a standard silicon photonics fabrication line would be highly desirable. Here, we combine two techniques to achieve this objective: We have previously shown that multi-mode fiber/laser-to-chip couplers initially developed to relax alignment tolerances [12],[13] can also be used to passively and permanently bias a Mach-Zehnder interferometer [14]. Moreover, we have adapted a device architecture in which collectively driven RRMs are utilized as phase shifters [15] to obtain optically broadband operation while maintaining a sizeable resonant enhancement [16] and high- speed, lumped element operation [6]. In the following, we will first review results pertaining to these two schemes and continue with a description of the combined device. II. PASSIVE BIASING OF MZMS BY MEANS OF MULTIMODE COUPLERS In [12],[13] we showed laser-to-chip edge couplers and fiber-to- chip grating couplers that have in common their ability to relax required alignment tolerances in one direction (transverse relative to the main optical axis of the devices) and the fact that they both have two single mode output waveguides. The couplers were architectured in such a way that the power coupled to both single mode output waveguides is substantially equal, but that the phase of the light coupled to either waveguide varies as a function of the placement of the input fiber or input laser. Since we sought to improve alignment tolerances without reducing the peak insertion efficiency, this degree of freedom was required so as to not violate the reciprocity theorem. The devices are multi-mode in nature and essentially combine a coupler and a Y-junction without forcing the light through a unique single mode optical path in-between. The operation principle of the devices can be found in the references [12],[13]. In a parallel single mode optics (PSM) transceiver, this varying phase at the output of the coupler is irrelevant, since the light is never recombined: One seeks to guarantee a minimum power in any of the downstream parallel communication channels (obtained by further splitting of the light). In [14] we repurposed these devices in order to use the phase shift at their output as a feature: We implemented a Mach- Zehnder interferometer (MZI), in which the input splitter was replaced by one such multi-mode grating coupler (MMGC). By moving the position of the input fiber across the MMGC, the relative phase at the two output waveguides is varied until the quadrature point (biasing at -3 dB MZI output power) is reached, so that tolerant after which the fiber is affixed in place by means of a UV- curable epoxy. sufficiently temperature Adequacy of this technique is contingent on a few elements: The fiber position can be assumed to be reliably fixed in place since packaging based on UV curable epoxy is an established technology. Moreover, the couplers themselves can be designed to be their functionalities, 3 dB splitting and a fixed phase offset at the output, is maintained over a wide temperature range. However, one also has to assume that the mismatch of the interferometer is of a sufficiently low order for the static correction of a fixed phase term to rebalance it over all relevant temperatures and wavelengths (which is easy to verify or to invalidate for a given interferometer). Lastly, one has to assume that the phase mismatch occurring in the two arms of the interferometer does not drift over time due to additional effects, such as for example the diffusion of fixed charges in surrounding dielectrics, that can not only cause losses, but also effective index changes [17]. This is the subtlest assumption to verify as it relates to long term reliability and remains an open question. Lastly, this scheme requires that at least one fiber per interferometer can be freely positioned, which would for example prevent packaging of a modulator array with a single fiber array. This may be the primary limitation of this approach. An important strength lies however in the fact that the phase correcting mechanism itself should be very robust and compatible with long term reliability. The concept of the alignment tolerant grating coupler is shown in Fig. 1. As can be seen in the bottom panel, the power coupled in either waveguide is equal and remains within 1 dB of its nominal value (centered fiber) within a ±7.2 µm misalignment range in the transverse/lateral direction (red arrow in the micrograph), a tolerance almost 3X larger than a conventional grating coupler. In the longitudinal direction, the Fig. 1. (top) Micrograph of an alignment tolerant grating coupler (MMGC) as reported in [13] and (bottom) corresponding coupling efficiency into either of the two output waveguides (black and red curves) as well as the cumulative coupling efficiency (solid blue curve) as compared to a regular single mode grating coupler (SMGC) fabricated in the same technology (dash dotted blue curve). The transverse misalignment, as shown by the red arrow in the micrograph, is indicated on the x-axis. and did not yet yield large phase shifts, they are interesting in that they target standard silicon photonics process flows without further modifications. The method shown in [18] in particular can be expected to feature long term stability. In all these cases, long term stability of the trimming mechanism itself and of other phase shifts within the device is a primary concern for the practicability of the trimming mechanism. III. TEMPERATURE TOLERANT, RESONANTLY ENHANCED MZMS it can be seen that the ring behaves as a linear waveguide whose finesse given as the ratio of the FSR to the full width at half !"#$=!&'()+,--./0 The phase shifters in the MZM are implemented in the form of collectively driven, highly overcoupled RRMs loaded on a common bus waveguide [16]. In the highly overcoupled regime, when the optical carrier is tuned on resonance, the transfer function of a single ring resonator can be given as [22] where !&', !"#$ are the field amplitudes in the bus waveguide, respectively before and after the ring, 1, 2 are respectively the average wave number and linear losses inside the ring, 3 is the maximum (FWHM) and 4 is the circumference of the ring. Thus, length 4 has been demultiplied by a factor 23 6, hence the coupling junction, the 7/4∙2 figure of merit of the phase shifter remains unchanged, wherein 7/4 is the drive voltage required to achieve a phase shift of 6 in a phase shifter of length 4. 23 6 (assuming the embedded phase shifter to cover the entire reducing the circumference 4 of the ring (this ought to make resonant enhancement factor applied to an embedded phase shifter section. Importantly, in the absence of excess losses due to e.g. waveguide bends or excess losses in the ring to waveguide In order to obtain a high optical bandwidth, an increased FWHM and thus a reduced resonator quality (Q-)factor are required. This also reduces the resonant enhancement factor circumference) since the finesse scales as the opposite of the FWHM. This can be compensated by increasing the FSR, i.e., sense intuitively, as the capacitance of the embedded phase shifter, and thus its power consumption, scales with its length). While this is conceptually simple, its reduction to practice is quite tricky: As the circumference shrinks, the sensitivity of the device to increased coupling losses rises (since the excess losses are distributed over the waveguide length to convert them into effective waveguide losses). In addition, bending losses go up, also contributing to increase 7/42. Finally, it becomes losses to maintain the same 7/42 and sufficiently overcoupling increasingly difficult to obtain sufficiently high coupling strengths over shrinking distances in order to maintain the high overcoupling regime required to spoil the resonator Q-factor without incurring excess losses. The design problem can thus be formulated as shrinking the device, while minimizing excess the ring. A diagram and a scanning electron microscope (SEM) image of the fabricated waveguides (prior to back-end-of-line fabrication) can be seen in Figs. 3(a) and 3(b). The overlay with implant regions and the position of the electrodes can be seen in Fig. 3(c) and a micrograph of the complete RRM in Fig. 3(d). Several tricks were applied to achieve the design goals: Away from the coupler region, the waveguide inside of the ring is Fig. 2. (a) MZI with multimode grating coupler as input splitter. (b) Measured MZI output power levels (relative to the maximum transmission) as a function of the input fiber position. (c) Imbalance between the two MZI output ports as a function of temperature after permanent fiber attachment. device behaves like a conventional grating coupler, i.e., the alignment tolerance is not enhanced. The total coupled power is only slightly worse to that of the conventional grating coupler, with the slight excess losses attributed in particular to the built- in Y-junction (highlighted with a black rectangle in the micrograph). Importantly, as the fiber is moved transversely across the grating coupler, the relative phase of the light coupled into the two output waveguides is varied, which is utilized in the following (in which case the alignment tolerance is again lost, as now placement is constrained by other considerations than coupling efficiency). The concept of the entire MZI is illustrated in Fig. 2(a), experimental results are shown in Fig. 2(b). As the fiber is moved along the transverse direction above the grating coupler, the phase in the two branches of the MZI is varied until the power at the output of the MZI, after an additional 2 by 2 multimode interferometer (MMI), is exactly balanced. The fiber is then epoxied in place and the temperature dependent characteristics of the interferometer are measured. As seen in figure 2(b), the initial imbalance of the device, as measured with the optical fiber centered on the optical axis of the grating coupler, was 13 dB as a consequence of fabrication induced mismatch between the two 1 mm long arms of the interferometer. After displacement and permanent attachment of the fiber this imbalance was maintained below 0.4 dB even while cycling the device temperature between 20oC and 80oC. This shows that in this case the imbalance order of the MZI was sufficiently low for the MZI to remain balanced over the entire temperature range and confirms that the temperature tolerance of the grating coupler, also in terms of its output phases, is sufficient for this purpose. For details on the devices, please refer to references [12]-[14]. A number of other techniques have been shown to modify the bias point of passive interferometers or resonant devices. These comprise local oxidation of the waveguide [18], optically induced modification of the refractive index of a thin chalcogenide film surrounding the waveguide by means of UV radiation [19], thermally induced refractive index change in the cladding [20], or thermally/current induced dopant diffusion [21]. While the method shown in [20] does not yet feature long term stability and the method shown in [21] was self-limiting Fig. 3. (a) Diagram of the waveguide in a single RRM layout from [16], (b) scanning electron micrograph after waveguide fabrication, (c) mask layout showing overlay with the PIN junction as well as electrode placement, and (d) micrograph of the complete RRM. widened in order to reduce bending losses. This, however, results in too small a coupling parameter, so that the waveguide has to be narrowed again in the coupling region. In order to suppress resulting bending losses, the silicon is fully etched on the opposite side of the bus waveguide, which can be done while maintaining electrical connectivity of the embedded PIN diode. This does result, however, in the waveguide region inside the coupling region to have an increased series resistance to the electrodes and thus a reduced RC limited cutoff frequency. In order to prevent this from impacting the electro-optic S21 of the device, as well as to prevent insertion losses associated to this inefficiently modulated region, it is left unimplanted (Fig. 3(c)). The device reported in [16] was fabricated in the standard silicon photonics process of IME ASTAR with a 220 nm silicon device layer thickness and a shallow etch depth of 130 nm (90 nm slabs). The outer radius of the resonators was chosen as R = 5 µm, the waveguide width varied between W1 = 425 nm (in the coupling section) and W2 = 1.575 µm. The bus waveguide width in the coupling regions was chosen as 370 nm and the bus waveguide to ring resonator gap as 200 nm to allow fabrication with 248 nm DUV lithography. p- and n-regions were doped with concentrations of respectively 2.5e18 cm-3 and 2e18 cm-3 and were spaced by an intrinsic region of nominally 20 nm (same phase shifter design as the third category of devices in [23]) and cover ¾ of the ring's circumference. the 7/42 of the phase shifter. In order to reduce these, a wiggle As mentioned above, excess losses occurring at the waveguide to ring junction are critical in order to avoid spoiling was introduced in the bus waveguide shape in order to smoothly taper the gap size [24]. As shown in Fig. 3(a), the bus waveguide follows a circular shape with the same center as the ring over a small angle q, outside of which it continues on a straight line prior to being bent back to the next resonator. Figs. 4(a) and 4(b) show simulated excess round trip losses and power coupling strengths for different resonator designs and q values, including for the nominal design described above with W1 = 425 nm, W2 = 1.575 µm, R = 5 µm and q = 5o. Figs. 4(c) and 4(d) show corresponding experimental data extracted fitted resonances, also for W1 = 425 nm, W2 = 1.575 µm and R = 5 µm. from Fig. 4. (a) Simulated bus to ring waveguide power coupling coefficient and (b) simulated round trip excess losses as a function of q. (c) and (d) show the corresponding experimental data overlaid with simulation data for the nominal ring design corresponding to W1 = 425 nm, W2 = 1.575 µm, and R = 5 µm. Fig. 5. (a) FWHM of the measured ring resonators (W1 = 425 nm, W2 = 1.575 µm, and R = 5 µm as in Fig. 4) as a function of q and (b) insertion losses as a function of the number of resonators N. The resulting FWHM and excess modulator insertion losses, with the optical carrier on resonance, are shown in Fig. 5(a) and 5(b), wherein the resulting resonator excess insertion losses already contain the resonant enhancement factor (these correspond to the losses incurred at the bottom of the Lorentzian resonator transfer function multiplied by the number of resonators). The complete modulator was implemented in a subsequent fabrication run with 5 collectively driven resonant ring modulators loaded on each arm, resulting in a cumulative PIN junction length of 110 µm on each interferometer arm, total on- resonance modulator insertion losses of 5.7 dB, only 1.2 dB of which were attributed to excess losses (the rest corresponding to the resonantly enhanced losses of the embedded phase shifters), total capacitance and resistance per modulator arm of respectively 72 fF and 48 W, and a resonator FWHM of ~2 nm. The reduction of the FWHM as compared to the 3.5 nm measured in the previous run is attributed to a reduction of the coupling coefficient k2 from ~0.6 to ~0.4 due to process biases. The actual optical bandwidth of the modulator, defined as the optical carrier wavelength range in which the modulation efficiency remains within a factor 2 of its maximum, was measured as 3.9 nm. The VpL for an on-resonance optical carrier was measured as 0.19 V(cid:1)cm (as normalized relative to the cumulative PIN junction length), 6.8 times lower than a linear phase shifter with the same PIN junction design. This enhancement was actually slightly below the actual resonant enhancement of 8, due to a slight reduction in phase shifter efficiency associated to the reduced optical field confinement in the wider sections of the ring (compared to the reference modulator with a 400 nm waveguide width). The modulator could be operated with a cutoff frequency of 23.5 GHz in a 50 W environment, as limited by the RC time constant associated to the capacitance of the 5 parallel rings and the sum of the 48 W total phase shifter resistance with the 50 W of the driving circuitry. With a low output voltage lumped element driver one would expect the RC time constant to be significantly improved, up to the 46 GHz intrinsic phase shifter cutoff frequency. We showed co-operability of the device with a wire bonded 25 Gbps low output impedance lumped element driver, but increased data rates could not yet be shown due to the speed limitations of the utilized driver. With 2 Vpp and 4 Vpp drive signals, the modulator yielded modulation penalties associated to the finite drive voltages (as measured as a reduction in the optical modulation amplitude) of respectively -4.8 dB and -2.2 dB for an on- resonance optical carrier. identical PIN junction design (embedded The factor 6.8 reported above essentially corresponds to the reduction in RF power consumption associated to the signal delivered to the modulator, as compared to a linear modulator with in high confinement, 400 nm wide waveguides) and sized to have identical modulation penalty. The insertion losses of the resonantly enhanced modulator, 1.2 dB higher than the equivalent linear modulator, however also have to be factored in to yield a fair comparison. Increasing the drive voltage from 2 Vpp to 2.7 Vpp would for example compensate for these additional insertion losses, provided the driver technology is able to support such a voltage increase, resulting in a reduced effective power enhancement slightly below 4. From this perspective, the improvement of the device seems very modest, particularly in view of the remaining thermal sensitivity. This would however be missing the main improvement, as the comparison above is really between the resonantly enhanced modulator and an equivalent linear modular, assuming it can be driven as a lumped element, which, with a cumulative length of > 500 µm would not be the case at RF frequencies above a few GHz. Thus, in order for this comparison to hold, the linear modulator would need to be driven by a distributed driver. In other words, the resonantly enhanced device provides the benefits of lumped element driving without requiring a The improved distributed driver. Compared to a TW wave device, the ~8.1 mW required by the resonantly enhanced device at 25 Gbps assuming a 3 Vpp drive voltage is a very substantial improvement (20X compared to a dual drive, 50 W, linear TW MZM driven with a 2 Vpp signal taking into account the reduction of the drive signal due to lower insertion losses). temperature tolerance should also be discussed: In [16] we showed that the optical modulation amplitude (OMA) at the output of the device remained within a factor two of its maximum in a 3.8 nm wavelength range at fixed temperature, or, equivalently, in a 55oC temperature range if the wavelength of the laser remains fixed. While this represents more than a factor 10 increase in wavelength / temperature tolerance compared to a typical high-speed RRM with a Q-factor of ~5000 [23], the question remains whether this is sufficient to operate a transceiver completely without tuning of the rings. The typical wavelength repeatability of off-the-shelf laser diodes is on the order of ±1 nm to ±2 nm. Since the repeatability of silicon photonics resonant devices across different dies and wafers is also on the order of ±1 nm, the initial wavelength tolerance can be seen to be largely eaten up by these sources of variabilities. While the remaining wavelength / temperature tolerance of below 1 nm / 14oC is too low on its own, a few elements may help: For one, since the laser and the modulator are operated in the same transmitter, temperatures can be assumed to largely co- vary in both devices, so that wavelength drifts will also be comparable given the similar thermo-optic coefficients of Si and InGaAsP [25]. Moreover, one may envision implementing the laser as an external cavity laser with the resonance wavelength determined by the silicon photonics chip [26], in which case variation of the wavelength as caused e.g. by varying Si film thicknesses might be applied to both devices, in particular if the tunable reflector closing the laser cavity and the modulator are laid out close to each other. Nonetheless, increasing the optical bandwidth of the device appears desirable, not only to increase the robustness of the system, but also so as to not incur the full additional 3 dB modulation penalty associated to operating the device at the edges of its optical passband. Fortunately, this can be improved by increasing the number of rings loaded on each of the modulator arms, with the only penalty being, at moderate increase, a corresponding increase in power consumption, which, starting from ~8 mW is quite acceptable. Practical difficulties associated to increasing the number of rings are expected to be related to driving the increased capacitance, for example increasing the difficulty of designing a fitting low output impedance driver due to increase current sourcing requirements or worsening the detrimental effect of parasitic inductances associated to packaging. Figure 6 shows the modeled optical OMA at the output of the temperature tolerant MZM described above, under the assumption of a 2 Vpp drive voltage and further assuming that the number of collectively driven RRMs is varied between 3 and Fig. 6. OMA as a function of the number of collectively driven RRMs per MZM branch, assuming a drive voltage of 2 Vpp. driven in dual drive configuration with a drive voltage swing Vd can be simply calculated as 9:!;< =10?@ABCDEF7G∙4∙6 7/4 −10∙2∙4 ?@A10 where a is the linear loss term associated to the phase shifters. Even when ignoring the nonlinearity introduced by the sine function, one can straightforwardly see that this results in an optimum modulator length due to the different scaling of ?@A4 and 4. This is also what drives the trends in Fig. 6: As the number shifter length at resonance 2I34JJK 6, where 4JJK is the of rings is increased, and thus also the total effective phase (2) is significantly circumference of a single RRM, the phase shifter length grows beyond it optimum for an on-resonance carrier frequency. For highly detuned carrier frequencies on the other hand the effective phase shifter length is lower, as the resonant enhancement these frequencies the effective phase shifter length is still below optimum and the phase shifter efficiency continues to grow with the number of rings. The OMA is maximized for two intermediate frequencies at which the effective phase shifter length is exactly optimum, with these two frequencies moving away from the center frequency as the number of rings is increased. reduced. Thus, for IV. COMBINED DEVICE As a final step, we combined the two concepts described in the following sections in a single device. A micrograph of the combined device can be seen in Fig. 7. Two fibers are mounted in a fiber array and are moved laterally, together with a fixed pitch, relative to underlying multi-mode grating couplers. The two branches of the interferometer are routed between the two multi-mode grating couplers in such a way that the resulting phase shifts in either MMGC add up rather than cancelling each Fig. 7. Micrograph of the passively biased, temperature tolerant resonantly enhanced Mach-Zehnder modulator. Two fibers with a fixed pitch constrained by a fiber array are positioned on top of multi-mode grating couplers also acting as splitter / combiner for the interferometer. One of the interferometer branches is routed around the first MMGC (red arrow) in order to maintain the phase sensitivity of the device [13]. 7 per MZM arm (5 being the nominal value for the experimental results reported above). Importantly, as mentioned above, as the number of RRMs is increased, the optical bandwidth of the device widens. At the same time, the OMA at the center frequency drops, which is not problematic at first if it remains above the minimum specified in the definition of the passband. This can be understood with the following considerations: The OMA of a linear modulator with phase shifters of length L Fig. 8. Recorded signal Q-factors and extinctions with an on-resonance optical carrier as a function of the transverse fiber array displacement. The maximum Q-factor / OMA are obtained for a fiber positioning corresponding to the quadrature point, at which the extinction is 3 dB above its minimum. Recorded eye diagrams are shown for selected fiber array positions. other out [13]. Note that this requires one waveguide to be routed around the first MMGC as indicated by a red arrow. The temperature tolerant MZM itself corresponds to exactly the same design as reported in the previous section, and has thus the same temperature and wavelength tolerance. Fig. 8 shows the signal Q-factor recorded at the output of the device, as well as the extinction of the in-coupled light occurring inside the MZM (grating coupler losses, as measured at the optimal lateral displacement, normalized out), as a function of the transverse fiber array displacement. The Q-factor was measured with a 13 dBm laser power and a Finisar/U2T XPRV2021A 40 GHz bandwidth photoreceiver with an input referred transimpedance amplifier noise density specified to be below 40 N! OP and the a responsivity between 0.5 A/W and 0.75 A/W. It can be seen that the signal Q-factor is maximized at a fiber displacement at which the on-resonance optical carrier excess loss is exactly 3 dB above its minimum, i.e., at a transverse displacement corresponding its quadrature point. Moreover, it can be seen that the sensitivity of the modulator bias point on fiber misalignment is approximately doubled relative to Fig. 2, by nature of two MMGCs being utilized in the device together with a fiber array and the incurred displacement induced phase shift thus being doubled. The signal Q-factor, and thus the OMA, remain within 3 dB of their maxima in a transverse fiber misalignment range of ±1.8 µm. interferometer being biased at to V. CONCLUSIONS In conclusion, we have shown a resonantly enhanced Mach- Zehnder modulator driven as a lumped element load with a ~20X reduction of power consumption compared to a linear travelling wave device with identical output optical modulation amplitude. The optical modulation amplitude is maintained within 3 dB of its maximum in a 55oC temperature range, which can be further increased by increasing the number of collectively driven RRMs on each branch. The Mach-Zehnder modulator was passively biased at its quadrature point by selective placement of the input and output fibers on top of multi-mode grating couplers also used as the input and output splitters of the interferometer. ACKNOWLEDGMENT The authors gratefully acknowledge funding from the European Research Council under grant agreement 279770 and from the Excellence Initiative of the German State and Federal governments. REFERENCES [1] Where http://www.greendataproject.org (2008). does power go? GreenDataProject, accessed at [2] N. Qi, X. Li, H. Li, X. Xiao, L. Wang, Z. Li, Z. Gao, Y. Yu, M. Moyal, P. Chiang, "A 25Gb/s, 520mW, 6.4Vpp Silicon-Photonic Mach-Zehnder Modulator with Distributed Driver in CMOS," in Proc. 2015 Opt. Fib. Comm. Conf. (OFC), Art. ID W1B.3. [3] A. Narasimha et al., "A 40-Gb/s QSFP optoelectronic transceiver in a 0.13µm CMOS Silicon-on-insulator technology," in Proc. 2008 Opt. Fib. Conf. (OFC), Art. ID OMK7. [4] S. Sharif-Azadeh et al., "Low Vp Silicon photonics modulators with highly linear epitaxially grown phase shifters," Opt. Expr., vol. 23, no. 18, pp. 23526-23550, Sep. 2015. [5] Q. Xu, B. Schmidt, S. Pradhan, M. Lipson, "Micrometer-scale silicon photonics electro-optic modulator," Nat., vol. 435, pp. 325-327 (2005). [6] A. Moscoso-Mártir et al., "Co-integration of a temperature tolerant low impedance resonantly enhanced silicon photonics modulator," in Proc. 2017 Conf. Group IV Photon. (GFP), pp. 101-102. [7] B. J. Frey, D. B. Leviton, T. J. Madison, "Temperature dependent refractive index of silicon and germanium," arXiv:physics/0606168. [8] S. K. Selvaraja, W. Bogaerts, P. Dumon, D. Van Thourhout, R. Baets, "Subnanometer Linewidth Uniformity in Silicon Nanophotonics Waveguide Devices Using CMOS Fabrication Technology," J. Sel. Top. Quant. Electron., vol. 16, no. 1, pp. 316-324, Jan. 2010. [9] Y. Zheng et al., "Power-Efficient Calibration and Reconfiguration for Optical Network-on-Chip," J. Opt. Commun. Netw., vol. 4, no. 12, pp. 955-966, Dec. 2012. [10] P. Dong et al., "Thermally tunable silicon racetrack resonators with ultralow tuning power," Opt. Expr., vol. 18, no. 19, pp. 20298-20304, Sep. 2010. [11] V. Raghunathan, W. N. Ye, J. Hu, T. Izuhara, J. Michel, L. Kimerling, "Athermal operation of Silicon waveguides: spectral, second order and footprint dependencies," Opt. Expr., vol. 18, no. 17, pp. 17631-17639 (2010). [12] S. Romero-García, B. Marzban, F. Merget, B. Shen, J. Witzens, "Edge couplers with relaxed alignment tolerance for pick-and-place hybrid inte- gration of III -- V lasers with SOI waveguides," J. Sel. Topics Quantum Electron., vol. 20, no. 4, pp. 369 -- 379, Jul./Aug. 2014. [13] S. Romero-García, B. Shen, F. Merget, B. Marzban, J. Witzens, "Alignment Tolerant Couplers for Silicon Photonics," J. Sel. Top. Quant. Electron., vol. 21, no. 6, Art. ID 8200214, Nov. 2015. [14] S. Romero-García, B. Shen, F. Merget, J. Witzens, "Packaged MZIs passively balanced by means of Multimode Grating Couplers," Proc. 2016 Photon. North Conf. [15] S. Akiyama, T. Kurahashi, K. Morito, T. Yamamoto, T. Usuki, S. Nomura, "," Opt. Express, vol. 20, no. 15, pp. 16321-16338, Apr. 2012. [16] S. Romero-García, A. Moscoso-Mártir, S. Sharif-Azadeh, J. Müller, B. Shen, F. Merget, J. Witzens, "High-speed resonantly enhanced silicon photonics modulator with a large operating temperature," Opt. Lett., vol. 42, no. 1, pp. 81-84, Jan. 2017. [17] S. Sharif-Azadeh, F. Merget, M. P. Nezhad, J. Witzens, "On the measurement of the Pockels effect in strained silicon," Opt. Lett., vol. 40, no. 8, pp. 1877-1880, Apr. 2015. [18] Y. Shen, I. B. Divliansky, D. N. Basov, S. Mookherjea, "Perfect set-and- forget alignment of silicon photonic resonators and inteferometers," in Proc. 2011 Opt. Comm. Conf. (OFC), Art. ID PDPC3. [19] S. Grillanda et al., "Post-fabrication trimming of athermal silicon waveguides," Opt. Lett., vol. 38, no. 24, pp. 5450-5453 (2013). [20] S. Spector, J. M. Knecht, P. W. Juodawlkis, "Localized in situ cladding annealing for post-fabrication trimming of silicon photonic integrated circuits," Opt. Expr., vol. 24, no. 6, pp. 5996-6003, Mar. 2016. [21] A. P. Knights, Z. Wang, D. Paez, L. Dow, "Electrical trimming of the resonance of a silicon micro-ring resonator," in Proc. 2017 Conf. Group IV Photon., pp. 29-30. [22] S. Romero-García, A. Moscoso-Mártir, S. Sharif Azadeh, B. Shen, J. Nojic, F. Merget, J. Witzens, "Passively Biased Resonantly Enhanced Silicon Photonics Modulator with High Optical Bandwidth," Proc. SPIE, vol. 10108, Art. ID 1010802-1, Jan. 2017. [23] J. Müller, J. Hauck, B. Shen, S. Romero-García, E. Islamova, S. S. Azadeh, S. Joshi, N. Chimot, A. Moscoso-Mártir, F. Merget, F. Lelarge, J. Witzens, "Silicon photonics WDM transmitter with single section semiconductor mode-locked laser," Adv. Opt. Technol., vol. 4, pp. 119- 145, Mar. 2015. [24] D. T. Spencer, J. F. Bauters, M. J. Heck, and J. E. Bowers, "Integrated waveguide coupled Si3N4 resonators in the ultrahigh-Q regime," Optica, vol. 1, pp. 153-157, Sep. 2014. [25] D. Melati et al., "Wavelength and composition dependence of the thermo- optic coefficient for InGaAsP-based integrated waveguides," J. Appl. Phys., vol. 120, Art. ID 213102, 2016. [26] S. Tanaka, S.-H. Jeong, S. Sekiguchi, T. Kurahashi, Y. Tanaka, K. Morito, "High-output-power, single-wavelength silicon hybrid laser using precise flip-chip bonding technology," Opt. Expr., vol. 20, no. 27, pp. 28057- 28069, Dec. 2012.
1806.02854	1	1806	2018-06-07T18:36:31	On optical-absorption peaks in a nonhomogeneous thin-film solar cell with a two-dimensional periodically corrugated metallic backreflector	[ "physics.app-ph", "physics.optics" ]	The rigorous coupled wave approach (RCWA) was implemented to investigate optical absorption in a triple-p-i-n-junction amorphous-silicon solar cell with a 2D metallic periodically corrugated backreflector (PCBR). Both total and useful absorptances were computed against the free-space wavelength $\lambda_o$ for both s- and p-polarized polarization states. The useful absorptance in each of the three p-i-n junctions was also computed for normal as well as oblique incidence. Furthermore, two canonical boundary-value problems were solved for the prediction of guided-wave modes (GWMs): surface-plasmon-polariton waves and waveguide modes. Use of the doubly periodic PCBR enhanced both useful and total absorptances in comparison to a planar backreflector. The predicted GWMs were correlated with the peaks of the total and useful absorptances. The excitation of GWMs was mostly confined to $\lambda_o < 700$ nm and enhanced absorption. As excitation of certain GWMs could be correlated with the total absorptance but not with the useful absorptance, the useful absorptance should be studied while devising light-trapping strategies.	physics.app-ph	physics	On optical-absorption peaks in a nonhomogeneous thin-film solar cell with a two-dimensional Faiz Ahmad,a Tom H. Anderson,b Benjamin J. Civiletti,b Peter B. Monk,b and Akhlesh Lakhtakiaa,∗ periodically corrugated metallic backreflector aPennsylvania State University, Department of Engineering Science and Mechanics, NanoMM -- Nanoengineered Metamaterials Group, University Park, PA 16802, USA bUniversity of Delaware, Department of Mathematical Sciences, 501 Ewing Hall, Newark, DE 19716, USA ∗[email protected] Abstract The rigorous coupled wave approach (RCWA) was implemented to investigate optical absorption in a triple-p-i-n-junction amorphous-silicon solar cell with a 2D metallic periodically corrugated backreflector (PCBR). Both total and useful absorptances were computed against the free-space wavelength λ0 for both s- and p-polarized polarization states. The useful absorptance in each of the three p-i-n junctions was also computed for normal as well as oblique incidence. Furthermore, two canonical boundary-value problems were solved for the prediction of guided-wave modes (GWMs): surface-plasmon-polariton waves and waveguide modes. Use of the doubly periodic PCBR enhanced both useful and total absorptances in comparison to a planar backreflector. The predicted GWMs were correlated with the peaks of the total and useful absorptances. The excitation of GWMs was mostly confined to λ0 < 700 nm and enhanced absorption. As excitation of certain GWMs could be correlated with the total absorptance but not with the useful absorptance, the useful absorptance should be studied while devising light-trapping strategies. 1 Introduction Amorphous silicon (a-Si) thin-film solar cells provide a viable option to the 1st-generation crystalline-silicon (c-Si) solar cells [1], due to their ease of manufacturing and low cost. But the typical efficiency of a-Si thin-film solar cells is not as high as of c-Si solar cells, due to the high electron-hole recombination rate and low charge- carrier diffusion lengths in a-Si [2, 3]. Consequently, light-trapping techniques are necessary to enhance the efficiency of a-Si thin-film solar cells. Several light-trapping strategies have been studied both experimentally and theoretically[4, 5]. Anti-reflection coatings [6, 7, 8], textured front faces [9, 10], metallic periodically corrugated backreflectors (PCBRs) [11, 12, 13], particle plasmonics[14], surface plasmonics [15, 16, 17] and multiplasmonics [18, 19, 20], and waveguide-mode excitation [21, 22, 23] are attractive for trapping light in solar cells. Of particular interest is the enhancement of the optical electric field through the excitation of two types of guided-wave modes (GWMs): surface-plasmon-polariton (SPP) waves and waveguide modes (WGMs). The periodically corrugated interface of a metal and a semiconductor that is periodically nonhomogeneous in the thickness direction (identified by the z axis in Sec. 2) can guide multiple SPP waves at the same frequency[18, 24]. Any open-face waveguide with an air/semiconductor/metal architecture can guide WGMs [22, 23, 25]. Therefore, the incorporation of nonhomogeneity along the thickness direction in the semiconductor layers of a solar cell with a PCBR can enhance photonic absorption [18, 21, 26]. That enhancement would increase the generation rate of electron-hole pairs [27, 26]. Much of the theoretical and experimental research done on thin-film solar cells with metallic PCBRs is confined to devices with a homogeneous semiconductor layer and a metallic backreflector with one- dimensionally (1D) periodic corrugation. An experimental report of broadband excitation of multiple SPP waves in a device comprising a 1D photonic crystal (PC) atop a 1D PCBR [28] confirmed theoretical predic- tions [29] and spurred research on solar cells containing piecewise nonhomogeneous semiconductor layers and 1D PCBRs [18, 26, 27, 30]. In a recent study, experimental excitation of multiple SPP waves and WGMs were reported in a device comprising a 1D PC atop a 2D PCBR [21]. Appropriately designed 2D PCBRs were found to be better for the excitation of GWMs than 1D PCBRs, after the broadband excitation of GWMs predicted by solving two canonical boundary-value problems was correlated with the experimentally measured absorption spectrums. 1 In solar-cell research, often the excitation of GWMs is correlated with the total absorptance ¯Atot of the device [18, 23], which however is not a good measure of useful photonic absorption in a solar cell, as photons absorbed in the metallic portions of a solar cell are not available for conversion into electric current. Therefore, the chief objective for the work reported in this paper was to determine the spectrums of both the total absorptance ¯Atot and the useful absorptance ¯Asc [31] in a tandem solar cell with a 2D PCBR exposed to either normally or obliquely incident linearly polarized light. The solar cell was taken to comprise three p-i- n solar cells made of a-Si alloys [32] that can be fabricated using plasma-enhanced chemical-vapor deposition over planar and patterned substrates. A top layer of aluminum-doped zinc oxide (AZO) was incorporated to provide a transparent electrode. Also, an AZO layer was taken to be sandwiched between the 2D PCBR and the stack of nine semiconductor layers in order to avoid the deterioration of the electrical properties of the a-Si alloy closest to the metal [33], which was chosen to be silver [34]. The total absorptance and the useful absorptance calculated using the rigorous coupled-wave approach (RCWA) [35, 36, 24] were correlated against the predicted excitations of GWMs. The plan of this paper is as follows. Section 2 is divided into four parts. Section 2.1 presents the boundary-value problem that can be solved to determine the optical electromagnetic fields everywhere in a device comprising a stratified, isotropic dielectric material atop a 2D PCBR, when the device is illuminated by a plane wave. The formulations for useful and total absorptances are discussed in Sec. 2.2. Section 2.3 provides brief descriptions of the underlying canonical problems to predict the excitation of SPP waves and WGMs. Excitation of GWMs is discussed in Sec. 2.4. Section 3 is divided into two parts. The wavenumbers of the predicted GWMs are presented in Sec. 3.1. Correlations of the absorptances with the predicted GWMs are discussed in Sec. 3.2. The paper concludes with some remarks in Sec. 4. An exp (−iωt) dependence on time t is implicit, with ω denoting the angular frequency and i = √−1. µ0ε0, λ0 = 2π/k0, and η0 = (cid:112)µ0/ε0, respectively, with µ0 being the permeability and ε0 the The free-space wavenumber, the free-space wavelength, and the intrinsic impedance of free space are denoted by k0 = ω permittivity of free space. Vectors are underlined; the Cartesian unit vectors are identified as ux, uy, and uz; and column vectors as well as matrixes are in boldface. √ 2 Theory in Brief 2.1 Boundary-value problem for tandem solar cell Let us consider the boundary-value problem shown in Fig. 1 for a tandem solar cell containing three p- i-n junctions. The solar cell occupies the region X : {(x, y, z) − ∞ < x < ∞,−∞ < y < ∞, 0 < z < Lt}, with the half spaces z < 0 and z > Lt occupied by air. The reference unit cell is identified as R : {(x, y, z) − Lx/2 < x < Lx/2, −Ly/2 < y < Ly/2, 0 < z < Lt}, the backreflector being periodic along both the x and y axes. The region 0 < z < Ld = Lw + Ls + La is occupied by a cascade of homogeneous layers and is compactly characterized by the permittivity εd(z, λ0), which is a piecewise constant function of z. The top layer 0 < z < Lw and the bottom layer Lw + Ls < z < Ld are made of AZO with permittivity εw(λ0). The semiconductor layers in the region Lw < z < Lw + Ls are identified in Fig. 1(b). The region Ld + Lg < z < Ld + Lg + Lm is occupied by a metal with permittivity εm(λ0). The region Ld < z < Ld + Lg, henceforth termed the grating region, contains a periodically undulating surface with period Lx along the x axis and period Ly along the y axis. In the grating region, X possesses rectangular symmetry in the xy plane. The permittivity εg(x, y, z, λ0) in the grating region can be stated as εg(x, y, z, λ0) = εm(λ0) + [εw(λ0) − εm(λ0)]U[z − g1(x)]U[z − g2(y)] , x < ζxLx/2 , y < ζyLy/2 , z ∈ (Ld, Ld + Lg) , where the unit step function U(σ) = (cid:26) 0 , σ < 0, 1 , σ ≥ 0, 2 (1) (2) Figure 1: (a) Schematic of the tandem solar cell comprising three p-i-n junctions of a-Si alloys on a 2D PCBR. The wavevector of the incident plane wave is inclined at angle θ with respect to the z axis and angle ψ with respect to the x axis in the xy plane. (b) Nine semiconductors layers of the three p-i-n junctions. and ζx ∈ [0, 1] as well as ζy ∈ [0, 1] are the duty cycles. We chose the grating-shape functions and g1(x) = g2(y) = (cid:40) Ld + Lg[1 − cos(2π πx (cid:40) Ld + Lg[1 − cos(2π πy Ld + Lg , ζxLx ζyLy Ld + Lg , )] , x ∈ [− ζxLx x /∈ [− ζxLx 2 , ζxLx 2 ], 2 , ζxLx 2 ], )] , y ∈ [− ζyLy y /∈ [− ζyLy 2 , ζyζy 2 ], 2 , ζyLy 2 ], (3) (4) to represent hillocks for all data reported in this paper. The grating-shape functions chosen here are only for illustration, many other choices fit for experimental study being also available [13]. Suppose that an arbitrarily polarized plane wave, propagating in the half space z < 0 at an angle θ ∈ [0◦, 90◦) with respect to the z axis and an angle ψ ∈ [0◦, 360◦) with respect to the x axis in the xy plane, is incident on the plane z = 0. The electric field phasor of this plane wave can be stated as (cid:16) (cid:17) (cid:104) (cid:16) (cid:17) • r (cid:105) Einc(r) = ¯as s(0,0) + ¯ap p(0,0) + exp i κ(0,0) + α(0,0) 0 uz , (5) where ¯as and ¯ap are the known coefficients of s- and p-polarized components, respectively. Here and hereafter, 3 the following quantities are used: 0 x = + y uy ux + k(n) κ(m,n) • κ(m,n) 0 − κ(m,n) • κ(m,n) κ(m,n) = k(m) k(m) x = k0 sin θ cos ψ + m(2π/Lx) (cid:112) k(n) y = k0 sin θ sin ψ + n(2π/Ly) = +(cid:112)k2 k(m,n) xy α(m,n) s(m,n) = − k(n) (cid:32) k(m,n) xy k(m) x k(m,n) xy k(m) x k(m,n) xy k(m) x k(m,n) xy k(n) y k(m,n) xy k(n) y k(m,n) xy p(m,n) + p(m,n)− = − (cid:32) ux + ux + ux + uy uy = y uy (cid:33) (cid:33) α(m,n) 0 k0 α(m,n) 0 k0 + uz + k(m,n) xy k0 k(m,n) xy k0 uz  , m ∈ Z , n ∈ Z . (6) As a result of the metallic PCBR being doubly periodic, the x- and y-dependences of the electric and magnetic field phasors are represented everywhere as an infinite series of Floquet harmonics as [35, 36, 24] (cid:88) (cid:88) m∈Z (cid:88) (cid:88) n∈Z m∈Z n∈Z (cid:16) (cid:16) E(x, y, z) = H(x, y, z) = e(m,n)(z) exp iκ(m,n) • r h(m,n)(z) exp iκ(m,n) • r where e(m,n)(z) = e(m,n) x h(m,n)(z) = h(m,n) (z)ux + e(m,n) (z)ux + h(m,n) (z)uy + e(m,n) (z)uy + h(m,n) y y z z (z)uz (z)uz are expansion coefficients. Accordingly, the incident and the reflected electric field phasors are represented as Einc(x, y, z) = a(m,n) s s(m,n) + a(m,n) p p(m,n) + (7) (8) (cid:17) (cid:17) (cid:41)  ,  , (cid:17) (cid:17) (cid:88) (cid:88) x (cid:88) (cid:88) (cid:110)(cid:16) (cid:110)(cid:16) 0 n∈Z m∈Z κ(m,n) + α(m,n) uz r(m,n) s m∈Z κ(m,n) − α(m,n) n∈Z uz 0 (cid:17) • r (cid:17) • r (cid:105)(cid:111) (cid:105)(cid:111) × exp × exp i (cid:104) (cid:16) (cid:104) (cid:16) (cid:88) i m∈Z (cid:110)(cid:16) (cid:104) (cid:16) (cid:88) n∈Z × exp (cid:88) s (cid:88) m∈Z n∈Z Eref (x, y, z) = s(m,n) + r(m,n) p p(m,n)− z < 0 , (9) and the transmitted electric field phasor as Etr(x, y, z) = t(m,n) s s(m,n) + t(m,n) p (cid:17) (cid:17) • (r − Ltuz) p(m,n) + (cid:105)(cid:111) , z > Lt , (10) i κ(m,n) + α(m,n) 0 uz where the coefficients a(m,n) p the Kronecker delta, but the coefficients r(m,n) , r(m,n) be determined. Finally, the permittivity ε(x, y, z) everywhere is represented by the Fourier series = ¯apδm0δn0 in Eq. (9)1 are known with δmm(cid:48) denoting in Eq. (9)2 and Eq. (10) have to = ¯asδm0δn0 and a(m,n) , and t(m,n) , t(m,n) p s s ε(x, y, z) = ε(m,n)(z) exp i , (11) (cid:104) (cid:16) p κ(m,n) − κ(0,0)(cid:17) • r (cid:105) 4 where ε(m,n)(z) are Fourier coefficients. m ∈ {−Mt, ..., Mt} and n ∈ {−Nt, ..., Nt}, with Mt ≥ 0 and Nt ≥ 0. Furthermore, a superindex Computational tractability requires the expansions in Eqs. (7) -- (11) to be truncated to include only τ = m(2Nt + 1) + n , m ∈ [−Mt, Mt] , n ∈ [−Nt, Nt] , (12) is defined for convenience. Then, τ ∈ [−τt, τt], where τt = 2MtNt + Mt + Nt. Also, both the mapping from (m, n) to τ and the inverse mapping from τ to (m, n) are injective [37]. Thereafter, column vectors eσ(z) = hσ(z) = σ e(−τt) h(−τt) (z), e(−τt+1) (z), h(−τt+1) (z), ..., e(τt−1) (z), ..., h(τt−1) σ σ σ σ (z), e(τt) σ (z) σ (z), h(τt) σ (z) σ ∈ {x, y, z}, (13) (cid:105)T (cid:105)T  , of length 2τt + 1 are set up, the superscript T denoting the transpose. The Toeplitz matrix [38] ε(−τt,−τt)(z) ε(−τt+1,−τt)(z) ε(τt−1,−τt)(z) ε(τt,−τt)(z) ··· ε(−τt,−τt+1)(z) ε(−τt+1,−τt+1)(z) ε(τt−1,−τt+1)(z) ε(τt,−τt+1)(z) ··· ··· ··· ··· ··· ··· ε(−τt,τt−1)(z) ε(−τt+1,τt−1)(z) ε(τt−1,τt−1)(z) ε(τt,τt−1)(z) ··· ε(−τt,τt)(z) ε(−τt+1,τt)(z) ε(τt−1,τt)(z) ε(τt,τt)(z) ···  . (cid:104) (cid:104)  ε(z) = contains the Fourier coefficients appearing in Eq. (8) with ε(τ,τ(cid:48))(z) = ε(m−m(cid:48),n−n(cid:48))(z). Finally, the (2τt + 1) × (2τt + 1) Fourier-wavenumber matrixes (cid:104)k(−τt) (cid:104)k(−τt) x y Kx = diag Ky = diag x , k(−τt+1) , k(−τt+1) y , ..., k(τt−1) , ..., k(τt−1) x y , k(τt) x , k(τt) y (cid:105) (cid:105)  are set up with k(τ ) x = k(m) x and k(τ ) y = k(n) y . The frequency-domain Maxwell curl postulates yield the matrix ordinary differential equation [24] where the 4(2τt + 1)-column vector d dz f (z) = i P(z) • f (z),  ex(z) ey(z) hx(z) hy(z)  f (z) =  0 0 0 0 0 − ε(z) ε(z) 0  I I µ0 0 0 0 0 −µ0 0 0 −1 • Ky − Kx −1 • Ky − Ky (14) (15) (16) (17) (18) and the 4(2τt + 1) × 4(2τt + 1) matrix P(z) = ω  + 1 ω 0 0 Kx Ky −µ−1 −µ−1 0 0 • Ky µ−1 • Ky µ−1 0 0 0 0 Kx Ky • Kx • Kx Kx Ky 5 • [ ε(z)] • [ ε(z)] 0 0  • [ ε(z)] • [ ε(z)] −1 • Kx −1 • Kx 0 0 contains 0 as the (2τt + 1) × (2τt + 1) null matrix and I as the (2τt + 1) × (2τt + 1) identity matrix. In order to solve Eq. (16), the region R is partitioned into a sufficiently large number of thin slices along the z direction [24]. Each slice is taken to be homogeneous along the z axis but it is either homogeneous or periodically nonhomogeneous along the x and y axes; thus, P(z) is assumed to be uniform in each slice. Boundary conditions are enforced on the planes z = 0 and z = Lt to match the fields to the incident, reflected, and transmitted waves, as appropriate. A stable numerical marching algorithm is then used to determine the Fourier coefficients of the electric and magnetic field phasors in each slice[24]. Finally, the z components of the electric and magnetic field phasors in the device can be obtained through ez(z) = − [ωε(z)] • hy(z)− Ky • ex(z)]. Thus, the electric field phasor can be determined everywhere. The entire procedure was implemented on the Mathematica R(cid:13) platform. • hx(z)] and hz(z) = (ωµ0)−1[ Kx • ey(z)− Ky −1 • [ Kx 2.2 Total and useful absorptances At any location inside the device, the absorption rate of the monochromatic optical energy per unit volume is given by Q(x, y, z) = ω Im{ε(x, y, z)}E(x, y, z)2 . 1 2 The useful absorptance [39] ¯Asc = LxLy (cid:16)¯as2 + ¯ap2(cid:17) 2η0 cos θ Rsc Q(x, y, z) dx dy dz is calculated by integrating Q(x, y, z) over the region Rsc ⊂ R occupied by the semiconductor layers. Like- wise, absorptance in the metal is given by (cid:16)¯as2 + ¯ap2(cid:17) 2η0 ¯Amet = LxLy cos θ Rmet Q(x, y, z) dx dy dz , where Rmet ⊂ R is the region occupied by the metal. The total absorptance is then the sum ¯Atot = ¯Asc + ¯Amet , if εw is purely real. Four reflection and four transmission coefficients of order (m, n) are defined as the elements in the 2×2 matrices appearing in the following relations [24]: (cid:35) (cid:34)r(m,n) s r(m,n) p (cid:34)r(m,n) ss r(m,n) ps = (cid:35) • (cid:35) (cid:34)¯as ¯ap , r(m,n) sp r(m,n) pp (cid:35) (cid:34)t(m,n) s t(m,n) p (cid:34)t(m,n) ss t(m,n) ps = (cid:35) • (cid:35) (cid:34)¯as ¯ap t(m,n) sp t(m,n) pp . (23) Coefficients of order (0, 0) are classified as specular, whereas coefficients of all other orders are nonspecular. Four reflectances and four linear transmittances of order (m, n) are defined as (19) (20) (21) (22) (cid:90)(cid:90)(cid:90) (cid:90)(cid:90)(cid:90) (cid:104) (cid:105) (cid:12)(cid:12)(cid:12)r(m,n) sp (cid:12)(cid:12)(cid:12)2 ∈ [0, 1] , Re R(m,n) sp = α(m,n) 0 α(0,0) 0 etc., and two absorptances as As = 1 − m=Mt(cid:88) Ap = 1 − m=Mt(cid:88) m=−Mt m=−Mt (cid:16) (cid:16) n=Nt(cid:88) n=Nt(cid:88) n=−Nt n=−Nt R(m,n) ss + R(m,n) ps + T (m,n) ss + T (m,n) ps R(m,n) pp + R(m,n) sp + T (m,n) pp 6 + T (m,n) sp (24) (25)  . (cid:17) ∈ [0, 1] (cid:17) ∈ [0, 1] These are total absorptances in that they contain the contributions of the semiconductors and the metal in the solar cell. Whereas ¯Atot, ¯Asc, and ¯Amet are defined for incident light of arbitrary polarization state, As is defined for incident s-polarized light and Ap for incident p-polarized light. All absorptances presented in Sec. 3 were calculated for a solar cell comprising just one triple p-i-n junction, as shown in Fig. 1. 2.3 Canonical boundary-value problems Two separate canonical boundary-value problems were solved to correlate peaks in the spectrums of various absorptances with the excitation of SPP waves and WGMs. Details on both canonical problems are available elsewhere [39] for the interested reader, but we note the following salient features of both canonical problems. 2.3.1 SPP waves The complex-valued wavenumbers q (cid:54)= 0 of SPP waves for a specific value of λ0 were obtained by solving a canonical boundary-value problem [24, 29], with the assumptions that the backreflector metal occupies the half space z < 0, a periodically semi-infinite cascade of three p-i-n junctions occupies the half space z > 0, and there are no AZO layers. 2.3.2 Waveguide modes An open-faced waveguide is formed by the three p-i-n junctions interposed between two half spaces, one oc- cupied by air and the other by the backreflector metal of thickness considerably exceeding the skin depth [40]. For a specific value of λ0, this waveguide can support the propagation of multiple WGMs (with wavenum- bers q (cid:54)= 0) which can play significant light-trapping roles [21, 22, 23]. We ignored the AZO layers for this canonical problem as well. 2.4 Excitation of SPP waves and WGMs Planewave illumination will excite a GWM of wavenumber q as a Floquet harmonic of order (m, n), provided that[24] ± Re [q] (cid:39) k(m,n) xy . (26) When Lx = Ly = L, the right side of Eq. (26) simplifies to yield ± Re [q/k0] (cid:39)(cid:110) [sin θ + (m cos ψ + n sin ψ)(λ0/L)]2 + [(m sin ψ − n cos ψ)(λ0/L)]2(cid:111) 1 2 . (27) Since the thickness Ld is finite, shifts in the predictions of θ for specific values of λ0 and ψ are possible for SPP waves. Also, shifts are possible for both SPP waves and WGMs, because both canonical problems were formulated and solved with Lw = La = 0. Finally, shifts can also be due to Lg (cid:54)= 0 [41]. Therefore, for all absorptance spectrums presented in this paper, we accepted predictions of θ from Eq. (27) with ±1◦ tolerance. However, let us note that not every possible GWM is strongly excited by planewave illumination. Finally, it is important to note that depolarization can occur because the PCBR is doubly periodic. Accordingly, illumination by a linearly polarized plane wave for a specific value of ψ can excite a GWM of a different polarization state propagating in a direction specified by the angle ϕ that may differ from ψ [21, 42]. 3 Numerical results and discussion All optical and geometric parameters were chosen only to illustrate the relationships of the WGMs to total and useful absorptances, but still are representative of actual tandem solar cells [30]. The compositions, bangaps, and thicknesses of the nine hydrogenated a-Si alloys for the nine semiconductor layers are presented in Table I. 7 The permittivity of each alloy was calculated as a function of λ0, using a model provided by Ferlauto et al. [18, 32]. The spectrums of all nine permittivities, normalized by ε0, are plotted in Fig. 2. The 2D PCBR was taken to be made of silver [34]. The refractive index of AZO was taken as a function of λ0 from Gao et al. [43]. Table I: Compositions, bandgaps, and thicknesses of hydrogenated a-Si alloys used for the nine semiconductor layers in the triple-p-i-n-junction tandem solar cell. Layer 1p 1i 1n 2p 2i 2n 3p 3i 3n Composition a-Si1−uCu:H a-Si:H a-Si:H a-Si1−uCu:H a-Si1−uGeu:H a-Si:H a-Si:H a-Si1−uGeu:H a-Si:H Bandgap (eV) 1.95 1.8 1.8 1.95 1.58 1.8 1.8 1.39 1.8 Thickness (nm) 20 200 20 20 200 20 20 200 20 Figure 2: Spectrums of the relative permittivity ε/ε0 of the different semiconductor alloys used in the triple-p-i-n-junction tandem solar cell. The following dimensions were chosen: Lw = 100 nm, La = 60 nm, Lg = 80 nm, Lm = 30 nm, Lx = Ly = 400 nm, and ζx = ζy = 1. We used Mt = Nt accordingly. Furthermore, we used Mt ≤ 12, which ensured the convergence of all non-zero reflectances and absorptances to within ±1% for every λ0 ∈ {500, 502, ..., 898, 900} nm. Here, convergence was defined to have occurred when there was a difference not exceeding 1% in magnitude between the results for Mt = N − 1 and Mt = N . Higher values of Mt were found to be necessary for higher λ0 as the chosen semiconductor alloys are then less absorbing and the effect of grating is more pronounced. 3.1 Prediction of GWM wavenumbers The real parts of the normalized wavenumbers q/k0 of SPP waves are presented in Fig 3 as functions of λ0 ∈ {500, 501, ..., 899, 900} nm. These wavenumbers are organized into three branches (labeled s1 -- s3) for s-polarized SPP waves and seven branches (labeled p1 -- p7) for p-polarized SPP waves. The real parts of the normalized wavenumbers q/k0 of the WGMs are presented in Fig 4. These wavenumbers are arranged into six branches for both s- and p-polarized WGMs labeled s1 -- s6 and p1 -- p6, respectively. 8 Figure 3: Real parts of the normalized wavenumbers q/k0 of s- and p-polarized SPP waves obtained after solving the relevant canonical boundary-value problem. Figure 4: Real parts of the normalized wavenumbers q/k0 of s- and p-polarized WGMs obtained after solving the relevant canonical boundary-value problem. 3.2 Absorptances and Correlation with Predictions Calculations of As and Ap as functions of λ0 ∈ [500, 900] nm were made for the chosen triple-p-i-n-junction tandem solar cell with a 2D PCBR. In addition, we computed the spectrums of the useful absorptances s = ¯Asc(cid:12)(cid:12)(cid:12)¯ap=0 p = ¯Asc(cid:12)(cid:12)(cid:12)¯as=0 ¯Asc ¯Asc  . (28) s , and ¯Asc The spectrums of As, Ap, ¯Asc p for λ0 ∈ [500, 900] nm for solar cells with and without corrugations (Lg = 80 nm and Lg = 0, respectively) were examined for several combinations of θ and ψ [39]. For the sake of illustration, data are presented in Figs. 5 -- 8 only for the following two directions of incidence: (1) {ψ = 1◦, θ = 1◦}, and (2) {ψ = 45◦, θ = 15◦}. The choices of 1◦ instead of 0◦ for the incidence angles help avoid spurious results associated with the computation of distinct eigenvalues of P(z) when the RCWA is implemented. Also shown in these figures in the mth p-i-n junction, m ∈ {1, 2, 3}, for are the spectrums of the useful absorptances ¯Ascm incident s- and p-polarized plane waves, respectively. The excitation of a GWM is marked by an absorptance peak. Therefore, values of λ0 ∈ [500, 900] nm for which the solutions of the two canonical problems (with the assumption that Lw = La = 0) predicted the excitation of SPP waves and WGMs for θ ∈ [0◦, 2◦] ∪ [14◦, 16◦] are also identified in Figs. 5 -- 8. Red arrows indicate the excitation of SPP waves that matched with both total absorptances (As and Ap) and useful and ¯Ascm p s 9 s and ¯Asc absorptances ( ¯Asc p ); black arrows indicate WGMs that matched with both total absorptances and useful absorptances; blue arrows indicate the excitation of SPP waves that correlated with total absorptances but not with useful absorptances; and purple arrows indicate the excitation of WGMs that correlated with total absorptances but not with useful absorptances. 3.2.1 Case 1: {ψ = 1◦, θ = 1◦} p for {ψ = 1◦, θ = 1◦} calculated with Lg = 80 nm are presented in Fig. 5. Spectrums of As, Ap, ¯Asc s and ¯Asc Also, spectrums of the same quantities calculated with Lg = 0 are presented in Fig. 6 for comparison. Tables II and III contain values of λ0 ∈ [500, 900] nm for which the excitation of either an SPP wave or a WGM as a Floquet harmonic of order (m, n) is predicted. s -peak at λ0 ≈ 728 nm in Fig. 5 occurs close to the wavelength λ0 ≈ 730 nm predicted for the excitation of an s-polarized SPP wave as a Floquet harmonic of order (1, 0) at θ = 0.856◦ in Table II. This is the only SPP wave that correlated with peaks of both ¯Asc The ¯Asc s and As. The As-peak in Fig. 5 at • λ0 ≈ 845 nm is due to the excitation of an s-polarized SPP wave as a Floquet harmonic of order (1, 1) predicted at θ = 0.088◦ in Table II, • λ0 ≈ 897 nm matches well with the excitation of a p-polarized SPP wave as a Floquet harmonic of order (1, 0) predicted at θ = 0.847◦ in Table II, • λ0 ≈ 754 nm is related with the excitation of a p-polarized WGM as a Floquet harmonic of order (−2, 0) predicted at θ = 1.145◦ in Table III, and • λ0 ≈ 892 nm is represents the excitation of a p-polarized WGM as a Floquet harmonic of order (−1, 1) predicted at θ = 1.198◦ in Table III, s , which indicates that not every ¯Asc Excitation of these GWMs correlated only with the total absorptance As but not with the useful absorptance ¯Asc s -peak can be matched to an As-peak that is correlated with the excitation of a GWM [21, 44]. Accordingly, useful absorptance, not the overall absorptance, needs to be studied for solar cells. Contributions to the overall absorptance are made both by the semiconductor layers and the metallic PCBR, but the contribution of the latter is useless for harvesting solar energy. p -peak in Fig. 5 at λ0 ≈ 786 nm is due to the excitation of an s-polarized WGM as a Floquet harmonic of order (−1, 0) predicted at θ = 0.948◦ in Table III. The Ap-peak at The ¯Asc • λ0 ≈ 834 nm is due to the excitation of a p-polarized WGM as a Floquet harmonic of order (−1, 1) predicted at θ = 0.838◦ in Table III, and • λ0 ≈ 845 nm matches well with the excitation of a p-polarized SPP wave as a Floquet harmonic of order (1, 1) predicted at θ = 0.088◦ in Table II, s and ¯Asc3 On comparing Figs. 5 and 6, we note that the GWMs are excited at λ0 > 700 nm. Also, the total absorptance for Lg = 80 nm exceeds that for Lg = 0 in the same spectral regime. This increase is largely due to the increases in ¯Asc3 p , i.e., in the p-i-n junction closest to the PCBR. Furthermore, increases in both total and useful absorptances for λ0 ∈ [634, 680] nm, regardless of the polarization state of the incident light, were observed with the use of the PCBR rather than a planar backreflector. In addition, depolarization due to the two-dimensional periodicity of the PCBR is evident from the excitation of WGMs that are not of the same polarization state as the incident light. 3.2.2 Case 2: {ψ = 45◦, θ = 15◦} Calculated spectrums of As, Ap, ¯Asc Fig 7. Also, the spectrums As, Ap, ¯Asc p for {ψ = 45◦, θ = 15◦} with Lg = 80 nm are presented in p calculated with Lg = 0 for the same incident direction are s , and ¯Asc s , and ¯Asc 10 Table II: Values of λ0 ∈ [500, 900] nm (calculated at 1-nm intervals) for which the excitation of an SPP wave as a Floquet harmonic of order (m, n) is predicted for θ ∈ [0◦, 2◦] and ψ = 1◦, for the tandem solar cell with a 2D PCBR. The SPP waves strongly excited in Fig. 5 are highlighted in bold. Pol. State s s p p p p λ0 (nm) Re{q/k0} 730 845 570 640 680 897 1.816 2.988 2.027 2.029 1.678 2.988 θ◦ 0.856 0.088 0.991 1.024 1.228 0.847 (m, n) (1, 0) (1, 1) (1, 1) (±1, 0) (1, 0) (1, 0) Table III: Same as Table II, except that the relevant excitations of WGMs are indicated. The WGMs strongly excited in Fig. 5 are highlighted in bold. Pol. State s s s s s p p p p λ0 (nm) Re{q/k0} 663 711 786 834 898 754 797 827 892 3.685 3.536 1.948 2.938 3.163 3.750 3.963 2.910 3.138 θ◦ 1.334 1.076 0.948 0.838 0.902 1.145 1.206 1.121 1.198 (m, n) (−2, 1) (−2, 0) (−1, 0) (−1, 1) (−1, 1) (−2, 0) (−2, 0) (−1, 1) (−1, 1) presented in Fig. 8. Tables IV and V contain values of λ0 ∈ [500, 900] nm for which the excitation of either an SPP wave or a WGM as a Floquet harmonic of order (m, n) is predicted from analysis of Figs. 3 and 4.. s -peak could be correlated with the excitation of a GWM. The As-peak at No ¯Asc • λ0 ≈ 800 nm is related to the excitation of a p-polarized SPP wave as a Floquet harmonic of order either (1, 0) or (0, 1) predicted at θ = 14.952◦ in Table IV, • λ0 ≈ 870 nm is due to the excitation of a p-polarized SPP wave as a Floquet harmonic of order either (−1, 0) or (0,−1) predicted at θ = 15.880◦ in Table IV, • λ0 ≈ 764 nm is associated with the excitation of a p-polarized WGM as a Floquet harmonic of order (−1,−2) predicted at θ = 15.141◦ in Table V. and • λ0 ≈ 778 nm arises due to the excitation of a p-polarized WGM as a Floquet harmonic of order either (−2, 0) or (0,−2) predicted at θ = 15.411◦ in Table V. p -peak at λ0 ≈ 647 nm is related with the excitation of an s-polarized SPP wave as a Floquet harmonic The ¯Asc of order (1, 1) at θ = 15.529◦ in Table IV. No other ¯Asc p - or Ap-peak was found to be correlated with GWM excitation. These results underscore the fact that useful absorptance is not necessarily enhanced by the excitation of a GWM. However, there are useful- and total-absorptance peaks which could not predicted by the canonical boundary-value problems. On comparing Figs. 7 and 8, increases in both total and useful absorptances for λ0 ∈ [640, 670] nm, regardless of the polarization state of the incident light, become evident with the use of the PCBR rather 11 , ¯Asc2 s s s , ¯Asc1 s Figure 5: Spectrums of (left) As, ¯Asc of the triple-p-i-n-junction tandem solar cell, when ψ = 1◦ and θ = 1◦. Red arrows indicate the excitation of SPP waves that matched with both total absorptances (As and Ap) and useful absorptances ( ¯Asc s and ¯Asc p ); black arrows indicate WGMs that matched with both total absorptances and useful absorptances; blue arrows indicate the excitation of SPP waves that correlated with total absorptances but not with useful absorptances; and purple arrows indicate the excitation of WGMs that correlated with total absorptances but not with useful absorptances. and (right) Ap, ¯Asc p , and ¯Asc3 , and ¯Asc3 p , ¯Asc1 p , ¯Asc2 p Figure 6: Same as Fig. 5 except that Lg = 0. than a planar backreflector. In comparison to normal illumination (Sec. 3.2.1) for which GWMs were excited only for λ0 > 700 nm, an SPP wave is excited at λ0 = 647 nm for oblique illumination. This is in accord with the blueshifts of SPP waves expected for oblique illumination [18, 42] as well as with the angular trends in Figs. 3 and 4. Apart from the SPP wave excited at λ0 = 647 nm, all other GWMs are excited at λ0 > 700 nm. The polarization state of an excited GWM may not be the same as that of the incident light, because the 2D PCBR is a depolarizing agent. Finally, the total absorptance increases in the same spectral regime with the use of the PCBR in comparison to a planar backreflector, for either polarization state of the incident light, largely due to the increases in ¯Asc3 p , i.e., in the p-i-n junction closest to the PCBR. and ¯Asc3 s 4 Concluding remarks The effect of a 2D PCBR on the absorptance of light in a triple-p-i-n-junction thin-film solar cell was studied using the RCWA. Total absorptances and useful absorptances for incident s- and p-polarized light were computed against the free-space wavelength for two different incidence directions. Calculations were also made of the useful absorptance in each of the three p-i-n junctions. Furthermore, two canonical boundary- value problems were solved for the prediction of GWMs. The predicted GWMs were correlated with the peaks of the total and useful absorptances for both linear polarization states. Numerical studies led to the following conclusions: 12 Table IV: Values of λ0 ∈ [500, 900] nm (calculated at 1-nm intervals) for which the excitation of an SPP wave as a Floquet harmonic of order (m, n) is predicted for θ ∈ [0◦, 2◦] and ψ = 45◦, for the tandem solar cell backed by a 2D PCBR. The SPP waves strongly excited in Fig. 7 are highlighted in bold. Pol. State s s p p p p p λ0 (nm) Re{q/k0} 647 690 565 620 750 800 870 2.581 2.193 2.272 1.734 2.404 2.190 2.376 θ◦ (m, n) 15.529 (1, 1) (−1,−1) 14.231 (1, 1) 15.955 14.365 (1, 0), (0, 1) (−1,−1) 14.315 14.952 (1, 0), (0, 1) 15.880 (−1, 0), (0,−1) Table V: Same as Table IV, except that the relevant excitations of WGMs are indicated. The WGMs strongly excited in Fig. 7 are highlighted in bold. Pol. State s s p p p p p λ0 (nm) Re{q/k0} 732 807 764 778 664 732 821 3.484 3.858 4.023 3.706 3.641 3.477 3.925 θ◦ (m, n) (−2, 0), (0,−2) 15.004 (−2, 0), (0,−2) 14.931 15.141 (−1,−2) 15.411 (−2, 0), (0,−2) (−2, 1) 14.287 (−2, 0), (0,−2) 15.355 (−2, 0), (0,−2) 15.098 • Regardless of the illumination direction and the polarization state of the incident light, increases in useful and total absorptances for λ0 < 700 nm arise from the replacement of a planar backreflector by a 2D PCBR. • The triple-p-i-n-junction tandem solar cell made of a-Si alloys is highly absorbing for λ0 < 700 nm, so that the excitation of SPP waves in this regime is unnoticeable. • Both SPP waves and WGMs are excited for λ0 > 700 nm for both normal and oblique illumination. An SPP wave excited at λ0 = 647 nm for oblique illumination is in accord with blueshifting of SPP waves with increasing obliqueness of illumination. • Some of the excited GWMs directly contribute to the increase in useful absorptance of the solar cell backed by a 2D PCBR. This increase is largely due to enhanced absorptance in the p-i-n junction closest to the 2D PCBR. • Depolarization due to the two-dimensional periodicity of the PCBR is evident from the excitation of GWMs that are not of the same polarization state as the incident light. • Excitation of certain GWMs could be correlated with the total absorptance but not with the useful absorptance. When devising light-trapping strategies, the useful, but not the total absorptance needs to be focused on. While reduction of reflectance is a worthwhile objective, meeting it will not necessarily boost the useful absorptance. We conclude with the recommendation that material and geometric parameters need to be optimized for efficiency enhancement. 13 Figure 7: Same as Fig. 5, except that θ = 15◦ and ψ = 45◦ . Figure 8: Same as Fig. 7, except that Lg = 0. Note. This paper is substantially based on a paper titled, "On optical-absorption peaks in a nonhomoge- neous dielectric material over a two-dimensional metallic surface-relief grating," presented at the SPIE Optics and Photonics conference Nanostructured Thin Films X, held August 5 -- 11, 2017 in San Diego, California, United States. Acknowledgments. F. Ahmad thanks the Graduate School and the College of Engineering, Pennsylvania State University, for a University Graduate Fellowship during the first year of his doctoral studies. A. Lakhtakia thanks the Charles Godfrey Binder Endowment at the Pennsylvania State University for ongoing support of his research. The research of F. Ahmed and A. Lakhtakia is partially supported by US National Science Foundation (NSF) under grant number DMS-1619901. The research of T.H. Anderson, B.J. Civiletti, and P.B. Monk is partially supported by the US National Science Foundation (NSF) under grant number DMS-1619904. References [1] R. Singh, "Why silicon is and will remain the dominant photovoltaic material," J. Nanophoton. 3(1), 032503 (2009). [2] D. E. Carlson and C. R. Wronski, "Amorphous silicon solar cell," Appl. Phys. Lett. 28(11), 671 -- 673 (1976). [3] M. A. Green, "Thin-film solar cells: review of materials, technologies and commercial status," J. Mater. Sci: Mater. Electron. 18(S1), 15 -- 19 (2007). [4] R. Singh, G. F. Alapatt, and A. Lakhtakia, "Making solar cells a reality in every home: Opportunities and challenges for photovoltaic device design," IEEE J. Electron. Dev. Soc. 1(6), 129 -- 144 (2013). 14 [5] R. J. Mart´ın-Palma and A. Lakhtakia, "Progress on bioinspired, biomimetic, and bioreplication routes to harvest solar energy," Appl. Phys. Rev. 4(2), 021103 (2017). [6] I. G. Kavakli and K. Kantarli, "Single and double-layer antireflection coatings on silicon," Turk. J. Phys. 26(5), 349 -- 354 (2002). [7] S. K. Dhungel, J. Yoo, K. Kim, S. Jung, S. Ghosh, and J. Yi, "Double-layer antireflection coating of MgF2/SiNx for crystalline silicon solar cells," J. Korean Phys. Soc. 49(3), 885 -- 889 (2006). [8] S. A. Boden and D. M. Bagnall, "Sunrise to sunset optimization of thin film antireflective coatings for encapsulated, planar silicon solar cells," Prog. Photovoltaics Res. Appl. 17(4), 241 -- 252 (2009). [9] W. H. Southwell, "Pyramid-array surface-relief structures producing antireflection index matching on optical surfaces," J. Opt. Soc. Am. A 8(3), 549 -- 553 (1991). [10] K. C. Sahoo, M.-K. Lin, E.-Y. Chang, T. B. Tinh, Y. Li, and J.-H. Huang, "Silicon nitride nanopillars and nanocones formed by nickel nanoclusters and inductively coupled plasma etching for solar cell application," Jpn. J. Appl. Phys. 48(12), 126508 (2009). [11] P. Sheng, A. N. Bloch, and R. S. Stepleman, "Wavelength-selective absorption enhancement in thin-film solar cells," Appl. Phys. Lett. 43(6), 579 -- 581 (1983). [12] C. Heine and R. H. Morf, "Submicrometer gratings for solar energy applications," Appl. Opt. 34(14), 2476 -- 2482 (1995). [13] M. Solano, M. Faryad, A. S. Hall, T. E. Mallouk, P. B. Monk, and A. Lakhtakia, "Optimization of the absorption efficiency of an amorphous-silicon thin-film tandem solar cell backed by a metallic surface- relief grating," Appl. Opt. 52(5), 966 -- 979 (2013); erratum: 54(3), 398 -- 399 (2015). [14] Y. Zhang, B. Jia, Z. Ouyang, and M. Gu, "Influence of rear located silver nanoparticle induced light losses on the light trapping of silicon wafer-based solar cells," J. Appl. Phys. 116(12), 124303 (2014). [15] L. M. Anderson, "Parallel-processing with surface plasmons, A new strategy for converting the broad solar spectrum," Proc. 16th IEEE Photovoltaic Specialist Conf. 1(1), 371 -- 377 (1982). [16] L. M. Anderson, "Harnessing surface plasmons for solar energy conversion," Proc. SPIE 408(1), 172 -- 178 (1983). [17] M. G. Deceglie, V. E. Ferry, A. P. Alivisatos, and H. A. Atwater, "Design of nanostructured solar cells using coupled optical and electrical modeling," Nano Lett. 12(6), 2894 -- 2900 (2012). [18] M. Faryad and A. Lakhtakia, "Enhancement of light absorption efficiency of amorphous-silicon thin-film tandem solar cell due to multiple surface-plasmon-polariton waves in the near-infrared spectral regime," Opt. Eng. 52(8), 087106 (2013); errata: 53(12), 129801 (2014). [19] M. E. Solano, G. D. Barber, A. Lakhtakia, M. Faryad, P. B. Monk, and T. E. Mallouk, "Buffer layer between a planar optical concentrator and a solar cell," AIP Advances 5(9), 097150 (2015). [20] L. Liu, G. D. Barber, M. V. Shuba, Y. Yuwen, A. Lakhtakia, T. E. Mallouk, and T. S. Mayer, "Planar light concentration in micro-Si solar cells enabled by a metallic grating -- photonic crystal architecture," ACS Photon. 3(4), 604 -- 610 (2016). [21] L. Liu, M. Faryad, A. S. Hall, G. D. Barber, S. Erten, T. E. Mallouk, A. Lakhtakia, and T. S. Mayer, "Experimental excitation of multiple surface-plasmon-polariton waves and waveguide modes in a one- dimensional photonic crystal atop a two-dimensional metal grating," J. Nanophotonics 9(1), 093593 (2015). 15 [22] F.-J. Haug, K. Soderstrom, A. Naqavi, and C. Ballif, "Excitation of guided-mode resonances in thin film silicon solar cells," MRS Symp. Proc. 1321(1), 123 -- 128 (2011). [23] T. Khaleque and R. Magnusson, "Light management through guided-mode resonances in thin-film silicon solar cells," J. Nanophotonics 8(1), 083995 (2014). [24] J. A. Polo Jr., T. G. Mackay, and A. Lakhtakia, Electromagnetic Surface Waves: A Modern Perspective, Elsevier, Waltham, MA, USA (2013). [25] A. W. Snyder and J. D. Love, Optical Waveguide Theory, Chapman and Hall, New York, NY, USA (1983). [26] T. H. Anderson, M. Faryad, T. G. Mackay, A. Lakhtakia, and R. Singh, "Combined optical-electrical finite-element simulations of thin-film solar cells with homogeneous and nonhomogeneous intrinsic lay- ers," J. Photon. Energy 6(2), 025502 (2016). [27] T. H. Anderson, T. G. Mackay, and A. Lakhtakia, "Enhanced efficiency of Schottky-barrier solar cell with periodically nonhomogeneous indium gallium nitride layer," J. Photon. Energy 7(1), 014502 (2017). [28] A. S. Hall, M. Faryad, G. D. Barber, L. Liu, S. Erten, T. S. Mayer, A. Lakhtakia, and T. E. Mallouk, "Broadband light absorption with multiple surface plasmon polariton waves excited at the interface of a metallic grating and photonic crystal," ACS Nano 7(6), 4995 -- 5007 (2013). [29] M. Faryad, A. S. Hall, G. D. Barber, T. E. Mallouk, and A. Lakhtakia, "Excitation of multiple surface- plasmon-polariton waves guided by a periodically corrugated interface of a metal and a periodic multi- layered isotropic dielectric material," J. Opt. Soc. Am. B 29(4), 704 -- 713 (2012). [30] M. Faryad, L. Liu, T. S. Mayer, and A. Lakhtakia, "Optical and electrical modeling of an amorphous- silicon tandem solar cell with nonhomogeneous intrinsic layers and a periodically corrugated back- reflector," Proc. SPIE 8823(1), 882306 (2013). [31] M. V. Shuba, M. Faryad, M. E. Solano, P. B. Monk, and A. Lakhtakia, "Adequacy of the rigorous coupled-wave approach for thin-film silicon solar cells with periodically corrugated metallic backreflec- tors: spectral analysis," J. Opt. Soc. Am. A 32(7), 1222 -- 1230 (2015). [32] A. S. Ferlauto, G. M. Ferreira, J. M. Pearce, C. R. Wronski, R. W. Collins, X. Deng, and G. Ganguly, "Analytical model for the optical functions of amorphous semiconductors from the near-infrared to ultraviolet: Applications in thin film photovoltaics," J. Appl. Phys. 92(5), 2424 -- 2436 (2002). [33] S. J. Fonash, Solar Cell Device Physics, Academic Press, Burlington, MA, USA (2010). [34] http://refractiveindex.info/?group=METALS/material=Silver (accessed 9 July 2017). [35] E. N. Glytsis and T. K. Gaylord, "Rigorous three-dimensional coupled-wave diffraction analysis of single and cascaded anisotropic gratings," J. Opt. Soc. Am. A 4(11), 2061 -- 2080 (1987). [36] M. Onishi, K. Crabtree, and R. A. Chipman, "Formulation of rigorous coupled-wave theory for gratings in bianisotropic media," J. Opt. Soc. Am. A 28(8), 1747 -- 1758 (2011). [37] E. Kreyszig, Advanced Engineering Mathematics, 10th edn., Wiley, Hoboken, NJ, USA (2011). [38] H. Lutkepohl, Handbook of Matrices, Wiley, Chichester, United Kingdom (1996). [39] F. Ahmad, T. H. Anderson, B. J. Civiletti, P. B. Monk, and A. Lakhtakia, "On optical-absorption peaks in a nonhomogeneous dielectric material over a two-dimensional metallic surface-relief grating," Proc. SPIE 10356(1), 103560I (2017). 16 [40] M. F. Iskander, Electromagnetic Fields and Waves, 2nd edn., Waveland Press, Long Grove, IL, USA (2013). [41] M. V. Shuba and A. Lakhtakia, "Splitting of absorptance peaks in absorbing multilayer backed by a periodically corrugated metallic reflector," J. Opt. Soc. Am. A 33(4), 779 -- 784 (2016). [42] J. Dutta, S. A. Ramakrishna, and A. Lakhtakia, "Characteristics of surface plasmon -- polariton waves excited on 2D periodically patterned columnar thin films of silver," J. Opt. Soc. Am. A 33(9), 1697 -- 1704 (2016). [43] X.-Y. Gao, L. Liang, and Q.-G. Lin, "Analysis of the optical constants of aluminum-doped zinc-oxide films by using the single-oscillator model," J. Korean Phys. Soc. 57(4), 710 -- 714 (2010). [44] M. Faryad and A. Lakhtakia, "Grating-coupled excitation of multiple surface plasmon-polariton waves," Phys. Rev. A 84(3), 033852 (2011). 17
1907.07365	1	1907	2019-07-17T07:31:03	Thickness determination of MoS2, MoSe2, WS2 and WSe2 on transparent stamps used for deterministic transfer of 2D materials	[ "physics.app-ph", "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Here, we propose a method to determine the thickness of the most common transition metal dichalcogenides (TMDCs) placed on the surface of transparent stamps, used for the deterministic placement of two-dimensional materials, by analyzing the red, green and blue channels of transmission-mode optical microscopy images of the samples. In particular, the blue channel transmittance shows a large and monotonic thickness dependence, making it a very convenient probe of the flake thickness. The method proved to be robust given the small flake-to-flake variation and the insensitivity to doping changes of MoS2. We also tested the method for MoSe2, WS2 and WSe2. These results provide a reference guide to identify the number of layers of this family of materials on transparent substrates only using optical microscopy.	physics.app-ph	physics	This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Thickness determination of MoS2, MoSe2, WS2 and WSe2 on transparent stamps used for deterministic transfer of 2D materials Najme S. Taghavi1,2, Patricia Gant1(), Peng Huang1,3, Iris Niehues4, Robert Schmidt4, Steffen Michaelis de Vasconcellos4, Rudolf Bratschitsch4, Mar García-Hernández1, Riccardo Frisenda1(), Andres Castellanos-Gomez1() 1 Materials Science Factory, Instituto de Ciencia de Materiales de Madrid (ICMM), Consejo Superior de Investigaciones Científicas (CSIC), Sor Juana Inés de la Cruz 3, 28049 Madrid, Spain. 2 Faculty of Physics, Khaje Nasir Toosi University of Technology (KNTU), Tehrān 19697 64499, Iran. 3 State Key Laboratory of Tribology, Tsinghua University, Beijing 100084, China. 4 Institute of Physics and Center for Nanotechnology, University of Münster, 48149 Münster, Germany. -- -- -- -- -- -- -- -- -- -- -- -- Address correspondence to Patricia Gant ([email protected]), Riccardo Frisenda ([email protected]) and Andres Castellanos-Gomez ([email protected]) ABSTRACT Here, we propose a method to determine the thickness of the most common transition metal dichalcogenides (TMDCs) placed on the surface of transparent stamps, used for the deterministic placement of two-dimensional materials, by analyzing the red, green and blue channels of transmission-mode optical microscopy images of the samples. In particular, the blue channel transmittance shows a large and monotonic thickness dependence, making it a very convenient probe of the flake thickness. The method proved to be robust given the small flake-to-flake variation and the insensitivity to doping changes of MoS2. We also tested the method for MoSe2, WS2 and WSe2. These results provide a reference guide to identify the number of layers of this family of materials on transparent substrates only using optical microscopy. MANUSCRIPT TEXT. Since the isolation of graphene in 2004[1], the mechanical exfoliation method (also called Scotch tape method) has established itself as one of the most used techniques to produce 2D materials [2 -- 5]. Its facile implementation combined with the high quality of the produced samples are most likely the reasons behind the success of this technique. Mechanical exfoliation, nonetheless, yields randomly distributed flakes of various thicknesses and sizes on the surface of the substrate. This limitation has been overcome to a great extent through the development of rapid methods to find and identify thin flakes based on the observation of the apparent color when they are transferred onto a SiO2/Si surface [6 -- 12]. On this substrate, there is a dependency of the apparent color of the flake with its thickness due to thin-film interference effects and many 1 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 epi-illumination (reflection mode) microscopy-based methods have been developed to identify 2D materials and to determine their number of layers [6 -- 12]. An alternative way to overcome the limitations of mechanical exfoliation relies on the use of experimental tools that allow for the deterministic placement of flakes onto any desired sample position with micrometer accuracy [13 -- 19]. These transfer techniques opened the field of van der Waals heterostructures [18,20 -- 23]. To carry out the deterministic placement, the isolated flakes have to be deposited onto the surface of a polymer-based stamp that is used as a release layer or sacrificial carrier substrate. These stamps are made of transparent material to allow for the accurate alignment of the flake to be transferred with the acceptor surface by inspection with an optical microscope. Unfortunately, on the surface of the transparent stamps most of the previously developed optical identification methods, based on the apparent colors, are less straightforward to implement and less effective and other complementary techniques such as Raman or photoluminescence (which are slower and more complex than simple optical microscopy) are typically used to determine the number of layers of the flakes deposited on the polymeric stamps [24 -- 27]. Due to the increasing interest on the use of deterministic placement methods to assemble nanodevices and van der Waals heterostructures, the development of alternative methods to determine the thickness of 2D materials on the surface of the transparent polymeric stamps is extremely relevant for the 2D community. That is exactly the goal of this work, to provide a fast and reliable method to identify TMDC flakes on transparent polymeric stamps and to determine their number of layers. Here, we demonstrate that the quantitative analysis of transmission optical mode images of TMDC flakes on the surface of transparent polydimethylsiloxane (PDMS) stamps is a reliable method to accurately determine their thickness. We compare the results of the quantitative analysis of the transmission mode optical images with results obtained via micro-reflectance spectroscopy [28], photoluminescence and Raman spectroscopy to verify its reliability. In order to test the limitations of this method, we probe its sensitivity concerning the doping level of the sample by measuring MoS2 samples with different intentional doping. We found that the determination of the thickness with the analysis of transmission mode images is rather independent on the doping level, giving similar results for all the studied samples. Finally, we extended the study to other TMDCs to provide a reference guide to identify the number of layers of this family of materials. As the measurement of the transmittance is a differential measurement, the effect of the substrate is accounted for and thus these results could be extrapolated to other transparent stamp surfaces like poly(methyl methacrylate) (PMMA) or polycarbonate/hexagonal boron-nitride (PC/hBN). Figure 1a shows a transmission mode optical microscopy image of a MoS2 flake transferred onto a Gel-Film substrate (a commercially available polydimethylsiloxane film, by Gel-Pak) which is typically used as a stamp for deterministic dry transfer of 2D materials [16,29]. Figure 1b displays line profiles of the intensity of the red, green and blue channels extracted from the image in Figure 1a. The transmittance of the different channels is calculated by normalizing the red, green and blue (RGB) channel images to the intensity of these channels measured on the bare substrate. From the line profiles, one can see that the transmittance of each channel changes stepwise with the number of layers. To determine the feasibility of using the transmittance to determine the number of layers one needs to characterize the statistical variations of the transmittance in different MoS2 flakes and the uncertainty associated to these measurements. Therefore, we acquired transmission mode optical microscopy images of 202 MoS2 flakes and we compiled three histograms by binning the measured transmittance of the flakes for each color channel. Figure 1c shows the histograms of the red, green and blue channels plotted versus 1-transmittance. Although all the histograms display prominent 2 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 peaks that correspond to different number of layers, the histogram of the blue channel shows a larger separation between the peaks (while maintaining a similar peak width) making it easier to distinguish between 1, 2, 3 or 4 layers of MoS2. The higher contrast of the blue channel is expected as MoS2 (and MoSe2, WS2 and WSe2 as well) present a strong excitonic feature (sometimes referred to as C exciton in the literature) in the blue part of the spectrum that yields a strong optical absorption in that wavelength range.[30,31] In order to verify the reliability of this method to determine the number of layers we benchmarked it with other commonly used spectroscopic techniques. In Figure 2a the experimental data employed to build up the histogram in Figure 1c is displayed in a scatter plot. We have selected 4 flakes with different thickness in the range between 1 layer and 4 layers, respectively labeled as 1L, 2L, 3L and 4L accordingly to their blue channel transmission, and we have carried out micro-reflection, Raman and photoluminescence measurements to have three independent methods to determine the thickness. Figure 2b shows the differential reflectance spectra acquired on these flakes where the intensity and the energy of the A exciton monotonically changes from 1L to 4L. The intensities and exciton positions obtained in the spectra agree with the values expected for 1L to 4L,[31] which means a correct assignment of the thickness estimated by the blue channel values [32]. Figure 2c compares the Raman spectra measured on the same flakes (vertically displaced by 0.1 for clarity) with a 532 nm excitation laser. We verify the number of layers of the samples from the difference between the A1g and E1 2g phonon [33,34] energies in Raman and check again the accuracy in the determination of the number of layers using the blue channel transmittance. Figure 2d shows photoluminescence measurements for 1L, 2L, 3L and 4L flakes, also measured with a 532 nm excitation laser. The intensity and the position of the A exciton dramatically depends on the number of layers and it confirms that the assignment of number of layers determined by the blue channel transmission is reliable.[35,36] We further test the analysis of the blue channel transmission images by studying MoS2 samples synthesized with intentional substitutional metal atoms at the Mo sites to enquiry about the robustness of this technique against a moderate variation of the chemical composition that lead to a big change in the electronic properties.[37,38] We follow the synthesis method described in references [37 -- 40] for the growth of Fe-doped, Ni-doped, Nb-doped and Co-doped MoS2 crystals. In all cases a 0.5% of dopant material has been added in the ampoule for the synthesis of these doped MoS2 samples which lead to a final doping level in the 0.3-0.4% range. Figure 3 compares the histograms of the blue channel transmittance constructed by analyzing 50 flakes of each doped MoS2 material. This comparison clearly illustrates how our method is rather insensitive to variations of the chemical composition (that induces strong changes in the doping level of the material). We note that other optical spectroscopic methods to determine the thickness of transition metal dichalcogenides such as Raman and photoluminescence spectroscopy are strongly dependent on the doping level of the samples and on slight variations of the chemical composition.[41 -- 45] We extended this method to other 2D materials of the TMDCs family to provide a general guide to identify these materials through the analysis of the blue channel of transmission images. Figure 4 compares the blue channel transmission histograms built up after analyzing 202 MoS2 flakes, 200 MoSe2 flakes, 200 WS2 flakes and 200 WSe2 flakes. In all the cases the histograms show well-resolved peaks indicating that one can unambiguously determine the number of layers through the quantitative analysis of the blue channel of the transmission mode optical images. 3 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 In summary, we have introduced a very simple and fast yet reliable method to determine the number of layers of MoS2, MoSe2, WS2 and WSe2 deposited on a PDMS stamp, used for deterministic placement of 2D materials. Moreover, being a differential measurement, the effect of the substrate is removed and it could be extended to other transparent substrates. We have demonstrated that the transmittance, extracted from the blue channel of transmission mode optical images, monotonically depends on the number of layers. We have benchmarked the layer assignment done with this method with other extended spectroscopic techniques (micro-reflection, Raman, photoluminescence) finding an excellent agreement. Interestingly, this method is robust enough to provide an accurate layer determination even for samples with different doping level. In view of all this, the quantitative analysis of the transmission mode images can become a powerful method to select the flakes of 2D materials deposited onto transparent polymeric stamps prior to their deterministic transfer. Experimental Materials: MoS2 samples were prepared out of a bulk natural molybdenite crystal (Moly Hill mine, Quebec, Canada). MoSe2 and WSe2 samples were prepared out of bulk synthetic crystals grown by physical vapour transport method (provided by Prof. Rudolf Bratschitsch). WS2 samples were prepared out of a bulk synthetic crystal grown by physical vapour transport method at Tennessee Crystal Center. The Fe-doped, Ni-doped, Nb-doped and Co-doped MoS2 crystals were grown following the protocols described in Refs. [37 -- 40]. A 0.5% of dopant material has been added during the synthesis of these doped MoS2 samples which leads to a final doping level in the 0.3-0.4% range. The transparent stamp substrate used in this work is a commercially available polydimethylsiloxane-based substrate manufactured by Gel-Pak® (Gel-Film® WF X4 6.0 mil). The TMDC flakes are exfoliated from the bulk crystals and transferred onto the surface of the Gel-Film stamp with a Nitto 224SPV adhesive tape. Optical microscopy: Optical microscopy images have been acquired with three different upright metallurgical microscopes, an AM Scope, a Motic BA MET310-T and a Nikon Eclipse CI, obtaining identical results. The transmission mode images have been acquired with an achromat condenser lens (N.A. 0.85) that ensures Köhler illumination, yielding to a homogeneous illumination spot of approximately 4 mm2 on the sample. The light was collected with a 50x magnification plan achromatic objective (NA 0.55). Two different digital cameras were tested during this work, an AM Scope mu1803 camera with 18 megapixels and a Canon EOS 1200D, also providing identical results. Image analysis: The quantitative analysis of the transmittance of the flakes and the rippling wavelength has been carried out using Gwyddion® and ImageJ software.[46,47] 4 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Acknowledgements We thank Prof. Der-Yuh Lin and Prof. Tsung-Shine Ko for providing the doped MoS2 samples. NT acknowledges to the Ministry of Science, Research and Technology of Iran. ACG and PG acknowledge funding from the European Commission Graphene Flagship (Grant Graphene Core 2 785219). RF acknowledges support from the Netherlands Organization for Scientific Research (NWO) through the research program Rubicon with project number 680-50-1515. ACG acknowledge funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement n° 755655, ERC-StG 2017 project 2D-TOPSENSE). D.-Y.L. acknowledges the financial support from the Ministry of Science and Technology of Taiwan, Republic of China under contract No. MOST 107-2112-M-018-002.. References [1] Novoselov, K.S.; Geim, A.K.; Morozov, S. V; Jiang, D.; Zhang, Y.; Dubonos, S. V; Grigorieva, I. V; Firsov, A.A. Electric Field Effect in Atomically Thin Carbon Films. Science, 2004, 306, 666 -- 669. [2] Bonaccorso, F.; Lombardo, A.; Hasan, T.; Sun, Z.; Colombo, L.; Ferrari, A.C. Production and Processing of Graphene and 2d Crystals. Mater. Today, 2012, 15, 564 -- 589. [3] Ferrari, A.C.; Bonaccorso, F.; Fal'ko, V.; Novoselov, K.S.; Roche, S.; Bøggild, P.; Borini, S.; Koppens, F.H.L.; Palermo, V.; Pugno, N.; Garrido, J.A.; Sordan, R.; Bianco, A.; Ballerini, L.; Prato, M.; Lidorikis, E.; Kivioja, J.; Marinelli, C.; Ryhänen, T.; Morpurgo, A.; Coleman, J.N.; Nicolosi, V.; Colombo, L.; Fert, A.; Garcia-Hernandez, M.; Bachtold, A.; Schneider, G.F.; Guinea, F.; Dekker, C.; Barbone, M.; Sun, Z.; Galiotis, C.; Grigorenko, A.N.; Konstantatos, G.; Kis, A.; Katsnelson, M.; Vandersypen, L.; Loiseau, A.; Morandi, V.; Neumaier, D.; Treossi, E.; Pellegrini, V.; Polini, M.; Tredicucci, A.; Williams, G.M.; Hee Hong, B.; Ahn, J.-H.; Min Kim, J.; Zirath, H.; van Wees, B.J.; van der Zant, H.; Occhipinti, L.; Di Matteo, A.; Kinloch, I.A.; Seyller, T.; Quesnel, E.; Feng, X.; Teo, K.; Rupesinghe, N.; Hakonen, P.; Neil, S.R.T.; Tannock, Q.; Löfwander, T.; Kinaret, J. Science and Technology Roadmap for Graphene, Related Two-Dimensional Crystals, and Hybrid Systems. Nanoscale, 2015, 7, 4598 -- 4810. [4] Radisavljevic, B.; Radenovic, A.; Brivio, J.; Giacometti, V.; Kis, A. Single-Layer MoS2 Transistors. Nat. Nanotechnol., 2011, 6, 147 -- 150. [5] Wang, Q.H.; Kalantar-Zadeh, K.; Kis, A.; Coleman, J.N.; Strano, M.S. Electronics and Optoelectronics of Two-Dimensional Transition Metal Dichalcogenides. Nat. Nanotechnol., 2012, 7, 699 -- 712. [6] Ni, Z.H.; Wang, H.M.; Kasim, J.; Fan, H.M.; Yu, T.; Wu, Y.H.; Feng, Y.P.; Shen, Z.X. Graphene Thickness Determination Using Reflection and Contrast Spectroscopy. Nano Lett., 2007, 7, 2758 -- 2763. [7] Inhwa Jung, †; Matthew Pelton, ‡; Richard Piner, †; Dmitriy A. Dikin, †; Sasha Stankovich, †; Supinda Watcharotone, †; Martina Hausner, § and; Rodney S. Ruoff*, †. Simple Approach for High-Contrast Optical Imaging and Characterization of Graphene-Based Sheets. 2007. [8] Li, H.; Wu, J.; Huang, X.; Lu, G.; Yang, J.; Lu, X.; Xiong, Q.; Zhang, H. Rapid and Reliable Thickness Identification 5 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 of Two-Dimensional Nanosheets Using Optical Microscopy. ACS Nano, 2013, 7, 10344 -- 10353. [9] Wang, X.; Zhao, M.; Nolte, D.D. Optical Contrast and Clarity of Graphene on an Arbitrary Substrate. Appl. Phys. Lett., 2009, 95, 81102 -- 31901. [10] Zhang, H.; Ran, F.; Shi, X.; Fang, X.; Wu, S.; Liu, Y.Y.; Zheng, X.; Yang, P.; Liu, Y.Y.; Wang, L.; Huang, X.; Li, H.; Huang, W. Optical Thickness Identification of Transition Metal Dichalcogenide Nanosheets on Transparent Substrates. Nanotechnology, 2017, 28, 164001. [11] Yu, Y.; Li, Z.; Wang, W.; Guo, X.; Jiang, J.; Nan, H.; Ni, Z. Investigation of Multilayer Domains in Large-Scale CVD Monolayer Graphene by Optical Imaging. J. Semicond., 2017, 38, 33003. [12] Wang, Y.Y.; Gao, R.X.; Ni, Z.H.; He, H.; Guo, S.P.; Yang, H.P.; Cong, C.X.; Yu, T. Thickness Identification of Two-Dimensional Materials by Optical Imaging. Nanotechnology, 2012, 23, 495713. [13] Dean, C.R.; Young, A.F.; Meric, I.; Lee, C.; Wang, L.; Sorgenfrei, S.; Watanabe, K.; Taniguchi, T.; Kim, P.; Shepard, K.L.; Hone, J. Boron Nitride Substrates for High-Quality Graphene Electronics. Nat. Nanotechnol., 2010, 5, 722 -- 726. [14] Zomer, P.J.; Dash, S.P.; Tombros, N.; Van Wees, B.J. A Transfer Technique for High Mobility Graphene Devices on Commercially Available Hexagonal Boron Nitride. Appl. Phys. Lett., 2011, 99, 232104. [15] Zomer, P.J.; Guimarães, M.H.D.; Brant, J.C.; Tombros, N.; Van Wees, B.J. Fast Pick up Technique for High Quality Heterostructures of Bilayer Graphene and Hexagonal Boron Nitride. Appl. Phys. Lett., 2014, 105, 13101. [16] Castellanos-Gomez, A.; Buscema, M.; Molenaar, R.; Singh, V.; Janssen, L.; van der Zant, H.S.J.; Steele, G.A. Deterministic Transfer of Two-Dimensional Materials by All-Dry Viscoelastic Stamping. 2D Mater., 2014, 1, 011002. [17] Pizzocchero, F.; Gammelgaard, L.; Jessen, B.S.; Caridad, J.M.; Wang, L.; Hone, J.; Bøggild, P.; Booth, T.J. The Hot Pick-up Technique for Batch Assembly of van Der Waals Heterostructures. Nat. Commun., 2016, 7, 11894. [18] Frisenda, R.; Navarro-Moratalla, E.; Gant, P.; De Lara, D.P.; Jarillo-Herrero, P.; Gorbachev, R. V; Castellanos-Gomez, A. Recent Progress in the Assembly of Nanodevices and van Der Waals Heterostructures by Deterministic Placement of 2D Materials. Chem. Soc. Rev., 2018, 47, 53 -- 68. [19] Masubuchi, S.; Morimoto, M.; Morikawa, S.; Onodera, M.; Asakawa, Y.; Watanabe, K.; Taniguchi, T.; Machida, T. Autonomous Robotic Searching and Assembly of Two-Dimensional Crystals to Build van Der Waals Superlattices. Nat. Commun., 2018, 9, 1413. [20] Liu, Y.; Weiss, N.O.; Duan, X.; Cheng, H.-C.; Huang, Y.; Duan, X. Van Der Waals Heterostructures and Devices. Nat. Rev. Mater., 2016, 1, 16042. [21] Geim, A.K.; Grigorieva, I. V. Van Der Waals Heterostructures. Nature, 2013, 499, 419 -- 425. [22] Novoselov, K.S.; Mishchenko, A.; Carvalho, A.; Castro Neto, A.H. 2D Materials and van Der Waals Heterostructures. Science, 2016, 353, aac9439. [23] Frisenda, R.; Molina-Mendoza, A.J.; Mueller, T.; Castellanos-Gomez, A.; van der Zant, H.S.J. Atomically Thin P -- n Junctions Based on Two-Dimensional Materials. Chem. Soc. Rev., 2018, 47, 3339 -- 3358. [24] Splendiani, A.; Sun, L.; Zhang, Y.; Li, T.; Kim, J.; Chim, C.-Y.; Galli, G.; Wang, F. Emerging Photoluminescence in Monolayer MoS2. Nano Lett., 2010, 10, 1271 -- 1275. [25] Pimenta, M.A.; del Corro, E.; Carvalho, B.R.; Fantini, C.; Malard, L.M. Comparative Study of Raman Spectroscopy in Graphene and MoS 2 -Type Transition Metal Dichalcogenides. Acc. Chem. Res., 2015, 48, 41 -- 47. 6 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 [26] Zhang, X.; Qiao, X.-F.; Shi, W.; Wu, J.-B.; Jiang, D.-S.; Tan, P.-H. Phonon and Raman Scattering of Two-Dimensional Transition Metal Dichalcogenides from Monolayer, Multilayer to Bulk Material. Chem. Soc. Rev., 2015, 44, 2757 -- 2785. [27] Zeng, H.; Cui, X. An Optical Spectroscopic Study on Two-Dimensional Group-VI Transition Metal Dichalcogenides. Chem. Soc. Rev., 2015, 44, 2629 -- 2642. [28] Frisenda, R.; Niu, Y.; Gant, P.; Molina-Mendoza, A.J.; Schmidt, R.; Bratschitsch, R.; Liu, J.; Fu, L.; Dumcenco, D.; Kis, A.; Perez De Lara, D.; Castellanos-Gomez, A. Micro-Reflectance and Transmittance Spectroscopy: A Versatile and Powerful Tool to Characterize 2D Materials. J. Phys. D. Appl. Phys., 2017, 50, 074002. [29] Yang, R.; Zheng, X.; Wang, Z.; Miller, C.J.; Feng, P.X.-L. Multilayer MoS2 Transistors Enabled by a Facile Dry-Transfer Technique and Thermal Annealing. J. Vac. Sci. Technol. B, Nanotechnol. Microelectron. Mater. Process. Meas. Phenom., 2014, 32, 61203. [30] Castellanos-Gomez, A.; Quereda, J.; van der Meulen, H.P.; Agraït, N.; Rubio-Bollinger, G. Spatially Resolved Optical Absorption Spectroscopy of Single- and Few-Layer MoS2 by Hyperspectral Imaging. Nanotechnology, 2016, 27, 115705. [31] Niu, Y.; Gonzalez-Abad, S.; Frisenda, R.; Marauhn, P.; Drüppel, M.; Gant, P.; Schmidt, R.; Taghavi, N.; Barcons, D.; Molina-Mendoza, A.; de Vasconcellos, S.; Bratschitsch, R.; Perez De Lara, D.; Rohlfing, M.; Castellanos-Gomez, A. Thickness-Dependent Differential Reflectance Spectra of Monolayer and Few-Layer MoS2, MoSe2, WS2 and WSe2. Nanomaterials, 2018, 8, 725. [32] Li, H.; Zhang, Q.; Yap, C.C.R.; Tay, B.K.; Edwin, T.H.T.; Olivier, A.; Baillargeat, D. From Bulk to Monolayer MoS2: Evolution of Raman Scattering. Adv. Funct. Mater., 2012, 22, 1385 -- 1390. [33] Lee, C.; Yan, H.; Brus, L.E.; Heinz, T.F.; Hone, J.; Ryu, S. Anomalous Lattice Vibrations of Single- and Few-Layer MoS2. ACS Nano, 2010, 4, 2695 -- 2700. [34] Tonndorf, P.; Schmidt, R.; Böttger, P.; Zhang, X.; Börner, J.; Liebig, A.; Albrecht, M.; Kloc, C.; Gordan, O.; Zahn, D.R.T.; Michaelis de Vasconcellos, S.; Bratschitsch, R. Photoluminescence Emission and Raman Response of Monolayer MoS_2, MoSe_2, and WSe_2. Opt. Express, 2013, 21, 4908. [35] Mak, K.F.; Lee, C.; Hone, J.; Shan, J.; Heinz, T.F. Atomically Thin MoS2: A New Direct-Gap Semiconductor. Phys. Rev. Lett., 2010, 105, 2 -- 5. [36] Splendiani, A.; Sun, L.; Zhang, Y.; Li, T.; Kim, J.; Chim, C.-Y.; Galli, G.; Wang, F. Emerging Photoluminescence in Monolayer MoS2. Nano Lett., 2010, 10, 1271 -- 1275. [37] Suh, J.; Park, T.-E.; Lin, D.-Y.; Fu, D.; Park, J.; Jung, H.J.; Chen, Y.; Ko, C.; Jang, C.; Sun, Y. Doping against the Native Propensity of MoS2: Degenerate Hole Doping by Cation Substitution. Nano Lett., 2014, 14, 6976 -- 6982. [38] Svatek, S.A.; Antolin, E.; Lin, D.-Y.; Frisenda, R.; Reuter, C.; Molina-Mendoza, A.J.; Muñoz, M.; Agraït, N.; Ko, T.-S.; de Lara, D.P.; Castellanos-Gomez, A. Gate Tunable Photovoltaic Effect in MoS 2 Vertical P -- n Homostructures. J. Mater. Chem. C, 2017, 5, 854 -- 861. [39] Reuter, C.; Frisenda, R.; Lin, D.-Y.; Ko, T.-S.; Perez de Lara, D.; Castellanos-Gomez, A. A Versatile Scanning Photocurrent Mapping System to Characterize Optoelectronic Devices Based on 2D Materials. Small Methods, 2017, 1, 1700119. [40] Wang, S.Y.; Ko, T.S.; Huang, C.C.; Huang, Y.S. Optical and Electrical Properties of MoS2 and Fe-Doped MoS2. Jpn. J. Appl. Phys., 2014, 53, 04EH07. [41] Chen, Y.; Dumcenco, D.O.; Zhu, Y.; Zhang, X.; Mao, N.; Feng, Q.; Zhang, M.; Zhang, J.; Tan, P.-H.; Huang, 7 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Y.-S. Composition-Dependent Raman Modes of Mo1− XWxS2 Monolayer Alloys. Nanoscale, 2014, 6, 2833 -- 2839. [42] Dumcenco, D.O.; Kobayashi, H.; Liu, Z.; Huang, Y.-S.; Suenaga, K. Visualization and Quantification of Transition Metal Atomic Mixing in Mo 1− x W x S 2 Single Layers. Nat. Commun., 2013, 4, 1351. [43] Mann, J.; Ma, Q.; Odenthal, P.M.; Isarraraz, M.; Le, D.; Preciado, E.; Barroso, D.; Yamaguchi, K.; von Son Palacio, G.; Nguyen, A. 2‐Dimensional Transition Metal Dichalcogenides with Tunable Direct Band Gaps: MoS2 (1 -- x) Se2x Monolayers. Adv. Mater., 2014, 26, 1399 -- 1404. [44] Zhang, M.; Wu, J.; Zhu, Y.; Dumcenco, D.O.; Hong, J.; Mao, N.; Deng, S.; Chen, Y.; Yang, Y.; Jin, C. Two-Dimensional Molybdenum Tungsten Diselenide Alloys: Photoluminescence, Raman Scattering, and Electrical Transport. ACS Nano, 2014, 8, 7130 -- 7137. [45] Mouri, S.; Miyauchi, Y.; Matsuda, K. Tunable Photoluminescence of Monolayer MoS2 via Chemical Doping. Nano Lett., 2013, 13, 5944 -- 5948. [46] Nečas, D.; Klapetek, P. Gwyddion: An Open-Source Software for SPM Data Analysis. Open Phys., 2012, 10, 181 -- 188. [47] Abràmoff, M.D.; Magalhães, P.J.; Ram, S.J. Image Processing with ImageJ. Biophotonics Int., 2004, 11, 36 -- 42. Figure 1. (a) Optical image in transmission mode of a MoS2 flake with different thicknesses. (b) Line profiles of the intensities of the red, green and blue channels of the image in 1a. (c) Histograms of the 1-transmission value in the RGB channels for several MoS2 flakes. 8 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Figure 2. (a) Scatter plot and histogram of the blue channel transmission values for several MoS2 flakes. (b) Differential reflectance spectra for MoS2 with different number of layers. (c) Raman spectra for MoS2 with different number of layers. (d) Photoluminescence of MoS2 for different number of layers. 9 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Figure 3. Histograms of the 1-transmission value in the blue channel for MoS2 flakes doped with (a) Co, (b) Ni, (c) Fe and (d) Nb. (e) Average of the 1-transmission values for 1L to 4L in each material. 10 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Figure 4. Histograms of the 1-transmittance value in the blue channel for (a) MoS2, (b) MoSe2, (c) WS2 and (d) WSe2 flakes. (e) Average of the 1-transmittance values for 1L to 4L in each material. 11 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Electronic Supplementary Material Thickness determination of MoS2, MoSe2, WS2 and WSe2 on transparent stamps used for deterministic transfer of 2D materials Najme S. Taghavi1,2, Patricia Gant1(), Peng Huang1,3, Iris Niehues4, Robert Schmidt4, Steffen Michaelis de Vasconcellos4, Rudolf Bratschitsch4, Mar García-Hernández1, Riccardo Frisenda1(), Andres Castellanos-Gomez1() 1 Materials Science Factory, Instituto de Ciencia de Materiales de Madrid (ICMM), Consejo Superior de Investigaciones Científicas (CSIC), Sor Juana Inés de la Cruz 3, 28049 Madrid, Spain. 2 Faculty of Physics, Khaje Nasir Toosi University of Technology (KNTU), Tehrān 19697 64499, Iran. 3 State Key Laboratory of Tribology, Tsinghua University, Beijing 100084, China. 4 Institute of Physics and Center for Nanotechnology, University of Münster, 48149 Münster, Germany. -- -- -- -- -- -- -- -- -- -- -- -- Address correspondence Castellanos-Gomez ([email protected]) to Patricia Gant ([email protected]), Riccardo Frisenda ([email protected]) and Andres INFORMATION ABOUT ELECTRONIC SUPPLEMENTARY MATERIAL. 1. Constructing a thickness map from transmission mode optical microscopy image 2. Blue channel transmittance for flakes thicker than 4 layers Constructing a thickness map from transmission mode optical microscopy image Interestingly one can directly convert a transmission mode optical microscopy image into a thickness map by using this quantitative analysis of the blue channel. First, the blue channel of the transmission mode image is extracted and the average intensity of the substrate (T0) is measured. Then, we calculate 1-T/T0 to each pixel of the blue channel image, where T is the blue intensity at each pixel of the image. Under such transformation the substrate value becomes ~0 and the value on the flakes can be directly compared to the values in the histograms displayed in Figure 4. Indeed, by selecting a colormap accordingly to the 1-T/T0 histograms in Figure 4 the map directly displays the number of layers. 12 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Figure S1. Transformation of a transmission mode optical microscopy image (top) into a thickness map (bottom) by applying the formula 1-T/T0, where T is the intensity of each pixel and T0 is the average intensity of the substrate. By selecting a colormap accordingly to the 1-T/T0 histograms in Figure 4 the map displays the number of layers. Blue channel transmittance for flakes thicker than 4 layers In the main text we focused on the analysis of the transmittance of flakes 1L to 4L thick because we had plenty of statistics in that thickness range and because we used complementary techniques like Raman spectroscopy, photoluminescence and micro-reflectance/transmittance to double-check the assessed number of layers. The uncertainty of the number of layers assessment done by those techniques increases substantially for flakes thicker than 4-5 layers. Figure S2 shows the analysis of the transmittance of the blue channel for TMDC flakes thicker than 4 layers. 13 This is the authors' version (post peer-review) of the manuscript: N.S. Taghavi et al. Nano Research (2019) 12: 1691 https://doi.org/10.1007/s12274-019-2424-6 That has been published in its final form: https://link.springer.com/article/10.1007/s12274-019-2424-6 Figure S2. 1 - transmittance of the blue channel measured for MoS2, MoSe2, WS2 and WSe2 flakes with thicknesses in the range of 1L to 8L. . 14
1910.02118	2	1910	2019-12-04T01:09:46	Stack Pressure Considerations for Room Temperature All-Solid-State Lithium Metal Batteries	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	All-solid-state batteries are expected to enable batteries with high energy density with the use of lithium metal anodes. Although solid electrolytes are believed to be mechanically strong enough to prevent lithium dendrites from propagating, various reports today still show cell failure due to lithium dendritic growth at room temperature. While cell parameters such as current density, electrolyte porosity and interfacial properties have been investigated, mechanical properties of lithium metal and the role of applied stack pressure on the shorting behavior is still poorly understood. Here, we investigated failure mechanisms of lithium metal in all-solid-state batteries as a function of stack pressure, and conducted in situ characterization of the interfacial and morphological properties of the buried lithium in solid electrolytes. We found that a low stack pressure of 5 MPa allows reliable plating and stripping in a lithium symmetric cell for more than 1000 hours, and a Li \| Li6PS5Cl \| LiNi0.80Co0.15Al0.05O2 full cell, plating more than 4 um of lithium per charge, is able to cycle over 200 cycles at room temperature. These results suggest the possibility of enabling the lithium metal anode in all-solid-state batteries at reasonable stack pressures.	physics.app-ph	physics	This is the pre-peer reviewed version of the following article: Doux, J.‐M., Nguyen, H., Tan, D. H. S., Banerjee, A., Wang, X., Wu, E. A., Jo, C., Yang, H., Meng, Y. S., Stack Pressure Considerations for Room‐Temperature All‐Solid‐State Lithium Metal Batteries. Adv. Energy Mater. 2019, 1903253. which has been published in final form at https://doi.org/10.1002/aenm.201903253. This article may be used for non-commercial purposes in accordance with Wiley Terms and Conditions for Use of Self-Archived Versions. 0 Stack Pressure Considerations for Room Temperature All- Solid-State Lithium Metal Batteries Jean-Marie Douxa,#, Han Nguyena,#, Darren H.S. Tana, Abhik Banerjeea, Xuefeng Wanga, Erik A. Wua, Chiho Joa, Hedi Yanga, Ying Shirley Menga,b. a Department of NanoEngineering, University of California San Diego, La Jolla, CA 92093. b Sustainable Power & Energy Center (SPEC), University of California San Diego, La Jolla, CA 92093. Correspondence to: [email protected] # These authors contributed equally: Jean-Marie Doux, Han Nguyen. ABSTRACT All-solid-state batteries are expected to enable batteries with high energy density with the use of lithium metal anodes. Although solid electrolytes are believed to be mechanically strong enough to prevent lithium dendrites from propagating, various reports today still show cell failure due to lithium dendritic growth at room temperature. While cell parameters such as current density, electrolyte porosity and interfacial properties have been investigated, mechanical properties of lithium metal and the role of applied stack pressure on the shorting behavior is still poorly understood. Here, we investigated failure mechanisms of lithium metal in all-solid-state batteries as a function of stack pressure, and conducted in situ characterization of the interfacial and morphological properties of the buried lithium in solid electrolytes. We found that a low stack pressure of 5 MPa allows reliable plating and stripping in a lithium symmetric cell for more than 1000 hours, and a Li Li6PS5Cl LiNi0.80Co0.15Al0.05O2 full cell, plating more than 4 µm of lithium per charge, is able to cycle over 200 cycles at room temperature. These results suggest the possibility of enabling the lithium metal anode in all-solid-state batteries at reasonable stack pressures. 1 INTRODUCTION All-solid-state batteries (ASSBs) using non-flammable solid electrolytes are attracting increasing interest from their potential to enable the metallic lithium anode, which would dramatically increase energy densities compared to their liquid electrolyte counterparts. This arises from the belief that solid electrolytes serve as a suitable barrier that prevents lithium dendrite propagation.[1 -- 3] The Monroe-Neumann criterion has postulated that a solid-state electrolyte (SSE) with a shear modulus twice that of lithium metal would be suitable to prevent such dendritic propagation.[4,5] However, internal shorting caused by lithium dendrite formation is still prevalent within SSEs that satisfy this criterion. Under ambient conditions, dendrites formed during plating are found to penetrate even near perfectly dense single crystal oxide solid electrolytes (such as Li7La3Zr2O12),[6 -- 8] shorting the battery after a few cycles as a result.[9] Others have attempted to enable lithium metal anodes in ASSBs by increasing electrolyte density,[10] by ensuring a good wetting at the lithium-electrolyte interface,[11] or by using protective coating layers.[12] As a result, most reported literature have continued to use Li-In alloys in ASSBs.[13,14] To address this problem, recent studies have focused on the mechanical properties of lithium metal, seeking to understand the dendrite penetration mechanism within the electrolyte.[15 -- 19] In this context, Masias et al. measured the Young's modulus, the shear modulus and the Poisson's ratio of lithium metal at room temperature.[15] Their findings shows that Li exhibits a yield strength of about 0.8 MPa, in accordance with former work by Tariq et al.,[16] over which the metal starts creeping. This has been confirmed by LePage et al., which showed that at room temperature, the yield strength of Li metal is creep-dominated.[18] This yield strength needs to be correlated with the stack pressure applied on ASSBs during cycling to fully understand the mechanical behavior of a lithium metal anode. Furthermore, contrary to liquid electrolytes where optical techniques can be used to observe the morphology of the plated lithium,[20 -- 23] observing dendrites buried inside a solid-state electrolyte requires the use of more advanced tools. Recently, Heon Kim et al. showed dendrites in Li6PS5Cl electrolyte using in situ Auger electron spectroscopy/microscopy and scanning electron microscopy.[24] However, this requires an open cell setup, utilizing the cross-section of a cell mounted on the in-situ sample holder, which allows lithium to protrude by creeping on the edge during cycling. Seitzman et al. also used synchrotron X-ray microscopy to observe the formation and evolution of voids and lithium dendrites in β-Li3PS4 during plating and stripping.[25] While pressure was applied, its mechanical effects on lithium and dendrite formation was not conclusive. There is still a lack of tools capable of high-resolution morphological imaging combined with chemical species identification within in situ buried interfaces to identify factors causing lithium dendrite formation with solid electrolytes. In this work, we study the influence of the applied stack pressure on the lithium metal anode in ASSBs, employing the argyrodite Li6PS5Cl sulfide electrolyte in a closed cell setup. We first use lithium symmetric cells to determine the necessary conditions to cycle Li metal over extended durations. Then, we demonstrate stable Li metal full cell cycling over 200 cycles at room temperature. High resolution X-ray tomography and X-ray diffraction were used to observe the 2 interfacial and morphological properties of dendrites formed during plating and stripping under higher stack pressures. Finally, we propose a mechanism for the dendrite growth in sulfide solid- state electrolytes based on the mechanical properties of Li metal. RESULTS & DISCUSSION To compare the features of typical Li metal batteries, a solid-state battery comprising Li6PS5Cl as the solid electrolyte and LiNbO3 (LNO) coated LiNi0.80Co0.15Al0.05O2 (NCA) as the cathode was used. This cell was compared against a similar one with Li-In alloy as the anode. Figure 1 shows the voltage profiles of both cells cycled at a rate of C/10 and with a stack pressure of 25 MPa. Unlike their liquid counterparts, the SSE cannot wet new surfaces that are formed during normal battery operation; thus, high pressures are thought to be necessary to ensure consistent interfacial contact between the electrolyte and the cathode. Figure 1: First two cycles of an all-solid-state battery using a Li-In alloy anode (blue) showing typical voltage profiles, and first charge cycle using a Li metal anode (red), showing characteristic shorting behavior. Both cells were prepared in the same conditions and cycled at a stack pressure of 25 MPa. The cell using Li-In alloy as the anode shows an expected cycling voltage behavior with a 1st cycle discharge capacity of approximately 140 mAh.g-1 and a Coulombic efficiency of 66.5%. The low 1st cycle Coulombic efficiency is attributed to initial electrolyte decomposition at the cathode.[26 -- 28] Subsequent cycles present an average Coulombic efficiency over 99% and the cell does not exhibit any shorting behavior. In contrast, the cell using Li metal anode exhibits significant voltage drop during its 1st charge cycle. The voltage then continues to plummet, and ultimately the cell fails to charge. This is consistent with short-circuiting behavior previously observed in the literature.[24,29] The features observed can be attributed to fundamental differences between Li metal and Li-In alloys; including: electrochemical potential, interfacial properties and mechanical 3 properties. Although several studies have characterized the interfacial products formed between sulfide solid electrolytes and Li or Li-In alloys,[30] the differences in their mechanical properties has not been studied yet. As such, we seek to investigate this by studying the effect of stack pressure on lithium metal ASSBs. Figure 2: a) Design of the solid-state Li symmetric cell used, allowing control and monitoring of pressure during cycling. b) Normalized voltage of Li symmetric cells as a function of time during plating and stripping at different stack pressure. At 75 MPa, the cell already mechanically shorts before cycling begins. At 5 MPa, no short was observed for over 1000 hours. c) Voltage profile of a full cell using Li metal anode: the 1st cycle was done at a stack pressure of 5 MPa. The stack pressure was subsequently increased on the same cell to 25 MPa before the 2nd cycle, during which the cell shorts. To investigate the effect of stack pressure on the shorting behavior of Li metal, a load cell has been added to the solid-state cell holder as shown in Figure 2a. The solid-state battery is then pressed 4 between two stainless steel plates, with the bottom end in direct contact with the load cell. The stack pressure can be accurately tuned by tightening the nuts accordingly. Insulating spacers are placed between the titanium current collectors and stainless-steel plates to avoid external short- circuiting. Next, plating and stripping of Li symmetric cells were conducted to determine stack pressure effects on dendrite formation and to determine an optimal operating pressure. Figure 2b shows cells plating and stripping at 75 µA.cm-2, with continuous 1-hour plating / stripping durations until short-circuiting was observed. The cell that was initially pressed to a stack pressure of 75 MPa was observed to have shorted before the plating and stripping test began. This short circuit is therefore determined to occur mechanically and not due to any lithium plating and stripping. Since the relative density of the cold-pressed electrolyte pellets was approximately 82% (Table S2), it is reasonable to expect a connecting network of pores within the electrolyte. Due to the low yield strength of Li metal, creeping under such high pressure allows lithium to flow within the pores, creating an electronic percolation pathway that shorts the cell internally. When the stack pressure is lowered to 25 MPa, the symmetric cell can be cycled for approximately 48 hours before short- circuiting occurs, as indicated by a sudden overpotential drop. It is noteworthy that the cell under 25 MPa only shorts during plating and stripping. The same cell does not short when no current is applied even over prolonged durations, indicating that Li creep-induced shorting does not occur at 25 MPa. Similar tests were conducted at stack pressures of 20, 15 and 10 MPa and similar shorting behavior was observed after 190, 272 and 474 hours, respectively. The overpotentials measured in all cells were constant throughout the entire process, which indicates that stable lithium-Li6PS5Cl interfaces are formed. All these results show that lithium metal shorting behavior is both a mechanical and electrochemical phenomenon; a trend can be observed between stack pressure and the time needed before short-circuiting occurs. However, at a stack pressure of 5 MPa, no short circuit was observed within 1000 hours of plating and stripping when the experiment was stopped. To confirm that this stack pressure could allow room temperature cycling of a Li metal anode, a full cell was constructed and the first cycle at a stack pressure of 5 MPa is shown on Figure 2c. Contrary to the Li anode cell shown previously, this battery shows a typical voltage profile without any short circuit. A specific capacity of 150 mAh.g-1 and a first cycle Coulombic efficiency of 69% was attained, similar to the cell constructed with Li-In. In order to verify that a high stack pressure was the cause of the short circuit in Figure 1, the pressure was increased to 25 MPa before starting the second cycle. As seen on Figure 2c, a small voltage drop is observed during the charging cycle and the cell fails to charge normally. This behavior is typical of Li metal cycling and has been attributed to lithium dendrite formation during plating, generating short circuits.[24,29] These results show that the stack pressure is a crucial parameter to enable cycling of Li metal anodes in all-solid-state batteries. 5 Figure 3: a) Voltage profile of the 1st, 2nd, 5th and 10th cycle and b) cycle life of a Li metal Li6PS5Cl LNO-coated NCA ASSB cycled at C/10 and at a stack pressure of 5 MPa. No shorting behavior was observed. Average Coulombic efficiency over 229 cycles is 98.86% and the cell shows a capacity retention of 80.9% over 100 cycles. The active material loading is 3.55 mg/cm². Figure 3a presents the voltage profile of the 1st, 2nd, 5th and 10th cycle of a Li metal Li6PS5Cl LNO-coated NCA cell cycled at C/10 and with a stack pressure of 5 MPa, at room temperature. This cell shows stable cycling over 229 cycles (Figure 3b) and exhibits a capacity retention of 80.9% over 100 cycles. This demonstrates the feasibility of Li metal anodes in all-solid-state batteries. There are only a few reports of full cells cycling Li metal at room temperature in the literature, a summary is reported in Table S1. Unfortunately, missing experimental details in the reported literature makes the reproduction of these results difficult, and the reported cycles are limited.[24,31,32] 6 Figure 4: Schematic of the effect of the stack pressure on the shorting behavior of Li metal solid- state batteries. a) During cell fabrication, the contact between the electrolyte and Li metal is poor before pressing the Li metal on the electrolyte pellet. b) Pressing the Li metal at 25 MPa allows for proper wetting of the electrolyte and induces a large drop in the symmetric cell impedance, as shown in c), even when the pressure is later released to 5 MPa. d) Plating and stripping at a stack pressure of 5 MPa: no creeping of Li inside the SSE pellet is observed and therefore the cell cycles for more than 1000 hours. e) At a stack pressure of 25 MPa, Li slowly creeps between the grains of the SSE and plating occurs on these dendrites, eventually shorting the cell after 48 hours. f) When the stack pressure is too high, Li creeps through the electrolyte and forms dendrites that mechanically short the cell. In order to understand the effect of the stack pressure on the plating and stripping of lithium in a Li symmetric cell, it is necessary to consider the creeping behavior of Li metal at each step of the cell fabrication and during cell cycling. This is detailed in Figure 4. First, when Li metal is added on both sides of the electrolyte pellet, interfacial contact between the two materials are poor (Figure 4a) and it is necessary to press the Li electrodes at 25 MPa to lower the initial cell impedance. This can be seen physically by the disappearance of voids at the interface between the Li metal and the electrolyte when using a clear polycarbonate pellet die (Figure S1). As shown in Figure 4c, the impedance of a Li metal symmetric cell depends principally of the pressure applied to improve the contact between the electrolyte and the lithium. If a pressure of only 1 MPa is used, the cell impedance exceeds 500 Ω, and this value decreases to ~110 Ω at 5 MPa, ~50 Ω at 10 MPa, 7 ~40 Ω at 15 MPa, 35 Ω at 20 MPa, and 32 Ω at 25 MPa. Upon releasing the pressure to 5 MPa, this cell impedance only goes up to ~50 Ω, which is less than half the initial impedance at the same pressure. This can be explained by the increased wetting between lithium and the electrolyte: a relatively high pressure of 25 MPa allows lithium to creep and conforms to the relatively rough surface on the electrolyte pellet, filling the pores along the interface (Figure 4b). After the fabrication of the cell using a pressure of 25 MPa, three scenarios are encountered: plating and stripping at low stack pressure (5 MPa), at intermediate stack pressure (25 MPa) and at high stack pressure (75 MPa). For each case, we consider the start and end conditions of cell cycling. For low pressure plating and stripping (Figure 4d), stack pressure applied is high enough to allow a good contact of lithium with the electrolyte during the cycling, but not high enough to cause lithium creep through the electrolyte and induce cell shorting. This explains why the cell shown in Figure 2b, when cycling under 5 MPa, was able to plate and strip for more than 1000 hours without any shorting behavior. For a stack pressure of 25 MPa, as shown on Figure 4e, lithium can slowly creep inside the pores of the electrolyte to form dendrites. As the distance between the two electrodes is reduced by these small protuberances of lithium, they become the preferred sites for plating lithium due to a slightly lower overpotential experienced. Therefore, after 48 hours of plating and stripping, dendrites develop and the cell shorts, as shown previously in Figure 2b. Merely applying a stack pressure of 25 MPa (without any plating and stripping) did not induce any shorting, indicating that plating and stripping is necessary to form dendrites at this pressure. Finally, when applying a stack pressure of 75 MPa (Figure 4f), lithium creeps through the pores of the electrolyte and the cell is shorted even before any plating and stripping starts. A porosity of 18% within the electrolyte provides connecting pathways across both electrodes, allowing lithium creep that shorts the cell. It is noted that a 75MPa stack pressure is much larger (around 100 times) than the yield strength of lithium metal. The lowest stack pressure used (5 MPa) is still high compared to the yield strength of lithium metal (0.8 MPa) and lithium creeping could therefore be expected. Nevertheless, this yield strength value has been measured in tension, and when working in compression, the increase of the surface of contact causes the stress to gradually increase because of the friction forces.[15] A similar behavior can prevent the creeping of lithium in the ASSB at a stack pressure of 5 MPa. For the three symmetric cell stack pressures of 5, 25, and 75 MPa, mechanical properties of the SSE pellet itself are not expected to have an influence on the shorting mechanism, as it has already been cold pressed at 370 MPa. At 75 MPa, lithium creeps through the pellet via interconnecting pores to ultimately create an electronic pathway. At 25 MPa, some lithium initially creeps into the pores but is insufficient to cause electronic short. With plating and stripping, additional force is exerted by the plated lithium deposited along the electrolyte grains, expanding the lithium filaments in the pellet until short-circuit occurs. At 5 MPa, plating of lithium only takes place on the surface of the pellet as the pressure is not high enough to allow lithium to creep into the pores. 8 Figure 5: Schematic of the cell used for X-ray tomography and X-ray diffraction, profile matching of the X-ray diffraction patterns, and X-ray tomography of a Li Li6PS5Cl Li symmetric cell cycled under a stack pressure of 25 MPa a) before plating and stripping and b) after shorting. Before plating and stripping, only Li6PS5Cl is detected in the electrolyte, and Li metal is present on both sides. The tomography pictures confirm that no lithium is present in the electrolyte. After shorting, several additional phases are detected inside the electrolyte: mainly Li2S, LiCl, P4 and Li3P8, all being components of the SEI formed when Li is in contact with Li6PS5Cl. Tomography pictures shows that large quantity of low-density dendrites have been formed in the electrolyte. Using a combination of laboratory X-ray tomography in conjunction with X-ray diffraction on the same solid-state Li symmetric cells in situ allowed us to obtain both morphological and chemical information of the buried dendrites. The cell was constructed with a diameter of 2 mm to allow a resolution of about 1 µm over the whole volume of the electrolyte with the X-ray tomography experiments. The use of Mo Kα radiation provided the X-ray diffraction patterns of the lithium metal and the electrolyte before and after plating and stripping. Figure 5a shows the X-ray tomography image of the cell before plating and stripping and XRD patterns in the lithium metal region and in the electrolyte region. The lithium metal electrodes show good contact with the electrolyte pellet; the contact interface is flat and without any voids. As expected in this pristine state, only Li6PS5Cl is present in the electrolyte region, and the lithium metal diffraction peaks can be detected in the electrode region. After plating and stripping at 25 MPa, as shown on Figure 5b, the tomography images show large low-density structures within the electrolyte. These dendritic formations propagate between the grains of the electrolyte along the grain boundaries and then expand within the local sites. X-ray diffraction of the electrolyte shows the presence of numerous phases: LiCl, Li2S, and reduced phosphorous species which are harder to identify because of their low concentrations. Such species have been previously identified in the literature to be the SEI formed between Li metal and Li6PS5Cl.[30] It is important to note that lithium metal dendrites are 9 not directly detected by X-ray diffraction due to the low amounts and low scattering efficiency of lithium metal in comparison to the electrolyte and SEI products formed. Both in situ tomography and diffraction experiments conducted on the same cell offers direct observation of lithium dendrite growth and its corresponding interface products within the solid electrolyte. This is consistent with the proposed mechanism of cell shorting seen with electrochemical measurements discussed earlier. Although there are recent reports of void formation during stripping metallic lithium in ASSB (at 3.5 and 7 MPa) due to limited lithium -- SSE contact, this issue was mitigated by improving this contact by an higher initial pressure (25 MPa) step followed by release at the working stack pressure (5 MPa). We believe that this initial high pressure step allows a homogeneous plating and stripping without formation of voids at the interface as no voids were observed by X-ray microscopy on our samples.[33] Conclusion In summary, the effect of stack pressure on the lithium metal anode in an all-solid-state battery was investigated. While stack pressure is needed to provide good initial contact between the electrolyte and the lithium by preventing the apparition of voids, a higher stack pressure can either short a cell immediately (75 MPa) or after a relatively short time of plating and stripping (25 MPa). We found that the ductility of lithium metal (due to its low stress yield) allows it to creep through the electrolyte's pores. To avoid this, a range of cycling stack pressures were studied and an optimal pressure of 5 MPa was found to allow long-term cycling of lithium metal in an all-solid- state battery. This was demonstrated in a full cell of Li Li6PS5Cl NCA which cycled at room temperature for more than 200 cycles without cell failure from dendrite formation. This work paves the way toward room temperature lithium metal ASSBs and helps shed light on the importance and role of stack pressure in preventing cell failure in ASSBs. 10 ASSOCIATED CONTENT Supporting Information The Supporting information contains: Experimental Details, Picture of the contact improvement after pressing the Li metal anode at 25 MPa, Literature Li metal anode ASSBs summary table, Relative density of electrolyte pellets table. AUTHOR INFORMATION Corresponding Author (Y.SM.) E-mail: [email protected] ORCID: Ying Shirley Meng: 0000-0001-8936-8845 Notes ◼ ACKNOWLEDGEMENTS This study was financially supported by the LG Chem company through Battery Innovation Contest (BIC) program. The authors would like to acknowledge UCSD Crystallography Facility. This work was performed in part at the San Diego Nanotechnology Infrastructure (SDNI) of UCSD, a member of the National Nanotechnology Coordinated Infrastructure, which is supported by the National Science Foundation (Grant ECCS-1542148). The authors would like to acknowledge the National Center for Microscopy and Imaging Research (NCMIR) technologies and instrumentation are supported by grant P41GM103412 from the National Institute of General Medical Sciences. 11 REFERENCES [1] A. Ferrese, J. Newman, J. Electrochem. Soc. 2014, 161, A1350. [2] Y. Guo, H. Li, T. Zhai, Adv. Mater. 2017, 29, 1700007. [3] L. Fan, S. Wei, S. Li, Q. Li, Y. Lu, Adv. Energy Mater. 2018, 8, 1702657. [4] C. Monroe, J. Newman, J. Electrochem. Soc. 2003, 150, A1377. [5] C. Monroe, J. Newman, J. Electrochem. Soc. 2005, 152, A396. [6] L. Porz, T. Swamy, B. W. Sheldon, D. Rettenwander, T. Frömling, H. L. Thaman, S. Berendts, R. Uecker, W. C. Carter, Y.-M. Chiang, Adv. Energy Mater. 2017, 7, 1701003. [7] Y. Ren, Y. Shen, Y. Lin, C.-W. Nan, Electrochem. Commun. 2015, 57, 27. [8] T. Swamy, R. Park, B. W. Sheldon, D. Rettenwander, L. Porz, S. Berendts, R. Uecker, W. C. Carter, Y.-M. Chiang, J. Electrochem. Soc. 2018, 165, A3648. [9] F. Aguesse, W. Manalastas, L. Buannic, J. M. Lopez del Amo, G. Singh, A. Llordés, J. Kilner, ACS Appl. Mater. Interfaces 2017, 9, 3808. [10] Y. Kim, H. Jo, J. L. Allen, H. Choe, J. Wolfenstine, J. Sakamoto, J. Am. Ceram. Soc. 2016, 99, 1367. [11] A. Sharafi, E. Kazyak, A. L. Davis, S. Yu, T. Thompson, D. J. Siegel, N. P. Dasgupta, J. Sakamoto, Chem. Mater. 2017, 29, 7961. [12] K. (Kelvin) Fu, Y. Gong, B. Liu, Y. Zhu, S. Xu, Y. Yao, W. Luo, C. Wang, S. D. Lacey, J. Dai, Y. Chen, Y. Mo, E. Wachsman, L. Hu, Sci. Adv. 2017, 3, e1601659. [13] A. L. Santhosha, L. Medenbach, J. R. Buchheim, P. Adelhelm, Batteries Supercaps 2019, batt. 201800149. [14] B. Wu, S. Wang, W. J. Evans IV, D. Z. Deng, J. Yang, J. Xiao, J. Mater. Chem. A 2016, 4, 15266. [15] A. Masias, N. Felten, R. Garcia-Mendez, J. Wolfenstine, J. Sakamoto, J. Mater. Sci. 2019, 54, 2585. [16] S. Tariq, K. Ammigan, P. Hurh, R. Schultz, P. Liu, J. Shang, in Proceedings of the 2003 Particle Accelerator Conference, 2003, pp. 1452 -- 1454 vol.3. [17] C. Campbell, Y. M. Lee, K. Y. Cho, Y.-G. Lee, B. Lee, C. Phatak, S. Hong, Sci. Rep. 2018, 8, 2514. [18] W. S. LePage, Y. Chen, E. Kazyak, K.-H. Chen, A. J. Sanchez, A. Poli, E. M. Arruda, M. D. Thouless, N. P. Dasgupta, J. Electrochem. Soc. 2019, 166, A89. [19] J. H. Cho, X. Xiao, K. Guo, Y. Liu, H. Gao, B. W. Sheldon, Energy Storage Materials 2019, DOI: 10.1016/j.ensm.2019.08.008. 12 [20] J. Steiger, D. Kramer, R. Mönig, Electrochim. Acta 2014, 136, 529. [21] J. Steiger, D. Kramer, R. Mönig, J. Power Sources 2014, 261, 112. [22] J. Steiger, G. Richter, M. Wenk, D. Kramer, R. Mönig, Electrochem. Commun. 2015, 50, 11. [23] K. N. Wood, M. Noked, N. P. Dasgupta, ACS Energy Lett. 2017, 2, 664. [24] S. Heon Kim, K. Kim, H. Choi, D. Im, S. Heo, H. Soo Choi, J. Mater. Chem. A 2019, 7, 13650. [25] N. Seitzman, H. Guthrey, D. B. Sulas, H. A. S. Platt, M. Al-Jassim, S. Pylypenko, J. Electrochem. Soc. 2018, 165, A3732. [26] J. Auvergniot, A. Cassel, D. Foix, V. Viallet, V. Seznec, R. Dedryvère, Solid State Ionics 2017, 300, 78. [27] J. Auvergniot, A. Cassel, J.-B. Ledeuil, V. Viallet, V. Seznec, R. Dedryvère, Chem. Mater. 2017, 29, 3883. [28] D. H. S. Tan, E. A. Wu, H. Nguyen, Z. Chen, M. A. T. Marple, J.-M. Doux, X. Wang, H. Yang, A. Banerjee, Y. S. Meng, ACS Energy Lett. 2019, DOI: 10.1021/acsenergylett.9b01693. [29] Y. S. Jung, D. Y. Oh, Y. J. Nam, K. H. Park, Isr. J. Chem. 2015, 55, 472. [30] S. Wenzel, S. J. Sedlmaier, C. Dietrich, W. G. Zeier, J. Janek, Solid State Ionics 2018, 318, 102. [31] J. E. Trevey, J. R. Gilsdorf, S. W. Miller, S.-H. Lee, Solid State Ionics 2012, 214, 25. [32] R. Xu, J. Yue, S. Liu, J. Tu, F. Han, P. Liu, C. Wang, ACS Energy Lett. 2019, 4, 1073. [33] J. Kasemchainan, S. Zekoll, D. S. Jolly, Z. Ning, G. O. Hartley, J. Marrow, P. G. Bruce, Nat. Mater. 2019, 1. 13 Supporting Information Stack Pressure Considerations for Room Temperature All- Solid-State Lithium Metal Batteries Jean-Marie Douxa, Han Nguyena, Darren H.S. Tana, Abhik Banerjeea, Xuefeng Wanga, Erik A. Wua, Chiho Joa, Hedi Yanga, Ying Shirley Menga,b. a Department of NanoEngineering, University of California San Diego, La Jolla, CA 92093. b Sustainable Power & Energy Center (SPEC), University of California San Diego, La Jolla, CA 92093. *Correspondence to: [email protected] This SI file includes: Supplementary text Fig. S1 to S6. Tables S1, S2. 14 Material & Methods All materials synthesis, processing, and testing was conducted in an Argon-filled glovebox (mBraun MB 200B) with H2O and O2 levels below 0.1 and 0.5 ppm, respectively. Synthesis Li6PS5Cl was purchased and used as is from NEI Corporation (USA). Li -- Indium (Li-In) alloy was prepared by mixing stabilized lithium nano powder (FMC) with indium powder (Sigma Aldrich, 99.99%) with a vortex mixer until the mixture was homogeneous. The cathode LiNi0.80Co0.15Al0.05O2 (NCA, Toda Chemicals) was surface modified with an amorphous 2 wt. % LiNbO3 (LNO) coating using a wet chemical method: NCA powder was added into an ethanol solution containing Li ethoxide (Sigma Aldrich) and Nb ethoxide (Sigma Aldrich) followed by stirring for an hour. The ethanol was dried in a rotary evaporator and then the powder was heated in air at 450 C in a box furnace (Lindberg Blue M) for one hour. Electrochemical measurements All-solid-state batteries were prepared using custom-made titanium plungers and a polyether ether ketone (PEEK) die mold. 200 mg of electrolyte powder was loaded into the 13 mm PEEK die and compacted at a pressure of 370 MPa using a hydraulic press. The obtained electrolyte pellet, with a thickness of ~1 mm, has an ionic conductivity measured between 2 and 2.5 mS/cm by Electrochemical Impedance Spectroscopy (EIS). For the Li symmetric cells, Li foil (0.5 mm thick, FMC) was cleaned with a brush to remove the surface oxidation. Li discs 12.7 mm in diameter were then cut and pressed on both side of the electrolyte pellet at a controlled pressure of 25 MPa. The cell stack pressure was then set to prescribed values of 5, 10, 15, 20, 25, 75 MPa during the plating and stripping test. Plating and stripping experiments were carried out using a 1-hour plating and stripping step, at a current of 75 µA.cm-2. For the full cells, a cathode composite was prepared with the LNO-coated NCA, Li6PS5Cl, and carbon black, in a 11:16:1 weight ratio. This mixture was mixed with an agate mortar and pestle. 12 mg of the cathode composite was then pressed on one side of the electrolyte pellet at a pressure of 370 MPa. For the anode, either a Li metal disc (12.7 mm diameter, 25 MPa) or Li-In powder (70 mg, 120 MPa) was pressed on the other side of the electrolyte pellet. Cells were cycled at a rate of C/10 between 2.5 and 4.3V vs. Li/Li+. All cells were cycled inside an Ar-filled glovebox using Landhe battery cyclers. Pressure monitoring and control Control of the stack pressure applied to the solid-state batteries during cycling was conducted with a special cell holder (detailed in Figure 1). A 3-ton load cell is mounted at the bottom of the holder, in the axis of the battery, and allows measurement of the pressure between 0 and 220 MPa. The 15 load cell was calibrated using a 100 kN Instron 5982 Universal Testing System mechanical testing frame by applying a known load at 100% of the capacity of the load cell. Linearity of the calibration was then verified over the whole range of the load cell. X-ray Tomography For the X-ray tomography measurements, a specially designed cell was used to allow higher resolution (i.e. small voxel size). A 3.2 mm diameter polycarbonate rod was bored at 2 mm internal diameter and 2 mm steel plungers were used to fabricate and cycle the cell. Around 5 mg of electrolyte (Li6PS5Cl) was pelletized at a pressure of 370 MPa, corresponding to a pellet thickness of around 2 mm. Lithium metal strips were then cut from a cleaned Li foil and pressed on both sides of the pellet at 25 MPa. Enough Li metal was used to ensure that there was more than 5 mm of Li on each side of the cell in order to prevent artifacts caused by the high absorption of the steel plungers on the X-ray microscopy pictures. X-ray computer-assisted tomography was done with a Versa 510 (Zeiss/Xradia) X-ray microscope, with a source voltage of 80 kV and a power of 6.5 W, using the LE2 filter. A 4X objective was used with an exposure of 0.5 s and 1601 projections, giving a resolution of about 1.18 µm. Analysis of the reconstruction data was performed using Amira 2019.1 (ThermoFisher Scientific) and Fiji software.[1,2] X-ray Diffraction X-ray diffraction experiments were performed on the solid-state Li symmetric cells used for the X-ray tomography experiments. As Cu Kα radiation are not energetic enough to work in transmission mode with this cell design, Mo Kα radiation (λ = 0.709320 Å) was used. A Bruker Apex II Ultra diffractometer was used in transmission mode equipped with a 2D CCD detector. Refinement of the diffraction patterns were carried out in profile matching mode using the FullProf software suite.[3] 16 SI figures Figure S1: Pictures of a Li metal Li6PS5Cl Li metal symmetric cell before (top) and after (bottom) pressing at a pressure of 25 MPa. The contact improvement between the Li metal and the electrolyte pellet can be seen by the disappearance of voids at the interface. Figure S2: Nyquist diagram of a Li symmetric cell before and after plating and stripping for 10 and 20 hours, at a stack pressure of 25 MPa. 17 Figure S3: Potential of a Li symmetric cell during plating and stripping at a stack pressure of 2 MPa. Figure S4: Voltage profiles of the Li symmetric cells during plating and stripping at different stack pressures. 18 Figure S5: X-ray tomography of a Li symmetric cell before (a) and after (b) 92 hours of plating and stripping at a stack pressure of 5 MPa. c) Voltage profile during the plating and stripping at 5 MPa. Figure S6: X-ray tomography of a Li symmetric cell mechanically shorted after pressing Li metal at 75 MPa. 19 Table S1: Summary of the reported Li metal cycling in ASSBs in the literature. Cathode Electrolyte Number of Charge/Discharge Cycles Rate Current Density Capacity retention Reference LiNi0.8Mn0.05Co0.15O2 Li6PS5Cl LiCoO2 Li2S-Li2O-P2S5 Li3PS4 Li2S 100 25 100 0.22C / 0.35C / 74% - 100 cycles / 50 µA/cm² ~50% - 25 cycles 0.2C / 0.2C / ~79% - 100 cycles [5] [6] [7] Table S2: Relative density calculation based on physical measurements of four Li6PS5Cl electrolyte pellets, using a theoretical density value of 1.860 g.cm-3.[4] Mass (mg) 200.0 201.2 200.3 200.3 Diameter (mm) 13 13 13 13 Average Thickness Exp. Density Relative Density (mm) (g.cm-3) 1.01 0.96 1.01 0.99 1.492 1.579 1.494 1.524 1.522 (%) 80.2 84.9 80.3 83.0 82.1 REFERENCES [1] J. Schindelin, I. Arganda-Carreras, E. Frise, V. Kaynig, M. Longair, T. Pietzsch, S. Preibisch, C. Rueden, S. Saalfeld, B. Schmid, J.-Y. Tinevez, D. J. White, V. Hartenstein, K. Eliceiri, P. Tomancak, A. Cardona, Nat. Methods 2012, 9, 676. [2] C. T. Rueden, J. Schindelin, M. C. Hiner, B. E. DeZonia, A. E. Walter, E. T. Arena, K. W. Eliceiri, BMC Bioinf. 2017, 18, 529. [3] J. Rodríguez-Carvajal, Phys. B 1993, 192, 55. [4] M. A. Kraft, S. P. Culver, M. Calderon, F. Böcher, T. Krauskopf, A. Senyshyn, C. Dietrich, A. Zevalkink, J. Janek, W. G. Zeier, J. Am. Chem. Soc. 2017, 139, 10909. [5] S. Heon Kim, K. Kim, H. Choi, D. Im, S. Heo, H. Soo Choi, J. Mater. Chem. A 2019, 7, 13650. [6] J. E. Trevey, J. R. Gilsdorf, S. W. Miller, S.-H. Lee, Solid State Ionics 2012, 214, 25. [7] R. Xu, J. Yue, S. Liu, J. Tu, F. Han, P. Liu, C. Wang, ACS Energy Lett. 2019, 4, 1073. 20
1910.01670	1	1910	2019-10-03T18:13:00	Micro-Computed Tomography Analysis of Damage in Notched Composite Laminates Under Multi-Axial Fatigue	[ "physics.app-ph" ]	The broad application of polymer composites in engineering demands the deep understanding of the main damage mechanisms under realistic loading conditions and the development of proper physics-based models. Towards this goal, this study presents a comprehensive characterization of the main damage mechanisms in a selection of notched composite structures under multiaxial fatigue loading. Thanks to a synergistic combination of X-ray micro-computed tomography ($\mu$-CT) and Digital Image Correlation (DIC), the main failure modes are identified while the crack volume associated to each mechanism is characterized. This study provides unprecedented quantitative data for the development and validation of computational models to capture the fatigue behavior of polymer composite structures.	physics.app-ph	physics	A&A Program in Structures William E. Boeing Department of Aeronautics and Astronautics University of Washington Seattle, Washington 98195, USA Micro-Computed Tomography Analysis of Damage in Notched Composite Laminates Under Multi-Axial Fatigue Yao Qiao, Marco Salviato INTERNAL REPORT No. 19-10/04E Submitted to Composites Part B: Engineering October 2019 Micro-Computed Tomography Analysis of Damage in Notched Composite Laminates Under Multi-Axial Fatigue Yao Qiaoa, Marco Salviatoa,∗ aWilliam E. Boeing Department of Aeronautics and Astronautics, Guggenheim Hall, University of Washington, Seattle, Washington 98195-2400, USA Abstract The broad application of polymer composites in engineering demands the deep understanding of the main damage mechanisms under realistic loading conditions and the development of proper physics-based models. Towards this goal, this study presents a comprehensive characterization of the main damage mechanisms in a selection of notched composite structures under multiaxial fatigue loading. Thanks to a synergistic combination of X-ray micro-computed tomography (µ-CT) and Digital Image Correlation (DIC), the main failure modes are identified while the crack volume associated to each mechanism is characterized. This study provides unprecedented quantitative data for the development and validation of computational models to capture the fatigue behavior of polymer composite structures. Keywords: Fatigue, Fracture, DIC, µ-CT, Damage mechanism 1. Introduction The broad application of polymer composites in engineering requires the insightful understanding of their damage mechanisms both under quasi-static and fatigue loading. These mechanisms affect the mechanical behavior of composites significantly [1 -- 5] and lead to important size effects that must be carefully considered in structural design [6 -- 14]. Towards this important goal, a range of non-destructive inspection techniques has been developed and applied in the last few decades. The most attractive techniques include acoustic emission [15 -- 17], infrared thermography [18 -- 21], ultrasonic C-scan [22], Digital Imaging Correlation [23, 24] and X-ray micro-computed tomography [25 -- 27]. An analysis to combine the mechanical behavior of the materials with the damage evaluation via the forgoing useful tools promotes a better understanding on the fracturing behavior of composite structures under a variety of loading conditions. This is a condition of utmost importance for the development of safe and reliable designs and certification guidelines. Among these techniques, acoustic emission method is widely used for monitoring damage initiation. An excellent investigation on the initiation of transverse cracking and delamination in textile composite under uni-axial tensile test was provided by Lomov et al. [17]. Similar studies were reported on the unidirectional and woven composites by Silversides et al. [16]. ∗Corresponding Author, Email address: [email protected] (Marco Salviato) Preprint submitted to Composites Part B October 7, 2019 However, infrared thermography and ultrasonic C-scan are typically employed to have a rough estimation of the damage progression in composite structures. A noticeable example was provided by Meola and Carlomagno [19] who evaluated impact damage in Glass Fiber Reinforced Polymer (GFRP) leveraging the thermographic technique. Thanks to this method, the impact damage progression throughout the thickness of the laminates was successfully reconstructed. By conducting an ultrasonic inspection on the damaged specimens, Scarponi and Briotti [22] identified significant development of delamination in thermoset polymer reinforced by various types of fibers. Although the forgoing techniques led to a remarkable progress in the damage characterization, the identi- fication of individual damage types is still not clear since the damage analyzed using the forgoing inspection tools is usually homogenized. In this context, microscopic techniques (Optical and Scanning Electron Mi- croscopy) are extensively used to visualize different damage types in specimens. Carraro et al. [28], for instance, conducted an optical microscopic inspection which was capable of capturing the delamination in- duced by transverse cracking in cross-ply laminates under tensile fatigue loading. On similar grounds, a micro-damage study leveraging Scanning Electron Microscopy (SEM) was provided by Nguyen et al. [29] who identified shear hackles between the fibers in specimens featuring [+45/−45]4s layup under un-axial tensile test. However, these techniques can only provide a local visualization of a detailed damage morphol- ogy since the specimens must be cut into a small portion which can be properly operated by microscopic techniques. Thanks to the emergence and development of X-ray micro-computed tomography (µ-CT), a detailed damage morphology throughout the entire composite structure can be obtained even under an in-situ in- spection test. An interesting study was provided by Yu et al. [26] who investigated the damage evolution in 3D woven composite under tensile fatigue loading. It was concluded that the dominating damage mechanism before final failure is the debonding cracks spreading along the interface between binder yarns and matrix. By leveraging µ-CT technique, the growth of different type of damage was also successfully quantified as a function of fatigue lifetime. A further step towards an even more insightful damage analysis requires a synergistic approach combing µ-CT technique and other aforementioned methods. In this way, this work presents a comprehensive damage analysis via X-ray micro-computed tomography and Digital Imaging Correlation techniques on notched quasi-isotropic [+45/90/−45/0]s and cross-ply [0/90]2s laminates under multi-axial quasi-static and fatigue loading. Thanks to the synergistic approach, a detailed analysis of the damage mechanisms and quantification of damage evolution was achieved. This result, which is of utmost importance for the development and validation of fatigue computational models for notched composite structures under multiaxial loading, has never been obtained before. 2 2. Materials and Methods 2.1. Material Preparation Following [30], quasi-isotropic [+45/90/−45/0]s and cross-ply [0/90]2s laminates made of a Glass Fiber Reinforced Polymer (GFRP) by Mitsubishi Composites [31] were investigated in this work. The material system included 7781 unidirectional E-glass fiber and NB301 epoxy resin leading to a 0.28 mm lamina with 68% fiber volume fraction. The glass transition temperature for epoxy resin is about 120◦C. The laminates were manufactured from prepreg sheets which were hand laid-up and then vacuum-bagged using a Vacmobile mobile vacuum system [32]. A Despatch LAC1-38A programmable oven was used to cure the panels by ramping up the temperature from room temperature to 135◦C in one hour, soaking for one hour, and cooling down to room temperature. 2.2. Test Method The application of the combinations of nominal normal and shear stresses on notched laminates was achieved by using an Arcan rig (Fig. 1a). The multi-axial quasi-static and fatigue tests were performed in a servo-hydraulic 8801 Instron machine with closed-loop control. The load control with a stress ratio of R = 0.1 and a low frequency of f = 5 Hz was conducted for the fatigue tests. To describe the multi-axial loading configuration, multiaxiality ratio can be defined as λ = arctan(τN /σN ) where σN = P cosθ/[(w−a0)t] is the nominal normal stress and τN = P sinθ/[(w− a0)t] is the nominal shear stress applied to the specimen. In the definition of stress, P is the instantaneous load, w is the specimen width, a0 is the crack length or hole diameter, t is the specimen thickness and θ is the angle between loading direction and axial direction of the specimen. A detailed description on the test method and loading conditions can be found in [30]. 2.3. Specimen preparation To study the damage evolution of notched laminates under multi-axial tests, specimens with the same dimensions but featuring different types of notches were prepared as shown in Figure 1. The detailed description of the manufacturing of these notches can be found in the previous work by Qiao et al. [30]. The procedure of damage evaluation leveraging Digital Image Correlation (DIC) and X-ray micro-computed tomography (µ-CT) is illustrated in Figure 1 and described in the following sections. 2.3.1. Digital Image Correlation (DIC) The strain distribution of notched laminates was investigated for three multiaxiality ratios (λ = 0, 0.785 and 1.571) by using an open source DIC system programmed in MATLAB software developed at the Georgia Institute of Technology [33, 34]. To this end, a recording of videos with a frame rate of 30 fps for speckled specimens was analyzed at roughly 97% of total quasi-static or fatigue life. The total quasi-static life corresponded to its final catastrophic failure whereas the percentage of fatigue life corresponded to its performance at the peak load of applied cyclic load (55% of quasi-static critical load). 3 2.3.2. X-ray micro-computed tomography (µ-CT) The sub-critical damage of notched laminates was studied in detail by means of a NSI X5000 X-ray micro-tomography scanning system [35] with a X-ray tube setting of 90 kV in voltage and 220 µA in current. To have a better understanding of the damage evolution under multi-axial tests, damaged specimens taken from the multi-axial quasi-static tests were analyzed at 90% and even later stage of the entire life whereas the ones taken from the multi-axial fatigue tests were analyzed at 40% and 70% of the entire life. It is worth mentioning here that the total quasi-static life corresponded to its life reaching the critical load whereas the total fatigue life corresponded to its lifetime to failure. Prior to the scanning, a dye penetrant composed of zinc iodide powder (250 g), isopropyl alcohol (80 ml), Kodak photo-flow solution (1 ml) and distilled water (80 ml) was used as a supplement to improve the visualization of the damage mechanisms [36, 37]. The specimens were soaked in the dye penetrant mixture for approximately one day. The sub-critical damage in each ply and interface were identified by slicing through the reconstructed 3D images of the specimens via the software ImageJ [38]. 3. Digital Image Correlation (DIC) Analysis The sub-critical, distributed damage occurring under fatigue loading of notched composite structures generally leads to a local reduction of the material stiffness with consequent increase of the local strain. Ac- cordingly, to have a qualitative idea of the overall damage distribution notched [0/90]2s and [+45/90/−45/0]s specimens were analyzed through DIC as described in section 2.3.1. As illustrated in Figure 2 which compares the maximum principal strain distribution of the quasi-static and fatigue loading, the overall damage distribution under fatigue loading appears to be more diffuse com- pared to the quasi-static case. A similar phenomenon was found by Fujii et al. [39 -- 41] who tested tabular specimens featuring a circular hole under the multi-axial fatigue loading of tension and torsion. This result is an indication of a larger non-linear Fracture Process Zone (FPZ) in the specimens subject to fatigue loading. It is worth mentioning that the maximum principle strains were normalized against their maximum values in the area of interest in order to have a comparison between quasi-static and fatigue loading condition. In fact, the focus was not on the magnitude of the strain field but the overall damage distribution. However, the critical magnitude in the maximum principle strain field for quasi-isotropic specimens featuring a central hole under multi-axial tests provides an insightful information on the quasi-static and fatigue fracturing behavior. As can be noted in Figure 3, the critical magnitude of the maximum principle strain in fatigue case is significantly lower than the one in quasi-static case almost before catastrophic failure. This result is an indication of substantially different failure criteria for quasi-static and fatigue case which is typically reported in fatigue fracturing behavior of thermoset polymers showing that the Mode I critical stress intensity factor in fatigue case is generally lower than the one in quasi-static case [42 -- 44]. On the other hand, before moving to the following section on the investigation of damage mechanisms via µ-CT technique, the fracturing features can be roughly known based on the overall strain distribution 4 leveraging DIC. As can be noted in Figure 2, the region of high deformation for notched laminates at various multiaxiality ratios does not show significant differences in terms of shape and location for quasi-static and fatigue case. For notched cross-ply laminates as shown in Figure 2a, the specimen under pure tension is characterized by a region of high deformation in 0◦ plies at the notch tip whereas the one under pure shear features a noticeable damage band in 90◦ plies ahead of the notch. This corresponds to splitting and fiber kinking for pure tension and pure shear case respectively. For notched quasi-isotropic laminates as shown in Figures 2b-c, the specimens at various multiaxiality ratios are mainly characterized by a highly-deformed region in +45◦ plies in front of the notch which is associated to significant splitting. In addition to this, a highly-deformed region due to shear strain almost perpendicular to the +45◦ plies was observed for the specimens under shear-dominated loading as shown in Figure 4. This indicates that the +45◦ plies are under compression leading to the potential micro-buckling of the fibers. 4. Quantitative analysis of the damage mechanisms To shed more light on the characteristics of the fracturing morphology of notched laminates under multi- axial quasi-static and fatigue loading, damaged specimens were analyzed leveraging X-ray µ-CT using the parameters described in the section 2.3.2. This non-destructive technique was particularly important to guarantee that no additional damage was created during the damage visualization process. Thanks to this technique, the sub-critical damage can be observed through the reconstructed 3D images in order to have a quantitative comparison on the crack volume and delamination area of the specimens under multi-axial tests. With the supplement of the dye penetrant, the sub-critical damage can be easily visualized as illustrated in Figures 6-14. Thanks to the great contrast provided by the use of the dye penetrant, the analysis of the grey- scale value can be used to roughly estimate the crack volume and delamination area of the specimen. Figure 5a shows an example of such analysis for a cross-ply specimen featuring a central crack under the quasi-static loading in pure tension. In this plot, the lower grey-scale values represent the sub-critical damage in the gauge volume of the specimen while the higher grey-scale values, constituting the most part of the percentage of pixels, characterize the remaining part of the gauge volume of the specimen. It is worth mentioning here that there are roughly 90 reconstructed images throughout the thickness of the specimen leading to about 2.5 × 107 pixels for the gauge volume of the specimen. To quantify the sub-critical damage, as illustrated in Figure 5b, the cumulative density function of grey-scale distribution up to a range of grey-scale end was treated as crack volume to provide a safe estimation. This range was based on a reference value with the upper and lower bounds of 5 grey-scale values while this reference value was about 144 to 179 for different specimens and reasonably taken on the location of the specimen where no significant damage was observed. 4.1. [0/90]2s laminate with a central crack 4.1.1. Fatigue loading condition (a) Mechanism A 5 In the case of specimens subjected to tension-dominated loading conditions (λ = 0, 0.262), the dominant mechanisms before sudden failure are the splitting in 0◦ plies at the notch tip and the splitting in 90◦ plies. The final failure is triggered by the additional splitting in 0◦ plies away from the notch tip and the breakage of 0◦ fibers. As illustrated in Figure 6, the damage evolution as a function of fatigue lifetime for mechanism A can be summarized as follows: (1) splitting initiates in 90◦ plies (Fig. 6D); (2) splitting initiates in 0◦ plies at notch tip (Fig. 6D); (3) splitting develops in the corresponding piles (Fig. 6E); (4) a small amount of delamination between 0◦ and 90◦ plies starts to initiate (Fig. 6E); (5) additional splitting in 0◦ plies away from the notch tip happens simultaneously with the breakage of 0◦ fibers (Fig. 6F). (b) Mechanism B When the multiaxiality ratio λ = 0.785, 1.309, a mix of nominal normal and shear stresses is applied on specimens and failure behavior follows Mechanism B. In this case, the dominant mechanism transitions from splitting to a combination of delamination and splitting. The final failure happens with the growth of splitting in 0◦ plies at the notch tip and the delamination between 0◦ and 90◦ plies. As illustrated in Figure 7, the damage evolution of mechanism B can be summarized by the following steps: (1) splitting initiates in 90◦ plies (Fig. 7D); (2) splitting initiates in 0◦ plies at notch tip simultaneously with the formation of delamination between 0◦ and 90◦ plies (Fig. 7D); (3) splitting and delamination develop in/between the corresponding plies (Fig. 7E); (4) fibers in 0◦ plies start to kink due to the shear stress (Fig. 7E); (5) delamination between 0◦ and 90◦ plies grows together with the significant splitting at notch tip (Fig. 7E-F). (c) Mechanism C When the specimens are only subjected to pure shear loading, the dominant mechanisms are the delam- ination between 0◦ and 90◦ plies and then fiber kinking in 0◦ plies. The final failure is due to the unstable growth of the inter-laminar crack enabling the splitting in 0◦ plies at notch tip. As illustrated in Figure 8, the damage evolution of mechanism C consists of the following phases: (1) a small amount of splitting initiates in 90◦ plies at notch tip (Fig. 8D); (2) delamination initiates between 0◦ and 90◦ plies (Fig. 8D); (3) fibers in 0◦ plies start to kink significantly due to the shear stress (Fig. 8E); (4) delamination develops between the corresponding plies (Fig. 8E); 6 (5) delamination grows dramatically and drives the significant splitting in 0◦ plies at notch tip (Fig. 8E-F). 4.1.2. Quantitative analysis: fatigue vs. quasi-static loading The crack volume and delamination area as a function of the percentage life were plotted in Figures 15a-d for specimens under quasi-static and fatigue loading. As illustrated in Figure 15c, in the fatigue case, the specimens following mechanism B and C feature larger delamination area compared to the specimens following mechanism A at 70% of total fatigue life, with 1.6% for λ = 0.785 and 2% for λ = 1.571 but only 1% for λ = 0. This is an indirect evidence of the foregoing damage mechanisms showing that delamination contributes to most of the energy dissipation in notched cross-ply laminates under shear-dominated loading. However, this is not the case for total crack volume in specimens at the corresponding life as shown in Figure 15a, with the lowest value (2.2%) featuring λ = 1.571 but 4.1% for λ = 0 and 0.785. This lower crack volume explains the less pronounced stiffness degradation of specimens under the fatigue of pure shear as shown in [30] and is associated to the significant reduction of splitting in 90◦ plies. With the comparison to the fatigue case, similar damage mechanisms were observed in the case of quasi- static loading for all the multiaxiality ratios as shown in Figures 6-8. However, the overall sub-critical damage during quasi-static loading is less diffused, consistent with the DIC results discussed in section 3 and the quantitative analysis via µ−CT. In Figures 15a-d, for all the multiaxiality ratios, both crack volume and delamination area for the quasi-static case above 90% of total quasi-static life are significantly less than the fatigue case at even lower percentage of the life. Only about 2.4% crack volume and 0.7% delamination area in average for different multiaxiality ratios were observed at 95% of total quasi-static life. Notwithstanding this, these sub-critical damage grows rapidly close to the end of total quasi-static life leading to the catastrophic failure of the specimen. It is worth mentioning here that the specimen only at 90% and even later stage of its quasi-static life was prepared for the quantitative analysis due to the less pronounced damage at the early stage of the quasi-static loading and the difficulties for the damage visualization. 4.2. [+45/90/ − 45/0]s laminate with an open hole 4.2.1. Fatigue loading condition (a) Mechanism A In the case of specimens subjected to tension-dominated loading conditions (λ = 0, 0.262), the dominant mechanisms before the dramatic failure are the delamination between all the plies except for the middle plies and the significant splitting in ± 45◦ plies. These damage is not the only contribution to the catastrophic failure and the breakage of 0◦ piles also plays a pivotal role. As illustrated in Figure 9, the damage evolution as a function of fatigue lifetime for mechanism A can be summarized in the following: (1) splitting initiates in 90◦ plies at notch tip (Fig. 9D); (2) splitting initiates in 0◦ and ± 45◦ plies at notch tip (Fig. 9D); 7 (3) delamination between the foregoing plies initiates (Fig. 9D); (4) splitting and delamination develop in/between the corresponding plies (Fig. 9E); (5) splitting in ± 45◦ plies grows significantly and delamination reaches critical condition accompanying the emergence of 0◦ plies breakage (Fig. 9E-F). (b) Mechanism B The specimens fail following Mechanism B when the shear load component is involved (λ = 0.785, 1.309 and 1.571). In contrast to mechanism A, a significant reduced delamination between all the plies was observed. On the other hand, the direction of shear loading was established so that the +45◦ plies are in compression while the -45◦ plies are in tension. Considering the fact that the compressive strength of an unidirectional ply is generally lower than its tensile strength, the dominant mechanism is the micro-buckling in +45◦ plies and the final failure is triggered by the significant growth of the micro-buckling due to the compression at the very late stage of the fatigue life. The paths of the micro-buckling usually start in a straight way at the two ends of the hole but not on the same line and then additional micro-buckling paths almost perpendicular to the +45◦ plies happen. These micro-buckling paths are usually in collaboration with the significant splitting in +45◦ plies as mentioned in the foregoing discussion of DIC analysis in Section 3. Similar conclusions were drawn by Tan et al. [45, 46] on the quasi-static fracturing behavior of notched Carbon Fiber Reinforced Polymer (CFRP) laminates. As illustrated in Figures 10-11, the damage evolution of mechanism B can be summarized in the following: (1) splitting initiates in 90◦ plies at notch tip (Fig. 10D) ; (2) splitting initiates in 0◦ and ± 45◦ plies at notch tip (Fig. 10D); (3) a small amount of delamination between 90◦ and ± 45◦ plies starts to initiate (Fig. 10D); (4) splitting and delamination develop in/between the corresponding plies (Fig. 10E); (4) micro-buckling initiates in +45◦ plies at notch tip (Fig. 10F); (5) micro-buckling in +45◦ plies grows unstably in collaboration with the significant splitting in +45◦ plies (Fig. 10F). 4.2.2. Quantitative analysis: fatigue vs. quasi-static condition The similar quantitative analysis on the sub-critical damage as discussed in section 4.1.2 was proceeded. As can be noted from Figure 15g for the fatigue case, the specimen following mechanism A has larger delamination area compared to the one following mechanism B at 70% of total fatigue life, with the highest value (1.13%) for λ = 0 and the lowest value (0.33%) for λ = 1.571. This is supported by the foregoing damage mechanisms showing that delamination takes less important role but micro-buckling dominates the fracturing behavior of notched quasi-isotropic laminates under shear-dominated loading. On the other hand, the evolution of total crack volume before 70% of total fatigue life is close for each multiaxiality ratio. This explains the observation, reported in the previous work by Qiao et al. [30], that the stiffness deteriorates 8 roughly 19 % to 25 % before catastrophic failure for various multiaxiality ratios. In the case of quasi-static loading, specimens follow the similar damage mechanisms for the foregoing fatigue case. However, in mechanism B, the straight paths of the micro-buckling in +45◦ plies at the two ends of the hole have a pronounced propagation with the increasing multiaxiality ratio as can be noted in Figures 10B and 11B . This difference is much more obvious in the quasi-static case compared to the fatigue case. Additionally, the overall sub-critical damage during quasi-static loading is also less diffused which is similar to the foregoing discussion on notched cross-ply laminates. As shown in Figures 15e-h, total crack volume only has 2.7% in average for different multiaxiality ratios and the highest delamination area has 0.13% for multiaxiality ratio λ = 0.785 at 90% of total quasi-static life but these sub-critical damage grows rapidly close to the catastrophic failure. 4.3. [+45/90/ − 45/0]s laminate with a central crack Having discussed the fracturing features and damage mechanisms for the quasi-isotropic specimens in presence of an open hole, similar damage mechanisms were observed for the same layup but featuring a central crack as shown in Figures 12-14. The only difference in mechanism B (micro-buckling dominant) is the location of the micro-buckling paths ahead of the notch. As can be noted from Figures 13-14, the straight micro-buckling paths in front of the notch are on the same line for multiaxiality ratios related to shear-dominated loading. 4.3.1. Quantitative analysis: fatigue vs. quasi-isotropic condition In the fatigue case as shown in Figure 15k, the larger delamination area was also observed for the specimen following mechanism A compared to the one following mechanism B before catastrophic fatigue failure. At 70% of total fatigue life, the delamination reaches approximately 1.4% for λ = 0 whereas a significant reduction characterizes the other multiaxiality ratios, with 0.8% for λ = 0.785 and 0.55% for λ = 1.571. This is similar to the quasi-isotropic specimens featuring an open hole as discussed in the foregoing section. Another similarity was found in terms of the evolution of total crack volume confirming a similar gradual stiffness degradation throughout the fatigue life for various multiaxiality ratios. Despite these similarities, the specimens weakened by a central crack have a significant larger amount of total crack volume compared to the ones weakened by an open hole. However, this does not occur for the quasi-static scenario which shows the central crack case having less amount of total crack volume. This indicates a different evolution of the Fracture Process Zone (FPZ) in quasi-static regime compared to fatigue which is significantly important for structural design since the previous study shows that specimens featuring a central hole behave better than the ones with the same layup but featuring a central crack in fatigue case but not for the quasi-static case [30]. It is worth mentioning again that less diffused sub-critical damage was also observed on the specimens featuring a central hole under quasi-static loading. As shown in Figure 15i-l, total crack volume has roughly 1.96% and the delamination area is significantly low (0.2%) for different multiaxiality ratios at 90% of total quasi-static life but these 9 sub-critical damage also grows rapidly close to the catastrophic failure similar to the foregoing discussion on the quasi-static fracturing behavior. 5. Conclusions Leveraging a synergistic combination of DIC and micro-computed tomography, this study investigated the fracturing morphology and the damage mechanisms of notched quasi-isotropic and cross-ply laminates under multi-axial loading. Based on the results obtained in this study, the following conclusions can be elaborated: 1. For the composite layups and specimen configurations investigated in this work, the sub-critical damage mechanisms under multi-axial fatigue were similar to the ones identified under quasi-static loading. However, it is worth mentioning that in notched quasi-isotropic laminates under shear-dominated loading a more pronounced straight micro-buckling paths ahead of the notch or hole was reported in the quasi-static case compared to the fatigue loading; 2. leveraging the µ-CT results, it is shown that the damage mechanisms of the notched cross-ply laminates are dominated by splitting in 0◦ plies under tension-dominated loading whereas a mix of delamination and fiber kinking in 0◦ plies occurs with increasing shear load; 3. in the case of notched quasi-isotropic laminates, significant delamination and the splitting in ± 45◦ plies characterizes the tension-dominated fatigue behavior whereas the main damage mechanism for the shear-dominated case is the micro-buckling in +45◦ plies in combination with splitting in the same plies; 4. The quantitative analysis of the crack volume via µ-CT and of the maximum principal strain distri- bution by means of DIC reveals a substantial difference in the damage evolution between the quasi-static and fatigue loading conditions. In fatigue, the sub-critical damage is distributed across a larger volume compared to the quasi-static case, leading to a significant strain redistribution. Further, the rate of crack volume increase throughout the life of the specimen is substantially different. While in the quasi-static case the crack volume increases quickly at almost a constant rate, in fatigue it exhibits a slow change followed by an abrupt increase towards final failure; 5. the foregoing results are of utmost importance for the structural design of polymer matrix composites under multi-axial loading condition but so far rarely investigated. This study provides a comprehensive damage evaluation on the fracturing behavior of notched laminates which can be used to validate the existing models for the design of composite structures under multi-axial stress states. Acknowledgments Marco Salviato acknowledges the financial support from the Haythornthwaite Foundation through the ASME Haythornthwaite Young Investigator Award and from the University of Washington Royalty Re- search Fund. The work was also partially supported by NSF CMMI 1428436 "MRI: Acquisition of a 3D 10 X-Ray Computed Tomography Scanner for Imaging of Large Size Infrastructure, Biological, and Mechanical Components" awarded to the University of Washington. References References [1] Salviato M, Chau VT, Li W, Bazant ZP, Cusatis G. Direct testing of gradual postpeak of fracture specimens of fiber composites stabilized by enhanced grip stiffness and mass. J Appl Mech 2016;83(11):111003. [2] Salviato M, Chau VT, Li W, Bazant ZP, Cusatis G. USA Patent NO.US20180259431A1, Grips For A Linear Fracture Testing Machine And Method of Designing Same. [3] Salviato M, Ashari SE, Cusatis G. Spectral stiffness microplane model for damage and fracture of textile com- posites. Compos Struct 2016;137:170-84. [4] Kirane K, Salviato M, Bazant ZP. Microplane triad model for simple and accurate prediction of orthotropic elastic constants of woven fabric composites. J Compos Mater 2015;50(9):1247-60. [5] Kirane K, Salviato M, Bazant ZP. Microplane-triad model for elastic and fracturing behavior of woven compos- ites. J Appl Mech 2016;83(4):041006. [6] Bazant ZP, Planas J. Fracture and size effect in concrete and other quasi-brittle materials. Boca Raton:CRC Press;1998. [7] Salviato M, Kirane K, Ashari SE, Bazant ZP. Experimental and numerical investigation of intra-laminar energy dissipation and size effect in two-dimensional textile composites. Compos Sci Technol 2016;135:67-75. [8] Mefford CH, Qiao Y, Salviato M. Failure and Scaling of Graphene Nanocomposites. Compos Struct 2017; 176:961-72. [9] Qiao Y, Salviato M, Study of the Fracturing Behavior of Thermoset Polymer Nanocomposites via Cohesive Zone Modeling. Compos Struct 2019;220:127-47. [10] Qiao Y, Salviato M. Strength and Cohesive Behavior of Thermoset Polymers at the Microscale: A Size-effect Study. Eng Fract Mech 2019;213:100-17. [11] Ko S, Yang J, Tuttle ME, Salviato M, Effect of the platelet size on the fracturing behavior and size effect of discontinuous fiber composite structures. Compos Struct 2019;227:111245. [12] Ko S, Davey J, Douglass S, Yang J, Tuttle ME, Salviato M. Effect of the Thickness on the Fracturing Behavior of Discontinuous Fiber Composite Structures. Compos Part A-Appl S 2019;125:105520. [13] Deleo AA, Salviato M. Computational study for size effect in composites and nanocomposites. Proceedings to 33rd Annual Technical Conference, 18th US-Japan Conference on Composite Materials ASTM D30. [14] Salviato M, Kirane K, Bazant ZP, Cusatis G. Mode I and II interlaminar fracture in laminated composites: a size effect study. J Appl Mech 2019;86(9):091008. [15] Bourchak M, Farrow IR, Bond IP, Rowland CW, Menan F. Acoustic emission energy as a fatigue damage parameter for CFRP composites. Int J Fatigue 2007;29:457-70. [16] Silversides L, Maslouhi A, Laplante G. Acoustic emission monitoring of interlaminar delamination onset in carbon fibre composites. Struct Hhealth Monit 2013;12(2):126-40. [17] Lomov SV, Ivanov DS, Truong TC, Verpoest I, Baudry F, Bosche KV, Xie H. Experimental methodology of study of damage initiation and development in textile composites in uniaxial tensile test. Compos Sci Technol 2008;68(12):2340-9. [18] Meola C, Boccardi S, Carlomagno GM. Infrared Thermography in the Evaluation of Aerospace Composite Materials. Woodhead Publishing, 2016. [19] Meola C, Carlomagno GM. Infrared thermography to evaulate impact damage in glass/epoxy with manufacturing defects. Int J Impact Eng 2014;67:1-11. [20] Meola C, Boccardi S, Carlomagno GM, Boffa ND, Monaco E, Ricci F. Nondestructive evaluation of carbon fibre reinforced composites with infrared thermography and ultrasonics. Compos Struct 2015;134:845-53. [21] Goidescu C, Welemane H, Garnier C, Fazzini M, Brault R, P´eronnet E, Mistou S. Damage investigation in CFRP composites using full-field measurement techniques: Combination of digital image stereo-correlation, infrared thermography and X-ray tomography. Compos Part B-Eng 2013;48:95-105. [22] Scarponi C, Briotti G. Ultrasonic technique for the evaluation of delaminations on CFRP, GFRP, KFRP com- posite materials. Compos Part B-Eng 2000;31:237-43. [23] Ivanov D, Ivanov S, Lomov S, Verpoest I. Strain mapping analysis of textile composites. Opt Laser Eng 2009;47:360-70. [24] Quek SC, Waas AM, Shahwan KW, Agaram V. Compressive response and failure of braided textile composite: part I-experiments. Int J Non-linear Mech 2004;39:635-48. [25] Garcea SC, Wang Y, Withers PJ. X-ray computed tomography of polymer composites. Compos Sci Technol 2018;156:305-19. [26] Yu B, Blanc R, Soutis C, Withers PJ. Evolution of damage during the fatigue of 3D woven glass-fibre reinforced composites subjected to tensiontension loading observed by time-lapse X-ray tomography. Compos Part A- Appl S 2016;82:279-90. [27] McCombe GP, Rouse J, Trask RS, Withers PJ, Bond IP. X-ray damage characterisation in self-healing fibre reinforced polymers. Compos Part A- Appl S 2012;43:613-620. 11 [28] Carraro PA, Maragoni L, Quaresimin M. Characterisation and analysis of transverse crack-induced delamination in cross-ply composite laminates under fatigue loadings. Int J Fatigue 2019;129:105217. [29] Nguyen MH, Davidson P, Waas AM. Particle-toughened interlayers enhance mechanical response of composite laminates. Compos Sci Technol 2019;182:107761. [30] Qiao Y, Deleo AA, Salviato M. A study on the multi-axial fatigue failure behavior of notched composite lami- nates. Compos Part A, in press, doi: 10.1016/j.compositesa.2019.105640. [31] Mitsubishi Chemical Carbon Fiber and Composites, Sacramento, California, USA http://www.mccfc.com/ [32] Vacmobiles, Auckland, New Zealand https://www.vacmobiles.com/ [33] Blaber J, Adair B, Antoniou A. Ncorr: open-source 2D digital image correlation matlab software. Exp Mech 2015;55(6):1105-22. [34] Harilal R, Ramji M. Adaption of open source 2D DIC software Ncorr for solid mechanics applications. In: Proceedings of 9th International Symposium on Advanced Science and Technology in Epxerimental Mechanics, New Delhi, November, 2014. [35] North Star Imaging, California, USA https://4nsi.com/ [36] Yu B, Bradley BS, Soutis C, Hogg PJ, Withers PJ. 2D and 3D imaging of fatigue failure mechanisms of 3D woven composites. Compos Part A- Appl S 2015;77:37-49. [37] Nexon-Pearson OJ, Hallet SR. An experimental investigation into quasi-static and fatigue damage development into bolted hole specimen. Compos Part B 2015;77:462-73. [38] ImageJ, USA https://imagej.nih.gov/ij/ [39] Fujii T, Shina T, Okubo K. Fatigue notch sensitivity of glass woven fabric composite having a circular hole under tension/torsion biaxial loading. J Compos Mater 1994;28(3):234-51. [40] Takemura K, Fujii T. Fatigue strength and damage progression in a circular-hole-notched GRP composite under combined tension/torsion loading. Compos Sci Technol 1994;52:519-26. [41] Takemura K, Fujii T. Fracture mechanics evaluation of progressive fatigue damage in a circular-hole-notched GRP composite under combined tension/torsion loading. Compos Sci Technol 1994;52:527-34. [42] Suresh S. Fatigue of Materials. Cambridge Univeisty Press, 2012. [43] Maiti S, Geubelle PH. A cohesive model for fatigue failure of polymers. Eng Fract Mech 2005;72:691-708. [44] Guo K, Qiao Y, Salviato M. Scaling of fatigue crack growth in pristine epoxy. Proceedings to 33rd Annual Technical Conference, 18th US-Japan Conference on Composite Materials ASTM D30. [45] Tan JLY, Deshpande VS, Fleck NA. Failure mechanisms of a notched CFRP laminate under multi-axial loading. Compos Part A-Appl S 2015;77:56-66. [46] Tan JLY, Deshpande VS, Fleck NA. The effect of laminate lay-up on the multi-axial notched CFRP panels: Simulation versus experiment. Eur J Mech A-Solid 2017;66:309-21. 12 Figures and Tables Figure 1: Notched specimen geometries and damage evaluation leveraging Digital Imaging Correlation (DIC) and X-ray micro- computed tomography (µ-CT). 13 Damage Evaluation ProcedureFigure 2: Digital Imaging Correlation (DIC) analysis of investigated notched specimens at 97% of total quasi-static or fatigue life. This figure compares three multiaxiality ratios. From the contour plots of the normalized maximum principle strain, it can be noted that the region of high deformation in fatigue is generally more spread compared to quasi-static loading. This is an indication that fatigue damage is generally more distributed than its quasi-static counterpart. Figure 3: The evolution of critical maximum principle strain at 97% of final failure for the specimen weakened by a central hole as a function of multiaxiality ratio. The graph compares quasi-static and fatigue results. 14 Fatigue ( 55 % 𝑃𝑚𝑎𝑥)Quasi-static Load≈97% of Final FailureNormalized Maximum Principle Strain00.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.8Specimen Configuration𝜆=0𝜆=0𝜆=0.785𝜆=0.785𝜆=1.571[0/90]2s with crack𝜃𝜆=0𝜆=0𝜆=0.785𝜆=0.785𝜆=1.571𝜆=1.571𝜆=0𝜆=0[+45/90/-45/0]s with hole[+45/90/-45/0]s with crack𝜆=0.785𝜆=1.571𝜃𝜃≈97% of Final Failure≈97% of Final Failure𝜆=1.571𝜆=0.785𝜆=1.57100.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.800.20.40.60.8(a)(b)(c)0°0°0°0°0°0°+45°+45°+45°+45°+45°+45°+45°+45°+45°+45°+45°+45°00.20.40.60.8100.40.81.21.62MultiaxialityRatio, 𝜆Critical Max Principle Strain[+45/90/-45/0]swith holeQuasi-static Fatigue𝜃Figure 4: Digital Imaging Correlation (DIC) analysis of notched quasi-isotropic specimens for multiaxiality ratio λ = 0.785 and λ = 1.571 at 97% of total fatigue life. Note that the shear strain was normalized against its maximum value in the area of interest. Figure 5: (a) Typical graph of the percentage of pixels vs. grey-scale value in the gauge volume of the [0/90]2s specimens weakened by a 18 mm central crack for a multiaxiality ratio λ = 0 based on the reconstructed 3D image obtained by micro- computed tomography; (b) The evolution of cumulative density function of grey-scale distribution as a function of grey-scale end. All the grey-scale values lower than this end was considered as the damage. In this case, the grey-scale end (159) was selected and ±5 was considered to account for the measurement errors on the accuracy of this selective end value. 15 +45°+45°≈97% of Final Life00.20.40.60.8𝜆=0.785Normalized Shear Strain𝜆=1.57100.20.40.60.8Fatigue (55 % Pmax)𝜃=90o𝜃=45oSpecimen Configuration≈97% of Final Life[+45/90/-45/0]s with hole[+45/90/-45/0]s with crack01234501002000204060801001200100200Grey-scale ValuePercentage of Pixels [%][0/90]2s with central crack𝜆=0Reconstructed 3D 𝜇-CT image[0/90]2s with central crack(a)(b)Quasi-static (90% life)𝜆=0Quasi-static (90% life)Crack VolumeCumulative Density Function [%]Grey-scale EndEnd (154-164)Figure 6: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi- static and fatigue damage in [0/90]2s specimens weakened by a 18 mm central crack for a multiaxiality ratio λ = 0 in the 0◦, 90◦ layers and at the interface. Note that the original colors of the images were inverted for a better visualization on the damage. Figure 7: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi- static and fatigue damage in [0/90]2s specimens weakened by a 18 mm central crack for a multiaxiality ratio λ = 0.785 in the 0◦, 90◦ layers and at the interface. Note that the original colors of the images were inverted for a better visualization on the damage. 16 Fatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈95%90%Multiaxiality Ratio[0/90]2swith central crack90°Splttingb0°Splittingc90°Splttingb0°SplittingcDelaminationa/b interfaceb90°Splttingc0°SplittingbDelaminationa/b interfacec90°Splitting0°Splitting𝜆=0•0°is the longitudinal direction.•90°is the transverse direction.5 mmSymmetrycdab100%100%5 mmSplitting DominatedSplitting DominatedMechanism A(A)(B)(C)(D)(E)(F)Multiaxiality Ratio[0/90]2swith central crack•0°is the longitudinal direction.•90°is the transverse direction. Fatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈95%90%bDelamination / 0°Splittinga/b interface90°Splittinga/b interfaceDelamination / 0°Splittingb90°Splitting90°Splittingba/b interfaceDelamination / 0°SplittingFiber kinking / 0°splittingaa/b interfacebFiber kinking / 0°splittingaDelamination / 0°Splitting90°Splitting𝜆=0.7855 mmFiber kinking Fiber kinking 1 mm1 mmSymmetrycdab100%100%5 mmSplitting / Delamination DominatedSplitting / Delamination Dominated(A)Mechanism B(B)(C)(D)(E)(F)Figure 8: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi- static and fatigue damage in [0/90]2s specimens weakened by a 18 mm central crack for a multiaxiality ratio λ = 1.571 in the 0◦, 90◦ layers and at the interface. Note that the original colors of the images were inverted for a better visualization on the damage. Figure 9: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi- static and fatigue damage in [+45/90/ − 45/0]s specimens weakened by a 10 mm open hole for a multiaxiality ratio λ = 0 in the +45◦, 90◦, −45◦, 0◦ layers and at the interfaces. Note that the original colors of the images were inverted for a better visualization on the damage. 17 Fatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈95%90%Multiaxiality Ratio[0/90]2swith central crack𝜆=1.571DelaminationFiber Kinking90°Splittingcba/b interfaceDelamination90°Splittingb90°SplittingbDelaminationa/b interfaceFiber Kinkingc90°SplittingbDelamination•0°is the longitudinal direction.•90°is the transverse direction.5 mmFiber KinkingFiber Kinking1 mm1 mma/b interfacea/b interfaceSymmetrycdab100%100%5 mmDelamination / Fiber KinkingDominatedDelamination / Fiber KinkingDominatedMechanism C(A)(B)(C)(D)(E)(F)Fatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈95%90%Multiaxiality Ratio•0°is the longitudinal direction.•90°is the transverse direction.+45°Splittinga90°Splittingb-45°Splittingc0°Splittingd[+45/90/-45/0]swith open hole𝜆=0Symmetryabcd+45°SplittingaDelaminationb/c interface0°/ -45°Splittingc and d90°SplittingbDelaminationc/d interfaceDelaminationDelaminationb/c interfacec/d interface+45°Splittinga0°Splitting90°Splittingbcd-45°SplittingDelaminationDelamination+45°Splitting0°Splitting90°Splitting-45°Splittingb/c interfacec/d interfaceabcd5 mm100%100%5 mmSplitting / Delamination DominatedSplitting / Delamination Dominated(A)(B)(C)(D)(E)(F)Mechanism AFigure 10: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi-static and fatigue damage in [+45/90/ − 45/0]s specimens weakened by a 10 mm open hole for a multiaxiality ratio λ = 0.785 in the +45◦, 90◦, −45◦, 0◦ layers and at the interfaces. Note that the original colors of the images were inverted for a better visualization on the damage. Figure 11: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi-static and fatigue damage in [+45/90/ − 45/0]s specimens weakened by a 10 mm open hole for a multiaxiality ratio λ = 1.571 in the +45◦, 90◦, −45◦, 0◦ layers and at the interfaces. Note that the original colors of the images were inverted for a better visualization on the damage. 18 Fatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈98%90%Multiaxiality Ratio•0°is the longitudinal direction.•90°is the transverse direction.[+45/90/-45/0]swith open hole𝜆=0.785Symmetryabcd+45°Splitting90°Splitting-45°Splitting0°Splittingabcd•+45°Micro-buckling•+45°Splitting•90°Splitting•Delamination0°/ -45°SplittingBetween a and bBetween b and cdb/c interfaceDelamination+45°Splitting90°Splitting-45°Splitting0°SplittingabcdDelaminationb/c interface-45°Splitting0°Splitting90°Splitting+45°Splittingacdba/b interfaceDelamination5 mma and b100%5 mm100%Micro-buckling DominatedMicro-buckling Dominated(A)(B)(C)(D)(E)(F)Mechanism BFatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈98%90%Multiaxiality Ratio•0°is the longitudinal direction.•90°is the transverse direction.Symmetryabcd[+45/90/-45/0]swith open hole+45°Splittingac-45°Splitting90°Splittingabd•+45°Micro-buckling•+45°Splitting•90°Splitting•Delamination0°Splittingda and b+45°Splitting-45°Splitting90°Splitting0°Splittingcabdc/d interfaceb/c interface+45°Splitting90°Splitting-45°Splitting0°SplittingabcdDelaminationDelaminationc and d0°/ -45°Splitting5 mm100%100%5 mm𝜆=1.571Micro-buckling DominatedMicro-buckling Dominated(A)(B)(C)(D)(E)(F)Mechanism BFigure 12: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi-static and fatigue damage in [+45/90/ − 45/0]s specimens weakened by a 10 mm central crack for a multiaxiality ratio λ = 0 in the +45◦, 90◦, −45◦, 0◦ layers and at the interfaces. Note that the original colors of the images were inverted for a better visualization on the damage. Figure 13: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi-static and fatigue damage in [+45/90/ − 45/0]s specimens weakened by a 10 mm central crack for a multiaxiality ratio λ = 0.785 in the +45◦, 90◦, −45◦, 0◦ layers and at the interfaces. Note that the original colors of the images were inverted for a better visualization on the damage. 19 Fatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load𝜆=040%% of life≈95%90%Multiaxiality RatioSymmetry•0°is the longitudinal direction.•90°is the transverse direction.abcd[+45/90/-45/0]swith central cracka/b interfaceb/c interfaceabdDelamination+45°Splitting90°Splitting0°SplittingDelaminationa+45°Splitting0°SplittingdDelaminationa/b interfacea+45°SplittingDelaminationa/b interfaceDelaminationb/c interface90°Splittingbb90°Splitting0°Splittingdb90°Splittingd0°Splittinga+45°Splittingb/c interfaceDelaminationc-45°Splittingc-45°Splittingc-45°Splitting-45°Splittingc5 mm100%100%5 mmSplitting / Delamination DominatedSplitting / Delamination DominatedMechanism A(A)(B)(C)(D)(E)(F)0°Splitting+45°Splitting / Micro-Buckling-45°SplittingaDelaminationa/b interface+45°Splitting90°Splitting-45°Splitting0°Splittingabcda/b interfaceb/c interfaceDelaminationDelamination+45°Splittingabcd90°Splitting-45°Splitting0°SplittingDelamination / -45°Splitting0°Splittingdb/c interface𝜆=0.785SymmetryabcdcdFatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈95%90%Multiaxiaity Ratio•0°is the longitudinal direction.•90°is the transverse direction.[+45/90/-45/0]swith central crackMicro-Buckling / Splitting90°SplittingbMicro-Buckling / Splittinga and b•+45°Splitting •+45°Micro-Buckling•90°Splitting 5 mm1 mm1 mm100%100%5 mmMicro-buckling DominatedMicro-buckling Dominated(A)(B)(C)(D)(E)(F)Mechanism BFigure 14: Analysis of quasi-static and fatigue damage mechanisms by micro-computed tomography. Comparison between quasi-static and fatigue damage in [+45/90/ − 45/0]s specimens weakened by a 10 mm central crack for a multiaxiality ratio λ = 1.571 in the +45◦, 90◦, −45◦, 0◦ layers and at the interfaces. Note that the original colors of the images were inverted for a better visualization on the damage. Figure 15: The evolution of the total crack volume and delamination area for all the investigated specimens weakened by a central crack or hole as a function of the percentage of life for three multiaxiality ratios. The graphs compare quasi-static and fatigue results. 20 Fatigue (55% 𝑃𝑚𝑎𝑥)70%% of lifeQuasi-static Load40%% of life≈98%90%Multiaxiality Ratio•0°is the longitudinal direction.•90°is the transverse direction.[+45/90/-45/0]swith central crack90°Splitting0°Splitting-45°Splitting+45°SplittingabcdDelaminationa/b interfaceDelamination+45°Splittingb/c interface90°Splittingbcda-45°Splitting0°Splittinga/b interfaceb/c interfaceDelaminationDelamination+45°Splittinga0°Splitting-45°Splittingcd90°Splittingb0°/ -45Splittingc and d𝜆=1.571Symmetryabcd•+45°Micro-buckling•+45°Splitting•90°Splitting•Delaminationa and b5 mm100%100%5 mmMicro-buckling DominatedMicro-buckling Dominated(A)(B)(C)(D)(E)(F)Mechanism B00.20.40.60.8185909510001234585909510000.511.522.505010001234567050100Fatigue (55% Pmax) 𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571[0/90]2swith crackQuasi-static[0/90]2swith crackQuasi-static[0/90]2swith crackFatigue (55% Pmax) [0/90]2swith crack0246885909510000.20.40.60.811.20501000123456050100Fatigue (55% Pmax) 𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571[+45/90/-45/0]swith holeQuasi-static[+45/90/-45/0]swith hole[+45/90/-45/0]swith holeQuasi-static[+45/90/-45/0]swith holeFatigue (55% Pmax) 0123485909510000.511.522.585909510000.511.50501000123456785909510001234567050100Fatigue (55% Pmax) 𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571𝜆=0𝜆=0.785𝜆=1.571[+45/90/-45/0]swith crackQuasi-static[+45/90/-45/0]swith crackFatigue (55% Pmax) [+45/90/-45/0]swith crackQuasi-static[+45/90/-45/0]swith crackDelamination Area [%]Crack Volume [%]Percentage of Life [%](a)(b)(c)(d)(e)(f)(g)(h)(i)(j)(k)(l)
1802.01761	1	1802	2018-02-06T01:58:14	Nonequilibrium theory of the conversion-efficiency limit of solar cells including thermalization and extraction of carriers	[ "physics.app-ph", "cond-mat.mes-hall" ]	The ideal solar cell conversion efficiency limit known as the Shockley-Queisser (SQ) limit, which is based on a detailed balance between absorption and radiation, has long been a target for solar cell researchers. While the theory for this limit uses several assumptions, the requirements in real devices have not been discussed fully. Given the current situation in which research-level cell efficiencies are approaching the SQ limit, a quantitative argument with regard to these requirements is worthwhile in terms of understanding of the remaining loss mechanisms in current devices and the device characteristics of solar cells that are operating outside the detailed balance conditions. Here we examine two basic assumptions: (1) that the photo-generated carriers lose their kinetic energy via phonon emission in a moment (fast thermalization), and (2) that the photo-generated carriers are extracted into carrier reservoirs in a moment (fast extraction). Using a model that accounts for the carrier relaxation and extraction dynamics, we reformulate the nonequilibrium theory for solar cells in a manner that covers both the equilibrium and nonequilibrium regimes. Using a simple planar solar cell as an example, we address the parameter regime in terms of the carrier extraction time and then consider where the conventional SQ theory applies and what could happen outside the applicable range.	physics.app-ph	physics	Nonequilibrium theory of the conversion-efficiency limit of solar cells including thermalization and extraction of carriers Kenji Kamide,1, ∗ Toshimitsu Mochizuki,1 Hidefumi Akiyama,2, 3 and Hidetaka Takato1 1Renewable Energy Research Center, National Institute of Advanced Industrial Science and Technology (AIST), Koriyama, Fukushima, 963-0298, Japan 2The Institute for Solid State Physics (ISSP), The University of Tokyo, Kashiwa, Chiba 277-8581, Japan 3OPERANDO-OIL, Kashiwa, Chiba 277-8581, Japan (Dated: February 7, 2018) The ideal solar cell conversion efficiency limit known as the Shockley-Queisser (SQ) limit, which is based on a detailed balance between absorption and radiation, has long been a target for solar cell researchers. While the theory for this limit uses several assumptions, the requirements in real devices have not been discussed fully. Given the current situation in which research-level cell efficiencies are approaching the SQ limit, a quantitative argument with regard to these requirements is worthwhile in terms of understanding of the remaining loss mechanisms in current devices and the device characteristics of solar cells that are operating outside the detailed balance conditions. Here we examine two basic assumptions: (1) that the photo-generated carriers lose their kinetic energy via phonon emission in a moment (fast thermalization), and (2) that the photo-generated carriers are extracted into carrier reservoirs in a moment (fast extraction). Using a model that accounts for the carrier relaxation and extraction dynamics, we reformulate the nonequilibrium theory for solar cells in a manner that covers both the equilibrium and nonequilibrium regimes. Using a simple planar solar cell as an example, we address the parameter regime in terms of the carrier extraction time and then consider where the conventional SQ theory applies and what could happen outside the applicable range. PACS numbers: 84.60.Jt, 88.40.-j, 85.30.-z I. INTRODUCTION Ee for electrons and nh Eh Shockley and Queisser (SQ) determined a theoreti- cal estimate for the upper limit of the conversion effi- ciency in an ideal solar cell1. The original SQ theory takes radiative recombination into account as a main cause of the current loss in solar cells in a simple man- ner. The energy distributions of the carriers, denoted by ne for holes, must be known to evaluate the radiative recombination rate because it is proportional to the sum of their product. SQ the- ory assumes that the carriers in the absorber are in thermal and chemical equilibrium with both the lattice phonons and the carriers in the electrodes at an ambi- ent temperature Tc and with chemical potentials of µc for the conduction electrons and µv for the valence elec- trons. The resulting current-voltage relationship given by radeeV /kBTc , where the voltage be- I = I(V ) = Isun − I 0 tween the electrodes is equal to the Fermi level separation within the absorber, eV = µc − µv, ultimately deter- mines the conversion efficiency during maximum power operation, where Isun is the photo-generated current pro- duced by absorption of sunlight. The detailed balance, which is the essential aspect of the SQ assumptions, has been used routinely in later analyses that further incorpo- rated various additional factors (including Auger recom- bination, light trapping, photon recycling, and Coulomb interactions 2). The requirements for the assumptions used in the SQ theory to be justified are commonly described as follows: 1) the photo-generated carriers lose their kinetic en- ergy via phonon emission and rapidly establish their thermal equilibrium distribution in a moment (which is called fast thermalization); 2) the carriers are extracted rapidly into carrier reser- voirs immediately after they are produced (which is called fast extraction). The latter assumption 2) is actually given explicitly in the original paper 1. However, the above requirements are not sufficiently clear and thus some quantitative is- sues remain. While the two time scales, i.e., the carrier thermalization time, τph, and the carrier extraction time, τout, are assumed to be short, the following questions are not addressed: first, how short should these times be, i.e., which timescales from other processes should be compared with these times, and second, how do τph and τout compare? The latter question relates directly to the concept of hot carrier solar cells operating out of equilib- rium3–7, where fast carrier extraction before the thermal- ization is complete can reduce the thermalization losses and ensure that device performance is not limited by a detailed balance. Record efficiencies of recent cell research are gradu- ally approaching the SQ limits in nonconcentrator-type single-junction solar cells, e.g., Kaneka's Si-based cell with 26.7 percent efficiency and Alta Devices' thin-film GaAs-based cell with 28.8 percent efficiency8. It is there- fore important to have a more precise understanding of the situation in which detailed balance theory provides a reliable estimate of the attainable upper efficiency limit. Quantitative estimation of the parameters to which the SQ theory applies will help to clarify the remaining en- ergy losses and push current device performance towards the SQ limit. Additionally, a more precise understanding of the energy conversion mechanisms from the detailed balance will lead to new strategies for future improve- ments that are intended to go beyond the SQ limit. The nonequilibrium dynamics of many particle sys- tems can be described in general terms using nonequi- librium Green's functions (NEGFs)9–11. These func- tions were initially applied to study electron transport in solids and in mesoscopic devices12,13, and later in semiconductor light-emitting devices (e.g. light-emitting diodes or LEDs14, semiconductor lasers15, quantum cas- cade lasers16, and polariton condensates17,18). More re- cently, the NEGF formalism was also used to study so- lar cells with nanostructured absorbers19–21, where the device characteristics are likely to be affected strongly by the quantum transport of the carriers. NEGFs were also used to study the conditions required to validate use of luminescence-based characterization of solar cells22, which is justifiable in terms of photovoltaic reciprocity under the detailed balance principle23,24. Despite the sound theoretical basis that is available, the device char- acteristics have not been explored for a sufficiently wide range of parameters via the NEGF approach, particularly for solar cells. This seems to be related to the complex- ity of the theory and high computational costs. Similar issues were found with an ab initio approach25 In this work, we present a nonequilibrium theory that does not assume any form for the distribution functions used for the carriers in the absorber, in a manner similar to the NEGF formulation. The carrier distribution func- tions in the absorber are determined using a set of rate equations that is derived from second-order perturbation theory based on the coupling between the absorber carri- ers and three baths (the phonon, electron, and hole reser- voirs). Spectral broadening of the microscopic states of the carriers is also included in the relevant cases. As a result, the theory describes solar cell operation for a wide range of parameters, including the situations where the photo-generated carriers are either in or out of thermal equilibrium. This paper is organized as follows. In section II, we formulate a nonequilibrium theory for solar cells based on the model shown in Fig. 1, and derive a set of rate equations for the microscopic carrier distribution func- tion in the absorber. The microscopic carrier distribution function is then determined as a steady-state solution to the rate equation. At the end of this section, general expressions are given for the total output current and the total output power to enable simulation of the solar cell device performance. In section III, the basic prop- erties required by the solution to the set of rate equa- tions are presented before the numerical analysis begins. These properties are useful when verifying the accuracy of the simulation. Classification of the parameter regime, specifically in terms of the carrier extraction time τout, is also presented in Sec. III. Before the set of equations is 2 FIG. 1: Nonequilibrium model of solar cell: electron and hole carriers in a semiconductor absorber (energy gap, Eg) interact with the carriers in the carrier reservoirs (Bath 1 and Bath 2) and the lattice phonons in the absorber (Bath 3). Popu- lation distributions of the carriers and phonons in the three reservoirs are given using the thermal equilibrium distribution function (Fermi-Dirac distribution with chemical potentials of µc and µv for conduction and valence electrons, respectively, at temperature Tc(= 300K), and Bose-Einstein distribution for phonons at lattice temperature Tph(= Tc)). Electron-hole pairs are generated in the absorber by solar ilumination (a blackbody spectrum at temperature TS(= 6000K) is assumed here), and the output current loss is due to radiative recom- bination in the absorber. solved, the equations themselves can be used to indicate the parameter regime where the assumptions of the SQ theory fail. In section IV, a device performance simula- tion based on our formulation is presented for a simple planar single-junction solar cell. The numerical simula- tions show what physically happens in the photovoltaic energy conversion processes in each of the regimes that were classified in Sec. III. In section V, we summarize these findings and discuss future issues and future appli- cations of the nonequilibrium theory. Finally, in this section, we list definitions for the sym- bols used in this paper. Parameters for bulk semiconduc- tors can be found in the standard textbook26. • c = speed of light = 3 × 108 m/s • =Planck constant/2π = 1.0545718 × 10−34 J s • RS = Sun's radius = 0.696 × 106 km • LES = average distance from the Earth to the Sun = 1.496 × 108 km • CR(cid:0)≤ (LES/RS)2 = 46200(cid:1) = concentration ratio • w = absorber thickness in a planar solar cell • A = absorber area in a planar solar cell • V = Aw = absorber volume in a planar solar cell • TS = surface temperature of the Sun = 6000 K • Tc = ambient temperature = room temperature = 300 K • Tph(= Tc) = absorber lattice temperature • kB = Boltzmann constant = 8.6 × 10−5 eV/K • βS = 1/(kBTS), βc = 1/(kBTc), βph = 1/(kBTph) • m∗e(h) = effective mass of electrons (holes) in the valleym2 absorber (Si: m∗e/me = (ν2 mk)1/3/me = ⊥ 1.08, m∗h/me = (m3/2 hh + m3/2 lh )2/3/me = 0.55 with νvalley = 6, m⊥/me = 0.19, mk/me = 0.98, mhh/me = 0.49, mlh/me = 0.16) • me = bare electron mass = 9.1 ×10−31 kg • De(h)(Ee(h)) = de(h)pEe(h) = density of states per unit volume for electrons (holes) in the ab- sorber with kinetic energy Ee (Eh) where de(h)(= (2m∗ e(h))3/2 2π23 ) • Eg = absorber bandgap (Si: 1.12 eV, GaAs: 1.42 eV) • τout = carrier extraction time • µc = Fermi level of electrons in electron reservoir + eV /2 (charge (Bath 1) = Eg/2 + β−1 neutrality condition) c /2 ln dh de • µv = Fermi level of electrons in hole reservoir (Bath de −eV /2 (charge neutrality c /2 ln dh 2) = Eg/2 + β−1 condition) q = electron-phonon coupling constant for con- • gc(v) duction (valence) band carriers (= adef,cq q 2V vAρA (LA) phonons with for longitudinal-acoustic q=phonon wave number) • adef,c = deformation potential for electrons in the bottom conduction band in the absorber (Si: ∼ 10 eV) • adef,v = deformation potential for electrons in the top valence band in the absorber (Si: ∼ adef,c/10 ∼ 1 eV) • vA = LA phonon velocity (Si: ∼ 104 m/s) • ρA = absorber mass density (Si: 2.3 g/cm3) • f F (B) µ,β (E) = Fermi-Dirac (Bose-Einstein) distribu- tion at chemical potential µ and inverse tempera- ture β. 3 FIG. 2: Energy diagrams for carriers in the absorber and Fermi distribution functions in the carrier reservoirs (Baths 1 and 2). II. NONEQUILIBRIUM THEORY (FORMULATION) In this section, we formulate the nonequilibrium theory for solar cells. As shown below, a set of rate equations for the microscopic distribution functions for the electrons (e) and holes (h) in the absorber, ({ne Eh}), are given in the following form: Ee, nh d dt d dt Ee = J e,sun ne Ee − J e,rad Ee − J e,out Ee + d dt Eh = J h,sun nh Eh − J h,rad Eh − J h,out Eh + d dt , (1) . (2) ne Ee(cid:12)(cid:12)(cid:12)(cid:12)phonon Eh(cid:12)(cid:12)(cid:12)(cid:12)phonon nh Here, Ee and Eh are the carrier kinetic energies measured from the bottom of the bands (Fig. 2). The first term, J e(h),sun , on the right-hand side of the rate equations rep- Ee(h) resents the carrier generation rate due to sunlight absorp- tion. The second term, J e,rad Ee(h), represents the carrier loss rate due to radiative carrier recombination. The third term, J e(h),out , represents the rate of carrier extraction to the electrodes. The last term, d , repre- Ee(h) sents the rate of electron scattering to other microscopic states within the same band due to phonon emission or absorption. For the solar cell characteristics simulation, this equation will be solved under the steady-state con- dition: dt ne(h) Ee(h)(cid:12)(cid:12)(cid:12)phonon d dt ne Ee = d dt nh Eh = 0. (3) , d Ee(h) Ee(h) Ee(h) , J e(h),out , J e(h),rad for, J e(h),sun via microscopic modeling of the carriers Ee(h)(cid:12)(cid:12)(cid:12)phonon In the following subsections, we will derive explicit and expressions dt ne(h) in the simple planar solar cell (thickness w, surface area A, volume V = Aw) shown in Fig. 3. The assumptions that were made in the original SQ model1 are also used here to simplify the discussion but will not alter the main conclusion of this paper. For example, we consider the absorber thickness w to be 4 FIG. 4: Absorption processes in (a) direct and (b) indirect gap semiconductors. 6000K under the AM0 condition28, jsun(E) = CR × 0,βS (E), (5) c D0 γ(E) × f B LES(cid:19)2 4(cid:18) RS concentration ratio, D0 1 the where CR is γ(E) = 3π2 (c)−3 × 3E2 is the photonic density of states in a vacuum, and f B 0,βS (E) = (exp(βSE) − 1)−1 is the Bose- Einstein distribution function at energy E with inverse temperature βS = 1/(kBTS). If necessary, for practical device simulations, the solar spectrum may be replaced appropriately, e.g., using the AM1.5 spectrum normal- ized at a total power of 1 kWm−2, which is not the case here (the 6000 K blackbody spectrum in Eq. (4) and Eq. (5) approximates the AM0 spectrum at a total power of 1.6 kW/m2 at 1 sun, with CR = 1). In the rate equa- tions in Eq. (1) and Eq. (2), the generation rates of the microscopic carrier distribution function J e(h),sun should be expressed using the solar spectrum, jsun(E). The ex- pression is dependent on whether the absorber is made from direct or indirect gap semiconductors (Fig. 4). Ee(h) For direct gap semiconductor absorbers (Fig. 4 (a)) - by considering momentum and energy conservations, we can equate the number of carriers that are generated in energy ranges of Ee < E′ < Ee + dEe for electrons and Eh < E′ < Eh + dEh for holes with the number of pho- tons absorbed in the energy range E < E′ < E + dE per unit time in the absorber as follows: Ajsun(E)dE = De(Ee)VJ e,sun = Dh(Eh)VJ h,sun Eh Ee dEe dEh, (6) where the energy conservation law gives E = Eg + Ee + Eh, and momentum conservation under the ef- fective mass approximation gives Ee = m∗ Eh. Here De(= de√Ee) and Dh(= dh√Eh) are the densities of states of electrons and holes per unit volume, respec- tively. The equation relates J e(h),sun and jsun(E) di- h m∗ e Ee(h) FIG. 3: Simple planar solar cell model: a planar absorber of thickness w is illuminated by sunlight from the normal direction and photo-generated carriers are extracted to the two electrodes (Baths 1 and 2) through tunneling contact. The photocarrier losses are due to radiative recombination. larger than the absorption length but less than the mi- nority carrier diffusion length. This allows us to consider perfect absorption of sunlight above the absorption edge (E > Eg) and a homogeneous carrier distribution in the absorber. Perfect anti-reflection behavior at the front surface and perfect passivation with zero surface recom- bination are also assumed here. An additional simplification is made in this work to the band structure of the carriers in the absorber. An effective two-band model is used to describe the micro- scopic carrier states under an effective mass approxima- tion (with infinite bandwidths), in which the effective masses for the electrons (m∗e) and holes (m∗h) were se- lected to reproduce the densities of states near the band extrema. Therefore, the effects of the band anisotropy, the valley degree of freedom within the degenerate bands (particularly in Si), and the contributions from other bands located away from the extrema were not taken into account correctly. In this sense, our analysis is far from but is not intended to be quantitatively accurate in simulations for specific systems, but is rather intended to produce a general picture of the main issue, i.e., nonequi- librium aspects of solar cells. A. Generation rate due to sunlight absorption: J e(h),sun Ee(h) Under the assumption of perfect absorption (where w ≫ absorption length), the number of photons absorbed into the absorber per unit time through surface area A is given by A ×Z ∞ Eg jsun(E)dE. (4) The solar spectrum for the photon number current (per unit area, per unit time, and per unit energy), jsun(E), is simply approximated using blackbody radiation at TS = rectly, as follows: 5 J e,sun Ee = J h,sun Eh = e m∗ h jsun(cid:16)E = Eg + (1 + m∗ jsun(cid:16)E = Eg + (1 + m∗ wDe(Ee) h m∗ e wDh(Eh) )Ee(cid:17) )Eh(cid:17) (cid:18)1 + (cid:18)1 + m∗e m∗h(cid:19) ,(7) m∗e(cid:19) ,(8) m∗h where we assume that all microscopic states of carriers with the same energy are generated with equal proba- bility, independent of their momentum directions. We therefore assume that the carrier distribution function is solely dependent on the kinetic energy of carriers and in- dependent of the momentum direction. This assumption is used throughout the paper. For indirect gap semiconductor absorbers (Fig. 4 (b)) - the absorption process accompanies photon emission or absorption. The energies of the electron-hole pairs deviate from the photon energy by the energy of one phonon ((ΩLA, ΩT A, ΩT O) = (50.9, 57.4, 18.6) meV for Si). We simply neglect the energy shift here because the phonon energy is much smaller than the spectral band- width, ∼ kBTS, of the incoming sunlight. However, be- cause the indirect transition accompanies a shift in the carrier momentum corresponding to the momentum car- ried by the phonons, momentum conservation among the photon and electron-hole pairs is not required in the ab- sorption process. As a result, electron-hole pairs with arbitrary combinations of the energies Ee and Eh can be created by absorption of one photon with energy E, as long as E = Eg + Ee + Eh is satisfied (where the phonon energy shift is neglected). The situation in indirect gap semiconductors means that the expressions for J e(h),sun from Eq. (7) and Eq. (8) for direct gap semiconductors must be altered. Assuming that all electron-hole pairs with Ee(< E − Eg) and Eh(= E − Eg − Ee) are cre- ated by absorption of one photon with energy E with equal probability, the probability pe Ee(E)dEe of finding electrons in a small energy window Ee < E′ < Ee + dEe immediately after absorption is Ee(h) Ee(E)dEe = De(Ee)Dh(∆E − Ee)dEe pe , (9) R ∆E 0 De(E′)Dh(∆E − E′)dE′ where ∆E ≡ E−Eg. Because the number of photons ab- sorbed in the absorber per unit time and per unit energy is Ajsun(E) for E > Eg, we find the following expression for the generation rate of electrons per microscopic state for indirect gap semiconductor absorbers: J e,sun Ee Eg Ajsun(E) × pe Ee(E)dEedE = R ∞ = Z ∞ = Z ∞ 0 Eg+Ee VDe(Ee)dEe jsun(E) × Dh(∆E − Ee)/w R ∆E 0 De(E′)Dh(∆E − E′)dE′ jsun(Eg + Ee + Eh) × √Eh dEh. πwde(Ee + Eh)2/8 FIG. 5: Schematic for calculation of recombination loss rate, J e(h),rad Ee(h) The hole generation rate is given in a similar manner as J h,sun Eh = Z ∞ 0 jsun(Eg + Ee + Eh) × √Ee πwdh(Ee + Eh)2/8 dEe. (11) B. Recombination loss rate: J e(h),rad Ee(h) Ee(h) The derivation of the expression for J e(h),rad presented in this subsection largely follows the derivations in the literature27,28. Because the absorber thickness w consid- ered here is much smaller than the minority carrier diffu- sion length, we can safely assume a homogeneous carrier distribution inside the absorber. For a given set of carrier distribution functions, {ne Eh}, the recombination ra- diation rate of photons Rsp(E) at photon energy E from the arbitrary position of a small volume inside the ab- sorber into the whole solid angle (4π), per unit volume, per unit energy, and per unit time, is Ee, nh Rsp(E) = ( c n γ (E) )M2Dcell 0 De(E′)Dh(∆E − E′)ne × Z ∆E (12) E ′ nh ∆E−E ′dE′, 1 for an indirect gap semiconductor absorber. Here, ∆E ≡ E − Eg, c n is the speed of light inside the absorber with refractive index n, M is proportional to the phonon- mediated transition matrix element that is approximated using the value at the absorption edge, and Dcell γ (E) = 3π2 (c/n)−3 × 3E2 is the photonic density of states in the absorber. It is important to note that only part of the radiation, i.e., the radiation into the limited solid angle within the critical angle of total reflection, can es- cape from the absorber, as shown in Fig. 5, and this re- sults in the photovoltaic current loss. The radiation rate Rsp(±)(E) into the escape cones in the ±x-direction, per unit volume, per unit energy, and per unit time, is then, dE (10) Rsp(±)(E) = h(~c/n) · ~e±xiθ<θc × M2Dcell ×Z ∆E ∆E−E ′dE′, De(E′)Dh(∆E − E′)ne E ′nh 0 γ (E) (13) where h ~c n · ~e±xiθ<θc represents the velocity of light in the ±x-directions when averaged inside the escape cones. The geometric average yields an expression for the rates that is independent of n, i.e., Rsp(±)(E) = c 4M2D0 γ(E) ×Z ∆E 0 De(E′)Dh(∆E − E′)ne E ′ nh ∆E−E ′dE′. (14) Under the same approximation using the constant transition matrix element M, the absorption coefficient α(E) is given microscopically by where Eq. (14) and Eq. (15) were inserted, and 6 ∆E−E ′dE′ E ′nh 0 De(E′)Dh(∆E − E′)ne hnenhi∆E ≡ R ∆E R ∆E 0 De(E′)Dh(∆E − E′)dE′ h1 − ne − nhi∆E ≡ R ∆E E ′ − nh R ∆E 0 De(E′)Dh(∆E − E′)dE′ 0 De(E′)Dh(∆E − E′)(1 − ne , (22) ∆E−E ′)dE′ . (23) α(E) = M2Z ∆E ×(1 − ne 0 De(E′)Dh(∆E − E′) E ′ − nh ∆E−E ′)dE′. Equation Eq. (21) is a generalized Planck's law for in- direct gap semiconductors. Given that the band filling effect can be neglected with ne(h) Ee(h) ≪ 1, it is approxi- mated using (15) and j(±) Using the homogeneous radiation rate and absorption coefficient, we consider continuity equations for the pho- ton number current density inside the absorber. Let nγ(±) E be the photon number density (per unit E volume and per unit energy) and the photon number cur- rent density (per unit area, per unit time, and per unit energy) with energy E propagating in the ±x-directions. Then, the continuity equations under the steady-state condition are given by ∂tnγ(±) E = −∂xj(±) E (x) + Rsp(±)(E) − α(E)j(±) E (x) = 0,(16) for 0 < x < w (Fig. 5). The continuity equations under the appropriate boundary conditions, j(+) E (x ≤ 0) = jsun E , j(−) E (x ≥ w) = 0, (17) (18) give a solution for the output photon-number current density, j(+) E (x = w) = Rsp(+)(E)/α(E), j(−) E (x = 0) = Rsp(−)(E)/α(E), (19) (20) where perfect absorption (α(E)w ≫ 1) and zero refrec- tion at the front surface were assumed again. From the results, the number of photons radiated out from the ab- sorber per unit time (≡ dN Rad /dt) through the front and back surfaces is γ dN Rad γ dt Eg = AZ ∞ 2π2 (cid:18) 1 = Ac (j(+) E (x = w) + j(−) E (x = 0))dE c(cid:19)3Z ∞ Eg E2hnenhi∆E h1 − ne − nhi∆E dE, (21) dN Rad γ dt 2π2 (cid:18) 1 ≈ Ac c(cid:19)3Z ∞ Eg E2hnenhi∆EdE. (24) Because the radiative loss of one photon is equal to the loss of one electron-hole pair, dN Rad /dt can be related to J e(h),rad γ . Ee(h) γ For indirect gap semiconductor absorbers - part of dN Rad /dt in Eq. (24) with Eq. (22), which comes from recombination of the electrons (holes) in a small en- ergy window, Ee(h) < E′ < Ee(h) + dEe(h), divided by the number of corresponding electron (hole) states, dEe(h), gives the radiation loss rates for the mi- croscopic carrier distribution functions. This produces the following expressions: VDe(h) Ee(h) J e,rad Ee J h,rad Eh 0 c = c(cid:19)3 2π2w (cid:18) 1 (Eg + Ee + Eh)2√Eh × ne × Z ∞ c(cid:19)3 2π2w (cid:18) 1 (Eg + Ee + Eh)2√Ee × ne × Z ∞ (π/8)dh(Ee + Eh)2 (π/8)de(Ee + Eh)2 = c 0 Eenh Eh Eenh Eh (25) dEh, (26) dEe. For direct gap semiconductor absorbers - taking the momentum conservation discussed in Sec.II A into ac- count, the radiation loss rates for the microscopic car- rier distribution functions are obtained using a similar analysis. Here we simply show the final results: (27) = = c c m∗ e m∗ h Ee m∗e J e,rad Ee J h,rad Eh de√Ee Eenh ne Eh 1 − ne Ee − nh c(cid:19)3 (Eg + Ee + Eh)2 2π2w (cid:18) 1 Eh(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)Eh= × (cid:18)1 + m∗h(cid:19) c(cid:19)3 (Eg + Ee + Eh)2 2π2w (cid:18) 1 Eh(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)Ee= m∗e(cid:19) × (cid:18)1 + Insertion of 1 − ne Ee − nh Eh ≈ 1 into the denominators in Eq. (27) and Eq. (28) gives approximate expressions for direct gap semiconductors that correspond to Eq. (25) and Eq. (26) for indirect gap semiconductors. dh√Eh ne Eenh Eh 1 − ne Ee − nh m∗h (28) m∗ h m∗ e Eh , . C. Carrier extraction rate: J e(h),out Ee(h) Ee(h) The relaxation dynamics of a system of interest caused by weak interaction with the environment can be de- scribed using a standard approach that is widely used in studies of open quantum systems29,30. The carrier extraction rate, J e(h),out , is also obtained using a sim- ilar approach. First, as shown in Fig. 1, complete so- lar cell systems are divided into four parts: the main electron-hole systems in the absorber (Carrier System (sys) ), the electron and hole reservoirs (Bath 1 and Bath 2), and the phonon reservoir (Bath 3). There- fore, the noninteracting Hamiltonian for the whole sys- tem, H0, can be given as the sum of four parts, i.e., H0 = H0,sys + H0,Bath1 + H0,Bath2 + H0,Bath3, where k′ c†k′ ck′ , ǫc H0,sys =Xk (cid:16)(Eg + Ee(k))e†kek + Eh(k)h†khk(cid:17) , (29) H0,Bath1 =Xk′ H0,Bath2 =Xk′ H0,Bath3 =Xq k′ d†k′ dk′ , ǫd q b†qbq. Eph (31) (32) (30) Here, ek and hk are fermionic annihilation operators that are defined using the anticommutation relations, [ek, e†k′]+ = [hk, h†k′ ]+ = δk,k′ ([X, Y ]+ ≡ XY + Y X), of the electrons and holes in the carrier system (absorber), with momentum k and energies of Eg + Ee(k) and Eh(k), respectively. ck′ and dk′ represent the fermionic anni- hilation operators when defined using the anticommu- tation relations [ck, c†k′ ]+ = [dk, d†k′ ]+ = δk,k′ for the electrons in Bath 1 and the holes in Bath 2 with mo- mentum k′, and energies of ǫc k′ , respectively. bq k′ and ǫd 7 represents a bosonic annihilation operator, which is de- fined using the commutation relation [bq, b†q′]− = δq,q′ ([X, Y ]− ≡ XY − Y X) for the phonons in Bath 3 (the crystal lattice in the absorber) with momentum q and energy Eph q . Under the assumption of weak carrier sys- tem interaction with the environments (Bath 1 + Bath 2 + Bath 3), the density matrix of the whole system ρ can be approximated using a product of the matrices for the subsystems: ρ = ρsys ⊗ ρBath1 ⊗ ρBath2 ⊗ ρBath3. (33) With this density matrix, the quantum and statistical average for any physical quantity O is given by hOi = Tr(Oρ). Here, the density matrix for the Carrier Sys- tem, denoted by ρsys, is the matrix of interest and will be determined using the von Neumann equation30. Ad- ditionally, the density matrices for the environments are assumed to be ρBath1 = exp (−βc(H0,Bath1 − µcNc)) /ZBath1, (34) ρBath2 = exp (−βc(H0,Bath2 − (−µv)Nd)) /ZBath2, (35) (36) ρBath3 = exp (−βphH0,Bath3) /ZBath3, which represent the matrices in their thermal equilibrium states, e.g., with temperature Tc and electron chemical potential µc for Bath 1, Tc and hole chemical potential µh(= −µv) for Bath 2, and phonon temperature Tph and a chemical potential of zero for Bath 3. The βs represent inverse temperatures. Nc = Pk′ c†k′ ck′ and Nd = Pk′ d†k′ dk′ represent the total numbers of carri- ers in Bath 1 and Bath 2, respectively, and the Zs are normalization factors used to ensure that Tr(ρBath1) = Tr(ρBath2) = Tr(ρBath3) = 1. (37) Using the density matrix, the distribution functions for the carriers in the absorber are defined as Ee(=Ee(k)) ≡ he†keki, Eh(=Eh(k)) ≡ hh†khki, ne k = ne nh k = nh (38) (39) while those for the particles in the baths are given by hc†k′ ck′i = f F hd†k′ dk′i = f F hb†qbqi = f B µc,βc(ǫc k′ ), −µv,βc(ǫv k′ ) 0,βph(Eph q ), (40) (41) (42) µ,β(E)(≡ 1/(eβ(E−µ) + 1)) and f B where f F µ,β(E)(≡ 1/(eβ(E−µ) − 1)) are the Fermi-Dirac and Bose-Einstein distribution functions, respectively, with inverse temper- ature β and chemical potential µ. The electron extraction rate appears as a perturbation expansion to the kinetic motion in the equations for the distribution functions of the second-order with respect to the weak interaction between the Carrier System and Bath 1, H′ = Hsys−Bath1, whereas the interaction Hamil- tonian is given using the form Hsys−Bath1 =Xk,k′ (T e k,k′ ekc†k′ + (T e k,k′ )∗ck′ e†k). (43) The coupling parameter T e k,k′ represents the tunnel- ing amplitude of an electron passing from the absorber through the tunnel barrier to the electron reservoir. Therefore, T e k,k′ , is a function of the overlap integral of the wavefunctions in the absorber and the reservoir, i.e., it is a function of k and k′, the height and width of the tunnel barrier, and the absorber thickness. Switching to the interaction picture, where OI (t) ≡ ei(H0/)tOe−i(H0/)t, the von Neumann equation for the density matrix is given as d dt ρI (t) = 1 i [H′I (t), ρI (t)]−, where H′I (t) =Xk,k′ (T e k,k′ ekc†k′ e−i(Eg+Ee−ǫc k′ )t/ + h.c.) (44) (45) the interaction Hamiltonian, given by H′(= is Hsys−Bath1), in the interaction picture. Successive itera- tions and time integration of Eq. (44) gives d dt ρI = (cid:18) 1 ∼ (cid:18) 1 i(cid:19)2Z ∞ i(cid:19)2Z ∞ 0 0 [H′I (t), [H′I (t − τ ), ρI (t)]−]−dτ,(46) where the initial time contribution from t = −∞ is ne- glected in the first equation, and a Markov approxima- tion, under the assumption that the main system dynam- ics are sufficiently slow when compared with the memory time in the environments, is used in the second equa- tion29,30. Because we are considering steady-state opera- tion of the solar cells, the Markov approximation can be used safely. Using Eq. (46), the extraction rate for an electron with kinetic energy Ee(= Ee(k)) for momentum k is given by d dthe†keki = Tr(cid:16)e†kek ρI (t)(cid:17) = Ee −J e,out =(cid:18) 1 i(cid:19)2Z ∞ 0 Tr(cid:16)e†kek[H′I (t), [H′I (t − τ ), ρI (t)]−]−(cid:17) dτ (47) Using the cyclic property of the trace, where Tr(XY Z) = Tr(ZXY ), the integrand on the right-hand side of the third equation is given explicitly by Tr(cid:16)(cid:16)e†kekH′I (t)H′I (t − τ ) − H′I (t − τ )H′I (t)e†kek −H′I(t − τ )e†kekH′I (t) + H′I (t)e†kekH′I (t − τ )(cid:17) ρI (t)(cid:17) . (48) Considering the assumption in Eq. (33), the trace for the whole system can be provided by successive partial [H′I (t), [H′I (t − τ ), ρI (t − τ )]−]−dτ (τ e out)−1 = traces of the subsystems. By retaining only the terms that do not vanish after the trace is taken, Eq. (48) can be rewritten as 8 k′ )τ /) k.k′2(ei(Eg +Ee−ǫc T e k′ )τ / − e−i(Eg+Ee−ǫc Xk′ ×Tr(cid:16)(e†keke†kekck′ c†k′ − eke†keke†kc†k′ ck′ )ρI (t)(cid:17) k′ )τ / − e−i(Eg+Ee−ǫc k.k′2(ei(Eg +Ee−ǫc = Xk′ T e µc,βc(ǫc Ee(1 − f F ×(cid:0)ne k′ )) − (1 − ne µc,βc(ǫc Ee)f F k′ )τ /) k′ )(cid:1) . (49) By inserting Eq. (49) into Eq. (47) and performing the integration with respect to time, we obtain J e,out Ee = 2π Xk′ ×(cid:0)ne T e k.k′2δ(Eg + Ee − ǫc k′ ) Ee − f F µc,βc(Eg + Ee)(cid:1) . We therefore derive a simple expression for the extraction rate: (50) (51) (52) J e,out Ee = with extraction time τ e 1 τ e Ee − f F out that is defined as out (cid:0)ne µc,βc(Eg + Ee)(cid:1) , T e(E)2Dc(E)(cid:12)(cid:12)(cid:12)(cid:12)E=Eg+Ee , 2π k′ )) is the density of states in Bath 1, and where Dc(E)(= Pk′ δ(E − ǫc T e(E)2 = Pk′ T e k,k′2δ(E − ǫc k′ ) Pk′ δ(E − ǫc k′ ) , (53) represents the strength of tunneling coupling when aver- aged over the states in Bath 1 at energy E. While τ e out is dependent on the energy of the carriers, we consider it to be a constant parameter in the following analysis for simplicity. In this sense, the carrier extraction time used here is an effective parameter representative for all the electrons tunneling between the absorber and the elec- trode. The same argument is also applicable to the hole extraction rate when using H′ = Hsys−Bath2 = k,k′ hkd†k′ + h.c.). We therefore have a similar ex- Pk,k′ (T h pression for holes: J h,out Eh = 1 τ h out (cid:0)nh Eh − f F −µv ,βc(Eh)(cid:1) , with a hole extraction time of (54) (τ h out)−1 = ≡ 2π 2π T h k.k′2δ(Eh − ǫd k′ ) Xk′ T h(E)2Dd(E)(cid:12)(cid:12)(cid:12)(cid:12)E=Eh , (55) out = τ h where T h(E)2 is an effective tunneling probability for a hole tunneling from the absorber to Bath 2, and Dd(E)(= Pk′ δ(E−ǫd k′ )) is the density of states in Bath 2 at energy E. In the following, we also assume the simplest case, i.e., where τ e out ≡ τout, which will not limit the generality of the main conclusion. As mentioned earlier, τout should be regarded as the effective time scale used for carrier extraction. In this sense, however, it could also be used to parametrize the time between photogeneration and extraction of the car- riers, which may be required for other reasons; e.g., when photogeneration occurs at the center of absorber, τout cannot be less than the time delay given by the distance from the point of generation to the contacts divided by the average carrier velocity. Such time delays would be important in thicker solar cells and appear to be critical in solar cells using nanocrystals, organic solar cells31,32 (including dye-sensitised33), and perovskite solar cells34, which have low carrier mobilities caused by disorders and the Frenkel-like localization35,36 of excitons. τout could be measured using specific characterization methods that are suitable for each system, e.g., by optical characteri- zation of the lifetimes of photo-generated carriers under short-circuit conditions and by transient measurement of the photocurrents37,38. The description based on the tunnel Hamiltonian given in Eq. (43) simply neglects the voltage drop at the con- tact. Ohmic contacts with energy losses at the contacts are outside the scope of this paper. D. Phonon scattering (thermalization) rate: d dt ne(h) Ee(h) (cid:12) (cid:12) (cid:12)phonon The phonon scattering (thermalization) rate in the rate equation for the microscopic carrier distribution functions, given by d , can be obtained us- ing an analysis similar to those presented in Sec. II C. In this case, interactions between the carriers and the phonons are considered using the perturbation Hamilto- nian H′ = Hsys−Bath3, which is given by dt ne(h) Ee(h)(cid:12)(cid:12)(cid:12)phonon Hsys−Bath3 =Xq,k +Xq,k q (bq + b† gv q(bq + b† gc −q)(e†k+qek + h.c.) −q)(h†k+qhk + h.c.), (56) q and gv where gc q are the electron-phonon coupling con- stants for the bottom-conduction-band and top-valence- band electrons in the absorber, respectively. The q- dependence of the coupling constants is dependent on the types of phonons involved. For an order of magnitude- level estimate of the thermalization rate, the LA phonons that originate from the deformation potential are consid- ered for Si absorbers: gc(v) q = adef,c(v)s q 2VvAρA , 9 (57) where adef,c(v), vA, and ρA are the deformation potential for the conduction (valence) electrons, the phonon veloc- ity, and the mass density in the absorber, respectively. A realistic estimate requires inclusion of the scattering caused by the other phonon modes, i.e., the transverse acoustic (TA), transverse optical (TO), and LO modes, and the intra-valley scattering (within the degenerate bands)39, based on realistic electron and phonon band structures26, which is far beyond the scope of this work. The interaction Hamiltonian in the interaction picture is qe†k+qekbqei(Ee(k+q)−Ee(k)−Eph H′I (t) =Xq,k (cid:16)gc +Xq,k (cid:16)gc +Xq,k (cid:16)gv +Xq,k (cid:16)gv qe†k+qekb† −qei(Ee(k+q)−Ee(k)+Eph q h†k+qhkbqei(Eh(k+q)−Eh(k)−Eph q h†k+qhkb† −qei(Eh(k+q)−Eh(k)+Eph q )t + h.c.(cid:17) q )t + h.c.(cid:17) q )t + h.c.(cid:17) (58) q )t + h.c.(cid:17) . Insertion of Eq. (58) into Tr(e†kek ρI (t)) with the second- order Born-Markov approximation given in Eq. (46) and performing a time integration, we obtain the phonon scattering rates as follows: q (cid:1)(cid:16)ne k−q(1 − ne k(1 − ne k)f B k−q) q )(cid:17) 0,βph(Eph q ) + 1) − ne q)2δ(cid:0)Ee(k + q) − Ee(k) − Eph q (cid:1) q ) + 1) k+q)f B 0,βph(Eph (59) q )(cid:17), (gc q)2 2π ne (gc d dt = − 0,βph(Eph Xq Ee(cid:12)(cid:12)(cid:12)(cid:12)phonon ×δ(cid:0)Ee(k) − Ee(k − q) − Eph ×(f B 2π Xq + ×(cid:16)ne k+q(1 − ne Eh(cid:12)(cid:12)(cid:12)(cid:12)phonon ×δ(cid:0)Eh(k) − Eh(k − q) − Eph ×(f B 2π Xq + ×(cid:16)nh k+q(1 − nh 0,βph(Eph k(1 − ne 2π Xq k)(f B −ne = − 0,βph(Eph k)(f B −nh 0,βph (Eph k(1 − nh d dt (gv nh (gv q )2 q (cid:1)(cid:16)nh k−q(1 − nh k(1 − nh k)f B k−q) q )(cid:17) 0,βph(Eph q ) + 1) − nh q )2δ(cid:0)Eh(k + q) − Eh(k) − Eph q (cid:1) q ) + 1) k+q)f B 0,βph(Eph (60) q )(cid:17). Scattering rates in this form are equivalent to those ob- tained using Fermi's golden rule calculation. The ex- pression above includes a momentum representation of the carrier distribution functions that can be further stated using simpler expressions in the energy represen- tation. First, we transform the q-summation into an integration over polar coordinates in the form Pq = V(2π)3 R ∞ −1 d(cos θ), where θ is the angle be- tween k and q. The angular integration is then performed using the dispersion relation Eph q = vAq, the coupling constants for LA phonons in Eq. (57), and 0 2πq2dqR 1 Ee(h)(k) − Ee(h)(k − q) = Ee(h)(k + q) − Ee(h)(k) = 2(2qk cos θ − q2) 2m∗e(h) , (61) 2(2qk cos θ + q2) 2m∗e(h) . (62) By making a change in the coordinates, where q = ǫ/(vA), we finally obtain cut d dt ne(h) ǫ2dǫ Ee(h)(cid:12)(cid:12)(cid:12)(cid:12)phonon ph Z ǫe(h),− 0,βph(ǫ)(cid:1) − ne(h) = −Ce(h) × (cid:0)1 + f B 0 + Ce(h) cut ph Z ǫe(h),+ 0 pEe(h)(cid:18)ne(h) pEe(h)(cid:18)ne(h) ǫ2dǫ × (cid:0)1 + f B 0,βph(ǫ)(cid:1) − ne(h) Ee(h) Ee(h) (1 − ne(h) Ee(h)−ǫ) 0,βph(ǫ)(cid:19) Ee(h)−ǫ(1 − ne(h) Ee(h) )f B ) Ee(h) Ee(h)+ǫ(1 − ne(h) (1 − ne(h) (ǫ) and 1 + f B Ee(h)+ǫ)f B 0,βph(ǫ)(cid:19). (63) 0,βph Terms proportional to f B (ǫ) repre- sent the scattering rates for phonon absorption and emis- sion, respectively. Here, we newly defined the coefficient Ce(h) ph ≡ 4π4v4 energies as follows: and the Ee(h)-dependent cutoff def,c(v)√m∗ AρA e(h)/2 0,βph a2 ǫe(h),− cut ≡ min(cid:18)Ee(h), ǫph cut, (64) 2vA(cid:0)q2m∗e(h)Ee(h) − m∗e(h)vA(cid:1)(cid:19), ǫe(h),+ cut ≡ min(cid:18)ǫph cut, 2vA(cid:0)q2m∗e(h)Ee(h) + m∗e(h)vA(cid:1)(cid:19), (65) where ǫph cut is the Debye cutoff energy for the LA phonons (∼ 50 meV for Si). The Ee(h) dependences in Eq. (64) and Eq. (65) stem from the condition that the argu- ments in the delta functions in Eq. (59) and Eq. (60) are zero, i.e., from the requirements for energy and mo- mentum conservation during the carrier-phonon scatter- ing processes. A situation also occurs in which the cut- off energy ǫe(h),− becomes negative. This occurs when q2m∗e(h)Ee(h) − m∗e(h)vA < 0 for small Ee(h), i.e., when cut 10 FIG. 6: Estimation of phonon relaxation rate of carriers via the LA mode in Si as estimated from Eq. (67) as a function of kinetic energy for the electrons (solid) and holes (dashed) at two different lattice temperatures: Tph = 300 K (thick) and 0 K (thin). The following parameters are used for Si: vA = 104 m/s, m∗ cut = 50 meV, adef,c = 10 eV, adef,v = 1 eV, and ρA = 2.3 g/cm3. h/me = 0.55, ǫph e/me = 1.08, m∗ A ≡ EPB e(h). Ee(h) < 1 2 m∗e(h)v2 In this case, carriers with Ee(h) < EPB e(h) cannot lose energy via phonon emissions (i.e., carrier cooling does not occur), even at the zero temperature of the lattice (Tph = 0), which is prohibited by the conservation law. However, this Effect, which is called the phonon bottleneck effect, is normally negligible because the threshold energy EP B e(h), known as the phonon bottleneck energy, is very low (e.g., vA = 104 m/sec, m∗e/me = 1.08, and m∗h/me = 0.55 give EPB e = 0.307 meV and EPB h = 0.156 meV in the model for Si). We have already implicitly assumed that Ee(h) > EPB e(h) in Eq. (63) (which sets ǫe(h),− > 0 and the lower domain boundary of the latter integration to zero). cut The time scale for thermalization of the photogen- erated carriers in the absorber can be estimated from Eq. (63). Consider the case where electrons (holes) with energy Ee(h) are generated at an initial time t = 0 under illumination by a narrow-band photon source at the corresponding energy. In this case, ne(h) Ee(h) 6= 0 and ne(h) = 0 at t = 0. When this condition is inserted E6=Ee(h) into Eq. (63), the initial population dynamics are given by d dt ne(h) Ee(h) = − 1 τ e(h) ph ne(h) Ee(h) , where the relaxation time τ e(h) ph is given by τ e(h) ph = Ce(h) 0 cut ph (cid:18)Z ǫe(h),− +Z ǫe(h),+ cut 0 ǫ2f B 0,βph ǫ2(cid:0)1 + f B pEe(h) dǫ(cid:19). (ǫ) 0,βph pEe(h) (66) (67) (ǫ)(cid:1) dǫ At the zero temperature of the lattice, f B 0,βph=+∞ sion insertion of (ǫ) = 0 into Eq. (67) gives the simple expres- 1 τ e(h) ph = Ce(h) ph (ǫe(h),− cut )3 3pEe(h) . (68) ph ph Of course, the time constant τ e(h) gives a timescale for the initial relaxation dynamics for such an ideal situa- tion, It therefore seems reasonable to consider that the carrier thermalization time in solar cells could also be es- timated using τ e(h) in Eq. (67). Fig. 6 shows the phonon relaxation rate that was estimated for Si as a function of the kinetic energies of the carriers. In our Si model, we found that the relaxation time ranges between 10−10 and 10−13.5 s in the relevant energy window for solar cells (as shown in Fig. 5 that corresponds to the band- width of the solar spectra, kBTS). This result is consis- tent with the measured timescale for the carrier cooling process, which ranges from sub-picosecond to hundreds of picoseconds39–41 and also with the timescales in the literature42. A major difference (two orders of magni- tude) between the results for electrons and holes origi- nates from differences in the deformation potentials for the LA phonons. The situation can therefore change if other phonon modes and the fast electron-hole equilibra- tion that was discussed in39 are taken into account in the calculations. From this perspective, we should stress here again that the results in Fig. 6 were given for an order of magnitude-level estimation. 11 include it in our model. Other factors occur because of interactions between the carrier system and the baths, or more explicitely, because of the carrier-phonon inter- action and carrier extraction processes that occur in our model. Spectral broadening of the electrons (holes) with Ee(h), given by Γe(h) (≡ −ImΣr(Ee(h))), is related to the Ee(h) out (= τout) and τ e(h) time constants τ e(h) ph , which we already determined in the preceding subsections: Γe(h) Ee(h) = 2τ e(h) ph + 2τ e(h) out . (69) out ) when τ e(h) ph ≪ τ e(h) Therefore, the broadening is given by /(2τ e(h) ph ) when out and /(2τ e(h) τ e(h) ph ≪ τ e(h) out ≪ τ e(h) ph . In the former case (τ e(h) out ), it is natural to consider that the carriers must be fully relaxed to estab- lish thermal equilibrium with the lattice in the steady state. This assumption can be checked by solving the rate equations in Eq. (1) and Eq. (2) as follows. When the thermalization term d dominates the rate equation, a steady state is achieved when all inte- grands in Eq. (63) disappear. This condition is fulfilled if dt ne(h) Ee(h)(cid:12)(cid:12)(cid:12)phonon ,   ne(h) Ee(h) 1 − ne(h) Ee(h)   = exp(−βphǫ)   ne(h) Ee(h)+ǫ 1 − ne(h) f B 0,βph 1 + f B (ǫ) (ǫ) 0,βph Ee(h)+ǫ = E. Effects of spectral broadening of the microscopic states In the previous four subsections (Sec. II A, Sec. II B, Sec. II C, Sec. II D), explicit forms of the rate equation for the microscopic carrier distribution function in both Eq. (1) and Eq. (2) appear to have been found. Actu- ally, as shown in the simulations below, direct use of the equations that have been derived thus far can be justi- fied in most cases. However, in certain situations, the equations should be modified slightly. Modifications are required when the spectral broadening of the microscopic carrier states becomes important. These situations occur when the carriers are far out of equilibrium. The spectral broadening Γ is taken into account in the NEGF formal- ism automatically as an imaginary part of the (retarded) self-energy (Γ = −ImΣr)9–11. Therefore, we also take the broadening effect into account in a satisfactory manner while keeping the calculations as simple as possible. Several factors can broaden the spectra of the sin- gle particle states in many-particle systems. One factor comes from Coulomb interaction between the carriers. However, we consider the Coulomb interaction to have a minor or secondary effect on the properties of solar cells, which normally work at low carrier densities (as shown for Si solar cells in2), and we have already neglected to = exp(−βph(Ee(h) + ǫ))/ exp(−βphEe(h)) (70) for every possible choice of Ee(h) and ǫ. This means that in the steady state, the distribution function is given by the Fermi-Dirac distribution, with some value for the chemical potential in the absorber µcell e(h): ne(h) Ee(h) = f F µcell e(h),βph (Ee(h)). (71) Therefore, these distribution functions are thermally dis- tributed over the energy range 0 < Ee(h) < kBTph. In this case, inclusion of the broadening within the single particle spectra does not modify the distribution func- tion as long as Γe(h) for 300 K). The condition appears to be satisfied well when we consider that a phonon relaxation time of 1 ps corresponds to broadening of 0.33 meV. In this way, we confirm that the broadening effect is negligible for τ e(h) ph ≪ τ e(h) out . ph )(cid:17) < kBTph (=26 meV Ee(h)(cid:16)= /(2τ e(h) The above consideration allows us to neglect broaden- ing due to electron-phonon interactions in Eq. (69) over the entire range of τ e(h) out . We can therefore safely use the following approximation: Γe(h) Ee(h) = 2τ e(h) out . (72) This approximation greatly simplifies the calculations be- cause τ e(h) out (= τout) is a given parameter that is identical for every carrier state in our model, whereas τ e(h) ph , which should be defined precisely from Eq. (63), is a function of the carrier distribution functions that are to be deter- mined finally. We therefore use an approximation based on use of Eq. (72) for broadening of the single-particle states in this work. We define the lineshape function of the states using AΓ(x), which is a Gaussian function with a half width at half maximum that is equal to Γe(h) . Us- Ee(h) ing Γ ≡ (log 2)−1/2Γe(h) AΓ(x) =r 1 , the function is given by πΓ2 exp(cid:0)−(x/Γ)2(cid:1) . A Lorentzian function is normally used for the lineshape of a single particle state (this also applies in the NEGF formalism13) rather than a Gaussian spectral profile be- cause the Gaussian distribution reflects the statistical fluctuations of the system; it is not derived naturally for ideal models without structural imperfections. We use the Gaussian function for a reason that is demonstrated later in the paper; however, it will not change our main conclusion. Using the line function, expressions for the generation and loss rates in the rate equation that has been derived thus far are modified as follows, where the (73) Ee(h) modification becomes important when τout ≪ τ e(h) ph . For the generation rate J e(h),sun - because the Sun's spectrum is much broader than the broadening of the states, kBTS ≫ Γ, we can neglect the effects of broaden- ing on J e(h),sun . We therefore use Eq. (7) and Eq. (8) for direct gap semiconductors and Eq. (25) and Eq. (26) for direct gap semiconductors. Ee(h) Ee(h) For the radiative recombination loss rate J e(h),rad - inclusion of the spectral broadening modifies the density of states of the carriers. This effect replaces De(h)(E) with De(h)(E) in the analysis that was presented in Sec. II B (particularly in Eq. (14), Eq. (15), Eq. (22), and Eq. (23)), where Ee(h) De(h)(E) ≡ (cid:0)De(h) ∗ AΓ(cid:1) (E) = Z ∞ 0 De(h)(E′)AΓ(E − E′)dE′, (74) is the density of states of the carriers convolved using the spectral function. For example, the denominator in both Eq. (22) and Eq. (23) can be modified as follows: −∞ Z ∞ = Z ∞ −∞ ×Dh(Eh)AΓ(∆E − E′ − Eh) = Z ∞ De(E′) Dh(∆E − E′)dE′ dE′Z ∞ dEeZ ∞ dEeZ ∞ 0 0 0 0 dEhDe(Ee)Dh(Eh) ×A√2Γ(∆E − Ee − Eh). dEhDe(Ee)AΓ(E′ − Ee) In the last equation, we used a general property of the convolution of Gaussian functions: 12 AΓ1 ∗ AΓ2 (E) ≡ Z ∞ −∞ = A√Γ2 1+Γ2 2 dE′AΓ1 (E′)AΓ2 (E − E′) (E). (76) A similar modification was made to the numerator in Eq. (22) and Eq. (23). As a result, the recombination loss rates for the indirect transition are given by , 0 c c = = (77) dE E2 Eenh E ′ J e,rad Ee J h,rad Eh 1 w × 1 w × ×Z ∞ c(cid:19)3 2π2 (cid:18) 1 × R ∞ 0 Dh(E′h)ne hA√2Γ(∆E − Ee − E′h)dE′h RR De(E′e)Dh(E′h)A√2Γ(∆E − E′e − E′h)dE′edE′h c(cid:19)3 2π2 (cid:18) 1 × R ∞ EhA√2Γ(∆E − E′e − Eh)dE′e 0 De(E′e)ne nh RR De(E′e)Dh(E′h)A√2Γ(∆E − E′e − E′h)dE′edE′h for the electrons and holes, respectively (where the ap- proximation 1 − ne Eh ≈ 1 was used). The above expressions can be simplified further by introducing the dimensionless functions Φ(x) and Θ(x, y): ×Z ∞ Ee − nh dE E2 (78) E ′ e 0 , ds √π + Φ(x) ≡ Z ∞ 0 xe−x2 2√π Θ(x, y) ≡ Z ∞ −x = s2e−(s−x)2 (1 + 2x2)(1 + erf(x)) (s + x)2 Φ(s) × 4 e−(s−y)2 √π ds, , (79) (80) where erf(x) is the Gaussian error function and Ψ(x) has following asymptotic forms: Φ(x ≫ 1) = x2, Φ(x ≪ −1) = e−x2 4√πx3 . (81) Using these functions, Eq. (77) and Eq. (78) can be rewritten as: J e,rad Ee J h,rad Eh , 8 = ne Ee c(cid:19)3 2π2(cid:19)(cid:18) 1 πde (cid:18) c/w ×Z ∞ Ee + Eh√2Γ (cid:19) nh 0 pEhΘ(cid:18) Eg√2Γ c(cid:19)3 2π2(cid:19)(cid:18) 1 πdh (cid:18) c/w ×Z ∞ Ee + Eh√2Γ (cid:19) ne 0 pEeΘ(cid:18) Eg√2Γ nh Eh = 8 , EhdEh, (82) EedEe. (83) Within the limit from Γ → +0, we find that Θ → (Eg + Ee + Eh)2/(Ee + Eh)2 in the integrands, which repro- duces the original expressions of Eq. (25) and Eq. (26). Equations (82) and (83), when derived in this way, are (75) the explicit forms of the radiation loss rates for indirect gap semiconductor absorbers. A similar analysis is also applied to direct gap semi- conductor absorbers. We find that the expressions in Eq. (27) and Eq. (28) are modified by the broadening effect as follows: (84) (85) J e,rad Ee = ×Z ∞ 0 J h,rad Ee = ×Z ∞ 0 c e m∗ Eenh Eh c(cid:19)3 (cid:16)1 + m∗ h(cid:17) 2π2w (cid:18) 1 de√Ee E2 √EehA√2Γ(∆E − Eeh)ne R ∞ 0 pE′ehA√2Γ(∆E − E′eh)dE′eh c(cid:19)3 (cid:16)1 + m∗ e(cid:17) 2π2w (cid:18) 1 dh√Eh √EehA√2Γ(∆E − Eeh)ne R ∞ 0 pE′ehA√2Γ(∆E − E′eh)dE′eh Ee − nh Eenh Eh E2 h m∗ c dE, dE, For the carrier extraction rate J e(h),out where we used the approximation 1 − ne Eh ≈ 1. Here, ∆E ≡ E − Eg and m∗eEe = m∗hEh for direct gap semiconductors, and Eeh ≡ Ee + Eh. - inclusion of the broadening effect in this term is straightforward. The incoming particle number rates from the electrodes into the absorber are modified because the single parti- cle states have a spectral width. After modification, we obtain Ee(h) J e,out Ee J h,out Eh = = 1 τout (cid:16)ne τout (cid:16)nh 1 Ee − f F Eh − f F µc,βc(Eg + Ee)(cid:17) , −µv,βc(Eh)(cid:17) , (86) (87) where we assume that τ e(h) out = τout, and the Fermi- Dirac distribution convolved with the lineshape function is given by f F µ,β(E) = (f F µ,β ∗ AΓ)(E) = Z ∞ −∞ f F µ,β(E′)AΓ(E − E′)dE′. (88) For the phonon scattering rate: - in- clusion of the broadening effect replaces the δ-functions in Eq. (59) and Eq. (60) with the convolved lineshape functions as follows: d dt ne(h) Ee(h)(cid:12)(cid:12)(cid:12)phonon (89) −∞ dE1Z ∞ δ(cid:0)Ee(h)(k) − Ee(h)(k − q) − Eph q (cid:1) → A√2Γ(cid:0)Ee(h)(k) − Ee(h)(k − q) − Eph q (cid:1), (cid:16)=Z ∞ dE2 δ(cid:0)E1 − E2 − Eph q (cid:1) ×AΓ(cid:0)E1 − Ee(h)(k)(cid:1)AΓ(cid:0)E2 − Ee(h)(k − q)(cid:1)(cid:17) δ(cid:0)Ee(h)(k + q) − Ee(h)(k) − Eph q (cid:1) → A√2Γ(cid:0)Ee(h)(k + q) − Ee(h)(k) − Eph q (cid:1). −∞ (90) 13 the After replacement, we can transform the q- summation into an integration over polar coordinates, and the angular integration can finally be performed, as was done in Sec. II D. The final result is 0 cut ph E− dE′ d dt ne(h) ǫ2dǫZ E+ Ee(h)(cid:12)(cid:12)(cid:12)(cid:12)phonon = −Ce(h) pEe(h) Z ǫph A√2Γ(Ee(h) − E′ − ǫ)(cid:18)ne(h) ×(cid:0)1 + f B pEe(h) Z ǫph A√2Γ(Ee(h) − E′ + ǫ)(cid:18)ne(h) ×(cid:0)1 + f B 0,βph(ǫ)(cid:1) − ne(h) ǫ2dǫZ E+ 0,βph(ǫ)(cid:1) − ne(h) Ce(h) dE′ Ee(h) E− + ph cut 0 Ee(h) (1 − ne(h) E ′ ) )f B E ′ (1 − ne(h) Ee(h) 0,βph(ǫ)(cid:19) (91) ) Ee(h) E ′ (1 − ne(h) (1 − ne(h) E ′ )f B 0,βph(ǫ)(cid:19), where we defined E± as E± ≡ qEe(h) ± ǫs 1 2m∗e(h)v2 A!2 ≥ 0. (92) Within the limit from Γ → +0 in Eq. (91), the Gaus- sian function A√2Γ becomes δ functions. The terms that remain after integration of E′ should come from the poles E′ = Ee(h) ± ǫ, which are located in the interval E− < E′(= Ee(h) ± ǫ) < E+. In this way, Eq. (91) safely reproduces Eq. (63) when Γ = +0. Here, we assumed again that Ee(h) > EPB e(h), i.e., that the phonon bottle- neck effect was neglected as it was in Eq. (63). The reason for adoption of the Gaussian lineshape- As mentioned earlier in this subsection, we have used a Gaussian lineshape for the microscopic states, despite the fact that it cannot be derived naturally for an ideal model without statistical fluctuations. This lineshape was adopeted for the following technical reason. We have derived rate equations for the microscopic distribution functions of the carriers in both direct and indirect gap semiconductor absorbers. In the derivation for the in- direct gap semiconductors, we introduced a probability density of pe(h)(Ee(h)) in Eq. (9) and used it to obtain the generation rate J e(h),sun and the recombination loss rate J e(h),rad 0 De(E′)Dh(∆E − E′)dE′ that Ee(h) is found in the denominator in each of Eq. (10), Eq. (22), and Eq. (23) can all be traced back to the same normal- ization factor in the definition of pe(h)(Ee(h)) in Eq. (9), i.e., the number of possible combinations of all electron- hole pairs with Ee and Eh that can emit a photon of energy E. If the spectral broadening of the microscopic states is included, the normalization factor should then . The factor R ∆E Ee(h) be modified by replacing De(h)(E) with De(h)(E) (as de- fined in Eq. (74)), as shown in Eq. (75). No problems arise with the mathematics here if Gaussian lineshape function AΓ(x) is used for the lineshape. However, if we use the Lorentzian lineshape function, AL Γ (x) ≡ 1 π Γ x2 + Γ2 , (93) the situation changes. Using a similar analysis, the nor- malization factor is calculated to be −∞ Z ∞ De(E′) Dh(∆E − E′)dE′ = Z ∞ dE′Z ∞ −∞ ×Dh(Eh)AL = Z ∞ dEeZ ∞ ×AL dEeZ ∞ Γ (∆E − E′ − Eh) dEhDe(Ee)Dh(Eh) 2Γ(∆E − Ee − Eh). 0 0 0 0 dEhDe(Ee)AL Γ (E′ − Ee) In the last equation, we used a general property of the convolution of Lorentzian functions: Γ1 ∗ AL AL Γ2(E) ≡ Z ∞ dE′AL −∞ = AL Γ1+Γ2 (E). Γ1 (E′)AL Γ2 (E − E′) (95) The integration can also be performed by changing the integration variable to Eeh ≡ Ee + Eh and Eh, which results in the following divergence: −∞ Z ∞ ∝ Z ∞ 8 Z ∞ ∝ Z ∞ 0 π = 0 0 0 De(E′) Dh(∆E − E′)dE′ dEehZ Eeh E2 ehAL E2 eh 2Γ(∆E − Eeh)dEeh eh + (2Γ)2 dEeh = +∞. E2 dEhpEh(Eeh − Eh)AL 2Γ(∆E − Eeh) This divergence represents a clear artifact that is derives from the simplification of our model of the semiconductor carriers. We adopted an effective two-band model with infinite bandwidths for carriers. Because of these infinite bandwidths, there is no upper domain bound for integra- tion over Eeh in Eq. (96). This prevents us from defining the probability measure for pe(h)(Ee(h)) and our formu- lation thus fails when we use the Lorentzian lineshape function and the infinite bandwidths for the carriers. Of course, this problem can be avoided by introducing an effective bandwidth parameter that should exist natu- rally. However, the introduction of an additional band- width parameter can then reduce the benefits of our sim- ple two-band model with the infinite bandwidths. The final results may also depend on the bandwidth parame- ter, the definition of which remains unclear. For these 14 reasons, we have used the Gaussian lineshape for the spectral function, not having encountered such an ar- tifact, and have thus benefitted greatly from use of our simplified model. In realistic devices, the semiconductor absorbers are not ideally prepared with structural imper- fections to cause statistical fluctuations. As a result, their spectral lineshapes will be more or less Gaussian-like, at least in the tails (which could also be Voigt functions). Even when the Lorentzian lineshape is used in the infi- nite bandwidth model, this divergence problem does not occur for direct gap semiconductors. Despite the above discussion on possible changes in the formulation, we still expect that the selection of the lineshape functions will not strongly affect the result or the main conclusion. F. Macroscopic properties: current, conversion efficiency, and energy flow The rate equations used to determine the microscopic carrier distribution functions were derived in the preced- ing subsections (from Sec. II A to Sec. II E). In this sec- tion, we briefly summarize the method used to calculate macroscopic quantities such as the output charge current I, the conversion efficiency η, and the energy flows into the different channels, denoted by JX (X =sun, T, rad, work, Qin, and Qout are defined below and are also shown in Fig. 7) in terms of the distribution functions that are to be obtained. The charge current I is defined as J e,out I = eXk = eZ ∞ Ee(k) = eXk 0 VDe(Ee)J e,out Ee J h,out Eh(k), dEe, (97) Ee Eh which represents the total charge per unit time (where e is the elementary charge) output by an absorber of volume V(= Aw) to an electrode. Here J e,out and J h,out are the functions of the distribution functions given by Eq. (51) and Eq. (54) without the broadening effect, and of the functions given by Eq. (86) and Eq. (87) with the broadening effect, respectively. The second equation in Eq. (97) comes from the steady-state condition based on the total number of charges in the absorber (and will be mentioned again as Eq. (117) in Sec. III). Given that the conduction electrons are extracted to the electrode with chemical potential µc and the valence holes are extracted to the electrode with potential µv (−µv for holes), the total output energy per unit time is given by Pwork = (µc − µv)I = V I, which defines the conversion efficiency as (98) (99) Pwork 0 Ejsun(E)dE × 100. η(%) = A ×R ∞ This expression indicates that the conversion efficiency is dependent on the absorber thickness w because I and Pwork are both proportional to V from Eq. (97). (94) (96) the integrand is given by 15 hnenhi∆E =Z ∞ 0 Z ∞ 0 p∆E Ee,Ehne Eenh EhdEedEh, (104) where p∆E Ee,Eh is a weight function that is defined by p∆E Ee,Eh (105) ≡ De(Ee)Dh(Eh)A√2Γ(∆E − Ee − Eh) RR De(Ee)Dh(Eh)A√2Γ(∆E − Ee − Eh)dEedEh for indirect gap semiconductors. For the direct gap semi- conductors, hnenhi∆E is given by Eehne p∆E Eh dEeh, Eenh (106) hnenhi∆E =Z ∞ 0 with a weight function that is defined by , (107) p∆E Eeh ≡ √EehA√2Γ(∆E − Eeh) R ∞ 0 pE′ehA√2Γ(∆E − E′eh)dE′eh where ∆E ≡ E−Eg, m∗eEe = m∗hEh, and Eeh ≡ Ee+Eh. In a similar manner to the discussion in Sec. II E, the expressions above can be used in the limit where Γ = +0 by changing the lineshape functions in the integrand into delta functions. Jwork is the energy that is extracted as work from the solar cell (per unit time per unit area) and transferred via the charge current, which we already evaluated earlier, and is given by Jwork = Pwork/A = weV Z ∞ 0 De(Ee)J e,out Ee dEe. (108) JQout is the heat that is carried by the charge current and lost in the electrodes outside the cell per unit time per unit area. It can be calculated as: JQout w w + = Ee − f F Eh − f F EeEAΓ(∆E − Ee)(cid:0)ne Eh EAΓ(E − Eh)(cid:0)nh τout ZZ De µc,βc(E)(cid:1) dEedE τout ZZ Dh −µv ,βc(E)(cid:1) dEhdE − Jwork, where ∆E ≡ E − Eg, and the integration runs over the ranges −∞ < E < ∞ and 0 < Ee(h) < ∞. The remaining channel for the energy loss in our model is the absorber thermalization loss that is lost in the phonon reservoir (Bath 3 in Fig. 1). JQin is this heat current, which is lost in the phonon bath, inside the cell per unit time per unit area. It is convenient to determine JQin using the energy conservation law given in Eq. (101), (109) JQin = Jsun − JT − Jrad − Jwork − JQout , (110) because all other terms on the right-hand side have been given above. It can also be evaluated directly using the phonon scattering rate in the rate equation; however, this complicates the calculation. FIG. 7: Energy current from the Sun Jsun, which is shared by five energy current channels: radiation Jrad, light transmis- sion JT, output work Jwork, heat passing into the electrodes JQout , and heat passing into the absorber crystal lattice JQin . The microscopic carrier distribution function offers more detailed information about the energy balance of the solar cells. This information is very helpful in under- standing what occurs inside solar cells during the energy conversion processes, which could also be addressed ex- perimentally 44. The energy current JX (per unit time per unit area) that flows into each channel X can be evaluated separately using the steady-state solution as follows (see also Fig. 7). The energy current per unit time per unit area that is carried by the photons that illuminate the solar cell is Jsun =Z ∞ 0 Ejsun(E)dE. (100) The energy flow from sunlight is shared by different flow channels (five channels are present in this model), where Jsun = JT + Jrad + Jwork + JQout + JQin. (101) The proportion of the solar energy, JT, per unit time per unit area, that is transmitted is given by JT =Z Eg 0 Ejsun(E)dE. (102) Here, the absorption spectrum of the absorber is approx- imated simply using the step function α(E) = α0θ(E − Eg) under the assumption of perfect absorption α0w ≫ 1 (or a large effective light path leff ≫ 1/α0 with light trap- ping textures43). The energy that is radiated outside the solar cell per unit time per unit area is then given by c Jrad = 2π2 (cid:18) 1 c(cid:19)3Z ∞ where ∆E ≡ E−Eg, and 1−ne Eh ≈ 1 is used. When the broadening effect is taken into accoint, hnenhi∆E in E3hnenhi∆EdE, Ee−nh (103) 0 III. BASIC PROPERTIES AND CLASSIFICATION OF THE PARAMETER where REGIME I ǫ Ee(h),E ′ 16 (113) In the previous section, the rate equations that deter- mine the microscopic distribution functions were derived based on our nonequilibrium model. In this section, be- fore we describe the numerical simulation, we give the basic properties related to the particle number conserva- tion laws that must be preserved in any steady states for the equations. This information can then be used in the first screening to validate the numerical results obtained. In the latter part of this section, we will determine the parameter regime in which the solar cells will work under or out of the detailed balance condition. The parameter regime can be classified from the equation itself without fully solving for it. A. Conservation of the total number of carriers within the band during phonon scattering processes The rate equations in Eq. (1) and Eq. (2) represent the net rates for the numbers of particles coming into the mi- croscopic states with energies Ee and Eh. Therefore, as shown in the previous section (Sec. II F), the summation of dne(h) /dt over all microscopic states gives the total Ee(h) net number of electrons (or holes) coming into the ab- sorber (or the cell) per unit time. In the steady state, the net total rate vanishes as a result of the balance between the four terms that are related to generation by sunlight absorption, the radiative recombination loss, extraction to the electrodes, and phonon scattering, on the right-hand side of the rate equation. Among these four terms, the last one, which leads to intraband carrier thermalization, should not change the total number of carriers contained within the band. This basic property must be preserved in our model formulation, which can be checked using the general expression given in Eq. (91) (with the broadening effect) as shown below. The net change in the total number of electrons (holes) in the absorber due to the phonon scattering processes per unit time is given by VZ ∞ 0 De(h)(Ee(h)) d dt ne(h) Ee(h)(cid:12)(cid:12)(cid:12)(cid:12)phonon dEe(h). (111) By inserting Eq. (91) into Eq. (111), we find that the integrand of the ǫ-integration is proportional to Z ∞ 0 dEe(h)Z E+ E− dE′(cid:16)I ǫ Ee(h),E ′ − I ǫ E ′,Ee(h)(cid:17) , (112) Ee(h) Ee(h) (1 − ne(h) E ′ ) )f B ≡ A√2Γ(Ee(h) − E′ − ǫ)(cid:18)ne(h) ×(cid:0)1 + f B 0,βph(ǫ)(cid:1) − ne(h) E ′ (1 − ne(h) 0,βph(ǫ)(cid:19). From the definition of E± given in Eq. (92), the integra- tion domain, E− < E′ < E+, is equivalent to A!2 A!2 Ee(h) < √E′ + ǫs 1 Ee(h) > √E′ − ǫs 1 2m∗e(h)v2 2m∗e(h)v2 (114) (115) , , which are both symmetrical with respect to interchange of the following variables: E′ ↔ Ee(h). In this way, we find that the two contributions in Eq. (112) cancel per- fectly after integration. This ensures that the total num- ber of carriers is preserved within the band during the carrier thermalization process. B. Conservation of total charge and charge neutrality sorber, Qtot(≡ ePk(nh The rate equation for the net total charge in the ab- Ee(k))), can also be ob- tained using the microscopic rate equations. Summation over all states and the difference between those states for the electrons and holes give Eh(k) − ne d dt = −eVZ ∞ +eVZ ∞ Qtot (116) 0 De(Ee)(J e,sun 0 Dh(Eh)(J h,sun Ee − J e,rad Eh − J h,rad Ee − J e,out Eh − J h,out Eh Ee )dEe )dEh, where the phonon scattering terms are absent, in line with the discussion in Sec. III A. Additionally, the terms for photon absorption and recombination radiation on the right-hand side should all cancel. This should be true because each single photon absorption and emission process generates or loses one electron-hole pair with no changes in the total charge. This statement can also be verified easily in our formulation using the general ex- pressions for the generation rate (Eq. (7) and Eq. (8) for direct gap semiconductors, and Eq. (10) and Eq. (11) for indirect gap semiconductors) and the radiative recombi- nation rate (Eq. (84) and Eq. (85) for direct gap semi- conductors, and Eq. (82) and Eq. (83) for indirect gap semiconductors). As a result, Eq. (116) can be rewritten dEe 17 as d dt 0 De(Ee)J e,out Qtot = eV(cid:16)Z ∞ −Z ∞ 0 Dh(Eh)J h,out Eh Ee dEh(cid:17) = 0, (117) where the steady state has been assumed in the second equation. The above equation, which shows that the net extraction rates for the electrons and holes are balanced in the steady state, was used in Eq. (97) in the previous section. Additionally, when using Eq. (86) and Eq. (87) along with the definition of the total charge, the total net charge that is present in the absorber is determined from Eq. (117) under steady-state conditions and is given by Qtot τout 0 Dh(Eh) f F τout(cid:16)Z ∞ = eV −Z ∞ 0 De(Ee) f F −µv,βc(Eh)dEh µc−Eg,βc(Ee)dEe(cid:17). (118) From this result, we see that the total charge that is present in the absorber is solely determined by the cell distribution functions (apart from the broadening effect). In the numerical simulations presented here, we have fo- cused only on the special case of charge neutrality, where Qtot = 0. When the condition ne(h) Ee(h) ≪ 1 is satisfied at low carrier densities, the charge neutrality condition then relates the chemical potentials in the electrodes, µc and µv, as follows: µc + µv 2 = Eg 2 + β−1 c 2 ln dh de . (119) Therefore, using the voltage between the electrodes (eV ≡ µc − µv), the chemical potentials can be given by (120) c µc = (cid:0)Eg + β−1 µv = (cid:0)Eg + β−1 c ln(dh/de) + eV(cid:1) /2, ln(dh/de) − eV(cid:1) /2, to enable charge neutrality to be realized in the absorber. (121) C. Classification of the parameter regime ph In Sec. II, we saw that the rate equation has two char- acteristic times: τout for carrier extraction (from Eq. (52) and Eq. (55)) and τ e(h) for carrier thermalization (from Eq. (67)), which will be determinate for the solar cell properties. As noted in Sec. I, the SQ theory assumes that these two characteristic times should be fast enough for the detailed balance analysis to be applicable, whereas the quantitative issue of how short these times should be remains to be solved. Here, using the rate equations that were derived using the nonequilibrium approach, we ad- dress this issue. In the following discussion, we consider τout to be a free parameter, while τ e(h) is the given material param- eter and is roughly of picosecond order (although the ph FIG. 8: Classification of parameter space 0 < τout < +∞ into three different regimes: (I) slow extraction, where τout > τ ul∗ out ; (II) normal extraction, where τ e(h) out; and (III) fast extraction, where τout < τ e(h) ph < τout < τ ul∗ ph parameter may be modified to a certain degree by intro- duction of phononic nanostructures 45,46). We therefore focus on the parameter space given by τout (in a simi- lar manner to parameterization using the conductance in hot carrier solar cells6). As shown in Fig. 8, we divide the parameter space given by 0 < τout < +∞ into three different regimes. In regime III, i.e., the fast extraction regime where τout < τ e(h) ph , solar cell operation is likely to be differ- ent from that of conventional solar cells because car- rier extraction before thermalization with the lattice will become possible3,4. In real devices, fast extraction within the ps scale requires an ultrathin absorber. With an average estimated velocity at room temperature of ve =p2m∗e × 3kBTc/2 ∼ 105m/s, the carriers can travel a maximum of 0.1 µm within 1 ps. Because the extrac- tion time cannot be less than the travel time from the center of the absorber (where photogeneration occurs) to its surface (where extraction occurs), τout of less than 1 ps requires absorbers with less than sub-µm thickness in planar solar cells. The situation can hardly be real- ized in Si solar cells with low absorption coefficients (a thickness of 10 µm will be required even with light trap- ping for perfect absorption which was assumed here). To achieve fast extraction in Si solar cells, a meander-like structure could be used (see Sec. 7.4 in28). However, for direct gap semiconductors with higher absorption coeffi- cients, sub-µm-scale absorbers could possibly be used to provide efficient solar cells. This is why hot carrier solar cells should use ultrathin nanostructures with strongly absorbing materials, such as GaAs, as described in the next section. While we acknowledge these realistic is- sues, we will however proceed with further discussions and simulation of fast extraction regime III, given that this scenario could even be realized in Si solar cells, to highlight the general properties of solar cells operating in this fast extraction regime. ph The SQ theory assumes fast carrier extraction, where the condition τout < τ e(h) is not required. An upper limit on the extraction time, τ ul∗out(> τ e(h) ph ), should there- fore exist, above which the SQ theory will fail to predict the conversion efficiency limit. We can therefore define these two regimes separately using a boundary as the normal extraction regime (II) and the slow extraction regime (I). In regime II, the SQ theory can be used to predict the solar cell properties. The boundary time τ ul∗out between regimes I and II, which will be dependent on de- vice parameters such as the material and thickness of the absorber, can be evaluated as shown in the remainder of this section. 18 dt ne(h) Ee(h)(cid:12)(cid:12)(cid:12)phonon In regimes I and II, we can safely assume that the thermalization (intraband carrier cooling) is completed within the absorber because the phonon scattering rate dominates the other terms in the rate equations, Eq. (1) and Eq. (2). In this case, the carrier distribution function in the cell is given by the Fermi-Dirac distribution func- tion in Eq. (71), while the chemical potentials in the cell can differ from those in the electrodes, i.e., µcell e(h) 6= µe(h). When the function in Eq. (71) has been assumed, we can insert d = 0 into the rate equations in Eq. (1) and Eq. (2). This is what we observed earlier in Sec. II E. We should also reconsider the meaning of the relation µcell e(h) 6= µe(h) here. This means that the carrier distri- bution functions in the cell differ from those in the elec- trodes, while they are assumed to be equal in the SQ theory when calculating the radiative recombination loss. Using the difference ∆ne(h) the microscopic rate equations for the steady states can be rewritten as Ee(h)(cid:0)≡ ne(h) (Ee(h))(cid:1), Ee(h) − f F µe(h),βc J e(h),out Ee(h) = ∆ne(h) Ee(h) τout = J e(h),sun Ee(h) − J e(h),rad Ee(h) Therefore, the difference is given by ∆ne(h) Ee(h) = τout(cid:16)J e(h),sun ≈ τoutJ e(h),sun Ee(h) (cid:17) Ee(h) − J e(h),rad Ee(h) , . (122) (123) Ee(h) Ee(h) ≫ J e(h),rad Where, in the second line, we have used the approxima- tion J e(h),sun , which is usually fulfilled at the maximum operating power point of V = Vmp (0.779 V for Si and 1.052 V for GaAs at 1 sun when using our 6000 K blackbody Sun), being focused from this point in this subsection. Now we will determine the condition by which the SQ conversion efficiency limit can be modified. When τout increases inside the regime II and gradually approaches I, we can fix the maximum operating power point because the maximum power condition given in the SQ theory will at least remain unchanged inside regime II. The SQ calculation using ne(h) µe(h),βc(Ee(h)) will Ee(h) be justified as long as ∆ne(h) µe(h),βc(Ee(h)) is sat- isfied at V = Vmp, which can be read as Ee(h) ≪ f F = f F τout ≪ f F µe(h),βc(Ee(h)) J e(h),sun Ee(h) ≡ τ ul out(Ee(h)). (124) When τout > τ ul out(Ee(h)), the conversion efficiency limit can then differ from the limit from the SQ theory. A suf- ficient condition for the SQ calculation to be justified is FIG. 9: Upper boundary of the extraction time τ ul∗ out when evaluated as 10 percent of τ ul out(Ee = kBTc) in Eq. (124), shown as functions of (a) cell thickness w and (b) the con- centration ratio for planar Si and GaAs solar cells. In (a), we assumed 1 sun illumination (CR = 1) and maximum power operation at V = Vmp (=0.779 V for Si and 1.052 V for GaAs). In (b), the absorber thickness w was fixed at 100µm for Si and 5µm for the GaAs cells. The CR-dependence of Vmp(CR) is calculated using the SQ theory. that where Eq. (124) is fulfilled over an energy range in which there is a non-negligible carrier distribution that makes a relevant contribution to the radiative recombi- nation; this is given approximately by 0 < Ee(h) < kBTc. (Given that f F (Ee(h)) decreases exponentially with Ee(h), it is almost impossible to satisfy this condition at higher energies.) In this way, the upper boundary of parameter regime II can be evaluated in principle using τ ul out(Ee(h)) in Eq. (124), e.g. by selecting Ee(h) = kBTc as a typical carrier energy scale. µe(h),βc In Fig. 9, we plotted the boundary time τ ul∗out that was defined using 10 percent of τ ul out(Ee = kBTc) as a func- tion of absorber thickness w for Si and GaAs solar cells. Fig. 9(a) clearly shows that τ ul∗out is dependent on the ma- terial parameters (effective mass and indirect/direct gap) and is linearly on w. For 100-µm-thick Si solar cells, τ ul∗out is estimated to be on the sub-ms scale. Additionally, τ ul∗out will decrease in the concentrator solar cells and is plotted as a function of concentration ratio CR in Fig. 9(b). τ ul∗out is roughly proportional to 1/√ The physical meaning of this condition still seems un- out(Ee(h)) in Eq. (124) is dependent on ener- clear since τ ul CR. gies of the microscopic carrier states. The physical mean- ing of the timescale will become clearer if the condition ∆ne(h) µe(h),βc(Ee(h)) is summed over all the micro- scopic states. This can be read as Ee(h) ≪ f F 19 τout ≪ ≈ (Ee(h))dEe(h) dEe(h) µe(h),βc VR De(h)(Ee(h))f F VR De(h)(Ee(h))J e(h),sun sc /e(cid:18)= sc /e × w(cid:19) , Nc I max nc imax Ee(h) (125) where Nc(= Ne = Nh) is the total number of carriers, nc ≡ Nc/V is the carrier density (where ncw = Nc/A is the areal density); the maximum available short-circuit current I max is given by sc I max sc ≡ e × A ×Z ∞ Eg jsun(E)dE, (126) from which the corresponding current density per unit area is defined using imax sc = I max sc /A. (127) The meaning of the condition in Eq. (125), which is equivalent to Nc/τout ≫ I max sc /e, is now much clearer. The number of carriers output from the absorber per unit time (which differs from the net current given by the out- flow minus the inflow) must be greater than the number of photons absorbed per unit time, i.e., the number of carriers that is generated per unit time in the absorber. This condition may have simply been assumed in the SQ theory, although it is not given explicitly in their original paper1. sc sc The final equation in Eq. (125) clearly explains the w-linear dependence of τ ul∗out shown in Fig. 9(a). Note here that imax and the carrier density nc are indepen- dent of w because imax is given solely by the Sun il- lumination conditions and nc is given by the chemical potential that is fixed at the maximum operating power point, V = Vmp. The CR-dependence of τ ul∗out ∝ 1/√CR shown by Fig. 9(b) can also be understood based on the following analysis. The radiative loss current Irad at the maximum operating power can be estimated from the SQ theory using the I-V relation, where I = Isun − Irad with Irad = I 0 radeβceV (see Sec. I). Using d(IV )/dV = 0, we can easily show that Irad ≈ Isun/(βcVmp) as long as βcVmp ≫ 1, which is normally fulfilled. In general, Vmp increases with the concentration ratio CR; at the same time, however, it can never exceed the absorber band gap Eg for solar cell operation (below the lasing condition47), i.e., Vmp(CR = 1) ≤ Vmp(CR) ≤ Eg. In Si, for example, this means that 0.779 eV ≤ Vmp(CR) ≤ 1.12 eV, and correspondingly, Irad (≈ Isun/(βcVmp)) ranges at most from 2.3 to 3.3 percent of Isun over the whole CR range (1 ≤ CR ≤ 46200 (full)). Therefore, Irad is almost pro- portional to Isun(∝ CR). In addition, we notice that the radiative loss current in general is Irad ∝ n2 c, or equiv- alently, nc ∝ I 1/2 rad ∝ CR1/2. Because imax is obviously sc FIG. 10: Summary of device simulation results for simple planar solar cell with a 100 − µm-thick Si absorber using our nonequilibrium theory; the maximum conversion efficiency is shown as a function of carrier extraction time τout for various concentration ratios (CR = 1, 101, 102, 103, 46200). The solid lines were obtained for the heat-shared phonon reser- voirs and the dashed lines were obtained for the heat-isolated phonon reservoirs (as explained in the text). The realistic limit for the maximum efficiency will lie between the solid and dashed curves. The SQ limit for each CR is also indi- cated using horizontal lines. ∝ Isun ∝ CR, another definition of τ ul∗out from Eq. (125) shows that τ ul∗out ∝ nc/imax sc ∝ CR−1/2. IV. NUMERICAL SIMULATION OF DEVICE PERFORMANCE AND CONVERSION EFFICIENCY LIMIT OF A SIMPLE PLANAR SOLAR CELL In this section, we present numerical simulation results for the device performance and the conversion efficiency limit of a simple planar solar cell. The results obtained here are summarized in Fig. 10, which shows the theo- retical conversion efficiency limit as a function of carrier extraction time τout for various values of concentration ratio CR. As shown in Fig. 10, two curves (solid and dashed lines) are presented for a given CR. The theoret- ical limit is dependent on how effectively the heat that is generated in the crystal lattices of the absorber and the electrodes can be delivered to each other via phonon transport. In the ideal limit case, where phonon trans- port between the absorber and the electrodes is very fast in either direct or indirect ways, which we shall refer to as "heat-shared phonon reservoirs " (see Fig. 11(a)), we have higher maximum conversion efficiencies (shown as solid curves in Fig. 10) when the carrier extraction becomes fast. At the opposite limit, where the phonon transport between absorber and electrodes so slow as to be negligible, which we refer to as "heat-isolated phonon reservoirs" (Fig. 11(b)), the maximum conversion effi- ciency is reduced when the carrier extraction becomes fast (see the dashed curves in Fig. 10). In realistic cases, the maximum conversion efficiency will lie somewhere be- tween these two curves. The precise position will be de- pendent on the phonon transport properties between the absorber and the electrodes, which will be sensitive to the formation conditions for the contacts between the absorber and the electrodes and/or the phonon environ- ment surrounding them. The differences in the two ideal cases are incorporated into the stability conditions for solar cell operation. To enable solar cell operation with Jwork > 0 to be self- sustained solely by solar illumination with Jsun > 0, the requirements are that JQin +JQout > 0 for the heat-shared phonon reservoirs shown in Fig. 11(a), and that JQin > 0 and JQout > 0 for the heat-isolated phonon reservoirs shown in Fig. 11(b). Otherwise, an additional external heat supply to the absorber and/or the electrodes will be required to sustain solar cell operation, and the definition of the solar cell conversion efficiency then becomes less clear. In the following subsections, we discuss the solar cell properties and the underlying physics for these two limiting cases. A. A limiting case: heat-shared phonon reservoirs We now present numerical simulation results for a limiting case involving heat-shared phonon reservoirs. The solid lines in Fig. 10 show the maximum conver- sion efficiencies for various concentration ratios (CR = 1, 10, 102, 103, 46200) plotted as a function of the carrier extraction time τout. We find flat regions for τout of more than 1 ps and less than τ ul∗out, which is dependent on the concentration ratio CR (Fig. 9 (b)), while the maximum conversion efficiency is equal to the SQ limit ηSQ. As expected, we can conclude that the SQ theory applies in the normal extraction regime II. Outside this regime, in both the slow and fast extraction regimes denoted by I and III, respectively, we found significant reductions in ηmax for the simple planar solar cell (Fig. 3). Because the solar cell properties are different for each regimes, our definition of the parameter classifications in Fig. 8 seems reasonable. We must now consider how the differ- ence can be understood. v Fig. 12 shows an increase in the Fermi levels of the electrons and holes in the absorber when compared with those in the electrodes, denoted by ∆µe ≡ µcell c − µc and ∆µh ≡ (−µcell ) − (−µv), respectively, when evaluated using the carrier distribution functions at the maximum operating power point V = Vmp under 1 sun illumina- tion (CR = 1). The increases in the Fermi levels found in regimes I and III clearly show that the carriers in the absorber form nonequilibrium populations. We checked numerically that similar results were also obtained for higher values of concentration ratio CR > 1. The in- creased Fermi levels in the different regimes originate from different mechanisms. In regime I, the carriers are accumulated in the absorber because of the slow extrac- 20 FIG. 11: Two ideal models that take the phonon environ- ments in the two limiting cases into account: (a) heat-shared phonon reservoirs, where the phonon transport between the absorber and electrodes is very fast, in either direct or indi- rect ways; and (b) heat-isolated phonon reservoirs, where the phonon transport between the absorber and electrodes is very slow. In the latter case, the heat generated in the absorber (JQin ) and in the electrodes (JQout ) are dissipated indepen- dently into different heat baths (heat baths A and B). TO enable solar cell operation, Jwork > 0 must be self-sustained solely by solar illumination Jsun > 0; the requirements are that JQin + JQout > 0 for (a) heat-shared phonon reservoirs, and that JQin > 0 and JQout > 0 for (b) the heat-isolated phonon reservoirs. Otherwise, an additional external heat supply to the absorber and/or the electrodes will be required to sustain solar cell operation; the definition of the solar cell conversion efficiency then becomes less clear. tion process, resulting in increases in the carrier density and Fermi level. In contrast, the increment in regime III is attributed to broadening of the microscopic states because Γ = /(cid:0)2√log 2τout(cid:1) is no longer negligible. In this regime, the dominant term in the rate equation is the carrier extraction term J e(h),out , which allows us to have J e(h),out ≈ 0. As a result, the carrier distribu- tion functions in the absorber can be approximated us- Ee(h) Ee(h) 21 FIG. 13: Energy balance evaluated at the maximum operating power point V = Vmp as a function of carrier extraction time (100-µm-thick Si under 1 sun AM0 illumination). (and is also fulfilled for 0 < V < Voc, which is not shown here). The differences between these loss mechanisms are re- flected in their current-voltage (I-V ) characteristics in each regime, as shown in Fig. 14. In slow extraction regime I (Fig. 14(a)), the short-circuit current Isc de- creases with increasing τout while the open-circuit volt- age Voc remains unchanged. In contrast, in fast extrac- tion regime III (Fig. 14(b)), the open-circuit voltage Voc decreases with decreasing τout while the short-circuit cur- rent Isc remains unchanged. These results agree with the consideration that the enhanced radiation loss in regime I accompanies a loss in the output charge current while the enhanced heat current in regime III does not accom- pany such a loss. We find no significant changes in the I-V characteristics for τout between 1 ps and 10−3.5 s, thus supporting the belief that the SQ theory works in normal extraction regime II. The reduction of ηmax in fast extraction regime III in Fig. 10 may be confusing because the hot carrier solar cells that are targeted in this regime can surpass the SQ limit of ηSQ theoretically3–7. Therefore, we may expect an increase in ηmax as τout decreases in regime III. How- ever, this discrepancy is not surprising and can be ex- plained as follows. The differences in the results stem from the differences between the carrier extraction pro- cesses. A hot carrier solar cell uses a filter to select the energies of the carriers that pass from the absorber to the electrodes, which can reduce the heat dissipation (ther- mal losses) in the electrodes. Tailored filtering can in- crease the output voltage while preventing large output current losses in hot carrier solar cells. However, our simple planar solar cell in Fig. 10 does not use such a filter, and the heat dissipation in the electrodes is there- fore not controlled. As already shown in Fig. 13 and as will be shown in the next subsection (Fig. 15), heat losses, and especially the heat loss in the electrodes, in- crease in regime III. This results in a strong reduction c −µc and ∆µh ≡ (−µcell FIG. 12: Increases in the Fermi levels of electrons and holes in the absorber when compared with that in the electrodes; ∆µe ≡ µcell v ) −(−µv) are evaluated at the maximum operating power point V = Vmp and are shown as functions of carrier extraction time (for 100-µm-thick Si under 1 sun AM0 illumination). The maximum conversion efficiency (the black solid line shown in Fig. 10) is also shown. µc,βc Ee ≈ f F (Eg + Ee) and nh ing ne (Eh) from Eq. (86) and Eq. (87). As long as the band filling effect remains negligible, i.e., when ne Eh ≪ 1, the distribution functions from Eq. (88) can be approximated using −µv ,βc Ee ≪ 1 and nh Eh ≈ f F ne nh Ee ≈ exp(cid:0)−βc(Eg + Ee − (µc + βcΓ2/4)(cid:1), (128) Eh ≈ exp(cid:0)−βc(Eh − (−µv + βcΓ2/4)(cid:1). Therefore, in this regime, the increases in the Fermi levels are fitted well using (129) ∆µe ≡ µcell ∆µh ≡ (−µcell v c − µc ≈ βcΓ2/4, ) − (−µv) ≈ βcΓ2/4. (130) (131) To see what happened in the nonequilibrium regimes directly, we simulated the energy balance at the maxi- mum operating power point, as shown in Fig. 13, which provides further information on the energy loss mecha- nism. In slow extraction regime I, the energy loss caused by radiative recombination (Jrad) increases greatly with increasing τout, while that due to thermal loss (JQin + JQout ) decreases. This can be understood easily because the slow extraction process increases the carrier density in the absorber, which thus enhances the radiative re- combination rate. In contrast, in fast extraction regime III, the thermal loss increases while the radiation loss de- creases as τout decreases because the excess energy of the photo-generated carriers is transferred quickly and dissi- pated rapidly in the electrodes before thermalization in the absorber is complete. In this sense, ∆µe in Eq. (130) and ∆µh in Eq. (131) can be regarded as additional ex- cess heat energy conveyed by the fast extraction of one carrier. As shown in Fig. 13, the requirement for sta- ble solar cell operation, given by JQin + JQout > 0 in this model with heat-shared phonon reservoirs, is fulfilled 22 heat-isolated phonon reservoirs, the stability condition for the solar cell requires JQin > 0 and JQout > 0. Oth- erwise, an additional heat supply must be added to the absorber or the electrodes, which is not suitable for our targeted device. To be more precise, let's consider a sit- uation where JQout < 0 (as found below) and what will happen next in the device. In this case, the electrodes require heat supply from others for the stable operation. However, supply from the absorber lattice is prohibited in the heat isolated model. With no heat supply, the lattice temperature of the electrodes, initially at the am- bient temperature, will decrease. Then, the cooled elec- trodes will start to collect heat from the ambient (e.g. surrounding air) at the higher temperature. Then, the cooled electrodes will cool the ambient next during the solar cell operation. Semiconductor devices with similar structure, which cool down the ambient for the opera- tion, can exist in reality as found in light-emitting diodes (LEDs) (namely, refrigerating LEDs48–52). However so- lar cells are the devices aimed for long-term stand-alone operation requiring the stability at a long-time scale. In our model simulation with heat isolated reservoirs, we consider the device with JQout < 0 inevitably requires an additional heat supply for the stable operation. In Fig. 15, the energy balance is shown as a function of bias voltage V for various carrier extraction times: (a) τout = 10−5, (b) 10−7, and (c) 10−11 s. To address the stability condition, we plotted the contributions from the heat flow into the absorber JQin and into the electrodes JQout separately. As shown by the plots, JQout decreases with V and changes sign from positive to negative at a point V = Vst, whereas JQin always remains positive. We call Vst the stability boundary here because the stability condition is fulfilled for V < Vst. When the carrier extraction is slow, e.g. when τout = 10−5 s in Fig. 15(a), Vst is almost the same as Voc. This means that the heat flow direction between absorber and electrodes is the same as the charge flow direction. We consider this to be the normal situation. However, when the extraction becomes faster, as shown in Fig. 15(b) and Fig. 15(c), we find a clear departure of Vst from Voc. In this case, we found the regime where Vst < V < Voc, in which Jwork > 0 but JQout < 0, i.e., where the heat flow and charge flow directions are different. In this regime, the solar cell generates electric power (Jwork > 0), but an external heat supply of no less than JQout must be additionally provided for the electrodes. Such a device would not provide a solar cell operating in a self-sustained manner using solar illumination alone. In this way, we have evaluated ηmax to be the maximum conversion effi- ciency under the stability conditions 0 < V < Vst. For example, in Fig. 15, ηmax is 29.5 percent for τout = 10−5 and 10−7 s, whereas ηmax is reduced to 25.0 percent for τout = 10−11 s. As already shown in Fig. 13, the sum JQin + JQout is positive as long as 0 < V < Voc. Therefore, if ther- mally linked, the depleted heat in the electrodes when Vst < V < Voc can be complemented by the heat in- FIG. 14: Current-voltage characteristics for various carrier extraction times (100-µm-thick Si under 1 sun AM0 illumi- nation): (a) curves obtained in slow extraction regime I are highlighted; (b) curves obtained in fast extraction regime III are highlighted. The solid black curves in (a) and (b) were obtained for 10−12sec < τout < 10−4sec in normal extraction regime II. in ηmax in our case. These contrasting results support the supposition that, unless a tailored carrier extraction process such as that using energy selection is used, it is difficult for fast carrier extraction before thermalization to be beneficial. B. Another limiting case: heat-isolated phonon reservoirs The maximum conversion efficiency ηmax is also de- pendent on the phonon environment that surrounds the solar cell. In this subsection, we focus on solar cell per- formance in another limiting case with the heat-isolated phonon reservoirs shown in Fig. 11(b). When exchange of phonons between the absorber and electrode crystals is prevented both directly and indirectly, i.e. when their phonon environments are isolated (e.g., Baths 3 and 4 in Fig. 11(b)), ηmax is significantly reduced from the val- ues obtained for heat-shared reservoirs for a small τout. Similar results are obtained, irrespective of the value of CR. Here we explain how these differences occur. In the 23 transport), ηmax will be located between the two ideal cases (the solid and dashed curves in Fig. 10). Because the difference between these two results is large, we pro- pose that ηmax is sensitive to the phonon transport be- tween the absorber and the electrodes (in either direct or indirect ways) in the fast carrier extraction regime. V. CONCLUSION AND FUTURE PROSPECTS We have formulated a nonequilibrium theory for solar cells that includes microscopic descriptions of the car- rier thermalization and extraction processes. This theory extends the Shockley-Queisser theory to nonequilibrium parameter regimes where the detailed balance cannot be applied. The theory provides detailed information about the solar cells, including nonequilibrium carrier distri- bution functions with the chemical potentials that are higher than those in the electrodes, and the energy bal- ance (including output work, radiation losses, transmis- sion losses, and heat dissipation in the absorber and the electrodes), which will provide a precise understanding of the loss mechanisms in various solar cell types for a wide range of parameters. Using the developed theory, we defined three differ- ent regimes in terms of their carrier extraction time that were bounded using two time scales: the thermalization time, τph, and τ ul∗out, at which the device characteristics should change. The upper boundary τ ul∗out is dependent on the absorber material parameters, and is more strongly dependent on the system parameters, e.g., the absorber thickness and solar light concentration ratio. Device simulations of simple planar solar cells have shown that the SQ limit is applicable in the normal ex- traction regime, denoted by regime II (τph < τout < τ ul∗out) in Fig. 8. Outside this regime (in regimes I and III), nonequilibrium carrier populations are found in the ab- sorber and the maximum conversion efficiency is signifi- cantly reduced from the SQ limit. While the reductions in ηmax were similar, the energy loss mechanisms in the fast and slow extraction regimes are different, which is clearly reflected in their I-V characteristics. The reduc- tion in ηmax in the fast extraction regime also indicates that unless a tailored carrier extraction procedure such as that based on energy selection was performed, it would be difficult for fast carrier extraction before carrier thermal- ization to be beneficial. This strong claim is consistent with the fact that hot carrier solar cells require energy selection during their carrier extraction processes in ad- dition to the fast extraction procedure. The nonequilibrium theory presented here covers only a few basic elements of solar cells and has only been tested in simple planar solar cells. The losses of photo- generated carriers in the absorber in this work are solely due to radiative recombination. Inclusion of nonradiative recombination may change the result, as will be discussed elsewhere. In the carrier extraction process, this paper does not consider energy losses at the junction. A case FIG. 15: Energy balance shown as a function of bias voltage V for various carrier extraction times: (a) τout = 10−5, (b) 10−7, and (c) 10−11 s (100-µm-thick Si under 1 sun AM0 il- lumination). For the ideal model with heat-isolated phonon reservoirs shown in Fig. 11(b), the stability condition is given by JQin > 0 and JQout > 0, which is fulfilled for V < Vst (green vertical lines). Note that the fraction from transmis- sion loss JT of the subbandgap light is 21.4 percent (not shown here). flow into the absorber denoted by JQin . The ideal limit in such a case with a strong thermal link corresponds to the heat-shared phonon reservoirs that were discussed in Sec. IV A. In real situations where the thermal link strength is moderate (i.e., heat depletion in the electrodes is partially complemented by the absorber via phonon of this type using ohmic contacts will be studied in fu- ture work. Application of the proposed theory to other types of solar cells, e.g., organic solar cells, perovskite solar cells, multi-junction solar cells, intermediate-band solar cells, and hot carrier solar cells will also be inter- esting. The most important and challenging aspect will be to provide feasible proposals for new solar cells using nonequilibrium features by which the SQ limit can be surpassed. 24 Acknowledgments Authors acknowledge Tatsuro Yuge, Makoto Yam- aguchi, Yasuhiro Yamada, Katsuhiko Shirasawa, Tetsuo Fukuda, Katsuhito Tanahashi, Tomihisa Tachibana, and Yasuhiko Takeda for discussion. This work is by JSPS KAKENHI (15K20931), and the New Energy and Indus- trial Technology Development Organization (NEDO). ∗ Electronic address: [email protected] 1 W. Shockley, and H. J. Queisser, Detailed balance limit of efficiency of p-n junction solar cells, J. Appl. Phys. 32, 510 (1961). 2 A. Richter, M. Hermle, and S. W. Glunz, Reassessment of the Limiting Efficiency for Crystalline Silicon Solar Cells, IEEE J. Photovoltaics 3, 1184 (2013). 3 R. T. Ross, and A. J. Nozik, Efficiency of hot-carrier solar energy converters, J. Appl. Phys. 53, 3813 (1982). 4 P. Wurfel, Solar energy conversion with hot electrons from impact ionisation, Sol. Energy Mater. Sol. Cells 46, 43 (1997). 5 Y. Takeda, T. Motohiro, D. Konig, P. Aliberti, Y. Feng, S. Shrestha, and G. Conibeer, Practical Factors Lower- ing Conversion Efficiency of Hot Carrier Solar Cells, Appl. Phys. Exp. 3, 104301 (2010). 6 Y. Takeda, A. Ichiki, Y. Kusano, N. Sugimoto, and T. Motohiro, Resonant tunneling diodes as energy-selective contacts used in hot-carrier solar cells, J. Appl. Phys. 118, 124510 (2015). 7 D. Suchet, Z. Jehl, Y. Okada, and J-F. Guillemoles, In- fluence of Hot-Carrier Extraction from a Photovoltaic Ab- sorber: An Evaporative Approach, Phys. Rev. Applied 8, 034030 (2017). 8 M. A. Green, Y. Hishikawa, W. Warta, E. D. Dunlop, D. H. Levi, J. Hohl-Ebinger, A. W. Ho-Baillie, Solar Cell Efficiency Tables (Version 50)" Prog. Photovoltaics 25, 668 (2017). 9 P. C. Martin, and J. Schwinger, Theory of Many-Particle Systems. I, Phys. Rev. 115, 1342 (1959). 10 G. Baym, and Leo P. Kadanoff, Conservation Laws and Correlation Functions, Phys. Rev. 124, 287 (1961). 11 L. V. Keldysh, Diagram Technique for Nonequilibrium Processes, Sov. Phys.-JETP 20, 1018 (1965). 12 J. Rammer, and H. Smith, Quantum field-theoretical methods in transport theory of metals, Rev. Mod. Phys. 58, 323 (1986). 13 S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, 1995). 14 S. Steiger, R. G. Veprek, and B. Witzigmann, Electrolu- minescence from a Quantum-Well LED using NEGF, in Proceedings of the 13th International Workshop on Com- putationalElectronics, IWCE 2009, Beijing, China (IEEE 2009). 15 K. Henneberger, and H. Haug, Nonlinear optics and trans- port in laser-excited semiconductors, Phys. Rev. B 38, 9759 (1988). 16 S. -C. Lee, and A. Wacker, Nonequilibrium Green's func- tion theory for transport and gain properties of quantum cascade structures, Phys. Rev. B 66, 245314 (2002). 17 M. H. Szyma´nska, J. Keeling, and P. B. Littlewood, Nonequilibrium Quantum Condensation in an Incoherently Pumped Dissipative System, Phys. Rev. Lett. 96, 230602 (2006). 18 M. Yamaguchi, K. Kamide, R. Nii, T. Ogawa, and Y. Ya- mamoto, Second Thresholds in BEC-BCS-Laser Crossover of Exciton-Polariton Systems, Phys. Rev. Lett. 111, 026404 (2013). 19 U. Aeberhard, Quantum-kinetic theory of photocurrent generation via direct and phonon-mediated optical tran- sitions, Phys. Rev. B 84, 035454 (2011). 20 U. Aeberhard, Theory and simulation of quantum photo- voltaic devices based on the nonequilibrium Green's func- tion formalism, J. Comp. Electronics 10, 394 (2011). 21 N. Cavassilas, F. Michelini, and M. Bescond, Modeling of nanoscale solar cells: The Green's function formalism, J. Renewable and Sustainable Energy 6, 011203 (2014). 22 U. Aeberhard, and U. Rau, Microscopic Perspective on Photovoltaic Reciprocity in Ultrathin Solar Cells, Phys. Rev. Lett. 118, 247702 (2017). 23 U. Rau, Reciprocity relation between photovoltaic quan- tum efficiency and electroluminescent emission of solar cells, Phys. Rev. B 76, 085303 (2007). 24 T. Kirchartz, and U. Rau, Electroluminescence analysis of high efficiency Cu(In,Ga)Se2 solar cells, J. Appl. Phys. 102, 104510 (2007). 25 M. Bernardi, D. Vigil-Fowler, J. Lischner, J. B. Neaton, and S. G. Louie, Ab Initio Study of Hot Carriers in the First Picosecond after Sunlight Absorption in Silicon, Phys. Rev. Lett. 112, 257402 (2014). 26 P. Y. Yu, and M. Cardona, Fundamentals of Semiconduc- tors: Physics and Material Properties, 3rd ed. (Springer, New York, 2005). 27 P. Wurfel, The chemical potential of radiation, J. Phys. C: Solid State Phys. 15, 3967 (1982). 28 P. Wurfel, and U. Wurfel, Physics of Solar Cells: From Ba- sic Principles to Advanced Concepts, 3rd ed., (Weinheim: Wiley-VCH, 2016). 29 H. J. Carmichael, Statistical Methods in Quantum Optics 1: Master Equations and Fokker-Planck Equations, 2nd ed." (Springer, 2003). 30 H. P. Breuer, and F. Petruccione, The Theory of Open Quantum Systems, (Oxford University Press, 2002). 31 C. W. Tang, Two-layer organic photovoltaic cell, Appl. Phys. Lett. 48, 183 (1986). 32 G. Yu, J. Gao, J. C. Hummelen, F. Wudl, and A. J. Heeger, Polymer Photovoltaic Cells: Enhanced Efficiencies via a Network of Internal Donor-Acceptor Heterojunctions, Sci- ence 270, 1789 (1995). 33 B. O'Regan and M. Gratzel, A low-cost, high-efficiency so- lar cell based on dye-sensitized colloidal TiO2 films, Nature 353, 737 (1991). 34 A. Kojima, K. Teshima, Y. Shirai, and T. Miyasaka, Organometal Halide Perovskites as Visible-Light Sensitiz- ers for Photovoltaic Cells, J. Am. Chem. Soc. 131, 6050 (2009). 35 J. Frenkel, On the Transformation of light into Heat in Solids. I, Phys. Rev. 37, 17 (1931). 36 M. Fox, in Chapter 4 of Optical Properties of Solids, 2nd ed., (Oxford University Press, 2010). 37 Y. Tanaka, Y. Noguchi, K. Oda, Y. Nakayama, J. Taka- hashi, H. Tokairin, and H. Ishii, Evaluation of internal po- tential distribution and carrier extraction properties of or- ganic solar cells through Kelvin probe and time-of-flight measurements, J. Appl. Phys. 116, 114503 (2014). 38 V. M. Le Corre, A. R. Chatri, N. Y. Doumon, and L. J. A. Koster, Charge Carrier Extraction in Organic Solar Cells Governed by Steady-State Mobilities, Adv. Energy Mater. 7, 1701138 (2017). 39 T. Suzuki, and R. Shimano, Cooling dynamics of photoex- cited carriers in Si studied using optical pump and tera- hertz probe spectroscopy, Phys. Rev. B 83, 085207 (2011). 40 J. R. Goldman and J. A. Prybyla, Ultrafast Dynamics of Laser-Excited Electron Distributions in Silicon, Phys. Rev. Lett. 72, 1364 (1994). 41 A. J. Sabbah, and D. M. Riffe, Femtosecond pump-probe reflectivity study of silicon carrier dynamics, Phys. Rev. B 66, 165217 (2002). 42 M. Green, Third Generation Photovoltaics: Advanced So- lar Energy Conversion, (Springer-Verlag: Berlin, Heidel- berg, 2003). 43 E. Yablonovitch, Statistical ray optics, J. Opt. Soc. Am. 25 72, 899 (1982). 44 S. Chen, L. Zhu, M. Yoshita, T. Mochizuki, C. Kim, H. Akiyama, M. Imaizumi, and Y. Kanemitsu, Thorough sub- cells diagnosis in a multi-junction solar cell via absolute electroluminescence-efficiency measurements, Sci. Rep. 5, 7836 (2014). 45 J.-K. Yu, S. Mitrovic, D. Tham, J. Varghese, and J. R. Heath, Reduction of thermal conductivity in phononic nanomesh structures, Nat. Nanotechnol. 5, 718 (2010). 46 P. E. Hopkins, C. M. Reinke, M. F. Su, R. H. Olsson, E. A. Shaner, Z. C. Leseman, J. R. Serrano, L. M. Phinney, and I. El-Kady, Reduction in the Thermal Conductivity of Single Crystalline Silicon by Phononic Crystal Patterning, Nano Lett. 11, 107 (2011). 47 M. G. A. Bernard, and G. Duraffourg, Laser Conditions in Semiconductors, phys. stat. solidi b 1, 699 (1961). 48 J. Xue, Z. Li, and R. J. Ram, Irreversible Thermodynamic Bound for the Efficiency of Light-Emitting Diodes, Phys. Rev. Applied 8, 014017 (2017). 49 J. Tauc, The share of thermal energy taken from the sur- roundings in the electro-luminescent energy radiated from a p-n junction, Czech. J. Phys. 7, 275 (1957). 50 G. C. Dousmanis, C.W. Mueller, H. Nelson, and K. G. Pet- zinger, Evidence of refrigerating action by means of photon emission in semiconductor diodes, Phys. Rev. 133, A316 (1964). 51 P. Berdahl, Radiant refrigeration by semiconductor diodes, J. Appl. Phys. 58, 1369 (1985). 52 P. Santhanam, D. J. Gray Jr., and R. J. Ram, Thermoelec- trically Pumped Light-Emitting Diodes Operating above Unity Efficiency, Phys. Rev. Lett. 108, 097403 (2012).
1812.05845	1	1812	2018-12-14T10:19:49	Ultrathin acoustic parity-time symmetric metasurface cloak	[ "physics.app-ph" ]	Invisibility or unhearability cloaks have been made possible by using metamaterials making light or sound flow around obstacle without the trace of reflections or shadows. Metamaterials are known for being flexible building units that can mimic a host of unusual and extreme material responses, which are essential when engineering artificial material properties to realize a coordinate transforming cloak. Bending and stretching the coordinate grid in space requires stringent material parameters, therefore, small inaccuracies and inevitable material losses become sources for unwanted scattering that are decremental to the desired effect. These obstacles further limit the possibility to achieve a robust concealment of sizeable objects from either radar or sonar detection. By using a elaborate arrangement of gain and lossy acoustic media respecting parity-time symmetry, we built an one-way unhearability cloak capable to hide objects seven times larger than acoustic wavelength. Generally speaking, our approach has no limits in terms of working frequency, shape, or size, specifically though, we demonstrate how, in principle, an object of the size of a human can be hidden from audible sound.	physics.app-ph	physics	Ultrathin acoustic parity-time symmetric metasurface cloak Hao-xiang Li,1 Maria Rosendo-L´opez,2 Yi-fan Zhu,1 Xu-dong Fan,1 Daniel Torrent,3 Bin Liang,1 Jian-chun Cheng,1 and Johan Christensen2 1Key Laboratory of Modern Acoustics, MOE, Institute of Acoustics, Department of Physics, Collaborative Innovation Center of Advanced Microstructures, Nanjing University, Nanjing, 210093, China 2Department of Physics, Universidad Carlos III de Madrid, ES-28916 Legan`es, Madrid, Spain 3GROC, UJI, Institut de Noves Tecnologies de la Imatge (INIT), Universitat Jaume I, 12080 Castell`e, Spain 8 1 0 2 c e D 4 1 ] h p - p p a . s c i s y h p [ 1 v 5 4 8 5 0 . 2 1 8 1 : v i X r a 1 Invisibility or unhearability cloaks have been made possible by using meta- materials making light or sound flow around obstacle without the trace of re- flections or shadows. Metamaterials are known for being flexible building units that can mimic a host of unusual and extreme material responses, which are essential when engineering artificial material properties to realize a coordinate transforming cloak. Bending and stretching the coordinate grid in space requires stringent material parameters, therefore, small inaccuracies and inevitable ma- terial losses become sources for unwanted scattering that are decremental to the desired effect. These obstacles further limit the possibility to achieve a robust concealment of sizeable objects from either radar or sonar detection. By using a elaborate arrangement of gain and lossy acoustic media respecting parity-time symmetry, we built an one-way unhearability cloak capable to hide objects seven times larger than acoustic wavelength. Generally speaking, our approach has no limits in terms of working frequency, shape, or size, specifically though, we demonstrate how, in principle, an object of the size of a human can be hidden from audible sound. I. INTRODUCTION For centuries people have dreamt of an invisibility cloak that can make someone indiscernible for the nacked eye when hidden underneath it. An ideal cloak would involve the suppression of back reflected light to render an object camouflaged, but the shadow behind it must also diminish for truly being able to make an object disappear. More than a decade ago an approach was brought forward based on transformation optics permitting the path of light to be bent around objects to be hidden1 -- 7. This transformational approach to engineer space has ever since been extended to other areas of wave physics for cloaks of unhearability comprising sound and mechanical vibrations8 -- 13. Despite those advances, it remains a fundamental challenge to create an unhearability cloak of ultrathin layer width involving minuscule loss-free materials. Here we propose theoretically and demonstrate experimentally that a parity-time symmetric metasurface incorporating acoustic gain and loss can act as such cloak when insonified from one direction. In contrast to transformation acoustics that implies a coordinate-transformation-based deformation of sound through the 2 accurate distribution of the material properties, our approach to perfectly absorb incoming sound and to re-emit it behind the hidden object solely implies the engineering of a complex acoustic metasurface-impedance. The use of parity-time symmetry enables unique cloaking properties useful in the audible range but also applicable to hide submarines from sonar detection. Non-Hermitian systems that respect parity-time (PT)-symmetry have recently become an active frontier in wave physics due to unprecedented possibilities in guiding both sound and light14,15. Most notably, designing complex eigenstates through appropriate balancing of gain and loss provides an unexpected paradigm for exploring non-Hermitian wave control in flourishing areas such as, waveguiding, sensing, communication, and topological insulators16 -- 23. In this work we demonstrate that a sizeable acoustically rigid obstacle (7 times larger than the wavelength) appears hidden to sound waves when coated by a PT symmetric metasurface due to the cancellation of reflections and re-radiation of the impinging field to the far-side. When adjusting the gain-loss contrast to a point where this scenario is reached, unidirectional invisibility (unhearability) is obtained, an effect commonly know as the anisotropic transmission resonance (ATR)24,25. Designing PT symmetry is based on manipulating absorption using judicious structures with gain regions and vice versa. The Hamiltonian commutes with the combined PT operator when loss and gain are equally balanced giving rise to entirely real eigenmode frequencies representing the unbroken or exact phase. When the loss and gain contrast exceeds a certain threshold to reach the broken phase, one of the complex eigenmodes exhibits exponential growth while the other one decays exponentially. The transition between these two phases is the non-Hermitian singularity, also known as an exceptional point (EP) where the modes coalesce. The ATR is associated to the flux-conservation process leading to full transparency, i.e., unity transmittance T = 1, but one-sided reflectionless wave propagation. A special case of the ATR is the unidirectional invisibility phenomenon that not only fulfills the condition of full transmission and vanishing of the reflection from either left RL or right RR incidence, but also implies a zero transmission phase signifying the apparent absence of an obstacle to be heard or seen25. Several studies, both in the fields of optics and acoustics, have already investigated unidirectional invisibility 3 in one dimensional PT symmetric structures enabling shadow-free acoustic sensors and Bragg-scattering suppression in photonic lattices26,27. Here we demonstrate the ability to acoustically cloak a rigid obstacle by covering it by an ultrathin PT metasurface as has been previously simulated for microwave radiation28. The approach consists in camouflaging the portion of the insonified metasurface through absorption and providing the time-reversed image, i.e., acoustic gain, to the shadow region behind the rigid obstacle (Fig. 1). II. RESULTS Complete absorption of sound We begin the study by designing the insonified portion of the cloak, whereas the the gain portion will be treated afterwards. Hence, in order to engineer complete scattering cancella- tion to an incoming plane wave we cover the rigid cylinder of radius a by a lossy metasurface as shown in Fig. 1. The aim is to impedance match the metasurface to the surrounding air to totally absorb the incident wave without reflection. Under the assumption that the air gap separating the ultrathin cloak from the rigid cylinder is substantially smaller than the wavelength of the sound wave, i.e., k(b − a) (cid:28) 1 we can write down the necessary complex surface impedance to fulfill camouflaging of the obstacle (see supplementary information for derivations): Re(Zs) ≈ Z0 Im(Zs) ≈ Z0 b a b a 1 cos θ k(b − a), (1) where θ is the angular position in polar coordinates and k = 2π/λ where λ is the wavelength. Eq. (1) states that an impedance match of the metasurface to its surrounding with respect the geometrical parameters a and b is essential in order to achieve complete acoustic energy absorption. In Eq. (1), the free space impedance Z0 = ρ0c0 where ρ0 and c0 are the mass density and speed of sound in air, respectively. There is a plethora of passive and active metamaterials solutions available capable of complete sound absorption13. For the realiza- tion of a one-sided compact and lightweight unhearability cloak we use Helmholtz resonators that can be fabricated to absorb sound at broad spectral windows. In the present case, we focus on the audible range and therefore begin the design by engineering sound absorption at a frequency of f = 3 KHz although the approach could be readily realized at other 4 desired frequencies. Strong air oscillations in the neck of these resonators in the presence of viscous losses are responsible for efficient energy dissipation. Conclusively, most absorbed acoustic energy is localized at the neck region, therefore, in order to fully camouflage an object, we pattern the rigid obstacle of radius a = 40 cm by Helmholtz resonators (Fig. 2(a)) and adjust the individual neck parameters w and t accordingly (See supplementary information for their values) to account for the angular variation. By computing the averaged acoustic pressure (cid:104)p(cid:105) and velocity normal to the resonator surface (cid:104)v⊥(cid:105) we are able to determine the impedance of the metasurface Zs = (cid:104)p(cid:105)/(cid:104)v⊥(cid:105). In the absence of acoustic backscattering, we predict total absorption of a plane wave at each individual Helmholtz resonator as displayed In Fig. 2(b) via their specific angular position. Correspondingly, we are able to explain full acoustic absorption via surface impedance matching as predicted in Eq. (1) where the real part of the relative metasurface impedance scales according to 1/ cos(θ) and its imaginary counterpart approaches zero for a vanishing gap separation a ≈ b. Acoustic gain adjustment Perfect absorption removes acoustic backscattering and is the first ingredient of a PT sym- metric system. The time-reversed image of this response constitutes acoustic amplification that we implement with an active electric circuit to control an semicircular array of loud- speakers (Fig. 3(a))21,27. In order to implement sound amplification of equal but opposite strength to the absorbing counterpart we must ensure to meet the condition of the ATR that dictates unidirectional-zero reflection at full transmission. Hence, beyond the need of balancing out the acoustic attenuation at the loss semi-shell with the gain counterpart that s (π− θ), is expressed through the PT symmetry of the entire metasurface cloak: Zs(θ) = −Z∗ we must ensure that the acoustic intensity profile is spatially symmetric. Thus,at the ATR it can be shown that I(θ) = I(π − θ), (2) which signifies that when the PT symmetric metasurface cloak is irradiated at the loss portion, the acoustic intensity in the nearest vicinity of an individual Helmholtz resonator (located at (π − θ)) equals the intensity at the exact opposite active loudspeaker (located at θ with respect to Fig. 3(a))25. This property accompanying the ATR condition is extremely useful when adjusting the individual loudspeakers to realize a one-way unhearability cloak. First, as detailed in the method section and the supplementary information, we placed the 5 two jointed semi-shells surrounding the rigid obstacle inside an acoustic waveguide whose rigid walls are covered by absorbing cotton. The acoustic source is formed by an array of loudspeakers that generate plane waves with frequency f = 3 KHz. In order to emit signals from the gain semi-shell perfectly synchronized in phase and amplitude with the impinging signal, a microphone measures the incoming sound field in front of the Helmholtz resonators whose phase and amplitude is processed through a phase shifting and amplifier circuit. The adjustment is performed in relation to the discrete and opposite locations of the Helmholtz resonator and loudspeaker couples whose intensity relation at the ATR is shown in Fig. 3(b). Due to size limitations of the source and the geometrical restrictions of the waveguide, our detection range exhibit an unitary intensity relation upto ±48◦ beyond which deviations start to grow. In other words, the subsequent cloaking experiment will be conclusively limited to this range. PT symmetry cloak Fig. 4(a) displays the metasurface cloak in its entirety comprising the jointed non-Hermitian semi-shells surrounding the acrylic obstacle of diameter 80 cm. At an operation frequency of 3 KHz corresponding to an acoustic wavelength of 11 cm, simulations and experimental measurements display how the pressure waves impinging the undecorated acrylic obstacle back-scatter at the side of irradiation but leave and almost soundless shadow at the obstacle's far-side (Fig. 4(b)). Contrary to this, when sound irradiates the non-Hermitian semi-shells that have been tuned to fulfill the aforementioned PT symmetry ATR condition, the acrylic obstacle, whose diameter is about 7 times larger than the acoustic wavelength, is acoustically camouflaged to match its surrounding via complete absorption, but, more importantly, the acoustic shadow gets eliminated through reconstruction of the impinging wave (Fig. 4(c)). The experimental measurements show that within the test areas, both in front and behind the obstacle, almost perfect plane waves have been sustained rendering the object to be hidden perfectly unhearable and concealed. Discussion Further improvement can be achieved by enlarging the loudspeaker array to launch a near- ideal plane wave. Also, decorating an object of arbitrary shape with gain and loss units in response to a point source or more complicated wave shapes greatly broadens the usage 6 of PT symmetry based acoustic cloaks. We implemented the proof of concept by means of Helmholtz resonators to suppress back reflected sound via resonant absorption. In analogy to the implemented active gain component, active loss control would enrich the possibility to eliminate back-scattering and to provide an acoustic camouflage dynamically at a wider spectral range. Extensions toward an acoustical concealment of three dimensional bodies by ultrathin non-Hermitian shells is another avenue worth pursuing. In conclusion, We have derived a theoretical recipe to realize an acoustic unhearability cloak via PT symmetry. By combining loss and gain structures, we showed that reflected sound is eliminated from an insonified body to be concealed and how it is reconstructed at the rear side of it via the anisotropic transmission resonance. Full wave simulations and measurement data support the theoretical predictions in creating a cloak based on a single but non-Hermitian shell structure. III. ACKNOWLEDGEMENTS J. C. acknowledges the support from the European Research Council (ERC) through the Starting Grant No. 714577 PHONOMETA and from the MINECO through a Ram´on y Cajal grant (Grant No. RYC-2015-17156). 7 IV. BIBLIOGRAPHY 1 Pendry, J. B., Schurig, D. & Smith, D. R. Controlling electromagnetic fields. Science 312, 1780 -- 1782 (2006). 2 Schurig, D. et al. Metamaterial electromagnetic cloak at microwave frequencies. Science 314, 977 -- 980 (2006). 3 Li, J. & Pendry, J. B. Hiding under the carpet: A new strategy for cloaking. Phys. Rev. Lett. 101, 203901 (2008). 4 Valentine, J., Li, J., Zentgraf, T., Bartal, G. & Zhang, X. An optical cloak mode of dielectrics. Nat. Mater. 8, 568 (2009). 5 Ergin, T., Stenger, N., Brenner, P., Pendry, J. B. & Wegener, M. Three-dimensional invisibility cloak at optical wavelengths. Science 328, 337 -- 339 (2010). 6 Ni, X., Wong, Z. J., Mrejen, M., Wang, Y. & Zhang, X. An ultrathin invisibility skin cloak for visible light. Science 349, 1310 -- 1314 (2015). 7 Al`u, A. Mantle cloak: Invisibility induced by a surface. Phys. Rev. B 80, 245115 (2009). 8 Cummer, S. A. & Schurig, D. One path to acoustic cloaking. New Journal of Physics 9, 45 (2007). 9 Torrent, D. & Sanchez-Dehesa, J. Acoustic cloaking in two dimensions: a feasible approach. New Journal of Physics 10, 063015 (2008). 10 Sanchis, L. et al. Three-dimensional axisymmetric cloak based on the cancellation of acoustic scattering from a sphere. Phys. Rev. Lett. 110, 124301 (2013). 11 Zigoneanu, L., Bogdan-Ioan, P. & Cummer, S. A. Three-dimensional broadband omnidirectional acoustic ground cloak. Nat. Mater. 13, 352 (2014). 12 Stenger, N., Wilhelm, M. & Wegener, M. Experiments on elastic cloaking in thin plates. Phys. Rev. Lett. 108, 014301 (2012). 13 Cummer, S. A., Christensen, J. & Alu, A. Controlling sound with acoustic metamaterials. Nat. Rev. Mat. 1, 16001 (2016). 14 Bender, C. M. & Boettcher, S. Real spectra in non-hermitian hamiltonians having pt symmetry. Phys. Rev. Lett. 80, 5243 -- 5246 (1998). 8 15 Bender, C. M., Brody, D. C. & Jones, H. F. Complex extension of quantum mechanics. Phys. Rev. Lett. 89, 270401 (2002). 16 Makris, K. G., El-Ganainy, R., Christodoulides, D. N. & Musslimani, Z. H. Beam dynamics in pt symmetric optical lattices. Phys. Rev. Lett. 100, 103904 (2008). 17 Guo, A. et al. Observation of pt-symmetry breaking in complex optical potentials. Phys. Rev. Lett. 103, 093902 (2009). 18 Ruter, C. E. et al. Observation of parity-time symmetry in optics. Nat. Phys. 6, 192 (2010). 19 Peng, B. et al. Parity-time-symmetric whispering-gallery microcavities. Nat. Phys. 10, 394 (2014). 20 Christensen, J., Willatzen, M., Velasco, V. R. & Lu, M.-H. Parity-time synthetic phononic media. Phys. Rev. Lett. 116, 207601 (2016). 21 Shi, C. et al. Accessing the exceptional points of parity-time symmetric acoustics. Nat. Commun. 7, 11110 (2016). 22 Wang, M., Ye, L., Christensen, J. & Liu, Z. Valley physics in non-hermitian artificial acoustic boron nitride. Phys. Rev. Lett. 120, 246601 (2018). 23 Gong, Z. et al. Topological phases of non-hermitian systems. Phys. Rev. X 8, 031079 (2018). 24 Lin, Z. et al. Unidirectional invisibility induced by PT-symmetric periodic structures. Phys. Rev. Lett. 106, 213901 (2011). 25 Ge, L., Chong, Y. D. & Stone, A. D. Conservation relations and anisotropic transmission resonances in one-dimensional pt-symmetric photonic heterostructures. Phys. Rev. A 85, 023802 (2012). 26 Regensburger, A. et al. Parity-time synthetic photonic lattices. Nature 488, 167 (2012). 27 Fleury, R., Sounas, D. & Alu, A. An invisible acoustic sensor based on parity-time symmetry. Nat. Commun. 6, 5905 (2015). 28 Sounas, D. L., Fleury, R. & Alu, A. Unidirectional cloaking based on metasurfaces with balanced loss and gain. Phys. Rev. Applied 4, 014005 (2015). 9 V. METHODS Numerical simulations. The full-wave simulations presented in the paper are performed using the finite-element solver Comsol Multiphysics. We employed the acoustic module comprising viscothermal losses and modelled the Helmholtz resonators as acous- tically rigid materials. The surface impedance is computed via the averaged acoustic pressure and normal velocity Zs = (cid:104)p(cid:105)/(cid:104)v⊥(cid:105) at the resonator surface. Acoustic gain has been imposed at the respective boundary through amplifying outgoing wave. Fabrication of the cloak. The obstacle to be cloaked was realized via a cylindrical acrylic shell with an outer diameter of 80 cm. The Helmholtz resonators responsible for the loss components and the active loudspeakers providing acoustic gain were mounted onto the shell. The individual Helmholtz resonators were 3D printed with thermoplastics whose geometrical parameters are tabulated in the supplementary information. Measurements. The experiment is carried out in a two dimensional waveguide with a uniform height of 6 cm. As can be seen in the illustration below, a line speaker array was FIG. 0: Experimental implementation of the parity-time symmetric metasurface cloak. employed to launch an incoming plane wave. In order to reduce unwanted reflections we covered the inner walls of the waveguide with sound-absorbing cotton. The experimental implementation is detailed in the supplementary information. 10 VI. FIGURES FIG. 1: (a) A rigid cylinder is covered by an ultrathin PT symmetric metasurface. The left (right) semicircle metasurface contain acoustic loss (gain) to fully absorb incoming (re-emit outgoing) sound waves. (b) Three scenarios are exemplified: No cloak, comprising strong back-scattering and shadow; camouflaging through complete absorption with a lossy semicircle metasurface only; cloaking via PT symmetry. 11 FIG. 2: (a) The insonified portion of the cloak (θ < π/2) is patterned by sound absorbing Helmholtz resonators. The angular dependence of their resonances has been tuned via the resonator neck width w and depth t. (b-d) The theoretically and numerically computed absorption and impedance match is presented in dependence to the angle θ, i.e, the position of the individual Helmholtz resonators. 12 FIG. 3: (a) The amplifying portion of the metasurface cloak has been decorated by loudspeakers that are controlled by gain circuits. (b) The anisotropic transmission resonance with omnidirectional full transmission and one-sided zero reflection, is accompanied by a spatial symmetry of the measured intensity profile I(θ) = I(π − θ), which was the experimental parameter for the gain adjustment. 13 FIG. 4: (a) Experimental realization of the PT symmetric metasurface cloak made out of two jointed semi-shells (radius a = 40 cm): (a) lossy Helmholtz resonator array and active loudspeakers controlled by gain circuits. (b) Full wave simulations of the pressure field of a bare rigid cylinder when insonified from the left by a plane wave at f = 3 kHz. The dashed test areas have been experimentally measured. (c) Simulations of the metasurface cloak surrounding the rigid obstacle and the corresponding measurements at the front and the backside of the decorated obstacle. 14
1809.05138	1	1809	2018-08-10T02:01:58	Microscope for X-ray orbital angular momentum imaging	[ "physics.app-ph", "physics.optics" ]	Orbital angular momentum (OAM) of photons is carried upon the wave front of an optical vortex and is important in physics research due to its fundamental degree of freedom. As for the interaction with materials, the optical OAM was shown to be transferred to the valence electron based on modified selection rules where a single ion is carefully aligned to the center of a vortex [Schmiegelow et al.(2016)]. We here demonstrate an elaborate way of extracting the distributed OAMs in the two-dimensional x-ray wave field at the exit-face of the specimen by a vorticity-sensitive microscopy, which detects vorticity in the reciprocal-space wave field downstream of the optical vortices. This method enables us to measure the distributed topological charges and OAMs by a single image with the wide field of view over 300 $\mu$m. Among various wavelength, the research of x-ray OAM is especially interesting since it could be closely related to the predicted OAM-induced x-ray dichroism [Veenendaal et al.(2007)].	physics.app-ph	physics	Microscope for X-ray orbital angular momentum imaging Yoshiki Kohmura,1 Kei Sawada,1 Masaichiro Mizumaki,2 Kenji Ohwada,3 and Tetsuya Ishikawa1 1RIKEN, SPring-8 Center, 1-1-1, Kouto, Sayo-gun, Sayo-cho, Hyogo 679-5148, Japan 2Japan Synchrotron Radiation Research Institute, 1-1-1 Kouto, Sayo, Hyogo, 679-5198, Japan 3National Institutes for Quantum and Radiological Science and Technology (QST), 1-1-1, Kouto, Sayo-gun, Sayo-cho, Hyogo 679-5148, Japan Orbital angular momentum (OAM) of photons is carried upon the wave front of an optical vortex and is important in physics research due to its fundamental degree of freedom. As for the interaction with materials, the optical OAM was shown to be transferred to the valence electron based on modified selection rules where a single ion is carefully aligned to the center of a vortex[1]. We here demonstrate an elaborate way of extracting the distributed OAMs in the two-dimensional x- ray wave field at the exit-face of the specimen by a vorticity-sensitive microscopy, which detects vorticity in the reciprocal-space wave field downstream of the optical vortices. This method enables us to measure the distributed topological charges and OAMs by a single image with the wide field of view over 300 µm. Among various wavelength, the research of x-ray OAM is especially interesting since it could be closely related to the predicted OAM-induced x-ray dichroism[2]. The intensity profile with a central zero, around an optical vortex, is elaborately utilized in a modern super- resolution microscope[3, 4] and in nano-fabrication well under the scale of the diffraction limit[5 -- 9]. On the other hand, Orbital Angular Momentum (OAM), carried by an optical vortex[10 -- 13], is accelerating its significance recently. Since quantized OAM number ranges between minus and plus infinity, OAM would be useful for enlarg- ing the information channels[14] in a remarked contrast with the spin angular momentum number of ±1. To detect OAM of photons, a vorticity-sensitive mea- surement of the wave field, e.g. downstream of the spec- imens, is highly required. Imaging microscopy, making use of the radial Hilbert transform (RHT), is best suited for this purpose due to its sensitivity to the gradient of complex amplitude, phase and of amplitude, at the exit- face of specimens[15]. We hereafter show our experiment using the RHT which enabled us for the first time to extract the distributed OAMs and to discriminate their topological charge at the exit-face of specimens. The RHT was invented to operate on the input com- plex amplitudes of two dimensional complex data in a similar manner as the Hilbert transform (HT) for one dimensional data. Three operations on the input com- plex amplitudes, Fourier transform, multiplication of a phase filter in the reciprocal space and inverse Fourier transform, results in complex amplitudes that equal to the convolution of the input complex amplitudes and the inverse Fourier transform of the phase filter. In the RHT, a spiral phase filter with the jump of 2lπ is used as the phase filter in the wave-number space, where l is a quantum number. This transform produces an edge en- hanced image of the complex transmissivity of the spec- imen [Fig. 1(a)]. The RHT is sensitive to the large gra- dient existing in the distribution of complex amplitude. In the RHT, the phase jump of 2lπ is preserved during the inverse Fourier transform on the spiral phase filter as shown in Fig.1 (a, b). This results in interesting sort- FIG. 1. (a) The filter G(ω) in HT is a step function of π in angular frequency space. g(t) is inverse Fourier transform of G(ω). (b) The filter in RHT is a spiral phase filter with jump of 2lπ in wave-number space. g((cid:126)r) is inverse Fourier transform of G((cid:126)k). Edge enhancement is observed. The function of Spi- ral Fresnel zone plate (SFZP) is indicated by the dotted box in purple. (c) The spiral phase anomalies with jump of 2mπ is discriminated in RHT when m=l is realized. Amplitude & phase are expressed by brightness & hue, respectively. ing possibilities of objects having spiral phase anomalies, e.g. with the exit-face wave field of ψ1((cid:126)r) ∝ eimφ hav- ing a phase jump of 2mπ where m is a quantum num- ber. The complex wave field at the reciprocal space is formed by the Fourier transform of the exit-face wave field [Ψ((cid:126)k), middle of Fig. 1(c)] which is multiplied by passing through the spiral phase filter with the complex transmissivity of G((cid:126)k) = eilφk , where φk is the azimuthal angle in the reciprocal space. Our new theoretical anal- ysis revealed that the output complex wave field, Ψml, formed by the RHT shows the intensity enhancements only when the quantum numbers of m and l coincide (expressed by δml) using a single image, which also has position dependences (expressed by Ψml((cid:126)r)) in the fol- lowing equation [see Fig. 1(b)], Ψml((cid:126)r) = F −1 [Ψ((cid:126)k)eilφk ] = δml Ψml((cid:126)r). (1) Note that these equations are for the pure spiral phase filter and the objects with the pure spiral phase anomaly, while the neglected effect due to the distribution of atten- uation will be reduced by using low-Z element for these structures. We moved on to show the possibility of deter- mining the quantum number of a specimen using a single image while the previous theory suggested the measur- ability of the rotation of probability current density of the exit wave front using two images[15]. In the present experiment, a Spiral Fresnel zone plate (SFZP)[16] was utilized to function both as the lens for performing the inverse Fourier transform and as a spiral phase filter[17]. In this paper, we verify for the first time that the RHT method is powerful to sort and analyze the OAM distri- bution formed by the specimens. FIG. 2. (a) Setup of topological charge microscope with an objective SFZP lens placed at the focal plane of the FZP. The magnification factor of this microscope is 1.09. (b) Scanning Electron Microscope image of the SFZP used in the exper- iment. The scale bar corresponds to 100 µm. (c) Edge en- hanced image of copper mesh, without hosting any finite topo- logical charge, using the topological charge microscope, with the setup shown in 2(a), exhibiting the rim of the aperture. The exposure time was 5 seconds. The scale bar corresponds to 50 µm. The quantum nature of OAM corresponds to (i) a complete circulation of the twisting of the vector nor- mal to the wave front that is Poynting vector, (cid:126)S = 2 (cid:126)E∗((cid:126)r) × (cid:126)H((cid:126)r) + c.c. and (ii) the energy current density around an anomaly. The orientation and the number of this circulation correspond to the sign and the topological charge of the formed optical vortex. We construct a RHT microscope using SFZP as the objective lens where the phase shift aside from the trans- missivity of the specimen is observed at the image plane, which enables us to discriminate the sign and the topo- logical charge of the generated vortices [Fig. 1(c)]. We hereafter call this microscope a topological charge micro- scope. We can use two topological charge microscope images (I l=1 and I l=−1) obtained with reversed orientations of the SFZP objective lens. The difference of these two topological charge microscope images exhibits the distri- bution of rotation of Poynting vector (cid:126)S at the exit-face of the specimen[15]. (cid:104) ∂ψ∗ ∂y (cid:105) I l=1−I l=−1 ∝ i((cid:126)∇× (cid:126)S)z ∝ i − ∂ψ∗ ∂x ∂ψ ∂x ∂ψ ∂y = Ωxy, (2) where Ωxy is a Berry curvature term. Berry curva- ture, the fictitious magnetic field acting on the wave[18], is an important concept applied in various fields of physics such as Hall effect, spin Hall effect[19], optical Hall effect[20], x-ray translation effect inside deformed crystals[21 -- 24]. X-ray microscopy experiment was performed at Ex- perimental Hutch 2 along BL29XUL[25], an undulator beamline of SPring-8 facility in Japan, using 7.71 keV x- rays from a double crystal monochromator. The higher order reflection light of monochromator was reduced by a pair of total reflection mirrors. The gap of the undulator source was set to 12.207 mm to equalize the peak in- tensity of 1st order harmonics to 7.71 keV. The x-ray images have been taken with an indirect x-ray image sensor[26, 27]. An objective SFZP lens was placed at the x-ray focal plane of the upstream Fresnel Zone Plate (FZP with focal length of 1.72 m) while the x-rays trans- mitted through the specimen between FZP and SFZP [Fig. 2(a)]. The magnification ratio of the microscope was set to be 1.09 by selecting the distances between sample-SFZP and SFZP-detector to be 1.40 m and 1.52 m, around twice the focal lengths of SFZP of 0.73 m[28]. The zone depth of 1.84 µm was chosen for the tantalum FZP to realize π-phase shift at 7.71 keV. Using this setup, the wave field that corresponds to Fourier transform of the specimen's complex transmissivity is irradiated on the SFZP plane[29, 30]. This enables us to observe ob- jects with a wide range of spatial length scales using a topological charge microscope [Fig. 1(b,c)]. Two blades were placed in front of the specimen and of the imaging detector so that the contributions from the other unnec- essary orders of light from SFZP, aside from 1st order, are excluded in the setup of Fig. 2(a). The field of view was reduced to approximately one half by this order selection of the light, approximately equalizing to the radius of SFZP which is around 324 µm. First, a copper mesh with the pitch of 12.5 µm, which contains no phase anomalies or nor finite topological charge, was observed using the topological charge mi- croscope. As shown in Fig. 1(b), only an edge enhaced image with the bright open squares at the rims of the mesh aperture was observed [Fig 2(c)]. We then move on to observe objects that yields OAMs or finite topological charges at the exit-face wave field us- ing the topological charge and its differential microscope. For this purpose, we fabricated a specimen having four spiral phase filters at the corners of a square on a silicon substrate[31] with the maximum depth of 19.5 µm, cor- responding to the phase shift of 2π for 7.71 keV x-rays. The radii of spiral phase filters and the distances between the neighbors were set to be 34 µm and 80 µm, respec- tively. Spiral orientation of the phase filters (specimen) was defined according to the phase advance in a clock- 3 wise (m=-1) or in an anti-clockwise (m=1) orientations from downstream. This specimen was placed towards the downstream and the x-ray vortices with reversed OAM were realized among the neighbouring spiral phase filter [Fig. 3(a)]. The images of this specimen were obtained for two set- ting of the SFZP objective lens [Fig. 2(a)], one in an anti- clockwise (l = 1) and the other in a clockwise (l = −1) orientation from downstream, where l corresponds to the topological charge of the SFZP [see Figs. 3(b) and 3(c)]. The result clearly indicates that the bright spots are formed at the centers of the spiral phase filters (specimen) when the quantum numbers of the spiral phase filters and SFZP coincides (m = l) and are not when they are differ- ent (m = −l). The results shown in Figs. 3(b) and 3(c) agree well with eq. (1). The attenuation through these structures was not taken into account in eq.(1), but this effect was not significant as shown in these figures. The differential image obtained from these two images, which visualizes the distribution of Berry curvature at the exit- face of the specimen, is shown in Fig. 3(d). Remarkable structures at the centers and at the rims of the spiral phase filters were exhibited which were also clearly visi- ble in the calculated differential image shown in Fig. 3(e), derived from eq. (2). Note that Fig. 3(e) was calculated by taking account of the attenuation through the struc- tures of spiral phase filters in the specimen. The developed microscope, which visualizes the spatial distributions of OAM from one or two images, is free from the extremely precise alignment. The present microscope will open the door of various OAM related researches e.g., x-ray OAM-induced dichroism[2] or further modification of the selection rules that govern the transfer of OAM between valence electrons and photons[1] and so on. The chirality research[32] in atomic resolution would be also accelerated. Acknowledgments Experiments at SPring-8 BL29XU have been per- formed with the approval of RIKEN (under proposal numbers of 20170065). Y. K. express a sincere gratitude to Drs. L. Szyrwiel and A. Takeuchi for discussions. (a) Image of multiple spiral phase filters on the cor- FIG. 3. ners of a square by Scanning Electron Microscope (SEM) with the magnification factor of M=550. Topological charge micro- scope images obtained with the SFZP objective lens having the quantum OAM number of (b) l=1 (anti-clockwise) and (c) l=-1 (clockwise from downstream) and the exposure time of 5 seconds. (d) Differential microscope image (I l=1-I l=−1) of two images in 3(b) and 3(c). (e) Calculated differential microscope image using eq. (2). [1] C. T. Schmiegelow, et al. Nature Communications 7, 129981-129986 (2016). [2] M. V. Veenendaal, and I. McNulty, Phys. Rev. Lett., 98, 157401 (2007). [3] S. W. Hell, and J. Wichmann, Opt. Lett. 19, 780 (1994). [4] S. Hell, Nature BioTechnology 21, 1347 (2003). [5] Y. Kohmura, et al. Appl. Physics Lett. 112, 1231031 (2018). [6] A. Ambrosio, et al. Nat. Com., 3, 989 (2012). [7] F. Takahashi, et al. Scientific Reports, 6, 21738 (2016). [8] K. Toyoda, et al. Nano Letters, 12, 3645 (2012). [9] K. Toyoda, et al. Phys. Rev. Lett., 110(14), (2013). [10] M. Padgett, J. Courtial and L. Allen, Physics Today, 57 (5), 35 (2004). [11] M. Zurch, et al. Nature Physics 8, 743 (2012). [12] A. G. Peele, et al. Opt. Lett. 27, 1752 (2002). [13] Y. Taira, et al. Scientific Reports 7, 5018 (2017). [14] Z-Y. Zhou, et al. Light: Science & Applications 5, e16019 (2016). [15] R. Juchtmans, et al., Phys. Rev. A, 94, 023838 (2016). [16] A. Sakdinawat, et al. Opt. Lett. 32, 2635-2637 (2007). [17] SFZP is a modified FZP, designed to function both as the lens and as the spiral phase filter, having the complex transmissivity expressed in the following equation SFZPl(r, φ) = exp(ilφ − iπr2 λf ), (3) where l, φ, λ and f are the OAM quantum number, the azimuthal angle around the center of vortex, the x- ray wavelength and the focal length of SFZP[16], respec- tively. The fabricated SFZP is binalized from the function expressed above. [18] H. Kurosawa, K. Sawada, and S. Ohno, Phys. Rev. Lett., 117, 083901 (2016). [19] O. Hosten, & P. Kwiat, Science. 319, 787 (2008). [20] M. Onoda, S. Murakami, & N. Nagaosa, Phys. Rev. Lett. 93, 083901 (2004). [21] K. Sawada, S. Murakami, & N. Nagaosa, Phys. Rev. Lett. 96, 154802 (2006). [22] Y. Kohmura, K. Sawada, & T. Ishikawa, Phys. Rev. Lett. 104, 244801 (2010). [23] Y. Kohmura, K. Sawada, S. Fukatsu, & T. Ishikawa, Phys. Rev. Lett. 110, 057402-1 (2013). [24] D. Takei, Y. Kohmura, K. Sawada, & T. Ishikawa, Optics Express 24(21), 24544 (2016). 4 [25] K. Tamasaku, et al. Nucl. Instrum. Methods Phys. Res. A467-468, 686-689 (2001). [26] T. Kameshima, et al., AIP Conf. Proc., 1741, 040033 (2016). [27] The x-ray images have been taken with an indirect x- ray image sensor composed of a fluorescence screen made from a LuAG single crystal, an optical lens system and a CMOS camera[26]. The effective pixel size of the detector was 325 nm. [28] The utilized FZP and SFZP were fabricated by the Shinko-Seiki Corp. (Kobe, Japan). The specifications of FZP and SFZP were the diameter of 2rN = 664 µm & 648 µm, the first zone radius of r1=16.6 µm & 10.8 µm, the outermost zone width of ∆rN = 0.415 µm & 0.18 µm, number of zones of 400 & 900, respectively. The 7.71keV x-ray focal lengths are 1.72 m and 0.73 m, re- spectively. These ZPs were formed on a membrane on the 500 µmt silicon substrate, back edged to have the thickness of (cid:39) 80 µm. [29] J. W. Goodman, Introduction to Fourier Optics, 2nd Edi- tion, McGraw-Hill (1996). [30] A topological charge microscope requires the Fourier transform of the specimen's complex transmissivity to be irradiated on the SFZP [Fig. 1(b, c)]. The propagated wave field between the specimen and the SFZP, how- ever, is affected by the additional Fresnel diffraction term downstream of the specimen. This term is cancelled by placing a focusing optics upstream of the specimen and by placing the SFZP objective lens at the corresponding focal plane[16, 29] as shown in Fig. 2(a). [31] The specimen with four spiral phase filters were fabri- cated by the Shinko-Seiki Corp. (Kobe, Japan) on a mem- brane formed on a silicon substrate back edged to have the thickness of (cid:39) 80 µm. [32] W. Brullot, et al. Sci. Adv. 2, e15013491 (2016).
