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
|
|---|---|---|---|---|---|---|---|---|---|
1806.03643 | 1 | 1806 | 2018-06-10T12:16:12 | Hybrid plasmonic nanostructures based on controlled integration of MoS2 flakes on metallic nanoholes | [
"physics.app-ph"
] | Here, we propose an easy and robust strategy for the versatile preparation of hybrid plasmonic nanopores by means of controlled deposition of single flakes of MoS2 directly on top of metallic holes. The device is realized on silicon nitride commercial membranes and can be further refined by TEM or FIB milling to achieve the passing of molecules or nanometric particles through a pore. Importantly, we show that the plasmonic enhancement provided by the nanohole is strongly accumulated in the 2D nanopore, thus representing an ideal system for single-molecule sensing and sequencing in a flow-through configuration. Here, we also demonstrate that the prepared 2D material can be decorated with metallic nanoparticles that can couple their resonance with the nanopore resonance to further enhance the electromagnetic field confinement at the nanoscale level. This method can be applied to any gold nanopore with a high level of reproducibility and parallelization; hence, it can pave the way to the next generation of solid-state nanopores with plasmonic functionalities. Moreover, the controlled/ordered integration of 2D materials on plasmonic nanostructures opens a pathway towards new investigation of the following: enhanced light emission; strong coupling from plasmonic hybrid structures; hot electron generation; and sensors in general based on 2D materials. Nanopore | physics.app-ph | physics | Hybrid plasmonic nanostructures based on controlled integration of
MoS2 flakes on metallic nanoholes
Denis Garoli1*, Dario Mosconi2, Ermanno Miele1, Nicolò Maccaferri1, Matteo Ardini1, Giorgia
Giovannini1, Michele Dipalo1, Stefano Agnoli2 and Francesco De Angelis1*
1 Istituto Italiano di Tecnologia, via Morego 30, I-16163, Genova, Italy.
2 Dipartimento di Chimica, Università degli Studi di Padova, Via Marzolo 1, 35131, Padova, Italy.
* Corresponding author: Dr. Denis Garoli, [email protected]; Dr. Francesco De Angelis,
[email protected]
Keywords: Self-organized formation, MoS2, nanopore, plasmonic nanopore, optical antennae, Raman
enhancement, SERS
Abstract
Here, we propose an easy and robust strategy for the versatile preparation of hybrid plasmonic nanopores
by means of controlled deposition of single flakes of MoS2 directly on top of metallic holes. The device
is realized on silicon nitride commercial membranes and can be further refined by TEM or FIB milling
to achieve the passing of molecules or nanometric particles through a pore. Importantly, we show that
the plasmonic enhancement provided by the nanohole is strongly accumulated in the 2D nanopore, thus
representing an ideal system for single-molecule sensing and sequencing in a flow-through configuration.
Here, we also demonstrate that the prepared 2D material can be decorated with metallic nanoparticles
that can couple their resonance with the nanopore resonance to further enhance the electromagnetic field
confinement at the nanoscale level. This method can be applied to any gold nanopore with a high level
of reproducibility and parallelization; hence, it can pave the way to the next generation of solid-state
nanopores with plasmonic functionalities. Moreover, the controlled/ordered integration of 2D materials
on plasmonic nanostructures opens a pathway towards new investigation of the following: enhanced light
emission; strong coupling from plasmonic hybrid structures; hot electron generation; and sensors in
general based on 2D materials.
Nanopore technology is the core of third-generation sequencing, and solid-state nanopores are
now one of the main topics of research in single-molecule sensing. To produce solid-state nanopores, a
valid alternative to the currently used commercial biological nanopores, as well as the current
nanofabrication methods and materials, must be developed. To date, one of the most advanced
generations of solid-state nanopores is represented by 2D materials. The atomically thin nature of
graphene and other materials, such as transition metal dichalcogenides (TMDCs) (MoS2, WS2, etc.)1
make them ideal translocation membranes for high-resolution, high-throughput, single-molecule
sequencing based on nanopores2−12. Electrical measurements are the main approach for single-molecule
sequencing by means of nanopores, but readout schemes that rely on optical spectroscopy can be
envisioned13-15. Within this context, a plasmonic nanopore15 represents an intriguing tool for enhancing
the signal-to-noise from the optical signal via the electromagnetic field enhancement that can be
generated by engineered metallic nanostructures. The integration of a 2D material with plasmonic
nanostructures leads to a new generation of hybrid nanopores16. In fact, the development of a hybrid 2D-
plasmonic nanoarchitecture that efficiently combines the benefits from a plasmonic field enhancement
with the intrinsic in-plane electric field localization from atomically thin materials would represent a step
towards the next generation of hybrid nanopores.
The preparation of a nanopore in an atomically thick layer of 2D material remains a challenging
task and requires the deposition of single layers on top of a larger solid-state pore/membrane with a
successive step involving a high-resolution electron beam sculpting/drilling process9 that often suffers
from process variability, precluding the platform from being scalable. Moreover, to prepare a hybrid
plasmonic / 2D material pore, the integration of metallic nanostructures must be achieved in close
proximity to the 2D material pore. The pioneering work of Nam et al.16 on a hybrid nanopore involved
the use of photothermal sculpting to create a nanopore in a graphene membrane with a nearby gold
nanoparticle acting as an optical antenna. In their paper, a significant enhancement in fluorescence was
detected during single-molecule DNA translocation though the nanopore, highlighting the potential of
hybrid systems. Here, we propose an easy and robust strategy for the versatile preparation of hybrid
plasmonic nanostructures by means of controlled deposition of single- or few-layer flakes of MoS2
directly on top of metallic holes. This method can be applied to any gold nanohole (with 2D or 3D
geometries, as will be demonstrated) and can pave the way to the next-generation fabrication of hybrid
systems. Compared to the more frequently used graphene, MoS2 presents several advantages in nanopore
applications and has been recently proposed for use in single-molecule sequencing4,5,11. The presence of
local defects (-S vacancies) on the MoS2 layer is used here to anchor the material locally on metallic
nanoholes by means of chemical functionalization via dithiol molecules. Thiol conjugation of MoS2 has
been explored in a few recent works17-19; to the best of our knowledge, this work represents the first
example of a deposition method based on that process. A pictorial representation of the process is
illustrated in Fig. 1 (Top Panel). The preparation of MoS2 flakes is based on chemical exfoliation20-22, as
described in the SI; Figures 1(b)-(d) show examples of the ordered deposition of single flakes on top of
2D and 3D gold holes. The method used for the deposition is based on the conjugation between a gold
(or another noble metal) surface and a dithiol-terminated organic chain as well as the same conjugation
between the MoS2 flake and the dangling –SH group of the same molecule (Fig. 1). In particular, to
perform a controlled deposition of MoS2 over metallic holes, we used a 1,12-dodecanedithiol molecule
as a linker between the gold surface and local –S vacancies in MoS2 flakes. The protocol of
functionalization is the following (illustrated in Fig. 1-Top Panel): 1) a 1 mM solution of dithiol is
prepared in EtOH; 2) the sample to be deposited is first cleaned in oxygen plasma for 60 seconds to
facilitate the process; 3) the plasmonic holes are prepared on a Si3N4 membrane, and only one side of the
substrate is covered with metal (see the SI for details on the nanohole fabrication process); 4) as we
expect the thiol deposition to occur only on the metal in contact with the solution, to functionalize only
the holes, we put the sample with the metallic side in contact with a MoS2 suspension in EtOH, i.e., we
keep the sample floating to avoid the complete wetting of the sample; 5) at the same time, we spot on the
dry side, opposite to the one we want to decorate with MoS2, 10 L of 1,12-dodecanedithiol diluted in
EtOH; 6) after a few seconds, the drop of dithiol starts to dry; and 7) the sample is rinsed in EtOH, and
the controlled deposition is achieved. This method can be used on every nanostructure involving a
metallic nanohole; in our case, we demonstrate that the link between the two materials can be achieved
with high reproducibility, both in a flat metallic hole and in 3D hollowed antennas. To control the % of
coverage of the holes and the quantity of deposited flakes in terms of number of layers, the most critical
parameter is the time of incubation. In our case, 20 seconds of incubation leads to a high percentage of
single-layer flakes deposited over large nanohole array. In fact, as will be reported later, over 80 % of
metallic holes present in the array can be covered with MoS2. The optimization of the preparation of a
MoS2 batch (see the SI), allows one to obtain a single-layer deposition almost over the different pores.
This deposition can be demonstrated by Raman measurements where the discrimination among single-
or few-layer flakes is possible23. As illustrated in Fig. 1(a), (b) and (c), the deposition over the 2D
nanopores results in small flakes covering the pores with dimensions spanning from 200 nm up to 500
nm. In the case of the 3D structures (Fig. 1(d), (e) and (f)), the flake deposition results in a partially
covering layer that crinkles around the metal in many different ways, ranging from small flakes covering
only the top hole to large flakes wrapping the 3D body of the structure. Note that, although it is beyond
the scope of this work, the obtained structures can find several interesting applications in all the present
fields of research in which 2D materials are the core. For example, the controlled/ordered integration of
2D materials on plasmonic nanostructures can pave the way to new investigations on enhanced light
emission from TMDCs24-31, strong coupling from plasmonic hybrid structures32, hot electron generation
33-34, and sensors in general based on 2D materials35-40.
Figure 1. SEM micrographs of MoS2 flakes deposited onto an array of plasmonic nanoholes. (top panel)
Illustration of the concept for controlled deposition of MoS2 flakes over metallic holes; (a) top view over
large flat gold holes array; (b)(c) detail of a single-layer flake on a 2D pore; (d) tilted view over large
array of 3D antennas covered with MoS2 flakes; (e)(f) details of the MoS2 flakes deposited onto an
antenna.
Next, we investigate the plasmonic properties of our archetypical 3D structure integrated with
MoS2 flakes by means of finite element method (FEM) simulations using the RF Module in Comsol
Multiphysics and taking into account the geometry that can be fabricated using our method 41,42. The
phenomena that will be illustrated resemble the phenomena we expect to observe in a flat metallic pore
integrated with the 2D flake (see SI). The optical properties of MoS2 flakes can be simulated considering
the experimental optical constants obtained by Zhang et al. 43 for a single-layer MoS2 film in the spectral
range of interest. First, we consider a hollow 3D antenna with optimized dimensions (height, diameter
and hole radius) for field confinement at the top area at wavelength of 633 nm. We used this antenna
structure for the Raman spectroscopy test reported below (details on fabrication are reported in the SI).
At this wavelength, it is possible to enhance the electromagnetic field at the top of our structure by almost
two orders of magnitude. As shown in Fig. 2(b), the same significant field enhancement can also be
obtained in the case of a hollow antenna with a monolayer of MoS2 covering the hole, as in the
experiment. Note that the presence of a top MoS2 layer does not significantly change the field distribution
in the region of interest. Because we are investigating the fabrication of a nanopore into a MoS2
monolayer, a 5 nm hole has been simulated as well. As seen from Fig. 2(c), the presence of such a hole
in the high-index MoS2 layer on top of the plasmonic antenna induces a strong field confinement and
intensity enhancement (by a factor of 50) inside the nanopore. Note that this high field enhancement is
not achievable if no plasmonic element is present, i.e., if we consider a hollow antenna without the gold
coverage, we have the field confinement and a small enhancement (by a factor of 3) inside the nanopore
(see the SI), whereas with the gold, we increase this enhancement by at least one order of magnitude.
Figure 2. (a) Electric field intensity distribution of a gold nanopillar made of a 300 nm wide and 400 nm
thick dielectric (S1813 optical resist41,42) structure covered with 35 nm of gold at =633 nm. (b)
Electromagnetic field distribution with a disk with a MoS2 monolayer on top of the pillar. (c) Left-panel:
same as in (b) but with a circular nanohole 5 nm wide in the MoS2 monolayer; right-panel: electric field
intensity distribution in the MoS2 nanohole.
As is well known, MoS2 is a layered material; thus, to obtain processable flakes, it must be exfoliated by
breaking the van der Waals interactions between the layers, a process that causes high stress to the
material in all the preparation methods developed until now20-22.
Considering that each MoS2 single-layer nanosheet consists of molybdenum atoms sandwiched between
two planes of sulfur anions, it is reasonable to suppose that several defects are present on the surface
where vacancies in –S bonds are expected44. In principle, these unsaturated bonds can represent not only
a favorite site for the thiol conjugation used for the deposition but also a site of nucleation for gold
nanoparticles, hence allowing the decoration of the flakes. In addition, if we consider the higher
sputtering rate of S respect to Mo atoms, if a MoS2 layer is drilled with an ion/electron beam, then we
can expect to have extra edges where free –S links (or partially unsaturated S anions) may be present.
Moreover, these latter free bonds can be used as nucleation sites for metallic nanoparticle growth or
deposition.
Consequently, here we extend our simulations considering the case of a 5-nm gold nanoparticle (AuNP)
in close proximity to a 5-nm pore prepared in the MoS2 flake. As reported below, this case resembles
well the experimental case where the feasibility of this fabrication is demonstrated.
Figure 3 illustrates the effect as obtained from our FEM simulations. As expected, the presence
of a metallic nanoparticle on the edge of the MoS2 nanopore strongly distorts the field confinement,
leading to significant additional enhancement due to coupling between the plasmonic nanopore and the
resonating AuNP. Moreover, switching the polarization appears to possibly modulate this coupling
phenomenon and hence the final enhancement. In fact, as illustrated in Fig. 3(a) and 3(b), once the
polarization of the incident light is oriented along the AuNP and the hole, the field reaches a value of up
to 90 at the pore exit, whereas in the cross polarized configuration (Fig. 3(c) and 3(d)), this effect cannot
be observed. This result suggests a possible means to switch the system based on this effect. For example,
this switching could be interesting for applications where single molecules pass through the pore for
sensing based on enhanced spectroscopy, such as SERS or metal enhanced fluorescence (MEF)14,15, both
of which are now of great interest for sequencing applications.
Figure 3. (a) Electric field intensity distribution at the nanopore when the incident electric field
is parallel to the AuNP-nanopore axis. (b) Electric field intensity distribution at the nanopore when the
incident electric field is perpendicular to the AuNP-nanopore axis.
From the perspective of the fabrication point, as is well known, a single layer of MoS2 is
approximately 0.7 nm thick, and a nanopore can be easily prepared by means of focused electron beam
exposure (using TEM)4, 5. Here, we first verify that the deposited MoS2 flakes can be sculptured by means
of a 100 keV focused electron beam (HRTEM Fig. 4(a) and (b)). However, TEM sculpturing is a very
time-consuming and expensive procedure and can be performed only on suitable small and fragile
substrates. Alternative strategies for narrow nanopore fabrication represent important contributions to
the nanopore topic2, 45, 46. Here, we report the preparation of a sub-10-nm hole prepared in a MoS2 flake
by means of FIB milling at low current (4.4 pA) with a single-pass exposure. The ability of the FIB
microscope to perform patterning on engineered arrays allows the preparation of multiple-point
nanopores on our substrate in a rapid and reproducible manner. The illustration of the process is reported
in Fig. 4(c). TEM- and FIB-fabricated nanopores in MoS2 flakes were characterized by means of TEM
micrographs; Figure 4 reports examples of the obtained data from selected samples. As expected, while
TEM sculpturing of a nanopore down to 2 nm can be easily achieved by controlling the duration of the
exposure (Fig. 4(a) and (b)), in the case of FIB-milled holes, diameters just below 10 nm are the lower
limit of this approach (Fig. 4(d)).
As described above, the holes in the MoS2 layer are expected to result in vacancies in –S bonds
that we use here as nucleation sites for the growth of Au nanoparticles (AuNPs). For the growth of
AuNPs, a 2 mM HAuCl4 solution was prepared in H2O, with 20 L dropped over the sculptured samples
for different time durations to allow the AuNPs to grow. The dimensions of the obtained AuNPs depend
on the duration of the deposition (see the SI for examples of different growths); AuNPs of approximately
5 nm in diameter were obtained using 30 seconds of incubation at room temperature and subsequent
rinsing in H2O. Under this condition, AuNPs were grown on both bare flakes and on flakes in which a
nanohole was created. Figures 4(e) and 4(f) show TEM micrographs of MoS2 decorated with AuNPs.
The possibility of decorating a MoS2 layer with metallic nanoparticles has been previously investigated
in several recent papers47-50, and it has also been demonstrated that the MoS2 exposed edges are
preferential nucleation sites. Consequently, in our case, it is possible to achieve AuNP growth in close
proximity to the nanopore in almost all the cases; however, additional AuNPs can be present on the
flakes. Control of the number of grown particles requires additional experimental optimizations.
However, here, we are interested in a plasmonic phenomenon that we expect to observe with one or more
AuNPs in close proximity to the pore.
Figure 4. (a) TEM micrograph of a nanohole sculptured into a MoS2 layer by means of HRTEM exposure
for 1 minute; (b) TEM micrograph of a nanohole sculptured into MoS2 layer by means of HRTEM
exposure for 3 minutes; (c) illustration of the concept used for FIB milling of the MoS2 nanopores array;
(d) example of a nanopore prepared by means of FIB; (e)(f) example of AuNPs grown on a MoS2 layer(s)
after FIB milling.
Considering the simulations reported above and the proof-of-concept fabrication obtained, we
expect a significant enhancement due to the presence of both the resonating antenna and AuNPs. To
verify this enhancement, we performed Raman spectroscopy on our samples at two different wavelengths
of excitation, i.e., 532 nm and 633 nm. The latter results appear to be resonant with the structure and are
able to excite the mode at the antenna apex. The measurements were performed by using a Rainshaw
InVia Microscope Raman system with a 50 × 0.95 NA objective, collecting the signal with a spectral
resolution of 2.5 cm-1 and an integration time of 1 second. The system was calibrated by using the
intensity of the standard peak at 520 cm-1 from a silicon substrate. Figure 5 reports the results of our
measurements. In the top panel, a map over a large array of 256 3D antennas is reported. Raman shifts
(excitation wavelength 532 nm) between 400 and 410 cm-1 (in correspondence of A1g Raman mode) have
been used to evaluate the coverage of the MoS2 over the 256 points. According to the figure, the signal
appears only in correspondence of the antennas. This result is a clear demonstration that the deposition
strategy covers the desired elements in almost all the cases: over 85 % of the antennas are decorated with
MoS2 flakes. Figure 5(b) reports the statistical analysis over the 256 points with which we evaluate the
number of layers corresponding to each MoS2 flake. As illustrated in SI, the measured points were fitted
by Lorentzian functions. The in-plane (E1
2g at ~ 380 cm-1) and out-of-plane (A1g at ~ 404 cm-1) Raman
modes were always clearly visible and used in the analysis. The difference between the E1
2g and A1g
modes (∆f) is known to steadily increase with the number of layers;51-55 hence, this parameter can be a
reliable quantity to count the number of layers of MoS2. We used f to evaluate the percentage of single-
layer flakes deposited on the considered array (Fig. 5(b), histogram). From our analysis, we can conclude
that a single layer can be deposited over approximately 50 % of the antennas and nanopores. Regardless,
we think that this can be improved acting on the exfoliation procedure to obtain better-quality, single-
layer flakes in solution. Indeed, herein, we chose to follow a Li-intercalation protocol because of the
clear advantages for our purposes with respect to other techniques. The resulting 1T-phase MoS2 is well
known to be more defective and, consequently, more reactive than the pristine, semiconductive
hexagonal phase is. 17,56,57 This enhanced reactivity clearly promotes both anchoring the flake via thiol
conjugation and nucleating AuNPs on MoS2 defects. In addition, with respect to liquid-phase exfoliation,
a Li-based method can provide stable suspensions without any surfactant (that may hamper both
anchoring and plasmonic behavior) and with higher exfoliation degree,58 which, in our case, was
maximized by doubling the Li dose. Unfortunately, MoS2 strongly tends to break up during the
exfoliation, resulting in a quite large size dispersion (see SI). Even if our deposition procedure was proved
to work with all nanosheet sizes, we believe that improving the synthetic procedure to have flakes with
homogenous thickness and a controlled lateral dimension would allow further optimization of the
deposition parameters and, consequently, enhancement of the performances of these hybrid systems.
Finally, although our synthetic procedure is highly time consuming and low yielding, Li-exfoliation may
be scaled-up by switching from chemical to electrochemical intercalation.59,60 This process would allow
the preparation of single-layer MoS2 with higher throughput, which is necessary for the application of
these types of systems on a large scale.
Finally, Raman spectroscopy has also been used to demonstrate, as a proof of concept, the
resonance coupling between AuNPs grown on MoS2 pores and the integrated plasmonic 3D antennas.
Figure 5(c) reports the Raman shift collected with excitation wavelengths of 532 (see SI) and 633 nm in
the presence of optimized 3D antennas (working at 633 nm) and with the addition of AuNPs. As seen in
all the cases, both E1
2g and A1g Raman modes are observed. This observation is not surprising because,
despite the use of 1T-MoS2 flakes, it has been already demonstrated that MoS2 may undergo a 1T 2H
phase transition under laser beam.61,62
Although a detailed study of the Raman spectrum is far from the scope of this work, it is
interesting to report that, using the excitation wavelength of 532 nm, a Lorentz function perfectly fits the
experimental data and determines the two E1
2g and A1g Raman modes at 383 and 402 cm-1, respectively,
i.e., a f below 20 cm-1 equivalent to single-layer MoS2. Moreover, this observation appears to be
verified after AuNP growth, when a higher intensity in the Raman modes appears with a slightly increase
in f that we ascribe to the presence of the nanoparticles.
The presence of AuNPs induces a more significant enhancement in the collected spectra in the
case of 633 nm excitation wavelength. This result can be caused by the combination of the additional
enhancement from the plasmonic antennas and the resonant condition on excitation. In this latter case, it
has not been possible to fit all the peaks with a single Lorentz function cause, as already reported65, and
additional modes appear at approximately 410 and 450 cm-1.
Figure 5. Raman analysis on MoS2 deposited on 3D metallic antennas. (a) Map over 256 antennas.
Signal integrated between 400 and 410 cm-1: 227 antennas give the expected signal. Note that the laser
is slightly misaligned with respect to the optical image; (b) histogram reporting the statistical analysis
on the number of layers based on f of the Raman modes; (c) Raman spectra of the same nanoantenna
decorated with a MoS2 flake before and after AuNP growth. Solid lines correspond to Lorentz curve fits,
and dots correspond to experimental data.
In summary, we presented a hybrid plasmonic 2D material structure able to generate a significant field
confinement in close proximity of the nanopores. The fabrication procedure allows the preparation of
ordered structures over large array using a low-cost procedure and without the use of complex
lithographic processes. This strategy can be applied to not only MoS2 but also many 2D materials for
which the (always present) defects over the layer can be used to anchor the linker between the metallic
nanopore and the flake. Moreover, the presence of defects and of edges on the 2D materials allows the
nucleation of metallic nanoparticles, hence paving the way to integrate additional plasmonic elements
over ordered hybrid structures.
We believe that such an architecture can be a key element for the realization of new hybrid devices for
use in several applications, including photoluminescence, strong coupling and valley-polarization64
studies, and single-molecule detection for DNA or protein sequencing. With respect to previously
reported hybrid plasmonic nanostructures, our scheme significantly reduces the complexity of
fabrication, leading to a more robust and low-cost approach for the integration of 2D materials with
plasmonic nanopores.
AUTHOR INFORMATION
* Corresponding authors: Dr. Francesco De Angelis, [email protected]
AUTHOR CONTRIBUTION
DG conceived the experiment and fabricated and characterized the structures; DM and SA prepared the
MoS2 flakes; EM proposed the functionalization protocol; NM performed the FEM simulations; MA and
GG performed TEM microscopy; MDP helped with the data analysis; and FDA supervised the work.
ACKNOWLEDGMENTS
The research leading to these results has received funding from the European Research Council under
the FET-Open: PROSEQO, Grant Agreement No. [687089].
References
1.
2.
Choi, W.; Choudhary, N.; Han, G. H.; Park, J.; Akinwande, D.; Lee, Y. E. j.mattod. 2016, 10, 002.
Feng, J.; Liu, K.; Graf, M.; Lihter, M.; Bulushev, R. D.; Dumcenco, D.; Alexander, D. T. L.;
Krasnozhon, D.; Vuletic, T.; Kis, A.; Radenovic, A. Nano Lett. 2015, 15, 3431−3438.
Danda, G.; Das, P. M.; Chou, Y.-C.; Mlack, J. T.; Parkin, W. M.; Naylor, C. H.; Fujisawa, K.; Zhang, T.;
3.
Fulton, L. B.; Terrones, M.; Johnson, A. T. C.; Drndic, M. ACS Nano 2017, 11, 1937−1945.
4.
Liu, K.; Feng, J.; Kis, A.; Radenovic, A. ACS Nano 2014, 8, 2504–2511.
5.
Feng, J.; Liu, K.; Bulushev, R. D.; Khlybov, S.; Dumcenco, D.; Kis, A.; Radenovic, A. Nature
Nanotech. 2015, 10, 1070.
6.
7.
8.
Wilson, J.; Sloman, L.; He, Z.; Aksimentiev, A. Adv. Funct. Mater. 2016, 26, 4830–4838.
Heerema S. J.; Dekker, C. Nature Nanotech. 2016, 11, 127.
Bonome, E. L.; Lepore, R.; Raimondo, D.; Cecconi, F.; Tramontano, A.; Chinappi, M. J. Phys. Chem. B
2015, 119, 5815-5823.
9.
10.
11.
12.
Arjmandi-Tash, H.; Belyaeva, L. A.; Schneider, G. F. Chem. Soc. Rev. 2016, 45, 476.
Qiu, H.; Sarathy, A.; Leburton, J.-P.; Schulten, K. Nano Lett. 2015, 15, 8322−8330.
Farimani, A. B:; Min, K.; Aluru, N. R. ACS Nano 2014, 8, 7914–7922.
Smolyanitsky, A.; Yakobson, B. I.; Wassenaar, T. A.; Paulechka, E.; Kroenlein, K. ACS Nano 2016, 10,
9009−9016.
13.
Gilboa, T.; Torfstein, C.; Juhasz, M.; Grunwald, A.; Ebenstein, Y.; Weinhold, E.; Meller, A. ACS Nano
2016, 10, 8861−8870.
14. McNally, B.; Singer, A.; Yu, Z.; Sun, Y.; Weng, Z.; Meller, A. Nano Lett. 2010, 10, 2237–2244.
15.
16.
Assad, O. N.; Gilboa, T.; Spitzberg, J.; Juhasz, M.; Weinhold, E.; Meller, A. Adv. Mater. 2017, 1605442.
Nam, S.; Choi, I.; Fu, C.-C.; Kim, K.; Hong, S.; Choi, Y.; Zettl, A.; Lee, L. P. Nano Lett. 2014, 14,
5584−5589.
17.
Chou, S. S.; De, M.; Kim, J.; Byun, S.; Dykstra, C.; Yu, J.; Huang, J.; Dravid, V. P. J. Am. Chem. Soc.
2013, 135, 4584−4587.
18.
19.
Chen X.; McDonald, A. R. Adv. Mater. 2016, 28, 5738-5746.
Benson, E. E.; Zhang, H.; Schuman, S. A.; Nanayakkara, S. U.; Bronstein, N. D.; Ferrere, S.; Blackburn,
J. L.; Miller, E. M. Am. Chem. Soc. 2018, 140, 441–450.
20.
Jian Zheng, J.; Zhang, H.; Dong, S.; Liu, Y.; Nai, C. T.; Shin, H. S.; Jeong, H. Y.; Liu, B.; and Loh, K.
P. Nat. Commun. 2014, 5:2995.
21.
Voiry, D.; Salehi, M.; Silva, R.; Fujita, T.; Chen, M.; Asefa, T.; Shenoy, V. B.; Eda, G.; Chhowalla, M.
Nano Lett. 2013, 13, 6222–6227.
22.
23.
24.
Acerce, M.; Voiry, D.; Chhowalla, M. Nature Nanotech. 2015, 10, 313–318.
Lee, C.; Yan, H.; Brus, L. E.; Heinz, T. F.; Hone, J.; Ryu, S. ACS Nano 2010, 25, 2695-700.
Chen, H.; Yang, J.; Rusak, E.; Straubel, J.; Guo, R.; Myint, Y. W.; Pei, J.; Decker, M.; Staude, I.;
Rockstuhl, C.; Lu, Y.; Kivshar Y. S.; Neshev, D. Sci. Rep. 2016, 6:22296.
25.
26.
Li, Z.; Li, Y.; Han, T.; Wang, X.; Yu, Y.; Tay, B.; Liu, Z.; Fang, Z. ACS Nano 2017, 11, 1165−1171.
Huang, J.; Akselrod, G. M.; Ming, T.; Kong, J.; Mikkelsen, M. H. ACS Photonics 2018, 5, 552−558.
27.
Janisch, C.; Song, H.; Zhou, C.; Lin, Z.; Elías, A. L.; Ji, D.; Terrones, M.; Gan, Q.; Liu, Z. 2D Mater.
2016, 3, 025017.
28.
Kang, Y.; Najmaei, S.; Liu, K.; Bao, Y.; Wang, Y.; Zhu, X.; Halas, N. J.; Nordlander, P.; Ajayan, P. M.;
Lou, J.; Fang, Z. Adv. Mater. 2014, 26, 6467–6471.
29. Wang, Z.; Dong, Z.; Gu, Y.; Chang, Y.-H.; Zhang, L.; Li, L.-J.; Zhao, W.; Eda, G.; Zhang, W.; Grinblat,
G.; Maier, S. A.; Yang, J. K.W.; Qiu, C.-W.; Wee, A. T. S. Nature Com. 2016, 7:11283.
30.
Butun, S.; Tongay, S.; Aydin, K. Nano Lett. 2015, 15, 2700−2704.
31. Mukherjee, B.; Kaushik, N.; Tripathi, R. P. N.; Joseph, A. M.; Mohapatra, P. K.; Dhar, S.; Singh, B. P.;
Pavan Kumar, G. V.; Simsek, E.; Lodha, S. Sci. Reports 2017, 7:41175.
32.
Liu, W.; Lee, B.; Naylor, C. H.; Ee, H.-S.; Park, J.; Johnson, A. T. C.; Agarwal, R. Nano Lett. 2016, 16,
1262−1269.
33.
Yu, Y.; Ji, Z.; Zu, S.; Du, B.; Kang, Y.; Li, Z.; Zhou, Z.; Shi, K.; Fang, Z. Adv. Funct. Mater. 2016, 26,
6394–6401.
34.
35.
Kang, Y.; Gong, Y.; Hu, Z.; Li, Z.; Qiu, Z.; Zhu, X.; Ajayanc P. M.; Fang, Z. Nanoscale 2015, 7, 4482.
Zhao, Y.; Tang, M.; Liao, Q.; Li, Z.; Li, H.; Xi, K.; Tan, L.; Zhang, M.; Xu, D.; Chen, H.-Y. ACS
Sensors 2018, 3, 806-814.
36.
37.
Srinivasan, K.; Subramanian, K.; Muruganb, K.; Dinakaran, K. Analyst 2016, 141, 6344.
Lin, J.; Li, H.; Zhang, H.; Chen, W. Applied Physics Letters 2013, 102, 203109.
38. Miao, J.; Hu, W.; Jing, Y.; Luo, W.; Liao, L.; Pan, A.; Wu, S.; Cheng, J.; Chen, X.; Lu, W. small 2015,
11, 2392–2398.
39.
Feng, J.; Graf, M.; Liu, K.; Ovchinnikov, D.; Dumcenco, D.; Heiranian, M.; Nandigana, V.; Narayana,
Aluru, R.; Kis, A.; Radenovic, A. Nature 2016, 536, 197.
40.
41.
Zhu, C.; Zeng, Z.; Li, H.; Li, F.; Fan, C.; Zhang, H. J. Am. Chem. Soc. 2013, 135, 5998−6001.
De Angelis, F.; Malerba, M.; Patrini, M.; Miele, E.; Das, G.; Toma, A.; Zaccaria, R. P.; Di Fabrizio, E.
Nano Lett. 2013, 13, 3553–3558.
42.
43.
44.
45.
Garoli, D.; Zilio, P.; Gorodetski, Y.; Tantussi, F.; De Angelis, F. Nano Lett. 2016, 16, 6636–6643.
Zhang, H.; Ma, Y.; Wan, Y.; Rong, X.; Xie, Z.; Wang, W.; Dai, L. Sci. Rep. 2015, 5, 8440.
Addou, R.; Colombo, L.; Wallace, R. M. ACS Appl. Mater. Interfaces 2015, 7 (22), 11921–11929.
Carson, S.; Wanunu, M. Nanotechnology 2015, 26, 074004.
46. Waduge, P.; Bilgin, I.; Larkin, J.; Henley, R. Y.; Goodfellow, K.; Graham, A. C.; Bell, D. C.;
Vamivakas, N.; Kar, S.; Wanunu, M. ACS Nano 2015, 9, 7352–7359.
47.
Zhou, H.; Yu, F.; Guo, C. F.; Wang, Z.; Lan, Y.; Wang, G.; Fang, Z.; Liu, Y.; Chen, S.; Sun, L.; Ren, Z.
Nanoscale 2015, 7, 9153.
48.
Shi, Y.; Huang, J.-K.; Jin, L.; Hsu, Y.-T.; Yu, S. F.; Li, L.-J.; Yang, H. Y. Scientific Reports 2013, 3 :
1839.
49.
Su, S.; Zhang, C.; Yuwen, L.; Chao, J.; Zuo, X.; Liu, X.; Song, C.; Fan, C.; Wang, L. ACS Appl. Mater.
Interfaces 2014, 6, 18735−18741.
50.
Jiang, S.; Guo, J.; Zhang, C.; Li, C.; Wang, M.; Li, Z.; Gao, S.; Chen, P.; Sib, H.; Xuc, S. RSC Adv.,
2017, 7, 5764.
51.
52.
Lee, C.; Yan, H.; Brus, L. E.; Heinz, T. F.; Hone, J.; Ryu, S. ACS Nano 2010, 4, 2695-700.
Cooper, R. C.; Lee, C.; Marianetti, C. A.; Wei, X.; Hone, J.; Kysar, J. W. Physical Review B 2013, 87,
035423.
53. Mak. K. F.; Lee, C.; Hone, J.; Shan, J.; Heinz, T. F. Physical Review Letters 2010, 105 136805.
54.
55.
Castellanos-Gomez, A.; Agrait, N.; Rubio-Bollinger, G. Applied Physics Letters 2010, 96, 213116-3.
Splendiani, A.; Sun, L.; Zhang, Y.; Li, T.; Kim, J.; Chim, C. Y.; Galli, G.; Wang, F. Nano letters 2010,
10. 1271-5.
56.
Voiry, D.; Goswani, A.; Kappera, R.; de Carvalho Castro e Silva, C., Kaplan, D., Fujita, T.; Chen, M.;
Asefa, T.; Chhowalla, M. Nat. Chem. 2015, 7, 45–49.
57.
Presolski, S.; Wang, L.; Huiling Loo, A.; Ambrosi, A.; Lazar, P.; Ranc, V.; Otyepka, M.; Zboril, R.;
Tomanec, O.; Ugolotti, J.; Sofer, Z.; Pumera, M. Chem. Mater. 2017, 29(5), 2066–2073.
58.
Eda, G.; Yamaguchi, H.; Voiry, D.; Fujita, T.; Chen, M.; Chhowalla, M. Nano Lett. 2011, 11, 5111–
5116.
59.
Zeng, Z.; Yin, Z.; Huang, X.; Li, H.; He, Q.; Lu, G.; Boey, F.; Zhang, H. Angew. Chem. Int. Ed. 2011,
50, 11093–11097.
60.
61.
Benavente, E.; Santa Ana, M.A.; Mendizàbal, F.; Gonzàlez, G. Coord. Chem. Rev. 2002, 224, 87–109.
Guo, Y.;Sun, D.; Ouyang, B.; Raja, A.; Song, J.; Heinz, T. F.; Brus, L. E. Nano Lett. 2015, 15,
5081−5088.
62.
Fan, X.; Xu, P.; Zhou, D.; Sun, Y.; Li, Y. C.; Nguyen, M. A. T.; Terrones, M.; Mallouk, T. E. Nano Lett.
2015, 15, 5956−5960.
63.
Li, H.; Zhang, Q.; Ray Yap, Q.-C.; Tay, B.-K.; Tong Edwin, T.-H.; Olivier, A.; Baillargeat, D. Adv.
Funct. Mater. 2012, 22, 1385–1390.
64.
Chen, Y.-J.; Cain, J. D.; Stanev, T. K.; Dravid, V. P.; Stern, N. Nature Phot. 2017, 11, 431- 436.
|
1907.02360 | 1 | 1907 | 2019-07-04T12:22:44 | Micron-size two-dimensional methylammonium lead halide perovskites | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Hybrid lead halide perovskites with 2D stacking structures have recently emerged as promising materials for optoelectronic applications. We report a method for growing 2D nanosheets of hybrid lead halide perovskites (I, Br and Cl), with tunable lateral sizes ranging from 0.05 to 8 microns, and a structure consisting of n stacked monolayers separated by long alkylamines, tunable from bulk down to n=1. The key to obtaining such a wide range of perovskite properties hinged on utilizing the respective lead halide nanosheets as precursors in a hot-injection synthesis that afforded careful control over all process parameters. The layered, quantum confined (n small than 4) nanosheets were comprised of major and minor fractions with differing n. Energy funneling from low to high n (high to low energy) regions within a single sheet, mediated by the length of the ligands between stacks, produced photoluminescent quantum yields as high as 49 percent. These large, tunable 2D nanosheets could serve as convenient platforms for future high efficiency optoelectronic devices. | physics.app-ph | physics | Micron-Size Two-Dimensional Methylammonium Lead Halide Perovskites
Eugen Klein1, Andres Black1, Öznur Tokmak2, Christian Strelow1, Rostyslav Lesyuk1,3,
Christian Klinke1,4,5*
1 Institute of Physical Chemistry, University of Hamburg, Martin-Luther-King-Platz 6, 20146
Hamburg, Germany
2 Fraunhofer Center for Applied Nanotechnology (CAN), Grindelallee 117, 20146 Hamburg,
Germany
3 Pidstryhach Institute for applied problems of mechanics and mathematics of NAS of
Ukraine, Naukowa str. 3b, 79060 Lviv & Department of Photonics, Lviv Polytechnic National
University, Bandery str. 12, 79000 Lviv, Ukraine
4 Department of Chemistry, Swansea University -- Singleton Park, Swansea SA2 8PP, United
Kingdom
5 Institute of Physics, University of Rostock, Albert-Einstein-Strasse 23, 18059 Rostock,
Germany
ABSTRACT
Hybrid lead halide perovskites with 2D stacking structures have recently emerged as
promising materials for optoelectronic applications. We report a method for growing 2D
nanosheets of hybrid lead halide perovskites (I, Br and Cl), with tunable lateral sizes ranging
from 0.05 to 8 µm, and a structure consisting of n stacked monolayers separated by long
alkylamines, tunable from bulk down to n=1. The key to obtaining such a wide range of
perovskite properties hinged on utilizing the respective lead halide nanosheets as precursors in
a hot-injection synthesis that afforded careful control over all process parameters. The
layered, quantum confined (n≤4) nanosheets were comprised of major and minor fractions
with differing n. Energy funneling from low to high n (high to low energy) regions within a
single sheet, mediated by
the
length of
the
ligands between stacks, produced
photoluminescent quantum yields as high as 49%. These large, tunable 2D nanosheets could
serve as convenient platforms for future high efficiency optoelectronic devices.
* Corresponding author: [email protected]
1
KEYWORDS: layered perovskites, large nanosheets, energy funneling, hot-injection, as
prepared lead halide precursors
The outstanding properties of perovskites, including low temperature processability, tunable
band gap,1 small exciton binding energy,2 narrow absorption edges and emission spectra, and
long charge carrier diffusion lengths,3 have been exploited in recent years for photovoltaic4
and optoelectronic applications.5 Bulk perovskites with the formula ABX3 have been
especially successful, where A is an organic ammonium cation or Cs+, B is Pb2+ and X is a
halide anion. In contrast, 2D layered perovskite nanosheets have a Ruddleson-Popper type
formula L2An-1BnX3n+1, where L is a long chain alkyl6 and n is the number of neighboring BX
monolayers between the organic spacers. This crystal structure makes 2D perovskite
nanosheets labile,7,8 and for n<4 they are in quantum confinement. Like quantum wells, the
low dielectric screening and large exciton binding energy in these quantum confined
perovskites enhances their radiative recombination properties compared to bulk (n=∞),
whereas the long alkyl ligands enhances their stability.9 Compared to films composed of small
nanoparticles, individual 2D nanosheets do not exhibit tunnel barriers or grain boundaries in
the lateral dimensions, which makes them interesting for optoelectronic studies10,11 and
flexible electronic devices.12
Perovskites nanosheets can be prepared through various approaches. Micron-sized sheets
were obtained via a chemical vapor deposition growth method.13 Colloidally stable
methylammonium lead bromide (MAPbBr) and MAPbI nanosheets, up to 500 nm in size, can
be prepared via exfoliation of bulk crystals with long chained ligands14 or via instantaneous
crystallization of precursor salts in an antisolvent.15 Despite the significant progress made in
the past years, synthesis of micron-size, colloidal, quantum-confined perovskites is limited to
CsPbBr.16 Such MAPbX nanosheets, where X is Cl, Br or I, have remained elusive.
We present in this work a colloidal method for producing MAPbX nanosheets for all halides,
with control of sheet thickness and structure down to n=1, as well as lateral dimensions from
<1 µm up to 8 μm. This hot-injection, colloidal synthesis method is based on dissolving as-
prepared PbX nanosheets17 and using them as precursor for the synthesis of the respective
MAPbX nanosheets. The hot-injection route allows tuning the lateral dimensions and
thickness of the nanosheets by controlling all aspects of the reaction process, from the ligand
and precursor concentration, to the temperature and reaction times. In contrast, colloidal anti-
2
solvent and exfoliation methods offer a much more limited control of the process parameters.
The perovskite structures were prepared either as single crystal 2D particles or as stacked
sheets consisting of repeatable MAn-1PbnX3n+1 layers separated by long ligands, as confirmed
by X-Ray Diffraction (XRD). The layered structures, which were more stable due to the dense
ligand shell,9 exhibited quantum confinement, as confirmed by UV-Vis absorption and
photoluminescent (PL) spectroscopy. Regions of different thickness within a single sheet
produced an energy/charge carriers funneling phenomenon,18 particularly strong in high PL
quantum yield (PLQY) MAPbBr, resulting in emission from the lowest energy (highest n)
sheets in the stack. The bottom-up synthesis approach for producing large area MAPbX
nanosheets presented herein provides opportunities for both fundamental and applied
optoelectronic research.
RESULTS AND DISCUSSION
Figure 1. Shape and structural characterization of (n>4) MAPbX nanosheets. TEM images of (a) MAPbCl (b)
MAPbBr and (c) MAPbI nanosheets. Insets show the SAEDs of the corresponding materials. (d), XRD patterns
of the three structures, displayed for comparison with the reference diffraction patterns (black).19 -- 21
3
Figures 1a, b and c present transmission electron microscope (TEM) images of bulk (n>4)
nanosheets of MAPbCl, MAPbBr and MAPbI, respectively. All these structures were
synthesized using the corresponding PbX nanosheets as precursors, shown in Figure S1. The
MAPbCl nanosheets have straight edges, uniform thickness and lateral sizes from 500 nm to
1.5 μm. The MAPbBr nanosheets shown in Figure 1b are more uniform and square-like
compared to the MAPbCl, with a size distribution between 400 and 600 nm. MAPbI
nanosheets show the most uniform shape and size distribution, with the ones depicted in
Figure 1c having lateral dimensions between 2 µm and 4 µm. Figures S2 to S4 show the wide
range of tunable lateral sizes that can be prepared for the bulk nanosheets, and is summarized
in Table 1. Whereas the structure of the MAPbCl changes from uniform squares to net-like
with increasing size, the shape of the MAPbBr and MAPbI sheets is independent of their
lateral size. Figure S5 depicts atomic force microscopy (AFM) images for all three materials.
The thickness fits well with the data calculated from XRD patterns and can be tuned from 2.5
nm to 80 nm for MAPbCl, 2.6 nm to 50 nm for MAPbBr and 20 nm to 56 nm for MAPbI
(Figure S6).
Table 1: Lateral size range of bulk and layered nanosheets.
Material
Bulk (µm)
Layered (µm)
MAPbI
0.05 -- up to 8
0.05 - 6
MAPbBr
0.05 - 2
0.05 -- up to 8
MAPbCl
0.3 -- up to 8
1 - 15
The insets in Figure 1a-c presents selected area electron diffraction (SAED) patterns for all
three materials, and confirm the single crystal appearance of the nanosheets. The lattice
constants obtained from the SAED analysis correspond to a=b≈5.75 Å for MAPbCl NSs,
a=b≈5.93 Å for MAPbBr NSs and a=b≈8.87 A for the MAPbI NSs. The appearance of only
(00l) signals indicates that the lateral growth and the alignment of nanosheets (NSs) is
orthogonal to the [001] direction. XRD patterns shown in Figure 1d reveal four signals for all
three materials, which fit with (00l) reflections and confirm the orientation of the crystal
lattice. The omission of the rest of the signals occurs due to the planar alignment of the sheets
on the XRD wafer and the resulting texture effect. The lattice constant c =5.71 Å, 5.95 Å and
12.62 Å was determined from (00l) patterns for MAPbCl, MAPbBr and MAPbI NSs.
4
Capillary XRD (Figure S7) agrees with the reference spectrum, indicating that no other
phases are present.
The bulk perovskite nanosheet structures with tunable lateral size and thickness were
synthesized using the corresponding PbX nanosheets prepared in nonanoic or oleic acid
(detailed protocols and instructions in the SI). The perovskite syntheses for all three materials
employ diphenyl ether (DPE) as the solvent and trioctylphosphine (TOP) together with long
chained amines as ligands. These chemicals were mixed, heated up to 80 °C and dried under
vacuum for 1 h. The lead halide nanosheets in toluene were injected at temperatures between
80 °C and 220 °C and stirred until all material was dissolved. The syntheses were started by
the addition of the methylammonium halides in dimethylformamide (DMF). The lateral size
can be tuned for the MAPbCl by varying the concentration of the reactants, and for MAPbBr
with the amount of ligands. For MAPbI, large sheets could be obtained by growing them
slowly via a controlled temperature increase, from 35 to 110 °C, and by decreasing the
amount of methylammonium iodide. In general, starting a reaction at high temperatures with a
ratio of the reactants nearly at 1:1 leads to many nuclei that grow into small particles. In
contrast to this, a slow increase in reaction temperature and a higher ratio between the two
reactants of 5:1 leads to a reaction mixture consisting of few nuclei that grow into big
structures.22,23 The thickness of the bulk sheets can be increased by increasing the amount of
methylammonium precursor for the MAPbCl and MAPbI nanosheets, and by increasing the
concentration of the two reactants for MAPbBr. Aliquots taken during the reaction (Figure
S8) show an agglomeration of small three dimensional particles arranged in the same size and
shape as the end product. SAED image of these structures are comprised of dot pattern that
indicates that the particles are all oriented in the same way and form quasi-crystal-units with
crystal lattice similar to MAPbI nanosheets. These findings indicate that the sheets are formed
either from small particles which agglomerate and merge, or that the precursors meet in some
sort of micelle in the shape and size of the end product prior to nucleating, similar to PbS
nanosheets24 and in contrast to the continuous growth mechanism of different colloidal
materials.25,26
5
Figure 2. Shape characterization of layered (n≤4) nanosheets. TEM images of (a) MAPbCl, (b) MAPbBr and (c)
MAPbI nanosheets. Insets show the SAEDs of the corresponding materials.
Figures 2a, b and c present TEM images of layered 2D nanosheets of MAPbCl, MAPbBr and
MAPbI, respectively. The wide range of lateral tunability is presented in Figures S9 to S11,
and summarized in Table 1. The MAPbCl sheets in Figure 2a show a large variation in shape
6
and lateral size, between 1 and 4 µm, whereas the MAPbBr sheets have a uniform square-like
shape and a smaller size distribution, between 1 and 2 µm. Similar to their bulk counterparts,
layered MAPbI nanosheets show the most uniform shape and size distribution, between 1.5
and 2.5 µm. Some MAPbBr and MAPbI layered nanosheets show pyramidal, shifted
pyramidal or squared spiral stacking, as shown in Figure S12. All of the layered nanosheets
produce a dot SAED pattern (Figure 2 insets), evidence of their monocrystallinity,
corresponding to a cubic or tetragonal lattice of MAPbX viewed from [001].
The general synthesis of layered and bulk nanosheets is similar. The critical difference for
obtaining layered sheets lies in utilizing a shorter alkyl chain ligand like hexadecylamine
(HDA), which favors layer stacking, and lower reaction temperatures, ensuring the stability of
the stacks. The lateral size of the MAPbCl sheets can be tuned by changing the amount of
amine ligands, whereas the amount of both TOP and amine ligands determined the size of the
MAPbI sheets. In contrast, careful control of the nucleation event is required to determine the
size of the MAPbBr sheets. Injecting the methylammonium bromide precursor into the
reaction mixture at 160 °C resulted in the immediate formation of 2 µm nanosheets.
Maintaining the temperature for 5 minutes completely dissolved the structures. Finally, letting
the solution cool slowly to 60 °C over 35 minutes leads to a nucleation around 80 °C and
sheets of 4 to 8 µm in size. In general, the variation of lateral sizes of MAPbX nanosheets
presented is constituted by several factors and is different for MAPbI, MAPbBr, MAPbCl,
respectively. In some cases, two or three parameters are responsible for the lateral growth
such as temperature and amount of ligand. Detailed protocols can be found in the SI. The
variation in thickness can be controlled for the chloride sheets with the chain length of the
amine, for bromide with the amount of the two reactants and for iodide with the temperature.
7
Figure 3. MAPbX synthesis. Schematic illustration depicting the importance of the precursor and its
purification, using PbI nanosheets as an example.
Maintaining the surface ligands of the PbX precursor was critical for a successful synthesis.
Therefore, as prepared PbX nanosheets were used, and were only centrifuged once, as shown
in the center row of Figure 3, showing a schematic of the general synthesis process. In order
to prepare MAPbX nanosheets, PbX nanosheets were dissolved in DPE along with TOP and
long chained amine ligands. In the case of PbI, the mixture turns from turbid yellow to a
slightly yellow pellucid solution. After dissolution, the resulting dissolved PbX nanoparticles
and Pb2+ ions were surrounded by a mixture of the original oleate and oleic acid ligands, and
the newly added TOP and amines. UV/VIS and PL spectra, shown in Figure S13, reveal
pronounced absorption and emission features in the range 330 -- 400 nm which we attribute to
tiny lead iodide nanoparticles with sizes below 1 nm. These eventually grew into the well-
defined MAPbX perovskite nanosheets. The synthesis did not work if PbX powders were
used, while excessive centrifugation (3 times) of the as prepared PbX nanosheets removed
many of their ligands and produced amorphous structures with undefined shapes, as shown at
the bottom right of Figure 3.
8
Figure 4. Structural characterization of layered MAPbX. (a) Schematic of layered MAPbX nanosheets for
n=1,2,3. Organic part of the structure illustrated schematically in yellow. (b) Experimental (color) and simulated
(black) XRD patterns for various n for (b) MAPbCl, (c) MAPbI and (d) MAPbBr. Red triangles correspond to
the presence of a minor fraction with different n, and red circles to a fractions coming from regions with
different stacking or bulk domains. Enlarged patterns of Bragg reflections for corresponding structures can be
found in Figure S14.
XRD, optical absorption and photoluminescent spectroscopy, shown in Figure 4 and 5,
respectively, were used to study the crystallographic and structural properties of the MAPbX
nanosheets, along with their resulting quantum confinement. Figure 4a depicts schematically
the PbX nanosheets, consisting of n perovskite monolayers between organic ligand spacers.
Due to the strong texture effect, the powder XRD patterns of these samples (prepared by
drop-casting diluted solutions, ensuring the lateral alignment of the sheets on the substrate)
consists of repeatable equidistant (00l) Bragg reflections, whose relative intensities are
determined by n. These spectra agree well for all materials and thicknesses with XRD patterns
calculated from the respective inorganic PbX crystal structures for n=1, 2 and 3. Capillary
XRD measurements, shown in Figure S15, eliminate the texture effects and confirm the
perovskite structure of the layered sheets, and the half-unit cell shift between adjacent n-
monolayer stacks within the same sheet, as reported previously for 2D Ruddleson-Popper
perovskite crystals.6 The XRD patterns reveal that the multi-layered nanosheets tend to form
mixtures of different n values, as indicated by the red triangles in the figures for the minor n
fraction. Relatively pure samples for different n values were obtained for both the MAPbI and
the MAPbBr based nanosheets, indicating the robustness and tunability of the synthesis
process for these systems. The Bragg reflections positions were used to calculate the unit
constant c, corresponding to the spacing between adjacent n-monolayer stacks, and
9
subsequently the thickness of the organic layer spacer between the stacks. For MAPbI
nanosheets, c=29.5 (n=1), 34.5 (n=2), and 40.6 Å (n=3), with an organic spacing between 2.1
and 2.3 nm, which agrees well with the 2.17 nm length of the HDA employed as ligand.
Likewise, for MAPbBr, c equals 29.6 (n=2), 35.4 (n=3), and 41.1 Å (n=4), and the organic
layer spacing of 1.7 nm agrees well with the length of the dodecylamine (DDA) ligand. In
both cases, the ligands between n-layer stacks appear to be interdigitated along the [001] axis,
as illustrated in Figure 4a. Only n=1 stacking was obtained in relatively pure form for the
MAPbCl nanosheets, with c=31.1 Å and an organic layer spacing of 2.5 nm, significantly
longer than the expected 1.9 nm of the tetradecylamine (TDA) ligand. This indicates that the
ligands between neighboring MAPbCl stacks are not fully interdigitated. The residual
intensity of narrow non-Bragg reflexes for a given structure (for example in Figure 4b, c as
indicated by the red circle) might originate from bulk regions within the sheets or regions with
a more complex built-up.
Figure 5. Optical properties of MAPbX nanosheets. (a) UV/VIS absorption and photoluminescence (PL) spectra
(solid and dashed lines, respectively). Time-dependent PL decay of MAPbBr at various wavelengths, for (b)
short (linear scale) and (c) long times (logarithmic scale). Arrows in (b) correspond to PL decay illustrated in (d)
for specific wavelengths corresponding to different n.
10
The presence of regions of different thicknesses within the same nanosheet batch is supported
by absorption and PL spectroscopy, from the near infrared for the MAPbI series to the near
UV for the MAPbCl, as shown in Figure 5a. As the sheets transition from bulk to quantum
confined, the absorption spectra reveal the emergence of sharp excitonic features.14,27 The
MAPbI n=1 sample shows excellent spectral purity, with a single absorption peak at 492 nm,
and a slightly offset PL peak at 502 nm. Likewise, the bulk spectrum shows an absorption
edge peaking at 730 nm, and PL at 759 nm. In contrast, the other two spectra in the MAPbI
series show mixed n fractions: the n=2 sample shows absorption at 574 nm and a shoulder at
600 nm, corresponding to the main n=2 and a minor n=3 fractions, with the respective PL
peaks Stokes-shifted by about 10 nm. The main fraction n=3 spectrum also shows minor
absorption features at longer wavelengths, with almost all of the PL coming from the thickest
minor fraction at 674 nm.
In the MAPbBr series, the main PL peak for all of the samples is centered between 515 and
525 nm, at significantly lower energy than the exitonic absorption peaks arising from
quantum confinement of the nanosheets. This PL emission at lower energy is likely due to an
energy cascade mediated by excitonic charge transfer from high energy, quantum confined
regions to lower energy bulk regions within the same nanosheets.28 This is also known as
energy funneling.29 In this process, more confined, low n regions with larger bandgap transfer
their energy via resonant energy or charge transfer processes to less confined, higher n regions
with lower bandgap, where radiative emission occurs. Regions with differing n could be
stacked either vertically or laterally next to each other. Figures 5b-d show time-resolved PL
spectroscopy of an ensemble in solution, with Figure 5b showing that upon excitation, the n=2
fraction emits a very short burst of light that very quickly dies off, as its energy is rapidly
transferred to higher n fractions. This is reflected in the much longer 1/e decay time of the
n=4 fraction, 88 ns, compared to the n=2 (150 ps) and n=3 fractions (140 ps), as shown in
Figure 5c, d and inset. These observations agree with previous time-resolved spectroscopic
studies for 2D perovskites.28,29 The short DDA ligands used to prepare the MAPbBr sheets
favored the energy funneling process. Synthesizing with longer TDA and HDA resulted in
emission at higher energies (Figure S16), away from the bulk, confirming that the energy
transfer process is much weaker. Longer octadecylamine (ODA) does not favor the formation
of layered structures, and produces bulk nanosheets with a corresponding emission.
11
Energy funneling towards lower bandgap regions, where radiative recombination processes
are more efficient, results in high PL quantum yields (PLQY) of up to 49% for the MAPbBr
sample with n=4 main fraction. The high PLQY value is striking, considering that the volume
of the bulk-like regions must be small, at most 10%, compared to that of the layered regions
based on optical absorption and XRD data (minor bulk-like features at 15 and 30° can be
deconvoluted from the XRD measurements for n=4, as shown in S17), and is evidence of the
efficiency of the energy funneling process. Indeed, such nanograins of bulk MAPbBr have
been engineered recently to enhance the efficiency of light emitting diodes by discouraging
the dissociation of excitons into unbound charge carriers.30 If energy funneling were not
present, it would require that the bulk regions have PLQYs over 100%, considering their
approximate volumes and that about 70% of the PL comes from these regions. The quantum
yield is significantly lower for MAPbBr nanosheets with lower n value, down to 11% for n=2
main fraction. Despite having an
increased exciton binding energy and radiative
recombination rate, charges in thinner layers are less effectively screened from surface
defects, resulting in lower quantum yields.9 The group of Tisdale et al. reported an increase of
PLQY for colloidal 2D perovskites from 6% for MAPbBr to 22% for formamidinium-based
analogue with n=2.31 A PLQY reaching 70% was reported by the group of Urban et al. for
mixed lead bromide-iodide perovskite nanoplatelets obtained by the exfoliation method.14
Recently, quasi-2D perovskites in films on the basis of 5-aminovaleric acid cross-linked
MAPbBr were presented with PLQY reaching 80%.32 Thus our nanosheets present high
quality and effective perovskites among solution-processable hybrid 2D MAPbBr
nanostructures obtained by hot-injection colloidal synthesis offering additionally the potential
of large lateral size.
An energy funneling process is observable in the MAPbI sample with n=3 as the main
fraction, having a PLQY of 1%. The generally lower PLQY can be partly attributed to its
lower exciton binding energy (about three times lower than MAPbBr).33 In addition, energy
funneling effects between regions of differing n will be significantly hampered by the longer
organic spacer separating them (2.2 nm for MAPbI compared to 1.7 nm for MAPbBr).
Indeed, in order to achieve efficient energy funneling in MAPbI films, Yuan et al. used the
relatively short phenylethylammonium (~0.8 nm) ligand. Similarly, the 2.5 nm organic spacer
for the MAPbCl series is expected to hinder energy funneling to low bandgap regions,
resulting in PL from all of the peaks in the nanosheets, as shown in Figure 5c. The organic
12
spacer ligands, longer in this work compared to those employed in other studies, enhances the
potential barrier around the perovskite layers, intensifying their quantum confinement.
Whereas the n=1 MAPbI PL in this work is centered at 502 nm, it was at 512,31 528 and 539
nm for similar n=1 nanosheets with shorter spacer ligands.6,14 These results show that the
ligand choice for the formation of 2D perovskite NSs might have dramatic influence on their
optical properties and should be taken into account during designing devices based on these
materials. Additionally, investigations regarding a change in the PL with different lateral sizes
showed no significant shift of the peak (Figure S18).
CONCLUSION
In summary, highly tunable, colloidal route protocols are presented for synthesizing micron-
sized nanosheets of MAPbX (X=I, Br, Cl) with 2D (layered, n=1 to 4) and 3D structure, as
well as good control over both the lateral size and thickness. Excitonic features in the optical
spectra confirmed the 2D nature of the nanosheets, and revealed an energy funneling process
from low to high n regions, mediated by the length of the organic ligands. This efficient
process, which resulted in strong emission from the high n regions (PLQY up to 49%) even if
these were present only in very small quantities, could be used to engineer high performance
optoelectronic devices.
METHODS
Chemicals and reagents. All chemicals were used as received: Lead(II) acetate tri-hydrate
(Aldrich, 99.999%), oleic acid (OA, Aldrich, 90%), nonanoic acid (Alfa Aesar, 97%), tri-
octylphosphine
(TOP; ABCR,
97%),
1,2-diiodoethane
(DIE; Aldrich,
99%),
methylammonium bromide (MAB; Aldrich, 98%), methylammonium chloride (MAC;
Aldrich, 98%), methylammonium iodide (MAI; Aldrich, 98%), diphenyl ether (DPE; Aldrich,
99%),
toluene
(VWR, 99,5%), dimethylformamide
(DMF; Aldrich, 99,8%), 1-
bromotetradecane (BTD; Aldrich, 97%), 1-chlorotetradecane (CTD; Aldrich, 98%),
octadecylamine
(ODA; Aldrich, 97%),
tetradecylamine
(TDA; Aldrich, 95%),
hexadecylamine (HDA; Aldrich, 90%), dodecylamine (DDA; Merck, 98%), oleylamine
(ACROS, 80-90%).
13
Synthesis of n>4 MAPbI3 nanosheets. Standard synthesis: A three neck 50 mL flask was
used with a condenser, septum and thermocouple. 10.5 mL of diphenyl ether (66.6 mmol),
0.48 mL of a 400 mg hexadecylamine (1.66 mmol) in 4 mL diphenyl ether precursor and 0.1
mL (0.22 mmol) of TOP were heated to 80 °C in a nitrogen atmosphere. Then vacuum was
applied to dry the solution. After 1 h the reaction apparatus was filled with nitrogen again and
2 mL of as prepared PbI2 nanosheets in toluene were added. The reaction temperature was
reduced to 35 °C after all of the PbI2 dissolved. The synthesis was started with the injection of
0.06 mL of a 600 mg methylammonium iodide (3.77 mmol) in 6 mL dimethylformamide
precursor. After the injection the temperature was slowly increased to 90 °C for a time period
of 9 minutes. At 90 °C the heat source was removed and the solution was left to cool down
below 60 °C. Afterwards, it was centrifuged at 4000 rpm for 3 minutes. The particles were
washed two times in toluene before the product was finally suspended in toluene again and
put into a freezer for storage.
Synthesis of n>4 MAPbBr3 nanosheets. Standard synthesis: A three neck 50 mL flask was
used with a condenser, septum and thermocouple. 10.5 mL of diphenyl ether (66.6 mmol),
0.06 mL of oleylamine (0.18 mmol) and 0.2 mL (0.44 mmol) of TOP were heated to 80 °C in
a nitrogen atmosphere. Then vacuum was applied to dry the solution. After 1 h the reaction
apparatus was filled with nitrogen again, the temperature was increased at 120 °C and 1.5 mL
of as prepared PbBr2 nanosheets in toluene were added. The reaction temperature was reduced
to 35 °C after all of the PbBr2 dissolved. The synthesis was started with the injection of 0.06
mL of a 300 mg methylammonium bromide (2.68 mmol) in 6 mL dimethylformamide
precursor. After the injection the temperature was increased to 120 °C. After 10 minutes the
heat source was removed and the solution was left to cool down below 60 °C. Afterwards, it
was centrifuged at 4000 rpm for 3 minutes. The particles were washed two times in toluene
before the product was finally suspended in toluene again and put into a freezer for storage.
Synthesis of n>4 MAPbCl3 nanosheets. Standard synthesis: A three neck 50 mL flask was
used with a condenser, septum and thermocouple. 10.5 mL of diphenyl ether (66.6 mmol),
0.04 mL of oleylamine (0.12 mmol) and 0.1 mL (0.22 mmol) of TOP were heated to 80 °C in
a nitrogen atmosphere. Then vacuum was applied to dry the solution. After 1 h the reaction
apparatus was filled with nitrogen again, the temperature was increased at 220 °C and 1 mL
of as prepared PbCl2 nanosheets in toluene was added. After all of the PbCl2 was dissolved
the temperature was reduced to 100 °C. The synthesis was started with the injection of 0.36
14
mL of a 50 mg methylammonium chloride (0.74 mmol) in 6 mL dimethylformamide
precursor. After 10 minutes the heat source was removed and the solution was left to cool
down below 60 °C. Afterwards, it was centrifuged at 4000 rpm for 3 minutes. The particles
were washed two times in toluene before the product was finally suspended in toluene again
and put into a freezer for storage.
Synthesis of n<4 MAPbI3 nanosheets. Standard synthesis: A three neck 50 mL flask was
used with a condenser, septum and thermocouple. 10.5 mL of diphenyl ether (66.6 mmol),
0.48 mL of a 400 mg hexadecylamine (1.66 mmol) in 4 mL diphenyl ether precursor and 0.1
mL (0.22 mmol) of TOP were heated to 80 °C in a nitrogen atmosphere. Then vacuum was
applied to dry the solution. After 1 h the reaction apparatus was filled with nitrogen again and
2 mL of as prepared PbI2 nanosheets in toluene were added. The reaction temperature was
reduced to 35 °C after all of the PbI2 dissolved. The synthesis was started with the injection of
0.06 mL of a 600 mg methylammonium iodide (3.77 mmol) in 6 mL dimethylformamide
precursor. After the injection the temperature was slowly increased to 60 °C for a time period
of 6 minutes. At 60 °C the heat source was removed and the solution was centrifuged at 4000
rpm for 3 minutes. The particles were washed two times in toluene before the product was
finally suspended in toluene again and put into a freezer for storage.
Synthesis of n<4 MAPbBr3 nanosheets. Standard synthesis: A three neck 50 mL flask was
used with a condenser, septum and thermocouple. 10.5 mL of diphenyl ether (66.6 mmol), 0.2
mL of a 500 mg dodecylamine (2.70 mmol) in 4 mL diphenyl ether precursor and 0.2 mL
(0.44 mmol) of TOP were heated to 80 °C in a nitrogen atmosphere. Then vacuum was
applied to dry the solution. After 1 h the reaction apparatus was filled with nitrogen again, the
temperature was increased to 160 °C and 2.5 mL of as prepared PbBr2 nanosheets in toluene
was added. The synthesis was started after all of the PbBr2 dissolved with the injection of 0.03
mL of a 300 mg methylammonium bromide (2.68 mmol) in 6 mL dimethylformamide
precursor. After 5 minutes the heat source was removed and the solution was left to cool
down below 60 °C. Afterwards, it was centrifuged at 4000 rpm for 3 minutes. The particles
were washed two times in toluene before the product was finally suspended in toluene again
and put into a freezer for storage.
Synthesis of n<4 MAPbCl3 nanosheets. Standard synthesis: A three neck 50 mL flask was
used with a condenser, septum and thermocouple. 10.5 mL of diphenyl ether (66.6 mmol),
15
0.24 mL of a 400 mg hexadecylamine (1.66 mmol) in 4 mL diphenyl ether precursor and 0.1
mL (0.22 mmol) of TOP were heated to 80 °C in a nitrogen atmosphere. Then vacuum was
applied to dry the solution. After 1 h the reaction apparatus was filled with nitrogen again, the
temperature was increased to 220 °C and 1 mL of as prepared PbCl2 nanosheets in toluene
was added. The reaction temperature was reduced to 100 °C after all of the PbI2 dissolved.
The synthesis was started with the injection of 0.24 mL of a 50 mg methylammonium
chloride (0.74 mmol) in 6 mL dimethylformamide precursor. After 10 minutes the heat source
was removed and the solution was left to cool down below 60 °C. Afterwards, it was
centrifuged at 4000 rpm for 3 minutes. The particles were washed two times in toluene before
the product was finally suspended in toluene again and put into a freezer for storage.
Variation of the dimensions for all three materials for bulk and for the layered structures is
described in detail in the Supporting Information.
Characterization. The TEM samples were prepared by diluting the nanosheet suspension
with toluene followed by drop casting 10 µL of the suspension on a TEM copper grid coated
with a carbon film. Standard images were done on a JEOL-1011 with a thermal emitter
operated at an acceleration voltage of 100 kV. X-ray diffraction (XRD) measurements were
performed on a Philips X'Pert System with a Bragg-Brentano geometry and a copper anode
with a X-ray wavelength of 0.154 nm. The samples were measured by drop-casting the
suspended nanosheets on a <911> or <711> cut silicon substrate. Atomic force microscopy
(AFM) measurements were performed in tapping mode on a JPK Nano Wizard 3 AFM in
contact mode. Images were taken of the as-prepared nanoring devices. UV/vis absorption
spectra were obtained with a Cary 5000 spectrophotometer equipped with an integration-
sphere. The PL spectra measurements were obtained by a fluorescence spectrometer
(Fluoromax-4, Horiba Jobin Yvon). Simulations of XRD spectra were carried out in
PowderCell 2.4 software using crystallographic data from the literature.19-21 The structure was
simplified excluding the organic part of each unit cell, thus simulating solely Pb-X (X=Cl, Br,
I) networks. Absolute quantum yield measurements were performed in solution with K-
Sphere 'Petite' Integrating Sphere (Horiba) and Fluorolog-3 with FluorEssence software.
Liquid samples were prepared as low-concentrated nanosheets solutions in toluene (optical
density in the range of 0.02 -- 0.05) in quartz cuvettes (QG). Excitation and emission spectra of
both QD solutions and pure toluene were recorded at the excitation wavelength of 420 nm and
16
recalculated in photons absorbed and emitted by the nanosheets according to the absolute 4-
step measurement method. Time-resolved PL measurements were performed with Picoquant
FT300 fluorescence spectrometer (1200 lines/mm grating, 30 cm focal length, PMA Hybrid
Detector). Excitation was carried out with SuperK FIANIUM FIA-15 white-light laser with
LLTF contrast tunable single line filter (1.5 nm bandwidth). Pulse duration was set to 60 ps,
excitation wavelength to - 410 nm.
ASSOCIATED CONTENT
*Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at
DOI:
Additional experimental details; optical absorbance spectra; emission spectra; characterization
of XRDs performed in a capillary for bulk and layered perovskite NSs; SAED patterns; TEM
images; UV/Vis and emission spectra of dissolved lead halide nanosheets; AFM images and
measured height images of synthesized perovskite nanosheets; TEM analysis and XRD
patterns; difference in lateral dimensions and thickness; TEM images of pyramidal
nanosheets; TEM images and SAED patterns of aliquots taken during a perovskite synthesis.
ACKNOWLEDGMENTS
The authors thank the Alf Mews group for providing the Confocal Microscopy setup. Further,
the German Research Foundation DFG is acknowledged for financial support in the frame of
the Cluster of Excellence "Center of ultrafast imaging CUI" and for granting the project KL
1453/9-2. The European Research Council is acknowledged for funding an ERC Starting
Grant (Project: 2D-SYNETRA (304980), Seventh Framework Program FP7). We further
acknowledge MINECO (Spain) for the project MAT2016-81118-P.
17
TABLE OF CONTENTS
REFERENCES
(1) Baikie, T.; Fang, Y.; Kadro, J. M.; Schreyer, M.; Wei, F.; Mhaisalkar, S. G.; Graetzel,
M.; White, T. J. Synthesis and Crystal Chemistry of the Hybrid Perovskite
(CH3NH3)PbI3 for Solid-State Sensitised Solar Cell Applications. J. Mater. Chem. A
2013, 1, 5628.
(2) Xing, G.; Mathews, N.; Sun, S.; Lim, S. S.; Lam, Y. M.; Gratzel, M.; Mhaisalkar, S.;
Sum, T. C. Long-Range Balanced Electron- and Hole-Transport Lengths in Organic-
Inorganic CH3NH3PbI3. Science 2013, 342, 344 -- 347.
(3) Dou, L. Emerging Two-Dimensional Halide Perovskite Nanomaterials. J. Mater.
Chem. C 2017, 5, 11165 -- 11173.
(4) NREL efficiency chart https://www.nrel.gov/pv/assets/images/efficiency-chart.png
(accessed June 6, 2019).
(5) Sutherland, B. R.; Sargent, E. H. Perovskite Photonic Sources. Nat. Photonics 2016,
10, 295 -- 302.
(6) Stoumpos, C. C.; Cao, D. H.; Clark, D. J.; Young, J.; Rondinelli, J. M.; Jang, J. I.;
Hupp, J. T.; Kanatzidis, M. G. Ruddlesden -- Popper Hybrid Lead Iodide Perovskite 2D
Homologous Semiconductors. Chem. Mater. 2016, 28, 2852 -- 2867.
(7) Mitzi, D. B. Progress in Inorganic Chemistry. Synthesis, Structure, and Properties of
Organic-Inorganic Perovskites and Related Materials, Wiley: New York, 2007; Vol
48, pp 1 -- 121.
(8) Mitzi, D. B. Templating and Structural Engineering in Organic -- Inorganic Perovskites.
J. Chem. Soc. Dalt. Trans. 2001, 1 -- 12.
(9) Weidman, M. C.; Goodman, A. J.; Tisdale, W. A. Colloidal Halide Perovskite
18
Nanoplatelets: An Exciting New Class of Semiconductor Nanomaterials. Chem. Mater.
2017, 29, 5019 -- 5030.
(10) Dogan, S.; Bielewicz, T.; Lebedeva, V.; Klinke, C. Photovoltaic Effect in Individual
Asymmetrically Contacted Lead Sulfide Nanosheets. Nanoscale 2015, 7, 4875 -- 4883.
(11) Bielewicz, T.; Dogan, S.; Klinke, C. Tailoring the Height of Ultrathin PbS Nanosheets
and Their Application as Field-Effect Transistors. Small 2015, 11, 826 -- 833.
(12) Vikulov, S.; Di Stasio, F.; Ceseracciu, L.; Saldanha, P. L.; Scarpellini, A.; Dang, Z.;
Krahne, R.; Manna, L.; Lesnyak, V. Fully Solution-Processed Conductive Films Based
on Colloidal Copper Selenide Nanosheets for Flexible Electronics. Adv. Funct. Mater.
2016, 26, 3670 -- 3677.
(13) Wang, Y.; Shi, Y.; Xin, G.; Lian, J.; Shi, J. Two-Dimensional van Der Waals Epitaxy
Kinetics in a Three-Dimensional Perovskite Halide. Cryst. Growth Des. 2015, 15,
4741 -- 4749.
(14) Hintermayr, V. A.; Richter, A. F.; Ehrat, F.; Döblinger, M.; Vanderlinden, W.; Sichert,
J. A.; Tong, Y.; Polavarapu, L.; Feldmann, J.; Urban, A. S. Tuning the Optical
Properties of Perovskite Nanoplatelets through Composition and Thickness by Ligand-
Assisted Exfoliation. Adv. Mater. 2016, 28, 9478 -- 9485.
(15) Sichert, J. A.; Tong, Y.; Mutz, N.; Vollmer, M.; Fischer, S.; Milowska, K. Z.; García
Cortadella, R.; Nickel, B.; Cardenas-Daw, C.; Stolarczyk, J. K.; Urban, A. S.;
Feldmann, J. Quantum Size Effect in Organometal Halide Perovskite Nanoplatelets.
Nano Lett. 2015, 15, 6521 -- 6527.
(16) Shamsi, J.; Dang, Z.; Bianchini, P.; Canale, C.; Di Stasio, F.; Brescia, R.; Prato, M.;
Manna, L. Colloidal Synthesis of Quantum Confined Single Crystal CsPbBr3
Nanosheets with Lateral Size Control up to the Micrometer Range. J. Am. Chem. Soc.
2016, 138, 7240 -- 7243.
(17) Klein, E.; Lesyuk, R.; Klinke, C. Insights into the Formation Mechanism of Two-
Dimensional Lead Halide Nanostructures. Nanoscale 2018, 10, 4442 -- 4451.
(18) Tian, W.; Leng, J.; Zhao, C.; Jin, S. Long-Distance Charge Carrier Funneling in
Perovskite Nanowires Enabled by Built-in Halide Gradient. J. Am. Chem. Soc. 2017,
139, 579 -- 582.
(19) Yamada, Y.; Yamada, T.; Phuong, L. Q.; Maruyama, N.; Nishimura, H.; Wakamiya,
A.; Murata, Y.; Kanemitsu, Y. Dynamic Optical Properties of CH3NH3PbI3 Single
Crystals As Revealed by One- and Two-Photon Excited Photoluminescence
Measurements. J. Am. Chem. Soc. 2015, 137, 10456 -- 10459.
(20) Elbaz, G. A.; Straus, D. B.; Semonin, O. E.; Hull, T. D.; Paley, D. W.; Kim, P.; Owen,
J. S.; Kagan, C. R.; Roy, X. Unbalanced Hole and Electron Diffusion in Lead Bromide
Perovskites. Nano Lett. 2017, 17, 1727 -- 1732.
(21) Brunetti, B.; Cavallo, C.; Ciccioli, A.; Gigli, G.; Latini, A. Corrigendum : On the
Thermal and Thermodynamic ( In ) Stability of Methylammonium Lead Halide
Perovskites. Nat. Publ. Gr. 2017, 3, 6 -- 7.
19
(22) De Mello Donegá, C.; Liljeroth, P.; Vanmaekelbergh, D. Physicochemical Evaluation
of the Hot-Injection Method, a Synthesis Route for Monodisperse Nanocrystals. Small
2005, 1, 1152 -- 1162.
(23) Bielewicz, T.; Ramin Moayed, M. M.; Lebedeva, V.; Strelow, C.; Rieckmann, A.;
Klinke, C. From Dots to Stripes to Sheets: Shape Control of Lead Sulfide
Nanostructures. Chem. Mater. 2015, 27, 8248 -- 8254.
(24) Schliehe, C.; Juarez, B. H.; Pelletier, M.; Jander, S.; Greshnykh, D.; Nagel, M.; Meyer,
A.; Foerster, S.; Kornowski, A.; Klinke, C.; Weller, H. Ultrathin PbS Sheets by Two-
Dimensional Oriented Attachment. Science 2010, 329, 550 -- 553.
(25) Lesyuk, R.; Klein, E.; Yaremchuk, I.; Klinke, C. Copper Sulfide Nanosheets with
Shape-Tunable Plasmonic Properties in the NIR Region. Nanoscale 2018, 10, 20640 --
20651.
(26) Ithurria, S.; Bousquet, G.; Dubertret, B. Continuous Transition from 3D to 1D
Confinement Observed during the Formation of CdSe Nanoplatelets. J. Am. Chem. Soc.
2011, 133, 3070 -- 3077.
(27) Tyagi, P.; Arveson, S. M.; Tisdale, W. A. Colloidal Organohalide Perovskite
Nanoplatelets Exhibiting Quantum Confinement. J. Phys. Chem. Lett. 2015, 6, 1911 --
1916.
(28) Bouduban, M. E. F.; Burgos-Caminal, A.; Ossola, R.; Teuscher, J.; Moser, J.-E. Energy
and Charge Transfer Cascade in Methylammonium Lead Bromide Perovskite
Nanoparticle Aggregates. Chem. Sci. 2017, 8, 4371 -- 4380.
(29) Yuan, M.; Quan, L. N.; Comin, R.; Walters, G.; Sabatini, R.; Voznyy, O.; Hoogland,
S.; Zhao, Y.; Beauregard, E. M.; Kanjanaboos, P.; Lu, Z.; Kim, D. H.; Sargent, E. H.
Perovskite Energy Funnels for Efficient Light-Emitting Diodes. Nat. Nanotechnol.
2016, 11, 872 -- 877.
(30) Cho, H.; Jeong, S.-H.; Park, M.-H.; Kim, Y.-H.; Wolf, C.; Lee, C.-L.; Heo, J. H.;
Sadhanala, A.; Myoung, N.; Yoo, S.; Im, S. H. Friend, R. H. Lee, T.-W. Overcoming
the Electroluminescence Efficiency Limitations of Perovskite Light-Emitting Diodes.
Science 2015, 350, 1222 -- 1225.
(31) Weidman, M. C.; Seitz, M.; Stranks, S. D.; Tisdale, W. A. Highly Tunable Colloidal
Perovskite Nanoplatelets through Variable Cation, Metal, and Halide Composition.
ACS Nano 2016, 10, 7830 -- 7839.
(32) Zhang, T.; Xie, L.; Chen, L.; Guo, N.; Li, G.; Tian, Z.; Mao, B.; Zhao, Y. In Situ
Fabrication of Highly Luminescent Bifunctional Amino Acid Crosslinked 2D/3D NH3
C4H9COO(CH3NH3PbBr3) n Perovskite Films. Adv. Funct. Mater. 2017, 27, 1603568.
(33) Yang, Y.; Yang, M.; Li, Z.; Crisp, R.; Zhu, K.; Beard, M. C. Comparison of
Recombination Dynamics in CH3NH3PbBr3 and CH3NH3PbI3 Perovskite Films:
Influence of Exciton Binding Energy. J. Phys. Chem. Lett. 2015, 6, 4688 -- 4692.
20
|
1905.03931 | 2 | 1905 | 2019-06-05T01:00:21 | A Plainified Composite Absorber Enabled by Vertical Interphase | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Interface constitutes a significant volume fraction in nanocomposites, and it requires the ability to tune and tailor interfaces to tap the full potential of nanocomposites. However, the development and optimization of nanocomposites is currently restricted by the limited exploration and utilization of interfaces at different length scales. In this research, we have designed and introduced a relatively large-scale vertical interphase into carbon nanocomposites, in which the dielectric response and dispersion features in microwave frequency range are successfully adjusted. A remarkable relaxation process has been observed in vertical-interphase nanocomposites, showing sensitivity to both filler loading and the discrepancy in polarization ability across the interphase. Together with our analyses on dielectric spectra and relaxation processes, it is suggested that the intrinsic effect of vertical interphase lies in its ability to constrain and localize heterogeneous charges under external fields. Following this logic, systematic research is presented in this article affording to realize tunable frequency-dependent dielectric functionality by means of vertical interphase engineering. Overall, this study provides a novel method to utilize interfacial effects rationally. The research approach demonstrated here has great potential in developing microwave dielectric nanocomposites and devices with targeted or unique performance such as tunable broadband absorbers. | physics.app-ph | physics | A Plainified Composite Absorber Enabled
by Vertical Interphase
Yuhan Li, Faxiang Qin*, Le Quan, Huijie Wei, Yang Luo, Huan Wang, Hua-Xin Peng
Institute for Composites Science Innovation, Department of Materials Science and Engineering,
Zhejiang University, Hangzhou, China
KEYWORDS: carbon nanotubes; nanocomposites; interphase; dielectric response
ABSTRACT: Interface constitutes a significant volume fraction in nanocomposites, and it requires
the ability to tune and tailor interfaces to tap the full potential of nanocomposites. However, the
development and optimization of nanocomposites is currently restricted by the limited exploration
and utilization of interfaces at different length scales. In this research, we have designed and
introduced a relatively large-scale vertical interphase into carbon nanocomposites, in which the
dielectric response and dispersion features in microwave frequency range are successfully adjusted.
A remarkable relaxation process has been observed in vertical-interphase nanocomposites, showing
sensitivity to both filler loading and the discrepancy in polarization ability across the interphase.
Together with our analyses on dielectric spectra and relaxation processes, it is suggested that the
intrinsic effect of vertical interphase lies in its ability to constrain and localize heterogeneous
charges under external fields. Following this logic, systematic research is presented in this article
affording to realize tunable frequency-dependent dielectric functionality by means of vertical
interphase engineering. Overall, this study provides a novel method to utilize interfacial effects
rationally. The research approach demonstrated here has great potential in developing microwave
dielectric nanocomposites and devices with targeted or unique performance such as tunable
broadband absorbers.
1 INTRODUCTION
Interface has a crucial influence on the overall performance of composites both mechanically and
functionally1-3. Especially in terms of dielectric properties, the permittivity, breakdown strength,
dielectric dispersion and relaxation dynamics of composites are very sensitive to interfacial
properties4-7. With the filler size decreasing to nanoscale, interfaces become more complex and
spacious, playing a pivotal role in the overall dielectric functionality8, 9. Thereby, great attention has
been paid to reveal the internal correlation between interface and dielectric properties both
theoretically and experimentally3, 10-12. Maxwell-Wagner-Sillars effect gives a general explanation
of interface-induced polarization, which originates from the differences in conductivity, permittivity
and relaxation time of charge carriers in the materials across the interface13-15. By tuning interfacial
properties with various chemical and engineering methods, the polarization ability and dielectric
response of nanocomposites are also changed accordingly16-20. For example, the permittivity of
carbon nanotube (CNT)/ polyvinylidene fluoride (PVDF) nanocomposites was reported to increase
remarkably with enhanced molecular interaction and huge interfacial area4. In the microwave
frequency range, dielectric relaxation peaks of CNT/silicone elastomer (SE) nanocomposites were
shown to be sensitive to different types of interfacial interaction between chemically modified CNTs
and polymer matrix21. While great attention has been given to explore these phenomena from
various perspectives, limited research is carried out on utilizing the interfacial effects to tailor the
dielectric properties rationally and effectively.
Based on the existed research that focuses on the filler-matrix interface, it would be promising if
we can amplify these effects by enlarging the interfacial region to realize the true potential of
interface engineering. Here, we propose "plainified composites" to indicate composites with
optimized performance achieved by only interface engineering (Figure 1). As shown in the figure,
canonical composites are normally composed of the matrix and the filler. Traditionally, certain
properties of composites can be improved by adding more fillers or various types of fillers. However,
the design/manufacture complexity and structural disintegrity will be increased accordingly.
Meanwhile, the lightweight feature of composites could also be compromised when the filler
loading becomes higher. In comparison, plainified composites are extremely potential as the
material system and concentration of fillers remain unchanged. In these composites, interfacial
2
effects can be fully utilized through various methods such as modifying filler-matrix interface,
introducing in-built interphase or large-scale interface without the penalty of density and structural
integrity. In metallic materials, the concept of "plainification" has also been highlighted, in which
tailored microstructures are achieved by modifying or altering grain boundaries instead of changing
the composition22. Thereby, the proposed research topic is of general interest towards a wide range
of applications. Through engineering interfaces at different length scales, we will be able to realize
more efficient design methodology and superior materials performance.
Figure 1. Plainified composites enabled by interface engineering (the X axis represents the
relative design/manufacture complexity and structural disintegrity, and the Y axis represents the
specific property of composites)
The implication of interface engineering and plainified composites can be magnificent in the
context of dielectric functionality. Previous research has shown that the interfacial area of
composites has a major influence on the dielectric performance. In storage capacitors, by building
multi-layer or sandwich structures, the interfacial area can be greatly increased5. The broad
interfacial area between layers could restrict the tunneling effect of electrons and delay the
breakdown of materials under high electric field. Joyce et.al introduced multi-layer structure into
polymer capacitors and increased the energy barrier at the interface, thereby minimizing tunneling
3
current and increasing the energy density of the capacitor effectively23. Meanwhile, there are charge
accumulations and polarizations at the interface between layers, enhancing the dielectric response
and improving the permittivity in a great degree24. Such kind of effect is similar to the interfacial
polarization that is originated from the distinct electrical properties between filler and the matrix. It
will introduce new polarization mechanisms and relaxation processes into the whole system and
change the overall dielectric response. Moreover, the large-scale interface in layered structures
opens up new possibilities as it can be easily engineered through varying the composition,
arrangement, and thickness of each layer. For instance, Wang et.al have designed bilayer high-k
composites by graphite/polyvinylidene fuluoride (PVDF) composites with positive and negative
respectively25. The permittivity of the bilayer samples can be easily tuned in a broad range by
adjusting the filler contents and the thickness of the two layers. Likewise, in multi-layer structures
composing of alternative PVDF and carbon black/PVDF layers, gradual enhancement of
permittivity in the frequency range of 102-107 Hz can be realized by increasing the number of
layers26. The position and intensity of the multilayer structure induced loss peak between 103-104
Hz also change upon layer multiplication. In this respect, the interface in these structures realize the
accumulation and confinement of heterogeneous charges at a higher length scale, which can be
further developed to tune the dielectric functionality stably and reliably.
To utilize interfacial effects rationally, it is also important to consider the relative position of the
interface and the propagating direction of electromagnetic waves to generate effective interaction
between them. In multi-layer dielectric functional materials, the relative direction between the
layered structure and the electric field plays a decisive role on dielectric properties. According to
Teirikangas et al, in the 'horizontal structure', the distribution of the electric field is relatively
homogeneous along the interface, while in the 'vertical structure' this continuity is broken by the
interface27, resulting in distinct dielectric response. In multiferroic oxide heterostructures, vertical
interface has also been explored to manipulate the electromagnetic properties. In this context, the
hetero-interface is vertical to the substrate surface, reducing the influence of the substrate and
increasing strain tenability at the same time. Hence, it is promising for precise control of mechanical
and electromagnetic properties28. To this end, large-scale interfaces that have effective interaction
with external fields will be critical for manipulating the overall performance of dielectric functional
4
nanocomposites.
In this paper, we take inspiration from the above perspectives and introduce a vertical interphase
in carbon nanotube/silicone elastomer nanocomposites to fully explore interfacial effects and
expand the tunability of electromagnetic properties (Figure 2). The vertical interphase is composed
of nanocomposites with different polarization abilities across the interphase. The interphase can
function by accumulating charges, restraining their local distribution and inducing dielectric
response that is distinguishable from that of homogeneous materials. The proposed method has great
potential in enhancing the high frequency dielectric response of nanocomposites and enables the
full utilization of microscopic and macroscopic interfacial properties. It provides new insights for
the design and fabrication of lightweight nanocarbon microwave absorbing materials and other
microwave functional materials. By introducing vertical interphase into nanocomposites, this
research brings the investigation of interface related dielectric response to a higher level and takes
full advantage of interfacial engineering in dielectric functionality context.
2 EXPERIMENTAL SECTION
2.1 Materials
Multi-walled carbon nanotubes (outer diameter: <8 nm, inner diameter: 2-5 nm, length: 10-30
m) were obtained from Chengdu Organic Chemicals Co., Ltd, Chinese Academy of Sciences. The
nanotubes were grown by Chemical Vapor Deposition method and had a purity of 95%.
SYLGARD(R) 184 silicone elastomer kit (Dow Corning) was used as the polymer matrix of carbon
nanocomposites. Sodium nitrate (NaNO3), potassium permanganate (KMnO4), and sulfuric acid
(H2SO4) were purchased from Sinopharm Chemical Reagent Co., Ltd. Dopamine hydroxide was
supplied by Aladdin Co., Ltd. γ-Methacryloxypropyl trimethoxy silane (KH570, coupling agent)
was purchased from Adamas-beta.
2.2 Design of vertical-interphase nanocomposites
To introduce a large-scale interphase region that could effectively interact with electromagnetic
waves, vertical-interphase carbon nanocomposites are designed as follows (Figure 1). The relative
position of the sample and the incident microwave is shown in Figure 1a. The distribution of electric
5
field under TE10 mode is shown in Figure 1b. Figure 1c is the schematic description of CNT/SE
nanocomposites. Various surface and interface modification methods are applied to achieve different
polarization abilities in CNT/SE nanocomposites. A relatively broad interphase area is formed with
two kinds of premixed CNT/SE nanocomposites through flow and diffusion (Figure 1d). As such,
the relatively broad interphase region is introduced and expanded along the Z direction so that it can
fully interact with microwave (distributed in the XY plane).
Figure 2. Schematic design of vertical-interphase nanocomposites: (a) sample for waveguide
measurements and the corresponding incident direction of electromagnetic waves (b) direction and
distribution of electric field under TE10 mode (c) schematic description of CNT/polymer
nanocomposites (d) building vertical interphase in CNT nanocomposites with different CNT
modification methods
2.3 Modification strategies towards nanocomposites with different polarization ability
In order to prepare nanocomposites with different polarization abilities, several CNT surface
modification methods were chosen. Silane coupling agent (KH570) modified CNT is marked as
CACNT. As-received CNTs were first dispersed into KH570/ ethanol solution (1 wt%). The solution
was sonicated for 1 h before drying in oven under 60 ℃. Oxidization of CNT was carried out with
strong oxidants (KMnO4, H2SO4). 0.5 g CNT was mixed with 0.375 g NaNO3, 1 g KMnO4 and
dispersed in 15 ml H2SO4 under room temperature and stirred for 24 h. 50 ml H2O and 3 ml H2O2
6
were then added. The mixture was washed with deionized water and dried under 80 ℃. The product
of chemical oxidation is marked as OCNT. Further modification was achieved by mixing OCNT
with dopamine hydrochloride aqueous solution (2 g/L) and stirring for 10 h under 60 ℃. The final
product is marked as DPACNT.
2.4 Preparation of vertical-interphase nanocomposites
The solution mixing method was utilized to prepare premixed CNT/SE nanocomposites. CNTs
were dispersed in tetrahydrofuran (THF) and silicone elastomer was also dissolved in THF
simultaneously, followed by mixing in a planetary centrifugal mixer. The solvent was then
evaporated and the premixed nanocomposites were degassed for use (marked as CNT/SE,
CACNT/SE, OCNT/SE, DPACNT/SE depending on the type of nanofillers). For each vertical-
interphase nanocomposites, two kinds of premixed nanocomposites were prepared at the same time
(e.g. CNT/SE and OCNT/SE). Equal amounts of CNT/SE and OCNT/SE were poured into the mold
from two sides. The premixed nanocomposites flowed and diffused from both sides of the mold and
blended in the central area, forming an interphase region. The vertical-interphase nanocomposites
were then cured at 125 ℃. The same method was used to prepare a set of nanocomposites with
different compositions. The dimension of the mold is 22.86 mm×10.16 mm×2 mm.
2.5 Characterization
A field emission scanning electron microscope (Zeiss, Utral 55) was used for observing the
morphologies of the samples. The dispersion and distribution of CNTs were monitored by an optical
microscope (Olympus BX53M). ImageJ (an open-source software) is applied to present the
skeletonized pictures of optical images and carry out the statistical analysis of the average area of
CNT aggregates to better illustrate the structure of the interphase. The chemical structures of CNTs
and modified CNTs were characterized by Fourier transform infrared spectroscopy (FTIR,
ThermoFisher). Raman spectroscopy (DXR smart Raman spectrometer, irradiation wavelength: 532
nm) was performed for the nanocomposites. A vector network analyzer (R&S, ZNB20) was used to
measure the scattering parameters. The complex permittivity in the frequency range of 8.2-12.4 GHz
(X band) was extracted by Nicolson-Ross-Weir method. 1stopt (an optimization software, developed
by 7D-Soft High Technology Inc.) was used for curve fittings and the extraction of characteristic
7
relaxation times.
3 RESULTS AND DISCUSSION
3.1 Carbon nanocomposites with different polarization abilities
Various chemical modification methods are first explored to achieve different polarization
abilities for building vertical-interphase in carbon nanocomposites. Silane functionalization of CNT
could improve its interaction with the silicone elastomer. The siloxy group on KH570 molecules
would interact with the oxygen-containing groups on CNTs through hydrolysis and condensation,
and the vinyl group can participate in the vulcanization reaction of silicone rubber29. Oxidation is
commonly adopted to introduce oxygen functional groups on the surface and ends of CNT, which
is helpful for improving its compatibility with polymer and reducing agglomeration30, 31. These
functional groups such as carboxyl can further take part in a variety of chemical reactions and realize
secondary modification and functionalization32. Dopamine has been considered to be extremely
adhesive to various surfaces since 2007, when Lee et.al first used it for coating a wide range of
materials33. Meanwhile, the characteristic self-oxidative polymerization of dopamine makes it
especially suitable for surface modification. For instance, dopamine modification was shown to
improve the dispersion of TiO2 nanofibres in PVDF and mitigate the concentration of electric field
efficiently by forming a shell around the fillers34.
The chemical structures of unmodified and modified CNTs were investigated by FTIR
spectroscopy as shown in Figure 3a. Comparing to raw CNTs, there appear absorption peaks at 933
cm-1, 1011 cm-1, and 1300 cm-1 for CACNT that can be attributed to C=C stretch vibrations, Si-O
stretch vibrations, and C-O-C stretch vibrations from KH570 molecules29. The oxidation process
introduced the characteristic peak at 1712 cm-1, corresponding to stretch vibrations of C=O from
carboxyl groups. Meanwhile, there is a significant peak at 2350 cm-1 that can be assigned to
vibrations of hydrogen bonding, suggesting the strong interaction between the increased oxygen-
bearing groups. Further modification of dopamine is evidenced by the appearance of peaks at 1427
cm-1, 1576 cm-1 attributed to aromatic C-C and N-H stretch vibrations respectively34. The modified
nanotubes were then incorporated into silicone elastomer to fabricate nanocomposites. The Raman
spectra of the as-prepared carbon nanocomposites are presented in Figure 3b. The D bands and G
8
bands are observed for all samples at 1342 cm-1 and 1592 cm-1 respectively, which represent the
disorder-induced double-resonance and in-plane vibrations of sp2 C-C bonds31, 35, 36. The relative
intensity ratios of D peak to G peak (ID/IG) is used for estimating the degree of defects and the
destruction of graphitic integrity. According to Figure 3b, the increases of ID/IG are all observed for
OCNT/SE, CACNT/SE, and DPACNT/SE comparing to CNT/SE. Thereby, different chemical
modification methods have all influenced the vibrations of carbon atoms and increased the degree
of defects on CNT, which would result in nanocomposites with distinct microstructures and
interfacial features.
Figure 3. (a) FTIR spectra of CNTs modified by different methods (the appearance of characteristic
peaks are indicated by blue arrows): (i) raw CNTs, (ii) CNTs modified with KH570, (iii) oxidized
CNTs, (iv) CNTs further functionalized with dopamine; (b) Raman spectra of CNT/silicone
elastomer nanocomposites with different modification methods (the positions of D peak and G peak
are indicated by the vertical lines): (i) CNT/SE, (ii) CACNT/SE, (iii) OCNT/SE, (iv) DPACNT/SE
The SEM images of CNT/silicone elastomer nanocomposites with different modification methods
are shown in Figure 4. From Figure 4a, e, i, it is observed that there exist many large-size
9
agglomerates (about 10-20 µm) in CNT/SE. In comparison, silane functionalization has improved
the dispersion of nanotubes, decreasing the agglomerates size to below 10 µm (Figure 4b, f, j). In
both OCNT/SE and DPCNT/SE nanocomposites, the dispersions of CNTs are greatly enhanced.
The agglomerates appear to be dotted-like in Figure 4c-d, and the distributions of CNTs are very
uniform. Under higher magnification, it could be seen that DPACNTs are more homogeneously
dispersed in the matrix than OCNTs (Figure 4k, i). Thereby, the relative dispersion abilities of the
nanotubes in silicone elastomer are as follows: DPACNT>OCNT>CACNT>CNT. Silane
functionalization is not as effective as oxidation in reducing agglomeration because there are limited
active sites on raw CNT. Meanwhile, wrapping OCNT with polydopamine has further improved the
dispersion degree. Thus, four distinct dispersion states are achieved in CNT nanocomposites by
diverse chemical modification methods and different degrees of modification.
Figure 4. SEM images of CNT/silicone elastomer nanocomposites (f=0.5 vol%): (a, e, h) CNT/SE;
(b, f, i) CACNT/SE; (c, g, h) OCNT/SE; (d, h, i) DPACNT/SE (CNT agglomerates are indicated by
red circles)
The differences in the chemical structures of nanofillers and the microstructures of
10
nanocomposites have resulted in distinct dielectric properties, which are illustrated as follows. In
Figure 5, the imaginary permittivities of the above CNT nanocomposites are plotted as a function
of filler loadings under selected frequencies. It is observed in the first place that the imaginary
permittivities for all the nanocomposites are all very small and show little dependence on frequency.
For each loading, various modification methods have resulted in different degrees of decrease in
permittivity. Generally, the for each filler loading follows the trend of CNT/SE> CACNT/SE>
OCNT/SE> DPACNT/SE as a result of different modification and functionalization mechanisms. These
chemical treatments influenced not only the conductivity of CNTs but also their dispersion ability
and interfacial interaction with polymer matrix.
Figure 5. Imaginary permittivities of CNT/silicone elastomer nanocomposites with different
modification methods and filler loadings: (a) 8.2 GHz; (b) 9.2 GHz; (c) 10.2 GHz; (d)10.2 GHz
To be specific, while silane functionalization partially affected the original structure of CNTs
through chemical bonding, the decrease of permittivity is relatively moderate due to limited active
sites on raw CNTs. When CNTs were treated with strong oxidants, the structure of nanotubes was
severely destroyed and the conductivity was greatly compromised. Thereby, the of OCNT/SE
11
and DPACNT/SE decrease to below 0.1 (f=0.25 vol%). It should be noticed that when OCNTs were
wrapped with polydopamine, the of nanocomposites further drops because polydopamine
functions as a protective shell restricting the dielectric loss at the interface. At this point, the
permittivity does not change with filler loading anymore, indicating that the dielectric response
becomes very insignificant. Except for the original structure of CNTs, the dispersion ability also
plays a role in influencing the dielectric properties. When the functional fillers are better dispersed,
there are less local conductive networks and the overall dielectric response can also be decreased.
As such, the conductivity and the dispersion of nanofillers are the primary causes of the ultimate
differences in dielectric properties. Real permittivities of the samples as function of filler loading
are presented in Figure S1 (Supporting Information), in which a similar pattern is observed, again
confirming the effectiveness of these methods in adjusting dielectric properties. To this end, a set of
nanocomposites with distinct polarization abilities are designed and fabricated.
3.2 Vertical interphase induced dielectric relaxation
CNT/SE nanocomposites developed
in 3.1 are used for building vertical-interphase
nanocomposites. The vertical-interphase nanocomposites are fabricated by diffusion and partial
mixing of premixed nanocomposites and are named correspondingly. For example, the vertical-
interphase nanocomposite constituted by CNT/SE and OCNT/SE is marked as CNT-OCNT/SE. The
original optical microscope images of CNT-OCNT/SE are shown in Figure S2 (Supporting
Information). The left side of the sample (OCNT/SE rich) is obviously much better dispersed than
the right side (CNT/SE rich) due to the difference in dispersion states. The mixed region could be
observed from the skeletonized image of the central area (Figure 5a). Statistical analyses on the
average area of agglomerates are carried out with ImageJ for the left (L), central (C), and right (R)
area of the sample respectively as a relative indication of dispersion degree37, 38. Each region is
divided into 6 sections and the results are presented in Figure 6b, in which L and R represents the
overall estimation of the left region and the right region. In comparison, the average area of
agglomerates in C1, C2, and C3 are similar to that of the L region. The dispersion is much worse in
C4, C5, and C6, indicating that these sections are mainly constituted by CNT/SE. There is sharp
increase in the average area of CNT agglomerates from C3 to C4, suggesting that the interphase is
12
mainly introduced around these two sections. While there is no clear boundary, the interphase is
supposed to be a relatively large area with a torturous path forming along the Y and Z direction.
Figure 6. Optical microscope images analyses of CNT-OCNT/SE nanocomposites (f=0.5 vol%): (a)
skeletonized optical image of the interphase region (marked by the dashed oval); (b) statistical
analysis of the average area of CNT aggregates on each section (the purple arrow indicates the
abrupt change of agglomerates area); (c, d): frequency dependence of complex permittivity of CNT
nanocomposites with or without in-built vertical interphase: (c) real part; (d)imaginary part;
Frequency dependence of permittivity is used for analyzing the dielectric properties of vertical-
interphase nanocomposites. For CNT-OCNT/SE nanocomposites, the complex dielectric spectra are
exhibited in Figure 6c-d. The and of CNT/SE and OCNT/SE are relatively stable over the
whole X band and there is a distinct discrepancy in their polarization abilities. The of CNT-
OCNT/SE is in between that of CNT/SE and OCNT/SE (Figure 6c). It first remains steady as the
frequency increases, and then starts to drop continuously over 10.8-11.5 GHz. Simultaneously, a
remarkable relaxation peak appears in the imaginary spectra of CNT-OCNT/SE over this frequency
range (Figure 6d). The peak value of CNT-OCNT/SE reaches 1.2 at 11.2 GHz, which is significant
comparing to CNT/SE (0.5) and OCNT/SE (0.15) at the same frequency. Such variation trend in
13
complex dielectric spectra of CNT-OCNT/SE represents a typical dielectric relaxation process. It
can be logically concluded that this process is closely correlated with the introduction of vertical
interphase, which induces strong and effective interaction between the nanocomposite and
electromagnetic waves.
3.3 Tunable microwave dielectric response enabled by vertical interphase
To further reveal the effect of introducing vertical interphase into nanocomposites, we built
several vertical-interphase nanocomposite systems with varied compositions and studied their
dielectric response under external fields. These systems are labeled as 1, 2, 3, 4 and 5 according to
the discrepancy in polarization ability of the two regions across vertical interphase (). It should
be noted that the label of each system is only a comparative description marking the difference in
dielectric properties of the premixed nanocomposites. Considering that should be frequency-
dependent, we could not specify it as an absolute value. Figure 7a presents the imaginary dielectric
spectra of nanocomposites with a single type of modified or raw CNTs. These nanocomposites are
overall homogeneous and their imaginary permittivities are relatively stable over X band. We first
fabricated vertical-interphase nanocomposites that are composed by OCNT/SE and DPACNT/SE,
which have the smallest discrepancy in polarization ability. It can be seen in Figure 7b that the
dielectric response of OCNT-DPACNT/SE is very weak (<0.2) and there is no obvious change in
its dielectric spectra comparing to homogeneous nanocomposites. As increases, there appears a
weak relaxation peak in the dielectric spectra of CNT-CACNT/SE (at 10.7 GHz) as shown in Figure
7c. For OCNT-CACNT/SE, a stronger relaxation peak is observed at 11.3 GHz, accompanied by
the peak value of increasing to around 0.8. Meanwhile, the blue shift of relaxation peak can be
attributed to improved dispersion state and reduced relaxation time with the addition of OCNT. In
this context, the of vertical-interphase nanocomposites is the key to the appearance and
enhancement of the characteristic relaxation peak. We then enlarged and built system 4 and 5, in
which significant changes of dielectric spectra are observed and the peak values of increase
remarkably to 1.2 and 1.1 (Figure 7d). Simultaneously, these relaxation peaks at around 11.2 GHz
become extremely sharp. It is thereby indicated that the relaxation process is enhanced greatly with
increasing , again validating that it is originated from the difference in dielectric properties
14
between the two regions across interphase. It is also noteworthy that the position and the intensity
of the relaxation peak remain almost unchanged due to limited increase in from CNT-OCNT/SE
to CNT-DPACNT/SE.
Figure 7. Frequency dependence of
imaginary permittivity of CNT/silicone elastomer
nanocomposites (f=0.5 vol%) for studying vertical interphase (8.2-12.4 GHz): (a) carbon
nanocomposites with a single type of modified or raw CNTs ( marked by the purple double
headed arrows); (b) OCNT-DPACNT/SE vertical interphase nanocomposites; (c) CNT-CACNT/SE
and OCNT-CACNT/SE nanocomposites;
(d) CNT-OCNT/SE
and CNT-DPACNT/SE
nanocomposites (the
insets:
the schematic descriptions of different vertical-interphase
nanocomposite systems, and various premixed nanocomposites are marked by different colors)
We further plot the frequency dependence of imaginary permittivity of vertical-interphase
nanocomposites (f=0.5 vol%) in Figure 8a to investigate the intrinsic mechanism of dielectric
enhancement and the new relaxation process enabled by vertical interphase. It is clear that the
relaxation peak varies significantly with increasing (from 1 to 5). The Cole-Cole plots of these
samples are exhibited in Figure 8b correspondingly. Firstly, the permittivity of OCNT-DPACNT/SE
is very small and its dispersion feature is not very obvious. Secondly, the shape of the Cole-Cole
15
plots for CNT-CACNT/SE and OCNT-CACNT/SE change to be partially arc-like, reflecting the
appearance of a typical relaxation process. Finally, when the is raised to 4 or 5, the radius of the
arc increases astoundingly to almost circular, indicating that the experiences fierce changes over
the investigated frequency range. Based on the above analyses, this characteristic relaxation process
is originated from the polarization induced by the interphase, as shown in the inset of Figure 8b. To
be specific, as there exist differences in the conductivity and permittivity between the left and right
region of the sample, the vertical interphase becomes the centre of charge accumulation and
restriction. A new polarization mechanism is induced and the relaxation occurs under alternating
electric fields. While each vertical-interphase system has a different , its ability to restrict and
accumulate charges are varied, enabling tunable dielectric response through controlling the
compositions.
Figure 8. (a) Frequency dependence of imaginary permittivity of CNT/silicone elastomer
nanocomposites (f=0.5 vol%) with in-built vertical interphase (8.2-12.4 GHz); (b) Cole-Cole plots
and schematic description of the polarization mechanism (the inset) of the investigated
nanocomposites
16
3.4 Comprehensive assessment of the vertical-interphase nanocomposites
According to the analyses above, the vertical interphase-induced relaxation relies on the
discrepancy in polarization ability inside the nanocomposites. The filler loading of nanocomposites
has an intrinsic influence on dielectric properties, so we focus on revealing the effect of vertical
interphase in CNT nanocomposites as a function of different filler loading in this section. A
relatively low filler loading (0.25 vol%) and a high one (1 vol%) are chosen to present a
comprehensive assessment of the dielectric functionality in vertical-interphase nanocomposites. The
frequency dependence of imaginary permittivity of CNT nanocomposites (f=0.25 vol%) is presented
in Figure 9a. As the overall dielectric response is very weak (< 0.3), we only display the dielectric
spectra of CNT-OCNT/SE and CNT-DPACNT/SE here. They are marked as system 4* and 5*
correspondingly. The of vertical-interphase nanocomposites are very similar to that of
homogeneous nanocomposites, and weak relaxation peaks appear at 11.7 GHz.
Figure 9. Frequency dependence of
imaginary permittivity of CNT/silicone elastomer
nanocomposites (with different filler contents) for studying vertical interphase ( marked by the
purple double headed arrows): (a) f=0.25 vol%; (b) f=1 vol%
17
The imaginary dielectric spectra of CNT nanocomposites (f=1 vol%) are shown in Figure 9b.
Four systems are built and labeled as 2#, 3#, 4#, and 5#. It can be observed that there appears a
significant dielectric relaxation peak (between 9.5-11.5 GHz) for each vertical-interphase sample
under this filler loading, suggesting that the introduction of vertical interphase induces a new
relaxation process. Unlike homogeneous nanocomposites, the ε" of these samples vary remarkably
with frequency. It is noteworthy that the position and intensity of the relaxation peak are distinct for
different systems. The relaxation peak shifts to higher frequency when the overall dispersion of
CNT is improved, contributing to a shorter relaxation time. Since the dispersion degrees of the
nanocomposites are as follows: DPACNT/SE> OCNT/SE> CACNT/SE> CNT/SE, blue shifts of
the relaxation peak are observed in 3# (to 11.2 GHz) and 5# (to 11.3 GHz) comparing to 2# and 4#
respectively. To better illustrate this relaxation process, the real dielectric spectra are fitted by Cole-
Cole equation24, 39, which takes the form of:
𝜀 = 𝜀∞ +
𝜀𝑠−𝜀∞
1+(𝑖𝜔𝜏0)1−𝛼 (1)
in which is the permittivity at high frequency limit, s is the static permittivity, 0 represents
the characteristic relaxation time and represents the dispersion of relaxation time. The fitted curves
are shown in Figure S3 (Supporting Information). The characteristic relaxation times of 2#, 3#, 4#
and 5# are extracted and displayed in Table 1. The characteristic relaxation times of 3# and 5# are
smaller than that of 2# and 4#, corresponding to the above analyses very well. Interestingly, although
better dispersion is expected in 4# comparing to 2#, the relaxation time is not decreased, which is
the result of stronger relaxation achieved by the enhanced ability of the interphase to restrict charges.
In other words, the relaxation times are not only dependent on the overall dispersion of the sample,
but are related to the interphase-induced relaxation process itself. In addition, the relaxation peak
becomes more significant with increasing (from 2# to 3#), accompanied by the peak value
increasing from 0.8 to 1.0. CNT-OCNT/SE shows the strongest relaxation peak with the ε" reaching
1.8 at 10.2 GHz. When the is further increased from 4# to 5#, the peak value of ε" remains almost
stable and only the blue shift of the relaxation peak is observed. In this case, the charge accumulation
at the vertical interphase reaches its limit at 4#.
18
Table 1. Fitted characteristic relaxation time of vertical-interphase nanocomposites
(f=1 vol%)
Sample
2#: CNT-CACNT/SE
3#: OCNT-CACNT/SE
4#: CNT-OCNT/SE
5#: CNT-DPACNT/SE
0
1.57e-11
1.42e-11
1.58e-11
1.46e-11
Comparative analyses are carried out to further display the ability of vertical interphase in tuning
microwave dielectric response. The imaginary dielectric spectra of CNT-OCNT/SE and CNT-
DPACNT/SE nanocomposites are plotted as a function of filler volume fraction in Figure 10a. With
increasing filler loading, the nanofillers are worse dispersed, leading to the red shift of the
characteristic relaxation peak. Simultaneously, the dielectric response is overall enhanced, which
means that for the same composition, there is a larger discrepancy in the polarization ability across
the interphase. As a result, the dielectric relaxation becomes even more predominant in
nanocomposites with a higher filler loading. This phenomenon is proved by the Cole-Cole plots
depicted in Figure 10b, in which circular curves with larger radius are observed upon increasing
filler volume fraction. In this sense, it provides a new dimension to adjust the dielectric response
within vertical-interphase nanocomposites.
19
Figure 10. (a) Frequency dependence of imaginary permittivity of CNT-OCNT/SE and CNT-
DPACNT/SE nanocomposites with different filler contents; (b) Cole-Cole plots of CNT-OCNT/SE
and CNT-DPACNT/SE nanocomposites
The idea of engineering vertical-interphase carbon nanocomposites falls rightly into the scope of
plainified composites, as in this research, only interface/interphase engineering methods are adopted
to tailor dielectric functionalities. The vertical interphase enabled plainified nanoƒcomposites can
be used as single-layer microwave absorbers or dielectric layers regulating the functionality of
multilayer absorbers. They could be widely applied as composing layers of novel multilayer
microwave absorbers to adjust the working frequency. For instance, vertical-interphase
nanocomposites with differences in relaxation peak can be assembled to achieve broad-band
microwave absorption (Figure 11) without introducing additional functional fillers and
compromising the lightweight characteristic of the structure. Also, such a multilayer structure can
maintain the structural integrity as all layers are essentially of the same CNT nanocomposite nature.
Thus, it would be highly efficient and of significant engineering implication.
20
Figure 11. Multilayer assembly of vertical-interphase nanocomposites towards efficient microwave
absorption: (a) Configuration of multilayer microwave absorber composed by vertical-interphase
nanocomposites (L1, L2, and L3 represent layers with differences in the position of characteristic
relaxation peak); (b) schematic representation of imaginary dielectric spectra for each layer; (c-d)
schematic description of frequency dependence of reflection loss for each layer (c) and the
multilayer assembly (d)
4 CONCLUSIONS
To summarize, this study presents a promising strategy to utilize interfacial effects to manipulate
the dielectric response of nanocomposites facilely and effectively. On top of the typical filler-matrix
interface, this study proposes the essential concept of plainified composites, in which only interface
engineering methods are utilized to improve material properties. In the context of tuning dielectric
functionality, a large-scale vertical interphase is rationally designed and successfully introduced into
carbon nanocomposites in this study. The in-built vertical interphase becomes the centre of charge
accumulation and confinement, triggering a strong and characteristic relaxation process in X band.
The
interphase-induced relaxation changes
the dielectric dispersion pattern of carbon
nanocomposites remarkably. Meanwhile, the intensity and position of the relaxation peak are highly
tunable by adjusting the filler loading and the discrepancy in polarization ability across the
interphase (). With large volume fraction of functional fillers or increasing , the relaxation
process is greatly enhanced. From this perspective, it is convenient and efficient to tune the
microwave dielectric properties of carbon nanocomposites. It is anticipated that stronger dielectric
21
response will be achieved with higher loading of nanofillers, larger enabled by other systems, or
multiple interphase introduced in nanocomposites with a considerable in future work. While the
typical filler-matrix interfacial effect is sometimes insignificant due to small filler volume fraction,
the ability to tailor dielectric response through vertical interphase without increasing the weight of
materials is of great significance for microwave functional materials40. The insights provided here
can be applied to reconfigurable microwave absorption to easily manipulate the microwave
absorbing frequency. Overall, this study reveals an intrinsic dependence of dielectric functionality
on interfacial properties at a higher length scale, thereby opening up new possibilities for the
designing and engineering of microwave functional polymer nanocomposites and devices.
Corresponding Author
*Email: [email protected]
REFERENCES
(1) Coleman, J. N.; Khan, U.; Blau, W. J.; Gun'ko, Y. K., Small but strong: A review of the
mechanical properties of carbon nanotube -- polymer composites. Carbon 2006, 44 (9), 1624-1652.
(2) Nan, C.-W.; Liu, G.; Lin, Y.; Li, M., Interface effect on thermal conductivity of carbon
nanotube composites. Appl. Phys. Lett. 2004, 85 (16), 3549-3551.
(3) Roy, M.; Nelson, J.; MacCrone, R.; Schadler, L. S.; Reed, C.; Keefe, R., Polymer
nanocomposite dielectrics-the role of the interface. IEEE Trans. Dielectr. Electr. Insul. 2005, 12 (4),
629-643.
(4) Yuan, J.-K.; Yao, S.-H.; Dang, Z.-M.; Sylvestre, A.; Genestoux, M.; Bai, J., Giant
Dielectric Permittivity Nanocomposites: Realizing True Potential of Pristine Carbon Nanotubes in
Polyvinylidene Fluoride Matrix through an Enhanced Interfacial Interaction. The Journal of
Physical Chemistry C 2011, 115 (13), 5515-5521.
(5) Wang, Y.; Cui, J.; Yuan, Q.; Niu, Y.; Bai, Y.; Wang, H., Significantly Enhanced
Breakdown Strength and Energy Density in Sandwich-Structured Barium Titanate/Poly(vinylidene
fluoride) Nanocomposites. Adv Mater 2015, 27 (42), 6658-63.
(6) Sanida, A.; Stavropoulos, S. G.; Speliotis, T.; Psarras, G. C., Development, characterization,
energy storage and interface dielectric properties in SrFe 12 O 19 / epoxy nanocomposites. Polymer
2017, 120, 73-81.
(7) Holt, A. P.; Bocharova, V.; Cheng, S.; Kisliuk, A. M.; White, B. T.; Saito, T.; Uhrig,
D.; Mahalik, J. P.; Kumar, R.; Imel, A. E.; Etampawala, T.; Martin, H.; Sikes, N.;
Sumpter, B. G.; Dadmun, M. D.; Sokolov, A. P., Controlling Interfacial Dynamics: Covalent
Bonding versus Physical Adsorption in Polymer Nanocomposites. ACS Nano 2016, 10 (7), 6843-
6852.
(8) Lewis, T. J., Interfaces are the dominant feature of dielectrics at the nanometric level. IEEE
Trans. Dielectr. Electr. Insul. 2004, 11 (5), 739-753.
(9) Nelson, J. K., Dielectric Polymer Nanocomposites. Springer: New York, 2010.
(10) Wang, Y.; Zhang, Y.; Zhao, H.; Li, X.; Huang, Y.; Schadler, L. S.; Chen, W.;
Brinson, L. C., Identifying interphase properties in polymer nanocomposites using adaptive
optimization. Compos. Sci. Technol. 2018, 162, 146-155.
(11) Dang, Z.-M.; Yuan, J.-K.; Zha, J.-W.; Hu, P.-H.; Wang, D.-R.; Cheng, Z.-Y., High-
permittivity polymer nanocomposites: Influence of interface on dielectric properties. Journal of
22
Advanced Dielectrics 2013, 3 (03), 1330004.
(12) Sousa, A. A.; Pereira, T. A. S.; Chaves, A.; de Sousa, J. S.; Farias, G. A., Interfacial
confinement in core-shell nanowires due to high dielectric mismatch. Appl. Phys. Lett. 2012, 100
(21), 211601.
(13) Maxwell, J., Electricity and Magnetism, vol. 1, Clarendon. Oxford: 1892.
(14) Wagner, K., The after effect in dielectrics. Arch. Electrotech 1914, 2 (378), e394.
(15) Sillars, R., J. Inst. Electr. Eng. 1937.
(16) Kang, D.; Wang, G.; Huang, Y.; Jiang, P.; Huang, X., Decorating TiO2 Nanowires with
BaTiO3 Nanoparticles: A New Approach Leading to Substantially Enhanced Energy Storage
Capability of High-k Polymer Nanocomposites. ACS Appl Mater Interfaces 2018, 10 (4), 4077-
4085.
(17) Feng, Y.; Deng, Q.; Peng, C.; Hu, J.; Li, Y.; Wu, Q.; Xu, Z., An ultrahigh discharged
energy density achieved in an inhomogeneous PVDF dielectric composite filled with 2D MXene
nanosheets via interface engineering. Journal of Materials Chemistry C 2018, 6 (48), 13283-13292.
(18) Watts, P.; Ponnampalam, D.; Hsu, W.; Barnes, A.; Chambers, B., The complex
permittivity of multi-walled carbon nanotube -- polystyrene composite films in X-band. Chem. Phys.
Lett. 2003, 378 (5-6), 609-614.
(19) Zhang, X.; Li, B.-W.; Dong, L.; Liu, H.; Chen, W.; Shen, Y.; Nan, C.-W., Superior
Energy Storage Performances of Polymer Nanocomposites via Modification of Filler/Polymer
Interfaces. Advanced Materials Interfaces 2018, 5 (11), 1800096.
(20) Sharma, M.; Madras, G.; Bose, S., Size dependent structural relaxations and dielectric
properties induced by surface functionalized MWNTs in poly (vinylidene fluoride)/poly (methyl
methacrylate) blends. PCCP 2014, 16 (6), 2693-2704.
(21) Li, Y.; Qin, F.; Estevez, D.; Wang, H.; Peng, H.-X., Interface Probing by Dielectric
Frequency Dispersion in Carbon Nanocomposites. Scientific reports 2018, 8 (1), 14547.
(22) Li, X.; Lu, K., Improving sustainability with simpler alloys. Science 2019, 364 (6442), 733-
734.
(23) Joyce, D. M.; Ouchen, F.; Grote, J. G., Re-engineering the Polymer Capacitor, Layer by Layer.
Advanced Energy Materials 2016, 6 (15), 1600676.
(24) Wang, B.; Liu, L.; Huang, L.; Chi, L.; Liang, G.; Yuan, L.; Gu, A., Fabrication and
origin of high-k carbon nanotube/epoxy composites with low dielectric loss through layer-by-layer
casting technique. Carbon 2015, 85, 28-37.
(25) Wang, J.; Shi, Z.; Mao, F.; Chen, S.; Wang, X., Bilayer Polymer Metacomposites
Containing Negative Permittivity Layer for New High-k Materials. ACS Appl Mater Interfaces 2017,
9 (2), 1793-1800.
(26) Zhu, J.; Shen, J.; Guo, S.; Sue, H.-J., Confined distribution of conductive particles in
polyvinylidene fluoride-based multilayered dielectrics: Toward high permittivity and breakdown
strength. Carbon 2015, 84, 355-364.
(27) Teirikangas, M.; Juuti, J.; Jantunen, H., Multilayer BST-COC Composite with Enhanced
High Frequency Dielectric Properties. Ferroelectrics 2009, 387 (1), 210-215.
(28) Li, W.; Zhang, W.; Wang, L.; Gu, J.; Chen, A.; Zhao, R.; Liang, Y.; Guo, H.;
Tang, R.; Wang, C.; Jin, K.; Wang, H.; Yang, H., Vertical Interface Induced Dielectric
Relaxation in Nanocomposite (BaTiO3)1-x:(Sm2O3)x Thin Films. Scientific reports 2015, 5, 11335.
(29) Yang, D.; Ruan, M.; Huang, S.; Wu, Y.; Li, S.; Wang, H.; Ao, X.; Liang, Y.;
Guo, W.; Zhang, L., Dopamine and silane functionalized barium titanate with improved
electromechanical properties for silicone dielectric elastomers. RSC Advances 2016, 6 (93), 90172-
90183.
(30) Kim, S. W.; Kim, T.; Kim, Y. S.; Choi, H. S.; Lim, H. J.; Yang, S. J.; Park, C. R.,
Surface modifications for the effective dispersion of carbon nanotubes in solvents and polymers.
Carbon 2012, 50 (1), 3-33.
(31) Datsyuk, V.; Kalyva, M.; Papagelis, K.; Parthenios, J.; Tasis, D.; Siokou, A.;
Kallitsis, I.; Galiotis, C., Chemical oxidation of multiwalled carbon nanotubes. Carbon 2008, 46 (6),
833-840.
(32) Spitalsky, Z.; Tasis, D.; Papagelis, K.; Galiotis, C., Carbon nanotube -- polymer composites:
Chemistry, processing, mechanical and electrical properties. Prog. Polym. Sci. 2010, 35 (3), 357-
401.
(33) Lee, H.; Dellatore, S. M.; Miller, W. M.; Messersmith, P. B., Mussel-inspired surface
23
in
induced by dopamine-modified BaTiO3 nanofibers
chemistry for multifunctional coatings. science 2007, 318 (5849), 426-430.
(34) Song, Y.; Shen, Y.; Liu, H.; Lin, Y.; Li, M.; Nan, C.-W., Enhanced dielectric and
ferroelectric properties
flexible
poly(vinylidene fluoride-trifluoroethylene) nanocomposites. J. Mater. Chem. 2012, 22 (16), 8063.
(35) Mu, M.; Osswald, S.; Gogotsi, Y.; Winey, K. I., An in situ Raman spectroscopy study of
stress transfer between carbon nanotubes and polymer. Nanotechnology 2009, 20 (33), 335703.
(36) Dresselhaus, M. S.; Dresselhaus, G.; Saito, R.; Jorio, A., Raman spectroscopy of carbon
nanotubes. Phys. Rep. 2005, 409 (2), 47-99.
(37) Shah, K.; Vasileva, D.; Karadaghy, A.; Zustiak, S., Development and characterization of
polyethylene glycol -- carbon nanotube hydrogel composite. Journal of Materials Chemistry B 2015,
3 (40), 7950-7962.
(38) Glaskova, T.; Zarrelli, M.; Aniskevich, A.; Giordano, M.; Trinkler, L.; Berzina, B.,
Quantitative optical analysis of filler dispersion degree in MWCNT -- epoxy nanocomposite. Compos.
Sci. Technol. 2012, 72 (4), 477-481.
(39) Cole, K. S.; Cole, R. H., Dispersion and absorption in dielectrics I. Alternating current
characteristics. The Journal of chemical physics 1941, 9 (4), 341-351.
(40) Qin, F.; Brosseau, C., A review and analysis of microwave absorption in polymer composites
filled with carbonaceous particles. J. Appl. Phys. 2012, 111 (6), 061301.
24
|
1810.05678 | 1 | 1810 | 2018-07-10T16:06:03 | Dynamic Response of Tunable Phononic Crystals and New Homogenization Approaches in Magnetoactive Composites | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | This research investigates dynamic response of tunable periodic structures and homogenization methods in magnetoelastic composites (MECs). The research on tunable periodic structures is focused on the design, modeling and understanding of wave propagation phenomena and the dynamic response of smart phononic crystals. High amplitude wrinkle formation is employed to study a one-dimensional phononic crystal slab consists of a thin film bonded to a thick compliant substrate. Buckling induced surface instability generates a wrinkly structure triggered by a compressive strain. It is demonstrated that surface periodic pattern and the corresponding large deformation can control elastic wave propagation in the low thickness composite slab. Simulation results show that the periodic wrinkly structure can be used as a smart phononic crystal which can switch band diagrams of the structure in a transformative manner. A magnetoactive phononic crystal is proposed which its dynamic properties are controlled by combined effects of large deformations and an applied magnetic field. Finite deformations and magnetic induction influence phononic characteristics of the periodic structure through geometrical pattern transformation and material properties. A magnetoelastic energy function is proposed to develop constitutive laws considering large deformations and magnetic induction in the periodic structure. Analytical and finite element methods are utilized to compute dispersion relation and band structure of the phononic crystal for different cases of deformation and magnetic loadings. It is demonstrated that magnetic induction not only controls the band diagram of the structure but also has a strong effect on preferential directions of wave propagation. Moreover, a thermally controlled phononic crystal is designed using ligaments of bi-materials in the structure. | physics.app-ph | physics | University of Nevada, Reno
Dynamic Response of Tunable Phononic Crystals and New Homogenization
Approaches in Magnetoactive Composites
A dissertation submitted in partial fulfillment of the
requirements for the degree of Doctor of Philosophy in
Mechanical Engineering
By
Alireza Bayat
Dissertation Advisor:
Dr. Faramarz Gordaninejad
August, 2015
ProQuest Number: 3724119
All rights reserved
INFORMATION TO ALL USERS
The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
Published by ProQuest LLC (2015). Copyright of the Dissertation is held by the Author.
ProQuest 3724119
This work is protected against unauthorized copying under Title 17, United States Code
Microform Edition © ProQuest LLC.
All rights reserved.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
Copyright by Alireza Bayat 2015
All Rights Reserved
THE GRADUATE SCHOOL
We recommend that the dissertation
prepared under our supervision by
ALIREZA BAYAT
Entitled
Dynamic Response Of Tunable Phononic Crystals
And
New Homogenization Approaches In Magnetoactive Composites
be accepted in partial fulfillment of the
requirements for the degree of
DOCTOR OF PHILOSOPHY
Faramarz Gordaninejad, Ph.D., Advisor
Emil Geiger, Ph.D., Committee Member
Wanliang Shan, Ph.D., Committee Member
Mark A. Pinsky, Ph.D., Committee Member
Ahmad M. Itani, Ph.D., Graduate School Representative
David W. Zeh, Ph. D., Dean, Graduate School
August, 2015
i
ABSTRACT
This research investigates dynamic response of tunable periodic structures and
homogenization methods in magnetoelastic composites (MECs). The research on tunable
periodic structures is focused on the design, modeling and understanding of wave
propagation phenomena and the dynamic response of smart phononic crystals. High-
amplitude wrinkle formation is employed to study a one-dimensional phononic crystal
slab consists of a thin film bonded to a thick compliant substrate. Buckling induced
surface instability generates a wrinkly structure triggered by a compressive strain. It is
demonstrated that surface periodic pattern and the corresponding large deformation can
control elastic wave propagation in the low thickness composite slab. Simulation results
show that the periodic wrinkly structure can be used as a smart phononic crystal which
can switch band diagrams of the structure in a transformative manner. A magnetoactive
phononic crystal is proposed which its dynamic properties are controlled by combined
effects of large deformations and an applied magnetic field. Finite deformations and
magnetic induction influence phononic characteristics of the periodic structure through
geometrical pattern transformation and material properties. A magnetoelastic energy
function is proposed to develop constitutive laws considering large deformations and
magnetic induction in the periodic structure. Analytical and finite element methods are
utilized to compute dispersion relation and band structure of the phononic crystal for
different cases of deformation and magnetic loadings. It is demonstrated that magnetic
induction not only controls the band diagram of the structure but also has a strong effect
ii
on preferential directions of wave propagation. Moreover, a thermally controlled
phononic crystal is designed using ligaments of bi-materials in the structure.
Temperature difference is used to generate large deformations and affect the elastic
moduli tensor of the structure. Phononic characteristics of the proposed structure are
controlled by the applied temperature difference. The effect of temperature difference on
the band diagrams of the structure is investigated.
Homogenization methods in periodic and random MECs are also investigated. A finite
element method (FEM)-based homogenization approach is presented to simulate the
nonlinear behavior of MECs under a macroscopic deformation and an external magnetic
field. Micro-scale formulation is employed on a characteristic volume element, taking
into account periodic boundary conditions. Periodic homogenization method is utilized
to compute macroscopic properties of the MEC at different mechanical and magnetic
loadings. A new efficient numerical scheme is used to develop the magnetoelastic
tangent moduli tensors. In addition, the effective response of a random MEC under
applied magnetic fields and large deformations is computed. The focus is on the spatially
random distribution of identically circular inclusions inside a soft homogenous matrix. A
FEM-based averaging process is combined with Monte-Carlo method to generate
ensembles of randomly distributed MECs. The ensemble is utilized as a statistical
volume element in a scale-dependent statistical algorithm to approach the desired
characteristic volume element size.
iii
ACKNOWLEDGEMENTS
I would like to express my deepest gratitude to my academic advisor Professor Faramarz
Gordaninejad, who supported me, advised me and provided me the opportunity to do the
research I liked to do. I would like to thank his personal assistance during my stay in
Reno and for helping me in adapting to the new environment. I look forward to many
years of friendship and professional collaboration with him.
I was fortunate to enjoy the assistance and friendship of the members of Composite and
Intelligent Materials Laboratory in the past few years. I am grateful to Xiaojie, Majid,
Nima, Nich, Sevki, Ling and all those who have helped my professional as well as
personal life.
I would like to thank my Ph.D. defense committee members for their time and
consideration: Prof. Mark Pinsky, Prof. Ahmad Itani, Prof. Emil Geiger and Prof.
Wanliang Shan.
Thank you to my lovely wife, Fatemeh, for being by my side throughout this process!
Thank you to my family, who helped me, encouraged me, supported me and gave me
cheer during all steps of my life!
TABLE OF CONTENTS
ABSTRACT………………………………………………………………………
ACKNOWLEDGEMENT………………………………………………………
TABLE OF CONTENTS……………………………………………………….
LIST OF FIGURES …………………………………………………………….
Chapter One-Introduction…………………………………………………………
1.1 Overview…………………………………………………………………...
1.2 Magnetoactive composites………………………………………………….
1.3 Homogenization……………………………………………………………
1.4 Phononic crystals………………………………………………………
1.5 Pattern change in soft structures……………………………………………
1.6 Notations………………………………………………………………
1.7 Motivations………………………………………………………………
1.8 Research contributions………………………………………………………
1.9 Structure of the dissertation………………………………………………
Chapter Two- Switching Band-Gaps of a Phononic Crystal Slab by Surface
Instability………………………………………………………………………
2.1 Modeling………………………………………………………………
2.2 Results and discussion……………………………………………………
2.3 Summary and conclusion ………………………………………………
Chapter Three- Dynamic Response of a Tunable Phononic Crystal under Applied
iv
i
iii
iv
vi
1
1
1
4
7
16
21
22
23
24
26
27
37
44
Mechanical and Magnetic Loadings……………………………………………
46
3.1 Governing equations…………………………………………………
3.2 Buckling analysis………………………………………………………
3.3 Wave propagation analysis……………………………………………
3.4 Results and discussion…………………………………………………
3.5 Summary and conclusion………………………………………………
Chapter Four-A Thermally Tunable Phononic Crystal…………………………
4.1 Modeling………………………………………………………………
4.2 Results and discussion…………………………………………………
4.3 Summary and conclusion………………………………………………
Chapter Five- A New Computational Method for Overall Tangent Moduli of a Soft
Magnetoactive Composite Using Periodic Homogenization……………
5.1 Modeling ……………………………………………………………………
5.2 Results and discussion……………………………………………………
v
48
53
55
59
73
75
75
79
83
84
85
94
5.3 Computation of macroscopic moduli tensors………………………………
104
5.4 Summary and conclusion………………………………………………
108
Chapter Three- Homogenization Approach in Random Magnetoelastic Composites
110
6.1 Modeling………………………………………………………………
6.2 Random homogenization framework…………………………………
6.3 Results and discussion…………………………………………………
6.4 Summary and conclusion………………………………………………
Chapter Seven- Concluding Remarks And Future Work……………………
111
119
123
134
135
References………………………………………………………………………
137
LIST OF FIGURES
Figure 1. SEM image of MREs with (a) random distribution of magnetic particles,
(b) particles aligned along the applied magnetic field used for curing, within the
silicon matrix [11] …………………………………………………………………..
Figure 2. An example of multiscale modeling for heterogeneous structures [52]…..
Figure 3. Examples of (a) a 1D [85], (b) a 2D [86] and (c) a 3D [87] PnC and
vi
3
5
corresponding unit cells chosen for wave propagation study……………………….
8
Figure 4. Application of PnCs as (a) acoustic lens [95] and (b) acoustic
waveguides [96] …………………………………………………………………….
10
Figure 5. A tunable 3D PnC and corresponding unit cells in undeformed and
deformed states, designed to manipulate elastic waves [99]. ………………………
11
Figure 6. A typical periodic structure with corresponding unit cell, direct lattice
vectors and reciprocal lattice vectors. ……………………………………………..
13
Figure 7. Examples of first Brillouin zone and irreducible Brillouin zone for 1D,
2D and 3D typical lattices [100]. …………………………………………………...
15
Figure 8. A thermally modulated reconfigurable structure [101, 105]……………..
17
Figure 9. Transverse contraction of the periodic lattice under compressive strain
due to microstructural buckling [106] ………………………………………………
18
Figure 10. Pattern change in a granular structure under compressive loading [107].
19
vii
Figure 11. Surface wrinkling of a 66nm thick film of gold bonded on a PDMS
substrate at different levels of applied compressive strain from at left
column to at the right side [126] …………………………………………..
20
Figure 12. (a) Schematic of a bilayer slab and corresponding boundary conditions,
and (b) the deformed wrinkly structure resulted from the compressive stretch, ε,
due to surface instability (not to scale). ……………………………………………
28
Figure 13. First and second mode shapes for a typical L-H structure………………
29
Figure 14. Effect of different imperfection weight (w) on the bifurcation. Post-
buckling does not occur for w = 0.01% and less. ………………………………….
30
Figure 15. Simulation results for plane strain wrinkling of (a) L-H structure, with
, , and (b) H-H
structure, with , , , at different levels of
compressive strain. Dashed rectangles show unit cells selected for wave
propagation analysis. ………………………………………………………………
32
Figure 16. Variation of the dimensionless amplitude, (left axis) and periodicity
of the wrinkles, (right axis) vs. applied strain ε, for (a) L-H and (b) H-H
composite slabs, computed form numerical simulations. ………………………….
33
Figure 17. Numerical results for band diagram plots for (a-d) L-H and (e-h) H-H
composite structure at different levels of the compressive strain, . Band-gaps are
shown in shaded regions. Dispersion curves are computed for eigenfrequencies
larger than . …………………………………………………………………..
38
Figure 18. Numerical results for band diagram plots for (a-d) L-H and (e-h) H-H
periodic structure for different heights of composite slabs. Band-gaps are shown
in shaded regions. Dispersion curves are computed for eigenfrequencies larger
viii
than . ………………………………………………………………………
40
Figure 19. Evolution of Phononic band-gap vs. density contrast ratio, for (a)
L-H and (b) H-H composites. Phononic band-gap vs. shear modulus contrast ratio,
for (c) L-H and (d) H-H composites. ……………………………………..
41
Figure 20. Line graphs showing the strain energy density along the middle surface
of the L-H composite slab compressed by ε=0.4 at different normalized
frequencies. Contour plots represent the vertical component of harmonic
perturbation. Modeling parameters are identical to the ones used for band diagram
investigation in Figure 17(d). ………………………………………………………
43
Figure 21. Schematic of the computational approach proceeded to study the
dynamic response of the structure. ………………………………………………
47
Figure 22. Reference (undeformed), deformed and superimposed incrementally
deformed configurations……………………………………………………………
48
Figure 23. The FEM results for pattern transformation in the phononic crystal due
to buckling instability under the applied deformation. (a) An undeformed 8×8
periodic structure, (b) the first buckling mode of the structure subjected to
uniaxially compressive stretch in horizontal direction, and (c) the first buckling
eigenmode of the enlarged unit cell. ………………………………………………..
55
Figure 24. (a) The RUC selected for the uniaxial compression, (b) the
corresponding deformed geometry for uniaxial stretch , and (c) reciprocal
lattice's unit cell selected for the wave propagation study. Irreducible Brilluine
zone
is
shown
in
the
region
bounded
by
ix
………………………………………………………………………………………
56
Figure 25. In-plane band diagrams for uniaxially compressive stretch at
(a) No magnetic induction, (b) 1.0T, (c) 2.0T, and (d) 3.0T unidirectionally
applied magnetic induction. PBGs are shown in shaded regions. The second and
third modes represented by red and blue bands, respectively, are selected for
directionality analysis. ……………………………………………………………..
63
Figure 26. Normalized iso-frequency contours associated with the second modes
for uniaxially compressive stretch at (a) No magnetic induction, (b)
0.25T, (c) 1.0T, and (d) 2.0T unidirectionally applied magnetic induction. The
contours are associated with the red band in Figure 25. …………………………
66
Figure 27. Normalized iso-frequency contours associated with the third modes for
uniaxially compressive stretch at (a) No magnetic induction, (b) 0.25T,
(c) 1.0T, and (d) 2.0T unidirectionally applied magnetic induction. The contours
are associated with the blue band in Figure 25…………………………………….
67
Figure 28. Directionality plots associated with the second modes for uniaxially
compressive stretch at (a) No magnetic induction, (b) 0.25T, (c) 1.0T,
and (d) 2.0T unidirectionally applied magnetic induction. Frequency and direction
is plotted in radial and angular directions, respectively. …………………………
68
Figure 29. Directionality plots associated with the third modes for uniaxially
compressive stretch at (a) No magnetic induction, (b) 0.25T, (c) 1.0T,
x
and (d) 2.0T unidirectionally applied magnetic induction Frequency and direction
is plotted in radial and angular directions, respectively. …………………………
29
Figure 30. (a) The RUC selected for the uniaxial compression, (b) the
corresponding deformed geometry for , (c) The RUC selected for equally
biaxial compression (d) the corresponding deformed geometry for , (e)
reciprocal lattice's unit cell selected for the wave propagation study. Irreducible
Brilluine zone is shown in the region bounded by ………………….
71
Figure 31. PBG vs. applied magnetic induction in y direction, for the square array
PnC in: (a) uniaxially compressive stretch and (b) in biaxially
compressive stretch ………………………………………………………..
72
Figure 32. A finite sample of the proposed periodic structure at (a) undeformed
state, (b) deformed by thermal actuation, . …………………………..
76
Figure 33. (a) The primitive unit cell, (b) deformed unit cell, under
and (c) First and irreducible Brillouin zone chosen for the wave propagation
analysis. ……………………………………………………………………………
79
Figure 34. (a) Band diagram of undeformed structure and (b) the first 6 mode-
shapes at 𝛤. Psudo-gaps and complete gaps are shown in color region in the band
diagram. …………………………………………………………………………….
81
Figure 35. (a) Band diagram of deformed structure and (b) the first 6 mode-shapes
at 𝛤. Psudo-gaps and complete gaps are shown in color region in the band
diagram. …………………………………………………………………………
82
Figure 36. (a) The homogenized body and corresponding boundary decomposition
in Lagrangian configuration, (b) corresponding CVE, attached to , selected for
homogenization study including a circular permeable particle inside a soft square
xi
matrix, (c) the deformed CVE in Eulerian configuration. ……………………….
90
Figure 37. Contour plots of normalized microscopic distribution of for
uniaxial loading case at (a) 0.8, (b) 1.1 and (c) 1.4 stretches. Normalized
microscopic distribution of for uniaxial loading case at (d) 0.8, (e) 1.1 and (f)
1.4 stretches. ……………………………………………………………………..
97
Figure 38. Contour plots of normalized microscopic distribution of for pure
shear loading case at (a) -0.3, (b) 0.1 and (c) 0.2 shear stretches. Normalized
microscopic distribution of for pure shear loading case at (d) -0.3, (e) 0.1 and
(f) 0.2 shear stretches. The red boundary lines represent the in undeformed
configuration. All deformed plots are scaled to 1. ………………………………
98
Figure 39. Contour plots of normalized microscopic distribution of for
equally biaxial loading case at (a) , (b) 5 and (c)
magnetic field. Normalized microscopic distribution of for equally biaxial
loading case at (d) , (e) 5 and (f) magnetic
field……..…………………………………………………………………………
99
Figure 40. Contour plots showing the periodic boundary conditions on the
fluctuation fields. The normalized microscopic distribution of for pure shear
loading case at (a) -0.3, (b) 0.1 and (c) 0.2 shear stretches. Normalized
microscopic distribution of for pure shear loading case at (d) -0.3, (e) 0.1 and
(f) 0.2 shear stretches. The red boundary lines represent the in undeformed
configuration. All deformed plots are scaled to 1. The Figure corresponds to
Figure 38 which sketches the deformed unit cells based on of microscopic position
xii
vector. ……………………………………………………………………………
101
Figure 41. Parametric analysis resulting from numerical study representing (a)
vs. stretch (b) vs. stretch for uniaxial loading case, (c) vs. stretch (d)
vs. stretch for equally biaxial loading case and (e) vs. stretch (f) vs.
stretch for pure shear loading case at different levels of macroscopic magnetic
field. ……………………………………………………………………………….
103
Figure 42. (a,b) Components of homogenized mechanical moduli tensor, (c,d)
Components of the homogenized coupling magneto-mechanical moduli tensor…..
107
Figure 43. (a) The homogenized body and corresponding boundary decomposition
in Lagrangian configuration, (b) corresponding SVE, attached to , selected for
statistical analysis including randomly distributed circular permeable particle
inside a soft matrix (c) the effective medium modeled using homogenization
approach. ………………………………………………………………………….
116
Figure 44. Schematic of the statistical algorithm used in the study. (a) selection of
random meso-scale ensembles of the heterogeneous MEC for a typical test
window size, (b) increasing the size of the test window and number of particles
towards the convergence window and (c) five typical realizations for All
ensembles are generated for ……………………
120
Figure 45. Computational algorithm used in the statistical approach……………..
123
Figure 46. Moduli component , results from the statistical analysis for (a) LD-
126
xiii
BC and (b) PF-BC versus number of simulations………………………………….
Figure 47. Moduli component , results from the statistical analysis for (a) LD-
BC and (b) PF-BC versus number of simulations………………………………
127
Figure 48. Coefficient of variation for (a) and (b) versus scale
factor, …………………………………………………………………………..
129
Figure 49. Average values of (a) and (b) versus scale factor, ….
131
Figure 50. FEM results for component of stress tensor for SVEs with (a)
, with LD-BC, (b) , with PF-BC, (c) , with LD-BC and
(d) , with PF-BC. All plots are normalized by their average value of
component. ………………………………………………………………………
132
Figure 51. FEM results for component of stress tensor for SVEs with (a)
, with LD-BC, (b) , with PF-BC, (c) , with LD-BC and
(d) , with PF-BC. All plots are normalized by their average value of
component. ……………………………………………………………………..
133
1
CHAPTER ONE
Introduction
1.1 Overview
This dissertation consists of two main topics: 1) wave propagation in tunable phononic
crystals, and 2) computational homogenization in periodic and random magnetoactive
composites. New tunable periodic structures are proposed and analyzed to manipulate
elastic wave propagation. Moreover, periodic and random homogenization approaches
are addressed in magnetoelasticity framework. New numerical methods are presented to
develop effective properties of magnetoelastic composites
(MEC) utilizing
homogenization approach. Based on the contents of the dissertation, the present Chapter
provides brief introductions and backgrounds on different topics that are presented in
following chapters.
1.2 Magnetoelastic composites
MECs are materials consisting of micro-size permeable particles randomly distributed in
an elastomeric matrix. Upon the application of an external magnetic field, the MEC
undergoes finite deformations due to the interaction of magnetizable particles and an
2
elastomeric matrix [1-10]. In addition to the nonlinear geometrical deformation, the
mechanical properties of the MEC can be controlled by the applied magnetic field.
Extensive works have been devoted to develop mathematical formulation of MECs [3-
28].
Recently, researches have been focused on media with electro-magneto-mechanical
coupling interactions. Specifically, magnetorheological elastomers (MREs) have been
developed as field-controllable composite materials capable of significant changes in
their mechanical properties. MREs consist of micron size magnetic particles suspended
in an elastic/hyperelastic elastomeric matrix that can endure finite deformations upon the
application of an external magnetic field [15-23].
Curing the MRE by magnetic field causes the particles to form a chain-like column
resulting in a transversely isotropic structure. Otherwise, a random distribution of
magnetic particles can be considered as an isotropic MRE. The deformation in MREs is
induced by the combined magnetic and mechanical interactions between iron particles as
well as the hyperelastic matrix. The applied magnetic field is an added parameter to
control the mechanical properties of the MRE [3, 14]. Figure 1 shows an SEM image of
MREs with random distribution of magnetic particles, and particles aligned along the
applied magnetic field used for curing, within the silicon matrix [11].
(a)
(b)
3
Figure 1. SEM images of MREs with (a) random distribution of magnetic particles, (b) particles
aligned along the applied magnetic field used for curing, within the silicon matrix [11].
Theoretical modeling of MREs has been extensively studied based on
the
electrodynamics of continuous media [2, 15]. Several studies have been done to develop
basic constitutive equations for nonlinear magnetoelastic relations [17-27]. Some efforts
have been focused on the effect of magnetic fields and finite deformations on the wave
propagation in MREs [20-28]. The effect of initial stress on the propagation of Rayleigh,
Love and Stoneley waves in a magnetoelastic medium has also been studied in [24-26].
Recently, general theoretical framework for the analysis of incremental motions
superimposed on a state of finite deformations subjected to the electromagnetic field has
been studied [20-22].
1.3 Homogenization
4
Different homogenization approaches have been utilized to investigate the effective
characteristics of MECs for random and periodic microstructures. Homogenization has
been used as a tool to study the overall response of the composite and heterogeneous
materials presumed to be statistically homogenous. Various techniques have been carried
out to simulate the heterogeneous materials as an equivalent effective medium [29-36].
Homogenization offers an averaging process on the microstructure's characteristic
volume element (CVE) to compute the effective behavior of the composite.
For periodic structures, CVE is known as the smallest spatial microstructural unit of the
structure (i.e. a unit cell). For random heterogeneous materials, the challenge is to find
an appropriate CVE to extract the overall properties of the composite material. For this
purpose, a statistical analysis is required to find the CVE with randomly dispersed
inclusions. The effective behavior of composite material depends on properties of
microstructure's constituents. Exact mathematical procedures as well as numerical
methods have been developed to study overall mechanical properties of adaptive
composites such as magnetoelastic and electroelastic media [37-44].
Different
types of micro-scale boundary conditions have been studied
in
the
homogenization process [45-51]. Recently, a finite element method (FEM) based
homogenization approach is used to study different mechanical and magnetic material
parameters and boundary conditions on the overall response of the structure [46-47].
5
Homogenization has been widely used as a powerful tool in multi-scale modeling and
FE2 analysis of heterogeneous structures [50]. An algorithm for computation of
consistent tangent moduli of electroelastic composites is presented in references [47-49,
51].
An example of multi-scale modeling in pure elasticity is shown in Figure 2. First, a
CVE is chosen for the study. The CVE is loaded by the macroscopic deformation by
incremental steps. Then, the effective constitutive laws and homogenized moduli tensors
are driven to characterize the macroscopic properties of the continuum. The process is
repeated in an iterative way until the convergence is achieved [52].
Figure 2. An example of multi-scale modeling for heterogeneous structures [52].
Many heterogeneous structures in engineering applications are identified by random
distribution of their constituents within a matrix [52-61]. For brevity, "random
composites" is referred to the heterogeneous structures with randomly dispersed
6
inclusions. Likewise, "periodic composites" is used for the periodic distribution of the
inclusions in a composite structure. Examples of random composites cover wide range of
polymer, ceramic or metal matrix surrounding inclusions of different materials. The
inclusions are mainly made of high material properties' contrast with respect to matrix to
enhance the effective properties of material. Recent studies on the heterogeneous
structures focuses on evaluation of overall response of composites. Overall properties are
also referred to effective or homogenized properties in the literature [52-61]. The
challenge in study of mechanics of random composites is to identify a characteristic
volume element (CVE) and appropriate boundary conditions. Since the classical CVE
concept in the homogenization of periodic composites is not valid for random media.
Several techniques have studied the homogenization of non-periodic media [52-69]. A
statistical based scale-dependent homogenization approach
is used
to simulate
mechanical behavior of a two phase particulate composite. A numerical convergence
scheme is used to define an algorithm for determining the CVE size [53]. Monte-Carlo
method is widely used to generate statistical ensembles of random composites [53-57].
Specific details are addressed in the determination of CVE size for linear elastic and
nonlinear regimes under different boundary conditions [57]. A scale-dependent
homogenization approach is presented for the study of linear and nonlinear thermoelastic
random composites which uses variational methods to define hierarchy bounds on
constitutive laws. This provides a statistical limit for determining CVE size [59-66].
Moreover, the mismatch between the material properties of composites constituents has
shown to have strong effect on the CVE size [56-60]. In a recent study, a multi-scale
strategy is proposed for nonlinear thermoelastic analysis of random composites with
7
temperature dependent properties [66]. Random magnetoelastic composites have also
been studied in [67-70] where different inclusion shapes, material properties and particle
concentration are used in the calculations. A closed-form expression is derived for the
effective constitutive laws through defining a homogenized energy function which takes
into account the concentration, aspect ratio of fibers cross-section and their distribution
[70].
1.4 Phononic crystals
Electromagnetic and elastic wave propagation through periodic structures has been
extensively studied in recent years. A phononic crystal (PnC) is a periodic structure
consists of different materials in an elastic medium designed to interact with elastic
waves. PnCs are structures with one-dimensional (1D), two-dimensional (2D) or three-
dimensional (3D) periodicity in their geometry and material properties. These structures
have interesting applications, such as, frequency filters, beam splitters, sound or vibration
isolators, acoustic mirrors and elastic waveguides [71-84].
Periodic structures can be designed to hinder wave propagation in some range of
frequencies; namely, band-gaps. Existence of band-gaps is due to the periodic change of
density and speed of wave in the structure [77-80]. Unit cell is the smallest structural
unit of the periodic material which is spatially assembled together in 1D, 2D or 3D to
build up the structure. Basically, the wave propagation characteristics in PnCs depend on
unit cell's geometry and material properties (i.e. density, stiffness and bulk modulus).
8
While different approaches have been used to study wave propagation phenomena in
periodic structures, the most common methods are plane wave expansion methods, finite
element methods, and finite-difference time domain methods to explore band diagrams
[72, 75, 77, 78]. Figure 3 shows examples of 1D, 2D and 3D PnCs and corresponding
unit cells [85-87].
(a)
(b)
(c)
Figure 3. Examples of (a) a 1D [85], (b) a 2D [86] and (c) a 3D [87] PnC and corresponding unit
cells chosen for wave propagation study.
Dynamic response of the structure can be controlled through the geometry and elastic
properties of the unit cell. Adaptive periodic structures promise the ability to design
tunable properties that can control band-gaps upon the application of an external
stimulus. The elastic band structure of PnCs made of piezoelectric materials has been
investigated through plane wave expansion method [83]. The phononic band-gap (PBG)
9
and acoustic characteristics of the structure have been shown to significantly shift due to
various electric and magnetic fields [84].
The band structure characteristics of lattices of three different geometries consisting of
piezoelectric and piezomagnetic media were studied to understand the effects of different
magnetic and electric fields through the plane wave expansion method [88-90]. It has
been shown that tunable PnCs can be employed as a transmission switch for elastic waves
when magnetic field passes a threshold [88]. Only limited work has been carried out on
the band diagram calculations of MRE PnCs. In all previous work on MRE PnCs
constant coefficients are presumed as magnetoelastic coefficients [91-94]. Parametric
study of the effects of different geometry and configuration on the band structure of a set
of parallel square-section columns regularly distributed in air has been investigated as a
tunable periodic structure [90]. Periodic structures have also been studied theoretically
and experimentally for super resolution at a narrow band of frequencies using flat lenses
made of PnCs [95]. Figure 4 shows some applications of PnCs as waveguides and
acoustic lenses [95-96].
10
(a)
(b)
Figure 4. Application of PnCs as (a) acoustic lens [95] and (b) acoustic waveguides [96].
Deletion of existing band-gaps and creation of new band-gaps were reported due to the
effects of large deformation and the microstructural elastic instability in periodic
elastomers subjected to different types of mechanical loadings. When deformation
reaches a critical value, strong revolution happens in the band structure due to
transformation in structural pattern resulting in a tunable behavior of the structure in
terms of the applied mechanical loading [97-98]. Figure 5 shows a tunable periodic
structure which its phononic characteristics is tuned by the applied macroscopic
deformation [99].
11
Figure 5. A tunable 3D PnC and corresponding unit cells in undeformed and deformed states,
designed to manipulate elastic waves [99].
1.4.1 Wave propagation analysis using Bloch theorem
Elastic wave propagation in PnCs is studied based on the fundamentals of solid state
physics in lattices utilizing Bloch theorem [74]. In PnCs, the desired quantities in wave
propagation are extracted from displacement and stress fields. In this work, a 2D infinite
size periodic lattice in plane, is considered. The periodic lattice is characterized by
a unit cell and direct lattice vectors and . Reciprocal lattice of any periodic structure
is defined by reciprocal lattice vectors and :
‖ ‖
‖ ‖
‖ ‖
12
Hence, the position of each point, Q, of the lattice -with respect to the reference unit cell-
can be expressed as
where are integers. The displacement of a point Q in the lattice satisfies the
periodic condition:
Bloch theorem states that the displacement of each point Q follows the Bloch theorem:
where k is the Bloch wavenumber. Comparing the Bloch statement with the periodicity
condition requires;
Since (where is the Kronecker delta) this statement is satisfied when k
is represented in terms of reciprocal lattice vectors:
Clearly:
.
13
Figure 6 shows a typical periodic structure with corresponding unit cell, direct lattice
vectors and reciprocal lattice vectors [74, 97-99].
𝑥
𝑥
Q
b
2
a
2
a1
b
1
Figure 6. A typical periodic structure with corresponding unit cell, direct lattice vectors and
reciprocal lattice vectors.
Bloch theorem states that the spatial field in each unit cell of the direct lattice presents the
same distribution and does not depend on the cell's location. This fact reduces the
problem of studying an infinite number of cells to one of considering only a single unit
cell by applying appropriate boundary conditions. Several techniques are demonstrated
on the wave propagation analysis in periodic structures. Mathematical methods such as
plane wave expansion method have been used for simple geometry lattices while finite
difference time domain method and FEM based solutions are usually used to study
complex microstructures [71-99]. In FEM based approach, the unit cell is meshed and
14
Bloch boundary conditions are applied on the opposite boundaries of the unit cell [77-
78]. The wave propagation in the unit cell is described through the discretized equation
of motion and by considering the cell interaction with the neighboring cells. Substituting
Bloch solution in the wave equation results in an eigenvalue problem whose eigenvalues
are the frequency of the solution. Wavenumber is a periodic function of the wavevector,
k in the reciprocal lattice.
Hence, the dispersion relations are obtained by investigating the variation of the
eigenfrequency, versus wavenumber, k, over a single period in reciprocal lattice. This
requires identifying the single unit cell in the reciprocal lattice called first Brillouin zone.
The dispersion relations are extracted from the eigenvalue problem where the
wavenumber is swept on the boundaries of the first Brillouin zone. It has been shown
that the calculation domain may further be reduced by taking the advantage of the
symmetry of the first Brillouin zone. The reduced domain is referred as irreducible
Brillouin zone (IBZ) [74-78]. Some examples of first Brillouin zone and IBZs are
depicted in Figure 7 for some typical lattices [100].
Once the Brillouin zone is identified, the wavenumber, k, is swept on the boundaries of
the IBZ and the corresponding eigenfrequencies are computed for each k. Band diagrams
are then plotted as versus k dispersion relations to analyze the band-gaps of the
structure.
15
Figure 7. Examples of first Brillouin zone and irreducible Brillouin zone for 1D, 2D and 3D
typical lattices [100].
1.5 Pattern change in soft structures
16
One of the concepts that have been utilized in the design of the PnCs in this dissertation
is the pattern change in the soft structures. Recently, interests have been attracted to
study the behavior of reconfigurable structures [101]. Once the loading condition is
vanished the structure tends to return to the initial state in a reversible manner. Advances
in reconfigurable materials have led to the next generation of adaptive and actuating
materials. Various parameters can be employed to modulate the shape and geometry of
the materials as a tunable material. The structures can be loaded by different external
stimuli such as mechanical [102], electrical [103], magnetic [104] and thermal [105]
excitations. The static and dynamic response of the structure then can be controlled
through the applied stimuli. In this section, recent findings on pattern change in the
reconfigurable structures are summarized.
Figure 8 shows a flower-shaped film that is capable of self-folding and resulting in a
different enclosed shape. The hinged structure is combined with thermo-responsive self-
folding capsules which are absorbed on the polymer bilayer at elevated temperatures.
Cooling results in swelling of thermo-responsive polymer and folding of the capsules.
Heating results in releasing the cells and unfolding the capsules provides a thermally
modulated reconfigurable scheme in the structure [105].
17
Figure 8. A thermally modulated reconfigurable structure [101, 105].
In a recent study, pattern change in a soft periodic lattice has shown to have negative
Poisson's ratio effect. Buckling in the microstructure has been employed to contract the
structure in the transverse direction when it is under compressive loading. The pattern
change arises due to microstructural instability in the material. Figure 9 shows the 3D
periodic lattice in undeformed and deformed state [106]. The structure is compressed by
which demonstrates the contraction in the transverse direction.
18
Figure 9. Transverse contraction of the periodic lattice under compressive strain due to
microstructural buckling [106].
Magnetic field has been used as an external stimulus to change the pattern in
magnetoactive polymers. The structure is made of 60 wt% vinyl cobalt nanoparticles
within a PDMS matrix, under the excitation of a permanent magnet. The structure
restores its initial shape once the magnetic field is vanished provides the opportunities to
utilize as a reconfigurable structure [104].
Pattern change in elastic periodic structures has also been utilized in wave propagation
systems [98, 107, 108]. The applied deformation transforms the structure to a new
geometry which can switch the band diagrams in a reversible manner. Figure 10 shows
the effect of compressive load on a granular structure. It has been shown that the
19
phononic characteristics of the structure significantly changes under the influence of the
large deformation and stress [107].
Figure 10. Pattern change in a granular structure under compressive loading [107].
One of the interesting pattern change phenomenon occurs in bilayer structures under
compressive strain, where the contrast in the material properties of a thin layer with a
thick substrate generates wrinkles at the surface of the structure. When a critical
compressive strain is applied to an elastic thin film bonded to a compliant substrate,
surface wrinkling occurs in the thin film due to surface buckling instability. Wrinkling
can be employed as a novel methodology for creating 1D and 2D periodic micro and
nanosurfaces [109-125]. Wrinkling has potential applications due to its highly ordered
pattern, unique structure and convenient fabrication methods; from surface patterning
[118-119] and smart adhesion [120-121] to optical surfaces [122-123] and flexible
electronic devices [124]. It has been shown that the amplitude and periodicity of
20
wrinkles depend on the material properties, geometry of layers and the applied
compressive strain [110-111]. A recent study has developed a new method to fabricate
high aspect ratio (amplitude over wavelength of wrinkles) wrinkles over flat surfaces
which enhances the aspect ratio from 0.3 to 0.6 [125]. This has improved the surface
contour area from 9% to 36%, doubled the local curvature, and increased the
stretchability of wrinkly flexible electronic devices from 20% to 40%. Formation of high
aspect ratio wrinkles results in a periodic pattern of scatterers at the surface of the
structure. Figure 11 shows the surface wrinkling evolution of a bilayer composite of thin
gold film on a PDMS substrate at different levels of macroscopic strain. Different
wrinkle modes and patterns are generated by the increasing strain beyond [126].
Figure 11. Surface wrinkling of a 66nm thick film of gold bonded on a PDMS substrate at
different levels of applied compressive strain from at left column to at the right
side [126].
21
In Chapter 2 the surface wrinkling pattern change is employed to design a smart PnC slab
and analyze the wave propagation in the resultant periodic structure.
1.6 Notations
Here, the vector and tensor notations mainly used throughout this manuscript, are
summarized. The summation notation for repeated indices is used. Scalars are in regular
letters. Bold faces are used for vectors and tensors. The scalar product of two vectors A
and B is denoted by: . The terms single contraction or dot product also refers
to the scalar product. Cross product of two vectors is defined by:
where is the third-order permutation tensor.
The double contraction or scalar product of two second-order tensors C and D is denoted
by: . Single contraction is also defined on two second-order tensors as;
. Single contraction of a second order tensor C with a vector A is also
denoted by; . Double contraction and single contraction of two third-order
tensors E and F are denoted by: and , respectively.
Double contraction of a third-order tensor E and a second-order tensor C is a vector
denoted by; . Scalar product of a third-order tensor E and a vector A is a
second-order tensor represented by; . For a vector A and a second-order
tensor C cross-product is a second-order tensor defined by; .
22
For three vectors M, N and P tensor product is defined as; . Tensor
product of two vectors A and B is a second-order tensor denoted by; .
Tensor product of two second-order tensors C and D is a fourth-order tensor denoted by;
. Tensor product is also defined for a vector A and a second-order tensors
C by: which represents a third-order tensor. Also . For a
second order tensor C with components , the transpose is denoted by with the
components . The trace of C is defined by; .
1.7 Research Goals
The goal for the tunable phononic crystals research is to explore:
Novel tunable PnCs to employ in wave propagation systems.
Combined effects of magnetic field and pattern change on the wave propagation
in a tunable PnC.
Surface acoustic band-gaps in a smart PnC.
Thermally modulated pattern change effects on wave propagation in smart PnCs.
The goal for the homogenization methods in magnetoactive media is to:
Identify the nonlinear magnetoelastic moduli tensors using a cost-effective
computational method and available commercial FEM packages. Moduli tensors
are of high importance in the study of instability, effective properties and multi-
23
scale modeling of MECs. Only few existing work are reported on the calculation
of the moduli tensors which requires high mathematical computations.
Understand new computational methods for homogenized properties of MECs
with randomly dispersed permeable inclusions using FE-based homogenization
methods.
1.8 Uniqueness of this research
Surface instability is employed to design a novel smart PnC slab. Band diagram
of the resultant tunable structure is significantly transformed by the applied
compressive strain.
Combined effects of pattern change and magnetic excitation has shown significant
effects on the band diagrams of the porous structure.
Band diagrams of the magnetoelastic PnC is examined through defining
magnetoelastic energy function. As a result, nonlinear moduli tensors are
considered as functions of deformation gradient tensor and magnetic induction
vector.
The effect of magnetic field on the elastic wave directionality in the
magnetoactive PnC is studied. Magnetic field has shown to strongly affect the
preferential directions of wave propagation in magnetoactive PnCs.
A thermally tunable PnC is designed and the effect of temperature difference on
the wave propagation is investigated. Temperature difference has shown to
transform the band diagram of the resulting tunable PnC.
24
A new cost-effective algorithm for computation of nonlinear moduli tensors of
MECs is presented.
A FEM-based homogenization approach is combined with Monte-Carlo method
to evaluate overall properties of MECs with random distribution of inclusions.
1.9 Structure of the dissertation
Chapter 2 introduces a new paradigm in designing tunable PnCs using surface instability.
Buckling induced surface instability is used to generate wrinkly scatterers at the surface
of a bilayer composite slab. The slab consists of a thin and stiff film bonded on a soft
substrate. The effect of applied compressive strain on band diagrams of the PnC slab is
investigated.
In Chapter 3 the effect of combined effects of mechanical deformation and magnetic field
on dynamic response of a magnetoactive PnC is studied. First, the theoretical
formulation of magnetoactive media is reviewed. A hyperelastic magnetic energy
function is considered to develop the constitutive laws. Variational method is used to
develop the coupled governing equations. Weak expressions of the coupled governing
equations are derived to use in the FEM framework. Nonlinear moduli tensors are used
as functions of deformation gradient tensor and magnetic field. The Chapter focuses on
dynamic response of the tunable PnC under applied magneto-mechanical loadings.
25
Chapter 4 introduces a new thermally controlled PnC. Temperature difference is used to
exploit large deformations and affect the phononic characteristics of the structure. Band
diagrams are plotted to investigate the effect of temperature difference on band-gaps.
Theoretical modeling and governing equations of magnetoactive media is presented in
Chapter 5. The Chapter employs the FEM-based homogenization approach to propose a
novel and computationally cost-effective method for evaluation of homogenized tangent
moduli tensors using sensitivity analysis.
Chapter 6 details the computational approach in the study of magnetoactive media with
randomly dispersed particles. An FEM based homogenization method is combined with
Monte-Carlo approach to evaluate the static response of the structure under macroscopic
loadings. A statistical approach is presented to find the appropriate CVE size to compute
the overall properties of the structure.
Finally, concluding remarks and some recommendations for future work are presented in
Chapter 7.
CHAPTER TWO
26
Switching Band-Gaps of a Phononic Crystal Slab by Surface Instability1
High-amplitude wrinkle formation is employed to propose a one-dimensional phononic
crystal slab consists of a thin film bonded to a thick compliant substrate. Buckling
induced surface instability generates a wrinkly structure triggered by a compressive
strain. It is demonstrated that surface periodic pattern and corresponding stress can
control elastic wave propagation in the low thickness composite slab. Simulation results
show that the periodic wrinkly structure can be used as a transformative phononic crystal
which can switch band diagram of the structure in a reversible manner. Results of this
study provide opportunities for smart design of tunable switch and frequency filters at
ultrasonic and hypersonic frequency ranges.
In the present study, one of the unexplored features of wrinkle-based structure is studied;
utilizing wrinkle formation at a surface of a slab as a tunable phononic crystal (PnC) in
elastic wave propagation systems. The effect of generation and control of wrinkles on
the propagation of elastic wave in thin film/substrate slabs is investigated. It is presumed
that global buckling is prevented. Wrinkles are formed when the compressive strain
1 Results of this chapter are published in:
Bayat A and Gordaninejad F 2015 Switching band-gaps of a phononic crystal slab by surface
instability Journal of Smart Materials and Structures 24 075009.
Bayat A and Gordaninejad F 2015 A Wrinkly Phononic Crystal slab Proc. SPIE San-Diego, USA,
9438, p. 943810.
27
reaches a critical strain. The compressive strain is used as a control parameter to tune the
periodicity, amplitude and pattern of the wrinkly slab. Phononic characteristics of the
periodic wrinkly slab will be controlled through both geometry and stress.
The wavelength of uniform wrinkles is defined by (
⁄
)
, where
are the elastic modulus, shear modulus and Poisson's ratio of the film,
respectively, and the elastic modulus, shear modulus and Poisson's ratio of the
substrate, respectively. The wavelength is defined as the length of the structure divided
by number of wrinkles in the undeformed state. The periodicity in a deformed
configuration is defined by where is the applied strain beyond the
bifurcation point. The amplitude of wrinkles relates to the applied strain by
√
where is the critical strain defined by
(
⁄
)
.
2.1 Modeling
The post-bifurcation state of the structure as well as wave propagation in the deformed
structure is investigated through numerical simulations. Simulations are performed using
Finite element methods (FEM) through COMSOL Multiphysics. A 2D plane strain
model is adopted to simulate the 1D wrinkling in a bilayer structure, beyond the
bifurcation point. The schematic of the model, parameters and boundary conditions are
shown in Figure 12(a). Loading of the structure will be applied through prescribed
compressive strain, ε in the direction at the right edge. Free shear traction is imposed
on all edges. Symmetry boundary condition is applied on the left and bottom edges to
28
prevent normal displacements. The top surface has a free boundary condition. The
thickness of the substrate is chosen to be sufficiently deep compared to the film
thickness.
(a)
Free
𝑓
H
Symmetry
B.C.
Symmetry B.C.
𝜺
𝒙𝟐
Λ
𝒙𝟏
(b)
𝒫
Figure 12. (a) Schematic of a bilayer slab and corresponding boundary conditions, and (b) the
deformed wrinkly structure resulted from the compressive stretch, ε, due to surface instability
(not to scale).