1811.01418	1	1811	2018-11-04T19:02:13	Magnetic Nanoparticles in Nanomedicine	[ "physics.app-ph" ]	Nanomaterials, in addition to their small size, possess unique physicochemical properties that differ from the bulk materials, making them ideal for a host of novel applications. Magnetic nanoparticle (MNP) is one important class of nanomaterials that have been widely studied for their potential applications in nanomedicine. Due to the fact that MNPs can be detected and manipulated by remote magnetic fields, it opens a wide opportunity for them to be used in vivo. Nowadays, MNPs have been used for diverse applications including magnetic biosensing (diagnostics), magnetic imaging, magnetic separation, drug and gene delivery, and hyperthermia therapy, etc. This review aims to provide a comprehensive assessment of the state-of-the-art biological and biomedical applications of MNPs. In addition, the development of high magnetic moment MNPs with proper surface functionalization has progressed exponentially over the past decade. Herein, we also reviewed the recent advances in the synthesis and surface coating strategies of MNPs. This review is not only to provide in-depth insights into the different synthesis, biofunctionalization, biosensing, imaging, and therapy methods but also to give an overview of limitations and possibilities of each technology.	physics.app-ph	physics	Magnetic Nanoparticles in Nanomedicine Kai Wu1, Diqing Su2, Jinming Liu1, Renata Saha1, and Jian-Ping Wang1,* 1Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, Minnesota 55455, USA 2Department of Chemical Engineering and Material Science, University of Minnesota, Minneapolis, Minnesota 55455, USA E-mail: [email protected] (Dated: November 4, 2018) Abstract: Nanomaterials, in addition to their small size, possess unique physicochemical properties that differ from the bulk materials, making them ideal for a host of novel applications. Magnetic nanoparticle (MNP) is one important class of nanomaterials that have been widely studied for their potential applications in nanomedicine. Due to the fact that MNPs can be detected and manipulated by remote magnetic fields, it opens a wide opportunity for them to be used in vivo. Nowadays, MNPs have been used for diverse applications including magnetic biosensing (diagnostics), magnetic imaging, magnetic separation, drug and gene delivery, and hyperthermia therapy, etc. This review aims to provide a comprehensive assessment of the state-of-the-art biological and biomedical applications of MNPs. In addition, the development of high magnetic moment MNPs with proper surface functionalization has progressed exponentially over the past decade. Herein, we also reviewed the recent advances in the synthesis and surface coating strategies of MNPs. This review is not only to provide in-depth insights into the different synthesis, biofunctionalization, biosensing, imaging, and therapy methods but also to give an overview of limitations and possibilities of each technology. Keywords: magnetic nanoparticle, nanomedicine, magnetic biosensing, magnetic imaging, magnetic separation. Table of Contents 1 2 Introduction .................................................................................................................................................... 3 Physical Properties of Magnetic Nanoparticles (MNPs).............................................................................. 3 2.1 2.2 From Bulk Material to Nanoparticle .................................................................................................. 3 From Multi-domain to Single-domain MNPs..................................................................................... 5 2.3 Superparamagnetism ............................................................................................................................ 7 2.4 Magnetic Susceptibility ........................................................................................................................ 9 2.5 Magnetic Relaxivity ............................................................................................................................ 10 1 2.6 Dipolar Interactions ............................................................................................................................ 12 2.7 Multicore MNPs .................................................................................................................................. 12 3 Synthesis of MNPs ....................................................................................................................................... 13 3.1 Ball Milling Method ............................................................................................................................ 14 3.2 Gas-phase Condensation (GPC) Method .......................................................................................... 17 3.3 Other Nanoparticle Fabrication Approaches ................................................................................... 20 4 Surface Coating Strategies .......................................................................................................................... 22 4.1 Organic Coating .................................................................................................................................. 22 4.2 Inorganic Coating ............................................................................................................................... 24 5 MNPs for Diagnosis ..................................................................................................................................... 25 5.1 Magnetoresistive (MR) Sensors ......................................................................................................... 25 5.1.1 Giant Magnetoresistive (GMR) Biosensors ...................................................................................... 25 5.1.2 Magnetic Tunnel Junction (MTJ) Biosensors ................................................................................... 28 5.2 Micro-Hall (μ Hall) Sensors ............................................................................................................... 29 5.3 NMR-based Diagnostics ..................................................................................................................... 30 5.3.1 Magnetic Relaxation Switching (MRSw) Assay ................................................................................ 30 5.3.2 Magnetic Resonance Imaging (MRI) ................................................................................................ 32 5.4 Superparamagnetism-based Diagnostics .......................................................................................... 35 5.4.1 Brownian Relaxation-based Assay ................................................................................................... 35 5.4.2 Magnetic Particle Imaging (MPI) .................................................................................................... 40 5.5 Surface Enhanced Raman Spectroscopy (SERS) Systems .............................................................. 43 6 MNPs for Therapeutic Applications............................................................................................................ 46 6.1 Drug Delivery ...................................................................................................................................... 46 6.2 Hyperthermia Therapy ...................................................................................................................... 49 7 MNPs for Bioseparation and Manipulation ............................................................................................... 52 7.1 Magnetic Separation ........................................................................................................................... 52 7.2 Magnetic Manipulation ...................................................................................................................... 54 8 Conclusions and Future Perspectives ......................................................................................................... 54 Author Information.............................................................................................................................................. 55 Acknowledgements ............................................................................................................................................... 55 References ............................................................................................................................................................ 55 2 1 Introduction Magnetic nanoparticles (MNPs), with the size between 1 nm and 100 nm, are one important nanomaterial for science and technology in the past two decades. The unique characteristics of MNPs, such as high surface to volume ratio and size-dependent magnetic properties, are drastically different from those of their bulk materials. MNPs have been receiving tremendous attention in multiple areas such as data storage, spintronics, catalyst, neural stimulation, and gyroscopic sensors, etc.1-20 Nowadays, many methods have been developed to fabricate MNPs due to their wide applications, and there are mainly two approaches to obtain MNPs: top-down approach and bottom-up approach. MNPs with a properly functionalized surface can be physically and chemically stable, biocompatible, and environmentally safe. In addition, biological samples exhibit virtually no magnetic background, thus high sensitivity measurements can be performed on minimally processed samples in MNP- based biological and biomedical applications. Thus, MNPs have been successfully applied as contrast enhancers in magnetic resonance imaging (MRI) 21-26, tracer materials in magnetic particle imaging (MPI) 27-33, tiny heaters in magnetic hyperthermia 34-37, carriers for drug/gene delivery 38-43, tags for magnetic separation 12, 44-49, etc. The ease of synthesis and facile surface chemistry, with comparable size to biologically important ligands, has generated much eagerness in applying MNPs to clinical diagnostics and therapy 50-58. Herein, we have reviewed different strategies for the synthesis and biofunctionalization of MNPs, as well as the diagnostic and therapeutic applications of MNPs. Our aim is to discuss the challenges of working with MNPs while giving an overall overview of the state-of-the-art. 2 Physical Properties of Magnetic Nanoparticles (MNPs) 2.1 From Bulk Material to Nanoparticle MNPs, with comparable sizes to biologically important objects 59, have demonstrated unique properties such as larger surface-to-volume ratio, excellent reactivity, exceptional magnetic response compared to their bulk materials 60, 61. Materials such as pure metals (Fe, Co, Ni, etc.), alloys (FeCo, alnico, permalloy, etc.), and oxides (Fe3O4, -Fe2O3, CoFe2O4, etc.) with high saturation magnetizations are preferred for the production of MNPs. Although pure metals are able to yield higher saturation magnetizations, they are not suitable for clinical use due to the high toxicity and oxidative properties. One of the most widely used MNP types is iron oxide considering the high chemical and colloidal stability, high biocompatibility, and low cost. Magnetic properties of MNPs, such as magnetic moment and anisotropy constant, depend strongly on their size and shape. Magnetic moment (𝜇) is the product of magnetization (𝑀) and magnetic core volume 𝑉𝑚 which is the 3 most important property of MNP for nanomedicine related applications. A higher magnetic moment yields a more pronounced detection signal (for magnetic biosensing and imaging-related applications) as well as a larger magnetic force (for magnetic manipulation and drug/gene delivery related applications) 62, 63. Due to the lack of translational crystal symmetry in the surface layer (also known as surface spin-canting effect) 64-67, the magnetic properties of surface layer are very different from the interior region of MNP, as a result, decreased saturation magnetizations 𝑀𝑠 and increased anisotropy constants are observed in MNPs compared to their corresponding bulk materials. This spin-canting effect is not affected by the organic capping but the magnetic core size of MNP 68 and the particle synthesis methods 69. It can be understood in terms of a core-shell structure (as shown in Figure 1(a)) where the spins in the magnetic core are fully aligned along the applied magnetic field while the spins in the shell are canted relative to the field. For a spherical MNP, its saturation magnetization can be modeled by 70: Where 𝑀𝑠𝑏 is the saturation magnetization of the bulk material, 𝐷 is the diameter of the magnetic core, and 𝛿 is 𝑀𝑠 = 𝑀𝑠𝑏(1 − 2𝛿 𝐷⁄ )3 (1) the thickness of the spin canting layer. Dutta et al. 71 have reported the fitted values of 𝛿 = 0.68 𝑛𝑚 and 𝑀𝑠 = 92 𝑒𝑚𝑢 ∙ 𝑔−1 (~4.76 × 105 𝐽 𝑚3𝑇 ) for magnetite nanoparticles based on experimental results. ⁄ Figure 1. (a) Schematic view of general spin canting geometry, the core-shell model. (b) Theoretical 𝑀𝑠 (blue solid line) and 𝐾𝑒𝑓𝑓 (orange solid line) versus magnetite nanoparticle diameter 𝐷 at 300 K. Horizontal dash-dot lines exhibit 𝑀𝑠𝑏 and 𝐾𝑏, respectively. Theoretical data are compared with experimental ones. Blue diamonds from 72, blue hexagons from 73, orange triangle from 74, orange squares from 75, orange pentagon from 31, and orange circle from 76. ((b) reprinted with permission from reference 67, copyright 2017 IOP Publishing) On the other hand, due to the spin-canting effect, the observed anisotropy constants from MNPs are always larger than their corresponding bulk materials 67. The effective anisotropy constant 𝐾𝑒𝑓𝑓 is modeled by 77: 4 𝐾𝑒𝑓𝑓 = 𝐾𝑏 + ( 6Φ 𝐷 ) 𝐾𝑠 (2) Where, 𝐾𝑏 and 𝐾𝑠 are the bulk and surface anisotropy constants, respectively. For spherical particles, Φ = 1. Demortiere et al. 77 have reported the fitted values of 𝐾𝑏 = 1.04 × 105 𝑒𝑟𝑔 ∙ 𝑐𝑚−3 (~1.04 × 104 𝐽 𝑚3⁄ magnetite, and 𝐾𝑠 = 3.9 × 10−2 𝑒𝑟𝑔 ∙ 𝑐𝑚−2 (~3.9 × 10−5 J m2⁄ ). The effective anisotropy is also affected by the ) for bulk MNP shape (i.e. shape anisotropy), for spherical nanoparticles, shape anisotropy is negligible compared to surface anisotropy. These two models in equations (1) & (2) have been applied to predict the magnetic properties of magnetite nanoparticles with respect to magnetic core diameters (𝐷) , and they were examined by comparing with experimental data from literature available (see Figure 1(b)). 2.2 From Multi-domain to Single-domain MNPs Bulk magnetic materials are composed of microscopic crystalline grains (also known as polycrystalline). Each grain is single crystal, with crystal lattices oriented in different directions, has an easy axis of magnetization. To reduce the magnetostatic energy, each grain spontaneously divides into many magnetic domains separated by domain walls, called multi-domain state. The magnetizations of different domains point in different directions in a more or less random manner while the magnetization within each domain points uniformly to one direction. As a single crystal grain splits into multi-domains, the magnetostatic energy reduces but meanwhile extra energy (domain wall energy) is required. Magnetic domains stabilize to a critical size 𝐷𝑐𝑟𝑖𝑡 when the energy cost of generating an additional domain wall is equal to the magnetostatic energy saved. This critical size varies for different magnetic materials (proportional to(𝐴 2𝐾⁄ )1 2⁄ ) which is determined by the material properties such as exchange stiffness (𝐴) and anisotropy constant (𝐾), usually, the critical size is between 10 nm and 100 nm. The critical size range is bounded by a lower limit (superparamagnetism) and a higher limit (multi-domain), see Figure 2(a). The critical sizes for the observation of superparamagnetism, single-domain, and multi-domain behaviors of a variety of common ferromagnetic fine nanoparticles can be found from reference 78. 5 Figure 2. (a) Transition from superparamagnetic to the multi-domain region. The inset figure shows qualitative behaviors of the size-dependent coercivities of MNPs. 𝐷𝑠𝑝 and 𝐷𝑐𝑟𝑖𝑡 are, respectively, the transition sizes from superparamagnetic to a blocked state, and from single-domain to multi-domain regions. (b) Magnetization curve of superparamagnetic nanoparticles measured at room temperature by VSM. Superparamagnetic nanoparticle shows zero magnetic coercivity. (c) Schematic view of energy barrier and thermal fluctuation. ((b) reprinted with permission from reference 79, copyright 2015 AIP Publishing LLC) Single-domain MNPs are particles with internal magnetizations pointing uniformly in one direction, thus, these particles bear magnetizations equal to their spontaneous magnetization 𝑀𝑠, and they have the largest possible magnetic moment of 𝜇 = 𝑉𝑚𝑀𝑠, where 𝑉𝑚 is the magnetic core volume of the particle. The magnetization of a single-domain particle rotates as one single giant magnetic moment (called the macro-spin approximation) under external magnetic fields and its hysteresis has been well described by the Stoner-Wohlfarth model. 6 Due to the magnetic anisotropy, the magnetic moment of a single-domain MNP has two preferred orientations which are antiparallel to each other and are both aligned along its "easy axis". These two preferred directions are energy minimums separated by an energy barrier of 𝐸𝑏 = 𝐾𝑒𝑓𝑓𝑉𝑚, which prevents the magnetic moment from flipping from one direction to the other. However, thermal fluctuations cause the magnetic moment to rotate in a random manner. As is shown in Figure 2(c), under a finite temperature 𝑇, if the energy barrier 𝐸𝑏 is comparable to or smaller than the thermal fluctuation energy 𝑘𝐵𝑇, the magnetic moment jumps from one direction to the other frequently during a measurement time period of 𝜏𝑚, then the observed net magnetization is zero, resulting in superparamagnetism. Under a specific measurement time 𝜏𝑚 and temperature 𝑇, there is a critical size 𝐷𝑠𝑝 at which the transition from single-domain to superparamagnetic nanoparticle occurs. Again, this critical size 𝐷𝑠𝑝 varies for different magnetic materials, usually in the range of several to several tens nanometers. Due to the fast flipping of magnetic moments in superparamagnetic nanoparticles, they show zero magnetic moments in the absence of an external magnetic field. When an external field is applied, their magnet moments tend to align along the field resulting in nonzero net magnetization, the magnetic moment of superparamagnetic nanoparticle vs applied field is a reversible S-shaped curve as shown in Figure 2(b). 2.3 Superparamagnetism At a finite temperature 𝑇 , there is a finite probability that the magnetic moment of a superparamagnetic nanoparticle flips between two preferred directions. The mean time between two flips is called zero-field Néel relaxation time expressed as: 𝜏𝑁 = 𝜏0𝑒𝑥𝑝 ( 𝐾𝑒𝑓𝑓𝑉𝑚 𝑘𝐵𝑇 ) (3) Where, 𝜏0 is attempt time (attempt period), which is around 10−10~10−9 seconds depending on the material, and 𝑘𝐵 is Boltzmann constant. It can be seen from equation (3) that the Néel relaxation time is an exponential function of the particle size, and it can vary from nanoseconds for nanoparticles to years for larger particles or bulk materials. Néel relaxation process is the rotation of magnetic moment inside a stationary MNP (see Figure 3(b)). 7 Figure 3. (a) Magnetic core diameter 𝐷 and organic capping layer with thickness 𝑑. (b) Néel relaxation process is the rotation of magnetic moment inside a stationary particle. (c) Brownian relaxation process is the rotation of entire particle with its magnetic moment. (d) Néel, Brownian, and effective relaxation time as a function of MNP core diameter, 𝑇 = 293 𝐾. Viscosity of the MNP solution is assumed to be 𝜂 = 1 𝑐𝑝. (e) The AC magnetic susceptibility has two components: in-phase and out-of-phase. The figure shows the normalized 𝜒′ and 𝜒′′ as functions of 𝜔𝜏 , the out-of-phase component 𝜒′′ reaches a maximum when 𝜔𝜏 = 1 . ((d) reprinted with permission from reference 79, copyright 2015 AIP Publishing LLC) However, for most biomedical applications, superparamagnetic nanoparticles are dispersed in suspensions (also called ferrofluid) where the magnetic relaxivity is a joint effect of both Néel and Brownian processes (see Figure 3(b) & (c)) 67, the zero-field Brownian relaxation time is given as: 𝜏𝐵 = 3𝜂𝑉ℎ 𝑘𝐵𝑇 (4) where 𝜂 is the fluid viscosity, 𝑉ℎ is the hydrodynamic volume of MNP. The Néel and Brownian relaxation models from equations (3) & (4) are quite simplified, representing a general guideline of non-interacting superparamagnetic nanoparticles under zero magnetic field. Within the assumption of independence of Néel and Brownian processes, the effective relaxation time is given as: 8 𝜏 = 𝜏𝑁𝜏𝐵 𝜏𝑁+𝜏𝐵 (5) Figure 2(b) shows the magnetization curve of an ensemble of 25 nm iron oxide superparamagnetic nanoparticles measured by Vibrating Sample Magnetometer (VSM) at room temperature 79. Since above the blocking temperature 𝑇𝐵, the measurement time 𝜏𝑚 is larger than 𝜏𝑁, so the nanoparticle appears to be in the superparamagnetic state and its magnetic moment flips several times during one measurement period, thus, the measured magnetization is averaged to zero when the external field 𝐻 = 0 𝑂𝑒. As the external magnetic field is applied, the magnetic moments of nanoparticles tend to align with the field direction, resulting in a net magnetization. Hence, superparamagnetic nanoparticles show paramagnetic behavior under small magnetic fields, and the magnetization curve is a reversible S-shape, which is usually simplified by the Langevin model: where 𝜇0 is the magnetic permeability of vacuum, 𝐿(𝑥) is the Langevin function, 𝐻 is the applied magnetic field. 𝑀(𝐻) = 𝑀𝑆𝐿 (𝜇0 𝜇𝐻 𝑘𝐵𝑇 ) (6) 2.4 Magnetic Susceptibility Magnetic susceptibility 𝜒 is a dimensionless proportionality constant that describes the degree of magnetization of a material in response to external magnetic field, it is the ratio of magnetization 𝑀 to the field 𝐻. When subjected to alternating current (AC) magnetic field, the magnetization of MNP may not be able to follow the AC field due to its finite rate of magnetic relaxation, thus, a phase delay between the AC field and the magnetization is introduced. This property introduces a complex magnetic susceptibility 𝜒(𝜔), which can be calculated by the Debye model 80: and 𝜒(𝜔) = 𝜒0 1+𝑗𝜔𝜏 = 𝜒′ + 𝑗𝜒′′ = 𝜒𝑒𝑗𝜑 (7) 𝜒′ = 𝜒′′ = 𝜒0 1+(𝜔𝜏)2 (8) 𝜒0𝜔𝜏 1+(𝜔𝜏)2 (9) 𝜑 = 𝑡𝑎𝑛−1(𝜔𝜏) (10) where 𝜒0 is the DC (direct current) field susceptibility, 𝜔 is the angular frequency of AC field, 𝜒′ and 𝜒′′ are the in-phase and out-of-phase components, 𝜑 is the phase delay. Figure 3(e) shows the normalized 𝜒′ and 𝜒′′ as functions of 𝜔𝜏, the out-of-phase component 𝜒′′ reaches a maximum when 𝜔𝜏 = 1, this property is exploited for monitoring the binding of target biomolecules to MNPs in magnetic relaxometry based biosensors. Furthermore, 𝜒′′ also holds great significance in magnetic hyperthermia applications. The power generation 𝑃 (also called specific absorption rate, SAR) by MNPs under an AC magnetic field is defined by Rosensweig theory (RT) 81: 9 𝑃 = 1 2 𝜇0𝜔𝜒′′𝐻0 2 (11) where 𝐻0 is the magnitude of AC magnetic field 𝑯. This equation leads to a global maximum of 𝑃 when 𝜔𝜏 = 1, which defines the critical frequency for the system 82. SAR is a parameter commonly used to characterize the goodness of a given combination of colloidal suspension and magnetic field characteristics to convert the magnetic field energy into thermal energy. 2.5 Magnetic Relaxivity The element that contains an odd number of protons and/or neutrons in its nucleus, such as 1H, 2H, 13C, 14N, 15N, 17O, 23Na, 31P, etc., exhibits intrinsic magnetic moment (namely, spin), which is the primary origin of the magnetic resonance signal. Single proton hydrogen 1H is one particularly favorable element for nuclear magnetic resonance (NMR) and magnetic resonance imaging (MRI) applications due to its high intrinsic sensitivity and high abundance in water and lipid molecules. For example, the magnetic resonance signals from water or fat within the patient's tissue are monitored for MRI applications, these magnetic resonance signals come from the 1H which is abundant in water and lipid molecules 83. Although NMR has relatively low sensitivity, the MNP provides inherent signal amplification to NMR since each MNP cluster affects billions of adjacent water protons. Thus, MNPs have been widely used as contrast agents in NMR and MRI applications 84, 85. Under an externally applied static magnetic field (𝑯0 along the 𝒛-direction), the water proton nuclei align parallel to 𝑯0 and precess with the Larmor frequency. As shown in Figure 4(a), when a resonant radio frequency (RF) pulse is applied perpendicular to 𝑯0, these nuclei are excited to antiparallel states. Upon the removal of RF pulse, these nuclei relax to equilibrium states. In the presence of MNPs, the dipolar magnetic field increases the local field inhomogeneity. When water molecules diffuse to the periphery of MNPs, the coherent precessions of water proton spins are perturbed. As a result, the net effect is a change of magnetic resonance signal, which can be measured as the shortening of the longitudinal and transverse relaxation time of surrounding water proton spins. The longitudinal relaxation time 𝑇1 (also known as spin-lattice, or relaxation in the z-direction), as shown in Figure 4(b), is a measure of the time taken for the 𝑧 component of the nuclear spin magnetization, 𝑀𝑧, to return to its thermal equilibrium value 𝑀0: 𝑀𝑧(𝑡) = 𝑀0 − [𝑀0 − 𝑀𝑧(𝑡 = 0)] ∙ 𝑒−𝑡 𝑇1⁄ (12) where the 𝑧 component magnetization recovers to 63% of its equilibrium value within a time period of 𝑇1. On the other hand, the transverse relaxation time 𝑇2 (also known as spin-spin, or relaxation in the x-y plane), as shown in Figure 4(c), is the decay constant of the net magnetization 𝑀𝑥𝑦 (magnetization perpendicular to 𝐻0): 𝑀𝑥𝑦(𝑡) = 𝑀𝑥𝑦(𝑡 = 0) ∙ 𝑒−𝑡 𝑇2⁄ (13) where the transverse magnetization drops to 37% of its original magnitude within a time period of 𝑇2. Correspondingly, the longitudinal and transverse relaxation rates are reciprocals of 𝑇1 and 𝑇2: 10 𝑅1 = 𝑅2 = 1 𝑇1 1 𝑇2 (14) (15) Magnetic relaxivity is the intrinsic property of MNP contrast agent, it reflects the MNP's ability to increase the longitudinal and transverse relaxation rates of the surrounding nuclear spins, denoted as 𝑟1 and 𝑟2, respectively: 𝑟1 = 𝑟2 = ∆𝑅1 ∆𝐶 ∆𝑅2 ∆𝐶 (16) (17) Where, 𝐶 is the concentration of MNPs. Thus, the relaxivities 𝑟1 and 𝑟2 have units of 𝑛𝑀−1 ∙ 𝑠−1. Figure 4. (a) The transition from parallel to antiparallel states upon the application of RF pulse. (b) Longitudinal relaxation time 𝑇1. (c) Transverse relaxation time 𝑇2. The sensitives of NMR and MRI are largely dependent on the relaxivities of the MNP contrast agents. Generally, the transverse relaxivity values 𝑟2 of MNPs are greater than longitudinal relaxivity 𝑟1, thus MNPs are mainly used as a 𝑇2 -modulating agent and MNPs with higher 𝑟2 are preferred for these applications 86. In addition, the relaxivity is also dependent on the magnitude of 𝐻0 87, temperature, and solvent (e.g., blood, water, plasma). Magnetic properties of MNPs such as saturation magnetization, size 88, and shape 89-91 are reported to affect the relaxivity. Furthermore, the aggregated and evenly dispersed MNPs can also lead to a 𝑇2 variation. In real NMR ∗) is always less than or equal to 𝑇2, which experiments, the measured transverse relaxation time (denoted as 𝑇2 arises from the inhomogeneities in the main magnetic field. In recent years, lots of work have been carried out to increase the 𝑟2 of contrast agents in order to improve the sensitivity of NMR and MRI 92-94. In a recent report by Zhou et al. 90, they artificially introduced local magnetic 11 field inhomogeneity by clustering of MNPs as well as designing MNPs with heterogeneous geometries (e.g., size and shape) to enhance the 𝑟2. Mohapatra et al. 95 have reported iron oxide nanorods with very high 𝑟2 relaxivity value of 608 𝑚𝑀−1𝑠−1, which is mainly attributed to the higher surface area and anisotropic morphology. Furthermore, alloy-based nanomaterials are good candidates for developing 𝑇2 contrast agents with high 𝑟2 relaxivity values, graphene oxide-Fe3O4 (GO-Fe3O4) MNP composite 94 and Zn2+ doping controlled MNPs 96 have been reported to effectively increase the 𝑟2 value. The classical outer-sphere relaxation theory points out that 𝑟2 𝑟1⁄ increases with increasing particle sizes 97, thus, larger MNPs or MNP clusters are more likely to be better 𝑇2-modulating agents. 2.6 Dipolar Interactions MNPs have been successfully applied in magnetic biosensing (i.e. diagnosing), trapping, and therapeutic platforms. Most of these applications are based on the magnetic properties of MNPs, which may vary depending on the MNP aggregation state, namely, the interparticle distances. In a cluster of MNPs, the dipolar interactions (also known as dipole-dipole interactions or dipolar coupling) affect substantially to the collective magnetic behaviors 67, 82, 98. To be specific, this dipolar interaction modifies the magnetic relaxation time 67, 82, susceptibility, remanence, coercivity, and blocking temperature 99 of MNPs. As a result, the performance of hyperthermia therapy 82, 100, 101 and magnetic particle imaging (MPI) 102 can be drastically modified. As the MNP concentration increases, the interparticle distance decreases, and the dipolar interaction increases, which alters the magnetic response of the whole ferrofluid. For MNPs under an applied magnetic field 𝑯, the total magnetic field acting on one MNP is the sum of external magnetic field 𝑯 and the dipolar field 𝑯𝑑𝑖𝑝 which is generated by surrounding MNPs: 𝑯𝑑𝑖𝑝 = 1 4𝜋 ∑ 𝑖≠𝑗 3(𝝁𝑗∙𝒆𝑖𝑗)∙𝒆𝑖𝑗−𝝁𝑗 3 𝑟𝑖𝑗 (18), Where, 𝒆𝑖𝑗 is the unitary vector joining two MNPs, 𝑟𝑖𝑗 is the distance between these two MNPs. This dipolar field is inversely proportional to the cube of interparticle distance. As is mentioned in section 2.1, the organic capping layer will not affect the magnetic properties of MNPs, which suggests a possibility of tailoring the interparticle distances by controlling the thickness of capping layer. This organic capping layer can effectively prevent direct surface contact and increase the average interparticle distance thus, reducing the dipolar interactions. 2.7 Multicore MNPs Superparamagnetic NPs show zero remanent magnetization due to thermal fluctuation, as the NP size increases they become ferrimagnetic and show hysteresis. For many MNP-based medical and biological applications, larger 12 MNPs are preferred to yield higher specific heating losses and higher magnetic moments. However, the non-zero remanent magnetization of larger MNPs leads to agglomeration in the absence of external magnetic field even with polymer coatings. Once the MNP agglomerates reach to the size of a red blood cell (which is around 6 m), they are at risk of blocking blood vessels and cause dangerous side effects to the patients 103. To prevent agglomeration, the multicore MNPs (also called superferrimagnetic multicore NPs, i.e., MCNPs) are proposed, they are clusters of smaller superparamagnetic NPs embedded in a polymer matrix. Typically, these multicore MNPs are between 20 and 100 nm and consist of superparamagnetic NPs about the size of around 10 nm. Multicore MNPs show much smaller remanent magnetization compared to the single core MNPs with similar sizes, which could effectively prevent the agglomeration. Multicore MNPs have shown excellent potential for applications in magnetic diagnostics and therapy. Lartigue et al. 104 reported magnetically cooperative multicore MNPs with enhanced hyperthermia efficiency and MRI 𝑇1 and 𝑇2 contrast effects. By applying electrostatic colloidal sorting while preserving a superparamagnetic-like behavior, they have successfully enhanced magnetic susceptibility and decreased surface disorder of multicore MNPs. Kratz et al. 105 presented a novel aqueous synthesis for generating multicore MNPs that are suited for both MRI and MPI and allows the combination of these two techniques for biomedical imaging. Lai et al. 106 presented multicore MnFe2O4@SiO2@Ag MNPs with both magnetic and plasmonic properties, which holds great promise for applications in magnetic/photo-thermal hyperthermia and surface-enhanced Raman spectroscopy. 3 Synthesis of MNPs Overall, there are two main approaches to prepare nanoparticles: top-down approach and bottom-up approach. The schematic drawing of these two approaches is shown in Figure 5. In the top-down approach, such as lithography and ball milling, bulk materials (or thin films) are broken down into micrometer or nanometer size. In the bottom-up approach, however, nanoparticles form from atoms followed by nucleation and growth process. There are several bottom-up methods used to prepare nanoparticles, such as gas-phase condensation, chemical vapor deposition, and wet chemical method. In this section, we will focus on these two methods: ball milling method as top-down approach and gas-phase condensation method as bottom-up approach. 13 Figure 5. Schematic drawing of a typical top down and bottom up approach for nanoparticle synthesis. (a) Bulk material. (b) Grinding ball. (c) Nanoparticles. (d) Clusters. (e) Atomic level. 3.1 Ball Milling Method Ball milling method developed by John Benjamin 107 in 1970 is used to prepare powders with reduced size. Fecht et al. 108 proposed the working mechanism of ball milling method. Schematic drawing of the working principle of ball milling method is shown in Figure 6. 14 Figure 6. Schematic representation of ball milling mechanism for the formation of crystalline nanoparticles. There are three stages for ball milling method to prepare nanoparticles. In the first stage, deformations and dislocations are introduced into bulk materials during collisions between balls and bulk materials. The dislocation density keeps increasing with milling time. In the second stage, small grains (nanoscale) are formed due to the accumulation, recombination or rearrangement of dislocations. In the third stage, the grain orientation became random. And then grains at the edge of bulk material are peeled off. Thus, the nanoparticles are obtained from bulk materials. The relationship between grain size of the prepared nanoparticle and the milling time could be 15 estimated by the equation 𝑑 = 𝑘𝑡−2 3⁄ , where 𝑑 is the grain size of the nanoparticle, 𝑘 a constant related with specific system and materials, and 𝑡 the milling time 109. However, there is a limitation to the particle size obtained in a specific ball milling system. If the milling time is too long the so-called cold-welding effect, where nanoparticles will be welded together resulting in big particles 110, 111. When particles have smaller sizes, for examples sub-micrometer, the surface energy of the particles plays a more important role, and particles are more likely to aggregate together to reduce their surface energy. In order to reduce the cold-welding effect, a surfactant is used to lower down the surface energy and reduce the cold- welding effects 112, 113. Chen et al. 110 reported a surfactant-assisted ball milling method to prepare de- agglomeration graphite nanoparticles. The agglomerate size varies from 1 µm to 30 µm and much smaller nanoparticles (< 100 nm) are obtained when using Phosphoric acid dibutyl ester as a surfactant. Even through surfactant-assisted ball milling could produce particles with smaller size, the size distribution is usually very wide. How to control the size distribution of particles prepared by ball milling method is another problem need to be solved. Liu et al. 114, 115 proposed an idea to select particles with different size via centrifugal separation with different "settling-down" time of a particle solution. In their experiment, different kinds of nanoparticles, such as Fe, Co, FeCo, SmCo, and NdFeB, are obtained with much narrower size distributions. Nanoparticles with high saturation magnetizations such as Fe, FeCo are focused due to their high magnetic moments. Figure 7 exhibits the TEM images of Fe and SmCo5 nanoparticles prepared by a surfactant-assisted ball milling approach. The Fe nanoparticle size ranges from 4 nm to 6 nm as shown in Figure 7(a) and (b). SmCo5 nanoparticles with size ranging from 3 nm to 13 nm are obtained after 5 h milling, and elongated shape SmCo5 nanoparticles are achieved with even longer milling time of 25 h, as shown in Figure 7(c) and (d). Poudyal et al. 116 reported Fe, Co, and FeCo nanoparticles using surfactant-assisted ball milling method. Uniform size (~ 6 nm) particles could be obtained by properly controlling the ball milling parameters and applying size selection process. For bio-applications, ball milling with a surfactant is used because of reduced cold-welding effect, small particle size and narrow particle size distribution, which makes surfactant-assisted ball milling a suitable way to prepare particles for bio-related applications. 16 Figure 7. TEM images of the nanoparticles prepared by milling Fe powders for (a) 1 h, and (b) 5h, and by milling SmCo5-based powder for (c) 5 h, and (d) 25 h. (Reprinted with permission from reference 115, copyright 2006 AIP Publishing LLC) 3.2 Gas-phase Condensation (GPC) Method Gas-phase Condensation (GPC) is a bottom-up approach in which atoms (from the sputtering or evaporation) nucleate and grow to form nanoparticles. The particle size and crystallinity are well-controlled by this method while the throughput is generally low compared to ball milling methods. Granqvist and Buhrman 117 firstly used the GPC method to prepare ultrafine particles in 1976. In this GPC system, the atoms are generated from an evaporation source. Then atoms nucleate and grow into nanoparticles in static inert gas. However, both the particle size and crystallinity are out of control. Sattler et al. 118 introduced differential pressure and a skimmer into the GPC system, and the size of nanoparticles obtained is under control, but this method is still unable to control the crystallinity of nanoparticles. Later in 1991, a sputtering source is adopted. Thus, more materials are suitable for 17 preparing nanoparticles 119. Yamamuro et al. 120 pointed out the key factor to get monodispersed nanoparticles is to separate nanoparticles' nucleation and growth process into different space regions. In 2006, Wang et al. 121 reported that field-controlled plasma heating effect could help control the phase and crystallinity of NPs. For instance, disordered A1 phase and ordered L10 phase FePt nanoparticles are successfully prepared using the GPC method by adjusting the plasma heating effects. Moreover, meta-stable phase body-centered tetragonal Fe nanoparticles are also successfully made by this method 122. We will focus on GPC system with sputtering sources in this section. In the GPC system, atoms are kicked off from a target forming atom vapor near the target surface. High sputtering pressure (several hundred mTorr) is applied to form nanoparticles instead of thin films since the mean free path of atoms is much smaller at high sputtering pressure (several hundred mTorr). In this case, these sputtered atoms will collide with argon atom, and energy will transfer from atoms to argon atoms. Thus, the temperature of atoms is cooling down and nucleation and growth will happen when the temperature is low enough. The size of the nanoparticle is dependent on the sputtering current, magnetic field intensity at the target surface, sputtering pressure and gas flow rate. Due to magnetron sputter sources, many kinds of particles have been successfully prepared, such as high magnetic moment MNPs (Fe, Co, FeCo), core-shell nanocomposite, and MNPs with the tunable magnetic property. First, well-crystallized high moment FeCo nanoparticles are successfully synthesized by a GPC method. The average size of FeCo nanoparticle is around 12 nm showing superparamagnetic properties, which is suitable for bio-related applications such as GMR based biosensors 123, 124. As shown in Figure 8, both spherical and cubic shape FeCo nanoparticles could be made using GPC method by adjusting the plasma heating effects. Some other high moment nanoparticles such as Fe and Co are also successfully synthesized using a GPC method. Figure 8. TEM bright field images of FeCo nanoparticles with different shapes, (a) spherical and (b) cubic. (Reprinted with permission from reference 125 , copyright 2007 Elsevier) 18 Second, to make nanoparticles easier for following surface chemical modification, core-shell nanoparticle structure is a good candidate. Gold is good for the surface modification, making it a good candidate for shell material for high moment magnetic nanoparticles. To make a gold shell for magnetic nanoparticles, atom diffusion at the nanoscale should be well-understood. There are two effects competing during shell formation: diffusion via concentration gradient and surface segregation. The first one is to make atoms distribution uniform and the other one is opposite. Therefore, surface segregation should be a major effect in obtaining core-shell nanoparticles 126. Figure 9 shows the TEM images of FeCo nanoparticles with Au shell prepared by the sputtering-based GPC method. Core-shell nanoparticles could also be prepared by multiple ion cluster source (MICS), in which three independent magnetron sources replace the single source as discussed above 127. In MICS setup, one source could be used as core NPs synthesize and others for shell materials. In this case, some core@shell, core@shell@shell structures are prepared 128. As shown in Figure 10, nanoparticles with Co/Ag/Au core@shell@shell structure are prepared, demonstrated by scanning TEM (STEM) image and energy dispersive X-ray spectroscopy (EDS) line scan and electron energy loss spectroscopy (EELS) element mapping of these nanoparticles. Figure 9. Morphology and composition distribution of FeCo -- Au nanocrystals. (a) High angle annular dark field (HAADF) image showing clear core-shell structure due to the contrast between the different materials of core and shell. (b) Composition distribution of a cross section of a single FeCo -- Au nanocrystal, acquired by EDS line scan. The nanocrystal is shown in the inset. The line indicates the path of the electron beam. (Reprinted with permission from reference 129, copyright 2007 AIP Publishing LLC) 19 Figure 10. Core@shell@shell Co@Ag@Au nanoparticles. a) Representation of the complex Co@Ag@Au structure together with the expected EDS intensity profiles. b) Cs-corrected STEM representative image of a Co@Ag@Au NP. c) EDS line scan performed at the Co, Ag and Au, along the line depicted in Figure 4b. d) EELS compositional analysis for the Co L3,2 edge. The dashed line represents the outer limit of the NP. e) EELS map for the Ag M4,5 edge. f) STEM image together with the corresponding Co and Ag EELS concentration maps superimposed. (Reprinted with permission from reference 128, copyright 2014 The Royal Society of Chemistry) 3.3 Other Nanoparticle Fabrication Approaches Besides ball milling method and gas-phase condensation system, there are some other approaches to prepare nanoparticles such as wet chemical way 130, 131, chemical vapor deposition 132, 133, thermal decomposition 134, 135, etc. Figure 11 shows the SEM and TEM images of FeCo nanocubes prepared by a wet chemical method. The size of FeCo nanocubes ranges from 60 nm to 270 nm. The morphology and dimensions of the FeCo nanotube can be adjusted by controlling the concentration of cyclohexane and PEG-400, the reaction time, and the molar ration of Fe2+ and Co2+ in the reaction solutions. The saturate magnetization of 68 ± 6 nm FeCo nanocubes is 211.9 emu/g. 20 Figure 11. (a) SEM and (b) TEM micrographs of FeCo nanocubes, obtained by reaction of 0.1 M Fe2+ and Co2+ with hydrazine for 30 min in the presences of 2.8 M PEG-400 and 0.14 M cyclohexane. The inset is an illustration of the nanocube. (c) TEM image and (d) SAED pattern of a single FeCo alloy nanocube oriented along [001]. (e) TEM image, (f) SAED pattern, and (g) HRTEM image of a FeCo alloy nanocube oriented along [110]. (Reprinted with permission from reference 130, copyright 2008 American Chemical Society) 21 4 Surface Coating Strategies The biocompatibility and chemical stability of MNPs can be enhanced by conjugating chemical compounds such as polyethylene glycol (PEG), chitosan, lipid, proteins, etc. 136. In most biological applications, the solubility and chemical stability of the MNPs should be well controlled in different environments both in vivo and in vitro. It is also necessary to prevent MNPs from aggregation and precipitation while maintaining their biocompatibility and chemical stability 136. To this end, surface coating strategies are needed to facilitate the application of MNPs in nanomedicine. In this section, both organic and inorganic coating strategies with regard to different areas of applications will be reviewed. 4.1 Organic Coating MNPs synthesized through organic solutions are monodisperse, single crystalline with the high magnetic moment. But since these particles often turn out to be hydrophobic, additional surface modification techniques are needed 137. In general, the organic coating techniques on MNPs can be divided into covalent and absorption processes. As is shown in Figure 12, the covalent coating can be further divided into grafting-to, grafting-from, and grafting- through techniques 138. Figure 12. Schematic representation of different coating techniques for MNPs. By reacting with hydrophilic molecules, the hydrophobic surfactants on the nanoparticles can be replaced with hydrophilic ligands. Polyethylene glycol (PEG) is the most commonly used polymer for such ligand exchange 22 process due to its biocompatibility and also the ability to reduce any non-specific reaction between MNPs and proteins 139. For example, Xie et al. 140 synthesized monodispersed 9 nm Fe3O4 nanoparticles and functionalized them with dopamine (DPA) terminated PEG. The coating process is shown in Figure 13. Since DPA moiety has higher affinity to the Fe3O4 surface, it was first linked to PEG and was then used to replace the oleate and oleyamine on the particle surface. It was found that not only did these MNPs have higher stability and no agglomeration in water and physiological environment, they were also undetectable by the body immune system, which made them promising candidates for drug delivery applications. Besides PEG, other polymers such as dendrimers 141, polyethylene oxides 142 and dextrans 143 have also been used in the ligand exchange. In addition, small molecules that have high affinity to the MNP surface, such as the previously mentioned DPA, can also be directly used in the ligand exchange 144, 145. Figure 13. Surface modification of Fe3O4 nanoparticles via DPA-PEG-COOH. X=CH2NHCOCH2CH2. Although PEG is commonly used for the requirement of minimizing non-specific protein absorption, it has a major drawback, which is the tendency of oxidative degradation. To solve this problem, polyzwitterions, which are made from zwitterionic repeat units without net charge, were developed 138. Since zwitterionic moieties are the main component of cell membranes and exhibit long circulation time, which can largely reduce the cytotoxic side effects. Several approaches are available for the functionalization of polyzwitterions. For example, von der Lühe et al. 146 successfully coated polydehydroalanine (PDha) on the surface of MNPs. The functionalization can also be accomplished by simply inducing polyzwitterions during the synthesis of MNPs. In addition to surface coating, MNPs can also be encapsulated in a shell to increase their biocompatibility and hydrophilicity. Cheng et al. 147 embedded MNPs in a copolymer matrix, i.e., poly(D,L-lactide -- co -- glycolide) -- block -- poly(ethylene glycol) (PLGA-b-PEG-COOH) and managed to tune the size of the MNPs by altering 23 solvent type, polymer concentration, and solvent : water ratio. They were the first group to report a linear correlation between polymer concentrations and the size of the resulting MNPs, which makes it possible to precisely control the size of the MNPs to fulfill different requirements for drug delivery in various organs. Other copolymers, such as polystyrene-co-poly-(acrylic acid) (PS-co-PAA) 148 and poly(D,L-lactide) -- block -- poly(ethylene glycol) (PLA -- b -- PEG) 149 are also used in similar applications. 4.2 Inorganic Coating Silica is one of the most widely used inorganic coating materials, which is biocompatible and can provide non- aggregated and stable suspensions 150. A silica outer shell also allows for the subsequent functionalization of alkoxysilanes. Chen et al. 151 demonstrated the nontoxicity of FePt NPs with a 17 nm thick silica shell and proved that FePt NPs can be internalized by tumor cells. Furthermore, it was found that these particles were strong 𝑇2 agents and thus had great potential for the design of diagnostic and therapeutic agents for drug delivery and hyperthermic tumor ablation. The capability of cellular imaging and in vivo MRI applications was also demonstrated. To obtain a better control of the magnetic properties of the silica-coated MNPs, especially to satisfy the temperature requirement for hyperthermia treatments, the influence of silica coating on the magnetization and the Currie temperature of the MNPs were studied. It was shown that at room temperature, the magnetization can be decreased by 32% by silica coatings under a magnetic field of 1 kOe due to the presence of diamagnetic shells 152. The Currie temperature can also be reduced by 7% due to the interaction between silica and the MNP surface. Thus, a trade-off between the biological stability and magnetic properties should be considered for silica coated MNPs. Compared to organic coatings, inorganic coatings are more frequently used in hyperthermia treatments. Since metals are conducting materials, metallic coatings allow for inductive heating under an AC magnetic field. Due to its easy conjugation with many biomolecules such as DNAs and proteins, gold has become the optimal metallic coating for MNPs 153. While gold coatings can simplify subsequent bio-functionalization processes and protect the internal MNPs from oxidation, they inevitably induce changes in the magnetic properties of the resulting NPs. Presa et al. 154 synthesized FePt nanoparticles from high-temperature solution phase and tried to optimize the number of gold coatings on the MNP surface. It was shown that at low temperature, the coercivity of the FePt- Au NPs decreases about 3 times and the blocking temperatures also reduce to the half compared to uncoated NPs, which was attributed to the reduction of surface anisotropy due to the interaction between gold atoms and the FePt NPs. Yano et al. 154 also did a similar study on this issue and pointed out the direct relationship between coercive field and magnetic anisotropy, but the mechanism underlying in the gold-FePt interaction is still not clear. 24 5 MNPs for Diagnosis Early detection of diseases allows for therapeutic intervention in the early stages, which is the key to successful treatment. With the rapid developments in nanotechnology, optical and electrochemical sensors have been applied as high sensitivity diagnosis platforms. However, considering their susceptibility to interferences from unprocessed biological samples, the optical and electrochemical sensors require complicated sample pretreatments 155. On the other hand, magnetic nanomaterial-based sensing systems are attracting much attention for the immunoassay applications since biological samples exhibit negligible magnetic susceptibility, which makes very high signal-to-noise ratio (SNR) magnetic sensing possible even in a minimally processed biofluidic sample. 5.1 Magnetoresistive (MR) Sensors 5.1.1 Giant Magnetoresistive (GMR) Biosensors Since its discovery in 1988, GMR sensors have been widely used in hard disk drives 156, 157, position sensing 158, electrical current measurement 159, 160, as well as biomarker detection 161, 162 163, 164. GMR effect is observed in structures with alternating ferromagnetic and nonmagnetic metal layers (see Figure 14(a)). When exposed to the external field, the magnetization orientation of the "free" ferromagnetic layer with respect to the "pinned" ferromagnetic layer will change, which will result in the change of the resistance of the whole device 165, 166. The ability of a GMR sensor to respond to the external field is characterized by the GMR ratio, which can be expressed as: 𝐺𝑀𝑅 = R𝐴𝑃−R𝑃 R𝑃 × 100% (19). The first GMR based sensing system for biomarker detection was performed by Baselt et al. 167 The Bead Array Counter (BARC), which includes multilayer GMR sensors and magnetic microbead tags, was developed and exhibited great potential in the measurement of intermolecular forces during biomolecular recognition processes. In the past 20 years, lots of research has been carried out to optimize the structure of the magnetic immunoassay and the performance of the GMR detectors. The most widely used magnetic immunoassay is built up based on antibody-antigen reactions. For example, in the GMR based probe station system developed by Wang et al.,168, 169 the MNPs were functionalized with streptavidin, which can bind to the biotinylated detection antibody. During the detection process, the analyte was captured by the capture antibody functionalized on the GMR sensor surface. Then, the detection antibody was added and can only bind to the sites with the analyte. The functionalized MNPs were subsequently attached to the detection antibody as the final layer. Under the external field, the MNPs were magnetized, whose dipolar field (also known as the stray field) can be sensed by the GMR sensors underneath 170 (see Figure 14(c)). To realize on-site diagnosis, several handheld systems were also developed 171, 172. With the size of a snack container and a user-friendly interface, the system can be used by non-technicians with minimum 25 expertise (see Figure 14(e)(f)). Apart from antibody-antigen reactions, aptamers can also be used in the immunoassay (see Figure 14(d)) 173. The MNPs used in the immunoassay are often embedded in a polymeric matrix (also known as multi-core MNPs) and their diameters can vary from tens of nanometers to hundreds of nanometers 174-177. Compared to other biomarker tags, MNPs exhibit superior performance due to low background noise, low possibilities of aggregation due to superparamagnetism, and high biological compatibility 178. To date, the detection of various biological targets has been demonstrated, such as DNAs 173, viruses 171, 179 and food pathogens 180, 181. Chugh et. al 182-184 has demonstrated that GMR in its IC form has also been quite active in non-invasive determination of primary healthcare parameters such heart rate, respiration rate and blood pressure. 26 Figure 14. (a) An example for GMR spin valve stacks. (b) An example for MTJ stacks. (c)Magnetic immunoassay on GMR sensor surface. (d) GMR detection based on aptamers. (e) and (f) are GMR-based handheld systems developed by two different groups. ((e) reprinted with permission from reference 171, copyright 2017 American Chemical Society, (f) reprinted with permission from reference 172 , copyright 2016 Elsevier) 27 However, a major problem associated with GMR-SV sensors is that they exhibit Barkhausen noise due to the formation and depletion of the magnetic domains during the spin direction reversal of the sensing layer. This causes changes in resistance which may occur in the absence of the MNPs, and when used for the detection of MNP-labels/biomarkers, it can result in false signal detection. Nevertheless, this problem can be resolved if necessary by applying a strong pre-magnetizing field is applied (e.g., by integrated microconductors) to reduce the order the internal magnetic domains prior to the sample run 185. 5.1.2 Magnetic Tunnel Junction (MTJ) Biosensors In addition to GMR sensors, MTJ sensors are also promising candidates as the detectors for the magnetic immunoassays. The basic structure of a MTJ consists of a thin insulating tunnel barrier sandwiched between two ferromagnetic layers (Figure 14(b)). The tunneling magnetoresistance (TMR) ratio is defined in the same way as the GMR ratio. While the GMR ratio of the spin valve sensors commonly used in the bio-detection is less than 20% 168, the TMR ratio of the MTJ sensors with MgO tunnel barriers can be over 200% at room temperature 186. MNPs and the immunoassays are integrated with MTJ sensors in the same way as with GMR detectors. Sharma et al. 187 have developed a sensing system based on MTJ sensors for the detection of pathogenic DNA. Firstly, the probe oligonucleotides complementary to the target DNA strands are immobilized on the sensor surface. After hybridization between probe DNA and target DNA, the streptavidin-coated MNPs were added. The real-time signal change of the injection, binding and washing steps of the MNPs can be read out from the MTJ sensor arrays. The detection limit for Listeria DNA can be as low as 1 nM. Cardoso et al. 188 studied the difference of MTJ sensors and spin valve (SV) sensors in the detection of 130 nm MNPs. It was found that MTJ sensors had a higher signal-to-noise ratio (SNR) than SV sensors but will reach a limit at a current of 900 μA. The SNR of SV sensors, however, will increase continuously with the applied current and will only be limited by the heat generation. In the detection of MNPs, SV sensors also exhibited higher signal level and sensitivity despite the high TMR ratio of the MTJ sensors, which can be attributed to the increased distance between MNPs and sensor surface due to the top electrodes of MTJ sensors. Despite greater linearity and sensitivity of TMR sensors over SV, a major problem for TMR is shot noise that arises from the discontinuity in the electron transport paths is present in MTJs unlike that in spin valves and GMR multilayers. This is due to the existence of an insulating barrier in MTJs; the conduction medium is discontinuous for MTJs. Another probable source of error in both SV and MTJ sensors can be from the sample itself in cases where there are MNPs from clusters, may be because of the presence of surface charges resulting in an erroneous detection 185. Such clustering can be avoided, with suitable surface modification and sheathing of the magnetic material as discussed previously in Section 4. Other possible sources of errors in MR sensors are the thermal noise (which are non-magnetic in origin but has a direct effect on the resistance of the sensor) and stray field from MNPs. New, exciting applications of these sensors could be in MCG and magnetic encephalography provided 28 there field of operation can be shifted to elevated frequencies, giving them better noise level, and multiple calibration steps are carried out before use. 5.2 Micro-Hall (μ Hall) Sensors Since Hall voltage is proportional to the external field, Hall sensors can provide a linear response to the dipolar/stray field from MNPs and are only sensitive to its perpendicular component. Thus, they can map out the trajectory of moving magnetic particles and have been used in many areas such as drug delivery, medical imaging, and biomarker detection 189. In 2002, Besse et al. 190 fabricated a highly sensitive silicon Hall sensor with the ability to detect one single magnetic microbead of 2.8 μm by standard complementary metal-oxide-semiconductor (CMOS) technology. Later, devices based on InAs quantum well (QW) semiconductor heterostructures were developed. Phase-sensitive detection was employed and demonstrated even better SNR for micrometer and nanometer-sized magnetic particles 191. Michele et al. 192 also performed single-particle detection using InSb double Hall cross with an area of 1 μm2. By sweeping DC field and taking the normalized difference between the in-phase signals, the susceptibility curve for the target magnetic particle can be measured. Aledealat et al. 193 was the first group to integrate microfluidic channel with micro-Hall sensors (see Figure 15). With a cross-shaped InAs quantum well micro-Hall sensor, the real-time signal from beads moving within and around the sensing area in the microfluid channel can be detected and distinguished by polarities. The first attempt to use micro-Hall sensors in biological detection was made by Issadore et al. 194, in their work, the MNPs were bound to the bacterial cell wall and acted as the magnetic tags. The stream with target cells was flowed through the hall sensor surface in a microfluidic channel and was confined in the vertical direction. It was shown that the sensing system can reliably distinguish Gram-positive from Gram-negative species and requires much smaller sample volume (1 μL) compared to flow cytometry. The detection limit of the system was ~10 bacteria, which was comparable to that of culture tests, but the assay time was 50 times faster. Sandhu et.al 195 had studied some practical sensors based on Hall Effect for biomedical instrumentation. 29 Figure 15. Schematic illustration of the microfluid channel integrated with micro-Hall sensor. (a) SEM image of the central region of an InAs micro-Hall sensor and immobilized SPM beads. (b) 𝑉𝐻 as a function of time for crosses (1) -- (4). The increase in 𝑉𝐻 for crosses (1) -- (3) and its drop for cross (4) as B was applied agrees with the expected signals based on a dipolar stray field representation of SPM beads. (Reprinted with permission from reference 193, copyright 2010 Elsevier) 5.3 NMR-based Diagnostics NMR-based diagnostics exploits MNPs as proximity sensors (contrast agents), which modulate the 𝑇2 of water molecules adjacent to the target-MNP aggregates. When MNPs specifically bind to their target molecules through affinity ligands, they form magnetic clusters which lead to a faster decay of NMR signal or shorter transverse relaxation time 𝑇2 of the surrounding water protons. The NMR-based biosensing technology collects signals directly from the whole volume of sample (volume-based biosensor), which effectively shorten the immunoassay time than that of surface-based sensors such as GMR, MTJ, and μHall sensors. This class of sensors (volume- based biosensors) are more flexible, smaller in size, and suitable for on-site diagnosis. 5.3.1 Magnetic Relaxation Switching (MRSw) Assay As is mentioned in section 2.5, the transverse relaxivity 𝑟2 of MNPs are greater than longitudinal relaxivity 𝑟1, thus MNPs are mainly used as 𝑇2-modulating agents for NMR related applications. In MRSw-based nanosensors, the change of 𝑇2 mainly comes from the aggregation degrees of MPNs in the presence of target analytes. To date, there are two types of detection mechanisms that have been reported so far 196-200, namely, 𝑇2 decreases with the aggregation of MNPs in type I system (see Figure 16(a)), and 𝑇2 increases with the aggregation in type II system (see Figure 16(b)). The outer sphere relaxation theory gives a theoretical explanation for these two systems. In type I system, the aggregation of MNPs cannot overcome the thermal randomization thus it's in a motional averaging (MA) model where the diffusional motion of water molecules is fast enough to average out the magnetic 30 field generated by MNP aggregations. In this MA model (type I system, Figure 16(a)), 𝑇2 is inversely proportional to the number of MNPs in aggregation and the concentration of aggregations. Thus, the more severely the MNPs aggregate, the lower the 𝑇2 will be. However, once the aggregation is big enough to overcome the thermal randomization, the system becomes a static dephasing model (SD) where the free water protons contribute mainly to 𝑇2. If the size of MNP aggregation is in SD regime, the 𝑇2 increases with the increasing size. In this SD model (type II system, Figure 16(b)), a small number of aggregates and large space between aggregates lead to most water protons' failure to experience the magnetic field inhomogeneity 155, in this diffusion-limited case, 𝑇2 increases as the size of MNP aggregates increases. This diffusion-limited case applied when larger sized MNPs are used. Koh et al. 201 explored the behavior of MNP-based type I and micrometer-sized particle (MP) based type II MRSw assay systems. Both systems successfully detected the presence of Tag peptide of influenza virus hemagglutinin (IVH) and a monoclonal antibody to that peptide (anti-Tag), while type II MP based assay shows better sensitivity than type I MNP based assays. Chung et al. 202 reported the detection of kidney injury markers KIM-1 (kidney injury molecule-1) and Cystatin C with as low concentration of 0.1 ng/ml and 20 ng/ml respectively. MRSw also found its applications in the detection of virus such as herpes simplex virus 1 (HSV-1) 203, ions such as Hg2+ 204, 205, Pb2+ 206, 207 and Cd2+ 208, and breast cancer cells, colon cancer cells, and lung cancer cells 209. Besides its applications in disease diagnosis, MRSw has also been applied as a sensitive and rapid method for detecting foodborne pathogens such as Listeria monocytogenes 210, Salmonella enterica 211, etc. Chen et al. 212 combined the NMR and magnetic separation into a one-step platform. Based on the difference in the separation speed of small (30 nm) and large (250 nm) MNPs, both MNPs are conjugated with antibodies which specifically recognize the targets, then large MNPs are employed for separation and small MNPs are probes for MRSw sensing. Luo et al. 213 have demonstrated a portable MRSw-based biosensor system for the detection of Escherichia coli (E. coli) O157:H7 from drinking water and milk samples within one minute. Their device reached a detection limit of 76 cfu/mL in water samples with a linear dynamic detection range of 4 orders of magnitude. Recently, some groups have developed miniaturized devices based on NMR for POC applications 202, 213-217, bringing down the cost, size, and sample use by orders of magnitude. As shown in Figure 16(c)&(d), a NMR system is developed by Lee et al. 217, it consists of a permanent magnet that generates stable magnetic field 𝑯0, a RF generator with coil close to sample to generate RF pulses, a signal receiver to amplify NMR signals and external electronics for synchronizing the different components and store the data 218. 31 Figure 16. (a) MRSw-based detection mechanisms (type I and type II) using small MNPs. (b) MRSw-based detection mechanism (type II) using micro-sized magnetic beads. (c) Schematic diagram of the portable NMR system developed by Lee et al. 217. This system consists of microfluidic networks for sample handling and mixing, an array of microcoils for NMR measurements, miniaturized NMR electronics and a permanent magnet. (d) A photograph of the portable NMR system. For user-inputs and data sharing, the system communicates with external devices. ((c) adapted with permission from reference 217, copyright 2008 Springer Nature, and (d) reprinted with permission from reference 219, copyright 2011 The Royal Society of Chemistry) 5.3.2 Magnetic Resonance Imaging (MRI) One of the most well-known applications of NMR in nanomedicine is MRI. MRI is a medical imaging technique that measures the NMR signals from protons in human bodies, and its performance can be significantly improved by administering contrast agents. The MRI contrast agents can be divided into two groups: positive (𝑇1-weighted) and negative (𝑇2-weighted) contrast agents. Positive contrast agents shorten the 𝑇1 of surrounding protons and result in brighter MR images, while negative contrast agents shorten the 𝑇2 of protons and lead to darker MR images. The mechanisms of MNP-based NMR have been discussed in the foregoing sections. After intravenous or oral administration, the MNPs (or USPIONs) can shorten the 𝑇2 (or 𝑇1) relaxation times of surrounding water protons inside various organs, leading to contrast in the MR images. 32 Since the first commercialization of 𝑇1-weighted positive MRI contrast agent Gd-DTPA (Magnevist, Schering AG) in 1988, there have been large numbers of gadolinium-based contrast agents dominating the market (see Figure 17(a)). However, some recent studies have pointed out the concern of gadolinium-associated nephrogenic systemic fibrosis (NSF) 220-223 and FDA has issued warnings to limit the usage of gadolinium-based contrast agents 224. Nowadays, MNPs have emerged as excellent MRI contrast agents, they have proved superior biocompatibility and safety profiles 225. MNPs with core sizes larger than 10 nm are used as 𝑇2-weighted MRI contrast agents, and ultra-small SPIONs (USPIONs) with core size less than 5 nm are reported as promising 𝑇1- weighted MRI contrast agents 226-228 (see Figure 17(a)). Based on the classical outer-sphere relaxation theory, larger MNPs with a high magnetic moment and 𝑟2 relaxivities are chosen as 𝑇2-weighted (negative) MRI contrast agents. There are two types of 𝑇2 -weighted (negative) MNP contrast agents that are clinically approved, namely: ferumoxides (Feridex in the USA, Endorem The principal effect of this negative contrast agents is on 𝑇2 in Europe) with particle sizes of 120 to 180 nm, and ferucarbotran (Resovist) with particle sizes of about 60 nm. ∗ relaxation and thus MR imaging is usually performed ∗-weighted sequences in which the tissue signal loss is due to the susceptibility effects of the iron using 𝑇2 − 𝑇2 oxide core 225. The main drawback of negative imaging agents is, however, inherently related to the contrast mechanism that they generate. MNPs produce a dark signal (a signal decreasing effect) which could be confused with other pathogenic conditions and render lower contrast compared to 𝑇1 contrasted images. This main drawback has limited their clinical usages in low signal body regions, in the presence of hemorrhagic events, and in organs with intrinsic high magnetic susceptibility (such as lung) 229. Some techniques such as spin-echo sequences 230, 231, inversion recovery ON-resonant water suppression (IRON)-MRI 232 and the usage of micron- sized iron oxide particles 233 have been proposed to overcome these challenges. On the other hand, the 𝑇1-weighted MRI contrast agents yield better imaging quality and they can effectively diagnose the normal and lesion tissues especially in blood imaging 234-236. Kim et al. 237 reported one type of efficient 𝑇1 contrast agent which is synthesized via the thermal decomposition of the iron-oleate complex in the presence of oleyl alcohol. These 3 nm-sized USPIONs with a high 𝑟1 value of 4.78 𝑚𝑀−1𝑠−1 and low 𝑟2 𝑟1⁄ ratio of 6.12. The synthesis methods for USPIONs include thermal decomposition, polyol, coprecipitation, or reduction precipitation and 𝑟1 values vary from 2 to 50 𝑚𝑀−1𝑠−1 are also reported by other groups 224, 227, 228, 238, 239. 33 Figure 17. (a) Schematic view of MRI contrast agents: 𝑇1 contrast agents such as Gd-DTPA and USPIONs, 𝑇2 contrast agents such as IONs, dual mode contrast agents such as the core/shell/shell structured NPs. (b) MRI images and their color-coded images of AFIAs and conventional contrast agents. Fe3O4@SiO2@Gd2O(CO3)2 (i), CoFe2O4@SiO2@Gd2O(CO3)2 (ii), CoFe2O4@SiO2@Eu2O(CO3)2 (iii), CoFe2O4@SiO2@Dy2O(CO3)2 (iv), MnFe2O4@SiO2@Gd(BTC)(H2O) (v), ZnFe2O4@SiO2@[ImH][Mn(BTC)-(H2O)] (vi), Gd-DTPA (vii), Feridex (viii), and H2O (ix). Images of contrast agents are taken by using 3.0 T MRI at the identical metal concentrations. (c) & (d) Plots of 𝑅1 and 𝑅2 vs concentration of the metal; 𝑟1 and 𝑟2 of mAFIA, ZnFe2O4@SiO2@[ImH][Mn(BTC)-(H2O)], are∼2-fold larger than those of conventional contrast agents. ((b)-(d) reprinted with permission from reference 240, copyright 2014 American Chemical Society) Recently, dual-mode contrast agents are emerging to eliminate the possible ambiguity of a single-mode contrast agent (either 𝑇1 or 𝑇2) when some of the in vivo artifacts are present, it is the combination of simultaneously strong 𝑇1 and 𝑇2 contrast effects in a single contrast agent (see Figure 17(b)). This dual mode contrast agent can potentially provide more accurate MRI via self-confirmation with better differentiation of normal and diseased areas. The core@shell-type 𝑇1 − 𝑇2 dual mode nanoparticle contrasts have been described by several groups 240- 245. For example, Choi et al. 242 reported the MnFe2O4@SiO2@Gd2O(CO3)2 core@shell@shell nanoparticles as dual-mode MRI contrast agents, where the SiO2 layer was used to modulate the magnetic coupling between 𝑇1 34 contrast material Gd2O(CO3)2 and 𝑇2 contrast material MnFe2O4. By increasing the SiO2 layer thickness (varies from 4 nm to 20 nm), the 𝑟1 value increased from 2.0 to 33.1 𝑚𝑀−1𝑠−1, and the 𝑟2 value decreased from 332 to 213 𝑚𝑀−1𝑠−1. In Figure 17(c)-(d), Shin et al. 240 measured different combinations of core@shell@shell dual mode artifact filtering nanoparticle imaging agent (AFIA) [𝑇2 core (superparamagnetic nanoparticle) @SiO2@𝑇1 shell (paramagnetic material)] and demonstrated its superior relaxivities (𝑟1 and 𝑟2) than Magnevist (𝑇1 contrast agent) and Feridex (𝑇2 contrast agent). Other structures such as iron core with ferrite shell nanoparticles 246, 247, ultrasmall mixed gadolinium-dysprosium oxide (GDO) nanoparticles 248 and core@shell structured manganese- loaded dual-mesoporous silica spheres (Mn-DMSSs) 249 have also been reported as 𝑇1 − 𝑇2 dual mode contrast agents. 5.4 Superparamagnetism-based Diagnostics 5.4.1 Brownian Relaxation-based Assay As is aforementioned in section 2.3, superparamagnetic nanoparticles exhibit non-linear magnetic response curve with zero coercivity. Under an external AC magnetic field 𝐻, the magnetic moment inside superparamagnetic nanoparticle tends to align with the field while this process is countered by Néel and Brownian relaxations that randomize its magnetic moment 250, 251. For superparamagnetic iron oxide nanoparticles (SPIONs) dispersed in liquid solution, the Brownian relaxation is the dominating obstructing factor when the magnetic core size is above 20 nm while Néel relaxation dominates when core size is below 20 nm 79, 252. For those Brownian relaxation- dominated SPIONs, the extent of the disorder is linked directly to environmental conditions 253, such as temperature 254, 255, viscosity 79, 256-258, and the ligand binding states 80, 259-262. In this section, we focus on the Brownian relaxation dominated SPIONs for immunoassay applications. For Brownian relaxation dominated SPIONs, the effective relaxation time is expressed as 𝜏 ≈ 𝜏𝐵 = 3𝜂𝑉ℎ 𝑘𝐵𝑇 . SPIONs are functionalized with ligands (such as aptamers, proteins, antibodies, etc.) which can specifically recognize and bind to target analytes from minimally processed biofluid samples, this successful binding process increases the hydrodynamic size of SPIONs thus the effective relaxation time. Hence, a larger phase delay between its magnetic moment and the AC magnetic field is detected, and the magnitudes of harmonics are attenuated as a result (see Figure 18(h)&(i)). In 2006, Krause group 263 and Nikitin group 264 have independently proposed a mixing frequency method based Brownian relaxation dominated SPIONs for immunoassay applications. As shown in Figure 18(a)-(g), two drive fields 𝐻 = 𝐴𝐻𝑐𝑜𝑠(2𝜋𝑓𝐻𝑡 + 𝜑𝐻) + 𝐴𝐿𝑐𝑜𝑠(2𝜋𝑓𝐿𝑡 + 𝜑𝐿) are applied to drive SPION suspension, the response signal generated at 𝑓𝐻 ± 2𝑓𝐿 (3rd harmonics), 𝑓𝐻 ± 4𝑓𝐿 (5th harmonics), etc., are collected. These harmonics are highly specific to the nonlinearity of the magnetization curve of the SPIONs. The amplitude of low frequency AC field, 𝐴𝐿, is chosen to periodically switch on and off the capability of SPIONs to further magnetization, while the 35 high frequency AC field is used to modulate the harmonics into high frequency regions where the pink noise (1 𝑓⁄ noise) is lower. Due to the time-varying magnetization of the SPIONs, a time-variant voltage is induced in the receive coil due to the SPION harmonic response (Figure 18(b)). On the other hand, Rauwerdink and Weaver et al. 255, 260, 262, 265 proposed a mono-frequency method based Brownian relaxation dominated SPIONs for immunoassay applications, its setup is similar to the 1D MPI: they applied an AC magnetic field 𝐻 = 𝐴𝑐𝑜𝑠(2𝜋𝑓𝑡 + 𝜑) to drive SPIONs, the nonlinear magnetic response induces higher harmonics at frequencies of 3𝑓 (3rd harmonic), 5𝑓 (5th harmonic), etc., the information such as SPION-target analyte binding states can be extracted from these harmonics in the same way as we described before. 36 Figure 18. (a) Schematic setup of mixing frequency method-based magnetometer. Microcontroller with electronics part provide sine waves to drive the excitation coils, the generated signals are picked up by receive coil and harmonics are filtered and read into the microcontroller via the ADC. (b) Sectional view of the coils. (c)- (g) SPIONs are exposed to two AC magnetic field with frequencies of 𝑓𝐻 and 𝑓𝐿 (d). The drive field spectrum in (g) exhibits two distinct lines. Due to the nonlinear magnetization curve of SPIONs (c), the resulting time- 37 dependent magnetization (e) saturates at higher fields, leading to higher harmonics and frequency mixing components in the Fourier-transformed response signal (f). (h) SPIONs are functionalized with ligands that will specifically bind to target analytes, this binding process increases the SPION's hydrodynamic size as well as phase lag. (i) The ligands on SPIONs blocks them from non-specific binding and prevents false signal change. Since the amplitudes of harmonics increase with the number of SPIONs within the fluid sample, so phase lags 80, 266 and harmonic ratio 265 (magnitude ratio of 5th over 3rd harmonic) are used as concentration-independent metrics to characterize the relaxation time as well as the binding state of SPIONs. Orlov et al. 267 applied the method of registration of SPIONs by their nonlinear magnetization to a novel immunoassay on 3D fiber solid phase for detection of staphylococcal toxins in unprocessed samples of virtually any volume (see Figure 19(a)). This 3D fiber structure serves as a reaction surface to selectively filter antigens as well as accelerate reagent mixing. Two different setups are reported, the limits of detection (LOD) reaches to 4 and 10 pg/mL for toxic shock syndrome toxin (TSST) and staphylococcal enterotoxin A (SEA), respectively, by using 30 mL samples, in 2 hours. Tu et al. 80 reported a real-time monitoring of the increase of phase delay in 3rd harmonic due to the binding of goat anti-human IgG to protein G coated SPIONs (see Figure 19(b)). Which is the first experimental demonstration of real-time detection of analyte binding process. Zhang et al. 262 demonstrated a rapid measurement (within 10 s) using SPIONs that conjugated with anti-thrombin ssDNA aptamers for the detection of thrombin, a LOD of 4 nM is reached (see Figure 19(c)). 38 Figure 19. (a) Top left: Magnetic immune-sandwich on 3D fiber filters used as a solid phase. Bottom left: SEM surface morphology of the cylindrical 3D fiber filters. Right: Detection of SPIONs by their nonlinear response at combinatorial frequencies from the whole volume of the 3D solid phase located inside a pipet tip. (b) Real time measurement of phase delay of IPG35 in blue curve and control sample SMG35 in red curve, with antibody injection at the 50th second. IPG35 are SPIONs coated with protein G that will specifically bind to IgG antibodies, this binding process gradually increase the hydrodynamic size of IPG35 SPIONs. SMG35 are SPIONs coated with PEG layer which prevents the binding of IgG onto the SPION surface, as a result, hydrodynamic size of SMG35 barely changed. IPG35 and SMG35 are SPIONs with average hydrodynamic sizes of 35 nm, thus, they exhibit same initial phase delays under the same driving fields. (c) Ratio of 5th to 3rd harmonic of 2 populations of SPIONs, conjugated with 15 mer and 29 mer anti-thrombin aptamer, harmonic ratios are measured as a function 39 of thrombin concentration. Three control samples are applied in this experiment to verify the specificity of the detection: SPIONs without aptamer functionalized, and only one population of SPION functionalized with either 15 mer or 29 mer anti-thrombin aptamer with the increasing concentration of thrombin added. ((a) adapted with permission from reference 267, copyright 2012 American Chemical Society, (b) adapted with permission from reference 80, copyright 2011 AIP Publishing LLC, (c) adapted with permission from reference 262, copyright 2013 Elsevier) 5.4.2 Magnetic Particle Imaging (MPI) Magnetic particle imaging (MPI) has shown great promise in many clinical applications ranging from angiography to cancer theranostics and molecular imaging. This technology was first reported in 2005 by Gleich and Weizenecker 268, which exploits the nonlinearity of magnetization curves of SPIONs and the fact that their magnetization saturates at some magnetic field strength. It's also a safer imaging method especially for chronic kidney disease (CKD) patients for whom iodine contrast agents are toxic. The dynamic 2D MPI and the first prototype of 3D real-time MPI were proposed by Gleich and Weizenecker et al. 269, 270, pushing a huge step toward its medical applications. A typical modality of 3D MPI system is shown in Figure 20, an AC modulation magnetic field 𝑯(𝒕) (it's also called drive field) excites SPIONs and the nonlinear magnetic response 𝑴(𝒕) is induced and picked up by a receive coil. This magnetic response signal 𝑴(𝒕) contains not only the modulation field frequency 𝑓 , but also a series of harmonic frequencies 3𝑓, 5𝑓, 7𝑓 etc. Higher harmonics instead of the fundamental frequency are used for analysis to avoid picking up the applied field. These higher harmonics can be easily separated from the received signal by filtering. In addition to the modulation field, an additional gradient field (also called selection field) is applied to localize received signals. This selection field features a field free point (FFP) where its magnitude is zero and increases towards the edges (see Figure 20(b)), this FFP is scanned through the sample following the scan trajectory. SPIONs in the FFP (red region in (b) and red curves in (c)) produce signals containing higher harmonics while those in the saturation region (the grey region in (b) and grey curves in (c)) contribute negligible signals. Tomographic MPI images are generated by scanning the FFP through the volume of interest. 40 Figure 20. (a) Schematic view of 3D MPI scanner. The biological sample was inserted into the x drive/receive coil cylinder. The selection field is generated by both the coil pair in the z-direction. The drive field coils can move the FFP in all three spatial directions. Only the SPIONs in the instantaneous location of the FFP generate MPI signals. For signal reception, each spatial component of the magnetization is detected by a respective receive coil. In the x-direction, the drive field coil is also used for signal reception. (b) Illustration of FFP scanning and distribution of magnetic field. SPIONs locate in the saturation region contribute negligible MPI signals. (c) The 41 response of the SPIONs to external magnetic fields. When the modulation field 𝑯 with a frequency of 𝑓 is applied, the magnetic response 𝑀(𝑡) exhibits higher harmonics (3𝑓, 5𝑓, 7𝑓, 𝑒𝑡𝑐.,) as shown in the Fourier Transformed signal (red curves). These higher harmonics are used for MPI image generation. When an additional selection field is added (grey curves), the SPIONs are magnetically saturated and the magnetic response 𝑀(𝑡) does not significantly change. ((a) reprinted with permission from reference 271, copyright 2017 American Chemical Society) MPI is an ideal tool for angiography, cellular and targeted imaging, cancer imaging, as well as imaging major organs and the finer coronary arteries, it allows 3D visualization, real-time imaging, and moreover, it's radiation- free 269, 272. MNP tracers can only be detected indirectly in MRI due to their effect on the magnetic resonance signal of the nuclear spin magnetization. On the other hand, MPI directly detects MNP (SPION) tracers based on the nonlinear magnetization response. Thus, MPI provides the distribution of tracer SPIONs deep in body with relatively high temporal and good spatial resolution, and the quality of MPI images are largely dependent on the properties of SPION tracers (core diameter, shell thickness) and scan speed 28, 273-275. In recent years, the development of customized SPIONs for optimized MPI is pursued with great effort. Dhavalikar et al. 274 reported that the MPI signal strength is related to the cubic power of the SPION core diameter 𝐷 as long as superparamagnetic characteristics are retained. In addition, the MPI spatial resolution is heavily dependent on the SPION's magnetic response curve and the gradient selection field. To be specific, a steeper magnetization curve confines a smaller FFP, similarly, a larger gradient selection field shrinks FFP, so a better spatial resolution is achieved 270, 276. Since the SPION's magnetic response to a time-variant magnetic field is governed by Brownian or/and Néel relaxations, the spatial resolution is highly dependent on the SPION's relaxation characteristics. In the first generation of MPI systems, the drive fields were around 10-20 mT in magnitude and frequency of 25 kHz. Theoretically, an ideal SPION core size of 30 nm was proposed for this drive field setting 268. However, Lüdtke-Buzug et al. 277 evaluated the commercially available SPION contrast agents for MRI regarding their MPI performances, and he Resovist (Bayer Pharma AG), with core size much smaller than 30 nm, showed the best MPI performance over other tracers. Later research found that the Resovist SPIONs form aggregates and each aggregate still behaves like one SPION with a core size equivalent to 24 nm 29. Besides this SPION aggregation factor, other factors such as the SPION size distribution, anisotropy constant, morphology, shell thickness, inter- particle interactions also affect the MPI performance 278. To sum up, the properties of SPION, as well as the interaction with the environment must be considered and modified for different drive field settings 33, 279, 280. 42 5.5 Surface Enhanced Raman Spectroscopy (SERS) Systems Raman scattering results from the radiation emitted by molecules or atoms after the bombardment of a primary radiation 281. The absorption of energy due to inelastic scattering of photons causes rotational and vibrational changes, which in turn leads to the change of the molecular dipole-electric polarizability. Raman Spectroscopy (RS) often uses a non-ionizing laser as the excitation source, and each peak in its signal can provide information on a specific chemical bond. While RS is a non-invasive and label-free technique, its signal is rather weak with a large fluorescence background. Moreover, there is also a limitation on the penetration depth 282. In pursuit of stronger signal magnitude, SERS was firstly introduced by Fleishmann et al. 283. The substrates of the SERS are nano-noble metals. As the target molecules are absorbed to the surface of the substrate, there will be a significant increase in the Raman signal due to the excitation of the surface plasmons, which makes it possible for SERS to detect trace biomarkers even at a large distance from the target tissues 281. However, when only plasmonic NPs (eg., Au or Ag NPs) are used, the non-uniform attachment of NPs on bacteria makes it difficult to quantitatively measure the concentration. Detection of targets with very low concentrations is also impossible due to the lack of approaches to concentrate the collected samples. To solve this problem, Zhang et al. 284 developed multifunctional magnetic-plasmonic Fe2O3@Au core@shell NPs. With the magnetic components inside the NPs, the plasmonic properties can be tuned by changing the inter-particle distance via the external magnetic field (see Figure 21) 285. Furthermore, MNPs can also facilitate the fast concentration of bacteria cells under a point magnetic field. It was demonstrated that the plasmonic MNPs can concentrate the bacteria to a level of ~ 60 times higher than non-concentrated samples, which greatly enhanced the SEPR signal. Figure 21. Schematics of the condensation process of Au MNPs and bacteria(left) and the biomolecular characteristics of the bacterial cell wall that can possibly be detected by SERS (right). (Reprinted with permission from reference 284, copyright 2011 Elsevier) 43 Due to their superior properties in SEPR detection, plasmonic MNPs were employed in the detection of a wide variety of pathogen bacteria 286-288 and also in immunomagnetic separations 289. For example, Guven et al. 286 used ion-oxide MNPs with Au nanorods (NRs) as substrates to detect E. coli, achieving a limit of detection at 8 cfu/mL. Drake et al. 287 developed the same MNPs together with Au NPs for the detection of S. aureus. The detectable concentration was shown to be as low as 1 cell/mL. For the convenience of the readers, different MNP-based immunoassay techniques that have been reviewed in this paper are summarized in Table 1. Table 1. Comparison of different MNP-based immunoassay techniques Sensor Type MNP Assay Target Analyte Limit of Evaluated Ref. GMR 50 nm 30 min Pregnancy-associated 1 ng/mL Blood serum 168 Time Detection (LOD) Matrices MNP plasma protein A (PAPP-A) Proprotein convertase 433.4 pg/mL subtilisin/kexin type 9 (PCSK9) Suppression of 40 pg/mL tumorigenicity 2 (ST2) GMR 50 nm 10 min Influenza A Virus 15 ng/mL Phosphate 171 MNP (IAV) Nucleoprotein buffered saline (NP) (PBS) Purified H3N3 Virus 125 TCID50/mL GMR 50 nm 1 hr DNA with NRAS and NA 20×saline sodium 173 MNP BRAF mutations citrate (SSC) MTJ 250 nm 30 min DNA from Listeria, 1 nM Phosphate buffer 187 MNP HEV and Salmonella MTJ 12 nm 12 hr Single strand DNA 10 nM PBS MNP Hall 20 nm 40 min S. aureus 10 bacteria Staphylococcus MNP broth 290 194 Hall 250 nm 24 hr Oligonucleotides NA NaCl+HCl+EDTA 291 MNP MRSw 30 nm NA Kidney injury 0.1 ng/mL Urine 202 MNP molecule-1 (KIM-1) 44 Cystatin C 20 ng/mL MRSw 50 nm 30 min S. enterica 103 cfu/mL MNP MRSw 30 nm 30 min S. enterica MNP S. enterica 104 cfu/mL 103 copy/mL MRSw 30 nm 90 min Newcastle disease 5 fM MNP virus Urine Milk Purified Purified Urine MS-MRSw a) 30 nm 30 min miR-21 from tumor 102 cfu/mL Purified MNP & 250 nm MB cells S. enterica 102 copy/mL Purified Brownian 50 nm 10 sec Newcastle disease 150 pM or 0.075 Purified Relaxation- MNP virus pmole based Assay Streptavidin 4 nM or 2 pmole Purified Thrombin 100 pM or 0.05 Purified pmole Brownian 50 nm 2 h Anti-thrombin 4 pg/mL, 30 mL Milk Relaxation- MNP aptamers based Assay Staphylococcal 10 pg/mL, 30 mL Milk enterotoxin A (SEA) 25 min Toxic shock syndrome 0.1 ng/mL, 150 Milk toxin (TSST) L SEA 0.3 ng/mL, 150 Milk L 211 212 292 212 262 267 SERS 200 nm 3 4-mercatopyridine Lower than 1 nM DI water 284 hours Au- MNP SERS 200 nm NA E. coli 2105 cfu/mL DI water 284 Au- MNP P. aeruginosa SERS 100 nm 2 S. aureus 103 cfu/mL Spinach wash 288 Silica- hours coated MNP MS-SERS b) 500 nm 2 TSST 1 pg/mL Purified 293 MB hours a). Magnetic Separation and Magnetic Relaxation Switching assay b). Magnetic Separation and Surface-enhanced Raman Scattering assay 45 6 MNPs for Therapeutic Applications 6.1 Drug Delivery The concept of using MNPs/magnetic microparticles as carriers for targeted therapeutic applications was first proposed in the late 1970s by Senyei and Widder et al. 294, 295. The therapeutic agents (such as cytotoxic drugs for chemotherapy and DNA for genetic therapy) are attached to the surface of the magnetic carrier or encapsulated within the polymer matrix which contains the magnetic carrier. These particle-drug complexes are injected into the subject either via oral or intravenous, then an externally applied high-gradient magnetic field guides and concentrates these complexes to the target sites (see Figure 22(a)). Once the complexes are concentrated at the target organ in the body, the therapeutic agents are released via enzymatic cleavage of the cross-linking molecules, charge interactions, pH change, degradation of the polymer matrix, or heating up the particles 296-299. This MNP- based targeting method effectively reduces the side-effects brought by the non-specific nature of chemotherapy, which attacks healthy cells in addition to primary targets. Furthermore, by accurately administering and delivering therapeutic agents to target sites, it reduces the systemic distribution of cytotoxic compounds and enables effective treatment with a lower dosage. SPION is one of the most commonly used magnetic carriers for drug delivery, it's often coated with an organic matrix such as polysaccharides, fatty acids, or other polymers to improve the colloidal stability 199, 300-302. SPIONs exhibit zero remnant magnetization upon the removal of the external field, which avoids the aggregations and facilitates their excretion from the body. Magnetic targeting is based on the theory that, a translational force will be exerted on the particle-drug complex in the presence of a magnetic field gradient, thus trapping the complex at the target site and pulling the complex towards the magnet 296, 303. A successful trapping of the particle-drug complex at the target site is largely dependent on the blood flow rate, surface characteristics of magnetic carriers, magnetic properties of particles, the magnetic field strength as well as magnetic field gradient. 46 Since the magnetic field strength decays rapidly with distance to the inner sites of the body become more difficult to target, which is one of the main barriers that impeding the scale-up of magnetic targeting from small animals to humans 297. Some groups have proposed to implant magnets (i.e. a magnetizable intraluminal stent, seed, etc.) in the body near the target site to circumvent this problem 304-309. Nacev et al. 310 reported pulsed magnetic fields to exploit the rotational dynamics of ferromagnetic rods and attain well-like curvature of the magnetic potential energy (see Figure 22(b)). Using this method, they focused a disperse concentration of ferromagnetic rods to a central target between eight external electromagnets (see Figure 22(c)&(d)). Figure 22. (a) Schematic representation of magnetic nanoparticle-based drug delivery system. (b) How forces generated from a magnet configuration affect particle concentration. A magnet configuration creates a magnetic potential energy surface (top row) that generates the magnetic forces. Magnetic forces (middle row) shape particle concentrations (bottom row). Particles will move from locations of high-energy states (white) to low-energy locations (black). Equivalently, particles will move due to either divergent forces (blue arrows) or convergent forces (red arrows). By applying Earnshaw's theorem to static magnetic fields, only unstable static magnetic potential energy configurations were theorized to be possible, e.g., (left) a peak energy configuration and (middle) a saddle; (right) through the use of pulsed magnetic fields, a magnetic potential energy well is generated which is capable of concentrating particles to a central target. (c) Pulse sequence element for inverting the energy surface 47 of ferromagnetic rods and concentrating them at the center of the sample area. This pulse sequence element is repeated many times for the four directions to push and concentrate the rods to the center. (d) Focusing of ferromagnetic rods to a central target. Four snapshots of concentrating cobalt rods to the center of the sample area using dynamic magnetic inversion. The rods began optically undetectable and dispersed throughout the region. After 09:06 min, the rods were concentrated at the center of the sample area. Video available as Supporting Information. (e) Schematic illustration of the MEMSN system for pH-responsive drug delivery and magnetic resonance imaging. (f) In vivo MR imaging of the mice after intravenous injection of MS@S-NPs at a different time period, the red circles point the liver. ((a) reprinted with permission from reference 311, copyright 2009 Elsevier, (b)-(d) reprinted with permission from reference 310, copyright 2014 American Chemical Society, (e)&(f) reprinted with permission from reference 298, copyright 2015 Elsevier) Besides the magnetic field issue, there are several other obstacles impeding the delivery of drug to target sites, such as unspecific uptake by healthy cells, rapid clearance, and aggregation during the circulation. To overcome these barriers, an ideal drug carrier should have the following properties: a protection layer to keep the carrier stable and avoid nonspecific cell uptake, tumor-targeting ability to accumulate at tumor sites, ligand-mediated cell adhesion for efficient cellular entry, and effective drug encapsulation to permit the drug release inside tumor cells only 312. In recent years, magnetic carriers with multiple functions including targeted anticancer drug delivery, hyperthermia, and imaging have been an ongoing hot topic 313-319. Hervault et al. 317 reported a dual-pH and thermo-responsive magnetic carrier with combined magnetic hyperthermia and anticancer drug delivery. In their study, a pH- and thermo-responsive polymer shell is coated onto SPION, then the anticancer drug doxorubicin hydrochloride (DOX) is conjugated via acid-cleavable imine linker. This complex provides advanced features for the targeted delivery of DOX via the combination of magnetic targeting, and dual pH- and thermo-responsive behavior which offers spatial and temporal control over the release of DOX. Chen et al. 298 reported a pH- responsive nano-gated multifunctional envelope-type mesoporous silica nanoparticle (MEMSN) which is capable of magnetic drug delivery and in vivo MRI. They immobilized acetals on the surface of mesoporous silica and coupled to ultra-small lanthanide doped nanoparticles (NaGdF4) coated with TaOx layer (S-NPs) as gatekeeper (see Figure 22(e)). The anticancer drug DOX is locked in the pores and its burst release can be achieved under acidic environment on account of the hydrolyzation reactions of acetals. This MEMSN system has demonstrated in vivo MRI (see Figure 22(f)), passive tumor targeting, increased tumor accumulation, and it can be harmlessly metabolized and degraded into apparently nontoxic products within a few days. Ye et al. 316 reported a nanocarrier system with multiple imaging agents and an anticancer drug, they encapsulated inorganic imaging agents of SPION, manganese-doped zinc sulfide (Mn:ZnS) quantum dots (QDs) and the anticancer drug into poly(lactic- co-glycolic acid) (PLGA) vesicles via an emulsion-evaporation method. Their PLGA vesicles exhibit high 48 𝑟2 ∗ (523 𝑠−1𝑚𝑀−1 𝐹𝑒) relaxivity and greatly enhanced 𝑇2 ∗-weighted MR imaging contrast. The Mn:ZnS QDs are used for fluorescence imaging to investigate the interaction between PLGA vesicles and cells. 6.2 Hyperthermia Therapy Magnetic hyperthermia is a promising method for the treatment of cancer, currently in clinical trials 320-323. In hyperthermia therapy, an AC magnetic field is remotely controlling MNPs to induce local heat, provoking a local temperature increase in the target tissues where the tumor cells are present. The specificity of this technique is based on that tumor cells are vulnerable to temperatures above 42 C, at which the natural enzymatic processes that keep tumor cells alive are destroyed, so allowing the selective killing 324. The AC field used in magnetic hyperthermia is in the range of radio-frequency, between several kHz and 1 MHz, which is completely healthy and shows enough penetration depth to access inner organs/tissues in the body. Several heating mechanisms are possible in magnetic hyperthermia, namely, hysteresis loss, Brownian and Néel relaxations (susceptibility loss), and viscous heating. In comparison to these aforementioned magnetic losses, the eddy current is only significant in materials at centimeter scale and it's negligibly small in MNP hyperthermia applications. For multi-domain MNPs, the hysteresis loss is determined by integrating the area of hysteresis loops, which is a measure of energy dissipated per cycle of magnetization reversal. This type of heating attributes to the shifting of domain walls and it's strongly dependent on the AC field amplitude and its magnetic prehistory. As the MNP size reduces to the single domain, hysteresis heating is accomplished by rotating the magnetic moment of each MNP against the energy barrier. As the MNP size further reduces to superparamagnetic NP where the thermal fluctuation becomes comparable to the energy barrier, the relaxation mechanism plays a major role in magnetic hyperthermia. During Brownian relaxation process the thermal energy is delivered through shear stress in the surrounding fluid (viscous heating) and during Néel relaxation process the thermal energy is dissipated by the rearrangement of atomic dipole moments within the crystal 321. For those superparamagnetic NPs in a liquid environment, each particle maintains a constant magnetic moment and its orientation is determined by the effective anisotropy of the particle. An external magnetic field switches the moment from its preferred orientation and the relaxation of the moment back to equilibrium state releases thermal energy which results in local heating 325. Susceptibility loss is associated with Brownian and Néel relaxations, recall equation (11), the specific absorption rate (SAR), 𝑃 (𝑊 ∙ 𝑚3), is related to 𝜒′′ according to 𝑃 = 1 2 𝜇0𝜔𝜒′′𝐻0 2, the global maximum 𝑃 is reached when 𝜔𝜏 = 1 (SAR is related to the specific loss power, namely, 𝑆𝐿𝑃 (𝑊 ∙ 𝑔−1), by the mean mass density of particles). The strong size dependence of relaxation time results in a very narrow maximum of the loss of power density. Accordingly, a good output of heating power occurs only in particle systems with narrow size and anisotropy distribution, as well as the mean diameter, being well adjusted in relation to the AC field frequency 49 326, 327. In addition, the efficiency of heating is largely dependent on the ability of MNPs to specifically accumulate in the target tissues and to achieve effective cancer therapy with a minimum dosage. Figure 23. (a) SAR as a function of the mean particle size 𝐷 for variable volumetric packing fraction  including non-interacting case (=0) and two values 𝑀𝑠 = 200 𝑒𝑚𝑢 𝑐𝑚3 ⁄ and 400 𝑒𝑚𝑢 𝑐𝑚3 ⁄ . The solid line represents the solutions by means of RT. (b) SAR versus 𝐷 for different packing fractions and the values of 𝐾 = 3 × 105 𝑒𝑟𝑔 𝑐𝑚3 ⁄ (curve set 1), 1.5 × 105 𝑒𝑟𝑔 𝑐𝑚3 ⁄ (curve set 2) and 0.5 × 105 𝑒𝑟𝑔 𝑐𝑚3 ⁄ (curves set 3). Calculations correspond to log-normal distributions of size and anisotropy constants both with standard deviations 0.1, spherically distributed anisotropy axes, and field amplitude 𝐻 = 300 𝑂𝑒. (c) Hysteresis loops for a spherical 50 (red circles, diameter 20 nm) and a cubic particle (blue squares, side 20 nm) obtained from MC simulations of an atomistic spin model of maghemite at low temperature. In both, uniaxial anisotropy at the core and surface anisotropy have been considered. Spins have been colored according to their projection into the magnetic field direction (z-axis) from red (+1) to blue (-1). (d) & (e) SAR values for two nanoparticle solutions of similar concentration (0.5 mg/mL) and size volume but different shape indicating enhancement of SAR values for the 20 nm square nanoparticles. (d) and (e) are experimental and MC simulation results for the macrospin model with dipolar interactions at 300K. (f) Representative hysteresis M(H) normalized loops corresponding to different lengths (N) of a chain of magnetic nanoparticles with uniaxial easy anisotropy axes distributed within a cone of angle α = 20°. The black hysteresis curve corresponds to the case of randomly distributed non-interacting particles, also with the easy axes distributed at random. The inset illustrates the distribution of the axes within the cone of angle α, for the N = 4 chain. (g) The main panel shows the dependence of hysteresis losses 𝐻𝐿 2𝐾⁄ vs 𝐻𝑚𝑎𝑥 for different chain lengths, for the same α = 20° case. Inset illustrates the optimal conditions for hyperthermia applications, as inferred by plotting 𝐻𝐿 2𝐾 ∙ 𝐻𝑚𝑎𝑥 ⁄ versus 𝐻𝑚𝑎𝑥. (h) Experimental hyperthermia power of the 0.5 % and 1 % agar cases measured at 765 kHz and 300 Oe maximum AC field, applied at different orientations with respect to the chains long axis (0°, 45°, and 90°; schematically illustrated in the top panel). (i) Simulated angle-dependence of the hysteresis losses for different chain lengths. The dotted line represents the hysteresis losses of a randomly distributed noninteracting system. ((a)&(b) from reference 328 under CC BY 4.0, (c)-(e) from reference 329 under CC BY 4.0, (f)-(i) reprinted with permission from reference 330, copyright 2014 American Chemical Society) In recent years, many studies have been focused on optimizing the heating efficiency in terms of MNP intrinsic properties such as particle size 327, 329, 331, 332, anisotropy constant 329, 333-335, saturation magnetization 328, easy axis orientations 330, 336, extrinsic properties such as AC field frequency 327 and amplitude 329, 334, and the role of dipolar interactions 322, 329, 330, 337-339. Ruta et al. 328 reported a kinetic Monte-Carlo (MC) method for the study of dipolar interaction, particle size, and saturation magnetization dependence of the SAR. The dependence of SAR on the particle diameter for two different values of 𝑀𝑠 is shown in Figure 23(a). For low 𝑀𝑠 = 200 𝑒𝑚𝑢 𝑐𝑚3 the dipolar interactions are weak ⁄ and SAR does not depend on volumetric packing fraction . The MC calculation also extended toward the hysteretic regime in Figure 23(b), for small anisotropy constant 𝐾, increasing  can lead to enhancement of SAR (see curve 3) while, on the other hand, for large 𝐾, the dipolar interactions are seen to decrease the SAR value (see curve 1). Martinez-Boubeta et al. 329 demonstrated that single-domain cubic iron oxide particles show superior magnetic heating efficiency compared to spherical particles of similar size, evidencing the beneficial role of surface anisotropy in the improved heating power. The observed area of the hysteresis loop of the cubic particle is bigger than that of spherical one as shown in Figure 23(c), which is further confirmed by experimental results 51 in Figure 23(d) and MC simulation in Figure 23(e). On the other hand, the oriented attachment of MNPs into chain-like arrangements, biomimicking magnetotactic bacteria, have been recognized as an important pathway in the magnetic hyperthermia therapy roadmap. Serantes et al. 330 reported that the MNP chain has superior heating performance, increasing more than 5 times in comparison with the randomly distributed system when aligned with the magnetic field. In their study, the hysteresis cycles of an ensemble of non-interacting chains formed by N particles each were firstly investigated, a random angular distribution of anisotropy easy axis within the angle  that defines a cone around the longitudinal direction of the chain is assumed as shown in Figure 23(f). The role of chain length is illustrated in Figure 23(g), for the extreme cases of N = 2 and 12, respectively, the hysteresis area increases with increasing chain length, and it strongly depends on the filed amplitude 𝐻𝑚𝑎𝑥. Furthermore, they found that an increase of solution viscosity (as shown in Figure 23(h), agar > 0.5%) results in a decrease in the SAR. Heating power is also dependent on the relative orientation between AC field direction and chain, as shown in Figure 23(i). 7 MNPs for Bioseparation and Manipulation 7.1 Magnetic Separation The separation and concentration of target analytes during sample preparation are primary steps in many biological studies such as disease diagnosis. Although magnetic diagnosis nowadays makes it possible to conduct bioassays on minimally processed biofluid samples, it remains a great challenge for the detection/quantification of analytes that are of very low-abundance due to the current instrumental detection limits and/or interferences from the complicated matrix. Based on this need, the sample preparation prior to an analytical process is introduced in order to improve the detection limit, this sample preparation step generally involves isolation of analytes from sample matrix, removal of interfering species and enrichment of analytes 46, 340-342. Magnetic separation technique has several advantages in comparison with conventional separation methods such as ultrafiltration 343, 344, precipitation345, 346, electrophoresis 347-350, etc. Furthermore, magnetic separation is usually gentle and nondestructive to biological analytes such as protein and peptides 351. As shown in Figure 24(a), in a typical magnetic separation process, MNPs (or magnetic beads, MB) are employed to specifically conjugate to target analytes (cells, proteins, pathogenic substances, etc.), then the conjugated complexes can be easily and selectively removed by a subsequent magnetic separation process 352. Magnetic separation technique has been applied in combination with the majority of other procedures used in biological analysis 353. Chen et al. 212 reported a one-step sensing methodology that combines Magnetic Separation and Magnetic Relaxation Switching (MS-MRSw) assay for the detection of bacteria and viruses with high sensitivity and reproducibility. As shown in Figure 24(b), this method employs the different separation speeds of 52 MB30 (30 nm magnetic bead) and MB250 (250 nm magnetic bead) in a small magnetic field (0.01 T), and the 𝑇2 of the water molecules around the unreacted MB30 as the readout of the immunoassay. This MS-MRSw based assay reached a LOD of 102 cfu/mL for the detection of S. enterica, compared to the conventional MRSw assay with a LOD of 104 cfu/mL. Yang et al. 293 reported a sensing methodology that combines Magnetic Separation and Surface-enhanced Raman Scattering (MS-SERS). They have successfully applied this method for the detection of chloramphenicol (CAP) and reached a LOD of 1 pg/mL with a detection range of 1 -- 104 pg/mL. Figure 24. (a) The general procedure of magnetic separation in sample preparation for biological analysis. (b) Schematic illustration of the MS-MRSw sensor. MB250 and MB30 can selectively capture and enrich the target to form the sandwich "MB250-target-MB30" conjugate. After the magnetic separation, the 𝑇2 signal of water molecules around the unreacted MB30 can be employed as the readout. (c) The principle of magnetic separation- based blood purification: elimination of pathogens. ((b) reprinted with permission from reference 212, copyright 2015 American Chemical Society) 53 7.2 Magnetic Manipulation In addition to magnetic separation, this technique has also been applied to manipulate and/or sort cells, such as blood purification 47, 354-357, where the direct removal of disease-causing compounds is inherently attractive treatment modality for a range of pathological conditions such as bloodstream infections. In this magnetic separation-based blood purification process, MNPs/MBs conjugated with capture agents are injected into an extracorporeal blood circuit (see Figure 24(c)). Then the pathogen loaded MNPs/MNs are removed from the blood by magnetic separation. The effectiveness of blood purification is largely dependent on the target-ligand binding affinity. 8 Conclusions and Future Perspectives The past three decades have been an exciting period in the synthesis of MNPs with interesting physical properties for their use in biological and biomedical applications, and many of these synthesis methods have been put in commercial production. To date, MNP is a burgeoning field as more improved techniques are being available for clinical therapy and diagnostics with increased sensitivity and efficiency. Furthermore, the surface functionalization of MNPs with polymers, biomolecules and ligands is crucial in order to impart biological recognition and interaction skills. Different kinds of MNPs have been developed to assist the bio-medical applications, such as MNPs with high magnetic moment, core@shell MNPs, core@shell@shell MNPs etc. The tunable properties due to the existence of the surface groups largely increases MNPs' compatibility with massive biomedical applications. As a matter of fact, it is also important that the toxicity of nanoparticles in biomedicine is also taken into consideration in future. It can be foreseen that in the near future the design of MNPs will revolutionize the medical healthcare field by their applications in the development of ultrasensitive and multiplexed diagnostic systems, in vivo imaging systems with high spatial and temporal resolutions, targeted and remotely controlled drug/gene delivery system for effective treatment of diseases, and gene therapy. MR and Hall sensors represent another equally important example which is also a product of interdisciplinary associations. Besides, there are some very encouraging nanotechnological methods that include molecular-ruler technique, nanotransfer printing and nanoskiving that are yet to be used to fabricate magnetic structures. Although recent in origin, micromagnetic tools for simulation and modelling of magnetic materials have opened a new horizon of research that helps predicting properties of biosensors and their approximate results in their applications even before fabrication. This makes the total process all the more cost effective, error free and therefore saves time. It is rational to expect that the future progress in magnetism-based nanomedicine lies in the development of the latest technological advances in the fields of micro- and nano-fabrication. Apart from using these magnetic biosensors as portable lab-on-chip devices, the fact that they were integrated with IoT platforms, 54 increased their commercial value to a great extent. This in turn had revolutionized the modern eHealth architecture. Author Information To be filled. Acknowledgements This study was financially supported by the Institute of Engineering in Medicine of the University of Minnesota, National Science Foundation MRSEC facility program, the Distinguished McKnight University Professorship, Centennial Chair Professorship, Robert F Hartmann Endowed Chair, and UROP program from the University of Minnesota. References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. G. Reiss and A. Hütten, Nature materials, 2005, 4, 725-726. X. Sun, Y. Huang and D. E. Nikles, International journal of nanotechnology, 2004, 1, 328-346. J.-W. Liao, H.-W. Zhang and C.-H. Lai, Magnetic Nanomaterials: From Fundmental to Applications, 2017. L. Yu, J. Liu, K. Wu, T. Klein, Y. Jiang and J.-P. Wang, Scientific reports, 2014, 4. I. S. Barreto, S. I. Andrade, F. A. Cunha, M. B. Lima, M. C. U. Araujo and L. F. Almeida, Talanta, 2018, 178, 384-391. Z. Li, G. Wang, Y. Shen, N. Guo and N. Ma, Advanced Functional Materials, 2018, 28, 1870089. Y. Jia, Y. Peng, J. Bai, X. Zhang, Y. Cui, B. Ning, J. Cui and Z. Gao, Sensors and Actuators B: Chemical, 2018, 254, 629- 635. Y. Sun, L. Fang, Y. Wan and Z. Gu, Sensors and Actuators B: Chemical, 2018, 259, 428-432. Y. Hu, W. Ma, M. Tao, X. Zhang, X. Wang, X. Wang and L. Chen, Journal of Applied Polymer Science, 2018, 135. F. M. Moghaddam, V. Saberi, S. Kalhor and N. Veisi, Applied Organometallic Chemistry, 2018. A. Kuwahata, M. Kaneko, S. Chikaki, M. Kusakabe and M. Sekino, AIP Advances, 2018, 8, 056713. T. Leshuk, A. Holmes, D. Ranatunga, P. Z. Chen, Y. Jiang and F. Gu, Environmental Science: Nano, 2018. A. K. P. Tay, in Acute and Chronic Neural Stimulation via Mechano-Sensitive Ion Channels, Springer, 2018, pp. 83-93. R. Dalpozzo, Green Chemistry, 2015, 17, 3671-3686. X. Bian, K. Hong, X. Ge, R. Song, L. Liu and M. Xu, The Journal of Physical Chemistry C, 2015, 119, 1700-1705. H. Wang, J. Covarrubias, H. Prock, X. Wu, D. Wang and S. H. Bossmann, The Journal of Physical Chemistry C, 2015, 119, 26020-26028. Q. Zhou, Z. Wan, X. Yuan and J. Luo, Applied Organometallic Chemistry, 2016, 30, 215-220. B. G. Krug and J. A. Asumadu, IEEE Sensors Journal, 2015, 15, 4174-4177. A. Tay and D. Di Carlo, Nano letters, 2017, 17, 886-892. M. B. Gawande, Y. Monga, R. Zboril and R. Sharma, Coordination Chemistry Reviews, 2015, 288, 118-143. 55 21. Y. Qiao, J. Gumin, C. J. MacLellan, F. Gao, R. Bouchard, F. F. Lang, R. J. Stafford and M. P. Melancon, Nanotechnology, 2018, 29, 165101. 22. M. A. Miller, S. Gadde, C. Pfirschke, C. Engblom, M. M. Sprachman, R. H. Kohler, K. S. Yang, A. M. Laughney, G. Wojtkiewicz and N. Kamaly, Science translational medicine, 2015, 7, 314ra183-314ra183. 23. W.-H. Chen, G.-F. Luo, Q. Lei, F.-Y. Cao, J.-X. Fan, W.-X. Qiu, H.-Z. Jia, S. Hong, F. Fang and X. Zeng, Biomaterials, 2016, 76, 87-101. 24. J. J. Fütterer, A. Briganti, P. De Visschere, M. Emberton, G. Giannarini, A. Kirkham, S. S. Taneja, H. Thoeny, G. Villeirs and A. Villers, European urology, 2015, 68, 1045-1053. 25. 26. E. H. Hamoen, M. de Rooij, J. A. Witjes, J. O. Barentsz and M. M. Rovers, European urology, 2015, 67, 1112-1121. I. G. Schoots, N. Petrides, F. Giganti, L. P. Bokhorst, A. Rannikko, L. Klotz, A. Villers, J. Hugosson and C. M. Moore, European urology, 2015, 67, 627-636. 27. C. Kuhlmann, A. P. Khandhar, R. M. Ferguson, S. Kemp, T. Wawrzik, M. Schilling, K. M. Krishnan and F. Ludwig, IEEE transactions on magnetics, 2015, 51, 1-4. 28. 29. 30. 31. 32. 33. Y. Du, P. T. Lai, C. H. Leung and P. W. Pong, International journal of molecular sciences, 2013, 14, 18682-18710. D. Eberbeck, F. Wiekhorst, S. Wagner and L. Trahms, Applied physics letters, 2011, 98, 182502. R. M. Ferguson, K. R. Minard and K. M. Krishnan, Journal of magnetism and magnetic materials, 2009, 321, 1548-1551. R. M. Ferguson, K. R. Minard, A. P. Khandhar and K. M. Krishnan, Medical physics, 2011, 38, 1619-1626. R. M. Ferguson, A. P. Khandhar and K. M. Krishnan, Journal of applied physics, 2012, 111, 07B318. R. M. Ferguson, A. P. Khandhar, S. J. Kemp, H. Arami, E. U. Saritas, L. R. Croft, J. Konkle, P. W. Goodwill, A. Halkola and J. Rahmer, IEEE transactions on medical imaging, 2015, 34, 1077-1084. 34. C. Blanco-Andujar, F. Teran and D. Ortega, in Iron Oxide Nanoparticles for Biomedical Applications, Elsevier, 2018, pp. 197-245. 35. M. L. Tebaldi, C. M. Oda, L. O. Monteiro, A. L. de Barros, C. J. Santos and D. C. F. Soares, Journal of Magnetism and 36. 37. Magnetic Materials, 2018. I. Andreu, E. Natividad, L. Solozábal and O. Roubeau, ACS nano, 2015, 9, 1408-1419. S. Kossatz, J. Grandke, P. Couleaud, A. Latorre, A. Aires, K. Crosbie-Staunton, R. Ludwig, H. Dähring, V. Ettelt and A. Lazaro-Carrillo, Breast Cancer Research, 2015, 17, 66. 38. D. Coluccia, C. A. Figueiredo, M. Y. Wu, A. N. Riemenschneider, R. Diaz, A. Luck, C. Smith, S. Das, C. Ackerley and M. O'Reilly, Nanomedicine: Nanotechnology, Biology and Medicine, 2018, 14, 1137-1148. 39. J. Huang, Y. Li, A. Orza, Q. Lu, P. Guo, L. Wang, L. Yang and H. Mao, Advanced functional materials, 2016, 26, 3818- 3836. 40. 41. 42. 43. 44. S. J. Mattingly, M. G. O'Toole, K. T. James, G. J. Clark and M. H. Nantz, Langmuir, 2015, 31, 3326-3332. Y. Zhao, Z. Luo, M. Li, Q. Qu, X. Ma, S. H. Yu and Y. Zhao, Angewandte Chemie International Edition, 2015, 54, 919-922. O. Mykhaylyk, Y. Sanchez-Antequera, D. Vlaskou, M. B. Cerda, M. Bokharaei, E. Hammerschmid, M. Anton and C. Plank, in RNA Interference, Springer, 2015, pp. 53-106. Y. Ding, S. Z. Shen, H. Sun, K. Sun, F. Liu, Y. Qi and J. Yan, Materials Science and Engineering: C, 2015, 48, 487-498. D. Zhang, J. P. Berry, D. Zhu, Y. Wang, Y. Chen, B. Jiang, S. Huang, H. Langford, G. Li and P. A. Davison, The ISME journal, 2015, 9, 603. 45. S. Kyeong, C. Jeong, H. Kang, H.-J. Cho, S.-J. Park, J.-K. Yang, S. Kim, H.-M. Kim, B.-H. Jun and Y.-S. Lee, PloS one, 2015, 10, e0143727. 46. J. He, M. Huang, D. Wang, Z. Zhang and G. Li, Journal of pharmaceutical and biomedical analysis, 2014, 101, 84-101. 56 47. 48. 49. 50. 51. 52. 53. 54. I. Herrmann, A. Schlegel, R. Graf, W. J. Stark and B. Beck-Schimmer, Journal of nanobiotechnology, 2015, 13, 49. S. Li, H. Liu, Y. Deng, L. Lin and N. He, Journal of biomedical nanotechnology, 2013, 9, 1254-1260. M. Iranmanesh and J. Hulliger, Chemical Society Reviews, 2017, 46, 5925-5934. M. Salehiabar, H. Nosrati, S. Davaran, H. Danafar and H. K. Manjili, Drug research, 2017. M. Gharibshahian, O. Mirzaee and M. Nourbakhsh, Journal of Magnetism and Magnetic Materials, 2017, 425, 48-56. V. Balan, M. Butnaru and L. Verestiuc, 2017. B. A. Kairdolf, X. Qian and S. Nie, Analytical chemistry, 2017, 89, 1015-1031. X. Tian, L. Zhang, M. Yang, L. Bai, Y. Dai, Z. Yu and Y. Pan, Wiley Interdisciplinary Reviews: Nanomedicine and Nanobiotechnology, 2018, 10. 55. A. Kirschning, L. L. Wang, K. Seidel, M. Ott, A. Balakrishnan, C. Janko and C. Alexiou, Chemistry-A European Journal, 2017. 56. H. Nosrati, M. Salehiabar, H. K. Manjili, H. Danafar and S. Davaran, International journal of biological macromolecules, 2017. 57. L. Burke, C. J. Mortimer, D. J. Curtis, A. R. Lewis, R. Williams, K. Hawkins, T. G. Maffeis and C. J. Wright, Materials Science and Engineering: C, 2017, 70, 512-519. 58. S. Shabestari Khiabani, M. Farshbaf, A. Akbarzadeh and S. Davaran, Artificial cells, nanomedicine, and biotechnology, 2017, 45, 6-17. 59. 60. 61. 62. 63. 64. 65. 66. 67. 68. P. Srivastava, P. K. Sharma, A. Muheem and M. H. Warsi, Recent patents on drug delivery & formulation, 2017. L. Hajba and A. Guttman, Biotechnology advances, 2016, 34, 354-361. J. B. Haun, T. J. Yoon, H. Lee and R. Weissleder, Wiley Interdisciplinary Reviews: Nanomedicine and Nanobiotechnology, 2010, 2, 291-304. H. Lee, T.-H. Shin, J. Cheon and R. Weissleder, Chemical reviews, 2015, 115, 10690-10724. K. Wu, D. Su, Y. Feng and J.-P. Wang, Clinical Applications of Magnetic Nanoparticles, 2018, 247. P. Hendriksen, S. Linderoth, C. Oxborrow and S. Morup, Journal of Physics: Condensed Matter, 1994, 6, 3091. M. d. P. Morales, S. Veintemillas-Verdaguer, M. Montero, C. Serna, A. Roig, L. Casas, B. Martinez and F. Sandiumenge, Chemistry of Materials, 1999, 11, 3058-3064. M. Morales, C. Serna, F. Bødker and S. Mørup, Journal of Physics: Condensed Matter, 1997, 9, 5461. K. Wu, L. Tu, D. Su and J.-P. Wang, Journal of Physics D: Applied Physics, 2017, 50, 085005. M. Darbandi, F. Stromberg, J. Landers, N. Reckers, B. Sanyal, W. Keune and H. Wende, Journal of Physics D: Applied Physics, 2012, 45, 195001. 69. A. Roca, D. Niznansky, J. Poltierova-Vejpravova, B. Bittova, M. González-Fernández, C. Serna and M. Morales, Journal of 70. 71. 72. 73. 74. Applied Physics, 2009, 105, 114309. T. Kim and M. Shima, Journal of applied physics, 2007, 101, 09M516. P. Dutta, S. Pal, M. Seehra, N. Shah and G. Huffman, Journal of Applied Physics, 2009, 105, 07B501-507B501. M. Abbas, B. P. Rao, S. Naga, M. Takahashi and C. Kim, Ceramics International, 2013, 39, 7605-7611. G. Goya, T. Berquo, F. Fonseca and M. Morales, Journal of Applied Physics, 2003, 94, 3520-3528. P. Guardia, B. Batlle-Brugal, A. Roca, O. Iglesias, M. Morales, C. Serna, A. Labarta and X. Batlle, Journal of Magnetism and Magnetic Materials, 2007, 316, e756-e759. 75. J. Vargas, E. Lima Jr, R. Zysler, J. S. Duque, E. De Biasi and M. Knobel, The European Physical Journal B, 2008, 64, 211- 218. 57 76. J. Park, K. An, Y. Hwang, J.-G. Park, H.-J. Noh, J.-Y. Kim, J.-H. Park, N.-M. Hwang and T. Hyeon, Nature materials, 2004, 3, 891-895. A. Demortiere, P. Panissod, B. Pichon, G. Pourroy, D. Guillon, B. Donnio and S. Begin-Colin, Nanoscale, 2011, 3, 225-232. K. M. Krishnan, IEEE transactions on magnetics, 2010, 46, 2523-2558. K. Wu, J. Liu, Y. Wang, C. Ye, Y. Feng and J.-P. Wang, Applied Physics Letters, 2015, 107, 053701. L. Tu, Y. Jing, Y. Li and J.-P. Wang, Applied Physics Letters, 2011, 98, 213702. R. E. Rosensweig, Journal of magnetism and magnetic materials, 2002, 252, 370-374. K. Wu and J.-P. Wang, AIP Advances, 2017, 7, 056327. J. P. Ridgway, Journal of cardiovascular magnetic resonance, 2010, 12, 71. S. Mornet, S. Vasseur, F. Grasset and E. Duguet, Journal of Materials Chemistry, 2004, 14, 2161-2175. E. U. Saritas, P. W. Goodwill, L. R. Croft, J. J. Konkle, K. Lu, B. Zheng and S. M. Conolly, Journal of Magnetic Resonance, 2013, 229, 116-126. H. Shao, T.-J. Yoon, M. Liong, R. Weissleder and H. Lee, Beilstein journal of nanotechnology, 2010, 1, 142. M. Rohrer, H. Bauer, J. Mintorovitch, M. Requardt and H.-J. Weinmann, Investigative radiology, 2005, 40, 715-724. A. Joos, N. Löwa, F. Wiekhorst, B. Gleich and A. Haase, Journal of Magnetism and Magnetic Materials, 2017, 427, 122- 126. R. A. Brooks, Magnetic resonance in medicine, 2002, 47, 388-391. Z. Zhou, R. Tian, Z. Wang, Z. Yang, Y. Liu, G. Liu, R. Wang, J. Gao, J. Song and L. Nie, Nature Communications, 2017, 8. S. H. Koenig and K. E. Kellar, Magnetic resonance in medicine, 1995, 34, 227-233. L. P. Hwang and J. H. Freed, The Journal of Chemical Physics, 1975, 63, 4017-4025. J. H. Freed, The Journal of Chemical Physics, 1978, 68, 4034-4037. N. Venkatesha, P. Poojar, Y. Qurishi, S. Geethanath and C. Srivastava, Journal of Applied Physics, 2015, 117, 154702. J. Mohapatra, A. Mitra, H. Tyagi, D. Bahadur and M. Aslam, Nanoscale, 2015, 7, 9174-9184. J. t. Jang, H. Nah, J. H. Lee, S. H. Moon, M. G. Kim and J. Cheon, Angewandte Chemie, 2009, 121, 1260-1264. M. Guéron, Journal of Magnetic Resonance (1969), 1975, 19, 58-66. S. Mørup, M. F. Hansen and C. Frandsen, Beilstein journal of nanotechnology, 2010, 1, 182-190. A. Weizenmann, M. Santos and W. Figueiredo, Physics Letters A, 2012, 376, 1535-1539. 77. 78. 79. 80. 81. 82. 83. 84. 85. 86. 87. 88. 89. 90. 91. 92. 93. 94. 95. 96. 97. 98. 99. 100. L. C. Branquinho, M. S. Carrião, A. S. Costa, N. Zufelato, M. H. Sousa, R. Miotto, R. Ivkov and A. F. Bakuzis, Scientific reports, 2013, 3, 2887. 101. D. F. Coral, P. Mendoza Zelis, M. Marciello, M. a. d. P. Morales, A. Craievich, F. H. Sanchez and M. B. Fernandez van Raap, Langmuir, 2016, 32, 1201-1213. 102. K. Them, Physics in Medicine & Biology, 2017, 62, 5623. 103. S. Dutz, IEEE Transactions on Magnetics, 2016, 52, 1-3. 104. L. n. Lartigue, P. Hugounenq, D. Alloyeau, S. P. Clarke, M. Levy, J.-C. Bacri, R. Bazzi, D. F. Brougham, C. Wilhelm and F. Gazeau, ACS nano, 2012, 6, 10935-10949. 105. H. Kratz, M. Taupitz, A. A. de Schellenberger, O. Kosch, D. Eberbeck, S. Wagner, L. Trahms, B. Hamm and J. Schnorr, PloS one, 2018, 13, e0190214. 106. J.-J. Lai, W.-R. Lai, C.-Y. Chen, S.-W. Chen and C.-L. Chiang, Journal of Magnetism and Magnetic Materials, 2013, 331, 204-207. 107. J. S. Benjamin, Metallurgical transactions, 1970, 1, 2943-2951. 108. H. Fecht, E. Hellstern, Z. Fu and W. Johnson, Metallurgical Transactions A, 1990, 21, 2333. 58 109. S. Li, K. Wang, L. Sun and Z. Wang, Scripta metallurgica et materialia, 1992, 27, 437-442. 110. C.-N. Chen, Y.-L. Chen and W. J. Tseng, Journal of materials processing technology, 2007, 190, 61-64. 111. Y. Jiang, J. Liu, P. K. Suri, G. Kennedy, N. N. Thadhani, D. J. Flannigan and J. P. Wang, Advanced Engineering Materials, 2016, 18, 1009-1016. 112. W. Kaczmarek, R. Bramley, A. Calka and B. Ninham, IEEE transactions on magnetics, 1990, 26, 1840-1842. 113. E. Kirkpatrick, S. Majetich and M. McHenry, IEEE Transactions on Magnetics, 1996, 32, 4502-4504. 114. Y. Wang, Y. Li, C. Rong and J. P. Liu, Nanotechnology, 2007, 18, 465701. 115. V. Chakka, B. Altuncevahir, Z. Jin, Y. Li and J. Liu, Journal of Applied Physics, 2006, 99, 08E912. 116. N. Poudyal, C.-b. Rong and J. P. Liu, Journal of Applied Physics, 2011, 109, 07B526. 117. C. Granqvist and R. Buhrman, Journal of applied Physics, 1976, 47, 2200-2219. 118. K. Sattler, J. Mühlbach and E. Recknagel, Physical Review Letters, 1980, 45, 821. 119. H. Haberland, M. Karrais, M. Mall and Y. Thurner, Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, 1992, 10, 3266-3271. 120. S. Yamamuro, K. Sumiyama, M. Sakurai and K. Suzuki, Supramolecular Science, 1998, 5, 239-245. 121. J.-M. Qiu and J.-P. Wang, Applied physics letters, 2006, 88, 192505. 122. J. Liu, K. Schliep, S.-H. He, B. Ma, Y. Jing, D. J. Flannigan and J.-P. Wang, Physical Review Materials, 2018, 2, 054415. 123. T. Vystavel, G. Palasantzas, S. Koch and J. T. M. De Hosson, Applied physics letters, 2003, 82, 197-199. 124. J. Liu, K. Wu and J.-P. Wang, AIP Advances, 2016, 6, 056126. 125. Y. H. Xu, J. Bai and J.-P. Wang, Journal of Magnetism and Magnetic Materials, 2007, 311, 131-134. 126. Y. H. Xu and J. P. Wang, Advanced Materials, 2008, 20, 994-999. 127. L. Martínez, A. Mayoral, M. Espiñeira, E. Roman, F. J. Palomares and Y. Huttel, Nanoscale, 2017, 9, 6463-6470. 128. D. Llamosa, M. Ruano, L. Martinez, A. Mayoral, E. Roman, M. Garcia-Hernandez and Y. Huttel, Nanoscale, 2014, 6, 13483-13486. 129. Y.-H. Xu and J.-P. Wang, Applied Physics Letters, 2007, 91, 233107. 130. X.-W. Wei, G.-X. Zhu, Y.-J. Liu, Y.-H. Ni, Y. Song and Z. Xu, Chemistry of Materials, 2008, 20, 6248-6253. 131. Y. Wang, Y. Zheng and S. Hu, Journal of Physics and Chemistry of Solids, 2017, 100, 78-82. 132. W. S. Seo, J. H. Lee, X. Sun, Y. Suzuki, D. Mann, Z. Liu, M. Terashima, P. C. Yang, M. V. McConnell and D. G. Nishimura, Nature materials, 2006, 5, 971. 133. Y. Xu, Y. Ma, Y. Liu, S. Feng, D. He, P. Haghi-Ashtiani, A. Dichiara, L. Zimmer and J. Bai, The Journal of Physical Chemistry C, 2018, 122, 6437-6446. 134. G. S. Chaubey, C. Barcena, N. Poudyal, C. Rong, J. Gao, S. Sun and J. P. Liu, Journal of the American Chemical Society, 2007, 129, 7214-7215. 135. N. J. Orsini, B. Babić-Stojić, V. Spasojević, M. Calatayud, N. Cvjetićanin and G. Goya, Journal of Magnetism and Magnetic Materials, 2018, 449, 286-296. 136. X. Li, J. Wei, K. E. Aifantis, Y. Fan, Q. Feng, F. Z. Cui and F. Watari, Journal of Biomedical Materials Research Part A, 2016, 104, 1285-1296. 137. N. Kohler, C. Sun, A. Fichtenholtz, J. Gunn, C. Fang and M. Q. Zhang, Small, 2006, 2, 785-792. 138. P. Biehl, M. von der Luhe, S. Dutz and F. H. Schacher, Polymers, 2018, 10, 28. 139. V. P. Torchilin and V. S. Trubetskoy, Advanced Drug Delivery Reviews, 1995, 16, 141-155. 140. J. Xie, C. Xu, N. Kohler, Y. Hou and S. Sun, Advanced Materials, 2007, 19, 3163-+. 59 141. E. Strable, J. W. M. Bulte, B. Moskowitz, K. Vivekanandan, M. Allen and T. Douglas, Chemistry of Materials, 2001, 13, 2201-2209. 142. F. Quaglia, L. Ostacolo, G. Nese, M. Canciello, G. De Rosa, F. Ungaro, R. Palumbo, M. I. La Rotonda and G. Maglio, Journal of Biomedical Materials Research Part A, 2008, 87A, 563-574. 143. L. M. Lacava, Z. G. M. Lacava, M. F. Da Silva, O. Silva, S. B. Chaves, R. B. Azevedo, F. Pelegrini, C. Gansau, N. Buske, D. Sabolovic and P. C. Morais, Biophysical Journal, 2001, 80, 2483-2486. 144. H. Wei, N. Insin, J. Lee, H. S. Han, J. M. Cordero, W. H. Liu and M. G. Bawendi, Nano Letters, 2012, 12, 22-25. 145. Y. Liu, T. Chen, C. C. Wu, L. P. Qiu, R. Hu, J. Li, S. Cansiz, L. Q. Zhang, C. Cui, G. Z. Zhu, M. X. You, T. Zhang and W. H. Tan, Journal of the American Chemical Society, 2014, 136, 12552-12555. 146. R. Mincheva, O. Stoilova, H. Penchev, T. Ruskov, I. Spirov, N. Manolova and I. Rashkov, European Polymer Journal, 2008, 44, 615-627. 147. J. Cheng, B. A. Teply, I. Sherifi, J. Sung, G. Luther, F. X. Gu, E. Levy-Nissenbaum, A. F. Radovic-Moreno, R. Langer and O. C. Farokhzad, Biomaterials, 2007, 28, 869-876. 148. L. Charoenmark, D. Polpanich, R. Thiramanas and P. Tangboriboonrat, Macromolecular Research, 2012, 20, 590-596. 149. S. E. Lupold, B. J. Hicke, Y. Lin and D. S. Coffey, Cancer Research, 2012, 72, 3887-3887. 150. P. Crespo, P. de la Presa, P. Marin, M. Multigner, J. M. Alonso, G. Rivero, F. Yndurain, J. M. Gonzalez-Calbet and A. Hernando, Journal of Physics-Condensed Matter, 2013, 25. 151. S. Chen, L. J. Wang, S. L. Duce, S. Brown, S. Lee, A. Melzer, S. A. Cuschieri and P. Andre, Journal of the American Chemical Society, 2010, 132, 15022-15029. 152. A. Villanueva, P. de la Presa, J. M. Alonso, T. Rueda, A. Martinez, P. Crespo, M. P. Morales, M. A. Gonzalez-Fernandez, J. Valdes and G. Rivero, Journal of Physical Chemistry C, 2010, 114, 1976-1981. 153. C. Kumar and F. Mohammad, Advanced Drug Delivery Reviews, 2011, 63, 789-808. 154. P. de la Presa, M. Multigner, M. P. Morales, T. Rueda, E. Fernandez-Pinel and A. Hernando, Journal of Magnetism and Magnetic Materials, 2007, 316, E753-E755. 155. Y. Zhang, H. Yang, Z. Zhou, K. Huang, S. Yang and G. Han, Bioconjugate Chem., 2017, 28, 869-879. 156. R. Lamberton, M. Seigler, K. Pelhos, H. Zhou, M. McCurry, M. Ormston, G. Yi, G. McClean, T. McLaughlin and P. J. I. t. o. m. Kolbo, 2007, 43, 645-650. 157. A. Glass and M. Cardillo, Holographic data storage, Springer, 2012. 158. C.-C. Huang, C.-L. Lin, J.-J. Kao, J.-J. Chang and G.-J. J. I. T. o. V. T. Sheu, 2018. 159. D. García-Romeo, N. Medrano, B. Calvo, P. Martínez, M. Cubells-Beltrán, C. Reig, S. Cardoso and P. Freitas, 2015. 160. Y. Ouyang, J. He, J. Hu and S. X. J. S. Wang, 2012, 12, 15520-15541. 161. Y. Xianyu, Q. Wang and Y. J. T. T. i. A. C. Chen, 2018. 162. D. A. Hall, R. S. Gaster, S. J. Osterfeld, K. Makinwa, S. X. Wang and B. Murmann, 2011. 163. G. Binasch, P. Grunberg, F. Saurenbach and W. Zinn, Physical Review B, 1989, 39, 4828-4830. 164. M. N. Baibich, J. M. Broto, A. Fert, F. N. Vandau, F. Petroff, P. Eitenne, G. Creuzet, A. Friederich and J. Chazelas, Physical Review Letters, 1988, 61, 2472-2475. 165. A. H. Wilson, Proceedings of the Royal Society of London Series a-Mathematical and Physical Sciences, 1938, 167, 0580- 0593. 166. Y. Feng, J. Liu, T. Klein, K. Wu and J.-P. Wang, Journal of Applied Physics, 2017, 122, 123901. 167. D. R. Baselt, G. U. Lee, M. Natesan, S. W. Metzger, P. E. Sheehan and R. J. Colton, Biosensors & Bioelectronics, 1998, 13, 731-739. 60 168. Y. Wang, W. Wang, L. N. Yu, L. Tu, Y. L. Feng, T. Klein and J. P. Wang, Biosens. Bioelectron., 2015, 70, 61-68. 169. B. Srinivasan, Y. Li, Y. Jing, Y. Xu, X. Yao, C. Xing and J. P. Wang, Angewandte Chemie International Edition, 2009, 48, 2764-2767. 170. S. X. Wang and G. Li, IEEE transactions on Magnetics, 2008, 44, 1687-1702. 171. K. Wu, T. Klein, V. D. Krishna, D. Q. Su, A. M. Perez and J. P. Wang, Acs Sensors, 2017, 2, 1594-1601. 172. J. Choi, A. W. Gani, D. J. B. Bechstein, J. R. Lee, P. J. Utz and S. X. Wang, Biosens. Bioelectron., 2016, 85, 1-7. 173. G. Rizzi, J. R. Lee, C. Dahl, P. Guldberg, M. Dufva, S. X. Wang and M. F. Hansen, Acs Nano, 2017, 11, 8864-8870. 174. Y. C. Liang, L. Chang, W. L. Qiu, A. G. Kolhatkar, B. Vu, K. Kourentzi, T. R. Lee, Y. L. Zu, R. Willson and D. Litvinov, Sensors, 2017, 17, 10. 175. J. Park, Sens. Actuator A-Phys., 2016, 250, 55-59. 176. Y. P. Li, Y. Wang, T. Klein and J. P. Wang, Applied Physics Letters, 2014, 104, 5. 177. D. J. Bechstein, J.-R. Lee, C. C. Ooi, A. W. Gani, K. Kim, R. J. Wilson and S. X. Wang, Scientific reports, 2015, 5, 11693. 178. J. Schotter, P. B. Kamp, A. Becker, A. Puhler, G. Reiss and H. Bruckl, Biosens. Bioelectron., 2004, 19, 1149-1156. 179. V. D. Krishna, K. Wu, A. M. Perez and J. P. Wang, Frontiers in Microbiology, 2016, 7, 8. 180. X. C. Sun, C. Lei, L. Guo and Y. Zhou, Sens. Actuator B-Chem., 2016, 234, 485-492. 181. V. D. Krishna, K. Wu, D. Su, M. C. Cheeran, J.-P. Wang and A. Perez, Food Microbiology, 2018. 182. V. K. Chugh, K. Kalyan, C. Anoop, A. Patra and S. Negi, 2017. 183. K. Kalyan, V. K. Chugh and C. Anoop, 2016. 184. V. K. Chugh, K. Kalyan and C. Anoop, 2016. 185. I. Giouroudi and E. J. J. o. A. P. Hristoforou, 2018, 124, 030902. 186. S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Hughes, M. Samant and S. H. Yang, Nature Materials, 2004, 3, 862- 867. 187. P. P. Sharma, E. Albisetti, M. Massetti, M. Scolari, C. La Torre, M. Monticelli, M. Leone, F. Damin, G. Gervasoni, G. Ferrari, F. Salice, E. Cerquaglia, G. Falduti, M. Cretich, E. Marchisio, M. Chiari, M. Sampietro, D. Petti and R. Bertacco, Sens. Actuator B-Chem., 2017, 242, 280-287. 188. F. A. Cardoso, J. Germano, R. Ferreira, S. Cardoso, V. C. Martins, P. P. Freitas, M. S. Piedade and L. Sousa, Journal of Applied Physics, 2008, 103. 189. V. Nabaei, R. Chandrawati and H. Heidari, Biosens. Bioelectron., 2018, 103, 69-86. 190. P. A. Besse, G. Boero, M. Demierre, V. Pott and R. Popovic, Applied Physics Letters, 2002, 80, 4199-4201. 191. G. Mihajlovic, P. Xiong, S. von Molnar, K. Ohtani, H. Ohno, M. Field and G. J. Sullivan, Applied Physics Letters, 2005, 87. 192. L. Di Michele, C. Shelly, P. de Marco, P. See, D. Cox and O. Kazakova, Journal of Applied Physics, 2011, 110. 193. K. Aledealat, G. Mihajlovic, K. Chen, M. Field, G. J. Sullivan, P. Xiong, P. B. Chase and S. von Molnar, Journal of Magnetism and Magnetic Materials, 2010, 322, L69-L72. 194. D. Issadore, H. J. Chung, J. Chung, G. Budin, R. Weissleder and H. Lee, Advanced Healthcare Materials, 2013, 2, 1224- 1228. 195. A. Sandhu and H. Handa, 2005. 196. D. Alcantara, S. Lopez, M. L. García-Martin and D. Pozo, Nanomedicine: Nanotechnology, Biology and Medicine, 2016, 12, 1253-1262. 197. I. Koh and L. Josephson, Sensors, 2009, 9, 8130-8145. 198. C. Min, H. Shao, M. Liong, T.-J. Yoon, R. Weissleder and H. Lee, ACS nano, 2012, 6, 6821-6828. 199. S. Laurent, D. Forge, M. Port, A. Roch, C. Robic, L. Vander Elst and R. N. Muller, Chemical reviews, 2008, 108, 2064-2110. 61 200. R. Hong, M. J. Cima, R. Weissleder and L. Josephson, Magnetic resonance in medicine, 2008, 59, 515-520. 201. I. Koh, R. Hong, R. Weissleder and L. Josephson, Analytical chemistry, 2009, 81, 3618-3622. 202. H. J. Chung, K. L. Pellegrini, J. Chung, K. Wanigasuriya, I. Jayawardene, K. Lee, H. Lee, V. S. Vaidya and R. Weissleder, PloS one, 2015, 10, e0133417. 203. J. M. Perez, F. J. Simeone, Y. Saeki, L. Josephson and R. Weissleder, Journal of the American Chemical Society, 2003, 125, 10192-10193. 204. W. Ma, C. Hao, W. Ma, C. Xing, W. Yan, H. Kuang, L. Wang and C. Xu, Chemical Communications, 2011, 47, 12503- 12505. 205. H. Yang, Z. Tian, J. Wang and S. Yang, Sensors and Actuators B: Chemical, 2012, 161, 429-433. 206. L. Xu, H. Yin, W. Ma, L. Wang, H. Kuang and C. Xu, The Journal of Physical Chemistry B, 2013, 117, 14367-14371. 207. Y. Yang, Y. Zhang, J.-C. Shen, H. Yang, Z.-G. Zhou and S.-P. Yang, Chinese Chemical Letters, 2016, 27, 891-895. 208. Y. Zhang, Y. Zhao, Y. Yang, J. Shen, H. Yang, Z. Zhou and S. Yang, Sensors and Actuators B: Chemical, 2015, 220, 622- 626. 209. J. B. Haun, N. K. Devaraj, S. A. Hilderbrand, H. Lee and R. Weissleder, Nature nanotechnology, 2010, 5, 660. 210. Y. Zhao, Y. Li, K. Jiang, J. Wang, W. L. White, S. Yang and J. Lu, Food Control, 2017, 71, 110-116. 211. S. Wang, Y. Zhang, W. An, Y. Wei, N. Liu, Y. Chen and S. Shuang, Food Control, 2015, 55, 43-48. 212. Y. Chen, Y. Xianyu, Y. Wang, X. Zhang, R. Cha, J. Sun and X. Jiang, ACS nano, 2015, 9, 3184-3191. 213. E. C. Alocilja and Y. Luo, Journal of Biological Engineering, 2017, 11, 14. 214. C. M. Castro, A. A. Ghazani, J. Chung, H. Shao, D. Issadore, T.-J. Yoon, R. Weissleder and H. Lee, Lab on a Chip, 2014, 14, 14-23. 215. N. Sun, T.-J. Yoon, H. Lee, W. Andress, R. Weissleder and D. Ham, IEEE Journal of Solid-State Circuits, 2011, 46, 342- 352. 216. K.-M. Lei, H. Heidari, P.-I. Mak, M.-K. Law, F. Maloberti and R. P. Martins, IEEE Journal of Solid-State Circuits, 2017, 52, 284-297. 217. H. Lee, E. Sun, D. Ham and R. Weissleder, Nature medicine, 2008, 14, 869. 218. S. S. Zalesskiy, E. Danieli, B. Blumich and V. P. Ananikov, Chemical reviews, 2014, 114, 5641-5694. 219. D. Issadore, C. Min, M. Liong, J. Chung, R. Weissleder and H. Lee, Lab on a Chip, 2011, 11, 2282-2287. 220. X.-H. Ma, A. Gong, L.-C. Xiang, T.-X. Chen, Y.-X. Gao, X.-J. Liang, Z.-Y. Shen and A.-G. Wu, Journal of Materials Chemistry B, 2013, 1, 3419-3428. 221. P. H. Kuo, E. Kanal, A. K. Abu-Alfa and S. E. Cowper, Radiology, 2007, 242, 647-649. 222. J. Ramalho, R. Semelka, M. Ramalho, R. Nunes, M. AlObaidy and M. Castillo, American Journal of Neuroradiology, 2016, 37, 1192-1198. 223. M. Rogosnitzky and S. Branch, Biometals, 2016, 29, 365-376. 224. Z. Shen, A. Wu and X. Chen, Molecular pharmaceutics, 2016, 14, 1352-1364. 225. Y.-X. J. Wang, Quantitative imaging in medicine and surgery, 2011, 1, 35. 226. Z. Liu, X. Wang, K. Yao, G. Du, Q. Lu, Z. Ding, J. Tao, Q. Ning, X. Luo and D. Tian, Journal of Materials Science, 2004, 39, 2633-2636. 227. M. Z. Iqbal, X. Ma, T. Chen, L. e. Zhang, W. Ren, L. Xiang and A. Wu, Journal of Materials Chemistry B, 2015, 3, 5172- 5181. 228. L. Zeng, W. Ren, J. Zheng, P. Cui and A. Wu, Physical Chemistry Chemical Physics, 2012, 14, 2631-2636. 229. E. Terreno, D. D. Castelli, A. Viale and S. Aime, Chemical reviews, 2010, 110, 3019-3042. 62 230. C. H. Cunningham, T. Arai, P. C. Yang, M. V. McConnell, J. M. Pauly and S. M. Conolly, Magnetic Resonance in Medicine, 2005, 53, 999-1005. 231. J. Huang, X. Zhong, L. Wang, L. Yang and H. Mao, Theranostics, 2012, 2, 86. 232. M. Stuber, W. D. Gilson, M. Schär, D. A. Kedziorek, L. V. Hofmann, S. Shah, E. J. Vonken, J. W. Bulte and D. L. Kraitchman, Magnetic Resonance in Medicine, 2007, 58, 1072-1077. 233. M. A. McAteer, N. R. Sibson, C. von Zur Muhlen, J. E. Schneider, A. S. Lowe, N. Warrick, K. M. Channon, D. C. Anthony and R. P. Choudhury, Nature medicine, 2007, 13, 1253. 234. H. B. Na and T. Hyeon, Journal of Materials Chemistry, 2009, 19, 6267-6273. 235. N. Lee, H. Kim, S. H. Choi, M. Park, D. Kim, H.-C. Kim, Y. Choi, S. Lin, B. H. Kim and H. S. Jung, Proceedings of the National Academy of Sciences, 2011, 108, 2662-2667. 236. S. Wagner, J. Schnorr, H. Pilgrimm, B. Hamm and M. Taupitz, Investigative radiology, 2002, 37, 167-177. 237. B. H. Kim, N. Lee, H. Kim, K. An, Y. I. Park, Y. Choi, K. Shin, Y. Lee, S. G. Kwon and H. B. Na, Journal of the American Chemical Society, 2011, 133, 12624-12631. 238. U. I. Tromsdorf, O. T. Bruns, S. C. Salmen, U. Beisiegel and H. Weller, Nano letters, 2009, 9, 4434-4440. 239. J. Y. Park, P. Daksha, G. H. Lee, S. Woo and Y. Chang, Nanotechnology, 2008, 19, 365603. 240. T.-H. Shin, J.-s. Choi, S. Yun, I.-S. Kim, H.-T. Song, Y. Kim, K. I. Park and J. Cheon, ACS nano, 2014, 8, 3393-3401. 241. M. Yang, L. Gao, K. Liu, C. Luo, Y. Wang, L. Yu, H. Peng and W. Zhang, Talanta, 2015, 131, 661-665. 242. J.-s. Choi, J.-H. Lee, T.-H. Shin, H.-T. Song, E. Y. Kim and J. Cheon, Journal of the American Chemical Society, 2010, 132, 11015-11017. 243. H. Yang, Y. Zhuang, Y. Sun, A. Dai, X. Shi, D. Wu, F. Li, H. Hu and S. Yang, Biomaterials, 2011, 32, 4584-4593. 244. A. Xia, Y. Gao, J. Zhou, C. Li, T. Yang, D. Wu, L. Wu and F. Li, Biomaterials, 2011, 32, 7200-7208. 245. Z. Zhou, D. Huang, J. Bao, Q. Chen, G. Liu, Z. Chen, X. Chen and J. Gao, Advanced materials, 2012, 24, 6223-6228. 246. M. D. Shultz, S. Calvin, P. P. Fatouros, S. A. Morrison and E. E. Carpenter, Journal of Magnetism and Magnetic Materials, 2007, 311, 464-468. 247. O. B. Miguel, Y. Gossuin, M. Morales, P. Gillis, R. Muller and S. Veintemillas-Verdaguer, Magnetic resonance imaging, 2007, 25, 1437-1441. 248. T. Tegafaw, W. Xu, M. W. Ahmad, J. S. Baeck, Y. Chang, J. E. Bae, K. S. Chae, T. J. Kim and G. H. Lee, Nanotechnology, 2015, 26, 365102. 249. D. Niu, X. Luo, Y. Li, X. Liu, X. Wang and J. Shi, ACS applied materials & interfaces, 2013, 5, 9942-9948. 250. L. Tu, T. Klein, W. Wang, Y. Feng, Y. Wang and J.-P. Wang, Magnetics, IEEE Transactions on, 2013, 49, 227-230. 251. L. Tu, Y. Feng, T. Klein, W. Wang and J.-P. Wang, Magnetics, IEEE Transactions on, 2012, 48, 3513-3516. 252. L. Tu, K. Wu, T. Klein and J.-P. Wang, Journal of Physics D: Applied Physics, 2014, 47, 155001. 253. M. Lucchini and F. Canepa, Journal of Nanoparticle Research, 2012, 14, 1-7. 254. I. Perreard, D. Reeves, X. Zhang and J. B. Weaver, 2013. 255. A. M. Rauwerdink, E. W. Hansen and J. B. Weaver, Physics in medicine and biology, 2009, 54, L51. 256. J. B. Weaver, M. Harding, A. M. Rauwerdink and E. W. Hansen, 2010. 257. K. Wu, C. Ye, J. Liu, Y. Wang, Y. Feng and J.-P. Wang, IEEE Transactions on magnetics, 2016, 52. 258. K. Wu, L. Yu, X. Zheng, Y. Wang, Y. Feng, L. Tu and J.-P. Wang, 2014. 259. S. Chung, A. Hoffmann, S. Bader, C. Liu, B. Kay, L. Makowski and L. Chen, Applied physics letters, 2004, 85, 2971-2973. 260. A. M. Rauwerdink and J. B. Weaver, Applied Physics Letters, 2010, 96, 033702. 63 261. S.-H. Chung, A. Hoffmann, K. Guslienko, S. Bader, C. Liu, B. Kay, L. Makowski and L. Chen, Journal of applied physics, 2005, 97, 10R101. 262. X. Zhang, D. B. Reeves, I. M. Perreard, W. C. Kett, K. E. Griswold, B. Gimi and J. B. Weaver, Biosensors and Bioelectronics, 2013, 50, 441-446. 263. H.-J. Krause, N. Wolters, Y. Zhang, A. Offenhäusser, P. Miethe, M. H. Meyer, M. Hartmann and M. Keusgen, Journal of magnetism and magnetic materials, 2007, 311, 436-444. 264. P. I. Nikitin, P. M. Vetoshko and T. I. Ksenevich, Journal of Magnetism and Magnetic Materials, 2007, 311, 445-449. 265. J. B. Weaver and E. Kuehlert, Medical physics, 2012, 39, 2765-2770. 266. A. M. Rauwerdink and J. B. Weaver, Medical physics, 2010, 37, 2587-2592. 267. A. V. Orlov, J. A. Khodakova, M. P. Nikitin, A. O. Shepelyakovskaya, F. A. Brovko, A. G. Laman, E. V. Grishin and P. I. Nikitin, Analytical chemistry, 2012, 85, 1154-1163. 268. B. Gleich and J. Weizenecker, Nature, 2005, 435, 1214-1217. 269. J. Weizenecker, B. Gleich, J. Rahmer, H. Dahnke and J. Borgert, Physics in medicine and biology, 2009, 54, L1. 270. B. Gleich, J. Weizenecker and J. Borgert, Physics in Medicine & Biology, 2008, 53, N81. 271. P. Ludewig, N. Gdaniec, J. Sedlacik, N. D. Forkert, P. Szwargulski, M. Graeser, G. Adam, M. G. Kaul, K. M. Krishnan and R. M. J. A. n. Ferguson, 2017, 11, 10480-10488. 272. P. W. Goodwill, J. J. Konkle, B. Zheng, E. U. Saritas and S. M. Conolly, IEEE transactions on medical imaging, 2012, 31, 1076-1085. 273. S. Zanganeh, M. Aieneravaie, M. Erfanzadeh, J. Ho and R. Spitler, in Iron Oxide Nanoparticles for Biomedical Applications, Elsevier, 2018, pp. 115-133. 274. R. Dhavalikar and C. Rinaldi, Journal of Applied Physics, 2014, 115, 074308. 275. N. Panagiotopoulos, R. L. Duschka, M. Ahlborg, G. Bringout, C. Debbeler, M. Graeser, C. Kaethner, K. Lüdtke-Buzug, H. Medimagh and J. Stelzner, International journal of nanomedicine, 2015, 10, 3097. 276. J. Rahmer, J. Weizenecker, B. Gleich and J. Borgert, BMC medical imaging, 2009, 9, 4. 277. K. Lüdtke-Buzug, J. Haegele, S. Biederer, T. F. Sattel, M. Erbe, R. L. Duschka, J. Barkhausen and F. M. Vogt, Biomedizinische Technik/Biomedical Engineering, 2013, 58, 527-533. 278. J. Weizenecker, B. Gleich, J. Rahmer and J. Borgert, Physics in Medicine & Biology, 2012, 57, 7317. 279. H. Arami, R. Ferguson, A. P. Khandhar and K. M. Krishnan, Medical physics, 2013, 40. 280. H. Arami, A. P. Khandhar, A. Tomitaka, E. Yu, P. W. Goodwill, S. M. Conolly and K. M. Krishnan, Biomaterials, 2015, 52, 251-261. 281. R. Ravanshad, A. K. Zadeh, A. M. Amani, S. M. Mousavi, S. A. Hashemi, A. S. Dashtaki, E. Mirzaei and B. Zare, Nano Reviews & Experiments, 2017, 9. 282. Z. J. Smith, T. R. Huser and S. Wachsmann-Hogiu, Analytical Cellular Pathology, 2012, 35, 145-163. 283. M. Fleischmann, P. J. Hendra and A. J. McQuillan, Chemical Physics Letters, 1974, 26, 163-166. 284. L. Zhang, J. J. Xu, L. Mi, H. Gong, S. Y. Jiang and Q. M. Yu, Biosens. Bioelectron., 2012, 31, 130-136. 285. L. Zhang, F. Q. Yu, A. J. Cole, B. Chertok, A. E. David, J. K. Wang and V. C. Yang, Aaps Journal, 2009, 11, 693-699. 286. B. Guven, N. Basaran-Akgul, E. Temur, U. Tamer and I. H. Boyaci, Analyst, 2011, 136, 740-748. 287. P. Drake, P. S. Jiang, H. W. Chang, S. C. Su, J. Tanha, L. L. Tay, P. L. Chen and Y. J. Lin, Analytical Methods, 2013, 5, 4152-4158. 288. Y. L. Wang, S. Ravindranath and J. Irudayaraj, Analytical and Bioanalytical Chemistry, 2011, 399, 1271-1278. 64 289. U. Tamer, I. H. Boyaci, E. Temur, A. Zengin, I. Dincer and Y. Elerman, Journal of Nanoparticle Research, 2011, 13, 3167- 3176. 290. S. G. Grancharov, H. Zeng, S. Sun, S. X. Wang, S. O'Brien, C. Murray, J. Kirtley and G. Held, The Journal of Physical Chemistry B, 2005, 109, 13030-13035. 291. A. Sandhu, Y. Kumagai, A. Lapicki, S. Sakamoto, M. Abe and H. Handa, Biosensors and Bioelectronics, 2007, 22, 2115- 2120. 292. W. Lu, Y. Chen, Z. Liu, W. Tang, Q. Feng, J. Sun and X. Jiang, Acs Nano, 2016, 10, 6685-6692. 293. K. Yang, Y. Hu and N. Dong, Biosensors and Bioelectronics, 2016, 80, 373-377. 294. A. Senyei, K. Widder and G. Czerlinski, Journal of Applied Physics, 1978, 49, 3578-3583. 295. K. J. Widder, A. E. Senyei and D. G. Scarpelli, Proceedings of the Society for Experimental Biology and Medicine, 1978, 158, 141-146. 296. J. Dobson, Drug development research, 2006, 67, 55-60. 297. J. Dobson, Gene therapy, 2006, 13, 283. 298. Y. Chen, K. Ai, J. Liu, G. Sun, Q. Yin and L. Lu, Biomaterials, 2015, 60, 111-120. 299. G. Unsoy, R. Khodadust, S. Yalcin, P. Mutlu and U. Gunduz, European Journal of Pharmaceutical Sciences, 2014, 62, 243- 250. 300. S. Laurent, A. A. Saei, S. Behzadi, A. Panahifar and M. Mahmoudi, Expert opinion on drug delivery, 2014, 11, 1449-1470. 301. L. H. Reddy, J. L. Arias, J. Nicolas and P. Couvreur, Chemical reviews, 2012, 112, 5818-5878. 302. S. C. McBain, H. H. Yiu and J. Dobson, International journal of nanomedicine, 2008, 3, 169. 303. Q. A. Pankhurst, J. Connolly, S. Jones and J. Dobson, Journal of physics D: Applied physics, 2003, 36, R167. 304. T. Kubo, T. Sugita, S. Shimose, Y. Nitta, Y. Ikuta and T. Murakami, International journal of oncology, 2000, 17, 309-324. 305. B. B. Yellen, Z. G. Forbes, D. S. Halverson, G. Fridman, K. A. Barbee, M. Chorny, R. Levy and G. Friedman, Journal of Magnetism and Magnetic Materials, 2005, 293, 647-654. 306. E. Lueshen, I. Venugopal, T. Soni, A. Alaraj and A. Linninger, Journal of biomedical nanotechnology, 2015, 11, 253-261. 307. W. He, Y.-L. Ji, C. Luo and Z. Xu, Research Journal of Applied Sciences, Engineering and Technology, 2014, 7, 3022-3029. 308. A. J. Rosengart, M. D. Kaminski, H. Chen, P. L. Caviness, A. D. Ebner and J. A. Ritter, Journal of magnetism and magnetic materials, 2005, 293, 633-638. 309. M. O. Avilés, A. D. Ebner and J. A. Ritter, Journal of Magnetism and Magnetic Materials, 2009, 321, 1586-1590. 310. A. Nacev, I. Weinberg, P. Stepanov, S. Kupfer, L. Mair, M. Urdaneta, M. Shimoji, S. Fricke and B. Shapiro, Nano letters, 2014, 15, 359-364. 311. J. H. Park, G. Saravanakumar, K. Kim and I. C. Kwon, Advanced drug delivery reviews, 2010, 62, 28-41. 312. J. Zhang, Z.-F. Yuan, Y. Wang, W.-H. Chen, G.-F. Luo, S.-X. Cheng, R.-X. Zhuo and X.-Z. Zhang, Journal of the American Chemical Society, 2013, 135, 5068-5073. 313. A. R. Chowdhuri, D. Bhattacharya and S. K. Sahu, Dalton Transactions, 2016, 45, 2963-2973. 314. P. J. Kempen, S. Greasley, K. A. Parker, J. L. Campbell, H.-Y. Chang, J. R. Jones, R. Sinclair, S. S. Gambhir and J. V. Jokerst, Theranostics, 2015, 5, 631. 315. T. Yadavalli, S. Ramasamy, G. Chandrasekaran, I. Michael, H. A. Therese and R. Chennakesavulu, Journal of magnetism and Magnetic Materials, 2015, 380, 315-320. 316. F. Ye, Å. Barrefelt, H. Asem, M. Abedi-Valugerdi, I. El-Serafi, M. Saghafian, K. Abu-Salah, S. Alrokayan, M. Muhammed and M. Hassan, Biomaterials, 2014, 35, 3885-3894. 317. A. Hervault, A. E. Dunn, M. Lim, C. Boyer, D. Mott, S. Maenosono and N. T. Thanh, Nanoscale, 2016, 8, 12152-12161. 65 318. X. Yao, X. Niu, K. Ma, P. Huang, J. Grothe, S. Kaskel and Y. Zhu, Small, 2017, 13. 319. G. Wang, Y. Ma, Z. Wei and M. Qi, Chemical Engineering Journal, 2016, 289, 150-160. 320. M. Bañobre-López, A. Teijeiro and J. Rivas, Reports of Practical Oncology & Radiotherapy, 2013, 18, 397-400. 321. A. E. Deatsch and B. A. Evans, Journal of Magnetism and Magnetic Materials, 2014, 354, 163-172. 322. C. Haase and U. Nowak, Physical Review B, 2012, 85, 045435. 323. R. R. Shah, T. P. Davis, A. L. Glover, D. E. Nikles and C. S. Brazel, Journal of magnetism and magnetic materials, 2015, 387, 96-106. 324. W. Andra and H. Nowak, Magnetism in medicine: a handbook, John Wiley & Sons, 2007. 325. A. Aharoni, Introduction to the Theory of Ferromagnetism, Clarendon Press, 2000. 326. R. Hergt, W. Andra, C. G. d'Ambly, I. Hilger, W. A. Kaiser, U. Richter and H.-G. Schmidt, IEEE Transactions on Magnetics, 1998, 34, 3745-3754. 327. R. Hergt, S. Dutz, R. Müller and M. Zeisberger, Journal of Physics: Condensed Matter, 2006, 18, S2919. 328. S. Ruta, R. Chantrell and O. Hovorka, Scientific reports, 2015, 5, 9090. 329. C. Martinez-Boubeta, K. Simeonidis, A. Makridis, M. Angelakeris, O. Iglesias, P. Guardia, A. Cabot, L. Yedra, S. Estradé and F. Peiró, Scientific reports, 2013, 3, 1652. 330. D. Serantes, K. Simeonidis, M. Angelakeris, O. Chubykalo-Fesenko, M. Marciello, M. D. P. Morales, D. Baldomir and C. Martinez-Boubeta, The Journal of Physical Chemistry C, 2014, 118, 5927-5934. 331. E. L. Verde, G. T. Landi, J. d. A. Gomes, M. H. Sousa and A. F. Bakuzis, Journal of Applied Physics, 2012, 111, 123902. 332. B. Mehdaoui, A. Meffre, J. Carrey, S. Lachaize, L. M. Lacroix, M. Gougeon, B. Chaudret and M. Respaud, Advanced Functional Materials, 2011, 21, 4573-4581. 333. G. Vallejo-Fernandez and K. O'Grady, Applied Physics Letters, 2013, 103, 142417. 334. H. Sohn and R. Victora, Journal of Applied Physics, 2010, 107, 09B312. 335. J. Carrey, B. Mehdaoui and M. Respaud, Journal of Applied Physics, 2011, 109, 083921. 336. D. B. Reeves and J. B. Weaver, Applied physics letters, 2014, 104, 102403. 337. C. Martinez‐Boubeta, K. Simeonidis, D. Serantes, I. Conde‐Leborán, I. Kazakis, G. Stefanou, L. Peña, R. Galceran, L. Balcells and C. Monty, Advanced Functional Materials, 2012, 22, 3737-3744. 338. D. Serantes, D. Baldomir, C. Martinez-Boubeta, K. Simeonidis, M. Angelakeris, E. Natividad, M. Castro, A. Mediano, D.-X. Chen and A. Sanchez, Journal of Applied Physics, 2010, 108, 073918. 339. G. T. Landi, Physical Review B, 2014, 89, 014403. 340. É. Kalmár, K. Ueno, P. Forgó, G. Szakonyi and G. Dombi, Journal of pharmaceutical and biomedical analysis, 2013, 83, 149-156. 341. Y. Zhang, R. Liu, Y. Hu and G. Li, Analytical Chemistry, 2009, 81, 967-976. 342. J. Pan, Y. Huang, L. Liu, Y. Hu and G. Li, Journal of Chromatography A, 2013, 1316, 29-36. 343. J. R. Wiśniewski, D. F. Zielinska and M. Mann, Analytical biochemistry, 2011, 410, 307-309. 344. S. D. Leskinen, M. Brownell, D. V. Lim and V. J. Harwood, Applied and environmental microbiology, 2010, 76, 4116-4117. 345. X. Xu, R. A. Sherry, S. Niu, D. Li and Y. Luo, Global change biology, 2013, 19, 2753-2764. 346. S. Unger and M. Jongen, in Progress in botany, Springer, 2015, pp. 347-393. 347. C. Zhou, Z. Du, G. Li, Y. Zhang and Z. Cai, Analyst, 2013, 138, 5783-5790. 348. G. K. Scriba and F. Belal, Chromatographia, 2015, 78, 947-970. 349. L. Dong-Mei, S. Yan-Ping and C. Juan, Chinese Journal of Analytical Chemistry, 2015, 43, 775-782. 350. M. Cheng and Z. Chen, Electrophoresis, 2017, 38, 486-493. 66 351. X. Hu, J. Pan, Y. Hu, Y. Huo and G. Li, Journal of Chromatography A, 2008, 1188, 97-107. 352. X. Zhang, J. Wang, R. Li, Q. Dai, R. Gao, Q. Liu and M. Zhang, Industrial & Engineering Chemistry Research, 2013, 52, 10152-10159. 353. K. Hoshino, Y.-Y. Huang, N. Lane, M. Huebschman, J. W. Uhr, E. P. Frenkel and X. Zhang, Lab on a Chip, 2011, 11, 3449- 3457. 354. C. W. Yung, J. Fiering, A. J. Mueller and D. E. Ingber, Lab on a Chip, 2009, 9, 1171-1177. 355. A. J. Mach and D. Di Carlo, Biotechnology and bioengineering, 2010, 107, 302-311. 356. R. S. Molday and D. Mackenzie, Journal of immunological methods, 1982, 52, 353-367. 357. I. K. Herrmann, M. Urner, F. M. Koehler, M. Hasler, B. Roth‐Z'Graggen, R. N. Grass, U. Ziegler, B. Beck‐Schimmer and W. J. Stark, Small, 2010, 6, 1388-1392. 67
1905.10932	1	1905	2019-05-27T01:47:14	Nanoscale spatially resolved mapping of uranium enrichment in actinide-bearing materials	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Spatially resolved analysis of uranium isotopes in small volumes of actinide-bearing materials is critical for a variety of technical disciplines, including earth and planetary sciences, environmental monitoring, bioremediation, and the nuclear fuel cycle. However, achieving sub-nanometer scale spatial resolution for such isotopic analysis is currently a challenge. By using atom probe tomography, a three dimensional nanoscale characterization technique, we demonstrate unprecidented nanoscale mapping of uranium isotopic enrichment with high sensitivity across various microstructural interfaces within small volumes (100 nm3) of depleted and low enriched uranium alloyed with 10 wt % molybdenum with different nominal enrichments of 0.20 and 19.75% 235U respectively. The approach presented here can be applied to study nanoscale variations of isotopic abundances in the broad class of actinide-bearing materials, providing unique insights into their origin and thermo-mechanical processing routes.	physics.app-ph	physics	Nanoscale spatially resolved mapping of uranium enrichment in actinide-bearing materials Elizabeth Kautza, Douglas Burkesa, Vineet Joshib, Curt Lavenderb, Arun Devarajc* aNational Security Directorate, Pacific Northwest National Laboratory, 902 Battelle Boulevard, P.O. Box 999, Richland, WA 99352, United States bEnergy and Environment Directorate, Pacific Northwest National Laboratory, 902 Battelle Boulevard, P.O. Box 999, Richland, WA 99352, United States cPhysical and Computational Sciences Directorate, Pacific Northwest National Laboratory, 902 Battelle Boulevard, P.O. Box 999, Richland, WA 99352, United States *corresponding author: [email protected] ABSTRACT Spatially resolved analysis of uranium isotopes in small volumes of actinide-bearing materials is critical for a variety of technical disciplines, including earth and planetary sciences, environmental monitoring, bioremediation, and the nuclear fuel cycle. However, achieving sub-nanometer scale spatial resolution for such isotopic analysis is currently a challenge. By using atom probe tomography, a three dimensional nanoscale characterization technique, we demonstrate unprecidented nanoscale mapping of uranium isotopic enrichment with high sensitivity across various microstructural interfaces within small volumes (100 nm3) of depleted and low enriched uranium alloyed with 10 wt % molybdenum with different nominal enrichments of 0.20 and 19.75% 235U respectively. The approach presented here can be applied to study nanoscale variations of isotopic abundances in the broad class of actinide-bearing materials, providing unique insights into their origin and thermo-mechanical processing routes. Key words: uranium enrichment, atom probe tomography, low-enriched uranium, nuclear fuel, U- Mo, isotopic abundance, actinide-bearing material 1 Uranium (U) is the heaviest element naturally occurring in the Earth's crust in significant amounts, and is used both in its natural, processed, and anthropogenic forms. U isotopes are central to a diverse set of scientific disciplines, most notably earth and planetary sciences [1-8], toxicology, environmental monitoring and bioremediation [9-11], and the nuclear fuel cycle, forensics, and safeguards [12-22]. Specifically in nuclear fuel cycle applications, the amount of the fissionable U isotope (235U) relative to all U (defined as enrichment) is a critical parameter in fuel performance, and thus impacts economic viability of nuclear power. Increasing U enrichment beyond natural abundance of 0.7% is necessary in order to allow for a self-sustaining fission chain reaction to proceed in a nuclear fuel [23]. The distribution of 235U in a nuclear fuel before irradiation can directly influence the nucleation and distribution of fission products and fission gas bubbles leading to impact on fuel swelling kinetics [24]. Fuel swelling in turn impacts geometric stability (e.g. bowing, deflection), robustness of fuel plates under irradiation, impacting phenomenon such as coolant channel closure, or release or fission gas products to the coolant [25]. These phenomena make it critical to analyze distribution of 235U in a nuclear fuel at a high spatial resolution to account for 235U enrichment variation across all possible nanoscale heterogenities in the microstructure. The small volume (nanometers to micrometers in diameter) of precipitates or interfacial regions that must be analyzed limits the ability of many bulk analysis techniques currently used for quantifying 235U enrichment, introducing a crucial technological and resultant knowledge gap. Additionally, the risk associated with handling radioactive materials is high, and analysis of small volumes is needed to minimize risk to researcher health and the spread of radioactive contamination. Our work aims to address this gap, while also demonstrating a methodology by which 235U isotopic abundances in actinide-bearing materials can be measured quantitatively with sub-nanometer scale spatial resolution to gain uniquely powerful insight into material radioactivity, origin, or processing history. As a part of U.S. high performance research reactor conversion program, there is an effort to replace all the highly enriched uranium fuel currently used in research reactors to low enriched nuclear fuels. Development of such a LEU fuel has significant implications to international nuclear non-proliferation, safeguards, and health and environmental contamination risks associated with continued handling of highly enriched uranium (HEU) fuels [18, 25-30]. Metallic low enriched fuel made of Uranium-10wt% Molybdenum alloys (LEU-10Mo) with less than 20% enrichment is a leading candidate chosen for this effort. As a part of fuel qualification and testing efforts, multiple batches of various LEU fuel plates (e.g. U-Mo, U-Si, U-Al) have been fabricated and irradiated in research reactors and several studies examined the nuclear fuel microstructure after neutron irradiation to improve understanding of material behavior in a reactor environment [25, 31, 32]. Microstructural features observable in fuels after irradiation via optical and electron microscopy techniques range from large fission gas bubbles to inclusions and elemental segregation. To develop a comprehensive atomic-level understanding of fuel performance in reactor, it is critical to understand the distribution of the fissionable 235U isotope in the starting fuel microstructure, at a nanoscale spatial resolution to determine the isotopic distribution in different phases and across interfaces in the microstructure. Measurement of U enrichment in small volumes of actinide-bearing materials at nanoscale spatial resolution is also crucial to a variety of other fields. For example, environmental monitoring and bioremediation efforts in the wake of nuclear accidents (such as Fukushima Daiichi, Three Mile Island, or Chernobyl) and detonation of nuclear weapons (for testing or in war time) crucially depends on detection and mapping of actinide speciea. As part of the environmental monitoring process, measurement of U concentration in geological materials (soil, sandstone ores, and other 2 organics materials) or particulate matter (produced from a detonation event) is performed. The measurement of U isotopic abundances can allow for improved understanding of atomic-scale transport of radionuclides in natural materials, the origin of the particles (including age and processing history), how radioactive particles spread, and how harmful they are to the surrounding environment and communities [9, 10, 22]. Another scientific area where U isotopic enrichment measurement is critical is in regulating the international transportation of actinide-bearing materials. Isotopic measurements in nuclear materials and comparison to existing models is necessary for understanding sample history, and ensuring safe handling of nuclear materials [14, 21, 22]. High spatial resolution analysis of U enrichment of small volumes can critically impact these wide variety of scientific areas. Current capabilities for measurement of U isotope abundances typically involve mass spectrometry or spectroscopy methods, including the following specific techniques: inductively coupled plasma mass spectrometry (ICP-MS) [33], time of flight and nano secondary ion mass spectrometry (SIMS) [34, 35], thermal ionization mass spectrometry (TIMS) [33], gamma spectroscopy [36], laser induced breakdown spectroscopy (LIBS), laser absorption spectroscopy (LAS), and laser induced fluorescence spectroscopy (LIFS) [37]. However, these techniques are not capable of providingsub-nanometer scale spatial resolution required to probe fine-scale microstructural features. Atom probe tomography (APT) is a 3D nanoscale compositional characterization method uniquely suited to analyze both composition and isotopic abundances in various material systems, with highspatial resolution [38-40]. Our work shows the capability to obtain nanoscale, spatially resolved quantitative 234,235,238U isotopic mapping, which has not been demonstrated before in anthropogenic U-bearing materials. We selected metallic U-10Mo nuclear fuels with two different enrichment values for demonstrating the capability of sub-nanometer scale spatially resolved, quantiative analysis of U enrichment [41-43]. Results 3D elemental distribution across precipitate-matrix interfaces In both depleted U-10Mo (DU-10Mo) and LEU-10Mo alloys, the main microstructural features are uranium carbide (UC) inclusions and the surrounding γ-UMo matrix [41]. Representative images of both DU-10Mo and LEU-10Mo microstructures are shown in Figure 1 (with additional micrographs provided in Supplementary Information, Figure S1, which provide more examples of UC morphologies). The UC phase is distributed throughout the γ-UMo matrix in both alloys, and was found to have two distinct morphologies in LEU-10Mo: (1) fine (approximately 500 nm in width), and (2) coarse (approximately 3-5 µm in width). Both UC morphologies, in addition to the γ-UMo matrix phase, were analyzed via APT in order to determine U enrichment and elemental distribution in both phases, and across γ-UMo matrix/UC interfaces. 3 Figure 1: Microstructure depleted and low enriched U-10Mo alloys: Back scattered SEM images of microstructural features in the U-Mo nuclear fuel analyzed via APT in (a) DU-10Mo and (b) LEU-10Mo. Two distinct carbide morphologies are visible in these micrographs: coarse and fine. The scale bar in each micrograph is 20 µm in length. 3D distributions of major alloying elements (U, Mo) and impurity elements (Si, C) from DU-10Mo and LEU-10Mo alloys are given in Figures 2 and 3 respectively with corresponding mass spectra for UC and γ-UMo phases. Mass spectra were normalized to the peak with the maximum number of counts, which corresponds to the 238U3+ peak in all data sets collected as part of this work. Normalized mass spectra were generated using manual ranging criteria. For each phase, 238U3+ peak is shown in the portion of the mass spectra between mass-to-charge-state ratios (Daltons or Da) of 76 and 84. For DU-10Mo, 235U3+ and 238U3+ peaks were detected, and in the LEU-10Mo alloy, the 234U3+ peak was also visible in the spectra, where 234U is known to be a decay product of 238U. In depleted U, 234U accounts for only 0.001% of all U isotopes (10 parts per million or ppm) which is approaching the detection limit for APT [38]. These results highlight the capability of APT in resolving each U isotopic peak individually which also can be spatialy resolved in the 3D reconstruction. Figure 2: APT results of depleted U-10Mo alloy: 3D element distribution maps of U isotopes (235,238U), Mo, C, and Si across a DU-10Mo/UC interface (indicated by the black arrow). Full mass spectra (0-150 Da) for matrix and UC phases are also provided, with the specific region of the mass spectra between 76-84 Da also shown to provide additional detail on the U3+ peaks analyzed. 4 Figure 3: APT results of low enriched U-10Mo alloy: 3D element distribution maps of U isotopes (234,235,238U), Mo, C, and Si across a LEU-10Mo/UC interface (indicated by the black arrow). Full mass spectra (0-150 Da) for matrix and UC phases are also provided, with the specific region of the mass spectra between 76-84 Da also shown to provide additional detail on the U3+ peaks analyzed. U and Mo were observed as the major species in DU-10Mo and LEU-10Mo samples, where 235U and 238U were found to field evaporate as doubly (U2+) and triply (U3+) charged ions. Complex ions composed of U and H (UH, or uranium hydride) were also present in mass spectra of both alloys. UH peaks were detected at approximately 118.18 and 119.66 Da (corresponding to the U2+ charge state for 235U and 238U respecively) in LEU-10Mo. In DU-10Mo, however, only the peak at 119.66 Da was observed. The mass spectra for U2+ for both alloys analyzed are provided in Supplementary Information, Figure S2, and show all detected UH peaks. The change in concentration of U and Mo, and other impurity elements C and Si were quantified using a one dimensional (1D) concentration profile across the γ-UMo matrix/UC interface (Figure 4). Peak deconvolution was performed to determine total U, Mo, C, and Si contributions from both elemental and complex peaks present in the mass-to-charge spectrum, a process further detailed in the Materials and Methods section. U and Mo concentrations are consistent with alloy composition of U alloyed with 10 wt% Mo (approximately 22 at% Mo) for the γ-UMo matrix phase. In the γ-UMo matrix, C and Si impurity elements were also detected at concentrations of less than 3 at%. UC composition measured by APT ranges from 51-54 at% U and 35-39 at% C. In the UC phase, concentration of impurity elements (Si,H,O,Ni,Al) account for the balance. Analysis of impurity elements including Si, H, O, Ni, Al was performed, and the concentration of these elements was plotted across the matrix/carbide interface and are reported in Supplementary Information, Figure S3. Concentration of Si is less than 3 at%, O, Ni, and Al are all less than 2 at% in samples analyzed, whereas H concentration is more significant, and was found in concentrations of 4-20 at%. The H concentration can be attributed to the formation of UH when U is in the presence of H [17]. Sources of H leading to the formation of UH could be water used in the bulk metallographic polishing procedure, FIB prepareation of APT needle specimens and/or from residual H background from the APT analysis chamber. 5 Figure 4: Compositional and isotopic profiles of DUMo and LEUMo. (a) 3D element distribution map that contains a γ-UMo/UC interface in a DU-10Mo alloy, (b) corresponding compositional profile across the matrix/UC interface, (c) map of 235U% relative to all U isotopes (enrichment) across the DU-10Mo γ-UMo /UC interface. Sub-figures (d), (e), and (f) present the same type of data provided in (a),(b), and (c), but for a LEU-10Mo alloy. Quantification of U enrichment of individual phases U isotope abundances in UC (fine and coarse morphologies) and γ-UMo phases were quantified from APT data, and the results are summarized in Table 1. U isotopic enrichment was estimated based on the following equation: 𝑈𝑒𝑛𝑟𝑖𝑐ℎ𝑚𝑒𝑛𝑡 = 𝑈𝑖% 𝑈𝑡𝑜𝑡𝑎𝑙% Where 𝑈𝑖% correspond to the atom % of a particular U isotope and 𝑈𝑡𝑜𝑡𝑎𝑙% corresponds to the total uranium atom % estimated from APT results. Sample volumes from which U isotopic abundances were calculated ranged from 8,000-250,000 nm3, which corresponds to volumes with dimensions of approximately 20 nm x 20 nm x 20 nm to 50 nm × 50 nm × 100 nm. U ion counts from the U3+ elemental peaks accounted for the majority of U ions collected, thus isotopic ratios were calculated using this peak count. Percentage of U elemental and complex species are reported in Supplementary Information, Table S1, and a comparison of calculated U enrichment from different charge state peaks are provided in Table S2. From the APT results we didn't observe any U isotopic bias when it came to preferentially evaporating as as molecular ions. Thus calculating isotopic abundances based on the major elemental peak (U3+) provided the most consistent and accurate approach for analysis of isotopic abundances. 6 Table 1: U isotopic abundances measured by APT: Tabulated results are from different phases from both depleted and low-enriched U-Mo alloys. All data in this table is reported as percent U isotope relative to all U with percent error. Error was calculated by propagating point counting error, and is subsequently detailed in the Materials and Methods section. Alloy Phase DU-10Mo γ-UMo matrix UC - coarse γ-UMo matrix LEU-10Mo UC - coarse UC - fine % 235U 0.19 ± 0.001 0.19 ± 0.005 0.21 ± 0.017 0.20 ± 0.016 0.21 ± 0.005 19.47 ± 0.185 19.26 ± 0.078 20.44 ± 0.051 20.66 ± 0.029 20.17 ± 0.034 20.80 ± 0.544 20.00 ± 0.042 20.04 ± 0.051 % 238U % 234U 99.81 ± 0.060 99.81 ± 0.208 99.79 ± 0.676 99.79 ± 0.673 99.79 ± 0.210 80.34 ± 0.527 80.51 ± 0.222 79.34 ± 0.138 79.11 ± 0.078 79.61 ± 0.093 78.97 ± 1.623 79.79 ± 1.075 79.74 ± 0.141 0.0003 ± 0.0001 0.0020 ± 0.0005 0.0007 ± 0.0010 0.0080 ± 0.0033 0.0007 ± 0.0003 0.189 ± 0.015 0.226 ± 0.007 0.218 ± 0.004 0.233 ± 0.003 0.219 ± 0.003 0.229 ± 0.053 0.211 ± 0.003 0.226 ± 0.005 Analysis of U enrichment across a -UMo grain boundary Grain boundaries in metallic nuclear fuels are another critical interface with significant role to play when it comes to interacting with irradiation induced defects. Grain boundaries directly influence the radiation damage recovery and defect accumulation, such as fission gas bubble nucleation and growth in the irradiated fuel plate [25]. Hence as a proof-of-principle demonstration, APT was additionally used to analyze variation of isotopic enrichment across the -UMo grain boundary in DU-10Mo alloy. The APT reconstruction across a -UMo grain boundary in DU-10Mo is given in Figure 5 along with the compositional profile showing concentration of major alloying elements (U, Mo), impurity elements (Si, C, Ni, Al, Mn), and 235U enrichment. It can be seen that there is clear solute segregation along the grain boundary, consistent with our previous work [41]. Interestingly there is a narrow band of increase in the 235U enrichment specifically at the grain boundary (highlighted by the dashed black line in figure 3(c)) where locally at one spot the enrichment increased to as high as 0.5, while the overall enrichment in either grain remained consistent with the nominal value of 0.2% 235U . This highlights the unique capability of APT in not only accurately probing the compositional segregation across grain boundaries in metallic nuclear fuel but also its capability in obtaining accurate undertanding of local variation in 235U enrichment across grain boundaries. 7 Figure 5: Compositional and isotopic profiles of a DUMo grain boundary, where (a) provides element distribution maps for major alloying and impurity elements, (b) corresponding quantitative analysis of composition across the grain boundary is plotted (concentration in at% versus distance), via a 1D concentration profile, and (c) 235U enrichment across the grain boundary. Discussion Characterization of UC and γ-UMo matrix phases in DU and LEU alloyed with Mo was performed to demonstrate the capability of APT for quantitative measurement of U isotopes in actinide-bearing materials with varying nominal enrichments. The purpose of this effort was to gain insight into the homogeneity of 235U across different microstructural features, which has implications in fuel swelling and thus stability of nuclear fuel in a reactor operating environment. Calculated 235U isotopic abundances for both DU-10Mo and LEU-10Mo (summarized in Table 1) indicate that U isotopic abundances are comparable between matrix and carbide phases, 8 with minor fluctuations below 1% of U enrichment reported. This result suggests that fuel fabrication processes generate a microstructure in which 235U is relatively homogenously distributed between carbide and matrix phases. Our results indicate that the analyzed carbides likely formed during the melting and casting processes and are not retained from HEU or DU feedstock materials which are melted together and casted to form the LEU-10Mo alloy. The work presented here demonstrates that it is possible to probe U isotope abundances in fine microstructural features, and across interfaces in nuclear fuels to establish improved microstructure-processing relationships. Our results allowed us to gain insight into the sample processing history, which is a concept that could now be applied to forensics and environmental remediation studies. Additionally, the result that 235U is nominally homogenously distributed at the nanoscale between the γ-UMo matrix and UC phases can simplify fuel performance modelling since a homogenous isotopic enrichment can be assigned to describe the fuel performance in reactor. Impurity elements (C, Si) were measured via APT, and some local increase in Si concentration was observed along the UC/ γ-UMo interface in DU-10Mo. A higher concentration of Si was observed in the UC phase of the LEU-10Mo alloy. From the U isotope abundances given in Table 1, it is clear that APT can measure the concentration of not only major isotopes of 235U and 238U, but also 234U (an indirect decay product of 238U). The APT reconstructions shown in Figures 2 and 3 from DU-10Mo and LEU-10Mo, respectively, reveal the ability to spatially resolve the isotope enrichment up to a spatial resolution of 0.2 nm in x, y and z directions. APT is also uniquely suited for analysis of enrichment across grain boundaries in nuclear fuels, which is a challenging task for most of the bulk isotopic analysis methods routinely used. Given this demonstration, our approach can now be utilized to analyze small particulates from nuclear accidents or at different stages of the nuclear fuel cycle to help analyze U enrichment and identify material origin, age, or processing history. This study opens up opportunities to utilize APT as a method of choice for a variety of other fields ranging from earth and planetary sciences to bioremediation, toxicology and environmental monitoring. In summary, by using site specific sample preparation and laser assisted APT analysis, we demonstrate the unique capability to perform spatially resolved nanoscale mapping of isotopic enrichment and impurity elements in U-Mo fuels with two different 235U enrichments. Based on these results, we believe APT has a strong prospect for use in a wide range of characterization efforts of actinide-bearing materials in which isotopic analysis in small volumes (and of specific microstructural features) is critical. Our work has implications for multiple disciples that could benefit from precise measurement of U isotope enrichment and fractionation to better understand sample origin or processing history. Methods and experimental protocols using APT presented here translate well to analysis of U enrichment in many other systems (geologic materials, metallic fuels, and glass fabricated as a nuclear waste form, for example). Thus, APT is promising as an impactful and versatile tool for measurement of isotopic abundances in actinide-bearing materials, at length scales previously unexplored by other more widely used mass spectrometry and spectroscopy methods. Materials and Methods Material Fabrication DU-10Mo and LEU-10Mo fuels analyzed in this work were fabricated at the Y12 National Security Complex at Oak Ridge National Laboratory. DU-10Mo samples were fabricated by melting DU with Mo and casting in a graphite mold. The cast DU-10Mo was then homogenized 9 at 900 C for 48 hours in an inert gas (argon) atmosphere. After the homogenization annealing treatment, samples were cooled by turning off the furnace and forcing argon gas into the chamber to assist in sample cooling. A cooling rate of 25 C per minute was achieved from 900 C to 650 C, then 3 C per minute from 500 C to 350 C. LEU-10Mo samples were homogenized at 900 C for 144 hours in the same atmosphere and furnace as described previously. The LEU-10Mo was subjected to subsequent thermo-mechanical processing treatments (hot and cold rolling) to form the final fuel foil. Additional sample fabrication details are reported in References [47, 19, 17] FIB-based sample preparation and APT analysis APT needles were prepared from standard metallographicaly polished bulk samples [32]. Site-specific FIB lift-outs and annular milling was performed according to methods described in References [8]. A CAMECA LEAP 4000X HR system equipped with a 355 nm wavelength UV laser was used for all APT data collection with the following user-selected parameters: 100 pJ laser energy, 100 kHz pulse frequency, 45 K specimen temperature, and 0.005 atoms/pulse detection rate. Data sets analyzed and reported here ranged in size between 1.4 and 34.5 million ions, and either contain a single phase (UC or γ-UMo matrix) or capture a UC/γ-UMo interface. All data sets were reconstructed and analyzed using the Interactive Visualization and Analysis Software (IVAS), version 3.8.2. For data sets containing a single phase, bulk composition analysis was completed, and for those data sets that contained an interface, composition in each phase was determined using spherical regions of interest (ROIs) with diameters of 20-30 nm. To examine composition across the interface, a cylindrical ROI with a diameter of 20 nm was positioned perpendicular to the planar interface, and a one dimensional (1D) composition profile along the z-axis with a 0.3 nm bin width was constructed. For all composition analysis, ionic and background corrected data was used. In order to quantify Mo and C concentrations accurately from raw data, peak deconvolution was performed according to methods presented in Reference [44]. Peak deconvolution was required to distinguish between 96Mo2+ and C4 species that both have a mass-to-charge state ratio of approximately 48 Da, and therefore cannot be ranged separately. The number of 96Mo2+ ions was estimated from the number of 98Mo2+ ions, assuming both 96Mo2+ and 98Mo2+are present in U- Mo at naturally abundant levels, similar to the process reported in Reference [40]. Equation 1 was used to determine the 96Mo2+ ion count: 0.1668 0.2413 𝑛 96𝑀𝑜2+ = 𝑛 98𝑀𝑜2+ × (1) where 𝑛 96𝑀𝑜2+ is the estimated ion count for the 96Mo2+, and 𝑛 98𝑀𝑜+2 is the ion count for the 98Mo2+ peak. The two fractions in Equation 1 (0.1668 and 0.2413) are the natural isotope abundances of 96Mo and 98Mo, respectively. Since the entire 96Mo2+ peak was ranged as a single species but not all counts correspond to 96Mo2+, the calculated value of 𝑛 96𝑀𝑜2+ was subtracted from the total ion counts corresponding to the ranged peak to obtain the ion count for the C4 species. In order to obtain an accurate estimation of C concentration in the volume analyzed, we used the 13C method detailed in Reference [45]. This approach involves multiplying the 13C background corrected ion count by 92.5 to determine a more accurate value for the 12C ion count. The 92.5 multiplier relies on the assumption that C exists in natural abundance levels. Peak deconvolution was performed via methods described above to analyze the total U, Mo, C, and Si concentration across the interface. Error reported in tables and composition profiles represent propagation of point counting error, defined as: 10 𝐸 = √𝐶𝑖(1−𝐶𝑖) 𝑁 (2) where 𝐶𝑖 is 𝑖 number of solute ions, 𝑁 is the total number of ions in a bin, and 𝑥𝑖 is the count for a specific element. 𝐶𝑖 is defined in mathematical form as: 𝐶𝑖 = (3) 𝑥𝑖 𝑁 Error reported as a percentage is simply the value calculated from Equation 2 multiplied by 100. Error associated with U enrichment was calculated with Equations 4-5, where 𝐸𝑇𝑜𝑡𝑎𝑙,𝑈 (Equation 4) is the total error associated with all U isotope counts, 𝑈𝑒𝑛𝑟𝑖𝑐ℎ𝑚𝑒𝑛𝑡 and 𝐸𝑈𝑒𝑛𝑟𝑖𝑐ℎ𝑚𝑒𝑛𝑡 (Equation 5) is the counting error associated with a U isotope enrichment value. 𝐸𝑇𝑜𝑡𝑎𝑙,𝑈 = √(𝐸 234𝑈)2 + (𝐸 235𝑈)2 + (𝐸 238𝑈)2 𝐸𝑈𝑒𝑛𝑟𝑖𝑐ℎ𝑚𝑒𝑛𝑡 = 𝑈𝑖 × √( 2 𝐸𝑈𝑖 𝑈𝑖 ) + ( 𝐸𝑈𝑡𝑜𝑡𝑎𝑙 𝑈𝑡𝑜𝑡𝑎𝑙,𝑐𝑜𝑛𝑐 2 ) (4) (5) ACKNOWLEDGEMENTS This work was supported by the U.S. Department of Energy (DOE), National Nuclear Security Administration under Contract DE-AC05-76RL01830. A portion of this research was performed using facilities at the Environmental Molecular Sciences Laboratory, a national scientific user facility sponsored by the DOE's Office of Biological and Environmental Research and located at Pacific Northwest National Laboratory (PNNL). The authors thank Mark Rhodes of PNNL for performing bulk metallographic sample preparation, and all the other staff directly or indirectly associated with producing the results featured in this publication. AUTHOR CONTRIBUTIONS A.D. conceptualized and directed all work presented here. E.K. and A. D. performed site-specific sample preparation for analysis of interfaces and conducted the APT experiments. E.K. and A.D. analyzed all APT data. C.L. and V.J. coordinated the fabrication and metallographic preparation of DU-10Mo and LEU-10Mo samples used in this work. D.B. contributed to manuscript writing, and discussions about results and its significance to in reactor performance of nuclear fuels. All authors contributed to the discussion of results and manuscript preparation. Additional Information Competing interests: The author(s) declare no competing interests. References [1] J. R. Bargar et al., "Uranium redox transition pathways in acetate-amended sediments," (in en), PNAS, vol. 110, no. 12, pp. 4506 -- 4511, March 2013 2013. G. A. Brennecka and M. Wadhwa, "Uranium isotope compositions of the basaltic angrite meteorites and the chronological implications for the early Solar System," (in en), PNAS, vol. 109, no. 24, pp. 9299 -- 9303, June 2012 2012. J. Hiess, D. J. Condon, N. McLean, and S. R. Noble, "238U/235U Systematics in Terrestrial Uranium-Bearing Minerals," Science, vol. 335, no. 6076, pp. 1610 -- 1614, 2012 2012. C. H. Stirling, "Keeping Time with Earth's Heaviest Element," Science, vol. 335, no. 6076, 11 [2] [3] [4] pp. 1585 -- 1586, 2012 2012. [5] M. Stylo et al., "Uranium isotopes fingerprint biotic reduction," (in en), PNAS, vol. 112, no. [6] [7] [8] [9] [10] [11] [12] [13] 18, pp. 5619 -- 5624, May 2015 2015. H.-D. Sues, "Dating the origin of dinosaurs," (in en), PNAS, vol. 113, no. 3, pp. 480 -- 481, January 2016 2016. G. A. Wagner et al., "Radiometric dating of the type-site for Homo heidelbergensis at Mauer, Germany," (in en), PNAS, vol. 107, no. 46, pp. 19726 -- 19730, November 2010 2010. R. E. Wood, C. Barroso-Ruíz, M. Caparrós, J. F. J. Pardo, B. G. Santos, and T. F. G. Higham, "Radiocarbon dating casts doubt on the late chronology of the Middle to Upper Palaeolithic transition in southern Iberia," (in en), PNAS, vol. 110, no. 8, pp. 2781 -- 2786, February 2013 2013. R. J. M. Konings, T. Wiss, and O. Beneš. (2015). Predicting material release during a nuclear reactor accident [Comments and Opinion]. X. Wang, Z. Xu, A. Soulami, X. Hu, C. Lavender, and V. Joshi, "Modeling Early-Stage Processes of U-10 Wt.%Mo Alloy Using Integrated Computational Materials Engineering Concepts," JOM, vol. 69, no. 12, pp. 2532 -- 2537, December 2017 2017. S. Takahashi, Ed. Radiation Monitoring and Dose Estimation of the Fukushima Nuclear Accident. Springer Nature, 2017. S. F. Boulyga et al., "Determination of 236U/238U isotope ratio in contaminated environmental samples using different ICP-MS instruments," Journal of Analytical ATomic Spectrometry, vol. 17, pp. 958-964, 2002 2002. J. N. Christensen, P. E. Dresel, M. E. Conrad, G. W. Patton, and D. J. DePaolo, "Isotopic Tracking of Hanford 300 Area Derived Uranium in the Columbia River," Environmental science & technology, vol. 44, no. 23, pp. 8855 -- 8862, 2010 2010. [15] [16] [14] A. J. Fahey, C. J. Zeissler, D. E. Newbury, J. Davis, and R. M. Lindstrom, "Postdetonation nuclear debris for attribution," Proceedings of the National Academy of Sciences, vol. 107, no. 47, pp. 20207 -- 20212, 2010 2010. S. K. Hanson et al., "Measurements of extinct fission products in nuclear bomb debris: Determination of the yield of the Trinity nuclear test 70 y later," (in en), PNAS, vol. 113, no. 29, pp. 8104 -- 8108, July 2016 2016. R. C. Marin, J. E. S. Sarkis, and M. R. L. Nascimento, "The use of LA-SF-ICP-MS for nuclear forensics purposes: uranium isotope ratio analysis," Journal of Radioanalytical and Nuclear Chemistry, vol. 295, no. 1, pp. 99 -- 104, January 2013 2013. T. L. Martin et al., "Atomic-scale Studies of Uranium Oxidation and Corrosion by Water Vapour," (in en), Scientific Reports, vol. 6, p. 25618, July 2016 2016. S. Van den berghe and P. Lemoine, "Review of 15 years of high-density low-enriched UMo dispersion fuel development for research reactors in Europe," Nuclear Engineering and Technology, vol. 46, no. 2, pp. 125 -- 146, April 2014 2014. J. R. Vavrek, B. S. Henderson, and A. Danagoulian, "Experimental demonstration of an isotope-sensitive warhead verification technique using nuclear resonance fluorescence," (in en), PNAS, vol. 115, no. 17, pp. 4363 -- 4368, April 2018 2018. J. D. Ward et al., "Identifying anthropogenic uranium compounds using soft X-ray near- edge absorption spectroscopy," Spectrochimica Acta Part B: Atomic Spectroscopy, vol. [18] [17] [19] [20] 12 127, pp. 20 -- 27, January 2017 2017. [22] [21] M. R. B. I. Andrew Conant, Anna Erickson, "Sensitivity and Uncertainty Analysis of Plutonium and Cesium Isotopes in Modeling of BR3 Reactor Spent Fuel," Nuclear Technology, vol. 197, no. 1, pp. 12-19, 2017 2017. E. Kuhn, D. Fischer, and M. Ryjinski, "Environmental sampling for IAEA safeguards: A five year review," International Atomic Energy Agency, Vienna, AustriaIAEA-SM -- 367, 2001 2001. L. J. H. James J. Duderstadt, Nuclear Reactor Analysis. Wiley, 1976. S. Hu, D. E. Burkes, C. A. Lavender, D. J. Senor, W. Setyawan, and Z. Xu, "Formation mechanism of gas bubble superlattice in UMo metal fuels: Phase-field modeling investigation," Journal of Nuclear Materials, vol. 479, pp. 202 - 215, 2016 2016. [23] [24] [27] [26] [25] M. K. Meyer et al., "Low-temperature irradiation behavior of uranium -- molybdenum alloy dispersion fuel," Journal of Nuclear Materials, vol. 304, no. 2, pp. 221 -- 236, August 2002 2002. Y. S. Kim and G. Hofman, "Fission product induced swelling of U-Mo alloy fuel," Journal of Nuclear Materials, vol. 419, pp. 291 -- 301, December 2011 2011. S. Neogy, A. Laik, M. T. Saify, S. K. Jha, D. Srivastava, and G. K. Dey, "Microstructural Evolution of the Interdiffusion Zone between U-9 Wt Pct Mo Fuel Alloy and Zr-1 Wt Pct Nb Cladding Alloy Upon Annealing," (in en), Metall and Mat Trans A, vol. 48, no. 6, pp. 2819 -- 2833, June 2017 2017. J. L. Snelgrove, G. L. Hofman, M. K. Meyer, C. L. Trybus, and T. C. Wiencek, "Development of very-high-density low-enriched-uranium fuels1Work supported by the US Department of Energy, Office of Nonproliferation and National Security, under contract No. W-31-109- ENG-38.1," Nuclear Engineering and Design, vol. 178, no. 1, pp. 119 -- 126, December 1997 1997. [28] [30] [29] M. Ugajin, A. Itoh, M. Akabori, N. Ooka, and Y. Nakakura, "Irradiation behavior of high uranium-density alloys in the plate fuels," Journal of Nuclear Materials, vol. 254, no. 1, pp. 78 - 83, March 1998 1998. Research Reactors: Safe Management and Effective Utilization (Proceedings Series - International Atomic Energy Agency). Vienna, Austria: IAEA, 2015. S. Van den Berghe, W. Van Renterghem, and A. Leenaers, "Transmission electron microscopy investigation of irradiated U -- 7wt%Mo dispersion fuel," Journal of Nuclear Materials, vol. 375, no. 3, pp. 340-346, 2008/04/30/ 2008. [31] [32] H. J. Ryu, Y. S. Kim, and G. L. Hofman, "Amorphization of the interaction products in U -- Mo/Al dispersion fuel during irradiation," Journal of Nuclear Materials, vol. 385, no. 3, pp. 623-628, 2009/04/15/ 2009. J. S. Becker, "Chapter 13 - Inorganic Mass Spectrometry of Radionuclides," in Handbook of Radioactivity Analysis (Third Edition), M. F. L'Annunziata, Ed. Amsterdam: Academic Press, 2012, pp. 833-870. [33] [34] D. Huang, X. Hua, G.-L. Xiu, Y.-J. Zheng, X.-Y. Yu, and Y.-T. Long, "Secondary ion mass spectrometry: The application in the analysis of atmospheric particulate matter," Analytica Chimica Acta, vol. 989, pp. 1-14, 2017/10/09/ 2017. [35] A. Krein et al., "Imaging Chemical Patches on Near-surface Atmospheric Dust Particles with NanoSIMS 50 to Identify Material Sources," Water, Air, & Soil Pollution: Focus, vol. 13 [36] [37] 8, no. 5, pp. 495-503, 2008/12/01 2008. C. T. Nguyen and J. Zsigrai, "Basic characterization of highly enriched uranium by gamma spectrometry," Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, vol. 246, no. 2, pp. 417-424, 2006/05/01/ 2006. S. S. Harilal, B. E. Brumfield, N. L. LaHaye, K. C. Hartig, and M. C. Phillips, "Optical spectroscopy of laser-produced plasmas for standoff isotopic analysis," Applied Physics Reviews, vol. 5, no. 2, p. 021301, 2018/06/01 2018. [38] A. Devaraj et al., "Three-dimensional nanoscale characterisation of materials by atom probe tomography," International Materials Reviews, vol. 63, no. 2, pp. 68 -- 101, February 2018 2018. S. Piazolo et al., "Deformation-induced trace element redistribution in zircon revealed using atom probe tomography," (in en), Nature Communications, vol. 7, p. 10490, February 2016 2016. L. White et al., Atomic-scale age resolution of planetary events. 2017. [40] [41] A. Devaraj et al., "Grain boundary engineering to control the discontinuous precipitation in multicomponent U10Mo alloy," Acta Materialia, vol. 151, pp. 181 -- 190, June 2018 2018. S. Jana et al., "Kinetics of cellular transformation and competing precipitation mechanisms during sub-eutectoid annealing of U10Mo alloys," Journal of Alloys and Compounds, vol. 723, pp. 757 -- 771, November 2017 2017. [39] [42] [43] A. Devaraj et al., "Phase transformation of metastable discontinuous precipitation products to equilibrium phases in U10Mo alloys," Scripta Materialia, vol. 156, pp. 70-74, 2018/11/01/ 2018. S. D. Taylor, J. Liu, B. W. Arey, D. K. Schreiber, D. E. Perea, and K. M. Rosso, "Resolving Iron(II) Sorption and Oxidative Growth on Hematite (001) Using Atom Probe Tomography," J. Phys. Chem. C, vol. 122, no. 7, pp. 3903 -- 3914, February 2018 2018. [44] [45] M. Thuvander et al., "Quantitative atom probe analysis of carbides," Ultramicroscopy, vol. 111, no. 6, pp. 604 -- 608, May 2011 2011. 14
1907.09999	1	1907	2019-05-02T11:08:57	Electrokinetic Power Harvesting from Wet Textile	[ "physics.app-ph" ]	Developing low-weight, frugal and sustainable power sources for resource-limited settings appears to be a challenging proposition for the advancement of next-generation sensing devices and beyond. Here, we report the use of centimeter-sized simple wet fabric pieces for electrical power generation, by deploying the interplay of a spontaneously induced ionic motion across fabric channels due to capillary action and simultaneous water evaporation by drawing thermal energy from the ambient. Unlike other reported devices with similar functionalities, our arrangement does not necessitate any input mechanical energy or complex topographical structures to be embedded in the substrate. A single device is capable of generating a sustainable open circuit potential up to 700 mV. This suffices establishing an inherent capability of functionalizing self-power electronic devices, and also to be potentially harnessed for enhanced power generation with feasible up-scaling.	physics.app-ph	physics	Electrokinetic Power Harvesting from Wet Textile Sankha Shuvra Das, Vinay Manaswi Pedireddi, Aditya Bandopadhyay, Partha Saha and Suman Chakraborty* Department of Mechanical Engineering, Indian Institute of Technology Kharagpur, Kharagpur 721302, INDIA. email: [email protected] Developing low-weight, frugal and sustainable power sources for resource-limited settings appears to be a challenging proposition for the advancement of next-generation sensing devices and beyond. Here, we report the use of centimeter-sized simple wet fabric pieces for electrical power generation, by deploying the interplay of a spontaneously induced ionic motion across fabric channels due to capillary action and simultaneous water evaporation by drawing thermal energy from the ambient. Unlike other reported devices with similar functionalities, our arrangement does not necessitate any input mechanical energy or complex topographical structures to be embedded in the substrate. A single device is capable of generating a sustainable open circuit potential up to ~700 mV. This suffices establishing an inherent capability of functionalizing self-power electronic devices, and also to be potentially harnessed for enhanced power generation with feasible up- scaling. Introduction In the milieu of global warming and energy crisis, frugal yet sustainable and renewable energy resources are critical to the advancement of human civilization1 -- 4. Generating sustainable and stable electricity from mundane and natural sources without necessitating mechanized inputs and sophisticated setups triggers wishful desires and compelling challenges, simultaneously5 -- 9. Natural drying of wet clothes is common in daily lives, which is a ubiquitous process ideally suited for harvesting thermal energy from the incipient ambient10 -- 12. During this drying process, one may potentially harness the motion of salt-water solution within the fabric matrices towards creating preferential charge segregation13, generating electrical power14 -- 18. By introducing fabrication of simply-designed channels on textile pieces for achieving guided transportation of saline water and drawing analogies of the later with evaporative transport across the parts of a living plant, here we demonstrate an extremely frugal, simple and viable approach of harvesting electrical power using cellulose-based fiber cloth. In sharp contrast to other reported devices deploying specially-structured and delicately-fabricated substrates towards achieving this feat1,8,19 -- 26, we deploy a regular cellulose-based wearable textile as a medium for ionic motion though the interlace fibrous network by capillary action, inducing an electric potential in the process. The device design inherently exploits a large transpiration surface for achieving a sustainable salt-ion-flux migration, through natural evaporation effect. Our approach essentially deploys the surface energy, an intrinsic property of the material, to convert into electricity, in this manner. Notably, upon solar irradiance, the device can effectively generate a significantly higher potential than that in the absence of sun-light. A single prototype fabric channel (FC) harvester (with stem width of 4 cm, stem length of 3 cm and leaf area of 7 cm by 7 cm) shows the ability to generate a maximum open circuit potential of ~700 mV. Featuring these functionalities, the device can be exploited to electrically functionalize paper- or fabric-based rapid kits for on-spot diagnostics of diseases in resource-limited settings. 1 Experimental details Chemicals Sodium chloride, NaCl (>99% purity) was purchased from Merk Life Science Pvt. Ltd. Fabric cloth (white color) of ~95% cotton was purchased from local garment shop. Copper sheet of ~300 µm thickness (>98%), plastic straw was purchased from local market. All the electrolyte solutions are prepared by dissolving the particular analyte at various concentrations in DI water (Millipore India, 18.2MΩ-cm). Device description and experimentation FCs are prepared from a piece of commercial wearable fiber cloth, where those are cut to the required channel design using scissors. FC has three segments such as root, stem and leaf (Fig. 1a) which is physically analogous to the transport mechanism in a leaf (Fig. 1b and Supplementary Fig. 1). Field Emission Scanning Electron Micros copy (FESEM) analysis shows the stitching pattern of the fibers with crisscross orientation having periodically alternating micropores, with fiber interspacing of ~5-6 µm (Fig. 2b). The electrodes made of commercial grade copper sheet (>98% purity), with dimensions ~2 cm × 1 cm and ~300 µm thickness, are embedded at either sides of the stem section (ensuring a proper ohmic contact between FC surface and electrode). In order to obtain a rising capillary motion, the channel is inserted vertically into an electrolyte (aqueous NaCl solution) filled beaker covering the root area of FC, and leaving the leaf area exposed to the atmosphere (Fig. 1b). The bigger leaf area promotes the evaporation of water while maintaining a continuous negative pressure gradient across the channel length. To limit the evaporative loss, the entire stem section is inserted into a plastic straw (diameter ~6 mm) with length equal to that of stem length. Prior to the experimentation, FC is cleaned with DI water followed by drying. a Leaf (downstream reservoir pad) Stem Root (upstream reservoir pad) b Broom stick Electrode Plastic straw Beaker filled with electrolyte Figure 1 Design of a fabric-based channel (FC). a, Photograph of a FC prepared from a commercial grade fiber cloth. b, Complete FC based electrokinetic power generator; demonstrating its various segments. 2 Evaporation through sun-light a Capillary migration of ions through cellulose pores c Leaf Broom stick Electrode Stem Beaker filled with electrolyte Root b Super-capacitor Figure 2 Capillarity-coupled-evaporation driven FC based electrokinetic power generator. a, Schematic representation of the experimental setup for measuring induced potential. b, FESEM image of FC showing the stitching pattern and crisscross orientation of fibers. c, Capillary transport of ions through individual pore, showing specific ion distribution inside the pore. 3 The schematic representation of the complete experimental setup used for capillarity-coupled-evaporation induced streaming potential measurement is illustrated in Fig. 2. In order to measure the induced electric potential and the corresponding short-circuit current, nanovoltmeter (Keithley 2182A) and picoammeter (Keithley 4185) are connected in series and parallel to the copper electrodes, respectively. The high-end of the nanovoltmeter probe is connected to the downstream electrode, whereas the low-end of the nanovoltmeter probe is connected to the upstream electrode, and we term it as straight polarity. Before the measurements, the channels are primed with the electrolyte solution for almost 1 hour. The entire set of experiments are performed under the sun-light of hot- summer days (from March to June, 2018 at Kharagpur, India; latitude 22.3460° N, longitude 87.2320° E), where the temperature, T and relative humidity, RH fluctuate between ~32-40 °C and ~60-85%. Hence, to maintain uniform experimental conditions such as T ~35 °C and RH ~60%, the experiments are further conducted inside a controlled temperature and humidity test chamber (Labard Instrument Pvt. Ltd.). All the tests are performed thrice to check the RA ) ~7×3 cm2, stem width repeatability of the measurements. The following are the FC dimensions; root area ( LA ) ~7×7 cm2 (optimized FC dimensions; and discussed in the SW ) ~4 cm, stem length ( ( subsequent section), followed during the entire course of experimentation. sL ) ~3 cm and leaf area ( Characterization The surface morphology of fabric cloth, structures of the fiber, pore size and distributions are characterized by Field Emission Scanning Electron Microscope, FESEM (JEOL, Model: JSM-7800 F). The electrical impedance of FC in the presence of 1 mM NaCl solution is measured using Potentiostat (GAMRY, Reference 600). Zeta-potential is measured using Electrokinetic analyzer for solid surface (Anton Paar GmbH, Model: SurPASS 3). Results and discussion Optimization of FC dimensions As the liquid imbibes through the porous and tortuous network of the cellulose fibers, an electrical potential is induced between two copper electrodes which gradually rises, and thereafter reaches a maximum value followed by LA . The ions (such as Na+, Cl -- , H+ and OH -- ) in the electrolyte a steady state upon absolute saturation of leaf area, solution (1 mM NaCl) creates an interfacial charge layer (also known as electrical double layer; EDL) upon interaction with negatively charged free surface groups (such as -- COOH, -- OH) of cellulose fabrics25,27. The counter-ions (Na+, H+) thus preferentially migrate uphill through the mobile layer of the EDL under a capillary driven pressure gradient, which eventually creates a charge polarization, leading to induced streaming potential. Initial fluctuations in the readings can be attributed from the fact of rapid rise of the capillary front which eventually equilibrates with the pore saturation. LA and SW ) on the induced streaming potential. Initially, the stem length (or the capillary length), We have performed sets of comprehensive experiments in order to understand the effect of FC dimensions SL SL , (such as LA ~5×5 cm2 constant. The individual saturation voltage (or steady-state voltage), SW ~3 cm and is varied, keeping scI satV and short-circuit current, is recorded in three different identical test-sets, TS, which ensures the repeatability of the measurement. The measurements are performed in straight polarity mode. It is observed that as the capillary length increases, the corresponding capillary rising time also increases with the consequence of reduced satV (Fig. 3a) and SL such as 3 cm, 5 cm, 7 cm, 9 cm, and 11 cm are approximately 10-12 min, 18-20 min, 30-35 min, 45-50 min and 60-65 min, respectively, 4 scI (Supplementary Fig. 2b). The recorded saturation time for different sL (Supplementary which further changes with the number of runs. The channel resistance, which is a function of Fig. 3; where Nyquist plot shows the variation of channel impedance with respect to stem length), increases with the channel length. This further impedes the transport of ions, and reduces . Thus, to sL , a commercial capacitor of ~0.1 mF is connected to the estimate the induced output power against different ; where C is the capacitance for all sL ~3 cm) delivers circuit. The output power is calculated by following the relation, value, sL (Supplementary Fig. 2a). Thus, a lower FC length ( ct is the capacitor charging time. We observed that cV is the capacitor charging voltage and satV and the corresponding P output  0.5 CV 2 c / t c these three sets decreases with increase in a stable open-circuit potential of ~350 mV with scI outputP scI of ~1.75 µA and outputP of ~110 nW. b Straight polarity d Reverse polarity a c Figure 3 Optimization of FC dimension. a, Saturation voltage with respect to stem length of FC. b, Temporal evolution of induced electric potential for OFC connected in straight polarity, inset shows time required for charging a capacitor of ~0.1 mF. c, Temporal evolution of short-circuit current for OFC. d, Temporal evolution of induced potential for OFC connected in reverse polarity investigated under environmental conditions such as T ~32-35 °C and RH ~65-75%. All the measurements are performed in presence of 1 mM NaCl solution. 5 is measured against different scI increases initially with To further explore the effect of leaf area, LA and stem width, SW on induced potential, outputP LA (Supplementary Fig. 2c and 2d) and as well as SW (Supplementary Fig. 2e and 2f) scI respectively. The bigger leaf area corresponds to a larger transpiration surface which essentially promotes the evaporation rate of water, resulting in a faster ion migration, and thereby induces a higher output power. The short- LA . However, it decreases after a certain critical value (such as ~5×5 circuit current, SW , an cm2), which can be attributed to increase in channel impedance only. On the other hand, the parameter upto certain critical value (such as intrinsic function of the fiber density, effectively can stimulate a higher SW ~4 cm). The increase in scI . However, further rise in SW , may amplify the channel impedance that leads to reduction in . Based on the sL ~3 cm, LA overall parametric study, an optimized FC (OFC) dimensions that deliver maximum ~7×7 cm2 and of ~580-700 mV (Fig. 3b; inset shows of ~153.5 nW (see inset of the charging characteristics of a capacitor) with Fig. 3b), maintained for ~8 hours of experiment under the sunlight, with environmental conditions such as T ~32-35 °C and RH ~65-75%. In order to further confirm the effect of EDL on induced streaming potential, we change the nanovoltmeter probes in reverse connection mode, where OCVV changes its polarity and thereby confirms the effect of negative surface charge on the induced electric potential (Fig. 3d). SW reduces the channel hydraulic resistance which eventually increases SW ~4 cm. The OFC is able to reliably generate a stable OCVV scI of ~2.1-3.4 µA (Fig. 3c) and outputP outputP outputP become: outputP Device performance In order to analyze the role of temperature and associated evaporative flux on device performance, we perform a day-night cyclic test which consists of an uninterrupted measurement for 3 days (Fig. 4a). The maximum OCVV is recorded as ~550-600 mV at peak-day hours (12:30 pm-2 pm), where the effect of solar-heat and water evaporation is observed to be significant. The device performance deteriorates gradually from its peak value as the day-light intensity reduces and reaches to a minimum value of ~180-200 mV during the mid-night (1 am-3 am), where both these effects are quite less. The device performance improves again in the next day as the sun starts rising and subsequently reaches to a maximum value during the peak-day hours (almost equivalent to previous day's VOCV value) and thus the cycle continues. It is worth noting that the overall performance of the device is almost same at the end of the cyclic test (i.e. after 72 hours). Thus, to ensure evaporation as the prime source of induced potential, we expose the leaf area to a hot-air (T ~65°C) blowing at a speed of ~2.5 m/s. It enhances the water evaporation rate from the leaf area, leading to gradual increase in voltage maximum up to ~585 mV (Fig. 4b). The induced potential drops to ~550 mV as the hot-air blower stops, and thus represents a significant correlation between the device performance and the rate of water evaporation. On the other hand, to appreciate the performance of the device for long-term usages, and hence to determine the robustness of the device, we conduct an experiment (exposed to day-light only) with three different channels, subjected to three different conditions for more than 30 days' (Fig. 4c). The first channel is continuously CTS ) for the total 30 days' duration of experiment; the second exposed to electrolyte solution (abbreviated as ); and the third channel is cleaned using DI water and dried after each day's measurement (abbreviated as CDTS 6 a b c Figure 4 Device performance in different ambient conditions. a, Day-night cyclic test, showing the variation of induced potential of 3 days' uninterrupted measurement. b, Effect of induced potential on periodic application of hot-air blower. c, Performance of OFC subjected to three different conditions against 30 days' measurement (measurement started from April 15, 2018). DTS ). The readings are channel is only dried after each day's measurement under room conditions (abbreviated as taken at different day-time intervals including the peak-day hours. It is observed that the device performance remains almost constant throughout the entire 30 days' span of experiment for all these three sets (Fig. 4c), with minor fluctuations in the readings, which can be ascribed from the fact of change in ambient conditions. Hence, the device can be treated as a 'greener power-plant' and can be used as a source of continuous power supply (at the expense of zero external energy consumption). Conclusions To summarize, we have shown that water evaporation from a centimeter-sized piece of wet cloth containing frugally-cut fabric channels can generate electrical power. Drawing analogies with evaporative transport in living plants and harnessing the consequent preferential migration of ions towards developing an electrical potential as mediated by capillary action, we have demonstrated that ordinary cellulose-based wet textile, may be capable enough for achieving this remarkable feat. As compared to previously reported methods of energy harvesting from 7 complex resources, the electricity generation occurs in natural ambience, directly converting the abundantly available thermal energy into electrical power. Further, in contrast to classical streaming potential generated by an applied pressure gradient or other external pumping resources, here the intrinsic surface energy of the fabric is used to drive the ionic current. Most importantly, our method paves the way of deploying regular fabric pieces as the sources of energy, with no special topographical manipulation of the cloth surface being demanded. Thus, the device does not necessitate any extensive fabrication protocol, unlike some recently reported evaporation driven energy harvesting devices. Finally, in a hot and dry environment, the natural evaporative transport gets spontaneously enhanced, so that the flow-induced electrical potential can be maximized. The device, thus, may turn out to be extremely effective in geographically warm and dry regions of the earth. Our results reveal that a single fabric channel can stably deliver a potential of ~700 mV in ambient conditions. This eventually culminates into a utilitarian paradigm of low-cost power harvesting in extreme rural settings. References 1. Chen, J. et al. Micro-cable structured textile for simultaneously harvesting solar and mechanical energy. Nat. Energy 1, 1 -- 8 (2016). 2. 3. 4. 5. 6. Wang, Z. L. & Song, J. Piezoelectric Nanogenerators Based on Zinc Oxide Nanowire Arrays. Science 312, 242 -- 247 (2006). Chu, S. & Majumdar, A. Opportunities and challenges for a sustainable energy future. Nature 488, 294-303 (2012). Tian, B. et al. Coaxial silicon nanowires as solar cells and nanoelectronic power sources. Nature 449, 885 -- 890 (2007). Yang, R., Qin, Y., Dai, L. & Wang, Z. L. Power generation with laterally packaged piezoelectric fine wires. Nat. Nanotechnol.4, 34-39 (2009). Gr, M. Photoelectrochemical cells. Nature 414, 338-344 (2001). 7. Wang, X., Song, J., Liu,J., and Wang, Z.L. Direct-Current Nanogenerator Driven by Ultrasonic Waves. Science 316, 102 -- 106 (2007). Chai, Z. et al. Tailorable and Wearable Textile Devices for Solar Energy Harvesting and Simultaneous Storage. ACS Nano 10, 9201 -- 9207 (2016). Zhao, F. et al. Highly efficient solar vapour generation via hierarchically nanostructured gels. Nat. Nanotechnol. 13, 489-495 (2018). Energy, D., Elimelech, M. & Phillip, W. A. The Future of Seawater and the Environment. Science 333, 712 -- 718 (2011). Yang, P. et al. Solar-driven simultaneous steam production and electricity generation from salinity. Energy Environ. Sci. 10, 1923 -- 1927 (2017). Xu, B. J. et al. Large-Scale Synthesis of Long Crystalline Cu2-xSe Nanowire Bundles by Water-Evaporation- Induced Self- Assembly and Their Application in Gas Sensing. Adv. Funct. Mater. 19, 1759 -- 1766 (2009). Xue, G. et al. Water-evaporation-induced electricity with nanostructured carbon materials. Nat. Nanotechnol. 12, 317 -- 321 (2017). 8 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. Van Der Heyden, F. H. J., Stein, D. & Dekker, C. Streaming currents in a single nanofluidic channel. Phys. Rev. Lett. 95, 9 -- 12 (2005). Van Der Heyden, F. H. J., Bonthuis, J., Stein, D., Meyer, C. & Dekker, C. Power Generation by Pressure- Driven Transport of Ions in Nanofluidic Channels. Nano Lett. 7, 1022-1025 (2007). Xie, Y. et al. Electric energy generation in single track-etched nanopores. Appl. Phys. Lett. 93, 8 -- 11 (2008). Gillespie, D. High energy conversion efficiency in nanofluidic channels. Nano Lett. 12, 1410 -- 1416 (2012). Fan, B., Bhattacharya, A. & Bandaru, P. R. Enhanced voltage generation through electrolyte flow on liquid-filled surfaces. Nat. Commun. 9, 1 -- 7 (2018). Li, C., Liu, K., Liu, H., Yang, B. & Hu, X. Capillary driven electrokinetic generator for environmental energy harvesting. Mater. Res. Bull. 90, 81 -- 86 (2017). Li, J. et al. Electricity generation from water droplets via capillary infiltrating. Nano Energy 48, 211 -- 216 (2018). Li, C. et al. Wearable energy-smart ribbons for synchronous energy harvest and storage. Nat. Commun. 7, 1 -- 10 (2016). Lee, S. et al. Triboelectric energy harvester based on wearable textile platforms employing various surface morphologies. Nano Energy 12, 410 -- 418 (2015). Li, S. et al. Cloth-based power shirt for wearable energy harvesting and clothes ornamentation. ACS Appl. Mater. Interfaces 7, 14912 -- 14916 (2015). Pu, X. et al. Wearable Self-Charging Power Textile Based on Flexible Yarn Supercapacitors and Fabric Nanogenerators. Adv. Mater. 28, 98 -- 105 (2016). 25. Das, S. S., Kar, S., Anwar, T., Saha, P. & Chakraborty, S. Hydroelectric power plant on a paper strip. Lab Chip 18, 1560 -- 1568 (2018). 26. Veerubhotla, R., Bandopadhyay, A., Das, D. and Chakraborty, S. Instant power generation from an air- breathing paper and pencil based bacterial bio-fuel cell. Lab Chip 15, 2580 -- 2583 (2015). 27. Mandal, P., Dey, R. and Chaktaborty, S. Electrokinetics with 'paper-and-pencil' devices. Lab Chip 12, 9 4026 -- 4028 (2012). Supplementary Information Electrokinetic Power Harvesting from Wet Textile Sankha Shuvra Das, Vinay Manaswi Pedireddi, Aditya Bandopadhyay, Partha Saha and Suman Chakraborty Department of Mechanical Engineering, Indian Institute of Technology Kharagpur, Kharagpur 721302, INDIA. *email: [email protected] Leaf Stem Supplementary Figure 1 Analogy of transport mechanism of a tree. Photograph representing the analogy of the various segment of a tree and fabric-based channel. Bigger leaf area promotes the evaporation rate. 10 a c e b d f Supplementary Figure 2 Parametric optimization of FC. (a) Output power, and (b) Short-circuit current with respect to stem length (with constant stem width of 3 cm and leaf area of 5×5 cm) of FC. (c) Output power, and (d) Short-circuit current with respect to leaf area (for optimized stem length of 3 cm and constant stem width of 3 cm) of FC. (e) Output power, and (f) Short- circuit current with respect to stem width (for optimized stem length of 3cm and optimized leaf area of 7×7 cm) of FC. 11 Supplementary Figure 3 Impedance measurement of FC. Nyquist plot represents the variation of impedance with respect to stem length of FC for 1 mM NaCl solution. Supplementary Figure 4 Pore size and distribution. FESEM image (at 30,000× magnification) of single fibre thread showing the pore size and pore distribution. The Scalebar on the panel represents 100 nm. 12
1901.05528	1	1901	2018-12-10T17:57:04	Metasurfaces Leveraging Solar Energy for Icephobicity	[ "physics.app-ph", "physics.optics" ]	Inhibiting ice accumulation on surfaces is an energy-intensive task and is of significant importance in nature and technology where it has found applications in windshields, automobiles, aviation, renewable energy generation, and infrastructure. Existing methods rely on on-site electrical heat generation, chemicals, or mechanical removal, with drawbacks ranging from financial costs to disruptive technical interventions and environmental incompatibility. Here we focus on applications where surface transparency is desirable and propose metasurfaces with embedded plasmonically enhanced light absorption heating, using ultra-thin hybrid metal-dielectric coatings, as a passive, viable approach for de-icing and anti-icing, in which the sole heat source is renewable solar energy. The balancing of transparency and absorption is achieved with rationally nano-engineered coatings consisting of gold nanoparticle inclusions in a dielectric (titanium dioxide), concentrating broadband absorbed solar energy into a small volume. This causes a > 10 {\deg}C temperature increase with respect to ambient at the air-solid interface, where ice is most likely to form, delaying freezing, reducing ice adhesion, when it occurs, to negligible levels (de-icing) and inhibiting frost formation (anti-icing). Our results illustrate an effective unexplored pathway towards environmentally compatible, solar-energy-driven icephobicity, enabled by respectively tailored plasmonic metasurfaces, with the ability to design the balance of transparency and light absorption.	physics.app-ph	physics	ACS Nano 2018, 12, 7009−7017 www.acsnano.org Metasurfaces Leveraging Solar Energy for Icephobicity Efstratios Mitridis, Thomas M. Schutzius, Alba Sicher, Claudio U. Hail, Hadi Eghlidi, and Dimos Poulikakos* Laboratory of Thermodynamics in Emerging Technologies, Department of Mechanical and Pro- cess Engineering, ETH Zurich, Sonneggstrasse 3, CH-8092 Zurich, Switzerland. Keywords: icephobicity, anti-icing, defrosting, renewable energy, plasmonic metasurface, met- amaterial Received: April 11, 2018 Accepted: June 22, 2018 Published: June 22, 2018 DOI: 10.1021/acsnano.8b02719 1 Abstract Inhibiting ice accumulation on surfaces is an energy-intensive task and is of significant im- portance in nature and technology where it has found applications in windshields, automobiles, aviation, renewable energy generation, and infrastructure. Existing methods rely on on-site elec- trical heat generation, chemicals, or mechanical removal, with drawbacks ranging from financial costs to disruptive technical interventions and environmental incompatibility. Here we focus on applications where surface transparency is desirable and propose metasurfaces with embedded plasmonically enhanced light absorption heating, using ultra-thin hybrid metal -- dielectric coat- ings, as a passive, viable approach for de-icing and anti-icing, in which the sole heat source is renewable solar energy. The balancing of transparency and absorption is achieved with rationally nano-engineered coatings consisting of gold nanoparticle inclusions in a dielectric (titanium diox- ide), concentrating broadband absorbed solar energy into a small volume. This causes a >10 °C temperature increase with respect to ambient at the air -- solid interface, where ice is most likely to form, delaying freezing, reducing ice adhesion, when it occurs, to negligible levels (de-icing) and inhibiting frost formation (anti-icing). Our results illustrate an effective unexplored pathway to- wards environmentally compatible, solar-energy-driven icephobicity, enabled by respectively tailored plasmonic metasurfaces, with the ability to design the balance of transparency and light absorption. 2 Icing is very common in nature and technology, and when not controlled or alleviated, it can have very negative consequences in a broad range of applications including automobiles,1 aviation,2 power distribution,3 shipping,4 road transportation networks,5 buildings,6 wind energy generators7 and photovoltaics.8 The energy requirements, cost and environmental impact of de- icing are surprisingly high. A very common example from everyday life is automobile windshield defrosting by means of hot air, that mandates the function of the engine for up to 30 min in cold climates.9 It is also estimated that the global aircraft de-icing market will be worth $1.30 billion by 2020,10 while the ice protection systems are expected to amount for $10.17 billion by 2021.11 A degree of optical transparency is a crucial property in many of these applications,1,2,6,8 achieved by -- often -- multifunctional windshields and windows, constituting indispensable architectural elements of commercial and residential buildings.12,13 Several passive anti-icing strategies, based on scientifically nano-engineered surfaces,14,15 have been developed since the late 1990s, includ- ing hierarchical superhydrophobic surfaces,16 -- 22 lubricant infused surfaces,23 and liquid infused polymers.24 Their icephobic properties are defined by the nucleation delay, droplet contact time reduction, reduced ice adhesion and defrosting time.16 -- 24 While promising and highly desirable, passive approaches are to date complementary to active systems such as resistive heaters7 and mechanical scraping.4 Such active systems, however, are energy intensive (requiring electricity),25 their operation is intrusive, have limited optical transparency and working tempera- ture,15,26 or require replenishment.15,23 What is less explored is harvesting the potential of ubiqui- tous sunlight to impart icephobicity, here through specifically tailored plasmonic metasurfaces. We show that by using rationally nano-engineered ultra-thin hybrid plasmonic metasur- faces, one can concentrate naturally occurring solar energy into a small volume, causing a greater than 10 °C temperature increase with respect to ambient at the air -- solid interface, where ice is most likely to form, delaying freezing, reducing ice adhesion to negligible levels (de-icing) and 3 inhibiting frost formation (anti-icing). Significant thermal responses can be achieved with trans- parent metasurfaces, paving the way to a wide range of applications where the benefit of icing resistance must be weighed against loss of transparency. We realize this by engineering an array of nanoscale noble metal particles embedded in a dielectric matrix -- an approach that is shown to be capable of tuning absorption and transparency in a systematic way based on film thickness and is well-suited for a fundamental study -- on industrially and commercially relevant substrates (e.g., glass, plastic), for a total film thickness in the sub-micron regime, ensuring maximum tempera- ture boost. The use of plasmonics in metal -- dielectric composites in applications such as water desalination,27 photovoltaics,28 solar water heating29 and photochemistry28,30 were explored be- fore. Plasmon resonances in metallic nanostructures can be damped radiatively (photon re- emission) or non-radiatively via Landau damping,31 resulting in rapid localized heating of the nanoparticles.32 -- 34 Here, exploiting the Landau damping of hot electrons32 in deeply sub- wavelength gold particles, and incorporating them in rationally designed metal-dielectric nano- composite metasurfaces, we show that a broadband absorption of solar energy, with adjustable levels of absorption and transparency, can be achieved within sub-wavelength films, confined in the surface. We demonstrate that such heating can be considerable and under harsh icing condi- tions it can significantly delay frost formation (anti-icing) and lead to the swift removal of frozen ice blocks from the surface (de-icing) for several freezing cycles. This is a straightforward ap- proach that leverages naturally occurring sunlight to achieve an impressive anti-icing and de- icing performance that does not rely on chemicals, mechanical action or electricity, translating into environmental and cost savings and operational facility. Results and Discussion We designed and fabricated plasmonic metasurfaces consisting of closely-packed, deeply 4 sub-wavelength metal particles, in a dielectric matrix.35,36 For the metal and dielectric, we chose gold (Au) and titanium dioxide (TiO2), respectively. Nanoscopic gold particles are very effective absorbers of sunlight at their plasmon resonance wavelengths. Embedding them in a dielectric matrix, with a volumetric concentration close to the percolation limit, leads to a very effective and ultra-broadband absorption. This absorption is attributed to the significantly increased imagi- nary part of the effective permittivity, Im( )ε -- equivalently effective electronic conductivity -- r,nc over an ultra-broad spectrum,35 in our designed nanocomposites. The increased conductivity leads to a boost in the photoexcited hot carriers,32 which generate heat through Landau damping.31 The selection of TiO2 as the dielectric is based on previous findings in which high absorption levels of over 80% across the visible wavelength range were demonstrated in Au -- TiO2 thin films. The TiO2 enhances the plasmon resonance of individual Au nanostructures and the plasmonic coupling of proximal nanostructures, enabling broadband light absorption.37 Other common dielectric materials, such as silicon dioxide or Teflon, are also employable, albeit using the high-refractive index TiO2 enables enhancement of the visible light absorption (see Support- ing Information, section 'Modeling light absorption' and Figure S1 for a comparison between TiO2 and other common dielectrics). Based on our theoretical evaluations and experimental results (discussed below), a metasurface composed of gold particles with sizes, d , of 5≈ nm, and a volumetric concentra- tion, Auv , of 40%≈ , embedded in a TiO2 matrix, exhibits a high level of absorption across the entire visible and near-infrared spectrum, and is the selected material system in this study. To realize the nanocomposites, we deposited Au and TiO2 via a layer-by-layer sputter deposition process, on fused silica and acrylic (PMMA) substrates. An experimental parametric study by changing the total film thickness ( L ) and the total number of deposited layers ( LN ) was 5 conducted for achieving a gold volumetric concentration, Au v ≈ 40% , and a desired level of ab- sorption and transparency. The individual layer thicknesses of Au and TiO2 were kept constant at 4.6 and 6.9 nm, respectively. Due to the anti-wetting properties of Au, the thin deposited gold layers formed isolated particles instead of a continuous film, effectively leading to a nanocompo- site with sub-wavelength, highly-absorptive inclusions (see Supporting Information, section 'Modeling light absorption' for the effect of the particle size). LN was varied from 4 to 44, pro- ducing film thicknesses ranging from L = 38 2 ± nm to L = 270 5 ± nm, respectively, including a 15-nm TiO2 top-layer that provided identical surface chemistry for all the fabricated samples. To enhance film -- substrate adhesion, a 2-nm chrome layer was deposited between the substrate and the metasurfaces. Figure 1a shows a picture of a plasmonic metasurface fabricated by the above procedure ( L =270 nm). It also details representative cross-sectional and top-view scanning electron micro- graphs of the coating. The cross-section consists of layers of Au nanoparticles embedded in a TiO2 matrix (shown with arrows). The pitch between adjacent Au layers, p , is constant ( p = 11 1 ± nm). To analyze the gold particle size distribution, we acquired a top-view scanning electron micrograph of an Au layer deposited on a TiO2 layer (Figure 1a, bottom-right; 8-layer metasurface, L = 45 nm, without a TiO2 top-layer). Figure 1b shows a histogram of the number of Au nanoparticles, N , normalized by the total number of particles, 0N , vs. their equivalent diameter, d , assuming spherical particles. We found that the particle size has a gamma distribu- tion with a mean value of 5.4 nm, variance of 2.1 nm and a range of 7.9 nm. The methodology of the particle size analysis is described in Supporting Information, section 'Nanoparticle size analy- sis'; see also Figure S2. In Figure 1c, the transparency of the metasurface (here, L = 60 nm) rela- tive to a control sample is demonstrated. The metasurface was placed on top of a printed logo and 6 illuminated with white light on the back side. Figure 1d shows the normalized light absorption spectra, , vs. wavelength of light, λ, for four metasurfaces with L = 38 nm, 60 nm, 95 nm, and 270 nm (wavelength range of 400 -- 800 nm). The mean absorption can be calculated as: ( ) ∫  λ λ λ λ min max = / ( − ) , where min λ = 400 nm, and max λ = 800 nm. From Figure 1d we     λ max λ min d     can see that the metasurfaces absorb light broadly across the visible spectrum and have a mean absorption value of = 28%, 37%, 63%, and 83% for L = 38 nm, 60 nm, 95 nm, and 270 nm, respectively, indicating the tunability of  with L . The broadband absorption is a result of the small sizes of the nanoparticles ( d λ<< ) and their collective behavior ( p and d are compara- ble) in the nanocomposite with a volumetric concentration close to the percolation limit. Figure 1e shows a plot of the normalized transmission spectra,  , vs. λ, for the same four nanocompo- sites used in Figure 1d. Here we see that for L = 38 nm, 60 nm, 95 nm, and 270 nm,  =51%, 36%, 15%, and 2%, designating also the tunability of  with L . Information regarding the in- dividual reflection and transmission spectra can be found in Supporting Information, section 'Op- tical spectroscopy' and Figure S3. Figure 1f shows the spectra of the absorbed, rI⋅ , and trans- mitted, rI⋅ , sunlight (standard solar irradiance at sea level) vs. λ, for the partially transparent metasurface with L = 60 nm. Also shown is the reference standard solar irradiance, rI ( 0%= and 100%= ). This nano-engineered metasurface exhibits a good balance of transparency and absorption, all within a deeply-subwavelength film. Figure 2a shows a schematic of the experimental setup we used to characterize the tem- perature change of the nano-engineered metasurfaces (deposited on glass substrates) due to visi- ble light illumination. For illumination we used a halogen light source, which we collimated and focused on the metasurfaces. (See Supporting Information, section 'Thermography' and Figure 7 S4 for further information on characterizing plasmonically enhanced light absorption heating.) The temperature increase in the metasurface relative to ambient, T∆ , was measured with a high- speed infrared camera (spectral range of 1.5 to 5.1 μm). The focused light diameter, D , was 6.0 0.3± mm and the power density, P , was 2.4 0.2± suns (kW m-2). We used a mechanical shutter to rapidly control illumination. Figure 2a (inset) also shows the spectrum of the broadband light source used in this study. Figure 2b shows T∆ vs. time, t , for the four samples with differ- ent values of  (28%, 37%, 63%, and 83%). Time-zero was when the metasurface was first il- luminated. Here, T∆ was measured at the center of the illuminated area on the surface. It is clear that all metasurfaces exhibit an appreciable change in temperature due to visible light illumina- tion, valid even for the highly transparent metasurfaces. We also note that there are transient and steady-state regimes for T∆ . T∆ vs. t curve is also shown for the reference sample (uncoated glass substrate). Figure 2c shows the corresponding spatial distributions of T∆ for the four metasurfaces at t = 180 s (steady state). The boundary of the illuminated area is marked with a dashed circle, and it is evident that the maximum value of T∆ occurs at the center of this area and that heat diffuses well beyond the illuminated area. The time to steady state is controlled by the characteristic length, L C = ( L D 2 − ) / 2 , and the substrate thermal diffusivity, α, where 2L is the side-length of the square sample. This time can be estimated as 2 C /L α. Substituting appropriate values yields ( 2 6 mm / 0.43 mm s )2 -1 84≈ s, which is comparable to the order of magnitude of the experimentally determined time ( 100 ≈ s). The temperature increase at the surface of the sample is determined by light irradiance and ab- sorption as well as heat losses due to conduction (in the film and substrate), convection (in the surrounding air), and radiation (to the surrounding environment); see Supporting Information, section 'Heat Transfer' and Figure S5 for a detailed analysis on the above. In summary, to under- 8 stand the relative importance of convection and radiation on determining the steady-state value of T∆ , we solved for the temperature distribution (sample with 37%= ) in a two-dimensional semi-infinite plate immersed in a gas that had a heated gas -- substrate interface and an adiabatic condition on its bottom interface. The boundary condition at the interface was modified to ac- count for radiation losses. We fixed a value of emissivity ( 0.8ε≈ ), based on the infrared meas- urements of T∆ , and varied the position of T∞ (aspect ratio of /L L ) until 1 2 sT T∞− (the tempera- ture difference between the gas -- substrate interface and the gas very far away) matched our exper- imentally determined value of T∆ . Based on the steady-state value of T∆ that we measured, we have determined that the percentage of cooling due to radiation is a mediocre 4% of the amount of heat provided by illumination, but it can also exceed 25% in certain cases and metasurfaces (see Supporting Information, section 'Heat Transfer'). To understand the importance of natural convection on cooling, we computed the Rayleigh number, LRa 1 = g 3 1 / TLβ ∆ ( ) να , where g is the acceleration due to gravity, β is the expansion coefficient (1/ T for an ideal gas), and ν is the kinematic gas viscosity. Below and above a critical value of 1LRa , Ra = c 1708 ,38 heat transfer is in the form of conduction and advection (convection and conduction), respectively. Substitut- ing appropriate values, we see that the value of 1LRa in this case is 3 7 10− ⋅ ( LRa 1 ≈ ⋅ 7 10 3 − for T∞ = 23 °C, P ≈ 2.4 kW m-2, 37%= , T∆ ≈ °C, 7 L L 1 2 / ≈ 1. 10 2 ⋅ 2 − , ν = 1.57 10 ⋅ 5 − m2 s-1, α = 22.07 10 ⋅ 6 − m2 s-1, L = 1 212 μm, β = 3.38 10 ⋅ 3 − K-1, g = 9.81 m s-2).39 Therefore, due to the fact that LRa 1 Ra< c , we conclude that heat transfer through conduction is the dominant mecha- nism and natural convection can be neglected. Figure 3a shows a schematic of the experimental setup used to investigate the effect of il- lumination (halogen lamp; D = 6.0 0.3 ± mm, P = 2.4 0.2 ± kW m-2) on film -- ice adhesion (de- 9 icing), on each metasurface, which was held in place by a holder on an x-y piezo stage. A hydro- phobic cylinder (inner radius of R = 1.5 mm), filled with water, was placed on top of it, concen- trically to the illuminated area. The experiments took place at T = − °C: a pin connected to a 4 piezoelectric force sensor was initially pressed at the base of the frozen ice cylinder parallel to the x-axis inducing a shear stress, yxτ . (The value of yxτ , yxτ = 90 5 ± kPa, was selected to be close to, but less than, the mean ice adhesion strength of a PVDF-coated substrate, 131 19± kPa; 3 ex- periments), in order to prevent premature detachment of the ice cylinder. Illumination was switched on with a mechanical shutter at t = and 0 yxτ vs. t was recorded. The temperature of the chamber was controlled by flowing cold nitrogen gas. All the samples were coated with a thin PVDF top-layer prior to the experiments, to ensure identical wetting properties (see Methods for details). The PVDF protective films exhibit very high optical transparency due to their sub- micron thickness. Moreover, they can act as single-layer anti-reflection coatings. This can be clarified by considering the relationship n ar = n n air TiO 2 , where arn is the refractive index of the single-layer anti-reflection coating, n ≈ is the refractive index of air and air 1 n 2TiO ≈ 2.5 is the re- fractive index of the enclosing TiO2 layer of the metasurface in the visible wavelength range. This leads to the desired refractive index of the anti-reflection coating of n ≈ ar 1.6 , which is close to the refractive index of PVDF, PVDF 1.35 ≈ n .37 Therefore, upon top-side illumination, we expect that the PVDF layer decreases the reflectivity on the top side of the surface, thus boosting the level of absorption and the plasmonic heating. This was also confirmed experimentally for a PVDF-coated partially-transparent metasurface with L ≈ 60 nm, resulting in 41%= , vs. 34%= for the same uncoated metasurface (see Supporting Information, section "Optical spec- troscopy" and Figure S3 for the effect of the protection layer on the absorption, reflection and 10 transmission of the metasurfaces). In our experiments, though, where bottom-side illumination is used, since the incident light does not pass through the PVDF layer before impinging on the metasurface, the levels of absorption and the plasmonic heating should not change considerably in the presence of the PVDF top-layer. Figure 3b shows a plot of yxτ vs. t for the illuminated metasurface ( 37%= , blue line) and control (black line) samples, for several de-icing cycles. The gray and blue shaded regions surrounding the blue (metasurface, 9 experiments) and black (control, 3 experiments) lines, re- spectively, are the minimum and maximum values of yxτ observed during the experiments. Two regimes appear for the metasurface: almost constant yxτ (prior to illumination) and sharply de- creasing yxτ (during illumination) until reaching the minimum measurable stress ( 2.5 1.0± kPa). For the control case, there is only one regime with constant yxτ . Figure 3c shows boxplots of de- icing times (i.e. time elapsed from 0P > kW m-2 until yxτ ≈ ), dt , vs. , at 0 T = − °C. The 4 mean de-icing times were 394 211 ± s, 264 80± s,76 18± s and 34 11± s, for metasurfaces with = 28%, 37%, 63% and 83% respectively. From the graph, it is evident that there is an order of magnitude decrease in the de-icing time by increasing the amount of solar energy that is absorbed (metasurfaces with 28%= vs. 83%= ). For more information on the setup and calibration process of the force sensor, see Supporting Information, section 'Ice adhesion setup' and Figure S6. In the above, complete de-icing was achieved in all cases, which we attribute to the formation of an intervening melt layer at the surface. We ascribed the de-icing time, dt , and the gradual reduction of yxτ with time to the formation of a melt layer at the center of the ice -- film contact area (warmest region) and subsequent radial outward propagation of the phase boundary towards the edge (coldest region). For considerations on the effect of viscous and capillary forces in re- 11 sisting the ice -- block motion, which we found to be insignificant relative to ice adhesion, see Supporting Information, section 'De-icing analysis'. Figure 4a shows the cold chamber -- integrated with the visible light illumination system -- that was used to characterize the plasmonically enhanced light absorption heating in a partially transparent metasurface ( 37%= , L = 60 nm) at sub-zero temperatures. For this part of the work, we chose to deposit the metasurface onto a thermally insulating substrate, poly(methyl methacrylate) (PMMA, k = s 0.2 W m-1 K-1, 1l = mm), in order to minimize thermal losses due to conduction. Both the control and metasurface were coated with a thin layer of PVDF (trans- parent) to ensure that the surface chemical composition is similar. Next, we characterize the freezing behavior of a single supercooled water droplet on the illuminated control and metasurface. We ran the experiments by first placing the coated substrate on the sample holder. Then, we turned on the light source and focused it on the metasurface ( 6D ≈ mm, P ≈ 2.4 kW m-2). Boiling liquid nitrogen was then flowed throughout the chamber to cool it down. We continuously measured the environmental gas ( 1T ) and surface ( 2T ) tempera- tures. To run the droplet freezing experiment, we first set 1 T ≈ − 26 °C. Then, a single water drop- let, initially at room temperature, was deposited on the substrate at the center of the illuminated spot. Figure 4b-c shows representative side-view image sequences of a water droplet on a (b) control and (c) metasurface cooled down at a rate of 1≈ °C min-1, until spontaneous nucleation and freezing. This transition is characterized by a sudden change from a transparent to opaque droplet state (recalescent freezing,40 see Supporting Information Video S1). Also indicated are the time, t , and 1T . Time-zero was considered as the time moment at which the droplet is in thermal equilibrium with the environment (to ensure that, we waited 5≈ min after droplet placement and the light was switched on). The change in droplet volume during the experiments 12 was relatively small and we estimated it to be 5%< h-1. We define the environmental gas tem- perature just prior to freezing as 1T . The metasurface freezes at a much lower * 1T compared to * the control ( t = 910 s, * T = − 1 48 °C, vs. t = 230 s and * T = − 1 33 °C). Figure 4d shows a plot of 2T vs. 1T for the control ( -- ) and metasurface (- - -). We see that the metasurface has a significantly higher temperature relative to the control case for a range of sub-zero temperatures (-53 to 30 °C). Figure 4e shows a plot (calibration curve) of 1T vs. sample type (metasurface and control). * For the control and metasurface, we measured 1T to be 34 2 * − ± °C (16 experiments) and 47 3 ± − °C (14 experiments), respectively. It is clear that there is a significant difference in 1T for the two * cases -- which has equally significant implications in the freezing delay time (explored next) -- that we can clearly attribute to the heating effect due to illumination. To understand the significance of these results, we can use the classical nucleation theory.17,41 We term the supercooled water droplet temperatures on an illuminated control and metasurface as d,0T and dT , respectively. If we assume that T 2 T≈ d ,0 and 2T T≈ d on the respective samples, and we set 1 T ≈ − 34 °C, then we have T = − d,0 26 °C and d T = − 20 °C (from the calibra- tion curve). In the case of the control sample, T = − d,0 26 °C is the spontaneous nucleation tem- perature, NT . The difference in droplet temperatures is then defined as T T ∆ = d − T d,0 . Previously, it was shown that T ∆ ∝ log 10 ( t av ) , where avt is the average time required for ice to nucleate in a supercooled droplet when the droplet is maintained at thermal equilibrium with its surroundings;17 therefore, for a six degree temperature difference, one can expect a six orders of magnitude increase in avt for the metasurface relative to the control case, which is associated with a very pronounced freezing delay. See also Supporting Information, section 'Frosting char- acterization', Figure S7 and Figure S8 for frosting experiments in harsh environmental condi- 13 tions: ambient humidity and high heat flux to the substrate. Next, the defrosting potential of the partially transparent metasurfaces is investigated. Figure 5a -- b shows an image sequence of a (a) frosted control sample and (b) metasurface ( = 37%) samples (substrate: PMMA, 1l = mm) in a cold dry environment ( 1 T = − 16 °C to 15− °C) that are illuminated with a halogen lamp ( P ≈ 2.4 kW m-2) for 0 t≤ ≤ 600 s. Figure 5a shows that the frost on the control sample is unaffected by illumination, while Figure 5b shows that the metasurface is completely defrosted at the illuminated area by t = 140 s (see Supporting Information Video S2 for a defrosting demonstration on a frosted control sample and metasur- face). To eliminate the effects of surface composition on frost growth, both the control and metasurface were coated with a thin layer of PVDF. Frost was grown on both samples under identical environmental conditions and for the same duration, ensuring similar frost thicknesses. Due to the lower thermal conductivity and increased thickness of the PMMA substrate relative to the glass, we should expect that heat transfer into the sample holder should be minimized. Fur- thermore, we note that there is an insulating frost layer on top of the sample; therefore, one should expect a higher steady-state temperature increase in the illuminated metasurface, allowing defrosting to occur in-spite of the relatively cold surrounding environment. Conclusions In closing, we showed that with rationally designed hybrid metamaterial films that bal- ance transparency and absorption, extreme icephobic surface performances can be achieved. Such films, here nanocomposites of gold and titanium dioxide, exhibit broadband visible light absorp- tion, while being sub-wavelength thin, enabling localized ice melting at the film -- ice interface. The plasmonically enhanced light absorption heating induced a temperature increase greater than 10 °C, compared to a control surface, for rapid de-icing within 30 s. Furthermore, we achieved a 14 6 °C decrease in the spontaneous nucleation temperature (resulting in 6≈ orders of magnitude increase in droplet freezing delay at -32 °C), and a defrosting time of 4≈ min, for a highly trans- parent metasurface ( 37%= , 36%= ). We presented a viable, passive, anti-icing and de- icing metamaterial platform harvesting the benefit of solar radiation, that can find a broad range of applications, especially where transparency is required, including water solar heating, automo- tive industry, residential and commercial buildings and construction or machinery infrastructure. We believe that although the present approach demonstrates both anti-icing and de-icing behavior while maintaining transparency, it could be further improved by incorporating other passive icephobicity designs based on surface nanoengineering.16 -- 24 Methods Substrate preparation. Double side polished, 4-in fused silica wafers ( l = 500 μm) were sourced from UniversityWafer, Inc. A 5-μm protective photoresist layer was spin coated and de- veloped on each wafer, which was subsequently cut into 18 mm by 18 mm square pieces, using an ADT ProVectus LA 7100 semi-automatic wafer dicer. The cut glass substrates were sonicated in acetone for 3 min, in order to remove the photoresist, followed by an equal-time sonication in isopropyl alcohol. Finally, they were dried in a nitrogen stream. PMMA substrates were prepared by manually cutting a PMMA sheet (Schlösser GmbH, 1l = mm) into rectangular pieces ( ≈18 mm by 18 mm), removing the protective membrane and sonicating in water. Adhesion layer and thin film deposition. A 2-nm chrome adhesion layer was deposited on the substrates, using an Evatec BAK501 LL thermal evaporator. The multilayer structure was then applied layer-by-layer via sputter deposition in argon atmosphere, by employing a Von Ar- denne CS 320 C sputter tool. An RF field at a power of 600 W was used at the TiO2 target, while a 50 W DC field was used in the case of the Au target. Deposition times were 43 s and 3 s, re- 15 spectively, at a pressure of 6 μbar. A pre-sputtering time of 30 s was necessary for stabilizing the plasma and thus the deposition rate in the chamber. The first layer was TiO2, followed by Au, and the alternation continued until the desired number of layers was reached. The deposition time for the TiO2 top-layer was 72 s. Film characterization. The samples were cleaved after scratching the glass substrate with a diamond tip on two opposite sides. Film thickness was extracted from cross-sectional im- ages of the metasurface with L = 270 nm (44 layers), taken by a FIE Nova NanoSEM 450 scan- ning electron microscope, at an acceleration voltage of 2 kV. This approach also gave us a visual- ization of the cross-sectional particle distribution of the same metasurface. In the case of the top- view image of the Au nanoparticles (8-layer metasurface, L = 45 nm), acceleration voltage was 0.5 kV. For the particle size distribution (equivalent diameter) analysis of the top-view image, ImageJ and MATLAB software packages were employed. Light absorption measurements took place in two steps: the optical transmission,  , and reflection, , spectra of the metasurface were individually recorded at the same spot on the sample, over the 400 -- 800 nm wavelength range, by a UV -- Visible spectrometer (Acton SP2500, Princeton Instruments), making the as- sumption of negligible light scattering. The absorption spectra were obtained by    . 1= − − Measurements from three different spots per sample were averaged to extract the absorption curves in Figure 1d and Figure S3a -- d. Polymer protective coating and characterization. A polyvinylidene fluoride (PVDF) protective layer was spin coated (Laurell WS-400B-6NPP/LITE) on top of the metasurfaces de- posited on fused silica substrates, to minimize surface -- ice interactions and provide mechanical durability. For this purpose, a PVDF solution (4 wt.%) in N,N-dimethylformamide (DMF) was prepared under rigorous stirring for 2.5 h. After cleaning the sample with acetone, under soni- cation, and isopropyl alcohol, the solution was spin coated (30 s at 3000 rpm) onto it, and then it 16 was heated for 3 h at 200 °C, over the melting temperature of PVDF, to reduce surface rough- ness. The supplier of the PVDF (beads, Mw ≈180,000) and DMF (anhydrous, 99.8%) was Sigma- Aldrich Co LLC. Advancing and receding water contact angle measurements were performed in a OCA 35 goniometer (DataPhysics), using the inflation/deflation technique (droplet volume of 8 -- 10 μL), equal to 86.7 0.9± ° (advancing) and 73.0 1.3± ° (receding). The respective contact angles of a PVDF-coated only control substrate were 87.9 0.5± ° and 72.3 0.9± °. In the case of metasurfaces on PMMA substrates, a similar process to the ones on fused silica was followed, with the differences of only cleaning with sonication in water and heating up to 80 °C for 3 h (below the glass transition temperature of PMMA). A 4 wt.% PVDF solution in NMP (anhy- drous, 99.5%, Sigma-Aldrich Co LLC) was used in this case. Simulation software. MATLAB software suite was used to estimate the imaginary part of electric permittivity of our nanocomposite films. The heat transfer simulations were performed numerically in COMSOL Multiphysics Modeling Software. IR thermal response measurements. The transient thermal response of the unprotected metasurfaces on fused silica substrate ( l = 500 μm), as well as the one of an uncoated reference fused silica substrate, were measured by means of an infrared camera (FLIR SC7500, 1.5 -- 5.1 um), equipped with a 50 mm F/2 lens, within 320 by 256 pixels (pixel pitch: 30 μm), at a framer- ate of 50 fps. A visible light illumination source (FLEXILUX 600 Longlife) consisting of a 50 W halogen lamp and a 5-mm diameter gooseneck fiber constituted the power source for the illumi- nation of the samples. Light from the bottom side (substrate), was collimated and focused on the top-surface with two convex, 2-in lenses, using a Thorlabs monochrome CCD camera (DCC1545M-GL). The light spot had a diameter of 6.0 0.3± mm, corresponding to a maximum power density of 2.4 0.2± kW m-2 (suns), measured with a Thorlabs S301C, 0.19 -- 25 μm power meter. For calculating the emissivity of the samples, these were heated up on a hot plate to three 17 discrete elevated temperatures, while the hot plate was kept at a low angle (less than 5°) with respect to the IR camera lens, in order to eliminate the Narcissus effect. A fast mechanical shutter (Melles Griot, 04 IMS 001) was used to cut the illumination on and off. The light was switched on at least 15 min prior to the experiments, for stabilization reasons. Five experiments were per- formed per sample, with a recording time of 200 s. De-icing time and ice adhesion measurements. A home-built, temperature-controlled, zero humidity chamber was used for the purpose of the de-icing experiments under illumination, consisting of: a bronze cooling pipe, where cold nitrogen at -150 °C was supplied by a Kaltgas cryogenic cooling system; a piezo-actuated stage made by Smaract, consisting of three SLC- 1730-S positioners; two 4-wire, class A, RTD temperature sensors (Pt-1000, class B, Sensirion); a humidity module (SHT30, Sensirion); a A201-1 piezoelectric force sensor (FlexiForce Quick- Start Board, Tekscan); a force transfer pin; and finally a 3-mm thick glass window that enabled the de-icing experiments due to illumination. The sensor values were recorded through a custom data acquisition box (Beckhoff). A hydrophobic polypropylene cylinder ( R = 1.5 mm) was filled with fresh deionized water (EMD Millipore Direct-Q 3) and placed on the sample, which was mounted on the stage. The force transfer pin -- sample distance was 1≈ mm. A vacuum-insulated double shell minimized thermal losses and forced convection inside the chamber was enabled with a fan. Humidity levels were kept at zero throughout the experiments, via a cold nitrogen recirculation stream. The force vs. displacement data were then recorded every 50 ms. The max- imum ice adhesion strength measurable with this setup for the given R is 280 ≈ kPa. Anti-icing and defrosting experiments. The same chamber as in the de-icing experi- ments was used. A cold nitrogen recirculation stream maintained dry conditions. The exposure time of the camera was necessary to be readjusted, due to severe changes in the intensity of inci- dent light, at the following time moments (defrosting experiments): (a) t = s and (b) 0 t = 600 s. 18 In the case, again, of the defrosting experiments (control: 3 experiments, metasurface: 3 experi- ments), frost was grown in ambient humidity conditions, by placing each sample on a cold block, at a temperature of -50 °C, for 45 min. Transfer of the frosted sample to the pre-cooled chamber was done in a fast manner to prevent melting of the formed layer. Anti-frosting experiments in ambient humidity conditions. An in-house setup was prepared for the anti-frosting experiments, consisting of a xenon light source (300 W 6258 Xe lamp in a 87005 enclosure, Newport), two objectives (4x, 10x) to collimate and focus the light on the sample surface, which was vertically mounted on a peltier element (38.6 W, PE-127-14-25-S, Laird), a cooling system (SST-TD02-LITE, Silverstone) and a Thorlabs CCD camera (DCC1545M-GL). A PID peltier control circuit (TEC-1089-SV, TEC Engineering) was used to regulate temperature (measured with a PT-100 type RTD). The power density of light on the metasurface was 1P ≈ kW m-2. The light was switched on for 30 min prior to the experiments for stabilization reasons. Recording framerate was 1 fps. The exposed area on the sample was dried with a nitrogen stream prior and after every frosting cycle. Supporting Information The Supporting Information is available online on the ACS Publications website. The fol- lowing sections are included: Modeling light absorption, nanoparticle size analysis, characteriz- ing the absorption, transparency, and reflection of metasurfaces, thermography of the illuminated metasurfaces, heat transfer calculations, ice adhesion setup and de-icing analysis, and frosting characterization. Moreover, two videos are included, demonstrating the effect of illumination on droplet nucleation temperature, and defrosting of a partially transparent metasurface. Author Information Corresponding author 19 E-mail : [email protected] E-mail : [email protected] E-mail: [email protected] Author Contributions Statement D.P. conceived the research idea, D.P., T.M.S., and H.E., designed research and provided scientific guidance in all aspects of the work. E.M., A.S., and C.H. conducted the experiments and analyzed the results. D.P., E.M., T.M.S., and H.E. wrote the paper draft and all authors par- ticipated in manuscript reading, correcting and commenting. Acknowledgements Partial support of the Swiss National Science Foundation under grant number 162565 and the European Research Council under Advanced Grant 669908 (INTICE) is acknowledged. E. M. thanks Reidt S. and Olziersky A. for their assistance in SEM image acquisition, Stutz R. for the sputtering deposition parameters, Drechsler U. for cleanroom introduction, Caimi D. for wafer dicing, Graeber G. for providing the blackbody sample, as well as Vidic J. and Feusi P. for tech- nical support. Additional Information Competing financial interests: The authors declare no competing financial interests. References (1) Petrenko, V. F.; Higa, M.; Starostin, M.; Deresh, L. Pulse Electrothermal De-Icing. Eng. Conf. 2003, 5, 435 -- 438. (2) Gent, R. W.; Dart, N. P.; Cansdale, J. T. Aircraft Icing. Philos. Trans. R. Soc. A Math. 20 Phys. Eng. Sci. 2000, 358, 2873 -- 2911. (3) Farzaneh, M. Atmospheric Icing of Power Networks; Springer Netherlands, 2008. (4) Rashid, T.; Khawaja, H. A.; Edvardsen, K. Review of Marine Icing and Anti-/de-Icing Systems. J. Mar. Eng. Technol. 2016, 15, 79 -- 87. (5) Norrman, J.; Eriksson, M.; Lindqvist, S. Relationships between Road Slipperiness, Traffic Accident Risk and Winter Road Maintenance Activity. Clim. Res. 2000, 15, 185 -- 193. (6) Tobiasson, W.; Buska, J.; Greatorex, A. Ventilating Attics to Minimize Icings at Eaves. Energy Build. 1994, 21, 229 -- 234. (7) Parent, O.; Ilinca, A. Anti-Icing and de-Icing Techniques for Wind Turbines: Critical Review. Cold Reg. Sci. Technol. 2011, 65, 88 -- 96. (8) Fillion, R. M.; Riahi, A. R.; Edrisy, A. A Review of Icing Prevention in Photovoltaic Devices by Surface Engineering. Renew. Sustain. Energy Rev. 2014, 32, 797 -- 809. (9) Farag, A.; Huang, L.-J. CFD Analysis and Validation of Automotive Windshield De-Icing Simulation; 2003. (10) MarketsandMarkets Research Private. Aircraft De-Icing Market worth $1.30 Billion by 2020 https://www.marketsandmarkets.com/PressReleases/aircraft-de-icing.asp (accessed Aug 8, 2017). (11) MarketsandMarkets Research Private. Ice Protection Systems Market worth 10.17 Billion USD by 2021 https://www.marketsandmarkets.com/PressReleases/ice-protection- system.asp (accessed Aug 8, 2017). (12) Yang, P.; Sun, P.; Chai, Z.; Huang, L.; Cai, X.; Tan, S.; Song, J.; Mai, W. Large-Scale Fabrication of Pseudocapacitive Glass Windows That Combine Electrochromism and Energy Storage. Angew. Chem., Int. Ed. 2014, 53, 11935 -- 11939. (13) Martın-Palma, R. Silver-Based Low-Emissivity Coatings for Architectural Windows: 21 Optical and Structural Properties. Sol. Energy Mater. Sol. Cells 1998, 53, 55 -- 66. (14) Deng, X.; Schellenberger, F.; Papadopoulos, P.; Vollmer, D.; Butt, H. J. Liquid Drops Impacting Superamphiphobic Coatings. Langmuir 2013, 29, 7847 -- 7856. (15) Wong, T.-S.; Kang, S. H.; Tang, S. K. Y.; Smythe, E. J.; Hatton, B. D.; Grinthal, A.; Aizenberg, J. Bioinspired Self-Repairing Slippery Surfaces with Pressure-Stable Omniphobicity. Nature 2011, 477, 443 -- 447. (16) Schutzius, T. M.; Jung, S.; Maitra, T.; Eberle, P.; Antonini, C.; Stamatopoulos, C.; Poulikakos, D. Physics of Icing and Rational Design of Surfaces with Extraordinary Icephobicity. Langmuir 2015, 31, 4807 -- 4821. (17) Eberle, P.; Tiwari, M. K.; Maitra, T.; Poulikakos, D. Rational Nanostructuring of Surfaces for Extraordinary Icephobicity. Nanoscale 2014, 6, 4874 -- 4881. (18) Boreyko, J. B.; Srijanto, B. R.; Nguyen, T. D.; Vega, C.; Fuentes-Cabrera, M.; Collier, C. P. Dynamic Defrosting on Nanostructured Superhydrophobic Surfaces. Langmuir 2013, 29, 9516 -- 9524. (19) Davis, A.; Yeong, Y. H.; Steele, A.; Bayer, I. S.; Loth, E. Superhydrophobic Nanocomposite Surface Topography and Ice Adhesion. ACS Appl. Mater. Interfaces 2014, 6, 9272 -- 9279. (20) Meuler, A. J.; Smith, J. D.; Varanasi, K. K.; Mabry, J. M.; McKinley, G. H.; Cohen, R. E. Relationships between Water Wettability and Ice Adhesion. ACS Appl. Mater. Interfaces 2010, 2, 3100 -- 3110. (21) Chen, X.; Ma, R.; Zhou, H.; Zhou, X.; Che, L.; Yao, S.; Wang, Z. Activating the Microscale Edge Effect in a Hierarchical Surface for Frosting Suppression and Defrosting Promotion. Sci. Rep. 2013, 3, 2515. (22) Kreder, M. J.; Alvarenga, J.; Kim, P.; Aizenberg, J. Design of Anti-Icing Surfaces: 22 Smooth, Textured or Slippery? Nat. Rev. Mater. 2016, 1, 15003. (23) Kim, P.; Wong, T. S.; Alvarenga, J.; Kreder, M. J.; Adorno-Martinez, W. E.; Aizenberg, J. Liquid-Infused Nanostructured Surfaces with Extreme Anti-Ice and Anti-Frost Performance. ACS Nano 2012, 6, 6569 -- 6577. (24) Zhu, L.; Xue, J.; Wang, Y.; Chen, Q.; Ding, J.; Wang, Q. Ice-Phobic Coatings Based on Silicon-Oil-Infused Polydimethylsiloxane. ACS Appl. Mater. Interfaces 2013, 5, 4053 -- 4062. (25) He, X.; Liu, A.; Hu, X.; Song, M.; Duan, F.; Lan, Q.; Xiao, J.; Liu, J.; Zhang, M.; Chen, Y.; Zeng, Q. Temperature-Controlled Transparent-Film Heater Based on Silver Nanowire -- PMMA Composite Film. Nanotechnology 2016, 27, 475709. (26) Irajizad, P.; Hasnain, M.; Farokhnia, N.; Sajadi, S. M.; Ghasemi, H. Magnetic Slippery Extreme Icephobic Surfaces. Nat. Commun. 2016, 7, 13395. (27) Zhou, L.; Tan, Y.; Wang, J.; Xu, W.; Yuan, Y.; Cai, W.; Zhu, S.; Zhu, J. 3D Self- Assembly of Aluminium Nanoparticles for Plasmon-Enhanced Solar Desalination. Nat. Photonics 2016, 10, 393 -- 398. (28) Clavero, C. Plasmon-Induced Hot-Electron Generation at Nanoparticle/Metal-Oxide Interfaces for Photovoltaic and Photocatalytic Devices. Nat. Photonics 2014, 8, 95 -- 103. (29) Boström, T.; Westin, G.; Wäckelgård, E. Optimization of a Solution-Chemically Derived Solar Absorbing Spectrally Selective Surface. Sol. Energy Mater. Sol. Cells 2007, 91, 38 -- 43. (30) Sarina, S.; Waclawik, E. R.; Zhu, H. Photocatalysis on Supported Gold and Silver Nanoparticles under Ultraviolet and Visible Light Irradiation. Green Chem. 2013, 15, 1814-1833. (31) Li, X.; Xiao, D.; Zhang, Z. Landau Damping of Quantum Plasmons in Metal 23 Nanostructures. New J. Phys. 2013, 15, 023011. (32) Brongersma, M. L.; Halas, N. J.; Nordlander, P. Plasmon-Induced Hot Carrier Science and Technology. Nat. Nanotechnol. 2015, 10, 25 -- 34. (33) Wu, K.; Chen, J.; McBride, J. R.; Lian, T. Efficient Hot-Electron Transfer by a Plasmon- Induced Interfacial Charge-Transfer Transition. Science. 2015, 349, 632 -- 635. (34) Qin, Z.; Wang, Y.; Randrianalisoa, J.; Raeesi, V.; Chan, W. C. W.; Lipinski, W.; Bischof, J. C. Quantitative Comparison of Photothermal Heat Generation between Gold Nanospheres and Nanorods. Sci. Rep. 2016, 6, 29836. (35) Shalaev, V.; Cai, W. Optical Metamaterials: Fundamentals and Applications; Springer: New York, 2010. (36) Hedayati, M. K.; Faupel, F.; Elbahri, M. Review of Plasmonic Nanocomposite Metamaterial Absorber. Materials. 2014, 7, 1221 -- 1248. (37) Hedayati, M. K.; Javaherirahim, M.; Mozooni, B.; Abdelaziz, R.; Tavassolizadeh, A.; Chakravadhanula, V. S. K.; Zaporojtchenko, V.; Strunkus, T.; Faupel, F.; Elbahri, M. Design of a Perfect Black Absorber at Visible Frequencies Using Plasmonic Metamaterials. Adv. Mater. 2011, 23, 5410 -- 5414. (38) Incropera, F. P.; DeWitt, D. P.; Bergman, T. L.; Lavine, A. S. Fundamentals of Heat and Mass Transfer, 6th ed.; John Wiley & Sons, 2006. (39) Perry, R. H.; Green, D. W.; Maloney, J. O. Perry's Chemical Engineers' Handbook, 7th ed.; McGraw-Hill: New York, 1997. (40) Jung, S.; Tiwari, M. K.; Poulikakos, D. Frost Halos from Supercooled Water Droplets. Proc. Natl. Acad. Sci. 2012, 109, 16073 -- 16078. (41) Hobbs, P. V. Ice Physics; Clarendon Press: Oxford, 1974. 24 Figures Figure 1 Characterizing the topography and optical properties of the plasmonic metasurfaces. (a) Macroscopic (top-row) and microscopic (bottom-row) images of a metasurface; cross-sectional (bottom-left; L = 270 nm, p = 11 nm) and top-view (bottom-right; L = 45 nm) micrographs of the metasurface. Bright regions ( L = 45 nm, obtained with backscattered and secondary elec- trons) correspond to gold nanoparticles. The volumetric concentration of gold is 40%≈ . Sample surface is 18 mm by 18 mm. (b) Relative frequency of gold nanoparticles, /N N , vs. nanoparti- 0 cle diameter, d (sample properties: L = 45 nm, p = 11 nm). (c) Demonstration of the transpar- ency of the metasurfaces (here, L = 60 nm), vs. a control sample, placed on a printed logo and under white backlight illumination. (d) Normalized absorption, , and (e) normalized transmis- sion,  , vs. wavelength of light (400 -- 800 nm), λ, for films with varying L : 38 nm (• • •), 60 nm 25 ( -- -- -- ), 95 nm (• -- •), and 270 nm ( -- -- -- ). (f) Absorbed, rI⋅ ( -- -- -- ), and transmitted, rI⋅ ( -- -- -- ) , sunlight (standard solar irradiance), vs. λ, for a metasurface ( L = 60 nm) and reference sample ( 0%= , 100%= , -- -- -- ). Scale bars: (a) top-row, 5 mm; bottom-left, 50 nm; bottom-right, 30 nm; (c) 2 cm. Figure 2 Characterizing the heating behavior of the plasmonic metasurfaces (fused silica sub- strate) due to visible light exposure. (a) Schematic of the setup used to characterize the thermal response of the metasurfaces due to illumination: 1, collimating lens; 2, mechanical shutter; 3, silver mirror; 4, beam splitter; 5,6, focusing lenses; 7, CMOS camera; 8, sample holder; 9, sam- ple. The spectrum of the broadband light source ( -- -- -- ) is shown in the inset. (b) Temperature change, T∆ , vs. time, t , for metasurfaces with varying values of  (28% • • •; 37% -- -- -- ; 63% • -- •; 83% -- -- -- ) and a control substrate ( -- -- -- ) after illumination ( P ≈ 2.4 kW m-2); time-zero is de- fined as the moment that the mechanical shutter was opened. (c) Spatial distribution of T∆ at steady state (t = 180 s). The dashed circle represents the illuminated area, with a diameter of 6D ≈ mm. Scale bar: (c) 2.5 mm. 26 Figure 3 Effect of visible light illumination on surface -- ice adhesion. (a) Schematic of the setup used for measuring ice adhesion: 1, visible light illumination path, same as in the infrared tem- perature measurements; 2, glass window; 3, piezo-stage and sample holder; 4, sample; 5, non- wetting ice cylinder, with an inner radius of R = 1.5 mm; 6, piezoelectric force sensor (0 -- 2 N, in- house calibration) and force transfer pin; 7, cold nitrogen vapor inlet. (b) De-icing curve (shear stress, yxτ , vs. time, t ) of the sample with 37%= ( -- -- -- ). At t = the sample was illuminated. 0 The corresponding yxτ vs. t of a control sample ( -- -- -- ) is also shown. The shaded areas show the minimum and maximum of the experimental measurements. (c) Boxplots of de-icing time, dt (time from maximum yx F τ ≈ max / ( R π )2 to noise level), vs. mean absorption, , of the metasur- faces. The substrate was fused silica. 27 Figure 4 Nucleation temperature of illuminated droplets. (a) Schematic of the environmental chamber used: 1, visible light illumination path; 2, glass window; 3, sample holder; 4, sample; 5, cold nitrogen vapor inlet; 6, CMOS camera. Two temperature sensors measured the gas ( 1T ) and sample ( 2T ) temperature. The illumination power density was P ≈ 2.4 kW m-2. Side-view image sequences of water droplets on a (b) control and (c) metasurface ( 37%= ) cooled down at a rate of 1≈ °C min-1; the final frames are when the droplets spontaneously nucleated and the sec- ond stage of freezing was progressing. The chamber gas temperature, 1T , is also shown. (d) Cali- bration curve: metasurface temperature, 2T , vs. gas temperature, 1T , in the case of the control ( -- -- -- ) and metasurface ( -- -- -- ). (e) Gas temperature at the moment of freezing, 1T , vs. sample type * (control and metasurface). The substrate was PMMA. Scale bar: (b) -- (c) 3 mm. 28 Figure 5 Light-induced defrosting. Angled-view image sequences of frosted PVDF-coated (a) control and (b) metasurface ( 37%= ) samples that were illuminated with a halogen lamp ( P ≈ 2.4 kW m-2) for 0 t< < 600 s. At t = s and 0 t = 600 s, the samples were in ambient light conditions, revealing the frost before and after illumination. The chamber gas temperature, 1T , is also shown. The substrate was PMMA. Camera tilt angle was 25≈ °. Scale bar: (a) -- (b) 4 mm. 29
1910.08698	1	1910	2019-10-19T04:12:05	Comparison studies between BHBT2:PC71BM composite nanotubes and its bulk-heterojunction properties	[ "physics.app-ph" ]	In this study, we investigate the morphological, structural, and optical properties of both bulk-heterojunction and composite nanotubes that composed of thiophene-based small molecules BHBT2 and fullerene PC71BM. Mix-blended and template-assisted methods were used to fabricate the bulk-heterojunction and composite nanotubes, respectively. A single material of BHBT2 thin films and nanotubes that fabricated via spin-coating and template-assisted methods, respectively, was also studied. Two different formations between bulk-heterojunction and composite nanotubes were compared to elaborate their advancement in properties. Absorption spectra of BHBT2:PC71BM bulk-heterojunction and composite nanotubes have fallen within the ultraviolet-visible (UV-vis) range. Field emission scanning electron microscope (FESEM) and high-resolution transmission electron microscope (HRTEM) images show that the carbon-rich of PC71BM, has successfully infiltrated into the BHBT2 nanotube to form the BHBT2:PC71BM composite nanotubes. However, due to the informality distribution of infiltration, charge carrier transfer is seen to be better in bulk-heterojunction rather than in the composite nanotubes.	physics.app-ph	physics	Comparison studies between BHBT2:PC71BM composite nanotubes and its bulk-heterojunction properties Muhamad Doris1, Azzuliani Supangat1, Teh Chin Hoong1, Khaulah Sulaiman1 and Rusli Daik2 Abstract In this study, we investigate the morphological, structural, and optical properties of both bulk-heterojunction and composite nanotubes that composed of thiophene-based small molecules 1,4-bis (2,2'-bithiopen-5-yl)2,5-dihexyloxybenzene (BHBT2) and fullerene [6,6]- phenyl C71 butyric acid methyl ester (PC71BM). Mix-blended and template-assisted methods were used to fabricate the bulk-heterojunction and composite nanotubes, respectively. A single material of BHBT2 thin films and nanotubes that fabricated via spin-coating and template-assisted methods, respectively, was also studied. Two different formations between bulk-heterojunction and composite nanotubes were compared to elaborate their advancement in properties. Absorption spectra of BHBT2:PC71BM bulk-heterojunction and composite nanotubes have fallen within the ultraviolet-visible (UV-vis) range. Field emission scanning electron microscope (FESEM) and high-resolution transmission electron microscope (HRTEM) images show that the carbon-rich of PC71BM, has successfully infiltrated into the BHBT2 nanotube to form the BHBT2:PC71BM composite nanotubes. However, due to the informality distribution of infiltration, charge carrier transfer is seen to be better in bulk- heterojunction rather than in the composite nanotubes. Keywords: Thiophene, alumina template, bulk-heterojunction, composite nanotube Corresponding author: [email protected] 1 Low Dimensional Materials Research Centre, Department of Physics University of Malaya, Kuala Lumpur 50603, Malaysia 2 School of Chemical Sciences and Food Technology, Faculty of Science and Technology, Universiti Kebangsaan Malaysia, 43600 UKM Bangi, Selangor, Malaysia 1 Background Organic materials have been the main interest to many of researchers due to their remarkable properties such as flexibility, abundancy of resources, and low production cost. Initially, the organic materials are only used as the electric insulator of electronic devices. However, this practice has changed after the discovery of metallic behavior of polyacetylene (PA) by Shirakawa et.al [1]. The finding has uncovered the emergence of others conducting organic materials such as polythiophene (Pth), polypyrrole (PPY), polyaniline (PANI), polycarbazole (PCz), polyparaphenylene vinylene (PPV), poly(3,4-ethylene dioxythiophene) (PEDOT), and polyfuran (PF) [2-9] that to date have been applied to numerous electronic devices. In addition, many others new organic materials are recently found and applied in electronic nanodevices (field effect transistor, light emitting diode) [10, 11], sensors (chemical, gas, optical, biosensor) [12-15], energy storage (solar cells, fuel cells, supercapacitors) [16-18], microwave absorption and electromagnetic shielding [19], and biomedical applications (drug delivery, protein purification, tissue engineering, neural interfaces, actuators) [20-24]. However, the small molecules based materials have the promising properties in terms of the viscoelasticity, toughness, crystallinity, and optical if compared to the polymer based materials [25-30]. Thiophene is a heterocyclic compound that consists of five-membered ring whose properties are still reliable and remarkable to date. Thiophene-based small molecules have shown extraordinary properties that may compete with its polymer in terms of enhancing efficiency to the perovskite-based solar cells and small molecules organic solar cells [28, 31]. Some applications such as small molecules organic solar cells [28, 30-32], inorganic-organic hybrid solar cells [33], and field effect transistor [34-36] have been realized from the thiophene-based small molecules. Therefore, the intensive studies of thiophene-based small molecules are as crucial as its thiophene-conjugated polymer. Thiophene-based small molecules, namely 1,4-bis (2,2'-bithiopen-5-yl)2,5- 2 dihexyloxybenzene (BHBT2) is a pentamer that consists of four thiophene rings and one benzene ring in which the benzene ring binds two hexyl molecules and the thiophene. BHBT2 has a profound solubility particularly in organic solvents such as chloroform, dichloromethane, and tetrahydrofuran. In addition, it is easily purified through a column chromatography and recrystallization techniques. The terminal bithiophene groups in BHBT2 pentamer is expected to provide a good stability and excellent charge transport properties in the pentamer backbone. Therefore, coupling of dihexyloxy-p-phenylene moiety with terminal bithiophene groups could improve the pi-electron conjugation path in the pentamer backbone without affecting the optical and electrochemical properties. One of the crucial explorations of BHBT2 properties is its nanostructures. Nanostructures of small molecules such as nanotubes, nanorods, nanowires, and nanoflowers have been commonly fabricated using template assisted-method [37-41]. The template is based on alumina oxide material that consists of orderly nanopores to infiltrate a solution, which in turn molds the nanostructured materials. Formation of nanostructures is important to be studied due to its exceptional properties over the planar structures. For instance, an active material in organic solar cells that fabricated into nanostructures with size corresponds to visible wavelength can enhance optical performance of the material [42]. In addition, nanostructuring an active material of solar cells into nanorods, nanotubes, and nanowires increases the surface area of active layer in absorbing more photons that results in higher efficiency [43, 44]. The template assisted-method is a low cost technique to produce nanostructured materials that the properties of the materials can easily be controlled [37, 45, 46]. The types of nanostructures are affected by the parameters of solution that infiltrate into the template's pores. Molecular weight, solution viscosity, solution concentration, infiltration mechanism, and drying process are some of the parameters that influence the formation of nanostructures [39, 40]. Template-assisted method can be used to fabricate the donor- 3 acceptor system of p-n junction composite nanotubes [47]. Unlike the well-known donor- acceptor system of bulk-heterojunction, the composite nanotubes that compose of ordered array of p-n core-shell junction could accommodate to the higher carriers mobility. p-n core- shell junction is widely used in applications such as in drug delivery and optoelectronic [48, 49]. Reported previously, a p-n core-shell junction nanowire that was applied to inorganic solar cells have improved the performance of device efficiency [44]. On top of that, p-n core- shell junction of composite nanotubes could also be able to enhance the charge carrier transfer if compared with its bulk-heterojunction [47]. In this study, the p-n core-shell junction of composite nanotubes has been successfully produced via template assisted-method. Composite nanotubes of BHBT2 and fullerene [6,6]- phenyl C71 butyric acid methyl ester (PC71BM) are fabricated and characterized. Characterizations are particularly emphasized on the optical, morphological, and structural properties of BHBT2:PC71BM. Two different formations of bulk-heterojunction and composite nanotubes of BHBT2:PC71BM are studied. To the best of our knowledge, there have been no studies that investigate the properties of thiophene-based small molecules BHBT2. Therefore, the studies on the p-n BHBT2:PC71BM properties may provide the informative and useful knowledge. 4 Methods Thiophene-based small molecules, 1,4-bis (2,2'-bithiophene-5-yl)2,5-dihexyloxybenzene (BHBT2) was synthesized and used directly as desired. As shown in Figure 1a, BHBT2 contains of single benzene with four thiophene ring flank the benzene ring on each side. 5 mg of BHBT2 was dissolved into the 1 ml of chloroform in which the solubility limit of BHBT2 within chloroform is 59 mg ml-1 [50]. Similar concentration of 5 mg/ml was applied for [6,6]- phenyl C71 butyric acid methyl ester (PC71BM), by dissolving in chloroform. Volume ratio of 1:1 was used to produce the BHBT2:PC71BM bulk-heterojunction and composite nanotubes. Porous alumina template with 200 nm and 60 µm of pores diameter and thickness, respectively, was purchased from Whatman Anodisc and was utilized to fabricate nanostructures. Prior to the infiltration, templates were cleaned up by means of immerse it into acetone under sonication for 15 min, and then rinsed using deionized water. Three different spin coating rates of 1000, 2000, and 3000 rpm were used in 30 s. Prior to the spin coating process, template was firstly attached onto the glass slide by using scotch tape to its right and left side (Figure 1b). The scotch tape was used to hold the template to be stuck on the glass slide during the spin coating process. Figure 1c shows the infiltrated BHBT2 after and before the spin coating and dissolution process, respectively. Figure 2 represents the schematic illustrations on the formation of BHBT2 nanotubes and BHBT2:PC71BM composite nanotubes. Porous alumina template was firstly cleaned up under sonication of acetone (i). Prior to the spin coating process, 50 µL of BHBT2 solution was dropped onto the cleaned template, which then allow the solution to infiltrate into the template (ii). To fabricate the BHBT2:PC71BM composite nanotubes, 50 µL of PC71BM were dropped on top of the infiltrated BHBT2 followed by the spin coating process (iii). Templates with infiltrated materials were then dried under the room temperature. These templates were stuck upside down on a copper tape (iv) before the dissolution of 6 h in 5 M of sodium 5 hydroxide (NaOH) was taken place (v). In order to fully wash out the remaining template, deionized water was used for rinse for several times. Finally, the obtained nanotubes that remain stuck on the copper tape are ready to be characterized (vi). Schematic illustrations shown in Figure 2 were adapted from the sample preparation set up shown in Figure 3. It is clearly seen that the infiltrated BHBT2 shown a yellow appearance (Figure 3a). BHBT2 nanotubes were be able to retain their adhesion on the copper tape although the sample was washed for several times (Figure 3b). The colour of BHBT2:PC71BM composite nanotubes turned brownish due to the infiltration of PC71BM (Figure 3c). As portrayed by BHBT2 nanotubes, BHBT2:PC71BM composite nanotubes were also be able to stick upside down on copper tape after several washing (Figure 3d). Several equipment such as spin coater model WS-650MZ-23NPP (Laurell Technologies Corp., North Wales, PA, USA), Field Emission Scanning Electron Microscope (FESEM) (Quanta FEG 450), High Resolution Transmission Electron Microscope (HRTEM) (Tecnai G2 FEI), Raman and photoluminescence spectroscopy (RENISHAW), UV-vis spectroscopy (Shimadzu UV-3101PC) and X-ray Diffraction Spectroscopy (XRD) were used in this studies. 6 Results and discussion In this study, investigation on the properties of single material and composite materials that consist BHBT2 (p-type) and fullerene (n-type) are reported. BHBT2 (1,4-bis (2,2'- bithiophene-5-yl)2,5-dihexyloxybenzene) is a novel small molecules with pentamer [50] with the advantage of high soluble with most solvents. By investigating its optical, morphological and structural properties, new applications within the organic electronics devices can be realized to improve the device performance. Based on the X-ray Diffraction (XRD) measurement of BHBT2 shown in Figure 4a, multiple peaks that represent the BHBT2 pristine indicate that the BHBT2 small molecule is a crystalline material [51]. Crystalline material with periodic structure may provide a light management that can outperform devices, for instance, enhancement of photon absorption [52-54]. To elaborate the properties of BHBT2 small molecules furthermore, incorporation between BHBT2 and p-type material of PC71BM is applied. Nanostructuring these materials into p-n composite may enrich knowledge in nano-morphology studies. In addition, the charge transfer between donor and acceptor materials can be more understand and digest in providing information on the carriers behaviors. Figure 4b shows the energy diagram of BHBT2 and PC71BM with HOMO-LUMO of 1.96-4.65 eV and 3.94-5.93 eV, respectively. HOMO-LUMO of donor-acceptor will enable the exciton (electron-hole) to dissociate at the interface [55]. Optical properties studies Figure 5a and Figure 5b show the absorption range of BHBT2 thin films and nanotubes spin- coated at the different rates of 1000, 2000, and 3000 rpm, respectively. Similar pattern of absorption is shown with the dissimilarity in absorption intensity is only different. This can be understood that cause of the spin coating rate is matter to the thickness of material in which the thicker layer will attenuate the penetrated photon. From the UV-vis absorption 7 spectra, BHBT2 thin films and nanotubes have portrayed five significant absorption peaks. There is no single peak shifted has taken place, due to the different spin coating rates, apart from the changes of absorption intensity of BHBT2 nanotubes that get higher at 350 nm if compared to its thin films. It is noticed that BHBT2 is favorably absorb photon only in the range of UV and visible light region. Due to its light absorption properties, BHBT2 may potentially be applied as UV-vis photodetector or a booster material of active layer in organic photovoltaic applications. Fabricating this material into highly ordered nanostructures may enhance its performance since the charge carriers transfer mobility can be improved via the highly ordered structures [56, 57]. As shown in Figure 6a, BHBT2 nanotubes reveal higher absorption intensity in comparison to their thin films [56]. BHBT2 nanotubes and thin films exhibit four shoulders with their significant peaks occur at 335, 350, 414, 429, and 461 nm. The longer absorption wavelength of both BHBT2 nanotubes and thin films are observed at 461 nm of Soret peak (B-band). This condition occur due to movement of an electron dipole that corresponds to nonbonding-antibonding (n-π) excitation among BHBT2 molecules. A small part of UV light that disclosed at around 320 - 350 nm is attributed to the bonding- antibonding (π -π) excitation. Generally, nanostructuring the BHBT2 either in single material or composite has increased its absorption performance. Incorporating BHBT2 with PC71BM slightly shift the intense absorption and render the absorption range at 320-350 nm and 400-460 nm become broader. As shown in Figure 6b, although the presence of π -π* transition is observed in the BHBT2 bulk-heterojunction, the better absorption is exhibited in the BHBT2 composite nanotubes. In an organic material, once a photon bombards onto a donor material, exciton which is a pair of electron-hole that bound with each other will be generated. In order to generate current, this exciton will need to be dissociated by dislodging the electron from its binding energy. One condition that has to be fulfilled is the affinity electron requirement of each 8 donor and acceptor materials. In our study, to investigate the potential application of BHBT2 into device, BHBT2 and electron acceptor of PC71BM is mixed. The HOMO-LUMO configuration of both materials should be possible to create a charge transfer phenomena since the electron affinity of acceptor material is bigger than the donor. Charge carriers transfer will only occur when the binding energy of exciton is lower than the electron affinity energy difference between donor and acceptor material [58]. Photoluminescence (PL) measurement is done to investigate the effectiveness of charge transfer between donor and acceptor. On the other words, photoluminescence provide information on how well can the exciton to reach donor-acceptor interfaces [59]. Exciton (pair of electron-hole) that has been generated by the donor material needs to be separated for the current extraction. Recombination of exciton will lead to the emission of radiative photon (radiative recombination) which is shown as intensity in the photoluminescence spectra [60]. Excitons that have successfully reached the donor-acceptor interfaces will be dissociated into hole and electron. This phenomenon can be indicated by the lower intensity exhibited by the photoluminescence peak (quenching) compared to the curve of radiative recombination which always exhibit the higher intensity. Quenching is attributed to the charge carriers transfer occurred between donor and acceptor that carriers are produced from the dissociated excitons at the donor-acceptor interfaces [59, 60]. As shown in Figure 7a, the quench phenomenon is occurred due to the incorporation between BHBT2 and PC71BM, which allowed electrons from BHBT2 to jump to the acceptor material. It is hardly to obtain the photoluminescence results for thin films that were spin-coated at 1000, 2000, and 3000 rpm with concentration of 5 mg/ml due to the thickness matter. To solve this hindrance, drop- casting of solution onto a glass substrate is applied in order to get the feasible films thickness. It is noticed that BHBT2 thin films has a higher photoluminescence intensity compared to the BHBT2 bulk-heterojunction and composite nanotubes. Mixing of two materials in the 9 formation of bulk-heterojunction and composite nanotubes has gained almost totally quenching. Figure 7b shows the photoluminescence spectra of BHBT2 bulk-heterojunction and composite nanotubes. BHBT2 bulk-heterojunction gets slightly quenched compared to its composite nanotubes. Charge transfer between BHBT2 and PC71BM in the formation of bulk- heterojunction is more effective than the composite nanotubes. Base on the morphology view of bulk-heterojunction, donor and acceptor will easily agglomerate and mix together in order to allow the electron to reach the donor-acceptor interface. Better quenching of BHBT2:PC71BM bulk-heterojunction has a contradict result with what has been found by others [45, 47] where nanostructuring a material gets better quenching than bulk- heterojunction. Morphology of BHBT2:PC71BM composite nanotubes could be responsible for the better quenching of bulk-heterojunction where it could be predicted that most of interfaces of BHBT2:PC71BM composite nanotubes are unevenly constructed. Structural properties studies Raman spectroscopy is applied to study the chemical and structural composition of materials. In its application, incident photons that interact with materials may lose or gain energy. The energy difference between the scattered photon and the incident photon is exactly similar to energy difference in molecular vibration. Therefore, the patterns of Raman spectra are generated from the molecular or lattice vibrations within the materials. A low frequency in Raman shift corresponds to a low energy vibration of atoms which means that heavy atoms are held together with the weak bonds. On the other hand, a high frequency is corresponded to light atoms which held together with strong bound [61]. Figure 8a and 8b show the Raman spectra of BHBT2 thin films versus nanotubes and Raman spectra of BHBT2:PC71BM bulk- heterojunction versus composite nanotubes, respectively, with their Raman peaks are tabulated in Table 1. Raman peak at 1445 cm-1 shows a slightly different in intensity between BHBT2 thin films and nanotubes indicating that thiophenes ring are dominantly stretching in 10 thin film compared to its nanotubes. The occurrence of thiophenes ring stretching supports the existence of BHBT2molecular structure with four thiophene rings is attached to one benzene ring. However, the incorporation of two materials has resulted shifting around 3 cm-1 to the lower frequency. This shifting could be due to the effect of incidence energy that is divided and served to the two different molecules, which in turn decreasing in scattered energy [61]. Most of the Raman peaks of BHBT2 nanotubes are shifted to higher frequencies except for the peak at 1564 cm-1 that has shifted to lower frequency for 2 cm-1. However, there is no CH deformation occurred in BHBT2 nanotubes. Ring breathing and ring vibration para-substituted benzene are formed from the incorporation of BHBT2 and PC71BM. These rings are found at 1188 cm-1 and 1227 cm-1, respectively. BHBT2 composite nanotubes have shown shifting to the higher frequencies of 1189 cm-1, and 1231 cm-1. Among the four Raman spectra (BHBT2 thin films, BHBT nanotubes, BHBT2:PC71BM bulk heterojunction, and BHBT2:PC71BM composite nanotubes), BHBT nanotubes and BHBT2:PC71BM composite nanotubes have somewhat a similar intensity although the highest peak intensity is dominated by BHBT2 thin films. Formation of nanostructured composite BHBT2 nanotubes have been synthesized via the template-assisted method. Figure 9a-c show the FESEM images of BHBT2 nanotubes obtained from the three different spin coating rates of 1000, 2000, and 3000 rpm. Generally, the nanostructures create bundles of nanotubes by collapsing their tips with each other instead of grown aligned as a single nanotube. Formation of nanotubes bundles could be due to the presence of attractive forces (van der Waals interactions) between the nanotubes [40]. At spin coating rate of 1000 rpm, the morphology of BHBT2 nanotubes are produced inconsistently with some of the tubes are longer than others. However, with the increase of spin coating rate to 2000 rpm, the homogeneous growth 11 of BHBT2 nanotubes is attained. Further increase to 3000 rpm, has caused to the thicker base layer and denser nanotubes. Due to the thicker base layer, observation on the nanotubes bundles becomes very intricate as the base layer almost covered the nanotubes' structure. In the further investigation, spin coating rate at 2000 rpm is considered as an optimum parameter for the fabrication of BHBT2:PC71BM composites. During the infiltration, solution that passes through the template nanochannels may experience two possible conditions along the process. The first possible condition is the solution will spread and wet over the nanochannels' wall which in turn produce the hollow nanotubes after the template dissolution. The other possible condition is the formation of nanorods, that due to the existence of force interaction between the molecules (solution) during infiltration which stronger than the adhesive force (wall). In this study, the growth mechanism of BHBT2 nanostructures is portrayed by the first possible condition (Figure 10 a- d). Since the optimum spin coating rate is 2000 rpm (BHBT2 nanotubes), BHBT2:PC71BM composite nanotubes are then produced at this optimum rate. Observation of hollow structure has supported the prediction of wall wetting and force interaction between the wall and solution. The wall of BHBT2:PC71BM composite nanotubes are thicker than the BHBT2 nanotubes due to the infiltration of two different materials (BHBT2 and PC71BM). Figure 11a and 11b show the HRTEM images of BHBT2 nanotubes and BHBT2:PC71BM composite nanotubes, respectively. BHBT2 nanotubes contain only single wall of tube whereas BHBT2:PC71BM composite nanotubes exhibit two different regions. These two different regions are corresponded to the PC71BM (inner) and BHBT2 (outer), respectively. PC71BM is infiltrated into the center of hollow BHBT2 which is shown as a darker region. PC71BM of carbon rich materials has successfully infiltrated into the BHBT2 nanotubes and wetted the inner wall of nanotubes which have led to the formation of p-n 12 junction. If assumption that all of the PC71BM has totally infiltrated into the BHBT2 nanotubes is made, more effective charge transfer will be taken place in composite nanotubes rather than in the bulk-heterojunction. However, based on the morphological images, it is notice that not many PC71BM has been infiltrated into the BHBT2 nanotubes. In addition to that, the size of nanotubes is not homogeneously constructed. The occurrence of this phenomenon may be due to the spin coating rate. In the spin coating process, some of the substances may have been swept away out of the alumina template surface. Ideally, PC71BM substances would have to fully infiltrate the BHBT2 nanotubes in order to form the homogeneous composite nanotubes and to have a well-mixed of donor-acceptor interfaces. Therefore, the composite nanotubes would provide more effective way for the charges to be transferred into the acceptor material. Conclusions We have successfully synthesized and characterized the properties of 1,4-bis (2,2'-bithiopen- 5-yl)2,5-dihexyloxybenzene (BHBT2) as single material and composites. The comparison studies between BHBT2 thin films, BHBT2 nanotubes, BHBT2:PC71BM bulk-heterojunction and BHBT2:PC71BM composite nanotubes were emphasized on their optical, structural and morphological properties. Nanostructuring the BHBT2 via template-assisted method has been successfully applied to produce nanotubes and composite nanotubes. Better quenching and improvement of charge carriers' transport is observed in BHBT2:PC71BM bulk- heterojunction due to the poor interfaces morphology of unevenly constructed BHBT2:PC71BM composite nanotubes. However, the enhancement of light absorption is observed in BHBT2 nanotubes and BHBT2:PC71BM composite nanotubes. PC71BM has been successfully infiltrated into the BHBT2 nanotubes due to the wetting properties possessed by both materials, although the uniformity of infiltration is poor. 13 Abbreviations BHBT2, 1,4-bis (2,2'-bithiopen-5-yl)2,5-dihexyloxybenzene; FESEM, field emission scanning electron microscopy; HOMO, highest occupied molecular orbital; HRTEM, high resolution transmission electron microscopy; LUMO, lowest unoccupied molecular orbital; NaOH, sodium hydroxide; PA, polyacetylene; PANI, polyaniline; PC71BM, [6,6]-phenyl C71- butyric acid methyl ester; PCz, polycarbazole; PEDOT, poly(3,4-ethylene dioxythiophene); PF, polyfuran; PL, photoluminescence: Pth, polythiophene; PPV, polyparaphenylene vinylene; PPY, polypyrrole; XRD, X-ray Diffraction Spectroscopy Competing interests The authors declare that they have no competing interests. Authors' contributions MD carried out the experiments, performed the analysis, and drafted the manuscript. AS participated in the design of the study, performed the analysis, and helped draft the manuscript. TCH carried out the experiments, and performed the analysis. KS and RD participated in the design of study and in the sequence alignment. Authors' information MD is a research assistant and applying for postgraduate studies at the University of Malaya. TCH is currently a postdoctoral fellowship at the University of Malaya. AS and KS is the senior lecturer and associate professor at the Department of Physics, University of Malaya, respectively, while RD is a professor at the National University of Malaysia. 14 Acknowledgements The authors would like to acknowledge the University of Malaya for the project funding under the University of Malaya High Impact Research Grant UM-MoE (UM.S/625/3/HIR/MoE/SC/26), University Malaya Research Grant (RG283-14AFR), and the Ministry of Education Malaysia for the project funding under Fundamental Research Grant Scheme (FP002-2013A). References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. Shirakawa H, Louis EJ, MacDiarmid AG, Chiang CK, Heeger AJ: Synthesis of electrically conducting organic polymers: halogen derivatives of polyacetylene, (CH). Journal of the Chemical Society, Chemical Communications 1977, 16:578-580. Kirchmeyer S, Reuter K: Scientific importance, properties and growing applications of poly(3,4-ethylenedioxythiophene). Journal of Materials Chemistry 2005, 15(21):2077-2088. Booth MA, Leveneur J, Costa AS, Kennedy J, Travas-Sejdic J: Tailoring the Conductivity of Polypyrrole Films Using Low-Energy Platinum Ion Implantation. The Journal of Physical Chemistry C 2012, 116(14):8236-8242. Ćirić-Marjanović G: Recent advances in polyaniline research: Polymerization mechanisms, structural aspects, properties and applications. Synthetic Metals 2013, 177(0):1-47. Chen Q, Liu DP, Zhu JH, Han BH: Mesoporous Conjugated Polycarbazole with High Porosity via Structure Tuning. Macromolecules 2014, 47(17):5926-5931. p-Type doping of Bajpai M, Srivastava R, Dhar R, Tiwari RS, Chand S: tetrafluorotetracynoquinodimethane (F4TCNQ) in poly(para-phenylene vinylene) (PPV) derivative "Super Yellow" (SY). RSC Advances 2014, 4(88):47899-47905. Sreeram A, Patel NG, Venkatanarayanan RI, McLaughlin JB, DeLuca SJ, Yuya PA, Krishnan S: Nanomechanical properties of poly(para-phenylene vinylene) determined using quasi-static and dynamic nanoindentation. Polymer Testing 2014, 37(0):86-93. Atanasov SE, Losego MD, Gong B, Sachet E, Maria J-P, Williams PS, Parsons GN: Highly Conductive and Conformal Poly(3,4-ethylenedioxythiophene) (PEDOT) Thin Films via Oxidative Molecular Layer Deposition. Chemistry of Materials 2014, 26(11):3471-3478. Sheberla D, Patra S, Wijsboom YH, Sharma S, Sheynin Y, Haj-Yahia A-E, Barak AH, Gidron O, Bendikov M: Conducting polyfurans by electropolymerization of oligofurans. Chemical Science 2015, 6(1):360-371. Lee SY, Choi GR, Lim H, Lee KM, Lee SK: Electronic transport characteristics of electrolyte- gated conducting polyaniline nanowire field-effect transistors. Applied Physics Letters 2009, 95(1):013113. Grimsdale AC, Leok Chan K, Martin RE, Jokisz PG, Holmes AB: Synthesis of Light-Emitting Conjugated Polymers for Applications in Electroluminescent Devices. Chemical Reviews 2009, 109(3):897-1091. Rajesh, Ahuja T, Kumar D: Recent progress in the development of nano-structured conducting polymers/nanocomposites for sensor applications. Sensors and Actuators B: Chemical 2009, 136(1):275-286. 15 13. Menegazzo N, Herbert B, Banerji S, Booksh KS: Discourse on the utilization of polyaniline coatings for surface plasmon resonance sensing of ammonia vapor. Talanta 2011, 85(3):1369-1375. 14. Michira I, Akinyeye R, Baker P, Iwuoha E: Synthesis and Characterization of Sulfonated Polyanilines and Application in Construction of a Diazinon Biosensor. International Journal of Polymeric Materials and Polymeric Biomaterials 2011, 60(7):469-489. 15. Wang X, Shao M, Shao G, Fu Y, Wang S: Reversible and efficient photocurrent switching of ultra-long polypyrrole nanowires. Synthetic Metals 2009, 159(3 -- 4):273-276. Fischer FSU, Trefz D, Back J, Kayunkid N, Tornow B, Albrecht S, Yager KG, Singh G, Karim A, Neher D, Brinkmann M, Ludwigs S: Highly Crystalline Films of PCPDTBT with Branched Side Chains by Solvent Vapor Crystallization: Influence on Opto-Electronic Properties. Advanced Materials 2015, 27(7):1223-1228. 16. 22. 20. 18. 19. 21. 17. Ma Y, Jiang S, Jian G, Tao H, Yu L, Wang X, Wang X, Zhu J, Hu Z, Chen Y: CNx nanofibers converted from polypyrrole nanowires as platinum support for methanol oxidation. Energy & Environmental Science 2009, 2(2):224-229. Chen L, Yuan C, Dou H, Gao B, Chen S, Zhang X: Synthesis and electrochemical capacitance of core -- shell poly (3,4-ethylenedioxythiophene)/poly (sodium 4-styrenesulfonate)-modified multiwalled carbon nanotube nanocomposites. Electrochimica Acta 2009, 54(8):2335-2341. Saini P, Choudhary V, Singh BP, Mathur RB, Dhawan SK: Polyaniline -- MWCNT nanocomposites for microwave absorption and EMI shielding. Materials Chemistry and Physics 2009, 113(2 -- 3):919-926. Kim S, Kim J-H, Jeon O, Kwon IC, Park K: Engineered polymers for advanced drug delivery. European Journal of Pharmaceutics and Biopharmaceutics 2009, 71(3):420-430. Zhu Y, Li J, Wan M, Jiang L: Superhydrophobic 3D Microstructures Assembled From 1D Nanofibers of Polyaniline. Macromolecular Rapid Communications 2008, 29(3):239-243. Srivastava S, Chakraborty A, Salunke R, Roy P: Development of a Novel Polygalacturonic Acid- Gelatin Blend Scaffold Fabrication and Biocompatibility Studies for Tissue-Engineering Applications. International Journal of Polymeric Materials and Polymeric Biomaterials 2012, 61(9):679-698. Kang G, Borgens RB, Cho Y: Well-Ordered Porous Conductive Polypyrrole as a New Platform for Neural Interfaces. Langmuir 2011, 27(10):6179-6184. 24. Otero TF, Sanchez JJ, Martinez JG: Biomimetic Dual Sensing-Actuators Based on Conducting Polymers. Galvanostatic Theoretical Model for Actuators Sensing Temperature. The Journal of Physical Chemistry B 2012, 116(17):5279-5290. Fischer I, Kaeser A, Peters-Gumbs MAM, Schenning APHJ: Fluorescent π-Conjugated Polymer Dots versus Self-Assembled Small-Molecule Nanoparticles: What's the Difference?. Chemistry -- A European Journal 2013, 19(33):10928-10934. Liu Y, Chen C-C, Hong Z, Gao J, Yang YM, Zhou H, Dou L, Li G, Yang Y: Solution-processed small-molecule solar cells: breaking the 10% power conversion efficiency. Nature Scientific Reports 2013, 3:3356. Zhang Q, Kan B, Liu F, Long G, Wan X, Chen X, Zuo Y, Ni W, Zhang H, Li M, Hu Z, Huang F, Cao Y, Liang Z, Zhang M, Russell TP, Chen Y: Small-molecule solar cells with efficiency over 9%. Nat Photon 2015, 9(1):35-41. Li H, Fu K, Boix PP, Wong LH, Hagfeldt A, Grätzel M, Mhaisalkar SG, Grimsdale AC: Hole- Transporting Small Molecules Based on Thiophene Cores for High Efficiency Perovskite Solar Cells. ChemSusChem 2014, 7(12):3420-3425. 23. 25. 26. 27. 28. 29. Mishra A, Bäuerle P: Small Molecule Organic Semiconductors on the Move: Promises for Future Solar Energy Technology. Angewandte Chemie International Edition 2012, 51(9):2020- 2067. 16 30. Zhou J, Wan X, Liu Y, Zuo Y, Li Z, He G, Long G, Ni W, Li C, Su X, Chen Y: Small Molecules Based on Benzo[1,2-b:4,5-b′]dithiophene Unit for High-Performance Solution-Processed Organic Solar Cells. Journal of the American Chemical Society 2012, 134(39):16345-16351. 31. Malytskyi V, Simon JJ, Patrone L, Raimundo JM: Thiophene-based push-pull chromophores for small molecule organic solar cells (SMOSCs). RSC Advances 2015, 5(1):354-397. 33. 32. Montcada NF, Pelado B, Viterisi A, Albero J, Coro J, Cruz Pdl, Langa F, Palomares E: High open circuit voltage in efficient thiophene-based small molecule solution processed organic solar cells. Organic Electronics 2013, 14(11):2826-2832. Freitas FS, Clifford JN, Palomares E, Nogueira AF: Tailoring the interface using thiophene small molecules in TiO2/P3HT hybrid solar cells. Physical Chemistry Chemical Physics 2012, 14(34):11990-11993. Park JI, Chung JW, Kim JY, Lee J, Jung JY, Koo B, Lee BL, Lee SW, Jin YW, Lee SY: Dibenzothiopheno[6,5-b:6′,5′-f]thieno[3,2-b]thiophene (DBTTT): High-Performance Small- Molecule Organic Semiconductor for Field-Effect Transistors. Journal of the American Chemical Society 2015. doi:10.1021/jacs.5b01108 Sun J-P, Hendsbee AD, Eftaiha AaF, Macaulay C, Rutledge LR, Welch GC, Hill IG: Phthalimide- thiophene-based conjugated organic small molecules with high electron mobility. Journal of Materials Chemistry C 2014, 2(14):2612-2621. 34. 35. 39. 37. 36. Wang C, Zang Y, Qin Y, Zhang Q, Sun Y, Di C-a, Xu W, Zhu D: Thieno[3,2-b]thiophene- Diketopyrrolopyrrole-Based Quinoidal Small Molecules: Synthesis, Characterization, Redox Behavior, and n-Channel Organic Field-Effect Transistors. Chemistry -- A European Journal 2014, 20(42):13755-13761. Al-Kaysi RO, Ghaddar TH, Guirado G: Fabrication of one-dimensional organic nanostructures using anodic aluminum oxide templates. Journal of Nanomaterials 2009, Article ID 436375 Hu J, Shirai Y, Han L, Wakayama Y: Template method for fabricating interdigitate p-n heterojunction for organic solar cell. Nanoscale Res Lett 2012, 7(1):1-5. Fakir MS, Supangat A, Sulaiman K: Templated growth of PFO-DBT nanorod bundles by spin coating: effect of spin coating rate on the morphological, structural, and optical properties. Nanoscale Res Lett 2014, 9(1):1-7. Bakar NA, Supangat A, Sulaiman K: Elaboration of PCDTBT nanorods and nanoflowers for augmented morphological and optical properties. Materials Letters 2014, 131:27-30. Jung WS, Do YH, Kang MG, Kang CY: Energy harvester using PZT nanotubes fabricated by template-assisted method. Current Applied Physics 2013, 13, Supplement 2 (0):S131-S134. Edman Jonsson G, Fredriksson H, Sellappan R, Chakarov D: Nanostructures for Enhanced Light Absorption in Solar Energy Devices. International Journal of Photoenergy 2011, Article ID 939807. Yu K, Chen J: Enhancing solar cell efficiencies through 1-D nanostructures. Nanoscale Res Lett 2009, 4(1):1-10 44. Wang S, Yan X, Zhang X, Li J, Ren X: Axially connected nanowire core-shell pn junctions: a 42. 38. 40. 41. 43. 45. 46. composite structure for high-efficiency solar cells. Nanoscale Res Lett 2015, 10(1):1-7. Supangat A, Kamarundzaman A, Bakar NA, Sulaiman K, Zulfiqar H: P3HT: VOPcPhO composite nanorods arrays fabricated via template -- assisted method: Enhancement on the structural and optical properties. Materials Letters 2014, 118:103-106. Kamarundzaman A, Fakir MS, Supangat A, Sulaiman K, Zulfiqar H: Morphological and optical properties of hierarchical tubular VOPcPhO nanoflowers. Materials Letters 2013, 111(0):13- 16. 47. Makinudin AHA, Fakir MS, Supangat A: Metal phthalocyanine: fullerene composite nanotubes via templating method for enhanced properties. Nanoscale Res Lett 2015, 10(1):1-8 Chatterjee K, Sarkar S, Jagajjanani Rao K, Paria S: Core/shell nanoparticles in biomedical applications. Advances in Colloid and Interface Science 2014, 209(0):8-39. 48. 17 49. Tchoulfian P, Donatini F, Levy F, Dussaigne A, Ferret P, Pernot J: Direct Imaging of p -- n Junction in Core -- Shell GaN Wires. Nano letters 2014, 14(6):3491-3498 59. 53. 54. 55. 56. 51. 52. 50. Wei LL, Hoong TC, Sulaiman K, Daik R, NM S: Synthesis and Characterization of Solution Processable Organic Ultraviolet (UV) Photodetector based on 2,2'-Bithiophene End-Capped Dihexyloxy Phenylene Pentamer. (will be published elsewhere) Cullity BD: Elements of X-ray diffraction. 1978, Addison-Wesley, Philippines Battaglia C, Hsu C-M, Söderström K, Escarré J, Haug F-J, Charrière M, Boccard M, Despeisse M, Alexander DTL, Cantoni M, Cui Y, Ballif C: Light Trapping in Solar Cells: Can Periodic Beat Random? ACS Nano 2012, 6(3):2790-2797. Sai H, Koida T, Matsui T, Yoshida I, Saito K, Kondo M: Microcrystalline silicon solar cells with 10.5% efficiency realized by improved photon absorption via periodic textures and highly transparent conductive oxide. Applied Physics Express 2013, 6(10):104101. Sai H, Kondo M, Saito K: Periodic structures boost performance of thin-film solar cells. http://spie.org/x91297.xml. Accessed 20 April 2015 Kietzke T: Recent advances in organic solar cells. Advances in OptoElectronics 2007, Article ID 40285. Kumar B, Kim SW: Energy harvesting based on semiconducting piezoelectric ZnO nanostructures. Nano Energy 2012, 1(3):342-355. Liang D, Kang Y, Huo Y, Chen Y, Cui Y, Harris JS: High-efficiency nanostructured window GaAs solar cells. Nano letters 2013, 13(10):4850-4856 Abdulrazzaq OA, Saini V, Bourdo S, Dervishi E, Biris AS: Organic solar cells: a review of materials, limitations, and possibilities for improvement. Particulate Science and Technology 2013, 31(5):427-442. Cates NC, Gysel R, Beiley Z, Miller CE, Toney MF, Heeney M, McCulloch I, McGehee MD: Tuning the properties of polymer bulk heterojunction solar cells by adjusting fullerene size to control intercalation. Nano letters 2009, 9(12):4153-4157. Hoppe H, Niggemann M, Winder C, Kraut J, Hiesgen R, Hinsch A, Meissner D, Sariciftci NS: Nanoscale morphology of conjugated polymer/fullerene‐based bulk‐heterojunction solar cells. Advanced Functional Materials 2004, 14(10):1005-1011. Bright WE, Decius JC, Paul CC: Molecular vibrations: the theory of infrared and raman vibrational spectra. 1955, McGraw-Hill Book Company, Inc. Francis RD, William GF, Freeman FB: Characteristic Raman frequencies of organic compounds. 1973, John Wiley & Sons Inc. 60. 57. 58. 61. 62. 18 (a) (b) (c) Figure 1 (a) Molecular structure of BHBT2. (b) Illustration of porous alumina template being stuck on glass substrate. (c) Illustration of infiltrated BHBT2. Alumina template BHBT2 PC71BM Copper tape Porous alumina template Spin Coating (i) (ii) (iii) Stick onto copper tape upside down (iv) Dissolved in NaOH for 6 h (v) Nanotubes bundles (vi) Figure 2 Schematic illustrations on the formation of BHBT2 nanotubes and BHBT2:PC71BM composite nanotubes. BHBT2 nanotubes BHBT2:PC71BM composite 19 (a) (a) (b) (b) (c) (d) Figure 3 (a) Infiltrated BHBT2 that spin coated at three different rates of 1000, 2000 and 3000 rpm. (b) BHBT2 nanotubes stick upside down on copper tape. (c) BHBT2:PC71BM composite nanotubes before dissolution. (d) BHBT2:PC71BM composite nanotubes stick upside down on copper tape. (a) (b) Figure 4 (a) XRD measurement of pristine BHBT2. (b) Vacuum level (energy diagram) of BHBT2 and PC71BM. 20 (a) 1000 rpm 2000 rpm 3000 rpm 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 (b) 1000 rpm 2000 rpm 3000 rpm 1.50 1.25 1.00 0.75 0.50 0.25 0.00 . ) u a ( y t i s n e t n I . ) u a ( y t i s n e t n I 300 350 450 400 Wavelength (nm) 500 550 300 350 450 400 Wavelength (nm) 500 550 Figure 5 (a) UV-vis absorption spectra of BHBT2 thin films. (b) UV-vis absorption spectra of BHBT2 nanotubes. . ) u a ( e c n a b r o s b a y t i s n e t n I 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 (a) Nanotubes Thin Film 350 400 450 500 550 Wavelength (nm) . ) u a ( e c n a b r o s b a n y t i s n e t n I 3.0 2.5 2.0 1.5 1.0 0.5 0.0 (b) PC71BM Bulk Heterojunction 0.3 0.2 . ) u a ( y t i s n e n 0.1 t 400 500 600 Wavelength (nm) I 0.0 300 Composite nanotubes 350 400 450 500 550 Wavelength (nm) Figure 6 UV-vis absorption spectra of (a) BHBT2 thin films and nanotubes (b) BHBT2 composite nanotubes and BHBT2 bulk heterojunction, PC71BM thin films (inset). 21 ) u a . 3 0 1 x ( y t i s n e t n i L P (a) BHBT2 pristine Bulk Heterojunction Composite Nanotubes ) u a . 2 0 1 x ( y t i s n e t n I L P 14 12 10 8 6 4 2 0 400 500 600 700 800 900 Wavelength (nm) 10 9 8 7 6 5 4 3 2 1 0 -1 400 (b) Bulk Heterojunction Composite Nanotubes 500 600 700 800 900 Wavelength (nm) Figure 7 (a) Photoluminescence spectra of BHBT2 thin films, bulk heterojunction and composite nanotubes. (b) Photoluminescence comparison spectra between BHBT2 bulk heterojunction and composite nanotubes. ) u a . 3 0 1 x ( y t i s n e t n I ) u a . 3 0 1 x ( y t i s n e t n I Thin film Nanotubes 1200 1400 1800 1600 2000 Bulk Heterojunction PC71BM pristine Composite nanotubes (a) 1000 (b) 60 50 40 30 20 10 40 30 20 10 0 1000 1200 1400 1600 1800 2000 Raman shift (cm-1) Figure 8 (a) Raman spectra of BHBT2 thin films and composite nanotubes. (b) Raman spectra of BHBT2 bulk heterojunction, composite nanotubes and PC71BM. 22 (a) (b) 1 µm (c) 1 µm 1 µm Figure 9 FESEM images of BHBT2 nanotubes at spin coating rate of (a) 1000 rpm (b) 2000 rpm (c) 3000 rpm. 23 (a) (c) 1 µm (b) (d) 100 nm 100 nm 100 nm Figure 10 (a) and (b) FESEM images of BHBT2 nanotubes spin-coated at 2000 rpm. (c) and (d) FESEM images of BHBT2:PC71BM composite nanotubes spin-coated at 2000 rpm. Figure 11 HRTEM images of (a) BHBT2 nanotubes (b) BHBT2:PC71BM composite nanotubes. 24 Table 1 Raman peak positions of BHBT2 and BHBT2 : PC71BM [62] BHBT2 BHBT2 PC71BM Raman Shift Thin Film Nanotubes Bulk heterojunction Composites (at 2000 rpm) (at 2000 rpm) (at 2000 rpm) (at 2000 rpm) Assignments Nanotubes Vibrational 1067 1068 - - 1297 1324 1344 1445 1503 1544 1566 1601 - - 1299 1325 - 1446 1503 1544 1564 1602 1067 1188 1227 1295 1328 1342 1443 1507 - 1565 1603 C=S stretch Ethylene trithiocarbonate Ring "breathing" Ring vibration Para-disubstituted benzenes CC bridge bond stretch Ring vibration CH deformation Ring stretch 2- Substituted thiophenes Symmetric C=C stretch C=C stretch C=C stretch C=C stretch 1068 1189 1231 1299 1322 1338 1443 1505 - 1567 1604 25
1907.00346	1	1907	2019-06-30T09:41:50	Picosecond pulses from a mid-infrared interband cascade laser	[ "physics.app-ph", "physics.optics" ]	The generation of mid-infrared pulses in monolithic and electrically pumped devices is of great interest for mobile spectroscopic instruments. The gain dynamics of interband cascade lasers (ICL) are promising for mode-locked operation at low threshold currents. Here, we present conclusive evidence for the generation of picosecond pulses in ICLs via active mode-locking. At small modulation power, the ICL operates in a linearly chirped frequency comb regime characterized by strong frequency modulation. Upon increasing the modulation amplitude, the chirp decreases until broad pulses are formed. Careful tuning of the modulation frequency minimizes the remaining chirp and leads to the generation of 3.2 ps pulses.	physics.app-ph	physics	Picosecond pulses from a mid-infrared interband cascade laser Johannes Hillbrand1,∗, Maximilian Beiser1, Aaron Maxwell Andrews1,2, Hermann Detz1,3, Robert Weih4, Anne Schade5, Sven H ofling5,6, Gottfried Strasser1,2, Benedikt Schwarz1,+ 1Institute of Solid State Electronics, TU Wien, Gusshausstrae 25-25a, 1040 Vienna, Austria 2Center for Micro- and Nanostructures, TU Wien, Gusshausstrae 25-25a, 1040 Vienna, Austria 3CEITEC, Brno University of Technology, Brno, Czech Republic 4Nanoplus Nanosystems and Technologies GmbH, 97218 Gerbrunn, Germany 5Technische Physik, Physikalisches Institut, University Wrzburg, Wrzburg, Germany 6SUPA, School of Physics and Astronomy, University of St Andrews, St Andrews, KY16 9SS, United Kingdom ∗e-mail: [email protected] +e-mail: [email protected] The generation of mid-infrared pulses in monolithic and electrically pumped devices is of great interest for mobile spectroscopic instruments. The gain dynamics of inter- band cascade lasers (ICL) are promising for mode-locked operation at low threshold currents. Here, we present con- clusive evidence for the generation of picosecond pulses in ICLs via active mode-locking. At small modulation power, the ICL operates in a linearly chirped frequency comb regime characterized by strong frequency modulation. Upon in- creasing the modulation amplitude, the chirp decreases un- til broad pulses are formed. Careful tuning of the modula- tion frequency minimizes the remaining chirp and leads to the generation of 3.2 ps pulses. Optical frequency combs (OFC) operating in the mid- infrared (MIR) spectral range are a powerful spectroscopic tool 1. Measuring the molecular fingerprint in the MIR re- gion allows to identify chemical species and to determine their concentration. MIR frequency comb sources based on the non-linear conversion of near-infrared mode-locked lasers 2,3 and microresonators 4 have reached a high level of maturity featuring octave spanning spectra 5. Semicon- ductor laser OFCs 6 are advantageous in applications re- quiring compactness and low power consumption. Quan- tum cascade laser (QCL) frequency combs 7,8 are among the most investigated technologies. However, gain band- width and dispersion 9,10 limit their spectral bandwidth on the order of 100 cm−1. One way to overcome this issue is spectral broadening in an external non-linear fiber or waveguide. Recent results 11 revealed, however, that the temporal output of QCLs is strongly chirped accompanied by the suppression of amplitude modulation. In fact, the ultrafast gain dynamics of QCLs are believed to be highly unfavorable for the formation of light pulses 12. Previous attempts of mode-locking in monolithic QCLs were lim- ited to cryogenic temperatures and low peak powers 13. This makes non-linear techniques for spectral broadening very inefficient. First results to enter the mid-infrared from shorter wavelengths were demonstrated using pas- sively mode-locked GaSb-based type-I cascade diode lasers with 10 ps pulse duration around 3.25 µm. Type-II interband cascade lasers (ICL) are an interest- ing alternative that already cover a major part of the mid- infrared up to 6 µm 14,15,16. They combine the carrier injec- tion and extraction scheme of QCLs with the advantages of an interband lasing transition. Hence, the upper-state lifetime of the optical transition in ICLs is significantly longer than the cavity round-trip time. This enables low dissipation operation and has important consequences for mode-locking of ICLs. While the short upper-state life- time of QCLs prevents the formation of short pulses, this issue is not present in ICLs. Furthermore, the ICL active material can be switched to absorption at the laser wave- length, which was shown by using them as photodetector at zero-bias 17. Together with the fast carrier injection scheme, this allows the realization of efficient high-speed modulators with cut-off frequencies of several gigahertz 18. Hence, ICLs exhibit all required properties for efficient ac- tive mode-locking via modulation of the gain at the cav- Figure 1 -- a: Scanning electron microscope picture of the modulation section of the ICL. The modulation section (left) and a ground contact (right) are optimized for RF injection via RF tips. b: Light-Current- Voltage (L-I-V) characteristics of the ICL at room temperature for a homogeneously biased laser (blue line) as well as for 2.5 V absorber bias with (dotted red line) and without (solid red line) RF modulation at frep ≈ 10.15 GHz. The RF power is 31 dBm. 1 0200400600800Current density [A/cm2]01234Voltage [V]b01234Power [mW]L-I-V characteristicssame2.5V2.5V+RFa19. Recent efforts aimed at ity round-trip frequency frep the generation of OFCs via passive mode-locking of ICLs. However, the experimental results did not show the for- mation of pulses 20. Instead, such passive ICL frequency combs are characterized by a continuous output intensity with a strong frequency modulation 18, similarly to what was found in QCLs 8,11. In this letter, we report on the generation of picosecond pulses in two-section Fabry-P´erot ICLs. The dry etched laser ridges are 6 µm wide and split into a 3520 µm long gain section and a 480 µm long modulation section (Fig. 1a). The modulation section was designed to minimize parasitic capacitance and allow efficient RF injection via coplanar RF tips. A 1.5 µm thick Si3N4 passivation layer was used for the modulation section while keeping its top contact area as small as possible. The passivation layer of the gain section is thinner (250 nm) in order to improve the thermal performance of the laser. The back facet of the de- vice was high-reflection coated using Si3N4 and gold, while the front facet was left uncoated. The active region is com- prised of 6 stages and operates at 3.85 µm (2600 cm−1). At room temperature, the ICL emits up to 4.2 mW of optical power in continuous wave operation when both sections are biased homogeneously (Fig. 1b). When the bias of the modulation section is set to 2.5 V additional loss is added to the cavity causing the threshold current density to increase and the maximum output power decreases to 1.9 mW. The injection of an RF signal at frep into the modulation section reduces the threshold by about 20%, showing that the laser is strongly influenced by the active modulation. The characterization of the temporal output intensity of mid-infrared semiconductor lasers is challenging. Due Figure 2 -- a: SWIFTS characterization of the ICL at -2 dBm injected RF power. The gain section is operated at 770 A/cm2 and the modulation section at 2.5 V. blue line: intensity spectrum. red line: SWIFTS spec- trum. blue dots: SWIFTS amplitudes expected for full phase-coherence of the ICL comb. green dots: intermodal difference phases of adjancent comb lines. The right part shows a zoom-in on the center burst of the intensity and SWIFTS interferograms. b: reconstructed intensity and instantaneous wavenumber of the ICL frequency comb. 2 to the high repetition rate and the relatively low average power, the peak power is expected to be too low for es- tablished non-linear pulse characterization techniques 21. Instead, we employ a linear phase-sensitive autocorrela- tion technique called 'SWIFTS' 22. This method uses a Fourier transform infrared (FTIR) spectrometer and a fast quantum well infrared photodetector (QWIP) to measure the amplitudes and phases of the beatings between ad- jacent laser modes (details can be found in Ref. 22). In this way, SWIFTS allows the reconstruction of both the temporal intensity and instantaneous frequency of the ICL frequency comb. This method is not restricted to mode- locked operation and is valid for arbitrary periodic signals. Furthermore, it should be noted that recent experiments with a passively mode-locked quantum dot laser proved that SWIFTS and conventional intensity autocorrelation in a non-linear crystal are able to retrieve the same pulse width 23. At 2.5 V bias of the modulation section, the laser generates a narrow beatnote at the cavity round-trip fre- quency, which can be extracted directly from the laser cur- rent. This beatnote results from the beating of adjacent cavity modes. Its narrow linewidth on the kHz level indi- cates that the cavity modes are phase-locked. In order to provide a stable reference for SWIFTS, we inject a weak RF signal at -2 dBm into the modulation section. Pre- vious experiments showed that such a weak modulation is able to lock the frequency of the beatnote and leaves the spectral phases of the free-running OFC unchanged 24. The SWIFTS analysis of the ICL in this state is displayed in Fig. 2a. The intensity spectrum spans over roughly 28 cm−1 and consists of several lobes. The SWIFTS spec- trum is commensurate with the values expected for full phase-coherence (blue dots in Fig. 2a) over the entire span of the spectrum, which proves frequency comb operation. The intermodal difference phases ∆φ retrieved from the SWIFTS data decrease linearly over a range of exactly 2π. This particular frequency comb state was found in QCLs operating at 8 µm 11, ICLs at 4 µm 18 and quantum dot lasers at 1.25 µm 23 and appears to be universal in semicon- ductor laser OFCs. Recent theoretical work attributes its origin to the interplay of dispersion and the Kerr effect in lasers with spatial hole burning 25. Both SWIFTS interfer- ograms (Fig. 2a right) have a local minimum at zero-path difference, which indicates the suppression of amplitude modulation 8. Indeed, the reconstructed intensity (Fig. 2b top) does not show isolated pulses. In contrast, the in- stantaneous wavenumber is strongly modulated and lin- early chirps through the entire spectrum within a cavity round-trip period. In the following, we will investigate the influence of an increased modulation strength on the ICL frequency comb dynamics. Fig. 3 shows the detailed SWIFTS analysis of the ICL for injection power levels from 6 dBm to 36 dBm. The modulation frequency is altered slightly around frep to account for a small detuning of the laser round-trip fre- 259026002610Wavenr. (cm1)00101SWIFTSIntensityaSWIFTS101Delay (mm)SWIFTS XSWIFTS YIntensityCenter burst0.00.51.0Norm. powerbTime domain signal0100Time [ps]259026002610Wavenr. (cm1)Figure 3 -- SWIFTS analysis of the ICL frequency comb for 6 dBm (a), 18 dBm (c), 30 dBm (e) and 36 dBm (g) injected power. The wavenumber in the top right corner indicates the bandwidth of the spectrum. The gain section was operated at 800 A/cm2 and the modulation section bias was 2.5 V. The GDD is deduced from the slope of the intermodal difference phases (red dotted line). b,d,f,h: The corresponding reconstructed temporal intensity and instantaneous wavenumber. The shaded grey area indicates the fraction of phase range and of the cavity round-trip period, respectively, which is occupied by the pulses. quency with temperature due to the high injection power. At 6 dBm (Fig. 3a), the ICL still operates in a similar OFC state as in Fig. 2a and the reconstructed time sig- nal (Fig. 3b) does not show isolated pulses. When the injected power is further increased to 18 dBm (Fig. 3c), the spectral bandwidth grows to 33 cm−1 and the multiple lobes of the spectrum start to disappear. Interestingly, the SWIFTS spectrum shows that the ICL is not fully phase- locked in this transition state to the actively mode-locked regime. The intermodal difference phases still decrease linearly, but only cover the range of 0.53 · 2π. Since ∆φ is directly proportional to the group delay with 2π cor- responding to one cavity round-trip, this means that the ICL emits pulses with roughly 0.53 · 2π/2π ≈53% duty cy- cle. Indeed, the reconstructed time signal (Fig. 3d) shows broad and linearly chirped pulses. At 30 dBm, the spec- trum consists of a single lobe spanning over 36 cm−1 (Fig. 3e). The slope of ∆φ decreases, which corresponds to a de- crease of the group delay dispersion (GDD) to -5.4 ps2 at 30 dBm compared to -19 ps2 at 6 dBm. The reconstructed time signal (Fig. 3f) shows a train of isolated pulses with 27 ps full width at half maximum (FWHM). When the in- jected power is increased by another 6 dB, only a minor change in the shape of the spectrum and the GDD is ob- served (Fig. 3g). Detuning the modulation frequency away from the round- trip frequency frep of the free-running laser is another knob to influence the temporal output of the laser. Fig. 4a shows the SWIFTS characterization at 32 dBm injected power, while the modulation frequency is 15 MHz higher than frep. The spectrum consists of a single lobe spanning over 45 cm−1 and is fully coherent. Indeed, the intermodal difference phases do not show a linear chirp as in Figs 3a-h and occupy the phase range of only 0.13 · 2π. The recon- structed time signal (Fig. 4b) displays much shorter pulses than in Fig. 3g. A zoom on a single pulse reveals a FWHM of 3.2 ps. The peak power is 114 mW, which is an enhance- ment of more than 40 with respect to the average power of 2.7 mW and proves that the energy can be efficiently stored by the gain medium over a round-trip. The inset in Fig. 4b shows that the pulse is not transformation-limited with several smaller pulses arriving shortly after the first intense pulse. This is because the intermodal difference phases in Fig. 4a do not synchronize perfectly. In con- clusion, our results show that ICLs provide all features necessary to generate MIR pulses within a single semicon- ductor laser ridge. The free-running or weakly modulated ICL frequency comb operates in a regime characterized by a strong linear chirp and suppression of amplitude mod- ulation, which was also found in QCLs and quantum dot lasers. Upon increasing the modulation amplitude, the laser spectrum broadens accompanied by the formation of 3 25802590260026102620Wavenumber [cm1]00101GDD=-19.0 ps229 cm1SWIFTSIntensityaPinj = 6 dBm0.00.51.0Norm. powerb050100150Time [ps]258026002620Instantaneouswaven. [cm1]25802590260026102620Wavenumber [cm1]0.532GDD=-8.8 ps233 cm1cPinj = 18 dBm0.53Trepd050100150Time [ps]25802590260026102620Wavenumber [cm1]0.312GDD=-5.4 ps236 cm1ePinj = 30 dBm0.31Trepf050100150Time [ps]25802590260026102620Wavenumber [cm1]0.272GDD=-4.4 ps238 cm1gPinj = 36 dBm0.27Treph050100150Time [ps]References 1. Schliesser, A., Picqu´e, N. & Hansch, T. W. Mid- infrared frequency combs. Nature Photonics 6, 440 -- 449 (2012). 2. Keilmann, F. & Amarie, S. Mid-infrared frequency comb spanning an octave based on an er fiber laser and difference-frequency generation. Journal of Infrared, Millimeter, and Terahertz Waves 33, 479 -- 484 (2012). URL https://doi.org/10.1007/ s10762-012-9894-x. 3. Leindecker, N. et al. Octave-spanning ultrafast OPO with 26-61µm instantaneous bandwidth pumped by femtosecond tm-fiber laser. Optics Express 20, 7046 (2012). URL https://doi.org/10.1364/oe. 20.007046. 4. Wang, C. Y. et al. Mid-infrared optical frequency combs at 2.5 µm based on crystalline microresonators. Nature Communications 4 (2013). URL https:// doi.org/10.1038/ncomms2335. 5. Kuyken, B. et al. An octave-spanning mid-infrared frequency comb generated in a silicon nanophotonic wire waveguide. Nature Communications 6 (2015). URL https://doi.org/10.1038/ncomms7310. 6. Scalari, G., Faist, J. & Picqu´e, N. On-chip mid- infrared and THz frequency combs for spectroscopy. Applied Physics Letters 114, 150401 (2019). URL https://doi.org/10.1063/1.5097933. 7. Faist, J. et al. Quantum Cascade Laser. Science 264, 553 -- 556 (1994). 8. Hugi, A., Villares, G., Blaser, S., Liu, H. C. & Faist, J. Mid-infrared frequency comb based on a quantum cascade laser. Nature 492, 229 -- 233 (2012). 9. Villares, G. et al. Dispersion engineering of quan- tum cascade laser frequency combs. Optica 3, 252 -- 258 (2016). 10. Hillbrand, J., Jouy, P., Beck, M. & Faist, J. Tunable dispersion compensation of quantum cascade laser fre- quency combs. Optics Letters 43, 1746 (2018). URL https://doi.org/10.1364/ol.43.001746. 11. Singleton, M., Jouy, P., Beck, M. & Faist, J. Ev- idence of linear chirp in mid-infrared quantum cas- cade lasers. Optica 5, 948 (2018). URL https: //doi.org/10.1364/optica.5.000948. 12. Gordon, A. et al. Multimode regimes in quantum cascade lasers: From coherent instabilities to spatial hole burning. Physical Review A 77 (2008). URL https://doi.org/10.1103/physreva.77.053804. 13. Wang, C. Y. et al. Mode-locked pulses from mid- infrared quantum cascade lasers. Optics Express 17, 12929 (2009). URL https://doi.org/10.1364/oe. 17.012929. 14. Yang, R. Q. Infrared laser based on intersubband transitions in quantum wells. Superlattices and Mi- crostructures 17, 77 -- 83 (1995). URL https://doi. Figure 4 -- a: SWIFTS characterization for 920 A/cm2 gain section cur- rent density and 3 V modulation section bias. The modulation power is 32 dBm and the modulation frequency is roughly 15 MHz higher than frep. In contrast to the weakly modulated ICL (Fig. 2a), both SWIFTS interferograms now have local maxima at zero delay of the FTIR mirrors. This is an indicator for strong amplitude modulation of the laser inten- sity. b: Reconstructed time domain signal. The inset shows a zoom on a single pulse, revealing a FWHM of 3.2 ps. broad and chirped pulses. By carefully tuning the modula- tion frequency away from the round-trip frequency of the free-running laser, the pulse duration decreases to 3.2 ps with a peak power enhancement of over 40. This illus- trates the large potential of ICLs as a compact source for mid-infrared pulses. From the spectral bandwidth one can assume that further optimization of the laser dispersion, modulation section length or adding additional segments will enable the generation of subpicosecond pulses 26. This work was supported by the Austrian Science Fund (FWF) within the projects "NanoPlas" (P28914-N27), "Build- ing Solids for Function" (Project W1243), "NextLite" (F4909- N23), as well as by the Austrian Research Promotion Agency through the ERA-Net Photonic Sensing program, project "ATMO-SENSE" (FFG: 861581). J.H. was supported by the 'Hochschuljubilaumsstiftung' of the city of Vienna. H.D. was supported by the ESF project CZ.02.2.69/0.0/ 0.0/16 027/0008371. A.M.A was supported by the projects COMTERA - FFG 849614 and AFOSR FA9550-17-1-0340. 4 257025802590260026102620Wavenumber [cm1]00101SWIFTSIntensity0.13245 cm1aSWIFTS analysis101Delay [mm]SWIFTS XSWIFTS YIntensityCenter burst0255075100125150175Time [ps]020406080100120Power [mW]bPavg=2.7 mWReconstructed time domain signal5053.16psorg/10.1006/spmi.1995.1017. 15. Vurgaftman, I. et al. Interband cascade lasers. Jour- nal of Physics D: Applied Physics 48, 123001 (2015). URL https://doi.org/10.1088/0022-3727/48/12/ 123001. 16. Canedy, C. L. et al. Interband cascade lasers with longer wavelengths. In Razeghi, M. (ed.) Quantum Sensing and Nano Electronics and Photonics XIV (SPIE, 2017). URL https://doi.org/10.1117/12. 2246450. 17. Lotfi, H. et al. Monolithically integrated mid-IR in- terband cascade laser and photodetector operating at room temperature. Applied Physics Letters 109, 151111 (2016). URL https://doi.org/10.1063/1. 4964837. 18. Schwarz, B. et al. A monolithic frequency comb plat- form based on interband cascade lasers and detectors. arXiv preprint arXiv:1812.03879 (2018). 19. Kuizenga, D. & Siegman, A. FM and AM mode locking of the homogeneous laser - part i: Theory. IEEE Journal of Quantum Electronics 6, 694 -- 708 (1970). URL https://doi.org/10.1109/jqe.1970. 1076343. 20. Bagheri, M. et al. Passively mode-locked inter- band cascade optical frequency combs. Scientific Re- ports 8 (2018). URL https://doi.org/10.1038/ s41598-018-21504-9. 21. Kane, D. & Trebino, R. Characterization of arbitrary femtosecond pulses using frequency-resolved optical gating. IEEE Journal of Quantum Electronics 29, 571 -- 579 (1993). URL https://doi.org/10.1109/3. 199311. 22. Burghoff, D. et al. Evaluating the coherence and time- domain profile of quantum cascade laser frequency combs. Optics Express 23, 1190 (2015). 23. Hillbrand, J. et al. Frequency comb dynamics of ul- trafast quantum dot lasers. currently being prepared (2019). 24. Hillbrand, J., Andrews, A. M., Detz, H., Strasser, G. & Schwarz, B. Coherent injection locking of quan- tum cascade laser frequency combs. Nature Photon- ics 13, 101 -- 104 (2018). URL https://doi.org/10. 1038/s41566-018-0320-3. 25. Opacak, N. & Schwarz, B. Theory of frequency mod- ulated combs induced by spatial hole burning, dis- persion and kerr. arXiv preprint arXiv:1905.13635 (2019). 26. Bowers, J., Morton, P., Mar, A. & Corzine, S. Ac- tively mode-locked semiconductor lasers. IEEE Jour- nal of Quantum Electronics 25, 1426 -- 1439 (1989). URL https://doi.org/10.1109/3.29278. 5
1711.09701	1	1711	2017-11-27T14:20:09	A continuously tunable acoustic metasurface for transmitted wavefront manipulation	[ "physics.app-ph" ]	Previously reported acoustic metasurfaces that consist of fixed channels, are untunable to meet the broadband requirement and alterable functionalities. To overcome this limitation, we propose screw-and-nut mechanism of tunability and design a type of continuously tunable acoustic metasurface with unit components of helical cylinders which are screwed into a plate. The spiral channel length can be tuned continuously by the screwed depth; and then a metasurface with continuously tunable acoustic phase at independent pixels is attained. Different distributions of the unit components can shape an arbitrary metasurface profile. We also developed an approximate equivalent medium model to predict the tunability of the unit component. As an example, we present the design details of a circular tunable metasurface for three-dimensional acoustic focusing of different airborne sound sources in a wide frequency region. A sample of the metasurface is manufactured by poly lactic acid (PLA) plastic with the helical cylinders being 3D-printed. Experiments of sound focusing are performed. It is shown that the results of the equivalent medium model, the finite element simulation and the experiments are in a good agreement.	physics.app-ph	physics	A continuously tunable acoustic metasurface for transmitted wavefront manipulation Sheng-Dong Zhao1, 2, Yue-Sheng Wang1,, and Chuanzeng Zhang2, 1Institute of Engineering Mechanics, Beijing Jiaotong University, Beijing 100044, China 2Department of Civil Engineering, University of Siegen, Siegen 57068, Germany Abstract: Previously reported acoustic metasurfaces that consist of fixed channels, are untunable to meet the broadband requirement and alterable functionalities. To overcome this limitation, we propose screw-and-nut mechanism of tunability and design a type of continuously tunable acoustic metasurface with unit components of helical cylinders which are screwed into a plate. The spiral channel length can be tuned continuously by the screwed depth; and then a metasurface with continuously tunable acoustic phase at independent pixels is attained. Different distributions of the unit components can shape an arbitrary metasurface profile. We also developed an approximate equivalent medium model to predict the tunability of the unit component. As an example, we present the design details of a circular tunable metasurface for three-dimensional acoustic focusing of different airborne sound sources in a wide frequency region. A sample of the metasurface is manufactured by poly lactic acid (PLA) plastic with the helical cylinders being 3D-printed. Experiments of sound focusing are performed. It is shown that the results of the equivalent medium model, the finite element simulation and the experiments are in a good agreement. Introduction Recently, manipulation of waves by metasurfaces becomes a research focus in the communities of electromagnetics and acoustics. A metasurface is a kind of artificial material layer with thickness being smaller than a wave length. It can modulate wavefront relying on the gradually accumulated phase changes along the metasurface. The shape of the modulated wavefront is governed by generalized Snell's law1,2; and needed phase profile is achieved by careful design of gradually varying sub-wavelength microstructures of the metasurface3-7. This concept was first proposed in the electromagnetic field1, and was later extended to acoustics. Metasurfaces show prospective applications in many fields, e.g. design of innovative devices, such as optical vertex1, light bending devices3, light axicon8, etc. due to their novel physical properties1,3,8-10. However, because there is no plasmon resonance in the acoustic field, design of an acoustic metasurface is a challenging task. In 2012, Liang and Li11 used the curled perforations to coil up the space and proposed a new metamaterials with extreme constitutive parameters. Inspired by their study, many researchers proposed various designs of acoustic metasurfaces with space-coiling structures and demonstrated a variety of novel features12-15. The space-coiling metasurface show an exceptional ability in controlling acoustic waves, including self-accelerating beam generators15, negative bulk modulus cell16, acoustic rectifiers17,18, optimal sound-absorbing * Corresponding author: [email protected] * Corresponding author: [email protected] 1 device19-21, acoustic rainbow trapping22, extraordinary transmission23,24, acoustic one-way device25-27, negative refraction11,28, unidirectional acoustic cloak29,30, and acoustic focusing31-34. Such metasurface functionalities are based on the dispersive wavefront modulation and the extraordinary acoustic transmission which is attributed to the interplay between Fabry-Perot resonances inside the sub-wavelength-scale channel and resonances of the surface waves propagating on the plate31,33. It is known that the traditional metamaterial with a fixed microstructure can only operate in a fixed frequency range with fixed functions. In order to break this restriction, researchers pay more attentions on tunable metamaterials35-39. However research on tunable metasurfaces is limited. Very recently, by introducing the thermo-optic effect40,41 or varactor diode42,43, etc. into the active unit component of the metasurface, the continuously changeable optical phase at independent pixel achieves a great breakthrough. Continuously tunable metasurfaces have the ability to generate arbitrary radiation patterns which have long been pursued. For acoustic waves, Xie et al.44 proposed the concept of coding acoustic metasurfaces that can combine simple logical bits to acquire sophisticated functions in the wave control. Memoli et al.45 developed the notation of quantal metasurface that is assembled by replaceable metamaterial bricks. The tunable mechanisms in above mentioned studies are based on assembly of various fixed components, and therefore are inflexible and not continuously adjustable. Continuous tunability is much helpful to realize broadband muti-functions of metasurfaces. In this paper, we propose a continuously tunable helical-structured metasurface (HSM) which is manufactured by screwing the helical cylinders into a plate. The spiral channel length can be tuned by the screwed depth into the plate. The HSM can slow down the propagating waves by introducing helical wave rotation and wavefront revolution15. Therefore, a tunable metasurface with a high effective refractive index and effective mass density is obtained. We particularly design a circular metasurface for three-dimensional acoustic focusing and perform the numerical simulation, develop the equivalent theoretical model and experimentally evidence the transmitted acoustic focusing phenomenon. The tunability of the metasurface regarding the acoustic source types (i.e. the plane wave source and the point source), the incident wave frequency, and the focal length is analyzed and discussed. Results Basic idea of tunable metasurface design. The design of the metasurface is illustrated schematically in Fig. 1. Fig. 1a shows the metasurface with distributed helical cylinders screwed into a plate. The tunability of the metasurface is based on the screw-and-nut mechanism with a tunable screwed depth, as shown in Fig. 1b. The plate is perforated with circular holes; and the inwall of each hole is cut into left-handed screw with shallow grooves (see Fig. 1d). After screwing the helical cylinder (Fig. 1c) into the hole, a spiral channel is formed between the inwall of the hole and the helical cylinder. The channel length and thus the transmitted phase can be continuously tuned by adjusting the screwed depth. So a new degree of freedom of controlling wavefronts can be attained by introducing arbitrary form of phase shift along the metasurface. The wavefront modulation is based on the Snell's law1. For the sake of simplicity, we suppose a one-way phase gradient along x-direction as shown in Fig. 1a. For an incident wave in xz-plane, the refraction angle generalized by Snell's law1 is (1) 2 tidarcsinsin2dxwhere is the incident angle; is the transmitted phase; and is the wave length, see the schematic diagram of Fig. 1a. Figure 1│ The tunable metasurface based on the screw-and-nut mechanism. (a) The schematic diagram shows that the anomalous refraction phenomenon achieved by the tunable metasurface is governed by the generalized Snell's law. The gradient phase is tuned by the screwed depth of the helical cylinders. (b) The helical cylinder with two opposite spiraling blades connected through a central slender column. The cylinder is screwed into a block and then a tunable unit component is combined. (c) The geometric details of the helical cylinder. (d) The drilled hole with the inwall cut into shallow grooves. One of the key steps in design of the tunable metasurface is to obtain the relationship between the phase changes and screwed depth of the helical cylinder under different frequencies with consideration of high transmission. As an example, we will next present a careful design of a circular tunable metasurface for three-dimensional acoustic focusing of different airborne sound sources in a wide frequency region. Design of a tunable three-dimensional focusing metasurface. The proposed tunable HSM is composed of a perforated circular plate and helical cylinders screwed into the plate. All the components are made of poly lactic acid (PLA) plastic which can be treated as a rigid material relative to the airborne sound. Each helical cylinder is 3D-printed with two opposite spiraling blades connected through a central slender column, see Fig. 2a. Its geometry can be defined by the outer diameter , which is the same as the plate the column diameter . The cutting thickness, the blade thickness shallow grooves on the inwall of the drilled hole are defined by in depth and in width (see Fig. 1d). All the above geometric parameters marked in Fig. 1c, d are smaller than the wavelength of the airborne sound. The plate with the radius of is perforated with circular holes distributed in four circular , and the thread lead , the length layers as shown in Fig. 2b, c. The diameter of the hole is . The first layer is in the plate center with one hole; and the other three layers from the center to the outside contain 6, 12 and 18 holes, respectively. The distance between two adjacent layers is 50mm (i.e., , see Fig. 2b). , , , The helical cylinders are screwed into the holes as shown in Fig. 2c. Then a tunable spiral channel is formed between the inwall of the hole and the helical cylinder. 3 i32mmD8mmd40mmL1mmh10mmP1mmt1mmhd190mmr30mmD10r250mmr3100mmr4150mmr Figure 2│ The tunable metasurface for three-dimensional acoustic focusing and the transmission property corresponding to the screwed depth. (a) The fabricated helical cylinder based on 3D printing technique. (b) The geometric details of the plate with distributed helical cylinders screwed into the holes. (c) The sample of the tunable metasurface composed of a perforated circular plate with four layers of the helical air channels formed by the helical cylinders. (d) The transmitted wave field from the numerical simulation model with the screwed depth of at the frequency of 5.5kHz. (e) and (f) The nephograms of the phase and transmittance in the plane of the frequency and screwed depth. (g) and (h) The cut lines of (e) and (f) at the frequency of 5.5kHz showing the variation of the phase and transmittance with the screwed depth. Introduction of working principle and performance. We consider acoustic waves which are propagating through the HSM along the helical paths. By tuning the screwed depth of the helical cylinders appropriately, we can obtain the proper radial gradient variation of the phase distribution, yielding acoustic focusing according to the generalized Snell's law. This is the basic idea of our tunable metasurface. In the present design, the tunable depth ranges from 15mm to 40mm inducing highly controllable phase shifts of the transmitted wavefront over the whole range within the focusing frequency between 3.9 kHz and 6.3 kHz. Based on the flexible wave-control capability, a variety of tunable items including the broadband frequency range, the tunable focusing length , and the alterable acoustic source (an incident plane wave source or a point source) are verified. To realize the acoustic focusing with a particular source at a particular frequency, we should first establish the relationship between the transmitted wave phase and the screwed depth of the helical cylinders. To this end, a finite element simulation model based on the commercial software COMSOL Multiphysics is developed, see Fig. 2d. 4 22mmL2fyThe calculated phase shift and the normalized transmittance of the transmitted waves at the observation point A, which has a distance of 90mm to the plate, are shown in Fig. 2e, f as functions of the frequency and the screwed depth. By properly selecting the phase profiles in the radial direction of the metasurface plate, several fascinating wavefront engineering capabilities can be realized. For acoustic focusing, the hyperboloidal phase profile is necessary34. Because only the phases at the four layers of the helical cylinders can be tuned, the continuous phase profile must be converted to a discrete phase profile. For an incident plane wave, we obtain the phase difference between adjacent layers as , (2) where is the wave length at the focusing frequency with being the airborne sound speed and is the wave path from the ith layer to the focal point, see Fig. 2b. For a point wave source, the associated phase difference is given by , (3) where is the wave path from the ith layer to the source point with being the distance between the point source and the hyperlens. Within the controllable phase shifts of the transmitted waves, if the phase degree of the ith layer ( ) is determined, other phase degrees are thereupon determined by following the Eq. (2) or (3). It should be noted here that the result obtained in this way is not unique. Among the numerous potential tunable results, we should find the optimal one to achieve a high transmittance. So we should always try to make as many helical cylinders as possible to have a high transmittance when we determine their screwed depth. As an example, we consider an incident plane wave with the frequency , we can find that the phase . For the focusing length differences between the first layer to the others are , and according to Eq. (2). The cut lines of the phase and transmittance nephogram (see the dashed lines in Fig. 2e, f, respectively) at the frequency of 5.5 kHz are plotted in Fig. 2g, h. Assuming that then we can obtain and as shown in Fig 2g. The corresponding transmittance , , and are plotted in Fig. 2h, which are all located around the resonance peaks. Finally we get the optimized screwed depths of the four layers (from the center to the outside) as: 28mm, 22.8mm, 28mm and 32.8mm. Fig. 3a shows the schematic diagram of the plane wave focusing model. The calculated is presented in Fig. 3b. It is seen that the normalized pressure intensity field selected screwed depths of the helical cylinders can indeed guarantee a satisfactory acoustic focusing at the frequency of . Following the same process as described above, the optimized screwed depth corresponding to different frequencies and focusing lengths for both plane wave and point sources are obtained, and the results are given in Table S1 of the Supporting Information. These data may be used to tune the focus of the designed HSM. 5 112()[,]iiiidd2,3,4i0cff10342msc221/2()ifidyr1112()[,]iiiiiidddd2,3,4i221/2s()iidyrsy2i5.5kHzff50mmy212.1310.027411.63310.8121.2942.441T2T3T4TL20()PP5.5kHzfFigure 3│ The focusing results for an incident plane wave source at the frequency of 5.5kHz with the focusing length . (a) The schematic diagram of the plane wave focusing model. (b) The simulation results. (c) The effective medium model results. (d) The experimental results. (e) and (f) The longitudinal and transverse pressure intensity fields at the focal spot along the y- and x-directions. The pressure intensity is normalized by the incident plane wave amplitude . Comparison of the numerical simulation, approximate effective medium model and experimental results. Instead of the above detailed numerical simulation, an approximate effective medium model is developed to design the tuning of the acoustic focusing. In our model, the helical cylinder with the air in the spiral channel is replaced by a cylindrical channel with an equivalent homogeneous medium which has the same size as the screwed part (with the diameter ). It is noted here that the pillow left out of the plate should be considered in the model. are calculated as The effective density (in kg/m3) and the refractive index and length functions of the screwed depth Supplementary Note 1. The results are as follows: . For more details of the model, we refer to the , (4) , (5) 6 f50mmy0P30mmDLeffeffnL5332()3.3710()3.4710()0.1293.85effnLLLL4322()3.0710()3.1710()1.1835.11effLLLLfrom which we can easily obtain the effective parameters associated with the screwed depth. The acoustic focusing field based on the approximate effective medium model is plotted in Fig. 3c. The experimentally measured pressure intensity field is shown in Fig. 3d. Fig. 3b-d all show a sharp focal spot that is about more than four times of the pressure amplitude of the incident plane wave. For a more quantitative comparison, the normalized longitudinal and transverse pressure intensity fields at the focal spot ) along the y- and x-directions corresponding to Fig. 3b-d are shown in Fig. 3e, ( f. All results from the numerical simulation, effective medium model and experimental measurement are in good agreement except that the intensities obtained from the effective medium model and the experimental measurement are a little lower than the numerical simulation result. Presumably, this discrepancy is caused by the fact that an energy loss always exists due to the air viscosity in the experiment, and the big entrance part (Supplementary Fig. 1) induces a low estimation of the resonant transmission peaks by the approximate effective medium model. The simulation and experimental results for some other selected values of the frequency (4.1kHz, 5.1kHz and 5.9kHz) and focusing lengths (50mm and 100mm) are plotted in Supplementary Fig. 3. These results show that a broadband acoustic focusing can be achieved by the tunable HSM for an incident plane wave. We next discuss the focusing of a point wave source with the source length of is also (Fig. 4a). In this example, the focusing frequency assumed but the focusing length is . The phase degrees are thereupon determined by following Eq. (3), and the verified phase differences between the first , respectively. layer to the others are and , Then the optimized screwed depths of the four layers in this example are selected from the Table S1 of the Supporting Information as 32mm, 27.6mm, 29mm and 28mm, respectively. The results of the numerical simulation, the effective medium model and the experiment are shown in Fig. 4b-d. The normalized pressure intensity in the y- and x-directions across the focusing point show that the fields of focusing length is indeed around 100mm (Fig. 4e, f). Similar to the plane wave incidence, the numerical simulation and experimental results for some other selected values of the frequency (4.3kHz, 5.1kHz and 6.1kHz) and focusing lengths (50mm and 100mm) are plotted in Supplementary Fig. 4 of the Supporting Information. The results sufficiently illustrate that the tunable HSM is effective for the broadband tunable acoustic focusing. 7 fyys150mmy5.5kHzff100mmy212.01310.97411.852s()PP and focusing length Figure 4│ The focusing results for a point source at the frequency of 5.5kHz with the source length . (a) The schematic diagram of the point source focusing model. (b) The simulation results. (c) The effective medium model results. (d) The experimental results. (e) and (f) The longitudinal and transverse pressure intensity fields at the focal spot along the y- and x-directions. The pressure intensity is normalized by the pressure amplitude of a semi-spherical surface source (with radius of 12.7mm). The detailed definition and calculation are given in Supplementary Note 3. Collection and analysis of the data of the focusing quality. To evaluate the focusing quality and the broadband tunability within the frequency range of 3.9~6.3kHz, we present the numerical simulation results including the focusing lengths, the focusing amplitudes and the focusing resolutions in Fig. 5. The data are collected into four groups corresponding to the two kinds of acoustic sources (plane wave source and point source) and the two preset focusing lengths (50mm and 100mm). The optimized 8 s150mmyf100mmysPvalues of the screwed depth are taken from Table S1. It is seen from Fig. 5a that the four groups of the data are located around their preset focusing positions of with a maximum error of 20%. The four groups of the and focusing amplitudes are shown in Fig. 5b in two scales. The pressure amplitudes of the focusing point with respect to the point source are around 12%, and the focusing amplitudes for the plane wave source are around 4.5 times of the incident plane wave amplitude. Fig. 5c, d show the longitudinal and transverse spatial resolutions which are defined as the full widths at the half maximum of the transmission peak measured crossing the focusing point along the y- and x-directions, respectively. The results demonstrate that the transverse resolution is higher than the longitudinal resolution, which means that the focusing point is not a circle. In particular, the near field focusing (50mm) acquires a higher resolution than the far focusing (100mm). The transverse resolution does not break but is close to the diffraction limit of 0.5λ. Figure 5│ All the details of the simulation results for the broadband tunable focusing effect. The numerical simulation results of the focusing lengths (a), the focusing amplitudes (b) and the longitudinal (c) and transverse (d) focusing resolutions for the plane wave and point sources with the focusing length of 50mm and 100mm within the frequency range of 3.9kHz~6.3kHz. Discussion In summary, we designed a novel tunable metasurface consisting of helical cylinders screwed into a circular plate. The screw-and-nut mechanism is used to tune the spiral channel length to obtain a flexible manipulated wavefront. Thus, a continuously tunable acoustic metasurface is achieved. The broadband tunable acoustic focusing property of the system is studied in details. Acoustic focusing with a relatively high transmittance is obtained, and the tunability is independent of the sound source type and the focusing length. Such flexible tenability is highly desirable in phase engineering applications. The proposed mechanism can improve the adjustment of the 9 f50mmyf100mmymetasurface having any kind of the phase profile, such as the acoustic axicon for the Bessel beam or Airy beam, tunable carpet cloak and even a lens with tunable helical wavefronts that carry an angular momentum. Furthermore, the proposed unit components can be assembled into a three-dimensional curved surface to design a tunable curved metasurface. Method Numerical simulations. Throughout the paper, all wave propagation simulations are performed by using COMSOL Multiphysics 4.4 with the pressure acoustics module. The surrounding gas applied in our simulations is air. The solid metasurface is treated as a rigid material relative to the airborne sound and sound hard boundaries are used on the surface. The plane wave radiation boundary condition is adopted on the outer boundaries to eliminate the reflected waves. Sample fabrication and experimental setup. Each helical cylinder was fabricated by using a 3D printer (Z500, HORI, China). The circular plate was fabricated through mechanical processing. In the experiment, the acoustic wave was incident to the HSM from the side with outer pillows (Supplementary Fig. 5a), and the pressure intensity field was measured on the other side (Supplementary Fig. 5b). The tunable lens was surrounded by the absorbing foam and placed on the axis of the sound radiation. The loudspeaker (HS8, YAMAHA, Japan) is surrounded by the absorbing foam box with the thickness of 50mm (Supplementary Fig. 5c). For the incident plane wave source, the loudspeaker was placed on the lens axis with the distance of 1300mm (marked by 1 in Supplementary Fig. 5c) which is far enough to generate a plane wave. For the point source the loudspeaker was placed at the source point with the distance of 150mm to the lens (marked by 2 in Supplementary Fig. 5c). In the experiment, we first used the software of Audacity to generate a single frequency signal, and then the signal was input to the sound card (AudioBox22VST, PreSonus, USA) and played out by the loudspeaker. On the other side of the lens, the pressure intensity field was scanned in the axial plane area in a rectangular region of for the focusing length of 50mm, and for the focusing length of 100mm. The scanned area was separated from the right lens edge by 10mm. The scanned signal was acquired and analyzed by Audacity. Data availability. The data that support the findings of this study are available from the corresponding author on reasonable request. Reference 1. Yu, N., et al. Light propagation with phase discontinuities: generalized laws of reflection and refraction. Science 334, 333-337 (2011). 2. Chen, H.T., Taylor, A.J. & Yu, N. A review of metasurfaces: physics and applications. Rep. Prog. Phys. 79, 076401 (2016). 3. Ni, X., Emani, N.K., Kildishev, A.V., Boltasseva, A. & Shalaev, V.M. Broadband light bending with plasmonic nanoantennas. Science 335, 427 (2012). 4. Sun, S.L., et al. Gradient-index meta-surfaces as a bridge linking propagating waves and surface waves. Nat. Mater. 11, 426-431 (2012). 5. Sun, S., et al. High-efficiency broadband anomalous reflection by gradient 10 2160140mm2160180mmmeta-surfaces. Nano. Lett. 12, 6223-6229 (2012). 6. Arbabi, A., Horie, Y., Bagheri, M. & Faraon, A. Dielectric metasurfaces for complete control of phase and polarization with subwavelength spatial resolution and high transmission. Nat. Nanotechnol. 10, 937-943 (2015). 7. Wolf, O., et al. Phased-array sources based on nonlinear metamaterial nanocavities. Nat. Commun. 6, 7667 (2015). 8. Lin, D., Fan, P., Hasman, E. & Brongersma, M.L. Dielectric gradient metasurface optical elements. Science 345, 298-302 (2014). 9. Li, G., et al. Continuous control of the nonlinearity phase for harmonic generations. Nat. Mater. 14, 607-612 (2015). 10. Raman, A.P., Anoma, M.A., Zhu, L., Rephaeli, E. & Fan, S. Passive radiative cooling below ambient air temperature under direct sunlight. Nature 515, 540-544 (2014). 11. Liang, Z. & Li, J. Extreme acoustic metamaterial by coiling up space. Phys. Rev. Lett. 108, 114301 (2012). 12. Xie, Y., Popa, B.I., Zigoneanu, L. & Cummer, S.A. Measurement of a broadband negative index with space-coiling acoustic metamaterials. Phys. Rev. Lett. 110, 175501 (2013). 13. Li, Y., et al. Acoustic focusing by coiling up space. Appl. Phys. Lett. 101, 233508 (2012). 14. Li, Y., et al. Three-dimensional Ultrathin Planar Lenses by Acoustic Metamaterials. Sci. Rep. 4, 6830 (2014). 15. Zhu, X.F., et al. Implementation of dispersion-free slow acoustic wave propagation and phase engineering with helical-structured metamaterials. Nat. Commun. 7, 11731 (2016). 16. Cheng, Y., et al. Ultra-sparse metasurface for high reflection of low-frequency sound based on artificial Mie resonances. Nat. Mater. 14, 1013-1019 (2015). 17. Liang, B., Yuan, B. & Cheng, J.C. Acoustic diode: rectification of acoustic energy flux in one-dimensional systems. Phys. Rev. Lett. 103, 104301 (2009). 18. Liang, B., Guo, X.S., Tu, J., Zhang, D. & Cheng, J.C. An acoustic rectifier. Nat. Mater. 9, 989-992 (2010). 19. Yang, M., Chen, S.Y., Fuab, C.X. & Sheng, P. Optimal sound-absorbing structures. Mater. Horiz. 4, 673-680 (2017). 20. Li, Y. & Assouar, B.M. Acoustic metasurface-based perfect absorber with deep subwavelength thickness. Appl. Phys. Lett. 108, 063502 (2016). 21. Ma, G., Yang, M., Xiao, S., Yang, Z. & Sheng, P. Acoustic metasurface with hybrid resonances. Nat. Mater. 13, 873-878 (2014). 22. Ni, X., et al. Acoustic rainbow trapping by coiling up space. Sci. Rep. 4, 7038 (2014). 23. Li, Y., Liang, B., Zou, X.Y. & Cheng, J.C. Extraordinary acoustic transmission through ultrathin acoustic metamaterials by coiling up space. Appl. Phys. Lett. 103, 063509 (2013). 24. Li, Y., Liang, B., Gu, Z.M., Zou, X.Y. & Cheng, J.C. Unidirectional acoustic transmission through a prism with near-zero refractive index. Appl. Phys. Lett. 11 103, 053505 (2013). 25. Jiang, X., et al. Acoustic one-way metasurfaces: Asymmetric Phase Modulation of Sound by Subwavelength Layer. Sci. Rep. 6, 28023 (2016). 26. Zhu, Y.F., Zou, X.Y., Liang, B. & Cheng, J.C. Acoustic one-way open tunnel by using metasurface. Appl. Phys. Lett. 107, 113501 (2015). 27. Li, Y., et al. Tunable Asymmetric Transmission via Lossy Acoustic Metasurfaces. Phys. Rev. Lett. 119, 035501 (2017). 28. Zhu, H.F. & Semperlotti, F. Anomalous Refraction of Acoustic Guided Waves in Solids with Geometrically Tapered Metasurfaces. Phys. Rev. Lett. 117, 034302 (2016). 29. Wang, X.P., Wan, L.L., Chen, T.N., Song, A.L. & Wang, F. Broadband unidirectional acoustic cloak based on phase gradient metasurfaces with two flat acoustic lenses. J. Appl. Phys. 120, 014902 (2016). 30. Wang, X., Mao, D.X. & Li, Y. Broadband acoustic skin cloak based on spiral metasurfaces. Sci. Rep. 7, 11604 (2017). 31. Al Jahdali, R. & Wu, Y. High focusing by impedance-matched acoustic meta-surfaces. Appl. Phys. Lett. 108, 031902 (2016). 32. Yuan, B.G., Cheng, Y. & Liu, X.J. Conversion of sound radiation pattern via gradient acoustic metasurface with space-coiling structure. Appl. Phys. Express 8, 027301 (2015). transmission acoustic 33. Moleron, M., Serra-Garcia, M. & Daraio, C. Acoustic Fresnel lenses with extraordinary transmission. Appl. Phys. Lett. 105, 114109 (2014). 34. Li, Y., Liang, B., Gu, Z.M., Zou, X.Y. & Cheng, J.C. Reflected wavefront manipulation based on ultrathin planar acoustic metasurfaces. Sci. Rep. 3, 2546 (2013). 35. Wang, Z.W., Zhang, Q., Zhang, K. & Hu, G.K. Tunable Digital Metamaterial for Broadband Vibration Isolation at Low Frequency. Adv. Mater. 28, 9857 (2016). 36. Li, F.M., Zhang, C.Z. & Liu, C.C. Active tuning of vibration and wave propagation in elastic beams with periodically placed piezoelectric actuator/sensor pairs. J. Sound. Vib. 393, 14-29 (2017). 37. Wang, P., Casadei, F., Shan, S.C., Weaver, J.C. & Bertoldi, K. Harnessing Buckling to Design Tunable Locally Resonant Acoustic Metamaterials. Phys. Rev. Lett. 113, 014301 (2014). 38. Zhang, H.K., Zhou, X.M. & Hu, G.K. Shape-adaptable hyperlens for acoustic magnifying imaging. Appl. Phys. Lett. 109, 224103 (2016). 39. Zhang, Q., Zhang, K. & Hu, G.K. Smart three-dimensional lightweight structure triggered from a thin composite sheet via 3D printing technique. Sci. Rep. 6, 22431 (2016). 40. Sun, J., Timurdogan, E., Yaacobi, A., Hosseini, E.S. & Watts, M.R. Large-scale nanophotonic phased array. Nature 493, 195-199 (2013). 41. DeRose, C.T., et al. Electronically controlled optical beam-steering by an active phased array of metallic nanoantennas. Opt. Express 21, 5198-5208 (2013). 42. Zhu, B.O., et al. Dynamic control of electromagnetic wave propagation with the equivalent principle inspired tunable metasurface. Sci. Rep. 4, 4971 (2014). 12 43. Zhu, B.O., Zhao, J.M. & Feng, Y.J. Active impedance metasurface with full 360 degrees reflection phase tuning. Sci. Rep. 3, 3059 (2013). 44. Xie, B.Y., et al. Coding Acoustic Metasurfaces. Adv. Mater. 29, 1603507 (2017). 45. Memoli, G., et al. Metamaterial bricks and quantization of meta-surfaces. Nat. Commun. 8, 14608 (2017). Acknowledgements This work is supported by the National Natural Science Foundation of China (Grant No. 11532001) and Joint Sino-German Research Project (Grant No. GZ 1355). Author Contributions S. D. Zhao performed the numerical simulations and drafted the manuscript. Y. S. Wang and C. Zhang designed the experiment and supervised the study. S. D. Zhao performed the measurement and the data processing Y. S. Wang and C. Zhang conceived and led the project and contributed to the writing and revision. All authors contributed to the analysis and discussions. Author Information The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to Y. S. Wang ([email protected]) and C. Zhang ([email protected]). 13 Supplementary Information A continuously tunable acoustic metasurface for transmitted wavefront manipulation Sheng-Dong Zhao1, 2, Yue-Sheng Wang1,, and Chuanzeng Zhang2, 1Institute of Engineering Mechanics, Beijing Jiaotong University, Beijing 100044, China 2Department of Civil Engineering, University of Siegen, Siegen 57068, Germany * Corresponding author: [email protected] * Corresponding author: [email protected] 14 Supplementary Note 1. Modified effective medium model This section presents a modified effective medium model considering the outer pillow of the screwed cylinder. According to the effective medium model, the transmission of a one-dimensional can be acoustic wave propagating through a homogeneous medium of length written as1: , (S1) where is the mass density of the air; is the wave propagation constant; and and are the effective density and refractive index, respectively. It can be seen from Eq. (S1) that the resonant peaks will appear when being an integer. This is the well-known Fabry-Pérot resonant with condition that defines the resonant frequencies as where is the airborne sound speed. Based on the resonant peaks in Eq. (S1) and taking we can retrieve the effective refractive index as , , (S2) where is the frequency of the first nonzero resonance peak. When , Eq. (S1) yields the minimum transmittance as . (S3) From Eq. (S3), we can retrieve the effective mass density as . (S4) From Eqs. (S2) and (S4) it can be seen that and are the two key parameters to be determined in the effective medium model. For the present tunable HSM, a pillow with the depth of will be left out of the plate which can be an obstacle and should be taken into account when the effective medium model is used. In order to include the effect of the pillow, we developed a modified model which has a wider entrance part with the diameter (we have verified numerically that a larger size of has a negligible effect on the results) than the original model, as shown in Supplementary Fig. 1. Supplementary Fig. 1a is the idealized model without the outer pillow. Supplementary Fig. 1b is the model with a wider entrance part. Supplementary Fig. 1c is the model with a wider entrance part and an outer pillow. Supplementary Fig. 1d is the modified effective medium model corresponding to model S1c. Without loss of generality, we consider a unit-cell with a helical cylinder screwed depth of . Supplementary Fig. 1a-d show the acoustic transmission fields of the four models at the frequency of 5.74 kHz, and the corresponding transmission ratios are plotted in Supplementary Fig. 1e where a rectangular box highlights the focusing frequency range. We can find that the peak 15 L222eff0eff0eff0eff0effeff44cos()sin()TnknLknLn301.21kgm0keffeffn0effknLmm02cfmL10342msc1m0eff12cnfL1f0eff(21)/2knLmmin2eff0eff0effeff4Tnn0effeff0effminmin11nnTT1fminTLL050mmD0D22mmLfrequencies have almost no change when a wider entrance part is introduced, see the black solid and dashed lines in Supplementary Fig. 1e. However, the outer pillow will obviously influence the resonant frequencies including the fundamental resonant peak frequency , see the lines marked with b and c in Supplementary Fig. 1e. Nevertheless, the outer pillow has a little impact on the transmission ratio at the resonance valley which is marked by an ellipse in Supplementary Fig. 1e. So the outer pillow just affects one of two key parameters, i.e. the fundamental resonance frequency , which will be modified and denoted as . The other parameter in the idealized model of Supplementary Fig. 1a is still valid in the modified effective medium model when the outer-pillow effect is considered. The resonance peaks and lowest transmittance of the modified effective medium model should be consistent with the model of Supplementary Fig. 1c in the focusing frequency range marked by the rectangular box in Supplementary Fig. 1e. The modified frequency is retrieved from the ith resonance peak located in the focusing frequency range (3.9k~6.3k). We assume that the ith resonant frequency is the nearest one to the central frequency (5.1 kHz) of the focusing frequency range, then we have . (S5) Therefore the modified effective refractive index can be written as . (S6) The effective mass density keeps the same form as Eq. (S4). The transmittance of the actual model (S1c) and the modified effective medium model (S1d) reach their peaks at the same frequency , which is called the coupled peak as shown by the red dash-dot and solid lines in Supplementary Fig. 1e. The two lines keep pace with each other in the focusing frequency range, and the two models of Supplementary Fig. 1c and d show the similar transmittance acoustic field. In the particular case of , so the effective parameters can be calculated as , the coupled resonance peak is and . With the screwed depth ranging from 15mm to 40mm, the discrete effective parameters are calculated based on the modified effective medium model. In order to obtain the approximate expressions, the effective parameters are fitted by polynomials of the 3th degree, which are functions of the screwed depth. The discrete and fitted results of the effective parameters are plotted in Supplementary Fig. 2. Then, in the acoustic focusing study we can easily calculate the effective material parameters associated with the screwed depth using these approximate expressions. 16 1f1fmfminTmfifmiffi0eff2icinfLif22mmL5.74kHz (=4)ifieff5.42n3eff49.4kgmLSupplementary Figure 1│ The acoustic transmission fields at the frequency of 5.74 kHz for the four models: the idealized model without the outer pillow (a), the model with a wider entrance part (b), the model with an outer pillow (c), and the effective medium model (d) corresponding to the model (c). The transmission ratios of the four models are plotted in (e). Supplementary Figure 2│ The discrete and fitted results of the effective material parameters and associated with the screwed depth . The fitting curves are given by and (in kgm-3). 17 effneffL5332eff()3.3710()3.4710()0.1293.85nLLLL4322eff()3.0710()3.1710()1.1835.11LLLLSupplementary Note 2. Additional results The following two figures show the numerical simulation and experimental results for some other selected values of the frequency and focusing length. Supplementary Figure 3│ The numerical and experimental focusing results for several selected frequencies when using the plan wave source. (a)-(f) Numerical simulation results for the incident plane wave source at the frequencies of 4.1 kHz, 5.1 kHz and 5.9 kHz with the focusing lengths of 50mm (a-c) and 100mm (d-f). (g) and (h) The experimentally measured intensity fields at the three frequencies along the y- and x-directions at the focal spots for the focusing lengths of 50mm (g, h) and 100mm (i, j). 18 Supplementary Figure 4│ The numerical and experimental focusing results for several selected frequencies when using the point source. (a)-(f) Numerical simulation results for the point source ( ) at the frequencies of 4.3 kHz, 5.1 kHz and 6.1 kHz with the focusing lengths of 50mm (a-c) and 100mm (d-f). (g) and (h) The experimentally measured intensity fields at the three frequencies along the y- and x-directions at the focal spots for the focusing lengths of 50mm (g, h) and 100mm (i, j). 19 s150mmySupplementary Note 3. Definition and calculation of In the numerical computation for the case of a point source, we suppose a time-harmonic acoustic pressure field at the source point and then calculate the surrounding sound field. However, in the experiment a semi-spherical surface sound source with the radius of 12.7mm is used. In order to compare the numerical results with the experimental results, we consider the semi-spherical surface sound source as the half of a radiated field from a point source. The radiated sound pressure from a point source is given by2 , (S7) where the common time-harmonic factor is suppressed; ; and specifies the intensity of the source by stating the volume flow rate from the source. on the semi-spherical surface is In the experiment, the pressure amplitude measured and used in the normalization of the acoustic intensity field. From Eq. (S7), we have . (S8) For comparison, the numerical results are also be normalized by . 20 sP00000()4jkrkcPrjQerejt1j0QsP00001(), 12.7mm28skcPPrQrrsPSupplementary Note 4. Samples and flow chart of the experiment Supplementary Figure 5│The samples and flow chart of the experiment. (a) The incident wave (front sample) side with the outer pillows of the HSM sample. (b) The back sample side where the focused wave field is measured; and (c) the flow chart of the experiment with the scanned area in the x-y plane and the absorbing foam surrounding the source and sample. 21 Supplementary Table 1│ The optimized screwed depth corresponding to variable frequencies (3.9kHz~6.3kHz), different wave sources (plane wave source and point source) and different focusing lengths (50mm and 100mm). f (kHz) Layer 1 (mm) Layer 2 (mm) Layer 3 (mm) Layer 4 (mm) Layer 1 (mm) Layer 2 (mm) Layer 3 (mm) Layer 4 (mm) Layer 1 (mm) Layer 2 (mm) Layer 3 (mm) Layer 4 (mm) Layer 1 (mm) Layer 2 (mm) Layer 3 (mm) Layer 4 (mm) 3.9 4.1 4.3 4.5 4.7 4.9 5.1 5.3 5.5 5.7 5.9 6.1 6.3 24.1 29.7 28.6 20.1 26.3 19.4 18.8 34.3 28 32.4 30.8 34 38 17 23.5 22.5 33.1 20.6 31 30 29 22.8 27 26 16.2 33 30.9 15.5 34.5 21.4 27 19.8 19.2 34.5 28 32 30.3 30.8 23.8 17 23.2 22 28 19.4 25.6 24.6 23.6 32.8 21.6 31.9 34.4 38 30.3 23.9 22.8 27.9 32.1 31.1 29.8 28.6 33 31.2 36 26 23.5 24.8 23 22 26.5 27 26 25 24 17.2 27 21 24.9 33 17 15.9 35 20.2 20.7 19.9 19 34.2 22.7 21.4 26.2 30 24.8 31.1 24.4 22.7 27.5 27.8 26.3 24.9 23.8 28 26.2 30.9 34.5 28.8 23.4 29.8 22.7 22.2 25.4 19.9 19.1 34.9 22.2 21.6 25.1 25.8 37.4 16.4 23 15.9 34.6 19.4 31.2 29.5 28.9 28.2 27.6 31 34 29 22.5 28.6 21.9 21.1 20.3 31.8 30 29 28 27 30.3 30.6 27.5 24 29.4 22 21 20 31 25.5 24.5 38.9 22.7 25.6 25.6 33.8 23.5 28.2 21.1 20.2 20.2 31.2 30.5 29.9 32 36.1 35.3 34.3 33.5 16.8 22.9 35 33.4 32.7 25.6 24.9 28 27.6 31 30.3 29.4 28.7 24.9 30 22 21 38.8 31 30 29.7 29 32 31 30 29 31 15.1 22.5 21.4 39 30.8 29.5 29 28 30.5 30.1 29.2 24.9 Plane wave source (𝑦f = 50mm) Plane wave source (𝑦f = 100mm) Point source (𝑦f = 50mm) Point source (𝑦f = 100mm) Supplementary references 1. J. F, Allard, N. Atalla, (2009) Propagation of Sound in Porous Media: Modelling Sound Absorbing Materials (John Wiley & Sons Ltd). 2. E. Skudrzyk, (1971) The Foundations of Acoustics: Basic Mathematics and Basic Acoustics. (Springer-Verlag New York Wien) 22
1809.09277	1	1809	2018-09-25T01:39:02	Re-estimation of thermal contact resistance considering near-field thermal radiation effect	[ "physics.app-ph" ]	Thermal contact has always been a hot issue in many engineering fields and thermal contact resistance (TCR) is one of the important indicators weighing the heat transfer efficiency among the interfaces. In this paper, the contact heat transfer of conforming rough surfaces is theoretically re-estimated considering both the heat transfer from contact and non-contact regions. The fluctuational electrodynamics (an ab initio calculation) is adopted to calculate the thermal radiation. The contribution of contact regions is estimated by the CMY TCR model and further studied by modelling specific surfaces with corresponding surface roughness power spectrum (PSD). Several tests are presented where aluminum and amorphous alumina are mainly used in the simulations. Studies showed that there exists a significant synergy between the thermal conduction and near-field thermal radiation at the interface in a certain range of effective roughness. When the effective roughness is near to the scales of submicron, the near-field radiation effect should not be neglected even at room temperature.	physics.app-ph	physics	Re-estimation of thermal contact resistance considering near-field thermal radiation effect Yaoqi Xian1, Ping Zhang,1, Siping Zhai1, Peipei Yang１, Zhiheng Zheng2 1 School of Mechanical and Electrical Engineering, Guilin University of Electronic Technology, No. 1 Jinji Road, Guilin, Guangxi 541004, China 2 College of Engineering, Shanghai Second Polytechnic University, Shanghai 201209, China corresponding author, E-mail: [email protected] Abstract Thermal contact has always been a hot issue in many engineering fields and thermal contact resistance (TCR) is one of the important indicators weighing the heat transfer efficiency among the interfaces. In this paper, the contact heat transfer of conforming rough surfaces is theoretically re-estimated considering both the heat transfer from contact and non-contact regions. The fluctuational electrodynamics (an ab initio calculation) is adopted to calculate the thermal radiation. The contribution of contact regions is estimated by the CMY TCR model and further studied by modelling specific surfaces with corresponding surface roughness power spectrum (PSD). Several tests are presented where aluminum and amorphous alumina are mainly used in the simulations. Studies showed that there exists a significant synergy between the thermal conduction and near-field thermal radiation at the interface in a certain range of effective roughness. When the effective roughness is near to the scales of submicron, the near- field radiation effect should not be neglected even at room temperature. Keywords: Thermal contact resistance; Near-field thermal radiation; Conforming rough surface; 1 Greek symbols damping factor, s-1 dielectric constant high frequency dielectric constant σ effective roughness, m mean energy of a Planck oscillator, J relative separation characteristic wavelength, m angle frequency, rad/s plasma frequency, rad/s monochromatic Subscripts/Superscripts p transverse magnetic, TM s transverse electric, TE evan evanescent wave LO longitudinal optical prop propagating wave TO transverse optical Nomenclature i complex constant, (-1)1/2 Im imaginary part apparent area, m2 real contact area, m2 surface roughness power spectrum, m4 separation, m fractal dimension Young's elastic modulus, N/m2 ℎ height, m ℏ Planck constant divided by 2π, J·s microhardness, GPa Boltzmann constant, J/K harmonic mean thermal conductivity, W/mK vacuum wavevector, rad/m ∥ wavevector parallel to the interface, rad/m wavevector normal to the interface, rad/m wavevector, 1/m effective absolute surface slope pressure, N/m2 heat flux, W/m2 heat flow, W Fresnel reflection coefficients at interface i-j R thermal contact resistance, m2K/W R thermal constriction resistance, m2K/W thermal resistance of gap media, m2K/W thermal radiation resistance, m2K/W temperature, K length parameter, m Poisson ratio Y separation of mean planes, m Re real part 1. Introduction Thermal contact resistance (TCR) is a core parameter weighing the heat transfer efficiency among the interfaces of different components. The phenomena of contact heat transfer are universal in many engineering fields such as aerospace, electronic packaging, cryogenics, and mechanical manufacturing [1-3]. Researches have pointed out that the thermal budget in thermal interface can account for half of the total in some microelectronic packages, which 2 directly restricts the reliability, performance and lifetime of products [4]. So, it is crucial to qualitatively and quantitatively predicting the TCR in order to rationally carry out thermal management and design. As shown in Fig.1, the mechanism of the TCR is very complex and it is a result of triple interaction among geometry, mechanics and thermal [5]. In thermal transfer problem, there exist three modes including conduction of discrete point contact, convection of gap medium and radiation from a perspective of macroscopic scale, and they can also be interpreted as the transport in the form of energy carriers (i.e., phonons, electrons and photons) from a perspective of micro-nano scale [6, 7]. Fig. 1. Schematic for mechanism of thermal contact resistance A large number of experiments have been implemented via diverse characterization methods to study the interfacial heat transfer that the interfacial separation is from atomic level to engineering application level (~μm) [2]. The researchers have investigated many factors with regard to the TCR, which includes the materials, the surface roughness and waviness, the interface temperature, the direction of the heat flux, the contact pressure, load cycle and the contact regions under different resolution [8-11]. However, the previous theoretical studies or models for the TCR usually ignore the influence of interfacial thermal radiation since the radiation effect governed by the Stefan-Boltzmann law on TCR is negligible compared to asperities contact heat conduction under the condition of low or room temperature. With the development of engineering technology, the minimum feature size of the device structure has been in the order of microns and continued to develop on the order of nanometers. At the micro- nano scale, the transport of matter and energy in any physical process takes place in confined microscopic geometries, such that the transport behavior of matter and energy exhibits different size effects at the macroscale [7]. In the late 1950s, Cravalho and Tien et al. [12] found a phenomenon that net radiative energy increases with decreasing separation and rapidly 3 attenuated as the surface spacing increasing between two dielectrics, that is, near-field thermal radiation. The first correct calculation of near-field thermal radiation was done by Polder and Van Hove employing fluctuational electrodynamics [13]. More and more papers have reported that the near-field thermal radiative heat flux between two planar surfaces separated by a nanosized vacuum gap can exceed several orders of magnitude of the blackbody limit or even achieve a similar magnitude of the thermal conduction, on account of the electromagnetic evanescent waves, photon tunneling effects and excitation of surface polaritons [14-17]. As mentioned above, there are three main paths for heat to flow across the interface. The conduction and conduction with interstitial fluid have been proved to be significant for interfacial heat transfer in low or normal temperature. The thermal radiation in near field indicates that the energy transfer is mainly dependent on dielectric function of material and separation distance, and one should analyze it using fluctuational electrodynamics rather than conventional radiative transfer equation (RTE) based on particle features [17, 18]. In practice, the direct contact area between rough surfaces is much smaller than the apparent area, e.g., the diameter of the contact regions observed at atomic resolution may be of the order of ~1 nm [19, 20]. The enhancement of thermal radiation occurs when emitter and receiver are separated within characteristic wavelength () obtained from Wien's displacement law. According to the formula (∙=2898 ∙ ), the characteristic wavelength of thermal radiation is about 10 μm at a temperature of 300 K. Obviously, the surface roughness of engineering interest is comparable or less than this magnitude. However, little attention has been paid to the near-field radiative effect on the TCR. The first work considering the near-field thermal radiation effect on the TCR was presented by Persson et al. [20, 21] using proximity approximation and their rough estimation of the MEMS device concludes that the non-contact contribution to heat transfer coefficient is larger than or of similar magnitude as the contribution from the area of real contact. In this paper, we analyze the interfacial heat transfer of four metals including Al, Cu, Ag and Pb using the classical CMY TCR model as well as an ab initio calculation of thermal radiation to account for the heat transfer contribution of the non-contact regions. Among them, aluminum and amorphous alumina are chosen to carry out a coupling simulation for revealing the contribution between thermal conduction and near-field thermal radiation at a thermal 4 contact interface. Finally, specific surfaces are constructed to further analyze and compare. 2. Methodology 2.1 Thermal contact resistance To meet a wide range of thermal management applications, a number of experimental, analytical, numerical models to estimate the TCR have been developed. However, it is hard to derive a general precise predictive model due to the complexity of thermal joint or interface [22]. The TCR is defined as: (1) where ∆ is the temperature drop at the interface, is the heat flux along the normal direction of the interface, is the heat flow and is the apparent area of the interface. The R=∆ = ∆ ⁄ reciprocal of the TCR is the thermal contact conductance (TCC). Herein, we focus on the contact of metal surfaces of engineering interest that the roughness is mostly lower than 10 μm. Assuming the surfaces are microscopically rough and macroscopically conforming (i.e., nominally flat rough surfaces). Yovanovich et al. [23] developed a thermal contact correlation based on the classical Cooper-Mikic-Yovanocich (CMY) model considering both microscopic and macroscopic characteristics of a joint, which gives simple relationships for three measurable parameters of the contact surfaces: the geometric parameters, the mechanical parameters, the thermal parameters. The correlation has been verified comparing with plenty of experimental results and it is quite accurate for optically conforming surfaces [22, 24-26]: R=2σ√2 . (2) thermal conductivity, = where =2/(+) is + is the effective absolute surface slope, =+ is the effective RMS surface roughness. is a dimensionless parameter linking the geometry and the mechanics of ==√2(2 ⁄ ) (3) where is the inverse transformation of the complementary error function, is the nominal pressure and is the microhardness of the softer metal. the joint and it is defined as: exp ( 2⁄)1−0.5√2⁄ the harmonic mean 5 [27, 28]: Most surfaces produced by machining or grinding are Gaussian surfaces, that is, asperities are randomly distributed over the surface but isotropic and their profile heights obey the Gaussian distribution [22]. For Gaussian surfaces the empirical correlations to relate the effective RMS surface roughness, , to the effective absolute surface slope, , is as follow (σ)=0.124σ., σ≤1.6 μm σ>1.6 μm (4) 0.076σ., Y=1.185σ−3.132. (5) The distance between the mean planes can be derived via a detailed geometric analysis about interacting of two rough surfaces and the explicit correlation is expressed as [26]: In this part, the classical thermal contact equations and other additional correlations are introduced to calculate the interface thermal resistance responsible for the sum of spot-to-spot contact in a vacuum condition. To be more explicit, Fig. 2 shows a schematic of a contact interface between conforming rough surfaces and the main geometry parameters within the equations above. Furthermore, the TCR can also be expressed using thermal resistance network: 1= 1+ 1+ 1 (6) where and are resistance produced by gap media and radiation resistance respectively. Here, we only consider the situation of contact in vacuum that no interstitial fluids are present. Fig. 2. Schematic of a contact interface between conforming rough surfaces 2.1.1 Surface roughness power spectrum The CMY model above is used to preliminarily compare the heat transfer coefficient of contact conduction with radiative heat transfer of non-contact regions estimated by 6 = =√ amplitude as a function of the PSD can be determined by fluctuational electrodynamics. Then, we perform the further calculation via simulating specific surfaces with corresponding surface roughness power spectrum (PSD). The PSD can provide the characteristic information of a surface in wavevector space rather than real space, which is a more effective approach to describe the contact situation [20]. The PSD is defined as [29, 30]: According to the contact mechanism proposed by Persson et al. [20, 31], in the case of low squeezing pressure, the heat transfer coefficient associated with the area of real contact is constructed: 1(2)〈ℎ()ℎ()〉∙ (7) where =(,) and =(,). ℎ() is the profile height measured from the mean plane. The sign 〈…〉 represents the ensemble averaging. The mean square roughness =〈ℎ〉=2 (8) () ℎ= ∗ (9) (10) () () where ∗=((1−)/+(1−)/)) is the effective Young's elastic modulus that is determined by the Young's elastic modulus and Poisson ratio of each solid. is a length For the case of a self-affine fractal surface with a fractal dimension ≤2.5, Refs. [32, 33] give the relation between the interfacial separation () and the normal load (): ≈0.7493∗ (11) 2 where is a parameter that seems to relate to the surface roughness. parameter which is also determined from the PSD. sum of the power spectra of the individual surfaces: In addition, Majumdar et al. [34] have proved that the PSD of the equivalent surface is the =+ (12) 2.2 Radiative heat transfer 7 In previous research on the thermal interface, the radiation heat conductance is always neglected due to the complex manner of the bonding solids, and the contribution is relatively small compared with thermal conduction unless the surfaces have a high emissivity and are formed by rough, low-conductivity solids under light contact pressures [26]. However, previous estimations for the heat flux of interfacial radiation are based on the Stefan-Boltzmann law which is not included evanescent mode. Here, the radiative heat transfer is investigated employing an ab initio calculation based on the stochastic Maxwell equations and fluctuational electrodynamics [16,35]. The two components contacting with each other can be considered as semi-infinite (one-dimensional approximation), isotropic and non-magnetic. The spectral heat radiative flux calculation accounts for both the far- and the near-field effects [35, 36]: ,= (,) ∫ ∥∥ ∫ ∥∥ ,=(,) () (13a) ()+ ()() () (13b) ()+ () ()() ()() monochromatic radiative heat flux between two bulks, respectively. =∥+ where is the wavevector in vacuum, ∥ and are respectively the wavevectors parallel to and normal to the interface. is the Fresnel reflection coefficients from medium to medium for polarization state ( for TE and for TM). The mean energy of a Planck oscillator in thermal equilibrium at temperature of the source medium and angle frequency , (,), is given by )−1 (14) where ℏ is the Planck constant divided by 2π and is the Boltzmann constant. ∆ (15) ,+, where is the total radiative heat flux obtained by integrating over and it depends on the where Eqs. (13a) and (13b) are the propagating and evanescent contributions to the (,)= =− ℏ exp(ℏ ⁄ = ∫ The radiative resistance is defined as: separation of two bulks, the temperature and the optical response of the materials. 2.2.1 Dielectric function and parameters of materials 8 The dielectric function ε reflects the response of the materials to the external electric + (16) where denotes high-frequency contributions, is the plasma frequency, is the field. The dielectric function of metal and semiconductor can be described by the Drude model [14]: ()≡+i=− damping frequency. Table 1 lists the parameters in regard to the dielectric function [37] and physical property [5, 24, 38] of the materials investigated commonly both in the TCR and near- field radiation. In practice, the metal surface always has an oxide layer. The amorphous alumina is taken as an example to estimate the effect of a dielectric on near-field radiative heat transfer. The dielectric function of the film is described by the classical oscillator model: (17) where is the number of oscillators, , , and , are the strength, TO mode frequency and LO mode frequency of the th oscillator, respectively. The value of the ()=+ , , −− parameters obtained from Ref. [39] is also list at table 1. Table 1 The parameters of simulation materials (1) (GPa) (W/m·K) 0.912 1.089 0.745 0.040 (1/s)∙10 3.196 3.327 174 381 429 35.3 , (rad/s)∙10 1.012 1.806 (rad/s)∙10 (1/s)∙10 1 1 1 1 (1) 2.8 2.8 2.242 1.202 1.366 1.168 (1) 3.75 1.46 1.219 0.524 0.273 2.731 , (rad/s)∙10 0.795 1.358 9 Parameter Material Al Cu Ag Pb Al2O3 n=1 n=2 2.3 The coupling simulation strategy Fig. 3. Schematic of the coupling simulation strategy Our intention is to investigate the influence of the near-field radiation at the interface when two bodies contacting together. CMY model and PSD method give the relation about conduction resistance accounting for the contact regions. The heat flux across these regions is determined by surface profile, roughness, apparent pressure, material, etc. On the other side, the radiative heat transfer accounting for non-contact regions are estimated using an ab initio calculation that it comprises the contribution of both the propagating and evanescent modes. In evanescent mode, the heat flux depends on material, temperature difference and separation distance. Obviously, some of the parameters of radiative heat flux are associated with the equations of thermal contact correlation, which means that there is a coupling relation. Fig. 3 describes the coupling simulation strategy to analyze interface heat transfer between conforming rough surfaces. The numerical algorithm of one-dimensional near-field thermal radiation problems is according to Francoeur et al. [36]. In this work, the integral terms are 10 as π∗ that is the distance between atoms. The convergence criteria of the calculated using a composite Simpson's rule and the upper limit of integral for Eq. (13b) is set integration are that the relative errors for propagating mode and evanescent mode are ≤1 and ≤0.01, respectively. The CMY model enables to roughly give us the insight that the near-field radiation plays an importance in which situation and then the specific surfaces are constructed to carry out a more rigorous investigation utilizing PSD. 3. Results and discussion 3.1 Heat transfer through the contact regions The case of interface heat transfer in regard to the contact regions in vacuum is first investigated. In this area, heat flow transfers across the interface via microchannels constructed by the discrete asperities distributing on the upper and lower surfaces. If the heat leaves the half-space (body) through a small area (asperity), it forms an additional thermal resistance called constriction resistance due to the contraction of the flux lines. The converse is called spreading resistance. Therefore, there is a very complicated thermal resistance network in the interfacial heat transfer and the resistance network can also be determined by the macro measurable quantities in Sec. 2. Fig. 4 shows the curves of the TCR versus apparent pressure under different orders of magnitude of effective roughness of Al. Obviously, the TCR decreases with increasing pressure but increases with roughness. This tendency is in conformity with previous experimental investigation and theoretical models. In Fig. 5, we calculate the TCR of different metal materials with the roughness of a nanometer versus clamping pressure. It is of interest that in relatively high pressure lead demonstrates a well thermal dissipation at the interface compared with other three kind of metals although it possesses the lowest thermal conductivity. Because of low microhardness of the lead, the number of heat channels (i.e., microcontact spots) increases rapidly when the interface is subjected a larger load. The TCR of other three metals is mainly affected by their intrinsic thermal conductivities in the case of the same effective roughness and pressure. The Fig. 4 and Fig. 5 indicate a range of magnitudes of the TCR change with the effective roughness at different scales, which are compared with following thermal radiation resistance and analyze the role of near-field radiation at interfacial heat transfer. 11 Fig. 4. Influence of clamping pressure on the TCR under different roughness Fig. 5. Influence of clamping pressure on the TCR for different materials at 1 nm effective roughness 3.2 Heat transfer through the non-contact regions According to the CMY TCR model, the ratio of real contact area to nominal area can be related to the apparent pressure and microhardness using the force balance condition: = (18) In this work, the maximum pressure is set as 10.0 MPa and the minimum microhardness is 0.04 GPa so that the real contact area takes up ≤ 25% of the apparent area. For aluminum and copper, the ratio is less than 1%. Thus, the radiation effect is estimated by modeling the heat transfer across the micro-gap as equivalent to near-field thermal radiation between two 12 infinite isothermal smooth plates, and the separation is assumed to equal to the distance between mean planes determined by effective roughness, pressure and microhardness. In the following simulation, the interfacial temperature difference used is 10 K, where the emitter (hot side) is 310 K and the receiver (cold side) is 300 K. The spectral radiative heat flux is enhanced in the near-field as shown in Fig. 6. At a separation of 1 nm, the spectral radiative heat flux comprised propagating mode and evanescent mode of the four metal materials are several orders of magnitude larger than the estimation for blackbodies (black line) that is independent with separation distance. The shapes of the curves are mainly determined by the dielectric function and the roughness effect on the near-field radiation is neglected. Fig. 6. Near-field spectral heat flux of different materials at a separation of 1 nm The near-field thermal radiation between amorphous alumina (α-Al2O3) is also estimated for a rough approximation of the oxidation of aluminum. Note that the oxide layer is also simulated as bulk materials to simplify the structure and calculation. It can be seen in Fig. 7 that there are two peaks in the mid-infrared and at a separation of 1 nanometer the spectral radiative heat flux is seven orders of magnitude larger than blackbodies radiation and three orders of magnitude larger than the aluminum. The peaks account for the dielectrics that can support surface phonon-polaritons (SPhP). These surface electromagnetic waves are resonantly excited and provide the considerable contribution to the density of energy in the near-field [40]. An inset graph in the upper right corner is monochromatic evanescent component of the radiative heat flux per unit ∥ at a separation of 100 nm. The two bright bands confined to a 13 narrow range of frequencies in the inset correspond to the peaks nearby the angle frequency of 1.18×1014 rad/s and 2.0×1014 rad/s, respectively. When the separation achieves micrometer scale, this enhancement effect induced by the SPhP fades away. Fig. 8 illustrates the radiation resistance of Al and Al2O3. The overall radiative heat transfer performance of amorphous alumina is higher than that of aluminum. Fig. 7. Spectral radiative heat flux of the Al2O3 at different separation Fig. 8. Radiation resistance of the Al and Al2O3 at different scales The interfacial temperature effect on the thermal radiation is also estimated at the separation of 0.1 μm and the cold side is fixed at 300 K. It can be seen in Fig. 9 the radiation 14 heat transfer coefficient is approximately linearly related to the temperature difference in a specific range and the coefficient of the Al is not sensitive to the temperature changes compared with the Al2O3. Fig. 9. Radiation heat transfer coefficient as a function of the interface temperature difference 3.3 Comparison and analysis of results 3.3.1 Results based on CMY As shown in Fig. 5 and Fig. 6, the Al possesses relative high TCR and near-field radiative heat transfer coefficient, which means that there is more likely to exist significant synergy at the interface. Furthermore, aluminum is a kind of frequently-used material in engineering. Therefore, the analysis and comparison are centered around the aluminum in this section. We first consider the heat transfer of both the thermal conduction and thermal radiation at the nanoscale as shown in Fig. 10. The effective roughness is fixed at 1 nm. As the applied pressure decreases, the mean plane spacing increases and the contact thermal conductance drops sharply. When the pressure is lower than 1 MPa, the near-field radiation of the Al2O3 is comparable to that of thermal conduction of the Al and gradually dominates the heat transfer. The inset with logarithmic plot of y axis shows the comparison between thermal conduction contribution and thermal radiation contribution of the Al, which enables to clearly display the case of extreme low pressure and large gap at the interface. 15 Fig. 10. Thermal conduction and thermal radiation at the nanoscale When the pressure is fixed, the mean plane separation between two same solids and the thermal conduction resistance are mainly determined by the effective roughness. As shown in Fig. 11, the thermal conduction resistance and radiation resistance as a function of the scales of effective roughness at a low pressure of 1 kPa are investigated. Fig. 11a illustrates that the thermal radiation of the Al2O3 is dominant when the effective roughness is lower than 0.1 μm and produces the comparable effect to the thermal conduction at microscale. The thermal radiation of Al is also dominant in the nanoscale due to the contribution of the evanescent mode but it rapidly decreases and can be neglected at microscale. Fig. 11. The TCR versus the effective roughness at a pressure of 1 kPa In order to clearly demonstrate the contribution of heat transfer from contact regions and non-contact regions, Fig. 12 shows the relative contribution from thermal radiation and thermal 16 conduction as a function of the effective roughness based on the simulations of Fig. 11. From the tendency shown in Fig. 12a, the contribution of the thermal radiation of the Al2O3 may be larger than the thermal conduction again when the effective roughness (i.e., gap distance) becomes larger. Fig. 12b indicates that the near-field thermal radiation effect of pure metal aluminum should not be ignored at the effective roughness of submicron-scale, which approximately takes up 50% of the contribution to the interfacial heat transfer. At the separation of 100 nm, both the thermal radiation of the Al (ℎ =110.288 W/m2K) and the Al2O3 (ℎ =235.569 W/m2K) still exceed the blackbody radiation limit (ℎ =64.366 W/m2K). It means that the near-field effect on the thermal radiation should be considered in the investigation of thermal contact. Fig. 12. Relative contribution from thermal radiation and thermal conduction as a function of 3.3.2 Results based on PSD the effective roughness The analysis results from CMY model shows that when the effective roughness lower than submicron the impact of near-field thermal radiation is significant. Herein, we constructed three artificial randomly rough surfaces over an area of 10 μm × 10 μm and only consider the material of Aluminum. As shown in Fig. 13, a 3D fractal surface with effective RMS of 10 nm is modelled (Fig. 13a) and Fig. 13b shows 2D FFT of the surface topography. The symmetric power spectra indicate that the surface is isotropic [41]. The PSD profile of this surface has a 17 cut-off wavevector () of 107 m-1 and a slope of -3.203. It means that the fractal dimension is 2.4 since for a self-affine surface the PSD has the power-law behavior [29] ~() . The wavevectors lower than the 107 m-1 determine the RMS. Fig 13d shows the height probability distribution of the surface, which is nearly Gaussian. The other two modeling surfaces (not shown) have the same properties but different effective RMS of 50 nm and 100 nm. We study such range of scale because the validity of fluctuational electrodynamics for sub-10-nm gap distances is still questionable. Fig. 13. Artificial randomly rough surface with effective roughness of 10 nm It can be seen in Fig. 14 that the near-field radiation is seem to be weak at the effective roughness of 50 nm and 100 nm, which differs from the results based on CMY. The main reason is that the intensity of the near-field effect decays with the square of the distance and the separation distances used in near-field calculation are estimated via Eq. (11) and this equation assumes asperities undergone elastic deformation. Therefore, the real separation is several times larger than the effective roughness. The calculation results of gap distance of these three cases are 35.5 nm, 197.6 nm and 412.5 nm, respectively. The Eq. (9) and Eq. (10) are used to calculate the thermal conduction from contact regions. 18 Fig. 14. Heat transfer coefficient comparison at three modeling surfaces Fig. 15 shows the case of the modelling surface with an RMS of 10 nm. With the increase of squeezing pressure, the interfacial separation and thermal conduction resistance decrease observably but the near-field effect is not sensitive to the pressure in a small range. In this condition, the TCR is determined by the interaction of thermal conduction and near-field thermal radiation. Fig. 15. The near-field thermal radiation effect on the TCR of modelling surface with an 4. Conclusion RMS of 10 nm The role of near-field thermal radiation in thermal contact resistance has been preliminarily analyzed via the classical TCR CMY model and then further investigation are carried out utilizing PSD for specific rough surfaces. In most cases the separation between rough surfaces of engineering interest satisfies the condition of producing the near-field radiation effect at room temperature. The non-contact area accounts for the absolute dominance 19 compared with the real contact area when two bodies come into bare contact. The simulation results show that the radiative heat flux of dielectric is obviously larger than that of metal because of the thermally excited SPhP in the near field. Therefore, the radiation effect of the dielectric on the TCR should be considered within a wider range of effective roughness than metal. At low pressure, a significant synergy exists between the thermal conduction and thermal radiation at the interface. When the effective roughness is lower than the scale of 0.1 μm, the near-field radiation in non-contact regions may make a difference to total heat transfer at room temperature. Simultaneously, the estimated radiative heat fluxes in such a spacing exceed the value predicted by the Stefan-Boltzmann law of blackbody. It easily achieves low pressure and low surface roughness conditions in microelectronics industry. The presented study is expected to provide a guidance for precisely predicting thermal contact problem. Acknowledgments The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors acknowledge the financial support provided by National Natural Science Foundation of China (Project No. 51506033), Innovation Project of GUET Graduate Education (Grant No. 2016YJCX18), Guangxi Natural Science Foundation (Grant No. 2017JJA160108)). References: [1] M. Grujicic, C.L. Zhao, E.C. Dusel, The effect of thermal contact resistance on heat management in the electronic packaging, Appl. Surf. Sci. 246(2005) 290-302. [2] Y. Xian, P. Zhang, S. Zhai, P. Yuan, D. Yang, Experimental characterization methods for thermal contact resistance: a review, Appl. Therm. Eng. 130(2018) 1530-1548. [3] S. Zhai, P. Zhang, Y. Xian, J. Zeng, B. Shi, Effective thermal conductivity of polymer composites: theoretical models and simulation models, Int. J. Heat Mass Tran. 117(2018) 358-374. [4] B. Smith, T. Brunschwiler, B. Michel, Comparison of transient and static test methods for chip-to-sink thermal interface characterization, Microelectron. J. 40(2009) 1379-1386. [5] M.M. Yovanovich, Four decades of research on thermal contact, gap, and joint resistance in microelectronics, IEEE transactions on components and packaging technologies 28(2005) 182-206. [6] P. Zhang, P. Yuan, X. Jiang, S. Zhai, J. Zeng, Y. Xian, H. Qin, D. Yang, A theoretical review on interfacial thermal transport at the nanoscale, Small 14(2018) 1702769. [7] G. Chen, Nanoscale energy transport and conversion: a parallel treatment of electrons, molecules, phonons, and photons, Oxford University Press, 2005. [8] R. Dou, T. Ge, X. Liu, Z. Wen, Effects of contact pressure, interface temperature, and surface roughness on thermal contact conductance between stainless steel surfaces under atmosphere condition, Int. J. Heat Mass 20 Tran. 94(2016) 156-163. [9] P. Zhang, T. Cui, Q. Li, Effect of surface roughness on thermal contact resistance of aluminium alloy, Appl. Therm. Eng. 121(2017) 992-998. [10] P. Zhang, Y. Xuan, Q. Li, A high-precision instrumentation of measuring thermal contact resistance using reversible heat flux, Exp. Therm. Fluid Sci. 54(2014) 204-211. [11] R. Joseph, J. Peter, S. Sunil Kumar, N. Asok Kumar, Effect of thermal and load cycle on thermal contact conductance across dissimilar joints at cryogenic temperature, Appl. Therm. Eng. (2016). [12] E.G. Cravalho, C.L. Tien, R.P. Caren, Effect of small spacing on radiative transfer between two dielectrics, Journal of Heat Transfer 89(1967) 351-358. [13] D. Polder, M. Van Hove, Theory of radiative heat transfer between closely spaced bodies, Physical Review B 4(1971) 3303. [14] S. Basu, Z.M. Zhang, C.J. Fu, Review of near-field thermal radiation and its application to energy conversion, Int. J. Energ. Res. 33(2009) 1203-1232. [15] M.P. Bernardi, D. Milovich, M. Francoeur, Radiative heat transfer exceeding the blackbody limit between macroscale planar surfaces separated by a nanosize vacuum gap, Nat. Commun. 7(2016) 12900. [16] A.I. Volokitin, B.N.J. Persson, Near-field radiative heat transfer and noncontact friction, 79(2007) 1291- 1329. [17] M. Francoeur, M. Pinar Mengüç, Role of fluctuational electrodynamics in near-field radiative heat transfer, Journal of Quantitative Spectroscopy and Radiative Transfer 109(2008) 280-293. [18] J.R. Howell, M.P. Menguc, R. Siegel, Thermal radiation heat transfer, CRC press, 2010. [19] Y. Mo, K.T. Turner, I. Szlufarska, Friction laws at the nanoscale, Nature 457(2009) 1116. [20] B.N.J. Persson, B. Lorenz, A.I. Volokitin, Heat transfer between elastic solids with randomly rough surfaces, The European Physical Journal E 31(2010) 3-24. [21] S.A. Biehs, J.J. Greffet, Influence of roughness on near-field heat transfer between two plates, Physical Review B 82(2010). [22] M. Bahrami, J.R. Culham, M.M. Yananovich, G.E. Schneider, Review of thermal joint resistance models for nonconforming rough surfaces, Appl. Mech. Rev. 59(2006) 411-431. [23] T.E. Horton, Spacecraft radiative transfer and temperature control, American Institute of Aeronautics and Astronautics, 1982. [24] H. Yüncü, Thermal contact conductance of nominaly flat surfaces, Heat Mass Transfer 43(2006) 1-5. [25] L. Dong-Huan, S. Xin-Chun, The physical-mechanism based high-temperature thermal contact conductance model with experimental verification, Chinese Phys. Lett. 30(2013) 36501. [26] A. Bejan, A.D. Kraus, Heat transfer handbook, John Wiley & Sons, 2003. [27] V.W. Antonetti, T.D. Whittle, R.E. Simons, An approximate thermal contact conductance correlation, J. Electron. Packaging 115(1993) 131-134. [28] M.A. Lambert, L.S. Fletcher, Thermal contact conductance of spherical rough metals, Journal of Heat Transfer 119(1997) 684-690. [29] B.N.J. Persson, O. Albohr, U. Tartaglino, A.I. Volokitin, E. Tosatti, On the nature of surface roughness with application to contact mechanics, sealing, rubber friction and adhesion, 17(2005) R1-R62. [30] P.R. Nayak, Random process model of rough surfaces, Journal of Lubrication Technology 93(1971) 398-407. [31] B.N.J. Persson, Contact mechanics for randomly rough surfaces, Surf. Sci. Rep. 61(2006) 201-227. [32] B.N.J. Persson, Relation between interfacial separation and load: a general theory of contact mechanics, Phys. Rev. Lett. 99(2007). [33] M. Mahboob Kanafi, A.J. Tuononen, Top topography surface roughness power spectrum for pavement 21 friction evaluation, Tribol. Int. 107(2017) 240-249. [34] A. Majumdar, C.L. Tien, Fractal network model for contact conductance, Journal of Heat Transfer 113(1991) 516-525. [35] J. Mulet, K. Joulain, R. Carminati, J. Greffet, Enhanced radiative heat transfer at nanometric distances, Microscale Thermophysical Engineering 6(2002) 209-222. [36] M. Francoeur, M. Pinar Mengüç, R. Vaillon, Solution of near-field thermal radiation in one-dimensional layered media using dyadic green's functions and the scattering matrix method, Journal of Quantitative Spectroscopy and Radiative Transfer 110(2009) 2002-2018. [37] M.A. Ordal, L.L. Long, R.J. Bell, S.E. Bell, R.R. Bell, R.W. Alexander, C.A. Ward, Optical properties of the metals al, co, cu, au, fe, pb, ni, pd, pt, ag, ti, and w in the infrared and far infrared, Appl. Optics 22(1983) 1099-1119. [38] C.T. Merrill, S.V. Garimella, Measurement and prediction of thermal contact resistance across coated joints, Exp. Heat Transfer 24(2011) 179-200. [39] Y.T. Chu, J.B. Bates, C.W. White, G.C. Farlow, Optical dielectric functions for amorphous al2o3 and γ‐ al2o3, J. Appl. Phys. 64(1988) 3727-3730. [40] A.V. Shchegrov, K. Joulain, R. Carminati, J.J. Greffet, Near-field spectral effects due to electromagnetic surface excitations, Phys. Rev. Lett. 85(2000) 1548. [41] W. Yan, K. Komvopoulos, Contact analysis of elastic-plastic fractal surfaces, J. Appl. Phys. 84(1998) 3617- 22 3624.