Lagrangian coordinates is chosen for the study. Two different cases of film-substrate
bilayer composite slabs are considered here: a linear elastic film on a soft material
(linear-hyperelastic) and a hyperelastic film on a soft substrate (hyperelastic-hyperelastic)
hereafter will be referred as L-H and H-H bilayer slabs, respectively. A nearly
incompressible neo-Hookean model is selected for hyperelastic materials. A linear
buckling study is carried out on the bilayer structure to predict the mode shapes at
compressive strains. To deal with small amplitude signals, a linear perturbation analysis
is used for the solver to predict the mode shapes. Surface deformation modes will be a
plane strain sinusoidal deflection in plane as schematically shown in Figure
12(b). Other mode shapes mainly differ in number of the periodicity. Two typical mode
shapes are pictured in Figure 13. The first mode shape resulted from the buckling
analysis for both L-H and H-H bilayer structures have similar patterns.
29
Figure 13. First and second mode shapes for a typical L-H structure
A nonlinear stationary study is carried out on the bilayer slab of finite thickness to predict
post-buckled wrinkly shape under compressive strain. Slight sinusoidal imperfections are
imposed on the geometry of the film layer to activate the buckling mode and capture the
post-transformation shape. In buckling mechanisms, post-bifurcation states cannot be
captured for an ideal system. External perturbations are required to excite the desired
buckling modes. Imperfections can be applied through perturbations in: (1) mesh, (2)
boundary conditions, and (3) geometry of the model. In this study, imperfections are
applied through the geometry of the structure. Slight sinusoidal perturbation is imposed
on the geometry of the film layer to activate the desired buckling mode and capture the
post-transformation shape. Post-bifurcated wrinkles' shape is highly sensitive to the
applied imperfection. Figure 14 shows the effect of imperfection amplitude on the
30
bifurcation. A sensitivity analysis is conducted to find the appropriate imperfection
weight. Similar post-bifurcation modes are resulted for the imperfection weight in the
range of 0.02%-0.2%. No bifurcation occurs below 0.02%.
𝛿
,
e
d
u
t
i
l
p
m
A
d
e
z
i
l
a
m
r
o
N
Bifurcation
point
Compressive Strain, 𝜀
Figure 14. Effect of different imperfection weight (w) on the bifurcation. Post-buckling
does not occur for w = 0.01% and less.
Results for wrinkling of the L-H and H-H bilayer structures at different levels of the
applied strain are shown in Figure 15. For L-H structure, , ,
, ,
and
, are used as input parameters, where
and are bulk and shear moduli of hyperelastic substrate, respectively. For H-H
structure, , , , ,
and
are used in the model. Once the critical strain is reached, the surface initiates
31
to buckle and evolves by the increasing applied strain. The amplitude of wrinkles
continuously increases by the compressive strain. For the L-H composite slab, a
sinusoidal mode appears in the structure for different levels of In Figure 15(b), one
observes that upon the increase of , different modes of post-bifurcation appears in the H-
H slab. Once the passes 0.33, the sinusoidal mode transfers to a second mode with a
wavelength and a periodicity twice those of the first mode [113]. Further increase of the
compressive strain beyond 0.37, results in a third mode with a wavelength and a
periodicity three times as those of the first mode. It must be noted that all three post-
bifurcation modes are triggered by the initial sinusoidal imperfection in both L-H and H-
H models.
Figure 16 shows the variation of the dimensionless amplitude
(left axis) and
periodicity of wrinkles (right axis) versus the applied strain, , where is the maximum
vertical displacement of the top surface boundary, obtained from the FEM modeling.
The periodicity of wrinkles continuously decreases by the increasing strain for L-H
(Figure 16(a)) and H-H (Figure 16(b)) structures and agrees with the theoretical
estimation. Discontinuity of the periodicity graph in Figure 16(b), appears due to
periodicity-doubling and tripling at and , respectively.
32
(a) L-H Structure
(b) H-H Structure
𝜀
𝜀
𝜀
𝜀
𝜀
𝜀
Figure 15. Simulation results for plane strain wrinkling of (a) L-H structure, with
𝟎 𝟐𝟓 and (b) H-H structure, with
𝝁𝒇
𝝁𝒔
𝟏𝟐𝟓,
𝝁𝒇
𝝁𝒔
𝟏𝟐𝟓,
𝜿𝒇
𝜿𝒔
𝟑,
𝒉𝒇
𝑯
𝟎 𝟏𝟓,
𝝂𝒇 𝟎 𝟐 𝝂𝒔 𝟎 𝟒𝟗
𝒉𝒇
𝑯
selected for wave propagation analysis.
𝑯
𝜸
𝟎 𝟐𝟓, at different levels of compressive strain. Dashed rectangles show unit cells
The periodicity continues to decrease for second and third post-bifurcation modes. For
L-H structure, the threshold strain to initiate the surface buckling is , which
slightly differs from the theoretical prediction . Likewise, for H-H structure,
the critical strain is , while the theoretical critical strain predicts .
𝛿
,
e
d
u
t
i
l
p
m
A
d
e
z
i
l
a
m
r
o
N
(a) L-H structure
(b) H-H structure
𝒫
,
d
o
i
r
e
P
𝛿 --
𝒫 **
𝛿 --
𝒫 **
𝛿
,
e
d
u
t
i
l
p
m
A
d
e
z
i
l
a
m
r
o
N
33
𝒫
,
d
o
i
r
e
P
Compressive Strain, 𝜀
Compressive Strain, 𝜀
Figure 16. Variation of the dimensionless amplitude, 𝜹 (left axis) and periodicity of the
wrinkles, 𝓟 (right axis) vs. applied strain ε, for (a) L-H and (b) H-H composite slabs,
computed form numerical simulations.
Consequently, it is demonstrated that the geometry, pattern and periodicity of wrinkles
are significantly transformed by the applied compressive strain, creating opportunities to
employ the wrinkly structure as a tunable 1D PnC. The elastic wave propagation in the
periodic wrinkly structure is investigated through the FEM. Wave propagation in
periodic structures depends on the unit cell's geometry, material properties and the
lattice's periodicity. The incremental elastic wave propagation is influenced by the
transformation of wrinkles' pattern due to large deformations and stress.
The deformation of the body is defined by the deformation gradient tensor,
,
which projects a point in the material coordinates, X, to its Eulerian coordinates, x. The
constitutive law for linear elastic film at large deformation is defined by a stiffness matrix
through
. The first Piola-Kirchoff stress tensor is defined
by:
34
(1)
The hyperelastic material is characterized by a neo-Hookean type strain energy function,
as follows:
Thus, the first Piola-Kirchhoff stress tensor,
is:
(2)
(3)
where is the transpose matrix of the inverse of and . Equation governing
the incremental motions superimposed on pre-deformed structures in Lagrangian
coordinates is:
(4)
where is the incremental displacemet, is the incremental first Piola-Kirchoff stress
tensor, and is the density of each layer, hereafter is referred to as and for film and
substrate layers, respectively. The increment of first Piola-Kirchoff stress tensor,
is a function of the incremental deformation gradient tensor,
. The
incremental moduli tensor is a fourth-order tensor defined by
. For the material
model defined in Equation (2) the incremental moduli tensor is:
(5)
Where
and is the tensor product symbol. A solution of the wave
35
equation in the form of plane wave is sought, where is the
amplitude vector and is the angular frequency. Thus, the stress can be written in the
following form:
Therefore, the equation of motion is an eigenvalue problem, as follows:
where is the eigenfrequency of the system.
(6)
(7)
Wave propagation in PnCs is investigated through the application of Bloch type boundary
conditions on parallel boundaries of the unit cell; the smallest repetitive structural
element of the structure. The deformed unit cells are shown in Figure 15 for each
compressive strain. To capture the first mode of the post-transformed unit cell, a
rectangular geometry with a width of (
⁄
)
, and a height of is
chosen as the unit cell to be deformed through the compressive strain. For the second
and the third post-bifurcation modes, unit cells with widths of and are modeled,
respectively. For each unit cell, appropriate perturbations are imposed on the geometry
of top surface to excite the desired mode. The periodicity, of the lattice equals the
wavelength of the wrinkle in the deformed state. For 1D wave propagation, Bloch type
36
displacement boundary conditions are applied on the left and
the right boundaries of the unit cell, where is the Bloch wave vector and is the
distance vector between parallel boundaries. Superscripts + and -- denote corresponding
right and left opposite boundaries, respectively. It follows that the traction on the right
and the left boundaries follows the Bloch relation. The only difference appears due to the
fact that tractions on parallel boundaries are in opposite directions. The wave vector
takes values of along the irreducible Brillouin zone (IBZ), where
is a real number and
is the lattice vector of the 1D reciprocal space [74].
It is noted that in FEM solvers, once the Bloch displacement boundary conditions are
applied, the traction boundary conditions are automatically satisfied. Implementation of
the plane wave form solution in the equation of motion and FEM discretization of the
problem results in an eigenvalue problem whose stiffness matrix is a function of
deformation gradient tensor and wavenumber, . Prestressed eigenfrequency study of the
software is used to take into account the effect of deformation and prestress on the wave
propagation analysis.
This analysis is provided by the software to compute
eigenfrequencies influenced by a prior static loading. By sweeping on the boundaries
of IBZ, dispersion relations are investigated and eigenfrequencies are extracted as a
function of wavenumber, to plot band diagrams. Dispersion curves are plotted as
normalized frequency,
versus reduced wavenumber, k, where
,
√
and √
. All dispersion curves and band diagrams are computed
for the eigenfrequencies larger than 1MHz.
2.2 Results and discussion
37
Band diagrams at different levels of the applied strain are shown in Figure 17, for both L-
H and H-H structures. Band-gaps are shown in shaded regions.
and are inputs of the L-H model. Geometry
and material parameters used in the H-H model are
and
where is the wavelength of wrinkles at first bifurcation mode.
Figure 17(a-d) demonstrates that band-gaps are generated and shifted by increasing
compressive strain for L-H periodic slab. Two gaps are observed at and .
Five band-gaps are created in frequency spectrum of the structure when the strain is
increased to and . Analogously, for H-H composite, number and width
of the band-gaps are switched by increasing strain. Figure 17(g-h) shows that when the
periodicity doubling and tripling take place at and 0, respectively,
significant transformation of the wave propagation characteristics is observed. Results
show that the dispersion curves tend to be flat bands at higher strains, providing more
band-gap properties at second and third post-bifurcation modes. All results presented in
Figure 17 are computed for a constant height of the slab,
38
Figure 17. Numerical results for band diagram plots for (a-d) L-H and (e-h) H-H composite
structure at different levels of the compressive strain, 𝜺. Band-gaps are shown in shaded regions.
Dispersion curves are computed for eigenfrequencies larger than 𝟏𝑴𝑯𝒛.
39
Figure 18 illustrates the effect of height of the slab on band diagrams of the structure.
Slabs with different heights of
and 3 are modeled for both L-H and H-H
composites. Input parameters for the L-H composite are
and For the H-H model,
model.
and are inputs of the
Figure 18(a-d) shows how dynamic response of the L-H composite is influenced by
height of the substrate. Band-gaps at higher frequencies vanish when . Similar
trend is observed for H-H model, as is shown in Figure 18(e-h). For both L-H and H-H
composites, band-gaps moves to lower frequencies and the high frequency gaps vanish
for . For no band-gap appears in band diagrams. Increasing the height of
the slab results in a high contrast between the wrinkles' amplitude and thickness of the
structure, so as wrinkles barely affect the wave propagation. One observes from Figures
18(d) and 18(h) that band diagrams are most likely a bulk material rather than a periodic
structure. Red dashed line in Figures 18(d) and 18(h) is the sound line limiting the sound
cone given by the transverse phase velocity of the substrate. Two narrow surface band-
gaps are observed for both L-H and H-H structure as shown in the Figures 18(d) and
18(h), respectively.
40
Figure 18. Numerical results for band diagram plots for (a-d) L-H and (e-h) H-H periodic
structure for different heights of composite slabs. Band-gaps are shown in shaded regions.
Dispersion curves are computed for eigenfrequencies larger than 𝟏𝑴𝑯𝒛.
41
Figure 19(a-b) shows the evolution of band-gaps by the increasing density contrast ratio,
. Inputs of the L-H model are
and Inputs of H-H model are,
and
. One observes that for both L-H and H-H structures, density contrast ratio has
slight effect on band-gap properties of the structure.
Figure 19. Evolution of Phononic band-gap vs. density contrast ratio,
composites. Phononic band-gap vs. shear modulus contrast ratio,
for (a) L-H and (b) H-H
for (c) L-H and (d) H-H
composites.
42
Effects of shear modulus contrast ratio,
on band diagrams of L-H and H-H structures
are shown in Figure 19(c-d).
are inputs of the L-H model and
and
and
are input parameters of the H-H model. It is important to note that the width of
the unit cell, , depends on the input parameters and . For low shear modulus
contrast ratio, more band-gap properties are observed at high frequencies. By increasing
ratio, band-gaps tend to shift to low frequencies. At
, high frequency band-
gaps vanish and only a single band-gap remains for H-H composite structure.
Figure 20 represents examples of harmonic response of an L-H slab compressed by
at different frequencies. Prestressed frequency domain study is utilized to
capture the dynamic response of the structure on a priori deformed structure. This study
takes into account the effect of large deformations and stress on the frequency response
of the structure. Harmonic perturbation is applied on the right edge of the model. The
amplitude of the perturbation is chosen as 10% of the applied displacement. Line graphs
show the strain energy density along middle surface of the structure. Contour plots
represent in-plane vertical component of displacement resulted from a harmonic
perturbation superimposed on the deformed structure. Figures 20(a-d) illustrates the
harmonic perturbation response at and , respectively. While
Figures 20(a) and 20(c) show the wave propagation in the deformed structure, Figures
20(b) and 20(d) demonstrate that the wave is attenuated along the wrinkly structure. This
is due to the fact that and correspond to the center of band-gaps of
the composite slab, as is shown in Figure 20(d). Accordingly, frequency domain
analysis verifies the results obtained from the Bloch wave analysis on the deformed unit
43
cell of the composite slab.
(a) 𝜴 𝟎 𝟎𝟑𝟎
(b) 𝜴 𝟎 𝟎𝟒𝟎
]
3
-
m
J
[
y
t
i
s
n
e
D
y
g
r
e
n
E
n
i
a
r
t
S
]
3
-
m
(d) 𝜴 𝟎 𝟎𝟖𝟓
J
[
y
t
i
s
n
e
D
y
g
r
e
n
E
n
i
a
r
t
]
3
-
m
J
[
y
t
i
s
n
e
D
y
g
r
e
n
E
n
i
a
r
t
S
]
3
-
m
(c) 𝜴 𝟎 𝟎𝟔𝟎
J
[
y
t
i
s
n
e
D
y
g
r
e
n
E
n
i
a
r
t
S
S
Figure 20. Line graphs showing the strain energy density along the middle surface of the L-H
composite slab compressed by ε=0.4 at different normalized frequencies. Contour plots
represent the vertical component of harmonic perturbation. Modeling parameters are identical to
the ones used for band diagram investigation in Figure 17(d).
2.3 Summary and conclusion
44
The ability of high aspect ratio wrinkle formation to be utilized as a highly tunable PnC is
designed and demonstrated through FEM simulations. Two material models are
considered for the analysis; L-H and H-H composite slabs. Amplitude, period and
geometry of wrinkles are controlled by the applied compressive strain. Wave
propagation is controlled through the combined effect of large deformations and stress in
the moderately thick periodic wrinkly slab. For H-H structure, three different mode
shapes are triggered at different levels of the applied compressive strain. Band diagram
analysis demonstrates that the applied compressive strain can control and transform band-
gap properties of the designed PnC.
Conversion of post-bifurcation modes of the H-H structure has shown to have significant
effects on phononic characteristics of the wrinkly slab. Moreover, the dynamic response
of the wrinkly structure for different thickness of the slab is presented and possibility of
surface elastic band-gaps is discussed. A frequency sweep analysis is performed on a
typical L-H structure to examine the harmonic response at different frequencies. Results
of the harmonic perturbation analysis verified the band diagram analysis.
The proposed structure demonstrates the possibility of creating large elastic band-gaps by
the applied compressive stretch. The designed PnC can function as a 1D pillar based PnC
[127] with capability of tuning the phononic band-gaps. Due to wide range of fabrication
methods and growing interest in micro and nano-wrinkle structures [109-112], results of
this study can be employed in developing micro and nano-scale tunable acoustic switches
[128], acoustic filters [128], elastic isolators and acoustic sensors [128-129] to be utilized
45
in ultrasonic and hypersonic applications. Moreover, the wrinkly tunable PnC can be
employed in developing PnC-based high quality factor micro-mechanical resonators
[131] which has high applications in optomechanical systems [132], sensing and
communication devices [130-132].
46
CHAPTER THREE
Dynamic Response of a Tunable Phononic Crystal
under Applied Mechanical and Magnetic Loadings1
Dynamic response of a tunable PnC consisting of a porous hyperelastic magnetoelastic
elastomer subjected to a macroscopic deformation and an external magnetic field is
theoretically investigated. Finite deformations and magnetic induction influence
phononic characteristics of
the periodic structure
through geometrical pattern
transformation and material properties. A magnetoelastic energy function is proposed to
develop constitutive laws considering large deformations and magnetic induction in the
periodic structure. Analytical and finite element methods are utilized to compute
dispersion relation and band structure of the PnC for different cases of deformation and
magnetic loadings. It is demonstrated that magnetic induction not only controls the band
diagram of the structure but also has strong effect on preferential directions of wave
propagation.
In this study, a porous periodic structure of a soft magnetoelastic medium is designed to
control wave propagation characteristics via the combined effects of finite deformations
1 Results of this chapter are published in:
Bayat A and Gordaninejad F 2015 Band-gap of a Soft Magnetorheological Phononic Crystal
ASME Journal of Vibrations and Acoustics 137 011013.
Bayat A and Gordaninejad F 2015 Dynamic Response of a Tunable Phononic Crystal under
Applied Mechanical and Magnetic Loadings Journal of Smart Materials and Structures 24,
065027.
Bayat A and Gordaninejad F 2014 A Magnetically Field-Controllable Phononic Crystal Proc.
SPIE, San-Diego, USA, 9057, p. 905713.
47
and magnetic fields. Figure 21 illustrates the solution procedure. A soft matrix provides
the reversible pattern transformation of the structure due to large deformations. The
magnetic field contributes to the change in mechanical properties of the structure. In
addition, constitutive relations for incremental motions superimposed on a predeformed
structure is developed following the approaches used in Ref. [19-22]. Moreover, wave
propagation analysis is performed and the variational formulation for magnetoelastic
wave equations considering the Bloch boundary conditions is developed. Finite element
(FE) methods are used to solve the coupling magnetoelastic equations and results are
presented as band diagrams as well as isofrequency contours at different cases of
loadings. The nonlinear FE solver, COMSOL Multiphysics 4.4 is utilized for numerical
simulations.
Figure 21. Schematic of the computational approach proceeded to study the dynamic response of
the structure.
3.1 Governing equations
48
The continuum is considered as a magnetoelastic hyperelastic body that is initially in an
undeformed state, denoted by with boundary as the reference configuration. When
the body is subjected to time-dependent magnetic and mechanical loadings, it deforms.
Thus, the region occupied by the continuum , with boundary , at a given time t is
the deformed configuration. Let and be the position vectors of the material particle at
reference and deformed configuration respectively, where and is
called the deformation mapping. In this study, a multi-scale approach is followed to
consider the geometrical nonlinearity in the deformed structure as well as a linear wave
propagation analysis in the deformed structure. Figure 22 illustrates the reference,
deformed and incrementally deformed configurations.
Figure 22. Reference (undeformed), deformed and superimposed incrementally deformed
configurations.
The continuum first undergoes large deformations due to the macroscopic deformation
gradient tensor and magnetic induction from to to take into account the pattern
49
change in the structure. Next, the incremental wave motion superimposed on the
deformed body is studied at configuration.
The deformation gradient tensor is defined as , where is
gradient operator with
respect
to material coordinates .
The notations
are used for differential operators in Lagrangian coordinate , and
are used as the corresponding operators in Eulerian coordinates with
respect to . Also, with . In Eulerian form, the magnetic field and
magnetic induction vectors are denoted by and , respectively.
It is assumed that the magnetic field is stationary and the non-conducting magnetoelastic
material is initially at the static configuration and subjected to only magnetic and
mechanical interactions. Thus, are independent of time and equations of
magnetostatics can be written, as follows:
, ,
(1)
In the absence of distributed charges or current density, the electric field and the
displacement vector can be neglected. Let us assign and for the
Lagrangian counterpart of the magnetic field and the magnetic induction vectors, (defined
in the reference configuration ) respectively, where is the transpose of the
deformation tensor. Using the standard kinematic identities: and
, Equation (1) can be written in the Lagrangian form, as follows:
, ,
(2)
50
The equation of motion governing a continuum in the presence of electromagnetic and
body forces are,
, in
(3)
where is the total Cauchy stress tensor which incorporates electromagnetic body forces,
is the density, and f is the mechanical body force density per unit mass. In Lagrangian
form, the equation of motion is,
, in
(4)
where T is the total nominal stress tensor and is the reference mass density, which are
related to their Eulerian counterparts by:
(5)
Here, a nonlinear magnetoelastic hyperelastic material is considered with total energy
density as a function of deformation tensor and magnetic induction vector,
defined per unit volume in . For a compressible material, the constitutive relations for
total nominal stress and magnetic field in the Lagrangian coordinates are:
, in
The Eulerian counterparts of Equation (6) are:
, in
(6)
(7)
For an isotropic magnetoelastic material, is an isotropic function of two tensors
(right Cauchy -- Green deformation tensor) and which can be expressed
in terms of six independent invariants, as follows:
51
[ ] ,
(8)
are principal invariants of and are invariants which are
function of . Thus, the total nominal stress and the Lagrangian magnetic field can be
expressed, as follows:
∑
, ∑
,
(9)
are the derivatives of the energy function with respect to independent
where
invariants.
Assuming that the deformation and the applied magnetic field in are
priori known, an incremental displacement is superimposed on the deformed body in the
form of , which causes both the magnetic field and deformation
undergo incremental changes within the material. In the following, a superimposed dot
, represents an infinitesimal change in the quantity concerned. The new configuration
is denoted by . Considering and as increments in independent variables and ,
by considering increments of constitutive Equations (8), one can express the linearized
form of these equations, as follows:
,
(10)
where and and are incremental magnetoelastic moduli
52
tensors defined by:
,
(11)
and are fourth-, third- and second-order tensors, with corresponding symmetries,
respectively. The expressions for incremental moduli tensors in terms of the six
invariants of the total energy function are well documented in the literature [19-21]. The
products in Equation (10) are defined as:
,
(12)
By considering the increment of the Lagrangian form of governing Equations (2) and (3),
one has:
(13)
The Eulerian counterparts of the Equation (13) can be obtained through the following
transformations:
,
(14)
where the superscript e indicates the Eulerian form of the quantity concerned, when the
reference configuration is updated from to , after the increments are formed. Hence
Equation (13) in the Eulerian form can be written as:
,
Combining Equations (10), (12) and (14), one has:
53
(15)
(16)
where is the displacement gradient and the updated magnetoelastic moduli
tensors are defined by:
,
(17)
It is noted that the updated moduli tensors possess the same symmetry as the ones in
Equation (11).
3.2 Buckling Analysis
In periodic structures, shape transformation due to buckling instabilities has been utilized
to explore tunable PnCs [97, 98]. Under a deformation, the periodicity of an infinite size
periodic porous structure breaks down, due to the first occurrence of bifurcation in the
microstructure. Pattern transformation of PnCs occurs due to either: i) local buckling
modes, or ii) global modes of instability, namely microscopic and macroscopic
instability. Microscopic instability is based on the investigation of all possible bounded
modes of instability within the range of wavelengths comparable to the unit cell's size.
54
The macroscopic instability is a measure of instability modes with wavelengths much
longer than the unit cell's size [133].
In this study, a buckling analysis is performed on finite size PnCs to find eigenmodes and
post-buckling periodicity of the deformed structure, following the approach documented
in References [133]. The structure consists of a two-dimensional (2D) square array of
circular holes of radius . The initial cell of the undeformed structure is a
square with a side of that is used as the direct lattice vectors in and
directions (Figure 23(a)). A nearly incompressible Neo-Hookean hyperelastic model is
assumed for numerical simulations. The density, , shear modulus,
and initial bulk modulus of are assumed for FE simulations
[30].
The deformation boundary conditions are applied on all boundaries of the models
through; , where , F and are the displacement vector, the
deformation gradient tensor and the position vector in Lagrangian coordinates,
respectively. The superscripts (+) and ( -- ) are associated with the nodes on the opposite
boundaries (right (top) and left (bottom) in Figure 23(a)) of the rectangular model. The
linear buckling study is employed to study the buckling eigenmodes of the structure. The
buckling analysis on several PnCs of , and cells, subjected to
deformation boundary conditions, agreed with the previous findings on the instability
modes [97, 98, 133]. Figure 23 shows the first mode pattern transformation results from
the buckling analysis of a finite size periodic structure under uniaxially
compressive stretch.
55
(a)
(b)
(c)
Figure 23. The FEM results for pattern transformation in the phononic crystal due to
buckling instability under the applied deformation. (a) An undeformed 8×8 periodic
structure, (b) the first buckling mode of the structure subjected to uniaxially compressive
stretch in horizontal direction, and (c) the first buckling eigenmode of the enlarged unit cell.
Hence, buckling instabilities changes the configuration to a new enlarged unit cell with
periodicity of times the initial unit cell (Figure 23(c)). The new unit cell will be
referred to as the representative unit cell (RUC). A post-transformation analysis is
performed on the RUC to obtain the deformed geometry for the wave propagation study.
3.3 Wave propagation analysis
In this study, the propagation of infinitesimal harmonic plane waves is considered in a
medium subjected to a pre-existing homogeneous deformation gradient tensor, F, and a
uniform magnetic induction vector, . In this work, the focus is on the two-dimensional
elastic wave propagation in periodic structure of infinite size. PnCs are characterized by
a unit cell that is defined through direct lattice vectors, and . These vectors are the
periodicity of the lattice in and directions. The RUC and the corresponding deformed
geometry for a horizontal uniaxial stretch , considered for wave propagation
study, are shown in Figure 24.
56
𝚳 𝚼
𝚾 𝚪
(a)
(b)
(c)
Figure 24. (a) The RUC selected for the uniaxial compression, (b) the corresponding deformed
geometry for uniaxial stretch 𝝀𝒙 𝟎 𝟗, and (c) reciprocal lattice's unit cell selected for the
wave propagation study. Irreducible Brilluine zone is shown in the region bounded by 𝜞 𝜰
𝜧 𝜲 𝜞
A solution of field Equation (15) in the form of:
,
(18)
is considered, where are amplitudes and is the angular frequency.
Substituting Equation (18) into Equation (15) in the absence of body forces yields the
updated Equations as a function of and To develop a FE model, the variational
formulation of the coupled Equations (15) is derived. is used to
define a vector potential such that . The weak form of governing
Equations is derived by taking the inner product of Equations (15)1 and (15)3 with an
arbitrary test function. Let us consider and be arbitrary variations of and ,
respectively, that satisfies the boundary conditions on . Taking the variational form of
the (15)1 and (15)3, integrating by parts and using the divergence theorem yields:
57
(19)
(20)
Equations (19) and (20) are defined on V, the RUC's domain, and represents the
perimeter of the RUC. It is noted that the weak form in Equation (20) results from a
stationary magnetic field condition and the non-conducting magnetoelastic medium in the
absence of the surface current density on . Moreover, is the
incremental surface traction normal to the boundaries of the deformed RUC, where n is
the unit vector normal to boundaries in the outward direction. Additionally, ( )
represents the tangent magnetic field at boundaries. Natural boundary conditions in
Equations (19) and (20) appear as work terms applied to the boundaries , which arise
from the tangent magnetic field and the normal traction force. It is evident from Figure
24 that the boundary integrals in Equation (19) and in
Equation (20) vanish, because the normal unit vector n acts in opposite directions on the
parallel boundaries of the deformed RUC. The final weak forms of the constitutive
equations are:
(21)
(22)
58
Weak forms in Equations (21) and (22) define the coupling constitutive behavior of
MREs. The nodal values of the vector potential [ ] and displacement
[ ] are the unknowns.
Bloch displacement boundary conditions are applied on the opposite boundaries of the
deformed RUC so that where k is the Bloch wave vector and
r denotes the distance vector between parallel boundaries. The superscripts (+) and ( -- )
denote the corresponding opposite boundaries; right (top) and left (bottom) in Figure 24,
respectively. It should be noted that in Equation (16) the tractions on boundaries follow
the Bloch relation. The only difference appears due to the fact that the tractions on the
parallel boundaries are in the opposite directions. Hence, the Bloch type boundary
conditions are . The deformed RUC and the corresponding
unit cell in reciprocal lattice including irreducible Brillouin zone (IBZ) are shown in
Figure 24. The wave vector takes the values of along the edges
, where are real numbers and are
the lattice vectors of the reciprocal space [74]:
‖ ‖
‖ ‖
‖ ‖
,
Since
the
boundary
conditions
are
in
complex
space,
thus
and
are generally complex. It is noted that in FE solvers, once the Bloch displacement
boundary conditions are applied, the traction boundary conditions are automatically
satisfied. The Bloch type boundary condition is applied through the essential or Dirichlet
boundary condition node in the partial differential equations (PDE) interface of the FE
59
solver. A computational code is implemented in the FE model to investigate the
incremental moduli tensors defined by Equation (17) in terms of deformation gradient
and the applied magnetic induction. The magnetoelastic interaction is characterized by a
stiffness matrix, which is a function of magnetic induction, stretch and wave vector.
The field equations for 2D motions are considered;
. Hence, solutions
which only depend on the in-plane variables and are sought. The FE discretization of
the coupling problem results in an eigenvalue problem. Eigenfrequency study is used for
investigating the eigenfrequencies of the coupled equations. Eigenfrequencies of the
wave equation are obtained to examine band diagrams and plot the iso-frequency
contours at different levels of the applied magnetic induction. Dispersion relations are
calculated sweeping the boundaries of IBZ. The reduced frequency
, is plotted as
a function of the wave vector where
and √ .
3.4 Results and discussions
A compressible Neo-Hookean type energy function is proposed to extract constitutive
relations and magnetoelastic moduli tensors, as follows:
(23)
60
where and are the two Lame constants of the hyperelastic
material in the absence of the magnetic induction and is assumed to be constant. is
principal invariant of right Cauchy -- Green deformation tensor and are
invariants dependent on the magnetic induction vector which exhibit the nonlinear
coupling behavior of the material. One observes that the parameter does not affect the
stress tensor, while if positive, enhances the material stiffness in the direction of the
applied magnetic induction. In the contrary, provides a measure of how the magnetic
field is affected by the macroscopic deformation. Magnetic permeability in vacuum,
, are assumed for numerical
simulations which accounts for 10% by volume of iron particles in the soft elastomer [17-
23]. Moreover, the dissipation is neglected.
The constitutive laws in Equations (16) are expanded as:
(24)
( )
( )
The energy function in Equation (23) is used to calculate the incremental magnetoelastic
moduli tensors (Equation (17)) in order to apply in corresponding constitutive laws in
Equation (24). For a uniaxial deformation gradient tensor [
61
] and uniaxial
magnetic induction vector [
], substituting Equation (11) in Equation (17) and
vanishing the zero terms, the corresponding moduli tensors will be:
,
,
,
,
,
,
,
(25)
,
,
,
,
where [
] is the left Cauchy-green deformation tensor.
To illustrate the effect of the applied magneto-mechanical loadings, the PnC is first
deformed through the macroscopic deformation tensor, F, and then is subjected to an
external magnetic field. A plane strain uniaxial compression is applied to deform the
RUC. The deformation gradient tensor is [ ], where and are the
applied stretches in and directions, respectively. Since the deformation occurs before
the magnetic field is applied, is determined from the equilibrium condition, such that
the pure mechanical stress component in direction vanishes. is assumed to be 0.9
for the numerical simulations, hence .
62
A nonlinear stationary study is performed on the RUC to capture the large deformation
effect. The deformation boundary conditions are applied on all boundaries of the RUC.
Small distortions are implemented to perturb the initial geometry and obtain the post-
bifurcation state of the RUC. The deformed RUC is extracted and imported in a new
model for the coupled wave propagation analysis. No pre-stress exists in the new model.
The RUC and the deformed RUC for uniaxially compressive stretch are shown
in Figures 24(a) and 24(b), respectively.
Once the deformation is applied, the RUC is subjected to different levels of magnetic
inductions. The corresponding band diagrams are shown in Figure 25. To illustrate the
effect of the applied magnetic induction, the first 30 in-plane modes are computed and
plotted in band diagrams. As can be seen in Figure 24, the range of the frequency
spectrum of the band structure expands as the magnitude of the applied magnetic
induction increases.
Ω
,
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
Γ Χ Μ Υ Γ
Γ Χ Μ Υ Γ
Γ Χ Μ Υ Γ
Γ Χ Μ Υ Γ
Wavenumber, k[1/m]
Wavenumber, k[1/m]
Wavenumber, k[1/m]
Wavenumber, k[1/m]
63
Figure 25. In-plane band diagrams for uniaxially compressive stretch at (a) No
magnetic induction, (b) 1.0T, (c) 2.0T, and (d) 3.0T unidirectionally applied magnetic induction.
PBGs are shown in shaded regions. The second and third modes represented by red and blue
bands, respectively, are selected for directionality analysis.
The range of the reduced frequency spectrum increases from in the absence
of magnetic stimulus, to for the 3.0T applied magnetic induction. Figure
25(a) shows the band diagram in the absence of magnetic induction. A Phononic
bandgap (PBG) is captured at the range of . No considerable change in
PBG is observed at 0.25T applied magnetic induction. As the magnetic induction
increases to 1.0T, 2.0T and 3.0T, the first PBG transforms to ,
64
and , respectively (Figure 25(b-d)). Upon the application
of the 1.0T magnetic induction a new PBG is created at but vanishes at
2.0T and 3.0T applied magnetic inductions. Also, a new PBG is generated at
at 2.0T and widens and shifted to as the magnetic induction
increases to 3.0T (Figure 25(c) and (d)). During the application of the magnetic
induction, the first and second PBGs orientation is widened and considerably shifted after
about . Additionally, band diagram investigation on the effect of magnetic
induction in the undeformed structure demonstrates that the first PBGs in the band
diagram appear due to the applied deformation while the second PBGs appear due to the
applied magnetic induction. In Figure 25, second and third modes, depicted by red and
blue lines, respectively, are selected to study the directional behavior of the structure.
While the band diagram exhibits the compact representation of the consecutive
eigenmodes, it cannot illustrate the directionality of the wave propagating in the structure.
Dispersion relations are also presented in the form of iso-frequency contours to fully
illustrate the wave propagation characteristics of the PnCs. The iso-frequency contours
identify the frequency of free wave motion of the PnC in the plane. The solution
of the magnetoelastic eigenvalue problem for all combinations of
and defines the dispersion surfaces of the structure. The
symmetry of the phase constant surfaces allows to limit the iso-frequency plots in the first
Brillouin zone.
Preferential directions of the free wave propagation, phase velocity and group velocity
can be derived from iso-frequency contours of the phase constant surfaces. By definition,
65
phase velocity is related to the speed of the wave crests while the group velocity
represents the speed of the wave envelope and lies along the normal to the corresponding
iso-frequency contour in the plane. Phase velocity and group velocity are
defined by
and
respectively, where k is the magnitude of the wave
vector [134-135]. The group velocity determines the direction of the wave propagation,
while directionality represents the flow of energy in the structure. Directionality of PnCs
also provides information about the directions in the structure that wave does not
propagate. These unique features of PnCs depend on the unit cell's geometry, material
properties and frequency and provide useful information on the wave propagation in the
periodic structures. Directionality can be considered as a measure of stiffness of a PnC
structure. For instance, in the vicinity of a long wavelength limit, the anisotropy of a PnC
can be expressed in terms of the phase and group velocity diagrams [135-136].
Figure 26 represents the effect of uniaxial compression, and different
magnitudes of magnetic induction on dispersion relations in terms of normalized iso-
frequency contours of the second eigenmode shown by red line in the band diagrams. It
is necessary to mention that isofrequency contours are represented for 0.25T to
demonstrate the transition in the contours for the loading from 0 to 1.0T magnetic
induction. Since the isofrequency plots are similar for the applied magnetic inductions
beyond 2.0T, the results for 3.0T are not presented in the paper. The dispersion relations
in case of no magnetic field are plotted in Figure 26(a) and shows that wave propagates at
different directions with different speeds. In this case, qualitative behavior of the contour
lines verifies the results documented in the literature [97].
66
By increasing the magnetic loading, the dispersive behavior of the contour lines changes
and tends to become parallel to the axis, particularly at higher frequencies. The
pattern of contour lines does not change qualitatively when the magnitude of the
magnetic induction vector increases beyond 2.0T. It is interesting to note that frequency
spectrum covers a higher range by increasing the magnetic excitation. In Figure 26,
arrows represent the direction and magnitude of the group velocity vectors. Group
velocity vector fields confirm that the structure shows a directional behavior at higher
levels of magnetic loading.
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
Figure 26. Normalized iso-frequency contours associated with the second modes for uniaxially
compressive stretch at (a) No magnetic induction, (b) 0.25T, (c) 1.0T, and (d) 2.0T
unidirectionally applied magnetic induction. The contours are associated with the red band in
Figure 25.
67
The iso-frequency contour plots of the third modes (represented by blue line in the band
diagrams) are shown in Figure 27 for uniaxial compression, and different levels
of magnetic induction. A totally different pattern is observed for the third modes. At
1.0T and 2.0T wave propagation has a directional behavior only at low frequencies.
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
y
c
n
e
u
q
e
r
f
d
e
c
u
d
e
R
Figure 27. Normalized iso-frequency contours associated with the third modes for uniaxially
compressive stretch at (a) No magnetic induction, (b) 0.25T, (c) 1.0T, and (d) 2.0T
unidirectionally applied magnetic induction. The contours are associated with the blue band in
Figure 25.
Directionality, quantified as
at each frequency, is represented by
polar plots in Figures 28 and 29 associated with results presented in Figures 26 and 27
68
respectively. Frequency and directionality is plotted in radial and tangential directions,
respectively.
Figure 28(a-b) illustrates that the frequency-dependent directionality of the structure
covers a large angular range. Upon the application of the 0.25T magnetic induction, the
directional behavior of the structure is initiated. At high frequencies an isotropic pattern
is observed. Figure 28(c-d) exhibits directional behavior of the structure due to the
increasing magnetic excitation. The directionality spans a narrow angular range
demonstrating a higher speed of wave in the direction of the applied magnetic field.
Figure 28. Directionality plots associated with the second modes for uniaxially compressive
stretch at (a) No magnetic induction, (b) 0.25T, (c) 1.0T, and (d) 2.0T unidirectionally
applied magnetic induction. Frequency and direction is plotted in radial and angular directions,
respectively.
69
Figure 29 shows the directionality plot associated with the third modes at different levels
of the magnetic induction. At 1.0T, structure shows directional behavior only for low
frequencies. No strong preference in directions is captured for third modes by increasing
magnetic induction beyond 1.0T.
Figure 29. Directionality plots associated with the third modes for uniaxially compressive stretch
at (a) No magnetic induction, (b) 0.25T, (c) 1.0T, and (d) 2.0T unidirectionally applied
magnetic induction Frequency and direction is plotted in radial and angular directions,
respectively.
Next, we study the effect of the macroscopic deformation loading case on the band
diagrams. Different macroscopic deformation tensors generate different patterns in the
porous structure. Here, a Mooney-Rivlin type energy function is used to extract
constitutive relations and magnetoelastic moduli tensors, as follows:
[ ] ,
70
(26)
where is the shear modulus in the absence of the magnetic induction,
and assumed to be constant. are principal invariants of right Cauchy -- Green
deformation tensor and are invariants dependent on the magnetic induction
vector. Here, it is assumed that for magnetic permeability in
vacuum, for numerical simulations [20-22].
To illustrate the effect of the applied magneto-mechanical loadings, the PnC is first
deformed through the macroscopic deformation tensor F, and then is subjected to an
external magnetic field. Two different cases of deformation are considered:
A plane strain uniaxial compression: In this case, the RUC is compressed in
direction. The deformation gradient tensor is defined as: [ ] .
Where and are the applied stretches in and directions, respectively.
Since the deformation is occurred before the magnetic field applies, the is
determined from the equilibrium condition so that the pure mechanical stress
component in direction vanishes. is assumed for the numerical
simulations, hence .
A plane strain equally biaxial compression: In this case, the RUC is equally
compressed in and directions. The deformation gradient tensor is defined as:
[ ] where is the applied stretch, and considered to be for
numerical simulations.
The RUC and the deformed RUC for uniaxially compressive stretch are shown
in Figures 30(a) and 30(b), respectively. For the biaxial compression, different RUC is
selected, due to convergence issue. The corresponding RUC and deformed RUC are
shown in Figures 30(c) and 30(d), respectively.
71
(a)
(b)
(e)
(c)
(d)
Figure 30. (a) The RUC selected for the uniaxial compression, (b) the corresponding deformed
geometry for , (c) The RUC selected for equally biaxial compression (d) the
corresponding deformed geometry for , (e) reciprocal lattice's unit cell selected for the
wave propagation study. Irreducible Brilluine zone is shown in the region bounded by
.
Once the deformation is applied, the RUC is subjected to different levels of magnetic
inductions. Figure 31 reports the evolution of PBGs with the applied magnetic induction
for different types of macroscopic deformation.
72
Figure 31. PBG vs. applied magnetic induction in y direction, for the square array PnC in: (a)
uniaxially compressive stretch and (b) in biaxially compressive stretch .
This plot is resulted from the band diagram simulation at different levels of the magnetic
induction. During the application of the magnetic induction on the uniaxially compressed
structure (Figure 31(a)), the first PBG orientation is widened and considerably shifted
after about . The width of the second and third PBGs remains fairly constant
with increasing the magnetic induction up to 2.0T. The three PBGs gradually widens
when the magnetic field reaches to . The PBG diagram obeys a fairly similar
trend for the case of biaxial compression (Figure 31(b)). PBGs gradually widen and shift
to higher frequencies during the increase of the magnetic induction.
3.5 Summary and conclusion
73
In pursuing of the smart design for isolators and frequency filters, a tunable PnC design is
illustrated through a porous pattern of the soft magnetoelastic PnC. In summary, the
ability of a periodic magnetoelastic structure to be used as a tunable PnC is demonstrated.
A buckling analysis is carried out to demonstrate the change of periodicity of the
structure due to the applied deformation. When the magneto-mechanical loading is
applied, the transformation of eigenmodes occurs due to both the geometric pattern
transformation and the material property changes. Results reveal that the designed PnC
has the potential to control and switch the PBGs with the applied macroscopic
deformation and magnetic stimulus. Upon the application of the magnetic field, new
PBGs is created and the position and width of the PBGs is altered. Interestingly, the band
diagram investigations reveal that by increasing the magnetic induction the frequency
spectrum of the bands widens and PBGs shift to higher frequencies. Strong preferential
directions in wave propagation are observed by increasing the magnetic induction,
particularly for the second modes. Results confirm that the considered PnC possesses
highly directional dispersive phononic characteristics compare to the previously studied
mechanically tunable PnC [97]. The benefit of this approach is that it combines the effect
of magnetic induction and deformation based pattern change in the microstructure and
provides an added degree of freedom in the control of the dynamic response of the
structure. However, 1.0T magnetic induction is required to capture the desired band
diagram and directionality in the deformed structure, which is a high value, but
74
achievable. Results demonstrate the potential of the proposed PnC to be utilized in wave
propagation systems, such as wave filters, beam steering and waveguides.
75
CHAPTER FOUR
A Thermally Tunable Phononic Crystal
A thermally tunable PnC is designed and analyzed through analytical and FEM
simulations. Bimaterial ligaments composed of two strips with contrast in their thermal
expansion coefficient are employed to design local resonators inside a periodic structure.
The thermally induced large deformations are utilized to exploit pattern change in the
structure to control elastic wave propagation. Once the temperature difference is
removed the structure tends to return to the initial state providing opportunities to be used
as thermally tunable acoustic switches and filters.
4.1 Modeling
Figure 32(a) shows the geometry of the periodic structure considered in this study. The
unit cell is a resonator consisting of a square core with four identical bimaterial
ligaments. The frame and core are modeled as elastic materials with negligible thermal
expansion coefficients. Once the temperature difference is applied on the structure the
bimaterial ligaments are deformed and generate pattern change in the periodic structure,
as is shown in Figure 32(b).
(a)
(b)
76
Figure 32. A finite sample of the proposed periodic structure at (a) undeformed state, (b)
deformed by thermal actuation, .
77
Ligaments are modeled as nearly incompressible neo-Hookean materials. When the
thermal expansion takes place, a volume change occurs in the structure. The total
deformation gradient tensor, can be decomposed into an elastic, and a thermal,
gradient tensors; . The total volume ratio is related to the mechanical and
thermal volume ratios as; or .
The thermal strain appears as , where is the thermal expansion
coefficient and T and are the current and reference temperatures, respectively. For
isotropic materials, the thermal gradient tensor is a diagonal matrix; . Hence,
. The hyperelastic model is characterized by a nonlinear energy density
function [136-137]:
(1)
Where and are the shear and bulk moduli, respectively, and is the first
invariant of the right Cauchy-Green deformation tensor
Thus, the first Piola-
Kirchhoff stress tensor,
is:
(2)
where
is the transpose matrix of the inverse of . We note that
and
. Equation governing the incremental motions superimposed on pre-deformed
structures in Lagrangian coordinates is:
(3)
78
where is the incremental displacemet, is the incremental first Piola-Kirchhoff stress
tensor, and is the density of the material. The increment of first Piola-Kirchhoff stress
tensor, is a function of the incremental deformation gradient tensor,
The incremental moduli tensor is a fourth-order tensor defined by
.
.
A solution of the wave equation in the form of plane wave is sought,
where is the amplitude vector and is the angular frequency. Thus, the stress can be
written in the following form:
Therefore, the equation of motion is an eigenvalue problem, as follows:
(4)
(5)
where is the eigenfrequency of the system. Wave propagation in PnCs is investigated
through the application of Bloch type boundary conditions on parallel boundaries of the
unit cell; the smallest repetitive structural element of the structure. The primitive unit
cell and the deformed unit cell are shown in Figure 33. Elastic modulus and Poisson's
ratio are input in the model as and and and
for frame and square core. and are used as the Lame constants
for the strip and and for the L-shaped ligament.
and are presumed for thermal expansion coefficient of the strip and L-
shaped ligament. The unit cell size is and the size of the core square is .
79
A 2D propagation of infinitesimal harmonic plane waves is considered in a periodic
structure subjected to a pre-existing homogeneous deformation gradient tensor. PnCs are
characterized by a unit cell that is defined through direct lattice vectors, and . These
vectors are the periodicity of the lattice in and directions. are
used as the lattice vectors in the study.
M
𝛤
X
(a)
(b)
(c)
Figure 33. (a) The primitive unit cell, (b) deformed unit cell, under and (c) First and
irreducible Brillouin zone chosen for the wave propagation analysis.
4.2 Results and discussions
The deformed unit cell is shown in Figure 33(b) for . Bloch type displacement
boundary conditions are applied on the opposite boundaries of the deformed unit cell so
that where k is the Bloch wave vector and r denotes the
80
distance vector between parallel boundaries. The superscripts (+) and ( -- ) denote the
corresponding opposite boundaries; right (top) and left (bottom) in Figure 33(b),
respectively. The band diagram and mode shapes for undeformed structure are shown in
Figure 34. 40 modes are shown in the diagram for frequencies larger than .
The band diagram shows narrow gaps of frequencies at , ,
, , and . Psudo-gaps in ,
and directions are shown separately in diagrams. The structure shows partial gaps
in the specific directions of symmetries, while complete band-gaps cover very narrow
range of frequencies. The band diagram and mode shapes for the deformed structure are
shown in Figure 35. 40 modes are shown in the diagram for frequencies larger than
. The band diagram shows narrow gaps of frequency at ,
and . Psudo-gaps in , and directions are also
shown in diagrams. The structure shows partial gaps in the specific directions of
symmetries, while complete band-gaps occur at very narrow range of frequencies.
Comparing the band diagram results in the undeformed and deformed structure
demonstrates that the band-gaps are suppressed by the applied temperature difference.
Moreover, the position and width of psudo-gaps in specific directions are changed by the
applied temperature difference. Figures 34(b) and 35(b) compare the effect of
temperature difference on the mode-shapes of the structure.
81
ω
Γ Χ Μ Γ
(a)
(b)
Figure 34. (a) Band diagram of undeformed structure and (b) the first 6 mode-shapes at 𝛤.
Psudo-gaps and complete gaps are shown in color region in the band diagram.
ω
Γ Χ Μ Γ
(a)
(b)
82
Figure 35. (a) Band diagram of deformed structure and (b) the first 6 mode-shapes at 𝛤. Psudo-
gaps and complete gaps are shown in color region in the band diagram.
4.3 Summary and Conclusion
83
A thermally tunable PnC is designed and analyzed through FEM modeling. The
proposed periodic structure utilizes temperature induced large deformations in bimaterial
ligaments of an in plane resonant unit cell to exploit pattern transformation in a periodic
structure. The propagation of elastic waves is studied on the prestressed deformed
structure. Complete band-gaps are shifted by the applied temperature difference.
Moreover, psudo-gaps are transformed in specific directions of symmetry of the unit cell.
Band diagram results demonstrate the ability of the thermally tunable periodic structure
to be used in the control of elastic wave propagation.
CHAPTER FIVE
84
A New Computational Method for Overall Tangent Moduli of a Soft
Magnetoactive Composite Using Periodic Homogenization1
A finite element methods based homogenization approach is presented to simulate the
nonlinear behavior of magnetoactive composites under a macroscopic deformation and an
external magnetic field. The coupled magnetoelastic constitutive law and governing
equations are developed in micro-scale for large deformations. Micro-scale formulation
is employed on a characteristic volume element, taking into account periodic boundary
conditions. Periodic homogenization method is utilized to compute macroscopic
properties of the magnetoelastic composite at different mechanical and magnetic loading
paths. A new and cost effective numerical scheme is used to develop the magnetoelastic
tangent moduli tensors. The sensitivity analysis is proposed to compute the overall
tangent moduli tensors of the composite through the finite difference method. The
presented approach is useful in characterization of magnetoactive and electroactive
composites and FE2 methods. Results are presented for typical equilibrium states.
In this study, a FEM-based homogenization method is employed to compute the effective
response of a periodic MEC under applied magnetic fields and large deformations. Due
1 Results of this chapter are published in:
Bayat A and Gordaninejad F 2015 A New Computational Method for Homogenized Tangent Moduli of a
Soft Magnetoelastic Composite Journal of Smart Materials and Structures 24 075010.
85
to high interest in magneto-active and electro-active composites, the characterization of
these composites needs to be realized through commercial software. This study differs
from the prior work in the proposed numerical approach for the computation of the
tangent moduli tensors using the commercial FEM package COMSOL which is capable
of multiphysics modeling. The overall tangent moduli are developed based on the
sensitivity analysis of deformation gradient tensor and magnetic field vector by utilizing
the finite difference method. Constitutive laws of the homogenized material are not
derived from a macroscopic energy function, but a CVE is attached to a material point to
extract the effective response through the volume average of microscopic quantities. It is
assumed that the principle of separation of scales is satisfied for the relative dimensions
of the microstructure and fluctuation field in contrast to that of the CVE. Theoretical
framework for constitutive laws and coupled governing equations for magnetoelastic
continuum is presented following the finite elasticity theory [21-28]. The FEM
discretization is carried out on the CVE consisting of a magnetically permeable particle
and a hyperelastic matrix considering periodic boundary conditions. The periodic
homogenization is employed to extract macroscopic constitutive laws of the nominal
stress tensor and magnetic induction vector.
5.1 Modeling
A direct micro to macro extraction of the material properties is defined through the FEM-
based homogenization approach. A CVE consisting of a magnetically permeable particle
and a soft matrix is used for micro-scale. The permeable particle and soft matrix are
86
characterized by a magnetoelastic energy function where the magnetic permeability of
the matrix is presumed as that of the free space.
A typical CVE chosen for micro-scale analysis is shown in Figure 36. The MEC is
considered to be initially at an undeformed state, denoted by with boundary as the
reference configuration. The body deforms when subjected to time-dependent magnetic
and mechanical loadings. The region occupied by the continuum , with boundary ,
at a given time t is the deformed configuration. Let and be the position vectors of the
material point at reference and deformed configurations, respectively, where
and is the deformation mapping. The deformation gradient tensor is defined
by , where is the gradient operator with respect to material
coordinates, . In this study, notations are used for micro-scale
differential operators in Lagrangian coordinates. A Lagrangian formulation is adopted to
develop magnetoelastic relations. The Lagrangian magnetic field and magnetic induction
vectors are denoted by and , respectively. It is assumed that the
magnetic field is stationary and the non-conducting MEC material is initially at the static
configuration and subjected to only magnetic and mechanical interactions. Thus,
are independent of time. Maxwell equations of magneto-statics can be written, as
follows:
, ,
(1)
It is worth mentioning that Equation (1) is resulted from no electric field, no free charge
and no current density assumptions on the continuum. Equation (1) is used to define a
scalar potential , such that , since .
Thus, the magneto-static differential equations are solved for the scalar potential.
87
In the absence of body forces, the equilibrium equation on the micro-scale reads:
,
(2)
where T is the nominal stress tensor defined at reference configuration. Equations (1)
and (2) are coupled governing equations of the magneto-elastic continuum. Constitutive
relations of the magneto-elastic medium are derived from a nonlinear magnetoelastic
energy density, , which is a function of the deformation gradient tensor and
magnetic field vector, defined per unit volume at . For a compressible material,
constitutive relations for the nominal stress and the magnetic induction are:
in
(3)
In macro-scale, the volume occupied by a body in reference (undeformed) configuration
is denoted by , which is bounded by and notations and are assigned for the
corresponding deformed configuration of the continuum. and are associated with the
macroscopic Lagrangian and Eulerian coordinates, respectively. The macroscopic
deformation mapping, follows . Accordingly, are
used for macro-scale differential operators in Lagrangian coordinates. Hence,
is the macroscopic deformation gradient tensor. Consequently,
coupled governing equations of the continuum are:
, ,
88
(4)
where are macroscopic magnetic field, magnetic induction and nominal
stress, respectively. Similarly, macroscopic quantities can be related to a macroscopic
energy function through:
in
(5)
Derivation of constitutive laws through a macroscopic energy function is beyond the
scope of this study. Computation of these macroscopic quantities is performed through
surface integrals of corresponding microscopic counterparts across the CVE's boundary.
Notations,
∫
and
∮
, are introduced for volume
and surface integrals on the body's domain and boundary, respectively, where V is the
volume of the domain in the reference configuration. Assuming the continuity of the
deformation gradient tensor and the magnetic induction vector on the boundary of the
CVE, the surface integral can be equivalently estimated as the volume integral of
corresponding properties on the CVE's domain. Using Gauss theorem, macroscopic
deformation and nominal stress tensors are given by:
,
(6)
where and are the position vector, traction vector, and normal vector on the
boundary of the CVE, respectively. Likewise, the corresponding magnetic field and
magnetic induction vectors in macro-scale are defined by integral equations:
89
,
(7)
where and are the magnetic potential and magnetic flux on the boundary of
the CVE, respectively.
In periodic homogenization, the microscopic position vector can be expressed as a linear
function of the macroscopic deformation gradient and a fluctuation field;
, where is the fluctuation field, a vector function of the position vector.
Similarly, the magnetic potential follows where is a scalar
function of . Boundary conditions of the CVE are derived from the classical Hill-
Mandel homogeneity condition. Hill-Mandel condition states that the increment of the
macroscopic energy function is equivalent to the volume average of the increment of the
microscopic energy function, . It is shown that periodic boundary conditions
for the deformation and magnetic potential on satisfy the Hill-Mandel condition
[46-48]. Fluctuation fields are assumed to be periodic on thus, the traction vector
and magnetic flux are anti-periodic, as follows:
on
(8)
90
where superscripts (+) and ( -- ) are associated with nodes on opposite boundaries (right
(top) and left (bottom) edges on Figure 36 (b-c)) of the . In order to define the
boundary value problem (BVP), the macro-scale boundary conditions on , for
mechanical and magnetic problem must be specified. Figure 36 shows a schematic of the
BVP. The boundary of the homogenous body is decomposed into different sections,
where corresponding Dirichlet and Neumann boundary conditions for mechanical and
magnetic problems are prescribed. Displacement type boundary conditions, on
as well as traction boundary conditions on are prescribed to define
mechanical BVP, where , and are the macroscopic displacement, traction and
normal to the surface vectors defined on the boundary of the body, respectively. The
corresponding magnetic BVP is defined through scalar magnetic potential boundary
condition; on and magnetic flux boundary condition; on ,
where and are the macroscopic magnetic potential and magnetic flux, respectively.
𝜕 𝜎
𝒏
𝜕 𝑢
(a)
𝜕 𝜑
{𝑭 𝑯 }
CVE
𝑭 𝑭 𝑭
𝑯 𝑯 𝑯
+
𝑿
𝜕 𝑏
𝒏
(b)
-
-
(c)
+
Figure 36. (a) The homogenized body and corresponding boundary decomposition in
Lagrangian configuration, (b) corresponding CVE, attached to 𝑿, selected for homogenization
study including a circular permeable particle inside a soft square matrix, (c) the deformed CVE
in Eulerian configuration.
91
The macroscopic deformation gradient tensor and magnetic field are obtained from
macro-scale BVP which are inputs of the micro-scale problem. The transition from
microscopic to macroscopic properties is defined on the micro-scale problem through
FEM-based averaging process. Thus, the effective nominal stress and magnetic induction
as well as macroscopic (effective) moduli tensors are computed from corresponding
microscopic properties on CVE domain.
To develop a finite element model, the variational formulation of the equilibrium and
magneto-statics equations is derived. To derive the weak form of governing equations,
inner product of Equations (1)2 and (2) with an arbitrary test function, is considered and
then integrated over the CVE domain. Let us consider and be arbitrary variations
of and , respectively, that satisfy the boundary conditions on . Taking the
variational form of Equations (1)2 and (2), using integration by part and then the
divergence theorem yields:
∫
∮
∫
∮
(9)
(10)
It is noted that the weak form of magneto-statics equation results from a stationary
magnetic field condition and a non-conducting MEC medium in the absence of surface
current density on . Natural boundary conditions in Equations (9) and (10) appear
as work terms applied to , which arise from the normal traction force and the
magnetic flux. Recalling Equation (8), it is evident from Figure 36(b) that boundary
92
integrals ∮
, in Equation (9) and ∮
, in Equation (10) vanish, since
the normal unit vector, n acts in opposite directions on the parallel boundaries of the
deformed CVE. Weak forms in Equations (9) and (10) define the coupled magnetoelastic
behavior of the model which is solved utilizing the nonlinear FEM solver. COMSOL
Multiphysics is used for numerical simulations which allows for direct implementation of
weak expressions.
To predict the overall response of the MEC through homogenization approach,
computation of the effective tangent moduli is required. Homogenized tangent moduli
tensors contribute to constitutive laws of macro-scale BVP. In contrast to macroscopic
quantities; which can be directly computed by Equations (6) and (7), the
overall tangent moduli
cannot be obtained through volume averaging of
microscopic counterparts. This is due to the fact that there is no explicit relation of
macroscopic nominal stress tensor and magnetic induction vector as a function of and
. Incremental (linearized) constitutive relations of the coupled magnetoelastic BVP can
be estimated, as follows:
,
(11)
where and are macroscopic mechanical, magneto-mechanical and magnetic moduli
tensors, respectively. Taking the gradient of displacement vector and magnetic potential
defined for periodic homogenization, the deformation gradient and magnetic field can be
decomposed into a constant and a fluctuation part and , where
and . Substituting in moduli tensor relations one has,
93
,
(12)
where and are microscopic counterparts of moduli tensors. In Equations (12), the
challenge is to compute partial derivatives of fluctuation fields;
and
, since there is
no explicit expression of microscopic fluctuation fields in terms of macroscopic
variables, and . Computation of the sensitivity of and , with respect to their
macroscopic counterparts needs to be performed through numerical methods. Sensitivity
analysis is a technique developed to measure incremental variations of a single input
parameter that affects a particular dependent variable of an objective function, while
remaining inputs are kept constant. The objective function is in general a function of the
solution to a multiphysics problem , which is manipulated by the control
variables . Using a Taylor expansion around the state , the sensitivity of y to
is defined by
. A finite difference method is used to compute macroscopic tangent
moduli and sensitivity of and with respect to and . Details are given in section 3.
5.2 Results and Discussion
94
The FEM-based homogenization approach presented in this study is carried out on a
typical two dimensional CVE consisting of a permeable inclusion and a matrix shown in
Figure 36(b). The matrix is a square of edge and the radius of circular inclusion is
. The CVE is analyzed under different states of deformation and magnetic field to
compute the distribution of microscopic nominal stress and magnetic induction as well as
effective counterparts. Both inclusion and matrix are modeled as a compressible neo-
Hookean magnetoelastic material. To conduct numerical analysis, a particular form of
energy function is required. Due to lack of experimental data on MECs, limited data are
available in the literature. In this study, a typical magnetoelastic energy function is
considered:
⁄
(13)
where is the right Cauchy -- Green deformation tensor, is the trace of c and
. and are Lamé constants and is the magnetic permeability. Material
properties of the inclusion are chosen as , and and
those of the matrix are , and where
is the magnetic permeability of the vacuum [45, 46, 47].
For the magnetoelastic energy function given in the Equation (13), the nominal stress
tensor is calculated as:
[ ]
(
)
(14)
95
where is the inverse matrix of and is the identity tensor. The magnetic
induction vector would be:
[
]
The microscopic moduli tensors are derived as:
[
] [
]
[
]
( (
))
(15)
(16)
[(
) (
)]
[
] [
]
[
]
(17)
[
] [
]
(18)
For clarity, distribution of microscopic stress and magnetic induction are investigated for
three different cases of magneto-mechanical loadings in plane:
96
A plane strain uniaxial stretch at constant magnetic field: three cases of
macroscopic deformation is considered while magnetic field is being kept
constant at direction:
[
] and *
+
is numerically computed at each loading step, so as .
A plane strain pure shear at constant magnetic field: three cases of macroscopic
deformation is considered while magnetic field is being kept constant at
direction:
[
] and *
+
A plane strain equally biaxial stretch at different levels of magnetic field: three
cases of magnetic loading is considered at direction while macroscopic
deformation is kept constant:
*
+ and *
+
Simulation results for uniaxial loading case are shown in Figure 37. All contour plots
presented are normalized by the corresponding volumetric average of the quantity
concerned. Figures 37(a-c) show the distribution of the microscopic nominal stress
component, , for uniaxial stretch at constant magnetic field . The
macroscopic stress changes as and , as the stretch is
increased by and , respectively. The microscopic stress is a nonlinear function
97
of the deformation gradient tensor and magnetic field. Distribution of the microscopic
stress is different for compressive and tensile stretches. Smaller stress occurs in the
matrix compared to the inclusion, as it is expected due to stiffer material properties of the
inclusion.
𝑇
𝑇
(a)
(b)
𝐵
𝐵
𝑇
𝑇
𝐵
𝐵
(c)
(d)
(e)
(f)
𝑇
𝑇
𝐵
𝐵
Figure 37. Contour plots of normalized microscopic distribution of 𝑻𝟐𝟐 for uniaxial loading
case at (a) 0.8, (b) 1.1 and (c) 1.4 stretches. Normalized microscopic distribution of 𝑩𝟐 for
uniaxial loading case at (d) 0.8, (e) 1.1 and (f) 1.4 stretches.
Distribution of the component of the microscopic magnetic induction, normalized by
, is shown in Figure 37(d-f). Due to high contrast between magnetic permeability of
the inclusion and matrix, high magnetic induction occurs at the inclusion area. Effective
magnetic induction varies by and , as the stretch is increased by
and , respectively. This increase is due to the fact that the magnetic induction
98
vector is a nonlinear function of stretch. Moreover, nonuniform deformation distribution
in the matrix area causes a higher magnetic induction at top and bottom edges of the
CVE.
𝑇
𝑇
𝐵
𝐵
(a)
(d)
𝑇
𝑇
(b)
(c)
𝐵
𝐵
(e)
(f)
𝑇
𝑇
𝐵
𝐵
Figure 38. Contour plots of normalized microscopic distribution of 𝑻𝟏𝟐 for pure shear loading
case at (a) -0.3, (b) 0.1 and (c) 0.2 shear stretches. Normalized microscopic distribution of 𝑩𝟏
for pure shear loading case at (d) -0.3, (e) 0.1 and (f) 0.2 shear stretches. The red boundary lines
represent the 𝝏𝑪𝑽𝑬 in undeformed configuration. All deformed plots are scaled to 1.
Figures 38(a-c) and 38(d-f) show the contour plots of the microscopic stress and
magnetic induction for case of pure shear loading, respectively. Macroscopic magnetic
field is applied in direction. The component of stress tensor and component of
the magnetic induction vector are shown for analysis. The macroscopic stress takes the
99
values and , as the shear stretch changes by and
, respectively. Similar to the case of uniaxial loading case, the highest stress and
magnetic induction occurs at the inclusion, due to stiffer mechanical properties and
higher magnetic permeability of the inclusion. The macroscopic magnetic induction
changes by and , as the shear stretch increases by and
, respectively.
𝑇
𝑇
𝑇
𝑇
(a)
(b)
(c)
𝐵
𝐵
𝐵
𝐵
(d)
(e)
(f)
𝑇
𝑇
𝐵
𝐵
Figure 39. Contour plots of normalized microscopic distribution of 𝑻𝟏𝟏 for equally biaxial
loading case at (a) 𝟏𝒆𝟓𝑨𝒎 𝟏, (b) 5𝒆𝟓𝑨𝒎 𝟏 and (c) 𝟏𝒆𝟔𝑨𝒎 𝟏 magnetic field. Normalized
microscopic distribution of 𝑩𝟏 for equally biaxial loading case at (d) 𝟏𝒆𝟓𝑨𝒎 𝟏, (e) 5𝒆𝟓𝑨𝒎 𝟏
and (f) 𝟏𝒆𝟔𝑨𝒎 𝟏 magnetic field.
100
Figures 39(a-c) and 39(d-f) show the contour plots of the microscopic stress and
magnetic induction for equally biaxial loading case, respectively. Macroscopic magnetic
field is applied as in direction. The component of stress
tensor and component of the magnetic induction vector are shown for comparison.
The macroscopic stress takes the values and , as the
magnetic field is increased by , respectively. The highest stress
and magnetic induction occurs at the inclusion, due to stiffer mechanical properties and
higher magnetic permeability of the inclusion. The macroscopic magnetic induction
changes by and , as the magnetic field is increased by
, respectively.
In all cases of magneto-mechanical loadings presented, the relative distribution of the
microscopic stress components mainly depends on the deformation state. This arises
obviously due to the dominant geometric nonlinearity term compared to the magnetic
nonlinearity term, in the microscopic stress formulation at considered magnetic field.
One might note that periodic boundary conditions are not appeared in the deformed unit
cells shown in Figures 37-39. This is because of the fact that periodic boundary
conditions are applied on the fluctuation fields but the deformed unit cells displays the
microscopic position vector . The fluctuation fields on the
boundaries are shown in Figure 40 for the pure shear loading case. Figure 40
corresponds to the Figure 38 which depicts the deformed unit cell at pure shear loading
case.
𝑇
𝑇
𝐵
𝐵
𝑇
𝑇
(b)
(c)
𝐵
𝐵
(e)
(f)
(a)
(d)
101
𝑇
𝑇
𝐵
𝐵
Figure 40. Contour plots showing the periodic boundary conditions on the fluctuation fields.
The normalized microscopic distribution of 𝑻𝟏𝟐 for pure shear loading case at (a) -0.3, (b) 0.1
and (c) 0.2 shear stretches. Normalized microscopic distribution of 𝑩𝟏 for pure shear loading
case at (d) -0.3, (e) 0.1 and (f) 0.2 shear stretches. The red boundary lines represent the 𝝏𝑪𝑽𝑬
in undeformed configuration. All deformed plots are scaled to 1. The Figure corresponds to
Figure 38 which sketches the deformed unit cells based on of microscopic position vector.
A parametric study is carried out to understand the effect of the increasing magnetic field
on the macroscopic quantities. Figure 41 shows the plots for macroscopic stress and
magnetic induction components with respect to the stretch at different levels of
macroscopic magnetic field. For comparison, magnetic field is applied in direction
for all loading cases. The model is parametrically swept on both stretch and magnetic
field component for some distinct equilibrium states.
102
Results for components of the stress tensor for the uniaxial, equally biaxial and pure
shear loadings are shown in Figures 41(a), 41(c) and 41(e), respectively. In Figure 41(a),
the average stress turns into compressive one and is highly increased when the
magnetic field reaches . The quadratic magnetic field-dependent term in
the stress function becomes a negative term and shows significant effect at higher
magnitude of the applied magnetic field, while the influence of the deformation only
dependent terms is dominant at lower magnetic fields. In Figure 41(c), the effect of the
magnetic term tends to switch the stress component to compressive stress. Analogous
to the uniaxial loading case, the deformation dependent term of the stress dominates at
low magnetic fields. Figure 41(e) demonstrates the effect of macroscopic magnetic field
on the stress component for pure shear loading case. The macroscopic magnetic field
has strong effect on the shear stress component as is expected from Equation (15).
Corresponding parametric study results for component of the effective magnetic
induction are depicted in Figures 41(b), 41(d) and 41(f) for uniaxial, equally biaxial and
pure shear loading cases, respectively. From Equation (16), for a constant deformation
state, the magnetic induction has a linear relation with the applied magnetic field for both
matrix and inclusion domains. In all loading cases, magnetic induction increases by the
increasing stretch.
103
𝑻
𝟏
𝑩
(a) Stretch 𝑓
(b) Stretch 𝑓
𝑻
𝟏
𝑩
(c) Stretch 𝑓
(d) Stretch 𝑓
𝑻
𝟏
𝑩
(e) Stretch 𝑓
(f) Stretch 𝑓
𝒂
𝑷
𝑴
𝟏
𝟏
𝑻
𝒂
𝑷
𝑴
𝟏
𝟏
𝑻
𝒂
𝑷
𝑴
𝟐
𝟏
𝑻
Figure 41. Parametric analysis resulting from numerical study representing (a) 𝑻𝟏𝟏 vs. stretch
(b) 𝑩 𝟏 vs. stretch for uniaxial loading case, (c) 𝑻𝟏𝟏 vs. stretch (d) 𝑩 𝟏 vs. stretch for equally
biaxial loading case and (e) 𝑻𝟏𝟐 vs. stretch (f) 𝑩 𝟏 vs. stretch for pure shear loading case at
different levels of macroscopic magnetic field.
5.3 Computation of macroscopic moduli tensors
104
Computation of the first term in right hand side of Equation (12) requires integration of
components of microscopic moduli tensors given by Equations (17-19), over CVE's
domain. Calculation of the second term of integral Equations (12) requires the
computation of
and
. For small perturbations, a suitable approach is the Taylor
expansion of fluctuation functions around macroscopic quantities, and . Thus,
and
are first-order sensitivity of and , with respect to their macroscopic counterparts,
respectively.
The finite difference method is used to find the first order sensitivity for two independent
cases of loading paths, deformation induced loading and magnetic induced loading.
Practically, for an arbitrary deformation loading path, an incremental loading is
performed on the model by sweeping the uniaxial stretch at gradually decreasing
incremental steps. This is performed through parametric sweep study of the software. At
each increment, the model is run for small perturbations, e.g. , of four components
of macroscopic deformation gradient tensor, while all other parameters of the model are
kept constant. The four components of are then computed for each independent
perturbation and the 4th- order sensitivity tensor is estimated through [
]
[ ]
[ ]
. Analogously, for magnetic loading case, an incremental parametric sweep is
applied on the macroscopic magnetic field. At each incremental level, the model is run
for small variation, e.g. of two components of the macroscopic magnetic field
vector, while the rest of parameters are kept constant. Two components of are
computed for each independent perturbation component and the 2nd -- order magnetic
sensitivity tensor is approximated by [
]
[ ]
[ ]
. In Equations (12), single and
double contractions are denoted by and , respectively.
105
In this study, LiveLinkTM for MATLAB® interface is used for the computation of the
macroscopic moduli tensors, which connects COMSOL Multiphysics to MATLAB
scripting environment. LiveLinkTM for MATLAB® is a Java® based interface that
increases the FEM modeling ability by using MATLAB commands and functions to set
up the model and physics from scripts, control the model and analyze the results.
Microscopic moduli tensors are calculated with direct implementation of Equations (17-
19), as local variables in the model. The model is created and saved as .mph file in the
MATLAB directory. Then the model is imported in the MATLAB script for further
processing.
In terms of MATLAB script implementation, a loop is used for each loading path to
perform the incremental sweep on the parameter concerned. At every increment step,
components of the deformation gradient tensor are perturbed by stretch, while
is chosen for the perturbation of components of the magnetic field vector.
Components of the microscopic moduli tensors and fluctuation fields are computed and
integrated on CVE's domain according to the Equation (12). Once the FEM model is
created and saved in the MATLAB directory, it is called in the MATLAB script for
evaluation of the homogenized moduli tensors. Main syntaxes used in the code are:
1. model = mphload('modelname')
2. model.param.set('parameter','value');
3. model.study('std1').run;
4. [v1,v2,...] = mphint2(model,{'q1','q2',... },'surface')
5. data = mpheval(model,{'q1','q2',... },'selection',1)
106
The first command loads the .mph file (COMSOL files extension) which is already saved
in the MATLAB working directory. The second command assigns the quantity in
'value' to the desired parameter in the model. The third command runs the model. The
4th command evaluates the surface integration of the string expressions 'qi's on the
CVE's domain. The 5th command evaluates the string expressions 'qi's as a field value
at each node points. All the data and integration evaluations returns in matrix format and
stored for plotting and further processing.
Results for selected components of overall mechanical tangent moduli tensors for a
typical uniaxial loading path are depicted in Figure 42. component of the deformation
gradient tensor is incrementally increased from 0.8 to 1.35 at a constant macroscopic
magnetic field, . From Equation (17), the components of are highly
nonlinear and complex functions of the deformation gradient tensor. As it is shown in
Figure 42(a), the components and are of order of 1010, since they are highly
dependent on component of the macroscopic magnetic field. While -dependent
terms are not dominant in and components, as demonstrated in Figure 42(b).
Consequently, , and their symmetric counterparts of the mechanical moduli
tensors are dominant terms in the stress tensor.
𝑎
𝑃
𝔸
𝐴
𝑚
𝑎
𝑃
𝔸 …
𝔸
(a) 𝑓
…
107
𝔸 …
𝔸
…
(b) 𝑓
𝑎
𝑃
𝔸
𝐴
𝑚
𝑎
𝑃
(c) H 𝑨𝒎 𝟏
(d) H 𝑨𝒎 𝟏
Figure 42. (a,b) Components of homogenized mechanical moduli tensor, (c,d) Components of
the homogenized coupling magneto-mechanical moduli tensor.
Results for selected components of the coupling moduli tensor for a uniaxial magnetic
loading path are shown in Figure 42(c-d). is incrementally increased form
to at a constant macroscopic deformation state [ ]. From
Equation (18), components of moduli are linear functions of . The nonlinearity
observed in the Figures 42(c,d) stems from the effect of sensitivity tensor as given by
108
Equation (12). The magnitude of tends to increase by the increasing magnetic field.
Moreover, as it is deduced from Equation (19), components of tensors are independent
of the macroscopic magnetic field. The nonzero components of macroscopic counterparts
are nearly computed as (not shown in the Figure). It has to be
mentioned that tangent moduli tensors are computed through a set of perturbation tests
conducted along a loading path about a reference macroscopic equilibrium state { }.
Hence, the overall properties presented here depend on the selected uniaxial loading path
and the equilibrium state.
5.4 Summary and Conclusions
A numerical tool is demonstrated to compute the homogenized properties of the magneto-
active composite structures through a commercial FEM package. The proposed
algorithm provides a computational approach to study the micro to macro transition of
the mechanical and magnetic properties of the MEC structures. The presented approach
differs from all prior work in two aspects. The methods presented in references [48-51]
for computation of effective tangent moduli tensors are carried out through a C++ based
in-house FE code and are of high computational cost. In this study, the FEM package -
COMSOL Multiphysics- is employed for all numerical simulations which allows for
direct implementation of weak forms of governing equations. It also takes the advantage
of MATLAB scripting environment for parametric study and control of the model's
variables and physics. More importantly, the finite difference method proposed in this
109
study offers a computationally cost-effective methodology to evaluate homogenized
tangent moduli tensors for different loading paths.
A neo-Hookean type magnetoelastic energy function is proposed to demonstrate the
nonlinear coupling behavior of the matrix and magnetic inclusion of the composite. A
CVE with periodic boundary condition is selected to extract the microscopic distribution
of stress and magnetic induction. Macroscopic properties are evaluated through
volumetric averaging of microscopic counterparts. Macroscopic properties of the
composite are extracted from the microscopic counterparts, through the homogenization
procedure. No effective (macroscopic) energy function is assumed in this process. For
both uniaxial and biaxial loading cases considered here, macroscopic stress results
confirm that the magnetic field dependent terms of stress are dominant at high
macroscopic magnetic fields. Moreover, magnetic induction increases by increasing
stretch and magnetic field. At constant magnetic field, both stress and magnetic
induction increase by increasing the stretch.
In conclusion, homogenization is an essential mechanism to compute the effective
properties of the magneto-active composites, especially when finding an effective
constitutive law is very difficult for complicated composites. Homogenized tangent
moduli tensors are useful and necessary characteristics of the magneto-active composites
for evaluating the overall response of the composite, FE2 modeling and macro-scale
instability analysis.
CHAPTER SIX
110
Homogenization Approach in Random Magnetoelastic Composites1
Homogenization is a numerical approach used as a tool to study the overall response of
the composite and heterogeneous materials presumed to be statistically homogenous. In
this study, a FEM-based homogenization method is employed to compute the effective
response of a random MEC under applied magnetic fields and large deformations.
Spatially random distribution of identically circular inclusions inside a soft homogenous
matrix is investigated. FEM-based averaging process is combined with Monte-Carlo
method (MCM) to generate ensembles of randomly distributed MECs. The ensemble is
utilized as a statistical volume element (SVE) in a scale-dependent statistical algorithm to
approach the desired characteristic volume element (CVE) size.
It is assumed that the principle of separation of scales is satisfied for the relative
dimensions of the microstructure and fluctuation fields in contrast to that of the CVE.
The overall tangent moduli tensors are developed based on the sensitivity analysis of
deformation gradient tensor and magnetic field vector by utilizing the finite difference
method. Theoretical framework for constitutive laws and coupled governing equations
for magnetoelastic continuum is presented following the finite elasticity theory [21-28].
The FE discretization is carried out on SVEs consisting of randomly distributed
1 Results of this chapter are submitted to the Journal of Computational mechanics.
111
magnetically permeable particles within a hyperelastic matrix.
The random
homogenization is employed to extract macroscopic constitutive laws of the nominal
stress tensor and magnetic induction vector.
6.1 Modeling
A direct micro-meso-macro extraction of material properties is defined through the FEM-
based homogenization approach. A SVE consisting of magnetically permeable particles
within a soft matrix is used as ensembles in meso-scale. The magnetic particles and the
soft matrix are characterized by a magnetoelastic energy function where the magnetic
permeability of the matrix is presumed as that of the free space.
A typical SVE, chosen for meso-scale analysis is shown in Figure 43(b). The
magnetoelastic composite (MEC) is considered to be initially at an undeformed state,
denoted by with boundary as the reference configuration. The body deforms
when subjected to time-dependent magnetic and mechanical loadings. The region
occupied by the continuum , with boundary , at a given time t is the deformed
configuration. Let and be the position vectors of the material point at reference and
deformed configurations, respectively, where and
is
the
deformation mapping. The deformation gradient tensor is defined by
, where is the gradient operator with respect to material coordinates, .
In this study, notations are used for micro-scale differential
operators in Lagrangian coordinates. A Lagrangian formulation is adopted to develop
112
magnetoelastic relations. The Lagrangian magnetic field and magnetic induction vectors
are denoted by and , respectively. It is assumed that the magnetic
field is stationary and the non-conducting MEC material is initially at the static
configuration and subjected to only magnetic and mechanical interactions. Thus,
are independent of time. Maxwell equations of magneto-statics can be written, as
follows:
, ,
(1)
It is worth mentioning that Equation (1) is resulted from no electric field, no free charge
and no current density assumptions on the continuum. Equation (1) is used to define a
scalar potential , such that , since .
Thus, the magneto-statics differential equations are solved for the scalar potential.
In the absence of body forces, the equilibrium equation on the micro-scale reads:
,
(2)
where T is the nominal stress tensor defined at reference configuration. Equations (1)
and (2) are coupled governing equations of the magneto-elastic continuum. Constitutive
relations of the magneto-elastic medium are derived from a nonlinear magnetoelastic
energy density, , which is a function of the deformation gradient tensor and
magnetic field vector, defined per unit volume at . For a compressible material,
constitutive relations for the nominal stress and the magnetic induction are:
in
(3)
113
In macro-scale, the volume occupied by a body in reference (undeformed) configuration
is denoted by , which is bounded by and notations and are assigned for the
corresponding deformed configuration of the continuum. and are associated with the
macroscopic Lagrangian and Eulerian coordinates, respectively. The macroscopic
deformation mapping, follows . Accordingly, are
used for macro-scale differential operators in Lagrangian coordinates. Hence,
is the macroscopic deformation gradient tensor. Consequently,
coupled governing equations of the continuum are:
, ,
(4)
where are macroscopic magnetic field vector, magnetic induction vector and
nominal stress tensor, respectively. Similarly, macroscopic quantities can be related to a
macroscopic energy function through:
in
(5)
Derivation of constitutive laws through a macroscopic energy function is beyond the
scope of this study. Computation of macroscopic quantities is performed through surface
integrals of corresponding microscopic counterparts across the ensemble's boundary.
Notations,
∫
and
∮
, are introduced for volume
and surface integrals on the body's domain and boundary, respectively, where V is the
volume of the domain in the reference configuration. Assuming the continuity of the
deformation gradient tensor and the magnetic induction vector on the boundary of the
SVE, the surface integral can be equivalently estimated as the volume integral of
114
corresponding properties on the SVE's domain. Using Gauss theorem, macroscopic
deformation and nominal stress tensors are given by:
,
(6)
where and are the position vector, traction vector, and normal vector on the
boundary of the SVE, respectively. Likewise, the corresponding magnetic field and
magnetic induction vectors in macro-scale are defined by integral equations:
,
(7)
where and are the magnetic potential and magnetic flux on the boundary of
the SVE, respectively.
The microscopic position vector can be expressed as a linear function of the macroscopic
deformation gradient and a fluctuation field; , where is the
fluctuation field, a vector function of the position vector. Similarly, the magnetic
potential follows where is a scalar function of . Boundary
conditions of the SVE are derived from the classical Hill-Mandel homogeneity condition.
Hill-Mandel condition, states that the increment of the macroscopic energy
function is equivalent to the volume average of the increment of the microscopic energy
function,.
,
(8)
115
To satisfy the Hill's condition, both the mechanical and magnetic boundary integrals in
Equation (8) are required to vanish through defining appropriate boundary conditions.
Two types of boundary conditions are considered here:
Linear displacement boundary conditions (LD-BC) with zero mean fluctuation
fields; and .
Periodic boundary condition (PF-BC) for fluctuation fields; and and
anti-periodic condition for t and b [46-48]:
on
(9)
where superscripts (+) and ( -- ) are associated with nodes on opposite boundaries (right
(top) and left (bottom) edges on Figure 43 (b)) of the . One may argue that the use
of periodic boundary conditions is not suitable for a random heterogeneous structure. It
has been shown that adopting periodic boundary conditions can estimate the effective
properties, even when the structure is characterized with random distribution of the
inclusions [53-61].
In order to define the boundary value problem (BVP), macro-scale boundary conditions
on for mechanical and magnetic problem must be specified. Figure 43(a) shows a
schematic of the BVP. The boundary of the homogenous body is decomposed into
different sections, where corresponding Dirichlet and Neumann boundary conditions for
mechanical and magnetic problems are prescribed. Displacement type boundary
116
conditions, on as well as traction boundary conditions on are
prescribed to define mechanical BVP, where , and are the macroscopic
displacement, traction and normal to the surface vectors defined on the boundary of the
body, respectively. The corresponding magnetic BVP is defined through scalar magnetic
potential boundary condition; on and magnetic flux boundary condition;
on , where and are the macroscopic magnetic potential and magnetic
flux, respectively.
𝜕 𝜎
𝒏
𝜕 𝑢
(a)
𝜕 𝜑
{𝑭 𝑯 }
+
Homogenization
𝑿
𝜕 𝑏
-
(b)
SVE
-
+
Effective
medium
(c)
Figure 43. (a) The homogenized body and corresponding boundary decomposition in
Lagrangian configuration, (b) corresponding SVE, attached to 𝑿, selected for statistical analysis
including randomly distributed circular permeable particle inside a soft matrix (c) the effective
medium modeled using homogenization approach.
The macroscopic deformation gradient tensor and magnetic field are obtained from
macro-scale BVP which are inputs of the micro-scale problem. The transition from
microscopic to macroscopic properties is defined on the micro-scale problem through
FEM-based averaging process. Thus, the effective nominal stress and magnetic induction
as well as macroscopic (effective) moduli tensors are computed from corresponding
microscopic properties on the SVE domain.
117
To develop a finite element model, variational formulation of the equilibrium and
magneto-statics equations is derived. To derive the weak form of governing equations,
inner product of Equations (1)2 and (2) with an arbitrary test function, is considered and
then integrated over the SVE domain. Let us consider and be arbitrary variations
of and , respectively, that satisfy the boundary conditions on . Taking the
variational form of Equations (1)2 and (2), using integration by part and then the
divergence theorem yields:
∫
∮
∫
∮
(10)
(11)
It is noted that the weak form of magneto-statics equation results from a stationary
magnetic field condition and a non-conducting MEC medium in the absence of surface
current density on . Natural boundary conditions in Equations (10) and (11) appear
as work terms applied to , which arise from the normal traction force and the
magnetic flux. For linear displacement boundary conditions, the traction t and magnetic
flux b are calculated from the applied displacement field and magnetic potential on the
boundaries of the ensemble. The FEM solver automatically computes the surface
tractions from the displacement type boundary conditions, and applies them as work
terms in the variational formulations.
For periodic boundary conditions, recalling Equation (9), it is evident from Figure 43(b)
that boundary integrals ∮
, in Equation (10) and ∮
, in Equation (11)
vanish, since the normal unit vector, n acts in opposite directions on the parallel
118
boundaries of the deformed SVE. Weak forms in Equations (10) and (11) define the
coupled magnetoelastic behavior of the model which are solved utilizing the nonlinear
FEM solver. COMSOL Multiphysics is used for numerical simulations which allows for
direct implementation of weak expressions.
To predict the overall response of the MEC through homogenization approach,
computation of the effective tangent moduli is required. Homogenized tangent moduli
tensors contribute to constitutive laws of macro-scale BVP. In contrast to macroscopic
quantities; which can be directly computed by Equations (6) and (7), the
overall tangent moduli
cannot be obtained through volume averaging of
microscopic counterparts. This is due to the fact that there is no explicit relation of
macroscopic nominal stress tensor and magnetic induction vector as a function of and
. Incremental (linearized) constitutive relations of the coupled magnetoelastic BVP can
be estimated, as follows:
,
(12)
where and are macroscopic mechanical, magneto-mechanical and magnetic moduli
tensors, respectively. Taking the gradient of displacement vector and magnetic potential
defined for periodic homogenization, the deformation gradient and magnetic field can be
decomposed into a constant and a fluctuation part and , where
and . Substituting in moduli tensor relations one has,
(13)
119
,
where and are microscopic counterparts of moduli tensors. In Equations (13), the
challenge is to compute partial derivatives of fluctuation fields;
and
, since there is
no explicit expression of microscopic fluctuation fields in terms of macroscopic
variables, and . Computation of the sensitivity of and , with respect to their
macroscopic counterparts needs to be performed through numerical methods presented in
chapter two.
6.2 Random homogenization framework
In this study, the homogenized response of a 2D magnetoactive heterogeneous material is
sought through random homogenization process. A scale dependent statistical approach
is employed to perform the convergence analysis on the ensembles of the random MEC
following the approaches documented in [53-61].
120
(a)
𝔀 𝟓 𝔀 𝟕 𝔀 𝟗 𝔀 𝟏𝟏 𝔀 𝟏𝟑 𝔀 𝟏𝟓 𝔀 𝟏𝟕
Increasing scale factor and number of inclusions
(b)
(c)
Figure 44. Schematic of the statistical algorithm used in the study. (a) selection of random meso-
scale ensembles of the heterogeneous MEC for a typical test window size, (b) increasing the size
of the test window and number of particles towards the convergence window and (c) five typical
realizations for 𝔀 𝟕 All ensembles are generated for 𝑨𝒇 𝟎 𝟑𝟓 𝒅 𝟏𝟎 𝐚𝐧𝐝 𝒅𝒄 𝟏.
121
The magnetoactive material is a two-phase composite consists of randomly distributed
hard and permeable inclusions in a soft non-magnetic matrix. It is assumed that the Hill's
general homogenization limit between the size of heterogeneity, d and that of CVE, L is
satisfied; .
The effective behavior of composite material depends on properties of microstructure's
constituents. SVE is an area of the random heterogeneous material that is chosen to
calculate the overall properties. In random homogenization, the problem is to find the
appropriate size of the SVE which can represent the heterogeneous material properties,
called CVE. Here, the goal is to develop a FEM-based random homogenization approach
which uses the Monte-Carlo method to generate random SVEs and capture the effective
properties.
Figure 44 shows the schematic of statistical approach and generation of the meso-scale
SVE ensembles in the random MEC. The statistical approach is equivalent to moving
limited number of square-shaped frames at different points of the heterogeneous
structure, where the size of the square frames is successively increased step by step till
the desired CVE size is approached (Figure 44(b)). Typical realizations of the random
structure for are shown in Figure 44(c).
The computational algorithm utilized in the simulation is depicted in Figure 45. First,
constant parameters of simulations: , , , , , Tol are input in the model. An
area fraction , is assigned to find the number of inclusions inside the square ensemble.
is defined as the ratio of total area of inclusions to the area of the SVE. Material
122
properties of the matrix and inclusions, , particle diameter, d, and distance between
particles, are held constant in simulations. Next, a window size, L is assigned to
generate the realization, , where
is a non-dimensional scale factor, discretely
increased in a loop till the convergence point is achieved. Each is generated through
random positioning of circular inclusions in the square matrix, using a so called hard-core
Poisson point field, so that the circles are prevented from overlapping, by assigning a
minimum distance between the boundaries of inclusions. For each ensemble , the
variational formulations (10) and (11) are solved to find the components of the
constitutive laws and moduli tensors. Steps 1-3 are repeated in a loop for each and for
maximum of 45 independent simulations. For each , the number of realizations, N is
determined
through
√
,
where
√
∑
is the standard deviation, is the average of the selected
component of the macroscopic moduli tensors and Tol is the assigned tolerance [54].
Dispersion of the data is checked through calculation of N at each step.
and
, are defined as the selected components of the moduli
tensors being monitored for statistical accuracy. The algorithm stops when the statistical
accuracy is achieved and the corresponding ensemble size, is qualified as the desired
CVE. The CVE is defined as the minimum window size for which the number of
required realizations N, is less than 5 for both parameters and . In other words, the
average components of moduli tensors remain in the tolerance interval and the minimum
dispersion in and are observed.
123
Assign inputs: , , , , , Tol.
Loop : 5 . Subsequently increase the size of SVE till the convergence is achieved.
Loop t:1 .
Calculate number of particles, .
Calculate size of SVE square .
Generate spatially random distribution of n particles in square ensembles.
Solve the variational formulations (10) and (11).
Compute constitutive laws and overall moduli tensors over each SVE.
Compute
and
Calculate number of realizations, N from
If
End Loop t
End Loop
.
√
for each parameters and .
Figure 45. Computational algorithm used in the statistical approach.
6.3 Results and Discussion
The FEM-based homogenization approach presented in this study is carried out on a
typical two dimensional SVE consisting of permeable inclusions and a matrix shown in
Figure 43(b). The SVE is analyzed under a deformation and magnetic field loading state
to compute the distribution of microscopic nominal stress and magnetic induction as well
as effective counterparts. Both inclusion and matrix are modeled as a compressible neo-
Hookean magnetoelastic material. To conduct numerical analysis, a particular form of
energy function is required. Due to lack of experimental data on MECs, limited data are
available in the literature. In this study, a typical magnetoelastic energy function is
considered:
⁄
(13)
where is the right Cauchy -- Green deformation tensor, is the trace of c and
. and are Lamé constants and is the magnetic permeability. Material
properties of the inclusion are chosen as , and and
those of the matrix are , and where
is the magnetic permeability of the vacuum [45, 46, 47].
124
The nominal stress tensor is calculated as:
[ ]
(
)
(14)
where is the inverse matrix of and is the identity tensor. The magnetic
induction vector would be:
[
]
The microscopic moduli tensors are derived as:
[
] [
]
[
]
( (
))
[(
) (
)]
[
] [
]
[
]
[
] [
]
(15)
(16)
(17)
(18)
A plane strain pure shear at constant magnetic field is considered while magnetic field is
kept constant at direction:
*
+ and *
+
125
and are assumed constant in the simulations. Input
parameters as well as magneto-mechanical loading state are kept fixed for all results
presented in this study. Two cases of boundary conditions; LD-BC and PD-BC are
considered. The ensemble's size is successively increased in a loop from 9 to the
convergence point. FEM simulations are performed using COMSOL Multiphysics. The
statistical algorithm is run through a computer code. LiveLinkTM for MATLAB®
interface is used for computer programming, which connects COMSOL Multiphysics to
MATLAB scripting environment. Ensemble size, number of inclusions and random
position of inclusions are calculated and the SVE is modeled in the COMSOL graphical
environment.
Weak expressions (10) and (11) are directly input in the FEM model. Constitutive laws
and micro-scale moduli components are input in the model from Equations (14)-(18).
The homogenized tangent moduli tensors are computed through the sensitivity approach
documented in the previous chapter. The model is run for each following the
approach explained in section 3.2. The convergence plots are shown in Figures 46 and
47 for moduli components, and and different boundary conditions, respectively.
and are normalized by their corresponding average value at the convergence point.
Figure 46(a) plots the statistical convergence plot for component versus the number of
simulations for different values of and LD-BC. The number of necessary realizations,
N is calculated from
√
for each and at each simulation loop. The
convergence is achieved for .
126
LD-BC
𝐸
𝑉
𝐶
𝔸
𝔸
(a)
Number of Simulations
PF-BC
𝐸
𝑉
𝐶
𝔸
𝔸
(b)
Number of Simulations
Figure 46. Moduli component , results from the statistical analysis for (a) LD-BC and (b) PF-
BC versus number of simulations.
127
LD-BC
Number of Simulations
PF-BC
𝐸
𝑉
𝐶
𝔹
𝔹
(a)
𝐸
𝑉
𝐶
𝔹
𝔹
(b)
Number of Simulations
Figure 47. Moduli component , results from the statistical analysis for (a) LD-BC and (b) PF-
BC versus number of simulations.
128
In all convergence plots presented here, the simulations are continued for to
demonstrate the scattering of the statistical data for different test windows. Figure 46 (b)
demonstrates that when PF-BC is adopted, the convergence is occurred at .
Figure 47 reports the convergence plots for component of magnetic modulus for LD-
BC and LF-BC cases. For both cases of boundary conditions, convergence is reached at
.
It is observed that in all convergence plots, by increasing the size of SVEs and number of
particles, less dispersion is observed in the magnitude of and and the convergence is
achieved with less number of simulations. It is emphasized that the ensemble size
achieved in the statistical approach is only valid for the particular form of the inputs,
loading state and boundary conditions. Different size and convergence plots may result
for various input and loading conditions. Figure 48 plots the coefficient of variation,
versus the scale factor, corresponds to the convergence plots in Figures 46 and 47.
is defined as the ratio between standard deviation to the mean of the statistical
data and is a measure of the statistical scattering. Figure 48 confirms that by increasing
the size of ensembles, less scattering is observed in the moduli components and
approaches to zero.
129
Scale factor, 𝓌
𝔸
𝑣
𝐶
(a)
𝔹
𝑣
𝐶
(b)
Scale factor, 𝓌
Figure 48. Coefficient of variation of (a) and (b) versus scale factor, .
130
Figure 49 compares the average of moduli components, and versus scale factor,
for both cases of boundary conditions. For , the moduli component shows the
same convergence trend for both cases of LD-BC and PF-BC. The convergence trend for
is shown in Figure 49(b). It is noted that the two types of boundary conditions used in
this study, do not define the hierarchies bounds for the corresponding moduli parameters
and and this plot cannot be used for identification of the CVE size. The CVE is
determined statistically through the algorithm demonstrated on Figure 45.
Figure 50 shows the distribution of the microscopic nominal stress component for two
typical SVE size resulted from different boundary conditions. Solution of the BVP for a
meso-scale ensemble with , and 57 number of inclusions for case of LD-BC is
shown in Figure 50(a). Result for the same SVE and PF-BC is shown in Figure 50(b).
Figure 50(c-d) reports the simulation results for SVE with , and 80 number of
inclusions for case of LD-BC and PF-BC, respectively. All contour plots presented here
are normalized by the corresponding volumetric average of the quantity concerned. One
can notice how the use of PF-BC results in periodic displacement boundary conditions.
For SVEs with shown in Figures 50(a) and 50(b) the macroscopic stress
component takes the values , respectively. For ensembles
with in Figures 50(c) and 50(d) takes the values
respectively.
131
Scale factor, 𝓌
𝑎
𝑃
𝔸
(a)
𝐴
𝑎
𝑃
𝔹
(b)
Scale factor, 𝓌
Figure 49. Average values of (a) and (b) versus scale factor, .
𝑇
𝑇
(a)
(b)
𝑇
𝑇
(c)
(d)
132
𝑇
𝑇
𝑇
𝑇
Figure 50. FEM results for component of stress tensor for SVEs with (a) , with LD-
BC, (b) , with PF-BC, (c) , with LD-BC and (d) , with PF-BC. All
plots are normalized by their average value of component.
Figure 51 reports the microscopic distribution of component of magnetic induction,
run on the same ensembles described in Figure 50. LD-BC is adopted on the SVEs with
and as shown in Figure 51(a) and 51(c), respectively. PF-BC is adopted
on the SVEs with and as shown in Figure 51(b) and 51(d), respectively.
Plots are normalized by the corresponding volumetric average component, . For SVEs
with , pictured in Figures 51(a) and 51(c) the macroscopic magnetic induction,
varies as , respectively. For ensembles with in Figures 51(c)
and 51(d), takes the values respectively.
133
𝐵
𝐵
𝐵
𝐵
𝐵
𝐵
𝐵
𝐵
(b)
(d)
(a)
(c)
Figure 51. FEM results for component of stress tensor for SVEs with (a) , with LD-
BC, (b) , with PF-BC, (c) , with LD-BC and (d) , with PF-BC. All
plots are normalized by their average value of component.
134
6.4 Summary and conclusion
Analytical and FEM approaches are combined with a scale-dependent statistical
method to identify a random MEC behavior under magneto-mechanical loadings.
Governing equations for magnetoelastic media are presented. BVP is defined on the
statistical meso-scale volume elements of the random MEC. SVEs are generated
through a computer code with random distribution of particles where particles are
prevented from overlapping. The focus is on identifying the minimum CVE size for
the random MEC. The CVE size is computed for a particular loading state through a
statistical algorithm which combines the FEM-based homogenization approach with
Monte-Carlo method.
Results show that for low values of scale factor, , the random positioning of
inclusions in each realization has high influence on the effective response of the
MEC, for both cases of boundary conditions. By increasing the scale factor and
number of particles (for a constant area fraction) the statistical dispersion and the
necessary number of realizations significantly reduces. It is observed that the CVE
size and are obtained for the and components respectively. The
resultant CVE for the multi-physics system should be identified at the window size
where both magneto-mechanical moduli components, and are converged.
Results show that for PF-BC case, convergence is achieved at a smaller SVE size.
FEM results for typical SVEs show that the highest stress and magnetic induction
occurs at the inclusions area, due to stiffer mechanical properties and higher magnetic
permeability.
135
CHAPTER SEVEN
Concluding Remarks and Future Work
The research presented in this dissertation has resulted in the following contributions:
High-amplitude wrinkle formation is employed to propose a one-dimensional phononic
crystal slab consists of a thin film bonded to a thick compliant substrate. Buckling
induced surface instability is employed to generate surface periodic scatterers to control
elastic wave propagation in the low thickness composite slab. Simulation results show
that the periodic wrinkly structure can be used as a transformative phononic crystal which
can switch band diagram of the structure in a reversible manner.
Dynamic response of a tunable phononic crystal consisting of a porous hyperelastic
magnetoelastic elastomer subjected to a macroscopic deformation and an external
magnetic field is investigated through considering a magnetoelastic energy function and
nonlinear moduli tensors for the medium. The band diagram of the structure is tuned by
combined effects of microstructural pattern change and magnetic field.
A thermally tunable phononic crystal is introduced which utilizes pattern change to
control band diagram of the structure.
A numerical scheme is demonstrated to compute the homogenized properties of the
periodic magnetoactive composite structures.
136
The finite difference method proposed in this study offers a computationally effective
methodology to evaluate homogenized tangent moduli tensors for different loading paths
for periodic magnetoactive composites.
An algorithm for determination of characteristic volume element size and effective
properties of random magnetoactive composites is presented which utilizes the FEM-
based homogenization approach with Monte-Carlo method.
Recommendations for future work:
Response of new tunable phononic crystals needs to be studied through combination of
control parameters, for example thermal and magnetic effects.
New pattern transformation paradigms need to be sought in order to propose new
periodic structures in controlling elastic wave propagation.
Dynamic instability in periodic structures has not been explored. Dynamic instability
can be occurred in finite amplitude wave propagation in the periodic structures.
The effect of macroscopic instability in periodic structures needs to be explored.
The proposed homogenization approach presented in this work needs to be extended to
study multi-scale analysis in the periodic and random magnetoelastic composites.
The proposed random homogenization approach presented in this work needs to be
further extended to study different material models and loading conditions. For each case
different boundary conditions may be required. The approach needs to be utilized in an
optimization process to identify the optimum matrix properties, the particle shape,
distribution and percentage in the design of magnetoactive composites.
REFERENCES
137
[1]. Kankanala S V
and Triantafyllidis N 2004 On
finitely
strained
magnetorheological elastomers J. Mech. Phys. Solids 52 2869-2908
[2]. Ramberg E 2012 Electrodynamics (New York: Burlington: Elsevier Science) 88-
96.
[3]. Bustamante R 2010 Transversely isotropic nonlinear magneto-active elastomers
Acta Mech. 210 183 -- 214.
[4]. Bustamante R, Dorfmann A and Ogden R 2006 Universal relations in isotropic
nonlinear magnetoelasticity Q. J. Mech. Appl. Math. 59 435 -- 450.
[5]. Chen L, Gong X L, and Li W H 2007 Microstructures and viscoelastic properties
of anisotropic magnetorheological elastomers Smart Mater. Struct. 16 2645 -- 2650.
[6]. Chen T, Nan C-W and Weng G J 2003 Exact connections between effective
magnetostriction and effective elastic moduli of fibrous composites and
polycrystals J. Appl. Phys. 94 491-495.
[7]. Danas K, Kankanala S V and Triantafyllidis N 2012 Experiments and modeling
of iron-particle-filled magnetorheological elastomers J. Mech. Phys. Solids 60
120-138.
[8]. Dorfmann A and Ogden R W 2005 Some problems in nonlinear magnetoelasticity
Z. Angew. Math. Phys. 56 718 -- 745.
[9]. Dorfmann A and Ogden R W 2010 Nonlinear electroelastostatics: incremental
equations and stability Int. J. Eng. Sci. 48 1 -- 14.
[10]. Farshad M and Benine A 2004 Magnetoactive elastomer composites Polym. Test
[11].
23 347-353.
Jolly M R Carlson J D and Munoz B C 1996 A model of the behaviour of
magnetorheological materials Smart Mater. Struct. 5 607 -- 614.
[12]. Rudykh S and Bertoldi K 2013 Stability of anisotropic magnetorheological
elastomers in finite deformations: a micromechanical approach J. Mech. Phys.
Solids 61 949 -- 967.
[13]. Yang J, Gong X, Deng H, Qin L and Xuan Sh 2012 Investigation on the
mechanism of damping behavior of magnetorheological elastomers Smart Mater.
Struct. 21 125015(1-11).
[14]. Ginder J M, Nicholes M E, Elie L D and Tardiff J L 1999 Magnetorheological
elastomers: properties and applications Proc. SPIE, Smart Structures and
Materials 1999: Smart Materials Technologies, Newport Beach, CA, Vol. 3675
131 -- 138.
[15]. Brigadnov I A and Dorfmann A 2003 Mathematical modeling of magneto-
sensitive elastomers Int. J. Solids Struct. 40 4659 -- 4674.
138
[16]. Dorfmann A and Ogden R W 2004 Nonlinear magnetoelastic deformations Q. J.
[17].
Mech. Appl. Math. 57 599 -- 622.
Jolly M R, Carlson J D and Munoz B C 1996 The magnetoviscoelastic response
of elastomer composites consisting of ferrous particles embedded in a polymer
Matrix Smart Mater. Struct 7 613 -- 621.
[18]. Dorfmann A, Ogden R W and Saccomandi G 2006 Universal relations for
nonlinear magnetoelastic solids Int. J. NonLin. Mech. 39 1699 -- 1708
[19]. Bustamante R, Dorfmann A and Ogden R W 2011 Numerical solution of finite
geometry boundary-value problems in nonlinear magnetoelasticity Int. J. Solids
Struct. 48 874 -- 883.
[20]. Saxena P 2012 On wave propagation in finitely deformed magnetoelastic solids
PhD thesis University of Glasgow, Glasgow.
[21]. Ogden R W 2009 Incremental elastic motions superimposed on a finite
deformation in the presence of an electromagnetic field Int. J. NonLin. Mech. 44
570-580.
[22]. Otténio M, Destrade M and Ogden R W 2008 Incremental magnetoelastic
deformations with application to surface instability J. Elasticity 90 19 -- 42
[23]. Galipeau E and Ponte Castañeda P 2013 A finite-strain constitutive model for
magnetorheological elastomers: magnetic torques and fiber rotations. J. Mech.
Phys. Solids 61 1065-1090.
[24]. Destrade M and Scott N H 2004 Surface waves in a deformed isotropic
hyperelastic material subject to an isotropic internal constraint Wave Motion 40
347 -- 357.
[25]. Dorfmann A L and Ogden R W 2014 nonlinear theory of electroelastic and
magnetoelastic interactions Springer US, 91-112.
[26]. Lee J S and Its E N 1992 Propagation of rayleigh waves in magneto-elastic media
J. Appl. Mech. 59 812 -- 818.
[27]. Hefni I A Ghaleb A F and Maugin G A 1995 One dimensional bulk waves in a
nonlinear magnetoelastic conductor of finite electric conductivity Int. J. Eng. Sci.
33 2067-2084.
[28]. Destrade M and Ogden R W 2011 On magneto-acoustic waves in finitely
deformed elastic solids Math. Mech. Solids 16 594 -- 604.
[29]. Kochmann D M and Venturini G N 2013 Homogenized mechanical properties of
auxetic composite materials in finite-strain elasticity Smart Mater. Struct. 22
084004(7pp).
[30]. Kanouté P, Boso D P, Chaboche J L and Schrefler B A 2009 Multiscale methods
for composites: a review Arch. Comput. Meth. Eng. 16 31 -- 75.
139
[31]. Kouznetsova V, Geers M G D and Brekelmans W A M 2002 Multi-scale
constitutive modeling of heterogeneous materials with a gradient-enhanced
computational homogenization scheme Int. J. Num. Methods Eng. 54 1235 -- 1260.
[32]. Miehe C 2002 Strain-driven homogenization of inelastic microstructures and
composites based on an incremental variational formulation Int. J. Num. Methods
Eng. 55 1285 -- 1322.
[33]. Miehe C and Koch A 2002 Computational micro-to-macro transitions of
discretized microstructures undergoing small strains Arch. Appl. Mech. 72 300 --
317.
[34]. Temizer I and Wriggers P 2008 On the computation of the macroscopic tangent
for multiscale volumetric homogenization problems Comput. Meth. Appl. Mech.
Eng. 198 495 -- 510.
[35]. Nemat-Nasser S 1999 Averaging theorems in finite deformation plasticity Mech.
Mater. 31 493 -- 523.
[36]. Smit R, Brekelmans W and Meijer H 1998 Prediction of the mechanical behavior
of nonlinear heterogeneous systems by multi-level finite element modeling
Comput. Meth. Appl. Mech. Eng. 155 181 -- 192.
[37]. Wang D, Chen J-S and Sun L 2003 Homogenization of magnetostrictive particle
filled elastomers using an interface-enriched reproducing kernel particle method
Finite Elem. Anal. Des. 39 765 -- 782.
[38]. Yin H M, Sun L Z and Chen J S 2002 Micromechanics-based hyperelastic
constitutive modeling of magnetostrictive particle-filled elastomers Mech. Mater.
34 505 -- 516.
[39]. Bravo-Castillero J, Rodríguez-Ramos R, Mechkour H, Otero J A, Hernández J,
Sixto- Camacho L M, Guinovart-Díaz R and Sabina F J 2009 Homogenization
and effective properties of thermomagnetoelectroelastic composites J. Mech.
Mater. Struct. 4 819 -- 36.
[40]. Li J and Dunn M 1998 Micromechanics of magnetoelectroelastic composite
materials: average fields and effective behavior J. Intell. Mater. Syst. Struct. 9
404 -- 16.
[41]. Bravo-Castillero J, Rodriguez-Ramos R, Mechkour H, Otero J and Sabina F 2008
Homogenization of magneto-electro-elastic multilaminated materials Quart. J.
Mech. Appl. Math. 61 311 -- 32.
[42]. Berger H, Gabbert U, Koppe H, Rodriguez-Ramos R, Bravo-Castillero J,
Guinovart-Diaz R, Otero J A and Maugin G A 2003 Finite element and
asymptotic homogenization methods applied to smart material composites
Comput. Mech. 33 61 -- 67.
140
[43]. Costanzo F, Gray G L and Andia P C 2005 On the definitions of effective stress
and deformation gradient for use in MD: Hill's macro-homogeneity and the viral
theorem Int. J. Eng. Sci. 43 533 -- 555.
[44]. Ponte Castañeda P and Galipeau E 2011 Homogenization-based constitutive
models for magneto-rheological elastomers at finite strain J. Mech. Phys. Solids
59 194 -- 215.
[45]. Chatzigeorgiou G, Javili A and Steinmann P 2014 Unified magnetomechanical
homogenization framework with application to magnetorheological elastomers
Math. Mech. Solids 19 193-211.
[46]. Labusch M, Etier M, Lupascu D C, Schröder J and Keip M-A 2014 Product
properties of a two-phase magneto-electric composite: Synthesis and numerical
modeling Comput. Mech. 54 71 -- 83.
Javili A, Chatzigeorgiou G and Steinmann P 2013 Computational homogenization
in magneto-mechanics. Int. J. Solids Struct. 50 4197 -- 4216.
[47].
[48]. Schröder J 2009 Derivation of the localization and homogenization conditions for
electro-mechanically coupled problems Comput. Mater. Sci. 46 595 -- 599
[49]. Schröder J and Keip M A 2012 Two-scale homogenization of electromechanically
coupled boundary value problems Comput. Mech. 50 229 -- 244.
[50]. Schröder J 2014 A numerical two-scale homogenization scheme: the FE2-method,
in: J. Schröder K Hackl (Eds.), Plasticity and Beyond, in: CISM International
Centre for Mechanical Sciences, vol. 550, Springer Vienna, 1 -- 64.
[51]. Keip M A, Steinmann P and Schröder J 2014 Two scale computational
homogenization of electro-elasticity at finite strains Comput. Methods Appl.
Mech. Eng. 278 62- 79.
[52]. Miehe C 2003 Computational micro-to-macro transitions for discretized micro-
structures of heterogeneous materials at finite strains based on the minimization
of averaged incremental energy Comput. Methods Appl. Mech. Engrg. 192 559 --
591.
[53]. Trovalusci P, De Bellis M L, Ostoja-Starzewski M and Murrali 2014 A particulate
random composites homogenized as micropolar materials Meccanica 49 2719 --
2727.
[54]. Trovalusci P, De Bellis M L, Ostoja-Starzewski M and Murrali A 2015 Scale-
dependent homogenization of random composites as micropolar continua Eur. J.
Mech. A. Solids 49 396-407
[55]. Ma J, Zhang J, Li L, Wriggers P and Sahraee S 2014 Random homogenization
analysis for heterogeneous materials with full randomness and correlation in
microstructure based on finite element method and Monte-Carlo method Comput.
Mech. 54 1395 -- 1414.
141
[56]. Gitman I M, Askes H and Sluys L J 2007 Representative volume: Existence and
size determination Eng. Fract. Mech. 74 2518 -- 2534.
[57]. Temizer I and Zohdi T I 2007 A numerical method for homogenization in non-
linear elasticity Comput. Mech. 40 281 -- 298.
[58]. Terada K, Horib M, Kyoyac T and Kikuchi N 2000 Simulation of the multi-scale
convergence in computational homogenization approaches Int. J. Solids Struct. 37
2285-2311.
[59]. Khisaeva Z F and Ostoja-Starzewski M 2007 Scale effects in infinitesimal and
finite thermoelasticity of random composites J. Therm. Stresses 30 587 -- 603.
[60]. Geers M G D, Kouznetsova V G and Brekelmans W A M 2010 Multi-scale
computational homogenization: Trends and challenges J. Comput. Appl. Math.
234 2175-2182.
[61]. Temizer I and Wriggers P 2011 An adaptive multiscale resolution strategy for the
finite deformation analysis of microheterogeneous structures Comput. Methods
Appl. Mech. Engrg. 200 2639 -- 2661.
[62]. Temizer I and Wriggers P 2011 Homogenization in finite thermoelasticity J.
Mech. Phys. Solids 59 344 -- 372.
[63]. Francfort G A 1983 Homogenization and linear thermoelasticity SIAM J. Math.
Anal. 14 696 -- 708.
[64]. Ostoja-Starzewski M, Du X, Khisaeva Z F and Li W 2007 Comparisons of the
size of the representative volume element in elastic, plastic, thermoelastic, and
permeable random microstructures Int. J. Multiscale Comput. Eng. 5 73-82.
[65]. Vel S S and Goupee A J 2010 Multiscale thermoelastic analysis of random
and
heterogeneous materials Part
homogenization of material properties Comp. Mat. Sci. 48 22 -- 38.
I: Microstructure
characterization
[66]. Zhang H W, Yang D S, Zhang S and Zheng Y G 2014 Multiscale nonlinear
thermoelastic analysis of heterogeneous multiphase materials with temperature-
dependent properties Finite Elem. Anal. Des. 88 97 -- 117.
of
2014 Computational modeling
[67]. Ethiraj G
ferromagnetics
and
magnetorheological elastomers PhD thesis University of Stuttgart, Stuttgart.
[68]. Castañeda P P and Siboni M H 2012 A finite-strain constitutive theory for electro-
active polymer composites via homogenization Int. J. Non Linear Mech. 47 293 --
306.
[69]. Ponte Castaneda P and Galipeau E 2011, Homogenization-based constitutive
models for magnetorheological elastomers at finite strain J. Mech. Phys. Solids 59
194-215.
[70]. Galipeau E, Rudykh S, deBotton G and Castañeda P 2014 Magnetoactive
elastomers with periodic and random microstructures Int. J. Solids Struct. 51
3012 -- 3024.
142
[71]. Elachi C 1976, Waves in active and passive periodic structures a review Proc.
IEEE 64 1666 -1698.
[72]. Liu Z, Chan C T and Sheng P 2000 Elastic wave scattering by periodic structures
of spherical objects theory and experiment Phys. Rev. B 62 2446-2458.
[73]. Olsson R H and El-Kady I 2009 Microfabricated phononic crystal devices and
applications Meas. Sci. Technol. 20 012002-012015.
[74]. Kittle C 1986, Introduction to solid state physics 8th Ed Wiley New York 27-53.
[75]. Sigalas M M and Economou E N 1993 Band structure of elastic waves in two
dimensional systems Solid State Commun. 86 141 -- 143.
[76]. Wang G, Wen X, Wen J, Shao L and Liu Y 2004 Two dimensional locally
resonant phononic crystals with binary structures Phys. Rev. Lett. 93 154302(1-4).
[77]. Åberg M and Gudmundson P 1997 The usage of standard finite element codes for
computation of dispersion relations in materials with periodic microstructure J.
Acoust. Soc. Am. 102 2007-2013.
[78]. Langlet P, Hladky-Hennion A C and Decarpigny J N 1995 Analysis of the
propagation of plane acoustic waves in passive periodic materials using the finite
element method J. Acoust. Soc. Am. 98 2792-2800.
[79]. Liu Z 2000 Locally resonant sonic materials Science 289 1734-1736.
[80]. Sanchez-Perez J V, Rubio C, Martinez-Sala R, Sanchez-Grandia R and Gomez
V 2002 Acoustic barriers based on periodic arrays of scatterers Appl. Phys. Lett.
81 5240-5242.
[81]. Vasseur J O, Deymier P A, Khelif A, Lambin P, Djafari-Rouhani B, Akjouj A,
Dobrzynski L, Fettouhi N and Zemmouri J 2002 Phononic crystal with low
filling fraction and absolute acoustic band gap in the audible frequency range: a
theoretical and experimental study Phys. Rev. E 65 056608(1-6).
[82]. Goffaux C and Vigneron J P 2001 Theoretical study of a tunable phononic band
gap system Phys. Rev. B 64 075118(1-5).
[83]. Yeh J 2007 Control analysis of
the
tunable phononic crystal with
electrorheological material Physica B Condensed Matter, 400 137-144.
[84]. Wang Y, Li F, Kishimoto K, Wang Y, Huang W and Jiang X 2009 Elastic wave
band gaps in magnetoelectroelastic phononic crystals Wave Motion 46 47-56.
[85]. Khelif A, Achaoui Y, Benchabane S, Laude V and Aoubiza B 2010 Locally
resonant surface acoustic wave band gaps in a two-dimensional phononic crystal
of pillars on a surface Phys. Rev. B 81 214303.
[86]. Gonella S and Ruzzene M 2008 Analysis of in-plane wave propagation in
hexagonal and re-entrant lattices J. Sound Vib. 312 125 -- 139.
[87]. Collet M, Ouisse M, Ruzzene M and Ichchou M N 2011 Floquet -- Bloch
decomposition for the computation of dispersion of two-dimensional periodic,
damped mechanical systems Int. J. Solids Struct. 48 2837 -- 2848.
143
[88]. Robillard J, BouMatar O, Vasseur J O, Deymier P A, Stippinger M, Hladky-
Hennion A, Pennec Y and Djafari-Rouhani B 2009 Tunable magnetoelastic
phononic crystals Appl. Phys. Lett. 95 124104(1-3).
[89]. Sukhovich A, Jing L and Page J H 2008 Negative refraction and focusing of
ultrasound in two-dimensional phononic crystals Phys. Rev. B 77 014301(1-9).
[90]. Goffaux C and Vigneron J P 2001 Theoretical study of a tunable phononic band
gap system Phys. Rev. B 64 075118(1-5).
[91]. Deymier P A 2013 Acoustic metamaterials and phononic crystals Springer Berlin
253 -- 372.
[92]. Xu Z, Wu F and Guo Z 2013 Shear-wave band gaps tuned in two-dimensional
phononic crystals with magnetorheological material Solid State Commun. 154 43 --
45.
[93]. Bou Matar O, Robillard J F, Vasseur J O, Hladky-Hennion A-C, Deymier P A
Pernod P and Preobrazhensky V 2012 Band gap tunability of magneto-elastic
phononic crystal J. Appl. Phys. 111 054901(1-14).
[94]. Ding R, Su X, Zhang J and Gao Y 2014 Tunability of longitudinal wave band
gaps in one dimensional phononic crystal with magnetostrictive material J. Appl.
Phys. 115 074104(1-8).
[95]. Sukhovich A, Jing L and Page J H 2008 Negative refraction and focusing of
ultrasound in two-dimensional phononic crystals Phys. Rev. B 77 014301(1-9).
[96]. Otsuka P H, Nanri K, Matsuda O, Tomoda M, Profunser D M, Veres I A,
Danworaphong S, Khelif A, Benchabane S, Laude V and Wright O B 2013
Broadband evolution of phononic-crystal-waveguide eigenstates in real- and k-
spaces Sci. Rep. 3 3351.
[97]. Wang P, Shim J and Bertoldi K 2013 Effects of geometric and material
nonlinearities on tunable band gaps and low-frequency directionality of phononic
crystals Phys. Rev. B 88 014304
[98]. Bertoldi K and Boyce M C 2008 Wave propagation and instabilities in monolithic
and periodically structured elastomeric materials undergoing large deformations
Phys. Rev. B 78 184107(1-16).
[99]. Babaee S, Wang P and Bertoldi K 2015 Three-dimensional adaptive soft
phononic crystals J. Appl. Phys. 117 244903.
[100]. Farhan R Lecture notes ECE 407- Spring 2009 - Cornell University.
[101]. Geryak R and Tsukruk V V 2014 Reconfigurable and actuating structures from
soft materials Soft Matter 10 1246.
[102]. Bertoldi K, Boyce M C, Deschanel S, Prange S M and Mullin T 2008 Mechanics
of deformation-triggered pattern transformations and superelastic behavior in
periodic elastomeric structures J. Mech. Phys. Solids 56 2642 -- 2668.
144
[103]. Kamat P V, Thomas K G, Barazzouk S, Girishkumar G, Vinodgopal K and
Meisel D 2004 Self-assembled linear bundles of single wall carbon nanotubes and
their alignment and deposition as a film in a dc field J. Am. Chem. Soc. 126
10757.
[104]. Fuhrer R, Athanassiou E K, Luechinger N A and Stark W J 2009 Crosslinking
metal nanoparticles into the polymer backbone of hydrogels enables preparation
of soft, magnetic field-driven actuators with muscle-like flexibility Small 5 383.
[105]. Ionov L 2011 Soft microorigami: self-folding polymer films Soft Matter 7 6786-
6791.
[106]. Babaee S, Shim J, Weaver J C, Chen E R, Patel N and Bertoldi K 2015 3D soft
metamaterials with negative Poisson's ratio Adv. Mater 25 5044-5049.
[107]. Goncu F, Luding S and Bertoldi K 2012 Exploiting pattern transformation to tune
phononic band gaps in a two-dimensional granular crystal J. Acoust. Soc. Am. 131
EL475.
[108]. Rudykh S and Boyce M C 2014 Transforming wave propagation in layered media
via instability-induced interfacial wrinkling Phys. Rev. Lett. 112 034301.
[109]. Cao Y and Hutchinson J W 2012 Wrinkling phenomena in neo-Hookean
film/substrate bilayers J. A Mech. 79 031019.
[110]. Saha S K and Culpepper M L 2012 Predicting the quality of one-dimensional
periodic micro and nano structures fabricated via wrinkling in Proceedings of the
ASME 2012 International Mechanical Engineering Congress and Exposition,
Houston TX USA.
[111]. Schweikart A and Fery A 2009 Controlled wrinkling as a novel method for the
fabrication of patterned surfaces Microchim. Acta 165 249.
[112]. Li Y, Kaynia N, Rudykh S and Boyce M C 2013 Wrinkling of interfacial layers in
stratified composites Adv. Eng. Mater. 15 921.
[113]. Zhao Y, Cao Y, W. Hong W, Wadee M K and Fen X-Q 2015 Towards a
quantitative understanding of period-doubling wrinkling patterns occurring in
film/substrate bilayer systems Proc. R. Soc. A 471 20140695.
[114]. Cai S, Breid D, Crosby A J, Suo Z and Hutchinson J W 2011 Periodic patterns
and energy states of buckled films on compliant substrates J. Mech. Phys. Solids
59 1094.
[115]. Huang Z, Hong W and Suo Z 2004 Evolution of wrinkles in hard films on soft
substrates Phys. Rev. E 70 030601.
[116]. Damil N, Potier-Ferry M and Hu H 2014 Membrane wrinkling revisited from a
multi-scale point of view Adv. Model. Simul. Eng. Sci. 1 6.
[117]. Taylor M, Bertoldi K and Steigmann D J 2014 Spatial resolution of wrinkle
patterns in thin elastic sheets at finite strain J. Mech. Phys. Solids 62 163.
145
[118]. Cerda E and Mahadevan L 2003 Geometry and physics of wrinkling Phys. Rev.
Lett. 90 074302.
[119]. Bowden N, Brittain S, Evans A, Hutchinson J W and Whitesides G 1998
Spontaneous formation of ordered structures in thin films of metals supported on
an elastomeric polymer Nature 393 146.
[120]. Chan E P, Karp J M, and Langer R S 2011 A self-pinning adhesive based on
responsive surface wrinkles J. Polym. Sci. B. 49 40.
[121]. Chen C M and Yang S 2012 Wrinkling instabilities in polymer films and their
applications Polym. Int. 61 1041.
[122]. Chandra D, Yang S and Lin PC 2007 Strain responsive concave and convex
microlens arrays Appl. Phys. Lett. 91 251912.
[123]. Chan E P and Crosby A J 2006 Fabricating microlens arrays by surface wrinkling
Adv. Mater. 18 3238.
[124]. Khang DY, Jiang H, Huang Y and Rogers J A 2006 A stretchable form of single-
crystal silicon for high-performance electronics on rubber substrates Science 311
208.
[125]. Chen Y and Crosby A J 2014 High aspect ratio wrinkles via substrate prestretch
Adv. Mater. 26 5626.
[126]. Sun J-Y, Xia S, Moon M-W, Oh K H and Kim K-S 2012 Folding wrinkles of a
thin stiff layer on a soft substrate Proc. Royal Soc. London Ser. A 468 932 -- 953.
[127]. Khelif A, Achaoui Y, Benchabane S, Laude V and Aoubiza B 2010 Locally
resonant surface acoustic wave band gaps in a two-dimensional phononic crystal
of pillars on a surface Phys. Rev. B 81 214303.
[128]. Pennec Y, Vasseur J-O, Djafari-Rouhani B, Dobrzyński L and Deymier P A 2010
Two-dimensional phononic crystals: Examples and applications Surf. Sci. Rep. 65
229-291.
[129]. Lucklum R 2014 Phononic crystals and metamaterials - Promising new sensor
platforms Procedia Eng. 87 40-45.
[130]. Olsson III R H, El-Kady I F, Su M F, Tuck M R and Fleming J G 2008
Microfabricated VHF acoustic crystals and waveguides Sens. Actuators A 145 87-
93.
[131]. Mohammadi S, Eftekhar A A, Pourabolghasem R and Adibi A 2011
Simultaneous high-Q confinement and selective direct piezoelectric excitation of
flexural and extensional lateral vibrations in a silicon phononic crystal slab
resonator Sens. Actuators A 167 524-530.
[132]. Safavi-Naeini A H and Painter O 2010 Design of optomechanical cavities and
waveguides on a simultaneous bandgap phononic-photonic crystal slab Opt.
Express 18 14926-14943.
146
[133]. Triantafyllidis N, Nestorovic M D and Schraad M W 2006 Failure surfaces for
finitely strained two-phase periodic solids under general in-plane loading J. Appl.
Mech.73 505 -- 516.
[134]. Ruzzene M, Scarpa F and Soranna F 2003 Wave beaming effects in two-
dimensional cellular structures J. Smart Mater. Struct. 12 363-372
[135]. Ruzzene M and Scarpa F 2005 Directional and band-gap behavior of periodic
auxetic lattices Phys. Status solidi B 242 665 -- 680.
[136]. Introduction to Comsol Multiphysics. User Guide.
[137]. M. Attard 2003 Finite strain - isotropic hyperelasticity Int. J. Solids Struct. 40
4353 -- 4378.
|
1908.00088 | 1 | 1908 | 2019-07-31T20:41:13 | Advanced material system for the design of an intermediate band solar cell: type-II CdTe quantum dots in a ZnCdSe matrix | [
"physics.app-ph"
] | We explore CdTe fractional monolayer quantum dots (QDs) in a ZnCdSe host matrix for potential application in an intermediate band solar cell device. Careful consideration has been taken during the initiation of the growth process of QDs by migration enhanced epitaxy, in order to avoid the formation of undesirable interfacial layers that can form due to the lack of common anion between the two materials. A superlattice structure of 100 periods of alternating QD and spacer layers is analyzed by high resolution X-ray diffraction (XRD) and photoluminescent (PL) spectroscopy. Simple arguments are used following continuum elastic theory to deduce the size of the dots and the strain within the superlattice from XRD data. This is further verified using PL and used in the energy calculations that yield the values of the intermediate band energy. The results suggest that the optimized materials are highly suitable for these high efficiency solar cells. | physics.app-ph | physics | Advanced material system for the design of an intermediate band solar
cell: type-II CdTe quantum dots in a ZnCdSe matrix
V. Deligiannakis1,3, G. Ranepura1, I. L. Kuskovsky2, 3, M. C. Tamargo1,3
1City College of CUNY, 160 Convent Ave., New York, New York 10031, USA
2Queens College of CUNY, 65-30 Kissena Blvd, Queens, New York 11367, USA
3The Graduate Center of CUNY, 365 Fifth Avenue, New York, New York 10016, USA
[email protected]
Abstract: We explore CdTe fractional monolayer quantum dots (QDs) in a ZnCdSe host matrix for potential application in an intermediate band
solar cell device. Careful consideration has been taken during the initiation of the growth process of QDs by migration enhanced epitaxy, in order
to avoid the formation of undesirable interfacial layers that can form due to the lack of common anion between the two materials. A superlattice
structure of 100 periods of alternating QD and spacer layers is analyzed by high resolution X-ray diffraction (XRD) and photoluminescent (PL)
spectroscopy. Simple arguments are used following continuum elastic theory to deduce the size of the dots and the strain within the superlattice
from XRD data. This is further verified using PL and used in the energy calculations that yield the values of the intermediate band energy. The
results suggest that the optimized materials are highly suitable for these high efficiency solar cells.
Introduction:
The goal of higher efficiency in solar cells has been driving
solar cell research since their inception, and its significance has
never been greater than it is today. One possible solution to
overcoming the single junction solar cell efficiency imposed
by the Schockly-Quisser limit1 of around 31% is by introducing
an intermediate band (IB) within a host material, as in the so
called intermediate band solar cell (IBSC)2. Using either
epitaxial quantum dots (QDs) or impurities one can introduce
such a band3,4. A two step-photon process occurs from the
valence band (VB) to the IB and from IB to the conduction
band (CB), ultimately increasing light absorption of the solar
spectrum without compromising the open circuit voltage (VOC)
of the device5. The proof of concept for an IBSC has been
realized, and it has been shown that the short circuit current can
be increased by the introduction of QDs6. However, with most
materials used so far there is a significant drop in the VOC,
ultimately limiting the overall efficiency of the device with
regards to a reference solar cell7.
InP has an bandgap of ~2.1 eV, in which the Zn(Cd)Te QDs
can form an intermediate band with an energy 0.3 -- 0.7 eV
above the VB edge. The similarity of these parameters with
those required for an ideal IBSC11 makes this material system
an outstanding candidate. However, it was recently shown12
that at the interface between the host material of ZnCdSe and
the QDs an unintentional highly strained ZnSe interfacial layer
is formed. If this is not accounted for, the strain accumulation
in the thick stacked QD superlattice can be significant enough
to lead to the formation of defects affecting the device
performance. As is the case for QDs grown by the Stranski-
Krastanov (SK) method, which require the formation of a
strained wetting layer, as the QD layers are repeated multiple
times the strain in the structure increases, decreasing the
overall quality of the material13,14,15. The presence of an
interfacial layer can also affect the bandstructure of the device.
Our group recently showed that using a new shutter growth
sequence we can significantly suppress the formation of the
interfacial layer16. Now that we are able to control the interface
formation, we set out to explore new QD compositions and
their potentially advantageous properties. Here we pursue a
new material system, based on sub-monolayer CdTe QDs
Type-II Zn(Cd)Te /ZnCdSe submonolayer QDs have been
explored by our group for their promising properties as
IBSCs8,9,10.The ZnCdSe host material when lattice matched to
embedded in the ZnCdSe host material. Besides providing a
platform in which the ZnSe interfacial layer is more fully
suppressed, this system has several advantages over the
ZnCdTe QD system previously studied. Two main advantages
are 1) the binary composition of the QD which makes it more
easily controlled and more uniform, and 2) the large
compressive strain in the QDs which produces a large valence
band offset with respect to the matrix material (ZnCdSe) that
can be exploited for better devices via band structure
engineering. This strain can be easily offset by strain
compensation in the spacer regions.
Growth:
A schematic of the structure investigated is shown in Fig. 1(a).
A CdTe/ZnCdSe QD superlattice (SL) is grown by a
combination of conventional molecular beam epitaxy (MBE)
and migration enhanced epitaxy (MEE). The initial buffer
layers and spacer material (ZnCdSe) were grown by MBE,
whereas to achieve the formation of sub- monolayer CdTe QDs
the MEE process was implemented.
The sample was grown on (001) oriented InP substrates with a
100 nm lattice matched InGaAs buffer layer in a dual chamber
MBE growth system. The substrate and buffer layer were
transferred to the II-VI chamber via ultra-high vacuum transfer
modules. The II-VI layer structure included a 100 nm ZnCdSe
buffer
layer, followed by a superlattice consisting of
alternating 13 to 14 monolayers of the ZnCdSe spacer layer and
the submonolayer CdTe QDs. The formation of CdTe QDs by
MEE was achieved by using a special shutter sequence of
alternating Cd and Te fluxes with short wait times between
them.
The shutter sequence recently developed by our group
entailed a growth interruption with exposure of only a Cd flux
for 5 s after the growth of the ZnCdSe spacer, followed by a
short wait time of 1 s without any shutters open, instead of the
5 s wait times previously performed.12 We have found that
terminating the surface with only Cd rather than with both
group II elements of Zn and Cd is an important and critical step
for the avoidance of the IF layer formation15. The surface
termination of a metal rich surface is observed by a change
from a VI-rich (2×1) terminated RHEED pattern, observed
during the ZnCdSe spacer layer growth, to a II-rich c(2×2)
pattern after the Cd-only exposure. To grow our CdTe QDs we
employed the same initial Cd-only exposure (5 s) and 1 s wait
(a)
(b)
Figure 1. (a) Schematic of type-II sub-monolayr QD stacks, that
would be implemented as an intermediate band region in an IBSC
device. (b) Growth sequence used for the formation of QDs grown
by migration enhanced epitaxy.
time without any impingent fluxes. After this the Cd shutter
was opened for 5s, and then closed and followed by a 1s wait
time. This sequence was repeated three times (three cycles).
The shutter opening and closing cycles used are shown in Fig.
1(b). At the end of the MEE sequence the Cd shutters were
opened for 5s before resuming the ZnCdSe spacer layer growth
(by opening the Se and Zn shutters). It should be noted that in
this sequence, in contrast to the previous sequence (e.g., Refs.
12) the Zn shutter remains closed throughout the entire MEE
cycle, further minimizing the likelihood of any unintentional
ZnSe being formed during the process.
(a)
(b)
(c)
Figure 2 (a) HR-XRD along the (004) reflection; (b) The (224) asymmetric reflection. In conjunction, these two plots allow us to
accurately measure the size and strain in the QDs. (c) Thickness of the QDs necessary to position the SL(0) peak under high compressive
strain with respect to the substrate as observed in HR-XRD.
Results:
In Figs. 2(a) and 2(b) a high-resolution X-Ray diffraction
(HR_XRD), for symmetric (004) and asymmetric (224) Bragg
reflections, respectively, 2θ-ω scans are presented. The Bragg
diffraction peaks of the different layers are clearly resolved.
They include (see Fig. 2(a)): the ZnCdSe buffer layer (63.18),
the zeroth (0th)-order superlattice peak SL(0) (62.40) as well
as higher order satellite peaks, which originate from the
periodic CdTe/ZnCdSe QD SL structure. Such strong and
sharp higher order satellite peaks suggest high quality materials
and interfaces: well-controlled separation of QD layers and
well-contained spacer segregation of materials. By combining
the symmetric (004) and asymmetric (224) Bragg reflections
the strain and composition of the different layers within the
structure can be accurately calculated.
The ZnCdSe buffer layer has an in-plane lattice parameter,
𝑆𝑝
𝑆𝑢𝑏 =
𝑎∥
5.869 Å. The layer is under slight compressive strain, as its
𝑆𝑝 = 5.882 Å, is slightly
out-of-plane lattice parameter, 𝑎⊥
larger. The in-plane and out-of-plane lattice parameters for the
𝑆𝐿 = 5.948 Å.
SL(0) are, respectively, 𝑎∥
These values suggest that the superlattice region is nearly
pseudomorphic to the InP substrate, due to the small difference
between the in-plane lattice parameters which is only 0.22%.
The SL period is made up of the combination of a spacer layer
of the same composition as that of the ZnCdSe buffer layer and
the CdTe QD layer, which we will show can be modelled as
, that is equal to that of the InP substrate, 𝑎∥
𝑆𝐿 = 5.882 Å and 𝑎⊥
fractional layers of CdTe. For a more complete understanding
of the energy at which the IB level forms within the host
material, a good understanding of the strain and the thickness
of the quantum dots is necessary. For this we utilize the
symmetric and anti-symmetric HR-XRD scans (Figs. 2(a) and
2(b)) using the following simple arguments and considerations.
From the (004) scan the out-of-plane thickness of the
period can be obtained by the distance between higher order
𝑆𝐿 = 43Å.
superlattice peaks and
Considering the superlattice structure as a pseudo-crystal, we
𝑆𝐿 as the weighted average of the
describe the lattice constant 𝑎⊥
strained individual layers that make up the period:
is calculated
to be 𝑡⊥
𝑆𝐿 =
𝑎⊥
𝑄𝐷𝑠𝑡⊥
𝑎⊥
𝑄𝐷𝑠 + 𝑎⊥
𝑆𝑝𝑡⊥
𝑆𝑝
𝑆𝐿
𝑡⊥
(1)
are the thickness of the ZnCdSe spacer and
𝑄𝐷
, and 𝑡⊥
𝑆𝑝
where, 𝑡⊥
thickness of effective CdTe QD
the average
𝑄𝐷 and 𝑎⊥
respectively; 𝑎⊥
the individual materials.
𝑆𝑝
are the strained lattice constants of
layer,
Fig. 2(c) relates the QD thickness and amount of strain
imparted on the QDs necessary to match the measured out-of-
plane lattice parameter of the SL from Eq. (1). For the
unstrained case (0% strain) a significantly thicker dot is
necessary, more than double than for the fully strained case
(100% strain). By using Eq. (2), which relates the out of-plane-
lattice parameter to the elastic constants of the SL, we can
accurately calculate the thickness and strain of the QDs for this
sample17:
52545658606264666870 − (deg.)SL(-4)SL(-3)SL(-2)SL(+2)ZnCdSeSL(0)SL(+1) Intensity (arb. units)(004)SL(-1)InP(sub.)7778798081 − (deg.)ZnCdSeSL(0)SL(+1)InP(sub.)(224) Intensity (arb. units)SL(-1)2.53.03.54.04.55.05.5020406080100 % Strain in QDsQD Thickness (A)
𝑆𝐿 = (1 +
𝑎⊥
𝑆𝐿
2𝐶12
𝑆𝐿 ) (𝑎0
𝐶11
𝑆𝐿 − 𝑎∥
𝑆𝐿) + 𝑎∥
𝑆𝐿
(2)
𝑆𝐿
𝑁𝑆𝑝 = 2𝑡⊥
(𝑎⊥
𝑆𝑝(𝑎⊥
𝑎⊥
𝑄𝐷 − 𝑎⊥
𝑆𝐿)
𝑄𝐷 − 𝑎⊥
𝑆𝑝)
(7)
𝑆𝐿 and 𝐶11
𝑆𝐿 are the weighted averages of the elastic
Here 𝐶12
constants of the individual unstrained CdTe and ZnCdSe layers
given by18:
Inserting Eqs. (6), (7) and (3) back into Eq. (2) an expression
𝑄𝐷
in which 𝑎⊥
formulated:
is the only unknown parameter, can be
(3)
(𝑎⊥
𝑆𝐿 − 𝑎∥
𝑆𝐿) = (1 + 2
𝛽𝑎⊥
𝛽𝑎⊥
𝑆𝑝𝐶12
𝑆𝑝𝐶11
𝑄𝐷𝑠 + 𝑥𝑎⊥
𝑄𝐷𝑠 + 𝑥𝑎0
𝑆𝑝𝐶12
𝑆𝑝𝐶11
𝑆𝑝
𝑆𝑝) ×
(8)
𝑆𝐿 =
𝐶𝑖𝑗
𝑄𝐷𝑠𝑡0
𝐶𝑖𝑗
𝑄𝐷𝑠 + 𝐶𝑖𝑗
𝑆𝑝𝑡0
𝑆𝑝
𝑆𝐿
𝑡0
𝑄𝐷𝑠
𝑆𝑝
, 𝑡0
, 𝑡0
𝑆𝐿 are thickness of the unstrained
where 𝑡0
corresponding layers. These thicknesses, as well as the
thickness of strained layers can be conveniently expressed in
terms of the number of corresponding monolayers, 𝑁(𝑖), which
stay constant between the strained and unstrained cases:
𝑡⊥,0
𝑆𝐿 =
𝑄𝐷𝑠
𝑎⊥,0
2
𝑆𝑝
𝑎⊥,0
2
𝑁𝑆𝑝
𝑁𝑄𝐷𝑠 +
Correspondingly,
𝑎0,⊥
𝑆𝐿 =
𝑄𝐷𝑠
𝑄𝐷𝑠 𝑎⊥
𝑎0,⊥
2
𝑁𝑄𝐷𝑠 + 𝑎0,⊥
𝑆𝑝
𝑆𝑝 𝑎0,⊥
2
𝑆𝐿
𝑡0,⊥
𝑁𝑆𝑝
(4)
(5)
and 𝑎⊥
We note here that using Eq. (4), weighted averages for elastic
constants are also expressed in terms of number of monolayers
and lattice constants, instead of layer thicknesses.
In Eq. 5 important parameters that are to be considered are
𝑆𝑝
𝑆𝑝. The former is known, since the composition of the
𝑎0
spacer is the same as that of the buffer; however, the latter is
not since we don't know the degree of relaxation, if any, in the
spacers. We, nevertheless, know the range in which the in-
𝑆𝑝
plane lattice parameter of the spacer, 𝑎∥
𝑆𝑝
between 𝑎𝑆𝑢𝑏 (fully strained) and 𝑎0
Therefore, we will take this, in further calculations, as a
varying parameter, which, as shown below, uniquely
𝑄𝐷𝑠) for our given
determines the unknowns (𝑎⊥
case via the measured SL lattice parameters and period.
𝑄𝐷𝑠, 𝑁𝑄𝐷𝑠, 𝑡⊥
can vary. This is
(fully relaxed).
Using Eqs. (4) and (5) we can get expressions for number
of monolayers as follows:
𝑆𝐿
𝑁𝑄𝐷 = 2𝑡⊥
(𝑎⊥
𝑄𝐷(𝑎⊥
𝑎⊥
𝑆𝑝)
𝑆𝐿 − 𝑎⊥
𝑄𝐷 − 𝑎⊥
𝑆𝑝)
(6)
(
𝛼𝑎0
𝑆𝑝
𝑄𝐷𝑠 + 𝑥𝑎0
𝛼 + 𝑥
− 𝑎∥
𝑆𝐿)
𝑄𝐷𝑠
𝑄𝐷𝑠(𝑎⊥
𝑄𝐷 − 𝑎⊥
𝑄𝐷(𝑎⊥
𝑆𝑝); 𝑎⊥
𝑆𝑝), 𝛽 =
𝑆𝐿), 𝛼 = 𝑎0
𝑆𝐿 − 𝑎0
𝑆𝑝
can be calculated from 𝑎∥
Here 𝑥 = 𝑎⊥
𝑄𝐷𝑠(𝑎⊥
𝑆𝐿 − 𝑎⊥
𝑎0
equation similar to Eq. (2). Eq. (8) can be solved for 𝑎⊥
and
ultimately for 𝑁𝑄𝐷𝑠, 𝑁𝑆𝑝 and the thicknesses of both the QDs
and the spacer.
via an
𝑄𝐷𝑠
The out-of plane lattice parameter for the QDs is calculated
𝑄𝐷𝑠 = 7.037 Å. This value does not change much when
to be 𝑎⊥
varying the spacer in-plane lattice parameter (effectively
changing the strain) within the range we are working in. But,
the fractional coverage of the CdTe QDs changes in our
calculation significantly, as illustrated in Figure 3. The
effective number of monolayers varies from 0.66 to 0.76. In
both scenarios (strained and unstrained spacer) the average
thickness is submonolayer in quantity for the CdTe QDs.
Figure 3 QD height (in monolayers) and effective QD thickness are
plotted with respect to the strain of the spacer. Sub-monolayer
quantities are found to explain the experimental data.
(a)
(b)
Figure 4 (a) Low temperature intensity dependent PL from the QDs as well from the ZnCdSe spacer and buffer region. The inset illustrates
the type-II band alignment in which holes are confined within the QDs and electrons within ZnCdSe spacer. (b) Intensity dependent PL
measurement confirms the type-II band alignment between QDs and host material. (inset) Plot of peak position of the QD PL as a function
of excitation intensity, exhibiting a 1/3 power law fit. The energy of the ZnCdSe peak is plotted for comparison.
In order to investigate the suitability of the structure for an
IBSC, we measured its luminescent properties. Figure 4a
shows the PL spectrum of the structure taken at 75 K. PL
emission was acquired by exciting the sample using a 50 mW
405 nm diode laser and the collected PL was analyzed with a
HR4000 Ocean Optics spectrometer. In figure 3a two peaks
can be identified from the spectrum. A sharp peak around 2.04
eV is due to the ZnCdSe buffer and spacer layers. The
difference in energy from the bandgap energy of a lattice
matched Zn0.51Cd0.49Se to InP (around 2.1 eV) is to be expected
since the buffer layer was grown with a slightly Cd rich
composition, as indicated by the XRD19. A broad peak
observed at lower energies, centered around 1.86 eV, is due to
the CdTe QDs. Such a broad peak is reasonable due to expected
size distribution and the type-II nature of the recombination
process. Juxtaposed is a spectrum taken at a lower excitation
intensity and a noticeable shift to lower energy of the QD peak
by as much as 40 meV is observed.
Intensity dependent PL shown in Figure 4(b) was taken to
verify the type-II band alignment for the low energy peak. In
a type- II heterostructure, with higher excitation intensity an
increased flux of electrons is promoted into the conduction
band, producing a band bending effect at the interface between
the QDs and the host material. The staggered band alignment
between the two materials forms a triangular potential well at
this interface in which the energy scales as a function of the
cube root of the excitation intensity20. In the inset in figure 4(b),
the peak position shift is plotted for the QDs and the ZnCdSe
spacer. There is a clear shift in energy with excitation power
for the QD PL that fits very well with a 1/3 power law fit,
whereas the peak of the maximum excitation position of the
ZnCdSe does not change with intensity. At the lowest
excitation intensity, assuming a flat band potential between the
QDs and the host material, a difference in energy of ~200 meV
is observed between the band to band transition in the barrier
layer (2.04 eV) and the type II transition arising from the QDs.
This value suggests that the QDs in our structure are very
small, thus shifting the confined hole energy level to near the
VB edge of the ZnCdSe spacer.
Using Vegard's law and the known band alignments for
these two materials21,22, the band structure for the QD region is
approximated in figure 5(a). There is a large valence band
offset of 0.86 eV between ZnCdSe and CdTe and the large
quantum confinement of the holes due to their very small size
gives rise to the observed 200 meV energy difference.
Increasing the size of the dots will allow the confined energy
level of the QDs to shift closer to the CdTe valence band edge,
thus reducing the type-II transition energy and increasing the
energy difference between the QDs and the host material to the
desired 0.5 -- 0.7 eV range based on the device design. The
large VBO of 0.86 eV provides sufficient tunability of the
confinement energy in order to achieve these values. However,
since the heavy hole energy level also depends strongly on the
strain of the QD within the host material, strain effects must be
1.31.41.51.61.71.81.92.00.00.20.40.60.81.01.21.41.61E-41E-30.010.111.821.831.841.851.861.871.88 CdTe QDsFitZnCdSeExcitation Intensity (arb. units)QD PL Peak Energy (eV)1.851.901.952.002.052.10 ZnCdSe PL Peak Energy (eV) Normalized Intensity (arb. units)Energy (eV) IMax IMaxx10-2 IMaxx10-1.5 IMaxx10-2 IMax x10-3 IMaxx10-41.83eV1.87eV1.41.51.61.71.81.92.02.12.20.00.20.40.60.81.01.2 IMax IMaxx10-4 Intensity (arb. units)Energy (eV)ZnCdSeSpacer/Buffer2.04eV75K1.83eV1.87eV (a)
Unstrained
(b)
Strained
Figure 5 (a) Schematic band diagram between unstrained CdTe
and ZnCdSe lattice matched to InP. (b) Band energy diagram
from calculated values of composition ascertained from HR-
XRD and taking into account for effects of strain.
considered. Due to the high lattice mismatch, between ZnCdSe
and CdTe, biaxial compressive strain gives rise to hydrostatic
strain component and to a first order approximation there is a
splitting of the light and heavy hole levels due to the shear
component of the strain that can be calculated by standard
deformation potential theory.23 The equation below calculates
the new heavy hole (HH) energy due to strain,
Δ𝐸ℎℎ = 𝑎𝜈(𝜀𝑥𝑥 + 𝜀𝑦𝑦 + 𝜀𝑧𝑧) +
(𝜀𝑥𝑥 + 𝜀𝑦𝑦 − 2𝜀𝑧𝑧)
(9)
𝑏
2
Where 𝑎𝜈 is the linear hydrostatic deformation potential for the
valence band maxima and 𝑏 is the shear uniaxial deformation
potential. The strain components are given by,
𝜀 = 𝜀𝑥𝑥 = 𝜀𝑦𝑦 =
𝑎𝑆𝑢𝑏 − 𝑎𝑄𝐷𝑠
𝑎𝑆𝑢𝑏
and
(10)
(11)
𝜀𝑧𝑧 =
−2𝐶12
𝐶11
𝜀
Typically, compressive strain increases the band gap of the
material and, depending on how large the strain is, it could
significantly increase the VBO in a heterostructure. In our case
this is highly desirable and allows for another tuning parameter
to obtain the idealized value for the IB band. Due to the large
lattice mismatch between the CdTe QDs and the spacer regions
a large shift in the heavy hole energy can ultimately be
observed that would be beneficial in being able to tune the
energy level of the IB band. From our calculation in HR-XRD
we find that the QDs are partially relaxed (35%). Thus, we find
the VBO is now ~ 1.0 eV. The HH energy is calculated, using
self-consistent variational calculation of the Schrödinger
equation,24 for a single monolayer of CdTe under these strain
conditions, to be 289 meV, illustrated in figure 5(b). This is in
close agreement of our PL results. Below is a list of the
materials parameters used for these calculations.
Table I.
Material Parameters used in calculations
c11 (x10 10 Pa)
c12 (x10 10 Pa) (eV)
(eV)
ZnSe
8.26
25
CdSe
6.67
27
CdTe
5.62
25
4.98
25
4.63
27
3.94
25
26
1.23
25
0.90
26
0.89
-1.20
-1.26
-1.20
26
27
25
Conclusion:
We have successfully grown submonlayer CdTe QDs within a
ZnCdSe host. The structural quality of the material was
confirmed by the HR-XRD and their optical properties were
established using intensity dependent PL. These materials have
interesting properties that could be used towards the design of
an ideal IBSC. The large VBO offset that exists between CdTe
and the ZnCdSe host material, which is sufficient to provide
the needed IB energy values, can be further tuned with strain.
To have an accurate understating of the strain of the QDs we
used HR-XRD to analyze the superlattice structure. We found
that the dots were partially relaxed, and their fractional
coverage could be extracted from the strain measured within
the spacer region. Even though the dots were partially relaxed,
our calculations show that we should expect an increase in the
VBO of about 140 meV due to strain, which is helpful towards
tuning the IB energy level to the desired value of 0.7eV. From
the PL measurements and energy calculations we concluded
that the dots in the current structure are too small (~1
monolayer in height) and thus the growth must be modified to
achieve larger (thicker) dots.
Acknowledgements: This work was supported by NSF CBET
award number 1512017 and NSF award number HRD-
1547830 (CREST-IDEALS).
References:
[1] W. Shockley, H.J. Queisser, Detailed balance limit of
efficiency of p -- n junction solar cells, Journal of Applied
Physics 32 (1961) 510519.
[2] A. Luque, A. Martí, Increasing the efficiency of ideal solar
cells by photon induced transitions at intermediate levels,
Physical Review Letters 78 (1997) 5014 -- 5017.
[3] A. Martí, L. Cuadra, A. Luque, Quantum dot intermediate
band solar cell, Conference Record of the 28th IEEE
Photovoltaics Specialists Conference, New York, USA, (2000)
940-943.
[4] A. Luque, A. Martí, E. Antolín, C. Tablero, Intermediate
bands versus levels in non-radiative recombination, Physica B
382 (2006) 320 -- 327.
[5] I. Ramiro, J. Villa, C. Tablero, E. Antolín, A. Luque, A.
Martí, Analysis of the intermediate-band absorption properties
of type-II GaSb/GaAs quantum-dot photovoltaics, Physical
Review B 96 (2017) 125422.
[6] (6) S.M. Hubbard, C.D. Cress, C.G. Bailey, R.P. Raffaelle,
S.G. Bailey, D.M. Wilt, Effect of strain compensation on
quantum dot enhanced GaAs solar cells, Applied Physics
Letters 92, (2008) 123512.
[7] L. Zhu, K.H. Lee, M. Yamaguchi, H. Akiyama, Y.
Kanemitsu, K. Araki, N. Kojima, Analysis of nonradiative
recombination in quantum dot solar cells and materials,
Progress in Photovoltaics: Research and Applications (2019),
doi.org/10.1002/pip.3110.
[8] S. Dhomkar, U. Manna, L. Peng, R. Moug, I.C. Noyan,
M.C. Tamargo, I.L. Kuskovsky, Feasibility of submonolayer
ZnTe/ZnCdSe quantum dots as intermediate band solar cell
material system, Solar Energy Materials & Solar Cells, 117
(2013) 604-609.
[9] S. Dhomkar, U. Manna, I.C. Noyan, M.C. Tamargo, I.L.
Kuskovsky, Vertical correlation and miniband formation in
submonolayer Zn (Cd) Te/ZnCdSe type-II quantum dots for
intermediate band solar cell application, Applied Physics
Letters 103 (2014)18, 181905.
[10] C.M. Imperato, G.A. Ranepura, L.I. Deych, I.L.
Kuskovsky, Theoretical Determination of Optimal Material
Parameters for ZnCdTe/ZnCdSe Quantum Dot Intermediate
Band Solar Cells, Journal of Electronic Materials, 47 (2018)
4325 -- 4331.
[11] M.Y. Levy, C. Honsberg, A. Marti, A. Luque, Quantum
Dot Intermediate Band Solar Cell Material Systems with
Negligible Valence Band Offsets, Conference Record of the
IEEE Photovoltaic Specialists Conference (2005) 90-93.
[12] S. Dhomkar, I.L. Kuskovsky, U. Manna, I.C. Noyan, M.C.
Tamargo, Optimization of growth conditions of type-II Zn
(Cd) Te/ZnCdSe submonolayer quantum dot superlattices for
intermediate band solar cells, Journal of Vacuum Science &
Technology B 31(3) (2013).
[13] A. Marti, N. López, E. Antolín, E. Cánovas, A. Luque, C.
R. Stanley, C. D. Farmer, and P. Díaz, Emitter degradation in
quantum dot intermediate band solar cells, Applied Physics
Letters 90 (2007) 233510.
[14] V. Popescu, G. Bester, M.C. Hanna, A.G. Norman, A.
Zunger, Theoretical and Experimental Examination of the
Intermediate-Band
Strain-Balanced
(In,Ga)As/Ga(As,P) Quantum Dot Solar Cells, Physical
Review B 78 (2008) 20532.
[15] A. Takata, R. Oshima, Y. Shoji, K. Akahane, Y. Okada,
Fabrication of 100
InAs/GaNAs strain-
compensated quantum dots on GaAs (001) for application to
intermediate band solar cell, 35th IEEE Photovoltaic
Specialists Conference (2010).
[16] V. Deligiannakis, S. Dhomkar, M.S. Claro, I.L.
Kuskovsky, M.C.Tamargo Interface Modification in Type-II
ZnCdSe/Zn(Cd)Te QDs for High Efficiency Intermediate
Band Solar Cells, Journal of Crystal Growth 512 15 (2019)
203-207.
[17] T. Rieger, T. Riedl, E. Neumann, D. Grutzmacher, J.
Lindner, A. Pawlis, Strain Compensation in Single ZnSe/CdSe
Quantum Wells: Analytical Model and Experimental
Evidence, ACS Applied Materials & Interfaces 9(9) (2017).
[18] S.J. Polly, C.G. Bailey, A.J. Grede, D.V. Forbes, S.M.
Hubbard, Calculation of strain compensation thickness for III --
V semiconductor quantum dot superlattices, Journal of
Crystal Growth 454 (2016).
layer-stacked
Concept
for
[19] A. Sweiti, F. Medina, L. Martinez, A. Lopez-Rivera,
Photoluminescence spectroscopy and effective concentration
determination of CdxZn1−xSe, Semiconductor Science and
Technology, 23(3) (2008) 035019.
[20] F. Hatami, M. Grundmann, N. N. Ledentsov, F.
Heinrichsdorff, R. Heitz, J. Böhrer, Z.I. Alferov, Carrier
dynamics in type-II GaSb/GaAs quantum dots, Physical
Review B, 57(8) (1998) 4635 -- 4641.
[21] M.C. Tamargo, II-VI Semiconductor Materials and their
Applications (Optoelectronic Properties of Semiconductors
and Superlattices) 1st Edition, CRC Press (1997)
[22] M. Ściesiek, J. Suffczyński, W. Pacuski, M. Parlińska-
Wojtan, T. Smoleński, P. Kossacki, A. Golnik, Effect of
electron-hole separation on optical properties of individual
Cd(Se,Te) quantum dots, Physical Review B 93, (2016)
195313.
[23] S.L. Chuang, Physics of Photonic Devices, 2nd ed.
Hoboken: Wiley (2012) 132 -- 142.
[24] I.V. Ponomarev, L.I. Deych, V.A. Shuvayev, A.A.
Lisyansky, Self-consistent approach for calculations of exciton
binding energy in quantum wells, Physica E 25 (2005) 539.
[25] Y.H. Li, X.G. Gong, S.H. Wei, Ab initio all-electron
calculation of absolute volume deformation potentials of IV-
IV, III-V, and II-VI semiconductors: The chemical trends,
Physical Review B 73 (2006) 245206.
[26] C.G. Van de Walle, Band lineups and deformation
potentials in the model-solid theory, Physical Review B 39
(1989) 1871.
(27) H. J. Lozykowski, V.K. Shastri, Excitonic and Raman
properties of ZnSe/ZnCdSe strained layer quantum wells
Journal of Applied Physics 69 (1991) 3235.
|
1907.07325 | 1 | 1907 | 2019-07-17T04:19:14 | Noise Analysis of Photonic Modulator Neurons | [
"physics.app-ph",
"cs.NE",
"eess.SP"
] | Neuromorphic photonics relies on efficiently emulating analog neural networks at high speeds. Prior work showed that transducing signals from the optical to the electrical domain and back with transimpedance gain was an efficient approach to implementing analog photonic neurons and scalable networks. Here, we examine modulator-based photonic neuron circuits with passive and active transimpedance gains, with special attention to the sources of noise propagation. We find that a modulator nonlinear transfer function can suppress noise, which is necessary to avoid noise propagation in hardware neural networks. In addition, while efficient modulators can reduce power for an individual neuron, signal-to-noise ratios must be traded off with power consumption at a system level. Active transimpedance amplifiers may help relax this tradeoff for conventional p-n junction silicon photonic modulators, but a passive transimpedance circuit is sufficient when very efficient modulators (i.e. low C and low V-pi) are employed. | physics.app-ph | physics | Noise Analysis of Photonic Modulator Neurons
Thomas Ferreira de Lima*, Alexander N. Tait, Hooman Saeidi, Mitchell A. Nahmias, Hsuan-Tung Peng,
Siamak Abbaslou, Bhavin J. Shastri, and Paul R. Prucnal
1
9
1
0
2
l
u
J
7
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
5
2
3
7
0
.
7
0
9
1
:
v
i
X
r
a
Abstract -- Neuromorphic photonics relies on efficiently emulat-
ing analog neural networks at high speeds. Prior work showed
that transducing signals from the optical to the electrical domain
and back with transimpedance gain was an efficient approach
to implementing analog photonic neurons and scalable networks.
Here, we examine modulator-based photonic neuron circuits with
passive and active transimpedance gains, with special attention
to the sources of noise propagation. We find that a modulator
nonlinear transfer function can suppress noise, which is necessary
to avoid noise propagation in hardware neural networks. In
addition, while efficient modulators can reduce power for an
individual neuron, signal-to-noise ratios must be traded off with
power consumption at a system level. Active transimpedance am-
plifiers may help relax this tradeoff for conventional p-n junction
silicon photonic modulators, but a passive transimpedance circuit
is sufficient when very efficient modulators (i.e. low C and low
V-pi) are employed.
Index Terms -- Neuromorphic Computing, Neuromorphic Pho-
tonics, Analog Links, Neural Networks
I. INTRODUCTION
T HE GAP between current computing capabilities and
current computing needs ushered research in the field
of neuromorphic computing [1] -- [5]. This new field aims to
bridge the gap between the energy efficiency of von Neumann
computers and the human brain [6], [7]. As a consequence, this
thrust spawned research into novel brain-inspired algorithms
and applications uniquely suited to neuromorphic processors.
These algorithms attempt to solve artificial intelligence tasks in
real-time while using less energy. We posit that we can make
use of the high parallelism and speed of photonics to bring
the same neuromorphic algorithms to applications requiring
multiple channels of multi-gigahertz analog signals, which
digital processing struggles to process in real-time.
By combining the high bandwidth and parallelism of pho-
tonic devices with the adaptability and complexity attained by
methods similar to those seen in the brain, photonic processors
have the potential to be at least ten thousand times faster than
state-of-the-art electronic processors while consuming less
energy per computation [8]. An example of such an application
is nonlinear feedback control; a very challenging task that
involves computing the solution of a constrained quadratic op-
timization problem in real time. Neuromorphic photonics can
enable new applications because there is no general-purpose
*Corresponding author: [email protected]
T.F.L., A.N.T., H.S., M.A.N., H.-T.P., S.A., B.J.S. and P.R.P. are with the
Department of Electrical Engineering, Princeton University, Princeton, NJ
08544, USA.
B.J.S. is with the Department of Physics, Engineering Physics & Astron-
omy, Queen's University, Kingston, ON KL7 3N6, Canada.
A.N.T. is currently with the Physical Measurement Laboratory, National
Institute of Standards and Technology, Boulder, CO 80305, USA.
hardware capable of dealing with microsecond environmental
variations [9].
These benefits can be accomplished by use of wavelength-
division multiplexing (WDM), which explores the enormous
bandwidth of optical waveguides (∼THz). Integrated neuro-
morphic circuits based on WDM can now be manufactured
using silicon photonics platforms. Tait et al. recently demon-
strated a way of performing neural computations on WDM
signals via a photodetector directly driving a modulator [10],
which is capable of integrating hundreds of wavelength chan-
nels carrying gigahertz signals. We call this an O/E/O-based
photonic neuron (Fig. 1).
In this architecture, photonic neurons output optical sig-
nals with unique wavelengths. These are multiplexed into a
single waveguide and broadcast to all others, weighted, and
photodetected. Each connection between a pair of neurons is
configured independently by one MRR weight [11], [12], and
the WDM carriers do not mutually interfere when detected by
a single photodetector. Consequently, the physics governing
the neural computation is fully analog and does not require any
logic operation or sampling, which would involve serialization
and sampling. Thus they exhibit distinct, favorable trends in
terms of energy dissipation, latency, cross-talk and bandwidth
when compared to electronic neuromorphic circuits [8, Sec.
5].
But the same physics also introduce new challenges, espe-
cially reconfigurability, integration, and scalability. Informa-
tion carried by photons is harder to manipulate compared to
electronic signals, especially nonlinear operations and memory
storage. Photonic neurons described here solve that problem by
using optoelectronic components (O/E/O), which can be mated
with standard electronics providing reconfigurability. However,
neuromorphic photonic circuits are challenging to scale up
because they do not benefit from digital information, memory
units and a serial processor, and therefore requires a physical
unit for each element in a neural network, increasing size, area
and power consumption. Here, integration costs must also be
considered, since the advantages of using analog photonics
(high parallelism and high bandwidth) must outweigh the costs
of interfacing it with digital electronics (requiring both O/E
and A/D conversion). The optimal cost-benefit tradeoff can
be computed for particular applications by engineers, but one
factor that has not been addressed in prior literature is the
accumulation of noise in O/E/O neuromorphic analog links.
This paper will provide a quantitative study in this dimension
of O/E/O neural networks.
A. Analog Processing
The main advantage of this O/E/O approach is that it solves
the problem of transferring energy between WDM signals
into a single wavelength output. This property enables O/E/O
neurons to be networked together via broadcast waveguides.
Fig. 1. Diagram of an O/E/O photonic neuron.
that
The fact
lightwaves are transduced into electrical
currents and back is very advantageous for implementing
nonlinear operations, compared to all-optical strategies. If all
devices worked perfectly, this scheme would work even with
extremely low power and voltage levels. At the low-energy
limit, one photon transforms into a electron-hole pair, whose
charge can be used to modulate the transmission of an electro-
optic device, sourcing a number of photons at the output of
the circuit (Fig. 1). This mode of operation would require
lossless optics, coupled with very sensitive photodetectors and
modulators.
Because of such loss and inefficiency, we need to operate
the neuron with stronger signals. In addition, due to the analog
nature of this communication scheme, it is not immune to
noise accumulation. Ultimately it is the shot noise and the
noise of circuit components that currently prevents driving
these O/E/O photonic neurons with quantum-scale power
levels at room temperature.
Fig. 2. Photonic neurons that receive inputs from one WDM broadcast
medium and transmit into another, corresponding to the next layer. Distinct
layers can thus reuse the same optical spectrum for broadcasting, much like
a cellular telephone network reuses spectrum geographically. This strategy of
using more waveguides for more layers can be extended, in principle, to an
arbitrary depth.
In order for one layer to physically drive the next layer,
enough laser power PL must be provided to compensate for
loss and power splitting due to fan-out
layer
(NFO). Alternatively, electrical gain via a high transimpedance
(RTIA) can provide the necessary amplification. This "gain
to the next
2
cascadability" condition can be written with the following
inequality:
RTIA(cid:124)(cid:123)(cid:122)(cid:125)
electrical
(cid:18) MD
(cid:19)
(cid:124) (cid:123)(cid:122) (cid:125)
Vpp
mod. sens.
· PL(cid:124)(cid:123)(cid:122)(cid:125)
optical
·
>
NFO
2T1/2Rdηpp
(1)
where MD is the modulation depth; T1/2, mean transmission,
Rd, photodiode's responsivity, and ηpp is the optical point-
to-point efficiency between connected neurons, representing
excess optical loss between the output of one neuron and the
input of the next, i.e. propagation loss and insertion loss of
weighting devices and couplers.
Equation 1 describes the relationship between different
kinds of gain: electrical, optical, modulator sensitivity. We
draw attention to the RTIA/Vpp ratio (transimpedance over
peak-to-peak voltage), which represents the neuron's sensitiv-
ity, because it quantifies how much photocurrent is necessary
to effect a full amplitude swing in the modulator. Since optical
pump power (PL) is an expensive resource, it is desirable to
maximize this sensitivity in order to minimize overall power
consumption. However, higher sensitivity comes at a price, as
noise accumulation degrades the signal-to-noise ratio at the
output (Sec. II).
B. Suppressing Noise Accumulation
Without careful design, analog circuits with long chains of
cascaded neurons can accumulate and amplify noise, eventu-
ally burying signals under the noise floor. The optoelectronic
devices shown in Fig. 2 are mostly linear and noisy, resulting
in a signal-to-noise ratio (SNR) degradation, i.e. a noise factor
greater than unity (F > 1). Fundamentally,
this happens
because not only they linearly amplify noise as well as signals,
they generate noise on top of the output. To counter that effect,
we need a device that can amplify signals while decreasing
noise. This can be achieved with a nonlinear device -- in our
case, a modulator with a nonlinear transfer curve. The more
nonlinear the modulator is, the more it can compensate for
accumulating noise (Fig. 3).
C. Organization of the Paper
In this paper, we will numerically analyze the noise as it
propagates across a neural network composed of O/E/O neu-
rons (Sec. II). We will quantify the nonlinearity requirements
for a neural circuit to suppress noise accumulation. We propose
a simple experiment involving the cascadability property of the
neuron which verifies that noise is properly suppressed.
Following the noise analysis, we show two options of con-
structing the O/E/O circuit: one with a passive electronic link,
and the other with an active transimpedance amplifier which
provides energy gain in the electronic domain (Sec. III). The
electronic gain can enhance the neuron's response for weak
input signals. The effect of this enhancement is a reduction
of overall optical power levels in the circuit, leading to more
energy efficiency.
inputsweightssummationnonlinearityoutputT(λ)tttλ1 λ2λ3 λ1 λ2λ3 λ4ModulatorneuronPhoto-detectorPhotonic weight bankI(t)i(t)I(t)tbiaspump...AWGλ1CW Laser Bank......λNFOSPLITTER1NFOSPLITTER1NFI1NFI5/95WDM MonitorMRRModulatorPLBalancedPhotodiodeMRR Weight BankPinfrom previ-ous layerto nextlayerPoutIpoff-chipon-chipopticalelectricalTIA II. NOISE PROPAGATION IN O/E/O NEURONS
Neural networks are known to be robust to noise [13]. In
fact, noise can be exploited to train neural networks when other
optimization algorithms might fail [14], [15]. Noise originating
in hardware was used to implement on-line learning to a VLSI
neural network [16].
There are two methods for avoiding noise propagation
across a network. The first involves a collective approach,
using redundant neurons encoding correlated information. This
is called population coding in neuroscience, and is it re-
quired by physiological neural networks to overcome the noise
generated by individual neurons [17], [18]. In essence, the
√
estimation error for information carried by N neurons scales
N [18]. This concept has been adapted to machine
with 1/
learning and has been proven to mitigate noise in multilayer
perceptron networks [13], the category into which photonic
neural networks described in this paper falls. The second
method, the focus of this paper, relies on every individual
neuron to have noise suppressing circuit. This section studies
the noise accumulation mechanisms within a single neuron and
describes how a modulator's nonlinearity can suppress noise
accumulation.
A. Modulator Nonlinearity as Noise Trimmer
Consider non-return-to-zero (NRZ) modulated signals at the
input and output of each neuron. Assume that the neuron
is biased so that zeros and ones fall on each side of its
S-shaped transfer function (Fig. 3). Because the derivative
of this function is relatively small in these regions, noise
variance is reduced and the output looks "cleaner" than the
input. This operating principle can be generalized to other
modulation schemes and other transfer functions, but exploring
all theoretical possibilities is beyond the scope of this paper.
Fig. 3. Modulator's transfer function showing the noise trimming principle.
Vpp and T (V ) represent the peak-to-peak voltage and transfer function of
this modulator, respectively. T (V ) is assumed to be a symmetric S-shape, so
T (cid:48)(V0) = T (cid:48)(V1). Here, an NRZ signal with probability distribution shown
in the x-axis can be transduced into an optical signal with lower noise (y-axis).
The inset illustrates how to extract the distribution empirically.
B. Sources of Noise in the OEO Link
The accumulation of thermal noise, shot noise, amplifier
noise, and relative intensity noise (RIN) must be counteracted
3
Fig. 4. Simplified neuron circuit showing the sources of noise in each step
of the photonic link.
by the modulator nonlinearity in order to guarantee that the
SNR at the output equals the SNR at the input. This condition
leads to the following equation:
1 − T 2
n
SNR
=
4T 2
n
V 2
pp
(cid:18) 4kBT ∆f R2
(cid:124)
(cid:123)(cid:122)
(cid:19)
(cid:125)
TIA∆f I 2
(cid:123)(cid:122)
thermal noise
RL
TIA,n
TIA
(cid:125)
TIA,n=TIA noise
V 2
+ R2
(cid:124)
+ RIN2
(cid:124)
+ 2q∆f VppRTIA
+
(cid:124)
shot noise
(cid:18)
(cid:123)(cid:122)
(cid:123)(cid:122)
RIN
1 +
(cid:125)
(cid:19)
(cid:125)
1
MD2
(2)
where four sources of voltage noise (depicted in Fig. 4)
are balanced against a potentially very low term Tn. The
expression derivation, assumptions, and precise meaning of
each term are described in Appendix A.
Equation 2 shows that the SNR converges to a finite level
which cannot be arbitrarily large, since the term on the right
is always positive. That is expected since noise is generated
at every stage. To increase the neuron's SNR, we need to
decrease the terms on the right side of the equation, in partic-
n (noise transmission) and (RTIA/Vpp) (sensitivity). The
ular T 2
modulator's nonlinearity is very important in guaranteeing a
high SNR. A completely linear modulator (leading to a noise
transmission factor Tn → 1) results in a completely noisy
signal (SNR → 0). This happens because noise increases more
strongly than the signal at every stage, eventually reducing
signal integrity at the infinite cascadability limit. However, a
completely nonlinear modulator (Tn → 0) results in a high
quality signal (SNR → (2RIN)−2), which can be over 100 dB
(assuming RIN = −160 dB/Hz [19]).
C. Noise vs. Gain Tradeoff
Modulators in reality have intrinsic nonlinearities, but are
often operated in their linear region, often limiting their
extinction ratio. However, nonlinearities are not only necessary
by the mathematics of the neural network but are also here ex-
ploited to suppress noise. The microring resonator modulator,
in particular, is known to have a Lorentzian-shape transfer
function, which provides an ideal high-sensitivity S-shape
for this application. Another candidate is a Mach-Zehnder
modulator, with a sinusoidal transfer function. In either case,
we expect their noise multipliers to lie strictly between 0 and
1.
Assuming 0 < Tn < 1 and a fixed Vpp, then SNR can
be increased by reducing transimpedance RTIA as much as
T0ΔT VPPV0V1T1NRZ hist.VintimeTIAi shot PinPin,nIMODRINPLVbiasT(V)PL . T(V) +Pout,n Vi thermal possible, at the expense of greater power consumption (Eq. 1).
In other words, a higher quality signal requires a lower gain
circuit, resulting in a higher power consumption. This tradeoff
can be exploited to save energy in neural network applications
for which SNR is not a critical factor. Section III shows a
few realistic estimations for silicon photonic modulator neuron
implementations.
D. Autapse Test as Cascadability Standard
We use the notion of cascadability to verify whether a par-
ticular photonic neuron design can be scaled up to form large
networks. There are three kinds of cascadability: physical, gain
and noise.
a) Physical: A photonic neuron device is physically
cascadable if the nature of its output can be directly connected
to another's input. For example, the neuron introduced in Fig. 1
outputs a lightwave of a single wavelength, while receiving
a number of inputs in parallel at different wavelengths. The
signals are carried in the amplitude envelope (not phase, or
polarization) of both input and output lightwaves. Because of
that, this neuron can be interconnected and form arbitrarily-
large neural networks so long as the fan-in to each neuron is
equal or less than the number of WDM channels it is designed
to support.
b) Gain: Beyond being physically cascadable, each neu-
ron must be able to provide enough optical power to excite
the next layer of neurons, if needed. Equation 1 provides an
estimate of the gain cascadability condition for the worst case
scenario: one neuron, alone, delivering enough optical energy
to NFO other neurons. In this subnetwork, NFO neurons can
"replicate" the output of the initial neuron, which allow for
NFO subnetworks to process the input in parallel. In practical
deep neural networks, however, multiple neurons share the
burden of providing enough optical energy for the next layer.
This metric can only be quantified if the shape and weight
configuration of the neural network is known in advance, i.e.
in the presence of an application benchmark, which is out of
the scope of this article.
c) Noise: The other potential scalability limitation is
noise accumulation. This is particularly important for deep net-
works. In the worst case scenario, the information contained in
a signal fed to the first layer's input must survive uncorrupted
as it goes through the remaining layers of the network, even
in the presence of noise. The calculations leading to Eq. 2
show that at the limit of infinitely deep neural networks, the
SNR stabilizes to a certain value (solution of the equation) by
balancing noise generation by the electronic O/E/O link and
the noise suppression by the modulator's nonlinearity.
d) Autapse Test: A simple experiment can be constructed
to demonstrate and quantify all
three conditions: a self-
connection, also referred to as an autapse. With an autaptic
connection with unity weight, the neuron emulates an infinite
chain of neurons, where each connection delay τ represents
a virtual neuron. The autapse experiment
thus allows to
study infinite cascadability without producing an infinite chain
of cascaded neurons. In this experiment, an initial pulse
perturbation is sent to the neuron at t = 0, triggering an
4
output pulse in response. This output pulse travels through
the autapse and, provided the gain cascadability condition is
met, excites another perturbation at t = τ. The evolution
of the pulse amplitude and shape at
times t = nτ will
determine whether this neuron has met both gain and noise
cascadability conditions. This experiment emulates an infinite
series of neurons connected on a one-to-one basis. However, it
can also emulate a one-to-N connectivity pattern if the autapse
weight is set to 1/N, which represents a 1/N loss in optical
power between consecutive layers. Many-to-one and many-to-
many connectivity can be extrapolated from this test but not
directly emulated. We also note that this experiment tests for
indefinite cascadability, which might not be required in small
neural networks.
Autapse experiments as described here were conducted
in both modulator-based [10, Sec. E] and laser-based [23]
photonic neurons, but they focused on demonstrating physical
and gain cascadability. Observing noise accumulation in an
autapse is the next logical step in testing these devices.
III. TRANSIMPEDANCE AMPLIFIER IN SILICON
PHOTONICS
Noise suppression relies on using the modulator's intrinsic
nonlinear transfer curve. But the nonlinearity is only observed
if the voltage swing to the modulator is large enough (Fig. 3).
The suite of standard silicon photonic components today are
based on Ge photodetectors and p-n junction index modulators
which possess a capacitance on the order of tens of femto-
farads and require a voltage swing of a few volts. This voltage
swing is provided by a transimpedance amplifier (TIA), which
transduces photocurrent into voltage swing. Equation 1 already
showed us that
increasing the transimpedance reduces the
optical pump power for the neuron. But increasing it too much
limits the bandwidth of the circuit (inversely proportional to
RC). In Sec. III-A we explore how the modulator's parameters
affect this power-bandwidth tradeoff. Sec. III-B discusses how
active TIA can be used to mitigate some of that tradeoff. We
show that an active TIA is no longer necessary to maintain
10 GHz bandwidth for sub-femtofarad nanophotonic devices.
A. Passive Transimpedance
An easy way to control the transimpedance in passive silicon
photonic chips is to use a simple resistor in parallel with the
modulator, whose value determines the transimpedance gain.
This design is simple and works well, but the photodetector
and modulator's junction capacitance add in parallel with
the transimpedance value. The achievable bandwidth is deter-
mined by the dominant pole of the circuit (∆f = 1/2πRTIAC)
(see Fig. 5). As a result, this limits how large the tran-
simpedance can be, since the capacitances add up to 50 fF,
which for a 10 GHz bandwidth corresponds to a maximum of
RTIA = 320 Ω. This is significant because large networks will
require on the order of 100 parallel wavelengths in a single
waveguide. Assuming a maximum safe power of 100 mW per
waveguide (avoiding nonlinear effects [24]), that gives us a
maximum of 1 mW per wavelength, generating only ∼0.3 V of
swing at the modulator, far from the typical Vπ∼2 V required
5
Cable Lumped Model
Modulator Model
R1,2
C
L
100 Ω
22 pF
220 nH
Balanced PD Model
Rdiode
Cdiode
Rparasitic
17 kΩ
20 fF
57 Ω
C1
C2
R2
34.7 fF
14.7 fF
19.3 kΩ
Inter Stage (TIA)
RTIA
Cbypass
Lpeak
200 Ω
1 nF
5.1 nH
Fig. 5. Circuit schematic of a silicon photonic neuron with a passive transimpedance circuit whose bandwidth is enhanced via inductive gain peaking [20].
The circuit parameters are typical of recent literature and have been experimentally verified [20] -- [22].
in silicon modulators [25]. Fig. 5 shows an implementation
compatible with standard silicon photonic foundry chips.
In this case, gain cascadability can only be achieved if (a)
one uses a modulator with Vπ∼0.3 V or (b) with a smaller
capacitance, or (c) one uses an active TIA component which
can decouple the transimpedance from the modulator's capac-
itance, allowing for a higher gain with the same bandwidth.
The solution involving improving modulators is promising,
as we are far from the fundamental limits of photonics [26].
Efficient modulators are key to cope with increasing demand
in data communications, so research in this direction abounds.
Exotic materials such as graphene are being used to reduce
the switching energy of nanophotonic modulators toward sub-
fJ [27]. Photonic crystals also offer an avenue for ultracompact
O/E/O conversion. For example, Nozaki et al. [28] have
demonstrated a nanophotonic (InP-based) O/E/O link with
1.6 fF capacitance and 25 kΩ transimpedance, with a voltage
swing of 0.5 V. 40 µW was sufficient to operate this O/E/O
device.
B. Active Transimpedance Amplifier
Fig. 6. Concept diagram of a silicon photonic integrated circuit packaged
with a CMOS-based TIA. A flip-chip bonded alternative would yield similar
electrical performance for the bandwidth of interest (10GHz).
Another way to provide gain without compromising band-
width is to use an active TIA circuit [29], [30] instead of a
RLC circuit. The TIA is designed to enhance the voltage swing
of the modulator when the photocurrent is limited. In the short
term, photonic integrated circuits can be coupled with CMOS-
based TIAs via wirebonds or flip-chip bonding (Fig. 6). In the
long run, however, these may be homogeneously integrated on
the same chip, via a zero-change platforms [31].
Figure 7 shows the bandwidth performance simulation of
the O/E/O circuit in Fig. 5 using two active TIA designs (one
Fig. 7. Transimpedance gain characteristics of the O/E/O module assuming
an AC current source at one of the photodetectors, while measuring AC
voltage amplitude across the modulator. The passive transimpedance circuit
parameters were introduced in Fig. 5. The commercial off-the-shelf TIA
ONET8531T (Texas Instruments) replaces RTIA and Lpeak. The ideal TIA
design is similar to that of ref. [30] but was optimized specifically for this
bandwidth and gain range. Note: The f3dB values are: 14.8 GHz (passive),
8.0 GHz (commercial), 21.9 GHz (ideal). The fpeak value for the passive
circuit is 11.8 GHz.
commercial and one ideal), and how they compare against
the passive transimpedance approach. They were all designed
to a bandwidth greater than ∼10 GHz. As expected, the use
of active TIA allowed us to achieve a ∼17 times higher gain-
bandwidth product (21.9 GHz×2.4 kΩ vs. 14.8 GHz×0.2 kΩ).
We note that TIAs fabricated with modern CMOS nodes
have their maximum output swing voltage limited by the
maximum VDD of the transistor gates, which,
is
limited by the breakdown voltage of the node. For example,
the 0.18 µm-CMOS node offer TIAs with 2.5 VDD [30], thus
limiting the maximum achievable Vpp to about 1.8 V. Since
the modulation depth (MD) is proportional to Vpp, this limit
does not impact the gain cascadability condition (Eq. 1). But
if Vpp ≤ 1.8 V < Vπ, the modulator will operate in a more
linear regime, which will then transmit more noise (higher
transmission factor Tn),
impacting the noise cascadability
condition (Eq. 2). As a result, neuromorphic photonics would
benefit from photodiodes and modulators with driving voltages
lower than the VDD of modern technology nodes.
in turn,
We study the effect of the ideal active TIA in the cascad-
ability condition of the O/E/O neuron. Equations 1 and 2 relate
ModulatorBypass CapBypass CapOn-Chip InductorCableCable+VDCVbiasCableCableC1C2R2R_ParasiticR_DiodeC_DiodeInter Stage R_TIACoplanar WaveguideCoplanar Waveguide-VDC+_+_+_R_DiodeC_DiodeR_ParasiticBalanced PhotodetectorCoplanar WaveguideCoplanar WaveguideCable Lumped modelL_CableC_CableR2_CableR1_CableVDC+TIACMOS Based TIABalanced PDModulatorVDDGndVDC-GndVbiasGnd+-1091010frequency(Hz)10−1100NormalizedOutputVoltageTransimpedanceGain(dV/dI)0.2kΩ/0.27V(passive)2.4kΩ/0.35V(ideal)2.4kΩ/0.35V(commercial)-3dB COMPUTED TRADEOFFS OF VARIOUS O/E/O DESIGNS.
TABLE I
Modulator
Class
p-n junction
[10], [22]
graphene [27]
[32], [33, Fig.6]
TIA
passive
active
passive
active
Vpp
V
4.8
4.8
0.75
0.1
0.75
0.1
RTIA
kΩ
SNR (dB)
PL
dBm (Tn = 0.5)
0.4
2.8
2.0
1.2
2.8
2.8
26
17
18
11
17
8
64
54
49
41
41
24
Hz (active), 0 (passive);
This table was computed by using equations (1 -- 2) with the following
parameters: Tn = 0.5; kBT = 4.11 × 10−21 J; ∆f = 10 GHz;
√
q = 1.6 × 10−19 C; ITIA,n = 20 × 10−12 A/
RIN = 1 × 10−6 [19]; MD = 0.61 (p-n junction), 0.33 (graphene);
NFO = 10; T1/2 = 0.5; Rd = 1 A/W (p-n junction), 0.35 A/W
(graphene [32]); ηpp = 0.5; RL = RTIA (passive), RL = ∞ (active). RTIA
values were chosen to fulfill 2πRC = ∆f−1, with capacitances listed in
the references.
the autapse SNR to multiple design parameters, such as Vpp,
PL, RTIA and bandwidth ∆f. There are two main trends to
bear in mind. With a passive transimpedance gain, there is
a clear tradeoff between bandwidth and power consumption
for a fixed modulator design. The larger the bandwidth ∆f,
the lower the load RTIA has to be, and therefore the larger
the optical power PL to guarantee gain cascadability (eq. 1).
The autapse SNR is constant over values of ∆f, since noise
terms are proportional to RTIA · ∆f terms. On the other
hand, an active TIA may provide a higher transimpedance-
bandwidth product (RTIA · ∆f), but introduces more noise
and consumes more power than a passive transimpedance.
However, a higher gain allows laser pump power to be lower.
Since off-chip lasers tend to be inefficient, this can result
in overall system-wide power savings. The TIA also allows
for dynamic transimpedance tuning, which is useful to reduce
power consumption at times when noise is not critical to the
application.
The dependence between SNR and individual parameters
shows very simple trends -- linear or quadratic. Therefore,
instead of displaying arrays of plots, we chose to use Table I to
show a few specific examples tied to the devices reported in the
literature. The table shows the effect of adding an active TIA
to a p-n junction standard-platform modulator vs. a graphene-
based ultra-sensitive modulator with Vπ of 0.1 V or 0.75 V.
A standard-platform silicon photonic neuron can benefit much
more from the TIA; the laser power requirement decreases
by 10-fold (from 26 dBm to 17 dBm), with the expense of a
decrease in the SNR limit (from 64 dB to 54 dB). In contrast,
almost no enhancement is seen by the graphene-based neuron,
because it features low capacitance and therefore the passive
RTIA can be high without impacting bandwidth. With that
said, TIAs will continue to be useful for applications requiring
marrying conventional modulators, and low-power and lower
bandwidth budgets, two factors that scale favorably to justify
their use.
IV. CONCLUSION
We presented a quantitative overview of the interplay be-
tween electrical gain, optical gain, and modulator sensitivity
6
in O/E/O photonic neurons. Although these metrics have
positive influences over the accuracy and stability of the
neural network, they can impose difficult power and noise
requirements.
We showed that current p-n junction-based silicon photonic
platforms do not support highly-interconnected photonic neu-
ral networks unless they use (a) more sensitive modulators,
(b) active transimpedance amplifiers (TIAs), or (c) operate at
a sub-GHz bandwidth.
This occurs because modulators need a large voltage swing
to reach the nonlinear threshold in their nonlinear transfer
function, which suppresses noise directly between one neural
layer and the next -- a requirement for cascadable analog
links. This swing can either be achieved by increasing optical
pump power at the modulator or by providing electric tran-
simpedance gain. However, optical gain is limited by optical
nonlinearities in waveguides (and potentially power budgets),
whereas transimpedance gain is inversely proportional to the
bandwidth of the circuit.
The analysis shown here also applies to other kinds of
electro-optic transducers, such as lasers. In particular, excitable
lasers [34] have a zero-or-one thresholding response that
serves as both an amplifier and a noise suppression device.
Cascadability properties have recently been demonstrated in
one such devices [23]. Future work will involve testing and
quantifying noise cascadability by means of an autapse test
experiment, setting another useful benchmark in the field of
analog neuromorphic photonics.
APPENDIX A
DERIVATION OF CASCADABILITY EQUATIONS
In this section, we will derive the expressions contained in
Eqs. 1 and 2. We will be using the following notations:
x(t) ≡ E [x(t)] ;
(cid:104)x(t)(cid:105) ≡
x(t) dt.
0
We start by computing the expressions of the photocurrent in
Fig. 4. The photodetector is often modelled as a linear, added
with shot noise and thermal noise terms following normal
distributions. This approximation is valid if the optical power
amplitude is much slower than the photodiode's response time.
(cid:90) T
I = Rd · (Pin + Pin,n) + Ishot + Ithermal
where
Pin,n ∼ N (0, P 2
n)
Ishot ∼ N (0, 2qe∆f I)
∼ N (0, 2qe∆f RdPin)
(cid:18)
(cid:19)
Ithermal ∼ N
0,
4kBT
RL
∆f
(3)
(4)
(5)
(6)
are random variables sampled in time. Note: Ishot should
technically not be a normal random variable, but in many
textbooks it is modelled that way. It is a good approximation
for large values of I/∆f qe (average event count).
Similarly, the transimpedance amplifier is modelled as a
linear component with input referred noise.
V = RTIA · I + Vbias + VTIA,n
VTIA,n ∼ N (0, R2
TIAI 2
TIA,n∆f ).
(7)
(8)
√
The ITIA,n is often presented in the literature in units of
Hz and is assumed invariant for TIAs of the same design
pA/
on the same technology node. Vbias is assumed to be noise-free.
So far, it can be easily shown that V is a random variable
with Gaussian noise, resulting in
V = V + Vnoise
V = RTIARdPin + Vbias
Vnoise ∼ N (0, V 2
n ).
(9)
(10)
(11)
So far all noise terms have been additive. The next expres-
sion for the power envelope of the output optical signal Pout.
Pout = PL(1 + nRIN) · T (V )
nRIN ∼ N (0, RIN2)
(12)
(13)
where T (V ) is the transfer function of the modulator (Fig. 3)
and RIN is the relative intensity noise -- r.m.s. integrated
from 0 to ∆f,(cid:82) RIN(dBc/Hz)df -- at the pump. There are
two nonlinear transformations in this expression that render
analytical treatment difficult: RIN is a multiplicative noise
term, and T (·) is a nonlinear function. That means that the
probability distribution function of the output noise does not
have an analytical form. but we can proceed to compute the
signal-to-noise ratios at the input and output.
(cid:68)
P 2
in
(cid:69) −(cid:10)Pin
(cid:11)2
P 2
n
P 2
in
(cid:68)
(cid:69) −(cid:10)Pin
(cid:11)2
(cid:28)(cid:0)Pin − Pin
(cid:1)2(cid:29) =
(cid:69) −(cid:10)Pout
(cid:68)
(cid:11)2
(cid:28)(cid:0)Pout − Pout
(cid:1)2(cid:29)
P 2
out
SNRin =
SNRout =
F =
SNRin
SNRout
= 1
7
(cid:19)
(cid:18)
We also make a few assumptions about how the modulator
is biased, as hinted in Fig. 3. First, we assume that Vbias is
picked so that
RTIARd · P1 + P0
T
+ Vbias
=
2
≡ T (V1/2).
1
2
Second, we denote T0,1 = T (V0,1) = T (V (P0,1)), ∆T ≡
T1 − T0, MD ≡ ∆T /(T1 + T0), ∆P ≡ P1 − P0, and Vpp =
RTIARd∆P .
Third, we assume that T (V ) is symmetric around V1/2 and
thus T (cid:48)(V1) = T (cid:48)(V0) ≡ T (cid:48) as hinted in Fig. 3. It is useful
to define the quantity we named the noise transmission Tn as
(cf. Fig. 3):
Tn ≡ T (cid:48) · Vpp
∆T
.
(18)
Here, we are able to define the gain cascadability condition.
We simply state that the output optical amplitude generated by
an input amplitude ∆P should be able to provide at least an
amplitude equivalent to ∆P to each of the fan-out neurons
(NFO). In other words:
PL∆T ηpp
NFO
> ∆P,
(19)
where ηpp is efficiency term from point to point, accounting for
insertion losses of the WDM networking grid. Eq. 19 should
lead directly to Eq. 1.
With these assumptions, Eq. 14 and the numerator of Eq. 15
become, respectively:
SNRin =
(cid:68)
P 2
out
(cid:69) −(cid:10)Pout
(cid:11)2
=
∆P 2
4P 2
n
L∆T 2
P 2
4
(20)
(21)
In order to complete the derivation, we must know how to
n in Eq. 12. As shown in
approximate the propagation of P 2
Fig. 3, if the variance of the input signal around their means
is small, we can linearize T (V ) such that this approximation
holds:
T (V ) ≈ T (V ) + T (cid:48)(V ) · Vnoise
(22)
Plugging Eq. 22 into Eq. 12 and observing that
nRIN · Vnoise = 0 because they are independent, then Eq. 15
should become equal to the desired Eq. 2.
REFERENCES
[1] P. A. Merolla, J. V. Arthur, R. Alvarez-Icaza, A. S. Cassidy, J. Sawada,
F. Akopyan, B. L. Jackson, N. Imam, C. Guo, Y. Nakamura, B. Brezzo,
I. Vo, S. K. Esser, R. Appuswamy, B. Taba, A. Amir, M. D. Flickner,
W. P. Risk, R. Manohar, and D. S. Modha, "A million spiking-neuron
integrated circuit with a scalable communication network and interface,"
Science, vol. 345, no. 6197, pp. 668 -- 673, 2014.
[2] S. B. Furber, F. Galluppi, S. Temple, and L. A. Plana, "The SpiNNaker
project," Proceedings of the IEEE, vol. 102, no. 5, pp. 652 -- 665, May
2014.
[3] B. Benjamin, P. Gao, E. McQuinn, S. Choudhary, A. Chandrasekaran,
J.-M. Bussat, R. Alvarez-Icaza, J. Arthur, P. Merolla, and K. Boahen,
"Neurogrid: A mixed-analog-digital multichip system for large-scale
neural simulations," Proceedings of the IEEE, vol. 102, no. 5, pp. 699 --
716, May 2014.
(14)
(15)
(16)
The noise cascadability condition is expressed in Eq. 16.
The idea is to compute Eq. 15 and then express the resulting
n term in terms of SNRin per Eq. 14.
P 2
Equation 16 is general; it should work for arbitrary ampli-
tude modulation schemes with known probability distribution.
From here onwards, we need to choose a convenient input
signal waveform that will allow us to compute an approximate
expression. We choose an unbiased non-return-to-zero (NRZ)
signal as modulation scheme, which is very common in optical
links and makes it easy to extract values from experimental
papers and plug them into these equations.
(cid:40)
0 ≤ t < T /2
P1
P0 T /2 ≤ t < T
Pin(t) = Pin,n +
(17)
8
"Low Vpp, ultralow-energy, compact, high-speed silicon electro-optic
modulator." Optics express, vol. 17, no. 25, pp. 22 484 -- 22 490, 2009.
[26] V. J. Sorger, R. Amin, J. B. Khurgin, Z. Ma, H. Dalir, and S. Khan,
"Scaling vectors of attoJoule per bit modulators," Journal of Optics,
vol. 20, no. 1, p. 014012, jan 2018.
[27] R. Amin, Z. Ma, R. Maiti, S. Khan, J. B. Khurgin, H. Dalir, and
V. J. Sorger, "Attojoule-efficient graphene optical modulators," Applied
Optics, vol. 57, no. 18, pp. D130 -- D140, 2018.
[28] K. Nozaki, S. Matsuo, T. Fujii, K. Takeda, E. Kuramochi, A. Shinya, and
M. Notomi, "Ultracompact O-E-O converter based on fF-capacitance
nanophotonic integration," in Conference on Lasers and Electro-Optics.
Optical Society of America, 2018, p. SF3A.3.
[29] J. Kim and J. F. Buckwalter, "A 40-Gb/s Optical Transceiver Front-End
in 45 nm SOI CMOS," IEEE Journal of Solid-State Circuits, vol. 47,
no. 3, pp. 615 -- 626, mar 2012.
[30] Jun-De Jin and Shawn Hsu, "A 75-dB · Ω 10-Gb/s Transimpedance
Amplifier in 0.18-µm CMOS Technology," IEEE Photonics Technology
Letters, vol. 20, no. 24, pp. 2177 -- 2179, dec 2008.
[31] V. Stojanovi´c, R. J. Ram, M. Popovi´c, S. Lin, S. Moazeni, M. Wade,
C. Sun, L. Alloatti, A. Atabaki, F. Pavanello, N. Mehta, and P. Bhargava,
"Monolithic silicon-photonic platforms in state-of-the-art CMOS SOI
processes [Invited]," Optics Express, vol. 26, no. 10, p. 13106, may
2018.
[32] I. Goykhman, U. Sassi, B. Desiatov, N. Mazurski, S. Milana, D. de Fazio,
A. Eiden, J. Khurgin, J. Shappir, U. Levy, and A. C. Ferrari, "On-Chip
Integrated, SiliconGraphene Plasmonic Schottky Photodetector with
High Responsivity and Avalanche Photogain," Nano Letters, vol. 16,
no. 5, pp. 3005 -- 3013, may 2016.
[33] K. Nozaki, S. Matsuo, A. Shinya, and M. Notomi, "Amplifier-Free Bias-
Free Receiver Based on Low-Capacitance Nanophotodetector," IEEE
Journal of Selected Topics in Quantum Electronics, vol. 24, no. 2, pp.
1 -- 11, mar 2018.
[34] P. R. Prucnal, B. J. Shastri, T. Ferreira de Lima, M. A. Nahmias,
and A. N. Tait, "Recent progress in semiconductor excitable lasers for
photonic spike processing," Advances in Optics and Photonics, vol. 8,
no. 2, p. 228, jun 2016.
[8] T. Ferreira de Lima, B. J. Shastri, A. N. Tait, M. A. Nahmias, and P. R.
Prucnal, "Progress in neuromorphic photonics," Nanophotonics, vol. 6,
no. 3, pp. 577 -- 599, jan 2017.
[9] T. Ferreira de Lima, H.-T. Peng, A. Tait, M. A. Nahmias, H. Miller,
B. J. Shastri, and P. Prucnal, "Machine Learning with Neuromorphic
Photonics," Journal of Lightwave Technology, p. 1, 2019.
[10] A. N. Tait, T. Ferreira de Lima, M. A. Nahmias, H. Miller, H.-T. Peng,
B. J. Shastri, and P. R. Prucnal, "A silicon photonic modulator neuron,"
arXiv preprint arXiv:1812.11898, pp. 8 -- 9, Dec. 2018.
[11] A. N. Tait, A. X. Wu, T. Ferreira de Lima, E. Zhou, B. J. Shastri, M. A.
Nahmias, and P. R. Prucnal, "Microring weight banks," IEEE J. Sel.
Top. Quantum Electron., vol. 22, no. 6, 2016.
[12] A. N. Tait, T. Ferreira de Lima, M. A. Nahmias, B. J. Shastri, and
P. R. Prucnal, "Multi-channel control for microring weight banks," Opt.
Express, vol. 24, no. 8, pp. 8895 -- 8906, Apr 2016.
[13] Y. L. Lee, "Input Noise Immunity of Multilayer Perceptrons," ETRI
Journal, vol. 16, no. 1, pp. 35 -- 43, apr 1994.
[4] K. Meier, "A mixed-signal universal neuromorphic computing system,"
in 2015 IEEE International Electron Devices Meeting (IEDM), 2015,
pp. 4.6.1 -- 4.6.4.
[5] M. Davies, N. Srinivasa, T. Lin, G. Chinya, Y. Cao, S. H. Choday,
G. Dimou, P. Joshi, N. Imam, S. Jain, Y. Liao, C. Lin, A. Lines, R. Liu,
D. Mathaikutty, S. McCoy, A. Paul, J. Tse, G. Venkataramanan, Y. Weng,
A. Wild, Y. Yang, and H. Wang, "Loihi: A neuromorphic manycore
processor with on-chip learning," IEEE Micro, vol. 38, no. 1, pp. 82 --
99, 2018.
[6] B. Marr, B. Degnan, P. Hasler, and D. Anderson, "Scaling energy per
operation via an asynchronous pipeline," Very Large Scale Integration
(VLSI) Systems, IEEE Transactions on, vol. 21, no. 1, pp. 147 -- 151, Jan
2013.
[7] J. Hasler and H. B. Marr, "Finding a roadmap to achieve large neuro-
morphic hardware systems," Front. Neurosci., vol. 7, no. 118, 2013.
[14] A. H. Abdelaziz, S. Watanabe, J. R. Hershey, E. Vincent, and D. Kolossa,
"Uncertainty propagation through deep neural networks," Interspeech,
sep 2015.
[15] Y. Bengio, N. L´eonard, and A. Courville, "Estimating or Propagating
Gradients Through Stochastic Neurons for Conditional Computation,"
pp. 1 -- 12, 2013.
[16] A. Jayakumar and J. Alspector, "A Cascadable Neural Network Chip
Set With On-chip Learning Using Noise And Gain Annealing," in
Proceedings of the IEEE Custom Integrated Circuits Conference.
IEEE,
1992, pp. 19.5.1 -- 19.5.4.
[17] H. P. Snippe and J. J. Koenderink, "Discrimination thresholds for
channel-coded systems," Biological Cybernetics, vol. 66, no. 6, pp. 543 --
551, apr 1992.
[18] H. S. Seung and H. Sompolinsky, "Simple models for reading neuronal
population codes." Proceedings of the National Academy of Sciences,
vol. 90, no. 22, pp. 10 749 -- 10 753, nov 1993.
[19] P. Resneau, M. Calligaro, M. Krakowski, H. Liu, M. Hopkinson,
A. Somers, A. Forchel, and J. P. Reithmaier, "High power and very
low noise operation at 1.3 and 1.5 µm with quantum dot and quan-
tum dash Fabry-Perot lasers for microwave links," in Advanced Free-
Space Optical Communication Techniques/Applications II and Photonic
Components/Architectures for Microwave Systems and Displays, L. J.
Sjoqvist, R. A. Wilson, and T. J. Merlet, Eds., vol. 6399, no. October
2006, sep 2006, p. 63990K.
[20] A. Novack, M. Gould, Y. Yang, Z. Xuan, M. Streshinsky, Y. Liu,
G. Capellini, A. E.-J. Lim, G.-Q. Lo, T. Baehr-Jones, and M. Hochberg,
"Germanium photodetector with 60 GHz bandwidth using inductive gain
peaking," Optics Express, vol. 21, no. 23, p. 28387, nov 2013.
[21] J. M. Lee and W. Y. Choi, "An equivalent circuit model for germanium
waveguide vertical photodetectors on Si," 2014 International Topical
Meeting on Microwave Photonics / the 9th Asia-Pacific Microwave
Photonics Conference, MWP/APMP 2014 - Proceedings, pp. 139 -- 141,
2014.
[22] Z. Xuan, Y. Ma, Y. Liu, R. Ding, Y. Li, N. Ophir, A. E.-J. Lim, G.-Q.
Lo, P. Magill, K. Bergman, T. Baehr-Jones, and M. Hochberg, "Silicon
microring modulator for 40 Gb/s NRZ-OOK metro networks in O-band,"
Optics Express, vol. 22, no. 23, p. 28284, nov 2014.
[23] H.-T. Peng, T. Ferreira de Lima, M. A. Nahmias, A. N. Tait, B. J.
Shastri, and P. R. Prucnal, "Autaptic circuits of integrated laser neurons,"
in Conference on Lasers and Electro-Optics, May 2019, pp. 1 -- 2.
[24] C. Koos, L. Jacome, C. Poulton, J. Leuthold, and W. Freude, "Nonlinear
silicon-on-insulator waveguides for all-optical signal processing," Optics
Express, vol. 15, no. 10, p. 5976, may 2007.
[25] P. Dong, S. Liao, D. Feng, H. Liang, D. Zheng, R. Shafiiha, C.-C. Kung,
W. Qian, G. Li, X. Zheng, A. V. Krishnamoorthy, and M. Asghari,
|
1704.07370 | 1 | 1704 | 2017-04-24T00:24:25 | Electrostatic properties and current transport of two-dimensional Schottky barrier diode | [
"physics.app-ph",
"cond-mat.mes-hall",
"cond-mat.mtrl-sci"
] | Recently demonstrated metal-semiconductor heterojunctions with few-atom thickness show their promise as 2D Schottky contacts for future integrated circuits and nanoelectronics. The theory for 3D Schottky contacts, however, fails on these low-dimensional systems. Here, we propose a new model that yields carrier distribution and potential profile across the 2D metal-semiconductor heterojunction under the equilibrium condition, based on the input from first-principle calculations. Our calculation also suggests that, at the same forward bias, the current density of a stack of 2D graphene-phosphorene Schottky diodes may be ten thousand times higher than that of a traditional 3D Schottky diode and offer less energy dissipation. | physics.app-ph | physics | Electrostatic properties and current
transport of two-dimensional Schottky
barrier diode
Fangbo Xu, Alex Kutana, Yang Yang, Boris I. Yakobson ()
Department of Materials Science & NanoEngineering, Rice University, Houston,
TX77005
() [email protected]
Abstract
Recently demonstrated metal-semiconductor heterojunctions with few-atom thickness
show their promise as 2D Schottky contacts for future integrated circuits and
nanoelectronics. The theory for 3D Schottky contacts, however, fails on these low-
dimensional systems. Here, we propose a new model that yields carrier distribution and
potential profile across the 2D metal-semiconductor heterojunction under the equilibrium
condition, based on the input from first-principle calculations. Our calculation suggests
that, at the same forward bias, the current density of a stack of 2D graphene-phosphorene
Schottky diodes may be ten thousand times higher than that of a traditional 3D Schottky
diode and offer less energy dissipation.
1
Schottky barrier diodes (SBD) have been indispensable in power-rectification and radio-
frequency applications1, 2, 3, 4, 5, 6, 7. As three-dimensional integrated circuits, which
vertically stack layers of electronic components, has been widely recognized in recent
years8, 9, 10, 2D materials hold great promise11, 12, 13. In particular, some new 2D
semiconductors, such as phosphorene and monolayer transition-metal dichalcogenides
(MoS2, WSe2, etc.), have moderate band gaps (1-2 eV) and high carrier mobilities14, 15, 16,
and may be used to form 2D Schottky contacts. Seamless lateral 2D heterojunctions of
graphene and 2D semiconductors have been demonstrated experimentally by employing
conventional photolithography17, 18.
However, the laws governing 3D SBDs fail qualitatively in low-dimensional cases. For
instance, it erroneously predicts non-trivial electric field outside the space charge region
(SCR)19. Also, first-principle calculations proved to be prohibitively expensive to study
energy band offsets 20, 21, 22. To this end, we derive a generic model of a 2D SBD and
evaluate its electrostatics and I-V curves (The system-defined carrier-capturing edge
states are neglected). Then we apply this model to characterize the novel features of a 2D
SBD formed by graphene and phosphorene.
In the 3D SBD model, the electric field becomes trivial outside a well-defined SCR.
When it becomes few-atom thick, however, the surface charges give rise to a field that
vanishes only at infinity. Imagine a heterojunction of two semi-infinite metal and
semiconductor sheets lying at
, if an electron is moved from
(metal) to
(semiconductor) along an electric field line, the work in this process makes up
the difference of Fermi levels. Since the path is an immense semi-circle normal to the
plane, we have
and hence the surface charge density
, resulting
in an indefinite total charge23. Accordingly, in this work, metal and semiconductor are
assumed to be strips with widths of 1 µm and 2 µm, respectively.
Previous work demonstrates that charged graphene behaves like classical conductor with
a constant potential profile24. By conformal transformation25, the charge density 𝜎𝑚(𝑥) of
metal strip induced by a coplanar semiconductor strip with
is evaluated as:
2
0zxx~1/zxxE~1/sxxsx
(1)
assuming the metal strip lies at
and
, while the semiconductor strip lies
between
, (
). Suppose
, due to electric neutrality, in metal there is
another positive charge with the equal amount, and its distribution, evaluated by
, ensures equipotential across the metal strip3. Also, the semiconductor
injects electrons which obey the same distribution, neutralizing the positive charge.
Therefore, Eq(1) gives the net charge on the metal strip.
Figure 1(a) shows the band diagram before and after metal and n-type semiconductor
reach equilibrium. The work function of semiconductor is assumed lower than metal. At
equilibrium, we have:
(2)
in which Φ is the electrostatic potential relative to the interface. For a given total
transferred charge,
, the semiconductor itself may achieve equilibrium as long as the
carrier density satisfies
(3)
where
. With an initial guess of
, one may obtain
, and then
by Eq(1), and further a new
as well as
. When
and
converge, we use Eq(2) to check if the junction achieves equilibrium, if negative,
needs to be adjusted until it equilibrates the heterojunction.
3
212222xsmxxxaxdxxxax0z,xaa12,xx1xa0sx1/222~ax0__*_bFmFsFmnnqVEEEqqqscQ1exp/nxnxqxkT10xxsxmxx_FmExnxscQ
Figure 1. (a) Band diagram of a heterojunction of metal and n-type semiconductor. (b) Carrier density across
the semiconductor in 3D and 2D junctions. The thickness of the silicon strip is assumed to be 0.5 nm. Inset:
zoom-in of 3D semiconductor near the interface, and the dashed line marks the boundary of the space charge
region. (c) The electrostatic potential profile across the semiconductor strip in a 2D W-Si junction. Inset: charge
distribution near the far end of the semiconductor strip.
To highlight the impact of merely lowering dimensionality, we work with an imaginary
0.5 nm thick SBD of tungsten and n-type silicon, while the bulk properties (e.g. Schottky
barrier, permittivity) are retained. Figure 1(b) shows the change of
in silicon at
cm-3 in contrast to 3D model, and the inset displays the zoom-in near the
3D interface. We find that in the 2D SBD the neutral region outside SCR no longer exists
and the whole semiconductor becomes charged. Figure 1(c) suggests that
exhibits
a zero-field plateau. Here we take the height of the plateau as the built-in potential barrier
. In the 3D model, we have
, which barely varies since
is almost
4
nx3180110DnxbiVbiBnVn invariant if
is high. In our 2D model, however,
significantly decreases as
increases. Deep insight reveals that, at a higher
, more electrons pour into the metal,
causing a larger
and hence a lower
. In addition, the inset of Figure 1(c)
displays that, at the distance far away from the interface (excluding the edge effect), the
charge distribution behaves approximately
, which is consistent with our argument
on the infinite 2D SBD.
5
0nbiV0n0n_FmEbiV~1/x Next we apply our model to a 2D SBD of graphene and phosphorene. Doping of
phosphorene has been explored both theoretically and experimentally recently26, 27, 28, 29 ,
but the doping mechanism is not yet well established. Therefore we employ the
conventional nearly-free electron model involving 2D anisotropy, see Supplementary
Information (S.I.). Necessary parameters regarding electrostatics are listed in Table 1.
Table 1 Parameters for electrostatic properties of phosphorene. εz = 5.27. The effective masses and
carrier mobilities are extracted from Ref. 12. The permittivities in each direction, independent of
carrier type, are fitted by the model of a parallel plate capacitor by varying the thickness of vacuum
slab.
𝒎∗/𝒎𝟎
µ
𝛆/𝜺𝟎
(103 cm2V-1s-1)
𝚽𝐁
(eV)
MFP
(nm)
armchair (e)
0.17
1.10-1.14
15.23
~1.50
~30
zigzag (e)
1.12
~0.08
11.62
1.0
~6
armchair (h)
0.15
0.64-0.70
zigzag (h)
6.35
10-26
-
-
~ 0.1
~17
0.5
~3300
6
Figure 2 The electrostatic properties of the 2D graphene-phosphorene heterojunction, with widths of
1 µm and 2 µm, respectively. The transport is along the zigzag direction of phosphorene. (a) The
atomic structure. (b) The density of states projected on individual P atoms located at the interface
(pink) and far from the interface (blue). EF_G marks the position of the Fermi level of the graphene. (c)
The carrier density across the semiconductor strip at various doping concentrations. (d) Potential
profile at various doping concentrations. Inset shows dependence of xd on the doping level. (e)
7
Potential profile and carrier density (inset) at a variety of semiconductor widths at 𝒏𝟎 = 𝟖 × 𝟏𝟎𝟏𝟎
cm-2, xd=1.07 µm.
The graphene-phosphorene (G-P) heterojunction is modeled as periodic arrays of
dislocations30. Figure 2(a) shows a possible structure of C(5,1)P(3,3) with 0.8% lattice
mismatch, accommodating the boundary perpendicular to the zigzag direction of
phosphorene. Meanwhile, C(2,1)P(2,0) is also created with 1.1% lattice mismatch for the
boundary perpendicular to the armchair direction (see Figure 1 of S.I.).
The Schottky barrier for n- and p-type SBD, namely ΦBn and ΦBp, are determined in the
following way. We perform density functional theory (DFT) calculation on graphene
alone to obtain its Fermi level relative to the lower bound of the projected density of
states (PDOS), and then locate it on the PDOS of P atoms at the interface, as illustrated
by Figure 2(b). Note that the DFT calculations were performed on intrinsic
semiconductor under 0 K, so the carrier transfer is negligible, and accordingly the band
bending is ascribed to the interfacial dipoles. In this paper we ignore the impact on ΦB
due to carrier transfer, since the local bending of bands due to the carrier transfer is
estimated less than 30 meV (< 1% ΦB) for all doping levels.
Figure 2(c) displays the carrier density in the n-doped phosphorene with a variety of
doping concentrations. Compared with the 2D W-Si junction, due to the higher ΦB,
has a less steep transition, a narrowed plateau and more significant edge effect. Similar
features are also found in the potential profile, as presented in Figure 2(d). We define the
distance from the interface where Ex < 10-5 V/nm as the width of SCR, xd, and have
𝑉𝑏𝑖 = Φ(𝑥1 + 𝑥𝑑) − Φ(𝑥1). Referring to the 3D model in which
, Figure
2(d) inset gives
. This faster decay agrees with the fact that increasing
lowers Vbi as explained in the W-Si example.
The semiconductor strip needs to be wider than SCR since otherwise Vbi becomes
unpredictable. Figure 2(e) shows the results for various widths of the phosphorene strip,
ws. Apparently Φ(x) becomes flat after it reaches Vbi if 𝑤𝑠 > 𝑥𝑑 , while otherwise the
potential barrier always exhibits lower than Vbi. One may also find that Vbi stays invariant
8
()nx.5330~DDdxn.77220~DDdxn0n irrespective of ws, and this indicates that Qsc, which determines ΔEF_m, also becomes
invariant. Thus the diverse ws gives distinct carrier distributions, as shown in the inset of
Figure 2(e).
Furthermore, the mechanism of equilibration turns out different as the number of carriers
varies. With a given n0, altering ws essentially modifies the number of carriers, which
determines upper limit of Vbi. When the amount of carriers is so adequate that the carrier
transfer hardly changes
, the potential Φ(x) accounts for the majority of ΔEF_s.
However, if carriers are too few, the increase of
becomes dominant in ΔEF_s, and the
semiconductor is almost drained under this circumstance.
9
nn Currents
As indicated in Table 1, in phosphorene the mean free path (MPF) of electrons is so
trivial that the classical model for current transport is still applicable. The formalism of
diffusion current31 is readily modified to adapt to our 2D model and gives:
(4)
where
is the potential profile in the presence of the bias
applied on the ends of
the semiconductor strip (see S.I.).
To develop the expression for the thermionic current of 2D SBD, we utilize the
anisotropic nearly-free electron model and eventually reach (see S.I. for derivation):
(5)
where
is the 2D Richardson constant. Note that the 3D thermionic
current holds the similar form but varies as
. Our calculation shows that the diffusion
current is much higher than the thermionic current in G-P SBD (see S.I.)
On the other hand, the holes possess an appreciable MPF (~3 µm) in the zigzag direction.
We estimate the tunneling current with the non-equilibrium Green's function approach
(NEGF) combined with DFT calculations32, along with the WKB approximation. The
NEGF+DFT approach may give
the
transmission coefficient,
, which
characterizes the intrinsic tunneling defined by the interfacial dipoles and coupling of
Bloch waves. The WKB approximation is used to estimate the hole tunneling probability
through the barrier across the SCR as
. Thus the
10
_2*SCR1expexpandifnqVkTJkTnxqxdxkT*xaV*3/2_expexp1BnanthmqqVJATkTkT**2222yqkkmA2~TME**22exp-xSCRmqEEdx total transmission coefficient is evaluated as
, and further the
tunneling current is obtained by Landauer formula33. Figure 3 shows the tunneling
current of holes in the zigzag direction, and
, from the intrinsic phosphorene to
the graphene (inset). Under an increasing forward bias but lower than
, the current is
found to increase exponentially. Higher doping concentration yields a shrunk
and a
lowered
, which jointly produce a raised
.
Compared with the n-type SBD, the p-type SBD yields higher current under the same
condition, implying a much lower turn-on voltage. For instance, at
cm-2 and
V,
µA/cm, higher than
by 6 magnitude orders. If the 2D
Schottky diodes are stacked up with a spacing of 2 nm, at 0.1 V the current density could
be 104 times higher than that of a traditional 3D W-Si junction (2.9 mA/cm2)34.
Additionally, the p-type SBD employs ballistic transport of carriers, promising low
energy dissipation. Note that the tunneling from metal to semiconductor is neglected, due
to the absence of states in the bandgap of semiconductor.
Figure 3 The quantum transport of holes in the zigzag direction of
phosphorene under forward bias. Inset shows the intrinsic
transmission coefficient of the graphene-phosphorene junction in
the absense of carriers at zero-bias regime.
In summary, we present a generic model for 2D SBD which shows that the charge
distribution and potential profile are subject to pronounced modification relative to the
11
TEMEEsmMEbiVdxbiV100510n0.1aV_6.2ptnlJ_ndifJ classical 3D model. Moreover, the graphene-phosphorene heterojunction is studied in
terms of electrostatics and current transport arising from diverse mechanisms. Specially,
due to the considerable MFP of holes, the ballistic current density is dramatically higher
than 3D model, and therefrom a promising quantum 2D SBD with low turn-on voltage
and less energy dissipation is revealed.
Methods
First-principle calculations have been carried out with density functional theory
implemented in VASP. In particular, the Perdew–Burke–Ernzerhof (PBE) exchange-
correlation functional is used to relax the atomistic structure with the maximum force less
than 0.01 eV/Å on each atom and the vacuum slab larger than 15 Å, while HSE06 hybrid
functional is employed for electronic structures, predicting a band gap of 1.58 eV for
phosphorene. The intrinsic transmission coefficients through the junction of graphene and
zigzag-oriented phosphorene were computed using non-equilibrium Green's function
formalism implemented in the TRANSIESTA code35, averaged on multiple k-points
perpendicular to the transmission direction in the zero-bias regime, including calculation
of self-energies describing the coupling of the scattering region with a semi-infinite
graphene lead and a semi-infinite phosphorene lead.
Acknowledgement
Reference
Calow JT, Deasley PJ, Owen SJT, Webb PW. A review of semiconductor
heterojunctions. J. Mater. Sci. 1967, 2(1): 88-96.
Saxena AN. Forward current-voltage characteristics of Schottky barriers on n-
type silicon. Surf. Sci. 1969, 13(1): 151-171.
1.
2.
12
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
Sze SM. Physics of Semiconductor Devices. John Wiley & Sons, 1981.
Fliegel FM. A Green's function analysis of Schottky barrier diodes in strip-
coupled transversal signal processing architectures. Ultrasonics Symposium,
1990. Proceedings., IEEE 1990; 1990 4-7 Dec 1990; 1990. p. 259-264.
Tung RT. Electron transport at metal-semiconductor interfaces: General theory.
Phys. Rev. B 1992, 45(23): 13509-13523.
Tung RT. Recent advances in Schottky barrier concepts. Mater. Sci. Eng. R-Rep.
2001, 35(1–3): 1-138.
Shuichi T. Schottky-Barrier Metal–Oxide–Semiconductor Field-Effect Transistors
as Resonant Tunneling Devices. Jpn. J. Appl. Phys. 2010, 49(10R): 104204.
Knickerbocker JU, Andry PS, Dang B, Horton RR, Interrante MJ, Patel CS, et al.
Three-dimensional silicon integration. IBM J. Res. Dev. 2008, 52(6): 553-569.
Emma PG, Kursun E. Is 3D chip technology the next growth engine for
performance improvement? IBM J. Res. Dev. 2008, 52(6): 541-552.
Topol AW, Tulipe DCL, Shi L, Frank DJ, Bernstein K, Steen SE, et al. Three-
dimensional integrated circuits. IBM J. Res. Dev. 2006, 50(4.5): 491-506.
Butler SZ, Hollen SM, Cao L, Cui Y, Gupta JA, Gutiérrez HR, et al. Progress,
Challenges, and Opportunities in Two-Dimensional Materials Beyond Graphene.
ACS Nano 2013, 7(4): 2898-2926.
Qiao J, Kong X, Hu Z, Yang F, Ji W. High-mobility transport anisotropy and
linear dichroism in few-layer black phosphorus. Nat. Commun. 2014, 5: 4475.
Liu H, Neal AT, Zhu Z, Luo Z, Xu X, Tománek D, et al. Phosphorene: An
Unexplored 2D Semiconductor with a High Hole Mobility. ACS Nano 2014, 8(4):
4033-4041.
Buscema M, Groenendijk DJ, Blanter SI, Steele GA, van der Zant HSJ,
Castellanos-Gomez A. Fast and Broadband Photoresponse of Few-Layer Black
Phosphorus Field-Effect Transistors. Nano Lett. 2014, 14(6): 3347-3352.
13
15.
16.
17.
18.
19.
20.
21.
22.
23.
Kim JS, Lee HS, Jeon PJ, Lee YT, Yoon W, Ju S-Y, et al. Multifunctional
Schottky-Diode Circuit Comprising Palladium/Molybdenum Disulfide Nanosheet.
Small 2014, 10: 4845-4850.
Kang J, Liu W, Sarkar D, Jena D, Banerjee K. Computational Study of Metal
Contacts to Monolayer Transition-Metal Dichalcogenide Semiconductors. Phys.
Rev. X 2014, 4(3): 031005.
Levendorf MP, Kim C-J, Brown L, Huang PY, Havener RW, Muller DA, et al.
Graphene and boron nitride lateral heterostructures for atomically thin circuitry.
Nature 2012, 488(7413): 627-632.
Liu Z, Ma L, Shi G, Zhou W, Gong Y, Lei S, et al. In-plane heterostructures of
graphene and hexagonal boron nitride with controlled domain sizes. Nature
Nanotech. 2013, 8(2): 119-124.
Hsu C, Wang Q, Tao X, Gu Y. Electrostatics and electrical transport in
semiconductor nanowire Schottky diodes. Appl. Phys. Lett. 2012, 101(18):
183103.
Tung RT. Formation of an electric dipole at metal-semiconductor interfaces. Phys.
Rev. B 2001, 64(20): 205310.
Gao Q, Guo J. Barrier height determination of silicide-silicon contact by hybrid
density functional simulation. Appl. Phys. Lett. 2011, 99(18): 183110.
Stengel M, Aguado-Puente P, Spaldin NA, Junquera J. Band alignment at
metal/ferroelectric interfaces: Insights and artifacts from first principles. Phys.
Rev. B 2011, 83(23): 235112.
Yu H, Kutana A, Yakobson BI. Carrier Delocalization in Two-Dimensional
Coplanar p–n Junctions of Graphene and Metal Dichalcogenides. Nano Lett. 2016,
16(8): 5032-5036.
24. Wang Z, Scharstein RW. Electrostatics of graphene: Charge distribution and
capacitance. Chem. Phys. Lett. 2010, 489(4–6): 229-236.
14
Pelizzari MA, Criswell DR. Differential photoelectric charging of nonconducting
surfaces in space. J. Geophys. Res. 1978, 83(A11): 5233-5244.
Liu Y, Xu F, Zhang Z, Penev ES, Yakobson BI. Two-Dimensional Mono-
Elemental Semiconductor with Electronically Inactive Defects: The Case of
Phosphorus. Nano Lett. 2014, 14: 6782-6786.
Kulish VV, Malyi OI, Persson C, Wu P. Adsorption of metal adatoms on single-
layer phosphorene. PCCP 2015: 992-1000.
Li L, Yu Y, Ye GJ, Ge Q, Ou X, Wu H, et al. Nature Nanotechnology. Nat Nano
2014, 9(5): 372-377.
Liu H, Du Y, Deng Y, Ye PD. Semiconducting black phosphorus: synthesis,
transport properties and electronic applications. Chem. Soc. Rev. 2015.
Yazyev OV, Louie SG. Electronic transport in polycrystalline graphene. Nature
Mater. 2010, 9(10): 806-809.
Zeghbroeck V. Principles of Semiconductor Devices and Heterojunctions.
Prentice Hall PTR, 2007.
Xu F, Sadrzadeh A, Xu Z, Yakobson BI. XTRANS: An electron transport
package for current distribution and magnetic field in helical nanostructures.
Comput. Mater. Sci. 2014, 83(0): 426-433.
Landauer R. Spatial Variation of Currents and Fields Due to Localized Scatterers
in Metallic Conduction. IBM J. Res. Dev. 1957, 1(3): 223-231.
Neamen D. Semiconductor Physics And Devices. McGraw-Hill, Inc., 2003.
Brandbyge M, Mozos J, Ordejón P, Taylor J, Stokbro K. Density-functional
method for nonequilibrium electron transport. Phys. Rev. B 2002, 65(16): 165401.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
15
|
1912.11556 | 1 | 1912 | 2019-12-24T22:22:32 | Effect of Functionalization on the Properties of Silsesquioxane; a Comparison to Silica | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | While similar in nature, the properties of silica and silsesquioxane are very different, but little is known about these differences. In this paper, functionalized silsesquioxane microparticles are synthesized by adapting the modified St\"ober method and post-functionalized with rhodamine-B. The as synthesized silsesquioxane particles are characterized by a variety of physical and chemical methods. The synthesized particles are amorphous and nonporous in nature and are less dense than silica. While silsesquioxane and silica have some similar physical properties from their siloxane core, the organic functional group of silsesquioxane and the one-half oxygen difference in its structure impact many other properties of these particles like their charging behavior in liquids. These differences not only allow for the ease surface modification as compared to that necessary to modify silica, but also the use in a variety of colloidal systems that due to pH or electrolyte concentrations may not be suitable for silica particles. Keywords: silsesquioxane, Stober method, density, morphology, zeta potential | physics.app-ph | physics | Effect of Functionalization on the Properties of Silsesquioxane; a Comparison to Silica
Marzieh Moradia, Bailey M. Woodsa, Hemali Rathnayakeb, Stuart J. Williamsc, Gerold A. Willinga,*
a Chemical Engineering, University of Louisville, Louisville, KY 40292, USA
b Department of Nanoscience, University of North Carolina at Greensboro, Greensboro, NC 27401, USA
c Mechanical Engineering, University of Louisville, Louisville, KY 40292, USA
* Corresponding author: E-mail address: [email protected]
While similar in nature, the properties of silica and silsesquioxane are very different, but little is known about these
differences. In this paper, functionalized silsesquioxane microparticles are synthesized by adapting the modified
Stöber method and post-functionalized with rhodamine-B. The as synthesized silsesquioxane particles are
characterized by a variety of physical and chemical methods. The synthesized particles are amorphous and nonporous
in nature and are less dense than silica. While silsesquioxane and silica have some similar physical properties from
their siloxane core, the organic functional group of silsesquioxane and the one-half oxygen difference in its structure
impact many other properties of these particles like their charging behavior in liquids. These differences not only
allow for the ease surface modification as compared to that necessary to modify silica, but also the use in a variety of
colloidal systems that due to pH or electrolyte concentrations may not be suitable for silica particles.
Keywords: silsesquioxane, Stöber method, density, morphology, zeta potential
1. Introduction
Colloidal particles and suspensions have been investigated widely due to their extensive applications from advanced
materials to drug delivery [1,2]. Widespread research on colloidal particles has taken place in a number of application
areas including coatings [3], assembly of ceramics [4], photonic materials [5], and pharmaceutical materials [6].
Different characteristics of colloidal suspensions and resulting products can be reached by tailoring both the
morphology and chemistry of the colloidal particles [7].
The most important properties of a colloidal system such as stability depend on the potential or electric charge of the
colloidal particles which is determined by dipolar molecules and adsorption of ions [8]. Interaction forces between
colloidal particles can be determined by the potential distribution created by the variation of these molecules and ions
between the particles themselves [1,2,9]. Interaction forces are responsible for the stability of particles against
coagulation and have an important role in determining properties such as the shelf life, stability, rheology, and the
overall behavior in several industrial processes (e.g. mixing and membrane filtration) [9]. Therefore, it is essential to
know the charge distribution of a colloidal system to understand the system chemistry and its impact on the interaction
between particles.
In colloidal systems, the surface of particles is surrounded by an electric double layer which is also known as the
electrical diffuse double layer [8]. This double layer is formed by the particle surface charge and its counter ions,
forming an ionic cloud surrounding the particle [9]. The electrical potential of the double layer is commonly referred
to as the zeta potential. In fact, zeta potential is the potential difference between the dispersion medium and the
stationary layer of fluid associated with the colloidal particle [10].
Silsesquioxanes have gained significant attention due to the fact that they can be easily synthesized and further
modified with organic or inorganic functionalities [11-17]. Silsesquioxane refers to structures with a ratio of 1.5 of
oxygen to silicon atoms resulting in the general formula of R-SiO1.5. In this case, the R can represent a hydrogen or
an organofunctional group [18,19]. Based on the chemistry of the silsesquioxanes with an organic-inorganic
composition, their properties are combination of ceramic like properties of silica and the soft nature of organic
materials [20]. Silsesquioxanes have potential applications as nanoscale fillers in polymer systems for use in
adhesives, coatings, composites, and dental fillings [7,21]. They have also been examined for semiconducting devices
[22], fuel cells [23], optical devices [24] and sensors [25]. But there has been limited effort aimed at chemical
modifications of silsesquioxane particles with reactive organic functional groups.
Recently, Rathnayake et al. [7] made novel benzyl chloride, and benzyl chloride-amine functionalized silsesquioxane
(BC-SSQ, BC-A-SSQ) particles by adopting a modified Stöber method. They also post-functionalized the synthesized
particles with a Rhodamine-B (RhB) fluorescent tag (BC-SSQ-RhB, BC-A-SSQ-RhB). Silsesquioxane with benzyl
chloride functionalization is useful for the tailoring and grafting of a wide variety of materials to the surface including
polymer ligands and other precursors. The post-functionalization capabilities of these particles are significantly
beneficial for numerous applications in colloid chemistry and nanoscience. Due to the nature of the reactive group
functionality which mimics strong dipole-dipole and hydrogen bonding interactions, a colloidal suspension of these
particles can assemble into monolayers and hollow colloidasomes on a variety of polymer-coated substrates to create
3D colloidal assemblies [7,26]. Considering the applications of silsesquioxanes in adhesives, coatings and
1
semiconductors, it is very important to determine a wide range of properties for these particles such as their charge,
density and porosity along with their morphology.
In this paper, we synthesized the BC-SSQ, BC-A-SSQ, BC-SSQ-RhB and BC-A-SSQ-RhB particles based on the
Rathnayake et al. method [7] and report our efforts to fully characterize these microparticles. Zeta potential
measurements of the particles were performed over a range of pH to determine the charging behavior of the particles
and to find their isoelectric point. The morphology and surface analysis of the particles was examined by a helium ion
microscope (HIM) and a scanning electron microscope (SEM). Average size of the particles was measured using a
particle size analyzer. Density was measured by a helium gas pycnometer and specific surface area of particles was
measured using a BET instrument.
2. Methods and Materials
2.1 Materials
Para-(chloromethyl)phenyltrimethoxy silane was purchased from Gelest Inc. Potassium carbonate (ACS reagent) was
purchased from Fischer Scientific. 3-aminopropyltriethoxy silane (3-APT), anhydrous ethanol (200 proof),
ammonium hydroxide (28%), and Rhodamine-B carboxylic acid were purchased from VWR international.
2.2. Particle Synthesis
For synthesizing BC-SSQ particles, 10 mL of anhydrous ethanol and 4 mL of 28% ammonium hydroxide were mixed
on a magnetic stirrer for 5 minutes. Then 1.8 mL of benzyl chloride trimethoxy silane was added to the reaction at a
rate of 0.08 mL/minute and was allowed to stir for 18 hours. Particles were separated by centrifuging at 3000 rpm for
20 minutes. Then, particles were washed multiple times with ethanol followed by distilled water to remove any
impurities. Finally, particles were dried under the hood for 48 hours to yield BC-SSQ particles.
For synthesizing BC-A-SSQ particles, 65 mL of anhydrous ethanol and 3.5 mL of 28% ammonium hydroxide were
mixed on a magnetic stirrer for 5 minutes. Then 2 mL of 3-APT silane was added to the flask dropwise followed
immediately by 1 mL of benzyl chloride trimethoxy silane at a rate of 0.08 mL/minute and was allowed to stir for 18
hours. Then the BC-A-SSQ particles were washed and dried the same way as BC-SSQ particles.
In order to functionalize the as synthesized particles with Rhodamine-B, 500 mg of particles were dispersed in 30 mL
of anhydrous ethanol and were mixed on a magnetic stirrer until the particles were dispersed completely. Then 79 mg
of potassium carbonate was added followed by 230 mg of Rhodamine-B and was allowed to stir for 18 hours. In order
to keep the reaction away from visible light, the reaction flask was covered with aluminum foil. The same procedure
was used to wash and dry the particles.
2.3. Structure, Morphology, and Size
In this work, the morphology of particles was investigated using a Zeiss Auriga Crossbeam FIB-FESEM scanning
electron microscope and a Zeiss Orion helium ion microscope. A TESCAN scanning electron microscope (SEM)
equipped with energy-dispersive X-ray spectroscopy (EDX) was used for surface analysis of the particles. X-ray
diffraction (XRD) measurements were carried out on a Bruker D8 Discover diffractometer. Patterns were recorded
over the range from 10 to 90° (2𝜃) in steps of 0.02° with a scan speed of 2s at each step. A Brookhaven 90 Plus-Zeta
particle size analyzer was used to measure the average sizes of particles and their size distributions.
2.4. Zeta Potential
The Brookhaven 90 Plus-Zeta particle size analyzer was used to measure the zeta potential of colloidal particles as
well. Zeta potential of particles was measured at 0.01 volume percent concentration of particles in solutions with pH
ranging from 3 to 11. Acidic and basic solutions were made by adding proper amount of nitric acid or potassium
hydroxide to DI water. Each sample was sonicated and well-suspended in the pH solution before zeta potential
measurements. These measurements also revealed the isoelectric point of the particles. A Nikon Eclipse Ti inverted
microscope was then used to observe any agglomerates that formed during settling.
2.5. Density
Particle density is required for sedimentation analysis, or calculations involving volumes or mass of particles. The
particle density of different SSQ particles was measured by a AccuPyc II 1340 Pycnometer which is a fully automatic
gas displacement pycnometer. Particles were first dried in the desiccator to obtain true sample mass and to avoid the
distorting effect of water vapor on the volume measurement. Mass was determined by precisely weighing particles on
an analytical balance. Knowing the mass of the particles, the density was measured by the pycnometer using a 1.0 cm3
cup.
2
2.6. Specific surface area
After drying the particles in a desiccator, particles were degassed using a SmartPrep degasser to remove any gas
trapped in the pores and on the other surfaces of the particles. To degas, particles were heated to 120C and degassed
for 2 hours. After degassing, a TriStar 3000 gas adsorption analyzer was used to measure the surface area and porosity
of the particles by measuring gas (N2) adsorption.
3. Results
3.1. Morphology
In the literature, silsesquioxanes have been reported to have a variety of different structures, including a random
structure, ladder structure, cage structures, and partial cage structure [13,27].
The X-ray diffraction patterns of the as synthesized BC-SSQ and BC-A-SSQ microparticles are shown in Fig. 1. The
XRD spectral traces show that a broad peak is present at 2θ = 10 -- 30∘ for both particles, verifying them to be of
amorphous structure [28,29].
Fig. 1. XRD patterns for amorphous (a) BC-SSQ and (b) BC-A-SSQ
SEM and HIM images of BC-SSQ and BC-A-SSQ particles are shown in Fig. 2 and Fig. 3. These images are taken
after 4 hours of reaction. As it is clearly seen in these pictures, both particles are mostly spherical with a smooth
surface which is in agreement with N. Neerudu's results [7].
3
Fig. 2. SEM images of BC-SSQ and BC-A-SSQ
4
Fig. 3. HIM images of BC-SSQ and BC-A-SSQ
In Table 1, the EDX descriptive statistics of the average amount of different elements in the surface layer of BC-SSQ
and BC-A-SSQ particles are presented. One may notice that the chlorine amount in the surface layer is much higher
in BC-SSQ particles compared to the BC-A-SSQ particles. This is expected as the amine functionalization will cover
and replace many of the exposed benzyl chloride groups on the particle surface.
Table 1. Descriptive statistics of the amount of different elements in the surface layer of BC-SSQ and BC-A-SSQ
particles.
Element
Atomic %
Ratios to Si
BC-SSQ BC-A-SSQ BC-SSQ BC-A-SSQ
27.77
26.94
1
40.52
43.33
31.71
17.83
1.46
1.14
0
11.90
0
1
1.61
0.66
0.44
Si
O
Cl
N
5
Particle size measurements by DLS method, as shown by volume in Fig. 4, showed BC-SSQ and BC-A-SSQ particles
with sizes in the range of 600-2700 nm and 1400-3600 nm, respectively. The median size was determined for BC-
SSQ particles as 1200 nm and for BC-A-SSQ as 2300 nm.
Fig. 4. Lognormal differential distributions of particle sizes weighted by volume for (a) BC-SSC and (b) BC-A-SSQ
particles
3.2. Zeta Potential
Fig. 5 shows zeta potential measurements as a function of pH for the BC-SSQ and BC-SSQ particles. It is noted that
both particles are positively charged at lower acidic pH while bifunctional BC-A-SSQ particles have a higher charge
at a lower pH. This confirms the presence of amino groups in the BC-A-SSQ particles surface in their protonated
form. At higher pH, that is towards the basic end of the scale, the value of the zeta potential changes from positive to
negative. The isoelectric point of the BC-SSQ and BC-A-SSQ were found to be near pH 4 and 7, respectively, while
the isoelectric point of silica has been reported to be between pH 1.7 and pH 2.5 [30,31]. This makes SSQ favorable
for conditions where highly acidic solutions cannot be used due to materials or safety concerns. It should also be noted
that RhB functionalization did not significantly affect the charge of particles [32]. This is extremely valuable for
imaging in confocal microscopy because one can easily fluorescently label these SSQ particles and study the structure
of agglomerates without changing their isoelectric point.
Fig. 5. Plot of effective zeta potential (ξ) of the BC-SSQ, BC-SSQ-RhB and BC-A-SSQ, BC-A-SSQ-RhB
microparticles as a function of pH
Fig. 6 shows microscope images of the BC-SSQ samples at pH 3, 4, and the isoelectric point. As it can be seen in this
figure, by decreasing the pH from isoelectric point to 3 which increases the electric charge of the particles, we observe
that the particles are more dispersed during settling. These images were analyzed using ImagaJ. For this purpose,
6
images were converted into 8-bit, and then an Iso-Data algorithm threshold was set to differentiate the particles from
the backdrop [33]. A binary close operation and a watershed separation was performed on the thresholded images .
The watershed function visually separated semi-agglomerated particles, and ImageJ output the particle count, total
blob area and average blob size which are listed in Table 2. As the pH increases, the number of aggregates, total blob
areda and average blob size increase accordingly. This is consistent with the lower stability and higher tendency of
the particles to agglomerate at their isoelectric point. Such behavior for silica particles would not be observed unless
the pH was less than 2. It should be noted that this is a result of the BC-SSQ particles taking on a strong positive
electric charge below pH values of 4, unlike silica particles that would not take on a positive charge unless the pH
value is less than 2.
pH 3
pH 4
IEP
Fig. 6. Images of BC-SSQ particles at pH 3, 4 and isoelectric point after 4 hrs using inverted microscopy (20x).
Table 2. ImageJ analysis data of BC-SSQ samples
pH Count Total blob area Average blob size
2
3
337
347
IEP
394
4921
6063
8967
13.635
17.473
22.759
3.3. Density
The measured particle density for BC-SSQ and BC-A-SSQ was 1.68±0.05 and 1.48±0.04 g/cm3 at 22oC, respectively.
This is lower than the density of silica which is reported to be 2.196 g/cm3 [34]. This result is not unexpected as the
structure of the silsesquioxane is highly amorphous and has a random cage structure which is more open than the
structure of amorphous silica.
3.4. Specific surface area
The BET surface area (SBET) of BC-SSQ and BC-A-SSQ particles with size of 1.2 µm and 2.3 µm was 3.83±0.07
m2/g and 4.72±0.04 m2/g, respectively. The average pore width of BC-SSQ and BC-A-SSQ particles was 7.97 nm
and 8.76 nm, respectively. Fig. 7 shows the physisorption and pore size distribution of the particles. It can be seen that
both BC-SSQ and BC-A-SSQ particles exhibit a typical type II isotherm [35]. Considering the size of the pores and
type II isotherms, both types of particles possess a nonporous structure [35,36]. The gradual curvature and less
distinctive Point B indicate a significant amount of overlap of monolayer coverage and the onset of multilayer
adsorption. The thickness of the adsorbed multilayer generally appears to increase without limit when P/P 0=1.
7
Fig. 7. N2 adsorption-desorption isotherm and pore size distribution of (a) BC-SSQ and (b) BC-A-SSQ
3.5. General Overall Structure
Based on the characterization results, all four types of functionalized silsesquioxane particles tested in this study were
spherical and had a nonporous, amorphous structure. It was also shown that the silsesquioxane particles were less
dense than silica. Compared to silica, the potential colloidal stability of silsesquioxane was different due to the
difference in the structure and surface charges. Benzyl chloride and amine functionalization of different silsesquioxane
particles significantly affected the charge of particles. The isoelectric points of all four types of silsesquioxane particles
were significantly higher than the isoelectric point of silica. These differences may make these particles more viable
in applications where an isoelectric point closer to a neutral pH or a lower density particle is necessary.
Silsesquioxane is most often treated the same as silica. But in this study, it was shown that although silsesquioxane
and silica have a number of similarities in their physical properties such as thermal stability and rigidity which result
from their siloxane core, the one-half oxygen difference in the silsesquioxane structure along with its' organic
functional groups significantly impact many other properties of these particles. The more open structure of
silsesquioxane makes it less dense when compared to silica. The difference in the number of oxygen molecules, as
well as the type of the functional groups in the structure of the silsesquioxane particles as compared to silica particles,
changes the overall charge of the particles in the solution and results in a significantly higher isoelectric point.
Silsesquioxane particles can be positively charged in higher range of pH while silica is almost always negatively
charged due to its low isoelectric point. Charging behavior of colloids impacts the stability of the suspension and is
very important for designing a particular colloidal suspension based on the desired properties of that suspension. The
post-functionalization capabilities of silsesquioxanes, based on both silane and halogen chemistry, are also beneficial
for numerous applications in colloid science and ultimately give them a wider range of application when compared to
silica.
4. Conclusion
Silsesquioxane microparticles with different functional groups were synthesized by the modified Stöber method and
post-functionalized with rhodamine-B. Both BC-SSQ and BC-A-SSQ particles were spherical with a smooth surface.
The XRD results showed that both particles had amorphous structure. The isoelectric point of the BC-SSQ particles
and BC-A-SSQ were near pH 4 and 7. Density of BC-SSQ and BC-A-SSQ was 1.68±0.05 and 1.48±0.04 g/cm3 at
22oC, respectively. Both BC-SSQ and BC-A-SSQ were nonporous with an average pore width of 7.97 nm and 8.76
nm and BET surface area of 3.83±0.07 m2/g and 4.72±0.04 m2/g, respectively. While these particles share some
similarities with silica particles, the elimination of the one-half oxygen atom and the inclusion of the organic ligands
into the overall structure changes the physical and chemical properties in significant ways. These property changes
will not only allow silsesquioxane particles to be used in a wider variety of applications and but will also provide a
greater means for tailoring system properties for desired behaviors and products.
Funding information
This work was financially supported by a grant from NASA EPSCoR (Grant No. NNX14AN28A).
8
from microsphere-nanoparticle mixtures. Langmuir 21
Data Availability
The raw data required to reproduce these findings are available and can be requested from the authors. The processed
data required to reproduce these findings are available and can be requested from the authors.
Conflicts of interest The authors declare that they have no conflict of interest.
References
1. Martinez CJ, Liu J, Rhodes SK, Luijten E, Weeks ER, Lewis JA (2005) Interparticle interactions and direct imaging
of colloidal phases assembled
(22):9978-9989.
doi:https://doi.org/10.1021/la050382s
2. Hong X, Willing GA (2009) Transition force measurement between two negligibly charged surfaces: a new
perspective on nanoparticle halos. Langmuir 25 (9):4929-4933. doi:10.1021/la804103g
3. Martinez CJ, Lewis JA (2002) Shape Evolution and Stress Development during Latex−Silica Film Formation.
Langmuir 18 (12):4689-4698. doi:10.1021/la0114833
4. Lewis JA (2000) Colloidal processing of ceramics. J Am Ceram Soc 83 (10):2341-2359. doi:10.1111/j.1151-
2916.2000.tb01560.x
5. Joannopoulos JD, Villeneuve PR, Fan S (1997) Photonic crystals: putting a new twist on light. Nature 386
(6621):143-149. doi:10.1038/386143a0
6. Muller RH (1991) Colloidal Carriers for Controlled Drug Delivery and Targeting: Modification, Characterization,
and In Vivo Distribution. CRC Press, Boca Raton, FL
7. Neerudu N, McNamara L, Hammer NI, Rathnayake H (2017) A Versatile Synthesis to Novel Binary Reactive
Groups Functionalized Silsesquioxane Microparticles. Sci Adv Today 3:25266
8. Hunter RJ (1988) Zeta potential in Colloid Science: Principles and Applications. Academic Press Inc., San Diego,
CA 92101
9. He Q (2014) Investigation of stabilization mechanisms for colloidal suspension using nanoparticles. Dissertation,
University of Louisville,
10. Tadros T (2013) Encyclopedia of Colloid and Interface Science. Springer, Berlin, Heidelberg. doi:10.1007/978-3-
642-20665-8
11. Provatas A, Matisons JG (1997) Silsesquioxanes: Synthesis and applications. Trends Polym Sci 5 (10):327-332
12. Feher FJ, Walzer JF (1991) Synthesis and Characterization of Vanadium-Containing Silsesquioxanes. Inorg Chem
30 (8):1689-1694. doi:10.1021/ic00008a005
13. Li GZ, Wang LC, Ni HL, Pittman CU (2001) Polyhedral oligomeric silsesquioxane (POSS) polymers and
copolymers: A review. Journal of Inorganic and Organometallic Polymers 11 (3):123-154. doi:Unsp 1053-
0495/01/0900-0123/0
Doi 10.1023/A:1015287910502
14. Eisenberg P, Erra-Balsells R, Ishikawa Y, Lucas JC, Mauri AN, Nonami H, Riccardi CC, Williams RJJ (2000)
Cagelike precursors of high-molar-mass silsesquioxanes formed by the hydrolytic condensation of trialkoxysilanes.
Macromolecules 33 (6):1940-1947. doi:DOI 10.1021/ma9912507
15. Gravel MC, Laine RM (1997) Synthesis and characterization of a new amino-functionalized silsesquioxane. Abstr
Pap Am Chem S 38 (2):155-156
16. Bronstein LM, Linton CN, Karlinsey R, Ashcraft E, Stein BD, Svergun DI, Kozin M, Khotina IA, Spontak RJ,
Werner-Zwanziger U, Zwanziger JW (2003) Controlled synthesis of novel metalated poly(aminohexyl)-
(aminopropyl)silsesquioxane colloids. Langmuir 19 (17):7071-7083. doi:10.1021/la034291b
17. Feher FJ, Budzichowski TA (1995) Silasesquioxanes as Ligands in Inorganic and Organometallic Chemistry.
Polyhedron 14 (22):3239-3253. doi:Doi 10.1016/0277-5387(95)85009-0
18. Mori H (2012) Design and Synthesis of Functional Silsesquioxane-Based Hybrids by Hydrolytic Condensation of
Bulky Triethoxysilanes. Int J Polym Sci 2012:17. doi:10.1155/2012/173624
19. Sulaiman S (2011) Synthesis and Characterization of Polyfunctional Polyhedral Silsesquioxane Cages.
Dissertation, University of Michigan,
20. Ro HW, Soles CL (2011) Silsesquioxanes in nanoscale patterning applications. Materials Today 14 (1-2):20-33.
doi:10.1016/s1369-7021(11)70019-0
21. Liu YZ, Wu XR, Sun Y, Xie WL (2018) POSS Dental Nanocomposite Resin: Synthesis, Shrinkage, Double Bond
Conversion, Hardness, and Resistance Properties. Polymers-Basel 10 (4). doi:ARTN 369
10.3390/polym10040369
22. Wang Y, Vaneski A, Yang HH, Gupta S, Hetsch F, Kershaw SV, Teoh WY, Li HR, Rogach AL (2013) Polyhedral
Oligomeric Silsesquioxane as a Ligand for CdSe Quantum Dots. J Phys Chem C 117 (4):1857-1862.
doi:10.1021/jp3112843
9
14
doi:Doi
for
alkaline
fuel
cells.
(18):1279-+.
J Power Sources
23. Elumalai V, Sangeetha D (2018) Anion exchange composite membrane based on octa quaternary ammonium
Polyhedral Oligomeric Silsesquioxane
375:412-420.
doi:10.1016/j.jpowsour.2017.06.053
24. Lee J, Cho HJ, Jung BJ, Cho NS, Shim HK (2004) Stabilized blue luminescent polyfluorenes: Introducing
polyhedral oligomeric silsesquioxane. Macromolecules 37 (23):8523-8529. doi:10.1021/ma0497759
25. Chanmungkalakul S, Ervithayasuporn V, Hanprasit S, Masik M, Prigyai N, Kiatkamjornwong S (2017)
Silsesquioxane cages as fluoride sensors. Chem Commun 53 (89):12108-12111. doi:10.1039/c7cc06647c
26. Smay JE, Gratson GM, Shepherd RF, Cesarano J, Lewis JA (2002) Directed colloidal assembly of 3D periodic
structures. Adv Mater
10.1002/1521-4095(20020916)14:18<1279::Aid-
Adma1279>3.0.Co;2-A
27. Baney RH, Itoh M, Sakakibara A, Suzuki T (1995) Silsesquioxanes. Chemical Reviews 95 (5):1409-1430. doi:DOI
10.1021/cr00037a012
28. Music S, Filipovic-Vincekovic N, Sekovanic L (2011) Precipitation of amorphous SiO2 particles and their
properties. Braz J Chem Eng 28 (1):89-94. doi:Doi 10.1590/S0104-66322011000100011
29. Nallathambi G, Ramachandran T, Rajendran V, Palanivelu R (2011) Effect of Silica Nanoparticles and BTCA on
Physical Properties of Cotton Fabrics. Materials Research 14 (4):552-559. doi:10.1590/S1516-14392011005000086
30. Parks GA (1965) The Isoelectric Points of Solid Oxides, Solid Hydroxides, and Aqueous Hydroxo Complex
Systems. Chemical Reviews 65 (2):177-198. doi:10.1021/cr60234a002
31. Tohver V, Chan A, Sakurada O, Lewis JA (2001) Nanoparticle engineering of complex fluid behavior. Langmuir
17 (26):8414-8421. doi:10.1021/la011252w
32. Kornprobst T, Plank J (2012) Photodegradation of Rhodamine B in Presence of CaO and NiO-CaO Catalysts. Int
J Photoenergy:6. doi:Artn 398230
10.1155/2012/398230
33. T.W. Ridler SC (1978) Picture Thresholding Using an Iterative Selection Method. IEEE Transactions on Systems,
Man, and Cybernetics 8 (8):630-632. doi:10.1109/tsmc.1978.4310039
34. Lide DR (2003-2004) CRC Handbook of Chemistry and Physics. 84th edn. CRC Press,
35. Thommes M, Kaneko K, Neimark AV, Olivier JP, Rodriguez-Reinoso F, Rouquerol J, Sing KSW (2015)
Physisorption of gases, with special reference to the evaluation of surface area and pore size distribution (IUPAC
Technical Report). Pure and Applied Chemistry 87 (9-10):1051-1069. doi:10.1515/pac-2014-1117
36. Lawrence M, Jiang Y (2017) Porosity, Pore Size Distribution, Micro-structure. In: Bio-aggregates Based Building
Materials, vol 23. RILEM State-of-the-Art Reports. pp 39-71. doi:10.1007/978-94-024-1031-0_2
10
|
1801.09523 | 2 | 1801 | 2018-04-03T14:21:06 | The solution to an unresolved problem: newly synthesised nanocollagen for the preservation of leather | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | A widespread problem in libraries is related to the preservation of book covers in leather that are often torn, powdery and abraded. The same problem is encountered in the conservation of leather goods. Until now a satisfactory solution to contrast the leather deterioration had not been found and the applied conservation methods offered only temporary solutions, without guaranteeing a real and durable effectiveness. At the Istituto centrale restauro e conservazione patrimonio archivistico e librario (Icrcpal) it was decided to research more durable results and to apply nanocollagen solutions to the leather. A new synthesis of nanocollagen was performed in collaboration with Tor Vergata University, and Fondazione INUIT and the newly synthesised nanocollagen was characterised by different spectroscopic and imaging techniques, then applied to laboratory samples and, at the end of the research, it was used in the restoration of the leather cover of a 18th book. All the measurements performed on the tested leathers did not show any colour change after nanocollagen application, an increase of all mechanical characteristics and, of paramount importance, an increase in the shrinkage temperature of the leather with a partial reconstitution of its lost elasticity and flexibility. | physics.app-ph | physics | The solution to an unresolved problem: newly synthesised nanocollagen for the
preservation of leather.
Marina Bicchieria*, Federica Valentinib,c, Francesca Pascalicchiod, Maria Luisa Riccardie,
Piero Colaizzif, Camilla Del Reg, Maurizio Talamoc
a. Istituto centrale restauro e conservazione patrimonio archivistico e librario, Chemistry Department, Via
Milano 76, 00184 Roma, Italy, [email protected]
b. Università Tor Vergata, Chemistry Department, Via della Ricerca Scientifica 1, 00133 Roma, Italy,
[email protected]
c. INUIT Università
Tor
Vergata,
Via
dell'Archiginnasio
snc,
00133 Roma,
Italy,
d. Istituto centrale restauro e conservazione patrimonio archivistico e librario, Technology Department, Via
[email protected]
Milano 76, 00184 Roma, Italy, [email protected]
e. Istituto centrale restauro e conservazione patrimonio archivistico e librario, Restoration Department, Via
f.
Milano 76, 00184 Roma, Italy, [email protected]
Istituto centrale restauro e conservazione patrimonio archivistico e librario, Biology Department, Via Milano
76, 00184 Roma, Italy, [email protected]
g. Istituto centrale restauro e conservazione patrimonio archivistico e librario, SAF, Via Milano 76, 00184
Roma, Italy, [email protected]
*Corresponding author: Marina Bicchieri
Abstract
A widespread problem in libraries is related to the preservation of book covers in leather that are
often torn, powdery and abraded. The same problem is encountered in the conservation of leather
goods. Until now a satisfactory solution to contrast the leather deterioration had not been found and
the applied conservation methods offered only temporary solutions, without guaranteeing a real and
durable effectiveness. At the Istituto centrale restauro e conservazione patrimonio archivistico e
librario (Icrcpal) it was decided to research more durable results and to apply nanocollagen solutions
to the leather. A new synthesis of nanocollagen was performed in collaboration with Università Tor
Vergata, and Fondazione INUIT and the newly synthesised nanocollagen was characterised by
different spectroscopic and imaging techniques, then applied to laboratory samples and, at the end
of the research, it was used in the restoration of the leather cover of a 18th book. All the
measurements performed on the tested leathers did not show any colour change after nanocollagen
application, an increase of all mechanical characteristics and, of paramount importance, an increase
in the shrinkage temperature of the leather with a partial reconstitution of its lost elasticity and
flexibility.
Keywords: Nanocollagen, Synthesis, Leather, Surface, Restoration, Mechanical tests
Article history:
Received 10 January 2018
Accepted 1st March 2018
https://doi.org/10.1016/j.culher.2018.03.002
1
the book
1. Introduction
The preservation of torn, powdery, lacunose, worn,
abraded, weak and friable book-covers in leather is
a real problem for library conservators. These kinds
of deterioration are linked to the ageing, the usage
and manipulation of the books, the interactions with
pollutants, but they are also connected with the
products used in the leather manufacture or in the
finishing treatments applied for special or decorative
purposes.
Moreover, the choice of the materials for the covers
was very often more related to their price than to their
durability and permanence. Sheep leather, a less
durable and stable material in respect to calf or goat
leathers, is one of the most widespread materials in
the history of bookbinding, especially from 17th
century, due to its lower market price.
When a cover book is severely deteriorated, a
normal manipulation of
is almost
impossible, without causing the detachment of small
fragments or "dust" of leather from the binding, and
a restoring treatment is needed.
At present no satisfactory and durable solutions to
contrast the leather deterioration have been found,
but it is possible to apply environmental control
strategies and preventive conservation practices [1].
If environmental control is always advisable, the
preventive conservation is rather applicable to
museum objects than to artifacts that should be
consulted and used, such as books.
Since
century
hydroxypropylcellulose [2] was frequently used for
the consolidation of leather. More recently a mixture
of waxes and acrylic resin (SC 6000) was proposed
and
or
in mixture with
hydroxypropylcellulose [3].
None of these products had the capability to
penetrate into the bulk of the leather that after the
treatment often showed color changes and
increased brittleness.
The challenge has always been the recovery of
flexibility and stability of the degraded leather,
without altering the appearance and the equilibrium
of the internal fats and moisture.
2. Aim
The Istituto centrale restauro e conservazione
patrimonio archivistico e librario (Icrcpal) decided, in
collaboration with
to
approach the problem in a different way by using
nanomaterials
the
conservation of leather. Nanomaterials can, in fact,
penetrate into the bulk of the treated material,
offering a deeper consolidation effect. The idea was
to
its same principal
component, the collagen, synthesised at nanoscale
dimension and, after a series of
laboratory
experiments, to apply it in a real restoration case
study.
A promising preliminary investigation was performed
in 2014 [4, 5], but not implemented. Recently a new
electrochemical synthesis was optimised and
the Fondazione
leather with
expressly
designed
INUIT,
treat
the
used
alone
70s
for
the
of
the
last
a
for
the
to determine
nanocollagen with
patented (Patent N. 102016000096336, 2016)
obtaining
enhanced
dispersibility in different working media and long-
term stability of the colloidal phase dispersion (over
1 year, at ambient temperature, without precipitation
of a solid phase).
To assess the final application procedures, the effect
on leather of the newly synthesised nanocollagen,
soluble both in isopropyl alcohol and in water, was
studied by optical measurements, chemical and
mechanical tests. Moreover different nanocollagen
tested on
concentrations and solvents were
laboratory samples,
the minimal
amount of product to be applied to the leather with
the best consolidation results.
After a preliminary study, an original book was
chosen for the real application.
3. Experimental
3.1. Materials
3.1.1. Reagents for the synthesis of Nanocollagen.
Bovine Collagen Type I (Sigma-Aldrich) was used as
molecular precursor
synthesis of
nanocollagen, performed in acetate buffer aqueous
solution 0.1 M, pH 4.7,
(Sigma-Aldrich).
Alumina/Al2O3 tracked etched template membranes
(Whatman® Anodisc Inorganic Membranes) were
used (membrane diameter 30 mm, pores diameter
200 nm, pores length 100 µm, pores density 1x1012
pores/cm2), as well as HNO3 and NaOH (Sigma-
Aldrich, analytical grade).
3.1.2. Laboratory leathers
In the preliminary tests, the vegetable tanned
calfskin leather (mean thickness 1.6 mm) of a 18th
century cover, contemporary with
the original
volume, was used. For further tests a calfskin grain
split leather (chrome tanned, coloured with soluble
thickness 0.66 mm) was
aniline dyes, mean
employed,
because
showed mechanical
characteristics similar to those of the original cover.
3.1.3. Original cover
The Estro Poetico Armonico by Benedetto Marcello
(Mus. 243, 18th century, Biblioteca Casanatense,
Roma) was a perfect case study, presenting all the
damages described in the introduction. It belongs to
a series of five books with the same binding (first
edition in folio, Venezia, 1724-26), thus allowing for
a comparison with other original specimens, after the
treatment. The cover of Mus. 243 (mean thickness
0.60 mm), a vegetable sheepskin leather, tanned
with hydrolysable tannins, after the manufacture was
mottled with an acidic solution.
3.1.4. Tanning detection
The tanning was detected by specific spot tests:
ferric chloride for vegetable tannins, rhodanine for
hydrolysable tannins, acid butanol for condensed
it
2
applied
was
applied
electrode
synthesis
the silver conductive
tannins, alizarin sulphonate for aluminium detection
[6, 7, 8].
3.2. Methods
3.2.1. Electrochemical synthesis of nanocollagen.
the electrochemical
ChronoAmperometry was
techniques
the
of
for
nanocollagen. The optimised parameters were
patented in 2016 (Patent N. 102016000096336,
2016) and are briefly discussed below.
The
tropocollagen precursor was used at
concentration 1 mM in 0.1 M acetate buffer solution
at pH 4.7. To assemble the Alumina Template
Working Electrodes (ATWEs), it was necessary to
make the Al2O3 membrane conductive. An Ag layer
(20 nm thickness) was then deposited by sputtering
for 2 min at 2 mA. During the electrochemical
synthesis, a constant and controlled working
-1.0 V/versus Ag/AgCl/Cl-
potential value of
reference
by
ChronoAmperometry with deposition time 3600 s
under N2 at flow rate of 0.3 cm3/min. During the
ChronoAmperometry deposition,
the electrolysis
solution was magnetically stirred, at ambient
temperature. After the electrochemical synthesis of
nanocollagen,
layer was
dissolved in concentrated HNO3 and the alumina
template membrane was removed with concentrated
NaOH solution unable to dissolve the nanocollagen
that was rinsed in water until neutrality.
3.2.2. TEM (Transmission electron Microscopy)
TEM Philips Electron Optics 301 was employed for
the morphological study of nanocollagen samples
prepared by coating Cu grids (φ=3mm), by deep
coating in 0.7 mg/ml of nanocollagen dispersion.
After immersion, the coated Cu grids were dried
under Wood lamp.
3.2.3. SEM (Scanning Electron Microscopy)
For
of
nanocollagen a FE-SEM/EDX, LEO 1550 equipped
with a sputter coater (Edwards Scan Coat K550X)
was used. A volume of 5 μL of the nanocollagen
dispersion was deposited on Si(111), allowing the
solvent to evaporate at room temperature, and then
fixed on aluminium stub with carbon tape. The
samples were then coated with Au layer (thickness
10-20 nm), deposited by sputtering for 2 min at 25
mA.
For the studies on the application of nanocollagen to
leather, the SEM analyses were performed with a
Carl-Zeiss EVO 50 instrument equipped with both a
detector for electron-backscattered diffraction (BSD)
and for Secondary Electron scanning in Variable
Pressure mode (VPSE). SEM observations were
performed at 20 kV accelerating voltage with a
tungsten filament. All the samples were mounted on
Al stubs and observed at different magnification
the morphological
characterisation
for
FT-IR
(Fourier
Transform-Infrared
ranging from 500X to 3000X, following a fixed
measurement grid, before and after the treatment
with nanocollagen, which was applied on the sample
directly in the SEM chamber, in order to repeat the
observations after the consolidation, exactly at the
same point observed before the treatment. It was
possible to perform the analyses also on some
original fragments spontaneously detached from the
binding and no more repositionable.
3.2.4.
Spectroscopy)
IR spectra
the nanocollagen structural
characterisation were recorded by a Perkin-Elmer
Spectrum One FT-IR spectrometer from KBr pellets
in N2 environment.
3.2.5. Colour coordinates
A Minolta Chroma Meter CR22 colorimeter was used
in the CIE L* a* b* space, averaging 3 measurement
for each analysed point. Delta E before and after the
treatments was calculated
in agreement with
ISO/CIE 11664-6:2014(E) [9].
3.2.6. Tear resistance
For the measurements a Buchel Van Der Korput
Tearing Tester (Elmendorf Type) was used. The
number of samples (h: 60 mm, w: 50 mm) varied as
a function of the amount of leather that could be
subjected to the tests as will be evidenced in the
Results section. Tear resistance was measured in
accordance with T 414 om-12 method [10] that was
used both to measure the resistance to the formation
of a tear (tear initiation) and the resistance to the
expansion of a tear (tear propagation).
It was decided to extend to leather a method
normally used for paper, because the standardised
methods for leather are conceived for commercial
products with higher mechanical characteristics in
respect to those of the samples chosen to simulate
the original cover.
3.2.7. Tensile force
For the measurements an Instron Tensile Tester
Model 1026 was used with: load cell 0.49 kN at full
scale, load range adjustable with a scale selector,
crosshead speed 100 mm/min. The dimensions of
the samples were: h 200 mm, w 15 mm; useful length
150 mm.
3.2.8. Bending resistance (Stiffness)
For the measurement a Lorentzen & Wettre, Type
10-1 Bending Tester was used.
The samples had dimensions 70 mm (h) and 38 mm
(w) and the measurements were carried out with a
bending angle φ 30° and bending length 25 mm. The
number of samples varied as a function of the
amount of leather that could be subjected to the tests
as will be evidenced in the Results section.
3
Fig. 1. Characterisation study of the new nanocollagen. (a): SEM micrograph of a typical bundle of
tropocollagen, molecular precursor for the synthesis of collagen nanotubes. (b): Single fibre from tropocollagen
bundles, before the electrochemical template synthesis route. (c): TEM of the new synthesised nanocollagen;
(d): FTIR spectrum of the new nanocollagen samples. Wavelength of the characteristic collagen peaks are
reported.
3.2.9. pH and pH difference
pH of the original cover was measured by a portable
Crison pH-25 equipped with a flat membrane glass
electrode 5027, after extraction of 0.25 g of leather in
5 ml water. pH difference was measured by diluting
10 times the solution used for pH measurements
(cold extraction standard test ISO 4045:2008 [11]).
3.2.10. Shrinkage temperature
Very small amount of fibres were removed from the
flesh side of the leather, wetted with distilled water for
10 min on a single concave microscope slide. The
fibres were then separated with a needle, covered
with water and a coverslip. The microscope slide was
placed on the hot table Mettler FP82 Hot Stage,
thermostatically controlled through a Mettler FP90
Central Processor, and heated at a rate of 2°C/min.
The samples were observed under a Leica DMLP
microscope in transmitted light, at 50X magnification
and video recorded by Leica MC 170 HD. Video
recording allowed for the accurate verification of the
temperature at which shrinkage events occurs. The
method is described in ISO 3380:2015 [12] but had
been modified in order to be applied in conservation
field, where only small amount of material can be
examined [13, 14]. In this work the modified method
was applied.
3.2.11. Microscopy
To determine the animal species, the analysis of the
hairs follicular patterns of the leathers was performed.
The different leathers were observed with a Leica
Macroscope M420, using cold incident light produced
by optical fibres.
4
the
4. Results and Discussion
The experimental work has been divided in different
steps that will be discussed in separate paragraphs.
4.1. Characterisation of nanocollagen
Fig. 1 shows the SEM and TEM morphological
characterisation of
tropocollagen precursor
before its modification into nanocollagen, and the
nanocollagen final product. Fig. 1a, shows a typical
bundle of tropocollagen fibres, one of which is
reported at higher magnification in Fig. 1b to evidence
its typical cylindrical shape. After the electrochemical
tracked etched membrane
template synthesis
approach, the cylindrical fibre exhibits nanometer
dimensions:
inner diameter 100 nm, external
diameter 260 nm (Fig. 1c, TEM micrograph). FTIR
spectrum (Fig. 1d) confirms that the chemical
structure of the tropocollagen precursor (Collagen
Type I) is preserved after the synthesis. In particular,
the band centred around 1635 cm-1, is the typical
absorption of Amide I [15, 16] due to the stretching
vibration of the peptide carbonyl group (-C(=O)-).
The spectral assignments are reported in Table S1,
ESI section.
4.2. Application on leather - preliminary investigation
To evaluate the effect of nanocollagen on the leather,
mechanical, optical and microscopic tests (tensile
force, stiffness, colour coordinates, SEM imaging)
were performed on a 18th century and no more usable
leather cover book, chosen to simulate the original
document. The cover was divided into 4 samples
treated with different
sets. Three sets were
nanocollagen solution (Table 1), one was
left
untreated, and used as control.
Table 1
Solutions used in the preliminary investigation on a
18th century leather cover
Solvent(s)
isopropyl alcohol/water
70/30 (V/V)
isopropyl alcohol/water
70/30 (V/V)
water
nanocollagen
concentration
0.7 mg/ml.
1 mg/ml
1 mg/ml
The different nanocollagen concentration(s) and
solvent(s) were chosen to determine both the best
solvent and the more suitable concentration to be
applied in the consolidation treatments. Due to the
reduced amount of leather it was only possible to
obtain 16 samples for the tensile force measurements
and 6 samples for stiffness. The results are reported
in Tables 2 and 3.
Concerning the tensile stress measurements (Table
2), the unevenness of the ancient cover, caused by
the widespread walkways across the surface, gave
randomly distributed values. For this reason, only the
recorded values are reported, without statistical
treatment, as well as for binding measurements. The
unique result that can be inferred is that the samples
treated with nanocollagen in hydroalcoholic solution
(0.7 mg / ml) presented a better homogeneity regards
the tensile stress.
Table 2
Tensile force and breaking time of samples
from a 18th century leather cover, before and
after treatment with nanocollagen
Non-treated samples
Sample Force (N)
Breaking time (s)
Samples treated with nanocollagen 0.7 mg/ml
in isopropyl alcohol/water
70/30 (V/V) solution
Sample Force (N)
Breaking time (s)
Samples treated with nanocollagen 1 mg/ml
in isopropyl alcohol/water
70/30 (V/V) solution
Sample Force (N)
Breaking time (s)
Samples treated with nanocollagen 1 mg/ml
in water solution
Sample Force (N)
Breaking time (s)
78.5
145.1
35.3
54.9
59.8
59.8
59.8
113.8
51.0
19.6
66.7
53.0
41.2
58.6
80.4
127.5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
12.2
14.5
4.8
6
8.3
8
7
11
6
2.5
8.2
7.1
4.2
6
8.2
13
The binding measurements (Table 3) did not show a
great variation in stiffness before and after all the
treatments. The slight increase noticed after the
nanoparticles application is ascribable to the addition
of nanocollagen fibres to the leather, which filled the
inter-fibres cavities. This result is particularly positive
because consolidation treatments should not cause
an excessive increase in the material rigidity that
could lead to the breaking of the leather under stress
and manipulation.
Table 4 contains the averaged values (4 samples for
each nanocollagen solution) of colour coordinate and
Delta E, before and after application of the different
solutions of nanocollagen. None of the treatments
induced noticeable colour variations, except a light
decrease of
for colorimetric
measures better results with lower optical impact
luminosity. Also
5
Sample
1
2
3
4
5
6
were obtained when the hydroalcoholic solution 0.7
mg/ml was used.
The positive effect of nanocollagen treatment with all
of the applied solutions was clearly highlighted by
SEM images, as shown in Fig. 2. After the application,
fibres appeared less tangled than before and the
leather surface seemed smoother.
Table 3
Stiffness of samples from a 18th century leather
cover, before and after treatment with nanocollagen
Nanocollagen 0.7 mg/ml
in isopropyl alcohol/water 70/30 (V/V)
Before treatment
After treatment
(mN)
877
797
1127
845
1197
917
(mN)
950
798
1187
939
1212
1216
Nanocollagen 1 mg/ml
in isopropyl alcohol/water 70/30 (V/V)
Sample Before treatment (mN) After treatment (mN)
Nanocollagen 1 mg/ml
in water
Sample Before treatment (mN) After treatment (mN)
Table 4
Colour coordinates of samples from a 18th century
leather cover, before and after treatment with
nanocollagen. Average on 4 samples for each
treatment
Nanocollagen 0.7 mg/ml
in isopropyl alcohol/water 70/30 (V/V)
Coordinate
L*
a*
b*
After
Before
treatment
treatment
28.32 ± 0.78 28.01 ± 0.79
14.31 ± 0.66 14.03 ± 0.75
12.66 ± 1.33 12.62 ± 0.41
Nanocollagen 1 mg/ml
Delta E CIE
2000
0.27
in isopropyl alcohol/water 70/30 (V/V)
Coordinate
L*
a*
b*
Coordinate
L*
a*
b*
After
Before
treatment
treatment
32.64 ± 4.05 30.19 ± 4.28
13.53 ± 1.08 14.32 ± .16
14.63 ± 2.93 14.56 ± 3.42
Nanocollagen 1 mg/ml
in water
After
Before
treatment
treatment
29.64 ± 1.31 29.30 ± 1.62
14.98 ± 0.62 15.38 ± 0.71
12.86 ± 0.84 13.09 ± 0.96
Delta E CIE
2000
1.92
Delta E CIE
2000
0.39
tests
to verify
the ability of
The SEM analyses showed the reconstructing effect
due to the nanoparticles application: the cover margin
that was disordered and spread apart prior to the
treatment (Fig. 2 bottom left) appeared rebuilt in its
integrity (Fig. 2 bottom right) after the treatment.
4.3. Application on leather - evaluation of the effects
of nanocollagen at the working concentration
The encouraging results obtained in the preliminary
investigation gave the possibility both to choose the
optimal concentration (0.7 mg/ml in hydroalcoholic
solution) and to perform a new series of tests on a
more homogeneous set of samples. In this second
phase of the research a modern industrial split leather
was used. This kind of leather was very thin, almost
lacking in the flesh layer and presented mechanical
characteristics very similar to those of the original
leather needing restoration. In this phase of the work,
tear resistance tests were added to the other
mechanical
the
nanocollagen solution to reinforce the leather, in
particular in regards to its capability to resist to tearing
stress, an important parameter for the manipulation
of real objects.
Table 5 contains the results of the mechanical and
optical tests. Only the averaged values are reported,
as well as the number of samples subjected to the
tests.
As can be seen in Table 5, there is a sensible
increase in stiffness after the treatment, indicating an
increment of
the material.
Fortunately, the recorded absolute values do not
correspond to a rigid and breakable leather and the
increase can be regarded as a positive effect for the
stability of the treated leather.
Very positive results were obtained from the tensile
stress measurements (Table 5). Apart the increase in
the force necessary to induce the breaking of the
sample, the most positive effect is the increment in
the elongation parameter
indicates an
augmented elasticity of the treated leather, which can
better resist to the mechanical stresses induced by
manipulation.
Concerning tear resistance measures, two different
kinds of tests were performed: tear propagation and
tear initiation. The first measure simulates a very
common problem: the existence of tears in a real
object that can prosecute when the object is
manipulated. The second kind of measure gives
information on the resistance of the margin of an
object, a cover in this case, both during the
manufacture/restoration and the usage.
The tear propagation was measured on single
sample and on two coupled samples.
The results (Table 5) show a very
important
increment of the tear resistance after the treatment
with nanocollagen, in particular regards the tear
initiation. This increase is very positive for the future
manipulation of the original restored artefact.
resistance of
that
the
6
Before treatment
2126 ± 53 (two coupled samples)
3567 ± 90 (single sample)
Before treatment
4081 ± 80 (single sample)
Coordinate
L+
a*
b*
Before treatment
28.38 ± 0.26
16.77 ± 0.25
22.76 ± 1.96
After treatment
2243 ± 87 (two coupled samples)
3782 ± 185 (single sample)
After treatment
6670 ± 92 (single sample)
Tear initiation (mN)
Colour Coordinates
After treatment
31.91 ± 0.37
15.08 ± 0.22
19.45 ± 0.47
Delta E CIE 2000
3.26
Stiffness: 48 samples. Tensile stress: 30 samples. Tear propagation: 24 samples. Tear initiation: 10 samples.
Colorimetric measurements: 24 samples.
7
Fig. 2. Ancient cover, 18th century. Left side: before treatment with aqueous nanocollagen 1 mg/ml. Right side:
after treatment. Top: flesh side of the leather; bottom: grain side. Magnification 500X.
Table 5
Mechanical and optical characteristics of the split-leather before and after the treatment with nanocollagen 0.7
mg/ml in isopropyl alcohol/water 70/30 (V/V)
Before treatment
7 ± 2
Before treatment
Breaking time
(s)
41 ± 2
Force
(N)
38.9 ± 6.2
Stiffness (mN)
Tensile stress
Elongation
(mm)
23 ± 2
Force
(N)
43.0 ± 5.1
Tear propagation (mN)
After treatment
19 ± 8
After treatment
Breaking time
(s)
48 ± 5
Elongation
(mm)
27 ± 2
Values of colour coordinates and Delta E (Table 5)
show that in the case of the split leather the final
colour difference is greater than those obtained on
the ancient 18th century cover, probably as a
consequence of the dye used for colouring the
leather: vegetable in the ancient cover and aniline for
the modern split leather. The main difference is
related
to
the
treatment.
SEM images of the split leather showed the same
improvement noticed for the ancient 18th century
cover.
increases after
luminosity
that
4.4. Application on leather - final work on the original
volume Mus. 243
All the very positive results, previously reported and
commented, allowed to apply the nanocollagen
solution (0.7 mg/ml in isopropyl alcohol/water 70/30
(V/V)) to the original cover, after performing SEM
imaging analysis on non-repositionable fragments. In
Fig 3 some images are reported and the arrows
highlight peculiar effects such as the relaxation of the
fibres (Fig 3 top) or the formation of bonds between
fibres (Fig. 3 middle) or
filling effect of
nanocollagen (Fig. 3 bottom).
the
Fig. 3. Mus. 243. Left side: before treatment with nanocollagen 0.7 mg/ml in isopropyl alcohol/water 70/30
(V/V). Right side: after treatment. Top: flesh side of the leather, magnification 1000X; middle: grain side 3000X,
bottom: grain side 3000X. The arrows highlight some peculiar features and the effect of nanocollagen.
8
Sample
non-
treated
treated
to mechanical and optical
appear as a shrinkage of the fibres, and the
temperature interval where the shrinkage takes place
is a measure of the physical stability of collagen.
The shrinkage activity of collagen can be divided into
five different intervals with different characteristics of
the activity and temperatures. The intervals are
conventionally [14] indicated as:
- A1 to B1: distinct shrinkage activity is observed in
individual fibres. TF (Tfirst) is the temperature at which
the first shrinkage activity is recorded; TB1 is the
initial temperature of subsequent B1 activity.
- B1 to C: shrinkage activity of one fibre (occasionally
more) immediately followed by shrinkage activity in
another fibre is observed in the temperature interval
from TB1 to TS, where TS (Tshrinkage) represents the
starting temperature of the main interval C.
- C to B2: is the main shrinkage interval, and the
temperature to which this activity begins, is indicated
as shrinkage temperature TS. In this interval almost
all fibres shrink simultaneously and continuously. The
ending temperature of the main shrinkage interval is
conventionally indicated as TE (Tend).
- B2 to A2: shrinkage activity of fibres, which did not
contract at lower temperature in the previously
described
do
shrink
simultaneously (TE to TA2 interval).
- A2 to the last event: the shrinkage is going to end,
but few fibres contract with a well distinct and not
simultaneous event. The activity stops at the end of
the
this
temperature at which the last shrinkage is observed.
The temperature range of the whole observed activity
is ΔTtotal = TL – TF, whereas the length ΔT= TE-TS
of the C phase corresponds to the shrinkage interval.
The hydrothermal stability of the original leather was
analysed before and after
the nanocollagen
treatment.
Three measures were performed on the original
leather before and after the treatment with the chosen
nanocollagen solution. Results are reported in Table
6; Fig. 4 shows the shrinkage activity of some fibres
not treated and treated with nanocollagen.
tests,
In addition
measurements of pH, pH difference and shrinkage
temperature were performed.
Mus. 243 had average pH 4.3. It is well known that in
the 3-4 pH range leather is unharmed by the acidity
[17], and the kind or amount of tannin used in
manufacturing has little effect on pH, which can also
be influenced by the interactions with external factors,
such as light, ionising radiations and pollutants. More
or less weak acids can therefore be formed into the
leather and their presence can be revealed by pH
difference measures, i.e. the difference between the
standardised pH measurement and the pH of the
same solution diluted ten times [11]. This difference
is ranging from 0 to 1, and a measure between 0.7
and 1 denotes the presence of a certain amount of
strong free acids, that can cause red rot degradation,
which is a severe chemical degradation of vegetable-
tanned leather due to the used tanning compounds
as well as to interactions with atmospheric pollutants
[1, 18, 19]. Red-rotted leather have powdery surface,
are quite weak and vulnerable to abrasion or tearing.
In the more advanced states, the red-rotted leathers
became quite red.
Even though the cover of Mus. 243 appeared
powdery and fragile, the pH difference was 0.3
indicating no risk of red rot degradation. The powdery
and friable appearance of the cover could be mainly
ascribable to the usage of the manuscript over time
and to the acidic treatment to which the cover had
been subjected to obtain a marbled aspect.
A more valuable indicator of the stability and the
hydrothermal stability of collagen is the measure of
the shrinkage temperature [14] that represents a
reliable measure of the degree of deterioration of
collagen fibres. For this measurement only very small
amount of material is needed.
The triple helix of collagen consists of polypeptidic
chains held together by hydrogen bonds and cross-
links to form elongated fibrils, which bond together to
give rise to fibres with ordered structure. When
heated in water, the hydrogen bonds break and
collagen deforms to randomly disordered chains in a
specific
interval. The deformations
Table 6
Shrinkage temperatures (°C) of the original leather of Mus. 243 cover, before and after the treatment with
nanocollagen 0.7 mg/ml in isopropyl alcohol/water 70/30 (V/V). Average on 3 samples for each treatment.
Shrinkage
temperature
total interval
Initial temperature of shrinkage intervals
interval TA2-TL, where TL
shrinkage
temperature
intervals. Fibres
temperature
(Tlast)
is
not
Final
TL
TB1 (B1)
TF (A1)
TA2 (A2)
33.1 ± 0.6 41.1 ± 1.3 47.6 ± 1.3 70.3 ± 1.1 80.6 ± 4.1
37.2 ± 3.7 41.6 ± 5.2 59.6 ± 4.0 76.8 ± 0.0 77.8 ± 0.5
TE (B2)
TS (C)
ΔTtotal
93.5 ± 3.6
79.9 ± 0.8
60.4 ± 3.0
42.7 ± 2.9
TF=Tfirst; TS=Tshrinkage; TE=Tend; TL=Tlast
The leather treated with the nanocollagen solution
showed a significant increase in the shrinkage
temperature: 47.6°C before the treatment and 59.6°C
after nanocollagen application.
The shrinkage
interval ΔT (59.6 - 76.8°C) is shorter for the treated
leather, indicating an increased uniformity in the
fibres length and arrangement, but occurs at higher
9
Fig. 4. Shrinkage activity of the original Mus. 243 leather. Top: non-treated sample before shrinkage (left) and
at the end of the shrinkage activity (right). Bottom: sample treated with nanocollagen solution 0.7 mg/ml in
isopropyl alcohol/water 70/30 (V/V) before shrinkage (left) and at the end of the shrinkage activity (right).
Magnification 50X. At the end of the shrinkage activity, the non-treated fibres are fragmented and collapsed,
while the treated partially maintain their integrity.
temperatures, highlighting the reconstitution of the
bulk of collagen fibres.
It is to underline that undamaged standard collagen
usually shrinks at TS around 65°C [20] not far from
the values reached by the original leather after the
treatment with nanocollagen.
At the end of all the experimental work, the
nanocollagen solution in isopropyl alcohol/water
70/30 (V/V) was applied by spraying to the flesh side
of the Mus. 243 cover.
Due to the high brittleness of the cover leather, to
ensure a more effective consolidation, it was decided
to apply the solution also to the hair side, but, to avoid
an excessive increase in stiffness, the solution was
applied at reduced concentration (0.35 mg/ml).
Moreover nanocollagen, due its capability to link
fibres together, was used to adhere small fragments
of the cover, which were partially lifted or detached
(Fig. 5).
5. Conclusions
The research presented in this paper allowed for the
synthesis of a new kind of nanostructured nanotubes
of collagen especially conceived for the preservation
of leather artefacts, and able to ensure a real and
durable effectiveness of
the treatment, without
altering the appearance and the equilibrium of the
internal fats and moisture of the leather.
The nanomaterial was developed to exhibit novel
characteristics, such as increased strength, flexibility
and solubility in unusual solvents.
The mechanical tests demonstrated a very positive
increase in tensile and tearing resistance, as well as
in bending resistance, high enough to contrast the
weakness of the original artefact, without creating a
rigid and breakable leather.
SEM imaging well evidenced the capability of the
nanocollagen solution to create bonds between the
collagen fibres and to fill the inter-fibres cavities.
Moreover, after nanocollagen application, the fibres
10
Fig. 5. Top: Area of the leather cover presenting partial detachment and lifting. Bottom: same area after the
nanocollagen application.
appeared less tangled than before and the leather
surface appeared smoother and rebuilt in its integrity.
The measurements of shrinkage
temperature
showed an improvement in the response of collagen
fibres to the degradation, with a partial reconstitution
of the lost elasticity and flexibility. This behaviour can
be explained as a rehydration and a partial restoration
of the triple helixes bonds and the cross-links
between the superhelices forming the quaternary
structure of the collagen molecule.
The aforementioned effects made
the new
synthesised nanocollagen particularly suitable for the
treatment of the volume Mus. 243, which complete
restoration is described in a Master Thesis for the
high training school (SAF) of the Istituto centrale
restauro e conservazione patrimonio archivistico e
librario [21].
Further
the developed product were
performed by applying it to a very deteriorated leather
armchair, with excellent results.
The mass production and marketing of the novel
product is under study.
Appendix A. Supplementary data
References
[1] C. Dignard, J. Mason, Caring for leather, skin
and fur,https://www.canada.ca/en/conservation-
institute/services/care-objects/leather-feathers-
bone/caring-leather-skin-fur.html#a21002. (Last
access March 12th 2018)
tests of
Conservation-booking
[2] G. Ruzicka, P. Zyats, S. Reidell, O. Primanis,
Leather
leather
consolidants, in M. Kite, R. Thompson (eds),
Conservation of leather and related materials,
Oxford, Elsevier Butterworth- Heinemann,
(2006), 230-232.
[3] A. Johnson, Evaluation of the use of SC6000 in
conjunction with Klucel G as a conservation
treatment for bookbinding leather: notes on a
[4]
[6]
[9]
preliminary study, Journal of the Institute of
Conservation, 36 (2013) 125-144.
Ilaria Camerini, La conservazione del fondo
Corsini. Volume 157 G 2. Nuove metodologie
per il trattamento del Red Rot, (2013-14) Master
Thesis at Università Tor Vergata (Roma, Italy),
Unpublished.
regular dimensions
[5] J. Landoulsi, C.J. Roy, C. Dupont-Gillain, S.D.
Champagne, Synthesis of collagen nanotubes
with highly
through
membrane-templated layer-by-layer assembly,
Biomacromolecules, 10 (2009) 1021-1024.
[L. Falcão, M.E. Machado Araújo, Tannins
characterisation in new and historic vegetable
tanned leathers fibres by spot tests, Journal of
Cultural Heritage 12 (2011) 149-156.
[7] K.H. Inoue, A.E. Hagerman, Determination of
Analytical
gallotannin
rhodanine,
Biochemistry 169 (1988) 363-369.
[8] L.J. Porter, L.N. Hrstich, B.G. Chan, The
conversion of procyanidins and prodelphinidins
to cyanidin and delphinidin, Phytochemistry 25
(1985) 223-230.
ISO/CIE 11664-6:2014(E) Colorimetry – Part 6:
CIEDE2000 Colour-Difference Formula.
[10] Test Method T 414 om-12 - Internal tearing
resistance of paper (Elmendorf-type method).
[11] [ISO 4045:2008 (IULTCS/IUC 11) - Leather -
with
Chemical tests - Determination of pH.
[12] [ISO 3380:2015 (IULTCS/IUP 16) - Leather -
Physical and mechanical tests - Determination of
shrinkage temperature up to 100 degrees C.
[13] J. Font, J. Espejo, S. Cuadros, M. R. Reyes, A.
Bacardit, S. Butì, Comparison of IUP 16 and
Microscopic Hot Table Methods for Shrinkage
Temperature Determination, Journal of
the
Society of Leather Technologists and Chemists
94 (2010) 59-64.
[14] R. Larsen, M. Vest, K. Nielsen, Determination of
hydrothermal stability (shrinkage temperature)
of historical leathers by the micro hot table
technique. Society of Leather Technologists and
Chemists Journal 77 (1995) 151-156.
11
of
Journal
[15] B. de Campos Vidal, M.L. S. Mello, Collagen
type I amide I band infrared spectroscopy,
Micron 42 (2011) 283-289.
[16] W.H. Tiong, G. Damodaran, H. Naik, J.L. Kelly,
A. Pandit, Enhancing Amine Terminals in an
Amine-Deprived Collagen Matrix, Langmuir 24
(2008) 11752-11761.
[17] E.L. Wallace, Method for measuring the pH of
leather using a simple glass-electrode assembly,
Journal of Research of the National Bureau of
Standards, 15 (1935) 5-11.
[18] Canadian Conservation Institute, Care of alum,
vegetable, and mineral tanned leather CCI
Notes, 8(2) (1992) 1-4.
[19] V. Dirksen, (1997), The Degradation and
Conservation
of
Conservation and Museum Studies 3 (1997) 6-
10. DOI: http://doi.org/10.5334/jcms.3972.
Leather,
[20] [M. Kite, R. Thomson, Conservation of Leather
and Related Materials, Elsevier Butterworth-
Heinemann, Oxford 2006.
[21] Camilla Del Re, Una nuova risorsa per il restauro
del cuoio: sperimentazione e applicazione del
nanocollagene nel restauro del volume Mus. 243
della Biblioteca Casanatense di Roma, (2015-
2016) Master Thesis at Istituto centrale per il
restauro e la conservazione del patrimonio
archivistico e librario, Scuola di Alta Formazione.
Unpublished.
12
Supplementary data
Confocal Scanning Laser Microscopy (CLSM).
Leather samples were analysed with confocal laser scanning microscope (CLSM), FV1000 (Olympus Corp.,
Tokyo, Japan), using laser channels at l excitation 488 nm and emission range 459-494 nm; and at l excitation
635 nm and emission range 606-686 nm.
CLSM data are a set of two-dimensional (2D) cross-sectional images in the x-y plane obtained with IMARIS
6.2 software (Bitplane AG, Zurich, Switzerland). A set of three-dimensional (3D) values and cross-sectional
images in the xz-yz plane, obtained with IMARIS 6.2 software, were also recorded.
Fig. SI1 displays two images of the flesh side of leather non-treated and after application of nanocollagen at
concentration 0.7 mg/ml in isopropyl alcohol/water 70/30 (V/V).
By comparison between the two images the filling and reconstructing effect of nanocollagen is quite evident;
furthermore, after application of the product, the leather surface appears smoother than before.
Fig SI1: CLSM images. Left: non-treated leather. Right: the same leather after application of nanocollagen.
Magnification 20X. Laser excitation: 635 nm.
Attribution of the characteristics bands of nanocollagen by FTIR.
Table S1. Infrared spectral absorption bands for the nanocollagen final products.
Sample
Nanocollagen
Wavenumber
(cm-1)
3316
1635
1548
1454
1237
Functional group
Reference
N-H stretching
C(=O) Amide I stretching
N-H Amide II bending
C-N stretching
Interaction between N-H
bending and C-N stretching
Spectrometric Identification of organic
compounds, Seventh Edition,
Robert M. Silverstein,
Francis X. Webster
David J. Kiemle
13
|
1804.00600 | 2 | 1804 | 2019-02-22T17:20:25 | From solar cells to ocean buoys: Wide-bandwidth limits to absorption by metaparticle arrays | [
"physics.app-ph"
] | In this paper, we develop an approximate wide-bandwidth upper bound to the absorption enhancement in arrays of metaparticles, applicable to general wave-scattering problems and motivated here by ocean-buoy energy extraction. We show that general limits, including the well-known Yablonovitch result in solar cells, arise from reciprocity conditions. The use of reciprocity in the stochastic regime leads us to a corrected diffusion model from which we derive our main result: an analytical prediction of optimal array absorption that closely matches exact simulations for both random and optimized arrays under angle/frequency averaging. This result also enables us to propose and quantify approaches to increase performance through careful particle design and/or using external reflectors. We show in particular that the use of membranes on the water's surface allows substantial enhancement. | physics.app-ph | physics |
Wide-bandwidth limits to absorption by metaparticle arrays
From solar cells to ocean buoys:
Mohammed Benzaouia,1 Grgur Toki´c,2 Owen D. Miller,3 Dick K. P. Yue,2 and Steven G. Johnson4
1Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA 02139, USA
2Department of Mechanical Engineering, MIT, Cambridge, MA 02139, USA
3Department of Applied Physics and Energy Sciences Institute, Yale University, New Haven, CT 06511, USA
4Department of Mathematics, MIT, Cambridge, MA 02139, USA
In this paper, we develop an approximate wide-bandwidth upper bound to the absorption en-
hancement in arrays of metaparticles, applicable to general wave-scattering problems and motivated
here by ocean-buoy energy extraction. We show that general limits, including the well-known
Yablonovitch result in solar cells, arise from reciprocity conditions. The use of reciprocity in the
stochastic regime leads us to a corrected diffusion model from which we derive our main result: an
analytical prediction of optimal array absorption that closely matches exact simulations for both
random and optimized arrays under angle/frequency averaging. This result also enables us to pro-
pose and quantify approaches to increase performance through careful particle design and/or using
external reflectors. We show in particular that the use of membranes on the water's surface allows
substantial enhancement.
I.
INTRODUCTION.
One of the most influential theoretical results for solar-
cell design has been the ray-optical Yablonovitch limit
[1 -- 8], which provides a bound to how much surface tex-
turing can enhance the performance of an absorbing film
averaged over a broad bandwidth and angular range.
In this paper, we obtain approximate broad-band/angle
absorption limits for a case in which the traditional
Yablonovitch result is not useful: dilute arrays of "meta-
particles"(synthetic absorbers/scatterers). Known limits
bound the absorption at every wavelength [9, 10], but
they tend to be loose when considering large bandwidths
since coherent effects average out [5, 11]. Here, we find
limits on the absorption for arrays of particles that can
be described by the radiative-transfer equation (RTE)
[12, 13]. In particular, we show that an isotropic diffu-
sive regime is optimal for maximizing absorption. This
allows us both to obtain analytical upper bounds (Eqs. 7,
10) and identify the ideal operating regime of absorbing
metaparticle arrays.
In optics contexts, scattering particles can be used to
enhance absorption in thin-film or dye-sensitized solar
cells [15 -- 18]. Most previous work focused on numerical
optimization using the full-wave equations [15, 16] or, in
the case of dye-sensitized solar cells, RTE for random
arrays [17, 18].
In Ref. 19, approximate analytical
estimations of absorption enhancement were given in
cases of optically-thin/thick layers under assumptions
of weak absorption, normal
incidence and isotropic
differential cross section.
In this work, we were actu-
ally motivated by arrays of buoys designed to extract
energy from ocean waves [20 -- 23] depicted in Fig. 1.
Previous numerical-optimization work [14, 24 -- 26],
in
particular a recent extensive computational study on
large arrays [14, 26], showed promising results through
the design of buoy positions. The question we are
trying to answer in this work is more general: given
FIG. 1. Upper left: We bound absorption for very general
arrays of "particles", including arrays of buoys that extract
energy from ocean waves. Upper right: Ocean surface dis-
placement η for a cylindrical buoy array [14] where A is the
amplitude of waves incident from left (arrow). Lower: Sketch
of RTE system.
the absorbing/scattering properties of an individual
metaparticle, is there a limit on the total enhancement
and how can it be reached? The Yablonovitch limit
cannot be applied to all metaparticle arrays since it
requires an effective-medium approximation, which is
only accurate for either dilute weakly interacting dipolar
particles [27] or for strongly interacting particles with
sufficiently subwavelength separation [28], neither of
which is true of the ocean-power problem. Moreover, the
Yablonovitch limit is independent of the precise nature
of the scattering texture, whereas in our case the whole
point is to extrapolate the array properties from the
individual-scatterer properties.
In this paper, we define the interaction factor q(θ)
[29, 30] as the ratio of the power extracted by the ar-
ray to that of the equivalent number of isolated particles
for a given incident angle θ. We first point out that pre-
viously known limits in both solar cells and ocean buoys
arise from reciprocity constraints on the full-wave equa-
tions (Section II). The use of reciprocity in the radiative-
transfer equation leads to a general limit (Eq. 7), valid
for any geometrical configuration in RTE regime, that
is reached through an isotropic distribution of intensity
in the ideal case of small absorption (Section III). This
optimal solution justifies the use of a corrected radiative-
diffusion model (Eq. 10) that predicts the frequency-
averaged performance of random arrays, but also the
angle/frequency-averaged performance of the optimized
periodic array with better than 5% accuracy. This cor-
rected model can be used to estimate the upper bound
on q (which is proportional to the spatially-averaged in-
tensity in RTE framework) even in regimes where the
standard diffusion model is not expected to be accurate.
This result allows us to quickly evaluate the performance
benefits of different metaparticle designs and array con-
figurations, and we show that substantial improvement
is possible if the scattering cross-section is increased (rel-
ative to the absorption cross-section) and/or if partially
reflecting strips are placed on either side of the array
(Section IV). More specifically, we show that the use of
bending membranes on the water's surface around the
buoys significantly increases the interaction factor. We
finally use the corrected radiative-diffusion model to find
optimal parameters that maximize q.
II. RECIPROCITY
the
The
behind
original
intuition
ray-optical
Yablonovitch limit is that the optimal enhancement
is achieved through an isotropic distribution of light
inside the device [1, 2]. This can be thought of as a
reciprocity condition. Reciprocity [13] implies that rays
at a given position cannot emerge in the same direction
from two different paths.
if a given
point in the absorber is to be reached from as many ray
bounces as possible, the rays must be entering/exiting
that point from all angles. More formally, we show in
Appendix-A that reciprocity can be applied to the full
Maxwell's equations in order to relate the enhancement
to the density of states (accomplished in another way by
Ref. 10), leading to:
In consequence,
(cid:104)q(cid:105) =
q(θ)f (θ)dΩ ≤ 4π
n
4π
(1)
where (cid:104)q(cid:105) refers here to the absorption enhancement com-
pared to the single pass, averaged over both polarizations
and over a directional spectrum f (θ) with normalized flux
max
f
θ
(cid:104)ρd(cid:105)
ρv
(cid:90)
((cid:82)
2
4π cos θf (θ)dΩ = 1), (cid:104)ρd(cid:105) is the average density of
states in the device, ρv the free space density of states and
n the index of the absorbing medium. The previous equa-
tion becomes an equality in the case of isotropic incidence
and small absorption. Yablonovitch limit can then be re-
covered in bulk media (ρd = n3ρv) for an incident field
confined to a cone of aperture 2θi (f = 1
δ(θ < θi)):
(cid:104)q(cid:105) ≤ 4n2
π sin θ2
i
.
A similar procedure can be followed in the ocean-buoy
problem. By applying the appropriate reciprocity re-
lation derived from the wave equation, the Haskind --
Hanaoka formula [31], to the absorption of an optimal
array of buoys [29], one can bound the interaction factor
2π f (θ)dθ=1)
(cid:104)q(cid:105) for a given directional spectrum f (θ) ((cid:82)
sin θ2
i
by [9, Appendix-B]:
(cid:104)q(cid:105) =
q(θ)f (θ)dθ ≤ M
kσa
2π max
θ
f
(2)
2π
(cid:90)
where k is the wavenumber, σa the single-buoy absorp-
tion and M the number of degrees of freedom for the buoy
motion (1 -- 6, e.g. 1 for only heave motion). This result
implies that for isotropic incidence, we have (cid:104)q(cid:105) ≤ 1 at
the resonance frequency (the frequency at which the sin-
gle buoy reaches its maximum absorption M/k), while it
can in principle be larger at other frequencies. Although
this sets a general limit valid at any frequency for any
structure, we show in the following that it is not tight
when considering the frequency-averaged performance.
III. RTE LIMITS
We consider a two-dimensional array of scatter-
ing/absorbing particles distributed inside a region S
bounded with a curve C (Fig. 1).
In the case of dilute and non-structured arrays, co-
herent scattering effects average out. This allows one to
use the radiative-transfer equation (RTE) that only in-
volves specific intensity I(r, θ), and that is applicable to
ensemble averages of random arrays at a single frequency
[12, 13]:
eθ · ∇rI = −ρσeI + ρσs
dθ(cid:48)p(θ, θ(cid:48))I +
(3)
(cid:90)
where σs, σa and σe denote respectively the scattering,
absorbtion and extinction cross sections of the individual
particles (σe = σs + σa), p the normalized differential
cross section, ρ the particles' density, eθ the unit vector
with direction θ and internal sources.
We conjecture that a similar averaging of coherent
effects arises from averaging over frequency and/or an-
gle, and below we demonstrate numerically that this al-
lows RTE to make accurate predictions even for a small
number of random samples or for optimized periodic ar-
rays. This is similar to optical light trapping where
Yablonovitch model can predict the frequency/angle-
average performance of textured solar cells even though it
cannot reproduce the exact spectral or angular response
[5, 11].
A. General limit
3
One can first define a surface Green's
Similarly to our previous discussion of reciprocity-
based limits from the wave equation, we now use reci-
procity constraints on RTE to obtain general limits on
the interaction factor q.
function
Gs(r, θ; r(cid:48), θ(cid:48)) [32] giving I(r, θ) for an incident field
Ii(ri, θi) = δ(θi − θ(cid:48))δ(ri − r(cid:48)) and no internal sources
= 0. Similarly, a volume Green's function Gp(r, θ; r(cid:48), θ(cid:48))
can be defined as the intensity I(r, θ) obtained with
no incident field Ii = 0 and a point source (ri, θi) =
δ(θi − θ(cid:48))δ(ri − r(cid:48)).
(cid:82)
(cid:82)
We recall that the flux density F is defined as
2π Ieθdθ. Conservation of energy [12] then leads to
we have Pe =(cid:82)
C F · noutdr = Pe − Pa where Pe and Pa are the gener-
(cid:90)
(cid:90)
S (r, θ)drdθ = 1 so that:
Gp(r, θ; r(cid:48), θ(cid:48))(eθ · nout)drdθ = 1 − Pa (4)
ated and absorbed power respectively. For a unit source,
C
eθ·nout>0
To bound this last expression, we need a lower -bound
for Pa. By noting that the intensity at any point is
larger than the single pass value (obtained after extinc-
tion without multiple scattering), we have:
2π
Gp(r, θ; r(cid:48), θ(cid:48))drdθ
e−ρσer−r(cid:48)
r − r(cid:48) δ[angle(r − r(cid:48)) − θ(cid:48)]dr
(5)
(cid:90)
(cid:90)
(cid:90)
S
S
Pa = ρσa
≥ ρσa
σa
σe
=
Hρσe (r(cid:48), θ(cid:48))
(cid:90)
(cid:90)
where Hρσe (r(cid:48), θ(cid:48)) defined in the previous equation can
be interpreted as the power absorbed by a medium with-
out scattering and with an absorption coefficient ρσe in
the presence of a unit source at the point r(cid:48) emitting in
direction θ(cid:48).
Finally, we relate Gs to Gp through reciprocity using
Gp(r, θ; r(cid:48), θ(cid:48))eθ · nout = Gs(r(cid:48), π − θ(cid:48); r, π − θ) [32]. We
conclude from Eq. (4) and Eq. (5) after a simple change
of variable that:
C
eθ·nout<0
Gs(r(cid:48), θ(cid:48); r, θ)drdθ ≤ 1− σa
σe
Hρσe (r(cid:48), π − θ(cid:48))
(6)
with equality always realized in the absence of absorp-
tion.
Since the interaction factor in RTE is given by q =
I(r(cid:48), θ(cid:48))dθ(cid:48)(cid:105)r(cid:48)/Ii where Ii is the incident intensity
and (cid:104).(cid:105)r(cid:48)
is the average over r(cid:48) in S, we can therefore
bound the interaction factor q for a given directional
0
(cid:104)(cid:82) 2π
FIG. 2. Upper: Frequency-averaged interaction factor qs
vs incident angle θ for Nx × Ny = 3 × 30 arrays of buoys
from exact solution [14] (solid lines), compared to standard-
diffusion (black dashed lines), corrected-diffusion (red dashed
lines) and radiative-transfer (RTE with Monte Carlo simu-
lation, dots) models. (q = array absorption / isolated-buoys
absorption.) The average buoy spacings (randomly chosen via
a Gamma distribution) are dx/h = 1.73, dy/h = 3.63, with h
= ocean depth (the density is ρ = 1/dydx). Numbers in leg-
end are qs averaged over θ for a typical ocean-wave directional
spectrum cos2s θ with s = 4 [33]. Inset: q vs. wavelength at
θ = 0, where shaded regions is one standard deviation from
mean value (blue line) for 100 random structures. Lower:
(cid:104)q(cid:105) for over isotropic incidence. Results compared to limit in
Eq. (7).
spectrum f (θ) [fraction of power incident from angle θ]:
(cid:104)q(cid:105) =
(cid:90)
(cid:90)
(cid:90)
(cid:20)
2π
eθ·nout<0
C
≤ 2π
1 − σa
σe
f (θ)(cid:104)Gs(r(cid:48), θ(cid:48); r, θ)(cid:105)r(cid:48)drdθ(cid:48)dθ
(cid:21)
h(ρσe)
max
θ
f
(7)
where h(ρσe) = (cid:104)Hρσe (r(cid:48), θ(cid:48))(cid:105)r(cid:48),θ(cid:48) ≥ 0. In the case of a
(cid:90) π/2
"slab" of thickness d, we can show that [Appendix-C]:
1 − e1(xd)
0
xd
, ei(x) =
h(x) = 1 − 2
π
e−x sec α cosi αdα
(8)
Note that the bound in Eq. (7) reaches its maximal
value 2π max f in the limit of small absorption. This
maximal value, which does not assume optimal single-
buoy absorption, generalizes then the previous ocean-
buoy bound, giving (cid:104)q(cid:105) ≤ 1 for isotropic incidence f =
In addition,
1/2π at any wavelength in RTE regime.
(cid:104)q(cid:105) = 1 is always realized in the small absorption limit.
This special case is sometimes referred to as Aronson's
theorem [34].
The equality in Eq. (7) is reached for:
Gs(r(cid:48), θ(cid:48); r, θ)dr =
1 − σa
σe
Hρσe(r(cid:48), θ(cid:48))
δ(θ−θm)
(cid:20)
(cid:21)
(cid:90)
C{eθ·nout<0}
(9)
where θm = arg max f . This means that the interac-
tion factor should be equal to zero for any incident an-
gle different from θm. In the ideal case of small absorp-
tion, the optimal Gs becomes independent of θ(cid:48), which
corresponds to isotropic interior intensity, similar to the
Yablonovitch model. Therefore, in order to explore opti-
mal solutions of RTE, we solve it under the assumption
of nearly isotropic intensity, which is well known to lead
to a diffusion model [12, 13, 32]. We emphasize that not
all RTE systems are diffusive, but our result above shows
that the optimal (cid:104)q(cid:105) is attained in an isotropic diffusive
regime.
B. Radiative-diffusion model
Appendix-E, but in the absence of reflecting walls (Ri =
0) they simplify to q0(θ) = ξ(υe sec θ) and:
4
D = − C(1 + e−υe sec θ) + β (C+γp1 cos2 θ)
(1 + e−υd ) + β
(1−p1) cos θ (1 − e−υe sec θ)
υe(1−p1) (1 − e−υd )
υd
,
(11)
where p1 = σsµ/σe and β = π/4 (resp. = 1) in 2D (resp.
3D).
Equation (10) with η = 1 is obtained from the stan-
dard diffusion model. However, it is also known that
the diffusion solution is inaccurate for small thicknesses
[36 -- 38]. A major problem is that it does not guarantee
(cid:104)q(cid:105) = 1 for isotropic incidence and negligible absorption,
even though we previously mentioned that this is the case
for any solution of RTE. The reason behind this problem
is that the term Iri is not isotropic even for an isotropic
incidence. For large thicknesses, however, the contribu-
tion of the term Iri becomes negligible and the diffuse
term Id can ensure an isotropic solution. This simply
means that the higher order terms in the expression of
Id cannot be neglected for small thicknesses. In order to
keep the simplicity of the diffusion solution, we suppose
that the effects of higher order terms can be incorpo-
rated by the introduction of a scalar term in the diffuse
intensity ηId instead of Id. η is then defined so as ensure
the condition (cid:104)q(cid:105) = 1 for isotropic incidence and zero
absorption. This procedure is somewhat similar to the
approach in Ref. 37 except that we use a constant factor
η since we are interested in the total q and not the spa-
tially resolved I. In order to define η, we study the limit
of negligible absorption for which υd → 0, C → −2 cos2 θ
and D → cos2 θ(1− e−υe sec θ) + π
4 cos θ(1− e−υe sec θ). Af-
ter simplification, the condition (cid:104)q(cid:105) = 1 allows to define
η as:
β + π
8 γ − 2γ
3υe
π
[1 − e1(υe)]
2 − 1
− βe1(υe) + γ
υe
2 e2(υe) + γ
υe
.
e3(υe)
(12)
We note that, as expected, η → 1 for an absorber that
is thick compared to the extinction length. From our
discussion above, this corrected radiative-diffusion model
can now be used to estimate the upper bound on the in-
teraction factor even in regimes where the standard diffu-
sion model is not expected to be accurate (optically thin
or large absorption).
IV. OCEAN-BUOY ARRAYS
A. Example
We now present a validation of the accuracy of Eq. (10)
in a model of ocean-wave energy converter (WEC) con-
sisting of a truncated cylinder in heave motion (Fig. 1).
The isolated-buoy properties can be obtained analytically
[39 -- 41] and are depicted in Fig. 3: they are designed
[14] to have an absorption resonance that matches the
Unless otherwise stated, we restrict ourselves to scat-
terers distributed inside a slab of thickness d (Fig. 1).
η =
In addition to RTE parameters and reflection coeffi-
cients at the boundaries (Ri), the radiative-diffusion so-
lution uses an asymmetry factor (µ) [35, Appendix-F] of
the single particle (Fig. 3). The intensity is then given
by I = Iri + Id: Iri is the reduced intensity, solution
of cos θ∂xIri = −ρσeIri, and Id is the diffuse intensity
π F(x) · eθ where U verifies a
approximated by U (x) + 1
diffusion equation with flux-matching boundary condi-
tions [Appendix-D]. By defining the cross sections per
unit of length as υs,a,e = ρdσs,a,e, the model predicts an
interaction factor q of:
ξ(υd)
η
D
+ 1
+ C
(10)
ξ(υe sec θ)
q(θ) = q0(θ)
d = γυa(υe − υsµ) is the diffusion coefficient [γ =
where υ2
2 (resp. = 3) in 2D (resp. 3D)], ξ(x) is the function (1−
d − (υe sec θ)2], D is given
e−x)/x, C = γ[υs(υe + µυa)]/[υ2
by the boundary conditions, q0(θ) is the reduced factor
and η is an additional correction term that we discuss
later. General formulas for q0(θ) and D are given in
(cid:18)
(cid:20)
(cid:21)
(cid:19)
typical Bretschneider spectrum [42] of ocean waves. We
choose the array density based on an earlier optimized pe-
riodic 3-row WEC arrangement [14]. For this density, we
then compare the exact numerical scattering solution cal-
culated for both random and optimized-periodic arrays
(using the method from Ref. 14) to both the analytical
radiation-diffusion q from Eq. (10), with and without the
correction η, and the numerical solution of RTE model
by a Monte Carlo method [43].
In Fig. 2 (upper plot), our corrected model agrees to
< 2% accuracy with exact solutions for random arrays
at θ < 80◦, as long as the results are frequency-averaged.
The importance of frequency averaging is shown by the
q frequency spectrum shown in the inset for θ = 0◦. For
an ensemble of random structures, this spectrum exhibits
a large standard deviation (gray shaded region), due to
the many resonance peaks that are typical of absorption
by randomized thin films [3, 5], but the frequency av-
erage mostly eliminates this variance and matches our
predicted q(θ). Precisely such an average over many res-
onances is what allows the Yablonovitch model to ac-
curately predict the performance of textured solar cells
even though it cannot reproduce the detailed spectrum
[5, 11].
At first glance, our model does not agree in Fig. 2
with the performance of the optimized periodic array
from Ref. 14: the periodic array, which was optimized for
waves near normal incidence, is better at θ near 0◦ and
worse elsewhere. However, when we also average over
θ (from a typical ocean-wave directional spectrum [33]),
the result (shown as a parenthesized number in the legend
of Fig. 2) matches Eq. (10) within 5%. If we average over
all angles assuming an isotropic distribution of incident
waves, the results match within 1%. Similar results have
been observed for thin-film solar cells, in which an opti-
mized structure can easily exceed the 4n2 Yablonovitch
limit for particular incident angles, but the Yablonovitch
result is recovered upon angle-averaging [4, 5, 8, 11].
Finally, we note in Fig. 2 (lower plot) that RTE re-
sults respect indeed the bound in Eq. (7) for isotropic
incidence. In particular, we confirm that random arrays
achieve (cid:104)q(cid:105) = 1 for small absorption (i.e.
small wave-
length in our case). The periodic array, on the other
hand, doesn't satisfy this relation unless it is frequency
averaged. We also mention that the limit Eq. (7) is very
loose for anisotropic incidence and cannot be reached
without using external reflectors as discussed in Section
IV-B below.
B. Larger interaction factor
Given this model, we can now explore ways to increase
the interaction factor q. By examining the dependence
of q in Eq. (10) on the parameters (Fig. 4), we find that
for a fixed scattering-to-absorption ratio σs/σa, q reaches
a maximum qmax for an intermediate value of scattering
per unit length ρdσs, whereas it increases monotonically
5
FIG. 3. Properties of a single truncated-cylinder wave en-
ergy converter (WEC) in heave (vertical) motion, with radius
a = 0.3h and draft H = 0.2h where h is the ocean depth. The
WEC has an isotropic response with respect to the direction
of the incident field. Left: Scattering and absorption cross
sections of a single buoy normalized to the cylinder diame-
ter (σ/2a). The ocean spectral energy density (energy per
horizontal surface) is chosen as Bretschneider [42] with res-
onant frequency matching that of the body and is shown in
units of ρgH 2T (ρ is the water density, g the acceleration of
gravity, T the mean wave period and H the significant wave
height). Right: Asymmetry factors, defined as the average of
cos φ and cos 2φ for the two-dimensional differential scatter-
ing cross section. These parameters enter into the diffusion
equation as µ = (µ1 − µ2)/(1 − µ2) and with σs replaced by
σs(1 − µ2) [35, Appendix-F].
with µ. A maximum µ is achieved by increasing µ1 (for-
ward scattering) and decreasing µ2 (lateral scattering).
The optimal value of ρdσs and qmax both increase with
σs/σa; as the single particle absorbs more, the interaction
factor decreases and the optimal configuration requires a
larger spacing between the particles. The maximum q is
then achieved in the limit of small absorption (ρdσa (cid:28) 1)
and large scattering (ρdσs (cid:29) 1) for which we obtain a
perfect isotropic diffuse intensity.
From Fig. 3, we see that we have σa/σs ≈ 1 at the
resonance of the WEC. In this case, the enhancement
is expected to be smaller than 1 around the resonance
and optimal structures will tend to have a large spacing
dy. (If the array were optimized for small wavelengths
λ, where σs (cid:29) σa, then a larger q could be obtained
at those wavelengths, but qs would be worse because
the optimal spacing in this case is too small for good
performance at the resonance.) Still, multiple scatter-
ing significantly improves the broadband performance of
our array: our (cid:104)q(cid:105) ≈ 0.99 is larger than the (cid:104)q0(cid:105) ≈ 0.78
that is obtained from RTE in absence of multiple scat-
tering (reduced factor q0). The performance is still lower
than the 1.65 that would be obtained for σs (cid:29) σa in
the ideal isotropic regime discussed below, essentially be-
cause σa/σs is too small and the structure is too thin
(as for example quantified by the transport mean-free
H h 2a path d/ltr = υs(1 − µ) ≈ 0.5 for 2a
achieve an isotropic diffuse intensity.
λ
(cid:38) 0.3) to practically
Alternatively, we show that q can be enhanced by
putting partially reflecting strips around the array. Sim-
ilar to light-trapping by total internal reflection [1, 2],
one possibility is to use a strip of a lower-"index" [31]
medium (compared to the array's ambient medium) on
either side of the array. In the ocean-buoy problem, this
can for example be achieved by either a depth change
or the use of a tension/bending surface membrane which
can lead to near-zero index [44, 45]. This modifies equa-
tions (2 -- 4) with additional reflection coefficients Ri, as
given in Appendix-E.
In Fig. 4, we show the effect of an increase in the scat-
tering cross section and/or the index contrast for the
same array studied before. By combining both effects,
a large (> 3) spectral interaction factor can be achieved
at normal incidence. At the same time, waves incident at
large angles will be reflected out, so that the interaction
factor integrated over isotropic incidence is still smaller
than 1. For a given directional spectrum and scattering
cross section of a single buoy, the optimal interaction fac-
tor is achieved for a specific value of the index contrast
as can be seen in Fig. 4 (right).
Finally, it is instructive to look at the ideal case of
small absorption and large scattering, for which Eq. (10)
simplifies to:
q(θ) = [1 − R1(θ)]
+ cos θ
cos θ
(13)
(cid:16) π
4α
(cid:17)
−π/2 cosi(θ)dθ.
(cid:82) π/2
−π/2 R1(θ) cosi(θ)dθ/(cid:82) π/2
where R1 is the reflection coefficient of the front-
surface and α = (1 − r1)/(1 + r2) with ri =
Equation (13)
still gives 1 when averaged over isotropic incidence, but
the interaction factor is larger near normal incidence.
Without reflectors (R1 = 0), the maximum value of q
at normal incidence is 1 + π
4 , and the previous direc-
tional spectrum gives (cid:104)q(cid:105) ≈ 1.65. This maximum value
of q(0) does not reach the arbitrarily large enhancement
allowed by Eq. (7). However, q(0) can still be made suf-
ficiently large by including a reflector designed for trans-
mission near normal incidence and reflection elsewhere
(since α → 0).
6
FIG. 4. Upper: Dependence of q(0◦) on parameters in absence
of reflecting boundaries. In the left plot, we take σs/σa = 5.
In the right plot, we show the optimal ρdσs and qmax for dif-
ferent values of σs/σa and µ. Lower: Effect of a change in the
index contrast and scattering cross section on the bandwidth-
averaged factor qs for the same array in Fig. 2. We tune the
index n1 along a strip surrounding the array, with n0 being the
index of the array's ambient medium. We suppose that the
WEC has new scattering cross section σs, but keep the same
absorption cross section. Left: qs at normal incidence. Right:
qs averaged over θ with a directional spectrum of cos2s θ and
s = 4.
C. Surface membrane
We now use a specific example to demonstrate a larger
interaction factor q using surface membranes surrounding
the WEC array. For large scale applications, such mem-
branes could be designed to have the desired properties
by connecting floating pontoons with elastic elements of
appropriate stiffness.
A thin bending membrane on the water surface changes
the "refractive index" (∼ k/ω) through the following dis-
persion relation (e.g. Ref. 46):
ω2 = gk tanh(kh)
1 + Cb(kh)4
1 + m · kh tanh(kh)
(14)
where ω is the frequency, g the acceleration of gravity,
k the wavenumber, Cb is a dimensionless bending coef-
ficient, m is the mass of the membrane relative to the
mass of the water beneath it and h is the water depth.
We simply assume m = 0 in the following.
At a fixed ω, the membrane decreases k (decreases the
"index") compared to the surrounding medium. This
change of index leads to a reflection off the membrane's
edges.
In particular, total internal reflection traps the
water waves similarly to light trapping in solar cells,
which increases the interaction factor q. The reflection
coefficient, which depends on ω, Cb, the incident angle
and the membrane's width w, can be computed by ap-
plying appropriate boundary conditions on either side of
7
the front and rear membranes, respectively, we find the
optimal membrane widths that maximize the radiative-
diffusion bound. The resulting optimized (cid:104)qs(cid:105) values are
shown are shown in Fig. 5 (upper plot). We first note that
the frequency/angle-averaged interaction factor (cid:104)qs(cid:105) in-
creases significantly (> 1.8) compared to the (cid:104)qs(cid:105) = 1.00
without the membranes. We also confirm that (cid:104)qs(cid:105) in-
creases with Cb2 (rear membrane) as expected. On the
other hand, there is an optimal value for Cb1 depending
on the directional spectrum f (θ). For a focused incident
field, only angles near normal incidence matter so that
Cb1 can be increased allowing more of the waves scat-
tered by the WECs to be trapped. On the other hand,
for a broad directional spectrum, a large value of Cb1 pre-
vents waves incident from wide angles from reaching the
WECs.
For our array, supposing for example that the max-
imal attainable value of Cb2 is equal to 2, the optimal
value for Cb1 is 0.048 with optimal widths equal to 1.6h
for both the front and rear membranes. The frequency-
averaged interaction factor qs for the optimal parameters
is shown in Fig. 5 (lower plot). Our predicted bound (red
dashed line = corrected diffusion) is indeed larger than
the actual performance of the array as modeled by RTE
(orange dots). That is mainly due to the relatively small
scattering cross section compared to the absorption cross
section. As illustrated in the inset of Fig. 5 at small wave-
lengths where σs is large (Fig. 3), we see that an increase
in the scattering cross section leads to arrays with per-
formance closer to the radiative-diffusion bound.
We finally mention that in the case of using a perfect
back-reflector, (cid:104)qs(cid:105) can reach a value of 2.26 for Cb1 =
0.06 and w1 = 1.65h.
V. CONCLUSION.
We believe that the angle/frequency-averaged limits
presented in this paper provide guidelines for future de-
signs to achieve a large q factor which may open the path
for the realization of large arrays of buoys for efficient
ocean energy harvesting. In particular, the use of exter-
nal reflecting elements such as surface membranes seems
a promising approach. The results are also applicable to
other problems where multiple scattering effects are used
to achieve enhancement, including scattering particles in-
side an absorbing layer. One can, for example, recover
the standard Yablonovitch-4n2 result from our approach
in an appropriate limit [Appendix-H], but the real power
of our result is that it allows to study the effect of single-
metaparticle properties, angle of incidence and reflecting
boundaries.
ACKNOWLEDGMENTS
This work was supported in part by the Army Research
Office under Cooperative Agreement Number W911NF-
FIG. 5. Upper: (cid:104)qs(cid:105) with a directional spectrum of cos2s θ
and s = 4 for different values of Cb1 and Cb2 correspond-
ing to the front and back membranes respectively. Each
point is obtained after optimizing over the membranes' thick-
nesses. Lower: Frequency-averaged interaction factor qs vs
incident angle θ for the previously studied array using ad-
ditional membranes with parameters (Cb1, Cb2) = (0.048, 2)
and w1 = w2 = 1.6h.
the membrane and using a transfer-matrix method as re-
viewed in Appendix-G. We note that evanescent modes
need to be included because of the change in dispersion
relations.
The index contrast increases with Cb (increasing stiff-
ness), which increases the range of angles undergoing to-
tal internal reflection, making a more effective mirror.
Since no waves are coming from the rear of the array,
the optimal membrane behind the array should be a per-
fect reflector (Cb → ∞, limited only by the attainable
practical Cb).
We can now use our corrected diffusion model to pre-
dict the upper-bound for the previously studied array as
we change Cb. For each value of Cb1 and Cb2 representing
Buoys !" !# !$ %2 %1 18-2-0048.
Appendix A: Enhancement from reciprocity of
Maxwell's equations
field of the point source:
fs(k)2dk =
(cid:90)
1
η
(cid:90)
Now, we use the Poynting theorem to compute the far
8
(A5)
a] · kdS
(cid:90)
Re[Ea × H∗
a · Ea]
≤ −
= Im[Ea(r0) · es]
1
ωµ
Re[J∗
(cid:90)
(cid:82)
(cid:90)
Although the end result is not new, we wish to empha-
size that the underlying ideas of the Yablonovitch and
LDOS limits are closely tied to reciprocity. This is an al-
ternative to the derivation in Ref. 10, which differs in that
it directly uses the reciprocity (or generalized reciprocity)
from Maxwell's equations. As was also emphasized in
Ref. 10, the result also applies to linear nonreciprocal
systems, since the density of states of transposed-related
materials is the same (G(r, r) = Gt
t(r, r) [13]).
Here for simplicity, we consider a reciprocal system in
the derivation. We have then:
[Ea× Hb− Eb× Ha]· k dS =
S∞
(cid:90)
V
[Ea· Jb− Eb· Ja]dV
(A1)
a = 0) and (Jb = 0,
a] · k dS = 0.
(cid:90)
a] = − 1
η
If we choose (Ja = 1
b = ejkk0·reb), then Ea = ¯¯GE(r0, r0)es.
Einc
The far field term can be written as Es
jµω esδr0, Einc
a =
fs(k) ejkr
, and sim-
ilarly for the far-field of the scattered field "b", so that:
S∞[Es
We then expand the integrand of the left term in A1
η (k × Es
b × Hs
r ea, Hs
a × Hs
(cid:113) µ0
a) with η =
b − Es
a = 1
0
to obtain:
a × Hinc
b − Einc
b × Hs
[Es
fs(k)ejkr(1+k·k0)
S∞
[(ea · eb)(1 − k · k0) + (ea · k0)(eb · k)]rdk
(A2)
The integral can be evaluated using the method of sta-
tionary phase [47]. The function g(θ, φ) = 1 + k · k0 =
1 + cos θ cos θ0 + sin θ sin θ0 cos(φ − φ0) has two extrema
at ± k0. The integrand is null at the first, so only the
second matters. The Hessian matrix at − k0 is given by:
. We then conclude that the integral we want
(cid:20)1
(cid:21)
0
0 sin θ2
0
to evaluate is equal to:
− 1
η
j
1
sin θ0/2
1
kr
[2(ea · eb)) − (ea · k0)(eb · k0)]
fs(−k0)r sin θ0 = − j
η
4π
k
(ea · eb)fs(−k0)
(A3)
where ea is evaluated at −k0.
We finally conclude from A1 that:
−es · Eb(r0) = 4π(ea · eb)fs(−k0)
(A4)
which is the reciprocity relation relating the far field of a
point source at r0 in the direction −k0 to the field at r0
due to an incoming plane wave from the same direction.
At this point we are able to combine A4 and A5 to find
our main result about the enhancement. We consider an
incoming angular distribution f (k0) with a normalized
4π cos θf (k0)dk0 = 1). By integrating over all
coming angles and polarizations of the "b" field, we have:
flux ((cid:82)
(cid:90) (cid:88)
Eb2f (k0)dk0 =
Eb · es2f (k0)dk0
(cid:90) (cid:88)
(cid:90) (cid:88)
(cid:90) (cid:88)
eb,es
eb
= (4π)2
= (4π)2
es
≤ (4π)2 max f
k
= (4π)2 max f
k
= (4π)2 max f
k
eb,es
ea · eb2fs(−k0)2f (k0)dk0
fs(−k0)2f (k0)dk0
Im[Ea(r0) · es]
(cid:88)
es
Tr[Im ¯¯GE(r0, r0)]
πc2
ωn2 ρd(r0)
(A6)
which relates rigorously the enhancement and the local
density of states.
We can use this result to compute the absorbed power
and deduce the enhancement compared to the single pass
for a cell of surface S and effective thickness d. We have:
(cid:104)Pabs(cid:105) =
ω(cid:48)(cid:48)0
1
2
(cid:90)
(cid:90) (cid:88)
V
eb
Eb2f (k0)dk0
(cid:90)
(A7)
ρd
≤ 1
2
(cid:48)(cid:48)0(4π)2 πc3
ωn2 max f
V
(cid:82) f (k0) cos θdk0 × 2 × S =
The total incident power, taking into account the two
polarizations, is given by 1
2η
S
η , and the normalized single pass absorption is αd =
(cid:48)(cid:48)
n
ω
c d. The enhancement is then given by:
(cid:104)q(cid:105) =
(cid:104)Pabs(cid:105)
Pincαd
≤ 4π
n
(cid:104)ρd(cid:105)
ρv
max f
(A8)
where ρv = ω2
2π2c3 is the free space density of states. This
inequality becomes an equality in the case of negligible
absorption and isotropic incidence (f = 1
For a bulk dielectric, we have ρd = n3ρv so that (cid:104)q(cid:105) ≤
2n2 for isotropic incident light which is the standard limit
in the absence of a back reflector.
2π ).
Appendix B: Interaction factor from reciprocity in
ocean waves
In this section we review the result in Ref. 9 and em-
phasize that it is also a consequence of reciprocity, which
shows the similarity with the LDOS limit in solar cells.
The problem of ocean wave energy extraction using
oscillating bodies is formally equivalent to the problem
where there are discrete sources of which the amplitude
can in principle be controlled externally (velocity of the
body that can be controlled through an external mechan-
ical mechanism). Considering the effect of the incoming
wave and interaction between bodies, the total absorp-
tion can be written as a quadratic function in terms of
the amplitudes of the different sources as in [29] for ex-
ample. Maximizing the absorption allows to find the op-
timal amplitudes as a function of the scattered field and
the radiated fields from the sources. This gives [29]:
Pmax =
∗(θ)R−1Fe(θ)
Fe
1
8
(B1)
where Fe(θ) is the force applied on the bodies for an
incident wave from the direction θ and R is the resistance
matrix (radiation damping matrix).
One would try to see the effect of the reciprocity re-
lations discussed before on the maximum absorption in
this context. The exact equivalent of Eq. (A4) is al-
ready known in the ocean waves problem as the Haskind-
Hanaoka formula that relates the force applied on a body
due to an incident wave to the radiated field when the
the body acts as a source [31]. It leads to:
Fe,i(θ) = − 4
k
ρogAcgAi(θ + π)
(B2)
where A is the amplitude of the incident wave, Ai is
the far-field amplitude of the radiation mode i, k is the
wavenumber, cg is the group velocity, ρo is the water
density, and g is the gravity of Earth.
(cid:82)
The use of this formula on the maximum absorbed
power by an array of oscillating bodies leads to the
bound on the power absorbed by the array. For a given
incident angular distribution f (θ) normalized so that
2π f (θ)dθ = 1:
(cid:104)Pmax(cid:105) =
f (θ)Pmax(θ)dθ
(cid:90)
9
a,max = Pmax/( 1
2 ρogA2cg) is the maximum ab-
where σN
sorption cross section of the array, N is the number of
buoys and M is the number of degrees of freedom for the
buoy motion (1 -- 6 [29], e.g. 1 for only heave motion).
This result is general and does not depend on assump-
tions on the scatterers.
It means that the interaction
factor q = (cid:104)σN
a,max(cid:105)/N(cid:104)σ1
a(cid:105) is bounded by M/(k(cid:104)σ1
a(cid:105)) for
isotropic incidence. For buoys in heave motion which are
studied in this paper, we have M = 1 and (cid:104)σ1
a(cid:105) = σ1
a
(the absorption cross section of the single buoy does not
depend on the incident angle).
Note that Eq. (B4) is also valid for a single buoy. De-
pending on the symmetries of the buoy, the actual ab-
sorption may be smaller (for an axisymmetric buoy, we
always have kσ1
It is important to realize that this bound is equal to 1
a(cid:105) reaches the
at the resonance frequency [the k where (cid:104)σ1
maximum M/k from (B4)], while it can in principle be
larger at other frequencies.
a ≤ 3 [31]).
Appendix C: General RTE limit for a "slab"
We compute the function h in Eq. (7) for a slab of
thickness d (with perfectly transmitting boundaries). We
assume that the slab is normal to the x -axis.
We first write the integral H using polar coordinates
(cid:90) x(cid:48)
cos θ
(r, θ):
(cid:90) π/2
(cid:90) d−x(cid:48)
cos θ
Hα(x(cid:48), θ(cid:48)) =
(cid:90) π/2
+
−π/2
0
0
−π/2
αe−αrδ(θ − θ(cid:48))dθdr
αe−αrδ(θ − θ(cid:48))dθdr
After simplification, we have then:
(cid:90) 2π
(cid:90) d
(cid:34)
0
0
h(α) =
1
2πd
Hα(x(cid:48), θ(cid:48))dx(cid:48)dθ(cid:48)
(cid:90) π/2
0
= 1 − 2
παd
1 −
e−αd sec θ cos θdθ
(C1)
(cid:35)
(C2)
(cid:90)
1
8
≤ max f
= max f
(cid:88)
Pmax(θ)dθ
[R−1]i,j
(cid:90)
i,j
2π
F ∗
e,iFe,jdθ
Using
πk ρogcg Re((cid:82)
B2
(cid:90)
2π A∗
2
(cid:104)σN
a,max(cid:105) =
the
and
i Aj) [29], we conclude that:
fact
that Ri,j
a,max(θ)f (θ)dθ ≤ N M
σN
k
2π
2π max f (B4)
(B3)
=
Appendix D: Diffusion equation
Here we reproduce the diffusion equation as in Ref. 12,
13 but adjusting the numerical coefficients for a two-
dimensional medium.
We first separate the intensity as I = Iri + Id where
Iri is the reduced (coherent) intensity and I = Id is the
diffuse (incoherent) intensity. The reduced intensity is
related to the single scattering and obeys: eθ · ∇rIri =
−ρσeIri. So from RTE equation, the diffuse intensity
obeys:
eθ · ∇rId = −ρσeId + ρσs
(cid:90)
(cid:90)
dθ(cid:48)p(θ, θ(cid:48))Id + J,
J = ρσs
dθ(cid:48)p(θ, θ(cid:48))Iri
(D1)
In order to obtain U and F we apply the operators
Now, considering the diffusion approximation, we
write: Id(r, θ) = U (r) + 1
as a first order series in θ. We also note that the diffuse
π F(r) · eθ. This could be seen
flux is: (cid:82) Ideθ dθ = F.
(cid:82) dθ and(cid:82) eθdθ on (D1). This leads to:
(cid:90)
∇r · F = −2πρσaU + 2πρσsUri
dθ Iri(r, θ)
(cid:90)
where σtr = σe(1 − p1) and σep1 =(cid:82) dθ(cid:48)p(s, s(cid:48))[s · s(cid:48)], so
ρσtrF +
dθ Js
1
Uri(r) =
2π
∇rU = − 1
π
(D2)
1
π
(cid:20)
(cid:21)(cid:20)A
(cid:21)
with Ri = Ri(θ) and Y = e−υe sec θ.
10
D is given through boundary conditions by D =
A+B
1+ R2Y
, where:
)e−υd
α1 + β υd
υtr
(cid:34)
(α2 − β υd
−
υtr
C(1 + R2Y 2)α1 + β υe
υtr
[C(1 + R2)α2 − β υe
(α1 − β υd
)e−υd
)
υtr
= X =
(cid:35)
(α2 + β υd
υtr
cos θ + γp1 cos θ)(1 − R2Y 2)
( C
cos θ + γp1 cos θ)(1 − R2)]Y
( C
B
υtr
p = (cid:82) π/2
−π/2 Ri(θ) cosp(θ)dθ/(cid:82) π/2
with υtr = υe − υsµ, αi = (1 − ri
2) and
ri
−π/2 cosp(θ)dθ. We recall
that (γ = 2, β = π/4) [resp. (γ = 3, β = 1)] in 2D [resp.
3D].
1)/(1 + ri
(E2)
The correction term η, which ensures that the interac-
tion factor for isotropic incidence and zero absorption is
1, is defined as:
(cid:90) π/2
π − 2(cid:80)
i=1
0
q(i)
0 dθ
q(i)
0 D(i)
0 (θ, υe, υtr)
ξ(υe sec θ)
− γ cos2 θq(i)
0
(cid:35)
dθ
(E3)
that p1 = σsµ/σe where µ is the average of the cosine of
the scattering angle.
η =
Equations (D2) allow to solve for U and F. Combining
them, we obtain a diffusion equation for U :
∇ ·
∇2U − (ρσd)2U = −2ρ2σtrσsUri +
1
π
(cid:90)
dθ Js (D3)
with:
(cid:90) π/2
(cid:34)
2(cid:80)
i=1
0
Now we need to add appropriate boundary conditions.
Supposing that we have a reflection coefficient R on the
surface, this should be: Id(r, θ) = R(θ)Id(r, π − θ) for s
directed towards the inside of the medium. However, con-
sidering the assumed formula for Id the condition cannot
be satisfied exactly . A common approximate boundary
condition is to verify the relation for the fluxes:
R(θ)Id(s · n)dθ
Id(s · n)dθ =
(cid:90)
(cid:90)
(D4)
s·n>0
s·n<0
where n is the normal to the surface directed inwards.
Using the formula for Id we obtain:
where ri =(cid:82) π/2
(1 + r2)
2(1 − r1)U +
−π/2 R(θ) cosi(θ)dθ/(cid:82) π/2
2
F · n = 0
−π/2 cosi(θ)dθ.
(D5)
Appendix E: General expression for the interaction
factor
We give the expression for q in the presence of re-
flecting boundaries with angle-dependent reflection coef-
ficients Ri (R1 refers to the boundary facing the incident
wave).
Using the same notation as in Section III, we have:
q0(θ) =
(1 − R1)(1 + R2Y )
1 − R1 R2Y 2
ξ(υe sec θ)
(E1)
(1+ R2Y )D0(θ, υe, υtr) =
(α2 + 2β
υtr
2β
υtr
)X0,1 + (α1 + 2β
υtr
(α1 + α2) + 2α1α2
)X0,2
where:
X0 =
(cid:20)γ cos2 θ(1 + R2Y 2)α1 + 2β cos θ(1 − R2Y 2)
(cid:21)
[γ cos2 θ(1 + R2)α2 − 2β cos θ(1 − R2)]Y
(E4)
Superscripts for q(i)
0
is facing the incident wave.
0 and D(i)
(E5)
refer to the boundary that
Appendix F: Asymmetry factor
is [12, 13] µ = µ1, where in general µi =(cid:82)
The asymmetry factor usually used in diffusion models
2π cos(iθ)p(θ)dθ
(where we take p(θ, θ(cid:48)) = p(θ − θ(cid:48))). Since the diffusion
result depends only on υs, υa and µ1, it can be seen as
approximating the differential scattering cross section by:
p(θ − θ(cid:48)) = 1
2π [1 + 2µ1 cos(θ − θ(cid:48))].
The Delta-Eddington approximation [35] allows to in-
corporate the second moment of p by including the for-
ward scattering peak using a "delta function" term so
that: p(θ, θ(cid:48)) = µ2δ(θ−θ(cid:48))+ 1−µ2
2π [1+2µ cos(θ−θ(cid:48))] where
µ = (µ1 − µ2)/(1− µ2). This approximation matches the
Fourier decomposition of p up to the second term. By in-
corporating this expression in RTE (Eq. 3), one recovers
2π [1 + 2µ cos(θ − θ(cid:48))]
a second RTE with p replaced by 1
and σs replaced by σs(1 − µ2). So the diffusion approx-
imation can be made more accurate by replacing µ by
(µ1 − µ2)/(1 − µ2) and σs by σs(1 − µ2). This is known
as the Delta-Eddington approximation [35].
three-dimensional medium,
a
In
4π Pi(cos θ)p(cos θ)dΩ where Pi
polynomial.
=
is the ith Legendre
µi
(cid:82)
deduce:
2ikn
x,1αn,1 =
2ikn
x,1βn,1 =
(cid:88)
(cid:88)
x,1 − km
(cid:2)i(kn
(cid:2)i(kn
m
+i(kn
x,1 + km
x,2)αm,2
x,2)βm,2
x,1 − km
x,2)αm,2
(cid:3)(cid:104)fn,1, fm,2(cid:105)
(cid:3)(cid:104)fn,1, fm,2(cid:105)
m
x,1 + km
+i(kn
x,2)βm,2
11
(G4)
Appendix G: Reflection coefficient with membranes
We consider a plane wave arriving from medium (1),
that is a free-surface ocean with finite depth h, at angle
θ with respect to the x-axis. We suppose that we have a
thin membrane (2) on the water surface extended from
x = 0 to x = w. Change in the dispersion relation leads
to different wavenumbers kn
i verifying:
This allows us to define the transfer matrix as X1 =
M12X2 where Xi = (α0,i, α1,i, ..., β0,i, ...). M21 is subse-
quently defined as M−1
12 .
We finally write the global transfer matrix as M =
M12MpM21, where Mp is a diagonal matrix that propa-
gates the modes along the membrane and that is defined
as:
Mp,(n,n) = eikn
x,2w,
x,2w, 0 ≤ n ≤ N
Mp,(n+N +1,n+N +1) = e−ikn
(G5)
ω2 = gkn
1 h) = gkn
1 h tanh(kn
2 h tanh(kn
2 h)(1 + Cb(kn
2 h)4)
(G1)
where Cb is a bending coefficient of the membrane. k0
i
corresponds to a (real) propagating wave while the other
kn
i correspond to (pure imaginary) evanescent waves.