1906.01690	1	1906	2019-06-04T19:23:59	Non-reciprocal Wave Phenomena in Energy Self-reliant Gyric Metamaterials	[ "physics.app-ph" ]	This work presents a mechanism by which non-reciprocal wave transmission is achieved in a class of gyric metamaterial lattices with embedded rotating elements. A modulation of the device's angular momentum is obtained via prescribed rotations of a set of locally housed spinning motors and are then used to induce space-periodic, time-periodic, as well as space-time-periodic variations which influence wave propagation in distinct ways. Owing to their dependence on gyroscopic effects, such systems are able to break reciprocal wave symmetry without stiffness perturbations rendering them consistently stable as well as energy self-reliant. Dispersion patterns, band gap emergence, as well as non-reciprocal wave transmission in the space-time-periodic gyric metamaterials are predicted both analytically from the gyroscopic system dynamics as well as numerically via time-transient simulations. In addition to breaking reciprocity, we show that the energy content of a frictionless gyric metamaterial is conserved over one temporal modulation cycle enabling it to exhibit a stable response irrespective of the pumping frequency.	physics.app-ph	physics	Non-reciprocal Wave Phenomena in Energy Self-reliant Gyric Metamaterials M. A. Attarzadeh,1 S. Maleki,1 J. L. Crassidis,1 and M. Nouh1, a) 1Mechanical & Aerospace Engineering Dept., University at Buffalo (SUNY), Buffalo, NY 14260-4400, USA (Dated: 6 June 2019) This work presents a mechanism by which non-reciprocal wave transmission is achieved in a class of gyric metamaterial lattices with embedded rotating elements. A modulation of the device's angular momentum is obtained via prescribed ro- tations of a set of locally housed spinning motors and are then used to induce space-periodic, time-periodic, as well as space-time-periodic variations which influ- ence wave propagations in distinct ways. Owing to their dependence on gyroscopic effects, such systems are able to break reciprocal wave symmetry without stiffness perturbations rendering them consistently stable as well as energy self-reliant. Dis- persion patterns, band gap emergence, as well as non-reciprocal wave transmission in the space-time-periodic gyric metamaterials are predicted both analytically from the gyroscopic system dynamics as well as numerically via time-transient simula- tions. In addition to breaking reciprocity, we show that the energy content of a frictionless gyric metamaterial is conserved over one temporal modulation cycle enabling it to exhibit a stable response irrespective of the pumping frequency. Keywords: Gyroscopic; metamaterial; wave propagation: reciprocity I. INTRODUCTION Inertial and elastic components represent the building blocks of metamaterials and phononic crystals which have been predominantly utilized to manipulate incident waves over the past few decades1 -- 4. Wave control mechanisms have thus far been limited to modulations of mass and stiffness in spatial domains5 -- 8, temporal domains9, or both10. Alternatively, angular momentum arising from the rotation of distributed masses combined with the gyroscopic effect can bring about intriguing mechanical features. Examples include gyroscopic stabilization11 -- 13, vibration control through distributed and discrete networks of gyroscopes14,15, spacecraft attitude control16, shielding cloaks17 and, most recently, back- scattering immune edge states18,19. In the latter, gyroscopic metamaterials are created by a two-dimensional network of objects rotating at a constant speed with the purpose of real- izing topological edge states inspired by the Quantum Hall Effect. Contrary to dissipative elements, gyroscopic effects (herein called Gyric after D'Eleuterio20) are of a conservative nature, and do not drain the system's mechanical energy since gyroscopic forces are always orthogonal to the velocity vector. Furthermore, unlike the mass and stiffness, which are understandably difficult to modulate, angular momentum is not an inherent property of the medium but rather an outcome of a rigid body rotation (e.g. a rotor) at a desired speed. As such, a system's angular momentum can take on different values (small or large, positive or negative) and is simply tuned by adjusting the rotation speed in real-time. In this effort, we examine dispersion characteristics, as well as non-reciprocal wave phe- nomena, in a class of gyric metamaterials (GMMs) which comprise a set of embedded spinning rotors. Unlike conventional rotating periodic structures21, GMMs presented here are capable of incorporating angular momentum modulation in space, time, or both space and time independent of its constitutive mass and stiffness matrices. Gyric effects appear in a)Electronic mail: [email protected] (Corresponding author) 2 the governing GMM dynamics as moments induced by a change in the angular momentum vector. A specific interest of this work is the illustration of non-reciprocal wave phenom- ena in GMMs stemming from breaking their wave transmission symmetry. Breaking wave reciprocity in linear systems has been a growing focus of recent acoustic and elastic meta- materials research22 -- 24. It has been shown that the induction of linear or angular motion, whether physically25 -- 27 or artificially28 -- 30 can instigate non-reciprocal wave propagation. On the realization front, some efforts have proposed the achievement of non-reciprocal sys- tems via external stimulation of adaptive structures, e.g. using electrical fields31, magnetic fields32, or alternatively via torsional mechanical waves33. Nonetheless, most if not all of these investigations target a modulation of the system's stiffness properties, which risks influencing structural aspects. Stiffness modulations also often require hard-wiring as well as electrical shunt circuits rendering them experimentally tedious34. In addition to operating with an unperturbed stiffness matrix, non-reciprocal GMMs have the ability to exhibit negative values of angular momentum h(x, t) at any location x or time instant t, simply by reversing the direction of rotation. The latter can be a cumbersome task in stiffness-modulated systems, or at least not feasible without an added layer of active control. As a result, they provide a much larger range of parameters and design flexibility. Finally, as will be shown, time-periodic GMMs reliably exhibit a stable response and a bounded energy content, irrespective of the pumping frequency required to onset the modulation; a feature which is not possessed by systems with time-periodic elastic fields35. FIG. 1. (a) Schematic of a gyric body: Rigid body B with a spinning rotor R. (b) A non-gyric body with dynamically equivalent rotational inertia. (c) GMM lattice comprising a set of interconnected gyric cells with embedded spinning rotors II. STRUCTURAL DYNAMICS OF A GYRIC BODY Inspired by previous efforts on gyro-elastic continua20 and the modified Newton's sec- ond law36, we start by investigating the linear dynamics of a gyric rigid body. The body comprises a distributed mass B with an embedded spinning rotor R, as depicted in Fig. 1a, which represents the building block of a periodic gyric structure. Unlike conventional elastic metamaterials where the outer body is connected to an internal resonator and transmits a force to it37,38, the interaction between the outer and inner components here is rather a gyroscopic moment emerging as a result of the change in the total angular momentum vec- tor of the gyric body (i.e. B + R). The same moment is also responsible for the precession phenomenon of a spinning top39. The magnitude of the gyroscopic moment is proportional yxJx ,MxJyRByJx,eff ,MxJy,effD(a)(b)unit cellbearings(c)GMMx3 to R's angular momentum as well as B's rate of rotation. The direction of the moment, however, is governed by the right-hand rule and is therefore binormal to both of the rota- tional directions of R (z-axis) and B (x-axis). As a result, it influences the dynamics of the system in the y-direction. The previous instigates cross-coupling between the dynamics of the body in the x- and y-directions. We assume the system is fixed about the z-axis and study the two rotational degrees of freedom θx and θy. In the absence of elastic (restor- ing) and frictional forces, the governing equations of motion are found by a simple moment balance as Jx Jy θx + h θy = Mx θy − h θx = 0 (1a) (1b) where Jx and Jy are the moments of inertia about the shown principal axes, and Mx is the external moment applied to the outer mass in the x-direction. Also, θx,y and θx,y rep- resent the angular velocities and accelerations in the respective directions. The coupling term appears solely as a result of including the rotor's angular momentum h in the sys- tem dynamics. The steady-state response amplitude of Eq. (1) to a harmonic moment of amplitude Mx and frequency ω is given by θx = Mx h2/Jy − Jxω2 (2) Using Eq. (2), the gyric body can be alternatively presented by a dynamically equivalent non-gyric one (D in Fig. 1b) with an effective inertia Jx,eff, given by (cid:20) (cid:21) 1 − ω2 h ω2 Jx Jx,eff = (3) which mimics the frequency-dependent behavior of the gyric body B. In Eq. (3), ωh denotes the equivalent rotational speed of R with ω2 . Note that a similar result could be obtained for Jy,eff by replicating this equivalence in the y-direction. Fig. 2 displays variation of the effective inertia Jx,eff with frequency for different values of ωh. The figure shows that the spinning rotor induces an artificial rotational stiffness h2/Jy in the dynamics of the system which in turn yields a negative effective inertia for frequencies in the range 0 ≤ ω < ωh; a phenomenon which is highly desirable in low-frequency vibroacoustic applications5. Given these underlying features, the rest of this work focuses on wave propagation aspects stemming from incorporating such gyric bodies in metamaterial lattices. h = h2 JxJy FIG. 2. Effective inertia Jx,eff of a gyric body for increasing values of ωh increasinghh= 0h4 III. WAVE PROPAGATION In this section, different types of gyric lattices are devised and studied in order to explore the full potential of GMMs in manipulating incident elastic waves. In its most general case, a GMM is formed by connecting multiple gyric unit cells with torsional springs, such as shown in Fig. 3a. In the provided schematic, every unit cell is connected to its adjacent ones through a universal joint allowing rotations in the two lateral directions x and y only and is supported with torsional springs in both directions. The entire lattice is fixed about the local z-axis. Throughout this analysis, the fundamental mechanical properties, i.e. stiffness and inertia, are kept completely unchanged in the different proposed GMM configurations, in addition to being invariant spatially and temporally. Instead, the angular momentum h is varied in different ways to onset band gaps, as well as break wave reciprocity and transmission symmetry. As a result, such GMMs provide the additional advantage of real- time tunability as well as a switchable platform between different functionalities, as will be detailed. Although asymmetric elastic and inertial properties about the x- and y-axes might provide an additional degree of design flexibility, it is rather challenging to control these properties in real-time. Therefore, we limit our analysis to kx = ky = k and Jx = Jy = J in order to isolate the angular momentum role in tuning the dispersion profile from other parameters. In the following subsections, we discuss the dispersion of elastic waves in GMMs with uniform, space-periodic, time-periodic and space-time-periodic angular momentum distributions, and follow that with a numerical time-transient verification of the fully simulated system in the last section. FIG. 3. (a) Schematic of a GMM lattice. 2-DOF universal joints and torsional springs across the x− and y− axes connect the adjacent unit cells. (b) A dynamically equivalent non-gyric lattice with a frequency-dependent effective stiffness A. Uniform Angular Momentum Distribution Consider an infinite lattice made of gyric unit cells, as shown in Fig. 3a, where the index n specifies the global position of each cell on the z-axis with respect to the origin. We start by setting an identical and time-invariant h for all cells. Consequently, the nth unit cell equations in the frequency domain can be written as −Jω2θ(n) −Jω2θ(n) x + iωhθ(n) y − iωhθ(n) y + k(cid:0)2θ(n) x + k(cid:0)2θ(n) x − θ(n−1) y − θ(n−1) x y − θ(n+1) − θ(n+1) x y (cid:1) = 0 (cid:1) = 0 √−1. Owing to the periodicity of lattice, a solution of the form θ(n) x,y = θx,ye−inκ where i = can be adopted, where κ is the non-dimensional wave number. Substituting the solution (4a) (4b) …………kxkykxkyuniversal joint(a)xyz…………ky,eff(b)kx,effky,effkx,effn(n−1)(n+1)5 (5) (6) (cid:21) (cid:20)θx θy (cid:20) 1±i (cid:21) = back in Eq. (4) gives cos κ = 1 − 1 2 (Ω2 ± ΩhΩ) with the corresponding Bloch modes (eigenvectors): tum of the spinning rotors, and ω0 =(cid:112)k/J is the natural frequency. The group velocity at is the dimensionless frequency, Ωh = ωh ω0 is the dimensionless angular momen- where Ω = ω ω0 the long wavelength limit (cg = dΩ/dκ) is zero, indicating that the introduction of angular momentum results in a vanishing speed of sound in GMMs with a uniform h distribution. We also note that the ± appears in both Eqs. (5) and (6) because two distinct wavenum- bers (κ+ and κ−) can coexist in each Irreducible Brillouin Zone (IBZ) for every incident frequency Ω. The Bloch modes in Eq. (6) reveal that transverse deflections about the x- 2 phase shift corresponding to each of the [1,−i] and [1, i] modes, re- and y- axes have a ± π spectively. In essence, for each mode, two coupled orthogonal transverse waves travel in the spanwise direction of a 1D medium with identical wavenumbers and magnitudes. Hence- forth, we refer to each of these coupled transverse wave pairs as a wave-mode. Therefore, the two wavenumbers κ+ and κ− correspond to two separate wave-modes propagating in the GMM, both expected to materialize in the lattice simultaneously for any given excitation. A frequency band gap forms in the range where κ is complex-valued. As such, given Eq. (5), it is evident that the [1, i] mode is attenuated in the ranges 0 < Ω < Ωh and h−Ωh). Ω > 1 Each of these cases corresponds to a partial band gap where one mode attenuates and the other freely propagates. A complete band gap, where both modes are suppressed, occurs in the shared frequency range 1 h + Ωh). These ranges as well as the corresponding attenuation degree, i.e. Im(κ), are graphically illustrated in Fig. 4 as a function of the spinning rotor angular momentum Ωh. As can be seen in Fig. 4c, the GMM maintains a bounded band gap only if Ωh < 2. Further, for the limiting case of Ωh = 0, the unbounded band gap (commonly referred to as a stop band 4) occurs at Ω > 2 which, as anticipated, matches the stop band of a discrete monatomic dispersive structure1. h − Ωh) < Ω < Ωh and Ω > 1 √ h +Ωh), while the [1,−i] mode is attenuated only if Ω > 1 2 ((cid:112)16 + Ω2 2 ((cid:112)16 + Ω2 2 ((cid:112)16 + Ω2 2 ((cid:112)16 + Ω2 FIG. 4. Attenuation degree Im(κ) as a function of Ωh corresponding to: (a) The [1, i] wave-mode, (b) the [1,−i] wave-mode, and (c) the combined effect showing complete band gaps. A bounded band gap exists only for Ωh > √ 2 The previous characteristics are also confirmed by the GMM's dispersion diagrams shown in Fig. 5a. The middle panel represents the unit cell's band structure given by the real com- ponent of the wavenumbers while the leftmost and rightmost panels show the corresponding attenuation degrees Im(κ+) and Im(κ−), respectively. The dashed lines correspond to zero Pass BandUnboundedBand GapBoundedBand Gap(a)(b)(c)6 FIG. 5. (a-c) Dispersion diagrams for a GMM with a uniform angular momentum in the frequency range 0 < Ωh < 5. Color levels indicate the value of Ωh. Dashed-line corresponds to a non-gyric monatomic lattice (h = 0). (d-f) Dispersion diagrams for Ωh = 50 showing the localized wave modes at Ω = 0 and Ω = Ωh for the case when Ωh (cid:29) 4 h − Ωh) ≈ 0 and 1 2 ((cid:112)16 + Ω2 2 ((cid:112)16 + Ω2 angular momentum (h = 0) or a non-gyric structure where Im(κ+) = Im(κ−). More im- portantly, we note that as Ωh increases, the bounded band gap forms in the leftmost panel which corresponds to the [1, i] mode. This signals that the two degenerate modes of the system start diverging until they eventually become flat single frequency bands at Ω = 0 and Ω = Ωh, as depicted in Fig. 5e. In which case, the bounded and unbounded band gaps coalesce to span the entire dispersion spectrum with the exception of these two frequencies. Upon examining the band gap ranges discussed earlier, we expect this limiting behavior to happen for Ωh (cid:29) 4 where 1 h + Ωh) ≈ Ωh. In practice, such localized flat bands, with near-zero group velocities, behave similar to stand- ing waves associated with natural frequencies of a finite structure. As a result, the GMM manages to selectively admit both frequencies from a wide-band excitation. Such behavior has been also recently reported in time-periodic systems where wavenumber as opposed to frequency band gaps emerge as a result of an external temporal modulation34. It is also noteworthy that, given the dependence on h in GMMs, this apparent band separation can be tuned in real-time by solely varying the rotational speed of the rotors. Finally, by re- versing the rotation direction of R, the dispersion profile switches between the wave-modes. Substituting Ωh with −Ωh does not influence the band structure shown in Fig. 5 (with the exception of swapping the subplots of Im(κ+) and Im(κ−)) while perfectly exchanging the two eigenvectors. The latter being an added feature of gyroscopic systems, which has been recently utilized to induce internal cell asymmetry between neighboring unit cells thus creating topologically protected interface modes based on the quantum valley hall effect (QVHE)40. To shed light on the band gap generation mechanism in GMMs, consider a lattice made of the equivalent non-gyric unit cells as shown in Fig. 3b. To establish dynamic equivalence, the non-gyric lattice exhibits a frequency-dependent effective stiffness keff to compensate for the embedded spinning rotor in its gyric counterpart. The dispersion relation for such h=0h=0(a)(c)(b)1, i[ ]wave-mode1, i[ ]wave-modeband structure5cg=0unboundedgapboundedgap(d)(f)(e)lattice becomes and by comparing Eqs. (5) and (7) and solving for keff, we get cos κ = 1 − Jω2 2keff keff = k 1 ± Ωh Ω 7 (7) (8) FIG. 6. Effective stiffness properties of a dynamically equivalent non-gyric lattice for increasing values of Ωh Fig. 6 captures the relationship between Ω and keff for different values of Ωh. It confirms that the stiffness of the equivalent non-gyric lattice exhibits negative effective values within the range 0 < Ω < Ωh for the [1, i] mode only. Further, the negativity switches between the the two wave-modes with an angular momentum sign change, which correlates well with the predicted band gaps. B. Space-Periodic Angular Momentum Variation Next, we examine a phononic GMM with a spatially periodic pattern by setting the rotor speed of every other cell equal to zero. In which case, we are able to model the steady-state dynamics of the nth and (n − 1)th successive cells via the system of equations given by Kd Θ − kΘr = 0 0 2k − ω2J 0 −k −k 0 2k − ω2J −iωh Kd = where and 2k − ω2J (cid:104) 0 −k 0 Θ = 0 −k iωh 2k − ω2J (cid:105)T  (9) (10) (11) (12) θ(n−1) x θ(n−1) y θ(n) x θ(n) y (cid:104) Θr = θ(n−2) x θ(n−2) y θ(n+1) x θ(n+1) y (cid:105)T 51, i[ ]wave-modes1, i[ ]increasing1, i[ ]8 Eqs. (9) through (12) constitute four coupled mechanical resonators. As a result, we expect at most four different modes to appear in the system's response. Similar to the previous section, the number of modes is reduced by half for a vanishing angular momentum (i.e. h = 0). Given the GMM's periodicity, a Bloch-wave solution of the form θ(n−2) x,y = e−iκ is employed. Upon substituting in Eq. (9), we obtain the θ(n) x,y eiκ and θ(n+1) dispersion relation x,y = θ(n−1) x,y cos κ = 1 2 (Ω2 − 2)(Ω2 ± ΩhΩ − 2) − 1 (13) FIG. 7. (a-b) Attenuation degree Im(κ) as a function of Ωh for partial band gaps. (c) Shared attenuation regions showing three complete band gaps (two of which are bounded) for a space- periodic GMM FIG. 8. (a-c) Dispersion diagrams for a GMM with a space-periodic angular momentum distribution within the frequency range 0 < Ωh < 2.5. Color levels indicate the value of Ωh. Dashed-line corresponds to a non-gyric lattice (h = 0). (d-f) Dispersion diagrams for Ωh = 25 showing the localized wave modes at Ω = 0, Ω = √ 2, and Ω = Ωh for the case when Ωh (cid:29) 4 It is evident that the ±ΩhΩ term in Eq. (13) vanishes as Ωh approaches zero, causing a reduction in distinct wave modes. Band gap regions and attenuation degrees for different (b)(c)(a)PassBandsBoundedBand Gapunboundedgapband structureh=0h=0(a)(c)(b)(d)(f)(e)=cg0boundedgaps===9 values of Ωh are shown in Fig. 7, with the corresponding dispersion diagrams in Fig. 8. As predicted, the space-periodic GMM exhibits four distinct wave-modes resulting from the split of each of the acoustic and optic modes. The dashed lines correspond to a zero angular momentum value associated with a conventional non-gyric lattice with acoustic and optic modes only and no band gaps. The bottom panel of Fig. 8 depicts the large angular momentum scenario where the GMM effectively acts as a wide-band filter for all frequencies with the exception of Ω = 0, Ω = 2 and Ω = Ωh (represented by the flat branches in the figure). In other words, the lower band approaches zero, the upper one tends to Ω = Ωh, and the two inner modes converge to a single band at Ω = √ √ 2. C. Time-Periodic Angular Momentum Variation Inspired by efforts in utilizing time modulated material properties for frequency conver- sion and amplification9,34,41, we next examine a time-harmonic GMM obtained by imposing an oscillating h with a pumping frequency ωp across the entire gyric lattice. The time varia- tion is given by h(t) = h0 +h1 cos ωpt, where h0 and h1 are the average (bias) and oscillation amplitudes, respectively. Consider a time-periodic GMM realized by means of a mechanical torque imparted to each spinning rotor and given by T (t) = dh/dt = −h1ωp sin ωpt. The torque can, for example, be supplied by an internally embedded motor with an instanta- neous power Pe(t) = hT Jr = −ωph1 2Jr (2h0 sin ωpt + h1 sin 2ωpt) (14) where Jr is the rotational inertia of the rotor about its principle z-axis. As such, the required power to overcome the rotational inertia of the rotor is approximately upper-bounded by ∼= ωph1(2h0 + h1)/2Jr (an exact upper bound exists, but is omitted here for brevity). Pe For a frictionless setup, the net amount of energy injected into the system per one temporal cycle is Pe/Cycle = Pedt = 0 (15) independent of ωp, which can be attributed to the π 2 phase shift between the angular momentum and the electromotor torque. This behavior is reminiscent of forced harmonic oscillations of an undamped spring-mass system, where zero energy input is required to maintain oscillations at a desired frequency. It should be noted, however, that in the presence of friction or damping, a net positive amount of energy is needed to sustain the time-periodic angular momentum profile. (cid:90) 2π ωp 0 The time-periodic GMM's motion equations are given by x − θ(n−1) y − θ(n−1) x = −h(t) θ(n) y = h(t) θ(n) J θ(n) y − k(cid:2)2θ(n) x − k(cid:2)2θ(n) J θ(n) x y − θ(n+1) − θ(n+1) x y (cid:3) (cid:3) (16a) (16b) In deriving Eq. (16), we still maintain uniform stiffness and inertia along both transverse axes and the spanwise direction of the GMM lattice. We also assume that the inertia of the unit cells are not significantly affected by the temporal variation of the rotor's angular momentum. In practice, this is achieved by tuning the angular momentum via controlling the speed, rather than inertia, of the spinning rotors. By taking spatial periodicity into account, i.e. θ(n+1) x,y (t), and dropping n, Eq. (16) simplifies to x,y (t), and θ(n−1) (t) = e−iκθ(n) (t) = eiκθ(n) x,y x,y J θx + h(t) θy + 2k(1 − cos κ)θx = 0 J θy − h(t) θx + 2k(1 − cos κ)θx = 0 (17a) (17b) 10 FIG. 9. Dispersion diagrams of a time-periodic GMM with a zero modulation bias (h0 = 0): (a) Ωp = 0.5, (b) Ωp = 1.0, and (c) Ωp = 1.5. First column: h1 = 0.4Jω0, second column: h1 = Jω0, and third column: h1 = 2.5Jω0. Shaded regions indicate band gap emergence at increased pumping frequencies Eq. (17) constitutes a homogeneous linear time-periodic (LTP) system with a period τp = 2π ωp . For any given κ, a proper ansatz given by the Floquet theory is42 θx,y(t; κ) = eλt φ(l) x,yeilωpt (18) ∞(cid:88) l=−∞ x and φ(l) where φ(l) y are the Fourier coefficients and λ is a generally complex constant known as the characteristic exponent43. Substituting Eq. (18) in (17) and exploiting harmonic balance results into a second order eigenvalue problem in terms of λ and κ. According to the Lyapanov-Floquet theorem44, the temporal nature -- and thus stability45 -- of the wave propagation is dictated by the characteristic exponents. As a result, it is beneficial to consider the free-wave case by casting the dispersion relation as a quadratic eigenvalue problem in terms of Λ = λ/ω0 (cid:26)(cid:20) I O (cid:21) O I (cid:20) Ψ(1) Ψ(2) −Ψ(2) Ψ(1) (cid:21) (cid:20) Ψ(3) Ψ(4) −Ψ(4) Ψ(3) (cid:21)(cid:27)(cid:26)Φx (cid:27) Φy Λ2 + Λ + = 0 (19) where I and O are identity and null matrices of a proper size. The entries of the dimen- sionless matrices Ψ(j) for j = 1, . . . , 4 are given by Ψ(1) l,q = 2iqΩpδl,q Ψ(2) h1 δl,q + l,q = h0 Jω0 l,q =(cid:2)2(1 − cos κ) − q2Ω2 2Jω0 p δl,q±1 (cid:3)δl,q Ψ(3) Ψ(4) (20a) (20b) (20c) l,q = iqΩpΨ(2) l,q (20d) where Ωp = ωp/ω0 is the dimensionless temporal modulation frequency, l and q ∈ Z, and δl,q is the Kronecker delta which is equal to one for l = q and zero otherwise. In addition, the rotation vectors (Φx and Φy) are φ(−∞) , . . . , φ(−1) (cid:105)T x,y, . . . , φ(∞) x,y , φ(0) x,y, φ(1) Φx,y = (21) (cid:104) x,y x,y (a)(b)(c)11 The vector in Eq. (21) can be truncated as long as the summation in Eq. (18) remains bounded; a guaranteed property of a Fourier series provided a sonic blow-up does not happen46. In other words, approximate dispersion diagrams can be generated by seeking non-trivial solutions of the truncated version of Eq. (19). Note that the appearance of Ψ(1) and Ψ(4) is solely attributed to the time-periodic angular momentum variation. The stiffness matrix of the system is also influenced since a centripetal term is subtracted from the diagonal terms in the third matrix, as evident by Eq. (20c). By setting Ωp = 0, we should reconstruct the problem outlined in section III A. As previously stated, unlike stiffness and density in non-gyric metastructures, the angular momentum h is not limited to positive values. Consequently, time-periodic GMMs can be generally classified into three groups based on the modulation amplitudes h0 and h1. The first group comprises GMMs with zero bias angular momentum (h0 = 0) and h1 > 0. The second group is systems with a nonzero bias (h0 > 0) and h1 < h0. The third group has h0 > 0 and h1 > h0. The previous classification is in addition to the traditional division of time-periodic systems into slow (Ωp < 1) and fast (Ωp ≥ 1) modulation regimes28. Fig. 9 shows the dispersion patterns corresponding to the first category for various modulation regimes. Note that the effect of h0 (cid:54)= 0 is a separation of the otherwise identical dynamic modes of the system as well as an up-shift of one of them by an amount h0/Jω0 in the Ω − κ plot. Consequently, we only consider the zero-bias modulation amplitude category since the behavior of the remaining cases can then be readily extracted. FIG. 10. (a) Band gap width ∆Ω and (b) band gap frequency range in time-periodic GMMs with h0 = 0 Note that the time-periodicity augments the dispersion profile with a periodicity along the frequency axis. That is, band diagrams at higher frequencies are repetitions of the frequency range [0, Ωp]. As a result, dispersion curves plotted within the ranges of κ ∈ [−π, π] and Ω ∈ [0, Ωp] provide sufficient information about the propagation metrics of the time-periodic GMM. Furthermore, there does not exist any wavenumber band gaps (or complex λ value with temporal instability47) regardless of the speed or amplitude of the modulation. The latter indicates that a time-periodic GMM does not support unstable solutions with exponentially growing components as a result of complex eigenvalues. An observation which is in compliance with Eq. (15) stating that the energy content of a time- periodic GMM is conserved over one modulation cycle. A detailed discussion of the stability of this class of GMMs will be separately addressed in a future effort. A parametric study on the effect of dimensionless pumping frequency Ωp on the band gap width and range is depicted in Fig. 10 for the three cases shown in Fig. 9. It is noted that the band gap width generally increases with the pumping frequency in every case, albeit not monotonically. 0.750.25(a)(b)12 D. Space-time-periodic Angular Momentum Variation: Breakage of Reciprocal Symmetry As outlined so far, a distinct advantage of GMM systems is the ability of an angular momentum (induced via a set of embedded spinning rotors) to influence elastic waves, onset band gaps and, more importantly, tune them in real-time by solely adjusting the rotation speed. In this section, we exploit such features to introduce a traveling-wave-like modulation of angular momentum with the intention of breaking the reciprocal symmetry of elastic waves in the GMM lattice. Consider the following space-time-periodic waveform for h: h(n)(t) = h0 + h1 cos(ωpt − κpn) (22) for a GMM consisting of an infinite number of "super cells", each of which containing N unit cells with the prescribed time-varying angular momentum. Also, n is the unit cell index and κp = 2π/N is the spatial modulation frequency. It is worth noting that, unlike the time-periodic lattice, the angular momentum of adjacent unit cells here carries a 2π/N phase difference. For convenience, we rewrite the governing motion equations for every N gyric unit cells in the following LTP form J Θ + G(t) Θ + K(κ)Θ = 0 (23) where Θ2N×1 = [θ(1) ]T is the rotation vector, J = JI is the inertia ma- trix, and I is an identity matrix of size 2N . In the absence of non-conservative constituents, G(t) takes the form x , ..., θ(N ) x θ(1) y , ..., θ(N ) y 2N×2N (24) from which it can be inferred that G(t) is a time-periodic skew-symmetric block matrix, i.e. G(t + τp) = G(t) = −G†(t) with (•)† denoting the complex-conjugate-transpose operation. In Eq. (24), H(t) = diag{h(1)(t), h(2)(t), ..., h(N )(t)}. Furthermore, the stiffness matrix K(κ) is explicitly where the Hermitian matrix Γ(κ) is found to have the following general form for N ≥ 3: In order to study propagation of linear elastic waves in the space-time-periodic GMM described by Eq. (23), we build on the plane wave expansion method established for space- time modulated systems with N (cid:54)= ∞29,48. Correspondingly, we utilize a Bloch solution of the form Θ(t) = Φ(t)eλt in which the amplitude vector Φ(t) is τp-periodic. As such, Φ(t) and G(t) can be replaced with their Fourier series representations Φ(t) = Φqeiqωpt , G(t) = Gleilωpt (27) ∞(cid:88) q=−∞ where Φl and Gq are the associated Fourier vector and matrix coefficients. By substituting in Eq. (23) and exploiting the harmonic balance, we get (AΛ2 + BΛ + C)V = 0 (28) (cid:20) O H(t) −H(t) O (cid:21) G(t) = (cid:20) Γ(κ) O (cid:21) O Γ(κ) K(κ) =  2 −1 −e−iκ −1 . . . . . . −eiκ . . . . . . −1 −1 2 Γ(κ) = k  ∞(cid:88) l=−∞ (25) (26) 13 FIG. 11. Dispersion diagrams for uniform, space-periodic, and space-time-periodic GMMs with 3 unit cells per super cell (i.e. N = 3) and a zero bias angular momentum value (i.e. h0 = 0). Dashed lines denote all possible solutions of the theoretical dispersion analysis. Solid lines correspond to the fundamental modes obtained via an eigenvector weighting method where A, B and C are block matrices with entries that are explicit functions of N , Ωp, h0/Jω0 and h1/Jω0 (see supplementary material for details49). In addition, the vector V is obtained by stacking all the Φq from q = −∞ to ∞ as follows VT = [ ΦT−∞, ..., ΦT−1, ΦT 0 , ΦT 1 , ..., ΦT∞] (29) Solving the eigenvalue problem in Eq. (28) yields dispersion diagrams of the space-time- periodic GMM. For any given κ, only 2(2N ) solutions are fundamental modes that can be identified using the magnitude of the associated eigenvector components48 (see supplemen- tary material for more on dispersion band reduction49). Fig. 11 shows the dispersion plots for a uniform, space-periodic, and space-time-periodic (with two different modulation speeds) for a GMM with 3 gyric unit cells per super cell (i.e. N = 3) and h0 = 0. The dispersion behavior in the first two cases is fully reciprocal, with band gaps emerging in Fig. 11b due to the spatial periodicity. By imposing a temporal modulation with a frequency Ωp = 0.2, the GMM's transmission symmetry is broken and the dispersion becomes non-reciprocal (Fig. 11c). Note that band gaps corresponding to left-propagating waves up-shift a/o down-shift by an amount Ωp; a behavior which is in agreement with the observed robustness33 and quantization50 of non-reciprocal band gaps in space-time modulated systems. As indicated earlier, the GMM uniquely maintains stability at fast modulation speeds, in contrast to stiffness modulation in elastic metamaterials29. An example of a fast modulation is Ωp = 1.2 in Fig. 11d. The result is a much greater degree of distortion and asymmetry in the dispersion curves, albeit without wavenumber band gaps or complex frequencies. Finally, it is noted that 10 harmonic modes are included in Fig. 11d, whereas one mode sufficiently predicts the dispersion behavior with a reasonable accuracy in the low-speed modulation case. IV. NUMERICAL VERIFICATION To verify the dispersion predictions detailed in Section III, the transient response of a finite 1D GMM lattice with 300 cells is numerically computed via full wave simulation and is used to evaluate the GMM's actual dispersion contours. The GMMs in the simulations are subjected to a broadband force imparted at the center of the lattice such that both forward and backward traveling waves can be captured simultaneously. The excitation is a (SlowModulation)(FastModulation)(a)(b)(c)(d)14 FIG. 12. Numerically reconstructed dispersion contours for a GMM with (a-c) uniform and (d-f) space-periodic angular momentum distribution. Dotted lines denote the theoretically predicted dispersion bands for comparison Gaussian wave packet whose frequency content spans the frequency range of interest. We solve for the time-dependent displacement field of the lattice using MATLAB's "ode45" function with 1e-6 relative tolerance and an output time step of 5 ms. A Hamming window is also incorporated to compensate for the truncation of signal to the first 15 seconds -- before it reaches spatial boundaries. A two-dimensional Fourier transformation (2DFT) is then applied in both space and time to numerically reconstruct the dispersion diagrams of each individual case. The 2DFT is given by 300(cid:88) n=1 (cid:90) tend 0 θ(κ, Ω) = eiκ n N θ(n)(t) e−iω0Ωt dt (30) where tend is the total simulation time. In here, a grid of 200 × 200 is used to represent the transformed displacement amplitude in the κ− Ω space. Figs. 12a-c graphically display the recovered dispersion behavior corresponding to a finite GMM lattice with time-invariant uniform distribution of angular momentum. The numerical contours are in strong agreement with the theoretical dispersion predictions, plotted via dashed lines for comparison. Note that as Ωh increases from zero, each dynamic modes of the system splits into two separate wave-modes, as shown in Fig. 12b for Ωh = 1. By further increasing the speed of the spinning rotors, they converge to two flat modes at Ω = 0 and Ω = Ωh effectively suppressing almost the entire frequency spectrum. The second set of simulations, shown in Figs. 12d-f, correspond to the spatially periodic gyric lattice which comprises a zero angular momentum for every other unit cell. In this case, the two initially repeated wave-modes of the non-gyric lattice (Fig. 12d) turn into four √ distinct wave-modes (Fig. 12e) as Ωh grows from zero to one. By further increasing Ωh, the two middle wave-modes coincide to form a single standing-wave (flat) mode at Ωh = 2 while the two other bands become nearly flat at Ω = 0 and Ωh, with the number of unique dynamic modes reduced to three (Fig. 12f). Once again, the GMM becomes an almost all- attenuation structure suppressing the bulk of the frequency spectrum. Interestingly in this case, the wave mode emerging at Ω = 2 appears to be dominant (since it's a coalescence of two modes as evident by the contour amplitudes). √ 15 FIG. 13. Numerically reconstructed dispersion contours for a time-periodic GMM (a-c) and a space-time-periodic GMM (d-f) with the following angular momentum modulations: (a) h0 = 0, h1 = Jω0, and Ωp = 1.5, (b) h0 = 0.4Jω0, h1 = 0.5h0, , and Ωp = 1.5, (c) h0 = 2.5Jω0, h1 = 1.5h0, and Ωp = 1.5, (d) h0 = 0, h1 = 0.4Jω0, and Ωp = 0.2, (e) h0 = 0, h1 = 0.4Jω0, and Ωp = 1.2, and (f) h0 = 0.8Jω0, h1 = 3 8 h0, and Ωp = 0.5. Theoretical dispersion curves are shown for comparison In keeping with the previous cases, numerical simulations are used to validate the disper- sion predictions for the time-periodic GMM. Fig. 13 shows the behavior of the time-periodic GMM for select modulation cases at Ωp = 1.5. Fig. 13a corresponds to h0 = 0 and h1 = Jω0. The repetition of the first extended Brillouin zone is clearly noticeable. As the theory sug- gests, one band gap appears in each extended Brillouin zone that is repeated periodically at higher frequencies. Figs. 13b and c are obtained for h0 = 0.4Jω0, h1 = 0.5h0, and h0 = 1.5Jω0, h1 = 2.5h0, respectively. A near perfect agreement between the theoretical predictions and the simulations can also be observed throughout. Finally, Fig. 13d-f shows the simulation results for the space-time-periodic GMMs. Dashed lines correspond to all the dispersion bands obtained from the theoretical analysis, while the solid black lines represent the modes with greatest weights. In all the three modulation cases shown, reciprocity is broken and waves behave differently on both sides of κ = 0. V. CONCLUSIONS This work has showcased gyric metamaterials (GMMs) as an alternative class of meta- materials which can effectively exhibit wave manipulation capabilities culminating in band gaps as well as non-reciprocal transmission without necessarily requiring a mass or stiffness modulation. By tuning the rotational speed or direction of a set of embedded spinning ro- tors, a wide range of spatiotemporal distributions of the angular momentum of the lattice's unit cells can be achieved. Such variations have been shown to onset band gaps in the frequency spectrum and/or induce one-way directivity as needed. A combination of theo- retical dispersion analyses as well as numerical time-domain simulations have been utilized to show the ability of the angular momentum modulation to achieve some of the funda- mental and most recent developments in non-reciprocal systems and metamaterial-based wave guides. Unlike inertia and stiffness, angular momentum is not an inherent property of 16 the material and can, therefore, be handily tuned by solely controlling the speed of rotors. As such, a gyric lattice can near-instantaneously switch between a set of different desirable functionalities by design. Such tunability becomes even more significant in the context of non-reciprocal wave phenomena and space-time-periodic systems, where a certain degree of online control over the system's mechanical properties is critical. Further, given their dependence on gyroscopic effects, in a frictionless setup, GMMs can be classified as con- servative systems which do not dissipate mechanical energy into heat or other disorganized energy forms. Finally, the analysis undertaken here illustrated that non-reciprocal GMMs remain reliably stable at high pumping frequencies. Future directions of this work include an experimental realization of this class of time-periodic gyric lattices as well as a rigorous Lyapunov-based investigation of the different stability metrics of such GMMs. VI. ACKNOWLEDGEMENT M. A. and M. N. acknowledge the support of this work from the Vibration Institute (VI) through the 2018 VI Academic Grant Program. 1L. Brillouin, Wave propagation in periodic structures: electric filters and crystal lattices (Courier Corpo- ration, 2003). 2D. Mead, "Wave propagation in continuous periodic structures: research contributions from southampton, 1964 -- 1995," Journal of sound and vibration 190, 495 -- 524 (1996). 3P. A. Deymier, Acoustic metamaterials and phononic crystals, Vol. 173 (Springer Science, 2013). 4M. I. Hussein, M. J. Leamy, and M. Ruzzene, "Dynamics of Phononic Materials and Structures: Historical Origins, Recent Progress, and Future Outlook," Applied Mechanics Reviews 66, 040802 (2014). 5H. H. Huang, C. T. Sun, and G. L. Huang, "On the negative effective mass density in acoustic metama- terials," International Journal of Engineering Science 47, 610 -- 617 (2009). 6M. Nouh, O. Aldraihem, and A. Baz, "Wave propagation in metamaterial plates with periodic local resonances," Journal of Sound and Vibration 341, 53 -- 73 (2015). 7H. Al Ba'ba'a and M. Nouh, "An investigation of vibrational power flow in one-dimensional dissipative phononic structures," Journal of Vibration and Acoustics 139, 021003 -- 10 (2017). 8P. F. Pai, "Metamaterial-based Broadband Elastic Wave Absorber," Journal of Intelligent Material Sys- tems and Structures 21, 517 -- 528 (2010). 9A. Cullen, "A travelling-wave parametric amplifier," Nature 181, 332 (1958). 10G. Trainiti and M. Ruzzene, "Non-reciprocal elastic wave propagation in spatiotemporal periodic struc- tures," New Journal of Physics 18, 083047 (2016). 11A. Seyranian, J. Stoustrup, and W. Kliem, "On gyroscopic stabilization," Zeitschrift fur angewandte Mathematik und Physik ZAMP 46, 255 -- 267 (1995). 12G. C. Gaudin, "Gyro-controlled pitch stabilizing system," (1981), uS Patent 4,261,278. 13L. Barkwell, P. Lancaster, and A. Markus, "Gyroscopically stabilized systems: A class of quadratic eigenvalue problems with real spectrum," Canad. J. Math 44, 42 -- 53 (1992). 14M. Brocato and G. Capriz, "Control of beams and chains through distributed gyroscopes," AIAA journal 47, 294 -- 302 (2009). 15Q. Hu, Y. Jia, and S. Xu, "Recursive dynamics algorithm for multibody systems with variable-speed control moment gyroscopes," Journal of guidance, control, and dynamics 36, 1388 -- 1398 (2013). 16H. Yoon and P. Tsiotras, "Spacecraft adaptive attitude and power tracking with variable speed control moment gyroscopes," Journal of Guidance, Control, and Dynamics 25, 1081 -- 1090 (2002). 17M. Brun, I. S. Jones, and A. B. Movchan, "Vortex-type elastic structured media and dynamic shielding," Proc. R. Soc. A 468, rspa20120165 (2012). 18L. M. Nash, D. Kleckner, A. Read, V. Vitelli, A. M. Turner, and W. T. Irvine, "Topological mechanics of gyroscopic metamaterials," Proceedings of the National Academy of Sciences 112, 14495 -- 14500 (2015). 19P. Wang, L. Lu, and K. Bertoldi, "Topological phononic crystals with one-way elastic edge waves," Physical review letters 115, 104302 (2015). 20G. DEleuterio and P. Hughes, "Dynamics of gyroelastic continua," Journal of applied mechanics 51, 415 -- 422 (1984). 21Y. Alsaffar, S. Sassi, and A. Baz, "Band gap characteristics of periodic gyroscopic systems," Journal of Sound and Vibration 435, 301 -- 322 (2018). 22J. Achenbach, Wave propagation in elastic solids, Vol. 16 (Elsevier, 2012). 23M. B. Zanjani, A. R. Davoyan, A. M. Mahmoud, N. Engheta, and J. R. Lukes, "One-way phonon isolation in acoustic waveguides," Applied Physics Letters 104, 081905 (2014). 24N. Swinteck, S. Matsuo, K. Runge, J. Vasseur, P. Lucas, and P. Deymier, "Bulk elastic waves with unidirectional backscattering-immune topological states in a time-dependent superlattice," Journal of Applied Physics 118, 063103 (2015). 25R. Fleury, D. L. Sounas, C. F. Sieck, M. R. Haberman, and A. Al`u, "Sound isolation and giant linear nonreciprocity in a compact acoustic circulator," Science 343, 516 -- 519 (2014). 17 26M. Attarzadeh and M. Nouh, "Elastic wave propagation in moving phononic crystals and correlations with stationary spatiotemporally modulated systems," AIP Advances 8, 105302 (2018). 27D. Beli, P. B. Silva, and J. R. de Fran¸ca Arruda, "Mechanical circulator for elastic waves by using the nonreciprocity of flexible rotating rings," Mechanical Systems and Signal Processing 98, 1077 -- 1096 (2018). 28H. Nassar, X. Xu, A. Norris, and G. Huang, "Modulated phononic crystals: Non-reciprocal wave propa- gation and willis materials," Journal of the Mechanics and Physics of Solids 101, 10 -- 29 (2017). 29M. Attarzadeh, H. Al Babaa, and M. Nouh, "On the wave dispersion and non-reciprocal power flow in space-time traveling acoustic metamaterials," Applied Acoustics 133, 210 -- 214 (2018). 30M. Attarzadeh and M. Nouh, "Non-reciprocal elastic wave propagation in 2d phononic membranes with spatiotemporally varying material properties," Journal of Sound and Vibration 422, 264 -- 277 (2018). 31C. Croenne, J. Vasseur, O. Bou Matar, M.-F. Ponge, P. Deymier, A.-C. Hladky-Hennion, and B. Dubus, "Brillouin scattering-like effect and non-reciprocal propagation of elastic waves due to spatio-temporal modulation of electrical boundary conditions in piezoelectric media," Applied Physics Letters 110, 061901 (2017). 32M. Ansari, M. Attarzadeh, M. Nouh, and M. A. Karami, "Application of magnetoelastic materials in spatiotemporally modulated phononic crystals for nonreciprocal wave propagation," Smart Materials and Structures 27, 015030 (2017). 33R. Chaunsali, F. Li, and J. Yang, "Stress wave isolation by purely mechanical topological phononic crystals," Scientific reports 6, 30662 (2016). 34G. Trainiti, Y. Xia, J. Marconi, A. Erturk, and M. Ruzzene, "Time-periodic stiffness modulation in elastic metamaterials for selective wave filtering: theory and experimental investigations," arXiv preprint arXiv:1804.09209 (2018). 35E. Cassedy, "Dispersion relations in time-space periodic media: Part ii, unstable interactions," Proceed- ings of the IEEE 55, 1154 -- 1168 (1967). 36G. W. Milton and J. R. Willis, "On modifications of newton's second law and linear continuum elas- todynamics," in Proceedings of the royal society of london A: Mathematical, Physical and Engineering Sciences, Vol. 463 (The Royal Society, 2007) pp. 855 -- 880. 37G. L. Huang and C. T. Sun, "Band Gaps in a Multiresonator Acoustic Metamaterial," Journal of Vibration and Acoustics 132, 031003 (2010). 38H. Al Ba'ba'a, M. Nouh, and T. Singh, "Formation of local resonance band gaps in finite acoustic metamaterials: A closed-form transfer function model," Sound and Vibration Journal 410, 429 -- 446 (2017). 39H. Crabtree, An elementary treatment of the theory of spinning tops and gyroscopic motion (Longmans, Green, 1914). 40M. Garau, G. Carta, M. Nieves, I. Jones, N. Movchan, and A. Movchan, "Interfacial waveforms in chiral lattices with gyroscopic spinners," Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 474, rspa20180132 (2018). 41P. Tien, "Parametric amplification and frequency mixing in propagating circuits," Journal of Applied Physics 29, 1347 -- 1357 (1958). 42N. M. Wereley, Analysis and control of linear periodically time varying systems, Ph.D. thesis, Mas- sachusetts Institute of Technology (1990). 43J. A. Richards, Analysis of periodically time-varying systems (Springer Science & Business Media, 2012). 44R. W. Brockett, Finite dimensional linear systems, Vol. 74 (SIAM, 2015). 45W.-C. Xie, Dynamic stability of structures (Cambridge University Press, 2006). 46E. Cassedy and A. Oliner, "Dispersion relations in time-space periodic media: Part i, stable interactions," Proceedings of the IEEE 51, 1342 -- 1359 (1963). 47E. Cassedy, "Temporal instabilities in traveling-wave parametric amplifiers (correspondence)," IRE Trans- actions on Microwave Theory and Techniques 10, 86 -- 87 (1962). 48J. Vila, R. K. Pal, M. Ruzzene, and G. Trainiti, "A bloch-based procedure for dispersion analysis of lattices with periodic time-varying properties," Journal of Sound and Vibration 406, 363 -- 377 (2017). 49"See supplementary material at [url will be inserted by aip] for dispersion band reduction method and relations for space-time-periodic gmm," . 50H. Nassar, H. Chen, A. Norris, and G. Huang, "Quantization of band tilting in modulated phononic crystals," Physical Review B 97, 014305 (2018).
1711.06925	1	1711	2017-11-18T20:44:59	Laser patterned polymer/nanotube composite electrodes for nanowire transistors on flexible substrates	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	Fabrication techniques such as laser patterning offer excellent potential for low cost and large area device fabrication. Conductive polymers can be used to replace expensive metallic inks such as silver and gold nanoparticles for printing technology. Electrical conductivity of the polymers can be improved by blending with carbon nanotubes. In this work, formulations of acid functionalised multiwall carbon nanotubes (f-MWCNT) and poly (ethylenedioxythiophene) [PEDOT]: polystyrene sulphonate [PSS] were processed, and thin films were prepared on plastic substrates. Conductivity of PEDOT: PSS increased almost four orders of magnitude after adding f-MWCNT. Work function of PEDOT:PSS/f-MWCNT films was ~ 0.5eV higher as compared to the work function of pure PEDOT:PSS films, determined by Kelvin probe method. Field-effect transistors source-drain electrodes were prepared on PET plastic substrates where PEDOT:PSS/f-MWCNT were patterned using laser ablation at 44mJ/pulse energy to define 36 micron electrode separation. Silicon nanowires were deposited using dielectrophoresis alignment technique to bridge the PEDOT:PSS/f-MWCNT laser patterned electrodes. Finally, top-gated nanowire field effect transistors were completed by depositing parylene C as polymer gate dielectric and gold as the top-gate electrode. Transistor characteristics showed p-type conduction with excellent gate electrode coupling, with an ON/OFF ratio of ~ 200. Thereby, we demonstrate the feasibility of using high workfunction, printable PEDOT:PSS/MWCNT composite inks for patterning source/drain electrodes for nanowire transistors on flexible substrates.	physics.app-ph	physics	Laser patterned polymer/nanotube composite electrodes for nanowire transistors on flexible substrates Kiron Prabha Rajeev; Michael Beliatis; Stamatis Georgakopoulos; Vlad Stolojan; John Underwood and Maxim Shkunov and carbon nanotubes (f-MWCNT) functionalised multiwall Advanced Technology Institute, Electrical and Electronic Engineering, University of Surrey, Guildford GU2 7XH, United Kingdom Abstract Fabrication techniques such as laser patterning offer excellent potential for low cost and large area device fabrication. Conductive polymers can be used to replace expensive metallic inks such as silver and gold nanoparticles for printing technology. Electrical conductivity of the polymers can be improved by blending with carbon nanotubes. In this work, formulations of acid poly (ethylenedioxythiophene) [PEDOT]: polystyrene sulphonate [PSS] were processed, and thin films were prepared on plastic substrates. Conductivity of PEDOT: PSS increased almost four orders of magnitude after adding f-MWCNT. Work function of PEDOT: PSS/f-MWCNT films was ~ 0.5eV higher as compared to the work function of pure PEDOT: PSS films, determined by Kelvin probe method. Field-effect transistors source-drain electrodes were prepared on PET plastic substrates where PEDOT: PSS/f-MWCNT were patterned using laser ablation at 44mJ/pulse energy to define 36µm electrode separation. Silicon nanowires were deposited using dielectrophoresis alignment technique to bridge the PEDOT: PSS/f-MWCNT laser patterned electrodes. Finally, top-gated nanowire field effect transistors (FET) were completed by depositing parylene C as polymer gate dielectric and gold as the top-gate electrode. Transistor characteristics showed p-type conduction with excellent gate electrode coupling, with an ON/OFF ratio of ~ 200. Thereby, we demonstrate the feasibility of using high workfunction, printable PEDOT: PSS/MWCNT composite inks for patterning source/drain electrodes for nanowire transistors on flexible substrates. Key Words Nanowires, conducting polymers, carbon nanotubes, polymer-nanocomposite printable inks, laser ablation, field effect transistors, flexible substrates Introduction Solution processed nanoelectronic materials in the recent years have opened up possibilities for printed electronics applications such as sensors1, flexible displays2, energy harvesting piezoelectric devices3. High performance transistors are required for displays and sensors, whereas dense nanomaterials coverage is request for energy harvesting, and, in all cases, low cost deposition and structuring techniques are essential for the devices fabrication. Conventional photo-lithographic fabrication provides high resolution electrode deposition, but it does not fit well with high throughput processes4. Low cost and efficient technologies should be favoured in the patterning of electrodes such as ink-jet printing5, laser patterning6, soft lithography7. In particular, laser-based techniques to pattern the electrodes can be cost effective and efficient. The laser ablation technique is a one step process which can be applied to metals as well as polymers, without the necessity for any treatments such as etching and striping8. Efficient material removal for electrode patterning can be achieved by high intensity pulsed lasers. The induced laser pulse breaks the chemical bonds and removes the unwanted material from the substrate, for pulse with delivered energy above the binding energy of the molecules within the films on the substrate. The material can be removed with one or multiple laser pulses depending on the absorption properties of the coating and base substrate material, chemical structure of the films, film layer thickness and laser irradiation power density8. poly including PEDOT: PSS, The use of solution processed highly conductive materials such as metal nanoparticle inks or polymers as electrodes for field effect transistors (FET), can significantly reduce the overall cost of fabrication. The deposition of such conductive materials can be achieved by large area printing techniques, which do not require special vacuum-based deposition equipment such as evaporation or sputtering. In the recent years, various solution processed conductive inks for printing technology have been demonstrated. The choice of metallic inks for printing technologies is relatively limited, and only gold, silver and copper inks are currently available. Gold inks offer high workfunction, close to 5eV, however, they are prohibitively expensive. Copper inks have lower workfunction, and printed layers are typically not very stable in ambient conditions. Air stable copper inks with antioxidants have been demonstrated recently, but higher temperature annealing is required for sintering9. Among only few available metallic nanoparticle inks, silver nanoparticle ones are the most common, however, the workfunction of printed layers (~ 4.5eV) is not high enough to provide high quality ohmic contact to many hole transport organic semiconductors, and solution processable silicon nanowires. The need for higher workfunction inks has led to the development of hybrid inks such as silver/organic electronic acceptors with reported workfunction value of ~5.1eV 10. Whereas, conducting polymers, poly (styrenesulphonate), offer higher workfunction ϕ of ~5.1eV 11, and PEDOT:PSS has been widely used in printed electronics12. However, lower conductivity of solution processed PEDOT: PSS, as compared to printable metal inks, limits its usability in electronic devices. Alternatively, the conductivity of PEDOT: PSS have been improved by blending it with nanoparticles14,15. Carbon nanotubes in a polymer matrix are known to increase the conductivity of the films due to their high aspect ratio, thereby providing highly conducting paths inside the polymer-nanotube network16,17. It has been previously reported that increasing the concentration of single wall carbon nanotube in poly-N-vinyl carbazole from 0.26% to 0.43% have increased the conductivity of the films by more than two orders of magnitude18. The electrodes patterned from thin films of carbon nanotubes and polymer can provide large active areas and high current output because of the presence of large number of carbon nanotubes for charge transport. A blend of PEDOT: PSS and carbon nanotube is an ideal candidate for replacing metal electrodes such as silver inks in printable FETs. Thin composite films of PEDOT: PSS and functionalised carbon nanotube are easily deposited using techniques like spin coating, drop casting, bar coating, slot-die printing etc. The conductivity of the electrodes based on thin films of polymer and carbon nanotube blends is also anticipated to be higher, due to the enhanced charge transport properties. Fast and efficient electrode patterning techniques are required for industrial scalability. Laser ablation technique is an ideal candidate for patterning electrodes using PEDOT: PSS/f- MWCNT inks. Laser ablation is a direct structuring technique, which can completely remove both PEDOT: PSS and CNTs from the substrate, resulting in narrow electrode gaps (short channel lengths) essential for FETs. Hence laser ablation of polymer/carbon nanotube blend (3, 4-ethelenedioxythiophene): devices will result in a one-step patterning of electrodes for the FETs useful for upscale industrial production. Solution processable semiconducting inorganic nanowires (NWs) are ideal candidates for active layers in printable electronic devices, offering the advantage of significantly higher charge carrier mobility as compared to printable organic semiconductors19. Inorganic semiconducting single crystal nanowires maintain most of their bulk single crystalline properties including very efficient charge transport and highly ordered crystal lattices. Additionally, Supercritical Fluid-Liquid-Solid (SFLS) growth method for the production of silicon and germanium nanowires is efficient, and it can offer high throughput of nanomaterials of up to few kgs/day20. From device fabrication point of view, SFLS grown nanowires offer compatibility with low temperature processes, due to the possibility of separating the synthesis and device fabrication procedures 21,22. Furthermore, alignment of nanowires across the electrodes using electric field assembly technique, dielectrophoresis (DEP), results in highly ordered array of nanowires, a big advantage for transistor device21,23. In this work we demonstrate the feasibility of using laser patterned PEDOT: PSS/acid functionalised multiwall carbon nanotubes (f-MWCNT) composite electrodes for SFLS grown silicon nanowire field effect transistors (FETs) on flexible plastic substrates. We have shown an increase in conductivity of PEDOT: PSS after blending with f-MWCNT, and an increase in absolute workfunction values for blended samples as compared to pristine PEDOT:PSS. We demonstrate the alignment of nanowires across laser patterned PEDOT: PSS/f-MWCNT electrodes using dielectrophoresis technique. Finally, we illustrate fully working Si NW-FETs with laser patterned PEDOT: PSS/f-MWCNT electrodes on flexible substrate, with excellent gate modulation and good transistor parameters. Results and Discussion PEDOT: PSS and f-MWCNT composite electrodes Functionalisation of carbon nanotubes and preparation of formulations are described in detail in Experimental sections Ⅰ, Ⅱ. Thin films of PEDOT: PSS and PEDOT: PSS/f-MWCNT were prepared on polyethylene terephthalate (PET) substrates, via drop casting technique, resulting in a film thickness of ~ 800nm. Such film thickness of PEDOT:PSS/f-MWCNT films makes it difficult for imaging techniques such as SEM of AFM to examine the presence of carbon nanotubes. Raman spectroscopy offers the potential to determine the presence of MWCNT in PEDOT:PSS polymer matrix by identifying characteristic Raman peaks associated with carbon nanotube vibrations. Films of PEDOT:PSS and PEDOT:PSS/f-MWCNT were examined using Renishaw system 2000 Raman spectrometer with 782nm excitation laser. Raman spectra of both samples were obtained between 95cm-1 to 3500 cm-1 as demonstrated in figure 1 (A). The PEDOT:PSS sample has a G-peak at 1592 cm-1 with a D-peak at 1435 cm-1, when compared to the MWCNT with a G-peak at 1586 cm-1 and a significant D-peak at 1348 cm-1. The G peak of PEDOT:PSS can be attributed to the C=C asymmetric stretching of the thiophene rings at the centre and the ends of the polymer chain, and the D-peak is due to the C-C stretching vibrations between the localised excitations14. The D peak for f-MWCNT is due to the defects caused by acid functionalisation, and the G-peak is due to the lattice vibration of the carbon atoms24. However, the PEDOT:PSS/MWCNT samples illustrated two significant D-peaks at 1359 cm-1 and 1435 cm-1 which correspond to the D-peak from PEDOT:PSS and MWCNT's. G-peak at 1582 cm-1 was also observed as shown in figure 1 (A). The ID/IG ratio was estimated to be 0.85, which gives the information about the amount of impurities in the sample. The presence of carbon nanotubes can be verified from the presence of a strong 2D peak at 2708 cm-1 in Raman spectra of PEDOT:PSS/MWCNT sample in comparison with the PEDOT:PSS sample, where this peak is absent. 2D peak at 2708 cm-1 is an overtone of the D peak and its existence is common in nanocarbons with sp2 hybridised carbon orbitals, such as carbon nanotubes in our sample25. Thus, Raman data in figure 1 (A) confirms the presence of carbon nanotubes in PEDOT:PSS / MWCNT composite film samples. The effect of the addition of f-MWCNT into PEDOT:PSS matrix on electrical properties of films was determined by performing two-point probe current-voltage characteristics measurements of PEDOT: PSS and PEDOT: PSS/f- MWCNT composite samples. Figure 1 (B) shows the current vs voltage characteristics of PEDOT:PSS/f- MWCNT and PEDOT:PSS thin films of the same thickness. The conductivity values were extracted from figure 1 (B), PEDOT: PSS/f- MWCNT film gave higher conductivity value of 3.9  10-3 Sm-1 when compared to the conductivity value of 5  10-7 Sm-1 achieved for the pristine PEDOT:PSS film. The increase in conductivity for the PEDOT:PSS/f-MWCNT samples can be due to the more conductive PEDOT components and depletion of PSS on the surface14, and due to the presence of highly conducting MWCNT. In addition to the current-voltage characteristic, we also determined the absolute work function of PEDOT:PSS/MWCNT and PEDOT:PSS films using Kelvin probe method. The absolute workfunction value for PEDOT:PSS was estimated to be ~ 4.9eV ± 0.1eV, whereas PEDOT:PSS/MWCNT blend films gave an absolute workfunction of ~ 5.4eV ± 0.1eV. Higher workfunction values obtained for PEDOT:PSS/MWCNT films makes them an ideal candidate for high workfunction printable inks for electrodes. Figure 1: Characterisation of PEDOT: PSS and PEDOT: PSS/MWCNT thin films, (A) Raman spectroscopy illustrates the presence of D and 2D peaks for PEDOT: PSS/MWCNT thin films; (B) Two point probe I-V measurements showing higher current of composite PEDOT: PSS/MWCNT thin film as compared to pristine PEDOT:PSS film. Laser Patterning of PEDOT:PSS/MWCNT electrodes Films of PEDOT: PSS/f-MWCNT (~800nm) on PET substrates were used to fabricate source/drain electrodes for transistors, via laser ablation technique. The fabrication conditions are described in Experimental section Ⅲ. Figure 2 (B) shows polarized optical microscope image of a typical sample, obtained by laser ablation of PEDOT:PSS/MWCNT film, resulting in 36µm channel gap between two electrodes. To check if the conducting material was completely removed in the channel gap area, current-voltage characteristics were measured between the two electrodes patterned by laser ablation, using Keithley SCS 4200 semiconductor analyzer. A higher resistance (~ 1013Ω) between the electrodes was observed after laser ablation of electrodes, as compared to un-treated PEDOT:PSS/MWCNT electrodes (~ 105Ω). High resistance in GΩ range across laser ablated electrodes illustrates the absence of PEDOT: PSS/MWCNT residue in the channel area, and an excellent isolation of the electrodes. For FET device fabrication, solution processed silicon nanowires, with 30- 50nm diameters and few tens of microns in length as shown in figure 2 (A), were deposited to bridge the channel, using dielectrophoresis (see Experimental section Ⅳ). A dense array of silicon nanowires was observed in the channel area between the laser patterned PEDOT: PSS/MWCNT electrodes as shown in figure 2 (C). Figure 2: (A) TEM image of SFLS grown silicon nanowires on a carbon grid, (B) Optical polarised microscope image of PEDOT: PSS/MWCNT electrodes patterned using laser ablation, (C) Dark-field polarised optical microscope image of DEP aligned silicon nanowires on laser patterned PEDOT: PSS/MWCNT electrodes. Silicon nanowire FETs Top-gate NW FETs were fabricated using the laser ablated PEDOT: PSS/f-MWCNT electrodes with DEP aligned Si-NWs (see Experimental section Ⅳ). The schematic of top-gated Si-NEW FET fabricated on top of the aligned nanowires is shown in figure 3 (A), with a channel length of ~ 36µm and channel width ~ 500µm (area of nanowire coverage). Transfer characteristics of Si-NW FETs were obtained by sweeping the gate voltage (VG) from +20V to -60V at 1V/s gate sweep rate, and measuring the drain current (ID) at a constant drain bias voltage (VD), as shown in figure 3 (A). From the transfer characteristics, it is observed that Si-NW FET showed p-type conduction and good modulation with negative gate voltages. The output characteristics obtained by measuring drain current for different drain voltage, at constant gate voltage as shown in figure 3 (B). An increase in drain current for different gate bias voltages in the output characteristics demonstrates good gate electrode - channel coupling. Transistor ON-state current was relatively modest, being in the range of 10s of nano- amperes, which could be attributed to a long channel length of the transistor, and also very dense coverage of nanowires. Such dense channel coverage could lead to a strong screening of the gate electric field by the nanowires, resulting in a modest charge carrier accumulation in Si nanowires channel. The effective device mobility was calculated using the following equation. (1) μ= × × 1× Where L and W are the respective channel length and channel width, VSD is the drain voltage and Cins is the capacitance of the parylene C (~ 2.7nF/cm2). Gm (~ ∂ID/∂VG) is the transconductance of the transistor. Due to a dense 'mat' of nanowires it was not possible to calculate the exact number of nanowire crossing the channel, and thus, only effective device mobility was estimated. Figure 3: Si-NW FET characteristics; (A) transfer characteristics obtained by scanning gate voltage from +20V to -60V at 1V/S, at constant drain bias voltages -5V and -10V, (B) Output characteristics obtained by scanning drain voltage from 0V to -10V, at constant gate voltages 0V, -10V, -20V, -30V and -40V. Inset shows a photographic image of FETs on flexible PET substrate. A transconductance value () of ~ 0.12nS, and an ION/IOFF ratio of ~ 200 were obtained from transfer characteristics, shown in figure 3(A). The effective device mobility value was calculated to be ~ 6.4×10-4cm2/Vs using equation 1. Conclusions In summary, we have developed a high workfunction, printable PEDOT:PSS/MWCNT composite ink for source-drain electrodes fabrication in nanowire FETs. Importantly, this seems to be the only low-cost inks with high workfunction compatible with silicon nanoparticles valence band edge (5.1eV) for p-type transport. Gold nanoparticle inks, although, are significantly more expensive, and their sintering temperature usually exceeds glass transition temperature of flexible PET substrates, making them incompatible with low temperature printed electronics. the We have illustrated a higher conductivity value of ~ 3.9  10-3 Sm-1 for PEDOT: PSS/MWCNT films as compared to the pristine PEDOT: PSS film (~ 5  10-7 Sm-1). Furthermore, it was also observed that the work function value of PEDOT: PSS films has increased from ~ 4.9eV ± 0.1eV to ~ 5.4eV ± 0.1eV with the addition of MWCNTs. Increase in workfunction values can help to improve the charge injection from the electrodes into the semiconductor for the p-type FETs that are based not only silicon nanowires, but also high ionization potential semiconducting small molecules and polymers. The source/drain electrodes were patterned from PEDOT: PSS/MWCNT films using laser ablation process on plastic substrates. We have routinely obtained ~36µm gap between the S/D electrodes. Top gate Si-NW FETs were fabricated from laser patterned composite electrodes. Devices showed an ION/IOFF ratio > 102, with an effective device mobility values of ~ 6.4×10-4cm2/Vs. Overall, silicon nanowire FETs with laser ablated PEDOTPSS/MWCNT electrodes demonstrated feasibility of using high workfunction composite PEDOT:PSS/MWCNT inks for nanowire transistors. In is envisaged that laser ablation of transistor electrodes can reduce the overall cost of printed electronic devices. Experimental I. Functionalisation of carbon nanotubes Functionalisation of the carbon nanotubes is an important process to: (i) make CNTs soluble in common solvents used for conducting polymers, and (ii) prevent the carbon nanotubes from agglomeration and forming bundles in the thin films, and also maintaining good dispersion within the polymer matrix. MWCNTs were purchased from Sigma Aldrich with the dimensions 7-15nm × 3-6nm  0.5-200m, and were treated using acid functionalisation technique. 500mg of MWCNT's were mixed with 30ml of concentrated sulphuric acid and concentrated nitric acid mixture (3:1). The solution was sonicated in an ultrasonic bath for 10 minutes, and refluxed over an oil bath for 1 hour at 130οC. The mixture was allowed to cool down and was diluted to 80ml using MilliQ deionised water, and was then transferred to two 50ml centrifuge tubes. The diluted solution was centrifuged at 8700rpm for 25min. A brown colored supernatant was observed which was then removed using a homemade vacuum pump, resulting in a black precipitate. The black precipitate was then diluted using MilliQ deionised water and the precipitate was suspended using a vortex mixer. The process was repeated twice to remove all the concentrated acid used for functionalisation, resulting in a black suspension. The solution was filtered using 0.1m polycarbonate membrane filter, washed with deionised water until the required pH (~7) level was reached, which was then finally washed with absolute ethanol (Fisher AR grade). The functionalized COOH-MWCNTs were dried in vacuum desiccator and stored as dry powder. For making formulations, dry COOH-MWCNTs were added into 50% ethanol/deionised water to a concentration of 0.5mg/ml. II. Preparation of the formulation: f-MWCNT / PEDOT The solution of 0.5mg/ml COOH-MWCNT was filtered using a 0.1 m polycarbonate membrane filter. 30ml of 0.5mg/ml was taken from the solution. 1.3% poly (3, 4- ethylenedioxythiophene) poly (styrenesulphonate) (PEDOT:PSS) purchased from Clevios was filtered using a 0.22 m filter followed by sonicating in an ultrasonic bath for 4 minutes. 1ml of PEDOT: PSS was mixed with 1ml of COOH-MWCNTs and transferred into a vial. The process was repeated so as to get 3 vials of PEDOT: PSS/MWCNTs with 1:1 ratio. III. Laser patterning of electrodes The Excimer Compex 205 (Lamda Physik) laser operating at 248 nm wavelength with repetition rate of 80Hz and a spot beam size (after focusing lens) of 36µm was used to pattern PEDOT:PSS/MWCNT electrodes. An 800nm thick film of PEDOT:PSS/MWCNT was prepared on the PET substrate using drop casting technique, and oven dried at 100C for 15min. The laser beam was homogenized to provide uniform intensity distribution in the beam cross- section. A fixed pattern to be drawn on the sample using the laser beam was pre-programmed. The sample was placed on an X-Y translation stage which was programmed to move at a speed of 10mm/sec. The beam was focused on the thin film and the programmed pattern was drawn on the film. Laser energy of 44mJ with a power density of 339.5 mJ/cm2 was required to remove all the material from the channel area for a thickness of 800nm. The channel length obtained was 36µm measured using the optical polarization microscope, and the channel depth was found out to be 855nm using profilometer. The profilometer measurement suggests that the laser beam is penetrating almost 55nm into the substrate, after removing the PEDOT: PSS/MWCNT thin film. IV. Fabrication of top-gated silicon nanowire transistor SFLS synthesised silicon nanowires with diameters of 30nm to 50nm and a few tens of microns in length were used to form semiconducting layer between laser patterned electrodes. Silicon nanowires were dispersed in anisole, and then aligned using dielectrophoresis method by applying AC voltage of 10Vpp to the PEDOT/MWCNT electrodes and then drop-casting nanowire dispersion on top. Excess anisole was removed, samples were dried, and 1µm thick layer of parylene C was deposited on top of the electrodes as the dielectric layer23, and 50nm thick gold gate electrode was evaporated using a Kurt Lesker thermal deposition system on top of the dielectric layer through a shadow mask. The transistor characteristics were measured in ambient air, at room temperature, using Agilent 4155C semiconductor parameter analyzer. Acknowledgements The authors would like to thank Dr. Simon Henley, (ATI, University of Surrey, UK) for his valuable insight and advice on laser ablation technique. The author would also like to extend their gratitude to Prof. Brian A. Korgel (University of Texas, Austin, USA) for providing the SFLS grown silicon nanowires. References (1) Tian, S.; Yang, F.; Zeng, D.; Xie, C. J. Phys. Chem. C 2012, 116 (19), 10586–10591. (2) Park, C. B.; Lee, J. E.; Na, H.; Kim, K. M.; Yoo, S. S.; Yang, M. S. Solid. State. Electron. 2015, 111, 227–233. (3) Chung, S. Y.; Kim, S.; Lee, J.-H.; Kim, K.; Kim, S.-W.; Kang, C.-Y.; Yoon, S.-J.; Kim, Y. S. Adv. Mater. 2012, 24 (45), 6022–6027. (4) Pease, R. F.; Chou, S. Y. Proc. IEEE 2008, 96 (2), 248–270. (5) Subho Dasgupta, Robert Kruk, Norman Mechau, and H. H. ACS Nano 2011, 5 (12), 9628–9638. (6) Green, R. a; Matteucci, P. B.; Dodds, C. W. D.; Palmer, J.; Dueck, W. F.; Hassarati, R. T.; Byrnes-Preston, P. J.; Lovell, N. H.; Suaning, G. J. J. Neural Eng. 2014, 11 (5), 56017. (7) Zhang, F.; Nyberg, T.; Ingana, O. Nano Lett. 2002, 2 (12), 1373–1377. (8) Schrödner, M.; Stohn, R.-I.; Schultheis, K.; Sensfuss, S.; Roth, H.-K. Org. Electron. 2005, 6 (4), 161–167. (9) Tsai, C.-Y.; Chang, W.-C.; Chen, G.-L.; Chung, C.-H.; Liang, J.-X.; Ma, W.-Y.; Yang, T.-N. Nanoscale Res. Lett. 2015, 10 (1), 357. (10) Whiting, G. L.; Arias, A. C. Appl. Phys. Lett. 2009, 95 (25), 253302. (11) Nardes, a. M.; Kemerink, M.; de Kok, M. M.; Vinken, E.; Maturova, K.; Janssen, R.A.J. Org. Electron. 2008, 9 (5), 727–734. (12) Leenen, M. a. M.; Arning, V.; Thiem, H.; Steiger, J.; Anselmann, R. Phys. Status Solidi 2009, 206 (4), 588–597. (13) Vacca, P.; Nenna, G.; Miscioscia, R.; Palumbo, D.; Minarini, C.; Sala, D. Della. J. Phys. Chem. C 2009, 113 (14), 5777–5783. (14) Ji, T.; Tan, L.; Hu, X.; Dai, Y.; Chen, Y. Phys. Chem. Chem. Phys. 2015, 17 (6), 4137– 4145. (15) Alshammari, A. S.; Shkunov, M.; Silva, S. R. P. Phys. Status Solidi - Rapid Res. Lett. 2014, 8 (2), 150–153. (16) Breuer, O.; Sundararaj, U. Polym. Compos. 2004, 25 (6), 630–645. (17) Curran, S. A.; Dalton, A. B.; Mccarthy, B. Phys. Rev. B 1998, 58 (12), R7492–R7495. (18) Tameev, a R.; Jiménez, L. L.; Pereshivko, L. Y.; Rychwalski, R. W.; Vannikov, A. V. J. Phys. Conf. Ser. 2007, 61, 1152–1156. (19) Cui, Y.; Zhong, Z. H.; Wang, D. L.; Wang, W. U.; Lieber, C. M. Nano Lett. 2003, 3 (2), 149–152. (20) Hanrath, T.; Korgel, B. a. Adv. Mater. 2003, 15 (5), 437–440. (21) Constantinou, M.; Stolojan, V.; Prabha Rajeev, K.; Hinder, S.; Fisher, B.; Bogart, T. D.; Korgel, B. a.; Shkunov, M. ACS Appl. Mater. Interfaces 2015, 7 (40), 22115–22120. (22) Prabha Rajeev, K.; Opoku, C.; Stolojan, V.; Constantinou, M.; Shkunov, M. Nanosci. Nanoeng. 2017, 5 (2), 17–24. (23) Opoku, C.; Hoettges, K. F.; Hughes, M. P.; Stolojan, V.; Silva, S. R. P.; Shkunov, M. Nanotechnology 2013, 24 (40), 1–7. (24) Zdrojek, M.; Gebicki, W.; Jastrzebski, C.; Melin, T.; Huczko, A. Solid State Phenom. 2004, 99–100 (265), 265–268. (25) Singh, B. P.; Samal, S.; Nayak, S.; Majhi, S. M.; Besra, L.; Bhattacharjee, S. Surf. Coat. Technol. 2017, 206 (6), 1319–1326.
1911.02317	1	1911	2019-11-06T11:16:53	A BaTiO3-Based Electro-Optic Pockels Modulator Monolithically Integrated on an Advanced Silicon Photonics Platform	[ "physics.app-ph", "physics.optics" ]	To develop a new generation of high-speed photonic modulators on silicon-technology-based photonics, new materials with large Pockels coefficients have been transferred to silicon substrates. Previous approaches focus on realizing stand-alone devices on dedicated silicon substrates, incompatible with the fabrication process in silicon foundries. In this work, we demonstrate monolithic integration of electro-optic modulators based on the Pockels effect in barium titanate (BTO) thin films into the back-end-of-line of a photonic integrated circuit (PIC) platform. Molecular wafer bonding allows fully PIC-compatible integration of BTO-based devices and is, as shown, scalable to 200 mm wafers. The PIC-integrated BTO Mach-Zehnder modulators outperform conventional Si photonic modulators in modulation efficiency, losses, and static tuning power. The devices show excellent V{\pi}L (0.2 Vcm) and V{\pi}L{\alpha} (1.3 VdB), work at high speed (25 Gbps), and can be tuned at low static power consumption (100 nW). Our concept demonstrates the possibility of monolithic integration of Pockels-based electro-optic modulators in advanced silicon photonic platforms. {\c} 2019 Optical Society of America. Users may use, reuse, and build upon the article, or use the article for text or data mining, so long as such uses are for non-commercial purposes and appropriate attribution is maintained. All other rights are reserved. https://www.osapublishing.org/jlt/abstract.cfm?URI=jlt-37-5-1456 Publication date: March 1, 2019 This work was supported in part by the European Union (EU) under Horizon 2020 grant agreements no. H2020-ICT-2015-25-688579 (PHRESCO) and H2020-ICT-2017-1-780997 (plaCMOS).	physics.app-ph	physics	> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1 A BaTiO3-based electro-optic Pockels modulator monolithically integrated on an advanced silicon photonics platform Felix Eltes, Christian Mai, Daniele Caimi, Marcel Kroh, Youri Popoff, Georg Winzer, Despoina Petousi, Stefan Lischke, J. Elliott Ortmann, Lukas Czornomaz, Lars Zimmermann, Jean Fompeyrine, Stefan Abel Abstract -- To develop a new generation of high-speed photonic modulators on silicon-technology-based photonics, new materials with large Pockels coefficients have been transferred to silicon substrates. Previous approaches focus on realizing stand-alone devices on dedicated silicon substrates, incompatible with the fabrication process in silicon foundries. In this work, we demonstrate monolithic integration of electro-optic modulators based on the Pockels effect in barium titanate (BTO) thin films into the back-end-of-line of a photonic integrated circuit (PIC) platform. Molecular wafer bonding allows fully PIC-compatible integration of BTO-based devices and is, as shown, scalable to 200 mm wafers. The PIC-integrated BTO Mach-Zehnder modulators outperform conventional Si photonic modulators in modulation efficiency, losses, and static tuning power. The devices show excellent VπL (0.2 Vcm) and VπLα (1.3 VdB), work at high speed (25 Gbps), and can be tuned at low static power consumption (100 nW). Our concept demonstrates the possibility of monolithic integration of Pockels-based electro-optic modulators in advanced silicon photonic platforms. Index Terms -- Electrooptic modulators, Monolithic integrated circuits, Silicon photonics I. INTRODUCTION S ILICON technology based photonic integrated circuits (Si PIC) are becoming essential for various applications in the domain of communication technologies [1]. For large data centers, Si PIC technology offers attractive features for transceivers data-center communication. Utilizing advanced manufacturing techniques for the co-integration of optics and electronics enables high- speed and cost-effective take transceiver solutions targeting intra and inter that This work was supported in part by the Swiss National Foundation under project 200021_159565 PADOMO, by the European Commission under grant agreement no. H2020-ICT-2015-25-688579 (PHRESCO) and H2020-ICT- 2017-1-780997 (plaCMOS), and by the Swiss State Secretariat for Education, Research and Innovation under contract no. 15.0285. F. Eltes, D. Caimi, Y. Popoff, L. Czornomaz, J. Fompeyrine, and S. Abel are with IBM Research -- Zurich, Säumerstrasse 4, 8803 Rüschlikon, [email protected], Switzerland [email protected], [email protected], [email protected]). [email protected], [email protected], (e-mail: advantage of device scaling opportunities. Targeting a truly monolithic integration with CMOS (or Bi-CMOS) is crucial for such transceivers. The co-use of the back-end of line (BEOL) by photonic and electronic devices results in the smallest possible parasitics, which is a pre-requisite for efficient RF driving. Standard Si PIC modulators are based on phase shifters using the free-carrier dispersion effect. In terms of phase-shifter properties this is not the optimum solution. Besides a rather low modulation efficiency, nonlinearity and high-loss also limit modulator performance. The impossibility of disentangling amplitude and phase modulation also restricts their use for higher modulation formats [2], [3]. In addition, the high junction capacitance limits the achievable bandwidth [3] and is detrimental for power consumption. It is therefore highly desirable to enable - in a silicon photonic technology - pure electro-optic phase shifters exploiting the Pockels effect, in order to provide a solution without residual amplitude modulation, yet with high linearity, high efficiency and low optical loss. Recently, this field of research experienced a renaissance, with several attempts to demonstrate Pockels modulators potentially compatible with Si-PIC. Different strategies are being followed, using either a strain-induced Pockels effect in silicon [4], using well-known Pockels materials such as LiNbO3, bonded onto silicon by direct wafer bonding [5], [6], or introducing novel materials with large Pockels coefficients [7], [8]. All these approaches have intrinsic weaknesses, coming either from a weak Pockels effect [4], the limited availability of large wafer sizes [5], [6], thermal stability issues [8], or incompatibility with standard fabrication processes [7]. Our approach utilizes single crystalline, ferroelectric BaTiO3 (BTO) as a material having a large Pockels coefficient and M. Kroh was with IHP, Im Technologiepark 25, 15236 Frankfurt (Oder), Germany. He is now with Silicon Radar GmbH, Im Technologiepark 1, 15236 Frankfurt (Oder), Germany (e-mail: [email protected]). C. Mai, G. Winzer, D. Petousi, S. Lischke, L. Zimmermann are with IHP, Im Technologiepark 25, 15236 Frankfurt (Oder), Germany (e-mail: cmai@ihp- microelectronics.com, [email protected], petousi@ihp- microelectronics.com, [email protected], [email protected]). L. Zimmermann is with Technische Universitaet Berlin, FG Si-Photonik, Einsteinufer 25, 10587 Berlin, Germany. Y. Popoff is with EMPA, 8600 Dubendorf, Switzerland. J. E. Ortmann is with the Department of Physics, The University of Texas, Austin, Texas 78712, United States (e-mail: [email protected]). > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2 where an integration path of single crystalline layers with silicon does exist [9] -- [11]. Over the past years, great progress has been made in developing a hybrid BTO/silicon technology, including passive structures with low-propagation losses [12], active electro-optic switching [13], [14], excellent thermal stability [15], and, very recently, large Pockels coefficients of r42 = 923 pm/V and high-speed modulation in photonic devices[10], [11]. However, previous work was developed on silicon-on-insulator substrates -- without attention to process integration in a standard PIC or electronic PIC (EPIC) process. Here, we overcome this limitation and demonstrate the integration of highly efficient BTO Pockels modulators in the BEOL of a silicon photonic process flow and show the scalability of our approach up to 200 mm, making this an attractive technology for high-speed transceivers. II. TECHNOLOGY CONCEPT Our concept of high-speed transceivers relies on the monolithic integration of BTO thin films via direct wafer bonding above an interlayer dielectric (ILD) in a standard EPIC flow (Fig. 1a) [16]. The bonding step can be performed on top of any ILD above the front-end-of-line (FEOL) structures. Using wafer bonding we can first deposit BTO epitaxially on a silicon substrate, and then transfer the epitaxial layer onto an amorphous substrate, such as an ILD. Having an epitaxial BTO film is of importance for two reasons: First, the low defectivity in single-crystalline films is crucial for achieving a large effective Pockels effect in the material [17]. Second, the low surface roughness of epitaxial BTO films is critical for obtaining high bonding yield. To fabricate BTO thin films, we use a deposition process based on molecular beam epitaxy [9], [18], which relies on the epitaxial growth on Si wafers and can thus be scaled to large wafer sizes. The availability of large substrates is major a benefit compared to the bonding of LiNbO3 on silicon or to the epitaxial growth on crystalline oxides, both approaches being limited by the available substrate or donor crystal sizes. The BTO devices are based on a strip-loaded waveguide geometry, where a Si strip on top of BTO guides the optical mode (Fig. 1b). Lateral electrodes for phase shifters are made using metal lines fabricated in the top metal level of the BEOL before BTO integration, combined with a final metallization after BTO integration. Optical simulations are used to inform the design of the BTO-Si waveguides and to ensure substantial overlap of the transverse electric (TE) optical mode and the BTO layer at a wavelength of 1550 nm. Using a 170-nm-thick BTO layer loaded with a 100-nm-thick Si strip, we achieve an optical overlap of 38% between the first order TE mode and the BTO layer. The BTO/Si phase shifters can be used in Mach- Zehnder modulators (MZMs), or ring modulators. In this work we used unbalanced MZMs, with multi-mode interference splitters. Grating couplers were used to couple light in and out of the devices (Fig. 1c). The magnitude of the refractive index change induced by the Pockels effect is strongly anisotropic and depends on the relative orientation of the crystalline axes, the optical electrical field, and the modulating electric field [10], [19]. To ensure the maximum response we designed phase shifters with waveguides along the BTO[110] direction. III. INTEGRATION AND FABRICATION We deposited BTO thin films on SrTiO3-buffered silicon-on- insulator (SOI) substrates with 100 nm top Si using a previously reported process [9], [18]. Deposition of BTO using molecular beam epitaxy ensures a high-quality single-crystal film. After BTO deposition, we transferred the BTO layer and the top Si layer onto a planarized acceptor wafer using thin alumina layers for adhesion at the bonding interface. The donor wafer was subsequently removed by a combination of mechanical grinding and chemical etching, resulting in a high transfer yield from the source wafer. To demonstrate the scalability of our approach, we transferred BTO layers grown on a 200 mm SOI substrate onto another 200 mm silicon wafer that had been capped with a thermal oxide (Fig. 2a). The transferred BaTiO3 layer was Fig. 1. Scheme for monolithic integration of BTO/Si on PIC platforms. (a) Schematic cross-sections of PIC with electrical and optical front-end, and BTO integration in the back-end (this work) or front-end (future route). (b) Cross-section of active BTO/Si waveguide used for electro-optic modulators. The electrodes (shown in grey) are fabricated in the BEOL of the PIC platform. (c) Schematic layout of BTO/Si electro-optic modulator reported in this work. The BTO/Si active waveguide is used as phase shifter in a Mach-Zehnder modulator. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3 Fig. 2. Wafer bonding transfer of thin-film BTO between 200 mm source and target wafers. (a) Photo of transferred BTO layer. (b) Radial measurements of the homogeneity of the transferred BTO layer. XRD and ellipsometry was used to measure the lattice parameter, rocking curve, and thickness of the BTO, all of which show good homogeneity across the 200 mm wafer. thoroughly characterized using X-ray diffraction (XRD) and ellipsometry (Fig. 2b). The out-of-plane lattice parameter and rocking curve show good crystalline homogeneity with only minor variations along the 100 mm radius. Additionally, the thickness of the BTO, measured by ellipsometry, varies only minimally across the wafer. The observed variations in thickness contain a significant uncertainty due to variations also in the other layers of the stack. For the fabrication of modulators (Fig. 3), we used 200 mm target wafers, processed following a PIC flow having the same BEOL processes as EPIC runs [20]. In this work, the BEOL process of the PIC run was interrupted at the 4th metallization level, top metal 1 (TM1), after ILD planarization. We transferred a 170-nm-thick BTO layer from a 50 mm SOI wafer onto the planarized PIC wafer. Si waveguides were patterned by dry etching. In order to ensure a homogenous electric field across the BTO and to avoid a voltage drop over the thin ILD layer between BTO and TM1, vias to TM1 were etched through the BTO and the ILD along the waveguides. With a final metallization step, we extended the buried RF lines on top of the BTO. A cross-sectional electron micrograph (Fig. 4) demonstrates the successful fabrication of BTO/Si modulators on the PIC substrate. Direct wafer bonding using Al2O3 adhesion layers has a temperature budget well within the limits of the BEOL process [21]. However, annealing steps at temperatures up to 350°C are needed to reduce the propagation losses in the BTO layer [12]. Fig. 3. Simplified process flow for integration of BTO modulators in the BEOL of a PIC process flow. Schematics of the cross-sections (left) are shown at various steps in the process (right). After wafer-bonding of the BTO and Si layers, the Si is patterned into a strip-loaded waveguide, after which vias and contacts are fabricated. Figures (c)-(d) show a magnification of the region within the dashed rectangle in (b). It is therefore necessary to verify that the BTO integration does not cause any degradation of FEOL components. As the Ge photodiodes fabricated in the FEOL are highly sensitive to thermal degradation, we characterized their performance before and after BTO device integration. We cannot detect any degradation in either dark current or high-speed signal detection performance (Fig. 5). The absence of such degradation confirms that our integration strategy is compatible with the thermal limitations of the FEOL and BEOL processes, making integration of BTO devices compatible with PIC platforms, and fulfills the prerequisites for compatibility with EPIC platforms. IV. DEVICE PERFORMANCE To characterize the device performance, we used both passive ring resonators and active MZMs. The ring resonators had a radius of 30 μm, to ensure negligible bending losses, and allow accurate extraction of propagation loss. From the high Q- factor (~50,000) of ring resonators we extract a propagation loss of 5.8 dB/cm. Since the BTO layer itself has only minimal contributions to the propagation losses [12], we are instead limited by scattering losses in the Si waveguides. Using an optimized patterning process the propagation losses can be reduced further. We performed electro-optic characterization Fig. 4. Cross-sectional STEM image of BTO modulator integrated after top metal 1 (TM1) in BEOL process of a Si PIC wafer. The schematic shows how the modulator was integrated in this work. The electron micrograph shows the successful integration of BTO/Si modulators. Intermediate metal levels (M1 to M3) as well as the FEOL levels can be identified. Fig. 5. Comparison of FEOL Ge photodiode performance before and after integration BTO modulators. The photodiodes were characterized by recording a modulated data signal, and by measuring the dark-current. No detectable degradation is caused by integration of BTO modulators, showing that the integration scheme is compatible with the PIC FEOL. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4 to less than 3 dB/cm, resulting the Si waveguide. By improved processing, propagation losses can be reduced in a VπLα < 0.7 VdB. The low VπLα shows one of the key advantages of the hybrid BTO/Si technology compared to alternative modulator concepts: BTO/Si shows both a large electro-optic response and low insertion losses, since neither high doping levels nor absorbing materials are needed in the modulator design. When sweeping the bias voltage, the phase shift of the MZM (Fig. 6a) exhibited a hysteretic behavior, consistent with the ferroelectric nature of BTO [10]. The hysteresis curve illustrates the need for poling the BTO layer with a bias above the coercive field (~ 1 V) to maximize the electro-optic response. For smaller bias voltages, mixed ferroelectric domain states result in a reduced effective Pockels effect reff, which ultimately vanishes for evenly populated domain states [9] (Fig. 6b). The cancellation effect of opposing domains causes a deviation from the linear phase response when varying a DC voltage: The total electro-optic response is the convolution of the linear Pockels effect and nonlinear domain switching effects. To isolate the Pockels effect from the electro-optical response we extract the Vπ at the extremes of the curve shown in Fig. 6a, where all domains have been poled. The re- orientation of ferroelectric domains is a relatively slow process (<<1 GHz) [24], which does not impact the operation of the modulator at high frequency -- even at a bias below the poling voltage. Moreover, as the Pockels effect is an electric-field effect, very low-power tuning of the MZMs is possible. The low leakage results in extremely low tuning powers, Pπ <100 nW (Fig. 7), compared to silicon thermo-optic tuning elements which typically have a Pπ >1 mW [25]. As the Pockels-effect is a linear EO effect the device bias can be used for tuning without changing propagation the modulation efficiency. losses and without affecting From the measured VπL it is possible to extract the effective Pockels coefficient reff of the BTO layer using eq. (1) 𝑟eff = 𝜆𝑔 3 𝛤BTO𝑉𝜋𝐿 𝑛BTO (1) as reff = 380 pm/V. Here, λ is the operating wavelength of 1.55 µm, g is the electrode-gap (2.6 µm), nBTO is the refractive index of BTO (2.29) as measured by ellipsometry on similar films, and ΓBTO is the EO interaction factor which can be estimated as the optical overlap with BTO (38%) assuming a homogenous electric field across the BTO. The magnitude of the extracted reff is in the range of expected values for BTO thin films: The electro-optic response exceeds values previously reported on MBE-grown BTO layers on silicon [9], [13], [17], but is smaller than those reported in ref. [10], where BTO films of very high crystalline quality with rocking curves of 0.3° are reported. The variation of the magnitude of the Pockels coefficients in similar material stacks is in agreement with the dependence of the electro-optic response on the actual crystalline quality and film morphology [17]. To determine the high-frequency response, small-signal electro-optic S21 measurements were performed on a MZM with 1-mm-long electrodes (Fig. 8). The 3-dB bandwidth is 2 GHz. Fig. 6. (a) Induced phase shift when applying voltage to one arm of a 2-mm- long MZM. The response is linear at large voltages but shows non-linear, hysteretic contributions for small biases due to ferroelectric domain switching in the BTO layer as visualized in (b): The yellow arrows correspond to the polarization of ferroelectric domains. In the middle configuration the EO response of oppositely oriented domains cancel out, resulting in a vanishing effective Pockels coefficient. A sufficient bias voltage can align the domains to maximize the EO response. The ferro-electric domain switching is a slow effect that does not occur at frequencies >1 GHz. (c) Transmission spectra at various bias voltages in the poled regime used for VπL extraction. on unbalanced MZMs with phase shifter lengths of 1-2 mm. By applying a voltage to one arm of the MZM and recording the induced phase shift as a function of the applied voltage (Fig. 6), we extracted the DC VπL value as 0.23 Vcm. This value is 10 times smaller than state-of-the-art Si depletion-type plasma- dispersion modulators (VπL~2 Vcm) [3, 17] and comparable to integrated semiconductor-insulator-semiconductor capacitor (SISCAP) modulators (VπL of ~0.2 Vcm) [23]. When taking into account propagation losses α we reach a VπLα of 1.3 VdB, which is significantly better than any available high- speed Si modulator (VπLα >10 VdB). In the current devices the propagation losses are limited by scattering from roughness in silicon Fig. 7. Power-voltage characteristics of a 2-mm-long MZM device. The leakage current is small in the full bias range (<100 nA), resulting in low static power consumption of ~100 nW at the operating point of 2 V. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5 was amplified (Vp ~2 V) and was then applied to one arm of the MZM along with a 2 V DC bias. The MZM was operated in a travelling wave configuration with an off-chip 50 Ω termination. The modulated optical signal was amplified (to compensate losses from grating couplers and from the experimental setup) and directly detected using a high-speed photodiode. Eye-diagrams were recorded on a sampling oscilloscope with 10, 20, and 25 Gbps data rates (Fig. 9). Non- closed eyes can be achieved even at 25 Gbps, however the result of the limited EO bandwidth of the modulator is qualitatively visible as a reduction eye opening from 10 to 25 Gbps. With an adapted electrode design, we expect to reach data rates >50 Gbps using MZMs. V. CONCLUSION We have shown how a material (BaTiO3) with the Pockels effect can be integrated into a silicon photonics platform in a scalable way using direct wafer bonding. The demonstrated Mach-Zehnder modulators show excellent performance, exceeding state-of-the-art silicon-based devices on several figures of merit, such as VπL and VπLα. The established integration concept provides a path for a novel generation of high-speed modulators and ultra-fast switches. The technology is however not limited to such existing components, but further enables entirely new types of devices on a silicon photonics platform. Using BTO, ultra-low-power tuning elements [14] and compact plasmonic devices [10], [29], as well as non- volatile elements for optical neuromorphic computing exploiting ferro-electric domain switching [30] are possible. REFERENCES [1] M. Hochberg and T. Baehr-Jones, "Towards fabless silicon photonics," Nat. Photonics, vol. 4, no. 8, pp. 492 -- 494, 2010. [2] G. T. Reed, G. Mashanovich, F. Y. Gardes, and D. J. Thomson, "Silicon optical modulators," Nat. Photonics, vol. 4, pp. 518 -- 526, 2010. [3] D. Petousi et al., "Analysis of Optical and Electrical Tradeoffs of Traveling-Wave Depletion-Type Si Mach -- Zehnder Modulators for High-Speed Operation," IEEE J. Sel. Top. Quantum Electron., vol. 21, no. 4, pp. 199 -- 206, 2015. [4] M. Cazzanelli et al., "Second-harmonic generation in silicon waveguides strained by silicon nitride," Nat. Mater., vol. 11, no. 2, pp. 148 -- 154, 2012. [5] P. Rabiei, J. Ma, S. Khan, J. Chiles, and S. Fathpour, "Heterogeneous lithium niobate photonics on silicon substrates," Opt. Express, vol. 21, no. 21, pp. 114 -- 115, 2013. [6] C. Wang et al., "Integrated lithium niobate electro-optic modulators operating at CMOS-compatible voltages," Nature, vol. 562, pp. 101 -- 104, 2018. [7] K. Alexander et al., "Nanophotonic Pockels modulators on a silicon nitride platform," Nat. Commun., vol. 9, p. 3444, 2018. [8] S. Koeber et al., "Femtojoule electro-optic modulation using a silicon -- organic hybrid device," Light Sci. Appl., vol. 4, p. e255, 2015. [9] S. Abel et al., "A strong electro-optically active lead-free ferroelectric integrated on silicon.," Nat. Commun., vol. 4, p. 1671, 2013. [10] S. Abel et al., "Large Pockels effect in micro- and nanostructured barium titanate integrated on silicon," Nat. Mater., vol. 18, no. 1, 2019. [11] F. Eltes et al., "Record High Pockels Coefficient in PIC-Compatible BaTiO3/Si Photonic Devices," in European Conference on Optical Communication, 2018, p. We4C.1. [12] F. Eltes et al., "Low-loss BaTiO3-Si waveguides for nonlinear integrated photonics," ACS Photonics, vol. 3, no. 9, pp. 1698 -- 1703, 2016. [13] C. Xiong et al., "Active silicon integrated nanophotonics : ferroelectric BaTiO3 devices," Nano Lett., vol. 14, no. 3, pp. 1419 -- 1425, 2014. [14] S. Abel et al., "A hybrid barium titanate-silicon photonics platform for ultra-efficient electro-optic tuning," J. Light. Technol., vol. 34, no. 8, pp. Fig. 8. Small signal frequency response of a 1-mm-long MZM, and a ring modulator with radius 10 µm. The bandwidth of the MZM is limited by mismatch between the optical and RF modes, whereas the ring modulator is limited by the photon lifetime. The reason for this limited bandwidth is a mismatch between the optical mode and the electrical mode in the travelling wave electrodes, caused by the high dielectric constant of the BTO. Our device designs are based on moderate BTO permittivity values of εBTO = 100 [26], which turned out to be strongly underestimated compared to recent reports of εBTO as high as 3000 in epitaxial BaTiO3 thin films [10], [27], [28]. To improve the bandwidth, the electrodes should be designed based on the actual properties of the BTO layer to achieve mode matching between the RF and optical modes. To show that the bandwidth is not limited by the electro-optic properties of the material but rather by the electrical design, we measured the bandwidth of a ring modulator with a 10 µm radius, the small radius induces bending losses resulting in a reduced Q-factor of ~15,000. The measured bandwidth of ~20 GHz (Fig. 8) is limited by the photon lifetime (Q-factor ~15,000) but demonstrates the potential for high bandwidth operation using BTO/Si devices -- as confirmed in previous reports [10], [15]. We further characterized the high-speed performance of the BTO/Si modulators with data-modulation experiments using a 1-mm-long MZM. An electrical pseudorandom binary sequence (PRBS) was generated with a bit-pattern generator, without pre-emphasis or any other signal processing. The signal Fig. 9. Eye-diagrams from back-to-back data transmission through 1-mm-long BTO MZM in single-drive mode at 10, 20, and 25 Gbps, respectively. A bias voltage of 2 V was applied during the experiments. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6 1688 -- 1693, 2015. [15] A. Messner et al., "Integrated Ferroelectric Plasmonic Optical Modulator," in Optical Fiber Communication Conference Postdeadline Papers, 2017, p. Th5C.7. [16] F. Eltes et al., "A novel 25 Gbps electro-optic Pockels modulator integrated on an advanced Si photonic platform," in 2017 IEEE International Electron Devices Meeting (IEDM), 2017, p. 24.5.1-24.5.4. [17] K. J. Kormondy et al., "Microstructure and ferroelectricity of BaTiO3 thin films on Si for integrated photonics," Nanotechnology, vol. 28, no. 7, p. 075706, 2017. [18] R. a. McKee, F. J. Walker, J. R. Conner, E. D. Specht, and D. E. Zelmon, "Molecular beam epitaxy growth of epitaxial barium silicide, barium oxide, and barium titanate on silicon," Appl. Phys. Lett., vol. 59, no. 7, pp. 782 -- 784, 1991. [19] P. Castera, D. Tulli, A. M. Gutierrez, and P. Sanchis, "Influence of BaTiO3 ferroelectric orientation for electro-optic modulation on silicon," Opt. Express, vol. 23, no. 12, p. 15332, 2015. [20] D. Knoll et al., "High-performance photonic BiCMOS process for the fabrication of high-bandwidth electronic-photonic integrated circuits," in Technical Digest - International Electron Devices Meeting, IEDM, 2015. [21] N. Daix et al., "Towards large size substrates for III-V co-integration made by direct wafer bonding on Si," APL Mater., vol. 2, no. 8, p. 086104, 2014. [22] D. Patel et al., "Design, analysis, and transmission system performance of a 41 GHz silicon photonic modulator," Opt. Express, vol. 23, no. 11, pp. 14263 -- 14287, 2015. [23] M. Webster, K. Lakshmikumar, C. Appel, C. Muzio, B. Dama, and K. Shastri, "Low - Power MOS- Capacitor Based Silicon Photonic Modulators and CMOS Drivers," in Optical Fiber Communication Conference, 2015, p. W4H.3. [24] W. Huang et al., "Ferroelectric domain switching dynamics and memristive behaviors in BiFeO3-based magnetoelectric heterojunctions," J. Phys. D. Appl. Phys., vol. 51, no. 23, 2018. [25] A. Masood et al., "Comparison of heater architectures for thermal control of silicon photonic circuits," in IEEE International Conference on Group IV Photonics GFP, 2013, vol. ThC2, pp. 83 -- 84. [26] S. Abel, "Electro-optic photonic devices based on epitaxial barium titanate thin films on silicon," Université de Grenoble, 2014. [27] T. Hamano, D. J. Towner, and B. W. Wessels, "Relative dielectric constant of epitaxial BaTiO 3 thin films in the GHz frequency range," Appl. Phys. Lett., vol. 83, no. 25, pp. 5274 -- 5276, 2003. [28] B. H. Hoerman, G. M. Ford, L. D. Kaufmann, and B. W. Wessels, "Dielectric properties of epitaxial BaTiO3 thin films," Appl. Phys. Lett., vol. 73, no. 16, pp. 2248 -- 2250, 1998. [29] A. Messner et al., "Plasmonic Ferroelectric Modulators," J. Light. Technol., 2018. [30] S. Abel, D. J. Stark, F. Eltes, J. E. Ortmann, D. Caimi, and J. Fompeyrine, "Multi-Level Optical Weights in Integrated Circuits," in 2017 IEEE International Conference on Rebooting Computing (ICRC), 2017.
1709.08752	1	1709	2017-09-25T23:32:59	Doping Graphene via Organic Solid-Solid Wetting Deposition	[ "physics.app-ph", "cond-mat.mtrl-sci", "physics.atm-clus" ]	Organic Solid-Solid Wetting Deposition (OSWD) enables the fabrication of supramolecular architectures without the need for solubility or vacuum conditions. The technique is based on a process which directly generates two-dimensional monolayers from three-dimensional solid organic powders. Consequently, insoluble organic pigments and semiconductors can be made to induce monolayer self-assembly on substrate surfaces, such as graphene and carbon nanotubes, under ambient conditions. The above factuality hence opens up the potential of the OSWD for bandgap engineering applications within the context of carbon based nanoelectronics. However, the doping of graphene via OSWD has not yet been verified, primarily owing to the fact that the classical OSWD preparation procedures do not allow for the analysis via Raman spectroscopy, one of the main techniques to determine graphene doping. Hence, here we describe a novel approach to induce OSWD on graphene leading to samples suitable for Raman spectroscopy. The analysis reveals peak shifts within the Raman spectrum of graphene, which are characteristics for p-type doping. Additional evidence for chemical doping is found via Scanning Tunneling Spectroscopy. The results open up a very easily applicable, low-cost, and eco-friendly way for doping graphene via commercially available organic pigments.	physics.app-ph	physics	Published in Journal Carbon (2017). DOI: 10.1016/j.carbon.2017.09.043. 1 © 2017. This manuscript version is made available under the CC-BY-NC-ND 4.0 license. Doping Graphene via Organic Solid-Solid Wetting Deposition Alexander Eberlea, Andrea Greinera, Natalia P. Ivlevab, Banupriya Arumugam c, Reinhard Niessnerb, Frank Trixlera,d,* a Department of Earth and Environmental Sciences and Center for NanoScience (CeNS), Ludwig-Maximilians-Universität München, Theresienstrasse 41, 80333 München, Germany b Institute of Hydrochemistry, Chair for Analytical Chemistry, Technische Universität München, Marchioninistr. 17, 81377 München, Germany c Institute for Nanoelectronics, Technische Universität München, Theresienstrasse 90, 80333 München, Germany d TUM School of Education, Technische Universität München and Deutsches Museum, Museumsinsel 1, 80538 München, Germany Abstract Organic Solid-Solid Wetting Deposition (OSWD) enables the fabrication of supramolecular architectures without the need for solubility or vacuum conditions. The technique is based on a process which directly generates two-dimensional monolayers from three-dimensional solid organic powders. Consequently, insoluble organic pigments and semiconductors can be made to induce monolayer self-assembly on substrate surfaces, such as graphene and carbon nanotubes, under ambient conditions. The above factuality hence opens up the potential of the OSWD for bandgap engineering applications within the context of carbon based nanoelectronics. However, the doping of graphene via OSWD has not yet been verified, primarily owing to the fact that the classical OSWD preparation procedures do not allow for the analysis via Raman spectroscopy – one of the main techniques to determine graphene doping. Hence, here we describe a novel approach to induce OSWD on graphene leading to samples suitable for Raman spectroscopy. The analysis reveals peak shifts within the Raman spectrum of graphene, which are characteristics for p-type doping. Additional evidence for chemical doping is found via Scanning Tunneling Spectroscopy. The results open up a very easily applicable, low-cost, and eco-friendly way for doping graphene via commercially available organic pigments. * Corresponding author. Tel.: +49 89 2179 509. E-mail: A. Eberle et al. / Carbon 00 (2017) 000–000 2 1. Introduction Nearly a decade ago, it became evident that the miniaturisation of silicon-based electronics is limited and that it will soon reach its termination [1-2]. As a result, numerous scientists began exploring the prospects of carbon-based nanoelectronics, to utilise the outstanding electronic properties of graphene, and thus to increase the performance of existing and to develop future electronic devices like the flexible or inkjet-printed electronics [2-10]. However, a deeper insight revealed the production of functional nano-systems, like the nanoscale transistors, to be quite a challenging task [10-14]. A promising approach nonetheless, for the fabrication of essential semiconductive sub-regions, came forward as providing a bandgap to the graphene substrate, by covering it with a monolayer of an organic semiconductor [6-8]. Such a monolayer, in turn, can be built up by the bottom-up technologies (as the mostly available the vapor deposition- [15] or the liquid phase deposition- [16] techniques), directing the self- assembly of organic molecules via the non-covalent interactions (hydrogen bonding, Van-der- Waals, π–π stacking, and electrostatics) [10-14]. However, the processing of organic semiconductor and pigment molecules imposes its own limitations: only few of these compounds survive the thermally enforced vacuum sublimation unscathed that is necessary to apply vapour deposition methods, as the organic molecular beam deposition technique [16-18]. Further, as most of the organic pigments with promising semiconductive properties are insoluble in common liquids, liquid phase deposition techniques like the drop-casting or spin-coating [18] call for an additional chemical functionalisation [18-21]. Nevertheless, in relation to the standard pigments employed usually in the industrial sector, the custom synthesis of functionalised semiconductors marks as an extensive and cost- intensive process. As an alternative approach, we hence developed the Organic Solid-Solid Wetting Deposition (OSWD) technology, an environmental friendly, cheap, up-scalable, and both, straightforward and quick to perform procedure [19-22]. The OSWD being based on the solid-solid wetting effect [23-25], it is the gradient of surface free energy that acts as the prime driving force behind the technology. Briefly summarising its basics, it can be said that the surface free energy of organic semiconductor crystals that physically contact an inorganic substrate like graphite, graphene, carbon nanotubes or MoS2 [20], gets modified when appropriate organic or aqueous dispersing agents are used. As a consequence, a solid-solid wetting process is triggered, detaching semiconductor molecules from the attached crystal and adsorbing them to the substrate surface. Subsequently, the adsorbed molecules assemble into supramolecular architectures, covering the substrate surface [19-21]. A. Eberle et al. / Carbon 00 (2017) 000–000 3 However, in this regard, worth investigating was, if the OSWD generated surface coverage dopes the graphene substrate and thus induces a bandgap. Since a bandgap alters the spectroscopic properties of a material, such a modification, if induced, can be detected via the Raman spectroscopic analysis [6-8]. Raman spectroscopy has been known as a fast and a non- destructive high-resolution technique, which can be employed to study the fundamental physical properties of carbon nanomaterials, such as determining their layer thickness, detecting structural defects, and verifying graphene doping [6,34-39]. It can thus be said to be a reliable and widely used method for investigating the doping of graphene. However, in order to perform an accurate Raman measurement, the substrate surface needs to be covered with adsorbate layers, freely accessible for the laser beam and featuring homogeneity of the order of few hundred nanometres magnitude. Unfortunately, the hitherto used standard OSWD preparation technique fails to generate such a covering, thereby calling for a modification of the approach. Hence, in an attempt to modify the OSWD technique for gaining samples suitable for the Raman spectroscopy analysis, a series of experimental tests were performed, their results being presented and discussed in the following sub-sections. In this regard, initial efforts were made to enhance the surface coverage of the sample substrate by incorporating a reworking step. Furthermore, a new 'thermally triggered' sample preparation technique was tested and is put forward, with an aim of potentially triggering the OSWD process without employing a catalysing dispersing agent. Post successful generation of suitable Raman samples, the results of the Raman spectroscopy analyses are discussed, as to determine whether the OSWD produced supramolecular surface covering modifies the substrate's electronic properties by providing a bandgap or not. Finally, the results of a series of Scanning Tunneling Spectroscopy (STS) tests are presented, as to verify the outcomes of the Raman analysis via an independent, additional experimental technique. STS is especially suited in this regard, owing to its sensitivity in probing the chemical doping of graphene, by providing an atomic resolution analysis of the local electronic properties of a surface [26-32]. 2. Results and discussions 2.1. Refinement of the standard sample preparation method by incorporating a reworking step As per the results of the previous investigations, the substrate surface coverage generated via the OSWD can be altered to a large extent by substituting the dispersing agent in use, i.e. without replacing the organic semiconductor itself [20]. Until now, for the samples fabricated via the standard preparation method, the maximum achievable surface coverage rate was A. Eberle et al. / Carbon 00 (2017) 000–000 4 limited to approx. 67 %, not being sufficient enough for the execution of Raman spectroscopy measurements. Thus, to accomplish such analysis and to enhance the surface coverage of the 'traditionally prepared' samples, the incorporation of a reworking procedure was considered. For this, and to explore the physio-chemical basics of the OSWD process, model systems out of highly oriented pyrolytic graphite (HOPG) as the substrate material, and the organic semiconductive pigment gamma quinacridone (γQAC) as the active phase were utilized. γQAC is known to be a cheap and commercially available pigment with promising electrical properties, low toxicity, an excellent physical and chemical stability, and with biocompatibility for applications in the living organism [40-46]. Furthermore, the gamma polymorph has been known to be the most stable out of the four possible, three-dimensional crystal structures, this polymorph being built up by the linear QAC molecules (refer Fig. 1 (a)) connected with their neighbours via four hydrogen bonds of the type NH···O=C [47]. Presuming the successful processing of three-dimensional γQAC crystals into substrate surface adsorbate structures through the OSWD approach, the quinacridone molecules (QAC) have been investigated to arrange themselves in one-dimensional supramolecular chains via the NH···O=C hydrogen bonds. Further, multiple parallel and side-by-side appearing chains have been reported to form supramolecular arrays (refer Fig. 1 (b)) [19-20]. Note: the abbreviation QAC is used for quinacridone in general, usually relating to either quinacridone molecules or quinacridone adsorbate structures, whereas the term γQAC is employed for the 3D gamma polymorph of quinacridone. (a) (b) Fig. 1. OSWD induced monolayer self-assembly of QAC on graphene. (a) Chemical structure of the QAC molecule. (b) Upper section: supramolecular QAC structures situated atop single layer graphene on copper as the substrate, with the observed structures having lattice A. Eberle et al. / Carbon 00 (2017) 000–000 5 parameters as: a = 0.72 ± 0.02 nm, b = 2.06 ± 0.02 nm, and an intermediate angle of 89 ± 2 °. Further, the superimposed inlay on the QAC structures (little right of the image) depicts the arrangement of single QAC molecules in one-dimensional chain-like structures. The bottom of the image, in addition, presents the underlying graphene substrate's structure, with lattice parameters of the graphene unit cell as: g1 = g2 = 0.246 nm. Besides, the white marked hexagon in the image (atop the graphene substrate) represents one carbon ring of the graphene structure, with an atom to atom distance of 0.142 nm. Force field calculations in this respect revealed that a γQAC crystal comprises of at least one crystal face, in which the QAC molecules have binding energy less than the binding energy of a molecule adsorbed on a graphene substrate [19]. From these calculations, it can be hence deduced that only on the condition that the γQAC crystal contacts the HOPG with one of its energetically favourable crystal faces, QAC molecules can detach and subsequently attach themselves to the HOPG substrate and thus initiate the self-assembly processes. In addition, experiments revealed that a complete coverage of the HOPG surface by the supramolecular QAC arrays could not be achieved, although the standard sample preparation technique covers the entire HOPG surface with a distinct layer of γQAC crystals (i.e. γQAC powder) (refer Fig. 2 (a)). From the above theoretical and experimental results, it can hence be deduced that the OSWD approach is an anisotropic process. The latter deduction, in turn, proposes a way of subsequently increasing the surface coverage of the sample, by gently rubbing the remaining γQAC powder against it. Such a procedure, supposedly, forces the γQAC crystals to roll over the HOPG surface, thereby significantly increasing the chances of specific crystal faces to contact the HOPG. For the execution of the above, the virgin HOPGs were hence initially treated with a dispersion of γQAC and the dispersing agent octylcyanobiphenyl (8CB), the latter is known to be one of the few dispersing agents that neither does vaporise at room temperatures nor disturbs the STM measurements (further information on the 8CB's chemical structure and its ability to self-assemble stable and well-ordered arrays being available in the supplementary data). Subsequent STM measurements of the samples prepared in the above manner revealed an overall surface coverage of 50 ± 4 %, including twice the standard deviation (refer Fig. 2 (a)). Thereinafter, using a metal spatula, the remaining γQAC powder was gently rubbed against the substrate, and consequently the results depict a greater overall surface coverage of 98 ± 2 % (refer example picture in Fig. 2 (b)). Hence, it can be said that the incorporated reworking step enabled almost complete surface coverage, though the surface covering displayed still various A. Eberle et al. / Carbon 00 (2017) 000–000 6 arrays with different orientations. Therefore, since the substrate surface covering obtained by the above sample preparation procedure does not exhibit homogeneity of the order of magnitude of a few hundred nanometres, the above sample preparation method cannot be hence used to prepare samples suited for accurate Raman spectroscopy measurements. (a) (b) Fig. 2. Large-area STM scans of supramolecular QAC arrays (white markings highlighting the borders of the QAC arrays). (a) Example STM image of a supramolecular surface covering, atop a HOPG substrate, generated using a dispersion of γQAC and 8CB as the dispersing agent; the average surface coverage rate being 50 ± 4 %. (b) Example STM image post gently rubbing the remaining γQAC powder with 8CB being still present; the average surface coverage in this case being 98 ± 2 %. 2.2. Triggering the OSWD without a catalysing dispersing agent Since all attempts to rework samples fabricated by the standard OSWD sample preparation method did not lead to sufficient Raman samples, it was hence thought upon to develop a new and adequate sample preparation technique that could supply and transfer the essential activation energy to trigger the OSWD in an alternative way, i.e. without the need of a catalysing dispersing agent. A way of doing so, as per literature, could be by employing the concept that the gradient of surface free energy at the solid-solid interface changes with an increase in temperature [50-51], thereby presenting the possibility of triggering the solid-solid wetting effect via a thermal sample treatment [23-25]. However, for implication of such a treatment, the thermal stability and the melting point of the involved pigment has to be taken A. Eberle et al. / Carbon 00 (2017) 000–000 7 into account. Since QAC crystals are thermally stable up to their melting point of 390 °C [46,52], a series of tests were performed where the virgin HOPGs covered with pure γQAC powder were heated up gently to different temperatures, as to trigger the OSWD process thermally. Results in this respect revealed, though generation of no supramolecular QAC structures for temperatures up to 160 °C, however, the detection of a significant surface coverage of 83 ± 13 % for samples being further heated up to 240 °C (refer Fig. 3). In addition, it was observed that the supramolecular QAC chains arranged themselves in a large-scale homogeneous monolayer, which changed its orientation almost exclusively by hitting the border to a new graphite plane. Such planes, in turn, are predetermined by the quality of the substrate surface. Further experiments were performed in this regard, where several HOPGs covered with γQAC powder were heated up to approx. 270 °C, yielding a greater surface coverage of 92 ± 6 %. Hence, it can be concluded that the above described thermally triggered sample preparation method marks as a promising approach towards fabricating samples, enabling accurate Raman spectroscopy measurements. Fig. 3. STM image of a HOPG substrate covered with a QAC monolayer. The sample was prepared by heating up dry γQAC powder on the HOPG substrate to a temperature of 240 °C; the surface coverage rate being 83 ± 13 % in this case. The close-up view in the bottom right corner highlights, how the QAC molecules arrange themselves within the adsorbate layer. In order to determine the underlying formation mechanism that results in the observed extended and well-ordered QAC adsorbate layers, worth recapitulating, initially, are the so far gained findings about the classical OSWD sample preparation. Summarizing briefly, when a three-dimensional semiconductor crystal contacts a HOPG substrate, molecules from the crystal A. Eberle et al. / Carbon 00 (2017) 000–000 8 detach and get attached to the substrate, provided the adsorption energy Ea is higher than the cohesive energy Ec of the semiconductor bulk (as derived from force field calculations in previous studies [19]). In this respect, having analysed HOPG samples via STM measurements over a period of several weeks (with semiconductor particles and 8CB being continuously present), no significant increase in the array dimensions could be detected over time. Further, except few results showing instable bilayer structures, no case of three-dimensional growth could be detected [20]. In this regard, an approximate energetic criterion was found to predict two-dimensional vs. three-dimensional growth under conditions of thermodynamic equilibrium [53-54]: the condition for the three-dimensional growth being Ec < Ea and the inverse being true for the two-dimensional growth. Hence, it can be said that for the HOPG and γQAC model system as used in the present study, growth of two-dimensional morphology is expected to be favoured, what corresponds to our findings. Generally speaking, the growth of a supramolecular surface adsorbate structures is proportional to the surface diffusion. Surface diffusion was observed for both single adsorbate molecules and compact adsorbate clusters containing numerous molecules [54], provided the diffusion barrier is overcome. Further, the surface diffusion process is thermally promoted, just as in the case of bulk diffusion, with diffusion rates (corresponding to the adsorbate mobility) increasing with increasing temperature. We can thus conclude from our experimental findings that the surface diffusion is limited in the temperature range of up to 160 °C, whereas significant diffusion is achieved for temperatures of 240 °C and above. Hence, provided the conditions for a thermally triggered OSWD prevail, the adsorbed QAC molecules migrate over the substrate surface in a direction away from the depositing γQAC crystal plane, with the concentration gradient and diffusion processes as the driving forces. As a result, further molecules can be deposited from the γQAC source, leading to an expansion of the QAC array. As the formed adsorbate layers show a high degree of order, an additional thermal annealing effect is presumed. In this respect, it is referred to experiments conducted by Wagner et al., analysing the thermal annealing of a two-dimensional surface covering formed by QAC arrays exhibiting different orientations, however using Ag(111) as the substrate material [46]. Their results revealed that for a temperature range between 550 – 570 K (i.e. 277 – 297 °C), the structural properties of the QAC covering change towards the formation of extended and well- ordered monolayers. Hence, in analogy with the above, it can be said that for our study the thermal annealing presumably plays a part, by triggering the rearrangement of QAC molecules and thus leading to the formation of extended and highly ordered monolayers. In addition, worth mentioning are the series of continuative experiments performed to test the stability of these A. Eberle et al. / Carbon 00 (2017) 000–000 9 thermally generated QAC adsorbate structures [20]. Our results yielded no signs of structural decomposition after the samples have been stored for 36 days under ambient conditions and furthermore, the QAC adsorbates were observed to be resistant towards humidity and direct water contact. Discussing the effect of sintering processes on the above, it can be said that first of all, the working temperatures (240 °C and 270 °C respectively) are way low than the melting temperature of γQAC (390 °C) [52]. Although early reports on lower temperature sintering observed with nanoscale particles conjectured a melting temperature reduction, this idea however has been dispelled by careful analysis, revealing further no new mechanism to be active in sintering nanoscale particles beyond known processes [55]. In this respect, as per the well-known viscous flow sintering model, the concept of sintering is analogous with the growth of sinter necks between contacting objects (i.e. grains in this context), connecting the contacting grains and forming a polycrystalline solid [55]. Hence, it can be said that a potential sintering would both interlink contacting semiconductor particles to strongly bonded crystalline structures and connect these structures to the substrate surface at the contact points. Further, in contrast to the OSWD process, the resulting sinter neck formation would be isotropic in nature, resulting in numerous contacting points, thereby establishing a permanent connection (besides, the type of chemical bonding between the substrate and the semiconductor is supposed to be π– π stacking). Thus, small-scale nanocrystals fixed to the substrate could be detected directly via STM, whereas the presence of large, permanently fixed nanocrystals would be noticed since they would considerably disturb the STM measurements, thereby making STM imaging hardly possible. However, STM measurements revealed no detection of sintered γQAC crystals of any kind, thereby highly limiting the influence of sintering processes on the above thermally triggered sample treatment approach. 2.3. Replacing QAC by DMQAC Successfully applied to a HOPG substrate, the newly developed, thermally triggered sample preparation method generated a surface covering that synced with the requirements of a Raman spectroscopic analysis. Testing the applicability of OSWD for single graphene layer on a copper foil as the substrate material (refer Fig. 1) revealed similar supramolecular structures as detected on a HOPG substrate [20]. Hence, the single QAC molecules arranged themselves in one- dimensional chain-like structures, leading to the coverage of the substrate surface by multiple parallel and side-by-side appearing chain-like formations. Further, the lattice parameters 'a' and 'b' of the supramolecular monolayer (compare Fig. 1) were observed to correspond to one of A. Eberle et al. / Carbon 00 (2017) 000–000 10 the array configurations as observed on the HOPG substrates, such an HOPG array configuration being termed as the 'relaxed QAC chain configuration' in this case [20]. Next, the initial Raman measurements of samples fabricated via thermally triggered OSWD, applied to graphene as the substrate material, revealed that due to the sole presence of sp2-bonded carbon atoms in both the γQAC and the graphene substrates, the location, the shape, and the intensity of the corresponding G peaks (described in the next sub-sections) was found to be quite similar for both the materials. Consequently, the spectra of both the samples could not be distinguished accurately, making hence the further, exact analysis quite a challenging task. Thus, to resolve the above, it was decided to replace γQAC by the quinacridone derivate dimethylquinacridone (DMQAC). In this regard, the linear DMQAC molecules (refer Fig. 4) generate three-dimensional crystal structures, iso-structural to the αI polymorph formed by the QAC molecules [47-49]. In this respect, to begin with, different samples were analyzed via STM measurements, to explore the processability of DMQAC via the OSWD technique. Results revealed, analogous to QAC, the DMQAC molecules arranging themselves in one-dimensional supramolecular chains, forming in turn two-dimensional arrays. Further, within the limits of accuracy, the lattice parameters of these supramolecular structures were found to be identical for both the substrates HOPG and single layer graphene (refer Fig. 5). Besides, the latter substrate depict the distinct honeycomb structure of single layer graphene [26] (refer Fig. 5 (b)). However, further analysis revealed that in contrast to the QAC adsorbate structures (refer Fig. 3) [20], the DMQAC chains arrange themselves solely in a close-packing chain configuration (on both the HOPG and the single layer graphene), leading to DMQAC arrays with high packing density (refer Fig. 5 and Fig. 6). In addition, regarding the adsorbate layer thickness, most of the observed DMQAC arrays were clearly determined to be two-dimensional monolayers (refer Fig. 5 and Fig. 9 (b)). However, previous studies using DMQAC revealed that, although rarely observed, bilayer and even trilayer structures could as well be detected, with their structure being similar to the ones seen when analysing QAC adsorbates [20]. Further, these structures were found to range in the size of single DMQAC chains up to arrays of a few dozen nanometres in diameter. Hence, for clarity purposes, the term 'DMQAC adsorbate layer' will be used hereinafter. Nevertheless, it should be noted that since the Raman spectroscopy is an averaging technique, the related analysis of potential doping effects is not disturbed by sporadically occurring small-sized bilayer or trilayer adsorbates. Furthermore, as extended multilayer adsorbates would generate fluorescence effects within the Raman signal, their occurrence could be detected, however, A. Eberle et al. / Carbon 00 (2017) 000–000 11 Raman analysis revealed no detection of multilayer DMQAC adsorbates (refer to the Raman discussion below). Results revealed that the thermally triggered sample preparation method yielded an overall surface coverage of 92 ± 8 % atop a HOPG substrate, with the DMQAC adsorbate layer found to be sufficiently homogeneous in nature (refer Fig. 6). However, worth noting here is that the structure of the copper foil leads to a rather uneven surface, making hence large-scale STM scans of the covered graphene samples impossible. Thus, we have not been able to determine the surface coverage of the graphene samples accurately, though, the promising results of the HOPG samples and the explored similar adsorbate structure properties on both the substrates indicated similar coverage rates. In addition, the STM analysis of graphene samples over a period of four weeks detected negligible decomposition of the DMQAC adsorbate structures, thus indicating the temporal stability and the resistance against humidity of the latter adsorbate layers being similar to that of QAC adsorbates [20]. Fig. 4. Chemical structure of the DMQAC molecule. (a) (b) Fig. 5. STM images of supramolecular DMQAC structures (upper sections) and the subjacent substrates (l ower section). The STM pictures have been equalized using the substrates unit cell parameters for calibration. (a) The substrate in use is HOPG and the lattice parameters of the A. Eberle et al. / Carbon 00 (2017) 000–000 12 adsorbate structures were found to be as: a = 0.68 ± 0.02 nm, b = 1.72 ± 0.02 nm, and an intermediate angle of 87 ± 2°. The HOPG unit cell is depicted via the parameters g1 and g2, having lengths as: g1 = g2 = 0.246 nm. (b) Using single layer graphene on a copper foil as the substrate, the lattice parameters of the adsorbates were determined to be as: a = 0.67 ± 0.02 nm, b = 1.72 ± 0.02 nm, and an intermediate angle of 88 ± 2°. Further, the marked hexagon in the image represents one carbon ring of the graphene structure; the atom to atom distance in this regard being 0.142 nm. Fig. 6. Example STM picture of a two-dimensional supramolecular adsorbate layer atop a HOPG substrate generated by DMQAC molecules, yielding an overall average coverage rate of 92 ± 8 %. Besides, the close-up inset in the bottom right corner highlights, how the DMQAC molecules arrange themselves within the adsorbate layer. 2.4. Raman spectroscopy measurements Generally speaking, the Raman spectrum of carbon-based substrates can be mainly characterised by three characteristic peaks [6,34-39], i.e. the D peak, the G peak, and the 2D peak. The D peak is typically observed at a Raman frequency of approx. 1350 cm-1 indicating a structural defect, owing to its activation due to A1g mode breathing vibrations of six- membered sp2 carbon rings, which are absent in defect-free graphene [34-35]. Hence, the D peak intensity increases with the amount of disorder present in the material [36]. The G peak, on the other hand, appears at approx. 1580 cm-1, being associated with the doubly degenerate E2g phonon at the Brillouin-zone centre [34-35]. Finally, the 2D peak is the second order of the D peak, found usually at about 2680 cm-1. Further, since the 2D peak originates from a process A. Eberle et al. / Carbon 00 (2017) 000–000 13 where momentum conservation is satisfied by two phonons with opposite wave vectors, it is always present for graphene, with its activation requiring no structural defects [34-36]. Moreover, variations of the above characteristic Raman peaks can be generated by the introduction of either mechanical strain or by chemical doping. However, both these sources cause specific changes in the Raman spectrum, making hence the respective variations distinguishable from each other [35-36]. Mechanical strain, for example caused by an adsorbate layer or by a previous thermal treatment of the sample, modifies the crystal phonons due to changes in the lattice constants and the resulting structural disorder activates the D peak [36]. It was further found that the compressive strain leads to an upshift of the G and the 2D peak, the tensile stress whereas, leading to the downshift of both these peaks. However, in either case, the 2D peak shift is several times greater, with the intensity ratio of the 2D to G peak (I2D / IG) remaining unaltered [35-36]. Nevertheless, in contrast to the above, the intensity ratio I2D / IG has been observed to be sensitive to chemical doping. Appropriate doping effects cause the above ratio to decrease monotonically with an increase in both the electron and the hole concentration [6,36-38]. Also, as per empirical findings, doping with electron-donating aromatic molecules (i.e. electron- or n-type doping) downshifts the G peak frequency, whereas the presence of electron-withdrawing molecules (i.e. hole- or p-type doping) leads to a G peak upshift. Nonetheless, the 2D frequency is reported to be upshifted, irrespective of the type of doping [37-38]. The results of the Raman test series are as presented in the Fig. 7. To begin with, the depicted Raman spectra were determined by averaging nine measurements, both for the pure graphene and the 'DMQAC powder on a graphene substrate' samples, and by averaging 16 measurements for the 'graphene covered with a DMQAC adsorbate layer' sample. Results revealed detection of no graphene-specific peaks at the appropriate peak locations (Fig. 7 (a)) within the spectrum of the DMQAC powder on a graphene substrate (treated with the identical, thermally triggered sample preparation method). However, a few DMQAC-specific peaks could be observed which could be determined precisely in the spectra of both the DMQAC samples (DMQAC powder and DMQAC adsorbate layer on graphene), their locations being determined as 1204 ± 2 cm-1, 1233 ± 2 cm-1, 1312 ± 1 cm-1, and 1567 ± 2 cm-1, respectively. Also, further analysis in this regard revealed no shift of the DMQAC-specific peaks in the spectra of both the DMQAC samples. Furthermore, worth noting here is that the spectrum of the sample 'graphene covered with a DMQAC adsorbate layer' exhibited additional peaks, which are as well related to DMQAC [56], their determined locations being as 1408 ± 1 cm-1, 1509 ± 1 cm-1, and 1648 ± 2 cm-1, respectively. Nevertheless, the precise location of these A. Eberle et al. / Carbon 00 (2017) 000–000 14 peaks could not be determined in the spectrum of the sample 'graphene covered with DMQAC powder', due to the occurrence of fluorescence (Fig. 7 (b)) [57-58]. In this regard, the revealed finding that fluorescence effects are quenched by the properties and conditions of the sample 'graphene covered with a DMQAC adsorbate layer', indicate though a structural transition of the three-dimensional DMQAC particles into a thin adsorbate layer (being triggered by the OSWD process), thereby acting no longer as a bulk solid [59-60]. This deduction affirms the findings of the previously presented STM measurements. Further investigation revealed that the spectra of the pure graphene substrate and that of the graphene sample covered with a two-dimensional DMQAC adsorbate layer comprises the graphene-specific peaks, i.e. the D, the G, and the 2D peaks (Fig. 7 (a)). The location of the D peak was found to be similar for both the samples (Fig. 7 (b)), however, owing to its significantly lower intensity in contrast to the other peaks, it being determined only at around 1348 ± 5 cm-1 (including twice the standard deviation). In addition, the D peak intensity was as well found to be similar for both the samples, thereby indicating the exclusion of thermally induced structural defects. Furthermore, for the pure graphene substrate, the G peak was found at 1592 ± 1 cm-1 and the 2D peak at 2691 ± 1 cm-1, respectively, whereas for the 'graphene plus DMQAC adsorbate layer' sample, their locations being 1595 ± 2 cm-1 and 2701 ± 3 cm-1, respectively (Fig. 7 (c) and (d)). Thus, the Raman spectrum of the DMQAC adsorbate layer sample revealed an upshift of both the G peak frequency (3 ± 2 cm-1) and the 2D peak frequency (10 ± 3 cm-1), indicating thereby chemical doping with electron-withdrawing aromatic molecules (i.e. p-type doping). The latter is further supported by literature findings, showing that DMQAC thin films act only as p-type materials [30,33]. Comparison of the Raman intensity ratio I2D / IG of the pure graphene substrate (I2D / IG = 1.15 ± 0.06) and the graphene substrate covered with a DMQAC adsorbate layer (I2D / IG = 0.53 ± 0.06) further revealed a significant decrease, indicating hence the chemical doping of graphene as well. In addition, mechanical strain was excluded as the probable source of the Raman peak shifts, since no increase of the D peak intensity was found after the thermally triggered sample preparation and due to the decrease in the Raman intensity ratio I2D / IG. A. Eberle et al. / Carbon 00 (2017) 000–000 Raman intensity [a.u.] Raman spectroscopy (a) Raman intensity [a.u.] DMQAC specific peaks 15 (b) D peak G peak 2D peak D peak G peak 1100 1300 1500 1700 1900 2100 2300 2500 2700 1100 1200 1300 1400 1500 1600 1700 Graphene + DMQAC Graphene DMQAC powder Raman Shift [cm-1] Graphene + DMQAC Graphene DMQAC powder Raman Shift [cm-1] Raman intensity [a.u.] G peak shift (c) Raman intensity [a.u.] 2D peak shift (d) DMQAC peaks 1540 1550 1560 1570 1580 1590 1600 1610 1620 2620 2640 2660 2680 2700 2720 2740 2760 Graphene + DMQAC Graphene DMQAC powder Raman Shift [cm-1] Graphene + DMQAC Graphene DMQAC powder Raman Shift [cm-1] Fig. 7. Data representing the averaged results of the performed Raman measurements. (a) Overview of the Raman spectra in the relevant frequency range between 1100 and 2800 cm-1. (b) Zoomed-in view of the frequency range between 1100 and 1700 cm-1. The Raman spectra of both the DMQAC powder and the two-dimensional DMQAC adsorbate layer atop a graphene substrate depict additional peaks. Further, a close-up view of the frequency range revealing: (c) the G peak and (d) the 2D peak, respectively. 2.5. Scanning Tunneling Spectroscopy (STS) analysis of DMQAC on graphene In order to further verify the results of Raman spectroscopy by an independent experimental technique, an STS test series was conducted in addition. STS, determining the current-bias spectra I(V) at a fixed tip position, is known as a sensitive technique to probe the local electronic properties of a surface [26-32]. At low tip–sample voltages, the tunneling differential conductance is proportional to the local density of states of conducting and semiconducting samples [26-29]. However, due to the dependence of sample-tip separation on the tunneling probability, the STS acquisition relies on the initial set point tunneling conditions [28]. Thus, the tunneling distance for the below discussed STS spectra was adjusted with identical tunneling parameters, whenever possible: bias = 50.1 mV, and tunnel current = 1 nA for the analysis of graphene, and bias = 1.5 V, and tunnel current = 501 pA for the analysis of both A. Eberle et al. / Carbon 00 (2017) 000–000 16 DMQAC arrays and the test series regarding potential chemical doping. Further, each spectrum was acquired within 100 ms. Appropriate samples were investigated via STS measurements, in order to explore the surface electronic structure of graphene substrates with DMQAC adsorbates atop. To begin with, STS spectra taken of pure graphene are as shown in the Fig. 8 (a). As can be seen, the curve progressions are in agreement with the reported metallic behaviour of the zero-gap semiconductor graphene [27]. Further, the spectra taken of DMQAC arrays feature a sample bias range with approximately zero current (refer Fig. 8 (b)), as expected of a semiconducting material [27]. However, the curves are subject to strong fluctuations that are related to the ambient measurement conditions, causing thermal fluctuations that affect the STS measurement accuracy [27]. Thereby, owing to the above, the direct determination of the tunneling differential conductance was hardly possible. Instead, a trend line of the type f(x) = a (x + b)³ (with 'a' and 'b' as constants) was fitted to the obtained spectra and differentiation yielded suitable parabolic shaped dI/dV curves (refer Fig. 8 (b) and Fig. 9 (a)). Assuming a differential conductance below 0.7 nA to be zero, the bandgap of DMQAC was estimated to be 2.4 ± 0.2 eV (including twice the standard deviation), the result hence being in good accordance with the reported HOMO – LUMO gap of DMQAC of 2.3 eV [30]. Regarding the analysis of a potential chemical doping, it was reported that a suitable semiconducting surface adsorbate modifies the electronic properties of the substrate beyond the physical borders of the adsorbate. Hence, it was found that the STS measurements yield a decreasing bandgap in the direction away from the chemical dopant [31-32]. The results of the related test series are shown in the Fig. 9. As can be seen, the determined band gap decreases almost linearly in the direction away from the DMQAC array, reaching the detection limit of 0.1 ± 0.2 eV at a distance of 8 nm. The latter result thus signalizes the chemical doping of graphene via supramolecular DMQAC adsorbates, generated via the OSWD. (a) STS - Graphene Current [nA] 100 80 60 40 20 0 -2 -1.5 -1 -0.5 0 0.5 1 -20 -40 -60 -80 -100 A. Eberle et al. / Carbon 00 (2017) 000–000 17 (b) STS - DMQAC array -2 -1.5 -1 -0.5 1.5 2 Sample bias [V] Current [nA] dI/dV [nA/V] 2.4 ± 0.2 eV 0 0.5 1 1.5 2 Sample bias [V] 5 4 3 2 1 0 -1 -2 -3 -4 -5 Current Trend line dI/dV Fig. 8. STS measurements at different locations of three different (but equally prepared) samples. (a) 10 STS measurements of pure graphene. (b) 18 STS measurements of supramolecular DMQAC adsorbates. Further, the indicated trend line and the dedicated dI/dV curve exemplarly reveal the determination of a bandgap (for further details, refer to the explanations in the text); the corresponding bandgap of a DMQAC monolayer being found as 2.4 ± 0.2 eV (including twice the standard deviation). (a) Scanning tunneling spectroscopy Current dI/dV [a.u.] Bandgap: 0.1 eV Distance: 8 nm (b) 0.2 eV 0.8 eV 1.5 eV 5 nm 3 nm 2 nm 1.6 eV 1 nm 2.4 eV DMQAC array -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 Sample bias [V] Fig. 9. (a) Example STS measurements, taken at different distances from a DMQAC array (including the calculated trend lines and the dI/dV curves). With the determined bandgaps indicated above the related spectrum, twice the standard deviation was deduced to be ± 0.2 eV. A. Eberle et al. / Carbon 00 (2017) 000–000 18 (b) STM image of the region, at which the STS measurements were taken. The upper section depicts a DMQAC monolayer and the lower section shows the bare graphene substrate. The measurement points were located on the highlighted white line. 3. Conclusions Our approach to thermally trigger the solid-solid wetting of crystalline carbon surfaces by organic semiconductor particles enables an easily applicable technique to achieve monolayers with high surface coverage rates and uniform adsorbate structures. Moreover, graphene samples generated in this way allows for the analysis of possible doping effects via techniques as Raman spectroscopy. Additionally, by using the commercially available pigment DMQAC for the new OSWD approach, clear spectral evidence of chemical doping effects of graphene could be obtained. This finding is further supported by STS analysis, showing evidence of chemical doping by DMQA adsorbate structures. The results hence bring forward new and straightforward to perform approaches for the fabrication and bandgap engineering of low-cost, but large-scale products based on pigment-functionalized graphene, like the printed and potentially flexible carbon based electronics [9]. Acknowledgements The authors would like to thank Michael Blum for his kind support with the STM measurements using graphene as the substrate material. Heartfelt acknowledgment as well to Michaela Wenner for her assistance in exploring the potentials of sample reworking. We also would like to extend our gratitude to Arthur Wesemann for his contribution with experiments and valuable experiences, in regard to the imaging of organic semiconductor adsorbates via the NaioSTM setup (as a precondition for our STS studies). Also, we would like to thank Neeti Phatak for her support with the proofreading and editing of this publication. The Bayerisches Staatsministerium für Umwelt und Verbraucherschutz is gratefully acknowledged for their funding. Appendix A. Supplementary data Supplementary data related to this article can be found at xxx. References A. Eberle et al. / Carbon 00 (2017) 000–000 19 [1] S. Thompson, S. Parthasarathy, Moore's law: the future of Si microelectronics, Mater. Today. 9 (2006) 20–25. doi:10.1016/S1369-7021(06)71539-5. [2] P. Avouris, Z. Chen, V. Perebeinos, Carbon-based electronics, Nat. Nanotechnol. 2 (2007) 605–15. doi:10.1038/nnano.2007.300. [3] W. Lu, C.M. Lieber, Nanoelectronics from the bottom up., Nat. Mater. 6 (2007) 841– 50. doi:10.1038/nmat2028. [4] H.-H. Chen, S.H. Su, S.-L. Chang, B.-Y. Cheng, S.W. Chen, H.-Y. Chen et al., Tailoring low-dimensional structures of bismuth on monolayer epitaxial graphene, Sci. Rep. 5 (2015) 11623. doi:10.1038/srep11623. [5] S. Tans, A. Verschueren, C. Dekker, Room-temperature transistor based on a single carbon nanotube, Nature. 672 (1998) 669–672. doi:10.1038/29954. [6] H. Medina, Y.C. Lin, D. Obergfell, P.W. Chiu, Tuning of charge densities in graphene by molecule doping, Adv. Funct. Mater. 21 (2011) 2687–2692. doi:10.1002/adfm.201100401. [7] A.K. Manna, S.K. Pati, Tuning the electronic structure of graphene by molecular charge transfer: A computational study, Chem. - An Asian J. 4 (2009) 855–860. doi:10.1002/asia.200800486. [8] B. Long, M. Manning, M. Burke, B.N. Szafranek, G. Visimberga, D. Thompson et al., Non-covalent functionalization of graphene using self-assembly of alkane- amines, Adv. Funct. Mater. 22 (2012) 717–725. doi:10.1002/adfm.201101956. [9] S. Park, M. Vosguerichian, Z. Bao, A review of fabrication and applications of carbon (2013) 1727–52. flexible electronics., Nanoscale. 5 nanotube film-based doi:10.1039/c3nr33560g. [10] N. Liu, Y. Zhou, N. Ai, C. Luo, J. Peng, J. Wang et al., High-performance, all- solution-processed organic nanowire transistor arrays with inkjet-printing patterned electrodes, Langmuir. 27 (2011) 14710–14715. doi:10.1021/la2033324. [11] N. Liu, Y. Zhou, L. Wang, J. Peng, J. Wang, J. Pei et al., In situ Growing and patterning of aligned organic nanowire arrays via dip coating, Langmuir. 25 (2009) 665–671. doi:10.1021/la8036633. [12] J.Y. Zheng, Y. Yan, X. Wang, Y.S. Zhao, J. Huang, J. Yao, Wire-on-wire growth of fluorescent organic heterojunctions, J. Am. Chem. Soc. 134 (2012) 2880–2883. doi:10.1021/ja209815f. A. Eberle et al. / Carbon 00 (2017) 000–000 20 [13] G. Yu, C.M. Lieber, Assembly and integration of semiconductor nanowires for functional nanosystems, Pure Appl. Chem. 82 (2010) 2295–2314. doi:10.1351/PAC- CON-10-07-06. [14] B.J. Jordan, Y. Ofir, D. Patra, S.T. Caldwell, A. Kennedy, S. Joubanian et al., Controlled self-assembly of organic nanowires and platelets using dipolar and 2074–2078. hydrogen-bonding interactions, Small. 4 (2008) doi:10.1002/smll.200800811. [15] M. Liu, Z. Ji, L. Shang, Top-down Fabrication of Nanostructures, in: L. Chi editor, Nanotechnology, Vol. 8: Nanostructured Surfaces, WILEY-VCH, Weinheim, 2010, pp. 17-35. [16] D. Zhong, H. Zhang, L. Chi, Surface-Supported Nanostructures Directed by Atomic- and Molecular-Level Templates, in: L. Chi editor, Nanotechnology, Vol. 8: Nanostructured Surfaces, WILEY-VCH, Weinheim, 2010, pp. 357-369. [17] S. Kowarik, A. Gerlach, F. Schreiber, Organic molecular beam deposition: fundamentals, growth dynamics, and in situ studies, J. Phys. Condens. Matter. 20 (2008) 184005. doi:10.1088/0953-8984/20/18/184005. [18] M. Treier, E. Orgiu, L. Zalewski, D. Cho, R. Rieger, K. Müllen et al., Solid–solid transfer of organic semiconductors for field-effect transistor fabrication, J. Mater. Chem. 20 (2010) 9018. doi:10.1039/c0jm02173c. [19] F. Trixler, T. Markert, M. Lackinger, F. Jamitzky, W.M. Heckl, Supramolecular Self- Assembly Initiated by Solid–Solid Wetting, Chem. - A Eur. J. 13 (2007) 7785–7790. doi:10.1002/chem.200700529. [20] A. Eberle, A. Nosek, J. Büttner, T. Markert, F. Trixler, Growing low-dimensional supramolecular crystals directly from 3D particles, CrystEngComm. 19 (2017) 1417- 1426. doi:10.1039/C6CE02348G. [21] F. Trixler, W.M. Heckl, Various approaches to control solid/solid wetting self- assembly of organic semiconductors with STM, in: A. Méndez-Vilas, J. Diaz editors, Modern Research and Educational Topics in Microscopy, Vol. 2, Formatex, Badajoz, 2007, pp. 534-541. [22] J.A.A.W. Elemans, S. Lei, S. De Feyter, Molecular and Supramolecular Networks on Surfaces: From Two-Dimensional Crystal Engineering to Reactivity, Angew. Chemie Int. Ed. 48 (2009) 7298–7332. doi:10.1002/anie.200806339. [23] J. Haber, T. Machej, T. Czeppe, The phenomenon of wetting at solid/solid interface, Surf. Sci. 151 (1985) 301–310. doi:10.1016/0039-6028(85)90468-6. A. Eberle et al. / Carbon 00 (2017) 000–000 21 [24] J. Leyrer, R. Margraf, E. Taglauer, H. Knözinger, Solid-solid wetting and formation of monolayers in supported oxide systems, Surf. Sci. 201 (1988) 603–623. doi:10.1016/0039-6028(88)90505-5. [25] F. Bertinchamps, C. Grégoire, E.M. Gaigneaux, Systematic investigation of supported transition metal oxide based formulations for the catalytic oxidative elimination of (chloro)-aromatics, Appl. Catal. B Environ. 66 (2006) 1–9. doi:10.1016/j.apcatb.2006.02.011. [26] G. Li, A. Luican, E.Y. Andrei, Scanning tunneling spectroscopy of graphene on graphite, Phys. Rev. Lett. 102 (2009) 1–4. doi:10.1103/PhysRevLett.102.176804. [27] A. Castellanos-Gomez, M. Wojtaszek, Arramel, N. Tombros, B.J. Van Wees, Reversible hydrogenation and bandgap opening of graphene and graphite surfaces tunneling spectroscopy, Small. 8 (2012) 1607–1613. probed by scanning doi:10.1002/smll.201101908. [28] R.J. Hamers, Atomic-Resolution Surface Spectroscopy with the Scanning Tunneling 531–559. Microscope, (1989) Chem. 40 Annu. Rev. Phys. doi:10.1146/annurev.pc.40.100189.002531. [29] F.W. Herbert, A. Krishnamoorthy, K.J. Van Vliet, B. Yildiz, Quantification of electronic band gap and surface states on FeS2(100), Surf. Sci. 618 (2013) 53–61. doi:10.1016/j.susc.2013.08.014. [30] S.-J. Lim, D.-S. Leem, K.-B. Park, K.-S. Kim, S. Sul, K. Na et al., Organic-on-silicon complementary metal-oxide-semiconductor colour image sensors., Sci. Rep. 5 (2015) 7708. doi:10.1038/srep07708. [31] P. Lauffer, K. V. Emtsev, R. Graupner, T. Seyller, L. Ley, Molecular and electronic structure of PTCDA on bilayer graphene on SiC(0001) studied with scanning 2064–2067. tunneling microscopy, Phys. Status Solidi. (2008) 245 doi:10.1002/pssb.200879615. [32] L. Zhao, R. He, K.T. Rim, T. Schiros, K.S. Kim, H. Zhou et al., Visualizing Individual Nitrogen Dopants in Monolayer Graphene, Science (80-. ). 333 (2011) 999–1003. doi:10.1126/science.1208759. [33] M. Hiramoto, S. Kawase, M. Yokoyama, Photoinduced hole injection multiplication in p-type quinacridone pigment films, Japanese J. Appl. Physics, Part 2 Lett. 35 (1996). doi:10.1143/JJAP.35.L349. A. Eberle et al. / Carbon 00 (2017) 000–000 22 [34] M.W. Iqbal, A.K. Singh, M.Z. Iqbal, J. Eom, Raman fingerprint of doping due to metal adsorbates on graphene, J. Phys. Condens. Matter. 24 (2012) 335301. doi:10.1088/0953-8984/24/33/335301. [35] W.X. Wang, S.H. Liang, T. Yu, D.H. Li, Y.B. Li, X.F. Han, The study of interaction between graphene and metals by Raman spectroscopy, J. Appl. Phys. 109 (2011) 15– 18. doi:10.1063/1.3536670. [36] C. Casiraghi, Probing disorder and charged impurities in graphene by Raman spectroscopy, Phys. Status Solidi - Rapid Res. Lett. 3 (2009) 175–177. doi:10.1002/pssr.200903135. [37] X. Dong, D. Fu, W. Fang, Y. Shi, P. Chen, L.J. Li, Doping single-layer graphene with aromatic molecules, Small. 5 (2009) 1422–1426. doi:10.1002/smll.200801711. [38] H. Liu, Y. Liu, D. Zhu, Chemical doping of graphene, J. Mater. Chem. 21 (2011) 3335. doi:10.1039/c0jm02922j. [39] Z. Ni, Y. Wang, T. Yu, Z. Shen, Raman Spectroscopy and Imaging of Graphene, Nano Res. 1 (2008) 273–291. doi:10.1007/s12274-008-8036-1. [40] E.D. Głowacki, R.R. Tangorra, H. Coskun, D. Farka, A. Operamolla, Y. Kanbur et al., Bioconjugation of hydrogen-bonded organic semiconductors with functional proteins, J. Mater. Chem. C. (2015) 6554–6564. doi:10.1039/C5TC00556F. [41] E.D. Głowacki, G. Romanazzi, C. Yumusak, H. Coskun, U. Monkowius, G. Voss et al., Epindolidiones-Versatile and Stable Hydrogen-Bonded Pigments for Organic Field-Effect Transistors and Light-Emitting Diodes, Adv. Funct. Mater. (2014) 1–12. doi:10.1002/adfm.201402539. [42] M. Sytnyk, E. D. Głowacki, S. Yakunin, G. Voss, W. Schöfberger, D. Kriegner et al., Hydrogen-Bonded Organic Semiconductor Micro- And Nanocrystals: From Colloidal Syntheses to (Opto-)Electronic Devices, J. Am. Chem. Soc. 136 (2014) 16522–16532. [43] E.D. Głowacki, M. Irimia-Vladu, M. Kaltenbrunner, J. Gsiorowski, M.S. White, U. Monkowius et al., Hydrogen-bonded semiconducting pigments for air-stable field- effect transistors, Adv. Mater. 25 (2013) 1563–1569. doi:10.1002/adma.201204039. [44] E.D. Glowacki, L. Leonat, G. Voss, M. Bodea, Z. Bozkurt, M. Irimia-Vladu et al., Natural and nature-inspired semiconductors for organic electronics, Org. Semicond. Sensors Bioelectron. IV. 8118 (2011) 81180M–81180M–10. doi:10.1117/12.892467. A. Eberle et al. / Carbon 00 (2017) 000–000 23 [45] E.D. Głowacki, M. Irimia-Vladu, S. Bauer, N.S. Sariciftci, Hydrogen-bonds in molecular solids – from biological systems to organic electronics, J. Mater. Chem. B. 1 (2013) 3742. doi:10.1039/c3tb20193g. [46] T. Wagner, M. Györök, D. Huber, P. Zeppenfeld, E.D. Głowacki, Quinacridone on Ag(111): Hydrogen bonding versus chirality, J. Phys. Chem. C. 118 (2014) 10911– 10920. doi:10.1021/jp502148x. [47] E.F. Paulus, F.J.J. Leusen, M.U. Schmidt, Crystal structures of Quinacridones, CrystEngComm. 9 (2007) 131–143. doi:10.1039/b613059c. [48] G. Lincke, On quinacridones and their supramolecular mesomerism within the crystal lattice, Dye. Pigment. 52 (2002) 169–181. doi:10.1016/S0143-7208(01)00085-7. [49] A.D. Squires, R.A. Lewis, A.J. Zaczek, T.M. Korter, Distinguishing Quinacridone Pigments via Terahertz Spectroscopy: Absorption Experiments and Solid-State Density Functional Theory Simulations, J. Phys. Chem. A. 121 (2017) 3423–3429. doi:10.1021/acs.jpca.7b01582. [50] Q. Zhao, Y. Liu, E.W. Abel, Effect of temperature on the surface free energy of amorphous carbon films, J. Colloid Interface Sci. 280 (2004) 174–183. doi:10.1016/j.jcis.2004.07.004. [51] S.K. Rhee, Critical Surface Energies of Al203 and Graphite, J. Am. Ceram. Soc. 55 (1972) 300–303. doi:10.1111/j.1151-2916.1972.tb11289.x. [52] D. R. Lide, CRC Handbook of Chemistry and Physics, CRC Press, Boca Raton, Internet Version 2005, pp. 243. [53] D. McDougall, H. Hattab, M.T. Hershberger, M. Hupalo, M. Horn von Hoegen, P.A. Thiel et al., Dy uniform film morphologies on graphene studied with SPA-LEED and STM, Carbon 108 (2016) 283–290. doi:10.1016/j.carbon.2016.06.083. [54] D. Appy, H. Lei, C.-Z. Wang, M.C. Tringides, D.-J. Liu, J.W. Evans et al., Transition metals on the (0001) surface of graphite: Fundamental aspects of adsorption, 219–238. and morphology, Prog. Surf. Sci. diffusion, 89 (2014) doi:10.1016/j.progsurf.2014.08.001. [55] R. M. German, Sintering: From empirical observations to scientific principles, Butterworth-Heinemann, Oxford, 2014, pp. 71-82 & 413-421 [56] E. del Puerto, C. Domingo, J. V. Garcia Ramos, S. Sanchez-Cortes, Adsorption Study and Detection of the High Performance Organic Pigments Quinacridone and 2,9- Dimethylquinacridone on Ag Nanoparticles By Surface-Enhanced Optical Spectroscopy, Langmuir. 30 (2014) 753–761. doi:10.1021/la403625u. A. Eberle et al. / Carbon 00 (2017) 000–000 24 [57] J.A. Smith, R.M. West, M. Allen, Acridones and quinacridones: Novel fluorophores (2004) 151–171. J. Fluoresc. 14 lifetime for fluorescence studies, doi:10.1023/B:JOFL.0000016287.56322.eb. [58] P.L. Gentili, C. Clementi, A. Romani, Ultraviolet-visible absorption and luminescence properties of quinacridone-barium sulfate solid mixtures., Appl. Spectrosc. 64 (2010) 923–9. doi:10.1366/000370210792080993. [59] M.R. Kagan, R.L. McCreery, Reduction of Fluorescence Interference in Raman- Spectroscopy Via Analyte Adsorption on Graphitic Carbon, Anal. Chem. 66 (1994) 4159–4165. doi:10.1021/ac00095a008. [60] L. Xie, X. Ling, Y. Fang, J. Zhang, Z. Liu, Graphene as a substrate to suppress fluorescence in resonance raman spectroscopy, J. Am. Chem. Soc. 131 (2009) 9890– 9891. doi:10.1021/ja9037593. Supporting Information Doping Graphene via Organic Solid-Solid Wetting Deposition Alexander Eberlea, Andrea Greinera, Natalia P. Ivlevab, Banupriya Arumugamc, Reinhard Niessnerb, Frank Trixlera,d, a Department für Geo- und Umweltwissenschaften and Center for NanoScience (CeNS), Ludwig-Maximilians-Universität München, Theresienstrasse 41, 80333 München, Germany b Institute of Hydrochemistry, Chair for Analytical Chemistry, Technische Universität München, Marchioninistr. 17, 81377, München, Germany c Institute for Nanoelectronics, Technische Universität München, Theresienstrasse 90, 80333 München, Germany d TUM School of Education, Technische Universität München and Deutsches Museum, Museumsinsel 1, 80538 München, Germany * Corresponding author. Tel.: +49 89 2179 509; fax: +49 89 2179 239. E-mail address: [email protected] (F. Trixler) 1 Materials and methods Sample preparation As a standard Organic Solid-Solid Wetting Deposition (OSWD) sample for scanning tunnelling microscope (STM) investigations, a dispersion with 2 wt% of the pigment γQAC (5,12-Dihydro-quino[2,3-b]acridine-7,14-dione, purchased as Hostaperm Red E5B02 from Clariant) dispersed in 4 ml of the dispersing agent 8CB (purchased as 4'-n- Octylbiphenyl-4-carbonitrile from Alfa Aesar, item no. 52709-84-9) was prepared. A few drops of this dispersion were then dispensed on a highly ordered pyrolytic graphite (HOPG, supplier NT-MDT, item no. GRBS/1.0), to trigger the OSWD at the interface between the dispersed pigment particles and the HOPG. Besides, single layer graphene on a copper foil (suppliers: Graphene Laboratories, item no. CVD-Cu-2X2, and Graphenea Inc.) was used as the substrate material for further tests. Also, as an alternative sample preparation method (to thermally trigger the OSWD), the HOPG substrate was fully covered with the powdered pigment, but without a catalysing dispersing agent. The covered substrate was then heated up to 240 °C and 270 °C, respectively, using a special hotplate, enabling an accurate temperature control and providing a smooth temperature increase (Stuart SD160, temperature accuracy ± 1.0 °C). In any case, the ready-made STM samples were investigated within days, and as per the previous tests, QAC arrays were observed unaltered in their structure for a minimum of four weeks, unless not influenced via any external forces [1]. Further, for the Raman experiments, single layer graphene on Si/SiO2 (purchased from Graphene Laboratories, item no. 1ML-SIO2-5P) was used as the substrate material and DMQAC (2,9-Dimethyl-5,12-dihydro-quino[2,3-b]acridine-7,14-dione, purchased as Hostaperm Pink E from Clariant) as the organic semiconductor. In order to induce thermally triggered OSWD on the substrate graphene/Si/SiO2, a small amount of powdered DMQAC, enough to cover the substrate surface, was added on top of the substrate. The substrate (along with the powdered organic semiconductor atop) was then placed on the previously mentioned hot plate, heating the sample up to approx. 270 °C. Once the sample was heated well for the given temperature, it was taken from the hot plate and the pigment powder was immediately removed from the substrate surface by mechanical shaking. Being still hot, the pigment powder does not adsorb humidity from the surrounding environment and thus does not stick to the substrate surface. Hence, the 2 appropriately prepared Raman samples in the above manner enabled an accurate Raman spectroscopy analysis, being free of measurements artefacts related to traces of bulk pigment particles. Moreover, before performing any Raman measurements, the Raman sample was chilled to room temperature under ambient conditions. STM and STS settings Two types of STM systems operating under ambient conditions were used for this study. First was a home-built STM combined with a SPM 100 control system, supplied by RHK Technology Inc.; the scans settings being: bias = 1 V, tunnel current = 300 pA, and the line time = 50 ms. Secondly, a commercial STM, type NaioSTM, supplied by Nanosurf GmbH, was employed for the measurements, as depicted by fig. 5, 8 and 9 within the main article; the scans settings for imaging DMQAC being: bias = 1.5 V, current = 501 pA, and line time = 80 ms, and for imaging graphene being: bias = 50.1 mV, tunnel current = 1 nA, and line time = 60 ms. In addition, the voltage pulses used to improve the scan quality were set in the range between 4.3 and 10 V. The STS measurements were performed using the NaioSTM. The tunneling distance was adjusted with tunneling parameters being similar to the ones as mentioned above. Each spectrum was acquired within 100 ms; both the STM and the STS measurements being performed under ambient conditions. All STS measurements were performed at randomly distributed positions and from 3 different (but equally prepared) samples, to exclude incidental findings. Further, the graphene samples used for the STS measurements were prepared via the standard OSWD sample preparation method, thereby using a dispersion of DMQAC and 8CB. This preparation method was preferred owing to two reasons: First, STM and STS measurements require a conductive sample, thus excluding graphene on SiO2/Si and making the use of single layer graphene on copper foil mandatory. However, according to the specifications of the manufacturer Graphenea Inc., single layer graphene on the copper foil is only thermally stable up to 60 °C. Tests further in this regard, applying the thermally triggered OSWD sample preparation method to single layer graphene on a copper foil, revealed significant damage of the substrate, thereby making accurate STM and STS measurements impossible for graphene/Cu samples prepared via thermally triggered OSWD. Secondly, OSWD induced by the dispersing agent 8CB enables to achieve STS under controlled ambient conditions without undefined contamination layers. STS experiments with pure 3 8CB on graphene/Cu yielded no evidence of chemical doping by 8CB adsorbate layers (refer example STS measurements in Fig. 1). STS - 8CB 150 Current [nA] -2 -1.5 -1 -0.5 100 50 0 0 -50 -100 -150 0.5 1 1.5 2 Sample bias [V] Fig. 1. Example STS measurements of 8CB adsorbate layers atop single layer graphene. Image processing For analysing the supramolecular adsorbate structures in the STM images, the software SPIP™ (Scanning Probe Image Processor, Version 2.3000; distributor: Image Metrology A/S) was used. Image distortions by the drift of a STM scan were analysed and corrected by applying two-dimensional Fast Fourier Transformation (FFT) to the images, done by using the known lattice parameters of the substrates as a reference. Autocorrelations of corrected images were employed for measuring the distances and angles in the substrates and the adsorbates. To minimize noise in the final STM images, a selective FFT filtering was applied with thresholding between 15-25 (min.) and 100 (max.). Determining the surface coverage To determine the coverage of the HOPG surface by the QAC arrays within a single STM picture, the software Gwyddion (64bit), version 2.42 was used. For this, initially, the QAC arrays via the tool "Mask Editor" were highlighted, followed by the export of single array dimensions by the "Grain distributions" tool, this finally being accompanied by the Microsoft Excel 2013 calculations to determine the coverage ratio. Further, to investigate the average coverage of a STM sample, per sample an area of about 0.7 µ m² 4 was analysed. This was done by using a number of STM pictures with high scan resolution and without measurement artefacts, the images were further randomly selected from at least five clearly separated positions on the covered substrate; the average coverage rates, including the double standard deviations, being specified in the current publication. Raman Spectroscopy The Raman experiments were performed using a LabRAM HR Evolution Raman System, provided by HORIBA Scientific and controlled via the software LabSpec 6. Further, the tests were performed using a frequency-doubled Nd:YAG laser, having a wave length of 532 nm and applying a laser power output of 0.84 mW on the sample. Besides, a diffraction grating with 600 lines per mm and a confocal pinhole with 100 mm diameter were employed. The wavelength calibration was realized by focusing the laser on a silicon wafer and analysing the first order phonon band of silicon at 520 cm-1. Since the DMQAC powder sample shows no first order phonon band of silicon at 520 cm-1, the wavelength calibration for this sample was done using the DMQAC peak at 1312 cm-1. Furthermore, the intensity correction algorithm of the LabSpec 6 software was used to adjust the Raman intensity variations caused by the Raman measurement system. To compensate for the occurrence of a strong fluorescence effect while analysing the 'graphene plus DMQAC powder' samples, the measurements were adjusted via a baseline correction, by applying a polynomial of the sixth degree. Additional information 8CB With respect to the result presented in the present publication, it should be noted that 8CB is known to self-assemble stable and well-ordered arrays that can be detected via STM [2] (for the chemical structure of 8CB refer Fig. 3). Having used 8CB in numerous STM experiments, it was found that the 8CB arrays and arrays built by semiconductor molecules can sometimes be imaged at the same time using identical STM scan settings, and sometimes not. In this regard, we propose that the difficulties in imaging both the adsorbate structures simultaneously are related to the orientation of the liquid crystal 8CB with respect to the substrate surface. STM detectable 8CB adsorbate structures form only when the liquid crystal is oriented in such a way that the molecules are aligned parallel to the substrate. So, although no 8CB can be seen in the 5 figures depicted in relation to the reworking experiments, 8CB arrays could however be found on the HOPG surface (refer Fig. 2). Further, we could never find bilayer or multiple layer structures built by alternating layers of QAC and 8CB arrays to date. Thus, it is assumed that the uncovered areas around QAC arrays most likely contain 8CB arrays, although they sometimes cannot be imaged via STM. In addition, results indicate that the QAC arrays exhibit a stronger affinity to the HOPG surface than the 8CB arrays. This is probably attributed to a strong π-π interaction between the fully condensed aromatic ring system of the QAC molecules and the graphene substrate, whereas the 8CB molecule providing just two phenyl groups enabling a π-π interaction and a weakly interacting alkyl chain (for the chemical structure of the 8CB molecule refer to the supporting information). Hence, it is assumed that the QAC molecules compete successfully for array formation, and furthermore that 8CB arrays are expelled by the expanding QAC arrays. The latter assumption is additionally supported by the finding that the reworked HOPG surface almost exclusively contains supramolecular QAC structures, whereas 8CB arrays being hardly found on such a sample. (a) (b) Fig. 2. STM image of a HOPG substrate treated with a dispersion of γQAC and 8CB. Fig. 3. Chemical structure of the 8CB molecule. (a) Supramolecular QAC array. (b) Array formed by 8CB molecules. References [1] A. Eberle, A. Nosek, J. Büttner, T. Markert, F. Trixler, Growing low-dimensional supramolecular crystals directly from 3D particles, CrystEngComm. (2017). doi:10.1039/C6CE02348G. 6 [2] T.J. McMaster, H. Carr, M.J. Miles, P. Cairns, V.J. Morris, Adsorption of liquid crystals imaged using scanning tunneling microscopy, J. Vac. Sci. Technol. A Vacuum, Surfaces, Film. 8 (1990) 672–674. doi:10.1116/1.576370. 7
1803.09255	1	1803	2018-03-25T13:57:50	Improvement of heat exchanger efficiency by using hydraulic and thermal entrance regions	[ "physics.app-ph", "physics.comp-ph" ]	This study investigates one of the possible approaches of improvement of heat exchangers efficiency. Literature review shows that most approaches of improvement are based on the heat transfer surface increasing and laminar-to-turbulent flow transition using different types of riffles forming and shaped inserts. In this article, a novel approach to the heat transfer intensification was employed. The main hypothesis is that applying of multi-chamber design of heat exchanger - ordinary shell-and-tube regions intersperse with common for all tubes regions - will help to improve the utilization of the entrance hydraulic and thermal regions thereby receive higher heat transfer coefficients and higher heat capacity of the heat exchange device. To prove the hypotheses we take the following steps. Firstly, development of the new geometry of the heat-exchanger design - multi chambers construction. Secondly, proving of the higher efficiency of novel design comparing to ordinary design by analytical calculations. Thirdly, numerical simulation of the heat exchange process and fluids flow in both types of heat exchangers that proves the analytical solution.	physics.app-ph	physics	Improvement of heat exchanger efficiency by using hydraulic and thermal entrance regions Alexey Andrianova, Alexander Ustinova, Dmitry Loginova,b aSkolkovo Institute of Science and Technology, Moscow, Russia bMoscow Institute of Science and Technology, Dolgoprudny, Russia Abstract This study investigates one of the possible approaches of improvement of heat exchangers efficiency. This problem was and still is a burning issue due to a wide variety of applications: aviation and aerospace, energy sector, chemical and food industry, refrigeration and cryogenic engineering, heating/cooling and conditioning services, thermal engines, housing and utility infrastructure etc. Among different types of the heat exchangers, shell-and-tube unit was chosen for the research, as its part in the overall number of heat exchangers is about 90% [3]. Literature review shows that most approaches of improvement are based on the heat transfer surface increasing and laminar-to-turbulent flow transi- tion using different types of riffles forming and shaped inserts. In this arti- cle, a novel approach to the heat transfer intensification was employed. The main hypothesis is that applying of multi-chamber design of heat exchanger -- ordinary shell-and-tube regions intersperse with common for all tubes re- gions -- will help to improve the utilization of the entrance hydraulic and thermal regions thereby receive higher heat transfer coefficients and higher heat capacity of the heat exchange device. To prove the hypotheses we take the following steps. Firstly, development of the new geometry of the heat- exchanger design - multi chambers construction. Secondly, proving of the higher efficiency of novel design comparing to ordinary design by analyti- cal calculations. Thirdly, numerical simulation of the heat exchange process and fluids flow in both types of heat exchangers that proves the analytical solution. Keywords: Energy systems, Heat exchanger, CFD modelling Email addresses: [email protected] (Aleksei Andrianov), [email protected] (Alexander Ustinov), [email protected] (Dmitry Loginov). 1. Introduction Technical-and-economic indexes of the different types of heating plants depend heavily on the heat exchangers [1, 2, 4, 9, 11]. The reason is sig- nificant part of the heat exchange units in the overall mass of the system. Furthermore, industrial growth in terms of both volume and capacity leads to the increase of heat exchangers capacity, overall dimensions and weight. As a result, the cost of the whole system increases. That is why the problem of rationalization and further development of the heat exchangers becomes more and more important and challenging issue. This challenge is important for a wide range of the industries due to the fact, that there are many different types of heat exchange units and they are essential part in such areas as: aviation and aerospace, energy sector, chemical and food industry, refrigeration and cryogenic engineering, heat- ing/cooling and conditioning services, thermal engines, housing and utility infrastructure etc [6, 7]. Many researchers paid attention to the problem of the heat exchangers efficiency improvement [5], describing the promising ways of the heat trans- fer intensification [8, ]. There are two main principles of the heat capacity increase. First, to increase the heat transfer surface area, allowing thermal energy change from the greater surface. In this case, a great variety of fins of different shapes and locations provides the better heat performance of the system. Second, to make the turbulence of the fluid flows. The laminar-to- turbulent transition can be achieved by using many kinds of shaped inserts in the inner space of the tubes, intensifying baffles of special design in the shell side -- space between tubes inside the shell -- of the heat exchanger, or special geometry of the tube itself for turbulent flow creation [12]. The aim of the current study is the development of the novel approach in the intensification process of the heat transfer. New construction -- multi- chamber design -- will help to achieve greater heat transfer coefficient and thereby will allow getting greater heat capacity of the system. The reason for that is special geometry that provides better utilization of the entrance hydraulic and thermal regions. This hypothesis is going to be proved by three interrelated steps. The first tool is analytical calculation of the heat capacity of the heat exchanger. The classical approach [10] is implemented for the ordinary type of the shell- and-tube heat exchanger. The novel unit capacity is estimated by dividing the system into sections and summing up the capacity of each section. The 2 second tool is numerical solution of both types of the heat exchanger with the same boundary conditions. The result is the pressure and temperature drop of fluid flows in two systems. The third tool is experimental solution, using specially constructed laboratory unit for testing ordinary and novel heat exchangers. Comparison of the results of all three steps for two systems will allow If so, getting answers if the novel geometry has better heat performance. how much the capacity of the new system is higher. 2. Calculations of the ordinary design and the novel multi-chamber design heat exchangers In this section shell and tube heat exchangers in novel and ordinary de- signs will be proposed with different boundary conditions, liquid-liquid type. The main hypothesis is that applying of multi-chamber design of heat exchanger -- ordinary shell-and-tube regions intersperse with common for all tubes regions -- will help to improve the utilization of the entrance hy- draulic and thermal regions thereby receive higher heat transfer coefficients and higher heat capacity of the heat exchange device. Ordinary design of the heat transfer unit is usual shell-and-tube heat ex- changer, which consists of shell with outlet and inlet of the cold heat transfer agent. Tubes are inside the shell with the common space for all tubes on both sides. These sides are inlet and outlet of the hot heat transfer agent. Values of all geometrical and thermal parameters are present below (Figure 1). Materials are water for hot and cold fluids; all walls are made of stainless steel. 3 Figure 1: Ordinary design of the shell-and-tube heat exchanger. Novel multi-chamber design is the development of the usual shell-and- tube heat exchanger. It has several qualitative changes. New heat exchange unit consist of three sections. Each section is the combination of two regions: first region is ordinary region of tubes inside the shell; second region is the common space for all tubes (Figure 2). This geometry can be described as the series of the three individual small shell-and-tube heat exchangers. This special design provides better utilization of the entrance hydraulic and ther- mal regions. The reason for that is relatively small length of each section -- fluid enters the pipes and boundary layers start to fill the inner space of the pipes until the end of the entrance region. Approximately at this point the common space for all tubes occurs. It means that we look at the process only in the entering regions where the highest heat transfer coefficient is. Values of all geometrical and thermal parameters are present below (Figure 2). 4 Figure 2: Novel multi-chamber design of the shell-and-tube heat exchanger. 2.1. Calculation of the heat capacity of the ordinary shell-and-tube heat ex- changer We need to define the heat capacity of the system. It depends on area of the heat transfer surface, temperature difference and coefficient of the heat transmission. The area of the heat transfer surface is sum of the outer surfaces of n number tubes: F = n · π · d · L = 10 · π · 0.025 · 1 = 0.785 m2 The logarithmic mean temperature difference is defined using equation: ∆T = h − T (cid:48) c) − (T (cid:48)(cid:48) h − T (cid:48)(cid:48) (T (cid:48) c ) h−T (cid:48) (T (cid:48) c) ln h −T (cid:48)(cid:48) (T (cid:48)(cid:48) c ) = (80 − 20) − (74 − 28) ln (80−20) (74−28) = 52.7 ◦C 5 We need to make preliminary calculations to define coefficient of the heat transmission: First, define the flow regime, using Reynolds number, for the inner tubes space and shell side. For the tubes, the characteristic dimension is equal to the inner diameter of the tube. For the shell side, the characteristic dimension is defined using hydraulic diameter equation. lshell = deq = ltube = din = 0.019m 4 · (π · (Din/2)2 − n · π · (d/2)2) π · Din + π · d · n 4 · f Π = = 0.0465 m Reynolds number for the inner tube space -- hot heat transfer agent: Retube = wtube · ltube vtube 0.3 · 0.019 3.62 · 10−7 = 15707 = Reynolds number for the shell side -- cold heat transfer agent: Reshell = wshell · lshell vshell 0.3 · 0.0465 9.19 · 10−7 = 15183 = The values of the Reynolds number both for inner tube space and shell side are higher than 10000. It means that the flow regime is turbulent and we need equation for Nusselt number [13, 15, ] in the turbulent flow. Prandtl number for both hot and cold fluids in inner tube space and shell side has the following form: P rtube = ctube · vtube · ρtube λtube P rshell = cshell · vshell · ρshell λshell = = 4194 · 3.62 · 10−7 · 975 0.68 = 2.2 4175 · 9.19 · 10−7 · 1000 0.59 = 6.4 Nusselt number for the turbulent flow has the following form in the inner tube space -- hot fluid: N utube = 0.021 · ε1 · Re0.8 · P r0.43 · (P r/P rw)0.25 = 62.75 Nusselt number for the turbulent flow has the following form in the shell side -- cold fluid: N ushell = 0.021 · ε1 · Re0.8 · P r0.43 · (P r/P rw)0.25 = 106.2 6 After Nusselt number calculation it is possible to define the heat transfer coefficient both for cold and hot heat transfer agents from the equation for Nusselt number: For the hot fluid in the inner tube space: 62.75 · 0.68 N utube · λtube α1 = ltube = 0.019 = 2248 W/(m2 ·◦ C) For the cold fluid in the shell side: α2 = N ushell · λshell lshell 106.2 · 0.59 0.0465 = = 1353 W/(m2 ·◦ C) Now we can find the value of the coefficient of the heat transmission: k = 1 λ + 1 + δ α2 1 α1 = 1 20 + 1 2248 + 0.003 1 1353 = 749 W/(m2 ·◦ C) Finally, we can define the heat capacity of the heat exchanger: Q = 749 · 52.7 · 0.785 = 31029 W 2.2. Estimation of the heat capacity of the novel multi-chamber shell-and-tube heat exchanger This paragraph called estimation rather than calculation due to specific cause. As the design of the multi-chamber heat exchanger is new, there is no special technique to calculate its heat capacity like in the previous para- graph. Therefore, we can only estimate it taken into consideration several assumptions. First, we will consider each section of the novel heat exchanger as a sepa- rate small ordinary shell-and-tube heat exchanger. Second, the distribution of the temperature between inlet region and outlet region will be taken as uniform -- the same temperature drop in each section. The temperature drop in the previous paragraph is 6◦C for hot fluid domain and 8◦C for cold fluid domain. For novel heat exchanger estimation, we will take the same temperature drops, paying attention to the distribution mentioned above. Therefore, the temperature drop in each section for the hot heat transfer agent will be 2◦C; for the cold heat transfer agent -- 2.6◦C. 7 The area of the heat transfer surface is sum of the outer surfaces of n number tubes: F = n · π · d · L = 10 · π · 0.025 · 1 = 0.2615 m2 The logarithmic mean temperature difference is defined using equation: ∆T = h − T (cid:48)(cid:48) c) − (T (cid:48)(cid:48) h − T (cid:48) (T (cid:48) c ) h−T (cid:48) (T (cid:48) c) ln h −T (cid:48)(cid:48) (T (cid:48)(cid:48) c ) = (80 − 20) − (74 − 28) ln (80−20) (74−28) = 57.66 ◦C We need to make preliminary calculations to define coefficient of the heat transmission: First, define the flow regime, using Reynolds number, for the inner tubes space and shell side. For the tubes, the characteristic dimension is equal to the inner diameter of the tube. For the shell side, the characteristic dimension is defined using hydraulic diameter equation. ltube = din = 0.019m 4 · (π · (Din/2)2 − n · π · (d/2)2) π · Din + π · d · n 4 · f Π = = 0.0465 m lshell = deq = Reynolds number for the inner tube space -- hot heat transfer agent: Retube = wtube · ltube vtube 0.3 · 0.019 3.62 · 10−7 = 16144 = Reynolds number for the shell side -- cold heat transfer agent: Reshell = wshell · lshell vshell 0.3 · 0.0465 9.19 · 10−7 = 14257 = The values of the Reynolds number both for inner tube space and shell side are higher than 10000. It means that the flow regime is turbulent and we need equation for Nusselt number in the turbulent flow. Prandtl number for both hot and cold fluids in inner tube space and shell side has the following form: P rtube = ctube · vtube · ρtube λtube = 4194 · 3.62 · 10−7 · 975 0.68 = 2.16 8 P rshell = cshell · vshell · ρshell λshell = 4175 · 9.19 · 10−7 · 1000 0.59 = 6.97 Nusselt number for the turbulent flow has the following form in the inner tube space -- hot fluid: N utube = 0.021 · ε1 · Re0.8 · P r0.43 · (P r/P rw)0.25 = 63.26 Nusselt number for the turbulent flow has the following form in the shell side -- cold fluid: N ushell = 0.021 · ε1 · Re0.8 · P r0.43 · (P r/P rw)0.25 = 104.12 After Nusselt number calculation it is possible to define the heat transfer coefficient both for cold and hot heat transfer agents from the equation for Nusselt number: For the hot fluid in the inner tube space: 62.75 · 0.68 N utube · λtube α1 = ltube = 0.019 = 2277 W/(m2 ·◦ C) For the cold fluid in the shell side: α2 = N ushell · λshell lshell 106.2 · 0.59 0.0465 = = 1316 W/(m2 ·◦ C) Now we can find the value of the coefficient of the heat transmission: k = 1 λ + 1 + δ α2 1 α1 = 1 20 + 1 2248 + 0.003 1 1353 = 741.5 W/(m2 ·◦ C) Finally, we can define the heat capacity of the heat exchanger: Q1 = k · ∆T · F = 749 · 52.7 · 0.785 = 11184 W We have three such sections so approximately the heat capacity of the system is three times higher: Q = 33552 W That is approximately 8% higher than the heat capacity of the ordinary shell-and-tube heat exchanger. However, the same set of the operations was 9 made with the conditions of laminar flow. All the initial geometry and ther- mal boundary conditions were the same, but the velocity of both hot and cold fluid were 0.03 m/s instead of 0.3 m/s. It changed the flow regime from turbulent to laminar and gave better results of heat capacity in the novel sys- tem. Heat capacity of the novel multi-chamber geometry system was about 30% higher than heat capacity of the ordinary geometry system. Here the calculation for one section will be shown. Other sections will not be shown. Nevertheless, all the sections have their estimation in the specially created Excel spreadsheet (Figure 3). It was done to make calculations in simpler and less time-consuming way. 10 Figure 3: Excel spreadsheet for the heat capacity calculation. 11 3. Numerical solution 3.1. Geometry. The geometry of two domains was prepared using SolidWorks software and can be seen on the Figure 4 -- 5. Figure 4: Geometry adapted for the numerical simulation. Ordinary shell-and-tube unit. Figure 5: Geometry adapted for the numerical simulation. Novel shell-and-tube unit. 3.2. Mesh. We conduct the mesh independent study for both types of the heat ex- changers and compare results for different sizes of the element of the mesh (Table 1 -- 2, Figure 6 -- 7). The method of the gradual mesh size reducing is used with the monitoring of two parameters of interest. In this case, they are temperatures of the outlet regions for hot and cold fluids [15 -- 18]. At moment when there is no significant change between values of the temperature on two steps, it can be considered the solution is not dependent on the mesh size. 12 Figure 6: Different mesh size for the ordinary shell-and-tube heat exchanger. Table 1: Mesh independent study for ordinary design. Type of the mesh in Comsol Multiphysics Extremely coarse Extra coarse Coarser Maximum/Minimum size of the element, mm Number of domain elements 48.7 / 10.3 29.5 / 7.38 19.2 / 5.9 124 885 169 201 268 046 Computational time 2 hours 01 min 2 hours 45 min Outlet temperature of cold domain, K Outlet temperature of hot domain, K 301.4 347.8 303.2 346.2 14 hours 16 min 303.1 346.0 As one can see from the table, there is almost no difference between the temperatures of the second and third steps (Extra coarse and Coarser types). At the same time, the third step is more resources consuming both in terms of time and computer memory. Thereby, we can use the second step conditions for numerical simulation, considering the results of the solution have enough accuracy. The same picture with the novel shell-and tube heat exchanger (Table 2, Figure 7). 13 Figure 7: Different mesh size for the ordinary shell-and-tube heat exchanger. Table 2: Mesh independent study for novel design. Type of the mesh in Comsol Multiphysics Extremely coarse Extra coarse Coarser Maximum/Minimum size of the element, mm Number of domain elements 48.7 / 10.3 29.5 / 7.38 19.2 / 5.9 132 718 190 003 280 830 Computational time 2 hours 32 min 4 hours 14 min Outlet temperature of cold domain, K Outlet temperature of hot domain, K 343.3 307.2 342.9 308.4 17 hours 52 min 342.9 308.3 3.3. Materials and boundary conditions. To have opportunity for the comparison of the performance of two types of the shell-and-tube heat exchangers, we are going to implement the same boundary conditions for both systems (Figure 8 -- 9). For numerical simula- tion, we need to define the inlet temperature of the hot and cold fluids and their inlet velocities. In our case, the heat transfer agents are water and wa- ter, the material of the metal part of the construction is stainless steel. Cold and hot water have 0.3 m/s velocity in the inlet region. The cold fluid has nearly room temperature and is 20◦C (293 K), hot water has temperature of 80◦C (353 K). 14 We also implement symmetry plane. The reason for that is axisymmetric problem and reduction of the element number of mesh to get the results quicker. Figure 8: Boundary conditions of the ordinary design of shell-and-tube heat exchanger. Figure 9: Boundary conditions of the novel design of shell-and-tube heat exchanger. 3.4. Solution results analysis. Problems with two different types of shell-and-tube heat exchangers were solved with the geometry, mesh and boundary conditions mentioned above. Computer was not the same as for previous sections. The main parameters of interest are outlet temperatures of hot and cold heat transfer agents. On the figure 10 -- 11 it can be seen the temperature distribution in the system of two fluids in thermal contact separated by metal heat transfer surface -- hot fluid domain and cold fluid domain. Figures of the temperature changing 15 within each domain show the heating of the cold heat transfer agent and the cooling of the hot heat transfer agent. Figure 10: Temperature distribution in the ordinary design of heat exchanger. Figure 11: Temperature distribution in the novel design of heat exchanger. The comparison of the results for two systems shows the better perfor- mance of the novel design of heat exchanger. The temperature difference be- tween inlet and outlet regions of cold and hot fluids are higher in case of novel design. For the hot fluid domain, ∆T1 = 353 − 346.2 = 6.8K -- ordinary de- sign; ∆T2 = 353−342.9 = 10.1K -- new design. The ratio is ∆T2/∆T1 = 1.49. The same case with the cold fluid domain, ∆T1 = 303.2 − 293 = 10.2K, ∆T2 = 308.4 − 293 = 15.4. The ratio is ∆T2/∆T1 = 1.51. This fact allows us making conclusion about higher efficiency of the novel shell-and-tube heat exchanger. However, along with the higher temperature drop novel design has the higher pressure drop (Figure 12 -- 13 -- pressure on the pictures is difference between pressure in the system and atmosphere pressure). The reason for that is com- plexity of the geometry. Novel heat exchanger has more regions with local resistances. In case of hot fluid domain, there is almost no difference between novel and ordinary designs. In case of cold fluid domain, the difference in 16 pressure drop is significant. It is almost six time higher if we compare the inlet regions. Figure 12: Pressure distribution in the ordinary design of heat exchanger. Figure 13: Pressure distribution in the novel design of heat exchanger. Numerical solution shows the strengths and weaknesses of the new multi- chamber design of heat exchanger. First, it really has better performance in terms of heat capacity. If we have the system where we need to cool down the hot fluid by using cooling liquid, the novel design will give 1.5 times less temperature of interest in the outlet region of the system. At the same time, pressure drop increases almost by six times. It means that more powerful equipment for pumping of the cooling liquid is needed. This obstacle needs consideration in each particular situation. For example, there are can be case when the space for the equipment installation is critical (e.g. automobile industry). The producer may not care about pumping, but the main aim is to install more heat effective exchanger unit and save the initial space. In this particular case, we do not worry about high pressure drop. However, in general case, we need to find the optimal ratio between the heat intensification and increase of the operating pressure. As it is seen from the figures above, we have the intermittent growth of the pressure in 17 the cold fluid domain on the stage of going from one chamber into another. The reason for pressure drop is significant decrease of the cross section of the channel -- first region is shell side space, but next region is only one medium tube with several times less area of cross section. The possible solution is increase of the diameter of medium tube. Then the cross section of the shell side space will be less two. On the one hand, it leads to lower pressure step between these two regions, on the other, less cross section of the shell side leads to greater pressure from the beginning of the system. Therefore, the iteration problem needs to be solved with monitoring of the pressure drop ratio in the medium tube and chamber space and monitoring of the ratio between pressure drop and heat capacity. This process is a part of future study continuing the current one. 4. Conclusion In the current study, the problem of the efficiency improvement in shell- and tube heat exchangers is investigated. Novel approach for heat transfer in- tensification with help of entrance hydraulic and thermal regions is described. According to hypothesis of new method, special multi-chamber geometry of the shell-and-tube heat exchanger provides better performance of the system in terms of heat capacity. This special new design was developed in order to prove the hypothesis by means of analytical, numerical and experimental solutions. The first step is analytical solution. It consists of two parts -- calculation of the heat capacity of ordinary heat exchanger and estimation of the heat capacity of novel heat exchanger. It was initial stage, but it was important to understand if the current problem and new approach of its solution really happen to be. Results of the analytical solution show 8% higher heat capacity of the novel design in case of turbulent fluid flow, and up to 30% higher heat capacity in case of laminar fluid flow. At this point, it is necessary to realize that the results are not one hundred percent accurate, taking into consideration all the assumptions, but they depict the characteristic behavior of the new system (efficiency improvement of the heat exchanger) and allow future speculations on this theme. The second step is numerical solution. It is numerical simulation of the fluids flow inside the ordinary and novel shell-and-tube heat exchangers with the thermal contact of the fluids in the Comsol Multiphysics software envi- ronment. The problems for both types of systems were solved with the same 18 boundary conditions in order to compare the results. Solution shows almost 1.5 times higher temperature drop of the heat transfer agents between inlet and outlet regions. It proves the hypothesis of better heat performance of the novel design. However, complex geometry of new system raises require- ments for pump equipment - it should be taken into consideration during the technical-and-economic indexes estimation in each particular case. The third step is experimental solution. It is set of tests on real equip- ment and comparison of the results -- outlet temperature measurements -- for both ordinary and novel designs. These tests will be possible only after the laboratory unit producing. Now we have almost half of the equipment ready, the rest part is on the stage of manufacturing. As the conclusion it can be said that hypothesis about higher heat ca- pacity of the multi-chamber design of shell-and-tube heat exchanger was successfully proved. Novel construction has visible advantages and can be even patented. 19 References. References [1] Heatric. Heat Exchanger Industry Applications. http://www.heatric. com/heat exchanger. [2] Applications of Heat Exchangers. http://www.barriquand.com/en/ heat-exchangers-industry. [3] Smart district heating and cooling, SETIS. https://setis.ec.europa.eu/ energy-research/publications/smart-district-heating-and-cooling. [4] Large Scale Solar District Heating, PlanEnergi. http:// solar-district-heating.eu. [5] Suhov V. V., Kazakov G. M. Basics of designing and calculating heat exchangers 2009. [6] Shell and tube heat exchanger. https://en.wikipedia.org/wiki/ [7] Rajiv Mukherjee Effectively design shell-and-tube heat exchangers Chem- ical Engineering Progress, 1998. [8] NPTEL, Chemical Engeneering Process design of heat exchanger, 1991. [9] http://russian.spiralfinnedtube.com/photo/pl5637163-aluminum. [10] http://studopedia.ru/4 35439 apparati-s-trubchatoy-poverhnostyu-teploobmena. html. [11] Bannih O. P. Main constructions and thermal calculation of heat ex- changer, 2012. [12] Pishchulin V. P. Calculation of shell and tube heat exchanger 2010. [13] Sadik Kakac Heat exchangers: selection, rating, and thermal design, 2012. [14] Gortishov U. F. Heat exchangers. 2012. [15] Kuppan Thulukkanam Heat exchanger design handbook, second edition 2013. 20 [16] Koch Heat Transfer Company. http://www.kochheattransfer.com/ products/twisted-tube-bundle-technology. [17] Eddify. http://www.eddyfi.com/wp-content/uploads/2014/10/ application-note-twisted-hx-tubes.pdf. [18] Trinvalco. http://www.trinvalco.com/Koch-Heat-Transfer-Company/ twisted-tube-technology. 21
1809.09641	1	1809	2018-09-25T18:12:58	Non-Reciprocal Willis Coupling in Zero-Index Moving Media	[ "physics.app-ph" ]	Mechanical motion can break the symmetry in which sound travels in a medium, but significant non-reciprocity is typically achieved only for very large motion speeds. Here we combine moving media with zero-index acoustic propagation, yielding extreme non-reciprocity and induced bianisotropy for modest applied speeds. The metamaterial is formed by an array of waveguides loaded by Helmholtz resonators, and it exhibits opposite signs of the refractive index sustained by asymmetric Willis coupling for propagation in opposite directions. We use this response to design a non-reciprocal acoustic lens focusing only when excitation from one side, with applications for imaging and ultrasound technology.	physics.app-ph	physics	Non-Reciprocal Willis Coupling in Zero-Index Moving Media Li Quan1, Dimitrios L. Sounas1,2, and Andrea Alù1,3,4,5,* 1Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX 78712, USA 2Department of Electrical and Computer Engineering, Wayne State University, Detroit, MI 48202, USA 3Photonics Initiative, Advanced Science Research Center, City University of New York, New York, NY 10031, USA 4Physics Program, Graduate Center, City University of New York, New York, NY 10016, 5Department of Electrical Engineering, City College of The City University of New York, USA NY 10031, USA *Corresponding author: [email protected] Mechanical motion can break the symmetry in which sound travels in a medium, but significant non-reciprocity is typically achieved only for very large motion speeds. Here we combine moving media with zero-index acoustic propagation, yielding extreme non-reciprocity and induced bianisotropy for modest applied speeds. The metamaterial is formed by an array of waveguides loaded by Helmholtz resonators, and it exhibits opposite signs of the refractive index sustained by asymmetric Willis coupling for propagation in opposite directions. We use this response to design a non- reciprocal acoustic lens focusing only when excitation from one side, with applications for imaging and ultrasound technology. Reciprocity in wave propagation demands that the response of a source remains the same when source and observation points are interchanged. Breaking this symmetry allows designing devices that exhibit different transmission for opposite propagation directions, which is important for protection of sensitive equipment from external interference and for full-duplex communications. For electromagnetic waves, several approaches have been proposed to break reciprocity, including biasing with external magnetic fields [1]-[2] transistors [3]-[4], angular momentum [5]-[6], spatiotemporal modulation [7]-[8], and non- linearities [9]-[10]. In acoustics, non-reciprocal devices have been mainly realized based on nonlinear mechanisms [11]-[13]. It is well known that sound traveling parallel or anti- parallel to a moving medium is transmitted non-reciprocally [14], however strong effects are typically achieved only when the velocity of the medium is large, or in highly resonant devices. Based on this principle, momentum bias applied through moving media was recently used to realize linear acoustic non-reciprocal devices [15]-[16]. In the following, we explore moving metamaterials operated around their zero-index operation, showing how in this scenario mechanical motion opens highly unusual scenarios for sound propagation. A moving medium exhibits different wave-vectors and for opposite propagation directions, which is a signature of nonreciprocity. As such, the nonreciprocal coefficient (1) measures the degree of asymmetry in wave propagation for opposite directions. In absence of motion, and are necessarily the same in magnitude and with opposite signs, leading to , while is nonzero when reciprocity is broken. In a moving non- kkReReReRekkkkkk0dispersive medium, for small velocities [17], where and are the reciprocal and non-reciprocal portions of the wavenumber, respectively, is the Mach number, defined as the ratio of flow speed to the background sound speed , and is the wavenumber in free space. Replacing these quantities in (1) yields , implying that non-negligible non- reciprocity can be expected only for large, often impractical Mach numbers. In order to break this trade-off and achieve large non-reciprocity with small flow speed, we explore the regime for which , i.e., zero-index propagation, so that the non- reciprocal portion of the wave-number dominates. In electromagnetics, epsilon-near- zero (ENZ) metamaterials [18]-[19] provide a reciprocal zero index of refraction, which has been shown to lead to extreme wave propagation properties [20]-[21]. Acoustic waves in a zero-index media may therefore provide a platform to boost nonreciprocal phenomena when modest medium speeds are considered. Density-near-zero metamaterials [22]-[23], the analogue of ENZ materials for sound, have been realized in the past using waveguides loaded by membranes. However, this approach is not suitable for our purpose, because membranes block material flow. We consider therefore extreme non-reciprocal responses by imparting air flow to waveguides loaded by a Helmholtz resonator array, inducing a near-zero refractive index at rest ( ) [24], while . In such a metamaterial, we expect that even a small mechanical motion yields a negative phase velocity (and refractive index) for propagation parallel to the fluid motion, and a positive one for propagation anti- parallel to it, hence . In the following, we verify this response for plane-wave incidence, observing opposite refraction angles for excitation from opposite RNRkkk20/1RkkM20/1NRkMkM00/MUc0U0c00kc/NRRkkM0RkNRk0Rk0NRk/NRRkkMsides. After explaining the physics around this unusual response, and its relation to non- reciprocal Willis coupling, we then utilize these properties to design a planar non-reciprocal lens, which can focus a source from one side, but acting as a divergent lens for opposite excitation. The geometry under analysis is shown in Fig. 1a and it consists of an array of parallel waveguides loaded with Helmholtz resonators. The green color indicates the region with air flow, and the geometrical parameters are provided in the caption. In order to impart air flow in each waveguide, a pipe is connected to both sides and loaded with a fan, (not shown in the figure), inducing a continuous air flow, as indicated by the arrows. To ensure that the incident acoustic wave only travels through the waveguide, we load two Helmholtz resonators at each pipe opening (not shown in the figure), designed to filter out the frequencies of excitation. The structure is supplied with two quarter-wavelength matching layers at both sides (yellow color), in order to eliminate impedance mismatch. Figure 1b shows the calculated Mach number along each waveguide, which starts at zero outside the moving segment, and it linearly increases to an approximately constant value of 0.1 inside the waveguide through a finite transition layer. The small fluctuations of the Mach number inside the waveguide are due to the presence of the Helmholtz resonators. The numerical simulations here and in the following are performed using COMSOL Multiphysics [25]. Consider now the excitation of this geometry with obliquely-incident plane waves from opposite sides, as shown in Fig. 2. In our simulations, we use the continuity boundary conditions for acoustic pressure and particle velocity at the interface between the region with no motion and the matching layer. At the interface between the region with no motion and the transition layer inside the waveguides, as well as at the one between the transition layer and the main body of the waveguide, we instead use continuity boundary conditions for the acoustic pressure and the air mass flow [26]. Figure 2a presents the acoustic pressure field distribution for excitation from the left side. Due to momentum conservation, the tangential component of the wave-vector is the same everywhere in space, while the normal component changes direction across the interface, resulting in negative refraction. Quite interestingly, the situation is opposite for excitation from the right side: Figure 2(b) presents the acoustic fields for an incident wave coming from the right side at the same incident angle. In this case, the direction of the normal component of the wave-vector does not flip as the wave enters the metamaterial, indicating positive refraction. This drastically different refraction response from opposite sides is a signature of extreme nonreciprocity, arising from the moving medium combined with the index-near-zero response in the metamaterial. We stress that this is achieved for modest values of M. A better understanding of the phenomenon can be gained by extracting the effective constitutive parameters of the metamaterial. The mass conservation equation and momentum equation in each waveguide with air flow are derived in detail in [27], and take the general form (2) , (3) where is the effective bulk modulus, the effective density, is the bulk modulus in air, is and are odd bianisotropy cross coupling coefficients arising from non-reciprocity, is a factor that depends on the geometry of the 1effeffuxiEpueffeffpxiup11120/1effEEFM20/1effM2000Ec120/1effMcM112000/1effMcFcM220waFSdMstructure, and , are the Helmholtz resonator resonance angular frequency and acoustical mass, respectively. The dispersion of the effective parameters versus frequency for the metamaterial geometry at hand is shown in [27]. The mechanical motion in each waveguide introduces highly unusual propagation properties around the zero-index propagation regime. For a stationary waveguide loaded with Helmholtz resonators [28], which corresponds to our scenario in the limit of , only the bulk modulus is affected by the loads, yielding near-zero , corresponding to a zero index of refraction, when . The effective density is not affected by the resonators. When a modest fluid motion is considered, the effective density and bulk modulus are weakly modified through the factor , but most importantly they are coupled together through the bianisotropy coefficients and , also known as Willis coupling, which is at the core of the described phenomena. Different from conventional Willis coupling [29]-[32], these coefficients do not obey reciprocity, , and are odd with respect to , i.e., they flip sign for opposite propagation directions, a clear sign of non-reciprocity. As we show in [27], around the zero-index operation goes through a resonance and flips sign, similar to , producing extremely asymmetric Willis coupling coefficients and non-reciprocal response at the zero-index operation. By combining Eqs. (2) and (3), we derive the dispersion relation yielding and . At the angular frequency , (4) 0aM0MeffEeffE110EF121MeffeffeffeffMeffeffE21422effeffEk2142ReffeffkE2NRk, , and , largely enhancing the non- reciprocal response, due to asymmetric Willis coupling induced by the mechanical motion. In this regime the wavenumber reads , (5) enabling opposite refractive index for opposite directions of propagation. Interestingly, the wavenumber in the metamaterial has the same real value for both propagation directions, i.e., no matter whether the incident wave is coming from left or right, the wavevector has the same direction and value, anti-parallel to the motion, consistent with our numerical simulations in Fig. 2. We stress that for the effective bulk modulus is negative, , and the effective density positive, , yet the acoustic wave travels in the metamaterial without decay because of the strong Willis coupling response. Figure 3 shows the dispersion of the wavenumber in Eq. (4). In absence of air flow (M = 0), the dispersion has a cut-off at the zero-index condition , and it is strictly even with respect to k, as expected from reciprocity. When a moderate air-flow is turned on (M = 0.1), the dispersion diagram is asymmetric, as expected for a nonreciprocal medium, and the cutoff frequency shifts down to . Around this frequency, waves propagating in opposite directions have a nonzero (negative) wavenumber, independent of the propagation direction, and the non-reciprocity coefficient is very large. 2200112waMESdM0Rk022111NRMkkkMM0Rk214effeffeffeffE20/1effM110EF2200112waMESdMThe non-reciprocal Willis coupling introduced here through mechanical motion at the zero-index frequency can be used to create a lens that focuses a source placed at one side, but with diverging properties when a source is placed on the other side. The focusing operation is achieved by imparting a phase shift across the structure that transforms a diverging circular wavefront to a converging one [27], which is achieved tailoring the air flow velocity across different channels to accumulate the required phase at each aperture. Our design is shown in Fig. 4(a), where we plot the relation between Mach number, maintained small in each channel, and the channel number n, with n = 0 being the channel on the same axis as the source (y = 0). The air flow in each channel is symmetric with respect to the y-axis (i.e., for channels N and -- N the air flow is the same), so we only show the imparted Mach number for positive n. Figure 4(b) presents the calculated acoustic pressure distribution when the sound source is located on the left of the lens: after travelling through the planar metamaterial, a focused image is constructed at the right side of the lens. Figure 4(c) presents instead the pressure distribution when the sound source is located at the right side of the lens: here we only get a divergent wave, demonstrating strong nonreciprocity with modest required flow velocities, enabled by operating around the zero- index frequency of the metamaterial. By adjusting the air flow in each channel, we can get different functionalities. In Figs. 4d-f we show a design that converts a point source to a plane wave only when the source is located on the right side of the metamaterial. Again, the relation between channel number and Mach number is presented in Fig. 4(d), while Figs. 4e-f show the acoustic pressure profile when the source is located at the two sides of the structure [27], highlighting again the strongly non-reciprocal response. In conclusion, we have presented here a Willis metamaterial operating near the zero- index operation, yielding extreme non-reciprocal responses with modest air flow. These unusual acoustic properties are the result of a non-reciprocal bianisotropic response dominating the effective properties of the metamaterial, as the reciprocal response tends to zero. For this reason, even modest air flows can provide highly unusual responses, including opposite (positive-to-negative) refractive index for opposite propagation directions, and non-reciprocal lenses. The air flow can be modulated in real-time to reconfigure the properties of the metamaterial, making it an exciting platform to control sound beyond the conventional limitations of natural materials. We envision a plethora of applications of these concepts, from ultrasound imaging to sonar technology, with possible extensions to the realm of phononics and surface acoustic waves. Reference [1] Z. Wang, Y. Chong, J. D. Joannopoulos, and M. Soljacic, Observation of unidirectional backscattering-immune topological electromagnetic states. Nature 461, 772-775 (2009). [2] F. D. M. Haldane, and S. Raghu, Possible realization of directional optical waveguides in photonic crystals with broken time-reversal symmetry. Phys. Rev. Lett. 100, 013904 (2008). [3] T. Kodera, D. L. Sounas, and C. Caloz, Artificial Faraday rotation using a ring metamaterial structure without static magnetic field. Appl. Phys. Lett. 99, 031114 (2011) [4] B. I. Popa, and S. A. Cummer, Nonreciprocal active metamaterials, Phys. Rev. B 85, 205101 (2012). [5] 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). [6] N. Estep, D. Sounas, J. Soric, and A. Alu, Magnetic-free non-reciprocity based on parametrically modulated coupled-resonator loops, Nat. Phys. 10, 923 (2014). [7] H. Lira, Z. Yu, S. Fan, and M. Lipson, Electrically driven nonreciprocity induced by interband photonic transition on a silicon chip. Phys. Rev. Lett. 109, 033901 (2012). [8] D. Wang, H. Zhou, M. Guo, J. Zhang, J. Evers, and S. Zhu, Optical diode made from a moving photonic crystal, Phys. Rev. Lett. 110, 093901 (2013). [9] S. Lepri, and G. Casati, Asymmetric wave propagation in nonlinear system. Phys. Rev. Lett. 106, 164101 (2011). [10] M. Soljacic, C. Luo, J. D. Joannopoulos, and S. Fan, Nonlinear photonic crystal microdevices for optical integration. Opt. Lett. 28, 637 (2003). [11] B. Liang, B. Yuan, J. C. Cheng, Acoustic diode: rectification of acoustic energy flux in one-dimensional system. Phys. Rev. Lett. 103, 104301 (2009). [12] B. Liang, X. S. Guo, J. Tu, D. Zhang, J. C. Cheng, An acoustic rectifier, Nat. Mater. 9, 989 (2010). [13] N. Boechler, G. Theocharis, C. Daraio, Bifurcation-based acoustic switching and rectification. Nat. Matter. 10, 665 (2011). [14] L. M. Brekhovskikh, I. P. Lysanov, Fundamentals of Ocean Acoustics (Springer, Berlin, 2003). [15] R. Fleury, D. L. Sounas, C. F. Sieck, M. R. Haberman, and A. Alu, Sound isolation and giant linear nonreciprocity in a compact acoustic circulator, Science 343, 516 (2014). [16] R. Fleury, D. L. Sounas, and A. Alu, Subwavelength ultrasonic circulator based on spatio-temporal modulation, Phys. Rev. B 91, 174306 (2015). [17] P. M. Morse and K. U. Ingard, Theoretical Acoustics, Princeton University Press, (1987). [18] A. Alu, M. G. Silveirinha, A. Salandrino and N. Engheta, Epsilon-near-zero metamaterials and electromagnetic sources: tailoring the radiation phase pattern, Phys. Rev. B 75, 155410 (2007). [19] B. Edwards, A. Alu, M. E. Young, M. Silverinha, and N. Engheta, Experimental verification of epsilon-near-zero metamaterial coupling and energy squeezing using a microwave waveguide, Phys. Rev. Lett. 100, 033903 (2008). [20] I. Liberal, and N. Engheta, Near-zero refractive index photonics, Nat. Photon. 11, 149-158 (2017). [21] I. Liberal, A. M. Mahmoud, Y. Li, B. Edwards, and N. Engheta, Photonic doping of epsilon-near-zero media, Science 355, 1058-1062 (2017). [22] R. Fleury and A. Alu, Extraordinary sound transmission through density- near-zero ultranarrow channels, Phys. Rev. Lett. 111, 055501 (2013). [23] Y. Gu, Y. Cheng, X. Liu, Acoustic planar hyperlens based on anisotropic density-near-zero metamaterials, Appl. Phys. Lett. 107, 133503 (2015). [24] L. Quan, X. Zhong, X. Liu, X. Gong and P. A. Johnson, Effective impedance boundary optimization and its contribution to dipole radiation and radiation pattern control, Nat. Commun. 5, 3188 (2014). [25] [26] COMSOL Multiphysics, https://www.comsol.com/ G. K. Bachelor, An Introduction to Fluid Dynamics, Cambridge University Press, (2000). [27] See Supplementary Material for a detailed derivation of the mass conservation and momentum equations, the dispersion relation of the effective parameters, and the design details of the non-reciprocal lens. [28] N. Fang, D. Xi, J. Xu, M. Ambati, W. Srituravanich, C. Sun and X. Zhang, Ultrasonic metamaterials with negative modulus, Nat. Mater. 5, 452 (2006). [29] J. R. Willis, Variational principles for dynamic problems for inhomogeneous elastic media, Wave Motion 3, 1 (1981). [30] A. N. Norris, A. L. Shuvalov, and A. A. Kutsenko, Analytical formulation of three-dimensional dynamic homogenization for periodic elastic system, Proc. R. Soc. A 468, 1629 (2012). [31] L. Quan, Y. Ra'di, D. L. Sounas and A. Alu, Maximum Willis coupling in acoustic scatterers, Phys. Rev. Lett. 120, 254301 (2018). [32] C. F. Sieck, A. Alu and M. R. Haberman, Origins of Willis coupling and acoustic bianisotropy in acoustic metamaterials through source-driven homogenization, Phys. Rev. B 96, 104303 (2017). [33] L. Quan, F. Qian, X. Liu and X. Gong, A nonlinear acoustic metamaterial: Realization of a backwards-traveling second-harmonic sound wave, J. Acoust. Soc. Am. 139, 3373 (2016). Figures Fig. 1. (a) Geometry of the nonreciprocal metamaterial, formed by an array of parallel waveguides loaded by Helmholtz resonators (green). A constant air flow inside the waveguides is generated by fans. The waveguides have a width Sw=3 mm, a distance d=8 mm between neighboring resonators, which have a neck length l=0.5 mm, a neck width a=1 mm, a cavity length b=7 mm and a cavity width h=3.5 mm. (b) Mach number distribution in the waveguide between two matching layers. Fig. 2. (a) Acoustic pressure distribution for an incident wave coming from the left. The yellow arrows indicate the wave vector in air and the green arrow indicates the wave vector in the metamaterial. (b) Acoustic pressure distribution for an incident wave coming from the right. Fig. 3. Dispersion diagram for the geometry of Fig. 1(a). (a) Real part. (b) Imaginary part. Fig. 4. (a) Modulation of the Mach number in each channel to synthesize a focusing lens. (b) Acoustic pressure distribution when the source is located on the left side of the lens. A focused image is obtained on the right. (c) Acoustic pressure distribution when the source is located on the right of the lens. (d) Modulation of the Mach number to synthesize a point- source to plane wave converter. (e) Acoustic pressure distribution when the source is located on the right. A plane wave is induced on the left side. (f) Acoustic pressure distribution when the source is located on the left side.