We first compute the transfer-matrix between medium
(1) and medium (2). We write the velocity potential in
each medium i as:
(cid:104)
N(cid:88)
n=0
(cid:105)
φi =
fn,i(z)
αn,ieikn
x,i + βn,ie−ikn
x,i
eikyy
(G2)
y = (kn
where (kn
i (z +
h) (z = 0 is the water's free surface). Nn,i =
1/
to ensure that
i )2 and fn,i(z) = Nn,i cosh kn
is defined so as
x,i)2 + k2
(cid:113)
(cid:104)fn,i, fn,i(cid:105) = (cid:82) 0
1 + sinh(2kn
i h
−h f 2
i h)
2kn
n,idz = 1. We also note that (fn,1)n
form an orthogonal basis while (fn,2)n are not orthogonal
but still complete (in the limit of N → ∞) [46]. Finally,
for a propagating wave incident from medium (1) with
angle θ, we have ky = k0
1 sin θ.
The boundary condition requires continuity of φ and
∂xφ at x = 0. We write then:
(cid:88)
(cid:88)
n
fn,1(αn,1 + βn,1) =
n
fn,1(αn,1 − βn,1)ikn
x,1 =
(cid:88)
(cid:88)
n
n
fn,2(αn,2 + βn,2)
fn,2(αn,2 − βn,2)ikn
x,2
(G3)
By projecting the previous equations on fn,1, we can
We can now write Xout = M Xin where Xin = (I, R) =
(1, 0, ..., r, β1,1...) and Xout = (T, 0) = (t, α1,1, ..., 0, ....).
By writing M =
, we have:
(cid:20)M1 M2
(cid:21)
M3 M4
T = M1I + M2R,
0 = M3I + M4R
(G6)
which allows us to compute the transmission and reflec-
tion coefficients as:
R = −M−1
4 M3I,
T = M1I + M2R
(G7)
We check of course that t2 + r2 = 1.
Appendix H: Scattering particles embedded in
low-absorbing layer
We consider scattering particles embedded in a layer
of index n and negligible absorption in the presence of
perfect back-reflector (R2 = 1).
In the limit of large
scattering we obtain:
q(θ) = 3 cos2 θ +
2
α1
cos θ
(H1)
where θ is the refraction angle (< θc = asin 1
n ) and α−1
1 =
2(cid:105)
(cid:1) 3
.
n2(cid:104)
1 +(cid:0)1 + 1
(cid:90)
n2
(cid:104)q(cid:105) =
4π
(cid:90) θc
0
n2
π
For isotopic incidence (f = n2
π δ(θ < θc)), we have:
q(θ)f (θ)dΩ = 2π
q(θ) sin θdθ = 4n2
(H2)
In the presence of bulk scattering, the Yablonovitch
limit is indeed maintained for isotropic incidence but can
be overcome at normal incidence.
12
[1] E. Yablonovitch, JOSA 72, 899 (1982).
[2] M. A. Green, Progress in Photovoltaics: Research and
EWTEC (2009).
[25] B. Child and V. Venugopal, Ocean Engineering 37, 1402
Applications 10, 235 (2002).
(2010).
[3] Z. Yu, A. Raman, and S. Fan, Proceedings of the Na-
[26] G. Toki´c and D. K. Yue, Journal of Fluid Mechanics 862,
tional Academy of Sciences 107, 17491 (2010).
34 (2019).
[4] Z. Yu and S. Fan, Applied Physics Letters 98, 011106
[27] T. C. Choy, Effective Medium Theory: Principles and
(2011).
Applications, Vol. 165 (Oxford University Press, 2015).
[5] X. Sheng, S. G. Johnson, J. Michel, and L. C. Kimerling,
[28] D. R. Smith, S. Schultz, P. Markos, and C. M. Soukoulis,
Optics Express 19, A841 (2011).
Physical Review B 65, 195104 (2002).
[6] K. X. Wang, Z. Yu, V. Liu, Y. Cui, and S. Fan, Nano
Letters 12, 1616 (2012).
[29] J. Falnes, Applied Ocean Research 2, 75 (1980).
[30] D. Evans, Annual review of Fluid mechanics 13, 157
[7] D. M. Callahan, J. N. Munday, and H. A. Atwater, Nano
(1981).
Letters 12, 214 (2012).
[8] V. Ganapati, O. D. Miller, and E. Yablonovitch, IEEE
Journal of Photovoltaics 4, 175 (2014).
[31] C. M. Chiang, M. Stiassnie, and D. K. Yue, Theory and
Applications of Ocean Surface Waves (World Scientific
Publishing Co Inc, 2005).
[9] H. Wolgamot, P. Taylor, and R. E. Taylor, Ocean Engi-
[32] K. M. Case and P. F. Zweifel, Linear Transport Theory
neering 47, 65 (2012).
(Addison-Wesley, 1967).
[10] S. Buddhiraju and S. Fan, Physical Review B 96, 035304
(2017).
[11] Z. Yu, A. Raman, and S. Fan, Optics Express 18, A366
(2010).
[33] H. Mitsuyasu, F. Tasai, T. Suhara, S. Mizuno,
M. Ohkusu, T. Honda, and K. Rikiishi, Journal of Phys-
ical Oceanography 5, 750 (1975).
[34] R. Aronson, Nuclear Science and Engineering 44, 449
[12] A. Ishimaru, Wave Propagation and Scattering in Ran-
(1971).
dom Media (Academic press New York, 1978).
[35] J. H. Joseph, W. Wiscombe, and J. Weinman, Journal
[13] L. Tsang, J. A. Kong, and K.-H. Ding, Scattering of
Electromagnetic Waves, Theories and Applications (Wi-
ley, 2000).
[14] G. Tokic, Optimal Configuration of Large Arrays of
Floating Bodies for Ocean Wave Energy Extraction,
Ph.D. thesis, MIT (2016), http://hdl.handle.net/
1721.1/104198.
[15] J. R. Nagel and M. A. Scarpulla, Optics Express 18, A139
(2010).
[16] J.-Y. Wang, F.-J. Tsai, J.-J. Huang, C.-Y. Chen, N. Li,
Y.-W. Kiang, and C. Yang, Optics Express 18, 2682
(2010).
[17] G. Rothenberger, P. Comte, and M. Gratzel, Solar En-
ergy Materials and Solar Cells 58, 321 (1999).
[18] F. E. G´alvez, P. R. Barnes, J. Halme, and H. M´ıguez,
Energy & Environmental Science 7, 689 (2014).
[19] R. Mupparapu, K. Vynck, T. Svensson, M. Burresi, and
D. S. Wiersma, Optics express 23, A1472 (2015).
[20] J. Falnes, Marine Structures 20, 185 (2007).
[21] J. Tollefson, Nature 508, 302 (2014).
[22] V. Stratigaki, P. Troch, T. Stallard, D. Forehand,
and
J. P. Kofoed, M. Folley, M. Benoit, A. Babarit,
J. Kirkegaard, Energies 7, 701 (2014).
[23] I. Penesis, R. Manasseh, J.-R. Nader, S. De Chowd-
hury, A. Fleming, G. Macfarlane, and M. K. Hasan, in
3rd Asian Wave and Tidal Energy Conference (AWTEC
2016), Vol. 1 (2016) pp. 246 -- 253.
[24] J. Cruz, R. Sykes, P. Siddorn, and R. E. Taylor, Proc.
of the Atmospheric Sciences 33, 2452 (1976).
[36] A. D. Kim, JOSA A 28, 1007 (2011).
[37] C. Chen, Z. Du, and L. Pan, AIP Advances 5, 067115
(2015).
[38] U. Tricoli, C. M. Macdonald, A. Da Silva, and V. A.
Markel, JOSA A 35, 356 (2018).
[39] C. Garrett, Journal of Fluid Mechanics 46, 129 (1971).
[40] R. W. Yeung, Applied Ocean Research 3, 119 (1981).
[41] D. Bhatta and M. Rahman, International Journal of En-
gineering Science 41, 931 (2003).
[42] C. L. Bretschneider, Wave Variability and Wave Spectra
for Wind-Generated Gravity Waves, Tech. Rep. 118 (US
Beach Erosion Board, Washington D.C., 1959).
[43] G. I. Marchuk, G. A. Mikhailov, M. Nazareliev, R. A.
Darbinjan, B. A. Kargin, and B. S. Elepov, The Monte
Carlo Methods in Atmospheric Optics, Vol. 12 (Springer,
2013).
[44] C. Zhang, C.-T. Chan, and X. Hu, Scientific Reports 4
(2014).
[45] T. Bobinski, A. Eddi, P. Petitjeans, A. Maurel,
and
V. Pagneux, Applied Physics Letters 107, 014101 (2015).
[46] C. Fox and V. A. Squire, Phil. Trans. R. Soc. Lond. A
347, 185 (1994).
[47] J. Choi,
of
stationary
"The method
phase,"
http://www.math.uchicago.edu/~may/VIGRE/
(2011),
VIGRE2011/REUPapers/Choi.pdf.
|
2002.07058 | 1 | 2002 | 2019-12-16T15:21:10 | Modulation of heterogeneous surface charge and flow pattern in electrically gated converging-diverging nanochannel | [
"physics.app-ph",
"physics.bio-ph",
"physics.chem-ph",
"physics.flu-dyn"
] | The present study aims at utilizing field effect phenomenon to induce heterogeneous surface charge and consequently changing the fluid flow in a solid state nanochannel with converging-diverging periodicity. It is shown that the proposed geometry causes non-uniform radial field adjacent to channel walls which is stronger around the diverging section and weaker next to the converging part of the wall. The later generates heterogeneous surface charge at channel walls depending on the applied gate potential i.e. applying low gate potential enables effective modulation of surface charge with the same polarity of the intrinsic charge at channel walls, while moderate gate potential causes charge inversion in diverging sections of the channel and generates reverse flow and thus results in fluid flow circulation. The potential application of flow circulation for trapping and rejection of particles is also demonstrated. | physics.app-ph | physics | Modulation of heterogeneous surface charge and flow pattern in electrically
gated converging-diverging nanochannel
Movaffaq Kateba,b,⁎, Mohammadreza Kolahdouzc, Morteza Fathipourb,⁎⁎
T
a BIOS lab on chip, MESA+ Institute for Nanotechnology, University of Twente, Enschede, Netherlands
b MEMS & NEMS Lab, School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran
c Nanoelectronic Center of Excellence, School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran
A R T I C L E I N F O
A B S T R A C T
Keywords:
Field effect
Heterogeneous surface charge
Nanochannel
Trapping
The present study aims at utilizing field effect phenomenon to induce heterogeneous surface charge and con-
sequently changing the fluid flow in a solid state nanochannel with converging-diverging periodicity. It is shown
that the proposed geometry causes non-uniform radial field adjacent to channel walls which is stronger around
the diverging section and weaker next to converging part of the wall. The later generates heterogeneous surface
charge at channel walls depending on the applied gate potential i.e. applying low gate potential enables effective
modulation of surface charge with the same polarity of the intrinsic charge at channel walls, while moderate gate
potential causes charge inversion in diverging sections of the channel and generates reverse flow and thus results
in fluid flow circulation. The potential application of flow circulation for trapping and rejection of particles is
also demonstrated.
1. Introduction
Electroosmosis describes how an external electric field can initiate
motion in an electrolytic fluid in the vicinity of immobile charged
surface, and how its corresponding flow rate depends on surface char-
acteristics. Recently electroosmotic flow in micro and nanochannel
with non-uniform wall potential has been shown promising in mixing
application [1]. In practice, the non-uniform surface potential may arise
from surface defects or adsorption of analyte to the walls. However, the
focus of the current study is on the locally altered surface potential to
achieve specific flow patterns and its applications.
Anderson and Idol [2] initially studied electroosmotic flow in mi-
crochannel of periodically varying ζ potential and found out reversion
of electroosmotic flow in different regions generates circulation. They
also determined mean fluid velocity accurately by putting average ζ
potential into Helmholtz-Smoluchowski equation ( = −u
εEζ η/ ). Ajdari
[3,4] showed symmetric altering ζ potential with zero average gives
only a circulation without net motion in axial direction. The later effect
is sometimes considered as drawback due to reduced flowrates [5] but
also has found use in passive mixers [1]. However, the tradeoff between
mixing and transport is not resolved i.e. excellent mixing may lead to
poor transport efficiency and vice versa [6].
The above-mentioned studies were assuming the thin EDL condition
i.e. the ion concentration is uniform and thus the Poisson equation
reduces to Laplace's equation. Ren and Li [7] using non-uniform ζ po-
tential with overlapped EDL produced different types of the velocity
profile in a microchannel. They also considered different ζ potential
profiles along the channel and concluded that the flow rate is only
determined by an average value of ζ potential independent of its dis-
tribution. The latter seems surprising since Helmholtz-Smoluchowski
equation derived for thin EDL assumption. Thus, Fu et al. [8] high-
lighted the assumption of the Boltzmann distribution of charge density
in these studies [1 -- 7]. The Boltzmann model is only applicable at
equilibrium i.e. constant electrochemical potential in the entire channel
without any imposed electric field. Fu et al. [8] studied a step change in
ζ potential by Nernst-Planck equation compared with classical Pois-
son -- Boltzmann model. The former gives a gradual change of EDL
around the ζ potential step while latter results in a sharp change in
concentration which cannot be true due to convective transport of ions.
They assumed zero ζ potential on one side of the channel which means
no EDL, hence their solution could not predict circulation.
A more sophisticated control over ζ potential and fluid flow is of-
fered by field effect i.e. applying a potential to a gate (VG) which is
separated from the fluid by a dielectric channel wall and affects ζ po-
tential through capacitance according to Debye-Hückel theory. The
later called ionic/fluidic field effect transistor (FET) in analogy to
⁎ Correspondence to: M. Kateb, Science Institute, University of Iceland, Dunhaga 3, IS-107 Reykjavik, Iceland.
⁎⁎ Corresponding author.
E-mail addresses: [email protected] (M. Kateb), [email protected] (M. Fathipour).
M. Kateb et al.
semiconductor MOSFET devices. Qian and Bau [9] proposed an ana-
lytical solution to FET structure consist of alternatively charged elec-
trodes. Various flow patterns were obtained by time-wise variation of ζ
potential e.g. overcome above mentioned mixing and transport tradeoff
by sequential mix and transport. However, they only assumed a sharp
variation of ζ potential instead of solving Poisson equation for the field
effect. However, in practice, surface charge and ζ potential changes
gradually between alternatively charged electrodes.
On the other hand, experimental studies are restricted to available
fabrication and characterization methods which are more evident at
nanoscale. For instance, these studies are still limited to straight in-
plane channels, with respect to the substrate, that eventually leads to
low throughput devices [10,11].
In the present study, BOSCH process was used for the preparation of
vertical (out-of-plane) channels in the large area. The sequential nature
of BOSCH process leaves scalloped sidewalls which is usually con-
sidered as a drawback but in the present study the advantage of such a
channel for generating heterogeneous ζ potential is exploited. The be-
havior of the device under different condition studied by simulation
since out-of-plane channels do not allow imaging along the channel.
2. Method
2.1. Fabrication
The fabrication process began with deposition of a 70 nm thick low
stress silicon nitride (SiRN) on a (100) p-silicon wafer. Then several
2 × 2 mm squares patterned in the wafaer's backside using photo-
lithography followed by immersion in a hot phosphoric acid bath to
open window in SiRN. Then the silicon was removed by KOH aniso-
tropic wet etching through the window to make suspended SiRN/Si
which becomes slightly transparent when the Si thickness reaches
~1 μm. The front face SiRN patterned by laser interference lithography
[12] which followed by anisotropic etching in SF6 and C4F8 mixture
using inductively coupled plasma system to serve as hard mask. The
BOSCH process consisted of 1, 1 and 2 s sequence of C4F8, O2 and SF6 at
−40 °C and 1 kW was utilized for drilling silicon through the SiRN
mask. The backside again was dry etched with SF6, O2 and C4F8 mixed
at 1 kW to open pore bottoms and reach desired membrane thickness.
Then an oxide was grown everywhere by dry oxidation at 945 °C to
desired thickness.
2.2. The model
Fig. 1A shows the scanning electron microscope image of a fabri-
cated channel consisting of periodic converging-diverging sections.
Fig. 1B illustrates the geometry used in the simulation with channel
walls and gate considered to be made of SiO2 and Si, respectively.
However, Si\-SiO2 interface obtained in the fabrication process sepa-
rated by SiOx layer i.e. interface charges with specific thickness which
was neglected here. The average diameter (Dave) of the channel as-
sumed to be 130 nm which smoothly changes between 100 and 160 nm
in the converging (Dmin) and diverging (Dmax) parts with a repetition
period of 100 nm. It can be seen the Dmin was sharper in the fabricated
channel. However, the fabricated channel with 200, 300 and 400 nm
present more smooth edge at Dmin and thus sharp edges were avoided in
the model for both gate and oxide. The gate and oxide thickness is
minimized at Dmax and maximized at Dmin in the model.
2.3. Numerical approach
The EDL was modeled by solving mass transport of ions considering
diffusion, ion-ion interaction and fluid convection as well as electro-
migration due to applied fields. These assumptions are of the prime
importance since the contribution of converging-diverging geometry
and field effect is still unknown and may alter charge distribution and
104
Fig. 1. (A) SEM image of the channel cross section obtained in the fabrication and (B) the
model used in the simulation indicating Dmin, Dave and Dmax as well as longitudinal, l, and
radial, r, axis.
EDL. The time dependent form of the Nernst-Planck equation is used to
describe the motion of ionic species inside the channel.
∂
c
i
∂
t
∇ = ∇ ⎡
c
D z e
i
i
k T
B
⎣⎢ ∇ +
D c
i
∇ ⎤
c ϕ
i
⎦⎥
(1)
U
+
.
.
i
i
where ci, Di and zi are concentration, diffusion coefficient and ionic
valence of the i species. U is the velocity vector of the fluid and ϕ is the
electric potential. The e, kB and T respectively stand for elementary
charge, Boltzmann constant and absolute temperature.
n
ϕ
= −
ρ
ch
ε
∇ = −
2
The mean field approximation of the electrostatic potential is de-
scribed by the Poisson equation, which relates the electrical potential to
the charge density.
∑
=
(2)
here ρch and ε are the charge density and electric permittivity with n
being the total number of species in the system. We assumed similar ε
for the bulk electrolyte and EDL and neglected its variation with VG
which is a valid assumption in non-overlapping EDL regime.
z c
i
e
ε
1
i
i
The Navier-Stokes equation is an expression of conservation of
linear momentum for an incompressible Newtonian fluid with constant
mass density. Now we allow for fluid motion due to electrostatic force
by adding a term to the Navier-Stokes to represent the body force
density due to the electrostatic force.
∂
U
∂
t
2
.
η
p
E
+
U
⎛
⎝
ρ
ch
∇ ⎞
U U
⎠
= −∇ + ∇ +
ρ
f
(3)
here U, ρf, η and p respectively are the velocity vector, density, viscosity
and pressure of the fluid. E is the electric field due to charge redis-
tribution around the walls and the electric field applied along the
channel as well as radial field imposed by VG. An electric field of 105 V/
m was generated along the channel by applying proper voltage to inlet
M. Kateb et al.
and outlet to generate electroosmotic flow.
Although no pressure gradient is imposed along the channel, ne-
glecting the pressure, p, term means skipping osmotic pressure effect
which is proportional to the charge density. Especially in the present
configuration, the ρch is not constant throughout the channel and thus it
changes p abruptly next to channel walls. This is an important issue
since it may lead to an artificial flow due to an unbalanced electro-
hydrostatic ion pressure stemming from the Maxwell stress term.
The left-hand side of the Eq. (3) represents the convective transfer of
linear momentum which can be simplified by neglecting unsteady
terms. This is an appropriate assumption unless there is forcing at high
frequency. In addition, the continuity equation for an incompressible
fluid leads to:
∇ =U 0
(4)
The later assumption has been made for mixers of heterogeneous ζ
potential with [9] and without [13] field effect (gate). For the case of
non-uniform ζ potential, it has been demonstrated that the results of
keeping [14] or skipping [8] the inertial term are in good agreement
especially for ion distribution which determines EDL and electro-
osmotic flow.
The fluid considered to be water which is proper solvent for most of
the electrolytes with a relative permittivity of 87.5 at room tempera-
ture. On each side of the channel, there is a reservoir with the constant
concentration of electrolyte equal to 0.01 M KCl.
The particles were assumed to be rigid spheres of 10 (P1) and 20 nm
(P2) in diameters, with the relative permittivity of 50 and 59. The
density of both particles was chosen to be 1050 kg/m3 and their elec-
trical conductivity was assumed to be 1 μS/m with opposite charge. The
EDL formed adjacent to the charged surface of a particle with a typical
thickness ranging from 0.1 to 10 nm, is so thin that will not be resolved
in detail. In the framework of the thin EDL approximation, the particles
and their adjacent EDL were considered as a single entity, with 1 nm
shell thickness and shell relative permittivity of 6 and 4.44 respectively
for P1 and P2. This is a true assumption since the particles are small
enough regarding the Dmin of the channel. For bigger particles, how-
ever, EDL around particles cannot be neglected [15,16].
The coupled system described above is simultaneously solved with a
commercial finite-element package COMSOL (version 5.2) and Matlab
R2016a. In order to validate the present computational method, we
made comparisons with existing analytical and experimental results of
electroosmotic flow in channels with a simple flat geometry.
3. Results and discussion
3.1. Device characteristics
Fig. 2 illustrates the regulation of net charge density, ρch, radial
electric field, Er, and potential, ϕ, at two different part of the channel
along Dmax and Dmin. The figure also contains the result of a cylinderical
channel (Dcylinder) with 100 nm in diameter for comparison. The fluid,
oxide and gate regeions are also labeled and separated by a yellow
highlight in the figure for more clarity. It can be seen that the channel
presents characteristic properties of MOSFET devices. For instance, zero
ρch in the bulk fluid which changes in the EDL followed by a sharp
change to zero in the oxide. However, the thickness variation of the EDL
seems to be negligible and thus we are in non-overlapping EDL regime.
In analogy to a semiconductor MOSFET, which the variation of deple-
tion zone width reduces at higher dopant concentration, here the high
strength of electrolyte reduces EDL thickness variation. It can be seen
that at zero VG, the same charge density was obtained at the wall in
Dmax, Dmin and Dcylinder which will be called intrinsic surface charge
hereafter. The effect of complex morphology is more pronounced at VG
of 0.6 V which present inverse polarity of EDL at Dmax compared to Dmin
and Dcylinder. Further increase in VG, would results in similar polarity at
Dmax, Dmin and Dcylinder. It is also worth noting that the Dcylinder presents
an intermediate charge between values at Dmax and Dmin which in-
dicates the importance of curvature.
Since there is no charge and charge transfer within the oxide, the Er
is expected to be constant which is the case for the cylinder. However,
in the current geometry the field lines starting from the gate are con-
verging toward Dmax and thus the magnitude of Er increases gradually
up to the channel walls. Inversely, the filed lines become diverging
toward Dmin that cusses reduction in the magnitude of Er.
The ϕ equals to VG in the gate which drops drastically across the
oxide but a slight change in ϕ at the channel wall is enough to change
the electroosmotic flow in the channel. Since Er is not constant, the ϕ
drop within the oxide shows deviation from linear behavior obtained in
the cylinder.
3.2. Space charge density
Fig. 3 illustrates the distribution of ρch at different applied VG in the
side view cross section of the channel. At zero VG, accumulation of
cations occurs at the channel walls due to intrinsic surface charge (σ0).
For VG of 0 -- 0.4 V, not shown here, the walls are still surrounded by
cations but the ρch values drops slightly around the channel walls. Ap-
plying 0.5 V to the gate overcomes σ0 at channel walls but only around
Dmax which causes anion accumulation as indicated by deep blue. While
cations still pile up around Dmin which is evident from the red line and
cyan halation. The VG of 0.6 V almost balances the density of anions
and cations at Dmax and Dmin, respectively. This effect enables efficient
mixing by inversion of electroosmotic flow in converging and diverging
part of the channel. Slightly higher VG (0.7 V), unbalances the ρch and
increases anions density around Dmax and this trend continues by fur-
ther increase in VG. It is worth noting that the application of VG cannot
provide uniform anions density along the channel walls since the
strength of the field is different at Dmax and Dmin.
To illustrate net charge distribution more quantitively, the exact
value of ρch for various VG is plotted in Fig. 4. At VG equazero, the ρch is
positive in all cases as result of σ0 those shift toward negative values by
applying positive VG. At VG ~0.6 V the ρch presents reversed values at
Dmax and Dmin.
The point of zero charge (PZC) is obtained at different gate voltages
namley around 0.4, 0.6 and 0.9 V respectively for Dmax, Dave and Dmin.
In other words, for VG between 0.4 and 0.9 V a specific part of the
channel between Dmax and Dmin is at PZC. The latter is practically im-
portant since the maximum solubility of the gate oxide occurs at PZC.
The ionic inversion voltage, compared to the VG = 0, is around 0.8,
1.2 and 1.8 V at Dmax, Dave and Dmin, respectively. The possibility of
ionic inversion obtained here called ambipolar effect. This is in con-
tradiction to the results of Karnik et al. [10] which claimed ambipolar
effect is not accessible using low dielectric constant oxide such as SiO2.
Thus, they asserted that the efficiency of FET device is limited to tuning
the charge with the same polarity of intrinsic charge e.g. cation density
at the channel walls. However, they used a planar gate on one side of
the channel which means one side of the channel experienced applied
field efficiently. Later studies by Lee et al. [11] showed a surrounded
gate structure enables charge conversion at channel walls or so called
ambipolar effect. Then the ambipolarity of fluidic FET is highly de-
pendent on the gate structure and how efficiently it applies the field
into the channel.
3.3. Zeta potential
Variation of the ζ potential with the VG is shown in Fig. 5 which
depicts the characteristic non-linear behavior of ζ potential [17].
However, the deviation is bigger for Dmax and it saturates at higher ζ
potential which can be explained by the difference in capacitance as
follows:
Assuming the EDL and the gate oxide as parallel plates the Gauss'
law gives:
105
M. Kateb et al.
Fig. 2. Radial variation of charge density, electric field and potential at two different part of channel along Dmax and Dmin in comparison with a flat cylinder (Dcylender).
Fig. 3. Charge distribution at different VG of 0 -- 0.9 V in the side view cross section of the
channel.
f
ε E
f
r
−
ox
ε E
ox
r
=
σ
ε
0
(5)
where ox and f superscript denotes Er in the gate oxide and fluid re-
spectively. ε0 is the permittivity of the vacuum, εox and εf are relative
permittivity of the gate oxide and fluid and σ is the surface charge
density on the channel-liquid interface.
Assuming that the potential in the bulk is zero, Eq. (5) can be re-
written as,
C ϕ
d
−
C V
(
ox
G
−
ϕ
)
=
σ
(6)
where Cox = ε0εox/tox is the capacitance per unit area of the oxide layer
having thickness tox, and Cd = dσ/dϕ is the differential capacitance. As
106
Fig. 4. Variation of ρch versus VG at the Dmax, Dave and Dmin adjacent to channel walls.
illustrated by Grahame [18], the Gouy-Chapman- Stern model divides
Cd to the capacitance of the charges held at the outer Helmholtz layer
(CH) and capacitance of the truly diffuse charge (Cdiff). The former is
independent of ϕ, but the later varies in a v-shaped fashion with the ϕ.
The composite capacitance shows a complex behavior and is governed
by the smaller of the two components. At larger electrolyte con-
centrations, or even at large polarizations in dilute media, Cdiff becomes
so large that it no longer contributes to Cd and one sees only the con-
stant capacitance of СH. However, in general, Cdiff is smaller and shows
the major contribution to the total capacitance in series. Here, the EDL
capacitance assumed to be equal to that of the diffused layer.
ζ
=
+
C V
σ
ox G
+
C
C
(
ox
diff
)
(7)
When there is no voltage on the gate:
=
ζ
0
(8)
where ζ0 is the intrinsic ζ potential of the channel wall. Now rewriting
σ
+
C
(
ox
C
diff
)
M. Kateb et al.
Fig. 5. Variation of ζ potential with VG at Dmax, Dave and Dmin.
Eq. (7) gives:
ζ
=
C
ox
+
C
diff
)
C
(
ox
V
G
+
ζ
0
(9)
Eq. (9) seems to predict liner variation for ζ potential but this re-
lation is more complicated since Cdiff would change with ζ potential.
However, this simplified model enables describing the difference in
Fig. 5. The EDL around Dmax experience stronger applied field which
results in dielectrophoretic saturation of counter ions and reduction of
εf. Consequently, capacitance becomes smaller at Dmax and the ζ po-
tential increases.
3.4. Velocity field
Fig. 6 shows the axial velocity, Ul, profile in the cross section of the
channel at Dmax and Dmin for a range of VG (0 -- 2 V). It can be seen that
applying proper VG enables efficient control over the fluid flow at both
Dmax and Dmin. However, in both cases at VG of 0.8 V, and 0.6 V along
Dmax, the Ul is reversed next to the wall with respect to center as result
of ionic inversion. The later generates circulation in both cases and is of
interest for efficient mixing.
Away from moderate VG, the velocity profile shows a deviation from
Fig. 6. Variation of axial velocity profile with VG along Dmax (line) and Dmin (dash-dot
line).
107
Fig. 7. The magnitude of velocity field in side view cross section of the channel at dif-
ferent VG. The arrows and indicating the flow direction.
the ideal plug flow i.e. constant Ul in the bulk with a drop to zero in EDL
within a flat channel. Here the velocity profile reaches a maximum
magnitude in the EDL with a drop toward the center of the channel.
Such a profile has been reported for transient development of the plug
flow in a flat micro-slit [19]. However, in the converging-diverging
channel, the fluid always experiences variation in cross section and thus
stable profile would never become flat in the bulk. It is worth noting
that the velocity drop in the bulk is more pronounced along Dmax. This
is associated with the combination of transient plug flow and continuity
since the latter compensates for the decrease in cross section.
To provide more evidence of circulation, electroosmotic velocity in
the side view cross section of the channel is shown in Fig. 7 at several
VG. In the lack of VG, the maximum velocity occurs in converging sec-
tions due to continuity Eq. (4). Thus, a moving particle experiences
acceleration and deceleration in the red and yellow regions, respec-
tively. At VG of 0.5 V, as result of slight anion pile up around Dmax a
slow current circulation occurs with zero velocity in the middle of the
blue region. The red region is indicating the maximum velocity also
confined close to Dmin where the cations are still available. It is evident
that the velocity is uniform in the center of the channel (r equal to
0 -- 15 nm) and thus there would be no acceleration-deceleration beha-
vior in the center. As discussed before, a strong enough field causes
anion accumulation at Dmax channel walls which eventually generates
considerable reverse flow. This is shown in VG of 0.6 V where reverse
flow forms a vortex in the diverging section with zero velocity in the
vortex center. The vortex act as an obstacle that decreases the channel
cross section and results in the maximum velocity below the vortex in
the center of the channel. This is very interesting since around the
center of the channel, acceleration and deceleration is swapped com-
pared to VG = 0. However, it is doubted whether velocity field in the
rest of the channel guides the particle into the vortex or not. VG of 0.7 V,
generates the maximum velocity around the Dmax by providing a higher
density of anion. It also causes further displacement of the vortex center
toward the center of the channel. The higher VG of 0.8 V holds the
maximum velocity around Dmax while moves the vortex center even
more and develops a very small reverse flow to the left in the blue
regions. Further increase in VG, as shown in VG = 0.9 V subplot, forms a
serpent like reverse flow and extends the vortex center toward the
This is very promising since previous mixers hold complicated struc-
ture, with at least two adjacent gate electrodes, while present structure
eliminates multi-electrodes and suggests simple fabrication techniques.
By introducing particles, the circulation causes different behaviors of
focusing, trapping and rejection at different gate voltages.
Acknowledgment
Authors would like to thank Henk van Wolferen and Johan Bomer
respectively from nanolab and BIOS in mesa + institute who shared
their expertise in nanofabrication. We would like to thank our collea-
gues Amirali Ebadi, Mohsen Hajari and Sedigheh Nikpay in the MEMS
& NEMS lab that greatly assisted the initial simulations.
References
[1] D. Erickson, D. Li, Influence of surface heterogeneity on electrokinetically driven
microfluidic mixing, Langmuir 18 (5) (2002) 1883 -- 1892.
[2] J.L. Anderson, W. Keith Idol, Electroosmosis through pores with nonuniformly
charged walls, Chem. Eng. Commun. 38 (3 -- 6) (1985) 93 -- 106.
[3] A. Ajdari, Electro-osmosis on inhomogeneously charged surfaces, Phys. Rev. Lett.
75 (4) (1995) 755.
[4] A. Ajdari, Generation of transverse fluid currents and forces by an electric field:
electro-osmosis on charge-modulated and undulated surfaces, Phys. Rev. E 53 (5)
(1996) 4996.
[5] A.E. Herr, J.I. Molho, J.G. Santiago, M.G. Mungal, T.W. Kenny, M.G. Garguilo,
Electroosmotic capillary flow with nonuniform zeta potential, Anal. Chem. 72 (5)
(2000) 1053 -- 1057.
[6] F. Tian, B. Li, D.Y. Kwok, Tradeoff between mixing and transport for electroosmotic
flow in heterogeneous microchannels with nonuniform surface potentials, Langmuir
21 (3) (2005) 1126 -- 1131.
[7] L. Ren, D. Li, Electroosmotic flow in heterogeneous microchannels, J. Colloid
Interface Sci. 243 (1) (2001) 255 -- 261.
[8] L.-M. Fu, J.-Y. Lin, R.-J. Yang, Analysis of electroosmotic flow with step change in
zeta potential, J. Colloid Interface Sci. 258 (2) (2003) 266 -- 275.
[9] S. Qian, H.H. Bau, A chaotic electroosmotic stirrer, Anal. Chem. 74 (15) (2002)
3616 -- 3625.
[10] R. Karnik, R. Fan, M. Yue, D. Li, P. Yang, A. Majumdar, Electrostatic control of ions
and molecules in nanofluidic transistors, Nano Lett. 5 (5) (2005) 943 -- 948.
[11] S.-H. Lee, H. Lee, T. Jin, S. Park, B.J. Yoon, G.Y. Sung, K.-B. Kim, S.J. Kim, Sub-
10 nm transparent all-around-gated ambipolar ionic field effect transistor, Nano 7
(3) (2015) 936 -- 946.
[12] H. Wolferen, L. Abelmann, Laser interference lithography, in: T.C. Hennessy (Ed.),
Lithography: Principles, NOVA Publishers, Hauppauge NY, USA, Processes and
Materials, 2011, pp. 133 -- 148.
[13] S. Hong, J.-L. Thiffeault, L.G. Fréchette, V. Modi, Numerical study of mixing in
microchannels with patterned zeta potential surfaces, ASME 2003 International
Mechanical Engineering Congress and Exposition, American Society of Mechanical
Engineers, 2003, pp. 573 -- 577.
[14] S. Bhattacharyya, A.K. Nayak, Electroosmotic flow in micro/nanochannels with
surface potential heterogeneity: an analysis through the Nernst -- Planck model with
convection effect, Colloids Surf. A Physicochem. Eng. Asp. 339 (1 -- 3) (2009)
167 -- 177.
[15] X. Xuan, B. Xu, D. Li, Accelerated particle electrophoretic motion and separation in
converging-diverging microchannels, Anal. Chem. 77 (14) (2005) 4323 -- 4328.
[16] S. Qian, A. Wang, J.K. Afonien, Electrophoretic motion of a spherical particle in a
converging -- diverging nanotube, J. Colloid Interface Sci. 303 (2) (2006) 579 -- 592.
[17] N. Hu, Y. Ai, S. Qian, Field effect control of electrokinetic transport in micro/na-
nofluidics, Sensors Actuators B Chem. 161 (1) (2012) 1150 -- 1167.
[18] D.C. Grahame, The electrical double layer and the theory of electrocapillarity,
Chem. Rev. 41 (3) (1947) 441 -- 501.
[19] C. Yang, C.B. Ng, V. Chan, Transient analysis of electroosmotic flow in a slit mi-
crochannel, J. Colloid Interface Sci. 248 (2) (2002) 524 -- 527.
M. Kateb et al.
Fig. 8. Variation of particle velocity and position along the channel with elapsed time.
center of the channel forming those blue triangles.
3.5. Particle tracing
Fig. 8 compares the velocity and position along the channel for two
particles of opposite charge at different VG. At zero VG, depicted by
hollow symbols, the velocity shows fluctuation between ± 3 mm/s and
reaches a maximum till particles
i.e. when
l = 600 nm in the lower plot. The time for reaching the maximum
velocity is 0.7 and 0.35 ms for P1 and P2, respectively. At VG of 0.8 V
shown by filled symbols, the P1 is rejected which is evident from its
position at channel inlet (l = 0) with its velocity become more unstable
at the longer time. While for P2 the velocity fluctuates at the very be-
ginning and nearly converges to zero which means it is trapped at
l = 150 according to the lower plot.
leave the channel
4. Conclusion
In conclusion, we have introduced a simple and large area technique
for fabrication of ionic FET device on the basis of using nanoscale
patterning and DRIE followed by dry oxidation. The converging-di-
verging geometry was obtained as result of the sequential nature of
DRIE which remains unchanged after oxidation. The field effect control
was achieved through the gate electrode which surrounded the outer
surface of SiO2 channel walls to generate radial electric field and ef-
fectively modulate the surface charge at the channel-liquid interface.
The results show the application of VG generates heterogeneous
surface charge in diverging and converging sections. This field induced
redistribution of ions affects the electrokinetic transport of fluids and
suspended particles in the channel. The intermediate VG causes circu-
lation of the fluid which can be utilized for efficient mixing on its own.
108
|
1807.04471 | 1 | 1807 | 2018-07-12T08:37:08 | Preparation and physical properties of soft magnetic nickel-cobalt nanowires with modulated diameters | [
"physics.app-ph",
"cond-mat.mes-hall"
] | We establish a method to produce cylindrical magnetic nanowires displaying several segments, with a large versatility in terms of segment diameter and length. It is based on electroplating in alumina templates, the latter being prepared by several steps of anodization, wet etching and atomic layer deposition to produce, widen or shrink pores, respectively. We propose an analytical model to analyze the in-plane and out-of-plane magnetization loops of dense assemblies of multisegmented wires. The model considers inter-wires dipolar fields, end-domain curling and predicts the switching field of individual wires with no adjustable parameter. Its ingredients are crucial to extract reliable parameters from the fitting of loops, such as magnetization or the porosity of the array. | physics.app-ph | physics | Preparation and physical properties of soft magnetic
nickel-cobalt nanowires with modulated diameters
Sebastian Bochmann,a Dirk Döhler,a Beatrix Trapp,b Michal Stano,b
Olivier Fruchart,b,c Julien Bachmanna,d,*
a Friedrich-Alexander University of Erlangen-Nürnberg, Inorganic Chemistry,
Egerlandstrasse 1, 91058 Erlangen, Germany
b Univ. Grenoble Alpes, CNRS, Grenoble INP, Inst NEEL,
38000 Grenoble, France
c Univ. Grenoble Alpes, CNRS, CEA, Grenoble INP, INAC-Spintec,
38000 Grenoble, France
d Institute of Chemistry, Saint Petersburg State University, 26 Universitetskii Prospect,
Saint Petersburg, Petergof 198504, Russia
Abstract
We establish a method to produce cylindrical magnetic nanowires displaying several segments,
with a large versatility in terms of segment diameter and length. It is based on electroplating in
alumina templates, the latter being prepared by several steps of anodization, wet etching and atomic
layer deposition to produce, widen or shrink pores, respectively. We propose an analytical model
to analyze the in-plane and out-of-plane magnetization loops of dense assemblies of
multisegmented wires. The model considers inter-wires dipolar fields, end-domain curling and
predicts the switching field of individual wires with no adjustable parameter. Its ingredients are
crucial to extract reliable parameters from the fitting of loops, such as magnetization or the porosity
of the array.
Keywords
Nanowires, modulation, nanomagnetism, galvanic plating, electrodeposition, nickel-cobalt
1
Introduction
The continuously decreasing size of magnetic bits in hard disk drives has reached a dimension
close to the fundamental superparamagnetic limit, which corresponds to the minimal volume of
particles, or "grains", of a certain material needed to maintain a stable magnetic moment.1 In
parallel, magnetic random access memory (MRAM) is shifting from a concept to commercial
devices combining the access speed and compactness of a solid-state device with the non-volatility
of magnetic materials. Still, this remains a two-dimensional technology and as such faces
technological and fundamental limitations. One possibility to overcome this ceiling and increase
the areal storage density beyond 1 Tbit/in² is to turn to a three-dimensional storage platform. This
is addressed by the concept of the racetrack memory by Parkin et al., in which bits are stored as
series of domain walls in dense arrays of tall, 'vertical' wires of magnetic material.2,3 For the sake
of simple and inexpensive implementation, there should be only one read/write cell per wire,
located for example at the wafer surface. This requires that domain walls be set in motion along
the wire, or "track", in order to bring the individual bits outside of the matrix. This could in
principle be achieved by spin transfer torque created by the spin polarization of conduction
electrons.4–7 One major challenge in this respect is the controlled motion of magnetic domain walls
in a digital manner to predefined positions along the wire. Otherwise, after a series of individual
motions in a smooth, unstructured wire, the spread of wall mobility may cause domain walls to
meet, interact, and possibly annihilate each other.8 Creating pinning sites along the wire length is
a strategy to digitize it by defining a finite number of well-identified domain wall positions. These
sites can be experimentally realized by a modulation of the material composition and properties,9
or by a modulation in the wire structure, such as the diameter.8 In planar systems, 'notched' strips
produced by standard lithographic methods have indeed exhibited successful pinning of domain
walls in motion.10 However, a transfer of this concept to a three-dimensional system is yet to be
demonstrated.
Individual tools needed towards this goal are available from the literature. Straights wires are
routinely prepared by electrodeposition of a magnetic metal in an appropriate template, such as
'anodic' alumina.11 The synthesis of anodic alumina templates with diameter modulations has been
demonstrated, as well, using periodic oscillations of the anodization voltage (or current density).
However, those systems feature gradual diameter changes instead of abrupt ones (which is
2
deleterious for domain wall pinning), and limited freedom for separating modulations along the
pore length.12–15 A similar system of 'anodic' pore arrays featuring a small number of well-defined,
abrupt changes has been obtained by the combination of successive "mild" and "hard"
anodization.16 The limitation of this system is that the diameters of the individual segments cannot
be tuned at will, since the interpore distance of mild and hard anodization must be adjusted to each
other to avoid instabilities.
In this paper, we present a preparative method for generating parallel arrays of magnetic nanowires
featuring abrupt changes in diameter at well-defined positions along their length and arbitrarily
defined values of the diameters. This is achieved by the combination of several steps involving the
anodization of aluminum (in mild conditions), isotropic pore widening, and atomic layer deposition
(ALD). Subsequently, we perform electrodeposition of the modulated magnetic nanowires. The
magnetization hysteresis of our samples, measured as ensembles of macroscopic size, displays a
complex behavior related to the distribution of wire segment lengths, diameters, and dipolar
interactions. We propose a model able to reproduce both the hysteretic and reversible parts of these
loops, from which accurate geometric and magnetic material parameters can be extracted.
Results and Discussion
1. Preparation of modulated wires
The anodization of aluminum enables the experimentalist to generate parallel arrays of straight
pores perpendicular to the surface, with well-defined diameters that can be set to values between
20 nm and 400 nm and pore lengths between 0.5 µm and 100 µm.17 Anodized aluminum has been
abundantly used as a template for further elaboration of nanowires and nanotubes by galvanic and
other methods.18 While the ratio of pore diameter versus pitch cannot be varied much through
anodization, the pore diameter can be adjusted after growth by an isotropic wet chemical etching
step.18,19 Conversely, atomic layer deposition (ALD) can be used to reduce the diameter of the
template's pores after growth, or confer them with specific chemical or physical properties.20–24
Atomic layer deposition is a thin film deposition technique from the gas phase, which is uniquely
suited to coating three-dimensional structures, including deep pores with high aspect ratio.25,26
3
Figure 1. Sketch of the preparative procedure devised.
We combine these methods to generate modulated magnetic wires as sketched in Figure 1. After
the first, sacrificial anodization step (a-b) used to generate the order, a first segment grown in 1%
phosphoric acid electrolyte for 4 h exhibits a pore diameter of 150 nm and length of approximately
10 µm (c) according to standard procedures.19 The pore is then submitted to isotropic widening in
4
10% phosphoric acid (d) to reach a diameter chosen here to be 400 nm, for example. A second
anodization is performed under the same conditions (e), yielding a second segment with the natural
diameter of 150 nm. The resulting pore structure featuring two distinct diameters is then coated
with 5 nm SiO2 via ALD (f).27 If this coating is sufficiently thin, a third anodization can be
performed to grow a third segment (g).28 After removing the remaining metallic aluminum
substrate and the Al2O3 barrier layer at the lower pore extremity (h), the diameter of the third
segment can finally be increased by an isotropic etching step, for instance to match the first segment
at 400 nm, whereas the first and second segments are protected by the chemically inert nature of
the SiO2 layer.
Figure 2. Scanning electron micrograph of a modulated template obtained with the
method presented in Figure 1, with segment diameters of 400 nm, 150 nm and 400 nm.
The false-color image is obtained from a combination of secondary and backscattered
electron signals.
Figure 2 shows the template structure obtained by this procedure. The pores are straight and
parallel, the segments have a homogeneous length, and the transitions between distinct diameter
5
values are abrupt. Note that our procedure is very flexible, in that each geometric parameter of the
modulated structure (that is, each segment length and diameter) can be adjusted individually and
independently of the others. For example, the diameter of the third segment can be either increased
by pore widening, or reduced by Al2O3 ALD. The length of each segment can be adjusted by the
anodization duration. If anodization is performed in 0.3 M oxalic acid (instead of 1% phosphoric
acid), the initial pore diameter is 40 nm (instead of 150 nm) and can be widened to 90 nm (instead
of 400 nm).11
Figure 3. Scanning electron micrograph of a bundle of modulated NiCo wires ("nano-
q-tips") isolated from the template, in false colors (combination of secondary and
backscattered electron signals). NiCo appears pink, the polished Cu substrate dark
green, and short Au stubs from the electrical contact neon green.
In the next step, an electrical contact is defined on one side of the alumina template by sputter-
coating (i). We deposit 6 nm aluminum metal (as an adhesion layer) followed by 50 nm gold. This
thin layer is then augmented by galvanic plating, performed in a two-electrode setup under –2.50 V
in a K[Au(CN)2] electrolyte. The thick galvanic layer closes the pore extremities completely, and
results in a short (1 µm long) Au stub inside the pores.9,29 The sample is then exposed with the
open pore extremity to an electrolytic solution containing cobalt and nickel. Soft magnetic Ni60Co40
alloy nanowires are then grown (k) in the modulated pores from the gold contact in a three-
electrode configuration under –1.10 V vs. Ag/AgCl.30 The resulting wires may optionally be
removed from the template with chromic acid (l). Figure 3 highlights how the diameter
modulations of the template are reproduced in the galvanic wires with high fidelity. The
6
characteristic shape suggests the nickname 'nano-q-tips' for such modulated wires. Backscattered
electron contrast allows for differentiation between the long, modulated Ni60Co40 wires (pink) and
a few broken off Au stubs (neon green).
As mentioned above, the diameter of each segment can be adjusted very precisely. Figure 3
displays a rather extreme diameter "contrast" of 400/150/400 nm, but shorter durations of the
isotropic pore widening steps allow one to generate more modest contrasts, for example the
200/150/200 nm sample presented in Figure 4a. Alternatively, performing the anodization steps in
oxalic acid electrolyte instead of phosphoric acid enable one to work with smaller diameters.
Figure 4b displays an excerpt of one such wire of the type 70/40/70 nm. This range of diameters
is of interest in the magnetic realm since it is on the order of ten times typical magnetic exchange
lengths, so that several topologies of domain walls may be expected.31
Figure 4. (a) Scanning electron micrograph of an isolated NiCo "nano-q-tip" with
modest diameter contrast between segments of 200 nm, 150 nm, and 200 nm.
(b) Transmission electron micrograph of a "nano-q-tip" isolated from pores grown in
oxalic acid instead of phosphoric acid: The diameters are 70 nm, 40 nm, and 70 nm.
2. Geometry of the diameter modulations
The diameter reduction between first and second segment and the diameter increase between
second and third segment result in distinct pore profiles (Figure 5). This is due to the distinct
experimental procedures for diameter decrease and increase, as sketched in Figure 5c and 5d. At
7
the diameter reduction, the hemispherical extremity of a wide pore concentrates the electric field
at the start of the subsequent anodization, so that the second segment continues straight from its
lower extremity. This second segment immediately recovers its natural, narrower diameter, but
does not affect the gradual, approximately hemispherical profile left at the end of the first segment.
In contrast to this, the diameter increase occurs abruptly given that the hemispherical end of the
narrow pore is lost completely upon growth of the next segment. The isotropic pore widening
performed as the last preparative step finds an abrupt etch stop defined by the SiO2 layer.
Figure 5. Transitions between two different pore diameters: Scanning electron
micrographs highlighting the geometric difference between the gradual, conical
diameter reduction from 200 nm to 150 nm (a) and the abrupt diameter increase from
150 nm to 200 nm (b). The corresponding anodization procedure for diameter decrease
(c) and increase (d) is sketched.
Wires grown inside the modulated template exhibit the same geometric properties as the template
(Figure 6). They exhibit diameter transitions with two distinct geometries: a smoother one obtained
upon diameter reduction of the template pores stands in stark contrast to the abrupt one yielded by
pores submitted to a diameter increase.
8
Figure 6. Transitions between two different wire diameters: (a) Gradual, conical
diameter reduction; (b) Abrupt diameter increase.
3. Experimental magnetic properties of modulated wires
Magnetic nanowires display highly anisotropic magnetic properties, as the magnetization of soft
magnetic materials tends to align along the long dimensions of an object.32 When homogeneously
magnetized wires are submitted to an external field oriented antiparallel to the magnetization, its
reversal implies the formation of a magnetic domain wall, which nucleates at one extremity and
then travels to the other. Its energy of formation, and thereby the reversal field measured
experimentally, is strongly dependent on the wire diameter.18 More important in our case, the
dipolar interactions are expected to be different within the three parts, leading to a different slanting
of the loops, which is the dominant feature in such dense arrays.33 Thus, our modulated wires may
be expected to reverse their magnetization in steps for individual segments, and to exhibit a stable
domain boundary in a range of external fields bounded by the two distinct values of switching
fields for each diameter.
Figure 7 presents magnetic hysteresis loops recorded on a sample consisting of an ordered array
of modulated NiCo nanowires embedded in their alumina matrix. The lengths of the thick, narrow
9
and thick segments are 10 µm, 13 µm, and 10 µm, respectively. The comparison of curves recorded
with the wires oriented parallel and perpendicular to the applied field highlights the anisotropy of
the sample. More interestingly, the loop exhibits kinks in the parallel configuration, which define
a segment of curve with large slope (near zero field) distinct from the rest of the reversal, happening
with a lower slope over a broad field window between approximately ‒40 kA/m and +40 kA/m.
Figure 7. Magnetic (SQUID) hysteresis loops recorded on a sample of NiCo wires with
diameters 400 nm (10 µm) / 150 nm (13 µm) /400 nm (10 µm) at room temperature
with field applied parallel to the wires' long axis (top) perpendicular to it (bottom). The
three parts with different slopes in the former curve are attributed to the distinct
switching behavior of segments of different diameters.
10
With this in mind, we may attribute the magnetization change occurring around zero field to the
magnetic reversal of the central, narrow segments, which are subject to a lower internal field,
whereas the thicker extremities reverse more gradually. This interpretation is confirmed
experimentally when three sample are compared which differ from each other solely by the length
of the central segment (Figure 8). When this segment is varied from 3 µm to 9 µm and 13 µm, the
steep section of hysteresis becomes more and more prevalent. This corresponds to the increase in
the total magnetic moment of this segment associated with the increase in its length.
Figure 8. SQUID magnetic hysteresis
loops presented as first derivatives
(susceptibility) of three samples of modulated NiCo wires, with the field applied
parallel to their long axis. The red, green and blue curves correspond to central (thin)
segment lengths of 3 µm (green), 9 µm (red) and 13 µm (blue), respectively.
The distinct switching fields of the three segments can be utilized to trap magnetic domain
walls at or near diameter modulations. Figure 9 displays the morphology and magnetic stray
fields of an isolated nanowire, recorded by atomic force and magnetic force microscopies,
respectively. The wire was prepared by an oscillatory quasistatic demagnetization procedure
performed from 1 T, with magnetic field applied out-of-plane (oop), i.e. perpendicular to the
wire axis. The MFM micrograph shows the presence of contrast (stray field) at the modulations,
as expected because the change of wire area implies that the excess induction gives rise to stray
field. More interestingly, however, it displays a feature located near one modulation, which we
can confidently attribute to a domain wall given that the two wire extremities have the same
contrast (possible only if the wire consists of two domains of opposite direction). In most wires
11
we evidence one or more such domain walls. This shows the effectiveness of the trimodulated
wires to confine domain walls in a well-defined area, suitable for further investigations.
Figure 9. Demonstration of a magnetic domain wall pinned near a modulation. A wire
isolated from the matrix displays three 10-µm long segments of distinct diameters (40
/ 30 / 40 nm) in atomic force microscopy (AFM, top). The corresponding magnetic
force microscopy (MFM) image (lower panel) shows the presence of magnetic stray
fields at the wire extremities and diameter modulations, as expected, but also at a point
situated approximately 1.5 µm from one modulation. That this stray field is caused by
the presence of a domain boundary is proven by the head-to-head magnetization
expressed by the other contrast points.
4. Modeling magnetic properties of modulated wires
a. Arrays of wires
A global hysteresis loop is the basic characterization for a magnetic material. Even though one may
in the end use local properties of single objects, a global hysteresis loop may inform about intrinsic
properties such as material magnetization and sample geometry, having an impact on dipolar
interactions. For instance, for a soft magnetic material, magnetization may be derived if the
geometry of the system is known, through demagnetizing coefficients. In practice, this is more
reliably achieved along hard-axis loops, which do not depend on hysteretic effects. Figure 10
shows hysteresis loops of macroscopic arrays of trisegmented wires with two geometries: the
central segment of diameter 120 nm and length either 3 µm or 13 µm, embedded between two
segments of diameter 420 nm and length 6 µm each, all this with pitch 490 nm. We name in-
plane (ip) the situation where the magnetic field is applied parallel to the membrane (i.e. across the
axis of the wires), and out-of-plane (oop) the situation where the magnetic field is applied
perpendicular to the membrane (i.e. along the axis of the wires). The pinpoint dipole approximation
12
can obviously not be used to describe long wires in a dense array, as proposed in the early days of
nanomagnetism.34 Yet, there exist simple models to fit hysteresis loops in the case of straight
nanowires considered uniformly magnetized, as first proposed by Wang et al33, and later
refined.19,20 In short, the ip loops may be fitted with high confidence as no hysteretic event is
involved, so that the analytical and exact coherent rotation model is perfectly relevant. From the
fitting of the loop, 𝑀s may be extracted if 𝑝 is known, or the reverse. Using as fixed input the
geometric parameters determined accurately by SEM, the resulting magnetization from the best fit
is 862 kA/m (1.08 T in induction units). This is in perfect agreement with CoNi alloys of the present
composition, CoNi being suitable for a linear interpolation on the Slater-Pauling curve.35
(a)
(b)
Figure 10. Hysteresis loops of samples with thick / narrow / thick segments of
(a) 6 µm / 3 µm / 6 µm and (b) 6 µm / 13 µm / 6 µm, recorded along the in-
plane (blue) and out-of-plane (red) directions. Experiments are shown with open
symbols. The dotted line stands for the simple model based on uniformly magnetized
wires, while the continuous line represents the curling model. The modelling
parameters are: magnetization 862 kA/m; pitch 490 nm; narrow segment with
diameter 120 nm and switching field 20 mT; broad segment with diameter 410 mT
and switching field 5 mT.
The case of oop loops is more complex to tackle, and cannot be done so exactly, as hysteresis is
involved. The wires are assumed to switch through nucleation at a wire end, and fast propagation
of a domain wall. Each wire is ascribed the same individual switching field 𝐻c, however the global
loop is slanted as inter-wire interactions induce a dipolar contribution to the magnetic field felt by
each wire. The dipolar field can be computed by considering the top and bottom charges on both
13
sides of the membrane.33 The saturation field is 𝐻c + 𝐻p. 𝐻p = γ𝑝𝑀s, with γ ≳
1
is a
2
phenomenological parameters comprised between 0.5 and 1, taking into account nucleation
effects (see Annex). For the ip direction, coherent and reversible rotation of magnetization occurs,
coercivity and remanence are zero, and saturation field is 𝑀s [(1 − 𝑝)/2].
Two complications arise here. First, we need to consider multiple segments. Second, the larger
diameter of the outer segments, which gives rise to large intra-wire dipolar fields, threatens the
validity of the hypothesis of uniform magnetization, and may possibly be responsible for the slow
saturation of the loops (Figure 10), as previously reported especially for rather short wires.36 No
analytical model has been proposed to describe either of these two aspects. We will now address
them in two steps and obtain a model describing such complex arrays with a remarkable accuracy.
Let us first discuss a way to deal with the diameter modulations. In our case, the outer segments
have a larger diameter, thus we expect nucleation to occur at rather low field.37 Second, the
modulations are sharp and very significant in diameter, so that we expect pinning of domain walls
before propagation into the central low-diameter segment.38 Thus, in practice it is as if the central
segment switches on its own under the effect of external plus all internal magnetostatic fields, albeit
with a switching field smaller than an isolated wire due to the facilitated nucleation through
injection of a domain wall into the wire (Figure 11a). Thus, the behavior of the array of
trimodulated wires should be similar to the one of three independent arrays described by the model
mentioned above for straight wires, each characterized by a distinct demagnetizing field. Indeed, a
uniformly magnetized slab with infinite lateral dimensions (one of the three arrays, considered in
the mean field as a medium with magnetization 𝑝𝑀s) gives rise to no stray field outside the slab,
whatever the direction of magnetization (Figure 11b). So, it is as if this slab was independent. The
resulting equations for hysteresis loops are provided in Annex. The dotted lines in Figure 10 show
the outcome of this first model compared with the present experimental data. The geometric
parameters (length and diameter of the various segments) are fixed in the model, as determined
with scanning electron microscopy. Magnetization is also fixed to the value 862 kA/m determined
from the ip loops. Only the two coercive fields are let free to best adjust the experiments. The
general shape of the loop is reproduced, however the match is not perfect: the slope is slightly
lower in the model than in the experiments. This residual imperfection will be lifted by the second
part of the model, described below.
14
(a)
(b)
(c)
Figure 11. Schematic illustration of processes implemented in the modeling.
(a) Nucleation at the end of the thick segments, followed by injection of a domain
wall into the narrow segments. (b) Magnetic charges arising from each layer,
responsible for a demagnetizing field in this sole layer. (c) Curling domain ends in
thick segments.
In this second stage, we consider deviations from uniform magnetization. Non-uniform
distributions of magnetization arise at wire ends, since they spread the magnetic charges and thus
decrease magnetostatic energy. Above a diameter of typically seven times the dipolar exchange
length (40 to 50 nm in our case), these deviations take the form of curling around the wire axis, in
the case of a perfect geometry.39,40 These so-called domain ends may be viewed as half a domain
wall (Figure 11). In a previous publication,41 we proposed an analytical model to predict the width
of domain walls in cylindrical nanowires, based on the balance of exchange and dipolar energies.
Although we estimated these energies in a handwaving fashion, the model was successful, e.g.
reproducing the variation of wall width with the square of the wire diameter beyond typically
100 nm found via numerical simulation. Considering domain ends as half a domain wall, here we
extend this model to include the Zeeman energy from the magnetic field, be it external or
dipolar (see Annex). This field tends to compress the domain end when it is parallel to the
magnetization in the body of the wire, and to stretch it when antiparallel. Thus, the curling model
provides three features. First, it rounds off the hysteresis loop edges because magnetization in the
15
domain ends is no longer strictly parallel to the applied field. Second, it brings an extra
susceptibility contribution to hysteresis loops, through the compression/stretching of the domain
ends. Third, it predicts the nucleation field, defined for the divergence of the width of the domain
end for a given value of the applied field.
(a)
(b)
Figure 12. (a) Hysteresis loops predicted by the analytical curling model for various
diameters, with no adjustable parameter. The fixed parameters are: magnetization
862 kA/m, exchange stiffness 1.5 ∙ 10−11 J/m, wire length 1 µm. (b) Hysteresis
loops of trisegmented wires with diameters 200 nm / 150 nm / 200 nm, segment
length 10 µm each, and pitch 410 nm, both ip (blue) and oop (red). Symbols show
experiments, fitted with a line by the curling model with magnetization as the sole
adjustable parameter, yielding 822 kA/m.
Thus, the model predicts the hysteresis loop of any wire with given length and diameter, both in
rounding and switching value, in the absence of any adjustable parameter (Figure 12a). Note that
curling domain ends should have no impact on ip loops, as the saturation field for a single curling
wire or uniformly magnetized wire is 𝑀s/ 2, in both cases. Curves for the refined model are shown
as continuous lines on Figure 11. The agreement with experiments is improved, despite having
now no adjustable parameter. Note the rounding at large field and the increased susceptibility at
low field, resulting from the compression and the stretching of domain ends, respectively. Thus,
while more refined techniques can be applied such as first-order reversal curves (FORC) and the
Preisach model,42 our model shows that the accurate analysis of a single hysteresis loop remains
important to provide a quick feedback on material parameters.Figure 12b shows hysteresis loops
for a sample with very different geometry, namely tri-segmented with diameters 200 nm, 150 nm,
16
and 200 nm, and segment length 10 µm for all three parts. The model perfectly fits these loops with
again no adjustable parameter except for a refinement of magnetization. This demonstrates that it
is a robust model, suitable for very different geometries.
Figure 13. Magnetic force microscopy of a single Co40Ni60 wire with diameter 200 nm,
freed from the membrane and supported on a Si surface. (a) Domain ends versus
longitudinal applied field. The height of all images is 1000 nm. The opposite contrast
on either side of the wire arises from the tilt of the tip axis and oscillation direction,
with respect to the normal to the supporting surface43, (b) Length of end domains as
determined from the MFM images as full-width at half-maximum. (c) Scaling of
experimental data shown in b, following Eq.(3).
b. Single wires
Here, we report on the magnetic microscopy of single wires, to support experimentally the claims
of the curling model introduced previously. We imaged the length L of curling domain ends,
17
identified from the distribution of charges along the wire axis, as a function of the longitudinal
applied field favoring magnetization in the wire (Figure 13a-b). As expected, L decreases with
increasing applied field. Plotting the left-hand side 1/L2 scaling law of Eq.(3), shows a linear
variation with the applied field (Figure 13c), although the intercept with the y axis is slightly
below 1. This supports the validity of the curling model.
Conclusions
The results presented in this paper establish a preparative strategy for the generation of smooth,
cylindrical magnetic nanowires featuring well-defined modulations of diameter along their length,
with for instance two segments of broad diameter encompassing one segment with narrower
diameter. This design allows magnetic domain walls to be kept in the system upon quasistatic
demagnetization with an oscillatory magnetic field applied across the axis of the wire. This
provides a model system for studying domain walls, such as their inner structure.44 Suitably
designed modulations should provide energy barriers confining the domain wall in the central
segment, which should prove useful for the investigation of domain wall mobility under magnetic
field or spin-polarized conduction electrons current, without annihilating the domain wall at the
end of the wire. Besides, we provide a simple yet robust analytical model fitting the in-plane and
out-of-plane hysteresis loops of modulated arrays, taking into account demagnetizing field and
curling effects and with no adjustable parameters. The model can be used reliably to extract
geometric and/or material parameters, such as magnetization.
18
ANNEX
We gather here technical aspects related to the model for hysteresis loops, and the means for its
implementation.
a. Modelling hysteresis loops for arrays of straight wires
Along the oop direction, the saturation field related to inter-wire dipolar interactions is the oop
demagnetizing field for a saturated sample: 𝑝𝑀s with 𝑝 =
π
2√3
2
(
d
)
D
the filling factor of the
membrane, which is easily derived from considering the magnetic charges on both sides of the
membrane, in a mean field approach.33 Yet, the maximum interaction field relevant for hysteresis
loops can be modeled as 𝐻p = γ𝑝𝑀s, with γ ≳
1
2
a phenomenological parameter taking into account
nucleation effects, where mostly the charges from the membrane side opposite to that where
nucleation occurs are relevant, due to a projection effect of the charges on the same side.20 The in-
plane direction is characterized by the gradual rotation of magnetization from wire axis to ip. The
resulting loop is derived from the balance between the intra-wire demagnetizing field
−(𝑀s/2) sin𝜃, and the inter-wire field 𝑝(𝑀𝑠/2) sin 𝜃, considering the Lorentz cavity around each
wire, with 𝜃 the angle between magnetization and wire axis. Thus, the ip saturation field of the
loop is 𝑀s [(1 − 𝑝)/2].
b. Modelling hysteresis loops with curling
Here we extend an earlier model,41 developed to derive a scaling law for the domain wall width in
cylindrical nanowires, to the present case of end domains in the presence of an axial external field.
In short, we consider a progressive curling state over a distance L (Figure 11c), and evaluate in a
quite crude fashion exchange, Zeeman and magnetostatic energy. Although the validity of the
estimation of the latter term is rather inaccurate for end domains much longer than the wire
diameter, this approximation allows a first insight into the phenomenon at play. One can obtain a
total energy such as:
𝐸t ≈ 10𝐿𝐴 +
𝜖
3
𝐿𝑑2𝜇0𝑀s𝐻 + 𝑘2𝐾d
𝑑4
𝐿
ln (
𝐿
𝑑
)
Eq(1)
19
The first term is exchange energy, the second one is Zeeman energy, and the third one is dipolar
energy. k is a phenomenological scaling constant introduced to compensate the crude
approximation for the dipolar energy, whose value is determined by fitting the domain wall width
determined by simulation: 𝑘 ≈ 0.18. The length of the curling domain end is found by minimizing
the above against L. Thanks to the slow variation of the logarithm function, one has:
𝐿 ≈
𝑘
√10
𝑑2
Δd
1
√1+
𝜖
30
𝑑
(
Δ𝑍
2
)
Eq(2)
Δd = √2𝐴/(𝜇0𝑀s
2) is the dipolar exchange length, and ΔZ = √𝐴/(𝜇0𝑀s𝐻) is the field-dependent
Zeeman exchange length. The above may also be written in the form of a scaling law versus a
linear variation of applied field:
𝑘2𝑑4
10𝐿2Δd
2 ≈ 1 +
𝜖𝜇0𝑀s𝐻𝑑2
30𝐴
Eq(3)
The linearity of this scaling law is checked from the value of L determined experimentally.
c. Implementation
We use slanted functions to adjust the hysteresis loops, reflecting interaction fields. We define
𝑚sl(𝐻, 𝐻0) as the symmetric slanted non-hysteretic function saturating at 𝐻0, i.e., a linear function.
For an array of single segments, considered along the oop direction in a mean-field approach, the
loop with rising field reads 𝑚sl(𝐻 − 𝐻c, γ𝑝𝑀s), while the one for decreasing field reads 𝑚sl(𝐻 +
𝐻c, γ𝑝𝑀s). These functions, combined with the proper interaction fields discussed in the main text,
are implemented in Wavemetrics IGOR procedures.
20
Acknowledgements
The research leading to these results has received funding from the European Community's Seventh
Framework Program under Grant No. 309589 (M3d).
References
(1) Piramanayagam, S. N.; Srinivasan, K. Recording media research for future hard disk drives.
Journal of Magnetism and Magnetic Materials 2009, 321, 485–494.
(2) Parkin, S. S. P.; Hayashi, M.; Thomas, L. Magnetic domain-wall racetrack memory. Science
(New York, N.Y.) 2008, 320, 190–194.
(3) Parkin, S.; Yang, S.-H. Memory on the racetrack. Nature nanotechnology 2015, 10, 195–198.
(4) Petit, D.?e.; Jausovec, A.-V.; Read, D.; Cowburn, R. P. Domain wall pinning and potential
landscapes created by constrictions and protrusions in ferromagnetic nanowires. Journal of Applied
Physics 2008, 103, 114307.
(5) Lewis, E. R.; Petit, D.; Thevenard, L.; Jausovec, A. V.; O'Brien, L.; Read, D. E.; Cowburn, R.
P. Magnetic domain wall pinning by a curved conduit. Appl. Phys. Lett. 2009, 95, 152505.
(6) Grollier, J.; Boulenc, P.; Cros, V.; Hamzi, A.; Vaur?s, A.; Fert, A.; Faini, G. Switching a spin
valve back and forth by current-induced domain wall motion. Appl. Phys. Lett. 2003, 83, 509–511.
(7) Vernier, N.; Allwood, D. A.; Atkinson, D.; Cooke, M. D.; Cowburn, R. P. Domain wall
propagation in magnetic nanowires by spin-polarized current injection. Europhys. Lett. 2004, 65,
526–532.
(8) Da Col, S.; Jamet, S.; Stano, M.; Trapp, B.; Le Denmat, S.; Cagnon, L.; Toussaint, J. C.;
Fruchart, O. Nucleation, imaging, and motion of magnetic domain walls in cylindrical nanowires.
Appl. Phys. Lett. 2016, 109, 62406.
(9) Bochmann, S.; Fernandez-Pacheco, A.; Mačković, M.; Neff, A.; Siefermann, K. R.; Spiecker,
E.; Cowburn, R. P.; Bachmann, J. Systematic tuning of segmented magnetic nanowires into three-
dimensional arrays of 'bits'. RSC Adv. 2017, 7, 37627–37635.
21
(10) Hayashi, M.; Thomas, L.; Rettner, C.; Moriya, R.; Jiang, X.; Parkin, S. S. P. Dependence of
current and field driven depinning of domain walls on their structure and chirality in permalloy
nanowires. Physical review letters 2006, 97, 207205.
(11) Nielsch, K.; Choi, J.; Schwirn, K.; Wehrspohn, R. B.; G?sele, U. Self-ordering Regimes of
Porous Alumina: ? The 10 Porosity Rule. Nano Lett. 2002, 2, 677–680.
(12) Losic, D.; Lillo, M. Porous Alumina with Shaped Pore Geometries and Complex Pore
Architectures Fabricated by Cyclic Anodization. Small 2009, 5, 1392–1397.
(13) Santos, A.; Forment?n, P.; Pallar?s, J.; Ferr?-Borrull, J.; Marsal, L.?s. F. Structural
engineering of nanoporous anodic alumina funnels with high aspect ratio. Journal of
Electroanalytical Chemistry 2011, 655, 73–78.
(14) Wang, B.; Fei, G. T.; Wang, M.; Kong, M. G.; Zhang, L. D. Preparation of photonic crystals
made of air pores in anodic alumina. Nanotechnology 2007, 18, 365601.
(15) Krishnan, R.; Thompson, C.?V. Monodomain High-Aspect-Ratio 2D and 3D Ordered Porous
Alumina Structures with Independently Controlled Pore Spacing and Diameter. Adv. Mater. 2007,
19, 988–992.
(16) Pitzschel, K.; Bachmann, J.; Martens, S.; Montero-Moreno, J. M.; Kimling, J.; Meier, G.;
Escrig, J.; Nielsch, K.; Görlitz, D. Magnetic reversal of cylindrical nickel nanowires with
modulated diameters. Journal of Applied Physics 2011, 109, 33907.
(17) Masuda, H.; Fukuda, K. Ordered metal nanohole arrays made by a two-step replication of
honeycomb structures of anodic alumina. Science (New York, N.Y.) 1995, 268, 1466–1468.
(18) Nielsch, K.; Wehrspohn, R. B.; Barthel, J.; Kirschner, J.; G?sele, U.; Fischer, S. F.;
Kronm?ller, H. Hexagonally ordered 100 nm period nickel nanowire arrays. Appl. Phys. Lett. 2001,
79, 1360–1362.
(19) Gemmer, J.; Hinrichsen, Y.; Abel, A.; Bachmann, J. Systematic catalytic current enhancement
for the oxidation of water at nanostructured iron(III) oxide electrodes. Journal of Catalysis 2012,
290, 220–224.
22
(20) Da Col, S.; Darques, M.; Fruchart, O.; Cagnon, L. Reduction of magnetostatic interactions in
self-organized arrays of nickel nanowires using atomic layer deposition. Appl. Phys. Lett. 2011,
98, 112501.
(21) Haschke, S.; Wu, Y.; Bashouti, M.; Christiansen, S.; Bachmann, J. Engineering Nanoporous
Iron(III) Oxide into an Effective Water Oxidation Electrode. ChemCatChem 2015, 7, 2455–2459.
(22) Wu, Y.; Assaud, L.; Kryschi, C.; Capon, B.; Detavernier, C.; Santinacci, L.; Bachmann, J.
Antimony sulfide as a light absorber in highly ordered, coaxial nanocylindrical arrays: Preparation
and integration into a photovoltaic device. J. Mater. Chem. A 2015, 3, 5971–5981.
(23) Assaud, L.; Monyoncho, E.; Pitzschel, K.; Allagui, A.; Petit, M.; Hanbücken, M.; Baranova,
E. A.; Santinacci, L. 3D-nanoarchitectured Pd/Ni catalysts prepared by atomic layer deposition for
the electrooxidation of formic acid. Beilstein journal of nanotechnology 2014, 5, 162–172.
(24) Katz, M. J.; Riha, S. C.; Jeong, N. C.; Martinson, A. B.F.; Farha, O. K.; Hupp, J. T. Toward
solar fuels: Water splitting with sunlight and ?rust?? Coordination Chemistry Reviews 2012, 256,
2521–2529.
(25) George, S. M. Atomic layer deposition: An overview. Chemical reviews 2010, 110, 111–131.
(26) Miikkulainen, V.; Leskel, M.; Ritala, M.; Puurunen, R. L. Crystallinity of inorganic films
grown by atomic layer deposition: Overview and general trends. Journal of Applied Physics 2013,
113, 21301.
(27) Bachmann, J.; Zierold, R.; Chong, Y. T.; Hauert, R.; Sturm, C.; Schmidt-Grund, R.;
Rheinländer, B.; Grundmann, M.; Gösele, U.; Nielsch, K. A practical, self-catalytic, atomic layer
deposition of silicon dioxide. Angewandte Chemie (International ed. in English) 2008, 47, 6177–
6179.
(28) Bae, C.; Zierold, R.; Montero Moreno, J. M.; Kim, H.; Shin, H.; Bachmann, J.; Nielsch, K.
Multisegmented nanotubes by surface-selective atomic layer deposition. J. Mater. Chem. C 2013,
1, 621–625.
23
(29) Haschke, S.; Pankin, D.; Petrov, Y.; Bochmann, S.; Manshina, A.; Bachmann, J. Design Rules
for Oxygen Evolution Catalysis at Porous Iron Oxide Electrodes: A 1000-Fold Current Density
Increase. ChemSusChem 2017, 10, 3644–3651.
(30) David R. Lide. Handbook of Chemistry & Physics, 73rd Edition; CRC Press: Boca Raton,
1992.
(31) Magnetic Nano- and Microwires; Elsevier, 2015.
(32) Park, B. C.; Kim, Y. K. Synthesis, microstructure, and physical properties of metallic barcode
nanowires. Met. Mater. Int. 2017, 23, 413–425.
(33) Wang, T.; Wang, Y.; Fu, Y.; Hasegawa, T.; Oshima, H.; Itoh, K.; Nishio, K.; Masuda, H.; Li,
F. S.; Saito, H. et al. Magnetic behavior in an ordered Co nanorod array. Nanotechnology 2008,
19, 455703.
(34) Fodor, P. S.; Tsoi, G. M.; Wenger, L. E. Modeling of hysteresis and magnetization curves for
hexagonally ordered electrodeposited nanowires. Journal of Applied Physics 2003, 93, 7438–7440.
(35) O'Handley, R. C. Modern magnetic materials: Principles and applications; Wiley: New York,
NY, 2000.
(36) Salem, M. S.; Sergelius, P.; Corona, R. M.; Escrig, J.; Görlitz, D.; Nielsch, K. Magnetic
properties of cylindrical diameter modulated Ni80Fe20 nanowires: Interaction and coercive fields.
Nanoscale 2013, 5, 3941–3947.
(37) Zeng, H.; Skomski, R.; Menon, L.; Liu, Y.; Bandyopadhyay, S.; Sellmyer, D. J. Structure and
magnetic properties of ferromagnetic nanowires in self-assembled arrays. Phys. Rev. B 2002, 65,
R4875.
(38) Allende, S.; Altbir, D.; Nielsch, K. Magnetic cylindrical nanowires with single modulated
diameter. Phys. Rev. B 2009, 80, 47.
(39) Zeng, H.; Zheng, M.; Skomski, R.; Sellmyer, D. J.; Liu, Y.; Menon, L.; Bandyopadhyay, S.
Magnetic properties of self-assembled Co nanowires of varying length and diameter. Journal of
Applied Physics 2000, 87, 4718–4720.
24
(40) Arrott, A. S.; Heinrich, B.; Templeton, T. L.; Aharoni, A. Micromagnetics of curling
configurations in magnetically soft cylinders. Journal of Applied Physics 1979, 50, 2387–2389.
(41) Jamet, S.; Rougemaille, N.; Toussaint, J. C.; Fruchart, O. Head-to-head domain walls in one-
dimensional nanostructures. Magnetic Nano- and Microwires; Elsevier, 2015; pp 783–811.
(42) Rotaru, A.; Lim, J.-H.; Lenormand, D.; Diaconu, A.; Wiley, J. B.; Postolache, P.; Stancu, A.;
Spinu, L. Interactions and reversal-field memory in complex magnetic nanowire arrays. Phys. Rev.
B 2011, 84, DOI: 10.1103/PhysRevB.84.134431.
(43) Ciuta, G.; Dumas-Bouchiat, F.; Dempsey, N. M.; Fruchart, O. Some Aspects of Magnetic
Force Microscopy of Hard Magnetic Films. IEEE Trans. Magn. 2016, 52, 1–8.
(44) Da Col, S.; Jamet, S.; Rougemaille, N.; Locatelli, A.; Mentes, T. O.; Burgos, B. S.; Afid, R.;
Darques, M.; Cagnon, L.; Toussaint, J. C. et al. Observation of Bloch-point domain walls in
cylindrical magnetic nanowires. Phys. Rev. B 2014, 89, 101.
25
|
1810.06772 | 1 | 1810 | 2018-10-16T01:04:01 | Studies of two-dimensional MoS2 on enhancing the electrical performance of ultrathin copper films | [
"physics.app-ph"
] | Copper nanowires are widely used as on-chip interconnects due to superior conductivity. However, with aggressive Cu interconnect scaling, the diffusive surface scattering of electrons drastically increases the electrical resistivity. In this work, we studied the electrical performance of Cu thin films on different materials. By comparing the thickness dependence of Cu films resistivity on MoS2 and SiO2, we demonstrated that two-dimensional MoS2 can be used to enhance the electrical performance of ultrathin Cu films due to a partial specular surface scattering. By fitting the experimental data with the theoretical Fuchs Sondheimer model, we obtained the specularity parameter at the Cu MoS2 interface in the temperature range 2K to 300K. Furthermore, first principle calculations based on the density functional theory indicates that there are more localized states at the Cu amorphous SiO2 interface than the Cu MoS2 interface which is responsible for the higher resistivity in the Cu SiO2 heterostructure due to more severe electron scattering. Our results suggest that Cu MoS2 hybrid is a promising candidate structure for the future generations of CMOS interconnects. | physics.app-ph | physics | Studies of two-dimensional MoS2 on enhancing the electrical performance of
ultrathin copper films
Tingting Shen1, 3*, Daniel Valencia2 ,4, Qingxiao Wang5, Kuang-Chung Wang2, Michael Povolotskyi3,
Moon J. Kim5, Gerhard Klimeck2, Zhihong Chen2 ,3, and Joerg Appenzeller2 ,3
1) Department of Physics and Astronomy, Purdue University, West Lafayette, IN, USA 47907
2) Department of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA 47907
3) Birck Nanotechnology Center, Purdue University, West Lafayette, IN, USA 47907
4) The Charles Stark Draper Laboratory Inc., Cambridge, MA, USA 02139
5) Department of Materials Science and Engineering, University of Texas at Dallas, Richardson, TX, USA 75080
Abstract:
Copper nanowires are widely used as on-chip interconnects due to superior conductivity. However, with
aggressive Cu interconnect scaling, the diffusive surface scattering of electrons drastically increases the
electrical resistivity. In this work, we studied the electrical performance of Cu thin films on different
materials. By comparing the thickness dependence of Cu films resistivity on MoS2 and SiO2, we
demonstrated that two-dimensional MoS2 can be used to enhance the electrical performance of ultrathin Cu
films due to a partial specular surface scattering. By fitting the experimental data with the theoretical Fuchs --
Sondheimer (FS) model, we claimed that the specularity parameter at the Cu/MoS2 interface is p ≈ 0.4 in
the temperature range 1.8K < T < 300K. Furthermore, first principle calculations based on the density
functional theory (DFT) indicates that there are more localized states at the Cu/amorphous SiO2 interface
than the Cu/MoS2 interface which is responsible for the higher resistivity in the Cu/SiO2 heterostructure
due to more severe electron scattering. Our results suggest that Cu/MoS2 hybrid is a promising candidate
structure for the future generations of CMOS interconnects.
Keywords: Cu interconnect, surface scattering, MoS2, resistivity, FS model, DFT
Introduction
Copper is widely used as the interconnect material due to superior conductivity [1-4]. Along with the scaling
of VLSI circuits, the scaling of the interconnects is also highly demanded. The scaling trends of the height
and width of the interconnects are shown in the International Technology Roadmap for Semiconductors
(ITRS) [4]. However, when the thickness of Cu film decreases down to the electron mean free path which
is 40nm at room temperature [5], the electrical resistivity will increase significantly due to the increased
electron scatterings from film surfaces [6-8] and grain boundaries [9-10]. This size effect impacts the time
delay of the interconnects severely and represents a major challenge for the development of nanoelectronics
[3-4]. To resolve this problem, novel Cu/barrier interfaces with specular rather than diffusive electron
scattering need to be developed to achieve high-conductivity interconnects [11-12]. Although pristine
atomically smooth Cu surface shows partial specular scattering [11, 13], there are many factors such as
surface roughness [8], oxidation in ambient environment [10-12] or coatings of secondary materials [14]
that will lead to completely diffusive surface scattering due to the randomization of the electron momentum
in the current flow direction. Furthermore, interconnect system requires not only good electrical
performance but also the capability to mitigate Cu diffusion into damascene structures. Conventional barrier
materials such as Ta/TaN and TiN have been used to isolate Cu from the surrounding dielectrics. Since the
resistivity of the conventional barrier materials are more than one order higher than that of Cu [15], the
thickness of the barrier material needs to be reduced as much as possible to maximize the Cu volume for
lower line resistance. But when the thickness of the conventional barrier material is scaled down below
3nm, the barrier cannot block Cu diffusion anymore. To achieve high performance integrated circuits, sub-
1nm novel diffusion barriers which yield specular electron surface scattering are highly demanded.
2D layered materials have attracted intense research interest for the application in copper interconnect
technology [12, 16-20] due to their ultra-thin body thickness. Recent studies show that atomic graphene not
only has excellent performance in blocking Cu diffusion [19-20] but also can enhance the electrical and
thermal conductivity of Cu [12]. R. Mehta et al. [12] reported a partial specular scattering of p = 0.23 at
graphene coated Cu surfaces. On the other hand, two-dimensional layered semiconducting transition metal
dichalcogenides (TMDs) like MoS2 have also been indicated to be good Cu diffusion barrier materials [16].
However, there are few studies on the electrical properties of Cu/MoS2 barrier hybrid [18]. In this letter, we
studied the electrical performance of Cu thin films on different materials. Our experimental results show
that for Cu films with the same thickness, Cu on MoS2 always show much lower resistivity than Cu on SiO2.
Analyzing the relationship between Cu resistivity and thickness at different temperatures, we demonstrate
that surface scattering is the main contribution to the total resistivity when Cu is thinner than 100nm, and
the Cu/MoS2 interface shows partial specular scattering with a temperature independent specularity p = 0.4
which is better than that reported in Cu/Ni [11] and Cu/graphene [12] structures. Furthermore, we studied
the electronic properties of four different Cu surfaces: pure Cu, Cu/amorphous SiO2, Cu/crystalized SiO2
and Cu/MoS2 by first principle calculations based on the density functional theory (DFT). It is found that:
(1) the DOS of Cu/MoS2 interface is similar with pure Cu surface; (2) the DOS of Cu/amorphous SiO2 and
Cu/crystalized SiO2 is much higher than that of Cu surface. It is worth to mention that the states at the
interfaces are localized which will trap electrons traversing near the interface. Upon subsequent release, the
electron momentum will be randomized in the current direction. Thus, the Cu/SiO2 heterostructure always
show higher resistivity than the Cu/MoS2 heterostructure since the high localized states at the Cu/SiO2
interface caused complete inelastic surface scattering. Our results indicate Cu/MoS2 hybrid has significantly
improved the electrical performance of thin Cu films which is highly desirable for future generations of
CMOS interconnects.
Results and Discussion:
Figure 1. (a) Schematic diagram and (b) representative SEM image of a Cu on MoS2 device. (c) Schematic
diagram and (d) representative SEM image of a Cu on SiO2 device.
We fabricated two types of devices, Cu on MoS2 and Cu on SiO2, to study the electrical performance of Cu
thin films on different materials. Figure 1(a)-(d) show the schematic diagrams and representative SEM
images of Cu/MoS2 and Cu/SiO2 devices respectively. To achieve fair Cu electrical performance
comparison, these two types devices were patterned into structures with the same dimensions on the same
Si/SiO2 substrates as the representative SEM images show in Figure 1(b) and Figure 1(d). Cu thin films of
different thickness were deposited using an e-beam evaporation system and the electrical resistance was
measured by four-probe methods in a probe station set-up. The measurement geometry is shown in Figure
1(a) and Figure 1(c). Details of the fabrication are described in the "Methods" section.
Figure 2. Cross-sectional STEM, EDS, and EELS map of a Cu on MoS2 (a)-(c) and Cu on SiO2 (d)-(f)
device. The blue, red and pink line represents the signal of Cu, O and Si respectively in the cyan arrow
direction. There is a C layer deposited on top of both devices during the sample preparation using focused
ion beam (FIB) micromachining.
Before discussing the electrical performance, more details of the device structure should be analyzed. We
carried out cross-sectional structure analysis and chemical mapping of a Cu on MoS2 and a Cu on SiO2
device by scanning transmission electron microscopy (STEM) in conjunction with energy dispersive X-ray
spectroscopy (EDS) and electron energy loss spectroscopy (EELS). As it is shown in Figure 2(a)-(f), O
signal was detected on the top surface of both devices because there was a CuOx layer formed due to Cu
oxidation in air. Besides, there is another O signal on the Cu/SiO2 interface as it is shown in Figure 2(e)
and Figure 2(f). It is worth noting that in Figure 2(e), the second O signal, which is marked with green
circle, appeared simultaneously with the Si signal, while the O and Si signals at the Cu upper surface marked
in the black circle are different. Hence, the O signal belongs to SiO2 rather than to CuOx, so no CuOx is
formed at Cu/SiO2 or Cu/MoS2 interfaces. The thickness of CuOx measured from the blue dashed lines in
Figure 2(a)-(f) is around 2.5nm for both devices. We assume the CuOx thickness is the same for all the Cu
films in both Cu/SiO2 and Cu/MoS2 structures. Besides, Atomic Force Microscope (AFM) was used to
characterize the total thickness of CuOx/Cu. After subtracting 2.5nm CuOx from the total thickness, we can
identify the real thickness of Cu which is significantly important for the Cu electrical resistivity calculation
from the measured resistance results in the electrical performance analysis.
Figure 3. The resistivity of Cu on SiO2 and Cu on MoS2 as a function of the inverse of Cu film thickness at
room temperature. The relationship between resistivity and Cu thickness is shown in the insert. Error bars
capture the uncertainty in film thickness determination and resistivity calculation. The dots are experimental
data and the solid lines are fitting results using FS analytical model.
Next, we will analyze the electrical performance of Cu thin films on different materials. Figure 3(a) shows
the resistivity of Cu on SiO2 and Cu on MoS2 as a function of the inverse of Cu film thickness at room
temperature. The corresponding relationship of resistivity and thickness is shown in the insert. The dots are
experimental data, and the lines are theoretical fitting results with Fuchs -- Sondheimer (FS) model. Each
data point represents the averaged resistivity obtained from more than 10 individual devices with the same
Cu thickness. Using four-probe method, we measured the resistance of Cu thin films in the geometry shown
in Figure 1. During the measurement, we obtained similar resistance values with 400µA DC current and
10µA AC current, thus the Joule heating effect is negligible in our experiment. Comparing the resistivity
of Cu on MoS2 and SiO2, we found: (1) the resistivity of Cu increases dramatically with the decrease of
thickness regardless the underlying material. (2) when the thickness is larger than 100nm, the resistivity of
Cu on MoS2 and SiO2 are similar. (3) for thinner Cu films, the resistivity on MoS2 is smaller than that on
SiO2 and the thinner the Cu the larger the difference between these two cases. There are two reasons which
may result in lower resistivity in Cu on MoS2 case: (i) the semiconducting MoS2 underneath Cu film works
as a parallel electron transport channel which decreases the total resistivity of Cu; (ii) the electron scattering
mechanism is different for Cu on different materials. To figure out which reason dominants, we measured
gate dependence of the current in one Cu on MoS2 device. It is well known that for a MoS2 field effect
transistor device, the on/off ratio should be 7-8 orders of magnitude [21-22]. If the MoS2 plays an important
role in the electron transportation, then the current through the device should be tuned dramatically by the
gate voltage. However, the current only changes within 1% for -40V < Vgate < 40V which means reason (i)
should not be considered in our analysis. Thus, the resistivity difference should be related to different
electron scattering in these two cases.
The contributions of surface scattering and grain boundary scattering to the total resistivity 𝜌 = 𝜌𝑆 + 𝜌𝐺
can be modeled by the Fuchs-Sondheimer (FS) equation (1) [23] and the Mayadas-Shatzkes (MS) equation
(2) [24] respectively:
𝜌𝑆 = 𝜌0[1 +
3
8
Λ0
𝑇
(1 − 𝑝)] (1)
𝜌𝐺 = 𝜌0[1 −
3𝛼
2
+ 3𝛼2 − 3𝛼3 ln (1 +
)]−1, 𝑤ℎ𝑒𝑟𝑒 𝛼 =
1
𝛼
𝜆
𝑑𝑔𝑟𝑎𝑖𝑛
𝑅
(
1−𝑅
) (2)
Here, 𝜌0 is the bulk resistivity of Cu, Λ0 is the electron mean free path, T is the thickness of the film, p is
the specularity parameter ranging from 0 (completely diffuse) to 1 (specular scattering), 𝑑𝑔𝑟𝑎𝑖𝑛 is the
average grain size and R is the grain-boundary reflection coefficient. The experimental resistivity shown in
Figure 3 is linear with the inverse of Cu film thickness. Since the trap states at the Cu-oxide interface
perturbs the smooth surface potential of Cu [14], we assumed the electron scattering at the Cu/CuOx and
Cu/SiO2 interfaces is completely diffusive (p = 0) and fitted the experimental results of Cu on SiO2 with FS
equation (Equation 1) to obtain the values of Cu bulk resistivity and electron mean free path. The fitting
result is the blue solid line shown in Figure 3 corresponds to 𝜌0 = 1.69*10-8 Ωm and 𝜌0Λ0 = 1.99*10-15
Ωm2. Since the fitted bulk Cu resistivity is the same with the reported bulk value, together with the linear
relationship between the resistivity and 1/Thickness, we regard the surface scattering as the dominant
contribution to the electrical resistivity and the effects of the grain boundary scattering is negligible.
However, the FS model has its intrinsic limitations because it is based on two approximations that are not
justified for small thickness. It is assumed that the electronic structure is as in bulk and the surface is smooth,
so the surface scattering occurs only at the surface. Thus, the FS model does not include the contribution
from roughness scattering, it is not adequate to describe thin films with surface roughness. This causes the
deviation of the fitted 𝜌0Λ0 from the acknowledged value 6.6 * 10-16 Ωm2 [9, 25].
Later studies have proposed other models to include the contribution of surface roughness to the electrical
resistivity [8, 13, 26-27]. Among them, the extended FS model [13] and the power law model developed
by T. Zhou et al [26] claimed that the resistivity contribution from surface scattering is still proportional to
1/Thickness which coincides with our experimental results. They include a numerical factor 𝛼 equal to or
larger than 1 which is related to the roughness to the second term of Equation 1:
𝜌 = 𝜌0[1 +
3
8
𝛼Λ0
𝑇
(1 − 𝑝)] (3)
For Cu/SiO2 devices, we used p = 0, 𝜌0Λ0 = 6.6 * 10-16 Ωm2 and fitted the experimental results with
Equation 3. The fitting results are 𝜌0 = 1.69*10-8 and 𝛼 = 3.02. According to atomic force microscopy
results, the surface roughness of Cu on MoS2 is similar with Cu on SiO2, thus 𝛼 is the same for both cases.
With all the parameters obtained from Cu on SiO2 devices, we fitted the experimental results of Cu on MoS2
with Equation 3 as the solid red line shown in Figure 3 and obtained p = 0.39 for the Cu/MoS2 interface.
Figure 4. (a) The resistivity of Cu on SiO2 and Cu on MoS2 as a function of temperature for different Cu
film thicknesses. (b) The resistivity of Cu on SiO2 and Cu on MoS2 as a function of the inverse of Cu film
thickness at 1.8K shown in the dashed box in (a). The dots are experimental data and the solid lines are
fitting results of FS analytical model. (c) The calculated α*(1-p) as a function of temperature of Cu on SiO2
and Cu on MoS2 extracted from (b) at different temperatures. (d) The calculated specularity parameter p as
a function of temperature of Cu on MoS2 from (c). Error bars capture the uncertainty in numerical
calculations and film thickness determination.
In addition, we studied the temperature dependence of the electron scattering in Cu thin films. Figure 4(a)
shows the experimental resistivity of Cu on MoS2 (red curves) and Cu on SiO2 (blue curves) as a function
of temperature ranging from 300K to 1.8K for Cu films with different thickness. The measurement was
carried out in a physical property measurement system (PPMS) with 10µA AC current. Each curve
represents one set of experimental data obtained from one device. When 50K < T < 300K, the resistivity is
linear with temperature and when the temperature is lower than 30K, the resistivity curves flatten out and
reach constant residual resistivities [8, 28]. This is because at higher temperature phonon scattering
dominants the total electoral resistivity. When the thermal energy becomes smaller than the phonon energy
at low temperature, the phonon scattering is negligible and the contributions to the resistivity becomes
temperature independent surface scattering, grain boundary scattering and impurity scattering. Moreover,
within each set of the blue and red curves, thicker Cu films always show lower resistivity and for similar
film thickness, the Cu resistivity on MoS2 is lower than that on SiO2 which agrees with the previous analysis.
Figure 4(b) is the residual resistivity of Cu in both cases as a function of 1/Thickness which is also labeled
in the dashed box of Figure 4(a). The symbols are experimental data and the solid lines are eye guide linear
relationship between Cu resistivity and 1/Thickness.
⁄
Λ0/𝑇)]. However, this prediction cannot describe
At low temperature, the FS model predicts 𝜌 ∝ 1 [𝑇𝑙𝑛(
the experimental data correctly due to its intrinsic limitation: in the limit of high-purity films at low
temperature, Λ0→∞, the FS model predicts a vanishing thin-film resistivity since surface scattering alone
cannot relax carriers within the FS model [26]. Later studies reported by T. Zhou et al [26] have proposed
another model which claims the resistivity contribution from surface scattering is temperature-independent
and proportional to 1/Thickness to replace the FS model. Accordingly, we fitted the experimental results at
different temperatures for Cu thin films on both MoS2 and SiO2 with Equation 3. Same as the analysis in
Figure 3, we fitted the Cu on SiO2 experimental data with p = 0 and 𝜌0Λ0 = 6.6 * 10-16 Ωm2 which is
temperature independent [29] and obtained the fitting values of α. Using the same α, we extracted p = 0.4
at the Cu/MoS2 interface at 1.8K according to the slope difference of the blue and red solid lines. Same
fitting was also done for experimental resistivity data at different temperatures. The fitted α*(1-p) for both
Cu on SiO2 and Cu on MoS2 is shown in Figure 4(c). For Cu on SiO2, the specularity is p=0 for all the
temperatures because of the diffusive surface scattering. Thus, the blue points shown in Figure 4(c)
represent α at different temperatures. The specularity of the Cu/MoS2 interface extracted from the difference
between the red and blue points is shown in Figure 4(d). Here, the fitting results show the specularity
parameter at the Cu/MoS2 interface is temperature independent with p ≈ 0.4 which means the Cu/MoS2 has
a temperature independent elastic surface scattering. Our experimental finding is consistent with the
theoretical results reported by T. Zhou et. al [26]. The temperature dependent α*(1-p) in Figure 4(c) means
the α is temperature dependent. Although α represent the surface roughness contribution to the total
resistivity and should be independent of temperature, the slightly increase with temperature is
understandable since as the increase of temperature, the trap charge density of states [30] at the Cu/CuxO,
Cu/SiO2 and Cu/MoS2 interfaces might be increased slightly, and the surface scattering becomes more
severe. As a result, the resistivity increases faster as a function of 1/Thickness at higher temperature. The
fitting result in Ref [30] also show α varies with temperature, but we still need more studies to investigate.
Figure 5. (a) Total DOS for Cu films with different interfaces. The red, blue, green and black corresponds
to the simulated interface between Cu and cristobalite SiO2, amorphous SiO2, MoS2 and a Cu interface with
no passivated atoms respectively. Projected DOS at the interface of (b) Cu/MoS2 and (c) Cu/amorphous
SiO2.
To explain the underlying mechanisms for the partial elastic surface scattering at the Cu/MoS2 interface,
we carried out first principle calculations based on the DFT to study the density of states at different Cu
interfaces. The details of the computational simulation are discussed in the "Methods" section. As shown
in Figure 5(a), the interface between Cu and MoS2 has similar DOS with the free Cu surface, while the
interfaces between Cu and cristobalite SiO2 and amorphous SiO2 have much higher DOS. Figure 5(b) and
(c) are the projected DOS at the interface of Cu/MoS2 and Cu/amorphous SiO2 respectively. They show
that the available states at these two interfaces are localized rather than continuous which means there are
more trapping states at the Cu/amorphous SiO2 interface than the Cu/MoS2 interface. When electrons
transport in thin Cu films, the probability to be trapped in the Cu/amorphous SiO2 is much higher than that
of the Cu/MoS2 interface. Upon subsequent release, the trapped electrons have randomized momentum in
the current flow direction. This explains why the Cu/SiO2 heterostructures show higher resistivity. Our
findings are in accordance with other works that demonstrate the interaction between MoS2 and Cu is very
weak [31] while the oxidation of the Cu surface or adsorption of foreign adatoms may cause perturbations
to the Cu surface potential and effectively results in severe surface scattering [14, 32].
In summary, we studied the resistivity of thin Cu films on different materials and demonstrated two-
dimensional MoS2 can be used to enhance the electrical performance of Cu. With the scaling of the film
thickness, the resistivity increases dramatically because of the diffusive surface scattering. However, by
inserting MoS2 under Cu, the resistivity can be decreased significantly due to the partial specular surface
scattering at the Cu/MoS2 interface. Our experimental results suggest a resistivity contribution from surface
scattering on Cu surfaces is proportional to 1/Thickness at the temperature 1.8K-300K. From the analytical
fitting results, we obtained a temperature independent specularity p ≈ 0.4 at Cu/MoS2 interface. According
to the DFT calculations, the higher resistivity in the Cu/SiO2 heterostructure is caused by the higher density
of localized states at the Cu/amorphous SiO2 interface than the Cu/MoS2 interface. Currently, only one
surface of Cu thin film has been coated with MoS2. If we could coat MoS2 on all Cu surfaces, the resistivity
of Cu can be reduced even further which is highly desirable for future generations of CMOS interconnects.
Methods
Few-layer MoS2 flakes were exfoliated on Si/SiO2 substrates followed by 200℃ annealing for 5h in high
vacuum. Subsequently, four probe test structures (4μm Length × 2μm Width) were fabricated on MoS2 and
SiO2 surfaces with Cu thickness ranging from 8.5nm to 102.5nm using e-beam lithography, e-beam
evaporation metal deposition and conventional lift-off. The thickness of Cu films was measured by an
atomic force microscope (AFM) set-up.
Computational Details
To quantify the effect of the SiO2 and MoS2 interface over the copper atoms, first principles calculations
were carried out by the density functional theory (DFT), using projector-argument waves (PAWs) as
implemented in the VASP code [33]. In these calculations, the generalized-gradient approximation (GGA)
with the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional were used [33]. In all the
calculations, an energy cut-off of 500 eV with a convergence criterion of 10^-8 eV and 0.1 eV Å-1 for
energies and forces, respectively, were used.
In this work, the Cu/SiO2-Cristob configuration corresponds to the simulated interface between Cu and
Cristobalite SiO2. The configuration was constructed following the same process reported by T. Shan et al
[34]. This interface was generated for copper oriented in the (001) direction which was matched to -
cristobalite (001). Based on the work in Ref 33, an oxygen terminated interface was chosen since this type
of termination has the strongest adhesion energy between both materials. To reduce the strain effects at the
edges of the interface, 8 atomic layers of each material were used in the x direction as show in Fig. 5 (b)-
(c) and only four layers on each side were relaxed while the rest of the atoms were fixed during the ionic
relaxation.
Making use of the structure previously described, Cu with amorphous SiO2 was also studied. The
amorphous silica used in the interface is prepared using the melt and quench method as suggested in Ref
35. This process was carried out with the ReaxFF potential modified for Cu/SiO2 interface as reported in
Ref 36 in the large-scale atomic/molecular massive parallel simulation (LAMMPS) [37]. During the
molecular dynamics (MD) process, the copper atoms are fixed, and the atoms are melted from 300 K to
2000 K at a constant pressure for 200ps to ensure complete melting. Afterwards, the structure is quenched
using a stepwise cooling scheme at a rate of 12.5mK/fs and the structure is equilibrated at 300 K for an
additional 10ps and then relaxed in DFT making use of the same parameters used for Cu/SiO2-Cristob
configuration.
Finally, the Cu MoS2 interface is obtained by straining the MoS2 atoms to match the Cu interface and then
the supercell is relaxed following the same process described for the SiO2 cristobalite interface.
AUTHOR INFORMATION
Corresponding Author
*E-mail: [email protected]
Author contributions
T.S. worked on the device fabrication, characterization and data analysis; D.V., K.W. and M.P. worked on
the DTF calculations; Z.C., J.A. and G.K. analyzed the data and oversaw the planning and execution of the
project; Q.W. and M.K. worked on the structure analysis; T.S. wrote the manuscript.
Notes
The authors declare no competing financial interest.
ACKNOWLEDGMENTS
The authors thank the staff at the Birck Nanotechnology center for their technical support. This work was
in part supported by the STARnet center LEAST, a Semiconductor Research Corporation program
sponsored by MARCO and DARPA. T. Shen is thankful for the help from Yuqi Zhu for discussions about
data analysis.
References
(1) Kapur, P., McVittie, J. P., & Saraswat, K. C. (2002). Technology and reliability constrained future
copper interconnects. I. Resistance modeling. IEEE Transactions on electron devices, 49(4), 590-597.
(2) Banerjee, K., Amerasekera, A., Dixit, G., & Hu, C. (1996, December). The effect of interconnect scaling
and low-k dielectric on the thermal characteristics of the IC metal. In Electron Devices Meeting, 1996.
IEDM'96., International (pp. 65-68). IEEE.
(3) Havemann, R. H., & Hutchby, J. A. (2001). High-performance interconnects: An integration overview.
Proceedings of the IEEE, 89(5), 586-601.
(4) http://www.itrs.net
(5) Graham, R. L., Alers, G. B., Mountsier, T., Shamma, N., Dhuey, S., Cabrini, S., ... & Peddeti, S. (2010).
Resistivity dominated by surface scattering in sub-50 nm Cu wires. Applied Physics Letters, 96(4),
042116.
(6) Fuchs, K. (1938, January). The conductivity of thin metallic films according to the electron theory of
metals. In Mathematical Proceedings of the Cambridge Philosophical Society (Vol. 34, No. 1, pp. 100-
108). Cambridge University Press.
(7) Sondheimer, E. H. (1952). The mean free path of electrons in metals. Advances in physics, 1(1), 1-42.
(8) Timalsina, Y. P., Horning, A., Spivey, R. F., Lewis, K. M., Kuan, T. S., Wang, G. C., & Lu, T. M.
(2015). Effects of nanoscale surface roughness on the resistivity of ultrathin epitaxial copper films.
Nanotechnology, 26(7), 075704.
(9) Mayadas, A. F., & Shatzkes, M. (1970). Electrical-resistivity model for polycrystalline films: the case
of arbitrary reflection at external surfaces. Physical review B, 1(4), 1382.
(10) Chawla, J. S., Gstrein, F., O'Brien, K. P., Clarke, J. S., & Gall, D. (2011). Electron scattering at
surfaces and grain boundaries in Cu thin films and wires. Physical Review B, 84(23), 235423.
(11) Zheng, P. Y., Deng, R. P., & Gall, D. (2014). Ni doping on Cu surfaces: Reduced copper resistivity.
Applied Physics Letters, 105(13), 131603.
(12) Mehta, R., Chugh, S., & Chen, Z. (2015). Enhanced electrical and thermal conduction in graphene-
encapsulated copper nanowires. Nano letters, 15(3), 2024-2030.
(13) Ke, Y., Zahid, F., Timoshevskii, V., Xia, K., Gall, D., & Guo, H. (2009). Resistivity of thin Cu films
with surface roughness. Physical Review B, 79(15), 155406.
(14) Zahid, F., Ke, Y., Gall, D., & Guo, H. (2010). Resistivity of thin Cu films coated with Ta, Ti, Ru, Al,
and Pd barrier layers from first principles. Physical Review B, 81(4), 045406.
(15) Kapur, P., Chandra, G., McVittie, J. P., & Saraswat, K. C. (2002). Technology and reliability
constrained future copper interconnects. II. Performance implications. IEEE Transactions on electron
devices, 49(4), 598-604.
(16) Lo, C. L., Catalano, M., Smithe, K. K., Wang, L., Zhang, S., Pop, E., ... & Chen, Z. (2017). Studies of
two-dimensional h-BN and MoS2 for potential diffusion barrier application in copper interconnect
technology. npj 2D Materials and Applications, 1(1), 42.
(17) Lo, C. L., Smithe, K. K., Mehta, R., Chugh, S., Pop, E., & Chen, Z. (2017, April). Atomically thin
diffusion barriers for ultra-scaled Cu interconnects implemented by 2D materials. In Reliability Physics
Symposium (IRPS), 2017 IEEE International (pp. MR-4). IEEE.
(18) Lo, C. L., Zhang, S., Shen, T., Appenzeller, J., & Chen, Z. (2017, June). BEOL compatible 2D layered
materials as ultra-thin diffusion barriers for Cu interconnect technology. In Device Research
Conference (DRC), 2017 75th Annual (pp. 1-2). IEEE.
(19) Li, L., Chen, X., Wang, C. H., Cao, J., Lee, S., Tang, A., ... & Wong, H. S. P. (2015). Vertical and
lateral copper transport through graphene layers. ACS nano, 9(8), 8361-8367.
(20) Mehta, R., Chugh, S., & Chen, Z. (2017). Transfer-free multi-layer graphene as a diffusion barrier.
Nanoscale, 9(5), 1827-1833.
(21) Das, S., Chen, H. Y., Penumatcha, A. V., & Appenzeller, J. (2012). High performance multilayer MoS2
transistors with scandium contacts. Nano letters, 13(1), 100-105.
(22) Shen, T., Penumatcha, A. V., & Appenzeller, J. (2016). Strain engineering for transition metal
dichalcogenides based field effect transistors. ACS nano, 10(4), 4712-4718.
(23) Sondheimer, E. H. (1949). Influence of a magnetic field on the conductivity of thin metallic films.
Nature, 164(4178), 920.
(24) Mayadas, A. F., & Shatzkes, M. (1970). Electrical-resistivity model for polycrystalline films: the case
of arbitrary reflection at external surfaces. Physical review B, 1(4), 1382.
(25) Josell, D., Brongersma, S. H., & Tőkei, Z. (2009). Size-dependent resistivity in nanoscale
interconnects. Annual Review of Materials Research, 39, 231-254.
(26) Zhou, T., & Gall, D. (2018). Resistivity scaling due to electron surface scattering in thin metal layers.
Physical Review B, 97(16), 165406.
(27) Timoshevskii, V., Ke, Y., Guo, H., & Gall, D. (2008). The influence of surface roughness on electrical
conductance of thin Cu films: an ab initio study. Journal of Applied Physics, 103(11), 113705.
(28) Matula, R. A. (1979). Electrical resistivity of copper, gold, palladium, and silver. Journal of Physical
and Chemical Reference Data, 8(4), 1147-1298.
(29) Gall, D. (2016). Electron mean free path in elemental metals. Journal of Applied Physics, 119(8),
085101.
(30) Rickman, J. M., & Barmak, K. (2012). Resistivity in rough metallic thin films: A Monte Carlo study.
Journal of Applied Physics, 112(1), 013704.
(31) Le, D., Sun, D., Lu, W., Bartels, L., & Rahman, T. S. (2012). Single layer MoS2 on the Cu (111)
surface: first-principles electronic structure calculations. Physical Review B, 85(7), 075429.
(32) Chawla, J. S., & Gall, D. (2009). Specular electron scattering at single-crystal Cu (001) surfaces.
Applied Physics Letters, 94(25), 252101.
(33) Kresse, G., & Furthmüller, J. (1996). Efficiency of ab-initio total energy calculations for metals and
semiconductors using a plane-wave basis set. Computational materials science, 6(1), 15-50.
(34) Shan, T. R., Devine, B. D., Phillpot, S. R., & Sinnott, S. B. (2011). Molecular dynamics study of the
adhesion of Cu/SiO2 interfaces using a variable-charge interatomic potential. Physical Review B, 83(11),
115327.
(35) Yu, W., Wu, L., & Shen, S. (2017). Adhesion properties of Cu (111)/α-quartz (0001) interfaces: A
molecular dynamics study. Materials Science and Engineering: A, 695, 239-248.
(36) Onofrio, N., Guzman, D., & Strachan, A. (2015). Atomic origin of ultrafast resistance switching in
nanoscale electrometallization cells. Nature materials, 14(4), 440.
(37) Plimpton, S. (1995). Fast parallel algorithms for short-range molecular dynamics. Journal of
computational physics, 117(1), 1-19.
|
1812.07420 | 1 | 1812 | 2018-12-18T15:04:21 | Petahertz Spintronics | [
"physics.app-ph",
"cond-mat.mtrl-sci",
"physics.optics",
"quant-ph"
] | The enigmatic coupling between electronic and magnetic phenomena was one of the riddles propelling the development of modern electromagnetism. Today, the fully controlled electric field evolution of ultrashort laser pulses permits the direct and ultrafast control of electronic properties of matter and is the cornerstone of light-wave electronics. In sharp contrast, because there is no first order interaction between light and spins, the magnetic properties of matter can only be affected indirectly on the much slower tens-of-femtosecond timescale in a sequence of optical excitation followed by the rearrangement of the spin structure. Here we record an orders of magnitude faster magnetic switching with sub-femtosecond response time by initiating optical excitations with near-single-cycle laser pulses in a ferromagnetic layer stack. The unfolding dynamics are tracked in real-time by a novel attosecond time-resolved magnetic circular dichroism (atto-MCD) detection scheme revealing optically induced spin and orbital momentum transfer (OISTR) in synchrony with light field driven charge relocation. In tandem with ab-initio quantum dynamical modelling, we show how this mechanism provides simultaneous control over electronic and magnetic properties that are at the heart of spintronic functionality. This first incarnation of attomagnetism observes light field coherent control of spin-dynamics in the initial non-dissipative temporal regime and paves the way towards coherent spintronic applications with Petahertz clock rates. | physics.app-ph | physics | Petahertz Spintronics
Florian Siegrist1,5, Julia A. Gessner1,5, Marcus Ossiander1, Christian Denker2, Yi-Ping Chang1,
Malte C. Schröder1, Alexander Guggenmos1,5, Yang Cui5, Jakob Walowski2, Ulrike Martens2, J.
K. Dewhurst4, Ulf Kleineberg1,5, Markus Münzenberg2, Sangeeta Sharma3, Martin Schultze1,5,*
1Max-Planck-Institute of Quantum Optics, Hans-Kopfermann-Strasse 1, 85748 Garching
2Institut für Physik, Universität Greifswald, Felix-Hausdorff-Strasse 6, 17489 Greifswald
3Max Born Institute for Nonlinear Optics and Short Pulse Spectroscopy, Max-Born-Strasse 2A, 12489 Berlin
4Max-Planck-Institute of Microstructure Physics, Weinberg 2, 06120 Halle (Saale)
5Fakultät für Physik, Ludwig-Maximilians-Universität München, Am Coulombwall 1, 85748 Garching, Germany
*Correspondence to: [email protected]
The enigmatic coupling between electronic and magnetic phenomena was one of the riddles
propelling the development of modern electromagnetism1. Today, the fully controlled
electric field evolution of ultrashort laser pulses permits the direct and ultrafast control of
electronic properties of matter and is the cornerstone of light-wave electronics2 -- 7. In sharp
contrast, because there is no first order interaction between light and spins, the magnetic
properties of matter can only be affected indirectly on the much slower tens-of-femtosecond
timescale in a sequence of optical excitation followed by the rearrangement of the spin
structure8 -- 16.
Here we record an orders of magnitude faster magnetic switching with sub-femtosecond
response time by initiating optical excitations with near-single-cycle laser pulses in a
ferromagnetic layer stack. The unfolding dynamics are tracked in real-time by a novel
attosecond time-resolved magnetic circular dichroism (atto-MCD) detection scheme
revealing optically induced spin and orbital momentum transfer (OISTR) in synchrony with
light field driven charge relocation17. In tandem with ab-initio quantum dynamical
modelling, we show how this mechanism provides simultaneous control over electronic and
magnetic properties that are at the heart of spintronic functionality. This first incarnation of
attomagnetism observes light field coherent control of spin-dynamics in the initial non-
dissipative temporal regime and paves the way towards coherent spintronic applications
with Petahertz clock rates.
1
While matter reacts to optical excitations by a virtually instantaneous and direct response of
electrons to the electric field oscillations, ultrafast magnetic switching is intrinsically slower due
to the need of mediating processes linking optical excitations to spin dynamics8,9,13,14.
Femtosecond laser pulses can modify magnetic properties at response times of several tens of
femtoseconds or longer (50 − 500 x 10−15s)15,16, the prospect of optical control of ferromagnetic
bits has been discussed18 and circularly polarized attosecond light pulses are proposed to generate
ultrafast magnetic field bursts as new tools for ultrafast magneto-optics19. However, direct
experimental evidence of the ability to link the response of a spin system to the quasi-instantaneous
opto-electronic response calls for attosecond temporal resolution and is missing to date.
A route to the novel regime of attosecond magnetism has been worked out theoretically and
proposes the use of ferromagnetic alloys or layer stacks where optical excitations result in the local
displacement of charge carriers between different atomic species or across layer interfaces, in
analogy to shift currents discussed in compound semiconductors20,21. In such a scenario, dubbed
OISTR, the spatially dislodged electron wave carries its spin away from the atomic species at
which its ground state resides17. Consequently, the resulting coherent spin transfer is linked
directly to the temporal evolution of the optical excitation field and modifies the spin moment of
the magnetic layers extended over macroscopic dimensions defined by the illuminated area.
Critical for advancing towards attosecond magnetism is an experimental scheme with the ability
to control and detect the sub-optical-cycle evolution of electronic excitations driven by ultrafast
electric fields, combined with simultaneous observation of modifications of the magnetic moment
of individual sample constituents. Here we achieve this by linking attosecond transient absorption
detection (atto-XAS), which is sensitive to the temporal evolution of electronic excitations2 -- 7, with
simultaneous atto-MCD sketched in Fig. 1, which is probing the local magnetic moment.
2
In our experiment, ultrafast near-infrared (NIR) few-cycle laser pulses serve as temporally well
confined trigger of electronic excitations in nickel. The metal exhibits broadband resonant
absorption properties (see density of states DOS , Fig. 1 c) and the strong electric field of the laser
pulse accelerates electrons around the Fermi energy in the band structure or promotes them to
states in higher energy bands. Similarly, for nickel sandwiched between platinum layers the optical
excitation is a virtually instantaneous response to the external electric field oscillations. However,
the adjacent heavy metal ad-layers open an additional excitation pathway where an electron is
coherently driven across the material interface from states around the Fermi energy of nickel into
vacant platinum states (Fig. 1 c, d).
The resulting depletion of majority spins oriented by placement of the samples in a static external
B-field well beyond magnetic saturation (𝐵𝑒𝑥𝑡 = 50 mT) leads to a macroscopic reduction of the
nickel layer magnetic moment. We employ atto-MCD as illustrated in Fig. 1 for the first real time
study of this light field induced coherent modification of a materials' magnetic moment.
Linearly polarized attosecond pulses are generated by the same ultrafast laser pulses used to pump
the electronic system of the samples via high-harmonic-generation and subsequent spectral
selection22 -- 24. To implement a time resolved variant of x-ray magnetic circular dichroism
detection25, we developed a grazing incidence, multi reflection phase-retarder26,27 (Fig. 1 a)
optimized for photon energies in the extreme ultraviolet (XUV) regime. Rotation of the phase-
retarder around the beam propagation axis allows either to transmit the initially linear polarized
light or to create circularly polarized attosecond pulses (see Methods).
In our experiment, the resulting circularly polarized attosecond pulses cover XUV photon energies
around the Nickel M2,3-transition (Fig. 1) and last ~20% of the half-cycle duration of the optical
pump field (𝜏𝑋𝑈𝑉 = 310 as) as confirmed by attosecond streak camera detection.
3
In a transient absorption scheme, these pulses permit to measure the dynamics of photo-excited
electrons and modifications of the spin structure. Final state interactions between electrons photo-
injected into conduction band (CB) states and electrons promoted into the CB by an XUV photon
turn the energy resolved XUV absorption into a sensitive probe for pump laser induced population
transfer between electronic states around the Fermi edge of Nickel28.
Recording the transmitted XUV spectral intensities 𝐼+/− for two opposite magnetization directions
oriented along the sample surface (or equivalently for two orthogonal helicities of the attosecond
pulses) yields the magnetic dichroism contrast Δ𝑀 = 𝐼+ − 𝐼− (atto-MCD, see Fig. 1 b, and
Methods). Detection of Δ𝑀 renders the method sensitive to element specific local magnetic
moments and allows the simultaneous tracking of sub-femtosecond time-resolved modifications
of both the electronic and magnetic implications of strong field induced optical pumping.
Fig. 2 shows the recorded atto-XAS and atto-MCD transients. Ultrashort NIR waveforms carried
at a central wavelength of 𝜆𝑁𝐼𝑅 = 800 nm, set to a peak intensity of 𝐼𝑁𝐼𝑅 = 4 × 1012 W
⁄
cm2
and
full-width-at-half-intensity-maximum duration of 𝜏𝑁𝐼𝑅 = 4 fs are used to excite electrons in nickel
(panel b, sample I, for details on sample preparation and static characterization see Methods).
Owing to the attosecond temporal resolution of the measurement, in synchrony with the electric
field oscillations (simulated waveform depicted as green line, top of panel a) a repetitive decrease
of the XUV absorbance is observed (red line)3. The attosecond pulse experiences this increased
absorption (energy interval Δ𝐸 = 65.5 − 66.5 eV) as a result of reduced Pauli blocking in
electronic states that gradually are liberated by the pump field induced promotion of carriers into
states above the Fermi energy. The stepwise behavior is indicative of the transition rate peaking at
the field crests of the light waveform and the number of excited carriers increasing in synchrony
4
with the half-cycle oscillations of the NIR pump laser pulses, reminiscent of interband
tunneling3,5,29.
Inspection of the simultaneously recorded atto-MCD contrast Δ𝑀 (blue line) yields no observable
change within the sampled time interval, indicating the conservation of magnetic moment in the
ferromagnetic nickel layer for the first ~10 fs during and after electronic excitation.
In striking contrast, panel a shows the magnetic moment of nickel sandwiched between platinum
layers (sample II, see Methods) under otherwise identical experimental conditions to respond as
fast as the electronic system to the optical excitation. Platinum surrounding nickel acts as efficient
spin absorber for optical excitations transferring charge across the interfaces. In these samples,
ultrafast changes of Δ𝑀 (blue line) > 40% of the initial value are immediately discernible during
the light field oscillations that cause optical excitation, attesting to the dominant role of OISTR in
the loss of magnetic moment. The reduction acts out with a time constant of 𝜏𝑐𝑜ℎ𝑒𝑟𝑒𝑛𝑡 =
4.5 fs ( 1
𝑒⁄ ), concurrent with the field induced excitation and identical to the response function
computed from the laser pulse duration. This synchrony is direct evidence for coherent, sub-
femtosecond magnetization control extended over the macroscopic dimensions probed by the
attosecond pulse spot size ∅𝑋𝑈𝑉 ≈ 40μm.
To shed light on the coherent electronic processes in the presence of the oscillating light field and
its influence on initial stages of ultrafast spin dynamics, we turn to theory. Fig. 3 compares the
experimental data with the theoretical results obtained using a state-of-the-art, full quantum
description of the dynamics of the non-equilibrium state of matter by means of time-dependent
density functional theory (TD-DFT), which accounts for the coupled dynamics of both, charge and
spins. The theory identifies the two major phenomena that contribute to the dynamics in the early
times (during and immediately after pumping) to be the flow of charge and spin currents across
5
the interface and spin-orbit-mediated spin-flips. The net result of which is coherent momentum
transfer (CMT) in space and time across the interface by redistribution of the interlinked spin S(t)
and orbital angular momentum L(t).
The combination of TD-DFT and atto-MCD, both capable of tracking ultrafast charge and spin
migration that outpace the slower lattice dynamics, is therefore a powerful framework to
investigate how different mechanisms conspire to turn the initially coherent optical excitation into
a macroscopic loss of magnetic moment. Fig. 3a shows the results of atto-MCD recording the
magnetization dynamics in Ni/Pt sandwich samples pumped at an intensity of 𝐼𝑁𝐼𝑅 = 2 ×
1012 W
⁄
cm2
(green line w. circles) in comparison with the results of ab-initio computations with
the sample geometry and experimental laser pulse parameters as input. Excellent quantitative
agreement between experimentally observed ∆𝑀 and the results of the full calculation (including
spin-orbit coupling) of the time evolution of the relative magnetic moment 𝑀(𝑡) 𝑀(0)
⁄
testifies
for the validity of the theoretical approach and confirms that the experimental observable is a direct
measure of the layer magnetization. Interestingly, excluding spin-orbit coupling from the
theoretical calculations (dark red line) causes a leveling off from the predicted magnetization
dynamics at 20 fs, which is not mirrored in the experimental data. Since the only process that can
cause a demagnetization of Ni apart from spin-orbit coupling is the flow of electrons (carrying
their spins) across the interface, these first 20 fs mark the hitherto unexplored time interval in
which OISTR-mediated CMT dictates the physics of spin-dynamics.
Our data indicate the presence of a second time scale of slower demagnetization (20 fs < 𝑡 <
50 fs) where theory identifies spin-orbit-mediated spin-flips to dominate the evolution of the
magnetic moment. Later after optical excitation, (𝑡 > 50 fs, panel b) the sample fully
demagnetizes with a timescale of 𝜏𝑖𝑛𝑐𝑜ℎ𝑒𝑟𝑒𝑛𝑡 = 112 ± 6fs which represents the transition to
6
stochastic dynamics rooted in magnetically correlated behavior and is in agreement with previous
experiments reporting the decay of magnetic moment in the range of 80 -- 400 fs depending on
system dependent quench rates10,12.
We observed light-wave induced coherent transfer of spin and orbital angular momentum in space
and time caused by the interplay of the few-cycle optical excitation and the spin-orbit interaction
in magnetic/non-magnetic multilayers. This excursion into the unexplored territory of ultrafast
(𝑡 < 20 fs) all-optical control of spin dynamics and macroscopic magnetic moments by and in
synchrony with the oscillations of ultrafast electric fields opens the door to the novel regime of
attosecond magnetism and is a benchmark for the design of future coherent magnetic control
protocols. Our ab initio theory predicts that in alloys or suitably chosen multilayer systems the
same mechanism can be tailored to cause a local and ultrafast increase in magnetic moment to
sustain reversible optical switching of magnetic moments bringing spintronics to the attosecond
regime. Finally, the observed sub-femtosecond magnetization control via optically induced spin
transfer, promises a new class of ultrafast spintronic applications, and suggests the feasibility of
light wave gated coherent spin transistors mimicking spintronic functionality orders of magnitude
faster than applicable today.
Methods:
Theory:
Computations rely on the Runge-Gross theorem30 establishing that the time-dependent external
potential is a unique functional of the time dependent density, given the initial state. Based on this
theorem, a system of non-interacting particles can be chosen such that the density of this non-
7
interacting system is equal to that of the interacting system for all times. The wave function is
represented as a Slater determinant of single-particle Kohn-Sham (KS) orbitals. A fully non-
collinear spin-dependent version of this theorem entails that these KS orbitals are two-component
Pauli spinors determined by the equations:
𝑖
𝜕𝜓𝑗(𝒓, 𝑡)
𝜕𝑡
= [
1
2
(−𝑖∇ +
2
𝑨𝑒𝑥𝑡(𝑡))
1
𝑐
+ 𝑣𝑠(𝒓, 𝑡) +
1
2𝑐
𝜎 ∙ 𝑩𝑠(𝒓, 𝑡) +
1
2𝑐
𝜎 ∙ (∇𝑣𝑠(𝒓, 𝑡) × −𝑖∇)] 𝜓𝑗(𝒓, 𝑡)
(1)
where the first term is the kinetic term and responsible for the flow of current across the interface31,
𝑨𝑒𝑥𝑡(𝑡) is a vector potential representing the applied laser field, and 𝜎 are the Pauli matrices. The
KS effective potential 𝑣𝑠(𝒓, 𝑡) = 𝑣𝑒𝑥𝑡(𝒓, 𝑡) + 𝑣𝐻(𝒓, 𝑡) + 𝑣𝑋𝐶(𝒓, 𝑡) is decomposed into the external
potential 𝑣𝑠(𝒓, 𝑡), the classical electrostatic Hartree potential 𝑣𝐻(𝒓, 𝑡) and the exchange-correlation
(XC) potential 𝑣𝑋𝐶(𝒓, 𝑡). Similarly, the KS magnetic field is written as 𝑩𝑠(𝒓, 𝑡) = 𝑩𝑒𝑥𝑡(𝒓, 𝑡) +
𝑩𝐻(𝒓, 𝑡) + 𝑩𝑋𝐶(𝒓, 𝑡) where 𝑩𝑒𝑥𝑡(𝒓, 𝑡) is the magnetic field of the applied laser pulse plus possibly an
additional magnetic field and 𝑩𝑋𝐶(𝒓, 𝑡) is the XC magnetic field. The final term is the spin-orbit
coupling term, presence of which ensures that the total spin angular momentum, S, is not a good
quantum number and hence spin-flip excitations can lead to loss in S. Eq. 1 is solved for the
electronic system alone in dipole approximation and by using adiabatic local spin density
approximation32 for the XC fields.
Dynamics of the nuclear degrees of freedom and radiative effects described by simultaneously
time-propagating Maxwell's equations, are not included in the present work limiting the predictive
power of the method to the first ~100 fs during and after pumping.
Calculations of a magneto-optical function for NiPt layer (see inset Fig. 3) were performed by a 3
step process: (i) The ground-state of NiPt multilayers was determined using DFT, (ii) a fully spin-
polarized GW calculation33 was performed to determine the correct position and width of Ni 3p
states and (iii) finally the response function was calculated on top of GW-corrected Kohn-sham
8
ground-state. This response function is calculated within linear response TD-DFT in which the
excitonic effects34 and local field effects can be easily included. As a merit of this treatment, no
experimental parameter was needed to determine the accurate magneto-optical functions.
Computational details:
A fully non-collinear version of TDDFT as implemented within the Elk code35 is used for all
calculations. All the implementations are done using the state-of-the art full potential linearized
augmented plane wave (LAPW) method.
A regular mesh in 𝒌-space of 8 × 8 × 1 is used and a time step of ∆𝑡 = 2 as is employed for the
time-propagation algorithm. To mimic experimental resolution, a Gaussian energy broadening is
applied with spectral width of 0.027 eV. The laser pulse used in the present work is linearly
polarized (out of plane polarization) with central frequency of 1.55 eV, full-width-at-half-
maximum duration 8fs and fluence of 5.4 mJ/cm².
Experiment:
The experiments are carried out with a phase-stabilized few-cycle near-infrared laser as driver
(FemtoPower). It delivers pulses with 4fs, 0.5mJ at a repetition rate of 4kHz. These pulses are
focused into a neon-filled ceramics target, generating high-harmonic radiation with a cut-off
energy around 70 eV. A Mach-Zehnder type interferometer is used to introduce a delay between
the NIR and the XUV pulses. We use a 150nm Al filter to block the driving NIR laser pulse, while
providing a constant transmission at the energies of interest. Isolated attosecond pulses are
achieved by spectrally filtering the cut-off regime with a Si/B4C multilayer mirror with a reflection
of 15% @66eV and a full-width-at-half-maximum bandwidth of 8eV under and angle of incidence
of 45°. Subsequently, the multi reflection phase-shifter changes the polarization from linear to
9
elliptical. Ellipticity of ε>0.75 is achieved with transmittance of >25% and without introducing
significant wavelength dispersion,
Four Mo/B4C multilayer mirrors are mounted under an angle of 78° with respect to the surface
normal, a setting optimized for high broadband transmission and maximized ellipticity.
To maximize the atto-MCD contrast while maintaining sufficient XUV transmission the samples
are mounted under an angle of ~35° between the propagation direction of the laser and the surface
normal. The surface normal and the magnetization direction of the sample span a plane parallel to
the propagation direction of the laser field. We use a gold coated grating with 2105 lines/mm in
reflection (Jobin-Yvon) with a 200 µm entrance slit as a spectrometer.
For every delay step, we measured the transmitted spectrum for both magnetization directions of
the sample and then scanned the delay over the region of interested. To check and compensate for
long timescale drifts in the XUV spectrum we took spectra of the transmitted XUV in the absence
of NIR laser light before and after every scan. to ensures that no irreversible changes were done
to the multilayer or the supporting polycrystalline Silicon substrate. Typical integration times were
on the order of 10-15s for every spectrum, leading to a measurement duration of 30-60 minutes.
We observe the OISTR effect for pump light intensities ranging between 𝐼𝑁𝐼𝑅 = 1.5 − 4 ×
1012 W
⁄
cm2
. At lower intensities, only the slow demagnetization component depicted in Fig. 3,
panel b is observed. At intensities 𝐼𝑁𝐼𝑅 > 5 × 1012 W
⁄
cm2
the time span between laser pulses
(250 µs) is found to be insufficient for the magnetic system to return to the unperturbed state.
Data Evaluation:
For the energy calibration of our spectrum we used the pronounced aluminum L2,3-edges at 72.7
eV and the Nickel M3-edge at 66.2 eV. The grating equation 𝑛 ∗ 𝜆 = 𝑑 ∗ (sin 𝜃𝑖 − sin 𝜃𝑓) gives a
10
relation between the position of a spectral feature on the XUV sensitive camera (Princeton
Instrument PIXIS) and the photon energy. We checked for the influence of the non-homogeneous
diffraction and reflection efficiency of the grating and found it to be negligible within the range of
photon energies relevant in this study. We applied an equal weight sliding average to the
spectrograms along the energy (ΔE = 50 meV) as well as along the delay dimension (Δt = 0.5 fs).
We determine the magnetic dichroism contrast according to Δ𝑀 = 𝐼+ − 𝐼−. Application of
alternative definitions of the contrast or computing the magnetic dichroism asymmetry has no
influence on the transient signals.
Sample preparation and magnetic characterization:
A number of Ni and Ni/Pt multilayer films were grown on Si membranes and optimized to grow
in fcc structure. The 8.4 nm Ni film and Pt(2)/[Ni(4)/Pt(2)]×2/Ni(2nm) multilayer were deposited
on 200 nm thick polycrystalline Silicon membranes (Norcada) at room temperature by electron
beam evaporation under ultrahigh vacuum (UHV) conditions with a base pressure less than 1x
10−9 mbar with high purity source materials (99.95%). The individual layer thicknesses in the
[Ni(4)/Pt(2)]×2 stack were optimized to increase the number of active interfaces for OISTR while
maintaining sufficient XUV transmittance and providing maximized atto-MCD contrast. The
sample was protected by an inert 2 nm Pt capping layer and an ultrathin Ni seed layer (nonmagnetic
at room temperature) was grown on the substrate to provide identical surrounding to all Pt layers
and ensure uniform growth of the layer stack. Both materials grow in fcc structure. The deposition
rate was monitored by a calibrated standard quartz crystal microbalance and adjusted to 0.02 nm/s.
Special care was taken for low tension mounting to protect the membranes from any mechanical
stress. The in-plane magnetic properties were characterized by a standard longitudinal Magneto-
11
Optical Kerr Effect (MOKE) setup with a 3 mW HeNe cw-laser, a photoelastic modulator (PEM)
and Glan-laser prisms in longitudinal geometry. Measurements on both, the frame and the samples
membrane window, yield similar ferromagnetic properties. For both types of samples, we found
coercive fields of a few mT and a remanence around 60-70% of the saturation magnetization.
MOKE measurements confirm that the Ni seed layer does not magnetize at room temperature and
thus the layer does not affect the magnetic moment of the sample. The recorded MOKE signal also
confirms that the ferromagnetic properties of the pure Ni layer does not suffer from an assumed
oxidic passivation of the sample surface.
12
Fig. 1: Atto-MCD to study coherent ultrafast magnetism
A quarter wave phase retarder/band-pass filter optimized for photon energies covering the M-edge
absorption of the most transition metals is used to turn incident linearly polarized attosecond pulse
13
trains (APT) into isolated circular polarized attosecond pulses (panel a). These pulses are applied
to record the magnetic circular dichroism contrast of nickel/platinum multilayer systems pumped
by few-cycle near-infrared waveforms in an external magnetic field. Before the few-cycle near-
infrared excitation, the nickel/platinum multilayer and the bulk nickel are in the magnetically
saturated state along the in-plane direction easy axis set by ± 𝑩.
Panel b displays the recorded absorption A+ and A- for the two orientations of the magnetic field
applied to the sample and the resulting atto-MCD contrast: before the arrival of the NIR pulse (t1)
which represents the unperturbed system, shortly after its arrival (t2) and hundreds of femtoseconds
after excitation (t3).
In the multilayer (as opposed to bulk nickel) the optical excitation causes a coherent trans-interface
charge current driven by the few-cycle light pulse, associated with which is the synchronous
transfer of spin into the heavy metal (panel c), where the spin is subjected to the strong spin-orbit
coupling of the heavy metal ad-layer (panel d).
14
Fig. 2: Attosecond-MCD of light-field induced coherent spin transfer
The strong electric field 𝐸(𝑡) (waveform fitted to an attosecond streak camera recording, green
line top of panel a) of an ultrashort laser pulse excites electrons around the Fermi energy in Ni/Pt-
multilayer (panel a) samples and bulk nickel (panel b). The resulting transfer of electronic
population in synchrony with the laser fields' half cycle oscillations is tracked with atto-XAS and
yields a stepwise modification of the XUV transmission evaluated at 66 ± 0.5 eV (red-dashed
curve in panel a and b). The simultaneously recorded atto-MCD contrast evaluated in the energy
interval of 66.5 − 68 eV, around the nickel M2,3-edge is a measure of the magnetization of the
nickel layer (blue solid curve in panel a and b). The reduction of the magnetic moment of the
multilayer system synchronously with the electronic response is clear experimental evidence for
ultrafast coherent spin transfer and the OISTR effect. This is evidenced further by bulk nickel
showing no noticeable change in magnetic moment in the first femtoseconds after excitation (panel
b), while also exhibiting an electronic response linked to the electric field oscillations of the
excitation pulse.
15
Fig. 3: Optically induced spin-transfer, post-excitation spin dynamics and quantum
dynamical modelling
Comparison between the experimentally recorded atto-MCD trace (green line w. circles) in Ni/Pt-
multilayers with the ab-initio simulation of light-field induced spin dynamics including spin-orbit
coupling (dark-blue line, panel a) reveals that for the first 10 fs the demagnetization of Ni layers
is entirely due to flow of majority spin current across the material interface. Theory without spin-
orbit interaction predicts a saturation of the demagnetization already 15 fs after optical excitation
(light-blue line). This is a clear indication that beyond this time all the demagnetization is caused
by spin flips while at earlier times spin-dynamics are governed by OISTR -- absent in the
theoretical prediction for bulk nickel (dashed line and cp. Fig 2). Experimental data recorded for
longer timescales (panel b) indicate the complete quenching of the magnetic moment due to
16
stochastic processes with a characteristic timescale of 112 fs. The recorded Δ𝑀 exhibits additional
modulations up to 300 fs after optical excitation, that are potentially due to coherent phonon modes
affecting the coupling to the lattice. The inset compares the recorded MCD signal ∆= log
𝐼+
𝐼− of
the Ni/Pt sandwich with the imaginary part of the computed magneto-optical function.
References
1.
2.
3.
Ørstedt, J. C. Experimenta circa effectum Conflictus electrici in Acum magneticam. J. für
Chemie und Phys. 29, 275 -- 281 (1820).
Schiffrin, A. et al. Optical-field-induced current in dielectrics. Nature 493, 70 -- 4 (2013).
Schultze, M. et al. Attosecond band-gap dynamics in silicon. Science 346, 1348 -- 1352
(2014).
4. Mashiko, H., Oguri, K., Yamaguchi, T., Suda, A. & Gotoh, H. Petahertz optical drive with
wide-bandgap semiconductor. Nat. Phys. 12, 741 -- 745 (2016).
5.
6.
7.
8.
9.
Lucchini, M. et al. Attosecond dynamical Franz-Keldysh effect in polycrystalline
diamond. Science 353, 916 -- 9 (2016).
Garg, M. et al. Multi-petahertz electronic metrology. Nature 538, 359 -- 363 (2016).
Reimann, J. et al. Subcycle observation of lightwave-driven Dirac currents in a
topological surface band. Nature 562, 396 -- 400 (2018).
Bigot, J.-Y., Vomir, M. & Beaurepaire, E. Coherent ultrafast magnetism induced by
femtosecond laser pulses. Nat. Phys. 5, 515 -- 520 (2009).
Boeglin, C. et al. Distinguishing the ultrafast dynamics of spin and orbital moments in
solids. Nature 465, 458 -- 461 (2010).
10. Walowski, J. & Münzenberg, M. Perspective: Ultrafast magnetism and THz spintronics. J.
Appl. Phys. 120, 140901 (2016).
11. Koopmans, B. et al. Explaining the paradoxical diversity of ultrafast laser-induced
demagnetization. Nat. Mater. 9, 259 -- 265 (2010).
12. Hellman, F. et al. Interface-induced phenomena in magnetism. Rev. Mod. Phys. 89,
025006 (2017).
13. Kirilyuk, A., Kimel, A. V. & Rasing, T. Ultrafast optical manipulation of magnetic order.
Rev. Mod. Phys. 82, 2731 -- 2784 (2010).
14. Battiato, M., Carva, K. & Oppeneer, P. M. Superdiffusive Spin Transport as a Mechanism
of Ultrafast Demagnetization. Phys. Rev. Lett. 105, 027203 (2010).
15.
Stamm, C. et al. Femtosecond modification of electron localization and transfer of angular
momentum in nickel. Nat. Mater. 6, 740 -- 743 (2007).
16. Rudolf, D. et al. Ultrafast magnetization enhancement in metallic multilayers driven by
superdiffusive spin current. Nat. Commun. 3, 1037 (2012).
17
17. Dewhurst, J. K., Elliott, P., Shallcross, S., Gross, E. K. U. & Sharma, S. Laser-Induced
Intersite Spin Transfer. Nano Lett. 18, 1842 -- 1848 (2018).
18. Lambert, C.-H. et al. All-optical control of ferromagnetic thin films and nanostructures.
Science 345, 1337 -- 40 (2014).
19. Bandrauk, A. D., Guo, J. & Yuan, K.-J. Circularly polarized attosecond pulse generation
and applications to ultrafast magnetism. J. Opt. 19, 124016 (2017).
20. Côté, D., Laman, N. & van Driel, H. M. Rectification and shift currents in GaAs. Appl.
Phys. Lett. 80, 905 -- 907 (2002).
21. Laman, N., Bieler, M. & van Driel, H. M. Ultrafast shift and injection currents observed in
wurtzite semiconductors via emitted terahertz radiation. J. Appl. Phys. 98, 103507 (2005).
22. Hentschel, M. et al. Attosecond metrology. Nature 414, 509 -- 13 (2001).
23.
24.
Schweinberger, W. et al. Waveform-controlled near-single-cycle milli-joule laser pulses
generate sub-10 nm extreme ultraviolet continua. Opt. Lett. 37, 3573 -- 5 (2012).
Fiess, M. et al. Versatile apparatus for attosecond metrology and spectroscopy. Review of
Scientific Instruments 81, 093103 (2010).
25. Carra, P., Thole, B. T., Altarelli, M. & Wang, X. X-ray circular dichroism and local
magnetic fields. Phys. Rev. Lett. 70, 694 -- 697 (1993).
26. Höchst, H., Patel, R. & Middleton, F. Multiple-reflection λ4 phase shifter: A viable
alternative to generate circular-polarized synchrotron radiation. Nucl. Instruments
Methods Phys. Res. Sect. A Accel. Spectrometers, Detect. Assoc. Equip. 347, 107 -- 114
(1994).
27. Willems, F. et al. Probing ultrafast spin dynamics with high-harmonic magnetic circular
dichroism spectroscopy. Phys. Rev. B 92, 220405 (2015).
28. Kaindl, G., Brewer, W. D., Kalkowski, G. & Holtzberg, F. M -Edge X-Ray Absorption
Spectroscopy: A New Tool for Dilute Mixed-Valent Materials. Phys. Rev. Lett. 51, 2056 --
2059 (1983).
29. Ghimire, S. et al. Strong-field and attosecond physics in solids. J. Phys. B At. Mol. Opt.
Phys. 47, 204030 (2014).
30. Runge, E. & Gross, E. K. U. Density-Functional Theory for Time-Dependent Systems.
Phys. Rev. Lett. 52, 997 -- 1000 (1984).
31. Krieger, K. et al. Ultrafast demagnetization in bulk versus thin films: an ab initio study. J.
Phys. Condens. Matter 29, 224001 (2017).
32. Barth, U. von & Hedin, L. A local exchange-correlation potential for the spin polarized
case. I. J. Phys. C Solid State Phys. 5, 1629 -- 1642 (1972).
33. Hedin, L. New Method for Calculating the One-Particle Green's Function with
Application to the Electron-Gas Problem. Phys. Rev. 139, A796 -- A823 (1965).
34.
Sharma, S., Dewhurst, J. K., Sanna, A. & Gross, E. K. U. Bootstrap Approximation for the
Exchange-Correlation Kernel of Time-Dependent Density-Functional Theory. Phys. Rev.
Lett. 107, 186401 (2011).
35. Dewhurst, K. et al. The Elk FP-LAPW Code. Http://Elk.Sourceforge.Net/ (2018).
18
Extended Data
Extended Data Figure 1: Hysteresis curves of the samples recorded using the longitudinal
magneto-optical Kerr effect. Both samples exhibit soft magnetic hysteresis and low saturation
fields needed to orient the macroscopic magnetization in the sample plane.
19
|
1912.05354 | 1 | 1912 | 2019-12-09T19:23:46 | Kinetic modelling of carrier cooling in lead halide perovskite materials | [
"physics.app-ph"
] | The relaxation of high-energy "hot" carriers in semiconductors is known to involve the redistribution of energy between (i) hot and cold carriers and (ii) hot carriers and phonons. Over the past few years, these two processes have been identified in lead-halide perovskites (LHPs) using ultrafast pump-probe experiments, but the interplay between these processes is not fully understood. Here we present a comprehensive kinetic model to elucidate the individual effects of the hot and cold carriers in bulk and nanocrystal $CsPbBr_{3}$ films obtained from "pump-push-probe" measurements. In accordance with our previous work, we observe that the cooling dynamics in the materials decelerate as the number of hot carriers increases, which we explain through a "hot-phonon bottleneck" mechanism. On the other hand, as the number of cold carriers increases, we observe an acceleration of the cooling kinetics in the samples. We describe the interplay of these opposing effects using our model, and by using series of natural approximations, reduce this model to a simple form containing terms for the carrier-carrier and carrier-phonon interactions. The model can be instrumental for evaluating the details of carrier cooling and electron-phonon couplings in a broad range of LHP optoelectronic materials. | physics.app-ph | physics | Kinetic modelling of carrier cooling in lead halide perovskite materials
Kinetic modelling of carrier cooling in lead-halide perovskite materials
Thomas R. Hopper,1, a) Ahhyun Jeong,1, a) Andrei Gorodetsky,1 Franziska Krieg,2, 3 Maryna I. Bodnarchuk,2, 3
Xiaokun Huang,4 Robert Lovrincic,4 Maksym V. Kovalenko,2, 3 and Artem A. Bakulin1, b)
1)Ultrafast Optoelectronics Group, Department of Chemistry, Imperial College London, London W12 0BZ,
United Kingdom
2)Laboratory of Inorganic Chemistry, Department of Chemistry and Applied Biosciences, ETH Zurich, Vladimir-Prelog-Weg 1-5/10,
8093 Zurich, Switzerland
3)Laboratory for Thin Films and Photovoltaics, Empa -- Swiss Federal Laboratories for Materials Science and Technology,
Überlandstrasse 129, CH-8600 Dübendorf, Switzerland
4)Institute for High-Frequency Technology, Technische Universität Braunschweig, Schleinitzstrasse 22, 38106, Braunschweig,
Germany
The relaxation of high-energy "hot" carriers in semiconductors is known to involve the redistribution of energy between
(i) hot and cold carriers and (ii) hot carriers and phonons. Over the past few years, these two processes have been
identified in lead-halide perovskites (LHPs) using ultrafast pump-probe experiments, but the interplay between these
processes is not fully understood. Here we present a comprehensive kinetic model to elucidate the individual effects
of the hot and cold carriers in bulk and nanocrystal CsPbBr3 films obtained from "pump-push-probe" measurements.
In accordance with our previous work, we observe that the cooling dynamics in the materials decelerate as the number
of hot carriers increases, which we explain through a "hot-phonon bottleneck" mechanism. On the other hand, as the
number of cold carriers increases, we observe an acceleration of the cooling kinetics in the samples. We describe the
interplay of these opposing effects using our model, and by using series of natural approximations, reduce this model to
a simple form containing terms for the carrier-carrier and carrier-phonon interactions. The model can be instrumental
for evaluating the details of carrier cooling and electron-phonon couplings in a broad range of LHP optoelectronic
materials.
I.
INTRODUCTION
Lead-halide perovskites (LHPs) are rapidly emerging as a
promising class of semiconducting material for future opto-
electronic applications. Since the first report of 3.8% power
conversion efficiency (PCE) in 2009,1 laboratory-scale LHP
solar cells can now routinely achieve PCEs above 20%, with
a record PCE of ∼25%.2 Compared to the top PCEs of other
third-generation solution-processable photovoltaic materials,
this value is impressively close to the ∼33% PCE given by
the theoretical "Shockley-Queisser" limit for single-junction
solar cells.3 A key factor underpinning this limit is the rapid
dissipation of energy from "hot" carriers following the pho-
toexcitation of the semiconductor above its bandgap. Early
work by Ross and Nozik postulated that the performance of
a solar cell could be dramatically enhanced if the hot carri-
ers could be utilised.4 The efficiency of this process depends
on how quickly hot carriers dissipate their excess energy, re-
flected by the cooling rate.
The cooling rate depends on many processes which con-
trol the carrier dynamics, including the interaction between
carriers (i.e. redistribution of heat between the carriers, some-
times called thermalisation), and the interactions between car-
riers and phonons (i.e. the removal of excess energy through
lattice vibrations).5 The latter process has been studied ex-
tensively in LHPs using pump-probe methods. A common
finding is a slowdown of the cooling kinetics with increas-
ing carrier density.6 -- 12 This phenomenon has also been ob-
served in traditional semiconductors such as GaAs,13,14 and is
a)These authors contributed equally to the manuscript
b)Corresponding author email: [email protected]
often referred to as the "hot-phonon bottleneck". For LHPs,
this behaviour is thought to occur when polarons (i.e.
lo-
calised distortions of the lattice surrounding a carrier) spa-
tially overlap, leading to the reabsorption of hot phonons by
adjacent carriers.15 We previously demonstrated that this be-
haviour in LHPs is sensitive to the material composition, and
is particularly pronounced in the all-inorganic CsPbBr3 sys-
tem compared to its hybrid counterparts with vibrationally ac-
tive organic cations.16 Most commercially available transient
absorption methods do not have the time resolution required to
observe the aforementioned thermalisation process, but high-
resolution two-dimensional electronic spectroscopy has been
employed to demonstrate this effect in LHPs.17,18 These stud-
ies show that the high-energy carriers produced by above-gap
excitation reach a thermal distribution through inelastic carrier
scattering at the early stage of cooling (<100 fs), followed by
the carrier-phonon interactions where the heat is dissipated in
the lattice (∼500 fs). Although these reports provide valuable
insight into the mechanisms of carrier relaxation, the methods
lack the ability to distinguish the individual effects of the hot
and cold carriers on carrier cooling.
Here, we propose a comprehensive kinetic model to study
the effect of the number of hot and cold carriers on the car-
rier cooling kinetics in CsPbBr3 bulk and nanocrystal (NC)
films determined by "pump-push-probe" measurements. Our
results convey two competing pathways for carrier cooling
based on the dissipation of the excess hot carrier energy into
either optical phonons or cold carriers. By applying simple
approximations, we reduce this model into a single equation
containing terms for the carrier-carrier and carrier-phonon in-
teractions. At sufficiently high cold carrier concentration, the
interactions between hot and cold carriers dominate over the
hot-phonon bottleneck behaviour, leading to an overall in-
crease in the rate of carrier cooling.
9
1
0
2
c
e
D
9
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
4
5
3
5
0
.
2
1
9
1
:
v
i
X
r
a
Kinetic modelling of carrier cooling in lead halide perovskite materials
2
the regenerative amplifier. Sum frequency generation yielded
the 500 nm ∼60 fs pump pulse. The idler output (2073 nm,
∼50 fs) of the other optical parametric amplifier was split in
a 9:1 ratio. The more intense portion was used as the push,
and the weaker portion was used as the probe. The pump and
probe beams were collinearly overlapped and focused onto a
∼0.2 mm diameter spot on the sample. The push was focused
less tightly (∼0.4 mm) to facilitate beam overlap and focused
in the same spot in non-collinear (∼5◦) geometry with respect
to the pump/probe. The transmission of the probe through
the sample was recorded with an amplified PbSe photodetec-
tor (PDA20H-EC, ThorLabs). This was connected to a lock-
in amplifier (SR830, Stanford Research Systems) coupled to
a chopper in the pump beam. The chopper operated at half
the repetition rate of the amplifier to block every other pump
pulse. The delay between the pump, push and probe beams
were controlled with mechanical delay stages. The sample
was kept in a nitrogen-purged quartz cuvette during measure-
ments.
III. EXPERIMENTAL RESULTS
The linear absorption spectra of the bulk and NC samples
is shown in Fig. 1. As expected, the NC sample exhibits a
slightly blueshifted absorption onset due to the effect of weak
quantum confinement.22 The NCs studied here have a cubic
shape and an average edge-length of 8±2 nm, as confirmed
by transmission electron microscopy (Fig. S1).
II. EXPERIMENTAL METHODS
A. Sample preparation
CaF2 substrates (12 mm diameter, 1 mm thickness,
EKSMA Optics) were first treated with O2 plasma to improve
wetting.
The bulk CsPbBr3 films were prepared by dissolving 1:1
mole ratio of PbBr2 (Alfa Aesar, 99.9%) and CsBr (Alfa Ae-
sar, 99.9%) in dimethyl sulfoxide (Sigma-Aldrich, 99.9%)
with concentration of 0.4 M and stirring at 50 ◦C overnight.
Afterwards, ∼35 µl of the as-prepared solution was dropped
onto the substrate and spin-coated at 3000 rpm for 2 min. The
solvent extraction method was then applied by using 300 µl
chloroform after 1 min spinning.19,20 The as-formed thin films
were then thermally annealed on a hot plate at 100 ◦C for
10 min inside the glovebox. The thin film layer thicknesses
were determined by UV-vis ellipsometry to be ∼300 nm.
The CsPbBr3 NCs were synthesised according to a pre-
viously reported method.21 1 mL lead-oleate (0.5 M in 1-
octadiene (ODE)), 0.8 mL Cs-oleate (0.4 M in ODE), 0.0445 g
of N,N-(dimethyloctadecylammonio)-propanesulfonate and
10 mL ODE were added to a 25 mL three-neck flask and
heated to 130 ◦C under vacuum. As soon as the reaction tem-
perature was reached, the atmosphere was changed to nitrogen
and 0.42 g of oleylammonium bromide, dissolved in 3 mL of
toluene, was injected. The reaction mixture was immediately
cooled to room temperature by means of an ice bath. The
crude solution was then centrifuged at 29500 g for 10 min.
The precipitate was discarded and ethylacetate (30 mL) was
added to the supernatant (ca. 15 mL). The mixture was cen-
trifuged at 29500 g for 10 min and the precipitate was dis-
persed in toluene (3 mL). This solution was subjected to three
additional rounds of precipitation with ethylacetate (6 mL)
and centrifugation at 29500 g for 1 min and subsequent re-
dispersion using toluene (3 mL). The film thickness was de-
termined to be ∼300 nm by spectroscopic ellipsometry.
B. UV-Vis spectroscopy
In order to determine the carrier density induced by the
pump, the absorption cross section of the samples was cal-
culated from the linear absorption spectra. These spectra
were obtained from a UV-Vis spectrometer (Shimadzu 2600,
equipped with an ISR-2600Plus integrating sphere). A 1 nm
sampling interval was used with a 5 nm slit width.
C. Ultrafast differential transmission measurements
A Ti:sapphire regenerative amplifier (Astrella, Coherent)
was used to pump two optical parametric amplifiers (TOPAS-
Prime, Coherent) with 800 nm pulses at a repetition rate
of 4 kHz and pulse duration of ∼35 fs. The signal out-
put (∼1300 nm) of one of the optical parametric amplifiers
was directed into a β -barium borate crystal (EKSMA Optics)
along with the residual part of the fundamental (800 nm) from
FIG. 1. UV-Vis absorption spectra of bulk and NC CsPbBr3.
To investigate the intraband cooling dynamics in the sam-
ples, we used a "visible pump-IR push-IR probe" pulse se-
quence, which has been outlined in previous works.16,23,24 As
depicted in Fig. 2, a 2.5 eV pump pulse generates excited
carriers which subsequently relax to the band edges. These
free carriers are monitored by the differential transmission of
a near-IR probe at 0.6 eV, which corresponds to an intraband
absorption in the LHPs.25 After a fixed delay (12 ps), an in-
tense 0.6 eV push pulse re-excites these cold states, thus trans-
forming them to hot carriers. The heating of the cold carriers
2.22.42.62.83.0Abs (norm.)Energy (eV) Bulk NC Kinetic modelling of carrier cooling in lead halide perovskite materials
3
FIG. 2. (a) Schematic for the pump-push-probe experiment. Note that the pump and probe are co-linear. (b) Three-level model (left) used to
interpret the differential probe transmission in response to the pump and push pulses (right).
is observed as a bleaching of the differential transmission sig-
nal. As the hot carriers cool, the pump-probe signal recovers.
We fit the recovery of the probe signal with a monoexponen-
tial curve to extract the cooling time, τcool, for the hot carriers.
The intensity of the pump and push pulses are used to con-
trol the cold and hot carrier densities, respectively. Fluence-
dependent pump-probe kinetics for the NC sample are shown
in Fig. S2.
The total number of (hot and cold) carriers per unit volume
in the system, nexc, is calculated using the Eq. (1) below. In
this equation, Einc is the incident energy per pump pulse, Eexc
is the energy per photon, AI is the absorbance of the sample
at the pump wavelength, R is the beam radius and d is the
sample thickness. The initial hot carrier density, nhot
, can be
0
obtained from the product of nexc and the ratio between the
amplitude of the push-induced bleach to the amplitude of the
signal before the arrival of the push pulse.
Einc(1− 10-AI)
π R2 d Eexc
nexc =
(1)
0
The dependence of τcool on the initial density of the hot and
cold carriers is outlined in Fig. 3. For a given value of ncold
(cold carrier density just before the push), we find that τcool
increases with increasing nhot
. This is in accordance with our
0
previous findings, and can be explained by the reduced rate
of carrier-phonon interactions through the hot-phonon bottle-
neck effect.16 Interestingly, we also find that τcool tends to
decrease with increasing nexc, suggesting that the hot-phonon
bottleneck effect plays less of a role in intraband cooling when
there are many cold carriers in the system. From this, we can
deduce that the carrier cooling occurs via at least two com-
peting pathways. When there are few cold carriers, carrier-
phonon interactions dominate over the hot-cold carrier inter-
actions, and the hot-phonon bottleneck effect primarily gov-
erns the carrier cooling dynamics. At higher cold carrier den-
sity, carrier cooling via carrier-carrier interactions starts to
dominate, resulting in the overall faster cooling and the re-
duced impact of the hot-phonon bottleneck effect.
FIG. 3. Plot of cooling time with respect to the hot carrier density
directly after the push (nhot
) and cold carrier density just before the
push (ncold
) in the CsPbBr3 NCs. Square points are experimental
data and the dotted lines are fits generated from the model described
in Section IV.
0
0
IV. KINETIC MODELLING
As discussed above, we interpret our experimental data as
the cooling of hot carriers in the LHP samples through carrier-
carrier and/or carrier-phonon interactions. To gain more in-
sight into the interplay between these processes, we have de-
veloped a numerical model describing each process that is
eventually reduced to a single equation, shown later. To fully
describe the data, we take the following approximations:
• Carriers can be either hot or cold, hence the sum of the
number of hot carriers (nhot) and the number of cold
carriers (ncold) is equal to the total number of carriers
(nexc);
12 psPush-probe delay (ps)-ΔT/T0-12𝜏coolPump2.5 eVPush0.6 eVProbe0.6 eVa)b)SampleProbeProbeColdHotPushCoolPumpRecombinePush offPush onPhoto-diode0.00.20.40.60.81.01.21.40.00.10.20.30.40.50.60.70.8ncold0(1018cm-3)0.110.350.621.041.40τcool(ps)nhot0(1018cm-3) Kinetic modelling of carrier cooling in lead halide perovskite materials
4
• Before the push arrives, the number of hot carriers is
negligible;
• The process of carrier cooling involves two co-existing
pathways, each with characteristic rates; (i) carrier-
carrier interaction, when a hot carrier scatters with an-
other hot or cold carrier and loses its excess energy; and
ii) carrier-phonon interaction, when the excess energy
is transferred into an unoccupied phonon mode. The
phonon mode becomes occupied, and the carrier cools
down;
• Total number of phonon modes (Nph) available is a con-
stant for the material. Phonon modes can be either oc-
cupied or unoccupied.
All simulations are presented for the unit volume (cm-3).
The initial system of equations is shown in Eq. (2) with the
initial conditions shown in Eq. (3):
dnhot
dt = Ipushncold − αnexcnhot − β nphnhot
dncold
dt = −dnhot
dnph
dt = −β nphnhot + ν(Nph − nph)
− εncold
dt
nhot = 0
ncold = ncold
0
nph = Nph
(2)
(3)
0
arises
The hot-phonon bottleneck effect
where nhot, nhot and nexc are the number of hot, cold and to-
tal number of carriers, respectively; nhot, nhot and nexc are the
total number of phonon modes and the number of unoccupied
phonon modes, respectively; ncold
is the cold carrier density at
t=0 (just before the push pulse arrives); Ipush is the number of
photons absorbed from the push pulse per unit time, expressed
as a time-dependent Gaussian envelope function; α is the rate
constant for the carrier-carrier interaction; β in the rate con-
stant for carrier-phonon interaction; ε is the rate constant for
the recombination of carriers; and ν is the rate constant for
phonon vacancy freeing.
from the
−β nphnhot component of the rate equation for nhot;
this
component is proportional to the unoccupied phonon density,
which depletes during the carrier-phonon interaction. The
hot-phonon bottleneck effect becomes prominent if the rate
of phonon depletion is greater than or comparable to the
rate of phonon scattering (the dissipation of hot phonons to
the lattice), i.e. −β nphnhot (cid:62) ν(Nph − nph). The effect of
carrier density is expressed in −αnexcnhot; an increased rate
of carrier-carrier interactions is expected for a greater number
of carriers. This system can be numerically solved to provide
a full picture of the process, as shown in Fig. 4(a). The
result is comprehensive but involves too many parameters for
the analysis of carrier dynamics. Considering that the push
pulse is short (∼50 fs) in comparison to the cooling times
(∼500 fs),16 we can assume that the intraband excitation is
immediate and hence Ipushncold ≈ 0 after 100 fs, i.e.
in the
time window of our interest. In addition, we notice that the
recombination rate is considerably slow (>100 ps),26 and
thus can be neglected (ε ≈ 0). Consequently, nexc becomes a
constant (nexc = ncold
) and the rate equation for ncold becomes
obsolete. The system simplifies into the form shown in
Eq. (4), with the initial conditions shown in Eq. (5):
0
dnhot
dt = −αnexcnhot − β nphnhot
dnph
dt = −β nphnhot + ν(Nph − nph)
(cid:40)
nhot = nhot
0
nph = Nph
(4)
(5)
is proportional to Ipush. The solution of these equa-
where nhot
0
tions provides the time dependence of nhot and nph, plotted in
Fig. 4(b).
FIG. 4.
(a) Representative plot of carrier dynamics from equa-
tions (Eq. (2)) and (Eq. (3)). Constant parameters used: α = 1.4×
10−18 cm3 ps-1, β = 2.7×10−17 cm3 ps-1, ε = 1.0×10−2 ps-1, ν =
1.0× 10−1 cm ps-1. Boundary conditions used: ncold
0 = 1.0× 1018,
0 = 6.0 × 1017, Nph = 2.0 × 1017. Pulse duration: 35 fs.
nhot
(b)
Representative plot of carrier dynamics from equations Eq. (4) and
Eq. (5), using the same boundary conditions as above. Constant pa-
rameters used: α = 1.4×10−18 cm3 ps-1, β = 2.7×10−17 cm3 ps-1,
ν = 1.0× 10−1 ps-1. The same boundary conditions as in Fig. 4(a)
were used. All parameters are in roughly the same order of magni-
tude as the experimentally determined values.
0.00.20.40.60.81.0 nhot ncold nexc nph Nph IPush(1018 cm-3), Ipusha)b) nhot ncold nexc nph Nph012340.00.20.40.60.81.0 (1018 cm-3)Time (ps) Kinetic modelling of carrier cooling in lead halide perovskite materials
5
The rate equation for nhot involves a variable (nph). This
can be eliminated using the following expression from the rate
equation of nhot and nph. All other parameters are constants.
=
≈ αnexc + β nph
−αnexcnhot − β nphnhot
−β nhotnph + ν(Nph − nph)
dnhot
dnph
We can neglect the ν(Nph−nph) term based on the assump-
tion that the phonon scattering is significantly slower than the
carrier-phonon interaction. This notion is supported by theo-
retical and experimental findings of >1 ps phonon lifetimes in
LHPs.27 -- 30
β nph
(6)
Using the initial condition shown in Eq. (5):
nhot(cid:90)
dn∗
hot =
nph(cid:90)
αnexc + β n∗
ph
β n∗
ph
dn∗
ph
nhot
0
This simplifies to:
nph = Nph exp
Nph
(cid:18) β
where ∆nhot = nhot − nhot
0
(cid:19)
(∆nhot − ∆nph)
αnexc
and ∆nph = nph − Nph.
(7)
(8)
FIG. 5. The dynamic occupied phonon density nph,occ = Nph − nph
from equations (Eq. (4)) and (Eq. (5)) is shown as the orange dashed
line, and the static occupied phonon density n∗
ph,occ from (Eq. (10))
is shown as the yellow dashed line. (a) and (b) show the effect when
the initial hot carrier density is set to 6.0× 1017 and 3.0× 1017 cm-3,
respectively. All other parameters are kept constant: α = 1.4×10−18
cm3 ps-1, β = 2.7 × 10−17 cm3 ps-1, ν = 1.0 × 10−1 ps-1, nexc =
1.0× 1018 cm3, Vp = 7.0× 10−18 cm3. Boundary condition: Nph =
2.0× 1017 cm-3.
We use an alternative approach to derive the equation for
further simplification and to find a physical quantity related
to polaron overlap. Let's assume that each hot carrier occu-
pies a specific volume, Vp, where carrier-phonon coupling can
occur.15 Then we divide a unit volume (V) of the lattice into
uniform cells with the volume Vp. The hot carriers can occupy
any of the cells and can either overlap or not overlap with each
other. The phonons are assumed to be evenly scattered across
the lattice, and only one of the overlapping hot carriers can in-
0 − nhot )V
teract with the phonon modes in the cell. When (nhot
carriers are coupled and dropped into the cells with the vol-
ume Vp in a lattice with Nph phonons per unit volume, then
the average available number of accessible phonons per unit
volume (nph) can be expressed as follows:
(cid:19)(nhot
(cid:18)
1− Vp
V
0 −nhot )V
nph = Nph
(9)
Given than Vp
imated to the following:
V is a small number, the expression is approx-
nph = Nph exp(Vp∆nhot )
(10)
Fig. 5 (a) and (b) compare the expression of nph from
Eqs. (4), (5) and (10) in the time domain. We demonstrate
that the use of Eq. (10) is a reasonable approximation of nph
in equations (4) and (5), especially at early times <1 ps.
The rate equation of nhot can be expressed as:
dnhot
dt = −αnexcnhot − β Nph exp(−Vpnhot
0 )nhot
(11)
where α, β , Vp and Nph are constants. In this expression,
α is the rate constant for the carrier-carrier interaction, β is
the rate constant for the carrier-phonon interaction, Vp is the
volume of a polaron and Nph is the number of unoccupied
phonon modes per unit volume in the lattice.
The equation Eq. (11) includes the product of two arbitrary
constants (β Nph), allowing the elimination of a parameter. β
is defined as the average volume of space around a hot car-
rier where a phonon mode will interact with the carrier with
∼100% probability in 1 ps. β Nph can be substituted by ρVp
where Vp is the volume of space around a hot carrier where
phonons are coupled and ρ is the number of carrier-phonon
interactions per unit volume per 1 ps. Both β Nph and ρVp
represent the number of carrier-phonon interactions per hot
carrier per 1 ps. Hence, the rate equation can be expressed as
the Eq. (12).
dnhot
dt = −αnexcnhot − ρVp exp(−Vpnhot
0 )nhot
(12)
The hot-phonon bottleneck arises from the reduced number
of available phonon modes with the increasing hot carrier den-
sity, expressed as ρVp exp(−Vpnhot
0 ). The carrier-carrier inter-
action is shown in the −αnexcnhot component; the rate of the
interaction is shown to be proportional to the carrier density.
The cooling time is calculated using the following expression:
e(cid:90)
nhot
0
nhot
0
τcool =
dnhot
dt
dt
(13)
0.00.10.20.30.40.50.6 nhot nph, occ Nph n*ph, occ (1018 cm-3)a)b)012340.00.10.20.3 nhot nph, occ Nph n*ph, occ (1018 cm-3)Time (ps) Kinetic modelling of carrier cooling in lead halide perovskite materials
6
Based on this equation, we fit the experimental data from
the CsPbBr3 NCs and bulk material for comparison. The fits
and experimental data are shown in Figs. 3 and 6 for the re-
spective samples. Material constants derived from the fits are
tabulated in Table I.
FIG. 6. Plot of experimental data (square points) and fits from the
model (dashed lines) for bulk CsPbBr3.
TABLE I. Material constants for the polaron volume (Vp), carrier-
carrier interaction (α) and carrier-phonon interaction (ρ) derived
from the fit.
Material
α
ρ
Vp
(10−18cm3)
(10−18cm3 ps−1)
(1018cm−3 ps−1)
Bulk CsPbBr3
CsPbBr3 NCs
7.58
7.00
1.95
1.08
0.46
0.43
Based on Vp, the polaron radius in bulk CsPbBr3 and the
CsPbBr3 NCs is 12.1 and is 11.9 nm, respectively. Although
"large" polarons are often invoked to explain the unique pho-
tophysics of LHPs,31 -- 38 we note these values are approxi-
mately double to the polaron size experimentally determined
by Munson et al. for MAPbI3 (∼4.5 nm radius),39 and larger
still for the computationally-derived values for CsPbBr3 re-
ported elsewhere.16,33 We posit that the relatively large values
obtained here are a consequence of energy transfer processes
which lead to an incorrect estimation of the local carrier den-
sity. In other words, we believe that energy transfer processes
result in "hot spots" where the local carrier density is higher
than the average carrier density of the photoexcited ensem-
ble. Experimental evidence for this phenomenon has been re-
ported in LHPs.40 We also note that the comparable values for
the bulk and NC samples are not entirely unexpected given the
bulk-like behaviour of weakly confined CsPbBr3 NCs.41
Taking that both the carrier-carrier and carrier-phonon in-
teractions occur on the sub-ps timescale, we can use the model
parameters to compare the carrier-carrier and carrier-phonon
interaction distances. The values above indicate that for
CsPbBr3 and its NC analogue, the typical distance at which
carrier-phonon interactions start playing a role is ∼12 nm.
This is slightly larger than the typical carrier-carrier interac-
tion distance which, when deduced from α, is equal to ∼7 nm.
As discussed above, these absolute values are higher estimates
of the actual numbers due to aforementioned incorrect estima-
tion of the local carrier density. However, the estimates of α
and Vp depend on carrier density in the same way, so the ra-
tio between these parameters should be captured by the model
correctly. The fact that the carrier-carrier interactions are ac-
tive at shorter distances than the polaron size is consistent with
the notion of polaronic screening in LHPs previously reported
in the literature.31 -- 33
V. CONCLUSION
In summary, we developed a numerical model which
includes "carrier-carrier" and "carrier-phonon" interaction
terms to describe the role of the hot and cold carriers in
the carrier cooling kinetics of bulk and nanocrystal CsPbBr3
films determined by "pump-push-probe" spectroscopy. The
observation of slower cooling kinetics at higher push inten-
sity (higher hot carrier density) is explained by invoking a
hot-phonon bottleneck mechanism where overlapping hot po-
larons compete for phonons in order to cool. This is incorpo-
rated in our model by the carrier-phonon term which describes
the spatial overlap between the hot polaron states. Meanwhile,
the observation of faster cooling kinetics for higher pump in-
tensity is attributed to the redistribution of energy from a hot
carrier to adjacent cold carriers, as described by the carrier-
carrier term. By fitting our model to the experimental re-
sults, we demonstrate the interplay between these contrasting
effects, and propose a way to estimate the polaron size, the
phonon density of the lattice and the rate constant for carrier-
carrier and carrier-phonon interactions.
ACKNOWLEDGMENTS
A.A.B. is a Royal Society University Research Fellow. A.J.
thanks the Royal Society for supporting the project. A.G.
thanks Mr. Amirlan Sekenbayev of Queen Mary University
for useful discussions. This project has also received fund-
ing from the European Research Council (ERC) under the
European Union's Horizon 2020 research and innovation pro-
gramme (Grant Agreement No. 639750). The authors declare
no competing financial interest.
1A. Kojima, K. Teshima, Y. Shirai, and T. Miyasaka, J. Am. Chem. Soc 131,
6050 -- 6051 (2009).
2NREL, "Best research-cell efficiencies chart," Accessed 10 September
2019.
3W. Shockley and H. J. Queisser, J. Appl. Phys 32, 510 -- 519 (1961).
4R. T. Ross and A. J. Nozik, J. Appl. Phys 53, 3813 -- 3818 (1982).
5S. Kahmann and M. A. Loi, J. Mater. Chem. C 7, 2471 -- 2486 (2019).
6M. B. Price, J. Butkus, T. C. Jellicoe, A. Sadhanala, A. Briane, J. E. Halpert,
K. Broch, J. M. Hodgkiss, R. H. Friend, and F. Deschler, Nat. Commun. 6,
8420 (2015).
0.00.10.20.30.40.50.60.70.00.10.20.30.40.50.60.7ncold0(1018cm-3)0.100.230.350.500.730.91τcool(ps)nhot0(1018cm-3) Kinetic modelling of carrier cooling in lead halide perovskite materials
7
7Y. Yang, D. P. Ostrowski, R. M. France, K. Zhu, J. van de Lagemaat, J. M.
Luther, and M. C. Beard, Nat. Photonics 10, 53 -- 59 (2016).
8J. Yang, X. Wen, H. Xia, R. Sheng, Q. Ma, J. Kim, P. Tapping, T. Harada,
T. W. Kee, F. Huang, Y.-B. Cheng, M. Green, A. Ho-Baillie, S. Huang,
S. Shrestha, R. Patterson, and G. Conibeer, Nat. Commun. 8, 14120 (2017).
9P. Papagiorgis, L. Protesescu, M. V. Kovalenko, A. Othonos, and G. Itskos,
J. Phys. Chem. C 121, 12434 -- 12440 (2017).
10M. Li, S. Bhaumik, T. W. Goh, M. S. Kumar, N. Yantara, M. Grätzel,
S. Mhaisalkar, N. Mathews, and T. C. Sum, Nat. Commun. 8, 14350 (2017).
11J. Chen, M. E. Messing, K. Zheng, and T. Pullerits, J. Am. Chem. Soc ,
3532 -- 3540 (2019).
12B. T. Diroll and R. D. Schaller, Adv. Funct. Mater. 1901725, 1901725
(2019).
(2017).
13J. Shah, Solid-State Electronics 21, 43 -- 50 (1978).
14R. Leheny, J. Shah, R. Fork, C. Shank, and A. Migus, Solid State Commun.
31, 809 -- 813 (1979).
15J. M. Frost, L. D. Whalley, and A. Walsh, ACS Energy Lett. 2, 2647 -- 2652
16T. R. Hopper, A. Gorodetsky, J. M. Frost, C. Müller, R. Lovrincic, and
A. A. Bakulin, ACS Energy Lett. 3, 2199 -- 2205 (2018).
17J. M. Richter, F. Branchi, F. Valduga de Almeida Camargo, B. Zhao, R. H.
Friend, G. Cerullo, and F. Deschler, Nat. Commun. 8, 376 (2017).
18T. Ghosh, S. Aharon, L. Etgar, and S. Ruhman, J. Am. Chem. Soc 139,
18262 -- 18270 (2017).
8308 -- 8315 (2016).
19R. A. Kerner, L. Zhao, Z. Xiao, and B. P. Rand, J. Mater. Chem. A 4,
20J. Endres, D. A. Egger, M. Kulbak, R. A. Kerner, L. Zhao, S. H. Silver,
G. Hodes, B. P. Rand, D. Cahen, L. Kronik, and A. Kahn, J. Phys. Chem.
Lett. 7, 2722 -- 2729 (2016).
21F. Krieg, S. T. Ochsenbein, S. Yakunin, S. ten Brinck, P. Aellen, A. Süess,
B. Clerc, D. Guggisberg, O. Nazarenko, Y. Shynkarenko, S. Kumar, C.-
J. Shih, I. Infante, and M. V. Kovalenko, ACS Energy Lett. 3, 641 -- 646
(2018).
22L. Protesescu, S. Yakunin, M. I. Bodnarchuk, F. Krieg, R. Caputo, C. H.
Hendon, R. X. Yang, A. Walsh, and M. V. Kovalenko, Nano Lett. 15, 3692 --
3696 (2015).
23P. Guyot-Sionnest, M. Shim, C. Matranga, and M. Hines, Phys. Rev. B 60,
R2181 -- R2184 (1999).
24F. T. Rabouw, R. Vaxenburg, A. A. Bakulin, R. J. A. van Dijk-Moes, H. J.
Bakker, A. Rodina, E. Lifshitz, A. L. Efros, A. F. Koenderink, and D. Van-
maekelbergh, ACS Nano 9, 10366 -- 10376 (2015).
25K. T. Munson, C. Grieco, E. R. Kennehan, R. J. Stewart, and J. B. Asbury,
ACS Energy Lett. 2, 651 -- 658 (2017).
26N. S. Makarov, S. Guo, O. Isaienko, W. Liu, I. Robel, and V. I. Klimov,
Nano Lett. 16, 2349 -- 2362 (2016).
27M. Wang and S. Lin, Adv. Funct. Mater. 26, 5297 -- 5306 (2016).
28L. D. Whalley, J. M. Skelton, J. M. Frost, and A. Walsh, Phys. Rev. B 94,
220301 (2016).
29A. N. Beecher, O. E. Semonin, J. M. Skelton, J. M. Frost, M. W. Terban,
H. Zhai, A. Alatas, J. S. Owen, A. Walsh, and S. J. L. Billinge, ACS Energy
Lett. 1, 880 -- 887 (2016).
30B. Li, Y. Kawakita, Y. Liu, M. Wang, M. Matsuura, K. Shibata, S. Ohira-
Kawamura, T. Yamada, S. Lin, K. Nakajima, and S. Liu, Nat. Commun. 8,
16086 (2017).
31X.-Y. Zhu and V. Podzorov, J. Phys. Chem. Lett. 6, 4758 -- 4761 (2015).
32H. Zhu, K. Miyata, Y. Fu, J. Wang, P. P. Joshi, D. Niesner, K. W. Williams,
S. Jin, and X.-Y. Y. Zhu, Science 353, 1409 -- 1413 (2016), 1609.07036.
33K. Miyata, D. Meggiolaro, M. T. Trinh, P. P. Joshi, E. Mosconi, S. C. Jones,
F. De Angelis, and X.-Y. Zhu, Sci. Adv 3, e1701217 (2017).
34L. Wang, N. P. Brawand, M. Vörös, P. D. Dahlberg, J. P. Otto, N. E.
Williams, D. M. Tiede, G. Galli, and G. S. Engel, Adv. Opt. Mater. 6,
1700975 (2018).
35G. Batignani, G. Fumero, A. R. Srimath Kandada, G. Cerullo, M. Gandini,
C. Ferrante, A. Petrozza, and T. Scopigno, Nat. Commun. 9, 1971 (2018),
1705.08687.
36Y. Lan, B. J. Dringoli, D. A. Valverde-Chavez, C. S. Ponseca, M. Sutton,
Y. He, M. G. Kanatzidis, and D. G. Cooke, Sci. Adv (2018).
37E. Cinquanta, D. Meggiolaro, S. G. Motti, M. Gandini, M. J. P.
Alcocer, Q. A. Akkerman, C. Vozzi, L. Manna, F. De Angelis,
A. Petrozza, and S. Stagira, Phys. Rev. Letters 122, 166601 (2019), Phys-
RevLett.122.166601 [10.1103].
38A. Mahata, D. Meggiolaro, and F. De Angelis, J. Phys. Chem. Lett. 10,
1790 -- 1798 (2019).
2826 -- 2843 (2018).
39K. T. Munson, E. R. Kennehan, G. S. Doucette, and J. B. Asbury, Chem 4,
40M. Vru´cini´c, C. Matthiesen, A. Sadhanala, G. Divitini, S. Cacovich, S. E.
Dutton, C. Ducati, M. Atatüre, H. Snaith, R. H. Friend, H. Sirringhaus, and
F. Deschler, Adv. Sci. 2, 1500136 (2015).
41J. Butkus, P. Vashishtha, K. Chen, J. K. Gallaher, S. K. K. Prasad, D. Z.
Metin, G. Laufersky, N. Gaston, J. E. Halpert, and J. M. Hodgkiss, Chem.
Mater 29, 3644 -- 3652 (2017).
|
1809.07231 | 1 | 1809 | 2018-09-19T15:02:47 | Overall constitutive description of symmetric layered media based on scattering of oblique SH waves | [
"physics.app-ph"
] | This papers investigates the scattering of oblique shear horizontal (SH) waves off finite periodic media made of elastic and viscoelastic layers. It further considers whether a Willis-type constitutive matrix (in temporal and spatial Fourier domain) may reproduce the scattering matrix (SM) of such a system. In answering this question the procedure to determine the relevant overall constitutive parameters for such a medium is presented. To do this, first the general form of the dispersion relation and impedances for oblique SH propagation in such coupled Willis-type media are developed. The band structure and scattering of layered media are calculated using the transfer matrix (TM) method. The dispersion relation may be derived based on the eigensolutions of an infinite periodic domain. The wave impedances associated with the exterior surfaces of a finite thickness slab are extracted from the scattering of such a system. Based on reciprocity and available symmetries of the structure and each constituent layer, the general form of the dispersion and impedances may be simplified. The overall quantities may be extracted by equating the scattering data from TM with those expected from a Willis-type medium. It becomes evident that a Willis-type coupled constitutive tensor with components that are assumed independent of wave vector is unable to reproduce all oblique scattering data. Therefore, non-unique wave vector dependent formulations are introduced, whose SM matches that of the layered media exactly. It is further shown that the dependence of the overall constitutive tensors of such systems on the wave vector is not removable even at very small frequencies and incidence angles and that analytical considerations significantly limit the potential forms of the spatially dispersive constitutive tensors. | physics.app-ph | physics |
Overall constitutive description of symmetric layered
media based on scattering of oblique SH waves
Alireza V. Amirkhizi and Vahidreza Alizadeh
Department of Mechanical Engineering
University of Massachusetts, Lowell
One University Avenue, Lowell, MA 01854
alireza [email protected]
Abstract
This papers investigates the scattering of oblique shear horizontal (SH) waves
off finite periodic media made of elastic and viscoelastic layers. It further
considers whether a Willis-type constitutive matrix (in temporal and spa-
tial Fourier domain) may reproduce the scattering matrix (SM) of such a
system. In answering this question the procedure to determine the relevant
overall constitutive parameters for such a medium is presented. To do this,
first the general form of the dispersion relation and impedances for oblique
SH propagation in such coupled Willis-type media are developed. The band
structure and scattering of layered media are calculated using the transfer
matrix (TM) method. The dispersion relation may be derived based on the
eigen-solutions of an infinite periodic domain. The wave impedances associ-
ated with the exterior surfaces of a finite thickness slab are extracted from
the scattering of such a system. Based on reciprocity and available sym-
metries of the structure and each constituent layer, the general form of the
dispersion and impedances may be simplified. The overall quantities may be
extracted by equating the scattering data from TM with those expected from
a Willis-type medium. It becomes evident that a Willis-type coupled consti-
tutive tensor with components that are assumed independent of wave vector
is unable to reproduce all oblique scattering data. Therefore, non-unique
wave vector dependent formulations are introduced, whose SM matches that
of the layered media exactly.
It is further shown that the dependence of
the overall constitutive tensors of such systems on the wave vector is not
removable even at very small frequencies and incidence angles and that an-
alytical considerations significantly limit the potential forms of the spatially
Preprint submitted to Wave Motion
June 8, 2021
dispersive constitutive tensors.
Keywords: Dynamic Overall Properties, Transfer Matrix, Homogenization,
Spatial Dispersion, Layered Media, Willis-type Media
1. Introduction
Spatial dispersion, i.e. the wave-vector dependence of wave propagation,
has been widely studied in electromagnetism, crystal optics, and photonic
crystals, but has received relatively little attention in stress wave propaga-
tion in solids [1, 2]. In electromagnetism, such phenomena are rooted in the
possible dependence of the electric displacement field at a given point, D,
on the electric field, E, in a neighborhood around that point, i.e. nonlocal
physics. Another mathematical approach to such phenomena is to consider
the dependence of D on spatial derivatives of E, which in the case of wave
propagation, or after performing a spatial Fourier transform, can be written
in terms of k functionality. The general non-local behavior is not always rep-
resentable in a wave-vector based formalism particularly for finite systems.
However, the focus of this paper is to determine whether the overall scatter-
ing response of certain heterogeneous systems may be fully represented using
overall constitutive tensors in Fourier domain, and to describe the process
of extracting the relevant components of such tensors. Similarly, for stress
waves a limited representation of non-locality after spatial Fourier transform
is written as the elastic constants exhibiting a dependence on the wave vector
besides the frequency, i.e. Cijkl(ω, k). The 4th order tensorial nature of the
elastic constants introduces more extensive mathematical complexity in stress
wave problems. Many interesting features in EM of conductive materials and
metamaterials, such as virtual surface waves [3], metasurfaces [4, 5], cloak-
ing [6], longitudinal polarization waves [7 -- 10], negative refractions [11, 12],
anti-resonance response [13, 14] among others, are either due or related to
spatial dispersion. Extensive studies has been also done on different types of
metamaterials in which spatial dispersion is a necessity to characterize the
system due to its periodicity (particularly when wavelengths of interest are
comparable with unit cell dimensions) [15 -- 23]. Experimentally, Hopfiled et
al. [24] observed spatial dispersion behavior in optical properties of crystals
and showed that classical optics of nonmagnetic crystal theory could not ac-
curately predict their empirically obtained scattering data. In another study
by Portigal et al. [25], it was shown that by considering the first-order spatial
2
dispersion "acoustical activity" can occur in crystals which exhibit optical ac-
tivity. Higher order theories have also been utilized in order to exactly satisfy
the impedance matching for all incident angles in perfectly matched layers
(PML), [26] or to homogenize for overall properties of metamaterials with
nonmagnetic inclusions [27]. Agarwal et al.
[28] proposed a theory related
to the nature of electromagnetic field in spatially dispersive media. They
concluded that without the introduction of any additional ad-hoc boundary
conditions, the scattering problem of a plane-parallel slab can be solved com-
pletely within the framework of classical electromagnetic theory. This theory
can be treated as a step forward in understanding and formulating of the spa-
tial dispersion phenomenon in general. Puri et al.
[29] also demonstrated
that the optical properties of finite bounded crystals, e.g. GaAs, CuCl, etc,
show nonlocal behavior for frequencies near an exciton resonance. In another
study on bianisotropic media Belov et al. [30] proved that for a 2D lattice of
infinitely long parallel conducting helices, strong spatial dispersion is observ-
able even at very low frequencies along the spiral axis of the bianisotropic
medium. Yaghjian et al. [31] developed an anisotropic homogenization the-
ory for spatially dispersive periodic arrays based on Maxwell's equations.
They mathematically proved that it is impossible to characterize metama-
terials formed by periodic arrays of polarizable inclusions without spatially
dispersive representation of permittivity and inverse transverse permeability.
In contrast, nonlocal stress wave propagation in elastic composites has
not received much attention in the literature. A number of authors have
looked at the extraction of the overall mechanical properties from scattering
data. Popa et al.
[32] showed that the anisotropic effective mass density
tensor components, which were extracted from the scattering data, can be
independently controlled by properly designing the inclusions geometry in
a host fluid. To enhance the applicability of retrieval methods, Castanie
et al.
[33] analyzed the problem of extracting the overall properties of an
orthotropic slab, whose normal is not one of the material principal axes.
Park et al.
[34] studied the behavior of anisotropic acoustic metamaterials
by considering non-diagonal effective mass density tensor in the presence of
spatial dispersion. Lafarge and Nemati [35, 36] developed a nonlocal theory
for sound propagation in fluid-filled rigid frame porous media. The nonlocal
approach was further expanded to include the effect of arrays of Helmoholtz
resonators in [37] and general non-local treatment of dissipative phononic
liquids in [38]. In these works, the focus has been on propagation of pressure
acoustic waves and the effect of microstructure on effective bulk modulus and
3
density. Lee et al. [39], extended the work in [34] to mechanical metamate-
rials, where the elasticity tensor components, beyond just the bulk modulus,
are incorporated. Recently a mathematically exact theory for Willis-type
media has been developed along with a matching field integration technique
for normal incidence in [40], where transfer matrix method (TMM) is used
to calculate the scattering exactly. The field averaging technique in [40] is
based on the explicit integration of the wave equations and it matches the
scattering response exactly as well. The method is applied to viscoelastic
and non-symmetric structures without any modification. It must be noted
that Nemat-Nasser [41] developed a variational approach to extract the band
structure of SH waves in periodic composite media, where group velocity and
energy flux vectors were calculated explicitly. In another study by Srivastava
[42], SH waves at an interface between a homogeneous material and a lay-
ered periodic composite were investigated. The laminate had a periodically
layered structure in x1-direction, with interface between the homogeneous
medium and layered structure was normal to the x2-axis and the SH waves
were polarized along the x3-axis. Negative refraction and beam steering was
observed in a wide range of frequency.
In this work, the scattering approach of [40] is used for the oblique inci-
dence of SH waves in layered media to calculate the scattering matrix (SM)
without any numerical (e.g. FE) effort. It is determined whether a Willis-
type constitutive tensor formalism in spatial Fourier domain can reproduce
the scattering response of such systems, and if so, how to retrieve these
constitutive parameters. Transfer matrix methods has been widely and suc-
cessfully used to calculate exact and approximate scattering for oblique SH
[43]. To simplify the derivations, the
waves; See, for example, Vinh et al.
work is focused on symmetric unit cells and structures.
In the following
sections, first, the general form of the dispersion relations and impedances
are extracted by solving the equations of motion, taking into account the
Willis-type constitutive law. The results are discussed in light of various
symmetry and reciprocity requirements. A general approach for extracting
the overall constitutive parameters from the scattering data (dispersion and
impedances) is developed. Next, a two-phase symmetric layup composite
is presented to apply the general approach. It is shown that the response
may not be fully represented by constitutive matrices that are independent
of wave vector, even when one allows for coupling terms in the Willis for-
malism. When the constitutive tensors are allowed to be functions of wave
vector, the SH wave propagation results are not enough to fully determine all
4
constitutive functions. Two special representations are discussed: one with
a diagonal constitutive tensor, and one with the diagonal terms that are not
functions of the wave vector (only off-diagonal components are allowed to
change with the wave vector). A third representation is studied for which
it is shown that analytical considerations eliminate any possible compatible
solution, pointing out that further limitations on the potential acceptable
forms of the constitutive tensors exist. Based on this, it is expected that
by considering all other propagation modes (P and SV) some of the poten-
tial solutions will further be ruled out. This approach and the treatment of
asymmetric structures will be taken up in future efforts.
2. SH wave dispersion, slowness, and impedance in coupled media
The elastodynamics of source-free media may be derived based on the
combination of the compatibility and equilibrium equations
(∂ivj + ∂jvi)/2 = ∂tεij,
∂jσji = ∂tpi,
(1)
(2)
with the constitutive law. Here v, σ, p, and ε denote particle velocity, stress,
momentum density (per unit volume), and strain, respectively, and subscripts
after ∂ represent partial differentiation with respect to that variable. For any
material or composite that has translational symmetry along an axis, e.g. x3,
SH-waves with particle velocity polarization along this symmetry axis and
propagation direction in the x1x2 plane normal to it may be studied inde-
pendently if all constituents have sufficient symmetry to decouple such shear
deformation from other tensor components of the dynamic and kinematic
quantities in Equations (1) and (2), which can then be collected as
0
∂2
∂1
∂2 ∂1
0
0
0
0
v3
= ∂t
τ4
τ5
,
p3
γ4
γ5
(3)
where the Voigt notation is used, i.e. γ4 = 2ε23 = 2ε32, γ5 = 2ε31 = 2ε13, τ4 =
σ23 = σ32, τ5 = σ31 = σ13.
In the following only systems are considered,
for which it is assumed that a Willi-type formalism in temporal and spatial
Fourier domain is capable of reproducing their overall (scattering) behavior.
This could be a homogeneous, layered, finite, or infinite system, and the
validity of such assumption is established by applying this formalism and
5
inspecting its actual success in this reproduction. To study oblique incidence
of SH waves, only a portion of the full Willis-type constitutive law is used
=
v3
τ4
τ5
η33 κ34 κ35
κ43 µ44 µ45
κ53 µ54 µ55
p3
,
γ4
γ5
(4)
(5)
where η33 represents the appropriate specific volume, µij, i, j = 4, 5 are the
relevant moduli of elasticity (e.g. µ12 = C2331), while κij denote particle
velocity/strain and stress/momentum density couplings. All components are
functions of the frequency and wave vector in general. Equation (4) is a
slightly transformed version of the formulation used in [44, 45] in that stress
is grouped with particle velocity instead of momentum density [40]. Matrix
inversion of Equation (4) gives the form in terms of density and compliance.
One reason for using this grouping is the similarity with Equations (1) and
(2), in that momentum density is grouped with strain components in a sin-
gle column vector, and particle velocity is grouped with stress components.
Furthermore and as discussed in [40] the integration of wave equations also
leads to volume integral definitions for overall strain and momentum density
quantities, while overall stress and particle velocities have natural definitions
as surface integrals. In frequency domain, the physical quantities described
above (β = v3, τj, p3, γj, j = 4, 5) are written as β(x, t) = (cid:60)(βc(x)e−iωt),
where ω = 2πf is the angular frequency and βc is the complex amplitude.
For a spatial Fourier component βc(x) = βc0eik·x,1 where k is the wave vector,
equation (3) can be written as
0
k2
k1
− 1
ω
v3
=
τ4
τ5
.
p3
γ4
γ5
k2 k1
0
0
0
0
Using the column vector β = (v3, τ4, τ5)T , these equations can be combined
as
−κψβ = β.
(6)
1The subscripts c and c0 will be dropped in the following unless there is potential for
confusion.
6
Here κ is the Willis-type constitutive matrix in Equation (4) and
0
s2
s1
ψ =
s2 s1
0
0
0
0
=
1
ω
0
k2
k1
,
k2 k1
0
0
0
0
(7)
is made of slowness vector components s = k/ω. The dispersion relation
between the wave vector components, kj, j = 1, 2, and the frequency ω can
now be found from
In the most general form, the dispersion surface equation will look like
det(I + κψ) = 0.
0 = D(ω, k1, k2) = ω2 + ω((κ35 + κ53)k1 + (κ34 + κ43)k2)
+ (κ34κ53 + κ43κ35 − η33(µ45 + µ54))k1k2
+ (κ35κ53 − η33µ55)k2
1 + (κ34κ43 − η33µ44)k2
2.
(8)
(9)
1 = sα
For any 2 selected variables, e.g. wave vector components k1 and k2, or even
one variable and a ratio, e.g. frequency ω and slowness component s2, there
may exist multiple points on the complex dispersion surface, which will be
formally identified by a superscript, e.g. ωα = ωα(k1, k2) or sα
1 (ω, s2).
While this allows for finding the general solutions, in most cases of interest
one or more of the variables are restricted. For example, steady-state cases
are limited to real frequencies. For example, in particular cases of interest in
this paper, i.e. scattering of oblique SH waves in systems that have transla-
tional invariance along the x2 axis, a single value of k2 is fixed everywhere in
the system. Furthermore, expectation of analyticity of constitutive tensors
(except at potentially discrete locations) may be utilized to conclude that for
most of the wave vector space, there are limited number of solutions to (9), in
fact only 2. This is in agreement with the physical observation of overall scat-
tering of layered media, calculated further below. Moreover, any potential
point for which the analyticity assumption fails should be treated carefully
and separately. Note that the Bloch multiplicity of k1 in these cases does
not provide a physical distinct new solution; this is further discussed below.
For each solution α (in the following indexed by ± for 2 solutions), the null
space of the matrix I + κψ is generated by vectors βα = (1,−zα
53)T ,
where zα
j3, j = 4, 5, are formally defined as impedances associated with the
points α on the dispersion surface.
43,−zα
7
Reciprocity and symmetries. A reversal in time coordinate, t → −t (along
with ω → −ω) would not change the physics of reciprocal systems and in
particular the shape of their dispersion surfaces. However, such a transfor-
mation will change the solutions of the dispersion relation (9) unless the
coefficient of ω vanishes, therefor: 2
κ34 + κ43 = 0,
κ35 + κ53 = 0.
(10)
(11)
Note that the analyticity considerations of the previous paragraph is implic-
itly used here, as this discussion is based on considering the lowest order
approximation of constant constitutive parameters in a small enough neigh-
borhood around the point of interest on the dispersion surface (which is also
assumed to be one with the normal topology observed almost everywhere
as shown later in the example results). The dispersion relation may now be
slightly simplified, e.g. in slowness space
(η33µ55 + κ2
35)s2
1 + (η33µ44 + κ2
34)s2
2 + (η33(µ45 + µ54) + 2κ34κ35)s1s2 = 1. (12)
If the system has a mirror or π rotational symmetry in or around the x1 or
x2 directions then the coefficient of k1k2 should vanish as well
η33(µ45 + µ54) + 2κ34κ35 = 0,
and the slowness/dispersion formulas will simplify further to
(η33µ55 + κ2
35)s2
1 + (η33µ44 + κ2
34)s2
2 = 1.
(13)
(14)
Even without the mirror symmetry simplification, for a reciprocal medium,
the impedances have the relatively compact forms
z43 = µ44s2 + µ45s1 + (κ34/η33)(1 + κ34s2 + κ35s1),
z53 = µ55s1 + µ54s2 + (κ35/η33)(1 + κ35s1 + κ34s2).
(15)
(16)
Mirror and rotational symmetries of the material or structure will further
restrict the constitutive tensors, particularly when they are considered as
2In a system where constitutive tensors are function of wave vector, this will lead to a
single yet more complex condition, instead of the two independent ones in the text.
8
functions of the wave vector.3 For example if x2 → −x2 is a symmetry
operation of the system, then z43 should switch sign under k2 → −k2, while
z53 should stay unchanged. A sufficient4 condition set for this is
Similarly if x1 → −x1 is a symmetry operation then
µ45 = µ54 = κ34 = 0.
µ45 = µ54 = κ35 = 0.
(17)
(18)
However, for systems where the coupled constitutive parameters are functions
of the wave vector, the conditions will take the form of parity restraints
on the functional k1 or k2 dependence of the coupling constants. If x2 →
−x2 is a symmetry operation, then µ44, µ55, η33, κ35 are even functions of
k2, while µ45, µ54, κ34 are odd. If x1 → −x1 is a symmetry operation, then
µ44, µ55, η33, κ34 are even functions of k1, while µ45, µ54, κ35 are odd.
3. Transfer matrix of layered structures for oblique SH waves
The transfer matrix of a material layer normal to the x1-axis, and identi-
fied by index l, for SH waves with velocity polarization along x3 axis relates
the particle velocity and traction components on the two boundaries:
(cid:18)v3(x1,l+1)
(cid:19)
τ5(x1,l+1)
(cid:18)v3(x1,l)
(cid:19)
τ5(x1,l)
= Tl
.
(19)
Tl is a function of the frequency and/or wave vector components, the layer's
constitutive parameters, and its thickness dl = x1,l+1 − x1,l, where x1,l and
x1,l+1 represent the coordinates of the two surfaces of the layer. To satisfy
continuity along all interfaces at all times, the frequency and the k2 compo-
nent of the wave vector should match across all layers. For each such pair,
3As pure SH waves are discussed here, at least mirror symmetry in the x3 direction is
needed. Note that none of the quantities considered here are pseudo-tensors. Therefore,
mirror symmetries on planes normal the other two axes are essentially equivalent to π
rotations around them. If there is no mirror symmetry normal to the x3 axis, it is unlikely
to get a pure SH wave without couple stresses and spins.
4And necessary except at discrete locations on the dispersion surface, e.g. s1 = 0.
9
the homogeneous wave equation has the general solution superposing two
independent SH waves
(cid:19)
(cid:18)v3(x1, kα
τ5(x1, kα
1,l)
1,l)
(cid:18)
1−zα
=
53,l(kα
1,l)
(cid:19)
Aα
l (kα
1,l)eikα
1,lx1,
x1,l ≤ x1 ≤ x1,l+1
(20)
where α = +,− represents the two solutions, Aα
plitudes of the two waves, the common phase ei(k2x2−ωt) is dropped, and
1,l) are the complex am-
l (kα
zα
53,l(kα
1,l) = − τ5(x1, kα
1,l)
v3(x1, kα
1,l)
,
(21)
is the impedance of layer l associated with the wave vector kα
metric and reciprocal system, the transfer matrix can be written as:
1,l. For a sym-
cos k1d
−iz53 sin k1d
T =
.
sin k1d
− i
z53
cos k1d
(22)
Using the continuity of the particle velocity and traction vectors, one may
calculate the transfer matrix of a cell consisting of nl layers as
and that of a finite slab consisting of nc cells as
Tc = Tnl...T2T1,
Ts = (Tc)nc.
(23)
(24)
Consider an elastic unit cell that generates an infinitely periodic structure.
The band structure of such a system may calculated by solving the eigenvalue
problem
Tcζ = eikα
1 dcζ,
(25)
where dc = x1,nl+1 − x1,1 and for finite impedances ζ = (1,−zα
53(kα
1 ))T . Note
the phase ambiguity in the overall wave vector component kα
1 , where any
integer multiple of 2π/dc may be added to or subtracted from it. Enforc-
ing continuity requirements on the wave vector allows one to remove this
ambiguity [40].
The scattering matrix of a multi-layered slab extending from x1,a to
x1,b = x1,a +ds = x1,a +ncdc and placed in between two half-spaces (also iden-
tified by subscripts a and b) may be calculated from its transfer matrix and
10
the properties of the two half spaces on by using the continuity conditions.
For the sake of simplicity consider the two media to be identical and sym-
metric, with characteristic impedance zb = za = z0. When the sample is also
symmetric and reciprocal and using the homogenized values in Equation (22)
[40]:
2
Sba = Sab =
Sbb = Saa =
2 cos k1ds − i(z53/z0 + z0/z53) sin k1ds
2 cos k1ds − i(z53/z0 + z0/z53) sin k1ds
i(z53/z0 − z0/z53) sin k1ds
,
.
(26)
53, k+
1 = −k−
53 = −z−
where Sba and Saa denotes transmission and reflection coefficients, respec-
tively, in terms of kinematic quantities (particle velocity). The symmetry
and reciprocity of the system removes the need for superscript α as for the
only 2 SH solutions z+
1 . Note that point-wise inversion
of the measured (or calculated) SM using these equations to determine k1
and z53 does not provide a unique solution for k1. However, this ambiguity
may be removed by enforcing the (physically motivated) continuity of k1 as a
function of frequency (similar to the traditional procedure of phase spectral
analysis; see for example [46]). The inter-relation of S and T is presented
in [40]. Therefore, one may define the overall properties of a heterogeneous
sample by finding the parameters k1 and z53 (assuming d = ds, of course)
with which Equation (22) reproduces its exact transfer matrix.
It can be
shown that these definitions are independent of the number unit cells in the
analysis [40]. If all layers have x2 mirror symmetry (or π rotational around
x1), the system does so as well, and therefore the simplifications discussed
above may be used. That means in the approximation where the constitu-
tive parameters are not functions of the wave vector, equations (17) will hold.
The dispersion relation and impedances will now simplify to:
(η33µ55 + κ2
z53 = µ55s1 + (κ35/η33)(1 + κ35s1),
1 + η33µ44s2
2 = 1,
35)s2
z43 = µ44s2.
(27)
(28)
(29)
Furthermore, if the system has mirror symmetry normal to x1 as well as
every layer (or π rotational around x2 or x3) then equations (18) will also
11
hold, enforcing a diagonal form for the constitutive tensor and
η33µ55s2
2 = 1,
1 + η33µ44s2
z53 = µ55s1,
z43 = µ44s2.
(30)
(31)
(32)
Reciprocity and symmetries with explicit dependence of constitutive tensor
on the wave vector. When the constitutive tensor is assumed a function of
the wave vector, the parity considerations should be used. The constitutive
tensor does not have to be diagonal, and in fact for asymmetric structures
off-diagonal terms may be necessary. Note that z43 is not accessible from
scattering, and in fact τ4 and v3 vary through the appropriate boundary in
the unit cell with complicated profiles. With a reasonable definition or calcu-
lation of z43 (e.g. considering scattering off an oblique cut or via integrating
along the thickness direction), more equations will be available to determine
constitutive parameters. Without further information however, at least one
of the parameters may be considered free. In the following the simplifying
Equations (10), (11), and (13) are used. The normal component of slow-
ness may be written as s1 = ±s1(ω, s2) based on the dispersion relation and
therefore any even function of s1 can be written as an even function of ω and
any odd function of s1 can be written as sign(s1)f (ω), with f (ω) vanishing
at zero frequency. With x1 symmetry, and based on Equation (16), κ35 will
have this latter (odd) form, which may be chosen as a free parameter and
set equal to zero. This is not feasible for a x1-asymmetric structure, but
it may be possible to get the value of κ35 for a normal incidence calculation
and use throughout. The other quantities can now be fully determined based
on the transfer matrix. With both x1- and x2-symmetries, and a first order
approximation in s2
5
µ44 ≈ µ44,0(ω), µ55 ≈ µ55,0(ω),
µ54(= −µ45) ≈ s2µ54,1(ω),
η33 ≈ η33,0(ω),
κ34 ≈ s2κ34,1(ω).
(33)
The left hand side of all thee approximations are general functions of ω and s2.
Even though we have set κ35 = 0, at each frequency there are infinitely many
5s2 has to be non-dimensionalized with some characteristic slowness s2/¯s for a formal
series consideration. But we neglect this process here since we only look at the first order
approximation.
12
potential constitutive functions that could match the observed dispersion and
impedances. Some cases in particular are of interest. First, one may assume
that the constitutive matrix is diagonal. Even in this case, there is no need
for η33 to be a function of the wave vector and after enforcing a non-local
form for it, one can calculate the diagonal functions as
µ55(ω, s2) =
z53(ω, s2)
s1(ω, s2)
,
η33(ω) =
1
µ55(ω, 0)s1(ω, 0)2 ,
1 − η33(ω)µ55(ω, s2)s1(ω, s2)2
η33(ω)s2
2
µ44(ω, s2) =
(34)
(35)
(36)
.
The second potential choice is to enforce that all of the diagonal quantities
are independent of the wave vector. In this case:
µ55 and η33 are easily determined as
z53 = µ55s1 + µ54s2.
µ55(ω) =
z53(ω, 0)
s1(ω, 0)
,
η33(ω) =
1
µ55(ω)s1(ω, 0)2 .
(37)
(38)
(39)
Furthermore,
µ54(ω, s2) =
z53(ω, s2) − µ55(ω)s1(ω, s2)
s2
.
(40)
To determine the other two functions, one may start with the dispersion
relation
η33µ55s2
1 + (η33µ44 + κ2
34)s2
2 = 1,
(41)
1 + s(cid:48)2
1 + µ44/µ55 + [(κ34κ(cid:48)(cid:48)
34 + κ(cid:48)2
then divide by η33µ55, and implicitly differentiate twice to get
s1s(cid:48)(cid:48)
34]/(η33µ55) = 0, (42)
where (cid:48) represents differentiation with respect to s2. At s2 = 0 this equation
is simplified by noting that κ34(ω, 0) = 0, since it is an odd function of s2,
and s(cid:48)
1(ω, 0) = 0, since s1(ω, s2) is an even function of s2. Therefore
2 + 4κ34κ(cid:48)
34s2 + κ2
34)s2
µ44(ω) = −µ55(ω)s1(ω, 0)s(cid:48)(cid:48)
1(ω, 0).
(43)
13
Finally
κ2
34(ω, s2) =
1 − η33(ω)[µ55(ω)s1(ω, s2)2 + µ44(ω)s2
2]
s2
2
.
(44)
from z43) but is irrelevant at this point.
The particular root of κ34 may not be determined without further information
(e.g.
It is possible to evaluate
the goodness of first order approximations µ54(= −µ45) ≈ s2µ54,1(ω) and
κ34 ≈ s2κ34,1(ω) at this point. Similarly, one may evaluate the goodness
of second order fits µ44 ≈ µ44,0 + s2
2µ55,2 in the
diagonal description. In summary, in this scenario, κ2
34, η33 and µ44 can only
be obtained using dispersion relation, Equation (41), and µ55 and µ54 are
merely extractable from impedance equation, Equation (37). This avoids
any inconsistency that may occur if the equations are coupled.
2µ44,2 and µ55 ≈ µ55,0 + s2
Such an inconsistency may occur if instead of assuming free parameter
κ35 = 0, one instead opts for selecting κ34 as the free parameter and set it to
zero, while keeping the diagonal elements independent of the wave vector. In
such a case, the dispersion relation in slowness domain, Equation (14), and
impedance z53, Equation (16), will simplify further to:
(η33µ55 + κ2
35)s2
1 + η33µ44s2
2 = 1,
z53 = µ55s1 + µ54s2 + (κ35/η33)(1 + κ35s1).
(45)
(46)
Having in mind the x1-symmetry of the system, which leads to κ35 being an
odd function of s2, the values of µ55 and η33 can be determined exactly from
Equations (38), (39). However, the value of µ54 will be a function of κ35 as:
µ54(ω, s2) = {z53(ω, s2) − µ55(ω)s1(ω, s2)
− [κ35(ω, s2)/η33(ω)][1 + κ35(ω, s2)s1(ω, s2)]}/s2.
(47)
The equation for µ44 can be derived by differentiating the dispersion relation,
Equation (45), twice with respect to s2 and simplifying it at s2 = 0:
µ44(ω) = −µ55(ω)s1(ω, 0)s(cid:48)(cid:48)
1(ω, 0) − κ(cid:48)2
35(ω, 0)s2
η33(ω)
1(ω, 0)
.
(48)
Both these quantities are written in terms of κ35 and its derivatives.
In
order to determine κ35, one can differentiate the dispersion relation one more
time, i.e. a total of three times, to eliminate non-dispersive µ44(ω) and
14
solve the resulting equation numerically. However, unlike the previous case
where impedance z53 and κ34 were uncoupled, κ35 appears in the impedance
equation as well. A cumbersome analysis may be performed here leading
to a necessary consistency condition between the dispersion and impedance
equations, with the final form:
12z(cid:48)2
53(ω, 0)s1(ω, 0)s(cid:48)(cid:48)
1(ω, 0) + 4z(cid:48)
[µ55(ω)/η33(ω)] × [3s(cid:48)(cid:48)
53(ω, 0)z(cid:48)(cid:48)(cid:48)
2(ω, 0) + s1(ω, 0)s(cid:48)(cid:48)(cid:48)(cid:48)
53(ω, 0)s2
1
1(ω, 0)+
1 (ω, 0)] = 0.
(49)
In the following a simple example is analyzed and for the sake of brevity here
we state without reproducing the results that such a condition was violated
almost everywhere in the frequency domain. Therefore, the potential free
parametric choices stated earlier are indeed limited quite significantly by the
physics of the problem. As a final case, note that if one considers either
µ45 + µ54 = 0 as the free parameter and set it equal to zero, based on
Equation (13), one would still require either of κ34 or κ35 to vanish as well,
leading to one of the cases considered above.
4. Example: Determining the overall constitutive parameters
Here, the proposed method is applied to an example of a symmetric sys-
tem in order to expand the determination of the constitutive parameters into
the stop bands and upper pass bands. The periodic layered medium of inter-
est here is [PMMA, Brass, PMMA] with (dj) = (2.5, 0.2, 2.5), (ρj) = (1, 8, 1),
and (µj) = (1.2, 40, 1.2) which is shown in Figure (1). The PMMA phase
is modeled as a lossy material with 5% loss in terms of wave speed, i.e.
3 = −0.05 (≈ 10% loss tangent in modulus). In the following,
1/c(cid:48)
c(cid:48)(cid:48)
the unit system [mm, µs, mg] for length, time, and mass are used which re-
sults in [km/s, GPa, g/cm3, MHz, MRayl] units for velocity, stress, density,
frequency, and impedance, respectively. The complex reflection and trans-
mission coefficients into PC are shown in Figure (2) for a single unit cell.
θ = 0, π/6 graphs for a five unit cell slab are also added in order to observe
how increasing the number of unit cells can affect the scattering results.
1 = c(cid:48)(cid:48)
3/c(cid:48)
Dispersion, wave vector, and impedance. Figures (3) and (4) depict the dis-
persion results extracted from the scattering data based on Equation (26).
Note that the results are independent of the number of unit cells in the slab
nc. Furthermore, using the eigenvalue method in Equation (25) also gives the
15
Figure 1: The schematic view of the periodic layered system with a symmetric 2-phase
unit cell of acrylic (PMMA) and brass. The incidence and transmission media are selected
as polycarbonate (PC). Arrows show the oblique SH incident wave along with the reflection
and transmission waves.
same exact wave vector component k1 and impedance z53. Here the normal-
ized phase advance, k1d, is shown as a function of frequency, f , for different
values of s2. The phase ambiguity is removed by adding 2π whenever needed
to maintain continuity, leading to a single positive phase advance solution
0 ≤ (cid:60)(k1d) ≤ ∞. Due to the symmetry of the structure, the second solution
is simply the negative of this result. In general, for higher values of s2 = k2/ω
the stop bands, associated with peaks in (cid:61)(k1d), become wider. However,
for the incident angle θ = π/6, the PC material constants requires s2 = 0.45,
for which the scattering calculation leads to consistently high amplitude of
transmission and low reflection. It is also observed that at this value of s2,
k1d phase advance is almost exactly a linear function of frequency, essentially
rendering the medium non-dispersive, i.e leading to constant s1 and z53 in
terms of frequency.
Figure (5) shows two representations of the dispersion surfaces. One can
observe that for high amplitudes of s2 > 0.5, (cid:61)(s1) increases significantly.
The sign of (cid:60)(s1) is changing for s2 > 0.789; see the insets. The isofrequency
contours appear to converge around s ≈ (±(0.750 + 0.049i),±0.447)T . This
16
(a)
(c)
(b)
(d)
Figure 2: (a) and (c) show the magnitude of reflection and transmission, as functions of
frequency, f , for different values of s2 (nc = 1). At θ = 0, π/6 graphs for 5 unit cells
(nc = 5, hollow markers) are also shown. (b) and (d) show the phase angle of the same
quantities. At θ = π/6 incident angle, or s2 = 0.447, the reflection is substantially lower in
comparison to the other incident angles, giving the appearance of an impedance matched
system.
is where the system appears non-dispersive. The value of impedance around
this point is also nearly independent of frequency z53 = 0.9563 − 0.0328i.
Constitutive functions. In both approaches discussed above, the inverse den-
sity quantity, η33, may be chosen to be independent of wave vector or slowness
s2 (Equations (35),(39) which give the same numerical value). The results
are shown in Figure (6). At very low frequencies, the overall density of the
media should match a simple volume average, which translates into the Reuss
average formula for η33. This is shown in Fig (6), where (cid:60)(η33) is tangent to
this value and (cid:61)(η33) → 0 at low frequencies.
For a diagonal constitutive tensors, the stiffness quantities will have to
be functions of the wave vector. In this case Equations (34) and (36) will be
used. As it can be seen from Figure (7), the value of µ55 is a strong function of
propagation direction and even becomes relatively constant around s2 = 0.45.
µ44 is shown in Figure (8) and also demonstrates significant non-locality and
17
SSSS (a)
(c)
(b)
(d)
Figure 3: (a) Real and (b) imaginary parts of the normalized wave number k1d, as functions
of frequency, f for different values of s2. Due to the symmetry of the system, the other
solution would be the negative of the presented values.
(c) and (d) are the real and
imaginary parts of the slowness component s1 = k1/ω, respectively. The dependence of
s1 on frequency demonstrates nonlocal physics, as otherwise the slowness will be constant
in frequency. The only such special case is happening at s2 = 0.45. The phase ambiguity
is removed through maintaining continuity of k1d by adding 2π when necessary.
(a)
(b)
Figure 4: (a) and (b) show the real and imaginary parts of the impedance, z53, as functions
of frequency, f , for different values of s2, respectively. Again, the impedance appears as
independent of frequency only at s2 = 0.45.
sensitivity to the wave vector, especially in and around stop bands.6
6µ44 is not calculated for s2 = 0 (Equation 36), and therefore is shown starting from
18
(a)
(c)
(b)
(d)
Figure 5: (a) and (b) are the contour plot representations of the dispersion surfaces for
real and imaginary parts of k1 vs. k2 and frequency (contour values). (c) and (d) show
isofrequency contours in the 2D spaces of s2 and real and imaginary parts of s1, respec-
tively. In (c) it is shown that for f = 0.01, (cid:60)(s1) is changing sign at s2 = ±0.789 which
results in (cid:60)(k1) sign change at k2 = ±0.05, (a).
In a spatially non-dispersive system,
the slowness graphs will be independent of frequency, therefore this graphs will only show
a single curve for SH waves. Note that all different contours converge near s2 = 0.45.
In other words, the system is nearly non-dispersive in the neighborhood of this point
s ≈ (±(0.7497 + 0.0493i),±0.4470)T . At this point z53 = 0.9563 − 0.0328i.
For the case when µ44 and µ55 are assumed independent of the wave
vector, their values match the s2 → 0 functions in the previous calculation.
They also approach Voigt and Reuss averages of the shear moduli of the
layers, respectively; See Figure (9). However, in this scenario, µ54 is non-zero
and a function of wave vector and can be calculated using Equation (40).
Figure (10) illustrates the variations of µ54 with respect to frequency for
different s2 values. The coupling function κ34 is also non-zero, a function
of the wave vector, and can be calculated from Equation (44). Note that
s2 = 0.001.
19
Figure 6: The real and imaginary parts of η33, as functions of frequency, f , are shown
here. The overall η33 would be equal to the Reuss average of component layers at very
low frequencies.
(a)
(b)
Figure 7: (a) and (b) show the real and imaginary parts of the µ55 as functions of frequency,
f , for different values of s2, respectively, when a diagonal constitutive tensor is used.
(a)
(b)
Figure 8: (a) and (b) show the real and imaginary parts of µ44 as functions of frequency,
f , for different values of s2, respectively, when a diagonal constitutive tensor is used.
there are two possible roots for κ34. For higher values of frequency, presence
of branch cuts make it difficult to enforce continuity of κ34. The exact root
20
selection may be resolved if z43 is known. There is, however, no ambiguity
in κ2
34 and the overall scattering and dispersion of the the medium is only
dependent on it. One continuous root selection is shown in Figure (11),
while its negative is similarly acceptable in this analysis. µ54 and κ34 are
odd function of s2 and can get linearized for small enough s2. The values
of µ54,1 and κ34,1 (one root) from Equation (33) are shown in Figure (12).
Note that the nonzero value of (cid:61)(κ34,1) at small frequencies indicate that
the nonlocality of the system is inherently not removable even as f → 0 and
s2 → 0.
(a)
(b)
Figure 9: (a) and (b) show the real and imaginary parts of the moduli µ55 and µ44 as
functions of frequency, f , when they are considered independent of the wave vector. It
should be noted that in this case, Reuss and Voigt averages provide accurate estimates of
overall, µ55 and µ44, respectively at very low frequencies as one expects for quasi-static
situations.
(a)
(b)
Figure 10: (a) and (b) show the real and imaginary parts of the moduli, µ54 as functions
of frequency, f .
One can provide an estimate for z43 based on the formulation here, even
though the exact value is not accessible based on scattering measurements. In
21
(a)
(b)
Figure 11: (a) and (b) show the real and imaginary parts of the coupling term, κ34, as
functions of frequency, f . Note that the negative of these values are also acceptable roots
based on the present analysis.
(a)
(b)
Figure 12: The real and imaginary parts of the (a) µ54,1 and (b) κ34,1 (one root), as
functions of frequency, f . Note the nonzero value of (cid:61)(κ34,1) at zero frequency indicating
the nonlocality of the system is not removable even as f → 0 and s2 → 0.
the diagonal tensor case, there is no ambiguity involved in calculating the z43
based on µ44, while the exact value of κ34 (and not just its square) is needed in
the other case. These estimates may be compared with the ones calculated
using an integration scheme of the wave equation to further elucidate the
acceptable mathematical approaches to overall properties determination of
the physical heterogeneous problem.
5. Summary and conclusions
The scattering of oblique SH waves from a periodic layered slab may
be calculated easily using the transfer matrix of such systems. It is shown
here that one can use the scattering data to calculate overall constitutive
tensors for any such symmetric slab within the coupled Willis formalism.
22
However, the presence of micro-structure immediately necessitates the de-
pendence of the constitutive tensors on the wave vector, even for simplest
cases. Without further assumptions on the constitutive structure and/or
other scattering analysis, one may not uniquely determine the full tensor.
A number of physically and mathematically attractive assumptions are used
here to demonstrate the process for a simple example. It is observed that
not all potential selections of free parameters are analytically acceptable, e.g.
κ35 = 0 may be identically assumed for symmetric structures, while κ34 = 0
leads to inconsistent results. Other 3D structural symmetries not considered
here may further limit the potential constitutive descriptions. Interestingly,
one can identify certain non-trivial points in the wave vector space, around
which the system becomes nearly non-dispersive. Further analysis of such
simple layered systems may require one or more of the following approaches.
One may consider the scattering of coupled P/SV waves in a similar man-
ner and utilize the expected symmetry of the response that demand more
restrictive forms on the quantities that were discussed here. For example,
the values of shear moduli for a transversely isotropic system are expected
to match for both SH and P/SV waves. Alternatively, one can utilize field
integration techniques to analyze through thickness quantities such as τ4 and
z43. Finally and in a sense related to the previous approach, one may con-
sider the physical problem of scattering off surfaces other than the parallel
plane faces of the layers. The last approach however brings an extra level of
complexity as one has to consider non-specular scattering.
23
REFERENCES
References
[1] L. D. Landau and E. M. Lifshitz. Electrodynamics of continuous media.
Pergamon Press, 2nd edition, 1984.
[2] V. M. Agranovich and V. Ginzburg. Crystal Optics with Spatial
Springer Series in Solid-State Sciences.
Dispersion, and Excitons.
Springer-Verlag, Berlin Heidelberg, 2 edition, 1984.
[3] A. A. Maradudin and D. L. Mills. Effect of Spatial Dispersion on the
Properties of a Semi-Infinite Dielectric. Physical Review B, 7(6):2787 --
2810, 1973.
[4] V. S. Asadchy, M. Albooyeh, S. N. Tcvetkova, A. D´ıaz-Rubio, Y. Ra'Di,
and S. A. Tretyakov. Perfect control of reflection and refraction using
spatially dispersive metasurfaces. Physical Review B, 94(7), 2016.
[5] A. D´ıaz-Rubio, V. S. Asadchy, A. Elsakka, and S. A. Tretyakov. From
the generalized reflection law to the realization of perfect anomalous
reflectors. Science Advances, 3(8), 2017.
[6] A. S. Shalin, P. Ginzburg, A. A. Orlov, I. Iorsh, P. A. Belov, Y. S.
Kivshar, and A. V. Zayats. Scattering suppression from arbitrary ob-
jects in spatially dispersive layered metamaterials. Physical Review B -
Condensed Matter and Materials Physics, 91(12), 2015.
[7] A. R. Melnyk and M. J. Harrison. Resonant Excitation of Plasmons
in Thin Films by Elecromagnetic Waves. Physical Review Letters,
21(2):85 -- 88, 1968.
[8] W. E. Jones, K. L. Kliewer, and R. Fuchs. Nonlocal Theory of the Op-
tical Properties of Thin Metallic Films. Physical Review, 178(3):1201 --
1203, 1969.
[9] A. R. Melnyk and M. J. Harrison. Theory of Optical Excitation of
Plasmons in Metals. Physical Review B, 2(4):835 -- 850, 1970.
[10] R. Ruppin. Optical properties of small metal spheres. Physical Review
B, 11(8):2871 -- 2876, 1975.
24
[11] V. M. Agranovich and Y. N. Gartstein. Spatial dispersion and negative
refraction of light. Physics-Uspekhi, 49(10), 2006.
[12] R. Chern. Spatial dispersion and nonlocal effective permittivity for pe-
riodic layered metamaterials. Optics Express, 21(14):85 -- 88, 2013.
[13] A. Al`u. Restoring the physical meaning of metamaterial constitutive pa-
rameters. Physical Review B - Condensed Matter and Materials Physics,
83(8), 2011.
[14] P. Alitalo, A. E. Culhaoglu, C. R. Simovski, and S. A. Tretyakov. Ex-
perimental study of anti-resonant behavior of material parameters in
periodic and aperiodic composite materials. Journal of Applied Physics,
113(22), 2013.
[15] P. Li, S. Yao, X. Zhou, G. Huang, and G. Hu. Effective medium theory of
thin-plate acoustic metamaterials. The Journal of the Acoustical Society
of America, 135(4):1844 -- 1852, 2014.
[16] J. Vehmas, S. Hrabar, and S. Tretyakov. Omega transmission lines with
applications to effective medium models of metamaterials. Journal of
Applied Physics, 115(13), 2014.
[17] S. Papantonis, S. Lucyszyn, and E. Shamonina. Dispersion effects
in Fakir's bed of nails metamaterial waveguides. Journal of Applied
Physics, 115(5), 2014.
[18] P. A. Belov, R. Marqu´es, S. I. Maslovski, I. S. Nefedov, M. Silveirinha,
C. R. Simovski, and S. A. Tretyakov. Strong spatial dispersion in wire
media in the very large wavelength limit. Physical Review B, 67(11),
2003.
[19] C. Rockstuhl, C. Menzel, T. Paul, T. Pertsch, and F. Lederer. Light
propagation in a fishnet metamaterial. Physical Review B, 78(15), 2008.
[20] Y. Tyshetskiy, S. V. Vladimirov, A. E. Ageyskiy, I. I. Iorsh, A. Orlov,
and P. A. Belov. Guided modes in a spatially dispersive wire medium
slab. Journal of the Optical Society of America B, 31(8), 2014.
25
[21] J. J. Miret, J. Aitor Sorn´ı, M. Naserpour, A. Ghasempour Ardakani, and
C. J. Zapata-Rodr´ıguez. Nonlocal dispersion anomalies of Dyakonov-
like surface waves at hyperbolic media interfaces. Photonics and
Nanostructures - Fundamentals and Applications, 18:16 -- 22, 2016.
[22] S. Yves, R. Fleury, T. Berthelot, M. Fink, F. Lemoult, and G. Lerosey.
Crystalline metamaterials for topological properties at subwavelength
scales. Nature Communications, 8, 2017.
[23] A. A. Ushkov and A. A. Shcherbakov. Concurrency of anisotropy and
spatial dispersion in low refractive index dielectric composites. Optics
Express, 25(1), 2017.
[24] J. J. Hopfield and D. G. Thomas. Theoretical and experimental effects of
spatial dispersion on the optical properties of crystals. Physical Review,
132(2):563 -- 572, 1963.
[25] D. L. Portigal and E. Burstein. Acoustical activity and other first-order
spatial dispersion effects in crystals. Physical Review, 170(3):673 -- 678,
1968.
[26] S. A. Tretyakov. Uniaxial Omega Medium as a Physically Realiz-
able Alternative for the Perfectly Matched Layer (Pml). Journal of
Electromagnetic Waves and Applications, 12(6):821 -- 837, 1998.
[27] A. Ciattoni and C. Rizza. Nonlocal homogenization theory in metamate-
rials: Effective electromagnetic spatial dispersion and artificial chirality.
Physical Review B - Condensed Matter and Materials Physics, 91(18),
2015.
[28] G. S. Agarwal, D. N. Pattanayak, and E. Wolf. Electromagnetic fields in
spatially dispersive media. Physical Review B, 10(4):1447 -- 1475, 1974.
[29] A. Puri and J. L. Birman. Pulse propagation in spatially dispersive
media. Physical Review A, 27(2):1044 -- 1052, 1983.
[30] P. A. Belov, C. R. Simovski, and S. A. Tretyakov. Example of bian-
isotropic electromagnetic crystals: The spiral medium. Physical Review
E, 67(5), 2003.
26
[31] A. D. Yaghjian, A. Al`u, and M. G. Silveirinha. Homogenization of spa-
tially dispersive metamaterial arrays in terms of generalized electric and
magnetic polarizations. Photonics and Nanostructures - Fundamentals
and Applications, 11(4):374 -- 396, 2013.
[32] B. Popa and S. A. Cummer. Design and characterization of broadband
acoustic composite metamaterials. Physical Review B, 80(17), 2009.
[33] A. Castani´e, J. F. Mercier, S. F´elix, and A. Maurel. Generalized method
for retrieving effective parameters of anisotropic metamaterials. Optics
Express, 22(24):29937 -- 29953, 2014.
[34] J. H. Park, H. J. Lee, and Y. Y. Kim. Characterization of anisotropic
acoustic metamaterial slabs. Journal of Applied Physics, 119(3), 2016.
[35] D. Lafarge and N. Nemati. Nonlocal Maxwellian theory of sound prop-
agation in fluid-saturated rigid-framed porous media. Wave Motion,
50(6):1016 -- 1035, 2013.
[36] N. Nemati and D. Lafarge. Check on a nonlocal Maxwellian theory of
sound propagation in fluid-saturated rigid-framed porous media. Wave
Motion, 51(5):716 -- 728, 2014.
[37] N. Nemati, A. Kumar, D. Lafarge, and N. X. Fang. Nonlocal descrip-
tion of sound propagation through an array of Helmholtz resonators.
Comptes Rendus M´ecanique, 343(343):656 -- 669, 2015.
[38] Navid Nemati, Yoonkyung E. Lee, Denis Lafarge, Aroune Duclos,
and Nicholas Fang. Nonlocal dynamics of dissipative phononic fluids.
Physical Review B, 95(22), June 2017.
[39] H. J. Lee, H. S. Lee, P. S. Ma, and Y. Y. Kim. Effective material
parameter retrieval of anisotropic elastic metamaterials with inherent
nonlocality. Journal of Applied Physics, 120(10), 2016.
[40] A. V. Amirkhizi. Homogenization of layered media based on scattering
response and field integration. Mechanics of Materials, 114:76 -- 87, 2017.
[41] S. Nemat-Nasser. Anti-plane shear waves in periodic elastic composites:
band structure and anomalous wave refraction. Proceedings of the Royal
Society A: Mathematical, Physical and Engineering Science, 471(2180),
2015.
27
[42] A. Srivastava. Metamaterial properties of periodic laminates. Journal
of the Mechanics and Physics of Solids, 96:252 -- 263, 2016.
[43] P. C. Vinh, T. T. Tuan, and M. A. Capistran. Explicit formulas for the
reflection and transmission coefficients of one-component waves through
a stack of an arbitrary number of layers. Wave Motion, 54:134 -- 144,
2015.
[44] G. W. Milton and J. R. Willis. On modifications of Newton's second law
and linear continuum elastodynamics. Proceedings of the Royal Society
A: Mathematical, Physical and Engineering Sciences, 463(2079):855 --
880, 2007.
[45] J. R. Willis. Exact effective relations for dynamics of a laminated body.
Mechanics of Materials, 41(4):385 -- 393, 2009.
[46] W. Nantasetphong, Z. Jia, M.A. Hasan, A.V. Amirkhizi, and S. Nemat-
Nasser. A new technique for characterization of low impedance materials
at acoustic frequencies. Experimental Mechanics, Jul 2018.
28
|
1907.10755 | 1 | 1907 | 2019-05-25T01:04:06 | Robust quantitative single-exposure laser speckle imaging with true flow speckle contrast in the temporal and spatial domains | [
"physics.app-ph",
"physics.optics"
] | A systematic and robust laser speckle contrast imaging (LSCI) method and procedure is presented, covering the LSCI system calibration, static scattering removal, and measurement noise estimation and correction to obtain a true flow speckle contrast and the flow speed from single-exposure LSCI measurements. We advocate to use as the speckle contrast instead of the conventional contrast K as the former relates simply to the flow velocity and is with additive noise alone. We demonstrate the efficacy of the proposed true flow speckle contrast by imaging phantom flow at varying speeds, showing that (1) the proposed recipe greatly enhances the linear sensitivity of the flow index (inverse decorrelation time) and the linearity covers the full span of flow speeds from 0 mm/s to 40 mm/s; and (2) the true flow speed can be recovered regardless of the overlying static scattering layers and the type of speckle statistics (temporal or spatial). The fundamental difference between the apparent temporal and spatial speckle contrasts is further revealed. The flow index recovered in the spatial domain is much more susceptible to static scattering and exhibit a shorter linearity range than that obtained in the temporal domain. The proposed LSCI analysis framework paves the way to estimate the true flow speed in the wide array of laser speckle contrast imaging applications. | physics.app-ph | physics | Robust quantitative single-exposure laser
speckle imaging with true flow speckle
contrast in the temporal and spatial domains
CHENGE WANG,1,# ZILI CAO,1,# XIN JIN,1 WEIHAO LIN,1 YANG ZHENG,1
BIXIN ZENG,1 AND M. XU1,2,*
1 Institute of Lasers and Biophotonics, School of Biomedical Engineering, Wenzhou Medical University,
Wenzhou, Zhejiang, China 325035
2Department of Physics and Astronomy, Hunter College, The City University of New York, 695 Park
Avenue, New York, NY 10065
#: These authors equally contributed to the work.
*[email protected]
Abstract: A systematic and robust laser speckle contrast imaging (LSCI) method and
procedure is presented, covering the LSCI system calibration, static scattering removal, and
measurement noise estimation and correction to obtain a true flow speckle contrast
and the
flow speed from single-exposure LSCI measurements. We advocate to use
as the speckle
contrast instead of the conventional contrast K as the former relates simply to the flow velocity
and is with additive noise alone. We demonstrate the efficacy of the proposed true flow speckle
contrast by imaging phantom flow at varying speeds, showing that (1) the proposed recipe
greatly enhances the linear sensitivity of the flow index (inverse decorrelation time) and the
linearity covers the full span of flow speeds from 0 mm/s to 40 mm/s; and (2) the true flow
speed can be recovered regardless of the overlying static scattering layers and the type of
speckle statistics (temporal or spatial). The fundamental difference between the apparent
temporal and spatial speckle contrasts is further revealed. The flow index recovered in the
spatial domain is much more susceptible to static scattering and exhibit a shorter linearity range
than that obtained in the temporal domain. The proposed LSCI analysis framework paves the
way to estimate the true flow speed in the wide array of laser speckle contrast imaging
applications.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1.
Introduction
When coherent light interacts with a turbid medium, the interference between the outgoing
waves produces grainy speckle patterns which encode the phase fluctuation of all rays (random
phasors) reaching a point. The contrast of laser speckles reduces with the motion of scatterers
inside the turbid medium. Laser speckle contrast hence can be used to infer the dynamic
property of the medium. Laser speckle contrast imaging (LSCI, see recent reviews [1-4]) has
now been widely used in monitoring blood flow in brain, skin, retina, arthrosis and etc due to
advantages including simplicity, high spatial and temporal resolution, and large field of view
without scanning [5-8].
Although LSCI has a wide range of applications and a long history, the recovery of absolute
flow velocity from LSCI measurements remains a challenge, especially when the measurement
is compounded by static scattering and noise. For static scattering in laser speckle imaging, Li
et al. showed that the static scattering effect can be partially suppressed by using the temporal
rather than spatial contrast analysis of laser speckles [9] as the static scattering is an invariant
quantity with time. Zakhraov et al. [10, 11] presented a data processing scheme to correctly
separate dynamic and static components within the speckle contrast based on their different
decorrelation behaviour across speckle patterns captured at consecutive times. Dunn et al. [6,
2fK2K 12-14] later demonstrated a multi-exposure laser speckle contrast imaging method, which
quantifies and eliminates the influence of static scattering from speckle contrasts measured
under different exposure times using a laser speckle contrast model. For LSCI measurement
noise, the correction of the variance of the shot noise and sensor dark currents were found to
be crucial to estimate the true speckle contrast [15, 16]. Yuan et al. [16] increased the signal-
to-noise ratio (SNR) of LSCI with noise correction to detect small blood flow changes caused
by brain activity. A systematic study and recommended practical recipe to obtain true flow
velocity from LSCI measurements addressing both static scattering and measurement noise is,
however, still lacking.
In this article, we analysed laser speckle flow imaging from the first principle and provided
a complete procedure covering the LSCI system calibration, static scattering removal, and
measurement noise estimation and correction to obtain a genuine flow speckle contrast and the
flow speed from single-exposure LSCI measurements. We demonstrated the power of our LSCI
analysis recipe by imaging phantom flow at varying speeds. Experimental results show that our
procedure greatly enhances the linear sensitivity of the flow index (defined as the inverse
decorrelation time) and the linearity covers the full span of flow speeds from 0 mm/s to 40
mm/s. The true flow speed is recovered regardless of the overlying static scattering layers and
the type of statistics (temporal or spatial). The proposed LSCI analysis framework hence paves
the way to estimate the true flow speed in the wide array of laser speckle contrast imaging
applications.
2. Theory and Data Analysis
2.1 Theoretical basis
The spatial intensity distribution of the speckle pattern fluctuates with the motion of the
scattering particles under the illumination of coherent light. The recorded pattern by a camera
is the integration of all instantaneous speckles over the exposure time. The faster the scattering
particles move, the more blurred the recorded pattern becomes. The degree of blurring is
quantified by the contrast [17] given by
(1)
represents the mean of light intensity I over a small region (spatial contrast) or over a
where
short durance of time (temporal contrast). For "fully developed" static speckles, the spatial
contrast K equals to 1.
We will assume the scattered electric field containing both dynamic and static components
(2)
with ω being the angular frequency of light. The dynamic component consists of photons which
have at least been scattered by moving scatterers (flow) once and the static component consists
of photons being scattered by static scatterers alone.
The electric field temporal autocorrelation function can be written as
(3)
where
means average over t,
is the electric field temporal
autocorrelation function related to flow, and
.
2()IIKIItt()[()]iifsEteEtEe*1()()()SEtEtGI*1()()()ffGEtEt2ssIE In practice only light intensity fluctuation signals can be recorded. The intensity
autocorrelation function is defined as
where
and
assuming the dynamic and static electric fields are uncorrelated, i.e.,
. In terms of the normalized dynamic electric field and full intensity
autocorrelations
expresses
and
, the full Siegert relation [18, 19]
(4)
where
,
is a parameter that accounts for the reduction in the measured contrast
due to averaging (by the detector) over uncorrelated speckles. Note
[19].
is real and non-negative
The
speckle
contrast
under
an
exposure
time
T,
expressed
as
, reduces now to
(5)
where the average intensity
and the dynamic fraction
. We will
as the speckle contrast instead of the conventional contrast K as the former relates
use
simply to the flow velocity and is with additive noise alone.
Equation (5) is the theoretical temporal contrast from the random process taken by the
photons migrating through a turbid medium. Some complexities arise when evaluating
from
measurement data. First the measurement noise (of zero mean) introduces an extra variance
. Second when using spatial ensemble average for the evaluation of the contrast rather
term
than temporal statistics, the extra terms appearing in
will be
with
being the
non-ergodic contribution from the static field. This motivates us to introduce the dynamic
(flow) contrast
defined in terms of the dynamic component alone.
The measured speckle contrast can be expressed as
in the temporal domain and
(6)
(7)
in the spatial domain. The value of
and
can be evaluated from calibration and
measurement data as shown later. The dynamic contrast remains the same when evaluated in
either the temporal or spatial domain.
A velocity distribution model for the moving particles is needed to relate
to the flow
speed. With the commonly used Lorentz velocity distribution model, the dynamic electric field
autocorrelation function can be written as
[12], yielding
2()()()GItItfsItItI2ffItEt*()0fsEtE11()()/fgGI222()()/(0)gGG22211212()ffsfsgIgIIgII()ffIIt11g222121200()()/1TTKTItItdtdtI222110022(1)()2(1)(1)()TTKgdgdTTTTfsIII/ffsIII2K2K2noise2K22noisene2ne222/1fffKII2222noisefKK22222noisenefKK2noise2ne2fK1)()exp(/cgt (8)
where
and
is the decorrelation time. The inverse decorrelation time increases with
the flow speed and can be regarded as the flow index.
In the next section, we will examine system calibration, sample measurement and data
analysis to provide a complete procedure for static scattering removal, and measurement noise
estimation and correction to obtain a true flow speckle contrast and the flow speed from single-
exposure LSCI measurements.
2.2 System calibration, measurement, and data analysis
Let's consider a set of speckle images
(
) at time
with an exposure time T,
i.e.,
(9)
Here the recorded image consists of the static component
, the dynamic component
and
the noise n. The noise [15] presented in the measurement is mainly comprised of the dark counts
and the shot noise
, i.e.,
.
The dark counts
and the variance of dark counts
can be easily
acquired by taking multiple dark frames at the same exposure time and camera gain (with all
light off). One could use the temporal average to get the dark count and its variance pixel by
pixel when the number of the dark frames is large (
) or use the spatial average over a
=7) pixel window otherwise. If the behavior of dark counts is
sliding
(typical
assumed uniform across the whole sensor frame, the mean and the variance of the dark counts
are given by further averaging over the whole sensor frame. In many cameras, the recorded
intensity has been pre-subtracted by certain base. In this case,
should be estimated by the
median value and the variance
where
with the negative
values of
replaced by zeros.
We would always subtract the dark counts from
before further analysis. After this pre-
processing, the speckle image becomes
and the noise term is
replaced by
. The noise
satisfies
and
. The shot noise obeys a Poisson distribution with the mean
and a variance equal to
as the camera converts the photoelectrons to
digital counts where
is the analog to the digital conversion factor[20]. The
factor is
typically the same across the sensor frame and thus is obtained by further averaging over the
sensor frame. Under such a shot noise model,
(10)
21221112112exp(2)12exp()14(1)221(21)(1)136171(43)(4),12,fxxxxKxxxxxxxx/cxTciI1iNititi()()()(),,,,isfiIxyIxyIxynxysIfIdnsn (,) (,) (,)idisinxynxynxydn22()dddVarnnn50PPNNPNdn2()2'ddVarnn'dddnnn'dniI(,)(,)(,)ciidIxyIxynxy(,)(,)(,)ciidnxynxynxy0cincisidVarnVarnVarn0sin(,)/ciIxy2()(,)(,)/()ccciiidVarnIxyIxyVarn where an extra quadratic term in
is added to account for other noise sources such as
the laser fluctuations.
The temporal and spatial averages of the sequence of single-exposure speckle images then
satisfy:
and
with temporal statistics and
and
(11)
(12)
(13)
(14)
(15)
(16)
with spatial statistics. Here
with
,
means averaging over the N temporal
instances and
means the spatial average over a sliding
pixel window. We have
used the fact
due to the ergodic nature of the dynamic component.
When the time difference satisfies
, the complete decorrelation between the
dynamic component measured at two different times ensures the important identities
and
as
is a zero-mean Gaussian variable.
These identities could serve as the data consistency check.
One important consequence is that for any two speckle images
and
taken at times
satisfying
, the following holds
(17)
where the coherence factor of the imaging system is defined as
(18)
associated with spatial averaging over the sliding window of size
. Equation (17) is also
correct for a pure static scattering sample producing fully developed speckles (ρ = 0) as long as
.
(,)ciIxy(,)(,)(,)cisfiiiIxyIxyIxy2(,)(,)(,)(,)(,)2(,)(,)ccijsfifjsfiiiiIxyIxyIxyIxyIxyIxyIxy2222(,)(,)(,)(,)2(,)(,)ccisfiisfiiiiiIxyIxyIxynxyIxyIxy(,)(,)(,)cisfixyxyxyIxyIxyIxy2(,)(,)(,)(,)(,)2(,)(,)ccijsfifjsfixyxyxyxyxyIxyIxyIxyIxyIxyIxyIxy2222(,)(,)(,)2(,)(,)(,)cisfisfiixyxyxyxyxyxyIxyIxyIxyIxyIxynxyji0ixy(,)(,)fifjxyxyIxyIxyjicttT,,(,)(,)(,)(,)fififiiiifiIxyIxyIxyIxy(,)(,)(,)(,)yfifjfifxxjyxyIxyIxyIxyIxy,fiExyiIjIcjittT22(,)(,)(,)(,)()(1)(,)cccciixyxyxypciyjxjIxyIxyIxyIxyNIxy22(,)()1(,)sxypsxyIxyNIxyPNjiPPNN 2.2.1 System calibration
As stated earlier, the dark counts
and the variance of dark counts
is first acquired
by taking multiple dark frames at the identical experimental condition (the same exposure
time, camera gain and etc) with all light off. The other system parameters (the coherence
factor
and the behavior of
vs
) can be directly evaluated from a set of fully
developed speckle images taken on a pure static scattering sample such as a reflection
standard. Indeed, according to Eq. (17), we have
(19)
for
in this case. Here the spatial average should use the largest window size (full image
if possible) due to the reason discussed in [21] for the temporal speckle contrast. For the spatial
speckle contrast
in Eq. (19) should take the same value used for the contrast calculation.
The
factor is typically the same across the sensor frame and thus obtained by further
averaging over the sensor frame.
The noise variance
associated with a particular
can be found as
or
(20)
(21)
with the temporal or spatial statistics, respectively. Multiple sets of such speckle images under
the identical experimental condition and varying incident intensities are measured to cover the
full range of
. By fitting to Eq. (10), the system noise behavior is then characterized.
We note the above results on
and
should be independent of the
. In
reality a slight dependence on ∆ can be observed due to the inevitable system instability. In
this case, the correct values of
and
are obtained by extrapolating to
.
2.2.2 Sample measurement and data processing
The sample containing both dynamic and static scatterers are then imaged under the identical
experimental condition to yield a new set of dark current removed speckle images
(
).
The dynamic fraction
can be determined using Eq. (17) as
(22)
for
.
The noise level can also be estimated directly from the set of speckle images via
dn()dVarncVarncI(,)(,)()1(,)(,)cciixypcciixyxyIxyIxyNIxyIxy0PNcVarncI22()(,)(,)iciiiccVarnIxyIxy22()(,)(,)(,)(,)yiiiicccccxyxyxVarnIxyIxyIxyIxycIcVarn0cVarn/2Tt(,)ciIxy1iN22(,)(,)1(1)1()(,)cciixycpixyIxyIxyNIxyiicttT in the temporal domain or
in the spatial domain (they are equal by ergodicity).
Using Eq. (13) and (16), the temporal speckle contrast is then
(23)
(24)
and the spatial speckle contrast is then
(25)
(26)
from which we can identify the non-ergodic contribution from the static field to be
(27)
Finally, the velocity information of the sample can be obtained by solving the flow contrast
and fitting to Eq. (8) to obtain the decorrelation time
and the flow index.
Figure 1 outlines the complete procedure of system calibration, sample measurement,
noise correction, and static scattering removal for robust quantitative single-exposure laser
speckle imaging.
222(,)(,)/()(,)cciidiinoiseciiIxyIxyVarnIxy222(,)(,)/()(,)cciidxyxynoisecixyIxyIxyVarnIxy222222(,)1(,)ciifnoiseciiIxyKKIxy222222(,)1()(1)(,)cixyfpnoisecixyIxyKKNIxy22ne()(1)pN2fKcDetermine thesystem noise characteristicsDetermine the coherence factor, β Compute measurement noise contrast, Compute the dynamic fraction, ρ Compute the flow speckle contrast in temporal domains Compute theflow speckle contrast in spatial domains System calibrationSample measurementReflection standardFlowDynamic samples TargetCameraProcedure2()(,)(,)/()ccciiidVarnIxyIxyVarn Determine the decorrelation time, and flow speedDetermine camera dark current Fig. 1 Experimental and data analysis framework for robust quantitative single-exposure laser speckle
imaging.
System calibration first determines the camera dark current
, the coherent factor
, and
the noise parameters and from measuring fully developed speckles produced by a pure static
reflection standard. The true flow contrast
of dynamic samples are afterwards obtained by
removing the static scattering and correcting the measurement noise from the measured
temporal or spatial speckle contrasts. The flow decorrelation time and speed are then be
determined from
with a proper flow velocity model such as Eq. (8).
3. Results
3.1 Experimental setup
Figure 2 shows the schematic diagram of the experimental setup. Light from a DPSS red laser
(LSR671ML, λ = 671nm, Lasever, Ningbo, China) illuminated the sample and the speckle
images were recorded by a 12bit camera (MER-125-30UM, Daheng Imaging, China, 1292×964
pixels, 3.75μm×3.75μm) with an exposure time set between 20 and 40 msec. The DMD
(DLC9500P24 0.95VIS) acted as a reflection mirror here. In system calibration, light
reflectance from a Lambertian reflection standard was recorded with the exposure time set at
40 msec and a total of 150 images captured at a frame rate of 15 fps. The system characteristics
under different levels of light illumination was obtained by varying the intensity attenuator and
the reflection ratio of DMD. In flow velocity measurement experiments, Intralipid-2%
suspension (scattering coefficient = 1.7
) inside a glass tube (inner diameter 1mm, outer
diameter 2mm) is used to simulate blood flow. The flow rate in the glass tube is set by adjusting
the driving speed of the fuel injection pump, covering the whole range from 0 to 40 mm/s in
this study. A stack of 250 raw speckle images of the dynamic sample was acquired with an
exposure time set at 40 msec and a frame rate of 15 fps for each flow speed.
Fig. 2 Schematic of the experimental setup.
3.2 Results of system calibration
2fK2fK1mmdn
Fig. 3 Dark current of the camera.
Figure 3 shows the dark current of the camera with a distribution centered at 0. This means that
the dark current of the camera has been pre-subtracted and
should be set to 0.
A set of 150 reflectance images from the reflectance standard were then recorded. The
coherence factor
of the system was then computed with Eq. (19) for different step size
(see Fig. 4). The coherence factor
reduces slightly with
owing to the inevitable system
instability. The proper system coherence factor is obtained by extrapolating to
(=0.540/67, determined by the exposure time 40 msec and the acquisition time 67 msec),
yielding
=0.3144.
Fig. 4 The coherence factor
reduces with the step size
.
By varying the intensity attenuator and the reflection ratio of DMD, multiple sets of 150
reflectance images from the reflectance standard were recorded. The noise variance was
computed with Eq. (20) or (21) in the temporal or spatial domain. The noise variance computed
with either approach agrees with each other. The noise variance in the spatial domain, however,
has lower standard error and is preferred (see Fig. 5). The computed noise variance increases
dn=0.3 with the step size
extrapolating to
and the light intensity. The proper noise variance is obtained by
as well as the determination of above.
Fig. 5 Noise variance increases with the step size
and (b) ∆=0.3 and 1.
and the light intensity for (a) two particular intensities,
Fig. 6 Noise characteristics of the imaging system: (a) Noise variance and (b) noise contrast
versus
light intensity extrapolated at
. The shadow represents the error range.
Figure 6 shows the noise characteristics of the imaging system by extrapolating to
.
The shadow represents the error range given by the standard deviation computed from five
separate sets of measurements. Table 1 displays the noise parameters of
by fitting with
Eq. (10).
Table 1 Fitted noise parameters of
0.89±0.03
99±2
In previous LSCI experiments, an analog-to-digital conversion factor [9]
=0.32noise=0.3=0.3()ciVarn()ciVarn()dVarn4(1.470.05)10 (28)
was often used. The value of this factor calculated from the measurement is observed to
decrease with the light intensity (see Fig. 7). The assumption of a constant
is thus not correct,
attributed to the nonzero mainly caused by the light source fluctuations.
Fig. 7 The analog-to-digital factor
decreases with the light intensity.
3.3 Results of dynamic sample measurements
3.3.1 Importance of static scattering removal and noise correction
A stack of 250 images were taken for the flow phantom at each flow speed ranging between 0
and 40 mm/s. The dynamic fraction was computed with Eq. (22). Fig. 8(a) shows the 2D
distribution of
with an average value of 0.871 over a region of interest (ROI) when flow
speed is 0 (Brownian motion alone). The extracted value of
stays unchanged when the flow
speed increases (see Fig. 8(b)). The non-uniformity of the dynamic fraction is caused by the
imperfect glass tube.
(,)'()()cicidIxyVarnVarn'' Fig. 8 The dynamic fraction
over a ROI. (a) 2D distribution when the flow speed is 0. (b) The average
dynamic fraction versus the flow speed.
The noise contrast
was computed using Eq. (23) or Eq. (24) in the temporal or spatial
domain, respectively. Fig. 9 shows the temporal and spatial
when the flow speed is 0. The
temporal
has much higher spatial resolution than the spatial one. The average temporal
is 0.00165±0.00001, agreeing with
is 0.00169±0.00015 and the average spatial
each other.
Fig. 9
computed in the (a) temporal and (b) spatial domain.
Figure 10 shows the temporal speckle contrast
computed from the data set (original,
after noise correction, after both noise correction and static scattering removal yielding
).
2noise2noise2noise2noise2noise2noise2K2fK Fig. 10 Correction of the temporal speckle contrast
.
after both noise correction and static scattering
removal yields
.
The flow speed is directly related to the decorrelation time. The inverse decorrelation time
increases with the flow speed and may serve as its proxy. The inverse decorrelation time
extracted from fitting Eq. (8) is shown in Fig. 11. The sensitivity of uncorrected
to the
flow speed is very poor and loses linearity around 5 mm/s whereas the corrected
shows
excellent linearity over the whole range up to 40 mm/s. The corrected one with both noise and
static scattering removal also exhibits the least standard deviation.
Fig. 11 The inverse decorrelation time from (a) uncorrected and (b) corrected temporal speckle contrast.
3.3.2 Effects of different static scattering
The efficacy of the static scattering removal is then investigated. One part of the glass tube was
coated with a scattering layer (dried colloidal suspension) and the same set of the measurements
were performed. The region A (=0.83, average light intensity = 670) in the green rectangle is
covered by the static scattering layer whereas the region B (=0.88, average light intensity =
810) in the red rectangle is directly exposed (see Fig. 12). Both regions should have identical
flow speed.
Fig. 12 ROI A and B (covered with an extra static scattering layer) are imaged.
Figure 13 compares the temporal speckle contrast and the inverse decorrelation time for
ROI A and B. The inverse decorrelation time from the uncorrected speckle contrast differs
2K2K2fK1/c1/c1/c significantly between A and B (see Fig. 13 (a,d)). After noise correction, the agreement
between A and B significantly improves although the discrepancy between their recovered
is appreciable (see Fig. 13 (b,e)). With a further static scattering removal, the gap between
for the two regions in (e) almost disappeared (see Fig. 13 (c,f)). The degrade in the
performance for faster flow speeds is caused by the loss of SNR at higher speeds. The above
results show that different static scattering can be successfully removed to obtain the true flow
velocities.
Fig. 13 (a-c) The temporal speckle contrast (uncorrected, after noise correction, after both noise correction
and static speckle removal) for ROI A and ROI B; (d-f) the recovered corresponding inverse decorrelation
time.
To further show the agreement of the flow speed in ROI A and B, the error in the recovered
can be directly estimated using the uncertainty in the noise variance. The noise contrast
depends on light intensity alone when the imaging system has been specified. At higher speeds,
the uncertainty in the noise variance starts to dominate as the flow contrast steadily reduces.
Fig. 14 shows the flow speeds in regions A and B indeed agree with each other within the
system uncertainty given in Fig. 6 and Table 1.
Fig. 14 The flow speed at region A and B agrees with each other within system uncerntainty.
3.3.3 Temporal speckles vs spatial speckles
1/c1/c1/c The speckle contrast analysis can not only be performed within the temporal domain presented
in Sec. 3.3.1 and 3.3.2 but also in the spatial domain. The two different approaches have their
own merits.
Fig. 15 (a-c) The temporal and spatial speckle contrast (uncorrected, after noise correction, after both noise
correction and static speckle removal) for ROI A; (d-f) the recovered corresponding inverse decorrelation
time.
Figure 15 compares the temporal and spatial speckle contrast and the inverse decorrelation
time for ROI A. The uncorrected temporal and spatial speckle contrasts and inverse
decorrelation times differ significantly caused by the non-ergodic static scattering
(see Fig.
14 (a,d)). After noise correction, the temporal speckle contrast and inverse decorrelation time
performs much better than the spatial counterparts which still retain
(see Fig. 14 (b,e)).
With a further static scattering removal,
and
in the temporal and spatial domains tend
to agree with each other (see Fig. 13 (c,f)). However, a careful examination of the recovered
in (f) reveals their difference. The inverse decorrelation time recovered in the spatial
domain shows much less variation yet the linearity range of the temporally recovered
expands to much higher speeds. The latter behavior can be attributed to the difficulty of static
scattering removal inside the spatial domain where a subtraction between the measured contrast
and
is required. At higher speeds, the error in
dominates and the
flow speckle contrast
computed in the spatial domain fails to obtain the true flow speed.
3.3.4 2D flow profile
In addition to the above LSCI analysis of the overall behavior of the flow contrast and inverse
decorrelation time versus flow speed, the result of LSCI imaging of a specific 2D region is
shown in Fig. 16 (v=3 mm/s). The flow speed is observed to increase closer to the center of the
tube. The flow speed cross sectional profile
marked in Fig. 16 (a) fits well by a Newtonian
flow profile [22].
2ne2ne2fK1/c1/c1/c2K22(1)nepN2ne2fK1/c
Fig. 16 (a) ROI selected for analysis. (b) Flow index
for the ROI. (c) The flow speed cross sectional
profile
marked in (a) fitted to a Newtonian flow profile (
).
4. Discussions
The measured temporal and spatial speckle contrasts for flow imaging are affected by both the
presence of static scattering and measurement noise. Their values always differ from each other
except for a pure dynamic medium without static scattering. The spatial speckle contrast
contains one extra term
due to the non-ergodic static scattering than the temporal
counterpart. Nevertheless, a common true flow speckle contrast
can be defined in both the
temporal and the spatial domains. A complete procedure covering the LSCI system calibration,
static scattering removal, and measurement noise estimation and correction to obtain the true
flow speckle contrast
and the flow speed from single-exposure LSCI measurements has
been detailed here. The recovered inverse decorrelation time
from
exhibits excellent
linearity against the flow speed over the full span from 0 to 40 mm/s. The true
is obtained
regardless of the overlying static scattering layers and the type of measured contrasts (temporal
or spatial speckle contrasts).
Comparing speckle contrasts in the temporal and the spatial domain, the latter contains one
. This fact explains the apparent increase of the spatial speckle contrast
additional term of
with the static scattering [23]. The inverse decorrelation time recovered in the spatial domain
shows much less variation yet with a much shorter linearity range than that obtained in the
in the spatial domain is
temporal domain. The rapid deterioration of the performance of
caused by the difficulty of static scattering removal which requires a subtraction between the
measured spatial speckle contrast
dominates and
. At higher speeds, the error in
and
the flow speckle contrast
can no longer be accurately estimated. This observation is
fundamental in selecting the appropriate statistics in analysing the LSCI measurements. A
general guideline is that the spatial speckle contrast should be avoided when
is not larger
than the error in
.
A Lorentz velocity distribution model for the moving particles is assumed to relate
to
the flow speed in our study. Different velocity distribution models may be assumed [24].
However, in the typical situation of much longer exposure time compared to the decorrelation
time (as in our study), the relation Eq. (8) stills holds other than a trivial pre-factor. The
1/c1/c220.5r2ne2fK2fK1/c2fK1/c2ne2fK2K2ne2ne2fK2fK2ne2fK1mmNewtonian flow (a)(b)(c) 0.5mm underlying flow speed estimated by
compares favorably with the input value where
is the vacuum wavelength of the incident light and n is the refractive index of the medium.
For example, it yields
mm/s
when the input flow speed is 10 mm/s (see Fig. 15).
mm/s when the input flow speed is 5 mm/s and
Finally, although our study is on the single-exposure laser speckle imaging, the same
analysis methodology can be carried over to the multiple-exposure LSCI. The latter gains the
advantage over the former when probing the flow velocity distribution. However, the much
simpler and faster single-exposure LSCI performs as well as multiple-exposure LSCI when the
detailed velocity distribution is not of interest as long as the exposure time is much larger than
the decorrelation time and the outlined analysis procedure is followed.
The code for the proposed analysis procedure has been provided in GitHub [25].
5. Conclusion
In summary, a systematic and robust laser speckle flow imaging method and procedure has
been presented, covering the LSCI system calibration, static scattering removal, and
measurement noise estimation and correction to obtain a true flow speckle contrast and the flow
speed from single-exposure LSCI measurements. The power of our LSCI analysis recipe has
been demonstrated by imaging phantom flow at varying speeds, showing that (1) our recipe
greatly enhances the linear sensitivity of the flow index and the linearity covers the full span of
flow speeds from 0 to 40 mm/s; and (2) the true flow speed is recovered regardless of the
overlying static scattering layers and the type of speckle statistics (temporal or spatial). The
difference and merits of the temporal and spatial speckle contrasts have been compared and a
guideline for selecting the appropriate statistics for LSCI has been provided. The proposed
LSCI analysis framework paves the way to estimate the true flow speed in the wide array of
laser speckle contrast imaging applications.
Funding
Natural Science Foundation of Zhejiang Province (LZ16H180002); National Natural Science
Foundation of China (81470081); Wenzhou Municipal Science and Technology; Bureau
(ZS2017022); National Science Foundation, USA (1607664).
Disclosures
The authors declare that there are no conflicts of interest related to this article.
References
1.
J. Senarathna, A. Rege, N. Li, and N. V. Thakor, "Laser Speckle Contrast Imaging: theory, instrumentation and
applications," IEEE Reviews in Biomedical Engineering. (IEEE, 2013), 99-110.
2. B. David, D. D. Duncan, H. Evan, S. J. Kirkpatrick, L. Marcus, S. Wiendelt, S. Tomas, and O. B. Thompson,
"Laser speckle contrast imaging: theoretical and practical limitations," J. Biomed. Opt. 18, 066018 (2013).
3. D. A. Boas and A. K. Dunn, "Laser speckle contrast imaging in biomedical optics," J. Biomed. Opt. 15, 011109
(2010).
4. M. Draijer, E. Hondebrink, T. V. Leeuwen, and W. Steenbergen, "Review of laser speckle contrast techniques
for visualizing tissue perfusion," Laser. Med. Sci. 24(4), 639-651 (2009).
5. C. Haiying, L. Qingming, L. Qian, L. Qiang, G. Hui, Z. Shaoqun, and Biology, "Laser speckle imaging of
6.
blood flow in microcirculation," Phys. Med. Biol. 49(7), 1347:1357 (2004).
S. M. Kazmi, S. Balial, and A. K. Dunn, "Optimization of camera exposure durations for multi-exposure
speckle imaging of the microcirculation," Biomed Opt Express 5(7), 2157-2171 (2014).
7. R. Ambrus, R. B. Strandby, L. B. Svendsen, M. P. Achiam, J. F. Steffensen, and M. B. Sondergaard Svendsen,
"Laser Speckle Contrast Imaging for Monitoring Changes in Microvascular Blood Flow," Eur. Surg. Res. 3(56),
87-96 (2016).
8. N. Hecht, M. M. Muller, N. Sandow, A. Pinczolits, P. Vajkoczy, and J. Woitzik, "Infarct prediction by
intraoperative laser speckle imaging in patients with malignant hemispheric stroke," J. Cerebr. Blood. F. Met.
36(6), 1022-1032 (2016).
9. L. Pengcheng, N. Songlin, Z. Li, Z. Shaoqun, and L. Qingming, "Imaging cerebral blood flow through the
intact rat skull with temporal laser speckle imaging," Opt. Lett. 31(12), 1824-1826 (2006).
/2cn2.00.14.20.2 10. Z. Pavel, V. L. Andreas, B. Alfred, W. Bruno, and S. Frank, "Quantitative modeling of laser speckle imaging,"
Opt. Lett. 31(23), 3465-3467 (2006).
11. P. Zakharov, A. C. Völker, M. T. Wyss, F. Haiss, N. Calcinaghi, C. Zunzunegui, A. Buck, F. Scheffold and B.
Weber, "Dynamic laser speckle imaging of cerebral blood flow," Opt. Express. 17(16), 13904 (2009).
12. A. B. Parthasarathy, W. J. Tom, G. Ashwini, Z. Xiaojing, and A. K. Dunn, "Robust flow measurement with
multi-exposure speckle imaging," Optics Express 16(3), 1975-1989 (2008).
13. A. K. Dunn, A. B. Parthasarathy, and S. Kazmi, "Quantitative imaging of ischemic stroke through thinned skull
in mice with Multi Exposure Speckle Imaging," Biomed. Opt. Express 1(1), 246-259 (2010).
14. S. M. S. Kazmi, A. B. Parthasarthy, N. E. Song, T. A. Jones, A. K. Dunn, "Chronic imaging of cortical blood
flow using Multi-Exposure Speckle Imaging," J. Cerebr. Blood. F. Met. 33(6), 798-808 (2013).
15. C. P. Valdes, H. M. Varma, A. K. Kristoffersen, T. Dragojevic, J. P. Culver, and T. Durduran, "Speckle contrast
optical spectroscopy, a non-invasive, diffuse optical method for measuring microvascular blood flow in tissue,"
Biomed. Opt. Express 5(8), 2769-2784 (2014).
16. S. J. D. Yuan, "Sensitivity, noise and quantitative model of Laser Speckle Contrast Imaging," Dissertations &
17.
T. Gradworks (2008).
I. Freund, "Joseph W. Goodman: Speckle Phenomena in Optics: Theory and Applications," J. Stat. Phys.
130(2), 413-414 (2008).
18. R. Bonner and R. Nossal, "Model for laser Doppler measurements of blood flow in tissue," Appl. Opt. 20(12),
2097-2107 (1981).
19. P. A. Lemieux and D. J. Durian, "Investigating non-Gaussian scattering processes by using nth-order intensity
correlation functions," J. Opt. Soc. Am. A. 16(7), 1651-1664 (1999).
20. C. P. Valdes, H. M. Varma, A. K. Kristoffersen, T. Dragojevic, J. P. Culver, and T. Durduran, "Speckle contrast
optical spectroscopy, a non-invasive, diffuse optical method for measuring microvascular blood flow in tissue,"
Biomed Opt Express 5(8), 2769-2784 (2014).
21. D. D. Duncan, S. J. Kirkpatrick, R. K. Wang,, "Statistics of local speckle contrast," J. Opt. Soc. Am. A. 25(1),
9-15 (2008).
22. A. K. Doolittle, "Studies in Newtonian Flow. I. The Dependence of the Viscosity of Liquids on Temperature,"
J. Appl. Phys. 22(8), 1031-1035 (1951).
23. J. C. Ramirez-San-Juan, R. Caitlin, C. O. Beatriz, and C. Bernard, "Spatial versus temporal laser speckle
contrast analyses in the presence of static optical scatterers," J. Biomed. Opt. 19(10), 106009 (2014).
24. D. D. Duncan and S. J. Kirkpatrick, "Can laser speckle flowmetry be made a quantitative tool?," J. Opt. Soc.
Am. A. 25(8), 2088 (2008).
25. Chenge Wang, Xin Jin, M. Xu, Robust laser speckle contrast imaging, https://github.com/wcg123123/Laser-
Speckle-Contrast, 4, 2019.
|
1906.07797 | 1 | 1906 | 2019-06-18T20:36:27 | Acoustically Driven and Modulation Inducible Radiating Elements | [
"physics.app-ph"
] | The low propagation loss of electromagnetic radiation below 1 MHz offers significant opportunities for low power, long range communication systems to meet growing demands for IoT applications. Especially in the very low frequency (VLF: 3-30 kHz) range and lower, propagation through tens of meters of seawater, hundreds of meters of earth, and hundreds of kilometers of air with only 2-3 dB of loss is possible. However, the fundamental reduction in efficiency as the size of electrical antennas decreases below a wavelength (30 m at 1 MHz) has made portable communication systems in the VLF and low frequency (LF: 30-300 kHz) ranges impractical for decades. A paradigm shift from electrical to piezoelectric antennas utilizing strain-driven currents at resonant wavelengths up to five orders of magnitude smaller than electrical antennas offers the promise for orders of magnitude efficiency improvement over the electrical state-of-the-art. This work demonstrates a lead zirconate titanate transmitter >6000 times more efficient than a comparably sized electrical antenna and capable of bit rates up to 60 bit/s using frequency-shift keying. Detailed analysis of design parameters offers a roadmap for significant future improvement in both radiation efficiency and data rate in the new field of acoustically driven antennas. | physics.app-ph | physics | Acoustically Driven and Modulation Inducible Radiating Elements
Ahmed E. Hassanien*, Michael Breen, Ming-Huang Li & Songbin Gong
University of Illinois at Urbana-Champaign
Abstract
The low propagation loss of electromagnetic radiation below 1 MHz offers significant opportunities for low
power, long range communication systems to meet growing demands for IoT applications. Especially in the very
low frequency (VLF: 3-30 kHz) range and lower, propagation through tens of meters of seawater, hundreds of
meters of earth, and hundreds of kilometers of air with only 2-3 dB of loss is possible. However, the fundamental
reduction in efficiency as the size of electrical antennas decreases below a wavelength (30 m at 1 MHz) has made
portable communication systems in the VLF and low frequency (LF: 30-300 kHz) ranges impractical for decades.
A paradigm shift from electrical to piezoelectric antennas utilizing strain-driven currents at resonant wavelengths
up to five orders of magnitude smaller than electrical antennas offers the promise for orders of magnitude
efficiency improvement over the electrical state-of-the-art. This work demonstrates a lead zirconate titanate
transmitter >6000 times more efficient than a comparably sized electrical antenna and capable of bit rates up to
60 bit/s using frequency-shift keying. Detailed analysis of design parameters offers a roadmap for significant
future improvement in both radiation efficiency and data rate in the new field of acoustically driven antennas.
Introduction
Portable wireless devices have become ubiquitous over the last decade, and with the growth of the internet of
things (IoT), demand for small, efficient wireless communication systems continues to accelerate. While the
development of wireless systems has kept pace with demand at higher frequencies, progress toward portable low-
frequency systems has been stagnant for nearly a century. Compact antennas at the very high frequency (VHF:
30-300 MHz) and ultra high frequency (UHF: 300-3000 MHz) are well developed and suited for transmitting data
at high bit rates. However, increased spectral crowding and relatively large propagation loss in those bands make
them unsuitable for widespread arrays of remote, low power sensors in rural areas or long-range communication
elements over rugged terrain desirable for internet of things (IoT) or defense1 applications. Compared to VHF
and UHF signals, radiation at the ultra-low (ULF: 0.3-3 kHz) and very low frequency (VLF: 3-30 kHz) ranges
exhibits relatively low propagation loss, enabling communication underwater up to 20 meters2 and through
hundreds of feet of earth3. Additionally, VLF radiation can propagate as ground waves which reflect back and
forth between the Earth surface and ionosphere with very low atmospheric attenuation of ~2-3 dB/100 km4.
However, while the desirable propagation properties ensure continued demand for portable, long-range VLF
transmitters, use of VLF antennas has been largely restricted to submarines5,6 and large base installations7,8 such
as the VLF transmitter in Cutler, Maine.
Despite decades of investigation, compact antennas in the VLF and low frequency (LF, 30 -- 300 kHz) bands have
remained an unattainable holy-grail considered impractical due to the fundamental tradeoff between antenna
efficiency and electrical size. Efficient electrical antennas require operation near electromagnetic resonance,
typically restricting the physical size to be larger than one-tenth of a wavelength λ/10. Fundamental analysis on
the tradeoff between antenna size and efficiency was first conducted by Wheeler9
and Chu10 in the 1940s and
extensive subsequent work11 -- 13 defined the size limit for an efficient electrically small antenna (ESA). Decreasing
the size of an ESA below that limit results in a diminished radiation resistance, which leads to a low radiation
efficiency as resistive losses begin to dominate14. Furthermore, as the size of electrical antennas becomes much
smaller than λ, the reactive component of the antenna impedance becomes increasingly large. The small radiation
resistance in conjunction with the much larger antenna reactance results in a large impedance mismatch with the
1
driving electronics. Tuning out the large reactance to improve the matching efficiency requires an impedance
matching circuit, but for frequencies below 1 MHz, the large size and lossy nature of the matching circuit have
made ESAs impractical to implement.
Recently, piezoelectric resonant acoustic antennas have been considered to circumvent the inefficiency of ultra-
sub wavelength ESAs required for portable VLF communication. First proposed by Mindlin15, piezoelectric
antennas couple mechanical vibration into electrical radiation. Acoustic waves propagate at velocities 105 times
lower than electromagnetic waves, enabling resonant operation for mechanical antennas at frequencies five orders
of magnitude lower than similar sized electrical counterparts. Resonant impedances of acoustically driven
antennas can be easily matched to driving electronics, removing the need for bulky, inefficient matching circuits
and circumventing the Wheeler-Chu limit for VLF antennas. More recently, additional studies on the radiation
properties of piezoelectric antennas16,17 and early prototypes at both VHF18 and VLF19 have been demonstrated
to show promise as compact antennas with efficiency advantages over ESA.
In this paper, we demonstrate an acoustically driven and modulation inducible radiating element (ADMIRE) using
lead zirconate titanate (PZT) as the piezoelectric material which redefines VLF transmitters by circumventing the
Chu bandwidth limit and demonstrating novel shaping of near and far-field regions using high-permittivity
materials. While the presented antenna efficiency is already more than 6000x that of an equivalently sized ESA,
it is still far from the anticipated limits of piezoelectric antennas. A full analysis of the design space for
piezoelectric antennas is detailed, paving the way for the subsequent development of compact, high-efficiency
piezo-transmitters with the potential for widespread use in low-frequency wireless communication systems.
Theory
Acceleration of charges, including dipole moment flipping, results in far-field electromagnetic (EM) radiation
with field components that are inversely proportional to the distance traveled away from the radiating element20.
Using this concept, any element that contains a time-varying dipole moment, such as the acoustically excited
piezoelectric materials described in this paper, can be considered a radiating element. Piezoelectric materials lack
inversion symmetry within its crystalline structure, resulting in a linear coupling between the electrical and the
mechanical domain parameters via the reversible piezoelectric effect. In particular, the direct piezoelectric effect
is the electrical polarization produced by an applied mechanical stress. For a time-varying stress, radiation is
produced with the time-varying electrical polarization21.
This concept is further explained in Fig. 1a, where a sinusoidal force, with period 𝜏, is exerted on a piezoelectric
material resulting in electric polarization with surface charge density 𝜎𝑞 which can be calculated using the
piezoelectric constitutive equations as follows17:
𝜎𝑞 = 𝑑𝑇 = 𝑑𝐶𝐸𝑆
𝐼 = 𝜎𝑞𝐴𝜔 = 𝑑𝐶𝐸𝑆𝐴𝜔
(1)
(2)
where 𝑑 is the piezoelectric strain constant, 𝑇 is the applied stress, 𝐶𝐸 is the stiffness at constant electric field and
𝑆 is the resulting strain. The effective dipole current is calculated in (2), where 𝐴 is the surface area of the
accumulated charges and 𝜔 is the angular frequency of the applied stress. The generated magnetic field density
in the far-field region due to the dipole current is then formulated as20:
𝐵𝑓𝑎𝑟 =
𝜎𝑞𝐴
4𝜋𝜀𝑜
𝐿𝜔2
𝑐3𝑅
(3)
where 𝐿 is the dipole moment length, 𝜀𝑜 is the permittivity of the free space, and 𝑐 is the speed of light. The
corresponding far-field electric field 𝐸𝑓𝑎𝑟 = 𝑐𝐵𝑓𝑎𝑟.
2
Fig. 1 Comparison between acoustically driven and modulation inducible radiating elements (ADMIREs) and electrically small
antennas (ESAs). a, ADMIRE-conceptual diagram with a sinusoidal time-varying mechanical stress applied, resulting in time-varying
electrical polarization. b, Butterworth Van-Dyke equivalent circuit model of the ADMIRE. c, ESA representation (infinitesimal dipole).
d, ESA equivalent circuit model at low frequencies (< 1 MHz) with large antenna reactance dominating the antenna input impedance
which requires an impractical matching network.
For comparison, both the ADMIRE and a generic electrically small antenna (ESA), are shown in Fig. 1a and 1c,
respectively. Both types of antennas can be modeled in the electrical domain with the equivalent circuit
representations shown in Fig. 1b and Fig. 1d, respectively. The ADMIRE is modeled with the Butterworth Van
Dyke (BVD) model22, where at the mechanical resonance the reactive components cancel out (𝜔𝑟𝐿𝑚 = 1 𝜔𝑟𝐶𝑚
)
. In the model, 𝑅𝑚 represents the mechanical
and the input impedance of the ADMIRE is 𝑅𝑚 + 𝑅𝑟𝑎𝑑 ≪ 1 𝜔𝐶𝑜
losses, 𝑅𝑟𝑎𝑑 is added to represent the radiated power, 𝐶𝑜 is the static capacitance between the input electrical
terminals, 𝜔𝑟 is the resonance frequency, 𝐿𝑚 is the motional inductance representing the mechanical mass effect,
and 𝐶𝑚 is the motional capacitance representing the mechanical stiffness effect. As previously explained, ESAs
at low frequencies (< 1 MHz) have a large reactive element that requires impractical matching compared with the
ADMIRE which is designed to be impedance matched.
⁄
⁄
It can be shown that the ADMIRE radiation efficiency, defined as the radiated power divided by the input power,
is proportional to the piezoelectric material properties and dimensions as follows17:
𝜉𝐴𝐷𝑀𝐼𝑅𝐸 ∝ 𝑑2𝐶𝐸𝑉𝑄𝜔3
(4)
where 𝑄 is the quality factor of the acoustic mode in the material and 𝑉 = 𝐿𝐴 is the volume of the ADMIRE. The
relative radiation efficiency for similarly sized ADMIRE and ESA can be formulated as17:
𝜉𝑟𝑒𝑙 =
𝜉𝐴𝐷𝑀𝐼𝑅𝐸
𝜉𝐸𝑆𝐴
∝
𝑑2𝐶𝐸𝑄𝜔
𝜎𝑐
(5)
3
Equivalent circuit at mechanical resonance Matching NetworkAntenna Reactance abdc + + StressPotentialSurface Charge DensityDipole MomentSurface Area Applied Stress Dipole length
Fig. 2 Comparison of electrical polarization response for different piezoelectric materials. All materials are assumed to have a
charge density of ±1 mC/m2 on the top and bottom surfaces with a thickness (distance between surfaces) of 1 cm. a, The electric field
and the corresponding surface voltage versus the piezoelectric material relative permittivity. The top gray region is the air breakdown
region where the electric field exceeds 3 MV/m. b, The voltage distribution due to the electric polarization and the electric field direction
represented by the black arrows.
where 𝜎𝑐 is the electrical conductivity of the ESA metallic material. From (5), the relative radiation efficiency of
the ADMIRE can be increased by selecting a material with larger stiffness, quality factor, and especially the
piezoelectric strain constant due to its squared behavior. However, the main advantage of mechanical antennas
arises from the mismatch efficiency of ADMIREs compared to ESAs at low frequencies below 1 MHz. The total
antenna efficiency is defined as the ratio of the power radiated relative to the maximum available power from the
source and is the product of the radiation and mismatch efficiencies. While ADMIREs can be designed to have
real resonant impedances that achieve high mismatch efficiency without the need for a matching network at low
frequencies, ESAs are well known to exhibit small radiation resistances and large reactive components which
result in an enormous mismatch efficiency. To improve total efficiency, ESAs require bulky impedance matching
circuits to tune out the reactive component. The relative total antenna efficiency of the ADMIRE, normalized
with respect to an impedance matched ESA can be expressed as:
𝑟𝑒𝑙 =
𝜉𝑡𝑜𝑡
𝐴𝐷𝑀𝐼𝑅𝐸
𝜉𝑡𝑜𝑡
𝐸𝑆𝐴 =
𝜉𝑡𝑜𝑡
𝐴𝐷𝑀𝐼𝑅𝐸
𝑅𝑟𝑎𝑑
𝐸𝑆𝐴
𝑅𝑟𝑎𝑑
(𝑅𝑟𝑎𝑑
𝐸𝑆𝐴 + 𝑅𝑙𝑜𝑠𝑠 + 𝑅𝑚𝑎𝑡𝑐ℎ + 𝑅𝑠)2
(𝑅𝑟𝑎𝑑
𝐴𝐷𝑀𝐼𝑅𝐸 + 𝑅𝑚 + 𝑅𝑠)2
(6)
𝐴𝐷𝑀𝐼𝑅𝐸 and 𝑅𝑟𝑎𝑑
𝐸𝑆𝐴 are the ADMIRE and the ESA radiation resistances respectively, 𝑅𝑙𝑜𝑠𝑠 is the ESA
where 𝑅𝑟𝑎𝑑
conduction/dielectric losses, 𝑅𝑚𝑎𝑡𝑐ℎ is the matching resistance resulting from the finite quality factor of the
matching inductor, and 𝑅𝑠 is the source resistance as shown in Fig. 1. Even with matching networks for the ESAs,
typically consisting of low-frequency inductors with quality factors less than a few hundred, the matched
impedance seen by the source remains in the kilo-ohms range, resulting in total antenna efficiencies more than
6400x greater in favor of ADMIREs over ESAs.
In addition to the material properties essential for efficient radiation, the relative permittivity of the piezoelectric
material bears crucial consideration for reliable antenna operation. As the bound charge densities on the top and
bottom surfaces of the ADMIRE are flipped to induce the dipole current for radiation in (2), an electric field 𝐸 is
produced. This electric field is inversely proportional to the relative permittivity as shown in equation (7):
𝐸 ∝
𝜎𝑞
𝜀𝑟𝜀𝑜
(7)
where 𝜀𝑟 is the relative permittivity of the piezoelectric material. the radiated field strength for an antenna is
determined by the maximum achievable current and its distribution. In the case of ADMIREs, the maximum
current limit is determined by the charge density that results in electric near-fields just below the breakdown limit
4
AlNLiNbO3LiTaO3PZTPMN-PTQuartzPotentialMaxMinAirPiezoelectric MaterialabElectric Field Direction of the surrounding environment. Therefore, the maximum radiated field strength is directly proportional to the
relative permittivity of ADMIREs. Fig. 2a compares a few commonly used piezoelectric materials with different
values of relative permittivity. The same charge density of 1 mC/m2 is assumed on the top and bottom surfaces
while the generated electric field and the corresponding surface potential are calculated for a piezoelectric material
with a thickness of 1 cm. Fig. 2b shows a piezoelectric material at resonance surrounded by air and its
corresponding voltage distribution, where the fringing electric field is represented by the black arrows. For
materials such as Quartz, AlN, LiTaO3, and LiNbO3 with low/moderate relative permittivity, the electric field is
higher than or very close to the air breakdown field (~3MV/m), thus imposing a fundamental limit on the
maximum radiation. Although one conceivable solution to this problem is non-metallic vacuum packaging, it
increases both the antenna volume and cost, making it bulky and less reliable. On the other hand, an ADMIRE
with a high relative permittivity such as PZT or PMN-PT (𝜀𝑟 > 1000) can be used to mitigate this issue. In addition
to enabling greater maximum radiation, better near-field confinement inside high permittivity piezoelectrics
results in the near-field region becomes shortened to a fraction of the distance compared to the near-field of an
equivalent infinitesimal electric dipole20. To facilitate future material selection for optimal antenna performance,
the following figure of merit for ADMIREs is defined:
𝐹𝑜𝑀 = 𝑑2𝐶𝐸𝜀𝑟𝑄
(8)
Orders of magnitude further improvement in radiation for acoustically driven antennas is expected with further
optimization of material choice and design.
In addition to efficient radiation, passband transmission requires a modulation technique to send information.
Simple and common digital modulation schemes can be utilized such as binary amplitude, frequency, and phase-
shift keying (BASK, BFSK, BPSK) to directly modulate the ADMIRE (carrier) amplitude, frequency or phase
with a modulating bit stream23. A mechanical antenna such as the ADMIRE has a settling time that is directly
proportional to its quality factor and limits the BASK (on-off keying) rate since the mechanical system must be
switched on and off corresponding to bit 1 and bit 0, respectively. The same applies to BPSK due to the phase
discontinuity that requires the system to resettle and synchronize with the driving signal every time the phase
changes. This presents a tradeoff between the material quality factor, which is required for efficient antenna
operation, and the maximum achievable data rate, which is required for bandwidth efficiency23. On the other
hand, BFSK can be designed to have a fixed amplitude and continuous phase, sometimes referred to as continuous
phase FSK (CPFSK), or minimum-shift keying (MSK), which mitigates the amplitude settling limitation and
dramatically improves the achievable data rates. An FoM presenting the characteristics of a BFSK modulator can
be expressed as follows:
𝐹𝑜𝑀𝑀𝑜𝑑 = ∆𝑓 × 𝐹𝑆𝐾𝑅𝑎𝑡𝑒
(9)
where ∆𝑓 is the separation between the two frequencies representing the binary message (∆𝑓 = 𝑓2 𝑓1) and
𝐹𝑆𝐾𝑅𝑎𝑡𝑒 is the maximum achievable FSK rate for switching between the two frequencies. For practical systems,
∆𝑓 must be as large as possible to allow for larger separation between the band-pass filters (BPF) in the receiver,
which relaxes the BPF design specifications and reduces the bit error rate (BER), while higher 𝐹𝑆𝐾𝑅𝑎𝑡𝑒 enables
higher bit rates (for BFSK 𝐵𝑖𝑡𝑅𝑎𝑡𝑒 = 2 × 𝐹𝑆𝐾𝑅𝑎𝑡𝑒).
Design
Depending on the design goals, different resonance modes and frequencies can be targeted based on the
piezoelectric material properties, dimensions, vibration direction, and excitation to meet performance metrics. In
this paper, a high FoM ADMIRE antenna is designed to operate at the upper bound of the VLF band. Emphasis
is placed on measuring the ADMIRE far-field radiation in the VLF band, and thus the FoM is constrained by
frequency and geometry considerations and well below the ultimate FoM achievable for the ADMIRE. A disc
resonator is designed with a high quality factor as shown in Fig. 3a and 3b, with 0.5 cm wide silver electrodes
patterned on the sides of the PZT disc. This structure forms an acoustic resonator that is mechanically free with
5
Fig. 3 Proposed design and simulation of the piezoelectric radiating element using PZT. a,b, The 3D view and the side view of the
PZT disc (8 cm diameter, 1 cm thickness), with patterned silver electrodes along the disc circumference (0.5 cm width), driving source
connected to the top and bottom electrodes, and dipole moment direction elaborated. c, Resonance mode showing the stress distribution
at resonance formed by the acoustic standing wave. d, Simulated surface voltage with an applied voltage amplitude of 1 V. e, Simulated
and measured impedances at the input terminals of the PZT disc.
metal electrodes to drive it into resonance via the 𝑑31 coefficient. The lateral vibration of the disc, also known as
contour mode or dilation mode, is excited by applying a time-varying voltage on the metalized edges of the PZT
disc. Upon excitation, the time-varying electric field introduced by the electrodes (configured as a pair of top and
bottom electrodes) excites the piezoelectric disc into vibration via the reverse piezoelectric effect. The excited
acoustic wave is reflected by the PZT disc boundaries, resulting in a standing acoustic wave with its maximum
stress at the disc center. Fig. 3c shows the resonance dilation mode at 33.6 kHz along with the stress distribution.
During vibration, the mechanical stress generates electrical charges via the direct piezoelectric effect. The charges
generated in the metalized electrode areas are neutralized by the electrodes, so the electrodes are designed around
the edge of the disc where stress is lowest, leaving the highest stress, highest charge density center of the disc free
to radiate. The density of the electrical charge is amplified by the quality factor at resonance, leading to a large
time-varying dipole moment (current) that causes EM radiation. Additional geometries can be used to excite
different high coupling piezoelectric materials in optimal resonant modes (such as dilation, thickness extensional
or shear) to maximize generated charge, and thus radiation, due to higher piezoelectric coupling coefficients.
The structure is simulated using finite element modeling (FEM). Fig. 3d shows the generated surface voltage due
to charge accumulation at resonance with an applied voltage amplitude of 1 V. A motional resistance of 63 Ω is
designed to match with typical 50 Ω systems at the 33.6 kHz resonance, as seen in Fig. 3e which shows the
impedance at the input terminals of the PZT disc (both simulated and measured). The motional resistance can be
further tailored for perfect matching with 50 Ω systems by changing the width of the electrodes, since 𝑅𝑟𝑎𝑑 is
negligible for matching consideration (𝑅𝑟𝑎𝑑 𝑅𝑚 <<1). According to the BVD model 𝑅𝑚 can be expressed as:
6
PZTDiscMetal ElectrodesEM Radiation Unmetallized Region (Top Surface ) = DisplacementDipole Moment8 cm1 cm+-+-+-+-StressMaxMinabcdeFEMMeasured 63 Ω63 Ω 16901340 1.6 nF1.8 nF 3.4 %3.7 % Fig. 4 Simulation and measurement of the PZT radiation. a, Measurement setup for detecting the magnetic field radiation of the
ADMIRE. b, Simulation comparing ESA (infinitesimal dipole) and PZT magnetic fields. Due to the high relative permittivity of PZT,
the magnetic field is confined within it, which dramatically reduces the near-field component relative to the equivalent ESA
approximation. The radiated far-field can be measured very close to PZT after passing the crossover point at 1.3 m. c, Measured
magnetic field vs. distance exhibiting clearly the far-field regime.
𝑅𝑚 =
𝜋2
8𝜔𝑟𝐶𝑜𝑘𝑡
2𝑄
(10)
where 𝑘𝑡
2 is the electromechanical coupling coefficient.
Results
To demonstrate the ADMIRE antennas, a prototype is created from a 1 cm thick, 8 cm diameter disc of PZT. A
20 µm thick, 0.5 cm wide silver ring electrode is patterned onto the top and bottom surfaces and driven to excite
the PZT in the dilation mode via the 𝑑31 piezoelectric coefficient. The resonant response is extracted from a direct
impedance measurement and yields the results shown in Fig. 3e. The radiation measurements of the ADMIRE
are complicated by the near-field confinement due to the high permittivity of the PZT. Unlike the far-field
radiation of the ADMIRE which is dependent only on the equivalent current caused by the flipping dipole
moments, the radiated near-fields are confined by the large relative permittivity of PZT within the dielectric. This
means that near-field radiation, characterized by 1/r3 for electric fields and 1/r2 for magnetic fields, is diminished
in both magnitude and distance. Compared to an equivalently sized 33 kHz ESA which radiates in the near-field
regime up to 1 km, the ADMIRE antenna reaches its far-field regime (magnetic) after around 1.3 meters. Due to
the respective distance scaling of 1/r2 vs. 1/r, equivalent magnetic field radiation from the ESA is 100 times larger
at ten meters than the ADMIRE radiating the same power. Therefore, both the PZT disc and the measurement
setup shown in Fig. 4a are designed to minimize RF interference from the leads and connections so that the PZT
radiation is not obscured.
The magnetic field vs. distance is measured in free space to minimize RFI and scattering using the setup shown
in Fig. 4a. As seen in Fig. 4c, the measured magnetic field decreases as 1/r as expected from the simulations in
Fig. 4b, confirming the PZT ADMIRE exhibits far-field radiation very close to the antenna. An input power of
1.2 W is supplied to excite the PZT disc. Radiation is measured using a passive loop antenna and the magnetic
7
50xLoop AntennaLock-in AmplifierFunction GeneratorAmplifierPZT DiscacbDistance (R)20.135-7.90621.632164.1750XCrossover at 1.3 mNear-field Confinement
Fig. 5 Different modulation schemes applied to the ADMIRE. a, Mock-up of the measurement setup. b, Frequency response of the
velocity (both magnitude and phase are measured) at the PZT disc edge. c, (top) 10 Hz bit stream, (middle) vibrometer velocity
measurement of the amplitude-shift keyed (ASK) signal, and (bottom) vibrometer velocity measurement of the frequency-shift keyed
(FSK) signal.
field is extracted from a lock-in amplifier measurement using the measured antenna factor 𝐴𝐹 = 𝐵𝑅𝑀𝑆 𝜇𝑜𝑉𝑅𝑀𝑆
,
where 𝐵𝑅𝑀𝑆 is the root mean square (RMS) magnetic field, 𝑉𝑅𝑀𝑆 is the voltage measured with the lock-in
amplifier, and 𝜇𝑜 is the free space permeability. In order to better distinguish the measured radiation from noise,
an average field reading is collected over two minutes at each distance. Extrapolating the measured data to 1 km
yields a magnetic field of 0.23 fTRMS with an ADMIRE driving power of 1.2 W compared to a simulated magnetic-
field of 0.5 fTRMS. The discrepancy between the simulated and measured field strengths is likely due to imperfect
earth ground effects14, shifts in resonance due to ambient temperature changes, and effects from nearby radiators
and reflectors.
⁄
The PZT disc is directly modulated using a function generator outputting both ASK and FSK signals with the
resonant response of the PZT disc captured using an optical vibrometer as shown in Fig. 5a. In both cases the 10
Hz binary bit stream at the top of Fig. 5c is used. With the ASK signal, as the driving signal is switched on and
off the resonator energy ramps up and down over a duration inversely proportional to the loaded quality factor
(𝑄𝐿≈850) The ramping time limits the fundamental modulation rate for direct BASK to approximately 1/2T,
where the time constant 𝑇 = 3 × (2𝑄𝐿 𝜔⁄ ) ≈ 24.4 ms (corresponding to 95% settling from the peak value) for the
demonstrated measurement. A fundamental design tradeoff must be considered to balance the inversely
proportional data rate with the high 𝑄 desired for the radiation FoM. BFSK modulation is conducted within the
3-dB bandwidth of the PZT resonator corresponding to BFSK frequencies of 𝑓1 = 33.218 kHz and 𝑓2 = 33.248
kHz. The input is a continuous-phase FSK with no discontinuity when switching between the two frequencies.
However, due to the phase difference of the mechanical resonator at the two frequencies, the mechanical
resonance is out of phase with the modulated driving signal when it is switched and ramping of the PZT edge
velocity occurs while energy is transferred from one resonant frequency to another as seen in Fig. 5c(bottom). As
the modulation frequency approaches the limit set by the frequency settling, although the amplitude of resonance
is not diminished, the demodulated output signal is distorted as seen in Fig. 6a and 6b. Multiple approaches can
be implemented to surpass the 𝑄-limited fundamental modulation rate of the resonator by ensuring that the phase
of the resonator and driving signal are in phase during modulation transitions.
8
acFunction GeneratorVelocityPZTExcitationSignal Source AnalyzerDisplaybBandwidth∆ = = …Vibrometer
Fig. 6 Measured FSK modulation of the PZT ADMIRE at different FSK data rates. a,b, 5 Hz, and 20 Hz FSK rate. a,b, (Top)
modulating signal, (middle) measured velocity, and (bottom) demodulated signal using signal source analyzer (SSA). c, Peak-to-peak
frequency difference (∆𝑓𝑑𝑒𝑚𝑜𝑑 = 𝑓2 𝑓1) after demodulation with SSA at different FSK rates. The upper limit for direct modulation
using the 3-dB bandwidth is limited by frequency settling, the maximum direct modulation frequency is approached resulting in a
distorted modulation waveform.
Conclusion
The presented ADMIRE demonstrates the potential for portable VLF transmitters that have been unattainable for
decades. Although significant work is needed to optimize both antenna efficiency and bit rate, this work provides
a framework for future development in the field of acoustically driven antennas.
Methods
Modeling
Piezoelectricity and EM radiation modeling require multidisciplinary understanding and coupling between the
electrical and the mechanical domains. This is achieved by using FEM available from "COMSOL Multiphysics"
that couples these domains in the "piezoelectric devices" toolbox. Such a model can be used to determine the
resonance frequency of the structure using eigenfrequency simulations followed by frequency domain simulations
to find out parameters such as induced stress/strain, the velocity of the edge of the disc, internal polarization,
surface voltages, and admittance. The internal polarization (charge density) can then be used to calculate the
polarization current and the radiated EM field from equations (2) and (3). PZT piezoelectric properties are
supplied by the vendor (see Supplemental Information) and input to the FEM model. The quality factor is
modified so that the motional resistance 𝑅𝑚, calculated using equation (10), matches the measured value. The
simulation time can be dramatically reduced since our designed PZT disc exhibits symmetry around its central
axis so, axisymmetric simulations are utilized. In addition, an air sphere is added to model the surrounding of the
ADMIRE which enables near-field simulations around the ADMIRE to compare air breakdown around different
piezoelectric materials. Simulated admittance is compared to measured admittance in Fig. 3e. Moreover, the
simulated near-field of PZT is compared with the infinitesimal dipole near-field in Fig. 4b.
PZT Fabrication and Characterization
The PZT discs are commercially fabricated by Physik Instrumente (PI) and made from their PIC181 material24.
The commercial discs have an 8 cm diameter and a 1 cm thickness with both top and bottom surfaces fully
metalized with ~20 µm of silver. Patterning of the silver is conducted using an end mill to remove the interior
metal until only the desired 0.5 cm ring along the edge remains. Two wire leads are split from a BNC cable and
9
abc FSK Rate = 15 HzFSK Rate = 45 Hz soldered to the top and bottom metal surfaces to provide electrical excitation, with the lead lengths minimized to
reduce near-field radiation from the current loop.
Characterization of the PZT is conducted by connecting a Tektronix AFG3152C function generator directly to
the electrodes via BNC cable. An Agilent E4445A spectrum analyzer connected in series with the PZT disc is
then used to characterize the impedance of the PZT as a function of frequency, from which the motional resistance,
2) and mechanical quality factor are extracted. The bottom surface of the PZT disc
electromechanical coupling (𝑘𝑡
rests freely on a 2x2 cm insulating cardboard lattice and the top and side surface are unconstrained. Multiple
clamping configurations were considered but yielded negligible changes in mechanical properties. Input power
to the PZT disc is characterized by removing the series spectrum analyzer and adding an Agilent MSO7104B
oscilloscope in parallel with the disc with the power dissipation measured from the voltage drop across the PZT.
Radiation Measurement
Wireless radiation measurements of the generated magnetic field are conducted in an open environment to
minimize scattering and noise. Confinement of the near-field component of the PZT radiation results in the far-
field component dominating beyond 2 m but current loops in the transmitter exhibit near-field dominate radiation
up to 1000 m. In order to minimize near-field radiation from current loops, leads and connections are minimized
and oriented to exhibit radiation orthogonal to the receiving antenna. The resulting total radiation exhibits a near-
to-far-field crossover between 1 and 2 meters. At the operating frequency, only the PZT radiation can exhibit a
1/r roll off at distances less than 1 km, therefore all measured radiation with a 1/r fit is attributed solely to the
PZT.
The transmitting system consists of a Tektronix AFG3152C function generator connected in series to a 50x Trek
model 2100HF amplifier to generate a sufficiently large excitation to measure the far field. The amplifier presents
a resistance of 200 Ω in series with the 63 Ω motional resistance of the PZT disc at resonance which results in a
diminished loaded quality factor where 𝑄𝐿 ≈ 𝑄𝑅𝑚 (𝑅𝑚 + 𝑅𝑠). From (4), the diminished 𝑄𝐿 results in a lower
radiation efficiency for the PZT and a higher power is needed to drive the loaded ADMIRE.
The magnetic field is measured using an AH-Systems SAS-565L 24" shielded passive loop antenna which is
oriented to receive the maximum signal from the PZT far-field component. Incident radiation induces an open
circuit voltage across the antenna terminals proportional to the field strength. The antenna factor of the loop
receiver is calibrated by the manufacturer post-production to be 1.74 Ω-1m-1 at 33 kHz and is used to extract the
measured B-field where 𝐵 = 𝜇0𝑉 ∗ 𝐴𝐹. The open-circuit voltage of the antenna is measured using a Stanford
Research Systems SR865A lock-in amplifier that is frequency locked to the transmitting PZT disc and employs
a 24 dB/octave bandpass filter to attenuate noise around the locked frequency. Measurements were made at 1 m
distance intervals for 2 minutes at a time using a 1 second time constant. The measured B-field strength is
extracted from the average terminal voltage over the 2-minute measurement window, with 1 standard deviation
error bars also provided to account for variance in the measured field strength due to noise. Between field
measurements, the noise floor is measured at 1-minute intervals with the input signal turned off (Supplementary
Fig. 3). Measurements beyond 6 m exhibit a signal-to-noise ratio < 2 and are not shown.
Modulation Measurement
Direct digital modulation of the ADMIRE can be done by altering amplitude, frequency, and phase of the
excitation signal which in turn modulates the mechanical resonance of the ADMIRE and thus radiated signal. In
this paper, we focus on BFSK since it has a continuous phase which lowers the mechanical settling time compared
to both BPSK and BASK. The modulation is evaluated using the measurement setup in Fig. 5a. The setup consists
of a Tektronix AFG3152C function generator that directly excites the ADMIRE with continuous phase BASK or
BFSK signals. A Polytec OFV-5000 laser vibrometer is used to measure the velocity of the PZT edge while
vibrating in the dilation mode. The two BFSK modulation frequencies are chosen to be within the 3-dB bandwidth
as shown in Fig. 5b. Fig. 5c shows the modulated velocity of BASK and BFSK with a 10 Hz modulation rate.
The velocity signal is then input to a ROHDE & SCHWARZ FSUP signal source analyzer with FM demodulation
10
capability to demodulate the signal as shown in Fig. 6a and 6b (bottom figures) for 5 Hz and 20 Hz BFSK rates,
respectively.
References
1.
2.
3.
4.
Roy Olsson, J. Microsystems Technology Office (MTO) Broad Agency Announcement: A MEchanically
Based Antenna (AMEBA) (Document HR001117S0007). (2016).
Singh, H. B. & Pal, R. Submarine communications. Def. Sci. J. (1993).
Yarkan, S., Güzelgöz, S., Arslan, H. & Murphy, R. Underground mine communications: A survey. IEEE
Commun. Surv. Tutorials (2009).
Hargreaves J. K., H. R. D. The High-Latitude Ionosphere and its Effects on Radio Propagation.
(Cambridge University Press, 2007).
5. Wheeler, H. A. Fundamental Limitations of a Small VLF Antenna for Submarines. IRE Trans. Antennas
Propag. (1958).
6. Waheed-uz-Zaman. DESIGN AND CONSTRUCTION OF VERY LOW FREQUENCY ANTENNA. J.
Basic Appl. Sci. (2011).
7.
8.
Ashdown, G. A. Study of a VLF distributed antenna array. in In AGARD, ELF/VLF/LF Radio
Propagation and Systems Aspects 13 p (SEE N93-30727 11-32) (1993).
Simpson, T., Roberts, M. & Berg, E. Developing a broadband circuit model for the Cutler VLF antenna.
IEEE Antennas Propag. Soc. AP-S Int. Symp. (2001).
9. Wheeler, H. A. Fundamental limitations of small antennas. Proc. IRE (1947).
10. Chu, L. J. Physical limitations of omni-directional antennas. J. Appl. Phys. (1948).
11. Collin, R. E. & Rothschild, S. Evaluation of Antenna Q. IEEE Trans. Antennas Propag. (1964).
12. Hansen, R. C. Fundamental Limitations in Antennas. Proc. IEEE (1981).
13. McLean, J. S. A re-examination of the fundamental limits on the radiation q of electrically small
antennas. IEEE Trans. Antennas Propag. (1996).
14. Balanis, C. A. Antenna Theory: Analysis and Design. (John Wiley & Sons, 2005).
15. Mindlin, R. D. Electromagnetic radiation from a vibrating quartz plate. Int. J. Solids Struct. (1973).
16. Lee, P. C. Y., Kim, Y. G. & Prevost, J. H. Electromagnetic radiation from doubly rotated piezoelectric
crystal plates vibrating at thickness frequencies. J. Appl. Phys. (1990).
17. Domann, J. P. & Carman, G. P. Strain powered antennas. J. Appl. Phys. 121, (2017).
18. Nan, T. et al. Acoustically actuated ultra-compact NEMS magnetoelectric antennas. Nat. Commun.
(2017).
19. Kemp, M. A. et al. A high Q piezoelectric resonator as a portable VLF transmitter. Nat. Commun.
(2019).
20. Griffiths, D. J. Introduction to Electrodynamics. (Pearson Education, 2013).
21. Auld, B. A. Acoustic fields and waves in solids. (John Wiley & Sons, 1973).
22. Larson, J. D., Bradley, P. D., Wartenberg, S. & Ruby, R. C. Modified Butterworth-Van Dyke circuit for
11
FBAR resonators and automated measurement system. Proc. IEEE Ultrason. Symp. 1, 863 -- 868 (2000).
23. Haykin, S. Communication Systems. (John Wiley & Sons, 2000).
24.
PI Piezo Technology. Available at: https://www.piceramic.com/en/.
12
|
1805.01976 | 1 | 1805 | 2018-05-05T00:37:19 | Biocompatible Two-dimensional Titanium Nanosheets for Efficient Plasmonic Photothermal Cancer Therapy | [
"physics.app-ph"
] | Photothermal therapy (PTT) has shown significant potential in cancer treatment. However, developing nanomaterial-based photothermal agents with good biocompatibility and high photothermal conversion efficiency (PTCE) remains a key challenge. Titanium has been widely employed as biomedical materials based on their excellent biocompatibility. In this work, the titanium based plasmonic photothermal therapy (Ti-PPTT) is demonstrated. Using the liquid-phase exfoliation (LPE), the metallic Ti can be fabricated into the two-dimensional (2D) nanosheets (NSs), similar to exfoliating other layered 2D materials. The 2D Ti NSs exhibited good biocompatibility, high extinction coefficient of 20.8 Lg-1cm1 and high PTCE of 73.4%, owing to localized surface plasmon resonances (LSPR); which is significantly higher than other photothermal agents, including Au (21%), MoS2 (24.4%), BP (28.4%) and Ti3C2 MXene (30.6%). Consequently, 2D Ti-PPTT exhibited a notable therapeutic effect in a human hepatocellular carcinoma model without side effects. Our study could pave a new avenue for PTT using metal Ti and arouse a wide interest in the potential efficient PTT for other elemental transition metals owing to their LSPR. Furthermore, the efficient photothermal effect and environmental friendliness of Ti NSs make them deserve more research toward other application, such as solar energy harvesting and sea water desalination. | physics.app-ph | physics | DOI: 10.1002/ ((please add manuscript number))
Article type: Full Paper
Biocompatible Two-dimensional Titanium Nanosheets for
Efficient Plasmonic Photothermal Cancer Therapy
Zhongjian Xiea,1, Shiyou Chena,1, Quan Liuc,d, Zhitao Lina, Jinlai Zhaod, Taojian Fana,
Dou Wangc,d, Liping Liub,*, Shiyun Baob, Dianyuan Fana, and Han Zhanga,*
aShenzhen Engineering Laboratory of Phosphorene and Optoelectronics, Collaborative
Innovation Center
for Optoelectronic Science & Technology,
International
Collaborative Laboratory of 2D Materials for Optoelectronics Science and Technology
of Ministry of Education, Shenzhen University, Shenzhen 518060, China
bDepartment of Hepatobiliary and Pancreatic Surgery, Shenzhen People's Hospital,
Second Clinical Medical College of Jinan University, Shenzhen, Guangdong Province,
P. R. China.
cIntegrated Chinese and Western Medicine Postdoctoral Research Station, Jinan
University, Guangzhou 510632, China
dCollege of Materials Science and Engineering, Shenzhen University, Shenzhen
518060, PR China
* To whom correspondence should be addressed.
E-mail: [email protected]
1 Z.X. and S.C. contributed equally to this work.
Abstract
Photothermal therapy (PTT) based on two-dimensional (2D) nanomaterials has shown
significant potential in cancer treatment. However, developing 2D nanomaterial-based
photothermal agents with good biocompatibility and high photothermal conversion
efficiency (PTCE) remains a key challenge. Titanium and its alloys have been widely
employed as biomedical materials based on their biocompatibility. In this work, the
elemental Ti based plasmonic photothermal therapy (Ti-PPTT) is demonstrated. Using
the liquid-phase exfoliation (LPE), it was shown that metallic Ti can be fabricated into
the 2D nanosheets (NSs), similar to exfoliating other layered 2D materials.
Furthermore, the 2D Ti NSs exhibited good biocompatibility, high extinction
coefficient of 20.8 Lg−1cm−1 and high PTCE of 73.4%, owing to localized surface
plasmon resonances (LSPR); which is significantly higher than other photothermal
agents, including Au (21%), MoS2 (24.4%), BP (28.4%) and Ti3C2 MXene (30.6%).
Interestingly, the Ti NSs can be degraded into TiO2 in water-based solution and made
themselves hydrophilic in the meantime. Consequently, 2D Ti-PPTT exhibited a
notable therapeutic effect in a human hepatocellular carcinoma model without side
effects. Our study could pave a new avenue for PTT using metal Ti and arouse a wide
interest in the potential efficient PTT for other elemental transition metals owing to
their LSPR.
Keywords: biocompatibility, titanium nanosheets, photothermal therapy, liquid-phase
exfoliation, plasmon resonance
1. Introduction
Materials used for biomedical applications cover a very broad range and must exhibit
specific properties depending on application. The prerequisite of all biomedical
materials however, is biocompatibility. Titanium (Ti) and its alloys are among the
newest and most attractive metallic biomaterials because of their biocompatibility.[1,2]
In both medical and dental fields, Ti and its alloys have been demonstrated as a
successful class of biomedical materials, which do not give rise to adverse effects
locally or systemically.[3] In medicine, they are frequently used as implant devices to
replace failed hard tissues, including artificial knee joints, artificial hip joints, bone
plates, cardiac valve prostheses and artificial hearts.[4] For dentistry devices, Ti and its
alloys are used for bridges, crowns, overdentures and dental implant prosthesis
components.[5]
Pure metallic Ti is considered to be one of the best biocompatible metallic
materials. The main physical properties of Ti responsible for its biocompatibility are:
high corrosion resistance,[3,4] low electronic conductivity, its thermodynamic state at
physiological pH values, an isoelectric point of the oxide of 5–6, and low ion-formation
tendency in aqueous environments.[2] Moreover, the inert oxide layer covered surface is
only slightly negatively charged at physiological pH. The dielectric constant of Ti, is
comparable to that of water resulting in the coulomb interaction of the charged species
being similar to that in water.
Photothermal therapy (PTT) based on the stimulus of near-infrared (NIR) laser
irradiation is a noninvasive therapeutic modality compared with traditional therapies
for cancer treatment.[6] In the past decade, many photothermal agents, such as noble
metal nanomaterials (NMs),[7] carbon-based NMs,[8] copper sulfide,[9] rare earth
compounds,[10] and many organic nanoparticles and polymers,[11] have been widely
explored. Recently, 2D NMs, including palladium (Pd),[12] MoS2,[13–16] WS2,[17,18] and
black phosphorus (BP),[19] have attracted attention as photothermal agents due to their
successful photothermal performances. An efficient photothermal agent should not
only possess an enhanced absorption in the NIR region, but also a high photothermal
conversion efficiency (PTCE). Importantly, the biocompatibility prerequisite for
photothermal agents should also be readily satisfied.
Plasmonic NMs with localized surface plasmon resonances (LSPR) in the NIR
region have been extensively explored as photo-sensitizers for PTT over the past
decade.[20–22] Compared with conventional photo-sensitizers, plasmonic nanoparticles
not only possess stronger NIR light absorption, owing to LSPR, but also exhibit higher
photostability.[22] Plasmonic NMs are composed of various materials, such as metals
(Au, Ag), and semiconductors. In view of the high biocompatibility of Ti alloys, they
have not only been widely investigated for biocompatible implants, but also for
emerging PTT-based applications. Up to now, several titanium compounds, such as
titanium carbide,[23–26] titanium oxide,[27–29] titanium nitride,[20,30,31] and TixTa1-xSyOz,[32]
have been investigated as photothermal agents for cancer therapy. They not only
possess excellent biocompatibility, but also exhibit high photothermal performance
owing to their plasmon resonance.
Titanium carbide (Ti3C2) is an example of an MXene, which is a new class of 2D
early transition-metal carbides. Ti3C2 MXene exhibited high antibacterial efficiency
with growth inhibition of 97.7%.[33] The TiO2-Ti3C2 nanocomposite has also been
shown to be a promising biosensor for the detection of H2O2 with high sensitivity.[34]
Ti3C2 based field-effect transistors, with high biocompatibility, were used to detect
dopamine, and to monitor spiking activity in hippocampal neurons.[35] Ti3C2 quantum
dots (QDs) have been shown to have great potential for multicolor cellular imaging,
and in the optical field in general.[36] These investigations revealed that Ti3C2 has great
potential in environmental and biomedical applications
Notably, Ti3C2 MXene was reported as an effective 2D light-to-heat conversion
material with high photothermal efficiency under sunlight irradiation.[23] Moreover, a
novel photothermal agent based on ultrathin Ti3C2 MXene showed strong optical
absorption in the NIR region.[24] Ti3C2 MXene also demonstrated highly effective
photothermal ablation of tumors.[25] In spite of the high PTT performance of Ti3C2
MXene, its fabrication is a hazardous and time-consuming process, using HF
pretreatment. Fortunately, a new and simple fluorine-free fabrication method has been
proposed and photoacoustic (PA) imaging and notable tumor ablation have been
demonstrated simultaneously.[26] The strong absorption and high conversion efficiency
of Ti3C2 MXene originate from the LSPR effect of its semimetal character. All of the
highlighted reports support the potential of Ti3C2 based MXene application in
biomedicine, especially for photothermal cancer treatment.
In the LSPR enhanced photothermal effect, heat generation is particularly high in
the case of metal nanoparticles, given that LSPR is a collective motion of numerous
electrons.[37] In comparison, heat generation is much weaker for semiconductor
nanoparticles, since heat dissipation occurs through the creation of single mobile
electrons and holes. Consequently, metallic NMs with plasmon resonance can be
developed as effective and promising photothermal agents.
In this work, based on the widely used biocompatible material Ti, we report a
photothermal agent of 2D metal Ti nanosheets (NSs), which exhibits excellent
biocompatibility and highly efficient photothermal performance both in vitro and in
vivo (Figure 1). It is interesting to find that bulk Ti can be processed to give the 2D
form (Ti NSs) using liquid-phase exfoliation (LPE), as has been demonstrated for
other layered 2D materials such as black phosphorus. The Ti NSs showed high
photothermal performance, evaluated by both extinction coefficient and PTCE.
Surface modification with PEG also led to colloidal stability in water, saline and
physiological conditions. In vitro and in vivo experiments were conducted to
systematically investigate the toxicity of the Ti NSs, and a hepatocellular carcinoma
model was established to evaluate the potential of Ti NSs for photothermal tumor
therapy.
2. Results and Discussion
Fabrication
Non-layered 2D materials, the other class of 2D materials, have been of great interest
in recent years because of their useful properties.[38,39] Distinct from 2D layered
materials, non-layered materials have chemical covalent bonds in three dimensions.
After exfoliation, the surfaces of non-layered 2D materials present dangling bonds,
which render their surfaces highly chemically active and enhance their catalytic
capability, sensing, and carrier transfer. These features are not present in the case of 2D
layered materials. Given these advantageous properties, many well-established
techniques have been developed to synthesize non-layered 2D materials, such as
template directed synthesis (TDS),[40,41] the lamellar hybrid intermediate method,[42,43]
and van der Waals epitaxial (vdWE) growth.[44,45]
LPE can produce 2D structures through the choice of an appropriate solvent, with
the aid of sonication. The liquid phase can be considered the most efficient and
versatile as it allows for preparation of many materials,[46] and is particularly suitable
for biomedical purposes.[26] LPE can also be used to produce specific surface
functionalization, which is particularly advantageous for stabilization of the 2D
nanosheets in solution.[46] However, using LPE to exfoliate non-layered 2D materials is
rarely investigated.
2D layered materials can be exfoliated through LPE because of their clear
anisotropic bonding nature in three dimensions; strong covalent bonding interactions
intralayer (the in-plane direction) and weak physical van der Waals' forces between
layers (the out-of-plane direction). While, for non-layered 2D materials, the chemical
covalent bonds in three dimensions still show some anisotropic bonding character.
Based on this anisotropy, they can also be exfoliated into a 2D form using the LPE
method. For instance, the large-area freestanding single layers of non-layered ZnSe
with four-atom thickness were exfoliated using LPE.[43] And ultrathin CoSe2
nanosheets exfoliated using LPE can be used to efficiently catalyze the oxygen
evolution reaction.[47] In our previous work, the non-layered VI A elements selenium
(Se) and tellurium (Te) were exfoliated into 2D form using LPE (not published).
It is clear that Ti metal is a non-layered material. It has a hcp crystal structure,
which presents mechanical anisotropy. Therefore, it can theoretically be exfoliated
using LPE, and interestingly, our attempts have found that typical 2D Ti NSs can be
obtained (Figure 2) using the typical LPE method (Figure 1). As in exfoliation of
other 2D layered materials, isopropyl alcohol (IPA) was chosen as the solvent for this
exfoliation. Compared with other liquid choices such as N-methyl-2-pyrrolidone
(NMP),[48] this solvent can be easily evaporated to ensure the surface of the obtained
Ti NSs is free of impurities for subsequent biological experiments.
Characterization
To obtain 2D Ti NSs with the desired dimensions, dispersions of Ti NSs subjected to
varying degrees of centrifugal force were characterized using transmission electron
microscopy (TEM) and atomic force microscopy (AFM). Figure 2a shows the
dimensions of the Ti NSs obtained from the supernatant of a dispersion subjected to a
centrifugal force of 2000 g. The TEM image shows a typical Ti NS lateral size of less
than 50 nm. Additionally, some Ti QDs can also be observed. The AFM image (Figure
2b) shows that the thickness of the Ti NSs falls within the range of 1–5 nm. Clear lattice
fringes with inter-atomic d-spacing of 0.21 nm, 0.24 nm and 0.26 nm can be observed
(Figure 2c). The selected-area electron diffraction (SAED) pattern suggests the typical
crystalline features of α-Ti.[49] Fast Fourier Transformation (FFT) of the HRTEM
image shows the expected crystallographic lattice reflections of the Ti NSs.
The chemical composition and crystal phase of bulk Ti and Ti NSs were
characterized using X-ray photoelectron spectroscopy (XPS) and X-ray diffraction
(XRD). In Figure 2d, the two XPS peaks at 458.58 eV and 454.33 eV are ascribed to
the Ti 2p3/2 and Ti 2p1/2 orbitals, respectively, for both bulk Ti and Ti NSs.[50,51] The
XPS peak at 464.33 eV indicates the presence of TiO2.[52] The XRD patterns for both
the bulk Ti and exfoliated Ti NSs exhibited typical Ti crystalline diffraction (Figure
2e).[51,53] In Figure 2f, the bulk Ti and Ti NSs exhibited similar Raman peaks. A sharp
peak at 140.5 cm−1 was observed for bulk Ti, which is assigned to the zone-center E2g
mode, which can be directly assigned to hcp titanium.[54] For exfoliated Ti NSs, a
peak at 143.8 cm−1 indicates a Raman shift for the ultrathin Ti NSs. The peaks at ~200
cm-1 for both bulk Ti and exfoliated Ti NSs arise as a result of TiO2.[55]
All characterizations confirm the dimension, composition and crystal features of
the exfoliated Ti NSs. It can therefore be concluded that the LPE method can be used
to exfoliate non-layered metal materials with anisotropic crystal structures in three
dimensions; and is not limited to layered materials showing more significant contrast
between dimensions, for example, between covalent bonding and van der Waals'
interactions. This widely expands the scope of the LPE method and means more 2D
materials with unknown properties can be explored. The exfoliated Ti NSs with
relatively small size (<50 nm lateral size and ~3 nm in thickness) can be used in PTT.
The stability of the Ti NSs in physiological medium was enhanced by surface
coating with polyethylene glycol (PEG). The success of this coating was validated by
Fourier transform infrared spectroscopy (FTIR) and scanning transmission electron
microscopy (STEM) with energy dispersive X-ray spectroscopy (EDS) mapping
(Figure 3). For the naked Ti NSs, the intense band at ~3400 cm−1 was reported to be
resulted from the OH stretching. The composition of TiO2 in the surface of Ti NSs has
been confirmed in the former XPS and Raman measurements (Figure 2). Thus, the
OH group was produced by TiO2 and made the surface of Ti NSs hydrophilic.[56,57]
Compared with the spectrum of naked Ti NSs, two additional FTIR peaks emerged at
~1100 cm−1 and ~2900 cm−1 for the PEGylated Ti NSs. Comparison with the pure
PEG spectrum (Figure 3a) shows that these two peaks can be assigned PEG. The
absorption band at ~2900 cm−1 is attributed to the C-H vibration, and that at ~1100
cm−1 is assigned to C-O stretching in the PEG unit, indicating the successful
PEGylation of Ti NSs. The thickness of PEGylated Ti NSs from the AFM image is
~10 nm (Figure 3b), which is slightly thicker compared with the naked Ti NSs
(Figure 3b). Dynamic Light Scattering (DLS) measurements showed that the size
distribution of PEGylated Ti NSs shifted to larger sizes compared with naked Ti NSs
(Figure 3c). Moreover, the STEM mapping shows the co-localization of four different
elements: Ti, C, O and N. The C, O and N contributions come from the surface coating
of PEG.
Photothermal performance
Extinction coefficient defines how strongly materials absorb light. A high extinction
coefficient is a prerequisite for effective photothermal agents. To characterize this
value for Ti NSs, their optical absorption at different concentrations was measured.
The photograph of Ti NSs dispersed in water at concentrations of 10, 25, 50 and 100
ppm is shown in Figure 4a. The concentration (C) was determined using inductively
coupled plasma atomic emission spectroscopy (ICP-AES). The 100 ppm dispersion
was completely opaque due to the strong absorption of Ti NSs. The optical absorption
spectra of Ti NSs at different concentrations are shown in Figure 4b. They show a
broad and strong absorption band spanning from the UV to NIR regions, similar to
those of other non-metallic, layered 2D NMs, such as GO[8], MoS2,[13]
WS2
[58] and
BP.[19] Strong absorption in the NIR region is necessary to take advantage of the NIR
transparent window (750–1000 nm) of biological tissue for PTT.[59] The normalized
absorbance over the light path length of the measurement cuvette (A/L, where A is
absorbance and L is length of the light path) at 808 nm for different concentrations, was
determined (Figure 4d). The extinction coefficient (k) at 808 nm was then calculated
to be 20.8 Lg−1cm−1, according to the Lambert-Beer law (A/L=kC). This is 5.3 times
greater than that of AuNRs (3.9 Lg−1cm−1).[19] Compared with previously reported 2D
NMs, the extinction coefficient of the Ti NSs is higher than that of GO NSs (3.6
Lg−1cm−1),[8] and even higher than that of popular photothermal agent BPQDs (14.8
Lg−1cm−1).[19] Although it is slightly lower than the extinction coefficient of Ti3C2
(25.2 Lg−1cm−1),[25] it still shows significant potential in photothermal conversion
efficiency (PTCE), another important property for evaluating photothermal agents.
Different concentrations of Ti NSs in water were exposed to an 808 nm NIR laser
with power density of 1.0 W.cm−2, which obeys the maximum permissible exposure
(MPE) for skin of 1 W.cm−2 (American National Standard for Safe Use of Lasers,
ANSI Z136.1−2007).[60,61] The temperature of the Ti NSs dispersion was measured as
irradiation time increased (Figure 4c). The power density of the laser was calibrated
before irradiation. At a low concentration of 50 ppm, the temperature increased from
25 °C to 58.5 °C after 600 s of irradiation. The PTCE (η) is the key parameter for
evaluating photothermal performance. PTCE of the Ti NSs was determined to be up to
73.4 % (Figure 4e) using a standard method,[62] indicating that the Ti NSs can
efficiently convert NIR light into heat. The determined value was significantly higher
than other photothermal agents, such as Au nanoparticles (21 %),[63,64] MoS2
(24.4 %),[14] BPQDs (28.4 %),[19] recently reported Ti3C2 MXene (30.6%)[25] and
antimonene quantum dots (AMQDs, 45.5 %).[65]
In addition to the extinction coefficient and PTCE, the photothermal stability is
an important property for PTT. Figure 5a shows six photothermal cycles at
concentrations of 25 ppm and 50 ppm. In one photothermal cycle, the sample was
irradiated for 10 min, followed by a further 10 min period when the laser was turned off.
It was found that the temperature initially increased to almost saturation level and then
decreased to room temperature by natural cooling (Figure 5a). For the 25 ppm
concentration, the highest temperature throughout the six cycles was approximately
consistent, illustrating that the Ti NSs did not appreciably deteriorate during the 2 h
photothermal process. While for the 50 ppm sample, it was observed with interest that
there was a rising trend in the highest temperature as irradiation cycle number
increased, shown by the dotted line. This was due to the increasing concentration of the
Ti NSs following evaporation of the water dispersant, as a result of the strong
photothermal heating. This was supported by the increased absorbance after 2h of
irradiation (Figure 5b). However, it was unexpectedly found that there was a slight
decrease in the absorbance after 2 h storage in water, indicating the degradability of Ti
NSs in water. The degradability of Ti NSs was further demonstrated after 30 days. It
was found that both the absorbance and photothermal temperature change decreased
significantly after 30 days (Figure 5a, b). Quantitatively, absorbance measurements
showed that 77% of the Ti NSs degraded, however the photothermal temperature only
degraded by 47%. The degradation product was TiO2, as established by the XPS
characterization in Figure 2. TiO2 may have a significant impact on the decrease in
absorbance, but led to only a minor decrease of photothermal temperature because it
can also contribute to the photothermal temperature rise.[66–68] The degradability of Ti
NSs is another advantage of Ti NSs for biomedical application.
Consequently,
the high extinction coefficient
(20.8 L.g−1.cm−1), high
photothermal conversion efficiency (73.4%), good photostability and degradability of
Ti NSs make Ti NSs a unique and high efficiency photothermal agent. Their excellent
biocompatibility and photothermal killing performance in the following studies
further support these assertions.
Toxicity assays
Ti metal is well known for its excellent biocompatibility, however the cytotoxicity of
the newly fabricated Ti NSs had to be ascertained to support their further clinical
application. The potential cytotoxicity of both the naked and PEG encapsulated Ti NSs
was evaluated. PEG was chosen in light of its biocompatibility and has been approved
for medical use by the FDA. Tumor microenvironments consist of various kinds of
cells, including tumor infiltrating immune cells. To model
the
true
tumor
microenvironment, both normal and cancer cells were incubated with the Ti NSs,
including SMMC-7721 (hepatocellular carcinoma), B16 (melanoma) and J774A.1
(macrophage). As shown in Figure 6a,b, both naked and PEGylated Ti NS-treated
cells exhibited trace cytotoxicity in CCK-8 assays post Ti NSs treatment; even at high
concentration (100 ppm) which contrasts with the concentration of 30 ppm required
for the total cell killing effect in the following study as shown in Figure 6.
The in vivo toxicity of both Ti NSs and PEGylated Ti NSs was also investigated.
Given the proof of concept objective of this study, we used subcutaneous injection of Ti
NSs for the toxicity assays study. Three groups of mice were used, which were
injected with naked Ti NSs, PEGylated Ti NSs or saline. Figure 6c shows that the
body weights of the mice dosed with Ti NSs or PEGylated Ti NSs showed no
difference to those injected with saline, which suggests that the Ti NSs did not
intrinsically affect the overall condition of the animals. In addition, H&E staining of the
major organs including heart, liver, spleen, lung and kidney, showed no damage
resulting from injecting the Ti NSs analogues (Figure 6d).
Consequently, our results demonstrate that the fabricated Ti NSs were non-toxic
both in vitro and in vivo, and were therefore biocompatible, supporting further
biomedical development.
In vitro photothermal experiments
Based on the high efficiency photothermal performance and biocompatibility of Ti
NSs, high tumor cell killing efficiency of Ti NSs was anticipated. As shown in Figure
7a, b, both Ti NSs and PEGylated Ti NSs showed a clear photothermal killing effect as
concentration increased. 75% of SMMC-7721 and B16 cells were killed at 20 ppm and
almost all of the cells were killed at a low concentration of 30 ppm. J774A.1 cells were
particularly susceptible, with only ~25% of the cells surviving after the photothermal
killing process, even at 10 ppm. This could be because J774A.1 cells can take up larger
amounts of Ti NSs than cancer cells due to phagocytosis (data not shown), resulting in
higher photothermal efficiency.
Additionally, we investigated the laser irradiation time required to achieve 100%
cell killing efficacy at 50 ppm. It was found that a short time of 2 min of laser
irradiation killed all of the cells. We attribute the efficient cell killing, to the sharp
increase in temperature, passing the fatal point, induced by NIR irradiation (Figure 7c).
Graphic illustrations of the photothermal killing effect for different concentrations of
Ti NSs are presented in Figure 7d. We demonstrated the high photothermal killing
effect of Ti NSs, and their excellent biocompatibility, making them promising for
further in vivo development for cancer therapy.
3. Conclusions
In this work, we found that non-layered metallic Ti can be fabricated into 2D
nanosheets (NSs) using liquid-phase exfoliation (LPE), with an average thickness of
~3 nm, and lateral size of less than 50 nm. The ultrathin Ti NSs exhibited a high
extinction coefficient of 20.8 L.g−1.cm−1. They also showed a good photothermal
conversion efficiency (PTCE) of 73.4% under NIR irradiation (808 nm laser), owing
to the localized surface plasmon resonances (LSPR) resulting from the metal nature of
the Ti NSs. This PTCE is significantly higher than those of classic photothermal
agents such as Au (21%) and emerging 2D NMs, such as MoS2 (24.4%), black
phosphorus (28.4%) and Ti3C2 MXene (30.6%). The Ti NSs also degraded in a
relatively short time (77% degradation over 30 days in water at ~30 °C). Thanks to
the oxidization of Ti NSs, the Ti NSs presented the hydrophilic feature. In addition to
their high photothermal performance, the as-prepared Ti NSs showed trace amounts of
toxicity both in vitro and in vivo, and could also be encapsulated using conventional
coating materials such as PEG, to improve their water dispersibility. As expected, the
Ti NSs exhibited excellent therapeutic effects in a photothermal tumor therapy study,
in hepatocellular carcinoma models. Consequently, the widely used biomedical
material of Ti was demonstrated as a promising photothermal agent with excellent
biocompatibility and high photothermal efficacy. Given the strengths of LSPR for
elemental metals, our study would arouse a wide interest for PTT using other
elemental transition metal.
4. Experimental section
Materials
Commercially available Ti powder was purchased from Macklin Company.
DSPE-PEG, 5000 Da was purchased from Nanocs Inc. (New York, USA). All cell lines
were obtained from American Type Culture Collection (ATCC). Acridine orange (AO)
and propidium iodide (PI) assay kits were obtained from Logos Biosystems. Ultrapure
water (18.25 MΩ.cm, 25 °C) was used to prepare water-based dispersions. All other
reagents used in this work were analytical reagent grade.
Synthesis of Ti NSs
The ultrathin Ti NSs were prepared from non-layered bulk Ti using liquid-phase
exfoliation. Typical exfoliation was mainly divided into two steps: probe sonication
and bath sonication. 500 mg of Ti powder was dispersed in 100 ml of IPA. The
suspension was then subjected to probe sonication for 10 h at a power of 240 W. To
avoid over heating during the sonication process, the sonication was set to an on/off
cycle of 2/4 seconds and the Ti dispersion was kept in ice water. Subsequently, the Ti
dispersion underwent bath sonication at a power of 360 W for 10 h. The water bath
temperature was controlled at 10 °C.
After sonication, the resulting dispersions were centrifuged at 2000 g for 30
min to remove the un-exfoliated component. The supernatant containing the Ti NSs
was decanted gently and then centrifuged for a further 30 min at 12000 g. The
precipitate was dried in a vacuum drying oven. To avoid oxidation, the Ti NSs were
packaged in tinfoil and stored at 4 oC in the refrigerator for further characterization or
use in bio-experiments.
The PEGylated Ti NSs were further prepared. 1 mg of DSPE-PEG was dispersed
in 1 ml water. 5 ml Ti NSs dispersion in water with concentration of 100 ppm was
involved sonication for 30 min and then mixed with PEG solution. The mixture
underwent bath sonication for several min and stir for 3 h. Then, in order to remove the
excess PEG molecules, the resulting mixture was ultrafiltered in Amicon tubes
(MWCO 100kDa; Millipore) at 1000 g until all water was filtered out, and was
washed 2 times using the same method. The pure PEGylated Ti NSs were re-suspended
in ultrapure water or culture media for further use.
Characterization
To confirm the three-dimensional morphology of Ti NSs, both atomic force
microscopy (AFM, Bruker, Dimension Fastscan) and transmission electron microscopy
(TEM, JEM1230) were used. AFM samples were prepared by dispersing on a silicon
substrate using the drop-casting method and AFM images were scanned at 512 pixels
per line. High-resolution TEM images and selected-area electron diffraction (SAED)
were obtained using a Tecnai G2 F30 with an acceleration voltage of 300 kV. X-ray
diffraction (XRD) patterns were acquired using a Philips X'Pert Pro Super
diffractometer. X-ray photoelectron spectra (XPS) were acquired using a VG Escalab
MK II spectrometer. UV-Vis-NIR absorption spectra were measured in the range 400–
1000 nm using a Cary 60 spectrometer from Agilent. The Fourier transform infrared
(FTIR) spectra were measured to verify the PEG coating of Ti NSs. An 808 nm
fiber-coupled continuous semiconductor diode laser, LSR808H from Lasever Inc., was
used as the laser source for the photothermal experiments. For the photothermal
temperature measurements, an infrared thermal imaging camera (FLIR E-60) and
thermocouple were used.
Cell culture assays
Mouse melanoma B16 cells and mouse macrophage J774A.1 cells were cultured in
Dulbecco's Modified Eagle Medium (DMEM) with high glucose (Hyclone). Human
hepatocellular carcinoma SMMC-7721 cells were maintained in a 1:1 mixture of
DMEM and Ham's F-12 medium (Hyclone). All of the culture media were
supplemented with 10% fetal bovine serum (Gibco) and 1% Pen/Strep (Gibco). Cells
were cultured in an incubator (Thermo Fisher Scientific) at 37 °C with 5% CO2.
In vitro experiments
SMMC-7721, B16 or J774A.1 cells were incubated in 96-well plates. After adherence
(~12 h later), cells were treated with different culture media containing naked Ti NSs
or PEGylated Ti NSs, or remained untreated (Mock). For in vitro cytotoxicity assays,
cells were directly subjected to cell-counting kit (CCK8) assays (Beyotime
Biotechnology) 24 hours post Ti NSs incubation. For in vitro photothermal experiments,
cells were incubated with Ti NSs for 4 h and were then irradiated with an 808 nm laser,
1 W.cm−2 for 10 min. After a further 24 h, cells were subjected to CCK8 assays
according to the manufacturer instructions. The relative cell viability was normalized to
the Mock samples (concentration of Ti NSs = 0 ppm) of each cell line. For fluorescent
imaging of photothermal effects, cells were treated with the indicated concentrations of
PEGylated Ti NSs for 4 h, and were subsequently irradiated as described above. 24 h
post irradiation, cells were subjected to calcein AM/PI staining (Sigma), live cells
(green, stained by calcein AM) and dead cells (red, stained by PI) were revealed.
Mouse experiments
All of the animal studies were performed in compliance with the guidelines approved
by the Animal Welfare and Research Ethics Committee at Shenzhen University (ID:
2017003). The mice used in this work were purchased from Guangdong Medical
Laboratory Animal Center (Guangzhou, China). Mice were euthanized before the
ACUSU maximum allowable tumor burden of 2 cm3.
In vivo toxicity experiments
For all in vivo studies, Ti NSs or PEGylated Ti NSs were dispersed in saline.
6-week-old female Balb/c nude mice were randomly divided into 3 groups (n=5 each
group). Mice were subcutaneously injected with saline, Ti NSs or PEGylated Ti NSs at
day 1. The dose of Ti NSs or PEGylated Ti NSs was adjusted to 0.25 mg.kg−1 with an
injection volume of 100 μl and concentration of 50 ppm. To monitor the in vivo
toxicity, body weight was measured every 2 days until day 15. The mice were then
euthanized and the major organs, heart, liver, spleen, lung and kidney were collected for
H&E staining as previously reported.[69]
In vivo photothermal tumor therapy
To study the PTT efficiency of PEGylated Ti NSs in treating tumor tissue, human
hepatocellular carcinoma models were established by subcutaneous injection of 5×106
SMMC-7721 cells in the left flank of Balb/c nude mice (6-week-old, female).
Approximately 10 days post injection, the tumor volumes reached 100–200 mm3. The
tumor bearing mice were randomly divided into 4 groups (n=5 each group) for different
treatments by intratumoral injection: group 1, saline; group 2, PEGylated Ti NSs;
group3, saline with NIR irradiation; group 4, PEGylated Ti NSs with NIR irradiation.
To achieve an injection dose of 0.25 mg.kg−1 of PEGylated Ti NSs, the injection
volume was 100 μl of 50 ppm in saline.
6 h post injection, the mice were anaesthetized and the tumors were irradiated with
the NIR laser (808 nm, 1 W.cm−2, 5 min). The distance between the laser point and
tumor was 6 cm, and the temperature of the tumor was monitored using an infrared
thermal imager. Following laser irradiation, the body weights and tumor volumes were
recorded every 2 days. Tumor dimensions were measured using a caliper and the
volumes were calculated using the formula (volume=length×width2/2, V=l×w2/2). At
day 15, all of the mice were euthanized and the major organs, heart, liver, spleen, lung
and kidney were isolated for H&E staining to assess the possible damage caused by the
PTT.
Statistical analysis
All of the data were analyzed using Graphpad Prism software and are presented as
means ± SD. For in vitro studies, cell viability was normalized to the mean of Mock
samples for each cell line, which was set to be 100%. Analysis of significance was
performed by student's t-tests. For animal experiments, the significance analysis of
tumor volumes between the four groups was performed using multiple t-tests. p<0.05
was considered statistically significant, *p<0.05, **p<0.01, ***p<0.001.
Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.
Acknowledgements
This research is partially supported by the National Natural Science Fund (Grant Nos.
61435010 and 61575089), Science and Technology Innovation Commission of
Shenzhen (KQTD2015032416270385 and JCYJ20150625103619275), and China
Postdoctoral Science Foundation (Grant No. 2017M612730). We thank Sarah Dodds,
PhD, from Liwen Bianji, Edanz Editing China (www.liwenbianji.cn/ac), for editing the
English text of a draft of this manuscript.
Reference
[1] W. F. Oliveira, I. R. S. Arruda, G. M. M. Silva, G. Machado, L. C. B. B.
Coelho, M. T. S. Correia, Mater. Sci. Eng. C 2017, 81, 597.
[2] C. N. Elias, J. H. C. Lima, R. Valiev, M. A. Meyers, JOM 2008, 60, 46.
[3] S. Spriano, M. Bronzoni, F. Rosalbino, E. Vernè, J. Mater. Sci. Mater. Med.
2005, 16, 203.
[4] C. E. Wen, Y. Yamada, K. Shimojima, Y. Chino, T. Asahina, M. Mabuchi,
J. Mater. Sci. Mater. Med. 2002, 13, 397.
[5] E. D. de Avila, B. P. Lima, T. Sekiya, Y. Torii, T. Ogawa, W. Shi, R. Lux,
Biomaterials 2015, 67, 84.
[6] K. Yang, L. Hu, X. Ma, S. Ye, L. Cheng, X. Shi, C. Li, Y. Li, Z. Liu, Adv.
Mater. 2012, 24, 1868.
[7] Y. Xia, W. Li, C. M. Cobley, J. Chen, X. Xia, Q. Zhang, M. Yang, E. C. Cho,
P. K. Brown, Acc. Chem. Res. 2011, 44, 914.
[8]
J. T. Robinson, S. M. Tabakman, Y. Liang, H. Wang, H. S. Casalongue, D.
Vinh, H. Dai, J. Am. Chem. Soc. 2011, 133, 6825.
[9] Q. Tian, F. Jiang, R. Zou, Q. Liu, Z. Chen, M. Zhu, S. Yang, J. Wang, J.
Wang, J. Hu, ACS Nano 2011, 5, 9761.
[10] Z. Li, Y. Zhang, X. Wu, L. Huang, D. Li, W. Fan, G. Han, J. Am. Chem. Soc.
2015, 137, 5304.
[11] Z. Zha, X. Yue, Q. Ren, Z. Dai, Adv. Mater. 2013, 25, 777.
[12] X. Huang, S. Tang, X. Mu, Y. Dai, G. Chen, Z. Zhou, F. Ruan, Z. Yang, N.
Zheng, Nat. Nanotechnol. 2011, 6, 28.
[13] T. Liu, C. Wang, X. Gu, H. Gong, L. Cheng, X. Shi, L. Feng, B. Sun, Z. Liu,
Adv. Mater. 2014, 26, 3433.
[14] W. Yin, L. Yan, J. Yu, G. Tian, L. Zhou, X. Zheng, X. Zhang, Y. Yong, J. Li,
Z. Gu, ACS Nano 2014, 8, 6922.
[15] S. S. Chou, B. Kaehr, J. Kim, B. M. Foley, M. De, P. E. Hopkins, J. Huang,
C. J. Brinker, V. P. Dravid, Angew. Chemie 2013, 52, 4160.
[16] S. Wang, X. Li, Y. Chen, X. Cai, H. Yao, W. Gao, Y. Zheng, X. An, J. Shi,
H. Chen, Adv. Mater. 2015, 27, 2775.
[17] L. Cheng, J. Liu, X. Gu, H. Gong, X. Shi, T. Liu, C. Wang, X. Wang, G. Liu,
H. Xing, Adv. Mater. 2014, 26, 1886.
[18] L. Cheng, C. Yuan, S. Shen, X. Yi, H. Gong, K. Yang, Z. Liu, ACS Nano
2015, 9, 11090.
[19] Z. Sun, H. Xie, S. Tang, Y. XF, Z. Guo, J. Shao, H. Zhang, H. Huang, H.
Wang, C. PK, Angew. Chemie 2015, 54, 11526.
[20] W. He, K. Ai, C. Jiang, Y. Li, X. Song, L. Lu, Biomaterials 2017, 132, 37.
[21] P. D. Howes, S. Rana, M. M. Stevens, Chem. Soc. Rev. 2014, 43, 3835.
[22] D. J. De Aberasturi, A. B. Serrano-Montes, L. M. Liz-Marzán, Adv. Opt.
Mater. 2015, 3, 602.
[23] R. Li, L. Zhang, L. Shi, P. Wang, ACS Nano 2017, 11, 3752.
[24] J. Xuan, Z. Wang, Y. Chen, D. Liang, L. Cheng, X. Yang, Z. Liu, R. Ma, T.
Sasaki, F. Geng, Angew. Chemie 2016, 55, 14569.
[25] H. Lin, X. Wang, L. Yu, Y. Chen, J. Shi, Nano Lett. 2017, 17, 384.
[26] Y. Chen, C. Tan, H. Zhang, L. Wang, Chem. Soc. Rev. 2015, 44, 2681.
[27] H. Zhao, M. Wang, P. Zhou, Q. Wang, Z. Zhou, D. Wang, H. Yang, S. Yang,
J. Mater. Sci. 2017, 52, 13356.
[28] G. Ou, Z. Li, D. Li, L. Cheng, Z. Liu, H. Wu, Nano Res. 2016, 9, 1236.