1904.11097	1	1904	2019-04-24T23:29:31	Gold/Parylene-C/Pentacene Capacitor under Constant-Voltage Stress	[ "physics.app-ph" ]	Degradation of metal-insulator-semiconductor (MIS) capacitors of gold/Parylene-C/Pentacene under constant voltage stress (CVS) was investigated to explore the electrical stability and reliability of Parylene C as a gate dielectric in flexible electronics. A stress voltage of fixed magnitude as high as 20 V, both negative and positive in polarity, was applied to each MIS capacitor at room temperature for a fixed duration as long as 10 s. The CVS effects on the capacitance-voltage curve-shift, the time-dependent leakage current, and the time-dependent dielectric breakdown were measured and analyzed. CVS is observed to induce charge in Parylene-C and its interfaces with gold and Pentacene. The net induced charge is positive and negative for, respectively, negative and positive gate bias polarity during CVS. The magnitude of the charge accumulated following positive gate CVS is significantly higher than that following negative gate bias CVS in the range of 4 to 25 nC cm$^{-2}$. In contrast, the leakage current during the negative gate stress is three orders of magnitude higher than that during the positive gate stress for the same bias stress magnitude. The charge buildup and leakage current are due to the trapping of electrons and holes near the Parylene-C/Pentacene interface as well as in the Parylene-C layer. Before the application of the CVS, a dielectric breakdown occurs at an electric field of 1.62 MV cm$^{-1}$. After the application of the CVS, the breakdown voltage decreases and the density of the trapped charges increases as the stress voltage increases in magnitude, with the polarity of the trapped charges opposite to that of the stress voltage. The magnitude and direction of the capacitance-voltage curve-shift depend on the trapping and recombination of electrons and holes in the Parylene-C layer and in the proximity of the Parylene-C/Pentacene interface during CVS.	physics.app-ph	physics	Gold/Parylene-C/Pentacene Capacitor under Constant-Voltage Stress Ibrahim H. Khawajia,b, Alyssa N. Brigemana, Osama O. Awadelkarimc,d,∗, Akhlesh Lakhtakiac a Department of Electrical Engineering, Pennsylvania State University, University Park, PA 16802, USA. b Department of Electrical Engineering, Taibah University, P.O. Box 344, Al-Madina Al Munawara, KSA. cDepartment of Engineering Science and Mechanics, Pennsylvania State University, University Park, PA dThe Center for Nanotechnology Education and Utilization, Pennsylvania State University, University 16802, USA. Park, PA 16802, USA. Abstract Degradation of metal-insulator-semiconductor (MIS) capacitors of gold/Parylene-C/Pentacene under con- stant voltage stress (CVS) was investigated to explore the electrical stability and reliability of Parylene C as a gate dielectric in flexible electronics. A stress voltage of fixed magnitude as high as 20 V, both negative and positive in polarity, was applied to each MIS capacitor at room temperature for a fixed duration as long as 10 s. The CVS effects on the capacitance-voltage curve-shift, the time-dependent leakage current, and the time-dependent dielectric breakdown were measured and analyzed. CVS is observed to induce charge in Parylene-C and its interfaces with gold and Pentacene. The net induced charge is positive and negative for, respectively, negative and positive gate bias polarity during CVS. The magnitude of the charge accumulated following positive gate CVS is significantly higher than that following negative gate bias CVS in the range of 4 to 25 nC cm−2. In contrast, the leakage current during the negative gate stress is three orders of magnitude higher than that during the positive gate stress for the same bias stress magnitude. The charge buildup and leakage current are due to the trapping of electrons and holes near the Parylene-C/Pentacene interface as well as in the Parylene-C layer. Before the application of the CVS, a dielectric breakdown occurs at an electric field of 1.62 MV cm−1. After the application of the CVS, the breakdown voltage decreases and the density of the trapped charges increases as the stress voltage increases in magnitude, with the polarity of the trapped charges opposite to that of the stress voltage. The magnitude and direction of the capacitance- voltage curve-shift depend on the trapping and recombination of electrons and holes in the Parylene-C layer and in the proximity of the Parylene-C/Pentacene interface during CVS. Keywords: Parylene C, flexible electronics, dielectric polymer, constant-voltage stress, charge buildup. 1. Introduction The stability of flexible devices is a major reliability concern. To enhance environmental stability, passivation/encapsulating layers are used. Another critical reliability concern is device stability against electrical stress. Flexible devices often operate under continuous voltage biases which can affect performance dramatically [1, 2]. The general reason for bias ∗corresponding author, [email protected] Preprint submitted to Journal of LATEX Templates April 26, 2019 9 1 0 2 r p A 4 2 ] h p - p p a . s c i s y h p [ 1 v 7 9 0 1 1 . 4 0 9 1 : v i X r a instability is charge trapping [2]; electrode, insulator, and semiconductor interfaces easily trap charges, resulting in device-parameter shifts and degraded performances. Therefore, processes that take place at bimaterial interfaces must be characterized and controlled. Parylene C is commonly used as a dielectric material for electronic applications [1, 3, 4, 5, 6, 7, 8, 9]. Pinhole-free films of this polymer function well as protection layers[10] to minimize device degradation caused by exposure to air and moisture. In recent years, this polymer has been used for flexible substrates due to its desirable mechanical properties (yield strength of 55.1 MPa and static Young's modulus of 2.76 GPa) and the ease of deposition in both micrometer- and nanometer-thickness regimes [11]. In addition, we have recently shown that Parylene-C columnar microfibrous thin films are viable candidates for use as interlayer dielectrics [3, 4, 5]. Moreover, Parylene-C thin films exhibit superior electrical insulation characteristics, high breakdown strength (∼2.5 MVcm−1), and low dielectric loss [11], not only as passivation layers but also as gate dielectrics for organic field-effect transistors (OFETs) [12, 13, 14]. In fact, there is a limited number of materials that can be simultaneously used as a substrate, gate dielectric, and encapsulation layer simultaneously while exhibiting a high-performance level comparable to materials dedicated to a specific application [1, 2]. Parylene-C films have been extensively studied [1, 12, 13, 14, 15, 16, 17, 18, 19, 20] as gate dielectrics in OFETs. However, very little is known about the electrical stability and reliability of Parylene C and its interfaces with the active layers in these devices. Current- voltage measurements are commonly made to study the electrical degradation of OFETs under constant-voltage stress (CVS) [20]. However, capacitance-voltage measurements are more sensitive than current-voltage measurements for investigating interface characteristics [21]. In order to address this shortfall, we investigated the electrical stability and reliability of Parylene C as a gate dielectric using capacitance-voltage measurements. We undertook a systematic analysis of gold/Parylene-C/Pentacene metal -- insulator -- semiconductor (MIS) capacitors. In this paper, the effects of CVS, the capacitance-voltage curve-shift, and time-dependent dielectric-breakdown (TDDB) are experimentally analyzed. Needless to add, the gold/Parylene-C/Pentacene system is the heart of the OFET. 2. Experimental Procedures P-type silicon (p-Si) substrates were ultrasonically cleaned in an ultrasonic cleaner (2200 Branson, Emerson, St. Louis, MO, USA) using, successively, acetone, DI water, isopropyl alcohol, and DI water for 10 min each. Subsequently, the cleaned substrates were etched for 20 min using a 1:4 mixture of hydrofluoric acid and DI water. A 150-nm-thick layer of silicon dioxide (SiO2) was then deposited atop the p-Si substrate using a PECVD tool (P-5000, Applied Materials, Santa Clara, CA, USA). Thereafter, a 50-nm-thick adhesion layer of chromium (Cr) was deposited using a sputter tool (Desktop Pro R(cid:13), Denton Vacuum, Moorestown, NJ, USA). Next, a 150-nm-thick gold (Au) layer was sputtered on top of the Cr layer using the same tool. Pentacene was purified in gradient-temperature sublimation system (MK-5024-S, Lind- berg Electric, Watertown, WI, USA) and then loaded in a tubular chamber. A thermal 2 gradient was maintained along that chamber maintained at about 10−5 Torr pressure. The material was sublimated at 300 ◦C and re-condensed down the tube at 165 ◦C in order to drive out impurities. A 150-nm-thick layer of purified Pentacene was then deposited atop the Au layer via vacuum thermal evaporation (Amod, Angstrom Engineering, Kitchener, ON, Canada) at a rate of 0.2 nm s−1. During that process, the Au/Cr/SiO2/p-Si structure was fixed to a rotating chuck and maintained at 0 ◦C in a chamber with a base pressure of 10−7 Torr. A 200-nm-thick insulating layer of Parylene C was deposited on top of the Pentacene layer using a physicochemical vapor deposition technique. An aluminum-foil boat loaded with 0.2 g of commercial Parylene-C dimer (980130-C-01LBE, Specialty Coatings and Systems, In- dianapolis, IN, USA) was placed inside the vaporizer of a Parylene Labcoater (PDS2010, Specialty Coatings and Systems, Indianapolis, IN, USA). Parylene-C dimer was first vapor- ized at 175 ◦C and then pyrolyzed into a reactive-monomer vapor at 690 ◦C. The reactive- monomer vapor was diffused onto the Pentacene/Au/Cr/SiO2/p-Si structure attached to a rotating platform in a vacuum chamber maintained at 28 mTorr pressure. Finally, a 150-nm-thick circular disk of Au was sputtered on top of the Parylene-C insulating layer using a shadow mask to form the gate of an Au/Parylene-C/Pentacene capacitor above the Au/Cr/SiO2/p-Si bottom structure. The area of the circular disk was set as 7.1×10−2 cm2. Capacitance-Voltage (C-V) measurements were carried out using a Precision LCR Meter (HP 4284, Hewlett-Packard, Palo Alto, CA, USA), while the parallel mode of 'C-D' option was selected. These measurements were made at 100 kHz frequency and room temperature with an applied gate voltage Vg ∈ [−2, 2] V and an oscillating voltage signal Vac = 24 mV. Using a Semiconductor Parameter Analyzer (HP 4155C, Hewlett-Packard, Palo Alto, CA, USA), we measured the time-dependent leakage current Istress as a function of time t while a stress voltage Vstress ∈ {±10,±15,±20} V was being applied at room temperature. A fresh Au/Parylene-C/Pentacene/Au/Cr/SiO2/p-Si structure was used for every value of Vstress. 3. Results 3.1. Capacitance-voltage characterizations before CVS The measured C-V characteristics of an Au/Parylene-C/Pentacene capacitor at 100 kHz are shown in Fig. 1. As this sample was not subjected to CVS, it served as our control sample. Its C-V characteristics were measured in small voltage sweeps from ±2 V to ∓2 V. The capacitance shows an apparent transition from accumulation to depletion. A small hysteresis is evident, the C-V curve-shift ∆V of 150 mV being very small [22, 23]. An MIS capacitor is generally modeled as two capacitors in series: the insulator ca- pacitance Ci and the semiconductor-depletion-layer capacitance Cs [21, 22, 24]; hence, the capacitance C = . (1) CiCs Ci + Cs 3 Figure 1: Measured C-V characteristics of a Au/Parylene-C/Pentacene capacitor, functioning as a control sample (i.e., Vstress = 0) at 100 kHz and room temperature. The Parylene-C layer is of thickness d = 200 nm and the top-electrode area A = 7.1 × 10−2 cm2. For Vg > 0, the measured capacitance reaches a constant value equivalent to C given by Eq. 1. In contrast, for Vg < 0, the capacitance saturates to a value close to (2) where o = 8.854 × 10−14 F cm−1 is the permittivity of vacuum. The Parylene-C layer is of thickness d = 200 nm and the top-electrode area A = 7.1 × 10−2 cm2. Ci = oκ A d , 3.2. MIS capacitor characterizations under CVS 3.2.1. Capacitance-voltage characterizations after CVS Figure 2 shows the C-V curves measured at 100 kHz in Au/Parylene-C/Pentacene ca- pacitors after the room-temperature application of (a) Vstress ∈ {−10,−15,−20} V and (b) Vstress ∈ {10, 15, 20} V for duration tstress = 10 s. After the 10-s application of Vstress ≷ 0, a C-V curve-shift ∆V ≷ 0 is observed for all three values of Vstress, which suggests a positive (resp. negative) charge buildup in the insulator of the MIS capacitor during the applica- tion of positive (resp. negative) CVS. For the same Vstress and tstress, ∆V is higher for Vstress > 0 than for Vstress < 0. 3.2.2. Time-dependent leakage current under CVS The time-dependent leakage current Istress was measured as a function of t ∈ [1, 11] s during the application of CVS on MIS capacitors for Vstress ∈ {±10,±15,±20} V. The data presented in Fig. 3 show that Istress is larger for larger Vstress . For the same Vstress, Istress is lower by three orders of magnitude for Vstress > 0 than for Vstress < 0. Irrespective of the polarity of Vstress, Istress decays as t increases. 4 Figure 2: Measured C-V characteristics of a Au/Parylene-C/Pentacene capacitor at 100 kHz and room temperature, after the application of (a) Vstress ∈ {−10,−15,−20} V and (b) Vstress ∈ {10, 15, 20} V for tstress = 10 s. 3.2.3. TDDB characterization after CVS The TDDB characteristics of the Au/Parylene-C/Pentacene capacitors were obtained by recording the current-voltage (I-V) response before and after the 10-s application of Vstress > 0. Before CVS, the I-V curve in Fig. 4(a) shows a dielectric breakdown occurs at an electric field of 1.62 MV cm−1. This value is comparable to the values of the breakdown electric field Ebd in the range 1.9 to 2.2 MV cm−1 reported for 200-nm-thick Parylene-C layers in MIM structures [9, 25], but for a much smaller gate electrode area of about 3×10−4 cm2. In Fig. 4(b), measured values of the breakdown field Ebd and the time-to-breakdown tbd are plotted as functions of Vstress ∈ {5, 10, 15, 20} V. Time-to-breakdown tbd is the time it takes for the Parylene C to breakdown under the application of each Vstress. As expected, both Ebd and tbd decrease as Vstress > 0 increases. The value of Ebd decreases from 1.62 MV cm−1 for the control sample (i.e., Vstress = 0) to 1.04 MV cm−1 for the sample stressed with Vstress = 20 V. Furthermore, tbd decreases from 1005 s for the control sample to 104 s for the sample stressed with Vstress = 20 V. 5 Figure 3: Measured time-dependent leakage current Istress in a Au/Parylene-C/Pentacene capacitor at room temperature as a function of time t ∈ [1, 11] s during CVS application. (a) Vstress ∈ {−10,−15,−20} V and (b) Vstress ∈ {10, 15, 20} V. 4. Discussion The C-V characteristics of organic-based MIS capacitors are limited by contact injection [21, 26, 27, 28, 29]. The C-V curve obtained in Fig. 1 can be explained in terms of charge accumulation arising from injection and contained within the semiconductor layer. For Vg > 0, a thin accumulated layer of injected holes occurs at the Au/Pentacene interface, while the Parylene-C/Pentacene interface is depleted and devoid of any significant free charge carriers. As a result, a depletion layer is created inside the Pentacene. Hence, the capacitance is given by C in Eq. 1. For Vg < 0, an accumulation of holes occurs near the interface of Pentacene/Parylene-C. As Vg further increases to more negative values, Cs increases and C approaches Ci given by Eq. 2. The C-V curve shifts observed in Fig. 2 are attributed to charge buildup ∆Qt in Parylene C [30, 31]. Charges of three different provenances are associated with ∆Qt [30, 31]; i.e., ∆Qt = Qm + Qb + Qi . 6 (3) Figure 4: Room-temperature TDDB characteristics of Au/Parylene-C/Pentacene capacitors: (a) Leakage current measured as a function of gate voltage Vg before CVS application (control sample). (b) Breakdown electric field Ebd and time tbd after application of Vstress ∈ {5, 10, 15, 20} V. Here, Qm is the charge density of mobile positive charges located in the bulk of the insulator and arising from ionic impurities such as Na+. The effect of these charges can be seen as a hysteresis in the C-V curve when sweeping Vg in a negative-positive-negative loop. Also, Qb is the charge density of charges trapped in the bulk insulator. It can be either negative or positive, depending on whether holes or electrons are trapped. Qi is the charge density of charges trapped in the semiconductor/insulator interface. It can also be either negative or positive. Because these charges are trapped at the interface, Qi has the largest effect on ∆V . As can be deduced from Fig. 1, Qm is negligibly small because a very small hysteresis (∆V = 150 mV) is detected as the gate voltage is swept from −2 V to +2 V to −2 V. Hence, Qm plays no role in the charge buildup observed in Parylene C after the application of CVS so that ∆Qt = Qb + Qi . (4) Let us now attempt a quantitative analysis of ∆Qt and its dependence on Vstress and tstress. Accordingly [32], 7 ∆Qt = −∆V Ci . (5) Table 1 provides the values of ∆Qt for Vstress ∈ {±10,±15,±20} V. Clearly, the charge buildup are positive for Vstress < 0 but negative for Vstress > 0. Furthermore, for fixed Vstress, the charge buildup is more than twice in magnitude for Vstress > 0 than for Vstress < 0. A Table 1: C-V curve-shift ∆V of and the charge buildup ∆Qt in a Au/Parylene-C/Pentacene MIS capacitor subjected to Vstress ∈ {±10,±15,±20} V for tstress = 10 s. −10 −0.35 −15 −0.56 ∆V (V) Vstress (V) −20 −0.89 +10 +0.76 +15 +1.37 +20 +2.10 ∆Qt (C cm−2) +4.10×10−9 +6.50×10−9 +1.04×10−8 −8.90 × 10−9 −1.60 × 10−8 −2.45 × 10−8 During the time that Vstress < 0, electrons are injected from the gate and holes are injected from the layer of accumulated holes near the Parylene-C/Pentacene interface. Electrons and holes transiting the Parylene-C layer can be trapped at defect sites and give rise to charge buildup. It is apparent from Table. 1 that hole trapping dominates and the net charge buildup is positive for Vstress < 0. In contrast and during the time that Vstress > 0, holes are only injected from the gate into Parylene-C. The resulting hole-leakage current is observed to be much smaller than Istress shown in Fig. 3(a). This is because a significant portion of the applied Vstress is dropped across the depleted (i.e., devoid of charge carriers) layer near the Pentacene/Parylene C interface; much less hole transport takes place across Pentacene for Vstress > 0. However, the charge-buildup sign is negative, which may indicate that the observed charge buildup is not entirely due to the trapping of charge carriers (electrons and holes) but could also be caused by defect generation. Presumably, these generated defects are electrons traps that are populated during the application of Vstress. The I-V curve in Fig. 4(a) shows a dielectric breakdown occurs at an electric field of 1.62 MV cm−1. This breakdown may indicate the formation of a defect-related conduction path [33, 34]. In other words, a higher applied voltage could induce defects that eventually form different conducting paths from the gate to the semiconductor in the Au/Parylene- C/Pentacene structure. As shown in Fig. 4(b), the decrease in Ebd suggests that more/longer conductive paths are formed with increasing Vstress > 0. Conductive paths result from defect-generation pro- cesses and, hence, more defects are presumably generated as Vstress further increases to higher values. This deduction is in agreement with earlier inference from Table 1 that defect generation dominates over charge trapping during Vstress > 0. 8 5. Concluding Remarks A systematic analysis of the reliability of Au/Parylene-C/Pentacene MIS capacitors un- der constant-voltage stress was performed, with focus on the effects of CVS on the stability of Parylene C as a gate dielectric. 200-nm-thick Parylene-C thin films were utilized as gate- dielectric layers of Au/Parylene-C/Pentacene MIS capacitors. Measurements and analysis of the C-V curve-shift, the time-dependent leakage current, and the time-dependent dielectric breakdown were performed before and after application of CVS. Positive and negative stress voltages of the same magnitude were applied for 10-s duration. A summary of our results is as follows: • The MIS capacitance shows an apparent transition from accumulation to depletion. • The C-V curve-shift is higher for positive stress voltage than for negative stress voltage of the same magnitude. • The time-dependent leakage current for positive stress voltage is three orders of mag- nitude lower than for negative stress voltage. • The charge density of trapped charges increases with the stress voltage, the polarity of the trapped charges being opposite to the polarity of the stress voltage. • As the application of CVS increases, the breakdown voltage decreases as does the time to breakdown. Therefore, our main conclusions are as follows: • The C-V characteristic can be explained in term of accumulation charges within the semiconductor layer. • This accumulation charge is due to contact injection. • Inside the insulating layer,the charge buildup resulting from the accumulation of trapped charges affects the stability of the the MIS capacitor by shifting its C-V curve. • The shift of the C-V curve is attributed to the trapping and recombination of electrons and holes inside Parylene C and its interface with Pentacene. • The dielectric-breakdown mechanism is defect dominated. Overall, the buildup of trapped charges in the Parylene-C layer and near the Parylene- C/Pentacene interface plays a major role in the degradation of Au/Parylene-C/Pentacene capacitors. Our analysis in this paper provides a first-level understanding of the charge buildup in Au/Parylene-C/Pentacene capacitors and, perhaps, will serve as the basis of future studies on the defect-generation process and the trapping of charge carriers within the insulator layer in OFETs. 9 Acknowledgment We thank Prof. Chris Giebink for providing access to the vacuum thermal evaporation tool in the Applied Optoelectronics & Photonics Lab at the Pennsylvania State Univer- sity. AL is grateful to the Charles Godfrey Binder Endowment at the Pennsylvania State University for ongoing support of his research activities. References References [1] T. Marszalek, M. Gazicki-Lipman, and J. Ulanski, Beilstein J. Nanotechnol. 8, 1532 (2017). [2] W.H. Lee, H.H. Choi, D.H. Kim, and K. Cho, Adv. Mater. 26, 1634 (2014). [3] I.H. Khawaji, C. Chindam, O.O. Awadelkarim, and A. Lakhtakia, IEEE Trans. Electron Dev. 64, 3360 (2017). [4] I.H. Khawaji, C. Chindam, O.O. Awadelkarim, and A. Lakhtakia, Flex. Printed Electron. 2, 045012 (2017). [5] I. H. Khawaji, O. O. Awadelkarim, and A. Lakhtakia, IEEE Trans. Dielectr. Electr. Insul. 26, 270 (2019). [6] C. Chindam, N.M. Wonderling, A. Lakhtakia, O.O. Awadelkarim, and W. Orfali, Appl. Surf. Sci. 345, 145 (2015). [7] C. Chindam, A. Lakhtakia, O.O. Awadelkarim, and W. Orfali, J. Appl. Phys 116, 134905 (2014). [8] J.-M. Hsu, L. Rieth, R.A. Normann, P. Tathireddy, and F. Solzbacher, IEEE Trans. Biomed. Eng. 56, 23 (2009). [9] E.-Y. Shin, E.-Y. Choi, and Y.-Y. Noh, Org. Electron. 46, 14 (2017). [10] X. Xie, L. Rieth, L. Williams, S. Negi, R. Bhandari, R. Caldwell, R. Sharma, P. Tathireddy, and F. Solzbacher, J. Neural Eng. 11, 026016 (2014). [11] Bulk Parylene C properties: https://scscoatings.com/docs/broch- ures/parylene properties.pdf (Ac- cessed on 6 December 2018). [12] J. Jakabovic, J. Kov´ac, M. Weis, D. Hasko, R. Srn´aek, P. Valent, and R. Resel, Microelectron. J. 40, 595 (2009). [13] P. Tewari, R. Rajagopalan, E. Furman, and M.T. Lanagan, J. Colloid Interface Sci. 332, 65 (2009). [14] B. Park, K.-J. Im, K. Cho, and S. Kim, Org. Electron. 9, 878 (2008). [15] C. Sheraw, D. Gundlach, and T. Jackson, Mater. Res. Soc. Symp. Proc. 558, 403 (1999). [16] V. Podzorov, V. M. Pudalov, and M. E. Gershenson, Appl. Phys. Lett. 82, 1739 (2003) [17] T. Yasuda, K. Fujita, H. Nakashima, and T. Tsutsui, Jpn J. Appl. Phys. 42, 6614 (2003). [18] I. N. Hulea, S. Fratini, H. Xie, C. L. Mulder, N. N. Iossad, G. Rastelli, S. Ciuchi, and A. F. Morpurgo, Nat. Mater. 5, 982 (2006). [19] N. Kawasaki, W. L. Kalb, T. Mathis, Y. Kaji, R. Mitsuhashi, H. Okamoto, Y. Sugawara, A. Fujiwara, Y. Kubozono, and B. Batlogg, Appl. Phys. Lett. 96, 113305 (2010). [20] K. Fukuda, T. Suzuki, D. Kumaki, and S. Tokito, Phys. Status Solidi A, 209, 2073 (2012). [21] A. Nigam, M. Premaratne, and P. R. Nair, Org. Electron. 14, 2902 (2013). [22] D.A. Neamen, Semiconductor Physics and Devices: Basic Principles (McGraw -- Hill, New York, NY, 2012). [23] C.C. Hu, Modern Semiconductor Devices for Integrated Circuits (Prentice -- Hall, Upper Saddle River, NJ, 2010). [24] W. Wondmagegn, N. Satyala, I. Mejia-Silva, D. Mao, S. Gowrisanker, H. Alshareef, H. Stiegler, M. Quevedo-Lopez, R. Pieper, and B. Gnade, Thin Solid Films 519, 4313 (2011). [25] S. Gowrisanker, M. Quevedo-Lopez, H. Alshareef, and B. Gnade, Org. Electron. 10, 1024 (2009). 10 [26] A. Nigam, P. R. Nair, M. Premaratne, and V. R. Rao, IEEE Electron Dev. Lett. 35, 581 (2014). [27] L. Diao, C. D. Frisbie, D. D. Schroepfer, and P. P. Ruden, J. Appl. Phys. 101, 014510 (2007). [28] G. Horowitz, J. Mater. Res. 19, 1946 (2004). [29] H. Klauk, Chem. Soc. Rev. 39, 2643 (2010). [30] D.K. Schroder, Semiconductor Material and Device Characterization (IEEE Press, Piscataway, NJ, 2006). [31] M. Dom´ınguez-Pumar, C.R. Bheesayagari, S. Gorreta, G. L´opez-Rodr´ıguez, I. Martin, E. Blokhina, and J. Pons-Nin, IEEE Trans. Ind. Electron. 64, 3023 (2017). [32] D.B. Farmer and R.G. Gordon, J. Appl. Phys. 101, 124503 (2007). [33] N. Raghavan, K.L. Pey, and K. Shubhakar, Microelectron. Reliability 54, 847 (2014). [34] J. Sun´e, G. Mura, and E. Miranda, IEEE Electron Dev. Lett. 21, 167 (2000). 11
1912.06026	1	1912	2019-12-12T15:24:40	Bioinspired Materials with Self-Adaptable Mechanical Properties	[ "physics.app-ph", "cond-mat.mtrl-sci", "cond-mat.soft", "nlin.AO" ]	Natural structural materials, such as bone and wood, can autonomously adapt their mechanical properties in response to loading to prevent failure. They smartly control the addition of material in locations of high stress by utilizing locally available resources guided by biological signals. On the contrary, synthetic structural materials have unchanging mechanical properties limiting their mechanical performance and service life. Here, a material system that autonomously adapts its mechanical properties in response to mechanical loading is reported inspired by the mineralization process of bone. It is observed that charges from piezoelectric scaffolds can induce mineralization from media with mineral ions. The material system adapts to mechanical loading by inducing mineral deposition in proportion to the magnitude of the loading and the resulting piezoelectric charges. Moreover, the mechanism allows a simple one-step route for making graded materials by controlling stress distribution along the scaffold. The findings can pave the way for a new class of self-adaptive materials that reinforce the region of high stress or induce deposition of minerals on the damaged areas from the increase in stress to prevent/mitigate failure. They can also contribute to addressing the current challenges of synthetic materials for load-bearing applications from self-adaptive capabilities.	physics.app-ph	physics	Bioinspired Materials with Self-Adaptable Mechanical Properties Santiago Orrego1,2,3, Zhezhi Chen,1,2, Urszula Krekora4, Decheng Hou1,2, Seung-Yeol Jeon1,2, Matthew Pittman1, Carolina Montoya3, Yun Chen1, Sung Hoon Kang1,2,5* 1Department of Mechanical Engineering, Johns Hopkins University, Baltimore, 21218, USA. 2Hopkins Extreme Materials Institute, Johns Hopkins University, Baltimore, 21218, USA. 3Department of Oral Health Sciences and Bioengineering Department, Temple University, Philadelphia, 19140, USA. 4Deparment of Chemical & Biomolecular Engineering, Johns Hopkins University, Baltimore, 21218, USA. 5Institute for NanoBioTechnology, Johns Hopkins University, Baltimore, 21218, USA. E-mail: [email protected] (S. H. Kang) Keywords: bioinspired materials, self-adaptive, mineralization, piezoelectric, graded materials Natural structural materials, such as bone and wood, can autonomously adapt their mechanical properties in response to loading to prevent failure. They smartly control the addition of material in locations of high stress by utilizing locally available resources guided by biological signals. On the contrary, synthetic structural materials have unchanging mechanical properties limiting their mechanical performance and service life. Here, a material system that autonomously adapts its mechanical properties in response to mechanical loading is reported inspired by the mineralization process of bone. It is observed that charges from piezoelectric scaffolds can induce mineralization from media with mineral ions. The material system adapts to mechanical loading by inducing mineral deposition in proportion to the magnitude of the loading and the resulting piezoelectric charges. Moreover, the mechanism allows a simple one-step route for making graded materials by controlling stress distribution along the scaffold. The findings can pave the way for a new class of self-adaptive materials that reinforce the region of high stress or induce deposition of minerals on the damaged areas from the increase in stress to prevent/mitigate failure. They can also contribute to addressing the current challenges of synthetic materials for load-bearing applications from self- adaptive capabilities. 1 The mechanical efficiency of natural (or biological) structural materials is remarkable. Self-optimized architectures,[1] outstanding mechanical properties with weak constituents,[2] hierarchical structuring,[3] and self-adaptability of stiffness in response to external stimuli[4] are just a few examples of inspiring features that should be included within synthetic structural materials. They provide inspiration for potential solutions to address challenges associated with traditional approaches of material selection.[5] In the classical approach, a material is selected for a specific application based on the expected loading conditions, design objective and constraints, and databases of material properties with fixed values (e.g., Ashby method[6]). This methodology does not account for unpredictable loading conditions. To address this issue, safety factors are added, which increases the associated costs and weight of the structure; thus, reduces its mechanical efficiency. A potential solution for this dichotomy is to have materials with self-adaptable properties responding to loading conditions. This feature can result in materials with improved mechanical efficiency and reduction in costs, resources and environmental impact. So far, there are few synthetic materials capable of increasing stiffness in response to external stimuli. For example, Capadona et al. and Ramirez et al. reported polymers that changed elastic moduli by changing the degree of crosslinking upon chemical, thermal or mechanical stimuli.[7,8] Agrawal et al. reported an increase of the elastic modulus of liquid crystal elastomers during cyclic mechanical stress due to the alignment of nematic directors.[9] Recently, a "self-growing" hydrogel responding to repetitive mechanical stress was reported through mechanochemical transduction.[10] Despite progress there are limited synthetic material systems adapting mechanical properties to external mechanical stimuli. Challenges of current self-stiffening materials include limited load-bearing capability, difficulty of material synthesis, high cost, lack of or limited biocompatibility, and need of additional energy for material property changes.[11] 2 Natural structural materials have resolved these challenges by taking different strategies.[12] Bone,[13] wood,[14] fish scales,[15] and coral reefs[16] are dynamic structural materials capable of self-adapting their mechanical properties in response to external loading to improve their mechanical efficiency and prevent failure.[13,17] Introducing these features to synthetic material systems has been challenging[18] since it requires active maintenance of living organisms. These natural materials utilize resources available in the environment to regulate their properties through mechanobiological mechanisms.[16,19] For example, bone and coral reefs utilize cellular signals to control the addition and removal of minerals harvested from surrounding media (e.g. blood or seawater) at specific locations.[20] During the process, organic matrices serve as templates for mineral growth. Specifically, it is reported that negatively charged carboxyl groups can bind calcium ions and induce nucleation of biominerals.[21] Inspired by these findings, there have been numerous reports[22] on mineral growth using negatively charged surfaces obtained by using electric field[23] or self-assembled monolayers.[24] However, previously synthesized minerals cannot change their mechanical properties upon external loading. Here we report a synthetic material system that can change its mechanical behaviors in response to external loading conditions inspired by natural mineralization processes in bone and coral reefs. We hypothesized that if we have a scaffold that can generate charges proportional to an external mechanical stimulus, the charges can work as signals to induce mineralization from resources (e.g. mineral ions) available in surrounding media. As a result, the material system can exhibit self-adapting mechanical behaviors. Our material system uses a new mechanism to realize self-stiffening materials and overcomes limitations of current synthetic materials with changing mechanical properties from its enhanced load-bearing capability, low cost, simple synthesis, no need of extra energy, and biocompatibility. To test our hypothesis, we utilized piezoelectric materials as scaffolds because they convert mechanical forces into electrical charges while other materials (e.g. triboelectric, 3 flexoelectric materials) could be utilized. We immersed piezoelectric polymer films (PVDF, (CH2CF2)n), d33~30 pC/N) with oppositely charged surfaces in a simulated body fluid (SBF)[25] mimicking the ionic concentrations of human blood plasma (Figure 1a). After one week of SBF incubation, we observed the minerals formed on the surfaces of the films using a scanning electron microscope (SEM). Minerals were preferentially deposited onto the negatively charged surfaces (Figure 1b). This was consistent with previous study.[23] To characterize chemical compositions of the formed minerals, we conducted energy dispersive X-ray spectroscopy (EDX) and X-ray diffraction (XRD) measurements. EDX results showed peaks of Ca, P, and O, which are elements consisting of hydroxyapatite (HAp) (Figure S2). XRD analysis showed the characteristic diffraction peaks of HAp (Figure 1d). We also observed similar mineral deposition behavior on different substrates including piezoelectric composites (See Supporting Information S3 for details). Moreover, we found that we can control the type of formed minerals by controlling the ion composition of the media (See Supporting Information S4). Refreshing the SBF solution everyday led to the formation of HAp minerals (Figure 1d), whereas using non-refreshed SBF led to a combination of Calcite and HAp (Figure 1c). The minerals found in bone (HAp) and coral reefs (Calcite) originate from blood and seawater, respectively.[21] Metallic medical devices are typically coated with these type of minerals to enable biocompatibility.[26] Usually these coatings wear out and do not regenerate. To understand the dynamics of mineralization, we conducted quantitative studies by comparing the mineral thickness at different conditions (Figure 2). First, we studied the effect of the polarity of the charged surface on the mineral formation rate (see Supporting Information S5). We found that the negatively charged surface showed about an order of magnitude faster mineral formation rate than the positively charged surface (Figure 2b). The mineral growth rate (~6.4 µm/day) was comparable to that reported by Yamashita et al., which was obtained in polarized negative surfaces.[23] We also studied the effect of the concentration of the SBF on mineral formation rate. We found higher concentrations of calcium in SBF resulted in faster 4 rates (Figure 2c), which is consistent with previous study that utilized other negatively functionalized surfaces.[27] In our study, we mainly utilized 10× calcium concentration in SBF as it provides faster mineral formation. We also found that we could modulate the mineral growth rate by controlling the external mechanical loading. Higher electrical charges generated by cyclic mechanical loads led to higher quantities of minerals compared to static or no mechanical loads (Figure 2d). Different from previous studies that utilized permanent chemical functionalization,[28] one can utilize "mechanical functionalization" using piezoelectric scaffolds to dynamically control the formation of minerals by modulating the loading condition. As an application of the mineralization mechanism controlled by mechanical stress, we made a graded material by using a simple one-step process. A piezoelectric PVDF film was subjected to a cantilever loading since it offers a gradual increase of stress (σ) from the free end to the fixation point (see Figure 2e, Supporting Information S13). In this beam configuration, the electrical charge changes proportionally to the beam length. We experimentally confirmed higher electrical charges at regions of high stress and vice versa at regions of low stress (Figure 2e) as a piezoelectric charge (Q) is proportional to the applied stress (Q = dσA, d: piezoelectric coefficient, A: area). Then, we submerged a PVDF specimen in SBF under repetitive loading (Supporting Information S7) and measured the amount of mineral along the beam (Supporting Information S8). Our results showed a gradual formation of minerals with higher amounts at regions of high stress (high charge) and vice versa at low stress regions (Figure 2f). This result indicates that our material system could autonomously reinforce regions that experience higher stress with more minerals. The mechanism allows materials or structures to self-adapt to the external loading conditions to minimize failure. We further investigated the correlation between the mechanical stress and the mineral distribution. We used a numerical model to calculate the stresses distribution along the beam and compared with the mineral thickness distribution. Interestingly, the stress distributions at different positions of the specimen showed similar trends as the mineral thickness (Figure 2f and Supporting Information S9). We also obtained 5 the quantitative relation between the mineral height and the charge along the beam as Equation (1) (Supporting Information S13) and found that the mineral height (MH) is approximately !≅#$$%(4.33×10-./0) (1) proportional to the square root of the charge (Q) or stress (σ). (Q: charge in C, MH: mineral height in mm, d: piezoelectric coefficient in pC/N, A: area in mm2). Furthermore, we investigated whether our material system can change its mechanical behavior by inducing mineral formation in response to the external loading akin to bone (Figure 3a). We prepared porous piezoelectric scaffolds (d33~2 pC/N) using electrospinning (Supporting Information S10) inspired by bone's porous architecture. The scaffolds were subjected to cyclic mechanical loading while submerged in SBF solution at room temperature (Supporting Information S11). The applied load and actuator displacement were measured to calculate changes in material property such as Young's modulus. From the measurements, our material system showed self-stiffening behaviors by changing its mechanical properties in response to periodic loadings (Figure 3b). The modulus increased ~20% after 3 days of immersion in SBF. A control sample submerged in deionized water (no mineral ions available in the solution) did not show a significant increase in the mechanical properties over time (Figure 3c). Also, comparison of the stress-strain behaviors of the scaffolds before and after the external stimulation in SBF showed that the mechanical property increased ~30% in modulus and ~100% in toughness (Figure 3c). SEM images confirmed the mineralization of the scaffold that enabled the increase in mechanical property (Figure 3d-e). EDX results confirmed the presence of minerals with traces of calcium, phosphorous, and oxygen (Figure 3f). Fluorescent microscopy images showed the formation of minerals around the fibers (Figure 3i and Supporting Information S12). Moreover, to further test the self-adaptive behavior of the material system, we subjected scaffolds to cyclic mechanical loading with different load magnitudes while submerged in SBF. 6 Before cyclic loading, we verified the different piezoelectric charges under different load magnitudes. We measured the stress-strain responses after samples were submerged in SBF. As expected, the modulus of the material system increased proportional to the magnitude of the load and the resulting piezoelectric charge (Figure 3g-h). Specifically, the modulus increased by ~180% after three days for the maximum 5N cyclic load, further demonstrating the self- adaptive capability of the material system. In summary, we report a synthetic material system capable of changing its mechanical properties depending on external loading conditions by utilizing scaffolds that generate "signals" to induce mineralization from media with mineral ions. The material system allows control of the location and the rate of mineral formation by modulating the loading conditions. Our findings can contribute to addressing current challenges of synthetic materials used for load- bearing applications from self-adaptive capabilities. To make the material system portable, one can utilize porous scaffolds with suitable surface properties[29] for making liquid-infused porous piezoelectric scaffolds inspired by bone, which holds blood within porous matrices. Beyond biominerals used as a model system in this study, we can further expand the material system based on the reported mechanism for broader applications. We envision that the strategy and the mechanism that we report will open new opportunities for technological advancement including smart coatings for bone implants to mitigate mismatch of mechanical properties, scaffolds for accelerating treatment of bone-related disease or fracture, smart resins for dental treatments requiring tissue regeneration, and 4D materials and structures reconfiguring to adapt to loading conditions. They will also contribute to fundamental advances in understanding dynamic behaviors of adaptive materials and structures and dynamics of mineralization induced by mechano-electric coupling. Moreover, built upon recent studies that reported piezoelectric charges could stimulate cell differentiation in various cell types and improve cell adhesion, bone growth and function in endothelial, nerve, and bone cells,[30] the material system can serve 7 as a model system for quantitative understanding of the process with adequate choice of medium and provide guides for facilitating the regeneration. Experimental Section SBF Incubation: Piezoelectric PVDF films (TE Connectivity, PN: 3-1003352-0) were submerged in a sealed polypropylene bottle (Dyn-A-Med, 80094) containing ~200 mL of SBF solution kept at 37 °C for the duration of the experiment (see Figure 1a). Every 24 hours, the sample was taken out of the beaker, gently rinsed with de-ionized water (Milli pore, Milli-D) and put back into a refreshed solution. After the incubation, the sample was repeatedly rinsed gently in a de-ionized water solution to remove salts of NaCl. The sample was taken out and dried at room temperature in air for further analysis including SEM, EDS, XRD, profilometer. Self-Stiffening Experiments: To test the self-stiffening capability of our material system, electrospun PVDF samples were submerged in SBF solutions and were subjected to cyclic mechanical loadings to activate the piezoelectric charge generation. We prepared electrospun PVDF samples of 30×15 mm2. Simulated body fluids (10×SBF) were prepared by following protocols described in Supporting Information S1. While submerged in the solution at room temperature, samples were subjected to cyclic compression loading using an electromechanical universal testing machine (TA Instruments Electroforce 5500 with a 10 N capacity load cell). The stiffness (S) was calculated from the load (P) and displacement (∆δ) amplitudes measured with the actuator using the follow relation: S=∆P/∆δ. The force and displacement signals were monitored and recorded during experiments. To prevent any change of mechanical properties of electrospun films due to viscoelastic and time-dependent behavior during experiments in SBF, we initially subjected samples to cyclic compression loading while submerged in DI water for 3 days. The modulus (E) was calculated by multiplying by the cross section area and the deflection E=S∆δ / A. 8 Acknowledgements We would like to thank Mr. Eugene Kang for the help fabricating the electrospinning apparatus, Mr. Ian McLane for the help and guidance during contact profilometer measurements, and Mr. Prashant Ray for his help during the concentration measurements. This work is supported by the Air Force Office of Scientific Research Young Investigator Program Award (Award number: FA9550-18-1-0073, Program manager: Dr. Byung-Lip (Les) Lee), Johns Hopkins University Whiting School of Engineering start-up fund, and Temple University Maurice Kornberg School of Dentistry start-up fund (PI: Orrego). Any opinions, finding, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the United States Air Force. Conflict of interest: A patent application has been filed on aspects of the described work. References [1] [2] [3] [4] [5] E. Stach, in Text. Compos. Inflatable Struct., Springer, 2005, pp. 285 -- 303. Z. Liu, M. A. Meyers, Z. Zhang, R. O. Ritchie, Prog. Mater. Sci. 2017, 88, 467. P. Fratzl, J. R. Soc. Interface 2007, 4, 637. S. Weiner, L. Addadi, J. Mater. Chem. 1997, 7, 689. K.-I. Jang, H. U. Chung, S. Xu, C. H. Lee, H. Luan, J. Jeong, H. Cheng, G.-T. Kim, S. Y. Han, J. W. Lee, Nat. Commun. 2015, 6, 6566. [6] M. F. Ashby, H. Shercliff, D. Cebon, Materials: Engineering, Science, Processing and Design, Butterworth-Heinemann, 2018. [7] J. R. Capadona, K. Shanmuganathan, D. J. Tyler, S. J. Rowan, C. Weder, Science 2008, 319, 5868. [8] A. L. B. Ramirez, Z. S. Kean, J. A. Orlicki, M. Champhekar, S. M. Elsakr, W. E. Krause, S. L. Craig, Nat Chem 2013, 5, 757. [9] A. Agrawal, A. C. Chipara, Y. Shamoo, P. K. Patra, B. J. Carey, P. M. Ajayan, W. G. Chapman, R. Verduzco, Nat. Commun. 2013, 4, 1739. [10] T. Matsuda, R. Kawakami, R. Namba, T. Nakajima, J. P. Gong, Science 2019, 363, 504. [11] J. F. Patrick, M. J. Robb, N. R. Sottos, J. S. Moore, S. R. White, Nature 2016, 540, 363. 9 [12] M. A. Meyers, J. McKittrick, P.-Y. Chen, Science 2013, 339, 773. [13] S. C. Cowin, Bone Mechanics Handbook, CRC Press, 2001. [14] G. C. Mattheck, Trees: The Mechanical Design, Springer Science & Business Media, 2012. [15] E. A. Zimmermann, B. Gludovatz, E. Schaible, N. K. N. Dave, W. Yang, M. A. Meyers, R. O. Ritchie, Nat. Commun. 2013, 4, 2634. [16] W. C. Dennison, D. J. Barnes, J. Exp. Mar. Bio. Ecol. 1988, 115, 67. [17] B. E. Brown, Adv. Mar. Biol. 1997, 31, 221. [18] U. G. K. Wegst, H. Bai, E. Saiz, A. P. Tomsia, R. O. Ritchie, Nat. Mater. 2015, 14, 23. [19] D. J. Hadjidakis, I. I. Androulakis, Ann. N. Y. Acad. Sci. 2006, 1092, 385. [20] P. Fratzl, R. Weinkamer, Prog. Mater. Sci. 2007, 52, 1263. [21] S. Mann, Biomineralization: Principles and Concepts in Bioinorganic Materials Chemistry, Oxford University Press On Demand, 2001. [22] A.-W. Xu, Y. Ma, H. Cölfen, J. Mater. Chem. 2007, 17, 415. [23] K. Yamashita, N. Oikawa, T. Umegaki, Chem. Mater. 1996, 8, 2697. [24] H. Cölfen, Nat. Mater. 2010, 9, 960. [25] T. Kokubo, H. Takadama, Biomaterials 2006, 27, 2907. [26] Y. Yang, K.-H. Kim, J. L. Ong, Biomaterials 2005, 26, 327. [27] S.-S. Kim, M. S. Park, S.-J. Gwak, C. Y. Choi, B.-S. Kim, Tissue Eng. 2006, 12, 2997. [28] M. Tanahashi, T. Matsuda, J. Biomed. Mater. Res. An Off. J. Soc. Biomater. Japanese Soc. Biomater. 1997, 34, 305. [29] T. Wong, S. H. Kang, S. K. Y. Tang, E. J. Smythe, B. D. Hatton, A. Grinthal, J. Aizenberg, Nature 2011, 477, 443. [30] A. H. Rajabi, M. Jaffe, T. L. Arinzeh, Acta Biomater. 2015, 24, 12. 10 Figure 1. Formation of minerals induced by piezoelectric charges. a) Schematic of the experimental setup. A PVDF piezoelectric film (d33~30 pC/N) was submerged in a simulated body fluid (SBF). The generation of negative charges controlled by external mechanical loading preferentially induced the formation of minerals on the negative side of the film. b) Micrographs of the negative (left) and positive (right) sides of the films showing significant formation of minerals on the negative side. c) The type of formed mineral was controlled by the medium used. Not refreshing the SBF solution everyday led to the formation of mainly Calcite minerals. d) Refreshing the SBF solution everyday led to the formation of mainly HAp minerals. Scale bars are 10 µm. 11 Figure 2. Dynamics of the formation of minerals by piezoelectric charges. a) The amount of mineral formed on films was quantified by measuring the mineral thickness. b) Time dependence test. A maximum of 60 µm of mineral was formed on the negative side after 2 weeks of incubation in SBF. No significant mineral was formed on the positive side. c) SBF concentration dependence test. Higher concentration of the SBF solution led to higher amounts of mineral. d) Mechanical loading dependence test. The higher charge generated by external loading led to higher amounts of mineral. e) Formation of graded materials. A piezoelectric film submerged in SBF and subjected to cantilever loading generated higher charge at regions of high stress (close to the foundation) compared to regions of low stress (low charge generation). f) Mineral was formed proportional to the mechanical stress (gray line) with approximately linear (R2=0.93) distribution form the foundation to the free end (orange line). The orange line is the filtered (1D- median filter) data of the mineral height shown as shades in orange. The error bars were obtained by having one standard deviation from N=5 samples. 12 Figure 3. Self-adaptable mechanical properties. a) Electrospun piezoelectric scaffolds (d33~ 2 pC/N) were submerged in SBF and subjected to controlled repetitive mechanical loading. The external activation of the electrical charges promoted the formation of minerals on the scaffold. The modulus of the mineralized scaffold was monitored for the duration of the experiment. b) The material showed a self-adapting mechanical property by increasing modulus under dynamic loading. A control sample submerged in deionized water (gray line) showed no increase of modulus as a function of time. c) After the repetitive loading experiment, samples were loaded-to-failure under quasi-static conditions to confirm the change of mechanical properties. The modulus (E) and toughness (U) increased ~30 and ~100%, respectively. d) Micrographs of the PVDF scaffold after submerged in SBF showing the formation of minerals, compared to e) the as-prepared scaffold. f) Chemical analysis (EDS) confirmed the formation of calcium phosphate minerals on the PVDF scaffold. g) To show the adaptability of mechanical properties, PVDF scaffolds were subjected to different magnitudes of mechanical stress and resulting electrical charge. h) The mechanical modulus increased proportional to the magnitude of the applied load during incubation. i) Confocal laser scanning microscopy showed the formation of minerals around the PVDF template. j) The average thickness of the formed minerals was 0.39 µm. The error bars were obtained by having one standard deviation from N=5 samples. 13 Supporting Information Bioinspired Materials with Self-Adaptable Mmechanical Properties Santiago Orrego, Zhezhi Chen, Urszula Krekora, Decheng Hou, Seung-Yeol Jeon, Matthew Pittman, Carolina Montoya, Yun Chen, Sung Hoon Kang Table of Contents • Supporting Information S1: Mineral Incubation Procedure • Supporting Information S2: SEM, EDS and XRD Evaluations • Supporting Information S3: Mineralization on Porous Piezocomposites • Supporting Information S4: Ion Concentration Measurements of Simulated Body Fluids • Supporting Information S5: Mineral Thickness Measurements • Supporting Information S6: Piezoelectric Coefficient Measurements • Supporting Information S7: Graded Material • Supporting Information S8: 3D Optical Profilometer Measurements • Supporting Information S9: Numerical Model • Supporting Information S10: Electrospinning of Polyvinylidene Fluoride (PVDF) • Supporting Information S11: Self-Stiffening Experiments • Supporting Information S12: Confocal Microscopy of Mineralized Electrospun Scaffolds • Supporting Information S13: Analytical Model • References 14 Supporting Information S1: Mineral Incubation Procedure Commercial piezoelectric PVDF films (TE Connectivity, PN: 3-1003352-0) were cut into 2×4 cm2 specimens and hanged vertically in a sealed polypropylene bottle (Dyn-A-Med, 80094) containing ~200 mL of SBF solution (Table S1) kept at 37 °C for the duration of the experiment (see Figure 1a). Every 24 hours, the sample was taken out of the beaker, gently rinsed with de-ionized water (Milli pore, Milli-D) and put back into a refreshed solution. After the incubation, the sample was repeatedly rinsed gently in a de-ionized water solution to remove salts of NaCl. The salinity of the rinsing solution was measured by a salinity meter (Extech, EC170) and after reaching a magnitude below 0.1 ppt, the sample was taken out and dried at room temperature in air for further analysis. To study the influence of ion concentration on mineralization induced by piezoelectric charges, different stable Calcium-saturated solutions were utilized including 1×,[1] 1.5×,[2] 5×[3] and 10×[4] SBF. The solutions were prepared as indicated by dissolving reagent-grade as received materials of NaCl, NaHCO3, KCl, K2HPO4·3H2O, MgCl2·6H2O, CaCl2 and Na2SO4 (All from Sigma-Aldrich) in the deionized water in accordance with the following ion concentrations (Table S1). 1×SBF and 1.5×SBF are further buffered with trishydroxymethylaminomethane (Tris) and HCl until pH=7.4 at 37 ºC in an incubator as indicated in the recipe. Table S1. Ion concentrations of 1×, 1.5×, 5× and 10× SBF Solutions (mM). Na+ K+ Ca2+ Mg2+ Cl- HCO3- HPO42- SO42- 1×SBF 142.0 5.0 2.5 1.5×SBF 142.0 5.0 3.75 5×SBF 714.8 - 12.5 1.5 1.5 7.5 148.8 148.8 723.8 4.2 4.2 21 1.0 1.5 5 10×SBF 1030.0 5.0 25.0 5.0 1065.0 10.0 10.0 0.5 0.5 - - 15 The change in pH of the SBF solutions was monitored. A pH meter (Extech PH220-C) was calibrated using a 3-point method including acid (pH=4.01), neutral (pH=7.00) and alkaline (pH=10.00) buffers. The probe was dipped in the solution and the pH value recorded at different time durations. Figure S1 shows the change of a 10×SBF solution during incubation process of a PVDF piezoelectric film. H p 6.4 6.2 6.0 5.8 5.6 5.4 0 10×SBF 400 600 200 Time (min) 800 Figure S1. Change of pH for a 10×SBF solution kept at 36 °C during the incubation of minerals on a piezoelectric PVDF film for 12 hours. Supporting Information S2: SEM, EDS and XRD Evaluations A scanning electronic microscope (Tescan, Mira3) was used to observe the morphology of the formed minerals at an accelerating voltage of 10 kV. Figure S2a shows a representative layer of minerals formed on a PVDF film after 7 days of incubation in 10×SBF. Chemical and structural evaluations of the newly formed minerals were conducted by energy dispersion of X- ray (EDS; EDAX Co., USA, Octane Plus) analysis and powder X-ray diffraction (XRD; Philips, X'Pert Pro). Figure S2b shows the chemical evaluation of the mineral layer showed in Figure S2a. Clear peaks of calcium and phosphorus were observed with a ratio of Ca/P=1.47. Then, mineral powders were scratched off from a dried sample using a razor blade for XRD. XRD measurements were conducted with 10°-65° incident angle and step size of 0.02° using CuKα radiation. A typical response of XRD results is shown in Figure S2c including a comparison 16 with standard hydroxyapatite (HAp) readings,[5] which showed good agreement between peaks of formed minerals and those from standard HAp. a) b) 2000 c) Formed mineral Standard HAp (ICDD 9-432) ) 1 1 2 ( ) 2 0 0 ( ) 0 1 3 ( ) 3 1 2 ( ) 4 0 0 ( ) 2 2 2 ( 400 300 200 100 ) s p c ( y t i s n e t n I 1500 1000 500 O 0 0 5 μm P Ca Ca . ) . u a ( y t i s n e t n I 2 1 4 Energy (keV) 3 0 10 20 5 30 40 2θ (°) 50 60 Figure S2. Mineralization of a piezoelectric PVDF film. a) SEM image of minerals formed on a negative side after 7-day incubation in 10×SBF with a 30° tilt view. b) Chemical evaluation of the mineral layer in the SEM image by EDS. c) XRD measurement of minerals from the specimen shown in the SEM image comparing with minerals formed on a PVDF film from the ICDD database. Supporting Information S3: Mineralization on Porous Piezocomposites Piezoelectric composites were utilized to check whether mineralizations induced by piezoelectric charges occur in a different material system. We employed a porous piezoelectric composite comprised of a PDMS matrix, carbon nanotubes and BaTiO3 fillers. The sample was prepared by utilizing a combination of the fabrication methods described in the work by Li et al. [6] and Chen et al.[7] Briefly, the piezocomposite samples were prepared via solvent-casting by thoroughly mixing PDMS (base material:crosslinking agent = 10:1 by weight) (Sylgard 184, Dow Chemical) with BaTiO3 nanoparticles (diameter: ∼200 nm, US3830, US Research Nanomaterials, Inc.) as 15 wt% of PDMS, carbon nanotubes (OD: 10-30 nm, US4500, US Research Nanomaterials, Inc.) as 1 wt% of PDMS and sacrificial NaCl microparticles (Sigma Aldrich S7653) as 26.5 wt% of PDMS. To thoroughly mix the compounds, a revolutionary mixer (KK-400W, Mazerustar) was utilized for 360 seconds. The mass ratio between BaTiO3, 17 NaCl and PDMS was adjusted to create polymer composites with different porosities. The mixture was then cured in an oven (Lindberg Blue M, Thermo Fisher Scientific) at 60 °C for 6 hours. After 24 hours, the solidified composite was immersed in water overnight to completely remove the NaCl grains. Microscale pores were then created. The synthesized microporous polymer composite was then cut with a razor blade into any shape and size according to the experimental requirements. Figure S3a shows a representative sample at the macroscale. The sample is soft and can allow high bending curvature without failure. The microstructure of a pristine sample is shown in Figure S3b. The average pore size is ~400 µm diameter. To improve the elcetromechanical output, samples were subjected to corona poling for 3 hours at 20 kV and 120 °C.[6] The measured piezoelectric coefficient of the porous piezocomposite is d33~30 pC/N and obtained as explained in Supporting Information S6. After fabrication, the piezocomposite was incubated in 10×SBF for 7 days under cyclic mechanical loading as explained in Supporting Information S1. After incubation, the sample was cut in half using a razor blade, and the fractured surface was evaluated in SEM and EDS as explained in Supporting Information S2. Figure S3c shows the fracture surface on the center of the sample after 7-day incubation under cyclic mechanical stimulation. The formation of the minerals inside the pores can be observed, which are denoted with white arrows. Chemical analysis of these sample area confirmed the presence of calcium phosphate minerals and peaks of Si corresponding to the PDMS matrix (Figure S3e). A higher magnification view in one of the pores can clearly show the minerals (see Figure S3d). The corresponding chemical analysis confirmed the presence of calcium and phosphorus corresponding to the ions available in the SBF solution (Figure S3f). 18 a) b) c) 500 μm d) e) 3000 ) s p c ( y t i s n e t n I 2000 1000 Si P O Na Ca 1 4 Energy (keV) 3 2 Si 0 3000 0 2000 1000 O Na P Cl Ca f) ) s p c ( y t i s n e t n I 500 μm 10 μm 0 0 2 1 4 Energy (keV) 3 5 5 Figure S3. Mineralization of porous piezoelectric composites. a) Macroscale view of the sample. b) SEM image of the cross-section of the as-prepared porous piezoelectric composite showing porous microstructure. c) Cross-section view of the porous microstructure after 7-day incubation in 10×SBF and under cyclic mechanical loading showing mineral formation inside the pores as marked by white arrows. d) Higher magnification view of the mineral. e) Chemical (EDS) evaluation of the section view in c) showing peaks of calcium, phosphorus and oxygen corresponding to the newly formed minerals, and peaks of Si corresponding to the PDMS matrix. f) Chemical (EDS) analysis of the section view in d) showing peaks of calcium, phosphorus and oxygen corresponding to minerals and peaks of Si corresponding to the PDMS matrix. Supporting Information S4: Ion Concentration Measurements of SBF To study the potential change of calcium and phosphate ions present in SBF over time as a result of mineral formation, we conducted ion concentration studies in the solutions. Calcium concentration in the SBF solution was measured using the protocol outlined by Gindler and King.[8] First, stock solutions of dye and base were prepared. The stock dye solution was prepared by combining 0.018 grams of methylthymol blue sodium salt (CAS 1945-77-3), 0.720 19 grams of 8-Hydroxyquinoline (Sigma Aldrich, CAS 148-24-3), and 1.00 mL of 12M HCl (CAS 7647-01-0) in 100 mL in deionized water. The stock base solution was prepared by combining 2.40 g of sodium sulfite (Fisher Scientific, CAS 7757-83-7) and 22.0 mL of monoethanolamine (Sigma Aldrich, CAS 7757-83-7) in 100 mL of deionized water. Working reagent was made by mixing equal amounts of stock dye solution and stock base solution. First, calibration measurements were done using the solutions containing known concentration of calcium. For each solution with known concentrations of calcium, 0.050 mL was mixed with 3.00 mL of working solution inside of a cuvette (Sigma Aldrich C5677). Using a UV-Vis photo spectrometer (LAMBDA 950, Perkin Elmer), absorbance values were measured from each sample. After confirming that the absorbance peak was at or near 612 nm, the absorbance reading was taken for each sample. After measuring absorbances of solutions with known concentrations treated with this assay, the relationship between calcium concentration and absorbance at 612 nm was found and described in Equation S1): 3=567.0-- 7.78$ (S1) where C is calcium concentration in mmol/L and A is absorbance. This relationship was used to estimate the calcium concentration values in SBF solutions from absorbance readings. To test calcium concentrations in SBF buffers, 0.050 mL of SBF solution was mixed with 3.00 mL of working solution in a cuvette. Calcium concentrations were correlated to absorbance reading using the calibration curve and Equation S1. The change of calcium concentration in an SBF solution is presented in Figure S4. Phosphate concentrations in solution were measured using the protocol outlined by Karl and Tien.[9] Stock potassium antimonyl tartrate solution was prepared by dissolving 0.2743 g of potassium antimonyl tartrate (Sigma Aldrich CAS 331753-56-1) in 100 mL of deionized water. Mixed reagent was prepared by combining 25 mL of 2.5M H2SO4 (Alfa Aeser CAS 7664-93-9) 0.300 grams of ammonium molybdate (Sigma Aldrich CAS 13106-76-8), 0.264 20 grams of ascorbic acid (Alfa Aesar CAS 50-81-7), and 2.5 mL of potassium antimonyl tartrate solution in a 50 mL volumetric flask and filling the remainder with deionized water and mixing by shaking until all solids dissolved. Similar to the case of the calcium ion concentration measurement, a calibration curve was obtained by using solutions containing known concentrations of phosphate ions. For each sample with known concentrations of phosphate, 4 mL of sample, 800 µL of mixed reagent, and 200 µL of deionized water were mixed in a cuvette. Using a UV-Vis photo spectrometer, absorbance readings were taken from each sample. For each sample with unknown concentration of phosphate, 4 mL of sample, 800 µL of mixed reagent, and 200 µL of deionized water was mixed in a cuvette. The absorbance readings were taken at 840 nm. Phosphate concentrations were correlated to absorbance reading using the calibration curve. After measuring absorbances of solutions with known concentrations treated with this assay, the relationship between calcium concentration and absorbance at 840 nm was found to be Equation 9=5:7.700 7.0;0 (S2) S2. where P is Phosphate concentration in mmol/L and A is absorbance. This relationship was used to calculate the phosphate concentration values of unknown solutions of SBF from absorbance readings. The change of phosphate concentration in an SBF solution is presented in Figure S4. Error was calculated using the following formula:[10] <==>(?=?5)0 <50 (S3) where SC is the calculated error in concentration measurements and SA is the standard deviation of the absorbance readings. For calcium concentration calculations, this simplified to Equation S4. (S4) <==> ABC7.78$C 21 For phosphate concentration calculations, this simplified to Equation S5. <==> ABC7.0;0C (S5) ) M m ( n o i t a r t n e c n o C m u i c l a C 30 20 10 0 0 Ca2+ 3- PO4 5 10 15 Time (days) 2.0 1.5 1.0 0.5 0.0 20 P h o s p h a t e C o n c e n t r a t i o n ( m M ) Figure S4. Change in concentration of calcium and phosphate ions over extended period of time in an SBF solution without being refreshed and maintained at 36 °C. The error bars are obtained by having one standard deviation from N=5 samples. Supporting Information S5: Mineral Thickness Measurements We utilized a contact stylus profilometer to quantify the height (thickness) of the new mineral layer formed on the piezoelectric films of PVDF. Minerals formed on the films in 10×SBF for different duration of times and mechanical loading conditions (see Supporting Information S1) were measured. The external edges of the sample were masked with Scotch® tape (7510000822520). After incubation period, the tape was gently de-bonded exposing the PVDF film. A step between the piezoelectric substrate and the newly formed mineral was created. The stylus of surface profilometer was placed on the mineral phase. Thickness measurements were performed on the mineral layer deposited on PVDF using the Dektak 3 surface profilometer with a 0.8 mm cutoff length according to ANSI standard B46.1. The step- height was measured by calculating the difference between the peak and valley (Ry) of the surface profile. Several profiles were measured along different locations of the mineral layer to 22 calculate an average step-height per sample. The time-dependent thickness measurements are summarized in Figure 2b. The similar method was used for the SBF concentration dependence study as summarized in Figure 2c. To show the influence of mechanical loading on mineralization rates, we conducted incubation in 10×SBF subjecting PVDF films to three different configurations including flat (no stress), fully rolled under static loading condition, and cantilevered bending under cyclic loading (see Supporting Information S7). We measured the mineral thickness for each configuration using the surface profilometer by following the aforementioned procedure as summarized in Figure 2d. Supporting Information S6: Piezoelectric Coefficient Measurements We measured the electromechanical response of the materials utilized in this investigation (i.e. commercial PVDF film, electrospun PVDF and piezoelectric composite) based on the Berlincourt's method.[11] Piezoelectric samples were sandwiched between compression platens covered with conductive carbon tape used as electrodes. The electrodes were wired to a charge amplifier (Piezo Film Lab Amplifier P/N 1007214, Measurement Specialties, Virginia). The amplifier was operated in Charge-Mode (100 pF) and bandwidth filter configured at 1 Hz and 10 Hz. Samples were subjected to cyclic compression loads (P) using an electromechanical universal testing machine (TA Instruments ElectroForce 5500). The load amplitude and frequency, and preload were adjusted to evaluate the electromechanical performance under different load conditions. The load cell and the charge amplifier signals were connected to a data acquisition system (NI-PCI 6251 and BNC-2110, National Instruments, Texas). Data was acquired with a sample rate of 1000 Hz ensuring our measures generated a sufficient representation of the acquired load and electrical charge signals. A schematic showing the wiring and configuration of the electromechanical measurements is showed in Figure S5a. 23 Piezo Amplifier DAQ Computer LabVIEW a) P Electrode Sample Electrode Load Cell P b) Figure S5. Piezoelectric coefficient measurements. a) Schematic of the configuration used to measure the electromechanical behavior of piezoelectric materials. Samples were sandwiched between two electrodes, which were wired to an amplifier and a data acquisition (DAQ). Signals were recorded and analyzed in LabVIEW. b) Electromechanical response of a commercial PVDF film. Signals recorded from mechanical load (blue) and electrical charge (red). LabVIEW software was used to process the measured signals and calculate the piezoelectric properties. The peak/valley function was used to calculate the electrical charge (DQ) and load cell (DP) amplitudes. The piezoelectric coefficient was calculated as Equation (S6) To validate electromechanical experiments, we used the commercial PVDF film (TE Connectivity, PN: 3-1003352-0) with known piezoelectric coefficient (d33 ~ 30 pC/N) before each measurement (see Figure S5b). 24 S6.[12] #$$=∆E∆F Supporting Information S7: Graded Material To investigate the potential capability of our material system to deposit different amount of minerals along a surface proportional to the amount of stress, we subjected piezoelectric films of PVDF to cantilever loading. With this loading configuration, regions of the film closer to the foundation experience high mechanical stress, thus high charge generation. Regions of the film closer to the point of load application will have low stress, thus low charge generation. This configuration can lead to having a graded material using a simple one-step procedure by generating graded stress distribution. We cut PVDF films of 5×40 mm2 with a razor blade. One side of the film was fixed to a foundation. The other side was bent by an actuator excited by a long stroke shaker (Electro-Seis APS 113, APS Dynamics, California). Films were subjected to strain-controlled actuation with a stroke of ~25 mm, guaranteeing that samples were sufficiently bent. The electrical charges were measured in the regions of high and low stresses by carbon tape electrodes of ~3×5 mm2. Electromechanical characterization was conducted as explained in Supporting Information S6. A schematic of the loading and charge measurement configuration is shown in Figure S6a. To allow mineral formation in response to the applied stress, the samples were submerged in a 10×SBF solution refreshed every day for 3 days under cyclic loading. After the incubation, samples were rinsed and prepared for surface profile measurements (Supporting Information S8). Supporting Information S8: 3D Optical Profilometer Measurements To measure the mineral distribution that was deposited on the PVDF films from the graded material experiments (Supporting Information S7), we utilized an optical profilometer. Samples were carefully attached to a silicon wafer by fixing with tape. A non-contact profilometer (Laser Scanning Microscope, Keyence VK-X100, Osaka, Japan) was used to scan samples over mineralized regions. The profilometer collected 3D information (x, y, z) from the scanned area. Samples were masked in their borders and a 3D distribution of the mineral was 25 obtained. Several images were taken along the length of the sample and neighboring images were stitched to obtain the final result. First, scanned images were initially processed by VK analysis application (Keyence, Osaka, Japan) to correct surface tilting by checking the level of the substrate layer. Common correction methods we used were the 2-point linear profile correction and the 3-point non-linear profile correction, depending on specific situations. Data was then imported to Gwyddion (Department of Nanometrology, Czech Metrology Institute). We utilized analysis tools to quantify height distribution of minerals along the length of sample. Specifically, we obtained the mineral height along the beam length by calculating the root mean square mineral height on the film surface. The mineral height vs. beam length data was then filtered using a 1-D median filtering of order 300. This allowed the mineral distribution to be less noisy. Supporting Information S9: Numerical Model A three-dimensional (3D) finite element model was developed to simulate piezoelectric beam deflection. The simulation was conducted using the commercial FEA package Abaqus (SIMULIA, Providence, RI). The Abaqus/Standard solver was employed and the model was built using linear solid elements (Element type: C3D8). The beam geometry was set as 15 mm (length) × 0.1 mm (thickness) × 5 mm (width). The material was assumed to be linearly elastic, with Young's modulus of 4 GPa and Poisson's ratio of 0.18. The one end of the beam was constrained in all degree of freedom and the bottom surface of the beam was loaded with 500 Pa in the vertical direction of the beam (Figure S6). The beam deflection was 6.5 mm and stress (S33) values on the top surface were extracted along the beam direction. 26 S33 S33 3 2 1 3 2 1 Figure S6. Finite element analysis (FEA) to calculate S33 according to the piezoelectric beam deflection. Left image shows stress map (S33) before deformation. Right image shows stress distribution (S33) of beam after subjected to cantilever bending mode. Supporting Information S10: Electrospinning of Polyvinylidene fluoride (PVDF) To make porous piezoelectric scaffolds, electrospinning was utilized. PVDF pellets (Mw=530,000), N,N-dimethylformamide (DMF, 99.8%) and acetone (99.5%) were purchased from Sigma-Aldrich (St. Louis, MO, USA). All the materials were used as received without further purification. The PVDF pellets were first dissolved in a DMF/acetone mixed solvent at a certain concentration and stirred at 80°C overnight until a transparent homogenous solution was obtained. The weight ratio of the DMF and acetone in the mixed solvent was 2:3 and the mass concentration of PVDF in the solution was 18%. Then, the PVDF/DMF/acetone solution was cooled down to the room temperature. To electrospin fibrous PVDF membranes, the PVDF/DMF/acetone solution was placed into a 10-mL glass syringe installed with a stainless steel-needle as the spinneret. A digitally controlled syringe pump (N1000 - New Era Pump Systems, Inc.) was used to feed the polymer solution into the needle tip at a constant feeding rate of 1 mL/hr. A rotary disk with the diameter of 30 cm was electrically connected to the ground and used as the fiber collector. The whole setup is shown in Figure S7a. During the electrospinning process, the PVDF solution 27 underwent a high DC electrical field of 138 kV/m. Such a high DC electrical field was generated by applying a positive voltage of 18 kV with a 13 cm gap between the spinneret and the fiber collector. After electrospinning, a non-woven PVDF fibrous mat was formed onto the rotary aluminum disk, which was peeled off as a thin fibrous PVDF membrane (Figure S7b). The thickness of the PVDF members was controlled around 100 - 200 µm by the electrospinning time, which was around 2 hours. The surface morphology of the as-prepared electrospun fibrous PVDF membranes (Figure S7c) were characterized by using a scanning electron microscope (SEM, Tescan Mira 3). Prior to SEM examination of the fibrous membranes, the membrane specimens were sputter-coated with Platinum to avoid possible charge accumulation onto the PVDF fibers during the test. d) 2000 ) s p c ( y t i s n e t n I 1500 1000 500 0 10 ) 0 0 2 , . 7 0 2 0 1 1 ( β . 6 6 3 ) 1 0 1 , 0 2 0 ( β 20 30 2θ (°) 40 Figure S7. Electrospinning of piezoelectric PVDF. a) Apparatus utilized to conduct electrospinning. b) A photo of an electrospun PVDF membrane. c) SEM of the microstructure of PVDF. d) XRD pattern of a pristine PVDF electrospun sample. The degree of crystallinity of the PVDF electrospun films (i.e. amount of β-phase) was estimated using XRD as described in Supporting Information 2. Figure S7d shows the XRD pattern of an electropun sample. The peaks observed 20.7°, 36.6°, and 56.9° are assigned to (1 1 0, 2 0 0) and (0 2 0, 1 0 1) diffractions of the β-PVDF crystal plane, respectively.[13] This 28 phase has a large spontaneous polarization within crystalline phase thus providing the piezoelectric effect.[14] Supporting Information S11: Self-Stiffening Experiments To test the self-stiffening capability of our material system, electrospun PVDF samples were submerged in SBF solutions and were subjected to cyclic mechanical loading (i.e. external stimulus) to activate the electrical charge generation and promote the nucleation and growth of minerals that can modulate the effective stiffness by adding minerals. We prepared electrospun PVDF samples of 30×15 mm2 according to Supporting Information S10. Simulated body fluids (10×SBF) were prepared by following protocols described in Supporting Information S1. While submerged in the solution at room temperature, samples were subjected to cyclic compression loading using an electromechanical universal testing machine (TA Instruments Electroforce 5500 with a 10 N capacity load cell). The stiffness (S) was calculated from the load (P) and displacement (d) amplitudes measured with the LVDT actuator using the follow relation: S=ΔP/Δδ. The force and displacement signals were monitored and recorded during experiments using Wintest software 7.6. To prevent any change of mechanical properties of electrospun films due to viscoelastic and time-dependent behavior during experiments in SBF, we initially subjected samples to cyclic compression loading while submerged in DI water for 3 days. The modulus (E) was then calculated by multiplying by the cross section area and the deflection Ε = SΔδ/A. After the modulus reached a constant value (Figure S8a), we switched the solution to the SBF and kept the same loading conditions. The force and displacement were continually monitored and the modulus was calculated based on the measured values as summarized in Figure 3b. To explore the modulus adaptation in response to the external mechanical loading, we subjected samples to different loading configurations in terms of frequency and load 29 amplitude. We chose loading conditions that offered a variety of charge conditions whose results are summarized in Figure 3g. a) ) m m N / ( s s e n f f i t S 160 140 120 100 80 0.0 b) 8 ) a P M ( s s e r t S 6 4 2 0 0.0 0.8 50 μm 0.2 0.1 0.4 Strain (mm/mm) 0.3 0.5 0.4 0.6 0.2 Time (days) Figure S8. Preparation of self-stiffening experiments. a) Stiffness change of a PVDF electrospun sample submerged in water. Initial response after cyclic loading was recorded. Due to viscoelastic effects, stiffness reached a plateau after 12 hours. b) Stress-strain response of as- prepared electrospun PVDF specimen without any incubation or mineralization. The fibers were aligned in the direction of loading. For measuring stress-strain responses of electrospun PVDF specimens, the specimens were subjected to quasi-static loading to failure under a tensile configuration using a universal testing system (TA ElectroForce 5500) with rate = 0.6 min-1 based on the ASTM standard.[15] The tensile strength (σ) of the specimens was calculated according to σ = P/A, where P is the measured load and A the cross-sectional area (width by thickness) of the specimen. The strain was calculated by measuring the gauge length divided by the displacement amplitude recorded by the actuator. Figure S8b shows the stress-strain response of a representative sample of electrospun PVDF. The elastic modulus of the sample is ~20 MPa and the tensile strength ~4 MPa, which are comparable with values found in literature.[16] The stress-strain responses after cyclic loadings are shown in Figure 3c and 3 h. 30 Supporting Information S12: Confocal Microscopy of Mineralized Electrospun Scaffolds Mineralized electrospun PVDF sample (7 days in 10×SBF, daily refreshed) was cut into 0.5×0.5 cm2 piece and stained with 1 mg/mL Alizarin Red S (Sigma -- Aldrich A5533) in aqueous buffer for 20 min at room temperature. After staining, the sample was gently rinsed with de-ionized water for 2 min and then kept in PBS solution (Lonza 17-516F) for imaging purpose. Imaging of mineral deposition was performed by confocal laser scanning microscopy on an inverted microscope system (Leica TCS SP8) with a 63x1.40 N.A. Plan Apo oil immersion objective lens. Illumination of the samples stained with Alizarin Red S was provided by a white light laser (Leica WLL), exciting at 540 nm. Images were captured by a hybrid detector (Leica HyD) with a bandpass filter of 605/70 nm. The Z-step size of the image (~200 nm) was automatically optimized by the system. Z-stacks were rendered in 3D in Leica LAS X software and analyzed in Fiji Image J. We confirmed that the dye preferentially stained minerals by comparing fluorescence images without and with minerals (Figure S9). To calculate the mineral thickness and fiber diameter, we did five cut section views along the field of view. Locations were sectioned randomly. We re-oriented the confocal view perpendicular to user. We utilized Image J measuring tool to manually measure the thickness of the mineral. In addition, we measured the hole diameter assuming it was a spaced occupied by the PVDF fiber. From each section cut, we averaged the fiber diameter and mineral thickness three times. The result summarized in Figure 3j showed good agreement with the data from the SEM analysis in Figure S7c. 31 a) PVDF with no minerals b) PVDF with minerals 20μm 20μm Figure S9. Staining process with Alizarin Red S. a) PVDF film without formed minerals showing no red fluorescence in the polymer. b) Mineralized PVDF film showing the expected red fluorescence in regions with formed minerals. Supporting Information S13: Analytical Model We further analyzed the data from the functionally graded material experiment (Supporting Information S7) to find the relation between the mineral height and the charge along the beam. Assuming small deformation, linear elastic response of a material and a linear relation between force and charge, we can obtain Eqnuation (S6): where Q is charge (unit: C), d33 is piezoelectric constant (unit: pC/N), G is stress (unit: MPa), (S6) Q = d33F Q = d33(σ33A) and A is area perpendicular to stress direction (unit: mm2).[12] From the stress distribution along the beam length in the Figure S10, the stress along regression of second order (R2=0.99), (see Figure S10b): the z-direction (σ33) at a given location (x) can be described as Equation (S7) after a polynomial G$$=0.1H0−4.1866H+37.183 (S7) 32 From the mineral formation distribution along the beam length, the mineral height at a location can be approximated as Equation. (S8) after a linear regression (R2=0.82), (see Figure S10c): MH = -1.52×10-3 x + 1.79×10-2 (S8) where MH is mineral height (mm), x is the beam length (mm). After solving for x in Equation (S8), and replacing it into Equation (S7), and then into Equation (S6), we can obtain Equation. (S9): Then, the Equation (S9) can be rewritten as Equation (S10): !≅#$$%(4.33×10-./0) (S9) ./=> -.$$×87NOPP5 (S10) E Because charge (Q) is linearly proportional to stress (σ) (Equation S6), the mineral height (MH) is proportional to Q1/2 or σ1/2. 33 a) L=15 Load b) z (3) x (1) A=w ⨯ L=0.5 mm2 h=0.1 w=5 δ=6.5 !33 = 0.1x2 -- 4.186x + 37.182 R² = 0.9803 40 30 ) a P M ( 3 3 σ 20 x 20 25 5 10 15 Beam Length, x (mm) Q= d33⨯A⨯43282.5⨯MH2 With d33= 30pC and A=0.5 mm2 10 30 0 Mineral Height, MH (μm) 20 10 0 0 500 400 300 200 100 0 c) ) m m ( H M , t h g i e H l a r e n M i 0.020 0.015 0.010 0.005 0.000 0 MH=-0.00152 (x) + 0.0179 R2=0.825 d) ) C p ( Q , e g r a h C x 5 10 Beam Length, x (mm) Figure S10. Analytical model. a) Cantilever configuration for a graded material experiment. b) Stress distribution along the beam length obtained from numerical model (Supporting Information S9). Higher stress at regions closer to the foundation. c) Mineral height along the beam length obtained from experiments (Supporting Information S7). Higher mineral deposited in regions of high stress of the cantilever beam. d) Analytical model obtained from experimental results in combination with numerical model and analytical equation describing the relationship between mineral height and electrical charge. Units in mm. References [1] [2] T. Kokubo, H. Takadama, Biomaterials 2006, 27, 2907. P. Zhu, Y. Masuda, K. Koumoto, Biomaterials 2004, 25, 3915. 34 [3] [4] [5] [6] S. Bharati, M. K. Sinha, D. Basu, 2005, 28, 617. F. Yang, J. G. C. Wolke, J. A. Jansen, Chem. Eng. J. 2008, 137, 154. P. D. F. JCPDS, PC PDFWIN 2000, 2, 4. J. Li, Z. Zhu, L. Fang, S. Guo, U. Erturun, Z. Zhu, J. E. West, S. Ghosh, S. H. Kang, Nanoscale 2017, 14215. [7] F. Chen, Y. Lu, X. Liu, J. Song, G. He, M. K. Tiwari, C. J. Carmalt, I. P. Parkin, Adv. Funct. Mater. 2017, 27, 1702926. [8] [9] E. M. Gindler, J. D. King, Am. J. Clin. Pathol. 1972, 58, 376. D. M. Karl, G. Tien, Limnol. Oceanogr. 1992, 37, 105. [10] J. Taylor, Introduction to Error Analysis, the Study of Uncertainties in Physical Measurements, 1997. [11] M. Stewart, M. G. Cain, in Characterisation Ferroelectr. Bulk Mater. Thin Film., Springer, 2014, pp. 37 -- 64. [12] A. Safari, E. K. Akdogan, Piezoelectric and Acoustic Materials for Transducer Applications, Springer Science & Business Media, 2008. [13] D. M. Esterly, B. J. Love, J. Polym. Sci. Part B Polym. Phys. 2004, 42, 91. [14] E. Kabir, M. Khatun, L. Nasrin, M. J. Raihan, M. Rahman, J. Phys. D. Appl. Phys. 2017, 50, 163002. [15] A. International, ASTM D638-14, Standard Test Method for Tensile Properties of Plastics, ASTM International, 2015. [16] R. Nakashima, K. Watanabe, Y. Lee, B. Kim, I. Kim, Adv. Polym. Technol. 2013, 32, E44. 35
1802.07500	1	1802	2018-02-21T10:30:27	Passive Decoupling of Two Closely Located Dipole Antennas	[ "physics.app-ph" ]	In this paper, we prove that two parallel dipole antennas can be decoupled by a similar but passive dipole located in the middle between them. The decoupling is proved for whatever excitation of these antennas and for ultimately small distances between them. Our theoretical model based on the method of induced electromotive forces is validated by numerical simulations and measurements. A good agreement between theory, simulation and measurement proves the veracity of our approach.	physics.app-ph	physics	Communication Passive Decoupling of Two Closely Located Dipole Antennas M. S. M. Mollaei, A. Hurshkainen, S. Kurdjumov, S. Glybovski, and C. Simovski 1 8 1 0 2 b e F 1 2 ] h p - p p a . s c i s y h p [ 1 v 0 0 5 7 0 . 2 0 8 1 : v i X r a Abstract-In this paper, we prove that two parallel dipole antennas can be decoupled by a similar but passive dipole located in the middle between them. The decoupling is proved for whatever excitation of these antennas and for ultimately small distances between them. Our theoretical model based on the method of induced electromotive forces is validated by numerical simulations and measurements. A good agreement between theory, simulation and measurement proves the veracity of our approach. Index Terms-Decoupling, Dipole antenna, Passive antenna. I. INTRODUCTION For many decades, usage of array antennas has the attention of researchers. Being employed in a variety of applications such as radars, Multi-Input Multi-Output (MIMO) systems and Magnetic Resonance Imaging (MRI) has made array antennas more interesting than any other time [1], [2]. Currently, decoupling of elements in the aforementioned applications has made the focus of research. For MIMO systems, a variety of techniques has been implemented and yet researchers strive to enhance those techniques [3]– [5]. For antennas used in MRI, perhaps not ideal but sufficient, reliable and easily tunable decoupling is an important issue. In the transmission regime, decoupling of the array antennas prevents the parasitic cross- talks and inter-channel scattering. In the reception regime, it prevents the noise correlation of channels that reduces the signal-to-noise ratio, one of key parameters of MRI. Finally, it makes the input impedances of two equivalent antennas 1 and 2 equivalent for arbitrary excitation magnitudes and phases. This simplifies the creation of the needed distribution of currents in the array elements and their impedance matching – no need to engineer very expensive adaptive properties in a decoupled array. In the most of antenna array applications where the received signal is rather weak, the array elements cannot be decou- pled in a straightforward way – using screens or absorbing sheets. In last two decades, a very successful technique of the passive decoupling was developed for microwave antennas – that based on the so-called electromagnetic band-gap (EBG) structures. EBG structures are planar periodical structures – microwave analogues of commonly known photonic crystals. Decoupling based on EBG structures works very well when the gap between two adjacent antennas may comprise a M. S. M. Mollaei, and C. Simovski are with Department of Electronics and Nanoengineering, Aalto University, PO Box 15500, FI-00076, Finland (e-mail: [email protected],[email protected] ). A. Hurshkainen, S. Kurdjumov, and S. Glybovski are with the Department of Nanophotonics ITMO University, 197101, St. Petersburg, Russia (e-mail: [email protected], [email protected], [email protected]) and Metamaterials, sufficient number of the EBG unit cells [6]. This is, for example, the case of [7]. In [7] 3 unit cells of the EBG structure in the gap of the width λ/10 (where λ is the operation wavelength) between two adjacent dipole antennas turned out to be sufficient for decoupling. This is, probably, the minimal gap for which the decoupling is possible using EBG structures. Judging upon [7]–[9] the minimal amount of unit cells required for decoupling is equal 3. EBG structures with tech- nically achievable miniaturization of the unit cell size allow one to place three unit cells into a λ/10 gap [9]. However, no one knew technical solution of EBG structures allowing the unit cell miniaturization up to λ/100, and if the gap is as small as d = λ/30 one has to find other solutions. The use of the adaptive circuitry based on operational amplifiers can be justified for radar systems but for MIMO and MRI applications it would be a very expensive and impractical way [10]. For the demands of such systems, the number of array elements needs to be increased, which necessarily leads to high inter-element coupling. Accordingly, one needs a passive decoupling for ultimately close antennas when d (cid:28) λ/10 in terms of the operational wavelength. Elegant technical solutions were found for the case when the array elements are loop antennas. Since the mutual inductance of two coplanar loops is negative and that of two coaxial loops is positive, the loop array is performed of partially overlapping loops [11], [12]. Another technical solution used for decoupling of loop antennas is capacitive decoupling [13]. Unfortunately, for dipole antennas where the coupling is not purely inductive or capasitive, these methods are not applicable. In the present paper we aim for the passive decoupling, namely, in the arrays of dipole antennas. This problem becomes especially difficult when the gap width is as small as d (cid:28) λ/10 and needs to be solved in two stages. We concentrate on two antennas – the basic case of any type of array. On the next stage we will extend our theory to more antennas. Our idea is to locate a passive scatterer in between two resonant dipoles. Basically, this idea is not fully new – in work [14] the authors revealed the decoupling offered to two resonant monopoles (vertical antennas fed by coaxial cables through a ground plane) by a similar monopole located in between them. The decoupling was obtained in presence of a human head phantom (stretched normally to the ground plane), the distance between the active monopoles was of the order of λ/10. Since the effects of the passive scatterer and of the body phantom were not studied separately, this technical solution was a heuristic finding. The same refers to the decoupling of two loop antennas using a passive resonant loop in work [15], [16]. 2 by other active antennas. It is possible if the impact of our passive scatterers compensates the impedance shared by the given antenna with the other active antennas of the array. B. Structure with a Passive Element In this part, we analytically prove that compensation of the electromotive force induced by antenna 1 in antenna 2 and vice versa is possible at a certain frequency if dipole scatterer 3 is introduced as it is shown in Fig. 1(b). In the regime of decoupling the input impedances of both antennas 1 and 2 are equivalent, and V2 = αV1 if I2 = αI1. The impedance of antenna 1 shared with both antenna 2 and scatterer 3 is as follows: Z s 1 = ZM I2 I1 + Z13 I3 I1 (3) In the decoupling regime it must be equal to the impedance of the second antenna shared with antenna 1 and scatterer 3: Z s 2 = ZM I1 I2 + Z23 I3 I2 (4) Here Z13 and Z23 are mutual impedances of, respectively, antennas 1 and 2 with scatterer 3 and ZM ≡ Z12. Current I3 in the center of scatterer 3 is induced by primary currents I1 and I2 (where by definition I2 = αI1). Due to linearity of electromagnetic interaction, the current induced in a scatterer by a primary current must be proportional to this primary current, and we have I3 = ξ13I1 + ξ23I2 (5) Here ξ13 and ξ23 are certain coefficients, determined by the system geometry and independent on currents I1 and I2. With these coeffcients the equivalence of (3) and (4) can be written in a form ZM α + Z13 ξ13I1 + ξ23I2 I1 = ZM α + Z23 ξ13I1 + ξ23I2 = (ZM + ξ23Z13) α + ξ13Z13 = (ZM + ξ13Z23) + ξ23Z23 (6) If the distances d1 and d2 are equivalent (the scatterer is symmetrically located) Z13 = Z23 and ξ13 = ξ23. In this case, (6) becomes an identity when α = 1. Of course, this equivalence of the input impedances does not mean their decoupling. For arbitrary α (6) is satisfied when ZM + ξ23Z13 = ZM + ξ13Z23 = 0. (7) Equation (7) is the needed condition of complete decoupling. If it is satisfied input impedance of antenna 1, Z1 = Z + Z s 1 does not depend on I2 and vice versa, the shared impedances of antennas 1 and 2 do not depend on I1 and I2 and are equal to each other: Z s 1 = Z s 2 = ξ13Z13 = ξ23Z23 (8) Equations (7) and (8) mean that the electromotive force induced in antenna 1 by antenna 2 (and vice versa) is always compensated by a part of the electromotive force induced in them by scatterer 3. If there is no mutual coupling of 1 and I2 1 α Fig. 1. dipole antenna between two active dipole antennas. (a) Two closely located active dipole antennas, (b) adding a passive In Section II, we suggest a concept of the complete passive decoupling of two active antennas by adding a parasitic dipole. Complete decoupling means the suppression of the power flux between the antennas obtained for arbitrary exciting magni- tudes and phases. In Section III, a numerical investigation is presented and the results are compared with the theory. In section IV we report an experiment that verifies both analytical and numerical results. II. DECOUPLING OF TWO CLOSELY LOCATED DIPOLES A. Reference Structure Here we consider the interaction of two identical parallel dipole antennas separated by a gap d. Fig. 1(a) specifies the most interesting case when these dipoles are resonant. Let them be fed by arbitrary voltages V1 and V2, respectively, and denote their self-impedances as Z11 = Z22 = Z. A system of Kirchhoff equations can be written in terms of the mutual impedances Z12, Z21 and, alternatively, via the shared (additional) impedances Z s 2 of antennas: 1 and Z s I1Z11 + I2Z12 ≡ I1 (Z11 + Z s I1Z21 + I2Z22 ≡ I2 (Z22 + Z s 1 ) = V1 2 ) = V2 (1) (2) 1 and Z s Here currents I1,2 refer to the centers (feeding point) of the dipoles. Mutual impedances, as follows from the reciprocity, are the same Z12 = Z21 ≡ ZM . The shared impedances of two 2 are different if I1 (cid:54)= I2. They express the antennas Z s electromotive forces induced in antenna 1 by antenna 2 (and vice versa) normalized to the current in antenna 1 (or 2). Let us write I2 = αI1, where the coefficient α is an unknown complex value. If it 1 = αZM 2 = ZM /α are different. Respectively, input impedances and Z s 1 and Z2 = Z + Z s of the equivalent antennas Z1 = Z + Z s 2 are also different. Therefore, mutual coupling of two antennas means that the relation α (cid:54)= 1 between the currents is different from the relation between their voltages V2 (cid:54)= αV1. is different from unity Z s For decoupled antennas we would have V2/V1 = I2/I1 for whatever α. Expanding this result to an array with N > 2 elements we observe that the distribution of currents in its elements repeats that of applied voltages. All we need for it is to nullify the electromotive force induced in the given antenna 2 via the electromotive force, there is no power flux between them. To express ξ13 through ZM ≡ Z12 = Z21, denote the self- impedance of scatterer 3 as Z0. For calculating ξ13 and ξ23, we need two different scenarios of excitation. In the first scenario, we assume that dipole 1 is active (V1 = V ) while dipole 2 is as passive as dipole 3 (V2 = 0). We may write the system of Kirchhoff's equations for our three dipoles as follows: I1Z + I2ZM + I3Z13 = V, I1ZM + I2Z + I3Z23 = 0, I1Z13 + I2Z23 + I3Z0 = 0, (9a) (9b) (9c) and easily obtain: κ13 ≡ I3 I1 = ZM Z23 − ZZ13 ZZ0 − Z 2 23 lkap1 (10) In the second scenario, we assume that dipole 2 is active and two other dipoles are passive: V1 = 0, V2 = V . Then we obtain κ23 ≡ I3 I2 = ZM Z13 − ZZ23 Z0Z − Z 2 13 lkap2 (11) 3 C. Decoupling of Two Half-Lambda Dipoles If Z0 = Z (self-impedances of all our dipoles are equivalent) the decoupling condition yields to: Z13 2 = ZZM (13) Formulas for the self-impedance of a dipole antenna located in free space and for the mutual impedances of two parallel dipole antennas are well known. Standard formulas represent converging power series [17], closed-form combinations of integral sine and cosine functions [18], or polylogarithm functions [19]. However, for our purposes we do not need these long formulas. We consider the special case of resonant dipoles and may use simple approximate relations, valid in the vicinity of the antenna resonance. First, let us recall that the straight-wire resonance holds at a frequency that is slightly lower than the one at which L = λ/2. By definition, at the resonance frequency ω0 the reactive part of the self-impedance vanishes. For a perfectly conducting wire of length L and cross section radius 10−5λ < r0 < 10−3λ (this interval of values for r0 is assumed below) the resonant wave number k0 ≡ 2π/λ equals to 0.992π/L [18]–[20]. At this frequency, the input resistance of the dipole is equal R0 ≈ 70 Ohm [18]–[20]. A very simple relation for the mutual impedance of two parallel antennas performed of wires with radius r0 < 10−3λ separated by a gap d < L was heuristically obtained in [21]. In our notations this relation can be written for both Z12 = ZM and Z13 = Z23 as follows: With substitutions of kap1 and kap2, condition (7) can be rewritten in terms of mutual and self-impedances: (cid:112) Z13 = Z0ZM = Z23. (12) This condition is that of complete decoupling of two active antennas by a passive one. Our term complete decoupling means the following: if antennas 1 and 2 are fed by arbitrary voltage sources V1 and V2 with internal impedances Zi1 and Zi2 we still have no mutual coupling. The impedances Z1 and Z2 connected to the voltages V1 and V2 comprise the antenna self-impedance Z, the source impedances Zi1,i2 and the shared 1,2 of antennas 1 and 2. If our condition (12) is impedances Z s satisfied we obtain from Kirchhoff's equations for the structure depicted in Fig. 1(b) the equivalence of the shared impedances 2. To obtain it we substitute in (9a) Z → Z + Zi1 and 1 = Z s Z s V → V1, and in Eq. (9b) Z → Z + Zi2 and 0 → V2. Then we substitute for Z13 and Z23 relation (12), solve Kirchhoff's equations for I1 and I2 and obtain V1/I1− Zi1 = V2/I2− Zi2 i.e. Z1 − Zi1 = Z2 − Zi2. Since Z1 = Z + Zi1 + Z s 1 and 2. This equivalence of Z2 = Z + Zi2 + Z s the shared impedances in the case when two arbitrary different generators are connected to our antennas may result only from their decoupling. 2 it means Z s 1 = Z s Now, we have to prove the feasibility of our condition (12). In our derivations we did not assume the exact symmetry of the location of scatterer 3. However, it is required by Eq. (12): Z13 = Z23. Let us show that (12) is feasible when scatterer 3 is the same resonant dipole located between dipoles 1 and 2 in the same plane. This case is shown in Fig. 1(b). ZM = η η 24πkd e−jkd e−jkd/2 (14a) (14b) Z13 = and 3 is d/2 and it is denoted η ≡ (cid:112)(µ0/0) (free-space Here it is taken into account that the distance between 1 12πkd impedance). Comparison with the known data for the mutual impedance of two parallel identical dipoles [19], shows that Eqs. (14a) and (14b) are sufficiently accurate when r0 (cid:28) d (cid:28) λ at frequencies located within the half-lambda resonance band. This resonance band is defined via the relative detuning γ as follows: γ ≡ (ω − ω0)/ω0 ≤ 0.01. If we show that Eq. (13) holds at a frequency within this band the use of approximations (14a) and (14b) for the mutual impedances will be justified. The input reactance X of a dipole within the above-defined resonance band is negligibly small compared to the input resistance R and we have Z ≈ R in (13). The dependence of R on the relative detuning in the resonance band of a half- lambda dipole can be modelled by a linear function [22]: (15) where β ≈ 59. Substituting Eqs. (14a) – (15) into (13), we obtain: Z = R0 (1 + βγ) R0 (1 + βγ) η e−jkd = η2e−jkd (12πkd)2 this In cancel Substituting η = 120 Ohm, we simplify (16) to a form: exponentials equation, 24πkd complex (cid:18) (cid:19) 70 1 + 59 k − k0 k0 = 20 kd (16) out. (17) 4 Fig. 2. Reference structure of two dipole antennas with ideal lossless matching circuits (removed in the mismatched regime). This is clearly a feasible condition. For example, for two dipoles of length L = 50 cm performed of a wire with radius r0 = 1 mm (such a dipole resonates, in accordance to the analytical theory, at 298 MHz) separated by a gap d = 3 cm, that is centered by a wire of the same radius and length, Eq. (17) is satisfied when γ = (ω − ω0)/ω0 ≈ 0.007, i.e. at frequency ω = 1.007ω0 ≈ 300 MHz. Thus, the prerequisite of (14a) and (14b) is respected: decoupling holds within the resonance band. III. NUMERICAL VERIFICATION In this part, the proposed method is validated numerically. The simulation has been carried out using CST Microwave Studio, Time Domain solver. As a reference, we consider the system of two dipole anten- nas performed of copper wires with geometric parameters L = 500 mm, r0 = 1 mm separated by the distance d = 3 cm. We simulate S-parameters of the system assuming our dipoles to be performed of copper wires. In these simulations, dipoles 1 and 2 were excited by lumped ports either through matching circuits or without them. This schematic of the reference antenna system is shown in Fig. 2. In the matched case lumped voltage sources V1,2 with internal resistances Ri1,i2 = 50 Ohms are connected to the dipoles centers through a lossless LC circuit whose parameters are chosen so that its reactance Xi1,i2 compensates the antenna input reactance at 300 MHz and the input resistance transforms into 50 Ohms. In this case in absence of scatterer 3 the coupling is maximal (S12 = −1.2 dB) at 293 MHz that is an evident consequence of the maximal antenna currents at this frequency. The band of matching in the reference structure defined by condition S11 ≤ −20 dB is equal [292.7,293.3] MHz (relative bandwidth 0.2%). It is desirable to keep this band in the regime of decoupling. Indeed, a scatterer located so closely to our dipoles obviously brings an extra mismatch. This mismatch can be compensated by adjusting the match- ing circuit. However,ideal matching with a reactive circuit is possible at a single frequency. Therefore, the decoupling element may severely shrink the band of the antenna matching that may become narrower than that of the transmitted signal. Then one has to introduce losses in the matching circuit that means the decrease of the antenna efficiency compared to the reference structure. Moreover, the band of decoupling can be Fig. 3. Simulation results for the reference and decoupled structures. Matching circuits are absent. even narrower than the band of lossless matching. Our Eq. (17) allows us to find a single frequency of decoupling and tells nothing about its matching band. These fine issues can be hardly covered by our approximate theory and we clarify them in extensive numerical simulations. In Figs. 3 and 4 we depict the results of numerically calcu- lated S-parameters of two dipoles in the presence of a passive scatterer without and with a matching circuit accordingly. Our simulations confirmed the prediction of the analytical model that decoupling of our dipoles 1 and 2 is granted by the straight wire (scatterer 3) of the same length and radius, located in the middle between 1 and 2. The decoupling is complete because it is achieved in both matched and mismatched cases. In the matched case Xi1,i2 is the value of the order of Ri1,i2, whereas in the mismatched case Xi1,i2 = 0. Therefore, the ratio of currents in the passive and active antennas (α = I2/I1) is different in these two cases. However, in both matched and mismatched cases we have obtained a local minimum of S12 at the same frequency. This is an evidence of the complete decoupling. Of course, in both these cases S12 cannot exactly vanish. Though our analytical model is approximate, and the simulation shows that the reachable isolation is incomplete, it can be seen that the complete decoupling condition is still valid. We note that in the most of applications the isolation S12 < −20 dB of two matched antennas is sufficient. As expected, the decoupled structure in the mismatched case manifests an extra mismatch compared to the reference one. In the reference structure the mutual coupling is not so harmful for matching. Mutual impedance of two closely located (d ≈ λ/30) dipoles in their resonance band has absolute values within the limits [20,50] Ohms (see e.g. in 1 is smaller than the mutual [20]), and the shared impedance Z s impedance because dipole 2 is passive and, therefore, α < 1. Consequently, for the reference structure we see a broadband (though poor) matching with the resonance frequency 293 MHz, almost equal to that of an individual dipole. In the de- coupling structure dipole 3 is distanced by d/2 = λ/60 = 1.5 cm from our active dipole 1. For a so small distance, the mutual resistance of two half-wave dipoles approaches the self-resistance and κ13 is almost real and negative. Therefore, 5 Fig. 4. Simulated S-parameters of the reference and decoupled structures. Matching circuits are present. Fig. 5. Measurement setup comprises two dipole antennas connected to a VNA and a dipole scatterer. The antenna system is supported by a foam board wrapped with the paper. The board design allowed to raise the passive scatterer. dipole 3 is excited in the opposite phase and the shared resistance of dipole 1 almost compensates its self-resistance. As a result, at the decoupling frequency 293 MHz the absolute value of the input impedance of dipole 1 turns out to be much smaller than Ri1 =50 Ohm which implies a very strong mismatch we observe for the decoupled structure in Fig. 3. However, whatever nonzero input impedance can be always matched at a single frequency using a lossless matching circuit. In Fig. 4 we show the S-parameters of the matched system calculated in absence and in presence of the scatterer. Again the decoupling holds at 293 MHz because within the band 292.7-293.3 MHz the minimum of S12 (equal to -20 dB) is achieved at 293 MHz. Beyond this band S12 also decreases versus detuning, but this is a consequence of the mismatch, and is not decoupling. From Fig. 3 and 4 we conclude the minimum of S12 at 293 MHz does not depend on the currents in antennas 1 and 2 and represents an evidence of the complete decoupling. Notice that the result for the decoupling frequency 293 MHz fits very well the prediction of the analytical model. The only drawback of our decoupling is squeezed operational band because our dipole 3 brings a strong mismatch. Namely, in accordance to Fig. 4, the relative band of matching defined via S11 ≤ −20 dB shrinks from 0.2% (reference structure) to 0.05%. The band of matching is practically equal to the band of decoupling defined on the level S12 = −15 dB. IV. EXPERIMENTAL VALIDATION AND DISCUSSION For the experimental validation of our theory we built a setup whose general views are pictured in Fig. 5. It comprised active dipoles 1 and 2 performed of a copper wire with L =500 mm and r0 = 1 mm split at the center with a small (0.6 mm) antenna gap. A vector network analyzer (VNA) Rohde Schwarz ZVB20 was connected to the arms of the dipoles through a logical symmetric ports with the wave impedance 100 Ohm each. In this scheme two identical coaxial cables are used to connect each of the dipoles. The symmetric connection allowed us to measure the S-parameters properly removing the cable effect in the whole frequency range. The distance between the dipole antennas was d = 3 cm. In the decoupling Fig. 6. Comparison of simulation and measurement results for the decoupled structure in mismatched regime. regime the similar copper wire (L = 500 mm, r0 = 1 mm but without a central split) was located in the middle of the antenna structure. A mechanical support was a board of foam wrapped with paper that allowed us to raise scatterer 3 to a height 0 < h ≤ 5 cm over the plane of dipoles 1 and 2. Varying h in both measurements and simulations, we found complete decoupling is possible also for h (cid:54)= 0. Measurements of S-parameters for the reference case (without scatterer 3) have shown an excellent agreement with these simulations. For the structure with the decoupling scatterer the agreement is still acceptable as one can see in Figs. 6 and 7. These plots correspond to h = 0. In the measurement whose result is depicted in Fig. 6 there is no matching circuits connected to antennas 1 and 2. In our simulations of this mismatched case the minimum of S12 occurs at 293 MHz, which is exactly the same as experiment. The same frequencies of the minima of S12 (simulated and measured ones) keep for the matched case. The corresponding plot is shown in Fig. 7, where the ideal two-port matching is supposed at all frequencies. The coincidence of the frequencies of these minima in Figs. 6 and 7 is the evidence of the complete decoupling. 6 question what factor higher reduces efficiency. However, for some applications (e.g. for MRI array coils) even a narrow operational band (relative bandwidth of the order of 0.1%) reported in the present paper may be sufficient. In our next paper we will expand the study to the case when the number of decoupled antennas is more than N = 2. VI. ACKNOWLEDGEMENT This work was supported by the Ministry of Education and Science of the Russian Federation (project No. 14.587.21.0041 with the unique identi fier RFMEFI58717X0041) and Euro- pean Unions Horizon 2020 research and innovation program under grant agreement No. 736937. Fig. 7. Comparison of simulation and measurement results for the decoupled structure when an ideal dual side matching of the two port is performed. REFERENCES Here, instead of real matching circuits we used MATLAB code based on the method from [23] which allows us to normalize S12 (was measured in the mismatched regime and depicted in Fig. 6) to the accepted power considered as a sym- metric passive two-port (dual-side matching). Mathematically, the obtained result is equivalent to the presence of an ideal matching circuit transforming the antenna input impedance into 50 Ohms at each of the plotted frequencies. This approach was dictated by the necessity to measure S12 for different h allowing us to avoid the fabrication of a tunable matching circuit and gives the reachable isolation between the dipoles at all frequencies, for which each of those a lossless matching circuit could be individually constructed. Simulations and measurements of S12 for h varying in the limits 0 ≤ h ≤ 50 mm have shown no complete decoupling for h (cid:54)= 0. Though the deepest minimum of S12 = −17 dB in the matched case was obtained for h = 10 mm, this was not our complete decoupling, because this minimum corresponds to frequency 294.2 MHz, whereas in the mismatched case for h = 10 mm S12 attains the minimum at 293.2 MHz. Only if h = 0 the frequency of the minimum of S12 keeps the same in both matched and mismatched cases which goes along with the theory. V. CONCLUSION In this work we have comprehensively studied the passive decoupling of two active dipole antennas by a single passive dipole scatterer. We aimed the complete decoupling – that holds for arbitrary relations of currents and deriving voltages in these antennas. We have proved analytically, numerically and experimentally that this decoupling is feasible for resonant dipole antennas. A drastic decrease of mutual coupling was obtained for the case when the distance between two antennas was much smaller than one tenth of the operation wavelength. The only drawback of this decoupling is the shrink of the lossless matching band by an order of magnitude (and the similarly narrow band of decoupling). If the signal band is not correspondingly narrow, this shrink may be harmful for the antenna efficiency. In this case, resistive elements may further significantly reduce the efficiency. This is an open [1] H. Li, "Decoupling and Evaluation of Multiple Antenna Systems in Compact MIMO Terminals," Ph.D. dissertation, Dept. Elect. Eng., KTH Univ., Stokholm, Sweden, 2012. [2] Y. Kawakami, R. Kuse, and T. Hori, "Decoupling of dipole antenna array on patch type meta-surface with parasitic cells," in 11th Euro. Conf. Antenna. Propag., Paris, France, 2017, 10.23919/EuCAP.2017.7928220. I. Ederra, and R. Gonzalo, "Decoupling of Multifrequency Dipole Antenna Arrays for Mi- crowave Imaging Applications," Int. J. Antenna Propag., Nov. 2009, org/10.1155/2010/843624. [3] E. Saenz, K. Guven, E. Ozbay, [4] J. H. Chou, H. J. Li, D. B. Limg, and C. Y. Wu, "A novel LTE MIMO antenna with decoupling element for mobile phone application," in 2014 Int. Symposium Electromagn. Compat., Tokyo, Japan, 2014. [5] S. M. Wang, L. T. Hwang, C. J. Lee, C. Y. Hsu, and F. S. Chang, "MIMO antenna design with built-in decoupling mechanism for WLAN dual-band applications," Electronics Lett.., vol. 51, no. 13, pp. 966–968, June. 2015. [6] Q. Li, "Miniaturized DGS and EBG structures for decoupling multiple antennas on compact wireless terminals," Ph.D. dissertation, Dept. Elect. Eng., Loughborough Univ., Loughborough, UK, 2012. [7] A. A. Hurshkainen, T. A. Derzhavskaya, S. B. Glybovski, I. J. Voogt, I. V. Melchakova, C. A. T. van den Berg, and A. J. E. Raaijmakers, "Element decoupling of 7 T dipole body arrays by EBG metasurface structures: Experimental verification," J. Magn. Res., vol. 269, pp. 87– 96, Aug. 2016. [8] F. Yang, and Y. Rahmat-Samii, "Microstrip Antennas Integrated With Electromagnetic Band-Gap (EBG) Structures: A Low Mutual Coupling Design for Array Applications," IEEE Trans. Antennas Propag., vol. 51, pp. 2936-2944, Oct. 2003. [9] M. F. Abedin and M. Ali, "Effects of a Smaller Unit Cell Planar EBG Structure on the Mutual Coupling of a Printed Dipole Array," IEEE Antennas Wirel. Propag. Lett., vol. 4, pp. 274-276, Feb. 2005. [10] RF Coils for MRI, J. T. Vaughan and J. R. Griffiths, Editors, Wiley, Chichester, UK, 2012. [11] P. B. Roemer, W. A. Edelstein, C. E. Hayes, S. P. Souza, O. M. Mueller, "The NMR phased array," Magnetic Resonance in Medicine, vol. 16, no. 2, pp. 192225, Nov. 1990. doi:10.1002/mrm.1910160203. [12] N. I. Avdievich, J. W. Pan and H. P. Hetherington, "Resonant inductive decoupling (RID) for transceiver arrays to compensate for both reactive and resistive components of the mutual impedance," NMR Biomed., vol. 26, 1547-1554, Nov. 2016. [13] C. von Morze, J. Tropp, S. Banerjee, D. Xu, K. Karpodinis, L. Carvajal, C. P. Hess, P. Mukherjee, S. Majumdar, and D. B. Vigneron, "An eight- channel, nonoverlapping phased array coil with capacitive decoupling for parallel MRI at 3 T," Concepts in Magnetic Resonance B: Magnetic Resonance Engineering, vol. 31, pp. 3743(1-5), 2007. [14] X. Yan, X. Zhang, L. Wei, and R. Xue, "Magnetic wall decoupling method for monopole coil array in ultrahigh field MRI: a feasibility test," Quantitative Imaging in Medicine and Surgery, vol. 214, pp. 79- 86, April 2014. [15] E. Georget, M. Luong, A. Vignaud, E. Giacomini, E. Chazel, G. Ferrand, A. Amadon, F. Mauconduit, S. Enoch, G. Tayeb, M. Bonod, C. Poupon, and R. Abdeddaim, "Stacked Magnetic Resonators for MRI RF Coils Decoupling," Journal of Magnetic Resonance, vol. 275, pp. 11-18, Nov. 2016. 7 [16] X. Yan, X. Zhang, B. Feng, C. Ma, L. Wei, and R. Xue, "7T trans- mit/receive arrays using ICE decoupling for human head MR imaging," IEEE Trans. Med. Imag., vol. 33 pp. 17811787, Sep. 2014. [17] C. T. Tai, "Coupled Antennas," Proc. IRE., vol. 36, pp. 487–500, Apr. [18] R. W. King, "The Theory of Linear Antennas," Cambridge, MA, USA: 1948. Harvard University, 1956. Prentice Hall, 1981, p. 225. York, NY, USA: Wiley, 2016. [19] R. S. Elliot, "Antenna Theory and Design," New York, NY, USA: [20] C. A. Balanis, "Antenna Theory: Analysis and Design," 4th ed., New [21] R. A. McConnell, "The effect of dipole coupled impedances on cali- bration," in Proc. IEEE Int. Conf. Electromagn. Compat., Denver, CO., USA, 1989, pp. 16–18. [22] A. Kazemipour and X. Begaud, "Numerical and analytical model of standard antennas from 30 MHz to 2 GHz," in Proc. IEEE Int. Symposium Electromagn. Compat., Montreal, Que., Canada, 2001, pp. 620–625. [23] J. Rahola and J. Ollikainen, "Removing the effect of antenna matching in isolation analyses using the concept of electromagnetic isolation," in Ant. Tech. Small Ant. Novel Meta., 2008, pp. 554–557.
1805.10589	1	1805	2018-05-27T06:57:59	Thickness dependence of space-charge-limited current in spatially disordered organic semiconductors	[ "physics.app-ph", "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Charge transport properties in organic semiconductors are determined by two kinds of microscopic disorders, namely energetic disorder related to the distribution of localized states and the spatial disorder related to the morphological features of the material. From a semi-classical picture, the charge transport properties are crucially determined by both the carrier mobility and the electrostatic field distribution in the material. Although the effect of disorders on carrier mobility has been widely studied, how electrostatic field distribution is distorted by the presence of disorders and its effect on charge transport remain unanswered. In this paper, we present a modified space-charge-limited current (SCLC) model for spatially disordered organic semiconductors based on the fractional-dimensional electrostatic framework. We show that the thickness dependence of SCLC is related to the spatial disorder in organic semiconductors. For trap-free transport, the SCLC exhibits a modified thickness scaling of $J\propto L^{-3\alpha}$, where the fractional-dimension parameter $\alpha$ accounts for the spatial disorder in organic semiconductors. The trap-limited and field-dependent mobility are also shown to obey an $\alpha$-dependent thickness scaling. The modified SCLC model shows a good agreement with several experiments on spatially disordered organic semiconductors. By applying this model to the experimental data, the standard charge transport parameters can be deduced with better accuracy than by using existing models.	physics.app-ph	physics	IEEE TRANSACTIONS ON ELECTRON DEVICES, 2018 1 Thickness dependence of space-charge-limited current in spatially disordered organic semiconductors Muhammad Zubair, Member, IEEE, Yee Sin Ang, and Lay Kee Ang, Senior Member, IEEE 8 1 0 2 y a M 7 2 ] h p - p p a . s c i s y h p [ 1 v 9 8 5 0 1 . 5 0 8 1 : v i X r a Abstract- Charge transport properties in organic semi- conductors are determined by two kinds of microscopic disorders, namely energetic disorder related to the dis- tribution of localized states and the spatial disorder re- lated to the morphological features of the material. From a semi-classical picture, the charge transport properties are crucially determined by both the carrier mobility and the electrostatic field distribution in the material. Although the effect of disorders on carrier mobility has been widely studied, how electrostatic field distribution is distorted by the presence of disorders and its effect on charge transport remain unanswered. In this paper, we present a modified space-charge-limited current (SCLC) model for spatially disordered organic semiconductors based on the fractional-dimensional electrostatic framework. We show that the thickness dependence of SCLC is related to the spatial disorder in organic semiconductors. For trap-free transport, the SCLC exhibits a modified thickness scaling of J ∝ L−3α, where the fractional-dimension parameter α accounts for the spatial disorder in organic semiconduc- tors. The trap-limited and field-dependent mobility are also shown to obey an α-dependent thickness scaling. The mod- ified SCLC model shows a good agreement with several experiments on spatially disordered organic semiconduc- tors. By applying this model to the experimental data, the standard charge transport parameters can be deduced with better accuracy than by using existing models. I. INTRODUCTION The mobility of charge carrier is a key parameter for the performance of optoelectronic devices [2], especially for devices using organic semiconductors and polymers. The mobility in organic semiconductors strongly depends on the nature, structure, purity of the materials and device operat- ing conditions. The charge transport in organic compounds occurs across various levels, ranging from within molecules, between molecules as well as between crystalline grains and amorphous and crystalline regions. The transport properties Muhammad Zubair is with the Department of Electrical Engineering at Information Technology University of the Punjab, Ferozpur Road, 54000 Lahore, Pakistan. He was affiliated until recently with the SUTD-MIT International Design Center, Singapore University of Technology and Design, 487372 Singapore (e-mail: [email protected]). Yee Sin Ang, and Lay Kee Ang are affiliated with the SUTD- MIT International Design Center, Singapore University of Technol- ogy and Design, 487372 Singapore (e-mails: yee [email protected], ricky [email protected]). The authors are thankful to P. W. M. Blom for providing the experi- mental data of Ref. [1] and helpful discussion. This work is sponsored by USA AFOSR AOARD (FA2386-14-1-4020), Singapore Temasek Lab- oratories (IGDS S16 02 05 1) and A*STAR IRG (A1783c0011). are determined by two kinds of microscopic disorders, namely the energetic disorder characterized by a broad distribution of localized states and the spatial disorder related to the morphological features of the material [3]. The space-charge- limited current (SCLC) is an important classical transport phenomenon in organic semiconductors where the quantum effects can be ignored at microscale and above [1]–[17]. The mobility of a given organic material sandwiched be- tween two planar electrodes with an applied voltage (V0), is commonly measured indirectly by fitting the measured current density-voltage (J-V) characteristics at high voltages to some SCLC models [4]. It is assumed that there is no barrier (ohmic contact) at the interface when the charges are injected from the electrode into the solid. The simplest SCLC model for a trap-free solid is known as the one-dimensional (1D) classical Mott-Gurney (MG) law [5], given by J = 9 8 µ V 2 0 L3 , (1) where = 0r is permittivity, and L is the thickness of the solid embedded between the metal electrodes. Once the values of J, V0, L and are determined, the mobility µ can be calculated by fitting the J-V characteristics at high V0 region to the 1D MG law. Note the assumptions in using the MG law include constant mobility (independent of the applied electric field and charge density) and the solid is trap-free. For a trap-filled solid with exponentially energy-distributed traps, the corresponding SCLC model is known as the trap- limited (TL) SCLC model [6]: (cid:21)l(cid:18) 2l + 1 (cid:19)l+1 V l+1 (cid:20) J = Ncµe1−l l Nt(l + 1) l + 1 0 L2l+1 , (2) where Nc is the effective density of states corresponding to the energy at the bottom of the conduction band, Nt is the total trapped electron density, and l = Tt/T ≥ 1 with Tt being a parameter controlling the trap distribution. For shallow traps or energetic disorder, the mobility varies with the electric field E that Murgatroyd's [7] model may be used to describe a field-dependent mobility in the form of √ µ = µ0 exp(γ E), (3) where µ0 is representing the mobility at zero field, and γ is a material-specific parameter that describes the strength of the field-dependence. The field-dependent mobility can 2 IEEE TRANSACTIONS ON ELECTRON DEVICES, 2018 also include other effects, such as carrier-density dependence (Gaussian disorder model (GDM) [8]) and deep traps, which can only be resolved by having a more comprehensive model to fit with experiments over a wide range of parameters [9]. By using Eq. (3), we get [7] J = 9 8 µ0 V 2 0 L3 exp(0.89γ √ E), (4) where 0.89γ is used as a fitting parameter. Other than field-dependent mobility, charge-carrier-density dependent mobility models have also been studied including a power-law dependence for energetically disordered semicon- ductors by Tanase et. al. [10], Blom et. al. [1], and others for disordered polymers [11]–[13]. The mobility is often extracted from J-V measurements by fitting the experimental data with the theoretical models of SCLC with different mobility terms. The mobility of a given sample is determined solely based on the goodness of the fit. Such empirical methodology may not always produce accurate physical picture. For example, an inconsistency of field-dependent mobility model with the experimental data has been raised in Ref. [1]. The conventional model of Eq. (4) could not fit the experimental data due to weaker thickness dependence of measured SCLC, and hence the carrier-density dependent mobility model was chosen to extract the mobility of PPV based diode device. However, assuming Gaussian density of states (DOS), in low carrier- density regime the mobility should be carrier-density inde- pendent [14]. Another recent experiment [3] has also reported that the charge transport in amorphous semiconductors is not charge density dependent but instead follows a field-dependent mobility model. In such scenarios, a new model of space- charge-limited transport is required to capture the correct thickness scaling of measured SCLC. is known that it The transport sites in organic semiconductors are distributed both in space and energy. The combined effect of spatial and energetic disorder on charge transport has already been studied in the previous works (see [18] for a comprehen- sive review). However, the SCLC is a transport phenomenon closely related with both the material properties (i.e., mobility) and the electrostatic field distribution inside these materials, and thus far, the non-uniform distribution of electrostatic field due to spatial complexity of the material and its implications on the macroscopic SCLC have not been fully addressed. The complex, spatially disordered and often self-organized microstructure, in which ordered microcrystalline domains are embedded in an amorphous domain, can be considered as fractal features having important consequences for electrical properties of these materials (refer to Fig. 2 in [19] and Fig. 1 in [20]). In this work, we present a modified SCLC model to account for the spatial disorder effect of a solid such as amorphous semiconducting polymer by treating the material as a fractal object. The key novelty of our proposed model is to utilize the fractional-dimension of space to effectively model the non- uniform distribution of electrostatic field inside these spatially disordered materials. Fig. (1) provides a schematic description of the concept that the spatially disordered organic semicon- ductor in real integer-dimensional space can be considered effectively as spatially ordered organic semiconductor embed- ded in the corresponding fractional-dimensional space using mathematical framework briefly introduced in the Appendix (also see [21], [22] and references therein). Such methods have been applied in other areas including quantum field theory [23], [24], general relativity [25], thermodynamics [26], mechanics [27], hydrodynamics [28], electrodynamics [29]– [38], and fractional charge transport [21], [22] to name a few. The proposed approach has been generalized to cover three types of SCLC models: trap-free model (or MG law), trap- limited (TL) model and the field dependent mobility models. We first analyze various experimental results to study the thickness (or L) dependence to show that the traditional L scalings from the traditional models are not valid for spatially disordered semiconducting materials. By using our proposed models, we are able to reproduce the experimental current- voltage measurements in [1] without using the carrier-density dependent model, and thus solving the issues raised by recent paper [3]. II. DERIVATION OF SCLC SCALING LAWS FOR SPATIALLY DISORDERED ORGANIC SEMICONDUCTORS A. Trap-free model Here, we derive the modified MG law for spatially disor- dered organic semiconductors with the assumption that the effect of non-uniform electrostatic field distribution inside spatially disordered material can be studied effectively by replacing the governing equations of classical SCLC model with the fractional-dimensional counterparts, by using the formulation described in Appendix (for more details see [21], [22] and references therein), where the fractional-dimension is related to the amount of spatial disorder. Provided that the thermal carriers are negligible in comparison to injected carrier, and assuming that the size of the electrode is much greater than the spacing L, thus the derivation is conducted only in x direction perpendicular to surface of the electrode. The following equations are solved in the α-dimensional space [39] with 0 < α ≤ 1: the concept Schematic description of Fig. 1. the spatially disordered organic semiconductor in real space [shown on left] can be considered effectively as spatially ordered organic semiconductor embedded in the corresponding fractional-dimensional space [shown on right] using fractional-dimensional space framework as description of complexity described in Appendix. Xk corresponds to the spatial coordinates of an α-dimensional space. that !"#$%&%'!!"#"#$#()!"()$%&%'!!$%#"()$µµZUBAIR et al.: SPACE-CHARGE-LIMITED TRANSPORT IN SPATIALLY DISORDERED ORGANIC SEMICONDUCTORS 3 J = −ρν, ∇α · E = ρ dE , dx E = −∇αV = − 1 dV dx c(α, x) c(α, x) = 1 , (5) (6) (7) Γ(α/2)xα−1 [21], ρ is the carrier charge where c(α, x) = πα/2 density, ν is the drift velocity, E is the electric field, and is the dielectric permittivity of the material, and V is the electric potential. Using ν = −µE, Eq. (5) gives J = ρµE. (8) It should be emphasized that µ is an averaged quantity independent of space variables. Now, inserting Eq. (6) into Eq. (8), we get J = µ 1 c(α, x) E dE dx , (9) and which can be rewritten in the form of Bernoulli differential equation: dE dx = c(α, x) µ JE−1. (10) Solving Eq. (10) with zero electric field condition (at SCL regime) at the injecting electrode, E(0) = 0, we have where l = Tc/T and Nc is effective density of states. In this case the relation between free n and trapped carrier density nt is given by , (16) where, N is total density of transport sites. By solving Eq. (6), the governing equation is (cid:18) nt (cid:19)l Nt n N = (cid:18) J Ncµ F = xα−1 = F dx el−1/l 0 Nt/ [E(x)]1/l πα/2 Γ(α/2) , . dE(x) (cid:19)1/l (cid:18) l + 1 (cid:19) l l+1 Integrating Eq. (17) on both sides gives E(x) = V (x) = πα/2 Γ(α/2) F α αl l+1 , x E(x)xα−1. l (cid:90) (cid:20) Γ(α/2) (cid:21)2l+1(cid:20) (cid:19)l+1 V l+1 πα/2 0 L2αl+α , (cid:21)l αl Nt(l + 1) J = Ncµe1−l (cid:18) 2αl + α l + 1 × The analytical evaluation of above integral leads to By using Eq. (11) in solving Eq. (7), we obtain E(x) = 2Jxαπα/2 αµΓ(α/2) . (cid:115) (cid:115)(cid:20) πα/2 (cid:20) Γ(α/2) πα/2 (cid:21)3 J (cid:21)3 µ J = 9α3 8 V 2 0 L3α . V (x) = Γ(α/2) µ x3α. which gives the modified MG law as a function of α: (17) (18) (19) (20) (21) (cid:21)l (23) (11) (12) (13) which reduces to Eq. (2) at α = 1. It should be noted that for a Gaussian distribution of traps, a similar equation for the trap-limited current is derived, except that l is then related to the depth and width of the trap distribution [40], [41]. In the case of Gaussian trap DOS centered at a distance Etc − Ea below the conduction-band edge ,the nondegenerate approximation gives [42] n N = nt Nt exp((Etc − Ea)/kTc) , (22) where, Ea = σ2 2kT , σ is the variance of Gaussian DOS. Finally, following the MH formalism, a current-voltage characteristic is obtained for Gaussian trap DOS, which is of the form J = Ncµe1−l (cid:19)l (cid:18) αl Nt exp((Etc − Ea)/kTc)(l + 1) (cid:20) Γ(α/2) (cid:18) 2αl + α (cid:21)2l+1(cid:20) (cid:19)l+1 V l+1 πα/2 l + 1 0 L2αl+α , × × which reduces to Eq. (7) in [42] at α = 1. C. Field-dependent mobility model If we simply combine the field-dependent mobility Eq. (3), and Eq. (13), the modified SCL model of field-dependent mobility for spatially disordered semiconductors is (cid:20) Γ(α/2) (cid:21)3 πα/2 µ0 V 2 0 L3α exp(0.89γ √ E). (24) For α = 1, Eq. (13) reduces to the classical MG Law [Eq. (1)]. B. Trap-limited (TL) model For a spatially disordered material with exponentially dis- tributed traps in energy, the trap-limited TL-SCLC injection is derived here. We assume that the mobility is field-independent, and that the density of the trapping states per unit energy range h(E) above the valence band is described by the distribution h(E) = (Nt/kTc) exp(−E/kTc), (14) where E is the energy measured upward from the top of the valence band, Nt is the total trap density, and Tc is a characteristic constant of the distribution. Following the Mark and Helfrich (MH) approach [6], we obtain (cid:18) J (cid:19)1/l Ncµ ρ(x) = el−1/l 0 NtE−1/l, (15) J = 9α3 8 4 IEEE TRANSACTIONS ON ELECTRON DEVICES, 2018 where γ is just a fitting parameter. In general, to include the field dependence of the mobility in SCLC model, coupled equations such as Eqs. (5-7) must be solved consistently [43]. It is however possible to derive an analytic solution if the field dependence of the drift mobility can be expressed in power law [44] given by with µ = µ0 at E = E0. By using this power law of mobility, we solve Eqs. (5-7) to obtain an analytical solution of (25) , E0 µ = µ0 (cid:19)n (cid:18) E (cid:21)n+3(cid:20) nα − n + α (cid:20) Γ(α/2) (cid:21)n+2 (cid:20) 2nα + 3α − n n + 2 (cid:21) µ0 En 0 × πα/2 n + 2 J = V n+2 0 L2nα+3α−n , (26) which reduces to Eq. (13) at n = 0. III. RESULTS AND DISCUSSIONS By analyzing the thickness (L) dependence of the classical SCLC models, we see the dependence of L−3 (at fixed V ), L−2l−1 (at fixed V ) and L−1 (at fixed E), respectively, from Eq. (1), Eq. (2), and Eq. (4). However, as predicted by corresponding modified SCLC models in Sec. II, the thick- ness dependence will be reduced by the fractional-dimension parameter α, which accounts to the spatial disorder in the underlying solids. In other words, the thickness dependence of the modified SCLC models will provide a tool to characterize the spatial disorder in the porous organic semiconductor. A. Implications of modified SCLC model on mobility extraction Before proceeding with the analysis of thickness depen- dence in some reported experimental data, it would be of interest to see the effect of variation in thickness dependence due to spatial disorder in the semiconductors on the mobility values extracted form experimental J-V curves taken from [1]. We denote the extracted values of mobility by µ, to distinguish them from actual mobility µ for this device. In Fig. (2), the extracted mobility (µ) is plotted as a function of thickness dependence parameter 3α. The fractional-dimension parameter α corresponds to the measure of spatial disorder in the semiconductor, with α = 1 corresponding to zero spatial disorder. It can be seen that µ is sensitively influenced by α. Thus it is important to check the L dependence rather than assuming α = 1 which may no longer be valid for complicated materials such as porous and amorphous organic materials. Fig. 2. The variation of the mobilities of PPV based devices as a function of varying thickness dependence (L3α). It is shown that a variation in thickness dependence leads to several orders of under- or over-estimation of mobility values. The J-V data is taken from [1] for L = 950 nm. thickness for a range of disordered organic semiconductors. It is observed that the thickness scaling of SCLC varies as predicted by our model, to which not much attention was paid previously and it was assumed trivially that thickness scaling follows standard MG law which turns out to be not true for organic semiconductor in several example cases reported in the following. Fig. (3) shows the corresponding thickness L dependence for various devices using different organic materials. The results shown in Fig. (3a-3d) are at fixed voltage regime (constant mobility) with varying thickness L. Based on the classical MG law, we will expect a scaling of L−3. However, due to spatial disorder, the results show a weaker thickness dependence in the range L−2 to L−3, which corresponds to about α ≈ 0.8 to α ≈ 1. (a) (c) (b) (d) B. Consistency of α extracted from experimental data Most organic semiconductors have spatial as well as en- ergetic disorder. The existing mobility models incorporate combined effect of energetic and spatial disorder using a range of mobility models including field- and density-dependent mobility. However, our model predicts that the spatial dis- order can affect the thickness scaling of SCLC. Here, we explore the available experimental data of SCLC versus device Fig. 3. The thickness dependency of the current density at fixed voltage for different polymers taken from current-voltage characteristics in the literature: (a) N RS − P P V [1] (b) P F O [45] (c) α-conjugated- Sexithienyl [46] (d) poly-fluorene-based [47]. For the results shown in Fig. (4a) for a trap-filled organic material we have l = Tt/T = 1500/273 = 5.49. Based on the classical TL-SCLC model (without any spatial disorder), the thickness dependence should be L−11.98. However, the 11.522.53thickness dependence (3)10-1010-5100102103L [nm], V0=1 [V]10-410-2100J [A/m2]expslope -2.8950.05 NRS-PPV102103L [nm], V0=1 [V]10-210-1100J [A/m2]expslope -0.60.04 PFO10-1100101L [m], V0=1 [V]10-5100J [A/cm2]expslope -2.60.4 -conjugated-Sexithieny50100150200L [nm], V0= 1 [V]10-210-1100101J [A/m2]expslope -2.60.4 polyfluorene-based copolymerZUBAIR et al.: SPACE-CHARGE-LIMITED TRANSPORT IN SPATIALLY DISORDERED ORGANIC SEMICONDUCTORS 5 THICKNESS DEPENDENCE OF ORGANIC SEMICONDUCTOR DEVICES BASED ON DISORDERED SEMICONDUCTING ORGANIC POLYMERS AND ITS RELATION TO PARAMETER α USED IN MODIFIED SCLC MODELS. TABLE I Polymer Type NRS-PPV PFO α-conjugated-Sexithienyl poly-fluorene-based OC1C10 − P P V NPB Average thickness dependence (extracted from experiment) 2.895 (Fig. (3a)) 2.9 (Fig. (3b)) 2.6 (Fig. (3c)) 2.5 (Fig. (3d)) 11 (Fig. (4a)) 0.52 (Fig. (4b)) Thickness dependence relation (modified MG law) Calculated α (measure of spatial disorder) Ref. (experimental data) 3α (Eq. (19)) 3α 3α 3α 2αl+α (Eq. (27)) 3α-2 (Eq. (30)) 0.965 0.967 0.86 0.83 0.918 0.84 [1] [45] [46] [47] [48] [49] experimental fitting shows again a weaker dependence, which corresponds to α = 0.918 based on Eq. (21), instead of α = 1. For results shown in Fig. (4b), the L scaling is calculated from measurements at fixed electric field. As mentioned earlier for field-dependent mobility, the thickness dependence should be L−1 at a fixed field for negligible spatial disorder at α = 1. However, we observe a weaker dependence, which corresponds to α = 0.86 based on Eq.(24). Table I summarizes the results of Fig. (3-4) for various disordered organic semi- conductor based devices and its relation to the α parameter used in our models to fit with the experimental results. Our analysis suggests that the traditional L scaling formulated in the classical SCLC models may not be suitable for organic semiconductors, and it will provide an inaccurate estimation of the mobility if such models are used. Note that the uncertainty in the measurement of L, which is about 5 nm from normal experimental setup, is not able to explain the variation from the expected α = 1 assuming the classical models are correct. SCLC varies at different applied voltages due to exp(γ(cid:112)V /L) slope of J versus L shows that the thickness dependence of factor in field-dependent mobility model of Eq. (24). Fig. (5b) shows the extracted thickness dependence at different voltages which immediately reveals that the value of the extracted α is inconsistent at different voltage. At high-voltage regime where field-dependence becomes non-negligible, the extracted thickness dependence even becomes stronger than L−3 which leads to an unphysical value of α > 1. This clearly reveals the fallacy of extracting α from the J-L curve at fixed voltage. Instead, the α should be extracted at fixed electric field strength as indicated by Eq. (24), i.e., J ∝ L3α−2 at fixed E. Fig. (5c) and (5d) shows the J-L characteristics and the extracted α at different E, respectively. In this case, a singular value of α ≈ 0.84 is extracted for all applied electric field strengths. More importantly, this value of α is consistent with that extracted from the low-voltage regime of Fig. (5b). (a) (b) Fig. 4. The thickness dependency of (a) trap-limited current density at fixed voltage for polymer OC1C10 − P P V taken from current-voltage characteristics in the literature [48] (b) current density at fixed electric field for polymer N P B taken from current-field characteristics in the literature [49]. In Fig. (3), we have extracted the thickness scaling of SCLC at low voltages to avoid the field-dependent SCLC regime. However, one must be careful while extracting the slopes at higher applied voltages. As the SCLC is field- dependent at high applied voltage, the extraction of α should be performed at fixed electric field strength, E, rather than at fixed voltage, V . To demonstrate this, we analyze the SCLC versus voltage data of polymer NPB based devices reported in Ref. [49], and plot the current density against device thickness at different voltages in Fig. (5a) . The varying (a) (c) (b) (d) Experimental data for polymer N P B based devices taken Fig. 5. from [49]. (a) The current density versus device thickness for varying voltages (applied voltage V and slope is shown in legend). (b) The extracted thickness dependence and the parameter α at different volt- ages. (c) The current density versus device thickness for varying electric field (applied electric field E and slope is shown in legend). (d) The extracted thickness dependence and the fractional dimension parameter α at different applied electric fields. 100200300400500L [nm], V0=13 [V]10-5100105J [A/m2]expslope -110.5 OC1C10-PPV102103L [nm], E = 0.2M [V/cm]101102103J [mA/cm2]expslope -0.520.22 NPB102103L [nm]100105J [mA/cm2]6.67 V, slope = -3.66 V, slope = -3.35 V, slope = -34 V, slope = -2.93 V, slope = -2.82 V, slope = -2.61.5 V, slope = -2.511234567V [V]2.533.54extracted L-dependence (3)00.511.522.5extracted 102103L [nm]101102103J [mA/cm2]0.2M [V/cm], slope = -0.520.175M [V/cm], slope = -0.490.15M [V/cm], slope = -0.510.125M [V/cm], slope = -0.480.1M [V/cm], slope = -0.510.10.120.140.160.180.2E [MV/cm]00.20.40.60.81extracted L-dependence (3-2)00.20.40.60.81extracted 6 IEEE TRANSACTIONS ON ELECTRON DEVICES, 2018 In Fig. (6a), the room-temperature current density versus voltage characteristics data from Ref. [45] is shown for PFO diodes of varying thickness together with various numerical models calculations. It should be noted that the classical model of Eq. (4) requires different values of γ to be used in order to fit with the experimental data despite the fact that the devices are composed of the same type of polymer. To address this inconsistency, we fitted the experimental data using our modified SCLC model with α = 0.967 [extracted from Fig. (3b)]. Remarkably, our modified model is able to fit the experimental J-V curves of all devices using a singular consistent value of γ = 1.2 × 10−4(m/V )1/2. Similarly, in Fig. (6b) the room-temperature current density versus voltage data from Ref. [48] for OC1C10 − P P V diodes with varying thicknesses is shown. The classical model in Eq. (2) fails to reproduce the experimental results with Nt fixed for all L. In contrast, by using our modified trap-limited SLC model with α = 0.918 extracted from Fig. (4a), a much better agreement with experimental results is obtained at a fixed Nt. These results show that the modified MG law can sufficiently describe the thickness dependence of SCLC for given range of applied voltages. SCL models including field dependent mobility (dashed and solid lines) are able to provide better agreements without the needs to use carrier-dependent mobility assumption that have been debated in recent years [18]. It is important to note that one of the direct consequence of modified MG law in Eq. (24) is that the mobility can be considered to have a thickness dependence along with field-dependence given by E)L3−3α. In Fig. (7b) we show the field µ = µ0 exp(0.89γ and thickness dependent mobility values for the same NRS- PPV based devices [1] using this model with the parameters shown in figure caption. √ Finally, we analyzed the thickness dependence of exper- imentally measured SCLC in hole-only devices based on diketopyrrolopyrole-based polymer (PDPPDTSE) [50]. The thickness dependence of current density at fixed voltage for PDPPDTSE based devices taken from experimental current- voltage data is shown in Fig. (8a). The thickness scaling of SCLC from standard L−3 to L−1.07. The observed thickness dependence corresponds to spatial disorder parameter α = 0.3567. In order to validate our model we also compared the reported mobility values for varying thickness of devices with the mobility scaling predicted by our model. It is shown in Fig. (8b)that the thickness scaling of measured mobility is in good agreement with the one predicted by our model (L3−3α). (a) (b) (a) Room-temperature current density vs voltage characteristics Fig. 6. data from [45] for P F O diodes with thicknesses of 100 (red), 150 (black), and 250 (blue) nm, respectively. Experiment (circles), Eq. 4 (dotted lines), Eq. 24 (dashed lines), Eq. 26 (solid lines). The parameters used are γ = 1.2 × 10−4(m/V )1/2, µ0 = 1.3 × 10−9m2/V s, α = 0.967, E0 = 0.1/Lα, n = 0.088. (b) Room-temperature current density vs voltage characteristics data from [48] for OC1C10 − P P V diodes with thicknesses of 200 (red), 220 (black), 300 (blue), and 440 (green) nm respectively. Experiment (circles), Eq. 2 (dotted lines), Eq. 21 (solid lines). The parameters used are α = 0.918, Nt = 8.5 × 1023m−3, Tt = 1500K and zero-field mobility of 5 × 10−11m2/V s. C. Fitting experimental current-voltage characteristics and mobility extraction using modified SCLC model In Fig. (7a) the experimental J-V characteristics [1] (circles) of the NRS-PPV based devices are shown together with vari- ous numerical models calculations: (i) (dotted lines) classical MG law based field-dependent mobility model of Eq. (4), (ii) (dashed lines) modified MG law based field-dependent mobility model of Eq. (24), and (iii) (solid lines) modified MG law based field-dependent mobility model of Eq. (26). From the figure, it is clear that the classical model (dotted lines) does not have a good agreement with the experimental results. As shown in Fig. (3a), NRS-PPV based devices show a thickness dependence of L−2.895 which corresponds the two modified to α = 0.965. Using this α = 0.965, (a) (b) (a) Room-temperature current density vs voltage characteristics Fig. 7. data from [1] for NRS-PPV hole-only diodes with thicknesses of 200 (red), 560 (black), and 950 (blue) nm, respectively. Experiment (circles), Eq. 4 (dotted lines), Eq. 24 (dashed lines), Eq. 26 (solid lines). The parameters used are γ = 4 × 10−4(m/V )1/2, µ0 = 5 × 10−12m2/V s, α = 0.965, E0 = 0.1/Lα, n = 0.10. (b) The calculated mobility values for devices with varying thickness using µ = µ0 exp(0.89γ E)L3−3α. √ (a) (b) (a) The thickness dependence of current density at fixed Fig. 8. voltage for diketopyrrolopyrole-based polymer (PDPPDTSE) taken from experimental current-voltage data in the literature [50]. The observed thickness dependence corresponds to spatial disorder parameter α = 0.3567. (b) The thickness dependence of measured mobility agrees with the one predicted by our model (L3−3α = L1.929). 012345V [V]10-210-1100101102103104J [A/m2]L=100 nmL=150 nmL=230 nm PFO051015202530V [V]10-1510-1010-5100105J [A/m2]L=200 nmL=220 nmL=300 nmL=440 nm OC1C10-PPV01020304050607080V [V]10-610-410-2100102104J [A/m2]L=200 nmL=560 nmL=950 nm NRS-PPV103104(E [V/m])1/210-1210-11mobility () [m2/Vs]L=200 nmL=560 nmL=950 nm NRS-PPV102103104L [nm], V0=1 [V]10-410-310-210-1J [A/m2]expslope -1.070.04 PDPPDTSE100105L [nm]10-610-410-2100mobility () [cm2 /Vs]expslope 1.920.04 PDPPDTSEZUBAIR et al.: SPACE-CHARGE-LIMITED TRANSPORT IN SPATIALLY DISORDERED ORGANIC SEMICONDUCTORS 7 IV. SUMMARY In summary, we have presented a modified thickness scaling in SCLC model to account for the spatial disorder in organic semiconductors by introducing a parameter α to imagine the solid as a fractal object sandwiched between two electrodes. The model has included different effects such as trap-free, trap-limited and field-dependent mobility. To provide an easy access to the main results of this work, we have summarized the modified SCLC equations in Table II. An analysis of multiple experimental results from literature reveals that the classical SCLC models might lead to incorrect extraction of mobilities due to weak thickness dependence arising from spatial disorder. For such materials, our proposed model here would be a better choice to extract the mobility for spatially disordered organic materials as we have shown that the traditional thickness scaling is not valid anymore. By applying our model with field-dependent mobility, we are able to reproduce the experimental results of SCLC transport in PPV derivative based device without using the carrier-density dependent mobility [1], agreeable with a recent report for amorphous polymers [3]. Note that the thickness dependence had been reported in others works. For example, Brutting et. al (see Fig. 2a in Ref. [51]) reported a weaker thickness dependence for Alq light-emitting devices than the expected L−1 at fixed electric field. John et. al (see Fig. (5-6) in Ref. [52]) reported a varying thickness dependence (L−2.7±0.46 to L−3.14±0.7) for plasma polymerized pyrrole thin films. Boni et. al. (see Fig. 12 in Ref. [53]) also reported a possible weaker thickness dependence for PZT ferroelectric based devices. Macdonald et. al (see Fig. 1b in Ref. [54]) also reported a weaker thickness dependence due to non-planar electrodes in conducting the experiment using tip atomic force microscopy (cAFM). This is a geometrical effect producing weaker thickness dependence of organic semiconductor devices [55] and is different from the physics studied here. It should be emphasized that our proposed models are based on a planar-diode geometry, thus such non-planar geometrical effects are not included. The extension of our models into non-planar geometries will be pursued in future works. Moreover, in this work we obtain the parameter α from the length scaling of SCLC in the experimental results, however a complete microscopic model can be created in further extensions to determine α directly from the knowledge of disorder either spatial or energetic or both. APPENDIX FRACTIONAL-DIMENSIONAL SPACE FRAMEWORK AS DESCRIPTION OF COMPLEXITY There is an increasing interest in the fractional modeling of complexity in physical systems [56], [57]. In recent years, the concept of fractional-dimensional space has been used as an effective physical description of restraint conditions in complex physical systems [24], [29], [39]. The approaches to describe the fractional dimensions include fractal geom- etry [58], fractional calculus [59], [60], and the integration over fractional-dimensional space [23], [61]. The axiomatic basis of spaces with fractional dimension with Euclidean metric were introduced by Stillinger [23]. The fractional- dimensional generalization of first order Laplace operators was then reported by Zubair et. al. [29] as approximations of the square of the second-order Laplace operator introduced in [23], [24]. Recently, a fractal metric based approach is considered by Tarasov [39] which provides a complete generalization of first and second order Laplace operators. In this work, we have utilized Tarasov's approach to vector calaculus in fractional- dimensional spaces, which is summarized in the following. space (F α ⊆ En) In fractional-dimensional frame- work [39], it is convenient to work with physically dimen- sionless space variables x/R0 → x, y/R0 → y, z/R0 → z, r/R0 → r, where R0 is a characteristic size of considered model. This provides a dimensionless integration and differen- tiation in α-dimensional space which leads to correct physical dimensions of quantities. We define a differential operator in the form of ∂αk,xk = ∂ ∂Xk = 1 c(αk, xk) ∂ ∂xk , (27) where c(αk, xk) corresponds to the non-integer dimensionality along the Xk-axis and it is defined by [39] c(αk, xk) = παk/2 Γ(αk/2) xkαk−1. (28) For the case of spatially disordered semiconductor or porous solid, the system can be effectively modeled by replacing the anisotropy with an isotropic continuum in an α-dimensional space, with a parameter 0 < α ≤ 1 to measure the anisotropy or disorder of the material. Using the operators in Eq. (27), we can generalize vector differential operators in an α-dimensional space. The gradient of a scalar function ϕ(r) in fractional-dimensional space is ∇αϕ(r) = ek∂αk,xk ϕ(r), (29) where ek are unit base vectors of the Cartesian coordinate system. The divergence of the vector field f (r) = ekfk(r) is ∇α · f (r) = ∂αk,xk f (r). (30) The curl for the vector field f (r) is 3(cid:88) k=1 3(cid:88) k=1 3(cid:88) k,i,l=1 ∇α × f (r) = eiεikl∂αk,xk f (r), (31) where εikl is the Levi-Civita symbol. Using Eqs. (29) and (30), the scalar Laplacian in the fractional-dimensional-space is written as [39] ∇2 αϕ(r) = ∇α · ∇αϕ(r) 3(cid:88) = (cid:18) ∂2 1 c2(αk, xk) ∂x2 k k=1 (cid:19) (32) . − αk − 1 xk ∂ ∂xk 8 IEEE TRANSACTIONS ON ELECTRON DEVICES, 2018 SUMMARY OF MODIFIED SCLC MODELS FOR SPATIALLY DISORDERED ORGANIC SEMICONDUCTORS PROPOSED IN THIS WORK. SEE MAIN TEXT FOR COMPLETE DESCRIPTION OF TERMINOLOGIES. TABLE II Description trap-free trap-limited (exponential trap density) trap-limited (Gaussian trap density) field-dependent (exponential) field-dependent (power-law) πα/2 µ V 2 0 L3α Modified SCLC Model J = 9α3 8 (cid:105)3 (cid:104) Γ(α/2) J = Ncµe1−l(cid:104) Γ(α/2) (cid:105)2l+1(cid:104) J = Ncµe1−l ×(cid:104) Γ(α/2) (cid:105)2l+1(cid:104) (cid:104) Γ(α/2) (cid:105)3 (cid:104) Γ(α/2) (cid:105)n+3(cid:104) nα−n+α J = 9α3 8 πα/2 πα/2 πα/2 µ0 πα/2 n+2 J = µ0 En 0 √ V 2 0 L3α exp(0.89γ E) (cid:105) ×(cid:104) 2nα+3α−n n+2 (cid:105)n+2 V n+2 0 L2nα+3α−n (cid:105)l ×(cid:16) 2αl+α l+1 (cid:17)l+1 V l+1 αl Nt(l+1) Nt exp((Etc−Ea)/kTc)(l+1) αl L2αl+α 0 (cid:105)l ×(cid:16) 2αl+α l+1 (cid:17)l+1 V l+1 0 L2αl+α Eqs. (13) (21) (23) (24) (26) REFERENCES [1] P. Blom, C. Tanase, D. De Leeuw, and R. Coehoorn, "Thickness scaling of the space-charge-limited current in poly (p-phenylene vinylene)," Applied physics letters, vol. 86, no. 9, p. 092105, 2005. [2] M. Kuik, G.-J. A. Wetzelaer, H. T. Nicolai, N. I. Craciun, D. M. De Leeuw, and P. W. Blom, "25th anniversary article: Charge transport and recombination in polymer light-emitting diodes," Advanced Mate- rials, vol. 26, no. 4, pp. 512–531, 2014. [3] A. J. Campbell, R. Rawcliffe, A. Guite, J. C. D. Faria, A. Mukher- jee, M. A. McLachlan, M. Shkunov, and D. D. Bradley, "Charge- carrier density independent mobility in amorphous fluorene-triarylamine copolymers," Advanced Functional Materials, vol. 26, no. 21, pp. 3720– 3729, 2016. [4] M. A. Lampert and P. Mark, "Current injection in solids," 1970. [5] N. F. Mott and R. W. Gurney, Electronic processes in ionic crystals. Clarendon Press, 1948. [6] P. Mark and W. Helfrich, "Space-charge-limited currents in organic crystals," Journal of Applied Physics, vol. 33, no. 1, pp. 205–215, 1962. [7] P. Murgatroyd, "Theory of space-charge-limited current enhanced by frenkel effect," Journal of Physics D: Applied Physics, vol. 3, no. 2, p. 151, 1970. [8] H. Bassler, "Charge transport in disordered organic photoconductors a monte carlo simulation study," physica status solidi (b), vol. 175, no. 1, pp. 15–56, 1993. [9] J. C. Blakesley, F. A. Castro, W. Kylberg, G. F. Dibb, C. Arantes, R. Valaski, M. Cremona, J. S. Kim, and J.-S. Kim, "Towards reliable charge-mobility benchmark measurements for organic semiconductors," Organic Electronics, vol. 15, no. 6, pp. 1263–1272, 2014. [10] C. Tanase, E. Meijer, P. Blom, and D. De Leeuw, "Unification of the hole transport in polymeric field-effect transistors and light-emitting diodes," Physical Review Letters, vol. 91, no. 21, p. 216601, 2003. [11] I. Fishchuk, V. Arkhipov, A. Kadashchuk, P. Heremans, and H. Bassler, "Analytic model of hopping mobility at large charge carrier concentra- tions in disordered organic semiconductors: Polarons versus bare charge carriers," Physical Review B, vol. 76, no. 4, p. 045210, 2007. [12] W. Pasveer, J. Cottaar, C. Tanase, R. Coehoorn, P. Bobbert, P. Blom, D. De Leeuw, and M. Michels, "Unified description of charge-carrier mobilities in disordered semiconducting polymers," Physical review letters, vol. 94, no. 20, p. 206601, 2005. [13] J. Cottaar, L. Koster, R. Coehoorn, and P. Bobbert, "Scaling theory for percolative charge transport in disordered molecular semiconductors," Physical review letters, vol. 107, no. 13, p. 136601, 2011. [14] C. Tanase, P. Blom, D. De Leeuw, and E. Meijer, "Charge carrier density dependence of the hole mobility in poly (p-phenylene vinylene)," physica status solidi (a), vol. 201, no. 6, pp. 1236–1245, 2004. [15] I. Fishchuk, A. Kadashchuk, J. Genoe, M. Ullah, H. Sitter, T. B. Singh, N. Sariciftci, and H. Bassler, "Temperature dependence of the charge carrier mobility in disordered organic semiconductors at large carrier concentrations," Physical Review B, vol. 81, no. 4, p. 045202, 2010. [16] I. Katsouras, A. Najafi, K. Asadi, A. Kronemeijer, A. Oostra, L. Koster, in poly (p- D. M. de Leeuw, and P. W. Blom, "Charge transport phenylene vinylene) at low temperature and high electric field," Organic Electronics, vol. 14, no. 6, pp. 1591–1596, 2013. [17] T. Leijtens, J. Lim, J. Teuscher, T. Park, and H. J. Snaith, "Charge density dependent mobility of organic hole-transporters and mesoporous tio2 determined by transient mobility spectroscopy: Implications to dye- sensitized and organic solar cells," Advanced Materials, vol. 25, no. 23, pp. 3227–3233, 2013. [18] A. Nenashev, J. Oelerich, and S. Baranovskii, "Theoretical tools for the description of charge transport in disordered organic semiconductors," Journal of Physics: Condensed Matter, vol. 27, no. 9, p. 093201, 2015. [19] N. Tessler, Y. Preezant, N. Rappaport, and Y. Roichman, "Charge transport in disordered organic materials and its relevance to thin-film devices: A tutorial review," Advanced Materials, vol. 21, no. 27, pp. 2741–2761, 2009. [20] R. Noriega, J. Rivnay, K. Vandewal, F. P. Koch, N. Stingelin, P. Smith, M. F. Toney, and A. Salleo, "A general relationship between disorder, aggregation and charge transport in conjugated polymers," Nature ma- terials, vol. 12, no. 11, pp. 1038–1044, 2013. [21] M. Zubair, Y. S. Ang, and L. K. Ang, "Fractional fowler-nordheim law for field emission from rough surface with nonparabolic energy dispersion," IEEE Transactions on Electron Devices, vol. 65, no. 6, pp. 2089–2095, 2018. [22] M. Zubair and L. K. Ang, "Fractional-dimensional child-langmuir law for a rough cathode," Physics of Plasmas (1994-present), vol. 23, no. 7, p. 072118, 2016. [23] F. H. Stillinger, "Axiomatic basis for spaces with noninteger dimension," Journal of Mathematical Physics, vol. 18, no. 6, pp. 1224–1234, 1977. [24] C. Palmer and P. N. Stavrinou, "Equations of motion in a non-integer- dimensional space," Journal of Physics A: Mathematical and General, vol. 37, no. 27, p. 6987, 2004. [25] M. Sadallah and S. I. Muslih, "Solution of the equations of motion for einsteins field in fractional d dimensional space-time," International Journal of Theoretical Physics, vol. 48, no. 12, pp. 3312–3318, 2009. [26] V. E. Tarasov, "Heat transfer in fractal materials," International Journal of Heat and Mass Transfer, vol. 93, pp. 427–430, 2016. [27] M. Ostoja-Starzewski, J. Li, H. Joumaa, and P. N. Demmie, "From fractal media to continuum mechanics," ZAMM-Journal of Applied Mathematics and Mechanics/Zeitschrift fur Angewandte Mathematik und Mechanik, vol. 94, no. 5, pp. 373–401, 2014. [28] A. S. Balankin and B. E. Elizarraraz, "Map of fluid flow in fractal porous medium into fractal continuum flow," Physical Review E, vol. 85, no. 5, p. 056314, 2012. [29] M. Zubair, M. J. Mughal, and Q. A. Naqvi, Electromagnetic fields and waves in fractional dimensional space. Springer Science & Business Media, 2012. [30] M. J. Mughal and M. Zubair, "Fractional space solutions of antenna radi- ation problems: An application to hertzian dipole," in Signal Processing and Communications Applications (SIU), 2011 IEEE 19th Conference on. IEEE, 2011, pp. 62–65. [31] Q. A. Naqvi and M. Zubair, "On cylindrical model of electrostatic potential in fractional dimensional space," Optik-International Journal for Light and Electron Optics, vol. 127, no. 6, pp. 3243–3247, 2016. ZUBAIR et al.: SPACE-CHARGE-LIMITED TRANSPORT IN SPATIALLY DISORDERED ORGANIC SEMICONDUCTORS 9 [32] M. Zubair, M. J. Mughal, and Q. A. Naqvi, "An exact solution of the cylindrical wave equation for electromagnetic field in fractional dimensional space," Progress In Electromagnetics Research, vol. 114, pp. 443–455, 2011. [33] H. Asad, M. J. Mughal, M. Zubair, and Q. A. Naqvi, "Electromagnetic greens function for fractional space," Journal of Electromagnetic Waves and Applications, vol. 26, no. 14-15, pp. 1903–1910, 2012. [34] H. Asad, M. Zubair, and M. J. Mughal, "Reflection and transmission at dielectric-fractal interface," Progress In Electromagnetics Research, vol. 125, pp. 543–558, 2012. [35] M. Zubair, M. J. Mughal, and Q. A. Naqvi, "An exact solution of the spherical wave equation in d-dimensional fractional space," Journal of Electromagnetic Waves and Applications, vol. 25, no. 10, pp. 1481–1491, 2011. [36] M. Zubair, M. J. Mughal, Q. A. Naqvi, and A. A. Rizvi, "Differential electromagnetic equations in fractional space," Progress In Electromag- netics Research, vol. 114, pp. 255–269, 2011. [37] M. Zubair, M. J. Mughal, and Q. A. Naqvi, "On electromagnetic wave propagation in fractional space," Nonlinear Analysis: Real World Applications, vol. 12, no. 5, pp. 2844–2850, 2011. [38] M. Zubair, M. J. Mughal, and Q. A. Naqvi, "The wave equation and general plane wave solutions in fractional space," Progress In Electromagnetics Research Letters, vol. 19, pp. 137–146, 2010. [39] V. E. Tarasov, "Anisotropic fractal media by vector calculus in non- integer dimensional space," Journal of Mathematical Physics, vol. 55, no. 8, p. 083510, 2014. [40] W. Hwang and K. Kao, "Studies of the theory of single and double injec- tions in solids with a gaussian trap distribution," Solid-State Electronics, vol. 19, no. 12, pp. 1045–1047, 1976. [41] D. Abbaszadeh, A. Kunz, G. Wetzelaer, J. Michels, N. Cra?ciun, K. Koynov, I. Lieberwirth, and P. Blom, "Elimination of charge carrier trapping in diluted semiconductors," Nature materials, vol. 15, no. 6, pp. 628–633, 2016. [42] M. Mandoc, B. de Boer, G. Paasch, and P. Blom, "Trap-limited electron transport in disordered semiconducting polymers," Physical Review B, vol. 75, no. 19, p. 193202, 2007. [43] I. Chen, "A model of charge injection at metal-insulator contacts," Solid State Communications, vol. 26, no. 6, pp. 359–363, 1978. [44] M. Abkowitz, J. Facci, and M. Stolka, "Time-resolved space charge- limited injection in a trap-free glassy polymer," Chemical physics, vol. 177, no. 3, pp. 783–792, 1993. [45] H. Nicolai, G. Wetzelaer, M. Kuik, A. Kronemeijer, B. De Boer, and P. Blom, "Space-charge-limited hole current in poly (9, 9- dioctylfluorene) diodes," Applied Physics Letters, vol. 96, no. 17, p. 172107, 2010. [46] G. Horowitz, D. Fichou, X. Peng, and P. Delannoy, "Evidence for a lin- ear low-voltage space-charge-limited current in organic thin films. film thickness and temperature dependence in alpha-conjugated sexithienyl," Journal de Physique, vol. 51, no. 13, pp. 1489–1499, 1990. [47] R. Coehoorn, S. Vulto, S. Van Mensfoort, J. Billen, M. Bartyzel, H. Greiner, and R. Assent, "Measurement and modeling of carrier transport and exciton formation in blue polymer light emitting diodes," in Photonics Europe. International Society for Optics and Photonics, 2006, pp. 61 920O–61 920O. [48] M. Mandoc, B. De Boer, and P. Blom, "Electron-only diodes of poly (dialkoxy-p-phenylene vinylene) using hole-blocking bottom elec- trodes," Physical Review B, vol. 73, no. 15, p. 155205, 2006. [49] T.-Y. Chu and O.-K. Song, "Hole mobility of n, n'-bis (naphthalen-1-yl)- n, n'-bis (phenyl) benzidine investigated by using space-charge-limited currents," Applied physics letters, vol. 90, no. 20, pp. 203 512–203 512, 2007. [50] K. H. Cheon, J. Cho, B. T. Lim, H.-J. Yun, S.-K. Kwon, Y.-H. Kim, and D. S. Chung, "Analysis of charge transport in high-mobility diketopyrrolopyrole polymers by space charge limited current and time of flight methods," RSC Advances, vol. 4, no. 67, pp. 35 344–35 347, 2014. [51] W. Brutting, S. Berleb, and A. Muckl, "Space-charge limited conduction with a field and temperature dependent mobility in alq light-emitting devices," Synthetic Metals, vol. 122, no. 1, pp. 99–104, 2001. [52] J. John, S. Sivaraman, S. Jayalekshmy, and M. Anantharaman, "Inves- tigations on the mechanism of carrier transport in plasma polymerized pyrrole thin films," Journal of Physics and Chemistry of Solids, vol. 71, no. 7, pp. 935–939, 2010. [53] A. Boni, I. Pintilie, L. Pintilie, D. Preziosi, H. Deniz, and M. Alexe, "Electronic transport in (la, sr) mno3-ferroelectric-(la, sr) mno3 epitaxial structures," Journal of Applied Physics, vol. 113, no. 22, p. 224103, 2013. [54] G. A. MacDonald, P. A. Veneman, D. Placencia, and N. R. Arm- strong, "Electrical property heterogeneity at transparent conductive oxide/organic semiconductor interfaces: mapping contact ohmicity using conducting-tip atomic force microscopy," ACS nano, vol. 6, no. 11, pp. 9623–9636, 2012. [55] O. G. Reid, K. Munechika, and D. S. Ginger, "Space charge limited current measurements on conjugated polymer films using conductive atomic force microscopy," Nano letters, vol. 8, no. 6, pp. 1602–1609, 2008. [56] B. J. West, Fractional calculus view of complexity: Tomorrow's science. CRC Press, 2015. [57] V. E. Tarasov, Fractional dynamics: applications of fractional calculus to dynamics of particles, fields and media. Springer Science & Business Media, 2011. [58] K. Falconer, Fractal geometry: mathematical foundations and applica- tions. John Wiley & Sons, 2004. [59] K. B. Oldham and J. Spanier, The Fractional Calculus. Academic Press, New York, 1974. [60] G. Calcagni, "Geometry and field theory in multi-fractional spacetime," Journal of High Energy Physics, vol. 2012, no. 1, pp. 1–77, 2012. [61] A. S. Balankin, "Effective degrees of freedom of a random walk on a fractal," Physical Review E, vol. 92, no. 6, p. 062146, 2015. Muhammad Zubair (S'13-M'15) received his Ph.D. degree in electronic engineering from the Politecnico di Torino, Italy, in 2015. From 2015 to 2017, he was with the SUTD-MIT International Design Center, Singapore. Since 2017, he has been with Information Technology University, La- hore, Pakistan. His current research interests include charge transport, electron device modeling, computa- tional electromagnetics, fractal electrodynamics, and microwave imaging. Yee Sin Ang his bachelors degree in medical and radiation physics in 2010, and his PhD de- gree in theoretical condensed matter physics in 2014 from the University of Wollongong (UOW), Australia. He is currently a Research Fellow with the Singapore University of Technology and Design, Singapore. His research interests include the theory and mathematical modelling of electron emission phenomena in 2D and topological materials, electron transport physics across 2D/3D, 2D ma- terial valleytronics, nanoelectronics and superconducting devices. (S'95-M'00-SM'08) Lay Kee Ang received the B.S. degree from the Department of Nu- clear Engineering, National Tsing Hua Univer- sity, Hsinchu, Taiwan, in 1994, and the M.S. and Ph.D. degrees from the Department of Nuclear Engineering and Radiological Sciences, Univer- sity of Michigan, Ann Arbor, MI, USA, in 1996 and 1999, respectively. Since 2011, he has been with the Singapore University of Technology and Design, Singapore. He is currently the Interim Head and Professor of the Engineering Product Development pillar and also the Ng Teng Fong Chair Professor of SUTD-ZJU IDEA.
1908.01576	1	1908	2019-07-03T23:24:14	Highly-Linear Magnet-Free Microelectromechanical Circulators	[ "physics.app-ph", "eess.SP" ]	This paper reports the first demonstration of a magnet-free, high performance microelectromechanical system (MEMS) based circulator. An innovative circuit based on the commutation of MEMS resonators with high quality (Q) factor using RF switches is designed and implemented. Thanks to the high Q factor, a much smaller modulation frequency can be achieved compared to the previous demonstrations, reducing the power consumption and enabling the use of high power-handling switches. Furthermore, the MEMS resonators greatly reduce the required inductance value, guaranteeing much smaller form factor compared to the previous LC demonstrations. The demonstrated circulator shows broad BW (15 dB-IX BW=34.7 MHz for an operational frequency around 2.5 GHz), low IL (4 dB), high IX (30 dB), high linearity (P1dB=28 dBm; IIP3=40 dBm) and at the same time low power consumption, addressing several of the current limitations hindering the full development of magnet-free circulators.	physics.app-ph	physics	Submitted to IEEE Journal of Microelectromechanical Systems 1 Highly-Linear Magnet-Free Microelectromechanical Circulators Yao Yu, Student Member, IEEE, Giuseppe Michetti, Ahmed Kord, Student Member, IEEE, Michele Pirro, Dimitrios L. Sounas, Senior Member, IEEE, Zhicheng Xiao, Student Member, IEEE, Cristian Cassella, Member, IEEE, Andrea Alù, Fellow, IEEE, and Matteo Rinaldi, Senior Member, IEEE.  Abstract -- This paper reports the first demonstration of a magnet-free, high performance microelectromechanical system (MEMS) based circulator. An innovative circuit based on the commutation of MEMS resonators with high quality (Q) factor using RF switches is designed and implemented. Thanks to the high Q factor, a much smaller modulation frequency can be achieved compared to the previous demonstrations, reducing the power consumption and enabling the use of high power-handling switches. Furthermore, the MEMS resonators greatly reduce the required inductance value, guaranteeing much smaller form factor compared to the previous LC demonstrations. The demonstrated circulator shows broad BW (15 dB-IX BW=34.7 MHz for an operational frequency around 2.5 GHz), low IL (4 dB), high IX (30 dB), high linearity (P1dB=28 dBm; IIP3=40 dBm) and at the same time low power consumption, addressing several of the current limitations hindering the full development of magnet-free circulators. Index Terms -- AlN, Circulator, FBAR, MEMS, Non- reciprocity, Resonators. I. INTRODUCTION C URRENT wireless communication systems are half- duplex, transmitting and receiving at different time slots or using different carrier frequencies to avoid self-interference between the transmitter (Tx) and the receiver (Rx) modules. Driven by the need of enhancing the spectral efficiency in an overly crowded wireless spectrum, full-duplex operation has been attracting more and more attention over the past few years [1-6]. In full-duplex mode, the wireless terminals transmit and receive signals at the same time over the same frequency band. Circulators, which are three-port non-reciprocal components, are crucial to enable full-duplex operation, due to their ability of prohibiting self-interference between the Tx and Rx nodes without sacrificing insertion loss, since they allow the signal to propagate only in one direction (i.e., from Tx to antenna (ANT) and from ANT to Rx). (CMOS) into transmission three categories: Conventional circulators break time-reversal symmetry, and consequently reciprocity, by applying a strong magnetic bias to ferrite cavities [7-9]. However, magnetic-biased circulators are bulky and incompatible with conventional complementary metal-oxide-semiconductor technologies. Active devices, such as transistors, are intrinsically non-reciprocal and hence have been employed to eliminate the requirement on magnetic bias [10-12]. However, these active circulators suffer from poor noise and linearity performances. Recently, a new class of magnet-free circulators based on linear-periodically- time-variant (LPTV) circuits has been proposed [13-37]. In LPTV-based circulators, periodic spatiotemporal modulation is applied to the system to break reciprocity. Based on the components being modulated, LPTV-based circulators can be classified line (TL) circulators, lumped element (LC) circulators and MEMS-based circulators. Refs. [13-15] demonstrated TL-based circulators by separating signals being excited from different ports into either different paths using CMOS switches, or different frequencies using distributed modulated varactors. Even though strong non- reciprocity has been achieved with broad BW and low IL, these demonstrations all require large modulation frequencies. This requirement in turn poses three challenges to the system: (i) large power consumption, (ii) leakage of RF power into the modulation paths, and (iii) limited linearity due to the use of either highly-scaled CMOS switches (to improve the switching speed) or solid-state varactors, both of which are limited in linearity. Even though in [13] the linearity from Tx to ANT path was improved by suppressing the voltage swing on the CMOS switches in charge of the modulation, this operation leads to an asymmetrical circulator response and increased sensitivity to impedance mismatch at RF ports, therefore requiring off-chip impedance tuners. Refs. [16]-[18] demonstrated LC circulators based on angular momentum biasing. In these cases, the resonant frequencies of three connected resonant tanks are periodically modulated in a rotating fashion, to break the time- Manuscript received April 10, 2019. This work was supported by the Defense Advanced Research Projects Agency (DARPA) SPAR program under contract no. HR0011-17-2-0002 and the Air Force Office of Scientific Research. Y. Yu, G. Michetti, M. Pirro, C. Cassella and M. Rinaldi are with the SMART center, Northeastern University, Boston, MA 02115 USA. (corresponding author: M Rinaldi. Phone: 617-373-2751; Fax: 617-373-8970; E-mail: [email protected]). A. Kord and Z. Xiao are with the Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, TX78712, USA. D. L. Sounas is with the Department of Electrical and Computer Engineering, Wayne State University, Detroit, MI 48202, USA. A. Alù is with the Department of Electrical and Computer Engineering, University of Texas at Austin, TX78712, USA, and also with the Advanced Science Research Center, City University of New York, New York, NY 10031, USA. He is currently the Chief Technology Officer of Silicon Audio RF Circulator. The terms of this arrangement have been reviewed and approved by The University of Texas at Austin and the City University of New York in accordance with its policy on objectivity in research. Submitted to IEEE Journal of Microelectromechanical Systems reversal symmetry. Low-loss non-reciprocal bands have been achieved with broad BW. Nevertheless limited by the low Q factor of the LC systems, the modulation frequencies required to achieve significant non-reciprocity are at least 10% of RF frequencies. The use of inductors also limits the form factor and integrability of the system. Furthermore, the use of solid-state varactors again limits the systems' linearity and modulation network complexity. Compared to TL or LC components, MEMS resonators have much larger Q factor and better integrability, thus having the potential to greatly reduce the modulation frequency, improve the circulator's insertion loss (IL), isolation (IX) and form factor. In this context, some attempts have been done to implement MEMS circulators [19- 22]. However, these demonstrations have all been characterized by severe limitations including high IL, limited IX, narrow bandwidth (BW) and/or intermodulation distortion. In order to address the aforementioned challenges, we introduce in this paper the first demonstration of a magnet-free circulator using high-Q AlN film bulk acoustic resonators (FBARs) centered at 2.5 GHz, which are spatiotemporally modulated to break time-reversal symmetry. In addition to low IL and high IX, the high Q factor of the system also guarantees much lower modulation frequency requirements (1.6% of RF frequency) compared to previous demonstration, which reduces power consumption and enables the use of high power handling switches. A small modulation frequency generates closely spaced harmonics, which risk to induce significant signal distortion. We address this issue by considering a differential geometry, which has been shown largely suppress intermodulation mixing [17]. Instead of using conventional varactors, we modulate the center frequencies of the resonant tanks by commutating between two different frequencies using RF switches, therefore achieving better linearity (IIP3=40dBm; P1dB=28 dBm) compared to previous LC demonstrations based on varactors. In fact, thanks to the high Q system and innovative modulation scheme, the achieved linearity is among the highest for all the magnet-free circulators. Finally, in order to improve the BW, the effective electromechanical coupling coefficient (kt2) is increased by connecting inductors in parallel to FBARs to resonate out the static capacitance (C0). Using this principle, we achieve a seven times wider BW. Compared to previous demonstrations based on LC systems, the required inductance values are also greatly reduced (4 nH in this work while hundreds of nH for previous demonstrations), therefore guaranteeing a much smaller form factor and lower cost. to II. DESIGN A. Angular momentum biasing The reciprocity of a symmetrical three-port resonant network can be broken by applying an angular momentum bias to the system [16]-[19]. As shown in Fig. 1, three resonant tanks with center frequency f0 are connected together, and the center frequencies are modulated periodically with proper phase delay 𝜑(cid:3041), modulation amplitude (i.e. center frequency shift) am and angular modulation frequency ωm. The center frequencies of the three resonant tanks are given by 2 ω(𝑡) = ω(cid:2868) + 𝑎(cid:3040) × cos (𝜔(cid:3040)𝑡 + 𝜑(cid:3041)) (1) where ω(cid:2868) is the static center frequency, 𝜑(cid:3041) is the phase of the modulation of the n-th tank and 𝜑(cid:2869), 𝜑(cid:2870), 𝜑(cid:2871) are equal to 0(cid:2868), 120(cid:2868) and 240(cid:2868), respectively. Fig. 1 Schematic of angular momentum biasing through modulation. Three LC tanks are connected to a common node, and the center frequencies of these three tanks are periodically modulated in a rotating fashion to break the degeneracy of the two counter-rotating current mode (I+ and I-) to achieve non- Fig. 2 Equivalent circuit model of an FBAR. Fig. 3 Circuit simulation of the admittance of an FBAR connected in series to different values of capacitors. Inset: schematic of an FBAR connect in series to a capacitor. The currents flowing through the three resonant tanks can be seen as a superposition of two counter-rotating modes, I- and I+ Submitted to IEEE Journal of Microelectromechanical Systems 3 [16-18], as shown in Fig. 1. Without modulation, these two counter-rotating modes are degenerate, therefore if a signal is excited from one port, transmission to the other two ports is equal and the network is reciprocal. However, when modulation is applied to the system with a phase shift specified by (1), a preferred sense of rotation is applied to the rotating modes, since one of them rotates in the same direction as the modulation while the other one rotates in the opposite direction. Therefore, degeneracy is lifted and, by choosing proper modulation amplitude and two modes destructively interfere at one port and constructively interfere at the other, thus achieving the operation of a circulator. frequency, the Fig. 4 Circuit simulation of the admittance of an FBAR connected in parallel to an inductor to resonate out C0 and then connect in series to different values of capacitors. Inset: schematic of an FBAR connected in parallel to an inductor Lp and then connected in series to different values of capacitors. B. FBAR resonant frequency modulation The equivalent circuit model of an FBAR is shown in Fig. 2, where Lm, Cm, Rm and C0 are motional inductance, motional capacitance, motional resistance and static capacitance, respectively. The values of Lm, Cm and Rm can be calculated as (cid:2869) (cid:3095)(cid:3118) (cid:2876) 𝐶(cid:3040) = 𝑅(cid:3040) = (cid:3104)(cid:3116)(cid:3004)(cid:3116)(cid:3038)(cid:3047)(cid:3118)(cid:3018) (2) 𝐶(cid:2868)𝑘𝑡(cid:2870) (3) (cid:3118)(cid:3004)(cid:3116)(cid:3038)(cid:3047)(cid:3118) (4) where 𝜔(cid:2868) is the center frequency, kt2 is the electromechanical coupling coefficient and Q is the quality factor. 𝐿(cid:3040) = (cid:3095)(cid:3118) (cid:2876) (cid:3095)(cid:3118) (cid:2876) (cid:3104)(cid:3116) (cid:2869) The center frequency of the FBAR is determined by the resonance between Lm and Cm, and can be shifted by connecting the FBAR to a series capacitor Cs. As shown in [19], as long as Cm<<C0, the amount of frequency shift can be expressed by (cid:3104)(cid:3294) (cid:4594)(cid:2879)ω(cid:3294) ω(cid:3294) = ∆(cid:3104)(cid:3294) ω(cid:3294) ≅ (cid:3004)(cid:3288) (cid:2870)((cid:3004)(cid:3116)(cid:2878)(cid:3004)(cid:3294)) (5) (cid:4593) is the shifted where ω(cid:3046) is the unshifted center frequency and 𝜔(cid:3046) center frequency. Furthermore, the introduction of the series capacitor will decrease the admittance magnitude of the FBAR at resonance by a factor 𝜒 [19], i.e., Assuming that 1+ (cid:3004)(cid:3294) (cid:3004)(cid:3116) ≫ 𝑘(cid:3047) (cid:2870) and (1 + )(cid:2870) ≪ (𝑘𝑡(cid:2870)𝑄)(cid:2870) , the (cid:3004)(cid:3294) (cid:3004)(cid:3116) 𝑌′(cid:3007)(cid:3003)(cid:3002)(cid:3019)(𝜔(cid:3046) (cid:4593)) = 𝜒 ∙ 𝑌(cid:3007)(cid:3003)(cid:3002)(cid:3019)(ω(cid:3046)) (6) admittance amplitude reduction factor 𝜒 can be expressed as (cid:2870) (cid:3004)(cid:3294) (cid:4673) (cid:3004)(cid:3294)(cid:2878)(cid:3004)(cid:3116) 𝜒 ≈ (cid:4672) (7) The above analysis highlights two challenges regarding the frequency modulation of FBARs compared to the case of simple LC resonators: first, according to (5), the modulation amplitude is limited by Cm/C0, i.e., the kt2 of FBARs, which is the reason why all the previous MEMS-resonator based circulator demonstrations showed narrow BW. Second, the frequency shift will cause a reduction in admittance at resonance, which is also determined by the value of C0, thus causing higher IL of the circulator. Fig. 3 shows the simulated results of frequency shift when an FBAR is connected to a series capacitor. The FBAR used in simulation is assumed to have a Co of 1 pF kt2 of 3%, Q of 600 and 𝜔(cid:2868) of 2𝜋 × 2.5 GHz. Consistent with the above analysis, the modulation amplitude is limited by the anti-resonance peak determined by the kt2 of the FBAR, and the admittance at peak drops with the decrease of Cs. The two challenges can be simultaneously addressed by decreasing the value of C0, or equivalently, increasing the kt2 of the FBAR resonators. However, the kt2 of the FBAR is determined by the piezoelectric coefficient of AlN and has a theoretical upper bound of ~7% for FBARs. Therefore, in this paper, in order to address the two challenges, inductors are connected in parallel to the FBARs. The value of the parallel inductors is chosen such that they can resonate with C0 at the center frequencies of the FBARs, thus we will have a larger effective kt2 around the center frequency. Therefore, when the center frequencies are shifted, the admittance at resonance will reduce by a much smaller ratio. Fig. 4 shows the simulated results of frequency shift when an FBAR is connected in parallel to an inductor. As explained before, the value of Lp is chosen to be 𝐿(cid:3043) = . As expected, with the same values of series capacitance Cs, the modulation amplitudes are significantly increased, and the peak admittance reduction is much smaller. (cid:2869) (cid:3118)(cid:3004)(cid:3116) (cid:3104)(cid:3116) C. Circulator design The circulator circuit combines two single-ended (SE) branches connected in a differential configuration with a modulation phase difference of 1800 (Fig. 5a). This configuration was proven to have the ability to cancel intermodulation products [17], therefore improving IL and IX. This is particularly important in this implementation, given the close proximity of these mixing products to the signal frequency. For each SE branch, three FBAR resonators are connected in a wye configuration. In order to simplify the printed circuit board (PCB) implementation, the three parallel inductors (Lp) are connected in a delta configuration, instead of Submitted to IEEE Journal of Microelectromechanical Systems wye. Using wye-to-delta transformation, this is equivalent to connecting them in wye configuration with three times smaller inductance (Fig. 5b). Instead of modulating the center frequencies by using varactors, three capacitors (Cs) are connected in series to the three FBARs respectively, and RF switches are connected in parallel to series capacitors and are modulated by square waves. Therefore, each of the resonant tank is commutated between two center frequencies. By synchronizing the modulation for each of the SE branch to have an increase of phase of 1200 towards either clockwise (CW) or counterclockwise (CCW) direction, a preferred rotation direction is formed and therefore, degeneracy of the two counter rotating modes is lifted and reciprocity is broken. Thanks to the ultra-low modulating frequency enabled by the use of high Q FBARs, switches with high power handling can be used. Compared to the conventional modulation mechanism using varactors, the use of highly-linear RF switches guarantees high linearity and much simplified modulation network. Fig. 5 (a) Circuit schematic of the proposed magnet-free circulator. The circuit contains two single-ended branch, with a modulation phase difference of 1800. For each of the single-ended branch, three FBARs are connected in a wye configuration, and the three parallel inductors are connected in a delta configuration, for easier PCB implementation. The FBARs are connected in series to switched capacitors to modulate the center frequency. (b) The equivalence of the connection of parallel inductors in wye and delta configuration. A three times larger inductance is needed for the delta connection. The value of series capacitors is chosen by circuit simulation, shown in Fig. 6. The circuit is simulated using the harmonic balance simulator in Keysight ADS. The IL and BW (defined by the BW at 15 dB of IX) versus series capacitance are plotted. For each of the series capacitance, the modulation frequency is chosen such that the maximum IX is 25 dB. When the capacitance is too small, the admittance reduction at resonance causes a high IL. When the capacitance is too large, the modulation amplitude is not enough to provide significant IX and low IL at the same time. The optimal value of series capacitance lies between 200 fF to 300 fF. On the other hand, BW increases with smaller series capacitance and therefore larger modulation amplitude. Therefore, the value of series capacitance in this paper is chosen to be 200 fF, in order to achieve low IL and high BW at the same time. Using the optimal value of series capacitance, circuit simulation shows an IL of 3.6 dB, IX of 25 dB and 15 dB-BW of 28 MHz (Fig. 7). 4 It is worth mentioning that the quality factor Q of the parallel inductors does not degrade the performance of the circuit too much. The simulated relationship between the circulator IL and Q of Lp is plotted in Fig. 8. Compared to Q of 500, using inductors with Q of 125 (commercially available) will only increase the IL by 0.1 dB. Fig. 6 Simulated results of IL and BW with different series capacitance values. For each of the capacitance value, modulation frequency is set such that the IX is 25 dB. Fig. 7 Circuit simulation of S-parameters with the optimum series capacitance (200 fF). Simulated results show a IL of 3.9 dB, IX of 25 dB and 15 dB-IX BW of 28MHz. Fig. 8 The relationship between IL and Q of the parallel inductors. Compared to a Q of 500, using a Q of 125 (commercially available) will only degrade the IL by 0.1 dB. D. PCB design A PCB was designed and fabricated in order to implement the described circulator (Fig. 9a). FBARs used in this paper (Fig. 9b) (Broadcom engineering sample) are monolithically integrated, showing a center frequency of 2.5 GHz, kt2 of ~3% and Q of ~600 (Fig. 9c) and are wire-bonded to the PCB. Submitted to IEEE Journal of Microelectromechanical Systems Parallel inductors are Coilcraft 0603HP series with inductance of 11.92 nH and Q of 125 at 2.5 GHz. RF switches are MACOM MASWSS0179 single-pole, double-throw (SPDT) switches with IL of ~1 dB at 2.5 GHz. Series capacitors are Murata GJW series with capacitance of 200 fF. 5 phase synchronization of modulation signals. Furthermore, if port 1, 2, and 3 are connected to Rx, Tx and ANT, the IL of signal transmission from ANT to Rx and from Tx to ANT and the IX between Tx and Rx, which are the most important metrics in full-duplex operation, can be optimized by slightly tuning the modulation phase. Fig. 12 (a) reports the optimized S-parameters, showing an IL of 4.0 dB, IX of 30 dB, and 15 dB- IX BW of 34.7 MHz (~1.4% of RF frequency). Fig. 9 (a) Designed PCB for the circulator. (b) FBAR chip (Broadcom engineering sample) and schematic of connection. Three FBARs are monolithically integrated on one chip. (c) The MBVD fitting of the FBARs. Result shows a C0 of 975 fF, kt2 of 2.67% and Q of 645. Fig. 10 Measurement set up. Three function generators are synchronized together to provide the modulation signal. The S-parameters of the device are measured using a 4-port VNA. III. MEASUREMENTS A. S-parameters The measurement of the S-parameters is shown in Fig. 10. In order to test the PCB, three two-channel function generators are synchronized to provide the square wave modulation signals. The modulation frequency is set to be 40 MHz. The S- parameters are measured using a 4-port vector network analyzer (VNA). The measurement of S-parameters shows IL of 4.5, 4.7 and 5.3 dB; IX of 24.5, 21.1 and 20.3 dB and RL of 23.2, 18.7 and 13.2 dB, respectively (Fig. 11). The slight difference is attributed to the random values fluctuations of components, parameters mismatch between different FBARs and imperfect Fig. 11. Measured S-parameters of the circuit when the modulation phases are set to be exactly 1200 between each other. (a)Measured IL. (b) Measured IX. (c) Measured RL. The S-parameters without parallel inductors are also measured (Fig. 12b). Due to a much smaller modulation amplitude, a lower modulation frequency is used (7 MHz). As is expected, the BW is almost 7 times narrower, and the IL is also higher, due to the limited modulation amplitude and large admittance reduction factor determined by C0. Note that in this case a switch with a lower IL (0.5 dB at 2.5 GHz) but faster switching speed is used (MASWSS0166), due to a less rigorous requirement on the switching speed. B. Output spectrum The output spectrum is measured using a spectrum analyzer (Fig. 13). Single tone signal is excited from port 1, and the Submitted to IEEE Journal of Microelectromechanical Systems output spectrum is measured at port 2, while port 3 is terminated by 50 ohm impedance. The measured output spectrum shows more than 30 dBc of first-order intermodulation products suppression, thanks to the use of differential configuration. The reason of imperfect intermodulation suppression is attributed to slight asymmetry of PCB designs and components values fluctuation between differential paths, and small modulation phase mismatch. 6 linearity is among the highest for all magnet-free circulators demonstrated to date, thanks to the use of RF switches with high power handling, enabled by the ultra-low modulation frequency. Fig. 14. Measured Linearity (a) IIP3 (b) P1dB. IV. CONCLUSION the highest for all magnet-free this circuits. Among Table I summarizes the performance metrics reported in this work compared to other relevant works on magnet-free circulators based on LPTV these demonstrations, Refs. [13] and [25] are based on the spatiotemporal modulation of TLs, Refs. [17] and [18] are implemented through angular momentum biasing using LCs, while Refs. [19], [20], [22], [26] and this work are based on spatiotemporal modulation of MEMS devices. In this work, for the first time without sacrificing other performance metrics including IL, BW and linearity, ultra-low modulation frequency (1.6%) has been achieved, thanks to the use of high Q FBARs. Besides a much smaller power consumption, low modulation frequency also enables the use of RF switches with high power handling, therefore leading to a power handling performance (P1dB of 28 dBm and IIP3 of 40 dBm) that is among circulator demonstrations. The achieved IL (4.0 dB) is much lower than all the other MEMS based circulators, and is also among the lowest compared to all previous work at a relatively high center frequency (2.5 GHz), again due to the use of high Q FBARs centered at 2.5 GHz. Compared to previous demonstrations based on varactors [17], [18], the use of RF switches significantly simplifies the modulation network and improves the linearity. Furthermore, by using parallel inductors to increase the effective kt2 of the FBARs, a broad BW (1.4%) is achieved, overcoming the narrowband issue from previous MEMS demonstrations. In summary, this paper shows the first demonstration of a 2.5 GHz highly-linear and broadband FBAR-based magnet-free circulator that shows low IL, high IX and time. The demonstrated response shows the potential towards high- performance RF non-reciprocal component with extremely small form factor in modern communication systems to achieve full-duplex operation. low power consumption at that can be integrated the same Fig. 12. (a) Measured IL from TX to ANT and from ANT to RX and measured IX from TX to RX. The modulation phases are slightly tuned to optimize these three parameters. (b) Measured S-parameters of the circulator without parallel inductors. A 7 times narrower BW is observed. Fig. 13. Measured output spectrum C. Linearity and power handling The linearity is evaluated by measuring the 1 dB compression point (P1dB) and input-referred third-order harmonic intersect point (IIP3). The circuit shows excellent linearity. Measured P1dB is 28 dBm and IIP3 is 40 dBm (Fig. 14). The measured ACKNOWLEDGMENTS The authors thank Dr. Rich Ruby from Broadcom Limited for providing FBARs engineering samples. COMPARISON TO OTHER LPTV CIRCULATORS TABLE I Submitted to IEEE Journal of Microelectromechanical Systems 7 Technology Center freq. Mod. Freq. a 25 GHz TL TL LC LC [13] [25] [18] [17] [22] [20] [19] [26] DC-3GHz c 1000 MHz 1000 MHz 155 MHz 2500 MHz 146 MHz 1165 MHz This work 2500 MHz a-b Defined by the ratio with center frequency. b Defined by the IX value. c Results are broadband measured from DC to 3 GHz. d-e Assuming center frequency is 1.5 GHz. f Baluns are de-embedded MEMS MEMS MEMS MEMS MEMS 33% 83% d 19% 10% 0.6% 0.1% 0.1% 0.1% 1.6% BW b 18.4% 93.3% e 2.4% 2.3% 5.8% 0.02% 0.2% 0.3% 1.4% IX 18.3 dB 20 dB 20 dB 20 dB 20 dB 20 dB 15 dB 15 dB 15 dB IL P1dB 3.3/3.2 dB 21.5/21 dBm 4.3 dB 3.3 dB 0.8 dB f 6.6 dB 11 dB 8 dB 12 dB 4.0 dB N/A 29 dBm 29 dBm N/A N/A -8 dBm N/A 28 dBm IIP3 N/A N/A 34 dBm 32 dBm 30 dBm N/A N/A N/A 40 dBm REFERENCES [1] J. Choi I, M. Jain, K. Srinivasan, P. Levis, and S. Katti, "Achieving single channel, full duplex wireless communication," In Proc. 16th Annu. Int. Conf. Mobile Computing and Networking, 2010, pp. 1-12. [2] M. Jain, J. Choi I, T. Kim, D. Bharadia, S. Seth, K. Srinivasan, P. Levis, S. Katti and P. Sinha, "Practical, real-time, full duplex wireless," In Proc. 17th Annu. Int. Conf. Mobile Computing and Networking, 2011, pp. 301- 312. [3] D. Bharadia, E. McMilin and S. Katti, "Full duplex radios," ACM SIGCOMM Comput. Commun. Rev., vol. 43, no. 4, pp. 375-386, Sept. 2013. [4] D. W. Bliss, P. A. Parker and A. R. Margetts, "Simultaneous transmission and reception for improved wireless network performance," In Proc. IEEE/SP 14th Workshop Stat. Signal Process., pp. 478-482, 2007. [5] M. Duarte and A. Sabharwal, "Full-duplex wireless communications using off-the-shelf radios: Feasibility and first results," in Proc. Conf. Rec. 44th ASILOMAR Conf. Signals, Syst. Comput., pp. 1558-1562, 2010. [6] M. Duarte, C. Dick and A. Sabharwal, "Experiment-driven characterization of full-duplex wireless systems," IEEE Trans. Wireless Commu., Vol. 11, no.12, 4296-4307, 2012. [7] D. M. Pozar, "Microwave engineering", John Wiley & Sons, 2009. [8] C. E. Fay and R. L. Comstock, "Operation of the ferrite junction circulator," IEEE Trans. Microw. Theory and Techn., vol.13, no. 1, pp. 15-27, 1965. [9] H. Bosma, "On stripline Y-circulation at UHF," IEEE trans. Microw. Theory and Techn., vol. 12, pp. 61-72, 1964. [10] S. Tanaka, N. Shimomura and K. Ohtake, "Active circulators: The realization of circulators using transistors," In Proc. IEEE, vol. 53, no.3, pp. 260-267, 1965. [11] Y. Ayasli, "Field effect transistor circulators," IEEE Trans. On Magnetics, vol. 25, no. 5, pp. 3243-3247, 1989. [12] S. A. Ayati, D. Mandal, B. Bakkaloglu and S. Kiaei, "Adaptive integrated CMOS circulator", Radio Frequency Integrated Circuits Symposium, pp. 146-149, 2016. [13] T. Dinc, M. Tymchenko, A. Nagulu, D. Sounas, A. Alu and H. Krishnaswamy, "Synchronized conductivity modulation to realize broadband lossless magnetic-free non-reciprocity," Nat. Commun., vol. 8, no. 1, 795, 2017. [14] S. Qin, Q. M. Xu and Y. E. Wang, "Nonreciprocal components with distributedly modulated capacitors," IEEE trans. Microw. Theory and Techn., vol. 62, no.10, pp. 2260-2272, 2014. [15] M. M. Biedka, R. Zhu, Q. M. Xu and Y. E. Wang, "Ultra-wide band non- reciprocity through sequentially-switched delay lines," Scientific reports, vol. 7, 40014, 2017. [16] N. A. Estep, D. L. Sounas, J. Soric and A. Alù, "Magnetic-free non- reciprocity and isolation based on parametrically modulated coupled- resonator loops," Nature Physics, vol.10, no.12, 923, 2014. [17] A. Kord, D. L. Sounas and A. Alù, "Pseudo-Linear Time-Invariant Magnetless Circulators Based on Differential Spatiotemporal Modulation of Resonant Junctions," IEEE trans. Microw. Theory and Techn., vol. 66, no. 6, pp. 2731-2745, Jun. 2018. [18] A. Kord, D. L. Sounas and A. Alù, "Magnet-less circulators based on spatiotemporal modulation of bandstop filters in a delta topology," IEEE trans. Microw. Theory and Techn., vol. 66, no. 2, 911-926, 2018. [19] Y. Yu, G. Michetti, A. Kord, D. Sounas, F. V. Pop, P. Kulik, M. Pirro, Z. Qian, A. Alu and M. Rinaldi, "Magnetic-free radio frequency circulator based on spatiotemporal commutation of MEMS resonators," In Proc. IEEE Micro Electro Mechanical Systems (MEMS), pp. 154-157, 2018. [20] M. M. Torunbalci, T. J. Odelberg, S. Sridaran, R. C. Ruby and S. A. Bhave, "An FBAR Circulator," IEEE Microwave and Wireless Components Letters, vol. 28, no. 5, 395-397, 2018. [21] C. Xu, E. Calayir and G. Piazza, "Magnetic-free electrical circulator based on AlN MEMS filters and CMOS RF switches," In Proc. IEEE Micro Electro Mechanical Systems (MEMS), pp. 755-758, 2018. [22] R. Lu, T. Manzaneque, Y. Yang, L. Gao, A. Gao, and S. Gong, "A Radio Frequency Non-reciprocal Network Based on Switched Acoustic Delay Lines," IEEE trans. Microw. Theory and Techn., to be published. [23] A. Kord, D. L. Sounas, Z. Xiao, and A. Alu, "Broadband Cyclic- Symmetric Magnet-less Circulators and Theoretical Bounds on their Bandwidth," IEEE trans. Microw. Theory and Techn., vol. 66, no. 12, pp.5472-5481. [24] A. Nagulu, Andrea Alù, and Harish Krishnaswamy, "Fully-Integrated Non-Magnetic 180nm SOI Circulator with> 1W P1dB,>+ 50dBm IIP3 and High Isolation Across 1.85 VSWR," In 2018 IEEE Radio Frequency Integrated Circuits Symposium (RFIC), pp. 104-107, 2018. [25] M. Biedka, Q. Wu, X. Zou, S. Qin, and Y. E. Wang, "Integrated time- varying electromagnetic devices for ultra-wide band nonreciprocity," In Proc. 2018 IEEE Radio and Wireless Symp. (RWS), pp. 80-83. IEEE, 2018. [26] C. Xu, and G. Piazza, "Magnet-Less Circulator Using AlN MEMS Filters and CMOS RF Switches," J. Microelectromech. Syst., to be published. [27] R. Lu, J. Krol, L. Gao, and S. Gong. "A Frequency Independent Framework for Synthesis of Programmable Non-reciprocal Networks," Scientific reports, vol. 8, no. 1, 14655, 2018. [28] A. Kord, D. L. Sounas, and A. Alù, "Achieving Full-Duplex Communication: Magnetless Parametric Circulators for Full‐Duplex Communication Systems," IEEE Microw. Mag., vol. 19, no. 1, pp. 84-90, Jan. 2018. [29] M. Pirro, C. Cassella, G. Michetti, G. Chen, P. Kulik, Y. Yu, and M. Rinaldi. "Novel topology for a non-reciprocal MEMS filter," In Proc. 2018 IEEE Int. Ultrasonics Symp. (IUS), pp. 1-3. IEEE, 2018. [30] G. Michetti, C. Cassella, F. Pop, M. Pirro, A. Kord, D. Sounas, A. Alú, and M. Rinaldi. "A quasi-LTI frequency-selective SAW circulato," In Proc. 2018 IEEE Int. Ultrasonics Symp. (IUS), pp. 206-212. IEEE, 2018. [31] M. Ghatge, G. Walters, T. Nishida, and R. Tabrizian. "A Non-Reciprocal Filter using Asymmetrically Transduced Micro-Acoustic Resonators," IEEE Electron Device Lett., to be published. [32] N. Reiskarimian, A. Nagulu, T. Dinc, and H. Krishnaswamy, "Nonreciprocal Electronic Devices: A Hypothesis Turned Into Reality," IEEE Microwave Magazine, vol. 20, no. 4, 94-111, 2019. [33] Y. Yu and M. Rinaldi, "Wi-Fi-band Acoustic Resonant Circulator, " In Proc. IEEE Micro Electro Mechanical Systems (MEMS), 2019. [34] P. Lulik, Y. Yu, G. Chen and M. Rinaldi, "Magnetic-free Isolator Based on a Single Differential Lattice Filter, " In Proc. IEEE Micro Electro Mechanical Systems (MEMS), 2019. [35] Y. Yu, F. V. Pop, G.Michetti, P. Kulik, M. Pirro, A. Kord, D. Sounas, A. Alu and M. Rinaldi, "2.5 GHz Highly-Linear Magnetic-Free Submitted to IEEE Journal of Microelectromechanical Systems Microelectromechanical Resonant Circulator," In Proc. IEEE Int. Freq. Cont. Symp, 2018. [36] D. Sounas, and A. Alù, "Non-Reciprocal Photonics Based on Time Modulation," Nature Photonics, Vol. 11, No. 12, pp. 774-783, November 30, 2017. 8 [37] Y. Yu, and M. Rinaldi, "Chip-scale micro-acoustic radio frequency gyrator," In. Proc. IEEE Int. Conf. Solid-State Sensors, Actuators and Microsystems (Transducers), 2019.
1810.06393	1	1810	2018-10-10T20:59:02	400%/W second harmonic conversion efficiency in $\mathrm{14 \mu m}$-diameter gallium phosphide-on-oxide resonators	[ "physics.app-ph", "physics.optics" ]	Second harmonic conversion from 1550~nm to 775~nm with an efficiency of 400% W$^{-1}$ is demonstrated in a gallium phosphide (GaP) on oxide integrated photonic platform. The platform consists of doubly-resonant, phase-matched ring resonators with quality factors $Q \sim 10^4$, low mode volumes $V \sim 30 (\lambda/n)^3$, and high nonlinear mode overlaps. Measurements and simulations indicate that conversion efficiencies can be increased by a factor of 20 by improving the waveguide-cavity coupling to achieve critical coupling in current devices.	physics.app-ph	physics	400%/W second harmonic conversion efficiency in 14 µm-diameter gallium phosphide-on-oxide resonators ALAN D. LOGAN,1,* MICHAEL GOULD,2, EMMA R. SCHMIDGALL,2 KARINE HESTROFFER,3 ZIN LIN,4 WEILIANG JIN,5 ARKA MAJUMDAR,1,2 FARIBA HATAMI,3 ALEJANDRO W. RODRIGUEZ,5 AND KAI-MEI C. FU1,2 1Department of Electrical and Computing Engineering, University of Washington, Seattle WA 98195 2 Department of Physics, University of Washington, Seattle WA 98195 3Department of Physics, Humboldt-Universitat zu Berlin, 12489 Berlin, Germany 4 John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA 02138 5 Department of Electrical Engineering, Princeton University, Princeton, NJ 08544, USA *[email protected] Abstract: Second harmonic conversion from 1550 nm to 775 nm with an efficiency of 400% W−1 is demonstrated in a gallium phosphide (GaP) on oxide integrated photonic platform. The platform consists of doubly-resonant, phase-matched ring resonators with quality factors Q ∼ 104, low mode volumes V ∼ 30(λ/n)3, and high nonlinear mode overlaps. Measurements and simulations indicate that conversion efficiencies can be increased by a factor of 20 by improving the waveguide-cavity coupling to achieve critical coupling in current devices. Introduction 1. Parametric nonlinear processes such as frequency conversion play a critical role in several established and emerging technological applications [1,2], including ultrashort-pulse shaping [3,4], generation of light in strategically important spectral windows [5 -- 7], spectroscopy [8, 9], and quantum information [10 -- 14]. In particular, efficient frequency conversion of single photons from optically accessible qubits is expected to enable long-range fiber transmission [15] and increase compatibility between dissimilar quantum memories [16]. Two important practical considerations dictating the performance of quantum networks are scalability and efficiency. Similar considerations arise in the context of frequency conversion, with typical devices exploiting resonators designed to exhibit high field strengths and long-lived modes to increase efficiency and reduce power requirements [17,18]. These conditions have been commonly met in macroscopic optical cavities (such as large-etalon resonators [19 -- 22]) and waveguides [23,24] that sacrifice spatial confinement for increased ability to engineer modes at the desired wavelengths and satisfy the requisite phase matching conditions [2]. However, there has been increased interest in exploring designs that exploit the small footprint, low mode volume, and wider bandwidth offered by integrated photonic devices, which could drastically improve scalability in size, cost, and power consumption [25 -- 33]. These include whispering gallery mode resonators [21,25], singly- resonant photonic crystal cavities [34], and nanoplasmonic and dielectric metasurfaces [35 -- 37]. More recently, integrated ring resonators in aluminum nitride (AlN) on sapphire demonstrated second harmonic generation (SHG) efficiencies as high as 17,000% W−1 [38], far exceeding the performance of prior devices. In this work, we study SHG in a gallium phosphide (GaP) on oxide platform that asphalts the path for achieving high-efficiency frequency conversion in robust, compact, and wide bandwidth integrated cavities. In particular, GaP-on-oxide ring resonators were designed for quasi-phase- matched SHG from the telecommunication C band (1550 nm) to the near-infrared (775 nm). This work represents at least an order-of-magnitude improvement over prior SHG GaP devices [27,39] while also extending device functionality and applicability via coupling to on-chip waveguides. Waveguide-to-waveguide SHG efficiencies of up to 400% W−1 are observed in 14 µm diameter rings. The conversion efficiency is primarily limited by overcoupling to both input and output waveguides. In addition to its significant χ(2) second-order nonlinear susceptibility (∼100 pm/V [40]), gallium phosphide offers a unique combination of properties useful for frequency conversion applications. GaP has a high index of refraction (n = 3.31 at 637 nm [41]) compared to AlN (n = 2.2 [42]) or most traditional nonlinear materials like LiNbO3 (n = 2.28 [43]), allowing fabrication of ultra-low mode volume resonators on a wide variety of substrates, including diamond (n = 2.4) [39,44 -- 46]. With a wide bandgap EG of 2.32 eV, GaP maintains transparency into the visible spectrum, permitting efficient frequency conversion of a wider range of wavelengths than other high-index materials such as GaAs (Eg=1.42 eV, χ(2) ∼ 220 pm/V [47]). Notably, the emission wavelength of the diamond nitrogen-vacancy center, a solid-state qubit candidate, falls within the transparency window of gallium phosphide [46]. 2. Model and Design At low input powers, the effect of down conversion on the efficiency and power requirements for achieving SHG in a doubly resonant cavity is negligible [18]. In this undepleted regime, the conversion efficiency corresponding to a cavity supporting modes at angular frequencies ω1,2 and coupled to waveguides, is given by [18,48]: χ(2)2 3 0λ 1 (Qc2 + Qi2)2 , (Qc1 + Qi1)4 χ(2)2 3 0λ 1 ¯β2 2 ¯β2 2 ω1 Q2 i1Q2 c1 Q2 i1Qi2 P2,out P2 1,in = Q4 1Q2 2 Q2 c1Qc2 = ω1 Qi2Qc2 (1) Here, Qk, Qik, and Qck denote the loaded, intrinsic, and coupling quality factors of mode k = {1, 2}, where mode 1 is the fundamental and mode 2 is the second harmonic. The coefficient ¯β is the dimensionless nonlinear overlap of the fundamental and second harmonic modes, 3 λ 1, (2) Þ ¯β = NL (cid:16)Þ 1E12d r i(cid:44)j(cid:44)k(E1iE∗ 2j E1k + E1iE1j E∗ 2k (cid:17)(cid:113)Þ 2E22d r (cid:113) )d r a generalization of the familiar phase matching figure of merit [2]. For zincblende crystals such as GaP, χ(2) is nonzero only for mutually perpendicular field components. [2]. Note that the integral in the numerator is only evaluated over the extent of the nonlinear medium. Notably, achieving high conversion efficiencies requires large χ(2), high intrinsic quality factors, critical coupling at both the fundamental and second-harmonic modes, and high nonlinear overlap ¯β. Previous work with GaP photonics on diamond [49] has indicated that the intrinsic quality factor is determined by process-dependent sidewall roughness, so the device design process was focused on achieving high mode overlap and coupling rather than increasing Q. A ring resonator topology was chosen to allow control over mode properties and independent coupling to on-chip waveguides with minimal design variables. To ensure sufficient confinement at 1550 nm, a 427 nm thick photonics layer of (100) GaP on a thermal SiO2 substrate was used. The ring structure imposes rotational symmetry on resonator modes. However, in (100) GaP, the effective nonlinear susceptibility changes sign every 90◦, so quasi-phase-matching is required to avoid back-conversion. Also, because only mutually perpendicular field components contribute to ¯β in GaP, the fundamental mode must be transverse-electric (TE) and the second harmonic mode must be transverse-magnetic (TM). To allow straightforward device optimization under these constraints, Eq. 2 was translated to cylindrical coordinates: Þ 2π (cid:113) (Þ 1E12 r drdz)(cid:113)Þ 2E22 r drdz β+ei(2m1−m2+2)θ + β 3 λ 1 1 0 ¯β = ± = β −ei(2m1−m2−2)θ dθ Þ (cid:0)2[E1r E1z(E∗ 2θ)](cid:1) r drdz N L 2r (3) + E∗ 2θ) + E1θ(E1r E∗ 2z + E1zE∗ 2r)] ± i[(E2 1r − E2 1θ)E∗ 2z + E1z(E1r E∗ 2r − E1θ E∗ where mk is the azimuthal mode number of mode k. In this coordinate system, it is readily aparent that ¯β vanishes if the quasi-phase-matching condition 2m1 = m2 ± 2 is not satisfied. Due 1r E∗ to the relatively small θ field component, both β+ and β− depend primarily on E2 , with the 2z main contributions to β+(−) weighted toward the outside (inside) edges of the ring. The resonator design process was based on two-dimensional eigenmode simulations of a ring cross-section. First, the fundamental TE00 at λ1 = 1550 nm was simulated for a ring of radius on the order of ten microns. Higher-order TM modes with similar effective index were simulated for λ2 = 775 nm, and the mode with the highest β+/− was selected. The ring width and radius was then modified via a gradient algorithm to minimize the phase-mismatch parameter (2m1−m2−2) for the selected modes. Through this process, high-efficiency resonator design candidates could be found with minimal computational resources and without bias toward higher-order harmonic modes. A final ring design of width w = 840 nm and radius r = 7.14 µm (measured from the center of the waveguide) was found, corresponding to an antisymmetric TM03 second harmonic mode shown in Fig. 2 (inset), which yields 2m1 − m2 = +2 and ¯β = 1.43 × 10−4. To allow testing of large device arrays on a single chip, the ring resonator was evanescently coupled to waveguides terminating in nearby grating couplers, as shown in Fig. 1a. Critical coupling of both modes is vital for maximizing conversion efficiency, so independent wraparound coupling regions were designed for each ring mode via supermode analysis. Waveguide widths, ring-to-waveguide separations, and coupling region lengths were co-optimized to theoretically provide coupling quality factors of Q1c ≈ 2 × 105 for the 1550 nm mode and Q2c ≈ 2 × 104 for the 775 nm mode. 3. Fabrication and Testing A 427 nm-thick GaP layer on a 300 nm Al0.8Ga0.2P sacrificial layer was grown by molecular beam epitaxy on a GaP substrate. A (2.5 mm)2 area GaP membrane was released from the substrate and transferred to a 10 µm thermal SiO2-on-Si substrate. Before transfer, the oxide surface was cleaned and treated with hexamethyldisilazane vapor. The GaP membrane was released from the sacrificial AlGaP layer in 3:100 HF:H2O and transferred to DI water. The membrane was then captured on a water droplet on the oxide substrate. A drying step at 80 ◦C completed the membrane transfer. The described transfer process was used due to its compatibility with transfer to mm-scale diamond chips for quantum information applications using the process described in Refs. [50 -- 52]. Recently, wafer-scale GaP membrane transfer to silicon oxide has been realized by other groups via direct wafer bonding followed by substrate removal [45,53]. In our devices, the resulting GaP-on-oxide chip was patterned with electron beam lithography, using ∼100-nm-thick HSQ as a resist. A final Cl/Ar/N2 (1.0/6.0/3.0 sccm) reactive ion etch step was used to transfer the mask into the GaP-on-oxide substrate. [49]. A device schematic and SEM image are shown in Fig. 1a. Two grating-coupled input/output waveguides, one for the fundamental and one for the second harmonic, are used to couple to the device. A 50-device array was fabricated varying the ring waveguide width from 839 nm to 847 nm to ensure quasi-phase matching and doubly resonant enhancement can be attained even in the presence of fabrication tolerances. The measured ring waveguide widths of the two devices (SHG01, SHG02) exhibiting doubly resonantly enhanced SHG are listed in Table 1. Fig. 1. (a) On-chip layout of the nonlinear ring resonator (yellow) coupled to two independent input/output waveguides for 775 nm (blue) and 1550 nm (pink) light. The proximity of the grating couplers allow any combination of inputs and outputs to be focused or collected simultaneously by a single microscope objective. Inset: SEM image of a fabricated GaP SHG device. (b) Free-space measurement setup for the device. Cross-polarization and the pinhole (PH) are used to eliminate reflected input light. PD: photodiode, Obj: objective, BS: beamsplitter, DC: dichroic mirror, HWP: half-wave plate Table 1. SHG device characteristics. w is the resonantor waveguide width. T is the transmission on-resonance. Uncertainty in w denotes the range of measured values. Q and T are determined by a Lorentzian fit with uncertainty representing the 95% confidence interval. Device SHG01 SHG02 w Q1 Q2 T1 T2 847 ± 23 nm 26, 500 ± 1500 847 ± 17 nm 40, 700 ± 10700 13, 600 ± 5400 16, 800 ± 3200 0.44 ± 0.10 0.81 ± 0.05 0.50 ± 0.05 0.52 ± 0.02 The devices were tested using the setup shown in Fig. 1(b). A scanning 1550 nm laser (Santec TSL-510) was used to excite the fundamental mode. Fig. 2(a) shows the telecom transmission measurement curve for SHG01. The Lorentzian fit to the the resonance dip corresponds to Q1 = 2.65 × 104, with Q1 denoting the total quality factor, 1/Q1 = 1/Qc1 + 1/Qi1. The transmission coefficient on resonance is T1 = 0.44. Transmission spectra of a broadband 775 nm source (supercontinuum laser or LED), shown in Fig. 2(b), were used to determine the second-harmonic mode quality factor Q2 and transmission T2. Quality factors and transmission coefficients for both modes and both devices are given in Table 1. Additionally, a cross-sectional mode profile for both modes are included as insets in Fig. 2a,b. Single wavelength transmission measurements at 1550 nm (775 nm) were used to measure the grating coupler efficiency as well as bulk transmission through the microscope set-up at each Fig. 2. The transmission dip from telecom TE00 (a,c) and near-infrared TM03 (b,d) resonances in devices SHG01 (a,b) and SHG02 (c,d), along with fitted Lorentzian curves. A cross-section of the mode profile is inset. The telecom resonance was measured on an infrared power meter with a tunable laser input, and the near-infrared resonance was measured using a supercontinuum laser and spectrometer. wavelength. A description of the grating coupler and microscope loss characteristics can be found in Appendix A. The telecom (SHG) grating couplers were designed for 33% (24%) efficiency at 1550 nm (775 nm). The measured efficiencies at the experimental resonances were 22% at 1523 nm (3.2% at 761 nm) in SHG01 and 19% at 1549 nm (7.5% at 774 nm) in SHG02. All measured values are derived by assuming identical efficiencies for the input and output gratings. For SHG measurements, the telecom input grating is used to excite the fundamental mode. Input power is continually monitored by a reference photodiode. Simultaneously, the SHG signal is collected from one of the SHG grating ports (both are tested). Any detuning between the fundamental and SHG excitation wavelengths and their respective resonant modes reduces conversion efficiency. To realize mutual resonance, the device is tested on a temperature- controlled stage. Heating the device causes both resonances to redshift at different rates, allowing relative tuning. SHG conversion efficiency is measured as a function of input wavelength at multiple temperatures to find the maximum conversion efficiency. Fig. 3. (a) SHG conversion efficiency of device SHG01 as a function of both temperature and input wavelength. Conversion efficiency profiles at 27 (green), 29.5 (black), and 32 ◦C (red) are inset. (b) Conversion efficiency of SHG02, with profiles at 26 (green), 29 (black), 32 (orange), and 36 ◦C (red). Asymmetry from thermal bistability is visible in the conversion efficiency profiles of both devices and becomes more pronounced with stage temperature and input laser power. Due to resonance splittings, SHG02 exhibits additional asymmetry as well as efficiency peaks at multiple temperatures. 4. Results and Discussion Quasi-phase matched resonances were found in device SHG01 at a fundamental wavelength of 1523.1 nm and in SHG02 at 1549.1 nm. The quasi-phase matching condition was identified by the strong and highly temperature-dependent second harmonic conversion of light at the fundamental resonance, as shown in Fig. 3. In devices in which only single-resonance enhancement is observed, both the conversion efficiency and the effect of temperature are far weaker. A maximum waveguide-to-waveguide conversion efficiency η of 175% W−1 was observed in SHG01. η includes SHG signals propagating in both directions of the SHG waveguide. The maximum conversion efficiency as a function of temperature followed a Lorentzian profile with a full-width- at-half-maximum of ∼ 4 ◦C. (Fig.4a), with the peak efficiency wavelength redshifting linearly to follow the fundamental resonance (Fig. 4a). The assymmetrical shape of the SHG efficiency curves (inset Fig. 3a) is attributed to a redshift of the resonance as it is heated by the laser. Device SHG02 exhibited both a higher maximum efficiency of 400% W−1 as well as a more complex dependence on the temperature and fundamental wavelength (Fig. 3b). Due to a splitting of both the fundamental and second harmonic resonances [54], SHG02 exhibited efficiency peaks at multiple temperatures instead of the single peak seen in SHG01. The double-humped structure in the peak efficiency curve (black curve, inset of Fig. 3b) is attributed to these split resonances. Fig. 4(c) shows the SHG efficiency, measured from only one SHG grating, as a function of the fundamental input power on double resonance for SHG02. As expected, the SHG power increases quadratically with the fundamental power, with no sign of depletion at waveguide input powers of up to 3 mW. The peak waveguide-to-waveguide conversion efficiency is 0.84% in device SHG02. Increasing the waveguide input power beyond 3 mW resulted in thermal optical bistability [55], causing a discrete hop of the resonance when tuning the fundamental on resonance. This relatively low absolute conversion efficiency points to the practical need to increase the per input power efficiency for quantum conversion applications as well as a limit on the total SHG power that may be produced for classical applications. Fig. 4. (a) Maximum SHG efficiency as a function of temperature for both devices. The red squares are the corresponding fundamental wavelength as a function of temperature. (b) Square root of the SHG output power as a function of fundamental input power, showing the expected linear dependence. Both input and output powers are calculated in-waveguide powers. To compare the experimental device performance to the simulations (Eq. 1), it is necessary to know the coupling Qc and intrinsic Qi of both modes. Based on transmission measurements, the fundamental and second harmonic resonances of both devices are not critically coupled, a condition that is met when Qc = Qi, and T = 0. The measured finite T gives us only the magnitude of the difference between Qc and Qi, T = Qi − Qc2/(Qi + Qc)2. Rewriting Eq. 1 in terms of the T1 and T2, we obtain (cid:18) 1 ±√ (cid:19)2 1 ±√ , (4) ηtheory = P2,out P2 1,in = χ(2)2 3 0λ 1 ¯β2 2 ω1 Q2 1Q2 T1 2 T2 2 SHG01 theory,oc = 230% W−1 and η in which the + signs correspond to the case where both modes are overcoupled, Qc > Qi, and the − signs to the case where both modes are undercoupled Qc < Qi. Using the measured Q's and T's in Table 1 and the calculated ¯β from Sec. 2, we are only able to obtain reasonable agreement between experiment and theory in the case where both modes are overcoupled. The theory,oc = 425% W−1 and corresponds SHG efficiency in this case is η to the highest theoretical efficiency. For SHG02, ¯β (Eq. 2) is calculated for a standing wave instead of a traveling wave due to the observed resonance splitting. To further investigate the coupling regime, the coupling regions of both devices were imaged and measured by scanning electron microscopy and simulated by finite-difference-time-domain and supermode analysis (Appendix B). Within the measurement uncertainty, the coupling quality factor for the 775 nm mode could be as low as 1.2× 103 and as high as 1× 106, so overcoupling of this mode is plausible. The lowest reasonably attainable coupling Q1c for the 1550 nm mode was 1.5 × 105, significantly higher than the measured loaded quality factors. This analysis indicates SHG02 the fundamental mode is most likely undercoupled, in which case theoretical calculations predict much lower conversion efficiencies than observed. Due to this uncertainty in the coupling factors, we are unable to reconcile the experimentally measured efficiencies with these simulations at this time. However, theoretical calculations do not include the effect of surface roughness or sidewall angles so other coupling mechanisms may be a factor. 5. Conclusion and Outlook In summary, we observe near 400%/W SHG conversion efficiency in waveguide-integrated GaP resonators. The high conversion efficiency is achieved with resonant enhancement of both the fundamental and second harmonic modes, meeting the quasi-phase matching requirement, and achieving high mode-overlap in small mode-volume structures. These experimental results indicate two areas in which device performance can be immediately and significantly improved. First, the coupling should be decreased to achieve critical coupling. Assuming intrinsic quality factors Qi of 105 (consistent with our current measurements), the expected SHG efficiency exceeds 8000%/W. Reducing the diameter of these simple rings to 5 µm should increase ¯β2 by a factor of 3. Under such a regime of operation, pump depletion would occur at modest powers << 1 mW, avoiding heating-induced optical bi-stability. Critical coupling to such small rings may prove challenging, but recent theoretical results utilizing inverse design methods have demonstrated single waveguide couplers capable of achieving critical coupling at multiple frequencies [56]. Finally, the ability to achieve higher nonlinear coupling factors ¯β in ring resonators is largely hampered by their diminishing capacity to confine light with decreasing sizes. While photonic crystals and associated structures can overcome such a tradeoff [17], they typically can only do so over narrow bandwidths. Recently proposed inverse design strategies [37, 57] point a way toward new kinds of multi-mode cavities capable of confining light at disparate wavelengths in ultra-small volumes, and exhibiting orders of magnitude larger ¯β factors, the subject of ongoing experimental efforts. A. Measurement Calibration All reported SHG conversion efficiencies are on-chip efficiencies, i.e. based on input and output power in the on-chip waveguides. On-chip powers were derived from off-chip efficiency measurements. Grating coupler efficiency was measured by comparing the transmission spectrum from each coupling photonic circuit to a reflection spectrum from the thermal oxide substrate. In Fig. 1a, the telecom coupling circuit is pink and the SHG/near-infrared coupling circuit is blue. Near-infrared spectra were measured using a supercontinuum laser for excitation and detected by a grating spectrometer with a CCD detector. Telecom spectra were measured using a scanning laser for excitation and a power meter for detection. Excitation polarization was adjusted using a λ/2 plate directly before the objective to excite TE (1550 nm) or TM (775 nm) modes. Reflected excitation light was filtered from the collection path using both a cross-polarizer and a spatial filter (a pinhole) to select the output grating coupler. The efficiency of a single grating was calculated from the power transmission spectrum assuming identical input/output grating couplers and negligible on-chip losses. This assumption of identical couplers is reasonable for SHG01 in which the SHG efficiency is similar when exciting either telecom grating coupler. There is a temperature-dependent discrepancy for SHG02 which is attributed to the splitting of the resonances due to backscattering [54]. Near the resonances of interest, the efficiency of the telecom grating coupler for device SHG01 (SHG02) was measured as 22% (19%). Efficiency of the near-infrared grating was 3.2% (7.5%). Transmission measurements through all off-chip optical components were conducted (depicted in Fig. 1b.) The total telecom power delivered to grating coupler input is 0.31 × PRefPD in which PRefPD is the power measured on the reference photodiode. Including the grating coupling efficiency, the total telecom power inside the waveguide is 0.068 × PRefPD (0.059 × PRefPD) for SHG01 (SHG02). For SHG experiments, the SHG power is measured at the visible photodiode (VisPD). In these measurements there is no pinhole in the collection path. Transmisson measurements through the bulk optics for both H and V polarized light (corresponding to the two grating orientations) were measured. Including the grating efficiency, for SHG01 we find the SHG power inside the waveguide corresponds to 114 × PVisPD (71 × PVisPD) for H (V) polarized light, while for SHG2 the power inside the waveguide is 48 × PVisPD (30 × PVisPD) for H (V) polarized light. B. Coupling region simulations Simulations of these devices are based on as-fabricated device dimensions measured using scanning electron microscopy (SEM). Measured features are illustrated in Fig. 2. We measured both the top and bottom dimensions for each feature. Fig. 2 shows the top dimensions (green brackets) and bottom dimensions (white brackets). Averages of the two measurements are given in Fig. 1. Assuming a trapezoidal line profile, the measured sidewall angle is ∼ 85 degrees. Due to this angled sidewall, the top dimensions of the ring and waveguide are smaller than the bottom dimensions and the top dimension of the gap is larger than its bottom dimension. For our device simulations, we used the mean width. Fig. 1. (Left) Device dimensions in the coupling region, shown for the telecom coupling region of device SHG01. White (green) brackets indicate the bottom (top) of the feature.(Right) The measured average dimensions (nm) for each feature for each coupling region for the two devices. The coupling efficiency of each wrapped waveguide region was simulated by supermode analysis. As shown in Fig. 2a, if the azimuthal mode number of the curved waveguide mode w(cid:105) is similar to the ring resonator mode r(cid:105), the two modes combine to form in-phase (+(cid:105)) and out-of-phase (−(cid:105)) supermodes when the structures are brought together in the coupling region. Light from the two original modes couples into the supermodes when the waveguide approaches the ring and then couples back into the original modes when the waveguide diverges. The relative phase of the supermodes determines how light is distributed between the original modes. Because the supermodes propagate with distinct azimuthal mode numbers m+ and m−, the relative phase changes along the coupling region, allowing energy to be transferred from ring to waveguide or vice versa. The field coupling strength κ of a single pass through the coupling region is (5) where θ is the angular length of the waveguide wrap. In our devices, the telecom (SHG) coupling κ = (cid:104)r+(cid:105) eim+θ (cid:104)+w(cid:105) + (cid:104)r−(cid:105) eim−θ (cid:104)−w(cid:105) , Fig. 2. (a) Separate mode profile cross-sections (λ = 775 nm) for the ring resonator (r(cid:105)) and waveguide (w(cid:105)) that compose the coupling region. When the two structures are combined, these modes split into two supermodes +(cid:105) and −(cid:105). (b) Coupling quality factors (logarithmic scale) for IR and NIR modes of a 860 nm wide ring, with gap width (top axis) decreasing as waveguide width (bottom axis) increases. Within measurement uncertainty, wider ring resonators reach slightly lower minimum coupling Q with narrower gaps. waveguide wraps around 51◦ (44◦) of the ring. The resulting coupling quality factor of the ring is Qc = 2Rng 4π 2 λ0κ , (6) where R is the ring radius, λ0 is the free-space wavelength of the resonance, and ng is the group index of the ring resonator mode. Coupling strengths and quality factors were simulated over the range of measured device dimensions summarized in Fig. 1. As shown in Fig. 2b, coupling to the second-harmonic mode is very sensitive to variations in device dimensions. Within the range of device measurements, Qc can vary from 1.2× 103 to 1× 106, indicating that this mode may be either overcoupled or undercoupled in each device. The telecom coupling region is less sensitive. Most configurations within the range of measurement variations yielded Qc between 1.5 × 105 and 5 × 105. Since this entire range is significantly larger than the measured loaded quality factors, the telecom mode is likely undercoupled in both devices. Finite-difference time-domain simulations of selected devices gave similar results for the plausible range of quality factors. Acknowledgements This material is based on work supported by the National Science Foundation under award no. 1640986 and DMR-1454836. We thank Taylor Fryett with assistance in optical testing, and N. Shane Patrick for advice on electron beam lithography. Part of this work was conducted at the Washington Nanofabrication Facility, a National Nanotechnology Coordinated Infrastructure (NNCI) site at the University of Washington, which is supported in part by funds from the National Science Foundation (awards NNCI-1542101, 1337840 and 0335765), the National Institutes of Health, the Molecular Engineering & Sciences Institute, the Clean Energy Institute, the Washington Research Foundation, the M. J. Murdock Charitable Trust, Altatech, ClassOne Technology, GCE Market, Google and SPTS. References 1. M. M. Fejer, "Nonlinear optical frequency conversion," Phys. Today 47, 25 (1994). 2. R. Boyd, Nonlinear Optics (Academic Press, 2008). 3. K. W. DeLong, R. Trebino, J. Hunter, and W. E. White, "Frequency-resolved optical gating with the use of second-harmonic generation," JOSA-B 11, 2206 -- 2215 (1994). 4. M. A. Arbore, A. Galvanauskas, D. Harter, M. H. Chou, and M. M. Fejer, "Engineerable compression of ultrashort pulses by use of second-harmonic generation in chirped-period-poled lithium niobate," Opt. Lett. 22, 1341 -- 1343 (1997). 5. P. S. Kuo, K. L. Vodopyanov, M. M. Fejer, D. M. Simanovskii, X. Yu, J. S. Harris, D. Bliss, and D.Weyburne, "Optical parametric generation of a mid-infrared continuum in orientation-patterned GaAs," Opt. Lett. 31, 71 -- 73 (2006). 6. K. L. Vodopyanov, M. M. Fejer, X. Yu, J. S. Harris, Y.-S. Lee, W. C. Hurlbut, V. G. Kozlov, D. Bliss, and C. Lynch, "Terahertz-wave generation in quasi-phase-matched GaAs," Appl. Phys. Lett. 89, 141119 (2006). 7. R. Krischek, W. Wieczorek, A. Ozawa, N. Kiesel, P. Michelberger, T. Udem, and H. Weinfurter, "Ultraviolet enhancement cavity for ultrafast nonlinear optics and high-rate multiphoton entanglement experiments," Nat. Photonics 4, 170 -- 173 (2010). 8. T. F. Heinz, C. K. Chen, D. Ricard, and Y. R. Shen, "Spectroscopy of molecular monolayers by resonant second- harmonic generation," Phys. Rev. Lett. 48, 478 (1982). 9. A. Ozawa, J. Rauschenberger, C. Gohle, M. Herrmann, D. R. Walker, V. Pervak, A. Fernandez, R. Graf, A. Apolonski, R. Holzwarth, F. Krausz, T. W. Hänsch, and T. Udem, "High harmonic frequency combs for high resolution spectroscopy," Phys. Rev. Lett. 100, 253901 (2008). 10. P. G. Kwiat, K. Mattle, H. Weinfurter, A. Zeilinger, A. V. Sergienko, and Y. Shih, "New high-intensity source of polarization-entangled photon pairs," Phys. Rev. Lett. 75, 4337 -- 4341 (1995). 11. A. Vaziri, G. Weihs, and A. Zeilinger, "Experimental two-photon, three-dimensional entanglement for quantum communication," Phys. Rev. Lett. 89, 240401 (2002). 12. S. Tanzilli, W. Tittel, M. Halder, O. Alibart, P. Baldi, N. Gisin, and H. Zbinden, "A photonic quantum information interface," Nature 437, 116 -- 120 (2005). 13. S. Zaske, A. Lenhard, C. A. Kessler, J. Kettler, C. Hepp, C. Arend, R. Albrecht, W.-M. Schulz, M. Jetter, P. Michler, and C. Becher, "Visible-to-telecom quantum frequency conversion of light from a single quantum emitter," Phys. Rev. Lett. 109, 147404 (2012). 14. D. E. Chang, V. Vuletić, and M. D. Lukin, "Quantum nonlinear optics -- photon by photon," Nat. Photonics 8, 685 15. J. S. Pelc, L. Yu, K. D. Greve, P. L. McMahon, C. M. Natarajan, V. Esfandyarpour, S. Maier, C. Schneider, M. Kamp, S. Höfling, R. H. Hadfield, A. Forchel, Y. Yamamoto, and M. M. Fejer, "Downconversion quantum interface for a single quantum dot spin and 1550-nm single-photon channel," Opt. Express 20, 27510 -- 27519 (2012). 16. K. De Greve, L. Yu, P. L. McMahon, J. S. Pelc, C. M. Natarajan, N. Y. Kim, E. Abe, S. Maier, C. Schneider, M. Kamp, S. Höfling, R. H. Hadfield, A. Forchel, M. M. Fejer, and Y. Yamamoto, "Quantum-dot spin -- photon entanglement via frequency downconversion to telecom wavelength," Nature 491, 421 EP -- (2012). nonlinear optics in photonic-crystal microcavities," Opt. Express 15, 16161 -- 16176 (2007). 17. J. Bravo-Abad, A. Rodriguez, P. Bermel, J. D. Johnson, Steven G. a nd Joannopoulos, and M. Soljacic, "Enhanced 18. A. Rodriguez, M. Soljačić, J. D. Joannopoulos, and S. e. G. Johnson, "χ(2) and χ(3) harmonic generation at a critical 19. J. U. Furst, D. V. Strekalov, D. Elser, M. Lassen, U. L. Andersen, C. Marquadt, and G. Leuchs, "Naturally phase-matched second-harmonic generation in a whispering-gallery-mode resonator," Phys. Rev. Lett. 104, 153901 (2010). power in inhomogeneous doubly resonant cavities," Opt. Express 15, 7303 -- 7318 (2007). 20. C. Wang, M. J. Burek, Z. Lin, H. A. Atikian, V. Venkataraman, I.-C. Huang, P. Stark, and M. Lončar, "Integrated high quality factor lithium niobate microdisk resonators," Opt. Express 22, 30924 -- 30933 (2014). 21. G. Lin, A. Coillet, and Y. K. Chembo, "Nonlinear photonics with high-Q whispering-gallery-mode resonators," Adv. EP -- (2014). Opt. Photonics 9, 828 -- 890 (2017). Opt. Express 25, 29927 -- 29933 (2017). 22. R. Wolf, I. Breunig, H. Zappe, and K. Buse, "Cascaded second-order optical nonlinearities in on-chip micro rings," 23. C. Wang, Z. Li, M.-H. Kim, X. Xiong, X.-F. Ren, G.-C. Guo, N. Yu, and M. Lončar, "Metasurface-assisted phase-matching-free second harmonic generation in lithium niobate waveguides," Nat. Commun. 8, 2098 (2017). 24. J.-Y. Chen, Y. M. Sua, H. Fan, and Y.-P. Huang, "Modal phase matched lithium niobate nanocircuits for integrated nonlinear photonics," OSA Continuum 1, 229 -- 242 (2018). 25. P. S. Kuo, J. Bravo-Abad, and G. S. Solomon, "Second-harmonic generation using -quasi-phasematching in a GaAs whispering-gallery-mode microcavity," Nat. Commun. 5, 3109 EP -- (2014). 26. S. Buckley, M. Radulaski, J. Petykiewicz, K. G. Lagoudakis, J.-H. Kang, M. Brongersma, K. Biermann, and J. Vućković, "Second-harmonic generation in gaas photonic crystal cavities in (111)b and (001) crystal orientations," ACS Photonics 1, 516 -- 523 (2014). 27. D. P. Lake, M. Mitchell, H. Jayakumar, L. F. dos Santos, D. Curic, and P. E. Barclay, "Efficient telecom to visible wavelength conversion in doubly resonant gallium phosphide microdisks," Appl. Phys. Lett. 108, 031109 (2016). 28. I. Roland, M. Gromovyi, Y. Zeng, M. El Kurdi, S. Sauvage, C. Brimont, T. Guillet, B. Gayral, F. Semond, J. Y. Duboz, M. de Micheli, X. Checoury, and P. Boucaud, "Phase-matched second harmonic generation with on-chip GaN-on-Si microdisks," Sci. Reports 6, 34191 EP -- (2016). 29. 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). 30. S. M. Hendrickson, A. C. Foster, R. M. Camacho, and B. D. Clader, "Integrated nonlinear photonics: emerging applications and ongoing challenges," J. Opt. Soc. Am. B 31, 3193 -- 3203 (2014). 31. M. S. Mohamed, A. Simbula, J.-F. Carlin, M. Minkov, D. Gerace, V. Savona, N. Grandjean, M. Galli, and R. Houdré, "Efficient continuous-wave nonlinear frequency conversion in high-Q gallium nitride photonic crystal cavities on silicon," APL Photonics 2, 031301 (2017). 32. A. Rao and S. Fathpour, "Second-harmonic generation in integrated photonics on silicon," physica status solidi (a) 215, 1700684 (2017). 33. G. Moille, S. Combrié, L. Morgenroth, G. Lehoucq, S. Sauvage, M. E. Kurdi, P. Boucaud, A. de Rossi, and X. Checoury, "Nonlinearities in gaas cavities with high cw input powers enabled by photo-oxidation quenching through ald encapsulation," Opt. Express 26, 6400 -- 6406 (2018). 34. Y. Li, C. Wang, and M. Loncar, "Design of nano-groove photonic crystal cavities in lithium niobate," Opt. Lett. 40, 2902 -- 2905 (2015). 35. J. Butet, P.-F. Brevet, and O. J. Martin, "Optical second harmonic generation in plasmonic nanostructures: from fundamental principles to advanced applications," ACS Nano 9, 10545 -- 10562 (2015). 36. T. K. Fryett, A. Zhan, and A. Majumdar, "Phase-matched nonlinear optics via patterning layered materials," Opt. K5 -- K8. Lett. 42, 3586 -- 3589 (2017). 37. C. Sitawarin, W. Jin, Z. Lin, and A. W. Rodriguez, "Inverse-designed photonic fibers and metasurfaces for nonlinear frequency conversion," Photonics Res. 6, B82 -- B89 (2018). 38. A. W. Bruch, X. Liu, X. Guo, J. B. Surya, Z. Gong, L. Zhang, J. Wang, J. Yan, and H. X. Tang, "17,000%/w second-harmonic conversion efficiency in single-crystalline aluminum nitride microresonators," Appl. Phys. Lett. 113, 131102 (2018). 39. K. Rivoire, S. Buckley, F. Hatami, and J. Vućković, "Second harmonic generation in gap photonic crystal waveguides," Appl. Phys. Lett. 98, 263113 (2011). 40. A. D. Corso, F. Mauri, and A. Rubio, "Density-functional theory of the nonlinear optical susceptibility: Application to cubic semiconductors," Phys. Rev. B 53, 15638 -- 15642 (1996). 41. W. L. Bond, "Measurement of the refractive indices of several crystals," J. Appl. Phys. 36, 1674 -- 1677 (1965). 42. J. Pastrnák and L. Roskovcová, "Refraction index measurements on AlN single crystals," physica status solidi (b) 14, 43. D. E. Zelmon, D. L. Small, and D. Jundt, "Infrared corrected Sellmeier coefficients for congruently grown lithium niobate and 5 mol.% magnesium oxide -- doped lithium niobate," J. Opt. Soc. Am. B 14, 3319 -- 3322 (1997). 44. P. E. Barclay, K.-M. Fu, C. Santori, and R. G. Beausoleil, "Hybrid photonic crystal cavity and waveguide for coupling to diamond nv-centers," Opt. Express 17, 9588 -- 9601 (2009). 45. K. Schneider, P. Welter, Y. Baumgartner, H. Hahn, L. Czornomaz, and P. Seidler, "Gallium phosphide-on-silicon dioxide photonic devices," J. Light. Technol. 36, 2994 -- 3002 (2018). 46. D. Englund, B. Shields, K. Rivoire, F. Hatami, J. Vućković, H. Park, and M. D. Lukin, "Deterministic coupling of a single nitrogen vacancy center to a photonic crystal cavity," Nano Lett. 10, 3922 -- 3926 (2010). 47. I. Shoji, T. Kondo, A. Kitamoto, M. Shirane, and R. Ito, "Absolute scale of second-order nonlinear-optical coefficients," J. Opt. Soc. Am. B 14, 2268 -- 2294 (1997). 48. Z.-F. Bi, A. W. Rodriguez, H. Hashemi, D. Duchesne, M. Loncar, K.-M. Wang, and S. G. Johnson, "High-efficiency second-harmonic generation in doubly-resonant χ (2) microring resonators," Opt. express 20, 7526 -- 7543 (2012). 49. M. Gould, S. Chakravarthi, I. R. Christen, N. Thomas, S. Dadgostar, Y. Song, M. L. Lee, F. Hatami, and K.-M. C. Fu, "Large-scale gap-on-diamond integrated photonics platform for nv center-based quantum information," J. Opt. Soc. Am. B 33, B35 -- B42 (2016). 50. N. Thomas, R. J. Barbour, Y. Song, M. L. Lee, and K.-M. C. Fu, "Waveguide-integrated single-crystalline gap resonantors on diamond," Opt. Express 22, 13555 (2014). 51. M. Gould, E. R. Schmidgall, S. Dadgostar, F. Hatami, and K.-M. C. Fu, "Efficient extraction of zero-phonon-line photons from single nitrogen-vacancy centers in an integrated gap-on-diamond platform," Phys. Rev. Appl. 6, 011001 (2016). 52. E. R. Schmidgall, S. Chakravarthi, M. Gould, I. R. Christen, K. Hestroffer, F. Hatami, and K.-M. C. Fu, "Frequency control of single quantum emitters in integrated photonic circuits," Nano Lett. 18, 1175 -- 1179 (2018). 53. D. J. Wilson, K. Schneider, S. Hoenl, M. Anderson, T. J. Kippenberg, and P. Seidler, "Integrated gallium phosphide 54. B. E. Little, J.-P. Laine, and S. T. Chu, "Surface-roughness-induced contradirectional coupling in ring and disk nonlinear photonics," ArXiv e-prints (2018). resonators," Opt. Lett. 22, 4 -- 6 (1997). 55. V. R. Almeida and M. Lipson, "Optical bistability on a silicon chip," Opt. Lett. 29, 2387 -- 2389 (2004). couplers," Opt. Express 26, 26713 -- 26721 (2018). 57. Z. Lin, X. Liang, M. Lončar, S. G. Johnson, and A. W. Rodriguez, "Cavity-enhanced second-harmonic generation via 56. W. Jin, S. Molesky, Z. Lin, K.-M. C. Fu, and A. W. Rodriguez, "Inverse design of compact multimode cavity nonlinear-overlap optimization," Optica 3, 233 -- 238 (2016).
1802.04661	2	1802	2018-05-15T02:30:46	Modelling of cavity optomechanical magnetometers	[ "physics.app-ph", "physics.optics" ]	Cavity optomechanical magnetic field sensors, constructed by coupling a magnetostrictive material to a micro-toroidal optical cavity, act as ultra-sensitive room temperature magnetometers with tens of micrometre size and broad bandwidth, combined with a simple operating scheme. Here, we develop a general recipe for predicting the field sensitivity of these devices. Several geometries are analysed, with a highest predicted sensitivity of 180~p$\textrm{T}/\sqrt{\textrm{Hz}}$ at 28~$\mu$m resolution limited by thermal noise in good agreement with previous experimental observations. Furthermore, by adjusting the composition of the magnetostrictive material and its annealing process, a sensitivity as good as 20~p$\textrm{T}/\sqrt{\textrm{Hz}}$ may be possible at the same resolution. This method paves a way for future design of magnetostrictive material based optomechanical magnetometers, possibly allowing both scalar and vectorial magnetometers.	physics.app-ph	physics	Modelling of cavity optomechanical magnetometers Yimin Yu,1 Stefan Forstner,1 Halina Rubinsztein-Dunlop,1 and Warwick P. Bowen1, ∗ 1ARC Centre for Engineered Quantum Systems, School of Mathematics and Physics, The University of Queensland, Brisbane, Queensland 4072, Australia (Dated: May 2, 2019) Cavity optomechanical magnetic field sensors, constructed by coupling a magnetostrictive material to a micro-toroidal optical cavity, act as ultra-sensitive room temperature magnetometers with tens of micrometre size and broad bandwidth, combined with a simple operating scheme. Here, we √ develop a general recipe for predicting the field sensitivity of these devices. Several geometries are analysed, with a highest predicted sensitivity of 180 pT/ Hz at 28 µm resolution limited by thermal noise in good agreement with previous experimental observations. Furthermore, by adjusting the composition of the magnetostrictive material and its annealing process, a sensitivity as good as 20 pT/ Hz may be possible at the same resolution. This method paves a way for future design of magnetostrictive material based optomechanical magnetometers, possibly allowing both scalar and vectorial magnetometers. √ I. INTRODUCTION Magnetometers with high spatial resolution are re- quired for many applications such as magnetoencephalog- raphy [1], measurements of topological spin configura- tions [2] and nuclear magnetic resonance spectroscopy to identify chemical composition, molecular structure and dynamics [3]. Optical readout of magnetometers can of- fer high sensitivity for a given resolution, while being well decoupled from the magnetic signal. Among optical magnetometers, an ensemble of nitrogen-vacancy (NV) centres with a volume size of 8.5×105 µm3 pushes the sensitivity down to 1 pT/ Hz [4]. However, NV mag- netometry generally requires high optical power for exci- tation (e.g., 400 mW in Ref. [4]), as well as complicated microwave decoupling sequences in NMR spectroscopy, and is limited by the sample fabrication reproducibility [5]. A magnetometer based on micro-sized Bose -- Einstein √ condensates has a volume of 90 µm3, but its quantum- enhanced sensitivity is limited to 1.86 nT/ Hz [6]. It is crucial yet challenging to reduce the size of magnetometers while maintaining competitive sensitivities. √ Among various types of magnetometers, optomechan- √ ical magnetometers [7, 8] reach sensitivities in the high pT/ Hz range at room temperature with sizes of tens of micrometres, comparable to the best cryogenic SQUID- magnetometer of the same size [9]. The principle of an optomechanical magnetometer is illustrated in Figure 1a. A magnetostrictive material converts the magnetic field to a force as a result of mechanical deformation. The magnetostrictive response has a nonlinear component, a property that has been utilised in previous work to mix low frequency magnetic fields up to megahertz frequen- cies and therefore evade low frequency noise [8]. How- ever, in general, it is far smaller than the linear com- ponent, so that the force may be well approximated by ∗[email protected] Ffield = cactBsig, where cact (N/T) is the actuation pa- rameter and Bsig (T) is the magnetic field to be mea- sured. The amplitude of the mechanical response to this force is greatly enhanced when the magnetostrictive material is driven resonantly at its mechanical eigenfre- quency by a modulated magnetic field. The mechanical response changes the path length of the optical cavity to which the magnetostrictive material is attached, allow- ing the magnetic field to be read out optically from the shift of the optical resonance [10]. While significant suc- cesses have been achieved in experimental demonstrations of optomechanical magnetometers [7, 8], modelling and sensitivity-prediction for these devices have been some- what ad hoc [11, 12]. Better modelling techniques are needed to both enhance understanding of previous exper- imental results and for design of future magnetometers. In this work, we present a model of magnetostrictive mag- netometers that accounts for arbitrary mechanical mode shape and device geometry. We modify the elastic wave equation, which describes the small-amplitude motion of elastic materials, by including magnetostrictive stress. This modified elastic wave equation is then numerically solved by finite element analysis (using COMSOL Multiphysics). Magnetomechanical overlap, describing the overlap between the magnetostrictive deformation induced by the signal mag- netic field and the excited mechanical eigenmode, is intrin- sically included in the matrix form of the modified elastic wave equation, with each matrix element containing direc- tional information. Mechanical properties are extracted from the solution to the modified elastic wave equation from COMSOL to be further combined with optomechani- cal analysis [10] to predict the sensitivity of a magnetometer for a given geometry. We apply this analysis to study the effect of the posi- tion of the magnetostrictive material on the sensitivity of devices similar to those reported in Ref. [7]. Using the piezomagnetic constant measured from a rod of the magnetostrictive material Terfenol-D [13], we model a magnetometer design, where the Terfenol-D is deposited directly on top of a standard silica toroid. From there, we employ single mode analysis (Appendix A) and dis- 8 1 0 2 y a M 5 1 ] h p - p p a . s c i s y h p [ 2 v 1 6 6 4 0 . 2 0 8 1 : v i X r a 2 II. CONCEPT OF OPTOMECHANICAL MAGNETOMETRY Optomechanical magnetometry can be schematically ex- plained via the example of a Fabry -- P´erot optical resonator coupled to a spring-mass mechanical oscillator as depicted in Figure 1a. An applied magnetic field Bsig causes a deformation to a magnetostrictive material attached to the mechanical oscillator (see Appendix B for details of how this field is generated in COMSOL). This induces a Ffield = cactBsig on one movable end mirror of the optical resonator, changing the optical path length and thus the optical resonance frequency. The shift in the optical reso- nance frequency is therefore proportional to the applied magnetic field. The transduction from magnetic field to mechanical motion is determined by the actuation param- eter cact depending on magnetomechanical overlap and magnetostrictive coefficient. The magnetic field signal en- coded on the motion of the mechanical element is read out by optically probing the the optical resonance frequency. This can be achieved with high precision by coupling a coherent optical field into the cavity, collecting the output field, and measuring the change in its amplitude or phase due to the modulation of the optical resonance frequency. For instance, directly detecting the output field, as in several reported experiments [7, 8], measures changes to the amplitude of the output optical field and enables simple operation. Alternatively, a homodyne scheme can be used, allowing an arbitrary quadrature of the optical field to be accessed as shown in Figure 1c. Here, the output field is interfered with a bright local oscillator field prior to detection. The transduction from mechanical displacement to optical signal can be quantified by the effective cooperativity Ceff [10]. The magnetic field sensitivity is limited by noise consist- ing primarily of thermal force and shot noise on the optical field. Thermal noise is explained by the equipartition the- orem, which states that each mechanical degree of freedom of an object has a mean energy of kBT /2 (kB is the Boltz- mann constant and T is the temperature). This energy excites incoherent mechanical vibration near mechanical eigenfrequencies. The bandwidth of the magnetometer depends on the visibility of the thermal noise over the optical shot noise. For the case of a single mechanical resonance, the sensitivity is flat over the frequency range where thermal noise dominates shot noise, and degrades outside of this region. Consequently, in this case, the bandwidth is given simply by the thermal-noise-dominant frequency band, which is typically on the order of a few megahertz [15]. The case of multiple mechanical modes is more complex due to variations in actuation constants, effective cooperativities and mechanical parameters, and due to interferences in the coherent response of the me- chanical modes. In this paper, as a test geometry for our model, we choose optomechanical magnetometers of the form re- ported in Refs. [7, 8]. They utilise a silica microtoroid as the optical resonator. The magnetostrictive material is FIG. 1: Concept of an optomechanical magnetometer. (a) illustration via a Fabry -- P´erot type optical resonator. The cou- pling of magnetostrictive material to an optical cavity is quan- tified by the effective cooperativity Ceff . The magnetostrictive material converts a magnetic field to a force Ffield = cactBsig with Bsig being an oscillating magnetic field. Thermal force and optical shot noise act as noise terms. κ (rad·s−1), Γ (rad·s−1), ω0 (rad·s−1), and ΩM (rad·s−1) are optical and mechanical decay rate, optical and mechanical resonance fre- quency, respectively; (b) sketch of a magnetometer with micro- toroidal structure coupled to a tapered optical fibre; (c) homo- dyne detection scheme. The signal arm couples a coherent light source in and out from a magnetometer via a tapered optical fibre through an evanescent optical field, and is mixed with a strong reference beam (local oscillator field) by a 3 dB coupler. The magnetometer is embedded in the signal magnetic field. cover that a bimetallic-stripe-like bending effect, similar to the bimetallic bending effect in a cantilever [14], greatly enhances the sensitivity when the magnetostrictive ma- terial is positioned off-centre. Optimisation of this effect may allow substantial improvements in sensitivity in fu- ture devices. Furthermore, we investigate the sensitivity achievable from a device comprised of a toroidal structure with a centre hole that is filled with the Terfenol-D, as √ studied experimentally in Ref. [8] and sketched in Fig- ure 1b . We predict a peak sensitivity of 180 pT/ Hz over a broad spectrum by using multi-mode analysis un- der optimised operational conditions, in good agreement with current experimental observations. This numerical model allows specification of the orien- tation of a sample to maximally enhance the magnetome- chanical overlap, thus amplifying the detected magnetic field signal, as well as characterization of the magne- tomechanical overlap in response to the variation of the magnetic field direction. This is crucial to vectorial mag- netometers that measure not only the intensity but also the direction of the magnetic field. thermalF =cactBfieldΩM Γ(a)Bsig(b)sigκoptical ω0shot noise3 dBlocal oscillatorhomodyne detectionsignal magnetic (cid:31)eld(c)forceCeffembedded in or deposited onto the microtoroid as sketched in Figures 1b and 2a, respectively. Combined, the silica microtoroid, the magnetostrictive material and the silicon pedestal serve as the mechanical oscillator. Using a ta- pered optical fibre placed next to the toroid, the optical field can be coupled in and out of the microtoroid through an evanescent optical field. This optomechanical mag- netometry platform offers a simple operational scheme and low energy consumption with state-of-the-art field sensitivity for a micro-magnetometer. FIG. 2: (a) sketch of the position offset of the magnetostric- tive material of the first experimentally realized optomechani- cal magnetometers [7]; (b) a second order crown mode without (left) and with 4 µm (right) Terfenol-D position offset. Arrows show the positions with maximum displacement; (c) strain of the magnetometer with centred (left) Terfenol-D, and with 4 µm offset (right). Note that the colourmaps of the strain have different scales; (d) Ceff and sensitivity as a function of the position of the Terfenol-D. III. NUMERICAL METHODS The primary objective of this work is to develop a ver- satile technique to numerically obtain a meaningful esti- mation of the magnetic field sensitivity for a wide range of sensor geometries. We consider the case of phase quadra- ture detection in a homodyne scheme and on-resonance optical probing of the cavity resonance, which maximises the signal-to-noise. We note, however, that simpler direct detection with off-resonance probing and an optimal de- tuning of , where κ is the optical cavity linewidth, only degrades the sensitivity by a factor of √ κ 3 33/2 ∼ 1.5. The sensitivity as a function of magnetic field frequency Ω can be determined from the finite-time sensor power spectrum S(Ω), which can be separated into a stochastic 8 2 3 noise term Snoise(Ω) and coherent signal term Ssignal(Ω) as S(Ω) = τ−1(cid:104)i∗(Ω)i(Ω)(cid:105) = Ssignal(Ω) + Snoise(Ω), (1) where i is the photocurrent, normalised so that the optical shot noise contribution to Snoise(Ω) is equal to 1/2 [10], and τ is the measurement time. At frequencies Ω (cid:29) 2π/τ and considering j mechanical modes, Snoise(Ω) is given by [10] Snoise(Ω) = (cid:88) + Ceff,j(Ω)(cid:105) 1 2 + j , 8ηΓ2 jCeff,j(Ω)χj(Ω)2(cid:104) kBT ΩM,j (2) where the first term is the optical shot noise and the second term constitutes the combination of mechanical thermal noise and quantum back-action noise. The detection effi- ciency η, consisting of the loss in the fibre-device coupling and detection process, is ideally taken to be 1 in the model. However, in the non-back-action dominated regime relevant here, reductions in efficiency can be exactly modelled by a proportionate decrease in the optomechanical cooperativity. Γj is the mechanical decay rate of mode j and Ceff,j is its effective cooperativity, which depends on the input laser power used, the decay rate of the optical field and mechan- ical excitation, and the radiation pressure coupling rate between them. The mechanical susceptibility of mode j is defined as χj(Ω) ≡ ΩM,j/(−Ω2 − iΩΓj + Ω2 M,j), with ΩM,j its mechanical resonance frequency. kBT/ΩM,j is the number of phonons thermally excited at room tem- perature, with being the reduced Planck constant. The mechanical motion induced by an alternating-current (AC) magnetic field is quantified by the finite-time power spec- trum Ssignal(Ω). This is calculated by replacing the thermal environment forcing Fth in the input momentum fluctuation Γ [10], which leads to Equation (2), with a coherent sinusoidal driving force Ffield(t) = cactB(t)sig at frequency Ω, and neglecting the incoherent noise terms (laser shot noise in amplitude and phase quadrature). This results in the expression Pin = xzpfFth/√ (cid:113)Ceff,j(Ω)χj(Ω) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:88) (cid:112)4meff,jΩM,jΓj cact,jBsignal,rms Γ j · (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 (3) , Ssignal(Bsig, Ω) =16τ πη where Bsignal,rms is the root-mean-square amplitude of Bsig(t), meff,j is the effective mass of mode j, and cact,j is the actuation constant associated with that mode. This finite-time power spectrum takes into account mechanical interference, as experimentally observed, for example in optoelectromechanical systems coherently driven by an electric field [16]. (b)(a)(d)(c)(cid:18)(cid:15)(cid:22)(cid:14)(cid:18)(cid:15)(cid:22)(cid:17)(cid:128)(cid:18)(cid:17)(cid:14)(cid:23)(cid:18)(cid:15)(cid:22)(cid:14)(cid:18)(cid:15)(cid:22)(cid:17)(cid:128)(cid:18)(cid:17)(cid:14)(cid:21)(cid:17)(cid:19)(cid:21)(cid:23)µ(cid:78)(cid:10)(cid:18)(cid:17)(cid:14)(cid:21)(cid:18)(cid:17)(cid:14)(cid:19)(cid:18)(cid:17)(cid:17)(cid:18)(cid:17)(cid:14)(cid:24)(cid:18)(cid:17)(cid:14)(cid:23)(cid:18)(cid:17)(cid:14)(cid:22)The frequency dependent signal-to-noise ratio (SNR) of the magnetic field measurement is given simply by SNR = Ssignal(Bsig, Ω) Snoise . (4) The minimum detectable field in the measurement time τ is defined as the field that produces a signal-to-noise ratio SNR of one, i.e., Bmin,τ = Bsig(SNR = 1). It should be noted that the stochastic noise power spec- tral density S(Ω)noise of Equation (2) is independent of integration time, whereas the integral of a coherent band- limited signal power spectrum, as described by Ssignal(Ω) in Equation (3), increases linearly with time. Conse- quently, Bmin,τ improves with measurement time as τ−1/2. To obtain a minimum detectable field in the conventional units of Tesla per root Hertz, independent of time, we multiply through by τ 1/2 with the result Bmin(Ω) = Bmin,τ (Ω) × τ 1/2 = Bsig (cid:115) S(Ω) × τ Ssignal(Bsig, Ω) . (5) To determine the minimum detectable field via finite element simulations, we use COMSOL Multiphysics. Sim- ulations detailed in the appendices allow us to extract each of the parameters in Equations (2) and (3) and therefore predict the sensitivity. These simulations involve both mechanical eigenmode solving to determine the resonance frequency, effective mass and effective cooperativity of each mechanical mode of a given device geometry; and magnetic field driving to determine the coherent response of the mechanical modes to a magnetic field and the inter- ferences between them. The approach is briefly sketched in what follows. placement maxr[Ψ(r)] as meff,j =(cid:82) The spatio-temporal mechanical modeshape is described by a separable function u(r, t) = Ψ(r)x(t). The effective mass meff,j for one mechanical resonance at an eigenfre- quency ΩM,j is calculated from the maximum physical dis- V ρnΨ(r)2dV [17], with normalization maxr,t[u(r, t)] = maxt[x(t)] and there- fore maxr[Ψ(r)]2 = 1. ρn is the density of the material and the subscript n denotes different parts of the device (for instance, silica for the optical resonator and Terfenol-D for the transducing medium). Note that, while this definition of effective mass is the convention for microelectromechani- cal systems, an alternative definition -- where the effective mass is defined with respect to the optical path length -- is commonly used in the optomechanical community [18]. This choice of convention has no effect on the ultimate predictions of our model. The magnetic field response Ssignal(Ω) of the sensor is determined by the eigenmode-dependent actuation pa- rameter cact. For a single mechanical eigenmode, the equation of motion is x(t) + Γ x(t) + Ω2 Mx(t) = cactBsig(t) meff . (6) 4 At the resonance frequency of each mechanical eigen- mode, cact can be extracted as a fitting parameter in the mechanical signal frequency response spectrum ob- tained from COMSOL. Taking the Fourier transform of Equation (6), we see that cact = x(Ω)ΩM meff χ(Ω)Bsig(Ω) . (7) This allows cact to be determined for each mechanical mode. Due to the magnetostrictive energy stored within com- pressed magnetostrictive materials, the extraction of meff and displacement from COMSOL for such materials re- quires modification of the elastic wave equation. To treat the magnetostrictive material in COMSOL, we built upon a previously used method [19 -- 21], including the magnetic field in a driving stress σdriv and adding a damping stress σΓ to the elastic stress σela, which describes the mechan- ical properties without driving force in the elastic wave equation [22], resulting in − ρnΩ2u = (cid:79) · (σela + σdriv + σΓ). (8) The modulated driving stress is linked to the magnetic field via the piezomagnetic constant [20], and a low value for the damping stress σΓ is chosen manually to avoid an artefactual infinity in the mechanical displacement at resonance (see Appendix C for technical details). Simula- tions reveal that the influence of a particular value chosen for σΓ on numerical results is negligible (Appendix D). To obtain the value of effective cooperativity, we quan- tify the effectiveness of transduction of mechanical motion to measurable optical path length change as the geomet- rical factor, as ξ ≡ δL max [u] , (9) where δL is the change of the optical path length due to the mechanical displacement. The extraction of the value of ξ from COMSOL is detailed in Appendix E. Within one mechanical mode, ξ is directly linked to the effective cooperativity (Appendix F) by Ceff (Ω) = ξ2 meff ΩMΓ L2(cid:0)κ2 + 4Ω2(cid:1) · 8ηescNinω2 0 (10) where κ is the optical decay rate, ω0 is the optical reso- nance frequency, Nin (photons·s−1) is the input optical photon number flux, L is the optical path length, and ηesc is the escape efficiency counting fibre-device coupling. The front part of the right hand side of Equation (10) is arranged to be mechanical mode dependent. The calcula- tion of the magnetic field sensitivity from Equations (2), (3) and (5) can then be obtained based on the value of the geometrical factor ξ. IV. SINGLE MECHANICAL MODE OPTOMECHANICAL ANALYSIS A. Bending Effect To verify the numerical model, we apply it to the first experimentally realized optomechanical magnetometer [7]. For simplicity, we begin the analysis considering only a single mechanical eigenmode (Appendix A). The magnetometer as sketched in Figure 2a consists of a silica micro-toroidal cavity with major radius of 33 µm. The Terfenol-D is glued on top of the silica and is modelled as a semi-sphere with a transverse radius of 18.5 µm and a height of 15 µm. The optical quality factor Qo = ωo/κ is taken to be 2×107 from the experiment. The mechanical quality factor QM = ΩM/Γ is assumed to be 200 for all modes which is a simplification, but is roughly in line with the experimentally observed quality factors. A continuous input laser is locked to the optical cavity resonance in the homodyne detection scheme, and the input laser power ensures that on mechanical resonances thermal noise dominates over optical shot noise. From available optical microscopic images, it is not clear whether the Terfenol-D is centred on the toroid or not. Therefore, we sweep the position of the Terfenol-D from the centre. Without loss of generality, we analyse the magnetic response for a second order crown mode because this mode has been commonly observed in exper- iments [15, 16, 23]. For the magnetometer with centred Terfenol-D, the effective motional mass is meff = 3.9 pg with eigenfrequency at 10.1 MHz. As the Terfenol-D is moved away from the centre as illustrated in Figure 2a, the mechanical eigenmode changes (Figure 2b). Gener- ally, the top of the Terfenol-D stretches more than the bottom part attached to a silica disk during a mechanical oscillation. This is also the case for silica where the top layer experiences the force from the Terfenol-D and the bottom layer is clamped to the silicon pedestal. There- fore, a bimetallic-like strain gradient is formed vertically. In the second order crown mode, as the major motion takes place at the silica layer instead of the Terfenol-D, the strain gradient can be viewed inside the silica disk at the edge of bottom Terfenol-D and top facet of silicon. With the centred Terfenol-D, the strain at the top layer of the silica is nearly two orders of magnitude smaller than that with 4 µm Terfenol-D position offset as shown in the red areas in Figure 2c. This local maximum strain leads to the maximum displacement of the device (pointed by the arrows on tori in Figure 2b) in the radial direction. Figure 2d shows the best sensitivity of 78 nT/ Hz, when driven by an in-plane magnetic field, takes place when the Terfenol-D offset is at 4 µm, nearly two orders of magnitude better than that of Terfenol-D centred (the same order of magnitude difference as that of the strain). We therefore see that the position of the Terfenol-D on the silica layer has strong influence on the bimetallic-like strain effect, and consequently the sensitivity. Moreover, the effective cooperativity Ceff of the crown mode expe- √ 5 riences four orders of magnitude enhancement with only a few micrometres Terfenol-D offset as plotted in Figure 2d. Ceff is chosen for evaluating mechanical mode shape induced characteristics. Terfenol-D with offset breaks the axial symmetry of the crown mode, creating a first order circumference difference of the toroid as the mechanical mode oscillates, and thus improves the value of Ceff. √ The numerical results show that asymmetry and the bimetallic-like bending effect helps to enhance the sensi- tivity. With an optimal offset of Terfenol-D, low nT/ Hz sensitivity is predicted, which is five times better than the experimental result [7]. It is likely that the experimental results were degraded not only due to a lack of Terfenol-D offset, but also by the epoxy used to fix the Terfenol-D on top of the toroid, reducing the expansion of the silica disk. B. Effect of the Size of the Terfenol-D The single mechanical mode analysis is then applied to a proposed [7] thin disk structure: 1 µm sputter coated Terfenol-D film on top of a 400 nm-thick silica disk. Mag- netometers with sputter coated Terfenol-D have the ad- vantage of a reproducible fabrication process. The silica disk has a radius of 30 µm and the pedestal has a top facet of 15 µm (sketched in Figure 3a inset top). The optical quality factor is taken to be 1×106 [24], a coherent laser source is again used to probe the system with zero detuning and measured via homodyne detection. The effective mass extracted from numerical simulation varies from 1 pg to 3.8 pg depending on the Terfenol-D size, for the radial breathing modes of the device. FIG. 3: (a) sensitivity vs. Terfenol-D disk size for the first order radial breathing-type modes of a thin film structure. The silica disk dominates the mechanical eigenmodes when the Terfenol-D (highlighted with white dashed line) is smaller (left) than the 15 µm radius top facet of silicon pedestal indicated by a vertical line. If the Terfenol-D is larger than the silicon facet, the mechanical motion is hybridized with the Terfenol-D mode (right). A power-law fit is applied to the right side data. Insets are sketches of a thin film magnetometer and of two mechanical eigenmodes; (b) deformation profile induced by axial magnetic field driving for Terfenol-D smaller and larger than the pedestal top facet. Figure 3a shows the relation of sensitivity to the size of the Terfenol-D for the first order radial breathing-type (a)10152025Terfenol-D radius (µm)Bmin(T/√Hz)(b)10-910-810-710-610-5mode. The signal magnetic field drives the radial breath- ing mode in the axial direction to create a magnetic field induced deformation profile as shown in Figure 3b. Unlike an isotropic magnetostrictive material breathing radially under axial magnetic field driving, the spatial profile from the non-isotropic Terfenol-D stretches only in one direc- tion. When the size of the Terfenol-D is larger than the top facet of the silicon pedestal, the silica disk is also significantly affected by the motion from the Terfenol- D. The part of Terfenol-D inside the top pedestal facet (Terfenol-D is highlighted with the white dashed line in the mechanical eigenmode simulation in Figure 3a inset) is motionless because it is obstructed by the silicon pedestal. When the rim of the Terfenol-D reaches outside the top pedestal facet, the device mechanical motion is hybridized with mechanical modes of the Terfenol-D. This leads to a bi-metallic-strip-like effect close to the edge of the top facet of the silicon pedestal across the silica layer, increas- ing the silica displacement and thus allowing for better sensitivity than in the cases where the Terfenol-D is con- fined inside the silicon pedestal. Generally, the sensitivity scales with the size of the motional part of the Terfenol-D. A sensitivity of 2.9 nT/ Hz is predicted when the diame- tre of the Terfenol-D disk covers more than 2/3 of the silica disk in Figure 3a. A power-law fit (y(x) = a · xb with fitting results of a = 7.9 × 10−7 and b = −1.7) is applied to the data with the Terfenol-D radius larger √ than that of the pedestal, predicting a 300 µm radius of Terfenol-D may lead to 50 pT/ Hz sensitivity. To achieve better sensitivity, the size of the Terfenol-D must be larger than the pedestal so as to have large portion of motional Terfenol-D and large bi-metallic-strip-like bend- ing effect, which could be realised by decreasing the size of the silicon pedestal and by increasing the size of the Terfenol-D. √ V. MULTI-MODE ANALYSIS Single mode analysis is limited, in that it only correctly predicts the performance of devices over frequency ranges where only one mechanical mode contributes significantly to the dynamics. In reality, this is rarely the case, and often there is a dense spectrum of mechanical modes (see e.g., Ref. [7, 8]). To extend our analysis to such sit- uations, we use multi-mode analysis from Section III. We first examine the limitations of the single mode anal- ysis and then predict an optimal driving direction of the magnetic field leading to a best predicted sensitivity of an ensemble of mechanical eigenmodes. We examine the limitations of single mode analysis by considering the magnetometer design reported in Ref. [8]. This type of magnetometer has a hole of 14 µm radius in the middle of a silica toroid, which has a 45 µm major ra- dius. A cross-sectional view is shown in Figure 4a, where the outer silicon undercut is 15 µm. The Terfenol-D is modelled as an ellipsoid having the same transverse radius as the silica hole and an axial radius of 16 µm. Mechanical 6 modes with resonant frequencies up to 45 MHz are selec- tively driven with the in-plane Bsig in accordance with the experimental conditions of Ref. [8]. Three windows (∼7 MHz, ∼26 MHz and ∼43 MHz) of interest are selected. Mechanical modes in between are not taken into consider- ation due to their small optomechanical coupling resulting from their symmetrical mode shapes. The power spectral density Snoise(Ω) and magnetic field sensitivity spectrum in Figure 4b are obtained, again choosing QM = 200 for all modes, and setting Qo = 2 × 106 and a coherent laser with power of 1 µW at 1550 nm in an on-resonance ho- modyne detection scheme. With these parameters, the sensor noise floor is dominated by mechanical thermal noise close to the mechanical resonance frequencies, and optical shot noise at other frequencies (Figure 4b top). A single mechanical mode at ΩM/2π=23 MHz has the largest actuation parameter (see Appendix D for cact spectrum) due to a relatively large spatial mode overlap between the mechanical eigenmode (Figure 4b inset) and the magnetic field induced deformation profile (Figure 4c) compared with other modes. However, this particular mode has a very weak optomechanical coupling when the device is modelled uniformly and axial-symmetrically. This prevents the mode from being optically resolved from the thermal noise of others, causing a large difference of the magnetic field sensitivity between the single mode and multi-mode analysis, as shown in triangles and lines in Figure 4b bottom, respectively. To achieve better sensitivity, the direction of the driv- ing magnetic field needs to be optimised. The mechanical mode under magnetic field driving should have both rel- ative large optomechanical coupling and relative good magnetomechanical overlap compared to other modes. As might be expected, and is shown in Figure 4b, top modes with radial-breathing-like motion (Figure 4d top insets show the eigenmodes) offer the largest optomechan- ical coupling. These mechanical modes are at 4.8 MHz, 26 MHz, 27 MHz, 43.2 MHz and 43.4 MHz. When driven axially, the deformation profile due to magnetostriction is also radial, as shown in Figure 4e. This suggests the magnetometer will perform well when axially driven near radial breathing modes. Choosing axial field magnetic field driving, we find the power spectral density, network response and sensitivity shown in Figure 4d. The radial breathing mode at ΩM/2π = 27 MHz, third from left in Figure 4d top inset, reaches a sensitivity of 180 pT/ Hz. We confirm that the result from multi-mode analysis (see Figure 4d bottom blue line) is consistent with single mode analysis (see Figure 4d bottom triangles) for this mode. The actuation parameter is 3200 times larger than if the same mechanical mode is driven by an in-plane magnetic field (see Appendix D for cact values), verifying a strong dependence of the magnetomechanical overlap on the magnetic field direction and the potential for vectorial magnetometry. √ With in-plane magnetic field driving, the sensitivity observed in the experiment 200 pT/ Hz [8] surpasses the modelled sensitivity by around two orders of magni- √ 7 FIG. 4: Multi-mode analysis with a device reported in Ref. [8]. (a) cross-sectional view of the optomechanical magnetometer; (b) top: the power spectral density Snoise(Ω) (blue) is the sum of individual thermal Brownian motion peaks (grey) and coherent laser shot noise on the optical phase quadrature (red); bottom: minimum detectable magnetic field from multi-mode (blue) and single mode (black triangles) analysis driven by in-plane magnetic field. The inset is the mechanical mode with the highest cact at ΩM/2π = 23 MHz; (c) deformation profile induced by in-plane magnetic field far away from mechanical resonance frequencies; (d) top: the power spectral density Snoise(Ω) of the radial-breathing-like mechanical modes. The insets show the mechanical eigenmodes corresponding to each resolved thermal Brownian motion peaks; middle: the magnetic field response Ssignal(Ω)/τ to the axial magnetic field driving; bottom: the sensitivity spectrum from multi-mode (blue) and single mode (black triangles) analysis driven by the axial magnetic field; (e) deformation profile induced by axial magnetic field far away from mechanical resonance frequencies. tude. This is likely due to the fact that the simulated mode at 23 MHz (Figure 4b bottom inset) is thermally resolved in the experiment, which is not the case in the model. This difference can be understood in terms of symmetry. In the model, the symmetry results in a very poor predicted optical transduction sensitivity. However, in the experiment, it can be expected that the symmetry is broken due to fabrication defects resulting in improved sensitivity [25]. VI. CONCLUSIONS We have developed a new versatile approach to model the sensitivity of optomechanical magnetometers, intro- ducing magnetostriction into the elastic wave equation used to solve for mechanical eigenmodes. By numerically solving a modified elastic wave equation for a range of geometries, we model the sensitivity for magnetometers both experimentally demonstrated and not-yet fabricated. The modelling predicts that at least one order of mag- nitude improvement from previous experimental results [8] is possible. The sensitivity of optomechanical mag- netometers can be significantly improved by optimising the size and the shape of the Terfenol-D, by utilizing the bending effect, which arises from a magnetic equivalent of the bi-metallic strip effect, and by optimizations of √ the composition and the annealing process of Terfenol-D, which may lead to sensitivity below 20 pT/ Hz using the piezomagnetic constant in Ref. [26] with micrometre-level resolution. The numerical method developed here is applicable to optomechanical magnetometers with a wide range of geometries and any magnetostrictive materials. A full characterization of the response of the magnetomechani- cal overlap to the variation of the signal magnetic fields direction may allow vectorial optomechanical magnetom- etry, complementary to vectorial optomechanical force sensors [27, 28]. Micro-optomechanical magnetometers with pT/ Hz sensitivity can potentially be applied to detect signals from neurons, similar to recent results with nitrogen-vacancy centre based magnetometers [29] and atomic magnetometers [30], but with benefits of a sim- pler, silicon-chip fabricateable approach, as well as high bandwidth. √ Acknowledgements We appreciate ETREMA Products, Inc. (Ames, IA, US) for providing advice on the choice of Terfenol-D pa- rameters for simulation, and we thank Christopher Baker, Bei-Bei Li, George A. Brawley, Kiran E. Kholsa and James S. Bennett for useful discussions. This research is funded by Australian Research Council Discovery Project DP140100734, and Defence Science and Technology Group projects CERA49 and CERA50. W.B. acknowledges the Australian Research Future Fellowship FT140100650. BsigΩ/2π (MHz)Bmin(T/√Hz)(a)(b)(c)10-610010610203040Snoise(Ω)Ω/2π (MHz)Bmin(T/√Hz)(d) (e)Bsig100103106106101010141020304010-1110-910-710-1010-810-610-410-2Ssignal(Ω)/τ (Hz)Snoise(Ω)8 F F To provide an idea of the relative magnitude of Stherm , Simp F F and Sba F F , we choose the geometry and parameters (in Section V in the main text) of the magnetometer reported in Ref. [8]. Not surprisingly, on mechanical resonances at room temperature, we find that back-action noise is always smaller than the thermal noise by a large margin. For each mechanical mode, this is quantified, roughly, by the ratio of thermal phonon occupancy to effective cooperativity, with the former being in the range of 105 -- 106 for our mechanical frequencies at room temperature, and the later not exceeding 100 for typical parameters. Far from mechanical resonance, the backaction noise will eventually exceed the thermal noise [10]. However, in this regime, the optical shot noise dominates. As a con- sequence, backaction noise can be safely neglected at all frequencies. Appendix B: COMSOL Implementation of the Magnetic Field The magnetic field Bsig is generated by a pair of Helmholtz coils whose axis can be freely rotated in a 4π solid angle as shown in Figure 5a for COMSOL lay- out. To enable the simulation of the magnetic field, the outermost sphere is filled with air. The amplitude of the magnetic field is controlled by inputting a known current in the pair of Helmholtz coils. The coils diametre is set to be more than 40 times larger than the lateral size of the Terfenol-D to ensure a uniform driving magnetic field. Therefore, the direction of Bsig is along the axial axis of the pair of coils. The magnetic field can be viewed by the intersected orthonormal slices on which the magnetic field amplitude is projected, with the colour refers to the amplitude of the magnetic field as shown in Figure 5b. Appendix A: Derivation of the Sensitivity for a Single Mechanical Mode (cid:113) For a single mechanical mode, the minimum detectable magnetic field can be obtained from the actuation pa- rameter cact and the noise force spectral density SF F . Calibrated in the medium of air, the sensitivity at indi- vidual mechanical eigenfrequencies can be written as (A1) Bmin = F F + Sba F F , F F + Simp Stherm 1√ 2πcact √ √ 2π ensures Bmin having the unit of where a factor of 1/ T/ Hz. Noise sources considered are thermal noise and noise from optical measurement including imprecision and back-action. Measurement imprecision comes from the laser shot noise in the optical phase quadrature. Back- action noise is due to the laser shot noise in the optical amplitude quadrature driving the mechanical oscillator. The power spectral density in the unit of force (specif- ically rad·s·N2) for individual mechanical modes driven by noise can be found in Ref. [10]. Here, we extend the calculation to include the back-action noise, as: Stherm F F Simp F F (ΩM) = (ΩM) = 4meff (ΩM)Γ(ΩM)kBT, meff (ΩM)QM(ΩM) 8ηχ(ΩM)2 (cid:12)(cid:12)(cid:12)Ceff (ΩM) (cid:12)(cid:12)(cid:12) , F F (ΩM) = 4meff (ΩM)Γ(ΩM)ΩMCeff (ΩM), Sba (A2) (A3) (A4) in which a factor of 4 in front of the classical thermal force spectrum in Equation (A2) is due to the definition of Γ being the full-width-half-maximum of the mechanical oscillator. Inspection of Equations (A3) and (A4) shows that, despite optomechanical coupling, the effective coopera- tivity Ceff also quantifies the trade-off between better measurement precision and large back-action noise due to the Heisenberg uncertainty relation. Ceff (Ω) is given by Ceff (Ω) ≡ 4g2 κΓ(cid:12)(cid:12)1 − 2iΩ 0(ΩM)N (cid:12)(cid:12)2 = κ 16ηescg2 Γ(cid:12)(cid:12)κ − 2iΩ(cid:12)(cid:12)2 0(ΩM)Nin , (A5) where N is the intra-cavity photon number, Nin (photons ·s−1) is the input photon number flux, and g0 (rad·s−1) is the vacuum optomechanical coupling rate quantifying the optical resonance frequency shift by the mechanical displacement at zero energy excitation. Fibre-device cou- pling here is idealized to be lossless where the intra-cavity and end mirror loss due to the scattering and/or absorp- tion of the light is neglected, leaving the optical decay only counted at the front mirror to be κ as shown in Figure 1a and thus making the cavity escape efficiency ηesc = κ/(κ + 0) = 1. FIG. 5: (a) COMSOL layout for a pair of Helmholtz coils used to generate signal magnetic field. The axis of the pair of coils can be freely rotated in a 4π solid angle. The di- ametre of the coils are more than 40 times larger than the lateral size of the Terfenol-D in the device under test (DUT); (b) intersected orthonormal slices are used to project the am- plitude of the magnetic field; (c) the effect of the eddy current inside Terfenol-D when the signal magnetic field Bsig is driven in plane with frequency below 100 MHz, at 1 GHz and at 10 GHz. Colourmap refers to the magnetic field inside the Terfenol-D. TBy<100 MHz1 GHz10 GHzDUT(a) (b)(c)At high frequencies, the eddy current induced magnetic field opposes the external magnetic field and thus rein- forces most of the magnetic field between the surface and skin depth, leaving most inner part of the magnetostric- tive material unused. This undesired effect is evaluated by varying the frequency of the signal magnetic field Bsig. The magnetometer [7] as illustrated in Figure 2a left is used to perform the eddy current simulation. Figure 5c shows that the skin depth effect starts to take place at frequencies above 1 GHz. The in-plane magnetic field Bsig characterized at the location of Terfenol-D is 7.7 µT, and the colourmap shows the magnetic field inside the Terfenol-D is ∼ 25 µT at low frequencies as shown in Figure 5c left. The value from the colourmap is consistent with the calculation from Bsig and the value of the relative permeability tensor in Table I. The simulation agrees with the simple relation for skin depth δskin = 1/ πf µ0µrσc, where the conductivity for Terfenol-D is σc = 1.67×106 S/m. Mechanical eigenfrequencies of interest of modelled devices are below 50 MHz, far detuned from the range in which eddy currents pose a problem. √ Appendix C: COMSOL Implementation of the Modified Elastic Wave Equation COMSOL's Solid Mechanics module allows users to modify the elastic wave equation and solve it numeri- cally. In the modified elastic wave equation given by i ui(r, Ω)ei, and stress n σnen in three dimensions Equation (8) displacement u =(cid:80) ij σijeij = (cid:80) tensor σ = (cid:80) σxx σxy σxz  = σ1 σ6 σ5 can be fitted into a 3 × 3 matrix as  . σ = σxy σyy σyz σxz σyz σzz σ6 σ2 σ4 σ5 σ4 σ3 9 (C2) (cid:12)(cid:12)(cid:12)(cid:12)H ∆kl + O(∆2 kl) (cid:123)(cid:122) ∆σij ela (cid:125) ∆σij ela + ∆σij driv = (cid:12)(cid:12)(cid:12)(cid:12) + ∂σij ∂Hk (cid:124) ∂σij ∂kl (cid:124) (cid:123)(cid:122) ∆Hk + O(∆H2 k) + ∆σij Maxw (cid:125) ≈ λH ijkl∆kl + e driv ∆σij ijk∆Hk + ∆σij Maxw, where the modulated Maxwell stress tensor ∆σij Maxw [31] describes the stress caused by the interaction of a magne- tized ferromagnetic material and the external magnetic field, and its value depends on the relative permeabil- ity of the magnetostrictive material. Hk is the magnetic field strength inside the Terfenol-D when the external magnetic field strength H0 = Bsig/µ0 (µ0 is the vacuum permeability) magnetizes the magnetostrictive material due to the effect of the demagnetization field. The internal magnetic field Hk decreases as the length of the magnetic rod is reduced from infinite length where Hk = H0 to zero-thickness where Hk = H0/µr, in which µr is the relative permeability of the magnetostrictive material. The last term in the bracket on the RHS of Equation (8) is the input damping σΓ. It is chosen to be proportional to the time derivative of strain as σΓ = ηdamp (the unit of ηdamp is Pa·s) so that the integral form of the elastic wave equation along one dimension (C1) (cid:90) V (cid:90) S ρn ∂2ui ∂t2 dV = (σij ela + σij driv + σij Γ ) · dS (C3) Up to the first term in the bracket on the right-hand side (RHS) of Equation (8) is the elastic stress σela. Elastic stress is connected to strain via a tensor coefficient λijkl as σij ela = λijklkl. The tensor with λijkl being its elements is termed elasticity matrix in COMSOL. For isotropic mate- rials, elements of the elasticity matrix are determined by an isotropic Young's modulus and an isotropic Poisson's ratio, while for anisotropic materials the number of inde- pendent elements can go up to 21 in the 6 × 6 matrix [22]. Terms σela + σdriv in the bracket on the RHS of Equa- tion (8) incorporate both the elastic stress and the stress under the magnetic field driving. Assuming (1) the vari- ation of the AC magnetic field is slow enough for the material to reach deformed equilibrium, (2) the mag- netostrictive material exhibits reversibility, and (3) the operational point is far below the magnetostrictive satu- rated strain defined as the ratio of the maximum material elongation to its original length, the stress-magnetic field relation can be linearly approximated [19]. σela + σdriv with small modulation can be expressed via first order Taylor expansion, when projected onto one dimension, as ela + σij driv + σij followed by a Fourier transformation to the frequency domain can have the damping coefficient Γ in front of the term linear with Ω in the mechanical susceptibility. Only the part of (σij Γ ) perpendicular to dS contributes to the integral. Roughly speaking (simplified to one dimension), meff Ω2 M is related to elasticity matrix and spatial profile of the mechanical eigenmode Ψ(r) and surface area S. The damping factor Γ includes the input damping ηdamp (Pa·s), meff , Ψ(r) and surface area S. The actuation parameter is determined by cact = eS⊥/µ0µr (S⊥ being the area perpendicular to the magnetic field induced stress). Note that Ψ(r) comes from an ansatz of Equation (8) as the displacement u(r, Ω) = Ψ(r)x(Ω) can be decoupled into a spatial and a frequency dependent term. Due to the transverse (axial) symmetry of the Terfenol- D, the independent elements of the elasticity matrix are reduced to only six [19], which are λH 33, λH 44 and λH 66. The explicit tensor form of stress, strain and magnetic field relation is presented as: 11, λH 12, λH 13, λH  =  σxx total σyy total σzz total σyz total σxz total σxy total  (cid:124) λH 11 λH 12 λH 0 0 0 13 λH 12 λH 11 λH 0 0 0 13 13 λH λH 13 λH 0 0 0 33 0 λH 0 0 0 0 44 0 λH 0 0 0 0 44 0 λH 0 0 0 0 66 (cid:123)(cid:122)    (cid:124) elasticity matrix σxx Γ σyy Γ σzz Γ σyz Γ σxz Γ σxy Γ 0 0 0 0 e 15 0 + (cid:124) (cid:123)(cid:122) (cid:125) damping 0 0 0 e 15 0 0 e 13 e 13 e 33 0 0 0 (cid:123)(cid:122) external driving stress 10 Appendix D: Calculation of cact by Lorentzian Fit Actuation parameter cact is extracted by fitting the equation of motion to the Lorentzian distribution. In COMSOL, the phase of the mechanical displacement spec- trum is in the range 0 to −π/2 at frequencies below the resonance frequency and π/2 to 0 above it, this generates an artefact that we remove by taking the absolute value of the displacement. The modified mechanical frequency response given by the Fourier transform of Equation (6) for each single mechanical mode is given by max[u] · meff Bsig = cact −Ω2 − iΓΩ + Ω2 M . (D1)  (cid:125)   xx yy zz yz xz xy  +  Hx (cid:125) Hy Hz , (C4) where small modulation ∆kl and ∆Hk in Equation (C2) is replaced with kl and Hk in the frequency domain. ijkl and piezomag- The input elasticity matrix elements λH netic constant elements e ijk are taken from an experimen- tal measurement biased at 60 kA/m and prestressed at 20 MPa [13] as summarised in Table I. In addition, the density of Terfenol-D and silica are taken as 9250 kg·m−3 and 2203 kg·m−3, respectively. The Maxwell stress tensor in Equation (C2) is neglected in Equation (C4). This is because the contribution of the Maxwell stress ten- sor and magnetostrictive stress (taking the value from Table I) is comparable only when the driving magnetic field is no less than the order of 10 Tesla under linear stress-magnetic-field approximation. Due to the large piezomagnetic constant of magnetostrictive materials, the influence from the Maxwell stress tensor in the Terfenol-D can be safely neglected in experimental condition where magnetic field is well below microtesla (magnetostrictive stress is ∼ 106 times larger than the Maxwell stress). The input external driving stress from Equation (C4) is fitted into a 3×3 matrix in the form of Equation (C1) in COMSOL as external stress after simple matrix product calculation. TABLE I: Coefficients in the magnetomechanical coupling biased at 60 kA/m and prestressed at 20 MPa [13]. λH is the elasticity matrix element, e is the piezomagnetic constant and µσ is the relative magnetic permeability. unit (GPa) λH λH 11 13 107 82.1 e e 90 −166 −168 13 15 λH 12 74.8 e 33 unit (T) λH 33 98.1 no unit λH 44 60 µσ 11 6.9µ0 λH 66 161 µσ 33 4.4 µ0 All the fitting parameters are on the RHS of Equa- tion (D1): cact, Γ and ΩM, while all the three parameters on the LHS of Equation (D1), maxi- mum displacement, effective mass and signal mag- netic field, can be drawn from COMSOL. The COM- SOL syntax for extracting the maximum displacement is sqrt(abs(u^2)+abs(v^2)+abs(w^2)) under volume maximum analysis and the meff is the quotient of solid.rho(abs(w^2)+abs(v^2)+abs(u^2)) under vol- ume integration analysis and maximum displacement where u,v,w are displacements in x, y, z directions. Since damping is input manually, it is important to check whether the damping affects cact or not. By chang- ing the input damping for a wide parameter range of 12500 times variation, it has been verified that the effect of the damping variation on cact is negligible. Figure 6a shows the fit to the Lorentzian distribution for an input damping factor 12.5 times smaller than that of in Fig- ure 6b, and the fit in Figure 6c has the damping factor 12500 times larger than that of in Figure 6a. The fitting is performed on a mechanical mode simulated in Figure 6d. A smaller input damping allows for a clearer Lorentzian fit as shown in Figure 6a. Therefore, in the implementation, input damping is chosen to be as small as possible limited by COMSOL's convergence error. Figure 7 shows the actuation parameter cact of the magnetometer [8] using the parameters in Section V in three frequency sections of mechanical modes. Blue dots present the cact for mechanical modes under in-plane mag- netic field driving, while orange dots are for mechanical modes under axial driving. As can be seen, the actuation parameter varies over many orders of magnitude both for different mechanical modes with the same driving mag- netic field direction (comparison of dots among the same colour) and for the same mode driving by the magnetic field in two orthogonal directions (in comparison between red and blue dots at the same mechanical eigenfrequency, connected via a black vertical line in Figure 7). 11 when real(u)/real(v)/real(w) is 0, and dx,dy,dz are displacements after the synchronization. A line integral with integrand 1 along the outermost circumference of the devices is followed after the phase synchronization, which results in the circumference at a particular phase. Linear optomechanical coupling is evalu- ated through δL by taking the difference of circumference synchronized at phase 0 and phase π where Equation (E1) is multiplied by a phase factor exp(ipi). Figure 6a,b shows the constant ξ across a mechanical resonance with 12.5 times variation of input damping, showing that ξ is insensitive to the mechanical quality. ξ is missing in Figure 6c due to the numerical error at large manually input damping where ξ spectrum is far away from con- stant. Plotting ξ spectrum offers a way of sanity check of possible numerical errors. Appendix F: From ξ to Optomechanical Coupling With the calculated geometrical factor ξ, the value of the parameters describing optomechanical coupling Ceff is easy to achieve. The optomechanical coupling rate is defined as g0 ≡ G · xzpf , where zero-point mo- tion xzpf =(cid:112)/(2meff ΩM), and G (rad·m−1s−1) is the optomechanical coupling strength quantifying the shift of optical resonance frequency δω0 by the mechanical displacement as G = δω0 max[u] . (F1) For a Fabry -- P´erot type and micro-toroidal structure cavity with length L, the shift of the optical resonance frequency is linked with the change of the cavity length by δL/L = δω0/ω0. Inserting the expression δL/L = δω0/ω0 into Equation (F1) leads to G = ω0 · ξ/L. (F2) Therefore, g0 and thereby Ceff can then be written as a function of the geometrical factor ξ. For an individual mechanical eigenmode at ΩM and based on Equation (A5), the expression for the effective cooperativity as a function of the geometrical factor ξ can be calculated to Ceff (Ω) = ξ(ΩM)2 meff (ΩM)ΩMΓ(ΩM) L2(cid:0)κ2 + 4Ω2(cid:1) , (F3) · 8ηescNinω2 0 where the front part on the RHS is mode dependent, ξ(ΩM) and meff (ΩM) can be accurately extracted from the numerical simulation. The optical resonance frequency ω0 and length of the cavity L are input parameters, and the only empirical parameters left are the optical decay κ and the mechanical damping factor Γ with the assumption of an idealized lossless cavity escape efficiency ηesc = 1. FIG. 6: (a) a fit to the Lorentzian distribution, and ξ spec- trum for a small input material damping; (b) the fit and ξ spectrum with input damping factor 12.5 times larger than that of a); (c) the fit with damping factor 12500 times larger than that of a); (d) all the fits and ξ spectrum are performed on the mechanical mode with Terfenol-D position offset from the centre. FIG. 7: Actuation parameter cact for the magnetometer [8] using the parameters in Section V. A black vertical line connects the same mechanical mode under magnetic field driving in both in-plane and axial directions. Appendix E: Calculation of Geometrical Factor ξ This numerical modelling offers an accurate way to compute the effective cooperativity Ceff from the geomet- rical factor ξ defined in Equation (9). The optical path length L corresponds to the outermost circumference in micro-toroidal resonators. The change of circumference δL(Ω) in a harmonic oscillation at a mechanical resonance can be obtained by synchronizing the output displace- ment to the maximum and minimum amplitudes. The implementation of the synchronization in COMSOL is to multiply a phase shift in all three Cartesian coordinates in the deformable mesh setting as dx=uexp(-iatan(imag(u)/(real(u)+1e-16))), dy=vexp(-iatan(imag(v)/(real(v)+1e-16))), dz=wexp(-iatan(imag(w)/(real(u)+1e-16))), (E1) where 1e-16 in the denominator is an example of adding a small value to eliminate the error of dividing by 0 1.080951.080960.20.610.0610.0660.07110×1.0809561.0809571.080958Ω/2π (Hz)×1074812-70.0610.0660.071×10-7 m/B(µm kg/ T)e(cid:31)ref(a) (b)(c) (d) 0.20.61×10-101.07 1.08 1.09ξ(Ω)Ω/2π (Hz)×107(µm kg/ T)Ω/2π (Hz)×107 max[u] m/Be(cid:31)ref max[u]10203040ΩM/2π (MHz)cact (N/T)in-plane Brefaxial Bref10-910-710-510-3[1] Hmlinen, M.; Hari, R.; Ilmoniemi, R.J.; Knuutila, J.; Lounasmaa, O.V. Magnetoencephalography - theory, in- strumentation, and applications to noninvasive studies of the working human brain. Rev. Mod. Phys. 1993, 65, 413 -- 497, doi:10.1103/RevModPhys.65.413. [2] Mehlin, A.; Xue, F.; Liang, D.; Du, H.F.; Stolt, M.J.; Jin, S.; Tian, M.L.; Poggio, M. Stabilized Skyrmion Phase Detected in MnSi Nanowires by Dynamic Can- tilever Magnetometry. Nano Lett. 2015, 15, 4839 -- 4844, doi:10.1021/acs.nanolett.5b02232. [3] Chary, K.V.R.; Govil, G. NMR in Biological Systems- From Molecules to Humans; Springer: Dordrecht, The Netherlands, 2008; pp. 130 -- 162, ISBN 987-1-4020- 6679-5. [4] Wolf, T.; Neumann, P.; Nakamura, K.; Sumiya, H.; Ohshima, T.; Isoya, J.; Wrachtrup, J. Subpicotesla Di- amond Magnetometry. Phys. Rev. X 2015, 5, 021009, doi:10.1103/PhysRevX.5.021009. [5] Jensen, K.; Kehayias, P.; Budker, D. Magnetometry with Nitrogen-Vacancy Centers in Diamond. In High Sensitivity Magnetometers; Grosz, A., Haji-Sheikh, M.J., Mukhopadhyay, S.C., Eds.; Springer International Pub- lishing: Cham, Switzerland, 2017; p. 570. [6] Muessel, W.; Strobel, H.; Linnemann, D.; Hume, D.B,; Oberhalter, M.K. Scalable Spin Squeezing for Quantum-Enhanced Magnetometry with Bose -- Einstein Condensates. Phys. Rev. Lett. 2014, 113, 103004, doi:10.1103/PhysRevLett.113.103004. [7] Forstner, S.; Prams, S.; Knittel, J.; van Ooijen, E.D.; Swaim, J.D.; Harris, G.I.; Szorkovszky, A.; Bowen, W.P.; Rubinsztein-Dunlop, H. Cavity Optomechanical Magnetometer. Phys. Rev. Lett. 2012, 108, 120801, doi:10.1103/PhysRevLett.108.120801. [8] Forstner, S.; Sheridan, E.; Knittel, J.; Humphreys, C.L.; Brawley, G.A.; Rubinsztein-Dunlop, H.; Bowen, W.P. Ultrasensitive optomechanical magnetometry. Adv. Mater. 2014, 26, 6348 -- 6353, doi:10.1002/adma.201401144. [9] Kirtley, J.R.; Ketchen, M.B.; Stawiasz, K.G.; Sun, J.Z.; Gallagher, W.J.; Blanton, S.H.; Wind, S.J. High- resolution scanning SQUID microscope. Appl. Phys. Lett. 1995, 66, 1138 -- 1140, doi:10.1063/1.113838. [10] Bowen, W.P.; Milburn, G.J. Quantum Optomechanics; CRC Press: London, UK, 2015; pp. 1 -- 92, ISBN 978-1- 4822-5915-5. [11] Forstner, S.; Knittel, J.; Sheridan, E.; Swaim, J.D.; Rubinsztein-Dunlop, H.; Bowen, W.P. Sensitivity and performance of cavity optomechanical field sensors. Pho- ton. Sens. 2012, 2, 259 -- 270, doi:10.1007/s13320-012-0067- 2. [12] Forstner, S.; Knittel, J.; Rubinsztein-Dunlop, H.; Bowen, W.P. Model of a microtoroidal magnetometer. Proc. SPIE Opt. Sens. Detect. II 2012, 8439, doi:1117/12.920057. [13] Claeyssen, F.; Bossut, R.; Boucher, D. Modeling and Characterization of the Magnetostrictive Coupling. In Proceedings: Power Transducers for Sonics and Ultrasonics; Hamonic, B.F., Wilson, O.B., Decarpigny, J.-N., Eds.; Springer: Berlin/Heidelberg, Germany, 1991; pp. 132 -- 151. 12 doi:10.3390/s7091757. [15] Schliesser, A.; Anetsberger, G.; Rivire, R.; Arcizet, O.; Kippenberg, T.J. High-sensitivity monitoring of mi- cromechanical vibration using optical whispering gallery mode resonators. New J. Phys. 2008, 10, 095015, doi:10.1088/1367-2630/10/9/095015. [16] Lee, K.H.; McRae, T.G.; Harris, G.I.; Knittel, J.; Bowen, W.P. Cooling and control of a cavity optoelectrome- chanical system. Phys. Rev. Lett. 2010 104, 123604, doi:10.1103/PhysRevLett.104.123604. [17] Hauer, B.D.; Doolin, C.; Beach, K.S.D.; Davis, J.P. A general procedure for thermomechanical calibration of nano/micro-mechanical resonators. Ann. Phys. 2013, 339, 181 -- 207, doi:10.1016/j.aop.2013.08.003. [18] Pinarda, M.; Hadjarb, Y.; Heidmannc, A. Effective mass in quantum effects of radiation pressure. Eur. Phys. J. D 1999, 7, 107 -- 116, doi:10.1007/s100530050354. [19] Engdahl, G. Modeling of Giant Magnetostrictive Mate- rials. In Handbook of Giant Magnetostrictive Materials; Academic Press: Los Angeles, CA, USA, 1999; pp. 127 -- 174, ISBN 978-0-12-238640-4. [20] Claeyssen, F.; Lhermet, N.; Le Letty, R.; Bouchilloux, P. Actuators, transducers and motors based on giant magnetostrictive materials. J. Alloys Compds. 1997, 258, 61 -- 73, doi:10.1016/S0925-8388(97)00070-4. [21] Kannan, K.S. Galerkin Finite Element Scheme for Mag- netostrictive Structures and Composites. Ph.D. Thesis, University of Maryland, College Park, MD, USA, 1997; pp. 30 -- 90. [22] Landau, L.D.; Lifshitz, E.M. Theory of Elasticity Vol- ume 7 of Course of Theoretical Physics, 3rd ed.; Elsevier: Burlington, NJ, USA, 1986; pp. 1 -- 15, 87 -- 94, ISBN 978-0- 08-057069-3. [23] Wilson, D.J.; Sudhir, V.; Piro, N.; Schilling, R.; Ghadimi, A.; Kippenberg, K.J. Measurement-based control of a me- chanical oscillator at its thermal decoherence rate. Nature 2015, 524, 325 -- 329, doi:10.1038/nature14672. [24] Kippenberg, T.J.; Spillane, S.M.; Armani, D.K.; Vahala, K.J. Fabrication and coupling to planar high-Q silica disk microcavities. Appl. Phys. Lett. 2003, 83, 797 -- 799, doi:10.1063/1.1593833. [25] Kippenberg, T.J.; Spillane, S.M.; Vahala, K.J. Modal coupling in traveling-wave resonators. Opt. Lett. 2002, 19, 1669 -- 1671, doi:10.1364/OL.27.001669. [26] Verhoeven, J.D.; Gibson, E.D.; Mcmasters, O.D.; Osten- son, J.E. Directional Solidification and Heat Treatment of TerfenoI-D Magnetostrictive Materials. Metall. Trans. A 1990, 21, 2249 -- 2255, doi:10.1007/BF02647887. [27] Rossi, N.; Braakman, F.R.; Cadeddu, D.; Vasyukov, D.; Tutuncuoglu, G.; Fontcuberta, I.M.A.; Pog- gio, M. Vectorial scanning force microscopy using a nanowire sensor. Nat. Nanotechnol. 2016, 12, 150 -- 155, doi:10.1038/nnano.2016.189. [28] De Lepinay, L.M.; Pigeau, B.; Besga, B.; Vincent, P.; Poncharal, P.; Arcizet, O. A universal and ultra- sensitive vectorial nanomechanical sensor for imaging 2D force fields. Nat. Nanotechnol. 2017, 12, 156 -- 162, doi:10.1038/nnano.2016.193. [14] Ramos, D.; Mertens, J.; Calleja, M.; Tamayo, J. Study of the Origin of Bending Induced by Bimetallic Ef- fect on Microcantilever. Sensors 2007, 7, 1757 -- 1765, [29] Barry, J.F.; Turner, M.J.; Schloss, J.M.; Glenn, D.R.; Song, Y.; Lukin, M.D.; Park, H.; Walsworth, R.L. Optical magnetic detection of single-neuron ac- tion potentials using quantum defects in diamond. Proc. Natl. Acad. Sci. USA 2016, 113, 14133 -- 14138, doi:10.1073/pnas.1601513113. [30] Jensen, K.; Budvytyte, R.; Thomas, R.A.; Wang, T.; Fuchs, A.M.; Balabas, M.V.; Vasilakis, G.; Mosgaard, L.D.; Staerkind, H.C.; Muller, J.H.; et al. Non-invasive detection of animal nerve impulses with an atomic mag- netometer operating near quantum limited sensitivity. Sci. Rep. 2016, 6, 29638, doi:10.1038/srep29638. [31] Griffiths. D.J. Introduction to Electrodynamics, 3rd ed.; Prentice Hall International. Inc.: Englewood Cliffs, NJ, USA, 1999; pp. 351 -- 355, ISBN 0-13-805326-X. 13
1801.07527	1	1801	2018-01-23T13:20:08	Graphene gas pumps	[ "physics.app-ph" ]	We report on the development of a pneumatically coupled graphene membrane system, comprising of two circular cavities connected by a narrow trench. Both cavities and the trench are covered by a thin few-layer graphene membrane to form a sealed dumbbell shaped chamber. Local electrodes at the bottom of each cavity allow for actuation of each membrane separately, enabling electrical control and manipulation of the gas flow inside the channel. Using laser interferometry, we measure the displacement of each drum at atmospheric pressure, as a function of the frequency of the electrostatic driving force and provide a proof-of-principle of using graphene membranes to pump attolitre quantities of gases at the nanoscale.	physics.app-ph	physics	Graphene gas pumps D. Davidovikj1,∗ D.Bouwmeester1, H. S. J. van der Zant1, and P. G. Steeneken1,2 1Kavli Institute of Nanoscience, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands 2Department of Precision and Microsystems Engineering, Delft University of Technology, Mekelweg 2, 2628 CD, Delft, The Netherlands 8 1 0 2 n a J 3 2 ] h p - p p a . s c i s y h p [ 1 v 7 2 5 7 0 . 1 0 8 1 : v i X r a We report on the development of a pneumatically coupled graphene membrane system, comprising of two circular cavities connected by a narrow trench. Both cavities and the trench are covered by a thin few-layer graphene membrane to form a sealed dumbbell shaped chamber. Local electrodes at the bottom of each cavity allow for actuation of each membrane separately, enabling electrical control and manipulation of the gas flow inside the channel. Using laser interferometry, we mea- sure the displacement of each drum at atmospheric pressure, as a function of the frequency of the electrostatic driving force and provide a proof-of-principle of using graphene membranes to pump attolitre quantities of gases at the nanoscale. Pumps have been of importance for humanity since early civilization. The Egyptians used a contraption called "shadoof" to take out water from the Nile that was used for irrigation. As technology progressed, better pumps usually meant higher pressure, larger flow, and hence, higher power. Micro- and nanofluidics in the past thirty years have substantially changed the way these devices are benchmarked. Microscale pumps are an es- sential ingredient in a microfluidic system, and the rapid advancements of biosciences require continually more de- vices capable of accurate micromixing and microdosing. This, in turn, imposes better controllability, better ac- curacy, lower operational power, and much smaller flow rates [1 -- 3]. With respect to the first electrostatically actuated membrane pumps [4, 5], that were presented more than 25 years ago, a tremendous reduction in size has been achieved. Pumps are also of interest for driving pneumatic actuators in micro- and nanoelectromechani- cal motors. The properties of graphene, like its atomic scale thickness and extreme flexibility, are very promising for further miniaturization of such nanofluidic devices. Since the first realization of mechanical graphene de- vices [6], suspended 2D materials have attracted increas- ing attention in the MEMS/NEMS communities. Many device concepts have been proposed, including pressure sensors [7, 8], gas sensors [9, 10], mass sensors [11, 12], and graphene microphones [13, 14]. The high tension and low mass of graphene membranes have also inspired their implementation as high-speed actuators in micro- loudspeakers [15]. Another attractive aspect of graphene membranes is their hermeticity [16] and the ability to controllably introduce pores that are selectively perme- able to gases [9]. Although gas damping forces limit graphene's Q-factor at high frequencies, they provide a useful but little explored route towards graphene pumps and nanofluidics. For efficient pumps and pneumatics it is essential that most of the available power is used to move and pressurize the fluid, while minimizing the power required to accelerate and flex the pump mem- FIG. 1. Working principle of the pump. a, A 3D schematic of the device: the graphene flake is covering two circular cavities that are connected through a narrow trench. b, Schematic cross-section of the pumps and actuation mech- anism for the case that pump 1 is actuated. brane while minimizing leakage of the fluid outside of the system. In these respects, the low mass and high flexibility, combined with the impermeability of graphene membranes [16] provide clear advantages. In this work we realise a system of two nanochambers (with a total volume of 7 fl) coupled by a narrow trench and sealed using a few-layer graphene flake. By design- ing a chip with individually accessible electrodes we con- struct a graphene micropump, capable of manipulating the gas flow between the two chambers using small driv- ing voltages (Vdc ≤ 1 V). Increasing the gas pressure in one of the nanochambers results in pneumatic actuation of the graphene drum that covers the other nanocham- abAuPd topAuPd electrode 1Quartz substratespin-onglassgas flowchannelpneumaticactuationAuPd electrode 2electrostaticactuation FACTgraphenepump 1VACTpump 2ber via the connecting gas channel. To measure the dis- placement of the drums, we use laser interferometry and demonstrate successful pumping of gas between the two pneumatically coupled graphene nanodrums. metallic island on which the dumbbell shape is patterned. Graphene is transferred last (as described in [17]) and it is visible in the image as a darker area on top of the metallic island. 2 Images of the fabricated device. a, Scan- FIG. 2. ning electron microscopy (SEM) image of the device prior to graphene transfer. b, Optical micrograph of the device after graphene transfer. Device description The device concept is presented in Fig. 1a. Two circular AuPd electrodes (thickness: 60 nm) at the cavity bottom (one for addressing each of the membranes) are separated by a thin layer (130 nm) of spin-on-glass (SOG) silicon oxide from the metallic (AuPd) top electrode (thickness: 85 nm). The few-layer (FL) graphene flake (black), with a thickness of 4 nm, is in direct electrical contact with this top electrode. The entire device is fabricated on top of a quartz substrate to minimize capacitive cross-talk. The device fabrication is described in detail in [17]. A cross- section along the direction of the trench of the device is shown in Fig. 1b, which illustrates the working principle. The actuation voltage VACT,1 is applied between AuPd electrode 1 and ground, while keeping AuPd electrode 2 and the AuPd top electrode grounded. As a result, pump 1 experiences an electrostatic force FACT, causing it to deflect downward. This compresses the gas under- neath the membrane and the induced pressure difference causes a gas flow through the channel between the two nanochambers. This results in a pressure increase that causes the other membrane (pump 2) to bulge upward. Figure 2a shows a false-coloured SEM image of the de- vice after fabrication. The AuPd is shown in light (bot- tom electrodes) and dark (top electrode) yellow. The diameter of each drum is 5 µm and the trench connect- ing them is 1 µm wide and 3 µm long. Figure 2b shows an optical image of the measured device. The image shows the two bottom electrodes, together with the top FIG. 3. Measurement setup. a, Setup for electrostatic actuation and interferometric motion readout of either of the pumps. PD: photodiode, OSC: oscilloscope, AWG: arbitrary waveform generator, BE: beam expander and PBS: polarized beam splitter. b, Offset photodiode voltage (red curve) for a triangular input signal VACT (black curve). The dashed blue curve represents the input voltage squared, normalized to the photodiode voltage: V 2 ACT,norm. = α(VACT + β)2. The term β = − 0.13 V accounts for residual charge on the graphene flake [18]. Readout The readout of the drum motion of the is performed using a laser interferometer, shown schematically in Fig. 3a. A red HeNe laser is focused on one of the graphene mem- branes, and the sample is mounted in a pressure cham- ber in a N2 environment at ambient pressure and room temperature. When the membrane moves, the optical interference between the light reflected from the bottom electrode and the light reflected from the graphene causes the light intensity on the photodiode detector to depend strongly on the drum position. By lateral movement of the laser spot, the motion of either of the pumps can be detected. The photodiode signal is read out via an inter- nal first-order low-pass filter with a cut-off frequency of 50 kHz. aAuPd bottomelectrode 1AuPd bottomelectrode 25 μmpump 1pump 2grapheneAuPd top electrode channelb5 μmPDOSCAWGBEPBSλ/450x632 nm4123ch1ch2HeNe laser Time (ms)VPD (mV)VACT (V)06-6V ACT, norm.2V ACTVPD, offset201510506420ab3 FIG. 4. Time domain measurements. a, Measured displacement (photodiode voltage) of pump 1 when actuating pump 1 (measurement 11) and drum 2 (measurement 21). b, Measured displacement (photodiode voltage) of drum 2 when actuating pump 1 (measurement 12) and drum 2 (measurement 22). The actuation voltage is shown on the right y-axis. Measurements of each of the drums are performed at constant laser position to ensure that the transduction of the system (VPD/z) is constant. plying VACT on pump 1 while keeping pump 2 grounded (dark blue curve) or when actuating pump 2 while keep- ing pump 1 grounded (light blue curve). Both curves show a main frequency component that is coinciding with the frequency of the driving signal, meaning that the de- tected motion is a consequence of the applied actuation. However, when switching the actuation to pump 2 it is seen that the photodiode voltage (VPD,21) is 180 degrees out of phase with respect to VPD,11. This is indicative of an out-of-phase motion of the two drums. Such effect is possible only if the actuation of pump 1 (in the 21 config- uration) is pneumatic, i.e., mediated by gas displacement from one chamber to the other. For electrostatic actuation, the two bottom electrodes are connected to two channels of an arbitrary waveform generator, where one is grounded and the other one is ac- tuated (Figs. 1b and 3). The actuation voltage (VACT) on each of the drums and the photodiode voltage (VPD) are measured using an oscilloscope. The top electrode (i.e., the graphene flake) is electrically grounded during the measurements. Since there are 2 pumps that can be actuated (pump 1 and pump 2) and either of them can be detected with the red laser there are 4 measurement con- figurations indicated by VPD,11,VPD,21,VPD,12 and VPD,22, where the first number indicates the pump that is actu- ated and the second number indicates the pump that is read out. We first characterize the responsivity of the system by applying a triangular voltage signal to one of the drums while measuring its motion with the laser. The mea- surement is shown in Fig. 3b. The force acting on the drum scales quadratically with VACT and therefore, for small amplitudes, it is expected that the amplitude of the drum would also depend quadratically on VACT (assum- ing FACT = −kz, see Supporting Information Section II). The fact that the voltage read out from the photodi- ode matches the scaled square of the input voltage (blue curve in Fig. 3b) confirms that the assumption of linear transduction (V 2 ACT ∝ VPD) of the motion is valid. Gas pump and pneumatic actuation Pneumatic actuation is one of the most efficient ways to transfer force over large distances in small volumes. At the microscale, the pneumatic coupling also has the advantage of converting the attractive downward elec- trostatic force on pump 1 to an upward force on the graphene membrane of pump 2 (Fig. 1b). Thus, proof for gas pumping and pneumatic actuation can be obtained by detecting that the drums move in opposite directions. The drums are actuated using a square-wave voltage input VAC,p−p = 1 V with a frequency of 1.3 kHz, plotted in Fig. 4a and Fig. 4b (grey curves). Figure 4a shows a measurement of the displacement of pump 1, when ap- The same experiments are repeated in Fig. 4b when moving the laser spot to pump 2. The red curve repre- sents the case when pump 2 is electrically actuated while keeping pump 1 grounded and the pink curve represents the case when pump 1 is electrically actuated and pump 2 is kept grounded. The same conclusion can be drawn: the two curves are 180 degrees out-of-phase, confirming that the drums move in opposite directions. The differences in signal amplitudes in Fig. 4 are at- tributed to differences in the effective cavity depths be- tween the pumps that affect the actuation/detection ef- ficiency (this may happen due to morphological imper- fections in the graphene flake). To confirm that the cou- pling is mediated by gas, the experiment is repeated at low pressure. After keeping the sample at 0.1 mbar for 48 hours, the gas is completely evacuated from the cav- ity [16]. In this case no sign of motion of the second drum is observed in the VPD,12 signal, showing that pneumatic actuation is absent in vacuum (see Supporting Informa- tion Section I). Assuming that the cavities are hermetically sealed by the graphene (valid for very low permeation rates [16]), the pneumatically coupled graphene pump system can be modelled in the quasi-static regime by a set of two lin- ear differential equations describing the pressure increase ∆Pi in each of the chambers. The pressure difference can then be related to the displacement zi of the drums (de- tails of the model and the derivation are given in the Sup- 2212V ACT1121V ACTVPD (mV)Time (ms)Time (ms)VACT (V)abVPD (mV)VACT (V)01234012341010012-1-20.511.5-0.5-104 FIG. 5. Frequency domain measurements. Bode plots (a, magnitude and b, phase) of the system for actuation of pump 2. The data points are coloured according to the measurement scheme: the red points represent actuation and detection at pump 2, while the light blue represent actuation at pump 2 and measurement of pump 1. The fit is according to the model represented with Equations (1) and (2). porting Information Section I). In the frequency domain the solutions of these differential equations can be writ- ten in terms of the Fourier transforms F of the solutions: z1(ω) = F(z1), z2(ω) = F(z2) and F(ω) = cF(V 2 2 ), where F is the actuation pressure and c is a function of the squeeze number and the gap size g0 = 155 nm. When the actuation signal is applied to pump 2, the response is given by: The resulting Bode plots are shown in Fig. 5. It can be seen that both the magnitude and phase of the result- ing frequency response curves are flat up to a frequency of 10 kHz. At higher frequencies the amplitude of the motion of the second drum drops, which suggests that at these frequencies the pumping efficiency starts to be- come limited by gas dynamics through the narrow chan- nel. Fits using the model described by Equations (1) and (2) show that the response of the pumps correspond to a first-order RC low-pass filter with a characteristic time constant of τ = 39.3 ± 3.4 µs, resulting in a cut-off frequency of 25.4 ± 2.2 kHz. z1(ω) = 1 2 1 1 + iωτ A k F(ω); (1) (cid:17) A 1 + 1 z2(ω) = −(cid:16)1 a 2 F(ω), 1 + iωτ (2) where z1 and z2 are the displacements of pump 1 and pump 2 respectively, ω is the actuation frequency, A is the area of each drum, k is the spring constant of the drums and a is the squeeze number. The time constant τ is then given by: k τ = 1 + a 2b , (3) where the constant b is related to the gas flow through the channel. Assuming a laminar Poiseuille flow, b is de- pendent on the geometry of the channel and the effective viscosity of the gas, in this case nitrogen (see Supporting Information Section II). To investigate the nanoscale gas dynamics experimen- tally, the frequency response of the system is measured. The actuation voltage is applied on pump 2. The fre- quency of the square-wave input signal (VACT(t), see Fig. 4) is varied from 510 Hz to 23 kHz. For each ac- tuation frequency, the Fourier transform is taken of both the input and output signal. By taking the ratio of the input and output at each of the driving frequencies a fre- quency response plot is obtained. We make use of the fact that the input square-wave contains higher harmon- ics to increase the amount of data acquired by a single time response signal, thereby increasing the frequency resolution. The demonstrated graphene-based pump system is not only of extraordinarily small size (total volume of 7 fl), but it is also capable of pumping very small amounts of gas: assuming the spring constant to be in the order of k ≈ 1 N/m, less than 80 al of gas is pumped through the channel each cycle. The thermal noise, due to charge fluctuations on the capacitor plates, sets a lower limit to √ Hz, which is equiva- the pump rate of less than 1 zl/ √ lent to less than 30 N2 molecules/ Hz at ambient pres- sure and room temperature. The maximum electrostatic pressure that can be generated by the graphene pump with the given geometry is 0.5 bar, limited by the break- down voltage of the dielectric (Vb = 16 V). The typical force exerted at VACT = 1 V is 4 nN, corresponding to an electrostatic pressure of 2 mbar. Besides the pneumatic actuation and pumping, the system also allows the study of gas dynamics in channels of sub-micron dimensions, where the free path length of molecules is smaller than the channel height, even at at- mospheric pressure. By controllably introducing pores in the graphene, the graphene pump can be used for molec- ular sieving of gases, or even aspiration and dispensing of liquids. The presented system can therefore be used as a platform for studying anomalous viscous effects in narrow constrictions as well as graphene-gas interactions at the nanoscale. It thus provides a route towards scaling down nanofluidic systems by using graphene membranes coupled by nanometre-sized channels. Frequency (Hz)Frequency (Hz)Magnitude (dB)Phase (rad)2122fitab10310410510310410510010-110-210-310-4-π/20π/2Acknowledgements This work was supported by the Netherlands Organisa- tion for Scientific Research (NWO/OCW), as part of the Frontiers of Nanoscience (NanoFront) program and the European Union Seventh Framework Programme under grant agreement n◦ 696656 Graphene Flagship. Parts of this manuscript have been published in the form of a proceeding at the IEEE 31th International Conference on Micro Electro Mechanical Systems [19]. ∗ [email protected] [1] D. J. Laser and J. G. Santiago, Journal of Micromechan- ics and Microengineering 14, R35 (2004). [2] B. D. Iverson and S. V. Garimella, Microfluidics and Nanofluidics 5, 145 (2008). [3] S. Lee, R. An, and A. J. Hunt, Nature Nanotechnology 5, 412 (2010). [4] W. Judy, T. Tamagawa, and D. L. Polla, Proceedings [5] R. Zengerle, A. Richter, and H. Sandmaier, Proceedings IEEE MEMS , 182 (1991). IEEE MEMS , 19 (1992). [6] J. S. Bunch, A. M. Van Der Zande, S. S. Verbridge, I. W. Frank, D. M. Tanenbaum, J. M. Parpia, H. G. Craighead, and P. L. McEuen, Science 315, 490 (2007). [7] A. Smith, S. Vaziri, F. Niklaus, A. Fischer, M. Sterner, A. Delin, M. Östling, and M. Lemme, Solid-State Elec- tronics 88, 89 (2013). 5 [8] R. J. Dolleman, D. Davidovikj, S. J. Cartamil-Bueno, H. S. J. van der Zant, and P. G. Steeneken, Nano Letters 16, 568 (2016). [9] S. P. Koenig, L. Wang, J. Pellegrino, and J. S. Bunch, Nature Nanotechnology 7, 728 (2012). [10] R. J. Dolleman, S. J. Cartamil-Bueno, H. S. van der Zant, and P. G. Steeneken, 2D Materials 4, 011002 (2016). [11] A. Sakhaee-Pour, M. Ahmadian, and A. Vafai, Solid State Communications 145, 168 (2008). [12] J. Atalaya, J. M. Kinaret, and A. Isacsson, Europhysics Letters 91, 48001 (2010). [13] D. Todorović, A. Matković, M. Milićević, D. Jovanović, R. Gajić, I. Salom, and M. Spasenović, 2D Materials 2, 045013 (2015). [14] Q. Zhou, J. Zheng, S. Onishi, M. Crommie, and A. K. Zettl, Proceedings of the National Academy of Sciences 112, 8942 (2015). [15] Q. Zhou and A. Zettl, Applied Physics Letters 102, 223109 (2013). [16] J. S. Bunch, S. S. Verbridge, J. S. Alden, A. M. van der and P. L. Zande, J. M. Parpia, H. G. Craighead, McEuen, Nano Letters 8, 2458 (2008). [17] D. Davidovikj, P. H. Scheepers, H. S. J. van der Zant, and P. G. Steeneken, ACS Applied Materials & Interfaces 9, 43205 (2017). [18] C. Chen, S. Rosenblatt, K. I. Bolotin, W. Kalb, P. Kim, I. Kymissis, H. L. Stormer, T. F. Heinz, and J. Hone, Nature Nanotechnology 4, 861 (2009). [19] D. Davidovikj, D. Bouwmeester, H. S. J. van der Zant, and P. G. Steeneken, Proceedings IEEE MEMS (2018). Supporting Information: Graphene gas pumps I. Model of the pump system Equations of motion 6 In this section the model for the two drum system will be explained, starting with the assumptions that were made in order to arrive at the model. The drums are modelled as simple harmonic oscillators. A parallel plate capacitor model is taken to model the electrostatic force on the drum, which holds for small deflections of the membrane with respect to gap. The gas inside the circular cavities is modelled as an ideal gas and gas inertia is neglected. The interactions of the gas are considered to be isothermal. Poiseuille flow through the trench between the two drums is considered. The mechanics of the drums are described using Newton's second law of motion. The forces that act on the drums are the tension force of the drums (assuming equal spring constants k and masses m), the pressure force acting on the drum and the electrostatic forces coming from the charge stored in the membrane-electrode capacitor. No damping is considered apart from damping due to the gas pressure. The electrostatic force is applied to the first drum (pump 1). We name deflection of drum i with respect to the gap zi, such that a positive value of zi corresponds to the drum bulging upward. We consider the outside air to be at ambient pressure P, while the pressure inside chamber i is Pi. The pressure difference across the drum is be called ∆Pi. The gap size is denoted as g0, ε0 is the vacuum permittivity and A = r2π is the area of each of the drums. In terms of these quantities the equations of motion are: dt2 = −kz1 + ∆P1A − V 2ε0A d2z1 2(g0 + z1)2 ; d2z2 dt2 = −kz2 + ∆P2A . (4) (5) m m  FIG. S1. Schematic of the device. a, Top view of the pump system. b, A side view of the cross section through the dashed blue line in a. A potential ∆V is applied on pump 1 that results in the membrane compressing the gas, causing an increase of pressure in the chamber P1 = P + ∆P2. This causes gas flow through the channel and results in pump 2 bulging upward. Since the channel is smaller than the chamber, it will obstruct the flow of gas, hence the pressure in the other chamber P2 = P +∆P2 will lag. With this, the mechanics are fully described. These equations have one driving force, the electrostatic force experienced by pump 1. The pressure force due to the gas is not a driving force and should react to the motion of the membrane. In order to describe the pressure in the drums, the ideal gas law is taken: P1 = P + ∆P1 = n1 ¯RT P2 = P + ∆P2 = n2 ¯RT V1 V2 A(g0 + z1); = n1 ¯RT = n2 ¯RT A(g0 + z2) . (6) (7) abP + ΔP1ΔVP + ΔP2η12Pg0lrlwThe quantity ni in these equations stands for the amount of moles of gas in chamber i. These two equations, together with Equations (4) and (5) give a set of equations in which the membranes are coupled to the gas pressure in the drums. The pressures are now coupled to one another using the Poiseuille flow equation. This equation determines the rate of pressure induced flow of a viscous fluid across a channel. In the pump system, this fluid is the nitrogen gas in the cavity. The Poiseuille flow equation describes both n1 and n2 through the following differential equation: 7 Φ = (1 − 0.63 g0 w )(∆P1 − ∆P2)g3 0w 12ηl = dn2 dt V2 n2 = dn2 dt ¯RT P + ∆P2 = − dn1 dt ¯RT P + ∆P1 . (8) In this equation Φ is the volumetric flux of gas through the channel, ¯R is the ideal gas constant, and η is the dynamic viscosity of the gas. The Poiseuille flow equation acts as the coupling between the two cavities. In using this equation to express the change in the amount of gas molecules in the cavities we have implicitly added the condition that the total amount of gas molecules in the pump system is conserved, which holds assuming no gas permeation outside the cavities. In order to incorporate Equation (8) into the model, the time derivatives of the ideal gas laws are taken: dz1 dt dz2 dt Filling in Equation (8) results in the following set of equations: V1 + (P + ∆P1)A V2 + (P + ∆P2)A d∆P1 dt d∆P2 dt = dn1 dt = dn2 dt ¯RT; ¯RT . (9) (10) (11) (12) ; . dz1 dt d∆P1 dt + (1 − 0.63 g0 w d∆P2 dt + (1 − 0.63 g0 w = −(P + ∆P1) g0 + z1 )(∆P2 − ∆P1)(P + ∆P1)g3 12ηlA(g0 + z1) dz2 dt = −(P + ∆P2) g0 + z2 )(∆P1 − ∆P2)(P + ∆P2)g3 12ηlA(g0 + z2) 0w 0w    To neatly express the model of the two drum system, the four differential equations that describe the system are given together. The following set of equations describe the two drum system: ∆P1 − V 2ε0A 2m(g0 + z1)2 ; z1 + A m z2 + A m ∆P2 ; dz1 dt m m d2z1 dt2 = − k d2z2 dt2 = − k d∆P1 dt + (1 − 0.63 g0 w d∆P2 dt + (1 − 0.63 g0 w = −(P + ∆P1) g0 + z1 )(∆P2 − ∆P1)(P + ∆P1)g3 12ηlA(g0 + z1) dz2 dt = −(P + ∆P2) g0 + z2 )(∆P1 − ∆P2)(P + ∆P2)g3 12ηlA(g0 + z2) 0w 0w (13) (14) (15) (16) ; . Quasi-static equations The quasi-static limit of these equations is taken. In this case the second derivatives in Equations (4) and (5) are negligible. Newton's law is now equivalent to a force balance, indicating that at all times the drums are at an equilibrium position. This equilibrium position changes in time due to the changing gas pressure and voltage. As such there is still a response to the driving force. In order to find approximate solutions to the differential equations, Equations (13) and (14) are linearised. The force balance that is found is: Now filling the force balance into this differential equation eliminates all displacement terms and yields the following kz1 = A∆P1 − V 2ε0A kz2 = A∆P2 . 2g2 0 ; = − P g0 = − P g0 dz1 dt dz2 dt + (1 − 0.63 g0 w + (1 − 0.63 g0 w )(∆P2 − ∆P1)P g2 )(∆P1 − ∆P2)P g2 12ηlA 12ηlA 0w 0w 8 (17) (18) (19) (20) ; . (1 + P A kg0 dV 2 ε0 2g2 dt 0 (1 + P A kg0 ) = (1 − 0.63 g0 w ; ) = (1 − 0.63 g0 w )(∆P2 − ∆P1)P g2 0w 12ηlA )(∆P1 − ∆P2)P g2 0w 12ηlA . (21) (22) The linearised differential equations for the pressure are: d∆P1 dt d∆P2 dt  differential equations for the pressure: d∆P1 dt + P A kg0 d∆P2 dt For simplicity, we define the following constants: a = P A kg0 ; b = (1 − 0.63 g0 w ) P g2 0w 12ηlA . Here a is the squeeze number and b is related to the gas flow dynamics through the channel. This allows us to put the differential equations into the following simple form: (cid:21) (cid:20)∆P1 ∆P2 d dt = −b 1 + a (cid:20) 1 −1 (cid:21) (cid:21)(cid:20)∆P1 ∆P2 −1 1 + c d dt (cid:21) (cid:20)V 2 0 ; c = a 1 + a ε0 2g2 0 . Frequency spectrum of the system (23) (24) In order to investigate the behaviour of this differential equation, a Fourier transform of the differential equations is taken: (cid:20)F(∆P1) (cid:21) F(∆P2) iω = −b 1 + a (cid:20) 1 −1 (cid:21)(cid:20)F(∆P1) (cid:21) F(∆P2) −1 1 (cid:20)F(V 2) (cid:21) 0 + ciω The frequency spectra found from these equations are given by F(∆P1) = 1 2 + iωτ 1 + iωτ cF(V 2) ; . (25) (26) F(∆P2) = 1 2 1 + iωτ cF(V 2) , 9 (27) 2b . This time also defines the cutoff frequency of the gas pump system ω0 = 2b with τ = 1+a 1+a. The function Fourier transform of ∆P2 takes the form of a low pass filter. We are more interested in the Fourier transform of P1, whose magnitude and phase of F(∆P1) are given below. F(∆P1) = 1 2 r1 + ω24τ 2 1 + ω2τ 2 cF(V 2) ; (cid:17) φ = Arg[F(∆P1)] = arctan(cid:16) ωτ 1 + ω22τ 2 . (28) (29) Finally, we examine the behaviour of the drum displacement. The algebraic equations found for the displacement allow us to directly calculate the Fourier transform of the displacement from the pressures and the square of the electrostatic potential. Once more it can be seen that F(z2) is given by applying a low pass filter on the driving force. The magnitude and phase of F(z1) and find: F(z1) = −(cid:16)1 (cid:17) A k 1 1 + iωτ + 1 a 2 cF(V 2) ; F(z2) = 1 2 1 1 + iωτ A k cF(V 2) . r 1 a2 +(cid:16)1 a (cid:17) F(z1) = φ1 = Arg(F(z1)) = π − arctan(cid:16) + 1 4 1 1 + ω2τ 2 cF(V 2) ; A k (cid:17) . aωτ 2 + a + 2ω2τ 2 (30) (31) (32) (33) We now use these equations to model the curves from Fig. 5. II. Measurement in vacuum 10 A comparison of "12" measurements in vacuum and in N2 is shown in Fig. 2. In ambient pressure, the motion of pump 2 responds to the actuation of pump 1, mediated by the gas in the chamber. The absence of motion of pump 2 in vacuum (orange curve in Fig. 2) is another confirmation of pneumatic actuation in the system. FIG. S2. Measurement in vacuum. Measurement of the motion of pump 2 when actuating pump 1 at atmospheric pressure (blue curve) and in vacuum (orange curve). VPD (mV)12 (1 bar N2)12 (0.1 mbar N2)Time(ms)024681012VACT (V)1010.51.50-0.5-1
1709.07926	2	1709	2018-05-24T00:08:37	Elastic wave cloaking via symmetrized transformation media	[ "physics.app-ph", "physics.class-ph" ]	Transformation media theory, which steers waves in solids via an effective geometry induced by a refractive material (Fermat's principle of least action), provides a means of controlling vibrations and elastic waves beyond the traditional dissipative structures regime. In particular, it could be used to create an elastic wave cloak, shielding an interior region against elastic waves while simultaneously preventing scattering in the outside domain. However, as a true elastic wave cloak would generally require nonphysical materials with stiffness tensors lacking the minor symmetry (implying asymmetric stress), the utility of such an elastic wave cloak has thus far been limited. Here we develop a means of overcoming this limitation via the development of a symmetrized elastic cloak, sacrificing some of the performance of the perfect cloak for the sake of restoring the minor symmetry. We test the performance of the symmetrized elastic cloak for shielding a tunnel against seismic waves, showing that it can be used to reduce the average displacement within the tunnel by an order of magnitude (and reduce energy by two orders of magnitude) for waves above a critical frequency of the cloak. This critical frequency, which corresponds to the generation of surface waves at the cloak-interior interface, can be used to develop a simple heuristic model of the symmetrized elastic cloak's performance for a generic problem.	physics.app-ph	physics	Seismic Invisibility: Elastic wave cloaking via symmetrized transformation media Sophia R. Sklan,1 Ronald Y.S. Pak,2 and Baowen Li1 1Department of Mechanical Engineering, University of Colorado Boulder, Colorado 80309 USA 2Department of Civil Engineering, University of Colorado Boulder, Colorado 80309 USA Transformation media theory, which steers waves in solids via an effective geometry induced by a refractive material (Fermat's principle of least action), provides a means of controlling vibrations and elastic waves be- yond the traditional dissipative structures regime. In particular, it could be used to create an elastic wave cloak, shielding an interior region against elastic waves while simultaneously preventing scattering in the outside do- main. However, as a true elastic wave cloak would generally require materials with stiffness tensors lacking the minor symmetry (implying asymmetric stress), the utility of such an elastic wave cloak has thus far been limited by the challenge of fabricating these materials. Here we develop a means of overcoming this limitation via the development of a symmetrized elastic cloak, sacrificing some of the performance of the perfect cloak for the sake of restoring the minor symmetry. We test the performance of the symmetrized elastic cloak for shielding a tunnel against seismic waves, showing that it can be used to reduce the average displacement within the tunnel by an order of magnitude (and reduce energy by two orders of magnitude) for waves above a critical frequency of the cloak. This critical frequency, which corresponds to the generation of surface waves at the cloak-interior interface, can be used to develop a simple heuristic model of the symmetrized elastic cloak's performance for a generic problem. The need to protect objects against unwanted mechanical vibration and wave incidence is a long-standing subject in engineering [1, 2]. The problem arises in multiple scenar- ios including blocking sound/elastic waves and eliminating unwanted mechanical resonances [1, 2] (vibration isolation) or guarding against nonlinear mechanical shock waves and preventing the collapse of structures under incident seismic waves [3, 4] (earthquake engineering). Techniques to ac- complish these goals have included strengthening structures, modifying structural resonances through additive engineer- ing or anti-resonance, developing flexible structures that can withstand large deformations, or including dissipative ele- ments [1 -- 4]. Recently, techniques from phononics and acous- tic wave engineering have been adapted to vibration isola- tion/earthquake engineering (VIEE), most prominently by the inclusion of phononic/sonic crystals as a means of shield- ing against seismic surface waves [1, 4 -- 11] (which contain a much smaller portion of energy from an earthquake than bulk waves, particularly shear waves [12]). However, one limitation of all these techniques is that they can only seek to mitigate elastic waves, reducing the local amplitude of the earthquake or allowing a structure to withstand an unmodified vibration. In addition, many of these techniques have focused upon narrow frequency bands (e.g. modifying resonances), particularly for low frequency applications. While these are primary concerns, minimization of broadband or higher fre- quency transmissions to a structure is of increasing engineer- ing interest nowadays as the reliance on electronic-computer control of critical equipment grows. An alternative framework to VIEE would be cloaking. Cloaks modify the environment around a region such that waves are refracted around a central domain [13 -- 18]. Since the energy is redirected, not dissipated, it could in principle be used to control vibrations of arbitrary amplitude or fre- quency (in practice, engineering limitations prevent perfect performance). Developing effective, realizable cloaks is an active field of research in optics and acoustics, but application to elastic waves and VIEE has remained in its infancy [19 -- 25]. The central challenge limiting the utilization of cloak- ing for these applications is that transformation media the- ory, the mathematical framework underlying the operation of the cloak, is not feasible for a generic elastic wave. As was shown by Milton et al., a cloaking transformation for an elas- tic wave would break the stiffness tensor's minor symmetry (cijkl = cjikl = cijlk) [26]. Since this symmetry exists for all commonplace solid materials, this has limited elastic wave cloaks to special cases where the loss of minor symmetry is irrelevant (e.g. planar structures, partial coordinate transfor- mations) or structures with significant fabrication challenges (e.g. Cosserat materials and auxetic/pentamode materials, hy- perelastic metamaterials, Willis materials) [27 -- 37]. Hypere- lastic metamaterials [38] deserve special mention as a system where the nonlinear pre-stress of the system does allow for the more ready construction of materials lacking minor elastic symmetry. However, fabrication challenges and the need for strong nonlinearity still limit the widespread application of the technique. Similarly, Willis materials [26] provide an explicit solution where the loss of minor symmetry is preserved, but at the cost of introducing highly dispersive materials which dis- tort the waves and thus limit the utility of broadband cloaking. Moreover, the incorporation of Willis materials at a theoreti- cal level has not noticeable reduced the fabrication challenges of constructing an elastic wave cloak. In this work we present a framework for an approximate elastic cloak, which preserves minor symmetries even in the most general case. This symmetrized elastic cloak (SEC) has the advantage of being, in principle, more readily realizable, but comes at the cost of no longer being a perfect cloak. How- ever, a cloak generally performs two tasks: preventing scatter- ing of an incoming wave as it propagates around and through the cloak (stealth, the primary concern in optical or acous- tic cloaking) and blocking waves from penetrating a central region (shielding, the primary concern in VIEE). The per- formance of an approximate cloak for both these tasks is an open question, but only the latter performance metric is im- portant for VIEE applications. As such, we characterize the performance of the cloak in the simplest physically realistic scenario − shielding a tunnel or a round shell buried in a soil- type medium from seismic waves. Through this analysis, we derive the limitations of the SEC and present a simple holistic model that captures the essential performance characteristics. To begin, the equations of motion in Cartesian coordinates for elastic waves in a solid are ρ∂ttui = ∂jσji σij = cijklkl ij = 1 2 (∂jui + ∂iuj), (1) (2) (3) (where ρ is density, u is the displacement vector, is the in- finitesimal strain tensor, σ is the Cauchy stress tensor, and c is the 4th order stiffness tensor). In cylindrical coordinates they become ρ∂ttur = ρ∂ttuθ = ρ∂ttuz = 1 r 1 r 1 r ∂rrσrr + ∂rrσrθ + ∂rrσrz + 1 r 1 r 1 r (∂θσθr − σθθ) + ∂zσzr (∂θσθθ + σθr) + ∂zσzθ (4) ∂θσθz + ∂zσzz, while σij = cijklij is unchanged except for a relabeling of coordinates, and  ∂rur er er ∂ruθ eθ er ∂ruz ez er = = 1 r (∂θur − uθ)er eθ ∂zur er ez 1 r (∂θuθ + ur)eθ eθ ∂zuθ eθ ez ∂zuz ez ez 1 r ∂θuz ez eθ   r r−a crrθr b b−a crrzr  crrrr  cθrrr  czrrr r r−a cθrθr b b−a cθrzr r r−a czrθr b b−a czrzr crrrθ r r−a crrθθ b b−a crrzθ cθrrθ r r−a cθrθθ b b−a cθrzθ czrrθ r r−a czrθθ b b−a czrzθ crrrz r r−a crrθz b b−a crrzz cθrrz r r−a cθrθz b b−a cθrzz czrrz r r−a czrθz b b−a czrzz c(cid:48) ijkl = r r−a crθθr b b−a crθzr   crθrr   cθθrr   czθrr r r−a cθθθr b b−a cθθzr r r−a czθθr b b−a czθzr 2 where we've dropped the explicit symmetry of and σ by requiring it be preserved in c. In general, c must possess both the major symmetry cijkl = cklij (6) (which comes from the symmetry of mixed partials) and the minor symmetry cijkl = cjikl = cijlk = cjilk (7) (which comes from the physical symmetry of stress and strain). The standard cylindrical cloaking transformation [13] is the dilation r(cid:48) = b − a b r + a (8) where r(cid:48) is the transformed radial coordinate, a is the inner ra- dius of the cloak, and b is the outer radius of the cloak. Under the cloaking transformation, we seek a set of c(cid:48) and ρ(cid:48) such that the equation of motion in the standard frame (i.e. equa- tion (4)) with these materials is the same as the equation of motion in the transformed frame (equation (4) with all factors of r replaced by r(cid:48) but (cid:126)u(cid:48) = (cid:126)u and σ(cid:48) (cid:54)= σ) with some trivial background material. Breaking the minor symmetry, this can be accomplished exactly using a density and stiffness tensor (cid:19)2 (cid:18) b b − a ρ(cid:48) = ρ0 r − a r (5)  r−a r 1 crθrθ r r−a crθθθ b b−a crθzθ cθθrθ r r−a cθθθθ b b−a cθθzθ czθrθ r r−a czθθθ b b−a czθzθ ×   crzrr   cθzrr   czzrr r r−a crzθr b b−a crzzr r r−a cθzθr b b−a cθzzr r r−a czzθr b b−a czzzr b b−a r−a r crθrz r r−a crθθz b b−a crθzz cθθrz r r−a cθθθz b b−a cθθzz czθrz r r−a czθθz b b−a czθzz crzrθ r r−a crzθθ b b−a crzzθ cθzrθ r r−a cθzθθ b b−a cθzzθ czzrθ r r−a czzθθ b b−a czzzθ crzrz r r−a crzθz b b−a crzzz cθzrz r r−a cθzθz b b−a cθzzz czzrz r r−a czzθz b b−a czzzz (9)     . (10) Note that this preserves major symmetry of c and that the transformation of cijkl is effectively independent of two in- dices (either i or j and either k or l). Because equation (10) has lost minor symmetry, however, it is no longer easily realizable. To restore this symmetry, we impose a symmetrization function (11) where S is an arbitrary function that preserves minor symme- cS ijkl = S(cijkl, cjikl, cijlk, cijilk) 3 try. (Note that a symmetrization technique was also applied in Ref. [39], but as the focus there was on the scattering field and the focus here is on shielding, their results and analysis differ from ours. In particular, when Ref. [39] considers cloaking, they measure the perfomance of the cloak with respect to the external scattered field. That is, their efficiency measure is the normalized r.m.s. deviation of the external field with respect to a homogeneous background. Any effect on the fields in the cloaked region is explicitly excluded from their analysis, fol- lowing Ref. [40]. This differs from our work, which considers the performance of the cloak in shielding an internal structure from an external source. As such, we focus on the deviation of the internal field with respect to an uncloaked object and ignore any effects on the external, scattered field. Thus, the focus of our work is complementary to Ref. [39] and the con- clusions we draw on the efficacy of symmetrized cloaks to shield vibrations does not conflict with the conclusions they draw on the efficacy of symmetrized cloaks to prevent scat- tering (i.e. that an anisotropic density tensor is a necessary condition to create a cloak that effectively reduces scattering while incorporating a symmetrized elasticity tensor).) For cS's simplicity, we select the geometric mean: SGM (x, y, z, w) = (xyzw)1/4 (12) for our SEC. (In principle some other symmetrization function could improve the performance of the cloak under some crite- ria, although it is intuitive that some average function is opti- mal.) With this symmetry imposed, our stiffness tensor is re- duced to the 21 components (in Voigt notation, rr = 1, θθ = 2, zz = 3, θz = 4, θr = 5, rθ = 6)  cS IJ = r−a r c11 c12 r r−a c22 r−a b r c13 b−a b b−a c23 b−a )2 r−a ( b r r−a r )1/2c14 r−a )1/2c24 ( b b−a ( b b−a b−a )3/2( r−a b b−a c44 r c33 ( b r )1/2c34 S Y M r c15 b−a )1/2 r−a ( b ( b b−a )1/2c25 b−a )3/2 r−a ( b r c35 b−a )1/2( r−a ( b r )1/4c45 r−a r c55 b b−a  ( r−a r )1/2c16 ( r r−a )1/2c26 b−a ( r−a b r )1/2c36 ( b b−a )1/4c46 r−a ( b r )1/2c56 b−a c66 (the lower half of the tensor omitted by symmetry). For a cloak embedded within an isotropic background, we can im- pose a further simplification to  cS(λ, µ) = r−a r (λ + 2µ) λ r−a r λ b b−a λ r r−a (λ + 2µ) b b−a λ r−a r λ b b−a b b−a λ b−a )2 r−a ( b r (λ + 2µ)  . b b−a µ b b−a r−a r µ µ (13) (14) Notably, this implies that the stiffness tensor for a cloak in an isotropic material should possess orthotropic symmetry. While this mathematical framework for the SEC is consis- tent − it is a set of material properties which mimic the perfect cloak in some respects but maintain the symmetries of com- mon materials − its utility is a separate matter. After all, when our material properties deviate from the perfect cloak its per- formance will degrade. To test the SEC's performance, it is necessary to impose a test case and a metric. Since our goal for the SEC is VIEE applications, we shall test it by special- izing to seismic waves and considering the highest symmetry case, namely protecting a hollow core concrete tunnel (con- crete: ρ = 2300kg/m3, µ = 9.34GPa, λ = 9.12GPa, air: ρ = 1.225kg/m3, µ = 0Pa, λ = 142kPa) buried in uni- form soil (ρ = 2203kg/m3, µ = 30MPa, λ = 120MPa) with a flat surface. For waves incident orthogonal to the length of the tunnel, then, we can approximate the system as two- dimensional (see Fig. 1). We consider a tunnel of radius a =10m with a hollow center of 5m radius (in practice, this concrete shell is likely thicker than a real tunnel, but was utilized for numerical stability) and an inhomogeneous cloak obeying equations (9), (10) with radius b =20m (this implies an SEC thickness of 10m, which is likely much thicker than required in practice − the thickness of the cloak being chiefly constrained by the feature size of the metamaterials used in its fabrication). This tunnel was placed in the center of a rect- angular domain (100m×300m) of soil. For boundary condi- tions, the top surface was left traction-free (σ · n = 0), with the sides left as impedance matched absorbing layers. (Note that a true impedance matching requires breaking the minor symmetry [39]. Instead, we only use the pre-defined "low re- flecting boundary" setting of COMSOL [41], which imposes a boundary that only approximately impedance matches ev- ery polarization and will introduce some amount of reflection. The simulation domain's boundaries were set to minimize the impact of these reflections on the performance of the SEC.) From the base, a vibration of ui = ui0 sin(2πωt)ei was im- posed, with ui0 =1cm and i = x, y for S or P waves respec- tively. Frequencies of the imposed waves were allowed to vary between 1 and 10 Hz, as those are the most relevant frequen- cies for typical earthquakes [42]. In addition, a second set of boundary conditions are considered where waves approach parallel to the surface, i.e. the displacement is imposed upon one of the sides and the base is left impedance matched. This second set of boundary conditions was used to account for any effects induced by the anisotropy of the boundary conditions; the operation of the cloak in infinite homogeneous space is expected to retain isotropy. To measure the efficiency of the cloak, we employ two metrics: reduction in average energy and reduction in average displacement. These are character- ized as efficiencies, and η(E) = 1 − (cid:104)E(cid:105)C/(cid:104)E(cid:105)U η(ui) = 1 − (cid:104)ui(cid:105)C/(cid:104)ui(cid:105)U (15) (16) where brackets denote averaging over the concrete shell and averaging over time, E is total energy, U refers to results with no cloak, and C refers to results with the cloak. Given that we consider two directions of incident waves (other directions can be decomposed into a linear superposition of these orthogonal cases), we define u(cid:107) as the response with polarization paral- lel to the driving force and u⊥ as the transverse response (in general u(cid:107) (cid:29) u⊥, η(u⊥), η(vi), η(ai) in supplement). Note that η ≤ 1 for a passive cloak, with η = 1 being perfect cloaking (100% efficiency), η = 0 being no improvement, and η < 0 being a degradation of performance due to the cloak. With this setup, we use COMSOL [41] to simulate the tunnel with and without the cloak for varying boundary condi- tions ((cid:126)u = 0 initial conditions, simulation interval of 180/ω). Simulations used a mesh grid of 9924 domain elements and 388 bounary elements (i.e. an "extremely fine mesh" setting), which showed a strong convergence and included a time res- olution of 1/20ω. Numerical integration was performed us- ing the pre-built COMSOL finite element differential equa- tion solver using MUMPS at COMSOL's default settings and displacements explicitly constrained to real values. Examining the performance of the SEC under this effi- ciency metric, we can separate out two different regimes for 4 FIG. 1: Schematic model of SEC tunnel shield simulation setup. The cloak and tunnel are placed at the center of simulation domain, with boundary conditions labeled at each end. (Inset) Material structure of the simulation: Soil (light brown) surrounds the SEC (dark brown ring). Inside the SEC is a concrete (grey ring) wall filled with air (white circle). each polarization. For S waves (see Figure 2A-B), the per- formance is relatively insensitive to the direction of the inci- dent waves, but shows a clear change in performance above and below approximately 1Hz≡ ωc,S. For ω (cid:46)1Hz, the SEC clearly reduces the performance of the concrete in blocking in- coming waves, reaching its minimum performance at ω ≈1Hz (η(u(cid:107),S) (cid:38) −2.00, η(ES) (cid:38) −8.11). On the other hand, for ω (cid:38)1Hz the SEC shows great im- provement in performance; the concrete shell is protected from incoming waves by an order of magnitude for displace- ment and two orders of magnitude for energy (η(u(cid:107),S) (cid:46) 0.849, η(ES) (cid:46) 0.988). P waves (see Figure 2C-D) are similar to S waves but the critical frequency separat- ing these regimes is shifted to 2Hz≡ ωc,P (η(u(cid:107),P ) (cid:46) 0.913, η(EP ) (cid:46) 0.993) due to the difference in speeds vP /vS = 6 ≈2.4 (where vi is the velocity, (cid:112)µ/ρ for S waves and(cid:112)(λ + 2µ)/ρ for P waves). Note that, for ω under √ the critical frequency, some of the simulations demonstrated small oscillations between positive and negative efficiency. However, as simulations far below the critical frequency pos- sess wavelengths on the order of kilometers or longer, and the cloak is on the order of meters, we expect that these effects are highly sensitive to boundary conditions and may be numerical artifacts. To explain the difference in performance above and below these critical frequencies, we examine the energy difference between the cloaked and uncloaked cases as a function of time for different incident frequencies (see Figure 3). Below the critical frequency (Figure 3A), we see a very clear buildup of energy near the inner surface of the cloak. While such an energy buildup exists for frequencies above the critical fre- quency (Figure 3B), it remains relatively localized in those cases. At or below the critical frequency, though, the surface energy concentration extends across nearly the entire length of the inner boundary, meaning that an effectively uniform σ•n=0AbsorbingDisplacement or AbsorbingAbsorbing orDisplacementSoilSECCementAir5 FIG. 3: Time domain simulations of SEC for VI S waves. Surface plot is energy difference EC − EU , i.e. positive values (reds) de- note lower efficiency and negative values (blues) denote higher ef- ficiency. Field generated by post-processing the displacement field u(x, t) from dynamical simulations in COMSOL. Simulations were paired, with EC and EU generated from different instantiations using identical boundary and initial conditions but with the SEC present or absent respectively. This allowed for the generation of the scattered energy field without any explicit scattering field methodology. (A) Simulation at resonance, ω =1Hz. Note buildup of energy within the concrete tunnel (inner annulus) and the surface wave surround- ing it. (B) Simulation above resonance, ω =8Hz. Note the surface wave is localized and does not surround the tunnel. View is zoomed out compared to (A) to display scattering field variation. field surrounds the interior domain. Since the elastic wave equations depend upon a Laplacian, they obey the mean value theorem, implying that such a uniform energy buildup is able to penetrate through any cloak, even a perfect one [43 -- 46]. However, for a perfect cloak, the penetration would merely imply η = 0, the presence of a negative efficiency implies that our SEC is still underperforming. In particular, a per- fect cloak would not produce the energy buildup that we ob- FIG. 2: Efficiency plots for the SEC as a function of frequency. Black dashed curve is data for vertical incidence (VI, wave from be- neath the SEC), grey dashed curve is data for horizontal incidence (HI, wave from the left side of the SEC), while solid blue and red curves are fitted harmonic oscillator models of the SEC for VI and HI respectively. Fits are generated by matching the data generated through direct COMSOL simulations of the dynamics with a sim- ple harmonic oscillator model given in equation (18). (A) Energy efficiency, S wave. Efficiency at the cutoff frequency of 1Hz (value given in main text) cropped to distinguish higher frequency variation. (B) Longitudinal displacement efficiency, S wave. (C) Energy effi- ciency, P wave. (D) Longitudinal Displacement efficiency, P wave. Note that fitted curves coincide in (C) and (D) due limited degrees of freedom in the harmonic oscillator model of equation (18). 12345678910−2−1.5−1−0.500.51ω [Hz]η(E) [1]SEC efficiency, S wave EnergyVertical Incidence, dataVertical Incidence, fitHorizontal Incidence, dataHorizontal Incidence, fitA)12345678910−1.5−1−0.500.51ω [Hz]η(u) [1]SEC efficiency, S wave DisplacementVI, dataVI, fitHI, dataHI, fitB)12345678910−3−2.5−2−1.5−1−0.500.51ω [Hz]η(E) [1]SEC efficiency, P wave EnergyVI, dataVI, fitHI, dataHI, fitC)12345678910−1−0.500.51ω [Hz]η(u) [1]SEC efficiency, P wave DisplacementVI, data VI, fitHI, dataHI, fitD)T=48.1s, ω=1Hz, Surface: E(Cloak)-E(Uncloak) (J/m3)Α)Time=7.3375s, ω=8Hz, Surface: E(Cloak)-E(Uncloak) (J/m3)Β)served in Figure 3. As has been observed in other approxi- mate cloaks [19, 22, 35], imperfections can induce resonant modes within the cloak. If we assume that the observed sur- face modes are resonantly excited by the SEC, we can predict a simple model for how the cloak should effect the response, using the standard 1D simple harmonic oscillator (SHO) dy- namical response (cid:112)(ω2 − ω2 F/m u = c )2 + ω2Γ2 (17) (18) where F/m is the effective acceleration and Γ is the damping rate. This would predict an efficiency of ηSHO(Ω, Q) = 1 − (cid:112)(Ω2 − 1)2 + Ω2/4Q2 1 where Ω = ω/ωc is the dimensionless frequency and Q = ωc/2Γ is the resonator quality factor. This predicts an effi- ciency of 0 for Ω = 0, 2 − 1/4Q2, a minimum of η = 1 − 2Q for Ω = 1, and η → 1 for Ω → ∞, which is qualitatively sim- ilar to our calculated efficiencies for η(E) (see solid curves in Figure 2). For comparison, a perfect cloak efficiency should resemble a step function, jumping from 0 to 1 at Ω = 1. While this harmonic oscillator model predicts the qualita- tive aspects of the SEC's performance, the exact value of ωc is not captured by that model. Since that determines the low frequency cutoff where the SEC's performance dramatically worsens, ωc is critical to determining where the SEC is ap- plicable and how it could be improved. If we substitute the equations (9) and (14) into equation (4) and apply some sim- ple mathematical operations (see supplement), we can derive an implicit series solution for u. Notably, this implicit solution possesses a set of dimensionless parameters ω2a2( b b − a )2 1 v2 i . (19) Plugging in the parameters used in our simulations gives ωc,S =0.9286Hz and ωc,P =2.275Hz, which closely match our numerical results. Notice too that b − a 1 a , b vi ωc,i = (20) which is equivalent to a wave with velocity vi(b − a)/b and wavelength 2πa, i.e. a surface wave with wavelength equal to the circumference of the cloaked region. Importantly, equa- tion (20) implies that ωc ∝ 1/a, so increasing the size of the cloaked domain will lower the cutoff frequency. To calculate Q from first principles is a less trivial problem, so instead we treat it as a fitting parameter in our model and fit the efficiency characteristics of our SEC to get (cid:104)QS(cid:105) = 1.57,(cid:104)QP(cid:105) = 3.89. In principle, for a real SEC, its performance could be mea- sured by finding the resonant frequency and performance of the SEC at that frequency. However, since no real material possesses the inhomogeneities of equations (9), (14), a real SEC would likely be constructed from constituent elements 6 that approximate this (as in the standard cloak design [15]). As such, it should possess an upper cutoff frequency with wavelength of approximately the size of constituent element. We therefore expect an SEC to be effective for blocking fre- quencies between ωc ∝ 1/a < ω < ωMM ∝ 1/L, where ωMM is the characteristic frequency for a metamaterial of characteristic length L. The design of seismic metamateri- als is a relatively new field [47 -- 49] but shows a great deal of progress, and when combined with techniques like additive manufacturing [50] it suggests the feasibility of realizing an SEC. Additionally, while the coordinate transformation used here necessitates the engineering of both stiffness and the den- sity of the metamaterials, more complicated transformations exist that simplify the material requirements. In particular, Ref. [51] would result in a design that only required engineer- ing the elasticity and would leave the density unchanged (i.e. uniformly equal to the surrounding medium's density). To determine the impact of this upper frequency cutoff and the feasibility of the SEC under more realistic conditions, we repeat our calculations using a discretized SEC. As is the stan- dard procedure in cloaking, we replace our inhomogeneous SEC given by equations (9) and (14) with a series of annular rings. Each ring's density and elasticity tensor are selected to be the average value of these equations in their respective region. In particular, we use 10 rings, each 1 meter thick, of uniform orthotropic material given by the mean value of the SEC in their 1 meter shells. Although one meter is still quite thick, a thinner discretization scheme was avoided to ensure that the effects of discretization weren't disguised by an unre- alistic number of layers and that the size of the rings was not significantly finer than the mesh grid used in our finite element simulations. On the other hand, even a single layer would be too thin to display the predicted cutoff due to wave scattering on the discretized structure. A 10 meter thick shell (i.e. a uni- form layer of thickness equal to our SEC) would have an upper cutoff frequency of approximately 73 Hz, far above the 10 Hz practical limit that we employ (a shell engineered to be thin enough to be practical, then, would likely have an upper cut- off frequency in the O(100) Hz range). In this regime, then, we expect that the principle limitation to the discretized SEC's performance is not the geometric effect of the shell thickness but simply the deviation of the material properties from con- tinuous limit. As 10 layers typically shows good agreement with the continuous limit in experimental tests of metamate- rial cloaks, we thus consider this case as the rigorous practical test of a realizable SEC. Given the uniformity of the contin- uous SEC's performance under different angles of incidence, polarizations, and efficiency measures, we focus on the en- ergy shielding of the discretized SEC for bulk S and P waves under vertical incidence. We find in Fig. 4 that the behavior is in good agreement with the continuous SEC case, even as the discretization removes the singular behavior of the cloak at the inner boundary. In particular, we find clear agreement in the location of the lower cutoff frequency, qualitative wave dynamics, quantitative trend, and maximum efficiency. More- over, we do not observe any decrease in performance due to 7 blocking seismic waves of any relevant frequency for a large structure. Alternatively, a smaller SEC could be used specifically to focus upon protecting equipment and critical infrastruc- ture against high frequency elastic waves (5-20+ Hz). While these high frequency waves are often considered negligible in earthquake engineering due to their stronger dissipation, they can be dangerous in certain situations such as the sur- vival of control equipment and circuits. Geological condi- tions can lead to larger content of higher frequencies [52]. Smaller structures can have their own higher resonant fre- quencies as well. Since this can include internal resonances of walls, excitation of strong ground motion in soil, or the failure of critical infrastructure like water pipes, generators, etc. [42, 53], the potential impact of high frequency waves can be immense and catastrophic. Furthermore, as the failure modes of the SEC are focused around a critical frequency cor- responding to surface elastic waves on the inner boundary of the SEC, the inclusion of damping or resonance shifting tech- niques practices in VIEE [1] could be used to create a hybrid SEC/traditional earthquake abatement system with improved performance. Moreover, as cloak designs have been devel- oped for various geometries [13 -- 16] and could in principle be adapted to arbitrary geometric configurations, it is likely that this SEC approach could find ready application in a variety of critical structures. ACKNOWLEDGEMENTS We would like to thank Prof. Kathryn Matlack and Prof. Fatemeh Pourahmadian for their helpful discussion and sug- gestions. SUPPLEMENTARY MATERIALS Velocity and Acceleration Performance The SEC's efficiency for shielding against velocity or accel- eration transmission closely matches its efficiency for shield- ing against displacement. This makes intuitive sense, as these quantities are expected to differ by factors of ω, which can- cel when we take the ratio of cloaked and uncloaked fields to calculate the efficiency. Transverse Component Efficiency While the overall trends for the efficiencies of the SEC for transverse components are similar to the longitudinal re- sponse of the main text, some deviations exist. However, these can be explained as arising from the relative magnitudes of the transverse and longitudinal components of the scattered waves, as only a tiny portion of the energy is relegated to these transverse modes, small fluctuations or numerical artifacts can FIG. 4: Comparison of efficiency for the continuous (dotted grey line) and discrete (solid blue line) implementation of the SEC as a funcion of frequency. All plots are for vertical incidence. (A) Energy efficiency, S wave. (B) Energy efficiency, P wave. the discretization, even at high frequencies, as the frequency range of interest is far lower than the upper cutoff frequency. We thus conclude that an SEC made from ordinary orthotropic material with tuned properties is an effective VIEE structure with much less limiting fabrication constriants. To summarize, we have developed a method of modulating the challenge to realize asymmetries of a perfect elastic wave cloak into a more readily realized symmetric elastic cloak. By testing the performance of the cloak as a shield against seis- mic waves in a tunnel, we have demonstrated the feasibility of this design for VIEE, reducing the displacement within the cloaked tunnel by an order of magnitude for most of the fre- quencies corresponding to common seismic wave resonances (1-10Hz, [42]) (compared to a tunnel with no cloak). Since the frequency range where our approximate SEC fails is de- termined by the size of the cloaked region and the characteris- tic length of the SEC, switching to shield even larger regions (a (cid:38)100m) and using sufficiently small building blocks for the SEC (L (cid:46)1m) should render this design quite effective for 12345678910 ω [Hz]-8-7-6-5-4-3-2-101η(E) [1]SEC Efficiency, Vertically Incident S waveDiscreteContinuousA)123456789 ω [Hz]-3-2.5-2-1.5-1-0.500.51 η(E) [1]SEC Efficiency, Vertially Incident P waveDiscreteContinuousB)8 FIG. 6: Transverse component, S wave efficiency plots for the SEC as a function of frequency. Red dashed curve is data for vertical incidence (VI, wave from beneath the SEC), blue dashed curve is data for horizontal incidence (HI, wave from the left side of the SEC). (A) Displacement, (B) Velocity, (C) Acceleration. have a large impact upon the calculated efficiency. Thus, these deviations from the longitudinal response are likely insignifi- cant. FIG. 5: Longitudinal component efficiency plots for the SEC as a function of frequency. Red dashed curve is data for vertical inci- dence (VI, wave from beneath the SEC), blue dashed curve is data for horizontal incidence (HI, wave from the left side of the SEC). (A) Velocity efficiency, S wave. (B) Acceleration efficiency, S wave. (C) Velocity efficiency, P wave. (D) Acceleration efficiency, P wave. Dimensionless Constants To find the dimensionless constants of the SEC we plug the parameters of equations (9) and (14) into equation (4). We assume that variation along z is negligible and thus specialize to polar coordinates, assuming a solution to the equations of 012345678910−2.5−2−1.5−1−0.500.51ω [Hz]η(v) [1]SEC efficiency, S wave VelocityVIHIA)012345678910−2.5−2−1.5−1−0.500.51ω [Hz]η(a) [1]SEC efficiency, S wave AccelerationVIHIB)12345678910−1.5−1−0.500.51ω [Hz]η(v) [1]SEC efficiency, P wave VelocityVIHIC)12345678910−1.5−1−0.500.51ω [Hz]η(a) [1]SEC efficiency, P wave AccelerationVIHID)012345678910−1.5−1−0.500.51ω [Hz]η(u⊥) [1]SEC efficiency, S wave Transverse DisplacementVIHIA)012345678910−1.5−1−0.500.51ω [Hz]η(v⊥) [1]SEC efficiency, S wave Transverse VelocityVIHIB)012345678910−1.5−1−0.500.51ω [Hz]η(a⊥) [1]SEC efficiency, S wave Transverse AccelerationVIHIC)equations of motion by r(r − a) gives −ω2(r − a)2( +(r − a)R(cid:48) (cid:18) λ + µ b ρ )2 b − a λ + 2µ r − Rr − (1 − a r Rr = (r − a)2R(cid:48)(cid:48) r )n2 µ λ + 2µ Rr (r − a)R(cid:48) θ − Rθ − µ (1 − a r )Rθ λ + 2µ λ + 2µ +in and 9 (22) (cid:19) −ω2(r − a)2( +(r − a)R(cid:48) (cid:18) λ + µ θ b Rθ = r(r − a)R(cid:48)(cid:48) )2 ρ b − a µ θ − (1 − a )Rθ − n2 λ + 2µ r λ + 2µ Rθ Rr + (1 − a r r + µ µ (r − a)R(cid:48) +in µ )Rr (23) (cid:19) which can be solved by the method of Frobenius for Rr, Rθ. Notably, though, the only terms which are independent of r or d/dr possess dimensionless parameters which can be used to define the scale of different variables. Some of these (e.g. n2µ/(λ + 2µ)) express obvious relations (e.g. the ratio of the speeds of sound times n2), but ξi = ω2a2 (24) relates the frequency of the incoming waves to the natural length scale of the cloak and is thus a non-trivial parameter for this problem. (cid:18) b b − a (cid:19)2 1 v2 i FIG. 7: Transverse component, P wave efficiency plots for the SEC as a function of frequency. Red dashed curve is data for vertical incidence (VI, wave from beneath the SEC), blue dashed curve is data for horizontal incidence (HI, wave from the left side of the SEC). (A) Displacement, (B) Velocity, (C) Acceleration. motion (cid:16) (cid:126)u = (cid:17) Rr(r)r + Rθ(r)θ eiωt+inθ (21) where ω, n are the temporal and azimuthal eigenvalues under separation of variables. Plugging these in and multiplying the [1] Mead, D.J., Passive vibration isolation (Wiley, Chichester, [2] Chopra, A.K., Dynamics of Structures, 3rd Edition (Pearson, 1998). 2007). [3] Reitherman, R.K., Earthquakes and Engineers: An Interna- tional History (ASCE Press, Reston2012). [4] Towhata, I., Geotechnical Earthquake Engineering (Springer, Berlin, 2008). [5] Brul´e, S., Javelaud, E.H., Enoch, S. and Guenneau, S., Experi- ments on seismic metamaterials: Molding surface waves. Phys. Rev. Lett. 112(13), 133901 (2014). [6] Yan, Y., Laskar, A., Cheng, Z., Menq, F., Tang, Y., Mo, Y.L. and Shi, Z., Seismic isolation of two dimensional periodic founda- tions. J. Appl. Phys. 116(4), 044908 (2014). [7] Krodel, S., Thom´e, N., and Daraio, C., Wide band-gap seismic metastructures. Extr. Mech. Lett. 4, 111 (2015). [8] Yan, Y., Cheng, Z., Menq, F., Mo, Y.L., Tang, Y. and Shi, Z., Three dimensional periodic foundations for base seismic isola- tion. Smart Mater. Struct. 24(7), 075006 (2015). [9] Aravantinos-Zafiris, N. and Sigalas, M.M., Large scale phononic metamaterials for seismic isolation. J. Appl. Phys. 118(6), 064901 (2015). [10] Dertimanis, V.K., Antoniadis, I.A. and Chatzi, E.N., Feasibility analysis on the attenuation of strong ground motions using finite periodic lattices of mass-in-mass barriers. J. Eng. Mech. 142(9), 04016060 (2016). 12345678910−2.5−2−1.5−1−0.500.51ω [Hz]η(u⊥) [1]SEC efficiency, P wave Transverse DisplacementVIHIA)12345678910−2.5−2−1.5−1−0.500.51ω [Hz]η(v⊥) [1]SEC efficiency, P wave Transverse DisplacementVIHIB)12345678910−2.5−2−1.5−1−0.500.51ω [Hz]η(a⊥) [1]SEC efficiency, P wave Transverse AccelerationVIHIC)[11] Palermo, A., Krodel, S., Marzani, A., and Daraio, C., Engi- neered metabarrier as shield from seismic surface waves. Sci. Rep. 6, 39356 (2016). [12] Perez-Campous, X., A Comprehensive Study of the Radiated Seismic Energy (Stanford University, 2002). [13] Pendry, J.B., Schurig, D. and Smith, D.R., Controlling electro- magnetic fields. Science 312(5781), 1780 (2006). [14] Leonhardt, U., Optical conformal mapping. Science 312(5781), 1777 (2006). [15] Schurig, D., Mock, J.J., Justice, B.J., Cummer, S.A., Pendry, J.B., Starr, A.F. and Smith, D.R., Metamaterial electromag- netic cloak at microwave frequencies. Science 314(5801), 977 (2006). [16] Li, J. and Pendry, J.B., Hiding under the carpet: a new strategy for cloaking. Phys. Rev. Lett. 101(20), 203901 (2008). [17] Liu, R., Ji, C., Mock, J.J., Chin, J.Y., Cui, T.J. and Smith, D.R., Broadband ground-plane cloak. Science 323(5912), 366 (2009). [18] Ma, H.F. and Cui, T.J., Three-dimensional broadband ground- plane cloak made of metamaterials. Nat. Commun. 1, 21 (2010). [19] Zhou, X., Hu, G. and Lu, T., Elastic wave transparency of a solid sphere coated with metamaterials. Phys. Rev. B 77(2), 024101 (2008). [20] Brun, M., Guenneau, S. and Movchan, A.B., Achieving con- trol of in-plane elastic waves. Appl. Phys. Lett. 94(6), 061903 (2009). [21] Kadic, M., Buckmann, T., Schittny, R., Gumbsch, P., and We- gener, M., Pentamode metamaterials with independently tai- lored bulk modulus and mass density. Phys. Rev. Appl. 2(5), 054007 (2014). [22] Diatta, A. and Guenneau, S., Controlling solid elastic waves with spherical cloaks. Appl. Phys. Lett. 105(2), 021901 (2014). [23] Diatta, A., Achaoui, Y., Brul´e, S., Enoch, S., and Guenneau, S., Control of Rayleigh-like waves in thick plate Willis metamate- rials. AIP Adv. 6(12), 121707 (2016). [24] Khlopotin, A., Olsson, P. and Larsson, F., Transformational cloaking from seismic surface waves by micropolar metama- terials with finite couple stiffness. Wave Motion 58, 53 (2015). [25] Diatta, A. and Guenneau, S., Elastodynamic cloaking and field enhancement for soft spheres. J. Phys. D Appl. Phys. 49(44), 445101 (2016). [26] Milton, G.W., Briane, M. and Willis, J.R., On cloaking for elas- ticity and physical equations with a transformation invariant form. New J. Phys. 8(10), 248 (2006). [27] Norris, A.N. and Shuvalov, A.L., Elastic cloaking theory. Wave Motion 48(6), 525 (2011). [28] Chang, Z., Hu, J., Hu, G., Tao, R. and Wang, Y., Controlling elastic waves with isotropic materials. Appl. Phys. Lett. 98(12), 121904 (2011). [29] Hu, J., Chang, Z. and Hu, G., Approximate method for control- ling solid elastic waves by transformation media.Phys. Rev. B 84(20), 201101 (2011). [30] Norris, A.N. and Parnell, W.J., Hyperelastic cloaking theory: transformation elasticity with pre-stressed solids. Proc. R. Soc. A 468(2146), 2881 (2012). [31] Chang, Z., Liu, X., Hu, G. and Hu, J., Transformation ray method: Controlling high frequency elastic waves (L). J. Acoust. Soc. Am. 132(4), 2942 (2012). [32] Parnell, W.J., Norris, A.N. and Shearer, T., Employing pre- stress to generate finite cloaks for antiplane elastic waves. Appl. Phys. Lett. 100(17), 171907 (2012). 10 [33] Parnell, W.J., Nonlinear pre-stress for cloaking from antiplane elastic waves. Proc. R. Soc. A 468(2138), 563 (2012). [34] Parnell, W.J. and Shearer, T., Antiplane elastic wave cloaking using metamaterials, homogenization and hyperelasticity. Wave Motion 50(7), 1140 (2013). [35] Hu, G. and Liu, H., Nearly cloaking the elastic wave fields. J. Math. Pure. Appl. 104(6), 1045 (2015). [36] Beveridge, A., Wheel, M., and Nash, D., The micropolar elas- tic behavior of model macroscopically heterogeneous materials. Int. J. Solids Struct. 50(1), 246266 (2013). [37] Wheel, M., Frame, J., and Riches, P., Is smaller always stiffer? On size effects in supposedly generalised continua. Int. J. Solids Struct. 6768, 8492 (2015). [38] Norris, A.N. and Parnell, W.J. Hyperelastic Cloaking theory. In: A handbook of metamaterials and nanophotonics. Editors: Maier, S., Shamonina, K., Guenneau, S., Hess, O., and Aizpu- rua., J. (World Scientific, 2017). [39] Diatta, A., Kadic, M., Wegener, M. and Guenneau, S., Scattering problems in elastodynamics. Phys. Rev. B 94(10), 100105(R) (2016). [40] Buckmann, T., Kadic, M., Schittny, R. and Wegener, M., Me- chanical cloak design by direct lattice transformation. Proc. Natl. Acad. Sci. USA 112(16), 49304 (2015). [41] COMSOL AB, COMSOL Multiphysics User's Guide, Version 4.3b (Burlington, 2013). [42] Meyer, P.K., The Impact of High Frequency/Low Energy Seis- mic Waves on Unreinforced Masonry, MS Thesis, MIT (2006). [43] Greenleaf, A., Lassas, M., and Uhlmann, G., Anisotropic con- ductivities that cannot be detected by EIT. Physiolog. Meas. 24(2), 413 (2003). [44] Greenleaf, A., Lassas, M., and Uhlmann, G., On nonuniqueness for Calderon's inverse problem. Math. Res. Lett. 10, 685 (2003). [45] Greenleaf, A., Kurylev, Y., Lassas, M. and Uhlmann, G., Invisi- bility and inverse problems. B. Am. Math. Soc. 46(1), 55 (2009). [46] Sklan, S., Cloaking of the momentum in acoustic waves. Phys. Rev. E 81(1), 016606 (2010). [47] Finocchio, G., Casablanca, O., Ricciardi, G., Alibrandi, U., Garesc`ı, F., Chiappini, M. and Azzerboni, B., Seismic metama- terials based on isochronous mechanical oscillators. Appl. Phys. Lett. 104(19), 191903 (2014). [48] Buckmann, T., Schittny, R., Thiel, M., Kadic, M., Milton, G.W. and Wegener, M., 2014. On three-dimensional dilational elastic metamaterials. New J. Phys. 16(3), 033032 (2014). [49] Zhou, X. and Hu, G., Analytic model of elastic metamaterials with local resonances Phys. Rev. B 79(19), 195109 (2009). [50] Gosselin, C., Duballet, R., Roux, P., Gaudilli`ere, N., Dirren- berger, J. and Morel, P., Large-scale 3D printing of ultra-high performance concretea new processing route for architects and builders. Mater. Design 100, 102 (2016). [51] Zareei, A. and Alam, M.R., Cloaking in shallow-water waves via nonlinear medium transformation. J. Fluid Mech. 778, 273 (2015). [52] Davis, P.M., Rubinstein, J.L., Liu, K.H., Gao, S.S. and Knopoff, L., Northridge earthquake damage caused by geologic focusing of seismic waves. Science 289(5485), 1746 (2000). [53] Massarsch, K.R. and Fellenius, B.H., Deep vibratory com- paction of granular soils. Elsevier Geo-Engineering Book Se- ries 3, 539 (2005).