[29] J. Wang, Y. Li, L. Deng, N. Wei, Y. Weng, S. Dong, D. Qi, J. Qiu, X. Chen,
T. Wu, Adv. Mater. 2017, 29.
[30] U. Guler, J. C. Ndukaife, G. V Naik, A. G. A. Nnanna, A. V Kildishev, V. M.
Shalaev, A. B. T.-L. and E.-O. Boltasseva, 2014; pp. 1–2.
[31] U. Guler, S. Suslov, A. V Kildishev, A. Boltasseva, V. M. Shalaev,
Nanophotonics 2015, 4, 269.
[32] H. Zhang, C. Tan, L. Zhao, P. Yu, Y. Huang, B. Chen, Z. Lai, X. Qi, M. H.
Goh, X. Zhang, Angew. Chemie 2017, 56, 7842.
[33] K. Rasool, M. Helal, A. Ali, C. E. Ren, Y. Gogotsi, K. A. Mahmoud, ACS
Nano 2016, 10, 3674.
[34] F. Wang, C. Yang, M. Duan, Y. Tang, J. Zhu, Biosens. Bioelectron. 2015,
74, 1022.
[35] B. Xu, M. Zhu, W. Zhang, X. Zhen, Z. Pei, Q. Xue, C. Zhi, P. Shi, Adv.
Mater. 2016, 28, 3411.
[36] Q. Xue, H. Zhang, M. Zhu, Z. Pei, H. Li, Z. Wang, Y. Huang, Y. Huang, Q.
Deng, J. Zhou, S. Du, Q. Huang, C. Zhi, Adv. Mater. 2017, 29, 1604847.
[37] A. O. Govorov, H. H. Richardson, Nano Today 2007, 2, 30.
[38] F. Wang, Z. Wang, T. A. Shifa, Y. Wen, F. Wang, X. Zhan, Q. Wang, K. Xu,
Y. Huang, L. Yin, Adv. Funct. Mater. 2017, 27.
[39] F. Wang, Z. Wang, Q. Wang, F. Wang, L. Yin, K. Xu, Y. Huang, J. He,
Crystengcomm 2015, 26, 292001.
[40] Z. Fan, Y. Zhu, X. Huang, Y. Han, Q. Wang, Q. Liu, Y. Huang, C. L. Gan,
H. Zhang, Angew. Chemie Int. Ed. 2015, 54, 5672.
[41] X. Huang, S. Li, Y. Huang, S. Wu, X. Zhou, S. Li, C. L. Gan, F. Boey, C. A.
Mirkin, H. Zhang, Nat. Commun. 2011, 2, 292.
[42] Y. Liu, H. Cheng, M. Lyu, S. Fan, Q. Liu, W. Zhang, Y. Zhi, C. Wang, C.
Xiao, S. Wei, J. Am. Chem. Soc. 2014, 136, 15670.
[43] Y. Sun, Z. Sun, S. Gao, H. Cheng, Q. Liu, J. Piao, T. Yao, C. Wu, S. Hu, S.
Wei, Nat. Commun. 2012, 3, 1057.
[44] Q. Wang, K. Xu, Z. Wang, F. Wang, Y. Huang, M. Safdar, X. Zhan, F.
Wang, Z. Cheng, J. He, Nano Lett. 2015, 15, 1183.
[45] Q. Wang, M. Safdar, K. Xu, M. Mirza, Z. Wang, J. He, ACS Nano 2014, 8,
7497.
[46] J. N. Coleman, M. Lotya, A. O'Neill, S. D. Bergin, P. J. King, U. Khan, K.
Young, A. Gaucher, S. De, R. J. Smith, Science (80-. ). 2011, 42, 568.
[47] Y. Liu, C. Hao, M. Lyu, S. Fan, Q. Liu, W. Zhang, Y. Zhi, C. Wang, X.
Chong, S. Wei, J. Am. Chem. Soc. 2014, 136, 15670.
[48] J. R. Brent, D. J. Lewis, T. Lorenz, E. A. Lewis, N. Savjani, S. J. Haigh, G.
Seifert, B. Derby, P. O'Brien, J. Am. Chem. Soc. 2015, 137, 12689.
[49] K. Y. Zhu, A. Vassel, F. Brisset, K. Lu, J. Lu, Acta Mater. 2004, 52, 4101.
[50] R. L. Kurtz, V. E. Henrich, Surf. Sci. Spectra 1998, 5, 179.
[51] Y. T. Sul, C. B. Johansson, S. Petronis, A. Krozer, Y. Jeong, A. Wennerberg,
T. Albrektsson, Biomaterials 2002, 23, 491.
[52] J. E. Gonçalves, S. C. Castro, A. Y. Ramos, M. C. M. Alves, Y. Gushikem,
J. Electron Spectrosc. Relat. Phenom. 2001, 114, 307.
[53] E. Galvanetto, F. P. Galliano, F. Borgioli, U. Bardi, A. Lavacchi, Thin Solid
Films 2001, 384, 223.
[54] R. J. Nemanich, C. C. Tsai, G. A. N. Connell, Phys. Rev. Lett. 1980, 44, 273.
[55] U. Balachandran, N. G. Eror, J. Solid State Chem. 1982, 42, 276.
[56] † Nobuyuki Sakai §, ‡ Akira Fujishima, † And Toshiya Watanabe, K.
Hashimoto†, J. Phys. Chem. B 2003, 107, 1028.
[57] † Nobuyuki Sakai, † Wang § Rong, Akira Fujishima †, ‡ Toshiya
Watanabe ⊥ And, K. Hashimoto‡, Langmuir 2012, 14, 5918.
[58] L. Cheng, J. Liu, X. Gu, H. Gong, X. Shi, T. Liu, C. Wang, X. Wang, G. Liu,
H. Xing, Adv. Mater. 2014, 26, 1886.
[59] H. Lin, S. Gao, C. Dai, Y. Chen, J. Shi, J. Am. Chem. Soc. 2017.
[60] X. Ding, C. H. Liow, M. Zhang, R. Huang, C. Li, H. Shen, M. Liu, Y. Zou,
N. Gao, Z. Zhang, J. Am. Chem. Soc. 2014, 136, 15684.
[61] M. F. Tsai, S. H. Chang, F. Y. Cheng, V. Shanmugam, Y. S. Cheng, C. H.
Su, C. S. Yeh, ACS Nano 2013, 7, 5330.
[62] D. K. Roper, W. Ahn, M. Hoepfner, J. Phys. Chem. C Nanomater.
Interfaces 2007, 111, 3636.
[63] B. Wang, J. H. Wang, Q. Liu, H. Huang, M. Chen, K. Li, C. Li, X. F. Yu, P.
K. Chu, Biomaterials 2014, 35, 1954.
[64] C. M. Hessel, V. P. Pattani, M. Rasch, M. G. Panthani, B. Koo, J. W.
Tunnell, B. A. Korgel, Copper selenide nanocrystals for photothermal therapy. Nano
Lett. 2011, 11, 2560–2566.
[65] W. Tao, X. Ji, X. Xu, M. A. Islam, Z. Li, S. Chen, P. E. Saw, H. Zhang, Z.
Bharwani, Z. Guo, Angew. Chemie 2017.
[66] C. Lee, C. Hong, H. Kim, J. Kang, H. M. Zheng, Photochem. Photobiol.
2010, 86, 981.
[67] C. Hong, J. Kang, J. Lee, C. Lee, H. Zheng, S. Hong, D. Lee, J. Cancer Ther.
2010, 1, 52.
[68] W. Ni, M. Li, J. Cui, Z. Xing, Z. Li, X. Wu, E. Song, M. Gong, W. Zhou,
Mater. Sci. Eng. C 2017, 81, 252.
[69] W. Tao, X. Zhu, X. Yu, X. Zeng, Q. Xiao, X. Zhang, X. Ji, X. Wang, J. Shi,
H. Zhang, Adv. Mater. 2017, 29.
Figure 1. Schematic representation of the liquid-phase exfoliation (LPE) of Ti nanosheets (NSs)
and Ti NSs based plasmonic photothermal therapy (Ti-PPTT) for in vitro and in vivo experiments.
Figure 2. Typical characterization of exfoliated Ti NSs, including transmission electron microscopy
(TEM), high-resolution transmission electron microscopy (HRTEM), atomic force microscopy
(AFM), X-ray photoelectron spectroscopy (XPS), X-ray diffraction (XRD) and Raman
spectroscopy. a) TEM and b) AFM image of Ti NSs and Ti QDs. c) Crystal lattice, selected-area
electron diffraction (SAED) and corresponding Fast Fourier Transformation (FFT) of Ti NSs. d), e)
and f) XPS, XRD and Raman spectra of bulk and exfoliated Ti NSs.
Figure 3. Characterization of PEGylated Ti NSs. a) FTIR spectra of pure PEG, PEGylated
Ti NSs and naked Ti NSs. b) AFM image of PEGylated Ti NSs. c) Dynamic Light Scattering
(DLS) size distribution of naked Ti NSs and PEGylated Ti NSs. d) STEM EDS mapping of
PEGylated Ti NSs.
Figure 4. Basic characterization of photothermal effect of Ti NSs. a) Photograph of Ti NSs
dispersed in water at different concentrations. b) Absorption spectra of Ti NSs. c)
Photothermal temperature increase of water dispersed Ti NSs. d) The normalized absorbance
intensity divided by the length of light path (A/L) at 808 nm. e), The linear fitting relationship
between -ln θ and cooling time in one photothermal cycle, to determine the photothermal
conversion efficiency (PTCE).
Figure 5. Photostability and temporal stability of Ti NSs. a) Photothermal stability for six
cycles. The dotted lines show the relatively stable photothermal performance of Ti NSs. b)
Laser stability and temporal stability of Ti NSs over 2 hours shown by absorbance. c) and d)
Temporal stability of absorbance and photothermal temperature over 30 days.
Figure 6. Toxicity assays. In vitro cytotoxicity of a) naked Ti NSs and b) PEGylated Ti NSs
was assessed using SMMC-7721, B16 and J774A.1 cells. Ti NSs dispersed in the
corresponding culture media with concentrations of 0, 10, 25, 50 and 100 ppm were incubated
with the cells. c-d) In vivo toxicity. c) The body weight of mice measured at the indicated times
and d) organ conditions assessed by H&E staining at day 15 post subcutaneous injection of
saline, Ti NSs or PEGylated Ti NSs.
Figure 7. In vitro photothermal experiments. The photothermal cell killing effect of a) naked Ti
NSs and, b) PEGylated Ti NSs on SMMC-7721, B16 and J774A.1 cells for different concentrations
under the same NIR irradiation conditions. Cells were incubated with Ti NSs or PEGylated Ti NSs
at concentrations of 0, 5, 10, 20, 30 and 50 ppm for 4 h, followed by NIR laser irradiation (1 W.cm−2,
10 min). c) Photothermal cell killing effect of PEGylated Ti NSs (50 ppm) for different irradiation
times. SMMC-7721 cells were incubated with PEGylated Ti NSs for 4 h, followed by NIR laser
irradiation (1 W.cm−2) for different times (0, 2, 5, 8, 10 min). The Mock samples were untreated. d)
Fluorescent image of photothermal effect on SMMC-7721 cells post irradiation. Cells were
incubated with the indicated concentrations of PEGylated Ti NSs, followed by NIR irradiation (808
nm, 1 W.cm−2, 10 min), and then stained with calcein AM/PI. n=3 biological replicates, *** p<0.001,
t-tests.
Titanium (Ti), a widely used biomedical material, is first demonstrated for
photothermal cancer therapy (PTT) and with high efficiency and excellent
biocompatibility. Using liquid-phase exfoliation (LPE), the non-layered metal Ti can
be fabricated into 2-dimentional (2D) Ti nanosheets (NSs). The 2D Ti NSs exhibited
high extinction coefficient of 20.8 Lg−1cm−1 and high photothermal conversion
coefficiency (PTCE) of 73.4%, owing to localized surface plasmon resonances
(LSPR). Our study could arouse a wide interest in the potential efficient PTT for other
elemental transition metals owing to their common LSPR.
Keywords: biocompatibility, titanium nanosheets, photothermal therapy, liquid-phase
exfoliation, plasmon resonance
Zhongjian Xiea,1, Shiyou Chena,1, Quan Liuc,d, Zhitao Lina, Jinlai Zhaod, Taojian Fana,
Dou Wangc,d, Liping Liub,*, Shiyun Baob, Dianyuan Fana, and Han Zhanga,*
Biocompatible Two-dimensional Titanium Nanosheets for
Efficient Plasmonic Photothermal Cancer Therapy
|
1909.09525 | 1 | 1909 | 2019-09-20T14:20:07 | Homogeneous enhancement of electric near-field in all-dielectric metasurfaces composed of cluster-based unit cells | [
"physics.app-ph",
"physics.optics"
] | In order to construct a dielectric analog of spaser, we study theoretically and experimentally several configurations of cluster-based unit cells for an all-dielectric metasurface characterized by resonant conditions of the trapped mode excitation. Excitation of the trapped mode is realized by either specific displacement of particles in the cluster, or perturbation of the equidistantly spaced particles by off-centered round holes or coaxial-sector notches. It turns out that the latter approach is more advantageous for enhancement of electric near-field with homogeneous distribution in-plane of the structure and its strong localization outside the high-refractive-index dielectric particles. This feature opens prospects for realization of highly desirable subwavelength flat lasing structures based on strong near-field interaction with substances exhibiting pronounced nonlinear characteristics and properties of gain media. | physics.app-ph | physics |
Homogeneous enhancement of electric near-field
Homogeneous enhancement of electric near-field in all-dielectric
metasurfaces composed of cluster-based unit cells
Anton S. Kupriianov,1 Kateryna L. Domina,2 Vyacheslav V. Khardikov,2, 3 Andrey B. Evlyukhin,4 and
Vladimir R. Tuz3, 5, a)
1)College of Physics, Jilin University, 2699 Qianjin St., Changchun 130012,
China
2)School of Radio Physics, V. N. Karazin Kharkiv National University, 4, Svobody Sq., Kharkiv 61022,
Ukraine
3)Institute of Radio Astronomy of National Academy of Sciences of Ukraine, 4 Mystetstv St., Kharkiv 61002,
Ukraine
4)Institute of Quantum Optics, Leibniz Universitat Hannover, 30167 Hannover,
Germany
5)International Center of Future Science, State Key Laboratory of Integrated Optoelectronics,
College of Electronic Science and Engineering, Jilin University, 2699 Qianjin St., Changchun 130012,
China
(Dated: 23 September 2019)
In order to construct a dielectric analog of spaser, we study theoretically and experimentally
several configurations of cluster-based unit cells for an all-dielectric metasurface characterized
by resonant conditions of the trapped mode excitation. Excitation of the trapped mode is
realized either by specific displacement of particles in the cluster, or by perturbation of the
equidistantly spaced particles by off-centered round holes or coaxial-sector notches. It turns
out that the latter approach is more advantageous for enhancement of electric near-field
with homogeneous distribution in-plane of the structure and its strong localization outside
the high-refractive-index dielectric particles. This feature opens prospects for realization of
highly desirable subwavelength flat lasing structures based on strong near-field interaction
with substances exhibiting pronounced nonlinear characteristics and properties of gain media.
Considerable interest in the study of metamateri-
als is due to the prospects of their use in practical
devices.1 Metamaterials can be a suitable platform for
many optical systems, such as sensors2 and perfect
absorbers.3 They allow one to enhance quantum dots
luminescence,4,5 realize optical switching6,7 and other re-
lated operations8 when combined with optically active
and nonlinear substances.9,10 In the latter case, thin pla-
nar metamaterials (metasurfaces) are of special interest,
due to their higher workability.11,12
In particular, it is proposed to combine metasurfaces
with optically active materials to obtain parametric gain
systems and develop amplifying or lasing devices13 (e.g.
spaser -- Surface Plasmon Amplification by Stimulated
Emission of Radiation14,15).
In a metasurface-based
spaser a regular array of subwavelength metallic res-
onators is supported by a slab of gain medium containing
quantum dots. A special type of symmetry-broken res-
onators is chosen to ensure excitation of a high-quality-
factor (high-Q) trapped mode with reduced radiative
losses.16,17 The collective plasmonic oscillations in such
resonators lead to the emission of spatially and temporar-
ily coherent light in the direction normal to the metasur-
face array. The spaser system is very thin and compact
and benefits from the strong electric field localization
near the surface associated with plasmons. Nevertheless,
although the concept of the metasurface-based spaser is
well developed, its practical implementation is difficult
due to requirement of high pumping power, which ad-
versely affects the system. It arises from excessive heat
a)Electronic mail: [email protected] and [email protected]
losses inherent in plasmonic nanostructures in infrared
and visible parts of spectrum. Moreover, asymmetric
plasmonic particles composing the metasurface typically
have a quite complicated form, so it is difficult to fabri-
cate them on the nanoscale.
can overcome
All-dielectric metasurfaces
above-
mentioned drawbacks of plasmonic structures while be-
ing simple in manufacturing.18 -- 20 The resonant behav-
ior of light in high-refractive-index (high-n) dielectric
nanoparticles21,22 makes it possible to reproduce many
subwavelength effects demonstrated in plasmonic sys-
tems due to the electric field localization, but without
much losses and energy dissipation into heat.
In ad-
dition, the coexistence of strong electric and magnetic
multipolar resonances, as well as their interference and
corresponding enhancement of near-fields in dielectric
nanoparticles23 bring much novel functionality to simple
geometries, especially for the nonlinear regimes24 -- 26 and
metadevices widespread applications.27 However, since
the electric near-field is mainly localized inside the high-
n particles, it interacts only with a small portion of the
surrounding gain medium, which limits the optical out-
put of the overall lasing system. This fact impairs ad-
vantageous use of such metasurfaces in construction of a
dielectric analog of spaser.
Combining dielectric particles into clusters makes it
possible to overcome this obstacle. For instance, in an
all-dielectric metasurface composed of bars of different
length a trapped mode can be excited.28,29 When reso-
nant conditions are satisfied for the trapped mode, large
electric near-field enhancement and localization in the
surrounding medium appears inside the nano gaps intro-
duced at the centre of bars.30 Moreover, it was recently
Homogeneous enhancement of electric near-field
2
reported31 -- 35 that a proper choice of asymmetric dielec-
tric particles and their ordering into clusters provides
advanced flexibility in obtaining the desired near-field
configuration, together with significant field localization
caused by the trapped mode excitation.
In development of the concept of cluster-based meta-
surfaces, in the present Letter we propose and character-
ize several configurations of an all-dielectric metasurface
whose array sustains resonant conditions of the trapped
mode. We demonstrate that under these conditions the
resonant electric near-field appears to be homogeneously
distributed in-plane of the structure and is strongly local-
ized outside the high-n dielectric particles. Our finding
is confirmed by both numerical simulations and proof-of-
principle microwave measurements of the near-fields in
the actual metasurface prototypes.
In what follows, we perform a numerical and ex-
perimental study of resonant characteristics of an all-
dielectric metasurface whose unit cell is composed of four
cylindrical resonators (disks) [see Fig. 1(a)]. The size of
a square unit cell is p. Disks are made of a nonmagnetic
dielectric material with permittivity εd. The radius and
thickness of the disks are rd and hd, respectively. The
disks are immersed into a dielectric substrate (host) with
permittivity εs and thickness hs.
It was theoretically shown21,36,37 and experimentally
confirmed38 that a metasurface consisting of an array
of equidistantly spaced dielectric particles demonstrate a
resonant response arising from excitation of electric and
magnetic multipole moments of individual particles. If
such particles (e.g. dielectric disks) are arranged into
clusters, the inter-particle coupling leads to a complex
collective behavior of modes which is different from that
of individual particles.31,39 Moreover, specific perturba-
tions of the particles can result in the appearance of addi-
tional ultra high-Q resonances in the overall metasurface
response. These resonances are related to the collective
trapped mode excitation.32 -- 34
Therefore, in order to differentiate the features of the
cluster-based configuration from those influenced by the
particles perturbations, further we consider three partic-
ular designs of the metasurface. The first design is re-
alized by displacing the non-perturbed disks toward the
unit cell center by a distance s along the cell's diagonals
[Fig. 1(b)]. In the second design, the disks of the array
are equally spaced, whereas they are perturbed by an ec-
centric through hole. The hole radius is indicated as rh,
the distance from the center of the disk to the center of
the hole is s. Within the unit cell all holes are oriented
inward the center of the cluster, as shown in Fig. 1(c).
In the third design, equally spaced disks are perturbed
by a coaxial-sector notch [Fig. 1(d), see also discussion
of benefits of such perturbation approach in Refs. 32 and
34]. The coaxial-sector notch is characterized by the ra-
dius of the middle line s, width rh, and opening angle α.
The notches of the disks are oriented inward the center of
the unit cell. For all designs we define the dimensionless
asymmetry parameter θ of the metasurface. The param-
eter θ varies in the range [0 − 1] and is associated with
the displacement of the disks toward the cluster center
[θ = 2s/(p − 2rd)], increase in the radius of the circular
hole (θ = 2rh/rd), and the opening angle of the notch
FIG. 1. (a) Schematic view of an all-dielectric metasurface
whose cluster-based unit cell is composed of (b) solid disks
shifted to the center of unit cell along its diagonals, (c) disks
with off-centered round holes, and (d) disks with coaxial-
sector notches.
[θ = sin(α/2)] for the first, second, and third design, re-
spectively.
The cluster forming metasurface corresponds to the
group C4v for all chosen designs.33 This group has the
four-fold symmetry axis with respect to rotation around
the z axis. For the metasurfaces whose unit cell symme-
try belongs to the rotational groups C4v, there is polar-
ization independence of the structure under normal in-
cidence conditions.34 Therefore, in what follows we con-
sider irradiation of the metasurface by a normally inci-
dent plane wave ((cid:126)k = {0, 0,−kz}), where the electric field
vector is directed along the x axis ( (cid:126)E = {Ex, 0, 0}).
In accordance with our available experimental equip-
ment, we have chosen the microwave part of the spectrum
to characterize the metasurface and confirm its features.
Thus, all the geometrical and material parameters of the
dielectric particles as well as the structure period are cho-
sen so that the metasurface operates in the specified fre-
quency band (1 − 15 GHz). All numerical simulations
of the electromagnetic response of the metasurface are
performed with the use of the rf module of the com-
mercial COMSOL Multiphysics R(cid:13) electromagnetic solver.
The results of simulation of the transmission character-
istic of the metasurface versus frequency and asymmetry
parameter are collected in Fig. 2 for the three proposed
metasurface designs. In these calculations, possible losses
in constitutive materials of the metasurface are ignored.
In the spectra of the metasurface composed of non-
perturbed disks an additional resonance of reduced trans-
mission appears in the frequency band of interest, pro-
vided that the equidistant arrangement of disks in the
array is violated [Fig. 2(a)]. Since this resonance is as-
sociated with some violation in the regular array, it can
be related to the excitation of the trapped mode. As
the asymmetry parameter increases, the quality factor of
hdhs(b)(a)(d)(c)αssrpyxzEkHdrdrhrsdrh Homogeneous enhancement of electric near-field
3
Intuition suggests that, the disks should be spaced
equidistantly in the array in order to achieve a more
homogeneous electric near-field distribution within the
In this case, the trapped mode can be ex-
cluster.
cited by perturbing directly the disks.
Indeed, the
cluster-based metasurfaces composed of asymmetrically
perturbed disks also support the excitation of the TE01
mode of the cylindrical dielectric resonator which behaves
as a trapped mode of the array.33,34 The transmission
characteristic for corresponding designs is presented in
Figs. 2(b) and 2(c) for the metasurfaces whose disks are
perturbed by the off-centered round hole and coaxial-
sector notch, respectively.
For these two designs, the resonance in transmission
arises as soon as the perturbation in the disks is intro-
duced and is typical for the trapped mode excitation.
The quality factor of the resonance decreases and the
resonant frequency shifts toward higher frequencies as
the asymmetry parameter increases. The quality factor
of the resonance degrades faster for the structure com-
posed of disks with notch. This effect can be explained
by vastly reduced extent of dielectric in the cluster, when
the notch size increases.
From the electric near-field patterns plotted in-plane
of the metasurface [see insets in Figs. 2(b) and 2(c)]
one can conclude that at the resonant frequency of the
trapped mode excitation there is a longitudinal electric
dipole moment and magnetic moments, which are in-
duced in a similar way as those of above-discussed ar-
ray of non-perturbed disks. However, for these two de-
signs, the resulting electric near-field appears to be more
homogeneously distributed in-plane of the metasurface
and is mainly concentrated outside the disks. The local
magnetic field is sufficiently enhanced inside the parti-
cles holes and notches. It is important to note, that such
near-fields configuration persists even when the asymme-
try parameter becomes sufficiently large.
We check experimentally all the distinctive features of
the near-field distribution and the degree of field con-
centration by direct measurements for the metasurface
prototypes based on three chosen designs. The Taizhou
Wangling TP-series microwave ceramic is used as a ma-
terial for particles fabrication (the losses in this material
are estimated as tan δ = 1 × 10−3 at 10 GHz). The sets
of particles are prepared with the use of a precise milling
machine. Particles are arranged in a custom holder made
of a rigid foam material. All geometrical and material
parameters of particles and holder correspond to those
listed in the caption of Fig. 2. All details on the exper-
imental method as well as the sketch and photo of our
measurement setup can be found in Refs. 34 and 41.
The measured transmission for all proposed designs is
depicted in Fig. 3 and compared with the results of our
numerical simulation, which takes into account estimated
material losses in the actual metasurface. The corre-
sponding resonances related to the trapped mode exci-
tation are well identified in the spectra, and they have
different quality factors depending on the metasurface
design, as predicted above.
We use a near-field scanning system for mapping both
electric and magnetic near-fields at the corresponding
resonant frequencies. These fields are measured above
FIG. 2. Transmission versus frequency and asymmetry pa-
rameter for the idealized (loss-less) all-dielectric metasurface
whose cluster-based unit cell is composed of (a) non-perturbed
disks (rd = 4.0 mm), (b) disks perturbed by a round off-
centered hole (rd = 4.0 mm, s = 2.0 mm), and (c) disks per-
turbed by a coaxial-sector notch (rd = 4.5 mm, rh = 2.0 mm,
s = 2.0 mm).
Insets present magnitude and vectorial pat-
terns of the electric near-field calculated in the cross section
along the midline of the resonators (z = 0) at corresponding
values of the resonant frequency and asymmetry parameter.
The geometrical and material parameters of the metasurface
are: p = 32 mm, hd = 2.5 mm, hs = 25.0 mm, εd = 23.0, and
εs = 1.1.
the resonance decreases and the resonant frequency shifts
toward the lower frequencies.
From the cross-section patterns of the electric near-
field calculated at the corresponding resonant frequency
[see inset of Fig. 2(a)] one can conclude that the reso-
nant field induced inside each particle resembles the char-
acteristics of the TE01 mode of the individual cylindri-
cal dielectric resonator.40 It arises from the electromag-
netic coupling between closely spaced disks in the array
with violated periodicity. The electric near-field distribu-
tion inside the unit cell exhibits the longitudinal electric
dipole moment oriented along the x axis parallel to the
direction of the vector (cid:126)E of the incident wave. The mag-
netic moment induced in each particle is oriented parallel
to the z axis and changes sign along the y axis from one
particle to another. The resulting magnetic field is con-
centrated inside the particles, while the electric near-field
is distributed in-plane of the metasurface. Electric field is
partially concentrated outside the disks, and is localized
mainly in the center of cluster.
8.79.38.859.09.157.59.0Asymmetry parameter Frequency (GHz)7.77.958.28.458.70.50.60.90.50.750.80.00.671.0(b)(a)(c)8.08.58.458.350.0T 21.00.5yx0.00.5Exy1.0 Homogeneous enhancement of electric near-field
4
FIG. 3. Comparison of simulated and measured characteristics of the actual (lossy) cluster-based all-dielectric metasurface.
Insets present fragments of the metasurface prototypes. The geometrical and material parameters of the metasurface as well
as corresponding values of the resonant frequency and asymmetry parameter coincide with those listed in the caption of Fig. 2.
In the color maps the electric and magnetic near-field magnitudes are normalized on their corresponding maximal values.
the metasurface prototypes starting from 1 mm distance
from their surface performing subsequent probe moving
with the increment of 1 mm in vertical and horizontal di-
rections. The resulting color maps of the measured near-
field distribution confirm discussed above trapped mode
resonant conditions, distribution of the electric near-field,
orientations of the out-of-plane magnetic moments, and
concentration of the magnetic near-field inside the par-
ticles holes and notches (Fig. 3).
In these color maps
one can see electric near-field enhancement as a func-
tion of the distance to the metasurface array plane. At
the resonant frequency, the maximal magnitude of the
electric near-field is 1.3 × 104 and 2.0 × 104 times that
of the incident field for the metasurfaces composed of
non-perturbed and perturbed disks, respectively. Thus,
the use of perturbed disks is more advantageous for a
stronger electric near-field confinement and furthermore
provides more homogeneous field distribution.
In conclusion, we demonstrate that in the cluster-based
metasurface made of an array of equidistantly spaced per-
turbed high-n dielectric disks, the trapped mode can be
excited. For this mode the electric near-field appears to
be sufficiently homogeneous and strongly concentrated
outside the disks. This effect is extremely promising
for applied optics especially for advantageous use of pro-
posed metasurfaces in conjunction with gain media.
V.R.T. acknowledges the hospitality and support of
the Jilin University. A.B.E. acknowledges the support
of Deutsche Forschungsgemeinschaft (DFG, German Re-
search Foundation) under Germany's Excellence Strategy
within the Cluster of Excellence PhoenixD (EXC 2122,
Project ID 390833453).
A.S.K. and K.L.D. contributed equally to this work.
1N. I. Zheludev and Y. S. Kivshar, "From metamaterials to
metadevices," Nat. Mater. 11, 917 -- 924 (2012).
2T. Chen, S. Li, and H. Sun, "Metamaterials application in sens-
ing," Sensors 12, 2742 -- 2765 (2012).
3Y. Ra'di, C. R. Simovski, and S. A. Tretyakov, "Thin perfect
absorbers for electromagnetic waves: Theory, design, and real-
izations," Phys. Rev. Applied 3, 037001 (2015).
4K. Tanaka, E. Plum, J. Y. Ou, T. Uchino, and N. I. Zheludev,
"Multifold enhancement of quantum dot luminescence in plas-
monic metamaterials," Phys. Rev. Lett. 105, 227403 (2010).
5I. Staude, V. V. Khardikov, N. T. Fofang, S. Liu, M. Decker,
D. N. Neshev, T. S. Luk, I. Brener, and Y. S. Kivshar, "Shaping
photoluminescence spectra with magnetoelectric resonances in
all-dielectric nanoparticles," ACS Photonics 2, 172 -- 177 (2015).
6V. R. Tuz, S. L. Prosvirnin,
and L. A. Kochetova, "Optical
bistability involving planar metamaterials with broken structural
symmetry," Phys. Rev. B 82, 233402 (2010).
7V. R. Tuz, V. S. Butylkin, and S. L. Prosvirnin, "Enhancement
of absorption bistability by trapping-light planar metamaterial,"
J. Opt. 14, 045102 (2012).
8X. Fang, K. F. MacDonald, and N. I. Zheludev, "Controlling
light with light using coherent metadevices: All-optical transis-
tor, summator and invertor," Light Sci. Appl. 4, e292 (2015).
9M. Lapine,
and Y. S. Kivshar, "Collo-
quium: Nonlinear metamaterials," Rev. Mod. Phys. 86, 1093 --
1123 (2014).
I. V. Shadrivov,
10V. Khardikov, P. Mladyonov, S. Prosvirnin, and V. Tuz, "Elec-
tromagnetic wave diffraction by periodic planar metamaterials
with nonlinear constituents," in Contemporary Optoelectronics:
Materials, Metamaterials and Device Applications, edited by
O. Shulika and I. Sukhoivanov (Springer, the Netherlands, Dor-
drecht, 2016) Chap. 5, pp. 81 -- 98.
11C. L. Holloway, E. F. Kuester, J. A. Gordon, J. O'Hara, J. Booth,
and D. R. Smith, "An overview of the theory and applications
Frequency (GHz)Frequency (GHz)SimulatedMeasured8.459.008.3581012θ = 0.6θ = 0.75θ = 0.67140Exy 0.5X (mm)0-16-321632X (mm)0-16-321632Re(Hz )1.01.00-1.0(c)SimulatedMeasured8101214X (mm)0-16-321632X (mm)0-16-321632(b)TransmissionFrequency (GHz)SimulatedMeasured810121401.0(a)X (mm)0-16-321632X (mm)0-16-321632Y (mm)0-16-321632 Homogeneous enhancement of electric near-field
5
of metasurfaces: The two-dimensional equivalents of metamate-
rials," IEEE Antenn. Propag. M. 54, 10 -- 35 (2012).
12S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S. Kivshar,
and C. R. Simovski, "Metasurfaces: From microwaves to visible,"
Phys. Rep. 634, 1 -- 72 (2016).
13N. I. Zheludev, S. L. Prosvirnin, N. Papasimakis, and V. A.
Fedotov, "Lasing spaser," Nat. Photonics 2, 351 (2008).
14D. J. Bergman and M. I. Stockman, "Surface plasmon amplifi-
cation by stimulated emission of radiation: Quantum generation
of coherent surface plasmons in nanosystems," Phys. Rev. Lett.
90, 027402 (2003).
15M. Premaratne and M. I. Stockman, "Theory and technology of
SPASERs," Adv. Opt. Photon. 9, 79 -- 128 (2017).
16V. A. Fedotov, M. Rose, S. L. Prosvirnin, N. Papasimakis, and
N. I. Zheludev, "Sharp trapped-mode resonances in planar meta-
materials with a broken structural symmetry," Phys. Rev. Lett.
99, 147401 (2007).
17V. V. Khardikov, E. O. Iarko, and S. L. Prosvirnin, "Trapping
of light by metal arrays," J. Opt. 12, 045102 (2010).
18U. Zywietz, A. B. Evlyukhin, C. Reinhardt, and B. N. Chichkov,
"Laser printing of silicon nanoparticles with resonant optical elec-
tric and magnetic responses," Nat. Commun. 5, 3402 (2014).
19A. I. Kuznetsov, A. E. Miroshnichenko, M. L. Brongersma, Y. S.
and B. Luk'yanchuk, "Optically resonant dielectric
Kivshar,
nanostructures," Science 354, aag2472 (2016).
20S. Kruk and Y. Kivshar, "Functional meta-optics and nanopho-
tonics governed by Mie resonances," ACS Photonics 4, 2638 -- 2649
(2017).
21A. B. Evlyukhin, C. Reinhardt, A. Seidel, B. S. Luk'yanchuk,
and B. N. Chichkov, "Optical response features of Si-nanoparticle
arrays," Phys. Rev. B 82, 045404 (2010).
22A. B. Evlyukhin, C. Reinhardt, and B. N. Chichkov, "Multi-
pole light scattering by nonspherical nanoparticles in the discrete
dipole approximation," Phys. Rev. B 84, 235429 (2011).
23A. B. Evlyukhin, S. M. Novikov, U. Zywietz, R. L. Eriksen,
C. Reinhardt, S. I. Bozhevolnyi, and B. N. Chichkov, "Demon-
stration of magnetic dipole resonances of dielectric nanospheres
in the visible region," Nano Lett. 12, 3749 -- 3755 (2012).
24D. Smirnova and Y. S. Kivshar, "Multipolar nonlinear nanopho-
tonics," Optica 3, 1241 -- 1255 (2016).
25W. Tong, C. Gong, X. Liu, S. Yuan, Q. Huang, J. Xia, and
Y. Wang, "Enhanced third harmonic generation in a silicon meta-
surface using trapped mode," Opt. Express 24, 19661 -- 19670
(2016).
26H. Liu, C. Guo, G. Vampa, J. L. Zhang, T. Sarmiento, M. Xiao,
P. H. Bucksbaum, J. Vuckovi´c, S. Fan, and D. A. Reis, "En-
hanced high-harmonic generation from an all-dielectric metasur-
face," Nat. Phys. 14, 1006 -- 1010 (2018).
27K. E. Chong, I. Staude, A. James, J. Dominguez, S. Liu, S. Cam-
pione, G. S. Subramania, T. S. Luk, M. Decker, D. N. Neshev,
I. Brener, and Y. S. Kivshar, "Polarization-independent silicon
metadevices for efficient optical wavefront control," Nano Lett.
15, 5369 -- 5374 (2015).
28V. V. Khardikov, E. O. Iarko, and S. L. Prosvirnin, "A giant
red shift and enhancement of the light confinement in a planar
array of dielectric bars," J. Opt. 14, 035103 (2012).
29J. Zhang, K. F. MacDonald, and N. I. Zheludev, "Near-infrared
trapped mode magnetic resonance in an all-dielectric metamate-
rial," Opt. Express 21, 26721 -- 26728 (2013).
30J. Zhang, W. Liu, Z. Zhu, X. Yuan, and S. Qin, "Strong field en-
hancement and light-matter interactions with all-dielectric meta-
materials based on split bar resonators," Opt. Express 22, 30889 --
30898 (2014).
31V. R. Tuz, V. V. Khardikov, and Y. S. Kivshar, "All-dielectric
resonant metasurfaces with a strong toroidal response," ACS
Photonics 5, 1871 -- 1876 (2018).
32V. R. Tuz, V. V. Khardikov, A. S. Kupriianov, K. L. Domina,
S. Xu, H. Wang, and H.-B. Sun, "High-quality trapped modes in
all-dielectric metamaterials," Opt. Express 26, 2905 -- 2916 (2018).
33P. Yu, A. Kupriianov, V. Dmitriev, and V. Tuz, "All-dielectric
metasurfaces with trapped modes: Group-theoretical descrip-
tion," J. Appl. Phys. 125, 143101 (2019).
34A. Sayanskiy, A. S. Kupriianov, S. Xu, P. Kapitanova,
V. Dmitriev, V. V. Khardikov, and V. R. Tuz, "Controlling high-
Q trapped modes in polarization-insensitive all-dielectric meta-
surfaces," Phys. Rev. B 99, 085306 (2019).
35A. S. Kupriianov, Y. Xu, A. Sayanskiy, V. Dmitriev, Y. S.
Kivshar,
and V. R. Tuz, "Metasurface engineering through
bound states in the continuum," Phys. Rev. Applied 12, 014024
(2019).
36J. van de Groep and A. Polman, "Designing dielectric resonators
on substrates: Combining magnetic and electric resonances,"
Opt. Express 21, 26285 -- 26302 (2013).
37V. E. Babicheva and A. B. Evlyukhin, "Analytical model of reso-
nant electromagnetic dipole-quadrupole coupling in nanoparticle
arrays," Phys. Rev. B 99, 195444 (2019).
38X. Liu, K. Fa, I. V. Shadrivov, and W. J. Padilla, "Experimen-
tal realization of a terahertz all-dielectric metasurface absorber,"
Opt. Express 25, 191 -- 201 (2017).
39P. D. Terekhov, V. E. Babicheva, K. V. Baryshnikova, A. S.
Shalin, A. Karabchevsky, and A. B. Evlyukhin, "Multipole anal-
ysis of dielectric metasurfaces composed of nonspherical nanopar-
ticles and lattice invisibility effect," Phys. Rev. B 99, 045424
(2019).
40R. K. Mongia and P. Bhartia, "Dielectric resonator antennas --
A review and general design relations for resonant frequency and
bandwidth," Int. J. RF Microw. Comput. Aided Eng. 4, 230 -- 247
(1994).
41S. Xu, A. Sayanskiy, A. S. Kupriianov, V. R. Tuz, P. Kapitanova,
H.-B. Sun, W. Han, and Y. S. Kivshar, "Experimental observa-
tion of toroidal dipole modes in all-dielectric metasurfaces," Adv.
Opt. Mater. 7, 1801166 (2019).
|
1910.07821 | 2 | 1910 | 2019-10-21T09:39:51 | Volume Fabrication of Quantum Cascade Lasers on 200 mm-CMOS pilot line | [
"physics.app-ph"
] | The manufacturing cost of quantum cascade lasers is still a major bottleneck for the adoption of this technology for chemical sensing. The integration of Mid-Infrared sources on Si substrate based on CMOS technology paves the way for high-volume low-cost fabrication. Furthermore, the use of Si-based fabrication platform opens the way to the co-integration of QCL Mid-InfraRed sources with SiGe-based waveguides, allowing realization of optical sensors fully integrated on planar substrate. We report here the fabrication and the characterization of DFB-QCL sources using top metal grating approach working at 7.4 microns fully implemented on our 200 mm CMOS pilot line. These QCL featured threshold current density of 2.5 kA/cm2 and a linewidth of 0.16 cm-1 with a high fabrication yield. This approach paves the way toward a Mid-IR spectrometer at the silicon chip level. | physics.app-ph | physics | Volume Fabrication of Quantum
Cascade Lasers on 200 mm-CMOS pilot
line
JG Coutard1, M Brun2, M Fournier1, O Lartigue1, F Fedeli1, G Maisons2, JM Fedeli1, S Nicoletti1, M
Carras2 and L Duraffourg1*
1 Univ. Grenoble Alpes, CEA, LETI, F38054 Grenoble.
² mirSense - Centre d'intégration NanoINNOV, Bâtiment 863, 8 avenue de la Vauve F91120 Palaiseau
* [email protected]
Abstract
The manufacturing cost of quantum cascade lasers is still a major bottleneck for the adoption of
this technology for chemical sensing. The integration of Mid-Infrared sources on Si substrate
based on CMOS technology paves the way for high-volume low-cost fabrication. Furthermore,
the use of Si-based fabrication platform opens the way to the co-integration of QCL Mid-InfraRed
sources with SiGe-based waveguides, allowing realization of optical sensors fully integrated on
planar substrate. We report here the fabrication and the characterization of DFB-QCL sources
using top metal grating approach working at 7.4 µm fully implemented on our 200 mm CMOS
pilot line. These QCL featured threshold current density of 2.5 kA/cm² and a linewidth of 0.16
cm-1 with a high fabrication yield. This approach paves the way toward a Mid-IR spectrometer
at the silicon chip level.
Introduction
These two last decades, hybrid photonics circuits using both silicon based materials and III-V
materials have been successfully developed for data communication and are a complementary
technology to advanced CMOS for high performance computers too. In 2006, Soref suggested
to consider a similar technological approach to make photonics circuits dedicated to the
mid/long wave infrared region [1]. This IR-wavelength range enables to address multiple
applications from free-space data communication [2], optical countermeasures [3], IR imaging
of biological tissues [4] to spectroscopy [5]. So far, the detection of chemical species in solid [6],
liquid [7] or gas mixtures [8] remains the most popular application that drives the technological
development. In fact, the spectral range between 3 μm and 12 μm corresponds to the first
harmonic resonance between rovibrational energy levels of the most chemical species, leading
to absorption cross sections that are order of magnitude stronger than overtone in the near IR
for instance. This is especially true for light molecules in gaseous phase. This unique feature
enables to detect a great number of gasses with extreme sensitivity and selectivity. A limit of
detection lower than 1 part per billion (ppb) and unequivocal identification of chemical species
can typically be reached through MIR absorption spectroscopy techniques [9].
Even if chemical detection using diode lasers has been developed since the mid-1960s, it is with
the advent of unipolar sources based on multiple quantum well stacks that spectroscopic
sensing in the MIR wavelength band has become commercially available. With the recent
progress in Quantum Cascade Laser (QCL) [10], [11] and Interband Cascade Laser (ICL)
technology, compact and reliable MIR light sources are now readily available [12]. In particular,
Distributed FeedBack (DFB) sources allow the selection of specific emission wavelengths to
target the detection of chemicals of interest [13], [14]. With these sources, a novel generation
of sensing tools suitable for real-time in-situ detection of trace element gasses is now available.
With the advent of MIR Si photonics, a novel class of integrated components has been developed
allowing the integration at chip level of the main building blocks required for chemical sensing,
i.e. the source, a photonics integrated circuits (PICs) and the detector [15]. At wavelengths
around 4.8 µm, Spott et al. developed a silicon on SiN waveguide coupled with a bonded QCL
material. DFB lasers with threshold currents as low as 80 mA and threshold current densities
below 1 kA/cm2 emitted more than 200 mW from a polished III-V/Si facet, and operated in
pulsed mode up to 100 °C [16]. They rely on InP type fabrication using III/V-manufacturing lines
on two inch format.
Such technological approaches cannot satisfy volume markets. Today manufacturing price may
be estimated around 1k€ per laser after fabrication, test, sorting and bonding of the die on its
holder. To date, even if III-V technologies are sufficient to meet the needs of niche markets they
cannot tackle the QCL volume manufacturing challenges: implementation of automatic testing
procedures at the wafer level, improvement of the reproducibility of electro-optical features,
simplification of the sorting operations and implementation of a quality-control system. These
elements enable bringing to market at a suitable manufacturing cost, most probably around few
€ with volume largely higher than 100 kunits per year.
So far, these challenges have not been properly addressed and prevent from a large
development and dissemination of these lasers beyond scientific community and niche
applications. In this paper, we present an original approach for the Mid-IR based upon the use
of microelectronic tools to realize the fabrication of QCL on 200 mm Si wafers. Doing so, we are
able to dramatically increase the laser reproducibility, while preserving the same performances
as those reached on InP and to set-up automatic test procedure for reducing or even removing
the sorting of lasers. This paves the way to the manufacturability of low cost devices suitable for
numerous applications from single laser to complete analysis system of chemicals, biologicals at
liquid or gas phases. This work address a fundamental building block for the co-integration of
QCL array with suitable MIR PICs and PhotoAcoustic detectors for making a spectrometer fully
miniaturized at chip level.
This paper reports on the design, fabrication, and characterization of DFB-QCL single sources
and arrays made on 200 mm Si wafers and compares their performances with QCLs made on InP
substrate, whose designs are similar to those used in this work. For the sack of clarity, here
following we will mainly focus on the devices emitting at 7.4 µm, even if similar experimental
results have been obtained at 4.5 µm.
Design
QCLs are commonly based on III-V materials depending on the chosen wavelength range.
Nowadays, most of the QCL emitting in the MIR region (in particular between 3 µm and 11 µm)
are made from a stack of InGaAs / AlInAs layers on InP substrate. These heterostructures are
significantly more efficient than the GaAs / AlGaAs stacks and have the optical power record by
reaching up to few watts [17], [18] in continuous mode and at room temperature. InAs based
QCL are also good candidates in particular for short wavelengths around 3 µm. More recently,
Baranov and co-workers have demonstrated InAs / AlSb QCL emitting at 15 µm wavelength with
a low threshold current density at room temperature [19].
The QCL developed within this work will be used for the chemical analysis in the 4 - 10 µm
wavelength range. The design has been derived from previous stacks developed a couple of
years ago and specifically designed to operate in the 7.4 µm emission wavelength bands. The
laser heterostructure is a standard lattice-matched GaInAs and AlInAs multilayers grown on InP.
Our active region is composed of 25 elementary periods, which are composed of a 3-quantum-
wells and a superlattice resonant injection regions. For more details on the heterostructure, the
reader can refer to [13].
The electroluminescence spectrum of the epitaxy layers, measured at 80 K, showed that the
emission band is actually centered at 1400 cm-1 (7.1 µm) with a full width at half
maximum, 𝐹𝑊𝐻𝑀 = 14.4 𝑚𝑒𝑉.
The multi-layer stack designs were however modified from DFB-QCL sources originally done on
InP to take into account the constraints set by CMOS processing tools. The bottom contacts of
the lasers are in particular no longer on the backside of the chips, but reported aside from the
ridge of the laser introducing doped layers below the active region.
Three lengths of ridge have been investigated (𝐿𝑟𝑖𝑑𝑔𝑒 = 1, 2, and 4 𝑚𝑚). We designed four
ridge widths per length ( 𝑤𝑟𝑖𝑑𝑔𝑒 = 6, 8, 10 and 12 µ𝑚 ) for studying the current density
threshold and the related overlap with the active region that should degrade the output power
at same injected current density level. The nominal ridge geometry is 10 µm-wide, 2 mm-long.
It corresponds to a good compromise between gain and losses and enables to have a single
spatial mode. This width is adjusted to increase the losses of the higher order modes to the
benefit of the first spatial mode. The DFB-QCL requires a wavelength selection mechanism,
which is done with a metal surface grating according to the approach coupling a surface
plasmon-polariton mode and the propagative modes [13].
Technology and fabrication
To fabricate the laser on the 200 mm CMOS pilot line, we directly bonded a 2 inch InP wafer
with the active layers on a 200 mm Si wafer via a thin oxide layer. This process step is a now
mature technology but it requires high-quality defect-free epilayers. This key process step has
been addressed first by the study of the impact of the surface quality of the InP wafers with
epitaxial QC stack from different providers. The bonding is very sensitive to defects and requires
a low roughness and a low bow. If the availability of defect-free epitaxial layers are primordial,
the quality of the InP substrates is also critical. Provided that the surface defects are low enough
on the epitaxial wafers, we achieved direct bonding of more than 90% of surface on oxidized
200 mm Si wafer. The InP substrate was first grinded and then fully removed by chemical etching
of the different sacrificial layers in order to leave a perfect upper surface for further processing.
It is worth to mention that a single Si wafer can receive up to four 2 inch wafers so that the
number of devices fabricate on a single wafer can be increased from 2700 up to 10000 single
lasers per wafer.
This QCL manufacturing is fully compatible with standard microelectronics processes as it
directly exploits the top metal grating technology, which does not require any further process
on the InP wafer before the molecular bonding, ensuring the high volume rate. The
implementation of QCL fabrication process on silicon benefit from the specific DFB technology
developed a few year ago [20], [16] and routinely implemented on InP [13], [21].
The manufacturing of DFB QCL is realized through a grating in the III/V stack exhibiting a periodic
index variation. This approach implies regrowth of III-V material after the grating etching [10].
Today, this regrowth is not possible in a CMOS-like manufacturing of QCL on silicon wafer. To
overcome this issue, the wavelength selection, presented in [16], is achieved on a silicon/nitride
waveguide. This approach requires to couple the QCL cavity to a silicon/oxide waveguide, which
exhibits two main drawbacks. It complicates the design to have an efficient coupling inside the
laser cavity, and reduces the useful wavelength range due to the presence of oxides in the
waveguide. Oxide usually exhibits high losses above 4 µm mid IR wavelength (likewise SiN above
7 µm). The top metal grating approach in double trench does not require regrowth and has
therefore provided the basis for the manufacturing process that we present in this paper.
A typical silicon wafer with the active 2" area including the QCLs is presented in Fig. 1 a. The
main challenge encountered in the fabrication of QCL within a 200 mm CMOS compatible facility
was the development of suitable processes matching the etching of thick layers up to 10 µm and
high resolutions down to 300 nm. For deep UV lithography at 248 nm, the thicknesses of the
resists, the coverage of steep and steps depth are the key points compared to the fabrication
of Near-IR lasers heterogeneously integrated on Si. The partial etch of InP and InGaAs layers
combined with precise stopping level are the challenges for the etching steps. The deposited
material should be optically and chemically compatible with QCL structure. We used SiN for
fabricating the hard mask for InP etching as well as for the deposition of conformal thick layers
on steep edges. Thick conformal gold layers are used for the DFB operation and for getting good
contacts. Finally, the complete process has been performed in our MEMS 200 mm Si platform
enabling the delivery of numerous DFB QCL lasers and QCL arrays. In this work, we manufactured
five wafers, three centered on 7.4 µm emission wavelength and two centered on 4.5 µm
emission wavelength. A 200 mm Si wafer with components (Fig. 1 a), a lithographic 22 x 22 mm²
field and a SEM zoom-in of a single QCL are shown in Fig. 1 b&c. The performance of lasers
depends not only on the quantum well stack but also on the morphology of the laser ridge and
the DFB filter. SEM characterizations were performed to verify the geometry of the ribbon and
the quality of its edges. The expected widths were larger by 1.4 µm. The manufacturing accuracy
is in the order of +/-100nm over the entire center of the 200mm wafer (Fig. 1a)). The roughness
of the etching flanks is less than 100nm (RMS). The ribbon edges are metallized with gold to
increase the losses of higher order optical modes and to have a single mode laser. It should be
noted that the metallization of the back mirror allows a 95% reflectivity and that an anti-
reflective layer (SiN) ensures a 95% transmission on the output facet.
Figure 1: a) 200 mm Si wafer with QCL components (the dotted circle refers to plate 2" InP & the dotted square
refers to a laser field) -- b) Field of (22 mm x 22 mm) of single QCL devices and QCL arrays -- c) SEM picture QCL laser
-- inset: Zoom in of the DFB grating (at the end of the ridge tip)
Electro optic characterizations
As aforementioned, the fabrication of QCLs on 200 mm substrate enables to use large-scale
characterization tools. Thus, 2700 devices (corresponding to single QCLs and QCL arrays) were
measured at wafer-level on an automatic 200/300 mm test prober to determine the emission
threshold levels. To do so, 𝑉(𝐼) and 𝑃(𝐼) characteristics were systematically measured using
synchronized IR detector (VIGO system, Poland) to measure the output light emission. The
measurements were performed in pulsed mode at low duty cycle (3%) and at 17 °C to avoid any
self-heating effect. The pressure exerted on the electrical contact pads as well as the positioning
of the prober tips remain delicate and may modify the series resistance (typically in the order of
1 or less). Calibration of our prober has shown an average contact resistance of a few tens of
m (and a dispersion in the order of 1 m), which is negligible compared to the differential
resistance of our QCLs that are typically around 40. We extracted the repartition of threshold
current densities, 𝐽𝑡ℎ , per wafer, which enables the selection of functional dies and the
evaluation of the fabrication yield. In particular, we computed the average threshold currents
and the standard deviations for each geometry of ridges.
In Figure 2, we have represented the threshold current density as a function of the length and
width of lasers. Each point corresponds to the average value of the current density and its
standard deviation computed over 225 identical lasers. Figure 2a) corresponds to measurements
made on one typical wafer. Figure 2b) shows the averages and standard deviations computed
from the cumulative measurements done on two wafers according to the ridge length. For
reference, we have superimposed the typical (non-statistical) threshold current densities
measured on lasers manufactured on InP. For the shortest lasers (1mm), the threshold current
density is around 4.6kA/cm² and drops for the longest lasers (4mm) down to 2.5 kA/cm². The
threshold current decreases rapidly with the length of the DFB grating, whose reflectivity tends
towards 1 from a value close to or greater than 4mm. The losses of the laser ribbon becomes
predominant from this length. The threshold current densities exhibited by QCL fabricated on
InP substrate remain lower (2.6 kA/cm² & 1.9kA/cm² for 2mm and 4mm respectively). As
reported in the fabrication section, the bottom contacts reported aside from the ridge of the
laser requires doped layers below the active region, which may induce additional losses, which
tend to increase the threshold current.
Figure 2c) shows the average values and standard deviations as a function of width (values from
SEM observations). We reported two sets of measurements corresponding on two different
wafers. The orange circles are for the typical wafer (same as in Fig 2a)) since the blue ones are
for a second wafer from the same fabrication batch. There is no clear trend in the variation of
the threshold current with the width. For the lengths of 1mm and 2mm, it nevertheless seems
that lasers around 10µm wide have the lowest current densities. This observation appears to be
consistent with the initial sizing of the laser, whose theoretical optimal width was set at 10µm
for the stack considered. The values from the two wafers remain very close, in particular for
1mm and 2mm long lasers. The 4mm long lasers of the second wafer show a larger dispersion
in the measured current densities.
/𝐽𝑡ℎ is about 3% except for the 4mm long lasers of the
The relative standard deviation 𝜎𝐽𝑡ℎ
second wafer. This weak dispersion demonstrates the pretty good robustness of our technology.
To go further, it is interesting to compare the disparity of the threshold current values with the
dimensional dispersion related to our lithography/entching processes, used to structure the
laser ribbon and DFB grating. The tolerance of the manufacturing process is better than 100nm
over a ridge width and does not induce the relative dispersion of few percent observed in Fig. 2.
Moreover, the DFB design uses a coupling of the surface plasmon mode at the metal / dielectric
interface with the guided mode [13], [20]. This approach enables a coupling strength of the
grating and losses that are almost constant over a wide range of etching depths [20]. We
estimated that the acceptable tolerance depth is +/-100nm. The variation of our etching process
remains below this limit (typically +/-50nm measured on few samples) what should not
significantly influence the threshold current.
With this systematic electro-optical characterization, we estimated a yield of 98% of functional
lasers per wafer (with similar electro-optical features).
Figure 2: Current density threshold: a) average and standard deviation values with the length and the width (225
dies per geometry) for one typical 200 mm-wafer at 7.4 µm emission -- b) average and standard deviation values
with width (orange circles: wafer 1, blue circles: wafer 2) -- c) average and standard deviation values as a function of
length for the two silicon wafers (orange circles) and for lasers made on InP (green stars)
After these first characterizations, the wafers were diced into discrete components (2700 dies /
wafer). 𝑃(𝐼) and 𝑉(𝐼) were once again measured according to the applied current through the
laser to extract 𝐽𝑡ℎ and the maximum current density 𝐽𝑚𝑎𝑥 . These measurements were
performed at four different operating temperatures (from 15°C to 45°C) at 1.5%-duty cycle.
When the current density reaches 𝐽𝑚𝑎𝑥 , the injection band level becomes higher than the
transition level and the Stark rollover effect appears. In this regime, the optical power drops
with the current. From the curve at different temperature, we estimated the variation of 𝐽𝑡ℎ(𝑇)
and extracted the characteristic temperature 𝑇0 according to (1).
𝐽𝑡ℎ(𝑇) = 𝐽0𝑒
𝑇
𝑇0
(1)
In the figure 3, we present typical 𝑃(𝐼) and 𝑉(𝐼) curves obtained for the nominal geometry
(theoretical ridge width= 8µm, measured width = 10.4 µm, ridge length = 2 mm). As shown in
inset of Fig. 3, 𝑇0 is close to 176 K, which corresponds to quite common value obtained on InP
substrate. This demonstrates that both the bonding process of quantum well stacks on Si wafers
and the ridge /DFB structuration using CMOS compatible tools do not degrade the electro-
optical performances of the QCL. The roll-off currents are quite common too. These
characterizations have however been done at low duty cycle preventing from large self-heating
occurrences and no systematic measurements of the heat diffusion through the Si substrate has
been made so far. Further measurements at the die level are in progress to better characterize
the QCL when working in pulsed mode up to 15%-duty cycle. In this case, the average optical
powers would be between 5 mW and 10 mW.
A differential optical power 𝑑𝑃
𝑑𝐼⁄ of ~83 mW/A can be easily estimated from measurements
presented in Fig. 3 considering the duty cycle of 1.5%. For comparison, typical differential
powers that can be found with commercial lasers on InP are around few 100mW/A in similar
conditions (in pulse mode and at 15°C) [22].
Figure 3: Typical characteristics 𝑃(𝐼), 𝑉(𝐼) of QCL lasers at 7.4µm emission wavelength (ridge width=8 µm, ridge
length=2mm) at four temperatures: blue: 15 °C, green: 25 °C, orange: 35 °C and red: 45 °C -- inset: 𝐽𝑡ℎ(𝑇) vs.
temperature
We also report the normalized typical emission spectra of an array of QCLs working in the 7.4
µm wavelength range for the nominal geometry in Fig. 4a). The output power was measured
operating the lasers at the same condition (applied voltage = 9.9 V, pulsed mode with a 50 ns-
pulse duration and 100 kHz-repetition rate (i.e 0.5%-duty cycle), operating temperature T=21
°C. The spectral resolution of our Fourier-transform infrared spectrometer (Thermofisher
Scientific Nicolet IS-50) is 0.0125 cm-1 (4375 points over the spectral range). Apart from sources
2 and 6, the devices showed single-mode emission and 25 dB side mode suppression ratios over
the working wavelength range (see Fig 4b) insert). The emission wavenumbers extracted from
this measurement are in good agreement with the expected wavelengths defined by the DFB
design, as shown in Fig 4 b). In the figure 4c), we plot the intensity emitted by a QCL at the middle
of the array (i.e. QCL12) with the wavenumber. The maximum power density is 199 µW at 1357.6
cm-1. The optical power integrated over the entire emission spectrum corresponds to 18 µW
(@0.5% duty cycle). This value is coherent with the power measured in Fig. 3 at 10 V at 1.5%-
duty cycle (i.e. 54 µW nearby the threshold). In this operating condition, the typical FWHM is
0.16 cm-1 with a current pulse duration of 50ns and a repetition rate at 100 kHz. The FWHM is
in particular affected by self-heating phenomenon that broadens the linewidth (FWHM of a free
running QCL is between 1MHz-10MHz in CW mode with an efficient thermal drain) [23]. Other
technical noises of the set-up results also in a spectral broadening, and the self-heating
contribution cannot be properly quantified with our current set-up. The emission spectrum even
widens rapidly for pulses exceeding 100ns up to 1 cm-1 (FWHM) for 100ns-pulse duration and
13V of applied voltage.
Figure 4: a) Typical spectral power densities of a QCL array (inset: photography of the QCL array, top view) -- b)
Intensity of a QCL (QCL in mid array; inset: Log representation) -- c) Comparison between expected nominal
wavenumber and experimental ones
Conclusion
In this paper, we report the fabrication of QCL sources on silicon substrate within 200 mm
CMOS/MEMS pilot. To do so, we have successfully transferred the top metal grating process on
an appropriate fabrication process flow that fully respects the design and the process rules of a
standard CMOS manufacturing line. This fabrication run achieves performance at the state of
the art, that are comparable with those of QCL fabricated on InP substrate. The first
characterizations done at wafer level have demonstrated average threshold current densities
between 3 kA/cm² and 2.5 kA/cm² with a relative dispersion around 3%. The measurements
have demonstrated fabrication yields of 98% per wafer having electro-optical features shown in
this paper. The optical power can reach 1 mW at ambient temperature, 1.5% duty cycle. This
value can easily be increased up to ten mW, which is enough to address many applications in
industrial process monitoring or even medical and health fields, by increasing up to reasonable
duty cycle value that is estimated around 15%. At this repetition rate, the self-heating will
probably affect the quantum wells stack and the electro-optical properties will be degraded.
Immediate works on this approach will hence consist of studying the self-heating processes and
propose a technological way to define a thermal drain toward the silicon tank. In next fabrication
batches, Si wafers will also receive up to four 2 inch wafers for multiplying the number of
functional dies. In parallel, development of InAs/AlSb layers grown by molecular beam epitaxy
(MBE) on (100 mm and then 200mm) silicon substrate [24] opens new opportunities that should
be further explored for a full integration into a CMOS line.
The fabrication in a CMOS/MEMS pilot line and wafer-level tests on probe stations should
greatly accelerate the commercialization of QCLs by further decrease the fabrication cost of such
components. Furthermore, the integration on a common technological platform implemented
on Si substrate is crucial to the realization of miniaturized and cost-effective MIR photonic
devices. MIR sources fabricated on Si should penetrate a number of new markets beyond the
gas sensing for professional applications.
With these preliminary results, we added a capital milestone to our works initiated a couple of
years ago on passive MIR photonics circuits and integrated photoacoustic detectors [25]. The
integration of MIR sources on common technological platforms based on IC/MEMS technology
is essential for the realization of miniaturized and cost-effective MIR optical sensors and paving
the way for the implementation of a spectrometer fully integrated on Si chip.
Acknowledgements
This research has been partially supported from the European Union through the H2020 project
RedFinch, N°780240 and the French ADEME project CIDO. The authors thank LETI Silicon
fabrication division and G. Lasfargues and L. Boutafa for their helpful supports with the
component fabrication and characterizations.
References
[1] R. A. Soref, S. J. Emelett, and W. R. Buchwald, "Silicon waveguided components for the long-wave infrared region", J.
Opt. Pure Appl. Opt., vol. 8, 840-848 (2006).
[2] G. Overton in http://www.laserfocusworld.com/articles/print/volume-46/issue-11/features/photonics-applied-free-space-
communications-sophisticated-optical-system-extend-reach-of-free-space-communications.html (2010)
[3] R.J. Grasso, "Defence and security applications of quantum cascade lasers", Proc. SPIE9933, Optical Sensing, Imaging, and
Photon Counting: Nanostructured Devices and Applications 2016, 99330F (2016)
[4] B. Bird and M. J. Baker, "Quantum Cascade Lasers in Biomedical Infrared Imaging", Trends in Biotechnology 33, 557-558
(2015)
[5] Q. Gaimard, M. Triki, T. Nguyen-Ba, L. Cerutti, G. Boissier, R. Teissier, A. Baranov, Y. Rouillard, and A. Vicet,
"Distributed feedback GaSb based laser diodes with buried grating: a new field of single-frequency sources from 2 to 3 µm
for gas sensing applications", Opt. Express 23, 19118-19128 (2015)
[6] P. Geiser, "New Opportunities in Mid-Infrared Emission Control" Sensors 15, 22724-22736 (2015)
[7] See for example: M.R. Alcaraz, A. Schwaighofer, C. Kristament, G. Ramer "External-cavity quantum cascade laser
spectroscopy for mid-IR transmission measurements of proteins in aqueous solution", Anal Chem. 87(13), 6980-6987
(2015)
[8] See for example: R. Ostendorf et al., "Recent Advances and Applications of External Cavity-QCLs towards Hyperspectral
Imaging for Standoff Detection and Real-Time Spectroscopic Sensing of Chemicals", Photonics 3(2), 28 (2016)
[9] See for example: F.K. Tittel, D. Richter and A. Fried, "Mid-Infrared Laser Applications in Spectroscopy", in Sorokina I.T.,
Vodopyanov K.L. (eds): Solid-State Mid-Infrared Laser Sources, Topics in Applied Physics 89, Springer, Berlin,
Heidelberg, 445-516 (2003)
[10] J. Faist, C. Gmachl, F. Capasso, C. Sitori, D. L. Sivco, J. N. Baillargeon, A. L. Hutchinson, and A. Y. Cho, Distributed
feedback quantum cascade lasers, Appl. Phys. Lett. 70, 2670 (1997).
[11] C. Gmachl, F. Capasso, D.-L. Sivco and A.Y. Cho, "Recent Progress in quantum cascade lasers and applications", Rep.
Prog. Phy. 64, 1533-1601 (2001)
[12] M. S. Vitiello, G. Scalari, B. Williams and P. De Natale, Quantum cascade lasers: 20 years of challenges, Optics Express
Vol. 23, No. 4, 5167- 5182 (2015)
[13] M. Carras, M. Garcia, X. Marcadet, O. Parillaud, A. De Rossi, and S. Bansropun, "Top grating index-coupled distributed
feedback quantum cascade lasers", Appl. Phys. Lett. 93, 011109 (2008)
[14] W. Zeller, L. Naehle, P. Fuchs, F. Gerschuetz, L. Hildebrandt, J. Koeth, "DFB Lasers Between 760 nm and 16 µm for
Sensing Applications", Sensors 10, 2492-2510 (2010)
[15] Sergio Nicoletti, Jean-Marc Fédéli, Maryse Fournier, Pierre Labeye, Pierre Barritault, Adrien Marchant, Alain Glière,
Alexandre Teulle, Jean-Guillaume Coutard, and Laurent Duraffourg "Miniaturization of mid-IR sensors on Si: challenges
and perspectives ", Proc. SPIE 10923, Silicon Photonics XIV, 109230H (4 March 2019)
[16] A. Spott, E.J. Stanton, N. Volet, J.D. Peters, J.R Meyer and J. Bowers, "Heterogeneous Integration for Mid-Infrared Silicon",
Photonics. IEEE Journal of Selected Topics in Quantum Electronics 23, 8200810 (2017)
[17] Y. Bai, N. Bandyopadhyay, S. Tsao, E. Selcuk, S. Slivken, and M. Razeghia, "Highly temperature insensitive quantum
cascade lasers", App. Phys. Lett. 97, 251104 (2010)
[18] See for instance: https://mirsense.com/products/powermir/, mirsense (2018)
[19] A.N. Baranov, M. Bahriz, and R. Teissier, Optics Express 24, 16, 18799-18806 (2016)
[20] M Carras, A De Rossi, Photonic modes of metallodielectric periodic waveguides in the midinfrared spectral range, Physical
Review B 74, 235120 (2006)
[21] M. Carras, G. Maisons, B. Simozrag, M. Garcia, O. Parillaud, J. Massies, and X. Marcadet, Room-temperature continuous-
wave metal grating distributed feedback quantum cascade lasers, Appl. Phys. Lett. 96, 161105 (2010)
[22] See for instance: Gangyi Xu, Aizhen Li, Yaoyao Li, Lin Wei, Yonggang Zhang, Chun Lin, and Hua Li, Low threshold
current density distributed feedback quantum cascade lasers with deep top gratings, Appl. Phys. Lett. 89, 161102 (2006)
[23] S. Schilt, L. Tombez, G. Di Domenico and D. Hofstetter, Frequency Noise and Linewidth of Mid-infrared Continuous-
Wave Quantum Cascade Lasers: An Overview, in The Wonders of Nanotechnology: Quantum and Optoelectronic Devices
and Applications (chapter 12), SPIE Press book, M. Razeghi, L. Esaki, and K. von Klitzing, Eds., 261-287 (2013)
[24] H. Nguyen-Van, A.N. Baranov, Z. Loghmari, L. Cerutti, J.-B. Rodriguez, J. Tournet, G. Narcy, G. Boissier, G. Patriarche,
M. Bahriz, E. Tournié and R. Teissier, "Quantum cascade lasers grown on silicon", Scientific Reports 8, 7206 (2018)
[25] J.-M. Fédéli, S. Nicoletti, Mid-Infrared (Mid-IR) Silicon-Based Photonics, PROCEEDINGS OF THE IEEE, Vol. 106, No.
12, 2302- 2312 (2018)
|
1902.10956 | 1 | 1902 | 2019-02-28T09:08:15 | A controllable superconducting electromechanical oscillator with a suspended membrane | [
"physics.app-ph",
"cond-mat.mes-hall",
"physics.optics",
"quant-ph"
] | We fabricate a microscale electromechanical system, in which a suspended superconducting membrane, treated as a mechanical oscillator, capacitively couples to a superconducting microwave resonator. As the microwave driving power increases, nonmonotonic dependence of the resonance frequency of the mechanical oscillator on the driving power has been observed. We also demonstrate the optical switching of the resonance frequency of the mechanical oscillator. Theoretical models for qualitative understanding of our experimental observations are presented. Our experiment may pave the way for the application of a mechanical oscillator with its resonance frequency controlled by the electromagnetic and/or optical fields, such as a microwave-optical interface and a controllable element in a superqubit-mechanical oscillator hybrid system. | physics.app-ph | physics |
A controllable superconducting electromechanical oscillator with a suspended
membrane
Yong-Chao Li,1, ∗ Jiang-shan Tang,2, 3, ∗ Jun-Liang Jiang,1, ∗ Jia-Zheng Pan,1 Xin Dai,1 Xing-Yu Wei,1 Ya-Peng
Lu,1 Sheng Lu,1 Xue-Cou Tu,1 Hua-bing Wang,1 Ke-yu Xia,3, 4, † Guo-Zhu Sun,1, ‡ and Pei-Heng Wu1
1Research Institute of Superconductor Electronics, School of Electronic
Science and Engineering, Nanjing University, Nanjing 210093, China,
2School of Physics, Nanjing University, Nanjing 210093, China
3National Laboratory of Solid State Microstructures,
College of Engineering and Applied Sciences, Nanjing University, Nanjing 210093, China
4Collaborative Innovation Center of Advanced Microstructures, Nanjing 210093, China
We fabricate a microscale electromechanical system, in which a suspended superconducting mem-
brane, treated as a mechanical oscillator, capacitively couples to a superconducting microwave res-
onator. As the microwave driving power increases, nonmonotonic dependence of the resonance
frequency of the mechanical oscillator on the driving power has been observed. We also demon-
strate the optical switching of the resonance frequency of the mechanical oscillator. Theoretical
models for qualitative understanding of our experimental observations are presented. Our experi-
ment may pave the way for the application of a mechanical oscillator with its resonance frequency
controlled by the electromagnetic and/or optical fields, such as a microwave-optical interface and a
controllable element in a superqubit-mechanical oscillator hybrid system.
There are intensive efforts for interfacing the mi-
crowave and optical domains by using various systems in-
cluding the opto-electro-mechanical systems [1 -- 5], solid-
state spins in resonators [6 -- 8], molecules or spins cou-
pled to a nanowaveguide [9, 10], electro-optical material-
based Whispering Gallery mode resonator [11] or cold
atoms [12]. Among these approaches, the opto-electro-
mechanical system is of particular interest because it is
efficient in signal conversion and can be integrated on a
chip.
In the opto-electro-mechanical resonator, the core ele-
ment is the mechanical oscillator, which has been attrac-
tive for its promising applications in both classical and
quantum regimes [13 -- 15]. The features of mechanical
oscillators have been engineered by using various mate-
rials and structures. One of the main characteristics of a
mechanical oscillator is its resonance frequency, which is
determined by its parameters such as the mass, the ge-
ometric shapes and dimensions and the spring constant.
These parameters are almost unchangeable after the sam-
ple is fabricated. In order to control the resonance fre-
quency, an electrostatic field is usually applied to in situ
change the tension of the mechanical oscillator [16 -- 18].
However adding the controlling component such as the
electrostatic electrode to the structure of the mechanical
oscillator greatly improves the complexity and difficulty
of the whole micro-fabrication process.
In this letter we report an experimental implementa-
tion of controlling the resonance frequency of a mechan-
ical oscillator with microwave and optical fields instead
of the electrostatic field. We fabricate a microscale elec-
∗These authors contributed equally to this work.
†Electronic address: [email protected]
‡Electronic address: [email protected]
tromechanical system composed of a microwave coplanar
waveguide (CPW) resonator and a mechanical oscillator,
which is a suspended membrane. Under a microwave
driving to the microwave subsystem, we observe a non-
linear and nonmonotonic frequency shift of the mechan-
ical subsystem due to the radiation pressure force of the
field. Also we use a laser beam pressing the mechanical
part to modify its resonance frequency. Because the mi-
crowave signal passing through the microwave resonator
is modulated by the motion of the mechanical oscillator,
we demonstrate the switching on/off of the transmission
of the sidebands of the microwave signal, which is a step
towards bridging the microwave and optical signals with-
out an optical cavity. The experimental observations are
qualitatively reproduced with our theoretical models.
The optical photograph of our sample is shown in
Fig. 1(a). A λ/4 microwave CPW resonator is capaci-
tively coupled to a transmission line on one end and is
terminated by the a vacuum-gap capacitor (VGC) [19 --
21] (see Fig. 1(b)) on the other end. The sample is made
from aluminum on the high-resistivity silicon substrate
evaporated in a multi-chamber evaporation system with
ultra-high vacuum. The main fabrication procedure in-
cludes three steps: (1) Fabricating the lower electrode
plate of VGC by lift-off process. (2) Preparing the sac-
rificial layer by using diluted S1813 UV photoresist. (3)
Forming the rest of the circuit, including the top elec-
trode plate of VGC, the λ/4 CPW resonator and the
CPW feed lines by wet-etching after depositing another
layer of 270 nm-thick aluminum film. The fabricated de-
vice is equivalent to an electric circuit with a mechanical
oscillator shown in Fig. 1(c). The CPW resonator couples
to the transmission line with characteristic impedance of
50 Ω via a capacitor of Cc = 4.8 fF. The value of VGC
is determined approximately by Cm = ε0A/d, where ε0
is the permittivity of vacuum, A = 45 × 45 µm2 is the
membrane area and d is the vacuum gap between the top
and lower plates of VGC. The gap d can be adjusted in
the process of fabrication and later be tuned with the
microwave and optical field. The displacement of the
fundamental vibration mode of the square membrane is
a spatial function given by
z(x, y) = z0sin(πx/Lx)sin(πy/Ly) ,
(1)
where Lx and Ly are the length and width of the sus-
pended membrane, respectively, and the spatial ranges
of vibration are x ∈ [0, Lx] and y ∈ [0, Ly], and z0 is the
oscillation amplitude. The fundamental-mode frequency
is evaluated as ω11 =
2πP/ρ, where P and ρ are the
tension per unit length and mass per unit area, respec-
tively.
√
The device is located in an Oxford Triton 400 dilu-
tion refrigerator below 20 mK, with magnetic shielding at
both mK and room temperature (see Fig. 2). To suppress
the background noise from the higher-temperature parts,
the microwave field input to the device is heavily attenu-
ated at each stage of the dilution refrigerator and filtered
by low-pass filters with a cutoff frequency of 12 GHz. The
output signal from the microwave resonator first passes
through microwave circulators at cryogenic temperature
to reject the back-action noise from amplifiers. Then, it
is amplified by a low-noise amplifier and microwave am-
plifiers at room temperature, respectively [22]. The low-
noise amplifier uses a high-electron-mobility transistor lo-
cated in the dilution refrigerator. A vector network ana-
lyzer is used to measure the transmission characteristics
of the device. To measure the resonance frequency of the
mechanical oscillator, we use frequency down-conversion
technology as described below: The input microwave sig-
nal is divided into two paths. One of the paths is fed
into the cryostat and the other one is for the local refer-
ence signal of a mixer. This mixer works as a frequency
down-converter for the amplified output microwave sig-
nal which is modulated by the mechanical oscillator. The
down-converted signal is then measured by a spectrum
analyzer. We apply a 1310 nm laser, generated by a
semiconductor diode laser source, to the mechanical os-
cillator to control its resonance frequency, thus switch
on/off the microwave sideband signals arising from the
modulation of the mechanical oscillator.
The transmission characteristic S21 of the CPW res-
onator is measured by a vector network analyzer shown
in Fig. 3(a). Clearly, a dip appears at the resonance fre-
quency of 8.06674 GHz. In our device, the impedance of
the CPW resonator is given by
Zall =
−j
ωCc
+ Z0
Zl + Z0 tanh γs
Z0 + Zl tanh γs
,
(2)
where Zl = 1/jωCm is the VGC impedance, Z0 = 50 Ω is
the characteristic impedance of the CPW resonator and
γ is the propagation constant of microwave field in the
vicinity of the resonance frequency. Under the supercon-
ducting condition and neglecting the loss, the propaga-
tion constant γ approximates to the phase constant. s
2
is the physical length of the λ/4 CPW resonator. Using
Imag(Zall) = 0, s = 4500 µm and γ = 428.8 rad/m, we
obtain Cm = 1.0 pF, in consistence with the estimate
from the fabrication parameters.
After obtaining the resonant frequency of the super-
conducting microwave CPW resonator, we measure the
frequency of the mechanical oscillator using the frequency
down-conversion technology as described previously. As
shown in Fig. 3(b), with an input microwave power of
19 dBm at room temperature, three evenly spaced peaks
appear at 5.76 MHz, 11.52 MHz and 17.28 MHz, re-
spectively. Note that the lowest frequency 5.76 MHz is
the resonance frequency of the mechanical fundamen-
tal mode. The higher frequencies correspond to the
higher-order harmonics of the fundamental mode. These
harmonics are caused by the nonlinear conversion of
the microwave resonator system [23]. The dependence
of the fundamental-mode frequency and its harmonics
on the input microwave power are obtained by scan-
ning the input microwave power. Hereafter we focus on
the fundamental-mode frequency.
It can be seen from
Fig. 3(c) that the fundamental-mode frequency of me-
chanical oscillator increases first from about 1 MHz to
a maximum 6.7 MHz and then decreases rapidly as the
input power increases further.
The observed nonmonotonic frequency shift of the me-
chanical oscillator can be understood by treating the de-
vice as an electromechanical resonator. Its motion is gov-
erned by the Hamiltonian [24]
H = Hm + Hmw + Hin ,
p2
+
1
2
mef f ωm
2z2 ,
Hm =
2meff
Hmw = −∆a†a + i√
Hin = g0za†a ,
2κe(αina† − αin
∗a) ,
(3a)
(3b)
(3c)
(3d)
where ωc is the resonance frequency of the microwave
resonator, meff and ωm are the effective mass and the
resonance frequency of the mechanical oscillator, p and
z are the mechanical momentum and displacement oper-
ators, a and a† are the annihilation and creation opera-
tors of the microwave resonator. a†a is the corresponding
photon-number operator. The input photon flux is de-
termined by the driving amplitude as αin2 = Pin/ωc,
related to the input power Pin. The microwave resonator
is driven by a microwave field with frequency ωin, yield-
ing a detuning ∆ = ωin − ωc. The single-photon coupling
rate of the electromechanical oscillator is g0 = ∂ωc/∂z.
For our device, g0 ≈ −αωc/d, where d ≈ 20 nm. When
considering the effective displacement of the whole area
of the membrane, we have α < 0.4. In a realistic experi-
ment, α can be even smaller. We use a rough estimate of
g0/2π ∼ −80 MHz/nm. The microwave loss of the sys-
tem includes two contributions: one due to the coupling
to the input and output channel yielding a decay rate κe,
the other from the intrinsic loss causing a decay rate κi.
For simplicity, we consider the critical-coupling case, i.e.
κe = κi.
¯a =
2κeαin
,
−i(∆ − g0 ¯z) + κ
¯z = − g0
2¯a2 ,
2
meffωm
(4a)
(4b)
where κ = κi + κe. Due to the displacement, the de-
tuning now becomes ¯∆ = ∆ − g0 ¯z. By calculating the
effective mechanical susceptibility, we obtain the effective
resonance frequency of the mechanical subsystem under
the driving αin as
ωeff =
ω2
0 + ω2
m ,
√
(cid:113)
For a constant driving αin leading to all time deriva-
tives ( a(t), p(t), z(t)) vanishing small, we can find the sta-
ble solutions a(t) = ¯a and z(t) = ¯z that [24]
(5)
(cid:35)
(cid:1)2
.
(cid:34)
where
ω0
2 =
2¯a2
g0
meff
( ¯∆ + ωm)2 +(cid:0) κ
¯∆ + ωm
2
(cid:1)2 +
¯∆ − ωm
( ¯∆ − ωm)2 +(cid:0) κ
2
(6)
In our case, the driving is strong that the frequency
change of the mechanical subsystem can be even larger
than its free oscillation frequency ωm. As a result,
the commonly applied fluctuation approximation,
i.e.
ω0 (cid:28) ωm, breaks. The effective frequency ωeff can be
found by numerically solving the joint Eq. 4 and then
substituting the solution into Eq. 6. In fact, when the
input power increases, the intra-cavity photon number
α2 increases, and so does the displacement ¯z. Thus, the
value of ¯∆ + ωm decreases rapidly. For a certain input
power, the frequency ωeff reaches its maximum. Cross-
ing this point, ¯∆ + ωm approaches zero. As a result, the
effective frequency reduces to a small value. Thus the
effective frequency has a nonmonotonic dependence on
the input power as observed in our experiment, which
is also confirmed by the theoretical result in Fig. 3(d).
It can be clearly seen from Fig. 3(d) that, by treat-
ing the device as a simple electromechanical resonator
and using ωc/2π = 8.06674 GHz, ωm/2π = 0.5 MHz,
κ/2π = 305 kHz, ∆ = 0, g0/2π = −80.7 MHz/nm and
meff = 100 pg, we reproduce qualitatively the nonmono-
tonic frequency shift of the mechanical oscillator as the
driving power increases. Due to the difficulty in the cal-
ibration on the power entering the microwave resonator,
the input power in our model is different from the exper-
imental data which is a nominal value at room tempera-
ture.
Then fixing the input microwave power at 19.2 dBm as
schematically illustrated by the white arrow in Fig. 3(c),
we apply a 1310 nm laser beam from a semiconductor
diode to the suspended membrane. As shown in Fig. 4(a),
we experimentally observe the laser-induced mechanical
frequency shift. When the laser power increases, the
mechanical fundamental-mode frequency deceases from
5.96 MHz to 4.95 MHz and then increase to 6.71 MHz,
wich also displays a nonmonotonic change. The first-
3
and second-order harmonics show similar nonmonotonic
behavior. Due to the difficulty in the calibration of the
optical power, we use the arbitrary unit in the y-axis.
To understand this optically-induced frequency shift,
we consider an optical pressure on the suspended mem-
brane of the electromechanical system. Unlike typical
optomechanical system [24], our model has an external
force from the laser beam and the equation of motion
of the mechanical oscillator with a microwave probe αin
takes the form
z(t) + Γm z(t) + ωm
2z(t) =
FL
mef f
− g0
a(t)2
mef f
,
(7)
where FL = 2ξPL/c is the optical force on the mechanical
oscillator applied by the control laser beam with power
PL, ξ is a coefficient describing how effective the laser
pressure acts on the moving part, and Γm is the mechan-
α2 indi-
ical decay rate. Here we simply set ξ = 1.
cates the intra-cavity photon number. c is the vacuum
light velocity. With this light force considered, the sta-
ble solutions, ¯aL and ¯zL, for the cavity mode a and the
mechanical displacement ¯z are determined by the joint
equations
meffω2
m ¯zL = −g0¯aL2 + FL ,
√
2keαin
¯aL =
−i(∆ − g0 ¯zL) + κ/2
.
(8a)
(8b)
Substituting the solutions into Eq. 5, we obtain the effec-
tive mechanical resonance frequency ωeff under the laser
pressure. The force FL attributed to the laser beam is
equivalent to the radiation pressure from the microwave
CPW resonator. It also strongly modulates the mechan-
ical resonance frequency. In this configuration, the me-
chanical oscillator vibrates initially at a resonance fre-
quency, biased by the microwave input. When the laser
beam is weak, the microwave radiation force is dominant.
As the laser power increases, the light pressure presses
the membrane and ¯zL becomes larger and larger. Thus,
the effective detuning, ¯∆ = ∆ − g0 ¯zL increases. As a
result, the intra-cavity photon number rapidly reduces,
leading to smaller radiation force from the microwave
field. When the laser beam is weak, the microwave radi-
ation force is dominant and the total force on the mem-
brane reduces. Thus the effective resonance frequency
of the mechanical oscillator decreases rapidly. At a cer-
tain laser power, the frequency ωm reaches its minimum.
When the laser power increases further to become dom-
inant over the microwave radiation force, the total force
pressing the membrane increases again. Thus the effec-
tive mechanical resonance frequency increases but at a
smaller rate. For a very strong laser beam, both the
light pressure and the displacement ¯zL increases at con-
stant rates, resulting in the saturation of ωm. Numeri-
cally solving Eq. 8, we find the effective mechanical reso-
nance frequency as a function of a constant laser power.
As shown in Fig. 4(b), our theoretical model reproduces
well the frequency shift of the mechanical oscillator tak-
ing ωc/2π = 8.06674 GHz, ωm/2π = 1.09 MHz, κ/2π =
1.614 MHz, g0/2π = −80.7 MHz/nm, meff = 100 pg and
∆ = 0. ωm and κ change a little in comparison with those
used in understanding the microwave-power-dependent
frequency shift. One of the reasons may be the simul-
taneous injection of the laser and the higher-power mi-
crowave to the mechanical oscillator.
With the laser, we can optically control the sideband
signals of the probe microwave field in our device. The
key idea is to apply a temporally modulated laser beam to
the movable membrane and thus to change the capacitor
Cm. In doing so, we dynamically modulate the resonance
frequency of the microwave CPW resonator and subse-
quently its sideband signals. Experimentally, a rectan-
gular voltage pulse is used to modulate the laser power.
When the rectangular voltage pulse is on a high level, a
laser beam is applied to press the suspended membrane.
This pressure causes a frequency shift to the mechani-
cal resonator, and thus the sideband signals of the res-
onant frequency in the microwave CPW resonator. The
duration of such "on"-state laser beam is controlled by
the duty ration of the rectangular voltage pulse. The
spectrum of down-converted microwave sideband signal
is presented in Fig. 4(c). The fundamental-mode fre-
quency shifts from 5.96 MHz
to 6.71 MHz, when the
laser is tuned from the "off" state to "on", corresponding
to the laser power being 0 and 30 as shown in Fig. 4(a),
respectively. We can switch on/off the microwave side-
band signal output from the microwave CPW resonator
with an extinction ratio of 31 dB.
Figure 4(d) shows the switching temporal distribution
of the transmitted "on"-state sideband signal with the
mechanical fundamental-mode frequency of 6.71 MHz.
This signal is switched from the "off"-state signal, cor-
responding to the sideband field modulated by the lower
mechanical fundamental-mode frequency of 5.96 MHz. A
time counter is triggered to start timing with the synchro-
nization output of the pulse generator, which modulates
the laser beam. The amplified down-converted signal is
4
filtered with a bandpass filter through which only the
signal associated with the mechanical fundamental-mode
frequency of 6.71 MHz can pass. This filtered signal is
then sent to the time counter to stop timing. We per-
form 2000 measurements for each repetition rate of the
laser pulse to obtain the corresponding switching time
statistic distribution. The highest repetition rate is at
least 1 kHz, which is limited by the modulation rate of
our laser. The average irradiation force of laser beam in-
creases as the rectangular pulse repeats faster. If there is
heating effect due to the laser, the mechanical resonant
frequency will change [25]. In this case, the distribution
will also change and even vanish as the repetition rate of
laser pulse increases. However, as seen from Fig. 4(c)(d),
little difference is observed in the distributions and the
line width of the mechanical resonator when the repeti-
tion rate increases from 10 Hz to 1 kHz. Therefore, the
heat effect due to the laser is not observable with the
measuring accuracy in our experiment.
To summarize, we fabricate a superconducting elec-
tromechanical system and modulate its mechanical fre-
quency with both the microwave and optical fields. We
first drive the microwave resonator strongly with the mi-
crowave field and observe large but nonmonotonic me-
chanical frequency shifts. By applying a laser beam to
the mechanical membrane, we further demonstrate the
optical control of the mechanical frequency and thus the
transmission of the sideband signals of the microwave
through the microwave CPW resonator. The developed
device may work as an interface for microwave and op-
tical domains. Also, the scenario demonstrated here has
the potential in controlling a superconducting qubit cou-
pled to a mechanical oscillator [19, 25 -- 28] with lasers
mediated by an optically modulated capacitor.
This work was
partially
supported
NKRDP of China (Grant Nos.
2017YFA0303703),
11474154,61521001,11874212,11574145), PAPD,
Dengfeng Project B of Nanjing University.
NSFC
by
the
2016YFA0301801,
(Grant
Nos.
and
[1] Y.-D. Wang and A. A. Clerk, Phys. Rev. Lett. 108,
[9] S. Das, V. E. Elfving, S. Faez, and A. S. Sørensen, Phys.
153603 (2012).
[2] L. Tian, Phys. Rev. Lett. 108, 153604 (2012).
[3] J. Bochmann, A. Vainsencher, D. D. Awschalom, and
A. N. Cleland, Nat. Phys. 9, 712 (2013).
[4] R. W. Andrews, R. W. Peterson, T. P. Purdy, K. Cicak,
R. W. Simmonds, C. A. Regal, and K. W. Lehnert, Nat.
Phys. 10, 321 (2014).
[5] A. P. Higginbotham, P. S. Burns, M. D. Urmey, R. W. Pe-
terson, N. S. Kampel, B. M. Brubaker, G. Smith, K. W.
Lehnert, and C. A. Regal, Nat. Phys. 14, 1038 (2018).
Rev. Lett. 118, 140501 (2017).
[10] K. Xia, F. Jelezko, and J. Twamley, Phys. Rev. A 97,
052315 (2018).
[11] A. Rueda, F. Sedlmeir, M. C. Collodo, U. Vogl, B. Stiller,
G. Schunk, D. V. Strekalov, C. Marquardt, J. M. Fink,
O. Painter, et al., Optica 3, 597 (2016).
[12] J. Han, T. Vogt, C. Gross, D. Jaksch, M. Kiffner, and
W. Li, Phys. Rev. Lett. 120, 093201 (2018).
[13] K. L. Ekinci and M. L. Roukes, Rev. Sci. Instrum. 76,
061101 (2005).
[6] C. O'Brien, N. Lauk, S. Blum, G. Morigi, and M. Fleis-
[14] M. Poot and H. S. van der Zant, Phys. Rep. 511, 273
chhauer, Phys. Rev. Lett. 113, 063603 (2014).
(2012).
[7] L. A. Williamson, Y.-H. Chen, and J. J. Longdell, Phys.
[15] M. Aspelmeyer, T. J. Kippenberg, and F. Marquardt,
Rev. Lett. 113, 203601 (2014).
Rev. Mod. Phys. 86, 1391 (2014).
[8] K. Xia and J. Twamley, Phys. Rev. A 91, 042307 (2015).
[16] I. Kozinsky, H. W. C. Postma, I. Bargatin, and M. L.
Roukes, Appl. Phys. Lett. 88, 253101 (2006).
[17] M. M. Parmar, P. R. Y. Gangavarapu, and A. K. Naik,
Appl. Phys. Lett. 107, 113108 (2015).
[18] K. Willick, X. Tang, and J. Baugh, Appl. Phys. Lett.
111, 223108 (2017).
[19] K. Cicak, D. Li, J. A. Strong, M. S. Allman, F. Altomare,
A. J. Sirois, J. D. Whittaker, J. D. Teufel, and R. W.
Simmonds, Appl. Phys. Lett. 96, 093502 (2010).
[20] J. D. Teufel1, D. Li, M. S. Allman, K. Cicak, A. J. Sirois,
J. D. Whittaker, and R. W. Simmonds, Nature 471, 204
(2011).
[21] J. Suh, A. J. Weinstein, C. U. Lei, E. E. Wollman, S. K.
Steinke, P. Meystre, A. A. Clerk, and K. C. Schwab,
Science 344, 1262 (2014).
[22] J. Pan, H. Z. Jooya, G. Sun, Y. Fan, P. Wu, D. A. Telnov,
S.-I. Chu, and S. Han, Phys. Rev. B 96, 174518 (2017).
[23] H. Rokhsari, T. J. Kippenberg, T. Carmon, and K. J.
Vahala, Opt. Express 13, 5293 (2005).
[24] in Advances In Atomic, Molecular, and Optical Physics,
edited by P. Berman, E. Arimondo, and C. Lin (Aca-
demic Press, 2010), vol. 58 of Advances In Atomic,
Molecular, and Optical Physics, pp. 207 -- 323.
[25] J. D. Teufel, J. W. Harlow, C. A. Regal, and K. W.
Lehnet, Phys. Rev. Lett. 101, 197203 (2008).
[26] M. D. LaHaye, J. Suh, P. M. Echternach, K. C. Schwab,
and M. L. Roukes, Nature 459, 960 (2009).
[27] J.-M. Pirkkalainen, S. U. Cho, J. Li, G. S. Paraoanu, P. J.
Hakonen, and M. A. Sillanpaa, Nature 494, 211 (2013).
[28] S. J. Bosman, M. F. Gely, V. Singh, D. Bothner,
A. Castellanos-Gomez, and G. A. Steele, Phys. Rev. B
95, 224515 (2017).
5
FIG. 1: (a) Optical photograph of the sample, composed of a
transmission line (TL), a λ/4 microwave coplanar waveguide
(CPW) resonator, and a suspended micromembrane (marked
by a red rectangular), fabricated on the end of the λ/4 CPW
resonator. (b) SEM image of the VGC. The superconducting
membrane is suspended from the substrate, forming a me-
chanical oscillator, and capacitively couples to the microwave
CPW resonator. (c) Equivalent circuit of the device. The me-
chanical oscillator couples to the microwave CPW resonator
via a capacitor Cm. The microwave decay is modeled by a
resistor R. A laser beam (red arrow) can be applied to the
mechanical oscillator to control its resonant frequency.
6
FIG. 2: Measurement setup. Experiments are performed be-
low 20 mK in an Oxford cryogen-free dilution refrigerator.
A microwave signal near the resonance frequency of the mi-
crowave CPW resonator is applied to the electromechanical
device through coaxial lines. Cryogenic attenuators (20 dB or
6 dB) mounted in the input channel and the microwave circu-
lators in the output channel are used to reduce noises. Both
the input and output signals are filtered by low-pass filters
(LPFs). The output signal from the device is first amplified by
a low-noise amplifier (LNA), made of a high-electron-mobility
transistor (HEMT), and microwave amplifiers at room tem-
perature (RT). Then it is demodulated by a mixer and read
out by a spectrum analyzer (SA), from which the knowledge
of the mechanical motion can be found.
7
FIG. 3: (a) Measured S21 of the transmission line. A reso-
nance dip is obtained due to the coupling of the supercon-
ducting microwave resonant circuit to the transmission line.
The resonant frequency and quality factor are 8.06674 GHz
and ∼ 29000, respectively.
(b) Measured down-converted
resonant frequencies with values of 5.76 MHz, 11.52 MHz
and 17.28 MHz, respectively, corresponding to the input mi-
crowave power 19 dBm at room temperature. The quality
factor of the mechanical resonator is about 670.
(c) De-
pendence of mechanical fundamental-mode frequency and its
harmonics on the input microwave power. The white ar-
row indicates the microwave power of 19.2 dBm where the
laser is applied to the mechanical oscillator to control its res-
onant frequency.
(d) Numerically calculated fundamental-
mode frequency of the mechanical oscillator versus the in-
put microwave power. Here we take parameters ωc/2π =
8.06674 GHz, ωm/2π = 0.5 MHz, κ/2π = 305 kHz, ∆ = 0,
g0/2π = −80.7 MHz/nm and meff = 100 pg.
8
FIG. 4: (a) Experimental dependence of the mechanical oscil-
lator's fundamental-mode frequency and its harmonics on the
optical intensity; (b) Theoretical dependence of the mechan-
ical oscillator's fundamental-mode frequency on the optical
intensity agrees qualitatively with the experimental data as
shown in the inset, which is enlarged from (a). ωc/2π =
8.06674 GHz, ωm/2π = 1.09 MHz, κ/2π = 1.614 MHz,
g0/2π = −80.7 MHz/nm, meff = 100 pg and ∆ = 0. Due to
the difficulty in the calibration of the optical power, we use the
arbitrary unit in the y-axis of (a) and (b). (c) Measured me-
chanical resonant frequencies when the laser is on (red) and off
(black). The data when the laser is on are shifted vertically
for clarity.
(d) The switching time statistic distribution of
the down-converted microwave sideband signal with the me-
chanical fundamental-mode frequency of 6.71 MHz switching
from the one with a lower fundamental-mode frequency of
5.96 MHz. The repetition rate of laser pulse is 10 Hz, 100 Hz
and 1 kHz, respectively.
|
1804.08699 | 1 | 1804 | 2018-03-29T09:46:58 | Particles redistribution and structural defects development during ice templating | [
"physics.app-ph",
"physics.flu-dyn"
] | The freezing of colloidal suspensions is encountered in many natural and engineering processes. It can be harnessed through a process known as ice templating, to produce porous materials and composites exhibiting unique functional properties. The phenomenon by itself appears simple: a solidification interface propagates through a colloidal suspension. We are nevertheless still far from a complete understanding and control of the phenomenon. Such lack of control is reflected in the very large scattering of mechanical properties reported for ice-templated ceramics, largely due to the formation of structural defects. Through systematic in situ investigations, we demonstrate here the role of the suspension composition and the role of particle-particle electrostatic interactions on defect formation during ice templating. Flocculation can occur in the intercrystal space, leading to a destabilisation of the solid/liquid interface triggering the growth of crystals perpendicular to the main ice growth direction. This mechanism largely contributes to the formation of structural defects and explains, to a large extent, the scattering of compressive strength values reported in the literature. | physics.app-ph | physics | Particles redistribution and structural
defects development during ice templating
Audrey Lasalle1, Christian Guizard1, Eric Maire2, Jérôme Adrien2,
Sylvain Deville1
1 Laboratoire de Synthèse et Fonctionnalisations des Céramiques, UMR 3080
CNRS/Saint-Gobain, 84306 Cavaillon, France
2 Université de Lyon, INSA-Lyon, MATEIS CNRS UMR5510, F-69621 Villeurbanne,
France
Abstract
The freezing of colloidal suspensions is encountered in many natural and engineering
processes. It can be harnessed through a process known as ice templating, to produce
porous materials and composites exhibiting unique functional properties. The
phenomenon by itself appears simple: a solidification interface propagates through a
colloidal suspension. We are nevertheless still far from a complete understanding and
control of the phenomenon. Such lack of control is reflected in the very large
scattering of mechanical properties reported for ice-templated ceramics, largely due to
the formation of structural defects. Through systematic in situ investigations, we
demonstrate here the role of the suspension composition and the role of particle-
particle electrostatic interactions on defect formation during ice templating.
Flocculation can occur in the intercrystal space, leading to a destabilisation of the
solid/liquid interface triggering the growth of crystals perpendicular to the main ice
growth direction. This mechanism largely contributes to the formation of structural
defects and explains, to a large extent, the scattering of compressive strength values
reported in the literature.
Keywords
Freeze-casting, ceramic material, cellular solids, mechanical properties, defects
10.1016/j.actamat.2012.02.023
1
Introduction
The solidification or freezing of colloidal suspensions is commonly encountered in a
variety of natural processes such as the freezing of soils and the growth of sea ice. It
is also seen in everyday life and engineering situations such as food engineering,
cryobiology, filtration, and water purification. In materials science, the solidification
of colloidal suspension is finding applications in various processes such as the
processing of particle-reinforced alloys and composites, and the processing of porous
materials, usually referred to as ice-templating or freeze-casting. This simple process,
where a colloidal suspension is simply frozen under controlled conditions and then
sublimated before sintering, provides materials with a unique porous architecture,
where the porosity is almost a direct replica of the frozen solvent crystals. When a
colloidal suspension is frozen unidirectionally, an initial transient regime is observed,
corresponding to the initial nucleation and growth of the ice crystals. After this
transient regime, a steady state regime is established, where lamellar ice crystals grow
steadily along the direction imposed by the temperature gradient.
Applications of ice-templating have been demonstrated for bone substitutes [1], drug
delivery [2], acoustic insulation [3], solid oxide fuel cells [4, 5] piezoelectric materials
[6] and ultra-sensitive sensors [7]. The great interest in this versatile technique comes
from the ease of implementation and the large range of porosity in terms of size (0.2
to 100µm), volume fraction (30 to 90%) and morphologies [8]. It was also shown that
the composition of the ice-templated suspensions influences the final microstructures
through the nature of additives [8] or the quantity of dispersant [9]. For any
application, a proper control of the structure is of critical importance. Yet, little is
understood about the dynamics of structure formation mechanisms during freezing.
The characteristics of the colloidal suspension are often critical to the behaviour of
the system during freezing, both in technological and natural occurrences of colloid
freezing, and have rarely been analysed or understood.
A wide range of compressive strength values is reported for ice-templated materials
(figure 1) when tested along their freezing direction. The compressive strength is of
course dependent on composition and is greater for porous titanium or zirconia than
for calcium phosphate, but the data show substantial variation even within identical
systems. Because of the unusual spread in the literature data, we performed a careful
review of the methods and microstructures in the literature. Microstructural
observations revealed that many of the lowest strength samples in the literature had
structural defects oriented perpendicular to the ice growth direction, as shown for
example in figure 2b or figure 8 of reference [10]. This orientation is the worst case
scenario for compressive strength measurements, and we believe that these defects are
the root cause of many anomalously low strength ice templated materials found in
the literature. High strength samples (figure 2a) are systematically free of such
10.1016/j.actamat.2012.02.023
2
defects. The absence of such defects is clearly a necessary but not sufficient condition
to obtain high compressive strength. Excessively large pore size can also lower the
strength.
What we understand so far of the solidification of colloidal suspensions is derived
primarily from analogies with dilute alloy systems or the investigated behaviour of
single particles (or cells) in front of a moving interface. Many geological, biological,
and industrial systems involve concentrated particle systems. In colloidal systems,
unlike alloys, the particle-particle electrostatic interactions can strongly determine
the behaviour of the system. Such aspects have not been taken into account so far.
Owing to their neglect of particle-particle interactions, isolated particle models are
not able to quantify the critical dependence of the final ice crystal morphologies on
the initial colloid concentration – a crucially important operating parameter for
industrial applications.
Through systematic in situ investigations, we demonstrate here the role of the
suspension composition and the role of particle-particle electrostatic interactions on
defect formation during ice templating. We performed in situ observations of crystal
growth and particles redistribution by X-ray radiography and tomography. We show
that particle-particle interaction can have a dramatic influence over the mechanisms
controlling the formation of the structure.
Alumina content
32vol%
32vol%
Nature of additive
D[NH4
D[NH4
+]
+] + PVA
Quantity of additive
0.2-0.4-0.7-1-2wt%
Respectively 0.2wt% +
0.5wt%
0.2-2wt%
0.2-1wt%
D[Na+]
+]
d[NH4
32 vol%
32 vol%
Table 1. Composition of the ice-templated alumina suspensions.
Experimental
We developed a panel of alumina suspensions (table 1), carefully characterized by
measurements of the zeta potential (figure 3a), viscosity (figure 3b), carbon organic
total (COT) and observations of the state of dispersion by Cryo-FEGSEM [12]. These
characterizations show that 0.2-0.4wt% of dispersant is the optimal range to obtain
the strong repulsive interactions between particles necessary for an optimal dispersion
state and stability, a condition traditionally required for ceramic processing routes to
minimize defect formation. Adding more dispersant compresses the thickness of the
diffuse layer around particles and reduces the effective range of the repulsive
10.1016/j.actamat.2012.02.023
3
interactions. This causes the zeta potential values to decrease and the viscosity to
increase (figure 3a).
Alumina powder (Ceralox SPA 0.5, Sasol, Tucson, AZ, USA), D50 = 0.3µm, specific
surface area (SSA)= 8m².g-1, was dispersed in distilled water with an organic
dispersant. Alumina content was held constant at 32vol%. Three sorts of suspensions
were prepared, each containing a different dispersant: (1) an ammonium
polymethacrylate (2) a sodium polymethacrylate and (3) an ammonium polyacrylate
(respectively, Darvan CN, Darvan 7Ns, Darvan 821A Vanderbilt, Norwalk, CT,
USA). These organic dispersants are respectively referred in the text as D[NH4
+],
D[Na+] and d[NH4
+]. With molecular weights of 13 000 and 3 500g.mol-1 respectively,
they are long organic chains of different lengths.
The dispersant concentration in each slurry was 0.2-2wt% with respect to the dried
powder. Dispersant was stirred with distilled water for 30min and then the alumina
powder was added. Alumina suspensions were ball-milled for 40h and de-aired before
being ice-templated. In some cases, 0.5wt% (with respect to the dried powder)
polyvinylic alcohol (PVA) was added as a binder.
The suspensions were first characterized by viscosity measurements performed in a
concentric cylinder system (Bohlin viscosimeter, Malvern, Worcestershire, UK). The
suspension was pre-sheared for 30 s followed by 30 s at rest. Viscosity was measured
at a constant gradient of 50s-1. Then a zetaprobe (Colloidal Dynamics, North
Attleboro, MA, USA) was used to measure the zeta potential.
We adapted a freezing set-up on the beamline ID-19 at the European Synchrotron
Radiation Facility in order to follow the freezing by X-ray radiography and
subsequently observe the frozen microstructure by X-ray tomography. Suspensions
were introduced into a polypropylene mold of 3mm of diameter with a syringe.
Particular attention was paid to not introduce air-bubbles in suspension. The mold
was placed onto a copper finger frozen from the bottom by a liquid nitrogen flux
pumped from a dewar. The cooling rate was controlled by the liquid nitrogen flow
rate and the temperature profile was monitored during the experiment, by a
thermocouple located near the copper finger surface. The cooling rates were in the
range 2-5°C.min-1. When the cooling began, a monochromatic highly coherent X-ray
beam with an energy of 20.5keV was sent through the sample. A CDD camera with
2048 x 2048 sensitive elements was placed 20mm behind the sample. The
advancement of the freezing front was followed by fast acquisition radiography. For
this we used a so called binning mode i.e. a reduction of the number of pixels in the
projection by averaging the measurement of four neighbouring pixels from the CCD
and combining them to create one pixel value. With this binning mode, we achieved
10.1016/j.actamat.2012.02.023
4
a spatial resolution of 2.8µm in the radiographs. For the tomography, the frozen
sample was maintained at a constant low temperature during the scan. An imaging
configuration with high resolution and low acquisition speed was preferred here so the
acquisition was performed without binning, with a spatial resolution of 1.4µm.
We used the sequence of radiographs to qualitatively investigate the local evolution
of the concentration of colloidal particles. Alumina absorbing more than water, the
X-rays absorption and thus the intensity of the signal on the radiograph is inversely
related to the concentration of colloidal particles in suspension. To determine the
change in colloid concentration in each image, we measured the change in intensity
relative to the previous radiograph. Any increase of intensity is thus accompanied
with a decrease of the local concentration of colloids (more beam coming through the
suspension). Conversely, a decrease of the intensity reveals an increase of the local
concentration. These variations can thus be measured qualitatively and dynamically,
although not quantitatively. This image analysis was performed using the ImageJ
software [13].
Results
In situ radiography of the advancing freezing front shows a behaviour drastically
dependent on the quantity of dispersant, D[NH4
+] in this case, introduced in
suspension (figure 4a-c). For suspensions with an optimal dispersion (low dispersant
quantity of 0.2-0.4wt%), the freezing front is composed of disoriented ice crystals
above which a 430µm's thick layer of particles is accumulated. Above this
accumulation, we observe a 20µm's thick particle-depleted region (figure 4a), where
the particle concentration is lower than the average concentration (32vol%). Figure
4b shows that increasing the dispersant concentration to 1wt% causes the
accumulated layer to decrease to 50µm. Ice crystals tend to align along the freezing
direction, but the particle-depleted region remains (figure 4b).
The introduction of dispersant in large excess (2wt%) yields a cellular interface with
no accumulated layer and no particle-depleted region (figure 4c). The corresponding
three-dimensional tomography reconstructions of frozen microstructures show
disoriented ice crystals at low quantity of dispersant (0.2wt%) (figure 4d). By
increasing the dispersant quantity to 0.7wt%, ice crystals are more and more oriented
(figure 4e) but defects are observed perpendicular to the freezing direction, as
illustrated in figure 5 and indicated by black arrows. Since these defects are oriented
perpendicular to the freezing direction, they resemble the defects found in the
literature review, and shown for example in figure 2b. These defects form pores that
traverse the dense ceramic walls, drastically affecting the integrity of the structure. A
typical defect-free lamellar microstructure is obtained at 2wt% of D[NH4
+] (figure 4f).
10.1016/j.actamat.2012.02.023
5
This behaviour is independent of the nature of the dispersants tested here. We obtain
similar microstructures with similar dispersant contents for both D[Na+] and d[NH4
+].
The counter ion (NH4
+ or Na+) and the chain length of the dispersant (3 500 to
13 000g.mol-1) do not seem to affect the orientation of ice crystals. The same change
of ice crystal morphology is observed with d[NH4
+], the dispersant with a shorter
chain length, with disoriented ice crystals at low quantity of dispersant (0.2wt%)
(figure 6a) and a lamellar microstructure at 2wt% (figure 6b). We also obtain a
similar morphological change from disoriented crystals to a lamellar microstructure
with the addition of 0.5wt% of an organic binder (polyvinyl alcohol, PVA). A binder
is usually required in ice-templated materials to ensure the cohesion between the
particles during the freeze-drying stage, in a suspension containing a low quantity of
dispersant. Thus, it is apparent that the formation of such defects is controlled
largely by the quality of the dispersion, rather than other variables like dispersant
counter-ion or dispersant chain length. The same morphological change is also
observed when the cooling rate is increased from 2-5°C.min-1 to 13°C.min-1. Growth
kinetics therefore also play a critical role in the mechanisms controlling the formation
of the microstructure.
Discussion
The development of defects perpendicular to the main ice growth direction during ice
templating must absolutely be avoided. The microstructures obtained in conditions
where such defects develop make such materials useless. We performed a review of
the literature on ice growth, in particular in geophysics, and found that the
perpendicular defects observed here in ice templated materials strongly resembles the
ice lenses observed in geophysics. Ice lenses are ice crystals observed in frozen soils or
during the directional solidification of colloidal suspension, growing perpendicular to
the temperature gradient direction (for example figure 2c). A schematic view is
represented in figure 2d. Ice lenses play a particularly important role in frost heave
[11], by determining the soil's heave rate. The typical ice templated microstructures
obtained in presence and in absence of perpendicular defects (shown in figures 2a and
2b) strongly suggest that such defects are indeed a replica of ice lenses. Since the
porosity is a replica of the ice crystal network obtained after freezing during ice
templating, ice lenses will result in the presence of crack-like pores perpendicular to
the main ice growth direction (determined by the temperature gradient). The
transverse ice crystals growing perpendicular to the main ice growth directions are
therefore ice lenses.
A physically intuitive model of ice lens formation was just proposed [11], whereby the
nucleation and growth of ice lens is controlled by the mechanical properties
10.1016/j.actamat.2012.02.023
6
(cohesion) of the concentrated colloidal suspension between the ice crystals. The
presence of heterogeneities in the concentrated suspension could therefore be a major
factor facilitating the nucleation and growth of ice lenses, locally reducing the
cohesion of the concentrated suspension. These heterogeneities can result from
flocculation and local formation of agglomerates, and conversely particle-depleted
regions. Through a systematic investigation of the panel of suspensions developed, we
identified three main parameters controlling the occurrence of particle depletion and
ice lenses (figure 7): the ionic strength, the viscosity of the suspension and the
velocity of the interface. These three factors are experimentally controlled by the
introduction of the dispersant, the quantity of additives (dispersant, binder) and the
imposed cooling rate, respectively. For each of these parameters, there is a threshold
value below which the behaviour of the system is changing.
The formation of particle-rich and particle-depleted regions requires a driving force
for particle redistribution during freezing. Based on our results, we will first discuss
the possible driving force for the formation of the accumulated particles layer, and
then propose a scenario for the formation of a particle-depleted region and the
nucleation and growth of ice lens, leading to the formation of structural defects.
Origin of the particles in the layer of accumulated particles
Unidirectional ice templating usually results in the growth of lamellar ice crystals,
oriented along the temperature gradient. A layer of accumulated particles above the
freezing front is observed clearly in figures 4a and 4b. This seems to be related to the
loss of an oriented lamellar microstructure. In addition, the presence of this layer
impacts the freezing dynamics by decreasing the front velocity (figure 8). It thus
plays a key role in the formation of the microstructure. The presence of regions with
high and low particle concentrations implies a redistribution of the particles during
freezing. The particles found in the accumulated particles layer can originate either
from below, in the inter-crystals region, in which case a diffusion mechanism is
probably involved, or from the non-frozen suspension above, whereby flocculation
would be the driving force.
The freezing front velocities measured experimentally are in the range 5 to 30µm.s-1.
Calculations of the diffusion velocity (Appendix A1) show that the major part of
particles diffuses slower than the freezing front advances. The diffusion model used
considers an ideal case of spherical particles. The real diffusion coefficient is certainly
lower than the one calculated here. The accumulation of particles above the tips of
the ice crystals by a diffusion mechanism from below seems therefore unlikely. We
propose that the accumulated particles flocculate from the depletion zone above.
10.1016/j.actamat.2012.02.023
7
Flocculation and depletion in the intercrystal regions
To explain the occurrence of a particle-depleted region, we propose a mechanism
based on the flocculation of particles induced by their progressive concentration in
the intercrystal regions, driven by the ice growth. The ice crystals grow in the
direction of the thermal gradient, rejecting and concentrating alumina particles
surrounded by the organic dispersant (figure 9a). The system can remain stable if the
velocity is high enough. In optimal condition of particle dispersion, a monolayer of
dispersant is adsorbed onto particles surface by the carboxylic groups –COO- and the
counter ions NH4
+ or Na+ are attracted by electrostatic interactions but are not
bonded to particles. When the dispersant is in large excess in suspension, a
monolayer is adsorbed onto particles, and the excess not adsorbed remains in
suspension. The solubility of any substance in ice being extremely low (10-6), any
compound or species in solution will be rejected by the growing crystals and thus
concentrated in the intercrystal regions.
When particles are concentrated between growing crystals, the non-adsorbed ionic
species yield a local increase of the ionic strength. The repulsive charge layer of the
alumina particles is thus compressed under ionic strength effect. The repulsive
interactions are diminished and particles can locally agglomerate. Once particle
agglomerates are formed, they may sediment rapidly. Calculations of the
sedimentation velocity of agglomerates (Appendix A2) show that this velocity
(11µm.s-1) is compatible with the typical interface velocity (5 to 30µm.s-1). The
formation of a particle-depleted region by flocculation and sedimentation is thus
compatible with the growth kinetics encountered experimentally.
Depletion and freezing temperature
Flocculation (figures 9c) is responsible for a local increase of the freezing temperature
in the particle-depleted region (figure 9d), due to particle volume fraction decrease
[19], along with a decrease of the cohesive strength of the concentrated colloidal
suspension. This depleted region is suddenly much more favourable to the growth of
an ice lens. By the repetition of the flocculation/depletion/nucleation, more and more
ice lenses are formed.
When the particle-depleted region moves above the tips of the growing ice crystals,
the top of the crystals is in a zone with a higher freezing temperature. They can grow
faster in this zone, resulting in an instability of the advancement of the freezing zone.
Such instability has been previously attributed to the extension of a supercooling
10.1016/j.actamat.2012.02.023
8
zone above the freezing front, favoured by the diffusion of particles. The formation of
the supercooled zone, that we reported previously [17], therefore originates not from
particles diffusion from below but by a particles flocculation from above.
For the suspensions containing a low quantity of dispersant (0.2-0.4wt%), which
corresponds to suspensions with a low viscosity (10-2Pa.s) and high zeta potential (-
75mV) (figure 3), we observe that the particle-depleted region is present above the
freezing front (figure 4 a,b). The displacement of the depleted region from the inter-
crystals space to above the freezing front is favoured by a low viscosity and a small
excess of organic dispersant in suspension.
The accumulation of particles above the freezing front and the associated
disorientation of the ice crystals do not occur directly after the transient regime of
freezing. We can reasonably assume that the depletion occurs first between ice
crystals and then moves above the freezing front because of the progressive
accumulation of particles. The conditions for this accumulation mainly depend on the
dispersant quantity (which controls the ionic strength), binder addition, cooling rate
and viscosity of the suspension.
When the particle-depleted region is above the freezing front, in a region free of
crystals (figure 7e), nucleation and growth of new ice lenses can take place, the
depleted region being in a highly supercooled state (figure 7f). These crystals grow in
the depleted region, thus in a direction perpendicular to the thermal gradient. This
ice lens repels and packs particles in the direction of the thermal gradient and favours
the accumulation of particles above the freezing front. This leads to a repetition of
the same flocculation/depletion/nucleation mechanism and the cellular morphology of
the crystals is lost. This explains the disoriented microstructure observed in figure 4d.
Parameters controlling the flocculation/depletion mechanism
This scenario provides explanations for the effect of the various parameters identified
as controlling the structuring mechanisms. The excess of dispersant or the presence of
an organic binder protects the alumina particles from the agglomeration/flocculation
by creating steric repulsions and it increases the cohesion of the concentrated
colloidal suspension between the crystals. Moreover the presence of ionic species in
large excess in aqueous media favours the depression of the freezing temperature and
the curvature of the freezing interface under Gibbs-Thomson effect. This increases
the supercooling degree in the area of the top of oriented ice crystals and favors
higher freezing front velocity. Then if the cooling rate is increased, the freezing front
10.1016/j.actamat.2012.02.023
9
velocity increases too and there is no time for flocculation to occur. In both cases, an
oriented microstructure is obtained.
This also explains the previous observations of the apparent influence of particle size
on the stability of the interface [17]. The previous set of experiments was performed
with various powders of different granulometry, but the same mass concentration of
dispersant. Decreasing the particle size with a constant dispersant mass concentration
is equivalent to increasing the dispersant mass concentration with a constant particle
size. For small particle sizes, all the dispersant is adsorbed onto the surface of the
particles. The flocculation/depletion mechanism described here can then possibly take
place, and it yields unstable interface propagation. When the particle size increases,
the surface area available for adsorption decreases. The surface of the particles tends
to saturate and the remaining dispersant is found in excess in the suspension,
protecting the particles from the flocculation mechanism when the particle
concentration increases.
Conclusions
The flocculation/depletion mechanism exposed here can facilitate the formation of ice
lenses in ice-templated materials, which are responsible to some extent of the
scattering of compressive strength values. The structures with low strength
correspond to processing conditions yielding a lot of ice lenses, turning into transverse
cracks in the final material. The presence of the defects is of course extremely
deleterious to the integrity and strength of the samples. The structures with high
strength correspond to defect-free structures, obtained with a stable freezing front
yielding no ice lens. Our results show that flocculation can be a viable mechanism to
facilitate ice lens nucleation and growth in such system, and are a first step towards
incorporating particle-particle electrostatic interactions into our understanding and
modelling of the freezing of colloids.
Ice-templating is a very unusual ceramic shaping route. Contrarily to the other
ceramic shaping routes, we show here that it is deleterious to optimize the dispersion
state of ceramic suspension, since these conditions lead to the destabilization
mechanism exposed here. Working with an excess of dispersant or binder is indeed
necessary to ensure the integrity of the obtained structures.
References
[1] Yoon BH, Choi WY, Kim HE, Kim JH, Koh YH Scripta Mater 2008;58:537.
[2] Szepes A, Ulrich J, Farkas Z, Kovács J, Szabó-Révész P Chem Eng Process
2007;46:230.
10.1016/j.actamat.2012.02.023
10
[3] Frank G, Christian E, Dietmar K. Int J Appl Ceram Tech 2011;8:646.
[4] Cable TL, Setlock JA, Farmer SC, Eckel AJ Int J Appl Ceram Tech 2010;8:1.
[5] Cable TL, Sofie SW J Power Sources 2007;174:221.
[6] Lee S-H, Jun S-H, Kim H-E, Koh Y-H J Am Ceram Soc 2007;90:2807.
[7] Zou J, Liu J, Karakoti AS, Kumar A, Joung D, Li Q, Khondaker SI, Seal S, Zhai
L ACS Nano 2010;4:7293.
[8] Munch E, Saiz E, Tomsia AP, Deville S J Am Ceram Soc 2009;92:1534.
[9] Zou J, Zhang Y, Li R. Int J Appl Ceram Tech 2011;8:482.
[10] Deville S Materials 2010;3:1913.
[11] Style R, Peppin S, Cocks A, Wettlaufer J Phys Rev E 2011;84:1.
[12] Lasalle A, Guizard C, Deville S, Rossignol F, Carles P J Am Ceram Soc
2010;94:244.
[13] Abramoff MD, Magalhaes PJ, Ram SJ Biophotonics Inter 2004;11:36.
[14] Glicksman ME Diffusion in solids. Field Theory, Solid-State Principles, and
Applications: Wiley-Interscience, 2000.
[15] Peppin SSL, Worster MG, Wettlaufer JS Proc R Soc London A 2007;463:723.
[16] Liu DM J Mater Sci 2000;35:5503.
[17] Deville S, Maire E, Bernard-Granger G, Lasalle A, Bogner A, Gauthier C, Leloup
J, Guizard C Nat Mater 2009;8:966.
[18] Wegst UGK, Schecter M, Donius AE, Hunger PM Philos Trans R Soc A
2010;368:2099.
[19] Peppin SSL, Wettlaufer JS, Worster MG Phys Rev Lett 2008;100:238301.
[20] Chino Y, Dunand DC Acta Mater 2008;56:105.
[21] Deville S, Saiz E, Nalla RK, Tomsia AP Science 2006;311:515.
[22] Deville S, Saiz E, Tomsia AP Biomaterials 2006;27:5480.
[23] Fu Q, Rahaman MN, Dogan F, Bal SB Biomed Mater 2008;3:025005.
[24] Fu Q, Rahaman MN, Dogan FB, Bal SB J Biomed Mater Res 2008;86B:514.
[25] Han J, Hong C, Zhang X, Du J, Zhang W J Eur Ceram Soc 2010;30:53.
[26] Han J, Hu L, Zhang Y, Zhou Y J Am Ceram Soc 2009;92:2165.
[27] Hong C, Zhang X, Han J, Du J, Han W Scripta Mater 2009;60:563.
[28] Hong C, Zhang X, Han J, Du J, Zhang W Mater Chem Phys 2010;119:359.
[29] Kim JH, Lee J, Yang T, Yoon SY, Kim BK, Park HC Ceram Int 2011;37:2317.
[30] Landi E, Valentini F, Tampieri A Acta Biomater 2008;4:1620.
[31] Lee EJ, Koh YH, Yoon BH, Kim HE, Kim HW Mater Lett 2007;61:2270.
[32] Liu G, Zhang D, Meggs C, Button TW Scripta Mater 2010;62:466.
[33] Liu X, Rahaman MN, Fu Q Acta Biomater 2010;7:406.
[34] Lu H, Qu Z, Zhou Y J Mater Sci Mater Med 2011;9:583.
[35] Macchetta A, Turner IG, Bowen CR Acta Biomater 2009;5:1319.
[36] Soon YM, Shin KH, Koh YH, Lee JH, Kim HE Mater Lett 2009;63:1548.
[37] Suetsugu Y, Hotta Y, Iwasashi M, Sakane M, Kikuchi M, Ikoma T, Higaki T,
Ochiai N, Tanaka J Key Eng Mater 2007;330-332 II:1003.
10.1016/j.actamat.2012.02.023
11
[38] Yang TY, Ji HB, Yoon SY, Kim BK, Park HC Resour Conserv Recy
2010;54:816.
[39] Yook SW, Kim HE, Yoon BH, Soon YM, Koh YH Mater Lett 2009;63:955.
[40] Yook SW, Yoon BH, Kim HE, Koh YH, Kim YS Mater Lett 2008;62:4506.
[41] Yoon BH, Koh YH, Park CS, Kim HE J Am Ceram Soc 2007;90:1744.
[42] Yoon HJ, Kim UC, Kim JH, Koh YH, Choi WY, Kim HE J Am Ceram Soc
2010;3:3.
[43] Zhang Y, Hu L, Han J, Jiang Z Ceram Int 2010;36:617.
[44] Zhao K, Tang YF, Qin YS, Wei JQ Ceram Int 2011;37:635.
[45] Zuo KH, Zeng YP, Jiang D Int J Appl Ceram Tech 2008;5:198.
[46] Zuo KH, Zhang Y, Zeng YP, Jiang D Ceram Int 2011;37:407.
Acknowledgements
Financial support was provided by the National Research Agency (ANR) of France,
project NACRE in the non-thematic BLANC program, reference BLAN07-2_192446.
Beamline access was provided by
the ERSF, under proposal MA997.
Acknowledgements are due, as usual, to local staff of the beamline: Elodie Boller,
Paul Tafforeau and Jean-Paul Valade for the technical and scientific support on ID-
19 at ERSF. We acknowledge Jérôme Leloup, Agnès Bogner, Catherine Gauthier,
Loïc Courtois and Stephen Peppin for their participation to the X-Ray experiments.
Thanks to Stephen Peppin and Robert Style for providing the ice lens picture in
figure 2c.
10.1016/j.actamat.2012.02.023
12
Figure 1. Compressive strength vs. total porosity, data from references [1, 20-46].
The colour code indicates the presence or absence of crack-like defects perpendicular
to the main ice growth direction, as identified from the corresponding published
figures. Such defects result from the ice lenses formation during freezing. The
presence of ice-lenses induced defects is systematically correlated to a low
compressive strength. A low strength can also results from excessively large pore size.
10.1016/j.actamat.2012.02.023
13
Figure 2. Occurrence of ice lenses and corresponding microstructures. Typical
microstructure (A) without and (B) with ice-lenses induced defects, (C) ice lens in
60wt% kaolinite clay suspension, frozen unidirectionally. The concentrated kaolinite
is in white; the ice lenses are the dark horizontal stripes. Arrows indicate the main ice
growth direction. Ice lenses grew approximately perpendicular to the main ice growth
direction. (D) Schematic representation of the typical lamellar ice crystal growing
along the temperature gradient and ice lenses growing perpendicularly. Scale bars (a,
b) 500µm (c) 2 mm.
Figure 3. Zeta potential values [A] and viscosity values measured at 50s-1 [B] for
suspensions dispersed with D[NH4
+] with a concentration ranging
from 0.2 to 2wt%.
+], D[Na+] or d[NH4
10.1016/j.actamat.2012.02.023
14
Figure 4. Crystal growth and particle redistribution during freezing and
corresponding three-dimensional particle concentration. (A-C) In situ radiographs
taken during freezing. The grey level is related to the concentration of particles. The
superimposed profiles indicate the corresponding decrease or increase of particle
concentration as the interface is moving from left to right, between two consecutive
radiographs (see details in the experimental section). The ice-templating suspensions
observed contain 0.2wt% [A], 1wt% [B] and 2wt% of D[NH4
+] [C]. FZ: frozen zone,
CT: crystals tips, APL: accumulated particle layer, PDR: particle-depleted region,
LS: liquid suspension. Scale bar = 200µm. (D-F) 3D reconstructions from
tomography of the particles rich phase regions for samples containing 0.2wt% [D],
0.7wt% [E] and 2wt% of D[NH4
+] [F]. Principal ice growth direction: left to right. The
black arrows show the cracks within the alumina walls, corresponding to the
transverse ice crystals grown in the particle-depleted region. Scale: 140x140x140µm3.
10.1016/j.actamat.2012.02.023
15
Figure 5: Close up view of the ice phase in presence of ice lenses. 3D reconstructions
from tomography. The ice lenses are growing perpendicular to the main ice growth
direction, bridging adjacent lamellar crystals. A few examples are pointed out by the
arrows. Scale: 140x140x140µm3.
Figure 6. 3D reconstructions from tomography of the particle rich phase regions for
samples containing 0.2wt% [A], and 1wt% [B] of d[NH4
+]. Principal ice growth
direction: left to right. Scale: 140x140x140µm3.
10.1016/j.actamat.2012.02.023
16
Figure 7. Occurrence of the accumulated particles layer as a function of ionic
strength of the suspension, showing the effect of the binder addition (0.5wt% PVA)
and
13°C.min-1.
2-5°C.min-1
increase
the
cooling
of
rate
from
to
Figure 8. Influence of the thickness of the accumulated particles on the ice/particles
interface velocity for suspensions dispersed with D[NH4
+]. All
points are on the same trajectory, independently of the nature of the dispersant. The
presence of an accumulated particles layer above the freezing front decreases the
freezing front velocity. This decrease is more important as the thickness of the
accumulated particles layer increases.
+], D[Na+] and d[NH4
10.1016/j.actamat.2012.02.023
17
Figure 9. Schematic representation of the various possible situations and
corresponding freezing temperature profiles. (A,B) Metastable situation with no
flocculation. (C,D) Apparition of a particle-depleted region in the inter-crystals space.
The growth of pre-existing crystals in the particle-depleted region is favoured by the
supercooling situation. (E,F) Apparition of a particle-depleted region above the
growing crystals. In absence of pre-existing crystals, nucleation and growth of an ice
lens in the particle-depleted region can occur, favoured by the supercooling situation.
10.1016/j.actamat.2012.02.023
18
Appendix
A1. Estimation of particles diffusion velocity
We can estimate the theoretical diffusion velocity of particles by using the generic
diffusion equation 1 [14].
This equation is used to determine the distance covered by a particle during a time t
with a coefficient of diffusion D. The diffusion coefficient is calculated from the
equation (2) proposed by Peppin & al [15]. D0 is the Stokes Einstein diffusivity and Z
is the compressibility factor.
(eq. 1)
𝐿=2(𝐷𝑡)0.5
𝐷𝜑 =𝐷0𝐷𝜑
𝐷𝜑 =(1−𝜑)6()
(eq. 2)
The particle size distribution (provided by the supplier) of the powder indicates that
80% of the particles are in the 50-350 nm range. By introducing the data
corresponding to our system in terms of volume fraction, diameter of particles and
maximal packing (obtained with the technique proposed by Liu [16] from viscosity
measurements), we expect the smaller alumina particles (50nm) to diffuse at 7.5.x10-
12m².s-1 and the larger (350 nm) at 1.2x10-12m².s-1. By using equation 1, the particle
diffusion velocity is estimated between 2.2 and 5.5 µm.s-1.
A2. Estimation of sedimentation velocity of agglomerates
We can estimate the sedimentation velocity, according to Wegst et al. [18]. The
sedimentation velocity vp of an agglomerate of radius r is given by:
𝑣=2−
9
²
(eq. 3)
where g is the gravitational acceleration, h the dynamic velocity of the suspension,
rA and rL the density of respectively the agglomerate and the liquid. We can estimate
a typical sedimentation velocity of 11µm.s-1 for 2 µm agglomerates.
10.1016/j.actamat.2012.02.023
19
|
1812.02417 | 1 | 1812 | 2018-12-06T09:30:14 | High efficient solar evaporation by airing multifunctional textile | [
"physics.app-ph"
] | Solar evaporation is important for many applications such as desalination, power generation and industrial drying. Recently, some studies on evaporation reported obtaining high energy efficiency and evaporation rate, which are based on floating evaporation setup (FES) with nanomaterials. Here, we proposed a new cheap and simple setup, named as airing evaporation setup (AES). It shows that the energy efficiency of AES reaches up to 87 % under 1 kW/m2 of solar irradiation, which is 14% higher than that of FES. Meanwhile, the total evaporation rate of AES is about 20% higher than that of FES. The theoretical analysis reveals that the main reason for a better performance of AES is the increasing evaporation area. More interesting, AES could be used for designing portable systems due to its simplicity and flexibility. Furthermore, we show that AES and the corresponding wick material can be used in solar desalination, textile quick-drying and warm-keeping. | physics.app-ph | physics | High efficient solar evaporation by airing multifunctional textile
Guilong Peng1,2, Shichen Deng1,2, Swellam W. Sharshir1,3,4, Dengke Ma1,2, A.E.
Kabeel5, Nuo Yang1,2,*
1 State Key Laboratory of Coal Combustion, Huazhong University of Science and
Technology, Wuhan 430074, China
2 Nano Interface Center for Energy (NICE), School of Energy and Power
Engineering, Huazhong University of Science and Technology, Wuhan 430074, China
3 School of Optical and Electronic Information, Huazhong University of Science
and Technology, Wuhan 430074, China
4 Mechanical Engineering Department, Faculty of Engineering, Kafrelsheikh
University, Kafrelsheikh, Egypt
5 Mechanical Power Engineering Department, Faculty of Engineering, Tanta
University, Tanta, Egypt
*Corresponding email: [email protected]
Abstract
Solar evaporation is important for many applications such as desalination, power
generation and industrial drying. Recently, some studies on evaporation reported
obtaining high energy efficiency and evaporation rate, which are based on floating
evaporation setup (FES) with nanomaterials. Here, we proposed a new cheap and
simple setup, named as airing evaporation setup (AES). It shows that the energy
efficiency of AES reaches up to 87 % under 1 kW/m2 of solar irradiation, which is 14%
higher than that of FES. Meanwhile, the total evaporation rate of AES is about 20%
higher than that of FES. The theoretical analysis reveals that the main reason for a better
performance of AES is the increasing evaporation area. More interesting, AES could be
used for designing portable systems due to its simplicity and flexibility. Furthermore,
we show that AES and the corresponding wick material can be used in solar desalination,
textile quick-drying and warm-keeping.
Keywords: solar energy; evaporation; airing; multifunctional textile; desalination.
1. Introduction
Enhancing the energy efficiency and evaporation rate of solar evaporation
has attracted great attention during the past decades due to its immense utility in
many fields such as water purification or desalination 1-3, power generation 4, phase
change cooling 5, industrial drying 6, 7, and so on. There are many traditional ways to
enhance solar evaporation, such as using dye, solar collectors or black plate to get high
solar energy absorption and high water evaporation rate 8, 9. On the other hand,
nanotechnology shows a better potential in improving solar evaporation. Therefore,
the solar absorber material and system design based on nanotechnology becomes
a hot topic during the past decade 10, 11.
Nanofluid evaporation setup (NES) is one of the strategies for improving
solar evaporation. Researchers found that dispersing metallic nanoparticles into water
can significantly enhance the solar energy absorption and heat transfer, which lead to
significant increase in the evaporation rate 12-16 . However, the high evaporation rate
requires extremely high irradiation density (>100 kW/m2), which is far from practical
application due to its high cost. Later, some researchers turned to carbon-based
nanofluid 17-19, and they found that the energy efficiency can reach up to 70% under
only one sun (1kW/m2 of solar irradiation) 19. However, the stability of nanofluid
remains a challenge, which limits the application of nanofluid in solar evaporation 20.
Another more effective strategy is utilizing floating evaporation setup (FES).
In FES, solar energy is absorbed on the top surface of floatable materials where creates
heat localization21. The floatable materials are floating particles 22, 23, foams 21, 24-26, or
films 27-30, which usually are hydrophilic and have high solar absorptivity. Low thermal
conductivity is also required in order to localize heat at the top surface. The energy
efficiency of evaporation reaches up to more than 80% under one sun, when using the
gold nanoparticles or structures coated foams 31, nano porous membrane 32, carbon
aerogels 26, 33-35, or carbonized biomaterials 36-38. It is reported that the energy efficiency
may reach up to 94% under one sun by using hierarchically nanostructured gels based
on polyvinyl alcohol and polypyrrole 39. Nevertheless, most of the floatable
nanomaterials for high efficiency evaporation are complex and costly, which limits its
application widely.
Herein, we propose a cheap, simple and portable solar evaporation setup,
named airing evaporation setup (AES), which also has high energy efficiency and
evaporation rate. Firstly, to verify the better performance of AES, the energy
efficiency and evaporation rate of AES are measured and compared with FES and NES.
Besides, the dependency of energy efficiency on the width of wings and concentration
of particles are also studied. Secondly, the theoretical analysis is established to
understand the mechanism of the high efficiency of AES. Finally, the potential
applications of AES in solar desalination, textile quick-drying and warm-keeping are
illustrated and discussed.
2. Experimental setup
Figure 1 shows the schematic diagrams of the experimental setup of both
AES and FES. As shown in Fig. 1a, the wick material hangs above the water in AES.
Here, linen is chosen as the wick material due to its excellent capillary action 40, 41. The
shape of linen is circular with two wings which are partly immersed in water (Fig. 1b).
Through the wings, water is absorbed up into the circular linen where is used to absorb
solar energy and evaporate water. To study the effect of wings, wings with different
width (W) were designed, while the diameter (D) is kept as the same (5 cm). The
characteristics of linen are shown in Table 1. To enhance the solar absorption and heat
transfer, carbon black nanoparticles are dispersed on the surface of linen. The size of
nanoparticles is around 40 nm, as measured by transmission electron microscopy (Fig.
2).
To compare with AES, the floating evaporation setup(FES), which uses
floatable materials, is also designed according to the references (Fig. 1c) 21, 42-45. In
FES, all the other surfaces of wick material are insulated with EPE (Expandable
Polyethylene) foam, except the top surface which is the only surface for evaporation.
The schematic diagram of AES shows that there is no insulation foam, that
is simpler than FES. The wick material in AES can be supported by wires or strings
which have negligible volume and cost. The flexible wick material can be easily folded
or rolled, and so, it is very convenient for transportation and storage compared to the
thick foam in FES 11, 46 and unstable nanofluids 47, 48. Therefore, AES is more suitable
for making portable systems, such as portable solar still. Removing the foams also
decreases the cost, because of the foams account for around 1/3 to 1/2 of the total cost
in FES, according to the retail price 45.
Figure 1. (a) The schematic diagram of AES. The wick material hangs 2 cm above the
water with two wings partly immersed in water. (b) The structure of the wick material.
The diameter of the circle for absorbing solar energy is fixed at 5 cm. the width of wings
varies from 0.4 cm to 3 cm. (c) The schematic diagram of FES. (d) The experiment
setup of the evaporation measurement. The surface of the bulk water is covered by a
layer of cling film. The wick material hangs above the water supported by wires.
The measurement setup for AES and FES is shown in Fig. 1d. A solar
simulator (CEL-S500, AM1.5 filter) was used to generate solar beam. The solar
intensity was adjusted to 1 kW/m2 by using a power meter (PM-150-50C). The mass
reduction during the evaporation process was measured by an electric balance
(Sartorius Practum 224), the data were recorded by a computer via a USB cable. The
ambient temperature and humidity during the experiment were controlled at around 24 ℃
and 50%, respectively.
Table 1 The characteristics of the wick material.
material
characteristic
Rate of moisture regain, (%)
linen
Mass density, (g/m2)
Thickness, (mm)
value
12.5
250
1
Figure 2. The TEM image of the carbon black nanoparticles. The scale bar is 200 nm.
3. Results and discussions
The amount of water evaporated under one sun (1 kW/m2 of solar irradiation)
was measured as the mass reduction of systems. Fig. 3a shows the mass reduction
along time of three different solar evaporation setups, traditional pure water evaporation
setup (PWES), NES, FES and AES. It should be noted that the values of natural
evaporation (the evaporation under dark environment) are removed in all cases, which
are around 0.07, 0.07, 0.09, 0.2 kg/(m2·h) for PWES, NES, FES and AES, respectively.
The reduction rate of AES is the highest, which is about 10% , 115% and 200% higher
than FES, NES and PWES, respectively. This means that AES not only simplifies the
structure, but also increases the evaporation rate, compared to the commonly used
effective FES.
To further enhance the evaporation rate and energy efficiency, the structure
of wick material in AES is optimized by changing width of wings. The energy
efficiency as a function of width of wings is shown in Fig. 3b. The energy efficiency,
𝜂, is defined as 21:
𝜂 =
𝛥𝑚∙ℎ𝐿𝑉
𝑄
(1)
where 𝛥𝑚 is the evaporation rate, i.e. the hourly mass reduction per unit area; ℎ𝐿𝑉 is
the total enthalpy of phase change (~2460 kJ/kg), which contains latent heat and
sensible heat; 𝑄 is the nominal direct solar irradiation, which is fixed at 1 kW/m2. In
AES, the efficiency is around 76% when the width of wings is 0.4 cm. Then the
efficiency increases when the width of wings increases, and finally the efficiency
converged to 87% when the width of wings is above 2 cm. In comparison, the efficiency
is only around 28%, 40% and 73% for PWES, NES and FES, respectively.
In addition to width of wings, the mass concentration of particles on wick
material also affect evaporation. The energy efficiency of AES is only 54% when
there are no carbon black nanoparticles, which is 30% lower than that of when the
concentration of particle is 15 g/m2 (Fig. 3d). It should be noticed that the efficiency
increases dramatically when the mass concentration of particles increases from 0 to
1.25g/m2. However, the efficiency only slightly increases for concentration of particles
further increases to 15 g/m2. Therefore, only a very low concentration (>1.25 g/m2) is
required for keeping a high efficiency, which means that the cost of nanoparticles is less
than $ 0.2/m2 according to the retail price of carbon black nanoparticles. The extremely
low cost of nanomaterials is very promising in practical applications.
Besides the above mentioned evaporation rate induced by solar, natural
evaporation rate is also very important in evaporation system. Fig. 3c shows that
AES also gives much higher natural evaporation rate than that of FES, which indicates
that AES can absorb more ambient energy for water evaporation. For the width of wings
at 3 cm, the water evaporated around 0.12 kg/m2 in 30 min, which is 150% higher than
that of FES. The results also demonstrate that the natural evaporation is an increasing
function of the width of wings, due to the increased evaporation area. The total
evaporation rate (natural evaporation + solar induced evaporation) of AES is around 20%
higher than that of FES.
Figure 3 (a) The mass reduction along time for PWES, NES, FES and AES for water
depth at 5 cm. The concentration of carbon black nanoparticles in NES is 0.5wt.%
according to Ref. 16 . The width of wings is 1 cm and the concentration of particles is
10 g/m2. (b) The effect of the width of wings on energy efficiency. The concentration
of particles is 10 g/m2. (c) The effect of the concentration of particles on energy
efficiency. The width of wings is fixed at 1 cm. (d) The mass reduction of natural
evaporation for FES and AES with different width of wings. The concentration of
particles is 10 g/m2.
To understand why AES is better than FES on theory, two setups have been
analyzed based on the classic evaporation theory 49, 50:
𝑚 = 𝜀(𝑃𝑠 − 𝑃𝑣)√
𝑀
2𝜋𝑅𝑇
𝑃𝑠 = 𝑒(25.317−
5144
𝑇
)
(2)
(3)
where 𝑚 is the evaporation rate of water at temperature T, 𝜀 is the evaporation
coefficient, which is a constant. 𝑃𝑠 is the saturated vapor pressure of water at
temperature T. 𝑃𝑣 is the vapor pressure of the ambient, which is 1486 Pa according to
the ambient temperature and humidity. 𝑀 is the relative molecular mass of water , 𝑅
is the universal gas constant (8.314 J·mol-1·K-1).
In AES, the structure of wick material will affect the evaporation. As shown in
Fig. 4a, the evaporation surface can be divided into 2 parts, the outer surface and inner
surface. For the outer surface, the evaporation rate, 𝑚 𝑜𝑢𝑡, is obtained by Eq. (2) and
(3) as:
𝑚 𝑜𝑢𝑡 = 0.587𝜀 (𝑒(25.317−
5144
𝑇
) − 1486) 𝑇−0.5
(4)
However, for the inner surface, the vapor diffusion is blocked by the wings, thus
the vapor from the inner surface is more difficult to diffuse into the ambient compared
to the outer surface. With the width of the wings increases, the diffusion area decreases,
hence the less evaporation rate per unit evaporation area. When wings fully cover the
inner surface, i.e., the total width of wings (2W) equals to the perimeter of the circle
(𝜋𝐷), vapor diffusion will be completely blocked and evaporation rate will be zero,
which is similar to the covered surface in FES (Fig. 4b) . Therefore, the evaporation
rate from the inner surface, 𝑚 𝑖𝑛, is defined as:
𝑚 𝑖𝑛 = 0.587𝜀 (1 −
2𝑊
𝜋𝐷
) (𝑒(25.317−
5144
𝑇
) − 1486) 𝑇−0.5
(5)
Thereby, the overall mass flux of evaporation in AES, 𝑀 , which is the integral
of 𝑚 , is given by:
𝑀 = ∫ 𝑚 𝑜𝑢𝑡 ∆𝐴𝑜𝑢𝑡𝑑𝐴𝑜𝑢𝑡+∫ 𝑚 𝑖𝑛 ∆𝐴𝑖𝑛𝑑𝐴𝑖𝑛
(6)
where 𝐴𝑜𝑢𝑡 and 𝐴𝑖𝑛 are the evaporation area of the outer and inner surface,
respectively.
Herein, we obtained the temperature distribution of water by using COMSOL
based on energy conservation:
𝑄𝑠𝑜𝑙 = 𝑄𝑟𝑎𝑑(𝑇) + 𝑄𝑐𝑜𝑣(𝑇) + 𝑄𝑒𝑣𝑎(𝑇) + 𝑄𝑐𝑜𝑑(𝑇)
(7)
where 𝑄𝑠𝑜𝑙 is the energy of solar irradiation, 𝑄𝑟𝑎𝑑(𝑇), 𝑄𝑐𝑜𝑣(𝑇) and 𝑄𝑐𝑜𝑑(𝑇) are
the radiation, convection and conduction loss, respectively, which are determined by
the ambient and water temperature. 𝑄𝑒𝑣𝑎(𝑇) is the energy carried away by vapor based
on Eq. (4) and Eq. (5). Herein, the slow mass transfer of water in the wick material is
ignored, which has negligible effect to the overall heat transfer process (<2%). 𝑄𝑒𝑣𝑎(𝑇)
is calculated as the following:
𝑄𝑒𝑣𝑎(𝑇) = 𝑚 ℎ𝑓𝑔
(8)
where ℎ𝑓𝑔 is the latent heat of phase change.
The theoretical temperature distribution of water in AES and FES are
shown in Fig. 4a and 4b, respectively. The water temperature in AES reaches up to
307 K which is 8 K lower than that of FES. AES has more evaporation area which
makes the water unable to reach high temperature. The water temperature at the wings
in AES is the same as or even lower than the ambient temperature. This is due to the
water evaporation at the wings carries a lot of heat away. However, the water
temperature at the wings in FES is much higher than the ambient temperature, thus
more heat is transferred to the bulk water compared to AES. Meanwhile, the high
surface temperature in FES also increases the convection and radiation heat loss.
Therefore, the efficiency of FES is lower than AES. The results imply that for high
efficiency evaporation, creating more evaporation surface might be more important
than creating high temperature.
Figure 4 (a) The water temperature in AES for width of wings at 1 cm. The outer (inner)
surface indicates that the vapor diffusion near the surface isn't (is) blocked by the wings.
(b) The temperature distribution of water in FES for the width of wings at 1 cm. The
covered surface means the surface is covered by insulating material. (c) The normalized
evaporation rate of AES. "Total" indicates the total evaporation rate, "in" and "out"
means the evaporation rate from the inner and outer surface, respectively. In simulation,
the water temperature at the end of the wings is 297.15 K, which is the same as the
temperature of the bulk water. The inlet heat flux on the top surface was fixed at 1
kW/m2. (d) Thermal image of HAS system, the hottest region is the wick material.
Based on the water temperature, theoretical evaporation rate can be
obtained by Eq. (4)-(6). Herein the evaporation rate includes both natural evaporation
and solar induced evaporation. Fig. 4c shows that the evaporation rate of the outer
surface in AES increase linearly with the width of wings. On the other side, the
evaporation rate of the inner surface decreases when the width of wings increases due
to the less area for vapor diffusion as discussed in Eq. (5). Therefore, the total
evaporation rate of AES increases first and then converges with the increasing of width
of wings. The theoretical calculations matched well with the experimental data (Fig. 4c
and 4d).
Due to the high efficient evaporation and simple structure of AES as shown
above, AES and the corresponding nano wick material have many potential
applications, such as solar desalination, textile quick-drying and warm-keeping. It is
measured that the evaporation rate of salt water and fresh water is almost the same by
using AES (Fig. 5a), which indicates that AES can be used in desalination efficiently.
Besides, the high water evaporation rate of nanoparticles coated wick material also
shows its potential in quick-drying textile. It is show that, when the moisture content
on wick material is around 80%, the wick material with nanoparticles saves 40% of
drying time compared to that without nanoparticles (Fig. 5b). Moreover, due to the high
solar absorptivity of carbon black nanoparticles 17, 43, the temperature of wick material
with nanoparticles can be 30 °C higher than that of without nanoparticles under 1
kW/m2 of irradiation (Fig. 5c). It means that nanoparticles coated textile could be used
in some solar heating process, such as warm-keeping in winter days. Therefore, this
work gives more possibility in solar energy utilization and multifunctional textile
designing.
Fig. 5 (a) The mass reduction of AES system by using fresh water and salt water (3.5
wt.% of salt) (b) The mass reduction of drying process by hang-airing wet wick material
(5 cm in diameter, 80% in moisture content) under 1 kW/m2 of irradiation. (c) The
temperature of different dry wick material under 1 kW/m2 of irradiation. The insert
figures are the thermal images of different wick material.
4. Conclusion
In conclusion, an airing evaporation setup (AES) has been proposed for high
efficient solar evaporation by using a multifunctional nano wick material. The results
show that the solar induced evaporation rate of AES is obtained around 14%, 120%,
and 210% higher than that of floating evaporation setup (FES), nanofluid evaporation
setup (NES) and traditional pure water evaporation setup (PWES), respectively.
Moreover, it is found that the energy efficiency of AES depends on the width of wings
and the nanoparticles concentration. The energy efficiency increases from 76 % to 87%
when the width changes from 0.4 cm to 3 cm. When using a very low concentration
(1.25g/m2) of carbon black nanoparticles, the efficiency can also be increased as high
as 20% compared to that without nanoparticles. The advantage of a low concentration
makes a low-cost of material.
Besides the solar induced evaporation rate, AES also shows a high natural
evaporation rate due to more evaporation area. Thus, the total evaporation rate of AES
is around 20% higher than that of FES. The theoretical results predict that the increase
of evaporation area of the inner surface in AES is the main reason of high efficient
evaporation. However, it's found that wider wings block vapor diffusion from the inner
surface, which limits the increasing of evaporation rate.
In the end, we show a discussion on the potential application of AES and the
corresponding nano wick material in in designing multifunctional textile and solar
energy utilization.
5. Conflicts of interest
There are no conflicts of interest to declare.
6. Acknowledgement
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 (2016YXZD006). The
authors thank the National Supercomputing Center in Tianjin (NSCC-TJ) and China
Scientific Computing Grid (ScGrid) for providing assistance in computations.
7. References
1.
2.
M. Elimelech and W. A. Phillip, Science, 2011, 333, 712-717.
S. W. Sharshir, G. Peng, L. Wu, F. A. Essa, A. E. Kabeel and N. Yang, Appl.
Energ., 2017, 191, 358-366.
3.
S. W. Sharshir, G. Peng, L. Wu, N. Yang, F. A. Essa, A. H. Elsheikh, S. I. T.
Mohamed and A. E. Kabeel, Appl. Therm. Eng., 2017, 113, 684-693.
N. S. Lewis, Science, 2016, 351, aad1920.
H. J. Cho, D. J. Preston, Y. Zhu and E. N. Wang, Nat. Rev. Mater., 2016, 2,
4.
5.
16092.
6.
G. Pirasteh, R. Saidur, S. M. A. Rahman and N. A. Rahim, Renew. Sust. Energ.
Rev., 2014, 31, 133-148.
7.
M. Kumar, S. K. Sansaniwal and P. Khatak, Renew. Sust. Energ. Rev., 2016, 55,
346-360.
8.
P. Durkaieswaran and K. K. Murugavel, Renew. Sust. Energ. Rev., 2015, 49,
1048-1060.
9.
A. K. Kaviti, A. Yadav and A. Shukla, Renew. Sust. Energ. Rev., 2016, 54, 429-
451.
10.
Z. Deng, J. Zhou, L. Miao, C. Liu, Y. Peng, L. Sun and S. Tanemura, J. Mater.
Chem. A, 2017, 5, 7691-7709.
11. M. Gao, L. Zhu, C. K. Peh and G. W. Ho, Energ. Environ. Sci., 2018, DOI:
10.1039/c8ee01146j, Advance Article.
12.
V. Kotaidis, C. Dahmen, G. von Plessen, F. Springer and A. Plech, J. Chem.
Phys., 2006, 124, 184702.
13.
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 and N. J. Halas,
Proc. Natl. Acad. Sci. U S A, 2013, 110, 11677-11681.
14.
Z. Fang, Y. R. Zhen, O. Neumann, A. Polman, F. J. Garcia de Abajo, P.
Nordlander and N. J. Halas, Nano Lett., 2013, 13, 1736-1742.
15.
N. J. Hogan, A. S. Urban, C. Ayala-Orozco, A. Pimpinelli, P. Nordlander and N.
J. Halas, Nano Lett., 2014, 14, 4640-4645.
16.
J. Wang, J. Zhu, X. Zhang and Y. Chen, Exp. Therm. Fluid. Sci., 2013, 44, 716-
721.
17.
G. Ni, N. Miljkovic, H. Ghasemi, X. Huang, S. V. Boriskina, C.-T. Lin, J. Wang,
Y. Xu, M. M. Rahman, T. Zhang and G. Chen, Nano Energy, 2015, 17, 290-301.
18.
X. Wang, Y. He, G. Cheng, L. Shi, X. Liu and J. Zhu, Energ. Convers. Manage.,
2016, 130, 176-183.
19.
X. Wang, G. Ou, N. Wang and H. Wu, ACS Appl. Mater. Interfaces, 2016, 8,
9194-9199.
20.
D. K. Devendiran and V. A. Amirtham, Renew. Sust. Energ. Rev., 2016, 60, 21-
40.
21.
H. Ghasemi, G. Ni, A. M. Marconnet, J. Loomis, S. Yerci, N. Miljkovic and G.
Chen, Nat. Commun., 2014, 5, 4449.
22.
Y. Zeng, K. Wang, J. Yao and H. Wang, Chem. Eng. Sci., 2014, 116, 704-709.
23.
Y. Zeng, J. Yao, B. A. Horri, K. Wang, Y. Wu, D. Li and H. Wang, Energ.
Environ. Sci., 2011, 4, 4074-4078.
24.
L. Tian, J. Luan, K. K. Liu, Q. Jiang, S. Tadepalli, M. K. Gupta, R. R. Naik and
S. Singamaneni, Nano Lett., 2016, 16, 609-616.
25.
H. Ren, M. Tang, B. Guan, K. Wang, J. Yang, F. Wang, M. Wang, J. Shan, Z.
Chen, D. Wei, H. Peng and Z. Liu, Adv. Mater., 2017, 29, 1702590.
26.
P. Zhang, Q. Liao, T. Zhang, H. Cheng, Y. Huang, C. Yang, C. Li, L. Jiang and
L. Qu, Nano Energy, 2018, 46, 415-422.
27.
L. Zhou, S. Zhuang, C. He, Y. Tan, Z. Wang and J. Zhu, Nano Energy, 2017, 32,
195-200.
28.
L. Zhou, Y. Tan, J. Wang, W. Xu, Y. Yuan, W. Cai, S. Zhu and J. Zhu, Nat.
Photonics, 2016, 10, 393-398.
29.
X. Yang, Y. Yang, L. Fu, M. Zou, Z. Li, A. Cao and Q. Yuan, Adv. Funct. Mater.,
2018, 28, 1704505.
30.
X. Liu, B. Hou, G. Wang, Z. Cui, X. Zhu and X. Wang, J. Mater. Res., 2018, 33,
674-684.
31. M. Chen, Y. Wu, W. Song, Y. Mo, X. Lin, Q. He and B. Guo, Nanoscale, 2018,
10, 6186-6193.
32.
Y. Ito, Y. Tanabe, J. Han, T. Fujita, K. Tanigaki and M. Chen, Adv. Mater., 2015,
27, 4302-4307.
33.
Y. Fu, G. Wang, T. Mei, J. Li, J. Wang and X. Wang, Acs. Sustain. Chem. Eng.,
2017, 5, 4665-4671.
34.
X. Hu, W. Xu, L. Zhou, Y. Tan, Y. Wang, S. Zhu and J. Zhu, Adv. Mater., 2017,
29, 1604031.
35.
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 and L. Hu, ACS Appl. Mater. Interfaces, 2018,
10, 1104-1112.
36.
C. Jia, Y. Li, Z. Yang, G. Chen, Y. Yao, F. Jiang, Y. Kuang, G. Pastel, H. Xie, B.
Yang, S. Das and L. Hu, Joule, 2017, 1, 588-599.
37.
T. Li, H. Liu, X. Zhao, G. Chen, J. Dai, G. Pastel, C. Jia, C. Chen, E. Hitz, D.
Siddhartha, R. Yang and L. Hu, Adv. Funct. Mater., 2018, 28, 1707134.
38.
K. K. Liu, Q. Jiang, S. Tadepalli, R. Raliya, P. Biswas, R. R. Naik and S.
Singamaneni, ACS Appl. Mater. Interfaces, 2017, 9, 7675-7681.
39.
F. Zhao, X. Zhou, Y. Shi, X. Qian, M. Alexander, X. Zhao, S. Mendez, R. Yang,
L. Qu and G. Yu, Nat. Nanotechnol., 2018, 13, 489-495.
40.
R. Caringella, A. Patrucco, M. Simionati, S. Gavignano, A. Montarsolo, R.
Mossotti, M. Zoccola, C. Tonin, R. Fabris and L. Floria, Text. Res. J., 2016, 88,
144-154.
41.
C.-I. Su, J.-X. Fang, X.-H. Chen and W.-Y. Wu, Text. Res. J., 2016, 77, 764-769.
42.
X. Li, W. Xu, M. Tang, L. Zhou, B. Zhu, S. Zhu and J. Zhu, Proc. Natl. Acad.
Sci. U S A, 2016, 113, 13953-13958.
43.
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 and Q. Gan, Global. Chall., 2017, 1,
1600003.
44.
G. Ni, S. H. Zandavi, S. M. Javid, S. V. Boriskina, T. A. Cooper and G. Chen,
Energ. Environ. Sci., 2018, 11, 1510-1519.
45.
G. Peng, H. Ding, S. W. Sharshir, X. Li, H. Liu, D. Ma, L. Wu, J. Zang, H. Liu,
W. Yu, H. Xie and N. Yang, Appl. Therm. Eng., 2018, 143, 1079-1084.
46.
P. Wang, Environ. Sci.- Nano., 2018, 5, 1078-1089.
47. W. Chen, C. Zou, X. Li and L. Li, Int. J. Heat. Mass. Tran., 2017, 107, 264-270.
48.
X. Liu, X. Wang, J. Huang, G. Cheng and Y. He, Appl. Energ., 2018, 220, 302-
312.
49.
R. Mareka and J. Straubb, Int. J. Heat. Mass. Tran., 2001, 44, 39-53.
50.
H. Ding, G. Peng, S. Mo, D. Ma, S. W. Sharshir and N. Yang, Nanoscale, 2017,
9, 19066-19072.
|
1908.07991 | 2 | 1908 | 2019-10-14T06:25:43 | Prospects of designing gold-nanoparticles-based soft terahertz radiation sources and terahertz-to-infrared converters for concealed object detection technology | [
"physics.app-ph",
"cond-mat.mes-hall",
"physics.ins-det"
] | The two-phonon scheme of generation of terahertz (THz) photons by gold nanobars (GNBs) is considered. It is shown that in GNBs, by choosing their sizes, it is possible to provide conditions for converting the energy of longitudinal phonons with THz frequencies into the energy of THz photons. The prospects of designing GNBs-based soft THz radiation sources (frequencies: 0.14; 0.24; 0.41 and 0.70 THz) with a large flow cross-section (diameter ~40 cm) intended for detection of hidden objects under clothing to ensure security in public places (airports, railway stations, stadiums, etc.) are assessed. The choice of the above frequencies is a compromise between the requirements of low absorption of THz radiation by water vapor in air, good penetration through the fabric of clothing, favoring a sufficient resolution of the imaging system, and an abundance of corresponding longitudinal phonons, capable of exciting Fermi electrons in GNBs. Estimates of the characteristics of the terahertz-to-infrared converter based on gold nanospheres (GNSs), which could work in tandem with these sources of THz radiation -- as a means of visualization of hidden objects -- are also given. | physics.app-ph | physics | Prospects of designing gold-nanoparticles-based soft terahertz radiation
sources and terahertz-to-infrared converters for concealed object detection
technology
K. A. Moldosanova)
Kyrgyz-Russian Slavic University, 44 Kiyevskaya St., Bishkek 720000, Kyrgyzstan
A. V. Postnikov
LCP-A2MC, University of Lorraine, 1 Bd Arago, F-57078 Metz, France
V. M. Lelevkin and N. J. Kairyev
Kyrgyz-Russian Slavic University, 44 Kiyevskaya St., Bishkek 720000, Kyrgyzstan
The two-phonon scheme of generation of terahertz (THz) photons by gold nanobars (GNBs) is considered. It
is shown that in GNBs, by choosing their sizes, it is possible to provide conditions for converting the energy
of longitudinal phonons with THz frequencies into the energy of THz photons. The prospects of designing
GNBs-based soft THz radiation sources (frequencies: 0.14; 0.24; 0.41 and 0.70 THz) with a large flow cross-
section (diameter ∼ 40 cm) intended for detection of hidden objects under clothing to ensure security in
public places (airports, railway stations, stadiums, etc.) are assessed. The choice of the above frequencies
is a compromise between the requirements of low absorption of THz radiation by water vapor in air, good
penetration through the fabric of clothing, favoring a sufficient resolution of the imaging system, and an
abundance of corresponding longitudinal phonons, capable of exciting Fermi electrons in GNBs. Estimates
of the characteristics of the terahertz-to-infrared converter based on gold nanospheres (GNSs), which could
work in tandem with these sources of THz radiation -- as a means of visualization of hidden objects -- are also
given.
Keywords: gold nanobar, infrared camera, longitudinal phonon, microwave, on body concealed weapon de-
tection, soft terahertz radiation source, terahertz-to-infrared converter
I.
INTRODUCTION
From the very beginning of practical exploring the ter-
ahertz (THz) range, attempts have been made to apply
it for remote detection of hidden threats on the human
body -- see, e.g., Ref. 1 -- 14 and references therein. Fig. 1
illustrates the geometry of active remote weapon detec-
tion (gun, bomb, knife, etc.) hidden under the clothes.
In this method of detecting threats on the subject's body,
the latter is irradiated by an external source of THz ra-
diation, and radiation reflected from hidden objects is
recorded. Previous works allowed to identify the main
problems complicating a successful application of THz
radiation for the remote detection of hidden objects,
namely, an attenuation of THz radiation in fabrics of
clothes, and its absorption by water vapor in the air. It
was found that the highest transmittance of clothing tis-
sue is in the low-frequency part (∼ 0.1− 0.5 THz) of the
THz range. In this range, the attenuation of THz radi-
ation in air water vapor is relatively low, but monotoni-
cally increases with increasing frequency and is saturated
with damping peaks due to the absorption by oscillating
hydrogen atoms in the water molecule. The latter limits
the detection range; however, in most hidden object de-
tection systems, the detection range is anyway relatively
short, from 1 to 20 m.13 As for the application for short-
range operation (< 4 m), such as airport security check,
a)Electronic mail: [email protected]
FIG. 1. Typical geometry of the remote threat detection. 1:
THz radiation source, 2: THz radiation detector.
the lower operating frequency (∼ 0.1 THz) can still pro-
vide diffraction-limited image resolution of ∼ 2 cm, which
should suffice for many threat scenarios.
The development of active methods for detecting hid-
den objects revealed a shortage of sources of soft THz
radiation with a wide radiation flux. Typically, the THz
radiation source would be required to create a spot on the
human body with a diameter of ∼ 40 cm. Accordingly,
the imaging system aperture A must be at least this size.
As a rule, THz radars with an aperture of ∼ 50− 100 cm
are used,6 -- 11 which is a compromise between the resolu-
tion ∆x of the imaging system, the range of observation
of the subject d and the THz radiation wavelength λ, in
9
1
0
2
t
c
O
4
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
2
v
1
9
9
7
0
.
8
0
9
1
:
v
i
X
r
a
TABLE I. Some numerical parameters for photons at four frequencies selected for our analysis
2
Frequency Wavelength Photon energy Photon momentum Objective resolution∗ Penetration
depth† (µm)
ν (THz)
0.14
0.24
0.41
0.70
pph (10−26 g cm s−1) ∆x≈ λd/A (cm)
(mm)
2.1
1.25
0.73
0.43
∗for d∼ 10 m, A∼ 0.5 m;
3.1
5.3
9.1
15.5
†skin layer thickness in bulk Au
4.2
2.5
∼ 1.5
∼ 0.9
hν (meV)
0.58
0.99
1.70
2.90
0.20
0.15
0.12
0.09
accordance with the formula for a diffraction-limited res-
olution: ∆ x ≈ λ d/A. Naturally, the wavelength λ is
chosen so that it is not at the same time at the peak of
attenuation in the atmospheric air and belongs to the re-
gion of acceptable transmission of clothing fabrics, that
is, the compromise wavelengths correspond to frequen-
cies ∼ 0.1 − 0.5 THz. The resolution accessible is within
the range ∆ x ∼ 0.5 − 2 cm.
In the present paper we suggest the novel soft terahertz
radiation sources based on gold nanobars (GNBs) being
irradiated with microwaves. This approach assesses the
feasibility of the idea to convert the energy of longitu-
dinal phonons into electromagnetic energy of THz pho-
tons. The approach seems to be reinforced by recent
reports15 -- 17 on an enhancement beyond the blackbody
limit of radiative heat transfer in nanometric-scale ob-
jects, as well as by our own estimates of the surface power
density of spontaneous THz radiation by gold nanopar-
ticles (see Appendix A).
To enhance the power of THz radiation, GNBs provide
the conditions (see Appendix B) to increase the number
of longitudinal phonons by heating GNBs with microwave
radiation. This approach would enable a wide beam of
THz radiation to be issued from a large number of GNBs
distributed on a substrate or within a matrix of suffi-
FIG. 2. Electromagnetic waves attenuation by an air layer of
1 m thickness, in the frequency range 0.01 through 1 THz, re-
produced from Figure 2.15 of Ref. 13. The frequencies chosen
for our analysis are 0.14; 0.24; 0.41 and 0.70 THz.
ciently large size.
Taking into account the data of Ref. 11 -- 13 on the at-
tenuation of THz radiation in the air, we have chosen the
following frequencies to detect hidden objects: 0.14; 0.24;
0.41 and 0.70 THz (the corresponding wavelengths, ener-
gies and momenta of THz photons are given in Table I).
These frequencies do not fall into the attenuation peaks of
THz radiation in air -- see Fig. 2, borrowed from Ref. 13.
In addition, at these frequencies the undesirable tissue
absorption is sufficiently low, whereas the desired trans-
mission, on the contrary, is high.11,18 Table I includes
also the objective's resolution ∆x, achievable at chosen
frequencies, with the distance to the subject d∼ 10 m
and the aperture A∼ 0.5 m.
The paper is organized as follows. Sec. II explains gen-
eration of THz radiation by gold nanobars, Sec. III sug-
gests the practical design of the THz source. In Sec. IV,
the parameters of the GNSs-based THz-to-IR converter,
which could be used to visualize hidden objects at the
security checkpoints, are briefly discussed. The paper is
concluded by Sec. V and contains two Appendices, A and
B.
II. SUGGESTED DESIGN OF SOFT TERAHERTZ
RADIATION SOURCES WITH GOLD NANOBARS AS
ACTIVE ELEMENTS
In the proposed approach, in a GNB, the Fermi elec-
tron is excited by absorbing a longitudinal phonon, and
relaxes by releasing a softer longitudinal phonon. The
energy difference is brought away by the electron and
emitted as a soft THz photon as the electron scatters
at the GNB boundary. The physical picture to be con-
sidered is somewhat similar to that discussed earlier in
Ref. 19 -- 21, with the only difference that in these works
the phonon energy was completely converted into the en-
ergy of THz phonon. As emphasized in the works cited,
the role of gold nanoobjects is to provide free electrons
and longitudinal phonons, their respective energy levels
being appreciably discretized due to spatial confinement.
The microwave radiation serves as a source of energy for
maintaining the "phonon bath" that heats the nanoob-
jects.
3
FIG. 3. Energy and momentum relations concerning absorption and subsequent emission of a longitudinal phonon by a Fermi
electron. (a) Phonon dispersion in bulk gold along X − Γ (red line), discretized as a sequence of spatial confinement (assuming
Nx,y = 13) in a "thin" GNB (vertical blue lines for wavevectors, horizontal blue lines for frequencies). (b) Energy of a Fermi
electron as function of (x, y) momenta prior to and after the absorption / emission of longitudinal phonons with energies E2 /
E1. pF is the Fermi momentum, nq the difference of phonon momenta, s the momentum of the electron after the absorption
/ emission of phonons. (c) The orientation of the three momenta pF, nq, s, making clear the meaning of the angle γ. See text
for detail.
A. General considerations
To streamline the argumentation, it seems useful to
consider a simplified picture in which the wavevector
of the primary (exciting) phonon, q2, and the phonon
released in the course of the electron relaxation, q1,
are collinear and directed along a short -- just several
lattice parameters of gold aAu -- GNB dimension, say
LX = Nx aAu or LY = Ny aAu (see Fig. 3), so that
the spatial confinement results in discretisation of (x, y)
phonon momenta. Now, we assume that the phonon
dispersion of bulk gold still approximately holds at the
nanoscale (this seems plausible since the vibration spec-
tra of bulk and nanoparticle gold are very close -- see
Fig. 2 of Ref. 19 and the original papers22 -- 24), and we
want to deduce some conclusions from matching the con-
ditions of electron and phonons momenta and energy con-
servation. Such an analysis has been done in Ref. 25, al-
beit with compact GNPs (not elongated GNBs) in mind.
The discussion around Fig. 2 in Ref. 25 parametrised the
experimental Γ− X dispersion relation for bulk gold and
elaborated on the matching conditions between the en-
ergy of absorbed/released phonons, ω(q1)−ω(q2), and
the quantisation step on the electron excitation.