1705.01097	1	1705	2017-05-02T09:55:26	Graphene-Flakes Printed Wideband Elliptical Dipole Antenna for Low Cost Wireless Communications Applications	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	This letter presents the design, manufacturing and operational performance of a graphene-flakes based screenprinted wideband elliptical dipole antenna operating from 2 GHz up to 5 GHz for low cost wireless communications applications. To investigate radio frequency (RF) conductivity of the printed graphene, a coplanar waveguide (CPW) test structure was designed, fabricated and tested in the frequency range from 1 GHz to 20 GHz. Antenna and CPW were screen-printed on Kapton substrates using a graphene paste formulated with a graphene to binder ratio of 1:2. A combination of thermal treatment and subsequent compression rolling is utilized to further decrease the sheet resistance for printed graphene structures, ultimately reaching 4 Ohm/sq. at 10 {\mu}m thicknesses. For the graphene-flakes printed antenna an antenna efficiency of 60% is obtained. The measured maximum antenna gain is 2.3 dBi at 4.8 GHz. Thus the graphene-flakes printed antenna adds a total loss of only 3.1 dB to an RF link when compared to the same structure screen-printed for reference with a commercial silver ink. This shows that the electrical performance of screen-printed graphene flakes, which also does not degrade after repeated bending, is suitable for realizing low-cost wearable RF wireless communication devices.	physics.app-ph	physics	This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. The final version of record is available at http://dx.doi.org/10.1109/LAWP.2017.2684907 Graphene-Flakes Printed Wideband Elliptical Dipole Antenna for Low Cost Wireless Communications Applications Antti Lamminen, Kirill Arapov, Gijsbertus de With, Samiul Haque, Henrik G. O. Sandberg, Heiner Friedrich, and Vladimir Ermolov (cid:3) Abstract-This letter presents the design, manufacturing and operational performance of a graphene-flakes based screen- printed wideband elliptical dipole antenna operating from 2 GHz up to 5 GHz for low cost wireless communications applications. To investigate radio frequency (RF) conductivity of the printed graphene, a coplanar waveguide (CPW) test structure was designed, fabricated and tested in the frequency range from 1 GHz to 20 GHz. Antenna and CPW were screen-printed on Kapton substrates using a graphene paste formulated with a graphene to binder ratio of 1:2. A combination of thermal treatment and subsequent compression rolling is utilized to further decrease the sheet resistance for printed graphene structures, ultimately reaching 4 (cid:525)(cid:18)(cid:401) at 10 µm thicknesses. For the graphene-flakes printed antenna an antenna efficiency of 60% is obtained. The measured maximum antenna gain is 2.3 dBi at 4.8 GHz. Thus the graphene-flakes printed antenna adds a total loss of only 3.1 dB to an RF link when compared to the same structure screen-printed for reference with a commercial silver ink. This shows that the electrical performance of screen-printed graphene flakes, which also does not degrade after repeated bending, is suitable for realizing low-cost wearable RF wireless communication devices. Index Terms-antenna, graphene, printing, RF, transmission line. I. INTRODUCTION P RINTED flexible graphene antennas for communications systems are at the focus of science and technology on account of their decent electrical, but more importantly, This work was supported by the EC under the Graphene Flagship under grant agreement no. 604391. A. Lamminen, H. G. O. Sandberg, and V. Ermolov are with VTT Technical Research Centre of Finland, Tietotie 3, FI-02044, Espoo, Finland (e-mail: [email protected]). K. Arapov was with the Laboratory of Materials and Interface Chemistry, Department of Chemical Engineering and Chemistry, Eindhoven University of Technology, Eindhoven, De Zaale, 5612AJ, The Netherlands. He is now with Johnson Matthey Advanced Glass Technologies BV, Fregatweg 38 Maastricht, Limburg 6222 NZ, The Netherlands. G. de With and H. Friedrich are with the Laboratory of Materials and Interface Chemistry, Department of Chemical Engineering and Chemistry, Eindhoven University of Technology, Eindhoven, De Zaale, 5612AJ, The Netherlands. S. Haque was with Nokia R&D UK, 21 JJ Thomson Avenue, Madingley Road, Cambridge, CB3 0FA, The United Kingdom. He is now with Emberion Limited, Sheraton House, Castle Park, Cambridge CB3 0AX, The United Kingdom. graphene-based excellent mechanical properties. Recently, graphene-based RFID antennas [1], [2], RF transmission lines and antennas [3], and fully integrated RFID devices [4](cid:3013)[6] have been demonstrated. Conductivity of printed graphene flakes is significantly lower than that of copper, aluminium and even that of printed metallic inks, which are the most commonly in flexible antennas to date. used conductor materials Nevertheless, several advantages such as low cost, chemical stability, mechanical flexibility and fatigue resistance. For comparison, the cost of silver ink depends strongly on the price level of bulk silver metal. The price of silver is very stable and is not expected to decrease in the foreseeable future. The cost for graphene ink raw materials is very low and the manufacturing process is simple and scalable. The cost can be estimated based on the complexity and scalability of the process and is expected to be significantly lower than commercially available silver inks. structures have The above mentioned planar dipole or slot-type antennas which have been realized [1](cid:3013)[6] are narrow band and can be used only in a limited frequency range. There is also a strong demand for low cost wideband antennas. Furthermore, for wideband applications such as ultra-wideband (UWB) wireless body area networks (WBAN) [7](cid:3013)[9], flexibility and bending fatigue resistance for antennas is required. Recently, a wideband graphene-printed triangular slot antenna design based on multiple resonances has been published in [10]. A downside of such antennas is usually multiple reflections in the antenna structure that may distort the UWB pulse in the time domain. We present in this letter a graphene-flakes printed non- resonant planar elliptical element dipole antenna with a reasonably high operational efficiency, designed by taking into consideration the moderate conductivity of printed graphene flakes. We also examine the electrical properties of a graphene-flakes printed transmission line up to 20 GHz. Finally, we demonstrate the flexibility of a graphene-flakes printed sheet by showing a stable resistance for the printed sheet even after numerous bending cycles. II. ANTENNA AND TRANSMISSION LINE DESIGN AND FABRICATION An important consideration for the antenna design is that screen-printed graphene-flake conductors have a relatively Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected]. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. The final version of record is available at http://dx.doi.org/10.1109/LAWP.2017.2684907 Fig. 1. Wideband quasi dipole antennas (WBQD) and coplanar waveguide (CPW) transmission lines. (a) EM simulation model of the WBQD antenna (a = 46 mm, b = 45 mm, w = 4 mm, g = 0.9 mm), (b) printed graphene WBQD antenna, (c) printed silver WBQD antenna, (d) printed graphene CPW, and (e) printed silver CPW. high sheet resistance. Hence, we chose a wideband dipole antenna with elliptical branches [11], where the impact of loss resistance is minimized by maximizing the width-to-length ratio. Thus, antenna radiation efficiency ((cid:536)r) is maximised by keeping the radiation resistance (Rr) high and the loss resistance (Rl) low, following the basics of the antenna theory equation (cid:536)r = Rr/(Rr+Rl) [12]. The antenna has a tapered impedance transformer from feed line to air and thus is not of a resonant type which is usually advantageous in wideband applications. The wideband quasi dipole antenna (WBQD) employs an axial ratio of 1.5 for the ellipse as this has been found to be an optimal value with respect to input matching and uniformity of the radiation pattern [11]. The final WBQD antenna layout (Fig. 1(a)) was designed using a commercial electromagnetic (EM) simulator (Ansys HFSS 2014) to operate at frequencies from 2 GHz up to 5 GHz. In the simulations a target DC sheet resistance of Rs = 4 (cid:525)(cid:18)(cid:401) for graphene-flakes printed conductors was used. The WBQD is fed by a coplanar waveguide (CPW) transmission line shown in Fig. 1(d) and (e). Due to the fact that printed graphene transmission lines have a rather high attenuation at 2(cid:3013)8 GHz [3], the CPW length was minimised by halving one branch of the antenna dipoles. In addition, the antenna design includes a cut-out in the halved branch for a RF connector which is required for RF testing. CPW transmission lines were designed to match with the WBQD antenna (CPW impedance is 93(cid:525), center line width w = 4 mm, gap width g = 0.9 mm, and length l = 44 mm) and were also used to test radio frequency conductivity of the printed graphene. Graphene WBQD and graphene CPWs were screen-printed onto polyimide substrates (Kapton HN; DuPont; USA; 76 µm thickness) using a DEK Horizon 03i (DEK International, UK) semiautomatic screen printer following earlier published procedures [13]. In brief, graphene-flake inks were prepared by gelation of graphene/binder dispersions induced by mild heating with a graphene to binder ratio of 1:2. After solvent exchange the graphene ink was applied to the substrate by screen printing using a 45° angle polyurethane squeegee, at a printing speed of 50 mm/s followed by drying in air at 100 °C for 5 minutes. Subsequently, printed structures were thermally annealed at 350 °C for 30 minutes in air and, finally, compression rolled [1], [14] to reach the target DC sheet resistance of 4 (cid:525)(cid:18)(cid:401) at 10 µm layer thickness. For comparison, WBQDs and CPWs were also screen printed using a commercial silver ink onto electronic grade PET film (ST506; DuPont Teijin Films; thickness 125 µm). Printing was done with EKRA E2 semi-automatic screen and stencil printer (ASYS Group) in air at controlled temperature and relative humidity. The silver structures were dried at 130 °C for 30 minutes to reach a DC sheet resistance of 41 m(cid:525)(cid:18)(cid:401) at an approximate thickness of 50 µm. The printed WBQD antennas are shown in Fig. 1(b) and (c). III. TESTING OF THE PRINTED SAMPLES Before RF testing the WBQD and CPW structures were attached to a 1.5 mm thick Rohacell foam sheet for mechanical support as the foam has (cid:304)r,R(cid:3) (cid:167) 1 and, hence, a negligible effect on RF performance. Using an Agilent 8722ES vector network analyser (VNA) the S-parameters of the CPW test structures were measured from 1 GHz up to 20 GHz. The attenuation (A) of the CPW was calculated from the measured S-parameters following A = (1-S112)/S212 [3]. A comparison of the simulated and measured attenuation for the graphene CPWs and the measured attenuation for the silver CPW are shown in Fig. 2(a). Graphene CPW simulations with a sheet resistance of Rs = 4 (cid:525)(cid:18)(cid:401) agree well with the measurements up to 5 GHz. At higher frequencies up to 13 GHz, agreement is good with simulations using Rs = 5 (cid:525)(cid:18)(cid:401) and above 13 GHz, the results agree well with simulations using Rs = 6 (cid:525)(cid:18)(cid:401) up to 19 GHz. The observed increase of sheet resistance with higher frequencies is probably caused by increased contact resistance of the highly porous graphene nanoflakes [1]. However, the results show that the sheet resistance of the printed graphene flakes is reasonable up to 20 GHz for developing antennas having moderate efficiencies. The reflection coefficients (S11) of the antennas were measured for the antenna frequency bandwidth from 1 GHz up to 5 GHz. A ferrite choke was used around the cable near the antenna feeding point to provide high impedance for the common-mode currents induced by the dipole antenna and this way reducing the cable effect in the measurements [15], [16]. The simulated and measured S11 of the antennas are presented in Fig. 2(b). It is seen that the measured S11 is below -8.5 dB from 1.7 GHz at least up to 5 GHz for both antennas, which indicates wideband performance. The results also show that over 85% of the input power is reaching the antenna and less than 15% is reflected back, so that the input matching is good Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected]. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. The final version of record is available at http://dx.doi.org/10.1109/LAWP.2017.2684907 and the antennas work as designed. Some deviation can be observed between the simulation and experimental results, which is due to common-mode currents that are not perfectly suppressed even when a ferrite choke is used. The graphene antenna has lower S11 around 2 GHz, i.e., the input impedance is closer to 50 (cid:525) than that of silver antenna. From 2.5 GHz to 5 GHz the results for silver and graphene antennas are very similar. The radiation patterns of the printed graphene-WBQD antenna were measured from 2 GHz up to 4.8 GHz. The ORT was included in simulations for precise comparison with the measurements. Simulated and measured co-polarised and cross-polarised gain patterns at 3 GHz are shown in Fig. 4(a) and (b) for two principal planes (cid:1483) = 0 degrees and (cid:1483) = 90 degrees, respectively. Typical dipole-like radiation patterns can be observed. Fig. 2. (a) Simulated and measured attenuation for graphene CPW and measured attenuation for silver CPW. (b) Simulated and measured reflection coefficient for graphene WBQD antenna, and measured reflection coefficient for silver WBQD antenna. The radiation patterns of the antennas were measured in an anechoic chamber. To minimise RF cable effects in the antenna measurements, an electro-optical link has been proposed [17]. An optical-to-RF transformer (ORT) with dimensions of 20 mm×23.5 mm×10 mm was also employed in this work. In the measurements, an optical signal is fed to the transmitting antenna under test (AUT) via an optical fiber to minimize the influence of RF cable radiation (Fig. 3(a)). The signal is transformed into an RF signal in the ORT block. The ORT output is directly connected to the AUT feeding RF connector. Here it should be noted that common-mode currents flow on the ORT block, which inevitably becomes a part of the antenna structure. Fig. 3. (a) Image of graphene WBQD antenna connected to an optical-to-RF transformer. (b) Image of radiation pattern measurements in anechoic chamber. The received signal was measured using a standard gain horn antenna and Agilent AG 8722E VNA analyser. The antenna is rotated by 360 deg. (-180 (cid:148)(cid:3) (cid:537)(cid:3) (cid:148) 180 deg.) in the horizontal plane with an antenna positioner by Scientific Atlanta and data is acquired in 2 degree steps. The antenna gain was determined using the gain comparison method with known standard gain reference antennas. Antenna efficiencies were determined from the measured directivity and the antenna gain. A photograph of the radiation pattern measurement setup is shown in Fig. 3(b). Fig. 4. Simulated and measured radiation patterns of the printed graphene WBQD antenna at 3 GHz in (a) (cid:1483) = 0 deg. plane and in (b) (cid:1483) = 90 deg. plane. An excellent match between simulation and measurement is found in the co-polarised beam. The maximum directivity and gain are 3.3 dBi and 0.6 dBi resulting in an antenna efficiency of 56 ± 5%. The antenna has omni-directional coverage which is very suitable for mobile applications. The simulated cross- polarisation level is below -40 dBi while the measured level is approximately -17 dBi at maximum, which is due to a RF leakage by the ORT, increasing the gain in cross-polarisation. Despite the measured cross- polarisation level is sufficiently low that to warrant no effect on the overall antenna performance. the observed differences, An important characteristics of the antenna is its directivity which describes the power density the antenna radiates over solid angle. Simulated and measured maximum directivity and maximum antenna gain are shown in Fig. 5(a) as a function of frequency. Overall, the simulations and measurements agree well. The curves are quite stable with frequency but some dips are observed at around 2(cid:3013)2.5 GHz. These dips are a measurement artefact and a result of a superposition of signals radiated by the antenna and the current flowing on the surface of the ORT metal housing. The effect was confirmed by simulations with and without ORT. Above 3 GHz, the surface currents on the ORT surface are very small and the radiation is excited only from the graphene-flakes printed antenna. It should be noted that in a real application an ORT would not be required. The antenna gain increases with frequency and has a maximum of 2.3 dBi at 4.8 GHz. Fig. 5. Simulated and measured (a) maximum antenna gain and maximum directivity, and (b) antenna efficiency as a function of frequency of the printed graphene WBQD antenna. Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected]. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. The final version of record is available at http://dx.doi.org/10.1109/LAWP.2017.2684907 Simulated and measured antenna efficiencies as a function of frequency are shown in Fig. 5(b). At frequencies from 2.8 GHz up to 4.8 GHz an antenna efficiency of 60% is measured. To the best of authors' knowledge, these are the highest gain and efficiency values published for graphene-flakes printed antennas to date. For comparison, the printed silver WBQD antenna was measured at 2 GHz and has a measured directivity and antenna gain of 2.7 dBi and 2.0 dBi, respectively, resulting in an antenna efficiency of 86%. Even though the sheet resistance of printed graphene flakes is two orders of magnitude higher than the sheet resistance of printed silver, a total loss in an RF link budget (including transmitting and receiving sides) is only 3.1 dB higher than for a silver-printed antenna. Such a performance is adequate for usage of our developed graphene-flakes printed antenna in wireless communications applications with ranges from meters up to tens of meters. Printed graphene layers were analyzed with respect to effects of repeated bending on their electrical resistance. For bending experiments a strip of material 11.3 cm long 0.9 cm wide was cut from the substrate and fixed with electrical contacts in the bending rig. During each of 32500 bending cycles (with a span of 40 mm and a bending radius of 42.5 mm per cycle) the resistance of the strip was measured. As seen in Fig. 6 we did not find any significant influence of repeated bending on the measured resistance. Fig. 6. Variation of resistance of graphene layer as a function of bending for 32500 cycles (with a span of 40 mm and a bending radius 42.5 mm per cycle). The absolute resistance of the strip decreased from 70.73 (cid:525) to 69.97 (cid:525), corresponding to about 1% gain in conductivity, indicating excellent fatigue resistance, at the described experimental conditions. For reference, screen printed silver on similar substrates have been shown to experience a gradual increase in resistance up to 10(cid:3013)20 % within a few hundred repetitions when bent to a radius of 6 mm [18]. least for A combination of a reasonable electrical performance with no degradation of conductive properties of the printed graphene-flakes layer even after 32500 bending cycles, renders the printed graphene flakes antennas suitable for low- cost wearable wireless communications devices such as health monitoring or smart clothing. ACKNOWLEDGMENT The authors would like to thank Mr. Ismo Huhtinen for the measurements and Mr. Alpo Ahonen for assembly work. Ms. Asta Pesonen is acknowledged for technical assistance with silver printing. REFERENCES [1] X. Huang, T. Leng T, X. Zhang, J. C. Chen, K. H. Chang, A. K. Geim, K. S. Novoselov, Z. Hu, "Binder-free highly conductive graphene laminate for low cost printed radio frequency applications," Appl. Phys. Lett. 106, 203105, pp.1(cid:3013)4, 2015. [2] T. Leng, X. Huang, K. Chang, J. Chen, M. A. Abdalla, Z. Hu, "Graphene nanoflakes printed flexible meandered-line dipole antenna on paper substrate for low-cost RFID and sensing applications," IEEE Antennas Wireless Propag. Lett. vol. 15, pp. 1565(cid:3013)1568, 2016. [3] X. Huang, T. Leng, M. Zhu, X. Zhang, J. Chen, K. Chang, M. Aqeeli, A. K. Geim, K. S. Novoselov, Z. Hu, "Highly flexible and conductive printed graphene for wireless wearable communications applications," Sci. Rep. 5:18298, pp. 1(cid:3013)7, 2015. [4] K. Arapov, K. Jaakkola, V. Ermolov, G. Bex, E. Rubingh, S. Haque, H. Sandberg, R. Abbel, G. de With, H. Friedrich, "Graphene screen-printed radio-frequency identification devices on flexible substrates," Phys. Status Solidi RRL, pp. 1(cid:3013)7, 2016. [5] M. Akbari, M. Waqas, A. Khan, M. Hasani, T. Björninen, L. Sydänheimo, L. Ukkonen, "Fabrication and characterization of graphene antenna for low-cost and environmentally friendly RFID tags," IEEE Antennas Wireless Propag. Lett. vol. 15, pp. 1569(cid:3013)1572, 2016. [6] P. Kopyt, B. Salski, M. Olszewska-Placha, D. Janczak, M. Sloma, T. Kurkus, M. Jakubowska, W. Gwarek, "Graphene-based dipole antenna for a UHF RFID tag," IEEE Trans. Antennas Propag. vol. 64, pp. 2862(cid:3013) 2868, 2016. J. Ryckaert, C. Desset, A. Fort, M. Badaroglu, V. D. Heyn, P. Wambacq, G. V. d. Plas, S. Donnay, B. V. Poucke, B. Gyselinckx, "Ultra-wide- Band transmitter for low-power wireless body area networks: design and evaluation," IEEE Trans. Circuits Syst. vol. 52, pp. 2515(cid:3013)2525, 2005. [8] Q. H. Abbasi, A. Alomainy, Y. Hao, "Recent Development of Ultra wideband body-centric wireless communications," in Proc. of the IEEE International Conference on Ultra-Wideband (ICUWB), Nanjing, China, 20(cid:3013)23 September 2010, pp. 1(cid:3013)4. [9] N. P. Gupta, R. Maheshwari, M. Kumar, "Advancement in ultra wideband antennas for wearable applications," IJSER, vol. 4, pp. 341(cid:3013) 348, 2013. [10] X. Huang, T. Leng, K. H. Chang, J. C. Chen, K. S. Novoselov, Z. Hu, "Graphene radio frequency and microwave passive components for low cost wearable electronics," 2D Mater., vol. 3, 025021, pp. 1(cid:3013)8, 2016. [11] H. G. Schantz, "Planar elliptical element ultra-wideband dipole [7] antennas," in IEEE AP-S Int. Symp. Dig. vol.. 3, pp. 44(cid:3013)47, 2002. [12] C. A. Balanis, Antenna Theory: Analysis and Design (John Wiley & Sons, Hoboken, 1997), p. 78. [13] K. Arapov, E. Rubingh, R. Abbel, J. Laven, G. de With, H. Friedrich, "Conductive screen printing inks by gelation of graphene dispersions," Adv. Funct. Mater. 26, pp. 586(cid:3013)593, 2016. [14] K. Arapov, G. Bex, R. Hendrix, E. Rubingh, R. Abbel, G. de With, H. Friedrich, "Conductivity enhancement of binder-based graphene inks by photonic annealing and subsequent compression rolling," Adv. Eng. Mater., pp. 1(cid:3013)6, 2016. [15] J. DeMarinis, "The antenna cable as a source of error in EMI measurements," in Proc. of the IEEE International Symposium on Electromagnetic Compatibility (ISEMC), Seattle, Washington, 2(cid:3013)4 August 1988, pp. 9(cid:3013)14. [16] T. W. Hertel, "Cable-current effects of miniature UWB antennas," in Proc. of the IEEE Antennas and Propagation Society International Symposium, Washington, DC, 3(cid:3013)8 July 2005, pp. 524–527. to reduce [17] R. R. Lao, W. Liang, Y.-S. Chen, J. H. Tarng, "The use of electro- optical in antenna measurement, in Proc. of the IEEE International Symposium on Microwave, Antenna, Propagation and EMC Technologies for Wireless Communications (MAPE), Beijing, China, 8(cid:3013)12 August 2005, pp. 427(cid:3013) 430. influence of RF cables [18] T. Happonen, T. Ritvonen, P. Korhonen, J. Häkkinen, T. Fabritius, "Bending reliability of printed conductors deposited on plastic foil with various silver pastes," The International Journal of Advanced Manufacturing Technology, 82, pp. 1663(cid:3013)1673, 2016. link the Copyright (c) 2017 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing [email protected].
1710.00040	2	1710	2017-12-28T00:11:10	Synthesis of Spherical Metasurfaces based on Susceptibility Tensor GSTCs	[ "physics.app-ph", "physics.class-ph", "physics.med-ph" ]	The bianisotropic susceptibility Generalized Sheet Transition Conditions (GSTCs) synthesis method is extended from planar to spherical metasurfaces. Properties specific to the non-zero intrinsic curvature of the spherical shape are highlighted and different types of corresponding transformations are described. Finally, the susceptibility-GSTC method and exotic properties of spherical metasurfaces are validated and illustrated with three examples: illusion transformation, ring focusing and birefringence.	physics.app-ph	physics	Synthesis of Spherical Metasurfaces based on Susceptibility Tensor GSTCs Xiao Jia, Yousef Vahabzadeh, Fan Yang, Senior Member, IEEE, Christophe Caloz, Fellow, IEEE 1 7 1 0 2 c e D 8 2 ] h p - p p a . s c i s y h p [ 2 v 0 4 0 0 0 . 0 1 7 1 : v i X r a Abstract-The bianisotropic susceptibility Generalized Sheet Transition Conditions (GSTCs) synthesis method is extended from planar to spherical metasurfaces. Properties specific to the non-zero intrinsic curvature of the spherical shape are highlighted and different types of corresponding transformations are described. Finally, the susceptibility-GSTC method and exotic properties of spherical metasurfaces are validated and illustrated with three examples: illusion transformation, ring focusing and birefringence. Index Terms-Spherical metasurface, Generalized Sheet Tran- sition Conditions (GSTCs), bianisotropic susceptibilities, synthe- sis, electromagnetic transformations. I. INTRODUCTION Metasurfaces are electrically thin 2D structures consisting of a subwavelength lattices of scattering particles that are capable to transform electromagnetic waves in unprecedented fashions [1]–[5]. Despite their recent emergence, they have already lead to an impressive number of applications, includ- ing ultra-thin optical lenses [6], high-resolution holograms [7], enhanced classical/quantum efficiency cavities [8], spatial an- gular filters [9], perfect absorbers [10], remote controllers [11], spatial operators [12], ultrafast processors [13] and surface plasmonic sensors [14]. The vast majority of metasurfaces reported to date were planar. However, many applications, such as cloaking, air- craft RCS reduction, vital signal detection, etc. would greatly benefit form other metasurface shapes. Shapes of interest may greatly vary, and even include most complex irregular shapes. However, irregular shapes typically do not admit mathematical solutions and may often be approximated by simpler and more insightful canonical shapes. Therefore, it makes sense to first consider such canonical shapes. Canonical shapes may be classified into two main categories [15]: a) shapes of zero intrinsic curvature, that may be obtained by folding flat sheets, such as corrugated surfaces, and cylinders with circular, elliptic, parabolic or hyperbolic cross sections, and b) shapes of non-zero-intrinsic curvature, which are fundamentally 3D, such as spheroidal, prolate, oblate, paraboloidal and conical shells. Xiao Jia is with Department of Electronic Engineering, Tsinghua and the Department of Electri- (e-mail: University, Beijing 100084, China cal Engineering, Polytechnique Montr´eal, Montr´eal, Qu´ebec. [email protected]). are with of (e- Y. Vahabzadeh the Department Electrical Engineering, Polytechnique Montr´eal, Montr´eal, Qu´ebec. mail:[email protected]; [email protected]). and C. Caloz Fan Yang is with Department of Electronic Engineering, Tsinghua Univer- sity, Beijing 100084, China. (e-mail:fan [email protected]). The simplest and most common of these canonical shapes are the cylinder with circular cross section, within category a), and the sphere, within category b). The former has been reported in leaky-wave antennas [16], 2D beam formers [17] and RCS reducers [18], while the later has been reported in 3D beam formers [19] and antenna decouplers [20]. The development of practical cylindrical and spherical metasur- faces will naturally require efficient design methods, and the Bianisotropic Susceptibility - Generalized Sheet Transition Condition (GSTC) approach [4], [21], [22], that has been successfully applied to planar metasurfaces [4], is a natural candidate for the design of such metasurfaces. The cylindrical case has already been treated in [23]. This paper presents the GSTC synthesis of spherical metasurfaces. The paper is organized as follows. Section II describes the spherical metasurface problem and its specificities. Next, Sec. III presents the extension of the susceptibility-GSTC method to spherical metasurfaces, with specific transformation types and scattering parameter mapping. Then, Sec. IV illus- trates and validates the method for some interesting spherical metasurface transformations. Conclusions are given in Sec. V. II. SPHERICAL METASURFACE PROBLEM The spherical metasurface structure and synthesis problem are represented in Fig. 1. A spherical metasurface is funda- mentally different from a planar metasurface, or from a curved metasurface with electrically large curvature1, which may be considered as electromagnetically quasi-planar. In the planar and quasi-planar cases, an incident wave is simply reflected and transmitted by the metasurface in a single scattering event, since the metasurface structure is open and smoothly varying. In contrast, a spherical metasurface is a closed structure, forming de facto a porous cavity, where the initial reflection- transmission event may be followed by multiple scattering events, and even resonance effects in the case of large local reflections2. Another fundamental difference, is that, while the planar and curved metasurfaces are practically finite and may hence diffract the incident wave at their edges and corners, the 1Under this condition, the Rayleigh hypothesis, according to which scat- tering is exclusively composed of outgoing waves [24] [25], holds. This practically means that a ray impinging on the metasurface directly scatters (reflects, refracts and/or diffracts) at its incidence point and does not get trapped and multiply scattered in the troughs of the structure. An example of such a curved metasurface is a periodic corrugated metasurface with a ratio of corrugation height over period much smaller than the wavelength. 2That would for instance be the case in the stop-bands of a spherical frequency selective surface (FSS), where that cavity becomes completely opaque. spherical metasurface is rotationally infinite and hence does not include edge or corner diffraction3. 2 Fig. 1. Spherical metasurface structure and synthesis problem. The structure consists of a deeply subwavelength (δ ! λ) spherical shell of radius a, centered at r " 0, and composed of spherically curved subwavelength scatter- ing particles. The synthesis problem consists in determining the metasurface susceptibility tensor, χpθ, φq, for transforming a specified arbitrary incident field, Ψi, into a specified arbitrary reflected field, Ψr, and a specified arbitrary transmitted field, Ψt. The problem of a general porous cavity is very complex, and we do not address it here4. Here, we restrict our attention to the particular case of a spherical metasurface that a) is excited from its inside, and b) is reflection-less, so as to avoid multiple internal scattering5 as planar and quasi-planar metasurfaces. While this is a major restriction, the inside- excitation reflection-less problem already represents an elec- tromagnetically rich and practically interesting metasurface, allowing unusual and exotic field transformations, as will be shown next. Figure 2 shows the different possible categories of spherical metasurfaces based on their azimuthal and elevation suscepti- bility variations. The susceptibility may be uniform in both elevation and azimuth [Fig. 2(a)], uniform in one angular direction and nonuniform in the other one [Figs. 2(b) and (c)], or nonuniform in both angular directions [Fig. 2(d)]. III. SYNTHESIS A. Susceptibility GSTCs The GSTC equations for the spherical metasurface problem in Fig. 1 are derived in Appendix A. They may be written as rr (cid:52)H " Js,totsr"a, rr (cid:52)E " ´Ks,totsr"a, (1a) (1b) where r is the unit vector normal the metasurface. In these relations, the metasurface discontinuity, i.e. the (cid:52) symbol represents the field jumps at (cid:52)Ψ " Ψ` ´ Ψ´ " Ψt ´ pΨi ` Ψrq, Ψ " E, H, 3This is true in the full spherical metasurface considered in this paper. However, a spherical metasurface cap, which would represent a further interesting and practically useful problem, would naturally also feature edge and corner diffraction. 4This problem, and other related interesting problems, such as that of a spherical-shell cavity, would naturally require more intensive studies, which are out of the scope of this paper and may be the object of later publications. 5This problem may be considered as the limiting case of a porous spherical (2) cavities with full porosity. Fig. 2. Categories of spherical metasurfaces based on their susceptibilities. (a) Uniform susceptibility. (b) θ-uniform and φ-varying susceptibitity. (c) φ- uniform and θ-varying susceptibitity (same problem as b), upon π{2 rotation). (d) Double nonuniform susceptibitity. with the superscripts referring to the points just before and just after the metasurface, i.e. at r " a, and the superscripts i, r and t referring to the incident, reflected and transmitted fields, respectively. Moreover, Js,tot " Js,imp ` Js,p ` Js,m " Js,imp ` BPs Bt ` ∇ Ms (A/m) and (3a) (3b) (3c) (3d) Ks,tot " Ks,imp ` Ks,m ` Ks,p " Ks,imp ` µ0BMs Bt ` ∇ pPs{0q (V/m) represent the total surface current densities on the metasurface, which are composed of the impressed electric and magnetic surface current densities, Js,imp and Ks,imp, the electric surface current densities due to electric and magnetic polarization densities, Js,p and Js,m, and the magnetic surface current densities due to electric and magnetic polarization densities, in (3) are surface Ks,p and Ks,m. Of course, Ps and Ms polarization densities (see Appendix A), measured in C/m and A, respectively. Note that Eqs. (3) are restricted to first-order discontinuities, with Ψs " Ψδpr ´ aq, Ψs " Js,p, Ks,p, Js,m, Ks,m In the particular case Ps " Ms " 0, only the impressed 6This restriction is valid for most practical metasurfaces. However, there are cases where it would not be acceptable. For instance, a metasurface transforming the incident field into a transmitted field being to its phase- reversed version could not be described by a series truncated to N " 0. the corresponding fields would include only the even δpr ´ aq Indeed, distribution whereas the field transformation is obviously odd in nature. In such a case, one should at least include the term N " 1, to include the odd distribution δ1pr ´ aq, following [4], [22], which would involve extra terms in (3). The application of higher-order GSTCs to metasurfaces is still an open research topic. 6. xyzφθθθθrǫ1,µ1ǫ2,µ2aaaaa¯¯χpθ,φq"?ΨiΨiΨiΨiΨiΨrΨrΨtδ!λδ!λδ!λδ!λδ!λδ!λδ!λxxxxyyyyzzzz¯¯χpθq¯¯χpφq¯¯χpθ,φq¯¯χ(a)(b)(c)(d)surface current densities, Js,imp and Ks,imp, survive in (3), and Eqs. (1) reduce to the usual boundary conditions at the interface between two media [26]. However, we are interested here in the opposite case, where Js,imp " Ks,imp " 0, assuming the inexistence of sources on the metasurface, and field dis- continuities only due to the polarization currents modeling the response of the metasurface scattering particles via the surface electric and magnetic polarization densities Js,p, Js,m, Ks,p and Ks,m, respectively7. In this paper, we shall restrict our attention to time- harmonic (ejωt) metasurfaces8. Inserting the time-harmonic versions (jω) of the polarization current densities (3), with Js,imp " Ks,imp " 0, into the GSTCs (1) yields (4a) (4b) rr (cid:52)H " jωPs,} ´ r ∇}Ms,rsr"a, rr (cid:52)E " ´jωµ0Ms,} ` ∇}pPs,r{0q rsr"a. The physical metasurface will actually be a spherical shell with finite thickness δ, as indicated in Fig. 1. However, this thickness is typically deeply subwavelength (δ ! λ). Therefore, the shell cannot support significant propagation or resonance effects along the r direction, and the metasurface may hence be safely modelled as a zero-thickness (δ " 0) sheet discontinuity through the bianisotropic surface suscep- tibility tensor functions χee, χmm, χem and χme, that relate the average fields at both sides of the metasurface, 2 Ψav " to the surface polarization densities as [4] , Ψ " E, H, Ψt ` pΨi ` Ψrq a P " 0 ¯¯χeeEav ` ?µ00 ¯¯χemHav, 0{µ0 ¯¯χmeEav ` ¯¯χmmHav. M " a r (cid:52)H " jωp0χeeEav ` ?µ00χemHavq} a ´ r ∇}rp 0{µ0χmeEav ` χmmHavqrs, 0{µ0χmeEav ` χmmHavq} r (cid:52)E " ´jωµ0p ` ∇}rp0χeeEav ` ?µ00χemHavqr{0s r, Inserting (6) into (4) finally provides the GSTC relations (5) (6a) (6b) (7a) (7b) explicitly expressed in terms of the difference fields in (2) on the left-hand sides and average fields in (5) on the right-hand sides through the surface susceptibility tensors, that read in spherical coordinates ¯¯χabpθ, φq " abpθ, φq χrθ abpθ, φq χθθ χθr ab pθ, φq χφθ χφr abpθ, φq χrφ ab pθ, φq χθφ ab pθ, φq χφφ ab pθ, φq ab pθ, φq ab pθ, φq with pa,bq " (e,e), (e,m), (m,e) and (m,m). In (7), we have dropped, for notational simplicity, the r. . .sr"a specification, 7The GSTCs (1) may thus be considered as a generalization of the usual boundary conditions including the effect of surface material polarization. Equations (3) were not common in the "pre-metasurface era" literature because, before the advent of metasurfaces, 2D materials (e.g. 2DEGs, graphene, etc.) and related computational sheets, polarization was essentially a volume concept, defined as the densities of electric and magnetic moments in 3D space, P and M, measured in (C/m2) and (A/m), respectively, which did not make sense in two dimensions. 8Polychromatic planar time-varying and nonlinear metasurface transforma- tions have been considered in [27], [28]. fifl , (8) »–χrr 3 which is implicitly assumed from now on. This information of the possibly different media surrounding the metasurface is implicitly present in (2) and (5), as pointed out in Appendix A. B. Synthesis Equations In this paper, we shall assume Pr " Mr " 0, which simpli- fies the coupled partial differential equations (8) to a simple algebraic linear system of equations. As extensively discussed in Secs. IV.A and IV.D of [29], this represents a restriction in terms of fabrication and separate transformation diversity, but no restriction in terms of an ideal metasurface performing a given transformation, including a transformation involving multiple simultaneous operations, since a metasurface with normal polarization components can always be reduced to an equivalent metasurface purely tangential polarization compo- nents.  Under the condition Pr " Mr " 0, Eqs. (7) with (8) reduce to„  (9a) ´(cid:52)Hφ (cid:52)Hθ " jω0 „ χθφ ee χφφ ee χθθ ee χφθ ee „  „ „ „ „ Eθ,av Eφ,av em χθφ χθθ em χφθ em χφφ em   and„ (cid:52)Eφ ´(cid:52)Eθ " jωµ0 ` jω?µ00 „ mm χθφ χθθ mm mm χφφ χφθ mm ` jω?µ00 Hθ,av Hφ,av Hθ,av Hφ,av me χθφ χθθ me χφθ me χφφ me „  Eθ,av Eφ,av . (9b) mm, χθφ Equations (9) represent a system of 4 equations in 16 un- ee , etc.). So, this is a heavily under-determined knowns (χθθ system. There are three approaches to solve this problem [29]. The first one is to reduce the number of unknowns from 16 to 4, in which case there would be 44 " 256 possible distinct susceptibility quadruplets, with only a subset of them repre- senting physically meaningful situations. The second approach is to increase the number of simultaneous field transformation specifications from 1 to 4. The last approach is a combination of the first two. To design an optimal metasurface, one has to make an educated choice of approach and susceptibility sets. Such an educated choice includes the conisideration of the following fundamental conditions: ‚ reciprocal (possibly with loss or gain) metasurface: ¯¯χT ee " ¯¯χee, ¯¯χT mm " ¯¯χmm, ¯¯χT me " ´ ¯¯χem, (10) which implies the suppression of 6 complex (i.e. 12 real numbers) susceptibility degrees of freedom; ‚ loss/gain-less reciprocal metasurface: ee " ¯¯χee, ¯¯χT mm " ¯¯χmm, ¯¯χT me " ¯¯χem, ¯¯χT (11) which implies the suppression of 16 real number degrees of freedom among the complex susceptibilities; ‚ non-gyrotropic metasurface: ee,mm " 0, ¯¯χθθ ¯¯χθφ ee,mm " 0, ¯¯χφθ em,me " 0, ¯¯χφφ em,me " 0, (12) which implies the suppression of 8 complex (i.e. 16 real numbers) susceptibility degrees of freedom; C. Transformation Types We derive here closed-form susceptibility solutions to (9) for a few types of transformations depending on the afore- mentioned approaches. The solutions to other types of trans- formations can be derived in a similar manner. In this section, we will give these closed-form solutions as functions of the difference fields (2) and average fields (5), i.e. as implicit synthesis relations, while explicit examples will be given in Sec. IV. 1) Monoisotropic Transformation: The simplest possible transformation is the monoisotropic transformation, which may be considered as a particular case of a 4-parameter transformation (approach 1) with χθ,θ ee,mm " χee,mm. In this case, Eqs. (9) reduce to „  ´(cid:52)Hφ „  (cid:52)Hθ (cid:52)Eφ ´(cid:52)Eθ „  ee,mm " χφ,φ  „ Eθ,av Eφ,av , " jω0χee " jωµ0χmm Hθ,av Hφ,av . (13a) (13b) and Solving this system for χee and χmm yields the closed-form synthesis solutions χee " ´ (cid:52)Hφ χmm " (cid:52)Eφ jω0Eθ,av " (cid:52)Hθ jωµ0Hθ,av " ´ (cid:52)Eθ jω0Eφ,av jωµ0Hφ,av , (14a) (14b) , showing that the corresponding metasurface performs the same transformation on θ-polarized and φ-polarized waves. 2) Monoanisotropic Transformation: For the metasur- face to perform different transformations on the θ- and φ-polarizations (birefringence), one may lift the previous (monoisotropic) restriction to monoanisotropy, involving the mm. This is another type susceptibilities χθθ of 4-parameter transformation (approach 1), but this time with 4 distinct susceptibilities. In this case, Eqs. (9) become mm and χφφ ee , χφφ ee , χθθ „ „   ´(cid:52)Hφ (cid:52)Hθ (cid:52)Eφ ´(cid:52)Eθ „ „ " jω0 " jωµ0 χθθ ee 0 χθθ mm 0 0 χφφ ee 0 χφφ mm „ „ Eθ,av Eφ,av   , Hθ,av Hφ,av , and their solution is χθθ jω0Eθ,av ee " ´ (cid:52)Hφ ee " (cid:52)Hθ χφφ jω0Eφ,av , χφφ , χθθ jωµ0Hφ,av mm " ´ (cid:52)Eθ mm " (cid:52)Eφ jωµ0Hθ,av which correspond to θ and φ polarizations, respectively. 3) Bianisotropic Transformation: As shown in [30] [31], perfect refraction, i.e. refraction without loss/gain and without spurious diffraction, requires bianisotropy, for which χem ‰ 0 and χme ‰ 0. If one further wishes to perform such a transfor- mation without field rotation (gyrotropy), the condition (12) (15a) (15b) . (16a) , (16b) 4 must be further enforced, which eliminates 8 complex suscep- tibilities (approach 1). This leaves out 8 complex susceptibility parameters (among which one must ensure χem ‰ 0 and χme ‰ 0), which further reduces to 4 complex susceptibility parameters if one cares for only one polarization (and the transformation of the other polarization is arbitrary). In this case, we also have two equations in (9) disappearing, reducing the total number of equations from 4 to 2. In the case of θ-polarization, we have then 2 equations for the remaining parameters are χθθ the system is under-determined, which allows us to specify a second transformation (approach 2). In this case, Eqs. (9) may be compactly written (cid:52)Hφ1 (cid:52)Hφ2 (cid:52)Eθ1 (cid:52)Eθ2 Eθ1,av Eθ2,av Hφ1,av Hφ2,av (17) where the subscripts 1 and 2 correspond to the two transfor- mations. The double transformation in (17) involves only 2 of the 4 equations in (9), and is hence a reduced-rank (from 4 to 2) transformation. ´jk0χθφ me ´jωµ0χφφ ´jω0χθθ ´jk0χφθ em and χφθ „ ee , χφφ mm, χθφ me. So,  „ „ " mm em ee Equation (17) represents a system of 4 equations in 4 unknowns, whose solution is  χθθ ee " χθφ em " χφθ me " χφφ mm " 1 jω0 1 jk0 1 jk0 1 (cid:52)Hφ2Hφ1,av ´ (cid:52)Hφ1Hφ2,av Eθ1,avHφ2,av ´ Eθ2,avHφ1,av (cid:52)Hφ2Eθ1,av ´ (cid:52)Hφ1Eθ2,av , Eθ2,avHφ1,av ´ Eθ1,avHφ2,av (cid:52)Eθ2Hφ1,av ´ (cid:52)Eθ1Hφ2,av , Eθ1,avHφ2,av ´ Eθ2,avHφ1,av (cid:52)Eθ2Eθ1,av ´ (cid:52)Eθ1Eθ2,av Eθ2,avHφ1,av ´ Eθ1,avHφ2,av jωµ0 , . (18a) (18b) (18c) (18d) Equation (18) generally represents a double transformation. If one further wanted a reciprocal metasurface, as is often the case both functionally and practically, then the third relation in (10) would demand a) χθφ me, and b) transformation 2 to be the reciprocal transformation of transformation 19. The combination of these 2 constraints leads to a new fully- determined system, which may seen as a single reciprocal transformation. em " ´χφθ 4) Full-rank Double Transformation: The double transfor- mation of (17) is a reduced-rank one because it specifies only one polarization specification. We shall now consider the case of a transformation with specifications for both polarizations. This leads to a full-rank system, involving the 4 equations in (9) and, without non-gyrotropy constraint, 16 unknowns. Using approach 1, we further specify here monoanisotropy, which leads, using the short-cut notation χee " jω0χee and χmm " jωµ0χmm, to the system »-–´(cid:52)Hφ1 ´(cid:52)Hφ2 (cid:52)Hθ1 (cid:52)Hθ2 (cid:52)Eφ2 (cid:52)Eφ1 ´(cid:52)Eθ1 ´(cid:52)Eθ2 fiffifl " »--– χθθ ee χφθ ee 0 0 χθφ ee χφφ ee 0 0 0 0 χθθ mm χφθ mm 0 0 χθφ mm χφφ mm fiffiffifl »-–Eθ1,av Eθ2,av Eφ1,av Eφ2,av Hθ1,av Hθ2,av Hφ1,av Hφ2,av fiffifl , (19) 9In this case, transformation 2 would be from the outside to the inside of the metasurface sphere, and the reciprocity specification would be physical only if the wave is strongly attenuated at the metasurface surface or/and within its filling medium. whose solution is 1 χθθ ee " χθφ ee " χφθ ee " χφφ ee " χθθ mm " χθφ mm " χφθ mm " χφφ mm " , , , (cid:52)Hφ2Eφ1,av ´ (cid:52)Hφ1Eφ2,av Eθ1,avEφ2,av ´ Eθ2,avEφ1,av (cid:52)Hφ2Eθ1,av ´ (cid:52)Hφ1Eθ2,av Eφ1,avEθ2,av ´ Eφ2,avEθ1,av (cid:52)Hθ1Eφ2,av ´ (cid:52)Hθ2Eφ1,av Eθ1,avEφ2,av ´ Eθ2,avEφ1,av (cid:52)Hθ1Eθ2,av ´ (cid:52)Hθ2Eθ1,av , Eφ1,avEθ2,av ´ Eφ2,avEθ1,av (cid:52)Eφ1Hφ2,av ´ (cid:52)Eφ2Hφ1,av Hθ1,avHφ2,av ´ Hθ2,avHφ1,av (cid:52)Eφ1Hθ2,av ´ (cid:52)Eφ2Hθ1,av Hφ1,avHθ2,av ´ Hφ2,avHθ1,av (cid:52)Eθ2Hφ1,av ´ (cid:52)Eθ1Hφ2,av Hθ1,avHφ2,av ´ Hθ2,avHφ1,av (cid:52)Eθ2Hθ1,av ´ (cid:52)Eθ1Hθ2,av Hφ1,avHθ2,av ´ Hφ2,avHθ1,av , , , . jω0 1 jω0 1 jω0 1 jω0 1 jωµ0 1 jωµ0 1 jωµ0 1 jωµ0 (20a) (20b) (20c) (20d) (20e) (20f) (20g) (20h) The corresponding metasurface exhibits birefringence, since it transforms the two polarizations differently. 5) Quadruple Transformation: In the absence of any con- straints, and particularly without requiring (10) to (12) – i.e. having a loss/gain, nonreciprocal and gyroropic structure – the spherical metasurface may achieve any arbitrary quadruple transformation10, as illustrated in Fig. 3. fiffiflT 4 equations »-–´(cid:52)Hφ1 ´(cid:52)Hφ2 ´(cid:52)Hφ3 ´(cid:52)Hφ4 " ‰„ ‰„ Eθ1,av Eθ2,av Eθ3,av Eθ4,av Eφ1,av Eφ2,av Eφ3,av Eφ4,av Hθ1,av Hθ2,av Hθ3,av Hθ4,av Hφ1,av Hφ2,av Hφ3,av Hφ4,av (21) χθθ ee χθφ ee `jk0 em χθφ χθθ em " jω0 " »-–Eθ1,av Eφ1,av Hθ1,av Hφ1,av Eθ2,av Eφ2,av Hθ2,av Hφ2,av Eθ3,av Eφ3,av Hθ3,av Hφ3,av Eθ4,av Eφ4,av Hθ4,av Hφ4,av whose solution is fiffiffifl " »--–jω0χθθ ee jω0χθφ ee jk0χθθ em jk0χθφ em fiffifl´1»-–´(cid:52)Hφ1 ´(cid:52)Hφ2 ´(cid:52)Hφ3 ´(cid:52)Hφ4 (22) 5   fiffifl . , D. Scattering Parameter Mapping In the holistic metasurface synthesis procedure described in Sec. I [29], [32], the susceptibility synthesis operation is followed by the determination of the physical metasurface structure via scattering parameter mapping. In the present case of a spherical metasurface, this mapping is different to that for the planar metasurface due to the different geometry. Instead considering wave scattering between two planar ports, we need to consider here scattering between two spherical-cap ports, as shown in Fig. 4. We will here only present the scattering parameter mapping procedure, without any specific physical metasurface design, which will be presented elsewhere with experimental results. Fig. 3. Illustration of a quadruple transformation, whereby the metasurface simultaneously and independently manipulates waves generated by four different sources. In this particular example, the four transformations are: 1) amplification, 2) refraction, 3) linear to circular polarization transformation, and 4) zero to nonzero orbital angular momentum transformation. In this case, Eqs (9) represent a system of 16 equations in 16 unknowns. For instance, the first line of (9a) splits into the 10In this case, a mathematical solution for the bianisotropic susceptibility tensor functions is always be found, but these mathematical functions will not necessarily be practically realizable. For instance, if the transformation is too drastic, the corresponding susceptibility functions may exhibit spatial variations that are too high for sampling by typical p « λ{5 particles. Or the required gain or/and nonreciprocity requirements may be unrealizable in the available technology, etc. Fig. 4. Unit cell spherical-cap port configuration for the scattering parameter mapping operation in the synthesis procedure. The scattering parameter mapping method consists in the following steps, followed in [4] for the case of planar meta- surface: 1) discretize the synthesized spherical susceptibility func- tions into subwavelength spherical-cap unit cells, as shown in Fig. 4, typically of size (lattice period) in the order of p « λ{5; transformation1transformation2transformation2transformation3transformation4PBCPBCPBCPBCmetasurfacePort1Port2△θ△φθ1φ1θ2φ26 ‚ First, we compute the susceptibility functions (8) corre- sponding to the specified fields using the general equa- tions (9), which is the essence of the synthesis procedure. ‚ Second, we validate this synthesis by comparing the fields scattered by the metasurfaces with such susceptibilities with the specified fields. The latter, which is an analysis operation, is performed by modelling the spherical surface susceptibility functions by volume-diluted susceptibility functions, following the proce- dure described in Appendix B, in the full-wave commercial full-wave FEM-based software COMSOL. The three examples will share the following features: ‚ Since COMSOL does not support bianisotropic media, i.e. assumes χem " χme " 0, the metasurface will be monoanisotropic13. ‚ Since COMSOL requires excessive memory for 3D sim- ulations, the metasurface will have variations only along one direction, corresponding to a transformation that is symmetric in the other direction. Specifically, the metasurface will belong to the category (c) in Fig. 2 (equivalent to category (b) upon π{2 rotation). assumption b) in the second paragraph of Sec. II. ‚ The metasurface will be reflection-less, according to ‚ For simplicity, and without loss of generality, the meta- surface will be surrounded by vacuum (1 " 2 " 0 and µ1 " µ2 " µ0). ‚ The sources will be infinitesimal vertical dipoles. ‚ Finally, the transformations will be specified in the far field. A. Illusion Transformation The first example is about illusion transformation. Specif- ically, the spherical metasurface is required to transform the field radiated by an off-centered source into the field produced by a virtual centered source. Assuming the source location px, y, zq " p0, 0, z0q, the corresponding incident field specification is Ei θr"a´ " ´ I(cid:96) jω0 4π a e´jk0?pa´ sin θq2`pa´ cos θ´z0q2 pa´ sin θq2 ` pa´ cos θ ´ z0q2 a e´jk0?pa´ sin θq2`pa´ cos θ´z0q2 pa´ sin θq2 ` pa´ cos θ ´ z0q2 k2 0 sin θ, (24a) jk0 sin θ, (24b) (25a) (25b) 2) for each unit cell, select a scattering particle geometry that is physically consistent with the synthesized sus- ceptibility at the corresponding point11; 3) compute the scattering parameters of that unit-cell, within periodic boundary conditions (PBCs) to approx- imate coupling between smoothly varying unit cells (assuming proper sampling), and between the two spherical-cap ports shown in Fig. 4 for polarizations corresponding to the tensorial nature of the synthesized susceptibility; 4) convert the resulting unit-cell scattering parameter func- tions to the corresponding susceptibility functions, as will be shown next by an example, and compare these functions with the synthesized susceptibility functions. 5) adjust the geometrical parameters of the physical unit cell until its susceptibility functions match the synthe- sized ones, and repeat this operation for all the unit cells; 6) combine the so-designed unit cells to form the final spherical metasurface structure. For instance, consider the monoanisotropic transforma- tion in Sec. III-C2, whose susceptibility functions are given by (16). After following steps 1) to 3) above, one needs to establish the proper conversion formulas for step 4). For this purpose, due to the absence of gyrotropy, we only need to consider the uncoupled orthogonal θ and φ ports, with reflection and transmission parameters Rθ " Er θ (S11 for θ input port), Tθ " Et θ (S21 for θ input and output ports), φ{Ei Rφ " Er φ (S21 for φ input and output ports)12. The sought after relations are found upon specifying the difference and average fields in terms of those parameters, for instance (cid:52)Hφ " TθHi φ ´p1´ RθqHi θs{2, into (16), which yields θ{Ei φ (S11 for φ input port) and Tφ " Et φ and Eθ,av " rTθEi θ `p1` RθqEi φ{Ei θ{Ei χθθ ee " ´2rTθ ´ p1 ´ Rθqs jω0ηpTθ ` 1 ` Rθq 2rTφ ´ p1 ´ Rφsq χφφ ee " jω0ηpTφ ` 1 ` Rφq 2ηrTφ ´ p1 ` Rφqs χθθ mm " jωµ0pTφ ` 1 ´ Rφq mm " ´2ηrTθ ´ p1 ` Rθqs χφφ jωµ0pTθ ` 1 ´ Rθq , , , . (23a) (23b) (23c) (23d) From this point, one proceeds to steps 5) and 6) above, which completes the synthesis. Hi φr"a´ " I(cid:96) 4π IV. ILLUSTRATIVE EXAMPLES In this section, we illustrate the spherical metasurface syn- thesis presented in Sec. III with the help of three examples in two steps: 11For instance, if the susceptibility at that point is non-gyrotropic and if one uses conducting particles, the particles should not include asymmetric bends, and one may then choose a straight cross or a Jerusalem cross, while if the susceptibility is chiral, one may choose a gammadion cross. tra scattering parameter computations, involving for instance Tθφ " Et i.e. S21 for φ input port and θ output port. 12If the two components were coupled, then one would need to perform ex- θ{Ei φ, while the transmitted field specification is ` e´jk0a 4πa` k2 Et θr"a` " ´T Ht φr"a` " T I(cid:96) I(cid:96) jω0 e´jk0a 4πa` jk0 sin θ, ` 0 sin θ, where I(cid:96) is the dipole moment (I: current, (cid:96): length) and k0 " ω{c is the free-space wavenumber (ω: angular frequency, c: velocity of light in vacuum) and T is the transmission 13We could naturally still have plotted the susceptibilities for such media, but these functions would not be very informative. coefficient. The incident and transmitted fields are related by the condition of local power conservation [30], 7 Fig. 6. Full-wave validation of the illusion transformation accomplished by the metasurface with susceptibilities plotted in Fig. 5. The 3D picture is obtained from revolving the 2D-computed fields about the z axis. Fig. 7. For such focusing, the total optical path from the source to the ring via any point P on the metasurface should be constant, namely ´jk0a ` ΦP ´ jk0dpθq " ´jk0ra ` dpθqs ` ΦP (28a) " const., a a2 ` c2 ´ 2ac sinpθq, with dpθq " (28b) where ΦPpθq " jk0ra ` dpθqs ` const. corresponds to the correction phase function to be provided by the metasurface. RerpEi θ H a T " RerpEt which sets the transmission coefficient to i,φ qs " 1 2 1 2 θ H t,φ qs, (26) a` . (27) pa´ sin θq2 ` pa´ cos θ ´ z0q2 mm ee " χφφ 14 plotted15 in Fig. 5 and χφφ Inserting these field specifications into (16) yields the sus- ceptibilities χθθ ee " mm " 0. The imaginary parts of susceptibilities are zero, χθθ which indicates that the metasurface performing the specified transformation is loss-less and gain-less. We see, with the help of the lattice in the inset of the top figure, that in this design, typical λ{5 particles or cells [29] can hardly sample the required susceptibility, except on the smooth parts of it. This issue may be resolved by increasing the electrical size the sphere or reduce the distance of the source to the center, if this is acceptable. Fig. 5. Susceptibility functions for the illusion transformation metasurface example with the parameters a " 10λ and z0 " 5λ. The normalized horizontal axis spans the entire elevation space, i.e. extends from θ " 0 to θ " π. The inset of the top figure shows the sampling of the susceptibility function with typical λ{5 scattering particles or periodic unit cells. Finally, Fig. 6 provides the full-wave validation of the metasurface synthesis, where the wave scattered from the metasurface clearly seems to be radiated by centered (virtual) source. Fig. 7. Focussing of the wave radiated by a centered point source on a ring of radius c. The corresponding incident and transmitted field specifica- tions are Ei θr"a´ " ´ Hi φr"a´ " I(cid:96) ´ e´jk0a 4πa´ k2 I(cid:96) jω0 e´jk0a 4πa´ jk0 sinpθq, ´ 0 sinpθq, (29a) (29b) (30a) (30b) (31) B. Ring Focusing The second example is a metasurface focusing the field radiated by a centered source onto a ring, as illustrated in and 14The equality is a result of the reflection-less specification, corresponding to electric and magnetic polarization currents canceling out at the input side (Huygens metasurface). 15The resulting relations are naturally closed-form expressions, but we do not give them here for the sake of brevity Et θr"a` " T ηejk0dpθq, Ht φr"a` " T ejk0dpθq, where T is found from (26) as I(cid:96) 4πa´ k0 sin θ. T " 051015202530-1001020304050051015202530-1001020304050χθθee(m)aθ{λrealrealimagimagχφφmm(m)aθ{λmeta-atomiclatticesampling01020-20-1001020-2-1012EV/mz{λρ{λmetasurfacemetasurfacerealsourcesourceVirtualad1cθ1focusedringspheremetasurfacePee " χθθ ee " χφφ mm plotted in Fig. 8 and χφφ Inserting these field specifications into (16) yields the sus- mm " 0. ceptibilities χθθ As expected from the symmetry of the transformation, these susceptibility functions are symmetry about the equator of the metasurface, i.e. at θ " π{2 or aθ{λ " 5π. Moreover, the susceptibilities are minimal in the vicinity of the equator where the required transformation is minimal given the doughnut radiation pattern of the vertical dipole source. Similar con- siderations is in the previous examples may be made about sampling. fields of virtual sources of the same nature place at the position of the other source, as illustrated in Fig. 10. 8 Fig. 10. Full-wave description and validation, for the susceptibilities plotted in Fig. 11, of a birefringent (double) transformation with electric and magnetic sources placed at px, y, zq " p0, 0, z1 " 5λ) and px, y, zq " p0, 0, z2 " ´5λ), respectively, and a " 10λ. Hi φ1r"a´ " I(cid:96)jk0 Ei θ1r"a´ " I(cid:96)jk0η 4π 4π a The fields corresponding to this transformations are e´jk0?pa´ sin θq2`pa´ cos θ´z1q2 pa´ sin θq2 ` pa´ cos θ ´ z1q2 a e´jk0?pa´ sin θq2`pa´ cos θ´z1q2 pa´ sin θq2 ` pa´ cos θ ´ z1q2 a e´jk0?pa´ sin θq2`pa´ cos θ´z2q2 pa´ sin θq2 ` pa´ cos θ ´ z2q2 a e´jk0?pa´ sin θq2`pa´ cos θ´z2q2 pa´ sin θq2 ` pa´ cos θ ´ z2q2 4π 4π Ei φ2r"a´ " IAηk2 0 Hi θ2r"a´ " ´IAk2 0 sinpθq, (32a) sinpθq, (32b) sinpθq, (33a) sinpθq (33b) and Et θ1r"a` " T1I(cid:96)η 4π Ht φ1r"a` " T1I(cid:96) 4π a e´jk0?pa` sin θq2`pa` cos θ´z2q2 pa` sin θq2 ` pa` cos θ ´ z2q2 a e´jk0?pa` sin θq2`pa` cos θ´z2q2 pa` sin θq2 ` pa` cos θ ´ z2q2 a e´jk0?pa´ sin θq2`pa´ cos θ´z1q2 pa´ sin θq2 ` pa´ cos θ ´ z1q2 a e´jk0?pa´ sin θq2`pa´ cos θ´z1q2 pa´ sin θq2 ` pa´ cos θ ´ z1q2 4π 4π jk0 sinpθq, (34a) jk0 sinpθq, (34b) sinpθq, (35a) sinpθq (35b) Et φ2r"a` " T2IAηk2 0 Ht θ2r"a` " ´T2IAk2 0 Fig. 8. Susceptibility functions for the ring focusing transformation illustrated in Fig. 7 with the same parameters as in Fig. 5. Finally, Fig. 9 provides the full-wave validation of the metasurface synthesis, where the wave scattered from the metasurface clearly focusses on the specified ring region. Fig. 9. metasurface with the susceptibilities plotted in Fig. 8. Full-wave validation of the ring focusing accomplished by the C. Birefringence The third example is a birefringent (double-transformation) metasurface transforming the fields radiated by two off- centered orthogonal electric and magnetic sources into the 051015202530-15-10-50510051015202530-15-10-50510χθθee(m)aθ{λrealrealimagimagχφφmm(m)aθ{λmeta-atomiclattice5π010-15-10-5051015-4-2024EV/mz{λρ{λmetasurfacemetasurfaceFocusedring010-15-10-5051015-0.2-0.100.10.2010-15-10-5015-1-0.500.51510electricdipolez{λρ{λEV/mmagneticdipolez{λρ{λEV/mwith d d T 1 " T 2 " pa` sin θq2 ` pa` cos θ ´ z2q2 a´ sin θq2 ` pa´ cos θ ´ z1q2 , pa` sin θq2 ` pa` cos θ ´ z1q2 a´ sin θq2 ` pa´ cos θ ´ z2q2 . (36a) (36b) Inserting these field specifications into (20) yields the four susceptibility functions plotted in Fig. 11, whose symmetry is again expected. Fig. 11. Susceptibility functions for the metasurface in Fig. 10. The full-wave simulation results, in Fig. 10, are in perfect agreement with the expectation. V. CONCLUSION This paper has extended the susceptibility-GSTC synthesis of planar metasurfaces to spherical metasurfaces. In contrast to 9 the cylindrical metasurface that has been the first non-planar metasurface modeled by susceptibility-GSTC, the spherical metasurface is has a non-zero intrinsic curvature and hence exhibits particularly interesting characteristics, some of which have been pointed out. The paper paves the way for the study of other canonical-shape metasurfaces of non-zero intrinsic curvature, and prompts for the exploration of irregular shaped metasurfaces combining GSTCs with conformal mapping tech- niques. REFERENCES [1] T. J. Cui, D. R. Smith, and R. Liu, Metamaterials. Springer, 2010. [2] 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, vol. 334, no. 6054, pp. 333– 337, 2011. [3] C. L. Holloway, E. F. Kuester, J. A. Gordon, J. O'Hara, J. Booth, and D. R. Smith, "An overview of the theory and applications of metasurfaces: The two-dimensional equivalents of metamaterials," IEEE Trans. Antennas Propag., vol. 54, no. 2, pp. 10–35, April 2012. [4] K. Achouri, M. A. Salem, and C. Caloz, "General metasurface synthesis based on susceptibility tensors," IEEE Trans. Antennas Propag., vol. 63, no. 7, pp. 2977–2991, July 2015. [5] S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S. Kivshar, and C. R. Simovski, "Metasurfaces: From microwaves to visible," Phys. Rep., vol. 634, pp. 1–72, 2016. [6] D. Lin, P. Fan, E. Hasman, and M. L. Brongersma, "Dielectric gradient metasurface optical elements," Science, vol. 345, no. 6194, pp. 298–302, 2014. [7] S. Larouche, Y.-J. Tsai, T. Tyler, N. M. Jokerst, and D. R. Smith, "Infrared metamaterial phase holograms," Nat. Mater., vol. 11, no. 5, pp. 450–454, 2012. [8] L. Chen, K. Achouri, E. Kallos, and C. Caloz, "Simultaneous enhance- ment of light extraction and spontaneous emission using a partially reflecting metasurface cavity," Phys. Rev. A, vol. 95, no. 5, p. 053808, 2017. [9] J. Ortiz, J. Baena, V. Losada, F. Medina, and J. Araque, "Spatial angular filtering by fsss made of chains of interconnected srrs and csrrs," IEEE Microw. Wirel. Compon. Lett., vol. 23, no. 9, pp. 477–479, 2013. [10] Y. Ra'di, V. S. Asadchy, and S. A. Tretyakov, "Total absorption of electromagnetic waves in ultimately thin layers," IEEE Trans. Antennas Propag., vol. 61, no. 9, pp. 4606–4614, Sept 2013. [11] K. Achouri, G. Lavigne, M. A. Salem, and C. Caloz, "Metasurface spatial processor for electromagnetic remote control," IEEE Trans. Antennas Propag., vol. 64, no. 5, pp. 1759–1767, 2016. [12] A. Silva, F. Monticone, G. Castaldi, V. Galdi, A. Al`u, and N. Engheta, "Performing mathematical operations with metamaterials," Science, vol. 343, no. 6167, pp. 160–163, 2014. [13] K. F. MacDonald, Z. L. S´amson, M. I. Stockman, and N. I. Zheludev, "Ultrafast active plasmonics," Nat. Photonics, vol. 3, no. 1, pp. 55–58, 2009. [14] A. Ishimaru, S. Jaruwatanadilok, and Y. Kuga, "Generalized surface plasmon resonance sensors using metamaterials and negative index materials," Prog. Electromagn. Res., vol. 51, pp. 139–152, 2005. [15] U. Leonhardt and T. Philbin, Geometry and light: the science of invisibility. Courier Corporation, 2012. [16] S. Pandi, C. A. Balanis, and C. R. Birtcher, "Curvature modeling in design of circumferentially modulated cylindrical metasurface lwa," IEEE Antenn. Wireless Propag. Lett., vol. 16, pp. 1024–1027, 2017. [17] B. O. Raeker and S. M. Rudolph, "Verification of arbitrary radiation pat- tern control using a cylindrical impedance metasurface," IEEE Antenn. Wireless Propag. Lett., vol. 16, pp. 995–998, 2017. [18] A. Shahi and A. Abdolali, "Cylinder scattering pattern manipulation for dual-polarized radar cross-section reduction using inhomogeneous metasurface," in 2017 Iranian Conference on Electrical Engineering (ICEE), May 2017, pp. 1691–1696. [19] B. O. Raeker and S. M. Rudolph, "Arbitrary transformation of radi- ation patterns using a spherical impedance metasurface," IEEE Trans. Antennas Propag., vol. 64, no. 12, pp. 5243–5250, 2016. [20] J. C. Soric, R. Fleury, A. Monti, A. Toscano, F. Bilotti, and A. Al`u, "Controlling scattering and absorption with metamaterial covers," IEEE Trans. Antennas Propag., vol. 62, no. 8, pp. 4220–4229, Aug 2014. 051015202530-20-1001020051015202530-20-1001020051015202530-20-1001020051015202530-20-1001020χθθee(m)aθ{λaθ{λaθ{λaθ{λχφφee(m)χθθmm(m)χφφmm(m)realrealrealrealimagimagimagimagmeta-atomiclattice[21] E. F. Kuester, M. A. Mohamed, M. Piket-May, and C. L. Holloway, "Averaged transition conditions for electromagnetic fields at a metafilm," IEEE Trans. Antennas Propag., vol. 51, no. 10, pp. 2641–2651, 2003. [22] M. Idemen, "Discontinuities in the electromagnetic field," IEEE Trans. Antennas Propag., vol. 56, no. 1, pp. 300–300, 2011. [23] M. Safari, A. Abdolali, H. Kazemi, M. Albooyeh, M. Veysi, and F. Capolino, "Cylindrical metasurfaces for exotic electromagnetic wave manipulations," in 2017 IEEE Int. Symp. on Antennas Propag. USNC/URSI National Radio Science Meeting, July 2017, pp. 1499– 1500. [24] P. van den Berg and J. Fokkema, "The rayleigh hypothesis in the theory of diffraction by a cylindrical obstacle," IEEE Trans. Antennas Propag., vol. 27, no. 5, pp. 577–583, Sep 1979. [25] J. W. S. B. Rayleigh, The theory of sound. Macmillan, 1896, vol. 2. [26] E. J. Rothwell and M. J. Cloud, Electromagnetics. CRC press, 2008. [27] N. Chamanara, Y. Vahabzadeh, K. Achouri, and C. Caloz, "Spacetime processing metasurfaces: Gstc synthesis and prospective applications," in 2016 IEEE Int. Symp. on Antennas Propag. (APSURSI), June 2016, pp. 365–366. [28] K. Achouri, Y. Vahabzadeh, and C. Caloz, "Mathematical synthesis and analysis of nonlinear metasurfaces," arXiv preprint, arXiv:1703.09082, 2017. [29] K. Achouri and C. Caloz, "Design, concepts and applications of elec- tromagnetic metasurfaces," arXiv preprint, arXiv:1712.00618, 2017. [30] G. Lavigne, K. Achouri, V. Asadchy, S. Tretyakov, and C. Caloz, "Sus- ceptibility derivation and experimental demonstration of refracting meta- surfaces without spurious diffraction," arXiv preprint, arXiv:1705.09286, 2017. [31] V. Asadchy, M. Albooyeh, S. Tcvetkova, A. D´ıaz-Rubio, Y. Ra'di, and S. Tretyakov, "Perfect control of reflection and refraction using spatially dispersive metasurfaces," Phys. Rev. B, vol. 94, no. 7, p. 075142, 2016. [32] Y. Vahabzadeh, N. Chamanara, K. Achouri, and C. Caloz, "Compu- tational analysis of metasurfaces," arXiv preprint, arXiv:1710.11264, 2017. [33] M. Albooyeh, S. Tretyakov, and C. Simovski, "Electromagnetic charac- terization of bianisotropic metasurfaces on refractive substrates: General theoretical framework," Ann. Phys. (Berl.), vol. 528, no. 9-10, pp. 721– 737, 2016. APPENDIX A. GSTCs Derivation Since GSTCs are local, they are the same for curved metasurfaces, including the spherical metasurface of interest here, as for planar metasurface16. They may be derived in different fashions. The most rigorous one, allowing for any discontinuity order, is given by Idemen in [22] and clarified in the appendix of [4]. However, this approach is mostly math- ematical and does not clearly reveal how to take into account the (possibly different) media surrounding the metasurface. An alternative approach, more physical and classical, is given in [33] for plane waves. We present here a derivation that is in the vein of [33], but that is more general, starting from Maxwell and constitutive relations, involving both volume and surface polarization densities, and applying to any type of waves. Assuming time-harmonic (e`jωt) waves, Maxwell equations take the form ∇ E " ´jωB ´ Kimp, ∇ H " jωD ` Jimp, where Jimp (A/m2) is the impressed electric current density and Kimp (V/m2) is the (fictitious) impressed magnetic current density, and where the fields and are related to the electric 16This is naturally under the assumption of Footnote 1. symmetric (37a) (37b) polarization density P and the magnetic polarization density M by the constitutive relations 10 D " 0E ` P, B " µ0pH ` Mq, or E " pD ´ Pq{0, or H " B{µ0 ´ M, (38a) (38b) with P and M measured in C/m2 and A/m, respectively. The essence of GSTCs is to model the metasurface as a thin sheet of equivalent polarization currents. Therefore, the fields D and E have to be written in terms of P, as in (38a), and B and H have to be written in terms of M, as in (38b). Inserting these relations into (37) yields ∇ rpD ´ Pq{0qs " ´jωµ0pH ` Mq ´ Kimp, ∇ pB{µ0 ´ Mq " jωp0E ` Pq ` Jimp. (39a) (39b) Assuming first-order metasurface discontinuity [4], the po- larization densities decompose into volume and surface parts as Inserting (40) into (39), and transferring the surface parts to the right-hand sides, yields D ´ Pv 0 P " Pv ` Psδprq, M " Mv ` Msδprq. „  " ´jωµ0H ´ jωµ0Mv Psδprq ´jωµ0Ms ` ∇ 0 (40a) (40b) ´ Kimp, (41a) (41b) ∇ ∇ B µ0 ´ Mv " jω0E ` jω0Pv `jω0Ps ` ∇ rMsδprqs ` Jimp. The metasurface discontinuity may now be analyzed by integrating (41) over the usual rectangular surface around the interface between the two media, that supports here the meta- surface, as shown in Fig. ??, and applying Stokes theorem. This yields ¿ ij ij 0 D ´ Pv ¿ ¿ „ δprqMs dS ` ¿ B µ0 ´ Mv δprqPs dS ` δprqPs ij pH ` Mvq dS Kimp dS, (42a)  dl " ´jωµ0 ij dl ´ ij pE ` Pvq dS Jimp dS. dl " jω0 rMsδprqs dl ` 0 ´jωµ0 `jω0 ij (42b) The integrands in the left-hand sides of (42) are, according to (38), nothing but the electric and magnetic fields in the two media, that be simply written as D ´ Pv 0 " E, B µ0 ´ Mv " H. (43) 11 Combining (46) and (47) finally yields r (cid:52)E " ´jωµ0Ms,} ` ∇}pPs,r{0q r ´ K},s,imp, (48a) (48b) r (cid:52)H " jωPs,} ´ r ∇}Ms,r ` J},s,imp, where the symbol } denotes the metasurface tangential com- ponents ξ and ζ in this appendix. Relations (48) are the final GSTC equations. To clearly see how Eqs. (48) account for different surround- ing media, as in Fig. ??, we rewrite them explicitly as r pE` ´ E´q " ´jωµ0Ms,} ` ∇}pPs,r{0q r ´ K},s,imp, (49a) r pE`{η2,eff ´ E´{η1,effq " jωPs,} ´ r ∇}Ms,r ` J},s,imp, (49b) with the effective impedances ηi,eff " ηi{ cos θi for TEr and ηi,eff " ηi cos θi for TMr (i " 1, 2). Finally, the surface polarization densities (6) may also be written in terms of the impedances of the surrounding media as a P " 0 ¯¯χeepE` ` E´q{2 ` ?µ00 ¯¯χempE`{η2,eff ` E´{η1,effq{2, 0{µ0 ¯¯χmepE` ` E´q{2 ` ¯¯χmmpE`{η2,eff ` E´{η1,effq{2. M " (50b) So, the information on the media surrounding the metasurface is not explicitly apparent in the usual form (48) of the GSTCs, incident but explicitly appears in the specified input (`: and reflected) and output (´: transmitted) fields, which must obviously be specified consistently with Maxwell equations. (50a) B. Derivation of Volume Equivalent Susceptibility No softwave is currently available to simulate curved meta- surfaces with zero thickness and hence, particularly, zero- thickness spherical metasurfaces. Therefore, we present here a technique allowing to model spherical surface susceptibilities by volume-diluted susceptibilities in a deeply subwavelength spherical shell in order to validate the synthesis presented in Sec. III. For simplicity, and without loss of generality, consider the case of an isotropic metasurface, with electric susceptibility χee. In this case, Maxwell-Amp`ere equation reads ∇ H " jω0p1 ` χeeqE. (51) The sought after modeling can be found by integrating this equation for both a metasurface sheet and a subwavelength shell, as shown in Fig. 13, and equating the results. In the case of the metasurface sheet [Fig. 13(a)], we have χee " χ2Dδpr ´ aq, and Eq. (51) integrates to ij ¿ H dl " jω0 r1 ` χ2Dδpr ´ aqsE dS, which yields pH`θ l2 ´ H´θ l1q " jω0ph ` χ2DqEφ l2 ` l1 2 , (52) (53) where the elevation distance has been taken as the average of the elevation distances on both sides of the metasurface (l1 Fig. 12. General curved boundary, supporting a metasurface, surrounded by two media with permittivity-permeability pairs p1, µ1q and p2, µ2q, respectively, with local coordinate system pξ, ζ, rq and rectangular integration surface for (41) with Ψ " E, H, D, B, P, M, labeled ´ at r " a´ (just below the metasurface in medium 1) and ` at r " a` (just above the metasurface in medium 2). With this, the ξ ´ r projection of Eqs. (42) integrate to pE`ξ ´ E´ξ q(cid:52)ξ ` p´Er,right ` Er,leftq(cid:52)r " ´ jωµ0pHζ ´ Mv,ζq(cid:52)ξ(cid:52)r ´ jωµ0Ms,ζδprq(cid:52)ξ(cid:52)r ´ pPs,r,right ´ Ps,r,leftqδprq(cid:52)r{0 ´ Kζ,v,imp(cid:52)ξ(cid:52)r ´ Kζ,s,impδprq(cid:52)ξ(cid:52)r, (44a) pH`ξ ´ H´ξ q(cid:52)ξ ` p´Hr,right ` Hr,leftq(cid:52)r "jω0pEζ ` Pv,ζq(cid:52)ξ(cid:52)r ` jω0Ps,ζδprq(cid:52)ξ(cid:52)r ´ pMs,r,right ´ Ms,r,leftqδprq(cid:52)r ´ Jζ,v,imp(cid:52)ξ(cid:52)r ´ Jζ,s,impδprq(cid:52)ξ(cid:52)r. (44b) Taking the limit (cid:52)r Ñ 0, replacing δprq(cid:52)r Ñ 1, and dividing by (cid:52)ξ, Eqs. (44) reduce to pE`ξ ´ E´ξ q " ´ jωµ0Ms,ζ pH`ξ ´ H´ξ q "jω0Ps,ζ ´ pPs,r,right ´ Ps,r,leftq{p0(cid:52)ξq ´ Kζ,s,imp, ´ pMs,r,right ´ Ms,r,rightq{(cid:52)ξ ` Jζ,s,imp, which, in the limit (cid:52)ξ Ñ 0, may be written as ´ Kζ,s,imp, (cid:52)Eξ " ´jωµ0Ms,ζ ´ BpPs,r{0q Bξ with (cid:52)Eξ " E`ξ ´ E´ξ , (cid:52)Hξ " jω0Ps,ζ ´ BMs,r Bξ ` Jζ,s,imp, with (cid:52)Hξ " H`ξ ´ H´ξ . Similarly, we find for the ζ ´ r projection of Eqs. (42) ` Kξ,s,imp, (cid:52)Eζ " jωµ0Ms,ξ ´ BpPs,r{0q Bζ with (cid:52)Eζ " E`ζ ´ E´ζ , (cid:52)Hζ " ´jω0Ps,ξ ´ BMs,r with (cid:52)Hζ " H`ζ ´ H´ζ . Bζ ´ Jξ,s,imp, (45a) (46a) (46b) (46c) (46d) (47a) (47b) (47c) (47d) ǫ2,µ2ǫ1,µ1ξζrΨ`Ψ´Ψ2Ψ1Ψ1△ξ△r12 Fig. 13. Integration parameters to derive the equivalence between surface and volume susceptibilities for a spherical metasurface: Eq. (56). (a) Ideal metasurface sheet with zero thickness (t " 0). (b) Corresponding metasurface shell with sub-wavelegnth thickness t ! λ. and l2). In the case of the metasurface shell [Fig. 13(b)], we have ¿ χee " χ3DΠrpr ´ aq{ts, where Πpq is the rectangular pulse function, and Eq. (51) integrates to ij H l " jω0 t1 ` χ3DΠrpr ´ aq{tsu E dS, which yields pH`θ l2 ´ H´θ l1q " jω0ph ` χ3DtqEφ l2 ` l1 2 . (54) (55) Equating (53) and (55), provides the surface-equivalent volume susceptibility χ3D " χ2D{t, (56) corresponding to the permittivity " 1` χee{t, and then also permeability µ " 1 ` χmm{t, which may be straightforwardly generalized to the anisotropic case in COMSOL. rrθθφφχee"χ2Dδpr´aqχee"χ3DΠpr´atql1l1hhl2l2t"0t!λ(a)(b)
1906.04677	1	1906	2019-06-11T16:17:44	Probing weakly hybridized magnetic molecules by spin-polarized tunneling	[ "physics.app-ph" ]	Advances in molecular spintronics rely on the in-depth characterization of the molecular building blocks in terms of their electronic and, more importantly, magnetic properties. For this purpose, inert substrates that interact only weakly with adsorbed molecules are required in order to preserve their electronic states. Here, we investigate the magnetic-field response of a single paramagnetic 5,5-dibromosalophenatocobalt(II) (CoSal) molecule adsorbed on a weakly interacting magnetic substrate, namely Fe-intercalated graphene (GR/Fe) grown on Ir(111), by using spin-polarized scanning tunneling microscopy and spectroscopy (SP-STM/STS). We have obtained local magnetization curves, spin-dependent tunneling spectra, and spatial maps of magnetic asymmetry for a single CoSal molecule, revealing its magnetic properties and coupling to the local environment. The distinct magnetic behavior of the Co-metal center is found to rely strictly on its position relative to the GR/Fe moire structure, which determines the level of hybridization between the GR/Fe surface pi-system, the molecular ligand pi-orbitals and the molecular Co-ion d-orbital.	physics.app-ph	physics	Probing weakly hybridized magnetic molecules by spin-polarized tunneling Emil Sierda1,2,, Micha Elsebach1, Roland Wiesendanger1 and Maciej Bazarnik1,2 1Dept. of Physics, University of Hamburg, Jungiusstrasse 11, D-20355 Hamburg, Germany 2Institute of Physics, Poznan University of Technology, Piotrowo 3, 60-965 Poznan, Poland [email protected] Advances in molecular spintronics rely on the in-depth characterization of the molecular building blocks in terms of their electronic and, more importantly, magnetic properties. For this purpose, inert substrates that interact only weakly with adsorbed molecules are required in order to preserve their electronic states. Here, we investigate the magnetic-field response of a single paramagnetic 5,5'-dibromosalophenatocobalt(II) (CoSal) molecule adsorbed on a weakly interacting magnetic substrate, namely Fe-intercalated graphene (GR/Fe) grown on Ir(111), by using spin-polarized scanning tunneling microscopy and spectroscopy (SP-STM/STS). We have obtained local magnetization curves, spin-dependent tunneling spectra, and spatial maps of magnetic asymmetry for a single CoSal molecule, revealing its magnetic properties and coupling to the local environment. The distinct magnetic behavior of the Co-metal center is found to rely strictly on its position relative to the GR/Fe moiré structure, which determines the level of hybridization between the GR/Fe surface π-system, the molecular ligand π-orbitals and the molecular Co-ion d-orbital. Molecular-based systems are promising candidates for nanospintronic devices, the main examples being single molecular magnets (SMMs) [Bogani, 2008, Schwöbel, 2012, Dreiser, 2016, Gragnaniello, 2017, Paschke, 2019], phthalocyanines [Iacovita, 2008, Javaid, 2010, Atodiresei, 2010, Brede, 2010, 2012, Mugarza, 2011, Avvisati, 2018, 2018, Czap, 2019], and carbon-based magnetic heterostructures [Gambardella, 2009, Kawahara, 2012, Brede, 2014]. In most of these studies the molecules' orbitals were strongly hybridized with the substrates' electronic states [Schwöbel, 2012, 1 Iacovita, 2008, Javaid, 2010, Atodiresei, 2010, Brede, 2010, 2012, Mugarza, 2011, Gambardella, 2009, Kawahara, 2012, Brede, 2014, Czap, 2019]. Based on this fact, there have been concerns about how to separate contributions of the molecules and the substrate to the measured magnetic signal. Recent experiments performed with molecules adsorbed on inert graphene (GR) [Dreiser, 2016, Gragnaniello, 2017, Paschke, 2019] have attempted to reduce the effect of hybridization on the magnetism of the molecule. However, these studies involved either spatially averaging techniques [Dreiser, 2016, Gragnaniello, 2017] or made use of inelastic spin excitations [Paschke, 2019] which do not address the spatial distribution of spin-polarized states of single adsorbed molecules. At the same time, spatially averaging techniques indicate that a Co-intercalated graphene (GR/Co) substrate exhibits significant hybridization with adsorbed molecules [Avvisati, 2018, 2018]. However, spatial variations of the electronic properties of intercalated graphene, due to its moiré structure [Decker, 2014], can result in a locally low level of substrate-molecule hybridization [Bazarnik, 2013]. In parallel, there have been a number of experimental as well as theoretical studies on the possibility to create planar and all-spin logic devices (SLD) using salophene-based molecules [DiLullo, 2012, Bazarnik, 2016, Sierda, 2017, García-Fernandez, 2017]. Here we take another step towards to this ultimate goal and demonstrate the detection and manipulation of the direction of the magnetic moment (µm) of individual CoSal molecules, based on the low level of hybridization with a GR/Fe substrate. The substrate stabilizes µm and provides a stable magnetic reference required for SP- STM/STS studies [Wiesendanger, 2009]. For the following investigations, we have intentionally prepared a spatially heterogeneous sample system consisting of areas of pristine GR/Ir(111) as well as areas of GR/Fe. The subsequently deposited CoSal molecules were found only on the latter regions and prefer to adsorb along step edges, followed by step-flow growth leading to molecular assemblies located near the step edges. The moiré structure of the GR/Fe substrate [Decker, 2014] leads to an additional preferential adsorption of the CoSal molecules away from the moiré's top sites. This behavior can be explained by the combined action of intermolecular Van-der-Waals forces and repulsive dipolar interactions between molecules and the substrate which drive the molecules away from the moiré's top sites and influence their 2 arrangement relative to one another [Bazarnik, 2013]. An example for such a molecular assembly consisting of 27 individual molecules is presented in Fig. 1a. In this STM topography image, starting from the left, we can observe an area of GR/Fe, an area of CoSal/GR/Fe and, separated by an upward step-edge, an area of GR/Ir(111). A schematic cross-section of the sample structure is provided in Fig. 1b. Its location within the STM image of Fig. 1a is marked by a white dashed line. At a sample bias voltage of U = +450 mV (as used for recording the data in Fig. 1a) electronic states located on the Co- metal centers of the molecules are dominant. Structural models of the molecules and the positions of the moiré top sites are superimposed onto the STM data in the lower part of Fig. 1a. Next, we focused on an area around a single molecule, as marked by a white square in Fig. 1a. Fig. 1c and Fig. 1d show topographic STM images of that region obtained with two different sample bias voltages: U = +50 mV and U = +450 mV, respectively. In order to enhance the spatial resolution, a molecule-terminated probe tip was used to obtain those images. By tunneling at a low bias of U = +50 mV (within the CoSal HOMO-LUMO gap) with such a probe tip, we can obtain intramolecular resolution of the CoSal molecule [Repp, 2005]. One can easily identify the protrusion in the middle of the molecule originating from the center Co atom, two protrusions arising from the Br end atoms and the molecular backbone formed by the rest of the atoms. A schematic drawing of that area is shown in Fig. 1e, where ball-and-stick models of the center molecule as well as of other surrounding molecules are depicted on top of a model of the GR/Fe substrate's atomic structure (red and black lines representing the GR and Fe lattices, respectively). The particular molecule we are focusing on in the following is adsorbed in between the hcp and fcc areas of the GR/Fe moiré structure. This area appears slightly protruded, creating a saddle-type line from one top-site of the moiré to another. The CoSal molecule of our focus is adsorbed slightly off the saddle line towards the hcp region. It is of central importance for the discussion later that this position exhibits a lower spin polarization of the electronic states than neighboring hcp and fcc regions [Decker, 2014]. Based on the bias-dependent differential tunneling conductance (dI/dU) maps we concluded that the maximum signal at the Co-centers for most of the molecules within the assembly is observed at U = 3 +450 mV. Therefore, we selected this particular bias to map the spin-resolved dI/dU signal as a function of an externally applied magnetic field (B = 0.75 T → 5.25 T → -5.25 T → 0.75 T) by using a magnetic SP-STM probe tip [Wiesendanger, 2009]. Based on that data we have obtained the spin- resolved differential tunneling conductance signal (G) as a function of magnetic field (details in supplementary information). It is spatially averaged over two areas; one including the position of the molecule's Co atom and the other covering an hcp area of GR/Fe. The resulting locally measured magnetization curves [Meier, 2008] are shown in Fig. 2a. For the interpretation of the measured data, it is important to note that the SP-STM probe tip used for these experiments is magnetically soft and requires only B = 0.2 T in order to fully align its magnetization direction with 𝐵⃗ . Therefore, while crossing B = 0 T one will always observe a change in the spin-resolved G-signal intensity due to a change of the SP-STM probe tip's magnetization direction. The measured values of G obtained for GR/Fe as a function of 𝐵⃗ -field reveal a characteristic magnetic hysteresis. There are two changes of the substrate's magnetization direction occurring: after increasing the 𝐵⃗ -field from 4.5 T to 5.25 T and decreasing from -4.5 T to -5.25 T. This behavior is expected for these 𝐵⃗ -field values as reported by Decker et al. [Decker, 2014]. The GR/Fe loop is inverted here due to the different sample bias voltage used to obtain the data. The deduced GR/Fe magnetization directions are  for forward [-5.25 T, 4.5 T] and  for backward [-4.5 T, 5.25 T] 𝐵⃗ -field sweeps (see Fig. 2a). In contrast, the CoSal response to the 𝐵⃗ -field variation is markedly different. It is still mirror-symmetric with respect to B = 0 T and is affected by the above-mentioned changes of the GR/Fe magnetization direction. However, we can observe a very pronounced increase in G-signal as the 𝐵⃗ -field rises from 1.5 T to 3.75 T. This suggests that the z-component of the magnetic moment µz of the molecule's Co-center is aligning with the 𝐵⃗ -field. Schematic drawings indicating the magnetization directions of all parts of the magnetic tunnel junction for B = 0.75 T, 3.75 T, and 5.25 T are provided in Fig. 2b. Upon reaching the value of the 𝐵⃗ -field for which the GR/Fe magnetization direction changes, the magnetic moment of the CoSal molecule aligns fully with the direction of the substrate's magnetization and the external magnetic field. A drop in G-signal intensity is strictly connected to changes of the substrate's magnetization. Subsequently, the external magnetic field is lowered and no significant changes are 4 observed until B = 0 T. Upon the SP-STM tip's magnetization reversal the behavior described above is repeated for the opposite 𝐵⃗ -field direction. Two spin configurations as outlined in Fig. 2b (for B = 0.75 T and 3.75 T) have been used for further SP-STS experiments in order to visualize the spin-dependent local density of states (LDOS) distributions and the differences between them. Before each SP-STS curve has been recorded, we stabilized the SP-STM probe tip above a non-magnetic part of the substrate, i.e. bare GR/Ir(111) (without the Fe-intercalation layer), in order to guarantee a constant and spin-polarization independent sample -- tip separation [Kubetzka, 2003]. The SP-STS data presented in Fig. 3a was measured on the Co-center of the CoSal molecule depicted in Fig. 1c at B-field values of 1.5 T and 4.5 T. The changes in the measured spectra reflect the behavior observed for the magnetization curves in Fig. 2a, and one can clearly distinguish three bias regions for which spin-dependent tunneling effects are most pronounced: one around U = -250 mV, one around the peak at U = +500 mV, and another around U = +1 V. It is important to note that at the same time, the SP-STS data on bare GR/Ir(111) (see Fig. 3b) does not show any changes when comparing the curves acquired at different 𝐵⃗ -field values. Hence, one can conclude that the changes observed for the CoSal molecule originate purely from its response to the external 𝐵⃗ -field. The change in the spectra of CoSal does not induce any change in the spectra of GR/Fe. A similar experiment in which the magnetization direction of GR/Fe has been inverted is presented in the supplementary information (Fig S1). There, one can clearly see a change in the SP- STS data of GR/Fe and how that influences the behavior of CoSal. After revealing the energetic positions of the spin-polarized states, we map their spatial distributions by spin-resolved dI/dU maps in Fig. 3c-e (HOMO at U = -250 mV, LUMO at U = +450 mV, and LUMO+1 at U = +950 mV, respectively) and their corresponding spin asymmetry in Fig. 3f-h. The spin asymmetry distribution for U = -250 mV is close to zero, for U = +450 mV it is of positive sign and particularly strong over the molecules' center atoms, while for U = +950 mV it is of negative sign and localized in the same area. 5 The single-molecule magnetization curves recorded in the present study and displayed in Fig. 2a are only part of a full hysteresis loop, i.e. magnetic saturation is not reached, which is caused by two factors. On the one hand, the GR/Fe magnetization direction changes in relatively low 𝐵⃗ -fields and CoSal follows this behavior. On the other hand, using a Brillouin function we have estimated that the 𝐵⃗ -field needed to fully align the molecule's magnetic moment (assuming it preserves spin ½) at our measurement temperature of 6.5 K would be as high as 29 T. In order to determine the full magnetization curve, the measurements would have to be performed at temperatures of ~1 K or lower for the experimentally accessible 𝐵⃗ -field (up to 6 T). At low 𝐵⃗ -field the molecule's magnetic moment is stabilized by superexchange interactions with the substrate. The same explanation has also been proposed for the similar system of phthalocyanine molecules adsorbed on a GR/Co substrate [Avvisati, 2018, 2018]. The spatially averaging techniques as used by Avvisati et al. suggested either ferro- or antiferromagnetic interactions of the molecule with the substrate, depending on the type of metal centers of different molecules. However, by using local probe techniques, we show here that the different adsorption sites are non-equivalent and exhibit different degrees of magnetic interaction strengths between molecule and substrate. Hence, it is possible to manipulate the molecule's magnetic moment by an external 𝐵⃗ -field. Based on previously reported DFT calculations for CoSal molecules we expect the dxz-orbital to carry a non-paired electron [DiLullo, 2012, Bazarnik, 2016]. Its geometry can lead to both ferro- and antiferromagnetic superexchange coupling between the GR/Fe substrate and the CoSal molecule. The latter is observed in our experiment. Therefore, its path is as follows: GR C π → CoSal O π + N π → Co dxz. The strength of this interaction can vary depending on the number of CoSal π-orbitals effectively interacting with the GR π-system. That is the reason why a sufficiently strong 𝐵⃗ -field can overcome this interaction and act on the molecule's magnetic moment in some of the adsorption configurations, including the one presented here. Following the equipartition theorem, the average thermal kinetic energy for this system is Ek(z) = ½kBT = 0.28 meV. The difference in the Zeeman energy for B = 1.5 T and 4.5 T, i.e. magnetic fields for which the SP- STS data of Fig. 3a has been obtained, is EZ = 0.3 meV. As EZ is comparable to Ek(z), no spin splitting is expected to be observed and Fig. 3a shows that indeed no spin splitting is visible. The spin 6 asymmetry observed in Fig. 3c-e is low because, as discussed above, we only partly align the magnetic moment of the CoSal molecule with the 𝐵⃗ -field. The effect is small and affecting only one direction (normal to the substrate). In Fig. 3f an almost vanishing spin asymmetry is observed, caused by a low spin polarization of the electronic states at that particular bias voltage. The signs of the spin asymmetries observed in Fig. 3g and Fig. 3h are in agreement with the SP-STS differences as revealed in Fig. 3a. They indicate that the two spin-polarized molecular orbitals located on the N, O and Co atoms of CoSal exhibit two directions with respect to GR/Fe: antiparallel originating from Co d orbitals and parallel originating from the O π and N π orbitals. In conclusion, we have observed spin-polarized tunneling to molecular orbitals of a paramagnetic molecule adsorbed on a ferromagnetic GR/Fe substrate. The distinct adsorption site and the effectively low hybridization resulted in relatively weak magnetic coupling of CoSal with the GR/Fe enabling us to address the spin-dependent behavior of the molecule independent from the substrate. Moreover, we mapped the spin polarization distribution of two molecular orbitals by bias-dependent SP-STS experiments. Being able to manipulate and read-out the magnetic state of individual adsorbed molecules independent from the substrate's magnetic state will be of great importance for the design and realization of molecular spintronic devices. Methods All experiments were performed in a UHV system equipped with a low temperature SP-STM and two preparation chambers: one for substrate cleaning and CVD growth and another for molecule as well as metal deposition [Wittneven, 1997]. Electrochemically etched tungsten tips cleaned by standard in situ procedures and coated with ~50 ML of Fe were used as probes for our SP-STM studies. The Ir(111) single crystals were cleaned by repeated cycles of Ar+ sputtering (800 V, 5E-6 mbar), annealing at temperatures ranging from 900 K to 1500 K in the presence of O2 and a flash annealing at ~1500 K. The graphene layer was grown in situ on Ir(111) by decomposition of ethylene molecules following the procedure described in ref. [N'Diaye, 2008]. Nearly full layer intercalation of Fe was 7 achieved in situ following the procedure of ref. [Bazarnik, 2015]. Molecules were deposited from thermally cleaned aluminum nitride (AlN) crucibles by thermal sublimation under UHV conditions directly onto the substrate held at room temperature. Samples after preparation were transferred in vacuo to the SP-STM setup and cooled down to the measurement temperature of 6.5 K. The exact positions of the molecules within the assembly presented in Fig. 1a have been determined by a combination of energy-dependent dI/dU maps revealing the electronic states of different functional groups within the molecules and high-resolution STM images obtained with a molecule terminated probe tip. Lattice constants for Gr and Fe lattices used in Fig. 1e have been extracted from ref. [Zeller, 2014]. A lock-in detection technique was used to obtain dI/dU maps and point spectroscopy data. The dI/dU maps were recorded simultaneously with the STM topography in the constant-current mode. The spin asymmetry is defined here as dI/dUasym(U) = dI/dU(U) - dI/dU(U) / dI/dU(U) + dI/dU(U) where the directions  and  refer to the magnetization of the SP-STM probe tip and the sample. The G values have been averaged over areas of ~0.1 nm2 and ~0.7 nm2 for the molecule's Co centers and the GR/Fe hcp regions respectively. All data has been processed using MATLAB and Gwyddion [Nečas, 2012] software. Tunneling parameters: Fig. 1a: U = +450 mV, I = 55pA; Fig. 1c: U = +50 mV, I = 50pA; Fig. 1d: U = +450 mV, I = 50pA; Fig. 2: Data is extracted from dI/dU maps recorded with: U = +450 mV, I = 55pA, fmod = 971 Hz, Vmod = 50 mVrms; Fig. 3a: Ustab = +1 V, Istab = 55pA, zoff = 100 pm, fmod = 971 Hz, Vmod = 50 mVrms, every line represents an average over 5 individual spectra; Fig. 3b: Ustab = +1 V, Istab = 200pA, zoff = 0 pm, fmod = 971 Hz, Vmod = 50 mVrms, every line represents an average over 5 individual spectra; Fig. 3c: U = -250 mV, I = 55pA, fmod = 971 Hz, Vmod = 50 mVrms; 8 Fig. 3d: U = +450 mV, I = 55pA, fmod = 971 Hz, Vmod = 50 mVrms; Fig. 3e: U = +950 mV, I = 55pA, fmod = 971 Hz, Vmod = 50 mVrms. Acknowledgements We gratefully acknowledge financial support from the Office of Naval Research via grant No. N00014-16-1-2900. M.B. additionally acknowledges the support of the National Science Centre, Poland under grant nr. 2017/26/E/ST3/00140. We are grateful to J. Wiebe for insightful discussions and to B. Bugenhagen for providing us with molecules and insightful discussions. Author contributions M.B. conceived the experiment. E.S. performed the measurements and analyzed the data. M.E. supported the measurements. R.W. and M.B. supervised the work. E.S., R.W., and M.B. wrote the manuscript. All authors discussed the results and commented on the manuscript. Competing interests The authors declare no competing financial interests. References Wiesendanger, R. Güntherodt, H. J. Güntherodt, G. Gambino, R. J. and Ruf, R. Observation of Vacuum Tunneling of Spin-polarized Electrons with the Scanning Tunneling Microscope. Phys. Rev. Lett. 65, 247-250 (1990). Wiesendanger, R. Spin mapping at the nanoscale and atomic scale. Rev. Mod. Phys. 81, 1495-1550 (2009). Decker R. et al. Local tunnel magnetoresistance of an iron intercalated graphene-based heterostructure. J. Phys.: Condens. Matter 26, 394004 (2014). Meier, F. Zhou, L. Wiebe, J. and Wiesendanger, R. Revealing Magnetic Interactions from Single- Atom Magnetization Curves. Science 320, 82-86 (2008). 9 Bogani, L. and Wernsdorfer, W. Molecular spintronics using single-molecule magnets. Nat. Mater.7, 179-186 (2008). Schwöbel, J. et al. Real-space observation of spin-split molecular orbitals of adsorbed single-molecule magnets. Nat. Commun. 3:953 doi: 10.1038/ncomms1953 (2012). Dreiser, J. et al. Out-of-Plane Alignment of Er(trensal) Easy Magnetization Axes Using Graphene. ACS Nano 10, 2887−2892 (2016). Gragnaniello, L. et al. Uniaxial 2D Superlattice of Fe4 Molecular Magnets on Graphene. Nano Lett. 17, 7177−7182 (2017). Paschke, F. Erler, P. Enenkel, V. Gragnaniello, L. and Fonin, M. Bulk-Like Magnetic Signature of Individual Fe4H Molecular Magnets on Graphene. ACS Nano 13, 780−785 (2019). Iacovita, C. et al. Visualizing the Spin of Individual Cobalt-Phthalocyanine Molecules. Phys. Rrev. Lett. 101, 116602 (2008). Javaid, S. et al. Impact on Interface Spin Polarization of Molecular Bonding to Metallic Surfaces. Phys. Rev. Lett. 105, 077201 (2010). Atodiresei, N. at al. Design of the Local Spin Polarization at the Organic-Ferromagnetic Interface. Phys. Rev. Lett. 105, 066601 (2010). Brede, J. et al. Spin- and Energy-Dependent Tunneling through a Single Molecule with Intramolecular Spatial Resolution. Phys. Rev. Lett. 105, 047204 (2010). Brede, J. and Wiesendanger, R. Spin-resolved characterization of single cobalt phthalocyanine molecules on a ferromagnetic support. Phys. Rev. B 86, 184423 (2012). Mugarza, A. et al. Spin coupling and relaxation inside molecule -- metal contacts. Nat. Commun. 2:490 doi: 10.1038/ncomms1497 (2011). Avvisati, G. et al. Superexchange pathways stabilize the magnetic coupling of MnPc with Co in a spin interface mediated by graphene. Phys. Rev B 98, 115412 (2018). Avvisati, G. et al. Ferromagnetic and Antiferromagnetic Coupling of Spin Molecular Interfaces with High Thermal Stability. Nano Lett. 18, 2268−2273 (2018). Czap, G. et al. Probing and imaging spin interactions with a magnetic single-molecule sensor. Science 364, 670 -- 673 (2019). 10 Gambardella, P. et al. Supramolecular control of themagnetic anisotropy in two-dimensional high- spin Fe arrays at a metal interface. Nat. Mater. 8, 189-193 (2009). Kawahara, S. L. et al. Large Magnetoresistance through a Single Molecule due to a Spin-Split Hybridized Orbital. Nano Lett. 12, 4558−4563 (2012). Brede, J. et al. Long-range magnetic coupling between nanoscale organic -- metal hybrids mediated by a nanoskyrmion lattice. Nat. Nano. 9, 1018-1023 (2014). Bazarnik, M. Brede, J. Decker, R. and Wiesendanger, R. Tailoring Molecular Self-Assembly of Magnetic Phthalocyanine Molecules on Fe- and Co-Intercalated Graphene. ACS Nano 7, 11341-11349 (2013). DiLullo, A. et al. Molecular Kondo Chain. Nano Lett. 12, 3174−3179 (2012). Bazarnik, M. et al. Toward Tailored All-Spin Molecular Devices. Nano Lett. 16, 577−582 (2016). Sierda, E. et al. On-Surface Oligomerization of Self-Terminating Molecular Chains for the Design of Spintronic Devices. ACS Nano 11, 9200−9206 (2017). García-Fernandez, C. at al. Exploring the Relation Between Intramolecular Conjugation and Band Dispersion in One-Dimensional Polymers. J. Phys. Chem. C 121, 27118−27125 (2017). Repp, J. Meyer, G. Molecules on Insulating Films: Scanning-Tunneling Microscopy Imaging of Individual Molecular Orbitals. Phys. Rev. Lett. 94, 026803 (2005). Kubetzka, A. Pietzsch, O. Bode, M. and Wiesendanger, R. Determining the spin polarization of surfaces by spin-polarized scanning tunneling spectroscopy. Appl. Phys. A 76, 873 -- 877 (2003). Wittneven, C. R. Dombrowski, R. Pan, S. H. and Wiesendanger, R. A low-temperature ultrahigh- vacuum scanning tunneling microscope with rotatable magnetic field. Rev. Sci. Instrum. 68, 3806- 3810 (1997). N'Diaye, A. T. Coraux, J. Plasa, T. N. Busseand, C. and Michely, T. Structure of epitaxial graphene on Ir(111). New Journal of Physics 10, 043033 (2008). Bazarnik, M. Decker, R. Brede, J. and Wiesendanger, R. Multi-layer and multi-component intercalation at the graphene/Ir(111) interface. Surf. Sci. 639, 70 -- 74 (2015). 11 Zeller, P. and Günther, S. What are the possible moiré patterns of graphene on hexagonally packed surfaces? Universal solution for hexagonal coincidence lattices, derived by a geometric construction. New Journal of Physics 16, 083028 (2014). Nečas, D. and Klapetek, P. Gwyddion: an open-source software for SPM data analysis. Cent. Eur. J. Phys. 10, 181-188 (2012). Figures 12 Figure 1: STM topographs of CoSal molecules adsorbed on GR/Fe and corresponding structure model of the imaged area. (a) STM overview image of the molecular assembly used for subsequent higher-resolution measurements (U = +450 mV, I = 55 pA). Ball-and-stick models of the molecules within the assembly are superimposed on the measured STM data, while the positions of the circles represent top sites of the moiré pattern originating from the intercalated GR layer. (b) A schematic 13 cross-section along the white dashed line in the STM image shown in (a). (c)-(d) Bias-dependent STM images of a single CoSal molecule within the assembly obtained with a molecular probe tip in the area marked with a white square in (a): U = +50 mV for (c) and U = +450 mV for (d), while I = 50 pA. (e) Ball-and-stick models of the molecules overlaid on the atomic lattice of the GR/Fe substrate (GR: red, Fe: black) as deduced from (a). The individual atomic species within the CoSal molecule as well as the specific regions of the GR/Fe moiré structure are labelled. Figure 2: Magnetic response of a single adsorbed CoSal molecule as imaged in Fig. 1c and the bare GR/Fe substrate to an external out-of-plane magnetic field 𝐵⃗ . (a) Spin-resolved differential tunneling conductance G extracted from the measured dI/dU signal averaged over the molecule's Co-metal center and over a bare GR/Fe area next to the molecular assembly. The error bars correspond to the standard deviation of the spatially averaged dI/dU signals. (b) Schematic drawings indicating the magnetization directions of the Fe-coated W-probe tip, the Co-center of the CoSal molecule and the GR/Fe layer for three different B-field values as color coded in (a): 1.50 T (blue), 3.75 T (green), and 5.25 T (red). 14 Figure 3: Local SP-STS data and SP-STM images of a single CoSal-molecule measured in different external B-fields. (a)-(b) Normalized SP-STS obtained on the CoSal molecule's Co-center (a) and on a bare GR/Fe area next to the molecular assembly (b) at different magnetic field strengths: B = 1.5 T and 4.5 T. The measured spin-resolved differential tunneling conductance dI/dU has been divided by I/U and plotted as a function of bias voltage U. (c)-(e) Spatially resolved dI/dU maps for three different bias voltages: U = -250 mV (c), +450 mV (d), and +950 mV (e) as marked in (a) by dashed lines (I = 55 pA). (f)-(h) Spin asymmetry maps for the bias voltages as presented above in (c)-(e). All maps in (c)-(h) have been superimposed by the structural model of the CoSal molecule. 15
1709.03692	1	1709	2017-09-12T05:05:48	UV Detector based on InAlN/GaN-on-Si HEMT Stack with Photo-to-Dark Current Ratio > 107	[ "physics.app-ph" ]	We demonstrate an InAlN/GaN-on-Si HEMT based UV detector with photo to dark current ratio > 107. Ti/Al/Ni/Au metal stack was evaporated and rapid thermal annealed for Ohmic contacts to the 2D electron gas (2DEG) at the InAlN/GaN interface while the channel + barrier was recess etched to a depth of 20 nm to pinch-off the 2DEG between Source-Drain pads. Spectral responsivity (SR) of 34 A/W at 367 nm was measured at 5 V in conjunction with very high photo to dark current ratio of > 10^7. The photo to dark current ratio at a fixed bias was found to be decreasing with increase in recess length of the PD. The fabricated devices were found to exhibit a UV-to-visible rejection ratio of >103 with a low dark current < 32 pA at 5 V. Transient measurements showed rise and fall times in the range of 3-4 ms. The gain mechanism was investigated and carrier lifetimes were estimated which matched well with those reported elsewhere.	physics.app-ph	physics	UV Detector based on InAlN/GaN-on-Si HEMT Stack with Photo-to-Dark Current Ratio > 107 Sandeep kumar1a,b), Anamika Singh Pratiyush1a), Surani B. Dolmanan2, Sudhiranjan Tripathy2, Rangarajan Muralidharan1, Digbijoy N. Nath1 1Centre for Nano Science and Engineering (CeNSE), Indian Institute of Science (IISc), Bangalore 560012, India 2Institute of Materials Research and Engineering (IMRE), Agency for Science, Technology, and Research (ASTAR), Innovis 08-03, 2 Fusionopolisway, Singapore 138634 Abstract: We demonstrate an InAlN/GaN-on-Si HEMT based UV detector with photo to dark current ratio > 107. Ti/Al/Ni/Au metal stack was evaporated and rapid thermal annealed for Ohmic contacts to the 2D electron gas (2DEG) at the InAlN/GaN interface while the channel + barrier was recess etched to a depth of 20 nm to pinch-off the 2DEG between Source-Drain pads. Spectral responsivity (SR) of 34 A/W at 367 nm was measured at 5 V in conjunction with very high photo to dark current ratio of > 107. The photo to dark current ratio at a fixed bias was found to be decreasing with increase in recess length of the PD. The fabricated devices were found to exhibit a UV-to-visible rejection ratio of >103 with a low dark current < 32 pA at 5 V. Transient measurements showed rise and fall times in the range of 3-4 ms. The gain mechanism was investigated and carrier lifetimes were estimated which matched well with those reported elsewhere. a) Sandeep Kumar and Anamika Singh Pratiyush have equally contributed for this work. b) Corresponding author email: [email protected], [email protected] With applications in strategic sector, biomedical research, medical science, and UV astronomy, semiconductor ultra-violet photodetectors (PDs)attract widespread attention of the device and material research communities1. Compared to the conventional UV enhanced silicon photodiodes which suffer from poor efficiencies and necessitates the use of filters to reject the visible band for solar blind applications which make them bulky, inefficient and cumbersome, wide band gap materials, such as III-Nitrides2–4 and Ga2O35–8 are intrinsically solar blind and offer high spectral responsivities. Different types of device geometries have been explored for GaN UV PDs (~365 nm) such as MSM4,9, PN3, PIN10, and Schottky2. Besides these, III-nitride high electron mobility transistor (HEMT) stacks with a 2D electron gas (2DEG) – more commonly AlGaN/GaN – which constitute a matured transistor technology, are also being reported for UV detection at ~ 365 nm in their various configurations such as AlGaN/GaN gated11, AlGaN/GaN 2DEG UV detectors12, AlGaN/GaN with MESA in meandering geometry13, GaN MSM with Al nanoparticles14 and GaN MSM with Ag nanoparticles15 have been reported earlier (listed in Table I). However, the spectral responsivity (SR) values of most of the PDs reported are comparatively low (< 5A/W at 1 V) while a few high SR values have also been reported in conjunction to high dark current (AlGaN/GaN 2DEG UV detectors). The high value of SR in meander geometry PDs13 comes with a complex device design. This high SR value is also attributed to good material quality, as the HEMT stack was grown on SiC substrate. In this letter, we report a UV PD at 367 nm realized on a recess etched InAlN/GaN HEMT stack on silicon (111) with a high spectral responsivity of 7.5 A/W at 1 V. This device seeks to exploit the advantages of the highly conducting 2D electron gas in a III-nitride HEMT while ensuring that the dark current of such a design remains low by recess etching the 'gate' region between the source and drain. The In0.17Al0.83N/GaN HEMT stack was used for fabricating UV PDs (schematic in Fig. 1(a)). The HEMT stack (total thickness ~1.9 µm) was grown on a Si (111) substrate 2 using a metal organic chemical vapor deposition technique. The epilayers consist of ~10 nm In0.17Al0.83N barrier, 1.0 nm AlN thin spacer, 150-nm undoped GaN channel, unintentionally doped GaN buffer, and AlGaN step-graded intermediate layers, overgrown on AlN/Si(111). The high-resolution reciprocal space mapping by x-ray diffraction confirmed an In concentration of ~17% in the HEMT barrier. The Hall data measured from these samples showed an average sheet resistance of 235 Ω/sq with a sheet carrier density of about 3.2 × 1013 cm-2. Atomic force microscopy measurements of such uncapped InAlN-based HEMT stack show an average surface rms roughness ~0.5 nm for 5.0 µm × 5.0 µm scan area. Following the standard lithographic process, Ti/Al/Ni/Au was evaporated for Ohmic contacts to In0.17Al0.83N/GaN stack. Ohmic contact annealing was performed at 850° C in N2 ambient for 30 s. Mesa isolation of devices was done by dry etching using BCl3/Cl2 gas chemistry in reactive ion etching (RIE). The Ohmic metal pads (350 m  350 m) of the fabricated devices were 300m spaced and the recess region was defined at the center, as shown in the Fig. 1 (b) between the Source and the Drain. Recess lengths of L = 3 m, 5 m, 7 m, and 9 m were opened using lithography for various devices, and 10 W of RF power in BCl3/Cl2 gas chemistry was used to recess etch to a depth of 20 nm, essentially etching the barrier + channel layer so as to completely deplete the 2D electron in this region. The SR measurement system used in this work is reported elsewhere5. Figure 2 shows SR versus wavelength () characteristics at different bias voltages for a device with 3 m recess length. The PD exhibits a peak SR value of 7.5 A/W and 55.2 A/W at 367 nm for 1 V and 15 V respectively. The UV-to-visible rejection ratio is defined as the ratio of peak SR value at 367 nm to the peak SR value at 420 nm which is > 103 and indicative of visible-blind nature of the PD. The high SR value could be attributed to the high quality GaN channel of the HEMT stack which is expected to lead to a high minority carrier lifetime for photo-generated carriers. The high 2DEG density at InAlN/GaN interface results 3 in RC = 0.26 ohm-mm and RSH = 262 ohm/(cid:0) from TLM measurements. Compared to Hall data, a small increase in the RSH value observed from the TLM measurement, which may be due to the thermal processing of Ohmic contacts. Hence there is a negligible voltage drop across the access regions and most of the voltage appears across the recessed region. In Fig. 3 the peak SR value first increases and then tend to saturate slowly at higher bias voltages. For a given bias, the SR decreases as the recess length increases. This indicates that the responsivity (or gain) in the devices is transit time limited. Figure 4 shows the photo (367 nm) and dark current versus bias voltage for the 3 m recess length device. The measured photo and dark currents were 1.2 mA and 58 pA respectively at 20 V. The photo-to-dark current ratio for device was > 107which is the highest for any type of III-nitride UV detector in this spectral range. Inset in Figure 4 shows the photo-to-dark current ratio versus recess length at a bias voltage of 20 V. The photo-to- dark current ratio (20 V) was found to be decreasing with increase in the recess length, which is attributed to the decrease in electric field with increase in recess length. Figure 5 shows the transient response of the PD measured at 5 V. The light was chopped at 100 Hz using optical chopper and then focused on to the DUT after being guided through the optical assembly. The PD was biased using SR570 current amplifier and the voltage transient response of the PD was measured using an oscilloscope. The rise and fall times (10% - 90% of value) were 3.6 ms and 4.2 ms respectively for devices with 3 µm recess lengths. There was no significant difference in the rise and fall times for PDs with recess lengths of 5 m, 7 m, and 9 m (not shown here). To further reveal the photo response of the InAlN/GaN-based UV photodetector, the voltage dependent photocurrent (367 nm) of the PD was measured at different light intensities. Figure 6 (a), shows a series of photo I-V curves under increasing incident light 4 intensity (from 4.9 to 11.5 mW/cm2). It can be observed from Fig. 6 (a) that the photocurrent increases with increasing light intensity at a particular bias voltage. Figure 6 (b) shows the photocurrent of the detector versus light intensity at 20 V for different recess lengths and their linear fit. The photocurrent increases linearly with the incident excitation light intensity at the wavelength of 367 nm for different recess lengths. The linear increase of photocurrent with light intensity shows that the detector has negligible trap related gain. Figure 7 shows gain of the PD versus bias voltage of different recess length devices (3 μm to 9 μm). Assuming the external quantum efficiency (η) to be 100%, the theoretical responsivity (RTh) of UV photodetectors having a detection range of 367-368 nm can be calculated using the expression below (where G is the gain): SRIdeal  q  h  (1(a)) SRMeasured  qG  h  (1(b)) The ideal SR value for 367 nm comes out to be 0.29 A/W. This value of ideal responsivity is well surpassed even at a bias voltage of 1 V, which is a clear indication of gain in the devices. The high gain (>102 at 15 V) resulting in high responsivity of the devices is due to photoconductive gain as the metal contacts have a non-rectifying (Ohmic) nature unlike in the more commonly reported MSM or Schottky geometries 16. The photoconductive gain of an intrinsic photoconductor with Ohmic contacts on both electrodes is given by17. G   s e  tr (2(a)) e   tr L 2  e V (2(b)) where, s is the excess carrier lifetime, e tr is the transit time, L is the recess length and V is the applied voltage across the recess length of the PD. Using equations (2(a)) , (2(b)) and 5 assuming the mobility of electron in GaN recess channel from the earlier reported value to be 100 cm2/V-s18, we have estimated the excess carrier lifetime values (~ 20 ns), which is in good agreement with the earlier reported values19. It can be observed from figure 7, the gain in devices with 3 µm recess length increases linearly till 5 V and then tend to saturate slowly at higher voltages whereas, gain in devices with recess length of 5 μm, 7 μm and 9 μm tend to saturate at lower voltages. The early gain saturation in devices with longer recess lengths (9 μm) compared to smaller recess lengths (3 μm) is attributed to higher channel resistance in the former, which limits the photocurrent20. Hence, we expect higher photocurrent and gain for smaller recess devices. As can be observed from figure 8, the Y-axis intercept (Gain ~ 0 V) for 3 μm device is higher than 9 μm device. This suggests that high gain and photocurrent can be expected for submicron recess devices. Fabrication of submicron recess length PDs are underway, which are expected to exhibit even higher responsivity (gain) as they would not suffer early saturation of gain. In summary, we demonstrate record high photo-to-dark current ratio exceeding 107 for UV photodetectors based on InAlN/GaN-on-Si HEMT stack. High responsivity of 32.9 A/W at 5 V is obtained with UV-to-visible rejection ratio >103. We have also demonstrated low dark current 32 pA at 5 V and transient response rise and fall time of 3.6 ms and 4.2 ms respectively. Photocurrent dependence on light intensity is also studied for devices with different recess lengths. The high gain is explained by photoconductive gain mechanism. This demonstration of a state-of-art UV detector in a HEMT stack holds promise towards physical integration of UV devices with high power/RF transistors on the same substrate which could open up exciting avenues to explore new devices with added functionalities. This work was funded by the Department of Science and Technology (DST) under its Water Technology Initiative (WTI), Grant No. DST 01519. This publication is an outcome of the Research and Development work undertaken in the Project under Ph.D. scheme of Media 6 Lab Asia. We would also like to thank Ministry of Electronics and Information Technology (MeiTY) for the financial support for the work. Authors would also like to acknowledge the National NanoFabrication Centre (NNFC) and Micro and Nano Characterization Facility (MNCF) at CeNSE, IISc for device fabrication and characterization. 7 REFERENCES: 1 M. Razeghi, Proc. IEEE 90, 1006 (2002). 2 O. Katz, V. Garber, B. Meyler, G. Bahir, and J. Salzman, Appl. Phys. Lett. 80, 347 (2002). 3 E. Monroy, E. Muñoz, F.J. Sánchez, F. Calle, E. Calleja, B. Beaumont, P. Gibart, J. a Muñoz, and F. Cussó, Semicond. Sci. Technol. 13, 1042 (1998). 4 E. Monroy, T. Palacios, O. Hainaut, F. Omnès, F. Calle, and J.F. Hochedez, Appl. Phys. Lett. 80, 3198 (2002). 5 A. Singh Pratiyush, S. Krishnamoorthy, S. Vishnu Solanke, Z. Xia, R. Muralidharan, S. Rajan, and D.N. Nath, Appl. Phys. Lett. 110, 221107 (2017). 6 S. Fujita, Jpn. J. Appl. Phys. 54, 30101 (2015). 7 X. Chen, K. Liu, Z. Zhang, C. Wang, B. Li, H. Zhao, D. Zhao, and D. Shen, ACS Appl. Mater. Interfaces 8, 4185 (2016). 8 T. Oshima, T. Okuno, N. Arai, N. Suzuki, S. Ohira, and S. Fujita, Appl. Phys. Express 1, (2008). 9 W.Y. Weng, T.J. Hsueh, S.J. Chang, S.B. Wang, H.T. Hsueh, and G.J. Huang, IEEE J. Sel. Top. Quantum Electron. 17, 996 (2011). 10 J.C. Lin, Y.K. Su, S.J. Chang, W.H. Lan, K.C. Huang, W.R. Chen, C.Y. Huang, W.C. Lai, W.J. Lin, and Y.C. Cheng, Appl. Phys. Lett. 91, 173502 (2007). 11 M. a. Khan, M.S. Shur, Q. Chen, J.N. Kuznia, and C.J. Sun, Electron. Lett. 31, 398 (1995). 12 T.M. Kuan, S.J. Chang, Y.K. Su, C.H. Ko, J.B. Webb, J.A. Bardwell, Y. Liu, H. Tang, W.J. Lin, Y.T. Cherng, and W.H. Lan, Japanese J. Appl. Physics, Part 1 Regul. Pap. Short Notes Rev. Pap. 42, 5563 (2003). 13 M. Martens, J. Schlegel, P. Vogt, F. Brunner, R. Lossy, J. Würfl, M. Weyers, and M. Kneissl, Appl. Phys. Lett. 98, 211114 (2011). 14 S. Chang, M. Chang, and Y. Yang, IEEE Photonics J. 9, 6801707 (2017). 15 D. Li, X. Sun, H. Song, Z. Li, Y. Chen, H. Jiang, and G. Miao, Adv. Mater. 24, 845 (2012). 16 S. Rathkanthiwar, A. Kalra, S. V. Solanke, N. Mohta, R. Muralidharan, S. Raghavan, and D.N. Nath, J. Appl. Phys. 121, 164502 (2017). 17 J.M. Liu, Photonic Devices (2015). 18 H. Kim, S. Han, W. Jang, C. Cho, and K. Seo, IEEE ELECTRON DEVICE Lett. 38, 1090 (2017). 19 Z.Z. Bandić, P.M. Bridger, E.C. Piquette, and T.C. McGill, Appl. Phys. Lett. 72, 3166 (1998). 20 J. Baek, M.S. Shur, K.W. Lee, and T. Vu, IEEE Trans. Electron Devices ED-32, 2426 (1985). 8 Figure and legends: Figure 1: (a) Schematic of fabricated device from HEMT stack (b) Optical micrograph of the fabricated device. Recess region (3 m) is shown by dashed line. Figure 2: Spectral response versus wavelength (300 nm- 420 nm) of GaN PD (3 m recess) (Log scale) at different bias voltages (1 V to 15 V), also showing UV to visible rejection ratio >103 at 5 V. The inset shows SR versus wavelength at different bias voltages (Linear Scale). 9 Figure 3: The peak SR versus bias voltage for different recess lengths (3 μm to 9 μm) PDs. Figure 4: Photo (367 nm) and dark current for 3 um recess (Till 20 V). Inset shows photo-to- dark current ratio versus recess length at 20 V. 10 Figure 5: Transient response of the detector measured using CRO and SR570 current amplifier. Figure 6: (a) Photocurrent (367 nm) versus voltage for 3 μm recess length device at different light intensities. (b) Photocurrent (367 nm) versus light intensity at 20 V for different recess length devices (3 μm to 9 μm). Dashed line shows liner fit to experimental data. 11 Figure 7: Gain versus bias voltage of different recess length devices (3 μm to 9 μm). Dashed line shows liner fit to experimental data up to 5 V. Tables: TABLE I. List of UV detectors (~365 nm, GaN band edge) reported earlier for different material systems. Material System GaN GaN GaN GaN GaN AlGaN/GaN AlGaN/GaN AlGaN/GaN GaN GaN Detector Type MSM PN PIN Lateral Schottky Vertical Schottky Gated HEMT HEMT 2DEG UV Meander MSM -Al Nanoparticle MSM- Ag Nanoparticle SR (Bias Voltage) 0.5 A/W (5 V) 0.15 A/W (-) 0.25 A/W (5 V) 0.02 A/W (0 V) 0.1 A/W (0 V) 3 A/mW (10 V) 5.2 x 109 A/W (-) 10 A/mW (5 V) 3 A/W (10 V) 0.14 A/W (5 V) IPhoto/IDark (Bias Voltage) Reference 107 (5 V), 103 (10 V) 50 (0.5 V) 102 (5 V) - - 1.25 (10 V) 2 (6 V) 104 (10 V) 106 (10 V) - 49 3 10 2 2 11 12 13 14 15 12
1811.01048	3	1811	2019-05-15T14:12:34	Mapping the global design space of nanophotonic components using machine learning pattern recognition	[ "physics.app-ph", "physics.optics" ]	Nanophotonics finds ever broadening applications requiring complex component designs with a large number of parameters to be simultaneously optimized. Recent methodologies employing optimization algorithms commonly focus on a single design objective, provide isolated designs, and do not describe how the design parameters influence the device behaviour. Here we propose and demonstrate a machine-learning-based approach to map and characterize the multi-parameter design space of nanophotonic components. Pattern recognition is used to reveal the relationship between an initial sparse set of optimized designs through a significant reduction in the number of characterizing parameters. This defines a design sub-space of lower dimensionality that can be mapped faster by orders of magnitude than the original design space. As a result, multiple performance criteria are clearly visualized, revealing the interplay of the design parameters, highlighting performance and structural limitations, and inspiring new design ideas. This global perspective on high-dimensional design problems represents a major shift in how modern nanophotonic design is approached and provides a powerful tool to explore complexity in next-generation devices.	physics.app-ph	physics	Mapping the global design space of nanophotonic components using machine learning pattern recognition Daniele Melati1,†, Yuri Grinberg2,†, Mohsen Kamandar Dezfouli1, Siegfried Janz1, Pavel Cheben1, Jens H. Schmid1, Alejandro Sánchez-Postigo3 and Dan-Xia Xu1,* 1Advanced Electronics and Photonics Research Centre, National Research Council Canada, 1200 2Digital Technologies Research Centre, National Research Council Canada, 1200 Montreal Rd., Ottawa, Montreal Rd., Ottawa, ON K1A 0R6, Canada 3Universidad de Málaga, Departamento de Ingeniería de Comunicaciones, ETSI Telecomunicación, ON K1A 0R6, Canada Campus de Teatinos s/n, 29071 Málaga, Spain *e-mail: [email protected] †These authors contributed equally to this work. Abstract Nanophotonics finds ever broadening applications requiring complex component designs with a large number of parameters to be simultaneously optimized. Recent methodologies employing optimization algorithms commonly focus on a single design objective, provide isolated designs, and do not describe how the design parameters influence the device behaviour. Here we propose and demonstrate a machine- learning-based approach to map and characterize the multi-parameter design space of nanophotonic components. Pattern recognition is used to reveal the relationship between an initial sparse set of optimized designs through a significant reduction in the number of characterizing parameters. This defines a design sub-space of lower dimensionality that can be mapped faster by orders of magnitude than the original design space. As a result, multiple performance criteria are clearly visualized, revealing the interplay of the design parameters, highlighting performance and structural limitations, and inspiring new design ideas. This global perspective on high-dimensional design problems represents a major shift in how modern nanophotonic design is approached and provides a powerful tool to explore complexity in next-generation devices. 1 A multitude of parameters determine the performance of a photonic device, encompassing the optical properties of the constituent materials, structural geometry and dimensions. Similarly, the choice of the best design to proceed to fabrication, integration and system implementation needs to take into account many performance criteria. These not only include the primary functionality, but also other metrics such as insertion loss, the effect of temperature, and the influence on other system components (e.g. back reflections), to name a few. Susceptibility of manufacturing yield to the inherent variability of the fabrication processes is another important consideration. Historically, conception of a new device relies on theoretical knowledge and physical intuition to identify the potential structure and design parameter range. The design parameter space is explored semi- analytically or numerically, and the relevant performance metrics are analyzed. This approach is constrained in scope by computational resources and limited to structures governed by only few parameters and where the evaluation process can be decomposed into sequential steps. As the scope of nanophotonics broadens in complexity and application range1,2, this conventional approach poses increasing challenges. For example, in devices employing metamaterials3 -- 6 or the complex geometries generated by inverse design and topology optimization7 -- 12 not only the number of design parameters vastly increase but they are often strongly inter-dependent. Sequential optimization is no longer applicable and simultaneous optimization of multiple parameters is required. Optimization tools such as genetic algorithm13 -- 15, particle swarm16,17, and gradient-based optimization18 -- 21 are now commonly used to search more efficiently for high-performance designs22. More recently, supervised machine learning methods such as the artificial neural network have begun to enter the fray in speeding up the search process23,24. While all these approaches represent significant improvements to the design flow, they still suffer constitutive limitations: usually a single performance criterion is optimized; only a single or a handful of optimized designs are discovered; and the optimization process needs to be repeated for new performance criteria. Furthermore, optimized designs in isolation reveal very little on the characteristics of the design space and the influence of the design parameters on the device performance. Consequently, careful balancing of different figures of merit becomes difficult. A global perspective on the design space of nanophotonic devices is presently missing. We propose here a methodology based on machine learning (ML) tools to map and characterize a multi- parameter design space. From an initial sparse set of optimized designs, unsupervised dimensionality reduction reveals a lower-dimensional design sub-space where good designs with high performance reside. Since computational effort grows exponentially with dimensionality, this sub-space can be mapped faster by orders of magnitude than the original design space, enabling the efficient evaluation and visualization of an arbitrary number of performance criteria. The comprehensive characterization of the continuous region that includes all possible good design solutions highlights their performance as well as structural differences and limitations. This provides a clear understanding of the design space, making possible the discovery of superior designs based on the relative priorities for a particular application. To the best of our knowledge, this is the first time such a global perspective is obtained by leveraging unsupervised machine learning techniques for high-dimensional design problems in nanophotonics. As a first demonstration-of-concept, we analyze a technologically important device, i.e. a vertical fiber grating coupler in the silicon-on-insulator (SOI) platform consisting of multiple segments whose dimensions need to be optimized simultaneously17,21,25 -- 29. We obtain a clear description of the device not 2 Fig. 1 Conceptual illustration of the three-stage approach for the characterization of a high-dimensional design space for nanophotonic devices. a An initial sparse collection of good designs (red circles) is found by optimization (here, random re-start - as indicated by blue circles - followed by local search) in the original high-dimensional design space. A trained machine learning predictor is used in conjunction with the optimizer to speed up the search by quickly identifying promising design candidates as starting points for the optimizer (see Methods). b Dimensionality reduction (e.g. principal component analysis) is employed to reveal the lower dimensional sub-space where the good designs with high performance reside, shown here as a hyperplane. c The low-dimensional design sub-space can be exhaustively mapped. For the continuum of the designs in the sub-space (that includes also the initial sparse set) a complete characterization is made possible by computing both the performance criterion selected for optimization and any additional metric. only in terms of fiber coupling efficiency but also back-reflections, minimum feature size, and sensitivity to dimensional variations. Furthermore, this global perspective on the design space is exploited to arrive at conclusive arguments as to whether certain features can be obtained with a given structure. Analysis of the considered grating geometry indicates that a minimum feature size larger than 88 nm is not achievable. This design bottleneck revealed by dimensionality reduction inspired a new grating structure that incorporates subwavelength metamaterial and allows a minimum feature size above 100 nm without compromising the device performance. This represents a crucial improvement in device manufacturability for volume- production. The composite design space of refractive index and segment dimensions involved here can be equally well characterized using the same global mapping procedure. This provides a clear indication that our novel optimization approach is generally applicable to a wide range of high-dimensional design problems. Strategy for characterizing a multi-parameter design space In designing multi-parameter devices, it is often difficult or impossible to obtain extensive information on device performance variation over the large parameter space due to constrains on computational resources. We tackle this problem by introducing the three-stage process illustrated in Fig. 1. In the first stage multiple iterations of an optimization algorithm are used to generate a sparse collection of different good designs, i.e. designs that optimize a primary performance criterion (Fig. 1a). Supervised ML techniques are exploited to speed up the search process by quickly providing promising design candidates as starting points for the optimizer (see Methods). In the second stage (Fig. 1b), dimensionality 3 Fig. 2 Schematic representation of the grating coupler structure under study. The guided light incident from the left is diffracted vertically by a grating periodically interleaving a pillar of height 220 nm and an L-shaped section partially etched to 110 nm. The L-shape approximates the angled facet of a conventional blazed grating21 in a way that can be fabricated with standard lithography and etch methods. The pillars are designed to suppress back-reflections by destructive interference. The five design parameters L1-L5 define the original five-dimensional parameter space and the grating period Λ. reduction is applied to analyse the relationship in the parameter space between these degenerate designs. The goal is to find a lower dimensional sub-space where all good designs reside. This design sub-space is described by significantly fewer parameters compared to the original design space. In the last stage (Fig. 1c) we map the design sub-space by computing across it all required performance criteria and identifying a continuous region of good design solutions. Through this process, the sparse initial set of good designs efficiently leads to the identification and comprehensive characterization of the continuum of all good designs in the sub-space. A vertical fiber grating coupler with five segments per period is taken as the study case. The considered grating structure is illustrated in Fig. 2. Although desirable, perfectly vertical emission makes the design a challenging problem due to the necessity to suppress the second order diffraction that reflects back into the waveguide26. In a recent work by Watanabe et al.17 a single optimized design was generated using particle swarm optimization, providing a good fiber-chip coupling efficiency and a fairly low level of back-reflections. Each period of the grating consists of a pillar of 220 nm in height and an L-shaped section with a partial etch to 110 nm17. The multiple segments in each period need to be simultaneously optimized and therefore provide a good target to demonstrate our machine-learning-based design approach. Discovery of a sparse collection of good designs In the first stage, an in-house optimizer launched from random starting points in the original parameter space is used to search for the initial sparse set of grating designs with state-of-the-art fiber coupling efficiency. A supervised machine learning predictor is trained to determine the diffraction angle of these complex gratings without performing a first principles Bloch mode calculation. The predictor is used to rapidly screen out random start designs that do not radiate close to the vertical direction, thereby speeding 4 up the search process by about 250% (see Methods). This algorithm achieves a wide coverage of the initial design space. For the grating illustrated in Fig. 2, the dimensions [L1 … L5] define the five-dimensional design parameter space we explore in this work. As primary optimization objective we choose the coupling efficiency η of the diffracted TE-polarized light to a standard single mode optical fibre (SMF-28) placed vertically on top of the grating. Low back-reflection r is another important criterion in minimizing the impact of the grating to other upstream components in the system30,31. These considerations lead to the formulation of the optimization problem for the first stage represented in Fig. 1a as: maximize L1⋯L5 𝜂(L1 ⋯ L5) (1) subject to 𝑟(L1 ⋯ L5) < −15 dB 400 nm < Λ < 1 μm; Li > 50 nm, i = 1 ⋯ 5. The optimization is guided by a single performance metric, the coupling efficiency η. Only solutions with η larger than 0.74 are retained, defined here as good designs. Back-reflection r is not optimized but simply constrained by rejecting design solutions with r > -15 dB. Additional constraints on the grating period Λ and the minimum feature size are included to confine the optimization to designs that are physically manufacturable. The wavelength of light is set at = 1550 nm. A highly-efficient Fourier-type 2-D eigenmode expansion simulator32 is used to compute the device performance. As detailed in the next section, the optimization stage is halted after a sufficient number of good designs is collected Each good design requires on average 1000 simulations to be identified (computational details in Methods). Sub-space identification through dimensionality reduction In this second and key step we study the relationship between the sparse set of good solutions obtained solving the optimization problem (1) through machine learning dimensionality reduction. The goal is to transform a set of correlated variables into a smaller set of new uncorrelated variables that retain most of the original information. Here the linear principal component analysis (PCA)33 is used obtaining a good level of accuracy (see PCA description in Methods). We find that two principal components are sufficient to accurately represent the entire pool of good designs each defined by five segment length in the original design space. That is, all good designs approximately lie on a 2D hyperplane -- the reduced design sub-space. The rest of the design space can be excluded from further investigation. As we discuss in Supplementary, post-processing error analysis demonstrates that 5 good designs are sufficient for PCA to provide an accurate result. In order to verify convergence we collect here 45 good designs. The linear design sub-space is defined by two orthogonal basis vectors V1αβ and V2αβ. Any design k with dimensions Lk = [L1,k … L5,k] can hence be written as 𝑳𝑘 = 𝛼𝑘𝑽1𝛼𝛽 + 𝛽𝑘𝑽2𝛼𝛽 + 𝑪𝛼𝛽. (2) 5 Fig. 3 Exhaustive exploration of the lower-dimensional parameter sub-space. a Reducing the design parameters from the original five Li to the two principal component coefficients α and β makes the exhaustive mapping of the sub-space of good designs achievable with modest computation resources. The map shows the coupling efficiency across the α-β hyperplane for η > 0.70. The large region of good designs with η > 0.74 is enclosed by the black contour line. Two designs with comparable coupling efficiency are marked along with the design reported in ref. 17. b, c The coupling efficiency simulated across two 2-D hyperplanes (Γ-Π and X-Π) orthogonal to the α-β hyperplane (whose intersections are shown with dashed white lines). Γ-Π and X-Π intersections with the α-β plane are shown in a with dashed black lines. Within these orthogonal planes, the cross-section of the sub-space of good designs reduces to a thin stripe confirming that it is approximately a 2-D geometrical structure. Design 3 represents the global optimum in both Γ-Π and X-Π projections: It has a coupling efficiency of 0.77, about 0.5% better than the top designs in the α-β hyperplane. The detailed structural parameters are reported in Table 1. d The back-reflection r simulated across the α-β hyperplane. e, f 2D finite-difference-time-domain (FDTD) simulations of (e) coupling efficiency and (f) back- reflection as a function of wavelength for designs 1-3. The values obtained by FDTD are slightly different than that reported in the maps, but showing consistent trends. All three designs have a 1-dB bandwidth exceeding the telecommunication C band (1530 nm -- 1565 nm, green shaded area). Design 2 affords very low back-reflections near 1550 nm but only within a narrow wavelength band. In contrast, the back-reflection of design 1 and 3 are less dependent on wavelength, but back-reflection lower than -26 dB cannot be achieved. Cαβ is a constant vector that defines the reference origin on the hyperplane. Two scalar coefficients αk and βk are thus sufficient to completely describe design k. Details of vector definitions are provided in Methods. 6 Table 1: Structural and performance parameters of selected grating designs as marked in Fig. 3 Design [α,β] [L1 … L5] [nm] 1 2 3 Ref. 17 [0.22,0.97] [1.93,-0.02] [0.97,0.68]1 [1.49,0.29] [77,84,115,249,171] [102,80,117,330,98] [82,87,111,283,139] [95,83,112,314,109] Λ [nm] 696 727 702 713 Distance [nm] - 216 78 149 η 0.76 0.76 0.77 0.752 r [dB] -21 -37 -20 -252 BW [nm] 44.8 48.9 45.8 46.22 Distance refers to the Manhattan distance with respect to design 1. The coupling efficiency η and reflection 𝑟 refer to the values at a wavelength of 1550 nm 1Closest projection on the hyperplane. 2The performance for the structure proposed in Ref. 17 is recalculated for consistency using the same Fourier-type 2D simulator and settings as the other structures. Now that the area of interest in the design space is limited to a 2D hyperplane, it becomes feasible to adopt a classical design approach and perform an exhaustive mapping of this sub-space, as illustrated in Fig. 1c. First, we generate a uniform grid of 6060 points covering the α-β hyperplane. Therefore, 3600 sampling points are sufficient to provide a wealth of information. As a comparison, sampling with the same resolution in the original design space would require approximately 1.5 million points, increasing the computation time by about over 400 times (details in Methods). For each point [αk, βk] we obtain the corresponding dimensions [L1,k … L5,k] in the original design space through equation (2) and compute the coupling efficiency η. The results are shown in Fig. 3a only for the designs with η > 0.7. Note that not all points on the α-β plane provide high coupling efficiency. A unit division in α or β corresponds to a Manhattan distance ∑ 𝐿𝑖,𝐴 − 𝐿𝑖,𝐵 resolution of 5 nm in Manhattan distance. This exhaustive mapping results in the discovery of a large and well defined region of degenerate designs with η > 0.74, highlighted by the black contour line in Fig. 3a. This region encloses a continuum of designs in addition to those discovered in stage 1. Remarkably, although all the good designs have similar coupling efficiencies ranging from η = 0.74 to η = 0.76, the actual structure of the gratings can vary quite significantly, as will be detailed in the next section. Without dimensionality reduction, there is no obvious way to discover these alternative designs with similar coupling efficiency but potentially different properties in other aspects. of 100 nm. The 60x60 grid covering the α-β hyperplane has a 5 𝑖=1 As the last step, we validate the PCA outcome by verifying that the projection on the low-dimensional design sub-space (the α-β hyperplane) is sufficient to represent the region of good grating designs. We generate two additional 2D hyperplanes Γ-Π and X-Π that are orthogonal to each other and to the α-β plane (details in Methods). They provide two different "cuts" through the α-β sub-space and their projections are shown in Fig. 3a with dashed black lines. We generate a uniform grid on Γ-Π and X-Π and simulate the coupling efficiency of the corresponding grating design. Coupling efficiencies η > 0.7 are plotted in Figs. 3b and 3c, respectively. The intersection with the α-β hyperplane is marked with a white dashed line. The axis use the same scale as in Fig. 3a: a unit division on X, Γ or Π corresponds to a Manhattan distance of 100 nm. When projected on Γ-Π and X-Π hyperplanes, the region of good designs essentially reduces to a thin stripe whose thickness depends on the range of accepted coupling efficiencies η. This confirms that this region has approximately a 2D geometry. Although it appears slightly curved (see the Γ-Π projection, Fig. 3b), it is still well approximated by the α-β hyperplane which has the advantage of being a simple linear structure. 7 Comprehensive characterization of the low-dimensional sub-space of good designs The exhaustive mapping of the sub-space of all good designs can now be readily extended to other performance metrics beyond the primary criterion originally chosen as the optimization objective (the coupling efficiency). This provides the designer a complete picture of the device behaviour, including the upper and lower limit of each performance metric. Informed trade-offs and identification of the best design that fits specific application needs are hence made possible. Along with the coupling efficiency, we evaluate here three additional criteria throughout the sub-space, i.e. back-reflections, minimum feature size, and tolerance to fabrication uncertainty, as presented in Fig. 3 and Fig. 4. All maps use the same axis scale, range and sampling as the α-β plane, and the black contour line marks the region with η > 0.74 for reference. Three designs are selected for further examination. Their structural and performance parameters are listed in Table 1. Design 1 and 2 are on the α-β plane (marked on Figs. 3a and 3d), while design 3 (marked in Figs. 3b and 3c) is the global optimum in both Γ- Π and X-Π projections (not exactly represented on the α-β plane). The design proposed in ref. 17, which was found through particle swarm optimization, also belongs to the sub-space of good solutions and its location on the α-β hyperplane is marked for reference. Despite the very different design parameter (especially for the L-shaped structure), all these gratings have a highly directional vertical emission and good overlap with the fiber mode, leading to a high coupling efficiency (η > 0.75 as listed in Table 1). On the other hand, the attainable back reflection differs markedly, from -21 dB for design 1 to -37 dB for design 2. The possibility to exhaustively map other metrics throughout the sub-space allows the designer to identify a design area with particularly low back-reflections around design 2. As an additional comparison of the performance for designs 1-3, Figs. 3e and 3f plot the two performance criteria η and r as a function of wavelength, now computed using 2D finite-difference-time- domain (FDTD) method as a cross-check. Results agree well with that predicted by the Fourier-type 2D simulator. All three designs have a 1-dB bandwidth larger than the telecommunication C band (1530 nm -- 1565 nm, green shaded area) with design 2 slightly out-performing the other two. Regarding back- reflections, the behaviour of the three designs is remarkably different (Fig. 3f). Back-reflections of design 2 near 1550 nm are very low which is particularly important for coupling to a laser. However, a reflection of less than -30 dB can only be achieved within a 7-nm wavelength band. In contrast, the reflection of design 1 and 3 oscillates between -26 dB and -17 dB within the entire C band. Minimum feature size determines the manufacturability of any nanophotonics device. Here feature sizes can be easily retrieved exploiting the α-β hyperplane through a query process without performing any additional photonic simulations. For each point the dimensions [L1 … L5] are computed with equation (2) and the shortest section is identified. Figure 4a shows the shortest segment among the 5 dimensional parameters. It is immediately evident that a design with minimum feature size above 88 nm does not exist for this grating structure, even accepting a small penalty in the coupling efficiency. This essential information can be easily retrieved because of the global perspective that our method offers, and it would be difficult to obtain with a conventional optimization procedure. Clearly, the bottleneck is predominantly in either L1 or L2. This finding inspired an improved grating structure (described in the next section) that allows a minimum feature size larger than 100 nm while maintaining similar performance. 8 Fig. 4 Comprehensive device characterization. The α-β hyperplane is exploited to investigate additional performance criteria. To ease comparison with Fig. 3, in each map the solid contour encloses the region of good designs (η > 0.74) while the two crosses mark design 1 and 2. a Minimum feature size for the different designs. The map highlights also for which segment this minimum size occurs, identifying three areas (dashed black lines) where L1, L2 or L3 is the shortest feature. Grating designs with minimum feature size above 88 nm do not exist. For almost the entire region of good designs the minimum feature is either L1 (e.g. for design 1) or L2 (e.g. for design 2). b-e Sensitivity of the designs to (b,c) width deviations δw of both shallow and deeply etched sections in the grating and (d,e) etch depth deviations δe for the shallow etched section. The maps show the value of the degradation derivative for (b,d) coupling efficiency η and (c,e) back-reflections r across the sub-space of good designs as defined in Methods. Coupling efficiency of design 1 is more sensitive to width deviations compared to design 2. The opposite occurs for back-reflections, where design 2 has a higher sensitivity than design 1. Both coupling efficiency and back-reflection have a low sensitivity to etch depth variations within most of the region of good designs. Another important aspect for nanophotonic devices is the robustness against unavoidable fabrication uncertainty. Here we examine two sources of common dimensional variability: A width deviation δw for both shallow and deeply etched sections and etch depth deviation δe from the nominal 110 nm for the shallow etched section. A good measure of the sensitivity of coupling efficiency and back-reflections to variability is provided by a quantity denoted here as degradation derivative, defined as the average of the two directional derivatives with respect to positive and negative values of δw or δe (definition in Methods). The computed values of the four degradation derivatives are shown in Figs. 4b-e, with a high value 9 indicating a high sensitivity. For width deviations, the coupling efficiency has a particularly sensitive region close to design 1. On the contrary, back-reflection is more sensitive to width deviations in the region close to designs 2. This region has a large overlap with the region of minimum back-reflection shown in Fig. 3d, making design 2 and surrounding designs high-performing when back-reflection is considered, but with stringent fabrication requirements. In Supplementary we provide a direct verification of these results through a polynomial-chaos-based stochastic analysis22. Regarding sensitivity to etch depth variations, the entire region of good designs largely overlaps with a region of low sensitivity for both coupling efficiency and back-reflection. Generality of the dimensionality reduction methodology and ML inspired geometry The proposed design approach requires no physical assumptions on the device under study, enabling its application to other design problems with different types of input parameters and/or objectives. A straightforward demonstration of its generality is carried out by designing vertical grating couplers for the optical communication O-band, center at 1310 nm (see Supplementary). We further demonstrate here this generality by investigating a new class of grating couplers utilizing subwavelength metamaterials3,4 to achieve designs with a minimum feature size larger than 100 nm in both the propagation and the transverse directions. This new grating geometry (see Fig. 5) is inspired by the global mapping of minimum feature size described in the previous section. The optimization now involves both dimensions and the effective material index used to represent the subwavelength segments. Dimensions and refractive index have different numerical magnitudes but they both significantly impact the device performance when varied. Below, we show that the method presented in the previous sections can successfully map out the high performance region of this new mixed design space. The subwavelength metamaterial grating structure, schematically shown in Fig. 5a, is represented by a mixed set of four geometrical parameters and one material parameter 𝐋 = [L1, L2, L3, L4, nswg]. These complex grating couplers can still be efficiently simulated using the 2D eigenmode expansion simulator. The formulation of the optimization problem remains the same as in the Eq. (1) but we additionally set 1.6 < nswg < 3. Good designs found by the optimization algorithm (η > 0.74) are used to perform both PCA and the corresponding error analysis (see Supplementary). Since the new design space includes parameters of different nature, before executing PCA it is essential to normalize the variables through their statistically estimated standard deviations. Performing the PCA analysis (stage 2) reveals that again only two principal components are sufficient to identify all the good designs within the original 5D design space (vectors are defined in Methods). In Figs. 5b and 5c, we exhaustively map out the coupling efficiency and the back-reflection over the 2D hyperplane (stage 3), discovering again a large continuous region with η > 0.74. The map covers the region where the coupling efficiency exceeds 0.7. The black contour encloses all devices with a minimum feature size larger than 100 nm in both propagation and transverse directions and maintains η  0.74. Three devices are marked on the maps for further investigation, where independent 2D FDTD simulations provide full wavelength analysis in Figs. 5d and 5e (see Supplementary for structural details and performance metrics). The results show that physically distinct devices with similar performances can be 10 Fig. 5 Vertical grating coupler with subwavelength metamaterial. a Schematic representation of the grating structure, with the subwavelength segment highlighted. The inset shows a 2D cross-section of the grating where the subwavelength metamaterial is modeled by an effective medium. Dimensionality reduction reveals that two principal components are sufficient to represent the good designs instead of the original five parameters. Plotted in b and c are the corresponding exhaustive maps of coupling efficiency and back reflection over the reduced parameters sub-space, respectively. Only designs with η > 0.7 are shown. The black contour encloses the design region that ensures η > 0.74 (as in Fig. 3) and also a minimum feature size of larger than 100 nm in both the propagation and transverse directions. These solutions were not possible with the grating structure shown in Fig. 2. d,e 2D-FDTD simulations of (d) coupling efficiency and (e) back-reflection as a function of wavelength for designs 1-3. All three have η > 0.74 and back-reflections below -15 dB within the C band. identified very efficiently. Delivering on such objectives would be very challenging by conventional optimization methods. Conclusion We have demonstrated a new approach for the design of complex photonics devices with a large number of parameters, case studied on a multi-parameter vertical fiber grating coupler. Rather than generating a single optimized design solution, our methodology exploits the dimensionality reduction technique from the suite of machine learning pattern recognition tools to identify within the large design space the lower- dimensional sub-space of good devices with high performance. This approach exponentially scales down the complexity of the problem, making it feasible to exhaustively map the continuous region of grating couplers with comparable fiber coupling efficiencies (𝜂 > 0.74 at 1550 nm). Significant differences emerge when different performance criteria are considered, such as back-reflection, minimum feature size, and tolerance to fabrication uncertainty. Such a global perspective also reveals performance and 11 structural limitations of the design geometry. In particular, we were able to conclude that good coupling efficiency and a minimum feature size larger than 88 nm could not be obtained simultaneously for this first structure. The analysis of this shortcoming inspired a new class of grating couplers that uses subwavelength metamaterial, achieving a minimum feature size of above 100 nm while maintaining state- of-the-art coupling efficiency and back-reflection. Given the generality of our implementation, the presented methodology can be readily exploited to navigate and comprehend a wide range of high-dimensional design spaces that photonic designers often encounter. While it is demonstrated here for two different design problems in nanophotonics, applications to photonic circuits or even sub-systems can be foreseen. This design methodology opens up new avenues in photonic device analysis and design where other dimensionality reduction methods such as Kernel PCA34, Principal manifolds35 and Autoencoders36 can deal with navigating through even more complex design spaces. Indeed, automation and integration of dimensionality reduction within the design flow will provide a powerful platform potentially transforming how advanced photonic devices are discovered and investigated. Methods Grating coupler simulation. The simulation of coupling efficiency and back-reflections for each design of the grating coupler is performed exploiting either a 2D vectorial Fourier eigenmode expansion simulator32 or a commercial 2D-FDTD solver. We consider a structure including silicon substrate, 2-μm buried oxide, 220-nm-thick silicon core and a silica upper cladding of 1.5 μm thickness. Silicon and silica refractive indices are 3.45 and 1.45 at λ =1550 nm. The mode of an SMF-28 single-mode optical fiber vertically coupled on top of the grating is modeled with a Gaussian function with a mode field diameter of 10.4 μm (λ = 1550 nm). The fiber facet is assumed to be in direct contact with the top of the upper cladding and its longitudinal position along the grating is optimized for each design to maximize the coupling efficiency. The latter is calculated as the overlap integral between the simulated field diffracted upwards by the grating and the Gaussian function. Machine learning enhanced optimization. We implemented a random restart local-search algorithm to solve problem (1), although other search methods could also be used. For each initial random design small perturbations are made until a better solution in terms of coupling efficiency is found and a line search is exploited to seek further improvement until convergence. The perturbation and line search process is repeated until no improvement is found in the perturbation stage. Following an initial optimization round with random-restart where a small collection of good designs was obtained, we trained a supervised machine learning model, specifically gradient boosted trees, to predict if the radiation angle is within 5° of vertical. Once trained, we used the predictor to sample random designs that are nearly vertical while rejecting other random designs. Only near vertical emitting designs proceed to the local search stage. The use of the predictor in the optimization led to approximately 250% increase in optimizer speed to find new designs meeting the coupling efficiency criteria η > 0.74. The performance of the general purpose machine learning predictor is comparable to that of a predictor based on the scalar grating equation relating the section lengths, effective indices and the radiation angle: 12 5 ∑ 𝑁𝑖𝐿𝑖 = 𝑐 + 𝑎 sin 𝜃 𝑖=1 5 , ⋅ ∑ 𝐿𝑖 𝑖=1 where 𝑎, 𝑐 are constants related to the wavelength and the overcladding effective index, 𝑁𝑖 is the effective index of the i-th section, 𝐿𝑖 is the length of the i-th section, and 𝜃 is the radiation angle. Given the simulated data (𝜃, 𝐿𝑖), one can use linear regression to estimate all the constants and then use those to predict the radiation angle of the structure for any combination of the section lengths. Despite the fact that the general machine learning predictor is not aware of this approximation, its predictions are comparable to those obtained using the scalar grating equation. Furthermore, this approach can be applied to predict other quantities that cannot be described by simple closed-form expressions. Principal Component Analysis. PCA is a dimensionality reduction technique that has been used widely and successfully across various engineering and science disciplines37 -- 40 and is implemented in most scientific computing platforms (e.g. Matlab, R, Python, etc.). Consider m data points in an n-dimensional design space. This can be written as a centered data matrix 𝐋 ∈ ℝm×n where the statistical average along each dimension is subtracted. PCA finds a sequence of best orthogonal linear projections (called principal components) that maximizes the corresponding variances. Mathematically, this is done by finding a set of vectors 𝐕1, 𝐕2, … , 𝐕n ∈ ℝn that are L2 normalized and orthogonal, ‖𝐕i‖2 = 1, 𝐕i ⊥ 𝐕j, and for every k < n, minimize ‖𝐋𝐑 − 𝐋[𝐑𝐤, 𝟎𝐧×(𝐧−𝐤)]‖ , 2 where, 𝐑 = [𝐕1, 𝐕2, … , 𝐕𝑛] ∈ ℝn×n is the transformation matrix that is formed by using the PCA vectors as its columns, 𝐑𝐤 = [𝐕1, 𝐕2, … , 𝐕𝑘] ∈ ℝn×k is similar but transforms to a lower dimensional space consisting of only k first components, and 𝟎𝐧×(𝐧−𝐤) is a null matrix used for padding. Such an exercise also returns the weights associated with PCA vectors from which the main principal components can be selected and the remaining can be dropped, with negligible amount of information lost. The original dataset can now be approximately represented by k < n effective parameters using a matrix product 𝐋PCA = 𝐋𝐑𝐤, where each row corresponds to a transformed representation of a single data point. The 𝐓 , which can be transformation back to the original n-dimensional space can be done using 𝐋est = 𝐋 𝐑𝐤𝐑𝐤 used to quantify approximation errors incurred by reducing the dimensionality. Hyperplanes definitions. The hyperplane approximation given in equation (2) and computed by PCA is defined by three 5D vectors V1, V2 and C, the latter being the reference origin within the hyperplane. The scaled vectors computed for the grating structure in Fig. 2 are: V1αβ = [-0.43, 3.78, -20.82, 44.77, -30.21] nm, V2αβ = [-25.80, 10.81, -37.69, -3.86, 21.93] nm and the origin Cαβ = [102,73,156, 243,156] nm. For the two orthogonal hyperplanes Γ-Π and X-Π described in Fig. 3b and 3c the vectors are found following linear algebra considerations. Γ-Π is defined enforcing the plane to pass through the two points (defined in the 5D design space) [77, 84, 115, 249, 171] nm (design 1 in Table 1) and [95, 83, 104, 336, 98] nm and to be orthogonal to the α-β hyperplane. The resulting vectors are V1ΓΠ = [9.45, -0.35, -6.13, 45.95, -38.35] nm, V2ΓΠ = [-22.48, 33.08, 9.87, -17.87, -28.78] nm, C ΓΠ = [85, 84, 110, 289, 138] nm. X- Π is defined as orthogonal to both α-β and Γ-Π and passing through the point [82, 87, 111, 283, 139] nm 13 (design 3 in Table 1). The vectors are V1XΠ = [-25.92, 11.96, -44.16, 10.05, 12.61] nm, V2XΠ = V2ΓΠ, and C XΠ = [85, 84, 110, 284, 142] nm. For the grating structure with sub-wavelength transverse metamaterial (Fig. 5a), the 5D vectors defining the 2D hyperplane are V1αβ = [13.37nm, -3.34nm, 17.21nm, -19.5nm, -0.03], V2αβ = [5.07nm, 14.07, -7.33nm, 2.53nm, -0.076] and Cαβ = [272nm, 71nm, 247nm, 120nm, 2.63]. Computational resources. In the proposed method, the largest fraction of the computational time is dominated by design simulations with a negligible overhead time for data processing. The initial optimization (stage 1) enhanced by the ML angle predictor required about 5000 photonic simulations to identify the initial 5 good designs used to find the reduced design sub-space through PCA. The exhaustive mapping was then performed by sampling 3600 points (design) arranged in a square grid in the sub-space and doing the corresponding 3600 photonic simulations to compute simultaneously coupling efficiency and back-reflections. Other sampling strategies or sparser grids can be used to reduce the number of simulations. Likewise, a different simulation approach could be used to retrieve additional metrics within the same simulation. The computation of additional performance metrics may require additional simulations, for example when fabrication tolerance is estimated through the degradation derivatives as reported in Figs. 4b-e. On the other hand, the computation of the minimum feature size does not require any additional simulation. It is worth calculating the number of points of a grid with the same resolution used for the α-β hyperplane but across the five grating dimensions and covering all designs with coupling efficiency η > 0.74 found in stage 1. The five dimensions of these good designs span a range of 60 nm, 27 nm, 86 nm, 138 nm, 118 nm, respectively. A grid with 5 nm resolution across all segments and covering all good designs, including two extra points to confirm the boundaries per segment, results in 14 x 7 x 19 x 30 x 26 ≈ 1.5 ∙ 106 points (designs). Compared to mapping directly the original 5D design space, mapping the α-β hyperplane thus reduces the computation time of about 400 times. For higher dimensions, the reduction can be even more significant. Uncertainty model. For the investigation of design tolerance to fabrication uncertainty we assume a width deviation δw for both shallow and deeply etched sections L1' = L1 -- δw; L2' = L2 + δw; L3' = L3 -- δw; L4' = L4; L5' = L5 + δw. We define a degradation derivative that is computed from two directional derivatives, assuming that over and under-etch are equally likely (positive and negative values of δw can in general affect the device performance differently). Calculating the common derivative would not be informative as for locally optimized devices it would be close to zero. For coupling efficiency, we are interested in calculating 𝛼𝜂 = − 1 2 ( 𝜕+𝜂 𝜕𝛿𝑤 − 𝜕−𝜂 𝜕𝛿𝑤 ) ≅ − 1 2𝛥𝛿𝑤 (𝜂+ + 𝜂− − 2𝜂0), where ∂+ and ∂- are the derivatives computed for positive and negative values of δw. The minus sign ensures that a positive value of αη indicates a worse (lower) coupling efficiency. Derivatives are numerically computed considering a small width variation and simulating the coupling efficiencies η+ (when Δδw = 5 nm) and η- (when Δδw = - 5 nm). η0 is the coupling efficiency for δw = 0. Similarly, for back-reflections 14 𝛼𝑟 = 1 2 ( 𝜕+𝜂 𝜕𝛿𝑤 − 𝜕−𝜂 𝜕𝛿𝑤 ) ≅ 1 2𝛥𝛿𝑤 (𝜂+ + 𝜂− − 2𝜂0). Also in this case a positive αr indicates worse (higher) back-reflections. The degradation derivatives plotted in Figs. 4b and 4c are finally computed as 𝑑 = { 𝛼 if 𝛼 > 0 0 otherwise When etch uncertainty is considered, δe represents the variability on the 110 nm etch depth in the fourth section of the grating. The same definitions apply for the degradation derivatives. Data availability The data that support the plots within this paper and other findings of this study are available from the corresponding author upon request. Author contributions D.M., Y.G., and D.X.X. conceived the design approach and developed the theoretical framework. Y.G. developed the machine learning algorithms. D.M. analysed the data and performed the stochastic analyses. P.C., A.S.P. and J.H.S. assisted in selecting the grating coupler study case. A.S.P. contributed to the development of the interface between the photonic simulator and machine learning algorithms. S.J., P.C., and J.H.S. provided theoretical and design guidance. D.X.X. and Y.G. supervised the project. M.K.D. and A.S.P. conceived and analysed the grating design with subwavelength patterning. All authors contributed to the discussion and manuscript preparation. References 1. Thomson, D. et al. Roadmap on silicon photonics. J. Opt. 18, 073003 (2016). 2. Smit, M. et al. An introduction to InP-based generic integration technology. Semicond. Sci. Technol. 29, 083001 (2014). 3. Cheben, P., Halir, R., Schmid, J. H., Atwater, H. A. & Smith, D. R. Subwavelength integrated photonics. Nature 560, 565 -- 572 (2018). 4. Halir, R. et al. Subwavelength-Grating Metamaterial Structures for Silicon Photonic Devices. Proc. IEEE 106, 2144 -- 2157 (2018). 15 5. Cheben, P. et al. Refractive index engineering with subwavelength gratings for efficient microphotonic couplers and planar waveguide multiplexers. Opt. Lett. 35, 2526 -- 2528 (2010). 6. Frellsen, L. F., Ding, Y., Sigmund, O. & Frandsen, L. H. Topology optimized mode multiplexing in silicon-on-insulator photonic wire waveguides. Opt. Express 24, 16866 (2016). 7. Jensen, J. S. & Sigmund, O. Topology optimization for nano-photonics. Laser Photonics Rev. 5, 308 -- 321 (2011). 8. Lu, J. & Vučković, J. Nanophotonic computational design. Opt. Express 21, 13351 (2013). 9. Piggott, A. Y. et al. Inverse design and demonstration of a compact and broadband on-chip wavelength demultiplexer. Nat. Photonics 9, 374 -- 377 (2015). 10. Shen, B., Wang, P., Polson, R. & Menon, R. Integrated metamaterials for efficient and compact free- space-to-waveguide coupling. Opt. Express 22, 27175 -- 27182 (2014). 11. Shen, B., Wang, P., Polson, R. & Menon, R. An integrated-nanophotonics polarization beamsplitter with 2.4 × 2.4 μm2 footprint. Nat. Photonics 9, 378 -- 382 (2015). 12. Liu, D., Gabrielli, L. H., Lipson, M. & Johnson, S. G. Transformation inverse design. Opt. Express 21, 14223 (2013). 13. Covey, J. & Chen, R. T. Efficient perfectly vertical fiber-to-chip grating coupler for silicon horizontal multiple slot waveguides. Opt. Express 21, 10886 (2013). 14. Håkansson, A. & Sánchez-Dehesa, J. Inverse designed photonic crystal de-multiplex waveguide coupler. Opt. Express 13, 5440 -- 5449 (2005). 15. Melati, D., Waqas, A., Xu, D.-X. & Melloni, A. Genetic algorithm and polynomial chaos modelling for performance optimization of photonic circuits under manufacturing variability. in Optical Fiber Communication Conference M3I.4 (Optical Society of America, 2018). doi:10.1364/OFC.2018.M3I.4 16. Ma, Y. et al. Ultralow loss single layer submicron silicon waveguide crossing for SOI optical interconnect. Opt. Express 21, 29374 -- 29382 (2013). 17. Watanabe, T., Ayata, M., Koch, U., Fedoryshyn, Y. & Leuthold, J. Perpendicular grating coupler based on a blazed antiback-reflection structure. J. Light. Technol. 35, 4663 -- 4669 (2017). 16 18. Jensen, J. S. & Sigmund, O. Systematic design of photonic crystal structures using topology optimization: Low-loss waveguide bends. Appl. Phys. Lett. 84, 2022 -- 2024 (2004). 19. Lalau-Keraly, C. M., Bhargava, S., Miller, O. D. & Yablonovitch, E. Adjoint shape optimization applied to electromagnetic design. Opt. Express 21, 21693 (2013). 20. Niederberger, A. C. R., Fattal, D. A., Gauger, N. R., Fan, S. & Beausoleil, R. G. Sensitivity analysis and optimization of sub-wavelength optical gratings using adjoints. Opt. Express 22, 12971 -- 12981 (2014). 21. Su, L. et al. Fully-automated optimization of grating couplers. Opt. Express 26, 4023 -- 4034 (2018). 22. Weng, T.-W., Melati, D., Melloni, A. & Daniel, L. Stochastic simulation and robust design optimization of integrated photonic filters. Nanophotonics 6, 299 -- 308 (2017). 23. Turduev, M. et al. Ultracompact photonic structure design for strong light confinement and coupling into nanowaveguide. J. Light. Technol. 36, 2812 -- 2819 (2018). 24. Peurifoy, J. et al. Nanophotonic particle simulation and inverse design using artificial neural networks. Sci. Adv. 4, eaar4206 (2018). 25. Michaels, A. & Yablonovitch, E. Inverse design of near unity efficiency perfectly vertical grating couplers. Opt. Express 26, 4766 (2018). 26. Wang, B., Jiang, J. & Nordin, G. P. Embedded slanted grating for vertical coupling between fibers and silicon-on-insulator planar waveguides. IEEE Photonics Technol. Lett. 17, 1884 -- 1886 (2005). 27. Roelkens, G., Thourhout, D. V. & Baets, R. High efficiency grating coupler between silicon-on- insulator waveguides and perfectly vertical optical fibers. Opt. Lett. 32, 1495 -- 1497 (2007). 28. Dabos, G. et al. Perfectly vertical and fully etched SOI grating couplers for TM polarization. Opt. Commun. 350, 124 -- 127 (2015). 29. Chen, X., Li, C. & Tsang, H. K. Fabrication-tolerant waveguide chirped grating coupler for coupling to a perfectly vertical optical fiber. IEEE Photonics Technol. Lett. 20, 1914 -- 1916 (2008). 30. Liu, A. Y., Komljenovic, T., Davenport, M. L., Gossard, A. C. & Bowers, J. E. Reflection sensitivity of 1.3 μm quantum dot lasers epitaxially grown on silicon. Opt. Express 25, 9535 -- 9543 (2017). 17 31. Melati, D., Melloni, A. & Morichetti, F. Real photonic waveguides: guiding light through imperfections. Adv Opt Photon 6, 156 -- 224 (2014). 32. Zavargo-Peche, L., Ortega-Moñux, A., Wangüemert-Pérez, J. G. & Molina-Fernández, I. Fourier based combined techniques to design novel sub-wavelength optical integrated devices. Prog. Electromagn. Res. 123, 447 -- 465 (2012). 33. Pearson, K. On lines and planes of closest fit to systems of points in space. Lond. Edinb. Dublin Philos. Mag. J. Sci. 2, 559 -- 572 (1901). 34. Schölkopf, B., Smola, A. & Müller, K.-R. Nonlinear component analysis as a kernel eigenvalue problem. Neural Comput. 10, 1299 -- 1319 (1998). 35. Principal Manifolds for Data Visualization and Dimension Reduction. 58, (Springer Berlin Heidelberg, 2008). 36. Kingma, D. P. & Welling, M. Auto-encoding variational bayes. ArXiv13126114 Cs Stat (2013). 37. M. Shenai, P., Xu, Z. & Zhao, Y. Applications of Principal Component Analysis (PCA) in materials science. in Principal Component Analysis - Engineering Applications (ed. Sanguansat, P.) (InTech, 2012). doi:10.5772/37523 38. Kassahun, Y. & Kebedee, T. Application of Principal Component Analysis in surface water quality monitoring. in Principal Component Analysis - Engineering Applications (ed. Sanguansat, P.) (InTech, 2012). doi:10.5772/38049 39. Pasini, G. Principal Component Analysis for stock portfolio management. Int. J. Pure Apllied Math. 115, (2017). 40. Dunteman, G. H. Principal Components Analysis. (SAGE, 1989). 18
1808.03259	1	1808	2018-08-09T17:53:58	Ultra-Conformable Free-Standing Capacitors Based on Ultrathin Polyvinyl Formal Films	[ "physics.app-ph" ]	Conformable Electronics refers to a class of electronic devices that have the ability to conformally adhere onto non-planar surfaces and materials, resulting particularly appealing for skin applications, such as the case of skin-worn unobtrusive (bio)sensors for healthcare monitoring. Conformability can be addressed by integrating basic electronic components on ultrathin polymeric film substrates. Among other basic electronic components, capacitors are fundamental ones for energy storage, sensing, frequency tuning, impedance adaptation and signal processing. In this work we present a novel approach for conformable capacitors based on a free-standing, ultrathin and ultra-conformable nanosheets of poly (vinyl formal) (PVF), which serve both as structural and dielectric component of the capacitor. A novel fabrication approach is proposed and applied to fully free-standing ultrathin capacitors fabrication, having an overall thickness as low as 200 nm; that represents, to the best of our knowledge, the thinnest free-standing capacitors ever reported. Thanks to the ultra-low thickness, the proposed capacitors are able to sustain flexure to extremely small curvature radii (as low as 1.5 {\mu}m) and to conform to complex surfaces, such as a nylon mesh with micrometric texture without compromising their operation.	physics.app-ph	physics	Pre-Print Manuscript Ultra-Conformable Free-Standing Capacitors Based on Ultrathin Polyvinyl Formal Films Jonathan Barsotti, Ikue Hirata, Francesca Pignatelli, Mario Caironi, Francesco Greco, Virgilio Mattoli* J. Barsotti, F. Pignatelli F. Greco, V. Mattoli Center for Micro-BioRobotics @SSSA, Istituto Italiano di Tecnologia Via R. Piaggio, 34, 56025 Pontedera, PI, (Italy) E-mail: [email protected]; [email protected] I. Hirata, M. Caironi Center for Nano Science and Technology @PoliMi, Istituto Italiano di Tecnologia Via Pascoli 70/3, 20133 Milano, MI, (Italy) F. Greco Institute of Solid State Physics, Graz University of Technology, Petersgasse 16, 8010 Graz, Austria E-mail: [email protected] F. Greco Department of Life Science and Medical Bioscience, Graduate School of Advanced Science and Engineering, Waseda University, 2-2 Wakamatsu-cho, Shinjuku-ku, 169-8480 Tokyo, Japan. Keywords: conformable capacitor; ultrathin dielectric; electronics; tattoo electronics. freestanding dielectric; conformable Abstract - Pre-Print Manuscript Conformable Electronics refers to a class of electronic devices that have the ability to conformally adhere onto non-planar surfaces and materials, resulting particularly appealing for skin applications, such as the case of skin-worn unobtrusive (bio)sensors for healthcare monitoring. Conformability can be addressed by integrating basic electronic components on ultrathin polymeric film substrates. Among other basic electronic components, capacitors are fundamental ones for energy storage, sensing, frequency tuning, impedance adaptation and signal processing. In this work we present a novel approach for conformable capacitors based on a free-standing, ultrathin and ultra-conformable nanosheets of poly (vinyl formal) (PVF), which serve both as structural and dielectric component of the capacitor. A novel fabrication approach is proposed and applied to fully free-standing ultrathin capacitors fabrication, having an overall thickness as low as 200 nm; that represents, to the best of our knowledge, the thinnest free-standing capacitors ever reported. Thanks to the ultra-low thickness, the proposed capacitors are able to sustain flexure to extremely small curvature radii (as low as 1.5 μm) and to conform to complex surfaces, such as a nylon mesh with micrometric texture without compromising their operation. 1. Introduction Organic electronics is enabling new technological applications in which electronic circuits are asked to be deformable, as in the case of flexible[1], stretchable[2] as well as conformable devices[3]. The great improvements in materials properties[4,5] and fabrication technologies[6 -- 9] enabled constant developments toward applications, such as flexible displays[10], electronic paper, radio-frequency identification (RFID) tags, smart cards[11], electronic skin[12,13], wearable monitoring devices [14]. Pre-Print Manuscript The term Conformable Electronics is used to refer to a class of electronic devices that, thanks to their reduced thickness and peculiar construction, conformally adhere onto non-planar surfaces and materials[15]. Thickness reduction also implies a total mass reduction and an increase in the aspect ratio values, factors that strongly affect adhesion and conformability. Under these conditions, Van der Waals forces become predominant, with a consequent adhesion improvement[16]. Conformability of devices is particularly appealing when skin is the target surface of interest. This is the case of skin-worn unobtrusive sensors, and more general of biosensors, to be used in biomedical, healthcare[15], sport activity and environment monitoring systems[15,17] applications. Indeed, in all cited applications it is highly demanded that devices are the least perceivable possible[18,19]. From a technological point of view empowering electronic devices with the ability to conformally adhere to a complex surface (such as skin) without compromising their functionality is very challenging. Mechanical instabilities such as buckling, crumpling, cracking, wrinkling, dewetting and swelling can set a severe limit to the deformation the device can sustain without losses of electrical performances, and eventually lead to complete failure. Moreover, since an obvious and main strategy to accomplish this task is to decrease the overall thickness and inherent stiffness of substrates and devices, there are intrinsic difficulties related to manipulation and processing of such thin systems. In order to try to overcome these difficulties, several interesting solutions have been proposed in literature. In this vision, a novel temporary tattoo approach that makes use of commercial tattoo paper as a temporary substrate for ultra-conformable electrodes enabling an easy water-based transfer directly on skin, was proposed by Zucca et al [15]. In their work, the authors developed an unconventional way to fabricate ultra-conformable EMG electrodes using ultrathin conductive poly(3,4- ethyllendioxythiophene): poly(styrenesulfonate) (PEDOT:PSS) nanosheets. Following the temporary Pre-Print Manuscript tattoo approach proposed by Zucca et al., Ferrari et al. proposed a novel dry and perforable ultra- conformable skin-contact electrode[17]. Thanks to the tattoo-like nature, the unperceivable electrode can be directly transferred on cleaned skin as well as on hairy zones. Indeed, the ultralow electrode thickness allow hairs to perforate and grow through it, without significant performances degradation over a period as long as 48h. An analogous temporary tattoo approach has recently enabled novel applications in edible electronics[20] and organic photovoltaics (OPV)[21]. Similar applications have been enabled by using a spin coating or even a roll-to-roll (R2R) process for preparing free/standing PEDOT:PSS and PEDOT:PSS/ poly (D,L lactic acid) nanosheets[22 -- 24]. Anyway, in order to address complex tasks, conformable sensors and devices, comprising Organic Light-Emitting Diodes (OLED), OPV and Organic Thin Film Transistors (OTFT)[25 -- 27], must be integrated onto ultrathin polymer films. Nevertheless, basic electrical components such as resistances, inductances and capacitors are required as well. In particular, capacitors are fundamental components for energy storage, sensing, frequency tuning functionalities, impedance adaptation and signal processing. To this aim, a few groups tried to fabricate conformable capacitors. A noteworthy example is provided by Luan et al., who fabricated a free-standing epidermal super- capacitor assembling a H2SO4-PVA based electrolyte with two hybrid electrodes made by single-walled carbon nanotubes coated with PEDOT:PSS deposited by cyclic voltammetry[28]. Authors reported a good capacitance density value of 56 F g-1, estimated considering the electrodes mass only, and demonstrated the possibility to bend and stretch the device, but they did not investigate conformability. Moreover, 1 μm was indicated as the lowest feasible thickness of these capacitors, since thinner films caused the occurrence of short circuit. Thickness (t) is indeed a key parameter in capacitor's design, which also strongly affects conformability. Maintaining suitable electrical properties when reducing the thickness of capacitors to very low values, such as sub-µm range is Pre-Print Manuscript challenging. In this work we propose a novel approach for fabricating conformable capacitors based on water- mediated (or air-mediated) lamination of free-standing ultrathin (UT) and ultra-conformable (UC) membranes, ensuring appealing impedance characteristics and spectral response. The capacitor is fabricated using free-standing, UT and UC nanosheets of poly (vinyl formal) (PVF), which serves both as structural (mechanical self-support, conformal adhesion) and as functional (dielectric) component of the capacitor. This approach enables a drastic reduction of the dielectric membrane thickness, down to 10 nm (170 nm of total thickness), and of mass, achieving a remarkable surface density of 1.48 mg/cm2, while retaining a large surface area. With the same approach we demonstrated here the thinnest free free-standing UT capacitors ever reported, having an overall thickness (including top and bottom electrodes) as low as 200 nm, also showing very nice electrical performances on relatively large area (centimeter square scale). Moreover, capacitors show a wide working frequency, displaying a constant capacitance up to 280 kHz, with very low leakage current in the order of 10-8 A/cm2 at 0.5 MV/cm, and with an average dielectric strength of 1 MV/cm. The broad operative frequency bandwidth, combined with a high breakdown voltage, make these devices suitable for conformable electronic applications in which higher frequencies and voltages are required. In fact, they can provide an alternative to conformable electrolytic capacitors, typically having much lower cut-off frequencies and operation voltage limits[28,29]. Finally, thanks to the ultra-low thickness, the capacitors are able to sustain flexure to extremely small curvature radii (as low as 1.5 μm) and to conform on complex surfaces[30], such as a nylon mesh with micrometric texture. Pre-Print Manuscript Overall this work illustrates the potentialities of a nanosheet based approach in conformable capacitors fabrication, envisaging it as a powerful tool in realization of other truly conformable devices. 2. Results and Discussions 2.1. Capacitor structure UT capacitors have been developed by performing multiple recollections of various PVF nanosheet layers. To produce freestanding PVF nanosheets we used the spontaneous delamination process developed by Baxamusa et al.[31]. We selected PVF as host material because it is has very good dielectric properties, it is already adopted at industrial level (e.g. as insulator for electrical wires), and because Baxamusa et al. demonstrated that it is suitable to obtain robust ultrathin freestanding films. Si wafer functionalization, implemented by spin coating deposition of a sub-nm layer of a cationic polyelectrolyte, Poly(diallyldimethylammonium chloride) (PDAC), enabled to deposit by spin coating PVF films that spontaneously delaminate and float when immersed into a water bath. A challenge we had to face in order to adopt such nanosheets to fabricate a capacitor was the deposition of the electrodes, since PVF nanosheets could sporadically display microscopic pin-holes, in which a metal electrode, such as sputtered Au, can easily penetrate, thus short-circuiting the device. Our strategy to avoid short circuit across electrodes was to pile up, through multiple recollection, three different layers for assembling a capacitor: a first PVF nanosheet carrying the bottom sputtered Au electrode (layer 1); an intermediate pure PVF dielectric membrane (layer 2); a top PVF nanosheet carrying the top Au electrode (layer 3). The full UT and UC capacitor fabrication process is shown in Figure 1 (a-c). By adopting this strategy, the probability of observing a short-circuit is drastically reduced, since it is very unlikely for two pin-holes of different layers to overlap when two nanosheets are recollected one on top of the other (see also Figure S1 in Supporting Information (SI) for PVF nanosheets pin-holes Pre-Print Manuscript analysis). Electrodes (layers 1 and 3) were fabricated by Au sputtering deposition (tAu = 40 nm) on a thin PVF nanosheet (nominal thickness tsup = 40 nm), before the release from the temporary substrate (total electrode thickness t1, t3 = 80 nm). A 40 nm-thick PVF nanosheet was found to be the thinnest one able to well support the Au sputtered electrode without being damaged by recollection/assembling process. A PVF layer ( 10 ≤ t2 ≤ 160 nm) composed the dielectric layer 2. After release in water, each nanosheet could be recollected onto solid surfaces (i.e. glass) or on plastic frames (i.e. PMMA rings) for further processing or assembling (see also Figure S3 in SI for floating nanosheets recollection sequence). After each recollection, all the nanosheets were dried in air in vertical position to facilitate water removal and under ambient conditions before recollection of subsequent layers. Performing multiple recollections of the various nanosheet layers, supported capacitors on glass were fabricated (Figure 1 (b)) in order to firstly asses devices functionality. Then, following the same recollection procedure, fully freestanding capacitors where assembled on plastic ring (Figure 1 (c)). The capacitor active area is determined by the orthogonal overlapping of the two electrodes forming a square with a nominal area of 25 mm2. This value was corrected taking into account the actual covered Au surface, estimated by SEM images analysis (Figure S2 in SI). In fact, homogeneously distributed cracks were found in the Au deposited electrodes, reducing the actual conducting surface. Pictures of supported and free-standing devices are shown in Figure 1 (d) and (e), respectively. Looking at the devices reported in Figure 1 (d), the PVF dielectric layer thickness variation can be appreciated thanks to the chromatic change due to light interference. The extreme colour uniformity in each sample is also a clear indication of the high uniformity of thickness of the dielectric nanosheets. The whole procedure is described in more detail in the Experimental Section (see also Pre-Print Manuscript Figures S4 and S5 for more details on device structure). 2.2. Capacitors electrical characterization Capacitance values will be reported in the following with the notation 𝐶𝑥 𝑡(𝑓), where x subscript indicates supported or free-standing (x = sup or free, respectively) capacitor, t is the dielectric (layer 2) thickness expressed in nm and f is the frequency. The same notation will be used for cut-off frequency, 𝑡, defined as the frequency value at which capacitance reduces to the 80% of its value at 1 kHz, i.e. 𝑓𝑥 𝐶𝑥 𝑡(𝑓𝑥 𝑡) = 0.8×𝐶𝑥 𝑡(1 kHz). In Figure 2 (a) we report the capacitance at 1 kHz versus capacitor dielectric thickness in the range 10 ≤ t2 ≤ 160 nm. Data can be simply fit by adopting a parallel plates capacitor model, thus considering the capacity C as follows: 𝐶 = 𝜀0𝜀𝑃𝑉𝐹 𝐴 𝑡+ℎ , (1), where 𝜀0 is the vacuum dielectric constant, 𝜀𝑃𝑉𝐹 is the PVF relative dielectric constant, A is the capacitor area and t is the thickness of dielectric (layer 2). The additional parameter h describes the presence of the supporting PVF layer of the top electrode that acts as an extra dielectric layer, actually increasing the dielectric thickness. Thanks to this extremely low thickness, capacitors' aspect ratio spans from 10-5 to 10-4, fulfilling the flat capacitor model requirements. The fit curve (red line, Figure 2 (a)) was obtained fixing A and h to the measured values, Am = 21.25 mm2 and hm= 41 ± 2 nm, respectively. The fit, performed using Eq. (1), allows to extract a dielectric constant value for 𝜀𝑃𝑉𝐹 of 3.76 ± 0.08, with a determination coefficient value of R2 = 0.9817. Such a determination coefficient value confirm the goodness of the fit. The calculated value of 𝜀𝑃𝑉𝐹 is in perfect agreement with the value of 3.7 found by Khare et al.[32] and with values reported for Pre-Print Manuscript various commercially available PVF materials[33]. In Figure 2 (b) a typical impedance module whit relative phase spectra for a capacitor with the thinnest PVF dielectric layer, t2 = 10 nm, is shown. This is the thinnest PVF nanofilm we are able to produce, by releasing it in water and recollecting it on substrate while maintaining the film integrity on a centimeter scale. With such dielectric thickness we measured an average low frequency capacitance of 12.3 ± 0.6 nF at 1 kHz, corresponding to 49 ± 3 nF/cm2. From 20 Hz to the cut-off frequency, falling at 𝑓𝑠𝑢𝑝 10 = 120 ± 40 kHz as average on six devices, the Z module linearly decreases with slope -1, and the phase remains approximately constant at the value of 89. For f > fc, the impedance of the device with t2 = 10 nm tends to the series resistance (Rs = 64 Ω) (originated by electrode and contact resistance) and with the phase correspondingly dropping down toward 0° (see also Figures S6 (a -- h) for other impedance module and phase spectra). The corresponding capacitance spectrum of the capacitor with t2 = 10 nm, which is flat until the cut-off frequency, is shown in the inset of Figure 2 (b). An operating range up to about 100 kHz is in the region of interest of many electronic applications, such as conformable sensors[34,35] and biosensors[36]. Indeed it is typically used in transduction of slowly varying physical and chemical quantities, typically lacking high frequency components. Despite all supported devices were manually assembled, the shape of the capacitance spectra of the supported capacitors having different dielectric thickness t2 showed very similar frequency dependence respect to the case with t2 = 10 nm (see Figures S7 (a - h) in the SI), showing fc decreasing when t increases (Figure 2 (c)), being cut-off frequency the inverse of the RC time constant of the devices. As an example, also the typical impedance module and phase spectra for a capacitor with the Pre-Print Manuscript thickest PVF dielectric layer, (t2 = 160 nm) is also reported in Figure 2 (d). 2.3. Leakage current and dielectric strength Leakage current density J of supported capacitors having different t2 was measured and reported as a function of the applied electric field E in Figure 2 (e). J is plotted versus the effective PVF thickness teff = t2 + tsup, where tsup = 48 nm is the PVF electrode-supporting nanosheet thickness for the specific samples run. Tested devices showed comparable and remarkably low current densities in the range of 1 -- 3×10-8 A/cm2 for E  0.5 MV/cm. As an exception, the thickest device (teff = 217 nm) showed even lower leakage current density, in the range of 6×10-9 < J < 8×10-9 A/cm2 in the same conditions. The highest J before device breakdown was observed in the thinnest device (teff = 63 nm) and it was 5.7×10-8 A/cm2 at a remarkable E = 0.84 MV/cm. Devices' dielectric strength was tested by applying a linear positive voltage ramp until electrical breakdown. The breakdown points of the capacitors were clearly visible in the sudden increase of the current density for each curve. We obtained a remarkable average dielectric strength up to E = 1.0 ± 0.3 MV/cm. Such result well compares with other commercially available PVF formulations such as Vinylec Series, with dielectric strength in the range of 0.26 -- 0.39 MV/cm[33]and Formex, with a reported dielectric strength of 1.57 MV/cm[37]. This fact indicates that, thanks to our device fabrication strategy, the intrinsic PVF dielectric strength is not reduced in ultra-thin films, thus underlining the substantial absence of severe structural defects (see also Figure S8 in SI). 2.4. Free-standing capacitors Pre-Print Manuscript Free-standing capacitors were assembled by recollecting the various nanosheet layers as suspended membranes across plastic frames (Figure 1 (c)).With this approach we successfully produced freestanding capacitor with a t2 down to 40 nm, thus with a total thickness of about 200 nm (see Figure S9 in SI for details). Typical impedance module and phase spectra for a free-standing capacitor with t2 = 160 nm are shown in Figure 2 (f). As observed for supported devices, Z module linearly decreases with slope -1 and phase remains approximately constant to the value of 89° until fc. Beyond cut-off the impedance tends to series resistance (Rs = 23 Ω) and phase drops down toward 0°. The corresponding capacitance spectrum is reported in the inset of Figure 2 (f). For the free-standing device we measured an average capacitance 𝐶𝑓𝑟𝑒𝑒 160 = 4.32 ± 0.01 nF, corresponding to 17.28 ± 0.01 nF/cm2 and a capacitance density of 0.4 mF∙g-1, computed on the whole device mass. The measure is in agreement with the value found for supported devices of the same thickness. Moreover, also free-standing capacitors show a flat capacitance curve from low frequencies until the cut-off point, consistently to what observed in supported devices. For the free-standing capacitors fc is generally higher than for the supported one at the same thickness. Indeed, the free-standing capacitors had an average cut-off frequency of 𝑓𝑓𝑟𝑒𝑒 160 = 1.43 ± 0.07 MHz, approximately one order of magnitude higher compared with supported devices (𝑓𝑠𝑢𝑝 160 = 0.28 ± 0.03 MHz). This difference is ascribable to the lower series resistance observed in the free-standing device probably due to contact differences originated by the device contacting (see Figure S10 in SI). 2.5. Mechanical test (SIEBIMM) Mechanical properties of PVF nanosheets were evaluated by the "Strain-induced buckling instabilities for mechanical measurements" (SIEBIMM)[38]. This method allows measuring the Young's modulus of UT nanosheets and it has been applied to several materials [39,40]. An elastomeric substrate (PDMS) was pre-stretched and the nanosheet was recollected on it. By relaxing the pre-imposed strain, a Pre-Print Manuscript quasi-periodic pattern of wrinkles was formed on the surface of the PVF nanosheet, aligned along a direction perpendicular to the applied pre-strain. By measuring the wavelength λ of the wrinkles, it was possible to calculate the Young's modulus E of the nanosheet from the equation 𝐸𝑃𝑉𝐹 = 3𝐸𝑠 2 1−𝜈𝑃𝑉𝐹 1−𝜈𝑠 2 ( 𝜆 2𝜋𝑡 3 ) , (2), where the subscript s stands for "substrate" (PDMS), 𝜈 is the Poisson's ratio, and t is the nanosheet thickness. In Equation (2) we used Es = 2.06 MPa (extrapolated from data reported in [41]) and 𝜈s = 0.5 for the PDMS substrate[41]. For PVF Poisson's ratio we used 𝜈PVF = 0.33, a value typically used for viscoelastic polymers (such as PVF)[42,43] . We performed SIEBIMM test on a PVF nanosheet with t2 = 140 ± 2 nm. Surface profile measurements were carried out on wrinkled samples (pre-strain released) with an optical profilometer (DCM 3D, Leica) and analyzed by evaluating the Power Spectrum Density Function (PSDF), as described in SI (see Figure S11). From the PSDF analysis, we found a single main peak at k = 0.76 ± 0.06 𝜇 m-1, corresponding to a wavelength 𝜆 = 8.3 ± 0.7 𝜇m. This value of 𝜆 was used in Equation (2) obtaining a Young modulus for the PVF nanosheet (EPVF) of 6 ± 2 GPa. Wang et al. found a Young's modulus value of 3.08 GPa for a membrane of poly(vinyl formal-acetal) as thick as 32 𝜇m[44]. In addition, apparent Young's modulus values in the range 2.7 - 3.1 GPa are reported for a specific commercial PVF formulation (Formvar)[45]. Since there is no prior reference in literature to PVF nanosheets and to similar SIEBIMM measurements, a direct comparison with our results is not possible. However, the value we obtained is higher compared to cited data. Such an increase in E could be explained by residual stress in the samples, easier to be induced because of the very low thickness. At the same Pre-Print Manuscript time, we cannot a priori exclude that some relaxation effect of the polymeric chains, induced by the nanostructure, could take place affecting the mechanical properties of PVF nanosheets. Indeed behavior of nanoscale polymer thin films can significantly deviate from that of bulk materials[46]. While an accurate investigation of this phenomenon is beyond the scope of this work, it could help in future to clarify whether PVF shows such "confinement" effects. 2.6. Capacitors transferred onto target surfaces In order to prove the transferability of our devices and the conformal adhesion to complex surfaces we transferred a free-standing capacitor sample with t2 = 90 nm onto a nylon mesh (see optical image on Figure 3 (a) and SEM image of active area on Figure 3 (b)). It can be noticed that there is no significant difference in impedance module and phase spectra (see Figure 3(c)), as well as in the capacitance (inset), if compared with the general trends observed in both free-standing and supported devices. The capacitance of the transferred device is 𝐶𝑓𝑟𝑒𝑒 90 = 5.9 nF, in agreement with the value of 𝐶𝑠𝑢𝑝 90 = 6.2 nF measured for supported device with same thickness. The cut-off frequency we measured for the transferred capacitor (𝑓𝑚𝑒𝑠ℎ 90 ) is 0.36± 0.01 MHz, higher than the one observed for supported device with similar thickness. A SEM image of a bare PVF nanosheet collected on the mesh and an image of the mesh (inset) are reported in Figure 3 (d) in order to show PVF conformability on such a complex surface. It is noteworthy how the Au electrode is affecting the conformability, by comparing micrographs of the electrode area (stacked PVF and double electrode layers, Figure 3 (c)) with the sole PVF layer (Figure 3 (d)). This demonstrates that the free-standing capacitor maintained its structural integrity and it was still perfectly working after the transfer onto a complex micro- structured surface such as the mesh's one. Pre-Print Manuscript In order to evaluate conformability characteristics of the device, a freestanding UT capacitor was transferred on an extremely sharp object, such as a scalpel (inset: Figure 3 (e)). By means of a Focused Ion Beam (FIB) milling we excavated a groove in the transferred device. In Figure 3 (e) a SEM image of the groove is reported. The image shows the device adaptation along the narrow scalpel cutting edge, where the curvature radius r reaches its smaller value of approximately 1.5 µm, as graphically estimated by SEM image. Along the clean groove edge the device layered structure is revealed, not showing any damage in the layers despite local bending at r  1.5 µm. The layer structure is visible with the darker PVF dielectric layer sandwiched between the two conductive and brighter Au electrodes. Top electrode had an additional Au layer, deposited after device transfer, for the sake of improving quality of SEM imaging. In order to compare capacitor structure in the bent case with the flat case of the supported one (on glass) a SEM image of a supported capacitor cross section is also reported in Figure 3 (f). 3. Conclusions In this work we proposed a novel approach to fabricate UT and UC free-standing capacitors based on a novel nanosheet approach, where a PVF layer is employed as both dielectric and structural conformable layer. PVF was selected as a candidate material thanks to its unique combination of good electrical properties with excellent mechanical resistance. By using this approach, free-standing capacitors with various PVF dielectric layer thickness in the range from 10 nm to 160 nm, and Au-PVF electrodes were fabricated. Free-standing capacitors showed extremely good conformability, sustaining micrometric curvature radii. This capability was combined with remarkable mass and capacitance densities of 1.48 mg/cm2 and 0.4 mF∙g-1 respectively, with RC limited cut-off frequency extending up to the MHz, leakage density currents as low as 6×10-9 A/cm2 and breakdown field of 1.0 MV/cm. To show actual devices capabilities a capacitor with a dielectric layer as thin as 90 nm was Pre-Print Manuscript transferred on a micro-structured fabric mesh, as an example of surface with complex topography at the microscale. Device characterization performed after the transfer revealed a correct device operation, consistent with that observed on same devices supported on solid substrates. Once released, the capacitors maintained their electrical characteristics while conforming to the target surface, thus confirming the success of the proposed approach. Finally, by transferring a free-standing capacitor onto a scalpel blade we quantitatively demonstrated the extremely high conformability of the structure that is able to sustain a curvature radius as low as 1.5 µm without breaking. This candidates our UC free standing capacitors to be used as deformation sensor, storage energy system as well as to be integrated in more complicated conformable circuits, tracing a promising path towards a new generation of truly unperceivable and ultra-conformable electronic circuits and systems. Finally, it is worth to note that ultrathin freestanding dielectric lamination approach can be directly extended to build other devices requiring high quality dielectric, such as low voltage organic transistors, an application that is under investigation in our group. Pre-Print Manuscript 4. Experimental Section 4.1 Materials Poly(vinyl formal) (PVF) was purchased from Sigma-Aldrich(trade name Vinylec K ). Ethyl lactate (EL, ≥ 98%, FCC, FG) and polydiallyldimethylammonium chloride (PDAC, 20 wt% in water solution) were purchased from Sigma-Aldrich. Si wafer (3" diameter and are 381 ± 25 µm thick) used for film fabrication were purchased by Silicon Materials (Si-Mat). Supported capacitor were assembled on Plain Micro-Slides, used as glass substrate and purchased by VWR International. Poly(dimethylsiloxane) (PDMS, Sylgard 184 silicone elastomer base and curing agent) was purchased by Dow Corning Corp. The nylon mesh (Sefar Nytal HC-58) was purchased from Sefar Co. Ltd., Switzerland. The Ag paste used for contacting all the devices was a bi-component Conductive Silver Epoxy paste, purchased by CircuitWorks®. 4.2 Methods PVF nanosheets preparation. A PDAC solution at 0.5 % wt was prepared by diluting the commercial 20 % wt water solution with deionized water. PDAC solution was filtered immediately before use with a cellulose acetate filter ( 0.45 µm pores diameter, MinisartTM). PVF solutions (1 %, 1.5 %, 2 %, 2.5 %, 3 %, 3.5 %, and 4 % in weight) were prepared by dissolving PVF in EL according to a procedure presented elsewhere[31]. Solutions were stirred at 650 rpm until all PVF was dissolved. PVF Solutions were filtered immediately before use, with a hydrophobic filter (0.2 µm pores diameter, MinisartTM). A Si wafer was first functionalized with a subnanometric PDAC layer[31]. PDAC solution was spin coated at 4000 rpm for 15 s and then baked at 100 °C for 10 s on a hotplate. About 2 ml of PDAC solution were deposited for each nanosheet spinning session. Extra PDAC was rinsed with deionized water and then the surface was dried with a compressed air gun. PVF solution was then spin coated over the Pre-Print Manuscript PDAC layer in a 2-steps process of a total duration of 10 s. During the first 5 s, spin coating speed was set at 300 rpm while during the remaining 5 s it was set at 3000 rpm. About 2 ml were deposited for each nanosheet PVF spinning session (see also Figure S11 in SI for PVF spinning characterization curve). PVF was then baked for 60 s at 50 °C on a hotplate in order to remove extra solvent. PVF nanosheets thickness characterization was performed by means of a P-6 stylus profilometer (KLA Tencor, USA) averaging over 15 estimated values randomly acquired on the film surface before the release from Si wafer. Same measurements were carried out on nanosheets after release in water and recollection onto a fresh Si wafer. Anyway, no differences were found in PVF nanosheet thickness before and after PVF nanosheet release/recollection. Capacitor assembly. We used plain micro-slide glass substrates for assembling supported capacitors and ring shaped polymethylmetacrylate circular frames (with 1 cm diameter of inner-hole) for recollecting and assembling the free-standing membrane capacitors. Both top and bottom electrodes were fabricated on 40 nm-thick PVF nanosheets supported on Si wafer, by Au sputtering deposition (70 W for 25 s at 10-2 mbar in Ar; nominal Au thickness tAu = 40 nm) through a physical mask (5 mm x 20 mm). A custom-made sputtering system assembled with Leybold components by Sistec was used. Rectangular electrode layers were shaped cutting the nanosheet with a surgical blade and then delaminated in water using a home-made immersing system for a controlled immersion of wafers at the fixed angle of 40°. Supported capacitors were manually assembled by immersing the glass substrate into water. Recollection of the floating bottom electrode was performed by pulling the glass slide out of the water with a fixed angle (ca. 90°) and allowing one edge of floating electrode to adhere to the glass and to be consequently pulled away from water surface; see Figure S2 in SI. A thin water film between recollected nanosheet and glass substrate facilitated film relaxation. The PVF dielectric layer was then Pre-Print Manuscript recollected on top of the bottom electrode (previously recollected and dried), following the same recollection procedure. Finally, top electrodes were recollected on top of the PVF dielectric layer with the same recollection procedure. Electrodes recollection was performed maintaining the right angle between electrode Au strips (approximately 90°) with an overlapping area of 25 mm2. Before each successive recollection step, the previously recollected layer was dried under ambient conditions to avoid nanosheet delamination during the successive recollection. Free-standing capacitors were also manually assembled with a similar stepwise procedure on top of circular frames. Once all the layers of capacitors (both supported and free standing) were recollected, electrical contacts were made by fixing electrical wire to Au electrodes by means of conductive Ag paste (CircuitWorks® Conductive Silver Epoxy). The percentage of the electrode area covered by Au was estimated to be around 85 % by analyzing the SEM images of electrode surfaces with MatLab software through thresholding operation (details reported in SI). An actual area of 25 mm2 x 0.85 = 21.25 mm2, was used in the calculations reported in the work. Capacitor electrical characterization. Impedance and capacitance spectra of on-glass supported capacitors were recorded in the frequency range 20 Hz - 2 MHz, by means of an E4980AL Precision LCR Meter (Agilent, now Keysight Technologies). Capacitors with 10 ≤ t2 ≤ 160 nm were characterized. Capacitance spectra were calculated from impedance module and phase, describing the device with an equivalent parallel circuit between an ideal resistor and an ideal capacitor. Capacitors transfer onto mesh. The free-standing capacitor was transferred on the nylon mesh with a wet process in order to attain improved conformability to the mesh surface. We previously moistened the mesh with water. We fixed the mesh onto a rigid transparent substrate of PMMA to manipulate it Pre-Print Manuscript and to electrically contact the device with conductive Ag paste. Mechanical test (SIEBIMM). Si wafers were functionalized in a bell desiccator with chlorotrimethylsilane vapours for 60 minutes in order to improve PDMS detachment. Crosslinking of PDMS was obtained by mixing the PDMS elastomer base with the curing agent (10:1 ratio by weight). The mixture was then degassed in a vacuum bell desiccator in order to remove air bubbles. PDMS was then deposited by spin coating onto a pre-functionalized Si wafer at a speed of 200 rpm for 60 s and cured at T = 95 °C for 60 min in a convection oven. The cured PDMS was cut into rectangular slabs (4 x 1.5 x 0.3 cm3) and pre-stretched to 5%. PVF nanosheets were recollected onto pre-stretched PDMS slabs from water and then dried in vacuum at RT overnight prior to the SIEBIMM test. The strain of the PDMS substrate was finally relaxed, producing the buckling of the PVF nanosheet. The buckling wavelength of the nanosheet was measured by an optical profilometer (DCM 3D, Leica), operating in confocal mode with an objective EPI 150X-L. The formula used to calculate the Young's modulus of the PVF nanosheets is reported in equation (2). Acknowledgements We acknowledge Mr. Carlo Filippeschi for Au sputtering deposition and FIB Dual Beam operations. M.C. acknowledges support by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program "HEROIC," Grant Agreement No. 638059. F.G. acknowledges financial support from Top Global University Program at Waseda University, Tokyo from MEXT Japan. V.M. acknowledges support by the RoboCom++ FLAG-ERA JTC 2016 project. References Pre-Print Manuscript [1] M. Caironi, Y.-Y. Noh, Eds. , Large Area and Flexible Electronics, Wiley-VCH, 2015. [2] T. Sekitani, T. Someya, Adv. Mater. 2010, 22, 2228. [3] T. Someya, S. Bauer, M. Kaltenbrunner, MRS Bull. 2017, 42, 124. [4] J. Onorato, V. Pakhnyuk, C. K. Luscombe, Polym. J. 2016, 49, 1. [5] Y. S. Rim, S. H. Bae, H. Chen, N. De Marco, Y. Yang, Adv. Mater. 2016, 28, 4415. [6] M. Eslamian, Nano-Micro Lett. 2017, 9, 1. [7] D.-H. Kim, R. Ghaffari, N. Lu, J. A. Rogers, Annu. Rev. Biomed. Eng. 2012, 14, 113. [8] a. a. Bessonov, M. N. Kirikova, Nanotechnologies Russ. 2015, 10, 165. [9] J.-H. Ahn, J. H. Je, J. Phys. D. Appl. Phys. 2012, 45, 103001. [10] Z. Li, H. Meng, Eds. , Organic Light-Emitting Materials and Devices, CRC Press, 2007. [11] F. M. Li, A. Nathan, Y. Wu, B. S. Ong, Organic Thin Film Transistor Integration, John Wiley & Sons, 2011. [12] S. Bauer, Nat. Mater. 2013, 12, 871. [13] M. L. Hammock, A. Chortos, B. C. K. Tee, J. B. H. Tok, Z. Bao, Adv. Mater. 2013, 25, 5997. [14] Y. Liu, M. Pharr, G. A. Salvatore, ACS Nano 2017, 11, 9614. [15] A. Zucca, C. Cipriani, Sudha, S. Tarantino, D. Ricci, V. Mattoli, F. Greco, Adv. Healthc. Mater. 2015, 4, 983. [16] B. N. Chapman, J. Vac. Sci. Technol. 1974, 11, 106. [17] L. M. Ferrari, S. Sudha, S. Tarantino, R. Esposti, F. Bolzoni, P. Cavallari, C. Cipriani, V. Mattoli, F. Greco, Adv. Sci. 2018, 1700771. [18] T. Sekitani, T. Someya, Mater. Today 2011, 14, 398. [19] M. Kaltenbrunner, T. Sekitani, J. Reeder, T. Yokota, K. Kuribara, T. Tokuhara, M. Drack, R. Schwödiauer, I. Graz, S. Bauer-Gogonea, S. Bauer, T. Someya, Nature 2013, 499, 458. [20] G. E. Bonacchini, C. Bossio, F. Greco, V. Mattoli, Y.-H. Kim, G. Lanzani, M. Caironi, Adv. Mater. 2018, 1706091. [21] N. Piva, F. Greco, M. Garbugli, A. Iacchetti, V. Mattoli, M. Caironi, Adv. Electron. Mater. 2017, 1700325, 1. [22] F. Greco, A. Zucca, S. Taccola, A. Menciassi, T. Fujie, H. Haniuda, S. Takeoka, P. Dario, V. Mattoli, Soft Matter 2011, 7, 10642. Pre-Print Manuscript [23] F. Greco, A. Zucca, S. Taccola, B. Mazzolai, V. Mattoli, ACS Appl. Mater. Interfaces 2013, 5, 9461. [24] A. Zucca, K. Yamagishi, T. Fujie, S. Takeoka, V. Mattoli, F. Greco, J. Mater. Chem. C 2015, 3, 6539. [25] M. S. White, M. Kaltenbrunner, E. D. Głowacki, K. Gutnichenko, G. Kettlgruber, I. Graz, S. Aazou, C. Ulbricht, D. A. M. Egbe, M. C. Miron, Z. Major, M. C. Scharber, T. Sekitani, T. Someya, S. Bauer, N. S. Sariciftci, Nat. Photonics 2013, 7, 811. [26] M. Kaltenbrunner, M. S. White, E. D. Głowacki, T. Sekitani, T. Someya, N. S. Sariciftci, S. Bauer, Nat. Commun. 2012, 3, 770. [27] T. Sekitani, U. Zschieschang, H. Klauk, T. Someya, Nat. Mater. 2010, 9, 1015. [28] P. Luan, N. Zhang, W. Zhou, Z. Niu, Q. Zhang, L. Cai, X. Zhang, F. Yang, Q. Fan, W. Zhou, Z. Xiao, X. Gu, H. Chen, K. Li, S. Xiao, Y. Wang, H. Liu, S. Xie, Adv. Funct. Mater. 2016, 26, 8178. [29] R. Kötz, R. Kötz, M. Carlen, M. Carlen, Electrochim. Acta 2000, 45, 2483. [30] L. B. Freund, S. Suresh, Thin Film Materials: Stress, Defect Formation and Surface Evolution, Cambridge University Press, 2003. [31] S. H. Baxamusa, M. Stadermann, C. Aracne-Ruddle, A. J. Nelson, M. Chea, S. Li, K. Youngblood, T. I. Suratwala, Langmuir 2014, 30, 5126. [32] P. K. Khare, R. S. Chandok, Polym. Int. 1995, 38, 153. [33] SPI Supplies, "Vinylec ® Resins data sheet," [34] W. Lee, D. Kim, N. Matsuhisa, M. Nagase, M. Sekino, G. G. Malliaras, T. Yokota, T. Someya, Proc. Natl. Acad. Sci. 2017, 114, 10554. [35] T. Yokota, Y. Inoue, Y. Terakawa, J. Reeder, M. Kaltenbrunner, T. Ware, K. Yang, K. Mabuchi, T. Murakawa, M. Sekino, W. Voit, T. Sekitani, T. Someya, Proc. Natl. Acad. Sci. 2015, 112, 14533. [36] A. M. Pappa, O. Parlak, G. Scheiblin, P. Mailley, A. Salleo, R. M. Owens, Trends Biotechnol. 2018, 36, 45. [37] M. M. Sprung, F. O. Guenther, M. T. Gladstone, Ind. Eng. Chem. 1955, 47, 305. [38] C. M. Stafford, C. Harrison, K. L. Beers, A. Karim, E. J. Amis, M. R. VanLandingham, H. C. Kim, W. Volksen, R. D. Miller, E. E. Simonyi, Nat. Mater. 2004, 3, 545. [39] S. Taccola, V. Pensabene, T. Fujie, S. Takeoka, N. M. Pugno, V. Mattoli, Biomed. Microdevices 2017, 19, 51. [40] F. Greco, A. Bellacicca, M. Gemmi, V. Cappello, V. Mattoli, P. Milani, ACS Appl. Mater. Interfaces 2015, 7, 7060. [41] I. D. Johnston, D. K. McCluskey, C. K. L. Tan, M. C. Tracey, J. Micromechanics Microengineering 2014, 24, 35017. [42] G. N. Greaves, A. L. Greer, R. S. Lakes, T. Rouxel, Nat. Mater. 2011, 10, 986. Pre-Print Manuscript [43] N. W. Tschoegl, W. G. Knauss, I. Emri, Mech. Time-Dependent Mater. 2002, 6, 3. [44] Q. Wang, Y. Yin, H. Xie, J. Liu, W. Yang, P. Chen, Q. Zhang, Soft Matter 2011, 7, 2888. [45] A. A. Tracton, Coatings Technology Handbook, 2006. [46] J. L. Keddie, R. A. L. Jones, R. A. Cory, Faraday Discuss. 1994, 98, 219. Pre-Print Manuscript Figure 1. a) Scheme of PVF nanosheet fabrication process and release in water. Left branch refers to bilayer Au-PVF ultrathin electrode (layers 1 and 3), while right branch refers to pristine PVF ultrathin dielectric (layer 2) fabrication. Scheme of assembly by consecutive recollection of free-standing floating nanosheets: b) ultrathin supported capacitor on glass substrate and c) free-standing capacitor membrane on a plastic ring frame. d) UT supported capacitor samples, ordered by increasing dielectric layer thickness t2 (structural colors due to thin film visible light interference). e) UT free-standing capacitor membrane on a plastic ring frame. Pre-Print Manuscript Figure 2. a) Capacitance at 1 kHz C of supported capacitors as a function of PVF dielectric thickness t2 (layer 2). Data fit (red line) was computed using flat capacitor model and returned a relative dielectric constant value of ε = 3.76. c) Cut-off frequency fc as a function of teff (t2 + tsup); fit curve (red line) highlights the linear dependence. e) Leakage current density J as a function of the applied electric field E for supported capacitors with different teff; device breakdown occurs in correspondence of sudden current increase in each device curve. Typical examples of impedance Z module and phase spectra of: b) supported capacitor with t2 = 10 nm; d) supported capacitor with t2 = 160 nm; f) free-standing capacitor with t2 = 160 nm (Insets: corresponding capacitance spectra computed applying series capacitor-resistor model (right), and optical image(left)). Pre-Print Manuscript Figure 3. Optical image (a) and SEM image (b) of a free-standing capacitor (PVF dielectric t2 = 90 nm) transferred onto a nylon mesh. The mesh was fixed to a plastic substrate for handling (Insets in (a): optical images the same capacitor at different illumination angles. c) Impedance module and phase spectra of the capacitor transferred onto the nylon mesh (Inset: corresponding capacitance spectrum computed applying series capacitor-resistor model, top-right corner). d) SEM image of a bare PVF nanosheet (t2 = 90 nm) conformally adhering onto the mesh pattern (inset: SEM image of the bare mesh). e) SEM image of a capacitor (t2 = 60 nm) recollected on a scalpel (Inset: optical image of the conformed free-standing capacitor on the scalpel, scale bar 4 mm). The image shows a trench obtained by Focused Ion Beam (FIB) milling of the sample surface in correspondence of the scalpel blade; the trench reveals the local curvature radius (white dashed semicircle) sustained by the device, of about 1.5 µm (yellow arrows). f) SEM image of a cross-section obtained by FIB milling on a capacitor supported on glass (t2 = 90 nm).
1903.08522	1	1903	2019-03-20T14:34:58	Limits of flexural wave absorption by open lossy resonators: reflection and transmission problems	[ "physics.app-ph" ]	The limits of flexural wave absorption by open lossy resonators are analytically and numerically reported in this work for both the reflection and transmission problems. An experimental validation for the reflection problem is presented. The reflection and transmission of flexural waves in 1D resonant thin beams are analyzed by means of the transfer matrix method. The hypotheses, on which the analytical model relies, are validated by experimental results. The open lossy resonator, consisting of a finite length beam thinner than the main beam, presents both energy leakage due to the aperture of the resonators to the main beam and inherent losses due to the viscoelastic damping. Wave absorption is found to be limited by the balance between the energy leakage and the inherent losses of the open lossy resonator. The perfect compensation of these two elements is known as the critical coupling condition and can be easily tuned by the geometry of the resonator. On the one hand, the scattering in the reflection problem is represented by the reflection coefficient. A single symmetry of the resonance is used to obtain the critical coupling condition. Therefore the perfect absorption can be obtained in this case. On the other hand, the transmission problem is represented by two eigenvalues of the scattering matrix, representing the symmetric and anti-symmetric parts of the full scattering problem. In the geometry analyzed in this work, only one kind of symmetry can be critically coupled, and therefore, the maximal absorption in the transmission problem is limited to 0.5. The results shown in this work pave the way to the design of resonators for efficient flexural wave absorption.	physics.app-ph	physics	Limits of flexural wave absorption by open lossy resonators: reflection and transmission problems J. Leng1, F. Gautier1, A. Pelat1, R. Pic´o2, J.-P. Groby1, V. Romero-Garc´ıa1 1Laboratoire d'Acoustique de l'Universit´e du Mans, LAUM - UMR 6613 CNRS, Le Mans Universit´e, Avenue Olivier Messiaen, 72085 LE MANS CEDEX 9, France 2Instituto para la Gesti´on Integral de zonas Costeras (IGIC), Universitat Polit`ecnica de Val`encia, Paranimf 1, 46730, Gandia, Spain E-mail: [email protected] Abstract. The limits of flexural wave absorption by open lossy resonators are analytically and numerically reported in this work for both the reflection and transmission problems. An experimental validation for the reflection problem is presented. The reflection and transmission of flexural waves in 1D resonant thin beams are analyzed by means of the transfer matrix method. The hypotheses, on which the analytical model relies, are validated by experimental results. The open lossy resonator, consisting of a finite length beam thinner than the main beam, presents both energy leakage due to the aperture of the resonators to the main beam and inherent losses due to the viscoelastic damping. Wave absorption is found to be limited by the balance between the energy leakage and the inherent losses of the open lossy resonator. The perfect compensation of these two elements is known as the critical coupling condition and can be easily tuned by the geometry of the resonator. On the one hand, the scattering in the reflection problem is represented by the reflection coefficient. A single symmetry of the resonance is used to obtain the critical coupling condition. Therefore the perfect absorption can be obtained in this case. On the other hand, the transmission problem is represented by two eigenvalues of the scattering matrix, representing the symmetric and anti-symmetric parts of the full scattering problem. In the geometry analyzed in this work, only one kind of symmetry can be critically coupled, and therefore, the maximal absorption in the transmission problem is limited to 0.5. The results shown in this work pave the way to the design of resonators for efficient flexural wave absorption. 9 1 0 2 r a M 0 2 ] h p - p p a . s c i s y h p [ 1 v 2 2 5 8 0 . 3 0 9 1 : v i X r a Limits of flexural wave absorption by open lossy resonators 2 1. Introduction Recent studies in audible acoustics have focused on wave absorption at low frequencies by means of subwavelength locally resonant materials. In particular, the design of broadband subwavelength perfect absorbers, whose dimensions are much smaller than the wavelength of the frequency to be attenuated, has recently been proposed [1 -- 5]. Such devices can totally absorb the energy of an incident wave and require solving the twofold but often contradictory problem: (i) increasing the density of states at low frequencies and (ii) matching the impedance with the background medium. On the one hand, the use of local resonators is a successful approach for increasing the density of states at low frequencies with reduced dimensions, as it has been shown in the field of metamaterials [6 -- 13]. On the other hand, the local resonators of such metamaterials are open and lossy ones, implying energy leakage and inherent losses. In these systems the impedance matching can be controlled by the ratio between the inherent losses of the resonator and the leakage of energy [14]. Particularly, the perfect compensation of the leakage by the losses is known as the critical coupling condition [15] and has been widely used to design perfect absorbers in different fields of physics [16, 17] other than acoustics. The critical coupling condition is also relevant for applications in vibrations owing to the increasing need for damping materials at low frequencies in several branches of industry [10]. Current passive solutions in this field are mainly based on the use of viscoelastic coatings [18]. Another solution yields in the tuned vibration absorber (TVA) [19 -- 21] that is used to control flexural waves in beams. The tuning of the resonance frequency of an undamped TVA has been analyzed [20], showing that complete suppression of the flexural wave transmission can be achieved. In most cases, TVA have been used to minimize the transmission of a propagating wave [20], resulting in practice in heavy treatments at low frequencies. Less attention has been paid to the case of maximizing the absorption in order to reduce simultaneously both the reflected and transmitted waves. The purpose of this work is to study the problem of perfect absorption of flexural waves in 1D elastic beams with local resonators by using the critical coupling condition. Particularly, the absorption of energy is analyzed through the balance between the energy leakage and the inherent losses in the resonators for the two scattering problems: the reflection and the transmission of flexural waves. The presented problem is related to the control of flexural waves in a beam using a passive TVA but with a physical insight that allows the interpretation of the limits of the flexural wave absorption based on both the critical coupling conditions and the symmetry of the excited resonances in the resonator. The analyzed systems are composed of a main beam and an open resonator simply consisting in a reduction of the thickness of the main beam. A thin viscoelastic coating is attached to it, leading to a composite material whose loss may be tuned. This composite material is modeled with the Ross-Kerwin-Ungar (RKU) method [22] and is embedded in the main beam. By tuning the losses, it is possible to analyze the different Limits of flexural wave absorption by open lossy resonators 3 Figure 1: Diagrams of the 1D configurations analyzed for the reflection and transmission problems for flexural waves. (b) Configuration for the transmission problem. (a) Configuration for the reflection problem. limits in both scattering problems. In practice, this type of resonator results in simpler geometries than that of the TVA which consists of complicated combinations of mass spring systems simulating a point translational impedance. The composite is studied by means of an analytical model based on the transfer matrix method. The analytical results, in accordance with the the experimental results, show the limits of the maximal values for the flexural wave absorption and their physical interpretations in both the reflection and transmission problems. The interpretations are based on the eigenvalues of the S-matrix for the propagating waves, represented in the complex frequency plane [1]. An experimental prototype is designed and measured for the reflection problem. The experimental results prove the perfect absorption of flexural waves and validate the analytical predictions. The work is organized as follows. In section 2, the theoretical model used to analyze the 1D scattering problems of flexural wave is presented. The physical analysis of the absorption coefficient in the complex frequency plane are presented in section 3. This analysis is based on an analytical model and the concept of critical coupling to obtain a perfect absorption of flexural waves. The experimental set-up used to validate the model for the reflection problem is then presented in section 4 as well as the experimental methodology and results. Finally, section 5 summarizes the main results and gives the concluding remarks. 2. Theoretical models This section describes the theoretical model used to study the absorption of flexural waves by open lossy resonators in 1D systems, following the approach of Mace [23]. The governing equations used in the model are first introduced. Two scattering problems are then presented. The first one is the reflection problem where the absorption by a resonator made of a thinner composite beam located at the termination of a semi-infinite beam is studied (figure 1a). The second one is the transmission problem where the absorption of the same resonator located between two semi-infinite beams is considered (figure 1b). The analytical results shown for the two problems have been tested by numerical simulations, but not shown in the article for clarity of the figures, later on the analytical results are validated experimentally in the section 4. (b)(a)Limits of flexural wave absorption by open lossy resonators 4 2.1. Flexural wave propagation in uniform beams Consider a thin uniform beam whose neutral axis is denoted by the x-axis. Assuming Euler-Bernoulli conditions, the flexural displacement w(x, t) satisfies [24]: D ∂4w ∂x4 + m ∂2w ∂t2 = 0, (1) where D = EI is the flexural rigidity, E the Young modulus, I the second moment of area and m the linear mass. Assuming time harmonic solution of the form eiωt, where ω is the angular frequency, the solution of Eq.(1) can be written in the frequency domain as the sum of four flexural waves: w(x) = a+e−ikx + a+ N e−kx + a−eikx + a− N ekx. (2) The complex amplitudes of the propagative and evanescent waves are a and aN respectively, and the signs + and − denote the outgoing and ingoing waves respectively. The evanescent component is a near field component, the amplitude of which decreases exponentially with distance. The flexural wavenumber k is given by k4 = , which is real and positive in the lossless case and complex when damping is accounted for. The wave amplitude is expressed in the vector form by convenience: mω2 D a+ = a− = , . (3) The relation between wave amplitudes along a beam with a constant thickness are then described by a+(x0 + x) = fa+(x0) and a−(x0 + x) = f−1a−(x0), where the diagonal transfer matrix f is given by (cid:34) (cid:35) a+ a+ N (cid:34) f = e−ikx 0 0 e−kx (cid:35) (cid:34) a− a− N (cid:35) . (4) (5) 2.2. Reflection coefficient in a pure reflection problem Consider an incident plane wave in the configuration described by the figure 1a, where the system is terminated by a free termination at one end. The displacement w at any point for x < 0 reads as w(x < 0) = a+ + a− = a+ + Rr · a+, (6) where Rr denotes the reflection matrix of the resonant termination of the beam at x = 0. The incident wave is transmitted into the resonant termination and reflected at its end, therefore the matrix Rr can be evaluated, using the displacement continuity at the interface and at the boundaries as [23]: Rr = a−a+−1 = r12 + t12 (cid:0)(f rf f )−1 − r21 (cid:1)−1 t21, (7) Limits of flexural wave absorption by open lossy resonators 5 where rij and tij represent the reflection and transmission matrices from section (i) to section (j) of the beam (see figure 1a). Considering continuity and equilibrium respectively at the section change, these matrices are given by (cid:34) 4 ∆ tij = (1 + β)(1 + γ) (−1 − iβ)(1 − γ) (cid:34)−2(β2 − 1)γ − 1iβ(1 − γ)2 (1 − i)β(1 − γ2) rij = 2 ∆ kj ki Djk2 j Dik2 i (cid:35) , (−1 + 1iβ)(1 − γ) (1 + β)(1 + γ) (1 + i)β(1 − γ2) −2(β2 − 1)γ + iβ(1 − γ)2 (cid:35) , (8) (9) and γ = where β = correspond to the ratios of wavenumbers and bending wave impedances, and ∆ = (1 + β)2(1 + γ)2 − (1 + β2)(1 − γ)2. The reflection matrix rf of the free termination is given by (cid:34) −ı rf = (1 − ı) (1 + ı) ı (cid:35) . (10) Rr is thus a 2 × 2 matrix where the diagonal components correspond to the reflection coefficients of the propagative and evanescent waves respectively. The study focuses on the reflection of waves in the far-field (x → −∞), i.e., on the propagative waves that carry the energy. The first term of the reflection matrix Rr(1, 1) ≡ Rr is therefore only considered since Rr(1, 2), Rr(2, 1), Rr(2, 2) → 0 when x → −∞. The absorption coefficient αR of propagating waves in the reflection problem can then be written as: αr = 1 − Rr2, x → −∞. (11) In the lossless case, i.e. without dissipation, Rr is simply equal to 1 for any purely real frequency as the energy conservation is fulfilled. 2.3. Reflection and transmission coefficients in a 1D symmetric and reciprocal transmission problem The transmission problem of the structure shown in figure 1b is described in this section, considering b− = 0. Due to the symmetry of the resonator and assuming propagation in the linear regime, the problem is considered as symmetric and reciprocal. The reflection and transmission matrices Rt and Tt at x = 0 and x = L are used to define the displacements on each side of the resonator such as: w(x < 0) = a+ + a− = a+ + Rt · a+, w(x > L) = b+ = Tt · a+. (12) (13) Using the displacement continuity at x = 0 and x = L in a similar way as in the previous section, Rt and Tt are written as (cid:0)(f r23f )−1 − r21 (cid:1)−1 t21, Rt = r12 + t12 (14) Limits of flexural wave absorption by open lossy resonators Tt = a−a+−1 = t23(I − (f r21f )r23)−1f t12. Therefore the absorption coefficient of the transmission problem is defined as αt = 1 − Tt2 − Rt2, x → ±∞ where Rt = Rt(1, 1) when x → −∞ and Tt = Tt(1, 1) when x → +∞. 6 (15) (16) 2.4. Viscoelastic losses in the resonator: the RKU model The inherent losses of the resonator are introduced by a thin absorbing layer of thickness hl as shown in figures 1a-1b and are considered frequency independent. The complex Young Modulus of the absorbing layer is El(1 + iηl), where ηl is its loss factor. Using the RKU model [22], this region is modeled as a single composite layer with a given effective bending stiffness Dc written as: Dc = E2I2 (1 + jη2) + ech3 c(1 + jηl) + 3 + (1 + hc)2echc[1 − η2ηl + j(η2 + ηl)] 1 + echc(1 + jηl) , (17) (cid:20) (cid:21) where the indices 2 and l stand for the parameters of the thin beam and of the absorbing layer respectively, ec = El/E2 and hc = hl/h2. The wave number kc of the composite material can then be written as k4 c = , where h = hl + h2 and ρch = ρ2h2 + ρlhl. ρchω2 Dc 3. Limits of absorption for the reflection and transmission problems This section describes the limits of absorption for flexural waves in the reflection and transmission problem by using open, lossy and symmetric resonators. It provides tools to design absorbers with a maximal absorption in both problems. For this purpose, the eigenvalues of the scattering matrix of the propagative waves are represented in the complex frequency plane as in Ref. [1]. The material and geometric parameters used in the following sections are described in table 1. 3.1. Properties of the S-matrix Consider a two-port, 1D, symmetric and reciprocal scattering process for the systems described in figure 1b in the case where x → ±∞. The relation between the amplitudes a+ and b− of the incoming waves and a− and b+ of the outgoing waves when x → ±∞ is given by (cid:32) (cid:33) a− b+ (cid:34) (cid:32) (cid:33) b− a+ = S(ω) = (cid:35)(cid:32) (cid:33) b− a+ Tt Rt Rt Tt x → ±∞, , (18) where S is the scattering matrix or the S-matrix of the propagative waves. The complex eigenvalues of the S-matrix are ψ1,2 = Tt ± Rt. An eigenvalue of the S-matrix equal to zero implies that the incident wave is completely absorbed (a− = b+ = 0). This happens when Tt = ±Rt and the incident waves [a+ , b−] correspond to the relevant eigenvector. Limits of flexural wave absorption by open lossy resonators 7 Table 1: Geometric and material parameters of the studied systems. The value of ηl depends on the experimental set-up used, see main text for the used values retrieved from experiments. Geometric parameters Material parameters Main beam h1 = 5 mm b = 2 cm Resonator beam h2 = 0.217 mm Coating layer b2 = 2 cm L = 1.6 cm hl = 1.5 mm bl = 2 cm Ll = 1.6 cm ρ = 2811 kg.m−3 E = 71.4 GPa η = 0 ν = 0.3 ρ2 = 2811 kg.m−3 E2 = 71.4 GPa η2 = 0 ν2 = 0.3 El = 6.86 × 10−3 GPa ρl = 93.3 kg.m−3 ηl νl = 0.3 When the eigenvalues are evaluated in the complex frequency plane [1], poles and zeros can be identified. The pole frequencies correspond to the resonances of the resonator (zeros of the denominator of the eigenvalues) while the zero frequencies (zeros of the numerator of the eigenvalues) correspond to the perfect absorption configuration. In the case of a reflection problem, the eigenvalues are reduced to the reflection coefficient. Since the systems analyzed in this work are invariant under time-reversal symmetry, the scattering matrix, as defined in Eq.(18), presents unitary property [25] in the lossless case (i.e., without dissipative losses): S∗S = I. (19) The complex frequencies of the eigenvalue poles of the propagative S-matrix are complex conjugates of its zeros. Poles and zeros appear therefore symmetric with respect to the real frequency axis in the lossless case. 3.2. Reflection problem 3.2.1. Lossless case. In the reflection problem, where no wave is transmitted, the reflection coefficient Rr represents the scattering of the system. Thus, Rr corresponds directly to both the S-matrix and its associated eigenvalue (ψ = Rr). Its zeros correspond to the cases in which the incident wave is totally absorbed. In the lossless case, Rr = 1 for any purely real frequency and the pole-zero pairs appear at complex conjugate frequencies. Figure 2a depicts log10(Rr) in the complex frequency plane. The main beam, the resonator beam and the coating layer have the geometric and material parameters given in table 1. Note that the Young moduli are purely real in Limits of flexural wave absorption by open lossy resonators 8 Figure 2: Analysis of the scattering in the reflection problem. (a) Representation of (b)-(d) log10(Rr) log10(Rr) in the complex frequency plane for the lossless case. in the complex frequency plane in the lossy case for configurations with ηl = 0.02, 0.15 and 0.4 respectively. The case when the critical coupling condition is fulfilled (ηl = 0.15) is represented in (c). (e) Trade-off of the absorption at the first resonance frequency of the resonator as the inherent loss ηl is increased in the system. The points along the absorption curve represent the values of the absorption for the configurations represented in figures (b)-(d). Red continuous (Black dashed) line represents the absorption (reflection) coefficient as a function of etal at 673 Hz, corresponding to the first resonance frequency of the termination. the lossless case (η = η2 = ηl = 0). As shown in section 3.1, the poles and zeros appear in pairs and are symmetric with respect to the real frequency axis. Moreover the value of Rr along the real frequency axis is equal to 1. It is also worth noting that the imaginary part of the pole in the lossless case represents the amount of energy leakage by the resonator through the main beam [1]. With the time dependence convention used in this work, the wave amplitude at the resonance frequency decays as e−Im(ωpole)t. Thus the decay time τleak can be related with the quality factor due to the leakage as Qleak = , where the leakage rate can be defined Re(ωpole)τleak 2 = Re(ωpole) 2Im(ωpole) as Γleak = 1/τleak = Im(ωpole). The imaginary part of the poles and zeros increases when the real part of the frequency increases, meaning that more energy leaks out through the resonator at high frequencies. 3.2.2. Lossy case. For the sake of clarity, this section only focuses on the first pole-zero pair of the system previously described. The discussion can nevertheless be extended to any pole-zero pair of the system in the complex frequency plane. Losses are now Limits of flexural wave absorption by open lossy resonators 9 introduced into the system by adding an imaginary part to the Young modulus of the damping material such that it can be written as El(1 + iηl). As a consequence, the symmetry between the poles and zeros with respect to the real frequency axis is broken, since the property of Eq.(19) is no more satisfied in the lossy case. Figures 2b-2d depict log10(Rr) in the complex frequency plane around the first resonance frequency for three different increasing values of ηl. Figure 2b represents the case for which the losses are small (ηl = 0.02). In this case, the pole-zero pair is quasi-symmetric with respect to the real frequency axis. As the losses in the damping layer increase (ηl = 0.15 in figure 2c and ηl = 0.4 in figure 2d), the zero moves to the real frequency axis. In particular, the zero of the reflection coefficient is exactly located on the real frequency axis in figure 2c. In this situation, the amount of inherent losses in the resonator equals the amount of energy leakage. This situation is known as the critical coupling condition [15] and implies the impedance matching, leading to a perfect absorption. The value of the absorption coefficient of the first resonant peak as a function of ηl is depicted in figure 2e. The position of the zero in the complex frequency plane is directly related to the value of the flexural wave absorption. When the zero approaches the real frequency axis, the value of the absorption is close to one, being equal to 1 when the zero is exactly located in the real frequency axis. It should be noted that the perfect absorption cannot occur once the zero has crossed the real frequency axis. This property might appear counterintuitive since it means that adding a large amount of losses in the system might lead to a deterioration of the absorbing properties of the structure. 3.2.3. Design of perfect absorbers for flexural wave in the reflection problem. A theoretical design for the perfect absorption of flexural waves is shown in this section based on the configuration represented in the figure 1b and the parameters given in table 1. Considering that there is no inherent losses in the main beam and the resonator beam (η = η2 = 0), the loss factor of the coating layer has to be ηl = 0.15 to obtain a perfect absorption at the first resonance frequency of the system. Figures 3a-3b depict log10(Rr) for the lossless and lossy configurations in the complex frequency plane respectively. Figure 3b shows particularly the first pole-zero pair of the system in the perfect absorption configuration where the critical coupling condition is fulfilled, showing the zero exactly located on the real frequency axis. Figure 3c shows the corresponding absorption (red continuous line) and reflection (black dashed line) coefficients according to real frequencies for the critical coupled configuration. These coefficients are calculated with the analytical model described in previous sections. The incident wave is totally absorbed at the first resonance frequency of the composite beam. Limits of flexural wave absorption by open lossy resonators 10 Figure 3: Representation of the perfect absorption for the reflection problem. (a), (b) show the representation of the log10(Rr) for the lossless and lossy configurations respectively. (c) Red continuous and black dashed lines show the analytical absorption and reflection coefficients for the critical coupled configuration respectively. 3.3. Transmission problem For the transmission problem, the S-matrix is defined in Eq. (18) and has two eigenvalues ψ1,2. The scatterer being mirror symmetric, the problem can be reduced to two uncoupled sub-problems: a symmetric problem where ψs = Tt + Rt and an anti- symmetric, where ψa = Tt − Rt. ψs corresponds to the reflection coefficient of the symmetric problem while ψa corresponds to the reflection coefficient of the anti-symmetric problem. The absorption coefficient can also be expressed as α = (αs + αa)/2 where αs = 1− ψs 2 and αa = 1− ψa 2. Similarly to the reflection problem, poles and zeros of ψs and ψa can be identified in the complex frequency plane. The following sections focuses on the first resonant mode of the beam resonator, the displacement distribution of which is symmetric. The interpretation of the results can nevertheless be applied to the higher order modes with anti-symmetric distributions of the displacement field. It is worth noting that the displacement distribution of the resonant modes changes from symmetric to anti-symmetric as the mode increases due to the geometry of the resonators [26]. 3.3.1. Lossless case. Figures 4a and 4b show the variation of log10(ψs) and log10(ψa) evaluated respectively in the complex frequency plane in the lossless case for the first resonant mode. The main beam, the resonator beam and the coating layer of the studied system have still the material and geometric parameters of table 1, where η = η2 = etal = 0 in the lossless case. The symmetric and anti-symmetric problems Limits of flexural wave absorption by open lossy resonators 11 Figure 4: Representation of the eigenvalues of the S-matrix for a transmission problem in the lossless and lossy case. (a) and (b) Lossless case for log10(ψs) and log10(ψa) in the complex frequency plane. (c) and (d) Lossy case for log10(ψs) and log10(ψa) in the complex frequency plane. (e) Trade-off of the transmission (blue dotted line), reflection (black dashed line) and absorption (red continuous line) for the maximum absorption of the first mode as the loss factor of the coating layer increases. (f) Red continuous, black dashed and blue dotted lines represent the absorption, reflection and transmission coefficients respectively for the half critically coupled configuration. exhibit pole-zero pairs similarly to the reflection problem in the lossless case. These pairs are also symmetrically positioned with respect to the real frequency axis. The absence of dissipation is shown along the real frequency axis where Tt ± Rt = 1 for any real frequency. This section focuses only on the first resonant mode which has a symmetric distribution of the displacement field. Therefore, only the symmetric problem presents a pole-zero pair at the corresponding resonance frequency, while the anti-symmetric one does not. 3.3.2. Lossy case. Unlike the reflection problem, the condition for perfect absorption is stronger in the transmission one and needs to place the zeros of both ψs and ψa at the same frequency in the real frequency axis. Once this condition is fulfilled, a+ and b− correspond to the relevant eigenvector and the system satisfies the Coherent Perfect Absorption (CPA) condition [2, 9, 27]. Losses are introduced in the system in the same way as for the reflection problem, i.e., by increasing the loss factor ηl of the material of the damping layer. Once the losses are introduced, the position of the pole-zero pair of the symmetric eigenvalue in the complex frequency plane shifts towards the upper half space while the anti- symmetric problem remains unchanged without pole-zero pairs, as shown in Figures 4c Limits of flexural wave absorption by open lossy resonators 12 and 4d. Therefore, only the zero of the symmetric problem can be placed on the real frequency axis, i.e., only half of the problem can be critically coupled. Figure 4e shows the dependence of the reflection, transmission and absorption coefficient on the inherent losses in the resonator for the first mode. The maximum absorption obtained is 0.5 since only the symmetric problem is critically coupled (α = (αs + αa)/2 (cid:39) (1 + 0)/2 = 1/2). 3.3.3. Design of maximal absorbers for flexural wave in the transmission problem. Based on the results discussed previously, a configuration with maximal absorption for flexural waves in the transmission problem is designed with the parameters given in table 1. As for the reflection problem, no inherent losses are considered in the main beam and the resonator beam (η = η2 = 0). The loss factor of the coating layer is ηl = 2.21. The reflection, transmission and absorption for this configuration is analyzed in figure 4(f), showing that the maximum absorption is 0.5 at the resonance frequency of the beam. This result is in accordance with the ones previously obtained [2, 9, 21], even if the resonator is not a point translational impedance. The absorption is limited to 0.5 since only one kind of geometry of resonant mode can be excited. The problem is therefore half critically coupled. To obtain a higher absorption, other strategies based on breaking the symmetry of the resonator [5] or on the use of degenerate resonators are needed [28]. In these cases, both eigenvalues present poles and zeros located at the same real frequencies. It would then be possible to fully critically couple the problem and so obtain a perfect absorption (i.e. α = 1) at the appropriate frequency. 4. Experimental results This section presents the experimental results of the reflection coefficient [29, 30] for an aluminum beam system with the configuration described in section 3.2.3. 4.1. Experimental set-up The beam is held vertically in order to avoid static deformation due to gravity. The extremity at which the reflection coefficient is estimated is oriented towards the ground (see figure 5a). The used coating layer have been experimentally characterized showing an ηl = 0.15, which is the value for which perfect absorption can be observed. A photograph of the resonator with the coating layer is shown in figure 5b. The measurements are performed along the beam at 21 points equidistant of 5 mm and located on its neutral axis in order to avoid the torsional component. The measurement points are also located sufficiently far from the source and the extremity of the beam to consider far-field assumption and neglect the contribution of evanescent waves. In this case, far-field assumption is fulfilled at a distance lf from both the source and the resonator for which the evanescent wave loses 90 % of its initial magnitude. The low frequency limit of the measurements is then estimated using eklf = 0.1. The shaker excites the beam with a sweep sine. The displacement field versus frequency is obtained Limits of flexural wave absorption by open lossy resonators 13 Figure 5: (a) Diagram of the experimental set-up. (b) Photograph of the resonator. (c) Black crosses and red open circles show respectively Rr2 and αr for the critical coupled configuration measured with the experimental set-up. Black dashed and red continuous lines show Rr2 and αr calculated with the analytical model. from the measurements of the vibrometer at each measure point. 4.2. Experimental estimation of the reflection coefficient Consider the flexural displacement W (xi, ω) measured at the point xi(i ∈ [0, 20]) for a given angular frequency ω as W (xi, ω) = A(ω)e−ikxi + B(ω)eikxi. (20) The set of W (xi, ω) for each measurement point can be written in a matrix format [30] such as  W (x0, ω) W (x1, ω) W (x2, ω) ... W (x20, ω)  =  e(−ikx0) e(−ikx1) e(−ikx2) ... e(ikx0) e(ikx1) e(ikx2) ... e(−ikx20) e(ikx20) (cid:32)  (cid:33) A(ω) B(ω) , (21) Limits of flexural wave absorption by open lossy resonators 14 The amplitudes A(ω) and B(ω) can then be derived from Eq. (21) which forms an overdetermined system. From these amplitudes, the reflection coefficient of the propagative waves can be deduced for any ω as: Rr(ω) = A(ω) B(ω) . (22) 4.3. Experimental evidence of perfect absorption for flexural waves Experimental results obtained with the experimental set-up are depicted in figure 5c. A drop of reflection is noticed at the first resonant frequency of the termination with a minimum value of Rr2 = 0.02 at 667 Hz for the experiment and Rr2 = 0 at 673 Hz for the analytical result. The gap between the analytical and experimental resonant frequency is 0.9%. This frequency shift between the model and the experiment is mainly due to the geometric uncertainty in the resonator thickness, induced by the machining process. This geometrical uncertainty induces also an estimation uncertainty of the energy leakage of the resonator. The absorption is then experimentally limited to αr = 0.98. Evidence of perfect absorption for flexural waves by means of critical coupling is shown experimentally here. Three experimental scans of the whole beam at 500 Hz, 670 Hz and 800 Hz have been measured ‡. At 500 Hz or 800 Hz the reflection coefficient is very close to one. The standing waves in the main beam are visible at these frequencies. At 670 Hz, the termination absorbs totally the incident waves. There is therefore no standing waves and the waves are propagating in the main beam. 5. Conclusions Absorption of propagative flexural waves by means of simple beam structures is analyzed in this work. The main mechanisms are interpreted in terms of both the critical coupling conditions and the symmetries of the resonances for both the reflection and the transmission problems. The positions of the zeros of the eigenvalues of the scattering matrix in the complex frequency plane give informations on the possibility to obtain the perfect absorption. The perfect absorption condition is fulfilled when these zeros are placed on the real frequency axis, meaning that the inherent losses are completely compensating the energy leakage of the system. In the reflection problem, the physical conditions of the problem lead to perfect absorption at low frequencies. In this case a single symmetry for the resonance is excited and perfect absorption can be obtained when the inherent losses of the system balance the energy leakage of the system. In the transmission problem, the requirement to obtain perfect absorption is stronger than for the case in reflection as two kinds of symmetries of the resonances are required to be critically coupled simultaneously. In the case presented in this work, or in the general case of point translational impedances, dealing only with one type of symmetry for the ‡ See supplementary material: videos 500Hz.avi, 670Hz.avi and 800Hz.avi Limits of flexural wave absorption by open lossy resonators 15 resonant modes [21] limits the absorption to 0.5. Therefore for the perfect absorption in the transmission case, two strategies are needed: (i) breaking the symmetry of the resonator in order to treat the full problem with a single type of symmetry of the resonance mode [5]; (ii) using degenerate resonators with two types of symmetries at the same frequency being critically coupled [28]. The resonator used in this study has been chosen as an integral part of the main beam for experimental set-up reasons. However, the presented approach can be applied to any class of 1D resonant-system provided that the resonators are local, open and lossy ones. These properties of the resonator are the essential points to achieve the perfect absorption at low frequency by solving the following problems: increasing the density of states at low frequencies and matching the impedance with the background medium. Acknowledgments The authors thank Mathieu S´ecail-Geraud and Julien Nicolas for the development of the experimental set-up. The work has been founded by the RFI Le Mans Acoustic (R´egion Pays de la Loire) within the framework of the Metaplaque project. This article is based upon work from COST action DENORMS CA 15125 , supported by COST ( European Cooperation in Science and Technology). The work was partly supported by the Spanish Ministry of Economy and Innovation (MINECO) and European Union FEDER through project FIS2015-65998-C2-2 and by project AICO/2016/060 by Conseller´ıa de Educaci´on, Investigaci´on, Cultura y Deporte de la Generalitat Valenciana. References [1] V. Romero-Garc´ıa, G. Theocharis, O. Richoux, and V. Pagneux. Use of complex frequency plane to design broadband and sub-wavelength absorbers. J. Acoust. Soc. Am., 139(6):3395 -- 3403, 2016. [2] A. Merkel, G. Theocharis, O. Richoux, V. Romero-Garc´ıa, and V. Pagneux. Control of acoustic absorption in one-dimensional scattering by resonant scatterers. App. Phys. Lett., 107(24):244102, 2015. [3] J.-P. Groby, R. Pommier, and Y. Aur´egan. Use of slow sound to design perfect and broadband passive sound absorbing materials. J. Acoust. Soc. Am., 139(4):1660 -- 1671, 2016. [4] N. Jim´enez, W. Huang, V. Romero-Garc´ıa, V. Pagneux, and J.-P. Groby. Ultra-thin metamaterial for perfect and quasi-omnidirectional sound absorption. Appl. Phys. Lett., 109(12):121902, 2016. [5] N. Jim´enez, V. Romero-Garc´ıa, V. Pagneux, and J.-P. Groby. Rainbow-trapping absorbers: Broadband, perfect and asymmetric sound absorption by subwavelength panels for transmission problems. Sci. Rep., 7(1):13595, 2017. [6] N. Fang, D. Xi, J. Xu, M. Ambati, W. Srituravanich, C. Sun, and X. Zhang. Ultrasonic metamaterials with negative modulus. Nat. Mater., 5(6):452, 2006. [7] Z. Liu, X. Zhang, Y. Mao, Y.Y. Zhu, Z. Yang, C.T. Chan, and P. Sheng. Locally resonant sonic materials. Science, 289(5485):1734 -- 1736, 2000. [8] E.A. Skelton, R.V. Craster, A. Colombi, and D.J. Colquitt. The multi-physics metawedge: graded arrays on fluid-loaded elastic plates and the mechanical analogues of rainbow trapping and mode conversion. New J. Phys., 20(053017), 2018. Limits of flexural wave absorption by open lossy resonators 16 [9] P. Wei, C. Croenne, S. Tak Chu, and J. Li. Symmetrical and anti-symmetrical coherent perfect absorption for acoustic waves. Appl. Phys. Lett., 104(12):121902, 2014. [10] Y. Duan, J. Luo, G. Wang, Z.H. Hang, B. Hou, J. Li, P. Sheng, and Y. Lai. Theoretical requirements for broadband perfect absorption of acoustic waves by ultra-thin elastic meta- films. Sci. Rep., 5(12139), 2015. [11] G. Theocharis, O. Richoux, V. Romero-Garc´ıa, A. Merkel, and V. Tournat. Limits of slow sound propagation and transparency in lossy, locally resonant periodic structures. New J. Phys., 16(9):093017, 2014. [12] A. Colombi, V. Ageeva, R.J. Smith, A. Clare, R. Patel, M. Clark, D. Colquitt, P. Roux, S. Guenneau, and R.V. Craster. Enhanced sensing and conversion of ultrasonic rayleigh waves by elastic metasurfaces. Sci. Rep., 7(6750), 2017. [13] R.J. Pal and M. Ruzzene. Edge waves in plates with resonators: an elastic analogue of quantum valley hall effect. New J. Phys., 19(025001), 2017. [14] K.Y. Bliokh, Y.P. Bliokh, V. Freilikher, S. Savel'ev, and F. Nori. Colloquium: Unusual resonators: Plasmonics, metamaterials, and random media. Rev. Mod. Phys., 80(4):1201, 2008. [15] A. Yariv. Universal relations for coupling of optical power between microresonators and dielectric waveguides. Electron. Lett., 36(4):321 -- 322, 2000. [16] Y. Xu, Y. Li, R.K. Lee, and A. Yariv. Scattering-theory analysis of waveguide-resonator coupling. Phys. Rev. E, 62(5):7389, 2000. [17] Ming Cai, Oskar Painter, and Kerry J Vahala. Observation of critical coupling in a fiber taper to a silica-microsphere whispering-gallery mode system. Phys. Rev. Lett., 85(1):74, 2000. [18] T.-L. Teng and N.-K. Hu. Analysis of damping characteristics for viscoelastic laminated beams. Comput. Method Appl. M., 190(29-30):3881 -- 3892, 2001. [19] J.P. Den Hartog. Mechanical vibrations. Courier Corporation, 1985. [20] M.J. Brennan. Control of flexural waves on a beam using a tunable vibration neutraliser. Jour. Sound Vib., 222(3):389 -- 407, 1999. [21] H.M. El-Khatib, B.R. Mace, and M.J. Brennan. Suppression of bending waves in a beam using a tuned vibration absorber. J. Sound Vib., 288(4):1157 -- 1175, 2005. [22] D. Ross, E.L. Ungar, and E.M. Kerwin. Damping of plate flexural vibrations by means of viscoelastic laminae. Structural damping, pages 49 -- 87, 1960. [23] B.R. Mace. Wave reflection and transmission in beams. J. Sound Vib., 97(2):237 -- 246, 1984. [24] A. W. Leissa. Tabulated numerical results of theories of plate vibration. NASA, Washington, DC, United States., 1969. [25] A. N. Norris. Reflection and transmission of structural waves at an interface between doubly curved shells. Acta Acust. United Ac., 84(6):1066 -- 1076, 1998. [26] N. Jim´enez, V. Romero-Garc´ıa, V. Pagneux, and J.-P. Groby. Quasiperfect absorption by subwavelength acoustic panels in transmission using accumulation of resonances due to slow sound. Phys. Rev. B, 95:014205, 2017. [27] Y.D. Chong, L. Ge, H. Cao, and A.D. Stone. Coherent perfect absorbers: time-reversed lasers. Phys. Rev. Lett., 105(5):053901, 2010. [28] M. Yang, C. Meng, C. Fu, Y. Li, Z. Yang, and P. Sheng. Subwavelength total acoustic absorption with degenerate resonators. Appl. Phys. Lett., 107(10):104104, 2015. [29] F. Gautier, M.-H. Moulet, and J.-C. Pascal. Reflection, transmission and coupling of longitudinal and flexural waves at beam junctions. part i: measurement methods. Acta Acust. United Ac., 92(6):982 -- 997, 2006. [30] V. Denis, F. Gautier, A. Pelat, and J. Poittevin. Measurement and modelling of the reflection coefficient of an acoustic black hole termination. J. Sound Vib., 349:67 -- 79, 2015.
1812.02287	4	1812	2019-07-30T04:37:24	Inverse-Designed Diamond Photonics	[ "physics.app-ph", "cond-mat.mes-hall", "physics.optics" ]	Diamond hosts optically active color centers with great promise in quantum computation, networking, and sensing. Realization of such applications is contingent upon the integration of color centers into photonic circuits. However, current diamond quantum optics experiments are restricted to single devices and few quantum emitters because fabrication constraints limit device functionalities, thus precluding color center integrated photonic circuits. In this work, we utilize inverse design methods to overcome constraints of cutting-edge diamond nanofabrication methods and fabricate compact and robust diamond devices with unique specifications. Our design method leverages advanced optimization techniques to search the full parameter space for fabricable device designs. We experimentally demonstrate inverse-designed photonic free-space interfaces as well as their scalable integration with two vastly different devices: classical photonic crystal cavities and inverse-designed waveguide-splitters. The multi-device integration capability and performance of our inverse-designed diamond platform represents a critical advancement toward integrated diamond quantum optical circuits.	physics.app-ph	physics	Inverse-Designed Diamond Photonics Constantin Dory,1,† Dries Vercruysse,1,∗Ki Youl Yang,1,∗ Neil V. Sapra,1 Alison E. Rugar,1 Shuo Sun,1 Daniil M. Lukin,1 Alexander Y. Piggott,1 Jingyuan L. Zhang,1 Marina Radulaski,1,2 Konstantinos G. Lagoudakis,1,3 Logan Su1 and Jelena Vuckovi´c1,† 1E. L. Ginzton Laboratory, Stanford University, Stanford, California 94305, USA. 2Also at: Electrical and Computer Engineering, University of California, Davis, CA 95616, USA 3Now at: Department of Physics, University of Strathclyde, Glasgow, G4 0NG, UK * These authors contributed equally to this work. †Corresponding authors: [email protected], [email protected] Diamond hosts optically active color centers with great promise in quantum compu- tation, networking, and sensing. Realization of such applications is contingent upon the integration of color centers into photonic circuits. However, current diamond quantum optics experiments are restricted to single devices and few quantum emit- ters because fabrication constraints limit device functionalities, thus precluding color center integrated photonic circuits. In this work, we utilize inverse design methods to overcome constraints of cutting-edge diamond nanofabrication methods and fab- ricate compact and robust diamond devices with unique specifications. Our design method leverages advanced optimization techniques to search the full parameter space for fabricable device designs. We experimentally demonstrate inverse-designed pho- tonic free-space interfaces as well as their scalable integration with two vastly different devices: classical photonic crystal cavities and inverse-designed waveguide-splitters. The multi-device integration capability and performance of our inverse-designed dia- mond platform represents a critical advancement toward integrated diamond quantum optical circuits. Diamond has excellent material properties for quan- tum optics,1,2 optomechanics,3,4 and nonlinear optics.5 Of particular interest is the variety of color centers that diamond hosts, some of which exhibit very long coherence times.1,6 The development of diamond photonic circuits7 has emerged as a promising route for implementing opti- cal quantum networks,8 -- 18 quantum computers,19 -- 21 and quantum sensors.22,23 However, a major challenge in dia- mond quantum photonics is the lack of high-quality thin films of diamond, as the production of electronic grade diamond can be achieved only in homoepitaxy, and thin- ning processes are not repeatable enough for photonic crystal cavity fabrication.24 As a result, state-of-the- art diamond cavity quantum photonics relies on angled- etching of bulk diamond.24 This technique naturally leads to triangular cross-sections with strongly constrained ge- ometries, which limit device design and functionality. Re- cent developments in diamond processing based on quasi- isotropic etching25 -- 28 (see Supplementary Note 1, Figure 1 and 2) allow the production of diamond membranes with rectangular cross-sections and variable dimensions from bulk diamond. Although rectangular cross-sections are a major step toward diamond integrated circuits, this fabrication technique comes with its own geometric con- straints, such as limitations on the range of fabricable feature sizes, which originate from a strong correlation of the initial etch depth and undercut thin-film area. Tra- ditional photonic designs that do not account for fabri- cation constraints are thus unable to take full advantage of this new fabrication technique. In this work, we overcome these fabrication and de- sign challenges by employing inverse design methods. In silicon nanophotonics these methods have recently at- tracted considerable attention for their efficient design of devices with superior performance over conventional designs.29 This optimization technique searches through the full parameter space of fabricable devices, thereby ar- riving at solutions previously inaccessible to traditional design techniques.30 We showcase the potential of in- verse design techniques for diamond integrated circuits by designing and fabricating several devices: A compact vertical coupler, an essential component for large-scale quantum photonic systems, and a small circuit consist- ing of inverse-designed vertical couplers and waveguide- splitters acting as interfaces for two nanoresonators. Our inverse-designed vertical coupler adheres to the diamond fabrication constraints and outperforms commonly used free-space interfaces. The fabricated devices show excel- lent agreement with simulations in terms of both perfor- mance and yield. In the second example, we illustrate the integration of such a vertical coupler into a diamond pho- tonic circuit consisting of two nanobeam resonators con- nected via inverse-designed waveguide-splitters -- a con- figuration that could be used to entangle two quantum emitters embedded inside such resonators. Inverse design of diamond nanophotonic devices In photonics, grating couplers are frequently used as op- tical free-space interfaces.31 -- 36 To achieve high coupling efficiencies, such designs typically use asymmetry along through partial etches31,32,36 or mate- the z-axis, e.g. rial stacks with varying refractive indices.32 In diamond quantum photonics many of these approaches cannot be ) % ( y c n e c i f f i E g n i l 25 20 15 10 2 900 a b c Grating Coupler Notch 0.5 m 0.5 m p u o C 5 0 650 700 750 800 850 Wavelength (nm) d z x e ) % ( y c n e c i f f i E g n i l f 30 25 20 15 10 y x 1 m p u o C 5 0 0 100 200 Iterations 300 400 1 m FIG. 1: Inverse design of efficient nanophotonic interfaces. Scanning electron micrograph of (a) a notch and (b) an inverse-designed vertical coupler with simulated fields superimposed in red. (c) Simulated performance of the vertical coupler (red) and the notch (green). (d) Design set-up, where the gray area indicates the design area. (e) In-coupling efficiency during design optimization; insets illustrate different optimization phases. The small performance drop beyond 200 iterations of optimization occurs when fabrication constraints are imposed. (f ) Final device design after optimization. employed because current thin-film diamond on silica substrate platforms33 -- 35 do not support state-of-the-art quantum optics experiments.20,21,24 Similar approaches with hybrid structures, such as gallium phosphide (GaP) membranes on diamond, offer a platform for efficient grating couplers.37 However, the optical field is confined in the GaP membrane and consequently emitters in dia- mond couple only evanescently to the field. A practical solution to these fabrication and design challenges are notches (Fig. 1a), which are a perturba- tion to a waveguide with ≈ 1 % scattering efficiency.20 In our work, we develop an inverse-designed vertical cou- pler (Fig. 1b) and use the notch for a baseline compar- ison. The couplers have a footprint of 1.0 × 1.0 μm2 and couple directly to a 400 nm wide waveguide with- out a tapering section, assuring compactness. As shown in Fig. 1c, the simulated peak efficiencies of the coupler (red) and the notch (green) are ≈ 25 % and ≈ 1 %,20 respectively. Furthermore, we optimize the vertical cou- pler to couple the light between the fundamental free- space mode TEM00 and the TE fundamental mode of the waveguide. Even for conventional grating couplers in mature photonics platforms the selective coupling to the TEM00 mode is a formidable challenge.31,32 The theoret- ical maximum coupling efficiency of our couplers is 50 %, because of the symmetry along the z-axis of our devices (i.e. the structure will couple light in +/- z direction equally). Inverse design problems in photonics are defined by an electromagnetic simulation, a design region and a figure of merit to optimize. The starting conditions of the simu- lation for vertical couplers are shown in Fig. 1d. A verti- cally incident Gaussian beam forms the radiative source and is centered above the 1.0 × 1.0 μm2 design region shown in gray. To the left of the design region are two black support bars to suspend the design, and to the right is a black output waveguide. The fraction of in- cident light coupled into the fundamental TE mode of the waveguide serves as the figure of merit and is maxi- mized during our optimization process (detailed in refer- ence 38). The coupling efficiency during the optimization is shown in Fig. 1e. At the start of the optimization, any permittivity value between that of air and diamond is allowed, which results in a continuous structure shown in the leftmost inset. After several iterations, this struc- ture is discretized, in which case the permittivity is that 3 SD . ) . u a ( s t n u o C Wavelength (nm) FIG. 2: Inverse-designed vertical couplers. (a) Scanning electron micrograph of two vertical couplers connected by a waveguide, which is used to characterize the efficiency of the vertical couplers. (b) Sideview of the vertical coupler, showing the undercut of the structure at an 85◦ angle. (c) Optical microscope image when focusing a Gaussian beam on the coupler on the left (Input) and detecting the transmitted light from the coupler on the right (Output). (d) Polarization scan shows that the vertical couplers preferentially couple to a Gaussian beam with a polarization perpendicular to the nanobeam. Simulated polarization dependence, shown as a blue line, are in good agreement with the measured data (red squares). (e) Efficiency of a single vertical coupler, peak efficiencies are ≈ 21 % for single-mode polarization-maintaining fibers (PMF, black) and ≈ 26.5 % for multimode fibers (MMF, red). Numerical simulation results are shown as a blue line. (f ) Transmission spectra of 15 different devices using a supercontinuum source. The solid black line corresponds to the mean value and the red shaded area corresponds to the standard deviation. of either air or diamond. This discrete structure is fur- ther optimized while also gradually imposing a penalty on infabricable features.39,40 As a result, the coupling ef- ficiency at a wavelength of 737 nm (silicon-vacancy color center zero-phonon line) peaks at a value of ≈ 27.5 %, which then decreases to ≈ 25 % to comply with fabrica- tion constraints.40 Characterization of diamond vertical couplers To characterize the coupling efficiency of the vertical cou- plers, we measure the device shown in Figs. 2a and b, in top-down and side view, respectively. An optical micro- scope image of the same structure, presented in Fig. 2c, qualitatively shows the high performance of the vertical couplers. We characterize the polarization dependence of the vertical couplers by sweeping the polarization of the input laser beam (Fig. 2d). The observed five-fold re- duction in the transmitted power when rotating the po- larization by π 2 corresponds well to our simulated results (blue line in Fig. 2d) and is experimental evidence for excellent coupling to a linearly polarized TEM00 mode. In Fig. 2e we present experimentally determined efficien- cies of the vertical couplers, which we acquire by cou- pling a tunable continuous-wave Ti:Sapphire laser to the structures in a cryostat using a 0.9 NA objective. We then collect the out-coupled beam with a single-mode polarization-maintaining fiber (PMF, black data points) and a multimode fiber (MMF, red data points). The ex- perimental results show peak efficiencies of ≈ 21 % for PMF and ≈ 26.5 % for MMF, with broadband perfor- mance of > 70 nm (PMF) and > 90 nm (MMF). The small discrepancy between the measurements with PMF and MMF suggests that we couple very efficiently from the waveguide mode back into the fundamental free-space mode TEM00. Moreover, the numerical simulation (blue line) agrees well with the experimental results. Imposing fabrication constraints, such as minimum a b c 1.00 ) . u . a ( s t n u o C 0.75 0.50 0.25 0.00 1x 10x 1x 4 Grating Coupler Notch 2 μm 2 μm 0.04 0.03 0.02 0.01 0.00 756 757 758 759 Grating Coupler Notch Notch d 1.00 0.75 0.50 ) . u . a ( s t n u o C 0.25 700 725 750 775 800 825 Wavelength (nm) 0.00 750 753 Wavelength (nm) 756 759 FIG. 3: Suspended rectangular diamond nanobeams with optical interfaces. Scanning electron micrographs of nanobeam photonic crystal cavities in (a) with inverse-designed vertical couplers and in (b) with notches as interfaces for in- and out-coupling. Fields inside the cavities are depicted in red. (c) Transmission measurements using a supercontinuum light source. The red spectrum corresponds to the coupler device in (a), black and green spectra to the notch device in (b). The spectra are offset for better visualization and the cavity resonances are indicated by blue arrows. (d) Spectra acquired by coupling a supercontinuum light source directly to the cavity and out through a vertical coupler (red line) or a notch (green line). Inset corresponds to the data inside the gray box. feature sizes, on the design optimization guarantees not only high fabrication yield but also robust performance, as we demonstrate in Fig. 2f. Here we overlay transmis- sion spectra of 15 different devices acquired with a super- continuum source. During the experiments we purposely constrained ourselves to coarse alignment to confirm the robustness to alignment imperfections. The result of our analysis is shown in Fig. 2f, where the solid black line is the mean value of all couplers and the red shaded area indicates the standard deviation at a given wavelength. Moreover, the average efficiency of 30 devices fabricated with various doses is 24.2 %. Diamond quantum optical interfaces The vertical coupler presented in this work provides a compact, robust, and efficient solution for free-space in- terfaces in cavity quantum electrodynamics. In partic- ular, our design is optimized to be compatible with si- multaneous fabrication of high-Q/V resonators for quan- tum optics experiments, as we avoid additional fabrica- tion steps31,32 that could impact the resonator perfor- mance. In this section we therefore investigate coupling to the modes of nanophotonic resonators, which are used in quantum optics to enhance light-matter interactions41 and to facilitate efficient integration of quantum emit- ters into optical circuits. We study nanobeam photonic crystal (PhC) cavities, which host a TE mode as shown in Figs. 3a and b. With a supercontinuum light source we acquire the transmission spectra shown in Fig. 3c by coupling a free-space laser beam into the TE fundamen- tal mode of the nanobeam and subsequently into PhC modes. The data in red correspond to the device with vertical couplers, while the black spectrum corresponds 5 2 μm a b ) . u . a ( s t n u o C 3.0 2.5 2.0 1.5 1.0 0.5 0.0 c Upper Beam Both Beams Lower Beam 746 748 750 752 754 756 758 Wavelength (nm) FIG. 4: Interfacing grating couplers with diamond photonic circuit. (a) Diamond photonic circuit, which could be used to entangle two emitters inside the two cavities. The circuit consists of a grating coupler, followed by a waveguide-splitter, and two resonators, the outputs of which are then recombined in a waveguide-splitter and coupled off-chip through a grating coupler. (b) Spectra of the nanobeams from the device shown in (a). Green, black and red data correspond to the upper, both and the lower nanobeam, respectively. (c) Demonstration that cavities with fabrication induced frequency offset can be tuned in resonance via gas tuning; colorbar corresponds to normalized counts. to a cavity with notches as the free-space interface for the same input power and integration time. The count rates of the device with notches as an interface are more than two orders of magnitude smaller, for which we com- pensate by integrating 10 times longer (data in green). When comparing the cavity resonances (blue arrows), we find a > 550-fold increase in counts of the vertical cou- pler over the notch device for comparable quality factors (Q ≈ 4000). This result matches well with the 625-fold enhancement that we expect from simulations. This im- provement in coupling efficiencies allows for dramatically decreased experimental times (in some cases from weeks to minutes of photon integration), thereby opening op- portunities for larger-scale experiments. In Fig. 3d we present spectra, where we couple the laser light directly to the cavity and optimize the alignment to collect maxi- mum counts from the vertical coupler (red) and the notch (green). From this measurement we can conclude that the extraction efficiency of light coupling from the cavity mode to the waveguide is ≈ 24 times greater for a verti- cal coupler than that for a notch, which corresponds well to the transmission experiment. Inverse-designed diamond photonic circuit For applications in quantum technologies, many nodes need to be connected to scale from single qubits to large, interconnected qubit arrays.8,10,18 This requires the ex- citation of emitters in multiple cavities, the interference of their emission on beamsplitters, and subsequently the efficient collection and detection of photons. However, up until now elements such as waveguide-splitters have posed a major challenge in suspended diamond photon- ics, as state-of-the-art fabrication using angled etch is not conducive to variations in the device geometry. In contrast, as shown in Fig. 4a, we can fabricate a con- ceptual circuit comprised of three components with com- pletely different geometries: vertical couplers, waveguide- splitters, and nanobeam PhC cavities. The device is de- signed to interfere the transmission of two nanobeam PhC cavities at an inverse-designed waveguide-splitter with a 50 : 50 splitting ratio and simulated efficiencies of 95 %. We address the cavities separately or simulta- neously by top-down excitation with a supercontinuum source focused on the cavities directly, as presented in Fig. 4b. The resonances of the two beams are detuned by < 1 nm because of fabrication imperfections. We tune the two cavities into and out of resonance via gas conden- sation, as shown in Fig. 4c. Comparing the amplitudes of the cavity on and off resonance suggests constructive in- terference, indicating that the cavities are approximately in phase and have the same polarization. With this 6 FIG. 5: Designs for high efficiency vertical couplers. (a) Vertically symmetric coupler suspended in air. Vertically asymmetric couplers employing partial etch with tilted incident laser beam (10◦): Vertical coupler (b) on SiO2, (c) suspended in air, (d) on SiO2 with aluminum back-reflector, and (e) suspended in air with aluminum back-reflector. (f ) Simulated efficiencies of the devices shown in (a)-(e): 44.7 %, 51.0 %, 67.9 %, 72.4 %, and 86.6 %. concept circuit, we show that inverse design can over- come limitations of classical photonics and enables large- scale on-chip quantum optics experiments. Extending this work we can increase compactness by combining sev- eral functionalities into a single device, design circuits for arbitrary emitter locations, assure phase-matching across different paths of the circuits, and optimize for specific bandwidths. Such a platform can then utilize the scalability that diamond color centers offer: site- controlled implantation of high quality color centers42,43 and small inhomogeneous broadening,44 which can be overcome by cavity-enhanced Raman emission17,20 or strain tuning.45 -- 48 Highly efficient free-space-waveguide interfaces Ultimately the implementation of scalable quantum net- works requires efficiencies of building blocks close to unity. Efficiencies of > 90 % can be achieved with fiber tapers,49 which have the drawback of significantly larger footprints. To achieve comparable efficiencies, we reduce the fabrication constraints to 60 nm feature sizes, in- crease the laser spot size, device footprint, and waveguide width. This allows us to improve the simulated efficiency to 44.7 %. However, vertically symmetric devices, such as shown in Fig. 5a cannot exceed 50 % efficiency. For fur- ther improvements, we tilt the incident laser beam by 10◦ and break the symmetry along the z-axis of the couplers via a partial etch.32 In Fig. 5b we show diamond devices on SiO2 with efficiencies of 51.0 %. Such devices could be achieved through diamond thin-film on SiO2 production5 or pick and place techniques50 and are a promising route for a range of applications including long-distance en- tanglement schemes, and nonlinear optics. Devices sus- pended in air (Fig. 5c) have a larger refractive index con- trast and show efficiencies of up to 67.9 %. Additionally employing back-reflectors31 as shown in Fig. 5d and e re- sults in efficiencies of 72.4 % and 86.6 %, for diamond on SiO2 and suspended structures, respectively. The back- reflector distance to the coupler (400 nm and 650 nm) is significantly shorter than the photon wave-packet and optimized to match the phase between reflected and di- rectly coupled photons. These findings are encouraging for the development of highly efficient and compact pho- tonic free-space interfaces as an alternative to fiber tapers for quantum photonic applications at the single-photon level. Moreover, many experiments will require optical driving of individual emitters to compensate for their spectral broadening via Raman processes.17,20 This in- dividual addressing is easier to implement in free-space coupling configurations than with many tapered fibers inside a cryostat. High efficiencies and compactness will be crucial in these experiments, as losses will be the lim- iting factor. Thus, inverse design is likely to play a major role in the development of such photonic circuits.30 Discussion In summary, we employ optimization-based inverse de- sign methods to overcome the constraints of cutting-edge diamond nanofabrication and to develop efficient build- ing blocks for diamond nanophotonic circuits.51 In op- tical free-space couplers and a small diamond photonic circuit we attain the crucial properties of high efficiency, compactness, and robustness. This work now enables more complex quantum circuits, where compact solutions for a variety of device components such as pulse shapers, splitter trees, phase delays,52 and mode converters53 are critical. Thus, this progress lays the foundation for scal- ing to larger quantum networks8,10,54 with spins6 embed- ded in quantum nodes. In addition, inverse design meth- ods can be applied to other promising material platforms that host quantum emitters and have challenging fabri- cation protocols, such as silicon carbide55 and yttrium orthovanadate.56 Acknowledgements We acknowledge the help of Usha Raghuram and Elmer Enriquez with RIE. This work is financially sup- ported by Army Research Office (ARO) (award no. 7 W911NF1310309), Air Force Office of Scientific Re- search (AFOSR) MURI Center for Attojoule Nano- Optoelectronics (award no. FA9550-17-1-0002), Na- tional Science Foundation (NSF) Division Of Electri- cal, Communications & Cyber Systems (ECCS) (award no. 1838976), and Gordon and Betty Moore Foundation; C.D. acknowledges support from the Andreas Bechtol- sheim Stanford Graduate Fellowship and the Microsoft Research PhD Fellowship. K.Y.Y. and M.R. acknowledge support from the Nano- and Quantum Science and En- gineering Postdoctoral Fellowship. D.M.L. acknowledges support from the Fong Stanford Graduate Fellowship. D.M.L. and A.E.R. acknowledge support from the Na- tional Defense Science and Engineering Graduate (ND- SEG) Fellowship Program, sponsored by the Air Force Research Laboratory (AFRL), the Office of Naval Re- search (ONR) and the Army Research Office (ARO). D.V. acknowledges funding from FWO and European Unions Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 665501. We thank Google for providing computational resources on the Google Cloud Platform. Part of this work was performed at the Stanford Nanofabrication Fa- cility (SNF) and the Stanford Nano Shared Facilities (SNSF), supported by the National Science Foundation under award ECCS-1542152. Author contributions C.D. and J.V. conceived and de- signed the experiment. C.D. developed the fabrication, fabricated the sample and measured and analyzed the data. D.V., N.V.S., C.D. and L.S. conducted inverse design optimization of photonic components. K.Y.Y., D.M.L. and A.Y.P. contributed to the sample fabrica- tion. A.E.R. and S.S. contributed to optical measure- ments. J.L.Z., M.R. and K.G.L. provided expertise. J.V. supervised the project. All authors participated in the discussion, understanding of the results, and the prepa- ration of the manuscript. 1 Atature, M., Englund, D., Vamivakas, N., Lee, S. Y. & Wrachtrup, J. Material platforms for spin-based photonic quantum technologies. Nat. Mater. 3, 38 -- 51 (2018). 2 Awschalom, D. D., Hanson, R., Wrachtrup, J. & Zhou, B. B. Quantum technologies with optically interfaced solid- state spins. Nat. Photonics 12, 516 -- 527 (2018). 3 Khanaliloo, B. et al. Single-crystal diamond nanobeam waveguide optomechanics. Phys. Rev. X 5, 041051 (2015). 4 Burek, M. J. et al. Diamond optomechanical crystals. Op- tica 3, 1404 -- 1411 (2015). 5 Hausmann, B. J. M., Bulu, I., Venkataraman, V., Deotare, P. & Loncar, M. Diamond nonlinear photonics. Nat. Pho- tonics 8, 369 -- 374 (2014). 6 Rose, B. C. et al. Observation of an environmentally in- sensitive solid-state spin defect in diamond. Science 361, 60 -- 63 (2018). 7 Aharonovich, I., Greentree, A. D. & Prawer, S. Diamond photonics. Nat. Photonics 5, 397 -- 405 (2011). 8 Cirac, J. I., Zoller, P., Kimble, H. J. & Mabuchi, H. Quan- tum state transfer and entanglement distribution among distant nodes in a quantum network. Phys. Rev. Lett. 78, 3221 ((1997). 9 Briegel, H.-J., Dur, W., Cirac, J. I. & Zoller, P. Quantum repeaters: The role of imperfect local operations in quan- tum communication. Phys. Rev. Lett. 81, 5932 ((1998). 10 Kimble, H. J. The quantum internet. Nature 453, 1023 -- 1030 (2008). 11 Bernien, H. et al. Two-photon quantum interference from separate nitrogen vacancy centers in diamond. Phys. Rev. Lett. 108, 043604 (2012). 12 Sipahigil, A. et al. Quantum interference of single photons from remote nitrogen-vacancy centers in diamond. Phys. Rev. Lett. 108, 143601 (2012). 13 Bernien, H. et al. Heralded entanglement between solid- state qubits separated by three metres. Nature 497, 8690 (2013). 14 Pfaff, W. et al. Unconditional quantum teleportation be- tween distant solid-state quantum bits. Science 345, 532 -- 535 (2014). 15 Hensen, B. et al. Loophole-free Bell inequality violation using electron spins separated by 1.3 kilometres. Nature 526, 682 -- 686 (2015). 16 Gao, W. B., Imamoglu, A., Bernien, H. & Hanson, R. Co- herent manipulation, measurement and entanglement of individual solid-state spins using optical fields. Nat. Pho- tonics 39, 363373 (2015). 17 Sun, S. et al. Cavity-enhanced Raman emission from a single color center in a solid. Phys. Rev. Lett. 121, 083601 (2018). 18 Lemonde, M. A. et al. Phonon networks with silicon- vacancy centers in diamond waveguides. Phys. Rev. Lett. 120, 213603 (2018). 19 Ladd, T. D. et al. Quantum computers. Nature 464, 45 -- 53 (2010). 20 Sipahigil, A. et al. An integrated diamond nanophotonics platform for quantum-optical networks. Science 354, 847 -- 850 (2016). 21 Evans, R. E. et al. Photon-mediated interactions between quantum emitters in a diamond nanocavity. Science 362, 662 -- 665 (2018). 22 Balasubramanian, G. et al. Nanoscale imaging magnetom- etry with diamond spins under ambient conditions. Nature 455, 648651 (2008). 23 Degen, C. L., Reinhard, F. & Cappellaro, P. Quantum sensing. Rev. Mod. Phys. 89, 035002 (2017). 24 Burek, M. J. et al. High quality-factor optical nanocavi- ties in bulk single-crystal diamond. Nat. Comm. 5, 5718 (2014). 25 Khanaliloo, B., Mitchell, M., Hryciw, A. C. & Barclay, P. E. High-Q/V monolithic diamond microdisks fabricated with quasi-isotropic etching. Nano Lett. 15, 5131 -- 5136 (2015). 26 Mouradian, S., Wan, N. H., Schroder, T. & Englund, D. Rectangular photonic crystal nanobeam cavities in bulk diamond. Appl. Phys. Lett. 111, 021103 (2017). 27 Wan, N. H., Mouradian, S. & Englund, D. Two- dimensional photonic crystal slab nanocavities on bulk single-crystal diamond featured. Appl. Phys. Lett. 112, 141102 (2018). 28 Mitchell, M., Lake, D. P. & Barclay, P. E. Realizing q >300 000 in diamond microdisks for optomechanics via etch optimization. APL Photonics 4, 016101 (2019). 29 Molesky, S. et al. Inverse design in nanophotonics. Nat. Photonics 12, 659 -- 670 (2018). 30 Su, L. et al. Fully-automated optimization of grating cou- plers. Opt. Express 26, 4023 -- 4034 (2018). 31 Van Laere, F. et al. Compact and highly efficient grating couplers between optical fiber and nanophotonic waveg- uides. J. Light. Technol. 25, 151 -- 156 (2007). 32 Vermeulen, D. et al. High-efficiency fiber-to-chip grat- ing couplers realized using an advanced CMOS-compatible silicon-on-insulator platform. Opt. Express 18, 18278 -- 18283 (2010). 33 Rath, P., Khasminskaya, S., Nebel, C., Wild, C. & Pernice, W. H. Grating-assisted coupling to nanophotonic circuits in microcrystalline diamond thin films. J. Light. Technol. 4, 300 -- 305 (2013). 34 Faraon, A. et al. Quantum photonic devices in single- crystal diamond. New J. Phys. 15, 025010 (2013). 35 Piracha, A. H. et al. Scalable fabrication of integrated nanophotonic circuits on arrays of thin single crystal di- amond membrane windows. Nano Lett. 16, 3341 -- 3347 8 (2016). 36 Zhou, X. et al. High-Efficiency Shallow-Etched Grating on GaAs Membranes for Quantum Photonic Applications. Appl. Phys. Lett. 113, 251103 (2018). 37 Gould, M., Schmidgall, E. R., Dadgostar, S., Hatami, F. & Fu, K.-M. C. Efficient extraction of zero-phonon-line pho- tons from single nitrogen-vacancy centers in an integrated gap-on-diamond platform. Phys. Rev. Applied 6, 011001 (2016). 38 Piggott, A. Y. et al. Inverse design and implementation of a wavelength demultiplexing grating coupler. Sci. Rep. 4, 7210 (2014). 39 Lu, J. & Vuckovic, J. Nanophotonic computational design. Opt. Express 21, 13351 -- 13367 (2013). 40 Piggott, A. Y., Petykiewicz, J., Su, L. & Vuckovi´c, J. Fabrication-constrained nanophotonic inverse design. Sci. Rep. 7, 1786 (2017). 41 Zhang, J. L. et al. Strongly cavity-enhanced spontaneous emission from silicon-vacancy centers in diamond. Nano Lett. 18, 1360 -- 1365 (2018). 42 Evans, R. E., Sipahigil, A., Sukachev, D. D., Zibrov, A. S. & Lukin, M. D. Narrow-linewidth homogeneous optical emitters in diamond nanostructures via silicon ion implan- tation. Phys. Rev. Applied 5, 044010 (2016). 43 Schroder, T. et al. Scalable focused ion beam creation of nearly lifetime-limited single quantum emitters in diamond nanostructures. Nat. Commun. 8, 15376 (2017). 44 Rogers, L. et al. Multiple intrinsically identical single- photon emitters in the solid state. Nat. Commun. 5, 4739 (2014). 45 Meesala, S. et al. Enhanced strain coupling of nitrogen- vacancy spins to nanoscale diamond cantilevers. Phys. Rev. Appl. 5, 034010 (2016). 46 Meesala, S. et al. Strain engineering of the silicon-vacancy center in diamond. Phys. Rev. B 97, 205444 (2018). 47 Sohn, Y.-I. et al. Controlling the coherence of a diamond spin qubit through its strain environment. Nat. Commun. 9, 2012 (2018). 48 Machielse, B. et al. Quantum interference of electrome- chanically stabilized emitters in nanophotonic devices. preprint at http://arXiv.org/abs/1901.09103 (2019). 49 Burek, M. J. et al. Fiber-coupled diamond quantum nanophotonic interface. Phys. Rev. Applied 8, 024026 (2017). 50 Kim, J.-H. et al. Hybrid integration of solid-state quantum emitters on a silicon photonic chip. Nano Lett. 17, 7394 -- 7400 (2017). 51 O'Brien, J. L., Furusawa, A. & Vuckovi´c, J. Photonic quantum technologies. Nat. Photonics 3, 687 -- 695 (2009). 52 Vampa, G., Fattahi, H., Vuckovic, J. & Krausz, F. Attosec- ond nanophotonics. Nat. Photonics 11, 210212 (2017). 53 Lu, J. & Vuckovi´c, J. Objective-first design of high-efficiency, small-footprint couplers between arbitrary nanophotonic waveguide modes. Opt. Express 20, 7221 -- 7236 (2012). 54 Wehner, S., Elkouss, D. & Hanson, R. Quantum internet: A vision for the road ahead. Science 362, 6412 (2018). 55 Koehl, W. F., Buckley, B. B., Heremans, F. J., Calusine, G. & Awschalom, D. D. Room temperature coherent con- trol of defect spin qubits in silicon carbide. Nature 479, 84 -- 87 (2011). 56 Zhong, T. et al. Nanophotonic rare-earth quantum mem- ory with optically controlled retrieval. Science 357, 1392 -- 1395 (2017).
1809.06261	1	1809	2018-08-06T04:18:55	Enhanced optical Kerr nonlinearity of graphene/Si hybrid waveguide	[ "physics.app-ph" ]	In this work, we experimentally study the optical kerr nonlinearities of graphene/Si hybrid waveguides with enhanced self-phase modulation. In the case of CMOS compatible materials for nonlinear optical signal processing, Si and silicon nitride waveguides have been extensively investigated over the past decade. However, Si waveguides exhibit strong two-photon absorption (TPA) at telecommunication wavelengths, which lead to a significant reduction of nonlinear figure of merit. In contrast, silicon nitride based material system usually suppress the TPA, but simultaneously leads to the reduction of the Kerr nonlinearity by two orders of magnitude. Here, we introduce a graphene/Si hybrid waveguide, which remain the optical properties and CMOS compatibility of Si waveguides, while enhance the Kerr nonlinearity by transferring patterned graphene over the top of the waveguides. The graphene/Si waveguides are measured with a nonlinear parameter of 510 W-1m-1. Enhanced nonlinear figure-of-merit (FOM) of 2.48 has been achieved, which is three times higher than that of the Si waveguide. This work reveals the potential application of graphene/Si hybrid photonic waveguides with high Kerr nonlinearity and FOM for nonlinear all-optical signal processing.	physics.app-ph	physics	Enhanced optical Kerr nonlinearity of graphene/Si hybrid waveguide Qi Feng1, Hui Cong1, Bin Zhang1,2, Wenqi Wei1,2,3, Yueyin Liang1,2, Shaobo Fang1, Ting Wang1* and Jianjun Zhang1 1Institute of Physics, Chinese Academy of Sciences, Beijing, 100190, People's Republic of China 2School of Physical Sciences, University of Chinese Academy of Sciences, Beijing, China, 3School of Physics and Technology, Wuhan University, Wuhan, 430072, China 100190 Abstract In this work, we experimentally study the optical kerr nonlinearities of graphene/Si hybrid waveguides with enhanced self-phase modulation. In the case of CMOS compatible materials for nonlinear optical signal processing, Si and silicon nitride waveguides have been extensively investigated over the past decade. However, Si waveguides exhibit strong two-photon absorption (TPA) at telecommunication wavelengths, which lead to a significant reduction of nonlinear figure of merit. In contrast, silicon nitride based material system usually suppress the TPA, but simultaneously leads to the reduction of the Kerr nonlinearity by two orders of magnitude. Here, we introduce a graphene/Si hybrid waveguide, which remain the optical properties and CMOS compatibility of Si waveguides, while enhance the Kerr nonlinearity by transferring patterned graphene over the top of the waveguides. The graphene/Si waveguides are measured with a nonlinear parameter of 510 W-1m-1. Enhanced nonlinear figure-of-merit (FOM) of 2.48 has been achieved, which is three times higher than that of the Si waveguide. This work reveals the potential application of graphene/Si hybrid photonic waveguides with high Kerr nonlinearity and FOM for nonlinear all-optical signal processing. Introduction Optical nonlinear effects in CMOS-compatible integrated optical devices are of great significance as they can be explored to realize a variety of functionalities ranging from all-optical signal processing to light generation [1-5]. Silicon-on-insulator (SOI) has been regarded as a popular platform for ultra-dense on- chip integration of photonic and electronic circuitry due to its compatibility with CMOS fabrication. In addition, nonlinear optical properties of silicon waveguides are also heavily explored over the past decade, such as stimulated Raman scattering (SRS), Raman amplification, self-phase modulation (SPM), four- wave mixing, super-continuum generation [2,6-8]. However, the existence of two-photon-absorption (TPA) at telecom wavelength (around 1550 nm) in Si platform leads to a strong degradation in the value of nonlinear figure-of-merit, which limits the power efficiency of nonlinear functionalities. In addition, TPA increases the photon loss in the process as well as generating carriers subsequently producing usually undesired free-carrier absorption (FCA) and free-carrier dispersion (FCD). TPA and FCA generally cause optical losses, which in turn lower the peak power inside the waveguide and therefore reduce the conversion efficiency of optical nonlinear process [9,10]. There are other promising platforms such as chalcogenide glass and AlGaAs, which possess high nonlinearity and low TPA, but highly challenging fabrication processes limit their usage in CMOS compatible applications [11-16]. Furthermore, CMOS compatible platforms such as Si3N4 and Hydex glass exhibit low TPA at telecom wavelength, thus efficiently reducing the nonlinear loss as well as linear loss, however, their nonlinear refractive index is approximately one order of magnitude smaller than that of silicon [17-22]. Therefore, the best way to fulfill the requirement of silicon nonlinear photonics is to integrate them with novel materials with high Kerr coefficient while keeping the silicon platform for its economic advantages. One of the best candidate is graphene, which has outstanding optoelectronic properties, while remians compatibilitiy for integration with silicon photonic devices. In the field of photonics, graphene has enjoyed widespread research attention in various optical devices, such as photodetectors, modulators, optical switches, optical gates, and lasers [23-29]. Some of the strengths of graphene manifest in its unique optical properties, the most common include its large tunable refractive index, high confinement factor, and a universal absorption of 0.3 %. One of the major optical properties of graphene is the giant optical Kerr nonlinearity which has been previously reported by several groups with Kerr-coefficients ranging from 10-7 to 10-13 m2/W [30-35]. In this paper, we demonstrate experimental studies of SPM process under picosecond pulses in Si waveguides and graphene/Si (G/Si) hybrid waveguides. The positioning of graphene on Si waveguides was done by precise transfer process. The effective Kerr coefficient n2 of graphene/Si hybrid waveguide is calculated to be~ 2×10-17 m2/W, which is three times higher than that of Si waveguide. Furthermore, the FOM has been enhanced from 0.7 in Si waveguides to 2.48 in G/Si hybrid waveguides. Design and Fabrication Fig. 1: (a) Schematic diagram of the graphene-Si hybrid waveguide. W, H., Lgr denote the width and height of Si wavegudie, and length of graphene covering on Si waveguide, respectively. Raman spectra (b) of the single layer graphene before (red solid curve) and after (blue dotted curve) transferring onto Si waveguide. Inset: optical microscope image of graphene on Si waveguide. Tilted SEM image of (c) sidewall and (d) inverse taper of Si waveguide, with scale bar of 1 μm and 500 nm, respectively. Here, we use single-mode silicon waveguides with dimensions of 500 nm × 340 nm × 3.5 mm, fabricated on a SOI wafer with a buried oxide layer of 3 µm. Standard e-beam lithography is used to pattern Si waveguides in JEOL-6300 (100kV) system with ZEP-520A EB resist. After development, the pattern is transferred by inductively coupled plasma (ICP) etching in Oxford PlasmaPro 100. The inverse taper with tip width of 100 nm and tip length of 50 µm are designed at each end of the waveguides to improve the efficiency of adiabatic coupling. Chemical vapor deposition (CVD)-grown monolayer graphene with a length of 60 µm was then transferred onto the waveguides by precise positioning [36]. The schematic diagram of the G/Si hybrid waveguide is shown in Figure 1(a), while (c) and (d) show good quality of sidewall of Si waveguide and inverse taper coupler. The explicit characterization of monolayer graphene is obtained by Raman spectroscopic measurements, pumped with a 488 nm laser. In Figure 1(b), Raman spectrum of the graphene layer coated waveguide device is measured by LabRAM HR 800 spectrometer. Both of the spectra show a G peak (~ 1586 cm-1) with a full width at half maximum (FWHM) of ~ 18 cm-1 and a 2D peak (~ 2700 cm-1), with a 2D-to-G peak intensity ratio of about 1.2, implying that the transferred graphene is monolayer and the corresponding chemical potential is around 0.2 eV. Raman spectra of graphene on different waveguides are measured for verification of graphene transfer reliability [34]. The results turn out to be similar as the Raman spectra presented in Figure 1(b). For comparison, identical silicon waveguides without graphene were fabricated under the same procedure. In order to investigate the third-order optical nonlinearity of G/Si hybrid waveguides, the dispersion and group velocity dispersion (GVD) are calculated using the standard Sellmeier equation by Lumerical MODE solutions, as shown in Figure 2. The group velocity dispersion is calculated to be -4.5 ps2m at the wavelength of 1550 nm. It is noted that comparing with the Si waveguide, the graphene has negligible effect on the GVD. Fig.2 : GVD and dispersion as a function of wavelength for G/Si hybrid waveguide. Inset: fundamental TE-mode optical intensity distribution of the G/Si hybrid waveguide, which were used for nonlinear refractive index calculation. Graphene layer is indicated as solid white line, and silicon core is indicated inside the surrounded by yellow lines. Results and Discussion Both Si and G/Si waveguides are pumped by a PriTel's FFL series of picosecond fiber laser. The laser produces pulses with a center wavelength of around 1548 nm at a repetition rate of 20 MHz, and a pulse duration of 1.5 ps, which are delivered with a polarization-maintaining (PM) fiber. The pulses were coupled into and out of the waveguides devices via lensed fibers and inverse taper mode-converter with a coupling loss of approximately 10 dB per facet. It is noted that one meter long PM fiber between the pulse laser and the waveguide devices was chosen in order to eliminate spectrum change induced from the nonlinear effects within the fiber. To monitor the input spectrum, a 90:10 coupler is inserted in the setup to split off 10 % of the laser power into optical spectrum analyzer (OSA, Yokogawa AQ6370D). The propagation loss measurements were carried out on Si waveguides by cut-back method. And the linear propagation loss in silicon waveguide is estimated to be (3.5±0.5) dB/cm. Here TE polarized light was coupled into the waveguide by PM fiber after polarization-controller, and the output power was monitored by fiber optic power meters (Thorlabs PM20). (a) (b) Fig.3: Experimental results of the transmission spectra of (a) comparison between the Si (magenta solid curve) and G/Si hybrid (blue solid curve) waveguides under the same input energy with 1.5 ps input pulse (spectrum denoted as red dashed curve). Inset: autocorrelation spectrum in time domain. (b) the output spectra of G/Si hybrid waveguide under various input energies. The red curve denotes the 1.5 ps pulse spectrum. Prior to the SPM measurements, as shown in inset of Figure 3(a), we measure the temporal characteristics of the input pulse by means of frequency-resolved optical gating (FROG) instrument (Coherent Solutions HR150). The Gaussian-shape pulse with a pulse width of 1.5 ps has been confirmed. The actual SPM measurements consist of simultaneously recording the spectral widths of the input and output spectra for the waveguides with graphene lengths of 60 μm. Here, the measured propagation loss induced from graphene absorption is 0.045±0.010 dB/μm. SPM measurements were carried out by measuring the transmission spectra in both Si waveguides and G/Si hybrid waveguides. It is shown in Figure 3(a) that the dotted red curve represents the spectrum of original input pulse, while the green and blue solid curves represent the output spectra for both Si waveguide and G/Si waveguide under identical input pulse energy, respectively. To note, both waveguides have the same length of 3.5 mm. The input pulse has spectral linewidth of 2.1 nm as shown in the dash curve of Figure 3(a). By comparing spectrum of Si waveguide with input pulse, the spectral broadening is assumed to be significantly less than the G/Si hybrid waveguides, which exhibits a strong 1.8π phase change. This result leads to the strong enhancement of third-order nonlinearity by introducing graphene decoration over the top of Si waveguides. The energy dependent measurements of G/Si hybrid waveguide are shown in Figure 3(b), with input pulase energies ranging from 5 pJ to 16 pJ. Since SPM alone is known to yield a symmetric spectral distribution around the injected laser frequency, the asymmetry must stem from other factors such as chirped injected laser pulses, self-steepening, GVD or changes in the free carrier density by TPA. Self-steepening can be ruled out. First, the change of n2 within narrow spectral bandwidth of the ps-pulse is negligible. Secondly, self-steepening induced spectral red- shift is absent in our experiments [37,38]. GVD should be of minor influence as well, as will be discussed in detail later. Therefore, FCA and TPA effects play dominant role in our case. Fig.4: Pin/Pout vs Pin for the Si waveguide and G/Si hybrid waveguide, denoted as blue stars and red circles, respectively. Linear fit for both Si waveguide and G/Si hybrid are denoted as dashed lines. Prior to the analysis of the nonlinear transmission, the relation between the average output power Pout and the average input power Pin is recalled in case of a dominant TPA effect: 𝑃𝑖𝑛 𝑃𝑜𝑢𝑡 = 2𝐼𝑚(𝛾)𝐿𝑒𝑓𝑓𝑒𝛼𝐿𝑃𝑖𝑛 + 𝑒𝛼𝐿 (1) where α=3.5 dB/cm is the linear propagation loss, Im(γ)=βTPA/(2Aeff) is the imaginary part of the γ nonlinear coefficient due to TPA, L is the waveguide length and Leff the effective optical path length reduced by the linear propagation loss through Leff=(1-e-αL)/α. Thus, this equation discloses a linear relation between the ratio Pin/Pout and the measured input power Pin with the slope being proportional to the nonlinear coefficient βTPA. From the similar slope shown in Figure 4, the TPA coefficients βTPA for both Si waveguide and G/Si hybrid waveguide can be extracted with similar values of 0.5 cm/GW. According to this result, it is verified that the effective Kerr nonlinearity is significantly larger for G/Si hybrid waveguides. The output spectra from G/Si hybrid waveguides are depicted in Figure 3(b). It shows that with increasing Pin, the optical Kerr effect induced self-phase modulation becomes asymmetric, which will be discussed in details as below. NLSE simulation We calculated the case where a Gaussian-shape laser pulse is coupled into the Si waveguide and G/Si hybrid waveguide. By using the nonlinear Schrodinger equation (NLSE) with the split-step Fourier method, the on-chip SPM process can be simulated with the equation below [37] ∂A ∂z = − 1 2 𝛼𝐴 + 𝑖 ∑ 10 𝑚=2 𝑖𝑚𝛽𝑚 𝑚! ∂𝑚𝐴 ∂τ𝑚 + 𝑖𝛾𝐴2𝐴 − 𝜎 2 (1 + 𝑖𝜇)𝑁𝑐𝐴 (2) where A(z,t) is the slowly varying temporal envelop along the length z of a nonlinear medium, γ0=ω0n2/cAeff is the nonlinear parameter, ω0 is the optical frequency, βm is the m-th order dispersion coefficient, n2 is nonlinear Kerr coefficient, βTPA is the two-photon absorption coefficient, Aeff is the effective mode area, Nc is the free carrier density, σ is the free carrier absorptin coefficient, μ is the free carrier dispersion coefficient, α is the linear loss parameter. Thus Nc can be obtained by solving: ∂N𝑐 ∂t = 𝛽𝑇𝑃𝐴 2ℎ𝜔 𝐴4 2 − 𝑁𝑐 𝜏𝑐 𝐴𝑒𝑓𝑓 (3) where τc is the effective lifetime of free carriers with an estimated value of 0.5 ns. Therefore, the profile of Nc(t) is calculated by solving Equation (3) approximately near the front end of the waveguide where E(z,t)2 is close to its input. Noting that pulse width T0<τc, the τc term can be ignored as carriers do not have enough time to recombine over the pulse duration. The carrier density is then given by: 𝑁𝑐(t) ≈ 𝛽𝑇𝑃𝐴𝐼0 2ℎ𝑣0 2𝑇0 √ 𝜋 8 [1 + erf⁡(√2𝑡 𝑇0 )] (4) The pulse dynamics are governed by the interplay of SPM and dispersion whose relative strengths can be determined by several characteristic lengths, namely the GVD and third-order dispersion (TOD) lengths, defined as LD = T0 3/β3, respectively, and the nonlinear length, defined as LNL= 2/β2 and 𝐿𝐷 ′ = T0 1/γP0. Beside TPA, free-carrier effect within silicon waveguides, such as FCD and FCA, could lead to the asymmetry of SPM spectrum of G/Si hybrid waveguides. Especially in the case of FCD effect, it can lead to the decrement of refractive index and thus cause the acceleration of the pulse. The last term of Equation 2 reveals that free carriers interact with the optical pulse by both modulating its phase through FCD (which acts couter to the Kerr effect), as well as reducing intensity by FCA. As depicted by the rate equation (Equation 3), the generation of free carriers in time follows the evolution of the pulse intensity squared -- that is, the free-carrier concentration will be negligible at the leading edge of the pulse and be significant at the trailing edge. Therefore, FCA causes nonliear absorption of the trailing edge of the pulse, generating pulse asymmetry. In the case of G/Si hybrid waveguides, the enhanced free carrier dentiy Nc introduced by graphene could lead to a strong spectral asymmetry in the SPM measurement. Here, the carrier density of 5.85×1016 cm-3 and 8.15×1016 cm-3, the free carrier absorption coefficient of 1.45×10-17 cm-2 and 6×10-17 cm-2, for Si and G/Si hybrid waveguides, respectively, were applied to our numerical calculation. Here, the nonlinear length LNL, dispersion length LD, are calculated to be 0.65 mm and 19.26 mm, respectively. Given that the LD is much longer and the LNL is much shorter than the waveguide length (3.5 mm) for both Si and G/Si hybrid waveguides, the pulse dynamics will be dominated by the third- order nonlinearity rather than the dispersion in the wavegudies. Equations (2) and (3) are then solved using a split-step Fourier transform method to model the behaviour of the pulse propagation in Si and G/Si hybrid waveguides. As shown in Figure 5, the simulated spectra have relatively good agreement with our experimental results under various input energies from 5-16 pJ. The extracted 𝑛2 value of G/Si hybrid waveguide is here calculated to be 2×10-17 m2/W, which is three times larger than that of the Si waveguide. Fig.5: Experimentally measured and numerically calculated spectra of the output picosecond pulse propagating along G/Si hybrid waveguide for various coupled energy, denoted as solid and dashed curves, respectively. As shown in Figure 6, we compared the spectral broadening for Si waveguide and G/Si hybrid waveguide as a function of coupled peak power. It is shown that spectral broadening of 3.5 mm long Si and G/Si hybrid waveguides are around 11.5 nm and 14.7 nm, respectively. The peak phase shift Φmax (in radians) experienced by the pulse is given by Φ𝑚𝑎𝑥 = 2𝜋 𝑛2𝑃𝐿 𝜆𝐴𝑒𝑓𝑓 , where P is the peak pulse power. Using the effective area Aeff of 0.16 μm for both Si and G/Si waveguides, the maximum phase shift Φ𝑚𝑎𝑥 are extracted as 0.54π and 1.8 π, respectively. The enhanced Kerr nonlinearity in G/Si waveguide results in additional 1.3π phase shift with calculated nonlinear parameter γ of 510 W-1/m. Nonlinear figure-of-merit can be defined as FOM= n2/(λβTPA), which is a measure of the optical nonlinear efficiency of the medium when both the nonlinear refractive index and nonlinear loss mechanisms are accounted for [39,40]. In addition, it provides a useful dimensionless measurement of suitability of the material for nonlinear switching. For a nonlinear directional coupler, the required nonlinear phase shift for optical switching is 4π and thus it must satify FOM > 2 for such devices [41]. For other devices, such as a nonlinear Mach-Zehnder interferometer, a π phase change is sufficient and the nonlinear FOM only needs to satisfy the condition FOM > 1/2. Silicon platform normally exhibit nonlinear FOM of ~0.6 at 1.55 μm, which is insufficient for optical switching applications. In this work, the FOM of G/Si hybrid waveguide is calculated to be approximately 2.48, which is higher than that of Si, SiGe and hydrogenated amorphous-Si waveguides, as shown in Table 1. In addition, although chalcogenide glass and AlGaAs possess high nonlinearity and low TPA, they are limited to applications where CMOS compatibility is not required due to the challenging fabrication for highly efficient waveguides. While for other platforms such as silicon nitride and Hydex glass, they are COMS compatible and exhibit much lower nonlinear loss and linear loss due to low TPA at telecom wavelength, however, the refractive index is ten times smaller than that of silicon. A key goal of all-optical chips is to reduce both the device footprint and the operation power. The significant improvement in both the optical kerr nonlinearity and nonlinear FOM in G/Si raise the prospect to provide a truly practical and viable platform for nonlinear photonic applications in the telecommunication wavelength window. This reveals that the corporation of single layer graphene can be employed to increase the nonlienar performance of silicon-based waveguides in all-optical signal processing. Fig.6: Measured SPM-induced spectral broadening for both Si and G/Si hybrid waveguides, denoted as red squares and blue circles, respectively, as a function of the coupled peak power. Conclusion In conclusion, enhancement of third-order nonlinearity in G/Si hybrid waveguide has been studied here by self-phase modulation experiments, and enhanced spectrum broadening has been observed in G/Si hybrid waveguide. Although the decorated graphene exhibits a relatively weak evanescent fields in such structure, three times magnitude enhanced Kerr nonlinearity is still achieved on G/Si hybrid waveguide with an overall optical nonlinear parameter of 510 W-1/m. The FOM has been improved as well from 0.7 to 2.48 comparing with Si waveguide. This work provides an insight that on-chip integration of graphene with CMOS-compatible silicon platform enables the realization of devices that possess many novel all- optical functions at telecommunication wavelength. Table 1: The parameters used for numerical simulation in NLSE. Platform Input pulse Waveguide Input peak 𝛽𝑇𝑃𝐴 FOM Width (fs) Length (mm) Power (W) (cm/GW) G/Si hybrid (this work) 1000 Si (bulk) [18] Si [6] Si-organic hybrid [42] Si0.3Ge0.7 [43] α-Si [44] α-Si:H [45] 130 376 1000 120 1600 180 3.5 - 71 6.9 6 10 7 20 4500 60 20 167 4 2.8×105 0.50 - 0.45 0.754 5.53 0.62 4.1 2.48 0.37 0.83 2.1 0.26 2.0 0.66 Acknowledgement The authors acknowledge the graphene transfer process by Dr. F.G. Yan from Institute of Semiconductors, Chinese Academy of Sciences in China. Financial support was provided by the National Natural Science Foundation of China (Grants 11504415, 11434041, 11574356 and 161635011), the Ministry of Science and Technology (MOST) of Peoples Republic of China (2016YFA0300600 and 2016YFA0301700), and the Key Research Program of Frontier Sciences, CAS (Grant No. QYZDB- SSW-JSC009). Reference [1] J. Leuthold, C. Koos and W. Freude, "Nonlinear silicon photonics", Nature Photon. 4, 535-544 (2010) [2] R. Claps, D. Dimitropoulos, V. Raghunathan, Y. Han, and B. Jalali, "Observation of stimulated Raman amplification in silicon waveguides", Opt. Express 11, 1731 -- 1739 (2003). [3] T. K. Liang, and H. K. Tsang, "Nonlinear absorption and Raman scattering in silicon-on-insulator optical waveguides", IEEE J. Selected Topics in Quant. Electron. 10, 1149 -- 1153 (2004). [4] H. Rong, et al. "An all-silicon Raman laser", Nature 433, 292 -- 294 (2005). [5] M. A. Foster, et al. "Broad-band optical parametric gain on a silicon photonic chip", Nature 441, 960 -- 963 (2006). [6] H. K. Tsang, C. S. Wong, T. K. Liang, I. E. Day, S. W. Roberts, A. Harpin, J. Drake, and M. Asghari, "Optical dispersion, two-photon absorption and self-phase modulation in silicon waveguides at wavelength", Appl. Phys. Lett. 80, 416 -418(2002) [7] R. L. Espinola, J. I. Dadap, R. M. Osgood, S. J. McNab, and Y. A. Vlasov, "C-band wavelength conversion in silicon photonic wire waveguides", Opt. Express 13, 4341 -- 4349 (2005). [8] L. H. Yin, Q. Lin, and G. P. Agrawal, "Soliton fission and supercontinuum generation in silicon waveguides", Opt. Lett. 32, 391 -394(2007). [9] L. H. Yin and G. P. Agrawal, "Impact of two-photon absorption on self-phase modulation in silicon waveguides", Opt. Lett. 32, 2031-2034 (2007) [10] Q. Lin, O. J. Painter, and G. P. Agrawal, "Nonlinear optical phenomena in silicon waveguides: Modeling and applications", Opt. Express 15, 16604-16644 (2007). [11] H. Lin, Y. Song, Y. Huang, D. Kita, S. K. Wang, L. Li, J. Li, H. Zheng, Z. Luo, H. Wang, S. Novak, A. Yadav, C-C. Huang, R-J. Shiue, D. Englund, T. Gu, D. Hewak, K. Richardson, J. Kong, J. Hu, "Chalcogenide glass-on-graphene photonics", Nature Photon.11, 798-806 (2017) [12] B. J. Eggleton, B. Luther-Davies, and K. Richardson, "Chalcogenide photonics", Nat. Photon. 5, 141 -- 148 (2011). [13] V. G. Ta'eed, et al. "Ultrafast all-optical chalcogenide glass photonic circuits", Opt. Express 15, 9205 -- 9221 (2007). [14] P. Apiratikul, J. J. Wathen, G. A. Porkolab, B. Wang, L. He, T. E.Murphy, and C. J. Richardson, "Enhanced continuous-wave fourwave mixing efficiency in nonlinear AlGaAs waveguides", Opt. Express 22, 26814 -- 26824 (2014). [15] C. Lacava, V. Pusino, P. Minzioni, M. Sorel, and I. Cristiani, "Nonlinear properties of AlGaAs waveguides in continuous wave operation regime", Opt. Express 22, 5291 -- 5298 (2014). [16] K. Dolgaleva, W. C. Ng, L. Qian, and J. S. Aitchison, "Compact highly-nonlinear AlGaAs waveguides for efficient wavelength conversion", Opt. Express 19, 12440 -- 12455 (2011). [17] D. T. H. Tan, K.Ikeda, P.C. Sun, and Y. Fainman, "Group velocity dispersion and self phase modulation in silicon nitride waveguides", Appl. Phys. Lett. 96, 061101 (2010). [18] K. Ikeda, R.E. Saperstein, N. Alic, and Y. Fainman, "Thermal and Kerr nonlinear properties of plasma-deposited silicon nitride/silicon dioxide waveguides", Opt. Express 16, 12987 -- 12994 (2008). [19] J. S. Levy, et al. "CMOS-compatible multiple-wavelength oscillator for on-chip optical interconnects", Nature Photon. 4, 37 -- 40 (2010). [20] J. P. Epping, et al. "High confinement, high yield Si3N4 waveguides for nonlinear optical applications", Opt. Express 23, 642 -- 648 (2015). [21] D. Duchesne, et al. "Supercontinuum generation in a high index doped silica glass spiral waveguide", Opt. Express 18, 923 -- 930 (2010). [22] D. J. Moss, R. morandotti, A. L. Gaeta and M. Lipson, "New CMOS-compatible platforms based on silicon nitride and Hydex for nonlinear optics", Nature Photon. 7, 597-607 (2013) [23] Z. Sun, A. Martinez, and F. Wang, "Optical modulators with 2D layered materials", Nature Photon. 10, 227 -- 238 (2016). [24] M. Liu, et al. "A graphene-based broadband optical modulator", Nature 474, 64 -- 67 (2011). [25] F. Xia, T. Mueller, T. Lin, A. Valdes-Garcia, and P. Avouris, "Ultrafast graphene photodetector", Nat. Nanotech. 4, 839 -- 843 (2009). [26] Q. Bao, et al. "Atomic-layer graphene as a saturable absorber for ultrafast pulsed lasers", Adv. Funct. Mater. 19, 3077 -- 3083 (2009). [27] A. Grigorenko, M. Polini, and K. Novoselov, "Graphene plasmonics", Nat. Photon. 6, 749 -- 758 (2012). [28] F. Bonaccorso, Z. Sun, T. Hasan, and A. Ferrari, "Graphene photonics and optoelectronics", Nat. Photon. 4, 611 -- 622 (2010). [29] X. Gan, et al. "Chip-integrated ultrafast graphene photodetector with high responsivity", Nat. Photon. 7, 883 -- 887 (2013). [30] A. K. Geim, and K.S. Novoselov, "The rise of graphene", Nature Mater. 6, 183 -- 191 (2007). [31] M. Breusing, C. Ropers, and T. Elsaesser, "Ultrafast carrier dynamics in graphite", Phys. Rev. Lett. 102, 086809 (2009). [32] T. Kampfrath, L. Perfetti, F. Schapper, C. Frischkorn, and M. Wolf, "Strongly coupled optical phonons in the ultrafast dynamics of the electronic energy and current relaxation in graphite", Phys. Rev. Lett. 95, 187403 (2005). [33] A. C. Ferrari, et al. "Raman spectrum of graphene and graphene layers", Phys. Rev. Lett. 97, 187401 (2006). [34] H. Li, Y. Anugrah, S. J. Koester, and M. Li, "Optical absorption in graphene integrated on silicon waveguides", Appl. Phys. Lett. 101, 111110 (2012). [35] E. Hendry, P. J. Hale, J. Moger, A. K. Savchenko, and S. A. Mikhailov, "Coherent Nonlinear Optical Response of Graphene", Phys. Rev. Lett. 105, 097401 (2010). [36] Faguang Yan, Lixia Zhao, Amalia Patanè, PingAn Hu, Xia Wei, Wengang Luo, Dong Zhang, Quanshan Lv, Qi Feng, Chao Shen, Kai Chang, Laurence Eaves and Kaiyou Wang, "Fast, multicolor photodetection with graphene-contacted p-GaSe/n-InSe van der Waals heterostructures", Nanotechnology 28, 27LT01(2017) [37] G. P. Agrawal, Nonlinear Fiber Optics (Academic, San Diego, 2001). [38] E. Dulkeith and Y. A. Vlasov, "Self-phase-modulation in submicron silicon-oninsulator photonic wires", Opt. Lett. 14, 5524-5534 (2006) [39] V. R. Almeida, C. A. Barrios, R. R. Panepucci, and M. Lipson, "All-optical control of light on a silicon chip", Nature 431, 1081-1084 (2004). [40] K. W. DeLong, K. B. Rochford, and G. I. Stegeman, Effect of two-photon absorption on all-optical guidedwave devices", Appl. Phys. Lett. 55, 1823-1825 (1989). [41] V. Mizrahi, K. W. DeLong, G. I. Stegeman, M. A. Saifi and M. J. Andrejco, "Two-photon absorption as a limitation to all-optical switching", Opt. Lett. 14, 1140-1142 (1989) [42] Thomas Vallaitis, Siegwart Bogatscher, Luca Alloatti, Pieter Dumon, Roel Baets, Michelle L. Scimeca, Ivan Biaggio, Franois Diederich, Christian Koos, Wolfgang Freude, and Juerg Leuthold, Optical properties of highly nonlinear silicon organic hybrid (SOH) waveguide geometries Opt. Lett. 14, 17357 (2009). [43] Samuel Serna, Vladyslav Vakarin, Joan-Manel Ramirez, Jacopo Frigerio, Andrea Ballabio, Xavier Le Roux, Laurent Vivien, Giovanni Isella, Eric Cassan, Nicolas Dubreuil and Delphine Marris-Morini, Nonlinear properties of Ge-rich Si1xGex materials with different concentrations, Sci. Rep. 7, 14692 (2017). [44] Jassem Safioui, Franois Leo, Bart Kuyken, Simon-Pierre Gorza, Shankar Kumar Selvaraja, Roel Baets, Philippe Emplit, Gunther Roelkens, and Serge Massar, Supercontinuum generation in hydrogenated amorphous silicon waveguides at telecommunication wavelengths, Opt. Lett. 22, 3089 (2014). [45] Karthik Narayanan and Stefan F. Preble, Optical nonlinearities in hydrogenated-amorphous silicon waveguides, Opt. Express 18, 8998 (2010).
1911.10250	1	1911	2019-11-22T21:03:38	100pTcm Sensitive MEMS Resonant Magnetometer from a Commercial Accelerometer	[ "physics.app-ph", "physics.bio-ph", "physics.ins-det" ]	Magnetic sensing is present in our everyday interactions with consumer electronics, and also demonstrates potential for measurement of extremely weak biomagnetic fields, such as those of the heart and brain. In this work, we leverage the many benefits of the micro-electromechanical systems (MEMS) devices to fabricate a small, low power, inexpensive sensor whose resolution is in the range of weak biomagnetic fields. The sensor works at room temperature, and is suitable for consumer electronics integration. At present, such biomagnetic fields can only be measured by expensive mechanisms such as optical pumping and superconducting quantum interference devices (SQUIDs). Thus, our sensor suggests the opening of a large phase space for medical and consumer applications. The prototype fabrication is achieved by assembling micro-objects, including a permanent micromagnet, onto a post-release commercial MEMS accelerometer. With this system, we demonstrate a room temperature MEMS magnetometer, whose design is only sensitive to gradient magnetic fields and is generally insensitive to the Earth's uniform field. In air, the sensor's response is linear with a resolution of 1.1 nT cm-1 and spans over 3 decades of dynamic range to 4.6 {\mu}T cm-1. In 1 mTorr vacuum with 20 dB magnetic shielding, the sensor achieved 100 pT cm-1 resolution at resonance. The theoretical floor of this design is 110 fT cm-1 Hz-1/2 with a resolution of 13 fT cm-1, thus these devices hold promise for both magnetocardiography (MCG) and magnetoencephalography (MEG) applications.	physics.app-ph	physics	100pT/cm Sensitive MEMS Resonant Magnetometer from a Commercial Josh Javor1*, Alexander Stange2, Corey Pollock1, Nicholas Fuhr2 and David J. Bishop1,2,3,4,5 Accelerometer 1. Department of Mechanical Engineering 2. Division of Materials Science and Engineering 3. Department of Electrical and Computer Engineering 4. Department of Physics 5. Department of Biomedical Engineering Boston University Boston, Massachusetts 02215 Abstract Magnetic sensing is present in our everyday interactions with consumer electronics, and also demonstrates potential for measurement of extremely weak biomagnetic fields, such as those of the heart and brain. In this work, we leverage the many benefits of the micro-electromechanical systems (MEMS) devices to fabricate a small, low power, inexpensive sensor whose resolution is in the range of weak biomagnetic fields. The sensor works at room temperature, and is suitable for consumer electronics integration. At present, such biomagnetic fields can only be measured by expensive mechanisms such as optical pumping and superconducting quantum interference devices (SQUIDs). Thus, our sensor suggests the opening of a large phase space for medical and consumer applications. The prototype fabrication is achieved by assembling micro-objects, including a permanent micromagnet, onto a post-release commercial MEMS accelerometer. With this system, we demonstrate a room temperature MEMS magnetometer, whose design is only sensitive to gradient magnetic fields and is generally insensitive to the Earth's uniform field. In air, the sensor's response is linear with a resolution of 1.1 nT cm-1 and spans over 3 decades of dynamic range to 4.6 µT cm-1. In 1 mTorr vacuum with 20 dB magnetic shielding, the sensor achieved 100 pT cm-1 resolution at resonance. The theoretical floor of this design is 110 fT cm-1 Hz-1/2 with a resolution of 13 fT cm-1, thus these devices hold promise for both magnetocardiography (MCG) and magnetoencephalography (MEG) applications. edge biomagnetic and large Introduction Magnetic sensing spans many scientific applications, from consumer electronics to cutting research. Smartphones utilize the Earth's magnetic field for navigation. Automobiles leverage non-contact magnetic sensing to determine position of components, such as in the crank shaft and braking systems. And, more recently, the highest resolution magnetic sensors have been used to measure biomagnetic fields of the brain and heart1,2. This range of applications is accomplished with various sensing mechanisms. Hall Effect sensors dominate the industrial market due to advantages in CMOS-compatible manufacturability power consumption, but are limited in resolution by the Earth's field at 50 µT (ref. 3). Fields emitted by the brain and heart begin at a million times smaller than Earth's field (approximately 100 pT for heart and 200 fT for brain) and detection must be accomplished by more sophisticated means1. Some of the first biomagnetic measurements were conducted by superconducting quantum interference devices (SQUIDs)4 but have since transitioned to optically pumped, spin-exchange relaxation free (SERF) magnetometers following the development of smaller, chip-scale sensors5. Most of these techniques, however, face the tremendous barrier of the Earth's field because they are measuring a uniform field, a magnetic field unchanging in position. For biomagnetic measurements, this requires heavily shielded rooms (typically 60 dB attenuation), averaging or triggering using EKG leads, and large costs (on order of $10k per sensor, QuSpin). To fully realize the clinical capabilities of biomagnetic sensing, arrays including many sensors are needed for biomagnetic mapping, enhancing the impact of cost6. In this work, we show that the marriage of a permanent micromagnet and a commercial accelerometer can accomplish both large range (1.1 nT cm-1 to 4.6 µT cm-1) and high resolution (100 pT cm-1) by directly sensing a gradient magnetic field, all with a total cost of goods less than $50. As the first sensor design of its kind, the theoretical floor is 110 fT cm-1 Hz-1/2, well within the range of biomagnetic field sensing7. Fig. 1 compares our sensor's experiment and theory to existing technology3,7,8. The key developments enabling the sensor discussed in this work are the highly engineered micro- electromechanical systems (MEMS) accelerometer and the permanent micromagnet. The capacitive accelerometer is the classic success story of the MEMS industry, fulfilling a need in the automotive market for sensitive, low-cost detection for airbag sensors9. Such lucrative applications have driven the development of MEMS accelerometers, reaching resolutions of 110 µg Hz-1/2 for the ADXL 203 Fig. 1 Existing Technology. Sensitivity of various magnetic sensors. Existing sensors measure uniform magnetic field and so are adjusted to measure a gradient field (difference between two sensors separated by 1 cm). E is Earth's gradient field and GMN is geomagnetic gradient noise. This work, an advance for MEMS technology (purple highlight), is shown at the bottom compared to all magnetic sensors3,7,8. The dark lines indicate measureable range while the dashed indicate theory. This work is the only intrinsically gradient field sensor. Fig. 2 Fabrication of Magnetometer. a Optical image of ADXL 203 accelerometer after hermetic lid removal. Octagonal proof-mass in the center is surrounded by integrated circuitry. b Colorized scanning electron microscope (SEM) image of top right quadrant of accelerometer sensor. The proof-mass (purple) is anchored through the springs (yellow) and position is sensed by interdigitated capacitive fingers (red). c Schematic of subassembly fabrication. The polysilicon plate (i) is designed in house and fabricated at a foundry. It is mechanically tethered by polysilicon springs (black). A microsphere is manipulated by a micropipette, dipped in UV glue, and placed in a corner of the plate (ii). This is repeated on all corners before the epoxy is cured with UV light (iii). The micropipette is used on one sphere to break the tethers and flip the assembly around to stand like legs under a table (iv -- v). A micromagnet is oriented, dipped in UV glue, and then cured in the center of the "table" (vi). d Fabrication on post-release MEMS. The subassembly is dipped lightly in UV glue and oriented above the ADXL 203 proof-mass before being lowered carefully until noise is sensed on the output, indicating contact. It is important that the subassembly is attached without interfering with the natural sensing mechanism of the ADXL 203. e Colorized SEM image of fully fabricated magnetometer (not the device used in this work). f Magnetometer assembled inside printed circuit board (PCB) antiparallel coils for experimental characterization. and 20 µg Hz-1/2 for the ADXL 354 at costs of $25 and $35 per sensor respectively10,11. At the same time, market demand from the hard disk drive industry and others has pushed development of rare- earth permanent magnets. High anisotropy, small size, high remanence, and a large variety of coatings for automotive, medical, and consumer products have led to diverse commercial availability12. At first look, the benefits of combining such technologies into a sensor are low power consumption, small size, and low cost. The greatest advantage, however, is that permanent magnets, when constrained from rotating, are only sensitive to forces from gradient fields. So while the Earth's field is large (50 µT), it varies only slightly across the Earth's surface, producing a much smaller gradient field at 500 fT cm-1 (ref. 13). Permanent magnets have been integrated into MEMS sensor design before, detecting deflection as a uniform field induces torque, much like in a compass3,12,13. The best of these achieved a resolution of 300 pT Hz-1/2 at 1 Hz with a large footprint (10 cm3) and electron tunneling feedback control14. Several other MEMS magnetic sensors have been designed based on the Lorentz force, achieving the best experimental resolution of 143 nT at 136 kHz (ref. 15,16). Results Magnetometer Fabrication from MEMS accelerometer Our sensor is a marriage of two recently matured technologies: the capacitive accelerometer and permanent micromagnets. The accelerometer is a sensing platform fit for adaptation to other measurands because it inherently senses the position of a movable polysilicon plate. Fig. 2a shows the ADXL 203 accelerometer with the hermetic lid removed revealing the silicon die underneath. The octagon in the center is the proof-mass, a quadrant of which is expanded in the Fig. 2b false-colored scanning electron microscope (SEM) image. The proof-mass (purple) is a polysilicon plate that can be mechanically coupled to a variety of microscale objects, functionalizing the device for other sensing applications17. A custom pick-and-place tool and procedure was developed for assembling of microscale objects on the proof-mass, illustrated in Fig. 2c and d (discussed further in Methods). A fully fabricated sensor is shown in Fig. 2e, an SEM image. A permanent magnet distorts the SEM image, so the device shown is for illustrative purposes only, where the magnet is completely demagnetized and the UV glue under the spheres overlaps some portions of the spring (compromising sensing ability). Similar fabrication techniques have been used to develop a MEMS Casimir force metrology platform17, a full hemisphere, tip-tilt micro-mirror18, and other MEMS at Nokia19. Fig. 2f shows the sensor fabricated within a custom, printed circuit board (PCB) coil for gradient field characterization. Electrostatic Characterization The experimental setup is illustrated in Fig. 3a. The sensor can be driven by two mechanisms: electrostatically (purple circuit) and magnetically (green circuit). Further detail is in Methods. The modified ADXL 203 is characterized by electrostatic actuation in Fig. 3bi and COMSOL simulation in Fig. 3b ii-v. Fig. 3b i displays results from a square wave frequency sweep (10 Hz to 3 kHz). The duty cycle is 20% in air and 0.02% in vacuum, maximized to achieve the strongest signal without overdriving at resonance. The root mean square magnetometer output is normalized so that the quality and relative peak magnitude can be compared. Two resonant peaks are shown near 500 Hz and 2.2 kHz and the quality factors (sharpness) of the peaks increase greatly from atmosphere to vacuum, as damping is decreased. The 500 Hz peak is expected as the resonant frequency will decrease from 5.5 kHz when mass is added (see Methods). The quality of the 500 Hz peak is 5 in atmosphere and 4000 in vacuum, demonstrating increased sensitivity at resonance when damping is reduced. Fig. 3b ii-v qualitatively illustrate the two modes using the COMSOL Eigenmodes tool. Materials and geometry are input to the model resulting in a calculation and visualization of mode frequency and deformation, respectively. A 3D computer automated design file is generated of the ADXL 203 proof-mass from an SEM image, where the thickness is measured to be 4 µm. The proof-mass is rigidly attached to the magnet-table subassembly, constrained by a roller in the XY plane, and anchored at four points in the center. For this input configuration, a translational mode at 600Hz and a torsional mode at 1.5kHz are found. Errors in the mode frequency are likely due to inaccuracy in model geometry and assumptions of material properties. Fig. 3b ii shows a full device view of deformation at the lower frequency mode and Fig. 3b iii shows the same mode, cropped and oriented so spring deformation in a quadrant of the proof-mass can be visualized (red is largest deformation, blue is least). The deformation is translational along the X-axis, the direction of magnetization. This is the type of deformation we would expect from a force imposed by a gradient magnetic field (see Methods, Eq. 5). Similarly, Fig. 3b iv-v show a torsional deformation at the higher frequency mode, where the assembly torques about Fig. 3 Experimental Setup and Electrostatic Characterization. a Magnetometer on PCB coil assembly is oriented upside down in a chamber with option to pull vacuum (yellow) and apply magnetic shielding (blue). Feedthroughs provide power to the sensor and PCB coil as well as sense the outputs of the functionalized ADXL 203. A pulse generator is used in combination with a built-in self-test (ST) functionality to electrostatically characterize the sensor mechanics in air and in vacuum (purple, 1). A waveform generator is used in combination with a precision current source and the PCB coil wired in antiparallel to magnetically characterize the sensor with gradient fields (green, 2). Both X (blue) and Y (red) outputs are filtered by lock-in amplifiers. b Characterization using electrostatic drive described in (a) and COMSOL Eigenmode simulation. Square wave frequency sweep (i) reveals two actuation modes and demonstrates increase in quality factor in vacuum. Full view of simulation deflection at first mode depicts translation (ii) and Quadrant view (iii) depicts spring deformation. Full view of simulation deflection at second mode depicts torsion (iv) and Quadrant view (iv) depicts spring and plate deformation. In all color maps, red is largest deformation and blue is least. c The simplified free body diagram of the translational mode (top view) resembles a damped harmonic oscillator. Applied force can be driven electrostatically or magnetically. Forces due to mechanical and magnetic noise are also shown (represented as Fn). the center of the x-y plane. This is the deformation we would expect from a uniform magnetic field in x-y plane, but not oriented along magnet's dipole axis (see Methods, Eq. 7). Since we are imposing a magnetic field with no uniform component and a constant gradient along the X-axis, we are primarily interested in the effect seen at the translational mode, and we can disregard the higher, torsional mode. Based on this electrostatic characterization, we can simplify our mechanistic understanding of the sensor to a one-dimensional, underdamped harmonic oscillator model (Methods), the free body diagram of which is illustrated in Fig. 3c. The collective mass of the subassembly and proof-mass are treated as a rigid body with mass, m. The four springs on the proof- mass are lumped into a single spring constant, k, and damping in air or vacuum modulate the constant, c. An applied force (electrostatic or magnetic) along the X-axis results in a displacement along the same axis. Forces from mechanical and magnetic noise are posited to limit the experimental resolution of the device (see Discussion). Magnetic Characterization The magnetometer's performance is dynamically characterized in three conditions: air (case 1), air with magnetic shielding (case 2), and vacuum with magnetic shielding (case 3). In all cases, the frequency of a gradient field sine wave, symmetric about zero, is swept as the magnetometer output signal is processed by a lock-in amplifier. The magnetometer output voltage is proportional to gradient magnetic field by Eq 9 (Methods). Fig. 4a shows results from case 1, where frequency is swept from 50 Hz to 1.1 kHz and field strength is swept from 4.6 µT cm-1 to 1.9 nT cm-1. Similar to electrostatic characterization, a low frequency peak is again present near 500 Hz, indicative of the translational mode and displacement along the X-axis. The largest applied field is 4.6 µT cm-1, as higher fields result in a clipped output signal at resonance by the ADXL 203 conditioning circuit. The sweeps follow a monotonic pattern, decreasing in signal output as the field magnitude is decreased. At lower field magnitudes, the signal to noise ratio visibly diminishes and is eventually overcome by noise. Fig. 4b shows results from case 3, where the applied field is swept in a narrower frequency range on the tip of the high quality peak (478.5 Hz to 480.5 Hz) and in a field range from 3.8 nT cm- 1 to 76.9 pT cm-1. Again, the largest applied field shown is 3.8 nT cm-1, above which output signals are clipped by the ADXL 203. The sweeps again follow a monotonic pattern corresponding to field magnitude. The resonance is approximately 479.2 Hz. Fig. 4c displays the magnetometer output at resonance with respect to applied gradient field, processed from the sweeps conducted in Fig. 4a and b. Here, data from sweeps below the experimental resolution of the sensor are included to characterize the experimental noise floor. Data from case 2 (green) is now included, and is largely similar to case 1, except with a lower resolution on the sensing axis. Data from case 1 and case 3 correspond in color to Fig. 4a and b, respectively. The characteristics from each case are also tabulated in Table 1. Circles represent the output from the X-axis (along the magnet's dipole axis), while diamonds are the sensor y-axis output. In all cases, the y-axis is also sensitive to the applied field, but is lower than 14% of the x-output in all cases, indicating good magnet alignment and reduced cross-axis sensitivity. The linear dynamic range of the magnetometer output in fT cm-1 is 3.3 decades in case 1 and 1.4 decades in case 3. The dotted lines show a linear least square fit of data above the experimental noise floor, where the sensitivity, 𝛾𝑚𝑎𝑔, is consistently linear and near 1 µVrms (fT cm-1)-1 in all cases. The black dashed line represents the theoretical noise floor, scaled from the ADXL 203 noise density with optimal lock-in filtering (Methods). The dotted lines are extrapolated to the theoretical floor to show the theoretical resolution in air (80 pT cm-1) and vacuum (40 fT cm-1). The dash-dot lines are a zero- order, least square fit of data below the experimental resolution, representing an experimental floor in each case. The intersection of the dotted line and dash-dot line are the experimental resolution of Fig. 4 Magnetic Sensing Performance. a Broad sine wave frequency sweep of gradient field in air (around translational mode). Sensor output decreases monotonically with the peak to peak of imposed gradient magnetic field. b Narrow sine wave frequency sweep of gradient field in vacuum (around translational mode on tip of high quality peak). Sensor output decreases monotonically with the peak to peak of gradient magnetic field. c Magnetometer output at peak of sweeps in (a) and (b) for air, shielded air, and shielded vacuum conditions. Signals along magnetic direction X-axis (circles) are over an order of magnitude higher than Y-axis (diamonds) indicating good magnet alignment and low cross-axis sensitivity. A linear least squares fit is conducted on data above the experimental floor to determine sensitivity (slope). All sensitivities are nearly 1 µ Vrms (fT cm-1)-1. Both axes in air reach a floor at the same sensor output, indicating a limitation of mechanical or electrical noise. Both axes in vacuum reach a floor at the same gradient magnetic field strength, indicating a limitation of gradient magnetic noise. In air with shielding, the X-axis reaches a lower sensor output floor than the Y-axis, demonstrating the attenuation of magnetic noise. Based on linear fits, the magnetometer's resolution is 100 pT/cm in vacuum and 1.1 nT cm-1 in air. Extrapolating the linear fit further to the theoretical floor, the theoretical resolution of this design configuration is 40 fT cm-1 in vacuum and 80 pT cm-1 in air, based on ADXL 203 noise density10 (see Methods). Table 1 Sensor Performance Metrics Based on Condition. Tabulated values are extracted from plot in Fig. 4c. the sensor (Table 1). It is most noteworthy that x- and y- outputs have the same experimental floor in case 1, and x- and y- outputs reach an experimental floor at the same magnetic field in case 3. In case 2, the x-output extends lower than both the y-output and case 1 data. These relationships are indications of resolution-limiting noise, expanded further upon in Discussion. The raw, unprocessed performance of the sensor in air is illustrated in Fig. 5, combining some of the performance metrics displayed in Table 1 (case 1). It is also important to highlight that data in this plot is not taken at resonance, where the signal to noise ratio is far more favorable. Rather, it is operating in a lower frequency regime where many common biomagnetic signals exist. An arbitrary waveform (black, dashed) resembling a magneto-cardiogram is imposed as a gradient field signal at the low frequency of 2.2 Hz. Within a period, the signal is composed of higher frequencies, mostly below 60 Hz. The blue data is the raw output from the magnetic axis and is shown to track the gradient signal very well with no distortion. The red data below the waveform shows the output of the y-axis, showing only mechanical noise and no features from the arbitrary waveform. The inset in the top left is an SEM of the magnetometer showing x- and y-axis direction, where the X-axis is the magnetized direction. Both axes are offset on the plot for ease of visualization. Biomagnetic signatures are typically in the hundreds of pT/cm, and the signal shown here is 20 µT cm-1 peak-to-peak (the smallest feature is a 250 nT cm-1 peak indicated at 0.7 s). While this is several orders of magnitude away, the Discussion expands on why this is a promising platform for such measurements in the future. Discussion Fabrication Error and Throughput In Fig. 2, we show that a commercially available accelerometer can be functionalized with a micromagnet using a custom pick-and-place procedure17,18. Previous prototypical works combining permanent magnets with MEMS structures did not employ commercially available platforms or such position-sensitive fabrication techniques3,14-16. While the fabrication method presented here is currently low-throughput and useful mainly for prototypical design, industrial scale fabrication techniques exist to accomplish similar tasks, such as pick-and-place or flip-chip-bonding20,21, and could be used to fabricate the sensor pre-release. It is also noteworthy that several asymmetries result from such a manual fabrication process that could limit the resolution of the sensor. Some of these include a displaced center of mass (anisotropy of cube magnet geometry, centering of magnet on table, table on proof-mass), variable sphere size and area of contact, magnet orientation, magnetization direction, and contamination of sensor from opening the sealed package. With the added weight of the micromagnet, such asymmetries may manifest themselves by pulling the proof- mass out of plane with respect to the capacitive fingers, which are designed to detect displacement in-plane only. Any rotational assembly error or uniform field (such as the Earth's field) may create an offset torque of the proof-mass, creating greater asymmetry in the springs and mechanical motion. Altogether, we show that the fabrication technique has minimized these errors and suggest that existing large-scale systems could reduce them even further. Exclusive Sensitivity to Gradient Magnetic Fields The mechanical modes of our magnetometer are characterized by electrostatic actuation in Fig. 3b i and by mode simulation in Fig. 3b ii-v. We argue that the fabricated sensor is only sensitive to gradient magnetic fields, which impose a force along the dipole axis of the magnet and result in a translational deformation (Methods). Sensitivity to uniform fields would result in a torque of the magnet. This is supported by the simulated mode deformation in Fig. 3b ii, where we show that the fundamental mode is a translational deformation along the dipole axis of the magnet. Thus any information from this mode will only be from gradient magnetic fields. We also explain the prediction of the 500 Hz fundamental mode from a simplified free body diagram in Fig. 3c and Eq 3 (Methods). Furthermore, we show there is a separate, higher mode of torsional deformation (Fig. 3 iv) about the center of the magnet that would correspond to a uniform field. This shows that information at the fundamental mode is not directly affected by uniform fields. The magnetic characterization of the device, then, is centered around the fundamental mode. Resolution-Limiting Noise The magnetic characterization in Fig 3c and Table 1 reveal differing resolutions in each case and on each output axis of the fabricated sensor. Analyzed together, the experimental noise floors are suggestive of the type of noise that limits the resolution. In air (case 1), the experimental floors of both x- and y- outputs coincide at the same sensor voltage, indicating that both axes are subject to a common noise. Since the noise floor is independent of magnet orientation, it cannot be due to gradient Fig. 5 Magnetic Sensing Demonstration. Raw output of magnetic sensor in air in response to an arbitrary waveform resembling an electrocardiogram (EKG) at 2.2 Hz and 20 µT cm-1 peak-to-peak, imposed along the X- axis by a PCB coil. The sensor output is displayed both in voltage and gradient magnetic field. Both X and Y outputs are offset for ease of visualization. The X output tracks the imposed field very well, while the Y output does not resolve any of the features in the magnetic signal. The inset (top left) illustrates the magnet alignment with respect to the sensitive X-axis and the insensitive Y-axis. magnetic field noise, which would predominantly actuate the magnet's central axis. Mechanical noise may result from the experimental set up or air pressure fluctuations in the ambient environment (as the sensor's lid is removed). Any asymmetries from fabrication discussed earlier could result in common electrical noise on the output from out-of-plane capacitive fingers. The directionality detection of the ADXL 203 is designed to modulate each axis differently, and out-of-plane deformation could enhance cross-talk of these signals. Furthermore, the device is unshielded in case 1 and an offset torque from the Earth's uniform field could add to the asymmetry. When shielded in case 2, the X-axis reaches a lower resolution (from 1.1 nT cm-1 in air to 700 pT cm-1in shield), but the y-axis remains the same. The effect of the shielding may be damping of mechanical noise (stabilizing the motion of the magnet) or attenuating any offset torque from the Earth's field that could be adding to asymmetry. It is unlikely that uniform geomagnetic noise plays a role here as geomagnetic noise is far lower at 100 pT (ref. 3). Results from the shielded vacuum condition (case 3) reveal a common experimental floor of both sensor axes at the same applied field, rather than the same sensor output voltage. This indicates that the limiting noise source is different than that in case 1 and case 2. The y-axis motion was still able to be detected at much lower displacements than the X-axis, suggesting that the limiting noise must be a gradient magnetic noise. Possible sources are either geomagnetic or the gradient coil driving system. Geomagnetic gradient noise has been reported to be much lower at 500 fT cm-1 (ref. 13), however measurements were conducted during daytime hours in a major city. Therefore, gradient noise from the environment may be larger than this report, but are likely not the limiting noise. The voltage controlled current source (CS580) is specified to have superb output noise characteristics (60 fA Hz-1/2 in the configuration at resolution measurement). However, the instrument is some distance away from the PCB coils, the wire is carried next to all other sensor leads with relatively small drive currents (<100nA drive, 11mA power), and connections are made at vacuum chamber feedthroughs. This may make the drive signal vulnerable to pick-up or cross talk, which is amplified and superimposed onto the magnetic driving force, thus limiting the sensor resolution. Current and Future Applications The sensor is characterized in three conditions (Table 1) to demonstrate its performance in the context of various applications. Most applications exist in ambient conditions and don't require the enhanced performance that vacuum and shielding provide. For example, dipole sources, such as planetary magnetic fields or ferromagnetic objects, have gradient signatures that are difficult to measure with a uniform field sensor alone. Among these applications, our magnetometer offers the key advantage of directly sensing the gradient field, rather than the difference between two uniform field sensors. Moreover, it is capable of doing all this with a small size, low power, low cost, and at room temperature. Finally, the realization of our theoretical resolution in air (80 pT cm-1) would offer the unique ability to sense biomagnetic fields in ambient conditions, an idea that is attractive for wearable sensing (such as signals illustrated in Fig. 5). For the most sensitive applications, vacuum or shielding can be applied. Vacuum can be pulled on the resonant sensor as shown in Fig. 3b i and 4b to increase the quality and resolution of the fundamental mode. At scale, vacuum packaging is a solved problem for MEMS packaging21, which allows for the benefits of enhanced resolution at a small scale. Resonant mode operation is typically a design tradeoff, limiting a sensor to a specific, narrow band of frequencies around resonance. However, a resonant sensing mechanism does not necessarily impede a resonant sensor from identifying features at other frequencies of interest. MEMS actuators with flux guides have been designed to modulate an arbitrary signal so it can be measured at the resonant frequency of a sensor22. Others have leveraged a nonlinear spring stiffness during cyclic resonant motion, ultimately reporting a shift in resonant frequency instead of oscillation amplitude23. Shielding becomes useful when the target of measurement can also fit inside the shield. For this, shielded rooms with 60 dB attenuation are common and often used for biogmagnetic measurements. We report the effect of just 20 dB magnetic attenuation, which demonstrates the potential for enhanced resolution with greater shielding. Finally, the combination of improved vacuum, shielding, and environment could realize a theoretical resolution (13 fT cm-1) directly on par with SQUIDs and optically pumped, atomic magnetometers. Such a sensor is disruptive in cost, size, and its gradient sensing mechanism, transforming the approach to the most sensitive applications such as biomagnetic fields. Materials and Methods MEMS Accelerometer Fig 2b shows an SEM image of a quadrant of the ADXL 203 from Analog Devices. The spring (yellow) is nearly symmetric on both X and Y axes. Displacement is sensed via capacitive fingers (red) in a differential configuration. We chose the ADXL 203 for its intrinsically low noise density (110 µG Hz-1/2), linearized sensitivity (1 V G-1), wide range (up to 104 G with optimized filtering), and accessible proof-mass10. From experimental observation, the two-axis accelerometer was found to have a resonant frequency of 5.5 kHz, a spring constant of 1 N m-1, a Q in air of 10, and Q in vacuum of 10,000. The maximum sensing range in one direction is 25 nm, giving a sensitivity of 10 nm V-1. This means the device has a noise density of 1 pm Hz-1/2, or 1 pN Hz-1/2. Permanent Micromagnets Permanent micromagnets are a sintered mix of rare-earth element powders and are typically coated for protection or passivation12. The micromagnets in this work are cubes and magnetized to N52 grade (SM Magnetics Co.). The smallest commercially available cube magnet of 250 µm side length is chosen to minimize gravitational forces. The powders consist of neodymium, iron, and boron. The standard coating of nickel, copper, and nickel is used to avoid degradation. Magnetometer Fabrication The magnetometer is fabricated in two stages. First, a subassembly is made that resembles a magnet on a table (Fig. 2c). Second, the subassembly is attached to the post-release MEMS (Fig. 2d). The separate subassembly comprises spheres, a polysilicon plate, and a micromagnet. A custom pick-and- place system is used to manipulate and assemble these micro-objects. Vacuum (typically -2psi) is pulled on a glass pipette with the orifice in contact with the object of interest. A micromanipulator on a probe station (Cascade Microtech EPS150FA) and a 3D printed part are used to direct the motion of the pipette in 3 dimensions. A straight pipette (WPI, 30 µm aperture) is used to manipulate microspheres and a custom, 45° angled pipette (Clunbury Scientific, 135 µm aperture) is used with the cube magnet. Borosilicate glass microspheres (Cospheric) of about 65 µm are used as they are sufficiently larger than the pipette, but still small enough to minimize contact with the MEMS proof- mass later. The plate is designed in-house and manufactured by the MEMSCAP foundry process, PolyMUMPs (Fig. 2c i). Mechanical tethers of polysilicon are attached to both the plate and the silicon handle so that the plate is suspended when a sacrificial layer of oxide is removed underneath by etching with hydrofluoric acid. Spheres are assembled on the plate to form a "table," which the magnet sits on. The spheres minimize contact surface area between the micro-objects and the sensitive proof-mass, allowing for repeatable, robust assembly. To assemble a sphere on the plate, vacuum is pulled while in contact with a sphere, which is then wetted on the bottom side with UV glue (Norland Adhesives, NOA 81). The sphere is then positioned above and lowered onto a corner of a plate (Fig 2c ii), where it is partially cured by UV light at the manufacturer recommended wavelength of 365nm (Dymax BlueWave) for 15 seconds (Fig 2c iii). This is repeated until all four corners of the plate contain spheres. Next, the vacuum pipette is put in contact with one sphere to form a "ball in socket" joint. The pipette is lifted to break the mechanical tethers holding the plate (Fig 2c iv). The newly assembled "table" is turned to sit on its legs (Fig 2c v). Next, a micromagnet is oriented on a vertical glass slide by an external magnet some distance behind the slide. The larger, angled pipette is brought into contact with the top face of the magnet, which is not one of the poles. The external magnet is removed, leaving magnet on the end of the pipette, held by vacuum. The bottom of the magnet is then dipped in UV glue, aligned on the standing table, and radiated with UV light. Now that the subassembly is a single rigid structure, the large, angled pipette can be used to attach it to the post-release MEMS. The sensor lid is removed and power is supplied so that the noise on the X and Y outputs can be monitored. The noise is very low normally, but spikes when contact is made by the pick-and-place system, presumably from vibrations in the pipette (Fig. 2d inset). The subassembly structure is lifted up and aligned over the center of the proof-mass under a brightfield microscope. The spheres are in predefined locations on the plate so that they make contact with a strip of the proof-mass between the capacitive fingers and the spring. The spheres on the bottom of the plate are dipped very lightly in UV glue, before the structure is carefully lowered toward the proof- mass (Fig. 2d). A spike in the accelerometer output signal is used to detect contact between the subassembly and the proof-mass (Fig. 2d inset). The spheres minimize contact with the proof-mass so that epoxy does not wick through the release holes patterned on the proof-mass, in which case the device would be rendered insensitive. The structure is then radiated with UV light for 15 seconds before cutting vacuum to the pipette and lifting off. The entire structure is then baked upside-down (to avoid unintended gluing) at 60C overnight to form a full cure (below manufacturer-recommended maximum operating temperature of magnet). Experimental Setup and Measurement The experimental setup comprises a custom PCB coil, custom vacuum chamber, MuMetal shield (Magnetic Shields Corp.), and an instrument drive system (Fig. 3a). The ADXL 203 is surface- mounted on a custom printed circuit board (PCB), on which a PCB coil for magnetic characterization is also attached (Fig. 2f). The PCB coil consists of two layers separated by the 1.6 mm thick PCB. The top copper traces can be seen in the image and the bottom traces are only different where the end of each line connects it to the next winding in sequence. The 0.1 mm vias, spaced 1 mm apart, connect the two layers to form a 10 turn coil on either side of the sensor. The PCB coil is rigidly connected to the PCB board to reduce mechanical noise in the output. The coil pair central axis is aligned with the micromagnet's dipole axis (along the accelerometer X-axis) for magnetic drive. Force on the magnet is proportional to the gradient of the magnetic field. By wiring the PCB coil in antiparallel fashion, the resulting magnetic field has a constant slope relative to position across the sensor, and thus a gradient magnetic field. The uniform field, then, is zero at the center of the coils, where the micromagnet is positioned. The sensor-coil assembly is fit into a dual inline pin (DIP) socket within a vacuum chamber built using standard parts (Kurt J. Lesker). For experimental results, the chamber is either at atmospheric pressure or in vacuum (1 mTorr). The sensor is positioned upside down to keep the proof-mass free from contacting the substrate underneath in the event of off-axis fields. The chamber is held between vibration isolating pads on a two-axis vertical stage, allowing for the chamber to be moved in and out of a MuMetal shield, which attenuates 20 dB of imposing fields (Magnetic Shields Corp). This entire assembly is built on a passive hydraulic vibration isolation table. There are two driving schemes of the magnetic sensor shown in Fig. 3a drive schematic: Electrostatic actuation and Magnetic actuation. Electrostatic actuation (circuit 1, purple) leverages the capacitively driven self-test functionality of the accelerometer. Originally designed to test whether the accelerometer is in normal working condition, this function may also be used to drive the sensor to arbitrary positions in one quadrant of the actuation range using pulse width modulation (PWM)24. In this work, self-test is used to non-magnetically actuate the magnetometer over a range of frequencies, characterizing the mechanics of the device after micro-objects are attached. The self-test pin is driven by a precision pulse generator (SRS DG645), as short duration pulses (<200 ns) are required in vacuum to avoid over-driving the MEMS at resonance. The pulse is 0 to 3 V and duty cycle is 20% in air and 0.02% in vacuum. Magnetic actuation (circuit 2, green) is achieved by driving low noise, small (>100 fA) currents through the PCB coil using a voltage-controlled current source (SRS CS580) and waveform generator (Agilent 33210A). The PCB coil's gradient field is linear with drive current (760 µA (µT cm-1)-1) and unchanging (< 3%) with frequency in the actuation range (DC to 1 kHz). In both drive schemes, the sensor output is only filtered using a lock-in amplifier (SR830). The reference signal is a 50% duty cycle square wave from either the pulse generator or the waveform generator. The equivalent noise bandwidth (ENBW) when using a 24 dB oct-1 roll-off and 300 ms time constant is equal to 0.26 Hz. The minimum ENBW for the most sensitive measurements is 0.008 Hz. Resolution at a given frequency is calculated from noise density, 𝜌, by Eq 1 (ref. 25), and is elaborated on further in Eq 10. 𝑅𝑒𝑠𝑜𝑙𝑢𝑡𝑖𝑜𝑛 = 𝜌 ∗ √𝐸𝑁𝐵𝑊 ∗ 1.6 (1) Measurement Theory Governing Mechanics A simplified, one-dimensional mechanical model is shown in Fig. 3c. The magnetometer behaves as an underdamped harmonic oscillator (Eq 2), with settling times of 20 ms in air and 3 s in vacuum (within 2% of final value). 𝑚𝑑 + 𝑐𝑑 + 𝑘𝑑 = 𝐹𝑎𝑝𝑝𝑙𝑖𝑒𝑑(𝑡) + 𝐹𝑛,𝑚𝑒𝑐ℎ(𝑡) + 𝐹𝑛,𝑚𝑎𝑔(𝑡) (2) The system may be thought of as one-dimensional because the ADXL 203 has low cross-axis sensitivity (1.5%) between X and Y10 and out-of-plane forces are minimized by centering the magnet on the proof-mass using the table subassembly. The magnet, table, and proof-mass are all considered one rigid body. Together, their mass is found to be 160 µg from the relationship between resonant frequency and mass shown in Eq 3. The resonant frequency is found from an electrostatic frequency sweep, shown in Fig. 3bi. This explains the effect of a decreased resonant frequency when mass is added. The spring constant, k, is estimated to be near 1 N m-1 from SEM images of the spring and a COMSOL model assuming polysilicon material. 𝑚 = 𝑘 (2𝜋𝑓0)2 (3) The constant, c, in Eq 2 represents damping. Vacuum decreases damping and is shown to increase the quality factor in Fig. 3b i. At resonance, this increases the amplitude of oscillation, 𝑑𝑓0. For a constant force at resonance, 𝐹𝑓0, the amplitude increases proportional to the quality factor, 𝑄 as shown in Eq 4 (ref.9). 𝑑𝑓0 = 𝑄 𝐹𝑓0 𝑚 (4) Forcing and Magnetics The applied force, 𝐹𝑎𝑝𝑝𝑙𝑖𝑒𝑑, is proportional to the gradient field26 generated by the PCB coil (Eq 5) as described in the previous section. The micromagnet has a moment, 𝑀⃗⃗ , of 15 µJ T-1 calculated from experimental data and confirmed by simulation. The permanent magnet is approximated as a dipole in Eq 6 and data is fit by cubic function26. Experimental magnetic field, B, is gathered from a hall sensor along the central axis of the magnet, r. The constants, including magnetic permeability of free space, M0, are condensed to 𝛼, and the moment of the magnet, M, is extracted. This is confirmed by a simulation using Finite Element Methods Magnetic (FEMM) software and a 250 µm cylindrical magnet of N52 grade. Again magnetic field, B, is collected at various distances along the central axis, r, and fit by cubic function to extract the moment, M. 𝐹𝑎𝑝𝑝𝑙𝑖𝑒𝑑(𝑡) = 𝑀 ∙ ∇𝐵(𝑡) 𝐵(𝑟) = 𝑟3 , 𝑟3 = 𝛼𝑀 2𝑀0𝑀 1 4𝜋 (5) (6) The main sources of noise are mechanical, Fn,mech, and gradient magnetic, Fn,mag, but these are insignificant for large fields and so are analyzed in Discussion as resolution-limiting terms. The effect of gravity is ignored since the magnetometer is held upside down and any gravitational forces would be outside of the sensing plane. This simplified model is only relevant to the fundamental mode in Fig 3b ii, as the mode described in Fig. 3b iv is deforming in two dimensions. This torsional deformation could be actuated by a uniform magnetic field, much like a compass. The relationship between a uniform field, B, and the magnet with moment, M, would be a torque, T, as is shown in Eq 7 (ref. 26). The ADXL 203 is designed to sense motion in either the x or y directions, however. And so in this case, the sensor would not have a meaningful output. (7) 𝑇 = 𝑀 × 𝐵 Sensor Transduction The ADXL 203 directly measures a differential capacitance, which is inversely proportional to a displacement of the proof-mass (in either X or Y directions, see Fig. 2b). The sensor is linear in the measurement range and so its signal output, 𝑆, is related to displacement, 𝑑, by a proportionality constant, 𝛾, in Eq 8. In the linear regime of the springs, Hooke's Law relates displacement to a force. Eq 5 earlier shows that ∇𝐵 is proportional to a force, 𝐹𝑎𝑝𝑝𝑙𝑖𝑒𝑑. Once functionalized with a micromagnet, the sensor output, now 𝑆𝑚𝑎𝑔, is then proportional to the applied ∇𝐵 by the sensitivity, 𝛾𝑚𝑎𝑔 (Eq 9). The sensitivity of the magnetometer is 1 µV (fT cm-1)-1 from experimental measurement in Fig 4c and tabulated in Table 1. 𝑆 = 𝛾𝑑 (8) 𝑆𝑚𝑎𝑔 = 𝛾𝑚𝑎𝑔∇𝐵 (9) Noise and Theoretical Resolution Using the understanding of mechanics and effect of magnetic fields on the sensor output, the noise floor can be calculated. The theoretical resolution is assumed to be limited by the noise floor of the sensor. The ADXL data sheet reports a noise density, 𝜌, of 110 µV Hz-1/2 (ref. 10). This accelerometer noise density can be scaled by the magnetic sensitivity, 𝛾𝑚𝑎𝑔, in Eq 10 to find magnetic noise density, 𝜌𝑚𝑎𝑔. This yields a magnetic noise density of 110 fT cm-1 Hz-1/2. Finally, Eq 1 earlier related noise density to sensor resolution when using a lock-in amplifier. This experimental setup, then, would reach a theoretical best measurement of 13 fT cm-1 in 1 mTorr vacuum and at room temperature. 𝜌𝑚𝑎𝑔 = 𝛾𝑚𝑎𝑔𝜌 (10) Conclusion The purpose of this work was threefold: 1) to build a magnetic sensor that is only sensitive to gradient magnetic fields, 2) to demonstrate the wide field and frequency space of this new class of magnetic sensors, and 3) to achieve 1 and 2 in a small, low-cost, and commercially available platform. We have demonstrated the performance of a gradient magnetometer by achieving a resolution of 100 pT cm-1 in shielded vacuum, and a range spanning over 3 decades in ambient conditions (1.1 nT cm-1 to 4.6 µT cm-1). Compared to existing designs of magnetic MEMS resonant sensors, our resolution surpasses the best found in literature14 by a factor of three experimentally and by over a factor of fifteen theoretically (110 fT cm-1 Hz-1/2). More sensitive accelerometers, such as the ADXL 354, would theoretically be able to improve this sensitivity by a factor of ten11. We have achieved all of this on a small, versatile platform, which is easily integratable into consumer technology integration. This new technology has potential to revolutionize magnetic sensing while also offering many advantages to other fields such as navigation, communication, and biomagnetic field mapping. Acknowledgements We would like to thank Pablo del Corro, Lawrence Barret, and Jeremy Reeves for helpful consultations regarding state-of-the-art magnetometry, theory, and fabrication. We would also like to thank Professor Anna Swan for the customization of her probe station for sensitive fabrication. This work was supported by the NSF CELL-MET ERC award no. 1647837 and a SONY Faculty Innovation Award. Conflict of Interests The authors declare that they have no conflict of interest. Author Contributions The device and experiments were conceived by D.J.B. and J.J. The fabrication was done by J.J. with assistance from A.S. All data were collected by J.J. and interpreted and analyzed by J.J., A.S., C.P., N.F., and D.J.B. The manuscript was written by J.J. with input from A.S. and edited by all authors. References 1. Shah, Vishal K., and Ronald T. Wakai. "A compact, high performance atomic magnetometer for biomedical applications." Physics in Medicine & Biology 58.22 (2013): 8153. 2. Shrivastava, Peeyush, et al. "Magnetocardiography-derived 3D Current Density Maps Show Increased Sensitivity and Localization of Ischemic Damage in High Risk Unstable Angina Patients." Circulation 134.suppl_1 (2016): A17076-A17076. 3. Lenz, James, and S. Edelstein. "Magnetic sensors and their applications." IEEE Sensors journal 6.3 (2006): 631-649. 4. Cohen, David. "Magnetoencephalography: detection of the brain's electrical activity with a superconducting magnetometer." Science 175.4022 (1972): 664-666. 5. Schwindt, Peter DD, et al. "Chip-scale atomic magnetometer." Applied Physics Letters 85.26 (2004): 6409-6411. 6. Malmivuo, Jaakko. "Biomagnetism." Wiley Encyclopedia of Electrical and Electronics Engineering (1999): 1-25. 7. Robbes, Didier. "Highly sensitive magnetometers -- a review." Sensors and Actuators A: Physical 129.1-2 (2006): 86-93. 8. Bahreyni, Behraad, and Cyrus Shafai. "A resonant micromachined magnetic field sensor." IEEE Sensors Journal 7.9 (2007): 1326-1334. 9. Liu, Chang. Foundations of MEMS. Pearson Education India, 2012. 10. Data Sheet ADXL103/ADXL203. Tech. Rep., Analog Devices (2019). URL http://www.analog.com/media/en/technical-documentation/data-sheets/adxl103_203.pdf 11. Data Sheet ADXL354/ADXL355. Tech. Rep., Analog Devices (2019). URL https://www.analog.com/media/en/technical-documentation/data-sheets/adxl354_355.pdf 12. Fischbacher, Johann, et al. "Micromagnetics of rare-earth efficient permanent magnets." Journal of Physics D: Applied Physics 51.19 (2018): 193002. 13. Sheng, Dong, et al. "A microfabricated optically-pumped magnetic gradiometer." Applied physics letters 110.3 (2017): 031106. 14. DiLella, D., et al. "A micromachined magnetic-field sensor based on an electron tunneling displacement transducer." Sensors and Actuators A: Physical 86.1-2 (2000): 8-20. 15. Yang, Henry H., et al. "Ferromagnetic micromechanical magnetometer." Sensors and Actuators A: Physical 97 (2002): 88-97. 16. Herrera-May, Agustín, et al. "Resonant magnetic field sensors based on MEMS technology." Sensors 9.10 (2009): 7785-7813. 17. Stange, Alexander, et al. "Building a Casimir metrology platform with a commercial MEMS sensor." Microsystems & nanoengineering 5.1 (2019): 14. 18. Pollock, C., et al. "Extreme angle, tip-tilt MEMS micromirror enabling full hemispheric, quasi-static optical coverage." Optics express 27.11 (2019): 15318-15326. 19. Aksyuk, Vladimir A., et al. "Method for forming micron-sized and smaller liquid droplets." U.S. Patent No. 5,961,767. 5 Oct. 1999. 20. Marenco, N., et al. "Vacuum encapsulation of resonant MEMS sensors by direct chip-to- wafer stacking on ASIC." 2008 10th Electronics Packaging Technology Conference. IEEE, 2008. 21. Fischer, Andreas C., et al. "Integrating MEMS and ICs." Microsystems & Nanoengineering 1 (2015): 15005. 22. Edelstein, A. S., et al. "Progress toward a thousandfold reduction in 1∕ f noise in magnetic sensors using an ac microelectromechanical system flux concentrator." Journal of applied physics 99.8 (2006): 08B317. 23. Seshia, Ashwin A., et al. "A vacuum packaged surface micromachined resonant accelerometer." Journal of Microelectromechanical systems 11.6 (2002): 784-793. 24. Pollock, Corey, et al. "Engineered PWM drives for achieving rapid step and settle times for MEMS actuation." Journal of Microelectromechanical Systems 27.3 (2018): 513-520. 25. Data Sheet Model SR830. Tech. Rep., Stanford Research Systems (2019). URL https://www.thinksrs.com/downloads/pdfs/manuals/SR830m.pdf 26. Jackson, John David. "Classical electrodynamics." (1999): 841-842.
1704.07226	1	1704	2017-04-24T13:47:43	Tunable, synchronized frequency down-conversion in magnetic lattices with defects	[ "physics.app-ph" ]	We study frequency conversion in nonlinear mechanical lattices, focusing on a chain of magnets as a model system. We show that by inserting mass defects at suitable locations, we can introduce localized vibrational modes that nonlinearly couple to extended lattice modes. The nonlinear interaction introduces an energy transfer from the high-frequency localized modes to a low-frequency extended mode. This system is capable of autonomously converting energy between highly tunable input and output frequencies, which need not be related by integer harmonic or subharmonic ratios. It is also capable of obtaining energy from multiple sources at different frequencies with a tunable output phase, due to the defect synchronization provided by the extended mode. Our lattice is a purely mechanical analog of an opto-mechanical system, where the localized modes play the role of the electromagnetic field, and the extended mode plays the role of the mechanical degree of freedom.	physics.app-ph	physics	Tunable, synchronized frequency down-conversion in magnetic lattices with defects Marc Serra-Garcia1, Miguel Moleron1 and Chiara Daraio2 1Department of Mechanical and Process Engineering, Swiss Federal Institute of Technology (ETH), Zürich, Switzerland 2Engineering and Applied Science, California Institute of Technology, Pasadena, California 91125, USA email: [email protected] 1 Abstract - We study frequency conversion in nonlinear mechanical lattices, focusing on a chain of magnets as a model system. We show that by inserting mass defects at suitable locations, we can introduce localized vibrational modes that nonlinearly couple to extended lattice modes. The nonlinear interaction introduces an energy transfer from the high- frequency localized modes to a low-frequency extended mode. This system is capable of autonomously converting energy between highly tunable input and output frequencies, which need not be related by integer harmonic or subharmonic ratios. It is also capable of obtaining energy from multiple sources at different frequencies with a tunable output phase, due to the defect synchronization provided by the extended mode. Our lattice is a purely mechanical analog of an opto-mechanical system, where the localized modes play the role of the electromagnetic field, and the extended mode plays the role of the mechanical degree of freedom. Introduction - Frequency converting processes have applications in a variety of problems, for example, in obtaining different wavelengths from a fixed-frequency laser1, harvesting energy from vibration sources2 and generating entangled photons3. Typically, frequency conversion is accomplished through wave mixing4 (which requires at least two input signals with a prescribed frequency difference), harmonic generation1 (which produces an output that is a multiple of the input signal) and subharmonic bifurcations5 (which produce an output that is an integer fraction of the original signal). In addition, these frequency conversion mechanisms prescribe the output's signal phase, which hinders the process of harvesting energy from multiple sources. Combination resonances6, processes that arise in the presence of multiple nonlinearly-interacting modes, can achieve frequency down-conversion between arbitrary input and output signals not related by a harmonic or subharmonic ratios. The resulting input and output frequencies can be tuned by adjusting the modes' frequencies. Combinational resonances can be found, for example, in vibrating beams7, membranes and plates8. In this paper, we show that nonlinear lattices have the potential to act as frequency- converting devices, due to the combination resonances arising from the nonlinear interaction between the lattice's normal modes. Chains of nonlinearly interacting elements have been studied for decades, beginning in the FPU problem9,10 . They present a wide variety of trapping14, breathers15,16, phenomena, unidirectional wave propagation17, negative or extreme stiffness18, localized modes with tunable profile19, shocks and rarefaction waves20. These phenomena can be used to realize acoustic rectifiers17, logic gates21, lenses22, filters23, vibration-attenuation24 and energy harvesting systems16. Nonlinear lattices can be implemented using a broad range of materials25, geometries26 and interactions27, allowing to tune the masses, coupling strengths and damping values of the particles, to optimize the performance under the required operating conditions. Because of this tunability, nonlinear metamaterials are a promising candidate for energy converting devices. including solitons11,12, band-gaps13, energy 2 Figure 1. (Color Online) Frequency-converting metamaterial concept. (a) Metamaterial design, consisting of a chain of nonlinearly-interacting magnets. The central particle of the chain is a defect, which has a lower mass. This magnet acts as the high-frequency input to the system. The down- converted energy can be extracted far away from the defect. In our experiments, the defect is driven by a wire carrying an electrical current (yellow arrow). (b) Cropped image of the experimental magnet chain, obtained using the same computer vision camera that is also used to track the magnets. Each magnet is enclosed in a 3D printed case, and has a random speckle pattern to facilitate its tracking by digital image correlation. (c) Extended mode of vibration. The red hollow circle is the defect particle, while the blue solid dots represent the other particles. (d) Experimental frequency response of the extended mode (blue dots) and Lorentzian fit (red solid line). (e) Localized mode of vibration. (d) Experimental frequency response of the localized mode (blue dots) and Lorentzian fit (red solid line). Experimental system – Our experimental setup consists of a chain of magnets27 floating on an air table (Fig. 1(a)). Each magnet is embedded in a 3D printed case that adds an additional mass, with different case designs resulting in different particle masses (!=0.45' for the non-defect particles, !()=0.197' for the first defect and !()=0.244' for the second defect). The presence of defects introduces localized modes around each defect particle (Fig 1(b,c)). When these modes are excited, the resulting motion is exponentially localized around the defect. In our experimental setup, the defects act as inputs for the frequency-conversion system. We excite them by passing current through a small conductive wire normal to the length of the chain (Fig. 1(a)). The wire is driven harmonically with a signal generator (Agilent 33220A) amplified by an audio amplifier (Topping TP22, class D). An extended mode of vibration (Figs. 1(c) and (d)) interacts with the localized mode to introduce frequency conversion. We monitor the motion of the magnets using a computer vision camera (Point Grey GS3-U3-41C6C-C), with a frame rate between 40 and 200 fps that allows us to resolve all particles' motion. We use the software VIC-2D from Correlated Solutions, to track the particles and determine their trajectory. 3 Experimental results for the system with a single defect – We start by studying a lattice of 21 magnets containing a singe defect (!()=0.197') in the middle position (i=11). The first exciting a combinational resonance (./+.1) between the extended and localized modes 6. parameters and is not an absolute limit. and last magnets are fixed. We set the excitation frequency to approximately the sum of the defect's frequency (Fig. 1(f)) and the extended mode's frequency (Fig. 1(d)), with the goal of We slowly increase the excitation amplitude until a threshold is reached and self-sustaining oscillations develop far away from the defect, indicating the transfer of energy between the localized mode and an extended mode (Fig. 2(a)). In this regime, the defect motion is modulated by the extended mode (Fig. 2(b)). Due to the exponential localization of the defect's motion, the Fourier transform of a far from the defect particle's displacement (Fig. 2(c)) does not reveal significant motion at the input frequency, and consists almost exclusively of down-converted energy. The frequency conversion efficiency, defined as the energy dissipated in the extended mode in comparison with the energy input into the system 2= , equals 10.8±0.9%. This efficiency arises from our particular system 3456578 34:6:78 ;<:<= Figure 2. (Color Online) Experimental response of the system under harmonic excitation. (a) Position of the magnets as a function of time. The red dotted line corresponds to the defect magnet, which acts as the input to the frequency-converting system. The green dashed line is taken as the output of the system. (b) Fourier transform of the defect magnet's position, which is modulated at the extended mode's frequency. The vertical dotted line represents the excitation frequency. (c) Fourier transform of the output magnet's position. This magnet's motion happens primarily at the second extended mode's frequency. 4 (1) Theoretical model – Our theoretical model describes the magnets as point masses. We model the interaction between particles using an empirical power-law model, AB =CBD, with C=3.378∙10G)HI!J.K)L and M=−4.316 determined experimentally (See Supplementary )YX3Q CBUV−W +SQ−SX D CBUW−V +SX−SQ D Information for the fitted force-displacement curves). This model does not have a straightforward physical justification in terms of the material properties and the geometry of the magnets, but it is chosen because it reproduces the experimental force law with very high precision and low complexity. Using this force-displacement law we can write the equation of motion for the system (the indices in parentheses indicate that no summation is performed over them): !(Q)S(Q)+T(Q)S(Q)− Where !(Q) and T(Q) are the mass and damping coefficient of the V-th particle, C and M are the magnetic force law parameters, AQ[ is the external driving force acting on the V-th particle (which may be zero if the particle is not externally driven), and BU is the distance BU is the same for all magnets. This is an approximation, since magnets that are not in the Runge-Kutta algorithm with a time step of 1 !\. center of the lattice are subject to asymmetric long-range forces. However, we have found this approximation to yield acceptable results. We emphasize that our theoretical model is not limited to nearest-neighbor interactions and takes into account the magnetic force between all pairs of magnets. All numerical integration in this paper is done using a 4th order between magnets at rest. When performing the reduced-order analysis, we will assume that =AQ[ Q3XYZ + Reduced modal description and frequency conversion mechanism – The mechanism responsible for the frequency conversion in our lattice becomes much clearer when we look at the evolution of the system in terms of the normal modes of the linearized system. We can obtain this description by approximating the force-displacement relation by a second order Taylor series. When we do this approximation, the system becomes: ]QXSX+MQXSX+^QXSX+_QX`SXS`=AQ[ Here, the indices W and a are summed over all degrees of freedom, ], M denote the mass and damping matrices defined conventionally, and the terms ^ and _ are obtained by Taylor ^QX= BBSX CBUV−! +SQ−Sb D + CBU!−V +Sb−SQ D )Yb3Q Q3bYZ BHBSXBS` _QX`=12 CBU!−V +Sb−SQ D + CBUV−! +SQ−Sb D Q3bYZ )Yb3Q Since ] is symmetric and positive-definite and ^ symmetric, we can find an invertible matrix c such that cd]c and cd^c are both diagonal. For simplicity, we assume that damping is expansion of the force law: proportional to the mass matrix and therefore also diagonalizable. In this basis, the equations of motion become: (2) (3) (4) 5 cQb]QXcXZeZ+cQbMQXcXZeZ+cQb^QXcXZeZ+_QX`cQbcXZc`feZef=Ab[ _QX`is highly non-local in the modal basis (i.e., modes far apart interact as strongly as nearby The diagonalized system in Eq. (5) does not provide any significant numerical advantage, since modes). However, we can see the motivation for this approach if we look at the experimental results in the modal basis (Fig. 3(a)). In this basis, most of the motion occurs in the second extended mode and in the localized mode. In fact, these modes hold around 90% of the system's energy (Fig. 3(b)). Therefore, we can restrict our description to these two modes without incurring a significant error. This reduced-order description is: (5) !1e1+T1e1+ a1−2ge/S1=Ahcos2lmh[ !/e/+T/e/+a/e/−ge1H=0 (6) (7) determined by performing a Taylor expansion of the interaction force, or by analyzing the frequency response of the local and extended modes (See Supplementary Information). We note that the reduced equations of motion present an asymmetry: There is no term In this description, e1 and e/ are the displacements of the localized and extended modes respectively, Ah is the input force, mh is the input frequency, !1, T1 and a1 are the effective mass, damping and stiffness of the localized defect mode, !/, T/ and a/ are the effective mass, damping and stiffness of the extended mode. The term g=_QX`cQ/cX1c`1=_QX`cQ1cX/c`1= _QX`cQ1cX1c`/ denotes the quadratic interaction between modes. This term can be proportional to S/H in Eq. (6) and there is no term proportional to S/S1 in Eq. (7). These terms square of the vibration amplitude, resulting in the term gS1H in the extended mode equation. introduces the parametric term na1=gS/, analogous to the modulation of the optical cavity do not appear in our lattice due to the location of the defect, but they are not generally zero (See Supplementary Information for a study on the relation between nonlinear terms and defect location). The interaction between modes can be understood in the following way: Due to nonlinearity, the vibration of the defect mode pushes against its neighbors, in a way that is analogous to thermal expansion of a crystal18 or the optical pressure in an opto- mechanical system (Fig. 3(c)). For small amplitudes, this expansion is proportional to the In addition, the motion of the extended mode modulates the distance between the defect particle and its neighbors (Fig. 3(d)). This affects the localized mode's effective stiffness and wavelength in an opto-mechanical system. This type of interaction appears in a variety of systems, such as phonon modes in superconductors28, and can result in stochastic heat engine opearation29. This reduced-order model can reproduce the experimentally-observed behavior (Figs. 3(e) and (f)) with remarkable accuracy. We highlight that the only fitting parameter used is the excitation amplitude. The particle mass, mode quality factor and inter- particle force law have all been measured experimentally. 6 Figure 3. (Color Online) Reduced-order description of the frequency conversion process. (a) Projection of the experimental time evolution (Fig. 2(a)) in the linear modal basis. (b) Average energy as a function of the mode number. The system's energy is highly concentrated in the second extended mode and the localized defect mode. (c) Dynamic expansion of the defect mode. When the defect vibrates, the nonlinear magnetic interaction results in a non-zero average force acting on the defect's neighbors. (d) The motion of the second extended mode modulates the distance between the defect particle and its neighbors, dynamically tuning the defect mode frequency. (e) Detail of the extended mode and localized mode evolution, measured experimentally. (f) Theoretical prediction for the extended and localized mode evolution, obtained from a reduced-order model considering only two modes (Eq. (6) and (7)). The numerical simulation in panel f corresponds to a system with !/= 0.45 ', !1=0.232 ', m/=0.5664 op, m1=3.913 op, mh=4.38 op, Ah=45 qI, r/= 4.518, r1=66.62 and g=1.801 I !H, where as=!s2lms H and Ts=!s2lms rs. The term gS1H acting on the extended mode plays the role of the optical pressure, while the term 2gS/S1 acting on the localized mode reproduces the modulation of the Fabry-Perot can be made explicit by expressing the motion of the defect mode as e1[ = 12[u[vQwx+u∗[vGQwx] and assuming that u[ changes slowly and that 1r1≪1. The two-mode system, described in Eq. 6 and Eq. 7, is a purely-mechanical analog of an opto- mechanical system30-32. The extended mode plays the role of the low-frequency mechanical motion, while the localized mode plays the role of the high-frequency electromagnetic field. resonance by the mechanical degree of freedom in an opto-mechanical system. This analogy With these assumptions, we arrive at the following equation (a detailed derivation and comparison with the full model are provided in the Supplementary Information): 7 !e/+T/e/+a/e/=guH2 u+u .U2r1−Vne/ =Ah (8) (9) Here, .U is the natural frequency of the localized mode and the detuning ne/ is the difference between the localized mode's natural frequency and the defect's excitation frequency, as a function of the extended mode's position. All other parameters have the same meaning than they did in Eq. (6) and Eq. (7). While being an approximation, this form has numerical advantages by not containing rapidly changing components at the frequency of the localized mode, and not requiring the evaluation of trigonometric functions for the excitation. Besides numerical reasons, the description provided in Eq. (8) and Eq. (9) is identical to the model of an opto-mechanical system30,32,33, for which there is extensive analytical literature32,34. This analogy provides a lucid interpretation of the frequency converting mechanism, whereby the self-sustaining oscillations of the extended more are the result of a feedback mechanism between the extended mode's motion and the localized mode amplitude. In this picture, the localized mode amplitude u depends on the extended mode displacement through the term ne/ . Equation (9) imposes a retardation between u and e/ and, as a consequence, the term guH has a quadrature component (shifted 90 degrees from e/([)) that results in negative damping31. When this negative damping exceeds the value of T/, the system develops self-sustaining oscillations, which saturate at a finite value due to non-linearity31. Multiple-defect synchronized frequency conversion – Systems containing multiple defects can present synchronized frequency conversion, where the motion of multiple defects is determined by the same extended mode, thereby synchronizing the defect's modulation envelopes and resulting in the conversion of energy from multiple input frequencies to a single output frequency. The use of an extended mode to synchronize multiple resonant elements appears in the context of Josephson junction arrays35, and here we use it to extract energy from multiple sources of mechanical vibrations. Our experimental system consists of 20 magnets, with defect particles in positions 7 (0.244') and 14 (0.197'). The initial and final particles are fixed. As in the case with a single defect, we set an excitation frequency for each defect equal to the defect's frequency plus an extended mode's frequency. This time we use the third extended mode instead of the second, because it presents two regions of maximum strain at the two defect's positons. As we did in the single defect case, we increase both defect's excitation amplitudes simultaneously, until we observe self-sustaining oscillations far from the defect. Figure 4a shows the trajectories of the magnets in the self-sustaining regime. We calculate the energy transfer between both defects and the extended mode, by utilizing the empirical force-displacement relation and the defect's trajectories, and we observe that both defects are contributing energy to the extended mode with a power (c=<AS/>=< gS1HS/>) of 16.9±1.5 ~ and 25.8±4.0 ~ respectively, indicating successful extraction of energy from multiple sources. The frequency conversion efficiency is 20.5±10.4%. As in the previous case, the motion of the defects presents sidebands indicating a modulation by the extended mode (Fig. 4(b)). Far from the defect, the motion takes place exclusively at the extended mode's frequency, as required for successful frequency conversion operation. 8 Figure 4. (Color Online) Synchronized frequency conversion. (a) Position of the magnets as a function of time. The yellow dotted line (particle 7) and the red dotted-dashed line (particle 14) are defect magnets that act as the high-frequency inputs of the system. The green dashed line is the low frequency output. (b) Fourier transform of the defects' positions, which are modulated at the extended mode's frequency. The vertical dotted line represents the excitation frequency. (c) Fourier transform of the output magnet's position. This magnet's motion happens primarily at the third extended mode's frequency. As in the case with a single defect, expressing the magnet's trajectories in terms of the lattice's linear normal modes reveals that the motion (Fig. 5(a)) and the energy (Fig. 5(b)) are primarily concentrated in an extended mode (Fig. 5(c), left) and in the two localized modes, the profiles of which are depicted in Fig. 5(c). This energy concentration allows us to formulate a reduced-order description following the same procedure as in the system with a single defect. The resulting system of equations has the form: !1)S1)+T1)S1)+ a1)−2g)S/S1)=Ah)cos2lmhH[ !1HS1H+T1HS1H+ a1H−2gHS/S1H=AhHcos2lmhH[ !/S/+T/S/+a/S/−g)S1)H −gHS1HH =0 (10) (11) (12) The model in Eqs. (10)-(12) is capable of qualitatively predicting the evolution of the modes (Fig. 5(d) and 5(e)), but under-estimates the output amplitude relative to the experiments. We attribute this difference to uncertainty in the system's resonance frequencies and quality factors. This is suggested by the difference between theory and experiment in the extended mode's frequencies (See Supplementary Information) and in the phase of the localized mode's vibration. 9 Figure 5. (Color Online) Reduced-order description of the synchronized frequency conversion. (a) Time evolution of the magnets in terms of the linear eigenmode basis. (b) Energy distribution in each normal mode. The energy is concentrated in the third extended mode and in the two localized defect modes. (c) Mode profiles of the three most relevant eigenmodes. (d) Experimental time evolution of the third extended mode e/ and the two localized modes e1) and e1H as a function of time. (e) Theoretical using a 3-DOF reduced order model. The numerical parameters used in panel (e) are: !/=0.45 ', !1)=0.2318 ', !1H=0.2915 ', m/=0.7494 op, m1)=3.404 op, m1H=3.063 op, mh)=4.1 op, mhH=3.81 op, Ah)=AhH=42 qI, r/=12.27, r1)=39.3, r1H=60.27, g)=−2.4293I !H, gH=2.5761 I !H. prediction for the time evolution of the eigenmodes. The theoretical predictions have been obtained Output phase tunability – In our lattice, the output signal's phase is not prescribed by the inputs and can be dynamically tuned while the system is operating. This offers an opportunity to synchronize multiple devices, create passive and tunable phased arrays or transfer information by modulating the output signal's phase. We theoretically demonstrate this output phase tunability in Fig. 6(a)-(c). The phase modulation is accomplished by perturbing the last particle following a Gaussian profile given by SZ[ =CÄvGÅÇÅÉ7 CÄ denotes the maximum perturbation amplitude, [U is the moment where the perturbation 7Ñ7 (Fig. 6(a)), where is applied and Ö represents the width of the perturbation. We choose a Gaussian profile 10 because it is highly localized in both time and frequency domains. Applying this perturbation results in a change in the output signal phase, that persists long after the perturbation has vanished. Figure 6(b) shows the extended mode's displacement 1790 seconds after a perturbation, for different perturbation amplitudes. In this calculation, the perturbation width Ö=30\ is much smaller than the wait time, ensuring that no effect remains by the phase tunability shown in Fig. 6(c): Firstly, this tunability covers the whole range (0°−360°). if we wait an additional 1000\ until [=2790\. In the experimental system, this phase Secondly, the perturbation-induced phase shift persists for a period of time that is much longer than any of the system's time constants, since the phase does not change significantly time the results are obtained. Two remarkable observations shall be made regarding the stability would be limited by the presence of external noise sources and Brownian motion. profile. (b) Extended mode signal 2790 seconds after the phase-shifting perturbation has been Figure 6. (Color Online) Theoretical investigation of phase and frequency tunability. (a) Phase tuning scheme. The output signal's phase is tuned by moving the last particle (SZ) following a Gaussian effected. The lines correspond to perturbations with CU equal to 0 mm (blue, solid), 6.2562 mm (red, equation of motion (Eq. (1)) with BU=16.3 !!, !Q,Qà))=0.45 ', !))=0.197 ', TQ,Qà))= 306.83 qI\ !, T))=42.62 qI\ !, AQ,Qà))=0I, A))=Ahsin2lmQ[, Ah=48.45 qI and mh= 4.38 op. The force-law parameters are as described in the theoretical model section. (d) Frequency dotted) and 6.5917 mm (yellow, dashed). (c) Output phase as a function of the maximum displacement of the phase-adjusting perturbation. The blue solid line is measured 1790 seconds after the perturbation, while the circles are measured 1000 seconds after the first measurement, 2790 seconds after the perturbation's peak. Panels b and c have been obtained by integrating the full down-conversion ratio (top) and input and output frequencies (bottom) as a function of the mass ratio between the defect and extended modes. These plots have been obtained by keeping the extended mode's mass constant and modifying the defect's mass. (e) Frequency down-conversion ratio (top) and input and output frequencies (bottom) as a function of the extended mode mass, while keeping 11 the modal mass ratio !/ !1 constant. In this section, all parameters except the masses are identical to those in Fig. 3(f). Tunability – A remarkable feature of our frequency-converting system is the possibility of tuning the input and output frequencies over a broad range, both during the design phase and dynamically once the system has already been built. Figure (6) theoretically explores the relationship between the input and output frequencies and the modal masses. We first explore the effect of the mass ratio by altering the mass of the defect mode without altering that of the extended mode (Fig. 6(a)). This results in a change in the optimal input frequency without a significant effect on the output frequency. We then proceed to alter the masses of the localized and extended mode simultaneously (Fig. 6(b)). This affects both input and output frequencies, while maintaining the down-conversion ratio constant. In the conversion ratio calculation, we identify the optimal input frequency by sweeping the input between the resonance frequency of the localized mode and the resonance frequency of the localized mode plus twice the resonance frequency of the extended mode, and finding the input frequency that results in the highest energy transfer. In addition to the particle's mass, there are many unexplored avenues for tuning the frequency conversion ratio. Examples include the static compression applied on the chain, the magnets' strength and geometry and the application of an external magnetic fields36. In addition, modern 3D printed materials allow us to engineer nonlinear inter-particle interactions37 beyond these offered by magnetic systems. Conclusions and outlook – We have demonstrated that lattices composed of magnetically interacting particles with defects are capable of converting energy from high frequencies to lower frequencies, which need not be linked by harmonic/subharmonic relations. This is possible through the nonlinear coupling between extended and localized normal modes. Such frequency-converting lattice, analogous to opto-mechanical systems, is highly tunable in both frequency and phase, and can extract energy from multiple signals at different frequencies to generate a single-component output. This work may motivate the design of innovative nonlinear metamaterials and devices with tunable energy conversion capabilities. References 1. 118-119 (1961). 2. 3. A. Saleh, Physical Review Letters 65 (11), 1348-1351 (1990). 4. (1974). 5. 6. York, 1979). 7. 8. 9. 10. P. Malatkar and A. H. Nayfeh, Nonlinear Dynamics 31 (2), 225-242 (2003). K. Oh and A. H. Nayfeh, Nonlinear Dynamics 11 (2), 143-169 (1996). J. P. E. Fermi, S. Ulam, M. Tsingou, 1955. J. Ford, Physics Reports 213 (5), 271-310 (1992). Y. C. Chen, H. G. Winful and J. M. Liu, Applied Physics Letters 47 (3), 208-210 (1985). A. H. Nayfeh, Nonlinear oscillations / Ali Hasan Nayfeh, Dean T. Mook. (Wiley, New P. A. Franken, A. E. Hill, C. W. Peters and G. Weinreich, Physical Review Letters 7 (4), S.-M. Jung and K.-S. Yun, Applied Physics Letters 96 (11), 111906 (2010). J. G. Rarity, P. R. Tapster, E. Jakeman, T. Larchuk, R. A. Campos, M. C. Teich and B. E. R. H. Stolen, J. E. Bjorkholm and A. Ashkin, Applied Physics Letters 24 (7), 308-310 12 N. Boechler, G. Theocharis and C. Daraio, Nat Mater 10 (9), 665-668 (2011). J. Lydon, M. Serra-Garcia and C. Daraio, Physical Review Letters 113 (18), 185503 C. M. Donahue, P. W. J. Anzel, L. Bonanomi, T. A. Keller and C. Daraio, Applied Physics N. Boechler, J. Yang, G. Theocharis, P. G. Kevrekidis and C. Daraio, Journal of Applied N. Boechler, G. Theocharis, S. Job, P. G. Kevrekidis, M. A. Porter and C. Daraio, Physical G. Theocharis, N. Boechler, P. G. Kevrekidis, S. Job, M. A. Porter and C. Daraio, Physical G. Gantzounis, M. Serra-Garcia, K. Homma, J. M. Mendoza and C. Daraio, Journal of C. Daraio, V. F. Nesterenko, E. B. Herbold and S. Jin, Physical Review E 72 (1), 016603 V. F. Nesterenko, Journal of Applied Mechanics and Technical Physics 24 (5), 733-743 A. L. Chen and Y.-S. Wang, Physica B: Condensed Matter 392 (1–2), 369-378 (2007). C. Daraio, V. F. Nesterenko, E. B. Herbold and S. Jin, Physical Review Letters 96 (5), H. Yasuda, C. Chong, J. Yang and P. G. Kevrekidis, ArXiv (2016). F. Li, P. Anzel, J. Yang, P. G. Kevrekidis and C. Daraio, Nature Communications 5, 5311 11. M. Toda, Journal of the Physical Society of Japan 23 (3), 501-506 (1967). 12. (1983). 13. 14. 058002 (2006). 15. Review Letters 104 (24), 244302 (2010). 16. Review E 82 (5), 056604 (2010). 17. 18. M. Serra-Garcia, J. Lydon and C. Daraio, Physical Review E 93 (1), 010901 (2016). 19. (2014). 20. 21. (2014). 22. Letters 104 (1), 014103 (2014). 23. Physics 109 (7), 074906 (2011). 24. Applied Physics 114 (9), 093514 (2013). 25. (2005). 26. 27. M. Molerón, A. Leonard and C. Daraio, Journal of Applied Physics 115 (18), 184901 (2014). 28. R. Mankowsky, A. Subedi, M. Forst, S. O. Mariager, M. Chollet, H. T. Lemke, J. S. Robinson, J. M. Glownia, M. P. Minitti, A. Frano, M. Fechner, N. A. Spaldin, T. Loew, B. Keimer, A. Georges and A. Cavalleri, Nature 516 (7529), 71-73 (2014). 29. M. Serra-Garcia, A. Foehr, M. Molerón, J. Lydon, C. Chong and C. Daraio, Physical Review Letters 117 (1), 010602 (2016). 30. M. Hossein-Zadeh and K. J. Vahala, IEEE Journal of Selected Topics in Quantum Electronics 16 (1), 276-287 (2010). 31. M. Aspelmeyer, T. J. Kippenberg and F. Marquardt, Reviews of Modern Physics 86 (4), 1391-1452 (2014). 32. 33. Letters 94 (22), 223902 (2005). 34. 5301 (2005). 35. L. Ozyuzer, A. E. Koshelev, C. Kurter, N. Gopalsami, Q. Li, M. Tachiki, K. Kadowaki, T. Yamamoto, H. Minami, H. Yamaguchi, T. Tachiki, K. E. Gray, W.-K. Kwok and U. Welp, Science 318 (5854), 1291-1293 (2007). 36. 161903 (2016). T. J. Kippenberg and K. J. Vahala, Opt. Express 15 (25), 17172-17205 (2007). T. Carmon, H. Rokhsari, L. Yang, T. J. Kippenberg and K. J. Vahala, Physical Review F. Li, D. Ngo, J. Yang and C. Daraio, Applied Physics Letters 101 (17), 171903 (2012). H. Rokhsari, T. J. Kippenberg, T. Carmon and K. J. Vahala, Opt. Express 13 (14), 5293- F. Allein, V. Tournat, V. E. Gusev and G. Theocharis, Applied Physics Letters 108 (16), 13 F. Wang, O. Sigmund and J. S. Jensen, Journal of the Mechanics and Physics of Solids 37. 69, 156-174 (2014). 14 Supplementary Information: Tunable, synchronized frequency down-conversion in magnetic lattices with defects Marc Serra-Garcia1, Miguel Moleron1 and Chiara Daraio2 1Department of Mechanical and Process Engineering, Swiss Federal Institute of Technology (ETH), Zürich, Switzerland 2Engineering and Applied Science, California Institute of Technology, Pasadena, California 91125, USA email: [email protected] Force-displacement relation We experimentally determine the parameters for the interaction force by measuring the force-displacement relation and fitting it using a power law function of the form !=#$%. The measurements and fitted curve are plotted in Fig. S1. Through this procedure, we obtain values of #=3.378±0.751 /012.345 and 6=−4.316± 0.0460. Derivation of the optomechanical equations of motion FIG S1: Magnetic force-displacement relation. The blue circles are the experimental measurements and the red line represents the power-law fit. Here we show that the equations governing the reduced-order dynamics of the frequency converting lattice (Eq. 6-7 in the main paper) can be approximated by those describing an optomechanical system, under realistic assumptions. The equations that we want to approximate are: 1:;:+=:;:+ >:−2@;A;:=!BcosFG Eq. S1 1A;A+=A;A+>A;A−@;:H=0 Eq. S2 We start our approximation by expressing the displacement of the localized mode as a harmonic function, with a slowly-changing amplitude and phase given by the complex function IG: ;:=12IGJKLM+IJNKLM Eq. S3 This allows us to rewrite Eq. S1 as: 1: OPOMP 4HIGJKLM+IGJNKLM +QRLSRTR Eq. S4 OOM4H IGJKLM+IGJNKLM + >:−2@;A 4H[IGJKLM+IGJNKLM]=4H[!WJKLM+!WJNKLM] For equation S4 to hold, terms multiplied by JKLM must be identical on both sides of the equation (Taking terms multiplied by JNKLM would yield an identical condition. This identity results in: 1:XHXGH IGJKLM +1:FW:Y: XXG IGJKLM + >:−2@;AIGJKLM=!WJKLM Eq. S5 By evaluating the derivative and both sides by JKLM we obtain: 1: −FHIG +2ZFIG +IG +1:FW:Y: ZFIG +IG Eq. S6 + >:−2@$AIG =!W Now we make our first assumption: That the localized mode amplitude and phase are slowly changing. This allows us to neglect the second derivative of I in Eq. S6, resulting in: 1: −FHIG +2ZFIG +1:FW:Y: ZFIG +IG + >:−2@$AIG =!B Eq. S7 Which can be regrouped as: −FH+FW:Y:ZF IG =!B1: FW:Y:+2ZF IG + >:−2@;A1: Eq. S8 Since Y: is a large value (40-60 in our experimental setup) and FW: is of the same order than F, we can neglect the contribution of FW: Y: in the right hand side. This results in: −FH +FW:2Y: IG = !B2F1: IG + −Z2F >:−2@;A1: Eq. S9 We note that the term >:−2@$A 1: is the square of the localized mode's angular frequency, as a function of the displacement of the extended mode. We call this term H to distinguish it from the natural frequency of the localized mode when FW:$A the extended mode is at rest ($A=0), which we termed FW:. By defining the detuning [$A =FW:$A −F as the difference between the natural frequency FW:$A and the excitation frequency, we obtain: −Z H− FW:;A −[;A 2(FW:;A −[;A) FW:;A IG + H Eq. S10 +FW:2Y: IG = !B2F1: Expanding the squares in the left hand side of Equation 10, we obtain: +FW:2Y: IG = !B2F1: IG + Z([;A −2FW:;A)[;A Eq. S11 2(FW:;A −[;A) Since [≪FW:$A we can neglect the additive term [$A in the numerator and denominator: IG + Z(−2FW:;A)[;A +FW:2Y: IG = !B2F1: Eq. S12 2(FW:;A) Which can be simplified to the equation for a classical optomechancial system: IG + FW:2Y:−Z[;A IG = !B2F1: Eq. S13 We now examine the equation for the extended mode, by replacing ;:G (Eq. S3) into Eq. S2. : Eq. S14 1A;A+=A;A+>A;A−@4 IHJKHLM+IJNKHLM+2II =0 By neglecting the rapidly varying degrees of freedom at 2ω, we arrive at the equation: Eq. S15 Equations S13 and S15 correspond to the optomechanical model presented in the main paper. 1A;A+=A;A+>A;A−@2IH=0 Tuning the nonlinear parameters by modifying the defect's location 1:;:+=:;:+>:;:−_:;:H−2@A;:;A−@:;AH=!BcosFG 1A;A+=A;A+>A;A−_A;AH−2@:;:;A−@A;:H=0 We can tune the nonlinear parameters in our reduced-order model by changing the defect location. The most general equation of motion for the system, truncated to contain only second-order terms, is given by: Eq. S16 Eq. S17 Figure S2 presents the nonlinear parameters as a function of the defect location. The selected defect locations maximize the @A coupling, while ensuring that all other nonlinear parameters are small. The point of maximal @A corresponds to the region where the mode's strain `K=$Ka4−$K is maximal, resulting in the highest change in the defect-neighbor distance during the motion of the extended mode. FIG S2: Nonlinear parameters as a function of the defect location. (a) Here, the extended mode is the second extended mode of the lattice, corresponding to the single-defect system in the main paper. (b) The extended mode is the third extended mode of the lattice, corresponding to the two-defects system in the main paper. In both panels, the dotted line represents the experimental light defect location. In panel b, the dashed line represents the heavy defect location. Comparison between full, reduced and optomechanical system Here we present a comparison between the system's evolution predicted by the full model (Eq. XX in the main paper), the two-mode reduced order model (Eq. 1 in the main paper) and the optomechanical model (Eq. 6 and Eq. 7 in the main paper). FIG S3: Comparison of full and reduced models. a Time evolution of the localized and extended modes calculated using the full system simulation. The modal description has been obtained by projecting the trajectories into the modal basis. b Modal time evolution calculated using the two-mode reduced order model. c Modal time evolution calculated using the optomechanical model. The G=0 point has been selected independently in each simulation, in order to present a consistent phase. We observe that the three models produce similar predictions. This allows us to conclude that a reduced-order modelling approach provides a good approximation, and that our nonlinear lattice accurately mimics the dynamics of an optomechanical system. Here we discuss the determination of the resonance frequencies and quality factors for the third extended mode and the two localized modes, used in the section Multiple-defect syncronized frequency conversion of the main paper. The frequency is determined by exciting the modes using a variable frequency signal. We monitor each particle's motion and project it into the theoretically-predicted modal basis. The amplitude is determined by calculating the RMS value of the modal coordinate after subtracting the average. We then fit the frequency response using a Lorentzian function to obtain the mode's frequency and quality factor. Determination of the natural frequencies in the two-defect system FIG S4: Fitting of the two-defect system parameters. a Frequency response of the third extended mode. b Frequency response of the first localized mode (Centered around the defect with mass 1b4=0.197 d. c Frequency response of the second localized mode (Centered around the defect with mass 1b4=0.244 d Frequency 0.7494±0.0197 ef Extended mode Quality factor 12.27±6.24 3.404±0.004 ef Frequency First Localized mode Quality factor 39.30±3.35 Frequency 3.063±0.004 ef Second localized mode 60.27±10.34 Quality factor Table 1: Two-defect system model parameters. In all of our paper's simulations, the nonlinear parameter @ has been determined by performing a Taylor expansion of the magnetic force-displacement relation presented in Fig. S1. In some circumstances (For example, in microscopic systems) it may not be possible to accurately measure the interaction potential. Here we calculate the nonlinear parameter @ from the frequency response curves (Fig. S5a), by simultaneously monitoring the displacement of the extended mode (Fig. S5b) during the frequency response characterization. The equation of motion for the extended mode is given by: 1A;A+=A;A+>A;A−@;:H=0 Eq. S18 For excitation amplitudes below the self-oscillation threshold, ;: follows a harmonic motion with constant amplitude. Under these conditions, ;A cannot follow the rapid changes of ;:H and reacts only to its average value. Since the displacement of ;A during the frequency response measurement is quasistatic, we can neglect the terms 1A;A and =A;A. This results in the equation: Determination of the nonlinear constant from the frequency response <;A>= @>A <;:H> @=1.81±0.42 0 1H obtained Eq. S19 Figure S5c presents the extended mode displacement as a function of the localized mode amplitude. By fitting this relation using a quadratic polynomial, we obtain a nonlinear coefficient @=1.79±0.56 0 1H which compares extremely well with the value force- displacement relation. from the experimental FIG S5: Experimental determination of the nonlinear parameter @. a Frequency response of the localized mode. b Displacement of the extended mode as a function of the localized mode excitation frequency, measured simultaneously with panel a. c Experimental relationship between localized mode amplitude and extended mode static displacement (Crosses), and polynomial fit (Red line).
1705.01096	1	1705	2017-05-02T01:44:06	Simultaneous summation and recognition effects for a dual-emitter light-induced neuromorphic device	[ "physics.app-ph" ]	We propose and fabricate a dual-emitter light-induced neuromorphic device composed of two light-induced devices with a common collector and base. Two InGaN multiple quantum well diodes (MQW-diodes) are used as the emitters to generate light, and one InGaN MQW-diode is used as the common collector to absorb the emitted light. When the presynaptic voltages are synchronously applied to the two emitters, the collector demonstrates an adding together of the excitatory post synaptic voltage (EPSV). The width and period of the two input signals constitute the code to generate spatial summation and recognition effects at the same time. Experimental results confirm that temporal summation caused by the repetitive-pulse facilitation could significantly strengthen the spatial summation effect due to the adding together behavior when the repetitive stimulations are applied to the two emitters in rapid succession. Particularly, the resonant summation effect occurs at the co-summation region when the two repetitive-pulse signals have a resonant period, which offers a more sophisticated spatiotemporal EPSV summation function for the dual-emitter neuromorphic device.	physics.app-ph	physics	Simultaneous summation and recognition effects for a dual-emitter light-induced neuromorphic device Yongchao Yang, Bingcheng Zhu, Jialei Yuan, Guixia Zhu, Xumin Gao, Yuanhang Li, Zheng Shi, Zhiyu Zhang, Yuhuai Liu, and Yongjin Wang  of and device fabricate composed neuromorphic Abstract-We propose a dual-emitter light-induced two light-induced devices with a common collector and base. Two InGaN multiple quantum well diodes (MQW-diodes) are used as the emitters to generate light, and one InGaN MQW-diode is used as the common collector to absorb the emitted light. When the presynaptic voltages are synchronously applied to the two emitters, the collector demonstrates an adding together of the excitatory post synaptic voltage (EPSV). The width and period of the two input signals constitute the code to generate spatial summation and recognition effects at the same time. Experimental results confirm that temporal summation caused by the repetitive-pulse facilitation could significantly strengthen the spatial summation effect due to the adding together behavior when the repetitive stimulations are applied to the two emitters in rapid succession. Particularly, the resonant summation effect occurs at the co-summation region when the two repetitive-pulse signals have a resonant period, which offers a more sophisticated spatiotemporal EPSV summation function for the dual-emitter neuromorphic device. Index Terms-InGaN multiple quantum well diode, dual-emitter light-induced neuromorphic device, excitatory postsynaptic voltage, resonant summation effect, adding together behavior. I. INTRODUCTION and is a hot The artificial synaptic device two-terminal memristors topic for brain-inspired neuromorphic systems [1-5], and synaptic electronics have gained considerable attention in recent years, including three-terminal ionic/electronic hybrid devices. Based on phase change materials, nanoelectronic programmable synapses have been developed for brain-inspired computing [6]. Dynamic logic and learning have been presented using a carbon nanotube synapse [7]. An Ag2S inorganic synapse has been reported to emulate the synaptic functions of both short-term plasticity and long-term potentiation characteristics [8]. Flexible metal oxide/grapheme oxide hybrid neuromorphic devices have demonstrated the realization of spatiotemporal correlated logics [9]. Proton-conducting grapheme oxide-coupled neuron devices have been proposed for brain-inspired cognitive systems [10]. Compared with other synaptic devices, the light-induced synaptic device uses photons rather than electrons or protons to induce excitatory postsynaptic voltage (EPSV) behavior for artificial synapse applications [11]. Here, we propose and fabricate a dual-emitter light-induced neuromorphic device on an III-nitride-on-silicon platform. Figure 1(a) shows a schematic illustration of the proposed dual-emitter light-induced neuromorphic device, which has a common base (B). The collector (C) absorbs the pulse light generated by the emitter (E) to achieve a photon-electron conversion, leading to an EPSV for the mimicking of synaptic activity with different signal sources. The period, shape and width of pulses constitute the code to transfer information in the biological nervous system, which is characterized by the EPSV summation two emitters are synchronously biased, the EPSVs happen at the same time and are added together, leading to a spatial summation. The adding together of EPSVs generated at the same emitter forms a temporal summation if they occur in a rapid succession. The spatial summation can be significantly reinforced by the temporal summation, which is investigated for emulating the complicated memory effect during the learning process. [12]. When the Fig. 1. (a) Schematic diagram of a dual-emitter neuromorphic device; (b) SEM image of fabricated dual-emitter neuromorphic device. II. EXPERIMENTAL RESULTS AND DISCUSSION is firstly thinned starting wafer The proposed dual-emitter neuromorphic device is fabricated on a 2-inch III-nitride-on-silicon wafer. The 1500-μm-thick to approximately 200 μm by chemical mechanical polishing. The emitter, collector and probing pad are patterned by photolithography and formed by induced coupled plasma reactive ion etching (ICP-RIE) of III-nitride epitaxial films with Cl2 and BCl3 hybrid gases at the flow rates of 10 sccm and 25 sccm, respectively. The Ni/Au (20nm/180nm) metal stacks are used as p- and n-type contacts. Then, waveguide structures are defined and etched by ICP-RIE. After protecting the top device structure with thick photoresist, silicon removal is conducted by deep reactive ion etching with alternating steps of SF6 etching and C4F8/O2 passivation. Subsequently, III-nitride backside thinning is carried out by ICP-RIE to obtain ultrathin membrane-type device architecture. Figure 1(b) shows a scanning electron microscope (SEM) image of the fabricated dual-emitter neuromorphic device. The suspended device architecture can form a highly-confined waveguide structure for the in-plane light coupling between the emitter and the collector. Two InGaN multiple-quantum-well diodes (MQW-diodes) are used as the emitters to generate light and one InGaN MQW-diode serves as a common collector to absorb through suspended waveguides. The two emitters are connected to the collector via two 100-µm-long, 2.47-µm high, and 6-µm-wide suspended in-plane guided light the 2 waveguides, and to 70-µm-diameter probe pads for device characterization. the electrodes are connected the simultaneously operate with 0.3mA&0.2mA. the injection currents of Fig. 2. (a) I-V curves of the four-terminal device, and the inset is electroluminescence (EL) spectra and optical power of the emitter; (b) Induced collector photocurrent versus forward current of the emitter. is controlled by the injection current of Figure 2(a) shows the measured EL spectra of single light emitter at different injection currents. The light emission intensity depends on the injection current, and the dominant emission wavelength locates around 452 nm, as shown in the inset of Fig. 2(a). The optical output power from single light emitter the MQW-emitter. The MQW-collector absorbs the emitted light to generate the photocurrent. The base contact is probably 150 µm away from emitter contact, and there will be a large resistance in BE. As a result, the current-voltage (I-V) curve of BE shows a relatively small current. According to the I-V curve of EE, the electrical isolation is obtained, indirectly indicating that the signal is coupled optically. The I-V curve of BC exhibits a typical rectifying behavior without light absorption, which is used to normalize the induced photocurrent with light absorption. When the two emitters are separately driven by the injection currents of 0mA&0.2mA to emit light, Figure 2(b) shows an increased photocurrent with increasing bias voltage of the collector from -4 to 4 V, in which the collector exhibits two light detection modes. Because the two emitters are symmetrical, an increased photocurrent is observed when the two emitters are synchronously driven by the injection currents of 0.3mA&0mA. The light intensity is modulated by the injection current of the emitter, so the induced photocurrent is dependent on the injection current of the emitter. Thus, the generated photons from different emitters can be accumulated when both emitters are operating simultaneously. The induced together when both emitters photocurrent is added Fig. 3. (a) Light coupling without separation gap; (c) coupling efficiency as a function of separation gap. Figure 3(a) shows the calculated light coupling between the emitter and the collector through suspended waveguide at a wavelength of 452 nm using beam propagation simulation, in which the refractive index of waveguide used is 2.45 and suspended waveguide is 100 µm long and 6 µm wide. Assuming that the suspended waveguide has a separation gap in the middle, Figure 3(b) shows the calculated coupling efficiency as a function of the separation gap. The coupling efficiency is significantly decreased with increasing the separation gap from 0 to 100 µm, indicating that the light being coupled in the waveguide does play a dominant role in the light coupling between the emitter and the collector because these optical components are fabricated on the same membrane [13]. Fig. 4. Spatial EPSV summation: (a) same width; (b) different widths. Figure 4(a) shows the spatial summation of the collector, in which two 8.0V@50μs pulse signals are separately applied to the two emitters simultaneously. Both the collector and the base operate at zero bias, and the two emitters are biased at 2.4 V. Compared to the EPSV generated by a single pulse signal from one emitter, the integrated EPSV amplitude increases, and the decay time extends, leading to an improved spatial summation effect. Moreover, the adding together of EPSVs is dependent on the pulse widths, as illustrated in Fig. 4(b). One 8.0V@50μs pulse signal and one 8.0V@30μs pulse signal are synchronously applied to the two emitters. The EPSV is integrated when two emitters are operating. The EPSV is then abruptly dropped and gradually increases when one pulse signal is terminated and only one emitter is driven by another pulse signal. Finally, the EPSV amplitude gradually returns towards its initial state due to the decay behavior when the later 3 pulse signal is also ended. The EPSV summation can enhance the signal amplitudes and remain their differences, and the integrated EPSV can be decoded into simple signals that distinctly identify the light source, leading to a recognition function. The dual-emitter light-induced device can achieve signal recognition through a spatial EPSV summation process when the signal pulse widths are different. Fig. 5. Spatiotemporal summation under different pulse widths. Repetitive-pulse facilitation (RPF) behavior occurs when the stimulation is continuously applied. Many EPSVs add together to produce a temporally integrated EPSV and finally reach a saturated value, which is determined by the initial amplitude of the pulse signal. However, it is still difficult to obtain a high temporal EPSV summation for the desired integration effect even if weak pulse signals are continuously applied [14]. The spatiotemporal EPSV could be achieved with the improved EPSV behavior through an adding together of spatial EPSV summation in a dual-emitter light-induced neuromorphic device. Two repetitive-pulse signals are separately applied to the two emitters simultaneously, in which the same pulse numbers are 50 and the same repetitive-pulse periods are 51 µs with the pulse widths of 50 and 21 µs, respectively. Figure 5 shows that the spatiotemporal EPSV amplitude increases and is eventually saturated as the pulse number increases. The spatiotemporal EPSV is improved due to the adding together of temporal and spatial summation. Compared to the first integrated EPSV amplitude of 113 mV, the saturated EPSV value is significantly increased to 402 mV. Signal recognition can simultaneously be obtained when repetitive stimulations with different pulse widths happen at the same period. The summation spatiotemporal a dual-emitter light-induced device is highly sensitive to the period of the repetitive-pulse signals. Compared to the spatial summation of two single signals from two emitters, the spatiotemporal summation behavior can be divided into a rapid summation region (RSR), a co-summation region (CSR), a temporal summation region (TSR), and a decay region (DR). Figure 6 shows the spatiotemporal summation behavior at different signal periods. One signal has a pulse width of 50 µs and a signal period of 51 µs, and the other has pulse width of 20 µs and a signal period of 21 µs. The pulse numbers of two repetitive signals are 50. Two RPF behavior patterns occur at RSR, and the integrated EPSV amplitude is quickly increased. for As the pulse number increases, the integrated EPSV amplitude enters into the CSR, in which the difference between the two signals leads to the signal recognition at a high EPSV level. When one repetitive signal is ended, the co-summation effect is stopped and the saturated EPSV amplitude is quickly dropped and enters into the TSR, in which only one RPF takes place. When the later signals are finished, the EPSV gradually returns to the initial state, resulting in a decay process. When the numbers of the two repetitive signals are increased, a periodic EPSV summation occurs, which relates to the resonant period. This sophisticated summation function occurs when the two repetitive pulse signals have a resonant period in the co-summation process. The two repetitive-pulse signals have the signal periods of 51 µs and 21 µs, respectively. Therefore, the two repetitive-pulse signals have a resonant period of 357 µs, leading to a distinct resonant summation effect (RSE). Taken together, it can be concluded that the RSE provides a more a multiple-emitter device, simultaneous summation and recognition of multiple signals at a high level. summation resulting complicated function the for in Fig. 6. Spatiotemporal summation behavior under different signal period. III. CONCLUSION In conclusion, we propose and fabricate a dual-emitter light-induced neuromorphic device to investigate the adding together of temporal and spatial EPSV summation effect. The integrated EPSV behavior occurs when pulse signals are synchronously applied to the two emitters. The period and width of signals lead to improved summation and recognition behavior at the same time. Experimental results confirm that the summation effect could be significantly strengthened due to the adding together of EPSV summation when repetitive stimulations are applied to the two emitters. Particularly, the RSE occurs at the CSR when the two repetitive-pulse signals have a resonant period, which provides a more sophisticated spatiotemporal EPSV summation function. REFERENCES [1] S. Song, K. D. Miller, and L. F. Abbott, "Competitive Hebbian Learning Through Spike-Timing Dependent Synaptic Plasticity," Nat. Neurosci. , Vol.3, pp.919-926, Sep. 2000, DOI: 10.1038/78829. [2] L. F. Abbott, and W. G. Regehr, "Synaptic Computation," Nature, Vol.431, pp.796-803, Oct. 2004, DOI: 10.1038/nature03010. [3] L. Q. Zhu, C. J. Wan, L. Q. Guo, Y. Shi, and Q. Wan, "Artificial synapse network on inorganic proton conductor for neuromorphic systems,". Nat. Commun. , Vol.5, pp.40-40, Jan.2014, DOI: 10.1038/ncomms4158. 4 [4] Q. X. Lai, L. Zhang, Z. Y. Li, W. F. Stickle, R. S. Williams, and Y. Chen, "Ionic/Electronic Hybrid Materials Integrated in a Synaptic Transistor with Signal Processing and Learning Functions," Adv. Mater. Vol. 22, pp. 2448-2453, May. 2010, DOI:10.1002/adma.201000282. [5] X. Yang, Y. C. Fang, Z. Z. Yu, Z. W. Wang, T. Zhang, M. H. Yin, M. Lin, Y. C. Yang, Y. M. Cai, and R. Huang, "Nonassociative Learning Implementation by Single Memristor-based Multi-terminal Synaptic Device," Nanoscale, Jan. 2016, DOI: 10.1039/C6NR04142F. [6] D. Kuzum, R. G. D. Jeyasingh, B. Lee, and H. S. P. Wong, "Nanoelectronic Programmable Synapses Based on Phase Change Materials for Brain-Inspired Computing," Nano Lett., Vol. 12, no. 5, pp. 2179-2186, 2011, DOI: 10.1021/nl201040y. [7] K. Kim, C. L. Chen, Q. Truong, A. M. Shen, and Y. Chen, "A Carbon Nanotube Synapse with Dynamic Logic and Learning," Adv. Mater. , Vol. 25, no. 12, pp. 1693-1698, 2013, DOI: 10.1002/adma.201370080. [8] T. Ohno, T. Hasegawa, T. Tsuruoka, K. Terabe, J. K. Gimzewski, and M. Aono, "Short-term plasticity and long-term potentiation mimicked in single inorganic synapses," Nature Mater. , Vol. 10, no. 8, pp. 591-595, 2011, DOI: 10.1038/nmat3054. [9] C. J. Wan, Y. H. Liu, P. Feng, W. Wang, L. Q. Zhu, Z. P. Liu, Y. Shi, and Q. Wan, "Flexible Metal Oxide/Graphene Oxide Hybrid Neuromorphic Transistors on Flexible Conducting Graphene Substrates," Adv. Mater. , Vol. 28, no. 28, pp. 5878-5885, 2016, DOI: 10.1002/adma.201600820. [10] C. J. Wan, L. Q. Zhu, Y. H. Liu, P. Feng, Z. P. Liu, H. L. Cao, P. Xiao, Y. Shi, and Q. Wan, "Proton Conducting Graphene Oxide-Coupled Neuron Transistors for Brain-Inspired Cognitive Systems," Adv. Mater. , Vol. 28, no. 18, pp. 3557-3563, 2016, DOI: 10.1002/adma.201505898. [11] Y. H. Li, Y. C. Yang, X. M. Gao, J. L. Yuan, G. X. Zhu, Z. Y. Zhang, and Y. J. Wang, "Light Induced Synaptic Transistor With Dual Operation Modes," IEEE Electron Device Lett. , Vol. 37, pp.1434-1437, 2016, DOI:10.1109/LED.2016.2607998. [12] M. F. Bear, B. W. Connors, and M. A. Paradiso, "Neuroscience: Exploring the Brain (3rd ed.)," Lippincott Williams & Wilkins, 2006. [13] Y. C. Yang, J. L. Yuan, Y. H. Li, X. M Gao, and Y. J. Wang, "On-chip integration for in-plane video transmission using visible light," SPIE Photonics Asia 2016, Vol. 10019, pp. 1001912-1001912, 2016, DOI: 10.1117/12.2247728. [14] G. X. Zhu, X. M. Gao, Y. H. Li, J. L. Yuan, W. Yuan, Y. C. Yang, Z. Y. Zhang, and Y. J. Wang, "Saturation behavior for a comb-like light-induced synaptic transistor," IEEE Electron Device Lett., Vol. 38, pp. 71-74, Jan. 2017, DOI: 10.1109/LED.2016.2627044.
1801.04114	1	1801	2018-01-12T09:55:48	Tunable elastic Parity-Time symmetric structure based on the shunted piezoelectric materials	[ "physics.app-ph" ]	We theoretically and numerically report on tunable elastic Parity-Time (PT) symmetric structure based on shunted piezoelectric units. We show that the elastic loss and gain can be archived in piezoelectric materials when they are shunted by external circuits containing positive and negative resistances. We present and discuss, as an example, the strongly dependent relationship between the exceptional points of a three-layered system and the impedance of their external shunted circuit. The achieved results evidence the PT symmetric structures based on this proposed concept can actively be tuned without any change of their geometric configurations.	physics.app-ph	physics	Tunable elastic Parity-Time symmetric structure based on the shunted piezoelectric materials Zhilin Hou†, Badreddine Assouar‡ †Department of Physics, South China University of Technology, Guangzhou 510640, China ‡CNRS, Institut Jean Lamour, Vandoeuvre-lès-Nancy F-54506, France and Université de Lorraine, Institut Jean Lamour, Boulevard des Aiguillettes, BP: 70239, 54506 Vandoeuvre-lès-Nancy, France Abstract We theoretically and numerically report on tunable elastic Parity-Time (PT) symmetric structure based on shunted piezoelectric units. We show that the elastic loss and gain can be archived in piezoelectric materials when they are shunted by external circuits containing positive and negative resistances. We present and discuss, as an example, the strongly dependent relationship between the exceptional points of a three-layered system and the impedance of their external shunted circuit. The achieved results evidence the PT symmetric structures based on this proposed concept can actively be tuned without any change of their geometric configurations. Keywords: elastic metamaterials, PT-symmetries, piezoelectric materials, tunablility. PACS：41.20.Jb,42.70.Qs, 46.40.Cd 1 I. Introduction In the past decades, extraordinary propagating behavior of acoustic waves in artificial structures has attracted much attention1-9. To control the wave propagation at will, various artificial structures with specially designed constructive units named as phononic crystal (PCs) and acoustic metamaterials (AMs) have been suggested and investigated. The efforts have produced tremendous achievements both on fundamental and engineering aspects. The discoveries include acoustic band gap1, subwavelength imaging2, acoustic cloaking3, negative refraction4, and so on. These phenomena are mostly achieved by the modulation of the real part of the acoustic parameters. Recently, parallel to the researches in optics10-18, a so-called Parity-Time (PT) symmetric structure has been suggested for the acoustic wave manipulation19-26. In contrast with the aforementioned PCs and AMs, the PT symmetric structure was built by elements with complex refraction index, which means the energy loss and gain are introduced into the wave propagation procedure. This new structure paves a new way for wave manipulation and associated applications. It has been demonstrated that novel wave behaviors existing in optical PT symmetric structures, such as coherent perfect absorption10, nontrivial anisotropic transmission resonances12, and unidirectional invisibility effects15, can also be realized in acoustic PT systems. However, because of the lack of freely controllable lossy and gain materials in nature, the realization of the PT symmetric structure for acoustic waves is remarkably difficult. To obtain the controllable acoustic gain and loss, several resolutions have 2 been proposed or used. For example, loudspeakers loaded by active circuits are used to fulfill the wave energy loss and gain in Ref.21. In Ref. 24, the loss and gain units were realized respectively by the leaky waveguide and two coupled acoustic sources, while in Ref.19, the acoustic loss and gain were obtained by the energy exchanging between the directional flow and sound through two specially designed diaphragms. Finally, in Ref.20 and Ref.25, the acoustoelectric and optomechanical effects are suggested to archive the elastic gain. We know that the acoustic loss or gain is just caused by the exchanging of acoustic energy in different forms, which means it can be archived through any materials if they can exchange the acoustic energy into other energy forms (for example, energy in electricity, mechanics, and so on). It can easily be found that piezoelectric material should be a suitable candidate for such purpose because it can give energy conversion between the elastic and electric forms. In fact, the shunted piezoelectric units have been well used in recent years in PC and AM structures27-30. For example, the piezoelectric units shunted by electric resistance can be used as elastic-wave-consuming structure28, 30. In addition, because the elastic parameters of the piezoelectric material strongly depend on the electric boundary conditions, it can also be used as constructing elements for PCs and AMs when it is shunted by an electric inductor or capacity29, 31. Remarkably, PCs and AMs based on the shunted piezoelectric units can have active tunable band gaps because the elastic property of the piezoelectric units can be freely adjusted by the shunted electric impedance27, 29. In this letter, we propose to realize elastic PT symmetry by shunted piezoelectric units. 3 As was shown in Refs.28, the acoustic loss can easily be obtained by shunting the piezoelectric unit with positive electric resistance, while to get the acoustic gain, the shunted external circuit should have negative electrical resistance, which can be archived by active non-Foster electrical circuits21, 32. The interest and added value of such design is that the loss/gain of the unit can actively be tuned by the external circuit, and when an inductance is added into the later, both of the real and imaginary parts of its effective parameter become tunable. This means that a tunable PT symmetric system for elastic waves can be realized. II. Calculation method To demonstrate this idea, we simply consider a system illustrated schematically in Fig. 1. It is constructed by two piezoelectric layers with a non-piezoelectric elastic layer inserted between them. The two piezoelectric layers with their surface normal to z direction are covered by thin enough electrodes and shunted by the external circuits with serial connected inductance 𝐿 and resistance ±𝑅. The thickness of all those three layers is set to be the same and denoted by l. The piezoelectric layers are polarized in z direction and the whole system is sandwiched by two half-infinite non-piezoelectric mediums. To obtain the tunable elastic loss and gain, the two piezoelectric layers are shunted by external circuits with the serial-connected resistance ±𝑅 and inductance L. Because the shunted positive R can introduce the elastic loss, the negative R can of course introduce the elastic gain. It is worth to mention that both the negative R and L can be achieved by active non-Foster circuits, the proposed structure can then be practically realized. 4 Fig. 1 Schematic illustration of the elastic PT-symmetic system. It is constructed by two piezoelectric layers with a non-piezoelectric elastic layer inserted between them. The left- and right-most half-infinite media are used for the inputting and outgoing wave. The system is taken to be uniform and infinite in x and y direction. "P" and "N" represent the piezoelectric and non-piezoelectric layers, respectively. The two piezoelectric layers with their surface normal to z direction are covered by thin enough electrodes and shunted by the external circuits with serial connected inductance L and resistance ±𝑅. The thickness of all three layers is set to be the same and marked by l. For this simple system, if we only consider the longitudinal polarized waves propagating in z direction, it can be treated as 1D system, and the transmission and reflection ratios can be solved by the transfer matrix method. Following our previous work 29, considering the piezoelectric layer with elastic constant 𝜆, mass density ρ, piezoelectric stress constant 𝑒33 and permittivity 𝜀33, the transfer matrix for the shunted piezoelectric layer that connect the total displacement U and stress 𝜒 at the right- and left-most boundaries of the layer, can be written as: , (1) 5 00rlprlUUTaawhere "a" is an arbitrary length introduced to normalize the thickness of the layers, and 𝜆0 is an arbitrary value with dimension of elastic constant introduced to normalize the elastic constants of materials in the system. We set it directly as the elastic constant of the non-piezoelectric materials in the followed calculations. The "r" ("l") appearing as superscript of U and 𝜒 mean the value at the right-(left-)most boundary of the layer. The components of the transfer matrix 𝑇𝑝 have the form as: , (2) where the symbols in the equations are defined as: , (3) with S as the area of the cross-section and as the thickness of the layer. By using the continuum condition of 𝜆0𝑈 and 𝜒𝑎 at the interfaces between each two nearest layers, we can get the total transfer matrix for the waves from the right-most boundary of the left half-infinite medium to the left-most boundary of the 6 '2''2'''sincos11,1cossin11sin1,2sinsin1cos12,1sinsin1sincos12,2cossin1ppppKfKfTKfKfKfiZCKfTKfKKfKfiZCKfTKKfKfKfiZCKfKfTKfKfKfiZC23333'0330002,//////eaKkaflaSCfCfaLRZiiLRaalright half-infinite one. By defining the values of 𝜆0𝑈 and 𝜒𝑎 at the right-(left-)most boundaries of the left (right) half-infinite medium as 𝜆0𝑈𝐿(𝑅) and 𝜒𝐿(𝑅)𝑎, we have the relationship , (4) where 𝑇 = 𝑇𝑝2𝑇𝑛𝑇𝑝1 is the total transfer matrix, 𝑇𝑝1and 𝑇𝑝2 are the transfer matrixes for the piezoelectric layers with shunted "+R" and "-R", respectively, and 𝑇𝑛, which can be obtained by setting 𝜅 = 0 in Eq.(2), is the transfer matrix for the inserted non-piezoelectric layer between the two piezoelectric ones. To calculate the reflection and transmission properties of the system, we need to decompose further the expression of U and 𝜒. Because the waves at the boundaries of the half-infinite non-piezoelectric materials can be expressed as: , (5) where k0a = Ω/√λ0/ρ0 is the dimensionless wave vector, and AL(R)(BL(R)) is the displacement amplitude for wave propagating in positive (negative) z direction, Eq.(5) can be rewritten as: , (6) by which the reflection and transmission properties can be calculated. In the analysis of PT symmetric system, Eq. (6) is usually rewritten as a scattering matrix form as . (7) The relationships between the elements of S and 𝑇′ are 7 00RLRLUUTaa0000000LRLRLRLRLRLRUAAMikaikaaBB1'RLLRLLAAAMTMTBBBRLLRAASBB . (8) With (7) and (8), the reflecting and transmitting ratios for the waves inputted from positive and negative z direction can be calculated respectively by and respectively. , (9) (10) III. Results and discussion To intuitively show that the elastic loss and gain are introduced into the piezoelectric layer by the shunted circuits, we can describe the layer as an equivalent non-piezoelectric elastic one with the effective parameters 𝜌𝑒 and 𝜆𝑒. If the elastic loss/gain is indeed introduced, they would be reflected as the non-zero negative/positive imaginary parts of 𝜌𝑒 and 𝜆𝑒. To obtain the effective parameters, we rewrite the transfer matrix for piezoelectric layer, Eq.(2), as the one for the non-piezoelectric layer as . (11) Then the effective normalized wave vector 𝐾𝑒 = Ω/√𝜆𝑒/𝜌𝑒 and elastic parameter (𝜆𝑒)′ = 𝜆𝑒/𝜆0 can be solved by equations 𝑇𝑝(1,1) = 𝑇𝑝 𝑒(1,1) and 𝑇𝑝(2,1) = 𝑇𝑝 𝑒(2,1) [or equivalently by 𝑇𝑝(2,2) = 𝑇𝑝 𝑒(2,2) and 𝑇𝑝(1,2) = 𝑇𝑝 8 𝑒(1,2)], and the '''''''''1,11,11,2*2,1/2,21,21,2/2,22,12,1/2,22,21/2,2STTTTSTTSTTST1,2;1,1rStS2,1;2,2rStS''1cossinsincoseeeeepeeeeKfKfKTKKfKfeffective elastic parameters 𝜌𝑒 and 𝜆𝑒 can then be obtained. As an example, we calculated the (𝜌𝑒)′ = 𝜌𝑒/𝜌0 and (𝜆𝑒)′ as function of Ω. Here and in all the followed calculations, the geometric parameters for the piezoelectric layers are fixed to be 𝑓 = 1 (i.e., l=a), 𝐶0 = 𝜀33𝑆 𝑎 = 10−8𝐹 (obtained by setting 𝑎 = 5𝑚𝑚 and 𝑆 = 10𝑎 × 10𝑎), and the parameters for piezoelectric and non-piezoelectric materials are chosen to be , and , , , , respectively. The results are shown in Fig.2. In Figs. 2(a) and 2(b), the inductance and resistance in the shunted circuit are set to be 𝐿0 = 0, 𝑅0 = 104Ω/𝑚 and L0 = 20H/𝑚2, R0 = 104Ω/m respectively. Note that by setting the thickness of the piezoelectric layer as 𝑎 = 5𝑚𝑚, the normalized values 𝐿0 = 20𝐻/𝑚 and 𝑅0 = 104Ω/𝑚 correspond to 𝐿 = 5 × 10−4𝐻 and R= 50Ω respectively, which are the reachable values in practice. From the figure, one can observe that both (𝜆𝑒)′ and (𝜌𝑒)′ have shunted-circuit-depended complex values, which means the elastic loss/gain can indeed be introduced and adjusted by the shunted +𝑅/−𝑅 and L. Remarkably, the values of (𝜆𝑒)′ [and (𝜌𝑒)′] for +R and –R appear always as conjugate pair no matter what value of L is selected. Those properties will be very useful for the active tunable PT-symmetric system constructing. 9 10212.210/Nm37200/kgm123321308.8510/Fm23323.3/eCm308900/kgm120016.8410/Nm Fig. 2 Normalized effective elastic parameters λe/λ0 and ρe/ρ0 of the shunted piezoelectric layer (with f = 1 and C0 = ε33S a = 10−8F) as function of the normalized frequency. The real and imaginary parts of the parameters are shown by solid and dashed line, respectively. (a) and (b) are for the layer with L0 = 0, R0 = 104Ω/m; (c) and (d) for the layer with L0 = 20H/𝑚2, R0 = 104Ω/m. One of the extensively studied phenomena for the PT-symmetric systems is the unidirectional reflectionless of wave at the so-called exceptional point (EP)22, 26. It has been shown that the eigenvalues of the S matrix for a PT-symmetric system can either be real or be complex conjugate pair, the EP refers to the frequency at which the 10 complex conjugate eigenvalue pair coalesces to the real one. At this special frequency, the wave can pass through the system without any reflection in and only in one direction. The phenomenon was suggested to be used in several circumstances. However, because the phenomenon only occurs is the structure when the loss and gain are elegantly balanced, the specially designed system can only works at several single fixed frequencies. For practical application, an active tunable system without any changing of its geometric configuration is needed. Before the discussion about the tunability of our system, we first present a picture about the eigenvalues of S matrix and the transmission property for a special system with fixed 𝐿0 and 𝑅0, by which the concept of EP and unidirectional-reflectionless phenomenon can be introduced and checked. For this purpose, we chose the inductance and resistance in the shunted circuit to be 𝐿0 = 0 and 𝑅0 = 104Ω/𝑚 respectively. Shown in Fig. 3(a) are the absolute values of eigenvalues of the S matrix as the function of Ω (calculated by𝜉1,2 = 𝑆(1,1) ± √𝑆(1,2) × 𝑆(2,1)) for the system. It can be seen from which that, the two branches of the eigenvalues keep overlapped in most of frequency region except the intervals between Ω = (2992,3699)𝑚/𝑠 and (9682,11244)𝑚/𝑠 . The phenomenon is called PT-symmetric phase transition in some references, and the exact frequencies from which the PT-symmetric phase been broken is called the EPs. In the Figure, we marked those EPs as 𝐸𝑃𝑙𝑖 and 𝐸𝑃𝑟𝑖 (i=1,2), which means the exceptional points at the left- and right-end of the first and second frequency intervals mentioned above. Shown in Fig.3(b) , (c), (d) and (e) are the total field distributions 𝑢 for the waves 11 inputting from the left- and right-most half-infinite medium with frequencies at 𝐸𝑃𝑙1 and 𝐸𝑃𝑟1 respectively. It can be seen that, at 𝐸𝑃𝑙1, the wave can pass through the system without reflection only when it is inputted from the right-most half-infinite medium, while at 𝐸𝑃𝑟1, the reflectionless transmission occurs only when the wave is inputted from the left-most half-infinite medium. Fig. 3 (a) Eigenvalues of the S matrix for the system with L0 = 0, R0 = 104Ω/m, the four EPs of the system are respectively marked as 𝐸𝑃𝑙1,2 and 𝐸𝑃𝑟1,2. (b) and (c) are the absolute value of the 12 total displacement field for the wave incident from left- and right-hand side of the system, respectively, the frequency of the incident wave is the one at 𝐸𝑃𝑙1. (d) and (e) are the same as (b) and (c) but for the wave with frequency at 𝐸𝑃𝑟1. Now we turn to discuss the tunable property of our suggested system. Because the unidirectional-reflectionless phenomenon only occurs at EPs, we just need to show how the EPs can be shifted by changing 𝐿0 and 𝑅0. In Fig.4(a) and (b) we show the EPs as function of 𝑅0 with 𝐿0 = 0 and 20𝐻/𝑚2, respectively. In Fig. 4(c), the EPs are shown as a function of 𝐿0 but with 𝑅0 = 104Ω/𝑚. In all the sub-figures of Fig. 4, the EPl and EPr are denoted by empty and full dots respectively. It can be seen from Fig.4(a) that in the considered frequency region (Ω < 8000), the PT-symmetry broken phase occurs when 𝑅0 = 100Ω/𝑚. As we increase 𝑅0, the frequencies for EPl and EPr are shifted accordingly. This means the unidirectional-reflectionless phenomenon can indeed be actively tuned by the external shunted circuits. However, we have to point out that, for the considered system, no matter what the value of 𝑅0 is chosen, the frequency range in which EPs can be shifted is limited between Ω = 2746~4074𝑚/𝑠. To make the system more tunable, we need to add the inductance into the external shunted circuits. Shown in Fig. 4(b) is the EPs as the function of 𝑅0 for the system with 𝐿0 = 20𝐻/𝑚2. It can be seen that, by adding such a small inductance, the EPs are pushed to a lower frequency region between Ω = 1796~3954𝑚/𝑠. In Fig. 4(b), the six branches of the curve appeared between 𝑅0 = 100~5500Ω/𝑚 mean there are three pair of EPs in this region. To see further how EPs can be shifted by the added inductance, in Fig. 4(c) we give the 13 result for the system with 𝑅0 = 104Ω/𝑚 but with changeable 𝐿0. It can be found that the EPl (EPr) can be shifted to the lower (higher) frequency region when a suitable value of 𝐿0 is selected. However, as 𝐿0 becomes larger and larger, the EPl and EPr will get closer and closer, which means the PT-symmetric-broken phase will finally be closed. Fig. 4 EPs as functions for different 𝑅0 and/or 𝐿0. (a) and (b) are for the systems with 𝐿0 = 0 14 and 𝐿0 = 20𝐻/𝑚2 but with tunable 𝑅0 , respectively; (b) is for the system with 𝑅0 = 104Ω/𝑚 but with tunable 𝐿0. IV Conclusions In conclusion, we provide in this letter an original concept of a tunable elastic PT-symmetric system based on shunted piezoelectric units. The conjugated elastic loss and gain units, which are necessary for the PT-symmetric system, can be achieved by the piezoelectric units shunted with positive/negative electric resistance. Because the electric impedances of the external shunted circuits can freely be tuned, the elastic PT-symmetric device based on this idea can actively be tuned without any geometric configuration changing. This will be useful in many practical fields and paves the way for the emerging PT-symmetry applications. Acknowledgements This work is supported by the National Natural Science Foundation of China (Grant No: 11274121). References Y. Chen, M. Zheng, X. Liu, Y. Bi, Z. Sun, P. Xiang, J. Yang and G. Hu, Physical Review B 95 (18), J. Liu, Z. Hou and X. Fu, Physics Letters A 379 (36), 2097-2101 (2015). 1. M. I. Hussein, M. J. Leamy and M. Ruzzene, Applied Mechanics Reviews 66 (4), 040802 (2014). 2. X. Yang, J. Yin, G. Yu, L. Peng and N. Wang, Applied Physics Letters 107 (19), 193505 (2015). 3. 180104 (2017). 4. 5. G. Ma, M. Yang, S. Xiao, Z. Yang and P. Sheng, Nature materials 13 (9), 873-878 (2014). 6. 7. 8. 9. 5, 5553 (2014). 10. Y. D. Chong, L. Ge and A. D. Stone, Physical review letters 106 (9), 093902 (2011). Y. Zhu, X. Fan, B. Liang, J. Cheng and Y. Jing, Physical Review X 7 (2), 021034 (2017). Y. Li and B. M. Assouar, Applied Physics Letters 108 (6), 063502 (2016). F. Lemoult, M. Fink and G. Lerosey, Physical review letters 107 (6), 064301 (2011). Y. Xie, W. Wang, H. Chen, A. Konneker, B. I. Popa and S. A. Cummer, Nature communications 15 11. R. Fleury, D. L. Sounas and A. Alu, Physical review letters 113 (2), 023903 (2014). 12. L. Ge, Y. D. Chong and A. D. Stone, Physical Review A 85 (2), 023802 (2012). 13. A. Guo, G. J. Salamo, D. Duchesne, R. Morandotti, M. Volatier-Ravat, V. Aimez, G. A. Siviloglou and D. N. Christodoulides, Physical review letters 103 (9), 093902 (2009). 14. V. V. Konotop, J. Yang and D. A. Zezyulin, Reviews of Modern Physics 88 (3), 035002 (2016). 15. Z. Lin, H. Ramezani, T. Eichelkraut, T. Kottos, H. Cao and D. N. Christodoulides, Physical review letters 106 (21), 213901 (2011). 16. F. Monticone, C. A. Valagiannopoulos and A. Alù, Physical Review X 6 (4), 041018 (2016). 17. S. Weimann, M. Kremer, Y. Plotnik, Y. Lumer, S. Nolte, K. G. Makris, M. Segev, M. C. Rechtsman and A. Szameit, Nature materials 16 (4), 433-438 (2017). 18. M. Ornigotti and A. Szameit, Journal of Optics 16 (6), 065501 (2014). 19. Y. Auregan and V. Pagneux, Physical review letters 118 (17), 174301 (2017). 20. J. Christensen, M. Willatzen, V. R. Velasco and M. H. Lu, Physical review letters 116 (20), 207601 (2016). 21. R. Fleury, D. Sounas and A. Alu, Nature communications 6, 5905 (2015). 22. R. Fleury, D. Sounas and A. Alu, IEEE Journal of selected Topics in Quantum Electronics 22 (5), 5000809 (2016). 23. A. V. Poshakinskiy, A. N. Poddubny and A. Fainstein, Physical review letters 117 (22), 224302 (2016). 24. C. Shi, M. Dubois, Y. Chen, L. Cheng, H. Ramezani, Y. Wang and X. Zhang, Nature communications 7, 11110 (2016). 25. X.-W. Xu, Y.-x. Liu, C.-P. Sun and Y. Li, Physical Review A 92 (1), 013852 (2015). 26. X. Zhu, H. Ramezani, C. Shi, J. Zhu and X. Zhang, Physical Review X 4 (3), 031042 (2014). 27. A. Bergamini, T. Delpero, L. D. Simoni, L. D. Lillo, M. Ruzzene and P. Ermanni, Advanced Materials 26 (9), 1343-1347 (2014). 28. P. Gardonio and D. Casagrande, Journal of Sound and Vibration 395, 26-47 (2017). 29. Z. Hou and B. M. Assouar, Applied Physics Letters 106 (25), 251901 (2015). 30. J.-Y. Jeon, Journal of Mechanical Science and Technology 23 (5), 1435-1445 (2009). 31. J. Hwan Oh, I. Kyu Lee, P. Sik Ma and Y. Young Kim, Applied Physics Letters 99 (8), 083505 (2011). 32. see "Negative resistance- Wikipedia" https://en.wikipedia.org/wiki/Negative_resistance. 16
1907.13069	1	1907	2019-07-30T16:54:34	Analysis of hydrogen distribution and migration in fired passivating contacts (FPC)	[ "physics.app-ph" ]	In this work, the hydrogenation mechanism of fired passivating contacts (FPC) based on c-Si/SiO$_{x}$/nc-SiC$_{x}$(p) stacks was investigated, by correlating the passivation and local re-distribution of hydrogen. Secondary ion mass spectroscopy (SIMS) depth profiling was used to assess the hydrogen (/deuterium) content. The SIMS profiles show that hydrogen almost completely effuses out of the SiC$_{x}$(p) during firing, but can be re-introduced by hydrogenation via forming gas anneal (FGA) or by release from a hydrogen containing layer such as SiN$_{x}$:H. A pile-up of H at the c-Si/SiO$_{x}$ interface was observed and identified as a key element in the FPC's passivation mechanism. Moreover, the samples hydrogenated with SiN$_{x}$:H exhibited higher H content compared to those treated by FGA, resulting in higher iV$_{OC}$ values. Further investigations revealed that the doping of the SiC$_{x}$ layer does not affect the amount of interfacial defects passivated by the hydrogenation process presented in this work. Eventually, an effect of the oxide's nature on passivation quality is evidenced. iV$_{OC}$ values of up to 706 mV and 720 mV were reached with FPC test structures using chemical and UV-O$_{3}$ tunneling oxides, respectively, and up to 739 mV using a reference passivation sample featuring a ~25 nm thick thermal oxide.	physics.app-ph	physics	Analysis of Hydrogen Distribution and Migration in Fired Passivating Contacts (FPC) Mario Lehmanna,, Nathalie Valleb, Jörg Horzelc, Alisa Pshenovad, Philippe Wyssa, Max Döbelie, Matthieu Despeissec, Santhana Eswarad, Tom Wirtzd, Quentin Jeangrosa, Aïcha Hessler-Wysera, Franz-Josef Hauga, Andrea Ingenitoa, Christophe Ballifa,c a Ecole Polytechnique Fédérale de Lausanne (EPFL), Institute of Microengineering (IMT), Photovoltaics and Thin Film Electronics Laboratory, Rue de la Maladière 71b, 2002 Neuchâtel, Switzerland b Characterization and Testing Platform (MCTP), Luxembourg Institute of Science and Technology (LIST), Materials Research and Technology Department, 41, rue du Brill, L-4422 Belvaux, Luxembourg c CSEM PV-Center, Jaquet-Droz 1, 2002 Neuchâtel, Switzerland d Advanced Instrumentation for Ion Nano-Analytics (AINA), Luxembourg Institute of Science and Technology (LIST), Materials Research and Technology Department, 41, rue du Brill, L- 4422 Belvaux, Luxembourg e ETH Zurich, Laboratory of Ion Beam Physics, Otto-Stern-Weg 5, 8093 Zürich, Switzerland Corresponding author: [email protected] Keywords Passivating contacts, silicon solar cells, firing, hydrogenation, SIMS, interfacial silicon oxide ABSTRACT High temperature passivating contacts for c-Si based solar cells are intensively studied because of their potential in boosting solar cell efficiency while being compatible with industrial processes at high temperatures. In this work, the hydrogenation mechanism of fired passivating contacts (FPC) based on c-Si/SiOx/nc-SiCx(p) stacks was investigated. More specifically, the correlation between passivation and local re-distribution of hydrogen resulting from the application of different types of interfacial oxides (SiOx) and post-hydrogenation processes were analyzed. To do so, the applied processing sequence was interrupted at different stages in order to characterize the samples. To assess the hydrogen content, deuterium was introduced (alongside/instead of hydrogen) and secondary ion mass spectroscopy (SIMS) was used for depth profiling. Combining these results with lifetime measurements, the key role played by hydrogen in the passivation of defects at the c-Si/SiOx interface is discussed. The SIMS profiles show that hydrogen almost completely effuses out of the SiCx(p) during firing, but can be re- introduced by hydrogenation via forming gas anneal (FGA) or by release from a hydrogen containing layer such as SiNx:H. A pile-up of H at the c-Si/SiOx interface was observed and identified as a key element in the FPC's passivation mechanism. Moreover, the samples hydrogenated with SiNx:H exhibited higher H content compared to those treated by FGA, resulting in higher iVOC values. Further investigations revealed that the doping of the SiCx layer does not affect the amount of interfacial defects passivated by the hydrogenation process presented in this work. Eventually, an effect of the oxide's nature on passivation quality is evidenced. iVOC values of up to 706 mV and 720 mV were reached with FPC test structures using chemical and UV-O3 tunneling oxides, respectively, and up to 739 mV using a reference passivation sample featuring a ~25 nm thick thermal oxide. 1 1. INTRODUCTION Over the past decade, the photovoltaic (PV) market has seen a tremendous growth. While the annual installed PV capacity was still below 7 GWp in 2008, it reached 100 GWp in 2018 and this trend is expected to continue [1 -- 3]. This evolution was enabled by a continuous increase in solar cell efficiencies [4] and cost reductions for solar cells and PV modules [1]. One of the key factors for high efficiencies is the suppression of recombination losses at the contacts, usually achieved by the deposition of a material that passivates the wafer surface, deactivating defects that act as recombination centers [5,6]. A well-known example is the heterojunction solar cell, where an intrinsic hydrogen-rich amorphous silicon layer is used to passivate interfacial defects. This cell design reaches conversion efficiencies up to 26.7 % in an interdigitated back contacted design, the current world record for single junction c-Si based solar cells [7]. As these heterojunction devices rely on hydrogenated amorphous silicon layers for passivation, they are not compatible with the most common industrial metallization processes that require high-temperatures. Recently, cells based on so called high temperature passivating contacts (HTPC) have attracted attention thanks to conversion efficiencies >25.5% [8,9] combined with compatibility with the high temperatures (>800 °C) typical of nowadays industrial processes. Most of these passivating contacts are made of a thin (1.2 -- 3.6 nm) silicon oxide (SiOx) layer capped with a doped poly-silicon layer (poly-Si). The stack is then annealed at high temperature and subsequently hydrogenated to provide chemical passivation [10 -- 15]. This annealing is usually performed in a tube furnace at temperatures > 800 °C and with heating ramps of 1-10 °C/min, leading to a crystallization of the deposited silicon layers and in-diffusion of dopants into the silicon wafer forming a shallow doped region below the poly-Si/SiOx stack [16]. In contrast to such approaches, the recently published fired passivating contact (FPC) is fabricated in a single rapid thermal processing (RTP) step, also called firing [17]. This step is used to metallize industrial solar cells [18,19]. Such a process typically requires temperature >750 °C, which are reached with ramps of ~50 °C/s and maintained for a few seconds only. The fabrication of an FPC thus requires a much lower thermal budget than HTPCs based on long annealings. As firing is too short to promote dopant in-diffusion, excellent interface passivation is needed to avoid recombination losses and to achieve high open circuit voltages (VOC). Further, the high temperature ramps lead to fast hydrogen effusion, which can lead to blistering. Avoiding such layer delamination is thus a challenge for the FPCs but could be overcome by the addition of carbon into the Si-network [17]. The C content was tuned in order to avert blistering while fostering layer crystallization, which was found to be beneficial for surface passivation and charge carrier extraction. The integration of the FPC as rear hole selective contact, co-fired with a screen printed Ag grid contacting a POCl3 diffused front emitter, resulted in a conversion efficiency of 21.9 % [17]. Hydrogenation is an essential processing step for HTPC (both annealed and fired), during which interfacial defects are passivated, allowing to reach high VOC values. It also plays a key role in surface passivation of many other type of solar cell architectures, and even in bulk quality improvement [20,21]. In this work, the distribution and migration of hydrogen in FPCs is analyzed by means of Secondary Ion Mass Spectrometry (SIMS). The impact of hydrogen re- distribution on surface passivation is studied. A special focus is set on the effect of the various processing steps as well as the influence of the oxide nature on passivation and hydrogen distribution. Deuterium has been incorporated in the samples analyzed by SIMS. The advantage of this being that, in contrast to hydrogen, the deuterium signal is not affected by residual air present in the chamber or humidity adsorbed on the sample surface. Moreover, the detection capability of SIMS is higher for deuterium than for hydrogen. 2 2. EXPERIMENTAL 2.1. Fabrication Symmetrical test structures were fabricated on double side polished (DSP) or shiny etched (SE) p-type float zone (100) wafers. The DSP wafers had a thickness of 280 μm and a resistivity of 3 Ωcm, while the SE wafers, purchased from a different supplier, had a thickness and resistivity of 200 μm and 2 Ωcm, respectively. The first processing step consisted of a wet chemical cleaning, ending with a hot HNO3 treatment (69 %, 80 °C, 10 min) growing a ~1.3 nm thin wet chemical SiOx layer on the wafer surfaces [22,23]. Next, a ~25 nm thick hydrogenated amorphous a-SiCx(p):H (~2.5 at.% of carbon [17]) layer was deposited by Plasma Enhanced Chemical Vapor Deposition (PECVD) at 200 °C. Subsequently, the samples were fired for 3 s at ~800 °C. During this step, the initially amorphous film (a-SiCx(p):H) crystallizes into nanocrystalline nc-SiCx(p) and hydrogen effuses from that layer. Re-hydrogenation was then performed either via a forming gas anneal (FGA) for 30 min at 500 °C, or via hydrogen diffusion from a ~70 nm thick sacrificial layer of SiNx:H. The latter was deposited at 250 °C in an in-house built PECVD tool and optimized for release of H during a subsequent 30 min hotplate annealing at 450 °C [16]. After this hydrogenation step, the SiNx:H layer was removed in a HF solution. Ellipsometry measurements indicate a refractive index of ~2.0 for these SiNx:H layers. More details about the fabrication process can be found in [17]. Note that the standard hydrogenation route used in this paper is the one by sacrificial SiNx:H layer. FGA was applied only once for comparison. For the samples to be analyzed by SIMS, deuterium was added into the layers by replacing the H2 gas flows by D2 during the FGA and a-SiCx(p):H/D and SiNx:H/D PECVD processes. Note that during these PECVD depositions, SiH4 and NH3 or trimethylborane (TMB) gas flows were present alongside D2. Thus, both deuterium and hydrogen have been incorporated in these layers. The term hydrogenation is used indifferently whether deuterium is diffused alongside hydrogen for passivation or not. Due to its higher mass, deuterium has a lower diffusivity than hydrogen [24]. Thus, the kinetics of the hydrogenation process are expected to be different. However, the passivation mechanism should be identical, as the nature of both isotope's bond with Si is the same. Fig. 1: Schematic illustration of the fabrication process using a sacrificial SiNx:H/D layer for hydrogenation. Note that a chemical cleaning was performed prior to the oxidation. Deuterium was incorporated into the layers by replacing the H2 gas flows by D2 in the two mentioned PECVD processes. Different experiments were performed, investigating the effect of various parameters on the hydrogenation of the FPC. In the first one, the distribution and migration of hydrogen during the processing sequence and its effect on passivation was studied. To do so, deuterium was incorporated into the samples and their passivation and chemical composition were measured at different steps of the processing sequence: 1) after a-SiCx(p) deposition, 2) after firing, and 3 3) after hydrogenation (FGA or SiNx:H/D). All samples for this study were fabricated on DSP wafers, as a flat surface is needed for SIMS measurements. In a second experiment, the wet chemical oxide (HNO3) was replaced by a ~25 nm thick thermally grown oxide (90 min in an oxygen ambient at 900 °C, applying N2 ambient while ramping the temperature up and down [25 -- 28]) in order to determine more accurately the location of the deuterium in this layer (1.3 nm being smaller than the SIMS's depth resolution). A reference sample without oxide layer (HF stripping of the SiOx before the SiCx(p) deposition) was also processed. For both samples SIMS and lifetime measurements were performed in the as-deposited and hydrogenated state, and compared to the previous samples featuring a wet chemical oxide. In a third experiment, the effect of the interfacial oxide's nature on the passivation quality was investigated. The samples were fabricated on SE wafers, as no SIMS analysis was performed, and their lifetimes measured after hydrogenation (deposition of SiNx:H, hotplate treatment, stripping of the SiNx:H in HF). Three different types of interfacial oxides were compared with each other, namely chemical oxide, grown in hot HNO3 [22,23], UV-Ozone (UV-O3) oxide, grown by exposing the wafer to UV radiation in ambient air (2 min each side) [29 -- 31], and thermal oxide, grown in a tube furnace (grown as detailed above) [25 -- 28]. The thicknesses of the interfacial oxides were ~1.3 nm for the chemical and the UV-O3 oxides and ~25 nm for the thermal oxide, as measured by ellipsometry. Note that the latter sample has to be considered as a reference, as such a stack with a homogenous 25 nm thick SiO2 layer could not be applied as contact. Unfortunately, growing a homogeneous, 1.3 nm thin thermal oxide at 900 °C is very challenging (and reducing the oxidation temperature affects the oxide quality [32]). Thus, a thickness of 25 nm was chosen, in order to enable comparison with the other experiments, where such thick thermal oxides were wanted. Finally, the effect of the SiCx layer on the hydrogenation process was studied. To do so, intrinsic, p-type and n-type a-SiCx layers were deposited on both sides of SE wafers covered with a ~25 nm thick thermal oxide. After firing and hydrogenation, the SiCx layer was stripped by a selective etch-back in a 20 % KOH solution at 60 °C. Lifetime measurements were performed after each processing step. Note that the samples with an intrinsic SiCx layer were not fired. Indeed, the SiCx(i) was found to be more prone to blistering than the doped SiCx layers. Thus, the firing step was replaced by a long hotplate (HP) anneal (7h @ 500 °C) to effuse hydrogen out of the SiCx(i). 2.2. Characterization The passivation quality of the samples was assessed by measuring the photoconductance decay using a Sinton WCT-120 instrument, recording the injection dependent effective minority carrier lifetime (τeff) and computing the implied open circuit voltage (iVOC) at 1 sun [33 -- 35], implementing the Auger correction published by Richter et al. [36]. The dark saturation current density (J0) was extracted from this data according to the method published by Kimmerle et al. [37]. The J0 values are given per wafer side. From the lifetime, the effective surface recombination velocity (Seff) was computed according to Sproul's equation [38]. The value of the diffusivity needed for this computation was determined with the help of PV Lighthouse's mobility calculator [39]. The chemical composition of the layers was measured by SIMS, using a CAMECA SC-Ultra instrument with a 1 keV Cs+ primary ions bombardment. Deuterium was analyzed as D- and +. Ions were collected from an area of 60 μm in diameter, with a depth resolution of ~4 nm DCs2 (not element dependent) [40]. A selection of samples was further characterized by Rutherford Backscattering Spectrometry (RBS) with a 2 MeV He ion beam [41]. Measurements were performed at the ETH Laboratory of Ion Beam Physics using a silicon PIN diode detector under 168°. The hydrogen and deuterium content of the samples' layers was determined by Elastic 4 Recoil Detection Analysis (ERDA) under 30° using a 2 MeV He beam and the absorber foil technique [41]. The collected RBS data was analyzed by the RUMP code [42]. Note that the depth resolution of H in Si of this measurement technique is about 50 nm [41]. Our layers being thinner than that, the hydrogen and deuterium contents are given as a surface concentration (at/cm2), corresponding to the total amount of H and D throughout the layer stack. Layer thicknesses were measured using a UVISELTM ellipsometer from HORIBA Jobin Yvon S.A.S. 3. RESULTS 3.1 Hydrogen distribution and migration as a function of the processing step 15 cm 15 cm 15 Looking at the lifetime curves throughout the individual steps of the processing sequence (Fig. 2a), it can be observed that the samples do not reveal any appreciable surface passivation after SiCx deposition. The firing process then slightly increases their iVOC (< 605 mV, corresponding to τeff@10 -3 < 50 μs). Finally, the post hydrogenation process provides a significant improvement. It is interesting to observe that higher iVOC values were reached when the hydrogenation was done with a sacrificial SiNx:H, rather than via FGA. SiNx:H -3 = 950 μs, J0 = 22 ± 5 fA/cm2), hydrogenation resulted in iVOC values up to 693 mV (τeff@10 whereas FGA treated samples reached only 649 mV (τeff@10 -3 = 190 μs). To gain a deeper understanding of these iVOC trends, deuterium profiles were measured by SIMS (Fig. 2b), analyzing negative secondary ions (high sensitivity to deuterium). First of all, it can be noticed that the deuterium content in the a-SiCx(p):H layer in the as deposited state is high and homogenous. Nevertheless, its iVOC is low due to the defective nature of the SiOx/c-Si interface. During firing, deuterium effuses out of the SiCx(p) and its concentration drops below the detection limit of the SIMS. Finally, the hydrogenation results in an increase of the deuterium content in the SiCx(p) and a strong peak at the position of the SiOx layer. As expected from lifetime results in Fig. 2a, the hydrogenation by a sacrificial SiNx:H layer introduces more deuterium than the FGA, explaining the observed trends. This is consistent with the work by Lelièvre et al. and Dekkers et al., showing that part of the hydrogen released from SiNx:H is in its atomic form, which diffuses more rapidly than molecular hydrogen from FGA [43,44]. Other parameters potentially affecting the hydrogen diffusion are the process temperature (450 °C for the hotplate treatment, vs. 500 °C for the FGA) and the concentration of hydrogen in the source (~18 at.% for SiNx:H, vs. ~4 at.% for FGA). cm The total deuterium concentration in the layer, measured by He ERDA, is given in Fig. 2b for the as-deposited, fired and hydrogenated (by a sacrificial SiNx:H/D layer) samples: (2.0 ± 0.4)∙1015 at/cm2, < 0.1∙1015 at/cm2 and (0.4 ± 0.2)∙1015 at/cm2, respectively. Besides the deuterium introduced through a D2 gas flow, there is also hydrogen incorporated into the layers through precursor gases like SiH4 and TMB. The hydrogen content measured by He ERDA for the as-deposited, fired and hydrogenated (by a sacrificial SiNx:H/D layer) samples were (44 ± 5)∙1015 at/cm2, (3 ± 1)∙1015 at/cm2 and (5.8 ± 0.6)∙1015 at/cm2, respectively. Note that these values could be biased by adsorption of humidity on the sample surfaces before ERDA measurements. This effect could explain the fact that the measured hydrogen content is above the detection limit after firing, whereas it drops below this limit for deuterium. Assuming a background signal of 3∙1015 at/cm2 of hydrogen, a H/D ratio of ~7 is measured after hydrogenation. He ERDA measurements on the SiNx:H layer reveal a H/D ratio of ~1.6, indicating a faster diffusion for hydrogen than for deuterium, in agreement with literature [24]. The as-deposited sample displays a homogeneous hydrogen and deuterium distribution corresponding to a total combined concentration of H + D of (1.8 ± 0.4)∙1022 at/cm3, i.e. >25 at.% according to [45]. This amount was found to be much lower after firing and hydrogenation, as the layer crystallized, containing thus less structural defects to be hydrogenated. 5 Fig. 2: (a): Minority carrier lifetime curves of selected samples at various processing stages, namely as deposited, fired, FGA, SiNx:H hydrogenation (after deposition, hotplate anneal and removal of the SiNx layer in HF), as a function of minority carrier density (MCD). The dashed line marks the MCD of 𝟏 ∙ 𝟏𝟎𝟏𝟓𝐜𝐦−𝟑 at -3in the text are determined. (b): Deuterium depth profiles (D- SIMS intensities) for samples at which τeff@10 different processing stages. The total concentration of deuterium within the samples (CD,total), measured by He ERDA, is given for the as deposited, fired and SiNx:H hydrogenated samples. cm 15 3.2. Hydrogen distribution and migration for different thicknesses of the interfacial oxide The SIMS depth profiles showed that the deuterium accumulates mainly at the position of the SiOx layer, which is in agreement with the work performed by Schnabel et al. [46] and Dingemans et al. [47] using Al2O3:D as deuterium donor layer. However, as the thickness of the chemical oxide is lower than the depth resolution of the SIMS (~4 nm), no conclusion about the exact location of the deuterium can be drawn from these measurements. For a deeper understanding of deuterium accumulation after the hydrogenation process, we analyzed the SIMS profile of a sample grown with a thick thermal oxide, and compared it to those of samples + with a thin chemical tunneling oxide and no oxide (Fig. 3). For these measurements, DCs2 secondary ions were analyzed, as this mode is less prone to matrix effects and thus a more suited approach to compare signals coming from different materials. As shown in Fig. 3a, the sample without the interfacial SiOx has a similar amount of D in the a-SiCx(p) than the other samples, in both as-deposited and hydrogenated states. Despite that, surface passivation is poor (iVOC < 600 mV after hydrogenation). The comparison with Fig. 3b illustrates that it is the previously observed accumulation of D at the SiOx layer that enables high iVOC values. Finally, Fig. 3c shows that after hydrogenation, D accumulates at both SiO2/nc-SiCx(p) and SiO2/c-Si wafer interfaces whereas its concentration is low within the SiO2. Such results are in agreement with the hypothesis that hydrogen accumulates at defective interfaces to passivate defects. In this specific case the H-accumulation at c-Si/SiOx enables to reach high iVOC values [48,49]. The especially high iVOC value of 728 mV (τeff@10 -3 = 3350 μs, J0 = 1.4 ± 0.5 fA/cm2) obtained for the sample with the thermal oxide layer indicates potential for improvement for the thin interfacial oxides. It is also interesting to note that the thick thermal oxide layer provides much better passivation in the as deposited state than the chemical oxide. cm 15 6 1014101510160.0010.010.11iVOC: 517 mViVOC: 604 mViVOC: 649 mVLifetime (ms)MCD (cm-3) As dep Fired SiNx-HF FGAiVOC: 693 mV05101520253035101102103104CD,total: < 0.1×1015 at/cm2CD,total: (0.4  0.2)×1015 at/cm2CD,total: (2.0  0.4)×1015 at/cm2Intensity (cnt/s)Depth (nm) As dep Fired SiNx-HF FGASiOx(a)(b) + SIMS intensities) for samples with (a) no interfacial oxide, (b) a chemical Fig. 3: Deuterium profiles (DCs2 tunneling oxide and (c) a thick thermal oxide, in the as deposited state (dashed line) or after hydrogenation (solid line). The yellow area indicates the region of the oxide. The purple region on the right side of the graph corresponds to the c-Si wafer, whereas the a-SiCx(p) layer is located in the green region on the left. The iVOC values of the samples (in mV) are given below the lines. 3.3. Effect of the interfacial oxide's nature on the passivation quality In this section the impact of various interfacial oxides on the passivation is studied. To do so, SiCx(p) layers were deposited on SE wafers covered with a chemical oxide (~1.3 nm), a UV-O3 oxide (~1.3 nm) or a thermal oxide (25 nm). The measured lifetime curves as a function of the injection level are shown in Fig. 4. A first observation is a trend to higher iVOC values when moving from chemical, to UV-O3, to thermal oxide: 706 mV, 720 mV and 737 mV, respectively, corresponding to τeff@10 -3 of 555 μs, 1100 μs and 3170 μs, and J0 of 8.3 ± 0.6 fA/cm2, 7.5 ± 0.3 fA/cm2 and 2.3 ± 0.5 fA/cm2. This improvement is believed to be linked to changes in the oxide's chemistry, which, according to literature, becomes closer to the stoichiometric ratio of 1:2 when switching from a chemical, to a UV-O3, and then to a thermal oxide [29,50]. cm 15 Note that while the samples with chemical and UV-O3 oxide can be compared directly, as the type of interfacial oxide was the only parameter varied, care has to be taken when comparing them with the sample with thermal oxide as the latter is thicker and has a different thermal history. Nevertheless, these results show that the nature of this oxide has a major influence on the final passivation quality. A second observation is that all iVOC values are ~10 mV higher than in the previous experiments, thus exceeding 700 mV even for the samples processed with a chemical oxide (as previously published in [17]). The J0 value decreases from 22 to 8.3 fA/cm2 for the samples featuring a chemical oxide on a DSP and SE wafer, respectively, and increases from 1.4 to 2.3 fA/cm2 for the DSP and SE samples with a thermal oxide. The reason behind this difference is unclear. Potential factors are the nature of the surface and different bulk lifetimes (as these wafers are provided by different suppliers). Furthermore, it should be mentioned that the computation of J0 becomes inaccurate and dominated by experimental uncertainties when approaching low values (< 4 fA/cm2) [37]. Finally, the 720 mV of iVOC reached for the sample with a UV-O3 tunnelling oxide layer, corresponding to a J0 of 7.5 ± 0.3 fA/cm2, confirm the high potential of the FPC. 7 02000400060008000100101102103(a) As deposited HydrogenatedIntensity (cnt/s)Sputtering time (s)(c)02000400060008000 As deposited HydrogenatedSputtering time (s)02000400060008000563 mV613 mV527 mV696 mV628 mV~25 nm thermal SiO2~1.2 nm chemical SiOx As deposited HydrogenatedSputtering time (s)No Oxide728 mV(b) Fig. 4: Minority carrier lifetime curves, after hydrogenation, of samples with various interfacial oxides (chemical, UV-O3, thermal) on p-type SE wafers. The dashed line marks the Minority Carrier Density (MCD) of 𝟏 ∙ 𝟏𝟎𝟏𝟓𝐜𝐦−𝟑 at which τeff@10 -3 in the text are determined. 15 cm 3.4. Study of the influence of the SiCx layer's doping on the hydrogenation process A set of samples with p-type, n-type and intrinsic SiCx layers deposited by PECVD on thick thermal oxides was prepared to investigate whether the doping of the nc-SiCx layer affects the hydrogenation process of the FPC, i.e. whether it influences the amount and charge state of the H diffused to the c-Si/SiOx interface, thus affecting the passivation quality. Such an effect has been reported by Yang et al. [51]. The usual processing sequence was completed with a selective etch-back of the SiCx layer in a KOH solution, in order to eliminate a potential field effect contribution by the doped layer to the surface passivation. Care was taken to selectively etch the partially crystallized SiCx layer and not the underlying oxide layer, such that the passivation of the interface was not compromised. The results are reported in Fig. 5. As can be observed, all iVOC values remain < 690 mV until the SiNx:H deposition. Excellent passivation can then be obtained thanks to the diffusion of hydrogen from the nitride layer towards the c-Si/SiO2 interface on a hotplate, reaching iVOC values > 740 mV for all SiCx layers studied here. This value drops slightly (by 4-8 mV) after stripping of the SiNx layer. Similarly, after etching off the SiCx layer, only a slight degradation in iVOC (by 1-8 mV) was observed, the exception being the SiCx(i) which presented local blistering that probably induced inhomogeneous etching and thus locally severe damage of the oxide layer altering the surface passivation, as indicated by an increase in J0 from 2.3 ± 0.5 fA/cm2 to 13.2 ± 0.9 fA/cm2. Just after PECVD of the SiNx:H layer, a difference of ~40 mV in iVOC can be observed between samples featuring a SiCx(p) layer and samples featuring a SiCx(i/n) layer. The origin of this effect is still unclear and requires further investigation. However, this difference vanishes after hydrogenation, indicating that the final amount of defects passivated by hydrogen (diffused for 30 min at 450 °C) is independent of the layer doping. Further experiments aiming at investigating the impact of the layer doping on the kinetics of this hydrogenation process are required. Moreover, the fact that the SiCx layer can be removed without major passivation loss indicates that, in the present case of a ~25 nm thick thermal SiO2, the doped SiCx layer does not contribute to the passivation. Assuming that the fixed charge density in the thermal oxide is low [52,53], the high iVOC values can be predominantly attributed to the accumulation of hydrogen at the c-Si/SiO2 interface. Whether 8 1014101510160.20.40.60.8241 Chem. SiOx UV-O3 SiOx Therm. SiO2iVOC: 706 mViVOC: 720 mVLifetime (ms)MCD (cm-3)iVOC: 737 mVit is chemical passivation alone or if, and to which extent, interfacial charges play a role, remains an open question. Fig. 5: iVOC values as a function of the processing step of samples fabricated with a p-type, n-type or intrinsic SiCx layer on p-type SE wafers with a 25 nm thick thermal oxide. The values for the best sample processed are also shown. The passivation qualities of the SiCx(p) samples after firing were too low to measure an iVOC. In samples with ultra-thin tunnelling oxides additional mechanisms might come into play such as superficial changes in carrier concentrations within the wafer, induced by the doped SiCx layer (leading to band-bending) [17]. But the observations from section 3.1 suggest that hydrogen passivation of the c-Si/SiOx interface is also the key element to reach high lifetime values with ultra-thin oxides. 15 cm 15 In this experiment, iVOC values up to 739 mV after hydrogenation and stripping of the SiNx:H -3 of 3260 μs, a J0 of 2.7 ± 0.7 fA/cm2 and a layer were reached, corresponding to a τeff@10 Seff@10 -3 of 3 cm/s. According to literature, these values correspond to state-of-the-art passivation levels of p-type silicon wafers [36,53 -- 61]. Note also that the present samples have no in-diffused doped region, as stated previously, and that these oxides were grown at 900 °C, a comparably low temperature, and without addition of trichloroethane (TCA). Both, an increased oxidation temperature and an addition of TCA may improve the passivation quality of the thermal oxide [27,62], but also increase the process' complexity. cm 4. CONCLUSION The combination of lifetime measurements with SIMS analysis elucidated the key role of hydrogen in passivating defects at the c-Si/SiOx interface to reach high iVOC values. Moreover, it could be observed that hydrogen almost completely effuses out of the SiCx(p) during firing and is later re-introduced during the hydrogenation step. Performing this hydrogenation step via a SiNx:H sacrificial layer was demonstrated to be more efficient than FGA, which could be correlated with a higher amount of deuterium diffused into the contact and especially to the oxide-wafer interface. Further investigations revealed that in the case of ~25 nm thick thermal 9 Therm. SiO2PECVD SiCxFired/HPPECVD SiNxHotplateHFKOH600620640660680700720740iVOC (mV)Processing step SiCx(p) SiCx(n) SiCx(i) SiCx(p) - bestoxides, the accumulation of hydrogen at the c-Si/SiO2 interface is the predominant factor enabling excellent passivation levels, and that the doping of the SiCx layer does not affect the amount of interfacial defects passivated by our hydrogenation process. Furthermore, it was observed that the nature of the interfacial oxide has a major impact on the passivation quality. iVOC values up to 720 mV could be reached using an ultra-thin UV-O3 tunneling oxide, and up to 739 mV on a reference passivation sample using a ~25 nm thick thermal oxide. Acknowledgments The authors gratefully acknowledge support by the Swiss National Science Foundation (SNF) under Grant no. 200021L_172924/1. The work was co-funded by the Luxembourg National Research Fund (FNR) through grant INTER/SNF/16/11536628 (NACHOS). Xavier Niquille is thanked for the wet chemical cleaning of the wafer and the growing of the UV-O3 oxides. The authors also thank Brahime El Adib (LIST) for his technical assistance with SIMS depth profiling, Christoph Peter (ISC Konstanz) for non-contact corona-Kelvin measurements (not shown), as well as Gizem Nogay and Josua Stückelberger for fruitful discussions. Conflict of interest The authors declare no conflict of interest. Copyright © 2019. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/ Accepted manuscript (peer-reviewed). Final article published under: https://doi.org/10.1016/j.solmat.2019.110018 References: [1] [2] [3] [4] [5] [6] [7] [8] W. Report 2018, 2018. Philipps, Warmuth, Photovoltaics VDMA, International Technology Roadmap for Photovoltaic (ITRPV) - Results 2018, ITRPV. 10th Editi (2019) 1 -- 38. https://itrpv.vdma.org/web/itrpv/download. G. Masson, I. Kaizuka, Trends 2018 in Photovoltaic Applications, 2019. S. https://www.ise.fraunhofer.de/content/dam/ise/de/documents/publications/studies/Photovoltaics- Report.pdf. NREL, Best Research-Cell Efficiencies, (2019). https://www.nrel.gov/pv/cell-efficiency.html (accessed March 21, 2019). A. Metz, D. Adler, S. Bagus, H. Blanke, M. Bothar, E. Brouwer, S. Dauwe, K. Dressler, R. Droessler, T. Droste, M. Fiedler, Y. Gassenbauer, T. Grahl, N. Hermert, W. Kuzminski, A. Lachowicz, T. Lauinger, N. Lenck, M. Manole, M. Martini, R. Messmer, C. Meyer, J. Moschner, K. Ramspeck, P. Roth, R. Schönfelder, B. Schum, J. Sticksel, K. Vaas, M. Volk, K. Wangemann, Industrial high performance crystalline silicon solar cells and modules based on rear surface passivation technology, Sol. Energy Mater. Sol. Cells. 120 (2014) 417 -- 425. doi:10.1016/j.solmat.2013.06.025. A. Cuevas, Physical model of back line-contact front-junction solar cells, J. Appl. Phys. 113 (2013) 164502. doi:10.1063/1.4800840. K. Yamamoto, D. Adachi, K. Yoshikawa, W. Yoshida, T. Irie, K. Konishi, T. Fujimoto, H. Kawasaki, M. Kanematsu, H. Ishibashi, T. Uto, Y. Takahashi, T. Terashita, G. Koizumi, N. Nakanishi, M. Yoshimi, Record-Breaking Efficiency Back-Contact Heterojunction Crystalline Si Solar Cell and Module, in: 33rd Eur. Photovolt. Sol. Energy Conf. Exhib. Rec., Amsterdam, Netherlands, 2017: pp. 201 -- 204. doi:10.4229/EUPVSEC20172017-2BP.1.1. A. Richter, J. Benick, F. Feldmann, A. Fell, M. Hermle, S.W. Glunz, N-Type Si solar cells with passivating electron contact: Identifying sources for efficiency limitations by wafer thickness and resistivity variation, 10 [9] Sol. Energy Mater. Sol. Cells. 173 (2017) 96 -- 105. doi:10.1016/j.solmat.2017.05.042. F. Haase, C. Hollemann, S. Schäfer, A. Merkle, M. Rienäcker, J. Krügener, R. Brendel, R. Peibst, Laser contact openings for local poly-Si-metal contacts enabling 26.1%-efficient POLO-IBC solar cells, Sol. Energy Mater. Sol. Cells. 186 (2018) 184 -- 193. doi:10.1016/j.solmat.2018.06.020. [10] Y. Tao, V. Upadhyaya, C.-W. Chen, A. Payne, E.L. Chang, A. Upadhyaya, A. Rohatgi, Large area tunnel oxide passivated rear contact n -type Si solar cells with 21.2% efficiency, Prog. Photovoltaics Res. Appl. 24 (2016) 830 -- 835. doi:10.1002/pip.2739. [12] [11] U. Römer, R. Peibst, T. Ohrdes, B. Lim, J. Krügener, E. Bugiel, T. Wietler, R. Brendel, Recombination behavior and contact resistance of n+ and p+ poly-crystalline Si/mono-crystalline Si junctions, Sol. Energy Mater. Sol. Cells. 131 (2014) 85 -- 91. doi:10.1016/j.solmat.2014.06.003. F. Feldmann, M. Bivour, C. Reichel, M. Hermle, S.W. Glunz, Passivated rear contacts for high-efficiency n-type Si solar cells providing high interface passivation quality and excellent transport characteristics, Sol. Energy Mater. Sol. Cells. 120 (2014) 270 -- 274. doi:10.1016/j.solmat.2013.09.017. J. Stuckelberger, G. Nogay, P. Wyss, M. Lehmann, C. Allebe, F. Debrot, M. Ledinsky, A. Fejfar, M. Despeisse, F.J. Haug, P. Loper, C. Ballif, Passivating contacts for silicon solar cells with 800 °C stability based on tunnel-oxide and highly crystalline thin silicon layer, in: Conf. Rec. IEEE Photovolt. Spec. Conf., IEEE, 2016: pp. 2518 -- 2521. doi:10.1109/PVSC.2016.7750100. [13] Passivating [14] G. Nogay, A. Ingenito, E. Rucavado, Q. Jeangros, J. Stuckelberger, P. Wyss, M. Morales-Masis, F.-J. Haug, P. Loper, C. Ballif, Crystalline Silicon Solar Cells With Coannealed Electron- and Hole-Selective SiCx 1478 -- 1485. doi:10.1109/JPHOTOV.2018.2866189. J. Stuckelberger, G. Nogay, P. Wyss, Q. Jeangros, C. Allebé, F. Debrot, X. Niquille, M. Ledinsky, A. Fejfar, M. Despeisse, F.J. Haug, P. Löper, C. Ballif, Passivating electron contact based on highly crystalline nanostructured silicon oxide layers for silicon solar cells, Sol. Energy Mater. Sol. Cells. 158 (2016) 2 -- 10. doi:10.1016/j.solmat.2016.06.040. Photovoltaics. Contacts, (2018) IEEE [15] J. 8 [16] G. Nogay, J. Stuckelberger, P. Wyss, E. Rucavado, C. Allebé, T. Koida, M. Morales-Masis, M. Despeisse, F.J. Haug, P. Löper, C. Ballif, Interplay of annealing temperature and doping in hole selective rear contacts based on silicon-rich silicon-carbide thin films, Sol. Energy Mater. Sol. Cells. 173 (2017) 18 -- 24. doi:10.1016/j.solmat.2017.06.039. [18] [17] A. Ingenito, G. Nogay, Q. Jeangros, E. Rucavado, C. Allebé, S. Eswara, N. Valle, T. Wirtz, J. Horzel, T. Koida, M. Morales-Masis, M. Despeisse, F.J. Haug, P. Löper, C. Ballif, A passivating contact for silicon solar cells formed during a single firing thermal annealing, Nat. Energy. 3 (2018) 800 -- 808. doi:10.1038/s41560-018-0239-4. F. Ye, W. Deng, W. Guo, R. Liu, D. Chen, Y. Chen, Y. Yang, N. Yuan, J. Ding, Z. Feng, P.P. Altermatt, P.J. Verlinden, 22.13% Efficient industrial p-type mono PERC solar cell, in: 2016 IEEE 43rd Photovolt. Spec. Conf., IEEE, 2016: pp. 3360 -- 3365. doi:10.1109/PVSC.2016.7750289. S. Duttagupta, N. Nandakumar, P. Padhamnath, J.K. Buatis, R. Stangl, A.G. Aberle, monoPolyTM cells: Large-area crystalline silicon solar cells with fire-through screen printed contact to doped polysilicon surfaces, Sol. Energy Mater. Sol. Cells. 187 (2018) 76 -- 81. doi:10.1016/j.solmat.2018.05.059. [19] [20] A.G. Aberle, Surface passivation of crystalline silicon solar cells: a review, Prog. Photovoltaics Res. Appl. 8 (2000) 473 -- 487. doi:10.1002/1099-159X(200009/10)8:5<473::AID-PIP337>3.0.CO;2-D. [21] B. Hallam, D. Chen, M. Kim, B. Stefani, B. Hoex, M. Abbott, S. Wenham, The role of hydrogenation and gettering in enhancing the efficiency of next-generation Si solar cells: An industrial perspective, Phys. Status Solidi Appl. Mater. Sci. 214 (2017). doi:10.1002/pssa.201700305. [22] H. Kobayashi, Asuha, O. Maida, M. Takahashi, H. Iwasa, Nitric acid oxidation of Si to form ultrathin silicon dioxide layers with a low leakage current density, J. Appl. Phys. 94 (2003) 7328 -- 7335. doi:10.1063/1.1621720. [23] N.E. Grant, K.R. McIntosh, Surface Passivation Attained by Silicon Dioxide Grown at Low Temperature 1676 -- 1679. 24th Eur. Photovolt. Sol. Energy Conf. (2009) 30 in Nitric Acid, doi:10.1109/LED.2009.2025898. [24] D.E. Carlson, C.W. Magee, A SIMS analysis of deuterium diffusion in hydrogenated amorphous silicon, [25] [26] Appl. Phys. Lett. 33 (1978) 81 -- 83. doi:10.1063/1.90153. J.G. Fossum, E.L. Burgess, High-efficiency p+-n-n+ back-surface-field silicon solar cells, Appl. Phys. Lett. 33 (1978) 238 -- 240. doi:10.1063/1.90311. J. Snel, The doped Si/SiO2 interface, Solid. State. Electron. 24 (1981) 135 -- 139. doi:10.1016/0038- 1101(81)90008-3. [27] A.W. Blakers, M.A. Green, Oxidation condition dependence of surface passivation in high efficiency silicon solar cells, Appl. Phys. Lett. 47 (1985) 818 -- 820. doi:10.1063/1.95994. 11 [28] E. Yablonovitch, T. Gmitter, R.M. Swanson, Y.H. Kwark, T. Gmitters, A 720 mV open circuit voltage SiO x :c-Si:SiO x double heterostructure solar cell, Appl. Phys. Lett. Appl. Phys. Lett. Appl. Phys. Lett. 47 (1985) 1211 -- 2510. doi:10.1063/1.96331. [29] A. Moldovan, F. Feldmann, M. Zimmer, J. Rentsch, J. Benick, M. Hermle, Tunnel oxide passivated carrier-selective contacts based on ultra-thin SiO2 layers, Sol. Energy Mater. Sol. Cells. 142 (2015) 123 -- 127. doi:10.1016/j.solmat.2015.06.048. [30] A. Fukano, H. Oyanagi, Highly insulating ultrathin SiO2 film grown by photooxidation, J. Appl. Phys. 94 (2003) 3345 -- 3349. doi:10.1063/1.1597940. [31] G. Krugel, J. Rentsch, M. Hermle, K. Kaufmann, C. Hagendorf, F. Feldmann, A. Roth-Fölsch, M. Zimmer, A. Moldovan, Simple Cleaning and Conditioning of Silicon Surfaces with UV/Ozone Sources, Energy Procedia. 55 (2014) 834 -- 844. doi:10.1016/j.egypro.2014.08.067. [32] Y. Lee, W. Oh, V.A. Dao, S.Q. Hussain, J. Yi, Ultrathin Oxide Passivation Layer by Rapid Thermal Oxidation for the Silicon Heterojunction Solar Cell Applications, Int. J. Photoenergy. 2012 (2012) 1 -- 5. doi:10.1155/2012/753456. [33] D.E. Kane, R.M. Swanson, Measurement of the emitter saturation current by a contactless photoconductivity decay method, in: Proc. 18th IEEE PVSC, Las Vegas, USA, 1985. [34] R.A. Sinton, A. Cuevas, M. Stuckings, Quasi-steady-state photoconductance, a new method for solar cell material and device characterization, in: Conf. Rec. Twenty Fifth IEEE Photovolt. Spec. Conf. - 1996, IEEE, 1996: pp. 457 -- 460. doi:10.1109/PVSC.1996.564042. [35] A. Cuevas, R.A. Sinton, Prediction of the open-circuit voltage of solar cells from the steady-state photoconductance, Prog. Photovoltaics Res. Appl. 5 (1997) 79 -- 90. doi:10.1002/(SICI)1099- 159X(199703/04)5:2<79::AID-PIP155>3.0.CO;2-J. [36] A. Richter, S.W. Glunz, F. Werner, J. Schmidt, A. Cuevas, Improved quantitative description of Auger recombination in crystalline silicon, Phys. Rev. B - Condens. Matter Mater. Phys. 86 (2012) 1 -- 14. doi:10.1103/PhysRevB.86.165202. [37] A. Kimmerle, J. Greulich, A. Wolf, Carrier-diffusion corrected J0-analysis of charge carrier lifetime measurements for increased consistency, Sol. Energy Mater. Sol. Cells. 142 (2015) 116 -- 122. doi:10.1016/j.solmat.2015.06.043. [38] A.B. Sproul, Dimensionless solution of the equation describing the effect of surface recombination on [39] carrier decay in semiconductors, J. Appl. Phys. 76 (1994) 2851 -- 2854. doi:10.1063/1.357521. PV (n.d.). https://www2.pvlighthouse.com.au/calculators/Mobility calculator/Mobility calculator.aspx (accessed March 20, 2019). Lighthouse Ltd., Pty. Mobility calculator, [40] A.R. Chanbasha, A.T.S. Wee, Ultralow-energy SIMS for shallow semiconductor depth profiling, Appl. Surf. Sci. 255 (2008) 1307 -- 1310. doi:10.1016/j.apsusc.2008.05.030. [41] M. Nastasi, J.W. Mayer, Y. Wang, Ion Beam Analysis: Fundamentals and Applications, CRC Press, 2015. L.R. Doolittle, A semiautomatic algorithm for rutherford backscattering analysis, Nucl. Instruments [42] Methods Phys. Res. Sect. B Beam Interact. with Mater. Atoms. 15 (1986) 227 -- 231. doi:10.1016/0168- 583X(86)90291-0. J.F. Lelièvre, E. Fourmond, A. Kaminski, O. Palais, D. Ballutaud, M. Lemiti, Study of the composition of hydrogenated silicon nitride SiNx:H for efficient surface and bulk passivation of silicon, Sol. Energy Mater. Sol. Cells. 93 (2009) 1281 -- 1289. doi:10.1016/j.solmat.2009.01.023. [43] [44] H.F.W. Dekkers, G. Beaucarne, M. Hiller, H. Charifi, A. Slaoui, Molecular hydrogen formation in hydrogenated silicon nitride, Appl. Phys. Lett. 89 (2006). doi:10.1063/1.2396900. [45] A.H.M. Smets, M.C.M. Van De Sanden, Relation of the Si H stretching frequency to the nanostructural Si (2007) 1 -- 4. - Condens. Matter Mater. Phys. 76 H bulk environment, Phys. Rev. B doi:10.1103/PhysRevB.76.073202. [46] M. Schnabel, B.W.H. van de Loo, W. Nemeth, B. Macco, P. Stradins, W.M.M. Kessels, D.L. Young, Hydrogen passivation of poly-Si/SiO x contacts for Si solar cells using Al 2 O 3 studied with deuterium, Appl. Phys. Lett. 112 (2018) 203901. doi:10.1063/1.5031118. [47] G. Dingemans, W. Beyer, M.C.M. Van De Sanden, W.M.M. Kessels, Hydrogen induced passivation of Si interfaces by Al2 O 3 films and SiO2/Al2 O3 stacks, Appl. Phys. Lett. 97 (2010) 2008 -- 2011. doi:10.1063/1.3497014. [48] A. Aberle, S. Glunz, W. Warta, J. Kopp, J. Knobloch, Si0 2 -passivated High Efficiency Silicon Solar Cells : Process Dependence of Si-SiO 2 Interface Recombination, in: A. Luque, G. Sala, W. Palz, G. Dos Santos, P. Helm (Eds.), Tenth E.C. Photovolt. Sol. Energy Conf., Springer, Dordrecht, Lisbon, Portugal, 1991: pp. 10 -- 14. doi:10.1007/978-94-011-3622-8_161. [49] B.E. Deal, E.L. MacKenna, P.L. Castro, Characteristics of Fast Surface States Associated with SiO2-Si 12 and Si3 N 4-SiO2-Si Structures, J. Electrochem. Soc. 116 (1969) 997 -- 1005. [50] B. Stegemann, K.M. Gad, P. Balamou, D. Sixtensson, D. Vössing, M. Kasemann, H. Angermann, Ultra- thin silicon oxide layers on crystalline silicon wafers: Comparison of advanced oxidation techniques with respect to chemically abrupt SiO2/Si interfaces with low defect densities, Appl. Surf. Sci. 395 (2017) 78 -- 85. doi:10.1016/j.apsusc.2016.06.090. [51] Y. Yang, P.P. Altermatt, Y. Cui, Y. Hu, D. Chen, L. Chen, G. Xu, X. Zhang, Y. Chen, P. Hamer, R.S. Bonilla, Z. Feng, P.J. Verlinden, Effect of carrier-induced hydrogenation on the passivation of the poly- Si/SiOx/c-Si interface, in: AIP Conf. Proc., American Institute of Physics, Lausanne, 2018: p. 040026. doi:10.1063/1.5049289. [52] A. Cuevas, T. Allen, J. Bullock, Skin care for healthy silicon solar cells, in: 2015 IEEE 42nd Photovolt. Spec. Conf., 2015: pp. 1 -- 6. doi:10.1109/PVSC.2015.7356379. [53] A.G. Aberle, S. Glunz, W. Warta, Impact of illumination level and oxide parameters on Shockley-Read- Hall recombination at the Si-SiO2interface, J. Appl. Phys. 71 (1992) 4422 -- 4431. doi:10.1063/1.350782. S.W. Glunz, A.B. Sproul, W. Warta, W. Wettling, Injection‐level‐dependent recombination velocities at the Si‐SiO 2 interface for various dopant concentrations, J. Appl. Phys. 75 (1994) 1611 -- 1615. doi:10.1063/1.356399. [54] [55] M.J. Kerr, A. Cuevas, Very low bulk and surface recombination in oxidized silicon wafers, Semicond. Sci. [56] Technol. 17 (2002) 35 -- 38. doi:10.1088/0268-1242/17/1/306. S. Mack, A. Wolf, C. Brosinsky, S. Schmeisser, A. Kimmerle, P. Saint-Cast, M. Hofmann, D. Biro, Silicon Surface Passivation by Thin Thermal Oxide/PECVD Layer Stack Systems, IEEE J. Photovoltaics. 1 (2011) 135 -- 145. doi:10.1109/JPHOTOV.2011.2173299. for Photovoltaic Applications, [57] N.E. Grant, T. Niewelt, N.R. Wilson, E.C. Wheeler-Jones, J. Bullock, M. Al-Amin, M.C. Schubert, A.C. Van Veen, A. Javey, J.D. Murphy, Superacid-Treated Silicon Surfaces: Extending the Limit of Carrier Lifetime (2017) 1574 -- 1583. doi:10.1109/JPHOTOV.2017.2751511. T. Niewelt, A. Richter, T.C. Kho, N.E. Grant, R.S. Bonilla, B. Steinhauser, J.I. Polzin, F. Feldmann, M. Hermle, J.D. Murphy, S.P. Phang, W. Kwapil, M.C. Schubert, Taking monocrystalline silicon to the ultimate 252 -- 259. doi:10.1016/j.solmat.2018.05.040. Energy Mater. IEEE J. Photovoltaics. 7 Sol. Cells. 185 (2018) [58] lifetime limit, Sol. [59] A. Cuevas, Y. Wan, D. Yan, C. Samundsett, T. Allen, X. Zhang, J. Cui, J. Bullock, Carrier population control and surface passivation in solar cells, Sol. Energy Mater. Sol. Cells. 184 (2018) 38 -- 47. doi:10.1016/j.solmat.2018.04.026. [60] A. Descoeudres, Z.C. Holman, L. Barraud, S. Morel, S. De Wolf, C. Ballif, >21% Efficient Silicon Heterojunction Solar Cells on N-and P-Type Wafers Compared, IEEE J. Photovoltaics. 3 (2013) 83 -- 89. doi:10.1109/JPHOTOV.2012.2209407. [61] R.S. Bonilla, B. Hoex, P. Hamer, P.R. Wilshaw, Dielectric surface passivation for silicon solar cells: A review, Phys. Status Solidi Appl. Mater. Sci. 214 (2017). doi:10.1002/pssa.201700293. [62] R.K. Bhan, S.K. Lomash, P.K. Basu, K.C. Cahabra, Interface Properties of Thermal SiO<inf>2</inf>using 1, 1, 1, Trichloroethane(TCA), J. Electrochem. Soc. 134 (1987) 2826 -- 2828. doi:10.1149/1.2100296. 13
1904.10902	1	1904	2019-04-24T16:10:30	An integrated cryogenic optical modulator	[ "physics.app-ph", "physics.optics" ]	Integrated electrical and photonic circuits (PIC) operating at cryogenic temperatures are fundamental building blocks required to achieve scalable quantum computing, and cryogenic computing technologies. Optical interconnects offer better performance and thermal insulation than electrical wires and are imperative for true quantum communication. Silicon PICs have matured for room temperature applications but their cryogenic performance is limited by the absence of efficient low temperature electro-optic (EO) modulation. While detectors and lasers perform better at low temperature, cryogenic optical switching remains an unsolved challenge. Here we demonstrate EO switching and modulation from room temperature down to 4 K by using the Pockels effect in integrated barium titanate (BaTiO3)-based devices. We report the nonlinear optical (NLO) properties of BaTiO3 in a temperature range which has previously not been explored, showing an effective Pockels coefficient of 200 pm/V at 4 K. We demonstrate the largest EO bandwidth (30 GHz) of any cryogenic switch to date, ultra-low-power tuning which is 10^9 times more efficient than thermal tuning, and high-speed data modulation at 20 Gbps. Our results demonstrate a missing component for cryogenic PICs. It removes major roadblocks for the realisation of novel cryogenic-compatible systems in the field of quantum computing and supercomputing, and for interfacing those systems with the real world at room-temperature.	physics.app-ph	physics	An integrated cryogenic optical modulator Authors: Felix Eltes1, Gerardo E. Villarreal-Garcia2, Daniele Caimi1, Heinz Siegwart1, Antonio A. Gentile2, Andy Hart2, Pascal Stark1, Graham D. Marshall2, Mark G. Thompson2, Jorge Barreto2, Jean Fompeyrine1, Stefan Abel1 1 IBM Research -- Zurich, Säumerstrasse 4, 8803 Rüschlikon, Switzerland 2 Quantum Engineering Technology Labs, H. H. Wills Physics Laboratory, University of Bristol, Bristol, BS8 1TL, United Kingdom Integrated electrical and photonic circuits (PIC) operating at cryogenic temperatures are fundamental building blocks required to achieve scalable quantum computing, and cryogenic computing technologies1 -- 4. Optical interconnects offer better performance and thermal insulation than electrical wires and are imperative for true quantum communication. Silicon PICs have matured for room temperature applications but their cryogenic performance is limited by the absence of efficient low temperature electro-optic (EO) modulation. While detectors and lasers perform better at low temperature5,6, cryogenic optical switching remains an unsolved challenge. Here we demonstrate EO switching and modulation from room temperature down to 4 K by using the Pockels effect in integrated barium titanate (BaTiO3)-based devices7. We report the nonlinear optical (NLO) properties of BaTiO3 in a temperature range which has previously not been explored, showing an effective Pockels coefficient of 200 pm/V at 4 K. We demonstrate the largest EO bandwidth (30 GHz) of any cryogenic switch to date, ultra-low-power tuning which is 109 times more efficient than thermal tuning, and high-speed data modulation at 20 Gbps. Our results demonstrate a missing component for cryogenic PICs. It removes major roadblocks for the realisation of novel cryogenic-compatible systems in the field of quantum computing and supercomputing, and for interfacing those systems with the real world at room-temperature. Cryogenic technologies are becoming essential for future computing systems, a trend fuelled by the world-wide quest to develop quantum computing systems and future generations of high- performance classical computing systems8,9. While most computing architectures rely solely on electronic circuits, photonic components are becoming increasingly important in two areas. First, PICs can be used for quantum computing approaches where the quantum nature of photons is exploited as qubits3,4. Second, optical interconnects can overcome limitations in bandwidth and heat leakage that are present in conventional electrical interconnect solutions for digital data transfer between cryogenic processors and the room temperature environment2. In addition, due to their low interaction with the environment, photons are the only viable carriers to transport quantum states over large distances. Optical interfaces are therefore essential for true quantum communication, necessary to connect multiple quantum computers10,11 and for secure remote operation of quantum computers12. Today, the realisation of such photonic concepts is hindered by the lack of switches and modulators that operate at cryogenic temperatures with low-loss, high bandwidth, and low static power consumption. So far, only two concepts for cryogenic EO switches have been investigated, based either on the thermo-optic effect13 or the plasma-dispersion effect14. Both mechanisms have physical limitations which intrinsically restrict the low-temperature performance of such devices. Thermo-optic phase shifters exploit Joule heating with large static power consumption and exhibit a bandwidth of less than a few MHz15. Plasma-dispersion-based devices require very high doping levels to compensate for carrier freeze-out at cryogenic temperatures. The high doping leads to large propagation losses and devices are limited to a bandwidth of <5 GHz in micro-disk modulators14. The use of EO switches based on the Pockels effect has been shown to offer low propagation losses and high-bandwidth, combined with low static power consumption at room temperature7,16 -- 18. Because of its electro-static nature, the Pockels effect has no intrinsic physical limitations for its application at cryogenic temperature. However, the lack of a Pockels effect in silicon means that heterogenous integration of new materials is needed to bring Pockels-based switching in PIC platforms. Pockels modulators have recently been demonstrated using organics19, PZT16, LiNbO3 20, and BaTiO3 7. Among them, BaTiO3 stands out due to having the largest Pockels coefficients7 and exhibiting compatibility with advanced silicon photonics platforms21. We complete this triumvirate by demonstrating that BaTiO3 is also an ideal candidate for cryogenic EO integration. Both the NLO properties and structural behaviour of BaTiO3 thin-films are entirely unknown at temperatures below 300 K. In fact, even in bulk BaTiO3 crystals the NLO behaviour is unexplored below 270 K, and the room temperature NLO behaviour of BaTiO3 thin-films has only recently been thoroughly investigated7. Predictions of the Pockels tensor at cryogenic temperatures based on data at higher temperatures is not possible because the temperature dependence of neither the Pockels coefficients nor the crystalline phase of thin-film BaTiO3 on Si is known. The phase transitions of thin-films are expected to differ from bulk crystals22 due to the structural mismatch and thermal stress that exists between the substrate and the BaTiO3 layer23 -- 25. Here, we determine the cryogenic behaviour of BaTiO3 thin films by analysing the performance of BaTiO3-based EO switches at temperatures down to 4 K. Our results show that efficient EO switching at cryogenic temperature is indeed possible and with bandwidths beyond 30 GHz. We also demonstrate the applicability of such devices for low-power switching and tuning as well as high-speed data modulation at 20 Gbps at 4 K. In this work, we use two waveguide designs fabricated on single crystalline BaTiO3 layers bonded to SiO2-buffered silicon substrates (Figure 1a, see Methods). In the first design, silicon nitride (SiN)-based waveguides allowed us to study the pure NLO properties of BaTiO3 in absence of mobile charge carriers which could result in an additional, non-Pockels EO response. In the second, silicon (Si) waveguides served as more efficient devices to demonstrate high-speed data modulation. The enhanced efficiency originates from a larger optical-mode overlap with the BaTiO3 layer (41 %) than with the SiN waveguides (18 %) (Figure 1b,c). We found that the propagation losses (5.6 dB/cm, SiN device) were not affected by the presence of BaTiO3 in the active section (Supplementary Note, SN 1) throughout the temperature range studied. To characterise the NLO behaviour of BaTiO3 at 4 K, we measured the induced resonance shift in a racetrack resonator as a function of the DC bias (Figure 1d,e), from which we determined the refractive index change of BaTiO3 (ΔnBTO) as a function of the applied electric field (see Methods). This dependence allows us to study two of the three expected features of Pockels-based switching7: NLO hysteresis and angular anisotropy, the third being the persistence of the Pockels effect at high frequencies (>10 GHz)26. The NLO response with a hysteretic behaviour (Figure 2a) indicates that a non-vanishing Pockels effect is preserved in BaTiO3 down to a temperature of 4 K. We determine the effective Pockels coefficient, reff, by analysing the hysteretic behaviour of the refractive index change (SN 2). The dependence of reff on device orientation (Figure 2b) reveals the second signature of the Pockels effect, its anisotropy. The reduced magnitude at 4 K compared to room temperature is due to a temperature dependence of the Pockels effect, as discussed below. While reff is reduced with temperature, the EO response is expected to be present at high frequencies also at low temperature. Indeed, we observe a constant EO response in racetrack resonators with a low Q factor (Q ~ 1,800) up to 30 GHz (Figure 2b). This constitutes the highest bandwidth for any cryogenic modulator reported to date. The frequency response is expected to remain flat at even higher frequency but could not be measured in our experiment (see Methods). The hysteretic behaviour, anisotropy, and high-speed response prove the presence of the Pockels effect in BaTiO3 at 4 K. We performed electrical characterisation of the material at low temperature using dedicated electrical test structures (SN 3). The resistivity of BaTiO3 at 4 K is very high, >109 Ωm. In fact, the measured current is dominated by capacitive charging and ferroelectric switching currents (Figure 2d). The field-dependent capacitance shows clear hysteretic characteristics (Figure 2e), consistent with ferroelectric domain switching. The measured reff at 4 K is lower than at room temperature (Figure 2b), which has two causes. First, the Pockels effect itself is generally temperature dependent due to changes in strain and polarisation of the crystal27. Second, the non-zero elements of the Pockels tensor depend on the crystal symmetry, which can change abruptly with temperature due to structural phase transitions. BaTiO3 bulk crystals are known to transition from a tetragonal phase at room temperature to orthorhombic and rhombohedral phases at lower temperatures (~270 K and ~200 K respectively)28. Such transitions change the elements of the Pockels tensor and modify the magnitude of the effective Pockels coefficients27. Because phase transitions of thin-film materials can be drastically affected by substrate strain23 -- 25, studying the properties of thin-film BaTiO3 becomes critical when considering cryogenic applications. To investigate the effects of possible phase transitions, we measured reff in a range from 4 to 340 K. Indeed, the magnitude of reff is strongly temperature-dependent (Figure 3). A peak around 240 K, with reff >700 pm/V, is consistent with the reported divergence of the r42 element of the Pockels tensor close to the tetragonal-orthorhombic transition27. Consistently, the permittivity of the BaTiO3 layer (see Methods) also shows a peak in the same temperature range (SN 4), confirming that the abrupt change in reff is caused by a phase transition. Below 240 K the magnitude of reff decreases gradually to around 200 pm/V at 4 K. In addition to the phase transition at 240 K, a second phase transition occurs below 100 K causing a rapid change in reff of 90° devices. This phase transition is also observed in the qualitative behaviour of the NLO hysteresis which shows that the transitions is induced by the electric field (SN 4). While reff of BaTiO3 is reduced at 4 K compared to room temperature, the value of ~200 pm/V is still larger than most other material systems at room temperature16,29,30. The effect of a reduced Pockels coefficient on the energy efficiency of EO switching is partially compensated for by a simultaneous reduction of the permittivity of BaTiO3 (SN 4). Additionally, the conductivity of BaTiO3 is reduced by more than four orders of magnitude (SN 3), resulting in a negligible static power consumption of BaTiO3-devices in cryogenic environments. We demonstrate the applicability of BaTiO3 for cryogenic photonic applications by two examples: low-power EO switching and high-speed data modulation. For switching we use a Mach-Zehnder interferometer with 2×2 multimode interference splitters, applying a voltage to one arm. Because the leakage current through BaTiO3 at 4 K is 104 times lower than at 300 K, less than 10 pW static power is consumed when inducing a π phase shift to switch between the two optical outputs (Figure 4a,b). Compared to state-of-the-art technology based on thermo-optic phase shifters13, static tuning using BaTiO3 is one billion times more power efficient. The dynamic energy of the switch is ~30 pJ, which could be reduced to ~2 pJ in an optimised device geometry (SN 5). As a second example, we performed data modulation experiments by sending a pseudo-random bit- sequence to the modulator and recording the optical eye-diagram (Figure 4c,d). Data transmission at rates up to 20 Gbps are achieved with our experimental setup using a drive voltage (Vpp) of just 1.7 V, resulting in an extremely low energy consumption of 45 fJ/bit. In conclusion, we have shown that BaTiO3 thin films can be used to realise EO switches and modulators for efficient cryogenic operation of silicon PICs. We have demonstrated low-power switching, as well as high-speed data modulation. Combining BaTiO3 with silicon photonic integrated circuits, we make a building block available that was previously inaccessible for any cryogenic circuits. We anticipate that such new components are a milestone for a versatile platform of cryogenic photonics for applications such as quantum computing and cryogenic computing technology, as well as quantum interconnects to room-temperature environments. 1. 2. 3. 4. 5. 6. 7. Gambetta, J. M., Chow, J. M. & Steffen, M. Building logical qubits in a superconducting quantum computing system. npj Quantum Inf. 3, 2 (2017). Holmes, D. S., Ripple, A. L. & Manheimer, M. A. Energy-Efficient Superconducting Computing -- Power Budgets and Requirements. IEEE Trans. Appl. Supercond. 23, 1701610 (2013). Silverstone, J. W., Bonneau, D., O'Brien, J. L. & Thompson, M. G. Silicon quantum photonics. IEEE J. Sel. Top. Quantum Electron. 22, 6700113 (2016). O'Brien, J. L., Furusawa, A. & Vučković, J. Photonic quantum technologies. Nat. Photonics 3, 687 -- 695 (2009). Pankove, J. I. Temperature Dependence of Emission Efficiency and Lasing Threshold in Laser Diodes. IEEE J. Quantum Electron. 4, 119 -- 122 (1968). Hadfield, R. H. Single-photon detectors for optical quantum information applications. Nat. Photonics 3, 696 -- 705 (2009). Abel, S. et al. Large Pockels effect in micro- and nanostructured barium titanate integrated on silicon. Nat. Mater. 18, 42 -- 47 (2019). 8. Mukhanov, O. A. Energy-Efficient single flux quantum technology. IEEE Trans. Appl. Supercond. 21, 760 -- 769 (2011). 9. Ladd, T. D. et al. Quantum Computing. Nature 464, 45 -- 53 (2010). 10. 11. Fan, L. et al. Superconducting cavity electro-optics: A platform for coherent photon conversion between superconducting and photonic circuits. Sci. Adv. 4, 1 -- 6 (2018). Javerzac-Galy, C. et al. On-chip microwave-to-optical quantum coherent converter based on a superconducting resonator coupled to an electro-optic microresonator. Phys. Rev. A 94, 1 -- 5 (2016). 12. Greganti, C., Roehsner, M. C., Barz, S., Morimae, T. & Walther, P. Demonstration of measurement-only blind quantum computing. New J. Phys. 18, 303 -- 309 (2016). 13. Elshaari, A. W., Zadeh, I. E., Jöns, K. D. & Zwiller, V. Thermo-Optic Characterization of Silicon Nitride Resonators for Cryogenic Photonic Circuits. IEEE Photonics J. 8, (2016). 14. Gehl, M. et al. Operation of high-speed silicon photonic micro-disk modulators at cryogenic temperatures. Optica 4, 374 (2017). 15. Harris, N. C. et al. Efficient, Compact and Low Loss Thermo-Optic Phase Shifter in Silicon. Opt. Express 22, 83 -- 85 (2014). 16. Alexander, K. et al. Nanophotonic Pockels modulators on a silicon nitride platform. Nat. Commun. 9, 3444 (2018). 17. Wang, C. et al. Integrated lithium niobate electro-optic modulators operating at CMOS- compatible voltages. Nature 562, 101 -- 104 (2018). 18. Eltes, F. et al. Low-loss BaTiO3-Si waveguides for nonlinear integrated photonics. ACS Photonics 3, 1698 -- 1703 (2016). 19. Lauermann, M. et al. 40 GBd 16QAM Signaling at 160 Gb/s in a Silicon-Organic Hybrid Modulator. J. Light. Technol. 33, 1210 -- 1216 (2015). 20. Wang, C., Zhang, M., Stern, B., Lipson, M. & Loncar, M. Nanophotonic Lithium Niobate Electro-optic Modulators. Opt. Express 26, 1547 -- 1555 (2018). 21. Eltes, F. et al. A BaTiO3-based electro-optic Pockels modulator monolithically integrated on an advanced silicon photonics platform. J. Light. Technol. 37, (2019). 22. Von Hippel, A. Ferroelectricity, domain structure, and phase transitions of barium titanate. Rev. Mod. Phys. 22, 221 -- 237 (1950). 23. He, F. & Wells, B. O. Lattice strain in epitaxial BaTiO3 thin films. Appl. Phys. Lett. 88, 152908 (2006). 24. He, F. et al. Structural phase transition in epitaxial perovskite films. Phys. Rev. B 70, 235405 (2004). 25. Tenne, D. A. et al. Absence of low-temperature phase transitions in epitaxial BaTiO3 thin films. Phys. Rev. B - Condens. Matter Mater. Phys. 69, 2 -- 6 (2004). 26. Cochard, C., Spielmann, T., Bahlawane, N., Halpin, A. & Granzow, T. Broadband characterization of congruent lithium niobate from mHz to optical frequencies. J. Phys. D. Appl. Phys. 50, (2017). 27. Bernasconi, P., Zgonik, M. & Gunter, P. Temperature dependence and dispersion of electro-optic and elasto-optic effect in perovskite crystals. J. Appl. Phys. 78, 2651 -- 2658 (1995). 28. Kay, H. F. & Vousden, P. XCV. Symmetry changes in barium titanate at low temperatures and their relation to its ferroelectric properties. London, Edinburgh, Dublin Philos. Mag. J. Sci. 40, 1019 -- 1040 (1949). 29. Damas, P. et al. Wavelength dependence of Pockels effect in strained silicon waveguides. Opt. Express 22, 1693 -- 1696 (2014). 30. Lu, H. et al. Enhanced electro-optical lithium niobate photonic crystal wire waveguide on a smart-cut thin film. Opt. Express 20, 2974 -- 2981 (2012). Acknowledgements This work has received funding from the European Commission under grant agreements no. H2020-ICT-2015-25- 688579 (PHRESCO) and H2020-ICT-2017-1-780997 (plaCMOS), from the Swiss State Secretariat for Education, Research and Innovation under contract no. 15.0285 and 16.0001, from the Swiss National Foundation project no 200021_159565 PADOMO, from EPSRC grants EP/L024020/1, EP/M013472/1, and EP/K033085/1, the UK EPSRC grant QuPIC (EP/N015126/1), and ERC grant 2014- STG 640079. JB thanks Dr. Döndü Sahin for her assistance with the experimental setup. Methods Device fabrication. Single crystalline BaTiO3 was deposited on top of an epitaxial 4-nm-thick STO seed layer by molecular beam epitaxy on 8" SOI wafers with 220-nm-thick device silicon layer for SiN-based devices, and on 2" SOI wafers with 100-nm-thick device silicon for Si-based devices, following a process described elsewhere7. Direct wafer bonding was used to transfer the BaTiO3 and device Si layers onto high-resistivity Si wafers capped with a 3-µm-thick thermal oxide. Specifics of the direct wafer bonding process can be found in ref. 7. For SiN-based waveguides the device Si layer was removed by dry etching, followed by chemical vapor deposition of SiN. The waveguide layer (Si or SiN) was patterned by dry etching. After waveguide patterning, a combination of SiO2 cladding deposition, via etching, and metallisation was used to form the final cross-section. Intermediate annealing steps were used to reduce propagation losses. The SiN-based waveguides use an 80-nm-thick BaTiO3 layer, and 150-nm-thick SiN layer. The strip-waveguide width is 1.1 µm. The electrode-to-electrode gap is 9 µm. The Si strip-waveguides were fabricated using 225-nm-thick BaTiO3 and 100-nm-thick Si. The waveguide width is 0.75 µm, and the electrode-to-electrode gap is 2.3 µm. Cryogenic measurements. The cryogenic electro-optic measurements were performed in a Lakeshore CPX cryogenic probe station, fitted with RF (40 GHz BW, K-type connectors) and optical feed-throughs. DC and RF signals were applied to the devices using RF probes, and optical coupling was achieved using a fibre array with polarisation maintaining fibres for 1550 nm. A tuneable laser (EXFO T100S-HP) and power meter were used to record transmission spectra (EXFO CT440). The cryogenic electrical measurements were performed in a Janis cryogenic probe station equipped with DC probes. Current-voltage and capacitance-voltage measurements were performed using a parameter analyser. Both cryogenic probe stations were cooled by liquid helium to a base temperature of 4.2 K. DC EO characterisation. The DC electro-optic response was extracted by applying a voltage to the electrodes of a racetrack resonator and measuring the shift in resonance wavelength (Δλ), compared to the unbiased case, as a function of the applied voltage. From the measured wavelength shift, the change in BaTiO3 refractive index (ΔnBTO) can be estimated as ∆𝑛BTO = 𝜆0 ∙ Δ𝜆 𝐹𝑆𝑅 ∙ 𝐿E ∙ 𝛤BTO where ΓBTO is the optical confinement in BaTiO3, FSR is the free spectral range of the resonator, LE is the electrode length, and λ0 is the resonance wavelength with no voltage.7 The effective Pockels coefficient, reff, was then determined according to the procedure described in SN 2. RF frequency response. To measure the EO frequency response (EO S21) a vector network analyser (VNA, Keysight PNA 50 GHz) was used to apply the electrical stimulus to a BaTiO3 ring modulator. The modulated optical signal was applied to a photodiode (Newport 1024) and the response recorded by the VNA. Electrical calibration was performed before the measurement, and the response of the photodetector was compensated for the data analysis. While the VNA could generate signals up to 50 GHz, the bandwidth of the photodetector was 26 GHz, which in combination with large frequency-dependent electrical losses in the cryogenic probe station (SN 6) makes it impossible to measure the bandwidth beyond 30 GHz. Devices for data modulation. For data modulation experiments, devices with Si strip-waveguide were used. The racetrack resonator that was used had a bending radius of 15 µm and straight sections of 30 µm. Data modulation experiments. The electrical signal was generated using an arbitrary waveform generator. A pseudo-random bit stream of 27-1 bits was used for modulation. The electrical signal was pre-distorted to compensate for the finite time-response of the electrical signal path (SN 6). The signal was amplified using a RF amplifier and sent to the cryogenic setup, with an estimated voltage swing on the device of 1.7 V (SN 6). A Pritel FA-23 EDFA was used to amplify the modulated optical signal which was applied to a photo diode and recorded on an oscilloscope. Figure 1. BaTiO3 electro-optic device concept. a, Schematic cross-section of the devices. A silicon or silicon nitride layer forms a strip-waveguide on top of an BaTiO3 layer. Lateral electrodes fabricated with W are used to apply an electric field across the BaTiO3. The devices are embedded in SiO2 layers on top of silicon substrates. b, Optical mode simulation of the SiN waveguide geometry and c the Si waveguide geometry, showing an optical confinement in BaTiO3 of 18 % and 41 % respectively. d, Optical micrograph of racetrack resonator devices used to characterise the nonlinear optical properties of BaTiO3. e, Characterisation principle of resonant electro-optic switches, showing example data of a shifted resonance. The shift in resonance wavelength is measured for an applied electric field and converted to the material properties of BaTiO3 (see Methods). Figure 2. Electro-optic and electrical response of BaTiO3-based optical switches at 4 K. a, Refractive index change of BaTiO3 as a function of applied electric field for a device in the 11.25° direction (as defined in b). The hysteretic behaviour originates from ferroelectric domain switching in the BaTiO3, as shown schematically (top). b, Angular anisotropy of the effective Pockels coefficient in BaTiO3. The angle is defined relative to the BaTiO3<100> direction. The same anisotropy as for BaTiO3 at room temperature is observed but with reduced magnitude. The error bars show the combined standard error of the fit and from averaging measurements of multiple devices with the same orientation. c, Electro-optic S21-parameter of BaTiO3 ring resonator showing a flat response up to a record frequency of 30 GHz at 4 K. d, Current measured as a function of electric field across the BaTiO3 layer showing extremely low current flowing through the material. The current is dominated by capacitive charging together with ferroelectric switching current resulting in a peak (SN 3). e, Capacitance as a function of electric field, showing characteristic ferroelectric hysteresis and field-dependent permittivity. Figure 3. Temperature dependence of the Pockels effect in BaTiO3. The effective Pockels coefficient along different crystal orientations at temperatures from 4 K to 340 K. The peak around 240 K is the signature of a phase transition in BaTiO3. A second, field-induced phase transition occurs around 100 K, causing a sharp drop of reff in 90° devices (indicated by horizontal dashed lines). This phase transition is also evident in the qualitative evaluation of the optical response (SN 4). The grey areas indicate the temperature ranges of the respective phase transitions. The error bars show the standard error of the fit used to extract the Pockels coefficients (SN 2). Figure 4. Demonstration of low-power switching and high-speed data modulation with BaTiO3-based devices at 4 K. a, Schematic of Mach- Zehnder (MZ) configuration used to switch between two ports. The inset shows the waveguide cross-section. b, Transmission from both ports of a MZ switch as a function of applied electric field, along with the static power consumption. When fully switching between outputs, less than 10 pW static power is consumed, and only 30 pJ of dynamic energy. c, Schematic of the experimental setup for data modulation. The electrical signal was amplified to compensate for losses into the cryogenic probe station, and then applied to the device. The modulated optical signal was detected using a photodiode and recorded on a high-speed oscilloscope. The inset show the waveguide cross-section. d, Eye diagram recorded at 20 Gbps with Vpp = 1.7 V, corresponding to modulation energy of 45 fJ/bit.
1911.05046	1	1911	2019-11-12T18:09:33	Tailored Molybdenum Carbide Properties and Graphitic Nano Layer Formation by Plasma and Ion Energy Control during Plasma Enhanced ALD	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	We demonstrate the extensive study on how film density and crystallinity of molybdenum carbide ($MoC_{x}$) can be tailored during plasma-enhanced ALD (PEALD) by controlling either the plasma exposure time or the ion energy. We investigated $MoC_{x}$ films grown using $Mo(^tBuN)_2(NMe_2)_2$ as the precursor and $H_2/Ar$ plasma as the co-reactant at temperatures between 150{\deg}C and 300{\deg}C. We discover a threshold for graphitic layer formation at high mean ion energies during the PEALD cycle. The supplied high energy dose allows for hybridised $sp^{2}$ carbon bonds formation, similar to high temperature annealing. The graphitisation of the $MoC_{x}$ surface takes place at temperature of 300$^{\circ}C$. The graphitic film show a (101) plane diffraction peak with dominant intensity in XRD, and a typical $sp^{2}$ C1s peak along with carbidic metal in XPS measurements. Surface roughness of the film lowers significantly at the graphitisation regime of deposition. This low temperature graphitisation by high energy plasma ions during PEALD shows a great promise to advancing graphene and graphite composites at low temperature by PEALD for future applications.	physics.app-ph	physics	Tailored Molybdenum Carbide Properties and Graphitic Nano Layer Formation by Plasma and Ion Energy Control during Plasma Enhanced ALD Eldad Grady,1 Marcel Verheijen,1 Tahsin Faraz,1 Saurabh Karwal,1 W.M.M. Kessels,1 and Ageeth A. Bol1 1Department of Applied Physics, Eindhoven University of Technology, Den Dolech 2, P.O. Box 513, 5600 MB Eindhoven, The Netherlands Abstract We demonstrate the extensive study on how film density and crystallinity of molybdenum carbide (M oCx) can be tailored during plasma-enhanced ALD (PEALD) by controlling either the plasma exposure time or the ion energy. We investigated M oCx films grown using M o(tBuN )2(N M e2)2 as the precursor and H2/Ar plasma as the co-reactant at temperatures between 150◦C and 300◦C. We discover a threshold for graphitic layer formation at high mean ion energies during the PEALD cycle. The supplied high energy dose allows for hybridised sp2 carbon bonds formation, similar to high temperature annealing. The graphitisation of the M oCx surface takes place at temperature of 300◦C. The graphitic film show a (101) plane diffraction peak with dominant intensity in XRD, and a typical sp2 C1s peak along with carbidic metal in XPS measurements. Surface roughness of the film lowers significantly at the graphitisation regime of deposition. This low temperature graphitisation by high energy plasma ions during PEALD shows a great promise to advancing graphene and graphite composites at low temperature by PEALD for future applications. 9 1 0 2 v o N 2 1 ] h p - p p a . s c i s y h p [ 1 v 6 4 0 5 0 . 1 1 9 1 : v i X r a 1 Preprint Preprint Preprint I. INTRODUCTION Graphene and graphitic based composites are of great interest in research of next gener- ation electronics, protective layer coatings, and flexible electronics [1, 2]. Transition metal of group IV-VI have shown to be highly suitable for the high temperature CVD graphene growth process due to its thermal stability, and low thermal coefficient matching Si sub- strates [3] These characteristics result in lower graphene film stress, as the mismatch between graphene and underlying catalytic substrate is significantly reduced. molybdenum carbide (M oCx) is a refractory metal carbide compound that combines the physical properties of ce- ramics and the electrical properties of metals, with hardness and mechanical strength (300 - 535 GPa), high thermal stability, chemical inertness and a metallic like electrical conductiv- ity [4, 5]. The atomic radii ratio of carbon to molybdenum of 0.556 makes it highly suitable to form interstitial carbides with a mixture of covalent, ionic and metallic bonds between the metal and the carbon atoms which are responsible for its unique set of properties [4]. These characteristics are of great interest for diffusion barriers in ICs, superconductors and various MEMS applications. Furthermore, molybdenum carbide has made a new addition to the 2D material family named MXenes [6]. As an IC diffusion barrier M oCx is ideally amorphous and dense, while for superconductivity the cubic δ − M oC0.75 shows the highest transition temperature (14.3 K) [7] for M oCx films. In order to accommodate the broad spectrum of usage, the ability to separately control the crystallinity and density of M oCx would be an asset for tailoring film properties to the specific application needs. Thus far, various techniques have been implemented to synthesize cubic M oCx, such as high energy ion bombardment of amorphous MoC, CVD, PVD and ultra high pressure synthesis, at high temperatures [8 -- 10]. While some techniques were successful in the synthesis, the high temperatures required make it unsuitable for integration in temperature sensitive devices. Atomic layer deposition (ALD) offers a cyclic soft deposition technique with a broad and low temperature window, submonolayer thickness control due to its self limiting nature, along with high uniformity and conformality. These merits allow for deposition on temper- ature sensitive applications as well as for deposition of layers sensitive for the kinetic ion impact during sputtering. While substantial work has been done on various molybdenum compounds such as molyb- 2 Preprint Preprint Preprint denum oxide [11] and nitride , so far, very little work has been done on ALD of M oCx. Recently, a PEALD process using M o(tBuN )2(N M e2)2 precursor with a H2/N2 plasma as co-reactant was developed with the aim of achieving low resistivity films [12]. Bertuch et al. demonstrated the deposition of a range of M oCx − M oCxNy compounds depending on the H2/N2 ratios used. A predominantly molybdenum carbide material with nitrogen impuri- ties was produced with pure H2 plasma at 150 ◦C, which showed the lowest (170 µΩ − cm) resistivity. Here, we present the first extensive study on how film density and crystallinity of M oCx can be tailored independently during plasma-enhanced ALD (PEALD) by control- ling either the plasma exposure time or the ion energy [13 -- 15]. We investigated M oCx films grown using (tBuN )2(N M e2)2M o as the precursor and H2/Ar plasma as the co-reactant at temperatures between 150◦C and 300◦C. Additionally, we present the effects of enhanc- ing the impinging ion energy on the physical and chemical properties of M oCx thin films, controlled via RF biasing of the substrate table during the plasma step of the PEALD. We discover a threshold for graphitic layer formation at high mean ion energies during the PEALD cycle. While plasma assisted graphene fabrication has attracted significant interest due to the low temperature synthesis [16], to the best of our knowledge this is the first report of nano graphitic layers formation at low temperature using PEALD. II. EXPERIMENTAL METHODS M oCx thin films have been deposited by plasma enhanced atomic layer deposition (PEALD) at various temperatures and plasma conditions. In this part the deposition process is explained and the analysis techniques described. A. Film depositions Plasma enhanced atomic layer deposition was performed on 100 mm Si (100) wafers coated with 450 nm of thermally grown SiO2. The depositions were performed in an Oxford instruments FlexAL2 ALD reactor, which is equipped with an inductively coupled remote RF plasma (ICP) source (13.56 MHz) with alumina dielectric tube. The reactor was pumped down to a base pressure of 1 · 10−6T orr with a turbomolecular pump. The samples were loaded and unloaded from a low pressure loadlock chamber, allowing for a cooldown after deposition in vacuum conditions. The reactor's deposition table was set to temperatures between 150◦C and 300◦C, and the wall temperature to 150◦C. 3 Preprint Preprint Preprint One ALD cycle consists of subsequently a precursor dose step, a purge step, a plasma exposure step and again a purge step. The (M o(tBuN )2(N M e2)2 precursor (98%, Strem Chemicals) was stored in a stainless steel container, which was kept at 50 ◦C. The precursor container was bubbled by an argon flow of 50 sccm during the precursor dose step, while flowing argon at the same rate from the ICP chamber to avoid deposition in the ICP tube. The total pressure in the reactor during the 6 seconds long precursor injection was set to 200 mTorr. During the plasma exposure step, the ICP RF power was set to 100W. H2/Ar with 4:1 ratio and total flow rate of 50 sccm was fed from the top ICP tube and the plasma was ignited for the desired exposure time, typically 20 to 80 seconds. Automatic pressure control (APC) valves were fully opened during plasma exposure to reduce the pressure to 7 mTorr. Purge steps (4 to 5 seconds) were set after precursor and plasma half cycles to evacuate residual precursor gas, reaction byproducts and plasma species with 100 sccm argon flow and open APC valves, at pressure 25 mTorr. B. Film analyses Film thickness and optical properties of the deposited films have been studied with a J.A. Woollam UV-spectroscopic ellipsometer (SE) using a model comprising one Drude and two Tauc-Lorenz oscillators. Data was obtained within the range of 190 nm -- 990 nm, and refractive index (n) and extinction coefficient (k) were determined. Growth per cycle (GPC) was calculated using in-situ J.A. Woollam IR-SE every 10 deposition cycles. Optical resistivity was derived from the first term of the Drude oscillator, corresponding to the imaginary part 2. Sheet resistance values of the M oCx films deposited on 450 nm SiO2/Si were determined using a Keithley 2400 SourceMeter and a Signatron probe. Film thickness was evaluated using ex-situ UV-SE, to calculate the electrical resistivity at room temperature. The film composition was analysed by X-ray photoelectron spectroscopy (XPS) with a Thermo Scientific KA1066 spectrometer, using monochromatic Al Kα x-rays with an energy of 1486.6 eV. The films have been sputtered with Ar+ ion gun prior to scans, in order to remove surface oxide and adventitious carbon. A continuous electron flood gun was employed during measurements to compensate for charging. XPS studies have been performed to evaluate the film composition as function of plasma conditions. Three main components were evaluated by their corresponding peaks, namely molybdenum by the mo3d peak, oxygen by the O1s peak and carbon by the C1s peak. The ratio between each peak area 4 Preprint Preprint Preprint to the sum of all peaks area gave the partial atomic percentage of the element. The precursor molecule used for deposition contains 4 nitrogen atoms bonded to the molybdenum atom, therefore nitrogen is expected to be found in the film. However, due to overlapping N1s and Mo3p peaks, deconvolution of the N1s peak is not reliable for low N1s peak intensity. Rutherford backscattering spectroscopy (RBS) and elastic recoil detection (ERD) were performed by AccTec BV, Eindhoven, The Netherlands, using a Singletron with a 2 MeV He+ beam to determine the chemical composition of the films. ERD was performed with the detector at a recoil angle of 30◦. RBS is performed at scattering angles of 150◦ and 105◦. The mass density was calculated using the measured mass density as obtained from RBS/ERD measurements and the film thickness as obtained from SE. RBS measurement complement the XPS estimation of film composition, and specifically provide further information on nitrogen content in the film. Film crystallinity and preferred crystal orientation was studied by Gonio x-ray diffraction. Experiments were conducted with PanAnlytical X'pert PROMRD diffractometer operated using CuKα(λ = 1.54A). A JEOL ARM 200 transmission electron microscope (TEM) at 200 kV was used to analyse the microstructure of films deposited on planar TEM windows. These windows consisted of ∼ 15 nm Si3N4 membranes coated with 5 nm of SiO2 grown using ALD. This ensured a SiO2 starting surface while maintaining transparency to the electron beam. High-angle-annular-dark-field TEM (HAADF-TEM) modes was employed to characterise the samples. III. RESULTS AND DISCUSSION Atomic layer film deposition has been performed with M o(tBuN )2(N M e2)2 precursor and 40/10 sccm of H2/Ar plasma. The growth per cycle (GPC) was measured with an in-situ SE by measuring the film's thickness every 10 cycles. Precursor saturation occurs after 6 seconds of dosing, and plasma time saturation after 20 seconds of exposure. The saturation curve for the precursor and the plasma time performed at 300◦C is presented in figure 1. Typical saturated GPC value is 0.365A. The effect of temperature on the GPC in saturated conditions has been studied as well for 150◦C and 250◦C yielding values of about 0.25A(see figure 2). The significant increase in GPC at 300◦C has been explained elsewhere [11] as precursor decomposition. However, we see no evidence to that effect, as the precursor 5 Preprint Preprint Preprint saturation curve confirms no increase in GPC after 6 seconds dosing. Furthermore, film thickness as measured across a wafer with ex-situ SE, show high uniformity typical for ALD. Additionally, previous work in our group using the same precursor for M oOx deposition showed no decomposition at the same table temperature [12].The characterisation of the M oCx properties with varying temperature provides an insight to observed rise in GPC. A. Plasma Time and Temperature effects on Films M oCx depositions at 150◦C with 20 seconds plasma have produced films with C/Mo ratio of 0.7 as estimated by XPS (see table I). The oxygen content was estimated to be 12.4% at 150◦C with 20 seconds plasma. Increasing the plasma exposure time to 80 seconds shows a slight decrease in C/Mo ratio to 0.68 with a sharp decrease of oxygen content to 5.7%, as demonstrated in figure 3. The growth per cycle (GPC) for both cases remains unchanged at 0.25A. All films deposited at 150◦C show no diffraction peaks in XRD measurements and are deemed to be completely amorphous. Deposition at 250◦C with 20 seconds plasma produced a film with 5% hydrogen, 5% nitrogen and C/Mo ratio of 0.63, however no [O] was detected in the bulk. Hydrogen content could originate both from the plasma coreactant gas, and the [H] rich precursor molecule, while nitrogen content originates solely from the precursor molecule, C12H30N4M o, that contributes 4 [N] atoms. Each of the [N] atoms is directly bonded to the [Mo], and to CH ligands. The H2/Ar plasma is able to break most N-Mo bonds with 20s plasma, explaining the relatively low rate of [N] in the film. The high content of [C] in the film on the other hand, suggests that the removed ligands are redeposited during the plasma step, ultimately producing a M oCx film. RBS measurements reveal a film mass density of 8.9 g/cm3, and [Mo] GPC of 1.33 nm2×cycle. The XRD pattern show diffraction peaks corresponding to the (111) and (220) lattice planes of the atom atom nm2×cycle and [C] of 0.83 delta-cubic M oCx. The (111) peak width appears broad and distorted, along with several sharper smaller peaks between 28-33◦ 2θ. With 80 seconds plasma at 250◦C the C/Mo ratio reduced from 0.63 to 0.56 and the nitrogen and hydrogen content from 5% to below detection limits (±1% and ±5% respectively), as table II shows. The removal of [N] content supports the role of the H2/Ar plasma in removal of N-Mo bonds, and redeposition of [C] from the ligands, allbeit [C] removal effects are revealed with the prolonged plasma exposure. As no [O] was present in the bulk, XPS shows no significant change in O1s peak (see figure 3) with 6 Preprint Preprint Preprint prolonged plasma, as oppose to film deposited at 150◦C. Moreover, the film's mass density has increased with the longer plasma exposure from 8.9 to 9.2 g/cm3. Film deposition at 250◦C and 80 seconds plasma exhibits a single diffraction peak of the (111) lattice plane, while other diffraction peaks are completely suppressed (see figure 4). The broad peak width indicates small order crystalline domains, embedded in mostly amorphous domains. GPC for both conditions at 250◦C is at 0.25A, similar to deposition at 150◦C within error margins. The growth of atoms per nm2cycle show a relatively unchanged growth in Mo atoms (1.33 and 1.35 for 20s and 80s respectively), while the growth of C was reduced from atom nm2×cycle to 0.76 atom 0.83 nm2×cycle which correlates to the slight increase in density. The removal of [C] content indicates more efficient removal of ligands during deposition with prolonged plasma time. Film deposition at 300◦C with 20 seconds plasma shows a significant increase in GPC from 0.25A for lower temperatures to 0.37A. RBS profiling reveal an increase in C/Mo ratio from 0.63 to 0.92 upon increasing the temperature from 250◦C to 300◦C. Atomic nm2×cycle at 300◦C), growth per cycle a slight increase in Mo (1.35 nm2×cycle. Moreover, 4% [N] content was detected in the film by RBS and 8% [H] by ERD measurements. The increase in film while [C] content increased by 60% from 0.83 to 1.34 nm2×cycle at 250◦C to 1.45 atom atom atom impurities content reflected on the mass density, which sharply declined from 8.9 g/cm3 to 7.0 g/cm3 with the increase in temperature. Prolonging the plasma exposure time to 80 seconds results in decreased C/Mo from 0.92 to 0.78, a sharper decrease than that observed at 250◦C. Correspondently, [Mo] atoms deposition rate increases from 1.45 to 1.63 nm2×cycle, and [C] decreases from 1.34 to 1.28 nm2×cycle. The increase of [Mo] growth is a factor of 10 higher at 300◦C than for the same plasma time at 250◦C (∆[M o]250 = 0.02 at./nm2cycle and ∆[M o]300 = 0.2 at./nm2cycle). Additionally, [N] and [H] content decreased below detection level for 80 seconds plasma. The lower C/Mo ratio and removal of additional impurities atom atom reflected in a higher mass density of 8.0 g/cm3. The film deposited with 80 seconds plasma show the dominant diffraction peak at 36.6◦ 2θ of the (111) plane. Comparison of film deposited with 80 seconds plasma exposure at different temperatures, shows C/Mo ratio of 0.7 at 150◦C which is reduced to 0.56 at 250◦C and then increased to 0.78 at 300◦C. By prolonging the plasma exposure time, the total ion energy dose to the deposition substrate is increased linearly with time (eVion/s×P lasmaseconds). The increased energy dose resulted in higher mass density and correspondently lower C/Mo ratio and mitigation of film impurities below detection levels. This reflected on the lowered film resistivity, with two fold decrease 7 Preprint Preprint Preprint from 272 µΩ − cm to 143 µΩ − cm at 300◦C for 20s and 80s plasma respectively. However, the effect on film crystallinity is not significant. Generally, the films deposited at 300◦C show higher peak intensities and narrower peak widths than films at 250◦C which are mostly amorphous. Hence, more crystalline material is present with larger crystal grains at higher temperature, whose effect on crystallinity is more significant than plasma exposure time. B. Biased Substrate effects on ALD Films The total energy dose to the deposition substrate can be alternatively altered by increas- ing the ion energy. By applying radio-frequency (RF) bias voltage to the substrate we can control the energy of ions impinging on the surface. An RF bias with time average voltage (Vbias) between -100V to -200V was applied to study the effects on the film properties. For these experiments we use 20 seconds of total plasma time, for a comparison with the saturated non biased plasma time. The plasma during biased ALD is comprised of 10s non biased followed by 10s biased plasma. The mean energy of Eion = e · Vsheath = e · (Vplasma − Vsurf ace) Where Vplasma is the ion energy for 0V bias, measured to be 25eV for 100W plasma power the ion energy is calculated from the following equation: at 7mTorr pressure, and Vsurf ace is the time averaged bias voltage. Thus for an applied time averaged Vbias ( < Vbias >) of -100V, the total mean ion energy is 125 eV. All bias experi- ments performed were done at 300◦C, as the largest fluctuation in film density and content occurs at this temperature. Figure 5 shows the effect of applied bias on the GPC. When −100Vbias was applied the GPC declined sharply from 0.37A at 300◦C with < 0Vbias > to 0.22A. RBS measurements reveal the cause for this significant decline in a lowered [Mo] and [C] GPC upon applying bias voltage. [Mo] GPC declined from 1.45 to 1.02 nm2×cycle and [C] with over 60% decline from 1.34 to 0.81 nm2×cycle. The corresponding C/Mo ratio reduced from 0.92 for the non biased case to 0.79 with −100Vbias. An effect of M oCx densification is noted, with mass density measured at 8.2 g/cm3 for −100Vbias compared to a 7.0 g/cm3 for < 0Vbias >. atom atom Physical alterations are also first seen with the application of surface bias voltage, giving the typical cubic δ − M oC0.75 structure with diffraction peaks at 35.75◦ and 41.42◦ 2θ (see 8 Preprint Preprint Preprint figure 7). The dominant peak is the (111) plane, as the (200) plane peak almost completely suppressed. The peak intensity increased by 2 orders of magnitude upon applying −100Vbias, and peak FWHM decreased by half (0Vbias = 1.03 ◦2θ and −100Vbias = 0.49 ◦2θ). Crystallite size has been calculated by fitting the (111) and analysing the peak broadening using Scherrer equation: τ = K · λ/(β · cos θ) The XRD data points to a highly crystalline material with crystallite size doubling in comparison for non biased deposition. Crystallite size for 20s pl without bias is calculated to be 9.0 nm while extending the plasma time yields a crystallite size of 7.6 nm, however this slight decrease could be attributed to increased strain in the denser film with 80s pl. When −100Vbias bias is applied, crystallite doubles in size to 19.4 nm. As can be seen in figure 8, the plan view high angle annular dark-field (HAADF) TEM images support these findings. The size of the visible structured material is significantly increased and voids previously seen for non biased deposition not detectable with −100Vbias. C. Graphitisation during PEALD With increased Vbias, further physical alterations and formation of graphitic nano layers are revealed in the M oCx film. With < −135Vbias > applied, the (111) lattice peak of the cubic δ − M oC0.75 is diminished, and new diffraction peak appears corresponding to crys- talline graphite. The emerging peak at 44.5 2θ◦ match the (101) graphite lattice plane and is the dominating peak intensity, indicating strong graphitization with higher ion energies (Eion = 155eV ). The observed emerging phase is consistent with increasing ion energy fur- ther (−187Vbias, Eion = 212eV ), however the (101) peak intensity is somewhat diminished in comparison to the previous bias step at −130Vbias. The GPC drops continuesly with in- creasing Vbias, from 0.22A for -100V to 0.15A and 0.14A for -135V and -187V respectively. While C/Mo ratio dropped with the application of −100Vbias from 0.92 for non biased ALD to 0.79, this ratio has stagnated for higher applied Vbias, and the C/Mo ratio remained at ∼0.80. 9 Preprint Preprint Preprint D. Plasma Time and Bias Effects on Film Resistivity The effect of plasma time on M oCx film resistivity has been studied using four-point probe (4PP) measurements. Film of comparable thickness of 30 nm have been deposited at 300◦C with plasma time ranging from 20 seconds (begin of saturation) up to 80 seconds. Difference in film thickness is within SE error margin. As shown in figure 10, a significant decrease in resistivity was measured with increase of plasma time. For 20 seconds plasma resistivity was 272 µΩ − cm and a decrease to 242 µΩ − cm is seen with double the plasma time (40 seconds). Further increase of plasma time to 60 seconds decreased resistivity to 181 µΩ − cm, and at the final plasma time of 80 seconds 143 µΩ − cm was measured. By increasing plasma time from 20 s to 40 s, XPS measurements revealed a reduction of C/Mo ratio from 0.96 to 0.85. With a total 80 s plasma time, the C/Mo ratio decreased further to 0.79. As was demonstrated above, the reduction in C/Mo ratio increased mass density of the M oCx film, and the higher ratio of [Mo] to [C] improves its electrical conductivity. RF bias of the substrate table of average −100Vbias has been applied for the duration of 10 seconds, after 10 seconds of non-biased plasma exposure. Film resistivity was measured for films of comparable thickness of 30 nm. We correlate the change in resistivity to GPC and density rates to gain further indications of chemical and physical effects to the film. Figure 5 depicts GPC for biased ALD, and figure 12 the resistivity for the corresponding bias voltages. With −100Vbias, film resistivity was decreased to 144 and GPC ratio was decreased from 0.36 (for 0Vbias)) to 0.22. and 146 µΩ − cm respectively. The GPC ratio was decreased from 0.22 A for -100V to 0.16A for -135V < Vbias >. As shown above, the C/Mo reduced upon application of bias voltage from 0.92 to ∼ 0.8, which reflected in higher mass density. The produced denser and more metallic film translated to better electrical conductivity. Increasing the bias voltage to -187V gives a slight rise in resistivity to 156 µΩ − cm and further decrease in GPC to 0.15A. At -210V < Vbias > a significant increase to 200 µΩ − cm is shown with 0.14A GPC. IV. CONCLUSIONS We presented here the ability to tailor M oCx film by controlling the ion energy and the total energy dose during PEALD. Additionally, we discover graphitic layer formation 10 Preprint Preprint Preprint during the PEALD cycle with mean ion energies between 150eV to 220eV. sp2 hybridised carbon bonds appear to form during the high energy dose to the film surface with average bias voltage is increased to -135Vbias. Furthermore, the effects of plasma time exposure, and the ion energy in the plasma during PEALD modification of the M oCx film electrical, chemical and physical properties were investigated. Both prolonged plasma and biasing are successful in mitigating [N] impurities in the film, and reducing C/Mo ratio. Consequently, a significantly lowered film resistivity was attained of 143 µΩ − cm. Increasing the total energy dose, which is a product of the mean Ei and the total plasma exposure time. With grounded substrate table (< 0Vbias >) we can increase the total energy dose by prolonging the plasma time, with mean ion energy fixed at 25eV. By increasing the total plasma time from 20s to 80s, we increased film density from 7 to 8, while retaining the film physical properties. With a fixed plasma exposure time of 20 seconds, we increased the mean ion energy by substrate biasing, thus increasing the total energy dose as well. We demonstrated an increase in mass density with increased total energy dose, while no significant change to crystallinity for the same Ei. With substrate bias we elevated the impinging ion energies, and demonstrated physical effects to the film. With mean ion energy of 125eV, a highly crystalline and dense film was fabricated corresponding the cubic δ − M oC0.79. We can achieve the same high density, low resistivity film in 20s plasma exposure time by biasing, as with long plasma exposure time of 80s without biasing. However, by increasing the mean ion energy, we see physical alterations similar to annealing effects on film, substituting the thermal energy with plasma ion energy. With mean ion energy of ∼ 200eV we see the most crystalline graphitic layer formation, indicating an optimal energy region for graphene formation. The diminishing graphite crystallinity beyond mean ion energy of 200 eV indicates deterioration of the graphite top layer by high energy ArH + ions. The increase in resistivity with further decrease in GPC for < Vbias > beyond -135V suggest microstructural changes to the film, more so for the higher -210V. These changes could be in form of voids due to increased ion energy bombardment. The bi-modal distribution of ion energy shows that part of the incident ion have energies higher than -210 eV. The diminished intensity and increased resistivity suggest that these high energy ions at the higher end of the spectrum, could have a sputtering effect on the graphitic top layer when impinging on the surface. Further work needs to be done in optimising the graphene layer by exploring ideal RF bias plasma exposure time, and mean ion energy, as the two determine the total energy dose to the film 11 Preprint Preprint Preprint surface and quality of graphitisation. We postulate, that the high energy ions are able to break Mo-C bonds and achieve a dominantly graphitic film in the [C] rich atmosphere. A trade-off between ion energy and the plasma exposure time needs to be explored, to fine tune the graphitisation. The applied RF bias voltage effect on film crystallinity and composition is shown to be critical. By controlling the mean ion energy of the impinging ions on the surface of the deposited film, we gain a powerful mean of influence on crystallinity and density alike. These coupled physical and chemical effects are derived from the composition of the coreactant H2/Ar gas in the plasma. [H] interaction with the surface is mainly chemical in nature, while the heavier Ar atoms contribute to the physical effects with their kinetic energy. The produced ArH + ions in the plasma combine both these features, hence the coupled chemical and physical effects. By varying the ion energy, we alter predominantly the kinetic impact effect on film, therefore the sharp changes to the M oCx crystallinity. In conclusion, we demonstrated the control of mass density independent from film crys- tallinity, and a method to control both features by PEALD, without post deposition thermal annealing treatments, or high temperature deposition. Furthermore, this work paves the way to fabrication of graphene layers with PEALD, crucial for the production of stacked layers, with atomic precision and high film purity. Moreover, the fabrication can be performed in temperatures as low as 300◦C. This effect could play a significant role in various applications that require low thermal budget. The ability of control in PEALD of M oCx opens a window for the integration of tailor-made thin M oCx, as well as low temperature graphene films in a wide range of applications previously unavailable. Acknowledgements E. Grady would like to acknowledge the financial support of the Dutch Technology Foun- dation STW (project number 140930), which is part of the Netherlands Organisation for Scientific Research (NWO). E. Grady thanks Cristian van Helvoirt for XRD measurements, 12 Preprint Preprint Preprint Jeroen Gerwen and Janneke Zeebregts for their technical support. [1] Jingjie Wu, Lulu Ma, Atanu Samanta, Mingjie Liu, Bo Li, Yingchao Yang, Jiangtan Yuan, Jing Zhang, Yongji Gong, Jun Lou, et al. Growth of molybdenum carbide -- graphene hybrids from molybdenum disulfide atomic layer template. Advanced Materials Interfaces, 4(4):1600866, 2017. [2] Jorge Guardia-Valenzuela, Alessandro Bertarelli, Federico Carra, Nicola Mariani, Stefano Biz- zaro, and Raul Arenal. Development and properties of high thermal conductivity molybdenum carbide-graphite composites. Carbon, 135:72 -- 84, 2018. [3] Zhiyu Zou, Lei Fu, Xiuju Song, Yanfeng Zhang, and Zhongfan Liu. Carbide-forming groups ivb-vib metals: a new territory in the periodic table for cvd growth of graphene. Nano letters, 14(7):3832 -- 3839, 2014. [4] Hugh O. Pierson. Handbook of Refractory Carbides and Nitrides. William Andrew Publishing, 1996. [5] Joseph Halim, Kevin M. Cook, Michael Naguib, Per Eklund, Yury Gogotsi, Johanna Rosen, and Michel W. Barsoum. X-ray photoelectron spectroscopy of select multi-layered transition metal carbides (mxenes). Applied Surface Science, 362:406 -- 417, 2016. [6] Babak Anasori, Maria R Lukatskaya, and Yury Gogotsi. 2d metal carbides and nitrides (mxenes) for energy storage. Nature Reviews Materials, 2(2):16098, 2017. [7] Clastin I. Sathish, Yuichi Shirako, Yoshihiro Tsujimoto, Hai L. Feng, Ying Sun, Masaki Akaogi, and Kazunari Yamaura. Superconductivity of δ − moc0.75 synthesized at 17gpa. Solid State Communications, 177:33 -- 35, 2014. [8] E. V. Clougherty, K. H. Lothrop, and J. A. Kafalas. A New Phase formed by High-Pressure Treatment : Face-centred Cubic Molybdenum Monocarbide. Nature (London), 191:1194, September 1961. [9] Jae Sung Lee, ST Oyama, and M Boudart. Molybdenum carbide catalysts: I. synthesis of unsupported powders. Journal of Catalysis, 106(1):125 -- 133, 1987. [10] F Okuyama, Y Fujimoto, S Kato, and T Kondo. Growth of molybdenum carbide particles from an amorphous phase induced by ion bombardment. Applied Physics A: Materials Science & Processing, 38(4):275 -- 279, 1985. 13 Preprint Preprint Preprint [11] Martijn F. J. Vos, Bart Macco, Nick F. W. Thissen, Ageeth A. Bol, and W. M. M. (Erwin) Kessels. Atomic layer deposition of molybdenum oxide from (ntbu)2(nme2)2mo and o2 plasma. Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, 34(1):01A103, 2016. [12] Adam Bertuch, Brent D. Keller, Nicola Ferralis, Jeffrey C. Grossman, and Ganesh Sundaram. Plasma enhanced atomic layer deposition of molybdenum carbide and nitride with bis(tert- butylimido)bis(dimethylamido) molybdenum. Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, 35(1):01B141, 2017. [13] HB Profijt, MCM Van de Sanden, and WMM Kessels. Substrate-biasing during plasma- assisted atomic layer deposition to tailor metal-oxide thin film growth. Journal of Vacuum Science & Technology A: Vacuum, Surfaces, and Films, 31(1):01A106, 2013. [14] Tahsin Faraz, Karsten Arts, Saurabh Karwal, Harm Knoops, and Erwin Kessels. Energetic ions during plasma-enhanced atomic layer deposition and their role in tailoring material prop- erties. Plasma Sources Science and Technology, 2018. [15] Tahsin Faraz, Harm CM Knoops, Marcel A Verheijen, Cristian AA Van Helvoirt, Saurabh Karwal, Akhil Sharma, Vivek Beladiya, Adriana Szeghalmi, Dennis M Hausmann, Jon Henri, et al. Tuning material properties of oxides and nitrides by substrate biasing during plasma- enhanced atomic layer deposition on planar and 3d substrate topographies. ACS applied materials & interfaces, 10(15):13158 -- 13180, 2018. [16] Igor Levchenko, Kostya (Ken) Ostrikov, Jie Zheng, Xingguo Li, Michael Keidar, and Kenneth B. K. Teo. Scalable graphene production: perspectives and challenges of plasma applications. Nanoscale, 8:10511 -- 10527, 2016. 14 Preprint Preprint Preprint XPS 4PP Temperature Plasma Oxygen time (s) (atomic%) (◦C) 150 150 C M o ratio Resistivity µΩ − cm 20 80 12.4 5.6 0.70 0.72 154 145 TABLE I. Atomic % of elements by XPS and electrical resistivity measurements 15 Preprint Preprint Preprint . s e m i t a m s a l p d n a s e r u t a r e p m e t s u o i r a v t a x C o M d e s a i B n o n f o s t n e m e r u s a e m S B R y b y t i s n e d m fi l d n a s t n e m e l e f o % c i m o t A . I I E L B A T y t i s n e D ] C [ S B R ] o M [ ] N [ ) 3 g m c ( ) e l c y m c × o t 2 a m n ( ) e l c y m c × o t 2 a m n ( ) % c i m o t a ( o i t a r o CM o i t a r o CM S P X -- -- 6 8 . 8 2 . 9 7 8 -- -- 3 8 . 0 6 7 . 0 4 3 . 1 8 2 . 1 -- -- 3 3 . 1 5 3 . 1 5 4 . 1 3 6 . 1 -- -- 1 ± 5 1 5 . 3 . l . d < -- -- 3 6 . 0 6 5 . 0 2 9 . 0 8 7 . 0 0 7 . 0 2 7 . 0 4 6 . 0 5 7 . 0 6 9 . 0 1 9 . 0 -- -- 5 ± 5 . l . d < 8 . l . d < . t n e m e l e e h t f o s t i m i l n o i t c e t e d w o l e b : . l . d < 0 2 0 8 0 2 0 8 0 2 0 8 ) % c i m o t a ( ) s ( e m i t D R E ] H [ a m s a l P e r u t a r e p m e T ) C ◦ ( 0 5 1 0 5 1 0 5 2 0 5 2 0 0 3 0 0 3 16 Preprint Preprint Preprint ERD Temperature Plasma [H] [N] (◦C) time(s) (atomic%) (atomic%) 0 0 -100 20 80 20 8 4 0.92 < d.l. < d.l. 0.78 < d.l. < d.l. 0.79 1.45 1.63 1.02 1.34 1.28 0.81 < d.l.: below detection limits of the elements. RBS [Mo] [C] Density C M o ratio ( atom ( nm2×cycle ) atom nm2×cycle ) ( g cm3 ) 7.0 8.0 8.2 TABLE III. Atomic % of elements and film density by RBS measurements of non Biased M oCx at various temperatures and plasma times. 17 Preprint Preprint Preprint FIG. 1. Growth per cycle (GPC) of M oCx deposited at 300◦C determined using a in-situ SE, as a function of (a) precursor dose time (b) plasma exposure time. The lines serve as a guide to the eye 18 Preprint Preprint Preprint FIG. 2. Growth per cycle (GPC) during initial 500 cycles. M oCx films deposited using (M o(tBuN )2(N M e2)2 precursor andH2/Ar plasma (without substrate biasing) as a function of deposition temperature 19 Preprint Preprint Preprint FIG. 3. XPS core level spectra of the O1s peak for ∼22 nm M oCx films deposited at (a) 150 ◦C and (b) 250 ◦C. The red and blue lines represent films deposited using H2/Ar plasma exposure times of 20s and 80s, respectively 20 Preprint Preprint Preprint FIG. 4. θ-2θ x-ray diffractograms for 30 nm MoC films deposited using different temperatures and H2/Ar plasma exposure conditions (different plasma exposure times without biasing) 21 Preprint Preprint Preprint FIG. 5. Growth per cycle (left axis) and C/Mo ratio (right axis) of M oCx films deposited at 300◦C as a function of the time-averaged bias voltage applied during the H2/Ar plasma exposure step. 22 Preprint Preprint Preprint FIG. 6. θ-2θ x-ray diffractograms for 30 nm M oCx films deposited at 300◦C without any substrate biasing with 20, and with a time-averaged bias voltage ranging from -100V to -187V applied during the last half (10 s) of the 20 s H2/Ar plasma exposure step. 23 Preprint Preprint Preprint FIG. 7. θ-2θ x-ray diffractograms for 30 nm MoCs films deposited at 300◦C without any substrate biasing with 20 and 80 s plasma, and with a time-averaged bias voltage of -100V applied during the last half (10 s) of the 20 second H2/Ar plasma exposure step. 24 Preprint Preprint Preprint FIG. 8. Plan-view high angle annular dark-field (HAADF) TEM images for molybdenum carbide films deposited at 300C with (a) 0 V or no bias, 20 s plasma and (b) 0 V 80 s plasma, and (c) -100 V (d) -135 V and (e) -210 V bias applied during the last half (10 s) of the 20 s H2/Ar plasma exposure step. 25 Preprint Preprint Preprint FIG. 9. Plan-view high angle annular dark-field (HAADF) TEM images for molybdenum carbide films deposited at 300C with (a) 0 V or no bias, 20 s plasma and (b) 0 V 80 s plasma, and (c) -100 V (d) -135 V and (e) -210 V bias applied during the last half (10 s) of the 20 s H2/Ar plasma exposure step. 26 Preprint Preprint Preprint FIG. 10. Resistivity of ∼ 30 nm M oCx films deposited at 300◦C expressed as a function of theH2/Ar plasma exposure time. The resistivities were measured using four point probe. 27 Preprint Preprint Preprint FIG. 11. Resistivity of ∼ 30 nm M oCx films deposited at 300◦C with time-averaged bias voltages ranging from 0 V (i.e., no biasing) to -210 V applied during the last 10 seconds of the 20 seconds H2/Ar plasma exposure step. The resistivities measured using four point probe are denoted by the stars while those derived using spectroscopic ellipsometry are denoted using squares. Centre: resistivity of ∼ 30 nm M oCx films deposited at 300◦C expressed as a function of the H2/Ar plasma exposure conditions. 28 Preprint Preprint Preprint FIG. 12. Properties of ∼ 30 nm M oCx films deposited at 300◦C with time-averaged bias voltages ranging from 0 V (i.e., no biasing) to 210 V applied during the last 10 seconds of the 20 seconds H2/Ar plasma exposure step. Graphitisation regime begins with mean ion energies higher than 125eV (-100Vbias), and shows a drastic decrease in surface roughness, and a peak in mass density at -135Vbias. 29
1902.02614	1	1902	2019-02-07T13:32:24	Exploring co-sputtering of ZnO:Al and SiO2 for efficient electron-selective contacts on silicon solar cells	[ "physics.app-ph" ]	In recent years, considerable efforts have been devoted to developing novel electron-selective materials for crystalline Si (c-Si) solar cells with the attempts to simplify the fabrication process and improve efficiency. In this study, ZnO:Al (AZO) is co-sputtered with SiO2 to form AZO:SiO2 films with different SiO2 content. These nanometer-scale films, deposited on top of thin intrinsic hydrogenated amorphous silicon films and capped with low-work-function metal (such as Al and Mg), are demonstrated to function effectively as electron-selective contacts in c-Si solar cells. On the one hand, AZO:SiO2 plays an important role in such electron-selective contact and its thickness is a critical parameter, thickness of 2 nm showing the best. On the other hand, at the optimal thickness of AZO:SiO2, the open circuit voltage (VOC) of the solar cells is found to be relatively insensitive to either the work function or the band gap of AZO:SiO2. Whereas, regarding the fill factor (FF), AZO without SiO2 content exhibits to be the optimal choice. By using AZO/Al as electron-selective contact, we successfully realize a 19.5%-efficient solar cell with VOC over 700 mV and FF around 75%, which is the best result among c-Si solar cells using ZnO as electron-selective contact. Also, this work implies that efficient carrier-selective film can be made by magnetron sputtering method.	physics.app-ph	physics	Exploring co-sputtering of ZnO:Al and SiO2 for efficient electron-selective contacts on silicon solar cells Sihua Zhong1,2,, Monica Morales-Masis1,3, Mathias Mews4, Lars Korte4, Quentin Jeangros1, Weiliang Wu1, Mathieu Boccard1, and Christophe Ballif 1 1 École Polytechnique Fédérale de Lausanne (EPFL), Institute of Microengineering (IMT), Photovoltaics and Thin-Film Electronics Laboratory (PV-lab), Rue de la Maladière 71b, CH-2002 Neuchâtel, Switzerland. 2. Shanghai Jiao Tong University, School of Physics and Astronomy, Institute of Solar Energy and Key Laboratory of Artificial Structures and Quantum Control (Ministry of Education), Shanghai 200240, P. R. China. 3. University of Twente, MESA+ Institute for Nanotechnology, 7500 AE Enschede, The Netherlands. 4. Helmholtz-Zentrum Berlin, Institute of Silicon Photovoltaics, Kekuléstraβe 5, D-12489 Berlin, Germany. Corresponding author: [email protected] Abstract In recent years, considerable efforts have been devoted to developing novel electron-selective materials for crystalline Si (c-Si) solar cells with the attempts to simplify the fabrication process and improve efficiency. In this study, ZnO:Al (AZO) is co-sputtered with SiO2 to form AZO:SiO2 films with different SiO2 content. These nanometer-scale films, deposited on top of thin intrinsic hydrogenated amorphous silicon films and capped with low-work-function metal (such as Al and Mg), are demonstrated to function effectively as electron-selective contacts in c-Si solar cells. On the one hand, AZO:SiO2 plays an important role in such electron-selective contact and its thickness is a critical parameter, thickness of 2 nm showing the best. On the other hand, at the optimal thickness of AZO:SiO2, the open circuit voltage (VOC) of the solar cells is found to be relatively insensitive to either the work function or the band gap of AZO:SiO2. Whereas, regarding the fill factor (FF), AZO without SiO2 content exhibits to be the optimal choice. By using AZO/Al as electron-selective contact, we successfully realize a 19.5%-efficient solar cell with VOC over 700 mV and FF around 75%, which is the best result among c-Si solar cells using ZnO as electron-selective contact. Also, this work implies that efficient carrier-selective film can be made by magnetron sputtering method. Keywords: Electron-selective contact, AZO, co-deposition, c-Si solar cells, magnetron sputtering 1. Introduction Carrier-selective contacts applied to crystalline silicon (c-Si) solar cells are attractive due to their high efficiency and simple fabrication process. For example, c-Si solar cells using p-type hydrogenated amorphous silicon (p a-Si:H) as hole-selective layer and n-type a-Si:H film as the electron-selective layer have already been shown to reach 25.1% efficiency [1], and even 26.6% by further adopting an interdigitated back contact structure [2]. Due to these successes, a growing number of companies now produce solar cells that feature doped a-Si:H films as carrier-selective contacts. In research, other 1 novel carrier-selective contacts based on polymer [3], metal oxide [4 -- 7], metal fluoride [8,9], metal nitride [10] films, etc. have attracted a significant interest as these contacts have the potential to further improve the cell performance by using more transparent or conducting layers, and to simplify the fabrication process. Until now, various materials have effective electron-selective layers, including LiFx [8], MgFx [9], MgOx [6], TiOx [4,5], TaOx [11], TaNx [10], alkali/alkaline-earth metal carbonates [12], and their combinations [13], in some cases combined with intrinsic a-Si:H (i a-Si:H) for passivation. ZnO (with or without doping), one of the most demonstrated been as widely used transparent conductive oxides [14 -- 16], has also been proposed as an electron-selective layer in c-Si solar cells [17 -- 20] due to the fact that the conduction band offset between c-Si and ZnO is beneficial for electron transport but the valence band offset forms a barrier for holes transport from c-Si to ZnO [18]. Combining i a-Si:H and ZnO:B grown by metal-organic chemical vapor deposition as the electron-selective layer, an efficiency of 16.6% was demonstrated by Wang et al [18]. Ding et al. [20] realized an 18.46%-efficient c-Si solar cell by using spin-coated ZnO:Al (AZO) on top of i a-Si:H as the electron-selective layer. The researchers from Ye group also demonstrated spin-coated AZO as an effective electron-selective film, achieving an efficiency of 13.6% [19]. Recently, they doped ZnO with Li to reduce its work function and adding intrinsic a-Si:H as passivation layer, promoting the efficiency to 15.1% [21]. In this study, magnetron sputtering method is utilized to prepare AZO as electron-selective film, which is a method easy to fabricate uniform films on large-size substrates and highly compatible to the present mass-production line of solar cells. We co-sputter AZO (2 wt% Al2O3) and SiO2 to form AZO:SiO2 films, which is demonstrated to be a low-work-function material [22]. We investigate the influence of the power applied to the SiO2 target (PSiO2) and the O2/(Ar+O2) flow ratio on the microstructure, conductivity, work function and band gap of the AZO:SiO2 films. When AZO:SiO2 capped with a thermally evaporated metal are applied to c-Si solar cells as electron-selective contacts, it is find that both the thickness of AZO:SiO2 and the work function of the capping metal are highly important. At the optimal thickness of AZO:SiO2, we show how the deposition conditions of AZO:SiO2 affects the open circuit voltage (VOC) and fill factor (FF), and thus pure AZO, i.e. PSiO2=0 W, is found to be the best choice. Finally, we realize a 19.5%-efficient c-Si solar cell by using an AZO/Al stack (on top of a passivating i a-Si:H layer) as electron-selective contact, which is the highest efficiency among the solar cells using ZnO as electron-selective film. In addition, it is worth mentioning that the previously successful electron-selective films are mainly made by either thermal evaporation [6,8,9,12] or atomic layer deposition [5,10,11]. Here our study firstly shows 2 that magnetron sputtering is also a feasible method to fabricate efficient electron-selective films. 2. Results and discussion Fig.1(a and b) compares the microstructure of AZO and AZO:SiO2 films by plane-view transmission electron microscopy (TEM) images. The film thickness was ~2 nm in both cases. The power applied to the AZO target was 35 W, which was maintained for the entire study. For Fig.1(b), PSiO2 was 25 W. Both AZO and SiO2 targets had a diameter of 100 mm. Completely different microstructures of the AZO and AZO:SiO2 are observed from the scanning TEM (STEM) images (Supporting information, Fig.S1). High resolution TEM images (Fig.1(a) and (b)) further reveal that the AZO film contains crystallites with a wurtzite structure and a diameter of a few nanometers in an amorphous matrix, whereas the AZO:SiO2 film is fully amorphous. A gradual decrease in crystallinity with increased SiO2 content could be observed in other studies [22,23]. The influence of the addition of SiO2 on the optoelectronic properties of these layers is presented below. We performed X-ray photoelectron spectroscopy (XPS) measurements to determine the ratio of Si to (Si+Zn) as a function of PSiO2 and O2/(Ar+O2) flow ratio during sputtering. As shown in Fig.1(c), on the one hand, Si/(Si+Zn) increases from 0 to ~0.4 when increasing the PSiO2 from 0 W to 35 W at a constant O2/(Ar+O2) flow ratio of 0.18%. On the other hand, Si/(Si+Zn) decreases with increasing the O2/(Ar+O2) at a constant PSiO2 of 25 W, which indicates that adding O2 the incorporation of SiO2 into AZO:SiO2. Fig.1(d) shows the conductivity of the different AZO:SiO2 films. Increasing PSiO2 leads to a drastic decrease of the conductivity, which is linked to the higher Si/(Si+Zn) ratio its amorphization [22]. When PSiO2 is 0 W, namely for a pure AZO film, the conductivity decreases with increasing O2/(Ar+O2) flow ratio, coinciding with the literature [24]. In other cases (PSiO2 > 0 W), the film conductivity increases with O2/(Ar+O2) flow ratio, owing to the lower SiOx incorporation at higher O2/(Ar+O2). the film and during sputtering reduces in Fig.1 High-resolution plane-view TEM images of (a) AZO film deposited under O2/(Ar+O2): 0.18%, and (b) AZO:SiO2 film deposited under PsiO2: 25 W, O2/(Ar+O2): 0.18%. Insets show the selected-area electron diffraction patterns. (c) Si/(Si+Zn) ratios (from XPS), (d) conductivity, (e) work function and (f) optical band gap of the AZO:SiO2 film versus the power applied to the SiO2 target (PSiO2) and for different O2 to (Ar+O2) flow ratio in the sputtering gas. investigated by Helium The work function of the AZO:SiO2 films was also ultra-violet photoelectron spectroscopy (He-UPS). The results are shown in Fig.1(e). Increasing the PSiO2 and hence the Si/(Si+Zn) ratio results in lower work functions, which agrees well with the result of Nakamura et al [22]. Furthermore, reducing the O2/(Ar+O2) flow ratio also leads to a lower work the more efficient SiO2 function, owing to 3 incorporation. To further study the influence of the deposition conditions (PSiO2 and O2/(Ar+O2) flow ratio) on the bandgap of AZO:SiO2 films, optical absorption coefficients were determined using UV-Vis-NIR spectroscopy. Through the Tauc plots assuming that the films have direct bandgap [25] (see Fig.S2, supporting information), the optical bandgap is obtained. As shown in Fig.1(f), both increasing the PSiO2 and reducing the O2/(Ar+O2) flow ratio lead to higher bandgap. In summary, variations of PSiO2 and O2/(Ar+O2) flow ratio in this study yield significant changes to the material properties. These changes are expected to affect the solar cell performance when using this material as electron-selective contact, which is discussed in the following. Fig.2 Schematic cross section (a) and band diagram at open-circuit conditions (b) of the solar cells using AZO:SiO2/Al as electron-selective contacts. (c) VOC and (d) FF of the solar cells with AZO:SiO2 films deposited at different conditions (different PSiO2 and O2/(Ar+O2) ratio). Average of 5 cells is displayed and error bar displays the the solar cells using different (~2-nm-thick) AZO:SiO2/Al as standard deviation. (e) J-V curves of electron-selective contacts. Fig.2(a) and (b) show the structure and a schematic band diagram of solar cells using AZO:SiO2/Al as electron-selective contact stack and p-type a-Si:H as hole selective contact. Intrinsic a-Si:H was used on both sides of the n-type c-Si wafer as passivation layer. For the AZO:SiO2 films, different thicknesses, different PSiO2 and O2/(Ar+O2) flow ratios have been studied. In this section, AZO:SiO2 includes pure AZO (i.e. films prepared with PSiO2 = 0 W). Fig.2(c) shows that the VOC is only around 580 mV if Al directly covers on i a-Si:H film. However, when a thin (1 nm-thick) AZO:SiO2 layer is inserted between the i a-Si:H and Al, VOC is greatly increased, demonstrating the significance of AZO:SiO2 on the electron-selective contact. The presence of AZO:SiO2 can remove the Fermi-level pinning between Al and i a-Si:H and 4 may also reduce the carrier recombination at the interface. For a thickness of AZO:SiO2 around 2 nm, best VOC values around 690-700 mV are obtained. Interestingly, despite the different deposition conditions (PSiO2 and O2/(Ar+O2)) and thus the different material properties as discussed in previous section, very similar VOC values can be reached. Additionally, very similar VOC values can also be obtained at a thickness of ~2 nm by using undoped ZnO capped with Al as electron-selective contact stack (see Supporting information, Fig. S3). Since VOC is determined by the difference between hole quasi-Fermi level at the positive electrode and electron quasi-Fermi level at the negative electrode, as shown in Fig.2(b), and based on the fact that i a-Si:H films capped with different AZO:SiO2 have to that in further insights the film PSiO2, as presented in Fig 1(d). Therefore, although the AZO:SiO2 deposition conditions have no influence on VOC, they do affect FF. To get the working into mechanisms of AZO:SiO2/metal electron-selective contact stacks, different capping metals have been studied for different AZO:SiO2 film thickness. Here, the PSiO2 was set to 15 W and the O2/(Ar+O2) flow ratio to 0.18%. The solar cells maintain the same structure as shown in Fig.2(a), but Mg, Al, Cu and Au are used as the negative electrodes. Based on literature data, their bulk work functions are 3.66 eV, 4.06-4.26 eV, 4.48-5.1 eV, 5.31-5.47 eV, respectively [28]. However, the effective work function of metals can change depending on the deposition conditions and the formation of the interface is contacted (e.g., formation of interface dipoles, Fermi-level pinning). Fig.3(a) schematically shows the possible energy band diagram near negative contact of the solar cells with the ultrathin AZO:SiO2 film/metal as electron-selective open-circuit conditions. Note, the AZO:SiO2 layers is not represented (although it is expected to be significant), and that the metal work function does not precisely correspond to the literature value, which will be explained in the following. As presented, higher effective work function of the metal leads to upwards band bending in the silicon wafer, i a-Si:H layer and AZO:SiO2. This reduces selectivity of the contact, leading to a slope of electron quasi-Fermi level in the i a-Si:H and AZO:SiO2 layers. The band bending in the silicon wafer also results in an increase in hole concentration at the electron contact, thus leading to smaller Fermi-level splitting in the absorber. Due to these reasons, the VOC is expected to be lower, which is confirmed in Fig.3(b). In addition, a higher upwards band bending means a higher energy barrier for electrons to be collected and thus lower FF, as verified in Fig.3(c). Similar result is also reported in the solar cells using TiOx capped with metal as carrier-selective contact [29]. contact that band bending increasing carrier recombination, under structures. Nevertheless, the same passivation quality (they have comparable implied VOC within the range of 736-741 mV), thus the measurement of very similar VOC values implies that electron quasi-Fermi levels at the negative contact are almost the same for the solar cells with different AZO: SiO2 films. We hypothesize that this is because the AZO:SiO2 films are too thin to screen the influence of Al, making the band diagram dominated by the Al properties and not by the AZO:SiO2 properties. Simulations were also carried out to observe the insensitivity of VOC to the work function of an ultrathin film, as shown in Fig.S4 of Supporting information. The results may help explain that other ultrathin films such as LiFx [8], MgFx [9], MgOx [6], SiO2 [26], etc., combined with a low work function metal can work similarly well as electron-selective contact despite their different energy-band this hypothesis does not imply that any material can work well as electron-selective contact, since different materials may have different abilities to screen the metal work function due to some different material effective conduction band density of states (see Supporting information, Fig.S4). Also, the way in which the material affects the effective work function of metal is important. Fig.2(d) further shows that similarly to Voc, FF increases first and then decreases with increasing the thickness of AZO:SiO2 films. Note that for 0 nm of AZO:SiO2, the FF is variable from run to run. For solar cells made with an AZO:SiO2 thickness of around 2 nm, FF decreases with increasing PSiO2. The current density-voltage (J-V) curves under air mass 1.5 global (AM1.5G) illumination of cells made with a 2-nm-thick AZO:SiO2 layer are shown in Fig.2(e), from which the influence of series resistance (RS) on J-V curves is observed. Based on the method proposed by Bowden [27], RS is calculated to increase from 1.7 Ωcm2 for cells with pure AZO to 4.1 Ωcm2 for cells with AZO:SiO2 (25 W SiO2), which results in the reduced FF. The increased RS is correlated with the decreased conductivity of the AZO:SiO2 film with increasing properties, e.g. 5 Fig.3 (a) Energy band diagram near negative contact of the solar cells using ultrathin AZO:SiO2 film and different capping metals as electron-selective contact stack at open-circuit conditions. Energy band bending in the AZO:SiO2 is neglected. (b) VOC and (c) FF varying with the metal and the thickness of the AZO:SiO2 film. Average of 3 cells is displayed and error bar displays the standard deviation. (d) J-V curves of the solar cells with 2-nm-thick AZO:SiO2 films but different capping metals. (e) J-V curves of the solar cells varying with the thickness of AZO:SiO2 films. The capping metal is Al. When in the absence of any AZO:SiO2 film, athough both VOC and FF obviously depend on the metal, the pinning factor of Fermi-level between the metal layers and the i a-Si:H film is estimated to be 0.3 according to the method described in literature [30]. When the thickness of the AZO:SiO2 film increases the Fermi-level pinning between the metal layers and the i a-Si:H film is removed but the pinning factor of Fermi-level between the metal and AZO:SiO2 is estimated to be 0.1, showing more severe Fermi-level pinning effect. However, from the significant improvement of the VOCs of solar cells with Al, Cu and Au, it can the pinning position of be speculated to ~2 nm, that 6 Fermi-level should move to a higher energy level. It is worth mentioning that inserting TiO2 between c-Si and metal is also reported to change the Fermi-level pining position [31]. The shifting of the Fermi-level pining position modifies the effective work function of metal, a reduction for Al, Cu and Au, but an increase for Mg. This makes the effective work function of Mg similar to that of Al, but lower than that of Cu and Au. Hence the VOCs of solar cells with Al and Mg are similar but higher than that of cells with Cu and Au when the thickness of the AZO:SiO2 is ~2 nm. Fig.3(d) further shows the J-V curves of the solar cells using different metals but with the same 2-nm-thick AZO:SiO2 film under AM1.5G illumination. The J-V curve of the solar cell with Au obviously deviates from that of a diode, suggesting that there is a strong n c-Si/Au Schottky diode opposing the solar cell diode. For AZO:SiO2 film thicknesses between 2 nm and 20 nm, both the VOC and FF decrease with increasing AZO:SiO2 thickness for any capping metal. Note that this decrease is not caused by the increased sputtering damage with time since sputtering damage is almost eliminated thanks to the low sputtering power and appropriate annealing process (see Supporting information, Fig.S5). For AZO:SiO2 thicknesses above 20 nm, both VOC and FF become insensitive to the thickness and appear to stabilize to poor values. Fig.3(e) shows the influence of the thickness of AZO:SiO2 film on the J-V curves of the Al contacted solar cell under AM1.5G illumination. With no AZO:SiO2, the J-V curve shows an S shape, probably because of a Schottky contact between Al and the n c-Si wafer passivated with intrinsic a-Si:H. When the AZO:SiO2 thickness increases to ~2 nm, the effective work function of Al is reduced as mentioned above, and a diode J-V curve is obtained. Further increasing the thickness, the J-V curves are influenced by the increased resistance and deviate from that of a single diode, which lead to the decrease of FF. Finally, to show the potential of sputtered AZO:SiO2 as electron-selective contact, a 4-cm2 cell was made using a 2-nm-thick simple AZO film capped with Al as electron-selective contact. A 19.5% efficiency is demonstrated with a VOC of 701 mV and a FF of 74.7%. The J-V curve of this device is shown in Fig.4. This is the highest efficiency reported for a solar cell that features ZnO as electron-selective contact. This result is comparable to successfully demonstrated electron-selective films [8 -- 12] and also the result suggests that magnetron sputtering method efficient electron-selective films, which the methods easy applied to mass production line. Nevertheless, it is necessary to point out that the FF is still limited by relatively high series resistance, even for this optimal composition and thickness. Also, a striking and universal problem for using nanometer-thin films combined with low work those of cells using other can be used is one of to make 7 function metal as carrier-selective contact is the relatively low JSC. One of the reasons is infrared absorption losses owing to metal close to the Si wafer [32]. We believe that utilizing a thick and conductive film as electron-selective is required to avoid this infrared absorption and fully benefit from the novel carrier-selective contact approach. layer Fig.4 J-V curve of the best solar cell using AZO/Al as electron-selective contact. 3. Conclusion In summary, ZnO:Al (AZO) is co-sputtered with SiO2 to form AZO:SiO2 films with different SiO2 content. These films, capped with different metals, have been applied as electron-selective contact in c-Si solar cells. The microstructure of the AZO film can be changed by incorporating SiO2. Both increasing the power applied to the SiO2 target and decreasing O2/(Ar+O2) flow ratio lead to higher Si/(Si+Zn), resulting in lower conductivity, lower work function and enlarged bandgap. On the one hand, thickness of these films was shown to be a critical parameter when applying them as electron-selective contacts, a thickness of ~2 nm appearing as optimal. At the performance of solar cells is significantly improved compared to that without AZO:SiO2 film. On the other hand, the open circuit voltage (VOC) was found to be insensitive to the deposition conditions of AZO:SiO2, despite the variation of work function and bandgap of the deposition condition of the AZO:SiO2 film was shown to greatly affect fill factor (FF). AZO, without SiO2 content, was thus shown to be the best the material. However, thickness, this electron-selective film. Furthermore, we showed that the effective work function of the capping metal has a significant influence on both VOC and FF, much more striking than the AZO:SiO2 material properties. Finally, a 19.5%-efficient c-Si solar cell with VOC of 701 mV and FF of 74.7% is demonstrated by using AZO/Al as study successfully shows that magnetron sputtering is capable of making efficient carrier-selective films. Further largely on improving the optical properties of this system, e.g. by inserting a >100-nm-thick low-refractive-index optical spacer between the wafer and the metal. improvements will contact. This rely 4. Experimental section The AZO:SiO2 films were deposited by RF-co-sputtering AZO (Al2O3: 2 wt%) and SiO2 at room temperature in a magnetron sputtering system (Leybold Univex). The target diameter was 100 mm. The sputtering power of AZO was fixed at 35 W, and the PSiO2 varied from 0 W to 35 W. The flow rates of Ar and O2 were changed to yield an O2/(Ar+ O2) flow ratio of 0%-0.35%, the working pressure was fixed at 2.7×10-3 mbar. The substrate was rotated at 10 rpm to obtain homogeneous films. Film thickness was controlled by the deposition time. For solar-cell fabrication, n-type float zone (FZ) c-Si wafers with resistivity of 1-5 Ω cm and thickness of either 240 μm (for Fig.2 and 4 ) or 180 μm (for Fig.3) were used. These wafers were firstly randomly etched pyramids-textured by wet-chemical cleaning and HF dipping. Intrinsic a-Si:H films were then deposited on both sides of the Si wafers as passivation layer by plasma enhanced chemical vapor deposition (PECVD). p-type a-Si:H was deposited on the front side (i.e. illumination side) of the wafer, and the AZO:SiO2 film was deposited on the back side by magnetron sputtering. The front side was then covered with an ~80-nm-thick indium-tin-oxide (ITO) film as the antireflection and conductive layer, and the front metal grids were prepared by either magnetron sputtering through a shadow mask for the 1.1-cm2 cells (Fig.3) or screen printing for the 4-cm2 cells (Fig.2 and 4), followed by annealing at 210 °C for 30 minutes. Finally, the back side of the wafer was form followed in KOH solution surface, to 8 covered with the metal film by thermal evaporation. Characterization: The Si to (Si+Zn) ratio of the AZO:SiO2 films were characterized by X-ray photoelectron spectroscopy (XPS) with Al-Kα excitation. To this end the Si 2p, O 1s and Zn 3p core levels were measured and fitted using a linear background and Voigt peaks with a 15 % Lorentz-contribution. The Zn 3p peak was fitted using two peaks with a fixed distance of 2.95 eV and the same full-width at half maximum. These two peaks represent the contributions from ZnO and ZnOH. The Si 2p signal was fitted with a single signal and the O 1s signal was fitted with two signals to account for SiO and ZnO contributions. The Si and Zn contents of the mixed layers were calculated using sensitivity factors, extracted from stochiometric ZnO and SiO2 samples. The ratio of the Si/Zn oxide peak area to the O 1s peak area, corrected by the stoichiometry of the respective element, was used as the sensitivity factor. These sensitivity factors were used to obtain the fraction of Si and Zn in the in the mixture. The work function of the films was characterized using Helium ultra-violet photoelectron spectroscopy (He-UPS). A bias voltage of 5 eV was applied and the secondary electron cut-off was measured and fitted using a Boltzmann-Sigmoid function to obtain the work function of the layers. The TEM observation of the AZO:SiO2 films were performed using an FEI Tecnai Osiris microscope. For that purpose, AZO and AZO:SiO2 thin films were directly sputtered onto Cu grids coating with a thin C film. High-resolution TEM top view recorded alongside selected-area electron diffraction patterns to assess the microstructure of the films. The reflectance spectra and transmittance spectra of samples were measured with a spectrophotometer (Lambda-950, Perkin Elmer) to extract the absorption coefficients. The thickness of the AZO:SiO2 film on planar surface was measured by ellipsometry, and a factor 0.66 was applied to obtain an estimate of that on the textured Si surface. Al electrodes with 1-mm spacing were deposited on the AZO:SiO2 film to measure the dark conductivity (The measured value is the lower limit of the real conductivity of the films since the contact resistance is included). Solar cell characterizations were carried out using a Wacom WXS-90S-L2 solar simulator, at standard images were test conditions (AM1.5G spectrum, 100 mW/cm2 and 25 °C). Acknowledgements and for silicon preparation, from CSEM thank Raphaël Monnard The authors for amorphous thank Christophe Allebe, Fabien Debrot and Nicolas Badel the high quality wet-processing and metallization. This project has received funding from the European Union's Horizon 2020 research and innovation programme under Grant Agreement No. 727523 (NextBase), as well as Swiss national science foundation under Ambizione Energy grant ICONS and the China Postdoctoral Science Foundation. Notes The authors declare no competing financial interest. Supporting information information associated with Supporting article can be found in the online version. this Reference [1] D. Adachi, L. Hernandez, K. Yamamoto, Impact of carrier recombination on fill factor for large area heterojunction crystalline silicon solar cell with 25 . 1 % efficiency, Appl. Phys. Lett. 107 (2015) 233506. [2] K. Yoshikawa, W. Yoshida, T. Irie, H. Kawasaki, K. Konishi, H. Ishibashi, T. Asatani, D. Adachi, M. Kanematsu, H. Uzu, K. Yamamoto, Exceeding conversion efficiency of 26% by heterojunction interdigitated back contact solar cell with thin film Si technology, Sol. Energy Mater. Sol. Cells. 173 (2017) 37-42. J. Sheng, K. Fan, D. Wang, C. Han, J. Fang, P. Gao, J. Ye, Improvement of the SiOx passivation high-efficiency Si/PEDOT:PSS heterojunction solar cells, layer for [3] 9 ACS Appl. Mater. Interfaces. 6 (2014) 16027 -- 16034. [4] X. Yang, P. Zheng, Q. Bi, K. Weber, Silicon heterojunction solar cells with electron selective TiOx contact, Sol. Energy Mater. Sol. Cells. 150 (2016) 32 -- 38. [5] X. Yang, Q. Bi, H. Ali, K. Davis, W. V. Schoenfeld, K. Weber, High-performance TiO2-based electron-selective contacts for crystalline silicon solar cells, Adv. Mater. 28 (2016) 5891 -- 5897. [7] [6] Y. Wan, C. Samundsett, J. Bullock, M. Hettick, T. Allen, D. Yan, J. Peng, Y. Wu, J. Cui, A. Javey, A. Cuevas, Conductive and stable magnesium oxide electron-selective contacts for efficient silicon solar cells, Adv. Energy Mater. 7 (2017) 1601863. J. Geissbühler, J. Werner, S. Martin De Nicolas, L. Barraud, A. Hessler-Wyser, M. Despeisse, S. Nicolay, A. Tomasi, B. Niesen, S. De Wolf, C. Ballif, 22.5% efficient silicon heterojunction solar cell with molybdenum oxide hole collector, Appl. Phys. Lett. 107 (2015) 081601. J. Bullock, M. Hettick, J. Geissbühler, A.J. Ong, T. Allen, C.M. Sutter-Fella, T. Chen, H. Ota, E.W. Schaler, S. De Wolf, C. Ballif, A. Cuevas, A. Javey, Efficient silicon solar cells with dopant-free asymmetric heterocontacts, Nat. Energy. 1 (2016) 15031. [8] [9] Y. Wan, C. Samundsett, J. Bullock, T. Allen, M. Hettick, D. Yan, P. Zheng, X. Zhang, J. Cui, J. McKeon, A. Javey, A. Cuevas, Magnesium electron-selective contacts for crystalline silicon solar cells, ACS Appl. Mater. Interfaces. 8 (2016) 14671 -- 14677. fluoride [10] X. Yang, E. Aydin, H. Xu, J. Kang, M. Hedhili, W. Liu, Y. Wan, J. Peng, C. Samundsett, A. Cuevas, S. De Wolf, Tantalum nitride electron-selective contact for crystalline silicon solar cells, Adv. Energy Mater. 8 (2018) 1800608. [11] Y. Wan, S.K. Karuturi, C. Samundsett, J. Bullock, M. Hettick, D. Yan, J. Peng, P.R. Narangari, S. Mokkapati, H.H. Tan, C. Jagadish, A. Javey, A. Cuevas, Tantalum oxide electron-selective heterocontacts for and silicon photovoltaics photoelectrochemical water reduction, ACS Energy Lett. 3 (2018) 125 -- 131. [12] Y. Wan, J. Bullock, M. Hettick, Z. Xu, C. Samundsett, D. Yan, J. Peng, J. Ye, A. Javey, A. Cuevas, Temperature and humidity stable alkali/alkaline-earth metal carbonates as electron silicon photovoltaics, Adv. Energy Mater. (2018) 1800743. doi:10.1002/aenm.201800743. heterocontacts for [13] J. Bullock, Y. Wan, Z. Xu, S. Essig, M. Hettick, H. Wang, W. Ji, M. Boccard, A. Cuevas, C. Ballif, A. Javey, Stable dopant-free asymmetric heterocontact silicon solar cells with efficiencies above 20%, ACS Energy Lett. 3 (2018) 508 -- 513. [14] K. Ellmer, A. Klein, B. Rech, Transparent conductive and applications in thin film solar cells, Springer, 2008. oxide. zinc basics [15] J. Perrenoud, L. Kranz, S. Buecheler, F. Pianezzi, A.N. Tiwari, The use of aluminium doped ZnO as transparent conductive oxide for CdS/CdTe solar cells, Thin Solid Films, 519 (2011) 7444-7448. [16] M. Morales-Masis, S. De Wolf, R. Woods-Robinson, J.W. Ager, C. Ballif, Transparent efficient optoelectronics, Adv. Electron. Mater. 3 (2017) 1600529. electrodes for the [17] B. Hussain, Improvement in open circuit voltage of n-ZnO/p-Si solar cell by using amorphous-ZnO at interface, Prog. Photovoltaics Res. Appl. 25 (2017) 919 -- 927. [18] F. Wang, S. Zhao, B. Liu, Y. Li, Q. Ren, R. Du, N. Wang, C. Wei, X. Chen, G. Wang, B. Yan, Y. Zhao, X. Zhang, Silicon solar cells with bifacial metal oxides carrier selective layers, Nano Energy. 39 (2017) 437 -- 443. [19] P. Liu, P. Gao, X. Liu, H. Wang, J. He, X. Yang, Y. Zeng, B. Yan, J. Fang, J. Ye, High-performance organic-silicon heterojunction solar cells by using Al-doped ZnO as cathode interlayer, Sol. RRL. 2 (2018) 1700223. [20] J. Ding, Y. Zhou, G. Dong, M. Liu, D. Yu, F. Liu, Solution-processed ZnO as the efficient passivation and electron selective layer of silicon solar cells, Prog Photovolt Res Appl. (2018) 1 -- 7. https://doi.org/10.1002/pip.3044. 10 [21] Z. Wang, Y. Yang, L. Zhang, H. Lin, Z. Zhang, D. Wang, S. Peng, D. He, J. Ye, P. Gao, Modulation-doped ZnO as high performance electron-selective for efficient silicon heterojunction solar cells, Nano Energy. 54 (2018) 99 -- 105. layer [22] N. Nakamura, J. Kim, H. Hosono, Material design of transparent oxide semiconductors for organic electronics: why do zinc silicate thin films have exceptional properties?, Adv. Electron. Mater. 4 (2018) 1700352. [23] E. Rucavado, M. Graužinytė, J.A. Flores-Livas, Q. Jeangros, F. Landucci, Y. Lee, T. Koida, S. Goedecker, A. Hessler-Wyser, C. Ballif, M. Morales-Masis, New Route for "Cold-Passivation" of Defects in Tin-Based Oxides, J. Phys. Chem. C. 122 (2018) 17612 -- 17620. [24] D.S. Kim, J.H. Park, S.J. Lee, K.J. Ahn, M.S. Lee, M.H. Ham, W. Lee, J.M. Myoung, Effects of oxygen concentration on the properties of Al-doped ZnO transparent conductive films deposited by pulsed DC magnetron sputtering, Mater. Sci. Semicond. Process. 16 (2013) 997 -- 1001. [25] A. Dabirian, S. Martin De Nicolas, B. Niesen, A. Hessler-Wyser, S. De Wolf, M. Morales-Masis, C. Ballif, Tuning the optoelectronic properties of ZnO:Al by addition of silica for in high-efficiency crystalline Si solar cells, Adv. Mater. Interfaces. 3 (2016) 1500462. trapping light Principles [26] Z. Yang, P. Gao, J. Sheng, H. Tong, C. Quan, X. Yang, K.W.A. Chee, B. Yan, Y. Zeng, J. Ye, dopant-free electron-selective contacts based on tunnel oxide/low work-function metal stacks and their applications in heterojunction solar cells, Nano Energy. 46 (2018) 133 -- 140. of [27] S. Bowden, A. Rohatgi, Rapid and accurate determination of series resistance and fill factor losses in industrial silicon solar cells, in: 17th Eur. Photovolt. Sol. Energy Conf., 2001. [28] D.R. Lide, CRC Handbook of Chemistry and Physics, Boca Raton, Florida, 2005. [29] T. Matsui, M. Bivour, P. Ndione, P. Hettich, of M. atomic-layer-deposited TiOx as selective Investigation Hermle, electron and hole contacts to crystalline silicon, Energy Procedia. 124 (2017) 628-634. [30] M. Bivour, Silicon heterojunction solar cells analysis and basic understanding, PhD thesis, University of Freiburg, 2015. [31] A. Agrawal, J. Lin, M. Barth, R. White, B. Zheng, S. Chopra, S. Gupta, K. Wang, J. Gelatos, S.E. Mohney, S. Datta, Fermi level depinning and contact resistivity reduction using a reduced titania interlayer in n-silicon metal-insulator-semiconductor ohmic contacts, Appl. Phys. Lett. 104 (2014) 112101. [32] Z.C. Holman, S. De Wolf, C. Ballif, Improving metal reflectors by suppressing surface plasmon polaritons: A priori calculation of the internal reflectance of a solar cell, Light Sci. Appl. 2 (2013) e106. 11
1711.07407	2	1711	2017-12-01T01:26:30	CFD investigation of thermal performance of aluminum oxide nanofluid in channel	[ "physics.app-ph", "physics.flu-dyn" ]	This paper shows the CFD investigation of pressure drop and thermal performance of aluminum oxide nano fluid inside square channel inserted a cylinder with and without fin using two-phase method. The constant heat flux on outer wall and laminar flow regime are applied in the considered domain using finite volume method. The results depict that the enhance of nano-particles volume concentration and Reynolds number have dramatic effects on heat transfer coefficient enhancement. Moreover, the increase of nano-particle diameter has opposite impact on heat transfer efficiency. The passive way leads to higher pressure drops. For all fluids under consideration, pressure drop escalated with Reynolds number. Injecting nano-particles into the water causes to increase in pressure drop and this impact is more significant in high nano-particle volume fraction.	physics.app-ph	physics	CFD Investigation of Thermal performance of Aluminum oxide nanofluid in channel Mehdi Jafari, Mohammad Khalili∗ Khomeinishahr Branch, Islamic Azad University, Khomeinishahr, Iran Abstract The present paper is a numerical study of heat transfer and pressure drop of two nano-fluids including water as base fluid with Al2O3 nano-particles inside a square channel having a cylinder inside, with and without fin under constant heat flux condition by using two-phase Euler-Lagrange approach. In this paper, numerical investigation has been done for various combinations of base fluid, nano-particle size, and concentration through a straight cylinder. Simulation has been performed in a laminar flow regime using finite volume method. Be- sides, the thermal boundary condition of constant uniform heat flux on the cylinder wall was applied. The results show that the increase of Reynolds num- ber and nano-particles volume concentration have considerable effects on heat transfer coefficient enhancement. For nano-particles, the heat transfer coeffi- cient decreases when nano-particles diameter increases. The passive way used in this study leads to higher pressure drops. For all fluids under consideration, pressure drop escalated with Reynolds number. Adding nano-particles to the base fluid leads to rise in pressure drop and this effect is more intensive for higher concentrations. Regardless of nano-particles type and their volume con- centration, the skin friction coefficient increases with a rise in Reynolds number. Keywords: Heat Transfer; of Aluminum oxide nanofluid; Inner Cylinder; Euler–Lagrange Approach 1. Introduction Lots of changes have happened in the field of enhanced heat transfer. Offset strip fins are designed to enhance heat transfer by enlarging surface. Some reviews of experimental studies on the effects of different surface geometries on the flow and heat transfer performance of offset strip fins are addressed in [1, 2, 3]. Suspensions of nano-particles also can increase the viscosity and decrease the specific heat, which means that the improvement in thermal conductivity of ∗Corresponding author Preprint submitted to Macromolecular Theory and Simulations December 4, 2017 nano-fluids may be counteracted by the negative effects of viscosity and specific heat. A wide variety of processes involving heat and energy, have low efficiency due to low thermal conductivity of working fluid such as water, ethylene glycol, and engine oil. An advanced method proposed to enhance heat transfer char- acteristic of fluid is dispersion of ultrafine solid particles in a base fluid known as nano-fluids. There are considerable researches on the superior heat transfer properties of nano-fluids especially on thermal conductivity and convective heat transfer [4, 5]. In the recent decade, numerous investigations are performed by using two- phase method in which mass, momentum, and energy conservation equations are solved for each phase individually or momentum and energy conservation equations coupled with mass conservation equation are solved for each phase. Ali Akbari et al. [6] numerically studied the heat transfer of CuO nano-fluid through a micro-tube using two phase approach. They showed that increasing nano-particles volume concentration leads to reduction in Nusselt number and friction factor. Moshizi et al. [7] Investigated heat transfer and pressure drop properties of Al2O3–H2O nano-fluid through a tube under constant heat flux condition. Results indicate that slip velocity at the tube walls improves coeffi- cient of heat transfer and enhances the ratio of pressure gradient. Heat transfer under nano-fluids flow with non-Newtonian base fluid inside a microchannel was studied by Esmaeilnejad et al. [8] applying two-phase mixture model. Results obtained illustrated impressive heat transfer enhancement of non-Newtonian flu- ids where nano-particles are distributed. By rising volume fraction, more heat transfer enhancement would be achieved. Due to a wide range of practical applications in heat transfer systems, many investigations on nano-fluids reported that nano-fluids have desirable proper- ties and behaviors such as enhanced wetting and spreading [9], increased heat transfer in forced convection [10, 11, 12]. Shuai and Chang [13] used a new type of non-continuous finned tubes that enhances heat transfer area. They found that the use of finite volume method is very helpful and important in design of three-dimensional pin fin tubes. Mir et al. [14] studied laminar forced convection in a finned annulus numerically. The simulation corresponded to a thermal boundary condition of uniform heat input per unit axial length with peripherally uniform temperature at any cross section. In the solution domain, heat transfer and fluid flow were investigated for different ratios of inner and outer pipes radius, fins height and number of fins. The results showed good agreement with other publications. K. Goudarzi and H. Jamali [15] used aluminum oxide in ethylene glycol as a nano-fluid so as to enhance heat transfer properties in a car radiator with wire coil inserts. The results showed enhancement in heat transfer rates up to 9% if using coils. Furthermore, the simultaneous use of coils with nano-fluids caused by the better thermal performance enhancement than coils alone. Ding et al. [16] presented a numerical study using TiO2-water as a nano-fluid. The flow was laminar under a constant wall heat flux prevailed. The study was carried out for both heat transfer coefficient and wall shear stress. By increasing nano-particles volume concentration, the shear stress increased. 2 Sohel et al. [5] experimentally showed that by increasing volumetric concentra- tion of Al2O3-water nano-fluid, thermal effectiveness increased at all flow rates. However, they found that thermal effectiveness was not necessarily increased with the increase of flow rate. They found 18% convective heat transfer co- efficient enhancement by using 0.25% concentrated Al2O3-water nano-fluid as compared to distilled water. Ho et al. [17] assessed Al2O3-water nano-fluid forced convection heat transfer. They observed 1% volumetric concentration nano-fluid was more efficient than 2% due to more variation occurrence in dynamic viscosity with temperature. Using 1% volume concentrated Al2O3-water nano-fluid, 70% enhancement was found in convective heat transfer coefficient. S. Akbarzadeh et al. [18] performed a study about the influence of particles volume fraction and temperature on the thermal conductivity and viscosity of nano-fluids and obtained new correlations. The results showed that viscosity of nano-fluids increases if any rise occurs in particles volume concentration. The thermal conductivity exhibited a nonlinear growth with the particle volume fraction. Furthermore, in high temperatures, thermal conductivity experienced an improvement. Cong et al. [11] studied heat transfer enhancement and behavior of Al2O3 nano-particles for use in cooling systems such as microprocessors or electronic components. Results showed a very good enhancement in heat transfer proper- ties. For example, for a special particle volume concentration of 6.8%, the heat transfer coefficient increased almost 40% in comparison with the base fluid. Zhang et al. [19] investigated heat transfer enhancement making use of micro-fin structures and nano-fluids. Data obtained presented that by increasing number of fins, Nusselt number and friction factor increase. In addition, using TiO2- water nano-fluid prospers heat transfer process, but causes the pressure to drop. The literature shows the importance of heat transfer enhancement using nano- fluids and fins each individually or combined together. Looking into conducted researches, it seems that there are very limited studies on non-circular channels for nano-fluids and heat transfer. Also, the effects of longitudinal fin and its shape in the channel on heat transfer have not been investigated so far. The objective of this paper is to study heat transfer and pressure drop of two nano- fluids including water as the base fluid with Al2O3 nano-particles inside a square channel having a cylinder inside, with and without fin under constant heat flux condition. 2. Physical Model 2.1. Geometry As mentioned earlier, the purpose of this paper is to study a nano-fluid flow in a channel having an inner cylinder. For this reason, Finite volume method has been used. In the current research, a square channel with a rigid cylinder located in its center and a flat and wavy separating sheet are used. In Fig. 1, a schematic of channel geometry is exhibited in non-real dimensions. Channel 3 dimensions are shown in Fig. 1 and presented in Table 1 in which L, A and B are length, width, and height of the channel, respectively and D is rigid cylinder diameter. 2.2. Governing Equations Numerous studies have been done in order to model two-phase flows in which acceptable results are achieved [20, 21, 22]. Two-phase Eulerian-Lagrangian model is considered as a continuous medium, while fluid and nano-particles phases are considered as discrete mediums. Thus, The Navier-Stokes equations are required to be solved for the fluid. Buongiorno [23] concluded that among diffusion mechanisms having an effect on nano-particles, thermophoretic and Brownian forces are amongst the most important ones. For this reason, effects of thermophoretic and Brownian forces are expressed as source terms while applying momentum equation. Therefore, mass, momentum, and energy con- servation equations for the liquid phase being exposed at a steady state flow and under the condition of temperature-dependent physical properties are written as Mass conservation equation is written as [24]: ∇ • (ρf Vf ) = 0 (1) Momentum conservation equation is written as [24]: ∇• (ρf Vf Vf ) = −∇P+∇• (µf∇Vf ) +ρ (FB+FD+FL+FV+FG+FP) (2) In which FB, FD, FL, FV , FG and FP are Brownian force, drag force, Saffman lift force, virtual mass force, gravitational force and gradient pressure force ap- plied on a particle, respectively. The Brownian force is applied on fine particles spread through the fluid. Collisions among particles and fluid molecules affect the dispersion of the particles when particles are fine enough and micron-sized. The influence of Brownian force is appropriated in an additional forces term. Using the process of Gaussian spectrum with the intensity spectrum of Sn,ij, the Brownian force components can be modeled. Based on this fact, one can write [25] In which δij is introduced as the Kronecker delta function possessing the Sn,ij = S0δij (3) following form. S0= 21νkBT (cid:17)2 (cid:16) ρP ρ π2ρd5 P 4 Cc (4) (cid:114) Where T is the absolute fluid temperature, ν is kinematic viscosity and kB is Stefan-Boltzmann constant. The magnitude of the Brownian force components can be computed by FBi=ζi πS0 ∆t (5) Where ζi is the Gaussian random numbers with the variance of non-unity and the mean value of zero. The drag force is calculated from the following equation. FD= 18µ d5 PρPCc (VF−VP) (6) Where ρP is density of the nano-particles, Cc is the Cunningham correction factor that is computed by Eq. (7), VF and VP are the speed of continuous phase and particles, respectively. Where λ is defined as molecular free path. The gravitational force and gradient pressure force are acquired through Eq. (8) and (9), respectively. (cid:16) (cid:17) Cc= 1+ 2λ dP 1.257 + 0.4e− 1.1dP 2λ g (ρP−ρf ) ρP FG= (cid:18) ρf (cid:19) FP= ρP VP•∇Vf (7) (8) (9) The force needed to accelerate the fluid around the particles is called the virtual mass force which is stated as follows (Ranz- Marshal [26]): FV= 1 2 ρf ρP d dt (VF−VP) (10) As a result of rotation due to the velocity gradient, a force is applied to a single particle called the Lift force that can be calculated via Eq. (11) (Saffman [27]): FL= 2kν0.5ρf VP dij dP (dijdji)0.25 (VF−VP ) (11) In which dij is the deformation tensor and k is considered to be 2.594. The equation utilized to calculate lift force is suitable for fine particles. It should be said that Eq. (11) is practicable for submicron-sized particles (Sommerfeld [28]). Energy equation is written as: ∇• (ρf CP,f VF Tf ) =∇• (kf∇Tf ) +VP Q (12) 5 In which Q is defined by the following relation that is the heat flux transferred between nano-particles and fluid. Q=hAP (TP−Tf ) (13) Where AP is defined as the particle surface area and h may be attained as follows. h = kf Nu dP (14) In Eq. (14), N u is obtained from Ranz –Marshal [28] relation as follows. Nu = 2 + 0.6Re 1 2 d Pr 1 3 (15) Considering a control volume around one particle and then applying energy conservation equation for the control volume leads to mPCP,P dTP dt = Q In the modeling of solid phase, one can write ∇2T = 0 (16) (17) 2.3. The Thermophysical properties of nano-fluid Determination of the physical properties of nano-fluids have been the center attention of many researchers in the past decades and many researches were done in this regard. In the present study, the base fluid is water and the particles are aluminum oxide which are the most commonly used particles in nano-fluids. The properties of these particles are brought in Table 2 [29]. As seen, the conductivity coefficient in nano-particles is much higher than pure water. 2.4. Boundary Conditions As shown in Fig. 1, a steady state and laminar fluid (pure fluid or nano- fluid) flow with uniform profile enters the channel at 25 C. Reynolds number is evaluated applying equation (18). In equation (18), properties are calculated at entrance temperature. Re = ρumDh µ Where Dh is the channel hydraulic diameter and is defined as below 4H2−πB2 16H + 4πB Dh= (18) (19) In equation (19), H and D are the channel height and rigid cylinder diameter, respectively. A constant and uniform 200 W/m2 heat flux on the channel walls is selected as the thermal boundary condition. Then, fluid is discharged to the atmosphere. (see Fig. 1) 6 2.5. Numerical Solution Numerical methods are very applicable in the field of nano-fluid flow [29, 30, 31, 32]. For discretization of the aforesaid nonlinear equations, the control volume method is used. For estimation of the diffusion and convection terms, the upwind scheme of second order is used and the SIMPLE algorithm is employed for coupling the velocity and pressure fields. 2.6. Grid Generation and Grid Sensitivity Grid generation is an important part of simulation, as it affects time, conver- gence, and results of the solution. Furthermore, regular gridding has better effects on the aforesaid parameters than irregular one dose. It is worth noting that grid should be fine enough near the walls so that one could investigate steep gradients of physical properties adjacent to the walls. These gradients occur perpendicular to the walls. Therefore, regular gridding with elements increment along the radius of the walls is used. In Fig. 2, a view of channel geometry gridding is depicted. In order to check grid sensitivity, local Nusselt variations at the channel exit due to an increase in the grid elements is evaluated. Water at 25 C and Re = 1000 enters the channel shown in Fig 1, possessing the properties listed in Table 1. A constant 20000 W/m2 heat flux is applied on the channel walls. Variations of Nusselt number with elements number at the channel exit is plotted in Fig. 3. By minifying grid, Nusselt number varies at the channel exit. As it can be observed, a grid having 1456446 elements is pertinent and no sensible change is seen minifying grid anymore. Elements number in five different cases is presented in Table 3. 2.7. Validation After determining elements number, by applying nano-fluid properties equations and boundary conditions into the software, simulation is completed. For mass, energy, and momentum conservation equations, the minimum divergence criteria is considered to be10−5. In order to verify the effect of nanofluids on heat transfer enhancement, an [33]. For validating the results of experimental study was done by Kim et al. the simulation, data presented in Ref. [35] are used. A laminar Al2O3-water with 3% volume concentration is modeled in a 2-meter long cylinder having 4.57 mm of diameter under2089.56 W/m2 Constant heat flux withRe = 1460. Results obtained are demonstrated in Fig. 4. Considering presented figures in this section and good agreement with theoret- ical and experimental data, it is found that simulation is accurate enough and results are confirmed. 3. Results and Discussion 3.1. Effects of Reynolds Number Fig. 5 represents axial variations of convective heat transfer coefficient and local Nusselt number of the base fluid versus Reynolds number. As Reynolds number 7 increases, convective heat transfer and local Nusselt number grow due to the reduction in the boundary layer thickness. In Fig. 6, axial variations of local convective heat transfer coefficient and local Nusselt number of Al2O3 nano-fluid is schemed against Reynolds number under 1% volume concentration and nano-particles diameter of 25 nm. As stated for the base fluid, an increase in Reynolds number causes the convective heat transfer coefficient to increase. For other concentrations, analogous graphs can be drawn. In all these graphs, variations of local convective heat transfer coefficient with Reynolds number is similar. Effects of Reynolds number on the nano-fluid flow was investigated in local graphs. Now, by computing mean convective heat transfer coefficient, a better comparison can be performed. In Fig. 7, variations of mean convective heat transfer coefficient of Al2O3 nano-fluid versus Reynolds number is plotted for five different volume concentrations of nano-particles having 25 nm diameter. Mean convective heat transfer coefficient for water increases up to 181 percent, if Reynolds number changes from 100 to 1000, while for Al2O3 nano-fluid in 1% and 5% volume concentration, up to 184.4 and 199.6 percent growth is achieved. Generally, by increasing Reynolds number, convective heat transfer for nano- fluid enhances as it did for the base fluid. 3.2. Effects of Nano-Particles Concentration Another effective parameter on nano-fluid heat transfer enhancement is volume concentration of particles. Note that by changing concentration, two phenom- ena affect mean heat transfer from walls. On one side, adding nano-particles increases boundary layer thickness and reduces temperature gradient next to the walls. On the other side, nano-fluid heat transfer increases with concentration. The combined net effect is convective heat transfer coefficient enhancement. Also, nano-fluid density increases, if a rise in the volume fraction of particles occurs which leads to increase in momentum and convective heat transfer. Den- sity gradient between particles and fluid temperature difference is the most important factors in naturally blended mixtures. Fluctuations in local density create a cavity adjacent to the fluid molecules that forces them to move and mix. In Figs. 8 and 9, effects of volume concentration on heat transfer coefficient and local Nusselt number of the nano-fluid are depicted for Re=100, 500 and 1000. Results show that heat transfer coefficient and local Nusselt number of the nano-fluid in different Reynolds numbers increase with any rise in volume con- centration. According to the relation of nano-fluid heat transfer with particles concentration increase, static and dynamic heat transfer of nano-fluid enhance which would increase heat transfer coefficient. In Fig. 10, variations of convective heat transfer coefficient for Al2O3-water nano-fluid is plotted versus volume concentration under nano-particles diameter of 25 nm and Re=100, 250, 500, 750 and 1000. As it can be observed, increasing concentration at a constant Reynolds number leads to increase in convective 8 heat transfer coefficient. For instance, in 5% volume concentration and Re=100, convective heat transfer coefficient increases up to 26.47 percent relative to the base fluid. Also, for Re=500 and 1000, up to 33.13 and 34.71 percent increase can be obtained. 3.3. Effects of Nano-Particles Size Another effective and basic parameter on convective heat transfer for nano-fluid is nano-particles size. According to the relation of heat transfer coefficient, one can infer that nano-particles size has negative effects on heat transfer and heat transfer coefficient. In Fig. 11, axial variations of convective heat transfer coefficient and local Nusselt number against size and volume concentration of nano-particles is demonstrated at Re=500. Results indicate that by increasing particles size at a constant volume concentration, local convective heat trans- fer coefficient decreases slightly. Variations of mean convective heat transfer coefficient for Al2O3-water nano-fluid versus size and volume concentration of nano-particles at Re=500 for water as base fluid is illustrated in Fig. 12. Alike local graphs, by decreasing particles size, nano-fluid heat transfer coefficient in- creases due to the rise in surface-to-volume ratio and properties improvement. For example, by changing particles diameter from 25 nm to 50 nm, mean con- vective heat transfer coefficient for Al2O3-water at 1% and 5% volume concen- tration reduces from 1.13 to 7.4 percent. It seems that at higher concentrations, nano-particles inside the base fluid show more reduction. 3.4. Effects of Pressure Drop Increasing particles density and volume fraction lead to rising in momentum and convective heat transfer. Also, nano-fluid viscosity increases, if particles volume concentration enhances. Therefore, by increasing concentration, nano- fluid pressure drop increases relative to the base fluid. Fig. 13 displays variations of pressure drop for Al2O3-water versus Reynolds number at five nano-particles volume concentrations. Similar to the base fluid, rising Reynolds number at constant concentration, increases nano-fluid pressure drop. For instance, by changing Reynolds number from 250 to 750 at 1% and 5% volume concentration, pressure drop increases up to 240.7 and 228.3 percent, respectively. Also, slope of pressure drop experiences a growth with volume concentration. Fig. 14 indicates that by increasing nano-particles volume concentration at constant Reynolds numbers, pressure drop rises. As an example, at Re=500 and 4% volume concentration, pressure drop for nano-fluid is 1.88 times bigger than that of base fluid. Variations of nano-particles diameter do not have sensible influence on pressure drop. 4. Conclusion A wide variety of processes involving heat and energy, have low efficiency due to low thermal conductivity of working fluid such as water, ethylene glycol, 9 and engine oil. An advance method proposed to enhance heat transfer charac- teristic of fluid, is dispersion of ultrafine solid particles in a base fluid known as nano-fluids. In this paper, heat transfer and effects of Reynolds number, nano-particles concentration, nano-particles size, pressure drop and wavy sheet AL2O3 and nano-particles and water as the base fluid inside a square channel having an inner cylinder, with and without fin under constant heat flux condi- tion were investigated. According to the research done, the following conclusions can be inferred. • As the thermal boundary layer grows, local Nusselt number decreases continuously. • Increasing mass flow and Reynolds number make a rise in heat transfer coefficient, base fluid, and nano-fluid Nusselt number. The more concen- trated nano-fluid, the more increase in aforesaid parameters. • Adding nano-particles increases boundary layer thickness and reduces tem- perature gradient next to the walls. Also, nano-fluid heat transfer in- creases with concentration. The combined net effect is convective heat transfer coefficient enhancement. • By decreasing particles size at a constant volume concentration, local convective heat transfer coefficient increases due to the rise in surface- to-volume ratio and properties improvement. • By increasing concentration, nano-fluid pressure drop increases relative to the base fluid. Similar to the base fluid, rising Reynolds number at constant concentration, increases nano-fluid pressure drop. • If Reynolds number rises, skin friction increases as well. • Flat separating sheet reduces convective heat transfer relative to the case where no sheet is used, while wavy sheet leads to convective heat transfer enhancement. The extension of this paper for nanofluids, according to previous studies [34, 35, 36, 37, 38, 39, 40, 41, 42, 30, 43, 44], affords engineers a good option for micro- and nano simulations. 10 P Q Re S0 Sn,ij t T v Pressure, N/m2 heat flux Reynolds Spectral intensity basis Spectral intensity Time (s) Temperature, K Velocity, m/s Greek symbols µ δij ∆ ζi Dynamic viscosity (N s/m2) Kronecker delta function Difference Zero-mean, unit-variance- independent Gaussian random number Molecular free path Kinematic viscosity (m2/S) λ υ ρ Density Subscripts f p Fluid Particle Notation A B Cc Cp D dp dij FD FL FV FG FP FB g h k kb L Nu Particle surface area, m2 channel height Cunningham correction factor to Stokes' drag law Specific heat, J/kg K rigid cylinder diameter Particle diameter, nm Deformation tensor Drag force Lift force Virtual mass force Gravity force Pressure gradient force Brownian force Gravity acceleration, m/s2 Convective heat transfer coeffi- cient, W/m2K thermal conductivity for Fluid, W/m.K Boltzmann constant (=1.3807 × 1023J/k) axial length, m Peripherally number average Nusselt 11 Table 1: Dimension of the channel under consideration L (m) A (mm) B (mm) D (mm) 3 50 50 25 Table 2: the properties of Al2O3 [31] Value 3600 765 36 50 Property Density (kg/m3) Specific heat( J kg K ) Thermal Conductivity ( w mk ) Diameter of particle (nm) Table 3: Cell number in cylinder for case 3 Grid Grid 1 Grid 2 Grid 3 Main Grid Grid 5 Cell Number 270600 416000 960000 1456446 2949442 12 Figure 1: Cases study geometry, Channel with an Inner Cylinder and without fin 13 Figure 2: channel griding, isometric view 14 Figure 3: variations of Nusselt number at channel exit under constant heat flux condition 15 Figure 4: axial variations of local convective heat transfer coefficient for Al2O3 nano-fluid in 3% volume concentration, nano-particles of 35 nm and Re = 1460. 16 Figure 5: axial variations of (a) convective heat transfer coefficient and (b) local Nusselt number of the base fluid with respect to the Reynolds number. 17 Figure 6: axial variations of (a) local convective heat transfer coefficient and (b) local Nusselt number of Al2O3 nano-fluid versus Reynolds number under 1% volume concentration and nano-particles diameter of 25 nm. 18 Figure 7: variations of mean convective heat transfer coefficient of Al2O3 nano-fluid versus Reynolds number for five different volume concentrations of nano-particles having 25 nm diameter. 19 Figure 8: axial variations of (a) local convective heat transfer coefficient and (b) local Nusselt number of Al2O3-water nano-fluid versus nano-particles volume concentration at Re=500. 20 Figure 9: axial variations of (a) local convective heat transfer coefficient and (b) local Nusselt number of Al2O3-water nano-fluid versus nano-particles volume concentration at Re=1000. 21 Figure 10: variations of convective heat transfer coefficient for Al2O3-water nano-fluid versus volume concentration under nano-particles diameter of 25 nm and Re=100, 250, 500, 750 and 1000. 22 Figure 11: axial variations of convective heat transfer coefficient against size and volume concentration of nano-particles at Re=500. 23 Figure 12: variations of convective heat transfer coefficient and mean Nusselt number against size and volume concentration of nano-particles at Re=500. 24 Figure 13: variations of Al2O3-water pressure drop versus Reynolds number at five nano- particles volume concentrations having diameter of 25 nm. 25 Figure 14: variations of pressure drop for Al2O3-water against volume concentration at two different diameters and Reynolds numbers. 26 Figure 15: variations of (a) convective heat transfer coefficient and (b) Nusselt number against separating sheet shape at Re=1000. 27 References [1] H. Alipour, A. Karimipour, M. R. Safaei, D. T. Semiromi, O. A. Akbari, Influence of t-semi attached rib on turbulent flow and heat transfer pa- rameters of a silver-water nanofluid with different volume fractions in a three-dimensional trapezoidal microchannel, Physica E: Low-dimensional Systems and Nanostructures 88 (2017) 60–76. [2] A. A. R. Darzi, M. Farhadi, A. M. Lavasani, Two phase mixture model of nano-enhanced mixed convection heat transfer in finned enclosure, Chem- ical Engineering Research and Design 111 (2016) 294–304. [3] M. I. Hasan, Investigation of flow and heat transfer characteristics in micro pin fin heat sink with nanofluid, Applied Thermal Engineering 63 (2) (2014) 598–607. [4] R. L. Webb, N. Kim, Enhanced heat transfer, Taylor and Francis, NY. [5] M. Sohel, S. Khaleduzzaman, R. Saidur, A. Hepbasli, M. Sabri, I. Mah- bubul, An experimental investigation of heat transfer enhancement of a minichannel heat sink using al 2 o 3–h 2 o nanofluid, International Journal of Heat and Mass Transfer 74 (2014) 164–172. [6] O. A. Akbari, M. R. Safaei, M. Goodarzi, N. S. Akbar, M. Zarringha- lam, G. A. S. Shabani, M. Dahari, A modified two-phase mixture model of nanofluid flow and heat transfer in a 3-d curved microtube, Advanced Powder Technology 27 (5) (2016) 2175–2185. [7] S. Moshizi, A. Malvandi, D. Ganji, I. Pop, A two-phase theoretical study of al 2 o 3–water nanofluid flow inside a concentric pipe with heat gen- eration/absorption, International Journal of Thermal Sciences 84 (2014) 347–357. [8] A. Esmaeilnejad, H. Aminfar, M. S. Neistanak, Numerical investigation of forced convection heat transfer through microchannels with non-newtonian nanofluids, International Journal of Thermal Sciences 75 (2014) 76–86. [9] A. Chengara, A. D. Nikolov, D. T. Wasan, A. Trokhymchuk, D. Hender- son, Spreading of nanofluids driven by the structural disjoining pressure gradient, Journal of colloid and interface science 280 (1) (2004) 192–201. [10] R. B. Mansour, N. Galanis, C. T. Nguyen, Effect of uncertainties in physi- cal properties on forced convection heat transfer with nanofluids, Applied Thermal Engineering 27 (1) (2007) 240–249. [11] C. T. Nguyen, G. Roy, C. Gauthier, N. Galanis, Heat transfer enhance- ment using al 2 o 3–water nanofluid for an electronic liquid cooling system, Applied Thermal Engineering 27 (8) (2007) 1501–1506. 28 [12] S. J. Palm, G. Roy, C. T. Nguyen, Heat transfer enhancement with the use of nanofluids in radial flow cooling systems considering temperature- dependent properties, Applied Thermal Engineering 26 (17) (2006) 2209– 2218. [13] S. Shi, C.-q. Yan, G.-l. Niu, Numerical study of heat transfer and pressure drop of integral pin-fin tubes, in: Power and Energy Engineering Confer- ence (APPEEC), 2011 Asia-Pacific, IEEE, 2011, pp. 1–4. [14] N. Mir, K. Syed, M. Iqbal, Numerical solution of fluid flow and heat transfer in the finned double pipe, Journal of Research (Science) 15 (3) (2004) 253– 262. [15] K. Goudarzi, H. Jamali, Heat transfer enhancement of al 2 o 3-eg nanofluid in a car radiator with wire coil inserts, Applied Thermal Engineering 118 (2017) 510–517. [16] Z. Ding, S. Cheah, N. H. Saeid, Parametric study of heat transfer enhance- ment using nanofluids, in: Energy and Environment, 2009. ICEE 2009. 3rd International Conference on, IEEE, 2009, pp. 294–298. [17] C.-J. Ho, L. Wei, Z. Li, An experimental investigation of forced convec- tive cooling performance of a microchannel heat sink with al 2 o 3/water nanofluid, Applied Thermal Engineering 30 (2) (2010) 96–103. [18] S. Akbarzadeh, M. Farhadi, K. Sedighi, M. Ebrahimi, Experimental in- vestigation on the thermal conductivity and viscosity of zno nanofluid and development of new correlations, Transport Phenomena in Nano and Micro Scales 2 (2) (2014) 149–160. [19] J. Zhang, Y. Diao, Y. Zhao, Y. Zhang, An experimental investigation of heat transfer enhancement in minichannel: Combination of nanofluid and micro fin structure techniques, Experimental Thermal and Fluid Science 81 (2017) 21–32. [20] E. Khodabandeh, A. Abbassi, Performance optimization of water-al2o3 nanofluid flow and heat transfer in trapezoidal cooling microchannel using constructal theory and two phase eulerian-lagrangian approach, Powder Technology 323 (2018) 103–114. [21] E. Khodabandeh, A. Rahbari, M. A. Rosen, Z. N. Ashrafi, O. A. Akbari, A. M. Anvari, Experimental and numerical investigations on heat transfer of a water-cooled lance for blowing oxidizing gas in an electrical arc furnace, Energy Conversion and Management 148 (2017) 43–56. [22] H. Aminfar, M. Mohammadpourfard, Y. N. Kahnamouei, A 3d numerical simulation of mixed convection of a magnetic nanofluid in the presence of non-uniform magnetic field in a vertical tube using two phase mixture model, Journal of Magnetism and Magnetic Materials 323 (15) (2011) 1963– 1972. 29 [23] J. Buongiorno, L.-W. Hu, S. J. Kim, R. Hannink, B. Truong, E. Forrest, Nanofluids for enhanced economics and safety of nuclear reactors: an eval- uation of the potential features, issues, and research gaps, Nuclear Tech- nology 162 (1) (2008) 80–91. [24] E. Khodabandeh, M. Pourramezan, M. H. Pakravan, Effects of excess air and preheating on the flow pattern and efficiency of the radiative section of a fired heater, Applied Thermal Engineering 105 (2016) 537–548. [25] A. Li, G. Ahmadi, Dispersion and deposition of spherical particles from point sources in a turbulent channel flow, Aerosol science and technology 16 (4) (1992) 209–226. [26] W. Ranz, W. Marshall, et al., Evaporation from drops, Chem. Eng. Prog 48 (3) (1952) 141–146. [27] P. Saffman, The lift on a small sphere in a slow shear flow, Journal of fluid mechanics 22 (2) (1965) 385–400. [28] C. T. Crowe, J. D. Schwarzkopf, M. Sommerfeld, Y. Tsuji, Multiphase flows with droplets and particles, CRC press, 2011. [29] R. S. Vajjha, D. K. Das, D. P. Kulkarni, Development of new correla- tions for convective heat transfer and friction factor in turbulent regime for nanofluids, International Journal of Heat and Mass Transfer 53 (21) (2010) 4607–4618. [30] M. Tehrani, M. H. Moshaei, A. S. Sarvestani, Network polydispersity and deformation-induced damage in filled elastomers, Macromolecular Theory and Simulations (1700038). [31] E. Khodabandeh, M. Ghaderi, A. Afzalabadi, A. Rouboa, A. Salarifard, Parametric study of heat transfer in an electric arc furnace and cooling system, Applied Thermal Engineering. [32] O. A. Akbari, D. Toghraie, A. Karimipour, Numerical simulation of heat transfer and turbulent flow of water nanofluids copper oxide in rectangular microchannel with semi-attached rib, Advances in Mechanical Engineering 8 (4) (2016) 1687814016641016. [33] D. Kim, Y. Kwon, Y. Cho, C. Li, S. Cheong, Y. Hwang, J. Lee, D. Hong, S. Moon, Convective heat transfer characteristics of nanofluids under lam- inar and turbulent flow conditions, Current Applied Physics 9 (2) (2009) e119–e123. [34] E. Khodabandeh, H. Davoodi, O. Zargar, S. A. Rozati, Evaluation of the performance and structure of an industrial bellows valve using finite ele- ment analysis, Asia Pac J 2. 30 [35] M. Tehrani, M. H. Moshaei, A. Sarvestani, Cell mechanotransduction is mediated by receptor diffusion, in: 54th annual technical meeting Society of Engineering Science, no. 10.13140/RG.2.2.34771.89124, 2017. [36] S. Asadi, A. A. Mokhtari, A. A. Suratgar, E. Khodabandeh, A. Karimi, A novel two-phased flow meter design using mems pressure meters array, in: Control, Instrumentation, and Automation (ICCIA), 2013 3rd Interna- tional Conference on, IEEE, 2013, pp. 89–94. [37] M. Tehrani, A. Sarvestani, Effect of chain length distribution on mechanical behavior of polymeric networks, European Polymer Journal 87 (2017) 136– 146. [38] M. Tehrani, M. H. Moshaei, A. Sarvestani, Revisiting the deformation- induced damage in filled elastomers: Effect of network polydispersity, arXiv preprint arXiv:1710.01846. [39] M. Tehrani, A. Sarvestani, Force-driven growth of intercellular junctions, Journal of Theoretical Biology 421 (2017) 101–111. [40] O. A. Akbari, D. Toghraie, A. Karimipour, M. R. Safaei, M. Goodarzi, H. Alipour, M. Dahari, Investigation of rib's height effect on heat trans- fer and flow parameters of laminar water–al 2 o 3 nanofluid in a rib- microchannel, Applied Mathematics and Computation 290 (2016) 135–153. [41] A. Kasaeian, F. Pourfayaz, E. Khodabandeh, W.-M. Yan, et al., Experi- mental studies on the applications of pcms and nano-pcms in buildings: a critical review, Energy and Buildings 154 (2017) 96–112. [42] M. Tehrani, Micromechanical analysis of strength of polymer networks with polydisperse structures, Ph.D. thesis, Ohio University (2017). [43] N. S. Moghaddam, A. Jahadakbar, A. Amerinatanzi, M. Elahinia, M. Miller, D. Dean, Metallic fixation of mandibular segmental defects: Graft immobilization and orofacial functional maintenance, Plastic and Re- constructive Surgery Global Open 4 (9). [44] R. Skoracki, M. Miller, A. Jahadakbar, M. T. Andani, N. S. Moghad- dam, C. Haberland, D. Dean, J. Walker, H. Karaca, M. Elahinia, Additive manufacturing of nitinol fixation hardware for reconstructing mandibular segmental defects, in: International Conference on Shape Memory and Su- perelastic Technologies, SMST 2015, ASM International, 2015. 31
1912.11322	1	1912	2019-12-24T12:43:35	Two-dimensional PS2: a promising anode material for sodium-ion batteries and a potential superconductor	[ "physics.app-ph" ]	Two dimensional materials as electrodes have shown unique advantages such as the infinite planar lengths, broad electrochemical window, and much exposed active sites. In this work, by means of density functional theory computations, we demonstrate that two dimensional PS2 with the 1T-Type structure as many two dimensional disulfides is a promising anode material for sodium ion batteries application. Different from many two dimensional disulfides (e.g. MoS2, TiS2, CrS2) compounds that are semiconducting, PS2 monolayer exhibits metallic character with considerable electronic states at the Fermi level, which can provide good electrical conductivity during the battery cycle and also suggest the potential superconductivity. Remarkably, PS2 monolayer has a considerably high theoretical capacity of 1692 mAh/g, a rather small sodium diffusion barrier of 0.17 eV, and a low average open circuit voltage of 0.18 V. These results suggest that PS2 monolayer can be utilized as a promising anode material for the application in sodium ion batteries with high power density and fast charge/discharge rates. In addition, estimated by first principle calculations, PS2 is found to be an intrinsic phonon-mediated superconductor with a relatively high critical superconducting temperature of about ~10 K.	physics.app-ph	physics	Two-dimensional PS2: a promising anode material for sodium-ion batteries and a potential superconductor Dawei Zhoua, Xin Tangb, Chunying Pua,* aCollege of Physics and Electronic Engineering,Nanyang Normal University,Nanyang 473061, China bCollege of Material Science and Engineering,Guilin University of Technology,Guilin 541004, China Abstract Two dimensional materials as electrodes have shown unique advantages such as the infinite planar lengths, broad electrochemical window, and much exposed active sites. In this work, by means of density functional theory computations, we demonstrate that two dimensional PS2 with the 1T-Type structure as many two dimensional disulfides is a promising anode material for sodium ion batteries application. Different from many two dimensional disulfides (e.g. MoS2, TiS2, CrS2) compounds that are semiconducting, PS2 monolayer exhibits metallic character with considerable electronic states at the Fermi level, which can provide good electrical conductivity during the battery cycle and also suggest the potential superconductivity. Remarkably, PS2 monolayer has a considerably high theoretical capacity of 1692 mAh/g, a rather small sodium diffusion barrier of 0.17 eV, and a low average open circuit voltage of 0.18 V. These results suggest that PS2 monolayer can be utilized as a promising anode material for the application in sodium ion batteries with high power density and fast charge/discharge rates. In addition, estimated by first principle calculations, PS2 is found to be an intrinsic phonon-mediated superconductor with a relatively high critical superconducting temperature of about ~10 K. †Correspondence should be addressed to: Chunying Pu ([email protected].) 1. Introduction As the clean energy storage technologies, Li-ion batteries (LIBs) exhibit many advantages such as high energy density, great power density, structural flexibility and stability, environmentally friendly and so on.[1-5] So LIBs have achieved great commercial success especially in the portable device market in past ten years.[6-7] However, the lack of lithium resources in the earth and the limited storage capacity impeders its further applications in large-scale.[8] In order to meet the demand for the next generation metal-ion batteries, extensive efforts have been made to develop new metal ion batteries.[9-17] Among various anode material candidates for metal ion batteries, two dimensional sodium-ion batteries (SIBs) have been paid more attentions.[12-17] Firstly, sodium is more abundant and more safety than lithium. Secondly, due to a large surface-area-to-volume ratio, two dimensional (2D) materials have a larger contact area between the electrolyte and electrode, which usually favor the improvement of energy density. So 2D SIBs are believed to be promising candidates for the next generation anode materials in metal ion batteries. Unfortunately, the high-performance anode materials in well developed LIBs generally are not applicable to SIBs because of the larger atomic radius of Na compared with Li. For example, by using Na instead of Li, the storage capacity of graphite falls from 372 mAh/g to just 35 mAh/g.[18,19] Therefore, design and preparation of 2D SIBs anode materials with desirable properties is indispensable. Theoretical calculations based first-principle method play an important role in understanding the charge/discharge mechanisms and electronic properties of electrode materials. To date, a tremendous number of 2D materials, including graphene systems,[20,21] phosphorene,[22-23] MXene,[24-25] transition-metal dichalcogenides and nitrides,[26-30] have been predicted to exhibit excellent performance in SIBs. For example, phosphorene as anode in SIBs achieves the theoretical capacity of 865 mAh/g,[22] and borophene gain a maximum theoretical capacity of 1984 mAh/g.[31] Other 2D materials which are studied by DFT calculations as potential anode materials for SIBs are defective graphene (1450 mAh/g),[32] borocarbonitride based anode (810 mAh/g),[33] silicene, germanene and stanene (954, 369, and 226 mAh/g, respectively),[34,35] B-doped graphene (762 mAh/g),[36] double layer graphene- phosphorene hybrid (DG/P) (372 mAh/g).[37] The ion diffusion of most these 2D materials is roughly between 0.1 to 0.6 eV. So searching 2D materials with good performance for applications in SIBs still face great challenge such as volume expansion problem, capacity restrict, diffusion barrier issue. Designing more 2D materials suitable for SIBs are still necessary. In this work, through first-principle calculation we study the PS2 monolayer as two-dimensional and investigate their electronic properties for superior anode materials of NIBs. The stability of PS2 monolayer is confirmed through the phonon spectra, ab initio molecular dynamics (AIMD) simulations, and in-plane stiffness constants. Our calculations to the electronic properties show that the PS2 monolayer is metallic, which is advantageous for the applications in Na-ion batteries. The storage capacity of the PS2 monolayer as Na-ion material is 1692 mAh/g, corresponding the ion diffusion barrier of 0.17 eV and the open current voltage of 0.18 V. Furthermore, we find that the PS2 monolayer is also exhibits superconducting behavior with the ~10 K superconductivity transition temperature by the calculations of electron-phonon coupling. 2. Computational Methods The first-principles calculation are done with the projector augmented wave (PAW) method[38,39] as implemented in the Vienna ab initio simulation package †Correspondence should be addressed to: Chunying Pu ([email protected].) (VASP).[40,41] The electron exchange-correlation energy was treated within the generalized gradient approximation (GGA), using the functional of Perdew, Burke, and Ernzerhof (PBE).[42] The energy cutoff of the plane wave was set to 380 eV and the Brillouin zone was sampled with a 12×12×1 Monkhorst-Pack k-point grid. All the atomic positions are fully optimized with the convergence of 10−5 eV and 10−3 eV/ Å for energy and force, respectively. A large vacuum space of 35 Å in the perpendicular direction of the sheet is used to avoid the interactions between periodic images. Phonon dispersion calculations were based on a supercell approach as used in the Phonopy code.[43] We take the 3×3 supercell for calculating the phonon spectra of the PS2 monolayer. In order to determine the dynamical stability of the PS2 monolayer, the thermal stability was analyzed by ab initio molecular dynamics (AIMD) simulations using the canonical ensemble (NVT) with a 3×3×1 supercell. In the calculation of sodium ions diffusion, we used nudged elastic band (NEB) method to get the ion diffusion/barrier.[44] The electron-phonon coupling (EPC) constants λ and the superconducting transition temperature were calculated in the framework of density functional perturbation theory as implemented in the QUANTUM ESPRESSO codes.[45] The ultrasoft pseudopotentials[46] and the GGA exchange-correlation potentials[42] were used to model the electron-ion interactions. After the full convergence test, the energy cutoff of the plane wave was set to be 60 Ry, and the Brillouin-zone mesh of 24 × 24 × 1 points for the self-consistent electron density calculation was used. The EPC coefficients were further calculated with a 12× 12 × 1 mesh of q-points. 3. Results and discussion Monolayer PS2 is one of the AB2 structure[47] and we found that it is energetically more stable in its 1T phase than its 1H phase. As shown in Figure 1a, the structure of PS2 consist of three atomic sub-planes similar to 1T-MoS2. The sub-plane of P atoms is sandwiched between the two sub-planes of sulfur atoms. The optimized lattice constants are a=b=3.288 Å with a layer thickness of 2.72 Å and the distance of P-S is 2.334 Å. To evaluate the chemical bonding, we computed the charge difference density, which is defined as the total electronic density of the PS2 monolayer minus the electron density of isolated P and S atoms at their respective positions. It is obviously seen that the non-polar covalent bonding character is evidenced by the presence of the electron density between P-S (Fig.1b). According to the Bader charge population analysis, the P atom and S atom in PS2 monolayer possess a charge of 0.88 and -0.44 e, respectively. Figure 1. (a)The structure of two dimensional PS2 monolayer. P and S atoms are represented by gray and yellow spheres, respectively. (b) Difference charge density of PS2. The gold color (i.e., 0.005 eÅ−3) in the plot indicates an electron density increase in the electron density after bonding, and the cyan color (i.e., 0.005 eÅ−3) indicates an loss. (c) Phonon dispersion curves of PS2 monolayer. (d) Fluctuation of total potential †Correspondence should be addressed to: Chunying Pu ([email protected].) energy of the PtS2 during the AIMD simulation at 600 K. The inset is the structure of PS2 monolayer at the end of the AIMD simulation. To explore the energetic stability of the PS2 monolayer, the binding energy is calculated, which is defined by the following formula: Eb=(EP+2ES−EPS2)/3 (1) where EP(ES) and EPS2 represents the total energies of a single P(S) atom and PS2 monolayer, respectively. The binding energy of the PS2 monolayer is 4.54 eV/atom, higher than that of silicene and germanene (3.98 and 3.26 eV/atom, respectively),[48] suggesting that the P-S bond in PS2 monolayer is robust. The dynamical stability of PS2 monolayer was tested by calculating the phonon dispersion curves along the high- symmetry paths. As shown in Fig.1c, all vibrational modes are found to be real, confirming that PS2 monolayer is dynamically stable. The highest frequency of the optical mode is up to 17.24 THz (~575cm-1), which can comparable to those of black phosphorene (~450 cm-1), [49] and MoS2 (~500 cm-1), [50] indicating the strong bonding characteristic in the PS2 monolayer. To investigate the thermal stability of the PS2 under ambient conditions, we performed AIMD simulations in NVT, running for 10 ps at 600 K with a time step of 1 fs. The fluctuation of the total potential energy with simulation time is plotted in Figure 1d, which shows that the average value of the total potential energy remains nearly constant during the entire simulation. The structure of PS2 at the end of the simulation is also plotted in Figure 1d, revealing that the structure does not experience serious structure disruptions, which confirms that PS2 monolayer possesses good thermal stability and can maintain structural stability at temperature of 600 K. We further examine the mechanical stability of PS2 monolayer by calculating its linear elastic constants. The elastic constants of PS2 are C11=C22=78 Nm-1, C12=45 Nm-1, and C66=16.5 Nm-1, respectively, which meet the necessary mechanical equilibrium conditions[51] for mechanical stability: C C 11 22 C− 2 12 > and 0 11C , 22C , C66 ＞0, confirming the mechanical stability of PS2. We also calculate the in-plane Young's modulus (Y) of PS2, which is defined as: Y=(C11−C122)/C22. The Young's modulus of PS2 is about 52 N/m. It is worth noting that this value is lower than that of silicene (62 N/m) and TiS2 (74 N/m). [52,53] Therefore, the structure of PS2 has better mechanical flexibility, which is propitious to the manufacture of flexible battery materials. Fig.2 Electronic band structure calculated and projected density of states (DOSs) calculated by using the PBE functional. The good electrical conductivity is essentially required for the excellent electrochemical performance of electrode materials and superconductivity. However, most of 2D disulfides(e.g. MoS2, TiS2, CrS2)[53,54] are semiconductor with poor conductivity, which limits the electrochemical reactions and the generation of superconductivity. To better understand the nature of the electronic properties of PS2, we calculate its electronic band structures and projected density of states. As shown in Figure 2, PS2 exhibits metallic character with severe energy levels crossing the Fermi †Correspondence should be addressed to: Chunying Pu ([email protected].) level. From the projected DOS analysis, the metallic states at the Fermi level are mainly contributed by S-2p states. Furthermore, there is also the contribution of P-2s orbitals at the Fermi level, indicating the formation of covalent bonds between P-S, which is consistent with the charge difference density analysis give above. Therefore, the metallic PS2 monolayer is not only a highly desirable anode materials for SIBs but also imply a potential superconductor. To estimate the superconducting transition temperature TC, we use the Allen- Dynes formula [55] coupling constant and the electron-electron Coulomb repulsion parameter, . (4) TC =ωlog1.20exp (− 1.04(1+λ) λ−𝜇∗(1+0.62λ) (2) where ωlog, λ , 𝜇∗are the logarithmic average of the phonon energy, electron-phonon ωlog=exp(2λ∫ α2F(ω)logωdωω ∞0 λ=2∫ α2F(ω)ω dω The electron-phonon spectral function α2F(ω) appear in the expression, which is the α2F(ω)= 12πN(εF)∑ δ(ω−ωqv) γqνℏωqν (5) where N(εF), γqν are the DOS at the Fermi level, and the linewidth of phonon mode ν at the wave vector q. As shown in Figure 3, the Eliashberg spectral function α2F(ω) and the electron-phonon coupling constant λ(ω) as functions of the phonon energy ω as well as the phonon dispersion and projected phonon density of states are plotted. We found λ=0.76, ωlog=238 meV, and Tc=10.0 K (setting µ∗ to 0.1) for 2D central quantity in the superconductivity theory can be calculated in terms of the PS2. Since the modes having the same symmetry can be mixed with each other, the phonon linewidth as qν respectively. The first two quantities were calculated as: ) (3) ∞0 phonon eigenvectors have a strongly mixed character of P and S atoms. As clearly phonons with energies from 0 to 280 cm-1. Due to the strong P-S interaction, we seen from Figure 3c, the contribution to the EPC constant λ come from the energy states to produce superconductivity in PS2 monolayer. The Tc is about 10.0 K for PS2, expect the out-of-plane vibrational modes sulfur coupled with the π* and interlayer which is higher than lithium and calcium doped graphene (8.1 and 1.4 K, respectively)[56, 57] and compared to Li-intercalated bilayer MoS2 superconductors (10.2 K). [58], but lower than intimate boron sheets (~12 -- 21 K)[59,60] and B2C (14.3 -- 19.2 K). [61] The comparatively strong EP coupling λ and higher DOS at the Fermi level in the PS2 monolayer are both favorable for the generation of superconductivity. Figure 3 (a) Calculated phonon dispersion, (b) Phonon density of states, and (c) Eliashberg function α2F(ω) with interaged EP coupling constant λ(ω) for PS2 monolayer. Since the PS2 monolayer is expected to be a potential anode materials in SIBs in view of its inherent metallicity, we firstly investigated the adsorption of a single Na atom on the surface of PS2 monolayer by constructing a 2×2×1 supercell associated with the chemical stoichiometry of NaP4S8. The adsorption energy of Na is defined as: Ead=EPS2Na−EPS2−µNa (6) †Correspondence should be addressed to: Chunying Pu ([email protected].) where EPS2Na and EPS2 are the total energies of the Na adsorbed PS2 monolayer and pristine PS2 monolayer, respectively, µNa is the chemical potential of Na and is taken as the cohesive energy of bulk Na. The negative value of adsorption energy means that the Na atom prefers to be adsorbed on the monolayer instead of forming a bulk metal and more favorable interaction between PS2 monolayer and Na. Considering the lattice symmetry of PS2 monolayer, three possible adsorption sites are considered, as shown in Figure 1a. After our geometrical optimization, three sites are remained and their adsorption energies are -0.89, -0.80, and -0.44 eV for V-, H-, and T-sites, respectively. The adsorption energies of three sites are negative, implying that Na atom prefers to be adsorbed on the host materials instead of forming a cluster. To further understand the adsorption of Na atom, we make the Bader charge analysis and the Na atoms possess a charge of 0.82 and 0.83 e at V- and H-sites, respectively, which means that the charge transfer from Na atom to adjacent sulfur atoms. The existence of charge transference by Na atom reveals that the adsorption is chemical and can be regarded as redox reaction during the battery operation. The existence of chemical adsorption can be confirmed by the charge density difference (Figure 4c and Figure 4d), which is defined by ∆ρ=ρ(NaPS2)−ρ(Na)−ρ(PS2). The density of states of the PS2 after adsorption of Na atom with V- and H-sites are also calculated, as shown in Figure 4a and 4b, the results show that the system still keeping metallic character, which is benefit for making electrode materials from the PS2 monolayer. Figure 4 The density of states of NaPS2 with Na atom at (a) V-site and (b) H-site. The charge density difference with the adsorption of Na atom at (c) V-site and (d) H-site. The gold color (i.e., 0.005 eÅ−3) in the plot indicates an electron density increase and the cyan color (i.e., 0.005 eÅ−3) indicates an loss. The Na ion diffusion energy barrier has a great effect on the charging and circuit rate capacity of SIBs. According to the above results, the data of adsorption energy shows a small difference of a Na ion at V- and H-sites (0.09 eV), which is due to existence of a similar S environment at their adsorption sites. So the Na ions are expected to have a minimum energy path for Na diffusion from a V-site to the nearest neighboring V-site via the H-site (path I). Furthermore, the path from V-site to adjacent V-site directly is taken into consideration (path II). The paths and the relative energy profile are shown in Figure 5, where the adsorption energy of Na atom at the V-site is taken as reference. The calculated diffusion barrier of Na ion along the path I †Correspondence should be addressed to: Chunying Pu ([email protected].) is 0.17 eV, which is the lowest of two possible circumstances. It can be explained that this path can reduces the influence of the energy variation at different sites. Figure 5 Relative energy profile for the diffusion of Na ion on the surface of monolayer PS2 along path I (a) and path II (b). The inset is the top views of the trajectory of Na ion diffusion over the surface of PS2 monolayer. The important parameters of NIBs as electrode materials are the open circuit voltage (OVC) and theoretical storage capacity. The theoretical storage capacity is directly concerned with the number of adsorbed atoms. The intrinsic advantage of the monolayer materials is double Na storage capacity through adsorbing multilayer Na on both sides. The average adsorption energy layer by layer can be obtained by: of Na atom. The negative adsorption energy means the adsorption of x layers is Ea=(ENa8xP4S8+Na8(x−1)P4S8−8ENa)/8. Here x represents the number of layers accessible and the maximum storage capacity can be obtained by CM= mFMPS2(mAhg-1) mAhg-1) is the Faraday constant, and MPS2 is the molar mass of PS2 per formula unit. where m is the number of adsorbed Na ions on the PS2 per formula unit , F (26801 As shown in Figure 6a, three layers of Na atom on each sides for Na ions adsorption on 2×2×1 supercell. The first Na atom layer is located at the V-site and the average adsorption energy is -0.45 eV. For the second layer, Na atoms prefer to be adsorbed at the T-site and the average adsorption energy becomes -0.06 eV. As for the third layer, Na atoms positions are the same as the first layer and the average adsorption energy is -0.02 eV. The adsorption of three-layer Na atoms on each side of PS2 monolayer can be understand by the distribution of the dispersive electron cloud (Figure 5b) acting as S ions. Electron clouds distributed around Na ion can effectively alleviated repulsive interactions between Na ions. The electron cloud and the negative adsorption energy indicate that at least two layers of Na atoms can be adorbed on the PS2 monolayer, which make the PS2 monolayer a high capacity. The maximal theoretical capacity of the PS2 monolayer can reach 1692 mAhg-1, which is only lower than the capacity for borophene (1984 mAhg-1),[31] but is even larger than other reported 2D disulfides (e.g.146 mAh/g for MoS2,[62] 466 and 233 mAh/g for VS2,[63,64] 479 mAh/g for TiS2 [53]) and comparable to that of NiC3(~1698 mAhg-1).[65] During the Na ions intercalation process, the lattice constants in the x-y plane only experience a tensile strain about 12.7%, which can comparable to the typical values that below 10% are acceptable. Figure 6 (a)The side view of the atomic structure of Na-intercalated PS2 monolayer, where three Na layers adsorbed on each side monolayer. (b) The electron local †Correspondence should be addressed to: Chunying Pu ([email protected].) function (ELF) of PS2 monolayer with three-layers of Na atom adsorbed on each side. In addition, open circuit voltage was also computed to estimate the performance of the PS2 monolayer as an anode material. A low OVC of the anode imply the possibility of a high net cell voltage. The charge/discharge process of PS2 monolayer change of volume and entropy during the adsorption process are neglected, the can be described as PS2+nNa++ne−↔NanPS2 . For this reaction, when the average open circuit voltage can be defined by Vave=(EPS2+nENa−ENanPS2)/ne, where EPS2 and ENanPS2 are the total energies of the PS2 monolayer before and after the adsorption of Na atom, ENa is the energy per Na atom in its stable bulk structure of Na metal, n is number of adsorbed Na content on a 2×2×1 supercell of PS2 monolayer. With the increase of the adsorbed Na concentration from 8 to 24 atoms on the 2×2×1 supercell, the OVC decreases from 0.45 to 0.18 V. The dropping voltage with the increasing Na ion concentration has also been reported for other anode materials.[11,66] In summary, PS2 possesses excellent stability and superior qualities and ultrahigh capacity for application as anode material in SIBs. 4. Conclusion In conclusion, we have proposed the two dimensional PS2 monolayer with dynamic, thermodynamic, and mechanical stability by first-principles investigations. Its electronic properties are investigated, and it keeps metallic feature before and after the adsorption of Na atom and thus has good electric conductivity. The relatively small lattice changes during the intercalation of Na and three Na atoms adsorbed steadily in each layer on both sides of the PS2 surface reveals good recyclability and the maximum storage capacity of the PS2 monolayer can reach to 1692 mAhg-1 for NIBs, which is quite high among two dimensional materials. The calculated diffusion energy barrier of 0.17 eV for Na indicates that PS2 monolayer can possess fast charge/discharge rates for Na atom in the SIBs. Moreover, we find that the superconducting state is characterized by an electron-phonon coupling constant and a superconducting critical temperature of 10.0 K. Our calculated results shown that two dimensional PS2 monolayer can be applied as nanoscale superconductor and a promising as electrode materials, and awaits experimental confirmation. Acknowledgements This research was supported by the National Natural Science Foundation of China (grant no. 51501093); The Henan Joint Funds of the National Natural Science Foundation of China (grant nos U1904179 and U1404608); The Key Science Fund of Educational Department of Henan Province of China (No. 20B140010). †Correspondence should be addressed to: Chunying Pu ([email protected].) References [1] Y. Idota, T. Kubota, A. Matsufuji, Y. Maekawa, T. Miyasaka, Science 1997, 276,1395. [2] K.T. Nam, D.W. Kim, P. J. Yoo, C.Y. Chiang, N. Meethong, P.T. Hammond, Y. M. Chiang, A.M. Belcher, Science 2006, 312, 885. [3] F. Ning, Y. He, B. Li, H. Du, D. Zhai, F. Kang, J. Alloys Compd. 2012, 513, 524. [4] Q. Zhang, J. Wang, J. Dong, F. Ding, X. Li, B. Zhang, S. Yang, K. Zhang, Nano Energy 2015, 13, 77. [5] W. Li, Y. Yang, G. Zhang, Y.W. Zhang, Nano Lett. 2015, 15,1691. [6] J. M. Tarascon, M. Armand, Nature 2001, 414, 359. [7] D.P. Dubal, O. Ayyad, V. Ruiz, P. Gómezromero, Chem. Soc. Rev. 2015, 44, 1777. [8] M. Endo, C. Kim, K. Nishimura, T. Fujino, K. Miyashita, Carbon 2000, 38, 183. [9] S. Li , Z. Zeng, J. Yang, Z. Han, W. Hu, L. Wang, J. Ma, B. Shan, J. Xie, ACS Appl. Energy Mater. 2019, 2, 2956 [10] Z. Zhang, Y. Zhang, Y. Li, J. Lin, D. G. Truhlar, S. Huang, Chem. Mater. 2018, 30, 10, 3208. [11] A. Samad , A . Shafique, Y. H. Shin, Nanotechnology 2017, 28(17),175401. [12] A X. Zhang, Z. Yu, S. S. Wang, S. Guan, H. Y. Yang, Y. Yao, S. A. Yang, J. Mater. Chem. A 2016, 4, 15224. [13] T. Yu, Z. Zhao, L. Liu, S. Zhang, G. Yang, J. Am. Chem. Soc. 2018,140, 5962 [14] C. Zhu, X. Qu, M. Zhang, J. Wang, Q. Li, Y. Geng,Y. Ma, Z. Su, J. Mater. Chem. A 2019, 7, 13356. [15] Z. Zhao, T. Yu, S. Zhang, H. Xu, G. Yang, Y. Liu. J. Mater. Chem. A 2019, 7, 405. [16] X. Zhang, L. Jin, X. Dai, G. Chen, G. Liu, ACS Appl. Mater. Interfaces 2018, 10, 38978. [17] S. H. Zhang, B.G.Liu, Nanotechnology 2018, 29, 325401. [18] S. Ullah, P. A. Denis, F. Sato, New J. Chem. 2018, 42, 10842. [19] M. D. Slater, D. Kim, E. Lee, C. S. Johnson, Adv. Func. Mater. 2013, 23,947. [20] X. Qian, X. Gu, M. S. Dresselhaus, R. Yang, J. Phys. Chem. Lett. 2016, 7, 4744. [21] L. Shi, T. Zhao, A. Xu, J. Xu, J. Mater. Chem. A 2016, 4, 16377. [22] V. V. Kulish, O. I. Malyi, C. Persson, P. Wu, Phys. Chem. Chem. Phys. 2015, 17, 13921. [23] K. P. Hembram, H. Jung, B. C. Yeo, S. J. Pai, S. Kim, K. R. Lee, S. S. Han, J. Phys. Chem. C 2015, 119, 15041. [24] Q. Meng, A. Hu, C. Zhi, J. Fan, Phys. Chem. Chem. Phys. 2017, 19, 29106. [25] D. Er, J. Li, M. Naguib, Y. Gogotsi, V. Shenoy, ACS Appl. Mater. Interfaces 2014, 6, 11173. [26] E. Yang, H. Ji, Y. Jung, J. Phys. Chem. C 2015, 119, 26374. [27] X. Huang, Z. Zeng, H. Zhang, Chem. Soc. Rev. 2013, 42, 1934. [28] M. Mortazavi, C. Wang, J. Deng, V. B. Shenoy, N. V. Medhekar, J. Power Sources 2014, 268, 279. [29] X. Lv, W.Wei, Q.Sun, B. Huang,Y. Dai, J. Phys. D: Appl. Phys. 2017, 50, 235501. [30] A. Samad, Y.-H. Shin, ACS Appl. Mater. Interfaces 2017, 9, 29942. [31] X. Zhang, J. Hu, Y. Cheng, H.Y. Yang, Y. Yao, S.A. Yang, Nanoscale 2016, 8,15340. [32] D. Datta, J.W. Li, V.B. Shenoy, ACS Appl. Mater. Interfaces 2014, 6,1788. [33] S. Banerjee, S. Neihsial, S.K. Pati, J. Mater. Chem. A 2016, 4, 5517. [34] J. Zhu, U. Schwingenschlögl, 2D Mater. 2016, 3, 035012. [35] B. Mortazavi, A. Dianat, G. Cuniberti, T. Rabczuk, Electrochim. Acta 2016, 213, 865. [36] C. Ling, F. Mizuno, Phys. Chem. Chem. Phys. 2014, 16, 10419. [37] L. Wang , Z. Jiang , W. Li , X. Gu, L. Huang , J. Phys. D: Appl. Phys. 2017, 50 , 165501. [38] P. E. Blöchl, Phys. Rev. B: Condens. Matter Mater. Phys. 1994, 50, 17953. [39] G. Kresse, D. Joubert, Phys. Rev. B 1999, 59, 1758. [40] G. Kresse, J. Hafner, Phys. Rev. B: Condens. Matter Mater. Phys. 1993, 47, 558. [41] G. Kresse, J. Furthmuller, Phys. Rev. B: Condens. Matter Mater. Phys. 1996, 54,11169. [42] J. P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 1996, 77, 3865. [43] A. Togo, F. Oba, I. Tanaka, Phys. Rev. B 2008, 78, 134106. [44] G. Henkelman, B. P. Uberuaga , H. Jonsson, J. Chem. Phys. 2000, 113, 9901. [45] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, et al. J. Phys.: Condens. Matter. 2009, 21, 395502. [46] D. Vanderbilt, Phys. Rev. B 1990, 41, 7892 . [47] M. Fukuda, J. Zhang, Y-T Lee, T. Ozaki, Structure map of AB2 type 2D materials by high- throughput DFT calculations, arXiv:1904.06047v1. †Correspondence should be addressed to: Chunying Pu ([email protected].) [48] L. M. Yang, V. Bacic, I. A. Popov, A. I. Boldyrev, T. Heine, T. Frauenheim, E. Ganz, J. Am. Chem. Soc. 2015, 137, 2757. [49] R. Fei, A. Faghaninia, R. Soklaski, J. A. Yan, C. Lo, L. Yang, Nano Lett. 2014, 14, 6393. [50] L. F. Huang, P. L. Gong, Z. Zeng, Phys. Rev. B: Condens. Matter Mater. Phys. 2014, 90, 045409. [51] F. Mouhat, F. Coudert, Phys. Rev. B 2014, 90, 224104. [52] Y. Ding, Y. Wang, J. Phys. Chem. C 2013, 117, 18266. [53]A. Samad, A. Shafique, Y. Shin, Nanotechnology 2017, 28, 175401. [54]Houlong L. Zhuang, Michelle D. Johannes, Michael N. Blonsky, Richard G. Hennig, Appl. Phys. Lett. 2014, 104, 022116. [55] P. B. Allen, N. I. Dynes. Phys. Rev. B 1975, 12, 905. [56] E. Margine, F. Giustino, Phys. Rev. B 2014, 90, 014518. [57] Profeta, Gianni , M. Calandra , F. Mauri, Nature Physics 2012, 8,131. [58] Y. C. Chang, A. V. Kildishev , E. E. Narimanov, T. B. Norris. Phys. Rev. B 2016, 94,155430. [59] E. S. Penev, A. Kutana, B. I. Yakobson, Nano Lett. 2016, 16, 2522. [60] R. C. Xiao, D.F. Shao, W.J. Lu, H.Y. Lv, J.Y. Li, Y.P. Sun, Appl. Phys. Lett. 2016, 109, 122604. [61] J. Dai, Z. Li, J. Yang, J. Hou, Nanoscale 2012, 4, 3032. [62]M . Mortazavi, C. Wang, J. Deng, V. B. Shenoy and N. V. Medhekar, J. Power Sources 2014, 268, 279. [63] D. Wang, Y. Liu, X. Meng, Y. Wei, Y. Zhao, Q. Pang, G. Chen, J. Mater. Chem. A 2017,5, 21370. [64] D. B. Putungan, S. H. Lin , J. L. Kuo, ACS Appl. Mater. Interfaces 2016, 8,18754. [65] C. Zhu, X. Qu, M. Zhang, J. Wang, Q. Li, Y. Geng, Y. Ma, Z. Su, J. Mater. Chem. A 2019, 7, 13356. [66] P. Liang, Y. Cao, B. Tai, L. Zhang, H. Shu, F. Li, D. Chao, X. Du, J. Alloys Compd. 2017, 704, 152.