In the following, these energy / momentum relations
are expressed in a more straightforward way, assuming
Nx = Ny = 13 for the reasons explained in Appendix
B. Then LX = LY = 5.3 nm, and the momenta of lon-
gitudinal phonons are quantized with the step h/LX ≈
1.25·10−20 g cm s−1. Fig. 3 summarizes relations between
momenta and energies of longitudinal phonons and of
Fermi electrons, in the course of absorption / emission of
phonons. Fig. 3a depicts the Au dispersion branch along
q1
2
3
4
5
3
4
5
4
5
q2
1
2
3
4
1
2
3
1
2
1
X−Γ, divided into h/LX steps. Different combinations of
absorption and emission phonon wavevectors are shown
in Table II, along with the corresponding energy differ-
ences, expressed also in units of frequency. Of primary
interest are the phonons whose frequencies fall into the
full width at half maximum (FWHM) of the longitudi-
nal phononic density of modes (shown, e.g., in Fig. 2 of
TABLE II. Energy differences on absorption / emission of
longitudinal phonons in gold, consistently with the dispersion
relation as shown in Fig. 3. See text for details.
Wavevectors in units of (X→Γ)
13
E2−E1 (meV) ν (THz) γ
q2−q1 = 1.25×10−20 g cm s−1
0.59
0.60
0.71
0.99
1.19
1.31
1.70
1.90
2.30
q2−q1 = 2.50×10−20 g cm s−1
q2−q1 = 3.75×10−20 g cm s−1
q2−q1 = 5.00×10−20 g cm s−1
5
2.90
0.14 87◦
0.15 87◦
0.17 87◦
0.24 87◦
0.29 84◦
0.32 84◦
0.41 84◦
0.46 82◦
0.56 82◦
0.70 79◦
Ref. 19) between ∼ 3.9 THz (16.2 meV) and ∼ 4.6 THz
(19.0 meV), the maximum of the peak in the density of
modes being at ≈ 4.2 THz (17.4 meV). The wavevectors
of such phonons are close to the Brillouin zone boundary,
e.g., separated by just few quantisation steps from the X
point. The slope of the frequency/wavevector curve, i.e.
the longitudinal speed of sound in this frequency region
which will be used in the following to estimate the energy
L ≈ 1·105 cm s−1, i.e., consid-
matching conditions, is v∗
erably reduced in comparison with the nominal (zone-
center) speed of sound in bulk gold. The combinations
of absorption / emission phonon energies give rise to the
energy gain to be transferred to the THz photon (see
discussion below); the frequency values appearing in Ta-
ble II cover the range of "useful" frequencies from 0.14
to 0.70 THz discussed in Section I.
Fig. 3b depicts the electron energy levels of the GNP,
indicating explicitly a promotion of the Fermi electron
(with momentum pF and energy EF), following an ab-
sorption and emission of two phonons with wavevectors
close to the Brillouin zone boundary, to an excited state
(with momentum s and energy EF+E2−E1). The relax-
ation from this excited state will bring about an emis-
sion of a THz photon, as argued below. For the moment,
we can make two remarks, to elaborate on the electron-
phonon part.
The first remark is that the energy of the electron ex-
cited state EF+E2−E1 must, in principle, match one of the
discrete levels quantized due to confinement within the
GNB, in the spirit of the Kubo formula.26,27 The step
in the "ladder" of electron excitation energies is about
∆Eel ≈ 3.38·10−3 meV, as argued in Appendix B. How-
ever, the uncertainty in the phonon energies due to the
/LX ≈ 0.125 meV,
Heisenberg ratio is about δEvibr ∼ v∗
that makes the above mentioned discreteness of the elec-
tron spectrum practically irrelevant, in the context dis-
cussed.
L
The second remark, illustrated by Fig. 3c, is that the
electron modifies the direction of its momentum only
weakly, and the Fermi momentum of an electron par-
ticipating at the phonon absorption / emission process is
roughly at right angle to the (difference) momentum of
the phonons involved. Indeed,
F = s2 + (nq)2 − 2s(nq) cos γ ,
p2
γ = arccos
s2 + (nq)2 − p2
F
2s(nq)
2m(E2 − E1) + (nq)2
and
.
(1)
= arccos
2s(nq)
4
in principle,
one can,
imagine the decay of the ex-
cited electron state via promoting an electron across the
nanoobject, or via emitting a low-energy phonon, or via
a radiative transition. The last process is of interest for
us, as it will be the source of soft THz radiation. Here
we try to present the argumentation why the two other
processes might turn up to be less likely.
The electron mean free path in gold nanoobjects, ex-
pectedly shorter than that in nearly perfect bulk crys-
tal, seems to be comparable with the GNB cross-section
size; Sec. 9.3.3 of Ref. 28 discusses this issue, in relation
to small gold nanoparticles. The situation is therefore
likely that the electron will arrive at the GNB surface
after none, or very few, scattering events. The elec-
tron exit from the GNB is precluded by a prohibitively
high value of the work function of gold (4.3 eV29), in
comparison with the energy values ((cid:39) 1 meV, see Ta-
ble II) under discussion. This energy could have been
spent onto an emission of a low-energy phonon. The
"problem" in this relation is that this ought to be a
nearly zone-center phonon, with the nearly linear disper-
sion characterized by the "nominal" longitudinal speed
of sound in gold, vL = 3.23·105 cm/s.30 However, the
slope of ω(q) near the center of the Brillouin zone is
therefore about 3 times more steep than that near the
phonon energies of (cid:39)4 meV, within the FWHM of the
peak in the longitudinal density of vibration modes. Due
to the quantisation of q depicted in Fig. 3a, the min-
imal energy "quantum" to excite a low-energy phonon
(near Γ) would be ∆Evibr = vL(h/LX ) (cid:39) 2.52 meV.
The inspection of (E2−E1) values in Table II shows that
such excitation energies do not come about before ar-
riving at combinations with q1 = 5. One can note in
this relation that the issue of the Heisenberg's uncer-
tainty, that was earlier helpful for us to demonstrate that
the exact (anyway difficult to control) quantisation of
electron energies is of little practical importance, won't
work the same way for phonons: the quantisation step
in the vibration energies is ∆Evibr = vL(h/LX ) whereas
the Heisenberg's uncertainty for these energies will be
δEvibr = vLδpvibr = vL(/LX ), hence the energy step is
always ∼2π times larger than the smearing of each energy
separated by this step, whatever the vL in question.
We note in conclusion that the penetration depth of
electromagnetic radiation at the relevant soft THz fre-
quencies (those shown in Table I) exceeds by far the
"short" dimensions of the GNBs.
The calculated values of γ for different combinations of
absorbed/emitted phonons are given in Table II.
B. Channeling the electron excitation energy into
emission of soft THz photons
Following an absorption / emission of "nearly zone-
boundary" longitudinal phonons as elaborated above,
C. Substrate or matrix with gold nanobars for THz source
For nanobars designed to generate soft THz photons, it
is important that they are either deposited on a substrate
or embedded in a matrix of a material (see Fig. 4) that
is not only transparent in the soft THz range, but also
withstands sufficiently high temperature heating. Heat-
ing nanobars would increase the number of longitudinal
phonons in them, which would increase the power of the
5
FIG. 4.
Soft THz radiation source in the form of a matrix
transparent in THz with randomly aligned embedded GNBs.
generated THz radiation. Specifically for the applications
in the field of hidden objects detection, the detection
range in the THz domain is limited by the absorption by
water vapor, therefore increasing the THz source power
might be an important issue.
Another way to increase the radiation power is to use
a large number of nanobars, deposited on / in a large
area of substrate (or matrix); this would also facilitate
generating a required large-diameter (∼ 40 cm) spot.
Heating nanobars could be carried out by microwave
radiation, namely, with the help of a household mi-
crowave oven (see Sec. III below); the dimensions of the
furnace chamber are large enough to accommodate the
substrate / matrix of significant size.
In addition, the
metal walls of the chamber, reflecting the THz radiation
primarily emitted in all directions, would also contribute
to a more efficient use of the generated radiation. Con-
ditions for heating the GNBs by microwave radiation see
in Appendix B.
easily available heat-resistant materials,
Teflon R(cid:13) is transparent in the THz31. The use of a
Teflon R(cid:13) matrix and application of a standard 2.45 GHz
microwave radiation would allow GNBs to be heated to
temperatures of about 260 ◦C. Some matters related to
the possible ways of manufacturing the Teflon R(cid:13) matrix
were considered in our work.32 An even greater increase
in the temperature of nanobars could be achieved by us-
ing for the substrate or matrix such materials as the high-
resistivity float-zone silicon, crystal quartz, or sapphire.31
Among
III. POSSIBLE DESIGN OF THE
GOLD-NANOBARS-BASED SOFT THz RADIATION
SOURCE
The soft THz radiation source could be designed as
shown in Fig. 5 (see also patents Ref. 20,21). The housing
1 with metal chamber 2 inside comprises a substrate 3
with GNBs 4 deposited on it (otherwise, a matrix can be
used with embedded GNBs). An electromagnetic emitter
in the form of a magnetron 5 with a waveguide 6 opens
into the chamber; the power, control and cooling systems
are not shown. In the housing 1, an opening 8 is made, in
which a resonant filter 7 for the outcoming THz radiation
is installed. Beyond the filter 7, a focusing system 9
can be placed, collecting the THz radiation emitted by
GNBs 4 and focusing it on a hidden object 10 on the
subject under examination. Filters and lenses for the
FIG. 5. Possible construction of the GNBs-based soft THz
radiation source. Upper panel: front view, lower panel: top
view. See text for discussion.
THz range are commercially available and manufactured,
for example, by the TYDEX R(cid:13).33,34
The source operates as follows. The magnetron gener-
ates microwave photons, which enter the cavity and heat
the GNBs, that is manifested as an increase in the num-
ber of longitudinal phonons.
As was argued above, those Fermi electrons in the
GNBs which undergo an absorption / emission of
phonons along the scenario elaborated in Sec. II end up
in an excited state the relaxation from which can not
proceed by exiting the sample (the work function of gold
being prohibitively high) nor via emission of a low-energy
(close to the zone center) phonon.
When using a THz radiation source to scan humans,
that is, for a non-covert examination at short distances
(≤1 m) to detect objects hidden under clothing, the fo-
6
tion of heat into the substrate or matrix, that will blur
the "pixel", and (ii) the distribution of GNSs within the
converter's plate to ensure sufficient sensitivity and spa-
tial resolution of thermal picture to be perceived by the
IR camera.
If the matrix' thickness δ is within the depth of fo-
cus ldof of the IR camera's objective, the matrix seems
preferable over single-layer deposition, because it allows
to achieve larger "projected" density of GNSs per surface
unit. In the following analysis, we assumed δ = 0.1 mm
(of the Teflon R(cid:13) film) and ldof ≈ 0.3 mm, therefore, we
have done all estimations for the matrix (Fig. 7).
Table III outlines the parameters of GNSs to be em-
bedded into the Teflon R(cid:13) matrix so that to yield the con-
verter's reasonable efficiency. The basic considerations,
including estimations of the GNSs diameters, are evoked
and discussed in Ref. 32.
In a nutshell, on absorbing
a THz photon by a GNS, the (presumed nearly free)
Fermi electron is excited across mel steps of (confinement-
imposed) energy ladder, and then relaxed by releasing a
longitudinal phonon with nvibr minimal steps in wavevec-
tor and/or energy, assuming the linear dispersion for
phonons. The mismatch in the momentum conserva-
tion on such process, that would normally come about
due to different dispersion relations for photons, elec-
trons and phonons, will be tolerated by force of the
Heisenberg's uncertainty relation, as argued in Ref. 32.
Specifically, the confinement-conditioned uncertainty of
the electron momentum in a particle of diameter D,
∆pD ≈ h/(2πD), should exceed ∆pF, the mismatch of
momentum of the Fermi electron on absorption of the
THz photon of frequency ν. The corresponding estimates
are listed in Table III for four target frequencies specified
in Sec. I. For all these frequencies, the soft THz radiation
easily penetrates the particle (cf. "skin depth" values in
Table I), so that all Fermi electrons can participate in
the absorption of THz photons.
The upper part of the Table III deals with isolated
GNSs; the lower part concerns the latter's embedding in
Teflon R(cid:13), as a likely candidate for the substrate material.
The data on the 0.1 mm thick Teflon R(cid:13) film transmissions
are from Ref. 31. The emissivity factor α, a phenomeno-
logical property of "real" nanoparticles, indicates which
part of the energy delivered will then emerge via IR ra-
diation; the values α=1 to 0.5 will presumably bracket
the realistic estimates. ∆Tα is the excess of temperature
over the background to be created in the GNS to make
FIG. 7. THz-to-IR converter in the form of the matrix with
embedded GNSs.
FIG. 6.
Detection and imaging of objects concealed un-
der clothing. 1: the person inspected; 2: objective focusing
the reflected THz radiation onto the converter; 3: THz-to-IR
converter; 4: IR camera.
cusing system 9 might be left out. In this case, instead,
an unfocused flow of THz radiation is required, which
would provide a spot with a diameter of ∼40 cm on the
human body. THz photons reflected from the metal parts
of the hidden object are recorded by the THz detector
(see Figs. 1 and 6). As a possible realization of the lat-
ter, a THz-to-IR converter could be used, similar to that
suggested for visualization of malignant tumour.35 More
detailed estimations of the expected parameters of this
device's performance can be found in Ref. 32.
IV. THz TO IR CONVERTER FOR VISUALIZATION OF
HIDDEN OBJECTS
The second essential component of the proposed con-
cealed object detection system is the setup for visualiza-
tion of scattered / reflected THz image; the idea is to con-
vert the distribution of intensities over the THz wavefront
into the map of temperatures, and inspect the latter with
the help of a standard IR camera. The working element
of the THz to IR converter is the plate covered with a
sufficiently dense array of gold nanospheres (GNS), to be
heated under the effect of incoming THz radiation, or the
matrix bearing such objects densely embedded within.
Fig. 6 outlines the general suggested scheme of the de-
tection setup. The subject 1 is irradiated with a soft
THz radiation source. The radiation reflected from the
object hidden on the subject's body is focused by the lens
2 on the THz-to-IR converter 3 and creates the hidden
object's image in THz rays on the converter plate. The
converter absorbs THz radiation and channels its energy
into excitation of longitudinal phonons, that amounts to
local heating and creation of a point source of IR radia-
tion associated with each single gold nanosphere (GNS).
The resulting two-dimensional picture of IR "pixels" is
perceived by a standard IR camera 4 and made visible
on the camera's display. Technical considerations to be
elaborated include (i) the choice of acceptable GNS sizes
to ensure a good "performance" of a single GNS as heat
emitter, in view of reaction time and the spatial dissipa-
TABLE III. Parameters of GNSs intended for registration of soft THz radiations, for the pixel size d = 15 µm and the thickness
of the Teflon R(cid:13) matrix δ = 0.1 mm. Concentration of GNSs in the matrix is (d2 δ)−1 = 4.44·104 mm−3.
7
Property
Phonon momentum
Phonon momentum
mel
nvibr
D ≈ 4.23(mel/nvibr)1/2
Thermal conductivity λ1p
∆pD
∆pF
Transmission of the 0.1 mm thick Teflon R(cid:13) film
Emissivity factor α
∆Tα
QT
Threshold number of GNSs within volume element (d2δ)
that maps onto a pixel of IR camera
Units
h/aAu
10−21 g cm/s
nm
W m−1K−1
10−21 g cm/s
10−23 g cm/s
%
Frequency (THz)
0.41
0.050
8.11
0.14
0.020
3.25
36
1
25.4
220.1
≥ 0.41
0.66
∼ 90
0.24
0.024
5.35
12
1
14.65
126.9
≥ 0.72
1.135
∼ 90
0.70
0.090
14.60
2
1
6.0
52.0
≥ 1.76
3.31
∼ 92
0.5
1
14
28
5
1
9.5
82.3
≥ 1.11
1.94
∼ 92
0.5
28
1
14
1
14
0.5
28
1
14
0.5
28
mK
10−10 W
5.96 11.91 3.40 6.79 2.19 4.38 1.38 2.75
0.08
0.13
0.21
0.33
Heating time
Cooling time
Radius at which the temperature in the Teflon R(cid:13)
matrix falls to 1/10 that in the GNS centre
Operating power of the 9.6×7.7×0.1 mm3 Teflon R(cid:13)
matrix with embedded GNSs of diameter D
µs
µs
nm
0.57 0.56 0.43 0.43 0.33 0.33 0.23 0.23
1.53 1.58 1.32 1.39 1.38 1.36 1.18 1.19
100.0
62.3
42.0
27.3
µW
195.6 390.9 111.6 222.9 71.9 143.8 45.3 90.3
FIG. 8. Time characteristics (left panel) and radial temperature distributions (right panel) around the GNS with a diameter
of 14.65 nm (for visualization at 0.24 THz) in the Teflon R(cid:13) film matrix for two values of the emissivity factor, α=1 (lower curve)
and α=0.5 (upper curve).
it detectable by the IR camera; the nominal sensitivity
value of ≈ 14 mK, characterizing modern IR cameras,
needs to be increased in case of reduced values of α. In
order to attain such temperature on the GNS, the power
QT needs to be delivered, which follows from the solu-
tion of the heat transfer equation. With QT as a source
term, the heat equation describes the rise of tempera-
ture with time and the radial temperature profile across
the GNS and its surrounding medium. These proper-
ties are depicted in Fig. 8 for the particle with diameter
D = 14.65 nm embedded in Teflon R(cid:13).
From Fig. 8 (left panel) and from Table III one can
conclude that THz-to-IR converter built along the con-
cept outlined should possess the heating / cooling times
acceptable for real-time operation, and operating power
in the range of accessible.36 -- 39 Thus, the THz sources
discussed, as well as the visualization system based on
the THz-to-IR converter, could find application in the
equipment for detection of hidden objects.
Note that some issues related to improving the ef-
ficiency of conversion of THz radiation into heat by
nanoobjects were discussed in our article,25 and issues
related to possible methods of manufacturing a matrix
from Teflon R(cid:13) were considered in Ref. 32. Since the ma-
trix with embedded GNSs is not commercially available,
we mention a method that may be useful in the labora-
tory conditions. For the matrix, one could take gelatin, a
water soluble protein transparent in the soft THz range.
For embedding into the matrix, one may find it conve-
nient to use commercial GNSs which are sold in the form
of water suspensions.
The data on the transparency of the pure gelatin in
the THz range are scarce, therefore for the estimates
one could use the graph of the transmission spectrum
of the silver-ion-doped gelatin matrix spanning a wave-
length range of 0.2 µm to 1.5 mm from Ref. 40. Figure
4 in that article shows that there is a transmission win-
dow in the THz range, which notably spans the wave-
lengths corresponding to "our" chosen soft THz frequen-
cies (see Table I). The transmission of gelatin is relatively
high (∼ 80%) at 1.25 mm wavelength, but drops down to
∼ 50% at 0.43 mm. Although at wavelengths larger than
1 mm the transmission seems to be gradually declining,
one can extrapolate the trend to an acceptable value of
∼ 75% at the 2.1 mm wavelength.
V. CONCLUSION
Summarizing, we outlined a concept of soft THz radi-
ation sources in the form of the matrix (high-resistivity
float-zone silicon, crystal quartz, sapphire or Teflon R(cid:13)),
into which the gold nanobars are embedded. A speci-
ficity of soft THz radiation, for which we selected for
our analysis the photons with frequencies of 0.14, 0.24,
0.41 and 0.70 THz is that it may come about in the
course of two-phonon processes: first, an excitation of
the Fermi electrons with the absorption of the longitudi-
nal phonon, and then the release of a softer longitudinal
phonon, whereby the excited electron retains the differ-
ence in the energies of the two phonons -- respectively,
0.58; 0.99; 1.70 and 2.90 meV for the above frequen-
cies. The relaxation of electrons possessing these excess
energy over the Fermi level occurs via emission of THz
photons. To increase the number of longitudinal phonons
(and, as a result, to increase the emitted THz power),
nanoobjects are suggested to be heated with microwave
radiation at a standard (domestic micro oven) frequency
of 2.45 GHz. To generate THz photons with the fre-
quencies indicated, the gold nanobars with dimensions
5.3 nm× 5.3 nm× 1.318 µm seem suitable. Such THz ra-
diation sources being used together with the THz-to-IR
converter could be, among other possible applications, be
used in the equipment for discovering and visualization
8
of concealed goods.
Appendix A: Enhanced surface power density of
spontaneous THz radiation by gold nanoparticles due to
confinement
Here we would like to show just a principal possibility
of the significant excess of the Planck limit in the THz
range in gold nanoparticles. Namely, the mechanism of
photon emission discussed in Sec. II may be characterized
by quite elevated surface power density of spontaneous
THz radiation. We estimate this effect only by order
of magnitude and within a very simplified model: we
assume the nanoparticle to be a sphere with diameter
D = 5.3 nm; further on, we suppose that the release
of a secondary phonon occurs simultaneously with the
emission of a THz photon by the excited electron.
The D value chosen lets us refer to Fig. 3 and to the
discussion related to quantization of states due to con-
finement on this linear size, without entering the details
depending on the exact nanoobject's shape. For numeri-
cal estimations, we limit ourselves by the case of 0.24 THz
emitted photons. We estimate the excess, due to spatial
confinement, of the radiation emitted over the predictions
of the Planck's law for the black body radiation for two
special cases, taking into account the uncertainty in the
energy of both the longitudinal phonons and the Fermi
electrons.
The uncertainty in the energy of longitudinal phonons
The spatial confinement at the length scale D brings
about the uncertainty of the phonon momentum ∆p,
∆p·D ≥ . By relation of p to the phonon energy via
the (longitudinal) speed of sound v∗
L within the "range of
interest" (throughout the FWHM centered at ∼4 meV,
in our case), the uncertainty of energy is
L·∆p ≥ v∗
L h
2πD
∆E ≈ v∗
.
(A1)
For the Fermi electron which absorbed energy (E2−E1),
the excited state will have an uncertainty ∼±(∆E/2)
around the energy (EF+E2−E1). This would amount to
this state being characterized by a finite lifetime ∆t such
that
∆E·∆t≥ h
2π
.
(A2)
the previous
Combining
order-of-
magnitude estimate of the characteristic lifetime of
the excited electron is
relations,
the
∆t(cid:39) h
2π∆E
(cid:39) D
v∗
L
,
(A3)
that gives a physically meaningful estimate of the "flight
time" of a phonon across the nanoparticle. The average
emitted THz power can be roughly related to the photon
energy E2 − E1±(∆E/2) released during ∆t:
(cid:104)PTHz(cid:105) =
(cid:39)
E2 − E1±(∆E/2)
(cid:20)
E2 − E1± v∗
Lh
4πD
(cid:21) v∗
L
D
∆t
.
(A4)
In the phononic energy range of FWHM, the speed of
L (cid:39) 105 cm/s. The "energy uncertainty" term
sound is v∗
±(v∗
Lh)/(4πD) ≈ 1·10−2 meV can be neglected com-
pared to E2− E1 = 0.99 meV. The emitted power is then
(cid:104)PTHz(cid:105)(cid:39) 3·10−11 W, and the surface density of the THz
power emitted by the gold nanosphere with the diameter
D = 5.3 nm:
(cid:104)PTHz(cid:105)
4π(D/2)2 ≈ 3.4·105 W/m2 .
(A5)
In order to make comparison with the prediction by the
Planck's theory, we integrate the function
(λ) =
2πc2
λ5
(cid:1) − 1
exp(cid:0) hc
h
kλT
over wavelengths throughout the FWHM, i.e., between
λmin = 6.53·10−5 m and λmax = 7.76·10−5 m, that yields
2.68 W/m2 (c is the speed of light in vacuum, h the
Planck constant, k the Boltzmann constant, T the tem-
perature assumed to be 300 K).
Note that the estimate Eq. (A5) was performed for a
single pair of phonons (with an energy difference of 0.99
meV) within the FWHM. This turned out already suf-
ficient to demonstrate that the Planck formula gravely
underestimates in the THz range on a nanometric scale,
as was already noted in the works15 -- 17 devoted to the
study of far-field thermal radiation transfer in nanoscale
objects with a size smaller than the Wien's wavelength
λW (≈ 10 µm at T = 300 K). Experimental studies
have shown that deviations from the predictions of the
Planck's theory can reach two to seven orders of magni-
tude.
Accounting for uncertainty in the energy of the Fermi
electrons
Since the Fermi electron velocity in gold vF =
1.4·108 cm/s is three orders of magnitude higher than
the speed of sound in the phononic energy range of
L (cid:39) 105 cm/s, a similar argumentation
FWHM, v∗
hints for an excess over the prediction by the Planck's
formula by three more orders of magnitude, yield-
ing, for gold nanosphere with D = 5.3 nm as above,
(cid:104)PTHz(cid:105)/[4π(D/2)]2 (cid:39) 2.7·109 W/m2.
(For comparison:
the surface density of power radiated by the Sun's sur-
face is (cid:39) 7·107 W/m2). Overcoming the Planck limit of
the surface power density of radiation inspires attempts
to generate soft THz radiation using a large number of
9
nanoparticles. So it would be possible to make a radia-
tion source of sufficiently large total power, unattainable
or difficult to achieve by other ways. By the order of
estimate, the total power of the distributed source ra-
diating into the solid angle 4π at frequency 0.24 THz,
with a matrix diameter of 200 mm, a thickness of 1 mm,
with a concentration of 105 mm−3 nanoparticles might
be (cid:39) 94 mW.
Appendix B: Choice of optimal size for gold nanobars
We discuss now the condition for a microwave pho-
ton with energy hν (assuming ν=2.45 GHz, the standard
microwave frequency of a domestic oven) to be most effi-
ciently absorbed by a gold nano-object with dimensions
LX = Nx aAu, LY = Ny aAu, LZ = Nz aAu. The absorp-
tion excites a Fermi electron into a state with the en-
ergy EF + mel∆Eel, several (mel) confinement-dependent
quantization steps, ∆Eel = 4
3 EF/N by force of the Kubo
formula,26,27 above the Fermi energy EF (N = 4NxNyNz
is the number of univalent gold atoms in the nanoparti-
cle). Following the excitation, the electron relaxes on
releasing a longitudinal phonon, whereby the nanopar-
ticle is heated. The vibration states are quantized, the
smallest step in the momenta values being that in the
direction in which the GNB is the longest, i.e., LZ. The
corresponding step in energy, assuming linear dispersion
law with the longitudinal speed of sound vL defining the
slope, will be ∆Evibr = vLh/LZ. The conservation of
energy imposes
hν = mel∆Eel = nvibr∆Evibr .
(B1)
for some mel and nvibr integer. The conservation of mo-
mentum cannot be exactly respected, because the dis-
persion relations for photons and phonons are markedly
different (see Fig. 9.2 and related discussion in Ref. 28).
However, the mismatch of momentum can be "absorbed"
in the uncertainty of the phonon momentum, ∆p(cid:39) /LZ,
which comes about as a (yet another) consequence of spa-
tial confinement. The condition for ∆p to help match the
momenta will read ∆p≥ ∆pF, whence the condition on
LZ (formulated for ν = 2.45 GHz):
LZ ≤ vF
2πν
≈ 91 µm .
(B2)
Making use of Eq. (B1) formulated in terms of Nx, Ny,
Nz, we get
mel =
3hν
EF
NxNyNz , nvibr =
νaAu
vL
Nz
and hence
mel
nvibr
=
3h vLNxNy
aAuEF
.
(B3)
(B4)
We are interested in the small size of the GNBs, because
heating a small GNB requires less microwave power. This
implies the smallness of the numbers Nx, Ny, Nz as well
as of mel and nvibr. Assuming for simplicity Nx = Ny and
applying the numerical values of h, vL = 3.23·105 cm/s,30
aAu = 0.408 nm29, EF = 5.53 eV,29 we obtain:
mel
nvibr
= 1.777·10−2·N 2
x .
(B5)
We select the minimum value of the nvibr parameter,
namely =1, and search for its compatible small enough
integer values of mel.
It turns out that at Nx = 13,
mel = 3.003 ≈ 3, and for Nx=15, mel = 3.998 ≈ 4. We
retain the smallest value Nx=13 for calculations in the
present work; in particular, we split the Γ− X interval in
Fig. 3a into 13 intervals. The GNB's width and height are
LX =LY =NxaAu=NyaAu=13×0.408 nm = 5.3 nm. From
the relations nvibr·∆Evibr = hν and ∆Evibr = vL(h/Lz),
we obtain for the length of GNB: Lz = nvibr vL/ν,
which for the frequency ν=2.45 GHz yields Lz=1.318 µm
(Nz ≈ 3230), whereby ∆Eel ≈ 3.38·10−3 meV.
The synthesis of single-crystal gold nanowires with
lengths of up to several microns has already been
mastered.41 -- 44 Manufacturing the GNBs of 1.318 µm
length will hopefully not be a problem. According to the
theoretical analysis,45 for gold nanowire to be "stable",
its diameter must be greater than 4.5 aAu, i.e., larger
than 1.84 nm. Our estimates of the minimal "thickness"
of the GNBs well respects this condition.
REFERENCES
1Bjarnason, J. E., Chan, T. L. J., Lee, A. W. M., Celis, M. A.,
and Brown, E. R., "Millimeter-wave, terahertz, and mid-infrared
transmissionthrough common clothing," Applied Physics Let-
ters 85(4), 519 (2004).
2Tribe, W. R., Newnham, D. A., Taday, P. F., and Kemp, M. C.,
"Hidden object detection: security applications of terahertz tech-
nology," Proceedings of SPIE 5354 (Apr 2004). Terahertz and
Gigahertz Electronics and Photonics III.
3Gatesman, A. J., Danylov, A., Goyette, T. M., Dickinson, J. C.,
Giles, R. H., Goodhue, W., Waldman, J., Nixon, W. E., and
Hoen, W., "Terahertz behavior of optical components and com-
mon materials," Proceedings of SPIE 6212, 62120E (May 2006).
Terahertz for Military and Security Applications IV.
4Kemp, M. C., "Millimetre wave and terahertz technology for
the detection of concealed threats: a review," Proceedings of
SPIE 6402, 64020D (Sep 2006). Optics and Photonics for Coun-
terterrorism and Crime Fighting II.
5Appleby, R. and Wallace, H. B., "Standoff detection of weapons
and contraband in the 100 GHz to 1 THz region," IEEE Trans-
actions on Antennas and Propagation 55, 2944 (Nov 2007).
6Sheen, D. M., McMakin, D. L., Barber, J., Hall, T. E., and Sev-
ertsen, R. H., "Active imaging at 350 GHz for security applica-
tions," Proceedings of SPIE 6948, 69480M (Apr 2008). Passive
Millimeter-Wave Imaging Technology XI.
7Sheen, D. M., McMakin, D. L., Hall, T. E., and Severtsen, R. H.,
"Active millimeter-wave standoff and portal imaging techniques
for personnel screening," in [2009 IEEE Conference on Technolo-
gies for Homeland Security ], 440 (May 2009).
8Moeller, L., "Standoff detection of concealed weapons using a
terahertz illuminator with an uncooled imager," tech. rep., Bell
Laboratories (Aug 2011).
10
9Robertson, D. A., Macfarlane, D. G., Cassidy, S. L., Bryllert,
T., Gandini, E., and Llombart, N., "Submillimetre wave 3D
imaging radar for security applications," in [IET Colloquium on
Millimetre-Wave and Terahertz Engineering Technology 2016 ],
1 (Mar 2016).
10Cooper, K. B., Dengler, R. J., Llombart, N., Talukder, A.,
Panangadan, A. V., Peay, C. S., Mehdi, I., and Siegel, P. H.,
"Fast high-resolution terahertz radar imaging at 25 meters," Pro-
ceedings of SPIE 7671, 76710Y (Apr 2010). Terahertz Physics,
Devices, and Systems IV: Advanced Applications in Industry and
Defense.
11Kemp, M. C., "Explosives detection by terahertz spectroscopy --
a bridge too far?," IEEE Transactions on Terahertz Science and
Technology 1, 282 (Sep 2011).
12Linden, K. J., Neal, W. R., Waldman, J., Gatesman, A. J., and
Danylov, A., "Terahertz laser based standoff imaging system,"
in [34th Applied Imagery and Pattern Recognition Workshop
(AIPR'05) ], 8 (Oct 2005).
13Daniels, D. J., [EM Detection of Concealed Targets ], John Wiley
& Sons, Inc. (Dec 2009).
14Li, C. and Fang, G., "Terahertz imaging for security -- algorithm
and system realization," Terahertz Science and Technology 9, 19
(Mar 2016).
15Thompson, D., Zhu, L., Mittapally, R., Sadat, S., Xing, Z.,
McArdle, P., Qazilbash, M. M., Reddy, P., and Meyhofer, E.,
"Hundred-fold enhancement in far-field radiative heat transfer
over the blackbody limit," Nature 561, 216 (Sep 2018).
16Thompson, D., Zhu, L., Mittapally, R., Sadat, S., Xing, Z.,
McArdle, P., Qazilbash, M. M., Reddy, P., and Meyhofer, E.,
"Author correction: Hundred-fold enhancement in far-field ra-
diative heat transfer over the blackbody limit," Nature 567, E12
(Mar 2019).
17Fern´andez-Hurtado, V., Fern´andez-Dom´ınguez, A. I., Feist, J.,
Garc´ıa-Vidal, F. J., and Cuevas, J. C., "Exploring the limits of
super-Planckian far-field radiative heat transfer using 2D mate-
rials," ACS Photonics 5, 3082 (Aug 2018).
18Moldosanov, K. A., Lelevkin, V. M., Kozlov, P. V., and Kaveev,
A. K., "Terahertz-to-infrared converter based on metal nanopar-
ticles: potentialities of applications," Journal of Nanophoton-
ics 6, 061716 (2012).
20Moldosanov, K. A. and Postnikov, A., "Source of
19Moldosanov, K. and Postnikov, A., "A terahertz-vibration to
terahertz-radiation converter based on gold nanoobjects: a feasi-
bility study," Beilstein Journal of Nanotechnology 7, 983 (2016).
ter-
ahertz
radiation. Russian patent RU 2622093. Priority:
13.05.2016, date of publication: 09.06.2017 (Bull. 16), corrected:
25.07.2017 (Bull. 21)." https://patents.google.com/patent/
RU2622093C1/en (2016). Accessed : 14 August 2019.
21Moldosanov, K. A. and Postnikov, A. V., "Converter of tera-
hertz vibrations into terahertz electromagnetic radiation. Rus-
sian patent RU 2650343. Priority: 20.03.2017, date of publi-
cation: 11.04.2018 (Bull. 11)." https://patents.google.com/
patent/RU2650343C1/en (2017). Accessed : 14 August 2019.
22Lynn, J. W., Smith, H. G., and Nicklow, R. M., "Lattice dynam-
ics of gold," Phys. Rev. B 8, 3493 (Oct 1973).
23Munoz, J. A., Lucas, M. S., Mauger, L., Halevy, I., Horwath, J.,
Semiatin, S. L., Xiao, Y., Chow, P., Stone, M. B., Abernathy,
D. L., and Fultz, B., "Electronic structure and vibrational en-
tropies of fcc Au-Fe alloys," Phys. Rev. B: Condens. Matter 87,
014301 (Jan 2013).
24Bayle, M., Combe, N., Sangeetha, N. M., Viau, G., and Carles,
R., "Vibrational and electronic excitations in gold nanocrystals,"
Nanoscale 6, 9157 (2014).
25Postnikov, A. V. and Moldosanov, K. A., "Suggested design of
gold-nanoobjects-based terahertz radiation source for biomedical
research," Nanotechnology 29, 285704 (Jul 2018).
26Kubo, R., "Electronic properties of metallic fine particles. I.," J.
Phys. Soc. Jpn. 17(6), 975 (1962).
27Kubo, R., "Discreteness of energy levels in small metallic parti-
cles," J. Phys. Colloques 38, C2 -- 69 (Jul 1977).
28Postnikov, A. and Moldosanov, K., "Phonon-assisted radiofre-
quency absorption by gold nanoparticles resulting in hyper-
thermia," in [Fundamental and Applied Nano-Electromagnetics ],
Maffucci, A. and Maksimenko, S. A., eds., The NATO Science
for Peace and Security Programme, Series B: Physics and Bio-
physics, 171 -- 201, Springer, Dordrecht, The Netherlands (2016).
Proceedings of the NATO Advanced Research Workshop on Fun-
damental and Applied Electromagnetics, Minsk, Belarus, 25-27
May, 2015.
29Ashcroft, N. W. and Mermin, N. D., [Solid State Physics ], Saun-
ders College (1976).
30Singh, R. N. and Ali, I., "Elastic moduli and phonon dispersion
curves for amorphous metals and alloys," International Journal
of Applied Physics and Mathematics 3, 275 (Jul 2013).
31"TYDEX R(cid:13) THz materials." http://www.tydexoptics.com/
products/thz_optics/thz_materials/. Accessed: 14 August
2019.
32Postnikov, A. V., Moldosanov, K. A., Kairyev, N. J., and
Lelevkin, V. M., "Prospects for terahertz imaging the human
skin cancer with the help of gold-nanoparticles-based terahertz-
to-infrared converter," in [Fundamental and Applied Nano-
Electromagnetics II ], Maffucci, A. and Maksimenko, S. A., eds.,
NATO Science for Peace and Security Series B: Physics and
Biophysics, 151, Springer, Dordrecht, The Netherlands (2019).
Proceedings of the NATO Advanced Research Workshop on Fun-
damental and Applied NanoElectroMagnetics II: THz Circuits,
Materials, Devices. Minsk, Belarus, 5-7 June, 2018.
33"TYDEX R(cid:13) THz band pass filters." http://www.tydexoptics.
com/products/thz_optics/thz_band_pass_filter/. Accessed:
14 August 2019.
34"TYDEX R(cid:13) THz
lenses."
http://www.tydexoptics.com/
products/thz_optics/thz_lens/. Accessed: 14 August 2019.
35Moldosanov, K. A., Postnikov, A. V., Lelevkin, V. M., and
Kairyev, N. J., "Terahertz imaging technique for cancer diag-
11
nostics using frequency conversion by gold nano-objects," Ferro-
electrics 509(1), 158 (2017).
36Gallerano, G. P. and Biedron, S., "Overview of terahertz radia-
tion sources," in [Proceedings of the 2004 FEL Conference ], 216
(2004).
37Tonouchi, M., "Cutting-edge terahertz technology," Nature Pho-
tonics 1, 97 (Feb 2007).
38Booske, J. H., "Plasma physics and related challenges of
millimeter-wave-to-terahertz and high power microwave gener-
ation," Physics of Plasmas 15, 055502 (May 2008).
39Armstrong,
C. M.,
"The
truth
about
terahertz."
https://spectrum.ieee.org/aerospace/military/
the-truth-about-terahertz (2012).
Aug 2012), accessed: 13 August 2019.
IEEE Spectrum (20
40Kang, S., Vora, K., and Mazur, E., "One-step direct-laser metal
writing of sub-100 nm 3D silver nanostructures in a gelatin ma-
trix," Nanotechnology 26, 121001 (Mar 2015).
41Huo, Z., Tsung, C.-k., Huang, W., Zhang, X., and Yang, P.,
"Sub-two nanometer single crystal Au nanowires," Nano Let-
ters 8, 2041 (Jul 2008). PMID: 18537294.
42Kang, Y., Ye, X., and Murray, C. B., "Size- and shape-selective
synthesis of metal nanocrystals and nanowires using CO as a
reducing agent," Angewandte Chemie International Edition 49,
6156 (Aug 2010).
43Kim, F., Sohn, K., Wu, J., and Huang, J., "Chemical synthesis
of gold nanowires in acidic solutions," Journal of the American
Chemical Society 130, 14442 (Nov 2008). PMID: 18850710.
44Pazos-P´erez, N., Baranov, D., Irsen, S., Hilgendorff, M., Liz-
Marz´an, L. M., and Giersig, M., "Synthesis of flexible, ultrathin
gold nanowires in organic media," Langmuir 24, 9855 (Sep 2008).
45Wang, F., Dai, Y., and Zhao, J., "The fracture stabilities of ul-
trathin gold nanowires," Superlattices and Microstructures 100,
237 (Dec 2016).
|
1905.05060 | 1 | 1905 | 2019-05-13T14:37:21 | Real-time monitoring of stress evolution during thin film growth by in situ substrate curvature measurement | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Strain engineering is the art of inducing controlled lattice distortions in a material to modify specific physicochemical properties. Strain engineering is applied for basic fundamental studies of physics and chemistry of solids but also for device fabrication through the development of materials with new functionalities. Thin films are one of the most important tools for strain engineering. Thin films can in fact develop large strain due to the crystalline constrains at the interface with the substrate and/or as the result of specific morphological features that can be selected by an appropriate tuning of the deposition parameters. Within this context, the in situ measurement of the substrate curvature is a powerful diagnostic tool allowing a real time monitoring of the stress state of the growing film. This manuscript reviews a few recent applications of this technique and presents new measurements that point out the great potentials of the substrate curvature measurement in strain engineering. Our study also shows how, due to the high sensitivity of the technique, the correct interpretation of the results can be in certain cases not trivial and require complementary characterizations and an accurate knowledge of the physicochemical properties of the materials under investigation. | physics.app-ph | physics | Citation: Journal of Applied Physics 125, 082513 (2019); doi: 10.1063/1.5054092
View online: https://doi.org/10.1063/1.5054092
Real-time monitoring of stress evolution during thin film growth by in situ substrate
curvature measurement
Elisa Gilardi1, Aline Fluri1, Thomas Lippert1,3,4, Daniele Pergolesi1,2*
1 Research with Neutrons and Muons Division, Paul Scherrer Institut, 5232 Villigen PSI,
Switzerland
2 Energy and Environment Research Division, Paul Scherrer Institut, 5232 Villigen PSI,
Switzerland
3 Department of Chemistry and Applied Biosciences, Laboratory of Inorganic Chemistry, ETH
Zürich, Vladimir-Prelog-Weg 1-5/10, 8093 Zürich, Switzerland
4 International Institute for Carbon Neutral Energy Research (WPI-I2CNER), Kyushu
University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan
*[email protected]
Abstract
Strain engineering is the art of inducing controlled lattice distortions in a material to modify
specific physicochemical properties. Strain engineering is applied for basic fundamental studies
of physics and chemistry of solids but also for device fabrication through the development of
materials with new functionalities. Thin films are one of the most important tools for strain
engineering. Thin films can in fact develop large strain due to the crystalline constrains at the
interface with the substrate and/or as the result of specific morphological features that can be
selected by an appropriate tuning of the deposition parameters. Within this context, the in situ
measurement of the substrate curvature is a powerful diagnostic tool allowing a real time
monitoring of the stress state of the growing film.
This manuscript reviews a few recent applications of this technique and presents new
measurements that point out the great potentials of the substrate curvature measurement in strain
1
engineering. Our study also shows how, due to the high sensitivity of the technique, the correct
interpretation of the results can be in certain cases not trivial and require complementary
characterizations and an accurate knowledge of the physicochemical properties of the materials
under investigation.
I. Introduction
Many examples are known where a change of the interatomic distance in the lattice of a material
leads to variations of certain physicochemical properties. Lattice strain can in fact modify
conducting,1-4 electronic and optical,5 catalytic6 or electrochemical,7,8 mechanical and thermal9
properties. The effect of strain can be in some case enabling new properties or functionalities that
were not present in the relaxed structure. Many studies on the effect of strain are conducted using
thin films as model systems where strain arises in consequence of the interfacial constrain at the
film/substrate interface or due to specific morphological features that can often be selected by an
appropriate tuning of the deposition parameters.
In the case of highly ordered epitaxial films, i.e. when film and substrate materials have similar
lattice parameter and suitable crystallographic matching, strain is induced by the film-to-
substrate lattice misfit. In the ideal case of a 1:1 match of all lattice planes at the interface, the
lattice mismatch is entirely converted into lattice strain. Often however, a large part of the
theoretical lattice misfit is compensated by introducing crystal defects to release the excess
strain. A typical mechanism of stress relaxation is the formation and migration of misfit
dislocations and for many oxides lattice strain exceeding a few percent cannot be elastically
accomodated.10
While for epitaxial films with very high crystallographic quality the origin of the strain is
obviously recognisable, for polycrystalline or textured films it is not straightforward. In this case
in fact, the strain state and extent depend on the kind of grain boundary formed which in turn
depends on the specific material, deposition method and experimental condition. Only the direct
observation of the local morphological features, for example by transmission electron
microscopy, can help to explain the measured strain state and identify its origin.
The quantitative analysis of strain in thin films is mostly performed ex situ, i.e. after the growth,
by X-ray diffraction (XRD) typically through 2/ scans and/or reciprocal space mapping. Of
course these methods provide the overall and average value of strain along the film and cannot
2
be used to investigate the evolution of the strain during the growth. This implies that the
presence of regions with different strain at different distances from the substrate for example
cannot be distinguished. Moreover, the analysis becomes very challenging in the case of ultra-
thin layers (below 10 nm), which is often the range of thickness where large strain are retained.
In situ XRD at synchrotron light sources11 or reflection high energy electron diffraction
(RHEED) were used as diagnostic tools to monitor the strain evolution. The first cannot
obviously be routinely applied at laboratory scale on a daily basis. Concerning RHEED, its
application as strain monitor is very rare.12 The sensitivity of RHEED to the changes of the in-
plane lattice parameter is limited by the spatial resolution of the diffracted electron spots. This is
especially the case at relatively high pressure (often used in pulsed laser deposition or sputtering
for example) and when the intensity of the spot changes with increasing thickness.
Optical measurements of the curvature of the substrate during the film growth may offer an
efficient and practical tool to monitor in situ the direction and the evolution of the strain along
the films.13 The mechanic constraint of the substrate does not allow the film to grow freely along
the plane of the surface of the substrate (in-plane). Volumetric changes without constraint in the
film are only possible in the direction normal to the surface (out-of-plane). This results in stress
generation along the growing film which exerts a force in-plane that bends the substrate. As the
system reacts in order to minimize the elastic energy, the substrate develops a positive curvature
(with respect to the substrate surface normal) in case of in plane tensile strain and negative
curvature in case of compressive strain. The substrate curvature variation can be therefore used
as a tool to identify the stress direction and evolution. Moreover, if the mechanical properties of
the substrate are known also a quantitative estimation of the stress is possible.
Very accurate measurements of curvature variations can be obtained by measuring the deflection
of a laser beam reflected from the bending substrate toward a CCD camera that records the
changes of the position of the laser spot. For this purpose substrate in the form of thin, flexible
cantilever were used.14 The main limitation of such an approach is the availability of cantilevers
made out of the desired material with the required crystallographic properties in terms of lattice
mismatch and surface termination. So far mainly Pt and Si cantilever were used.
A more advanced experimental setup based on the same working principle is the so-called multi-
beam optical stress sensors (MOSS). The MOSS uses an array of laser beams that are reflected
from the surface of the substrate toward a CCD camera. Any change of the curvature of the
3
substrate changes the relative distance among the laser beams at the CCD camera as
schematically shown in Figure 1.
Such an experimental setup allows the use of any kind of substrates and using an n × m array of
laser beams the variation of the distance between the spots can be measured accurately by
averaging over multiple spots.
Figure 1. Schematic representation of the working principle of the Multi-beam Optical Stress
Sensor. An n × m array of laser beams is reflected at the substrate surface toward a CCD camera.
The mean differential spacing between the laser spots is recorded during the film deposition.
This allows for monitoring the substrate curvature which is directly related to the stress.
The variation of the substrate curvature during the growth of the film can be measured by
measuring with the MOSS the change of the mean differential spacing (m.d.s.) between the laser
spots using the following equation:
1
𝜌
= −
cos 𝛼
2𝐿
𝛿𝑑
𝐷0
where is the curvature, is the angle of incidence of the laser beams with respect to the
substrate surface normal, L is the optical path of the laser beams, Do is the initial value of the
m.d.s. and d is the variation of the m.d.s. as measured in situ by MOSS.
4
Once the changes of curvature can be measured, and knowing the elastic properties of the
substrate, the Stoney equation can be used to calculate the stress thickness product (σ ∙ τ) of the
growing film:
1
𝜌
=
6
2
𝜏𝑠
1 − ν
𝑌
σ ∙ τ
, Y, and s being respectively the Poisson ratio, Young modulus and thickness of the substrate.
Literature reports several studies of stress generation and evolution in this films based on the
MOSS analysis. Mainly metals and semiconductor films grown by ultra-high vacuum
evaporation and sputtering15-21 and pulsed laser deposition22 (PLD) were investigated, and more
recently also epitaxial oxides films made by PLD. 1,2,10 In situ investigation of the stress
evolution in oxide films are indeed quite scarce, which is surprising considering the scientific
and technological interest of oxide materials and the effect that strain can have in oxides.
In our previous studies1,2 we used oxygen ion and proton conducting oxides as model systems to
monitor real time by MOSS the evolution of stress. These solid state ionic conductors are
materials of great scientific and technological interest for sustainable energy conversion.23,24 The
investigated films were epitaxially oriented and showed high crystallographic quality with only
small-angle grain boundaries. Basically, surface energy and stress govern the initial stage of the
growth, while the growth mode (layer-by-layer or island-like) and the nucleation and migration
of dislocations lines determine the subsequent stress relaxation.10
With the present manuscript we report more details about the highest sensitivity we could
achieve for the MOSS measurement of the substrate curvature during the growth of oxide
materials by PLD (probably the most widespread method for growing oxide films) at high
temperature in an oxygen background pressure. We also present new observations of the stress
evolution in textured oxide films, with columnar polycrystalline morphology very commonly
observed in oxide film deposited by PLD. Finally, we report on the importance of the choice of
the substrate material for a meaningful interpretation of the in situ curvature measurements.
II. Methods
Thin films of 15% Sm-doped CeO2 (SDC) and 8% Y2O3 stabilized ZrO2 (YSZ) were grown by
pulsed laser deposition using sintered ceramic pellets prepared in our laboratories as targets for
5
ablation. Commercially available 10×10×0.5 mm3 single crystal of Al2O3, LaAlO3, and SrTiO3
were used as substrates. The vacuum chamber has a base pressure of about 10-6 Pa and is
equipped with a multi-beam optical stress sensor (MOSS) for in situ measurement of changes of
the substrate curvature during the growth of thin films. For MOSS measurements a square 3×3
array of laser beams was directed to the centre of the substrate under an incidence angle of 30°.
The distance between the spots of the laser beams in the array was in the range of 1 mm. O2 was
used as the background gas during ablation setting a partial pressure in the range between 2 and
5 Pa. The target to substrate distance was set at 5 cm. A radiant heater was used to set the
deposition temperature at 750 °C. The measurements of the changes of the curvature of the
substrates were always performed at constant temperature and background pressure. The MOSS
curvature measurement does not allow the use of a metal paste to provide the required thermal
contact between the substrate and heating stage. For this, the back (unpolished) side of the
substrates was coated by a sputtered Pt film, about 500 nm thick, acting as the heat absorber for
the otherwise transparent substrates. The deposition temperature was read out using a pyrometer
pointing at the centre of the substrate setting the emissivity value of 0.97 for black Pt. A 248 nm
KrF excimer laser with pulse width of 25 ns was focused onto the targets on a spot of about
1 mm2 with an energy density of about 1.3 J cm-2. In these experimental conditions, with a laser
frequency of 2 Hz, a deposition rate of about 0.1 and 0.07 Å per pulse was found for SDC and
YSZ, respectively. The deposition rate was calibrated by X-ray reflectometry. The structural
characterization of the films was performed by X-ray diffraction.
III. Results and Discussion
We focus first on the sensitivity that could be achieved with our experimental setup for the
measurement of the relative change of the curvature of the substrate during the growth.
We make use of a thin film of 8% Y2O3 stabilized ZrO2 (YSZ) grown on Al2O3 substrate as an
example. Figure 2a shows the XRD analysis of the film revealing the polycrystalline nature with
a textured microstructure characterized by grains (100) and (111) out-of-plane oriented.
Figure 2b shows the X-ray reflectometry (XRR) measurement used for the calibration of the
deposition rate which was found to be 0.07 Å per pulse with the selected deposition parameters.
The red open circles in Figure 2c draw the progress of the substrate curvature with time. The
curvature is calculated by averaging the m.d.s. measured by MOSS. The continuous black line
6
shows how the thickness of the film changes with time. The slope of the curve is calculated on
the base of the XRR measurement of Figure 2b.
The deposition starts at 1000 seconds and the MOSS shows a negative curvature of the substrate
which indicates the development of an in-plane compressive stress along the growing film
(Figure 2c). The curvature increases almost linearly with increasing thickness. According to the
Stoney equation this indicates an almost constant stress value.
Figure 2. Analysis ex situ of the crystal structure and in situ of the stress evolution in a thin film
of YSZ on Al2O3 (0001) a) 2θ/θ scan of the polycrystalline thin film. Film peaks correspond to
the (100) and (111) orientations. b) XRR scan used for thickness measurement and deposition
rate calibration of YSZ thin films. c) Red circle: Substrate curvature variation monitored by
MOSS as a function of time. The negative substrate curvature indicates in-plane compressive
strain. Black line: film thickness calculated from the XRR measurement of the deposition rate.
Thin film deposition occurs between 1000 and 2800 s and between 3600 and 4000 s. Variation in
the substrate curvature are already distinguishable at 8 -- 10 Å thickness. d) Red circle: Mean
differential spacing of the laser beam spots recorded by MOSS (MOSS raw data) during the
deposition of the topmost 3 nm of the same sample. Deposition occurs between 400 and 800 s.
Black line: Calculated thickness as a function of the time.
7
The deposition is stopped after 2800 seconds and the MOSS shows that also the curvature stops
increasing. The elastic energy accumulated in the film in the form of a compressive in-plane
strain stays constant when no more material is added. It is important to note that no evidence of
stress relaxation can be observed. At this point the thickness of the YSZ film is about 17 nm.
After 1000 seconds the deposition was resumed and the MOSS clearly shows that the sapphire
substrate continues bending in the same direction. After the growth of additional 3 nm the
deposition is ended with a total thickness of about 20 nm and the MOSS shows again that no
further changes of the substrate curvature are detected. In Figure 2d the open red circles indicates
the in situ MOSS measurement of the m.d.s. (the MOSS raw data) as a function of time and the
black line is the evolution of the thickness of the film. This data refers to the last 3 nm added
after the first stop of the deposition process shown in Figure 2c. We would like to highlight the
high sensitivity of the technique, which depends on the background noise, the substrate elastic
modulus and the stress generated in the film during the deposition.
For Al2O3 substrate and the observed signal-to-noise ratio (depending on the substrate material
and the background vibrations) the curvature resolution leads to a detectable variation of the
strain-thickness product of -2.5 GPa nm already at thickness 1 nm; which means that for strains
close to 0.4% and an elastic modulus of 350 GPa (as in the case of Al2O3), clear changes of
wafer curvature can be detected already for thicknesses as small as of 8-10 Å (corresponding to
less than 2 unit cells of YSZ). We would like also to highlight here that MOSS analysis not only
provides a real-time diagnostic of the stress state of the growing films (whether compressive or
tensile) but it provides reliable information for film thicknesses that would be almost impossible
to analyse by standard (ex situ) XRD.
The use of C-cut sapphire substrates can promote the epitaxial growth of YSZ films along the
(111) crystallographic direction.25 Also the textured morphology with mixed (100) and (111)
orientation is reported.26 In general, as for many oxides, films prepared by PLD show a typical
columnar morphology consisting on parallel pillars with relatively narrow size distribution
separated by grain boundary regions.25,27 An example of such morphological feature is given in
Figure 3 for a film of doped ceria grown on a sapphire substrate.
8
Figure 3: SEM micrograph of a thin film of doped ceria on Al2O3 (0001). The typical columnar
morphology of the film is clearly visible.
In the case of epitaxial films of YSZ on C-cut sapphire, due to the lattice mismatch an in-plane
compressive strain of the film is expected, which is the same strain state detected by MOSS for
our samples. The out-of-plane lattice parameter calculated from the XRD measurement was
about 5.16 Å indicating an out-of-plane tensile strain consistently with the in-plane compressive
strain observed by MOSS. However, due to the columnar morphology, even in the case of
epitaxial films it would be questionable to ascribe a measured in-plane compressive strain to the
lattice misfit. Complementary characterizations (transmission electron microscopy, for example)
would be needed to investigate what type of grain boundary is present between adjacent grains
and how the grain boundary regions contribute in determining the final stress of the film. This
consideration is even more important in the case of textured films showing multiple orientations,
as those reported here.
C-cut sapphire substrates were also used for the growth of SDC films. Epitaxial films28 as well as
films showing multiple orientations are reported in the literature. The large lattice mismatch can
favor the formation of interfacial misfit dislocations leading to the growth of relaxed epitaxial
films with almost no evidence of grain separation.28 Instead, when polycrystalline films were
grown with the typical columnar morphology using SiO2 substrates, the strain was found to be
compressive in-plane (tensile out-of-plane).28
Figure 4a shows the X-ray diffraction pattern of a thin film of SDC grown on (0001)-oriented
sapphire substrate. The film is polycrystalline with multiple orientations. Figure 4b shows the
MOSS measurement of the substrate curvature (red open circles) during the growth and the
evolution with time of the film thickness (black line), as calibrated by XRR. In agreement with
literature,28 for polycrystalline films the in-plane stress is compressive, as can be observed real-
9
time by MOSS (negative substrate curvature). Under the selected deposition parameters, up to a
maximum thickness of about 20 - 25 nm the curvature remains almost constant. The same
behavior was observed for YSZ films, as can be seen in Figure 2c. Growing the thickness larger
(above 25 nm), the slope of the curve in Figure 4b decreases indicating that the stress is released.
The total thickness of this film is about 35 nm but the topmost 10 nm do not contribute to the
overall strain. The XRD analysis reveals an average out-of-plane tensile strain of about 0.15%
which confirms the MOSS observation of an in-plane compressive stress. The comparison of
these measurements with the structural analysis reported in reference [28] suggests that for both
ceria and zirconia the typical columnar morphology obtained by PLD leads to the development
of an in-plane compressive stress.
Figure 4. a) 2θ/θ scan of a polycrystalline thin film of SDC on Al2O3 (0001) substrate. Film
peaks correspond to the (100) and (111) orientations. b) Substrate curvature (red circles)
recorded by MOSS during the deposition of the same sample. Black line indicates the films
10
thickness calculated according to the deposition rate. Thin film deposition occurs between 500
and 2500 s. The variation in the curvature line slope around 1000 s indicates that the stress is
released.
Finally, we would like to report a remarkable effect observed using SrTiO3 (STO) as substrate.
STO is among the most commonly used substrates for thin film growth; however the importance
of the stability of its chemical composition, in term of oxygen content, is often underestimated.
STO, as other cubic or pseudocubic perovskite substrates such as LaAlO3 (LAO) or NdGaO3
(NGO), is frequently used for the growth of highly ordered epitaxial ceria films.1 STO was also
used as buffer layer deposited on MgO substrates for the growth of ceria films 29,30 or
ceria/zirconia multilayers.30-32
The potential problem of this material is that at high temperature and relatively low oxygen
partial pressure (i.e. in the typical condition for thin film growth by PLD) STO is easily reduced
creating oxygen vacancies and thus enabling ionic and electronic conductivities. The ionic
mobility allows oxygen ions to diffuse easily across the interface with the growing film. Using
18O-labelled STO substrates it was shown that the STO substrate itself can become the main
source of oxygen for the growing film, even more than the oxygen molecules in the surrounding
gaseous environment of the target material when low background pressure is used.33,34 This may
have very important consequences as far as the stress generation and evolution in the growing
film is concerned, as described in Figure 5.
Figure 5a shows the XRD analysis of two films of approximately the same thickness (36 --
38 nm) of SDC grown on LAO (100) and STO (100). In both cases we have films (100)-
oriented. The inset in Figure 5a shows the magnification of the angular region around the (200)
reflexes. The dashed lines indicates the angular position of the (200) diffraction peak of the
relaxed structure of 15% Sm-doped CeO2 with a lattice parameter of about 5.43 Å. 35-37
As can be seen, compared to the relaxed structure both films show an out-of-plane tensile strain
which is about 0.40% for the film grown on LAO and 0.15% for that grown on STO. This
implies the presence of an in-plane compressive strain of SDC which is in agreement with what
one would expect on LAO (with a lattice mismatch of about 0.49%) but it does not agree with
what is expected on STO. On STO in fact, the SDC film should develop an in-plane tensile strain
as the consequence of a lattice mismatch of about 1.6%.
11
The MOSS measurements anticipated in situ the conclusion obtained ex situ by XRD, as shown
in Figure 5b. In both cases we measured a positive change of the m.d.s. between the laser spots
of the MOSS, indicating the development of an in-plane compressive stress.
On the LAO substrate the stress rises at the very early stage of the growth. The slope of the curve
describing the evolution of the m.d.s. vs. time decreases slightly after 1000 seconds
(corresponding to a film thickness of 25 nm) indicating that part of the strain is released.
Instead, MOSS shows a very different stress evolution during the SDC deposition on the STO
substrate. Negligible curvature variation of the substrate was detected for the first 10 - 15 nm.
Figure 5. a) XRD diffraction pattern of thin films of SDC on STO and LAO. In the inset:
magnification of the (200) reflex. b) Circles: MOSS analysis of the SDC layers during growth on
STO and LAO substrates. Lines: thickness increment as a function of time. Line slope was
determined according to the deposition rate.
12
After that, the m.d.s. starts increasing with almost constant slope. The m.d.s. remains then
constant when the deposition ends. The same qualitative behaviour was observed for several
SDC films on STO (always in-plane compressive stress was measured), though the details of the
curvature vs. time curve can vary quite significantly from sample to sample. On the contrary, the
MOSS measurements were very well reproducible using LAO or NGO substrates,1 materials that
are more difficult to reduce. It is also interesting to note that on LAO and NGO, not only the
stress state of SDC is in line with the lattice misfit (in-plane compressive on LAO and tensile on
NGO), but also the strain measured ex situ by XRD showed values very similar to the lattice
misfit.1
Summarizing, using STO as a substrate the in situ MOSS measurements are less reproducible
and, in agreement with ex situ XRD, show a stress state of the film opposite to that expected
considering the lattice misfit. High quality ceria films are typically obtained with deposition
parameters similar to those used in the present study using cubic or pseudo-cubic perovskite
substrates such as NGO, LAO and STO. The compressive, instead of tensile, stress observed for
the SDC films on STO is not ascribed to the effect of the textured polycrystalline morphology, as
in the cases of the ceria and zirconia films grown on Al2O3 discussed above.
A possible way to rationalize the experimental observation is the aforementioned high oxygen
ion mobility in STO and its tendency to be reduced when exposed to the typical deposition
conditions of high temperature and relatively low oxygen background pressure. As observed in
[33] and [34] for films of LAO and YSZ, the STO substrate can become the main source of
oxygen for the growing film which act as a sort of oxygen pump for the substrate. Due to the fast
oxygen ion diffusion in SDC, the oxygen ion concentration is expected to equilibrate through the
film thickness (40 nm) quite fast and no gradient of the oxygen ion concentration is expected in
the film after the deposition, as observed for YSZ.34 Conversely, oxygen ion will be depleted
from an STO layer at the film/substrate interface and in this interfacial layer the more reduced
STO will have a slightly larger lattice parameter than the less reduce STO in the bulk. Within
this scenario, the formation of such a layer would have the equivalent effect of the deposition of
thin film of a material in compressive in-plane stress that would bend the substrate inducing a
positive variation of the m.d.s. of the laser beams of the MOSS (equivalent to negative substrate
curvature variation), as shown in Figure 5b. This mechanism could thus explain the sign of the
curvature variation of the substrate detected by MOSS, and may be the cause of the final in-plane
13
compressive stress of the SDC films. The mechanisms through which the tensile strain in SDC is
released cannot be directly identified and can vary for different deposition parameters and
therefore different lattice parameter of STO at the interface. Wafer curvature was in fact reported
to be a reliable way to study the variation of oxygen concentration in metal oxides.38
Higher oxygen background partial pressure, in the range of a few tens of Pa, may be enough to
keep the oxygen content of the substrate more stable. Also, the use of O3 or N2O as the
background gas may increase the amount of oxygen in the film originating from the gaseous
environment, thus reducing the oxygen ion exchange with the substrate.
The measurements reported here show that only materials with highly stable oxygen content are
recommended as the substrate for application in oxide strain engineering. Very common
materials such as STO may lead to unpredictable results.
IV. Conclusions
In situ wafer curvature measurements offer a powerful tool for strain engineering to monitor real-
time the evolution of stress along a growing film. The application of this diagnostic method for
oxide materials is still quite rare, though a remarkable sensitivity on the stress state of the film
can be achieved. We have shown here that changes of substrate curvature induced by the
deposition of less than 2 unit cells of the growing film can be clearly detected.
The multi-beam optical stress sensor provides invaluable qualitative insights into the evolution of
the stress in-plane and in situ to complement the quantitative measurement of the out-of-plane
strain performed ex situ by standard 2/ scan.
Wafer curvature measurements are highly reliable and reproducible when stable oxide materials
are used as the substrates. This is the case for example for Al2O3, MgO, LaAlO3, NdGaO3.
Special care has to be taken when the chemical composition of the substrate is not stable, as it is
the case for instance for SrTiO3, one of the most commonly used single crystal substrate also for
application in strain engineering.
Acknowledgements
The authors gratefully thank the Swiss National Science Foundation under grant agreement
number v200020_147190.
14
References
A. Fluri, D. Pergolesi, V. Roddatis, A. Wokaun, and T. Lippert, Nature Communications 7, 10692
(2016).
A. Fluri, A. Marcolongo, V. Roddatis, A. Wokaun, D. Pergolesi, N. Marzari, and T. Lippert,
Advanced Science 4, 1700467 (2017).
J. A. Kilner, Nature Materials 7, 838 (2008).
B. Yildiz, MRS Bulletin 39, 147 (2014).
N. Healy, S. Mailis, N. M. Bulgakova, P. J. A. Sazio, T. D. Day, J. R. Sparks, H. Y. Cheng, J. V.
Badding, and A. C. Peacock, Nature Materials 13, 1122 (2014).
P. Strasser, S. Koh, T. Anniyev, J. Greeley, K. More, C. Yu, Z. Liu, S. Kaya, D. Nordlund, H.
Ogasawara, M. F. Toney, and A. Nilsson, Nature Chemistry 2, 454 (2010).
S. E. Temmel, E. Fabbri, D. Pergolesi, T. Lippert, and T. J. Schmidt, ACS Catalysis 6, 7566 (2016).
S. Temmel, E. Fabbri, D. Pergolesi, T. Lippert, and J. Schmidt Thomas, Advanced Materials
Interfaces 3, 1600222 (2016).
S. Li, X. Ding, J. Ren, X. Moya, J. Li, J. Sun, and E. K. H. Salje, Scientific Reports 4, 6375 (2014).
A. Fluri, D. Pergolesi, A. Wokaun, and T. Lippert, Physical Review B 97, 125412 (2018).
S. Bauer, S. Lazarev, A. Molinari, A. Breitenstein, P. Leufke, R. Kruk, H. Hahn, and T. Baumbach,
Journal of Synchrotron Radiation 21, 386 (2014).
J. Zhu, Y. R. Li, Y. Zhang, X. Z. Liu, and B. W. Tao, Ceramics International 34, 967 (2008).
M. F. Doerner and W. D. Nix, Critical Reviews in Solid State and Materials Sciences 14, 225
(1988).
J. Premper, D. Sander, and J. Kirschner, Applied Surface Science 335, 44 (2015).
E. Chason and J. A. Floro, MRS Proceedings 428, 499 (2011).
E. Chason, J. W. Shin, C. H. Chen, A. M. Engwall, C. M. Miller, S. J. Hearne, and L. B. Freund,
Journal of Applied Physics 115, 123519 (2014).
J. A. Floro, E. Chason, R. C. Cammarata, and D. J. Srolovitz, MRS Bulletin 27, 19 (2011).
G. Abadias, A. Fillon, J. J. Colin, A. Michel, and C. Jaouen, Vacuum 100, 36 (2014).
E. Chason, J. W. Shin, S. J. Hearne, and L. B. Freund, Journal of Applied Physics 111, 083520
(2012).
J. A. Floro and E. Chason, Applied Physics Letters 69, 3830 (1996).
S. Michotte and J. Proost, Solar Energy Materials and Solar Cells 98, 253 (2012).
T. Scharf, J. Faupel, K. Sturm, and H.-U. Krebs, Journal of Applied Physics 94, 4273 (2003).
S. J. Skinner and J. A. Kilner, Materials Today 6, 30 (2003).
E. Fabbri, A. Magrasó, and D. Pergolesi, MRS Bulletin 39, 792 (2014).
J. Jiang, X. Hu, W. Shen, C. Ni, and J. L. Hertz, Applied Physics Letters 102, 143901 (2013).
M. Gerstl, G. Friedbacher, F. Kubel, H. Hutter, and J. Fleig, Physical Chemistry Chemical Physics
15, 1097 (2013).
A. Infortuna, A. S. Harvey, and L. J. Gauckler, Advanced Functional Materials 18, 127 (2007).
M. C. Göbel, G. Gregori, X. Guo, and J. Maier, Physical Chemistry Chemical Physics 12, 14351
(2010).
S. Sanna, V. Esposito, D. Pergolesi, A. Orsini, A. Tebano, S. Licoccia, G. Balestrino, and E.
Traversa, Advanced Functional Materials 19, 1713 (2009).
P. Daniele, R. Vladimir, F. Emiliana, W. S. Christof, L. Thomas, T. Enrico, and A. K. John, Science
and Technology of Advanced Materials 16, 015001 (2015).
D. Pergolesi, E. Fabbri, S. N. Cook, V. Roddatis, E. Traversa, and J. A. Kilner, ACS Nano 6, 10524
(2012).
15
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
D. Pergolesi, E. Gilardi, E. Fabbri, V. Roddatis, G. F. Harrington, T. Lippert, J. A. Kilner, and E.
Traversa, ACS Applied Materials & Interfaces 10, 14160 (2018).
C. W. Schneider, M. Esposito, I. Marozau, K. Conder, M. Doebeli, Y. Hu, M. Mallepell, A. Wokaun,
and T. Lippert, Applied Physics Letters 97, 192107 (2010).
D. Stender, S. Cook, J. A. Kilner, M. Döbeli, K. Conder, T. Lippert, and A. Wokaun, Solid State
Ionics 249-250, 56 (2013).
A. Fluri, D. Pergolesi, V. Roddatis, A. Wokaun, and T. Lippert, Nature Communications 7, No.
10692 (2016).
D. Pergolesi, V. Roddatis, E. Fabbri, C. W. Schneider, T. Lippert, E. Traversa, and J. A. Kilner,
Science and Technology of Advanced Materials 16, No. 015001 (2015).
S. Sanna, V. Esposito, D. Pergolesi, A. Orsini, A. Tebano, S. Licoccia, G. Balestrino, and E.
Traversa, Advanced Functional Materials 19, 1713 (2009).
B. W. Sheldon, S. Mandowara, and J. Rankin, Solid State Ionics 233, 38 (2013).
16
32
33
34
35
36
37
38
|
1804.10137 | 2 | 1804 | 2018-08-22T08:23:15 | Coupled Hydro-Mechanical Aging Of Short Flax Fiber Reinforced Composites | [
"physics.app-ph",
"physics.class-ph"
] | One of the challenges in the widespread use of biocomposites for engineering applications is the influence of environmental conditions on their mechanical properties, particularly for a combination of aging factors such as temperature, moisture, and mechanical stresses. Thus, the purpose of this paper is to study the influence of coupled aging factors by focusing on a 100% bio-based and biodegradable composites made of flax/poly(lactic acid) with several fiber contents. The development of a specific testing setup enabled continuous in-situ measurements and allowed comparing the effects of combined aging factors to those of uncombined aging factors. It was confirmed that the aging temperature in wet conditions led to a loss of elastic properties, especially for higher fiber fractions. While creep tests in dry conditions resulted in little decrease of elastic properties, it was observed that mechanical loading of the materials combined with water immersion resulted in a strong synergistic effect on the loss of stiffness. Finally, the presence of fibers reduced environmental stress cracking mechanisms and increased the time to failure. | physics.app-ph | physics | Coupled hydro-mechanical aging of short flax fiber reinforced
composites
Arnaud Regazzi, Stephane Corn*, Patrick Ienny, Anne Bergeret
Ecole des Mines d'Ales, C2MA, 6 avenue de Clavieres, F-30319 Ales Cedex, France
a r t i c l e i n f o
a b s t r a c t
Keywords:
Durability
Flax
Short-fibre composites
Hygrothermal effect
Creep test
One of the challenges in the widespread use of biocomposites for engineering applications is the in-
fluence of environmental conditions on their mechanical properties, particularly for a combination of
aging factors such as temperature, moisture, and mechanical stresses. Thus, the purpose of this paper is
to study the influence of coupled aging factors by focusing on a 100% bio-based and biodegradable
composites made of flax/poly(lactic acid) with several fiber contents. The development of a specific
testing setup enabled continuous in-situ measurements and allowed comparing the effects of combined
aging factors to those of uncombined aging factors. It was confirmed that the aging temperature in wet
conditions led to a loss of elastic properties, especially for higher fiber fractions. While creep tests in dry
conditions resulted in little decrease of elastic properties, it was observed that mechanical loading of the
materials combined with water immersion resulted in a strong synergistic effect on the loss of stiffness.
Finally, the presence of fibers reduced environmental stress cracking mechanisms and increased the time
to failure.
1. Introduction
1.1. Context
Nowadays, sustainable development is of major concern, and
with it, biomaterials have become part of our lives. In the field of
composites, a wide range of biopolymers combined with numerous
natural fibers allows to meet various criteria in terms of functional
properties. Besides using renewable resources, bio-based compos-
ites present many advantages such as low environmental impact,
low density, and in some cases biodegradable ability [1]. In several
industrial applications (i.e. automotive, sports and leisure, con-
struction and infrastructure [2]), these materials are considered as a
promising alternative to the usual oil-based materials reinforced
with glass fiber [3]. For these reasons they are likely to increase
their market share. However, a widespread use of these composites
is curbed by technical difficulties [4]. Among them, the most
important are fiber cultivation vagaries, manufacture of compos-
ites, and imprecise knowledge of their behavior. But most of all,
using such materials in real life conditions (e.g. in the prospect of
* Corresponding author.
E-mail address: [email protected] (S. Corn).
outdoor applications) leads to specific problems of aging due to
numerous factors like temperature, water, radiations, bacteria and/
or mechanical loadings. These factors are meant to decrease ma-
terial properties, and to the worst case scenario they may lead to its
complete degradation.
Previous works described the various physico-chemical pro-
cesses that may take place into composites during aging (i.e. plas-
ticizing, swelling, hydrolysis, oxidation, interfacial decohesion) [5].
Although the mechanisms of
these processes are properly
described, their interactions and their effects on material proper-
ties are still poorly known. Consequently the interdependencies
between the aging processes render the prediction of mechanical
properties in given condition utterly complex. That is why envi-
ronmental and mechanical aging are mostly studied separately.
Yet some studies focused on the assessment of the coupled in-
fluence of water and stress [6e9], mainly on thermosets compos-
ites [10e15]. These studies mostly underlined the acceleration of
water diffusion in polymer resulting from the applied stress [16]. As
a result,
loadings accelerates physico-
chemical aging mechanisms and may even induce others [17].
repeated mechanical
In the literature, studies related to biobased composites dura-
bility as a consequence on their mechanical properties have
focused mostly on laminate composites made of
thermoset
matrices [18e23] But thermoplastic composites may also require
substantial mechanical properties [24e27]. In this paper, it was
proposed to assess the aging of poly(lactic acid), a thermoplastic
matrix also named PLA, reinforced with short flax fibers. The main
advantages of this composite are its 100% bio-based origin, its
biodegradability,
its low price, and its mechanical properties
equivalent to those of some oil-based polymers (e.g. polyethylene
terephthalate (PET)). Several papers have referred to the interesting
properties of PLA reinforced with natural fibers [1,28,29].
However, these composites are sensitive to environmental
conditions. The thermo-hydric aging of flax/PLA composites was
thoroughly discussed in a previous work [30]. Several papers also
underline the sensitivity of PLA to damage mechanisms in a
hygrothermal environment resulting from its low glass transition
temperature [25,31]. Like any polymer, it undergoes plasticizing
and swelling during sorption, but temperature may also trigger
irreversible mechanisms such as hydrolysis [32]. This process re-
sults in polymeric chain scissions and consequently in permanent
modifications of its properties [33]. Natural fibers behavior is also
decisive in the aging process. Indeed, temperature and humidity
hold sway also over the mechanical properties of plant fibers
[34,35]. Therefore a wet environment induces mostly a decrease of
the mechanical properties of composites reinforced with plant fi-
bers [36,37].
Concerning the influence of long term mechanical loading on
the aging of PLA/flax composites, no assessment has been reported
yet.
1.2. Methodology
The purpose of this paper is to study the coupled thermo-hydro-
mechanical aging of flax/PLA composites. It presents firstly the
impact on elastic properties of each aging factor considered sepa-
rately (i.e. water immersion or mechanical loading) and then the
impact of the factors concomitantly.
In a first phase of the study, composites with different fiber
contents were immersed in order to study the sole influence of
water at several temperatures. In a second phase, other samples of
the same materials were submitted to creep tests at these same
temperatures in order to evaluate the effects of a long term me-
chanical loading only. Finally, thermo-hydro-mechanical aging was
assessed for these materials, and the evolution of their mechanical
properties was compared to the direct addition of the separate
effects of hydric and mechanical aging.
Mechanical properties were assessed in situ in order to simulate
service conditions.
2. Materials and methods
2.1. Materials
2.1.1. Poly(lactic acid)
PLA Ingeo™ 7000D resin was produced by NatureWorks® LLC
(Blair, NE, USA). This grade of PLA, designed for injection stretch
blow molded applications, had a density of 1.24 g/cm3, a glass
transition temperature between 55 and 60 C and a melting tem-
perature between 155 and 165 C [38]. Preliminary tensile tests of
injected samples showed a Young's modulus of 3.8 ± 0.1 GPa, a
strength of 65 ± 1 MPa, and a strain at break of 4.2 ± 0.6%.
2.1.2. Flax fibers
The short flax fibers (Linum usitatissimum) FIBRA-S®6A used for
this study were provided by Fibres Recherches Developpement®
(Troyes, France). According to the technical datasheet [39], fiber
bundles were 6 mm long with a diameter of 260 ± 150 mm and their
density was between 1.4 and 1.5 g/cm3. The Young's modulus of
bundles was 36 ± 13 GPa, maximum stress was 750 ± 490 MPa and
strain at break was 3.0 ± 1.9%.
2.2. Materials and techniques
2.2.1. Processing conditions
Several fiber weight contents were used: 0% (neat PLA) hereafter
named PLA, 10% hereafter named PLA-F10, and 30% hereafter
named PLA-F30. Polylactic acid granules were dried at 80 C for 24
h and flax fibers were vacuum dried at 120 C for 4 h. Composite
granules were obtained with a corotative twin-screw extruder
(Clextral BC21, screw length ¼ 900 mm; temperature profile along
the screw and at the die ¼ 180 C). After a second drying step under
vacuum at 80 C during 24 h, compounded granules were molded
with an injection molding machine (Krauss Maffei KM50-180CX)
into dog-bone samples according to the standard ISO 527-2 1BA.
The temperature profile was increasing up to 200 C and the mold
was kept at 25 C. After processing, samples were stored at room
temperature and 2%rh (relative humidity) before characterization
or aging. This equilibrium state was considered as the reference for
evaluating the effects of aging on the materials [30].
2.2.2. Size exclusion chromatography
The molecular mass of PLA was evaluated by size exclusion
chromatography (SEC) with Optilab® rEX™ of Wyatt Technology
(CIRAD, UMR 1208 (IATE), Universite de Montpellier, France). 90 mg
of each aged material was diluted in tetrahydrofuran stabilized
with butylated hydroxytoluene, and then kept at 30 C during 40 h
in a water bath. After a 0.45 mm-filtration, each solution was
injected in the column for measurement. The reproducibility was
evaluated on 3 samples.
2.2.3. Mechanical characterization
In order to assess the long term behavior of materials in real use
conditions, tensile mechanical loadings were applied. One can
consider two types of long-term loading: fatigue [40] and creep
tests [41]. The latter was chosen for its ease of interpretation but
also to avoid any dynamic effect on the behavior of materials.
The machine used for these tests was a Dartec model 100 kN
monitored by a Tema Concept® control system. A specific set-up
including a sealed polycarbonate tank was designed to allow
these tests to be performed at constant relative humidity as well as
underwater (cf. Fig. 1):
i. For atmospheric tests, its top was covered with an elastomeric
film to avoid air transfer between the enclosure and the room. In
order to regulate relative humidity, the bottom of the tank was
filled with silica gels. Temperature was monitored with a
thermo-regulated circulator (Julabo CF-31) and water circulated
in a flexible hose coiled inside the tank.
ii. For immersed tests, the tank was filled with distilled water and
directly connected to a thermo-regulated circulator (Julabo CF-
31).
The axial displacement was measured with a linear variable
differential transformer (LVDT) fixed on the upper clamp above the
tank. Its core was attached to a rod linked to the lower clamp. The
measurements resolution were 1 mm for the LVDT and 0.1 N for the
50 daN force sensor.
Given that the ultimate stress of PLA was around 65 MPa (cf.
2.1.1), the creep tests were carried out at a constant stress of 10 MPa
during 144 h or until failure of the sample. Their stiffness was
monitored during this aging by performing partial unloading-
reloading cycles on the samples every 30 min (ramps being set at
50 N/s). The elastic modulus was determined during unloading
LVDT
sensor
Sample
To load cell
Tank filled with
thermo-regulated
water
Clamps
Fig. 1. The set-up used for environmental creep tests.
from the slope of the stress-strain curve between 10 and 5 MPa.
The duration of the aging tests led to limit the assessment of the
reproducibility of the measurements to only one set of aging con-
ditions (i.e. for PLA-F10 in immersion at 20 C). The total strain and
the elastic modulus of two samples of flax/PLA composites during
hydro-mechanical aging are presented in Fig. 2. Creep strain and
elastic modulus turned out to be quite reproducible, the relative
deviation being less than 10% in term of total strain and 2% in term
of elastic modulus. Finally, the lifetime of samples differed only
from 6%.
3. Results
3.1. Thermo-hydric aging
Prior
to any assessment of
the coupled thermo-hydro-
mechanical aging, it was necessary to understand the sole influ-
ence of water on elastic properties in relation with aging temper-
ature. The set-up described in 2.2.3 was used in its hydrothermal
configuration. The morphology of PLA was assessed during aging by
evaluating its number average molar mass by size exclusion chro-
matography [30]. Fig. 3 shows the number average molar mass of
PLA for each material and at each aging temperature before and
after 24 h and 144 h of aging.
The initial molar mass of materials slightly decreased with the
fiber content because of the natural presence of water in flax fibers
despite drying, resulting in a more significant hydrolysis during
processing. During aging, the loss of PLA molar mass caused by
]
%
[
n
i
a
r
t
s
l
a
t
o
T
1.2%
1.0%
0.8%
0.6%
0.4%
0.2%
0.0%
4.8
4.5
4.2
3.9
3.6
3.3
3.0
]
a
P
G
[
s
u
l
u
d
o
m
c
i
t
s
a
l
E
0
25
50
75
Time [h]
100
125
hydrolysis was globally limited and concerned only composites. For
PLA-F10, the decrease only occurred at 50 C. For PLA-F30, a sig-
nificant decrease was observed for all temperatures during the first
24 h, but after this period the drop stopped at 20 and 35 C.
Fig. 4 displays the evolution of the in situ elastic modulus during
immersion in water for each material and each aging temperature.
Concerning initial moduli, as expected, results showed a significant
improvement with fiber content (þ17% and þ97% for an addition of
10 and 30%wt, respectively). During thermo-hydric aging, the
modulus of every material decreased. This decrease was more
significant as the fiber content increased. But the bath temperature
turned out to be the most decisive factor responsible for the drop of
modulus for all materials. At 50 C, rigidity became negligible for all
materials after only a few hours. Due to this drastic change of
behavior, this temperature was not suitable for thermo-hydro-
mechanical aging since any test would end up after only few mi-
nutes. Consequently, mechanical aging was only conducted at 20
and 35 C.
3.2. Thermo-mechanical aging
In this section, the study is focused on the influence of the sole
mechanical loadings. The set-up described in 2.2.3 was used in the
atmospheric configuration for the following discussion. Fig. 5 pre-
sents the total strain of composites during creep tests at 10 MPa in
dry air at 20 and 35 C. For each test, the stress-strain response
exhibited a quick primary creep (corresponding to the decrease of
the creep rate) and an extended secondary creep (where the creep
rate remains constant). Neither break nor tertiary creep (corre-
sponding to an increase of the creep rate) was observed on any
samples.
Strain rate was influenced by both temperature and fiber con-
tent. Indeed, a lower fiber content resulted in a higher strain rate.
However, increasing the temperature resulted in a more significant
increase of the strain rate. Regarding the impact on the elastic
modulus, Fig. 6 shows that no change occurred for all materials at
20 C. However, at 35 C, it decreased slightly and even more as
fiber content was increased.
As a result, based on the absence of both tertiary creep and
modulus change at 20 C, one can assume that no damage was
induced on materials when applying a 10 MPa stress at this tem-
perature. However, at 35 C, the damage on elastic behavior was
real but still limited.
3.3. Thermo-hydro-mechanical aging
After studying the influence of thermo-hydric and thermo-
r
a
l
o
m
e
g
a
r
e
v
a
r
e
b
m
u
N
]
l
o
m
/
g
k
[
s
s
a
m
100
80
60
40
20
0
0
25
50
75
100
125
Time [h]
20°C
35°C
50°C
PLA
PLA-F10
PLA-F30
Fig. 2. Evaluation of the reproducibility of the total strain (continuous lines) and the in
situ elastic modulus (dash-dotted lines) for PLA-F10 during creep tests at 10 MPa in
water.
Fig. 3. Number average molar mass of PLA in flax/PLA composites depending on im-
mersion time in water and temperature.
]
a
P
G
l
[
s
u
u
d
o
m
c
i
t
s
a
l
E
8
7
6
5
4
3
2
1
0
0
25
50
75
Time [h]
100
125
Fig. 4. In situ elastic modulus of flax/PLA composites depending on immersion time in
water and temperature.
]
%
[
n
i
a
r
t
s
l
a
t
o
T
0.8%
0.7%
0.6%
0.5%
0.4%
0.3%
0.2%
0.1%
0.0%
0
25
50
75
100
125
Time [h]
Fig. 5. Total strain of composites flax/PLA during creep tests at 10 MPa in dry air (10%
rh); the legend is the same as in Fig. 3.
mechanical aging separately, it was finally possible to apprehend
the coupled thermo-hydro-mechanical aging of flax/PLA compos-
ites. The set-up described in 2.2.3 was used in its immersed
configuration.
3.3.1. Strains
Fig. 7 highlights the differences of the creep responses at 20 C
measured in the presence or the absence of water (i.e. the differ-
ence between mechanical and hydro-mechanical aging). For all
]
a
P
G
[
s
u
l
u
d
o
m
c
i
t
s
a
l
E
8
7
6
5
4
3
2
0
25
50
75
100
125
Time [h]
Fig. 6. In situ elastic modulus of composites flax/PLA during creep tests at 10 MPa in
dry air (10%rh); the legend is the same as in Fig. 3.
materials, the presence of water resulted in a drastic increase of the
strain rate as a result of plasticizing.
Besides, the lifetime of materials subjected to coupled aging
turned out to be shortened compared to tests carried out in dry air.
As shown in Table 1, virgin PLA broke after about one day of im-
mersion, although diffusion was still incomplete [30]. PLA-F10
withstood barely 5 days of hydro-mechanical aging before failure.
However, no failure was recorded for PLA-F30 before ending the
experiments. Consequently, lifetime was strongly dependent of the
fiber content and it increased drastically with fiber content. How-
ever, no clear trend could be established regarding the influence of
the fiber content on strain rate. Neat PLA was the only material
which exhibited tertiary creep before failure. A visual inspection of
the samples showed by transparency the occurrence of crazes
oriented transversely to the loading direction (c.f. Fig. 8). Their
distribution seemed homogeneous along the length of the samples,
but, during tertiary creep, their concentration increased in the vi-
cinity of the failure area.
At 35 C, the observations were different. Fig. 9 shows the creep
response of samples immersed in water at 35 C. A drastic increase
of the strain rate was exhibited for samples which underwent
hydro-mechanical aging. The total strain exceeded 10%. During all
creep steps, lower fiber contents exhibited faster strain rates. Sec-
ondary creep was very short, even nonexistent, and gave way to an
extensive and strong tertiary creep. As shown in Table 2, the life-
time of samples was increased compared to samples immersed at
20 C. Indeed, PLA only failed after 3 days of thermo-hydro-
mechanical aging which was about 3 times longer than PLA in
water at 20 C. Finally, composites aged at 35 C did not break in the
duration of the experiments.
3.3.2. Elastic modulus
The evolution of elastic modulus was also assessed during these
experiments. In the first place, the case of the PLA-F10 at 20 C can
be taken as an example. Fig. 10 shows separately the influence of
the creep stress, of the presence of water, and of the coupling of
both these aging factors on the evolution of the elastic modulus of
PLA-F10 at 20 C. As mentioned in 3.2, the influence of the sole
creep stress was minor whereas water had a much larger impact on
elastic modulus. When aging factors were coupled, the modulus
loss was obviously greater than the losses induced by each aging
factor taken separately but also greater than their addition. The
difference between this addition and the actual loss was the evi-
dence of what we call a synergistic degradation.
In the purpose of an easier understanding, only modulus losses
]
%
[
n
i
a
r
t
s
l
a
t
o
T
1.2%
1.0%
0.8%
0.6%
0.4%
0.2%
0.0%
0
25
50
75
Time [h]
100
125
Fig. 7. Total strain of composites flax/PLA during creep tests at 10 MPa and 20 C, in
dry air (dotted lines) and in water (continuous lines); the legend is the same as in
Fig. 3.
Table 1
In situ modulus loss and time to failure of composites flax/PLA caused by the different aging factors during thermo-hydro-mechanical aging at 20 C.
Material
PLA
PLA-F10
PLA-F30
Creep
1.3%
3.8%
1.3%
Immersion
5.0%
10.7%
31.9%
Coupled creep and immersion
27.4%
25.5%
38.3%
Synergistic degradation
21.1%
11.0%
5.1%
Failure
20 h
115 h
none
]
a
P
G
[
s
u
l
u
d
o
m
c
i
t
s
a
l
E
4.6
4.4
4.2
4.0
3.8
3.6
3.4
3.2
Creep
Immersion
Synergistic degradation
Coupled
creep
and im-
mersion
0
25
50
75
Time [h]
100
125
Fig. 8. Apparition of crazes into a sample of PLA exposed to a creep stress of 10 MPa in
immersion at 20 C.
Fig. 10. In situ elastic modulus of PLA-F10 at 20 C depending on immersion time in
water (dashed line), and during creep tests at 10 MPa in dry air (dotted line) and in
water (continuous line).
]
%
[
n
i
a
r
t
s
l
a
t
o
T
8%
7%
6%
5%
4%
3%
2%
1%
0%
0
25
50
75
Time [h]
100
125
Fig. 9. Total strain of composites flax/PLA during creep tests at 10 MPa and 35 C, in
dry air (dotted lines) and in water (continuous lines); the legend is the same as in
Fig. 3.
are reported in Tables 1 and 2. They were evaluated at failure, or at
the end of the experiment (144 h) if failure did not occur. Globally,
the modulus drop increased with both fiber content and temper-
ature. Finally, the synergistic degradation seemed to be relatively
independent of the aging temperature. However,
it decreased
significantly with fiber content.
4. Discussion
It has been shown that the evolution of the elastic modulus
during thermo-hydric aging was the consequence of several
simultaneous phenomena which were reversible or irreversible
[30]. At 20 C, the only occurring phenomenon was plasticizing due
to the physical presence of water. Consequently, the limited evo-
lution of the PLA modulus at low aging temperature indicated that
matrix plasticizing was relatively limited. However, the significant
plasticizing of fibers suggested in the literature [3,42,43] explained
the decrease of the elastic modulus with the fiber content. When
increasing aging temperature, both reversible and irreversible ef-
fects grew. At 50 C, the proximity of the PLA glass transition
temperature combined with plasticizing (which lowers the tem-
perature of this transition [44]) were both responsible for a loss of
interaction between macromolecular chains making materials
malleable and viscous. The proximity of the PLA glass transition
temperature was also responsible for the hydrolysis of PLA as
shown by the decrease of its number average molar mass.
Several results lend credence to the hypothesis that the creep
behavior of composites was mostly influenced by the matrix
behavior. Firstly, the change in strain rate at 20 C was found to be
rather limited when varying the fiber content. The significant ac-
celeration of the strain rate at 35 C corroborated this assumption
on account of the proximity of the PLA glass transition temperature.
Besides, the strain rate of hemp fibers turns out to be particularly
low [35]. Consequently, assuming the creep behavior of flax and
hemp fibers is similar, it could explain why the strain rate of
composites decreased with the fiber content (particularly at 35 C).
In the meantime, the results in Fig. 6 showed that the decrease
of the elastic modulus was independent from the strain rate.
Indeed, the higher strain rate of PLA at 35 C compared to its
composites was not concomitant to a more significant decrease of
its elastic modulus. Since most of the decrease of the elastic
Table 2
In situ modulus loss and time to failure of composites flax/PLA caused by the different aging factors during thermo-hydro-mechanical aging at 35 C.
Material
PLA
PLA-F10
PLA-F30
Creep
4.2%
10.1%
18.0%
Immersion
17.1%
25.1%
50.7%
Coupled creep and immersion
43.4%
47.7%
68.8%
Synergistic degradation
22.1%
12.5%
0.1%
Failure
75 h
none
none
modulus of composites cannot be attributed to the behavior of PLA,
it originated either from the behavior of fibers themselves or from
the degradation of their interface with PLA. But at 10 MPa (far
below the tensile strength), loading and unloading of flax fibers did
not induce damage regarding the elastic modulus. On the contrary,
fibrils reorientations tend to result in an increase of the elastic
modulus [45]. As a result, the damage responsible for the decrease
of elastic modulus was most probably occurring at the fiber/matrix
interface.
During thermo-hydro-mechanical aging, the creep behavior in
water was very different from the one in air. The increase of strain
rate was attributed to the plasticizing of both matrix and fibers.
However, the sudden growth of the crazes amount in the polymer
matrix during tertiary creep was probably the reason of the ac-
celeration of the strain rate. The high concentration of crazes in the
vicinity of the fracture is supposed to be responsible for the brittle
failure of samples during this coupled aging. Nevertheless, the
presence of fibers prevented an important local concentration of
crazes, thus limiting the amplitude of tertiary creep and especially
extending the lifetime of samples. Besides, results showed that
increasing aging temperature allowed extending lifetime during
hydro-mechanical aging. It is most likely that this phenomenon was
linked to the plasticizing of composite components by both water
and temperature. Its main effect was a more ductile behavior of the
material, making it more accommodating to large deformations.
The possible increase of the strain rate with the fibre content in
water at 20 C observed in Fig. 7 could be attributed to the greater
sensitivity of the flax fibre modulus to the presence of water at this
temperature [42]. This could also result from the lower PLA molar
mass in PLA-F30 samples compared to PLA-F10 samples as shown
in Fig. 3.
Finally, the synergistic degradation reported in Tables 1 and 2
was attributed to environmental stress cracking for several rea-
sons [46]. Firstly, the appearance of crazes on the polymer was
typical of this phenomenon. Secondly, the fragile failure of mate-
rials was also characteristic, and excluded any process caused by
plasticizing. Then, this phenomenon usually affects amorphous
polymers. Indeed PLA was mostly amorphous at this stage of aging
[30]. Finally, the highly probable orientation of polymer chains in
the direction of the stress due to the sample forming process was
consistent with the mechanisms of environmental stress cracking
[47].
Regardless to the prevailing mechanism responsible for the
synergistic degradation, it was considerably limited by the pres-
ence of fibers, probably because they postponed crack propagation
into the matrix. Indeed, neat PLA turned out to be very sensitive to
this phenomenon, and even more when temperature was
decreased. Its fragile behavior at low temperature contributed to
this phenomenon, while it became less sensitive close to the glass
transition due to the raise of its ductility.
5. Conclusions
This study examined the influence of thermo-hydro-mechanical
aging on elastic and creep behaviors of poly(lactic acid) (PLA)/flax
fibers composites.
Firstly, the sole influence of water was assessed depending on
temperature. This thermo-hydric aging turned out to have a major
impact on the mechanical properties. The occurrence of both
reversible and irreversible phenomena explained the loss of rigidity
of these composites. In this case, the elastic behavior was driven by
the sensitivity of flax fibers to this type of aging. However, close to
its glass transition temperature (about 56 C), PLA lost its rigidity,
making it useless in these conditions for any structural application.
Then the sole influence of a constant stress was evaluated
depending on temperature. A creep stress of 10 MPa induced
limited damage on all materials in a dry environment, especially
compared to thermo-hydric aging. However, the little damage
observed mostly on composites was attributed to fiber/matrix
interface degradation.
Finally, the coupling of these aging factors was assessed. In this
case, hydro-mechanical aging induced a synergy attributed to
environmental stress cracking. This effect was characterized by a
decrease of the elastic modulus which was more significant than
the addition of the combined effects of the two aging factors taken
separately. Both temperature and fiber content tended to induce a
more significant decrease of the elastic modulus, yet they resulted
in extended time to failure. The increase in ductility caused by
temperature and the limitation of crack propagation resulting from
the presence of fibers were responsible for these observations. As a
result, fibers played a predominant role in the durability of com-
posites by limiting environmental stress cracking.
References
[1] F. Vilaplana, E. Str€omberg, S. Karlsson, Environmental and resource aspects of
sustainable biocomposites, Polym. Degrad. Stab. 95 (2010) 2147e2161, http://
dx.doi.org/10.1016/j.polymdegradstab.2010.07.016.
[2] D.B. Dittenber, H.V.S. GangaRao, Critical review of recent publications on use
of natural composites in infrastructure, Compos. Part A Appl. Sci. Manuf. 43
(2012) 1419e1429, http://dx.doi.org/10.1016/j.compositesa.2011.11.019.
[3] E. Bodros, I. Pillin, N. Montrelay, C. Baley, Could biopolymers reinforced by
randomly scattered flax fibre be used in structural applications? Compos. Sci.
Technol.
http://dx.doi.org/10.1016/
j.compscitech.2006.08.024.
462e470,
(2007)
67
[4] K.G. Satyanarayana, G.G.C. Arizaga, F. Wypych, Biodegradable composites
based on lignocellulosic fibers e an overview, Prog. Polym. Sci. 34 (2009)
982e1021, http://dx.doi.org/10.1016/j.progpolymsci.2008.12.002.
[5] B.A. Acha, M.M. Reboredo, N.E. Marcovich, Creep and dynamic mechanical
behavior of PP-jute composites: effect of the interfacial adhesion, Compos.
Part A Appl. Sci. Manuf. 38 (2007) 1507e1516, http://dx.doi.org/10.1016/
j.compositesa.2007.01.003.
[6] Y.J. Weitsman, M. Elahi, Effects of fluids on the deformation, strength and
durability of polymeric composites e an overview, Mech. Time Dependent
Mater 4 (2000) 107e126, http://dx.doi.org/10.1023/A:1009838128526.
[7] C. Suri, D. Perreux, The effects of mechanical damage in a glass fibre/epoxy
composite on the absorption rate, Compos. Eng. 5 (1995) 415e424, http://
dx.doi.org/10.1016/0961-9526(94)00014-Z.
[8] A.-L. Durier, Contribution a l'etude de l'interaction contraintes-diffusion dans
les polymeres, Ecole Arts et Metiers ParisTech, 2008. http://pastel.archives-
ouvertes.fr/pastel-00003698.
[9] F. Jacquemin, S. Freour, R. Guillen, Prediction of local hygroscopic stresses for
composite structures e analytical and numerical micro-mechanical ap-
proaches, Compos. Sci. Technol. 69 (2009) 17e21, http://dx.doi.org/10.1016/
j.compscitech.2007.10.031.
[10] P. Davies, G. Germain, B. Gaurier, A. Boisseau, D. Perreux, Evaluation of the
durability of composite tidal turbine blades, Philos. Trans. A. Math. Phys. Eng.
Sci. 371 (2013) 1e15, http://dx.doi.org/10.1098/rsta.2012.0187.
[11] A. Chateauminois, B. Chabert, J.P. Soulier, L. Vincent, Hygrothermal ageing
effects on the static fatigue of glass/epoxy composites, Composites 24 (1993)
547e555, http://dx.doi.org/10.1016/0010-4361(93)90268-D.
[12] E. Vauthier, J.C. Abry, T. Bailliez, A. Chateauminois, Interactions between
hygrothermal ageing and fatigue damage in unidirectional glass/epoxy com-
posites, Compos. Sci. Technol. 58 (1998) 687e692, http://dx.doi.org/10.1016/
S0266-3538(97)00149-8.
[13] J. Mercier, A. Bunsell, P. Castaing, J. Renard, Characterisation and modelling of
aging of composites, Compos. Part A Appl. Sci. Manuf. 39 (2008) 428e438,
http://dx.doi.org/10.1016/j.compositesa.2007.08.015.
[14] R. Leger, Contribution a l'etude de la durabilite des structures collees soumises
a des efforts mecaniques apres vieillissement humide, Ecole Nationale
Superieure de Mecanique et d'Aerotechnique, 2010. http://tel.archives-
ouvertes.fr/tel-00518367/.
[15] A. Boisseau, P. Davies, F. Thiebaud, Fatigue behaviour of glass fibre reinforced
composites for ocean energy conversion systems, Appl. Compos. Mater. 20
(2012) 145e155, http://dx.doi.org/10.1007/s10443-012-9260-0.
[16] Y.J. Weitsman, Fluid Effects in Polymers and Polymeric Composites, Springer,
US, Boston, MA, 2012, http://dx.doi.org/10.1007/978-1-4614-1059-1.
[17] R. Martin, Ageing of Composites, Woodhead P, Taylor & Francis, 2008.
[18] D. Scida, M. Assarar, C. Poil^ane, R. Ayad, Influence of hygrothermal ageing on
the damage mechanisms of flax-fibre reinforced epoxy composite, Compos.
Part
http://dx.doi.org/10.1016/
j.compositesb.2012.12.010.
51e58,
(2013)
Eng.
[19] H. Dhakal, Z. Zhang, M. Richardson, Effect of water absorption on the me-
chanical properties of hemp fibre reinforced unsaturated polyester
B
48
composites, Compos. Sci. Technol. 67 (2007) 1674e1683, http://dx.doi.org/
10.1016/j.compscitech.2006.06.019.
[20] A. Roudier, Analyse multi-echelle du comportement hygro-mecanique des
fibres de lin, Ecole Doctorale Sciences pour l'Ingenieur de Clermont-Ferrand,
2012.
[21] S. Alix, L. Lebrun, C. Morvan, S. Marais, Study of water behaviour of chemically
treated flax fibres-based composites: a way to approach the hydric interface,
Compos. Sci. Technol. 71 (2011) 893e899, http://dx.doi.org/10.1016/
j.compscitech.2011.02.004.
[22] P.A. Sreekumar, J.-M. Saiter, S. Thomas, Water absorption and ageing behav-
iour of natural fiber composites, in: Nat. Fibre Reinf. Polym. Compos. From
Macro to Nanoscale, Archives Contemporaines, 2009, pp. 446e472. http://
www.oldcitypublishing.com/servlet/Detail?no¼584 (accessed 7.02.2011).
[23] N. Nosbi, H.M. Akil, Z.A. Mohd Ishak, A. Abu Bakar, Degradation of compres-
sive properties of pultruded kenaf fiber reinforced composites after immer-
sion in various solutions, Mater. Des. 31 (2010) 4960e4964, http://dx.doi.org/
10.1016/j.matdes.2010.04.037.
[24] V. Placet,
Influence de traitements hygrothermiques sur les proprietes
mecaniques de composites a fibres vegetales, Comptes Rendus Des JNC 16,
Toulouse, 2009.
[25] A. Le Duigou, P. Davies, C. Baley, Seawater ageing of flax/poly(lactic acid)
biocomposites, Polym. Degrad. Stab. 94 (2009) 1151e1162, http://dx.doi.org/
10.1016/j.polymdegradstab.2009.03.025.
[26] Z.N. Azwa, B.F. Yousif, A.C. Manalo, W. Karunasena, A review on the degrad-
ability of polymeric composites based on natural fibres, Mater. Des. 47 (2013)
424e442, http://dx.doi.org/10.1016/j.matdes.2012.11.025.
[27] P.V. Joseph, M.S. Rabello, L.H. Mattoso, K. Joseph, S. Thomas, Environmental
effects on the degradation behaviour of sisal fibre reinforced polypropylene
composites, Compos. Sci. Technol. 62 (2002) 1357e1372, http://dx.doi.org/
10.1016/S0266-3538(02)00080-5.
[28] K. Oksman, Natural fibres as reinforcement in polylactic acid (PLA) compos-
ites, Compos. Sci. Technol. 63 (2003) 1317e1324, http://dx.doi.org/10.1016/
S0266-3538(03)00103-9.
[29] B. Bax, J. Müssig, Impact and tensile properties of PLA/Cordenka and PLA/flax
composites, Compos. Sci. Technol. 68 (2008) 1601e1607, http://dx.doi.org/
10.1016/j.compscitech.2008.01.004.
[30] A. Regazzi, S. Corn, P. Ienny, J.-C. Benezet, A. Bergeret, Reversible and irre-
versible changes in physical and mechanical properties of biocomposites
during hydrothermal aging, Ind. Crops Prod. 84 (2016) 358e365, http://
dx.doi.org/10.1016/j.indcrop.2016.01.052.
[31] A.K. Bledzki, A. Jaszkiewicz, Mechanical performance of biocomposites based
on PLA and PHBV reinforced with natural fibres e a comparative study to PP,
Compos. Sci. Technol. 70 (2010) 1687e1696, http://dx.doi.org/10.1016/
j.compscitech.2010.06.005.
[32] C.S. Proikakis, N.J. Mamouzelos, P.A. Tarantili, A.G. Andreopoulos, Swelling and
hydrolytic degradation of poly(D,L-lactic acid) in aqueous solutions, Polym.
Degrad.
http://dx.doi.org/10.1016/
614e619,
(2006)
Stab.
91
j.polymdegradstab.2005.01.060.
[33] J.D. Badia, L. Santonja-Blasco, A. Martínez-Felipe, A. Ribes-Greus, Hygro-
thermal ageing of reprocessed polylactide, Polym. Degrad. Stab. 97 (2012)
1881e1890, http://dx.doi.org/10.1016/j.polymdegradstab.2012.06.001.
[34] C. Baley, A. Le Duigou, A. Bourmaud, P. Davies, Influence of drying on the
mechanical behaviour of flax fibres and their unidirectional composites,
Compos. Part A Appl. Sci. Manuf. 43 (2012) 1226e1233, http://dx.doi.org/
10.1016/j.compositesa.2012.03.005.
[35] V. Placet, Characterization of the thermo-mechanical behaviour of hemp fi-
bres intended for the manufacturing of high performance composites, Com-
pos. Part A Appl. Sci. Manuf. 40 (2009) 1111e1118, http://dx.doi.org/10.1016/
j.compositesa.2009.04.031.
[36] M.S. Islam, K.L. Pickering, N.J. Foreman, Influence of accelerated ageing on the
physico-mechanical properties of alkali-treated industrial hemp fibre rein-
forced poly(lactic acid) (PLA) composites, Polym. Degrad. Stab. 95 (2010)
59e65, http://dx.doi.org/10.1016/j.polymdegradstab.2009.10.010.
[37] M.J.A. van den Oever, B. Beck, J. Müssig, Agrofibre reinforced poly(lactic acid)
composites: effect of moisture on degradation and mechanical properties,
Compos. Part A Appl. Sci. Manuf. 41 (2010) 1628e1635, http://dx.doi.org/
10.1016/j.compositesa.2010.07.011.
[38] NatureWorks, NatureWorks PLA Polymer 7000D, Datasheet, 2005, pp. 1e3.
[39] FRD, Fibres et preformes vegetales e Solutions composites, Datasheet, 2011.
www.f-r-d.fr.
[40] P. Rabbe, H.-P. Lieurade, A. Galtier, Essais de fatigue Partie I, Tech. L'Ingenieur.
M, 2000, pp. 1e22.
[41] F. Saint-Antonin, Essais de fluage, Tech. L'Ingenieur. M, 1995, pp. 1e14.
[42] A. Roudier, K. Charlet, F. Moreno, E. Toussaint, C. Geneau-Sbartaï,
S. Commereuc, et al., Caracterisation des proprietes biochimiques et hygro-
scopiques d'une fibre de lin, Materiaux Tech. 100 (2012) 525e535, http://
dx.doi.org/10.1051/mattech/2012044.
[43] R. Joffe, J. Andersons, L. Wallstr€om, Strength and adhesion characteristics of
elementary flax fibres with different surface treatments, Compos. Part A Appl.
Sci. Manuf. 34 (2003) 603e612, http://dx.doi.org/10.1016/S1359-835X(03)
00099-X.
[44] K.V. Pochiraju, G.P. Tandon, G.A. Schoeppner (Eds.), Long-Term Durability of
Polymeric Matrix Composites, Springer US, Boston, MA, 2012, http://
dx.doi.org/10.1007/978-1-4419-9308-3.
[45] C. Baley, Analysis of the flax fibres tensile behaviour and analysis of the tensile
stiffness increase, Compos. Part A Appl. Sci. Manuf. 33 (2002) 939e948, http://
dx.doi.org/10.1016/S1359-835X(02)00040-4.
[46] J. Arnold, The influence of liquid uptake on environmental stress cracking of
glassy polymers, Mater. Sci. Eng. A 197 (1995) 119e124, http://dx.doi.org/
10.1016/0921-5093(94)09759-3.
[47] A. Lustiger, Understanding environmental stress cracking in polyethylene, in:
R.C. Portnoy (Ed.), Med. Plast. Degrad. Resist. Fail. Anal., Plastics D, William
Andrew, Norwich, NY, 1998, pp. 65e71, http://dx.doi.org/10.1016/B978-
188420760-0.50010-7.
|
1805.00292 | 1 | 1805 | 2018-05-01T12:43:22 | Nanopore fabrication and characterization by helium ion microscopy | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | The Helium Ion Microscope (HIM) has the capability to image small features with a resolution down to 0.35 nm due to its highly focused gas field ionization source and its small beam-sample interaction volume. In this work, the focused helium ion beam of a HIM is utilized to create nanopores with diameters down to 1.3 nm. It will be demonstrated that nanopores can be milled into silicon nitride, carbon nanomembranes (CNMs) and graphene with well-defined aspect ratio. To image and characterize the produced nanopores, helium ion microscopy and high resolution scanning transmission electron microscopy were used. The analysis of the nanopore's growth behavior, allows inferring on the profile of the helium ion beam. | physics.app-ph | physics | Nanopore fabrication and characterization by helium ion microscopy
D. Emmrich1, A. Beyer1, A. Nadzeyka2, S. Bauerdick2, J. C. Meyer3, J.
Kotakoski3, A. Gölzhäuser1
1Physics of Supramolecular Systems and Surfaces, Bielefeld University, 33615 Bielefeld, Germany
2Raith GmbH, Konrad-Adenauer-Allee 8, 44263 Dortmund, Germany
3Faculty of Physics, University of Vienna, 1090 Vienna, Austria
Abstract:
The Helium Ion Microscope (HIM) has the capability to image small features with a resolution down
to 0.35 nm due to its highly focused gas field ionization source and its small beam-sample interaction
volume. In this work, the focused helium ion beam of a HIM is utilized to create nanopores with
diameters down to 1.3 nm. It will be demonstrated that nanopores can be milled into silicon nitride,
carbon nanomembranes (CNMs) and graphene with well-defined aspect ratio. To image and
characterize the produced nanopores, helium ion microscopy and high resolution scanning
transmission electron microscopy were used. The analysis of the nanopores' growth behavior, allows
inferring on the profile of the helium ion beam.
Nanopores in atomically thin membranes can be used for biomolecule analysis,1 electrochemical
storage,2 as well as for the separation of gases and liquids.3 All of these applications require a precise
control of the size and shape of the nanopores. It was shown that the focused beam of a transmission
electron microscope (TEM) is able to create nanopores in membranes of silicon nitride and graphene
with diameters down to 2 nm.4,5 Pores can be further shrunk in a TEM by areal electron impact.6
However, the preparation of such nanopores in a TEM is time-consuming and is limited to small
samples (~3 mm diameter) that fit into the microscope. Focused ion beam tools (FIB) offer more
flexibility concerning the sample size and a higher milling speed. Among these FIB tools gallium
liquid metal ion sources (LMIS) are widely used, allowing minimum sizes of 3 nm diameter for
nanopores.7 The development of a reliable gas field ionization source (GFIS) type allowed the
construction of the helium ion microscope which surpasses the imaging and milling resolution of
LMIS-based FIB tools.8,9 First studies about milling with helium ions reported sample damage by
amorphization and helium implantation during milling on bulk substrates.10 The latter effect is absent
on membranes, where nanopores with diameters of 2.6 nm were milled by HIM.11 In all these reports,
pores were created by single spot exposures. Here we present a different route to create small
nanopores in membranes by milling circular patterns. Furthermore we are able to connect the growth
of nanopores to the ion beam profile.
1
The work was performed with a Zeiss ORION Plus helium ion microscope, equipped with a Raith
ELPHY Multibeam pattern generator. Pores were typically milled at beam energies of 35-40 keV at a
working distance of 7-10 mm. Depending on the substrate and its cleanliness, the helium ion beam
current was set to values of 0.5-6.2 pA. Secondary electrons (SE) as well as transmitted ions were used
as signals for imaging. For the latter, a scanning transmission ion microscopy (STIM) extension was
utilized.12 For the high resolution imaging of nanopores, an ultra-high vacuum scanning transmission
electron microscope (Nion UltraSTEM 100) provided atomic resolution images at a beam energy of
60 keV, while minimizing sample damage.13 At the STEM the high-angle annular dark field signal
(HAADF) was chosen. As membranes we used 30 nm thick silicon nitride membranes (Silson Ltd),
single layer graphene sheets grown by CVD,14 and 1 nm thick carbon nanomembranes (CNM). CNMs
were made by electron irradiation of aromatic self-assembled monolayers leading to a two-dimensional
cross-linking and the formation of mechanically free-standing membranes.15-19
Figure 1: Three different nanopore arrays in silicon nitride imaged by HIM (transmission detection, STIM). (a)
Writing a pattern of circular areas in concentric inward (top row) and concentric outward direction (bottom row)
is accompanied by hydrocarbon deposition, depending on the writing path. (b) When repeating the pattern writing
of (a) at another membrane position after cleaning the sample, no difference between inward and outward direction
is observed. Note that (a) and (b) show fine details, but –unexpectedly- pores appear darker than the membrane.
(c) Nanopores imaged with a cleaned, i.e. hydrocarbon free, STIM detector, showing an –expected- contrast with
"bright" pores in a "dark" membrane.
Two approaches were made to mill nanopores: (i) keeping the beam at a constant position and
unblanking it for a desired exposure time and (ii) exposing the sample with a predefined pattern. Fig. 1
shows nanopore arrays in silicon nitride which were created by patterns of circular areas. The arrays
were then imaged with a HIM in the transmission (STIM) mode. At first sight, it is striking that
although Fig. 1 (a) and (b) are bright field transmission images, the milled holes appear darker than
the SiN membrane. This inversion of contrast is most likely due to hydrocarbons on the STIM detector.
This detector generates a secondary electron signal from the transmitted helium ions. In presence of a
2
high ion intensity, i.e. at openings in the SiN membrane, hydrocarbons on the detector can charge,
which deflects the (low energy) secondary electrons. Surprisingly, this disturbance happens in a way
that small features in the STIM images of nanopores, cf. Fig1a, b, can be better identified than with a
clean STIM detector. After cleaning the sample holder in an argon plasma at 10 W for 30 min, the
STIM detector operates "as expected" and the corresponding image shows "bright" open pores in a
"dark" SiN membrane, cf. Fig, 1c.
Milling circular areas facilitate to obtain perfectly round nanopores of different sizes. Furthermore the
membranes are thinner at the edges, creating pores with low aspect ratios. The milled pattern consists
of circles with decreasing diameters from 40 nm (left) down to 2 nm (right) and a final 40 nm circle
on the far right as a marker. The ion dose of 2.4 × 1018 ions/cm in Figure 1 (a) and (b) is too low to
mill the complete area, so that the 40 nm diameter in the exposure results in a 30 nm pore on the
sample. Our aim is to thin the membrane until a small hole is breaking off right in the middle of the
area. Figure 1a shows two lines of nanopores, patterned with the same ion dose but with different
writing strategies. In the upper line, the beam moves in concentric circles inwards, in the lower line
outwards. It can be seen that there is a clear difference between these writing directions. Writing
inwards leads to closed or partially closed pores while writing in outward direction leads to more
opened nanopores. The reason for this difference is the deposition of contaminant hydrocarbons on the
pores during the milling process. In Fig. 1a the sample is introduced into the specimen chamber of the
HIM after an air plasma treatment in the load lock for 8 min at 10 W, while in Fig. 1b the sample was
cleaned for 23 min and stayed an additional 12 h in vacuum (3 × 10-7 mbar). While in Fig. 1b the same
milling parameters were used as in Fig. 1a, the differences between the writing directions are not
visible. Hence, the direction dependence must be related to the presence of hydrocarbons as well as to
their surface diffusion on the sample. If the milling starts in the center of the pore, hydrocarbons are
not able to diffuse into the center while the beam proceeds outwards. This keeps the pore center free
of contamination and the milling rate is not reduced by hydrocarbon deposition. Conversely, if the
writing starts at the outer boundary of a feature, hydrocarbons diffuse with the inwards moving beam.
On clean samples nanopores of 5 nm in diameter can be built by this writing strategy. When the ion
dose is increased, the diameter of all pores increases and pores of diameters below 5 nm open at
locations that had been closed before. However, during HIM imaging these nanopores further grow as
the very thin material at their thin edge is sputtered away faster than material at the rim of larger high
aspect ratio pores. We therefore expect that smaller pores can be detected by utilizing e.g. TEM below
an electron energy of 120 keV which is known to be non-destructive on silicon nitride.1
A patterning strategy that employs the filling of geometrical objects thus provides reasonable results
for a range of pore sizes and aspect ratios. However, for very thin membranes the aspect ratio is
3
determined by the thickness of the membrane. In the case of graphene, taking control of the aspect
ratio becomes meaningless, as the thickness is fixed. The following section thus deals with the creation
of very small nanopores by spot exposures in carbon nanomembranes, graphene and silicon nitride. In
this case the focused ion beam is kept at a fixed position and the pore size is controlled by the exposure
time.
Figure 2 shows STEM micrographs of a nanopore array milled with the HIM into a CNM. The image
shows a typical dose array used for determining the smallest pores which are opening at the
breakthrough dose, just sufficient to form a hole. In case of Figure 2 the beam current was 1.4 pA and
the shortest exposure time was 5 ms at the first spot in the 10x10 array in the upper right corner of (a).
The exposure time increases from pore to pore by 5 ms from left to right in a single line, and by each
line from top to bottom so that the highest exposure time is 500 ms in the lower right corner. For the
STEM images the grey values of the HAADF signal are directly related to the thickness of the
membrane. It has been shown that the HAADF signal is very sensitive to hydrocarbon contamination.20
The CNM is an amorphous layer of cross-linked biphenyl molecules, and it also contributes to the
thickness variation seen in the images. Light patches in (a) are assigned to residues of a polymer layer
used in the transfer process.16 Figure 2b shows a magnified area of the array with pores from two
different lines. In horizontal direction the dose is increased in each line by the smallest step of 5 ms,
corresponding to 4.7 × 104 ions/point while the dose difference between the lines is a factor of ten
higher. The lower line shows what happens at doses high above the breakthrough dose. The upper line
shows the behavior near the breakthrough dose: At the left position the membrane was not thinned
while on the right position a pore was milled at a slightly increased dose. Even at lower dose on a
different position (not shown here) it was possible to sputter through the membrane. The behavior at
the breakthrough dose shows that slight differences in the membrane thickness determine whether or
not a pore can be created. Figure 2c shows the smallest pore in this array at atomic resolution. The
actual pore (see dashed circle) has a diameter of ~0.8 nm on the shorter axis and ~1.6 nm on the longer
axis while its surrounding was thinned to atomic mono- or bilayers by the ion beam. The symmetry of
the exposed area appears not circular. However, larger pores at longer exposure times indicate that the
ion beam has a circular symmetry. Hence, it appears as the local stability of the molecular membrane
influences the pore shape at very low doses.
Characteristic results for all investigated membranes are summarized in Table I. In all membranes
nanopores below 4 nm could be created and detected. HIM and STEM were used for the analysis of
the pore arrays. A comparison between HIM and STEM on both carbon based membranes shows that
the smallest detectable pore size is around 3 nm in HIM, whereas STEM provides atomic resolution.
4
The breakthrough dose correlates with the thickness of the membranes and is the lowest for graphene.
In contrary the silicon nitride membrane shows the highest sputter rate while graphene and CNM have
lower ones. We attribute this to the higher thickness of the silicon nitride which increases the
probability of a helium ion to hit a target atom. CNM and graphene show similar sputter rates, in
agreement with their carbon based composition and their thinness.
at the breakthrough. At the upper left position no pore is visible, at the upper right position the dose (𝟓.𝟔⋅𝟏𝟎𝟓
Figure 2: STEM images (HAADF signal) of nanopores in a CNM. (a) Overview of a spot exposure array. The dose
increases from the top left to the bottom right corner. Bright patches result from residual polymer from the transfer.
(b) The lower line pores received doses far beyond the breakthrough dose, while the upper line pores received doses
ions/point) is just sufficient to create a defect in the membrane that is highlighted by a dashed circle in (c).
Table I: Properties of the investigated membranes
Membrane
SiN
CNM
Graphene
Thickness Microscope
nm
Non-conductive 30
Non-conductive 1
Conductive
HIM
STEM
HIM
Smallest diameter Breakthrough dose
nm
3.8
1.3
2.9
ions/point
1.1 × 106
4.5 × 105
3.7 × 105
Sputter yield
0.005
0.001
0.001
0.34
In Figure 3 the pore diameters are plotted against the required ion doses. If the pores were not of
circular shape, the diameter was averaged from a perfect circle with the same pore area. For graphene
and silicon nitride membranes the curves were extracted from HIM images, STEM data were taken
for CNMs. The curves show a steep increase of the pore size which gradually slows with increasing
dose. The comparison of graphene and CNM shows the detection limits of the HIM, which obscures
identification of pores below 3 nm, although both curves show a similar growth behavior. For silicon
nitride we observe a slower pore size growth in agreement with the calculated sputter yield.
5
)
m
n
(
r
e
t
e
m
a
i
D
e
r
o
P
12
10
8
6
4
2
0
CNM
Graphene
SiN
0.0
2.0x106
4.0x106
6.0x106
Dose (Ions/Point)
8.0x106
1.0x107
Figure 3: Diameters of nanopores milled with helium ions into different membranes extracted from microscope
images. In the case of graphene and silicon nitride the sizes were determined on HIM images, for CNM STEM data
was used.
The growth behavior of nanopores is further related to the profile of the milling ion beam, i.e. its point
spread function (PSF). In resist based lithography with light and ions, it has been shown that the PSF
can be deduced from spot exposure arrays.21,22 We can adopt this concept for nanopore milling: When
the breakthrough dose is exceeded at a certain point in the PSF, a hole is created. The diameter of a
nanopore is therefore related to the width of the PSF at a certain height. Varying the exposure time
changes the amplitude of the PSF:21
Here 𝑟𝑛 is the radius of a pore, 𝑡𝑛 the exposure time for this pore and 𝑇𝐻 the breakthrough dose. The
𝑃𝑆𝐹(𝑟𝑛)×𝑡𝑛=𝑇𝐻 ⇒ 𝑃𝑆𝐹(𝑟𝑛)=𝑇𝐻/𝑡𝑛
PSF can be calculated by plotting the reciprocal exposure time (which works as a normalization factor)
against the pore radius. In our case we did the normalization not on the exposure time but on the ion
dose to compensate different beam currents on different samples as shown in Figure 4. Since the
membranes are ultimately thin, this measures the pristine ion beam profile without any contribution
from secondary electrons or backscattered ions that may broaden the beam.
Figure 4 (a) shows beam profiles for CNM and silicon nitride, derived from the data in Figure 3. In a
first approximation a Gaussian distribution is fitted to the data, having its maximum at 0 nm. The beam
currents in (a) are comparable for both data sets, meaning the beam profile is expected to be the similar.
In both cases the full width at half maximum (FWHM) gave identical results to the second decimal
place. We thus could deduce identical beam profiles from milling membranes of different composition,
sputter yield and thickness. We also found the dose of the HIM imaging to be sufficiently low to not
widen the pores, as the STEM analysis for CNMs and the HIM analysis for SiN show similar beam
profiles.
Figure 4 (b) compares beam profiles derived from arrays in graphene written at different beam
currents. To vary the beam current, adjustments on the condenser lens of the HIM were done while we
still use the 10 µm aperture. One can see that the lower beam current results in a much steeper slope
having a FWHM of 3.82 nm, while the higher beam current has a FWHM of 7.90 nm.
6
(a)
(b)
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
y
t
i
s
n
e
t
n
i
d
e
z
i
l
a
m
r
o
N
CNM
He+ current 1.5pA, STEM data
FWHM: 4.75 nm
SiN
He+ current 1.6pA, HIM data
FWHM: 4.75 nm
Graphene
He+ current 0.5pA, HIM data
FWHM: 3.82 nm
Graphene
He+ current 6.2pA, STEM data
FWHM: 7.90 nm
0
2
4
Pore radius (nm)
6
8
10
Figure 4: Beam profiles of the helium ion beam determined from nanopore arrays on different substrates. As a first
approximation a Gaussian distribution is assumed. The same beam setup at the HIM (a) leads to identical results
for the FWHM of 4.75 nm, even on different membranes. An increase of the beam current is at the expense of the
beam width (b).
In summary, we demonstrated two approaches of creating nanopores into membranes of different
material. With milling geometrical patterns we can vary the thickness of the pore's rim in silicon
nitride, while realizing pore diameters down to 5 nm. With spot exposures we are able to mill smaller
features and achieve pore diameters down to 1.3 nm. We evaluated the sputter yield for our membranes
and deduced the beam profile from the growth of the nanopores. Remarkably, the profile can be also
determined if we use the HIM for both milling and imaging. Milling well defined nanopores in the
demonstrated size range as well as the improved knowledge of the helium ion beam characteristic will
facilitate the application of such small pores in other scientific research areas like the analysis of
biomolecules with nanopore-based approaches.
Acknowledgements:
This work was financially supported by the German Bundesministerium für Wirtschaft. This work
was further conducted within the framework of the COST Action CM1301 (CELINA). The research
leading to these results has also received funding from the European Union Seventh Framework
Programme under grant agreement n°604391 Graphene Flagship. We acknowledge Wiener
Wissenschafts-, Forschungs- und Technologiefonds (WWTF) project MA14-009 and European
Research Council (ERC) Project No. 336453-PICOMAT.
7
1J. Yang, D.C. Ferranti, L.A. Stern, C.A. Sanford, J. Huang, Z. Ren, L.-C. Qin, and A.R. Hall,
2C. Liu, E.I. Gillette, X. Chen, A.J. Pearse, A.C. Kozen, M.A. Schroeder, K.E. Gregorczyk, S.B. Lee,
Nanotechnology 22, 285310 (2011).
and G.W. Rubloff, Nature nanotechnology 9, 1031 (2014).
H.G. Park, Science (New York, N.Y.) 344, 289 (2014).
(2010).
3K. Celebi, J. Buchheim, R.M. Wyss, A. Droudian, P. Gasser, I. Shorubalko, J.-I. Kye, C. Lee, and
4M.J. Kim, M. Wanunu, D.C. Bell, and A. Meller, Adv. Mater. 18, 3149 (2006).
5S. Garaj, W. Hubbard, A. Reina, J. Kong, D. Branton, and J.A. Golovchenko, Nature 467, 190
6A.J. Storm, J.H. Chen, X.S. Ling, H.W. Zandbergen, and C. Dekker, Nature materials 2, 537 (2003).
7J. Gierak, E. Bourhis, G. Faini, G. Patriarche, A. Madouri, R. Jede, L. Bruchhaus, S. Bauerdick, B.
8B.W. Ward, J.A. Notte, and N.P. Economou, J. Vac. Sci. Technol. B 24, 2871 (2006).
9G. Hlawacek, V. Veligura, R. van Gastel, and B. Poelsema, J. Vac. Sci. Technol. B 32, 20801
10R. Livengood, S. Tan, Y. Greenzweig, J. Notte, and S. McVey, J. Vac. Sci. Technol. B 27, 3244
Schiedt, A.L. Biance, and L. Auvray, Ultramicroscopy 109, 457 (2009).
(2014).
(2009).
11J. Buchheim, R. Wyss, I. Shorubalko, and H.G. Park, Nanoscale (2016).
12A.R. Hall, Microsc Microanal 19, 740 (2013).
13J.C. Meyer, F. Eder, S. Kurasch, V. Skakalova, J. Kotakoski, H.J. Park, S. Roth, A. Chuvilin, S.
Eyhusen, G. Benner, A.V. Krasheninnikov, and U. Kaiser, Phys. Rev. Lett. 108 (2012).
Banerjee, L. Colombo, and R.S. Ruoff, Science (New York, N.Y.) 324, 1312 (2009).
14X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner, A. Velamakanni, I. Jung, E. Tutuc, S.K.
15W. Eck, A. Küller, M. Grunze, B. Völkel, and A. Gölzhäuser, Adv. Mater. 17, 2583 (2005).
16A. Turchanin, A. Beyer, C.T. Nottbohm, X. Zhang, R. Stosch, A. Sologubenko, J. Mayer, P. Hinze,
17A. Turchanin, D. Käfer, M. El-Desawy, C. Wöll, G. Witte, and A. Gölzhäuser, Langmuir : the ACS
T. Weimann, and A. Gölzhäuser, Adv. Mater. 21, 1233 (2009).
journal of surfaces and colloids 25, 7342 (2009).
18P. Angelova, H. Vieker, N.-E. Weber, D. Matei, O. Reimer, I. Meier, S. Kurasch, J. Biskupek, D.
19X. Zhang, C. Neumann, P. Angelova, A. Beyer, and A. Gölzhäuser, Langmuir : the ACS journal of
Lorbach, K. Wunderlich, L. Chen, A. Terfort, M. Klapper, K. Müllen, U. Kaiser, A. Gölzhäuser,
and A. Turchanin, ACS Nano 7, 6489 (2013).
surfaces and colloids 30, 8221 (2014).
letters 15, 5944 (2015).
20J. Kotakoski, C. Brand, Y. Lilach, O. Cheshnovsky, C. Mangler, M. Arndt, and J.C. Meyer, Nano
21R. Menon, D. Gil, and H.I. Smith, J. Opt. Soc. Am. A 23, 567 (2006).
22D. Winston, B.M. Cord, B. Ming, D.C. Bell, W.F. DiNatale, L.A. Stern, A.E. Vladar, M.T. Postek,
M.K. Mondol, Yang, J. K. W., and K.K. Berggren, J. Vac. Sci. Technol. B 27, 2702 (2009).
8
|
1807.07086 | 1 | 1807 | 2018-07-18T18:01:02 | Generalized dissipation dilution in strained mechanical resonators | [
"physics.app-ph",
"cond-mat.mes-hall"
] | Mechanical resonators with high quality factors are of relevance in precision experiments, ranging from gravitational wave detection and force sensing to quantum optomechanics. Beams and membranes are well known to exhibit flexural modes with enhanced quality factors when subjected to tensile stress. The mechanism for this enhancement has been a subject of debate, but is typically attributed to elastic energy being "diluted" by a lossless potential. Here we clarify the origin of the lossless potential to be the combination of tension and geometric nonlinearity of strain. We present a general theory of dissipation dilution that is applicable to arbitrary resonator geometries and discuss why this effect is particularly strong for flexural modes of nanomechanical structures with high aspect ratios. Applying the theory to a non-uniform doubly clamped beam, we show analytically how dissipation dilution can be enhanced by modifying the beam shape to implement "soft clamping", thin clamping and geometric strain engineering, and derive the ultimate limit for dissipation dilution. | physics.app-ph | physics | Generalized dissipation dilution in strained mechanical resonators
S. A. Fedorov,1 N. J. Engelsen,1 A. H. Ghadimi,1 M. J. Bereyhi,1 R. Schilling,1 D. J. Wilson,2 and T. J. Kippenberg1, ∗
1Institute of Physics (IPHYS), ´Ecole Polytechnique F´ed´erale de Lausanne, 1015 Lausanne, Switzerland
2IBM Research - Zurich, Saumerstrasse 4, 8803 Ruschlikon, Switzerland
(Dated: July 20, 2018)
Mechanical resonators with high quality factors are of relevance in precision experiments, ranging
from gravitational wave detection and force sensing to quantum optomechanics. Beams and mem-
branes are well known to exhibit flexural modes with enhanced quality factors when subjected to
tensile stress. The mechanism for this enhancement has been a subject of debate, but is typically
attributed to elastic energy being "diluted" by a lossless potential. Here we clarify the origin of
the lossless potential to be the combination of tension and geometric nonlinearity of strain. We
present a general theory of dissipation dilution that is applicable to arbitrary resonator geometries
and discuss why this effect is particularly strong for flexural modes of nanomechanical structures
with high aspect ratios. Applying the theory to a non-uniform doubly clamped beam, we show
analytically how dissipation dilution can be enhanced by modifying the beam shape to implement
"soft clamping", thin clamping and geometric strain engineering, and derive the ultimate limit for
dissipation dilution.
I.
INTRODUCTION
Mechanical resonators with high quality factors are of
both fundamental and applied interest. They are em-
ployed in gravitational waves detector [1], cavity optome-
chanics [2], quantum [3] and classical [4] signal conver-
sion, tests of wavefunction collapse models [5] and nu-
merous sensing applications [6, 7]. In all these endeav-
ors, dissipation can be a limiting factor. As known from
the fluctuation-dissipation theorem [8], dissipation intro-
duces noise, which limits force sensitivity, frequency sta-
bility and results in decoherence of quantum states. Re-
duction of mechanical dissipation is practically challeng-
ing, however, because intrinsic and surface loss mecha-
nisms are often not well understood or not possible to
control. The quality factor, Q, of a mechanical resonator
typically does not exceed the inverse of the material loss
angle, φ, characterizing the delay between stress and
strain. Flexural modes of beams and membranes under
tension are notable exceptions to this rule: they can have
Qs far in excess of 1/φ due to a phenomenon known as
dissipation dilution.
The origin of dissipation dilution has been a subject
of debate. The concept was introduced in the gravita-
tional wave community when, to explain the enhanced
Q of test mass suspension wire, Gonzalez et. al. [9, 10]
reasoned that the lossy elastic energy of the wire was "di-
luted" by the conservative gravitational potential of the
test mass. A decade later, similar behavior was observed
in nanometric strings and membranes made of highly-
strained materials (most notably, silicon nitride [11–13]);
however, the lack of an external potential in this case
necessitated a rethinking of the physical model. In later
works the quality factors of flexural modes of uniform
beams [14] and membranes [15] were calculated from a
∗ [email protected]
FIG. 1. A) Dissipation dilution factors for vibrational modes
of a 3D resonator, doubly clamped to two quarter-sphere pads
(hatched gray) and subjected to tension. The total length is
20 µm, the block size is 8.5 × 7 × 4 µm, the bridge diameter
is 100 nm and the material pre-strain is 0.4%. B) Distri-
bution of effectively lossless elastic energy in a thin bridge
during flexural vibration. C) Schematic illustrating how the
cycle-averaged dynamic strain (cid:104)∆(cid:105) can be non-zero due to
geometric nonlinearity.
structural mechanics perspective and shown to be much
greater than 1/φ-in excellent agreement with experi-
ments [14–17]. These results partially demystified dissi-
pation dilution, but due to their lack of generality, the
understanding of this effect remains incomplete. It is still
not fully clear what causes dissipation dilution to emerge
in a resonator (aside from the mere presence of tensile
strain), if any modes except for flexural experience dilu-
tion and to what extent it can be engineered to produce
practical high-Q resonators.
8
1
0
2
l
u
J
8
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
6
8
0
7
0
.
7
0
8
1
:
v
i
X
r
a
∆lzxBADisplacement magnitudeResonator geometry0maxLongitudinalωn/2� = 8.1 MHzQ/Qint = 1.0xyzTorsionalωn/2� = 80 kHzQ/Qint = 1.0yzxFlexuralωn/2� = 560 kHzQ/Qint = 24xyzyzxu(x)‹∆ε› = ≠ 0‹∆l›llLossless elastic energy density wdilC0max Very recently, dissipation dilution has attracted signif-
icant interest as it enabled nanomechanical resonators, in
the form of patterned membranes and beams, to achieve
exceptionally high Q factors [18, 19]. In particular, by
localizing a beam mode away from its supports with a
phononic crystal (the "soft clamping" approach intro-
duced by Tsaturyan et al.
[18]) and using geometric
strain engineering [20] to enhance strain in the beam con-
striction, Q factors as high as 8× 108 were demonstrated
at room temperature [19]-surpassing even the highest
values measured in macroscopic sapphire bars [21]. These
advances suggest that a more detailed understanding of
dissipation dilution may be beneficial for optimizing ex-
isting designs and finding new ones, in addition to the
open questions mentioned above.
Here we address these questions with a general and
consistent theory which does not resort to the concept of
an a priori lossless potential. We derive the dissipation
dilution factors for modes of a mechanical resonator of
arbitrary geometry. We identify geometric nonlinearity
of strain in deformations to be a key component which,
together with static strain, enables dissipation dilution.
We extend the classic treatment of Q dilution in flexural
vibrations of a doubly-clamped beam to the case where
the beam has a non-uniform width. Using this theory we
show how a non-uniform width can be used to enhance
Q with three strategies: mode localization with phononic
crystals [18], both alone and in combination with adia-
batic tapering [19] and "thin clamping", introduced here.
We show that in a number of cases engineering dissipa-
tion dilution is related to geometric strain engineering
[22, 23]. We also derive the ultimate limit of dissipation
dilution set by the material yield strain. Our numerical
analysis of beams is based on the one-dimensional Euler-
Bernoulli equation and is in excellent agreement with a
full 3D treatment. The numerical routines for nanobeam
Q factor calculations are implemented in a freely avail-
able Mathematica package [24].
II. GEOMETRIC ORIGIN OF DISSIPATION
DILUTION
Dissipation dilution is commonly illustrated by a har-
monic oscillator subjected to an external lossless poten-
tial [9], as in the case of optically-trapped mirrors [25, 26]
or massive pendula in a gravitational field [9]. If ωint is
the oscillator natural frequency, φ is its loss angle [27]
and ωdil is the frequency of motion in the lossless poten-
tial, then the oscillator Q factor is increased compared to
the intrinsic value Qint ≡ 1/φ by the "dilution factor",
DQ ≡ Q
Qint
=
int + ω2
ω2
dil
ω2
int
.
(1)
For flexural vibrations of tensioned beams or membranes,
the Q enhancement takes place similarly to Eq. 1 with
the important distinction that here the potential energy
is stored only as elastic energy. Instead of introducing
2
an external potential, the elastic energy is divided into
lossy "bending" and lossless "tension" parts [14, 15], re-
lated to the curvature and gradient of the mode shape,
respectively. It is not evident a priori, however, how to
make this separation in a general case and under which
conditions the lossless part of energy is non-zero. Here
we answer both questions and show that the effectively
lossless elastic energy emerges if two conditions are sat-
isfied: a) static strain is non-zero in the resonator and b)
the average of strain variation over the oscillation period
is non-zero, i.e. the geometric nonlinearity of strain is
significant.
We now derive the dissipation dilution factor of a
generic vibrational mode. For this we compute the Q fac-
tor as the ratio of the elastic energy stored by the mode to
the energy dissipated per vibrational period. We assume
that static deformation is present in the structure along
with a part oscillating at the frequency ωn. Denoting the
total deformation field as Ui(x, y, z, t) (i = x, y, z), the
strain tensor eij [28] is given by
(cid:32)
(cid:33)
eij =
1
2
∂ Ui
∂xj
+
∂ Uj
∂xi
+
∂ Ul
∂xi
∂ Ul
∂xj
,
(2)
where summation over repeating indices is implied. The
last term in Eq. 2 is nonlinear in the displacement and can
be identified as the geometric nonlinearity. We empha-
size here that this nonlinearity is not due to a nonlinear
stress-strain relation and is not always negligible even for
infinitesimally small vibrations.
The strain tensor can be split into static eij and time-
dependent ∆eij(t) contributions
eij(t) = eij + ∆eij(t).
(3)
For brevity, when treating the 3D case we present a sim-
plified model where Poisson's ratio, ν, is neglected, so
that the stress-strain relation is given by
σij[ω] = Ee−iφeij[ω].
(4)
A full treatment accounting for Poisson's ratio can be
found in the Supplementary Information and ν is in-
cluded below when treating flexural modes of beams.
We find the time-averaged elastic energy density stored
by the mode as
(cid:104)∆wel(t)(cid:105) = E
(cid:104)eij(t)eij(t)(cid:105)
2
(cid:18)
eij(cid:104)∆eij(t)(cid:105) +
E
eijeij
− E
(cid:104)∆eij(t)∆eij(t)(cid:105)
=
2
(cid:19)
,
(5)
2
and the dissipated power density pdiss as
pdiss = (cid:104)σij(t) (eij)(cid:48)
The dilution factor of the vibrational mode is given by
the ratio of the resonator quality factor to Qint as
t(t)(cid:105) = ωn φ E(cid:104)∆eij(t)∆eij(t)(cid:105).
(6)
(cid:82) 2eij(cid:104)∆eij(t)(cid:105)dV
(cid:82) (cid:104)∆eij(t)∆eij(t)(cid:105)dV
DQ = 1 +
.
(7)
Eq. 7 reveals the peculiar effect of static strain eij
on dissipation.
If the static strain is zero then DQ =
Q/Qint = 1 irrespective of the mode shape (we empha-
size that corrections due to the imaginary part of Pois-
sons ratio are here neglected). In contrast, DQ can be
higher (or lower) than unity if eij (cid:54)= 0 and (cid:104)∆eij(t)(cid:105) (cid:54)= 0,
the latter being possible due to geometric nonlinearity in
Eq. 2.
Comparing Eq. 7 to Eq. 1, one recognizes
eij(cid:104)∆eij(t)(cid:105)dV
(cid:104)Wdil(t)(cid:105) ≡ E
(cid:90)
III. DISSIPATION DILUTION OF BEAM
RESONATORS
3
For the rest of the paper we consider in detail the flex-
ural modes of beams, as extreme dissipation dilution is
achievable in this case and it is possible to obtain an-
alytical results [10, 16]. Applying Eq. 7 we arrive at a
dilution factor given by
(cid:82) 2(cid:104)∆(t)(cid:105)dV
(cid:82) (cid:104)∆(t)2(cid:105)dV
(8)
DQ = 1 +
,
(11)
as an effectively lossless potential that generalizes the
"tension energy" in treatment of beams and membranes
[10, 15]. The lossy part of the energy is given by
(cid:104)∆eij(t)∆eij(t)(cid:105)dV,
(cid:90)
(cid:104)Wlossy(t)(cid:105) ≡ E
2
(9)
which generalizes the "bending energy" [10, 15] and cor-
responds to ω2
int in Eq. 1. Unlike the toy model, how-
ever, Wlossy in general depends on the static strain, which
implies that the intuitive picture that tension increases
stored energy without affecting dissipation is not correct
in general.
To give an example, we apply Eq. 7 to a doubly-
clamped 3D resonator made of pre-strained material as
shown in Fig. 1A and calculate dilution factors for a few
representative modes from different families. It can be
seen that among these modes only the flexural ones ex-
periences dissipation dilution, whereas the torsional and
longitudinal modes do not. A visualization of lossless
energy density (cid:104)wdil(t)(cid:105) in Fig. 1B shows that the loss-
less potential is concentrated in thin bridges between the
blocks. This is explained by a) static strain concentration
in constrictions and b) relatively large geometric nonlin-
earity of strain in flexural deformations, as opposed to
torsional or longitudinal deformations.
Strong dissipation dilution of flexural modes in high-
aspect-ratio beams and membranes [12, 14] is thus due
to the combination of tension and a large geometrically
nonlinear contribution to the dynamic strain. The latter
can be illustrated by considering flexural deformation of
an idealized infinitely thin beam shown in Fig. 1C. If the
beam is oriented along the x-axis and vibrates along the
z-direction with magnitude u, only the diagonal compo-
nent exx ≡ is relevant and the dynamic variation of
strain is quadratic (i.e. fully nonlinear) in the displace-
ment magnitude:
∆(x, t) = (u(cid:48)
x(x, t))2/2.
(10)
The role of geometric nonlinearity of strain in dissipation
dilution provides a warning: it is not correct to assume
that the mere presence of tensile strain in a mechanical
resonator increases its Q-for example, torsional modes
of the same structures that have high-Q flexural modes
usually do not experience any appreciable dissipation di-
lution (see Fig. 1A).
where is the static strain along the beam, terms pro-
portional to (cid:104)∆(t)(cid:105) and (cid:104)∆(t)2(cid:105) correspond to the loss-
less "tension" and lossy "bending" energy, respectively
[10, 15]-both are of elastic origin. Note that while Eq. 7
neglects Poisson's ratio, Eq. 11 does not, and is formally
exact in the 1D case.
So far we have not made any assumptions about the
beam cross-section, but in the following we focus on ge-
ometries directly accessible by nanofabrication. Specifi-
cally, we assume that the beams are made of a suspended
film with thickness h and pre-strain exx = eyy = film
(which redistributes upon suspension). The beam width
w(x) is in general non-uniform and its variation can be
used to improve vibrational quality factors.
For modes of a uniform rectangular beam evaluation
of Eq. 11 yields the well-known result [10, 16]
DQ,n =
1
2λ + π2n2λ2 .
Here n is mode number and λ is defined as [15, 16]
λ2 =
1
12avg
h2
l2 ,
(12)
(13)
where avg is the volume-averaged static tensile strain
and l is the beam length.
The derivation of Eq. 12 is based on a key insight:
the flexural modes of a beam contain two vastly differ-
ent length scales [9, 10]. Away from the clamping points
(clamps), modes form standing waves with wavelengths
on the order of 2l/n, while near the clamping points they
experience sharp bending at the length scale of λl, which
is responsible for fulfilling the clamped boundary con-
ditions u(cid:48) = 0. As a result, the majority of the elastic
energy is distributed over the mode away from the clamp-
ing points, while the small regions around them make a
large (dominant for lowest-frequency modes) contribu-
tion to the intrinsic losses [14, 15]. The energy dissipa-
tion around the clamping points is commonly referred
to as "clamping losses" [15], which, should not be con-
fused with losses due to modal coupling to the supporting
frame [29–31] or acoustic radiation [17, 32]. In the fol-
lowing we refer to the intrinsic loss occurring away from
the clamps as "distributed contribution".
We now generalize the multi-length scale approach for
the case of non-uniform beams and derive dissipation di-
4
strain distribution along the beam, (x), is equivalent as
these quantities are uniquely related as (see SI for details)
(x)/avg = wavg/w(x),
(18)
through the condition that the tension force must be con-
stant along the beam.
IV. DISSIPATION DILUTION LIMIT
Before showing how dissipation dilution can be en-
hanced in a non-uniform beam, we derive a rigorous up-
per bound for DQ. This bound is set by the yield strain,
material parameters, beam thickness and the frequency
of vibration, but does not depend on the beam length nor
the mode order. We assume that the clamping losses are
negligible (αn = 0) and evaluate the distributed loss co-
efficient βn using the strain-width relation (Eq. 18) and
the condition that the maximum strain in the beam can-
not exceed the yield strain yield. As a result we obtain
(see SI for details)
(cid:18) avg
(cid:19)2
yield
βn ≥
,
(19)
FIG. 2. Geometry, strain distribution and DQ in micropat-
terned beams, illustrating the concepts of soft-clamping, thin
clamping and strain-engineering. Dilution factors (DQ) are
calculated assuming beam length l = 3 mm and thickness
h = 20 nm. A) Beams with thin (above) and thick (be-
low) clamps, resulting in enhanced and reduced dissipation
dilution, respectively. DQ,max is maximum over modes. B)
Strain (top) and localized mode displacement field (bottom)
in a tapered phononic crystal beam.
lution factors as (see details in SI)
DQ,n =
1
2αnλ + βnΩ2
nλ2 ,
and thus the ultimate dissipation dilution bound is given
by
(14)
where we introduced dimensionless frequency of n-th
mode Ωn given by
Ω2
n =
ρl2ω2
n
avgE
,
(15)
and beam shape-dependent clamping and distributed loss
coefficients αn and βn are found as
√
(cid:16)(cid:82) 1
cl,n)2
vcl(u(cid:48)
0 v(s)un(s)2ds
(cid:17) ,
αn =
βn =
2Ω2
n
(cid:82) 1
(cid:82) 1
0 v(s)3un(s)2ds
0 v(s)un(s)2ds
.
(16)
(17)
Here s = x/l is the scaled coordinate along the beam,
un(s) is the mode shape, v(s) = w(s)/wavg is the beam
width variation normalized to its average width and
quantities with subscript "cl" are computed near the
clamps (see SI).
Dissipation dilution of a non-uniform beam can be dis-
cussed entirely in terms of the reduction of the αn and
βn coefficients by varying the beam shape w(x); however,
some results are more intuitively interpreted from the
prospective of geometric strain engineering [19, 22, 23],
a technique that exploits relaxation of a suspended film
to locally enhance the strain. Formally, the treatment in
terms of the transverse beam shape, w(x), or the static
DQ ≤ 12E2
ρh2ω2
yield
.
(20)
This limit is formally equivalent to the dissipation dilu-
tion of a clampless uniform beam strained to the yield
strain.
V. NON-UNIFORM BEAMS WITH ENHANCED
DISSIPATION DILUTION
We consider three beam designs that produce vibra-
tional modes with enhanced dissipation dilution com-
pared to uniform beam - phononic crystal (PnC) beams,
beams with thin clamps and tapered PnC beams. We
first analytically estimate the attainable DQs with these
designs and then numerically calculate them by solving
the Euler-Bernoulli equation [24] (see SI). Numerical cal-
culations are presented in Fig. 3 for beams with length
l = 3 mm and thickness h = 20 nm. We show dissi-
pation dilution factors, which are material independent,
along with absolute Q factors assuming parameters typi-
cal to stoichiometric Si3N4 films (E = 250 GPa, ν = 0.23,
σfilm = 1.14 GPa, Qint = 1.4 × 103 for h = 20 nm),
a well-established material for strained nanomechanics
[16]. Note that with these extreme parameters the max-
imum dilution factor is large (DQ > 104) even for a uni-
form beam.
The first strategy we consider is soft clamping [18, 19]
- suppression of clamping losses by localizing a flexural
01234500.20.40.60.8100.10.51.01.52.0sε [%]σ [GPa]Avcl = 0.14vcl = 5.200.20.40.60.81usBxyzxzyxyzDQ,max = 0.64×104DQ,max = 7.9×104DQ = 7.3×104 5
where nmax is the optimum localized mode order that in-
creases only logarithmically slowly with 1/λ (see SI for
the explicit expression). This result demonstrates that
patterning a beam with a phononic crystal can provide
an improvement in DQ by a factor of ∼ 1/(n2
maxλ) com-
pared to a uniform beam of the same size. Note that
the maximum attainable DQ is far below 1/λ2-the en-
hancement expected from clamping loss suppression for a
fundamental mode-as nmax is in practice much greater
than 1. It also follows from Eq. 22 that in order for soft
clamping to provide an increased quality factor, λ needs
to be much smaller than 1, i.e. dissipation dilution fac-
tors needs to be high even for non-localized modes.
√
The second strategy we consider is reduction of the
beam width near the clamps, vcl = w(0)/wavg, in order
to create local strain enhancement in clamping regions
(see Fig. 2A top). Eq. 16 shows that αn is proportional
to
vcl and thus can be reduced by thinning down the
clamps (u(cid:48)
cl,n and Ωn are almost unaffected by vcl as long
as the clamping region length is small). This can be
interpreted as an effective decrease of λ over the clamping
region to
λcl =(cid:112)h2/12cll2,
(23)
where cl = avg/vcl is the local strain. The dissipation
dilution of beams with thin clamps is thus given by
DQ,n ≈
1
2λcl + (nπ)2λ2 .
(24)
In contrast to the PnC approach, thin-clamping beams
are predicted to have improved quality factors for low-
order beam modes, including the fundamental mode (see
Fig. 3, green points).
One caveat needs to be mentioned when considering
the effect of local strain on dissipation dilution: geo-
metric concentration of strain in one region unavoid-
ably results in the reduction of strain elsewhere. To im-
prove dilution factors beyond those of a uniform beam,
the region(s) of enhanced strain must overlap with the
region(s) which dominate dissipation in the vibrational
mode, in this case the clamps. A common beam geome-
try which does not satisfy this requirement, a beam with
filleted (thick) clamping points, is shown in the bottom of
Fig. 2A. This result is at odds with recently reported en-
hanced Qs in trampoline membranes with filleted tethers
[33].
In both uniform PnC and thin-clamped beams, the
clamping loss is reduced, but distributed loss is not. The
latter can be addressed by co-localization of both flexu-
ral mode and strain away from the clamps as shown in
Fig. 2B. Following the strategy described in [19], here the
width of the PnC is changed cell-wise according to
wcell,i ∝ 1 − (1 − a) exp(−i2/i2
0),
(25)
FIG. 3. Dissipation dilution in beams with different trans-
verse profiles, assuming a fixed length l = 3 mm and thickness
h = 20 nm. Points correspond to DQ (left axis) and Q (right
axis) for specific flexural modes, assuming Qint = 1.4 × 103.
Blue and green points correspond to modes of uniform and
thin-clamped (vcl = 0.14) beams. Dark red and red points
correspond to localized modes of PnC beams and tapered
PnC beams, respectively. Note that each localized mode cor-
responds to a different beam profile. Blue line: ideal limit for
a soft-clamped beam (Eq. 21). Gray line:
ideal limit for a
clamp-free beam strained to the yield point (Eq. 20).
mode in a phononic crystal. A 1D phononic crystal can
be formed by periodically modulating the beam width
[17] (with wmax = 2wmin for the design in Fig. 3). Lo-
calized modes of a PnC beam can closely approach the
performance of idealized clamp-less beams, with dilution
factors given by
DQ =
12E2
film
(1 − ν)2ρh2ω2 .
(21)
Here Poisson's ratio accounts for relaxation of film stress
in transverse direction upon suspension.
Importantly,
the strong suppression of mechanical mode amplitude
near the clamps requires a large number of PnC unit
cells and thus a high order n of the localized mode. For
high-order modes, distributed losses increase as n2 due
to increased bending curvature for shorter acoustic wave-
lengths and at some point exceed the suppressed clamp-
ing losses. These trends can be seen in Fig. 3, where
the DQ factor of the localized mode is plotted versus fre-
quency. DQ can be optimized by changing the localized
mode order n while keeping all the parameters except
for the unit cell length fixed. The amplitude of a local-
ized mode decays exponentially with the distance from
the defect, such that the clamping loss coefficient can be
estimated as αn = e−(n−1)/nL , where nL is the mode am-
plitude decay length in units of acoustic half-wavelengths.
Optimization of DQ in Eq. 14 with respect to n, yields
DQ,max ≈
1
maxλ2 ,
π2n2
(22)
where i = 0, 1 ... is the cell index starting from the beam
center, a and i0 respectively define the transverse and
0.050.100.501510103DQQωn/2π [MHz]104105107108109106ε>εyield √
longitudinal sizes of the waist region. Importantly, the
PnC cell lengths must also be scaled proportional to
1/
wcell in order to compensate for the bandgap fre-
quency shift due to the non-uniform strain distribution.
An estimate of DQ for the tapered PnC is obtained by
assuming that the mode is localized in the waist region
of width vwaist and that clamping losses are negligible:
DQ,n ≈
1
Ω2
waistλ2
waist
,
(26)
where
Ωwaist =(cid:112)ρl2ω2/(waistE),
λwaist =(cid:112)h2/(12waistl2),
(27)
(28)
and waist = avg/vwaist. It follows that by increasing the
waist strain to yield value, the ultimate limit of dissi-
pation dilution (Eq. 20) is attainable with tapered PnC
beam designs, in contrast to the previous two methods.
A practical limitation for dissipation dilution enhance-
ment by strain concentration in this case originates from
the tradeoff between waist and the waist length. Substan-
tially increased strain is only achievable over a small frac-
tion of the beam length, therefore only short-wavelength
and high-frequency modes can benefit from such global
geometric strain engineering. In Fig. 3 we plot DQ versus
frequency for localized modes of tapered beams, where
the taper waist has been adjusted to match the wave-
length of the localized mode. It can be seen that as the
mode frequency increases, its dilution is progressively en-
hanced relative to conventional soft-clamped modes (red
points).
VI. CONCLUSIONS AND OUTLOOK
We have presented a theoretical framework to analyze
the quality factors of strained mechanical resonators of
6
arbitrary three dimensional geometry and shown that a
lossless contribution to the elastic energy -giving rise
to Q-enhancement by dissipation dilution - emerges in
the presence of static strain and geometric nonlinearity.
High aspect ratio beams and membranes can produce
particularly large dissipation dilution, though it is not
impossible that other geometries can do it as well.
For the specific case of variable cross-section beams
subjected to axial tension we presented an analytical
model. We showed that by corrugating the beam it is
possible to create modes with quality factors enhanced
by more than an order of magnitude compared to a uni-
form beam. We interpret the Q enhancement in terms
of clamping loss suppression and local strain engineer-
ing, deriving the limits of each approach, and estimat-
ing practically achievable absolute Q factors for beams
made of high-stress Si3N4 . The numerical results re-
ported for beams were obtained using a freely available
Mathematica package [24].
We note that while Si3N4 is currently the most popular
material for strained nanomechanics - particularly for
applications in optomechanics [34–37] - the principles
described here apply to resonators made of any material
under strain, whether produced by external force [38],
lattice mismatch (e.g. during epitaxial growth) [39] or
mismatch of thermal expansion coefficients [40].
ACKNOWLEDGEMENTS
We thank Alexander Tagantsev for useful discussions.
This work was supported by the EU Horizon 2020 Re-
search and Innovation Program under grant agreement
732894 (FET Proactive HOT), the SNF Cavity
no.
Quantum Optomechanics project (grant no.
163387)
and DARPA grant HR0011181003. M.J.B. is supported
by MSCA ETN-OMT (grant no. 722923). T.J.K ac-
knowledges support from ERC AdG (QuREM, grant no.
320966). Code to reproduce data in Fig. 2 and Fig. 3 is
available on Zenodo[24].
[1] L. Ju, D. G. Blair, and C. Zhao, Reports on Progress in
[7] Y. T. Yang, C. Callegari, X. L. Feng, K. L. Ekinci, and
Physics 63, 1317 (2000).
M. L. Roukes, Nano Letters 6, 583 (2006).
[2] M. Aspelmeyer, T. J. Kippenberg, and F. Marquardt,
[8] H. B. Callen and T. A. Welton, Physical Review 83, 34
Reviews of Modern Physics 86, 1391 (2014).
(1951).
[3] R. W. Andrews, R. W. Peterson, T. P. Purdy, K. Cicak,
and K. W. Lehnert,
R. W. Simmonds, C. A. Regal,
Nature Physics 10, 321 (2014).
[4] T. Bagci, A. Simonsen, S. Schmid, L. G. Villanueva,
E. Zeuthen, J. Appel, J. M. Taylor, A. Srensen, K. Us-
ami, A. Schliesser, and E. S. Polzik, Nature 507, 81
(2014).
[5] A. Vinante, R. Mezzena, P. Falferi, M. Carlesso, and
A. Bassi, Physical Review Letters 119, 110401 (2017).
[6] D. Rugar, R. Budakian, H. J. Mamin, and B. W. Chui,
Nature 430, 329 (2004).
[9] G. Gonzlez, Classical and Quantum Gravity 17, 4409
(2000).
[10] G. I. Gonz´alez and P. R. Saulson, The Journal of the
Acoustical Society of America 96, 207 (1994).
[11] S. S. Verbridge, J. M. Parpia, R. B. Reichenbach, L. M.
Bellan, and H. G. Craighead, Journal of Applied Physics
99, 124304 (2006).
[12] B. M. Zwickl, W. E. Shanks, A. M. Jayich, C. Yang,
A. C. Bleszynski Jayich, J. D. Thompson, and J. G. E.
Harris, Applied Physics Letters 92, 103125 (2008).
[13] Q. P. Unterreithmeier, E. M. Weig, and J. P. Kotthaus,
Nature 458, 1001 (2009).
[14] Q. P. Unterreithmeier, T. Faust, and J. P. Kotthaus,
Physical Review Letters 105, 027205 (2010).
T. P. Purdy, W. Wieczorek, C. A. Regal, et al., App.
Phys. Lett. 104, 201908 (2014).
[15] P.-L. Yu, T. P. Purdy, and C. A. Regal, Physical Review
[40] C. A. Regal, J. D. Teufel, and K. W. Lehnert, Nature
Letters 108, 083603 (2012).
Physics 4, 555 (2008).
[16] L. Villanueva and S. Schmid, Physical Review Letters
7
113, 227201 (2014).
[17] A. H. Ghadimi, D. J. Wilson, and T. J. Kippenberg,
Nano Letters (2017), 10.1021/acs.nanolett.7b00573.
[18] Y. Tsaturyan, A. Barg, E. S. Polzik, and A. Schliesser,
Nature Nanotechnology 12, 776 (2017).
[19] A. H. Ghadimi, S. A. Fedorov, N. J. Engelsen, M. J.
Bereyhi, R. Schilling, D. J. Wilson, and T. J. Kippen-
berg, Science 360, 764 (2018).
[20] R. A. Minamisawa, M. J. Sess, R. Spolenak, J. Faist,
C. David, J. Gobrecht, K. K. Bourdelle, and H. Sigg,
Nature Communications 3, 1096 (2012).
[21] V. B. Braginsky, V. P. Mitrofanov, and V. I. Panov,
Systems with Small Dissipation (University of Chicago
Press, 1985).
[22] T. Zabel, R. Geiger, E. Marin, E. Mller, A. Diaz, C. Bon-
zon, M. J. Sess, R. Spolenak, J. Faist, and H. Sigg, Jour-
nal of Materials Research 32, 726 (2017).
[23] R. Zhang, C. Ti, M. I. Davano, Y. Ren, V. Aksyuk,
Y. Liu, and K. Srinivasan, Applied Physics Letters 107,
131110 (2015).
[24] Mathematica package
is available at
zenodo.com,
DOI:10.5281/zenodo.1296925.
[25] T. Corbitt, C. Wipf, T. Bodiya, D. Ottaway, D. Sigg,
and N. Mavalvala, Physical
N. Smith, S. Whitcomb,
Review Letters 99, 160801 (2007).
[26] K.-K. Ni, R. Norte, D. J. Wilson, J. D. Hood, D. E.
Chang, O. Painter, and H. J. Kimble, Physical Review
Letters 108, 214302 (2012).
[27] P. R. Saulson, Physical Review D 42, 2437 (1990).
[28] L. D. Landau and E. M. Lifshitz, Theory of elasticity
(London Pergamon Press, 1970).
[29] I. Wilson-Rae, R. A. Barton, S. S. Verbridge, D. R.
Southworth, B. Ilic, H. G. Craighead, and J. M. Parpia,
Physical Review Letters 106, 047205 (2011).
[30] G. D. Cole, I. Wilson-Rae, K. Werbach, M. R. Van-
ner, and M. Aspelmeyer, Nature Communications 2, 231
(2011).
[31] Y. Tsaturyan, A. Barg, A. Simonsen, L. G. Villanueva,
S. Schmid, A. Schliesser, and E. S. Polzik, Optics Ex-
press 22, 6810 (2014).
[32] J. Chan, A. H. Safavi-Naeini, J. T. Hill, S. Meenehan,
and O. Painter, Applied Physics Letters 101, 081115
(2012).
[33] R. Norte, J. Moura, and S. Grblacher, Physical Review
Letters 116, 147202 (2016).
[34] J. D. Thompson, B. M. Zwickl, A. M. Jayich, F. Mar-
quardt, S. M. Girvin, and J. G. E. Harris, Nature 452,
72 (2008).
[35] D. J. Wilson, C. A. Regal, S. B. Papp, and H. J. Kimble,
Physical Review Letters 103, 207204 (2009).
[36] T. P. Purdy, R. W. Peterson, and C. A. Regal, Science
339, 801 (2013).
[37] D. J. Wilson, V. Sudhir, N. Piro, R. Schilling,
A. Ghadimi, and T. J. Kippenberg, Nature 524, 325
(2015).
[38] S. S. Verbridge, D. F. Shapiro, H. G. Craighead, and
J. M. Parpia, Nano Letters 7, 1728 (2007).
[39] G. D. Cole, P.-L. Yu, C. Gartner, K. Siquans,
R. Moghadas Nia, J. Schmole, J. Hoelscher-Obermaier,
Supplementary information for "Generalized dissipation dilution in
strained mechanical resonators"
8
I. DISSIPATION DILUTION IN A GENERIC MECHANICAL RESONATOR MADE OF ISOTROPIC
ANELASTIC MATERIAL
In the main manuscript Poisson's ratio is neglected to derive a simplified expression derived for the dissipation
dilution of modes of a 3D resonator. Here we remove this simplification and present a more general formula taking
Poisson's ratio, ν, into account. As in the main text, E is the Young's modulus, φ is the loss angle and summation
over repeating indices is assumed.
The time-dependent displacement field is denoted by Ui(x, y, z, t), where i = x, y, z is the coordinate index. The
train field is derived from it as[1]
eij =
1
2
∂ Ui
∂xj
+
∂ Uj
∂xi
+
∂ Ul
∂xi
∂ Ul
∂xj
and the stress σij is related to strain via Hooke's law[1]
(cid:32)
(cid:18)
(cid:33)
,
(cid:19)
ellδij
,
σij =
E
1 + ν
eij +
ν
1 − 2ν
(S1)
(S2)
where δij is the Kronecker delta. Following the main text, we now assume that the deformation field consists of a
static part, Ui(x, y, z), and a dynamic part due to mechanical vibrations, ∆Ui,n(x, y, z, t), which is given by
∆Ui(x, y, z, t) = ∆Ui,n(x, y, z)e−iωnt + c.c,
(S3)
where ∆Ui,n(x, y, z) and ωn is the complex envelope and frequency of n-th mode. Strain, stress and elastic energy
can be separated into static and time-dependent contributions accordingly:
eij(t) = eij + ∆eij(t),
σij(t) = σij + ∆σij(t),
w(t) = w + ∆w(t).
The instantaneous elastic energy density is then given by
w =
1
2
σij eij =
E
2(1 + ν)
(cid:18)
(cid:19)
,
eij eij +
ν
1 − 2ν
(ell)2
(S4)
(S5)
(S6)
(S7)
and the average of its variation, ∆w(t), which is the elastic energy stored by the vibrational mode, is found as
(cid:104)∆w(t)(cid:105) =
1
2
(cid:18)
(σij(cid:104)∆eij(t)(cid:105) + eij(cid:104)∆σij(t)(cid:105) + (cid:104)∆σij(t)∆eij(t)(cid:105))
E
(2eij(cid:104)∆eij(t)(cid:105) + (cid:104)∆eij(t)∆eij(t)(cid:105)) +
(2ell(cid:104)∆ekk(t)(cid:105) + (cid:104)(∆ekk(t))2(cid:105))
(cid:19)
(S8)
.
=
2(1 + ν)
ν
1 − 2ν
We can then find the dissipated power density as
(cid:28)
(cid:29)
∂eij
∂t
(cid:28) ∂
∂t
(cid:29)
(cid:28)
∆eij(t)
+
∆σij(t)
∂
∂t
(cid:29)
∆eij(t)
.
(S9)
pdiss =
σij
= σij
Here, the second term, (cid:104)∆σij(t)∂∆eij(t)/∂t(cid:105), yields non-zero dissipated power if a delayed strain response to stress
is introduced as a perturbation by the substitution ∆eij[ω] → (1 + iφ)∆eij[ω] and the average over time is found
using the unperturbed ∆eij. Unlike Eq. S8 for the stored energy, the extra term which arises in the presence of static
deformation, σij(cid:104)∂∆eij(t)/∂t(cid:105), is always zero as
(cid:28) ∂
∂t
(cid:29)
(cid:90) T
0
1
T
∆eij(t)
=
∆eij(t)dt = ∆eij(T ) − ∆eij(0) = 0,
∂
∂t
(S10)
9
FIG. S1. A) Shapes of third order flexural modes of beams assuming the values of λ = 10−1, 10−3, 10−3. The zoomed-in region
shows the mode in the clamping regions, which illustrates an increase in mode curvature around clamps with the reduction of
λ. B) Deformation of a segment vibrating beam.
where T is the oscillation period. Overall the dissipated power density is found as
pdiss = φωn(cid:104)∆σij(t)∆eij(t)(cid:105) = φωn
1 − 2ν
We find the quality factor of the mode from the stored energy and dissipation rate as
(1 + ν)
(cid:104)∆eij(t)∆eij(t)(cid:105) +
ν
(cid:19)
(cid:104)(∆ekk(t))2(cid:105)
.
(S11)
and then find the dissipation dilution ratio as
DQ =
Q
Qint
= 1 +
= 1 +
(cid:104)Wdil(t)(cid:105)
(cid:104)Wlossy(t)(cid:105) ,
where Qint = 1/φ and the dilution and lossy energies are given, respectively, by
Q =
E
2ωn
(cid:18)
(cid:82) (cid:104)∆w(t)(cid:105)dV
(cid:82) pdissdV
(cid:82) (cid:104)∆σij(t)∆eij(t)(cid:105)dV
(cid:90)
,
2(cid:82) (σij(cid:104)∆eij(t)(cid:105) + eij(cid:104)∆σij(t)(cid:105))dV
(cid:104)Wdil(t)(cid:105) =
(cid:104)Wlossy(t)(cid:105) =
(cid:90)
(σij(cid:104)∆eij(t)(cid:105) + eij(cid:104)∆σij(t)(cid:105))dV,
1
2
(cid:104)∆σij(t)∆eij(t)(cid:105)dV.
(S12)
(S13)
(S14)
(S15)
Eq. S14-S15 generalize the expressions for dilution and lossy elastic energies presented in the main text for the case
of non-zero Poisson's ratio and reproduce them if ν = 0.
II. DERIVATION OF DISSIPATION DILUTION IN A DOUBLY CLAMPED NON-UNIFORM BEAM
We consider flexural vibrations of thin doubly-clamped beam resonators and use the general result from the previous
section to derive more useful expressions for quality factors and dissipation dilution factors. The beam will have
dimensions h, w and l corresponding to the thickness (z-direction), width (y-direction) and length (x-direction). We
assume that h, w (cid:28) l, but we do not impose restrictions on the beam cross-section and do not assume h and w are
constant. The beam is suspended between two clamps and experiences a tensile force T , which creates an equilibrium
axial strain of (x) ≡ exx and a stress given by σ(x) ≡ σxx = E (x). Due to the high aspect ratio of the beam, we
can neglect stresses in all directions other than the x axis.
We now consider the displacement of a beam segment in z direction u(x) ≡ ∆Uz, as illustrated in Fig. S1B. We
then find the instantaneous variation of strain ∆ ≡ ∆exx and stress ∆σ ≡ ∆σxx to be given by
∆(x, y, z, t) = −u(cid:48)(cid:48)
xx(x, t)z +
(u(cid:48)
x(x, t))2
2
,
∆σ(x, y, z, t) = E ∆(x, y, z, t).