1802.01574	1	1802	2018-02-05T10:28:38	MBE grown Self-Powered \b{eta}-Ga2O3 MSM Deep-UV Photodetector	[ "physics.app-ph" ]	We demonstrate self-powered \b{eta}-Ga2O3 deep-UV metal-semiconductor-metal (MSM) photodetectors (PD) with 0.5% external quantum efficiency (EQE) at zero bias. 150 nm thick (-201)-oriented epitaxial \b{eta}-Ga2O3-films were grown on c-plane sapphire using plasma-assisted MBE. Ni/Au and Ti/Au metal stacks were deposited as contacts to achieve asymmetric Schottky barrier heights in interdigitated finger architecture for realizing self-powered photodetectors. Current-voltage characteristics (photo and dark), time-dependent photocurrent and spectral response were studied and compared with conventional symmetric MSM PD with Ni/Au as the Schottky metal contact, fabricated on the same sample. The asymmetric, self-powered devices exhibited solar-blind nature and low dark current < 10 nA at 15 V with high photo-to-dark current ratio of ~ 103. The dark and photocurrents were asymmetric with respect to the applied bias and the responsivity in the forward bias was characterized by gain. The detectors (asymmetric-MSM) were found to exhibit a responsivity of 1.4 mA/W at 255 nm under zero-bias condition (corresponding to an EQE ~ 0.5 %), with a UV-to-Visible rejection ratio ~ 102 and ~105 at 0 V and 5 V respectively.	physics.app-ph	physics	MBE grown Self-Powered β-Ga2O3 MSM Deep-UV Photodetector Anamika Singh Pratiyush1a), Sriram Krishnamoorthy2,3, Sandeep Kumar1, Zhanbo Xia2, Rangarajan Muralidharan1, Siddharth Rajan2, Digbijoy N. Nath1 a) 1Centre for Nano Science and Engineering (CeNSE), Indian Institute of Science (IISc), Bangalore 560012 2Department of Electrical and Computer Engineering, The Ohio State University, 3 Electrical and Computer Engineering, The University of Utah, Columbus, OH, 43210 Salt Lake City, UT, 84112 Abstract: We demonstrate self-powered β-Ga2O3 deep-UV metal-semiconductor-metal (MSM) photodetectors (PD) with 0.5% external quantum efficiency (EQE) at zero bias. 150 nm thick (-201)-oriented epitaxial β-Ga2O3-films were grown on c-plane sapphire using plasma-assisted MBE. Ni/Au and Ti/Au metal stacks were deposited as contacts to achieve asymmetric Schottky barrier heights in interdigitated finger architecture for realizing self-powered photodetectors. Current-voltage characteristics (photo and dark), time-dependent photocurrent and spectral response were studied and compared with conventional symmetric MSM PD with Ni/Au as the Schottky metal contact, fabricated on the same sample. The asymmetric, self-powered devices exhibited solar- blind nature and low dark current < 10 nA at 15 V with high photo-to-dark current ratio of ~ 103. The dark and photocurrents were asymmetric with respect to the applied bias and the responsivity in the forward bias was characterized by gain. The detectors (asymmetric-MSM) were found to exhibit a responsivity of 1.4 mA/W at 255 nm under zero-bias condition (corresponding to an EQE ~ 0.5 %), with a UV-to-Visible rejection ratio ~ 102 and ~105 at 0 V and 5 V respectively. a) Corresponding author email: [email protected], [email protected] β-Ga2O3 is thermodynamically the most stable phase among five other known phases (α, β, γ, δ, and, ε) of Ga2O3 with a wide bandgap of 4.6 eV 1,2, which makes it an attractive candidate for deep-UV detectors3–9 and power transistors10–15. β-Ga2O3 also offers economic advantage as conventional crystal growth techniques such edge-defined film-fed growth (EFG)16,17, float-zone18–20 and Czochralski21,22 methods can be employed towards enabling scalable and large-area single crystal wafers. Thin film growth of β-Ga2O3 on foreign substrates has also been widely reported using different growth techniques including Molecular beam epitaxy (MBE)9,23–25, Metalorganic chemical vapour deposition(MOCVD)26– 28, mist CVD29, radio frequency magnetron sputtering30,31 and microwave irradiation approach32 for different applications. β-Ga2O3-based deep-UV photodetectors with Schottky33–35 and MSM9,36–40 architectures on bulk and foreign substrate have been studied in the recent years. However, no report exists on β-Ga2O3 self-powered MSM detectors. In this letter, we report on β-Ga2O3 self-powered lateral MSM PDs (asymmetric- MSM) with external quantum efficiency (EQE) of 0.5 % at zero bias. The current-voltage (I- V) characteristics, transient response and spectral response of asymmetric MSM (A-MSM) has been studied and compared with conventional symmetric Schottky MSM (S-MSM). This work reports the first zero-bias spectral responsivity for any type of UV detector based on epitaxial and planar β-Ga2O3. Growth of β-Ga2O3 thin film on c-plane sapphire was carried out by plasma-assisted MBE equipped with a standard effusion cell for gallium and a Veeco Uni-bulb O2 plasma plasma source. Following the substrate cleaning, the sapphire substrates were indium bonded to a silicon wafer and degassed at 400 °C for 1 hour in the buffer chamber before the actual growth run. Ga2O3 was grown for 3 hours at a substrate temperature of 700 C with a Gallium flux of 1.5 X 10-8 and RF plasma power of 300 W. The β-Ga2O3 film was confirmed to be single phase (-201)-orientated with thickness of 150 nm from X-ray diffraction (XRD) and 2 X-ray reflectivity (XRR) measurements. The material characterization details have been reported earlier.9 Following standard lithographic process, Ti (20 nm)/Au (100 nm) stack was e-beam evaporated on the β-Ga2O3 films to form one side of the Schottky contact in the asymmetric MSM structure (Device: A-MSM). Post metal evaporation, a rapid thermal annealing (RTA) of Ti/Au metal stack was done at 470°C for 1 minute in nitrogen ambient resulted in improvement of the metal contacts in terms of current values. Subsequently, Ni (20 nm)/Au (100 nm) metal stack was evaporated for the other Schottky metal contact to obtain photodetectors of A-MSM geometry as shown in fig. 1(a). Conventional Schottky symmetric- MSM (Device: S-MSM) were also fabricated on the same sample with Ni (20 nm)/Au (100 nm) metal stack for both side of interdigitated architecture as shown in the fig. 1(b). Each device consisted of 36 interdigitated fingers with finger widths of 4 µm and spacing of 6 µm, resulting in an active area of 260 x 300 µm2. The photo current of the devices was measured using Sciencetech, Inc. Quantum Efficiency (QE) setup consisting of a 150-W Xenon lamp, monochromator and Keithly-2450 source meter. For intensity/power values Xenon lamp was calibrated with a standard silicon photodiode attached to the QE set-up. Figure 1(c) shows spectral response (SR) versus wavelength () at an applied bias voltage of 5 V for the A-MSM device in forward and reverse bias. SR was calculated using the expression below: (1) where, IPHOTO is the photocurrent, IDARK is the dark current, P is the optical power density, and A is the active area (A = 300 x 260 µm2). For A-MSM, the measurements in forward bias were carried out with positive potential on Ni/Au contact. The detectors exhibited a cutoff in responsivity at 253-255 nm. The peak SR values for A-MSM detector 3 APIISRDarkPHOTOCalculated.were measured to be 9.4 A/W and 1.1 A/W at 5 V in the forward (FB) and the reverse bias (RB) regimes, respectively. High internal gain in A-MSM detector was observed in the forward bias region as evident from Fig. 1(c).The zero-bias spectral responsivity is shown in the inset to figure 1(c) and a peak responsivity value of 1.4 mA/W was obtained corresponding to an EQE ~ 0.5 %, indicating its self-powered nature. It is more than an order of magnitude higher than the zero-bias responsivity of 0.01 mA/W33 reported for β-Ga2O3- nanowire Schottky detectors, which is the only report till date on zero-bias SR for any kind of Ga2O3 UV detector. The UV to visible rejection ratio of the devices in this study was estimated by dividing the responsivity at 255 nm by that at 450 nm and was found to be ~105 at 5 V and ~102 at zero bias respectively, testifying the solar-blind nature of the photodetectors. Further, voltage-dependent spectral response for both A-MSM and S-MSM was measured. Figure 2(a) shows SR versus wavelength at different applied biases in forward and reverse regions for A-MSM photodetectors. Figure 2 (b) shows the variation of peak SR (at 255 nm) with applied bias for both A-MSM and S-MSM PDs. With increasing applied biases in the forward/reverse region, the peak SR values (at 255 nm) were also found to be increasing while the overall SR in the forward bias regime was higher indicating a higher internal gain. Figure 3(a) shows the variation of photo current and dark current with voltage (I-V) for the A-MSM detectors. The photo current was measured at an illumination of 255 nm while the bias on Ni/Au was swept with the Ti/Au contact grounded. Both the photocurrent and the dark I-V exhibited asymmetric natures with more than one order of rectification indicating asymmetric Schottky barrier heights for Ni/Au and Ti/Au contacts unlike the photo and dark currents of the S-MSM detectors (Fig. 3(b)) which exhibited symmetric behavior with applied bias. The asymmetric Schottky barrier heights for A-MSM detectors arise from 4 the difference in the work functions of Ni and Ti metals used to form the interdigitated contacts on β-Ga2O3 thin film resulting in an asymmetric I-V. The photo currents at +15 V (FB) and -15 V (RB) for A-MSM detectors were thus measured to be 98 µA and 22 µA respectively while the corresponding dark currents were 100 nA and 1.7 nA. This can be contrasted with the corresponding photo (dark) current values of the S-MSM detectors (Fig. 3(b)). A-MSM devices were found to exhibit a photo to dark current ratio of ~ 103 at a bias of 5 V. An approximate calculation of the Schottky barrier height (SBH) at the junction was done from the dark I-V characteristics, under the assumption that almost all the potential drop happens at the reverse-biased junction. SBH in forward and reverse junction was extracted using the modified Schottky equation below:41 (2(a)) (2(b)) where, I0 is the dark current, n is the ideality factor, e is electronic charge, V is applied bias, k is Boltzmann constant, T is temperature in Kelvin, I is the measured current, φB is the barrier height, A is the area of the detector metal contact and A* is the Richardson constant. Based on a further simplification of the expression for MSM devices, the equation (2(a)) can be modified as below:42 (2(c)) Using the above equations 2(a), (b) and (c) and taking Richardson constant value ~ 41 A/cm2K2 for β-Ga2O3, SBHs for the A-MSM and S-MSM samples were extracted. The SBHs for A-MSM detector for Ti/Ga2O3 and Ni/Ga2O3 contacts were found to be 0.71 eV and 0.82 5 ]Texp1[Texp I0keVnkeVITexpTAA ,2*0keIandBTexp TexpI0nkeVIkeVeV respectively whereas, the SBH for Ni/Ga2O3 contacts in S-MSM detector was found to be 0.82 eV. Thus, the lower Schottky barrier height at the Ti/β-Ga2O3 junction compared to the Ni/ β-Ga2O3 junction leads to a higher measured dark current in the forward bias region for A-MSM PD. Similar observations are reported in literature for GaN and AlGaN systems.43–47 Figure 4 (a) and (b) show the band diagram for A-MSM photodetectors in forward and reverse bias conditions (under illumination) respectively. When a negative bias is applied to the Ti/β-Ga2O3 junction, the photogenerated electrons drift towards the Ni contact and holes gets accumulated at the Ni/β-Ga2O3 junction. The junction thus gets positively charged. To maintain charge neutrality, more electrons from the Ti metal will flow towards the β- Ga2O3 leading to photo-induced lowering of the Schottky barrier height 9,48,49. This results in a higher gain in the forward bias condition compared to reverse bias condition. Figure 5 (a), (b) and (c) shows the transient current characteristics for reverse biased S-MSM (at -15 V), S-MSM (at -15 V) and A-MSM (0 V) detectors respectively. Rise times (10% - 90% of value) were estimated to be 2.5 s, 2.3 and 1.9 s while the fall times were found to be 0.4 s, 0.9 and 0.5 s for S-MSM (at -15 V), S-MSM (at -15 V) and A-MSM (0 V) detectors respectively. The on/off ratio for all samples was ~ 103. The slow response time can be attributed to hole trapping at the junction and in the bulk as reported in literature earlier.9,48,49 Table I shows a comparison of β-Ga2O3-based deep UV photodetectors in terms of spectral response, dark current and self-powered ability for various device architectures as reported in the literature. Although a zero-bias responsivity value of 0.01 mA/W33 has been reported for Ga2O3-nanowire Schottky detectors, yet there has been no report on zero-bias responsivity for planar and epitaxial β-Ga2O3 UV photodetectors in either MSM or Schottky geometry other than this work. 6 In conclusion, we have demonstrated self-powered lateral MSM deep UV photodetectors based on β-Ga2O3. We have also carried out a comparative study of symmetric conventional MSM and asymmetric MSM in terms of dark current characteristics, voltage- dependent photoresponse, and time-dependent photoresponse. A spectral response of 1.4 mA/W at zero bias was demonstrated for asymmetric MSM photodetectors. Further, the spectral response analysis showed true solar-blind nature in conjunction with high photo-to- dark current ratio of ~ 103. This work reports zero-bias spectral responsivity for any type of epitaxial β-Ga2O3-based UV detector, and is expected to aid in the development of self- powered solar blind devices. Acknowledgement This work was funded by Department of Science and Technology (DST) under its Water Technology Initiative (WTI), Grant No. DSTO1519 and Space technology cell (STC/ISRO). We would also like to thank Micro and Nano Characterization Facility (MNCF) and NNFC staff at CeNSE, IISc for their help and support in carrying out his work. Z.X. and S.R. acknowledge funding from the U.S. Office of Naval Research EXEDE MURI (Program Manager: Dr. Brian Bennett). This work was supported in part by The Ohio State University Materials Research Seed Grant Program, funded by the Center for Emergent Materials, an NSF-MRSEC, Grant No. DMR-1420451, the Center for Exploration of Novel Complex Materials, and the Institute for Materials Research. 7 REFERENCES: 1 T. Onuma, S. Fujioka, T. Yamaguchi, M. Higashiwaki, K. Sasaki, T. Masui, and T. Honda, Appl. Phys. Lett. 103, 41910 (2013). 2 H.H. Tippins, Phys. Rev. 140, A316 (1965). 3 S. Nakagomi, T. Momo, S. Takahashi, and Y. Kokubun, Appl. Phys. Lett. 103, 72105 (2013). 4 T. Oshima, T. Okuno, and S. Fujita, Jpn. J. Appl. Phys. 46, 7217 (2007). 5 T. Oshima, T. Okuno, N. Arai, N. Suzuki, S. Ohira, and S. Fujita, Appl. Phys. Express 1, 11202 (2008). 6 R. Suzuki, S. Nakagomi, Y. Kokubun, N. Arai, and S. Ohira, Appl. Phys. Lett. 94, 222102 (2009). 7 S. Kumar, A.S. Pratiyush, S.B. Dolmanan, S. Tripathy, R. Muralidharan, and D.N. Nath, Arxiv ID 1709.03692 (2017). 8 T.-C. Wei, D. Tsai, P. Ravadgar, J.-J. Ke, M.-L. Tsai, D.-H. Lien, C.-Y. Huang, R.-H. Horng, and J.-H. He, IEEE J. Sel. Top. Quantum Electron. 20, 3802006 (2014). 9 A.S. Pratiyush, S. Krishnamoorthy, S.V. Solanke, Z. Xia, R. Muralidharan, S. Rajan, and D.N. Nath, Appl. Phys. Lett. 110, 221107 (2017). 10 H. Zhou, M. Si, S. Alghamdi, G. Qiu, L. Yang, and P.D. Ye, IEEE Electron Device Lett. 38, 103 (2017). 11 A.J. Green, K.D. Chabak, E.R. Heller, R.C. Fitch, M. Baldini, A. Fiedler, K. Irmscher, G. Wagner, Z. Galazka, S.E. Tetlak, A. Crespo, K. Leedy, and G.H. Jessen, IEEE Electron Device Lett. 37, 902 (2016). 12 M.H. Wong, K. Sasaki, A. Kuramata, S. Yamakoshi, and M. Higashiwaki, IEEE Electron Device Lett. 37, 212 (2016). 13 W.S. Hwang, A. Verma, H. Peelaers, V. Protasenko, S. Rouvimov, H.G. Xing, A. Seabaugh, W. Haensch, C. Van De Walle, Z. Galazka, M. Albrecht, R. Fornari, and D. Jeena, Appl. Phys. Lett. 104, 203111 (2014). 14 M. Higashiwaki, K. Sasaki, A. Kuramata, T. Masui, and S. Yamakoshi, Appl. Phys. Lett. 100, 13504 (2012). 15 K. Matsuzaki, H. Yanagi, T. Kamiya, H. Hiramatsu, K. Nomura, M. Hirano, K. Matsuzaki, and H. Honso, Appl. Phys. Lett. 88, 92106 (2006). 16 H. Aida, K. Nishiguchi, H. Takeda, N. Aota, K. Sunakawa, and Y. Yaguchi, Jpn. J. Appl. Phys. 47, 8506 (2008). 17 T. Oishi, Y. Koga, K. Harada, and M. Kasu, Appl. Phys. Express 8, 31101 (2015). 18 Y. Tomm, J.M. Ko, A. Yoshikawa, and T. Fukuda, Sol. Energy Mater. Sol. Cells 66, 369 (2001). 19 E.G. Villora, K. Shimamura, Y. Yoshikawa, K. Aoki, and N. Ichinose, J. Cryst. Growth 270, 420 (2004). 20 S. Ohira, N. Suzuki, N. Arai, M. Tanaka, T. Sugawara, K. Nakajima, and T. Shishido, Thin Solid Films 516, 5763 (2008). 21 Z. Galazka, K. Irmscher, R. Uecker, R. Bertram, M. Pietsch, A. Kwasniewski, M. Naumann, T. Schulz, R. Schewski, D. Klimm, and M. Bickermann, J. Cryst. Growth 404, 184 (2014). 22 K. Irmscher, Z. Galazka, M. Pietsch, R. Uecker, and R. Fornari, J. Appl. Phys. 110, 63720 (2011). 23 T. Oshima, T. Okuno, and S. Fujita, Jpn. J. Appl. Phys. 46, 7217 (2007). 24 X.Z. Liu, P. Guo, T. Sheng, L.X. Qian, W.L. Zhang, and Y.R. Li, Opt. Mater. 51, 203 (2016). 25 D. Guo, Z. Wu, P. Li, Y. An, H. Liu, X. Guo, H. Yan, G. Wang, C. Sun, L. Li, and W. Tang, Opt. Mater. Express 4, 1067 (2014). 26 M.J. Tadjer, M.A. Mastro, N.A. Mahadik, M. Currie, V.D. Wheeler, J.A.F. Jr, J.D. Greenlee, J.K. Hite, K.D. Hobart, C.R.E. Jr, and F.J. Kub, J. Electron. Mater. Electron. Mater. 45, 2031 (2016). 27 W. Mi, J. Ma, Z. Zhu, C. Luan, Y. Lv, and H. Xiao, J. Cryst. Growth 354, 93 (2012). 28 V. Gottschalch, K. Mergenthaler, G. Wagner, J. Bauer, H. Paetzelt, C. Sturm, and U. Teschner, Phys. Status Solidi A 206, 243 (2009). 29 K. Kaneko, H. Ito, S. Lee, and S. Fujita, Phys. Status Solidi C 10, 1596 (2013). 30 Y. Wei, Y. Jinliang, W. Jiangyan, and Z. Liying, J. Semicond. 33, 73003 (2012). 31 Z. Wu, G. Bai, Q. Hu, D. Guo, C. Sun, L. Ji, M. Lei, L. Li, P. Li, J. Hao, and W. Tang, Appl. Phys. Lett. 106, 171910 (2015). 32 P. Jaiswal, U. Muazzam, A.S. Pratiyush, N. Mohan, S. Raghavan, R. Muralidharan, and D.N. Nath, Appl. Phys. Lett. 112, 021105 (2018) 33 X. Chen, K. Liu, Z. Zhang, C. Wang, B. Li, H. Zhao, D. Zhao, and D. Shen, Appl. Mater. Interfaces 8, 4185 (2016). 34 T. Oshima, T. Okuno, N. Arai, N. Suzuki, S. Ohira, and S. Fujita, Jpn. J. Appl. Phys. 1, 11202 (2008). 35 R. Suzuki, S. Nakagomi, Y. Kokubun, N. Arai, and S. Ohira, Appl. Phys. Lett. 94, 222102 (2009). 36 D.Y. Guo, Z.P. Wu, P.G. Li, Y.H. An, H. Liu, X.C. Guo, H. Yan, G.F. Wang, C.L. Sun, L.H. Li, and W.H. Tang, Opt. Mater. Sexpress 4, 1067 (2014). 37 D. Guo, Z. Wu, P. Li, Q. Wang, and M. Lei, RSC Adv. 5, 12894 (2015). 38 G.C. Hu, C.X. Shan, N. Zhang, M.M. Jiang, S.P. Wang, and D.Z. Shen, Opt. Express 23, 13554 (2015). 39 T. Oshima, T. Okuno, and S. Fujita, Jpn. J. Appl. Phys. 46, 7217 (2007). 8 40 W.Y. Weng, T.J. Hsueh, S. Chang, G.J. Huang, and S.C. Hung, IEEE Trans. Nanotechnol. 10, 1047 (2011). 41 E. H. Rhoderick and R. H. Williams, Metal–Semiconductor Contacts ~Oxford University Press, New York, 1988. 42 S. Averine, Y.C. Chan, and Y.L. Lam, Appl. Phys. Lett. 77, 274 (2000). 43 M. Brendel, M. Helbling, A. Knigge, F. Brunner, and M. Weyers, Electron. Lett. 51, 1598 (2015). 44 M. Brendel, F. Brunner, and M. Weyers, J. Appl. Phys. 122, 174501 (2017). 45 X. Chen, K. Liu, Z. Zhang, C. Wang, B. Li, H. Zhao, D. Zhao, and D. Shen, Appl. Mater. Interfaces 8, 4185 (2016). 46 D. Li, X. Sun, H. Song, Z. Li, H. Jiang, Y. Chen, G. Miao, and B. Shen, Appl. Phys. Lett. 99, 261102 (2011). 47 X. Sun, D. Li, Z. Li, H. Song, H. Jiang, Y. Chen, G. Miao, and Z. Zhang, Scientific Reports. 5, 16819 (2015). 48 A.M. Armstrong, M.H. Crawford, A. Jayawardena, A. Ahyi, and S. Dhar, J. Appl. Phys. 119, 103102 (2016). 49 O. Katz, V. Garber, B. Meyler, G. Bahir, and J. Salzman, Appl. Phys. Lett. 79, 1417 (2001). 9 Figure and Tables: (a) (b) Figure 1: Schematic of (a) asymmetric-MSM (A-MSM) and (b) symmetric-MSM (S-MSM) device architecture respectively. (c) Spectral responsivity versus wavelength at an applied bias of 5 V for A-MSM detectors in forward and reverse bias condition. Inset of the figure 1(c) shows spectral response at zero bias for A-MSM photodetectors (EQE= 0.5%). Fig.2 (a) Variation of spectral responsivity with the wavelength as a function of applied bias for A-MSM photodetector (Forward bias and Reverse bias) in log scale. (b) Peak response values (at 255 nm) versus bias voltages for A-MSM and S-MSM detectors. 10 Sapphire Substrate150 nm β-Ga2O3Ni/AuTi/AuSapphire Substrate150 nm β-Ga2O3Ni/AuNi/Au24028032036040044010-410-310-210-1100101(c)Spectral Responsivity (A/W)Wavelength (nm) A-MSM (FB) A-MSM (RB)Log Scale (5 V)2403003604200.00.40.81.21.6Linear ScaleA-MSM SR (mA/W)Wavelength (nm) 0 V24028032036040044010-410-310-210-1100101102 15 V FB 10 V FB 15 V RB 10 V RB 5 V RBSR (A/W)Wavelength (nm)(a)Forward Bias (FB)Log ScaleA-MSMReverse Bias (RB)-15-10-50510151x10-51x10-41x10-31x10-21x10-11x1001x1011x102Peak SR (A/W)Voltage (V)(b) A-MSM S-MSM Fig.3 (a) Photo and dark I-V characteristics at room temperature for A-MSM detector (log scale). (b) Photo and dark I-V characteristics at room temperature for S-MSM detector (log scale). Illumination was done at 255 nm. Fig.4 (a) Schematic of band diagram under UV illumination (255 nm) under (a) forward bias and (b) reverse bias condition for A-MSM photodetectors. (EFNi and EFTi denotes fermi level at different junction). Fig. (a) Photo induced lowering of barrier at metal-semiconductor interface by ΔφB. (Dotted line represents new lowered barrier height after UV illumination in forward bias). 11 Ni/AuIllumination-Electron++E FNiTi/AuE FTiHole+++----++(a)ΔφB+_Ni/AuIllumination--Electron++E FNiTi/AuE FTiHole--++(b)+_--15-10-505101510-1310-1210-1110-1010-910-810-710-610-510-410-3(b)S-MSM Current (A)Voltage (V) Photocurrent (255 nm) Dark Current-15-10-505101510-1210-1110-1010-910-810-710-610-510-4A-MSM Current (A)Voltage (V) Photocurrent (255 nm) Dark current Fig.5 Time-dependent photo-response (transient) under the illumination of 255 nm (a) reverse biased S-MSM at 15 V, (b) reverse biased A-MSM at 15 V and (c) A-MSM at zero bias respectively in linear scale. Tables: (Double column) TABLE I. List of β-Ga2O3-based deep UV detectors reported earlier for different architecture devices. Device Structure Self-Powered (Bias Voltage) SR Dark current (Bias Voltage) Reference MSM MSM Schottky MSM Schottky MSM MSM MSM Schottky A-MSM No No Yes No Yes No No No Yes Yes 37 mA/W (10 V) 1200 pA (10 V) 1.7 A/W (20 V) 620 pA (20 V) 8.7 A/W (-10 V) 10 nA (-10 V) - 128 nA (10 V) 1000 A/W (-3 V) 0.1 nA (-3 V) 0.37 mA/W (5 V) 133 pA (5 V) - 1.5 A/W (4 V) 5 nA (20 V) 7 nA (20 V) 2.9 mA /W (-50 V) 10 pA (-30 V) 39 38 34 36 35 40 37 9 33 1 mA/W (0 V) 0.3 A/W (-5 V) 7.3 nA (-15 V) This Work 12 040801201600.06.0x10-61.2x10-51.8x10-50.08.0x10-61.6x10-52.4x10-5040801201600.04.0x10-98.0x10-91.2x10-8 S-MSM (RB @ 15 V) Time (sec)(b) Transient Current (A) A-MSM ( RB @ 15 V)(c)(a) A-MSM (0 V)
1810.07612	1	1810	2018-10-17T15:28:05	Highly efficient optical transition between excited states in wide InGaN quantum wells	[ "physics.app-ph", "cond-mat.mes-hall" ]	There is a lack of highly efficient light emitting devices (LEDs) operating in the green spectral regime. The devices based on (In,Al)GaN show extremely high efficiencies in violet and blue colors but fall short for longer emission wavelengths due to the quantum confined Stark effect (QCSE). In this paper we present a design of the active region based on wide InGaN quantum wells (QWs) which do not suffer from QCSE and profit from an enhancement in the internal quantum efficiency (IQE). The design exploits highly efficient optical transitions between excited states. It is shown that, counterintuitively, the devices with higher InGaN composition exhibit a higher enhancement in IQE. Experimental evidence is provided showing a gradual change in the nature of the optical transition with increasing thickness of the QW. Moreover, optical gain in long wavelength LDs incorporating standard and wide QWs is investigated to show the utilization of our concept.	physics.app-ph	physics	Highly efficient optical transition between excited states in wide InGaN quantum wells G. Muziol,1,* H. Turski,1 M. Siekacz,1 K. Szkudlarek1, L. Janicki,2 S. Zolud,2 R. Kudrawiec,2 T. Suski,1 and C. Skierbiszewski,1,3 1Institute of High Pressure Physics PAS, Sokolowska 29/37, 01-142 Warsaw, Poland 2Faculty of Fundamental Problems of Technology, Wroclaw University of Science and Technology, Wybrzeze Wyspianskiego 27, 50-370 Wroclaw, Poland 3TopGaN Ltd, Sokolowska 29/37, 01-142 Warsaw, Poland *E-mail: [email protected] Abstract There is a lack of highly efficient light emitting devices (LEDs) operating in the green spectral regime. The devices based on (In,Al)GaN show extremely high efficiencies in violet and blue colors but fall short for longer emission wavelengths due to the quantum confined Stark effect (QCSE). In this paper we present a design of the active region based on wide InGaN quantum wells (QWs) which do not suffer from QCSE and profit from an enhancement in the internal quantum efficiency (IQE). The design exploits highly efficient optical transitions between excited states. It is shown that, counterintuitively, the devices with higher InGaN composition exhibit a higher enhancement in IQE. Experimental evidence is provided showing a gradual change in the nature of the optical transition with increasing thickness of the QW. Moreover, optical gain in long wavelength LDs incorporating standard and wide QWs is investigated to show the utilization of our concept. 1 Introduction The optoelectronic devices based on III-nitrides had found many applications including general lighting and displays thanks to their high quantum efficiency [1-7]. One of the key challenges in the design of the active region of the devices are caused by the strong built-in electric fields arising both from the spontaneous and piezoelectric polarization [8-11]. Although this built-in electric field enabled the realization doping-free high electron mobility transistors [12-16] and polarization-induced doping [17] it can be detrimental in the optoelectronic devices like light emitting devices (LEDs) and laser diodes (LDs). The active region of LEDs and LDs consists of GaN/AlGaN or InGaN/GaN multi quantum well (MQW) structures. Due to the lattice mismatch between the alloys a strong piezoelectric field is present in the QWs. It causes the spatial separation of electron and hole wavefunctions leading to red-shift of the emission spectra, called the quantum confined Stark effect (QCSE), and reduction of the wavefunction overlap [10, 18-26]. The latter leads to an increase of the carrier density because of a lower probability of carrier recombination. This is unwanted because as the carrier density increases a growing part of the carriers recombine through the nonradiative Auger process causing the reduction of the quantum efficiency [27-31]. Additionally, in case of the InGaN MQWs the higher the indium content the higher the polarization field and thus the separation of carrier wavefunctions. This is the primary reason for the loss of efficiency of the III-nitride devices in the green spectral range [5, 32]. From the point of view of reducing the impact of Auger recombination it would be ideal to reduce the carrier density. A simple increase of the active region volume by increasing the number of QWs does not provide the expected efficiency enhancement because of the non- uniform hole distribution among QWs [33]. On the other hand, the increase of the QW thickness, due to the polarization-induced separation of carrier wavefunctions, decreases the transition probability severely [19, 34]. The wider the QW the lower the wavefunction overlap is. Several methods of overcoming the limitation arising due to the built-in electric field had been proposed. The two major are growth of staggered QWs [35, 36] and growth on semipolar and nonpolar crystal orientations [37-42]. There are a few reports in the literature studying the thick InGaN QWs on the polar c- plane orientation [19, 22, 34, 43, 44] and, although the calculations predict a severe decrease of the transition probability [19, 34], some of the results show an increase in the efficiency [44]. Additionally, the current state-of-the-art violet LDs incorporate wide (6.6 nm thick) InGaN QW [45]. There clearly is a discrepancy between the theoretical understanding and the experimental results. 2 In this paper it will be shown that, counterintuitively, the LEDs and LDs with wide InGaN QWs can have higher efficiency than with the thin QWs. The enhancement in the efficiency is caused by two factors: (i) screening of the built-in electric field by carriers occupying the ground states and (ii) emergence of exited states with highly efficient optical transition. Together with the lower carrier density in the wide QWs it significantly reduces the non-radiative Auger recombination making the wide InGaN QWs a good candidate for efficient sources of light in the green spectrum. Additionally, it will be presented, contrary to the common understanding, that the higher the composition of the QW the higher this increase in the efficiency is. The scheme of this paper is as following: (1) theoretical calculations will be presented to study the influence of thickness on the optical transitions inside the QW and explicate the idea behind the increased efficiency, (2) photoluminescence spectra of QWs with various thicknesses showing the experimental evidence of emergence of excited states will be shown, (3) the influence of indium composition on internal quantum efficiency (IQE) will be studied theoretically and (4) a comparison of optical gain of blue and cyan LDs with thin and wide QW will be shown to validate the predicted change in the efficiency with the InGaN composition. Appearance of efficient transitions through excited states in wide InGaN QWs The intriguing effect which triggered our interest in wide InGaN QWs was a peculiar dependence of photoluminescence (PL) intensity on QW thickness. Fig 1(a) presents the PL intensities for In0.17Ga0.83N single QW samples grown by plasma assisted molecular beam epitaxy (PAMBE) with the thickness varied from dQW=1.2 to 25 nm. The structure of the single QW is shown as an inset to Fig 1(a). Details on the PAMBE growth technique can be found elsewhere [46, 47]. The indium content in InGaN quantum barriers (QBs) was chosen to be 8% as the LDs presented in the latter part of this paper also have In0.08Ga0.92N QBs. The presence of indium in QBs slightly lowers the built-in electric field but does not change the qualitative trends. The In content in the QWs was corroborated by X-ray diffraction and is constant among the samples. The QWs were grown using high nitrogen flux RF plasma source which, as we demonstrated earlier, leads to high quality InGaN growth in PAMBE [48-50]. Fig 1(b) shows the transmission electron microscopy picture of the 15.6 nm QW indicating that both QW interfaces are sharp and the overall good structural quality. It is important as it shows that the samples of study can be treated as uniform and high quality QWs and not for example quantum dots. Depending on the excitation power used of the 325 nm He-Cd laser the measured intensity 3 Fig 1. (a) Dependence of photoluminescence intensity on QW thickness for two excitation power densities. The dashed lines are guides to the eye and present two contradictory trends. (b) TEM image of the 15.6 nm QW showing sharp bottom and top interfaces. was observed to decrease or increase with the increasing QW thickness. According to the QCSE as the QW thickness is increased the wavefunction overlap and thus intensity of PL should drop. This is observed if a relatively small 0.4 W/cm2 excitation power is used. However, when the excitation power is increased to 60 W/cm2 an opposite trend is seen which comes from screening of the built-in electric field by generated carriers and emerging of an efficient transition through excited states. To gain insight into the mechanisms responsible for the observed changes in the trends the SiLENSe 5.4 package has been used for calculation of the potential shape and wavefunction distribution at various excitations [51]. Although SiLENSe package requires an LED structure as input it can be used to a certain extent to mimic the behavior of QW samples under optical excitation. The comparison of the carrier wavefunction overlaps between thin and wide QWs has been shown in Fig 2(a). The ground transition <e1h1> of thin QWs has an initial overlap of 0.31 at j=0 A/cm2 which increases with current density due to screening of the built-in polarization fields. In the case of wide QW, as it had been shown earlier, the transition between ground states is significantly lower than the thin counterpart due to separation of electron and hole wavefunctions driven by QCSE. However, there is a significant transition path including the excited states e2 and h2. The wavefunction overlap of the <e2h2> transition, although initially nearly equal to 0, peaks to a 0.63 at j=300 A/cm2 which is higher than the overlap of the ground transition of the thin QW. The crucial factor in observing this high efficiency transition is that the excited states in the wide QW get populated with carriers. This will be covered later. 4 Figure 2. (a) Wavefunction overlaps within QWs of two thicknesses: a thin 2.6 nm and a wide 10.4 nm as a function of current density. Only three types of transitions <e1h1>, <e2h1> and <e2h2> are shown to clearly present the idea behind the high efficiency of wide QWs. (b) The full evolution of the wavefunction overlaps of four transitions as a function of well thickness. The region marked with a green rectangle shows the commonly used QW thicknesses. The emerging of a high efficiency transition <e2h2> is observed for large thickness of the QW. The full evolution of the wavefunction overlaps with the QW thickness is shown in Fig 2(b) for a fixed current density j=2kA/cm2. This value is chosen on purpose to show the QW band profile close to lasing thresfhold. The standard operating conditions for LEDs are much lower (j=10A/cm2) due to the efficiency droop. However, the efficiency droop of LEDs with wide QWs, as will be shown later, is lower. Therefore they can be operated with success at higher current densities. The green-colored region, ranging from 1.5 to 3.5 nm, marks the thicknesses of the QWs commonly used in LEDs and LDs. In the thin QW regime, even up to a thickness 4.1 nm, the overlap between the ground states <e1h1> decreases with the QW thickness leading to a higher carrier concentration for the same radiation rate. However, above 4.1 nm an excited state in the conduction band e2 appears with an initially high overlap with the valence band ground state h1. The high recombination probability of the <e2h1> transition is counterintuitive as it would violate the selection rules of a rectangular QW. However, the selection rules do not apply to the case of InGaN QWs with high built-in electric field which breaks the symmetry of the potential. As the thickness is increased the overlap between the excited states <e2h2> rises whereas other overlaps drop. The value of the <e2h2> overlap of wide QW becomes higher than the value of <e1h1> overlap of thin QW. At the same time the carrier density of the wide quantum well can be much lower as compared to the thin counterpart. This leads to a reduction of the non-radiative Auger recombination and a much more efficient radiative transition. 5 Figure 3. Band profiles and carrier wavefunctions of 17% InGaN QWs in the (a) no injection and (b) high injection regimes. The full screening of polarization field and high wavefunction overlap of excited states is observed only for the wide QW. The e2 state does not form in case of the 2.6 nm thick QW at high injection due to the insufficient barrier height. Three regimes can be distinguished from Fig 2(b): thin QWs up to 4.1 nm where only <e1h1> and <e2h1> transitions can be observed, intermediate QWs up to 10 nm with all four transitions, and wide QWs above 10 nm where the <e2h2> transition exceeds the others. To understand the interplay between these regimes the band profiles and carrier wavefunctions are presented in Figures 3(a) and 3(b) for current densities of j=0 A/cm2 and j=2×103 A/cm2, respectively. QWs of three thicknesses: 2.6 nm, 5.2 nm and 15.6 nm, have been presented to explain the behavior of the three regimes presented in Fig 2(b). First, the case in which no current flow is present will be discussed. As can be seen in Figure 3(a) a strong built-in field of ~1.7 MV/cm is present in the QWs which causes the separation of the electron and hole wavefunctions and the QCSE. The wavefunction overlaps between the electron and hole ground states are 0.19, 0.0006 and 10-24 for the QWs with thicknesses of 2.6, 5.2 and 15.6 nm, respectively. In the second case presented in Figure 3(b) in which j=2×103 A/cm2 the wavefunction overlaps between the ground states are equal to 0.52, 0.23 and 0.0007. Although 6 Figure 4. A schematic of the band profile and carrier distribution in the 15.6 nm wide QW at j=2 kA/cm2. The carriers occupying the e1 and h1 do not recombine due to almost zero overlap between their wavefunctions and build up their concentrations. This lead to screening of the piezoelectric sheet charges, marked with red circles. The red shaded area depicts the part of the QW in which there is still some residual electric field. The carriers on e2 and h2 states are involved in the radiative transition. The distribution along the whole QW thickness of carriers on e2 and h2 states ensures the low carrier density and reduced recombination through the Auger process. the overlaps are slightly higher due to the screening of the built-in fields still a dramatic decrease can be observed with increase of the QW thickness. This decrease is the cause of the common opinion that the thicker the InGaN QW the lower the efficiency of any device build based on it. On the other hand, the excited states show a remarkably high wavefunction overlaps. The nature of the built-in electric field is responsible for this discrepancy and will be discussed below. The built-in field is caused by appearance of fixed electric charges at the QW interfaces. For the 17% InGaN QW used in this example the polarization-induced sheet charges of 1×1012 cm-2 form at the QW interfaces as indicated in Fig 4. To screen the polarization charges a matching number of electrons and holes has to be introduced. However, the electron and hole wavefunctions have a nonzero spatial dimension. Therefore, even if a matching number of electrons and holes is introduced, the remains of the polarization field are present locally. This causes the notable difference between the thin (2.6 nm) and thick (15.6 nm) QWs. The distance at which the influence of fixed charges is present in case of the thin QW is comparable to its thickness. The screening of the fixed electric charges changes the values of the wavefunction overlaps but the qualitative "triangular-like" appearance of the band profile of QW is similar (compare Fig 3 (a) and Fig 3(b)). On the other hand, in case of the wide QW the traces of the screened built-in field are present only in the bending of the conduction and valence bands at the QW interfaces (see shaded area in Fig 4). In the middle of the wide QW no field is present and, in this region, its qualitative appearance is as of a square QW. Therefore, the ground 7 transition <e1h1> is weak but the transitions between excited states <e2h2> is strong almost as in the case of a rectangular QW without piezoelectric fields. The carrier density required to match and fully screen the polarization charges is 5x1019 cm-3. To benefit from the high-overlap excited states transition <e2h2> can be achieved only if the polarization field is screened. The required carrier density seems to be extremely high as it is comparable to the one observed in III-nitride LDs at threshold [52]. However, considering the low transition probability before the screening of piezoelectric field this high carrier density can be achieved easily as long as no other carrier loss mechanisms such as recombination via defects or thermal escape play a significant role. The initial carrier accumulation behavior of a thick QW can be thought of as a capacitor. Afterwards, a strong <e2h2> transition emerges and the build-up of carriers is suppressed. Below the mechanism of filling of the excited states with carriers will be discussed. At no carrier injection, the case shown in Fig 3(a), the energetic distances between the ground and excited states are slightly increasing with QW thickness and are equal for the 2.6 nm thick QW to 200 and 90 meV for electrons and holes, respectively. However, after the screening of the polarization field the difference in the energy level between the ground and excited states decrease with the increase of QW thickness as expected for a square-like potential shape (compare Fig 3(b)). In the case of the thin 2.6 nm QW the e2 excited state does not even form because of its energy level being higher than the quantum barrier while the difference in the energy levels of h1 and h2 is still 90 meV whereas in the case of 15.6 nm thick QW the energy level differences are equal to 39 and 13 meV for electrons and holes, respectively. Taking into account the high carrier density, required to screen the piezoelectric field, and the density of states of e1 and h1 it can be clearly shown that the carriers will start to occupy the e2 and h2 states. Assuming a bulk density of states the quasi-Fermi levels of the conduction and valence band for a carrier density of 5x1019 cm-3 would be 230 and 20 meV, respectively. This is far above the energy level difference between the ground and excited states, therefore the excited states will be occupied as soon as the carrier density reaches value required to screen the polarization charges and screen the built-in electric field. This is feasible as the carriers occupying the ground states do not recombine because of the low transition probability. Eventually, carriers will start to occupy the excited states. Photoluminescence experiments To verify the predictions of the model presented above a set of 17% InGaN QWs with thicknesses ranging from 1.2 to 25 nm had been prepared by PAMBE. In this section only three 8 Figure 5. Power dependent photoluminescence for QWs with three thicknesses: (a) 2.6 nm, (b) 5.2 nm and (c) 15.6 nm. In case of the thin 2.6 nm QW only the <e1h1> transition is observed, whereas in case of the intermediate 5.2 nm QW a second transition interpreted as the <e2h1> appears and is depicted with a green dashed line. The thick QW starts to emit light at a relatively high excitation through the <e2h2> transition. QWs with thicknesses representing the three regimes observed in Fig 2(b) are shown. However, the behavior shown below is persistent in all of the studied samples. The PL data of all of the studied samples is provided in the Supplementary Figure 1. The power dependent PL of the three samples are shown in Fig 5(a), 5(b) and 5(c). The excitation power regime is extremely low ranging from 40 µW/cm2 up to 6 W/cm2 and has been chosen to show the behavior at the lowest carrier densities at which the radiative transitions were detectable. The thin 2.6 nm QW, shown in Fig 5(a), starts to emit light at λ=450nm and slightly shifts to λ=447nm due to screening of the polarization field and the band-filling effect. At all excitation powers only one transition <e1h1> is observed as expected when analyzing Fig 2(b). The peak observed at 490 nm comes from the residues in the He-Cd laser used to excite the samples and can be observed at extremely low light intensities. The peaks at λ=400-405 nm observed also in all of the other samples come from the In0.08Ga0.92N layers. In the case of the intermediate 5.2 nm QW the emission starts at a wavelength of λ=495 nm. The large difference between the emission wavelength of 2.6 and 5.2 nm thick QWs is due to the QCSE which red-shifts the emission wavelength of the latter. However, with the increase of excitation power the emission blue- shifts due to screening of the built-in electric fields. Additionally, starting from the excitation power of 1 W/cm2, emerging of a second higher energy transition can be seen indicated with a green dashed line in Fig 5(b). Its intensity is growing extremely fast with power and at the maximum power used 6 W/cm2 is even higher than the ground transition. This is expected as the wavefunction overlap of the transitions including excited states is much higher than the <e1h1> transition. The energy difference between the two transitions is 100 meV. This value together with the fact that the <e2h1> has the highest wavefunction overlap suggest that it is the observed transitions. However, the calculated energy difference between h1 and h2 levels is small which does not allow to unambiguously rule out the <e2h2> transition. The 9 luminescence of the thickest, 15.6 nm, QW starts at λ=444 nm and shifts to λ=442 nm at higher excitation power. This peak is attributed to be the <e2h2> transition based on the fact that it is the only allowed transition as seen in Fig 2(b). The important issue is the power density at which the emission from the QW starts to be observed. The excitation density of 600 mW/cm2 is much higher than in the case of thinner QWs in which the <e1h1> transition could be observed. Our interpretation is that until the polarization field is screened the radiative recombination is suppressed due to nearly zero wavefunction overlap. Due to thermal escape and non-radiative recombination via defects a certain level of excitation is needed for the carrier density to build up and screen the polarization charges. Afterwards, the profile of the thick QW become square- like, as in Fig 3(b), and the highly efficient <e2h2> starts to dominate. The above experiment proves the predictions of the model of efficient recombination via excited states in wide InGaN quantum well. Influence of composition on wavefunction overlap -- towards efficient green emitters In this section the behavior of the transitions including the excited states for high indium content QWs will be discussed. It will be shown that the efficiency drop observed in the green spectral regime can be compensated by the use of wide InGaN QWs. This is contrary to the common understanding as the higher indium content generates a higher misfit between GaN substrate and InGaN QW and thus enhances the piezoelectric fields. The increased piezoelectric field causes a growing separation of electron and hole wavefunction and decreases their overlap. Fig 6(a) presents the calculated dependence of wavefunction overlap between the <e1h1> on Figure 6. The dependence of wavefunction overlap on the composition of the: (a) thin 2.6 nm and (b) wide 10.4 nm QW. The <e1h1> transition probability drops with indium content in both cases whereas the <e2h2> highly increases for the wide QW. Inset to Fig 5(a) presents the band profile and carrier wavefunctions of a 30% InGaN QW. The energetic distance between the h1 and h2 levels is large enough to prevent occupation of the h2 level. 10 the composition of a thin 2.6 nm QW. The commonly observed drop of efficiency is attributed mainly to the decrease of wavefunction overlap with indium content [32]. Naturally, one would expect to observe a similar behavior if the thickness of the QW is changed. However, as shown in Fig 6(b), the transitions including excited states behave differently. Although the probability of the transition between ground states <e1h1> decreases, the <e2h2> overlap increases with thickness. This is surprising and counterintuitive. The <e2h2> overlap of the wide 10.4 nm QW reaches a maximum value of 0.65 at 30% indium content. This value is strikingly high and together with the low carrier density characteristic for a wide QW makes it a good candidate for efficient green light sources. To compare the efficiency of the thin and wide QWs the following calculation has been 𝛤𝛤𝑒𝑒ℎ𝐵𝐵𝑛𝑛2 carried out. The IQE is given by: 𝛤𝛤𝑒𝑒ℎ𝐴𝐴𝑛𝑛+𝛤𝛤𝑒𝑒ℎ𝐵𝐵𝑛𝑛2+𝛤𝛤𝑒𝑒ℎ𝐶𝐶𝑛𝑛3 (1), 𝐼𝐼𝐼𝐼𝐼𝐼= where 𝛤𝛤𝑒𝑒ℎ is the overlap between electron and hole wavefunctions, 𝑛𝑛 is the carrier density and 𝐴𝐴, 𝐵𝐵 and 𝐶𝐶 are the Shockley-Read-Hall, radiative and Auger recombination 7(a). As can be seen from Eq. (1) the 𝛤𝛤𝑒𝑒ℎ term cancels out and thus the IQE is independent on coefficient, respectively. The IQEs for In0.17Ga0.83N (blue color regime) and In0.30Ga0.70N (green color regime) has been calculated for the ABC parameters reported in [53] and plotted in Fig the overlap nor on the thickness of the QW. To understand how the overlap and QW thickness influence the efficiency it is important to look at the relationship between carrier density and current densities: 𝑗𝑗=𝑞𝑞𝑑𝑑𝑄𝑄𝑄𝑄(𝛤𝛤𝑒𝑒ℎ𝐴𝐴𝑛𝑛+𝛤𝛤𝑒𝑒ℎ𝐵𝐵𝑛𝑛2+𝛤𝛤𝑒𝑒ℎ𝐶𝐶𝑛𝑛3) (2), where 𝑞𝑞 is the elementary charge and 𝑑𝑑𝑄𝑄𝑄𝑄 is the QW thickness. The higher the thickness and the higher the overlap the lower the actual carrier density is as can be derived from Eq (2). The reduction in carrier density cannot be realized by an increase of the number of QWs because the carrier recombination takes place only at the last QW due the high difference in electron and hole mobilities [33, 44]. Fig 7(b) presents the calculated dependence of carrier density on current density in 2.6 nm and 10.4 nm thick InGaN QWs with two In compositions. The calculated change of wavefunction overlap with current density has been taken into account. For each QW only the highest value wavefunction overlap has been taken into account meaning that for the 2.6 nm In0.17Ga0.83N the carrier density is calculated assuming only the <e1h1> transition while for the 10.4 nm In0.17Ga0.83N initially the <e2h2> overlap is taken and is changed to the <e2h1> transition above j= 1.6 kA/cm2 (please compare Fig 2(a)). The transitions for 2.6 and 10.4 nm 11 Figure 7. (a) Dependence of the internal quantum efficiency on carrier density for In0.17Ga0.83N and In0.30Ga0.70N QWs. It is important to stress that it is independent on the thickness of the QW. (b) Relation between carrier density and current density for 4 kinds of QWs calculated taking into account the change in active region volume and wavefunction overlap. (c) Resultant dependence of internal quantum efficiency on current density. Only one transition with the highest wavefunction overlap is taken into account. In case of 10.4 nm In0.17Ga0.83N QW the dominant transition changes from <e2h2> to <e2h1> which is indicated in (b) and (c). In0.30Ga0.70N are <e1h1> and <e2h2> in the whole current density range, respectively. This approximation is sufficient and does not change qualitatively the results presented in this paragraph. The difference in carrier densities at a given current density causes changes to the efficiency. The calculated dependence of IQE on current density is presented in Fig 7(c). The results show a well-known decrease of the IQE with the current density for all QWs (commonly referred to as "droop") and a decrease of the IQE with In content of the QW (commonly referred to as "the green gap"). Additionally, it can be seen that the wide QWs provide a much higher IQE in the high current regime for both cases of blue (In0.17Ga0.83N) and green (In0.30Ga0.70N) emitting LEDs. The reason is the decreased carrier density and thus the part of carriers lost to recombination through the nonradiative Auger process. The increase in IQE coming due to the use of 10.4 nm QW at j=1 kA/cm2 is 40% and 70% for the In0.17Ga0.83N and the In0.30Ga0.70N QW, respectively. It is surprising and counterintuitive that the increase in IQE is higher for the higher indium content QW. 12 Figure 8. (a) Optical gain of 4 LDs with different QWs. The MQW LDs consist of three 2.6 nm thick QWs separated by 8 nm thick In0.08Ga0.92N barriers. The SQW LDs have a single 10.4 nm wide QW. Solid and dashed lines are used to extract the differential gain. The arrows indicate the increment achieved due to the use of the wide QW. Optical gain of wide InGaN QWs To verify the theoretical predictions of high efficiency long wavelength emitters based on wide QWs a set of four LDs had been manufactured. The measurement of optical gain is used to show the change in the efficiency and that the wide QWs can be used in devices requiring high carrier concentration. The optical gain in InGaN LDs has been widely studied and it was shown that it suffers from the green gap problem just as LEDs do [54-57]. There are two major causes of the decrease of optical gain for long wavelength LDs: (i) droop of quantum efficiency and (ii) lower optical confinement factor (Γ). The decrease of Γ is due to lower refractive index contrast between the alloys as the operating wavelength increases [58]. It cannot be easily increased by increasing the QW number because of the, mentioned above, non- uniform carrier distribution among QWs. However, the increase of the QW thickness should allow for the increase in Γ as long as there exists the proposed efficient transition through excited states. The LD structures and calculated confinement factors are provided the Supplementary Figure 2 and Supplementary Table 1, respectively. Fig 8(a) presents the measured maxima of optical gain as a function of current density of the studied LDs. A linear fit is used to extract the differential gain. The change in differential gain with the emission wavelength of the LDs with the MQW design are substantial. The differential gain falls from the value of 10.4 to 2.2 cm/kA when the emission wavelength is changed from blue to cyan. This is in agreement with other reports on the long wavelength LDs [54-57]. One of the major in 13 reasons of the drop is the decrease of the <e1h1> wavefunction overlap, depicted in Figs 2(a) and 6(a), and consequently the decrease of the IQE shown in Fig 7(c). The differential gain of the LDs incorporating a wide 10.4 nm InGaN QWs are 12.7 and 6.5 cm/kA for the blue and cyan LDs, respectively. The increase in both cases is significant but is larger in the case of the higher indium content QW as predicted in the previous section. These results, showing a high optical gain, are of vast importance as they indicate that the real-life optoelectronic devices can benefit from the increased IQE of wide InGaN QWs. Conclusions It was shown that, counterintuitively and contrary to the common opinion, wide InGaN QWs have a higher efficiency than their thin counterparts. The increase in the efficiency comes from a transition through excited states with high wavefunction overlap and lower carrier density. It is shown that in a wide QW the carriers occupying the ground states do not recombine but rather screen the built-in electric field. Even after full screening of the field the wavefunction overlap is small enough to prevent recombination through this transition. Instead carriers start to occupy the excited states with a high wavefunction overlap between them. Additionally, the lower carrier density in a wide InGaN QW, compared to normal QWs at the same current densities, ensures a reduction in the nonradiative Auger recombination responsible for the efficiency droop. Furthermore, it was shown that the higher the indium content the higher the wavefunction overlap between the excited states becomes. This result opens a new way to solving the green gap problem in LED efficiency with wide InGaN QWs. Two sets of experiments have been presented to verify the theoretical predictions. In the first one power dependent photoluminescence was measured for QW structures with various thicknesses. A clear transition from the recombination through the ground states to the excited states was observed with increased thickness of QW proving the predictions of the model. In the second experiment the optical gain of LDs with wide QWs was measured. It was shown that an enhancement of the material gain was achieved for wide QWs emitting in both blue and cyan colors. A more significant increase in the efficiency was observed in the case of higher indium content QWs as predicted by the model. We hope these findings will initiate interest and give rise to further research on wide InGaN QWs. Methods Epitaxial growth 14 The epitaxial growth was carried out using plasma-assisted molecular beam epitaxy. The MQW samples and blue LDs were grown in a customized VG V90 MBE reactor while the cyan LDs were grown in a customized Veeco Gen20A MBE reactor. Both machines were equipped with two Veeco RF plasma sources. The MQW samples and green LDs were grown on bulk GaN substrates obtained by hydride vapor phase epitaxy while the blue LDs were grown on bulk Ammono-GaN substrates. The InGaN layers were grown at 650C in indium-rich conditions whereas GaN and AlGaN layers at 730C at gallium-rich conditions. The high quality InGaN growth was carried out by suppling a high active nitrogen flux of up to 2 µm/h in units of equivalent GaN growth rate. Details of InGaN growth mechanism by PAMBE can be found in Ref. [46, 47]. Photoluminescence measurement The PL was measured at room temperature (298 K) with a 325 nm He-Cd laser. The system consisted of a 0.55 m long monochromator coupled with a liquid nitrogen cooled Symphony Si CCD detector. A microscope objective was used to focus the laser light on the sample surface and collect the emitted light. The excitation spot was circular with a diameter of 70 µm. Calculations Commercially available SiLENSe 5.4 [51] package has been used to calculate the wavefunction overlaps. It calculates the potential shape in an LED at a given voltage by solving the Poisson equation. Afterwards, the Schrödinger equation is solved to calculate the wavefunction distribution and the wavefunction overlap. The default material properties provided with SiLENSE 5.4 had been used. It is important to stress that in our opinion any simulation tool should provide similar results regarding the dynamic screening of polarization charges in wide QWs. The IQE has been calculated taking into account the ABC model and the change in wavefunction overlap with current density calculated using SiLENSe. The A, B and C parameters have been taken from Ref [53]. Optical gain measurement The optical gain was measured using the Hakki-Paoli technique [59, 60]. The LDs were electrically driven below threshold and the high resolution amplified spontaneous emission (ASE) spectra were collected using a 1 m spectrometer equipped with a 3600 mm-1 grating and a CCD camera. An exemplary ASE spectrum of the cyan wide QW LD is presented in Supplementary Figure 3(a). Its magnification showing the adjacent peak in ASE spectrum is presented in Supplementary Figure 3(b). For comparison the lasing spectra collected above threshold is demonstrated in Supplementary Figure 3(c). The gain spectra were calculated based 15 on the ratio of valley to peak from the ASE spectra for each current density and are presented in Supplementary Figure 3(c). Next the maxima of each spectra were extracted and plotted as a function of current density in Figure 8. The internal and mirror losses were subtracted from the values of maximum gain in Figure 8 to clearly show the variation of the differential gain between the LDs. Acknowledgements The authors would like to thank J. Borysiuk for TEM imaging. This work has been partially supported by the Foundation for Polish Science grant TEAM TECH/2016-2/12, the National Centre for Research and Development grants PBS3/A3/23/2015 and LIDER/29/0185/L- 7/15/NCBR/2016. 16 References: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. S. Nakamura, G. Fasol, and S. Pearton, The blue laser diode: the complete story (Springer, Berlin, 2000). S. Nakamura, T. Mukai, and M. Senoh, "Candela‐class high‐brightness InGaN/AlGaN double‐heterostructure blue‐light‐emitting diodes," Appl. Phys. Lett. 64, 1687-1689 (1994). S. Nakamura, M. Senoh, S.-i. Nagahama, N. Iwasa, T. Yamada, T. Matsushita, H. Kiyoku, and Y. Sugimoto, "InGaN-Based Multi-Quantum-Well-Structure Laser Diodes," Jpn. J. Appl. Phys. 35, L74 (1996). O. Ambacher, "Growth and applications of Group III-nitrides," Journal of Physics D: Applied Physics 31, 2653 (1998). M. R. Krames, O. B. Shchekin, R. Mueller-Mach, G. O. Mueller, L. Zhou, G. Harbers, and M. G. Craford, "Status and Future of High-Power Light-Emitting Diodes for Solid-State Lighting," Journal of Display Technology 3, 160-175 (2007). M. H. Crawford, "LEDs for Solid-State Lighting: Performance Challenges and Recent Advances," IEEE Journal of Selected Topics in Quantum Electronics 15, 1028-1040 (2009). S. Pimputkar, J. S. Speck, S. P. DenBaars, and S. Nakamura, "Prospects for LED lighting," Nat Photon 3, 180 (2009). F. Bernardini, V. Fiorentini, and D. Vanderbilt, "Spontaneous polarization and piezoelectric constants of III-V nitrides," Physical Review B 56, R10024-R10027 (1997). T. Takeuchi, C. Wetzel, S. Yamaguchi, H. Sakai, H. Amano, I. Akasaki, Y. Kaneko, S. Nakagawa, Y. Yamaoka, and N. Yamada, "Determination of piezoelectric fields in strained GaInN quantum wells using the quantum-confined Stark effect," Appl. Phys. Lett. 73, 1691- 1693 (1998). R. Langer, J. Simon, V. Ortiz, N. T. Pelekanos, A. Barski, R. André, and M. Godlewski, "Giant electric fields in unstrained GaN single quantum wells," Appl. Phys. Lett. 74, 3827- 3829 (1999). O. Ambacher, J. Majewski, C. Miskys, A. Link, M. Hermann, M. Eickhoff, M. Stutzmann, F. Bernardini, V. Fiorentini, V. Tilak, B. Schaff, and L. F. Eastman, "Pyroelectric properties of Al(In)GaN/GaN hetero- and quantum well structures," Journal of Physics: Condensed Matter 14, 3399 (2002). "High electron mobility transistor based on a GaN‐AlxGa1−xN heterojunction," Appl. Phys. Lett. 63, 1214-1215 (1993). U. K. Mishra, P. Parikh, and W. Yi-Feng, "AlGaN/GaN HEMTs-an overview of device operation and applications," Proc. IEEE 90, 1022-1031 (2002). "Two-dimensional electron gases induced by spontaneous and piezoelectric polarization charges in N- and Ga-face AlGaN/GaN heterostructures," J. Appl. Phys. 85, 3222-3233 (1999). "Polarization effects, surface states, and the source of electrons in AlGaN/GaN heterostructure field effect transistors," Appl. Phys. Lett. 77, 250-252 (2000). "Two dimensional electron gases induced by spontaneous and piezoelectric polarization in undoped and doped AlGaN/GaN heterostructures," J. Appl. Phys. 87, 334-344 (2000). J. Simon, V. Protasenko, C. Lian, H. Xing, and D. Jena, "Polarization-Induced Hole Doping in Wide -- Band-Gap Uniaxial Semiconductor Heterostructures," Science 327, 60-64 (2010). V. Fiorentini, F. Bernardini, F. Della Sala, A. Di Carlo, and P. Lugli, "Effects of macroscopic polarization in III-V nitride multiple quantum wells," Physical Review B 60, 8849-8858 (1999). F. Della Sala, A. Di Carlo, P. Lugli, F. Bernardini, V. Fiorentini, R. Scholz, and J.-M. Jancu, "Free-carrier screening of polarization fields in wurtzite GaN/InGaN laser structures," Appl. Phys. Lett. 74, 2002-2004 (1999). S. F. Chichibu, A. C. Abare, M. S. Minsky, S. Keller, S. B. Fleischer, J. E. Bowers, E. Hu, U. K. Mishra, L. A. Coldren, S. P. DenBaars, and T. Sota, "Effective band gap inhomogeneity and piezoelectric field in InGaN/GaN multiquantum well structures," Appl. Phys. Lett. 73, 17 due to fields strength piezoelectric 2006-2008 (1998). N. Grandjean, B. Damilano, S. Dalmasso, M. Leroux, M. Laügt, and J. Massies, "Built-in electric-field effects in wurtzite AlGaN/GaN quantum wells," J. Appl. Phys. 86, 3714-3720 (1999). P. Lefebvre, A. Morel, M. Gallart, T. Taliercio, J. Allègre, B. Gil, H. Mathieu, B. Damilano, N. Grandjean, and J. Massies, "High internal electric field in a graded-width InGaN/GaN quantum well: Accurate determination by time-resolved photoluminescence spectroscopy," Appl. Phys. Lett. 78, 1252-1254 (2001). M. Leroux, N. Grandjean, M. Laügt, J. Massies, B. Gil, P. Lefebvre, and P. Bigenwald, "Quantum confined Stark effect due to built-in internal polarization fields in (Al,Ga)N/GaN quantum wells," Physical Review B 58, R13371-R13374 (1998). J. Seo Im, H. Kollmer, J. Off, A. Sohmer, F. Scholz, and A. Hangleiter, "Reduction of oscillator in ${\mathrm{G}\mathrm{a}\mathrm{N}/\mathrm{A}\mathrm{l}}_{x}{\mathrm{Ga}}_{1\e nsuremath{-}x}\mathrm{N}$ quantum wells," Physical Review B 57, R9435-R9438 (1998). T. Tetsuya, S. Shigetoshi, K. Maki, K. Miho, T. Hideo, A. Hiroshi, and A. Isamu, "Quantum- Confined Stark Effect due to Piezoelectric Fields in GaInN Strained Quantum Wells," Jpn. J. Appl. Phys. 36, L382 (1997). S. Chichibu, T. Azuhata, T. Sota, and S. Nakamura, "Spontaneous emission of localized excitons in InGaN single and multiquantum well structures," Appl. Phys. Lett. 69, 4188- 4190 (1996). M.-H. Kim, M. F. Schubert, Q. Dai, J. K. Kim, E. F. Schubert, J. Piprek, and Y. Park, "Origin of efficiency droop in GaN-based light-emitting diodes," Appl. Phys. Lett. 91, 183507 (2007). P. Joachim, "Efficiency droop in nitride-based light-emitting diodes," Phys. Status Solidi A 207, 2217-2225 (2010). E. Kioupakis, P. Rinke, K. T. Delaney, and C. G. Van de Walle, "Indirect Auger recombination as a cause of efficiency droop in nitride light-emitting diodes," Appl. Phys. Lett. 98, 161107 (2011). J. Iveland, L. Martinelli, J. Peretti, J. S. Speck, and C. Weisbuch, "Direct Measurement of Auger Electrons Emitted from a Semiconductor Light-Emitting Diode under Electrical Injection: Identification of the Dominant Mechanism for Efficiency Droop," Phys. Rev. Lett. 110, 177406 (2013). Y. C. Shen, G. O. Mueller, S. Watanabe, N. F. Gardner, A. Munkholm, and M. R. Krames, "Auger recombination in InGaN measured by photoluminescence," Appl. Phys. Lett. 91, 141101 (2007). M. Auf der Maur, A. Pecchia, G. Penazzi, W. Rodrigues, and A. Di Carlo, "Efficiency Drop in Green InGaN/GaN Light Emitting Diodes: The Role of Random Alloy Fluctuations," Phys. Rev. Lett. 116, 027401 (2016). A. David, M. J. Grundmann, J. F. Kaeding, N. F. Gardner, T. G. Mihopoulos, and M. R. Krames, "Carrier distribution in (0001)InGaN∕GaN multiple quantum well light-emitting diodes," Appl. Phys. Lett. 92(2008). N. G. Young, R. M. Farrell, S. Oh, M. Cantore, F. Wu, S. Nakamura, S. P. DenBaars, C. Weisbuch, and J. S. Speck, "Polarization field screening in thick (0001) InGaN/GaN single quantum well light-emitting diodes," Appl. Phys. Lett. 108(2016). H. Zhao, G. Liu, J. Zhang, J. D. Poplawsky, V. Dierolf, and N. Tansu, "Approaches for high internal quantum efficiency green InGaN light-emitting diodes with large overlap quantum wells," Optics Express 19, A991-A1007 (2011). R. A. Arif, Y.-K. Ee, and N. Tansu, "Polarization engineering via staggered InGaN quantum wells for radiative efficiency enhancement of light emitting diodes," Appl. Phys. Lett. 91, 091110 (2007). P. Waltereit, O. Brandt, A. Trampert, H. T. Grahn, J. Menniger, M. Ramsteiner, M. Reiche, and K. H. Ploog, "Nitride semiconductors free of electrostatic fields for efficient white light- emitting diodes," Nature 406, 865 (2000). A. E. Romanov, T. J. Baker, S. Nakamura, and J. S. Speck, "Strain-induced polarization in wurtzite III-nitride semipolar layers," J. Appl. Phys. 100, 023522 (2006). E. Yohei, Y. Yusuke, K. Takashi, A. Katsushi, U. Masaki, A. Masahiro, S. Takamichi, T. Shinji, I. Takatoshi, K. Koji, and N. Takao, "531 nm Green Lasing of InGaN Based Laser Diodes on Semi-Polar {20 {bar 2} 1} Free-Standing GaN Substrates," Appl. Phys. Express 18 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 2, 082101 (2009). C. S. Mathew, K. Kwang-Choong, M. F. Robert, F. F. Daniel, A. C. Daniel, S. Makoto, F. Kenji, S. S. James, P. D. Steven, and N. Shuji, "Demonstration of Nonpolar m -Plane InGaN/GaN Laser Diodes," Jpn. J. Appl. Phys. 46, L190 (2007). D. F. Feezell, J. S. Speck, S. P. DenBaars, and S. Nakamura, "Semipolar (20-2-1) InGaN/GaN Light-Emitting Diodes for High-Efficiency Solid-State Lighting," Journal of Display Technology 9, 190-198 (2013). T. Tetsuya, A. Hiroshi, and A. Isamu, "Theoretical Study of Orientation Dependence of Piezoelectric Effects in Wurtzite Strained GaInN/GaN Heterostructures and Quantum Wells," Jpn. J. Appl. Phys. 39, 413 (2000). Y. L. Li, Y. R. Huang, and Y. H. Lai, "Efficiency droop behaviors of InGaN∕GaN multiple- quantum-well light-emitting diodes with varying quantum well thickness," Appl. Phys. Lett. 91(2007). N. F. Gardner, G. O. Müller, Y. C. Shen, G. Chen, S. Watanabe, W. Götz, and M. R. Krames, "Blue-emitting InGaN -- GaN double-heterostructure light-emitting diodes reaching maximum quantum efficiency above 200A∕cm2," Appl. Phys. Lett. 91(2007). J.-I. Chyi, H. Fujioka, H. Morkoç, Y. Nanishi, U. T. Schwarz, J.-I. Shim, M. Kawaguchi, O. Imafuji, S. Nozaki, H. Hagino, S. Takigawa, T. Katayama, and T. Tanaka, "Optical-loss suppressed InGaN laser diodes using undoped thick waveguide structure," in Gallium Nitride Materials and Devices XI, (2016). C. Skierbiszewski, H. Turski, G. Muziol, M. Siekacz, M. Sawicka, G. Cywiński, Z. R. Wasilewski, and S. Porowski, "Nitride-based laser diodes grown by plasma-assisted molecular beam epitaxy," Journal of Physics D: Applied Physics 47(2014). H. Turski, M. Siekacz, Z. R. Wasilewski, M. Sawicka, S. Porowski, and C. Skierbiszewski, "Nonequivalent atomic step edges -- Role of gallium and nitrogen atoms in the growth of InGaN layers," J. Cryst. Growth 367, 115-121 (2013). M. Siekacz, M. Sawicka, H. Turski, G. Cywiński, A. Khachapuridze, P. Perlin, T. Suski, M. Boćkowski, J. Smalc-Koziorowska, M. Kryśko, R. Kudrawiec, M. Syperek, J. Misiewicz, Z. R. Wasilewski, S. Porowski, and C. Skierbiszewski, "Optically pumped 500 nm InGaN green lasers grown by Plasma-Assisted Molecular Beam Epitaxy," J. Appl. Phys. 110, 063110 (2011). C. Skierbiszewski, M. Siekacz, H. Turski, G. Muziol, M. Sawicka, P. Wolny, G. Cywiński, L. Marona, P. Perlin, P. Wiśniewski, M. Albrecht, Z. R. Wasilewski, and S. Porowski, "True- Blue Nitride Laser Diodes Grown by Plasma-Assisted Molecular Beam Epitaxy," Appl. Phys. Express 5(2012). H. Turski, G. Muziol, P. Wolny, S. Grzanka, G. Cywiński, M. Sawicka, P. Perlin, and C. Skierbiszewski, "Cyan laser diode grown by plasma-assisted molecular beam epitaxy," Appl. Phys. Lett. 104(2014). "SiLENSe package", soft.com/products/SiLENSe/. 2018, http://www.str- 5.4 retrieved 10th September, factor of GaN-based laser diodes from UV to green," Appl. Phys. Lett. 97, 021102 (2010). T. Lermer, A. Gomez-Iglesias, M. Sabathil, J. Müller, S. Lutgen, U. Strauss, B. Pasenow, J. Hader, J. V. Moloney, S. W. Koch, W. Scheibenzuber, and U. T. Schwarz, "Gain of blue and cyan InGaN laser diodes," Appl. Phys. Lett. 98(2011). M. Funato, Y. S. Kim, Y. Ochi, A. Kaneta, Y. Kawakami, T. Miyoshi, and S.-i. Nagahama, "Optical Gain Spectra of a (0001) InGaN Green Laser Diode," Appl. Phys. Express 6(2013). T. Lermer, M. Schillgalies, A. Breidenassel, D. Queren, C. Eichler, A. Avramescu, J. Müller, W. Scheibenzuber, U. Schwarz, S. Lutgen, and U. Strauss, "Waveguide design of green InGaN laser diodes," Phys. Status Solidi A 207, 1328-1331 (2010). 19 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 53. 54. 56. 57. 58. 52. W. G. Scheibenzuber, U. T. Schwarz, L. Sulmoni, J. Dorsaz, J. F. Carlin, and N. Grandjean, "Recombination coefficients of GaN-based laser diodes," J. Appl. Phys. 109(2011). D. Schiavon, M. Binder, M. Peter, B. Galler, P. Drechsel, and F. Scholz, "Wavelength- dependent determination of the recombination rate coefficients in single-quantum-well GaInN/GaN light emitting diodes," physica status solidi (b) 250, 283-290 (2013). J. Müller, U. Strauss, T. Lermer, G. Brüderl, C. Eichler, A. Avramescu, and S. Lutgen, "Investigation of long wavelength green InGaN lasers on c-plane GaN up to 529 nm continuous wave operation," Phys. Status Solidi A 208, 1590-1592 (2011). 55. W. G. Scheibenzuber, U. T. Schwarz, T. Lermer, S. Lutgen, and U. Strauss, "Antiguiding 59. 60. B. W. Hakki and T. L. Paoli, "cw degradation at 300°K of GaAs double‐heterostructure junction lasers. II. Electronic gain," J. Appl. Phys. 44, 4113-4119 (1973). B. W. Hakki and T. L. Paoli, "Gain spectra in GaAs double−heterostructure injection lasers," J. Appl. Phys. 46, 1299-1306 (1975). 20 Supplementary Figure 1. Power dependent photoluminescence for QWs with seven thicknesses: (a) 1.2 nm (b) 2.6 nm, (c) 5.2 nm, (d) 7.8 nm, (e) 10.4 nm, (f) 15.6 nm and (g) 26 nm. The dashed lines depict our interpretation of the observed optical transitions. 21 Supplementary Figure 2. The structure schematics of LDs used for the optical gain experiments. The LDs are identified by their active region in Figure 8 in manuscript: (a) 3x2.6 nm 17% In, (b) 10.4 nm 17% In, (c) 3x3 nm 24% In and (d) 10.4 nm 24% In. 22 Supplementary Table 1. The calculated optical confinement factor of LDs used for the optical gain experiments together with the measured differential gain. The change in the differential gain cannot be justified by the change in confinement factor. Measured differential gain (cm/kA) 10.4 12.7 2.2 6.5 LD sample 3x2.6nm In0.17GaN 10.4nm In0.17GaN 3x3nm In0.24GaN 10.4nm In0.24GaN Calculated optical confinement factor (%) 2.763 3.531 1.914 2.795 23 Supplementary Figure 3. (a) Amplified spontaneous emission (ASE) spectra of the LD with 10.4 nm 24% In quantum well collected below lasing threshold. (b) Magnification of the ASE showing the adjacent peaks used to calculate the optical gain. (c) Lasing spectra collected above lasing threshold. (d) Optical gain spectra collected for current densities ranging from 0.33 to 5.33 in steps of 0.33 kA/cm2. The green diamonds are maximum optical gain used to show the dependence of optical gain on current density in Figure 8. 24
1906.01077	1	1906	2019-05-07T12:50:41	Discussion about Different Methods for Introducing the Turbulent Boundary Layer Excitation in Vibroacoustic Models	[ "physics.app-ph", "math.NA", "physics.class-ph", "physics.comp-ph", "physics.flu-dyn" ]	For controlling the noise radiated from vibrating structures excited by turbulent boundary layer (TBL) it is relevant to develop numerical tools for understanding how the structure reacts to TBL excitation. Usually, the wall pressure fluctuations of the TBL are described through statistical quantities (i.e. space-frequency or wavenumber-frequency spectra) which depend on the TBL parameters. On the other hand, the vibro-acoustic models (i.e. Finite Elements, Boundary Elements, Transfer Matrix Methods, Analytical models, etc) evaluate deterministic transfer functions which characterise the response of the considered structures. The first part of this paper focuses on the coupling between the stochastic TBL and the deterministic vibro-acoustic models. Five techniques are presented. Numerical applications on an academic marine test case are proposed in order to discuss the calculation parameters and the interests / drawbacks of each technique. In the second part of the paper, the high frequency modelling with the Statistical Energy Analysis (SEA) method is considered. The focus is placed on the estimation of an important input of this method: the injected power by the TBL into the structure for each third octave band.	physics.app-ph	physics	DISCUSSION ABOUT DIFFERENT METHODS FOR INTRODUCING THE TURBULENT BOUNDARY LAYER EXCITATION IN VIBROACOUSTIC MODELS Laurent Maxit, Marion Berton, Christian Audoly, Daniel Juvé Abstract For controlling the noise radiated from vibrating structures excited by turbulent boundary layer (TBL) it is relevant to develop numerical tools for understanding how the structure reacts to TBL excitation. Usually, the wall pressure fluctuations of the TBL are described through statistical quantities (i.e. space-frequency or wavenumber-frequency spectra) which depend on the TBL parameters. On the other hand, the vibro-acoustic models (i.e. Finite Elements, Boundary Elements, Transfer Matrix Methods, Analytical models, etc) evaluate deterministic transfer functions which characterise the response of the considered structures. The first part of this paper focuses on the coupling between the stochastic TBL and the deterministic vibro-acoustic models. Five techniques are presented. Numerical applications on an academic marine test case are proposed in order to discuss the calculation parameters and the interests / drawbacks of each technique. In the second part of the paper, the high frequency modelling with the Statistical Energy Analysis (SEA) method is considered. The focus is placed on the estimation of an important input of this method: the injected power by the TBL into the structure for each third octave band. Laurent Maxit Laboratoire Vibrations- Acoustique, INSA de Lyon, Bât. St. Exupéry, 25 bis av. Jean Capelle, 69621 Villeurbanne Cedex ; e-mail : [email protected] Marion Berton Centre Acoustique, LMFA UMR 5509, Ecole Centrale de Lyon, 36 av. Guy de Collongue, 69134 Ecully Cedex ; e-mail : [email protected] Christian Audoly DCNS Research, Rond-point de l'Artillerie de la Marine, 83000 Toulon email: [email protected] Daniel Juvé Centre Acoustique, LMFA UMR 5509, Ecole Centrale de Lyon, 36 av. Guy de Collongue, 69134 Ecully Cedex ; e-mail : [email protected] Introduction 1. Structures excited by the turbulent boundary layer (TBL) are very common in practical applications. Car, airplanes, trains, and submarines may be excited by pressure fluctuations due to the turbulent flow induced by their motions. In order to reduce the noise radiated from these structures, it is important to understand at the design stage how the structure reacts to the TBL excitation. It is then necessary to develop numerical tools allowing predicting the vibration or the radiated pressure from the structure excited by the turbulent flow. Usually, the calculation process is decomposed in 3 steps: 1 - A stationary hydrodynamic model is used to estimation the TBL parameters over the surface of the structure from its geometry and the flow conditions; 2 - The spectrum of the wall pressure fluctuations is evaluated from the TBL parameters estimated in the previous step and by using one of the models proposed in the literature. Some of them are expressed in the space - frequency domain (like the famous Corcos model [1]) whereas as others are expressed in the wavenumber - frequency domain (like the no less famous Chase model [2]); Discussion about different models and comparison with experiment can be found in [3, 4] for the frequency auto spectrum and in [5, 6] for the normalized wavenumber cross spectrum; 3 - The last step consists in using a vibro-acoustic model to estimate the response of the structure to the pressure fluctuations. The choice of the model depends on the frequency range of interest: - For the low frequencies, deterministic models considering harmonic excitations are generally considered. For example, it can be a standard Finite Element Model (FEM) for a structural problem or FEM coupled with a Boundary Element Model (BEM) for an acoustic radiation problem. The coupling between the statistical model used to describe the wall pressure fluctuations and the deterministic vibroacoustic model constitute a difficulty in the calculation process described above (i.e. the transition from step 2 to step 3). This topic is specifically addressed in the first part of this paper. Five approaches will be proposed and discussed in Sec. 3 after having recalled the mathematical formulation of the problem in Sec. 2. - For high frequencies, the Statistical Energy Analysis (SEA) method [7] is generally used to represent the vibro-acoustic behavior of complex structures. As the excitation is characterized in SEA by its time-averaged injected power for each frequency band, it is necessary to evaluate this quantity when considering the TBL excitation. We propose and discuss in Sec. 4 a formula allowing estimating the injected power from the wall pressure spectrum expressed in the wavenumber-frequency space. A methodology is also proposed to take the spatial variations of the TBL parameters into account. 2. Vibrating structures excited by random pressure fluctuations 2.1 Presentation of the problem Sp Figure 1. Baffled simply supported plate excited by a homogeneous and stationary TBL. Let us consider a baffle panel of surface Sp excited by a TBL as shown in Fig. 1. Three assumptions are considered: - The TBL is assumed to be fully developed, stationary, and homogeneous over Sp; - The plate and the boundary layer are supposed weakly coupled. It is then supposed that the vibration of the plate does not modify the TBL wall pressure excitation. Spectra of the wall pressures over a rigid surface can then be considered; It is assumed that the propagation of the acoustic waves into the fluid is not affected by the turbulent flow. Moreover, for the marine applications (i.e. low Mach number), we could also neglect the convective effect on the acoustic wave propagation. - The marine test case considered for the numerical application is composed of a thin rectangular plate simply supported along its four edges and immersed in water on one-side. The flow direction is parallel to the longest edges of the plate (i.e. about x-axis). Numerical values of the physical parameters considered for this test case are given on Tab. 1. The parameters characterizing the turbulent boundary layer are supposed to be known: is the flow velocity, , the convection velocity, δ is the boundary layer thickness, and the wall shear stress. From these parameters and the wall pressure models proposed in the literature [2-5], we can define the spectrum of the wall pressure fluctuations acting on the plate. The auto spectrum density of the wall pressure is evaluated here considering Goody's model [8] whereas the normalised cross spectrum density is evaluated using the Corcos's model [2]. The later is considered for it simplicity because it provides an analytical expression of the cross spectrum both in the space-frequency domain and in the wavenumber-frequency domain , both. The spectrum of the wall pressure fluctuations is then given by - in the physical space (i.e. spatial separation): , and, (1) UcUwppS,pp,ppk,xy,,TBLppppppSS- in the wavenumber space : The frequency band of interest is fixed here to [10 Hz -- 1 kHz] and is above the hydrodynamic coincidence frequency. It results that the wavelength associated to the convection velocity is always smaller than the flexural wavelength of the plate. . (2) The objective of the present paper is to estimate the panel response induced by the wall pressure fluctuation defined by its spectrum. In the next section, we give the outlines of the formulation which is described in details in the literature [9-11]. Parameters Flow velocity Numerical value =7 m/s Convection velocity Boundary layer thickness Wall shear stress Corcos' parameters Panel thickness Panel length in the streamwise direction Panel length in the crosswise direction Panel Young's modulus Panel Poisson's ratio Panel mass density Panel damping loss factor Fluid sound speed Fluid mass density = 5 m/s = 9.1 cm = 2.52 Pa =0.11 ; =0.77 h=1 mm =0.455 m =0.375 m E=2.1 1011 Pa =0.3 = 7800 kg/m3 =0.01 =1500 m/s = 1000 kg/m3 Table 1. Physical parameters of the marine test case. 2.2 Mathematical formulation represents the wall-pressure fluctuations due to the TBL on the plate at point x as a function of time t. The plate velocity at point x due to wall-pressure fluctuations, can be expressed as the convolution product , (3) where is the velocity impulse response at point x for a normal unit force at point . The improper integral corresponds to the convolution product between the impulse response and it gives the plate exerted on an elementary surface and the force velocity at point x due to this force (for a time-invariant system). The surface integral over ,xykkk,,TBLppppppkSk2/ccUUcUwLb0c0,bpxt,vxt,,,,pvbSvxthxxtpxddx,,vhxxtx,,vhxxt,bpxdxdxcorresponds to the summation of the effect of the elementary forces over the plate surface and it gives (based on the principle of superposition for a linear system). As the turbulent flow produces random fluctuations, the plate response is characterised by the . Supposing that the process is stationary and auto-correlation function of the velocity, ergodic (i.e. expectation replaced by the limit of a time average), can be written as: . (4) The Auto Spectrum Density (ASD) of the velocity at point x is defined as the time Fourier transform of : . (5) The same definition is used for the ASD of the wall-pressure fluctuations, . Note that: - the Fourier transform of a function , is defined as whereas others conventions can be used (for example ). A special attention should be given on this point when the ASD of the wall-pressure fluctuations is extracted of the literature; (ζ,ω) is here a double-sided spectrum and is a function of the - Moreover, the angular frequency . The relation with a single-sided spectrum expressed as a function of only the positive frequency f is . Introducing (3) in (4), and the result in (5), we obtain after some manipulations of integrals: , (6) where is the Frequency Response Function (FRF) in velocity at point x for a normal force at point In the same manner, we can obtain the ASD of the radiated pressure at point z into the fluid . , (7) where is the Frequency Response Function (FRF) in pressure at point z for a normal force at point . pS,vxtvvRvvR/2/21,lim,,TvvTTRxtvxvxtdTvvR(,),, jtvvvvSxRxtedt(,)CLTppSfftjtfftedt12jtfftedtCLTppS(,)CLTppSf1(,)4(,)CLTCLTppppSSf(,),,,,,ppTBLvvvppvSSSxHxxSxxHxxdxdx,,,,jtvvHxxhxxtedtx(,),,,,,ppTBLppppppSSSzHzxSxxHzxdxdx,,pHzxxThese two equations are the starting point of the following techniques for coupling a wall pressure model with a deterministic vibroacoustic model. In the next section, five different techniques are presented to estimate the vibration response of the panel from (6). These techniques are also applicable to estimate the radiated pressure into the fluid from (7). 3. Different approaches to couple a stochastic wall pressure field to a deterministic vibroacoustic model 3.1 Preamble: calculation of Frequency Response Functions Different vibroacoustic models can be used to estimate the Frequency Response Functions (FRF) of complex panels radiated into a fluid: - FEM using Perfectly Matched Layers (PMLs) [12]; - FEM coupled with BEM [13]; - FEM coupled with Infinite Elements [14]; - Transfer Matrix Method (TMM) for infinite multi-layers panels [15]; - Etc… In these models, different types of harmonic excitations can be considered: - A normal point force as illustrated on Fig. 2 for estimating a point to point FRF; - A wall plane wave excitation; - A specified pressure field over the panel surface. - Etc… Figure 2. Illustration of the problem for evaluating the FRF between point x and : . Basically, for an angular frequency , the equations of motions of the vibroacoustic problem can be written in the matrix form: where the considered load case. is the dynamic stiffness matrix; , (8) , the response vector; and , the force vector of The response vector is obtained by inverting the dynamic stiffness matrix: . (9) The FRF can then been determined by extracting the appropriate information in the response vector. In order to simulate the effect of the TBL excitation, many FRFs should be calculated with the vibroacoustic model, and consequently, many load cases should be considered for the process described above (i.e. Eq. (8) and (9). The management of multi load cases is then an x,,/vvxHxxVFDx=FDxF1x=DFimportant issue when dealing with TBL excitation. For example, it is generally more efficient by a force matrix containing the different load cases (i.e. matrix -- matrix to multiply product) than to achieve a loop over the different load cases and to multiply by the force vector of each considered load case (i.e. loop + matrix -- vector product). Moreover, in some situations, for example when using of commercial software, it is not always possible to have this optimal management of the multi load cases. This is why in the following, we will not only indicate the computing time observed on the present test case, but we also indicate the number of considered load cases. For this present test case, the FRFs have been evaluated using an in-house code based on the PTF (Path Transfer Function) approach ([16-18]). It allows us to have an optimal management of the multi load case under the MATLAB environment. This substructuring method consists in decomposing our problem in two parts: the panel and the semi-infinite fluid. The coupling surface is divided into patches which sizes depend on the considered wavelengths. Each part is characterised separately by PTFs (i.e. Path mobilities for the panel using the modal expansion method, path impedances for the fluid using the Rayleigh integral). Writing the continuity conditions at the coupling interface allows us to assemble the two parts. The particularity of the present model compared to [17, 18] is that the fluid added mass effect is taken into account through the "wet modal frequencies" (which are estimated by assuming the fluid incompressible) instead of using the imaginary part of the acoustic impedance of the fluid domain. This permits to overcome the convergence issue evoked in [18] concerning the patch size criterion. Here, a patch size lower than half the flexural wavelength gives results with good numerical convergence. The numerical process based on the PTF approach has been validated for the test case considered by comparison with results published in the literature [19]. We do not describe more in details these calculations which are out of the scope of the present paper. 3.2 The spatial method Fi δxi x Figure 3. Illustration of the spatial discretization of the panel surface. The first method for coupling the wall pressure spectrum and the FRFs calculated with a vibroacoustic model is simply based on a regular spatial discretization of the panel surface as shown on Fig. 3. Eq. (6) becomes: , (10) where is the number of discrete points and is the elementary surface attributed at the discrete point i. 1D1D11(,),,,,,TBLvvvippijvjijijSxHxxSxxHxxxxixEq. (10) can be rewritten in the matrix form: with , , (11) (12) . The point-to-point FRFs, , should be evaluated using the . vibroacoustic model. The number of load cases corresponds to the number of discrete points, A key parameter of this method is the spatial resolution of the discretisation. Results for the test case are plotted in Fig. 4 with different resolutions given as a function on the convective wavelength λc (which depends on the frequency). The coarser mesh (i.e. δ=λc) gives poor results excepted at low frequency. It does not allow representing correctly the convective part of the pressure fluctuations. A spatial resolution corresponding to one third of the convective wavelength seems to be a good compromise between the results accuracy and the computing times. Even if the spanwise turbulence wall pressure correlation length of the Corcos model is lower than the convective wavelength (and the streamwise correlation length), a parametric survey shows us that the use of the same criterion for the spatial resolution in the spanwise direction than in the streamwise direction gives relevant results. We emphasize that the calculation process (based on the matrix form (11)) requires high memory capacity, in particular to store the wall pressure CSD matrix . This is why the calculations have not been performed above 500 Hz with our computer (although 1 kHz was initially expected). Figure 4. Velocity ASD at point x=(0.05, 0.18) for different spatial resolutions: blue, δ=λc; green, δ=λc /2; red, δ=λc /3; black, δ=λc /4; TBLppHSHtvvSTBLppS,TBLppijSxx1H,,viiHxxx,,, 1,viHxxiTBLppS3.3 The Choslesky method The second method is based on a Choslesky decomposition of the wall pressure CSD matrix [20, 21]: , (13) where is a lower-triangular matrix of dimensions and superscript T indicates the transpose of the matrix. Iin a first step, the method consists in achieving different realizations of the stochastic field characterized by is given by, . The wall pressure vector of the kth realization [21], , (14) where is a phase vector of random values uniformly distributed in . So, an ensemble average over a set of realizations of the pressure field approximates the wall pressure CSD matrix: , (15) where the bar over the complex value indicates the complex conjugate. In a second step, the vibroacoustic model is used to estimate , the panel velocity at point x when the panel is excited by the pressure field, . This calculation is achieved for a given number of realizations, K. The number of load cases considered in the vibroacoustic simulations corresponds then to the number of realization. Finally, the ASD of the velocity at point x is estimated by an ensemble average of the velocity responses, : We illustrate this approach on the present test case. The velocity responses . (16) of 20 realizations are plotted in grey in Fig. 5. A large dispersion of these responses can be observed. The ensemble average over these 20 realizations (Eq. (16)) is plotted with a black curve on Fig. 5 and compared with the result of the first method on Fig. 6. We can observe a good agreement between the two calculations even when only 20 realizations have been considered. With this approach, the number of load cases is then relatively small. TBLppSTBLppS,LLTTBLppijSxxLTBLppSkpkjkpLek0,2TBLppSkkEpp,kvxkp,kvx,,,kkvvSxEvxvx,kvxFigure 5. Velocity ASD at point x=(0.05, 0.18). Grey, Results of 20 realizations; Black, Average over the 20 realizations. Figure 6. Comparison of the results of method 1 (red) and method 2 (black). 3.4 The wavenumber method The third method is based on a formulation in the wavenumber space of Eq. (6). Let us consider the space Fourier transform of the wall pressure spectrum, . With our definition of the Fourier transform, it is related to the wall pressure spectrum in the physical space by Introducing Eq. (17) in Eq. (6) gives with . , . (17) (18) (19) is generally called the sensitivity function [22]. The interpretation of Eq. (19) indicates that this quantity corresponds to the velocity at point x when the panel is excited by an unit wall plane wave with wavevector k (i.e. by a wall pressure field ). We emphasize that theses wall plane waves can be generated by travelling acoustic plane waves only for wavenumbers k inside the acoustic domain (i.e. , the acoustic , with wavenumber). For wavenumbers inside the subsonic domain (i.e. ), the acoustic plane waves are evanescent and it is then more complex to generate them physically. The Source Scanning Technique proposed in [23] is one solution. From a numerical point of view, this problem disappears. The pressure field of this excitation can be directly applied as the panel loading. When using a numerical vibroacoustic model (like FEM, BEM, etc), it is however necessary to check that the spatial discretization of the model allows to represent the spatial variation of this pressure field. The third method proposed on this paper is based on a truncation and a regular discretization of the wavenumber space k. We suppose that the discrete space is composed of I points which are noted . The ASD of the velocity at point x can then be approximated by , (20) where is the elementary surface in the wavenumber domain attributed to the discrete wavenumber . The truncation and the discretisation of the wavenumber space should be done carefully in order to avoid the loss of information: - the wavenumber resolutions in the two directions should be defined such as to permit a correct representation of the spatial variations in the wavenumber space of the sensitivity function and the wall pressure spectrum. For the present test case, an analytical calculation of ,TBLppk,TBLppSxx21,,4jkxxTBLTBLppppSxxkedk221,,,,4TBLvvppvSxkHxkdk,,,,pjkxvvSHxkHxxedx,,vHxk, jkxpexS0kk0k0kk, 1,ikiI2211,,,,4ITBLvvppiviiiSxkHxkkikikthe sensitivity function for an invacuo panel gives us an order of magnitude of these spatial variations (inversely proportional to the panel lengths). The wavenumber resolutions are fixed to 1 rad/m, independently of the frequency. For a more complex panel, a trial and error process would be necessary to fix these parameters; - the cut-off wavenumbers in the two directions should be defined such as the main contributions of the integrant of (18) are well taken into account. This point is illustrated on Fig. 7 for the present test case at a given frequency. The highest values of the sensitivity functions are obtained for wavenumbers close to , the natural flexural wavenumber of an equivalent infinite plate taking the fluid added mass effect into account. On another hand, the wall pressure spectrum exhibits the highest values for wavenumbers close to the convective . In theory, the cut-off wavenumber should be defined in the streamwise wavenumber, direction by and in the crosswise direction by where and are margin coefficient. As the considered frequency is well above the hydrodynamic coincidence frequency, we have and . This last criterion can lead to huge computing costs (because the spatial discretisation of the vibroacoustic model should be able to describe the "small" wavelength ). However, it is well known [24] that in many cases, the structure plays a role of filtering of the excitation which is dominant. This is illustrated on Fig. 7c where the product of the sensitivity function with the wall pressure spectrum (i.e. integrant of Eq. (18)) has been plotted. It can be observed that the contribution of the convective domain is negligible. Then, for this case, the cut-off wavenumber in the streamwise direction can be reduced to . This permits to save huge computing times. It should be emphasized that this restriction is not always valid. In particular, it depends on the frequency (compared to the hydrodynamic frequency), on the considered wall pressure model (see [25]), and the boundary conditions of the panel (see [26]). Here again, a trial and error process could be necessary at certain frequencies to fix the cut-off wavenumber; wetfkckmax,wetxxfckkkwetyyfkkxywetfckkxxckk2/xkwetxxfkk kc (a) (b) (c) Figure 7. Different quantities in the wavenumber space : (a), the sensitivity function at point x, ; (b), the wall pressure spectrum, ; (c), the product between the sensitivity function and the wall pressure spectrum, . Results presented at 100 Hz. With this approach, the number of load cases corresponds to the number of wall plane waves considered for the calculation of the sensitivity functions, . The present method respecting the previous criteria for the wavenumber resolutions and the cut-off wavenumbers is compared with the spatial method in Fig. 8. We can observe that the spatial method gives results slightly higher than the wavenumber method (excepted for the first peaks). This can be attributed to the spatial resolutions of the first method (i.e. δ=λc /4) which is not sufficiently small to ensure a full convergence of the method. Contrary to the spatial and Cholesky methods, the wavenumber method allows us to obtain results up to 1 kHz. This is mainly due to the fact that the convective ridge which can be supposed negligible is not described with this method when using appropriate cut-off wavenumbers. ,xykkk,,vHxk,TBLppk2,,,TBLppvkHxk,,vHxkwetfkFigure 8. Velocity ASD at point x=(0.05, 0.18). Comparison of the results of method 1 (red) and method 3 (blue). 3.5 The reciprocity method This fourth approach has been proposed in [11] for predicting the noise radiated by stiffened structures excited by TBL. It is based on a reciprocity principle which gives a second interpretation of the sensitivity functions. Indeed, the Lyamshev reciprocity principle [27] for vibro-acoustic problems indicates that the ratio of the normal velocity of the plate at point x over the applied normal force at point is equal to the ratio of the normal velocity of the over the normal force applied at point x. With the present notation, we can plate at point write: . (21) This expression can be injected in the definition of the sensitivity function (i.e. Eq. (19)) that allows us writing . (22) One recall that represents the velocity response at point when the panel is excited at point . Then, can be interpreted as the spatial Fourier transform of the velocity response of the panel excited at point . Consequently, the power spectrum density of the velocity of the plate at point x excited by the TBL can be calculated with Eq. (20) on the basis of the response of the plate excited by a normal force at point x and expressed in the wavenumber space by a discrete spatial Fourier transform. That is to say that the plate response at a given point due to TBL can be estimated from the vibratory field of the plate excited by a point force at this same point. xx,,,,vvHxxHxx,,,,pjkxvvSHxkHxxedx,,vHxxxx,,vHxkxWe can emphasize that this technique remains available even if the point of observation is into the fluid domain (for example for dealing with transmission loss problem). In this case, the radiated pressure at a point z by the TBL-excited panel would be estimated from the velocity field of the panel excited by an acoustic monopole located at point z and having unit volume flow rate [11]. The main advantage of this approach is that the number of load case is very small in general because it corresponds to the number of receiving points for which the response to the TBL excitation should be estimated. As this excitation is spatially distributed, it is generally not necessary to consider a large number of receiving points, the stochastic vibratory field being relatively homogeneous. We compare the sensitivity functions obtained with these two interpretations on Fig. 9. Of course, the results are very similar. The wavenumber resolutions differ as the ones of the reciprocity method depends directly on the panel dimensions (as a consequence of the discrete spatial Fourier transform). The comparison of the wavenumber and reciprocity methods on Fig. 10 shows a good agreement. The slight differences can be attributed to the different wavenumber resolutions. We can emphasize that this method requires few load cases but it requires evaluating the spatial distributions of the vibratory field in order to perform the spatial Fourier transform. When the vibratory field is evaluated from a numerical model (like the PTF approach used in the present paper), this task can be relatively time consuming and can reduce the efficiency of this approach. At its origin, this approach has been developed for dealing with stiffened structures like plate or shell stiffened in 1 direction or in 2 orthogonal directions [11]. For these cases, it is possible to calculate analytically the sensitivity functions thanks to the reciprocity principle described in this section. The computing times are then very short. Figure 9. Sensitivity function at point x=(0.05, 0.18) in function of the wavevector . Results at 1 kHz. Two calculations: upper, with Eq. (19) (i.e. direct interpretation); lower, with Eq. (22) (i.e. using the reciprocity principle). ,xykkkFigure 10. Velocity ASD at point x=(0.05, 0.18). Comparison of the results of method 3 (blue) and method 4 (green). 3.6 Method based on the sampling of uncorrelated wall plane waves The last of the five methods presented in this paper has been presented recently in [28]. It has some similarities with the method 2 (Sec. 3.3). But, contrary to the latter, it does not require a Cholesky decomposition (which can be time consuming). Basically, it consists in rewriting Eq. (20) in the following form: with , . (23) (24) This expression can be interpreted as the panel response to a set of uncorrelated wall plane waves of stochastic amplitudes represents the ASD of the amplitude of the ith waves. These wall plane waves are uncorrelated because Eq. (23) corresponds to the case where (see [23] for details). . This interpretation is similar to the one generally considered for describing an acoustic diffuse field: a set of uncorrelated acoustic waves of equiprobable incident angles and equal intensities. In the present case with the TBL excitation, the waves are not limited to the acoustic domain and their amplitudes are not constant; they depend on the wall pressure fluctuations, from Eq. (24). 21,,,iiIvvAAviiSxSHxk2,4iiTBLppiiAAkkS, 1,iAiIiiAAS0, ijAASij,TBLppikFrom this interpretation, we can define the wall pressure field of the kth realization, by, , (25) where , are random phase values uniformly distributed in . As for the method 2, the panel velocity at point x when the panel is excited by the pressure is then estimated by using the vibroacoustic model. This process is repeated for field, a given number of realizations, K; and, finally, the ASD of the velocity at point x is estimated by an ensemble average of the velocity responses, (see Eq. (16)). In Fig. 11, the velocity responses of 20 realizations are plotted in grey and the ensemble average over these 20 realizations is plotted in black. We can observe that this figure is similar to Fig. 5 related to the method 2, excepted that the calculation is achieved up to 1 kHz. The advantages of the present method compared to the method 2 are: (a), it is not necessary to use a Cholesky decomposition to define the wall pressure field of each realization. Eq. (25) with Eq. (24) can be applied directly from the wall pressure spectrum expressed in the frequency-wavenumber space (with a Corcos or a Chase model for example); (b), the use of adapted cut-off wavenumbers permits to neglect easily the effect of the convective ridge and then to save computing time. The good agreement between the results of method 3 and 5 on Fig. 12 allows us validating the present approach. Figure 11. Velocity ASD at point x=(0.05, 0.18). Grey, Results of 20 realizations; Black, Average over the 20 realizations. ,kipx1,kiiiiIjjkxkiAAipxSee, 1,kiiI0,2kp,kvxFigure 12. Comparison of the results of method 3 (blue) and method 5 (black). 3.7 Synthesis Five methods for coupling a TBL wall pressure model with a deterministic vibro-acoustic model have been presented: - The first two methods are adapted for a wall pressure spectrum expressed in the physical space (like given by the Corcos model); the spatial resolution criterion which permits to describe correctly the convective ridge requires a very fine discretization of the panel surface and it can limit these methods to low frequencies and/or to small panels. We have noticed that if this criterion is not respected, these two methods overestimate significantly the vibratory field; - The three other methods are adapted for a wall pressure spectrum expressed in the wavenumber space. In some situations (e.g. for frequencies well above the hydrodynamic frequency), the effect of the convective ridge can be neglected which enables to reduce the cut-off wavenumber (with a criterion based on the panel characteristics and not on the TBL ones). It permits to save computing time. Thanks to that, a higher frequency range has been reached by comparison with the spatial methods. It should however be emphasized that if the wall pressure spectrum in the physical space was filtered by a low pass filter in order to suppress the convective peak corresponding to the small spatial separations, the spatial methods would certainly have a similar efficiency than the wavenumber methods. As an indication, we give on Tab. 2 the number of load cases and the CPU time per frequency observed on the present test case for the 5 methods. We should emphasize that these computing times do not represent strictly the efficiency of each method; they depend strongly on the calculation algorithm (that we try to optimize), the computing environment (MATLAB for us), the management of the input/output of the vibroacoustic code (use of PTF in-house code). Anyway, the method based on the realizations of uncorrelated wall plane waves gave us smaller computing time. The reciprocity method is the one which necessitates the lowest number of load case. It could be the most efficient if the management of the input/output with the vibroacoustic code would not be optimal. The presented results and the discussion focus on the vibratory response of the panel. Of course, all the methods described in this paper can be used to evaluate the radiated pressure from the panel excited by TBL (for dealing with transmission loss problem for example). Moreover, they can be applied to more complex structures than the rectangular thin plate considered for illustration. Method Spatial Cholesky Wavenumb. Reciprocity Uncorrelated Number of load cases CPU time / frequency (s) 27300 5.8 20 16.4 10000 2 .5 1 2.3 waves 20 1.9 Table 2. Synthesis of the number of load cases and the computing times. 4. High frequency modelling 4.1 Statistical Energy Analysis Statistical energy analysis (SEA) allows the vibro-acoustic behaviour of complex structures in high-frequency range to be predicted [7, 29]. The method is based on a fundamental relationship relating the power flow exchanged by two-coupled subsystems to their total subsystem energies by the coupling loss factor (CLF). Basically, SEA consists in decomposing the global subsystem in different subsystems as illustrated in Fig. 13 for a Sonar self noise issue on a ship. This substructuring should be done in order to fulfil several conditions [30-33]. In particular, each subsystem should exhibit several (many) modes in the frequency band of interest and the couplings between subsystems should be weak [31]. For the case presented in Fig. 13, the coupling between the dome and the Sonar cavity filled of water may be a problem because it does not respect strictly the weak coupling assumption [34]. In this case, SEA can be seen as a first approximate model which is valuable for practical studies [35]. In a second step, SEA consists in writing the power balance for stationary motion in each subsystem using the fundamental SEA relation to evaluate the power flows. It produces a linear equation system where the unknowns are the total energies of subsystems. Then, the difficulty in applying SEA is not due to solving complicated equations, but in evaluating the SEA input parameters such as the damping loss factors, the coupling loss factors [36] and the injected power [37-39]. In this section, we focus the discussion on the evaluation of the injected power when the SEA subsystem is excited by a TBL. For the illustration case of Fig. 13, it consists in estimating the injected power by the turbulent flow in three subsystems (i.e. the Sonar dome and the two parts of the hull) for each frequency band (typically, third octave bands). We suppose that the TBL parameters characterizing the turbulent flow have been obtained from a hydrodynamic code and an appropriate model allows us describing the spectrum of the wall pressure fluctuations. (b) (a) Figure 13. (a), Illustration of the self noise issue on the bow of an Anti-Submarine Warfare surface ship; (b), SEA model describing the energy sharing between subsystems. Before going into the details about the injected power calculation, it is necessary to make a break on two points: - First, it should be remembered that the injected power depends not only on the excitation, but also on the receiving structure. If the structure was infinitely rigid, the injected power would be null. In the case of SEA model for which the global system has been decomposed in weakly coupled subsystems, one can suppose that the injected power in a given subsystem can be evaluated by neglecting the couplings with the others subsystems. This assumption is valid if the weak coupling condition is well respected with the others subsystems; - The calculation of the injected power should be performed in the framework of SEA hypothesis. That is to say that the frequency is relatively high, the excited structure presents many modes, and the different SEA quantities (especially the injected power) are time- averaged for a considered frequency band. Under these hypotheses, it may be reasonable to evaluate the injected power in a complex structure from the one in an equivalent academic structure [7]. Generally, this latter is a rectangular thin plate. It could be surprising at the first sight to "replace" a complex structure like the Sonar dome by a thin plate. However, it is well- know that in the high frequencies, the effects of curvature of the dome and of the boundary conditions on its vibratory behaviour can be neglected. If the dome is made of an isotropic material and of constant thickness, a thin plate can then be reasonably considered to evaluate SEA parameters like the modal density or the injected power. For more complex structures like the stiffness hull of the ship, the thin plate alone is probably not sufficient to represent correctly the behaviour at high frequencies. In particular, the propagation of the Bloch- Floquet waves due to the periodic stiffeners would not be described. This aspect of approximation is part of the difficulties in applying SEA to manufactured structures and is also part of the expertise of the SEA specialists. Anyway, in this paper, we have decided to focus our attention on the estimation of the power injected by the TBL in an equivalent thin plate. 4.2 Estimation of the time-average injected power Let us consider a thin plate subjected to a TBL excitation. The plate is made of an isotropic elastic material and has a constant thickness. M, D, h, are, respectively, the mass per unit and of central angular frequency area, the flexural rigidity, the thickness, and the damping loss factor of the plate. The TBL is fully developed, stationary and homogeneous. We consider a frequency band of angular bandwidth which is well above the hydrodynamic coincidence angular frequency. The energy balance equation consists in writing that the injected power by the TBL is dissipated by the plate. The time average of the injected power in the considered frequency band, can then be evaluated from , (26) is the time and space average of the quadratic velocity of the plate. where In the high frequencies, the shape of the plate and the type of boundary conditions do not influence the SEA parameters [7]. Then, a rectangular simply-supported plate was considered in [24, 37] for evaluating from a modal expansion. An alternative consists in considering an infinite plate (which is excited by a 'fictive' homogeneous CLT) and in evaluating of the plate. This "equivalence" of vibratory behaviour for a given area, in the high frequencies between a finite structure and an infinite one is often used in SEA. For its simplicity in the mathematical developments, we adopt it in the present paper. As the infinite plate is theoretically excited by a homogeneous CLT, the vibratory field is assumed to be spatially homogeneous. The ASD of the velocity at a given point x is independent of the point position: The time and space average of the quadratic velocity of the plate is obtained from . . The wavenumber formulation of Sec. 3.4 allows us writing the ASD of the velocity , (27) (28) (29) where are wavenumbers in the streamwise and spanwise directions, respectively. The sensitivity functions, can be calculated using the reciprocity principle described in Sec. 3.5. It corresponds to the transversal velocities of the plate expressed in the wavenumber space when the plate is excited by a normal point force at an arbitrary point. (We chose the coordinate origin for convenience.) Considering the Kirchhoff-Love's dynamic plate equation, we obtain: sinj2injsMV2V2V2VpS,, vvvvpSxSxS/22/212vvVSd221,,,,4TBLvvppxyvxyxySkkHkkdkdk,xykk,,vxyHkk, (30) with , the natural flexural wavenumber of the plate. We can notice that these sensitivity functions have the most important magnitudes for wavenumbers close to the flexural wavenumber (i.e. when ) and their magnitudes decrease quickly when the wavenumbers deviate from these values (see example Fig. 14a). On the contrary, the wall pressure spectrum varies relatively slowly in the subconvective wavenumber domain (see Fig. 14b). Then, the more significant contributions of the integrand of (29) correspond to the wavenumbers close to the flexural wavenumber (taking account that the convective ridge can be neglected seeing that the frequency band of interest is well above the hydrodynamic coincidence frequency). Supposing moreover that the wall pressure spectrum is relatively flat for wavenumbers close to the flexural wavenumber, we can write: , (31) This approximation is illustrated in Fig. 14c by comparing (full line) with (dashed line); as discussed in Sec. 3.4, it should be emphasized that this restriction is not always valid. The validity of this approximation depends on the frequency (compared to the hydrodynamic frequency), on the considered wall pressure model, and the boundary conditions of the panel. In the 'high' frequency, it is generally well respected. The integral of this expression can be approximated by . (32) Introducing Eqs. (28, 31, 32) in Eq. (26) and supposing that obtain an estimation of the injected power by the TBL in the plate: (and then ), we . (33) An expression which differs only by a factor was obtained in [37] considering a simply supported plate and a modal calculation. This factor is only due to the difference of definition of the space-time Fourier transforms between the two papers. We notice that this power is independent from the plate damping. This may have consequences for vibration and noise control. As it can be expected, it is also proportional to the area excited by the turbulent flow. 2224,,1vxysxyfjHkkDjkkk1/41/2fMkD22xyfkkk22,0,,,4TBLppfvvvxyxykSHkkdkdk2,,,,TBLppxyvxykkHkk2,0,,,TBLppfvxykHkk2226,,8vxyxysfHkkdkdkDk,0,4TBLinjppfSkMD32kf (a) (b) (c) Figure 14. (a), Sensitivity function for an infinite steel 1mm-thick plate; (b), Corcos wall pressure spectrum; (c), Product between the sensitivity function and the wall pressure spectrum: full, without approximation; dotted, with the approximation used in Eq. (31). Results at 200 Hz for ky=0 rad/m. Different approximations have been made to obtain this formula. In particular, the frequencies should be well higher than the hydrodynamic coincidence frequency and the wall pressure spectrum should be considered relatively flat in the wavenumber region concerned by the plate characteristic wavenumbers (i.e. flexural wavenumber). Comparison in [37] with an "exact calculation" for a present test case and considering the Corcos model showed that the discrepancies were very small in the frequency domain for which the SEA can be applied. For aeronautical application, the calculation of the injected power proposed in Ref. [38-39] can be more accurate for frequencies lower than or close to the aerodynamic coincidence frequency. Expression (33) has been obtained considering a flat plate and it may give a fair approximation of the injected power in subsystems composed of sheets having a high radius of curvature, roughly constant thickness and made of isotropic material. For stiffened structures like the ship hull, it could be seen as a first approximation. A more accurate prediction could be obtained by considering the sensitivity functions of a periodically stiffened plate [11]. In this way, it would be difficult to obtain an analytical expression of the injected power but a numerical process could be developed. SEA results could be compared to the approach proposed in [40-42] to estimating broadband levels of acoustic power radiated due to rib/panel interaction under TBL-like excitation. 4.3 A methodology for taking into account the spatial variation of the TBL parameters Hydrodynamic codes [37] permit to estimate spatial variations of the TBL parameters due to static pressure gradients or development of the TBL. An illustration is given on Fig. 15 for the bow of an Anti-Submarine Warfare surface ship. The TBL parameters can then vary on the surface of a given SEA subsystem (for example the Sonar dome of the ship). This can be an issue for evaluating the SEA injected power. However, if these variations are relatively slow compared to the wavelengths of the flexural motions, a numerical process taking these variations into account can be proposed. Indeed, we have noticed previously that the relation (33) has been obtained independently of the boundary conditions of the panel and it remains valid as long as many wavelengths are contained along each edge of the panel. Then, it can be use to evaluate the injected power in a part of a subsystem for which TBL parameters does not vary significantly. The process consists in dividing the subsystem surface (excited by TBL) in K patches having , we can evaluate the roughly constant TBL parameters. For each patch of surface injected power per unit area, : , (34) where is the wall pressure spectrum depending on the TBL parameters on the kth patch. An approximation of the injected power can then be obtained from . (35) An illustration of this process is given on Fig. 16. The injected power in the Sonar dome is evaluated by integrating over the dome surface the injected power by unit area. The latter has been calculated from the parameters of Fig. 15 and it exhibits significant spatial variations. This highlights the importance to taken the TBL parameters into account. kkSkinj,1,0,, 1,4kTBLinjppkfkkKMD,TBLppk1KkkinjinjkSFigure 15. Illustration of hydrodynamic calculation of the TBL parameters for the front of a ship [37]: (a), Boundary layer thickness; (b), Hydrodynamic friction velocity. Dome area Figure 16. Values of the injected power by unit area obtained from the TBL parameters of Fig. 15 [37]. 5. Conclusions This paper is focused on the coupling between TBL excitations and vibro-acoustic models. In the low frequency range, different techniques have been presented to make the relationship between the stochastic excitation and the deterministic model. The efficiency of these techniques in terms of computation time depends on various parameters such that the choice of the wall pressure models (in spatial or wavenumber form), the values of the physical parameters of the considered case, the efficiency of the vibro-acoustic code (in particular, its ability to manage multi-load cases), etc. For the marine test considered in this paper, the method consisting in the realizations of uncorrelated wall plane waves was found to be the fastest one. This method is easy to implement and it requires a small number of vibro-acoustic calculations (i.e. the number of load cases is equal to the number of realization). These methods offer a large possibility for coupling the wall pressure spectrum of the CLT excitation with the transfer functions describing the vibro-acoustic behaviour of the considered structure. In the future a more detailed study of the influence of the spatial variations of the TBL parameters should be undertaken. 1KkkinjinjkSIn the high frequency range, a formulation of the injected power in a SEA subsystem subjected to a TBL excitation has been proposed as a function of the wall pressure spectrum expressed in the frequency-wavenumber space. It has been obtained considering an infinite flat plate and several assumptions which are generally valid for high frequencies. Investigations should be performed in the future to extend these developments to more complex cases such as the stiffened structures frequently met in industrial applications. 6. References [1] G. M. Corcos, « Resolution of pressure in Turbulence », J Acoust Soc Am 35 (2), 192- 199 (1963). [2] Chase D.M., The character of the turbulent wall pressure spectrum at subconvective wavenumbers and a suggested comprehensive model. J Sound Vib 112 (1), 125-147 (1987). [3] Y. Hwang, W. Bonness, S. Hambric, Comparison of semi-empirical models for tuburlent boundary layer wall pressure spectra, J Sound Vib 319 (1) 199-217 (2009). [4] Y. Rozenberg, G. Robert, et S. Moreau, Wall-Pressure Spectral Model Including the Adverse Pressure Gradient Effects, AIAA Journal 50 (10), 2168‑2179 (2012). [5] W. Bonness, D. Capone, S. Hambric, Low-wavenumber turbulent boundary layer wall-pressure measurements from vibration data on a cylinder in pipe flow, J Sound Vib 329 (20) 4166-4180 (2010). [6] W. R. Graham, A comparison of models for the wavenumber-frequency spectrum of turbulent boundary layer pressures, J Sound Vib 206 (4) 541-565 (1997). [7] R.H. Lyon, R.G. Dejong, Theory and application of statistical energy analysis, Second ed. -- Boston: Butterworth-Heinemann, 277p. (1995). [8] M. Goody, Empirical spectral model of surface pressure fluctuations, AIAA journal 42, 1788‑1794 (2004). [9] W. A. Strawderman, Wavevector-Frequency Analysis with Applications to Acoustics, Naval Underwater Systems Center, Technical Report, NUSC 8209 (1988). [10] C. Maury, P. Gardonio, et S. J. Elliott, A wavenumber approach to modelling the response of a randomly excited panel, Part I : General Theory, J. Sound Vib. 252, 83‑113 (2002). [11] L. Maxit, V. Denis, Prediction of flow induced sound and vibration of periodically stiffened plates, J Acoust Soc Am 133 (1) 146-160 (2013). [12] I. Harari, M. Slavutin, E. Turkel, Analytical and Numerical Studies of a Finite Element PML for the Helmholtz Equation, Journal of Computational Acoustics 8 121- 137 (2000). [13] P. Bettess, Infinite elements. Int. J. Num. Meth. Engng 11, 53-64 (1977). [14] D. W. Herrin, T. W. Wu, A. F. Seybert, Boundary element modelling, Handbook of noise and vibration control, Paper 8, Malcolm J. Crocker, ed., John Wiley and Sons, Inc., New York, pp. 116-127 (2007) [15] M. Abid, M. S. Abbes, J. D. Chazot, L. Hammemi, L. Hammemi, M. Haddar, Acoustic Response of a Multilayer Panel with Viscoelastic Material international Journal of Acoustics and Vibration 17, 82-89 (2012). [16] M. Ouisse, L. Maxit, C. Cacciolati, J.L. Guyader - Patch Transfer Functions as a tool to couple linear acoustic problems. Journal of Vibration and Acoustic (ASME) 127, 458-466 (2005) [17] M. Aucejo, L. Maxit, N. Totaro, J.L. Guyader, Convergence acceleration using residual shapes technique when solving structure-acoustic coupling with Patch Transfer Functions method, Computers&Structures 88, 728-736 (2010). [18] L. Maxit, M. Aucejo, J.L. Guyader, Improving the Patch Transfer Function approach for fluid-structure modelling in heavy fluid. Journal of Vibration and Acoustics, ASME 134, 1-14 (2012). [19] A. Berry, A new formulation for the vibrations and sound radiation of fluid-loaded plates with elastic boundary conditions, J Acoust Soc Am 96 (2), 889‑901 (1994). [20] « ACTRAN 12.2 User's Guide. Vol. 1: Installation, operations, theory and utilities. Chap. 20: Random excitation. » Free Field Technologies SA, Belgium (2012). [21] L.E. Wittig, A.K. Sinha, Simulation of multicorrelated random processes using the FFT algorithm, J Acoust Soc Am 58, 630-634 (1975). [22] F. Birgersson, N.S. Ferguson, S. Finnveden, "Application of the spectral finite element method to turbulent boundary layer induced vibration of plates", J Sound Vib 259 (4), 873 -- 891 (2003). [23] M. Aucejo, L. Maxit, J.-L. Guyader, Experimental simulation of turbulent boundary layer induced vibrations by using a synthetic array, J Sound Vib 331 (16), 3824 -- 3843 (2012). [24] W.K. Blake, "Mechanics of flow-induced sound and vibration. Vol. I: General concept and elementary sources. Vol. II: Complex flow-structure interaction. Orlando, Florida, Academic press, Inc. (1986). [25] A.O. Borisyuk, V.T. Grinchenko, Vibration and noise generation by elastic elements excited by a turbulent flow. J Sound Vib 204 (2), 213-237 (1996). [26] S. Hambric, Y. Hwang, W. Bonness, Vibrations of plates with clamped and free edges excited by low-speed turbulent boundary layer flow, Journal of fluids and structures, 19, 93-110 (2004). [27] F. Fahy, Some applications of the reciprocity principle in experimental vibroacoustics, Acoust. Phys. 49, 262 -- 277 (2003). [28] M. Berton, L. Maxit, D. Juvé, C. Audoly, Prediction of flow-induced sound and vibration: on different methods for introducing the TBL excitation in the vibroacoustic model, Proceedings of Acoustics2013, New Delhi, India, November 2013. [29] F. Fahy, Statistical energy analysis: a critical overview, Phil. Trans. R. Soc. Lond. A 34, 431-447 (1994). [30] A. Le Bot, Derivation of statistical energy analysis from radiative exchanges Journal of Sound and Vibration 300 (3), 763-779 (2007). [31] A. Le Bot, V. Cotoni, Validity diagrams of statistical energy analysis, Journal of Sound and Vibration 329 (2), 221-235 (2010). [32] T. Lafont, N. Totaro, A. Le Bot, Review of statistical energy analysis hypotheses in vibroacoustics, Proc. R. Soc. A 470: 20130515 (2013). [33] A. Culla, A. Sestieri, Is it possible to treat confidentially SEA the wolf in sheep's clothing?, Mechanical systems and signal processing 20, 1372-1399 (2006). [34] L. Maxit, J.-L. Guyader, Estimation of SEA coupling loss factors using a dual formulation and FEM modal information. Part I: Theory, J Sound Vib 239 (5), 907- 930 (2001). [35] L. Maxit, Analysis of the modal energy distribution of an excited vibrating panel coupled with a heavy fluid cavity by a dual modal formulation, J Sound Vib 332 (25), 6703-6724 (2013). [36] C. Audoly, S. Beretti, Prediction of turbulent flow noise inside an acoustic sounder cavity. Proceedings of Euromech, Cargèse, France, April 1999. [37] L. Maxit, C. Audoly, Hydrodynamic noise prediction inside a Sonar dome: estimation of injected power from the wavenumber -- frequency spectrum of the turbulent wall pressure, Proceedings of NOVEM 2005, St Raphael, France, April 2005. [38] N. Totaro, J.L. Guyader, Model of frequency averaged injected power into a plate excited by a turbulent boundary layer, Acta Acustica united with Acustica 89 , 647-657 (2003). [39] N. Totaro, G. Robert, J.L. Guyader, Frequency averaged injected power under boundary layer excitation: an experimental validation, Acta Acustica united with Acustica 94, 534-547 (2008) . [40] M. L. Rumerman, Estimation of broadband acoustic power due to rib forces on a layer-like pressure excitation. I. reinforced panel under Derivations using strong model, J. Acoust. Soc. Am. 109, 563 -- 575 (2001). turbulent boundary [41] M. L. Rumerman, Estimation of broadband acoustic power due to rib forces on a layer-like pressure excitation. II. reinforced panel under Applicability and validation, J. Acoust. Soc. Am. 109, 576 -- 582 (2001) turbulent boundary [42] M.L. Rumerman, Estimation of broadband acoustic power radiated from a turbulent boundary layer-driven reinforced finite plate section due to rib and boundary forces. J. Acoust. Soc. Am. 111, 1274 -- 1284 (2001).
1802.04152	1	1802	2018-02-12T16:03:47	Mid Infrared Nonlinear Plasmonics using Germanium Nanoantennas on Silicon Substrates	[ "physics.app-ph", "cond-mat.mes-hall" ]	We demonstrate third harmonic generation in plasmonic antennas made of highly doped germanium and designed to be resonant in the mid infrared. Owing to the near-field enhancement, the result is an ultrafast, sub-diffraction, coherent light source tunable between 3 and 5 micrometer wavelength on a silicon substrate. To observe nonlinearity in this challenging spectral region, a high-power femtosecond laser system equipped with parametric frequency conversion in combination with an all-reflective confocal microscope setup is employed. We show spatially resolved maps of the linear scattering cross section and the nonlinear emission of single isolated antenna structures. A clear third order power dependence as well as the mid-infrared emission spectra prove the nonlinear nature of the light emission. Simulations support the observed resonance length of the double rod antenna and demonstrate that the field enhancement inside the antenna material is responsible for the nonlinear frequency mixing.	physics.app-ph	physics	Mid Infrared Nonlinear Plasmonics using Germanium Nanoantennas on Silicon Substrates Marco P. Fischer1, Aaron Riede1, Kevin Gallacher2, Jacopo Frigerio3, Giovanni Pellegrini4, Michele Ortolani5, Douglas J. Paul2, Giovanni Isella3, Alfred Leitenstorfer1, Paolo Biagioni4, and Daniele Brida1,* 1 Department of Physics and Center for Applied Photonics, University of Konstanz, D-78457 Konstanz, Germany 2 School of Engineering, University of Glasgow, Rankine Building, Oakfield Avenue, Glasgow, G12 8LT, UK 3 L-NESS, Dipartimento di Fisica del Politecnico di Milano, Via Anzani 42, 22100 Como, Italy 4 Dipartimento di Fisica, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milano, Italy 5 Department of Physics, Sapienza University of Rome, 00185 Rome, Italy * e-mail: [email protected] We demonstrate third harmonic generation in plasmonic antennas made of highly doped germanium and designed to be resonant in the mid infrared. Owing to the near-field enhancement, the result is an ultrafast, sub-diffraction, coherent light source tunable between 3 and 5 µm wavelength on a silicon substrate. To observe nonlinearity in this challenging spectral region, a high-power femtosecond laser system equipped with parametric frequency conversion in combination with an all-reflective confocal microscope setup is employed. We show spatially resolved maps of the linear scattering cross section and the nonlinear emission of single isolated antenna structures. A clear third order power dependence as well as the mid-infrared emission spectra prove the nonlinear nature of the light emission. Simulations support the observed resonance length of the double rod antenna and demonstrate that the field enhancement inside the antenna material is responsible for the nonlinear frequency mixing. Plasmonic nano-antennas1,2, i.e. resonant metallic structures with sub-optical-wavelength size, are one of the key components for advanced nano-optics applications. By directing and concentrating the far-field electromagnetic radiation into sub-diffraction-limited near-field volumes, they are the ideal tools to access single quantum systems with light. Another benefit of the light concentration capabilities of resonant antennas consists in the ability to access nonlinear optical phenomena3 even with minute amplitudes of the electromagnetic fields. This approach has been exploited successfully in the near-IR spectral range to generate second-harmonic4, third-harmonic5,6, or even higher- order phenomena7-9 as well as incoherent, octave-spanning multi-photon photoluminescence6,10 from gold nano- antennas. Alternatively, mid-IR plasmonic metamaterials have been employed to improve the coupling of far field radiation with the nonlinearity of quantum wells11. 1 In recent years, highly-doped semiconductors like InAs12, InAsSb13, InP14 and Ge15-17have been introduced as high- quality and tunable mid-IR plasmonic materials for integrated devices. The mid-IR frequency range is of high interest for chemical and biological identification of molecules for environmental, healthcare and security sensing applications, since it includes the so-called molecular fingerprint region16,18-20 with countless unique vibrational absorption lines. In this context, nonlinear plasmonics can be a sensitive and a versatile tool for achieving a near-field tunable source that can directly interact with molecules at the nanoscale. Up to now, however, intrinsic nonlinearities from plasmonic structures have been demonstrated only in the near-IR and employing standard metals that are not effective at longer wavelengths for this kind of application21. For several reasons Ge is the ideal material for linear and nonlinear plasmonic applications in the mid-IR spectral range since, in the last years, extremely heavily-doped materials with a tunable plasma frequency of up to 95 THz (3.1 µm wavelength) and high crystalline quality became available through epitaxy on silicon wafers17. The reflectivity and dielectric functions of the material employed in the reported experiments are depicted in Fig. 1a,b. In particular, full CMOS-compatibility of Ge on Si promises easy integration with the potential for cheap mass production. Moreover, the third order nonlinear coefficient22,23 is comparable to that displayed by gold at visible and near-IR frequencies24. Since Ge is a non-polar elementary semiconductor, the lack of optical phonon absorption reduces losses in the mid-IR compared to III-V semiconductors. Finally, the plasma frequency can also be controlled dynamically by optical excitation of electron-hole pairs25. Figure 1 Plasmonic Antennas from heavily doped germanium. a, Complex dielectric function of heavily doped germanium film extracted from b the reflectivity spectrum. c, Scanning electron micrograph of a single germanium double rod antenna on silicon substrate with an arm length of 3.5 µm and a gap width of 300 nm. In this work we target the local generation of third harmonic radiation in the mid-IR at sub-wavelength dimensions. Our experiments enable several exciting perspective applications from the nonlinear-enhanced sensing of molecular species26 to ultrafast near-field microscopy27,28 with unprecedented temporal and spatial imaging resolution for 2 molecular-selective imaging in life sciences as well as the direct porting of a large variety of near-infrared nonlinear plasmonic applications3 in the mid infrared. Furthermore the reported technique offers new ways to study the fundamental mechanisms of nonlinearity in condensed matter, especially in nano-structures24,29. Nevertheless, the study of nonlinear plasmonics in the mid-IR wavelength range poses major challenges with respect to the near-IR and visible cases. Diffraction, that limits the excitation field in the focus, combines with the increased antenna interaction volume leading to a complex, unfavourable wavelength scaling of the expected third harmonic generation (THG). In addition, the longer pulse durations at low frequencies further decrease the peak fields and thus the nonlinear emission. These difficulties have to be mitigated by scaling the laser system used for the optical excitation. Under these conditions, Ge displays the additional advantage of having a strong durability while in metals strong electro-migration and diffusion prevents high-field excitation for nonlinear mid-IR studies21. Figure 2 Linear and nonlinear confocal microscope images of a single antenna. a, The spatially resolved transmission map of a heavily doped Ge double rod antenna with an arm length of 4.5 µm at an excitation wavelength of 12 µm recorded by moving the antenna sample through the common objective focus. The darker regions around the antenna are diffraction artefacts of the reflective objectives complex point spread function. b, Cut of the transmission map through the center of the antenna (along the white dashed line). c, Spatially resolved third harmonic emission intensity from this antenna normalized to the substrate background emission. d, Cut through the emission map. A high-power femtosecond laser system is employed to obtain narrowband, intense, mid-IR transients via difference frequency generation tunable over the wavelength range from 7 to 20 µm. At a pulse energy of up to 100 nJ, the 3 mid-IR driving pulses feature a duration of 300 fs at a bandwidth of 1.5 THz. Single antenna structures (see Fig. 1c) are excited in a dispersion free Cassegrain-geometry reflective microscope setup with these multi-THz transients reaching peak electric fields of up to 5 MV/cm. By scanning the sample through the focal region in a transmission geometry, maps with high spatial resolution can be recorded. Figure 2a demonstrates the transmission map of a double rod antenna with an arm length of 4.5 µm. The excitation wavelength is set to about 12 µm corresponding to 25 THz. The extinction of this single resonant subwavelength structure exceeds 8% (Fig. 2c). Due to the sub- wavelength dimension of the antennas, the geometric width roughly represents the point spread function of the optical excitation. To discriminate between the fundamental and third-harmonic radiation, crystalline filters are employed. This allows the strongly localized third-harmonic emission from the plasmonic structure to be selected (Fig. 2b,d). The emission exceeds the substrate background by a factor of 2.5, which is substantial considering the small interaction volume of the driving beam with the subwavelength structure, in comparison to the bulk silicon wafer interaction path. Both the excitation and the emitted third harmonic generated radiation are polarized along the antenna axis. a, The excitation Figure 3 Third harmonic emission spectra and nonlinear power dependence. the corresponding third harmonic emission spectra from single resonant double rod antennas at excitation wavelengths of 10.7 µm, 12 µm and 14 µm (red, blue and green, respectively). c, The nonlinear power dependence of the THG emission in double- logarithmic scale for 12 µm excitation wavelength (circles). The line of cubic proportionality proves the third harmonic character of the radiation and shows the start of saturating behavior for high excitation powers beyond 1.5 mW. spectra and b 4 Figures 3a,b show the exact emission spectrum under three different excitation conditions. The power dependence of the emitted THG radiation from the antenna centre (Fig. 3c) with a cubic power exponent further proves that we indeed observe third-harmonic emission in this experiment. At high pump intensity levels, the curve demonstrates that the efficiency of the nonlinear emission starts to deviate from a third order power law. This could be caused by transient heating of the antenna that modifies the dielectric behaviour of Ge or by charge carrier excitations that slightly increases the reflection losses. Taking into account the losses of the optical elements employed for the collection of the third harmonic, we estimate a yield of the nonlinear frequency conversion (defined as the number of third-harmonic photons generated per excitation photon) that is higher than 10-6 for a single antenna emitter illuminated by 25 nJ driving pulses. This calculation does not include the limited collection aperture of the condenser objective with respect to the emission pattern of the THG. If a standard dipole emitter coupled to the antenna gap is considered to mimic the third-harmonic radiation, the total energy conversion ratio can be estimated to be approximately 10-5 with a THG energy in the order of 1 pJ per pulse. Finally, we investigate the dependence of both the linear scattering cross section and the third harmonic emission on the antenna arm length. Extinction and emission data are extracted from confocal microscopy raster scans similar to Figure 2 for a set of different antennas. The arm length is varied between 1.0 µm and 6.0 µm at a constant gap width of 300 nm. Figure 4a demonstrates the measurement results for an excitation wavelength of 12 µm. By normalizing over the antenna volume, a clear resonance for an arm length of 3.5 µm becomes visible. As expected from the nonlinear scaling of the excitation, this effect becomes more prominent in the THG emission. For antennas below 1.5 µm arm length, no third harmonic could be detected, whereas antenna arms longer than 3.5 µm become less efficient because of off-resonance conditions. The measurement is repeated in Figure 4c for an excitation wavelength of 14 µm displaying a shift of the resonance to longer antennas of approximately 5.0 µm arm length. Figures 4b,d display the results of three-dimensional finite-difference time-domain simulations that reproduce the observed behaviour with high fidelity, taking into account the numerical aperture of the two objectives and evaluating the THG signal by considering the third power of the field intensity integrated over the whole antenna volume. As previous studies revealed, the relatively large thickness of the antenna arms leads to two distinct resonances at the Ge-Si interface and at the Ge-air interface16. The latter of these resonances is excited in this experiment. 5 Figure 4 Measurement and simulation of extinction and THG emission for antennas of various arm lengths. a,c, Normalized extinction (red circles) and third harmonic emission (blue diamonds) per unit antenna volume measured for double rod antennas of various lengths at a gap width of 300 nm excited at 12 µm and 14 µm central wavelength, respectively. b,d, Corresponding FTDT simulation results for both excitation conditions. In conclusion, we were able to demonstrate for the first time the coherent nonlinear emission from plasmonic subwavelength-structures in the mid-IR frequency range. This achievement is enabled by the crucial advancements in the growth of epitaxial group-IV semiconductors and of heavily-doped germanium on silicon substrates in particular. Our results pave the way for new methods in mid-IR near-field microscopy employing germanium nanostructures30 and for the sensing of molecules based on their vibrational absorption fingerprints. This capability combined with the integration in a CMOS platform will allow targeting sensitive applications in biomedicine, international security for the detection of hazardous compounds and even environmental protection for example by 6 monitoring the emissions of combustion engines. In addition, semiconductor plasmonics is promising for fundamental studies of nonlinear light-matter interaction. In perspective, new insight into controversial questions about the origin of nonlinear susceptibility in nanostructures24,29 and its spectral dispersion6 can be gained by tuning the plasma edge of doped germanium thus providing a degree of freedom not accessible with standard metals. Finally, combined with the all-optical ultrafast generation of free carriers in intrinsic Ge nanostructures25, this work opens the possibility for active control over coherent mid-infrared light sources with unprecedented spatiotemporal confinement. Methods: Heavily doped Ge film growth and material characterization. As a first fabrication step, a highly n-doped Ge epilayer was grown on an intrinsic Si(001) substrate by low-energy plasma-enhanced chemical vapour deposition (LEPECVD)31. The deposition was performed at 500 °C with a growth rate of about 1 nm/s. The n-type doping was obtained in-situ by adding 0.035 standard cubic centimetre per minute (sccm) of PH3 to 20 sccm of GeH4, achieving an active doping of 2.5×1019 cm-3, which corresponds to a plasma frequency of about 31 THz (9.7 μm wavelength). The thin film reflectivity spectrum is recorded using Fourier-transform infrared spectroscopy and allows extracting the dielectric function of the material (Fig. 1a,b) by a combination of Drude-like modeling and the use of the Kramers-Kronig relations17. It should be noted here that recent works also demonstrated that the same growth technique, when combined with post-growth annealing procedures, allows plasma frequencies up to about 95 THz (3.1 µm wavelength) to be reached32,33. Antenna fabrication. From these heavily doped Ge films, isolated antenna structures (Fig. 1c) are fabricated via electron-beam lithography with hydrogen silsesquioxane resist and anisotropic reactive ion etching with fluorine chemistry34,35. The half-wavelength length of the two arms in the double-rod antenna geometry is selected in order to maximize the resonantly-enhanced currents. Considering the refractive index of the material and of the environment, numerical simulations predict that the first-order localized plasmon-polariton resonance occurs for an arm length of 3 to 4 µm when excited at a wavelength around 12 µm with linearly polarized light. By the use of state- of-the-art nanofabrication technology, the relatively large structures can be fabricated close to perfection with steep sidewalls, sharp edges and reproducibility to within a nm of the design34. 7 Mid-IR laser microscopy setup and antenna characterization To excite the resonant antenna structures, we developed a high-power femtosecond laser system36 based on a regenerative Yb:KGW amplifier with a repetition rate of 50 kHz. The fundamental pulses at a wavelength of 1028 nm with a duration of 250 fs drive a non-collinear optical parametric amplifier (NOPA) 37,38 that generates broadband pulses tunable between 1050 and 1400 nm. Intense mid-IR pulses are subsequently produced via nonlinear phase-matched difference frequency mixing of the NOPA pulses (pulse energy up to 2.2 µJ) with 36-µJ-pulses from the Yb:KGW amplifier in a GaSe crystal39. Depending on the crystal thickness, phase-matching angle and NOPA wavelength, either broadband mid-IR transients spanning from 8 to 20 µm wavelength or narrowband but intense pulses of up to 100 nJ tunable over the same spectral range can be generated. For these experiment, driving mid-IR pulses are set to a duration of 300 fs at a bandwidth of 1.5 THz. A custom-build confocal microscope consisting of two dispersion-free gold-coated reflective Cassegrain objectives with a numerical aperture of 0.5 and working in transmission geometry is used to study the linear and nonlinear mid- IR response of individual antenna structures. The first objective focusses the driving pulses onto the antenna samples yielding excitation fields of up to 5 MV/cm and reaching a diffraction limited spot size of approximately one wavelength. The second objective collects the transmitted radiation as well as the nonlinear emission. A pinhole with a diameter of 150 µm in the image plane reduces the detection field of view to enhance the lateral resolution, rejects stray light coming from outside the focal volume and lowers the depth of field to better discriminate between the substrate bulk emission background and the antenna signal. The detection of the transmitted and emitted nonlinear radiation is performed via electro-optic sampling (EOS) in 80-µm-thick GaSe40 or with a liquid nitrogen cooled mercury cadmium telluride (MCT) detector with low-noise lock-in readout. To discriminate between the fundamental and third-harmonic radiation, crystalline filters of InSb (a long-pass filter with a transmission edge at 7 µm wavelength), sapphire (a 5 µm short-pass filter), and CaF2 (a 9 µm short-pass filter) are employed. A monochromator with blazed mid-IR gratings was employed to record the exact emission spectrum from the samples. Simulation of the antenna response. Three-dimensional finite-difference time-domain simulations (FDTD solutions, Lumerical Inc.41) were performed to reproduce the observed extinction and emission spectra. The dielectric constant of the heavily-doped Ge material and of the underlying Si substrate was experimentally characterized as described above and used for the simulations. The antennas are illuminated with a Gaussian source at the fundamental wavelength, featuring the same numerical aperture as the illumination objective in the experimental setup. The influence of the numerical aperture of the collection objective in the extinction maps was taken into account as well by considering the radiation pattern of the antennas at the fundamental wavelength. In order to model the nonlinear signal and reproduce its dependence on the arm length, we calculate the third power of the electric field intensity integrated over the whole antenna volume, mimicking a standard third-order bulk nonlinear process inside the antennas. This simple approach already reproduces the experimental data with high fidelity. 8 References 1. Novotny, L. & van Hulst, N. Antennas for light. Nat. Photonics 5, 83–90 (2011). 2. Biagioni, P., Huang, J.-S. & Hecht, B. Nanoantennas for visible and infrared radiation. Rep. Prog. Phys. 75, 24402 (2012). 3. Kauranen, M. & Zayats, A. V. Nonlinear Plasmonics. Nat. Photonics 6, 737–748 (2012). 4. Celebrano, M., Wu, X., Baselli, M., Grossmann, S., Biagioni, P., Locatelli, A., De Angelis, C., Cerullo, G., Osellame, R., Hecht, B., Duò, L., Ciccacci, F. & Finazzi, M. Mode matching in multiresonant plasmonic nanoantennas for enhanced second harmonic generation. Nat. Nanotechnol. 10, 412–417 (2015). 5. Hanke, T., Krauss, G., Träutlein, D., Wild, B., Bratschitsch, R. & Leitenstorfer, A. Efficient Nonlinear Light Emission of Single Gold Optical Antennas Driven by Few-Cycle Near-Infrared Pulses. Phys. Rev. Lett. 103, 257404 (2009). 6. Knittel, V., Fischer, M. P., Vennekel, M., Rybka, T., Leitenstorfer, A. & Brida, D. Dispersion of the nonlinear susceptibility in gold nanoantennas. Phys. Rev. B 96, 125428 (2017). 7. Pfullmann, N., Waltermann, C., Kovačev, M., Knittel, V., Bratschitsch, R., Akemeier, D., Hütten, A., Leitenstorfer, A. & Morgner, U. Nano-antenna-assisted harmonic generation. Appl. Phys. B 113, 75–79 (2013). 8. Rybka, T., Ludwig, M., Schmalz, M. F., Knittel, V., Brida, D. & Leitenstorfer, A. Sub-cycle optical phase control of nanotunnelling in the single-electron regime. Nat. Photonics 10, 667–670 (2016). 9. Vampa, G., Ghamsari, B. G., Siadat Mousavi, S., Hammond, T. J., Olivieri, A., Lisicka-Skrek, E., Naumov, A. Y., Villeneuve, D. M., Staudte, A., Berini, P. & Corkum, P. B. Plasmon-enhanced high-harmonic generation from silicon. Nat. Phys. 13, 659–662 (2017). 10. Biagioni, P., Brida, D., Huang, J.-S., Kern, J., Duò, L., Hecht, B., Finazzi, M. & Cerullo, G. Dynamics of Four-Photon Photoluminescence in Gold Nanoantennas. Nano Lett. 12, 2941–2947 (2012). 11. Lee, J., Tymchenko, M., Argyropoulos, C., Chen, P.-Y., Lu, F., Demmerle, F., Boehm, G., Amann, M.-C., Alù, A. & Belkin, M. A. Giant nonlinear response from plasmonic metasurfaces coupled to intersubband transitions. Nature 511, 65–69 (2014). 12. Law, S., Yu, L., Rosenberg, A. & Wasserman, D. All-semiconductor plasmonic nanoantennas for infrared sensing. Nano Lett. 13, 4569–74 (2013). 13. N'Tsame Guilengui, V., Cerutti, L., Rodriguez, J.-B., Tournié, E. & Taliercio, T. Localized surface plasmon resonances in highly doped semiconductors nanostructures. Appl. Phys. Lett. 101, 161113 (2012). 14. Panah, M. E. A., Han, L., Norrman, K., Pryds, N., Nadtochiy, A., Zhukov, A. E., Lavrinenko, A. V. & Semenova, E. S. Mid-IR optical properties of silicon doped InP. Opt. Mater. Express 7, 2260 (2017). 15. Soref, R. Mid-infrared photonics in silicon and germanium. Nat. Photonics 4, 495–497 (2010). 16. Baldassarre, L., Sakat, E., Frigerio, J., Samarelli, A., Gallacher, K., Calandrini, E., Isella, G., Paul, D. J., Ortolani, M. & Biagioni, P. Midinfrared Plasmon-Enhanced Spectroscopy with Germanium Antennas on Silicon Substrates. Nano Lett. 15, 7225–7231 (2015). 17. Frigerio, J., Ballabio, A., Isella, G., Sakat, E., Pellegrini, G., Biagioni, P., Bollani, M., Napolitani, E., Manganelli, C., Virgilio, M., Grupp, A., Fischer, M. P., Brida, D., Gallacher, K., Paul, D. J., Baldassarre, L., Calvani, P., Giliberti, V., 9 Nucara, A. & Ortolani, M. Tunability of the dielectric function of heavily doped germanium thin films for mid- infrared plasmonics. Phys. Rev. B 94, 85202 (2016). 18. Neubrech, F., Pucci, A., Cornelius, T. W., Karim, S., García-Etxarri, A. & Aizpurua, J. Resonant plasmonic and vibrational coupling in a tailored nanoantenna for infrared detection. Phys. Rev. Lett. 101, 2–5 (2008). 19. Adato, R., Yanik, A. A., Amsden, J. J., Kaplan, D. L., Omenetto, F. G., Hong, M. K., Erramilli, S. & Altug, H. Ultra- sensitive vibrational spectroscopy of protein monolayers with plasmonic nanoantenna arrays. Proc. Natl. Acad. Sci. U. S. A. 106, 19227–32 (2009). 20. Brown, L. V., Zhao, K., King, N., Sobhani, H., Nordlander, P. & Halas, N. J. Surface-Enhanced Infrared Absorption Using Individual Cross Antennas Tailored to Chemical Moieties. J. Am. Chem. Soc. 135, 3688–3695 (2013). 21. Strikwerda, A. C., Zalkovskij, M., Iwaszczuk, K., Lorenzen, D. L. & Jepsen, P. U. Permanently reconfigured metamaterials due to terahertz induced mass transfer of gold. Opt. Express 23, 11586 (2015). 22. Hon, N. K., Soref, R. & Jalali, B. The third-order nonlinear optical coefficients of Si, Ge, and Si1-xGex in the midwave and longwave infrared. J. Appl. Phys. 110, 11301 (2011). 23. Zhang, L., Agarwal, A. M., Kimerling, L. C. & Michel, J. Nonlinear Group IV photonics based on silicon and germanium: from near-infrared to mid-infrared. Nanophotonics 3, 247–268 (2014). 24. Boyd, R. W., Shi, Z. & De Leon, I. The third-order nonlinear optical susceptibility of gold. Opt. Commun. 326, 74– 79 (2014). 25. Fischer, M. P., Schmidt, C., Sakat, E., Stock, J., Samarelli, A., Frigerio, J., Ortolani, M., Paul, D. J., Isella, G., Leitenstorfer, A., Biagioni, P. & Brida, D. Optical Activation of Germanium Plasmonic Antennas in the Mid- Infrared. Phys. Rev. Lett. 117, 47401 (2016). 26. Mesch, M., Metzger, B., Hentschel, M. & Giessen, H. Nonlinear Plasmonic Sensing. Nano Lett. 16, 3155–3159 (2016). 27. Höppener, C. & Novotny, L. Imaging of membrane proteins using antenna-based optical microscopy. Nanotechnology 19, 384012 (2008). 28. Palomba, S. & Novotny, L. Near-field imaging with a localized nonlinear light source. Nano Lett. 9, 3801–3804 (2009). 29. Ginzburg, P., Krasavin, A. V., Wurtz, G. A. & Zayats, A. V. Nonperturbative Hydrodynamic Model for Multiple Harmonics Generation in Metallic Nanostructures. ACS Photonics 2, 8–13 (2015). 30. Sakat, E., Giliberti, V., Bollani, M., Notargiacomo, A., Pea, M., Finazzi, M., Pellegrini, G., Hugonin, J.-P., Weber- Bargioni, A., Melli, M., Sassolini, S., Cabrini, S., Biagioni, P., Ortolani, M. & Baldassarre, L. Near-Field Imaging of Free Carriers in ZnO Nanowires with a Scanning Probe Tip Made of Heavily Doped Germanium. Phys. Rev. Appl. 8, 54042 (2017). 31. Rosenblad, C., Deller, H. R., Dommann, A., Meyer, T., Schroeter, P., and von Känel, H. Silicon epitaxy by low- energy plasma enhanced chemical vapor deposition. J. Vac. Sci. Technol. A 16, 2785-2790 (1998). 32. Frigerio, J., Ballabio, A., Gallacher, K., Gilberti, V., Baldassarre, L., Millar, R. W., Milazzo, R., Maiolo, L., Minotti, A., Bottegoni, F., Biagioni, P., Paul, D. J., Ortolani, M., Pecora, A., Napolitani, E. & Isella, G. Optical properties of 10 highly n-doped germanium obtained by in situ doping and laser annealing. J. Phys. D. Appl. Phys. 50, 465103 (2017). 33. Paul, D. J., Gallacher, K., Millar, R. W., Giliberti, V., Calandrini, E., Baldassarre, L., Fischer, M. P., Frigerio, J., Ballabio, A., Sakat, E., Pellegrini, G., Brida, D., Isella, G., Ortolani, M. & Biagioni, P. n-Ge on Si for mid-infrared plasmonic sensors. in 2017 IEEE Photonics Soc. Summer Top. Meet. Ser. 125–126 (IEEE, 2017). doi:10.1109/PHOSST.2017.8012682 34. Mirza, M. M., Zhou, H., Velha, P., Li, X., Docherty, K. E., Samarelli, A., Ternent, G. & Paul, D. J. Nanofabrication of high aspect ratio (~50:1) sub-10 nm silicon nanowires using inductively coupled plasma etching. J. Vac. Sci. Technol. B Microelectron. Nano. Struct. 30, 06FF02 (2012). 35. Samarelli, A., Frigerio, J., Sakat, E., Baldassarre, L., Gallacher, K., Finazzi, M., Isella, G., Ortolani, M., Biagioni, P. & Paul, D. J. Fabrication of mid-infrared plasmonic antennas based on heavily doped germanium thin films. Thin Solid Films 602, 52–55 (2016). 36. Grupp, A., Budweg, A., Fischer, M. P., Allerbeck, J., Soavi, G., Leitenstorfer, A. & Brida, D. Broadly tunable ultrafast pump-probe system operating at multi-kHz repetition rate. J. Opt. 20, 14005 (2018). 37. Antipenkov, R., Varanavičius, A., Zaukevičius, A. & Piskarskas, A. P. Femtosecond Yb:KGW MOPA driven broadband NOPA as a frontend for TW few-cycle pulse systems. Opt. Express 19, 3519–24 (2011). 38. Brida, D., Manzoni, C., Cirmi, G., Marangoni, M., Bonora, S., Villoresi, P., De Silvestri, S. & Cerullo, G. Few-optical- cycle pulses tunable from the visible to the mid-infrared by optical parametric amplifiers. J. Opt. 12, 13001 (2010). 39. Junginger, F., Sell, A., Schubert, O., Mayer, B., Brida, D., Marangoni, M., Cerullo, G., Leitenstorfer, A. & Huber, R. Single-cycle multiterahertz transients with peak fields above 10 MV/cm. Opt. Lett. 35, 2645–2647 (2010). 40. Kübler, C., Huber, R., Tübel, S. & Leitenstorfer, A. Ultrabroadband detection of multi-terahertz field transients with GaSe electro-optic sensors: Approaching the near infrared. Appl. Phys. Lett. 85, 3360 (2004). 41. FDTD Solutions, version 8.17.1157; Lumerical Solutions, Inc.: Canada, 2013. Acknowledgements The authors acknowledge support from European Commission via the Marie Curie Carrier Integration Grant and the Seventh Framework Programme under Grant No. 613055, from the Deutsche Forschungsgemeinschaft through the Emmy Noether programme and EPSRC under Grant No. EP/N003225/1. The authors would like to thank the staff of the James Watt Nanofabrication Centre for help with fabricating the devices. 11
1710.06162	3	1710	2017-11-12T19:00:34	Probing dynamic behavior of electric fields and band diagrams in complex semiconductor heterostructures	[ "physics.app-ph" ]	Modern bandgap engineered electronic devices are typically made of multi-semiconductor multi-layer heterostructures that pose a major challenge to silicon-era characterization methods. As a result, contemporary bandgap engineering relies mostly on simulated band structures that are hardly ever verified experimentally. Here, we present a method that experimentally evaluates bandgap, band offsets, and electric fields, in complex multi-semiconductor layered structures and it does so simultaneously in all the layers. The method uses a modest optical photocurrent spectroscopy setup at ambient conditions. The results are analyzed using a simple model for electro-absorption. As an example, we apply the method to a typical GaN high electron mobility transistor structure. Measurements under various external electric fields allow us to experimentally construct band diagrams, not only at equilibrium, but also under any other working conditions of the device. The electric fields are then used to obtain the charge carrier density and mobility in the quantum well as a function of the gate voltage over the entire range of operating conditions of the device. The principles exemplified here may serve as guidelines for the development of methods for simultaneous characterization of all the layers in complex, multi-semiconductor structures.	physics.app-ph	physics	arXiv:1710.06162v3 physics.app-ph 12 November 2017 12 November 2017 Probing dynamic behavior of electric fields and band diagrams in complex semiconductor heterostructures Yury Turkulets and Ilan Shalish* Ben Gurion University of the Negev, Beer Sheva, Israel Modern bandgap engineered electronic devices are typically made of multi-semiconductor multi-layer heterostructures that pose a major challenge to silicon-era characterization methods. As a result, contemporary bandgap engineering relies mostly on simulated band structures that are hardly ever verified experimentally. Here, we present a method that experimentally evaluates bandgap, band offsets, and electric fields, in complex multi-semiconductor layered structures and it does so simultaneously in all the layers. The method uses a modest optical photocurrent spectroscopy setup at ambient conditions. The results are analyzed using a simple model for electro-absorption. As an example, we apply the method to a typical GaN high electron mobility transistor structure. Measurements under various external electric fields allow us to experimentally construct band diagrams, not only at equilibrium, but also under any other working conditions of the device. The electric fields are then used to obtain the charge carrier density and mobility in the quantum well as a function of the gate voltage over the entire range of operating conditions of the device. The principles exemplified here may serve as guidelines for the development of methods for simultaneous characterization of all the layers in complex, multi-semiconductor structures. 1 Introduction While silicon technology is facing the challenges of quantum mechanical tunneling, bandgap-engineered devices thrive on quantum mechanical effects to produce faster switching.1 Yet, while silicon technology can do with a single semiconductor, bandgap engineered devices typically require a stack of several nanometer-scale thin-films of semiconductors of different bandgaps. These complex heterostructures present a challenge to silicon-era characterization tools, which are mostly capable of characterizing structures made of a single semiconductor material. Photoemission spectroscopies, e.g., x-ray and ultra- violet photoelectron spectroscopies, have been successfully used in studying band-structure of single heterojunctions,2,3 but fall short of characterizing stacks of more than a single heterojunction, such as those in a typical light emitting diode, laser diode, or a high-electron-mobility transistor. So far, this shortage in characterization methods has been compensated for mostly by pure simulations, semi-empirical simulations, and theoretical calculations.4,5,6 The advantage of using junctions of more than a single semiconductor was already recognized by Shockley in his patent of the bipolar junction transistor, while the foundations for the use of semiconductor heterostructures were laid later by Kroemer,7 and dubbed "bandgap engineering" by Capasso.8,9 The early work on heterostructures at Bell Labs was also the ground for the invention of modulation doping by Dingle10 followed by the invention of the high electron mobility transistor (HEMT).11 The HEMT evolved from a single GaAs- AlGaAs heterojunction in 1980 to the state-of the-art of a multi- layer AlGaN-GaN of today.12,13 One of the challenges faced by bandgap engineers today is to verify that the engineered band structure of the multi-layer heterostructure was actually accomplished. Layer thickness and composition are used to estimate the bandgap and band-offsets that should result, while a more complex process is required for the estimate of the built- in electric fields. Eventually, when the structure is materialized, many methods can be used to give partial validation of some of the parameters. In practice, however, the designer will rarely bother to use several complimentary methods to obtain a rough estimate of the band structure, and in most cases, will make do with the final electrical tests of the device. The purpose of this work was to develop an experimental tool to measure the band structure: bandgaps, band offsets, and built-in electric fields, in a multi-layer heterostructure, at equilibrium, and also under external electric fields – all using a single method, and in a single measurement. Here, we propose a tool to characterize, simultaneously, all the layers in a multi- semiconductor structure and to construct energy band diagrams of complex heterostructures, not only at equilibrium, but also over the entire range of operating conditions. We demonstrate the method on a HEMT structure without limiting the generality, as we assume that the required test structure can always be fabricated on any semiconductor heterostructure. 2 Proposed Method and Model The approach taken here is to use an optical spectroscopy in conjunction with an electrical measurement. While the most commonly used optical spectroscopy is photoluminescence, it will reveal, in most cases, only the lowest bandgap in the structure. This is because carriers tend to descend to the lowest bandgap before recombining. To avoid this limitation, one may use a spectroscopy that is based on absorption rather than emission. In a structure comprised of nanometer-scale-thin layers, photons can reach and be absorbed in any layer in the 1 arXiv:1710.06162v3 physics.app-ph structure. The absorption of photons takes place mainly at the bandgap energy of each layer, and therefore, responses of layers of different bandgaps should be observed at different photon energies within the same spectral response curve. This way, all or most of the responses may be recorded in a single measurement of a spectral response curve. The absorption of photons at the bandgap energy generates electron-hole pairs, and the availability of photo-generated carriers gives rise to changes in various electrical properties. In to detect electric current. This this work, we chose configuration is commonly used in spectral photo-conductivity and in internal photoemission.14,15 In the latter method, the current flows perpendicular to a potential barrier formed between two materials, and only when the photon energy is great enough to excite electrons over the barrier, can an electric current be detected. Thus, in a stack of more than a single heterojunction, a current can be observed only at photon energies exceeding the highest barrier. This renders the internal photoemission configuration inadequate for our purposes.16 A better configuration would be to fabricate two parallel contacts to the lowest bandgap in the structure. Optically generated carriers will typically descend to the lowest bandgap layer, and therefore, the responses of all the layers may be detected as steps in the electric current – a single step for each bandgap- energy in the structure. A spectrum taken from a typical HEMT device, showing a set of such steps, is shown in Figure 1. The layer structure of the device is shown in the inset. Such a spectrum contains all the possible band-to-band transitions, whether within the same material (the bandgap), or between adjacent materials. The energy differences between these two types of transitions can be used to calculate band offsets. Each spectral step contains information not only on the optical transition energy (bandgap), but also on the electric field in the corresponding layer. Careful inspection reveals that each step commences well below the actual bandgap and rises in a sloped manner. This early photocurrent response reflects the shape of the absorption edge in semiconductors. The absorption edge, as reflected in the photocurrent, is affected mainly by the typically strong electric fields present at semiconductor interfaces.17 The electric field assists photons with energy smaller than the bandgap to excite electrons across the forbidden gap by adding energy from the electric field. The effect of electric fields on the absorption edge in semiconductors is commonly known as the Franz-Keldysh effect.18,19 We have previously modeled this effect on photoconductivity in single-material structures.20 Using the same model on the data of each step produces a linear curve that intercepts the photon energy axis at the exact bandgap (or optical transition) energy, and the slope of this curve may be used to obtain the maximum electric field in that layer. Each spectrum is thus analyzed for its various steps to produce a band diagram. The electrical current at each step of the photo-response was modelled using Eq. 120 2 12 November 2017 Fig. 1 Typical photocurrent spectrum measured between the drain and source contacts as a function of photon energy (Red curve). The spectrum is comprised of a series of steps, each saturating around its corresponding optical transition energy. When an optical transition takes place between the valence band and the conduction band of the same material, the photon energy equals the bandgap of this material. The photocurrent rise associated with each transition precedes the energy of the actual transition due to an effect of the electric field in the corresponding layer. Thus, the preceding slope can be used to measure the electric field in the corresponding layer. We added a sketch of the steps without their preceding slope to aid the eye (green curve). The inset shows the structure of the high electron mobility transistor used in this experiment. The top 40 nm of such device typically different semiconductors. Using the proposed method, it is possible to obtain for each layer the bandgap and the maximum electric field under any external electric field. It is also possible to get the band offsets from interlayer transitions. contains several active layers of  hv  I  I D   I S  I D   1   hvR     exp       E hv  g E     3 2     ( 1 ) where ID – dark current (practically it is the current preceding the rise), IS – current following the rise, and R(hv) – spectral reflectance from the surface of the sample (in practice, we took into account only the reflection of the air interface, because we found the effect of reflection in general to be rather minor and negligible for that we intra-layer reflections inter- and arXiv:1710.06162v3 physics.app-ph neglected), Eg is the bandgap or the energy of the involved optical transition, hv is the photon energy, and E is given by E  2 3 qE 3  24 m       ( 2 ) where E is the maximum of the electric field in the layer, q is the electron charge, m is the reduced effective mass, and ħ is the reduced Plank constant. Rearranging Equation 1, we get  hv  y      ln   I  S   hv I  D I I D     ln 1   hvR 3 2      E hv  g E  ( 3 ) The advantage of Eq. 3 is that its right-hand side is a linear expression of the photon energy. Presenting the data this way, each step response transforms into a linear curve that intercepts the photon energy axis at the exact optical transition energy. From the slope of this line, one can extract the maximum electric field in that semiconductor layer. The same treatment is given independently to each of the layers of a complex semiconductor heterostructure. As a matter of fact, the basic method does not require more than two Ohmic contacts to construct the equilibrium band diagram. However, the use of a complete transistor affords an additional experimental handle, the gate. The gate can be used to apply an external electric field during the spectral acquisition. Using a set of spectra, acquired over a range of gate voltages, it is possible to construct a set of band diagrams for a range of operating conditions of the device. Figure 2 shows a surface plot of a set of 150 photo-current spectra obtained over a photon energy range of 2.9 to 4.425 eV (5-meV steps) and a gate voltage range of 0 to -7.5 Volts (in 50-mV increments) from the heterostructure described in Figure 1. The curves were obtained under source to drain voltage of 0.1 Volts – within the linear mode of the transistor. Similarly, we also applied the method for source-drain voltage of 8 Volt corresponding to the saturation mode of the transistor (not shown). The top panel of Figure 3 shows overlapped plots of normalized photocurrent spectra at the photon-energy range near the bandgap of GaN. Each plot was acquired under a different external field. The plots were cut from the same spectra shown in Figure 2 to emphasize the effect of the applied external field on the various photocurrent steps – the slopes preceding the band-edge are observed to decrease with the increasing applied field. Applying the model (Equation 3) to each of the spectra on the top panel of Figure 3 produces a corresponding set of linear curves (Figure 3 – bottom panel). This graphic method provides an easy confirmation to validate the assumption of Franz-Keldysh electro-absorption. If the Franz-Keldysh effect does not take place, the use of Equation 3 is unlikely to produce a linear curve. All the curves intersect the photon energy axis at the GaN bandgap, while the slopes are observed to decrease with the increasing field. A similar analysis was also carried out for the step response of the AlGaN. The linear portion of each curve is observed over a range of about 0.1 eV preceding the bandgap. Below this range, 12 November 2017 Fig. 2 A series of 150 channel photocurrent spectra acquired from the high electron mobility transistor (HEMT) shown in Fig. 1 under a range of external electric fields. An external field was applied by applying voltage to the gate contact of the transistor (VGS), while the voltage between the drain and the source (VDS) was kept constant. In this figure, VDS was 0.1 V. Fig. 3 Top panel: Overlapped plots of photocurrent spectra at the photon-energy range near the bandgap of GaN. Each plot is for a different external field. The plots were cut from the spectra shown in Fig. 2 to show the effect of the applied external field on the GaN step response – the slope preceding the band-edge is seen to decrease with the increasing applied field. Bottom panel: A corresponding set of straight lines is obtained by applying the model (Eq. 3) to each of the spectra in the top panel. In both plots, the sampling has been reduced for clarity. 3 arXiv:1710.06162v3 physics.app-ph the data deviate from the linear course, because our model approximates a parabolic barrier with a triangular barrier. This approximation is good close to the bandgap but departs from reality further away. At the bandgap, the spectra deviate again from the linear course due to the different above-gap physics. 3 Experimental Details The semiconductor heterostructure was grown by metal organic chemical vapor deposition (MOCVD) on c-plane sapphire. The layer sequence was an AlN nucleation layer, 2 µm of undoped GaN, 11 nm of Al30Ga70N, and 1.5 nm GaN cap layer. For device isolation, shallow mesas were dry etched in chlorine- based plasma. After removing the GaN cap layer, 100 nm of Si3N4 was deposited on top of the structure by plasma enhanced chemical vapor deposition (PECVD). Contact pads and 3 µm wide gate trenches were dry etched in Si3N4. All the metal contacts were deposited using e-beam thermal evaporation. Source and drain Ohmic contacts were Ti(30nm)/Al(70nm)/Ni(30nm)/Au(100nm) annealed at 900 C for 1 min in nitrogen ambient. The gate contact was a Ni(30nm)/Au(100nm) Schottky barrier. For spectral data acquisition, the samples were placed in a dark and shielded box at atmospheric room temperature conditions. Illumination was applied from the gate side. For illumination, we used a 300 Watt Xe light source, monochromitized using a Newport Corp. double MS257 monochromator and further filtered by order- sorting long-pass filters. During spectral acquisition, a constant voltage was applied between the source and drain contacts. Electrical measurements were carried out using two Keithley 2400 source-meters. To avoid the effect of light on the measured electric field, we worked at a photon flux small enough, so that the maximum produced photocurrent is about two orders of magnitude smaller than the dark current. The intensity of light was 6.5 W/cm2 at 280 nm. To avoid features resulting from the spectral distribution of the lamps, we operated the spectrometer in a closed control loop maintaining a constant photon flux throughout the spectral range of the measurement. The wavelength was stepped at equal photon energy steps. At each photon energy point, a full scan of the gate voltage range was performed. Each data point is an average of 30 consecutive measurements. 4 Results and Discussion A full analysis of a typical transistor is given in Fig. 4. The figure has two columns. The left column shows an analysis at the linear mode, while the right column shows the same for the saturation mode. The first row shows the maximum electric fields in the AlGaN and GaN layers as a function of the applied gate voltage, as calculated from the optical response curves. The AlGaN layer is fully depleted, and therefore the field does not vary with the position within the layer. On the other hand, 12 November 2017 the GaN is in a state of accumulation at the heterojunction, forming a triangular quantum well with 2-dimensional electron gas (2DEG). The field in the GaN layer reaches its maximum within the quantum well and gradually decreases to zero as one gets away from the junction. The point where the optically induced band-to-band transitions take place in the layer is (always) the point where the electric field is the largest, provided there exist allowed states for electrons. For the GaN, this point is where the first discrete level (eigenstate) in the well meets the GaN conduction band. Given the evolution of electric fields on both sides of the 2DEG, it is now possible to calculate its sheet charge density as a function of the gate voltage. For example, in our AlGaN/GaN structure, the 2DEG charge density, qnS, is given by the discontinuity in the electric displacement field at the boundary between the AlGaN and GaN layers. The electric displacement field on the AlGaN side, DAlGaN, is the sum of the known spontaneous polarization in AlGaN, PSP,AlGaN, the piezoelectric polarization resulting from the mismatch to the GaN layer underneath it, PPE,AlGaN, and the measured electric field, EAlGaN. In the GaN, the electric displacement field outside the quantum well is created by PSP,GaN and the measured electric field, EGaN.21,22 The values of the polarization vectors are -0.034 Cb/m2, -0.0464 Cb/m2 and -0.00983 Cb/m2 for PSP,GaN, PSP,AlGaN and PPE,AlGaN (Al composition of 30%), respectively.23 Hence, we get24,25  P ,SP  P GaN,SP  P ,PE  E(   E(  ( 4 ) qn S   GaN  AlGaN  AlGaN )  AlGaN ) Substituting the measured electric fields in the AlGaN and GaN layers in Eq. 4, we can now graph the evolution of 2DEG charge density as a function of the gate voltage (second row in Fig. 4). Approaching the threshold voltage, the 2DEG gradually diminishes and the resistance of the channel increases. Since we measure the channel current, the photocurrent diminishes as well at this range, until, at a certain low value, the vanishing signal to noise ratio makes the results less reliable. This range is shaded in gray in the first row. We extrapolated the trend preceding this range and used this extrapolation in the calculation of the 2DEG charge density within the uncertain range. The third row of Fig. 4 shows drain current measured in the dark over the same range of gate voltages in both modes. As expected, it roughly follows a similar trend as the 2DEG charge density. Using this drain current and the 2DEG charge density, we were now able to draw the channel mobility as a function of the gate voltage in each of the transistor modes. The mobility is shown at the bottom, fourth row, of Fig. 4. Using the method on our AlGaN/GaN structure provided us with the bandgaps, band offsets, and built-in electric fields. This is basically all that is required to construct a band diagram. Since we get a band diagram for each of the applied gate voltages, we can actually draw the evolution of the band diagram with the applied external field. Figure 5 shows 3 specific band diagrams for 3 specific gate voltages. To 4 arXiv:1710.06162v3 physics.app-ph 12 November 2017 construct the band diagram, we assumed linear bands in the AlGaN layer. The GaN layer required a one-dimensional Poison-Schroedinger equation solver. The solution used our measured values of the 2DEG charge density and the GaN electric field at the first sub-band in the GaN quantum well. Fig. 4 Two sets of data for two transistor modes shown as a function of the gate voltage: 1 Linear Mode – Left column figures, and 2 Saturation mode – Right column. Row 1: Peak (maximum) electric fields in the GaN and the AlGaN layers obtained from the measured spectra. Note that the measured field is the maximal value of the field also in the lateral dimension, i.e., along the channel. The maximal field along the channel occurs in this device under the gate, at its drain side. Row 2: 2DEG charge density calculated from the electric fields of Row 1. Row 3: Drain current measured in the dark. Row 4: Channel mobility calculated from the 2DEG charge density of Row 2 and the drain current of Row 3. Note the grayed regions in Row 1. At these regions the channel is almost closed, the drain current drops, and therefore the sensitivity of our method reaches a limit. The data in these ranges (open circles) is not reliable. We have extrapolated the data as shown in the full-line curves. The 2DEG charge densities over these ranges are calculated from the extrapolated curves. The same caveat goes to the channel mobility of Row 4. However, mobility is meaningless where no conduction is possible. So far, we have treated only the case of electron-hole generation. Excitons are likely to be generated as well. Dow and Redfield showed that excitonic absorption followed a model different from the Franz-Keldysh model.26 If excitonic absorption affected the photocurrent, our graphic method would not yield a straight line. Therefore, if we do obtain a straight line, it serves to confirm the adequacy of the Franz-Keldysh model. The absence of exciton expression in our photocurrent is probably due to the fact that excitons are electrically neutral and require significant dissociation to be able to contribute. Dissociation of excitons is not always an effective process and poses a major bottleneck in photovoltaic efficiency. Another case not addressed so far is the absorption in quantum wells and the adequacy of the Franz-Keldysh model to various cases falling under this category. Franz-Keldysh electro- absorption requires that at least one type of carrier will not be confined. Hence, in the case shown here, of a triangular quantum well, the model is clearly adequate, because only electrons are confined. However, there is still the case of a double heterostructure. It has been shown by Miller et al. that 5 arXiv:1710.06162v3 physics.app-ph high electric fields give rise to the quantum confined Stark effect.27 This effect relates to excitonic absorption affecting the exciton binding energy and increasing its survival. As excitons do not carry charge, the Stark effect may not have a direct effect on photocurrent. Yet, another work by Miller et al. suggests that even if the effect of excitons is altogether excluded, bulk-like Franz-Keldysh effect, i.e. the smearing of the absorption edge to low energies, cannot take place in narrow quantum-wells.28,29 However, one should bear in mind that their calculations were carried out under the assumption of infinite energy barriers. Most of the practical quantum wells are actually very far from meeting this assumption. Furthermore, valence band wells are typically extremely shallow with the energy separation between eigenstates on the order of the phonon energy, kT. Clearly, these are not true eigenstates. No real hole confinement can take place in most of these real cases, and hence, it may not be unpractical to expect a bulk Franz- Keldysh photocurrent response in most of the real quantum wells at room temperature. The following limitations should be kept in mind when using the proposed method. First, the method can yield only the magnitude of the electric field but not its direction. Second, there is a practical lower bound to the detectable field. This is because various other effects may cause broadening of the photoresponse resulting in a minor slope of the photocurrent step. The most obvious source for such broadening is, for example, the resolution of the spectrometer. In this work, the lowest detectable electric field was ~0.1 MV/cm. In any case, the fields in heterojunctions are typically much greater than this bound (see e.g., the electric fields in Fig. 4). In field-effect transistors, the electric field varies in the lateral direction (along the channel) as well. Since electro-absorption takes place at the point of strongest electric field, our band diagrams show a cross-section of the transistor under the gate at its drain side, which is the point of strongest field in the HEMT. Finally, in the case of a HEMT, very often, a metal field-plate is placed on top of the gate that may optically screen the gate and its vicinity. In this case, it may be difficult to illuminate the point of highest electric field. 5 Conclusion The proposed method provides the energies of the various optical transitions, which can be used to evaluate the energy gaps at the various layers and the interlayer band offsets. It also provides the built-in electric fields at each of the layers, which can be used to construct the band diagram of the structure and the interface charges, e.g., the 2DEG charge density in quantum wells, and the channel mobility. It may also be used under external electric fields to explore the dynamic behavior of the band structure and of the interface charges and mobility. The model is based on the Franz-Keldysh effect, and it provides a graphical confirmation for the validity of this assumption. In this work, we have used the channel current as the measured electrical property. Various other electrical properties may also 12 November 2017 be measured using the same setup, and their physics may be used for the evaluation of various other semiconductor material properties in complex bandgap-engineered semiconductor layer-stacks. Fig. 5 Band diagrams for 3 specific gate voltages (from zero to threshold voltage), calculated from the data of Fig. 4. The proposed method allows one to draw a band diagram for each single gate voltage (e.g., a diagram for each of the 150 spectra of Fig. 2) to visualize the evolution of the band diagram over the entire range of operation of the transistor. Acknowledgements This work was funded by a research grant from the Israeli Ministry of Defence (MAFAT). References 1 J.A. del Alamo. Nanometre-scale electronics with III–V compound semiconductors. Nature 2011, 479, 317. DOI:10.1038/nature10677 2 L. Sun, R. Haight, P. Sinsermsuksakul, S.B. Kim, H.H. Park, R.G. Gordon. Band alignment of SnS/Zn(O,S) heterojunctions in SnS thin film solar cells. Appl. Phys. Lett. 2013, 103, 181904. DOI: 10.1063/1.4821433 3 T.D. Veal, P.D.C. King, S.A. Hatfield, L.R. Bailey, C.F. McConville, B. Martel, J.C. Moreno, E. Frayssinet, F. Semond, J. Zúñiga-Pérez. Valence band offset of the ZnO/AlN heterojunction determined by x-ray photoemission spectroscopy. Appl. Phys. Lett. 2008, 93, 202108. DOI: 10.1063/1.3032911 4 O. Moshe, D.H. Rich, B. Damilano, J. Massies. Effect of uniaxial stress on the polarization of light emitted from GaN/AlN quantum 6 arXiv:1710.06162v3 physics.app-ph 12 November 2017 18 K. Seeger. Semiconductor Physics 9th edn. Springer, Berlin 2004. p. 340. 19 N. Bachrach-Ashkenasy, L. Kronik, Y. Shapira, Y. Rosenwaks, M.C. Hanna, M. Leibovitch, P. Ram. Surface photovoltage spectroscopy of quantum wells and superlattices. Appl. Phys. Lett. 1996, 68, 879. DOI: 10.1063/1.116217 20 Y. Turkulets, T. Bick, I. Shalish. Double surface effect causes a peak in band-edge photocurrent spectra: a quantitative model. J. Phys. D: Appl. 10.1088/0022- 3727/49/36/365104 365104. DOI: Phys., 2016, 49, 21 O. Ambacher, J. Smart, J.R. Shealy, N.G. Weimann, K. Chu, M. Murphy, W.J. Schaff, L.F. Eastman, R. Dimitrov, L. Wittmer, M. Stutzmann, W. Rieger, J. Hilsenbeck. Two-dimensional electron gases induced by spontaneous and piezoelectric polarization charges in N- and Ga-face AlGaN/GaN heterostructures. J. Appl. Phys. 85, 3222 (1999). DOI: 10.1063/1.369664 22 P. Waltereit, O. Brandt, A. Trampert, H. T. Grahn, J. Menniger, M. Ramsteiner, M. Reiche, K. H. Ploog. Nitride semiconductors free of electrostatic fields for efficient white light-emitting diodes. Nature 406, 865 (2000). 23 O. Ambacher, J. Majewski, C. Miskys, A. Link, M. Hermann, M. Eickhoff, M. Stutzmann, F. Bernardini, V. Florentini, V. Tilak, B. Schaff, L. F. Eastman, Pyroelectric properties of Al(In)GaN/GaN hetero- and quantum well structures, J. Phys. Condens. Matter 14, 3399 (2002). DOI: 10.1088/0953-8984/14/13/302 24 O. Ambacher , V. Cimalla, "Polarization Induced Effects in GaN- based Heterostructures and Novel Sensors", in: C. Wood, D. Jena (Eds), "Polarization Effects in Semiconductors", Springer, New York, 2007, pp. 27-110. 25 U. Mishra, J. Singh. Semiconductor Device Physics and Design. (Springer, Netherlands, 2008), pp. 375-394. 26 J.D. Dow, D. Redfield. Electroabsorption in Semiconductors: The Excitonic Absorption Edge. Phys. Rev. B 1, 3558 (1970). DOI: 10.1103/PhysRevB.1.3358 27 D.A.B. Miller, D.S. Chemla, T.C. Damen, A.C. Gossard, W. Wiegmann, T.H. Wood, C.A. Burrus. Band-Edge Electroabsorption in Quantum Well Structures: The Quantum-Confined Stark Effect. Phys. Rev. Lett. 53, 2173 (1984). DOI: PhysRevLett.53.2173 28 D.A.B. Miller, D.S. Chemla, S. Schmitt-Rink. Relation between electroabsorption in bulk semiconductors and in quantum wells: The quantum-confined Franz-Keldysh effect. Phys. Rev. B 33, 6976 (1986). DOI: PhysRevB.33.6976 29 Smearing of the photocurrent step could still result from other effects, though not the Franz-Keldysh effect. dots grown on Si(111). Phys. Rev. B 2008, 77, 155322. DOI: 10.1103/PhysRevB.77.155322 5 A. Franceschetti, A. Zunger. The inverse band-structure problem of finding an atomic configuration with given electronic properties. Nature 1999, 402, 60. DOI: 10.1038/46995 6 S-H. Wei, A. Zunger. Valence band splittings and band offsets of InN. Appl. Phys. Lett. 1996, 69, 2719. AlN, GaN, and DOI:10.1063/1.117689 7 Kroemer, H. Nobel Lecture: Quasielectric fields and band offsets: Teaching electrons new tricks. Rev. Mod. Phys. 2001, 73, 783. DOI: 10.1103/RevModPhys.73.783 8 F. Capasso. Band-Gap Engineering: From Physics and Materials to New Semiconductor Devices. Science 1987, 235, 172. DOI: 10.1126/science.235.4785.172 9 J. Heber. The staircase to flexibility (An interview with Federico Capasso). Nature Mater. 2010, 9, 374. DOI: 10.1038/nmat2758 10 R. Dingle, H. L. Störmer, A. C. Gossard, and W. Wiegmann. Electron mobilities semiconductor heterojunction superlattices. Appl. Phys. Lett. 1978, 33, 665. DOI: 10.1063/1.90457 in modulationdoped 11 T. Mimura. The Early History of the High Electron Mobility Transistor (HEMT). IEEE Trans. Microw. Theor. Tech. 2002, 50, 3, 780. DOI: 10.1109/22.989961 12 H.W. Hou, Z. Liu, J.H. Teng, T. Palacios, S.J. Chua. High Temperature Terahertz Detectors Realized by a GaN High Electron Mobility Transistor. Sci. Rep. 2017, 7, 46664. DOI: 10.1038/srep46664 13 U.K. Mishra, L. Shen, T.E. Kazior, Y.-F. Wu. GaN-based RF power devices and amplifiers. Proc. IEEE 2008, 96, 287. DOI: 10.1109/JPROC.2007.911060 14 M. Heiblum, M. I. Nathan, M. Eizenberg. Energy band discontinuities internal photoemission. Appl. Phys. Lett. 1985, 47, 503. DOI: 10.1063/1.96107 heterojunctions measured by in 15 S.T. Bradley, S.H. Goss, J. Hwang, W. J. Schaff, L.J. Brillson. Surface cleaning and annealing effects on Ni/AlGaN interface atomic composition and Schottky barrier height. Appl. Phys. Lett. 2004, 85, 1368 (2004). DOI: 10.1063/1.1785287 16 The model that is commonly employed in analyzing internal photoemission data (Fowler's model) is based on physics of above- barrier emission, and, in the case of band-to-band transitions, relates to photon energies that exceed the bandgap energy. In essence, it does not take into account the Franz-Keldysh effect. For this reason, Fowler's model is inadequate for below-bandgap data. Its adequacy for above-bandgap data has also been questioned, because Fawler's method was actually developed for external photoemission from metals. See, e.g., I-S Chen, T.N. Jackson, C.R. Wronski. Characterization of semiconductor heterojunctions using internal photoemission. J. Appl. Phys. 1996, 79, 8470 (1996). DOI: 10.1063/1.36252 17 L. Kronik, Y. Shapira. Surface photovoltage phenomena: theory, experiment, and applications. Surf. Sci. Rep. 1999, 37, 1. DOI: 10.1016/S0167-5729(99)00002-3 7
1907.09258	1	1907	2019-04-13T14:30:48	Effects of external mechanical loading on stress generation during lithiation in Li-ion battery electrodes	[ "physics.app-ph" ]	Li-ion batteries are ineluctably subjected to external mechanical loading or stress gradient. Such stress can be induced in battery electrode during fabrication and under normal operation. In this paper, we develop a model for stresses generated during lithiation in the thin plate electrode considering the effects of external mechanical loading. It is found that diffusion-induced stresses are asymmetrically distributed through the thickness of plate due to the coupling effects of asymmetrically distributed external mechanical stress. At the very early stage during Li-ions insertion, the effects of the external mechanical loading is quite limited and unobvious. With the diffusion time increasing, the external mechanical loading exerts a significant influence on the evolution of stresses generated in the electrode. External compressed electrode is inclined to increase the value of stresses generated during lithiation, while external tensed electrode tends to decrease the value of stresses, and as the diffusion time increases, the effects of the external mechanical loading on the stresses generated during lithiation become more obvious.	physics.app-ph	physics	Effects of external mechanical loading on stress generation during lithiation in Li-ion battery electrodes Department of Mechanical Engineering, Imperial College London, London SW7 2AZ, UK Wenbin Zhou* Abstract Li-ion batteries are ineluctably subjected to external mechanical loading or stress gradient. Such stress can be induced in battery electrode during fabrication and under normal operation. In this paper, we develop a model for stresses generated during lithiation in the thin plate electrode considering the effects of external mechanical loading. It is found that diffusion-induced stresses are asymmetrically distributed through the thickness of plate due to the coupling effects of asymmetrically distributed external mechanical stress. At the very early stage during Li-ions insertion, the effects of the external mechanical loading is quite limited and unobvious. With the diffusion time increasing, the external mechanical loading exerts a significant influence on the evolution of stresses generated in the electrode. External compressed electrode is inclined to increase the value of stresses generated during lithiation, while external tensed electrode tends to decrease the value of stresses, and as the diffusion time increases, the effects of the external mechanical loading on the stresses generated during lithiation become more obvious. Keywords: Lithium-ion battery; Diffusion-induced stress; External mechanical loading; Thin plate electrode *Corresponding author. Email address: [email protected], [email protected] (W. Zhou) 1 1. Introduction Li-ion batteries have been widely used in mini-type electric instruments such as cell phones and other small portable electronic devices due to their high specific capacity, light weight, and no memory effects. However, their further applications such as in the fields of electromotive vehicle, large-scale energy storage and aerospace power supply are limited by the poor cycle of life and the capacity loss of themselves. The main reason is due to the deformation and fracture of the electrode caused by the stress generated in electrodes during cycling [1-4], which can result in electrical disconnects that render electrode active materials incapable of storing Li-ions. One of the critical challenges in Li-ion battery studies is the internal stress field distribution produced in the process of charging and discharging. Much work has been devoted to studying the stresses resulting from cycling of Li-ion batteries due to the intercalation and deintercalation of Li-ions [5-11], however, studies for effects of external mechanical loading on stress generation during lithiation in Li-ion battery electrodes are less. Actually, Li-ion batteries are ineluctably subjected to external mechanical loading or stress gradient during standard usage or storage. Such stress can arise during fabrication of battery materials, which usually involves compression of the electrode to control its porosity [12-14]. Also, such stress can be induced in electrode materials under normal operation, such as batteries operating in high pressure environments, electric/hybrid vehicles or flexible applications [15]. In many cases, the influence of external stresses on the diffusion process and the stress generation during lithiation in Li-ion battery electrodes is so significant that it cannot be neglected. Some researchers have studied the effects of electrode compression on capacity and efficiency of various electrode materials. Novak et al. [16] and Gnanaraj et al. [17] found both the capacity and first cycle efficiency of batteries had a decrease when subjected to high levels of compression. They explained that it was caused by electrode particle fracture for the largest pressures as well as transport limitations within the liquid path. By studying low-loading electrodes prepared with a natural graphite from Superior Graphite and compressed at a range of moderate pressures, Shim et al. found both the reversible capacity and irreversible capacity loss (ICL) had a decrease with increases in pressure [18]. They concluded that the decreased reversible capacity was due to increased stresses generated within the graphite electrode, which also slowed down in Li-ions diffusion process [19]. However, much of these work were conducted using electrochemical characterization. Theoretical studies for the effects of external mechanical loading on stress generation during lithiation in electrodes which are quite necessary for Li-ion batteries have been rare. 2 In this work, we develop a model for stress generation during lithiation in Li-ion battery electrodes with planar geometries coupled with the effects of external mechanical loading. To begin with, two basic approximations for material properties are introduced as follows [9,10]. Isotropic elasticity analysis of infinitesimal deformation is carried out, and large deformation related to plasticity is out of the scope of this research [20]. Mechanical properties and the diffusion coefficient are independent of Li-ion concentration, which means that Young's modulus, Poisson's ratio and the diffusion coefficient are constants. Yang et al. [21] studied the effects of composition-dependent modulus, finite concentration and boundary constraint on Li-ion diffusion and stresses in a bilayer Cu-coated Si nano-anode. This is an important work to show that mechanical properties of electrode materials will change and have an effect on diffusion-induced stresses during the lithiation/delithiation process. Since we focus on studying the effects of external mechanical loading, changes in material properties during the lithiation/delithiation process still need to be further considered by establishing more sophisticated model. 2. Basic Theory 2.1. Diffusion equation We model the insertion and extraction of Li-ions as a diffusion process. The species flux can be defined as [22] (1) where is the mobility of Li-ions, c is the Li-ion concentration and is the chemical potential, which is given by (2) where is the initial chemical potential and is assumed to be a constant. R is a gas constant, is absolute temperature, is the partial molar volume of the Li-ion, is the hydrostatic stress. Since atomic diffusion in solids is much slower than elastic deformation, mechanical equilibrium is established much faster than that of diffusion. Mechanical equilibrium is, therefore, treated as a static equilibrium problem, thus the linear elasticity theory is applicable for the coupling of diffusion-induced stresses and external stresses. The hydrostatic stress can be expressed as (3) 3 JMcM0lnhRTc0Thh,,1()3hdieiixyzwhere is the diffusion-induced stress due to the concentration gradient and denotes the external mechanical loading induced stress, respectively. Substituting Eq. (2) into Eq. (1), the species flux can be expressed as (4) where is the effective diffusion coefficient in a stressed isotropic solid. Substituting Eq. (3) into Eq. (4), can be given by (5) where is the diffusion coefficient in a stress-free isotropic solid. Combining Eq. (4) with the mass conservation equation , we obtain the diffusion equation as follows (6) 2.2. Diffusion-induced stresses Consider an electrode plate of thickness l subjected to a constant uniform charging current density on both of its side faces [2, 23], as shown in Fig. 1. The electrode plate is considered to be an isotropic linear elastic solid and is mainly composed of the active particles. The effect of the electrolyte is neglected though it can be included using a more sophisticated model such as the work conducted by Zhang et al. [24] and Renganathan et al. [25], and the aim of our work is to study the effects of external mechanical loading. Analogous to thermal stresses [26], for a given concentration profile, the nonzero diffusion-induced stress components due to the insertion of solute atoms into host are only and , which are two equal transverse stresses and can be given by (7) where E and  are Young's modulus and Poisson's ratio, respectively. 4 diei(1)heffcJMRTcDcRTceffDeffD,,()(1)13dieiixyzheffccDDDRTcRTcDMRT0cJt22,,,,()()33dieidieiixyzixyzcccDctRTRTdydz30012(2)()()3(1)2(1)lldydzExllcdxcEcxdxllIn general, the thin electrode plate is subjected to a unidirectional gradient stress field due to the mechanical loading combination of bending and tension . As illustrated in Fig. 1, the unidirectional gradient stress field can be expressed as (8) where denotes the stress gradient due to mechanical bending and is the tensile stress. It should be noted that Fig.1 only shows the case of , in fact, can be negative and zero and are also studied. Substituting the diffusion-induced stress Eq. (7) and external loading induced stress Eq. (8) into Eq. (6) yields where . Hence, the effective diffusion coefficient in Eq. (5) becomes (9) (10) where . It is a positive constant and reflects the degree of the effects of both the concentration gradient and external stress gradient on diffusion flux. Two operations reflect the insertion and extraction of Li-ions: surface Li-ion concentration keeps the maximum concentration and surface Li-ion flux remains a constant flux [7]. In this work, we assume that the surface Li-ion flux is uniform, namely galvanostatic operation, and the initial and boundary conditions are given by (11) where are the Li-ion flux at the bottom and top surface, respectively. To perform numerical calculations conveniently, the following dimensionless variables for thickness , concentration , time , parameter , current density , stress ,and external mechanical stress gradient , are also introduced: (12) Finite difference method is adopted here to solve the non-linear Eq. (9) combined with definite conditions in Eq. 5 zMyF000,exezeypxa0p0a00p0p2203012(1)()()23lpclDccccxcdxctRTl229(1)ERTeffD030121()(1)23leffcpclDDcxcdxDcRTclc030121()23lplxcdxRTclc0(,0),0,0,effbefftcxctDcJxDcJxl,bJtJxctjPmax02maxmaxmax3(1),,,,,,3xctDJllxctcjPplcDcEcRTl(11), where the time derivative and space derivative of the Li-ion concentration both need to be discretized. The time derivative and space derivative are calculated by the forward difference and central difference, respectively. The Simpson's rule is used to calculate the integral. The dimensionless time increment is set as . The dimensionless thickness of the plate is divided into 200 equal parts, i.e. . Once the diffusion equation is solved, Li-ion concentration is substituted into Eq. (7) for stresses generated in the electrode. 3. Results and Discussion We take the Mn2O4 system as an example, whose material properties are listed in Table 1 [10]. For simplicity, the surface flux of Li-ions at the bottom surface is assumed to be identical with that of the top surface . As an example, we take the dimensionless current density at the bottom and top surface as , and . We state that different fluxes could influence the stress profiles because of different Li-ion concentrations at the two surfaces. Eq. (11) can be applied to the Li-ion diffusion during charging and discharging: , , and for the process of Li-ions insertion, while , , and for Li-ions deinsertion. Fig. 2 shows the results of insertion in the absence of external mechanical stress gradient: (a) normalized concentration profile through the thickness of the plate, (b) normalized stress profile at different locations and (c) normalized stress over the surface at different charging times. At the early period during Li-ions insertion in Fig. 2(a), Li-ions transfer across the two surfaces first and gradually move towards the centre, showing that Li- ion concentration is much higher around the surface compared with the region close to the centre. Therefore, surface region experiences larger volume changes, which causes tensile stress in the centre and compressive stress in the two surface regions as shown in Fig. 2(b). Also, both the concentration and stress are symmetrically distributed through the thickness of the plate in the absence of external mechanical stress gradient, and the maximum stress is always located at the entry surface for a given time. Fig. 2(c) illustrates the evolution of normalized stress at the entry surface versus time, we can see that the magnitude of this stress increases to its peak with increasing time and then gradually decreases as the charging process proceeds. This is because stress is relevant to the concentration profile of Li-ions inserted into the plate, at the initial charging stage, the concentration profile is nonuniform, as more and more Li-ions insert into the plate, the concentration profile becomes more and more uniform, resulting in gradually decreasing stress. 6 =0.001t=0.005xbJtJmax/0.1bJlDcmax0.4c00cbbsJJttsJJ0maxccbbsJJttsJJTo study the influence of external mechanical stress, we set five different : 0, -0.5, -1, 0.5 and 1. is the dimensionless form of , which is given by Eq. (8) and denotes the stress gradient. Here the negative stress gradient means applying external compressive loading to the electrode, while positive means applying external tensile loading. Fig. 3 shows the results of insertion with different negative external mechanical stress gradients: (a) normalized concentration profile at the stage of dimensionless time , (b) normalized stress profile at the early stage of dimensionless time during Li- ions insertion and (c) normalized stress profile at the later stage of dimensionless time during Li- ions insertion. From Figs. 3(b) and 3(c), we can see that stress profile through the thickness of the plate is asymmetric due to the coupling effects of asymmetrically distributed external mechanical stress. At the early stage during Li-ions insertion ( ), the effects of the external mechanical loading is quite limited and unobvious. With the charging time increasing, the external stress gradient exerts a significant influence on the evolution of stresses. Such an influence will increase with increasing external stress gradient, and larger external negative stress gradient tends to increase the value of stresses generated in the electrode compared with that of lower external negative stress gradient as time increases. This result is in accordance with the previous experiments [16-18] that compressed electrode has a decrease in capacity. It also implies that when the external negative stress gradient is applied and becomes larger, the charging at the entry surface will be slower. The external negative stress gradient will impede the diffusion of solute atoms in this case. Consequently, the value of the concentration decreases and thus leads to more unevenly distributed concentration, as shown in Fig. 3(a). The results of insertion with different positive stress gradients are illustrated in Fig. 4: (a) normalized concentration profile at the period of dimensionless time , (b) normalized stress profile at the early period of dimensionless time during Li-ions insertion and (c) normalized stress profile at the later period of dimensionless time during Li-ions insertion. In contrast to Fig. 3, here external tensile loading is applied to the electrode. As expected, the external positive stress gradient will accelerate the solute penetration and thus leads to more uniform concentration profile, seen in Fig. 4(a). Therefore, the external tensile stress can be employed to modify diffusion barriers and help to decrease strains/stresses originating from unevenly distributed concentration. Yen et al. found tensile mechanical stress will enlarge atom spacing of silicon and thus enhance the oxidation rate [27]. Moreover, Sanchez et al. showed that an external tensile stress of 2GPa decreased diffusion barriers by about 9%, improving diffusion rates by about 30% at room temperature [28]. As shown in Figs. 4(b) and 4(c), the external positive stress gradient tends to decrease the value of stresses generated in the electrode, and as the charging time increases, the effects of external stress gradient become more obvious. The deformation and fracture of the electrode caused by the excessive internal stress during cycling can result in electrical disconnects, which renders electrode active material incapable of 7 P03lPpRT0pPP2/0.16Dtl2/0.032Dtl2/0.07Dtl2/0.05Dtl2/0.12Dtl2/0.032Dtl2/0.07Dtlstoring lithium-ion and lows the utilization of the active material. From Figs. 3(c) and 4(c), at the same diffusion time during Li-ions insertion, the maximum tensile stress ( =0.03) of electrode subjected to positive external stress gradient P=1 are reduced by as much as 86% compared to electrode subjected to negative external stress gradient P=-1 ( =0.22). Therefore, tensed electrode is superior in fracture resistance, and thus the utilization of the active material should be significantly increased because of the decrease of the maximum tensile stresses generated during cycling. By the results mentioned above, diffusion-induced stresses can be tailored by the external stress gradient in order to be kept below material strengths and avoid mechanical fracture. To quantify the effects of the non-uniform distribution of the stresses on the utilization of the active material, more data such as the critical fracture strength of electrode materials and fracture energy are needed from a range of further experiments. 4. Conclusions In summary, we develop a model for diffusion-induced stresses of the thin plate electrode and consider the effects of external mechanical loading. The results show that stress profile through the thickness of the plate is asymmetric due to the coupling effects of asymmetrically distributed external mechanical stress. At the early stage during Li-ions insertion, no significant change is observed for different external stress gradients. With the increase of the diffusion time, the external stress gradient exerts significant effects on the evolution of stresses. Such effects become more obvious with increasing the external stress gradient, larger external negative stress gradient leads to greater value of stresses as time increases, while larger external positive stress gradient tends to decrease the value of stresses generated in the electrode, and the effects become more obvious as time increases. References [1] K. Zhao, M. Pharr, J. J. Vlassak, Z. Suo, Fracture of electrodes in lithium-ion batteries caused by fast charging, J. Appl. Phys., 108 (2010) 073517. [2] T. K. Bhandakkar, H. Gao, Cohesive modeling of crack nucleation under diffusion induced stresses in a thin strip: Implications on the critical size for flaw tolerant battery electrodes, Int. J. Solids Struct., 47 (2010) 1424. [3] X. Xiao, P. Liu, M. W. Verbrugge, H. Haftbaradaran, H. Gao, Improved cycling stability of silicon thin film electrodes through patterning for high energy density lithium batteries, J. Power Sources, 196 (2011) 1409. [4] H. Haftbaradaran, X. Xiao, M.W. Verbrugge, H. Gao, Method to deduce the critical size for interfacial delamination of patterned electrode structures and application to lithiation of thin-film silicon islands, J. Power Sources, 206 (2012) 357. [5] J. Christensen, J. Newman, J. Electrochem. Soc., A mathematical model of stress generation and fracture in lithium manganese oxide, 153 (2006) A1019. 8 2/0.07Dtl[6] Y.-T. Cheng, M. W. Verbrugge, J. Appl. Phys., The influence of surface mechanics on diffusion induced stresses within spherical nanoparticles, 104 (2008) 083521. [7] Y.-T. Cheng, M. W. Verbrugge, Evolution of stress within a spherical insertion electrode particle under potentiostatic and galvanostatic operation, J. Power Sources, 190 (2009) 453. [8] Z. Cui, F. Gao, J. Qu, J. Mech. Phys. Solids, A finite deformation stress-dependent chemical potential and its applications to lithium ion batteries, 60 (2012) 1280. [9] F. Hao, X. Gao, D. Fang, Diffusion-induced stresses of electrode nanomaterials in lithium-ion battery: The effects of surface stress, J. Appl. Phys., 112 (2012) 103507. [10] F. Hao, D. Fang, Tailoring diffusion-induced stresses of core-shell nanotube electrodes in lithium-ion batteries, J. Appl. Phys., 113 (2013) 013507. [11] W. Zhou, F. Hao, D. Fang, The effects of elastic stiffening on the evolution of the stress field within a spherical electrode particle of lithium-ion batteries, Int. J. Appl. Mech., 05 (2013) 1350040. [12] C.-W. Wang, Y.-B. Yi, A. M. Sastry, J. Shim, K. A. Striebel, Particle compression and conductivity in Li-ion anodes with graphite additives, J. Electrochem. Soc., 151 (2004) A1489. [13] K. A. Striebel, A. Sierra, J. Shim, C.-W. Wang, A. M. Sastry, The effect of compression on natural graphite anode performance and matrix conductivity, J. Power Sources, 134 (2004) 241. [14] Y.-B. Yi, C.-W. Wang, A. M. Sastry, Compression of packed particulate systems: Simulations and experiments in graphitic Li-ion anodes, J. Eng. Mater-T ASME, 128 (2006) 73. [15] W. Van Schalkwijk, B. Scrosati, Advances in Lithium-Ion Batteries, Kluwer Academic Publishers, Norwell, 2002. [16] P. Novak, W. Scheifele, M. Winter, O. Haas, Graphite electrodes with tailored porosity for rechargeable ion-transfer batteries, J. Power Sources, 68 (1997) 267. [17] J. S. Gnanaraj, Y. S. Cohen, M. D. Levi, D. Aurbach, The effect of pressure on the electroanalytical response of graphite anodes and LiCoO2 cathodes for Li-ion batteries, J. Electroanal. Chem., 516 (2001) 89. [18] J. Shim, K. A. Striebel, Effect of electrode density on cycle performance and irreversible capacity loss for natural graphite anode in lithium-ion batteries, J. Power Sources, 119-121 (2003) 934. [19] J. Shim, K. A. Striebel, The dependence of natural graphite anode performance on electrode density, J. Power Sources, 130 (2004) 247. [20] K. Zhao, M. Pharr, S. Cai, J. J. Vlassak, Z. Suo, Large plastic deformation in high-capacity lithium-ion batteries caused by charge and discharge, J. Am. Ceram. Soc., 94 (2011) S226. [21] B. Yang, Y.-P. He, J. Irsa, C.A. Lundgren, J.B. Ratchford, Y.-P. Zhao, Effects of composition-dependent modulus, finite concentration and boundary constraint on Li-ion diffusion and stresses in a bilayer Cu-coated Si nano-anode, J. Power Sources, 204 (2012) 168. [22] J. C. M. Li, Chemical potential for diffusion in a stressed solid, Scr. Metall., 15 (1981) 21. 9 [23] H. Haftbaradaran, H. Gao, W.A. Curtin, A surface locking instability for atomic intercalation into a solid electrode," 2010, Applied Physics Letters, Appl. Phys. Lett., 96 (2010) 091909. [24] J. Zhang, B. Lu, Y. Song, X. Ji, Diffusion induced stress in layered Li-ion battery electrode plates, J. Power Sources, 209 (2012) 220. [25] S. Renganathan, G. Sikha, S. Santhanagopalan, R.-E. White, Theoretical analysis of stresses in a lithium-ion cell, J. Electrochem. Soc., 157 (2010) A155. [26] S. Prussin, Generation and distribution of dislocations by solute diffusion, J. Appl. Phys., 32 (1961) 1876. [27] J. Y. Yen, J. G. Hwu, Enhancement of silicon oxidation rate due to tensile mechanical stress, Appl. Phys. Lett., 76 (2000) 1834. [28] J. Sanchez, J. Fullea, C. Andrade, P. L. Andres, Hydrogen in α-iron: Stress and diffusion, Phys. Rev. B, 78 (2008) 014113. 10 Tables Table 1 Material parameters of Mn2O4 Name Diffusion coefficient Young's modulus Poisson's ratio Partial molar volume Max Li-ion concentration Symbol D/m2 s-1 E/GPa  /dm3 mol-1 cmax/mol dm-3 Value 7.0810-15 10 0.3 3.49710-3 22.9 11 Figures and Figure Captions Fig.1 Sketch of a plate electrode under galvanostatic charging, namely a uniform current density on surfaces. 12 Fig.2 (a) Evolution of the normalized concentration profile, (b) Evolution of the normalized stress profile and (c) Evolution of the normalized stress over the surface. 13 Fig.3 The effects of the negative stress gradient on: (a) concentration profile at the diffusion time , (b) stress profile at and (c) stress profile at . 14 2/0.16Dtl2/0.032Dtl2/0.07DtlFig.4 The effects of the positive stress gradient on: (a) concentration profile at the diffusion time , (b) stress profile at and (c) stress profile at . 15 2/0.12Dtl2/0.032Dtl2/0.07Dtl
1707.06287	1	1707	2017-07-19T20:37:22	Elastic Wave Eigenmode Solver for Acoustic Waveguides	[ "physics.app-ph" ]	A numerical solver for the elastic wave eigenmodes in acoustic waveguides of inhomogeneous cross-section is presented. Operating under the assumptions of linear, isotropic materials, it utilizes a finite-difference method on a staggered grid to solve for the acoustic eigenmodes of the vector-field elastic wave equation. Free, fixed, symmetry, and anti-symmetry boundary conditions are implemented, enabling efficient simulation of acoustic structures with geometrical symmetries and terminations. Perfectly matched layers are also implemented, allowing for the simulation of radiative (leaky) modes. The method is analogous to eigenmode solvers ubiquitously employed in electromagnetics to find waveguide modes, and enables design of acoustic waveguides as well as seamless integration with electromagnetic solvers for optomechanical device design. The accuracy of the solver is demonstrated by calculating eigenfrequencies and mode shapes for common acoustic modes in several simple geometries and comparing the results to analytical solutions where available or to numerical solvers based on more computationally expensive methods.	physics.app-ph	physics	Elastic Wave Eigenmode Solver for Acoustic Waveguides Nathan Dostart,1 Yangyang Liu,1 and Milos A. Popovi´c2, a) 1)Department of Electrical, Computer, and Energy Engineering, University of Colorado, Boulder, CO 80309, USA 2)Department of Electrical and Computer Engineering, Boston University, Boston, MA 02215, USA (Dated: 21 July 2017) A numerical solver for the elastic wave eigenmodes in acoustic waveguides of inhomogeneous cross-section is presented. Operating under the assumptions of linear, isotropic materials, it utilizes a finite-difference method on a staggered grid to solve for the acoustic eigenmodes of the vector-field elastic wave equation. Free, fixed, symmetry, and anti-symmetry boundary conditions are implemented, enabling efficient simulation of acoustic structures with geometrical symmetries and terminations. Perfectly matched layers are also implemented, allowing for the simulation of radiative (leaky) modes. The method is analogous to eigenmode solvers ubiquitously employed in electromagnetics to find waveguide modes, and enables design of acoustic waveguides as well as seamless integration with electromagnetic solvers for optomechanical device design. The accuracy of the solver is demonstrated by calculating eigenfrequencies and mode shapes for common acoustic modes in several simple geometries and comparing the results to analytical solutions where available or to numerical solvers based on more computationally expensive methods. I. INTRODUCTION Recent advances in several fields have attracted growing interest in the design of chip-scale acoustic devices that can interface with electrical and optical integrated components. Optomechanics is a prime example, where interacting acoustic and optical fields enable novel functionalities, such as ultra-sensitive quantum measurements1,2, narrow- linewidth lasers3,4, optomechanical memory5,6, non-reciprocity and optical diodes7–9, optical cooling10–12, phononic topological insulators13,14, optical amplifiers15, improved gravity wave detection16,17, microwave filters18, and quantum state transfer19–21. The field of RF micro-electromechanical systems is another important example where electroa- coustic transduction of bulk and surface acoustic waves in acoustic resonators enables some devices which outperform conventional RF electronics. These include reconfigurable filters22, narrowband signal filtering23, and high quality fac- tor (Q) resonators24. These devices are also being integrated into microelectronic systems for improved performance25. For all these acoustic wave based devices, good performance requires confining the acoustic energy to a small cross- sectional area (waveguides) or volume (resonators), phase-matching the acoustic wave to transducer arrays and/or optical waves, and optimizing transduction efficiency. Numerical tools for designing and simulating acoustic waveguide modes are thus necessary to enable efficient device designs, intricate nanoscale coupling schemes, and novel device architectures. Previous work has predominantly focused on full-wave simulation of 2D and 3D domains due to the importance of these problems in geophysics. 3D solvers have been developed for anisotropic, heterogeneous domains using both finite- difference26 and finite-element27 methods and are currently the predominant method for designing acoustic devices in the GHz frequency range. While many commercial software tools28 allow the design of acoustic waveguides using a full 3D solver, sometimes more efficiently by reducing the volume using Floquet (periodic) boundary conditions, the most efficient approach is the maximally reduced, 2D problem formulation (disregarding cross-section symmetries). The 2D formulation returns an orthogonal, complete set of modes (field and frequency) at a specified propagation constant. In contrast, the solution of a 3D volume returns all resonant modes including those in higher order Brillouin zones, which are an artifact of the 3D formulation and usually undesired. This rigorous reduction of a 3D problem to a 2D simulation domain is referred to as a 2+1D simulation. Papers utilizing a 2+1D version of the finite-element method have been implemented to solve free, isotropic waveguide geometries in the ultrasound regime29 and extended to axially symmetric waveguides30, embedded waveguides31, and viscoelastic materials32. It should be noted that generic FEM tools such as COMSOL, which provide an interface to solve arbitrary partial differential equations, are in principle capable of solving the 2+1D Cartesian acoustic waveguide simulation (by entering custom equations). In this work, we demonstrate an acoustic waveguide mode solver based on the finite-difference method, analogous to electromagnetic (EM) mode solvers, which solves the linear isotropic elastic wave equation. The mode solver combines 7 1 0 2 l u J 9 1 ] h p - p p a . s c i s y h p [ 1 v 7 8 2 6 0 . 7 0 7 1 : v i X r a a)[email protected] 2 all capabilities of FEM mode solvers with the efficiency of finite-differencing, all physical boundary conditions as well as perfectly matched layers, directly overlaps with the optical Yee grid33, and is verified for acoustic frequencies ranging from 1 MHz to 10 GHz. We utilize a staggered-grid discretization that preserves second-order accuracy for all physical quantities of interest, by analogy with the Yee scheme first used in EM33 and identical to the staggered grid of34,35. The solver finds the acoustic eigenmodes of a structure that is invariant along one Cartesian dimension. The source-free vectorial elastic wave equation in an inhomogeneous, isotropic, linear medium is formulated as an eigenvalue equation. Given a specific material configuration and the propagation constant, a unique set of eigenmodes with corresponding modal frequencies can be found. The acoustic problem is only a linear eigenvalue problem when formulated such that the propagation constant β is given and the eigenfrequency ω is the eigenvalue. This is notably different than the analogous electromagnetic problem which, due to Gauss's Law, can be formulated as a linear eigenvalue problem with either ω or β given and the other variable as the eigenvalue. A finite-difference method is used to sample the inhomogeneous medium and acoustic field over the computational domain, the cross-section of the structure. This operation transforms the continuous eigenvalue problem into a sparse matrix which can be solved by standard numerical sparse matrix eigen solver methods (e.g. shifted inverse method via Arnoldi iteration and a sparse linear solver). An example case is the 'eigs' function in MATLAB36. The solver calculates, to second order accuracy on the finite-difference grid, the acoustic eigenmodes of a straight acoustic waveguide with specified cross-section and propagation constant. The use of 2+1D mode solvers in both optical and acoustic domains allows for accurate and efficient calculation of propagation parameters and coupling terms, which can be input to the acousto- optic simulations for accuracy comparable to full 3D simulations. To this end, we have made our mode solver code, in MATLAB implementation, freely available37. We present the theoretical and mathematical basis for the elastic wave equation eigenmode decomposition in Sec. II. This section also includes implementation details such as the finite-differencing scheme and boundary conditions. We then implement the method in MATLAB and validate its accuracy by analysis of the matrix construction, convergence tests, example cross-sections with analytical solutions, and other examples which can be solved numerically in 3D using a commercial FEM solver28 in Sec. III. II. THEORY, MATHEMATICAL FRAMEWORK, AND IMPLEMENTATION OF THE MODE SOLVER The mathematical basis for the eigenmode solver and its implementation are described in this section. The source- free, linear, isotropic elastic wave equation is cast as an eigenmode problem with frequency eigenvalues and discretized. The only assumptions made are that the cross-section is invariant along one linear direction, z; all materials are linear, isotropic, and time-invariant; and that a computational domain that is finite in cross-section represents well the modes of the structure (justified for 'confined' modes, such as in electromagnetics38). The isotropic assumption is made for convenience and is not essential; the method can be applied to anisotropic media. This approach to solving for waveguide modes is analogous to electromagnetic mode solvers, where instead of an electric or magnetic field we solve for the elastic displacement field. Notably, in the elastic equation, there is no equivalent of Gauss's Law and thus we solve for three field components and have three polarization families. In EM, Gauss's Law reduces the eigen problem to two polarization mode families and specification of two field components fully defines the mode (e.g. Ex, Ey). A. Derivation of Isotropic Linear Elastic Wave Equation We begin with Newton's 2nd law written in a density formulation, the strain-displacement relation, and generalized Hooke's law t u = ∇ · ¯¯σ ρ∂2 ¯¯ = ∇su (1) (2) (3) where ρ(r, t) ≡ ρ(r) is the spatial distribution of material density, u(r, t) is the elastic displacement field, ∇· is the tensor divergence, ¯¯σ(r, t) is the stress tensor, ¯¯(r, t) is the strain tensor, ∇s is the symmetric spatial vector gradient, C(r, t) ≡ C(r) is the fourth order stiffness tensor, and : denotes a tensor inner product. Note that while stress and strain are second order tensors, due to symmetry considerations they can be unwrapped as six-vectors following the Voigt notation (¯¯σ = [σxx σyy σzz σyz σxz σxy]T )39, which is the form used in this paper. The strain tensor has ¯¯σ = C : ¯¯ 3 an analogous form. The operator ∇s, which acts on a vector to give a rank 2 tensor, is the adjoint of the tensor divergence operator (−∇·). C is a symmetric operator (even in the presence of loss) due to our assumption of linear, time-invariant materials39, and we use the notation ∂i to refer to the partial derivative ∂/∂i. Substituting Eqs. (2)-(3) into Eq. (1) in order to factor out the stress tensor yields the linear elastic wave equation written in terms of the displacement field t u = ∇ · C : ∇su. We define a weighted displacement field u ≡ √ ρu. This allows Eq. (4), a generalized eigenvalue problem, to be recast as an ordinary eigenvalue problem with a Hermitian operator in the absence of loss. Invariance of ρ and C with time means that the system has a spectrum defined by a linear eigenvalue problem by setting ∂t → jω. The elastic wave equation can be written as ρ∂2 (4) ω2 u = −1√ ρ ∇ · C : ∇s 1√ ρ u. (5) This has the form of an eigenvalue equation (Λx = ¯¯Hx), so we identify the eigenvalue as Λ ≡ ω2, the eigenvector as x ≡ u, and the symmetrized elastic resonance operator ¯¯H as ∇ · C : ∇s −1√ ¯¯H = 1√ ρ . (6) To clarify this expression, the modified differential operators (in the form that operate on the reduced, six-vector notation used for ¯¯σ and ¯¯) can be written in matrix form  ∂x 0 0 0 ∂y 0 0 0 ∂z 0 ∂z ∂y ∂z 0 ∂x ∂y ∂x 0 ρ ∇s =  ∂x 0 ∇· = 0 0 ∂z ∂y 0 ∂y 0 ∂z 0 ∂x 0 0 ∂z ∂y ∂x 0  . (7) (8) To this point, we have defined the acoustic resonator problem (in 3D). Next, when solving for eigenmodes of a structure with z-invariant geometry, by Fourier transformation along z the z-directed derivative becomes the prop- agation constant (∂z = −jβ). This reduces the problem to one on the cross-sectional plane and ensures that only modes with the specified propagation constant will be returned by the solver. Next, making the isotropic assumption, the stiffness tensor (in the form that operates on the six-vector notation) can be reduced to C =  λ + 2µ λ λ + 2µ λ 0 0 0 λ λ 0 0 0 0 0 0 λ + 2µ 0 0 0 µ 0 0 0 µ 0 0 0 µ 0 0 0 λ λ 0 0 0  (9) where λ(r, t) ≡ λ(r) and µ(r, t) ≡ µ(r) are the first and second Lam´e parameters. B. Discretization Scheme In order to solve the eigenvalue problem in an arbitrary cross-section geometry (ρ, C) numerically, Eq. (5) is dis- cretized to arrive at a form with a finite number of degrees of freedom, and is cast as a matrix eigenvalue problem. An appropriate 3D grid is first formulated which accurately captures the physics to second-order accuracy in the discretiza- tion and the grid is then collapsed to the 2D simulation domain. The collapse is performed such that the cross-sectional locations of all grid points are unchanged while the z-coordinates of all grid points are set to a single value. The z- derivatives, which would be performed between two points in the 3D case, amount to multiplying a single grid point by 4 FIG. 1. (a) Unit cell of the 3D discretization grid used in this paper, where the unit cell coordinates (i, j, k) define the corner of the unit cell and are co-located with the principal stresses. Each point is denoted with the quantities which are sampled at that point.(b) Discrete 'material cube' which provides the basis for an intuitive choice of discretization grid. The 'material cube' is offset from the unit cell by a half-step along each dimension. (c) 2D unit cell which can be obtained by collapsing the 3D unit cell along the z-dimension. (d) Representative implementation of the boundary conditions, including choice of location (red line) and values sampled on the boundary (or removed). (e) 2D discretization grid with schematic depiction of component grid locations. −jβ in the 2D case. More formally, for a 3D grid with z-invariance, ∂z → −jβsin(β∆z/2)/(β∆z/2) exp(−jβ∆z/2)40, but as ∆z → 0 then ∂z → −jβ. The discretization grid used is depicted in Fig. 1. This grid is based on physical intuition from solid mechanics: the state of an element cube is primarily described by the principal stresses (σxx, σyy, σzz), which we define to reside at the center of the cube. The principal stresses lead to the deformation of the faces of the cube, such that the center of each face is the sampling location of the associated normal displacement. If grid point (i, j, k) is associated with the principal stresses (and the center of the cubic volume elements), the corresponding displacements are ux : (i + 1/2, j, k), uy : (i, j + 1/2, k), and uz : (i, j, k + 1/2). An appropriate grid for the shear stresses is found to be a further half-step offset, such that shear stresses are located at σxy : (i + 1/2, j + 1/2, k), σxz : (i + 1/2, j, k + 1/2), and σyz : (i, j + 1/2, k + 1/2). The strain grid is co-located with the stress grid. The finite-difference operators that approximate the spatial partial derivatives transform quantities from the stress/strain grid to the displacement grid and back, as expected and desired. Because isotropic materials are assumed, the material operator C does not induce a change of grid coordinates (the strain at a specific grid point is only related to the stresses at the same grid point). This method can also be extended to anisotropic materials at the cost of additional complexity in the C matrix, which must then have different coefficients sampled on different grids. The staggered-grid scheme used here is analogous to the Yee grid33, commonly used in EM solvers, which preserves second-order accuracy for all fields due to centered differencing (when all materials/coefficients vary spatially on the scale of the discretization). Our discretization grid is slightly more complicated in that the elastic displacement field and shear stress directly replicate a Yee grid, while the principal stresses occupy an additional position at the center of each 'cube'. C. Finite Difference Operators Centered differences in 2nd order differential equations can be formed by appropriate combinations of forward and backward differences on a staggered grid40. Denoting forward differences as ∂i and backward differences as ∂i, the differential operators can be rewritten in terms of forward and backward differences (where the propagation constant (a)(b)3D Unit CellMaterial Cube(c)2D Grid2D Unit Cell(e)(d)Boundary ConditionsNo BoundaryFixed BoundaryFree BoundarySymmetry BoundaryAnti-Symmetry Boundaryhas been substituted for ∂z) as ∇· = ∇s =   ∂x 0 0 0  ∂x 0 0 ∂y 0 0 0 −jβ 0 0 −jβ ∂y −jβ ∂x 0 ∂x ∂y 0 0 −jβ ∂y ∂x 0 0 0 −jβ ∂y 0 ∂x ∂y 0 −jβ 5 (10) (11)  . It should be noted that, because ∂i † = − ∂i, ∇s and −∇· remain adjoints in discrete form (∇† s = −∇·). D. Matrix Operator Construction (cid:104) (cid:105)T (12) Referring to Eq. (6), the resonance operator can be rewritten as a matrix. This matrix can be found by discretizing the displacement field and unwrapping the field into a single column vector. A convenient ordering was found as u =(cid:2){ux} ,{uy} ,{uz}(cid:3)T {ux} = u(0,0) x u(1,0) x (13) Thus, each vector component is unwrapped in the x − y plane along the x-dimension first and the components are then concatenated. This results in a vector of approximate length 3nxny from an nx × ny × 3 matrix. A similar method is used to unwrap the stress tensor in a vector of approximate length 6nxny of the form x x . u(0,1) x u(1,1) x ··· u(nux x ,1) ··· u x ,nux y ) (nux x ··· u(nux x ,0) ¯¯σ =(cid:2){σxx} {σyy} {σzz} {σyz} {σxz} {σxy}(cid:3)T (14) with the strain tensor having an equivalent form. Note that the different components σij and ui are of different lengths since they are sampled at different locations. Shown in Table I are the number of grid points in both the x- and ncomp y-directions (ncomp . ) where the length of the vectorized form of each component is simply ncomp , ncomp vec = ncomp x y x y Component ux uy nx uz σxx σyy σzz nx nx nx nx σyz nx nx + 1 ny ny + 1 ny ny ny ny ny + 1 x ncomp ncomp y σxz σxy nx + 1 nx + 1 ny + 1 ny TABLE I. Grid points of each elastic wave component. The differential operator hereby becomes a matrix operator where each entry in Eqs. (9), (10), (11) is now a block matrix which relates one component to another. After populating the ∇·, C, and ∇s matrices with the block matrices, they are multiplied together to give  ∂xc11 ¯¯H = −1√ ρ ∂x − µβ2 + ∂yµ ∂y ∂yλ ∂x + ∂xµ ∂y −jβλ ∂x − jβ ∂xµ ∂yc11 ∂xλ ∂y + ∂yµ ∂x ∂y − µβ2 + ∂xµ ∂x −jβλ ∂y − jβ ∂yµ −jβ ∂xλ − jβµ ∂x −jβ ∂yλ − jβµ ∂y −β2c11 + ∂yµ ∂y + ∂xµ ∂x  1√ ρ (15) 39. where we have shortened λ + 2µ to the Voigt notation c11 We here note that the matrix, currently in a Hermitian form in the absence of loss, can be cast as a real symmetric matrix ¯¯H(cid:48) (a special case of Hermitian matrices) by defining a modified eigenvector form u(cid:48) ≡ [ux, uy, −j uz], i.e. we expect that the uz component will be in quadrature with the ux, uy components. Hermitian matrices have real eigenvalues (energy conservation) and their eigenmodes form a complete, orthogonal set. Casting the operator matrix into a symmetric form, which has eigenvectors that can be chosen to be entirely real, additionally implies that uz is guaranteed to be in quadrature with ux and uy. If loss is present, then this modified matrix form will be complex symmetric rather than real symmetric and we would expect to obtain complex eigenvalues and non-orthogonal eigenmodes. Defining the modified operator matrix ¯¯H(cid:48) such that ω2 u(cid:48) = ¯¯H(cid:48) u(cid:48), we can calculate this modified operator (cid:90)(cid:90) ρu∗ i · ujdA = uiujδij (cid:19) 1 (cid:18)(cid:90)(cid:90) i · uidA ρu∗ . 2 ui = (17) (18) 6 matrix by recognizing that u(cid:48) = ¯¯Ru and ¯¯H(cid:48) = ¯¯R ¯¯H ¯¯R−1 where ¯¯R is a diagonal matrix with diag( ¯¯R) = [1, 1, −j]. We can then write the modified operator matrix as ∂x − µβ2 + ∂yµ ∂y ∂yλ ∂x + ∂xµ ∂y βλ ∂x + β ∂xµ ∂xλ ∂y + ∂yµ ∂x ∂y − µβ2 + ∂xµ ∂x βλ ∂y + β ∂yµ −β ∂xλ − βµ ∂x −β ∂yλ − βµ ∂y −β2c11 + ∂yµ ∂y + ∂xµ ∂x  ∂xc11  1√ ¯¯H(cid:48) = −1√ ∂yc11 (16) ρ . ρ We remind the reader here that, even for complex finite difference operators, ∂T i = − ∂i so that this modified matrix ¯¯H(cid:48) is symmetric. It is additionally real if both material parameters (λ, µ) and finite difference operators ( ∂i, ∂i) are real, corresponding to no material loss and no complex coordinate stretching radiation (absorbing boundaries), respectively. E. Modal Orthogonality The orthogonality condition for vectors (in the sense of 1D tensors) is x i xi)1/2. These two equations correspond to (cid:82)(cid:82) xi(rT )∗ · xj(rT )dA = xixjδij and xi = ((cid:82)(cid:82) xi(rT )∗ · xi(rT )dA)1/2 when x represents a 3-vector field in a 2D domain. Noting that that u(cid:48)∗ is defined as xi ≡ (x can then replace x → u and write the orthogonality condition and normalization as † i xj ≡ xi2δij where the vector norm i · uj, we j = ρu∗ · u(cid:48) † i to obtain measures of the kinetic energy(cid:0)EK = ρv2/2(cid:1). Alternatively, because the potential energy(cid:0)V = ¯¯†C¯¯/2(cid:1) Noting that the particle velocity field is vi ≡ ∂tui = jωiui, we can substitute ui = −jvi/ωi into Eqs. (17)-(18) is equivalent to the kinetic energy when averaged39, this gives the orthogonality condition that would have been obtained if the wave equation were formulated with ¯¯ as the free variable. This is analogous to the electromagnetic case, where resonator problems use an energy formulation to determine modal orthogonality41. F. Boundary Conditions Four boundary conditions are implemented: fixed boundary (uBound = 0), free boundary (¯¯σ · n = 0), symmetry boundary (∂nun = 0, utBound = 0), and anti-symmetry boundary (unBound = 0, ∂nut = 0). Perfectly matched layers (PMLs), radiation-absorbing regions which permit simulation of radiating structures42, are also implemented. For all boundary conditions, we locate the boundary such that the boundary-normal displacements (un), corresponding out-of-plane shear stresses (σnz), and in-plane shear stresses (σxy) lie on the boundary. The choice of boundary location is depicted in Fig. 1(d) as the location of the red line. We additionally allow for vacuum (no material) in the simulation domain. The fixed boundary condition requires the displacement field be set to zero on the boundary. An efficient and computationally stable implementation is to remove the corresponding boundary-normal displacements un (since they are the only ones coincident with the boundary) from the matrix operator and solution vector, which is the method used here. We validate this boundary condition in Sec. III C by confirming that a beam with fixed boundary conditions returns the correct eigenmode shapes and frequencies (Fig. 4). For the free surface boundary condition, the boundary-normal stress components must go to zero at the boundary. In a similar manner to the implementation of the fixed boundary condition, we remove the shear stresses σnz, σxy on the boundary. This boundary condition is also validated in Sec. III C by confirming that the eigenmode shapes and frequencies are accurately calculated (Fig. 4). Symmetry and anti-symmetry boundary conditions are hybrids of free and fixed boundary conditions. Similar to the case in electromagnetics, symmetry of the normal displacement corresponds to anti-symmetry of the transverse dis- placement, while anti-symmetry of the normal displacement corresponds to symmetry of the transverse displacement. A symmetry (anti-symmetry) boundary therefore refers to symmetry (anti-symmetry) of the normal displacement and anti-symmetry (symmetry) of the transverse displacement. 7 The symmetry and anti-symmetry boundary conditions are implemented by combinations of the free and fixed boundary conditions. The proper implementation of these boundary conditions is validated in Fig. 5 by comparison of the free beam modes calculated with and without the symmetry/anti-symmetry boundary conditions. We enable vacuum within the simulation domain by removing both stress and displacement grid points within the vacuum region. We also remove the stress sampled on the vacuum-material boundary and double boundary-crossing derivatives as appropriate. This exactly emulates a free surface boundary condition along the interface, as desired. We additionally implement PMLs in order to support simulation of leaky modes. The PML is not formally a boundary condition and is instead a region within the computational domain42–45. A desired use for this acoustic mode solver is to estimate propagation losses of a particular acoustic mode in a waveguide, and a key loss mechanism can be radiation loss (analogous to a distributed version of clamping loss in resonators, e.g. cantilevers). We refer to acoustic modes exhibiting radiation loss as a leaky modes, which can still be considered confined modes when the radiation loss rate is much smaller than the oscillation frequency. The PML boundary condition can be implemented as a complex coordinate stretching near the boundary45. This can be achieved by making the grid discretization components ∆x and ∆y complex within the PML region. We choose to use a fixed boundary condition adjacent to the PML in order to force the acoustic field to go to zero at the boundary. In order to replicate the PML used by the commercial solver used for comparison, a linear coordinate stretching was used of the form (cid:18) (cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:19) ∆xP M L = ∆x 1 − jα x − xP M L,begin xP M L,end − xP M L,begin xP M L,end ≥ x ≥ xP M L,begin (19) where α is the scaling parameter of the PML. To demonstrate the accuracy of the PMLs implemented in this solver, we constructed a leaky acoustic waveguide that tunnels radiation across a barrier into the continuum on both sides (Fig. 6). III. MODE SOLVER VALIDATION AND DEMONSTRATION In order to demonstrate the accuracy of the proposed acoustic mode solver, we examine the resonance matrix to verify its Hermitian nature in Sec. III A, then compare several simple cases in our MATLAB implementation to both analytical solutions (Sec. III B) and a commercial, numerical 3D solver (Sec. III C). We verify the symmetry and anti-symmetry conditions in Sec. III D and demonstrate support of leaky modes in Sec. III E. A. Mathematical Validation: Matrix Properties To test our discretization scheme, we examine the properties of resonance operator matrices generated by the mode solver code. To provide an example of the resonance operator matrix, we implement a trial cross-section with two free and two fixed boundary conditions on a 4 × 4 pixel grid in Fig. 3(a). We additionally verify in Fig. 3(b) second order scaling of the error with discretization. For this case, we chose an anti-symmetric Lamb wave with a 10 µm wavelength in a 1 µm thick free slab and varied the discretization along the slab thickness direction. The calculated eigenfrequency and mode shape errors are relative to the analytical solutions, described further in Sec. III B. Our convergence plots indicate that the mode solver eigenfrequency error scales as ε ∝ (∆y)2, i.e. that our mode solver is second-order accurate as expected. The mode shape error scales as ε ∝ (∆y)4 due to the squaring in the mode shape error formula. We additionally expect that, in the absence of material loss or PMLs, the resonance operator would be real and symmetric even after discretization which we verified using a 200 × 200 pixel grid simulation with both free and fixed boundary conditions. We also confirmed that the operator is complex symmetric when PMLs are used. Other hallmarks of Hermitian operators are real eigenvalues, orthogonal eigenvectors, and that the eigenvectors form a complete basis. We have shown an example demonstrating the orthogonality of the eigenvectors in both a free beam and a waveguide without PMLs, and lack thereof when PMLs are implemented, in Fig. 3(c). B. Analytical Validation: Free Slab Waveguide We choose a slab waveguide, which can be analytically solved, to compare the proposed mode solver with elastic wave theory. The simulation implementation and results are shown in Fig. 2. We investigate three fundamental acoustic waves in a thin plate [Fig. 2(b)], namely pure shear waves, anti-symmetric Lamb waves, and symmetric Lamb waves39. We calculate the theoretical dispersion relations [Fig. 2(c)] of these waves compared with the mode solver solutions as the wavelength is varied from 100 µm to 0.5 µm to achieve full coverage of the thin plate regime and 8 FIG. 2. Three acoustic modes (pure shear, anti-symmetric Lamb, and symmetric Lamb) in a 1 µm thick silicon slab were simulated and compared to the theoretical modal distributions and frequencies: (a) Configuration schematic. (b) Illustration of the three acoustic modes simulated. (c) Modal dispersion curves; solid lines are theoretical dispersions and diamond markers are mode solver solutions (shear and longitudinal bulk wave asymptotes also shown). (d) Relative error in frequency and mode εM S = 1 − ((cid:12)(cid:12)(cid:82) UM S · U∗ (cid:17) o dx(cid:12)(cid:12)2)/((cid:82) UM S2 dx(cid:82) Uo2 dx) (cid:16) shape mode shape error is near the level of numerical error (∼ 10−14) and is therefore not shown. between the mode solver and theory. The pure shear wave FIG. 3. Validation of resonance operator matrix properties. (a) Example of the operator matrix. (b) Convergence tests of the mode solver: the 20 dB/decade slope for eigenfrequency error (40 dB/decade for mode shape) indicates second-order accuracy. (c) Tests of modal orthogonality in three configurations, supporting the Hermitian nature of the resonance operator in the absence of PMLs (and non-Hermitian nature with PMLs). into the thick plate (infinite half-space) regime. For the mode solver solutions, we use a plate which is much wider than its thickness (1 × 1000 µm) with the appropriate symmetry boundary conditions on the edges to approximate an infinitely long (1D cross-section) plate. We then calculate the error of the mode solver relative to theory, both in terms of mode shape and eigenfrequency, in Fig. 2(d). We sample the acoustic mode distribution at the center of the plate as an approximation to the theoretical 1+1D case. The mode shape error used here is formulated as )/((cid:82)(cid:82) UM S2 dA(cid:82)(cid:82) Uo2 dA) where Uo is the true mode shape and UM S is the mode εM S = 1 − ((cid:12)(cid:12)(cid:82)(cid:82) UM S · U∗ o dA(cid:12)(cid:12)2 shape found by the solver. For the slab case, we integrate over only a single dimension as the mode shapes are invariant along the second cross-sectional dimension. For the general case, the integration is carried out over the entire cross-section. The mode solver reproduces the key aspects of these acoustic waves with < 1% error across a range of wavelengths which captures the thick, thin, and wavelength-scale plate thickness regimes, indicating that the mode solver faithfully models linear elastic physics. Dispersion Relation(d) ErrorλPure ShearλSymmetric LambλAnti-Symmetric LambApproximate1+1D PlaneSlab SimulationSimulationDomain(b)1 𝜇𝑚1 𝑚𝑚Slab ModesSymmetry PlaneSymmetry Plane(c)(a)Mode ShapeEigenfrequencyWavelength (𝝁𝒎)Frequency (Hz)ErrorWavelength (𝝁𝒎)Wavelength (𝝁𝒎)Pure ShearAnti-Symmetric LambSymmetric LambMode Solver SolutionsTheoryBulk Shear VelocityBulk Longitudinal VelocityOperator Matrix(a)Mode ShapeMode Solver Convergence(b)Eigenfrequency∆𝒚(𝝁𝒎)∆𝒚(𝝁𝒎)npixelErrornpixel(c)Mode OrthogonalityFree BeamWaveguide, PMLsWaveguide, No PMLsdBdBLinearnmodenmodenmodenmodenmodenmodeSolver∝∆𝒚𝟐Solver∝∆𝒚𝟒9 FIG. 4. Four acoustic modes (x- and y-shear, torsion, and pressure waves) in suspended (a) and fixed (b) silicon beams were simulated across a wavelength range of 100 µm to 0.5 µm in both the mode solver and the 3D solver. Shown in each box for the suspended and fixed configurations are a schematic of the simulation implementation, depictions from the 3D solver of the acoustic modes, modal dispersions, and relative error between the mode solver and the numerical solver. The solid lines in the dispersion plots are the mode solver solutions and the diamond markers are the numerical solver solutions. C. Numerical Validation: Free and Fixed Beam Waveguides We now consider a rectangular beam waveguide (in both suspended and fixed, i.e. encased in an infinitely rigid medium, configurations) to compare the proposed mode solver with numerical solutions obtained from a commercial 3D solver28. We investigate the lowest order mode in each of four basic mode families: vertically (y) polarized shear waves, horizontally (x) polarized shear waves, torsional waves, and pressure waves. We choose as our cross-section a 2 × 1 µm silicon beam so as to avoid a degeneracy of the x- and y-shear modes. The same geometry is implemented in the commercial 3D solver with the beam length chosen equal to the simulated wavelength and Floquet periodic boundary conditions applied to the z-oriented boundaries. Both solvers are then used to find these four acoustic modes and corresponding eigenfrequencies across a wavelength range of 100 µm to 0.5 µm. These comparisons are made in Fig. 4 where we depict the comparative error of the mode solver relative to the 3D solver in terms of frequency and modeshape. For the majority of wavelengths, both the mode shape and frequency are very close (< 1%) to the 3D solver results which are defined here as the reference value. D. Symmetry/Anti-symmetry Boundary Condition Validation Having demonstrated the accuracy of the solver, we can use the free beam modes as a basis for comparison with modes computed in a reduced simulation domain that takes advantage of geometrical symmetries. For this case we halve (or quarter) the simulation domain and use a symmetry/anti-symmetry boundary condition on the cut plane to recreate the same mode. We have chosen to use both boundary conditions to recreate the same mode as a redundant test of the boundary conditions, and the pressure mode is chosen as a demonstration that two orthogonal boundary conditions can be used at the same time without loss of accuracy. This comparison is shown in Fig. 5, where examples of the modes and symmetry conditions are shown in Fig. 5(a) and the errors are shown in Fig. 5(b). The jump in error in both eigenfrequency and mode shape at small wavelengths is due to the simulated modes being nearly degenerate in frequency at these wavelengths, causing the solver to return mixed mode shapes and eigenfrequencies when the full simulation domain is used. The halved/quartered simulation domains did not have mixing due to the imposed symmetry requirements. E. Solver Demonstration: Evanescent Waveguiding, Leaky Modes, and PMLs We now construct an acoustic waveguide formed of two materials (a 'core' and a 'cladding') and find evanescently confined acoustic modes. Adding a radiation layer (formed of core material) to the waveguide cross-section creates a leaky acoustic waveguide that tunnels radiation across a (cladding) barrier into the radiation-mode continuum on X-ShearY-ShearPressureTorsional𝜆=10𝜇𝑚EigenfrequencySimulationMode DispersionErrorAcoustic Modes(b)Mode ShapeErrorFrequency (Hz)Wavelength (𝝁𝒎)Fixed BeamX-ShearY-ShearPressureTorsional𝜆=10𝜇𝑚SimulationDomain1𝜇𝑚2𝜇𝑚VacuumEigenfrequencySimulationMode DispersionErrorAcoustic Modes(a)Mode ShapeWavelength (𝝁𝒎)Wavelength (𝝁𝒎)ErrorFrequency (Hz)Wavelength (𝝁𝒎)Free BeamX-ShearY-ShearTorsionPressureNumerical SolutionsMode Solver SolutionsBulk Shear VelocityBulk Longitudinal VelocitySimulationDomain1𝜇𝑚2𝜇𝑚InfinitelyRigid MaterialWavelength (𝝁𝒎)Wavelength (𝝁𝒎)10 FIG. 5. Three acoustic modes (x- and y-shear and pressure waves) are simulated in a suspended silicon beam in a truncated simulation domain using symmetric and anti-symmetric boundary conditions and compared with the same modes simulated in the full simulation domain. (a) Schematic depictions of the symmetric and anti-symmetric boundary conditions and simulated mode shapes. (b) Comparison of mode shape and eigenfrequency error between the modes obtained with the full simulation domain, and the truncated simulation domain using symmetry/anti-symmetry boundary conditions. Increased errors at the shortest wavelengths are due to the mode solver returning mixed modes when the eigenfrequencies are nearly degenerate. FIG. 6. An embedded strip acoustic waveguide (1 × 0.4 µm) is simulated to demonstrate evanescent confinement, wave-guiding, radiation tunneling, leaky modes, and PML operation. (a) Schematic depiction of an acoustic waveguide and log-scale, cross- sectional plot of the horizontal displacement of the quasi-Love wave guided mode with linear scale inset. (b) The same waveguide after silica slabs have been added (with a 0.5 µm silicon barrier) with plots of the leaky mode. (c) An extended 1D slice of the leaky acoustic mode in (b) showing the displacement amplitude. (d) Propagation loss comparison between the 3D solver and the mode solver. (e) Fourier transform of displacement amplitudes in the leaky section of the 1D slice from (c), demonstrating coupling into two radiation modes. (f) Mode propagation and radiated wavefronts (shown schematically) for both horizontal and vertical displacement components. both sides (see analogous optical slab waveguide46). Implementation of PMLs on the simulation domain edges then allows outward radiating waves to be absorbed without reflection, enabling the simulation of leaky modes analogous to electromagnetic solvers47 and calculation of waveguide mode radiation losses. The configuration, modal amplitude plots, radiation patterns, and calculated radiation losses are shown in Fig. 6. First, an evanescently guiding waveguide is constructed [Fig. 6(a)] by embedding a strip of silica within a silicon substrate. This waveguide supports two 'polarizations' of acoustic modes, analogous to the Love and Rayleigh sur- face waves39. The former wave is chosen and the depicted log-scale plot of the horizontal displacement amplitude demonstrates that the wave is evanescently confined. We then add silica slabs to either side of the silica waveguide, separated by variable width silicon barriers, which cause the waveguide and guided acoustic modes to couple to the EigenfrequencyMode ShapeErrorSymmetry Boundary ConditionsX-Symm. BCSimulation DomainNot SimulatedY-Symm. BCY-Anti-Symm. BCX/Y-Anti-Symm. BC(b)(a)X-Shear Mode, 𝒖𝒙X-Anti-Symm. BCY-Shear Mode, 𝒖𝒙Y-Shear Mode, 𝒖𝒚X-Shear Mode, 𝒖𝒚Pressure Mode, 𝒖𝒛𝜆=1𝜇𝑚Wavelength (𝝁𝒎)ErrorErrorX-Shear, X-SymmX-Shear, Y-Anti-SymmY-Shear, Y-SymmY-Shear, X-Anti-SymmPressure, X/Y-Anti-SymmWaveguideSilicon SubstratePMLs(a)(c)Mode SolverNumerical SolverRadiation QBarrier Width (𝝁𝒎)Propagation Loss(d)Leaky WaveguideVariable Width BarriersSilicaSlabsSilicaWaveguidePMLMode AttenuationMode LeakagePML(b)Leaky WaveguideBarrier DecayMain LobeCross-Sectional SliceLog ScaleLog ScaleWaveguide ModeLeaky ModedBHorizontal Displacement Amplitude (10log10𝑢𝑥)PMLPMLLinear ScaleLinear ScaleX (𝝁𝒎)𝒖𝒙𝒖𝒚AmplitudeAngle (o)𝒖𝒙𝒖𝒚(e)Radiation Spectrum(f)Mode RadiationAmplitudeX (𝝁𝒎)X (𝝁𝒎)𝒖𝒙Z (𝝁𝒎)𝒖𝒚11 slab radiation continuum on both sides and become 'leaky' [Fig. 6(b)]. Oscillation of the field within the silica layer can be seen in the inset, and non-decaying intensity within the leaky region can be seen in the log-scale plot; both are indicative of the radiative loss of the guided mode into the adjacent silica layers. An extended simulation domain with a longer silica layer further demonstrates the presence of this mode leakage in Fig. 6(c) where a cross-sectional slice plots both horizontal and vertical displacement amplitudes within the different material regions. The mode displacement is predominantly confined in the waveguide with an evanescent tail in the silicon barrier. Some of the acoustic energy tunnels across the barrier into the silica slab, where it becomes oscillatory, and then enters the PML and is attenuated as expected. A comparative plot of this propagation loss, as a function of silicon barrier width, is shown in Fig. 6(d) as numerical validation of the accuracy of the PML. By taking a Fourier transform of the mode leakage section we show that two primary radiation field components are excited in the leaky wave [Fig. 6(e)], corresponding to quasi-Love and quasi-Rayleigh waves in the silica slab. Fig. 6(f) shows a cross-section in the x − z plane depicting the horizontal (left) and vertical (right) displacement amplitudes, where the wavefronts have been outlined for both polarizations and overlaid on both plots (gray for horizontal displacement, black for vertical). IV. CONCLUSION We have developed an eigenmode solver based on a finite-difference scheme on a staggered-grid for the linear, isotropic elastic wave equation. We have verified the accuracy of the solver by comparison with both theory and a 3D numerical solver (COMSOL). We have also used these comparisons to demonstrate correct implementation of fixed and free boundary conditions. We then used a leaky acoustic waveguide to demonstrate the solver's ability to simulate evanescent guiding, leaky modes, and numerically accurate PMLs. We expect that the mode solver will be of utility in the design of on-chip acoustic wave devices. It is particularly suited to interfacing with electromagnetic solvers on the Yee grid for applications based on acousto-optics and optomechanics. ACKNOWLEDGMENTS The authors acknowledge helpful discussions with Yossef Erhlichman, Cale Gentry, and Bohan Zhang. This work was supported by a National Science Foundation Graduate Research Fellowship (Grant #1144083) and a 2012 Packard Fellowship for Science and Engineering (Grant #2012-38222). 1G. Anetsberger, E. Gavartin, O. Arcizet, Q. P. Unterreithmeier, E. M. Weig, M. L. Gorodetsky, J. P. Kotthaus, and T. J. Kippenberg, "Measuring nanomechanical motion with an imprecision below the standard quantum limit," Physical Review A 82, 061804 (2010). 2J. Teufel, T. Donner, M. Castellanos-Beltran, J. Harlow, and K. Lehnert, "Nanomechanical motion measured with an imprecision below that at the standard quantum limit," Nature Nanotechnology 4, 820–823 (2009). 3H. Lee, T. Chen, J. Li, K. Y. Yang, S. Jeon, O. Painter, and K. J. Vahala, "Chemically etched ultrahigh-Q wedge-resonator on a silicon chip," Nature Photonics 6, 369–373 (2012). 4I. S. Grudinin, A. B. Matsko, and L. Maleki, "Brillouin lasing with a CaF2 whispering gallery mode resonator," Phys. Rev. Lett. 102, 043902 (2009). 5M. Bagheri, M. Poot, M. Li, W. P. Pernice, and H. X. Tang, "Dynamic manipulation of nanomechanical resonators in the high-amplitude regime and non-volatile mechanical memory operation," Nature Nanotechnology 6, 726–732 (2011). 6C.-H. Dong, Z. Shen, C.-L. Zou, Y.-L. Zhang, W. Fu, and G.-C. Guo, "Brillouin-scattering-induced transparency and non-reciprocal light storage," Nature Communications 6 (2015). 7C. G. Poulton, R. Pant, A. Byrnes, S. Fan, M. Steel, and B. J. Eggleton, "Design for broadband on-chip isolator using stimulated Brillouin scattering in dispersion-engineered chalcogenide waveguides," Optics Express 20, 21235–21246 (2012). 8S. Weis, R. Rivi`ere, S. Del´eglise, E. Gavartin, O. Arcizet, A. Schliesser, and T. J. Kippenberg, "Optomechanically induced transparency," Science 330, 1520–1523 (2010). 9J. Kim, M. C. Kuzyk, K. Han, H. Wang, and G. Bahl, "Non-reciprocal Brillouin scattering induced transparency," Nature Physics 11, 275–280 (2015). 10G. Bahl, M. Tomes, F. Marquardt, and T. Carmon, "Observation of spontaneous Brillouin cooling," Nature Physics 8, 203–207 (2012). 11J. Chan, T. M. Alegre, A. H. Safavi-Naeini, J. T. Hill, A. Krause, S. Groblacher, M. Aspelmeyer, and O. Painter, "Laser cooling of a nanomechanical oscillator into its quantum ground state," Nature 478, 89–92 (2011). 12J. Teufel, T. Donner, D. Li, J. Harlow, M. Allman, K. Cicak, A. Sirois, J. D. Whittaker, K. Lehnert, and R. W. Simmonds, "Sideband cooling of micromechanical motion to the quantum ground state," Nature 475, 359–363 (2011). 13V. Peano, C. Brendel, M. Schmidt, and F. Marquardt, "Topological phases of sound and light," Physical Review X 5, 031011 (2015). 14M. Schmidt, S. Kessler, V. Peano, O. Painter, and F. Marquardt, "Optomechanical creation of magnetic fields for photons on a lattice," Optica 2, 635–641 (2015). 15E. A. Kittlaus, H. Shin, and P. T. Rakich, "Large Brillouin amplification in silicon," Nature Photonics (2016). 16O. Arcizet, P.-F. Cohadon, T. Briant, M. Pinard, and A. Heidmann, "Radiation-pressure cooling and optomechanical instability of a micromirror," Nature 444, 71–74 (2006). 17T. J. Kippenberg and K. J. Vahala, "Cavity optomechanics: back-action at the mesoscale," science 321, 1172–1176 (2008). 12 18A. Byrnes, R. Pant, E. Li, D.-Y. Choi, C. G. Poulton, S. Fan, S. Madden, B. Luther-Davies, and B. J. Eggleton, "Photonic chip based tunable and reconfigurable narrowband microwave photonic filter using stimulated Brillouin scattering," Optics Express 20, 18836–18845 (2012). 19E. Verhagen, S. Del´eglise, S. Weis, A. Schliesser, and T. J. Kippenberg, "Quantum-coherent coupling of a mechanical oscillator to an optical cavity mode," Nature 482, 63–67 (2012). 20T. Palomaki, J. Harlow, J. Teufel, R. Simmonds, and K. Lehnert, "Coherent state transfer between itinerant microwave fields and a mechanical oscillator," Nature 495, 210–214 (2013). 21J. Bochmann, A. Vainsencher, D. D. Awschalom, photons," Nature Physics 9, 712–716 (2013). and A. N. Cleland, "Nanomechanical coupling between microwave and optical 22X. Lu, J. Galipeau, K. Mouthaan, E. H. Briot, and B. Abbott, "Reconfigurable multiband saw filters for lte applications," in Radio and Wireless Symposium (RWS), 2013 IEEE (IEEE, 2013) pp. 253–255. 23D. Psychogiou, R. G´omez-Garcı, D. Peroulis, et al., "Acoustic wave resonator-based absorptive bandstop filters with ultra-narrow bandwidth," IEEE Microwave and Wireless Components Letters 25, 570–572 (2015). 24J. R. Clark, W.-T. Hsu, M. A. Abdelmoneum, and C.-C. Nguyen, "High-Q UHF micromechanical radial-contour mode disk resonators," Journal of Microelectromechanical Systems 14, 1298–1310 (2005). 25D. Weinstein and S. A. Bhave, "The resonant body transistor," Nano Letters 10, 1234–1237 (2010). 26N. A. Petersson and B. Sjogreen, "Wave propagation in anisotropic elastic materials and curvilinear coordinates using a summation-by- parts finite difference method," Journal of Computational Physics 299, 820–841 (2015). 27K. Gao, S. Fu, R. L. Gibson, E. T. Chung, and Y. Efendiev, "Generalized multiscale finite-element method (gmsfem) for elastic wave propagation in heterogeneous, anisotropic media," Journal of Computational Physics 295, 161–188 (2015). 28COMSOL Multiphysics, version 4.3a (COMSOL Inc., Burlington, Massachusetts, 2012). 29L. Gavri´c, "Computation of propagative waves in free rail using a finite element technique," Journal of Sound and Vibration 185, 531–543 (1995). 30P. Wilcox, M. Evans, O. Diligent, M. Lowe, and P. Cawley, "Dispersion and excitability of guided acoustic waves in isotropic beams with arbitrary cross section," in AIP Conference Proceedings, Vol. 615 (AIP, 2002) pp. 203–210. 31M. Castaings and M. Lowe, "Finite element model for waves guided along solid systems of arbitrary section coupled to infinite solid media," The Journal of the Acoustical Society of America 123, 696–708 (2008). 32I. Bartoli, A. Marzani, F. L. di Scalea, and E. Viola, "Modeling wave propagation in damped waveguides of arbitrary cross-section," Journal of Sound and Vibration 295, 685–707 (2006). 33K. S. Yee et al., "Numerical solution of initial boundary value problems involving maxwells equations in isotropic media," IEEE Trans. Antennas Propag. 14, 302–307 (1966). 34R. Madariaga, "Dynamics of an expanding circular fault," Bulletin of the Seismological Society of America 66, 639–666 (1976). 35L. Etemadsaeed, P. Moczo, J. Kristek, A. Ansari, and M. Kristekova, "A no-cost improved velocity–stress staggered-grid finite-difference scheme for modelling seismic wave propagation," Geophysical Journal International 207, 481–511 (2016). 36MATLAB, version R2015a (The MathWorks Inc., Natick, Massachusetts, 2015). 37N. Dostart and M. Popovi´c, "Matlab central file exchange: Elastic mode solver," (Mathworks, 2016), [Online; accessed 14-April-2017]. 38A. W. Snyder and J. D. Love, Optical Waveguide Theory (Springer Science & Business Media, 2012). 39B. A. Auld, Acoustic Fields and Waves in Solids, 2nd ed. (Krieger, Malabar, 1990). 40W. C. Chew, "Electromagnetic theory on a lattice," Journal of Applied Physics 75, 4843–4850 (1994). 41J. D. Joannopoulos, S. G. Johnson, J. N. Winn, and R. D. Meade, Photonic Crystals: Molding the Flow of Light (Princeton university press, 2011). 42W. C. Chew and Q. H. Liu, "Perfectly matched layers for elastodynamics: a new absorbing boundary condition," Journal of Computa- tional Acoustics 4, 341–359 (1996). 43W. C. Chew and W. H. Weedon, "A 3D perfectly matched medium from modified Maxwell's equations with stretched coordinates," Microwave and Optical Technology Letters 7, 599–604 (1994). 44J.-P. Berenger, "A perfectly matched layer for the absorption of electromagnetic waves," Journal of computational physics 114, 185–200 (1994). 45F. Teixeira and W. Chew, "Analytical derivation of a conformal perfectly matched absorber for electromagnetic waves," Microwave and Optical technology letters 17, 231–236 (1998). 46M. Popovi´c, Theory and design of high-index-contrast microphotonic circuits, Ph.D. thesis, Massachusetts Institute of Technology (2008). 47N.-N. Feng, G.-R. Zhou, C. Xu, and W.-P. Huang, "Computation of full-vector modes for bending waveguide using cylindrical perfect- lymatched layers," Journal of lightwave technology 20, 1976 (2002).
1803.07065	1	1803	2018-03-19T17:58:36	Sensorless Resonance Tracking of Resonant Electromagnetic Actuator through Back-EMF Estimation for Mobile Devices	[ "physics.app-ph", "eess.SY", "eess.SP", "math.DS" ]	Resonant electromagnetic actuators have been broadly used as vibration motors for mobile devices given their ability of generating relatively fast, strong, and controllable vibration force at a given resonant frequency. Mechanism of the actuators that is based on mechanical resonance, however, limits their use to a situation where their resonant frequencies are known and unshifted. In reality, there are many factors that alter the resonant frequency: for example, manufacturing tolerances, worn mechanical components such as a spring, nonlinearity in association with different input voltage levels. Here, we describe a sensorless resonance tracking method that actuates the motor and automatically detects its unknown damped natural frequency through the estimation of back electromotive force (EMF) and inner mass movements. We demonstrate the tracking performance of the proposed method through a series of experiments. This approach has the potential to control residual vibrations and then improve vibrotactile feedback, which can potentially be used for human-computer interaction, cognitive and affective neuroscience research.	physics.app-ph	physics	Preprint Sensorless Resonance Tracking of Resonant Electromagnetic Actuator through Back-EMF Estimation for Mobile Devices Youngjun Cho UCL Interaction Centre, University College London, London, United Kingdom * [email protected] Abstract Resonant electromagnetic actuators have been broadly used as vibration motors for mobile devices given their ability of generating relatively fast, strong, and controllable vibration force at a given resonant frequency. Mechanism of the actuators that is based on mechanical resonance, however, limits their use to a situation where their resonant frequencies are known and unshifted. In reality, there are many factors that alter the resonant frequency: for example, manufacturing tolerances, worn mechanical components such as a spring, nonlinearity in association with different input voltage levels. Here, we describe a sensorless resonance tracking method that actuates the motor and automatically detects its unknown damped natural frequency through the estimation of back electromotive force (EMF) and inner mass movements. We demonstrate the tracking performance of the proposed method through a series of experiments. This approach has the potential to control residual vibrations and then improve vibrotactile feedback, which can potentially be used for human-computer interaction, cognitive and affective neuroscience research. Keywords: Sensorless resonance tracking, Back EMF, Estimation, Sensorless drive, Resonant electromagnetic actuator, Resonant frequency detection, Vibrotactile. I. Introduction While the visual display of consumer electronics such as smartphones has become one of the most fundamental and effective means in connecting a person to graphical contents and virtual worlds, mechanical keypads in such devices have been disappearing, allowing the screen to be larger and an embedded vibration motor to instead provide vibration feedback. Although there are individual differences in preferences for vibro-tactile effects, having a better vibration actuator and its feedback has been regarded as an important factor in designing state-of-the-art mobile devices and user experiences [1–6]. Various mechanisms have been proposed to make vibration actuators capable of effectively stimulating Pacinian Corpuscles, one of human mechanoreceptors, which responds to rapid vibration of low frequency (up to approximately 500 Hz) [7,8], through a mobile device. Along with the type of mechanisms, actuators can generally be categorized into three groups [4]: i) linear electromagnetic actuators which produce linear actuations enabled by a one-degree-of-freedom mechanical oscillator, ii) rotary 1 Preprint electromagnetic actuators that convert direct current (DC) into rotary force, and iii) non-electromagnetic actuators based on the use of smart materials (e.g., piezoelectric materials [9,10], electro-active polymers [11]). In the consumer electronics industry, the linear and rotary electromagnetic actuators have been dominantly used because of their better mechanical durability and inexpensive cost than those of the latter. In particular, the resonant electromagnetic actuator and eccentric rotating mass motor, which are special types of the linear electromagnetic actuators and the rotary electromagnetic actuators, respectively [4], are known as the most commercially successful vibration motors, given that they consume relatively low electrical power, but are still capable of generating sufficient vibration force [11,12]. By contrast with the eccentric rotating mass motors in which the vibration force is coupled with the frequency element [11,13], the force of the resonant electromagnetic actuator is controllable at a specific range of frequency. This is enabled by its mechanism that employs the mechanical linear resonance phenomenon to maximize its vibration force at a given frequency and limited driving energy. This also helps the resonant electromagnetic actuator, which is also commercially branded Linear Resonant Actuator (LRA), to respond faster than the rotary motors. Hence, the actuator has been employed more and more recently in mobile devices. To take an advantage from the mechanical resonance on which a resonant electromagnetic actuator relies, its resonant frequency needs to be either known prior to the use or automatically detectable. Although most manufacturers of the actuator set a specific natural frequency and provide the information, the frequency is often shifted due to many external and internal factors. For instance, driving the actuator for a long period of time or dropping it tear down its internal components such as a mechanical spring, leading to changes of its stiffness and then natural frequency. In addition, nonlinearity in the mechanical mass-spring-damper system exists in reality given many reasons such as the heat production, also leading to shifts of its resonance [14]. The mismatch between frequencies of electrical driving signals and the resonance induces a drop in the vibration force and difficulty in controlling residual vibrations [15]. Thus, an automatic resonance tracking approach is required to address such issue. A vibration is produced by the movement of the inner mass, which can be tracked through the use of additional sensors such as the hall effect sensors [16] and piezoelectric material [10]. In the real world, however, it is difficult to add an additional sensing channel to inner compact spaces of a mobile device and of an actuator. Hence, we focus on the automatic resonance tracking without the use of a physical sensor. This can be achieved by the estimation of back electromotive force, so-called, back EMF, which is the voltage induced by the magnetic linkage flux variation in accordance with Lenz's law, as used in other mechanical actuating systems for controlling torque and displacements [17–20]. For this, we first analyze the actuation model and build a new sensorless resonance tracking algorithm that drives an actuator, makes an inner coil to be in high impedance, and then estimates the back EMF so as to automatically detect the damped (unknown or shifted) natural frequency. This can be of help in keeping the vibration performance from being deteriorated and in extending lifetimes of actuators. The proposed drive scheme produces further potential benefits in designing vibrotactile effects. In particular, a residual vibration, one of the challenges in controlling a mechanical oscillation system [15,21], can be controlled through the estimation of back EMF, potentially helping to 2 Preprint improve a person's vibrotactile perception. This paper is organized as follows: first, we introduce details of commercialized resonant electromagnetic actuators which have been widely used in mobile devices in Section II. Section III discusses the mechanical actuation model commonly used for the actuators. Next, we analyze the electro-magnetic circuit combined with the mechanical structure to derive the relationship between the natural frequency and the back electromotive force, and propose the sensorless resonance tracking algorithm in Section IV. Section V describes conducted experiments and results. Finally, we conclude the paper in Section VI. II. Resonant Electromagnetic Actuators for Mobile Devices A resonant electromagnetic actuator is a typical type of linear electromagnetic actuators developed for handheld consumer electronics mobile devices to convey vibration feedback. The schematic mechanical structure of the actuator is illustrated in Fig.1. This is on the basis of the linear oscillation mechanism. Based on this structure, two types are commercially available in accordance with different coil compositions: the first type is a voice coil based actuator (i.e., coil without ferromagnetic component), such as, Linear vibration motor (also called Linear resonant actuator) [22], the other is a solenoid based one (i.e., coil with ferromagnetic materials), such as [23]. By reflecting the sensitive frequency range of the human mechanoreceptor [8,24], the resonant frequency of both types is generally set to a value between 150Hz to 250Hz. Fig. 1. Schematic mechanical structure of the resonant electromagnetic actuator. Both types of actuators tend to have a high Q factor, which draws a high resonance sharpness [25], so as to maximize their vibration force at a given natural frequency. However, the high Q value, at the same time, narrows their operable frequency range. As discussed above, a shifted damped natural frequency dramatically decreases the vibration force of an actuator with the high Q value when it is driven at a preset frequency provided by its manufacturer. For example, the natural frequency can be easily shifted along with different peak voltage levels of input driving signals as shown in Fig. 2. Here, we tested a commercial resonant electromagnetic actuator (LRA, SEMCO 1036) whose resonant frequency is designed to be 175Hz. The frequency responses collected through the use of an accelerometer and a signal 3 Preprint generator indicate the higher the peak voltage of the input signal, the lower the actual resonant frequency (e.g., 175.5Hz at 1.0 Vpeak and 173.4Hz at 1.5 Vpeak from this pilot test). Therefore, it is required to automatically track the varied resonant frequency when using the actuator. Fig. 2. Measured the shifted resonant frequency of the resonant electromagnetic actuator (Product: LRA – SEMCO 1036) along with the different peak voltage of input signal on the frequency response: (a) 175.5Hz - 1.0Vpeak, (b)174.1Hz - 1.2Vpeak, (c)173.4Hz - 1.5Vpeak. III. Actuator Model Fig. 3 shows the electrical-mechanical model of the resonant electromagnetic actuator mounted in the mobile device. The combined system consists of an electrical circuit of a single coil and an equivalent mechanical model of its surroundings. The external excitation force relies on the electrodynamic force generated by the magnetic flux between the coil and the permanent magnet. Given the one degree of freedom structure, we can assume that a direction of current flow inside the coil is vertical to the magnetic flux linkage and the electrodynamic force can be estimated by Fleming's rule. Thus, the external excitation force can be controlled by the electrical input u : tF )( = tuK )( f )(tu is the time-varying input signal, where fK is the force constant. In consideration of the operating principle of the actuator that employs a harmonic voltage source, Eq (1) can be rewritten as (1) (2) tF )( = VK f A sin( t ω ) 4 Preprint where AV is the peak voltage of the input signal, ω is a frequency of the applied harmonic input source. Fig. 3. Combined electrical circuit and mechanical model of the actuator embedded in mobile devices. In the Laplace domain, the dynamic equation of the linear model illustrated in Fig. 3 can be expressed with initial conditions (i.e., initial displacement velocity ) and Eq (2): )0(x! )0(x , initial sX )( = 2 ( s + 2 ω VK ω Af ms )( 2 + mcs + 2 ω n ( + ) xcms )0() + mcs + + + ms 2 xm )0( ! 2 ω n )(sX (3) where of the inner mass m , k is the stiffness of the spring and c is the constant of proportionality, the natural frequency denotes the Laplace transform of the displacement nω can be determined by )(tx n =ω . k m (4) Given the mechanical characteristics based on resonance, a resonant electromagnetic actuator is required to be driven by nωω= to maximize )(sX . )(tu at a resonant frequency, i.e., Lastly, the Q factor of the mechanical structure, which determines the sharpness of resonance, needs to be set to a certain value that enhances the vibration force of the actuator: where the damping ratio is given by =ζ =Q 1 ζ2 c km 2 (5) (6) . Empirically, in the manufacturing process, this value is set to lower than approximately 0.05 to have a high Q , contributing to having a strong vibration force at the same time while narrowing the operable frequency range. Once the natural frequency is shifted, it cannot help but depreciating the actuation performance. 5 Preprint IV. Sensorless Drive for Resonant Frequency Tracking In this section, we analyze electrical dynamics of the resonant electromagnetic actuator to derive the relationship between the natural frequency and the electric response of the actuation system in order to build a drive scheme that helps to estimate the back EMF, automatically track the natural frequency and produce resonance without the use of an extra physical sensor. A. Estimation of Back EMF and Damped Natural Frequency Derived from the electrical and magnetic characteristics of the coil in Fig.3, a voltage signal u driven to the actuator can be expressed as tu )( = = tRi )( tRi )( ε+ N + ΔΦ t Δ where R is the resistance of the coil, ε is the potential difference across the system, N is the number of the coil turns, Φ is the combined magnetic flux produced from the coil (7) (8) (9) 6 and a permanent magnet (PM) =Φ c Ni ℜ TH . AB=Φ r m r rB has a constant value which can be decided The remnant flux density of the magnet by the type of PM material, and the total magnetic reluctance influenced by composition of flux paths such as the dimension of PM area permeability of magnetic components and leakage fluxes over the air gap between the coil and the PM. Suppose that the displacement x in Eq (3) is associated only with the linear actuation so moves in a vertical direction, the flux combination Φ can be expressed in conjunction with the flux linkage λ: THℜ could be a function of x and THℜ in Eq (8) is rA , a Φ=Φ c ix ),( ix ),( λ=Φ+ N m . (10) By introducing Eq (10) into Eq (7), we can obtain tu )( = tRi )( ix d ),( λ+ dt ix ),( λ ∂+ x ∂ b + ! dx tRi )( dt diLxKtRi dt )( + = = di dt ix ),( ∂+ λ i ∂ (11) where the first term in the bottom of Eq (11) is a resistive voltage drop, the second term is a voltage source internally generated within the coil of the actuator, also called back electromotive force (EMF), and the last term is the inductive voltage from current variations. Here, L is the inductance. Assuming that the flux linkage is linearly proportional to the displacement, the back EMF term can be assumed to have Preprint a linear relation to the velocity of the movable mass x! , so it can be estimated with a constant bK . Suppose that the electrical input supply to the actuation system is cut off, i.e., 0 , making the coil to be in high impedance, the back EMF bV can be estimated )( =tu by measuring potential difference between each terminal of the coil in the actuator. This can be expressed as tV )( b = xK ! b −= ! iL − Ri (12) removing the first term in Eq (3), called the steady-state Given Eq (6) and response, we can also obtain 0=AV sX )( = ( s + s x x 2 )0() ζω + ! n s 2 2 2 + ωζω + n n )0( (13) which represents the transient response which depends on initial conditions. Given the initial conditions cannot be negligible during the actuation mode, Eq (13) can be expressed in the time domain as tx )( ℘= 0 e tn ζω − sin( ϕω 0 + d t ) (14) where the magnitude constant initial displacement transient response depends on the damped natural frequency less than 0℘ and the phase shift )0(x! and the initial velocity )0(x nω: 0ϕ can be determined by the , and the periodic property of the dω which is equal to or . (15) 1 ζωω 2 = n − d The time derivative of Eq (14) is given by tx )( ! tn e = − ℘= d ( ζωζω ℘− n d 0 e t sin( ζω − ϕω n d sin( t + d t ) + ϕω 0 ) ϕωω ℘+ 0 0 cos( + t d d )) (16) where d℘ and dϕ are the magnitude and the phase delay of the velocity, respectively. Finally, by substituting Eq (16) into Eq (12), we can derive the main result from the mathematical analysis in this section, i.e., the relationship between the back EMF and the damped natural frequency: tV )( b ℘= d K b ζω − n t e sin( ϕω d + d t . ) (17) Given this, we can extract periodic characteristics of the actuation system, that is, the resonant frequency, through the estimation of the back EMF. As discussed above, this deduction can be valid when the moving mass is excited by a certain external voltage input before the moment when the input is stopped to supply, contributing to nonzero initial conditions. 7 Preprint B. Automatic Resonant Frequency Tracking Algorithm Based on the relationship expressed in Eq (17), we propose a new sensorless drive scheme for resonant electromagnetic actuators through the back EMF estimation for the real-time tracking of the resonant frequency during the actuation mode. This approach does not require the use of a physical sensor. The proposed method also does not require to know, or manually check, the natural frequency preset by a manufacturer. A unit step input can be used to oscillate the actuation system, and the induced transient response Eq (13) can be monitored through probing the tip of each input terminal of the actuator coil and estimating the back EMF. The estimated signal can be used to synchronize the next input aligned with the periodical motion of the inner mass. In the actuation model shown in Fig. 3, the velocity of the moving mass is maximized at equilibrium given that the restoring force of the elastic spring resists against the applied external force in accordance with Hooke's law [26]. Given this, our idea is to amplify the velocity at the moment when the peak of x! is detected, by generating the external electromagnetic force as shown in Fig. 4. This helps to make the time interval between current and previous step inputs fit into the damped natural period, in turn driving the actuator at a self-sensed resonant frequency. Here, we assume that the phase shift in Eq (17) can be ignorable in reality given that the value is small during the actuation of the system designed to produce vibration feedback at the low frequency range and our focus is laid on the exploration of the periodicity. Fig. 4. Schematic graph for the periodic monitoring of the velocity of the inner mass of the resonant electromagnetic actuator estimated by the back electromotive force. As can be seen in Fig. 4, iterative transitions to the high impedance state is required so as to drive the actuator and estimate the back EMF. The connection of each physical terminal of the actuator to a controller needs to be controllable for connecting it to a high impedance node for the observation of the mass movement and connecting it to an external power source for the driving. Fig. 5 describes the block diagram of the proposed sensorless drive system. A switch module in the motor driver, which is linked to shut-down ports of amplifiers (i.e., an audio amplifier that amplifies the current level of input sources and a voltage amplifier that increases the amplitude of the back EMF signals), helps the transition. 8 Preprint Fig. 5. Block diagram of the proposed sensorless resonant electromagnetic actuator drive system. An additional application of this sensorless drive method is the residual vibration control. A vibration that lasts after the entire actuation period stops is called residual vibration. This affects a person's haptic perception. For the sensorless system, model- based feed-forward control methods such as the input shaping technique [15] can be used to reduce the residual oscillation. This type of techniques, however, cannot address the resonant frequency shift issue. On the other hand, the proposed approach could support the residual vibration reduction given that the system is capable of monitoring the periodical characteristics. Algorithm 1 describes details of the proposed algorithmic flow for the automatic resonance tracking and for the reduction of residual vibrations. The differential amplified back EMF (in Fig. 5) signal can be read through ad0_float. This is low- pass filtered with a cutoff frequency of 500Hz, considering the sensitivity range of human mechanoreceptors [7,8] and the operable frequency range of resonance-based actuators commercialized for mobile devices. The unit step input is driven by calling a function diff_DAC_out() with a parameter to set a current direction within the actuator coil. The duration of each step input, PULSE_DURATION, is set to be equal to or less than a half period of resonant frequency nωπ for the monitoring of the mass movement during the high impedance state. A function high_Impedance() makes the audio amplifier (in Fig. 5) shutdown for the transition to the high impedance state. Additionally, for the residual vibration reduction, a parameter named isBreakingTriggered is used to invert the step input direction, and the iterative operation lasts until the filtered back EMF value becomes lower than a threshold value (e.g., 10% of the maximum value of the filtered signal). 9 Preprint high_Impedance(OFF) if isBreakingTriggered then diff_DAC_out(POSITIVE) else if ~isBreakingTriggered then diff_DAC_out(NEGATIVE) end if isBackEMFmonitoring← false time_tic←0 end if else if ~isBackEMFmonitoring then if isBreakingTriggered then diff_DAC_out(NEGATIVE) else if ~isBreakingTriggered then diff_DAC_out(POSITIVE) end if if ++time_tic> PULSE_DURATION then end if high_Impedance(ON) isBackEMFmonitoring← true time_tic←0 driving_mode← BEMF_AND_UNIT_PULSE_II end if else if driving_mode == BEMF_AND_UNIT_PULSE_II then if isBackEMFmonitoring then high_Impedance(OFF) if isBreakingTriggered then else if ~isBreakingTriggered then diff_DAC_out(NEGATIVE) diff_DAC_out(POSITIVE) end if end if isBackEMFmonitoring← false time_tic←0 else if ~isBackEMFmonitoring then if isBreakingTriggered then diff_DAC_out(POSITIVE) else if ~isBreakingTriggered then diff_DAC_out(NEGATIVE) end if if ++time_tic> PULSE_DURATION then end if high_Impedance(ON) isBackEMFmonitoring← true time_tic←0 driving_mode← BEMF_AND_UNIT_PULSE_I end if end if if isBreakingTriggered && back_emf_y[0] < DESIRED_VALUE then return ▷ To finish the residual vibration control end if if (back_emf_y[0]-back_emf_y[1])>thr && (back_emf_y[1]-back_emf_y[2])<-thr then ▷ detecting a negative peak Algorithm 1. Auto-tracking of resonant frequency & Reduction of residual vibration. function TIMER_ISR() ▷ being called along with the sampling frequency, fs ad0_float←readDiffADC() back_emf_y←lowpassfilter(ad0_float, 500, fs) ▷ applying LPF with cut-off frequency 500Hz if driving_mode == BEMF_AND_UNIT_PULSE_I then if isBackEMFmonitoring then if (back_emf_y[0]-back_emf_y[1])<-thr && (back_emf_y[1]-back_emf_y[2])>thr then ▷ detecting a positive peak end function V. Experimental Validation and Results We conducted experiments to evaluate the performance of the proposed sensorless drive. In this validation, we aim to test the accuracy in tracking the resonant frequency and the response time from the residual vibration control. Fig. 6 shows the overall 10 Preprint experimental setup that includes the proposed sensorless driver and a resonant electromagnetic actuator. This setup was used both for collecting the frequency response profile of a resonant electromagnetic actuator using the current amplifier and driving the actuator based on the proposed scheme using the designed motor driver. An actuator sample was attached on a mock-up device which weighs 150g and has a similar dimension of smartphones. We placed this on sponge surfaces to best minimize energy loss due to the friction force and sound. The vibration force was measured through the use of an accelerometer (Bruel and Kjaer 4524) and the measured signals were amplified by a charge amplifier (Bruel and Kjaer 2692). For the conversion from the analog signal to the digital one, a data acquisition board (NI USB 6259) was used. An oscilloscope (HP 54622A) was further used to display measured signals. The sampling frequency was set to 10kHz for both the driving of the actuator and the monitoring of the back EMF and the acceleration signals. Fig. 6. Experimental setup to measure output accelerations of a resonant electromagnetic actuator along with the sweeping frequency and evaluate the proposed sensorless drive scheme. A. Automatic Tracking of Resonant Frequency Fig. 7 compares the system electrical output consisting of the input driving signal and the back EMF with the acceleration wave measured from the accelerometer. An actuator sample (SEMCO LRA1036, resonant frequency: 175 Hz) was driven by the proposed method for approximately 40ms. After the driving period, the actuator was made to be in the high impedance state so as to solely observe its back EMF. The top graph in Fig. 7 shows the voltage signals measured from the actuator, and the bottom graph in Fig. 7 shows the vibration acceleration wave. As in Eq (12), both measured signals had a similarity in the periodicity. Given this preliminary test result, we highlight that the proposed method is capable of driving the actuator without the requirement to know the preset natural frequency. 11 Preprint Fig. 7. Measured voltage (i.e., combined input voltage and output back electromotive force) between the terminals of the actuator (LRA1036, SEMCO) (top) compared with acceleration signals (bottom) while driven by our proposed method. To validate the automatic resonance tracking accuracy, we tested two resonant electromagnetic actuators samples that have different mechanical natural frequencies (sample S1: LG LRA-SMA which has a high resonant frequency; sample S2: LG DMA which has a low resonant frequency [23]). The size dimension of both samples is 10mm x 10mm x 3mm (300 mm3). Fig. 8a and 8b show the frequency response of each sample actuator which was measured through inputting sweeping sinusoidal signals of 2.5Vpeak to the actuator (i.e., (a) S1, (b) S2). Fig. 8c and 8d show automatically detected resonant frequencies of S1 and S2, respectively. Compared with the ground truth from the frequency response measurements, the errors of tracked resonant frequencies of S1 and S2 were 0.9 Hz and 0.72Hz, respectively. Given the operable frequency ranges of S1 and S2 within which the detected frequencies fall (i.e., S1: 1/(4.785ms)=209Hz, S2: 1/(6.604ms)=151.42Hz), the result demonstrates the performance of the proposed sensorless scheme in the simultaneous resonance detection and actuation. By contrast with the traditional drive scheme for resonant electromagnetic actuators that operates with a given, fixed resonant frequency (described on a datasheet provided by a manufacturer), our proposed method does not demand to know the predefined frequency and is able to adaptively cope with situations where the natural frequency shift occurs, in turn maintaining an adequate level of the vibration force. 12 Preprint Fig. 8. Sensorless drives and detected damped natural frequency in accordance with the resonant electromagnetic actuators, each with a different resonant frequency (S1: LRA-SMA, LG (high mechanical resonance), LG; S2: DMA, LG (low mechanical resonance)): (a) frequency response and (c) measured input/output signals of S1, (b) frequency response and (d) measured input/output signals of S2. B. Residual Vibration Control As part of an effort to extend use cases of the proposed sensorless resonance tracking method, we conducted further experiments to test its capability in reducing the residual vibration. It is evident that residual oscillation produces differences between a designed vibrotactile pattern and a physical vibration feedback, in turn affecting a person's haptic perception. Fig. 9a shows the residual vibration produced from the actuation of the resonant electromagnetic actuator (SEMCO LRA1036) without the residual vibration control. As observed, the residual wave (> 0.9s) was yielded after the actuation period of 1.28s. From the same settings except for the parameter isBreakingTriggered set to true (see Algorithm 1), activating the residual vibration control, the duration of the residual oscillation was dropped to 0.09s as shown in Fig. 9b. It is remarkable that the proposed method reduced the stopping time of residual waves to less than 10% of the original duration. This can help to bridge the gap between the designed and actual vibration patterns. 13 Preprint Fig. 9. Measured voltage between the terminals of the actuator (LRA1036, SEMCO) compared with acceleration signals while driven by our proposed method: (a) without, (b) with the proposed residual vibration control. To highlight an advantage of the residual vibration control, we applied the method to the creation of short haptic feedback for a virtual button to mimic the feedback from a physical button, one of interesting topics in tactile interaction for mobile devices (e.g., [27]). Fig. 10 compares the vibration pattern produced from the residual vibration control based on the sensorless resonance tracking (left in Fig. 10) with the pattern from the traditional method (right in Fig. 10). The duration of each actuation input signal driven to the actuator (SEMCO LRA1036) was approximately 30ms. The proposed method was capable of controlling the residual vibration, quickly dropping the vibration amplitude lower than the steady-state error margin (10% of the maximum acceleration, here, approximately 0.05G). Compared to the traditional actuation method without the residual vibration control, the proposed approach reduced the residual duration from 200ms to 65ms (by 67.5%). Fig. 10. Creation of short vibration feedback for a virtual button in mobile devices - conveying 'button click' sensation: the proposed method with the residual vibration control (left), the conventional drive (i.e., using a sine wave input at a fixed resonant frequency) (right). 14 Preprint VI. Conclusion In this paper, we have proposed a new sensorless drive technique to automatically track the resonant frequency of a resonant electromagnetic actuator which is designed for mobile devices. The method is grounded in the mathematical analysis of the actuation model, which derives the relation of the back electromotive force to the damped natural frequency of the moving mass inside an actuator. The developed drive system and algorithm are to drive the actuator and automatically track its unknown damped natural frequency through the back EMF estimation so as to generate vibration feedback. This does not require to use additional sensors. Conducted experiments have demonstrated the robustness of the proposed approach in automatic resonance tracking with a good accuracy result (lower than 0.5% errors). Given this, it is expected that the method can improve an actuator's lifetime by adaptively addressing the resonant frequency shift issue, and potentially increase yield rates during manufacturing processes (for example, decreasing the failure rate from drop tests [28]), helping to reduce production costs. Lastly, we have demonstrated the capability of the method in the residual vibration control for the creation of short and strong vibrotactile effects. We believe that the ability can help to create rich tactile feedback that can assist a person's cognitive process (e.g., pen writing [29]) or emotional self-regulation [30,31]. References 1. Brewster S, Chohan F, Brown L. Tactile Feedback for Mobile Interactions. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. New York, NY, USA: ACM; 2007. pp. 159–162. doi:10.1145/1240624.1240649 2. Hwang S-M, Lee H-J, Chung S-U, Hwang G-Y, Kang B-S. Development of solenoid- type vibrators used for mobile phones. IEEE Transactions on Magnetics. 2003;39: 3262–3264. doi:10.1109/TMAG.2003.816742 3. Poupyrev I, Maruyama S, Rekimoto J. Ambient Touch: Designing Tactile Interfaces for Handheld Devices. Proceedings of the 15th Annual ACM Symposium on User Interface Software and Technology. New York, NY, USA: ACM; 2002. pp. 51–60. doi:10.1145/571985.571993 4. Choi S, Kuchenbecker KJ. Vibrotactile Display: Perception, Technology, and Applications. Proceedings of the IEEE. 2013;101: 2093–2104. doi:10.1109/JPROC.2012.2221071 5. Halmai A, Lukács A. New Linear-Electromagnetic Actuator Used for Cellular Phones. Periodica Polytechnica Engineering Mechanical Engineering. 2007;51: 19–22. doi:http://dx.doi.org/10.3311/pp.me.2007-1.03 6. Kleinman J. There's a New Way to Force Restart on the iPhone 8. 10 Dec 2017. Available: https://lifehacker.com/theres-a-new-way-to-force-restart-on-the-iphone-8- heres-1819404928 7. Johnson KO. The roles and functions of cutaneous mechanoreceptors. Current Opinion in Neurobiology. 2001;11: 455–461. doi:10.1016/S0959-4388(00)00234-8 15 Preprint 8. Verrillo RT. Temporal Summation in Vibrotactile Sensitivity. The Journal of the Acoustical Society of America. 1965;37: 843–846. doi:10.1121/1.1909458 9. Nam J, Oh H, Jang G. Externally Leveraged Resonant Piezoelectric Actuator With Fast Response Time for Smart Devices. IEEE/ASME Transactions on Mechatronics. 2016;21: 2764–2772. doi:10.1109/TMECH.2016.2590564 10. Cho Y-J, Yang T, Kwon D-S. A New Miniature Smart Actuator based on Piezoelectric material and Solenoid for Mobile Devices. Proceedings of the 5th International Conference on the Advanced Mechatronics, ICAM. 2010. pp. 615–620. Available: http://ci.nii.ac.jp/naid/110009957786 11. Kwon D-S, Yang T-H, Cho Y-J. Mechatronics technology in mobile devices. Industrial Electronics Magazine, IEEE. 2010;4: 36–41. Available: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5486392 12. Hayward J. Haptics 2017-2027: Technologies, Markets and Players. IDTechEx; Available: https://www.idtechex.com/research/reports/haptics-2017-2027-technologies- markets-and-players-000532.asp 13. Wakiwaka H, Ezawa M, Kato H, Yoshimura W, Ito H, Fukuda N, et al. Development of high-performance millimeter-size vibrator for cellular phones. Journal of Magnetism and Magnetic Materials. 2002;242–245, Part 2: 1195–1197. doi:10.1016/S0304- 8853(01)01300-2 14. Mangussi F, Zanette DH. Internal Resonance in a Vibrating Beam: A Zoo of Nonlinear Resonance Peaks. PLOS ONE. 2016;11: e0162365. doi:10.1371/journal.pone.0162365 15. Cutforth CF, Pao LY. Adaptive input shaping for maneuvering flexible structures. Automatica. 2004;40: 685–693. doi:10.1016/j.automatica.2003.11.013 16. Samoylenko N, Han Q, Jatskevich J. Dynamic Performance of Brushless DC Motors With Unbalanced Hall Sensors. IEEE Transactions on Energy Conversion. 2008;23: 752–763. doi:10.1109/TEC.2008.921555 17. Genduso F, Miceli R, Rando C, Galluzzo GR. Back EMF Sensorless-Control Algorithm for High-Dynamic Performance PMSM. IEEE Transactions on Industrial Electronics. 2010;57: 2092–2100. doi:10.1109/TIE.2009.2034182 18. Chappuis B, Gavin S, Rigazzi L, Carpita M. Speed Control of a Multiphase Active Way Linear Motor Based on Back EMF Estimation. IEEE Transactions on Industrial Electronics. 2015;62: 7299–7308. doi:10.1109/TIE.2015.2463762 19. Asai Y, Hirata K, Ota T. Amplitude Control Method of Linear Resonant Actuator by Load Estimation From the Back-EMF. IEEE Transactions on Magnetics. 2013;49: 2253–2256. doi:10.1109/TMAG.2013.2243419 20. Ozturk SB, Toliyat HA. Direct Torque and Indirect Flux Control of Brushless DC Motor. IEEE/ASME Transactions on Mechatronics. 2011;16: 351–360. doi:10.1109/TMECH.2010.2043742 21. Hyde JM, Cutkosky MR. Contact transition control: an experimental study. [1993] Proceedings IEEE International Conference on Robotics and Automation. 1993. pp. 363–368 vol.1. doi:10.1109/ROBOT.1993.292008 16 Preprint 22. Kweon SD, Park IO, Son YH, Choi J, Oh HY. Linear vibration motor using resonance frequency. US Patent, patent number: US7358633 B2.. US7358633 B2, 2008. Available: http://www.google.com/patents/US7358633 23. An S. Dual mode vibrator. US Patent, patent number: US8461969 B2. US8461969 B2, 2013. Available: http://www.google.com/patents/US8461969 24. Murakami K, Tsuruno R, Genda E. Multiple illuminated paper textures for drawing strokes. Computer Graphics International 2005. 2005. pp. 156–161. doi:10.1109/CGI.2005.1500408 25. Bhiladvala RB, Wang ZJ. Effect of fluids on the Q factor and resonance frequency of oscillating micrometer and nanometer scale beams. Phys Rev E. 2004;69: 036307. doi:10.1103/PhysRevE.69.036307 26. Eringen AC. On differential equations of nonlocal elasticity and solutions of screw dislocation and surface waves. Journal of Applied Physics. 1983;54: 4703–4710. doi:10.1063/1.332803 27. Kaaresoja T, Brewster S, Lantz V. Towards the Temporally Perfect Virtual Button: Touch-Feedback Simultaneity and Perceived Quality in Mobile Touchscreen Press Interactions. ACM Trans Appl Percept. 2014;11: 9:1–9:25. doi:10.1145/2611387 28. Choi B, Kwon J, Jeon Y, Lee MG. Development of Novel Platform to Predict the Mechanical Damage of a Miniature Mobile Haptic Actuator. Micromachines. 2017;8: 156. doi:10.3390/mi8050156 29. Cho Y, Bianchi A, Marquardt N, Bianchi-Berthouze N. RealPen: Providing Realism in Handwriting Tasks on Touch Surfaces Using Auditory-Tactile Feedback. Proceedings of the 29th Annual Symposium on User Interface Software and Technology. New York, NY, USA: ACM; 2016. pp. 195–205. doi:10.1145/2984511.2984550 30. Smith J, MacLean K. Communicating emotion through a haptic link: Design space and methodology. International Journal of Human-Computer Studies. 2007;65: 376–387. doi:10.1016/j.ijhcs.2006.11.006 31. Meule A, Fath K, Real RG, Sütterlin S, Vögele C, Kübler A. Quality of life, emotion regulation, and heart rate variability in individuals with intellectual disabilities and concomitant impaired vision. Psych Well-Being. 2013;3: 1. doi:10.1186/2211-1522-3-1 17 Preprint Supporting Information Figure legends Fig. 1. Schematic mechanical structure of the resonant electromagnetic actuator. Fig. 2. Measured the shifted resonant frequency of the resonant electromagnetic actuator (Product: LRA – SEMCO 1036) along with the different peak voltage of input signal on the frequency response: (a) 175.5Hz - 1.0Vpeak, (b)174.1Hz - 1.2Vpeak, (c)173.4Hz - 1.5Vpeak. Fig. 3. Combined electrical circuit and mechanical model of the actuator embedded in mobile devices. Fig. 4. Schematic graph for the periodic monitoring of the velocity of the inner mass of the resonant electromagnetic actuator estimated by the back electromotive force. Fig. 5. Block diagram of the proposed sensorless resonant electromagnetic actuator drive system. Fig. 6. Experimental setup to measure output accelerations of a resonant electromagnetic actuator along with the sweeping frequency and evaluate the proposed sensorless drive scheme. Fig. 7. Measured voltage (i.e., combined input voltage and output back electromotive force) between the terminals of the actuator (LRA1036, SEMCO) (top) compared with acceleration signals (bottom) while driven by our proposed method. Fig. 8. Sensorless drives and detected damped natural frequency in accordance with the resonant electromagnetic actuators, each with a different resonant frequency (S1: LRA-SMA, LG (high mechanical resonance), LG; S2: DMA, LG (low mechanical resonance)): (a) frequency response and (c) measured input/output signals of S1, (b) frequency response and (d) measured input/output signals of S2. Fig. 9. Measured voltage between the terminals of the actuator (LRA1036, SEMCO) compared with acceleration signals while driven by our proposed method: (a) without, (b) with the proposed residual vibration control. Fig. 10. Creation of short vibration feedback for a virtual button in mobile devices - conveying 'button click' sensation: the proposed method with the residual vibration control (left), the conventional drive (i.e., using a sine wave input at a fixed resonant frequency) (right). Algorithm legends Algorithm 1. Auto-tracking of resonant frequency & Reduction of residual vibration. 18
1811.03347	1	1811	2018-11-08T10:51:23	Multilevel nonvolatile optoelectronic memory based on memristive plasmonic tunnel junctions	[ "physics.app-ph", "physics.optics" ]	Highly efficient information processing in brain is based on processing and memory components called synapses, whose output is dependent on the history of the signals passed through them. Here we have developed an artificial synapse with both electrical and optical memory effects using reactive tunnel junctions based on plasmonic nanorods. In an electronic realization, the electrons tunneled into plasmonic nanorods under low bias voltage are harvested to write information into the tunnel junctions via hot-electron-mediated chemical reactions with the environment. In an optical realization, the information can also be written optically by external light illumination to excite hot electrons in plasmonic nanorods. The stored information is non-volatile and can be read in both realizations either electrically or optically by measuring the resistance or inelastic-tunnelling-induced light emission, respectively. These memristive light-emitting plasmonic tunnel junctions can be used as memory, logic units or artificial synapses in future optoelectronic or neuromorphic information systems.	physics.app-ph	physics	Multilevel nonvolatile optoelectronic memory based on memristive plasmonic tunnel junctions Pan Wang, Mazhar E. Nasir, Alexey V. Krasavin, Wayne Dickson and Anatoly V. Zayats Department of Physics and London Centre for Nanotechnology, King's College London, Strand, London WC2R 2LS, UK Correspondence to: [email protected], [email protected] Highly efficient information processing in brain is based on processing and memory components called synapses, whose output is dependent on the history of the signals passed through them. Here we have developed an artificial synapse with both electrical and optical memory effects using reactive tunnel junctions based on plasmonic nanorods. In an electronic realization, the electrons tunneled into plasmonic nanorods under low bias voltage are harvested to write information into the tunnel junctions via hot-electron-mediated chemical reactions with the environment. In an optical realization, the information can also be written optically by external light illumination to excite hot electrons in plasmonic nanorods. The stored information is non-volatile and can be read in both realizations either electrically or optically by measuring the resistance or inelastic- tunnelling-induced light emission, respectively. These memristive light-emitting plasmonic tunnel junctions can be used as memory, logic units or artificial synapses in future optoelectronic or neuromorphic information systems. 1 Highly efficient information processing in brain utilises multilevel (as opposed to binary used in modern digital computers) logic components called synapses. A memristive device (or memristor) is a resistive electrical element with resistance depending on the history of the applied electrical signals1,2, and can, therefore, be used as a memory element for the storage of information or as an artificial synapse to emulate biological synapses. Since the first experimental realization based on a metal/oxide/metal (Pt/TiO2/Pt) structure in 2008 [2], memristors have attracted extensive interest due to their important application in next-generation non-volatile memory, signal processing, reconfigurable logic devices, and neuromorphic computing3‒14. By integrating light-emitting or plasmonic properties with memristive devices, optical memristors have also been demonstrated15‒18. Most of the memristive devices are based on metal- insulator-metal structures to achieve resistance switching by dynamically configuring the insulating layer (e.g., the formation/annihilation of nanoscale conductive filament), showing low operation voltages (several volts), short set/reset times (<100 ns), and good endurance. The oxide-based insulating layer is typically limited to thicknesses greater than 3 nm6‒9,12‒18. Further reduction of the insulating layer thickness is demanded in order to reduce the device size, operating voltage and energy consumption3,5. However, in metal-insulator-metal structures with insulator thickness of less than 2 nm, the occurrence of quantum-mechanical tunnelling effect can cause electron leakage through the insulating layer, which is believed to be a detrimental effect for further downscaling of electronic components such as transistors in integrated circuits. Here, we demonstrate optoelectronic memristive devices by taking advantage of the electron tunnelling effect. Based on metal-polymer-metal tunnel junctions, we show the simultaneous multistate switching of the resistance and built-in light emission of the junctions, which is realized both electrically and optically by programming the junctions via hot-electron-mediated chemical reactions controlled by 2 the environment. Light-emitting tunnel junction (Fig. 1a), an optoelectronic analog of a biological synapse, was constructed based on a plasmonic nanorod. During the tunnelling process (Fig. 1b), the inelastically tunneled electrons excite plasmons in the nanorod which can subsequently decay radiatively into photons, while those electrons that tunnel elastically, generate hot electrons in the tips of nanorods which can be harvested for the multilevel writing of the junction state. The information stored is non-volatile and can be read both electrically and optically by interrogating the resistance and emission intensity. Optical coding of the tunnel junctions is also possible using the hot-electrons generated in the tunnel junctions by an external illumination. Controlling a gas environment of the tunnel junctions can be used to program the memristor response. Experimentally, tunnel junctions were constructed in a plasmonic nanorod array (Fig. 1c), which was fabricated by electrodeposition of Au into porous alumina templates (see Supplementary Section 1). Figure 1d presents the cross-sectional view of a plasmonic nanorod array, clearly showing the Au nanorods embedded in the alumina template. The diameter, length, and separation of the nanorods are approximately 65, 480, and 105 nm, respectively. Metal-polymer-metal tunnel junctions were constructed on the surface of the nanorod metamaterial by using a monolayer of poly-L-histidine (PLH) as the tunnel barrier and 'storage' layer (used as a reactant to store information via reconfigurable chemical reactions), and a droplet of eutectic gallium indium (EGaIn) as the top electrode (see Methods and Supplementary Section 2 for details). Each Au nanorod forms a tunnel junction with the top EGaIn contact (Fig. 1a), creating an array of tunnel junctions (Supplementary Section 3) with density determined by the density of Au nanorod array on the order of ~1010 cm-2, which is close to the density of synapses in human brain (~1014 in total). Nonlinear character of current-voltage characteristic (Fig. 1e) confirms the tunnelling of electrons through the metal-polymer-metal junctions19. Upon the application of a forward bias, light 3 emission was observed from the substrate side of the device, which is due to the radiative decay of plasmons excited in the nanorod metamaterial (Fig. 1b)20‒25. The recorded emission spectra (having a linewidth of ~200 nm) as a function of the applied bias are shown in Fig. 1f. With the increase of the bias, the emission intensity increases gradually, accompanied by a blue-shift of the emission peaks following the quantum cut-off law ℎ𝜈(cid:3043)(cid:3035)(cid:3042)(cid:3047) ≤ 𝑒𝑉(cid:3029) [20]. Figure 1 Memristor structure and light emission properties. a, Schematic diagram of the memristive light-emitting tunnel junction (right), which is an optoelectronic analogue to a synapse (left). b, An energy level diagram of the tunnel junction with a bias of Vb. c, Schematic diagram of the plasmonic nanorod array used to realize multiple tunnel junctions. d, Cross-sectional SEM view of a nanorod array. e, Measured current-voltage characteristic of a tunnelling device fabricated using the array shown in d. f, Measured emission spectra of the tunnelling device as a function of the applied forward bias. 4 During the tunnelling process, the majority of electrons (~99%) tunnel elastically (Fig. 1b)20‒25, appearing as hot electrons26,27 in the tips of Au nanorods, which can be used for programming the state of the tunnel junctions via hot-electron-activated chemical reactions28,29. To use the hot-electron effects, the tunnelling device was put into a gas chamber under a bias of 2.5 V, with the tunnelling current and emission spectrum monitored simultaneously. The device was first stabilized in 2% H2 in N2, then, upon switching of a chamber environment to air, the tunnelling current decreased gradually down to two thirds of the original value (Fig. 2a). At the same time, the integrated light emission intensity increased gradually to twice the original value. The changes in the tunnelling current and emission intensity reflect a change in the junction state, which is due to the oxidization of the tunnel junctions by oxygen molecules in air mediated by hot electrons as a PLH monolayer undergoes oxidative dehydrogenation and coupling reactions25. The resistance and emission intensity of the tunnelling device depends on the total number of the tunneled electrons (Fig. 2b) since the state of tunnel junctions is dependent on the history of the tunnelling process, particularly on how many electrons have traversed the junctions before, demonstrating the memory effect similar to biological synapses. During the reaction, the device was brought from a low resistance state (~20 Ω) to a high resistance state (~29 Ω), with a simultaneous change in the light emission from a low intensity to high intensity state (~80% increase in intensity). In this case, the written state of the tunnel junctions can be read out both electrically and optically, which is attractive for use as memory devices or artificial synapses, not only in electronic but also in optoelectronic systems. Moreover, compared with the existing optical memristors which require external light sources for the optical readout16‒18,30, the plasmonic tunnel junctions have nanoscale built-in plasmonic light sources, providing advantages for the dramatic reduction in device size and power consumption. Normally, the emission 5 intensity changes linearly with the tunnelling current, however, the emission intensity shows an opposite trend to that of the current during the reaction. This can be understood considering the evolution of the estimated inelastic tunnelling efficiency during the programming process (Fig. 2c, see Supplementary Section 4 for details). During the reaction of the tunnel junctions with oxygen molecules, the inelastic tunnelling efficiency increases gradually, resulting in the increased light emission intensity despite the gradual decrease of the tunnelling current. Figure 2 Hot-electron-mediated programming of electrical and optical properties. a, Time-dependent evolution of the tunnelling current and integrated emission intensity during the hot-electron-mediated reaction of tunnel junctions with oxygen (environment switched from 2% H2 gas to air). b,c, Dependence of the resistance and integrated emission intensity (b), and the inelastic tunnelling efficiency (c) on the total number of the tunneled electrons. d, Time-dependent evolution of the tunnelling current and integrated emission intensity during the hot-electron-mediated reaction of oxidized tunnel junctions with hydrogen (environment switched from air to 2% H2). e,f, Dependence of the resistance 6 and integrated emission intensity (e), and the inelastic tunnelling efficiency (f) on the total number of the tunneled electrons. Vb is fixed at 2.5 V in all the measurements. The tunnelling device can be programmed back to the original status by introducing hydrogen molecules into the cell via the hot-electron-mediated reduction of the oxidized tunnel junctions (Fig. 2d‒ f). The resistance, integrated emission intensity, and inelastic tunnelling efficiency (Fig. 2e,f) decreased gradually back to the original value with the continuous supply of the hot electrons and hydrogen molecules, highlighting the ability to reversibly programme the tunnelling device. The dynamics of the light-emitting reactive tunnel junctions can be associated with long-term potentiation/depression processes of synapses in a biological neural network. Different from ferroelectric or magnetic tunnel junction based memristors exploiting tunnel electroresistance or magnetoresistance effects10,11, the light- emitting plasmonic tunnel junctions exploit elastically tunneled electrons for the writing of information and inelastically tunneled electrons for the optical readout, providing programmability of the response and sensitivity to the environment. As discussed above, the state of the tunnel junctions is highly dependent on the number of the tunneled electrons and the environment. By controlling the supply of hot electrons or reactants (oxygen or hydrogen), the tunnelling device can be latched to different intermediate states. For example, as shown in Fig. 3a, the resistance of the device was switched from ~20 Ω (level L) to 22 (level 1), 26 (level 2), and 29 Ω (level H), respectively, by controllably introducing oxygen molecules into the chamber for the oxidization of the tunnel junctions. When the required state was achieved, pure nitrogen (employed as a nonreactive environment) was introduced into the chamber to remove oxygen molecules to latch the state of the junctions. Under the nonreactive environment of nitrogen, the state of the junctions was maintained 7 when the bias was switched off, showing the non-volatility. Accordingly, the light emission from the device was also latched to different intermediate levels (Fig. 3b). Benefitted from the programming mechanism of the reactive tunnel junctions, the states of the junctions may, in principle, be controlled on single electron or molecule level. Instead of carrying out computations based on binary in digital chips, the artificial synapses based on reactive tunnel junctions work in an analog way like neurons in brain that activate in various way depending on the type and number of ions that flow across a synapse. Figure 3 Multilevel programming of the electrical and optical properties. a, Continuous switching of the resistance of the tunnelling device by controlling the environment. The device was first stabilized in 2% H2 (Vb = 2.5 V), then the resistance was switched by the introduction of oxygen molecules and latched by replacing oxygen environment with nitrogen. b, Corresponding latched emission spectra of the tunnelling device. 8 Apart from the electrical programming, the state of the tunnel junctions can be programmed optically. Under external light illumination of the nanorod metamaterial from the substrate side, the plasmonic modes in the metamaterial are excited. Figure 4a shows simulated electric field and current distributions in the unit cell of the metamaterial for the illumination wavelength of 600 nm (see Methods for details). The plasmonic excitation exists across the whole nanorod length and hot electrons are generated in both tips of the Au nanorods, which can be used for the activation of chemical reactions in the tunnel junctions. In order to demonstrate this (Fig. 4b), the device was first stabilized in 2% H2 in N2 under 2.5 V (period 1). The state of the junctions was unchanged when the environment was switched to air under zero bias (period 2) due to the lack of hot electrons for the reaction (under applied bias the gradual rise of the tunnel resistance to level H was observed as expected (period 3)). The state was programmed back to the low resistance level (period 4) by introducing 2% H2 back into the chamber under applied bias. However, when the environment was switched to air under zero bias but the metamaterial was illuminated by a white light (period 5), the state of the junctions was programmed to the high resistance level (confirmed by the stable resistance after the removal of illumination under a bias of 2.5 V (period 6)). The spectra of latched light emission correspond to the level L and optically switched level H' agree well with the emission spectra from the electrically programmed low and high resistance levels (cf., Fig. 4c and Fig. 3b). The ability of optical coding provides an alternative choice for the writing of information with advantages such as wireless and wavelength-dependent control. 9 Figure 4 Optical programming of the tunnelling device. a, Simulated electric field E and current J distributions in the unit cell of the metamaterial for the illumination wavelength of λ = 600 nm. b, Evolution of the resistance of the tunnelling device under different conditions as indicated at the bottom. White light illumination in the spectral range 500-750 nm and power density ~0.03 W cm-2 was used during period 5. c, Emission spectra of the tunnelling device measured at the resistance level L and the optically programmed resistance level H'. In conclusion, we have investigated the electrical and optical memory effects in reactive tunnel junctions. The high density of tunnel junctions and scalability provided by the plasmonic nanorod array make the proposed approach an attractive platform for the construction of 'brain on a chip' and neuromorphic computing devices. Flexibility of the approach can be further exploited using new chemical reactions and other switching mechanisms, such as formation/annihilation of conductive filament, to further control the operation. The reactive plasmonic tunnel junctions can be integrated directly with plasmonic or silicon waveguides for the application as memory, logic units, and artificial synapses in optoelectronic systems, scaled down to single junctions if required. The light-emitting reactive tunnel junctions have the potential to become important building blocks of memories, logic units, or artificial synapses in optoelectronic or neuromorphic computing systems. 10 Acknowledgement This work has been funded in part by the Engineering and Physical Sciences Research Council (UK) and the European Research Council iPLASMM project (321268). A.V.Z. acknowledges support from the Royal Society and the Wolfson Foundation. Author contributions A.V.Z. and P.W. conceived the study. P.W. constructed the experiment, performed the measurement and analysed the data. M.E.N. and W.D. fabricated the nanorod metamaterials. A.V.K. performed numerical simulations. All the authors discussed the results and co-wrote the paper. Additional information Correspondence and requests for materials should be addressed to A.V.Z and P.W. Competing financial interests The authors declare no competing financial interests. 11 References 1. Chua L. O. Memristor-the missing circuit element. IEEE Trans. Circuit Theory 18, 507‒519 (1971). 2. Strukov D. B., Snider G. S., Stewart D. R. & Williams R. S. The missing memristor found. Nature 453, 80‒83 (2008). 3. Yang J. J., Strukov D. B. & Stewart D. R. Memristive devices for computing. Nat. Nanotechnol. 8, 13‒24 (2013). 4. Parkin, S. & Yang, S. -H. Memory on the racetrack. Nat. Nanotechnol. 10, 195‒198 (2015). 5. Zidan M. A., Strachan J. P. & Lu W. D. The future of electronics based on memristive systems. Nat. Electron. 1, 22‒29 (2018). 6. Yang J. J. et al. Memristive switching mechanism for metal/oxide/metal nanodevices. Nat. Nanotechnol. 3, 429‒433 (2008). 7. Li C. et al. Analogue signal and image processing with large memristor crossbars. Nat. Electron. 1, 52‒59 (2018). 8. Borghetti J. et al. 'Memristive' switches enable 'stateful' logic operations via material implication. Nature 464, 873‒876 (2010). 9. Jo S. H. et al. Nanoscale memristor device as synapse in neuromorphic systems. Nano Lett. 10, 1297‒ 1301 (2010). 10. Krzysteczko P., Münchenberger J., Schäfers M., Reiss G., Thomas A. The memristive magnetic tunnel junction as a nanoscopic synapse-neuron system. Adv. Mater. 24, 762‒766 (2012). 11. Kim D. J. et al. Ferroelectric tunnel memristor. Nano Lett. 12, 5697‒5702 (2012). 12 12. Wu C. X., Kim T. W., Choi H. Y., Strukov D. B. & Yang J. J. Flexible three-dimensional artificial synapse networks with correlated learning and trainable memory capability. Nat. Commun. 8, 752 (2017). 13. Schneider M. L. et al. Ultralow power artificial synapses using nanotextured magnetic Josephson junctions. Sci. Adv. 4, e1701329 (2018). 14. Choi S. et al. SiGe epitaxial memory for neuromorphic computing with reproducible high performance based on engineered dislocations. Nat. Mater. 17, 335‒340 (2018). 15. He C. L. et al. Tunable electroluminescence in planar graphene/SiO2 memristors. Adv. Mater. 25, 5593‒5598 (2013). 16. Emboras A. et al. Nanoscale plasmonic memristor with optical readout functionality. Nano Lett. 13, 6151‒6155 (2013). 17. Emboras A. et al. Atomic scale plasmonic switch. Nano Lett. 16, 709‒714 (2016). 18. Cheng Z. G., Ríos C., Pernice W. H. P., Wright C. D. & Bhaskaran H. On-chip photonics synapse. Sci. Adv. 3, e1700160 (2017). 19. Simmons J. G. Generalized formula for the electric tunnel effect between similar electrodes separated by a thin insulating film. J. Appl. Phys. 34, 1793 -- 1803 (1963). 20. Lambe J. & McCarthy S. L. Light emission from inelastic electron tunneling. Phys. Rev. Lett. 37, 923 -- 925 (1976). 21. Kern J. et al. Electrically driven optical antennas. Nat. Photon. 9, 582 -- 586 (2015). 22. Parzefall M. et al. Antenna-coupled photon emission from hexagonal boron nitride tunnel junctions. Nat. Nanotechnol. 10, 1058 -- 1063 (2015). 13 23. Du W. et al. On-chip molecular electronic plasmon sources based on self-assembled monolayer. Nat. Photon. 10, 274 -- 280 (2016). 24. Du W., Wang T., Chu H. -- S. & Nijhuis C. A. Highly efficient on-chip direct electronic-plasmonic transducers. Nat. Photon. 11, 623 -- 627 (2017). 25. Wang P., Krasavin A. V., Nasir M. E., Dickson W. & Zayats A. V. Reactive tunnel junctions in electrically driven plasmonic nanorod metamaterials. Nat. Nanotechnol. 13, 159‒164 (2018). 26. Brongersma M. L., Halas N. J. & Nordlander P. Plasmon-induced hot carrier science and technology. Nat. Nanotech. 10, 25 -- 34 (2015). 27. Harutyunyan H. et al. Anomalous ultrafast dynamics of hot plasmonic electrons in nanostructures with hot spots. Nat. Nanotechnol. 10, 770 -- 774 (2015). 28. Mubeen S. et al. An autonomous photosynthetic device in which all charge carriers derive from surface plasmons. Nat. Nanotechnol. 8, 247 -- 251 (2013). 29. Zhai Y. M. et al. Polyvinylpyrrolidone-induced anisotropic growth of gold nanoprisms in plasmon- driven synthesis. Nat. Mater. 15, 889 -- 895 (2016). 30. Hoessbacher C. et al. The plasmonic memristor: a latching optical switch. Optica 1, 198‒202 (2014). 14 Methods Fabrication. Plasmonic nanorod metamaterials were fabricated by electrodeposition of Au into substrate- supported porous alumina templates25,31. Metamaterial-based light emitting tunnel junctions were fabricated as follows: firstly, a nanorod metamaterial was chemically etched in a 3.5% H3PO4 solution at 35 ℃ to make the surrounding Al2O3 matrix slightly lower than the nanorod tips; secondly, the metamaterial with exposed nanorod tips were functionalized with a monolayer of PLH (Mw = 5,000 -- 25,000, Sigma-Aldrich) via self-assembly; finally, a droplet of EGaIn (≥ 99.99% trace metals basis, Sigma-Aldrich) was added onto the surface of the metamaterial to form an array of metal-PLH-metal tunnel junctions. Numerical simulations. Numerical simulations of the near-field distributions of the electric field inside the nanorod metamaterial and the associated electric current in the nanorods were performed using a finite element method (Comsol Multiphysics software). The metamaterial was illuminated from the substrate side by a plane wave at an angle of incidence of 45o. The distribution of the nanorods in the metamaterial was approximated with a square array, which allowed to simulate the entire system modeling a unite cell with properly defined Floquet boundary conditions on the unit cell sides, determined by a phase delay acquired by the incident plane wave while travelling between the corresponding faces. To ensure the absence of back-reflection, a perfectly matched layer was implemented on the illumination side. At the opposite (EGaIn) side, this was not needed due to metallic nature of the latter. Experimentally measured data with a mean free path correction of 3 nm, related to the properties of electrochemically derived Au, were used for gold32, experimental tabulated data was also used for Al2O3 matrix33, SiO2 substrate34 and Ta2O5 adhesion layer35. Optical properties of EGaIn were approximated by the Drude model, while 15 refractive indices of gallium oxide, naturally appearing on the EGaIn surface, and PLH polymer were taken as non-dispersive in the studied wavelength range and equal to 1.89 and 1.565, respectively. References 31. Evans P. et al. Growth and properties of gold and nickel nanorods in thin film alumina. Nanotechnology 17, 5746 -- 5753 (2006). 32. Johnson P. B. & Christy R. W. Optical constants of the noble metals. Phys. Rev. B 6, 4370 -- 4379 (1972). 33. Malitson I. H. & Dodge M. J. Refractive index and birefringence of synthetic sapphire. J. Opt. Soc. Am. 62, 1405 (1972). 34. Malitson I. H. Interspecimen comparison of the refractive index of fused silica. J. Opt. Soc. Am. 55, 1205-1209 (1965). 35. Bright T. J. et al. Infrared optical properties of amorphous and nanocrystalline Ta2O5 thin films. J. Appl. Phys. 114, 083515 (2013). 16 Supplementary Information Multilevel nonvolatile optoelectronic memory based on memristive plasmonic tunnel junctions Pan Wang, Mazhar E. Nasir, Alexey V. Krasavin, Wayne Dickson and Anatoly V. Zayats* Department of Physics and London Centre for Nanotechnology, King's College London, Strand, London WC2R 2LS, UK Correspondence to: [email protected], [email protected] 17 S1. Fabrication of plasmonic nanorod metamaterials The plasmonic nanorod metamaterials were fabricated by electrodeposition of Au into substrate-supported porous alumina templates [31]. The substrate is a multilayered structure comprised of a glass slide (1 mm in thickness), a tantalum oxide adhesive layer (10 nm in thickness), and an Au film (7 nm in thickness) acting as a working electrode for the electrochemical reaction. An aluminum film (~500 nm in thickness) is then deposited onto the substrate by planar magnetron sputtering, which is subsequently anodized in oxalic acid (0.3 M) at 40 V to produce the porous alumina template by two-step anodization. The diameter, separation and ordering of the Au nanorods in the assembly are controlled by the conditions of anodization. The electrodeposition of Au is performed with a three-electrode system using a non-cyanide solution. The length of Au nanorods is controlled by the electrodeposition time. In this work, the length of Au nanorods is slightly shorter than the height of alumina template. The metamaterials were then washed several times in deionized water (DI water) and stored in 200 proof ethanol for future use. S2. Fabrication of metal-polymer-metal tunnel junctions Figure S1 shows the schematic diagram of the fabrication of metal-polymer-metal tunnel junctions based on a plasmonic nanorod metamaterial. In the first step, a wet chemical etching method was used to remove some of the alumina matrix to make the tips of Au nanorods slightly higher than the surrounding alumina. Briefly, the nanorod metamaterial stored in ethanol was firstly dried under N2 and then put into an aqueous solution of H3PO4 (3.5 %) at 35 ℃ to start the etching. The etching depth can be precisely controlled by the etching time. After the chemical etching, the metamaterial was washed several times in DI water. In the second step, the exposed Au nanorod tips were functionalized with a monolayer of poly-L- histidine (PLH, Mw 5,000-25,000, Sigma-Aldrich). Briefly, the etched nanorod metamaterial was 18 submerged into a PLH solution (1 mg/mL, pH ~5 -- 6) and incubated for 0.5 h. Due to the high affinity of PLH to Au surface and the positive charging of protonated PLH in solution, a monolayer of PLH self- assembled onto the exposed Au nanorod tips. The metamaterial was then washed several times in DI water to remove weakly bound PLH and dried under N2. Finally, a droplet of EGaIn was added onto the surface of the nanorod metamaterial to form an array of metal-PLH-metal tunnel junctions, and then two Au wires were connected to the bottom Au film and the EGaIn droplet for the application of bias across the tunnel junctions. Figure S1. Schematic diagram showing the steps for the fabrication of metal-polymer-metal tunnel junctions. 19 S3. Nanorod-metamaterial-based tunnelling device Figure S2 Nanorod-metamaterial-based tunnelling device. a, Schematic diagram of tunnelling junction array constructed on a Au nanorod metamaterial. b, Photograph of a nanorod-metamaterial-based tunnel device. S4. Estimation of inelastic tunnelling efficiency In the nanorod-metamaterial-based tunnel junctions, the light emission is due to the radiative decay of plasmonic modes excited by the inelastic tunnelling electrons. In this case, the relation between inelastic tunnelling efficiency (𝜂(cid:3036)(cid:3041)(cid:3032)(cid:3039) = Γ(cid:3036)(cid:3041)(cid:3032)(cid:3039)/Γ(cid:3047)(cid:3042)(cid:3047), where Γ(cid:3036)(cid:3041)(cid:3032)(cid:3039) and Γ(cid:3047)(cid:3042)(cid:3047) are inelastic and total tunnelling rates, respectively) and electron-to-photon conversion efficiency (𝜂(cid:3032)(cid:3039)(cid:2879)(cid:3043)(cid:3035) ) can be written as: 𝜂(cid:3032)(cid:3039)(cid:2879)(cid:3043)(cid:3035) = 𝜂(cid:3036)(cid:3041)(cid:3032)(cid:3039) ∙ 𝜂(cid:3028)(cid:3041)(cid:3047), where 𝜂(cid:3028)(cid:3041)(cid:3047) is the antenna radiation efficiency (𝜂(cid:3028)(cid:3041)(cid:3047) = 𝑃(cid:3045)(cid:3028)(cid:3031)/𝑃(cid:3047)(cid:3042)(cid:3047), defining how much power from the excited plasmonic modes is radiated in light). The electron-to-photon conversion efficiency can be estimated from the ratio of emitted photons (evaluated from the measured emission power, assuming all the emitted photons have the same wavelength of 850 nm) to injected electrons (evaluated from the measured tunnelling current under 2.5 V forward bias) during each period of measurement. The antenna radiation efficiency 𝜂(cid:3028)(cid:3041)(cid:3047) can be evaluated from the numerical simulations to be ~3.5 × 10-4 at 850 nm. 20 We can plot the evolution of inelastic tunnelling efficiency during the memristor programming process using the formula: 𝜂(cid:3036)(cid:3041)(cid:3032)(cid:3039) = 𝜂(cid:3032)(cid:3039)(cid:2879)(cid:3043)(cid:3035)(cid:3042)(cid:3047)/𝜂(cid:3028)(cid:3041)(cid:3047). 21
1810.10634	1	1810	2018-10-24T21:42:05	A frequency-preserving and time-invariant metamaterial-based nonlinear acoustic diode	[ "physics.app-ph" ]	We present the realization of an acoustic diode or rectifier, exploiting symmetry-breaking nonlinear effects like harmonic generation and wave mixing and the filtering capabilities of metamaterials. The essential difference and advantage compared with previous acoustic diode realizations is that the present is simultaneously a time invariant, frequency preserving and switchable device. This allows its application also as an on-off or amplitude-tuning switch. We evaluate its properties by means of a numerical study and demonstrate its feasibility in a preliminary experimental realization. This work may provide new opportunities for the practical realization of structural components with one-way wave propagation properties.	physics.app-ph	physics	A frequency-preserving and time-invariant metamaterial-based nonlinear acoustic diode A. S. Gliozzi,1, ∗ M. Miniaci,2, † A. O. Krushynska,3 B. Morvan,4 M. Scalerandi,1 N. M. Pugno,5, ‡ and F. Bosia3, § 1Department of Applied Science and Technology, Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy 2D. Guggenheim School of Aerospace Engineering, Georgia Institute of Technology, North Avenue, GA 30332 Atlanta, USA 3Department of Physics, University of Torino, Via Pietro Giuria 1, 10125 Torino, Italy 4University of Le Havre, Laboratoire Ondes et Milieux Complexes, UMR CNRS 6294, 75 Rue Bellot, 76600 Le Havre, France 5Laboratory of Bio-Inspired and Graphene Nanomechanics, Department of Civil, Environmental and Mechanical Engineering, Universit di Trento, via Mesiano, 77, I-38123 Trento, Italy (Dated: July 6, 2021) We present the realization of an acoustic diode or rectifier, exploiting symmetry-breaking nonlinear effects like harmonic generation and wave mixing and the filtering capabilities of metamaterials. The essential difference and advantage compared with previous acoustic diode realizations is that the present is simultaneously a time invariant, frequency preserving and switchable device. This allows its application also as an on-off or amplitude-tuning switch. We evaluate its properties by means of a numerical study and demonstrate its feasibility in a preliminary experimental realization. This work may provide new opportunities for the practical realization of structural components with one-way wave propagation properties. INTRODUCTION In acoustics, the invariance of the wave equation with respect to time inversion, also known as reciprocity, has been exploited in many applications, e.g. for wave fo- cusing in Time Reversal experiments [1]. However, reci- procity is not necessarily desirable in all cases, especially when the goal is to isolate a source from its echos. Re- moval of unwanted reflections could indeed find numerous applications, such as acoustic one-way mirrors to prevent an ultrasound source from being disturbed by reflected waves [2, 3], unidirectional sonic barriers to block envi- ronmental noise in a predefined direction [4], control of acoustic energy transmission in medical applications us- ing focused ultrasound [5], and energy harvesting [6]. To achieve this, researchers in the field of acoustics and ul- trasonics have drawn inspiration from electromagnetism, in the quest for a simple and efficient realization of an Acoustic Diode (AD) or rectifier. However, as illustrated by Maznev et al. [7], linear elastic systems cannot be exploited to create ADs or isolators because they do not violate the reciprocity principle, so that the symmetry needs to be broken, for instance by periodically varying the elastic properties in space and/or time or by means of the introduction of nonlinearity coupled with some other mechanism (e.g. attenuation) [7]. Examples of the first approach (periodically varying elastic properties in space and time) are provided by the- oretical and numerical studies of one-dimensional system described by the discrete nonlinear Schrodinger equa- tion with spatially varying coefficients embedded in a linear lattice [10], of continuous elastic systems with periodically-modulated elastic properties in space and time [8], or of a non-reciprocal active acoustic metama- terial [9]. Examples of the second approach (introduction of non- linearity), instead, include 1-D design of a "superlattice" structure coupled with a nonlinear elastic medium [2], later realized experimentally using a contrast agent mi- crobubble suspension to generate the nonlinearity [3], converting energy from the fundamental frequency to higher harmonics [7]. Since then, several experimental realizations of ADs or rectifiers based on different mech- anisms have been achieved. For instance, in [11], unidi- rectional transmission was obtained through mode con- version, using a sonic crystal, rather than elastic nonlin- earity. In [12], a mechanical energy switch and transis- tor are implemented by exploiting nonlinear dynamical effects of a granular crystal chain. To break the trans- mission symmetry, Ref. [13] proposed to use a subwave- length acoustic resonant ring cavity filled with a circu- lating fluid, splitting the degenerate azimuthal resonant modes, in analogy with the Zeeman effect in electromag- netism. In [14], a thin brass plate with single-sided peri- odical gratings immersed in water was shown to provide unidirectional transmission in a broad frequency range. Finally, in [15], a passive multi-port structure with asym- metric sound transmission between neighbouring ports was presented. Comprehensive reviews of these and other approaches can be found in [7, 16], in the latter with spe- cial reference to information processing in phononic com- puting, while the optimization of a rectifier efficiency in periodic mass -- spring lattices is discussed in [17]. Many of these approaches are based on designing pe- riodic structures, mainly phononic crystals and elastic metamaterials, which have attracted much attention for their wave manipulation capabilities, including negative refraction [18], frequency Band Gap (BG) formation [19 -- 21], wave filtering or focusing [22 -- 25], scattering free propagation [26] and acoustic cloaking [28]. Recent stud- ies have shown how structural instabilities induced in "static" mechanical metamaterials can be exploited to achieve highly nonlinear dynamic response that can be tailored to requirements [29, 30] and how weakly nonlin- ear monoatomic lattice chains can provide active control on elastic waves in phononic crystals [31]. These or other approaches can be exploited to generate the type of non- linearity required to violate spatial reciprocity in elas- tic wave propagation [32]. On the other hand, phononic crystals and metamaterials are ideal candidates to effi- ciently realise large BGs [33, 34] or to concentrate energy in to selected frequency ranges [25, 35]. In this paper, we propose the realization of an AD, based on the use of linear phononic crystals and elas- tic metamaterials, embedded between elastic nonlinear regions. The novelty of the device is that it is simultane- ously time -- invariant (in the sense that its physical prop- erties are not modified externally from the forward to the backward propagation direction [27]) and frequency preserving. Furthermore, besides its functionality as a diode, for other applications the device could be activated or deactivated at will, transforming it into a switch with the additional possibility to tune the amplitude of the output signal. These characteristics are in general not concurrently present in other AD designs that exploit nonlinearity to break the propagation symmetry and to transfer energy from the fundamental to the harmonics, with a frequency variation from input to output. The originality of our approach also resides in the exploita- tion of the combined effects of two different features of nonlinear elastic wave propagation, i.e. higher order har- monic generation and wave mixing, which allow to pre- serve the operating frequency taking place in two differ- ent zones separated by the periodic (filtering) structure. Wave mixing occurs when two longitudinal waves propa- gating through a nonlinear elastic zone interact and gen- erate another longitudinal wave with a frequency given by the difference (and sum) of the frequencies of the two original waves. MODEL AND METHOD The working principle of the AD proposed in this study is illustrated in Fig. 1 and can be described as follows: i) Propagation from left to right (LtR, Fig. 1a): an input signal is injected (from S1) into the device where it encounters a passband filter FB1 that selects a range of frequencies around f1. These waves then travel through a first nonlinear elastic zone, named NL1, where a sec- 2 ond frequency f2 = 3 2 f1 can be injected from the source S2. In this case, the presence of nonlinearity generates higher harmonics and the sum and difference frequencies (wave mixing), including f2 − f1 = f0 = f1 2 , which is a subharmonic of f1. The next portion of the device, FB2, is a low-pass filter, which eliminates frequencies above f0, and a second nonlinear zone, NL2, where the second harmonic f1 = 2f0 is generated. Finally, another pass- band filter (FB3) filters out f0 and the harmonics higher than f1, giving an overall output signal f1 at the same frequency of the input. ii) Propagation from right to left (RtL, Fig. 1b): in this case, the input signal at f1 travels through FB3 and through NL2 where higher harmonics are generated (but not f0), and where no wave mixing process takes place (this breaks spatial reciprocity). The next portion of the device, FB2, filters out the full signal, so that no signal propagates through NL1 and FB1, generat- ing no output from the device. Notice that the source S2 is present both in the forward and in the backward propagation (in this sense the AD is time invariant in its physical characteristics) and its role is to break spa- tial symmetry in the device. This mechanism allows us to overcome some of the difficulties in the practical realization encountered in other theoretical works that propose frequency-preserving ADs [36 -- 38]. The present model/configuration has been conceived for monochro- matic inputs, as usually done for nonlinearity-based ADs. More complicated designs can be considered by imposing a non-monochromatic wave injected by the source S2. However, this is beyond the scope of this work. RESULTS To verify the feasibility and functionality of the device, we will first simulate its behaviour numerically (Figs.2-4) and then discuss its experimental realization (Figs.5 and 6). Numerical verification In the numerical simulation (Fig.1c), we model the device as an Aluminum plate with mass density ρ1 = 2700 kg/m3, Young modulus E = 70 GPa, and Poisson ratio ν = 0.33 and in-plane dimensions L = 105 mm and d = 6.6 mm. The core of the device, in which reciprocity is broken, is composed by two nonlinear zones (NL1 and NL2 in Fig.1), separated by a metamaterial (FB2). The nonlinear sections NL1 and NL2 are realized by considering a zone of diffuse nonlinearity, and the nu- merical nonlinear parameters are set in order to produce about 10% of harmonics and subharmonics. These two nonlinear zones are placed between two filters made of metamaterials or phononic crystals, which confine the 3 FIG. 2. Fast Fourier Transforms of the signal recorded in the input (a, d), in the first cavity on the left (b, e) and at the output (c, f), for the two propagation directions (LtR and RtL in the first and second rows, respectively). tary Material [43] together with its dispersion character- istics. The scalability of the results is guaranteed by the fact that the geometry of the constituent elements can easily be tuned to shift the pass bands to the desired frequencies. The excitation signal (a sinusoidal wave) is uniformly applied at the left boundary of FB1 (for LtR propaga- tion) or at the right boundary of FB3 (for RtL propaga- tion). We assume reflecting condition at the boundary that are free from excitation. With this configuration we perform wave propaga- tion simulations to demonstrate the effective feasibility of the AD. For the LtR (RtL) propagation, we inject a monochromatic wave of frequency f1 = 600 kHz on the left (right) side of the device and the corresponding f2 = 900 kHz in the left cavity. The output signal is recorded on the left (right) side of the sample (T1 in Fig- 1c). Fig. 2 shows the Fast Fourier Transform (FFT) of the signals for LtR (a-c) and RtL (d-f) respectively. The signals are recorded at the input of the device (a,d), in the first cavity on the left (b,e) and at the output (c,f). While f1 propagates from LtR, no signal is detected at the receiver when the propagation is in the other direc- tion. The difference between the two cases (reported in the upper and lower parts of Fig. 2, respectively) lies in the generation in the left cavity of the frequency f0, which is the only component that can propagate from NL1 to NL2. Although any mechanism able to generate sub- harmonics [44, 45] of f1 can be appropriate, the mech- FIG. 1. Schematic representation of the basic concept of the proposed AD for (a) left to right and (b) right to left prop- agation, respectively. f1 and f2 = 3 2 f1 are the injected wave components, while f0 = 1 2 f1 is generated by wave mixing. The red barriers represent the frequency BGs, the gray zones are the nonlinear cavities (NL1 and NL2) where harmonic generation and wave mixing take place. The schematic rep- resentation of the numerical sample is represented in subplot (c). frequency components of the wavefield falling in their BGs, creating a sort of resonant cavity (also named left and right cavities in the following). The dimensions of these regions and of the nonlinear elements can be tai- lored to enhance the desired frequencies through reso- nance effects (f0 in the left, and f1 in the right cavity). A nonclassical nonlinear model [39 -- 41], implemented us- ing a Preisach-Mayergoyz [42] space representation, is adopted to simulate the nonlinear elastic response of these zones [43]. The structure of each metamaterial/phononic crystal part (FB1-FB3) is described in detail in the Supplemen- f0f12f1f2f02f0=f13f04f02f2=Injected=Propagating=Harmonics=Wave mixing......FrequencyMaterialf1f1NL 1NL 2FB 1FB 2FB3a)=S1=S2Left to Right propagation (LtR)NL 1NL 2FB 1FB 2FB 3f22f12f2=Injected=Propagating=Harmonics=Wave mixing......FrequencyMaterialf1f1=S1=S2b)Right to Left propagation (RtL)FB1FB2FB3S1S2T1dLc)NL2NL1Left cavityRight cavity010002000Frequency [kHz]020406080100120FFT Magnitude [dB]010002000Frequency [kHz]020406080100120FFT Magnitude [dB]Propagation left to right010002000Frequency [kHz]020406080100120FFT Magnitude [dB]InputOutputf1Left cavitya)b)c)f0f1f2f1010002000Frequency [kHz]020406080100120FFT Magnitude [dB]010002000Frequency [kHz]020406080100120FFT Magnitude [dB]Propagation right to left010002000Frequency [kHz]020406080100120FFT Magnitude [dB]f2Output2f2f1InputLeft cavityd)e)f)4 FIG. 3. Acoustic on-off switch. By switching on/off the source S2, the wave generated by S1 can/cannot propagate through the device. This is visible both in the a) time and b) frequency domain. The FFT performed over different time windows (highlighted with different colors in subplot (a)) shows the different frequency content of the propagating wave. anism based on wave mixing adopted here to generate f0 has several advantages. The first is that wave mixing is an extremely efficient way to produce sub-harmonics and no threshold mechanism seems to be at play. More- over, the source S2 can be tuned in order to decrease or increase the amplitude of the f0 component, and in the limit case to suppress it. Thus, the device can be used as an on-off or an amplitude-tuning switch. Two different simulations are presented to demonstrate these applications. In the first case, the source S2 (the pump) is switched on/off at regular time intervals and the corresponding output recorded (see Fig.3). As shown in Fig. 3a, the signal is prevented from propagating when the source S2 is switched off. This is also evident in the FFT analysis performed by windowing the time signal for the two dif- ferent cases (S2 on/off in Fig. 3b). This demonstrates the use of the AD as a switch. The same numerical experiment is then repeated at in- creasing amplitude of the pump (S2) while keeping the amplitude of S1 fixed. Since the amplitude of the f0 com- ponent (the subharmonic of the input) is proportional to the product of the two mixed frequency amplitudes, it is possible to change the output signal amplitude by tuning the amplitude of the pump (S2, in this case), as shown in Fig. 4. This generates the possibility to realize a switch with a variable amplitude output. Moreover, from a the- oretical point of view this opens the way to the possibility of considerably increasing the efficiency of the device by pumping energy from S2 and increasing the output am- plitude at will. From a practical point of view, a large amplification may be limited by spurious nonlinear ef- fects and by the large amount of energy required. This FIG. 4. Amplitude-tuning switch. (a) Increase in output am- plitude as a function of the pump amplitude S2, for constant input amplitude (S1); (b-g) Corresponding outputs signals. limitation could be partially overcome by finding a very efficient nonlinear system. Work is in progress on this aspect. Experimental realization The discussed design of the AD is quite general and can be realized with different nonlinearity types, filter- ing characteristics or optimized properties. We demon- strate its feasibility through the experimental realization of a prototype, representing the central part of the de- vice, which is responsible for the breaking of reciprocity. We use a 380 × 40 × 6 mm3 aluminium plate (ρ = 2700 kg/m3, E = 70 GPa and ν = 0.33) with a phononic crystal region representing the filtering barrier (FB2 in Fig.1). The phononic crystal is located between two re- gions that represent the left (with NL1) and right (with NL2) cavities in Fig. 1. FB2 consists of a 2D array of 4×8 cross-like cavities [47], fabricated using waterjet cut- ting, with a lattice parameter of a = 10 mm (see [43] for geometrical details). Dimensions have been designed so to suppress frequencies from 124 kHz to 175 kHz and 191 kHz to 236 kHz, in the propagation from one cavity to the other (see [43] for further details). It follows that the working frequency of this AD is f1 = 150 kHz, while the pump S2 needs to be set at a frequency f2 = 225 kHz. In this simplified realization, in the LtR propagation, the two sources (S1 and S2) are located in the same cavity on the left, while the receiver (T1)is situated in the right cavity, as shown in Fig. 5a. The nonlinearity is gener- ated in the two cavities, by superposing onto the plate a small object coupled with a drop of water [25]. The clapping of the surfaces, due to the action of the elastic wavefield propagating in the plate gives rise to typical nonlinear effects (i.e. harmonics and wavemixing). In the experiments, the emitting piezoelectric contact transducer was connected to an arbitrary waveform gen- 2004006008001000Frequency [kHz]020406080FFT Magnitude [dB]020040060080010001200Time [ s]-1.5-1-0.500.511.5Velocity Normalisedb)OFFa)ON05001000Time [ s]-0.05-0.0200.020.0500.10.20.30.40.50.60.70.8Amplitude of the source S200.020.040.06Output Amp (at f1)05001000Time [ s]-0.05-0.0200.020.0505001000Time [ s]-0.05-0.0200.020.0505001000Time [ s]-0.05-0.0200.020.0505001000Time [ s]-0.05-0.0200.020.0505001000Time [ s]-0.05-0.0200.020.05Output Amp (at f1)a)b)c)d)e)f)g)5 FIG. 5. Experimental realization of AD: a) central part of the device with the two nonlinear cavities and the central barrier. The positions of the two transducers S1 and S2 and of the receiver T 1 are reported. (b-d) Spectral content detected with a laser scan in three different points (A1, A2 and A3) of the device in each of the three zones indicated. erator (Agilent 33500B) through a 50dB linear amplifier (FLC Electronics A400). The receiving transducer/laser interferometer was connected to an oscilloscope (Agilent Infiniium DSO9024H) for data acquisition. Figs. 5(b-d) show the change in the spectral content of the signal at three different points in the sample detected by the laser vibrometer: only the nonlinearity in the first cavity (NL1) is activated (no reciprocity breaking is ex- pected in this case), so the only frequency that can travel from left to right is the frequency f0 = 75 kHz, generated by wave mixing in the cavity on the left (see [43]). In the second experiment (Fig. 6), the nonlinearity is activated in both cavities in order to demonstrate break- ing of reciprocity, and wave propagation is studied in both directions of the device by placing the transducer S1 in one of the two cavities and the receiver T1 in the other, and subsequently inverting them, leaving the position of S2 unchanged. The filtering action of FB1 and FB3 in the complete device is realized here in post-processing by imposing a numerical band-pass filter (centered around f1 = 150 kHz) to the output signals (third column in Fig. 6). Despite the relatively small amplitudes, the symme- try breaking in the wave propagation for the frequency f1 is evident in the two cases reported in the two rows of Fig. 6. The FFT of the input signal injected at the source S1 and of the output are shown in the first and in the second columns, respectively. In the third column the output is band-pass filtered, simulating the action of the phononic barriers. The difference in the output ob- tained in the left and right propagation demonstrates the functionality of the AD. FIG. 6. Experimental results for LtR (a,b) and RtL (d,e) propagation. In the first and second columns the FFT of the injected signal (a,d) and of the output (b,e) are reported, respectively. The third column represents the simulated effect on the FFT of the output of the phononic barriers in the full device. CONCLUSIONS In summary, we have presented numerical and experi- mental results demonstrating the feasibility of an acous- tic diode based on alternating nonlinear elastic and meta- material frequency-filtering regions, with time-invariant, frequency preserving characteristics. The design concept is sufficiently general to allow flexibility in its realization, involving different combinations of nonlinearity and BG mechanisms, and the use of phononic crystals or resonant metamaterials provides the opportunity to tune and scale results to the desired device sizes and frequency ranges. Additionally, the adoption of an input monochromatic driving signal allows the adaptation of the concept to different types of devices, such as switches or transistors, which can be exploited in practical applications in the field of acoustics or ultrasonics [46]. These can poten- tially be coupled and integrated with recently introduced metamaterial-based sensors for damage detection and lo- calization [25] or for other advanced signal manipulation purposes. However, for an integrated use in a more ad- vanced apparatus, an improvement in the efficiency and stability of the experimental results is required. For this, improved nonlinear elastic solutions are currently under study. 0100200300Frequency [kHz]-20020406080100FB3 action0100200300Frequency [kHz]-20020406080100Left to Right - Experimental0100200300Frequency [kHz]-20020406080100FFT Magnitude [db]a)f0f1b)f1InputOutputc)f1Output(fulll device prediction)0100200300Frequency [kHz]-20020406080100 FB1 action0100200300Frequency [kHz]-20020406080100FFT Magnitude [db]0100200300Frequency [kHz]-20020406080100Right to Left - Experimental f1f2d)InputOutpute)f)(fulll device prediction)OutputACKNOWLEDGMENTS [17] C. Ma, R. G. Parker and B. B. Yellen, J. Sound and 6 M. M. has received funding from the European Unions Horizon 2020 research and innovation program under the Marie Skodowska-Curie Grant Agreement No. 754364. A. K. acknowledges financial support from the Depart- ment of Department of Civil, Environmental and Me- chanical Engineering, University of Trento. NMP is sup- ported by the European Commission under the Graphene Flagship Core 2 grant No. 785219 (WP14 Composites) and FET Proactive "Neurofibres" grant No. 732344 as well as by the Italian Ministry of Education, University and Research (MIUR) under the "Departments of Excel- lence" grant L.232/2016. FB is supported by "Neurofi- bres" grant No. 732344 and by project "Metapp", (n. CSTO160004) cofunded by Fondazione San Paolo. ∗ [email protected] † Also at: EMPA, Laboratory of Acoustics and Noise Con- trol, Uberlandstrasse 129, 8600 Dubendorf, Switzerland ‡ Also at: School of Engineering and Materials Science, Queen Mary University of London, Mile End Road, Lon- don E1 4NS, United Kingdom; Ket Lab, Edoardo Amaldi Foundation, Italian Space Agency, Via del Politecnico snc, 00133 Rome, Italy § [email protected] [1] M. Fink, D. Cassereau, A. Derode, C. Prada, P. Roux, M. Tanter, J.-L. Thomas and F. Wu, Rep. Progr. Phys., 63, 1933 -- 1995 (2000). [2] B. Liang, B. Yuan and J.-C. Cheng, Phys. Rev. Lett., 103, 104301 (2009). [3] B. Liang, X.-Y. Zou, D. Zhang and J.-C. Cheng, Nat. Mater., 9, 989 -- 992 (2010). [4] B. Li, Nat. Mater., 9, 962 -- 963 (2010). [5] G. ter Haar, C. Coussios, Int. J. Hyperthermia, 23(2), 89 -- 104 (2006) [6] K. Liu, S. He, Scien. Rep., 6, 30206 (2016) [7] A. Maznev, A. Every and O. Wright, Wave Motion, 50, 776 -- 784 (2013). Vibration, 332, 4876 -- 4894 (2013). [18] B. Morvan, A. Tinel, A.-C. Hladky-Hennion, J. Vasseur and B. Dubus, Appl. Phys. Lett., 96, 101905 (2010). [19] M. S. Kushwaha, P. Halevi, L. Dobrzynski and B. Djafari- Rouhani, Phys. Rev. Lett., 71, 2022 (1993). [20] R. Martinez-Sala, J. Sancho, J. V. Sanchez, V. Gomez, J. Llinares and F. Meseguer, Nature, 378, 241 (1995). [21] F. Fraternali, G. Carpentieri and A. Amendola, J. Mech. Phys. Solids, 99, 259-271, (2017). [22] S. Yang, J. H. Page, Z. Liu, M. L. Cowan, C. T. Chan and P. Sheng, Phys. Rev. Lett., 93, 024301 (2004). [23] M. Brun, S. Guenneau, A. B. Movchan and D. Bigoni, J. Mech. Phys. Solids, 58, 1212 (2010). [24] A. S. Gliozzi, M. Miniaci, F. Bosia, N. M. Pugno and M.Scalerandi, Appl. Phys. Lett., 107, 161902 (2015). [25] M. Miniaci, A.S. Gliozzi, B. Morvan, A. Krushynska, F. Bosia, M. Scalerandi, and N.M. Pugno, Phys. Rev. Lett. 118, 214301 (2017). [26] M. Miniaci, R. K. Pal, B. Morvan, and M. Ruzzene, Phys. Rev. X, Phys. Rev. X 8, 031074 (2018). [27] K.L.Tsakmakidis, L. Shen, S. A. Schulz, X. Zheng, et al., Science, 356, 1260 (2017) [28] S. Zhang, C. Xia, and N. Fang, Phys. Rev. Lett., 106, 024301 (2011). [29] N. Nadkarni, C. Daraio, and D.M. Kochmann, Phys. Rev. E 90, 023204 (2014). [30] Bertoldi K. , Annu. Rev. Mater. Res., 47, 51-61 (2017). [31] Wang, Y.Z., Li, F.M., Wang, Y.S., Int. J. Mech. Sci., 106, 357 -- 362 (2016). [32] M. Scalerandi, A. S. Gliozzi and C. L. E. Bruno, J. Acoust. Soc. Am., 131, EL81 (2012). [33] P. Deymier, Acoustic Metamaterials and Phononic Crys- tals, Berlin: Springer, 2013. [34] A. Krushynska, M. Miniaci, F. Bosia and N. Pugno, Ext. Mech. Lett., 12, 30 -- 36 (2017). [35] M. Carrara, M. Cacan, J. Toussaint, M. Leamy, M. Ruzzene and A. Erturk, Smart Mater. Struct., 22, 065004 (2013). [36] C. Liu, Z. Du, Z. Sun, H. Gao, and X. Guo, Phys. Rev. Appl. 3, 064014 (2015). [37] Z.-m. Gu, J. Hu, B. Liang, X.-y. Zou, J-c. Cheng, Sci. Rep. 6, 19824 (2016). [38] H. Chen, A.N. Norris, M.R. Haberman, G.L. Huang, Proc. R. Soc. A 473, 20170188 (2017). [39] T. J. Ulrich, P. A. Johnson, R. A. Guyer, Phys. Rev. [8] G. Trainiti and M. Ruzzene, New J. Phys., 18, 083047 Lett., 98, 104301 (2007). (2016). [40] A. S. Gliozzi and M. Scalerandi, J. Acoust. Soc. Am., [9] B. I. Popa and S.A. Cummer, Nat. Commun. 5, 3398 136 (4), 1530 (2014). (2014). [41] P. P. Delsanto, M. Scalerandi, Phys. Rev. B, 68, 064107 [10] S. Lepri and G. Casati, Phys. Rev. Lett., 106, 164101 (2003). (2011). [11] X.-F. Li, X. Ni, L. Feng, M.-H. Lu, C. He and Y.-F. Chen, Phys. Rev. Lett., 106, 084301 (2011). [42] I. D. Mayergoyz, J. Appl. Phys., 57, 3803 (1985). [43] See supplemental material for further details. [44] A. Alippi, G. Shkerdin, A. Bettucci, F. Craciun, E. Moli- [12] F.Li, P.Anzel, J. Yang, P.G. Kevrekidis and C. Daraio, nari, and A. Petri, Phys. Rev. Lett., 69, 3318 (1992) Nat. Comm.,5, 5311 (2014). [45] F. Bosia, N. Pugno, A. Carpinteri, Wave Motion 43, 689- [13] R. Fleury, D. L. Sounas, C. F. Sieck, M. R. Haberman 699 (2006). and A. Al`u, Science, 343, 516 -- 519 (2014). [46] O. R. Bilal, A. Foehr, and C. Daraio, PNAS, 114, 4603 -- [14] H.-X. Sun, S.-Y. Zhang and X.-J. Shui, Appl. Phys. Lett., 4606 (2017). 100, 103507 (2012). [15] Y.-F. Zhu, Z.-M. Gu, B. Liang, J. Yang, J. Yang, L.- L. Yin and J.-C. Cheng, Appl. Phys. Lett., 109, 103504 (2016). [16] S. R. Sklan, AIP Advances, 5, 053302 (2015). [47] M. Miniaci, A. Marzani, N. Testoni, L. De Marchi, Ul- trasonics 56, 251 -- 259 (2015).
1901.10980	1	1901	2019-01-30T18:16:00	Meeting the Scaling Challenge for Post-Silicon Nanoelectronics using CaF2 Insulators	[ "physics.app-ph" ]	Two-dimensional (2D) semiconductors have been suggested both for ultimately-scaled field-effect transistors (FETs) and More-than-Moore nanoelectronics. However, these targets can not be reached without accompanying gate insulators which are scalable into the nanometer regime. Despite the considerable progress in the search for channel materials with high mobilities and decent bandgaps, finding high-quality insulators compatible with 2D technologies has remained a challenge. Typically used oxides (e.g. SiO2, Al2O3 and HfO2) are amorphous when scaled, while two-dimensional hBN exhibits excessive gate leakages. To overcome this bottleneck, we extend the natural stacking properties of 2D heterostructures to epitaxial fluorite (CaF2), which forms a quasi van der Waals interface with 2D semiconductors. We report scalable single-layer MoS2 FETs with a crystalline CaF2 insulator of about 2 nm thickness, which corresponds to an equivalent oxide thickness of less than 1 nm. While meeting the stringent requirements of low leakage currents, our devices exhibit highly competitive performance and record-small hysteresis. As such, our results present a breakthrough for very large scale integration towards commercially competitive nano-electronic devices.	physics.app-ph	physics	Meeting the Scaling Challenge for Post-Silicon Nanoelectronics using CaF2 Insulators Yury Yu. Illarionov,∗,†,‡ Alexander G. Banshchikov,‡ Dmitry K. Polyushkin,P Stefan Wachter,P Theresia Knobloch,† Mischa Thesberg,† Michael Stoger-Pollach,§ Andreas Steiger-Thirsfeld,§ Mikhail I. Vexler,‡ Michael Waltl,† Nikolai S. Sokolov,‡ Thomas Mueller,P and Tibor Grasser∗,† †Institute for Microelectronics (TU Wien), Gusshausstrasse 27 -- 29, 1040 Vienna, Austria ‡Ioffe Physical-Technical Institute, Polytechnicheskaya 26, 194021 St-Petersburg, Russia PInstitute for Photonics (TU Wien), Gusshausstrasse 27 -- 29, 1040 Vienna, Austria §University Service Center for Transmission Electron Microscopy (TU Wien), Wiedner Hauptstrasse 8-10/052, 1040 Vienna, Austria E-mail: [email protected]; [email protected] Abstract Two-dimensional (2D) semiconductors have been suggested both for ultimately- scaled field-effect transistors (FETs) and More-than-Moore nanoelectronics. However, these targets can not be reached without accompanying gate insulators which are scal- able into the nanometer regime. Despite the considerable progress in the search for channel materials with high mobilities and decent bandgaps, finding high-quality insu- lators compatible with 2D technologies has remained a challenge. Typically used oxides (e.g. SiO2, Al2O3 and HfO2) are amorphous when scaled, while two-dimensional hBN exhibits excessive gate leakages. To overcome this bottleneck, we extend the natural stacking properties of 2D heterostructures to epitaxial fluorite (CaF2), which forms a quasi van der Waals interface with 2D semiconductors. We report scalable single-layer 1 arXiv:1901.10980v1 [physics.app-ph] 30 Jan 2019 MoS2 FETs with a crystalline CaF2 insulator of about 2 nm thickness, which corre- sponds to an equivalent oxide thickness of less than 1 nm. While meeting the stringent requirements of low leakage currents, our devices exhibit highly competitive perfor- mance and record-small hysteresis. As such, our results present a breakthrough for very large scale integration towards commercially competitive nano-electronic devices. Various two-dimensional (2D) semiconductors are now considered as channel materials in next-generation field-effect transistors (FETs), which are potentially suitable to extend the life of Moore's law by enabling scaled channel geometries below 5 nm. For instance, several groups have reported FETs with graphene, 1 silicene, 2 black phosphorus 3,4 and transition metal dichalcogenides, such as MoS2, 5 -- 9 MoSe2, 10 MoTe2, 11 WS2 12 and WSe2. 13,14 Excellent transistor characteristics have already been obtained for MoS2 FETs, such as on/off current ratios 7 up to 1010 and sub-threshold swing values down to 69 mV/dec. 9 Furthermore, some attempts on circuit integration of MoS2 FETs have already been undertaken. 15,16 However, one of the major research problems of 2D technologies is their miniaturization without considerable loss in already achieved device performance thresholds. While fabrica- tion of competitive 2D FETs with scaled channel dimensions is already possible, 17 scaling and precise control of the insulator thickness and quality remain a challenge. Typically, oxides known from Si technologies (e.g. SiO2, Al2O3 and HfO2) are used. These materials are amorphous when grown in thin layers, which makes the fabrication of high-quality inter- faces with the channel difficult. In order to address this problem, different insulators have to be identified for next-generation 2D technologies. In particular, hexagonal boron nitride (hBN) has been intensively discussed in the literature. 18 However, hBN has a small bandgap of about 6 eV, 19 a small dielectric constant of 5.06, 20 and unfortunate band offsets to most 2D materials. As scaled technologies require equivalent oxide thicknesses (EOT) below 1 nm (corresponding to a physical thickness of below 1.3 nm in hBN), hBN will result in excessive thermionic and tunneling leakage currents (see Supplementary Section 1). To overcome this bottleneck, we suggest here to use epitaxially grown (and thus crystalline) calcium fluoride 2 (fluorite, CaF2) which has outstanding insulating properties even for a physical thickness of just about 2 nm (EOT of about 0.9 nm). Epitaxially grown CaF2 is a high-k crystalline material with a very favourable combina- tion of dielectric properties, 21 such as a high dielectric constant (ε = 8.43), wide bandgap (Eg = 12.1 eV) and high effective carrier mass (m∗ = 1.0m0). Also, its nearly perfectly match- ing lattice constant (0.546 nm) with Si (0.543 nm) and similarities between the fluorite and Si lattice structures allow growth of CaF2 layers on Si and Ge substrates using molecular- beam epitaxy (MBE) 22,23 with very high quality. Furthermore, the CaF2(111) surface is terminated by F atoms, which makes it chemically inert and free of dangling bonds with H passivation 24 known to result in numerous reliability challenges. Owing to this, heteroepitaxy of 2D materials on the 3D CaF2 surface is possible even for considerable lattice mismatch, 25 while leading to high-quality quasi van der Waals interface. In particular, recent studies have reported epitaxy of MoSe2 26 and MoTe2 27 on CaF2(111) bulk substrates. This further underlines that CaF2 is fully compatible with 2D semiconductors and thus can be considered an extremely promising insulating material for very large scale integration of 2D devices. Note that CaF2 is only one material out of a wide class of epitaxial fluorides. 22 Some of these materials (e.g. anti-ferromagnetic NiF2 28 and MnF2, 29 diamagnetic ZrF2, 29 ferro- electric BaMgF4 30) have additional fascinating properties which may revolutionize future electronic device technologies. Nevertheless, the real potential of fluorides in modern elec- tronic devices is still not fully exploited. For instance, previously, CaF2 films have been mostly used as barrier layers in resonant tunneling diodes 31 and superlattices, 32 together with CdF2 films. As for the use of CaF2 as a gate insulator in FETs, only a few working transistors with 640 nm thick CaF2 and poor reproducibility were reported. 33 The consider- able progress in MBE grown tunnel-thin CaF2 layers achieved in the last decade has revived the idea of fluoride insulators in FETs. 34 In this study we combine the epitaxial growth of CaF2 with chemical vapour deposition (CVD) of MoS2 in a single device. In order to study the variability and reproducibility 3 of these devices, we fabricated hundreds of electrically stable single-layer MoS2 FETs with about 2 nm thick CaF2 gate insulators. Ultra-thin CaF2 layers were deposited onto an atom- ically clean Si(111) surface using our well-adjusted MBE growth technique 23 at 250oC (for more details see the Methods section). The growth process and crystalline quality of CaF2 were controlled in real time using reflection high-energy electron diffraction (RHEED), 35 the corresponding images are provided in Supplementary Section 2. The thickness of CaF2 mea- sured by a quartz oscillator is 6 -- 7 monolayers (ML, 1 ML = 0.315 nm), which is close to the values measured using transmission electron microscope (TEM). The first step of transistor fabrication consisted in formation of SiO2(5 -- 10 nm)/Ti/Au source/drain contact pads using sputtering. Then, a single-layer MoS2 film grown by CVD 36 at 750oC was transferred onto the substrate according to the method of [37] (for more details see the Methods section) and MoS2 channels with L and W varied between 400 and 800 nm were shaped. Finally, additional e-beam evaporated Ti/Au layers were deposited to contact the channels. More details about the structure of our devices can be found in Supplementary Section 3. In Fig. 1a we show that the atomic structure of CaF2(111) is rather similar to that of 2D materials, with F-Ca-F monolayers having a thickness of 0.315 nm. This makes fluorite a natural candidate for the integration into 2D process flows. An essential ingredient of our devices is the virtually defect-free CaF2(111)/MoS2 interface, which is formed by the F-terminated fluorite substrate, a quasi van der Waals gap and an atomically flat MoS2 layer (Fig. 1a). This interface is present in the channel area and under the source/drain electrodes, as marked in the scanning electron microscope (SEM) image in Fig. 1b. In order to verify the layer structure of our device, we cut a 70 nm thick specimen using focused ion beam (FIB) along the line marked in Fig. 1b and performed TEM measurements. In Fig. 1c we show a TEM image obtained for the channel area. We can clearly see the interface between single-layer MoS2 and layered crystalline CaF2 of about 8 ML, which corresponds to a physical thickness of about 2.5 nm. For different substrates, the number of CaF2 monolayers varies between 6 and 8 (thickness between 1.9 and 2.5 nm). By recording electron energy 4 Figure 1: (a) Atomistic structure of the quasi van der Waals interface between F-terminated CaF2(111) and MoS2 in the channel area of our devices. (b) SEM image of our MoS2 FET. The black box indicates the channel area with the source/drain electrodes, where MoS2 is on top of CaF2, as shown in (a). The red line indicates the approximate location of the cut for the TEM sample, with two locations where the images were collected. (c) TEM image obtained in the channel area (location 1). (d) Low resolution TEM image obtained ouside the channel area (location 2). The structure is the same as for the contact pads, with an SiO2 isolation layer deposited on top of CaF2 and MoS2 sandwiched between two metal layers. (e) High resolution TEM image obtained in location 2 where the CaF2 layers are visible. (f) Gate leakage through the CaF2 layer is negligible compared to the drain current, which underlines the high quality of our MoS2 FETs. loss spectra (EELS, see Supplementary Section 5) at the interface between CaF2 and the Si substrate we observe some SiO2 (less than 1 nm thick), which is formed by oxidation 5 resulting from prolonged exposure to air of our CaF2/Si substrates before device fabrication. Taking into account thickness fluctuations 23 of CaF2 and the presence of a thin thermal oxide layer, we model the tunnel leakages measured for numerous devices and find that the effective gate insulator thickness is about 2 nm (see Supplementary Section 4). The layered structure of our CaF2 films is clearly visible in the TEM image obtained using low dose imaging. Although electron irradiation is known to be destructive for CaF2 samples, 38 these investigations indicate the extremely high stability of our thin CaF2 layers, where the desorption of F by the electron beam and subsequent formation of CaO are not as favourable as in CaF2 bulk crystals. 38 Nevertheless, we found that TEM measurements can destroy the Si substrate a few nanometers below the Si/CaF2 interface, which, however, only occurs within the channel area (see Supplementary Section 5). Interestingly, outside the channel area our sample is unaffected by TEM irradiation. There we can clearly see MoS2 sandwiched between two metal layers and an SiO2 isolation layer on top of the CaF2/Si substrate (Fig. 1d,e). As a final verification of the properties of our 2 nm thick CaF2 insulator, in Fig. 1f we show that the measured gate leakage is negligible compared to the drain current of our MoS2 FET. (a) ID − VG characteristics of 50 MoS2/CaF2 FETs from two different Si/CaF2 Figure 2: substrates. (b) Distributions of measured on/off current ratios and subthreshold swings for these devices. (c) Some devices exhibit SS down to 90 mV/dec. Using the process flow described above, we fabricated over 100 devices on two different CaF2 substrates. In Fig. 2a we show the gate transfer (ID − VG) characteristics measured 6 for 50 devices from both substrates. The typical on-currents vary from 1 nA to nearly 10µA, which is likely because of the non-homogeneous nature of the CVD MoS2 film and different effective channel widths. At the same time, the measured on/off current ratios of some devices approach 107 (Fig. 2b), which is excellent for back-gated MoS2 FETs with a tunnel- thin gate insulator. Note that for the devices with overall lower currents, the measured on/off current ratios are probably underestimated due to the limited measurement resolution, which affects the off current. At the same time, the subthreshold swing (SS) values of most devices are smaller than 150 mV/dec, while being close to 90 mV/dec for some devices (Fig. 2b,c). These values are among the best ever reported for back-gated MoS2 FETs. Although in these prototypes SS∼ 90 mV/dec is achieved mostly for the devices with lower current (Fig. 2c), several high-current devices also exhibit small SS values (e.g. Fig. 3). As such, we believe that further optimization of CVD MoS2 FETs on epitaxial fluorides and transition to more versatile configurations, such as top-gated devices and perhaps negative capacitance FETs with ferroelectric fluorides, 30 will lead to further improvements of these emerging devices. Figure 3: (a) ID − VG characteristics measured for the best high current device. (b) On/off current ratio and SS extracted at different VD. The best performance of this device is achieved at VD = 1 V. (c) ID − VD characteristics recorded on the same device exhibit saturation. In Fig. 3a we show typical ID− VG characteristics measured for a device which simultane- ously exhibits high drain currents and steep SS. The best transistor performance is achieved at VD = 1 V, with maximum measured on current about 5 µA (or about 6 µA/ µm if normal- ized to the channel width), on/off current ratio close to 107 and SS as small as 93 mV/dec (Fig. 3b). The output (ID − VD) characteristics measured for different VD (Fig. 3c) also 7 show promising behaviour with a large degree of current control and saturation. All these results confirm the promising nature of our devices and thus the high potential for further development. Figure 4: (a) Ultra-slow sweep ID − VG characteristics measured for our MoS2/CaF2(2 nm) FETs and Al2O3 encapsulated MoS2/SiO2(25 nm) devices reported in our previous work. 7 (b) For comparable sweep times, the hysteresis width has to be normalized to the insulator field factor ∆VG/dins. For our MoS2/CaF2(2 nm) devices it is smaller compared to other 2D FETs and already close to commercial Si/high-k FETs. Finally, we verify the electrical stability of our MoS2/CaF2 FETs by performing ultra- slow sweep hysteresis measurements with total sweep times tsw of several kiloseconds. In Fig. 4a we compare the ID − VG characteristics measured for our MoS2/CaF2 devices with those of MoS2/SiO2 FETs with Al2O3 encapsulation reported in our previous work, 7 which were found to exhibit the smallest hysteresis ever reported in 2D technologies. In addition to comparable values of ID normalized to the channel width and considerably smaller SS for our MoS2/CaF2 FETs, we found that the hysteresis width near Vth is about 0.2 V in both cases. However, for a fair comparison these values need to be normalized to the insulator field factor ∆VG/dins, 7 with ∆VG being the width of VG sweep range. While our devices with ultra-thin fluorite operate at considerably higher insulator fields of up to 15 MV/cm (against 6.4 MV/cm for the devices with thick SiO2), their hysteresis stability is even better than for encapsulated MoS2/SiO2 FETs and close to commercial Si/high-k FETs, not to mention less advanced devices (Fig. 4b). This confirms both the high electrical stability of CaF2 as a 8 gate insulator and the reduced number of slow insulator traps near the CaF2/MoS2 interface. The latter indicates that in addition to an excellent transistor performance, 2D FETs with CaF2 are also more than competitive in terms of their reliability, which is a fundamental requirement for commercial applications. In summary, we have reported single-layer CVD MoS2 FETs on an epitaxially grown CaF2 insulator of record small 2 nm thinness and demonstrated that our fully scalable technology allows fabrication of numerous transistors on a single chip. We have found that already these prototype devices can exhibit SS down to 90 mV/dec and on/off current ratios up to 107, which are among the best values ever reported for back-gated devices. Furthermore, we have demonstrated that our devices are of very high electrical stability even for insulator fields of 10 -- 15 MV/cm and exhibit record small hysteresis ever measured for 2D devices. All this is due to the outstanding dielectric properties of CaF2 and its good compatibility with MoS2, which leads to a virtually defect-free quasi van der Waals interface between these materi- als. Together with the recent demonstration of epitaxial growth of 2D semiconductors on CaF2(111), 26,27 our findings present a breakthrough towards enabling ultra-scaled dielectric layers for next-generation 2D nanoelectronics. Methods MBE growth of CaF2 insulators Ultra-thin CaF2 layers were epitaxially grown on weakly doped single-crystal n-Si(111) substrates with ND = 1015 cm−3 and a misorientation of 5 to 10 angular minutes. Before the growth process, a protective oxide layer was formed after chemical treatment by following the procedure suggested by Shiraki. 39 Then, this layer was removed by annealing for 2 minutes at 1200oC under ultra-high vacuum conditions (∼10−8−10−7 Pa). This allowed us to obtain an atomically clean 7×7 Si(111) surface. After this, the CaF2 film was grown on this surface by MBE at 250oC, which is known to be the optimum temperature to produce pinhole-free 9 homogeneous fluorite layers. 23 The deposition rate of fluorite measured by a quartz oscillator was about 1.3 nm/min. The growth processes and crystalline quality of the CaF2 layers were monitored using RHEED with an electron energy of 15 keV (see the diffraction images shown in Supplementary Section 2). Device fabrication A single-layer MoS2 film serving as a channel was grown on c-plane sapphire using the CVD process described by Dumcenco et al . 36 Namely, CVD growth was performed at atmo- spheric pressure and 750oC using sulfur and MoO3 as powder precursors and ultra-high-purity Ar as the carrier gas. All lithography steps were done using E-Beam lithography. First we deposited SiO2(5 -- 10 nm)/Ti/Au contact pads using sputtering. An isolation with the SiO2 layer is required to minimize parasitic leakage currents through the 15 -- 20 µm sized square electrodes, which have to be so large for a reliable contact with the probe. Then 7×7 mm CVD-grown MoS2 films were transferred onto the CaF2(111) substrate with pre-shaped isolated contact pads using the process suggested by Gurarslan et al . 37 In particular, we used a polystyrene film as a carrier polymer and dissolved it in toluene after the transfer process. The transferred MoS2 film was subsequentily etched by reactive ion etching, in order to define the transistor channels with L and W between 400 and 800 nm. Finally, the channels were contacted by e-beam evaporated Ti/Au pads deposited on top of MoS2 in the contact areas. This second layer of Ti/Au was slightly extended to contact MoS2 on top of the bare CaF2 surface. TEM measurements In order to achieve a high contrast in TEM measurements, the devices were covered by a 10 nanometers thick carbon layer deposited using sputtering. After this a TEM lamella preparation process was performed with a dual beam system. First a thicker granular plat- inum protective layer was deposited using a focused electron beam followed by a focused ion beam deposition. Then a TEM lamella was cut out along the channel of the device. Finally, the samples were examined using a TEM setup at the pressure of about 10−5 Pa. During 10 the measurements, we recorded EELS spectra to verify the layer structure of our device. Electrical characterization Electrical characterization of our MoS2/CaF2 FETs consisted in the measurements of ID − VG and ID − VD characteristics. These measurements were conducted using a Keithley 2636 parameter analyzer in the chamber of a Lakeshore vacuum probestation (∼5×10−6 torr) In order to correctly resolve the on/off at room temperature and in complete darkness. current ratio, we used the autorange measurement mode. The hysteresis of the ID − VG characteristics was investigated by doing double sweeps with a constant sweep rate. Acknowledgement The authors thank for the financial support through the Austrian Science Fund FWF grant n◦ I2606-N30. T.M., D.K.P. and S.W. acknowledge financial support by the Austrian Sci- ence Fund FWF (START Y 539-N16) and the European Union (grant agreement No. 785219 Graphene Flagship). This work was partly supported by the Russian Foundation for Basic Research (grant No. 18-57-80006 BRICS t). We also gratefully acknowledge useful discus- sions with Mr. Markus Jech. Y.Y.I. is a member of Mediterranean Institute of Fundamental Physics (MIFP). Author contributions Y.Y.I. introduced the idea of MoS2 FETs with ultra-thin CaF2 insulator, performed their characterization and prepared the manuscript. A.G.B. performed MBE growth of CaF2 and provided the substrates. D.K.P. and S.W. fabricated MoS2 FETs. M.S.-P. did TEM measure- ments. T.K. and M.T. contributed to preparation of figures. M.S-P. and A.S.-T. performed TEM measurements and sample preparation, respectively. M.I.V. performed quantitative analysis of gate leakage currents using tunnel models. M.W. programmed electrical mea- surements. N.S.S., T.M. and T.G. supervised this work. All authors regularly discussed the 11 results and commented on the manuscript. References (1) Guerriero, E.; Pedrinazzi, P.; Mansouri, A.; Habibpour, O.; Winters, M.; Rorsman, N.; Behnam, A.; Carrion, E.; Pesquera, A.; Centeno, A.; Zurutuza, A.; Pop, E.; Zirath, H.; Sordan, R. High-Gain Graphene Transistors with a Thin AlOx Top-Gate Oxide. Sci. Rep. 2017, 7, 2419. (2) Tao, L.; Cinquanta, E.; Chiappe, D.; Grazianetti, C.; Fanciulli, M.; Dubey, M.; Molle, A.; Akinwande, D. Silicene Field-Effect Transistors Operating at Room Tem- perature. Nature Nanotechnol. 2015, 10, 227. (3) Li, L.; Yu, Y.; Ye, G.; Ge, Q.; Ou, X.; Wu, H.; Feng, D.; Chen, X.; Zhang, Y. Black Phosphorus Field-Effect Transistors. Nat. Nanotechnol. 2014, 9, 372 -- 377. (4) Chen, X.; Chen, C.; Levi, A.; Houben, L.; Deng, B.; Yuan, S.; Ma, C.; Watanabe, K.; Taniguchi, T.; Naveh, D.; Du, X.; Xia, F. Large-Velocity Saturation in Thin-Film Black Phosphorus Transistors. ACS Nano 2018, 12, 5003 -- 5010. (5) Kang, J.; Liu, W.; Banerjee, K. High-performance MoS2 Transistors with Low Resis- tance Molybdenum Contacts. Appl. Phys. Lett. 2014, 104, 093106. (6) Chuang, S.; Battaglia, C.; Azcatl, A.; McDonnell, S.; Kang, J.; Yin, X.; Tosun, M.; Kapadia, R.; Fang, H.; Wallace, R.; Javey, A. MoS2 P-Type Transistors and Diodes Enabled by High Work Function MoOx Contacts. Nano Lett. 2014, 14, 1337 -- 1342. (7) Illarionov, Y.; Smithe, K.; Waltl, M.; Knobloch, T.; Pop, E.; Grasser, T. Improved Hysteresis and Reliability of MoS2 Transistors with High-Quality CVD Growth and Al2O3 Encapsulation. IEEE Electron Device Lett. 2017, 38, 1763 -- 1766. 12 (8) Smithe, K.; Suryavanshi, S.; Munoz-Rojo, M.; Tedjarati, A.; Pop, E. Low Variability in Synthetic Monolayer MoS2 Devices. ACS Nano 2017, 11, 8456 -- 8463. (9) Bolshakov, P.; Zhao, P.; Azcatl, A.; Hurley, P.; Wallace, R.; Young, C. Improvement in Top-Gate MoS2 Transistor Performance due to High Quality Backside Al2O3 Layer. Appl. Phys. Lett. 2017, 111, 032110. (10) Liao, W.; Wei, W.; Tong, Y.; Chim, W. K.; Zhu, C. Electrical Performance and Low Frequency Noise in Hexagonal Boron Nitride Encapsulated MoSe2 Dual-Gated Field Effect Transistors. Appl. Phys. Lett. 2017, 111, 082105. (11) Cho, Y.; Park, J.; Kim, M.; Jeong, Y.; Ahn, J.; Kim, T.; Choi, H.; Yi, Y.; Im, S. Fully Transparent p-MoTe2 2D Transistors Using Ultrathin MoOx/Pt Contact Media for Indium-Tin-Oxide Source/Drain. Adv. Fun. Mater. 2018, 28, 1801204. (12) Yang, L.; Majumdar, K.; Liu, H.; Du, Y.; Wu, H.; Hatzistergos, M.; Hung, P.; Tieck- elmann, R.; Tsai, W.; Hobbs, C.; Ye, P. Chloride Molecular Doping Technique on 2D materials: WS2 and MoS2. Nano Lett. 2014, 14, 6275 -- 6280. (13) Liu, W.; Kang, J.; Sarkar, D.; Khatami, Y.; Jena, D.; Banerjee, K. Role of Metal Con- tacts in Designing High-Performance Monolayer n-type WSe2 Field Effect Transistors. Nano Lett. 2013, 13, 1983 -- 1990. (14) Prakash, A.; Appenzeller, J. Bandgap Extraction and Device Analysis of Ionic Liquid Gated WSe2 Schottky Barrier Transistors. ACS Nano 2017, 11, 1626 -- 1632. (15) Wang, H.; Yu, L.; Lee, Y.-H.; Shi, Y.; Hsu, A.; Chin, M.; Li, L.-J.; Dubey, M.; Kong, J.; Palacios, T. Integrated Circuits Based on Bilayer MoS2 Transistors. Nano Lett. 2012, 12, 4674 -- 4680. (16) Das, T.; Chen, X.; Jang, H.; Oh, I.-K.; Kim, H.; Ahn, J.-H. Highly Flexible Hybrid 13 CMOS Inverter Based on Si Nanomembrane and Molybdenum Disulfide. Small 2016, 12, 5720 -- 5727. (17) Xie, L.; Liao, M.; Wang, S.; Yu, H.; Du, L.; Tang, J.; Zhao, J.; Zhang, J.; Chen, P.; Lu, X.; Wang, G.; Xie, G.; Yang, R.; Shi, D.; Zhang, G. Graphene-Contacted Ultrashort Channel Monolayer MoS2 Transistors. Adv. Mater. 2017, 29, 1702522. (18) Hui, F.; Pan, C.; Shi, Y.; Ji, Y.; Grustan-Gutierrez, E.; Lanza, M. On the Use of Two Dimensional Hexagonal Boron Nitride as Dielectric. Microelectron. Eng. 2016, 163, 119 -- 133. (19) Cassabois, G.; Valvin, P.; Gil, B. Hexagonal Boron Nitride is an Indirect Bandgap Semiconductor. Nature Photonics 2016, 10, 262 -- 266. (20) Geick, R.; Perry, C.; Rupprecht, G. Normal Modes in Hexagonal Boron Nitride. Phys. Rev. 1966, 146, 543. (21) Hayes, W. Crystals with the Fluorite Structure; Clarendon Press, 1974. (22) Sugiyama, M.; Oshima, M. MBE Growth of Fluorides. Microelectronics Journal 1996, 27, 361 -- 382. (23) Illarionov, Y.; Vexler, M.; Fedorov, V.; Suturin, S.; Sokolov, N. Electrical and Optical Characterization of Au/CaF2/p-Si(111) Tunnel-Injection Diodes. J. Appl. Phys. 2014, 115, 223706. (24) Foster, A.; Trevethan, T.; Shluger, A. Structure and Diffusion of Intrinsic Defects, Adsorbed Hydrogen, and Water Molecules at the Surface of Alkali-Earth Fluorides Calculated using Density Functional Theory. Phys. Rev. B 2009, 80, 115421. (25) Koma, A.; Saiki, K.; Sato, Y. Heteroepitaxy of a Two-Dimensional Material on a Three- Dimensional Material. Appl. Surf. Sci. 1990, 41, 451 -- 456. 14 (26) Vishwanath, S.; Liu, X.; Rouvimov, S.; Mende, P.; Azcatl, A.; McDonnell, S.; Wal- lace, R.; Feenstra, R.; Furdyna, J.; Jena, D.; Xing, H. Comprehensive Structural and Optical Characterization of MBE Grown MoSe2 on Graphite, CaF2 and Graphene. 2D Mater. 2015, 2, 024007. (27) Vishwanath, S. et al. MBE Growth of Few-Layer 2H-MoTe2 on 3D Substrates. J. Cryst. Growth 2018, 482, 61 -- 69. (28) Banshchikov, A.; Golosovskii, I.; Krupin, A.; Koshmak, K.; Sokolov, N.; Cher- nenkov, Y.; Yagovkina, M.; Ulin, V.; Tabuchi, M. Epitaxial Layers of Nickel Fluoride on Si (111): Growth and Stabilization of the Orthorhombic Phase. Phys. Solid State 2015, 57, 1647 -- 1652. (29) Kaveev, A.; Anisimov, O.; Banshchikov, A.; Kartenko, N.; Ulin, V.; Sokolov, N. Epitax- ial Growth on Silicon and Characterization of MnF2 and ZnF2 Layers with Metastable Orthorhombic Structure. J. Appl. Phys. 2005, 98, 013519. (30) Ravez, J. The Inorganie Fluoride and Oxyfluoride Ferroelectrics. Journal de Physique III 1997, 7, 1129 -- 1144. (31) Watanabe, M.; Funayama, T.; Teraji, T.; Sakamaki, N. CaF2/CdF2 Double-Barrier Resonant Tunneling Diode with High Room-Temperature Peak-to-Valley Ratio. Jpn. J. Appl. Phys. 2000, 39, L716. (32) Suturin, S.; Basun, S.; Gastev, S.; Langer, J.; Meltzer, R.; Sokolov, N. Optical Detection of Electron Transfer through Interfaces in CaF2: Eu -- CdF2 SLs. Appl. Surf. Science 2000, 162, 474 -- 478. (33) Smith III, T.; Phillips, J.; Augustyniak, W.; Stiles, P. Fabrication of Metal-Epitaxial Insulator-Semiconductor Field-Effect Transistors using Molecular Beam Epitaxy of CaF2 on Si. Appl. Phys. Lett. 1984, 45, 907 -- 909. 15 (34) Tyaginov, S.; Illarionov, Y.; Vexler, M.; Bina, M.; Cervenka, J.; Franco, J.; Kaczer, B.; Grasser, T. Modeling of Deep-Submicron Silicon-based MISFETs with Calcium Fluo- ride Dielectric. J. Computational Electronics 2014, 13, 733 -- 738. (35) Sokolov, N.; Alvarez, J.; Yakovlev, N. Fluoride Layers and Superlattices Grown by MBE on Si(111): Dynamic RHEED and Sm2+ Photoluminescence Studies. Appl. Surf. Sci. 1992, 60, 421 -- 425. (36) Dumcenco, D.; Ovchinnikov, D.; Marinov, K.; Lazic, P.; Gibertini, M.; Marzari, N.; Sanchez, O.; Kung, Y.-C.; Krasnozhon, D.; Chen, M.-W.; Bertolazzi, S.; Gillet, P.; Fontcuberta i Morral, A.; Radenovic, A.; Kis, A. Large-Area Epitaxial Monolayer MoS2. ACS Nano 2015, 9, 4611 -- 4620. (37) Gurarslan, A.; Yu, Y.; Su, L.; Yu, Y.; Suarez, F.; Yao, S.; Zhu, Y.; Ozturk, M.; Zhang, Y.; Cao, L. Surface-Energy-Assisted Perfect Transfer of Centimeter-Scale Mono- layer and Few-Layer MoS2 Films onto Arbitrary Substrates. ACS Nano 2014, 8, 11522 -- 11528. (38) Jiang, N. On the Oxidation of CaF2 in Transmission Electron Microscope. Micron 2012, 43, 746 -- 754. (39) Ishizaka, A.; Shiraki, Y. Low Temperature Surface Cleaning of Silicon and its Applica- tion to Silicon MBE. J. Electrochem. Soc. 1986, 133, 666 -- 671. 16 Meeting the Scaling Challenge for Post-Silicon Nanoelectronics using CaF2 Insulators (Supporting Information) Yu.Yu. Illarionov,∗ A.G. Banshchikov, D.K. Polyushkin, S. Wachter, T. Knobloch, M. Thesberg, M. Stoger-Pollach, A. Steiger-Thirsfeld, M.I. Vexler, M. Waltl, N.S. Sokolov, T. Mueller, and T. Grasser∗ E-mail: [email protected]; [email protected] 1. Tunnel leakage currents through CaF2 and other in- sulators In Fig. S1a we show the energetic alignment of CaF2, hBN and SiO2 relative to Si and several single-layer TMD semiconductors obtained using the band offsets known from literature. 1,2 Clearly, CaF2 has a considerable advantage in terms of bandgap, which is twice as large as that of hBN. This fact alone should lead to considerably smaller tunnel currents for the same equivalent oxide thickness (EOT). The other parameters which affect tunnel leakages are the dielectric constant ε, the electron and hole effective masses (me and mh, respectively) and the conduction band offset with the Si substrate χe. The values known from the literature 2 -- 8 for CaF2, hBN, SiO2 and high-k oxides HfO2 and La2O3 are summarized in Fig. S1b. Using these parameters and our well-established modeling technique, 9,10 we modeled the tunnel currents through the Au/insulator/n-Si(111) system. Our approach is based on a WKB calculation of the tunneling probability and accounts for transverse momentum conservation 9 -- 11 in the case of crystalline CaF2 on Si(111). As for the case of oxide insulators and hBN, the tunneling 1 arXiv:1901.10980v1 [physics.app-ph] 30 Jan 2019 (b) Dielectric parameters for different insulators. Figure S1: (a) Band alignment of CaF2, hBN and SiO2 relative to several single-layer TMD semiconductors. (c) Simulated tunnel leakages for a Au/insulator/n-Si(111) system with EOT = 0.5 nm (left) and EOT = 1.0 nm (right). The doping level of the substrate was ND = 1015 cm−3. In order to maintain a proper comparison of different insulator systems, the condition χm = χe + 0.25 eV is used for the band offset with Au. current is supposed to occur as if silicon were a direct-band semiconductor (otherwise a huge difference between the Si(111) and Si(100) cases would have been predicted, which is never observed). The results obtained for EOT = 0.5 nm and EOT = 1.0 nm are shown in Fig. S1c. Clearly, in both cases the tunnel leakage through CaF2 is even smaller than for high-k oxides, not to mention hBN, which is closer to SiO2 rather than to high-k materials A. Together with a defect-free substrate and a quasi van der Waals interface with 2D semiconductors, small leakages through CaF2 layers with sub-1 nm EOT make this material the most promising for scaled next-generation 2D devices. ANote that for hBN we use the best case values of χe ∼3 eV 2 and ε = 5.06. 3 However, this material is not perfectly parameterized. For instance, other suggested χe values are about 1 eV smaller 12 than in Fig. S1a and an alternative ε would be 3.76. 13 As such, our model provides the results for minimum possible tunnel leakages through hBN. 2 2. RHEED control of CaF2 epitaxy Figure S2: RHEED patterns of the Si(111) surface showing the 7×7 superstructure (a) and a 2 nm thick CaF2 layer grown at 250oC (b). Azimuth of the electron beam is [1¯10]. The MBE growth process and the crystalline quality of the CaF2 layers were monitored using reflection high energy electron diffraction (RHEED) 14 with an electron energy of 15 keV. In Fig. S2 we show the typical RHEED patterns obtained for atomically clean Si(111) before the beginning of fluorite epitaxy (a) and from a deposited CaF2 layer at the final stage of the MBE process (b). The presence of distinct reflections in the CaF2 pattern indicates its high crystalline quality even at a comparably low MBE growth temperature of 250oC. Note that although the use of a higher growth temperature (e.g. 750oC) would further improve the crystalline quality of fluorite, the layer would then become non-homogeneous with a number of pinholes. 3 (a) (b) 3. Geometry of our MoS2/CaF2 FETs Figure S3: (a) Schematic cross-section of our MoS2/CaF2 FETs. (b) SEM image of two de- vices with contact pads. (c) SEM image of the channel area with marked channel dimensions. (d) AFM image of the channel area. In Fig. S3a we show a schematic cross-section of our MoS2/CaF2 FETs. Reliable contact of the test devices with the probes requires contact pads as large as 15 -- 20 µm. However, the CaF2 layer may contain some thickness fluctuations and local pinholes with the depth of several angstroms, which may locally increase the tunnel leakage by orders of magnitude. The density of these pinholes typically depends on the quality of the Si(111) substrate (e.g. 4 misorientation angle, roughness, etc.) and the MBE growth parameters. It is commonly assumed that for a well adjusted MBE process of CaF2 on Si(111) there should be less than one pinhole per 100 µm2. As such, the probability of having some pinholes under the 225 -- 400 µm2 sized pads is quite high, which may lead to a smaller number of functional devices. Thus, in order to reduce parasitic leakages, the contact pads received some additional insu- lating layer with 5 -- 10 nm thick SiO2. At the same time, the contact of the SiO2 layer with the MoS2 film is completely avoided, which is necessary to block possible charge trapping events at the MoS2/SiO2 interface and thus make the transistor characteristics more stable. As such, within the contact area, the MoS2 film is sandwiched between two Ti/Au layers, in which a thin (few nanometers) Ti layer is used as an adhesion layer. A typical view of our devices obtained using a scanning electron microscope (SEM) is shown in Fig. S3b. A detailed SEM image of the channel area (Fig. S3c) allows to estimate the channel dimensions. For our devices both L and W are typically between 400 and 800 nm. However, the CVD MoS2 film contains some imperfections, which may lead to different effective channel widths for different devices. Within the channel area, the MoS2 film is just on top of the CaF2 insulator. As shown in Fig. S3d, the channel area of our devices can be also nicely resolved using atomic-force microscope (AFM). 4. Estimation of the effective gate insulator thickness in our MoS2/CaF2 FETs Taking into account that thickness fluctuations may be present in tunnel-thin CaF2 films, 15 the effective thickness of the insulator can be different from the physical thickness which can be visually seen in TEM images. For a qualitative estimation of the effective thickness, we theoretically model the tunnel leakages and compare the results with our experimental data. First we model the tunnel leakages through Au/SiO2/CaF2/n-Si structures which form our contact pads. In Fig. S4a we show that already for a SiO2 thickness of 3 nm the 5 (a) Simulated tunnel leakages for Au/SiO2/CaF2/n-Si(111) structures rep- Figure S4: resenting the contact pads of our MoS2 FETs. leakages for MoS2/CaF2/SiO2/n-Si(111) structures representing the channel and experimental data mea- sured for 17 devices. It appears that the effective thickness of our gate insulator is about 2 nm, which roughly corresponds to 1.5 nm of CaF2 and 0.3 -- 0.7 nm of SiO2 due to oxygen penetration. (b) Simulated tunnel leakage becomes negligible, while an increase of the insulator thickness by 1 nm leads to a decrease of the tunnel current by about 5 orders of magnitude. As such, our 15 -- 20 µm sized contact pads with 5 -- 10 nm SiO2 can not contribute to the measured tunnel leakage. This means that the measured gate current is mostly given by the tunnel leakage within the channel area. The corresponding modeling results are shown in Fig. S4b. Based on our TEM measurements, we assume that the physical thickness of CaF2 is 7 ML (1 ML=0.315 nm), which is about 22 A. The root mean square (rms) amplitude of fluctuations σ measured for this fluorite thickness using atomic-force microscope (AFM) 15 is about 2.7 A. This gives an effective thickness (deff = dphys − σ2, all values in angstroms) of CaF2 deff ∼ 1.5 nm. In agreement with our previous observations, 16 ambient storage of CaF2 films grown on Si leads to some oxidation at the CaF2/Si interface, since oxygen is able to penetrate through the thinnest places in the CaF2 layers. As a result, a thin SiO2 layer is formed underneath CaF2. While previously we observed this oxidation as a decrease of the tunnel currents after several months of device storage B, in this study the presence of the thin SiO2 layer was confirmed by TEM measurements (see Fig. S5c). As such, in our model we assume MoS2/CaF2/SiO2/n-Si BOur MoS2 FETs were fabricated after 8 -- 12 months following the epitaxial growth of CaF2. 6 structures and vary the thickness of SiO2. Fitting of our modeling results with the tunnel gate currents measured for numerous devices in Fig. S4b allows us to conclude that the effective thickness of the SiO2 oxidation layer is about 0.3 -- 0.7 nm. Therefore, the total effective thickness of the gate insulator in our MoS2 FETs is about 2 nm. Note that the oxidation layer underneath CaF2 does not affect the device performance, since it is far away from defect-free CaF2/MoS2 interface. 5. EELS analysis and sample degradation during TEM In Fig. 5Sa we show the TEM image measured at the beginning of our study. We observe that already during the first measurement the TEM irradiation leads to a partial damage of Si substrate a few nanometers below the Si/CaF2 interface. The origin of this issue is not clear yet, but could be a chemical modification of Si (e.g. by acetone penetrating through CaF2 during MoS2 transfer). Additionally, this damaged region is less stable with respect to irradiation. In particular, this damage appears only in the areas where MoS2 is right on top of CaF2. However, it does not appear away from the channel, where CaF2 is isolated by sputtered SiO2. After two more minutes of irradiation and focusing of the beam, the damaged area expands considerably (Fig. 5Sb). Nevertheless, the Si/CaF2 and CaF2/MoS2 interfaces remained stable enough to allow for recording an electron energy loss spectrometry (EELS) line scan employing the scanning mode of the TEM. In this situation a low dose electron beam is focused and scanned along the line shown in Fig. 5Sb. In each position an EELS spectrum is recorded, thus having a spatial resolution of 0.5 nm. As shown in Fig. 5Sc, the EELS profile clearly confirms the layered structure of our device, though the measured thickness is broadened by an unknown factor K due to the non-planar surface. Also, the presence of the thin SiO2 layer at the Si/CaF2 interface is clearly confirmed by observation of the fine structure of the S-L ionization edge at the respective position of the linescan. This thin oxide layer appears as a result of several months of storage of the Si/CaF2 substrates 7 Figure S5: (a) TEM image obtained at the beginning of our measurements. The Si substrate is damaged by TEM a few nanometers below the CaF2/Si interface. (b) TEM image obtained after 2 more minutes. The damage of the Si substrate is progressing. However, the features of the Si/CaF2/MoS2/Ti structure can be clearly resolved. (c) The cross-sectional EELS spectrum measured along the line marked in (b) confirms the layer alignment and suggests the presence of a thin SiO2 layer at the Si/CaF2 interface. (d) EELS spectra of each particular material detected. before device fabrication and has been accounted for in the model above. Interestingly, in the damaged area the Si signal drops, which confirms the removal of the material by TEM irradiation. The EELS spectra of all the materials detected are shown in Fig. 5Sd and agree well with their reference shapes from the EELS atlas. 8 Figure S6: TEM images of the channel measured with high electron dose rate. Images a-d are taken within 4 minutes, which is enough to destroy the channel completely. Finally, we record TEM images of the channel with an illuminated area of about 75 nm in diameter, a beam current of 26 nA giving an electron dose rate of 3.7×107 electrons/s/nm2, which is at least an order of magnitude larger than for all our previous TEM measurements. As shown in Fig. S6, already the first measurement with such a high electron dose rate leads to a complete amorphisation of CaF2 and partial damage of MoS2. A further increase of irradiation time leads to a severe damage of the Si substrate, and after about 4 minutes all layers are completely destroyed (Fig. S6d). These results confirm that TEM is in general destructive for MoS2/CaF2 FETs and reasonable results can be obtained only with moderate electron dose rates. References (1) Illarionov, Y.; Vexler, M.; Suturin, S.; Fedorov, V.; Sokolov, N. Characteristics of Thin Calcium Fluoride Barrier Layers for Field-Effect Transistors and Functional Electronic Devices. Techn. Phys. Lett. 2010, 36, 404 -- 407. (2) Rasmussen, F.; Thygesen, K. Computational 2D Materials Database: Electronic Struc- ture of Transition-Metal Dichalcogenides and Oxides. J. Phys. Chem. C 2015, 119, 13169 -- 13183. (3) Geick, R.; Perry, C.; Rupprecht, G. Normal Modes in Hexagonal Boron Nitride. Phys. Rev. 1966, 146, 543. 9 (4) Hayes, W. Crystals with the Fluorite Structure; Clarendon Press, 1974. (5) Robertson, J. Band Offsets of Wide-Band-Gap Oxides and Implications for Future Electronic Devices. J. Vac. Sci. Technol. B 2000, 18, 1785 -- 1791. (6) Hou, Y.; Li, M.; Yu, H.; Kwong, D.-L. Modeling of Tunneling Currents through HfO2 and (HfO2)x(Al2O3)/sub 1-x/Gate Stacks. IEEE Electron Device Lett. 2003, 24, 96 -- 98. (7) Miranda, E.; Molina, J.; Kim, Y.; Iwai, H. Effects of High-Field Electrical Stress on the Conduction Properties of Ultrathin La2O3 Films. Appl. Phys. Lett. 2005, 86, 232104. (8) Vexler, M.; Tyaginov, S.; Shulekin, A. Determination of the Hole Effective Mass in Thin Silicon Dioxide Film by Means of an Analysis of Characteristics of a MOS Tunnel Emitter Transistor. J. Phys.: Cond. Matter 2005, 17, 8057. (9) Vexler, M.; Sokolov, N.; Suturin, S.; Banshchikov, A.; Tyaginov, S.; Grasser, T. Elec- trical Characterization and Modeling of the Au/CaF2/nSi(111) Structures with High- Quality Tunnel-Thin Fluoride Layer. J. Appl. Phys. 2009, 105, 083716. (10) Vexler, M.; Illarionov, Y.; Suturin, S.; Fedorov, V.; Sokolov, N. Tunneling of Electrons with Conservation of the Transverse Wave Vector in the Au/CaF2/Si(111) System. Phys. Solid State 2010, 52, 2357 -- 2363. (11) Illarionov, Y.; Vexler, M.; Suturin, S.; Fedorov, V.; Sokolov, N.; Tsutsui, K.; Taka- hashi, K. Electron Tunneling in MIS Capacitors with the MBE-grown Fluoride Layers on Si(111) and Ge(111): Role of Transverse Momentum Conservation. Microelectron. Eng. 2011, 88, 1291 -- 1294. (12) Lee, G.-H.; Yu, Y.-J.; Lee, C.; Dean, C.; Shepard, K.; Kim, P.; Hone, J. Electron Tunneling through Atomically Flat and Ultrathin Hexagonal Boron Nitride. Appl. Phys. Lett. 2011, 99, 243114. 10 (13) Laturia, A.; Van de Put, M.; Vandenberghe, W. Dielectric Properties of Hexagonal Boron Nitride and Transition Metal Dichalcogenides: from Monolayer to Bulk. npj 2D Mater. & Applications 2018, 2, 6. (14) Sokolov, N.; Alvarez, J.; Yakovlev, N. Fluoride Layers and Superlattices Grown by MBE on Si(111): Dynamic RHEED and Sm2+ Photoluminescence Studies. Appl. Surf. Sci. 1992, 60, 421 -- 425. (15) Illarionov, Y.; Vexler, M.; Fedorov, V.; Suturin, S.; Sokolov, N. Electrical and Optical Characterization of Au/CaF2/p-Si(111) Tunnel-Injection Diodes. J. Appl. Phys. 2014, 115, 223706. (16) Illarionov, Y. Tunnel Carrier Transport and Related Physical Phenomena in Gold - Calcium Fluoride - Si (111) Structures. Ph.D. thesis, Ioffe Physical-Technical Institute, St-Petersburg, 2015; in Russian. 11
1906.09316	1	1906	2019-06-21T20:52:22	Field Dependent Conductivity and Threshold Switching in Amorphous Chalcogenides -- Modeling and Simulations of Ovonic Threshold Switches and Phase Change Memory Devices	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	We model electrical conductivity in metastable amorphous $Ge_{2}Sb_{2}Te_{5}$ using independent contributions from temperature and electric field to simulate phase change memory devices and Ovonic threshold switches. 3D, 2D-rotational, and 2D finite element simulations of pillar cells capture threshold switching and show filamentary conduction in the on-state. The model can be tuned to capture switching fields from ~5 to 40 MV/m at room temperature using the temperature dependent electrical conductivity measured for metastable amorphous GST; lower and higher fields are obtainable using different temperature dependent electrical conductivities. We use a 2D fixed out-of-plane-depth simulation to simulate an Ovonic threshold switch in series with a $Ge_{2}Sb_{2}Te_{5}$ phase change memory cell to emulate a crossbar memory element. The simulation reproduces the pre-switching current and voltage characteristics found experimentally for the switch + memory cell, isolated switch, and isolated memory cell.	physics.app-ph	physics	> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1 © 2019 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2 Field Dependent Conductivity and Threshold Switching in Amorphous Chalcogenides -- Modeling and Simulations of Ovonic Threshold Switches and Phase Change Memory Devices Jake Scoggin, Helena Silva, Senior Member, IEEE, and Ali Gokirmak, Senior Member, IEEE Abstract -- We model electrical conductivity in metastable amorphous Ge2Sb2Te5 using independent contributions from temperature and electric field to simulate phase change memory devices and Ovonic threshold switches. 3D, 2D-rotational, and 2D finite element simulations of pillar cells capture threshold switching and show filamentary conduction in the on-state. The model can be tuned to capture switching fields from ~5 to 40 MV/m at room temperature using the temperature dependent electrical conductivity measured for metastable amorphous GST; lower and higher temperature dependent electrical conductivities. We use a 2D fixed out-of- plane-depth simulation to simulate an Ovonic threshold switch in series with a Ge2Sb2Te5 phase change memory cell to emulate a crossbar memory element. The simulation reproduces the pre- switching found experimentally for the switch + memory cell, isolated switch, and isolated memory cell. fields are obtainable using different characteristics current and voltage Index Terms -- Phase change memory, amorphous semiconductors, finite element analysis I. INTRODUCTION P HASE change memory (PCM) is a non-volatile memory that stores information as the conductive crystalline or resistive amorphous phase of a material. The crystalline-to-amorphous phase transition is controllable and reversible, with PCM attaining 103x faster write times and 104x better endurance than flash memory [2]. PCM is CMOS back-end-of-line compatible, allowing memory integration on-chip with CMOS circuitry to eliminate latency from off-chip memory access [3]. PCM can be implemented as a crossbar array, allowing high device density (4F2) layers and efficient neuromorphic of perpendicular word and bit lines with memory elements sandwiched between these lines at the cross-points (Fig. 1). Each word or bit line is connected to Vdd or ground through a transistor, and cross-points can be randomly accessed by activating their corresponding word and bit lines. Non-selected in multiple memory computing [4]. Crossbars consist devices can form undesirable current sneak paths between selected and non-selected lines; hence access devices with non- linear current-voltage (I-V) characteristics, high Ion/Ioff ratios (Ion/Ioff ~106 for a 1000x1000 device array), and high drive capabilities to write (Iwrite ~MA/cm2) are needed at each cross- point [5]. Ovonic threshold switches (OTS) made from amorphous chalcogenides are one such access device. Amorphous chalcogenides are highly resistive under low electric fields and exhibit "threshold switching" to a highly conductive on-state at a threshold voltage (Vth). Once switched, these materials remain in the on-state while a minimum holding current or voltage (Ihold or Vhold) is maintained (Fig. 2) [6]. Fig. 1: A schematic illustration of a cross-point cell with an Ovonic threshold switch in series with a phase change memory element. The shown cell structure is used for 2D analysis to compare modeling results to the experimental results in [1]. Submitted on XX/XX/XXXX. This work was supported by AFOSR MURI under Award No. FA9550-14-1-0351. [email protected]; [email protected]). The authors are with the Department of Electrical and Computer Engineering, University of Connecticut, Storrs, CT 06269 USA (email: [email protected]; MiddleElectrodeOTSPCMBottomElectrode300 K300 KVappRLoadTopElectrode45 nmTiNaGSTcGSTHomogeneous Melting10 nmLiquidCrystal OrientationAmorphous(e)(a)(b)(c)(f)(g)(d)(h)TiNSiO2Heterogeneous MeltingVDeviceI> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3 device thickness, suggesting a field-based mechanism at a threshold Eth [15]. Theoretical arguments and the presence of crystalline filaments in failed devices suggest that on-state conduction is filamentary [12], [15]. (I-𝐸⃗ ) measurements on Current-field amorphous chalcogenides typically show an ohmic regime at low 𝐸⃗ , an 1 intermediate regime where ln(𝐼) ∝ 𝐸⃗ or 𝐸⃗ 2, and a high field regime where I increases at a super-exponential rate with 𝐸⃗ [20]. Reference [20] reviews conduction mechanisms in amorphous materials and shows that multiple mechanisms can fit measured data through the tuning of parameters which are otherwise difficult to validate (e.g. trap-to-trap distance, effective carrier mass, and carrier mobility). Models for these mechanisms have been proposed which define carrier concentrations (n) and mobilities (μ) as functions of 𝐸⃗ and temperature (T) such that all three I-𝐸⃗ regimes are captured with an electrical conductivity 𝜎 = 𝑞𝑛𝜇, but such techniques are computationally expensive in addition to relying on multiple unknown fitting parameters. Reference [13] proposes a field- based switching model where carrier concentrations rapidly increase once trap states near the Fermi band are filled and fits the model to amorphous (a-) Ge2Sb2Te5 (GST) measurements. References [18], [19] propose a field-assisted thermal model based on multiple trap barrier lowering and fit the model to a- GeTe and doped a-GST measurements. References [14], [21] ascribe switching to crystalline filaments which form under high 𝐸⃗ and fit the model to a-GST using relaxation oscillations. They suggest that these filaments become unstable at low 𝐸⃗ in OTS materials but remain stable in PCM materials. Here, we model conductivity in a-GST as the sum of T and 𝐸⃗ dependent terms (σa = σT + σE). This model does not require a computationally expensive evaluation of the (density of states × Fermi function) integral at every T, 𝐸⃗ combination where conductivity is needed; hence, it is appropriate for transient finite element simulations with dynamic T and 𝐸⃗ . While this model trades accuracy for ease of computation, simulations show that it (i) can be tuned to fit a wide range of switching fields, (ii) captures the appropriate changes in threshold switching as we systematically vary ambient conditions, geometries, and the rise and fall times of applied pulses, and (iii) can reproduce the behavior of a series PCM+OTS device when used with a finite element phase change model [22] -- [25]. II. COMPUTATIONAL MODEL We use a-GST material parameters as in [23], [25] to simulate an OTS material. Of particular interest to this work is σa, which we model as in [26] and cap at σmax = σT(930 K) (Fig. 3): 𝜎𝑎(𝑇, 𝐸⃗ ) = min( 𝜎𝑇(𝑇) + 𝜎𝐸(𝐸⃗ ) , 𝜎𝑚𝑎𝑥 ), (1) which is equivalent to assuming that free carriers are excited to Fig. 2: (a) 3D and (b) 2D-rotational OTS geometries used in this work. T = 300 K is used as the initial condition and as the boundary condition at the top and bottom of the TiN contacts. (c) When a 3 V / 60 ns triangular pulse is applied at Vapp, the device (i) is highly resistive until Vswitch ~ 1.75 V, (ii) switches on in ~10 ps, (iii - iv) remains on until the voltage and current drop below Vhold ~ 0.4 V and Ihold ~ 0.25 mA, and (v) returns to the high resistance off-state. Symmetric switching behavior is observed when a negative ramped pulse is applied. The 2D- rotational and 3D simulations give similar results, as expected for rotationally symmetric filamentary switching. Inset in (c) is the first quadrant in logarithmic scale. Crystallization dynamics of these materials determine whether they are more suitable for an OTS or a PCM. PCM materials include various stoichiometries of Ag-In-Sb-Te and Ge-Sb-Te, with typical crystallization times on the order of 10 ns [7]. OTS materials remain in their amorphous phase during normal operation and are often characterized by the number of switching cycles they withstand before failure (through, e.g., material damage or crystallization): > 600 for GeTe6 [8], > 108 for AsTeGeSiN [9], and unknown for As-doped Se-Ge-Si, a material used in a commercial OTS+PCM crossbar array [10]. There is still debate on the mechanism(s) underlying threshold switching despite many studies investigating this phenomenon [1], [6], [19], [11] -- [18]. Vth scales linearly with 2rSiO21 µmhOTS= 50 nmhTiN= 1 µmrOTS= 100 nmRLoadVapp(t)IVDevice3D2D -Rotational(a)(b)-1.5-1.0-0.50.00.51.01.5-2-1012 3D (~2 days) 2D Rotational (~5 min) I (mA)VDevice (V)Vdevice(V)00.511.5-0.5-1-1.5I (mA)-1-210-1.5-1.0-0.50.00.51.01.5-2-1012 3D (~2 days) 2D Rotational (~5 min) I (mA)VDevice (V)3D (~2 days)2D-Rot. (~5 minutes)(c)iiiiiiivvSiO2TiNaGST0110-610-3100 0110-610-3100> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4 dependent thermoelectric effects [34]: 𝑑𝑚𝑐𝑝 𝑑𝑇 𝑑𝑡 − ∇ ⋅ (𝑘𝛻𝑇) = −𝛻𝑉 ⋅ 𝐽 − 𝛻 ⋅ (𝐽𝑆𝑇) + 𝑞𝐻 (4) ∇ ⋅ 𝐽 = 𝛻 ⋅ (−𝜎𝛻𝑉 − 𝜎𝑆𝛻𝑇) = 0 (5) where dm is mass density, cp is specific heat, t is time, k is thermal conductivity, V is electric potential, J is current density, S is the Seebeck coefficient, and qH accounts for the latent heat of phase change. We use temperature dependent parameters for a-GST, crystalline GST (c-GST), TiN, and SiO2 as in [23], [25]. We model phase change as d𝐶𝐷⃗⃗⃗⃗⃗ dt = 𝑁𝑢𝑐𝑙𝑒𝑎𝑡𝑖𝑜𝑛 ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ + 𝐺𝑟𝑜𝑤𝑡ℎ ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ + 𝑀𝑒𝑙𝑡 ⃗⃗⃗⃗⃗⃗⃗⃗⃗ , (6) where 𝐶𝐷⃗⃗⃗⃗⃗ is a 2-vector whose magnitude (CD) corresponds to phase (CD = 0 or 1 for the amorphous or crystalline phase) and whose orientation (θCD) corresponds to grain orientation (tan(θCD) = CD2/CD1), capturing nucleation, growth, and grain boundary melting [23], [25]. We solve for (6) in PCM devices but not OTS devices, which we assume do not crystallize in our simulations. III. SIMULATIONS We first simulate switching in 3D and 2D-rotational geometries by applying a 3V / 60 ns triangular pulse (Fig. 2a,b). Results show filamentary switching with current confined to the central portion of the device (Fig. 4a-j), with practically identical for 3D and 2D-rotational simulations (Fig. 2c). 3D simulations show some instability of the filament location (slightly off-center in Fig. 4d,i) and filament migration over time. I-V characteristics We next evaluate the impact of σE by simulating switching with σE = 0 (Fig. 5a) and compare the results with σE defined as in (3) (Fig. 5b). Vswitch and Iswitch are the values at which VDevice begins to decrease. Threshold switching occurs even with σE = 0 due to thermal runaway. However, defining σE as in (3) gives a switching field (Eswitch = Vswitch / hOTS) that is smaller and less dependent on hOTS (Fig. 5c). Some hOTS dependence is still observed due to changing thermal conditions. Reference [33] reports switching fields from 8.1 MV/m (as- deposited a-Ge15Sb85) to 94 MV/m (as-deposited 4 nm thick a- Sb). We examine the tunability of our model by varying Eth in (3) from 5.6 to 560 MV/m (Fig. 6). Results show Eswitch varying from 5 to 42.5 MV/m. Eswitch = 25.01 MV/m when Eth = 56 MV/m, similar to the Eswitch = 28.75 MV/m measured in [13] for melt-quenched a-GST. σE becomes negligible compared to σT when Eth > 200 MV/m even for high fields: the σT used in this work precludes Eswitch > 42.5 MV/m; a reduced σT is required for higher switching fields. Iswitch decreases by ~100x as Vswitch increases, resulting in a decrease in switching power (Pswitch) from ~100 to 20 μW (Fig. 6c). Fig. 3: (a) T and 𝐸⃗ dependent contributions to electrical conductivity in (1); temperature is uncertain for T > Tmelt. (b) Conductivity-Field behavior using the model in this work (metastable a-GST) and the model in [19] (drifted, doped a-GST) at various temperatures. We use Eswitch(300 K) = 25.01 MV / m in this work and Eswitch(300 K) = 155 MV/m for the curves in [19]. a band edge via independent thermal and electrical processes: 𝜎𝑎(𝑇, 𝐸⃗ ) = min( 𝑞(𝑛𝑇 + 𝑛𝐸)𝜇 , 𝜎𝑚𝑎𝑥 ), (2) where nT and nE are carriers excited via thermal or electrical processes and 𝜇 is the free carrier mobility. We fit σT to low-𝐸⃗ measurements of metastable a-GST wires [27] and molten GST thin films [28], as described in [29] (Fig. 3a). Measurements of liquid GST show a semiconductor-to- metal transition near 930 K [30], with σ becoming practically independent of T. We therefore limit σa(T, 𝐸⃗ ) to σT(930 K) = 4.1×105 [Ω-1 m-1], which the highest conductivities measured in molten GST [30] -- [32](Fig. 3a). line with is in σE is assumed to be an exponential which contributes 1% of σT(300 K) at zero field and 10% of σT(Tmelt) at Eth: 𝜎𝐸(𝐸⃗ ) = 𝜎𝑇(300 𝐾) 100 exp(𝐸⃗ ⋅ 𝐶1) (3) where C1 = 2.42×10-7 m/V is chosen such that σE(Eth) = σT(Tmelt) × 10%. We use Eth = 56 MV/m, the breakdown field measured in as-deposited a-GST [33], and Tmelt = 858 K [28] (Fig. 3a). We include σ-𝐸⃗ curves at various temperatures calculated using the models in this work (for metastable a-GST) and the models in [19] (for drifted, doped a-GST) for comparison (Fig. 3b). σT dominates at low fields, while σE begins to dominate at higher and higher fields with increasing T. We couple heat transfer and current continuity physics to simulate transient device operation, including temperature 10-21001021041064006008001000T(930 K)[32][30][31][28]E  (-1 m-1)T[26]0255075T (K)TmeltEthE (MV/m)0.00.51.01.510-1100101102103104[27]This Work700 K300 K700 K300 K (-1 m-1)E/Eswitch(300 K)[11] (a)(b)> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5 Fig. 6: (a) The switching current (voltage) decreases (increases) and (c) the switching power decreases with increasing Eth in (3). The device switches thermally before the field contribution becomes significant for Eth > 1×108 V / m. As a result, further increases to Eth result in the same switching characteristics. (Fig. 2b: RLoad = 5 kΩ, hOTS = 100 nm, rOTS = 100 nm, Vapp = 5 V / 5 s triangular pulse). Vswitch has been shown to decrease with increasing ambient temperature (Tambient), while the temperature behavior of Iswitch and Pswitch are less clear [19]. We simulate switching while varying Tambient (the initial temperature and the fixed top and bottom TiN boundary temperatures, Fig. 2b) from 300 to 400 K (Fig. 7). Vswitch decreases as expected (Fig. 7a). Iswitch at first decreases and then increases with increasing Tambient, while Pswitch monotonously decreases in the Tambient range simulated but is beginning to flatten with increasing T by 400 K. Next, we systematically vary rOTS, hOTS, and the rise time (τrise) of Vapp in the geometry shown in Fig. 2b (Fig. 8). Results agree with expected OTS behavior: Vswitch approximately doubles as hOTS doubles [6], Vswitch decreases with increasing τrise, approaching a minimum value [16], Ihold and Vhold are only weakly dependent on hOTS [15], and switching characteristics are only weakly dependent on rOTS due to filamentary conduction in the on-state. Finally, we simulate an OTS and PCM in series (OTS+PCM, Fig. 1) based on the devices fabricated and characterized in [1]. We use a 2D, 45 nm fixed out-of-plane-depth simulation instead of a 2D-rotational simulation to more appropriately model phase change dynamics in the PCM with (6). We use a 500 nm depth in the bit line to account for its large thermal mass (Fig. 1), set Tambient = 300 K as the initial and fixed TiN boundary temperatures, and reset the device with a 5 V / 5 ns square pulse (1 ns rise and fall times) at Vapp followed by 1 μs for thermalization (starting at Fig. 1 and ending at Fig. 9a). We then sweep Vapp from 0 to 2.5 V over 50 ns to characterize the OTS + reset PCM. We also simulate an isolated OTS and isolated reset PCM in the same way by replacing the PCM or OTS, respectively, with TiN (Fig. 9c,d). We plot the I-V characteristics before switching, dividing currents and voltages by the isolated OTS Iswitch and Vswitch values in order to compare Fig. 4: (a-e) x-y and (f-j) x-z temperature cut planes while switching the 3D OTS in Fig. 2a illustrate filamentary on-state conduction. (j) Current and (k) device voltage transients resulting from the applied Vapp used to generate the I-V in Fig. 2c. Superscripts "-" and "+" refer to the time steps (1 ns increments) before and after the subscripted event. (Fig. 2a: RLoad = 1 kΩ, hOTS = 50 nm, rOTS = 100 nm, Vapp = 5 V / 60 ns triangular pulse). Fig. 5: The switching voltage increases with hOTS both (a) without and (b) with field dependent conductivity. Including field dependent conductivity reduces the switching field's (c) magnitude and (d) sensitivity to hOTS. (Fig. 2b: RLoad = 5 kΩ, hOTS = 25 to 100 nm, rOTS = 100 nm, Vapp = 5 V / 5 s triangular pulse). tswitch+T (K)300950-tswitchtpeak200 nm012I (mA)050100024tholdtpeaktswitchVDevice V (V)time (ns)Vapp(a)(b)(c)(d)(e)(f)(g)(h)(i)(j)(k)(l)-tholdthold+2550751000.000.250.500.751.000204060 E as in (3)E = 0 ESwitch/E25 nmSwitchhOTS (nm) E as in (3)E = 0ESwitch (MV/m)(c)012340200400600800(b) I (A)VDevice (V)increasing hOTS01234increasing hOTS(a)E as in (3)E = 0VDevice (V)(d)1071080255075100 PSwitch (W)EThreshold (V/m)101102ISwitch (A)01234 VSwitch (V)(a)(b)Eth(V/m)> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6 Fig. 7: The (a) voltage, (b) current, and (c) power required to switch as the ambient temperature changes. Vswitch decreases monotonically, but Iswitch and Pswitch have more complex relationships with Tambient. (Fig. 2b: RLoad = 5 kΩ, hOTS = rOTS = 100 nm, Vapp = 5 V / 5 s triangular pulse). Fig. 8: Switching characteristics of OTS devices with varying radii, ramp times, and heights. The simulated holding currents and voltages are weakly dependent on hOTS, but the switching voltage ~doubles as as hOTS doubles. (Geometry in Fig. 2b). Fig. 9: (a) Reset OTS+PCM, (b) isolated OTS, and (c) isolated reset PCM. (d) Pre-switching I-V characteristics scaled by Vswitch and Iswitch in the OTS using the model in this work (spheres) and experimental data extracted from [1] (squares). I-V characteristics from this work show similar switching voltages but lower scaled switching currents. The schematic of the simulation setup is shown in Fig. 1. The OTS+PCM reset animation for this simulation is available in supplementary material. between aGST and the (unreported) OTS material used in [1]. IV. CONCLUSION in amorphous semiconductors The coupling of thermal and electric field contributions to electrical conductivity is complex, as evidenced by the large number of physical models proposed to explain the same characteristics. Our modeling results show that threshold switching and 'snap-back' observed in OTS and PCM devices can be explained through electro- thermal phenomena giving rise to thermal run-away and filamentary conduction and can be modeled efficiently with a finite element framework. V. ACKNOWLDEGEMENTS The authors would like to thank Ilya Karpov of Intel Corporation, Martin Salinga of RWTH Aachen University, Abu Sebastian of IBM Zurich, and Geoffrey Burr of IBM Almaden for valuable discussions. 𝑂𝑇𝑆 our results to those presented in [1] (Fig. 9e). The results are similar, with the OTS limiting the current in the reset PCM until 𝑉𝐷𝑒𝑣𝑖𝑐𝑒 / 𝑉𝑠𝑤𝑖𝑡𝑐ℎ > 2 (Fig. 9d). This limits the current during read in a reset cell while allowing high current in a set cell, creating a large read margin. The smaller scaled currents for the PCM and OTS+PCM in our simulation could be due to a difference in the amorphous volume in the reset PCM; the fixed out-of-plane depth in our simulations, which cannot capture filaments smaller than 45 nm in depth and may thus overestimate Iswitch in the OTS; or parameter differences [1] [2] [3] REFERENCES D. Kau, S. Tang, I. V. Karpov, R. Dodge, B. Klehn, J. A. Kalb, J. Strand, A. Diaz, N. Leung, J. Wu, S. Lee, T. Langtry, K. W. Chang, C. Papagianni, J. Lee, J. Hirst, S. Erra, E. Flores, N. Righos, et al., "A stackable cross point phase change memory," in Technical Digest - International Electron Devices Meeting, IEDM, 2009, pp. 1 -- 4 DOI:10.1109/IEDM.2009.5424263. F. Xia, J. Xiong, and N.-H. Sun, "A Survey of Phase Change Memory Systems," J. Comput. Sci. Technol., vol. 30, no. 1, pp. 121 -- 144, 2015 DOI:10.1007/s11390-015-1509-2. H.-S. P. Wong, S. Raoux, S. Kim, J. Liang, J. P. Reifenberg, B. Rajendran, M. Asheghi, and K. E. Goodson, "Phase ChangeMemory," Proc. IEEE, vol. 98, no. 12, pp. 2201 -- 2227, 2010 3003203403603804000102030(c) PSwitch (W)TAmbient (K)051015(b)ISwitch (A)0123 [2016 Gallo] This WorkVSwitch (V)(a)[11]0200400600800hOTS = 25 nmhOTS = 50 nmrOTS = 25 nm5 V / 1 s5 V / 100 nsI (A)5 V / 10 ns0200400600800rOTS = 50 nmI (A)012340200400600800hOTS = 100 nmrOTS = 100 nm VDevice (V)I (A)01234VDevice (V)01234VDevice (V)01201This WorkOTS+PCMPCMOTSV / VOTSSwitchI / IOTSSwitch[2009 Kau] (b)OTS(c)(a)PCMOTS+PCM01201This WorkPCM+OTSPCMOTSV / VOTSSwitchI / IOTSSwitch[2009 Kau] (d)SiO2TiNaGSTcGSTHomogeneous Melting10 nmLiquidCrystal OrientationAmorphous(e)(a)(b)(c)(f)(g)(d)(h)TiNSiO2Heterogeneous Melting01201This WorkOTS+PCMPCMOTSV / VOTSSwitchI / IOTSSwitch[2009 Kau] 150 nm[5]> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7 [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] DOI:10.1109/JPROC.2010.2070050. A. Chen, Memory Selector Devices and Crossbar Array Design: A Modeling Based Assessment, vol. 16, no. 4. Springer US, 2017, pp. 1186 -- 1200 DOI:10.1016/j.cub.2015.10.018. G. W. Burr, R. S. Shenoy, K. Virwani, P. Narayanan, A. Padilla, B. Kurdi, and H. Hwang, "Access devices for 3D crosspoint memory," J. Vac. Sci. Technol. B, Nanotechnol. Microelectron. Mater. Process. Meas. Phenom., vol. 32, no. 4, p. 040802, 2014 DOI:10.1116/1.4889999. S. R. Ovshinsky, "Reversible electrical switching phenomena in disordered structures," Phys. Rev. Lett., vol. 21, no. 20, pp. 1450 -- 1453, Nov. 1968 DOI:10.1103/PhysRevLett.21.1450. T. Matsunaga, J. Akola, S. Kohara, T. Honma, K. Kobayashi, E. Ikenaga, R. O. Jones, N. Yamada, M. Takata, and R. Kojima, "From local structure to nanosecond recrystallization dynamics in AgInSbTe phase-change materials," Nat. Mater., vol. 10, no. 2, pp. 129 -- 134, 2011 DOI:10.1038/nmat2931. M. Anbarasu, M. Wimmer, G. Bruns, M. Salinga, and M. Wuttig, "Nanosecond threshold switching of GeTe 6 cells and their potential as selector devices," Appl. Phys. Lett., vol. 100, no. 14, 2012 DOI:10.1063/1.3700743. M. J. Lee, D. Lee, H. Kim, H. S. Choi, J. B. Park, H. G. Kim, Y. K. Cha, U. I. Chung, I. K. Yoo, and K. Kim, "Highly-scalable threshold switching select device based on chaclogenide glasses for 3D nanoscaled memory arrays," Tech. Dig. - Int. Electron Devices Meet. IEDM, no. December, pp. 10 -- 13, 2012 DOI:10.1109/IEDM.2012.6478966. J. Choe, "Intel 3D XPoint Memory Die Removed from Intel OptaneTM PCM (Phase Change Memory)," Tech Insights, 2017. [Online]. Available: https://www.techinsights.com/about- techinsights/overview/blog/intel-3D-xpoint-memory-die-removed- from-intel-optane-pcm/. [Accessed: 10-Dec-2018]. R. W. Pryor and H. K. Henisch, "Nature of the on-state in chalcogenide glass threshold switches," J. Non. Cryst. Solids, vol. 7, no. 2, pp. 181 -- 191, 1972 DOI:10.1016/0022-3093(72)90288-8. K. E. Petersen, "On state of amorphous threshold switches," J. Appl. Phys., vol. 47, no. 1976, p. 256, 1976 DOI:10.1063/1.322309. D. Ielmini, "Threshold switching mechanism by high-field energy gain in the hopping transport of chalcogenide glasses," Phys. Rev. B - Condens. Matter Mater. Phys., vol. 78, no. 3, pp. 1 -- 8, Jul. 2008 DOI:10.1103/PhysRevB.78.035308. M. Nardone, V. G. Karpov, D. C. S. S. Jackson, and I. V. Karpov, "A unified model of nucleation switching," Appl. Phys. Lett., vol. 94, no. 10, pp. 10 -- 12, Mar. 2009 DOI:10.1063/1.3100779. W. Czubatyj and S. J. Hudgens, "Invited paper: Thin-film Ovonic threshold switch: Its operation and application in modern integrated circuits," Electron. Mater. Lett., vol. 8, no. 2, pp. 157 -- 167, 2012 DOI:10.1007/s13391-012-2040-z. M. Wimmer and M. Salinga, "The gradual nature of threshold switching," New J. Phys., vol. 16, 2014 DOI:10.1088/1367- 2630/16/11/113044. A. Athmanathan, D. Krebs, A. Sebastian, M. Le Gallo, H. Pozidis, and E. Eleftheriou, "A finite-element thermoelectric model for phase-change memory devices," Int. Conf. Simul. Semicond. Process. Devices, SISPAD, vol. 2015 -- Octob, pp. 289 -- 292, 2015 DOI:10.1109/SISPAD.2015.7292316. M. Le Gallo, M. Kaes, A. Sebastian, and D. Krebs, "Subthreshold electrical transport in amorphous phase-change materials," New J. Phys., vol. 17, no. 9, 2015 DOI:10.1088/1367-2630/17/9/093035. M. Le Gallo, A. Athmanathan, D. Krebs, and A. Sebastian, "Evidence for thermally assisted threshold switching behavior in nanoscale phase-change memory cells," J. Appl. Phys., vol. 119, no. 2, 2016 DOI:10.1063/1.4938532. M. Nardone, M. Simon, I. V. Karpov, and V. G. Karpov, "Electrical conduction in chalcogenide glasses of phase change memory," J. Appl. Phys., vol. 112, no. 7, 2012 DOI:10.1063/1.4738746. I. V. Karpov, M. Mitra, D. Kau, G. Spadini, Y. A. Kryukov, and V. G. Karpov, "Evidence of field induced nucleation in phase change memory," Appl. Phys. Lett., vol. 92, no. 17, pp. 4 -- 6, 2008 DOI:10.1063/1.2917583. Z. Woods and A. Gokirmak, "Modeling of Phase-Change Memory: Nucleation, Growth, and Amorphization Dynamics During Set and Reset: Part I -- Effective Media Approximation," IEEE Trans. Electron Devices, vol. 64, no. 11, pp. 4466 -- 4471, Nov. 2017 DOI:10.1109/TED.2017.2745506. [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] Z. Woods, J. Scoggin, A. Cywar, L. Adnane, and A. Gokirmak, "Modeling of Phase-Change Memory: Nucleation, Growth, and Amorphization Dynamics during Set and Reset: Part II - Discrete Grains," IEEE Trans. Electron Devices, vol. 64, no. 11, pp. 4472 -- 4478, Nov. 2017 DOI:10.1109/TED.2017.2745500. J. Scoggin, R. S. Khan, H. Silva, and A. Gokirmak, "Modeling and impacts of the latent heat of phase change and specific heat for phase change materials," Appl. Phys. Lett., vol. 112, no. 19, p. 193502, 2018 DOI:10.1063/1.5025331. J. Scoggin, Z. Woods, H. Silva, and A. Gokirmak, "Modeling heterogeneous melting in phase change memory devices," Appl. Phys. Lett., vol. 114, no. 4, pp. 1 -- 6, Oct. 2019 DOI:10.1063/1.5067397. A. Faraclas, N. Williams, A. Gokirmak, and H. Silva, "Modeling of set and reset operations of phase-change memory cells," IEEE Electron Device Lett., vol. 32, no. 12, pp. 1737 -- 1739, Dec. 2011 DOI:10.1109/LED.2011.2168374. F. Dirisaglik, G. Bakan, Z. Jurado, S. Muneer, M. Akbulut, J. Rarey, L. Sullivan, M. Wennberg, A. King, L. Zhang, R. Nowak, C. Lam, H. Silva, A. Gokirmak, and others, "High speed, high temperature electrical characterization of phase change materials: metastable phases, crystallization dynamics, and resistance drift," Nanoscale, vol. 7, no. 40, pp. 16625 -- 16630, Oct. 2015 DOI:10.1039/C5NR05512A. L. Adnane, N. Williams, H. Silva, and A. Gokirmak, "High temperature setup for measurements of Seebeck coefficient and electrical resistivity of thin films using inductive heating," Rev. Sci. Instrum., vol. 86, no. 10, p. 105119, Oct. 2015 DOI:10.1063/1.4934577. S. Muneer, J. Scoggin, F. Dirisaglik, L. Adnane, A. Cywar, G. Bakan, K. Cil, C. Lam, H. Silva, and A. Gokirmak, "Activation energy of metastable amorphous Ge2Sb2Te5 from room temperature to melt," AIP Adv., vol. 8, no. 6, p. 065314, 2018 DOI:10.1063/1.5035085. R. Endo, S. Maeda, Y. Jinnai, R. Lan, M. Kuwahara, Y. Kobayashi, and M. Susa, "Electric resistivity measurements of Sb2Te3 and Ge2Sb2Te5 melts using four-terminal method," Jpn. J. Appl. Phys., vol. 49, no. 6, p. 5802, 2010 DOI:10.1143/JJAP.49.065802. T. Kato and K. Tanaka, "Electronic Properties of Amorphous and Crystalline Ge 2 Sb 2 Te 5 Films," Jpn. J. Appl. Phys., vol. 44, no. 10, pp. 7340 -- 7344, 2005 DOI:10.1143/JJAP.44.7340. K. Cil, F. Dirisaglik, and L. Adnane, "Electrical resistivity of liquid based on thin-film and nanoscale device measurements," on Electron Devices, 2013. D. Krebs, S. Raoux, C. T. Rettner, G. W. Burr, M. Salinga, and M. Wuttig, "Threshold field of phase change memory materials measured using phase change bridge devices," Appl. Phys. Lett., vol. 95, no. 8, pp. 1 -- 4, 2009 DOI:10.1063/1.3210792. G. Bakan, N. Khan, H. Silva, and A. Gokirmak, "High-temperature thermoelectric transport at small scales: Thermal generation, transport and recombination of minority carriers," Sci. Rep., vol. 3, no. 1, p. 2724, 2013 DOI:10.1038/srep02724. Jake Scoggin received his B.S. degree in Nanosystems Engineering from Louisiana Tech University in 2012 and his M.S. degree in Electrical Engineering from Louisiana Tech University in 2014. He is a PhD student in Electrical & Computer Engineering at the University of Connecticut since 2014. Helena Silva received her B.S. degree in Engineering Physics from the University of Lisbon, Portugal in 1998 and her M.S. and Ph.D. degrees in Applied Physics from Cornell University in 2002 and 2005. She is currently an Associate Professor of Electrical and Computer Engineering at the University of Connecticut. Her research focuses on electronics devices for logic, memory, and energy conversion. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8 Ali Gokirmak received his B.S. degrees in Electrical Engineering and Physics from University of Maryland at College Park in 1998 and received his Ph.D. in Electrical and Computer Engineering from Cornell University in 2005. He joined the Electrical & Computer Engineering Department of the University of Connecticut in 2006, where he is currently an associate professor .
1906.08461	1	1906	2019-06-20T06:55:49	Micro/nanomaterials for improving solar still and solar evaporation -- A review	[ "physics.app-ph" ]	In last decades, solar stills, as one of the solar desalination technologies, have been well studied in terms of their productivity, efficiency and economics. Recently, to overcome the bottleneck of traditional solar still, improving solar still by optimizing the solar evaporation process based on micro/nanomaterials have been proposed as a promising strategy. In this review, the recent development for achieving high-performance of solar still and solar evaporation are discussed, including materials as well as system configurations. Meanwhile, machine learning was used to analyze the importance of different factors on solar evaporation, where thermal design was founded to be the most significant parameter that contributes in high-efficiency solar evaporation. Moreover, several important points for the further investigations of solar still and solar evaporation were also discussed, including the temperature of the air-water interface, salt rejecting and durability, the effect of solid-liquid interaction on water phase change.	physics.app-ph	physics	Micro/nanomaterials for improving solar still and solar evaporation - Guilong Peng1,#, Swellam W. Sharshir1,2,3, #, Yunpeng Wang1, Meng An4, A.E. Kabeel5, A review Jianfeng Zang2, Lifa Zhang6, Nuo Yang1* 1 State Key Laboratory of Coal Combustion, School of Energy and Power Engineering, Huazhong University of Science and Technology, Wuhan 430074, China 2 School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan 430074, China 3 Mechanical Engineering Department, Faculty of Engineering, Kafrelsheikh University, Kafrelsheikh, Egypt 4 College of Mechanical and Electrical Engineering, Shaanxi University of Science and Technology, Xi'an, 710021, China 5 Mechanical Power Engineering Department, Faculty of Engineering, Tanta University, Tanta, Egypt 6 Center for Quantum Transport and Thermal Energy Science, School of Physics and Technology, Nanjing Normal University, Nanjing, 210023, China #Guilong Peng and Swellam W. Sharshir contribute equally on this work. Corresponding email: [email protected] 1 Abstract In last decades, solar stills, as one of the solar desalination technologies, have been well studied in terms of their productivity, efficiency and economics. Recently, to overcome the bottleneck of traditional solar still, improving solar still by optimizing the solar evaporation process based on micro/nanomaterials have been proposed as a promising strategy. In this review, the recent development for achieving high-performance of solar still and solar evaporation are discussed, including materials as well as system configurations. Meanwhile, machine learning was used to analyze the importance of different factors on solar evaporation, where thermal design was founded to be the most significant parameter that contributes in high-efficiency solar evaporation. Moreover, several important points for the further investigations of solar still and solar evaporation were also discussed, including the temperature of the air-water interface, salt rejecting and durability, the effect of solid-liquid interaction on water phase change. Keywords：micro/nanomaterials; micro/nanoparticles; solar still; solar evaporation; desalination; phase change. 2 1. Introduction Safe freshwater is essential for urban expansion, human civilization, and industrial development. Billions of people are suffering freshwater scarcity while seawater covers 70% of the earth in area. Given the both of environment pollution and water scarcity, it is desirable to develop eco-friendly technology for seawater desalination [1]. However, due to population growth and industrial development, human demand for freshwater resources has risen rapidly day after day. Nowadays, the problem of drinking water is one of the major problems that both developed and developing countries are facing [2]. Most health problems are caused by the scarcity of clean drinking water [3]. In recent decades, the lack of rainfall all over the world has led to an increase in the salinity of water bodies. Environmental pollution further exacerbates the scarcity of clean freshwater [4]. One of most important ways to address the freshwater scarcity is seawater desalination. Solar desalination, which uses renewable solar energy, is considered as a promising desalination technology to provide clean water with no or minimal environmental impacts. Compared to high-grade energy such as fuel and electric energy, solar energy does not depend on the long-range transportation due to worldwide distributed. At the same time, solar energy is eco-friendly energy, which is in agreement with the environmental protection policies all over the world. Therefore, solar desalination is getting more and more interest [5]. As a convenient solar thermal desalination technology, solar still is very useful in some places where insufficient electrical power and complicated desalination plants (such as RO plants) are not available [6]. For instance, poor, remote coastal area and emergency water supply in outdoor. The system and working principle of solar still are very simple compared to other desalination technologies, which make it no need skilled labor for operation and maintenance [7]. Meanwhile, the efficiency of the photo-thermal process is much higher than the photo-electric process, which gives solar still a potential to outperform electric based desalination process. However, the application of solar still is limited by its low productivity. Investigation of suitable 3 materials and system designs might make photo-thermal process based solar still system develop to a new stage and contribute more in desalination field. During past decades, many traditional ways have been developed to improve the efficiency of solar still, including using cotton cloth [8], sponge [9], charcoal [10] and so forth to improve the solar absorption. External area and power source, such as collectors [11], condenser[12] or reflectors [13], were also used and broadly studied. However, the energy efficiencies of traditional solar stills remain low (about 30-45%), due to the inefficient evaporation and condensation process, as well as the large heat dissipation to the ambient of system [14]. Recently, micro/nanotechnology shows a great potential in improving solar desalination and solar evaporation, due to high thermal conductivity, large surface area and high solar absorptivity of micro/nanomaterials [15]. Various nanostructured solar absorber materials, such as plasmonic metals [16], carbon-based materials [17], polymers [18] and semiconductors [19] with efficient photothermal conversion capabilities have been studied. Several effective concepts have been proposed to fully take the advantages of nanotechnology. For example, nanobubble formation [20] and heat localization [21], which direct the design of volumetric evaporation system and surface heating evaporation system, respectively [5, 22]. In this paper, the state-of-the-art developments in micro/nanotechnology for improving solar still and solar evaporation is reviewed. Various factors, such as material form, materials types, and thermal design and so forth, are compared and discussed in details. To evaluate the importance of different factors, a dataset of reported results is built and the machine learning method is carried out. The important gaps between research results of solar evaporation and solar still are also highlighted and discussed, which points out the future directions of this field. 2. Solar still by micro/nanoparticles 2.1 Passive solar still The schematic diagram of how to use suspended nanoparticles or nanofluids in a basic solar desalination system, i.e. solar still (SS), is shown in Fig. 1a and 1b. It 4 consists of a well-insulated container where brackish/saline water with nanoparticles is collected at its base. The upper surface of the container is covered by a tightly sealed glazier cover to minimize vapor leakage. Insolation entering the still through the glass is absorbed by nanofluids, hence the nanofluid is heated up. Nanoparticles absorb a large amount of solar radiation and discharge the heat to a base fluid (water) and the water molecules begin to evaporate faster. Vapor molecules are carried up from the surface of water to the air inside the still by natural convection and the air becomes saturated by vapors. When this saturated air strikes the cool inner glass surface, condensation of vapor molecules began to occur. This condensate water slides down due to gravity and accumulates outside the still as freshwater. Fig. 1 (a) Mixing nanoparticles with base water and (b) Working conventional solar still with nanofluid. As illustrated in Fig. 1b, passive solar still indicates the most basic solar still, which has no extra power input and additional equipment. In this section, the modifications done in the passive solar still with nanoparticles are summarized. Oxide nanoparticles based nanofluids are the most common type used in solar still, due to the excellent durability and low cost. Al2O3 was found to be an effective nanoparticle among various oxide nanoparticles in passive solar still. One of the reasons might be that higher thermal conductivity of nanofluid gives higher efficiency and productivity under sunny days in April at Kovilpatti (9° 11′ N, 77° 52′ E) Tamil Nadu, India, Al2O3 nanofluid gave 29.95% enhancement, pursued by 18.63% for SnO2 and 12.67% for ZnO. Thermal conductivity of Al2O3, SnO2, and ZnO nanofluids are 0.6355 5 W/(m2·K), 0.6215 W/(m2·K), 0.6105 W/(m2·K), respectively [23]. The trend of thermal conductivity agrees well with the enhancement. The experiments of CuO, TiO2 and Al2O3 under climatic conditions of New Delhi for the month of March also give the same conclusion[24]. The test has been carried out at various concentrations (0.2, 0.25 and 0.3%wt) in passive double slope SS to obtain the best concentration (0.25%) of the three nanoparticles. The best energy efficiency was gained for nanofluid (CuO 43.81%, TiO2 46.10%; and Al2O3 50.34%;) in compared with the pure water (37.78%). The results also agree with the trend thermal conductivity which are 0.6901 W/(m2·K), 0.7261 W/(m2·K) and 0.7863 W/(m2·K), for CuO, TiO2 and Al2O3, respectively. The experimental results agree with the theoretical analysis, which further proves the importance of thermal property of nanofluid in solar still. Nevertheless, the concentration of nanofluid in the above-mentioned works are relatively low (0.05-0.3%wt), hence the enhancement is not very high. Enhance the concentration might further enhance the productivity of solar still [3]. However, given the white color of Al2O3, TiO2, ZnO and SnO2 nanoparticles, the high reflectance of these nanoparticles under high concentration might decrease the productivity. Therefore, nanoparticles with black color might be a better choice for a higher concentration of nanofluid in solar still, such as CuO and carbon-based nanoparticles. Sharshir et al. investigated the effect of graphite micro-flakes (GMF) and copper oxide (CuO) with glass cover cooling on the SS performance on September in Wuhan, China. The GMF and CuO with various weight concentrations ranged from 0.125% to 2% was studied. Various brine depths ranged from 0.25 to 5 cm., and various glass cover cooling flow rates ranged from 1 to 12 kg/h were examined to achieve the best performance. The obtained results using CuO with glass cooling enhanced the still productivity by approximately 44.91% as well as using GMF with glass cooling improved the productivity by 53.95%, compared with the conventional one. When using the CuO and the water exit from glass cover cooling as feed water to the still the freshwater is enhanced by approximately 47.80%. Eventually, the daily efficiencies of the modified SSs using GMF and CuO microparticles with glass cooling are 49% and 6 46%, respectively [25]. Furthermore, when GMF nanofluid is combined with glass film cooling and phase change material the daily productivity is enhanced by 73.80%, as compared with the conventional SS [26]. 2.2 Active solar still The active solar still is similar to the passive solar still but integrated with the external energy dissipation devices such as external condenser and fan or collectors and pump as illustrated in Fig. 2. In this section, the modifications done in the active solar still with micro/nanoparticles are summarized. Due to the enhanced the heat and mass transfer rate, which is a benefit to the evaporation and condensation process, active solar still is found to be able to take better advantage of nanofluids than passive solar still. External condenser is one of the most popular and effective components used in active SS for enhancing the condensation rate. By combining nanofluid and external condenser, both high evaporation rate and condensation rate can be achieved. Kabeel et al. studied the effect of Al2O3 and Cu2O on the performance of SS with and without external condenser. The performance was examined at various concentrations without and with providing vacuum. The large improvement in output freshwater productivity was gained with using Cu2O nanoparticles at the optimal concentration of 0.2% together with the vacuum fan as an external condenser. The experimental results showed that using Al2O3 nonmaterial increased the freshwater output approximately by 88.97% and 125.0% without and with electric fan (external condenser), respectively, as compared with that of conventional stills. Cu2O nanoparticles improve freshwater output by about 93.87% and 133.64% without and with electric fan (external condenser), respectively [27]. Omara et al. investigated experimentally a hybrid distillation unit including corrugated plate absorbers, wick material, reflectors, external condenser and nanoparticles (Cu2O and Al2O3). The productivity of modified SS was enhanced by approximately 285.10% and 254.88% for Cu2O and Al2O3, respectively, at 1cm brine depth and 1.97%wt. of concentration [28]. Clearly, active SSs with nanofluids show a much better performance than traditional passive SSs. 7 However, the cost might increase due to the need of extra devices and energy. The nanofluid-based solar collector is another popular component for improving active solar still. Compared to directly use nanoparticles in the basin of solar still, nanoparticles in solar collector can be more stable due to the stir of pumps. The loss of materials can be also avoided when the flow cycle is closed as illustrated in Fig. 2. Nevertheless, the productivity enhancement of using nanofluid-based solar collector is not very significant according to the reported results. Mahian et al. conducted investigation to improve SSs by using two types of nanofluids (SiO2 and Cu /water nanofluid) with two flat plate collectors joint in series, as illustrated in Fig. 2. It shows that the productivity of SS improved by only about 1% when using SiO2 nanofluid at 4% volume fraction and nanofluid temperature at 70 ºC, although the convection heat transfer coefficient is enhanced by about 15.4% [29]. Sahota et al. also investigated the active SS integrated with water collectors and heat exchanger. The freshwater output of using CuO, TiO2, and Al2O3 was enhanced by about 31.49%, 7.26% and 26.4%, respectively, compared with that of pure water [30]. Fig. 2 Diagrammatic of the experimental setup of solar still with the solar collector and nanofluid [29]. Copyright 2017 Elsevier. 2.3 Challenges of using nanofluids in solar still 8 In solar stills, especially in active solar stills, metallic surfaces are important to improve the heat transfer process. However, nanofluid might enhance the erosion and corrosion on a metallic surface by both chemical and physical ways [31]. When the fluid's characteristics fall in the typical chemical corrosion range, even for a limited time interval, a marked and fast consumption of the metallic surface will be observed [32]. Meanwhile, nanofluid also shows erosion due to the collision between the metallic surface and particles in the bended pipes [33]. Therefore, to avoid the erosion and corrosion, the nanofluid should be preliminarily tested before its adoption in desalination system and be assessed on possible negative interactions with components [31]. It is also effective to take advantage of nanofluid itself to decrease erosion and corrosion, such as promoting the formation of a compact protective film on the metallic surface by nanoparticles [34]. Proper system design and maintenance are also necessary to decrease effects of erosion and corrosion [35]. Besides erosion and corrosion phenomena, stability of nanofluid and pressure drop are another two problems when using nanofluids for solar desalination. Long term stability is still one of the major challenges for nanofluids and requires more researches. Poor stability results in aggregation and settlement of particles as well as chemical dissolutions, which lead to the failure of nanofluid [36]. For passive device, where there is no pump to circulate and stir the nanofluids, high agglomeration of nanoparticles will occur, specially at height temperature gradients [37]. Pressure drop and pumping power problems will appear in active solar still when using nanoparticles. The increase of nanofluid concentration results in an increase in pressure drop under turbulent regime [38]. The increased pressure drop will inevitably increase the operation cost of system. 2.4 Section summary Nanofluids have great potential to enhance the energy efficiency and heat transfer in solar still system, which increases the freshwater output of solar still system. Many kinds of nanoparticles are studied in the previous works, such as Al2O3, ZnO, SnO2, CuO, TiO2 and graphite particles (Table 1). The enhancement of productivity is due to 9 the several merits compared to base fluid (i.e. water), such as high thermal conductivity and high solar absorptivity. However, the stability of nanofluid, erosion and corrosion by nanofluid, as well as pressure drop of pump by nanofluid are three important challenges in this field. Meanwhile, oxide nanoparticles such as Al2O3, ZnO and TiO2 are white, which should have low solar absorptivity thus to the disadvantages of enhancement in productivity. Therefore, more detailed and fundamental studies should be explored to further understand the reported enhancement by these white nanoparticles. 10 Table 1. Summary of solar stills with different types of micro/nanoparticles for enhancing the production Types of SS Types of particles Size Concentration, % Modifications Enhancement, % Criticism and disadvantages ZnO [23] Al2O3 SnO2 Graphite [25a] CuO Graphite [26] Al2O3 [27b] Cu2O Al2O3 [27a] Al2O3 [3] CuO [24] TiO2 Al2O3 Al2O3 [27b] Cu2O Al2O3 [27a] Al2O3 [28] Cu2O CuO [30] TiO2 Al2O3 SiO2 [29] Cu 9.3-16 nm 240-395 nm 114-115 nm 1.3 μm 1 μm 1.3 μm 10-14 nm 10-14 nm 20 nm 20 nm 10-14 nm 10-14 nm 0.1 1 0.5 0.2 0.2 0.12 0.25 0.2 0.2 - Glass cooling Glass cooling and PCM External condenser External condenser - - External condenser With fan External condenser With fan Wick absorbers，internal 10-14 nm 1.97 reflectors and external condenser. 20 nm 0.25 Flat plate collectors Flat plate collector and heat exchanger 7 nm 4 11 12.67 29.95 18.63 57.60 47.80 73.8 88.97 93.87 76 12.2 43.81 46.10 50.34 125.0 133.64 116 285.10 254.88 31.49 7.26 26.4 0.66 9.86 missing efficiency missing cost missing efficiency missing efficiency missing efficiency missing cost missing cost missing efficiency missing efficiency missing efficiency missing efficiency missing cost Passive solar stills Active solar stills 3. Solar evaporation by micro/nanomaterials Recently, solar evaporation draws much more attention compared to the whole solar still system, particularly during the last five years. Much effort has been done to improve solar evaporation by nanomaterials. The solar evaporation system can be divided into two categories: a volumetric system and interface system, which contains nanofluid and floatable porous materials, respectively (Fig. 3). Porous materials are foams and membranes, which have micro/nanostructures. Fig. 3 Schematic diagram of micro/nanomaterials in solar evaporation system. 3.1 Volumetric system by nanofluids In a volumetric system, solar irradiation is mostly absorbed by suspended nanoparticles. Water is heated by nanoparticles and then evaporates. Metallic nanoparticles, carbon-based nanoparticles and are the two main kinds of nanoparticles for improving solar evaporation during the past decade. Metallic nanoparticles convert solar irradiation to heat mostly based on plasmonic effect, i.e. photons induced electronic resonance in metallic nanoparticles which generate heat due to 12 electron-phonon scattering. Metallic nanoparticles strongly absorb light when the optical frequency matches the resonance frequency of electron in nanoparticles[5]. Meanwhile, the plasmonic effect in nanoparticles can be designed and adjusted by changing the shape, size, location, surface chemistry of particles, and so on [39]. Therefore, tuning the optical property of metallic nanoparticles becomes a hot topic during the last several years. In the beginning, metal-based nanofluids arise much interest in evaporation field due to the laser-induced generation of vapor bubbles [40]. When the laser intensity overs a typical threshold, the water around the particles reaches a high temperature, resulting in explosive evaporation. Thus, a bubble around the nanoparticles will be formed [41]. The nanoparticle-generated bubbles may temporally and spatially localize laser-induced thermal field and prevent residual heating of the bulk media, which indicates a high energy efficiency, due to a lower bulk temperature and a less heat loss [20, 42]. Inspired by laser-induced nanobubbles generation, metal-based nanofluids were further studied under concentrated solar irradiation to show its potential for solar evaporation. By using golden nanofluids, Halas et al. found that 80% of the absorbed sunlight was converted into water vapor and only 20% of the absorbed light energy was converted into heating of the surrounding liquid [20, 43]. To understand the mechanism, subsequent researches were carried out. The results show that the high-efficiency evaporation is caused by the collective effect mediated by multiple light scattering from dispersed nanoparticles. Randomly positioned nanoparticles that both scatter and absorb light can concentrate light energy into mesoscale volumes near the illuminated surface of the liquid. The resulting light absorption creates intense localized heating and efficient vaporization of the surrounding liquid [44]. A similar conclusion was obtained by Jin et al. [45]. It means that the nanobubble doesn't exist in the solar irradiation-based system, which is different from the laser-based system. To further improve the energy efficiency of metal-based nanofluid system, many 13 efforts have been devoted to optimize the optical property of the nanofluid, such as controlling morphology [46], concentration [47], compounds of particles [48]and diameters [49]. However, the required power density for efficiently using metal-based nanofluid is always hundreds of suns, thus high equipment cost is required for the solar concentration. Meanwhile, the material cost of metal nanoparticles is quite high. Therefore, the potential of using metal-based nanofluid in industrial solar evaporation process is questionable. Different from metal-based nanoparticles, carbon-based nanoparticles have much lower materials cost and higher solar absorptivity. The cost of producing 1 g/s vapor by gold nanoparticles is found to be ~300 folds higher than that produced by carbon black nanoparticles [50]. Carbon-based materials absorb solar energy mainly by thermal vibrations, hence a high and broadband absorptivity can be achieved under low solar intensity [5]. Therefore, some researchers turned to carbon-based nanoparticles or nanocomposites which shows a high efficiency under a relatively low solar concentration [51]. For example, graphene-silver nanoparticle composites exhibit a high efficiency as 80% under around 60 suns [52], and graphene oxide-gold nanoparticle composites have an efficiency up to 59.2% under 16.77 suns [53]. The pristine carbon nanoparticles such as carbon black, graphite and carbon nanotube have an efficiency up to 69% under only 10 suns [22]. Thereby, carbon-based nanoparticles seem to be a more suitable material for solar evaporation compared to metallic nanoparticles. 3.2 Interface system based on floatable porous materials Compared to nanofluid, an essential benefit of floatable porous materials is creating solar heating at air-water interface. Key components are solar absorber, floating evaporation structure and porous materials. A solar absorber that can efficiently absorb and convert the solar radiation into heat. While allowing the vapor to permeate through the front face, a floating evaporation structure that can simultaneously maximize the evaporation rate and supply liquid to the heated region [54]. Porous materials, such as foams and films, concentrate thermal energy and fluid 14 flow were needed for phase change and minimizes dissipated energy [21]. As shown in Fig. 3, solar energy is absorbed on the top surface of porous material and creates a hot layer. The porous foam, which has low thermal conductivity, prevents the heat transferring from hot layer to bulk water. That is, heat is localized at the hot surface. To reach a high evaporation rate, the setup for heat localization needs to have four main characteristics: high absorption in the solar spectrum, low thermal conductivity to suppress heat dissipation, hydrophilic surfaces to leverage capillary forces and promote fluid flow, and interconnected pores for fluid flow [21]. Moreover, compared to nanofluid, floatable porous materials can be easily recycled. Hence, the maintenance cost is low. 3.2.1 Materials design (i) Films Due to the thin thickness (mostly micrometer scale), films show a potential to achieve high evaporation efficiency with very few amounts of materials. Paper-based film is one of most popular materials used in the air-water interface evaporation system, because it is cheap and scalable [55]. A common method to fabricate paper-based photo-thermal film is depositing nanoparticles on airlaid paper. For example, gold nanoparticles and graphene oxide nanoparticles [56] [57]. Due to the improved solar absorption by higher surface roughness of papers, paper-based AuNP films give a much higher evaporation rate compared to pure AuNP films. The energy efficiency of paper-based AuNP film is able to reach 77.8% under 4.5 suns [58]. Other kinds of porous films, such as silica membrane [19], noble metal membrane [59], aluminum oxide membrane [60], zeolite membrane [61], wood membrane [62], polymer and fibers membrane[63], are also well studied. Some films exhibit good performance with a thin thickness. The nitrogen doped 3D porous hydrophilic graphene membrane (thickness 35 μm) enables an efficiency reaches up to 80 % under one sun of solar irradiation [64]. The evaporation efficiency of MXene thin membranes with only several micrometers in thickness can reach up to 84% under 15 one sun irradiation [65]. Based on alumina nonporous membrane (50 μm in thickness), Zhu's group developed several plasmon-enhanced solar desalination devices. Firstly, gold nanoparticles were used as plasmonic absorbers and found that the energy efficiency can reach up to over 90% under 4 suns. Later, to decrease the materials cost, gold nanoparticles were replaced by aluminum nanoparticles. The energy efficiency remains over 90% under 6 suns. [39, 66]. It is also found that films with nanometer thickness also show high efficiency. The ultrathin 2D porous photothermal film (120 nm in thickness) based on MoS2 nanosheets and single-walled nanotube (SWNT) gives an energy efficiency reach up to 91.5% under 5 sun [67]. Fig. 4 Different nanostructure morphologies of membrane. (a) Nanocage of black Titania nanoparticles. (b) Cauliflower-shaped hierarchical surface nanostructure on a copper surface. (c) Vertically aligned graphene sheets membrane (VAGSM). (a) Reprinted with permission from Ref. [68]. Copyright (2016) American Chemical Society. (b) Reprinted with permission from Ref. [69]. Copyright (2016) The Royal Society of Chemistry. (c) Reprinted with permission from Ref. [70]. Copyright (2017) American Chemical Society. Besides the use of different materials, the designing of micro/nanoscale 16 membrane surface structures is also a popular strategy to enhance evaporation. The main purpose of structure designing is to enhance solar absorptivity. It is challenging to achieve high solar absorptivity in a very thin film for most materials, due to shortened light path. Therefore, to absorb as much light as possible within a limited thickness, the light trapping effect has been proposed in many works, which enhances residence time and length of light path [68]. For example, a black Titania film with unique nanocage structure on the surface can increase the energy efficiency by more than 30% under solar intensity of 1 kW/m2, due to the dramatically enhanced absorptivity (Fig. 4a) [68]. Other special structures, such as cauliflower-shaped hierarchical surface (Fig. 4b), vertically aligned graphene sheets membrane (VAGSM) (Fig. 4c) were also proved very effective in improving solar evaporation. [69-70] (ii) Foams Although films exhibit excellent performance, its thin thickness limited the heat localization effect; hence the heat loss to bulk water remains relatively high. Therefore, to further prevent the heat transfer between hot interface and bulk water, foams (centimeter scale in thickness) are a better candidate than films. Aerogels are ideal materials for solar evaporation, due to their extremely low thermal conductivities and porous structures. One of the ways to fabricate suitable aerogels is integrating cellulose with metallic nanomaterials, such as dispersing gold nanorods into highly porous bacterial nanocellulose based aerogels [71]. Another way is integrating cellulose with carbon materials which have broader absorption band and more cost effective compared with metallic materials. The bilayer structure is a common method to use carbon materials in cellulose. A layer of the carbon material is on the top for solar absorption, while a layer of aerogel under is used for water transport, thermal insulation and supporting the carbon layer. The evaporation efficiency can reach up to 78% at 1 kW/m2 solar irradiation, which is much higher than metallic nanomaterials based cellulose aerogel [72]. Fabricating carbon aerogel directly is also well studied. GO aerogel and graphene aerogel is the most popular carbon aerogel used for solar evaporation [73]. The evaporation efficiency is up to 17 around 80%-90% under only 1 sun. [74]. Interestingly, hierarchically structured aerogels show a greater potential in enhancing solar evaporation compared to other aerogels. The hierarchical graphene foam (h-G foam) with continuous porosity grown via plasma-enhanced chemical vapor deposition, shows energy efficiency of solar evaporation up to 93.4% (Fig. 5a)[75]. The hierarchically nanostructured gels (HNG) evaporates water with a record high rate of 3.2 kg/(m2·h), and the energy efficiency reaches up to 94% at one sun irradiation[76]. This extremely high evaporation rate was 2.1 times that of the traditional limitation of evaporation rate, which is around 1.5 kg/(m2·h) under 1 sun [77]. Coincidentally, the hierarchical graphdiyne-based architecture also provides a high energy efficiency as 91% under 1 sun[78]. Those results indicate that the hierarchical structure may have a great impact on the solar evaporation process. However, the underlying mechanism remains to be uncovered. Apart from the artificially synthesized aerogels showed above, natural biomaterial-based foam was also proved efficient in solar distillation, such as mushroom and wood (Fig. 5b and 5c) [79]. The cell walls in biomaterial form natural porous structure which is similar to that of aerogel, thus biomaterial also has a low thermal conductivity which is important for heat localization. Meanwhile, the cellulose in biomaterial is hydrophilic and provides strong capillary force for water replenishment to evaporation surface. Those characters of biomaterial and inspired a broad research of using woods in solar evaporation and desalination. However, although the solar thermal efficiency could be relatively high by using biomaterials, the durability of biomaterial is questionable, due to corrosivity of seawater to biomaterials. 18 Fig. 5 (a) Schematic diagram of hierarchical graphene foam. G foam means graphene foam, h-G foam is hierarchical graphene foam [75]. (b) Surface carbonized wood with the tree-growth direction parallel to the water surface [80]. (c) Schematic of a mushroom-based solar steam-generation device [79c]. (a) Reprinted with permission from Ref. [75]. Copyright 2018 Wiley-VCH Verlag GmbH & Co. KGaA. (b) Reprinted with permission from Ref. [80]. Copyright 2018 Wiley-VCH Verlag GmbH & Co. KGaA. (c) Reprinted with permission from Ref. [79c]. Copyright 2017 Wiley-VCH Verlag GmbH & Co. KGaA. One of strategies by using biomaterials in solar evaporation and desalination is coating surface of biomaterials by other materials, such as graphene oxide[81], carbon nano tubes [82], metal nanoparticles [83] and polymers [84]. Another more convenient 19 and cost-effective strategy is carbonizing biomaterials directly [79b, 79c, 80]. Porous woods float on water spontaneously and function as insulation layer and water channel, while the black layer absorbs solar irradiation for water evaporation. Based on these principle, various biomaterials are studied, such as basswood [81], poplar, pine, and cocobolo wood [85], mushroom [79c], sugarcane [86] and leaf [87]. The most efficient biomaterials show an evaporation efficiency as high as 87% under one sun, which is comparable to artificial aerogels [84, 86]. 3.2.2 Thermal design Besides material design, thermal design of the system is also an effective way to improve the performance of solar evaporation. The thermal design of the evaporation system can be divided into three categories, 3D, 2D and 1D system, based on the dimension of water channel. In 3D water channel systems, all the foam is penetrable, and water is transported through the bottom to the top surface of the foam for evaporation via the connected holes (Fig. 6a). Both aerogels and biomaterial-based systems, as discussed above belong to 3D water channel systems. Although the thermal conductivity of dry penetrable foam is low, the thermal conductivity increases obviously when the foam is filled with water, which weakens heat localization effect [21, 81]. Whereas in 2D water channel systems, a penetrable layer wraps over an impenetrable foam. The penetrable layer functions as a water channel and the impenetrable foam functions as an insulation layer. Because of reduced dimensionality of water channel, the heat dissipation through water will be decreased compared to 3D water channel systems. Li et al. firstly reported that with a 2D water channel, both efficient water supply and suppressed parasitic heat dissipation could be achieved simultaneously. The efficiency was reached to 80% by using graphene oxide film and polystyrene foam (Fig. 6b) [88]. Due to excellent thermal performances, the evaporation efficiency of the 2D water channel system can reach up to 88% by using cheap and scalable materials. However, the reduced dimensionality of the water channel might give rise to the problem of salt accumulation on the evaporation 20 surface, when applied to seawater evaporation system[77]. In order to avoid salt accumulation, the molecular diffusion and water convection in the 2D water channels must be sufficient[89]. As for 1D water channel systems, there are two categories: bundle type system and truck type system (Fig. 6c and 6d). In the bundle type system, the water is transported through a bundle of 1D water channels, which is similar to the jellyfish (Fig. 6c). By using porous carbon black/graphene oxide (CB/GO) composite as the top solar absorption layer, aligned GO pillars as water channels, expanded polystyrene (EPS) as an impenetrable insulation layer. The assembled jellyfish-like evaporator can display a high energy efficiency of 87.5% under one-sun illumination [90]. On the other hand, truck type system contains only one 1D water channel, similar to the tree trunk [91]. Based on this concept, Liu et al. achieved 91.3% of energy efficiency under 1 sun by using airlaid paper as the truck and carbonized wood as solar absorber (Fig. 6d) [92]. Generally, both 2D and 1D water channel systems can reach a high energy efficiency with simpler material fabrication process, due to the minimized heat dissipation. 21 Fig. 6 Different thermal designs of solar evaporation system. (a) 3D water channel system, porous foam is full of connected holes which function as water channels during the evaporation process[64]. Copyright 2015 Wiley-VCH Verlag GmbH & Co. KGaA. (b) 2D water channel system, a penetrable layer wraps over impenetrable insulation foam. Water is transported along the wrapping layer[88] Copyright 2016, National Academy of Science.. (c) Root type 1D water channel system, water is transported along the root-like pillar array[90]. Copyright 2017, Elsevier (d) Trunk type 1D water channel system, the water channel is limited in the centre of impenetrable insulation foam[92]. Copyright 2018, Elsevier 3.3 Factors analyzing by machine learning In the experiments of solar evaporation, the evaporation efficiency is affected by many factors, such as materials, thermal design, ambient temperature, and solar intensity, and so forth. Therefore, it is difficult to directly conclude the importance of different factors, and different works can't be compared with each other. Herein, the 22 machine learning algorithm of random forest (RF) is used to analyze the importance of various factors (i.e. descriptor in RF). RF has been widely applied in many scientific and engineering fields [93], the main step of RF is shown in Fig. 7. After model construction, the importance of certain descriptor can be calculated by intrinsic metric. Experimental data used in RF are collected from articles since 2014 (Table S2). The detail of method and dataset are listed in Supporting Information. According to the results from RF, among all calculated factors, thermal design is the most important factor in solar evaporation, no matter how many descriptors are used in the calculation. The importance of thermal design is around 2 times higher than other descriptors (Fig. 8). This result shows that optimizing the heat transfer process in solar evaporation system is essential for enhancing the efficiency of solar evaporation. Solar absorptivity of materials is also relatively important, which is reasonable due to higher absorptivity enables more available energy for evaporation. However, most of the reported works have a very high absorptivity (>90%), hence its importance might be underestimated in the calculation. Meanwhile, due to the optimized thermal and material design, high efficiency can be obtained under low solar intensity as reported in many works (Table S2). Therefore, solar intensity is not important and similar to a random descriptor (i.e. a set of random data which should have no relationship to evaporation efficiency). The temperature of ambient (Tamb) and evaporation interface (Tinterface) are also insignificant, which might due to the small difference of Tamb and Tinterface between most of works as summarized in Table S2. The stable trend of importance of different factors in Fig. 8 proves that the calculated results are reasonable. However, it should be noted that some papers do not provide information about all descriptors. Data of ambient temperature, the diameter of evaporation surface, absorptivity, and the temperature of evaporation interface are missing in some papers. Therefore, to obtain a more accurate result, it requires authors to provide complete information on experimental factors in their papers. On the other side, some other important factors, such as the material design, can't be considered and calculated in 23 the current stage, because the detailed properties of materials are not offered in most of papers, such as thermal conductivity, contact angle, specific area, porosity, characteristic size and chemistry properties and so forth. Fig. 7 Schematics of applying the random forest in studying the importance of different factors. 24 Fig. 8 Results of descriptor importance. The values are normalized as the sum of all values of descriptor importance equals 1. (a)-(d) are the results of using 2, 3,4, and 6 different descriptors, respectively. Surface diameter is the diameter (length) of the evaporation surface. Tamb is the temperature of ambient, Tinterface is the temperature of evaporation interface. Random descriptor is a set of random values and should have no relationship to solar evaporation, which is calculated for comparisons. 3.4 Section summary In summary, interface system exhibits better performance than volumetric system, due to heat localization effect. Porous materials are a better candidate for solar evaporation compared to nanofluids. Various micro/nanostructured porous materials are suitable for solar evaporation, such as paper-based film, artificial aerogel and natural biomaterials. The most efficient materials, especially the hierarchically structured aerogels, exhibit a more than 90% of energy efficiency. Meanwhile, the thermal design shows a significant role in solar evaporation according to calculation results by machine learning. Decrease the heat loss by decreasing the dimension of water channel generally enhances the evaporation efficiency. However, the narrowed water channel may also increase the possibility of salt accumulation, which is quite harmful to seawater based solar evaporation. 25 4. Research gaps between solar still and solar evaporation 4.1 Low efficiency of solar still with floatable structure It is important to prove the effectiveness of applying materials in solar desalination by using solar still with floatable structure. However, there is only a few works carried outdoor experiments by using solar stills. It is found that although the evaporation rate is very high, low productivities and efficiencies are obtained for solar still. The daily productivity of solar still is usually about 1-4 L/day and the corresponding efficiency is below 40% under natural sun, which is far below the evaporation efficiency (usually as high as 60%-90%) [77, 89, 94]. It shows that there is a large research gap between solar evaporation and solar desalination using solar still. The reasons of lower productivity and efficiency in solar still are usually attributed to the lower solar intensity of natural sunlight, the optical impedance by condensed water drops and low condensation rate of vapor [94a, 95]. In the laboratory, the solar intensity can be maintained at from one sun to as high as more than ten suns. However, the solar intensity of the natural sun, varies from 0 at night to around one sun at noon. The evaporation efficiency will decrease when the solar intensity decreases, due to the lower temperature reached [89]. Meanwhile, when vapor condenses on the transparent cover as a droplet, the reflectance of cover will be increased, which reduces the solar energy received. The optical impedance might be avoided by using a more hydrophilic cover or a vapor-facing-down design as suggested by Liu et al. [95]. On the other hand, the low condensation rate of water vapor is due to the high cover temperature and insufficient condensation area [25a, 96]. 26 Fig. 9 (a) The efficiency versus the air-water interface temperature (details in Table S2). (b) Calculated productivity at different water temperature and glass temperature. Besides, there is another essential reason for low productivity and efficiency in solar still, which is usually ignored by most researchers: the low air-water interface temperature. In most of recent works, the temperature of the air-water interface in solar evaporation is usually maintained very low to decrease the heat loss and therefore enhance the energy efficiency. The temperature of the air-water interface under one sun is summarized at Fig. 9 (a) and Table S2. The air-water interface temperature in most works is at around 30-50 °C with the efficiency at around 70-90%. Only few of them reported a temperature of more than 80 °C [64, 68, 97]. But, for solar desalination using solar still, a higher air-water interface temperature will lead to higher productivity and efficiency [14, 98]. A very low air-water interface temperature will result in very low productivity and efficiency in solar still, although a good evaporation rate may be achieved. This is because, the vapor with lower temperature is more difficult to condense on the condensation plate (glass cover) due to the exponentially decreased difference of vapor pressure. Meanwhile, only a few parts of water can be extracted from the vapor when the temperature of vapor is close to that of the condensation plate. To illustrate this point, the productivity of solar still at different water temperature is calculated according to the half empirical equation (1) [7, 99]: (1) where is the heat transfer rate between glass and water by condensation, which is 27 proportional to the productivity of solar still. and are the temperature of water evaporation surface and inner surface of glass cover, respectively. is the saturation vapor pressure of water surface at temperature . is saturation vapor pressure of the inner surface of glass cover at temperature (2) (3) For a given temperature difference between the water surface and glass cover, the productivity increase with the increase in water temperature (Fig. 9b). The productivity almost doubled when the water temperature increases 10°C. The productivity of solar still is unable to reach a high value if the water temperature is lower than 40°C, due to the limited heat and mass transfer rate between water and glass cover. Therefore, materials and thermal designs for both high energy efficiency and high air-water interface temperature are very necessary in the future. It's not only important to increase the productivity of solar still, but also important to some other processes that solar evaporation is involved, such as sterilization [100], latent heat recovery [101] and power generation [54], where high vapor temperature is preferred. 4.2 Salt rejecting and durability Besides the unsatisfied productivity and efficiency of solar still by using foams and films, salt rejecting is another issue that draws a lot of attention. The rapid loss of water in the air-water interface leads to a significant increase in local concentration of salt. When the concentration of salt is saturated, the crystallized salt will reflect the solar irradiation and block the evaporation surface, and then slow down the evaporation[89, 102]. Therefore, whether salt crystallization will occur and affect the evaporation or not must be considered when design materials for solar still. The reported experiments about salt rejecting can be divided into two categories, cycle test and continuous test, shown in Table 2. In the cycle test, the materials are dried 28 or washed before the next cycle [75, 103], each cycle sustains 0.5-5 hours. Most of the cycle test shows that there is no salt crystallization which might be due to that the salt is removed timely or the good salt rejecting ability of materials. Although salt crystallization is observed in some works, the evaporation is not affected and remains stable [103a, 104]. Nevertheless, it is inconvenient to wash or dry materials circularly in solar still. Therefore, carrying out the long-time continuous test is necessary to further explore the salt rejecting performance of foams and films. It is found that salt crystallization is easy to occur at long-term continues test [102b, 105]. To avoid the salt crystallization, the high concentration of salt on the evaporation interface must be diluted or removed timely. One way is using materials with relatively larger holes (millimeter scale), which ensures that the generated salt particles at the interface can fall to the bulk liquid through holes instead of accumulated on the interface [63a]. The holes also provide dilute solution which exchanges salt ions with the high concentration brine on the interface to avoid salt crystallization, as shown in Fig 10a [106]. Janus structure is another effective strategy to solve salt crystallization (Fig 10b) [102a]. In Janus structure, solar absorption and water pumping, are decoupled into different layers. There is an upper hydrophobic layer (CB/PMMA) for light absorption, and a lower hydrophilic layer (PAN) for pumping water. Therefore, salt can be only deposited in the hydrophilic layer and be dissolved quickly due to continuous water pumping. A more thorough way to overcome salt crystallization on evaporation interface is utilizing contactless structure [97b]. In this method, the solar absorbing structure absorbs solar radiation and re-radiates infrared photons to heat the water (Fig 10 c). The heat transfer between the solar absorbing structure and water is completely by thermal irradiation, instead of heat conduction or convection. Therefore, fouling is entirely avoided due to the physical separation from the water. However, the evaporation efficiency of the aforementioned three strategy is relatively low (25%-75%). A better way with high efficiency might be enhancing the water absorption ability of materials. For example, due to the free solutions exchange enabled by strong capillary effect, melamine resin sponge achieves an evaporation 29 efficiency of 85% under one sun and 90% under ten suns without salt crystallization after 11 hours of continuing test [105]. Hydrogel can also achieve a water evaporation rate of ∼2.6 kg/(m2·h) at ∼91% energy efficiency under one sun, without salt crystallization after 100 hours of continuing test [107]. Therefore, improving the water absorption ability of materials might be a very potential way to enhance the salt rejecting performance without losing energy efficiency. Fig. 10 (a) a diagram showing new design self-regenerating solar evaporator (left), and multipath ways mass transfer in the evaporator (right) [106], Copyright 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) Structures of Janus absorber during solar desalination [102a] Copyright 2018 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim (c) Mechanism of contactless solar evaporation structure. As the absorber heats up, it emits thermal radiation to the water, which is absorbed beneath the water/vapor interface [97b]. Copyright Springer Nature. 30 Categories Durations of test Total illuminated time Salt crystallization Water type Solar intensity Table 2: Durability and salt rejecting performance of materials 15 cycles [103b] 10 cycles [77] 20 cycles [75] 10 cycles [55a] 10 cycles [68] 50 cycles [108] 15 cycles [104] 100 cycles [103a] 24 cycles [66a] 20 hours [63a] 100 hours [106] 10 hours [92] 10 hours [74b] 48 hours [107] 10 hours [109] 100 hours [79a] 11 hours [105] 28 days [76] 16 days [102a] 30 days [110] 25 days [111] of test Circular test Continuous test Immersing- Circular test /hours Unknown 10 Unknown 5 50 No Yes No No No Salt water (3.5 wt.% NaCl) Salt water (3.5 wt.% NaCl) Sea water (salinity 2.75%) Salt water (3.5 wt.% NaCl) Sea water (East Sea) Unknown Unknown Salt water (salinity unknown) 15 100 24 20 100 10 10 48 10 100 11 11 12 30 Yes Yes No No No Yes Salt water (3.5 wt.% NaCl) Salt water (3.5 wt.% NaCl) Sea water (Bohai Sea) Salt water (15 wt.% NaCl) Salt water (20 wt.% NaCl) Sea water (salinity 3.5%) Unknown Sea water (East Sea) No No Yes No No No No Sea water (salinity 3.5%) Salt water (3.5 wt.% NaCl) Sea water (Chesapeake Bay) Salt water (3.5 wt.% NaCl) Sea water (Gulf of Mexico) Salt water (3.5 wt.% NaCl) Sea water (Yellow Sea) Unknown Unknown Sea water (East Sea) /kw 1 1 1 1 1 1 1 1 2 1 1 1 1 1 2 5 10 1 1 1 2 31 The durability of materials is another essential point in solar still. The cost and difficulty of maintenance will be increased and the practical application will be limited if materials are required to be replaced frequently, especially for some remote and developing areas. Sea water contains not only salt but also some other contaminants such as minerals, bacteria, and sands, which make the durability of materials more challenging. Similar to the salt rejecting experiments, most of the durability results are obtained by cycle test or continuous test in salt water or sea water (shown in Table 2). Although the evaporation rate is stable, the duration remains just a couple of days, which is too short to prove the durability of materials. Therefore, several works extended their durability test to several weeks by immersing the samples in seawater [76, 102a, 110-111]. After days or weeks of storage, the samples were taken out and irradiated under one sun irradiation to verify the salt resistance by measuring the evaporation rate. Stable evaporation is observed and no obvious degradation of materials appeared in these tests. Although the samples contact with sea water for weeks, the total time of materials under solar irradiation is only tens of hours. Durability test under both long times of solar irradiation and seawater environment is still lacking. Therefore, the durability of materials needs more investigations to meet the requirement of practical application in solar still, where solar irradiation and seawater environment might last for months. To simulate the real work condition, it is necessary to carry out long-term experiments by testing the performance of materials in outdoor solar stills or vapor generators for weeks or months. 4.3 Effect of solid-liquid interaction on water phase change Traditional phase change model only focuses on vapor-liquid interaction. The Hertz-Knudsen (HK) equation and its modified forms have proved very effective to predict the phase change rate from droplets [112]. However, when water evaporates from a porous structure, the solid-liquid interaction becomes significant, due to the abundant solid-liquid interfaces and the short distance between the evaporation surface and solid-liquid interface. Many important details which are helpful to 32 manipulate phase change process will be ignored by only focus on vapor-liquid interaction. Thin-film evaporation theory proves the importance of solid-liquid interaction from the aspect of heat transfer at the microscale level[113]. At the hydrophilic surface of porous foams or films, the water channel is filled with water meniscuses, which is constructed by three regions: (I) adsorbed or non-evaporation region, where water is adsorbed on the graphite due to the high disjoining pressure; (II) thin-film or transition region where effects of long-range molecular forces are felt; (III) intrinsic meniscus region, where the thickness of the water layer increases very fast [114]. In the adsorbed region, water sticks to the graphite tightly, and no mass/heat transfer occurs. Whereas in the thin-film region, the disjoin pressure is weak, while the thickness of the water layer is still thin enough to assure a low thermal resistance. Therefore, the most heat current runs through the thin-film region, hence the fast evaporation and low heat loss [115]. Although the area of thin film region is relatedly small compared to intrinsic meniscus region, the evaporation heat transfer may account for more than 80 % of the total evaporation heat transfer [113a]. At the nano or molecular level, solid-liquid interaction endows fluids unconventional structural properties and dynamical behaviors, which greatly affect the phase change process. For example, the evaporation from water capillary in two-dimensional nanochannels surpasses the traditional Hertz -- Knudsen limit by an order of magnitude [116]. The solar evaporation assisted by nanomaterials also shows unconventional phenomenon. The solar evaporation rate under one sun is able to be 110% higher than the traditional limit due to the reduction of phase change enthalpy, which shows a great opportunity for the further improvement of phase change based solar desalination[76, 102b]. One possible reason of the enhanced evaporation is "cluster evaporation", i.e. water evaporates as molecular cluster, instead of evaporates one molecular by one molecular. However, this explanation is questionable, due to the lack of convincible fundamental theory and experimental observation of water clusters undergoing "evaporation"[95]. Another possible reason is the reduced free energy barrier for evaporation across the liquid-vapor interface [117]. Molecular 33 dynamics reveals that the graphene edges of nanopores can boost the overall evaporative flux by more than 100%. The interaction between the water molecules is weaken by the charged edges of graphene nanopores. This local-field-driven dipole moment destabilizes the H-bond network of water at the interface, and may responsible for the enhancement in the evaporation rate. Nevertheless, many works have to be done to give us a reliable fundamental theory in this field, which is very important for the development of evaporation and solar desalination. 5. Conclusions In this paper, a detailed review of current developments in solar stills with nano/micro materials is presented. Most of recent efforts were interested in improving solar evaporation, which is just one of basic process in solar still. Various materials, such as paper-based film, artificial aerogel and natural biomaterials, were applied to enhance the system efficiency and productivity. By combining heat localization at air-water interface and optimized thermal design, proper materials might achieve an efficiency more than 90% under 1 kw/m2 of solar irradiation. However, the productivity remains low when these materials were applied. This show the large gap between solar still and solar evaporation systems. Except lower natural solar irradiation and optical impedance by condensate droplets, the low water and vapor temperature might be another important reason in these evaporation systems. Meanwhile, salt rejecting and durability of materials are also two important challenges for the practical application of nano/micro materials. Enhance the water absorption ability of materials might be a very effective way to avoid salt crystallization. Besides the engineering and materials studies, more fundamental research is still missing and should be done to uncover the phase change mechanism at nano and molecular level. Especially, the new understanding on effect of solid-liquid interaction might open a new avenue in solar desalination field. 34 Acknowledgments N.Y. was sponsored by National Natural Science Foundation of China (No. 51576076 and No. 51711540031), Natural Science Foundation of Hubei Province (2017CFA046) and Fundamental Research Funds for the Central Universities (2019kfyRCPY045). M. A. was supported by Natural Science Research Start-up Fund of Shaanxi University of Science and Technology (2018GB-10). 35 References [1] [2] [3] [4] [5] [6] [7] [8] [9] M. Elimelech, W. A. Phillip, Science 2011, 333, 712. M. S. S. Abujazar, S. Fatihah, E. R. Lotfy, A. E. Kabeel, S. Sharil, Desalination 2018, 425, 94. L. Sahota, G. N. Tiwari, Sol Energy 2016, 130, 260. P. Vishwanath Kumar, A. Kumar, O. Prakash, A. K. Kaviti, Renewable and Sustainable Energy Reviews 2015, 51, 153. M. Gao, L. Zhu, C. K. Peh, G. W. Ho, Energ Environ Sci 2019, 12, 841. A. E. Kabeel, M. Abdelgaied, M. B. Feddaoui, Desalination 2018, 439, 162. S. W. Sharshir, N. Yang, G. Peng, A. E. Kabeel, Appl Therm Eng 2016, 100, 267. K. Kalidasa Murugavel, K. Srithar, Renew Energ 2011, 36, 612. B. A. K. Abu-Hijleh, H. M. Rababa'h, Energ Convers Manage 2003, 44, 1411. [10] M. M. Naim, M. A. Abd El Kawi, Desalination 2003, 153, 55. [11] R. Sathyamurthy, S. A. El-Agouz, P. K. Nagarajan, J. Subramani, T. Arunkumar, D. Mageshbabu, B. Madhu, R. Bharathwaaj, N. Prakash, Renewable and Sustainable Energy Reviews 2017, 77, 1069. [12] [13] A. Rahmani, A. Boutriaa, Int J Hydrogen Energ 2017, 42, 29047. Z. M. Omara, A. E. Kabeel, A. S. Abdullah, Renewable and Sustainable Energy Reviews 2017, 68, 638. [14] P. Durkaieswaran, K. K. Murugavel, Renewable and Sustainable Energy Reviews 2015, 49, 1048. [15] a) V. Trisaksri, S. Wongwises, Renewable and Sustainable Energy Reviews 2007, 11, 512; b) A. H. Elsheikh, S. W. Sharshir, M. E. Mostafa, F. A. Essa, M. K. Ahmed Ali, Renewable and Sustainable Energy Reviews 2018, 82, 3483. [16] M. Gao, P. K. N. Connor, G. W. Ho, Energ Environ Sci 2016, 9, 3151. [17] [18] [19] [20] [21] V.-D. Dao, H.-S. Choi, Global Challenges 2018, 2, 1700094. X. Yin, Y. Zhang, Q. Guo, X. Cai, J. Xiao, Z. Ding, J. Yang, ACS Appl Mater Interfaces 2018, 10, 10998. Y. Wang, L. Zhang, P. Wang, Acs Sustain Chem Eng 2016, 4, 1223. O. Neumann, A. S. Urban, J. Day, S. Lal, P. Nordlander, N. J. Halas, Acs Nano 2012, 7, 42. H. Ghasemi, G. Ni, A. M. Marconnet, J. Loomis, S. Yerci, N. Miljkovic, G. Chen, Nat Commun 2014, 5, 4449. [22] G. Ni, N. Miljkovic, H. Ghasemi, X. Huang, S. V. Boriskina, C.-T. Lin, J. Wang, Y. Xu, M. M. [23] [24] [25] Rahman, T. Zhang, G. Chen, Nano Energy 2015, 17, 290. T. Elango, A. Kannan, K. Kalidasa Murugavel, Desalination 2015, 360, 45. L. Sahota, G. N. Tiwari, Desalination 2016, 388, 9. a) S. W. Sharshir, G. Peng, L. Wu, N. Yang, F. A. Essa, A. H. Elsheikh, S. I. T. Mohamed, A. E. Kabeel, Appl Therm Eng 2017, 113, 684; b) S. W. Sharshir, G. Peng, A. H. Elsheikh, E. M. A. Edreis, M. A. Eltawil, T. Abdelhamid, A. E. Kabeel, J. Zang, N. Yang, Energ Convers Manage 2018, 177, 363. [26] S. W. Sharshir, G. Peng, L. Wu, F. A. Essa, A. E. Kabeel, N. Yang, Appl Energ 2017, 191, 358. [27] a) A. E. Kabeel, Z. M. Omara, F. A. Essa, Energ Convers Manage 2014, 78, 493; b) A. E. Kabeel, Z. M. Omara, F. A. Essa, Energ Convers Manage 2014, 86, 268. 36 [28] [29] [30] [31] Z. M. Omara, A. E. Kabeel, F. A. Essa, Energ Convers Manage 2015, 103, 965. O. Mahian, A. Kianifar, S. Z. Heris, D. Wen, A. Z. Sahin, S. Wongwises, Nano Energy 2017, 36, 134. L. Sahota, Shyam, G. N. Tiwari, Energ Convers Manage 2017, 135, 308. G. P. Celata, F. D'Annibale, A. Mariani, S. Sau, E. Serra, R. Bubbico, C. Menale, H. Poth, Chemical Engineering Research and Design 2014, 92, 1616. [32] R. Bubbico, G. P. Celata, F. D'Annibale, B. Mazzarotta, C. Menale, Chemical Engineering Research and Design 2015, 104, 605. [33] a) S. M. Shinde, D. M. Kawadekar, P. A. Patil, V. K. Bhojwani, International Journal of Ambient Energy 2019, DOI: 10.1080/01430750.2019.16211981; b) S. Shamshirband, A. Malvandi, A. Karimipour, M. Goodarzi, M. Afrand, D. Petković, M. Dahari, N. Mahmoodian, Powder Technol 2015, 284, 336. [34] J.-Y. Sha, H.-H. Ge, C. Wan, L.-T. Wang, S.-Y. Xie, X.-J. Meng, Y.-Z. Zhao, Corros Sci 2019, 148, 123. [35] B. G. N. Muthanna, M. Amara, M. H. Meliani, B. Mettai, Ž. Božić, R. Suleiman, A. A. Sorour, Eng Fail Anal 2019, 102, 293. N. Sezer, M. A. Atieh, M. Koç, Powder Technol 2019, 344, 404. R. A. Taylor, P. E. Phelan, R. J. Adrian, A. Gunawan, T. P. Otanicar, Int J Therm Sci 2012, 56, [36] [37] 1. [38] W. Duangthongsuk, S. Wongwises, Int J Heat Mass Tran 2010, 53, 334. [39] [40] L. Zhou, S. Zhuang, C. He, Y. Tan, Z. Wang, J. Zhu, Nano Energy 2017, 32, 195. Y. Wang, M. E. Zaytsev, H. L. The, J. C. Eijkel, H. J. Zandvliet, X. Zhang, D. Lohse, Acs Nano 2017, 11, 2045. [41] V. Kotaidis, C. Dahmen, G. von Plessen, F. Springer, A. Plech, J Chem Phys 2006, 124, [42] [43] 184702. D. Lapotko, Opt Express 2009, 17, 2538. a) Z. Fang, Y. R. Zhen, O. Neumann, A. Polman, F. J. Garcia de Abajo, P. Nordlander, N. J. Halas, Nano Lett 2013, 13, 1736; b) O. Neumann, C. Feronti, A. D. Neumann, A. Dong, K. Schell, B. Lu, E. Kim, M. Quinn, S. Thompson, N. Grady, P. Nordlander, M. Oden, N. J. Halas, Proc Natl Acad Sci U S A 2013, 110, 11677. [44] N. J. Hogan, A. S. Urban, C. Ayala-Orozco, A. Pimpinelli, P. Nordlander, N. J. Halas, Nano Lett 2014, 14, 4640. H. Jin, G. Lin, L. Bai, A. Zeiny, D. Wen, Nano Energy 2016, 28, 397. a) H. Wang, L. Miao, S. Tanemura, Solar RRL 2017, 1, 1600023; b) M. S. Zielinski, J. W. Choi, T. La Grange, M. Modestino, S. M. Hashemi, Y. Pu, S. Birkhold, J. A. Hubbell, D. Psaltis, Nano Lett 2016, 16, 2159. X. Wang, Y. He, X. Liu, L. Shi, J. Zhu, Sol Energy 2017, 157, 35. Y.-T. Yen, C.-W. Chen, M. Fang, Y.-Z. Chen, C.-C. Lai, C.-H. Hsu, Y.-C. Wang, H. Lin, C.-H. Shen, J.-M. Shieh, J. C. Ho, Y.-L. Chueh, Nano Energy 2015, 15, 470. A. Guo, Y. Fu, G. Wang, X. Wang, Rsc Adv 2017, 7, 4815. A. Zeiny, H. Jin, G. Lin, P. Song, D. Wen, Renew Energ 2018, 122, 443. X. Wang, G. Ou, N. Wang, H. Wu, ACS Appl Mater Interfaces 2016, 8, 9194. B. Sharma, M. K. Rabinal, J Alloy Compd 2017, 690, 57. Y. Fu, T. Mei, G. Wang, A. Guo, G. Dai, S. Wang, J. Wang, J. Li, X. Wang, Appl Therm Eng 37 [45] [46] [47] [48] [49] [50] [51] [52] [53] 2017, 114, 961. [54] P. Tao, G. Ni, C. Song, W. Shang, J. Wu, J. Zhu, G. Chen, T. Deng, Nature Energy 2018, DOI: 10.1038/s41560-018-0260-7. [55] a) Z. Wang, Q. Ye, X. Liang, J. Xu, C. Chang, C. Song, W. Shang, J. Wu, P. Tao, T. Deng, J Mater Chem A 2017, 5, 16359; b) X. Liu, B. Hou, G. Wang, Z. Cui, X. Zhu, X. Wang, J Mater Res 2018, 33, 674. [56] J. Lou, Y. Liu, Z. Wang, D. Zhao, C. Song, J. Wu, N. Dasgupta, W. Zhang, D. Zhang, P. Tao, W. Shang, T. Deng, ACS Appl Mater Interfaces 2016, 8, 14628. G. Wang, Y. Fu, X. Ma, W. Pi, D. Liu, X. Wang, Carbon 2017, 114, 117. Y. Liu, S. Yu, R. Feng, A. Bernard, Y. Liu, Y. Zhang, H. Duan, W. Shang, P. Tao, C. Song, T. [57] [58] Deng, Adv Mater 2015, 27, 2768. [59] a) Z. Wang, Y. Liu, P. Tao, Q. Shen, N. Yi, F. Zhang, Q. Liu, C. Song, D. Zhang, W. Shang, T. Deng, Small 2014, 10, 3234; b) Y. Liu, J. Lou, M. Ni, C. Song, J. Wu, N. P. Dasgupta, P. Tao, W. Shang, T. Deng, ACS Appl Mater Interfaces 2016, 8, 772. [60] [61] K. Bae, G. Kang, S. K. Cho, W. Park, K. Kim, W. J. Padilla, Nat Commun 2015, 6, 10103. J. Wang, Z. Liu, X. Dong, C.-E. Hsiung, Y. Zhu, L. Liu, Y. Han, J Mater Chem A 2017, 5, 6860. [62] F. Chen, A. S. Gong, M. Zhu, G. Chen, S. D. Lacey, F. Jiang, Y. Li, Y. Wang, J. Dai, Y. Yao, J. Song, B. Liu, K. Fu, S. Das, L. Hu, Acs Nano 2017, 11, 4275. [63] a) V. Kashyap, A. Al-Bayati, S. M. Sajadi, P. Irajizad, S. H. Wang, H. Ghasemi, J Mater Chem A 2017, 5, 15227; b) X. Wang, Y. He, X. Liu, G. Cheng, J. Zhu, Appl Energ 2017, 195, 414; c) M. Chen, Y. Wu, W. Song, Y. Mo, X. Lin, Q. He, B. Guo, Nanoscale 2018, 10, 6186; d) M. Kaur, S. Ishii, S. L. Shinde, T. Nagao, Acs Sustain Chem Eng 2017, 5, 8523. Y. Ito, Y. Tanabe, J. Han, T. Fujita, K. Tanigaki, M. Chen, Adv Mater 2015, 27, 4302. R. Li, L. Zhang, L. Shi, P. Wang, Acs Nano 2017, 11, 3752. a) L. Zhou, Y. Tan, J. Wang, W. Xu, Y. Yuan, W. Cai, S. Zhu, J. Zhu, Nat Photonics 2016, 10, 393; b) L. Zhou, Y. Tan, D. Ji, B. Zhu, P. Zhang, J. Xu, Q. Gan, Z. Yu, J. Zhu, Sci. Adv. 2016, 2, [64] [65] [66] e1501227 [67] X. Yang, Y. Yang, L. Fu, M. Zou, Z. Li, A. Cao, Q. Yuan, Adv Funct Mater 2018, 28, 1704505. [68] [69] [70] [71] G. Zhu, J. Xu, W. Zhao, F. Huang, ACS Appl Mater Interfaces 2016, 8, 31716. P. Fan, H. Wu, M. Zhong, H. Zhang, B. Bai, G. Jin, Nanoscale 2016, 8, 14617. P. Zhang, J. Li, L. Lv, Y. Zhao, L. Qu, Acs Nano 2017, 11, 5087. L. Tian, J. Luan, K. K. Liu, Q. Jiang, S. Tadepalli, M. K. Gupta, R. R. Naik, S. Singamaneni, Nano Lett 2016, 16, 609. [72] a) F. Jiang, H. Liu, Y. Li, Y. Kuang, X. Xu, C. Chen, H. Huang, C. Jia, X. Zhao, E. Hitz, Y. Zhou, R. Yang, L. Cui, L. Hu, ACS Appl Mater Interfaces 2018, 10, 1104; b) Q. Jiang, H. Gholami Derami, D. Ghim, S. Cao, Y.-S. Jun, S. Singamaneni, J Mater Chem A 2017, 5, 18397. [73] Y. Yang, R. Zhao, T. Zhang, K. Zhao, P. Xiao, Y. Ma, P. M. Ajayan, G. Shi, Y. Chen, Acs Nano 2018, 12, 829. [74] a) X. Hu, W. Xu, L. Zhou, Y. Tan, Y. Wang, S. Zhu, J. Zhu, Adv Mater 2017, 29; b) P. Zhang, Q. Liao, T. Zhang, H. Cheng, Y. Huang, C. Yang, C. Li, L. Jiang, L. Qu, Nano Energy 2018, 46, 415. [75] H. Ren, M. Tang, B. Guan, K. Wang, J. Yang, F. Wang, M. Wang, J. Shan, Z. Chen, D. Wei, H. 38 Peng, Z. Liu, Adv Mater 2017, 29, 1702590. [76] F. Zhao, X. Zhou, Y. Shi, X. Qian, M. Alexander, X. Zhao, S. Mendez, R. Yang, L. Qu, G. Yu, Nat Nanotechnol 2018, 13, 489. [77] Z. Liu, H. Song, D. Ji, C. Li, A. Cheney, Y. Liu, N. Zhang, X. Zeng, B. Chen, J. Gao, Y. Li, X. Liu, D. Aga, S. Jiang, Z. Yu, Q. Gan, Global Chall 2017, 1, 1600003. [78] X. Gao, H. Ren, J. Zhou, R. Du, C. Yin, R. Liu, H. Peng, L. Tong, Z. Liu, J. Zhang, Chem Mater 2017, 29, 5777. [79] a) M. Zhu, Y. Li, G. Chen, F. Jiang, Z. Yang, X. Luo, Y. Wang, S. D. Lacey, J. Dai, C. Wang, C. Jia, J. Wan, Y. Yao, A. Gong, B. Yang, Z. Yu, S. Das, L. Hu, Adv Mater 2017, 29; b) G. Xue, K. Liu, Q. Chen, P. Yang, J. Li, T. Ding, J. Duan, B. Qi, J. Zhou, ACS Appl Mater Interfaces 2017, 9, 15052; c) N. Xu, X. Hu, W. Xu, X. Li, L. Zhou, S. Zhu, J. Zhu, Adv Mater 2017, 29. [80] H. Liu, C. Chen, G. Chen, Y. Kuang, X. Zhao, J. Song, C. Jia, X. Xu, E. Hitz, H. Xie, S. Wang, F. Jiang, T. Li, Y. Li, A. Gong, R. Yang, S. Das, L. Hu, Adv Energy Mater 2018, 8, 1701616. [81] K. K. Liu, Q. Jiang, S. Tadepalli, R. Raliya, P. Biswas, R. R. Naik, S. Singamaneni, ACS Appl Mater Interfaces 2017, 9, 7675. [82] C. Chen, Y. Li, J. Song, Z. Yang, Y. Kuang, E. Hitz, C. Jia, A. Gong, F. Jiang, J. Y. Zhu, B. Yang, J. Xie, L. Hu, Adv Mater 2017, 29. [83] M. Zhu, Y. Li, F. Chen, X. Zhu, J. Dai, Y. Li, Z. Yang, X. Yan, J. Song, Y. Wang, E. Hitz, W. Luo, M. Lu, B. Yang, L. Hu, Adv Energy Mater 2018, 8, 1701028. [84] X. Wu, G. Y. Chen, W. Zhang, X. Liu, H. Xu, Advanced Sustainable Systems 2017, 1, 1700046. [85] C. Jia, Y. Li, Z. Yang, G. Chen, Y. Yao, F. Jiang, Y. Kuang, G. Pastel, H. Xie, B. Yang, S. Das, L. Hu, Joule 2017, 1, 588. [86] J. Liu, Q. Liu, D. Ma, Y. Yuan, J. Yao, W. Zhang, H. Su, Y. Su, J. Gu, D. Zhang, J Mater Chem A 2019, 7, 9034. [87] S. Zhuang, L. Zhou, W. Xu, N. Xu, X. Hu, X. Li, G. Lv, Q. Zheng, S. Zhu, Z. Wang, J. Zhu, Adv Sci (Weinh) 2018, 5, 1700497. [88] X. Li, W. Xu, M. Tang, L. Zhou, B. Zhu, S. Zhu, J. Zhu, Proc Natl Acad Sci U S A 2016, 113, 13953. [89] G. Ni, S. H. Zandavi, S. M. Javid, S. V. Boriskina, T. A. Cooper, G. Chen, Energ Environ Sci 2018, 11, 1510. [90] Y. Li, T. Gao, Z. Yang, C. Chen, Y. Kuang, J. Song, C. Jia, E. M. Hitz, B. Yang, L. Hu, Nano Energy 2017, 41, 201. [91] [92] X. Li, R. Lin, G. Ni, N. Xu, X. Hu, B. Zhu, G. Lv, J. Li, S. Zhu, J. Zhu, Natl Sci Rev 2018, 5, 70. P.-F. Liu, L. Miao, Z. Deng, J. Zhou, H. Su, L. Sun, S. Tanemura, W. Cao, F. Jiang, L.-D. Zhao, Materials Today Energy 2018, 8, 166. [93] a) L. Breiman, Mach Learn 2001, 45, 5; b) K. T. Butler, D. W. Davies, H. Cartwright, O. Isayev, A. Walsh, Nature 2018, 559, 547; c) R. Ma, D. Huang, T. Zhang, T. Luo, Chem Phys Lett 2018, 704, 49; d) C. Maltecca, D. Lu, C. Schillebeeckx, N. P. McNulty, C. Schwab, C. Shull, F. Tiezzi, Scientific Reports 2019, 9, 6574; e) S. Subramanian, S. Huq, T. Yatsunenko, R. Haque, M. Mahfuz, M. A. Alam, A. Benezra, J. DeStefano, M. F. Meier, B. D. Muegge, M. J. Barratt, L. G. VanArendonk, Q. Zhang, M. A. Province, W. A. Petri Jr, T. Ahmed, J. I. Gordon, Nature 2014, 510, 417. [94] a) L. Yi, S. Ci, S. Luo, P. Shao, Y. Hou, Z. Wen, Nano Energy 2017, 41, 600; b) L. Zhang, B. 39 Tang, J. Wu, R. Li, P. Wang, Adv Mater 2015, 27, 4889. [95] H. Liu, Z. Huang, K. Liu, X. Hu, J. Zhou, Adv Energy Mater 2019, DOI: 10.1002/aenm.2019003101900310. [96] [97] R. Bhardwaj, M. V. ten Kortenaar, R. F. Mudde, Appl Energ 2015, 154, 480. a) G. Ni, G. Li, Svetlana V. Boriskina, H. Li, W. Yang, T. Zhang, G. Chen, Nature Energy 2016, 1, 16126; b) T. A. Cooper, S. H. Zandavi, G. W. Ni, Y. Tsurimaki, Y. Huang, S. V. Boriskina, G. Chen, Nat Commun 2018, 9, 5086. [98] a) A. E. Kabeel, Z. M. Omara, M. M. Younes, Renewable and Sustainable Energy Reviews 2015, 46, 178; b) A. K. Kaviti, A. Yadav, A. Shukla, Renewable and Sustainable Energy Reviews 2016, 54, 429; c) K. H. Nayi, K. V. Modi, Renewable and Sustainable Energy Reviews 2018, 81, 136; d) S. W. Sharshir, A. H. Elsheikh, G. Peng, N. Yang, M. O. A. El-Samadony, A. E. Kabeel, Renewable and Sustainable Energy Reviews 2017, 73, 521. [99] V. Manikandan, K. Shanmugasundaram, S. Shanmugan, B. Janarthanan, J. Chandrasekaran, Renewable and Sustainable Energy Reviews 2013, 20, 322. [100] a) Y. Zhang, D. Zhao, F. Yu, C. Yang, J. Lou, Y. Liu, Y. Chen, Z. Wang, P. Tao, W. Shang, J. Wu, C. Song, T. Deng, Nanoscale 2017, 9, 19384; b) J. Li, M. Du, G. Lv, L. Zhou, X. Li, L. Bertoluzzi, C. Liu, S. Zhu, J. Zhu, Adv Mater 2018, DOI: 10.1002/adma.201805159e1805159. [101] G. Xue, Q. Chen, S. Lin, J. Duan, P. Yang, K. Liu, J. Li, J. Zhou, Global Challenges 2018, 2, 1800001. [102] a) W. Xu, X. Hu, S. Zhuang, Y. Wang, X. Li, L. Zhou, S. Zhu, J. Zhu, Adv Energy Mater 2018, 8, 1702884; b) F. Liu, B. Zhao, W. Wu, H. Yang, Y. Ning, Y. Lai, R. Bradley, Adv Funct Mater 2018, 28, 1803266. [103] a) N. Han, K. Liu, X. Zhang, M. Wang, P. Du, Z. Huang, D. Zhou, Q. Zhang, T. Gao, Y. Jia, L. Luo, J. Wang, X. Sun, Sci Bull 2019, 64, 391; b) F. Gong, H. Li, W. Wang, J. Huang, D. Xia, J. Liao, M. Wu, D. V. Papavassiliou, Nano Energy 2019, 58, 322. [104] H. Su, J. Zhou, L. Miao, J. Shi, Y. Gu, P. Wang, Y. Tian, X. Mu, A. Wei, L. Huang, S. Chen, Z. Deng, Sustainable Materials and Technologies 2019, 20, e00095. [105] Y. Liu, Z. Liu, Q. Huang, X. Liang, X. Zhou, H. Fu, Q. Wu, J. Zhang, W. Xie, J Mater Chem A 2019, 7, 2581. [106] Y. Kuang, C. Chen, S. He, E. M. Hitz, Y. Wang, W. Gan, R. Mi, L. Hu, Adv Mater 2019, DOI: 10.1002/adma.201900498e1900498. [107] Y. Guo, F. Zhao, X. Zhou, Z. Chen, G. Yu, Nano Lett 2019, 19, 2530. [108] P. Qiao, J. Wu, H. Li, Y. Xu, L. Ren, K. Lin, W. Zhou, ACS Appl Mater Interfaces 2019, 11, 7066. [109] P.-F. Liu, L. Miao, Z. Deng, J. Zhou, Y. Gu, S. Chen, H. Cai, L. Sun, S. Tanemura, Appl Energ 2019, 241, 652. [110] C. Li, D. Jiang, B. Huo, M. Ding, C. Huang, D. Jia, H. Li, C.-Y. Liu, J. Liu, Nano Energy 2019, 60, 841. [111] Q. Gan, T. Zhang, R. Chen, X. Wang, M. Ye, Acs Sustain Chem Eng 2019, 7, 3925. [112] A. H. Persad, C. A. Ward, Chem Rev 2016, 116, 7727. [113] a) R. Ranjan, J. Y. Murthy, S. V. Garimella, Int J Heat Mass Tran 2011, 54, 169; b) F. Su, H. Ma, X. Han, H. H. Chen, B. Tian, Appl Phys Lett 2012, 101, 113702; c) J. Yang, Y. Pang, W. Huang, S. K. Shaw, J. Schiffbauer, M. A. Pillers, X. Mu, S. Luo, T. Zhang, Y. Huang, G. Li, S. 40 Ptasinska, M. Lieberman, T. Luo, Acs Nano 2017, 11, 5510. [114] H. Wang, S. V. Garimella, J. Y. Murthy, Int J Heat Mass Tran 2007, 50, 3933. [115] G. Peng, H. Ding, S. W. Sharshir, X. Li, H. Liu, D. Ma, L. Wu, J. Zang, H. Liu, W. Yu, H. Xie, N. Yang, Appl Therm Eng 2018, 143, 1079. [116] Y. Li, M. A. Alibakhshi, Y. Zhao, C. Duan, Nano Lett 2017, 17, 4813. [117] S. Feng, Z. Xu, Nanotechnology 2019, 30, 165401. 41 Supporting information Micro/nanomaterials for improving solar still and solar evaporation - Guilong Peng1,#, Swellam W. Sharshir1,2,3, #, Yunpeng Wang1, Meng An4, A.E. Kabeel5, A review Jianfeng Zang2, Lifa Zhang6, Nuo Yang1 1 State Key Laboratory of Coal Combustion, School of Energy and Power Engineering, Huazhong University of Science and Technology, Wuhan 430074, China 2 School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan 430074, China 3 Mechanical Engineering Department, Faculty of Engineering, Kafrelsheikh University, Kafrelsheikh, Egypt 4 College of Mechanical and Electrical Engineering, Shaanxi University of Science and Technology, Xi'an, 710021, China 5 Mechanical Power Engineering Department, Faculty of Engineering, Tanta University, Tanta, Egypt 6 Center for Quantum Transport and Thermal Energy Science, School of Physics and Technology, Nanjing Normal University, Nanjing, 210023, China #Guilong Peng and Swellam W. Sharshir contribute equally on this work. *Corresponding email: [email protected] 42 Method Random forest (RF) is a typical ensemble method, which combines multiple decision trees (DT) into one model to improve performance [1]. Given an initial dataset S, a random forest model consist of K decision trees can be established as follows: First, the bootstrap resampling method [2] is used to randomly generate K sets of data from the initial dataset S. Second, K decision trees will be grown based on K sets by specific random selection algorithm in each node [3]. Finally, the finally prediction is made by a weighted vote of all decision trees. In the present study, classification and regression trees (CART) [4] were used to construct the forest. For each tree in the forest, the Gini index defined in Eq (1) is used to evaluate the purity of a node in the tree. If all samples of a node belong to the same class, the Gini index equals to 0. (1) Where G(t) is the Gini index of node t, is the relative frequency of class i in the node t. After model construction, RF can calculate importance of certain descriptor. For a single tree, a descriptor's importance is defined as the sum of Gini index reduction over all nodes in which the certain descriptor is chose to split [5]. The final descriptor importance is averaged among all trees. Dataset 86 Experimental data used in current study were acquired by collecting from articles since 2014. Due to the lack of details in original articles, there are some missing data of Surface diameter, Absorptivity, Tamb and Tinterface. The amount of missing data is 16, 22, 33, 24, respectively. The method of Mean Completer was used for filling missing data. Each descriptor was divided into three labels. The details of data representation are listed in Table S1. 43 Table S1 Details of data representation Descriptors Classification Labels Number of samples 51 25 10 53 21 12 30 29 29 23 31 34 21 40 27 37 31 20 28 31 28 1 Kw Solar intensity 1-10 Kw >10 Kw 3D interface Thermal design 2D\1D interface Volumetric <3 cm Surface diameter 3-4 cm Absorptivity >4 cm <0.95 0.95 >0.95 <24℃ Tamb 24-25℃ >25℃ <50℃ Tinterface 50-70℃ >70℃ <75 % Efficiency 75 %~85 % >85% 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 0 1 2 44 Table S2 Dataset of machine learning Ref Categories I_solar Tamb/RH (kW/m2) (℃/ ) Water type Thermal design Absorptivity Tinterface/vapor E_water Ev_rate Ev_dark Calibrated with E.F. (℃) (kg/(m2·h) (kg/(m2·h) (kg/(m2·h) Ev_dark (%) 1.355 25 /50 salt water (3.5%) 3D interface 1 25 /50 salt water (3.5%) 3D interface 1 2.3 0.56 1.28 (%) -- -- 68 46 95 76 -- 33.4 100 -- -- -- >100 -- -- -- -- 57.8 80 100 80 39 10.9 -- -- -- -- 8.5 1.12 -- -- 15.74 0.86 -- 1.1 -- 15.95 1.5 4.9 -- -- -- -- -- -- -- -- -- 4.4 -- -- -- -- -- 0.36 0.78 0.39 -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0.114 1.13 -- -- Unknown -- Unknown Unknown -- -- -- -- Unknown Yes -- -- Unknown Unknown -- Unknown -- Yes Unknown -- 69 35.6 51 -- 64.1 46.8 70 69 84 59.6 53.6 60.3 66 44 57 80 Unknown 77.8 0.92 0.088 Yes 58 floating particles floating particles nanofluid nanofluid nanofluid nanofluid nanofluid 206 nanofluid nanofluid nanofluid 10 1 11 20.6 /20 nanofluid 280 -- nanofluid 16.7 nanofluid nanofluid nanofluid 1 10 1 25 /55 28 /49 25 /25 -- film film film film film 50.9 30.4 / -- 20 1 4.5 1 24 /14 27 / -- 22 /50 [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] 10 353 0.8 0.8 24 / -- -- 21.4 /41 20 /40 35 / -- -- water water water water water water volumetric 99.5 (350-1900nm) volumetric volumetric volumetric volumetric -- -- -- -- volumetric 90 (200-2500nm) 26 / -- salt water (3.5%) volumetric 95 (220-2000nm) water water water water water water water water water water water volumetric volumetric volumetric volumetric volumetric volumetric 3D interface -- -- -- -- -- -- -- 3D interface 91 (400-2500nm) 3D interface 97 (250-2000nm) 3D interface 87 (400-800nm) 3D interface -- 45 -- water Bottom Heating 95 (200-2000nm) 22 / -- salt water (3.5%) 3D interface film film film film film film film film film film film film film film film film film film film film film film film film film 1 1 6.7 1 1 4 4 1 10 5 1 1 1 4 1.95 10 1 1 15 4 2.94 1 1 6 5 [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] -- -- -- 24 /42 24 /48 -- -- -- 26 /20 -- -- 25 / -- -- -- 22 /60 23 /55 22 /36 25 /22 25 /50 -- 25 / -- -- -- water water water water water water water water water water water water water water water water water water water water water water water -- -- 3D interface 3D interface 98 (200-800nm) 3D interface -- 3D interface 99 (200nm-10μm) 3D interface 96 (400-1500nm) 3D interface 90 (400-1800nm) 3D interface 98 (300-1200nm) 108 3D interface 85 (200-2000nm) 3D interface 94 (300-1500nm) 3D interface -- 3D interface >97 (350-1450nm) 61 -- 40 35 3D interface 95 (300-1200nm) 45.4 3D interface -- 3D interface 90 (400-1500nm) 3D interface 94 (400-2500nm) 3D interface 90.2 (250-2500nm) 3D interface 99 (280-820nm) 1D interface 98 (400-2500) 3D interface -- 3D interface 67.4 (400-1000nm) 3D interface 90-96 (400-2500) 3D interface 97 (200-2500nm) 3D interface 95 (300-2500nm) 46 50 -- -- -- 100 73.5 98.1 31 38.5 60 50 -- -- 62 -- -- 65 68 80 -- 0.59 -- 0.4 -- 1.55 2.37 0.59 2.48 3 0.42 0.58 0.44 0.89 0.45 2.8 0.1 0.509 2.1 2.4 1.83 0.1 0.51 3.96 1.3 -- 1.01 -- 1.31 -- 5.6 5.7 1.13 11.22 4.95 1.55 1.01 1.18 4 2.14 11.8 0.47 1.24 21 6.01 3.81 1.42 1.43 8.24 6.6 -- -- -- 0.08 0.117 -- -- -- -- -- 0.1 -- -- -- -- -- 0.06 -- 0.16 0.24 -- -- 0.24 -- -- -- 60 Unknown 63.6 -- Unknown -- Yes Yes -- 82 77.1 90 88 Unknown 70.9 Unknown Unknown Yes 81 62 91 Unknown 62.7 Unknown Unknown Unknown Unknown Yes 76 71.8 72.5 85 48 Unknown 77.5 Yes Yes 90 94 Unknown 81.4 Yes Yes Unknown 83 90.4 83 Unknown 91.5 foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam 10 1.7 10 1 1 24 51 10 1 10 1 1 1 1 1 12 1 10 1 1 1 10 1 3.5 1 3D interface 97 (250-2250nm) 3D interface 94 (300-800nm) 3D interface 96 (400-1100nm) 2D Interface 94(250-2500nm) 3D interface 93 (200-2000nm) 3D interface 97 (250-2500nm) 3D interface -- 3D interface 95 (250-2500nm) 3D interface 92 (200-2500nm) 3D interface 92 (250-2500nm) 3D interface 98(450nm-750nm) 3D interface 2D interface -- -- 1D interface 99 (250-2500nm) 2D interface 97 ((250-1200nm)) 24 /31 -- -- -- 20 / -- -- -- 26 /52.5 25 /45 20 / -- 25 / -- 20.5 /47 -- 20 /25 20 /30 water water water water water water water water water water water water water water water 27 / -- salt water (3%) 3D interface -- 21 /10 -- 20 /60 water water water 2D interface 98 (250-2500nm) 1D interface 92 (300-2500nm) 3D interface -- 28 / -- salt water (2.75%) 3D interface 90-95(295-2000nm) 100 48 100 38.8 100 100 100 -- -- 92 43 42 -- 38.4 36.5 67 44.2 82 -- -- -- -- 25 / -- 21.2 / -- 28 /41 water water water water water 2D interface 90 (300-800nm) 43.9 3D interface 91 (300-1500nm) 2D interface 90 (200-2500nm) 3D interface 98 (200-800nm) 1D interface 96 (250-2500nm) -- 40 75.7 38 47 [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] 5 0.6 4.8 0.24 -- -- -- -- 0.5 2.9 0.46 -- 0.478 0.36 0.39 -- 0.43 -- 0.23 0.4 0.42 0.95 0.05 1.8 0.45 12 1.53 11.8 1.45 -- -- 51.8 14.36 1.622 12.1 1.13 -- 1.33 1.27 1.25 14.02 1.28 12.6 0.83 1.4 1.31 11.24 1.282 7.54 1.27 0.074 Yes -- -- 0.065 -- -- -- -- -- Unknown Unknown Yes -- -- Unknown Unknown Unknown 0.12 Yes -- -- -- 0.132 0.1 -- Unknown -- Unknown Yes Yes Unknown 0.125 Yes -- -- 0.1 0.1 -- 0.492 -- 0.2 Unknown Unknown Yes Yes Unknown Yes Yes Yes 85 60 83 80 20 -- 76.3 82.7 83 86.7 78 80 83.9 87.5 85.6 82.8 88 89 52.2 91.4 83 81 9 135 78 3D interface 98 (300-2500nm) Bottom Heating -- 3D interface 99 (250-2500nm) 3D interface 95.2 (250-2250nm) 1D interface 98 (250-2000nm) 1D interface 98 (400-2500nm) 1D interface 97 (400-2500nm) 43 -- -- 47 45 48.5 42 3D interface 97.5 (300-1200nm) 32.7 2D interface 94 (300-2500nm) water water water water water water water water water salt water (3.5%) 2D interface -- water water water water water water water water water water water water water water water 3D interface 95 (200-2500nm) 1D interface 95 (250-2500nm) 3D interface 96 (400-1200nm) 1D interface 90-95(250-2500nm) 3D interface 95 (300-2500nm) 3D interface 99(400-2500nm) 1D interface 97 (250-2500nm) 2D interface 95 (400-1100nm) 1D interface 99.4 (250-2500nm) 3D interface 98 (250-2000nm) 1D interface 99 (250-2500nm) 3D interface 100 (200-2500nm) 3D interface 96.5(250-2500nm） 1D interface 98 (400-2500nm) 3D interface 97 (250-2000nm) 48 48 42 38 32.7 43 42.2 103 51.5 50 48.5 37 34 46.5 46 34.3 43 38 foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam foam 1 1 10 1 1 1 1 1 1 1 1 1 1 1 10 5 1 1 1 1 1 1 1 1 1 26 /40 -- -- 24 /10 25 / -- 23 /40 21.5 /55 -- -- -- 22 / -- 30 / -- 25 /50 25 / -- 20 / -- -- 30 /60 24 /40 21.5 /55 25 / -- 22 /20 -- /45 25 /45 25 / -- -- [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] 0.45 0.33 3.3 0.43 0.48 -- -- 0.24 0.55 -- 0.3 -- 0.462 0.502 -- 1.27 -- -- -- 0.15 0.45 0.5 0.3 0.39 0.44 1.05 0.83 12.2 1.16 1.3 1.09 1.37 1.11 1.22 -- -- -- -- 1.58 12.26 6.6 1.45 1.13 1.45 1 1.45 3.2 2.15 1.79 2.6 -- -- 0.08 -- 0.13 0.286 0.23 0.05 0.2 -- -- 0.47 -- 0.224 -- -- 0.156 0.11 0.59 0.015 0.35 0.025 -- 0.19 -- Unknown Unknown Yes Yes Yes Yes Yes Yes Yes -- Unknown -- 72 -- 87 81 86.5 73 100 76.3 79.4 56 80 85 Unknown 87.3 Yes 84.95 Unknown Yes Yes Yes Yes Yes Yes Yes No Yes Unknown 89 86.2 91.3 78 100 65 91 94 -- 92 91 20 /60 salt water (20%) 3D interface 98 (250-2500nm) water 3D interface 94 (400-2500nm） [101] [102] [103] foam foam bulk water 1 1 1 -- -- salt water (3.5%) contactless 92 (250-1500nm) 133 -- 43 -- 0.463 -- 1.04 -- -- -- -- -- Yes 75 Unknown 90.4 -- 25 Note: "E_water" is the evaporation rate of pure water under solar irradiation, "Ev_rate" is the evaporation rate of using micro/nanomaterials under solar irradiation, "Ev_dark" is the evaporation rate of using micro/nanomaterials under dark condition. "Calibrated with Ev_dark" means whether the "Ev_dark" is subtracted from "Ev_rate". "E.F." is the efficiency of evaporation 49 References [1] [2] [3] [4] [5] J. Carrete, W. Li, N. Mingo, S. Wang, S. Curtarolo, Phys Rev X 2014, 4, 011019. A. C. Davison, D. V. Hinkley, Bootstrap methods and their application, Cambridge university press, 1997. L. Breiman, Mach Learn 2001, 45, 5. D. Steinberg, P. Colla, The top ten algorithms in data mining 2009, 9, 179. X. Chen, M. Wang, H. Zhang, Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 2011, 1, 55. [6] Y. Zeng, J. Yao, B. A. Horri, K. Wang, Y. Wu, D. Li, H. Wang, Energ Environ Sci 2011, 4, 4074. [7] [8] Y. Zeng, K. Wang, J. Yao, H. Wang, Chem Eng Sci 2014, 116, 704. G. Ni, N. Miljkovic, H. Ghasemi, X. Huang, S. V. Boriskina, C.-T. Lin, J. Wang, Y. Xu, M. M. Rahman, T. Zhang, G. Chen, Nano Energy 2015, 17, 290. [9] D. Zhao, H. Duan, S. Yu, Y. Zhang, J. He, X. Quan, P. Tao, W. Shang, J. Wu, C. Song, T. Deng, Sci Rep 2015, 5, 17276. [10] S. Ishii, R. P. Sugavaneshwar, K. Chen, T. D. Dao, T. Nagao, Opt Mater Express 2016, 6, 640. [11] S. Ishii, R. P. Sugavaneshwar, T. Nagao, The Journal of Physical Chemistry C 2016, 120, 2343. [12] [13] [14] H. Jin, G. Lin, L. Bai, A. Zeiny, D. Wen, Nano Energy 2016, 28, 397. X. Wang, Y. He, G. Cheng, L. Shi, X. Liu, J. Zhu, Energ Convers Manage 2016, 130, 176. X. Wang, G. Ou, N. Wang, H. Wu, ACS Appl Mater Interfaces 2016, 8, 9194. [15] M. S. Zielinski, J. W. Choi, T. La Grange, M. Modestino, S. M. Hashemi, Y. Pu, S. Birkhold, J. A. Hubbell, D. Psaltis, Nano Lett 2016, 16, 2159. [16] M. Amjad, G. Raza, Y. Xin, S. Pervaiz, J. Xu, X. Du, D. Wen, Appl Energ 2017, 206, 393. [17] Y. Fu, T. Mei, G. Wang, A. Guo, G. Dai, S. Wang, J. Wang, J. Li, X. Wang, Appl Therm Eng [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] 2017, 114, 961. H. Li, Y. He, Z. Liu, Y. Huang, B. Jiang, Appl Therm Eng 2017, 121, 617. L. Shi, Y. He, Y. Huang, B. Jiang, Energ Convers Manage 2017, 149, 401. X. Wang, Y. He, X. Liu, L. Shi, J. Zhu, Sol Energy 2017, 157, 35. Z. Wang, Y. Liu, P. Tao, Q. Shen, N. Yi, F. Zhang, Q. Liu, C. Song, D. Zhang, W. Shang, T. Deng, Small 2014, 10, 3234. K. Bae, G. Kang, S. K. Cho, W. Park, K. Kim, W. J. Padilla, Nat Commun 2015, 6, 10103. Y. Ito, Y. Tanabe, J. Han, T. Fujita, K. Tanigaki, M. Chen, Adv Mater 2015, 27, 4302. Y. Liu, S. Yu, R. Feng, A. Bernard, Y. Liu, Y. Zhang, H. Duan, W. Shang, P. Tao, C. Song, T. Deng, Adv Mater 2015, 27, 2768. L. Zhang, B. Tang, J. Wu, R. Li, P. Wang, Adv Mater 2015, 27, 4889. P. Fan, H. Wu, M. Zhong, H. Zhang, B. Bai, G. Jin, Nanoscale 2016, 8, 14617. Z. Hua, B. Li, L. Li, X. Yin, K. Chen, W. Wang, The Journal of Physical Chemistry C 2016, 121, 60. [28] Y. Liu, J. Lou, M. Ni, C. Song, J. Wu, N. P. Dasgupta, P. Tao, W. Shang, T. Deng, ACS Appl Mater Interfaces 2016, 8, 772. [29] Y. Wang, L. Zhang, P. Wang, Acs Sustain Chem Eng 2016, 4, 1223. 50 [30] [31] C. Zhang, C. Yan, Z. Xue, W. Yu, Y. Xie, T. Wang, Small 2016, 12, 5320. L. Zhou, Y. Tan, D. Ji, B. Zhu, P. Zhang, J. Xu, Q. Gan, Z. Yu, J. Zhu, Sci. Adv. 2016, 2, e1501227 [32] L. Zhou, Y. Tan, J. Wang, W. Xu, Y. Yuan, W. Cai, S. Zhu, J. Zhu, Nat Photonics 2016, 10, 393. [33] [34] G. Zhu, J. Xu, W. Zhao, F. Huang, ACS Appl Mater Interfaces 2016, 8, 31716. C. Chen, Y. Li, J. Song, Z. Yang, Y. Kuang, E. Hitz, C. Jia, A. Gong, F. Jiang, J. Y. Zhu, B. Yang, J. Xie, L. Hu, Adv Mater 2017, 29. [35] D. Ding, W. Huang, C. Song, M. Yan, C. Guo, S. Liu, Chem Commun (Camb) 2017, 53, 6744. [36] X. Gao, H. Ren, J. Zhou, R. Du, C. Yin, R. Liu, H. Peng, L. Tong, Z. Liu, J. Zhang, Chem Mater 2017, 29, 5777. [37] V. Kashyap, A. Al-Bayati, S. M. Sajadi, P. Irajizad, S. H. Wang, H. Ghasemi, J Mater Chem A 2017, 5, 15227. [38] C. Liu, J. Huang, C.-E. Hsiung, Y. Tian, J. Wang, Y. Han, A. Fratalocchi, Advanced Sustainable Systems 2017, 1, 1600013. G. Wang, Y. Fu, X. Ma, W. Pi, D. Liu, X. Wang, Carbon 2017, 114, 117. J. Wang, Z. Liu, X. Dong, C.-E. Hsiung, Y. Zhu, L. Liu, Y. Han, J Mater Chem A 2017, 5, 6860. X. Wang, Y. He, X. Liu, G. Cheng, J. Zhu, Appl Energ 2017, 195, 414. J. Yang, Y. Pang, W. Huang, S. K. Shaw, J. Schiffbauer, M. A. Pillers, X. Mu, S. Luo, T. Zhang, Y. Huang, G. Li, S. Ptasinska, M. Lieberman, T. Luo, Acs Nano 2017, 11, 5510. L. Yi, S. Ci, S. Luo, P. Shao, Y. Hou, Z. Wen, Nano Energy 2017, 41, 600. Z. Yin, H. Wang, M. Jian, Y. Li, K. Xia, M. Zhang, C. Wang, Q. Wang, M. Ma, Q. S. Zheng, Y. Zhang, ACS Appl Mater Interfaces 2017, 9, 28596. P. Zhang, J. Li, L. Lv, Y. Zhao, L. Qu, Acs Nano 2017, 11, 5087. T. F. Chala, C. M. Wu, M. H. Chou, Z. L. Guo, ACS Appl Mater Interfaces 2018, 10, 28955. [39] [40] [41] [42] [43] [44] [45] [46] [47] M. Chen, Y. Wu, W. Song, Y. Mo, X. Lin, Q. He, B. Guo, Nanoscale 2018, 10, 6186. [48] F. Tao, Y. Zhang, K. Yin, S. Cao, X. Chang, Y. Lei, D. S. Wang, R. Fan, L. Dong, Y. Yin, X. Chen, ACS Appl Mater Interfaces 2018, DOI: 10.1021/acsami.8b11786. [49] W. Xu, X. Hu, S. Zhuang, Y. Wang, X. Li, L. Zhou, S. Zhu, J. Zhu, Adv Energy Mater 2018, 8, 1702884. [50] X. Yang, Y. Yang, L. Fu, M. Zou, Z. Li, A. Cao, Q. Yuan, Adv Funct Mater 2018, 28, 1704505. [51] H. Ghasemi, G. Ni, A. M. Marconnet, J. Loomis, S. Yerci, N. Miljkovic, G. Chen, Nat Commun 2014, 5, 4449. [52] [53] F. M. Canbazoglu, B. Fan, A. Kargar, K. Vemuri, P. R. Bandaru, Aip Adv 2016, 6, 085218. Q. Jiang, L. Tian, K. K. Liu, S. Tadepalli, R. Raliya, P. Biswas, R. R. Naik, S. Singamaneni, Adv Mater 2016, 28, 9400. [54] X. Li, W. Xu, M. Tang, L. Zhou, B. Zhu, S. Zhu, J. Zhu, Proc Natl Acad Sci U S A 2016, 113, 13953. [55] G. Ni, G. Li, Svetlana V. Boriskina, H. Li, W. Yang, T. Zhang, G. Chen, Nature Energy 2016, 1, 16126. 51 [56] S. M. Sajadi, N. Farokhnia, P. Irajizad, M. Hasnain, H. Ghasemi, J Mater Chem A 2016, 4, 4700. [57] L. Tian, J. Luan, K. K. Liu, Q. Jiang, S. Tadepalli, M. K. Gupta, R. R. Naik, S. Singamaneni, Nano Lett 2016, 16, 609. [58] [59] [60] Y. Fu, G. Wang, T. Mei, J. Li, J. Wang, X. Wang, Acs Sustain Chem Eng 2017, 5, 4665. X. Hu, W. Xu, L. Zhou, Y. Tan, Y. Wang, S. Zhu, J. Zhu, Adv Mater 2017, 29. C. Jia, Y. Li, Z. Yang, G. Chen, Y. Yao, F. Jiang, Y. Kuang, G. Pastel, H. Xie, B. Yang, S. Das, L. Hu, Joule 2017, 1, 588. [61] Q. Jiang, H. Gholami Derami, D. Ghim, S. Cao, Y.-S. Jun, S. Singamaneni, J Mater Chem A 2017, 5, 18397. [62] M. Kaur, S. Ishii, S. L. Shinde, T. Nagao, Acs Sustain Chem Eng 2017, 5, 8523. [63] [64] R. Li, L. Zhang, L. Shi, P. Wang, Acs Nano 2017, 11, 3752. Y. Li, T. Gao, Z. Yang, C. Chen, Y. Kuang, J. Song, C. Jia, E. M. Hitz, B. Yang, L. Hu, Nano Energy 2017, 41, 201. [65] Y. Li, T. Gao, Z. Yang, C. Chen, W. Luo, J. Song, E. Hitz, C. Jia, Y. Zhou, B. Liu, B. Yang, L. Hu, Adv Mater 2017, 29. [66] K. K. Liu, Q. Jiang, S. Tadepalli, R. Raliya, P. Biswas, R. R. Naik, S. Singamaneni, ACS Appl Mater Interfaces 2017, 9, 7675. [67] Z. Liu, H. Song, D. Ji, C. Li, A. Cheney, Y. Liu, N. Zhang, X. Zeng, B. Chen, J. Gao, Y. Li, X. Liu, D. Aga, S. Jiang, Z. Yu, Q. Gan, Global Chall 2017, 1, 1600003. [68] Z. Liu, Z. Yang, X. Huang, C. Xuan, J. Xie, H. Fu, Q. Wu, J. Zhang, X. Zhou, Y. Liu, J Mater Chem A 2017, 5, 20044. [69] S. Ma, C. P. Chiu, Y. Zhu, C. Y. Tang, H. Long, W. Qarony, X. Zhao, X. Zhang, W. H. Lo, Y. H. Tsang, Appl Energ 2017, 206, 63. [70] H. Ren, M. Tang, B. Guan, K. Wang, J. Yang, F. Wang, M. Wang, J. Shan, Z. Chen, D. Wei, H. Peng, Z. Liu, Adv Mater 2017, 29, 1702590. L. Shi, Y. Wang, L. Zhang, P. Wang, J Mater Chem A 2017, 5, 16212. G. Wang, Y. Fu, A. Guo, T. Mei, J. Wang, J. Li, X. Wang, Chem Mater 2017, 29, 5629. Z. Wang, Q. Ye, X. Liang, J. Xu, C. Chang, C. Song, W. Shang, J. Wu, P. Tao, T. Deng, J [71] [72] [73] Mater Chem A 2017, 5, 16359. [74] X. Wu, G. Y. Chen, W. Zhang, X. Liu, H. Xu, Advanced Sustainable Systems 2017, 1, 1700046. [75] [76] N. Xu, X. Hu, W. Xu, X. Li, L. Zhou, S. Zhu, J. Zhu, Adv Mater 2017, 29. G. Xue, K. Liu, Q. Chen, P. Yang, J. Li, T. Ding, J. Duan, B. Qi, J. Zhou, ACS Appl Mater Interfaces 2017, 9, 15052. [77] J. D. Yao, Z. Q. Zheng, G. W. Yang, Nanoscale 2017, 9, 16396. [78] M. Zhu, Y. Li, G. Chen, F. Jiang, Z. Yang, X. Luo, Y. Wang, S. D. Lacey, J. Dai, C. Wang, C. Jia, [79] [80] [81] J. Wan, Y. Yao, A. Gong, B. Yang, Z. Yu, S. Das, L. Hu, Adv Mater 2017, 29. Q. Chen, Z. Pei, Y. Xu, Z. Li, Y. Yang, Y. Wei, Y. Ji, Chem Sci 2018, 9, 623. J. Fang, J. Liu, J. Gu, Q. Liu, W. Zhang, H. Su, D. Zhang, Chem Mater 2018, 30, 6217. X. Gao, H. Lan, S. Li, X. Lu, M. Zeng, X. Gao, Q. Wang, G. Zhou, J.-M. Liu, M. J. Naughton, K. Kempa, J. Gao, Global Challenges 2018, DOI: 10.1002/gch2.2018000351800035. [82] S. Hong, Y. Shi, R. Li, C. Zhang, Y. Jin, P. Wang, ACS Appl Mater Interfaces 2018, 10, 52 28517. [83] F. Jiang, H. Liu, Y. Li, Y. Kuang, X. Xu, C. Chen, H. Huang, C. Jia, X. Zhao, E. Hitz, Y. Zhou, R. Yang, L. Cui, L. Hu, ACS Appl Mater Interfaces 2018, 10, 1104. H. Li, Y. He, Y. Hu, X. Wang, ACS Appl Mater Interfaces 2018, 10, 9362. G. Ni, S. H. Zandavi, S. M. Javid, S. V. Boriskina, T. A. Cooper, G. Chen, Energ Environ Sci [84] [85] 2018, 11, 1510. [86] T. Li, H. Liu, X. Zhao, G. Chen, J. Dai, G. Pastel, C. Jia, C. Chen, E. Hitz, D. Siddhartha, R. Yang, L. Hu, Adv Funct Mater 2018, 28, 1707134. [87] [88] X. Li, R. Lin, G. Ni, N. Xu, X. Hu, B. Zhu, G. Lv, J. Li, S. Zhu, J. Zhu, Natl Sci Rev 2018, 5, 70. X. Lin, J. Chen, Z. Yuan, M. Yang, G. Chen, D. Yu, M. Zhang, W. Hong, X. Chen, J Mater Chem A 2018, 6, 4642. [89] F. Liu, B. Zhao, W. Wu, H. Yang, Y. Ning, Y. Lai, R. Bradley, Adv Funct Mater 2018, 28, 1803266. [90] H. Liu, C. Chen, G. Chen, Y. Kuang, X. Zhao, J. Song, C. Jia, X. Xu, E. Hitz, H. Xie, S. Wang, F. Jiang, T. Li, Y. Li, A. Gong, R. Yang, S. Das, L. Hu, Adv Energy Mater 2018, 8, 1701616. [91] H. Liu, C. Chen, H. Wen, R. Guo, N. A. Williams, B. Wang, F. Chen, L. Hu, J Mater Chem A 2018, 6, 18839. [92] P.-F. Liu, L. Miao, Z. Deng, J. Zhou, H. Su, L. Sun, S. Tanemura, W. Cao, F. Jiang, L.-D. Zhao, Materials Today Energy 2018, 8, 166. [93] G. Peng, H. Ding, S. W. Sharshir, X. Li, H. Liu, D. Ma, L. Wu, J. Zang, H. Liu, W. Yu, H. Xie, N. Yang, Appl Therm Eng 2018, 143, 1079. [94] Y. Shi, R. Li, Y. Jin, S. Zhuo, L. Shi, J. Chang, S. Hong, K.-C. Ng, P. Wang, Joule 2018, 2, 1171. [95] [96] C. H. Xiao Luo, Shang Liu, Jinxin Zhong, Int. J. Energy Res. 2018, 42, 4830. P. Zhang, Q. Liao, T. Zhang, H. Cheng, Y. Huang, C. Yang, C. Li, L. Jiang, L. Qu, Nano Energy 2018, 46, 415. [97] F. Zhao, X. Zhou, Y. Shi, X. Qian, M. Alexander, X. Zhao, S. Mendez, R. Yang, L. Qu, G. Yu, Nat Nanotechnol 2018, 13, 489. [98] Z. Deng, L. Miao, P.-F. Liu, J. Zhou, P. Wang, Y. Gu, X. Wang, H. Cai, L. Sun, S. Tanemura, Nano Energy 2019, 55, 368. [99] F. Gong, H. Li, W. Wang, J. Huang, D. Xia, J. Liao, M. Wu, D. V. Papavassiliou, Nano Energy 2019, 58, 322. [100] Y. Guo, F. Zhao, X. Zhou, Z. Chen, G. Yu, Nano Lett 2019, 19, 2530. [101] Y. Kuang, C. Chen, S. He, E. M. Hitz, Y. Wang, W. Gan, R. Mi, L. Hu, Adv Mater 2019, DOI: 10.1002/adma.201900498e1900498. [102] C. Li, D. Jiang, B. Huo, M. Ding, C. Huang, D. Jia, H. Li, C.-Y. Liu, J. Liu, Nano Energy 2019, 60, 841. [103] T. A. Cooper, S. H. Zandavi, G. W. Ni, Y. Tsurimaki, Y. Huang, S. V. Boriskina, G. Chen, Nat Commun 2018, 9, 5086. 53
1709.01794	1	1709	2017-08-22T21:50:43	A pico-Tesla magnetic sensor with PZT bimorph and permanent magnet proof mass	[ "physics.app-ph", "physics.ins-det" ]	Ferromagnetic-ferroelectric composites have attracted interests in recent years for use as magnetic field sensors. The sensing is based on magneto-electric (ME) coupling between the electric and magnetic subsystems and is mediated by mechanical strain. Such sensors for AC magnetic fields require a bias magnetic field to achieve pT-sensitivity. Here we discuss measurements and theory for a novel passive, AC magnetic sensor that does not require a bias magnetic field and is based on a PZT bimorph with a permanent magnet proof mass. Mechanical strain on the PZT bimorph in this case is produced by interaction between the applied AC magnetic field and remnant magnetization of the permanent magnet, resulting in an induced voltage across PZT. Our studies have been performed on sensors with a bimorph of oppositely poled PZT platelets and a NdFeB permanent magnet proof mass. Magnetic floor noise N on the order of 100 pT per sqrHz and 10 nT per sqrHz are measured at 1 Hz and 10 Hz, respectively. When the AC magnetic field is applied at the bending resonance of 40 Hz for the bimorph, the measured N 700 pT per sqrHz. We also discuss a theory for the magneto-electro-mechanical coupling at low frequency and bending resonance in the sensor and theoretical estimates of ME voltage coefficients are in very good agreement with the data.	physics.app-ph	physics	A pico-Tesla magnetic sensor with PZT bimorph and permanent magnet proof mass G. Srinivasan,1 G. Sreenivasulu,1 and Peng Qu2 and Hongwei Qu2 1Physics Department and 2E&CE Oakland University Rochester, MI 48309, USA Abstract-Ferromagnetic-ferroelectric composites have attracted interests in recent years for use as magnetic field sensors. The sensing is based on magneto-electric (ME) coupling between the electric and magnetic subsystems and is mediated by mechanical strain. Such sensors for ac magnetic fields require a bias magnetic field to achieve pT-sensitivity. Here we discuss measurements and theory for a novel ac magnetic sensor that does not require a bias magnetic field and is based on a PZT bimorph with a permanent magnet proof mass. Mechanical strain on the PZT bimorph in this case is produced by interaction between remnant magnetization of the permanent magnet, resulting in an induced voltage across PZT. Our studies have been performed on sensors with a bimorph of oppositely poled PZT platelets and a NdFeB permanent magnet proof mass. Magnetic floor noise N on the order of 100 pT/√Hz and 10 nT/√Hz are measured at 1 Hz and 10 Hz, respectively. When the ac magnetic field is applied at the bending resonance of ~ 40 Hz for the bimorph, the measured N ~ 700 pT/√Hz. We also discuss a theory for the magneto-electro- mechanical coupling at low frequency and bending resonance in the sensor and theoretical estimates of ME voltage coefficients are in very good agreement with the data. the applied ac magnetic field and Keywords-Ferroelectric; bimorph; permanent magnet; proof mass; piezoelectric; magnetoelectric; bending resonance I. INTRODUCTION A new generation of magnetic field sensors based on layered composites of ferromagnetic and piezoelectric phases has been reported in recent years [1-4]. These magneto-electric (ME) composites are capable of converting magnetic fields into electrical fields in a two-step process: magnetic field induced mechanical strain and stress induced electric field. An applied ac magnetic field H produces a magnetostrictive strain in the ferromagnetic layer, leading to an induced voltage V in the ferroelectric layer of thickness t. The ME voltage coefficient (MEVC) = V/(t H) and the ME sensitivity S=V/H are directly proportional the piezomagnetic and piezoelectric coefficients. Such ferroelectric-ferromagnetic composite sensors, however, require a bias magnetic field in order to enhance the piezomagnetic coupling coefficient in the composite to achieve pico-Tesla sensitivity [2-8]. The composite sensors have pT-sensitivity, passive, and are miniature in size and show superior performance and cost the product of to Research supported by a grant from the National Science Foundation (ECCS- 1307714). Vladimir Petrov Institute for Information Systems Novgorod State University Veliky Novgorod, Russia advantage over traditional flux-gate magnetometers or Hall effect sensors [3-6]. This report is on a novel pT-magnetic sensor based on ME coupling in a PZT bimorph with a permanent magnet proof mass and has the advantage of not requiring a bias magnetic field for operation. Mechanical strain on the PZT bimorph in this case is produced by interaction between the applied ac magnetic field and static remnant magnetization of the permanent magnet, resulting in an induced voltage across PZT [5]. Sensors with a bimorph of oppositely poled PZT platelets and NdFeB permanent magnet proof mass have been studied. A giant magneto-electric effect with MEVC of 20 V/cm Oe at low frequencies and enhancement to ~ 500 V/cm Oe at bending resonance have been measured for the sensor. The measured magnetic floor noise is on the order of 100 pT/√Hz to 10 nT/√Hz at 1-10 Hz. When the ac magnetic field is applied at the bending resonance for the bimorph the measured equivalent magnetic noise is ~ 700 pT/√Hz. A model is developed here and the strain and electric displacement of piezoelectric bimorph due to strain produced by interaction between H and remnant magnetization M. For finding low frequency and resonance ME voltage coefficients, we solve elastostatic and electrostatic equations in PZT, taking into account boundary conditions. In this case, the theoretical modeling in a ferromagnetic and ferroelectric bilayer [1]. The MEVC has been estimated as a function of frequency and is found to be in very good agreement with the data. is based on equations for to ME coupling is similar the II. EXPERIMENT The sensor fabricated and characterized in the present study is schematically shown in Fig.1 and consisted of a cantilever of two oppositely poled piezoelectric layers of length 50 mm, width 10 mm, and thickness 0.15 mm. We used commercial PZT (No.850, APC International) platelets and were poled by heating to 400 K and cooling to room temperature in a field of 30 kV/cm. The two oppositely poled PZT platelets were bonded to each other with a 2 m thick epoxy. The bimorph was clamped at one end and a magnet assembly of two NdFeB magnets was epoxy bonded to top and bottom of the bimorph at the free end. The magnets were circular discs of diameter 5 mm, 10 mm in height and a mass of 2.5 g each. The remnant magnetization M of NdFeB magnet assembly (along direction 3) was measured to be 15 kG. An ac magnetic field H generated by a pair of Helmholtz coils was applied parallel to the bimorph length (direction 1) so that interaction with M gives rise to a piezoelectric strain in PZT and a voltage V across the bimorph thickness. Since the PZT platelets are poled in opposite directions and the strain produced is compressive in one of them and tensile in the other the ME voltage in PZT layers (measured along direction 3) will be of opposite sign so that the overall ME response is enhanced with the use of a bimorph [5]. Results on MEVC vs f over 25-50 Hz are shown in Fig.3. One observes frequency independent MEVC in Fig.3 except for the frequency range 33-43 Hz over which a resonance enhancement is clearly evident. The MEVC increases sharply with increasing f from 30 V/cm Oe at 30 Hz to a peak value of 480 V/cm Oe at 38 Hz. With further increase in f, the MEVC decreases rapidly and levels of at ~ 10 V/cm Oe for f > 47 Hz. The peak in MEVC is associated with bending resonance in the the bimorph with proof mass [1,5]. Similar resonances in MEVC are reported for bending modes and longitudinal and thickness acoustic resonance in ferromagnetic-piezoelectric composites [1-6]. The ME coupling at resonance is due to the Fig.1. Diagram showing a cantilever of PZT-bimorph with NdFeB permanent magnet proof mass. Measurements of ME sensitivity and magnetic noise were carried out by placing the sample in a plexiglass holder in magnetically shielded -metal chamber surrounded by an acoustic shield. The sample clamped at one end was subjected to an ac magnetic field H produced by a pair of Helmholtz coils powered by a constant current source (Keithley, model 6221). The ME voltage generated across the thickness of the bimorph was measured with a signal analyzer (Stanford Research Systems, model SR780). Since the ME voltage across PZT is nonuniform along the length of the bimorph, we measured the ME voltage V close to the clamped end where one expects maximum value [8]. The ME sensitivity S = V/H and the ME voltage coefficient MEVC = S/t (t is the bimorph thickness) were measured as a function of frequency and at room temperature. Measurements of sensor noise were performed with the signal analyzer and was converted to equivalent magnetic noise. Fig.2.ME sensitivity S vs. frequency f profile showing the sensor response for H at 1 Hz. traditional strain mediated coupling, but the ac field is applied at the bending mode frequency so that the overall strain and the strength of ME coupling are enhanced. Data in Figs. 2 and 3 reveal an increase in MEVC at resonance by a factor of 24 compared to low frequencies values. Thus our PZT bimorph- magnet proof mass system show a giant ME effect both at low frequencies and at bending resonance. It is noteworthy here that the permanent magnet proof mass provides an avenue for control of the resonance frequency. The bending mode frequency is found to decrease with increasing proof mass. III. RESULTS AND DISCUSSION A. Magnetoelectric sensitivity The ME sensitivity S and MEVC were measured by measuring the voltage induced in the bimorph due to the applied ac filed H. Figure 2 shows representative results on S vs f for the specific case of H = 1 mOe at 1 Hz. The ME voltage at 1 Hz measured across the bimorph was V = 68 V, corresponding to S = 6800 V/T and MEVC = 23 V/cm Oe. Figure 2, in addition to voltage response at 1 Hz, also shows the noise spectra for frequencies up to 14 Hz and one notices a relatively large background noise over 5-7 Hz and 9-10 Hz. Measurements of MEVC as a function of frequency showed a constant MEVC over the frequency interval f = 1-25 Hz. Fig.3. MEVC vs. f data showing resonance enhancement in MEVC at the bending mode for the bimorph-proof mass system. Next we compare the results on MEVC in Fig.2 and 3 with reported values for similar systems and composites. Xing, et. al., investigated the ME coupling in PZT-bimorph loaded with permanent magnet tip mass and measured MEVC of 16 V/cmOe and 250 V/cm Oe at low frequency and bending resonance, respectively [5]. Thus the MEVC in our case are much higher than reported values in Ref.5. Past studies in the case of ferromagnetic-ferroelectric composites include ferrites, lanthanum manganites, 3-d transition metals and rare earths and alloys for the ferromagnetic phase and PZT, lead magnesium niobate-lead titanate (PMN-PT), quartz and AlN for ferroelectric/piezoelectric phase [1]. The ME sensitivity at 1 Hz in Fig. 2 is two orders of magnitude higher than reported values for bulk ferrite-piezoelectric composites and for bilayers and trilayers of ferrite-PZT and lanthanum manganite-PZT [1]. And it compares favorably with MEVC of 3 - 52 V/cm Oe at 1 kHz for Metglas composites with PZT fibers and single crystal PMN-PT [2]. The resonance MEVC in Fig.3 is higher than for ferrite based composites, but is smaller than the best value of ~ 1100 V/cm Oe reported for Metglas-PMN-PT [1, 2]. B. Sensor noise measurements We also measured the noise in the system for possible use as magnetic sensors. Data on equivalent magnetic noise floor were obtained over 0.5 Hz-50 Hz. The equivalent magnetic noise N in terms of T/√Hz was estimated from the measured noise (in V/√Hz) and the ME sensitivity S (in V/T) from data in Figs.2 and 3. Results on low-frequency N vs f are shown in Fig.4 for our samples with PZT bimorph and magnet tip mass. One notices a general increase in N from 100 pT/√Hz at 1 Hz to ~ 1 nT/√Hz at 10 Hz. Fig.4. Equivalent magnetic noise N as a function of frequency for the sensor. The data on noise N vs f over 30-60 Hz in Fig.5 shows a constant value of N=10 nT/√Hz away from bending resonance frequency and N decreases sharply to ~ 700 pT/√Hz close to the bending mode frequency. Minor peaks of unknown origin are seen above and below the resonance frequency. Now we compare the N-values for our sensor with reported values for The best N values ferromagnetic-piezoelectric sensors. reported to-date are for PZT or PMN-PT fibers and Metglas based sensors. Gao, et al., in their work on comparison of sensitivity and noise floor for ME sensors reported N ranging from 20 to 150 pT/√Hz (at 10 Hz), respectively, for Metglas with PZT fibers or single crystal PMN-PT [9]. Wang, et al., reported a further reduction in N to 5 pT/√Hz at 1 Hz for Metglas/PMN-PT fiber sensors [2]. Thus the magnetic noise for the sensor studied here compare favorably with reported values for multiferroic composite sensors. Fig.5. Results as in Fig.4, but for frequencies centered around the bending resonance in the sensor. The minimum in N occurs close to bending mode frequency for the cantilever sensor. C. Theory A model for the magneto-electric response of the bimorph with permanent magnet proof mass is considered next. The specific focus is on low-frequency and voltage versus frequency characteristics around the bending resonance frequency. A cantilever with PZT layers in (x,y) or (1,2) plane as in Fig.1 is assumed with one end clamped and the permanent magnet assembly on the free end. The thickness of PZT along z-direction (direction 3) is assumed to be small compared to its length or width. The interaction between the ac magnetic field along direction 1 and remnant magnetization of the magnet along direction 3 gives rise to a piezoelectric strain in PZT. Based on equation of bending vibrations [8,10], the general expression for displacement w in z direction perpendicular to the sample plane can written as w=C1 sinh(kx)+C2 cosh(kx)+C3 sin(kx)+C4 cos(kx), (1) with the wave number k is defined by the expression , (2) where ω is the angular frequency, t is the thickness of each PZT layer, ρ is the density, and D is cylindrical stiffness of the cantilever. The constants in Eq. 1 should be determined from boundary conditions that have the following form for the cantilever with attached permanent magnet at free end: Dtk224w=0 and ∂w/∂x =0 for x=0; Ay=∂w/∂x I ω2/b +μ0 JHv/b and Vy=- m w ω2/b at x=L (3) where m, v, I, and J are mass, volume, moment of inertia of magnet with respect to axis that is positioned in the middle plane along y axis, and remanent magnetization, respectively, Ay is the torque moment relative to y-axis produced by internal stresses in the bimorph per unite width, Vy is the transverse force per unite width, H is applied ac magnetic field, and L is the sample length and b is its width. The induced electric field in PZT can be found from the open circuit condition where 1,2D3 is electric induction in first and second piezoelectric layers, G is the z-axis, and cross-section of sample normal to the . Here d31 and ε33 are piezoelectric coupling coefficient and permittivity of PZT and 1,2E is internal electric field in layers. The stress components 1,2T can be expressed in terms of strain components 1,2S from elasticity equations where Y is the modulus of elasticity of piezoelectric component at constant E and (z1,2 is distance of current point of first or second layer from the middle plane). One obtains the following equation for the low frequency MEVC for the condition that the bimorph length is much higher than the width or thickness: , (4) influence of where d31 is the piezoelectric coupling coefficient, and 0 and 33 are the permeability and permittivity, respectively. The the magnet mass on resonance frequencies and MEVC is specified by the ratio of tip mass m to bilayer mass m0 and the dependence of m vanishes when the mass is much than bimorph mass. A significant decrease in the bending mode frequency is expected when the tip mass is of the order of bilayer mass. The fundamental bending mode frequency is given by The peak ME voltage coefficient at bending resonance frequency is estimated to be . (5) (6) with Q denoting the quality factor for bending resonance. We applied the model to estimate the ME voltage coefficient for the piezoelectric bimorph cantilever with attached magnet at free end. Resonance losses are taken into account by a using a complex frequency +i with /=1/Q, and Q was estimated from observed resonance profiles. The following material parameters were used the calculations: Y=0.65·1011 N/m2, density of PZT ρ=7.7·103 kg/m3, d31=- 1750·10-12 m/V, ε33/ε0=1750, m=5 g and 0J=1.5 T. for Fig.6. Theoretical MEVC as a function of frequency for the PZT bimorph with permanent magnet tip mass. Measured values are also shown for comparison. Fig.7. Calculated bending mode frequency as a function of the mass of permanent magnet. Theoretical estimates of MEVC vs. frequency are shown in Fig. 6. Measured values in Fig.3 are also shown for comparison. One observes a very good agreement between theoretical MEVC vs f profile and the data. Both the values of MEVC and the bending mode frequency are within 2% of the measured value. Calculated values of the bending mode frequency are plotted as a function of the mass of the permanent magnet in Fig. 7. One infers the following from 032,1dxDG32,13312,13132,1ETdD)(32,13112,112,1EdSYT222,112,1xwzS33231043btJdvMEVC312302mmYLtrf144015902330310mmbtmmJdQvMEVCrthe results in Fig.7: (i) The cantilever arrangement facilitates electromechanical resonance at low-frequencies compared to longitudinal or thickness acoustic modes; (ii) it is possible to control the resonance frequency with proper choice for the mass of the permanent magnet; and (iii) assuming a linear increase in M with the magnet mass, any decrease in the resonance frequency with increasing m will be accompanied by an increase in the peak MEVC. IV. CONCLUSION A sensor of ac magnetic fields consisting of a PZT bimorph with a permanent magnet for proof mass has been designed and characterized. The sensor operation is based on magneto- electric interaction mediated by mechanical strain. The applied ac magnetic field interacts with the remnant magnetization of the permanent magnet resulting in a strain that gives rise to a voltage response from the PZT bimorph. Magneto-electric characterization of the sensor clamped at one end indicate a giant ME effect both at low-frequencies and at bending resonance and the MEVC are comparable to best values reported for ferromagnetic and ferroelectric composites. The equivalent magnetic noise range from 100 pT/√Hz to 10 nT/√Hz , depending on the frequency. A model for the sensor has been developed and estimates of low frequency and resonance MEVC are in very good agreement with the data. The key advantages of the sensor are (i) the elimination of the need for a dc magnetic bias field that is required for high sensitivity ferromagnetic-ferroeleectric magnetic sensors and (ii) potential for control of the sensitivity by controlling the MEVC and bending resonance frequency with proper choice of proof mass. It is possible to decrease the resonance frequency and increase MEVC by increasing the proof mass so that high sensitivity could be achieved by operating the sensor under frequency modulation [11]. The research was supported by a grant from the National Science Foundation (ECCS-1307714). REFERENCES [1] G. Srinivasan, S. Priya, and N. X. Sun, Composite Magnetoelectrics: Materials, Structures, and Applications, Elsevir: Woodhead Publishing, 2015. [2] Y. Wang, D. Gray, D. Berry, J. Gao, M. Li, J. Li, and D. Viehland, "An extremely low equivalent magnetic noise magnetic sensor," Advanced Materials, vol. 23, pp. 4111-4114, September 2011. [3] R. Jahns, H. Greve, E. Woltermann, E. Quandt, and R. Knochel, "Sensitivity enhancement of magnetoelectric sensors through frequency conversion," Sensor and Actuators, vol. 183, pp. 16-21, August 2012. [4] P. Zhao, Z. Zhao, D. Hunter, R. Suchoski, C. Gao, S. Mathews, M. Wuttig, and I. Takeuchi, "Fabrication and characterization of all-thin- film magnetoelectric sensors," Appl. Phys. Lett, vol. 94, pp. 243507, June 2009. [5] Z. Xing, J. Li, and D. Viehland, "Giant magnetoelectric effect in Pb(Zr,Ti)O3-bimorph/NdFeB laminate device," Appl. Phys. Lett. Vol. 93, pp. 013505, July 2008. [6] X. Zhuang, S. Saez, M. Lam Chok Sing, C. Cordier, C. Dolabdjian, J. Li, D. Viehland, S. K. Mandal, G. Sreenivasulu, and G. Srinivasan, "Investigation on the magnetic noise of stacked magnetostrictive- piezoelectric laminated composites," Sensors Lett. Vol 10, pp. 961-965, March/April 2012. [7] Y. K. Fetisov, A. A. Bush, K. E. Kamentsev, A. Y. Ostashchenko, and G. Srinivasan, "Ferrite-piezoelectric multilayers for magnetic field sensors," IEEE Sensors J., vol. 6, pp. 935-938, August 2006. [8] G. Sreenivasulu, P. Qu, V. M. Petrov, H. Qu, and G. Srinivasan, "Magneto-electric interactions at bending resonances in an asymmetric multiferroic composite: Theory and experiment on the influence of electrode position," J. Appl. Phys. Vol. 117, pp. 174105, May 2015. [9] J. Gao, J. Das, Z. Xing, J. Li, and D. Viehland, "Comparison of noise floor and sensitivity for different magnetoelectric laminates," J. Appl. Phys. Vol. 108, pp. 084509, Oct. 2010. [10] S. P. Timoshenko and D. H. Young, Vibration problems in engineering, 3rd ed., Van Nostrand Co., NY, 1955. [11] S. M. Gillette, A. L. Geiler, D. Gray, D. Viehland, C. Vittoria, and V. G. Harris, IEEE magn. Lett. Vol.2, pp. 2500104, June, 2011. ACKNOWLEDGMENT
1904.01397	1	1904	2019-01-10T20:41:16	Plasmonic antennas with electric, magnetic, and electromagnetic hot spots based on Babinet's principle	[ "physics.app-ph" ]	We theoretically study plasmonic antennas featuring areas of extremely concentrated electric or magnetic field, known as hot spots. We combine two types of electric-magnetic complementarity to increase the degree of freedom for the design of the antennas: bow-tie and diabolo duality and Babinet's principle. We evaluate the figures of merit for different plasmon-enhanced optical spectroscopy methods: field enhancement, decay rate enhancement, and quality factor of the plasmon resonances. The role of Babinet's principle in interchanging electric and magnetic field hot spots and its consequences for practical antenna design are discussed. In particular, diabolo antennas exhibit slightly better performance than bow-ties in terms of larger field enhancement and larger Q factor. For specific resonance frequency, diabolo antennas are considerably smaller than bow-ties which makes them favourable for the integration into more complex devices but also makes their fabrication more demanding in terms of spatial resolution. Finally, we propose Babinet-type dimer antenna featuring electromagnetic hot spot with both the electric and magnetic field components treated on equal footing.	physics.app-ph	physics	Plasmonic antennas with electric, magnetic, and electromagnetic hot spots based on Babinet's principle M. Hrto n1, A. Konecn´a2, M. Hor´ak1, T. Sikola1,3, and V. Kr´apek1,3,* 1Central European Institute of Technology, Brno University of Technology, Purkynova 123, 612 00 Brno, Czech Republic 2Materials Physics Center CSIC-UPV/EHU, Paseo Manuel de Lardizabal 5, 20018 San Sebasti´an, Spain 3Institute of Physical Engineering, Brno University of Technology, Technick´a 2, 616 69 Brno, Czech Republic *[email protected] ABSTRACT We theoretically study plasmonic antennas featuring areas of extremely concentrated electric or magnetic field, known as hot spots. We combine two types of electric-magnetic complementarity to increase the degree of freedom for the design of the antennas: bow-tie and diabolo duality and Babinet's principle. We evaluate the figures of merit for different plasmon-enhanced optical spectroscopy methods: field enhancement, decay rate enhancement, and quality factor of the plasmon resonances. The role of Babinet's principle in interchanging electric and magnetic field hot spots and its consequences for practical antenna design are discussed. In particular, diabolo antennas exhibit slightly better performance than bow-ties in terms of larger field enhancement and larger Q factor. For specific resonance frequency, diabolo antennas are considerably smaller than bow-ties which makes them favourable for the integration into more complex devices but also makes their fabrication more demanding in terms of spatial resolution. Finally, we propose Babinet-type dimer antenna featuring electromagnetic hot spot with both the electric and magnetic field components treated on equal footing. Introduction Plasmonic antennas are metallic particles widely studied for their ability to control, enhance, and concentrate electromagnetic field1. Strikingly, the field in the vicinity of plasmonic antennas, so-called near field, can be focused to deeply subwavelength region. At the same time, the field is strongly enhanced (in comparison to the driving field, which can be e.g. plane wave). The mechanism behind the focusing of the field are localised surface plasmons (LSP) -- quantized oscillations of the free electron gas in the metal coupled to the evanescent electromagnetic wave propagating along the boundary of the metal. In judiciously designed plasmonic antennas, local spots of particularly enhanced electric or magnetic field can be formed, referred to as hot spots. They typically arise from the interaction between adjacent parts a plasmonic antenna separated by a small gap2, 3 but they can be also based on the lightning rod effect (a concentration of the field at sharp features of the antenna)4 -- 6 or combination of both. In various studies, electric hot spots have been reported over a broad spectral range from THz5 (hot spot size λ /60000 predicted from electromagnetic simulations) to visible7 (hot spot size λ /600 and enhancement > 500). Depending on the enhanced field, hot spots can be classified as electric, magnetic, or electromagnetic. A variety of plasmonic antennas with specific shapes, sizes, and materials exists for both electric and magnetic hot spots. Electric hot spots have been observed in the nanorod dimer antennas, bow-tie antennas8, or chains of plasmonic nanoparticles2, 9. Magnetic hot spots are formed in diabolo antennas10, nanorings11, or split-ring resonators12. Electromagnetic hot spots with simultaneous enhancement of both electric and magnetic field are unique for plasmonic antennas13. Their formation has been observed in the dielectric resonators (silicon nanodimers)14. Hot spots can be involved in many application including surface enhanced Raman scattering2, 15, 16, improved photocataly- sis17, or fluorescence of individual molecules18. Metallic resonators with enhanced magnetic field (magnetic hot spots) are regularly used to increase the efficiency of magnetic spectroscopies such as electron paramagnetic resonance19. Electromagnetic hot spots can be useful for studies of materials with combined electric and magnetic transitions such as rare earth ions20, 21. combined enhancement of electric and magnetic field finds applications also in optical trapping22, metamaterials23, or non-linear optics24. For experimental characterization of plasmonic hot spots, the available methods is scanning near-field optical mi- croscopy14, 25, 26, photon scanning tunneling microscope3, or photothermal-induced resonance27. Bow-tie geometry of plasmonic antennas features particularly strong electric hot spot. Bow-tie antennas are planar antennas consisting of two metallic triangular prisms (wings) whose adjacent apexes are separated by a subwavelength insulating gap. The hot spot arises from the interaction between the apexes combined with the lightning rod effect (the charge of LSP accumulates at the apexes). When the insulating gap is replaced with a conductive bridge, a diabolo plasmonic antenna is formed. Instead of charge accumulation, electric current is funneled through the bridge, resulting into magnetic hot spot. Both the bow-tie and diabolo antennas have been frequently studied. Various optimization and modification approaches have been proposed with the aim to enhance the properties of the bow-tie and diabolo antennas, including the gap optimization28, fractal geometry29, or Babinet's principle. Babinet's principle relates the optical response of a (direct) planar antenna and a complementary planar antenna with interchanged conductive and insulating parts. Both the direct and complementary antennas shall support LSP with identical energies, but with interchanged electric and magnetic near field30, 31. Consequently, when the direct antenna features electric hot spot, the complementary antenna features magnetic hot spot and vice versa. The validity of Babinet's principle for the plasmonic antennas has been experimentally verified32, 33, although some quantitative limitations have been found in particular in the visible34, 35. A unique combination of Babinet's complementarity and bow-tie/diabolo duality extends a degree of freedom for the design of plasmonic antennas featuring hot spots. In our contribution we compare the two antennas featuring electric hot spot (bow-tie and complementary diabolo) and the other two featuring magnetic hot spot (diabolo and complementary bow-tie). By electromagnetic modeling we retrieve the characteristics of the hot spots and figures of merit of relevant plasmon-enhanced optical spectroscopy methods. Finally, we design Babinet dimer antennas featuring electromagnetic hot spots, rather unique in the field of plasmonics. Results and discussion Figure 1. Schemes of four plasmonic antennas featuring hot spots: (a) Bow-tie, (b) diabolo, (c) inverted bow-tie, (d) inverted diabolo. Metallic and hollow parts are represented by yellow and white color, respectively. Driving electric field is indicated by red arrows. Charge or current accumulation and formation of the electric or magnetic hot spot are shown as well. Dotted white line in (c) indicates the qualitative correspondence between the diabolo and inverted bow-tie antennas, while dotted white line in (d) indicates a similar correspondence between the bow-tie and inverted diabolo antennas. Dimensions of the antennas are shown in (e). 2/13 Plasmonic antennas, modes and hot spots Plasmonic antennas involved in the study and their operational principle are illustrated in Fig. 1. Bow-tie antenna consists of two disjoint triangular gold prisms. Oscillating electric field applied along the long axis of the antenna drives the oscillations of charge that is funneled by the wings of the antenna and accumulated at the adjacent tips [Fig. 1(a)]. Combined effects of plasmonic field confinement, charge funneling, and charge concentration (lightning rod effect) give rise to an exceptionally high field in the area between the triangles, by orders of magnitude higher than the driving field. In diabolo antenna, the triangles are connected with a conductive bridge, through which a concentrated current flows instead of charge accumulation [Fig. 1(b)]. A magnetic hot spot is formed around the bridge. Inverted bow-tie antenna is formed by two disjoint triangular apertures in otherwise continuous gold film. Babinet's principle predicts that for a complementary illumination (i.e., transverse oscillating electric field) a complementary magnetic hot spot is formed. This can be understood also intuitively as the antenna resembles a rotated diabolo antenna [see dotted line in Fig. 1(c)]. Finally, inverted diabolo antenna, which on the other hand resembles the bow-tie antenna, features electric hot spot Fig. 1(d). The dimensions of the antennas are schematically depicted in Fig. 1(d). The thickness of the gold film is set to H = 30 nm. The size of the right isosceles triangles (i.e. ϑ = 90◦) is described by the wing length v. Opposite triangles share a common apex. The isolating gap in bow-tie antennas and the conductive bridge in diabolo antennas have the length G equal to the width W . These dimensions do not scale with the size of the antenna (the only scalable parameter is thus v) and are set to 30 nm to reflect common fabrication limits. In general, one could expect stronger hot spots for narrower gaps or bridges due to stronger charge or current concentration. All edges are rounded with a radius of 10 nm. The antennas are situated on a semi-infinite glass substrate (refractive index 1.47). The dielectric function of gold is taken from Johnson and Christy.36. Figure 2. Spectral dependence of the scattering efficiency Qscat of (a) bow-tie, (b) diabolo, (c) inverted bow-tie, (d) inverted diabolo PAs for several values of the wing length v of the antennas. One of the quantities characterizing plasmonic response of antennas is their scattering efficiency Qscat. It describes the power Pscat scattered by the antenna illuminated with a monochromatic plane wave with an intensity I0 and is defined as Qscat = Pscat/(I0 S), where S denotes the geometrical cross-section of the antenna. Spectral dependencies of Qscat for all four PA types are shown in Fig. 2 and the energies of the lowest scattering peak corresponding to a dipole plasmonic mode are shown in Fig. 3. We observe that Babinet's principle holds reasonably well. Peak energies of the scattering efficiency in the complementary PA (i.e., bow-tie and inverted bow-tie, diabolo and inverted diabolo) of the same size differ by less than 11 %. The difference is less pronounced for large antennas, in line with the requirements of Babinet's principle: perfectly thin and 3/13 Figure 3. (a) Peak energies of the scattering efficiency Qscat of bow-tie (red full circles), diabolo (blue full squares), inverted bow-tie (blue empty circles), and inverted diabolo (red empty squares) PAs as functions of the PA dimension -- length of the wing v. (b) Quality factors of LSP (represented by the peak energy of scattering cross-sections divided by their full width at half maximum). Notice that circles and squares correspond to bow-tie and diabolo PAs, full and empty symbols correspond to particles (direct PAs) and apertures (inverted PAs), and red and blue color corresponds to electric and magnetic hot spots, respectively. opaque metal.35 For the bow-tie geometry, the scattering peaks of inverted PAs are less intense and red-shifted with respect to direct PAs (as is the case also for disc-shaped antennas35), while opposite is true for the diabolo geometry. Not surprisingly, the peak energies of the scattering cross-section for the diabolo PAs are considerably smaller than that for the bow-tie PAs of the same wing length v. In other words, for the same energy, the diabolo PAs are smaller by a factor of more than 2 than the bow-tie PAs. This effect is explained by larger effective size of connected (i.e., diabolo) antennas in comparison with disjoint ones (bow-tie) and has been observed previously37. There is an important practical consequence. Bow-tie geometry allows to achieve high energies for which diabolo-type PAs can be too small for involved fabrication technique. Considering the minimum wing length of 50 nm, diabolo antennas cover the LSP energy range up to 1.2 eV while bow-tie antennas operate up to 2.0 eV. On the other hand, diabolo geometry allows for a more compact PA design and better integration to more complex devices, such as a scanning near-field probe with the electric hot spot.38 Diabolo antennas, either direct or inverted, feature considerably narrower scattering peaks corresponding to larger quality factor than bow-tie antennas (Fig. 3). This is probably related to lower radiative losses due to their smaller volume. bow-tie diabolo 0.8 eV 300 nm 95 nm 1.8 eV 75 nm - inverted bow-tie inverted diabolo 270 nm 55 nm 100 nm - Table 1. Dimensions (wing length) of the antennas featuring the lowest LSPR at the energy of 0.8 eV and 1.8 eV. In the following, we compare the properties and performance of all four types of PAs. We adjust the dimensions of the compared PAs so that they all feature the LSPR at the same energy. Table 1 shows the dimensions of the antennas are listed in Table 1 for two specific energies: 1.8 eV corresponding to the minimum absorption of gold (i.e., minimum of the imaginary part of dielectric function) and 0.8 eV corresponding to one of the optical communication wavelengths (1550 nm). We note that the former energy is accessible only with bow-tie antennas. We have therefore focused at the energy of 0.8 eV. Figure 4 shows the formation of the hot spot. PAs featuring the lowest LSPR at the energy of 0.8 eV are illuminated by a plane wave with the same photon energy. Bow-tie and inverted diabolo PAs feature the electric hot spot and delocalized magnetic field, while diabolo and inverted bow-tie PAs feature the magnetic hot spot and delocalized electric field. Interestingly, the volume of all the hot spots is comparable despite pronounced differences in the dimensions of PAs. The fields exhibit clear Babinet's duality: their spatial distribution in direct and complementary antennas is qualitatively similar with interchanged electric and magnetic components. Nevertheless, the magnitudes of the complementary fields differ rather significantly. As an 4/13 Figure 4. Planar cross-sections of the electric ((cid:126)E) and magnetic ((cid:126)H) field magnitudes divided by related magnitudes of the driving plane wave. The top two sub-plots in each panel show field distributions in the plane parallel to the PA plane, 10 nm above the upper PA boundary. The bottom two sub-plots then show field distributions in the vertical plane with the orientation indicated by white dotted lines in the in-plane field plots. The size of all PAs was set so that they feature the lowest LSPR at 0.8 eV, which is also the photon energy of the driving field. Solid white lines indicate antenna boundaries, while hot spots are marked by the green point and the numbers correspond to the field enhancement in the hotspot. The figures show only the central part of the antennas with the metallic parts being denoted as Au for clarity. example, the electric field within the hot spot of the bow-tie antenna has the relative magnitude around 25 while the magnetic field of the inverted bow-tie has maximal relative magnitude less than 10, i.e., almost three times weaker than expected. This observation is attributed to the finite thickness and conductivity of gold, which both limit the validity of Babinet's principle. As for the direct and inverted diabolo, the difference in the magnitudes of the electric and magnetic fields is less pronounced, but still quite significant. The bow-tie/diabolo duality can be observed for the field forming the hot spot (e.g., electric for bow-tie and magnetic for diabolo) which has very similar spatial distribution in both cases. However, the distribution of the delocalized field (e.g., electric for bow-tie and magnetic for diabolo) differs. In general, magnetic fields are weaker than electric fields since the energy of the electric field is only partly converted to the energy of magnetic field and the other part converts into kinetic energy of electrons35, 39. Figures of merit for optical spectroscopy Plasmonic antennas can be used to enhance absorption and emission of light. Consequently, they enhance the signal of interest inthe signal of interest in various optical spectroscopy techniques, including absorption spectroscopy, Raman spectroscopy, photoluminescence spectroscopy, and absorption spectroscopy of magnetic transitions. Here we define figures of merit (FoM) for plasmonic enhancement of different spectroscopy techniques and evaluate them for all four type of PAs. We will consider a 5/13 small volume of the analyzed material (e.g. molecule, quantum dot, nanostructured material, or just nanosized crystal) that fits into the size of the hot spot. In case of absorption spectroscopy, absorbed power can be expressed by Fermi's golden rule as P = 0.5ℜ[σ (ω)]E2 where ω is the frequency of the probing radiation (in the following referred to as light), σ is the complex conductivity of the analyte, and E is the magnitude of the electric component of light. For simplicity we consider both the electric field and transition dipole moment to be polarized along the axis of the PA. The presence of plasmonic antennas alters the magnitude of electric field exciting the analyte. For the driving field (a plane wave) with the electric field intensity E0, the electric intensity in the hot spot reads EHS. We define the electric field enhancement ZE = EHS/E0. Clearly, absorbed power is enhanced by the factor of Z2 E, which is thus suitable FoM for plasmon enhanced absorption spectroscopy. Raman scattering is a two-photon process, where each of the subprocesses, i.e., absorption of the driving photon and re-emission of the inelastically scattered photons, is enhanced by Z2 E (spectral dependence of ZE can be neglected considering low relative energy shift in the Raman scattering and large energy width of plasmon resonances). Therefore, FoM for the plasmon enhanced Raman spectroscopy reads Z4 E. Absorption spectroscopy of magnetic transitions is relevant for the study of rare earth ions in the visible20, 21. Electron paramagnetic resonance is in principle also absorption spectroscopy involving magnetic dipole transitions in the microwave spectral range. Absorbed power can be expressed as P = 0.5ωℑ[µ(ω)]H2 where ω is the frequency of light, µ is the complex permeability of the analyte, and H is the magnitude of the magnetic component of light. For the magnetic field enhancement ZH defined analogously to ZE, the FoM for absorption spectroscopy of magnetic transition reads Z2 H. Figure 5. Electric (red) and magnetic (blue) field enhancements for bow-tie, diabolo and their Babinet complements in their respective hotspots. The electric field enhancement ZE, defined as the ratio of the local electric field intensity (cid:126)EHS and the amplitude of the driving field E0, enters the figures of merit for plasmon-enhanced absorption spectroscopy (∼ Z2 E) and Raman scattering (∼ Z4 intensity (cid:126)EHS and the magnetic field intensity H0 of the driving field, is important for plasmon-enhanced absorption spectroscopy of magnetic transitions (∼ Z2 H). E). On the other hand, the magnetic field enhancement ZH, defined as the ratio of the local magnetic field We should note that the choice of the location in which we should evaluate the enhancement factors is somewhat arbitrary. In the case of the bow-tie and inverted diabolo, we decided to take the values from a spot positioned in the center of the gap, 10 nm above the substrate, while for the inverted bow-tie and diabolo, the spot is situated 10 nm above the center of the bridge. This choice gives us reasonable estimates that are close to the average values over the whole hot spots and it also ensures sufficient separation from the metal which is relevant for dipolar emitters and their quenching. Ultimately, we can afford this slight inconsistency in the definition of the hot spot as we always compare bow-tie with inverted diabolo and inverted bow-tie with diabolo, i.e. PAs with the same definition of the hotspot. With this in mind, we can turn our attention back to the field enhancements ZE and ZH. The inspection of Figure 5 shows that both ZE and ZH decrease with increasing energy as a consequence of decreased funneling effect (the size of the wings decreases while the size of the bridge or the gap is kept constant). The electric field enhancement ZE ranges between 18 and 34 with the inverted diabolo PA having slightly better performance than bow-tie. The magnetic field enhancement ranges between 10 and 17 for the diabolo PA but only between 4 and 8 for inverted bow-tie. Thus, inverted diabolo presents an excellent option for electric field enhancement while inverted 6/13 bow-tie does not perform particularly well for the magnetic field enhancement. Luminescence spectroscopy is another important method that can benefit from plasmon enhancement. We will consider a simple model based on the rate equation model. A metastable excitonic state with the degeneracy g is populated through external excitation with the rate γG. The generation is only efficient when the metastable state is unoccupied. For its population n, the total generation rate reads (g−n)·γG. Excitons decay into the vacuum state via radiative and non-radiative recombination paths with the rates γR0 and γNR0, respectively. The rate equation reads dn dt = (g− n)· γG − n· γR0 − (g− n)· γNR0. In steady state, dn/dt = 0 and n = g· γG/(γG + γR0 + γNR0). Two regimes can be distinguished. In the linear (weak pumping) regime, γG (cid:28) γR0 + γNR0 and i.e., population is proportional to pumping. In the saturation (strong pumping) regime, γG (cid:29) γR0 + γNR0) and n ≈ g, i.e., metastable state is fully occupied. Emitted power reads n ≈ gγG/(γR0 + γNR0), PPL = n· γR0 · ¯hω, where ¯hω is the photon energy. In the linear regime, the emitted power can be expressed using internal quantum efficiency η0 = γR0/(γR0 + γNR0) as Pl = gγGη0 · ¯hω and in the saturation regime Ps = gγR0 · ¯hω. The presence of plasmonic antennas affects all three processes (generation, radiative decay, and non-radiative decay). The effect on generation varies from very important in the case of photoluminescence18 to negligible in the case of electrolumi- nescence. In general, generation is sequential inelastic process and cannot be described by a simple model. For that, we will not consider plasmon enhancement of generation in the following and focus on its influence of the radiative and non-radiative decay rates. Spontaneous emission is affected via Purcell effect40. The emitter transfers its energy to PA where it is partially radiated into far field and partially dissipated. It is customary to express the rates of both processes in multiples of the spontaneous emission rate γR0: ZR being the radiative enhancement and ZNR the non-radiative enhancement41, 42. Total radiative and non-radiative decay rates in the presence of plasmonic particles read γR = ZR · γR0 and γNR = ZNR · γR0 + γNR0, respectively. The figure of merit for plasmon enhanced luminescence (only its emission part) is the rate of the powers emitted with and without the presence of the PA. For linear regime, FoM is while for the saturation regime it reads simply Fl = η/η0 Fs = ZR. Consequently, only emitters with poor internal quantum efficiency can benefit from plasmon enhancement in the linear regime while the emitters with high internal quantum efficiency will suffer from the dissipation in metallic PA. On the other hand, in the saturation regime plasmon enhancement is benefitable as long as ZR > 1. Figure 6 shows spectral dependence of radiative and non-radiative enhancement factors (ZR and ZNR, respectively) for different types of PA with the maximum field enhancement at 0.8 eV. The point-like isotropic emitter (i.e., all polarizations are involved with the same intensity) was positioned to the centre of the PA 10 m above the substrate (bow-tie and inverted diabolo) or 10 nm above the surface of gold (diabolo and inverted bow-tie). Such a separation shall suppress emission quenching due to non-radiative decay of the emitter. For the electric dipole transitions, large radiative enhancement (several hundreds) is obtained for both bow-tie and inverted diabolo. The inverted diabolo offers approximately twice larger peak enhancement. However, bow-tie benefits from much lower non-radiative enhancement and is thus preferable for most emitters in the linear regime. For magnetic dipole transitions, diabolo provides considerably larger radiative enhancement than inverted bow-tie, but it also suffers from the considerably larger non-radiative enhancement. In addition, the resonance of the inverted bow-tie is considerably wider which can prioritize this type of antenna for emitters with broad spectral bands. The preferred PA type therefore depends on specific application. We note that the peaks in the enhancement are spectrally shifted from the maximum field enhancement; the effect is particularly pronounced for the inverted bow-tie. 7/13 Figure 6. Radiative and non-radiative enhancement factors ZR and ZNR, respectively, as functions of the photon energy for the (a) electric dipole transition and (b) magnetic dipole transition. The values are averaged over all possible polarizations of the transitions. Panels (c) and (d) then show enhancement of the overall quantum efficiency for two values of the internal quantum efficiency, namely η0 = 0.9 (good emitter) and η0 = 0.05 (poor emitter). Note that even though the radiative enhancement for direct and inverted diabolos is significantly higher than for their bow-tie counterparts, their enhancement of quantum effieciency is due to their equally larger non-radiative enhancement more or less the same. Babinet dimer with electromagnetic hot spot Bow-tie and diabolo PAs enhance either electric or magnetic component of the field, while the other component is only weakly enhanced and spatially focused. In this section we propose Babinet dimer antenna that forms an electromagnetic hot spot enhancing and focusing both components of the electromagnetic field equally. The Babinet dimer antenna is formed by a direct and an inverse PA, vertically stacked so closely that their individual electric and magnetic hot spots overlap. We explore and compare two configurations, namely the Babinet bow-tie dimer (BBD) [schematically depicted in Figure 7 (a)], consisting of a bow-tie on top of an inverted bow-tie PA, and the Babinet diabolo dimer (BDD) [sketched in Figure 7 (c)] made up by an inverted diabolo on top of a diabolo PA. In both configurations, the upper PA is rotated with respect to the bottom one by 90 degrees so that both of them can be excited by the same source polarization (oriented along the long axis of the direct PA) and the upper and bottom PAs are separated by a 10 nm spacer layer with refractive index equal to 1.5. As the individual modes in the closely spaced PAs exhibit strong interaction, the dimensions of the dimer constituents have been adjusted so that the maximum field enhancement occurs at 0.8 eV for both the electric and magnetic component. For the BBD, the wing lengths of the top (↑) and bottom (↓) PAs were set to v↑ = 110 nm and v↓ = 200 nm, while for the BDD, the optimal dimensions read v↑ = 200 nm and v↓ = 110 nm. Note that the antenna providing the magnetic enhancement is in both cases situated underneath the one with the electric enhancement so that the electromagnetic hot spot is directly accessible from the top. Figure 7(b) demonstrates the formation of the electromagnetic hot spot in the BBD. The dimer is illuminated by the field 8/13 Figure 7. (a) Schematical drawing of the Babinet bow-tie dimer (BBD). A direct bow-tie (wing length 110 nm) lies on top of an inverted bow-tie (wing length 200 nm), they are mutually rotated by 90 degrees and separated by a 10 nm spacer layer. (b) Distribution of the electric (left) and magnetic (right) fields in the vicinity of the BBD. The top two subplots show the fields in a plane parallel to the individual PAs, namely inside the spacer layer with 5 nm distance from both of them [indicated by a black dotted line in panel (a)]. The bottom two subplots then show the fields in a vertical plane perpendicular to the metal bridge of the inverted bow-tie. To avoid any confusion regarding its orientation, it is outlined by a white dotted line in the schematical drawing in panel (a) and also in the top two subplots showing the fields in the horizontal plane. (c) Schematical drawing of the Babinet diabolo dimer (BDD). A direct diabolo (wing length 110 nm) lies underneath an inverted bow-tie (wing length 200 nm), they are mutually rotated by 90 degrees and separated by a 10 nm spacer layer. The distribution of the electric (left) and the magnetic (right) fields around the BBD is plotted in (d), with planar cross-sections positioned and oriented in the same manner as in the case of BBD. Note that both BBD and BDD were illuminated with a plane wave polarized along the long axes of the direct PAs and the green point marks the position, in which the electric and magnetic field enhancements are equal in magnitude (with value specified by the number). polarized along the long axis of the bow-tie and perpendicular to the long axis of inverted bow-tie, which results into formation of an electric hot spot (field enhancement 14) around the direct and a magnetic hot spot (field enhancement 13) around the inverted bow-tie. Closely spaced hot spots overlap, yielding maximum simultaneous enhancement of both fields close to 8.4 at the position indicated by the green point in Fig. 7. The inspection of Figures 3 and 5 reveales that these values are similar to those obtained for single PAs of comparable size. This indicates that the enhancement mechanism based on the charge accumulation (or the current funneling) at the wing apices is rather robust and resistant to changes in surroundings of the PA. 9/13 The other proposed design, BDD, possesses electromagnetic hotspot as well, with maximum simultaneous enhancement of 11, while the individual maxima read 21 (electric enhancement) and 16 (magnetic enhancement) [see Figure 7(d)]. These values are again close to those encountered in single PAs, despite the partial screening of the bottom diabolo by its upper counterpart. On the whole, the better performance of isolated diabolos (at least in terms of local field enhancement) imprints itself also into Babinet dimers. So far we have altogether disregarded the vectorial nature of electromagnetic fields, which can be important in certain applications. In the designs proposed above, the electric and magnetic fields in the hot spot are perpendicular to each other, but one can achieve also other mutual orientations simply by rotating the vertically stacked antennas with respect to each other. Such control over the local polarization state of the light is quite valuable, especially when we consider the aforementioned robustness with both field amplitude and orientation tightly bound to the geometry of the PAs. In comparison to previous proposals43, 44 and realizations13 of plasmonic electromagnetic hot spots, our proposal brings two benefits. (i) It enhances both field on equal basis, i.e., with the same amplitude, resonance frequency, and lateral spatial distribution. (ii) It involves two isolated antennas which can be adjusted independently, allowing extended tunability of the hot spot. Conclusion We have focused on the plasmonic antennas featuring electric, magnetic, and electromagnetic hot spots: bow-tie and inverted diabolo, diabolo and inverted bow-tie, and their dimers, respectively. We have combined two types of electric-magnetic complementarity: bow-tie/diabolo duality and Babinet's principle. For a specific resonance frequency, diabolo antennas were significantly smaller than bow-tie antennas, and thus harder to fabricate but easier to integrate. For the minimum wing length of 50 nm, bow-ties covered energy range up to 2.0 eV while diabolos only up to 1.2 eV. Diabolo antennas also exhibit considerably narrower resonances related to higher Q factor as a consequence of lower scattering cross-section. We have evaluated figures of merit for different methods of optical spectroscopy. One of the most important is the field enhancement in the hot spot, which was larger for the diabolo antennas (and also for the electric field). For the luminescence, the key figure of merit is the radiative and non-radiative decay enhancement. Here, diabolo antennas exhibited slightly stronger radiative decay enhancement but also pronouncedly stronger non-radiative enhancement, making bow-tie antenna a preferred option for the electric dipole transitions and inverted bow-tie and equivalent alternative of diabolo for the magnetic dipole transitions. Finally, we have proposed Babinet dimer antennas enhancing both the electric and magnetic field on equal basis and forming electromagnetic hot spot, which finds applications in studies of rare earth ions, optical trapping, metamaterials, or non-linear optics. Methods Simulations In all simulations, the bow-tie and diabolo antennas have been represented by two gold triangles or triangular apertures (as shown in Fig. 1) of 30 nm height on a semiinfinite glass substrate. Babinet dimers are formed by two complementary PAs (direct and inverted, each of 30 nm height) vertically separated by a 10 nm thick layer with reractive index equal to 1.5. The whole dimer lies on a semiinfinite glass substrate. The dielectric function of gold was taken from Ref.36 and the refractive index of the glass was set equal to 1.47. The electromagnetic field has been calculated with finite-difference in time-domain (FDTD) method using a commercial software Lumerical. Scattering efficiencies and the near-field distribution have been calculated using plane wave as an illumination. Transition decay rates have been calculated as the decay rate of the power radiated by oscillating electric or magnetic dipole into its surrounding (total decay rate) and into far field (radiative decay rate). The dipole has been positioned at the vertical symmetry axis of the antenna with polarization parallel with the polarization of the plasmonic near field. Its height above the antenna plane has been set to 10 nm. Data availability The datasets analysed during the current study are available from the corresponding author on reasonable request. References 1. Novotny, L. & van Hulst, N. Antennas for light. Nat. Photonics 5, 83, DOI: 10.1038/nphoton.2010.237 (2011). 10/13 2. Nie, S. & Emory, S. R. Probing single molecules and single nanoparticles by surface-enhanced raman scattering. Science 275, 1102 -- 1106, DOI: 10.1126/science.275.5303.1102 (1997). 3. Krenn, J. R. et al. Squeezing the optical near-field zone by plasmon coupling of metallic nanoparticles. Phys. Rev. Lett. 82, 2590 -- 2593, DOI: 10.1103/PhysRevLett.82.2590 (1999). 4. Weeber, J.-C. et al. Near-field observation of surface plasmon polariton propagation on thin metal stripes. Phys. Rev. B 64, 045411, DOI: 10.1103/PhysRevB.64.045411 (2001). 5. Gao, H., Cao, Q., Zhu, M., Teng, D. & Shen, S. Nanofocusing of terahertz wave in a tapered hyperbolic metal waveguide. Opt. Express 22, 32071 -- 32081, DOI: 10.1364/OE.22.032071 (2014). 6. Urbieta, M. et al. Atomic-scale lightning rod effect in plasmonic picocavities: A classical view to a quantum effect. ACS Nano 12, 585 -- 595, DOI: 10.1021/acsnano.7b07401 (2018). 7. Benz, F. et al. Single-molecule optomechanics in "picocavities". Science 354, 726 -- 729, DOI: 10.1126/science.aah5243 (2016). 8. Zhou, L., Gan, Q., Bartoli, F. J. & Dierolf, V. Direct near-field optical imaging of uv bowtie nanoantennas. Opt. Express 17, 20301 -- 20306, DOI: 10.1364/OE.17.020301 (2009). 9. Lee, A. et al. Probing dynamic generation of hot-spots in self-assembled chains of gold nanorods by surface-enhanced raman scattering. J. Am. Chem. Soc. 133, 7563 -- 7570, DOI: 10.1021/ja2015179 (2011). 10. Grosjean, T., Mivelle, M., Baida, F. I., Burr, G. W. & Fischer, U. C. Diabolo nanoantenna for enhancing and confining the magnetic optical field. Nano Lett. 11, 1009 -- 1013, DOI: 10.1021/nl103817f (2011). 11. Lorente-Crespo, M. et al. Magnetic hot spots in closely spaced thick gold nanorings. Nano Lett. 13, 2654 -- 2661, DOI: 10.1021/nl400798s (2013). 12. Bicket, I. C., Bellido, E. P., Meuret, S., Polman, A. & Botton, G. A. Correlative electron energy loss spectroscopy and cathodoluminescence spectroscopy on three-dimensional plasmonic split ring resonators. Microscopy 67, i40 -- i51, DOI: 10.1093/jmicro/dfy010 (2018). 13. Chen, Y., Chen, Y., Chu, J. & Xu, X. Bridged bowtie aperture antenna for producing an electromagnetic hot spot. ACS Photonics 4, 567 -- 575, DOI: 10.1021/acsphotonics.6b00857 (2017). 14. Bakker, R. M. et al. Magnetic and electric hotspots with silicon nanodimers. Nano Lett. 15, 2137 -- 2142, DOI: 10.1021/acs. nanolett.5b00128 (2015). 15. Itoh, T., Yamamoto, Y. S., Kitahama, Y. & Balachandran, J. One-dimensional plasmonic hotspots located between silver nanowire dimers evaluated by surface-enhanced resonance raman scattering. Phys. Rev. B 95, 115441, DOI: 10.1103/PhysRevB.95.115441 (2017). 16. Tian, Y. et al. Plasmonic heterodimers with binding site-dependent hot spot for surface-enhanced raman scattering. Small 14, DOI: 10.1002/smll.201800669 (2018). 17. Yang, T.-H. et al. Ultrahigh density plasmonic hot spots with ultrahigh electromagnetic field for improved photocatalytic activities. Appl. Catal. B: Environ. 181, 612 -- 624, DOI: https://doi.org/10.1016/j.apcatb.2015.08.014 (2016). 18. Kinkhabwala, A. et al. Large single-molecule fluorescence enhancements produced by a bowtie nanoantenna. Nat. Photonics 3, 654, DOI: 10.1038/nphoton.2009.187 (2009). 19. Blank, A., Dikarov, E., Shklyar, R. & Twig, Y. Induction-detection electron spin resonance with sensitivity of 1000 spins: En route to scalable quantum computations. Phys. Lett. A 377, 1937 -- 1942, DOI: https://doi.org/10.1016/j.physleta.2013. 05.025 (2013). 20. Kasperczyk, M., Person, S., Ananias, D., Carlos, L. D. & Novotny, L. Excitation of magnetic dipole transitions at optical frequencies. Phys. Rev. Lett. 114, 163903, DOI: 10.1103/PhysRevLett.114.163903 (2015). 21. Malakhovskii, A. V., Gnatchenko, S. L., Kachur, I. S., Piryatinskaya, V. G. & Gudim, I. A. Transformation of the HoFe3(BO3)4 absorption spectra at reorientation magnetic transitions and local properties in the excited 5F5 states of the Ho3+ ion. Phys. Rev. B 96, 224430, DOI: 10.1103/PhysRevB.96.224430 (2017). 22. Juan, M. L., Righini, M. & Quidant, R. Plasmon nano-optical tweezers. Nat. Photonics 5, 349, DOI: 10.1038/nphoton. 2011.56 (2011). 23. Smith, D. R., Pendry, J. B. & Wiltshire, M. C. K. Metamaterials and negative refractive index. Science 305, 788 -- 792, DOI: 10.1126/science.1096796 (2004). 11/13 24. Klein, M. W., Enkrich, C., Wegener, M. & Linden, S. Second-harmonic generation from magnetic metamaterials. Science 313, 502 -- 504, DOI: 10.1126/science.1129198 (2006). 25. Kusch, P. et al. Dual-scattering near-field microscope for correlative nanoimaging of sers and electromagnetic hotspots. Nano Lett. 17, 2667 -- 2673, DOI: 10.1021/acs.nanolett.7b00503 (2017). 26. Caselli, N. et al. Deep-subwavelength imaging of both electric and magnetic localized optical fields by plasmonic campanile nanoantenna. Sci. Reports 5, 9606, DOI: 10.1038/srep09606 (2015). 27. Lahiri, B., Holland, G., Aksyuk, V. & Centrone, A. Nanoscale imaging of plasmonic hot spots and dark modes with the photothermal-induced resonance technique. Nano Lett. 13, 3218 -- 3224, DOI: 10.1021/nl401284m (2013). 28. Dodson, S. et al. Optimizing electromagnetic hotspots in plasmonic bowtie nanoantennae. The J. Phys. Chem. Lett. 4, 496 -- 501, DOI: 10.1021/jz302018x (2013). 29. Cakmakyapan, S., Cinel, N. A., Cakmak, A. O. & Ozbay, E. Validation of electromagnetic field enhancement in near-infrared through sierpinski fractal nanoantennas. Opt. Express 22, 19504 -- 19512, DOI: 10.1364/OE.22.019504 (2014). 30. Falcone, F. et al. Babinet principle applied to the design of metasurfaces and metamaterials. Phys. Rev. Lett. 93, 197401, DOI: 10.1103/PhysRevLett.93.197401 (2004). 31. Hentschel, M., Weiss, T., Bagheri, S. & Giessen, H. Babinet to the half: Coupling of solid and inverse plasmonic structures. Nano Lett. 13, 4428 -- 4433, DOI: 10.1021/nl402269h (2013). 32. Bitzer, A., Ortner, A., Merbold, H., Feurer, T. & Walther, M. Terahertz near-field microscopy of complementary planar metamaterials: Babinet's principle. Opt. Express 19, 2537 -- 2545, DOI: 10.1364/OE.19.002537 (2011). 33. Zentgraf, T. et al. Babinet's principle for optical frequency metamaterials and nanoantennas. Phys. Rev. B 76, 033407, DOI: 10.1103/PhysRevB.76.033407 (2007). 34. Mizobata, H., Ueno, K., Misawa, H., Okamoto, H. & Imura, K. Near-field spectroscopic properties of complementary gold nanostructures: applicability of babinet's principle in the optical region. Opt. Express 25, 5279 -- 5289, DOI: 10.1364/OE.25.005279 (2017). 35. Horák, M. et al. Babinet's principle for plasmonic antennas: complementarity and differences. submitted . 36. Johnson, P. B. & Christy, R. W. Optical constants of the noble metals. Phys. Rev. B 6, 4370 -- 4379, DOI: 10.1103/PhysRevB. 6.4370 (1972). 37. Scholl, J. A., García-Etxarri, A., Koh, A. L. & Dionne, J. A. Observation of quantum tunneling between two plasmonic nanoparticles. Nano Lett. 13, 564 -- 569, DOI: 10.1021/nl304078v (2013). 38. Mivelle, M. et al. Bowtie nano-aperture as interface between near-fields and a single-mode fiber. Opt. Express 18, 15964 -- 15974, DOI: 10.1364/OE.18.015964 (2010). 39. Zhou, J. et al. Saturation of the magnetic response of split-ring resonators at optical frequencies. Phys. Rev. Lett. 95, 223902, DOI: 10.1103/PhysRevLett.95.223902 (2005). 40. Purcell, E. M. Spontaneous emission probabilities at radio frequencies. Phys. Rev. 69, 681, DOI: 10.1103/PhysRev.69.681 (1946). 41. Atwater, H. A. & Polman, A. Plasmonics for improved photovoltaic devices. Nat. Mater. 9, 205, DOI: 10.1038/nmat2629 (2010). 42. Édes, Z., Krápek, V. & Šikola, T. Modeling of plasmon-enhanced photoluminescence of si nanocrystals embedded in thin silicon-rich oxinitride layer. Acta Phys. Polonica A 129, A -- 70, DOI: 10.12693/APhysPolA.129.A-70 (2016). 43. Yu, P. et al. Co-enhancing and -confining the electric and magnetic fields of the broken-nanoring and the composite nanoring by azimuthally polarized excitation. Opt. Express 21, 20611 -- 20619, DOI: 10.1364/OE.21.020611 (2013). 44. Roxworthy, B. J. & Toussaint, K. C. Simultaneously tuning the electric and magnetic plasmonic response using capped bi-metallic nanoantennas. Nanoscale 6, 2270 -- 2274, DOI: 10.1039/C3NR05536A (2014). Acknowledgement We acknowledge the support by the Czech Science Foundation (grant No. 17-25799S), Ministry of Education, Youth and Sports of the Czech Republic (projects CEITEC 2020, No. LQ1601, and CEITEC Nano RI, No. LM2015041), and Brno University of Technology (grant No. FSI/STI-J-18-5225). 12/13 Author information Contributions V.K. conceived and coordinated research with help of T.S. Ma.H. and A.K. performed numerical simulations. All authors were involved in the data processing and interpretation. V.K. and Ma.H. wrote the manuscript. Competing interests The authors declare no competing interests. 13/13
1801.01670	1	1801	2018-01-05T08:34:43	Large current modulation and tunneling magnetoresistance change by a side-gate electric field in a GaMnAs-based vertical spin metal-oxide-semiconductor field-effect transistor	[ "physics.app-ph", "cond-mat.mtrl-sci" ]	A vertical spin metal-oxide-semiconductor field-effect transistor (spin MOSFET) is a promising low-power device for the post scaling era. Here, using a ferromagnetic-semiconductor GaMnAs-based vertical spin MOSFET with a GaAs channel layer, we demonstrate a large drain-source current IDS modulation by a gate-source voltage VGS with a modulation ratio up to 130%, which is the largest value that has ever been reported for vertical spin field-effect transistors thus far. We find that the electric field effect on indirect tunneling via defect states in the GaAs channel layer is responsible for the large IDS modulation. This device shows a tunneling magnetoresistance (TMR) ratio up to ~7%, which is larger than that of the planar-type spin MOSFETs, indicating that IDS can be controlled by the magnetization configuration. Furthermore, we find that the TMR ratio can be modulated by VGS. This result mainly originates from the electric field modulation of the magnetic anisotropy of the GaMnAs ferromagnetic electrodes as well as the potential modulation of the nonmagnetic semiconductor GaAs channel layer. Our findings provide important progress towards high-performance vertical spin MOSFETs.	physics.app-ph	physics	Large current modulation and tunneling magnetoresistance change by a side-gate electric field in a GaMnAs-based vertical spin metal-oxide-semiconductor field-effect transistor Toshiki Kanaki1,a), Hiroki Yamasaki1, Tomohiro Koyama2, Daichi Chiba2, Shinobu Ohya1,3,4,b) & Masaaki Tanaka1,3,c) 1Department of Electrical Engineering and Information Systems, The University of Tokyo, 7-3-1 Hongo, Bunkyoku, Tokyo 113-8656, Japan 2Department of Applied Physics, The University of Tokyo, 7-3-1 Hongo, Bunkyoku, Tokyo 113-8656, Japan 3Center for Spintronics Research Network (CSRN), Graduate School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan 4Institute of Engineering Innovation, Graduate School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan A vertical spin metal-oxide-semiconductor field-effect transistor (spin MOSFET) is a promising low-power device for the post scaling era. Here, using a ferromagnetic-semiconductor GaMnAs-based vertical spin MOSFET with a GaAs channel layer, we demonstrate a large drain-source current IDS modulation by a gate-source voltage VGS with a modulation ratio up to 130%, which is the largest value that has ever been reported for vertical spin field-effect transistors thus far. We find that the electric field effect on indirect tunneling via defect states in the GaAs channel layer is responsible for the large IDS modulation. This device shows a tunneling magnetoresistance (TMR) ratio up to ~7%, which is larger than that of the planar-type 1 spin MOSFETs, indicating that IDS can be controlled by the magnetization configuration. Furthermore, we find that the TMR ratio can be modulated by VGS. This result mainly originates from the electric field modulation of the magnetic anisotropy of the GaMnAs ferromagnetic electrodes as well as the potential modulation of the nonmagnetic semiconductor GaAs channel layer. Our findings provide important progress towards high-performance vertical spin MOSFETs. a) Corresponding author: [email protected] b) Corresponding author: [email protected] c) Corresponding author: [email protected] 2 Reducing the power consumption in integrated circuits is an important issue that we have to tackle in the 21st century. Making volatile components non-volatile is one of the most promising approaches to this issue; various non-volatile technologies, such as reconfigurable logic circuits1, non-volatile power gating2 and magnetic random access memory, are applicable to low-power-consumption electronics. A spin metal-oxide-semiconductor field-effect transistor (spin MOSFET)3,4, in which source and drain electrodes are ferromagnetic materials, is a key component for realizing those applications because of its unique output characteristics and its compatibility with existing semiconductor technologies. For practical operation of spin MOSFETs, both large current modulation by applying a gate electric field and large magnetoresistance (MR) by magnetization reversal are required. At present, lateral and vertical types of spin MOSFETs have been proposed. The lateral spin MOSFETs, in which a current flows parallel to the substrate plane and is controlled by a gate electric field applied from the top of the channel, have a large current modulation capability (5×106%5, 400%6 and 107%7); however, the problem is the small MR ratio (0.1%5, 0.005%6 and 0.027%7). Meanwhile, a vertical spin field-effect transistor (FET)8–13, in which a current flows perpendicular to the film plane and is controlled by a gate voltage applied from the side surface of the channel, is promising for large MR. Previously, we reported ferromagnetic-semiconductor GaMnAs-based vertical spin FETs that exhibit large MR ratios (60%14 and 5%15). A GaMnAs-based heterostructure is one of the most ideal material systems, because we can obtain high-quality single-crystalline GaMnAs / nonmagnetic semiconductor (GaAs) / GaMnAs trilayers and thus can suppress spin relaxation at the interfaces16–19. However, in those vertical spin FETs, the current modulation ratio by the gate voltage was small (0.5%14 and 20%15). Thus, vertical spin 3 FETs with a large current modulation are strongly required. In addition, to further improve the performance of the vertical spin FETs, we need more profound understanding of the gate electric field effect on the spin-dependent transport. In vertical spin FETs, as shown by our electric field simulations later, the gate electric field influences the electric potential profile only within 10 nm from the side surfaces of the intermediate channel layer, which limits the current modulation. Thus, the lateral size of vertical FETs should be decreased as much as possible for obtaining high-performance vertical spin FETs. In this study, to enhance the current modulation and to understand the electric field effect on the spin-dependent transport, we reduced the lateral size (= width of the mesas as explained later) of the GaMnAs-based vertical spin MOSFET to ~500 nm. We have successfully obtained a large current modulation by the gate electric field with a modulation ratio up to 130%, which is the largest value that has been ever reported for vertical spin FETs14,15. Furthermore, using the electric field simulations, we find that indirect tunneling mainly contributes to the observed large current modulation. These new findings are important steps to further improve the performance of the vertical spin FETs. Results Samples. Our vertical spin MOSFET has a thin GaAs channel (9 nm) and ferromagnetic-semiconductor GaMnAs source and drain electrodes [Fig. 1(a)] (See the Methods section). To increase the current modulation, we reduced the width of the mesas down to ~500 nm. As a gate insulator, a 40-nm-thick HfO2 film was used since it has a large relative permittivity, which also contributes to the increase of the current modulation. In this device, tunneling of holes occurs between the source and drain, 4 because GaAs is a potential barrier with a height of ~100 meV for holes in the GaMnAs layers20,21, as shown in Fig. 1(b). When the gate-source voltage VGS < 0 V (VGS > 0 V), the tunneling current flowing at the side surfaces of the mesas is increased (decreased), as shown in Fig. 1(c). MOSFET operation and its analyses. To investigate the MOSFET characteristics of this device, we measured the drain-source current IDS as a function of the drain-source voltage VDS for various VGS [Fig. 2(a)]. Nonlinear IDS–VDS characteristics were observed for each VGS (black curves), indicating that tunneling transport is dominant. Furthermore, IDS was largely controlled by VGS. We note that the gate leakage current and electric field effect on parasitic resistances (the resistances of the top/bottom GaMnAs layers, GaAs:Be layer and Au/Cr electrodes), which may induce unintended modulation of IDS, were negligibly small (see Supplementary Note 1). When VGS = 20 V, the IDS modulation ratio by VGS, which is defined by [IDS(VGS) – IDS(VGS = 0 V)] / IDS(VGS = 0 V), is around –20% [see the blue points in Fig. 2(b)]. On the other hand, when VGS = – 20 V, it reached ~130% [see the red points in Fig. 2(b)]. This IDS modulation ratio (~130%) is the largest among the values reported for vertical spin FETs thus far14,15. To understand the modulation of the band alignment in detail, we measured IDS as a function of VGS at VDS = –10 mV [Fig. 2(c)]. IDS normalized at VGS = 0 V (γ) was changed from 1 to 1.28 when VGS was changed from 0 V to –3 V [see the right axis in Fig. 2(c)], meaning that IDS was increased by 28% when VGS was changed from 0 V to – 3 V. As shown below, this large modulation of IDS cannot be understood by the electric field effect on direct tunneling. To obtain the potential distribution and to calculate IDS normalized at VGS = 0 V (γcalc), we performed electric field simulation varying the 5 electric potential at the side surface of the mesa and investigated the effect of the side-gate electric field [Fig 2(e,h)] (see Supplementary Note 2). Here, we define EV (S) as the valence band top energy EV at the side surface (interface between the side-gate electrode and the GaAs channel) with respect to the Fermi level EF in terms of hole energy. The potential profile of EV when EV (S) = 0.75 eV is shown in Fig. 2(e), which corresponds to the case of VGS = 0 V, because EF at the side surface of the GaAs channel is pinned at the middle of the band gap22. With decreasing EV (S) from 0.75 eV, the electric potential near the side surface of the mesa is decreased [Fig. 2(f,i)], whereas the electric potential in the inner region of the mesa (10 nm ≤ x) is not influenced [Fig. 2(g,j)]. As shown in Fig. 2(d), γcalc remains almost unchanged between Fig. 2(e) (γcalc = 1 when EV (S) = 0.75 eV) and (h) (γcalc = 1.028 when EV (S) = 0.15 eV) because GaAs is a potential barrier for holes in both cases. On the other hand, when EV (S) < 0.15 eV, because EV of the GaAs channel at the side surface becomes lower than EV inside the mesa, γcalc starts to increase with decreasing EV (S) [Fig. 2(d)]. This feature is different from the experimental data shown in Fig. 2(c); when VGS is changed from 0 V to –10 V, γ starts to increase at VGS = 0 V. We can see the significant difference in the curve shapes of Fig. 2(c) and Fig. 2(d). This analysis indicates that the electric field effect on the direct tunneling current cannot reproduce the experimental IDS–VGS characteristic. Instead, the main origin of the large modulation ratio observed in our device is the electric field effect on the indirect tunneling current23. The indirect tunneling current is probably caused by a large amount of Mn atoms (~1018 cm-3), which are diffused to the intermediate GaAs layer from the top and bottom GaMnAs layers and form defect states in the band gap of GaAs. Furthermore, GaAs grown at low temperature (200 °C) is known to have a large amount of arsenic 6 antisite defects (1018 – 1019 cm–3)24,25. In fact, the IDS–VDS characteristics of our device show a strong temperature dependence (see Supplementary Note 3), which indicates that indirect tunneling via defect states takes place. (If IDS were dominated only by the direct tunneling current, no temperature dependence would be observed.) Therefore, the electric field effect on indirect tunneling via defect states is the most probable origin for the large IDS modulation ratio. Tunneling magnetoresistance and its change by VGS. To investigate the spin-dependent transport of our device, we measured the drain-source resistance RDS as a function of μ0H applied along the [110] direction in the film plane with VDS = –5 mV and VGS = 0 V [Fig. 3(a)]. Here, RDS is defined by VDS/IDS, μ0 is the permeability of a vacuum and H is an in-plane external magnetic field. In the major loop (black circles), clear tunnel magnetoresistance (TMR) was observed, indicating that IDS can be controlled by the magnetization configuration. The TMR ratio, which is defined by [RDS(μ0H) – RDS(μ0H = 0 mT)]/RDS(μ0H = 0 mT)×100 (%), reached ~7% at μ0H = 20 mT, where RDS(μ0H) is the drain-source resistance at H in the major loop. This value is more than 70 times larger than the MR ratios obtained in the lateral spin MOSFETs5–7. We also observed a clear minor loop (red circles), indicating that the antiparallel magnetization configuration is stable even at μ0H = 0 mT. (In the minor loop, RDS increases with increasing μ0H from –20 mT to 60 mT, probably because the magnetizations of the top and bottom GaMnAs layers are not completely antiparallel at the peak of RDS (at μ0H = –20 mT in the major loop) and they become close to the perfect antiparallel configuration with increasing μ0H to 10 mT in the minor loop.) To investigate the influence of VGS on the spin-dependent transport, we 7 measured the VGS dependence of the TMR ratio [Fig. 3(b)]. Here, the TMR ratio corresponds to the maximum value obtained in the major loop at each VGS. The TMR ratio tends to increase as VGS is changed from 0 V to -10 V. In our device, the gate electric field can modulate the electronic states of the top/bottom GaMnAs layers as well as those of the intermediate GaAs layer, both of which can modulate TMR. Applying VGS causes the change in the hole density of the GaMnAs layers near the side surfaces of the mesas, which can change the spin polarization and magnetic anisotropy15. To understand the modulation of the magnetic anisotropy by VGS in our device, we measured TMR applying H in various in-plane directions with an angle θ with respect to the [100] axis in the counterclockwise rotation when VGS = 0, –5 and –10 V [Fig. 3(c– e)] (see Measurements section). The observed TMR ratios showed dominant uniaxial anisotropy along the [110] direction in addition to biaxial anisotropy along the <100> directions for any VGS [see the four red peaks in Fig. 3(c–e)]. With changing VGS from 0 V to –10 V, the easy axes of our device were slightly rotated toward the [010] direction (the red-colored region is extended toward the [010] direction). Furthermore, the coercive force of the top GaMnAs layer, which has a larger coercivity than the bottom GaMnAs layer, increases as VGS is changed from 0 V to –10 V (the red-colored region slightly expands outward). These results indicate that the magnetic anisotropy constants are modulated by applying negative VGS, which can also contribute to the modulation of the TMR ratio. In addition, VGS modifies the electric potential of the GaAs layer. As we discussed in the previous paragraph, the modulation of the indirect tunneling current via defect states is the most probable mechanism for the obtained large modulation of IDS. The TMR induced by indirect tunneling via defect states depends on many factors such as energy levels of defect states, band width of defect states, and life time of carries and 8 so on. The modulation of the electric potential can influence the indirect tunneling and thus TMR. Therefore, the electric field effect both on the top/bottom GaMnAs layers and on the intermediate GaAs layer contributes to the modulation of the observed TMR ratio. Surprisingly, the VGS dependence of the TMR ratio shown in Fig. 3(b) is completely opposite to the one obtained in our previous study, i.e. the TMR ratio decreased as negative VGS is applied in our previous study14. This may be caused by the difference of the easy axes between the present device and the previous one (the biaxial anisotropy was dominant in our previous work) or by the different direction of an external magnetic field (along the direction with an angle 10-degree from the [100] direction toward the [110] in our previous work). Summary In summary, we have investigated the electric field effect on the spin-dependent transport properties in a GaMnAs-based vertical spin MOSFET. We obtained a large current modulation ratio up to 130 %, which is the largest value that has ever been reported thus far for the vertical spin FETs14,15. By comparing the experimental data with the calculated results, we concluded that this large IDS modulation does not originate from the modulation of direct tunneling between the source and the drain but from the modulation of the indirect tunneling current via defect states in the intermediate GaAs layer. The TMR ratio tends to increase as negative VGS is applied, which is attributed to the electric field effect both on the top/bottom GaMnAs layers and on the intermediate GaAs layer. These results provide an important insight into the device physics for realizing high-performance vertical spin MOSFETs. 9 Methods Growth. The heterostructure composed of, from the top to the bottom, Ga0.94Mn0.06As (10 nm) / GaAs (9 nm) / Ga0.94Mn0.06As (3.2 nm) / GaAs:Be (50 nm, hole concentration p = 5 × 1018 cm-3) on a p+-GaAs (001) substrate by low-temperature molecular beam epitaxy. The growth temperatures of the top Ga0.94Mn0.06As, GaAs, bottom Ga0.94Mn0.06As and GaAs:Be layers were 195 °C, 180 °C, 200 °C and 520 °C, respectively. Process. After the growth, we partially etched the grown films and buried the etched area with a 100-nm-thick SiO2 layer for the separation of the drain electrode and the substrate. Then, a comb-shaped Au (40 nm) / Cr (5 nm) layer, whose width of the comb teeth is ~500 nm and length of them is 50 μm, was formed by electron-beam lithography and a lift-off technique. We chemically etched the area that is not covered by the Au/Cr layer and then the magnetic tunnel junctions only beneath the comb teeth area of the Au/Cr drain electrode remained after the etching. We formed a 40-nm-thick HfO2 film as a gate insulator using atomic layer deposition at a substrate temperature of 150 °C and deposited a gate electrode composed of Au (50 nm) / Cr (5 nm) by electron-beam deposition. Measurements. After the device was bonded with Au wires and indium solder, we measured the spin-dependent transport properties of our spin MOSFET with varying VGS and H at 3.8 K. To measure the θ dependence of TMR, we applied a strong magnetic field of 1 T in the opposite direction of θ to align the magnetization directions 10 and we decreased H to zero. Then, we started to measure RDS while increasing H from zero in the direction of θ. The measurements were performed at every 10° step of θ. Acknowledgements This work was partly supported by Grants-in-Aid for Scientific Research (No. 26249039, No. 16H02095), CREST program of Japan Science and Technology Agency, and Spintronics Research Network of Japan (Spin-RNJ). T. Kanaki was supported by JSPS through the program for leading graduate schools (MERIT). T. Kanaki thanks the JSPS Research Fellowship Program for Young Scientists. Author contributions Sample preparation: T. Kanaki, H. Y., T. Koyama., and D. C.; measurements: T. Kanaki and H. Y.; data analysis: T. Kanaki and H. Y.; writing and project planning: T. Kanaki, S. O., and M. T. 11 References 1. Matsuno, T., Sugahara, S. & Tanaka, M. Novel Reconfigurable Logic Gates Using Spin Metal–Oxide–Semiconductor Field-Effect Transistors. Jpn. J. Appl. Phys. 43, 6032–6037 (2004). 2. Sugahara, S. & Nitta, J. Spin-Transistor Electronics: An Overview and Outlook. Proc. IEEE 98, 2124–2154 (2010). 3. Sugahara, S. & Tanaka, M. A spin metal–oxide–semiconductor field-effect transistor using half-metallic-ferromagnet contacts for the source and drain. Appl. Phys. Lett. 84, 2307–2309 (2004). 4. Tanaka, M. & Sugahara, S. MOS-Based Spin Devices for Reconfigurable Logic. IEEE Trans. Electron Devices 54, 961–976 (2007). 5. Nakane, R., Harada, T., Sugiura, K. & Tanaka, M. Magnetoresistance of a Spin Metal–Oxide–Semiconductor Field-Effect Transistor with Ferromagnetic MnAs Source and Drain Contacts. Jpn. J. Appl. Phys. 49, 113001 (2010). 6. Sasaki, T. et al. Spin Transport in Nondegenerate Si with a Spin MOSFET Structure at Room Temperature. Phys. Rev. Appl. 2, 034005 (2014). 7. Tahara, T. et al. Room-temperature operation of Si spin MOSFET with high on/off spin signal ratio. Appl. Phys. Express 8, 113004 (2015). 8. Yamada, S., Tanikawa, K., Miyao, M. & Hamaya, K. Atomically Controlled Epitaxial Growth of Single-Crystalline Germanium Films on a Metallic Silicide. Cryst. Growth Des. 12, 4703–4707 (2012). 9. Jenichen, B., Herfort, J., Jahn, U., Trampert, A. & Riechert, H. Epitaxial Fe3Si/Ge/Fe3Si thin film multilayers grown on GaAs(001). Thin Solid Films 556, 120–124 (2014). 12 10. Wong, P. K. J. et al. Spin-Dependent Transport in Fe/GaAs(100)/Fe Vertical Spin-Valves. Sci. Rep. 6, 29845 (2016). 11. Matsuo, N., Doko, N., Takada, T., Saito, H. & Yuasa, S. High Magnetoresistance in Fully Epitaxial Magnetic Tunnel Junctions with a Semiconducting GaOx Tunnel Barrier. Phys. Rev. Appl. 6, 034011 (2016). 12. Sakai, S. et al. Low-temperature growth of fully epitaxial CoFe/Ge/Fe3Si layers on Si for vertical-type semiconductor spintronic devices. Semicond. Sci. Technol. 32, 094005 (2017). 13. Kawano, M. et al. Electrical detection of spin accumulation and relaxation in p -type germanium. Phys. Rev. Mater. 1, 034604 (2017). 14. Kanaki, T., Asahara, H., Ohya, S. & Tanaka, M. Spin-dependent transport properties of a GaMnAs-based vertical spin metal-oxide- semiconductor field-effect transistor structure. Appl. Phys. Lett. 107, 242401 (2015). 15. Terada, H., Ohya, S., Anh, L. D., Iwasa, Y. & Tanaka, M. Magnetic anisotropy control by applying an electric field to the side surface of ferromagnetic films. Sci. Rep. 7, 5618 (2017). 16. Tanaka, M. & Higo, Y. Large Tunneling Magnetoresistance in GaMnAs / AlAs / GaMnAs Ferromagnetic Semiconductor Tunnel Junctions. Phys. Rev. Lett. 87, 026602 (2001). 17. Chiba, D., Matsukura, F. & Ohno, H. Tunneling magnetoresistance in (Ga,Mn)As-based heterostructures with a GaAs barrier. Phys. E Low-dimensional Syst. Nanostructures 21, 966–969 (2004). 18. Saito, H., Yuasa, S. & Ando, K. Origin of the Tunnel Anisotropic Magnetoresistance in Ga1-xMnxAs / ZnSe / Ga1-xMnxAs Magnetic Tunnel 13 Junctions of II-VI/III-V Heterostructures. Phys. Rev. Lett. 95, 086604 (2005). 19. Elsen, M. et al. Spin transfer experiments on (Ga,Mn)As / (In,Ga)As / (Ga,Mn)As tunnel junctions. Phys. Rev. B 73, 035303 (2006). 20. Ohno, Y., Arata, I., Matsukura, F. & Ohno, H. Valence band barrier at (Ga,Mn)As/GaAs interfaces. Phys. E Low-Dimensional Syst. Nanostructures 13, 521–524 (2002). 21. Ohya, S., Muneta, I., Hai, P. N. & Tanaka, M. GaMnAs-based magnetic tunnel junctions with an AlMnAs barrier. Appl. Phys. Lett. 95, 242503 (2009). 22. Shen, H. et al. Fermi level pinning in low‐temperature molecular beam epitaxial GaAs. Appl. Phys. Lett. 61, 1585–1587 (1992). 23. Vandooren, A. et al. Analysis of trap-assisted tunneling in vertical Si homo-junction and SiGe hetero-junction Tunnel-FETs. Solid. State. Electron. 83, 50–55 (2013). 24. Kaminska, M. et al. Structural properties of As‐rich GaAs grown by molecular beam epitaxy at low temperatures. Appl. Phys. Lett. 54, 1881–1883 (1989). 25. Look, D. C. et al. Anomalous Hall-effect results in low-temperature molecular-beam-epitaxial GaAs: Hopping in a dense EL2-like band. Phys. Rev. B 42, 3578–3581 (1990). 14 Figure captions FIG. 1. (Color online) (a) Schematic illustration of the vertical spin MOSFET investigated in this study. The backside of the substrate is the source electrode, the comb shaped Au/Cr layer is the drain electrode and the Au/Cr layer above the HfO2 layer is the gate electrode. (b)(c) Schematic device operation of our vertical spin MOSFET when a gate voltage VGS is not applied (b) and when a negative gate voltage is applied (c). The orange arrows represent a drain-source current IDS. FIG. 2. (Color online) (a) Drain-source current IDS as a function of the drain-source voltage VDS with the gate-source voltage VGS ranging from –20 V to 20 V with a step of 5 V at 3.8 K. (b) IDS modulation ratio as a function of VDS with various VGS at 3.8 K. (c) Drain-source current (–IDS) (left axis) and the IDS value normalized at VGS = 0 V (γ) (right axis) as a function of VGS with VDS = –10 mV at 3.8 K. (d) Calculated IDS normalized by the one at VGS = 0 V (γcalc) as a function of EV (S). (e)(h) Calculated valence band top energy EV with respect to the Fermi level when EV (S) = 0.75 eV (e) and EV (S) = 0.15 eV (h). Here, the Fermi level corresponds to 0 eV. The vertical axis expresses the hole energy. The inset in (e) and (h) shows the structure used in our calculation. Here, the x axis represents the distance from the side surface of the mesa and the y axis denotes the distance from the interface between the bottom GaMnAs layer and the intermediate GaAs layer. The calculation was performed in the region surrounded by the dashed line. In (e,h), only the region of 0 nm ≤ x ≤ 15 nm is shown because it is sufficient to see how the gate electric field influences the electric potential in the GaAs layer. (f,g) EV vs. y at x = 1 nm (f) and 15 nm (g) when EV (S) = 0.75 eV. (i,j) EV vs. y at x = 1 nm (i) and 15 nm (j) when EV (S) = 0.15 eV. 15 FIG. 3. (Color online) (a) Drain-source resistance RDS as a function of the in-plane external magnetic field μ0H applied along the [110] direction at 3.8 K. Here, the drain-source voltage VDS was –5 mV and the gate-source voltage VGS was 0 V. The black circles correspond to the major loop and the red circles correspond to the minor loop. The black (red) arrows are the sweep directions in the major (minor) loop. The magnetization states in the major loop are indicated by the white arrows above the graph. (b) TMR ratio as a function of the gate-source voltage VGS at 3.8 K. Here, the drain-source voltage VDS was fixed at –5 mV and the external magnetic field H was applied along the [110] direction. The TMR ratio is the maximum value obtained in the major loop at each VGS. (c–e) Magnetic-field-direction dependences of the TMR ratios at VDS = –10 mV with VGS = 0 V (c), –5 V (d) and –10 V (e). 16 17 18 19 Supplementary information Large current modulation and tunneling magnetoresistance change by a side-gate electric field in a GaMnAs-based vertical spin metal-oxide-semiconductor field-effect transistor Toshiki Kanaki, Hiroki Yamasaki, Tomohiro Koyama, Daichi Chiba, Shinobu Ohya and Masaaki Tanaka Supplementary Note 1. Influence of the gate leakage current and electric field effect on the parasitic resistances The observed gate-source leakage current ranged from -400 pA to 200 pA. It is much smaller than the experimentally observed drain-source current (IDS) modulation (~0.36 μA at the drain-source voltage (VDS) of 80 mV and the gate-source voltage (VGS) of -20 V). Thus, the gate leakage current is negligibly small. The influence of the gate electric field on the parasitic resistances is also negligibly small. The resistances of the top GaMnAs layer, bottom GaMnAs layer, GaAs:Be layer and Au/Cr layers are estimated to be 1.6 mΩ, 0.5 mΩ, 20 mΩ and 10 µΩ, respectively. Here, the resistances of the top and bottom GaMnAs layers, GaAs:Be layer and Au/Cr layers are obtained using the area of the mesas, the thickness of each layer and the resistivity of GaMnAs, GaAs:Be and Au/Cr. The resistivity of the GaMnAs, GaAs:Be and Au/Cr layers was measured to be 4 mΩ.cm, 10 mΩ.cm and 5.8 μΩ.cm, respectively, using Hall-bar structures. The modulation of resistance in the GaAs:Be layer by a gate electric field was also measured to be ~3% at 3.8 K using a Hall-bar structure with a gate electrode. Because the carrier concentration of the GaMnAs layers and Au/Cr layers are much larger than that of the GaAs:Be layer, the modulation of resistance in the top and bottom GaMnAs layers and Au/Cr layers is expected to be much less than 3%. Considering that the drain-source resistance changes from 80 kΩ (at VDS = 200 mV) to 1.5 MΩ (at VDS = 5 mV) at 3.8 K (see the purple curve in Fig. S2), the modulation of the parasitic resistances is (1.6 mΩ + 0.5 mΩ + 20 mΩ + 10 µΩ)×3% / 80 kΩ ~0.8×10-6% at most, which is much smaller than the experimentally obtained value (~130% at VDS = 80 mV and VGS = -20 V). Thus, the influence of the gate leakage current and electric field effect on the parasitic resistances is negligibly small. 1 Supplementary Note 2. Details of the calculation of the electric potential profile and the normalized drain-source current in our vertical spin MOSFET The device structure used in our simulation is shown in Fig. S1. To calculate the electric potential profile with various gate voltages, we solved the following two-dimensional Poisson equation. ∂2𝐸V 𝜕𝑥2 + 2 𝜕𝐸V 𝜕𝑦2 = 𝜌 𝜖 . (S1) Here, EV is the valence band top energy of GaAs in terms of hole energy, ρ is the charge density and 𝜀 is the dielectric constant of GaAs, respectively. Because all the experimental results presented in the main text were obtained at 3.8 K, activation of donors/acceptors and thermally excited carriers can be neglected in GaAs. Thus, we set the charge density at 0 in GaAs, meaning that the right side of Equation (S1) is 0. Considering that the potential barrier height of GaAs is ~0.1 eV for holes in the GaMnAs layers, Equation (S1) is solved under the following boundary conditions.  At x = 0, EV – EF = EV  At x = W, EV – EF = 0.1 eV.1,2 (S). Here, EF is the Fermi level, EV (S) is the valence band top energy with respect to EF at x = 0 (the interface between the side-gate electrode and GaAs), L is the channel length and W is the width of the calculated region [see Fig. S1]. As the boundary condition at the interfaces of GaMnAs/GaAs (y = 0 and L), we used the EV profile obtained for GaMnAs; in the surface depletion region of GaMnAs (0 ≤ x ≤ WD), where WD is the width of the depletion layer of GaMnAs, EV has a parabolic form (S) at x = 0. When WD ≤ x, EV – EF was approximated to be 0.1 that satisfies EV – EF = EV eV, which is the same as the barrier height of GaAs for holes, in GaMnAs. Considering the above conditions, the energy profile at y = 0 and L can be expressed by the following equation. 𝐸V − 𝐸F = { (S) − 0.1) (1 − (𝐸V 2 ) 𝑥 𝑊D 0.1 (𝑊D ≤ 𝑥) + 0.1 (0 ≤ 𝑥 ≤ 𝑊D) . (S2) 2 The depletion layer width WD is expressed by 𝑊D = √ (S) 2𝜖𝐸V 𝑒𝑁A . (S3) Here, NA is an acceptor concentration of GaMnAs and we set NA at 1020 cm-3. To (S). When a gate voltage is (S) = Eg/2, where Eg is the band gap of GaAs, because of the Fermi level introduce the effect of the gate electric field, we changed EV not applied, EV pinning at the surface of GaAs. We set W = 30 nm, which is larger than the depletion layer width in the GaAs layer (2–15 nm) [see Fig. 2(e,h)]. The parameters used in the potential calculation are summarized in Table S1. Equation (S1) was solved using Jacobi's iterative method so that the potential difference between the present step and the previous step at all (x, y) becomes less than 10-9 eV. Each mesh is a rectangle with a width in the x direction (Δx) of 0.1 nm and a width in the y direction (Δy) of 0.1 nm. The drain-source current IDS at EV (S) can be expressed by the following equation. 𝑊mesa 2 𝐼DS = 2 ∫ 𝑊D (S)(𝑥)𝑑𝑥 𝐽 𝐸V × 𝐿mesa × 𝑁mesa. (S4) Here, Wmesa is the width of our mesa (500 nm), 𝐽 (𝑆)(𝑥) is the current density at x and 𝐸V EV (S), Lmesa is the length of our mesa (50 μm) and Nmesa is the number of the mesas (10). When the applied voltage between the source and the drain V is much smaller than the barrier height, the carrier energy E dependence of tunneling probability can be neglected. Thus, 𝐽 (𝑆)(𝑥) can be described by 𝐸V 0 (S)(𝑥) ∝ ∫ 𝐷top(𝐸)𝐷bot(𝐸 + 𝑒𝑉)𝑇 𝐽 𝐸V (S)(𝑥) 𝐸V −𝑒𝑉 0 𝑑𝐸 (S5) = 𝑇 (S)(𝑥) × ∫ 𝐷top(𝐸)𝐷bot(𝐸 + 𝑒𝑉) 𝐸V −𝑒𝑉 𝑑𝐸. Here, Dtop, Dbot and 𝑇 (S)(𝑥) are the density of states in the top GaMnAs layer, the 𝐸V 3 density of states in the bottom GaMnAs layer and the tunneling probability at x and EV (S), respectively. Using equation (S4) and (S5), we can obtain the following relationship between IDS and 𝑇 (S)(𝑥). 𝐸V 𝑊mesa 𝐼DS ∝ 2 ∫ 2 𝑊D 𝑇 (S)(𝑥)𝑑𝑥 𝐸V . (S6) Note that we consider that Dtop and Dbot do not depend on EV (S). Therefore, only 𝑇 (S)(𝑥) is dependent on EV 𝐸V (S). 𝑇 (S)(𝑥) can be calculated using the Wentzel-Kramers-Brilluion approximation. The 𝐸V right side of Equation (S6) can be calculated as follows. 𝑊mesa 2 2 ∫ 𝑊D 𝑇 (S)(𝑥)𝑑𝑥 𝐸V 𝑊 = 2 ∫ exp 𝑊D 𝐿 −2 ∫ 0 ( + 2 ( 𝑊mesa 2 − 𝑊) √2𝑚ℎ (𝐸 (S)(𝑥, 𝑦) − 𝐸F) 𝑉,𝐸V ℏ 𝑑𝑦 𝑑𝑥 ) (S7) × exp 𝐿 −2 ∫ 0 ( √2𝑚ℎ (𝐸 (S)(𝑥 = 𝑊, 𝑦) − 𝐸F) 𝑉,𝐸V ℏ 𝑑𝑦 . ) Here, 𝐸 (S)(𝑥, 𝑦) is the valence band top energy at (x, y) and EV (S), and ħ is the 𝑉,𝐸V reduced Planck constant. Thus, the calculated IDS normalized at VGS = 0 V, corresponding to EV (S) = Eg/2, γcalc can be described as follows. 4 𝑇 (S) 𝐸V = 𝐸g 2 (𝑥)𝑑𝑥 . (S8) 𝑊mesa 2 𝛾calc = ∫ 𝑊D 𝑊mesa 2 𝑇 (S)(𝑥)𝑑𝑥 𝐸V ⁄ ∫ 𝑊D 5 FIG. S1. Schematic illustration of the device structure used in our electric potential calculation. Here, L is the GaAs-channel length (9 nm) and W is the width of the calculated area surrounded by the dashed lines. In our calculation, W was set at 30 nm, which is large enough to see how the gate electric field influences the electric potential profiles. 6 TABLE S1. Parameters used in the calculation of the electric potential profiles and γcalc. Parameters (unit) L (nm) W (nm) Δx (nm) Δy (nm) mh (kg) Eg (eV) Values 9 30 0.1 0.1 0.45m0 1,2 1.51914 7 Supplementary Note 3. Temperature dependence of the drain-source resistance RDS. We show RDS vs. VDS at various temperatures in Supplementary Fig. S2. If IDS were dominated by direct tunneling, RDS vs. VDS characteristics would not depend on the temperature. In our vertical spin MOSFET, when the temperature was changed from 300 K to 3.8 K, RDS was changed from 2 kΩ to 1.2 MΩ at VDS = -10 mV. The strong temperature dependence of RDS is experimental evidence of the indirect tunneling via defect states. 8 FIG. S2. Drain-source resistance RDS as a function of VDS at various temperatures in our vertical spin MOSFET. Here, VGS = 0 V. 9 References S1. Ohya, S., Muneta, I., Hai, P. N. & Tanaka, M. GaMnAs-based magnetic tunnel junctions with an AlMnAs barrier. Appl. Phys. Lett. 95, 242503 (2009). S2. Ohno, Y., Arata, I., Matsukura, F. & Ohno, H. Valence band barrier at (Ga,Mn)As/GaAs interfaces. Phys. E Low-Dimensional Syst. Nanostructures 13, 521–524 (2002). 10
1905.02315	1	1905	2019-05-07T01:27:39	A Polarization-insensitive and High-speed Electro-optic Switch Based on a Hybrid Silicon and Lithium Niobate Platform	[ "physics.app-ph", "physics.optics" ]	We propose and demonstrate a polarization-insensitive and high speed optical switch unit based on a silicon and lithium niobate hybrid integration platform. The presented device exhibits a sub nano-second switching time, low drive voltages of 4.97 V, and low power dissipation due to electrostatic operation. The measured polarization dependence loss was lower than 0.8 dB. The demonstrated optical switch could provide as a building block for polarization-insensitive and high-speed optical matrix switches.	physics.app-ph	physics	A Polarization-insensitive and High-speed Electro- optic Switch Based on a Hybrid Silicon and Lithium Niobate Platform SHENGQIAN GAO,1 MENGYUE XU,1 MINGBO HE,1 BIN CHEN,2 XIAN ZHANG,2 ZHAOHUI LI,1LIFENG CHEN,1 YANNONG LUO,3 LIU LIU,2 SIYUAN YU,1,4 AND XINLUN CAI1,* 1 State Key Laboratory of Optoelectronic Materials and Technologies and School of Electronics and Information Technology, Sun Yat-sen University, Guangzhou 510000, China 2 South China Academy of Advanced Optoelectronics, South China Normal University, Guangzhou 510000, China 3Guangxi Key Laboratory of Thalassemia Research, Guangxi Medical University, Nanning, 530021, China 4Photonics Group, Merchant Venturers School of Engineering, University of Bristol, Bristol BS8 1UB, UK. * [email protected], [email protected], [email protected] Received XX Month XXXX; revised XX Month, XXXX; accepted XX Month XXXX; posted XX Month XXXX (Doc. ID XXXXX); published XX Month XXXX We propose and demonstrate a polarization-insensitive and high speed optical switch unit based on a silicon and lithium niobate hybrid integration platform. The presented device exhibits a sub nano-second switching time, low drive voltages of 4.97 V, and low power dissipation due to electrostatic operation. The measured polarization dependence loss was lower than 0.8 dB. The demonstrated optical switch could provide as a building block for polarization-insensitive and high-speed optical matrix switches. extinction ratio and large port number, but require a high turn-on voltage of >50 V, which complicates the driver design and limits its application. To date, silicon photonic switches with high switching speeds, high extinction ratio and low drive voltage remains a challenging research objective. Previously, we have demonstrated ultra-high speed and low loss Mach -- Zehnder (MZ) modulators based on hybrid integration of lithium niobate (LN) phase shifter with passive silicon circuitry [25]. The devices exhibited low insertion loss of <2.5 dB, a modulation efficiency of 2.2 Vcm, and EO bandwidth of more than 70 GHz. In this paper, we demonstrate a polarization-insensitive MZ switch based on hybrid integration of LN phase shifter with silicon photonic circuits. The presented devices show sub-nanosecond switching speed, low drive voltages of around 5.97 V and low polarization dependence of <0.8 dB. Moreover, the present devices feature energy-efficient electrostatic operation with no power dissipation when holding the switch in the cross or bar state. Fast optical switch (FOS), with switching times in the micro- to nano-second region, is one of the key enabling optical components for optical packet switching (OPS) [1] and optical burst switching [2]. It brings the benefits of bit rate and format transparency which provide greater agility and flexibility to optical networks. Recently, the rise of high-performance computing and Data Centre Networks (DCNs) in recent years has created a need for FOS, which could enable high bandwidth, low latency, energy-efficient optical interconnect among the servers and racks [3-5]. Leveraging an advanced complementary metal-oxide-semiconductor (CMOS) manufacturing process, silicon photonics has emerged as a powerful platform for high-density photonic integrated circuits to the possibility of low-cost and high-volume production of photonic integrated circuits (PICs) [6-10]. In the past few years, silicon photonic switches have been reported exploiting the thermo-optic (TO) effect [11-15], the free-carrier dispersion effect [16-21], and micro-electro-mechanical-systems (MEMS) technology [22-24]. TO switches suffer from slow switching speed in the order of tens of microseconds or even milliseconds. To achieve nanosecond-scale switching times, free-carrier dispersion effect through carrier injection or depletion is widely exploited for high-speed electro- optic (EO) silicon switch fabrics. Unfortunately, free-carrier dispersion is intrinsically absorptive, degrading not only the insertion loss but also the extinction ratio of the switches. MEMS- actuated optical switch fabrics exhibit low insertion loss, excellent Fig. 1. (a) Schematic structure of the polarization insensitive FOS unit; (b) Cross section view of the X(Y)-polarization MZ switches. A schematic diagram of the polarization insensitive FOS unit is shown in Fig. 1(a). The device consists of a bottom silicon waveguide layer, a top LN waveguide layer and vertical adiabatic couplers (VACs) which transfer the optical power between the two layers. The top waveguides, formed by dry-etching of an X-cut LN thin film, serve as high-speed EO phase shifters where ultra-fast Pockels effect occurs. The bottom silicon circuit supports all of the passive functions, consisting of 3 dB multimode interference (MMI) couplers that split and combine the optical power, and two- dimensional grating couplers (2D-GC) for polarization-insensitive off-chip coupling. The VACs, formed by silicon inverse tapers and superimposed LN waveguides, serve as interfaces to couple light up and down between the silicon waveguides and LN waveguides. A mode calculation result (using finite difference eigenmode solver, Lumerical Mode Solution [26]) indicates that nearly 100% optical power can be transferred from silicon waveguide to the LN waveguide, and vice versa [25]. The input signal, coupled through the 2D-GC, is decomposed into two orthogonal polarization components, X-polarization (X-pol) and Y-polarization (Y-pol), and are coupled into a pair of orthogonal waveguides, both in the TE mode (see in Fig. 1(a)). Sharing the same polarization the they are switched by and dispersion, corresponding X- and Y-polarization MZ switches designed for only TE mode, and then sent to the two output 2D-GC (Output-1 and Output-2). The Cross section the device is shown in Fig. 1(b). The LN waveguides have a top width of w = 1 μm, a slab thickness of s = 420 nm, a rib height of h = 180 nm. The thickness of electrodes was set to t = 600 nm, and the gap between the waveguides and electrodes was set to 2.75 μm. The electrodes are designed in a single-drive push -- pull configuration, so that applied voltage induces a positive phase shift in one arm and a negative phase shift in the other. The length of the arms of the MZ switches are designed to be 4 mm. The device fabrication process is shown in Fig. 2. The device was fabricated in a silicon-on-insulator (SOI) wafer with 3-μm thickness buried oxide (BOX) and 220-nm thickness silicon waveguide. Firstly, a shallow etched 70 nm 2D-GCs and a 220nm Si waveguide were defined by e-beam lithography (EBL) and inductively coupled plasma (ICP) using hydrogen bromide (HBr), successively. Then a X-cut LN on insulator (LNOI) wafer with silicon substrate, commercially available from NANOLN, was filp-bonded to the patterned SOI wafer through an adhesive bonding process using benzocyclobutene (BCB). After that, the substrate of the LNOI was removed by mechanical grinding and ICP. Then, the BOX layer was removed by a dry etching process. Hydrogen silsesquioxane (HSQ, FOX-16 by Dow Chemical) was then spin-coated on the 600-nm thick LN membrane followed by EBL patterning. Through plasma etching in an inductively coupled (ICP) etching system, the waveguide patterns are transferred into LN. Finally, a liftoff process was performed to produce the Au electrodes. The scanning electron microscope (SEM) image of the fabricated electrode and LN waveguide are shown in Fig. 3(a). Fig. 3(b) shows the cross-section of the fabricated LN waveguides with a sidewall angle of 60°. The total footprint of the device is about 6.0 mm×1 mm. Fig. 2. Fabrication flow of the proposed polarization insensitive FOS unit. Fig. 3 SEM image of (a) LN waveguide, and (b) cross section of X(Y)-polarization MZ switches. Fig. 4. (a) Optical micrograph of 2D-GCs; (b) Optical micrograph of the devices in back to back configuration; (c) The measured P- and S- coupling spectra of the proposed 2D-GCs. The 2D-GCs are the key components for realizing the polarization insensitive operation [27-28]. The optical micrograph of the 2D-GCs used in the present device is shown in Fig. 4 (a). To measure the PDL, one of the most important performance metrics, two identical 2D- GCs were connected in a back-to-back configuration as shown in Fig. 4 (b). The measured coupling spectra for P- and S- polarization, illustrated in Fig.4 (c), indicate that the PDL is less than 0.8 dB for C- band. The P- or S-polarized input light was calibrated by measuring the transmission of a TE grating coupler for S-polarization, which was co-fabricated with the present device. As shown is Fig. 5, the measured coupling efficiencies is -6.9 dB at the central wavelength of 1547 nm and the 1-dB and 3-dB bandwidth are measured to be 27 nm and 43nm, respectively. The coupling efficiency of the present 2D-GCs is relatively low due to the unoptimized thickness of the BOX layer in the current SOI wafer (3 μm). The coupling efficiency can be significantly improved by using a substrate transfer technique as demonstrated in ref. [28]. obtained for S-polarization, which further confirms the broadband polarization insensitive operation of the circuit. Fig. 5. The measured transmission of (a) X-, and (b) Y-polarization MZ switches at different driving voltages. The measured transmission of the X- and Y-polarization MZ switches at different driving voltages and a fixed wavelength of 1550 nm are shown in Fig. 5. Light wave from a wavelength tunable laser was coupled to the waveguides of the device via a polarization controller (PC) and a single mode fiber. Several TE grating couplers were co-fabricated on the chip in order to calibrate the input polarization states to be X-, or Y-polarization. The transmittances at the Output-1 and Output-2 ports, when the X- or Y- polarization was introduced to Input-1 port, are plotted in Fig. 5 as a function of the voltages applied to the electrode. The X-polarization switch takes the cross/bar state at a voltage of 8.59 V/3.62 V, while Y- polarization switch takes the cross/bar state at a voltage of 9.52 V/4.55 V. The measured Dc 𝑉𝑉𝜋𝜋 is 4.97 V. Thus, the circuit would give a polarization insensitive cross or bar state when both of the X- and Y-polarization switches take their cross or bar state. Both of the extinction ratio of the switches for X- and Y-polarization was measured to be > 40 dB, as shown in Fig. 5. The on-chip insertion loss of the polarization diversity switch was estimated to be around 2 dB by subtracting the coupling loss of 2D-GCs. It should be noted that the circuit is very energy efficient because it consumes power only when the state changes, and no power is consumed when holding the switch in the cross/bar state. To examine the spectral response of the circuit, the transmittance spectra of the cross and bar ports in the cross and bar states for X-, Y-, and a mixed polarization are shown in Fig. 6. Extinction ratios of 26 dB and 28 dB have been achieved in the C-band at both cross and bar output ports for X- and Y-polarizations, respectively. In addition, Fig. 6(c) shows the transmission spectra for the S-polarization, which forms an angle of approximately 45 degrees to the X- or Y- polarization directions. An extinction ratio of at least 28 dB was Fig. 6. The transmittance spectra of the cross and bar ports in the cross and bar states for (a) X-, (b) Y-, and (c) S-polarization. Fig. 7. Temporal response of the EO switches. Finally, we characterized the dynamic switching properties of the EO switches. A square-wave electrical signal with a repetition rate of 500MHz and a duty cycle of 50%, generated from an arbitrary signal generator (MICRAM), was applied to the electrode of the phase shifter through a RF probe. A 50 GHz broadband 19. N. Dupuis, Optical Fiber Communication Conference (Optical Society of America, 2016), paper 1-3. 20. N. Dupuis, B. G. Lee, A. V. Rylyakov, D. M. Kuchta, C. W. Baks, J. S. Orcutt, D. M. Gill, W. M. Green, and C. L. Schow, J. Lightwave Technol. 33, 3597 (2015). 21. X. Wang, L. Zhou, R. Li, J. Xie, L. Lu, K. Wu and J. Chen, Optica 4, 507 (2017). 22. T. J. Seok, N. Quack, S. Han, R. S. Muller, and M. C. Wu, Optica 3, 64 (2016). 23. S. Han, T. J. Seok, N. Quack, B.-W. Yoo, and M. C. Wu, Optical Fiber Communication Conference (Optical Society of America, 2016), paper M2K- 2. 24. T. J. Seok, K. Kwon, J. Henriksson, J. Luo, and M. C. Wu, Optica 6, 490 (2019). 25. 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, Nat. Photon. 13, 359 (2019). 26. Lumerical Solutions, retrieved Inc. https://www.lumerical.com. 27. B. Chen, X. Zhang, X. Wen, Z. Ruan, Y. zhu, and L. Liu, Asia Communications and Photonics Conference, paper 1-2 (2018). 28. Y. Luo, Z. Nong, S. Gao, H. Huang, Y. Zhu, L. Liu, L. Zhou, J. Xu, L. Liu, S. Yu, and X. Cai, Opt. Lett. 43, 474 (2018). amplifier (SHF 807) was used to amplify the driving signal to the switch together with a DC bias. The optical output intensity is recorded using an oscilloscope (Tektronix DSA8300). As shown is Fig. 7, the rising and falling times were measured to be 100 ps and 312 ps, respectively, indicating an ultra-fast switching speed. In conclusion, we have designed and demonstrated a polarization-insensitive and high-speed optical switch circuit based on the hybrid silicon and LN platform. The polarization-insensitive operation was achieved with a polarization-diversity technique by using 2D-GC with low PDL of less than 0.8 dB in C-band. The demonstrated device exhibits switching speed of less than 1 ns, an insertion loss of 2 dB and a low drive voltage of around 4.97 V. The switch demonstrated here could provide as a building block for polarization-insensitive, silicon photonic matrix switches. Funding. National Natural Science Foundation of China (NSFC) (11690031, 61575224, 61622510, 61675069), Local Innovative and Research Teams Project of Guangdong Pearl River Talents Program (2017BT01X121), Guangzhou Science and Technology Program (201707010444) References 1. S. Yao, B. Mukherjee, S. Dixit, IEEE Commun. Mag. 38, 84 (2000). 2. G. Zervas, R. Nejabati, Z. Wang, D. Simeonidou, S. Yu, M. O'Mahony, Optical Fiber Communication and the National Fiber Optic Engineers Conference (Optical Society of America, 2005), paper OWC2. high-speed large-scale and 3. A. Biberman, K. Bergman, Rep. Prog. Phys. 75, 1 (2012). 4. C. Kachris, K. Kanonakis, I. Tomkos, IEEE Commun. Mag. 51, 39 (2013). 5. Q. Cheng, S. Rumley, M. Bahadori, and K. Bergman, Opt. Express 26, 16022 (2018). 4, 518 (2010). 6. D. A. B. Miller, Integrated Photonics Research, Silicon and Nanophotonics and Photonics in Switching (Optical Society of America, 2010), paper PMB3. 7. G. T. Reed, G. Mashanovich, F. Y. Gardes, and D. J. Thomson, Nat. Photon. 8. M. J. R. Heck, H-W. Chen, A. W. Fang, B. R. Koch, D. Liang, H. Park, M. N. Sysak, and J. E. Bowers, IEEE J. Sel. Top. Quantum Electron. 17, 333 (2011). 9. W. Bogaerts, R. Baets, P. Dumon, V. Wiaux, S. Beckx, D. Taillaert, B. Luyssaert, J. V. Campenhout, P. Bienstman, and V. Thourhout, J. Lightwave Technol. 23, 401 (2005). 10. C. Sun, M. T. Wade, Y. Lee, J. S. Orcutt, L. Alloatti, M. S. Georgas, A. S. Waterman, J. M. Shainline, R. R. Avizienis, S. Lin, B. R. Moss, R. Kumar, F. Pavanello, A. H. Atabaki, H. M. Cook, A. J. Ou, J. C. Leu, Y-H. Chen, K. Asanović, R. J. Ram, M. A. Popović, and V. M. Stojanović, Nat. 528, 534 (2015). 11. Y. Shoji, K. Kintaka, S. Suda, H. Kawashima, T. Hasama, and H. Ishikawa, Opt. Express 18, 9071 (2010). 12. L. Chen and Y.-K. Chen, Opt. Express 20, 18977 (2012). 13. S-H. Kim, K. Tanizawa, Y. Shoji, G. Cong, K. Suzuki, K. Ikeda, H. Ishikawa, S. Namiki, and H. Kawashima, Opt. Express 22, 29818 (2014). 14. H. Yang, Y. Kuan, T. Xiang, Y. Zhu, X. Cai, and L. Liu, Opt. Express 26,14340 (2018). 17, 24020 (2009). 15. S. Wang and D. Dai, Opt. Lett. 43, 11, 2531 (2015). 16. J. V. Campenhout, W. M. Green, S. Assefa, and Y. A. Vlasov, Opt. Express 17. M. Yang, W. M. Green, S. Assefa, J. V. Campenhout, B. G. Lee, C. V. Jahnes, F. E. Doany, C. L. Schow, J. A. Kash, and Y. A. Vlasov, Opt. Express 19, 47 (2011). 18. B. G. Lee, A. V. Rylyakov, W. M. Green, S. Assefa, C. W. Baks, R. Rimolo- Donadio, D. M. Kuchta, M. H. Khater, T. Barwicz, C. Reinholm, E. Kiewra, S. M. Shank, C. L. Schow, and Y. A. Vlasov, J. Lightwave Technol. 32, 743 (2014).
1908.06393	2	1908	2019-08-25T00:50:42	Electron g-factor engineering for non-reciprocal spin photonics	[ "physics.app-ph", "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	We study the interplay of electron and photon spin in non-reciprocal materials. Traditionally, the primary mechanism to design non-reciprocal photonic devices has been magnetic fields in conjunction with magnetic oxides, such as iron garnets. In this work, we present an alternative paradigm that allows tunability and reconfigurability of the non-reciprocity through spintronic approaches. The proposed design uses the high-spin-orbit coupling of a narrow-band gap semiconductor (InSb) with ferromagnetic dopants. A combination of the intrinsic and a gate-applied electric field gives rise to a strong external Rashba spin-orbit coupling (RSOC) in a magnetically doped InSb film. The RSOC which is gate alterable is shown to adjust the magnetic permeability tensor via the electron g-factor of the medium. We use electronic band structure calculations (k$\cdot$p theory) to show the gate-adjustable RSOC manifest itself in the non-reciprocal coefficient of photon fields via shifts in the Kerr and Faraday rotations. In addition, we show that photon spin properties of dipolar emitters placed in the vicinity of a non-reciprocal electromagnetic environment is distinct from reciprocal counterparts. The Purcell factor (F$_{p}$) of a spin-polarized emitter (right-handed circular dipole) is significantly enhanced due to a larger g-factor while a left-handed dipole remains essentially unaffected. Our work can lead to electron spin controlled reconfigurable non-reciprocal photonic devices.	physics.app-ph	physics	Electron g -factor engineering for non-reciprocal spin photonics Parijat Sengupta1, Chinmay Khandekar1, Todd Van Mechelen1, Rajib Rahman2 and Zubin Jacob1 1Department of Electrical and Computer Engineering Birck Nanotechnology Center Purdue University, West Lafayette, IN 47907, USA 2School of Physics, The University of New South Wales, Sydney, NSW 2052, Australia We study the interplay of electron and photon spin in non-reciprocal materials. Traditionally, the primary mechanism to design non-reciprocal photonic devices has been magnetic fields in conjunction with magnetic oxides, such as iron garnets. In this work, we present an alternative paradigm that allows tunability and reconfigurability of the non-reciprocity through spintronic approaches. The proposed design uses the high-spin-orbit coupling (soc) of a narrow-band gap semiconductor (InSb) with ferromagnetic dopants. A combination of the intrinsic soc and a gate-applied electric field gives rise to a strong external Rashba spin-orbit coupling (RSOC) in a magnetically doped InSb film. The RSOC which is gate alterable is shown to adjust the magnetic permeability tensor via the electron g-factor of the medium. We use electronic band structure calculations (k·p theory) to show the gate-adjustable RSOC manifest itself in the non-reciprocal coefficient of photon fields via shifts in the Kerr and Faraday rotations. In addition, we show that photon spin properties of dipolar emitters placed in the vicinity of a non-reciprocal electromagnetic environment is distinct from reciprocal counterparts. The Purcell factor (Fp) of a spin-polarized emitter (right-handed circular dipole) is significantly enhanced due to a larger g-factor while a left-handed dipole remains essentially unaffected. Our work can lead to electron spin controlled reconfigurable non-reciprocal photonic devices. I. Introduction Non-reciprocal photonic materials such as ferrites and magnetized plasmas are central to the design of optical isolators and circulators1. While technology exists in the microwave regime, there is a major impetus driving on-chip miniaturization of non-reciprocal devices for quantum2 to classical3 applications. A particular frontier in this regard is connected to time modulation as a possible pathway to achieve non-reciprocity as an alternative to using magnetic materials. However, significant challenges remain - primarily, insertion loss and the high speed modulation of such effects - which makes it an area of active interest to carry out a search for new materials exhibiting non-reciprocity. There is an intimate connection between photon spin4 and non-reciprocal materials exhibiting gyrotropy. A classical analysis5 of gyrotropic media reveals that the eigen states of such a medium are circularly polarized with differing phase velocities; however, the role of spin in the near-field of gyrotropic media has not been fully analyzed. In this work, we put forth approaches to probe the near-field spin properties of non-reciprocal media. It is pertinent to note here that the special case of moving media which displays magneto-electric non-reciprocity has fundamental similarities to the Kramers theorem in the near-field regime.6 Recently, gyrotropy was demonstrated to be equivalent to effective photon mass through a direct comparison equation.7 -- 9. with an optical-analog of Gyrotropy, similar to Dirac mass, is accompanied by a low energy (frequency) band gap for propagating waves. Within this band gap, Maxwellian spin waves can exist with unidirectional propagation which are closely the Dirac the edge magneto-plasmons. related to Jackiw-Rebbi waves that occur at the interface of positive and negative mass media. In addition, the gyro-electric phase of atomic matter combines the principles of non-locality and non-reciprocity to achieve skyrmionic texture of photonic spins in momentum space. This non-local topological electromagnetic phase can host helicity-quantized unidirectional10 edge waves fundamentally different from their classical counterparts - This advancement illustrates how hitherto unexplored forms of gyrotropy can lead to creation of intriguing Maxwellian spin waves as well as spin-1 photonic skyrmionic textures. An equally fundamental application of non-reciprocal materials lies in controlling heat transport11,12; thermal energy density in the near-field of a planar slab of gyrotropic media has been predicted to show unidirectional transport behavior even under equilibrium conditions13. from universal spin-momentum locking of evanescent waves14,15 in the near-field of a non-reciprocal slab. arises effect This this paper the potential The focus of is electron-spin control of gyrotropy which has to utilize spintronic devices with applications requiring photonic non-reciprocity.16 Typically, conventional gyro-electric media rely on cyclotron orbits and orbital angular momentum of electrons interacting with a fixed magnetic field; gyro-magnetic media, on the other hand, obtain their non-reciprocal behavior from electron spin angular momentum interaction with the static magnetic bias. These materials are also widely known as magneto-optic media. Here, we couple band structure calculations - performed using an eight-band k·p Hamiltonian adapted17 to quantum wells - to the theory of magnetic permeability tensors. This leads to a computation of the non-reciprocity coefficient inside matter for photon fields. We propose nanoscale thick InSb quantum well structures18 exhibiting optical non-reciprocity. Our structures are more amenable to use in small sized integrated systems and unlike YIG, the growth of quantum well devices is established easily through molecular beam epitaxy. We emphasize that leveraging the spin of the electron with a gate field for non-reciprocal photonics remains unexplored heretofore. InSb has been previously explored19 for its non-reciprocity with emphasis on its gyro-electric behavior, the present work shows that it is possible to design "multi-gyroic" materials which have non-reciprocity in both the electric and magnetic off-diagonal permeability and susceptibility tensor components. We further note while similar analyses with gyroelectric media exist in literature1,20 wherein non-reciprocity has been demonstrated21, such realizations however are generally incumbent solely upon the external magnetic field and offer no recourse to further modulations via microscopic device rearrangements. Furthermore, throughout the manuscript, we do not invoke the terminology of chirality22. Chirality (i.e. traditional optical activity) is a reciprocal phenomenon, and the fields of metamaterials, plasmonics, and chemistry define it as a coupling coefficient of electric and magnetic fields. Gyrotropic non-reciprocity, in contrast, associated with photon spin inside matter, couples the orthogonal components of the electric (or magnetic) fields. This effectively changes The present work, as mentioned above, combines a large spin-orbit coupling, narrow band gap, and crystalline asymmetry of the target nanostructure, materializing in a significant external Rashba spin-orbit field23,24. the material response to an impinging light beam in the presence of an external magnetic field which is discernible from appropriate magneto-optical data. We now give a succinct description of the arrangement on which the theoretical and computational analysis of the latter sections is centred. The model structure is a magnetically-doped InSb (Fig. 1a) slab with a permanent axis of magnetization (M) normal (aligned to the z -axis) to the x-y plane and forms the optically active component. The slab (Fig. 1a) is also placed under an external magnetic field parallel to M while a gate electrode is affixed to the top. The non-reciprocity of the magnetized InSb slab is captured by the non-zero off-diagonal elements in the permeability (µ) tensor matrix. However, beyond the influence of the magnetic fields, the extent to which such non-reciprocity manifests, is also functionally dependent on the gyromagnetic ratio (γ = g (e/2m∗ e)). Here, e stands for the electronic charge and m∗ e is the effective electron mass. The g-factor, therefore, evidently via γ determines the solutions to Maxwell equations that govern the light-matter interaction in this setup. The middle figure (Fig. 1b) denotes this process wherein a tailor-able g-factor arises 2 as the light beam propagates through a medium with significant external Rashba field identified through the spin-momentum locked states on a equi-energy circular contour. As tangible illustrations of such synergy - albeit indirect - between a photon beam and the Rashba spin orbit coupling (RSOC), we show 1) variations in the characteristic magneto-optical measurements (MO), in particular, the Kerr and Faraday rotation with a varying electric field and 2) the Purcell factor of non-reciprocal photon spin-polarized dipole emission. Briefly, we note that changes to the Rashba coupling parameter (λR) through a gate electric field and the dispersion relation (through additional confinement and strain etc.) revises the g-factor profile; a higher λR leading to an enhanced value, and revealed as greater Kerr and Faraday rotations.25 We also show electron spin control of photon-spin dependent Purcell factor26,27. Before we proceed to a complete analysis of the g-factor engineered non-reciprocal phenomena, a note about the organization of the paper is in order: In Section II steps are outlined for the g-factor calculation beginning with the model Hamiltonian for the InSb slab; this is followed by a quantitative discussion on electron spin-orbit coupling governed Kerr and Faraday rotations that characterize the viability of non-reciprocity driven magneto-optical devices (Section III). The Purcell factor, and its numerical determination is taken up next in Section IV and we close by summarizing the key findings in Section V that also touches upon the possibilities of extending the current work to include aspects of material and structural optimization. II. Theory The basis of all calculations presented in this paper begins with two essential steps : 1) Constructing the permeability (µ) tensor matrix that ties its behaviour to the extrinsic Rashba spin-orbit coupling and 2) band dispersion of the two-dimensional (2D) FM. In this section, their analytic expressions are presented in the same order below. Note that at this stage the steps are generalized and no target material is specified; however, we will allude to possibilities during a numerical evaluation of the µ matrix and the overall band dispersion later in the manuscript. We begin by writing the Landau-Lifshitz equation that governs all magnetization (M) behaviour in a magnet. In presence of Gilbert damping, and in an external magnetic field (H) it takes the form28,29 = γµ0 (M × H) + (M × (M × H)) , (1) ∂M ∂t αγ M where, γ = ge 2m∗ e . (2) In Eq. 1, g is the Lande factor, m∗ e is the electron's effective mass, and α is the Gilbert damping. The magnetic pemeability in vacuum is µ0. Without loss of 3 FIG. 1. The schematic represents the arrangement considered in this work. The left figure (a) shows a unit cell of ferromagnetically doped InSb (red atom denotes In while blue stands for Sb) irradiated with a beam of light (wavy line) that traverses its body and emerges on the opposite side. The passage of the light beam is governed by the constitutive parameters, 1 and µ1, of InSb, which is gyrotropic with an inherent magnetization. Note that for gyrotropy to be observed, an out-of-plane magnetic field (Hz) is applied to the device. The permeability tensor in this case is significantly modified by the external Rashba spin-orbit coupling (RSOC) that exists on the InSb slab. The amplitude of transmission of an incident beam through the slab, marked as an angled wavy blue line in the middle figure (b), is therefore linked to the strength of the RSOC. The RSOC in (b) is identified by its characteristic spin-momentum locking, where the tangential green lines indicate the spin-polarization vectors. The right figure (c) is a possible realization of a gyrotropic and non-reciprocal optical device. It is fitted with a metal gate that allows a dynamic tuning of RSOC, leading to the necessary modulation of the light beam. We elucidate here, via demonstration of such optical control, on an indirect but robust connection between the electron spin and diverse photonic applications. (cid:48) generality, we let the magnetic field vector point along the z -axis and superimpose a small and identically directed ac-field, H exp (iωt). The ac-field imparts a frequency dependence to the structure of the µ tensor matrix. Analogously, the M vector is also assumed to point along the z -axis in addition to an induced ac-component, M exp (iωt). Inserting the complete expressions for the magnetization and magnetic field in Eq. 1, the tensor components assume the form30 (cid:48) where the individual entries are defined as µxx = 1 + κxy = − (ω0 + iαω) ωm (ω0 + iαω)2 − ω2 . ωωm (ω0 + iαω)2 − ω2 ,  µxx −iκxy iκxy µxx 0 0 µ =  , 0 0 µzz in absence of M, the intrinsic magnetization vector. Additionally, it is a Hermitian tensor, since µik = µ∗ ki. The next comment pertains to the matrix dependence on the electron g-factor via the gyromagnetic ratio (γ), a number that is manifestly material-driven; as a case in point, it is determined to be -0.44 for GaAs31 conduction electrons while reaching ≈ 50 in 2D InSb.32 Notice that the free-electron value of g = 2.0023 does not apply for a crystal. The g-factor of an electron bound to a lattice, inter alia, is primarily governed by the intrinsic spin-orbit coupling (soc) and therefore must be computed for each nanosystem including the appropriate quantization effects, which are reflected via the dispersion (electronic) relations through altered (from bulk values) band gaps and effective masses. We will expound on this point in greater detail in the following sub-section and present a path that ties soc-effects and their influence on the overall non-reciprocal behaviour. A. Determination of the g-factor (3) (4) Finally, µzz = 1 + M/H, ωm = γµ0M , and ω0 = γµ0H. This completes the form of the tensor matrix for a gyromagnetic material. A set of remarks is in order here: Firstly, the structure of the µ matrix in Eq. 4, whose off-diagonal elements vanish (the medium therefore turns isotropic, assuming no gyroelectricity is present) We remarked above about the functional relationship between the structure of the µ tensor and crystal soc. In what follows, we make explicit use of band dispersion to formalize this connection. We consider an InSb slab which crystallizes under zinc blende symmetry and displays a substantial RSOC. A minimal Hamiltonian xyε2 μ2 Metal electrodeHzInsulatorε1 μ1kxky(a)(c)(b)InSbrepresenting the Γ6 conduction bands under RSOC is expressed as H0 = p2 2m∗ + λR (σxky − σykx) , (5) 4 √ kx + iky = 2/lBa† and k− = kx − iky = where λR > 0 is the Rashba coupling parameter. The effective mass in Eq. 5 is m∗. In presence of a z -directed magnetic field, carrying out the usual Peierl's transformation, the momentum terms are re-written as : k → k − eA (t), where A is expressed by a Landau gauge of the form (0, Bzx, 0). The momentum terms in Eq. 5, following this change, can be expressed via creation (cid:0)a†(cid:1) and annihilation (a) operators, k+ = (cid:19) (cid:18) Here, lB = (cid:112)/eBz, the magnetic length.  2 these Hamiltonian (Eq. 5) in matrix form is while k2 is now 0.5 (k+k− + k−k+) = transformed momentum representations,  . Inserting the a†a + 2/lBa, (cid:19) (cid:19) (cid:18) (cid:18) 2 l2 B 2 lB √ (6) 1 2 1 2 i . m∗l2 −i B λRa a†a + a†a + 2 lB λRa 2 m∗l2 B 1 2 Φ2 , (7) ΨLL n (x) (cid:19) Φn (x) exp (ikyy) n (x, y) = x -axis, Φn = n−1 (x) + Φ2 (cid:19)1/4 (cid:113) (cid:18) eB where the harmonic oscillator eigen function along the The diagonal elements in Eq. 6 represent a harmonic oscillator. To solve for eigen states, we let the wave function be of the form (assuming translational invariance along the y-axis) (cid:18)Φn−1 (x) (cid:16)−x (cid:17) (cid:0)x − kyl2 dxn exp(cid:0)−x2(cid:1). expression: Hn (x) = (−1)n exp(cid:0)x2(cid:1) dn (cid:114) 2neB eB  . (8) (cid:19) Employing the standard raising and lowering operator relations, a†Φn = nΦn−1, the Hamiltonian in Eq. 6 transforms to Hermite polynomials, Hn (x), have the usual analytic (cid:19) (cid:114) 2neB is the short-hand notation for n + 1Φn+1 and aΦn = √ 1 2n/2 n − 1 2 (cid:48)(cid:17) (cid:48)2/2 (cid:18) (cid:18) . The and x H0 = (cid:16) + ∆ Hn x π (cid:1) exp n! B lB iλR √ √ (cid:48) m∗ −iλR eB m∗ n + − ∆ 1 2 1 2 g0µBB, accounts for The additional term, ∆ = the Zeeman-splitting of spin-states in a z -axis pointed magnetic field. Note that we set g0 = 2.0 and µB is the standard Bohr magneton. It is now straightforward to diagonalize Eq. 8 to obtain eigen states for the nth FIG. 2. The Landau dispersion for the conduction electrons of a 15.0 nm InSb slab for several values of an external z -axis directed magnetic field is shown here. The left figure (a) was prepared by diagonalizing the Hamiltonian (Eq. 8); the desired InSb band parameters such as the effective mass and the fundamental band gap were obtained from a 8 x 8 k.p Hamiltonian adapted for slab-like structures. A note about the band structure calculations and their numerical implementation can be found in the Appendix and Ref. 17. The upper (lower) set of curves in red (blue) denote the dispersion of the spin-up (down) conduction electrons. The figure on the right (b) is the effective g-factor of the conduction electrons computed directly from the Landau dispersion curves. They are shown for two values of the Rashba parameter, a dynamically tunable quantity, an attribute which we harness to describe the coupling between electron spin and optical non-reciprocity in this paper. quantum level; it is simply (cid:115)(cid:18) ∆ − eB 2m∗ (cid:19)2 + RB 2neλ2 . (9) En = eB m∗ n ± The upper (lower) sign is for the spin-up (down) electron. The effective g-factor that an electron experiences can then be approximated as gef f = . (10) E1 − E−1 2µBB Notice that we limit our analysis to n = 1 Landau level for the computation of the effective g-factor. In Fig. 2, the Landau levels (up to n = 8) is shown; in addition, the difference in energies between the spin-up and spin-down states for the n = 1 level is marked on the plot - the precise quantity desired in Eq. 10 to ascertain the g-factor. As a way of elucidation, an additional comment must be included here: The g-factor, evidently a function of the Rashba parameter, influences the µ tensor (Eq. 4) and the concomitant magnetic anisotropy linked optical phenomena. In particular, supplementary degrees-of-freedom in optical manipulation can manifest through alterations made to the strength of the Rashba coupling coefficient, which is λR = λ0(cid:104) E (z)(cid:105). Here, (cid:104) E (z)(cid:105) serves as the average electric field. The material-dependent λ0 is given as33 λ0 = 2 2m∗ ∆so Eg 2Eg + ∆so (Eg + ∆so) (3Eg + 2∆so) . (11) In Eq. 11, the fundamental band gap is Eg and ∆so denotes the intrinsic spin-orbit coupling. It is therefore easy to see how a tuning of the essential dispersion parameters - principally, the band gap and electron effective mass - can adjust λR and thereby the electric and magnetic response of the system. Elucidating further, the electromagnetic response forms the solution to Maxwell's equations that are reliant on the electric permittivity and magnetic permeability of the medium, of which the latter in our case can be transformed via the RSOC-assisted g-factor. The set of plots (Fig. 2b) reinforces this reasoning. Before we proceed to discuss magneto-optical setups harnessing the embedded utility of the g-factor, an explanatory set of statements must be added to dispel any ambiguity: The g-factor is typically a tensor quantity and direction-dependent; however, for the case shown here, we assumed the electrons are located at the base of the conduction band which is spherically symmetric (Γ6) allowing a single number to fully represent this inherently tensor quantity. For methods that carry greater rigor and include contributions from higher-energy bands, see for example, Refs. 34 and 35, a more accurate modeling of the g-factor is possible. The Appendix contains a brief note on this point. Lastly, observe that Landau levels derived from a pure parabolic model (λR = 0) ensures the g-factor is independent of the magnetic field - the dependence here otherwise (Fig. 2) is simply an outcome of including a linear Rashba spin-orbit Hamiltonian. III. Magneto-optical phenomena A wide variety of functionalities can be accomplished through the inclusion of non-reciprocal photonic devices; however, as we pointed in the opening paragraphs, geometric considerations hinder integration into silicon photonic systems necessitating the need for planar and dimensionally shrunken devices. While magnetic oxide films have been put forward as suitable material systems in this regard, here we seek to explore a class of strongly spin-orbit coupled and narrow band gap zinc-blende materials with embedded magnetic (cf. Fig. 1). The usefulness of a magneto-optical material is typically gauged by a figure-of-merit (ξ) defined as36 Faraday degree of rotation per dB absorption; more concisely, ξ = θF /ζ, where θF is the Faraday rotation and ζ gives the absorption coefficient (per unit length) of the material. It may therefore appear prudent to measure θF and the related Kerr rotation (θK) in the InSb-based setup taken up in this work. The Kerr and Faraday rotation are sketched in Fig. 3. A numerical calculation of θF and θK can be carried out by examining the Fresnel coefficients. In matrix form, for Kerr rotation, we have37 impurities (cid:19) (cid:18)Ep r Es r (cid:18)rpp rps (cid:19)(cid:18)Ep (cid:19) = rsp rss i Es i . (12) Here, rss, rsp, rps, and rpp are the Fresnel coefficients and the superscript s(p) stands for s(p)-polarized incident (i ) and reflected (r ) electric field. A similar equation 5 FIG. 3. The twin optical phenomena of Kerr and Faraday rotation is shown here. The solid lines contained within the ellipses represent the polarization axes which suffer rotation (drawn separately with respective angles marked as θK and θF ) as an incident light beam on the InSb slab is partly reflected and transmitted. Note that this configuration describes the polar magneto-optical Kerr effect (PMOKE) where the magnetization (M) is oriented normal to the plane. can be written connecting the incident and transmitted components of the electric field by introducing another set of Fresnel coefficients, which are, tss, tsp, tps, tpp. Note that in this nomenclature, the off-diagonal coefficients (rsp, rps, tsp, tps) point to the inter-mixing of the s- and p-components. We can numerically ascertain the reflection and transmission behaviour for a completely generalized case of a planar stratified and bianisotropic media that follows the constitutive relations38 D = εε0E + ξ 1 c H, B = ζ 1 c E + µµ0H. (13) For our case, we set the magneto-electric coupling tensors, ξ and ζ, to zero while ε and µ are the dimensionless permittivity and permeability tensors. The permeability tensor has non-zero off-diagonal components. The incident, reflected and transmitted fields are then obtained by matching tangential components at the interface, which here straddles the vacuum and the InSb slab. The electric fields must therefore be computed, which we do by first writing the complete wave vector (cid:0)k = (k(cid:107),±kz (cid:1) expression for the respective conserved parallel (cid:0)k(cid:107)(cid:1) and perpendicular reflected and incident plane waves consisting of their (±kz) components. The '+' and '-' signs indicate waves propagating away and toward the interface respectively. A simple application of Maxwell's equations gives the 0 = (ω/c)2, where is real while kz can assume both real dispersion relation k2(cid:107) + k2 k(cid:107) = k(cid:107) (cid:1) and complex (cid:0)k(cid:107) > k0 (cid:1) values. Note that (cid:0)k(cid:107) < k0 z = k2 k(cid:107) = (k(cid:107) cos φ, k(cid:107) sin φ) where φ is the angle subtended (cid:104) (cid:105)T (cid:113) µ0 ε0 substitute the ansatz by k(cid:107) with x-axis. With this notation in mind, we ei(k(cid:107)·R+kzz−iωt) in Maxwell's equations (Eq. 13) to construct the following dimensionless dispersion relation inside the material E, H det(M + Mk) = 0, for M = . (14) The matrix, Mk, is defined by the auxiliary relation (cid:34) (cid:35) ε ζ ξ µ (cid:35) (cid:34)  Mk = k = 0 k/k0 0 −k/k0 0 kz , −kz 0 −k(cid:107) sin φ k(cid:107) cos φ  . k(cid:107) sin φ −k(cid:107) cos φ 0 (15) The 6 × 6 material tensor M expresses the constitutive relations and Mk encapsulates the result of the curl operator on the plane waves. For a completely generalized anisotropic system, we obtain kz numerically by setting det(M + Mk(kz)) = 0 for a given (k(cid:107), φ). The fields inside the material are linear combinations of these eigen states described by polarization vectors ej± for j = {s, p} given as  sin φ− cos φ  , ep± = 0 ±kz cos φ ±kz sin φ−k(cid:107)  . −1 k0 es± = (16) The upper (lower) sign is for a wave propagating along the +ez (−ez) direction. It is now a straightforward task to calculate the Faraday and Kerr rotation by simply noting the appropriate ratios of the Fresnel coefficients. For Faraday (F) and Kerr (K) rotation, we have39 ΘF = θF + iηF = tps tss , ΘK = θK + iηK = rps rss . (17) where θF/K is the Faraday/Kerr rotation and ηF/K stands for the ellipticity of the p-polarized wave. Note that the Fresnel coefficients can be in general complex quantities as seen from the form of Eq. 17. Moreover, θF = Re(cid:2)tan−1 (tps/tss)(cid:3) with a similar relation holding for θK, the Kerr rotation. This brief digression aside, which outlined the steps underpinning a numerical assessment of the Faraday and Kerr rotation, it is now possible to study their dependence on the g-factor that impacts the permeability tensor. We show such a calculation in Fig. 5 and elucidate further: First of all note, that both θK and θF shift with an electric field, an observation easily reconcilable by recalling that the g-factor (via the RSOC) undergoes a change leading to a quantitatively different permeability tensor (cf. It is therefore of interest that an electric (gate) field by acting upon the spin of the electrons for a given magnetic field arrangement (applied and intrinsic) serves as an Fig. 4). 6 FIG. 4. The permeability dispersions for two different values of the g-factor, where we made use of Eqn. 4 and set the external z -axis directed magnetic field to 0.8 T are shown in the above plots. The dispersion curves that use a g-factor value of 22 (25) is depicted by a dotted (solid) set of lines. Additionally, the intrinsic magnetization (parallel to the external magnetic field) and the dimensionless Gilbert damping constant were assumed to be 0.3 T and 0.04 T , respectively. The dispersion on the left (a) shows the real and imaginary components of the diagonal elements of the permeability tensor while the right figure (b) furnishes the corresponding curves for the off-diagonal entries. Note that the dispersions for both the diagonal and off-diagonal components besides displaying a functional dependence on the g-factor also peak at a resonant frequency. A switch of signs is also observed for a frequency range in both cases. ions effective control mechanism to regulate the θF -governed figure-of-merit (ζ) for magneto-optical devices. It is pertinent to mention here that the key to the adaptability of a non-reciprocal photonic device design is the ζ parameter, whose optimization until now has relied on the macroscopic alignment of the total angular momentum of magneto-optical (magneto-optic effects are principally an outcome of electronic states with different angular momentum) as a pathway to a high Faraday rotation. A typical arrangement generally brings into play a combined role for the intrinsic spin-orbit coupling of the magneto-optical material and an external magnetic field to achieve a ζ commensurate with a level desirable for applications. While in principle, a magnetic field controlled adjustment of material properties is feasible, electromagnetic compatibility and its lack thereof with the adjoining integrated circuitry (in a device environment) makes it a less propitious design guideline. The suggested procedure in this work also involves control of the spin-orbit coupling (external) for a higher Faraday rotation, but with an electric bias that significantly mitigates the severity of electromagnetic incompatibility in case of a magnetic field. IV. Spin-polarized Purcell effect and the g -factor We showed how a re-calibration of the permeability tensor via an altered g-factor offers promise of tangible dynamic control in magneto-optical measurements. The genesis of such results, which lay in a re-arrangement can also of be observed in a different the Purcell the surrounding electromagnetic field, setting - 7 FIG. 5. We numerically calculate the Kerr (a) and Faraday (b) rotation which arises from reflected and transmitted rays for two gate fields and several incoming frequencies. The incident light is assumed to make an angle of π/4 with the normal to the plane of incidence. A higher electric field (which augments the g-factor) widens the Kerr rotation angle and also pushes the peak past the one obtained for a lower bias. In addition, the Kerr angle is negative in the same frequency range for which the permeability plots dip below the zero mark (see Fig. 4). The inset in (a) quantitatively assesses the ellipticity of the reflected beam and a profile in agreement with that of the Kerr rotation. The Faraday rotation in (b) which quantifies the plane of rotation of electric field for transmitted waves exhibits a similar behavior for a higher gate bias and records a minimum at the same frequency as noted for its Kerr counterpart. Note that the Kerr and Faraday rotation and the measure of ellipticity are evaluated using the transmission formalism whose governing equations are summarized in Eq. 17 in the main text. The material system used in these calculations is a 30.0 nm wide InSb well under an external magnetic field of 0.8 T and intrinsic magnetization of 0.3 T. The Gilbert damping constant, as usual, is set to 0.04. effect (PE). This effect is characterized by alterations to the spontaneous emission lifetime of a quantum source whose dynamical properties are induced by its interaction with the environment. From an application standpoint, the PE aids in the construction of nano-scale probes and development of newer light sources, for example, lasers and LEDs. The quantitative prediction of PE, therefore, especially where emission-controlled design parameters are of importance. A traditional approach to securing an optimal PE draws upon the geometry and optical attributes of the medium surrounding the emitter, notably, the electromagnetic local density-of-states (LDOS), determined in part, by the constitutive parameters, and µ. Here, to exemplify the role of the g-factor in amendments to the PE, we consider a dipole placed close to the InSb slab and numerically compute the emitter (dipole) decay rate. Nominally, for a dipole moment p located at a distance z0 above the first interface, the PE can be written as26 (The frequency and speed of light in vacuum are ω and c, respectively.) P = 1 + 6π0 Imp∗Gscat(z0)p ω3c−3p2 (cid:48) , (18a) where Gscat(z0) is the scattered dyadic Green's function of the dipole near the InSb slab that starts at z = 0 and extends below. We write it as (cid:90) d2k(cid:107) (cid:20) (2π)2 i 2kz Gscat(z0) = (cid:122) scattered/reflected part gref (cid:125)(cid:124) (cid:125) (cid:124) (cid:123)(cid:122) (cid:124) (cid:123)(cid:122) ei2kzz0 [(rsses+ + rpsep+)eT s− + (rspes+ + rppep+)eT p− reflection of es− wave reflection of ep− wave (cid:21) . (18b) (cid:123) (cid:125) ] rate of the dipole(cid:0)d1 = 1/ 2 [x + iy](cid:1) in vicinity of the √ A plot of the Purcell factor (Fp) that features the decay InSb slab (which serves as a model two-dimensional array of scattering centres) normalized to its value in free space is presented in Fig. 6. Clearly, as the g-factor is increased, changing the localized electromagnetic setting through the µ tensor, a stronger field-dipole interaction is revealed as a concomitant rise in the Purcell factor. Further, we carried out the same calculation for a second orientation of the dipole, (cid:0)d2 = 1/ 2 [x − iy](cid:1), √ that yielded no definitive gain for the Fp. A marginal rise in the decay rate (or equivalently the Fp) for both 8 values of the g-factor points to no significant modification of the localized electric field in presence of the d2 dipole placed above the InSb slab. We make a comment on the connection of the Purcell effect to the non-reciprocity of the optical medium. Firstly, notice that the scattering matrix in the √ Purcell formulation identified through the dyadic Green's function (Eq. 18b), say for the dipole d1 = 1/ 2 [x + iy], is related to dipole d2 = d∗ 1 through the simple relation T Gscat (z0, d1) = Gscat (z0, d2) = G scat (z0, d1) . (19) The above relation, however, is untrue in a non-reciprocal medium such that the Purcell factors for dipoles d1 and d2 are unequal. Furthermore, since the two dipoles are distinguished through the spins of their emitted light (see Fig. 6a and accompanying caption), and display contrasting behaviour, it is conceivable to view this as an instance of photonic spin tied to non-reciprocity. V. Final Remarks We explored the prospects of magneto-optical devices that epitomize the phenomenon of non-reciprocity and showed a newer class of design guidelines can be laid down wherein the electron's spin degree-of-freedom is the primary determinant through the inclusion of the external Rashba spin-orbit-coupling (RSOC) assisted g-factor. A set of further advancements can be planned in which the usually weaker Dresselhaus spin-orbit-coupling may actively influence the g-factor in tandem40,41 with RSOC, and therefore requires an examination of a large variety of material systems using ab-initio techniques. In addition, pursuant to the former objective of suitable candidate materials, a more systematic study of the current setup will aid us to quantitatively correlate (via first-principles simulations) various sample slabs of InSb with strain, magnetized-dopants, defects, and vacancies to magneto-optical phenomena discussed here. Here, we may note that perovskites and its thin film derivatives which are strongly magnetoelectric42,43 and can carry a robust RSOC is an encouraging alternative to foresee as a starting point for further expanding the design space of magneto-optical structures (and upgrade the FoM (ζ) parameter) through a conjoined action of the principles of multi-ferroics and electron spin-orbit coupling. The theme of non-reciprocity allied to photon spin was carried over to Purcell factor calculations, where we established using the theory of dyadic Green's function, the decay rate of a dipole held close to an InSb slab. This framework also allows us to assess situations with a randomized configuration of electromagnetic scatters or plasmonic nano-antennas replacing the InSb slab, essentially building a general theory of decay rates in a Purcell factor calculation of emitters (dipoles) near a 2D array of scattering centres. A more comprehensive set of results that suggests structures and emitter orientations maximizing the Purcell effect is planned for a future publication. 1 A. Eroglu, Wave propagation and radiation in gyrotropic and anisotropic media (Springer Science & Business Media, 2010). 2 A. Kamal, J. Clarke, and M. Devoret, Nature Physics 7, 311 (2011). 3 L. Bi, J. Hu, P. Jiang, D. H. Kim, G. F. Dionne, L. C. Kimerling, and C. Ross, Nature Photonics 5, 758 (2011). 4 M. V. Berry, Journal of Optics A: Pure and Applied Optics 11, 094001 (2009). 16 C. Caloz, A. Al`u, S. Tretyakov, D. Sounas, K. Achouri, and Z.-L. Deck-L´eger, Physical Review Applied 10, 047001 (2018). 17 P. Sengupta, H. Ryu, S. Lee, Y. Tan, and G. Klimeck, Journal of Computational Electronics 15, 115 (2016). 18 R. f. Van Welzenis and B. Ridley, Solid-state electronics 27, 113 (1984). 19 K. Chen, P. Santhanam, S. Sandhu, L. Zhu, and S. Fan, Physical Review B 91, 134301 (2015). 5 J. A. Kong, New York, Wiley-Interscience, 1975. 348 p. 20 E. Kamenetskii, IEEE Transactions on Antennas and (1975). 6 S. Pendharker, F. Kalhor, T. Van Mechelen, S. Jahani, N. Nazemifard, T. Thundat, and Z. Jacob, Optics express 26, 23898 (2018). 7 S. M. Barnett, New Journal of Physics 16, 093008 (2014). 8 S. Horsley, Physical Review A 98, 043837 (2018). 9 T. Van Mechelen and Z. Jacob, Optical Materials Express 9, 95 (2019). 10 T. Van Mechelen and Z. Jacob, Nanophotonics (2019). 11 L. Zhu, Y. Guo, and S. Fan, Physical Review B 97, 094302 (2018). 12 M. G. Silveirinha, Physical Review B 95, 115103 (2017). 13 C. Khandekar preprint Jacob, arXiv and Z. arXiv:1905.02745 (2019). 14 T. Van Mechelen and Z. Jacob, Optica 3, 118 (2016). 15 K. Y. Bliokh, A. Y. Bekshaev, and F. Nori, Nature communications 5, 3300 (2014). Propagation 49, 361 (2001). 21 F. Lima, T. Dumelow, E. Albuquerque, and J. Da Costa, JOSA B 28, 306 (2011). 22 Y. Tang and A. E. Cohen, Physical review letters 104, 163901 (2010). 23 I. Zuti´c, J. Fabian, and S. D. Sarma, Reviews of modern physics 76, 323 (2004). 24 A. Manchon, H. C. Koo, J. Nitta, S. Frolov, and R. Duine, Nature materials 14, 871 (2015). 25 P. N. Argyres, Physical Review 97, 334 (1955). 26 L. Novotny and B. Hecht, Principles of nano-optics (Cambridge university press, 2012). 27 F. Khosravi, C. L. Cortes, and Z. Jacob, Optics express 27, 15846 (2019). 28 M. Lakshmanan, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 369, 1280 (2011). 9 FIG. 6. The numerically determined Purcell factor (Fp) for two sets of circularly-polarized dipoles of opposite handedness 2 [x ± iy](cid:1) placed at a certain distance from a 30.0 nm InSb slab with (assumed) intrinsic magnetization is shown (cid:0)d1,2 = 1/ √ here. We select two values of the g-factor for this calculation, where the lower (higher) number corresponds to an electric field of 8 (5) × 106 V/m. For the case of d1 dipole, a large enhancement in the Purcell factor is observed, which correlates to an increase in the electromagnetic density-of-states arising from a constructive interference of electric field in the vicinity of the dipole, d1. The placement of the second dipole (d2) however, in contrast, leads to no significant uptick in the electric field and the Fp remains close to unity. The Fp, also in case of the d1 dipole reaches a higher value for a g-factor pushed upwards through a stronger gate bias. For smaller distances (z0) from the slab, a more intense electric field operates that gives rise to a more robust Fp uptick; this trend falls off for larger z0 values in agreement with the usual inverse square law for electric fields. Note that the arrow curving around the dipole d1 in sub-figure (a) represents the emission of a right circularly-polarized light; for d2, the sense of polarization of emitted light is the exact opposite. 29 Y. Tserkovnyak, A. Brataas, and G. E. Bauer, Physical 197603 (2011). review letters 88, 117601 (2002). 30 L. D. Landau, J. Bell, M. Kearsley, L. Pitaevskii, E. Lifshitz, and J. Sykes, Electrodynamics of continuous media, vol. 8 (elsevier, 2013). 31 J. Hubner, S. Dohrmann, D. Hagele, and M. Oestreich, Physical Review B 79, 193307 (2009). 32 B. Nedniyom, R. Nicholas, M. Emeny, L. Buckle, A. Gilbertson, P. D. Buckle, and T. Ashley, Physical Review B 80, 125328 (2009). 33 R. Winkler, Spin-Orbit Coupling in Two-Dimensional Electron and Hole Systems, vol. 41 (Springer, 2003). 34 C. Hermann and C. Weisbuch, Physical Review B 15, 823 (1977). 35 C. E. Pryor and M.-E. Pistol, Journal of Applied Physics 118, 225702 (2015). 36 S. Jacobs, K. Teegarden, and R. Ahrenkiel, Applied optics 13, 2313 (1974). 37 G. Sz´echenyi, M. Vigh, A. Korm´anyos, and J. Cserti, Journal of Physics: Condensed Matter 28, 375802 (2016). 38 A. Ishimaru, S.-W. Lee, Y. Kuga, and V. Jandhyala, IEEE Transactions on Antennas and Propagation 51, 2550 (2003). 39 H. Da, Q. Bao, R. Sanaei, J. Teng, K. P. Loh, F. J. Garcia-Vidal, and C.-W. Qiu, Physical Review B 88, 205405 (2013). 40 D. V. Bulaev and D. Loss, Physical Review B 71, 205324 (2005). 41 L. Meier, G. Salis, I. Shorubalko, E. Gini, S. Schon, and K. Ensslin, Nature Physics 3, 650 (2007). 42 L. Yan, Y. Yang, Z. Wang, Z. Xing, J. Li, and D. Viehland, Journal of Materials Science 44, 5080 (2009). 43 E. Bousquet and N. Spaldin, Physical review letters 107, Appendix A Band structure calculations H (cid:19) , ∂ ∂z (cid:18) kx, ky,−i We include material that were left out of the main text and brief explanatory notes that clarify and expand on the discussion presented in the paper. The 8-band k.p band structure calculations are performed by discretizing the InSb slab (modeled as a quantum well) on a cubic grid. The quantum well is assumed to be grown along the [001]-axis. The quantized direction is aligned to [001] which is also the z -axis. The InSb slab Hamiltonian, is of size 8Nz × 8Nz, where Nz represents the number of discretized points along the z -axis. The finite-difference discretization scheme for the 8-band k.p Hamiltonian has been explained fully in Ref. 10 of the manuscript. The k.p parameters for this work were obtained from I. Vurgaftman et al., Journal of Applied Physics, 89, 5815 (2001). The parameters are also collected in Table I for easy reference. The conduction band profile of a 6.0 nm InSb quantum well which is spin-split by the Rashba coupling is shown in Fig. 7. In preparing Fig. 7, the effective mass (cf. Eq. 5) of the conduction electrons were obtained from the eight-band k.p-calculation. A direct approach to ascertain the g-factor (gf in Eq. A1) using k.p theory is from the following result gf = g0 1 − Ep 3 1 E6c − E8v − 1 E6c − E7v . (A1) (cid:20) (cid:18) (cid:19)(cid:21) 10 FIG. 7. The Rashba spin-orbit coupling (RSOC) leads to two non-degenerate Fermi concentric energy contours for the spin-up and spin-down ensemble (a). The right figure (b) shows the band structure of conduction electrons of a 6.0 nm InSb quantum well obtained from a k.p calculation. The two winged-profiles in the right figure (b) denote the energy contours for the spin-up (higher energy) and spin-down electrons. Notice that InSb is an ideal candidate material to observe RSOC as it satisfies the twin criteria of a large intrinsic spin-orbit-coupling (0.78 eV) and a small band gap (0.43 eV at Brillouin zone centre). In the present case, the Rashba coupling parameter was artificially enhanced to 4.0 eVA for a more vivid portrayal of the spin-splitting. In Eq. A1, g0 ≈ 2 is the free electron g-factor while the subscripts 6c, 7v, and 8v designate the symmetries of the bottom (top) of the conduction (valence) bands in a crystal with Td symmetry. All remote contributions from higher-order bands have been ignored. Note that E6c − E8v is the fundamental band gap (Eg) and E6c − E7v = Eg + ∆so. Here, ∆so is the splitting from the intrinsic spin-orbit coupling. While in principle, it is possible to derive a similar expression with Rashba coupling term that explicitly accounts for Eg, ∆so, and the effective mass, the approximate estimation procedure outlined in Section II A indirectly includes the foregoing quantities through the Rashba parameter (cf. Eq. 11). Finally, in context of the eight-band k.p Hamiltonian based g-factor calculations, it is relevant to mention here that the use of only the lowest conduction band is a reasonable approximation for InSb; the next p-like conduction band (Γ7) is much above the fundamental direct band gap. A more accurate model, however, must include the Γ7 and Γ8 conduction bands, for instance, in GaAs, suggesting a 14-band k.p-calculation as our starting point. The g-factor formula (Eq. A1 must reflect this modification through terms of the form.34 TABLE I. 8-band k.p parameters for InSb. Ev, Eg, Ep, and Vso are in units of eV. The remaining Luttinger parameters are dimensionless constants and the effective mass is in units of the free electron mass. Material Ev γ1 γ3 m∗ Eg Ep Vso 0.28 34.8 15.5 16.5 0.0135 0.235 18 0.81 γ2 InSb

Subsets and Splits

No community queries yet

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