The elastic energy density stored by the flexural mode is found as a sum of two terms
(cid:104)∆w(t)(cid:105) =
E
2
(2(cid:104)∆(t)(cid:105) + (cid:104)∆(t)2(cid:105)) = (cid:104)wtens(t)(cid:105) + (cid:104)wbend(t)(cid:105),
(S16)
(S17)
(S18)
xzu(x)u(x+dx)u'(x)u'(x+dx)00.010.020.000.040.0800.40.60.81.0012-2ssλ = 10-1λ = 10-2λ = 10-3-10.2ABu [arb. units] 10
FIG. S2. Beam resonator shapes with uniform thickness and non-uniform width demonstrating enhanced dissipation dilution.
A) Phononic crystal beam with "soft-clamped" localized mode, B) Tapered phononic crystal beam with soft-clamped mode
and engineered local strain enhancement C) Beam with thin clamps and its fundamental mode.
where the first term is tension energy and the second is bending energy. Inserting Eq. S16 into Eq. S18 and performing
integration over the beam volume, we find the total energy contributions
(cid:104)Wtens(t)(cid:105) =
(cid:104)Wbend(t)(cid:105) =
A(x)(x)(u(cid:48)
x(x, t))2dx =
I(x)(u(cid:48)(cid:48)
xx(x, t))2dx.
l
(u(cid:48)
x(x, t))2dx,
(S19)
(S20)
(cid:90)
T
2
(cid:90)
(cid:90)
l
E
2
E
2
l
Here I(x) = w(x)h(x)3/12 is the geometrical moment of inertia, A(x) = w(x)h(x) is the cross-section area and we
used the fact that the tension T = EA(x)(x) is constant along the beam. Provided that, according to Eq. S11, the
dissipation power density is given by
pdiss = φ ωnE(cid:104)∆(t)2(cid:105) = 2φ ωn(cid:104)wbend(t)(cid:105).
we find dissipation dilution factor of a flexural beam mode as
DQ = 1 +
(cid:104)Wtens(t)(cid:105)
(cid:104)Wbend(t)(cid:105) .
(S21)
(S22)
A. Nanobeams and equilibrium strain distribution in a suspended film
Until now we have considered beams of arbitrary variable transverse cross-section.
In the following we impose
geometrical constraints, consistent with nanomechanical resonators fabricated by locally suspending a micropatterned
thin film. Although qualitatively most of our conclusions are not affected by this assumption, it considerably simplifies
notations while allowing the theory to be directly applied to a very broad range of practical high-strain resonators. In
particular, we assume the yz cross section of the beam to be rectangular, the width w(x) be, in general, x-dependent
and the thickness to be constant (representative geometries are shown in Fig. S2). Strain can be present in a material
film used for microresonator fabrication due to lattice mismatch[2] between the film and substrate or by mismatch
in their thermal expansion coefficients[3]. Upon suspension, the originally homogeneous strain inside the film is
redistributed. The strain is locally enhanced in constrictions and reduced elsewhere[4–6].
from the unsuspended film strain film. This can be done by noting that (a) the total elongation of the beam(cid:82) l
The analysis of the vibrational properties of a beam in this case requires the axial tension force T to be found first
0 (x)dx
is constant over the relaxation process, as it is defined by separation of the beam clamping points and (b) that the
balance of tensile force requires
(x)w(x) = const =
T
hE
.
(S23)
From the initial condition
(cid:90) l
(S24)
where ν is the Poisson's ratio and the factor (1 − ν) accounts for transverse relaxation of the strain, one finds the
equilibrium tension as
0
(x)dx = film(1 − ν)l,
T = filmE(1 − ν)h
dx
.
(S25)
(cid:90) l
(cid:32)
1
l
1
w(x)
0
(cid:33)−1
ABCxyzxzyxyzxzyxyzxzy One other useful relation for the strain distribution follows from Eq. S23
(x) = avg/v(x).
For the following calculation, we also introduce a few auxiliary quantities:
1. Mean beam width
w0 =
1
l
(cid:90) l
0
w(x)dx
2. Relative width variation function
v(x) = w(x)/w0
3. Static stress σavg and strain avg, averaged over the beam volume
σavg = E avg,
avg =
1
hw0l
0
hw(x)(x)dx =
T
w0h
.
B. Vibrational modes
(cid:90) l
(cid:19)
(cid:18)
11
(S26)
(S27)
(S28)
(S29)
(S30)
(S32)
(S33)
(S34)
In order to proceed with explicit calculation the dissipation dilution factors, we first need to find the eigenfrequencies
ωn and the vibrational mode shapes un of a beam. For an elastic beam with high aspect ratio (l/h and l/w much
larger than one), these quantities can be found by solving the Euler-Bernoulli equation[1]
d2
dx2
I(x)E
d2un
dx2
− T
d2un
dx2 − ρl(x)ω2
nun = 0,
(S31)
where n is the mode index, ρl(x) = ρhw(x) is the linear mass density and I(x) is the geometric moment of inertia.
In order to simplify the notation, it is convenient to introduce a normalized length, s = x/l, taking values from 0 to
1, and use it to transform Eq. S31 to a new form
(cid:18)
(cid:19)
v(s)
d2un
ds2
− 1
v(s)
d2un
ds2 − Ω2
nun = 0,
λ2 1
v(s)
d2
ds2
where Ω is the dimensionless frequency
and λ is the strain dilution parameter given by
Ω2 =
ρl2ω2
avgE
,
λ2 =
1
12avg
h2
l2 .
The high-strain limit corresponds to λ being much smaller than 1. For a doubly clamped beam, the eigenvalue problem
in Eq. S32 is supplemented with boundary conditions
u(0) = u(1) = 0, u(cid:48)(0) = u(cid:48)(1) = 0.
(S35)
C. Derivation of distributed and clamping losses
The evaluation of the integrals in Eq. S19-S20 provides us with a general formula for the dissipation dilution of the
n-th mode
DQ,n = 1 +
1
λ2
(cid:82) 1
(cid:82) 1
0 (u(cid:48)
n(s))2 ds
0 v(s) (u(cid:48)(cid:48)
n(s))2 ds
,
(S36)
in which one can separate the contributions due to the distributed and the clamping parts of the mode. Here we are
interested in the strong dilution limit, where DQ (cid:29) 1 and therefore we neglect the first term in Eq. S36. In order to
find the distributed energy, we neglect the bending term in Eq. S32, which only weakly perturbs the solution in the
region away from the clamping points (see Fig. S1A), and find the mode shapes un from
12
− 1
v(s)
d2un(s)
ds2 = Ω2
nun(s).
The tension and bending energy integrals in Eq. S36 can be transformed to a new form
(cid:90) 1
(cid:90) 1
0
0
(u(cid:48)
n(s))2 ds = Ω2
n
v(s)un(s)2ds,
v(s) (u(cid:48)(cid:48)
n(s))2 ds = Ω4
n
v(s)3un(s)2ds.
(cid:90) 1
0
(cid:90) 1
0
where
DQ,n =
1
2αnλ + βnΩ2
nλ2 ,
√
(cid:16)(cid:82) 1
cl,n)2
vcl(u(cid:48)
0 v(s)un(s)2ds
(cid:17) ,
αn =
βn =
2Ω2
n
(cid:82) 1
(cid:82) 1
0 v(s)3un(s)2ds
0 v(s)un(s)2ds
.
With the help of Eq. S44-S46, the optimization of dissipation dilution can be performed by shaping v(s) to reduce
n = (πn)2 and
αn (clamping losses) and βn (distributed losses). For a uniform rectangular beam v(s) = 1, Ω2
un =
2 sin(πn s), which yields αn = 1, βn = 1 and reproduces the result from [7]
√
Drect.beam
Q,n
=
1
2λ + (nπ)2λ2 .
(S47)
In addition to the distributed contributions given by Eq. S39, the bending energy includes a contribution from the
clamping regions. The tensile energy stored in these regions is negligibly small. Near the clamping points the bending
term in Euler-Bernoulli equation is significant due to the boundary condition u(cid:48)(0) = u(cid:48)(1) = 0, but, on the other
hand, u is close to 0 so that Ω2
nu can be neglected. In the region around s = 0, assuming that the beam width is
approximately constant here, such that v(s) = vcl, Eq. S32 reduces to
λ2vclu(cid:48)(cid:48)(cid:48)(cid:48)(s) − u(cid:48)(cid:48)(s) = 0.
(S40)
The general solution is given by
u(s) = C1 + C2s + C3e−s/(λ
vcl) + C4es/(λ
vcl),
√
√
where the constants C1−4 can be found from the boundary conditions: u(0) = 0, u(cid:48)(0) = 0 and u(cid:48)(s (cid:29) λ
For the solution, un, to the wave equation given by Eq. S37, u(cid:48)
condition u(cid:48)
cl,n.
n(0). un does therefore not satisfy the boundary
cl,n = u(cid:48)
vcl) = u(cid:48)
n(0) = 0 per se. Explicitly,
√
(cid:17)(cid:17)
.
(cid:16)
(cid:16)
u(s) = u(cid:48)
cl,n
√
s + λ
vcl
√
e−s/(λ
vcl) − 1
(cid:90) ∞
0
v(s) (u(cid:48)(cid:48)(s))2 ds =
√
vcl(u(cid:48)
cl,n)2.
1
2λ
and the contribution of the clamping point into the curvature integral is found as
√
vcl (cid:28) 1 and the bending energy stored here is proportional to
Note, that the clamping region is small ∆xcl/l = λ
the magnitude of the mode envelope at the beam boundaries. Combining the clamping (assumed to be equal at both
clamping points, s = 0 and s = 1) and the distributed contributions, we arrive at
(S37)
(S38)
(S39)
(S41)
(S42)
(S43)
(S44)
(S45)
(S46)
13
FIG. S3. A) Top: localized flexural mode shape u(x) (blue) and its exponential envelope (orange). Bottom: Geometry of a
beam (red). PnC unit cell is highlighted blue within the beam. B) Red dots: spectrum of the out-of-plane flexural vibrations
of the beam shown in panel A. Blue lines: band diagram of a phononic crystal with the cell highlighted in panel A. Blue dots:
spectrum of a uniform rectangular beam with same l, h as the PnC beam. C) Dissipation dilution of PnC beam modes (red
dots) compared to the modes of a uniform beam (blue dots) and a uniform beam without clamping losses (blue line). D)
Variation of localized mode DQ (shades of red) with frequency and beam length, and comparison to modes of uniform beams
of the same lengths. Beam thickness here is h = 20 nm. Localized mode frequency is changed by the variation of the number
of PnC unit cells within the beam (together with the unit cell length as the beam length is fixed) while keeping the ratio of the
central defect to the unit cell length constant.
III. ABSOLUTE QUALITY FACTORS OF SI3N4 NANOBEAMS
If the resonator dissipation is due to intrinsic losses, the absolute mode quality factors can be calculated according
to Eq. S13 from the intrinsic material quality factor Qint and DQ as
Q = DQ × Qint.
(S48)
In the high-strain limit (λ (cid:28) 1) DQ depends only on the beam geometry, mode order and strain, but not on any of the
material parameters. Dissipation dilution can therefore be understood without ever specifying a material. However,
we present calculations of absolute Q factors assuming the material is stoichiometric Si3N4, as it is by far the most
popular platform for strained high-Q nanomechnical resonators (see, for example Villanueva et al.[8], and references
therein). In particular, we assume parameters consistent with the Si3N4 deposited by low pressure chemical vapor
deposition, as used in[9]: deposition strain film = 0.46% (stress σfilm = 1.14 GPa), Young's modulus E = 250 GPa,
Poisson's ratio ν = 0.23 and density ρ = 3100 kg/m3.
The intrinsic quality factor of Si3N4 was found to be almost frequency independent within the range 100 kHz–50
MHz[10] but increasing with thickness due to surface losses. For Si3N4 of smaller thickness than 100 nm it was
phenomenologically established[8, 9, 11] that the intrinsic quality factor is proportional to thickness.
Qint(h) = 6900
h
[100 nm]
.
(S49)
IV. SOFT CLAMPED MODES IN PHONONIC CRYSTAL BEAMS
Fig. S3 shows an example of a Si3N4 PnC nanobeam featuring a soft-clamped vibrational mode for which clamping
loss contribution is suppressed. The calculation is made for a 20 nm thick beam that consists of two phononic crystal
1 µmlcell = 100 μmA0204060n02468100.10.51510ωn/2π [MHz]0�/2�ΘFωn/2π [MHz]105104103DQ108107106QBC0.1110100102103104105106105106107108109DQQωn/2π [MHz]Dλ = 1·10-5 (l = 10 mm)λ = 1.4·10-4 (l = 700 µm)λ = 2·10-3 (l = 50 µm)ldef = 120 μmε > εyieldxyzxzy 14
FIG. S4. A) Distribution of axial strain and stress along beams with strain-enhancing fillets, with geometries corresponding to
the panel B). Here lwaist/ltot = 0.08, 0.41, 0.74, 1 for red, orange, light- and dark green curves correspondingly. These geometries
feature reduced dissipation dilution as discussed in the main text and only serve for illustration of strain redistribution.
barriers, each incorporating nine 100 µm-long unit cells, and a 120 µm-long central defect region. The band diagram
for the unit cell vibrations in the out-of-plane z-direction is plotted in Fig. S3B, showing the frequencies of the
stopbands. The mode spectrum of the finite beam from panel A is shown next to the band diagram in Fig. S3B and
it can be seen that a mode localized around the defect exists in the first bandgap. The quality factors of the beam
modes, plotted in Fig. S3C, show improved Q for the localized mode due to the suppression of clamping losses. Here
the Qs of the first localized mode is approaching the ideal value Q = Qint/(nπλ)2 that a uniform beam mode would
have without clamping losses (blue line), while the Qs of the modes with frequencies outside of the PnC bandgaps
are similar to those of a regular uniform beam (blue points).
The relative advantage from using the soft clamping is thus greater for beams with smaller λ. In order to illustrate
this numerically, in Fig. S3D, we plot the localized mode DQs and absolute Qs as their frequency is varied by changing
the number of unit cells within the beam . The center defect length is fixed to be 1.2 of the unit cell length in order
to keep the soft clamped mode frequency approximately in the bandgap center. As the dissipation dilution parameter
λ is varied by changing the beam length from 50 µm to 10 mm, one can observe that both the maximum absolute
Qs and the relative Q enhancement of a soft clamped mode compared to uniform beam modes increase dramatically.
At the same time, the frequency of the highest-Q localized mode necessarily shifts down with increasing l, thus never
entering the shaded red area and taking values between
12E2
film
(1 − ν)2ρh2ω2 < DQ <
12E2
yield
ρh2ω2
.
(S50)
The area forbidden by the breaking strain is hatched gray in Fig. S3D.
V. GEOMETRICAL STRAIN ENGINEERING
Taking insights from Eq. S26, it can be seen that shaping the transverse profile of the beam allows the increase
of strain in regions where the beam is thinner than on average. To gain some intuition into this geometric strain
enhancement, it is useful to consider the evolution of strain distribution in the bowtie structure shown in Fig. S4 as
the waist is being reduced compare to the triangular supports. It can be seen from the figure that larger supports
result in larger peak strain, but smaller spatial extent of the high-strain region in the beam center. This is a general
rule that can be made quantitative as follows: strain enhancement in a thin waist between two supports of arbitrary
shapes is estimated as
(cid:18) lwaist
ltot
(cid:19)
max
avg
(cid:46) min
,
wwaist
wsupp
.
(S51)
Eq. S51 is useful not only for the evaluation of geometric strain enhancement in beams, but also in membranes. For
example, it predicts that no significant enhancement of strain takes place in the tethers of a trampoline membrane
with fillet radius smaller than the tether length.
VI. SPECTRA OF BEAMS WITH LOCAL STRAIN ENHANCEMENT
Spectra of out-of-plane vibrations of beams with engineered local strain enhancement are shown in Fig. S5. As
shown in Fig. S5C, all low-frequency flexural modes of thin-clamping beam have enhanced Qs, while the Qs of thick-
00.20.40.60.81.00.10.20.5120.20.512500.51sε [%]σ [GPa]w [arb. units]ABs 15
FIG. S5. A) Geometry and color-coded strain distribution in beams with thin (top) and thick (bottom) clampings). B)
Variation of the strain in the clamping region as a function of clamp width wcl relative to the beam center wb. C) Dissipation
dilutions and quality factors of modes of a thin-clamp beam (red), a uniform rectangular beam (blue) and a thick-clamp beam
(green). Here l = 1 mm, h = 20 nm. D) Top: strain distribution in a non-uniform PnC beam. Bottom: vibrational mode,
localized in the high-strain region of the PnC. E) Variation of the maximum strain (in the beam center) with localized mode
frequency. F) Red: modes of a l = 3 mm h = 20 nm beam with the shape from panel D. Blue: modes of uniform beam with the
same l, h. The egion with red background shows the range of DQ values that exceed the DQ of an idealized clampless beam,
but is not forbidden by the breaking strain.
clamping beams are decreased. In Fig. S5C the spectrum of a 3 mm long 20 nm thick tapered PnC beam is shown.
As in the case of a bare PnC beam, it features a "soft clamped" localized mode with dissipation dilution (and thus Q)
significantly higher than the normal value in a uniform beam, but in addition, here the localized mode Q even exceeds
that of an ideal clamp-less beam tensioned by the material strain and enters the area shaded pink. This confirms the
increase in effective strain experienced by the soft-clamped mode.
VII. COMPARISON BETWEEN 1D AND 3D SIMULATIONS OF DISSIPATION DILUTION AND
NON-FLEXURAL MODES OF PNC BEAMS WITH ENGINEERED STRAIN
In the main text we employ a one dimensional model to treat the case of doubly clamped beams and consider only
out-of-plane flexural modes. These modes are our primary interest as they exhibit the highest dissipation dilution
factors and, correspondingly, Qs. The vibrational spectrum of a beam, however, also includes other mode families like
in-plane flexural, torsional and longitudinal modes. To give an example of a complete spectrum, we simulate first 100
modes of a PnC beam with strain engineering using commercial FEM software and calculate the dissipation dilution
factors for these modes using Eq. S13. The results are presented in Fig. S6A. We have to limit the aspect ratio of the
geometry in this case to a relatively moderate value set by the length l = 500 mum and thickness h = 100 nm as we
find that the results of 3D simulation of mode shapes may not be reliable for higher aspect ratios. We also plot in
0.10.5151000.20.40.60.81.00.515100.00.51.01.52.0012345012345ε [%]σ [GPa]105104103DQQFεmax [%]σmax[GPa]ωn/2π [MHz]ωloc/2π [MHz]Eus2.01.51.00.5D108107106ε > εyield0123450.20.51250.10.20.5120.20.512500.20.40.60.81.000.10.51.01.52.0sε [%]σ [GPa]Aywclwbεcl [%]σcl [GPa]wcl/wbB0.5151050102103104DQ105106107Qwcl/wb = 0.14wcl/wb = 5.2ωn/2� [MHz]C 16
FIG. S6. A) Dissipation dilution versus mode frequency calculated using 3D vibrational mode shapes for the first 100 modes of
a doubly-clamped beam with length l = 500 µm, thickness h = 100 nm, center width w = 400 nm and transverse profile shown
in B). Orange, green, red and blue dots correspond to different mode families as explained in the caption. For comparison,
calculations using the 1D model [12] are also presented for out-of-plane flexural modes (dark blue dots). The red circle highlights
the localized "soft clamped" mode. The orange circle highlights a hybridized flexural-rotational mode, the shape of which is
shown in D). C) Equilibrium axial stress along the beam center as calculated with the 1D model[12] and 3D simulation software.
Fig. S6A the predictions of our 1D model for out-of-plane flexural modes, which are in an excellent agreement with
the orders-of-magnitude more time consuming full 3D simulations. The most prominent deviation between the 1D
and 3D calculations that we observe takes place for a flexural mode presented in Fig. S6D, that happened to hybridize
with a low-Q torsional mode and correspondingly have a reduced quality factor.
[1] L. D. Landau and E. M. Lifshitz, Theory of elasticity (London Pergamon Press, 1970).
[2] G. D. Cole, P.-L. Yu, C. Gartner, K. Siquans, R. Moghadas Nia, J. Schmole, J. Hoelscher-Obermaier, T. P. Purdy,
W. Wieczorek, C. A. Regal, et al., App. Phys. Lett. 104, 201908 (2014).
[3] A. Moridi, H. Ruan, L. C. Zhang, and M. Liu, International Journal of Solids and Structures 50, 3562 (2013).
[4] R. A. Minamisawa, M. J. Sess, R. Spolenak, J. Faist, C. David, J. Gobrecht, K. K. Bourdelle, and H. Sigg, Nature
Communications 3, 1096 (2012).
[5] T. Zabel, R. Geiger, E. Marin, E. Mller, A. Diaz, C. Bonzon, M. J. Sess, R. Spolenak, J. Faist, and H. Sigg, Journal of
Materials Research 32, 726 (2017).
[6] T. Capelle, Y. Tsaturyan, A. Barg, and A. Schliesser, Applied Physics Letters 110, 181106 (2017).
[7] G. I. Gonz´alez and P. R. Saulson, The Journal of the Acoustical Society of America 96, 207 (1994).
[8] L. Villanueva and S. Schmid, Physical Review Letters 113, 227201 (2014).
[9] A. H. Ghadimi, S. A. Fedorov, N. J. Engelsen, M. J. Bereyhi, R. Schilling, D. J. Wilson, and T. J. Kippenberg, Science
360, 764 (2018).
[10] S. A. Fedorov, V. Sudhir, R. Schilling, H. Schtz, D. J. Wilson, and T. J. Kippenberg, Physics Letters A Special Issue in
memory of Professor V.B. Braginsky, 382, 2251 (2018).
[11] Y. Tsaturyan, A. Barg, E. S. Polzik, and A. Schliesser, Nature Nanotechnology 12, 776 (2017).
[12] Mathematica package is available at zenodo.com, DOI:10.5281/zenodo.1296925.
A00.20.40.60.8100.51.01.52.0sRelaxed stress, 1D modelRelaxed stress, 3Dsimulation05101520151050100σxxωn/2π [MHz]DQfilm deposition stress σfilm BDCOut-of-plane flexuralOut-of-plane flexuralIn-plane flexuralTorsionalLongitudinal3D simulation1D modelMode familyMethodDisplacement magnitude0max |
1802.03454 | 1 | 1802 | 2018-02-09T22:40:10 | Hybrid integration of silicon photonics circuits and InP lasers by photonic wire bonding | [
"physics.app-ph",
"physics.optics"
] | Efficient coupling of III-V light sources to silicon photonic circuits is one of the key challenges of integrated optics. Important requirements are low coupling losses, as well as small footprint and high yield of the overall assembly, along with the ability to use automated processes for large-scale production. In this paper, we demonstrate that photonic wire bonding addresses these challenges by exploiting direct-write two-photon lithography for in-situ fabrication of three-dimensional freeform waveguides between optical chips. In a series proof-of-concept experiments, we connect InP-based horizontal-cavity surface emitting lasers (HCSEL) to passive silicon photonic circuits with insertion losses down to 0.4 dB. To the best of our knowledge, this is the most efficient interface between an InP light source and a silicon photonic chip that has so far been demonstrated. Our experiments represent a key step in advancing photonic wire bonding to a universal integration platform for hybrid photonic multi-chip assemblies that combine known-good dies of different materials to high-performance hybrid multi-chip modules. | physics.app-ph | physics | Hybrid integration of silicon photonics circuits
and InP lasers by photonic wire bonding
Muhammad Rodlin Billah,1,2 Matthias Blaicher,1,2 Tobias Hoose,1,2
Philipp-Immanuel Dietrich,1,2,3 Pablo Marin-Palomo,2 Nicole Lindenmann,1,2
Aleksandar Nesic,2 Andreas Hofmann,4 Ute Troppenz,5 Martin Moehrle,5
Sebastian Randel,2 Wolfgang Freude,2 and Christian Koos1,2,3,*
1Institute of Microstructure Technology (IMT), Karlsruhe Institute of Technology (KIT), Hermann-von-Helmholtz-Platz 1, 76344
Eggenstein-Leopoldshafen, Germany
2Institute of Photonics and Quantum Electronics (IPQ), Karlsruhe Institute of Technology (KIT) Engesserstrasse 5, 76131
Karlsruhe, Germany
3Vanguard Photonics GmbH, Karlsruhe, Germany
4Institute of Applied Computer Science (IAI), Karlsruhe Institute of Technology (KIT), Hermann-von-Helmholtz-Platz 1, 76344
Eggenstein-Leopoldshafen, Germany
5Fraunhofer Institute for Telecommunications, Heinrich Hertz Institute (HHI), Einsteinufer 37, 10587 Berlin, Germany
*Corresponding author: [email protected]
Abstract: Efficient coupling of III-V light sources to silicon photonic circuits is one of the key challenges of integrated
optics. Important requirements are low coupling losses, as well as small footprint and high yield of the overall
assembly, along with the ability to use automated processes for large-scale production. In this paper, we demonstrate
that photonic wire bonding addresses these challenges by exploiting direct-write two-photon lithography for in-situ
fabrication of three-dimensional freeform waveguides between optical chips. In a series proof-of-concept experiments,
we connect InP-based horizontal-cavity surface emitting lasers (HCSEL) to passive silicon photonic circuits with
insertion losses down to 0.4 dB. To the best of our knowledge, this is the most efficient interface between an InP light
source and a silicon photonic chip that has so far been demonstrated. Our experiments represent a key step in
advancing photonic wire bonding to a universal integration platform for hybrid photonic multi-chip assemblies that
combine known-good dies of different materials to high-performance hybrid multi-chip modules.
REFERENCES
1. C. R. Doerr, "Silicon photonic integration in telecommunications," Front. Phys. 3, 1–16 (2015).
2. D. Thomson, A. Zilkie, J. E. Bowers, T. Komljenovic, G. T. Reed, L. Vivien, D. Marris-Morini, E. Cassan, L. Virot, J.-M. Fédéli, J.-M.
Hartmann, J. H. Schmid, D.-X. Xu, F. Boeuf, P. O'Brien, G. Z. Mashanovich, and M. Nedeljkovic, "Roadmap on silicon photonics,"
J. Opt. 18, 73003 (2016).
3. A. E. J. Lim, J. Song, Q. Fang, C. Li, X. Tu, N. Duan, K. K. Chen, R. P. C. Tern, and T. Y. Liow, "Review of Silicon Photonics
Foundry Efforts," IEEE J. Sel. Top. Quantum Electron. 20, (2014).
4. M. J. R. Heck, J. F. Bauters, M. L. Davenport, D. T. Spencer, and J. E. Bowers, "Ultra-low loss waveguide platform and its
integration with silicon photonics," Laser Photonics Rev. 8, 667–686 (2014).
5. K. Yamada, "Silicon Photonics II," 119, 1–29 (2011).
6. G. T. Reed, G. Z. Mashanovich, F. Y. Gardes, M. Nedeljkovic, Y. Hu, D. J. Thomson, K. Li, P. R. Wilson, S. W. Chen, and S. S.
Hsu, "Recent breakthroughs in carrier depletion based silicon optical modulators," Nanophotonics 3, 229–245 (2014).
7. D. Thomson, F. Gardes, S. Liu, H. Porte, L. Zimmermann, J. M. Fedeli, Y. Hu, M. Nedeljkovic, X. Yang, P. Petropoulos, and G.
Mashanovich, "High Performance Mach Zehnder Based Silicon Optical Modulators," IEEE J. Sel. Top. Quantum Electron. 19,
3400510 (2013).
8. M. Casalino, G. Coppola, R. M. De La Rue, and D. F. Logan, "State-of-the-art all-silicon sub-bandgap photodetectors at telecom
and datacom wavelengths," Laser Photonics Rev. 10, 895–921 (2016).
9. J. Michel, J. Liu, and L. C. Kimerling, "High-performance Ge-on-Si photodetectors," Nat. Photonics 4, 527–534 (2010).
10. P. O. Brien, L. Carrol, C. Eason, and J. S. Lee, Silicon Photonics III, Topics in Applied Physics (Springer Berlin Heidelberg, 2016),
Vol. 122.
11. N. Higashitarumizu and Y. Ishikawa, "Enhanced direct-gap light emission from Si-capped n^+-Ge epitaxial layers on Si after post-
growth rapid cyclic annealing: impact of non-radiative interface recombination toward Ge/Si double heterostructure lasers," Opt.
Express 25, 21286 (2017).
12. S. Wirths, R. Geiger, N. von den Driesch, G. Mussler, T. Stoica, S. Mantl, Z. Ikonic, M. Luysberg, S. Chiussi, J. M. Hartmann, H.
Sigg, J. Faist, D. Buca, and D. Grützmacher, "Lasing in direct-bandgap GeSn alloy grown on Si," Nat. Photonics 9, 88–92 (2015).
13. S. Chen, W. Li, J. Wu, Q. Jiang, M. Tang, S. Shutts, S. N. Elliott, A. Sobiesierski, A. J. Seeds, I. Ross, P. M. Smowton, and H. Liu,
"Electrically pumped continuous-wave III–V quantum dot lasers on silicon," Nat. Photonics 10, 307–311 (2016).
14. J. Wang, X. Ren, C. Deng, H. Hu, Y. He, Z. Cheng, H. Ma, Q. Wang, and Y. Huang, "Extremely Low-Threshold Current Density
InGaAs / AlGaAs Quantum-Well Lasers on Silicon," 33, 3163–3169 (2015).
15. S. Chen, M. Tang, J. Wu, Q. Jiang, V. G. Dorogan, M. Benamara, Y. I. Mazur, G. J. Salamo, A. Seeds, and H. Liu, "1.3??m
InAs/GaAs quantum-dot laser monolithically grown on Si substrates using InAlAs/GaAs dislocation filter layers," Conf. Dig. - IEEE
Int. Semicond. Laser Conf. 22, 88–89 (2014).
16. G.-H. Duan, S. Olivier, C. Jany, S. Malhouitre, A. Le Liepvre, A. Shen, X. Pommarede, G. Levaufre, N. Girard, D. Make, G. Glastre,
J. Decobert, F. Lelarge, R. Brenot, and B. Charbonnier, "Hybrid III-V Silicon Photonic Integrated Circuits for Optical Communication
Applications," IEEE J. Sel. Top. Quantum Electron. 22, 379–389 (2016).
17. A. Abbasi, S. Keyvaninia, J. Verbist, X. Yin, J. Bauwelinck, F. Lelarge, G.-H. Duan, G. Roelkens, and G. Morthier, "43 Gb/s NRZ-
OOK Direct Modulation of a Heterogeneously Integrated InP/Si DFB Laser," J. Light. Technol. 35, 1235–1240 (2017).
18. M. Buffolo, M. Meneghini, C. De Santi, M. L. Davenport, J. E. Bowers, G. Meneghesso, and E. Zanoni, "Degradation Mechanisms
of Heterogeneous III-V/Silicon 1.55-µm DBR Laser Diodes," IEEE J. Quantum Electron. 53, 1–8 (2017).
19. J. Justice, C. Bower, M. Meitl, M. B. Mooney, M. A. Gubbins, and B. Corbett, "Wafer-scale integration of group III–V lasers on
silicon using transfer printing of epitaxial layers," Nat. Photonics 6, 612–616 (2012).
20. X. Luo, Y. Cao, J. Song, X. Hu, Y. Cheng, C. Li, C. Liu, T.-Y. Liow, M. Yu, H. Wang, Q. J. Wang, and P. G.-Q. Lo, "High-
Throughput Multiple Dies-to-Wafer Bonding Technology and III/V-on-Si Hybrid Lasers for Heterogeneous Integration of
Optoelectronic Integrated Circuits," Front. Mater. 2, 1–21 (2015).
21. A. De Groote, P. Cardile, A. Z. Subramanian, A. M. Fecioru, C. Bower, D. Delbeke, R. Baets, and G. Roelkens, "Transfer-printing-
based integration of single-mode waveguide-coupled III-V-on-silicon broadband light emitters," Opt. Express 24, 13754 (2016).
22. B. Snyder, B. Corbett, and P. Obrien, "Hybrid integration of the wavelength-tunable laser with a silicon photonic integrated circuit,"
J. Light. Technol. 31, 3934–3942 (2013).
23. H. Lu, J. S. Lee, Y. Zhao, C. Scarcella, P. Cardile, A. Daly, M. Ortsiefer, L. Carroll, and P. O'Brien, "Flip-chip integration of tilted
VCSELs onto a silicon photonic integrated circuit," Opt. Express 24, 16258–16266 (2016).
24. S. Lin, X. Zheng, J. Yao, S. S. Djordjevic, J. E. Cunningham, J.-H. Lee, I. Shubin, Y. Luo, J. Bovington, D. Y. Lee, H. D. Thacker, K.
Raj, and A. V. Krishnamoorthy, "Efficient, tunable flip-chip-integrated III-V/Si hybrid external-cavity laser array," Opt. Express 24,
21454 (2016).
25. B. Song, P. Contu, C. Stagarescu, S. Pinna, P. Abolghasem, S. Ristic, N. Bickel, J. Bowker, A. Behfar, and J. Klamkin, "3D
integrated hybrid silicon laser," in 2015 European Conference on Optical Communication (ECOC) (IEEE, 2015), Vol. 2015–Novem,
pp. 1–3.
26. G. de Valicourt, C.-M. Chang, M. S. Eggleston, A. Melikyan, C. Bolle, N. Kaneda, M. P. Earnshaw, Y.-K. Chen, A. Maho, R. Brenot,
and P. Dong, "Hybrid-Integrated Wavelength and Reflectivity Tunable III–V/Silicon Transmitter," J. Light. Technol. 35, 1376–1382
(2017).
27. A. Moscoso-Mártir, J. Müller, J. Hauck, N. Chimot, R. Setter, A. Badihi, D. E. Rasmussen, A. Garreau, M. Nielsen, E. Islamova, S.
Romero-García, B. Shen, A. Sandomirsky, S. Rockman, C. Li, S. Sharif Azadeh, G. Q. Lo, E. Mentovich, F. Merget, F. Lelarge, and
J. Witzens, "Silicon Photonics Transmitter with SOA and Semiconductor Mode-Locked Laser," Sci. Rep. 7, 1–15 (2017).
28. Z. Zhou, B. Yin, and J. Michel, "On-chip light sources for silicon photonics," Light Sci. Appl. 4, e358 (2015).
29. Y. Urino, T. Nakamura, and Y. Arakawa, Silicon Photonics III, Topics in Applied Physics (Springer Berlin Heidelberg, 2016), Vol.
122.
30. Z. Wang, A. Abbasi, U. Dave, A. De Groote, S. Kumari, B. Kunert, C. Merckling, M. Pantouvaki, Y. Shi, B. Tian, K. Van Gasse, J.
Verbist, R. Wang, W. Xie, J. Zhang, Y. Zhu, J. Bauwelinck, X. Yin, Z. Hens, J. Van Campenhout, B. Kuyken, R. Baets, G. Morthier,
D. Van Thourhout, and G. Roelkens, "Novel Light Source Integration Approaches for Silicon Photonics," Laser Photonics Rev. 11,
1–21 (2017).
31. M. J. R. Heck and J. E. Bowers, "Energy efficient and energy proportional optical interconnects for multi-core processors: Driving
the need for on-chip sources," IEEE J. Sel. Top. Quantum Electron. 20, (2014).
32. I. Lucci, M. Bahri, Y. Leger, and C. Cornet, "Thermal management of monolithic and heterogeneous integrated lasers," in 2016
Compound Semiconductor Week (CSW) [Includes 28th International Conference on Indium Phosphide & Related Materials (IPRM)
& 43rd International Symposium on Compound Semiconductors (ISCS) (IEEE, 2016), Vol. 22, pp. 1–1.
33. J. S. Lee, L. Carroll, C. Scarcella, N. Pavarelli, S. Menezo, S. Bernabe, E. Temporiti, and P. O'Brien, "Meeting the Electrical,
Optical, and Thermal Design Challenges of Photonic-Packaging," IEEE J. Sel. Top. Quantum Electron. 22, 409–417 (2016).
34. B. Pezeshki, J. Heanue, D. Ton, T. Schrans, S. Rangarajan, S. Zou, G. W. Yoffe, A. Liu, M. Sherback, J. Kubicky, and P. Ludwig,
"High performance MEMS-based micro-optic assembly for multi-lane transceivers," J. Light. Technol. 32, 2796–2799 (2014).
35. D. Liang and J. E. Bowers, "Recent progress in lasers on silicon," Nat. Photonics 4, 511–517 (2010).
36. N. Hatori, T. Shimizu, M. Okano, M. Ishizaka, T. Yamamoto, Y. Urino, M. Mori, T. Nakamura, and Y. Arakawa, "A hybrid integrated
light source on a silicon platform using a trident spot-size converter," J. Light. Technol. 32, 1329–1336 (2014).
37. N. Lindenmann, I. Kaiser, G. Balthasar, R. Bonk, D. Hillerkuss, W. Freude, J. Leuthold, and C. Koos, "Photonic Waveguide Bonds -
A Novel Concept for Chip-to-Chip Interconnects," Opt. Fiber Commun. Conf. Fiber Opt. Eng. Conf. 2011 PDPC1 (2011).
38. A. Nesic, M. Blaicher, T. Hoose, and M. Lauermann, "Hybrid 2D / 3D Photonic Integration for Non-Planar Circuit Topologies,"
Eceoc 2016 2, 707–709 (2016).
39. N. Lindenmann, G. Balthasar, D. Hillerkuss, R. Schmogrow, M. Jordan, J. Leuthold, W. Freude, and C. Koos, "Photonic wire
bonding: a novel concept for chip-scale interconnects.," Opt. Express 20, 17667–77 (2012).
40. N. Lindenmann, S. Dottermusch, M. L. Goedecke, T. Hoose, M. R. Billah, T. P. Onanuga, A. Hofmann, W. Freude, and C. Koos,
"Connecting Silicon Photonic Circuits to Multicore Fibers by Photonic Wire Bonding," Light. Technol. J. 33, 755–760 (2015).
41. M. R. Billah, T. Hoose, T. Onanuga, N. Lindenmann, P. Dietrich, T. Wingert, M. L. Goedecke, A. Hofmann, U. Troppenz, A.
Sigmund, M. Möhrle, W. Freude, and C. Koos, "Multi-Chip Integration of Lasers and Silicon Photonics by Photonic Wire Bonding,"
in Conference on Lasers and Electro-Optics (OSA, 2015), p. STu2F.2.
42. K. Wörhoff, R. G. Heideman, A. Leinse, and M. Hoekman, "TriPleX: a versatile dielectric photonic platform," Adv. Opt. Technol. 4,
189–207 (2015).
43. M. R. Billah, M. Blaicher, J. N. Kemal, T. Hoose, H. Zwickel, P.-I. Dietrich, U. Troppenz, M. Möhrle, F. Merget, A. Hofmann, J.
Witzens, S. Randel, W. Freude, and C. Koos, "8-channel 448 Gbit/s Silicon Photonic Transmitter Enabled by Photonic Wire
Bonding," in Optical Fiber Communication Conference Postdeadline Papers (OSA, 2017), p. Th5D.6.
44. M. R. Billah, J. N. Kemal, M. Blaicher, Y. Kutuvantavida, and C. Kieninger, "Four-Channel 784 Gbit / s Transmitter Module Enabled
by Photonic Wire Bonding and Silicon-Organic Hybrid Modulators," in 2017 European Conference on Optical Communication,
2 of 9
p. Th.PDP.C.1.
45. "IP-PHOTORESISTS," http://www.nanoscribe.de/files/1814/0662/4393/IP-Resist_IP-Dip_web.pdf.
46. M. Moehrle, J. Kreissl, W. D. Molzow, G. Przyrembel, C. Wagner, A. Sigmund, L. Moerl, and N. Grote, "Ultra-low threshold 1490
nm surface-emitting BH-DFB laser diode with integrated monitor photodiode," Indium Phosphide Relat. Mater. (IPRM), 2010 Int.
Conf. 1–4 (2010).
47. K. Kikuchi, "Characterization of semiconductor-laser phase noise and estimation of bit-error rate performance with low-speed offline
digital coherent receivers," Opt. Express 20, 5291 (2012).
48. "CST - Computer Simulation Technology," https://www.cst.com/.
1. INTRODUCTION
The silicon photonic platform has evolved into a mainstay of high-density photonic integration [1], exploiting
advanced CMOS processes for high-yield mass fabrication [2,3] of a wide variety of passive photonic
devices [4,5], electro-optic modulators [6,7], or photodetectors [8,9]. Cost-efficient and technically viable
integration of high-performance light sources into silicon photonic circuits, however, still remains a challenge [10].
Such light sources should combine low power consumption and high efficiency with small footprint and good
thermal coupling to package-level heat-sinks while still maintaining good manufacturability and the potential for
automated large-scale production. Moreover, as system complexity increases, testing of active components prior to
integration into the final system is becoming more and more important to achieve high fabrication yield. While
substantial progress has been made
[11,12] or
heteroepitaxial [13–15] growth of direct-bandgap compound semiconductors on silicon substrates, wafer-level [16–
18] or die-level [19–21] transfer of III-V layer stacks or even complete devices [22–26] to silicon substrates is still
considered the most practical path towards an efficient light source integration. In this context, two main approaches
have been pursued, which are often referred to as heterogeneous integration and hybrid integration [27–30].
light sources by direct epitaxial
towards
realizing
Heterogenous integration is based on bonding of unpatterned III–V dies onto pre-processed silicon photonic
(SiP) wafers such that light generated in the epitaxial layers of the III–V die is evanescently coupled to the SiP
waveguides [16–21]. III-V devices are then fabricated by wafer-scale processing of the dies, where all structures
are lithographically aligned with highest precision. High-precision positioning of dies is hence not necessary – a key
advantage compared to hybrid integration. However, while heterogeneous integration is particularly well suited for
large-scale mass fabrication of III-V light sources on SiP circuits, the associated technical complexity is still
considerable, in particular with respect to the stringent requirements of ultraclean and extremely smooth surfaces.
Moreover, heterogeneous integration does not permit testing of light sources prior to integration into more complex
systems and hence requires tight process control to maintain high yield. As a consequence, heterogeneous
integration is mainly suited for high-volume applications that justify associated technological overhead, e.g., in the
context of optical on-chip networks that require integration of tens of light sources onto a single chip [31]. Moreover,
heterogeneously integrated light sources may consume considerable real estate on the SiP chip, and heat-sinking
is challenging due to the high thermal resistance introduced by the III-V-to-silicon bonding layer and by the buried
oxide [32].
Hybrid integration [22–26], in contrast, relies on optically connecting readily processed III-V lasers [22–24],
gain chips [25,26], or even photodiodes [27] to silicon photonic (SiP) circuits, where the III-V device may either be
mounted on top of the silicon substrate [22–25,27] or next to it [26]. Hybrid integration maintains the superior
performance characteristics of native III-V light sources and allows for testing of devices prior to system assembly,
but comes with fabrication challenges. In particular, efficient optical coupling of the III-V to the SiP waveguides
usually relies on alignment with precisions in the lower micrometer or even sub-µm range. This often requires slow
and expensive [33] active alignment techniques, where the coupling efficiency is continuously monitored while
optimizing the position of the devices [10,34]. Moreover, additional devices such as micro-lenses, prisms, or micro-
mechanical carriers are needed to adapt the mode field size and the emission direction of the III-V light source to
that of the SiP circuit [22], leading to comparatively big assemblies. In many cases, the III-V devices are mounted
on top of the SiP die. This approach does not only consume substantial on-chip real estate, but also poses
challenges with respect to heat-sinking of the III-V devices through the underlying silicon-on-insulator (SOI)
substrate due to the relatively poor heat conductivity of the buried oxide [23,28,35]. In hybrid integration, optical
coupling losses of 2.3 dB have been previously demonstrated for butt coupling of a III-V laser diode array to an
array of SiP waveguides equipped with trident spot-size converters [36] that relax alignment tolerances to ± 0.7 µm.
In this paper, we demonstrate that the technique of photonic wire bonding [37–41] enables highly flexible low-loss
coupling of InP light sources to SiP chips, maintaining all performance and flexibility advantages of hybrid
integration approaches while offering a path towards highly scalable automated mass production. Photonic wire
bonding exploits in-situ additive nanofabrication of freeform polymer waveguides between pre-positioned photonic
chips. The three-dimensional shape of the photonic wire bonds can be adapted to the exact positions of the chips
such that high-precision alignment of chips becomes obsolete, rendering the technique amenable to automated
mass production. Building upon our previous results [37–41], we demonstrate highly efficient coupling between InP-
based horizontal-cavity surface-emitting lasers (HCSEL) and silicon photonic chips with coupling losses down to
3 of 9
Fig. 1. Vision of a photonic multi-chip transmitter for wavelength-division multiplexing (WDM) communications. (a) The system exploits
photonic wire bonds to combine the distinct advantages of different photonic integration platforms: Distributed-feedback (DFB) lasers
based on direct-bandgap InP-substrates, whereas silicon photonic (SiP) chips lend themselves for implementing electro-optic
modulators. For dense packing of optical channels, high-quality arrayed waveguide gratings (AWG) are needed, which are best
realized on medium index-contrast material systems such as TriPleX [42]. The functionality of the multi-chip system depends vitally on
efficient chip-to-chip and chip-to-fiber interconnects, which are realized by photonic wire bonds (PWB). The focus of this work is on low-
loss PWB links between InP lasers and SiP chips (red colored). (b) SEM image of a laser-chip interface. (c) Chip-to-chip-connection
between two SiP dies [39]. (d) Fiber-to-chip interface using PWB to connect the individual cores of a multi-core fiber (MCF) to an array
of planar SOI waveguides [40].
0.4 dB. In these assemblies, the laser source is placed side-by-side to the SiP chip, thus allowing for good thermal
coupling to an underlying chip-level heat sink without consuming any on-chip real estate. We experimentally confirm
that the emission performance of the lasers is not impaired by the photonic wire bonding process. Combined with
previously demonstrated chip-to-chip [39] and fiber-to-chip [40] connections, our experiments represent a key step
in advancing photonic wire bonding to a universal integration platform for hybrid photonic multi-chip assemblies that
combine known-good dies of different materials while maintaining their individual high-performance characteristics.
2. CONCEPT AND BACKGROUND
A vision of a hybrid photonic multi-chip module enabled by photonic wire bonds is illustrated in Fig. 1(a) using a
wavelength-division multiplexing (WDM) transmitter as an example. The system combines the distinct advantages
of different photonic integration platforms: Distributed-feedback (DFB) lasers are used as optical sources for the
various wavelength channels and implemented on indium phosphide (InP) substrates, whereas silicon photonics
(SiP) chips are used to realize densely integrated electro-optic IQ modulators which encode information on the
various optical carriers. For dense packing of WDM channels, high-quality optical filters, e. g., arrayed waveguide
gratings (AWG), are needed. These devices are best realized on the basis of a medium index-contrast material
system such as the commercial integration platform TriPleX [42]. Hybrid multi-chip integration allows on-chip
subsystems to be tested prior to integration which significantly increases yield as compared to monolithic or
heterogeneous integration approaches, where failure of a single component impairs the functionality of the entire
system. In previous experiments [39,40], we have shown photonic wire bonds to enable particularly efficient single-
mode chip-to-chip and fiber-to-chip connections, Fig. 1(c) and (d), and the viability of the concept has been recently
demonstrated by realizing multi-chip transmitter modules for high-speed communications [43,44]. The focus of this
paper is on low-loss coupling of InP-based light sources to SiP chips as one of the most important interfaces for
practical applications, Fig. 1(b).
4 of 9
Fig. 2. Multi-chip assembly demonstrating a PWB between an InP laser and a SiP chip. (a) Assembly concept: The laser and the SiP
die are mounted on a common carrier by adhesive bonding. The carrier compensates differences of die thickness. Precise alignm ent
between the laser emission window and the Si nanowire is not required. Tapers are utilized at the interfaces at both ends of the PWB
for a low-loss transition between the PWB waveguide (WG) section and the attached components. A grating coupler (GC) is used to
interface the SiP waveguide to a standard single-mode fiber (SMF). The WG section is designed to have a rectangular cross-section of
2.0 µm × 1.4 µm. (b) HCSEL interface: The HCSEL consists of an in-plane InGasAsP DFB laser cavity and an etched 45° mirror to
deflect the light to the surface-normal direction. The light is then captured by a (rectangular) polymer PWB taper. At the HCSEL
emission window, the taper cross section corresponds to a square with a side length of 4 µm, which is linearly converted to a
rectangular cross section with size of 2.0 µm × 1.4 µm at the transition to the PWB waveguide section. (c, d) Side and top view of the
transition to the SiP nanowire waveguide. The polymer PWB taper starts with a rectangular 2.0 µm × 1.4 µm cross section of the PWB
waveguide section, which is linearly converted to a final width of 0.8 µm and a height of 0.5 µm along a length of 60 µm. Alignment
markers are used to exactly locate the coupling interface to the position of the targeted nanowire.
3. REALIZATION OF MULTI-CHIP MODULES
A. Assembly
In our experiment, we realize simple multi-chip modules (MCM) that use photonic wire bonds (PWB) to connect
passive SiP circuits to InP light sources, see Fig. 2(a). The assembly of the MCM requires a common carrier where
laser and SiP chips are mounted first with medium precision by adhesive bonding. Height differences between the
chips are compensated by the carrier to roughly align the top surfaces. The output port of the InP chip is then
interconnected to a silicon (Si) nanowire on the SiP chip by a 3D freeform PWB. At the end of the Si nanowire, a
grating coupler (GC) is used to direct the light out of plane into a single-mode fiber (SMF). Reference nanowires
with GC at both ends are also located on the same SiP chip for calibration. The PWB is fabricated in-situ by two-
photon lithography [39] and comprises two tapered sections to minimize mode-field mismatch at both interfaces.
The shape of the PWB is adapted to the position of these interfaces, thereby compensating tolerances in chip
placement.
B. Photoresist materials and devices
Photonic wire bonds are fabricated from a commercially available negative-tone photoresist (IP-Dip, Nanoscribe
GmbH, unexposed refractive index nPWB = 1.52 at 780 nm [45]). The MCM are built with so-called horizontal-cavity
surface-emitting lasers (HCSEL) [46] which combine an in-plane InGaAsP-based buried heterostructure
distributed- feedback (DFB) laser with an etched 45° mirror to deflect the emitted light to surface-normal direction,
Fig. 2(b). At the HCSEL facet, the 1/e2-diameter of the modal intensity profile amounts to 3 µm, extracted from a
measurement of the far-field intensity distribution by using a scanning-slit optical beam profiler (BP209-IR/M,
Thorlabs). The HCSEL deflection mirror is fabricated by chemically assisted ion beam etching [46]. Note that the
mirror inclination is subject to fabrication tolerances such that the light emission direction might not be perpendicular
to the chip surface, i.e., θ ≠ 90° in Fig. 2(b).
The PWB consists of a waveguide (WG) section, which is connected to the HCSEL and the SiP chip by dedicated
tapers, Fig. 2(a). On the HCSEL side, a taper with a rectangular cross section is used to convert the mode of the
laser to the fundamental mode of the PWB waveguide section. At the HCSEL emission window, Fig. 2(b), the taper
cross section corresponds to a square with a side length of 4 µm, which is then linearly converted into a rectangular
cross section of 2.0 µm × 1.4 µm at the transition to the PWB waveguide section. At the interfaces to the SiP circuit,
low-loss coupling is achieved by a tapered Si nanowire, which is embedded into the polymer PWB taper [39], see
Fig. 2(c, d). The Si nanowires (nSi = 3.48) were produced on a CMOS pilot line using 193 nm deep-ultraviolet (DUV)
lithography with standard height of 220 nm. The thickness of buried oxide (SiO2, nSiO2 = 1.44) amounts to 2 µm. The
Si nanowire starts with a tip width of 0.13 µm, which then is tapered up to the final SiP waveguide width of 0.5 µm
along a length of 60 µm. The polymer PWB taper starts with the rectangular 2.0 µm × 1.4 µm cross section of the
PWB waveguide section, which is linearly converted to a final width of 0.8 µm and a height of 0.5 µm along the
same length of 60 µm. The smallest radius of curvature typically used in the WG section of the PWB is 40 µm.
5 of 9
Fig. 3. Hybrid MCM combining passive silicon photonic (SiP) waveguides with an InP DFB laser array. (a) Scanning electron
microscope (SEM) micrograph of the MCM, comprising four laser diodes (LD1 … LD4), each connected to a SiP on-chip waveguide
via a photonic wire bond (PWB1 … PWB4). The insertion losses of the photonic wire bonds are denoted in the figure. PWB2 shows an
insertion loss of (0.4 ± 0.3) dB, including the coupling losses of both the HCSEL-PWB interface and the transition to the SiP nanowire
as well as the propagation loss in the freeform waveguide. Slightly higher losses of (1.3 ± 0.4) dB, (0.6 ± 0.3) dB, and (0.6 ± 0.3) dB
are observed for PWB 1, 3, and 4, respectively. The variations are attributed to uncertainties of the HCSEL emission spot size, of the
emission direction, and of the PWB shape. Still, these losses are well below the 2.3 dB obtained for other concepts that rely on active
alignment [36]. (b) Measurement of emission direction of a HCSEL before photonic wire bonding. The measurement was taken by
using a goniometric radiometer for three HCSEL chips. The Gaussian fit (red) reveals an emission angle θair = (83 ± 1)°, which deviates
from the ideal 90° due to fabrication tolerances of the HCSEL deflection. (c) Side view of PWB1. For efficient coupling into the PWB
taper (nPWB = 1.52), the taper axis has to be inclined by approximately θPWB = 85° according to Snell's law.
C. Photonic wire bond fabrication
Several steps are required for fabricating the PWB [37–40]. After mounting the components on a common carrier
with typical accuracies of 10 µm, the negative-tone photoresist is drop-cast onto the MCM, covering both the
HCSEL emission window and the coupling region of the targeted Si nanowire. The position of the coupling
interfaces is then detected in the volume of the resist through the observation camera of the 3D lithography system
(Photonic Professional GT, Nanoscribe GmbH). These positions may be derived from dedicated alignment markers,
see Fig. 2(d). The start and end points of the PWB allow to define its trajectory along with its 3D shape, and the
structure is then fabricated through two-photon polymerization by exposing the photoresist. The lithography laser
has an emission wavelength of λ = 780 nm and emits pulses with a full-width at half-maximum (FWHM) below 100
fs and with a repetition frequency of 80 MHz. In our current experiments, we have not yet optimized the writing
speed, leading to an exposure time of approximately three minutes for a single PWB with vast potential for further
acceleration. The unexposed photoresist is finally removed in a two-step development process using propylene-
glycol-methyl-ether-acetate (PGMEA) for the first 15 minutes followed by isopropyl-alcohol (2-propanol) inside a
critical point dryer (CPD 300, Leica Microsystems GmbH) for 50 minutes. Low-refractive index matching liquid
(Cargille Laser Liquid 3421, noil = 1.3 at 1550 nm) is drop-cast onto the assembly to emulate the low-refractive-
index cladding of the PWB. The liquid can easily be replaced with a long-term protective cladding material.
4. RESULTS AND DISCUSSION
A. Photonic wire bonding of laser array
To demonstrate the viability of the photonic wire bonding approach, we fabricate a hybrid MCM that combines
passive silicon photonic waveguides with an array of DFB lasers that feature different wavelengths. An SEM
micrograph of the assembly is shown in Fig. 3(a). Note that due to fabrication tolerances of the HCSEL deflection
mirror inclination, the angle θ between the light emission direction and the chip surface is not exactly 90°, and
hence the direction of the PWB trajectory has to be adapted accordingly, see Fig. 3(b, c). Using a goniometric
radiometer (LD 8900, Ophir Spiricon Europe GmbH), we find that actual emission angle amounts to θair = (83 ± 1)°
when measured in air (nair = 1), Fig. 3(b). For coupling into the polymer taper of the PWB (nPWB = 1.52), this
changes to approximately θPWB = 85° according to Snell's law, and the axes of the polymer tapers have to be
adapted accordingly, Fig. 3(c).
6 of 9
Fig. 4. PWB performance characterizations. (a) Experimental setup for measuring insertion losses of the PWB as well as laser
linewidth. The electrical path (green) consists of a current source to drive the laser diode (LD). The optical path comprises the PWB
(orange) as well as a SiP nanowires grating coupler (GC) and an optical single-mode fiber (all black). At the end of the single-mode
fiber (SMF), the light is coupled into a calibrated integrating sphere (IS) to precisely measure the optical power. Alternati vely, the
emitted light can be coupled to a heterodyne coherent receiver, which contains a highly stable external-cavity laser (ECL, not depicted)
as a local oscillator for linewidth measurements. A polarization controller (Pol. Contr.) is used for optimizing the beat signal between the
recorded light and the LO reference. (b) Power-current characteristic of LD2 and LD3 measured at the laser facet before photonic wire
bonding (P0, blue squares), and inside the respective SiP nanowire after photonic wire bonding (P1, red squares). The power level P1
was corrected by taking into account the losses caused by propagation through the SiP chip, the corresponding GC, and the SMF patch
cord. The dashed lines represent linear fits. The PWB insertion losses amount to 0.4 dB and to 0.6 dB respectively. (c) Optical linewidth
of LD5 measured before and after photonic wire bonding. The full-width at half-maximum of the power spectrum amounts to
∆f = 2.7 MHz without and ∆f = 3.4 MHz with a PWB. Within the measurement accuracy, we do not see a significant increase of the laser
linewidth. These findings were confirmed using a variety of other devices.
B. Insertion loss of PWB
In order to precisely measure the insertion loss (IL) of each PWB, we first determine the current-dependent output
power at each HCSEL facet before and after photonic wire bonding, see Fig. 4(a). This is done by electrically
pumping each laser diode (LD) with an adjustable current ILD, derived from a laser diode driver (LDX 3620, ILX
Lightwave). In a first measurement, we record the total output power of the bare HCSEL before photonic wire
bonding. To this end, we bring an integrating sphere (IS, S145C, Thorlabs) in close proximity to the HCSEL facet
such that the entire emitted power P0 is captured. In a second measurement, which we perform after photonic wire
bonding, we use the same integrating sphere to determine the optical power after propagation through the SiP chip,
the corresponding grating coupler (GC) and a single-mode fiber (SMF) patch cord. In this measurement, we simply
insert the SMF end facet into the IS to measure the power. For this measurement, we also consider the 4 % back
reflection at the SMF end facet. Taking into account the on-chip propagation losses of the 472 µm-long SiP
nanowire, the GC losses, and the connector losses of the SMF, we can then estimate the power P1 delivered to the
SiP nanowire. The wavelength-dependent GC losses and the on-chip propagation losses are obtained from
reference SiP waveguides on the same chip. The transmission spectra of these waveguides are measured using a
tunable laser and a synchronously swept optical spectrum analyzer (OSA). We investigate waveguides of different
lengths, leading to on-chip propagation losses of approximately 3 dB/cm and grating coupler losses of 4.1 dB for the
optimum wavelength of 1550 nm. For calculating the PWB insertion losses, we use the GC insertion loss at the
emission wavelength of the respective HCSEL.
Figure 4(b) shows the measurement results for the emitted HCSEL power P0 (blue, w/o PWB) and P1 (red, w/
PWB) for LD2 and LD3. From the ratio of P0 and P1, we estimate the PWB IL to be (0.4 ± 0.3) dB and (0.6 ± 0.3)
dB, respectively. Comparing the thresholds from both curves with and without PWB in Fig. 4(b), we confirm that the
HCSEL thresholds stay at 10 mA before and after photonic wire bonding. Using the same method, we estimate the
remaining PWB IL of LD1 and LD4 to be (1.3 ± 0.3) dB and (0.6 ± 0.3) dB, respectively, see Fig. 3(a). The IL
differences are attributed to uncertainties of the HCSEL emission spot size and the emission direction as well as to
slight variations of the PWB shape. Note that these losses are well below the 2.3 dB, which were obtained for other
concepts that rely on active alignment [36].
C. Laser performance without and with PWB
When connecting InP lasers to SiP chips, the use of intermediate optical isolators is not practical. An important
question is then whether back-reflection of optical power into the laser cavity will degrade the emission
performance, in particular with respect to the optical linewidth. This is investigated by first measuring the emission
7 of 9
Fig. 5. Numerical verification and benchmarking of the PWB. (a) Calculated normalized intensity of the PWB. Simulations were
performed using a vectorial finite-integration technique (FIT, CST Microwave Studio [48]). Light is launched into Port 1, which is located
at the bottom of the PWB input taper connected to the HCSEL, and Port 2 extracts the power guided by the fundamental mode of the
SiP waveguide. Losses are mainly caused by the transition between the PWB and the Si nanowire, indicated by field portions that are
radiated away from the waveguide structure, see inset. (b) Transmission spectrum obtained for the propagation from Port 1 to Port 2.
The insertion loss amounts to approximately 0.3 dB and is flat over the entire frequency range. This is in good agreement with the
measured insertion losses that range from (0.4 ± 0.3) dB to (1.3 ± 0.4) dB.
spectra to confirm that the lasers continue to emit in a single longitudinal mode after wire bonding. In a second step,
we measure and compare the linewidth of the HCSEL without and with PWB. The laser linewidth measurement
relies on superimposing the emitted LD light with light from a highly stable external-cavity reference laser acting as
a local oscillator (LO). The beat signal is then detected using a heterodyne coherent receiver (N4391A, Keysight
Technologies GmbH), see Fig. 4(a) and recorded by an oscilloscope (not shown). We record the intermediate-
frequency signal, evaluate the corresponding variance of the phase increments and retrieve the Lorentzian
linewidth [47]. The LO linewidth amounts to 14 kHz and can safely be neglected compared to the linewidth of the
HCSEL, which is specified between 3 MHz and 5 MHz.
Figure 4(c) shows the power spectrum of the beat signal obtained from another LD (LD5) without (blue) and with
PWB (red). The optical linewidth Δf of this laser can be inferred from the variance of the phase increments for small
time delays and amounts to 2.7 MHz w/o and 3.4 MHz with PWB. This measurement was repeated for the other
three devices (LD6 … LD8), leading to comparable results – measured linewidths changed from (2.9, 5.8, 4.2) MHz
without PWB to (3.5, 3.1, 5.7) MHz with PWB. Note that the linewidth measurements of LD5 … LD8 were obtained
from a previous MCM generation, in which the PWB IL amounted to 4 dB since the erroneous inclination of the
HCSEL deflection mirror was yet not taken into account. This insertion loss is significantly higher than that of the
low-loss MCM depicted in Fig. 3(a) for which we missed to measure the linewidth prior to photonic wire bonding.
However, to confirm that the laser performance is also maintained for low-loss bonds, we also measure the
linewidth of LD1 … LD4 after photonic wire bonding, leading to values of (1.9, 3.4, 2.8, 2.3) MHz. All these values
are in the same range as the linewidths obtained for LD5 … LD8, which were fabricated on the same wafer. Based
on these results, we conclude that the HCSEL performance is not altered by photonic wire bonding.
D. Numerical verification of measurements
To benchmark and support our experimental results, we also performed a simulation of a complete PWB structure
including the transition to the Si nanowire using a commercially available numerical solver (CST Microwave
Studio [48]). We consider a simple case of a PWB having a fully planar trajectory, Fig. 5(a). Note that, in general,
the positioning of the HCSEL chip with respect to the SiP chip is subject to mechanical tolerances that might, e. g.,
lead to a lateral offset. In this case, the axes of the HCSEL emission beam and of the Si nanowire waveguide do not
lie in the same plane, and the PWB trajectory is non-planar.
In the simulation, we use the refractive indices and the device dimensions specified in the previous section along
with a cladding of refractive index noil = 1.3. We consider a frequency range between 190 THz and 196 THz. Waves
are launched and detected at two simulation ports, which are located at the bottom of the PWB input taper
connected to the HCSEL and at the end of Si nanowire. The ports are marked with red lines in Fig. 5(a). The field
launched at Port 1 corresponds to the fundamental mode calculated for the starting cross section of the HCSEL
taper. At Port 2, the power guided by the fundamental mode of the SiP waveguide is extracted to determine the
coupling efficiency. The polarization of the launch field was chosen along the y-direction, leading to the excitation of
a quasi-TE mode in the SiP nanowire waveguide. Note that this simulation does not take into account any scattering
loss caused by surface roughness or any mode mismatch between the mode field at the HCSEL facet and the
launched excitation field. Using the measured 1/e² diameter of 3 µm for the rotationally intensity profile of the
HCSEL emission, we find a loss of only 0.03 dB for the transition to the square PWB input taper.
8 of 9
The plot in Fig. 5(a) shows the normalized intensity distribution obtained from the simulation at a frequency of
193 THz. The main losses originate from the transition of the PWB to the Si nanowire, indicated by radiated field
patterns that propagate away from the Si nanowire, see inset. In contrast to that, the bends of the PWB WG section
do not cause much loss. The simulated IL from Port 1 to Port 2 amounts to approximately 0.3 dB over the entire
frequency range, see Fig. 5(b). The simulated losses are slightly smaller than their measured counterparts, which
range from (0.4 ± 0.3) dB to (1.3 ± 0.4) dB.
5. SUMMARY
We have demonstrated that photonic wire bonding enables highly efficient coupling between InP-based horizontal-
cavity surface-emitting lasers (HCSEL) and silicon photonic chips. We achieve very low coupling losses down to
(0.4 ± 0.3) dB, which are in very good agreement with numerical simulations. To the best of our knowledge, this is
the most efficient interface between an InP light source and a SiP chip that has so far been demonstrated. We
further confirm experimentally that no detrimental effect on the DFB laser linewidth is introduced by the photonic
wire bond. The photonic wire bonding approach enables flexible hybrid integration of best-in-class known-good
devices with high density in a fully automated fabrication process. The technique lends itself to both rapid
prototyping of small batch sizes and to fully automated large-scale production. Combined with previously
demonstrated chip-to-chip [39] and fiber-to-chip [40] connections, we expect photonic wire bonding to evolve in a
universal integration platform for hybrid photonic multi-chip assemblies.
Funding. The work was supported by BMBF joint projects PRIMA (grant 13N14630) and PHOIBOS (grant 13N12574), the H2020
Photonic Packaging Pilot Line PIXAPP (grant 731954), the European Research Council (ERC Starting Grant 'EnTeraPIC' 280145),
the EU-FP7 project BigPIPES, the Alfried Krupp von Bohlen und Halbach Foundation, the Helmholtz International Research School
for Teratronics (HIRST), the Karlsruhe School of Optics & Photonics (KSOP), and Karlsruhe Nano-Micro Facility (KNMF).
Acknowledgment. We acknowledge support by the Open Access Publishing Fund of Karlsruhe Institute of Technology (KIT),
Florian Rupp for the SEM micrographs, Marco Hummel for the mechanical carrier, and Oswald Speck for the electric wire bonds.
9 of 9
|
1805.05403 | 2 | 1805 | 2018-07-03T14:18:40 | A novel EM concentrator with open-concentrator region based on multi-folded transformation optics | [
"physics.app-ph",
"physics.optics"
] | Conventional concentrators with inhomogeneous coating materials that fully enclose the destined region pose great challenges for fabrication. In this paper, we propose to design an EM concentrator with homogeneous materials. Distinguished from conventional ones, the elaborately designed EM concentrator features a concentrator region that is open to the outer-world, which is achieved with multi-folded transformation optics method by compressing and folding the coating materials to create window(s). Based on this concept, we also investigate open-rotator and open rotational-concentrator devices, which could simultaneously rotate and store the EM waves in the central destined region. Due to the open nature of our proposed designs, we believe they will find potential applications in remote controlling with impressive new functionalities. | physics.app-ph | physics | Received: 2 May 2018
Accepted: 13 June 2018
Published: xx xx xxxx
OPEN
A novel EM concentrator with
open-concentrator region based on
multi-folded transformation optics
Hamza Ahmad Madni1,2, Khurram Hussain1, Wei Xiang Jiang1, Shuo Liu1, Asad Aziz1, Shahid
Iqbal1, Athar Mahboob3 & Tie Jun Cui1
Conventional concentrators with inhomogeneous coating materials that fully enclose the destined
region pose great challenges for fabrication. In this paper, we propose to design an EM concentrator
with homogeneous materials. Distinguished from conventional ones, the elaborately designed EM
concentrator features a concentrator region that is open to the outer-world, which is achieved with
multi-folded transformation optics method by compressing and folding the coating materials to create
window(s). Based on this concept, we also investigate open-rotator and open rotational-concentrator
devices, which could simultaneously rotate and store the EM waves in the central destined region. Due
to the open nature of our proposed designs, we believe they will find potential applications in remote
controlling with impressive new functionalities.
The mathematical tool of transformation optics (TO)1–6 has proved that any object can be made invisi-
ble for electromagnetic (EM) waves if the object is fully coated with properly designed inhomogeneous ani-
sotropic1 or isotropic inhomogeneous2 materials. Besides the invisibility cloak7–16, EM field-rotator17–20 and
field-concentrator21–31 have also attained remarkable attention that can rotate the internal propagation direction
of EM waves, and increases the EM energy in the inner region, respectively, while the external field remains
undisturbed. Therefore, the foregoing development in TO-based devices1–31 endows materials with very complex
properties5,8,18 and on the other hand, the demands of manipulating the EM waves in the sense of remote control
is rapidly increasing in modern world that require compact open-devices with good exposure, easy to upgradea-
ble, and importance of inner-outer world32–34.
Start from concentrator21, Fabry Perot resonances were used to design and fabricate the concentrators35 that
achieved progressive attraction in novel energy devices. In addition, some unidirectional and angle-dependent
based concentrators29 have been investigated that work efficiently only in some specific incident wave-angles.
Beyond this, rotators20 are designed for the polarization and have wide applications in the antennas and
wave-guiding that provides a reason of linearly polarized into a circularly polarized nature18. By taking the com-
bine effects of concentrators and rotators, this phenomenon can be made in practice for Smart Grid applica-
tions36, large area wide band imaging37, commonly in laser LED-pumped devices38 and also in micro solar cells
arrays39. Specifically, parabolic concentrators provides practical impacts in radio broadcasting & motion picture
recording40, while polarization based rotators manipulates effects in some add-drop filter systems41. Meanwhile,
the TO-based EM concentrators and rotators yield straight path in the manufacturing and applications of
metamaterials18.
Despite all these facts EM concentrators and rotators possess few discrepancies by representing some known
reasons. As metamaterials provides very low practical outputs due to their low characteristics and narrow band-
widths5,8,18. Previous work concluded that in the designing of the TO based devices, the material parameters
were set in inhomogeneous coordinate transformation which is hard to predict the realization phenomenon and
difficult for fabrication42,43. Beside this, early EM concentrators and rotators show complete invisibility of the des-
tined (concentration and/or rotator) region because of the designing method of these conventional devices that
involves in the construction of enclosures the destined region by coating with properly designed materials17–31.
1State Key Laboratory of Millimeter Waves, Department of Radio Engineering, Southeast University, Nanjing,
210096, China. 2Department of Computer Engineering, Khwaja Fareed University of Engineering & Information
Technology, Rahim Yar Khan, 64200, Pakistan. 3Department of Electrical Engineering, Khwaja Fareed University
of Engineering & Information Technology, Rahim Yar Khan, 64200, Pakistan. Correspondence and requests for
materials should be addressed to H.A.M. (email: [email protected]) or T.J.C. (email: [email protected])
1
SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4www.nature.com/scientificreports However, in such conventional recipes, even a pinhole gap in the coating material will lead a deterioration of
desired functionality. To solve these issues, we have the necessity to design such EM concentrators and rotators,
which provide clear path for the better outcomes18 when the destined region is not fully coated with any EM
medium constructed by homogeneous materials that are easy to fabricate.
Now the question raised that whether we can manipulate the EM waves when the air-gap in the homogenous
coating materials exists. As an answer, in this paper, we propose an easy way to design "open-coating" devices
by utilizing the linear homogenous transformation method and this interesting phenomenon is associated with
concentrators, field rotators, and rotational-concentrator. Our proposed work is not only able to rotate but also
store the EM waves simultaneously and most importantly, the destined region is open to outer world.
Unlike inhomogeneous transformation, here, we divide the imaginary space into several triangular shaped
segments. Our proposed idea is based on the compression of coating materials and then map it to the transformed
medium, which leads to creation of window(s) in front of the destined region and the constitutive parameters of
the proposed device can be obtained by applying multi-folded TO33,34. To understand and validate the proposed
concept, we will consider three different examples. In the first example, we introduce a different scheme to design
concentrator with homogenous coating-materials that contains air-gaps in front of the concentrator region but
the functionality remains unchanged. In this case, the concentrating region will become open to outer-world and
can be easily used for matter exchange without scratching the whole coating-material. Based on this, we extend
our idea to construct open-coating rotators to artificially change the propagating direction of EM waves when the
rotation-region is open to outer world. Finally yet importantly, we proposed an open-coating bi-functional device
that can rotate and store the EM waves simultaneously. The proposed open-coating concept is devoted to remote
controlling applications of EM waves and in the other fields of engineering and in each example; the proposed
open-coated device is placed at a certain distance from the destined region. In the following, full wave finite ele-
ment method is used to demonstrate and validate the expected behavior of our proposed concepts.
Results and Discussions
Conventional concentrators13,21 are based on the radial mapping that store the EM energy in the smaller concen-
trator region. Meanwhile the concentrator device itself is also the example of invisibility as there is no scattering
for the outside observers. Therefore, the expanding technique is applied in ref.21 and the folding method is used in
ref.13 to design complementary media based concentrator. Thus, in previous works, the properly designed coating
material is used to fully enclose the destined region and that coating material is composed by anisotropic and
inhomogeneous materials that are difficult for practical implementations.
Distinguishing from the aforesaid methods, in our approach, we divide the virtual space of N segments into
N2 triangles and then apply two steps to obtain the desired goal. Figure 1 depicts the schematic diagram of the
proposed open-coating concentrator device in which a square shaped virtual space of total four segments are
8. In the first step, the region 1 of ∆ABC
further divided into eight different triangular regions labeled as
and the region 2 of ∆CDB in the virtual space x y z
CDB
′
labeled as ′2 in the physical space
), respectively. Similarly, the other regions in virtual space labeled by
(
8 are accordingly folded into colored regions 2′, 3′… 8′ to design the concentrator. It can be seen in Fig. 1,
2, 3
that the air-gap came to exist between the concentrator-region and concentration-coating. Thus, by taking advan-
tage of this facility, in the second step, a compression and folding method is applied by compressing the bigger
dashed lines into the compressed region I. For example, here, we applied compression technique to regions ′7 , ′8
and the whole segment is compressed into regions as ″7 and ″8 respectively, so that the coating materials turn into
open that do not require to enclose the concentration-region.
) is folded into region ∆ ′
′A B C labeled as ′1 and ∆
x y z
′
1, 2, 3
...
( ,
...
,
,
,
′
′
(
,
1
,
3
1 ) and region II (denoted by x y z
,
)
4 ), and region V (denoted by x y z
,
)
In the second step of Fig. 1, consequently, we compressed the whole black dashed lined space into the region I
(represented by x y z
2 ). Whereas, the material parameters of region II
)
are obtained from the first step of Fig. 1. Thereafter, to overcome the discontinuity occur due to the compression,
the compressed region is further coated with the folded regions such as region III (denoted by x y z
3 ), region
)
IV (denoted by x y z
5 ). For detail, taking the second quadrant as an
)
PQA is folded into ∆PQR which can be seen in the purple color of zoom-in view.
example of region III, the ∆
′QtA C into the light
Similarly, region IV and region V in the first quadrant are obtained by folding the space of
green colored region of QtRS and ∆PtC1
. Due to the symmetric structure,
other parts of these complementary regions in third and fourth quadrants can be achieved by rotating all of the
corresponding tensors by π. The materials used in folded regions are known as complementary materials that are
also used in lens designs to compensate the space44. xn, yn, and zn indicate the coordinate system of each region,
where =n
In the following, full wave simulations of finite element method (COMSOL) are performed in 2D by adopting
the transverse electric (TE) mode at the frequency of GHz
. The geometric structure parameter for the both steps
a
0 21, c = 0.25 and the coordinates of
in SI units are as follows: the total width of the squares = .
B (0 125, 0 125),
A ( 0 125, 0 125),
A( 0 085, 0 085),
.
′ − .
′
P( 0 21, 0), − .
t( 0 105, 0 084),and − .
S( 0 105, 0 0735).The
.
obtained material parameters for each region can be seen in method summary.
0 17, = .
D(0 105, 0 105),
.
.
Q( 0 125, 0 1), − .
, − .
into the golden colored region of ∆PtS1
R( 0 125, 0 0875), − .
.
P( 0 09, 0)
1
B(0 085, 0 085), − .
C( 0 105, 0 105).
In Fig. 2, we consider to examine the working performance of proposed open-coating concentrator with the
proposed close-coating concentrator under the illumination of plane waves with the two different incident angles
of 0° and 90°, respectively. The electric field (Ez) distributions in all cases (Fig. 2(a–c)) indicate that the electric
field's amplitude in the concentrator region is equivalent to that in free space near the outer boundary of concen-
trator. In Fig. 2a, the working functionality of the designed closed-coating concentrator is similar to that in all the
1, 2, 3, 4 and 5.
− .
− .
b
3
,
4
,
2
,
5
,
1
,
3
,
4
.
.
.
.
.
.
(
5
(
2
(
′
.
(
.
2
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 Figure 1. Scheme illustration of the proposed open concentration-coating device that concentrates EM waves
while the concentrating region is not fully enclosed with the concentration-coating. The mapping of
concentration-coating is defined as the dashed lines' regions in virtual space labeled by
transformed into a proposed concentrator with eight different regions (here labeled ′
words, the ∆ABC is folded into ∆ ′
a smaller square region of radius a. It should be clearly noticed that the air gap exists between the coated
material and concentrator region. In the second step, a multi-folded transformation method is applied to create
a window. As an example, after applying the second transformation function, the regions ′7 and ′8 turns into ″7
and ″8 , respectively, that is compressed and shifted towards the compressed region and that region is further
coated with the folding regions.
′A B C and so on. Meanwhile, the bigger square of radius c is compressed into
....
8 ). In easy
′
1, 2, 3
′....
′
1 , 2 , 3
8 are
previous reported concentrators21. The beauty of proposed concentrator is that the −z direction space is
y plane only. Therefore, the total energy flow is
unchanged and concentrator region compresses the space in −x
towards the concentration region and that region is fully covered with the properly designed coated materials, as
shown in Fig. 2a. Another case is observed in Fig. 2b, to design a concentrator with open-coating materials and
by comparing the field patterns of Fig. 2a and Fig. 2b, it is evident that the concentration region can be made open
for the outside observers while the overall performance remains unchanged. Furthermore, to prove the proposed
device have no limitations of incident waves' direction, the incident waves have applied at 90° angle, as seen in
Fig. 2c. The field patterns of Fig. 2c is identical to Fig. 2b and this result is the bonus point to validate the proposed
concept of this paper. Furthermore, point source (Fig. 2d) and linear source (Fig. 2g) are placed in free space and
further applied to demonstrate the effectiveness of the proposed concentrator in Fig. 2(d–f) and Fig. 2(g–i),
respectively. The functionality of proposed devices as shown in Fig. 2(e–f) and Fig. 2(h,i) represents the concen-
tration of the EM wave's energy in the central destined region. It can be clearly observed that the concentrator
region is open to outer-world in all cases but functionality remains unchanged. Whereas, the location of point
( 0 3, 0 3) with the current density of A1 and on the other side, the height of linear source is 0.2 m at the
source is − .
origin point − .
( 0 4, 0) with the electric field intensity of V1 .
A tradeoff is that the material parameters of the above example involves negative values of anisotropic and
homogeneous materials that have two windows in-front of the concentrator region and these windows can be
increased as per demand for remote controlling phenomenon. Since the designed concentrator can only store EM
energy that raise a question that whether we are able to rotate and store EM energy simultaneously, for example,
make a device that can be used for solar cells and for antenna applications as well. As an answer, we further extend
the concept of open-coating concentrator into the open-coating rotator first and then combine these two devices
to obtain the combinational effect.
The schematic diagram of the open-coating rotator is given in Fig. 3 with two different steps. At the first step,
A B C of label ′1 in physical space
8 are accordingly transformed into
the ∆ABC of region 1 in virtual space x y z
x y z
(
′
) and similarly, other regions in virtual space labeled by
) is transformed into region ∆ ′
2, 3
...
( ,
′
′
.
,
′
,
′
,
3
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 Figure 2. Simulation results of open-coating concentrator under different incident waves. (a) Electric field (Ez)
distributions for the proposed concentrator under the illuminations of plane waves with incident angles of 0°
and same incident waves are applied in (b) for the open-coating concentrator. (c) The incident angle of
propagating waves are changed to 90° for (b). In all cases, the functionality of proposed concentrator is
equivalent to previously designed TO-based concentrators21. (d) The EM waves of point source is concentrated
in (e) closed concentrator and (f) open-concentrator. Similarly, emission of linear source in (g) is concentrated
in (h) closed concentrator and (i) open-concentrator. In all cases, the results give the immediate validation of
the proposed open-concentrator device.
′ ...
2 , 3
8 to design the conventional rotator20 made by homogenous materials. Apart from this
colored regions ′
′
particular step of transformation, the second step of this device is to achieve the isotropic materials of the
open-coating device. In this regard, based on the specific choice13 of =c
, the bigger dashed line circle of
radius c (the origin point is represented by black dot O) is compressed into region I of radius a and that com-
pressed region also contain some mapped regions. As a reference, we compressed the segments of ′6 and ′7 into
the compressed region II and they became as ″6 and ″7 , respectively. Therefore, the folding materials to cover the
compressed region is obtained by folding the region −c
b
) into −b
b a/2
a
).
(
(
Hereafter, the proposed designs are analyzed by using COMSOL Multiphysics in transverse electric (TE)
. The geometric structure parameters for the both steps in SI units are as follows:
, and the coordinates of
O( 0 14, 0),
. The obtained material parameters for each region can be seen in
mode at the frequency of GHz
the total width of the outer and inner squares are
A
( 0 175, 0 175), = ′ = .
(0 175, 0 175), − .
a
m
b
, = .
method summary.
. m0 25
C (0 125, 0 125), − .
′
B
0 0625
, and = .
A
= ′ = − .
m0 04
= .
C( 0 125, 0 125),
The simulation results are shown in Fig. 4(a–c). Figure 4a shows the rotator's field pattern in which the rota-
tion region is fully encapsulated with the homogenous coating materials. Anyway, the functionality of our pro-
posed rotator is equivalent to previously reported work20. Meanwhile, the Fig. 4b is the case of open-coating
.
m0 05
. m0 35
.
.
.
3
B
c
.
,
4
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 Figure 3. Scheme illustration of the proposed open rotation-coating device that rotates the EM field while the
rotation region is not fully enclosed with the rotation-coating. A linear homogeneous coordinate
transformation is used to design the rotator by constructing several triangle regions in virtual space. At the first
step, the mapping of rotation-coating is defined as the dashed lines' regions in virtual space labeled by
1, 2, 3
′
In easy words, the ∆ABC is mapped into ∆ ′
designed from this method are homogeneous and easy to fabricate. In the second step, bigger dashed lines circle
is compressed into region I and region II (contained mapped medium of ′6 and ′7 regions), and then isotropic
folded material is used to coat the compressed region.
′.... ′
A B C and so on. The material parameters of the proposed rotator
8 are transformed into a conventional rotator with eight different regions (here labeled ′
1 , 2 , 3
8 ).
....
′
′
rotator in which the rotation region is open to the outer world. It can be clearly seen from Fig. 4(a,b) that the
incident plane waves are impinging the devices with 0° angle and the devices rotate the EM waves with 90° angle
in the destined region. In order to validate the bi-directional property of our proposed design, the incident wave
is propagating along the 90° angle and as a result, the open-coating rotator's functionality is efficiently achieved,
which can be seen in Fig. 4c.
Compared with the open-coating concentrator as discussed previously in Fig. 1, in this case, the EM waves are
rotated but not concentrated. By taking advantage from these two proposed designs, here, we plan to merge the
first step of Fig. 1 into the first step of Fig. 3, which can be clarified in Fig. 5. In simple words, the outer coating
layers are composed by rotators as the rotation region is free space, so in that free space the concentrator device is
merged. The red lines are used to differentiate the concentrator and rotator boundaries. This bi-functional device
is proposed to have the rotation and concentration of EM waves at the same time and we assume that device as
a virtual space for the next step to make it like open-coating device. The procedure to make it open, again the
second step of Fig. 3 is recalled. The geometric structure, constitutive parameters and working frequency of this
particular device are same as described earlier for Figs 2, 4. For more details see method summary.
After that, simulations are performed and the results are shown in Fig. 6(a–c). Figure 6a validates the working
performance of proposed bi-functional device that is rotating and concentrating the EM field at the same time
in the destined region. In this case, the destined region is fully covered with the combination of proposed rotator
(Fig. 3) and proposed concentrator (Fig. 1). Figure 6b is a case of open-coating device in which the destined
region is open to the outer world. It can be observed in Fig. 6(a,b) that the incident plane waves are impinging
the devices with 0° angle and the device rotate and concentrate the EM waves with 90° angle in the destined
region. In Fig. 6c, the incident wave propagates along the 90° angle and the efficiency of proposed device remains
unchanged.
It can be observed that there exist some negligible distortions in EM waves due to the resonance of the inci-
dent wave between the positive index material and negative index metamaterials to compensate each other.
5
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 Figure 4. Simulation results of open-coating rotators. (a) Electric field (Ez) distributions for the proposed
rotator under the illuminations of plane waves with incident angles of 0° and same incident waves are applied in
(b) for the open-coating concentrator. (c) The incident angle of propagating waves are changed to 90° for (b). In
all cases, the functionality of proposed rotator is equivalent to previously designed TO-based rotators20.
Figure 5. Scheme illustration of the proposed open-coating bi-functional device that first rotates and then
concentrates the EM waves at the same time while the concentrator region is not fully enclosed with the coating
materials. In the first step, the proposed bi-functional device is achieved by combining the first step of both
Fig. 1 and Fig. 3, distinguished by the red lines boundaries. For second step, Fig. 3 (second step) has been
recalled to make the proposed device open.
Theoretically, there should be no distortions according to the wave theory because of the existence of both folding
space and resonance. However, for simulation case, it requires finest mesh to vanish the wave's distortions.
Conclusion
Based on multi-folded TO, we proposed devices of homogeneous and anisotropic materials to manipulate the EM
waves and further created air-gap between the coated materials in order to make destined region open to outer
6
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 Figure 6. Simulation results of open-coating bi-functional device. (a) Electric field (Ez) distributions for the
proposed bi-functional device under the illuminations of plane waves with incident angles of 0° and same
incident waves are applied in (b) for the open-coating bi-functional device. (c) The incident angle of
propagating waves are changed to 90° for (b).
Figure 7. Representation of the linear homogeneous coordinate transformation in the virtual and real space.
(a) The assumed triangular region of free space ε µ(
transformed into (b) another triangular shape.
0 0 in the Cartesian coordinate system is geometrically
)
world. Meanwhile, the structure of open-coating devices are composed by mapped EM medium merged inside
the compressed region that is further coated with negative index materials. The simulation results validated each
case of different proposed structures and verified that the dielectric region can store and rotate the EM waves
without any physical connectivity of coating materials. These features of proposed concept will be very helpful for
future remote controlling applications in microwave and optical engineering.
Methods
Before moving to the next steps, we consider the mapping shown in Fig. 7, which uniquely maps every point x y
)
of Fig. 7a in a new region
) can be
x y
,
(
′
′
related to its original space x y
( ,
) as shown in Fig. 7b. The transformation material of the new space
(
) by:
x y
′
( ,
,
′
ε
′ =
µ
′ =
T
J
J
ε
0
J
det
J
J
µ
0
J
det
T
(1)
where J is the Jacobean tensor and for triangular shape only in 2D transformation, the Jacobean matrix can be
found as:
x
y
x
′ − ′
3
1
y
′ − ′
3
1
x
y
x
′ − ′
2
1
y
′ − ′
2
1
J
=
1
x
y
1
x
2
y
2
0
0
1
x
3
y
3
−
−
0
1
x
y
1
−
−
0
1
−
0
0
1
0
(2)
In accordance to achieve the ultimate goal, in this paper, we focused on to achieve homogenous materials by
using triangular shape regions. Thus, Eqs (1,2) will play an important role in designing the proposed devices.
We start from the first step of open-coating concentrator as mentioned in Fig. 1. Here, the bigger square of size
m0 25
with the following transformation equations:
is compressed into = .
m0 17
= .
a
0
c
7
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 x
′ =
′ =
y
z
x
y
a
c
a
c
z
Thus, the constitutive parameters of that region will become as: ε
(3)
2 . In the
same step, when ∆ABC of region 1 and ∆CDB of region 2 is folded into region ∆ ′
CDB
′
named as ′2 respectively, the coordinates parameters of each point is known so this phenomenon is similar as
mentioned in Fig. 7. So, by taking help from Eqs (1–2), the constitutive parameters will become as:
diag[1, 1, (0 25/0 17) ]
′A B C named as ′1 and ∆
′ = ′ =
a
′ =
µ
a
.
.
0
0
1
−
The same procedure is applied to find the material parameters of all other regions such as:
5 6212
− .
1 68
.
0
2
5
− −
1
2
− −
0
0
1 68
.
0 68
− .
0
0
0
0 68
− .
ε
′ = ′ =
2
′ = ′ =
ε
1
, and
µ
µ
1
2
′
′
′
′
.
′ = ′ =
ε
3
µ
3
′
′
ε
′ = ′ =
5
µ
5
′
′
′ = ′ =
ε
7
µ
7
′
′
0 68
− .
1 68
− .
0
1 68
− .
5 6212
− .
0
0
0
0 68
− .
,
5 6212
− .
1 68
.
0
1 68
.
0 68
− .
0
0
0
0 68
− .
,
0 68
− .
1 68
− .
0
1 68
− .
5 6212
− .
0
0
0
0 68
− .
,
ε
′ = ′ =
4
µ
4
′
′
ε
′ = ′ =
6
µ
6
′
′
ε
′ = ′ =
8
µ
8
′
′
1
−
2
0
2
5
−
0
0
0
1
−
,
2
5
− −
1
2
− −
0
0
0
0
1
−
,
1
−
2
0
2
5
−
0
0
0
1
−
.
In the second step (Fig. 1), the bigger black dashed lines are compressed with the following transformation
equations:
where κ is the compression ratio of .0 7. So, the material properties of region I and region II will become as:
x
′ =
y
′ =
z
′ =
x
κ
z
y
(4)
′ = ′ =
ε
I
µ
I
diag[1/0 7, 0 7, 1/0 7],
.
.
.
ε
″
II
7
″
= ″
µ
II
7
″
− .
0 68/0 7
.
1 68
− .
0
− .
1 68
− .
5 6212
0
0 7
× .
0
0
− .
0 68/0 7
.
,
ε
″
II
8
″
µ
= ″
II
8
″
1/0 7
− .
2
0
0 7
− × .
5
2
0
0
0
1/0 7
− .
.
After that folding transformation is applied and for each region as described in Fig. 1 (second step), the mate-
rial parameters will become as:
′ = ′
ε
µ
III
III
2
−
3 5294
− .
0
3 5294
6 7284
− .
− .
0
0
0
2
−
,
ε
µ
′ = ′
IV
IV
2 4
2
−
.
3 38
2 4
. − .
0
0
0
0
2
−
, and
ε
µ
′ = ′
V
V
2
−
16 80
.
0
.
16 80
141 62
− .
0
0
0
2
−
.
where as, the subscripts of ε′ and µ′ indicate the material parameters of that concerned region only. It can be
clearly observed from the constitutive parameters that the inhomogeneity of designed concentrator is completely
removed, and only anisotropy is required to obtain the desired functionality.
For open-coating rotator (Fig. 3), in the first step, the conventional rotator is designed by simply mapping the
8 in virtual space
8 . Thus, by seeking help from Eqs (1,2), the constitutive
A B C of label ′1 and similarly, other regions labeled as
region 1 as ∆ABC into region ∆ ′
are accordingly transformed into colored regions ′
parameters of the device will become as:
2 , 3
′... ′
2, 3
...
′
′
′ = ′
ε
µ
1
1
′
′
,
26
5
−
0
26
5
−
0
5 0
−
0
1
0
1
5 0
−
0
1
0
1
,
,
1 7 0
7 50 0
0 0 1
1 7 0
7 50 0
0 0 1
,
′ = ′ =
ε
5
µ
5
′
′
ε
′ = ′ =
6
µ
6
′
′
ε
′ = ′ =
7
µ
7
′
′
ε
′ = ′ =
8
µ
8
′
′
ε
′ = ′ =
2
µ
2
′
′
ε
′ = ′ =
3
µ
3
′
′
ε
′ = ′ =
4
µ
4
′
′
,
1 5
0
5 26 0
0 0 1
1 5
0
5 26 0
0 0 1
,
50
7
−
0
50
7
−
0
,
7 0
−
0
1
0
1
7 0
−
0
1
0
1
.
In the second step (Fig. 3), the bigger dashed lines circle is compressed into smaller blue colored circle with
the following transformation equations:
8
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 x
′ =
y
z
′ =
′ =
+ −
x
0
x a
c
0
x
y
a
c
a
c
z
So, the material properties of region I and region II will become as:
′ = ′ =
ε
I
µ
I
diag
[1, 1, ( / ) ],
c a
2
ε
″
II
7
″
µ
= ″
II
7
″
1 5
5 26
0 0 (
0
0
c
2
)
a
,
ε
″
II
6
″
µ
= ″
II
6
(5)
.
″
0
0
c
2
)
a
1 7
7 50
0 0 (
′ = ′ =
III
µ
III
r
− − b r
( / ) ]4 while =
Similarly the, material parameters of region III are obtained from ref.13 that are: ε
1,
For open-coating bi-functional device (Fig. 5), the constitutive parameters for both first and second step will
remain similar as that of the obtained material parameters of Figs 1 and 3 except the folding regions. It should be
noticed that the second step for this proposed design is achieved by recalling the Fig. 3 (second step), with the
same geometric and material parameters except the value of = − .
2 and = − .
diag[
0 14
.
0 15
2
)0
1,
−
(
x
+
−
x
0
x
,
y
.
x
0
093901 (2009).
24–46 (2010).
References
1. Pendry, J. B. et al. Controlling electromagnetic fields. Science 312, 1780–1782 (2006).
2. Leonhardt, U. Optical conformal mapping. Science 312, 1777–1780 (2006).
3. Liu, Y. et al. Recent advances in transformation optics. Nanoscale 4, 5277–5292 (2012).
4. Yan, M. et al. Cylindrical superlens by a coordinate transformation. Phys. Rev. B 78, 125113 (2008).
5. Kwon, D.-H. et al. Transformation electromagnetics: An overview of the theory and applications. IEEE Antenn. Propag. M. 52(1),
6. Chen, H. et al. Transformation optics and metamaterials. Nat. Mater. 9, 387–396 (2010).
7. Yang, Y. et al. Full‐Polarization 3D Metasurface Cloak with Preserved Amplitude and Phase. Adv. Mater. 28(32), 6866–6871 (2016).
8. Schurig, D. et al. Metamaterial electromagnetic cloak at microwave frequencies. Science 314, 977–980 (2006).
9. Li, J. et al. Hiding under the carpet: a new strategy for cloaking. Phys. Rev. Lett. 101, 203901 (2008).
10. Leonhardt, U. et al. Broadband Invisibility by Non-Euclidean Cloaking. Science 323, 110–112 (2009).
11. Zhang, B. et al. Macroscopic Invisibility Cloak for Visible Light. Phys. Rev. Lett. 106, 033901 (2011).
12. Shi, X. et al. Electromagnetic Detection of a Perfect Carpet Cloak. Sci. Rep. 5, 10401 (2015).
13. Lai, Y. et al. Complementary media invisibility cloak that cloaks objects at a distance outside the cloaking shell. Phys. Rev. Lett. 102,
18731–18738 (2008).
14. Yang, Y. et al. Full-polarization 3D metasurface cloak with preserved amplitude and phase. Adv. Mater. 201600625 (2016).
15. Lan, C. et al. Electrostatic Field Invisibility Cloak. Sci. Rep. 5, 16416 (2015).
16. Yang, Y. et al. A metasurface carpet cloak for electromagnetic, acoustic and water waves. Sci. Rep. 6, 20219 (2016).
17. Luo, Y. et al. Design and analytical full-wave validation of the invisibility cloaks, concentrators, and field rotators created with a
general class of transformations. Phys. Rev. B 77(12), 125127 (2008).
18. Chen, H. et al. Design and experimental realization of a broadband transformation media field rotator at microwave frequencies.
Phys. Rev. Lett. 102(18), 183903 (2009).
19. Chen, H. et al. Transformation media that rotate electromagnetic fields. Appl. Phys. Lett. 90(24), 241105 (2007).
20. Kwon, D.-H. et al. Polarization splitter and polarization rotator designs based on transformation optics. Opt. Express 16(23),
21. Rahm, M. et al. Design of electromagnetic cloaks and concentrators using form-invariant coordinate transformations of Maxwell's
equations. Photon. Nanostruct. Fundam. Appl. 6, 87 (2008).
22. Yaghjian, A. D. et al. Alternative derivation of electromagnetic cloaks and concentrators. New J. Phys. 10, 115022 (2008).
23. Navau, C. et al. Magnetic energy harvesting and concentration at a distance by transformation optics. Phys. Rev. Lett. 109, 263903
24. Bian, B. R. et al. Cylindrical optimized nonmagnetic concentrator with minimized scattering. Opt. Express 21, A231–A240 (2013).
25. Sadeghi, M. M. et al. Perfect field concentrator using zero index metamaterials and perfect electric conductors. Front. Phys. 9, 90–93
26. Yu, G. X. et al. Non-rotationally invariant invisibility cloaks and concentrators of EM waves. Eur. Phys. J. Appl. Phys. 44, 181–185
(2012).
(2014).
(2008).
27. Lin, L. et al. A cone-shaped concentrator with varying performances of concentrating. Opt. Express 16, 6809–6814 (2008).
28. Yang, J. J. et al. Metamaterial electromagnetic concentrators with arbitrary geometries. Opt. Express 17, 19656–19661 (2009).
29. Li, W. et al. Homogeneous-materials-constructed electromagnetic field concentrators with adjustable concentrating ratio. J. Phys. D:
Appl. Phys. 44, 125401 (2011).
18, 119–130 (2011).
28, 1573–1577 (2011).
6, 36846 (2016).
Sci. Rep. 7, 13171 (2017).
30. Li, T. H. et al. Three dimensional electromagnetic concentrators with homogeneous material parameters. Prog. Electromagn. Res. M.
31. Zhang, K. et al. Cylindrical electromagnetic concentrator with only axial constitutive parameter spatially variant. J. Opt. Soc. Am. B
32. Zheng, B. et al. Concealing arbitrary objects remotely with multi-folded transformation optics. Light-Sci. Appl. 5, e16177 (2016).
33. Madni, H. A. et al. Non-contact radio frequency shielding and wave guiding by multi-folded transformation optics method. Sci. Rep.
34. Madni, H. A. et al. Non-contact method to freely control the radiation patterns of antenna with multi-folded transformation optics.
35. Sadeghi, M. M. et al. Optical Concentrators with Simple Layered Designs. Sci. Rep. 5, 11015 (2015).
36. Schroeder, J. et al. Adaptive Data Collection Mechanisms for Smart Monitoring of Distribution Grids. arXiv 1608, 06510v1 (2016).
37. Hunter, R. et al. Large area W-band quasi-optical Faraday rotators for imaging applications. in Infrared and Millimeter Waves.13th
38. Pichon, P. et al. Alexandrite laser LED-pumped via Ce-doped luminescent concentrators. The European Conference on Lasers and
International Conference on Terahertz Electronics (2005).
Electro-Optics. Optical Society of America (2017).
9
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 39. Jutteau, S. et al. Micro Solar Concentrators: Design and fabrication for microcells arrays. Photovoltaic Specialist Conference (PVSC)
40. Coile, R. C. The parabolic sound concentrators. J. Acoust. Soc. Am. 11, 167 (2005).
41. Chen, G. et al. Polarization rotators in add-drop filter systems with double-ring resonators. IEEE Photon. Technol. Lett. 26(10),
(2015).
976–979 (2014).
(2015).
42. Jiang, W. X. et al. Design of arbitrarily shaped concentrators based on conformally optical transformation of nonuniform rational
B-spline surfaces. Appl. Phys. Lett. 92(26), 264101 (2008).
43. Sun, F. et al. Homogenous optic-null medium performs as optical surface transformation. Prog. Electromagnetics Res. 151, 169–173
44. Pendry, J. B. et al. Near-field lenses in two dimensions. J. Phys.: Condens Matter 14, 8463–8479 (2002).
Acknowledgements
This work was supported in part from the National Science Foundation of China under Grant Nos. 61631007,
61571117, 61501112, 61501117, 61522106, 61722106, 61701107, and 61701108, and 111 Project under Grant
No.111-2-05. H.A. Madni acknowledges the support of the Higher Education Commission's Start-Up Research
Grant Program, Pakistan under Grant No. 21-1742/SRGP/R&D/HEC/2017, and the support of the Postdoctoral
Science Foundation of China at Southeast University, Nanjing, China, under Postdoctoral number 201557.
Author Contributions
H.A. Madni designed the devices and carried out the simulations. K. Hussain, S. Liu, W.X. Jiang, A. Aziz, S.
Iqbal, A. Mehboob and T.J. Cui analyzed the data and interpreted the results. H.A. Madni, S. Liu, W.X. Jiang, A.
Mehboob and T.J. Cui draft the manuscript with the input from the others. T.J. Cui supervised the project.
Additional Information
Competing Interests: The authors declare 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
10
www.nature.com/scientificreports/SciENtific REPORTs (2018) 8:9641 DOI:10.1038/s41598-018-28050-4 |
1811.08729 | 1 | 1811 | 2018-11-21T13:43:51 | Network of thermoelectric nanogenerators for low power energy harvesting | [
"physics.app-ph"
] | We report the design, elaboration and measurements of an innovative planar thermoelectric (TE) devices made of a large array of small mechanically suspended nanogenerators (nanoTEG). The miniaturized TE generators based on SiN membranes are arranged in series and/or in parallel depending on the expected final resistance adapted to the one of the load. The microstructuration allows, at the same time, a high thermal insulation of the membrane from the silicon frame and high thermal coupling to its environment (surrounding air, radiations). We show a ratio of 60% between the measured effective temperature of the membrane, (and hence of the TE junctions), and the available temperature of the heat source (air). The thermal gradient generated across the TE junction reaches a value as high as 60 kelvin per mm. Energy harvesting with this planar TE module is demonstrated through the collected voltage on the TE junctions when a temperature gradient is applied, showing a harvested power on the order of 0.3 $\mu$Watt for a 1 cm 2 chip for an effective temperature gradient of 10 K. The optimization of nanoTEGs performances will increase the power harvested significantly and permit to send a signal by a regular communication protocol and feed basic functions like temperature measurement or airflow sensing. | physics.app-ph | physics | Network of thermoelectric nanogenerators for low power energy
harvesting
Dimitri Tainoff,1, 2 Anaıs Proudhom,1, 2 C´eline Tur,1, 2 Thierry Crozes,1, 2 S´ebastien
Dufresnes,1, 2 Sylvain Dumont,1, 2 Daniel Bourgault,1, 2 and Olivier Bourgeois1, 2
1Institut N ´EEL, CNRS, 25 avenue des Martyrs, F-38042 Grenoble, France
2Univ. Grenoble Alpes, Inst NEEL, F-38042 Grenoble, France
(Dated: November 22, 2018)
Abstract
We report the design, elaboration and measurements of an innovative planar thermoelectric (TE)
devices made of a large array of small mechanically suspended nanogenerators (nanoTEG). The
miniaturized TE generators based on SiN membranes are arranged in series and/or in parallel
depending on the expected final resistance adapted to the one of the load. The microstructuration
allows, at the same time, a high thermal insulation of the membrane from the silicon frame and
high thermal coupling to its environment (surrounding air, radiations). We show a ratio of 60%
between the measured effective temperature of the membrane, (and hence of the TE junctions),
and the available temperature of the heat source (air). The thermal gradient generated across the
TE junction reaches a value as high as 60 kelvin per mm. Energy harvesting with this planar TE
module is demonstrated through the collected voltage on the TE junctions when a temperature
gradient is applied, showing a harvested power on the order of 0.3 µWatt for a 1 cm2 chip for an
effective temperature gradient of 10 K. The optimization of nanoTEGs performances will increase
the power harvested significantly and permit to send a signal by a regular communication protocol
and feed basic functions like temperature measurement or airflow sensing.
8
1
0
2
v
o
N
1
2
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
9
2
7
8
0
.
1
1
8
1
:
v
i
X
r
a
1
Introduction
Wireless sensor network (WSN) are an association of various system being able to measure
and transfer information over a network without requiring human intervention. Focusing on
the sensor part, the major problem preventing the massive spread of WSN is their power
supply. Indeed depending on their location, sensors cannot be wired to the grid and/or the
maintenance costs induced by the battery change are not acceptable. The smartest solution
to get over this problem is to harvest energy from the environment of sensors in order to
make them autonomous.
Since the environment of sensors are changing drastically depending on their final use,
energy harvesting solutions that have been developed until now are devoted to one or very
few use cases. Each of these technologies has their own advantages and drawbacks which
have been reviewed recently [1, 2]. However, it is difficult to find a general solution which
could be sufficiently versatile to cover lots of WSN use cases [3].
In that context, ther-
moelectricity has a great potential because thermal gradients are present all around even
intermittently, being most of the time a byproduct of other energies coming from mechanical
motion, electromagnetic dissipation, or chemical reaction... Despite this advantage, the use
of thermoelectric generator (TEG), developed since more than fifty years, is still limited
to niche markets [4, 5]. This is due to numerous reasons among which one finds moderate
efficiency, the use of rare materials and, above all, the bulk nature of thermoelectric (TE)
module.
With the emergence of low power communication protocols dedicated to the WSN, there
is now a real demand for micro-sources of energy that can deliver power on the order of
100 µWatt. That range of power can be obtained from standard TE module with a tem-
perature gradient of few degrees under stationary operating conditions. However, these
conditions of use are not necessarily adapted to small wireless sensors especially when the
energy source is intermittent and/or the available volume too limited. This has given a new
direction for research that is clearly oriented towards miniaturization of TE modules [6, 7].
One of the first miniaturized TEG was based on the structuration of micro-pillars of TE
materials, it allowed to strongly limit the volume of the modules [8]. However, the need to
2
evacuate the heat flow for preserving the thermal gradient has led to the use of large heat
sink making the whole device quite voluminous. The use of such a massive heat sink is
the direct consequence of the 3D architecture of the miniaturized TEG and particularly the
fact that the thermal coupling of the face in contact with ambient air is not optimized for
convection.
Consequently, a growing interest has emerged regarding new designs of nanoTEG using
planar and suspended configurations, essentially thanks to pioneering works using microfab-
rication technologies [9 -- 15]. The major advantage of these suspended planar configurations
is that the free standing part is well coupled to the surrounding air by convection permit-
ting to the device to work without heat sink. However, at least one major issue remains
to be solved before using that kind of module as low power generator. Indeed, using TE
modules based on micropatterned membrane leads to designs involving very thin films of
TE materials and finally to a sizable internal electrical resistance. Since standard low power
electronic is optimized for low electrical impedance energy source, the actual solutions ex-
hibiting internal resistance of more than 10 kΩ are not well adapted. To summarize, the
major challenges for a planar microscaled TE module to be an efficient power source is to
create a significant thermal gradient upon small distance, to have low electrical impedance
and to collect thermal energy at the micro and nanoscale through a thermoelectric voltage.
Here we report the design, the elaboration and the measurement of innovative planar
thermoelectric devices made of a large array of suspended nano-thermoelectric generators
(single cell nanoTEG). These nanoTEGs are small enough to be duplicated per hundreds
upon centimeter square surface and arranged in series and/or in parallel depending on the
expected final resistance adapted to the load. The resistance of the measured devices is
on the order of 103 kΩ in the studied configuration, opening the door to massive parallel
wiring of nanoTEG which could lead to resistance of the order of few tens of ohms. We
demonstrates a harvested power on the order of 0.3 µWatt for a 1 cm2 chip for an effective
temperature gradient of 10 K. Under optimized performances, they can produce enough
power (below 100 µWatt) to send a signal using common communication protocols and feed
basic functions, for instance temperature sensing or air flow measurements.
3
Design and realization.
The concept at the basis of this work is twofold: first, to take advantage of the reduced
thermal coupling between the suspended thermoelectric junction on the membrane and the
heat bath (silicon frame) to favor heat exchange with the surroundings (air or radiation)
and second to decrease the electrical resistance of the final device in order to use it as
generator.
It is made possible thanks to the elaboration of small suspended nanoTEG
acting like individual nanosource of energy obtained by micro and nanostructuration (see
Fig. 1). These nanoTEGs can then be duplicated in series and/or parallel depending on the
expected internal resistance; the microstructuration favoring at the same time the thermal
insulation of the central part of the membrane and heat exchange through air or radiation.
To optimize the efficiency of the device, we have used bismuth telluride alloys as thermo-
electric materials for their high conversion efficiency close to room temperature [17, 18]. As
their mechanical properties do not allow to make self- suspended structure, silicon nitride
(widely used in MEMS industry) is used as a physical support taking profit of its mechanical
stability and its low thermal conductivity. The use of large membrane and planar config-
uration boost surface dependent heat transfers like convection or radiation. The sensitive
thermoelectric junctions are installed on the membrane and on the silicon frame to convert
temperature gradients in TE voltage.
As compared to a regular macroscopic TE modules that are quite massive, this technical
solution allows collecting energy from various small sources. In a regular TEG, the thermal
conductance is given by the n and p TE pillars themselves. In a macroscopic module, the
thermal conductance between the hot side and the cold side is given by the TE legs; it is of
10−4 W.K−1 for typical leg dimensions of 1 mm×1 mm×5 mm. Here, since the nanTEGs
are planar by construction, the thermal coupling of the sensitive part is only ensured by the
TE thin films and the SiN suspending arms. Indeed, first, the aspect ratio of the suspending
arms (small thickness, width and long length) is highly favorable for thermal insulation.
Second, by using low thermal conductivity materials like amorphous silicon nitride and
bismuth telluride, the thermal link to the heat bath is severely reduced as compared to the
use of polysilicon [11, 12, 14]. This leads to thermal conductance of the order of 10−7 W.K−1,
4
FIG. 1. Scanning Electron Microscope (SEM) image and functioning scheme of a nanoTEG cell
composed of a single membrane (SiN membrane suspended by four SiN arms). Each SiN arm
supports a n or p-type Bi2XTe3 thin films: the p-type (orange) is Bi2−xSbxTe3 and the n-type
(blue) is Bi2−xTe3Sex. One cold (hot) thermoelectric junction is located on the membrane and the
other hot (cold) junction is on the bulk silicon frame. The temperature of the membrane is free to
vary under an external source of heat (circulating air or radiation). Under appropriate conditions,
the thermal gradient between the membrane and the silicon frame can reach 10 K over a distance
of 150 µm. The thermoelectric voltage V generated by the temperature gradient between the
membrane and the silicon frame is collected by the two external contacts.
few orders of magnitude smaller than their macroscopic counterparts. In other words, this
very low thermal conductance and the very loss mass of the membrane allows to create
a thermal gradient of one degree when the membrane exchange a tenth of picojoule with
its environment. These characteristics and the very small mass of the thin silicon nitride
membrane lead to significantly enhanced thermal gradients when air convection is present
along with a very fast thermal time response.
Let us first described the fabrication of an individual single nanoTEG cell. It is built
by using two kinds of Bi2XTe3 thin TE films 300 nm thick (X being either antimony (Sb)
for the p-type Bi2−xSbxTe3 or selenium (Se) for the n-type is Bi2−xTe3Sex). The TE thin
films are deposited by reactive sputtering, structured by regular clean room processes and
5
FIG. 2. Sketch of the main steps of the elaboration process of the nanoTEG devices : a) Si substrate
with SiN top layer, b) structuration of the n-type Bi2−xTe3Sex thin film by lift-off process, c) idem
for the p-type Bi2−xSbxTe3 thin film, d) and the Ni metallic contact, e) etching of the SiN top
layer using SF6 RIE etching, the electrical contact being protected by photoresist f) Suspension of
the nanoTEGs with XeF2 dry etching process.
connected on-chip using a contact made of a metallic nickel thin film. As it can be seen in
Fig. 1, the membrane (100 nm thick) and the suspending arms (150 µm long, 6 µm wide)
are made of SiN. The suspending arms are the mechanical support on which the TE thin
films are deposited (150 µm long, 5 µm wide). After several microfabrication steps dedicated
to the structuration of the TE junctions, membrane and supporting arms, via are opened
into silicon nitride using SF6 RIE etching. Membrane and arms are finally suspended using
XeF2 dry etching process. The main microfabrication steps are described in Fig. 2. Then,
by construction, if the silicon chip is installed on a hot surface, a temperature gradient will
appear between the silicon frame and the suspended membrane, the temperature of the
latter remaining close to the on of air thanks to convection. A thermoelectric voltage will
then be generated across the TE junctions and collected on the two external contacts.
The final nanoTEG module is then made out of thousands of identical single cells [16].
They are assembled in such a way (series and parallel) that the total electrical impedance
can be significantly reduced if needed. Actually, this arrangement is very flexible and ver-
6
FIG. 3. SEM picture of a network of hundred of nanoTEG cells. The nanoTEGs are assembled in
series and parallel for adapting the internal electrical resistance to the impedance of the electrical
load. Ns is the number of nanoTEG in series and Np the number of nanoTEG in parallel. In inset,
a close-up view of three membranes of the network.
satile, it can be adjusted to fit the targeted electrical resistance of the load. As pointed
out by numerous works, the internal electrical resistance of the TE module is a point at
least as important as thermoelectric performances. Reducing the internal resistance helps
maximizing the power generated by the nanoTEG and permits its use with standard low
power electronic components [12]. An example of a large array of such nanoTEG cells is
shown in Fig. 3. Since the fabrication of these TE modules is made using microelectronic
technologies, it can be up-scaled easily to make large surface of planar nanoTEGs.
Measure, characterization and performance.
More than ten different nanoTEG networks were fabricated and tested going from single
cell TEG as shown in Fig. 1 to large array of membrane as shown in Fig. 3. Several different
physical properties are of great importance for a deep characterization of our nanoTEGs:
7
Materials
Bi2−xSbxTe3
Bi2−xTe3Sex
Bi2−xSbxTe3 in [17]
Bi2−xTe3Sex in [17]
e (nm) Trec (◦C) ρ (mΩ.cm) S (µV.K−1) S2σ (Watt.m−1.K−2)
300
300
300
300
275
175
275
250
5
2
3
1
137
-54
250
-200
3.89 × 10−4
1.43 × 10−4
2.1 × 10−3
4.0 × 10−3
TABLE I. Deposition parameters and thermoelectric properties of the two n and p-type bismuth
telluride thin films. e is the thickness of the TE materials, Trec is the temperature at which the
TE materials are annealed, ρ the resistivity of the TE film, S the Seebeck coefficient measured
at room temperature, and S2σ the power factor. The first two lines are related to parameters as
measured on TE thin films in this work. As a comparison, the last two bottom lines of the table
show the optimized parameters as obtained in previous studies made by Bourgault et al. [17]. The
Seebeck of the TE thin films used in this work are smaller than the accepted state-of-the-art value,
showing the room of improvement that still exists for the nanoTEGs.
the thermal conductance between the suspended structure and the heat bath, the resistance
of the electrical contact at the junction between the p and n types TE materials, the See-
beck coefficient of the bismuth telluride and finally, the overall performance of the TEG to
generate an electrical power harvested from its thermal environment.
First, the thermal properties of each constituting materials of the TEG have been mea-
sured independently. The thermal conductivity of SiN and Bi2XTe3 based TE materials has
been measured using on membrane 3ω measurements [19, 20]. At room temperature, SiN
exhibits a thermal conductivity below 3 W.m−1.K−1 and a value around 1 W.m−1.K−1 has
been found for thin films of Bi2Te3 based materials [19 -- 21]. This gives a thermal conduc-
tance of 10−7 W.K−1 for the four suspended beams showing that the membrane is highly
thermally isolated from the frame. The electrical resistivity of each thermoelectric legs (ρp
and ρn) has been measured at room temperature at 5 mΩ.cm for p-type and 2 mΩ.cm for
n-type using standard four probe experiment carried out on dedicated samples. Finally, the
Seebeck coefficients have been measured using a home-made experimental setup for each
8
composition of the Bi2XTe3 films on separated samples having the same geometry and con-
tacts [17]. The TE thin films are not yet fully optimized because the Seebeck coefficient
and the electrical conductivity that are summarized in Table I are below the state-of-the-art
value (see Table I) [17, 18]. The internal resistance of the nanoTEG assembly Rsim can be
estimated from the relation:
Rsim = Ns(ρpL + ρnL)/(Npwe)
(1)
where L is the length of the n and p-type arms, w their width and e their thickness; Np
and Ns being respectively the number of nanoTEGs in parallel and series. The expected re-
sistance Rsim for the studied array composed of 35 membranes in parallel and 12 membranes
in series is 880 Ω (see Table II). The differences observed between the simulated values from
Eq. 1 and the measurements of the actual resistance of the four samples lies within 10%.
Intrinsic variation of resistance between samples can originate from inhomogeneity in the
layer geometry, broken membrane or could be due to contact resistance. Regarding the con-
tribution of the contact to the overall resistance, we can estimate that the contact resistance
between metallic contact and Bi2XTe3 thin film to be below 10−6 Ω.cm2, a value in fair
agreement with the state-of-the-art [22]. This estimated value leads to contact resistances
of several Ohms for a contact surface of few micrometer square; a resistance that will be
negligible as compared to the resistance of the TE films themselves. After XeF2 etching,
the resistance of the whole device is increased by a factor of two. This increase has been
observed for all the samples. If the true origin of that change is still under investigation, it
may be assigned to the slight deterioration of the electrical properties of the TE thin films
or the electrical contact to the nickel thin film.
The performance of the nanoTEG has been measured on single membranes, as shown
in Fig. 4, as well as on a large network of membranes. A temperature gradient is imposed
by the mean of a cold air flow blown on the chip carefully glued on a hot plate regulated
initially at a temperature of 320 K. The cold junction located on the membrane is cooled
down by the air flow while the silicon substrate coupled to the hot plate will serve as the hot
heat sink. The temperature difference between the membrane and the Si frame is probed
by measuring the voltage appearing across the TE junctions. When the air flow speed is
9
Samples S1
S2
S3
S4
Rbf (Ω) 1050 970 1030 910
Raf (Ω) 1980 2160 2590 2110
TABLE II. Electrical characteristics of samples containing an array of 420 membranes over 0.5 cm2.
The resistance measurements have been performed before (bf ) and after (af ) the XeF2 etching of
Si for releasing the membranes. The XeF2 etching fabrication step is influencing the value of the
resistance of the Bi2XTe3 film or its contact resistance with the Ni.
FIG. 4. Power generated by a single nanoTEG cell as a function of the bias current when the
membrane is used in Peltier mode. Temperatures are determined from the thermoelectric voltage.
The temperature differences between the membrane and the silicon frame used in the experiment
are mentioned. In inset, the membrane on which the measurement has been done is shown.
increased the voltage measured at the end of the device increases. This shows that the
nanoTEG device is very sensitive to forced convection. Fig. 5 shows the variation of the
ratio between the effective temperature gradient as measured in the experiment and the
available temperature gradient (the gradient between the temperature of the hot plate and
10
FIG. 5. Ratio between the actual effective temperature gradient as measured on the membranes
and the temperature gradient available for different air-flow speeds. Results show that up to 60%
of the available temperature gradient is converted into a useful thermal gradient leading to the
appearance of a 10 K temperature difference between the membrane and the silicon frame.
the temperature of the air flow). It is worth noticing that this ratio reaches 60 percent for
airflow speed of 25 m.s−1.
These conditions of air flowing are frequently found in moving parts that necessitate on-
board sensor application. The power generated by the nanoTEGs has been measured for
different temperature gradients. Results are displayed in Fig. 6; it shows the expected linear
variation of the generated thermoelectric voltage as function of the effective temperature
gradient. The power output generated by an assembly of 420 membranes covering 0.5 cm2
surface subjected to a temperature gradient of 10 K has been measured to be of 0.15 µWatt,
giving for 1 cm2, a power of 0.3 µWatt.
This result is significant since it is obtained with not-fully optimized thermoelectric ma-
terials as shown in Table II. Indeed, by considering the standard TE properties of bismuth-
telluride alloys a power output of more than 10 times higher can be generated, meaning
3 µWatt for a 10 K temperature gradient over a 1 cm2 surface. Improvement in the techno-
11
FIG. 6. Variation of the thermoelectric voltage as a function of the effective thermal gradient for
three different samples. As expected, the variation is linear.
logical process will permit increasing the membrane density by a factor of, at least, three.
By working with larger surface (10 cm2) along with larger temperature gradient, the power
generated will be sufficient to feed a low energy communication protocol and exchange in-
formation on temperature or airflow speed.
As a concluding remarks, it can be mentioned that such miniaturized planar suspended
thermoelectric device could also be used for cooling applications, and not only for energy
harvesting. For example, small systems located at the center of the suspended membranes
can be cooled down by current injection in the nanoTEGs device, using such device in Peltier
mode.
Conclusion
Miniaturiazed planar suspended thermoelectric devices have been developed for low power
energy harvesting. This device is designed to convert small intermittent gradient of tem-
perature into electrical power by using thermally isolated membranes as heating or cooling
platform. The geometry of the device permits to adapt the resistance of the thermogen-
12
erator for each particular use case. The efficiency of the nanoTEG is sufficient to expect
its application in real wireless sensor network. By construction, the architecture of the de-
vice allows the appearance of high thermal gradients over very small distance (60 K.mm−1).
Using optimized TE materials, more than few tens of microWatt can be generated hence
permitting a wireless communication between the sensor and the network.
Regular microelectronic fabrication processes are used to elaborate such nanoTEG mod-
ules, meaning that the production can be easily upscaled at relatively low cost along with
a high integrability. Finally, it has to be stressed that better performances and evolution
of the technology are foreseen. Denser array of membranes, optimized new geometry and
higher Seebeck coefficients can be easily obtained boosting the performances of the nan-
oTEGs towards higher generated power. This shows that membrane based thermoelectric
energy harvesters are very promising building blocks for applications where hot or cold air
is present offering excellent perspectives for autonomous and connected physical sensing.
Acknowledgments
We thank the micro and nanofabrication facilities of Institut N´eel CNRS: the Pole Cap-
teurs Thermom´etriques et Calorim´etrie (E. Andr´e, G. Moiroux and J.-L. Garden) and
Nanofab (L. Abbassi, B. Fernandez, T. Fournier, G. Juli´e and J.-F. Motte) for their ad-
vices in the preparation of the samples. The authors acknowledge the financial support
from EU MERGING Grant No. 309150, the MODULO project Premat CNRS program in
2015-2016, and the SATT Linksium financial support for maturation of the Moız project in
2017 (https://moiz-eh.com/).
[1] A. Harb, Renewable Energy 36, (2011) 2461.
[2] R.J.M Vullers, R. van Schaijk, I. Doms, C. Van Hoof, R. Mertens, Solid-State Electronics 53,
(2009) 684.
[3] F. K. Shaikh and S. Zeadally, Renew. Sustain. Energy Rev. 55, (2016) 1041.
13
[4] Sub-Watt Thermoelectric Energy Harvesting Industry and Market Report,
Infinergia,
http://www.infinergia.com/en/energy-harvesting (2016).
[5] S. LeBlanc, S.K. Yee, M.L. Scullin, C. Dames, and K.E. Goodson, Renew. Sustain. Energy
Rev. 32, (2014) 313.
[6] B. Kyun, J. Seung, W. Han, and J.-Y. Kim, Microelectronic Engineering 88, (2011) 775.
[7] L.W. da Silva and M. Kaviany, International Journal of Heat and Mass Transfer 47, (2004)
2417.
[8] H. Bottner, J. Nurnus, A. Gavrikov, G. Kuhner, M. Jagle, C. Kunzel, D. Eberhard, G.
Plescher, A. Schubert, K.-H. Schlereth, Journal of Microelectromechanical Systems 13, (2004)
414-420.
[9] T. Huesgen, P. Woias, and N. Kockmann, Sensors and actuators A 145-146, (2008) 423.
[10] D. Davila, A. Tarancon, C. Calaza, M. Salleras, M. Fernandez-Regulez, A. San Paulo, and L.
Fonseca, Nano Energy 1 (2012) 812-819.
[11] A.P. Perez-Marin, A.F. Lopeandia, L. Abad, P. Ferrando-Villaba, G. Garcia, A.M. Lopez,
F.X. Munoz-Pascual, and J. Rodriguez-Viejo, Nano Energy 18 (2014) 73-80.
[12] Z. Yuan, K. Ziouche, Z. Bougrioua, P. Lejeune, T. Lasri and D. Leclercq, Sensors and Actua-
tors A 130221 (2015) 67-76.
[13] Y. Nakamura, M. Isogawa, T. Ueda, S. Yamasaka, H. Matsui, J. Kikkawa, S. Ikeuchi, T.
Oyake, T. Hori, J. Shiomi, A. Sakai, Nano Energy 12 (2015) 845-851.
[14] M. Sakata, T. Hori, T. Oyake, J. Maire, M. Nomura, J. Shiomi, Nano Energy 13 (2015)
601-608.
[15] G. Pennelli, J. Nanotechnol. 5, (2014) 1268.
[16] O. Bourgeois, D. Tainoff, D. Bourgault, Thermoelectric device, patent number WO
2017140975A1.
[17] D. Bourgault, C.G. Garampon, N. Caillault, L. Carbone, and J.A. Aymami, Thin Solid Films
516 (2008) 8579-8583.
[18] D. Bourgault, C.G. Garampon, N. Caillault, and L. Carbone, Sensors and Actuators A 273
(2018) 84-89.
14
[19] A. Sikora, H. Ftouni, J. Richard, C. H´ebert, D. Eon, F. Omn`es, and O. Bourgeois, Rev. Sci.
Instrum. 83 (2012) 054902.
[20] A. Sikora, H. Ftouni, J. Richard, C. H´ebert, D. Eon, F. Omn`es, and O. Bourgeois, Rev. Sci.
Instrum. 84 (2013) 029901.
[21] H. Ftouni, C. Blanc, D. Tainoff, A.D. Fefferman, K.J. Lulla, J. Richard, E. Collin, and O.
Bourgeois, Phys. Rev. B 92 (2015) 125439.
[22] R.P. Gupta, R. McCarty, and J. Sharp, Journal of Electronic Materials 43 (2014) 1608 -- 1612.
15
|
1709.08812 | 1 | 1709 | 2017-08-14T22:10:15 | Giant THz surface plasmon polariton induced by high-index dielectric metasurface | [
"physics.app-ph",
"cond-mat.mes-hall"
] | We use computational approaches to explore the role of a high-refractive-index dielectric TiO2 grating with deep subwavelength thickness on InSb as a tunable coupler for THz surface plasmons. We find a series of resonances as the grating couples a normally-incident THz wave to standing surface plasmon waves on both thin and thick InSb layers. In a marked contrast with previously-explored metallic gratings, we observe the emergence of a much stronger additional resonance. The mechanism of this giant plasmonic resonance is well interpreted by the dispersion of surface plasmon excited in the air\TiO2\InSb trilayer system. We demonstrate that both the frequency and the intensity of the giant resonance can be tuned by varying dielectric grating parameters, providing more flexible tunability than metallic gratings. The phase and amplitude of the normally-incident THz wave are spatially modulated by the dielectric grating to optimize the surface plasmon excitation. The giant surface plasmon resonance gives rise to strong enhancement of the electric field above the grating structure, which can be useful in sensing and spectroscopy applications. | physics.app-ph | physics | Giant THz surface plasmon polariton induced by high-index
dielectric metasurface
Shuai Lin1, Khagendra Bhattarai2, Jiangfeng Zhou2, Diyar Talbayev1,*
1Department of Physics and Engineering Physics, Tulane University, New Orleans, Louisiana
70118, USA
2Department of Physics, The University of South Florida, Tampa, Florida 33620-7100, USA
*corresponding author: [email protected]
Abstract
We use computational approaches to explore the role of a high-refractive-index dielectric TiO2
grating with deep subwavelength thickness on InSb as a tunable coupler for THz surface
plasmons. We find a series of resonances as the grating couples a normally-incident THz wave
to standing surface plasmon waves on both thin and thick InSb layers. In a marked contrast with
previously-explored metallic gratings, we observe the emergence of a much stronger additional
resonance. The mechanism of this giant plasmonic resonance is well interpreted by the
dispersion of surface plasmon excited in the air\TiO2\InSb trilayer system. We demonstrate that
both the frequency and the intensity of the giant resonance can be tuned by varying dielectric
grating parameters, providing more flexible tunability than metallic gratings. The phase and
amplitude of the normally-incident THz wave are spatially modulated by the dielectric grating to
optimize the surface plasmon excitation. The giant surface plasmon resonance gives rise to
strong enhancement of the electric field above the grating structure, which can be useful in
sensing and spectroscopy applications.
Introduction
Terahertz (THz) frequency range of electromagnetic spectrum has attracted significant
fundamental research interest and technological development due to various potential
applications, such as spectroscopy1,2, bio sensing3,4, high speed communication5, and
subwavelength imaging6,7. The extraordinary properties of surface plasmon polaritons (SPPs),
such as strong electromagnetic field confinement, make them an important topic in THz science
and technology8,9. THz frequency SPPs can exist on two main classes of surfaces: one is a
surface of a semiconductor with a low electron or hole density that brings the bulk plasma
frequency down to THz range10,11; the other is a microscopically corrugated or perforated metal
surface on which spoof THz SPPs can occur12. Semiconductor indium antimonide (InSb) is a
perfect candidate for the study of THz SPPs13 because of its high electron mobility and intrinsic
electron density at room temperature of 1016 cm-3, which corresponds to about 1.8 THz bulk
plasma frequency. The bulk electron density and plasma frequency are easily tunable by
temperature14,15. In recent years, THz SPPs on InSb have been the focus of research that
addressed both fundamental16 and applied plasmonics questions, including promising sensing
1
and spectroscopy uses17,18. The propagation and properties of spoof THz SPPs are determined
entirely by the microscopic structure of the supporting metallic surface and cannot be easily
tuned. By contrast, THz SPPs on InSb offer tunability by both temperature and very moderate
magnetic field19–21.
Investigations and utility of THz SPPs rely on a means to excite and detect them via coupling to
a free-space electromagnetic wave. Various coupling schemes have been proposed, such as
prism coupling22, grating coupling23, or a knife edge24, among others25. Spoof SPPs on
periodically structured surfaces can get excited directly by an incident free space wave, as the
periodicity of the structure acts to select the SPP wave vector26–30. In this work, we explore the
role of a dielectric grating as a tunable coupler between a normally-incident free-space wave and
THz SPPs on thin and thick InSb layers. In a previous study, we established micrometer-thin
InSb layers with metallic gratings as a platform for THz plasmonic devices; the metal grating
couples the free-space wave to SPPs17. Most metals, including gold, silver, and copper, can be
well approximated by a perfect conductor in THz range. Metallic grating offers little tunability
in terms of the coupling between SPPs and free-space waves. A dielectric grating allows spatial
modulation of both the phase and amplitude of the incident THz wave to manipulate the
excitation of SPPs. By varying the thickness, period, and refractive index of the grating, we are
able to control both the strength and frequency of the SPP resonance in THz transmission and
reflection of our structure. We compute the transmission and reflection spectra of InSb layers
with a dielectric grating and find that in addition to the SPP resonances present in metallic
grating/InSb structures, dielectric grating/InSb structures exhibit another much stronger
resonance in both transmission and reflection. This new resonance is also a SPP mode that is
sustained even on thick InSb layers where the other SPP modes are no longer observable. The
dispersion of the strong new resonance is well interpreted by the SPP dispersion on air/TiO2/InSb
trilayer structure. The dielectric grating structure exhibits strong surface electric field
enhancement typical of SPPs.
Figure 1 shows the geometric configuration of our THz plasmonic structure, which consists of a
grating on InSb layer. The grating has a period d=60 m with equal alternating 30-m-wide
grating strips and gaps that are periodic in the x direction. The x-polarized THz wave is incident
vertically from the top of the structure. The grating period d sets the wavevector 𝛽0 = 2𝜋/𝑑 of
the excited THz SPPs. The thickness and period of the dielectric grating are variable parameters
that allow the tuning of the coupling to THz SPPs. While the THz refractive index of the
dielectric material is also a tuning parameter, we choose titanium dioxide (TiO2) as the high-
refractive-index grating material and keep the refractive index fixed at n=9.531. We vary grating
thickness in the 0.2-2.4 m range and we study the transmission and reflection spectra of this
structure on 2 and 5 m thin InSb layers and a 500m thick InSb wafer. The geometric
parameters of this structure allow fabrication and processing using established photolithography
and microfabrication methods.
2
Figure 1. Schematic of a thin InSb slab with dielectric grating. The incident THz wave is in negative
z direction with x polarization. The structure stretches indefinitely in the x and y direction. The thickness
of the grating and InSb substrate varies for different simulations.
Results
Figure 2(a,b) shows the transmission spectrum of our plasmonic structure with InSb layer
thicknesses of 2 and 5 m. The thickness of the dielectric grating is 1 m. The grey line is the
transmission spectrum of this structure with no electrons in InSb or the bulk plasma frequency
𝜔𝑝 = 0. The solid color lines correspond to the transmission spectra with different bulk plasma
frequencies, which are labeled on the spectra. The dash color lines are the transmission spectra
of the same InSb layer with a metallic grating of the same period and 200 nm thickness17. The
transmission spectra display an overall band-pass shape. The high-pass part of transmission
results from absorption and reflection of low-frequency light by conduction electrons described
by the Drude model. The low-pass part of transmission results from the Fabry-Perot effect in the
micrometer-thin InSb slab, and the diminished high-frequency transmission is due to the onset of
the first Fabry-Perot minimum. Thus, the band-pass transmission peak is broader for the thinner
InSb slab, in agreement with the Fabry-Perot description17.
Superimposed on the band-pass shape of the transmission spectra, we find two clear high-
frequency sharp transmission dips in both 5 m and 2 m InSb which are almost the same
frequency and strength as the metallic grating spectra (dash lines), Fig. 2(a,b). These two
resonance modes correspond to the odd and even SPP modes on air/InSb/air trilayers and we
have studied them in detail in a previous article17. In addition to the high-frequency SPP
resonances on thin InSb layers, a much stronger low-frequency absorption resonance is observed
in the dielectric grating transmission spectra that is not present in the spectra of the metallic
grating on InSb. This giant absorption resonance shows strong dependence on the bulk plasma
frequency, which is indicative of its SPP origin. (We use the term giant to highlight the strength
of this new SPP resonance and contrast it with the much weaker high-frequency resonances that
are excited in the structure with both metallic and dielectric gratings.) In the following, we
investigate the detailed properties of the giant low-frequency SPP resonant absorption.
3
Figure 2. Transmission and reflectance of dielectric grating/InSb structures. (a) Transmission of 5
µm-thick InSb layers with different bulk plasma frequency (solid lines). The grey solid line is the
transmission of the structure with plasma frequency 𝜔𝑝 = 0. (b) Transmission of 2 µm-thick InSb layers
with different bulk plasma frequency (solid lines). The grey solid line is the transmission of the structure
with 𝜔𝑝 = 0. Dashed lines in graph (a,b) show the transmission of structures with metallic gratings. (c)
TiO2 grating on InSb layers of different thickness. The bulk plasma frequency of InSb is 𝜔𝑝 = 1.44 THz.
(d) TiO2 grating on 500 µm-thick InSb layer with different bulk plasma frequency. The thickness of the
dielectric grating is 1 µm and the period is d=60 µm in all graphs.
The weak high-frequency SPP resonances vanish very quickly as the InSb slab thickness
increases, Fig. 2(a,b). The transmission of the InSb structure vanishes as well due to Drude
absorption and reflection by conduction electrons, so we need to compute reflectance in order to
explore the behavior of the giant SPP resonance with InSb thickness. As Figure 2(c) shows, the
giant SPP resonance is clear and strong in the reflection spectra of the dielectric grating on thin
and thick InSb layers. The frequency of the giant resonance increases as the InSb layer thickness
grows from 2 m to 20 m and then stays stable as the InSb thickness grows from 20 m to 500
m. The weak high-frequency SPP modes disappear when InSb thickness becomes larger than
about 10 m. This is because the two high-frequency modes result from mixing between SPP
resonances on the upper and lower air/InSb interfaces, and this mixed mode on a thin InSb layer
can sustain much larger wave vector k than a single propagating mode on the air/InSb interface.
When InSb layer becomes sufficiently thick, the SPP modes on the upper and lower InSb
surfaces become decoupled and cease to exist at high wave vectors. Effectively, the SPP loss
4
becomes much higher for thick InSb layers and the high-frequency modes cannot get excited. In
this context, the thickness of about 10 m separates thin and thick InSb layers. For thick InSb
layers, the SPP modes on the top and bottom surfaces of InSb can be considered as decoupled
non-interacting waves. Figure 2(d) shows how the giant SPP resonance observed in reflectance
depends on the bulk plasma frequency of InSb.
The existence of the giant SPP resonance on thick InSb layers (there is no limit on the possible
InSb thickness) is one of the main findings in our work, in contrast to the vanishing of the weak
high-frequency SPP resonances on InSb layers thicker than 10 m. The robustness of the giant
SPP resonance means that plasmonic devices can be fabricated on thick InSb wafers, which
allows us to avoid the difficulty of manufacturing suspended grating/InSb structures with very
thin InSb layers required for metallic gratings.
Figure 3. Computational (open circles) vs. theoretical (solid lines) dispersion of the giant SPP
resonance. The bulk plasma frequency of InSb is 𝜔𝑝 = 1.44 THz. The InSb layer thickness is 500 m.
The vertical dash line shows the wave vector for d = 60 µm grating.
We now demonstrate that the behavior of the giant SPP resonance in the reflection spectra agrees
well with SPP theory of air/TiO2/InSb trilayer structure in which air and InSb have infinite
thicknesses. We calculate the theoretical dispersion relation of this trilayer from32
𝑒−4𝑘1𝑎 =
𝑘1/𝜖1+𝑘2/𝜖2
𝑘1/𝜖1−𝑘2/𝜖2
𝑘1/𝜖1+𝑘3/𝜖3
𝑘1/𝜖1−𝑘3/𝜖3
(1)
and
𝑘𝑖
2 = 𝛽2 − 𝑘0
2𝜖𝑖, (2)
where i=1,2,3, 𝑘0 = 𝜔/𝑐 is the wavevector of THz wave in vacuum, a is the thickness of the
TiO2 layer, 𝛽 is the SPP wavevector along the air/TiO2/InSb interface, 𝜖1, 𝜖2 and 𝜖3 are the
dielectric permittivities of TiO2, InSb and air, respectively. The quantities k1, k2, and k3 are the
5
imaginary wave vectors for TiO2, InSb, and air in the direction perpendicular to the interface.
The complex permittivity of InSb is described by the Drude model:
𝜖1(𝜔) = 𝜖∞(1 −
2
𝜔𝑝
𝜔2+𝑖𝜔𝛾
), (3)
where 𝜖∞ is the background dielectric constant, which is 15.6 for InSb, 𝜔𝑝 is the bulk plasma
frequency, 𝛾 = 0.3 × 1012 THz is the scattering rate for InSb at low temperature33.
Figure 4. Spatial distribution of electric field at the frequency of the giant SPP resonance. (a)
Distribution of the z component of the electric field in the standing SPP wave in a 60x60 m square around
a grating strip representing one grating period. The TiO2 strips are 1 m thick and cover 15-m lengths on
each side of the panel for a total strip width of 30 m. (b) Distribution of the x component of the total
electric field. InSb bulk plasma frequency in both graphs is 𝜔𝑝 = 1.44 THz. InSb thickness is 500 m.
The color scale is the same for both graphs and uses arbitrary units.
Figure 3 compares the theoretical trilayer SPP dispersion (solid lines) calculated from equations
(1)-(2) with the dispersion (open circles) extracted from the simulated reflection spectra of the
grating/InSb structure. The bulk plasma frequency is 1.44 THz in Figure 3. To obtain the
dispersion relation from our numerical reflection simulations, we need to identify the wavevector
of the standing SPP mode induced by the grating. By inspecting z component of the electric field
at the resonance frequency, we find that the giant SPP resonance in the reflection spectra, Fig. 2
(d), corresponds to the standing wave mode with wave vector 𝛽0 = 2𝜋/𝑑, where d is the grating
period. We extract the SPP dispersion from the simulated reflection spectra by varying the
grating period 𝑑. This dispersion matches very well the theoretical dispersion calculated from
the air/TiO2/InSb trilayer structure, Eqs. (1)-(2). The trilayer theory confirms our assignment of
the giant absorption resonance as a SPP mode. It also demonstrates the tunability of the
absorption resonance frequency via the dielectric grating period.
Why does the theory of an SPP resonance on the continuous air/TiO2/InSb trilayer describe so
well the dispersion of an SPP resonance found on the TiO2 grating/InSb structure? The answer is
provided by Fig. 4(a) that shows the distribution of the z component of the electric field at the
6
SPP resonance frequency. The field Ez is the hallmark of the SPP resonance, e.g., this field is
zero at off-resonance frequencies. The electric field Ez is concentrated above and below the TiO2
strips of the grating, as shown in Fig. 4(a). While the grating is necessary to excite the giant SPP
resonance, the presence of the air gaps does not appreciably perturb the dynamics of the standing
SPP wave, which is why it closely resembles the dynamics of the propagating SPP wave on a
continuous air/TiO2/InSb structure. We tentatively ascribe the weaker resonances in the
reflection spectra found below the frequency of the giant SPP resonance, Fig. 2 (d), to higher
order SPP modes with wavevectors that are integer multiples of 𝛽0. As shown by the shape of
the dispersion curve in Fig. 3, the derivative d/dk is negative for high wave vectors;
qualitatively, higher order modes of larger k are expected to have a lower frequency than the
fundamental SPP mode. In the discussion below, we focus only on the fundamental mode.
Discussion
A major difference between the structure with the dielectric grating and a similar structure with a
metallic (gold) grating is the presence and unique properties of the giant SPP resonance. Figure
5(a) shows that the resonance strength increases quickly as the dielectric grating thickness
increases from 0.2 m to 1.4 m; the resonance strength remains unchanged until the grating
thickness reaches about 2.2 m. The grating couples the incident free-space wave to the
standing SPP modes by modulating the spatial distribution of the THz electric field along the
InSb surface. To achieve better coupling, the modulated electric field should mimic as close as
possible the electric field of the standing SPP wave. The gold grating modulates the THz field
by imposing perfect-metal boundary conditions periodically along the InSb surface (gold is well
approximated by the perfect metal at THz frequencies)17. An extended dielectric layer atop InSb
allows a control of amplitude and phase of the THz field underneath the InSb surface. The high
refractive index of TiO2 (n=9.5) is the main reason why such a thin (~ 1 m) dielectric layer
efficiently modulates both the amplitude and phase of THz field. In the simplest optical model,
the THz amplitude and phase underneath the TiO2 layer inside InSb can be computed from
Fresnel reflection coefficients and the dielectric layer thickness (the optical path of the THz
wave) and compared to the amplitude and phase of the wave incident on bare InSb. Figure 5(b)
shows how both amplitude and phase of THz field Ex inside InSb at the SPP resonance frequency
change with the dielectric layer thickness; the THz phase and amplitude are compared to the
phase and amplitude of the same wave in the absence of any dielectric, which corresponds to the
gaps of the dielectric grating. Figure 4(b) shows the spatial distribution of electric field Ex in the
standing SPP wave. The Ex oscillation is characterized by a phase difference of between the
gaps and strips of the dielectric grating; the oscillation is completely out of phase between these
two regions. Therefore, we can expect strongest grating coupling efficiency when the phase
difference between gaps and strips for the incident Ex inside InSb is close to , the maximum
phase contrast. This phase contrast approaches 0.83 at around 2.5 m grating thickness and the
amplitude contrast reaches it maximum at 1.5 m grating thickness, Fig. 5(b). The combined
thickness dependence of the amplitude and phase contrast at low grating thicknesses largely
explains the fast rise of the coupling efficiency from 0.2 m to 2 m, Fig. 5(a).
7
Figure 5. Dependence of optical properties on the dielectric grating thickness. (a) Reflectance of
dielectric grating/InSb structures with different grating thickness. InSb thickness is 500 µm. (b) Intensity
and phase contrast of electric field just underneath InSb surface between the grating strips and air gaps.
The frequency is chosen at the resonance frequency of 1 µm grating, 1.1 THz. (c) Reflectance of
continuous TiO2 layers on InSb. A broad anti-reflection feature develops with increasing thickness. InSb
bulk plasma frequency in all graphs is 𝜔𝑝 = 1.44 THz.
In addition to the intensity changes, the resonance frequency also becomes lower as the thickness
of grating increases. Such frequency shift can be well explained by the dispersion of
air/TiO2/InSb trilayer system as shown in Fig. 3, where the SPP resonance frequency decreases
as the thickness of TiO2 layer increases. Another remarkable feature is that the thickness of TiO2
grating is much smaller than the wavelength of incident wave. For instance, the resonance
wavelength for 1.4 m grating is 𝜆 = 300 m, resulting in the wavelength-to-thickness ratio
𝜆𝑇𝑖𝑂2/𝑡 ≈ 23, where 𝜆𝑇𝑖𝑂2 = 𝜆/9.5 is the wavelength inside the TiO2 medium. Our grating is
extremely thin in contrast to dielectric gratings in other applications34, where the wavelength-to-
thickness ratio, 𝜆𝑔𝑟𝑎𝑡𝑖𝑛𝑔/𝑡~1, in order to create the necessary phase retardance for a wave
propagating through the grating layer. The ultra-thin grating greatly relieves the challenges for
film deposition and the patterning process in the fabrication.
The remarkable feature of the giant SPP resonance is its strength relative to the strength of the
weaker high-frequency SPP modes that are also present in the metallic grating sensor, Fig.
2(a,b). This strength is derived partly from the very efficient coupling between the free space
and the SPP waves, as described in the previous paragraph. Another contribution to the SPP
8
resonance strength is illustrated in Fig. 5(c), where we plot the reflectance of continuous TiO2
layers of different thickness on InSb. Below the plasma reflectance edge, we observe a broad
feature with a thickness-dependent center frequency. This feature results from the anti-
reflection-coating effect of the continuous TiO2 film. The energy that is not reflected is absorbed
inside InSb by conduction electrons. While this broad anti-reflection feature is not evident in the
reflection spectra of the grating/InSb structure, Fig. 5(a), it likely plays a role in enhancing the
giant SPP resonance by allowing a larger fraction of the incident energy to couple and be
absorbed as the SPP oscillation.
The excitation of SPPs is typically accompanied by an enhancement of the total electric field
intensity at the resonance frequency. In our structure, the amplitude of Ex can be taken as an
indicator of the excited SPP strength and the surface electric field enhancement. In the simulated
reflection response with 500 m InSb, we determine the amplitude of the x component of the
electric field above the dielectric grating strips at the SPP resonance frequency. We compare the
Ex amplitude for dielectric gratings of different thickness on 500 m InSb and for a continuous
dielectric layer with the same thickness on 500 m InSb. Figure 6 shows the Ex amplitude ratio
between the dielectric grating and the continuous dielectric layer. The surface electric field
enhancement due to the excitation of SPPs increases quickly as the grating thickness grows from
0.2 m to about 1 m; the enhancement ratio decreases slowly above the 1.5 m thickness, Fig.
6. Thus, this trend agrees well with the phase and intensity modulation in Fig. 5(b). The
dielectric grating achieves around 4 times electric field enhancement compared with the bare
dielectric layer without grating structure. We expect that higher enhancement ratios will be
possible via the optimization of other grating parameters, such as refractive index, period, and
gap/strip width ratio. Surface-plasmon enhanced electric field generally leads to stronger light-
matter interaction, which can be useful for spectroscopy and sensing applications35–37.
Figure 6. Electric field enhancement due to the giant SPP resonance. Enhancement is plotted as the
ratio of the x component Ex between the TiO2 grating and an equal-thickness continuous TiO2 layer on
500 µm InSb. The grating period is d=60 µm. The bulk plasma frequency is 𝜔𝑝 = 1.44 THz.
9
To summarize, we investigated the properties of a tunable dielectric grating coupler of a free-
space THz wave to standing SPP modes on thin and thick InSb layers. We found a giant SPP
resonance in transmission and reflection spectra of this plasmonic structure and successfully
described its dispersion. The frequency and strength of the giant resonance can be manipulated
by varying the geometric parameters of the grating, such as period and thickness. The refractive
index of the grating material offers another tuning variable. The dielectric grating/InSb structure
achieves about 4 times surface field enhancement over the continuous dielectric layer on InSb.
The normal-incidence transmission or reflection mode operation of the plasmonic structure and
the photolithography-friendly fabrication make our structure a flexible and competitive platform
in THz plasmonic applications.
Methods
This dielectric grating/InSb structure is modeled by a commercial finite element simulation
platform COMSOL Multiphysics. This work is simulated in the frequency domain of COMSOL
RF module. Considering the complexity of this structure and to maximize the computation
speed, the simulation is modeled in 2D by setting the structure infinite and translationally
invariant in the y direction. In simulation, we set both the superstrate and substrate as air (or
vacuum) for convenience. The use periodic boundary conditions in our simulation.
Acknowledgements
The work at Tulane was supported by NSF Award No. DMR-1554866. The work at USF was
supported by the Alfred P. Sloan Research Fellow grant BR2013-123 and by KRISS grant
GP2016-034.
References
1. Beard, M. C., Turner, G. M. & Schmuttenmaer, C. A. Terahertz Spectroscopy. J. Phys. Chem.
B 106, 7146–7159 (2002).
2. Vandrevala, F., Karmakar, A., Jornet, J. M. & Einarsson, E. Extracting complex optical
properties of ultra-thin conductors using time-domain THz spectroscopy. in Infrared,
Millimeter, and Terahertz waves (IRMMW-THz), 2016 41st International Conference on 1–2
(IEEE, 2016).
3. Anker, J. N. et al. Biosensing with plasmonic nanosensors. Nat. Mater. 7, 442–453 (2008).
10
4. Wang, S. et al. Terahertz biosensing based on a polarization-insensitive metamaterial. IEEE
Photonics Technol. Lett. 28, 986–989 (2016).
5. Koenig, S. et al. Wireless sub-THz communication system with high data rate. Nat. Photonics
7, 977–981 (2013).
6. Zhang, X. & Liu, Z. Superlenses to overcome the diffraction limit. Nat. Mater. 7, 435–441
(2008).
7. Liu, S., Mitrofanov, O. & Nahata, A. Near-field terahertz imaging using sub-wavelength
apertures without cutoff. Opt. Express 24, 2728–2736 (2016).
8. Liang, Y., Yu, H., Zhang, H. C., Yang, C. & Cui, T. J. On-chip sub-terahertz surface plasmon
polariton transmission lines in CMOS. Sci. Rep. 5, 14853 (2015).
9. Kaur, S., Kaur, S., Kaur, M. & Kumar, G. Numerical analysis of terahertz surface plasmon
polaritons propagating in a parallel plate configuration. in Journal of Physics: Conference
Series 759, 012049 (IOP Publishing, 2016).
10. Mandel, I. M., Bendoym, I., Jung, Y. U., Golovin, A. B. & Crouse, D. T. Dispersion
engineering of surface plasmons. Opt. Express 21, 31883–31893 (2013).
11. Rivas, J. G., Kuttge, M., Bolivar, P. H., Kurz, H. & Sánchez-Gil, J. A. Propagation of surface
plasmon polaritons on semiconductor gratings. Phys. Rev. Lett. 93, 256804 (2004).
12. Liu, Y.-Q., Kong, L.-B., Du, C.-H. & Liu, P.-K. Spoof surface plasmon modes on doubly
corrugated metal surfaces at terahertz frequencies. J. Phys. Appl. Phys. 49, 235501 (2016).
13. Howells, S. C. & Schlie, L. A. Transient terahertz reflection spectroscopy of undoped InSb
from 0.1 to 1.1 THz. Appl. Phys. Lett. 69, 550–552 (1996).
11
14. Oszwaldowski, M. & Zimpel, M. Temperature dependence of intrinsic carrier concentration
and density of states effective mass of heavy holes in InSb. J. Phys. Chem. Solids 49, 1179–
1185 (1988).
15. Handbook series on semiconductor parameters. Vol. 2: Ternary and quaternary III-V
compounds. (World Scientific, 1999).
16. Stepanenko, O. et al. Compact mid-IR isolator using nonreciprocal magnetoplasmonic InSb
mirror. in 2016 41st International Conference on Infrared, Millimeter, and Terahertz waves
(IRMMW-THz) 1–2 (2016). doi:10.1109/IRMMW-THz.2016.7758556
17. Lin, S., Bhattarai, K., Zhou, J. & Talbayev, D. Thin InSb layers with metallic gratings: a
novel platform for spectrally-selective THz plasmonic sensing. Opt. Express 24, 19448–19457
(2016).
18. Liu, H. et al. Tunable Terahertz Plasmonic Perfect Absorber Based on T-Shaped InSb Array.
Plasmonics 11, 411–417 (2016).
19. Hu, B., Zhang, Y. & Wang, Q. J. Surface magneto plasmons and their applications in the
infrared frequencies. Nanophotonics 4, 383–396 (2015).
20. Deng, L. et al. Direct optical tuning of the terahertz plasmonic response of InSb
subwavelength gratings. Adv. Opt. Mater. 1, 128–132 (2013).
21. Rivas, J. G., Janke, C., Bolivar, P. H. & Kurz, H. Transmission of THz radiation through
InSb gratings of subwavelength apertures. Opt. Express 13, 847–859 (2005).
22. O'Hara, J. F., Averitt, R. D. & Taylor, A. J. Prism coupling to terahertz surface plasmon
polaritons. Opt. Express 13, 6117–6126 (2005).
23. O'Hara, J. F., Averitt, R. D. & Taylor, A. J. Terahertz surface plasmon polariton coupling on
metallic gratings. Opt. Express 12, 6397–6402 (2004).
12
24. Isaac, T. H., Barnes, W. L. & Hendry, E. Determining the terahertz optical properties of
subwavelength films using semiconductor surface plasmons. Appl. Phys. Lett. 93, 241115
(2008).
25. Hanham, S. M. et al. Broadband terahertz plasmonic response of touching InSb disks. Adv.
Mater. 24, (2012).
26. Azad, A. K., Zhao, Y. & Zhang, W. Transmission properties of terahertz pulses through an
ultrathin subwavelength silicon hole array. Appl. Phys. Lett. 86, 141102 (2005).
27. Azad, A. K., Zhao, Y., Zhang, W. & He, M. Effect of dielectric properties of metals on
terahertz transmission in subwavelength hole arrays. Opt. Lett. 31, 2637–2639 (2006).
28. Rivas, J. G., Schotsch, C., Bolivar, P. H. & Kurz, H. Enhanced transmission of THz radiation
through subwavelength holes. Phys. Rev. B 68, 201306 (2003).
29. Yang, Y. & Grischkowsky, D. R. High-Resolution THz Transmission and Reflection
Measurements and Consequent Understanding of Resonant Hole-Arrays. IEEE Trans.
Terahertz Sci. Technol. 3, 151–157 (2013).
30. Bhattarai, K. et al. Angle-dependent Spoof Surface Plasmons in Metallic Hole Arrays at
Terahertz Frequencies. IEEE J. Sel. Top. Quantum Electron. 23, 1–6 (2017).
31. Matsumoto, N. et al. Analysis of Dielectric Response of TiO2 in Terahertz Frequency
Region by General Harmonic Oscillator Model. Jpn. J. Appl. Phys. 47, 7725 (2008).
32. Maier, S. A. Plasmonics: fundamentals and applications. (Springer, 2007).
33. Isaac, T. H., Rivas, J. G., Sambles, J. R., Barnes, W. L. & Hendry, E. Surface plasmon
mediated transmission of subwavelength slits at THz frequencies. Phys. Rev. B 77, 113411
(2008).
13
34. Deguzman, P. C. & Nordin, G. P. Stacked subwavelength gratings as circular polarization
filters. Appl. Opt. 40, 5731–5737 (2001).
35. Masatoshi Osawa. Surface-Enhanced Infrared Absorption. in Near-field optics and surface
plasmon polaritons (ed. Kawata, S.) (Springer, 2001).
36. Bhattarai, K. et al. A Large-Area, Mushroom-Capped Plasmonic Perfect Absorber:
Refractive Index Sensing and Fabry-Perot Cavity Mechanism. Adv. Opt. Mater. 3, 1779–1786
(2015).
37. Hu, X. et al. Metamaterial absorber integrated microfluidic terahertz sensors. Laser
Photonics Rev. 10, 962–969 (2016).
14
|
1905.01277 | 2 | 1905 | 2019-05-09T18:38:57 | Polariton Nanophotonics using Phase Change Materials | [
"physics.app-ph",
"cond-mat.mes-hall",
"physics.optics"
] | Polaritons formed by the coupling of light and material excitations such as plasmons, phonons, or excitons enable light-matter interactions at the nanoscale beyond what is currently possible with conventional optics. Recently, significant interest has been attracted by polaritons in van der Waals materials, which could lead to applications in sensing, integrated photonic circuits and detectors. However, novel techniques are required to control the propagation of polaritons at the nanoscale and to implement the first practical devices. Here we report the experimental realization of polariton refractive and meta-optics in the mid-infrared by exploiting the properties of low-loss phonon polaritons in isotopically pure hexagonal boron nitride (hBN), which allow it to interact with the surrounding dielectric environment comprising the low-loss phase change material, Ge$_3$Sb$_2$Te$_6$ (GST). We demonstrate waveguides which confine polaritons in a 1D geometry, and refractive optical elements such as lenses and prisms for phonon polaritons in hBN, which we characterize using scanning near field optical microscopy. Furthermore, we demonstrate metalenses, which allow for polariton wavefront engineering and sub-wavelength focusing. Our method, due to its sub-diffraction and planar nature, will enable the realization of programmable miniaturized integrated optoelectronic devices, and will lay the foundation for on-demand biosensors. | physics.app-ph | physics | Polariton Nanophotonics using Phase Change Materials
Kundan Chaudhary1‡, Michele Tamagnone1‡*, Xinghui Yin1‡*, Christina M. Spägele1‡, Stefano L.
Oscurato1,2, Jiahan Li3, Christoph Persch4, Ruoping Li1, Noah A. Rubin1, Luis A. Jauregui5, Kenji
Watanabe6, Takashi Taniguchi6, Philip Kim5, Matthias Wuttig4, James H. Edgar3, Antonio Ambrosio7,
Federico Capasso1*
1 Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge,
Massachusetts 02138, USA.
2 Department of Physics "E. Pancini", University of Naples "Federico II", Complesso Universitario di
Monte S. Angelo, Via Cinthia 21, 80126, Naples, Italy.
3 Department of Chemical Engineering, Kansas State University, Manhattan, Kansas 66506, USA.
4 1. Physikalisches Institut IA, RWTH Aachen University, 52056 Aachen, Germany.
5 Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA.
6 National Institute for Materials Science, 1-1 Namiki, Tsukuba, 305-0044, Japan
7 Center for Nanoscale Systems, Harvard University, Cambridge, Massachusetts 02138, USA.
*Correspondence to [email protected], [email protected],
[email protected]
‡These authors contributed equally to this work.
1
Polaritons formed by the coupling of light and material excitations such as plasmons, phonons, or
excitons enable light-matter interactions at the nanoscale beyond what is currently possible with
conventional optics1-18. Recently, significant interest has been attracted by polaritons in van der Waals
materials, which could lead to applications in sensing,15 integrated photonic circuits4 and detectors3.
However, novel techniques are required to control the propagation of polaritons at the nanoscale and
to implement the first practical devices. Here we report the experimental realization of polariton
refractive and meta-optics in the mid-infrared by exploiting the properties of low-loss phonon
polaritons in isotopically pure hexagonal boron nitride (hBN)5, which allow it to interact with the
surrounding dielectric environment comprising the low-loss phase change material, Ge3Sb2Te6
(GST)19-28. We demonstrate waveguides which confine polaritons in a 1D geometry, and refractive
optical elements such as lenses and prisms for phonon polaritons in hBN, which we characterize using
scanning near field optical microscopy. Furthermore, we demonstrate metalenses, which allow for
polariton wavefront engineering and sub-wavelength focusing. Our method, due to its sub-diffraction
and planar nature, will enable the realization of programmable miniaturized integrated
optoelectronic devices, and will lay the foundation for on-demand biosensors.
Phonon polaritons (PhP) in thin films of hBN behave as confined guided optical modes, which extend as
evanescent waves into the semi-spaces above and below the hBN. Therefore, the propagation of PhPs is
affected by the refractive indices of the superstrate and substrate3,4,6,8. These are not interface modes but
exist inside the volume of hBN. The permittivity values are of opposite signs along different principal axes
and thus polaritons exhibit hyperbolic dispersion3,8,12. The degree to which the optical energy density of
polaritons extends into the substrate and superstrate depends on the wavelength and thickness of hBN.
Therefore, the excitation wavelength can be controlled to the point where the polariton is affected even by
the very first few nanometres of the substrate and superstrate17. This suggests the feasibility of substrate-
2
engineered polariton optics where, instead of nanopatterning the polaritonic material itself, optical functions
such as waveguiding and focusing are conferred through engineering the refractive index of the substrate.
A heterostructure comprising the phase change material Ge3Sb2Te6 (GST) and isotopically pure h11BN
(referred to hBN hereinafter) is the ideal system for a proof-of-concept demonstration of substrate
engineered polariton optics: hBN possess low-loss polaritons with large propagation lengths5 and GST can
support two very different refractive indices in its amorphous and crystalline phases (𝑛𝑎 = 4.2 and 𝑛𝑐 =
6.1), which can co-exist at room temperature. Additionally, GST allows for reconfigurability in terms of
writing, erasing, and re-writing optical components because it can be reversibly switched between its two
phases through electrical or optical stimuli. This phenomenon is used in commercial rewritable optical
discs21.
While GST has been previously employed to demonstrate switchable phonon polariton resonators in
quartz26, more complex applications including metasurfaces remained elusive due to limited propagation
lengths of phonon-polaritons in quartz. On the other hand, VO2 has been used as a substrate with hBN to
achieve temperature dependent polariton dispersion27. However, this approach suffers from limitations
because VO2's different phases cannot co-exist at the same temperature, and thus in close spatial proximity
as is desirable for the realisation of optical devices.
Here, we show an hBN-GST heterostructure in which arbitrary patterns can be written, erased, and re-
written to control the PhP propagation. We achieve this by defining several structures, ranging from
waveguides to diffraction-limited focusing metalenses. Specifically, we use low-loss PhPs in isotopically
pure hBN (11B isotopes with >99% purity5) with longer propagation lengths, which we combine with
Ge3Sb2Te6, a GST stoichiometry with particularly low absorption in the mid-infrared (mid-IR)28. We work
in the second Reststrahlen band (from 1361-1610 cm-1), which is associated with in-plane optical phonons5,6.
3
Figure 1 Reconfigurable polaritons in hBN-GST heterostructures. a, Writing setup and device cross-
section. A 405 nm focused laser beam is used to write and reconfigure devices on GST underneath hBN
(transparent at 405 nm). b, Longer, low power laser pulses are used to crystallise GST and shorter high-
power pulses are used to melt it to restore the amorphous phase. c, Electric field profile of polaritons for the
a-GST and c-GST cases. The electric field confinement is larger in c-GST (due to its larger refractive index)
than in a-GST. Ex represents the electric field along the direction of polariton propagation. Thicknesses for
each layer are 195 nm for hBN, 15 nm for ZnS:SiO2, 55 nm for GST and 1 mm for CaF2, which is then
considered semi-infinite. Refractive indices are 1.7 for ZnS:SiO2, 4.2 and 6.1 for GST in amorphous and
crystalline phases, respectively, 1.37 for CaF2 while hBN is modelled with the Lorentz model presented in
the Supplementary Information. d, Calculated dispersion relation of the effective index 𝑛eff for different
hBN thicknesses on a-GST and c-GST. e, Polaritons are launched by the hBN edge when light impinges on
the sample. The launched polaritons interact with the written devices and their propagation is imaged using
scattering-type scanning near field optical microscopy (s-SNOM). f, Example of optical and s-SNOM
images. Scale bar is 5 µm.
As PhPs are strongly confined, a thin layer of 55 nm of GST below hBN (195 nm of thickness) is sufficient
to significantly alter polariton propagation. To create the heterostructure, we sputter a thin film of GST on
a CaF2 substrate (in an amorphous phase as-deposited), protect it with a 15 nm layer of ZnS:SiO2 against
oxidation and then transfer exfoliated hBN to form the top layer. A pulsed laser diode is used to write and
erase patterns by selectively crystallising or re-amorphising GST21.
The PhP mode profile is affected by the state (either crystalline or amorphous) of the GST beneath it (Figure
1c), which can be quantified by using the effective index 𝑛eff = 𝑐/𝑣ph, where c is the speed of light and
4
𝑣ph is the PhP's phase velocity. Figure 1d shows the dispersion of 𝑛eff for different hBN thicknesses on
both a-GST (𝑛eff,a) and c-GST (𝑛eff,c) respectively. We use scattering-type scanning near field optical
microscopy (s-SNOM) to characterize the polaritons launched at hBN edges which propagate across the
optical elements (Figure 1e,1f) 8,13,17.
Polariton propagation in heterostructures with a- or c-GST is analogous to light propagation in two different
materials (such as air and glass). The continuity of the electric field at the boundary between two regions
implies Snell's law30:
𝑛eff,c
𝑛eff,a
=
sin(𝜃a)
sin(𝜃c)
(1)
where 𝜃 is the propagation angle in corresponding regions with respect to the interface normal.
Many conventional optical devices (such as lenses and prisms) are governed by Snell's law, suggesting that
similar components can be implemented in our hBN-GST heterostructure. The first example to illustrate
this principle is a refractive lens, specifically, a plano-convex semi-circular lens to focus PhPs (Figure 2a).
We write and erase a semi-circular plano-convex lens (radius (𝑅) = 5 𝜇m) twice and subsequently replace
it with a plano-concave lens of the same radius (Figure 2b-f). The straight hBN edge launches linear waves
(the planar equivalent of three-dimensional plane waves), which are either focused by the plano-convex lens
or diverged by the plano-concave lens (Figure 1g,h). The lateral size of the focal spot is 2 𝜇m (29 % of the
free space wavelength), which is diffraction-limited according to the Abbe limit computed for the 2D waves
(i.e., 2.08 𝜇m). The numerical aperture (NA) with respect to polaritons in a-GST is 0.55, while the NA with
respect to vacuum is 2.11, which is higher than unity due to the high-confinement of PhPs.
5
Figure 2 Rewritable flat polaritonic lenses. a, Plano-convex lens schematics for 3D and 2D semi-
spherical and semi-circular lenses. In the 2D case the material refractive index is replaced by the effective
index of the polaritons on amorphous or crystalline GST. b-f, Optical images of the written lens. The written
patterns are clearly visible in the pictures because the refractive index of a- and c-GST also differs at visible
wavelengths. First a plano-convex semi-circular lens (radius R = 5 µm) is written and measured, then it is
erased, rewritten (with same dimensions), erased again and finally the same area is reconfigured into a
plano-concave lens (R = 5 µm). g-h, Diagram of wavefronts for 2D plano-convex and plano-concave lenses
respectively. i, s-SNOM image of the plano-convex lens after the first writing. A focal spot is clearly visible.
j, Dependence of the focal length on the wavenumber. k, s-SNOM scan after first erasing. l, s-SNOM scan
of the re-written plano-convex lens. m, s-SNOM scan of the plano-concave lens (third writing). s-SNOM
images in panels i, l and m have been processed to remove the fringes outside the main beam, see
Supplementary Information for details. Scale bars are 5 µm.
We performed phase resolved s-SNOM measurements after each writing and erasing step. Using amplitude
and phase information, the wavefronts of the polaritons can be clearly imaged (see Supplementary
Information for more details on measurements and image processing). The resulting images confirm
focusing, which is shown by the narrow waist in the transmitted beam (Figure 1i). Characteristic curved
wavefronts can be seen before and after the focal spot. The position of the focal spot of the lens is measured
from the images and compared with theoretical computation with two different methods (Figure 2j). The
6
first method is based on computing the focal spot using ray optics, while the second, more accurate, method
models the focused beam as a Gaussian beam, and identifies the focal spot as the beam-waist29 (see more
details in the Supplementary Information).
After erasing, only polaritons with linear wavefronts that are launched by the hBN edge are visible, whereas
focusing is fully restored when the lens is rewritten (Figure 2k,l). Furthermore, reconfiguring the same area
to a plano-concave lens results in curved wavefronts associated with diverging PhPs (Figure 2m).
Figure 3 Prism and waveguides. a, Snell's law for 2D prisms determines deflection of polaritons. b,
Optical image of the written prism, an isosceles right-angled triangle with edges of 7.5 µm. The flake edge
is also visible. c, Optical image of the written waveguides (top 0.7 µm wide, bottom 1.1 µm wide). The
distance between the waveguides is 8.5 µm, which ensures no coupling between them. d, Diagram of
wavefronts for the prism. e, Schematics of wavefronts for a waveguide. Polaritons propagating inside the
waveguide have smaller fringe spacing due to the additional confinement of the waveguide mode. f, s-
SNOM image of prism showing a clear deflection angle of the outgoing wavefronts. g, s-SNOM image of
waveguides, showing the expected confinement of the modes inside of them. The fringe spacings are
different for waveguides with different widths, confirming that the spacing is determined by the mode of
the waveguide. h, Simulated and measured effective indices of the waveguides. The effective indices are
between 𝑛eff,a and 𝑛eff,c. i, Cross-section of the guided mode of the 0.7 µm waveguide at different
frequencies (out-of-plane Poynting vector). Scale bars are 5 µm.
The successful implementation of lenses can be extended to other common devices such as prisms and
waveguides (Figure 3). Planar prisms are simply triangles and follow Snell's law (Figure 3a). We wrote a
7
prism and two waveguides (with different widths) close to the hBN edge so that edge-launched waves can
interact with them (Figure 3b-3e). The prism is designed to be an isosceles right triangle with one side
parallel to the hBN edge such that edge-launched waves enter orthogonal to it. When traversing the
hypotenuse, the polariton propagation direction (k-vector) is bent downwards (as expected from Snell's
law), as is clearly visible in the s-SNOM measurements in the form of bent fringes (Figure 3f).
The waveguides consist of c-GST lines with widths (0.7 and 1.1 µm) smaller or comparable to the guided
polariton wavelength. They provide additional in-plane confinement such that the propagating mode is truly
one-dimensional and is confined along the waveguide. Here, the c-GST line acts as the waveguide core,
while a-GST serves as cladding. The s-SNOM measurement in Figure 3g shows that the wavefront spacing
decreases inside the waveguides, as expected from confined modes. Furthermore, the compression is greater
for the wider waveguide. This implies that the waveguide effective index 𝑛eff,wg is larger when the width
of the waveguide increases, which agrees with the behaviour known from conventional dielectric
waveguides where the core size affects the effective index of the mode. In both the conventional and the
polariton cases, the value of the waveguide effective index lies between the indices of the core and cladding
material, i.e. 𝑛eff,a ≤ 𝑛eff,wg ≤ 𝑛eff,c. We verified this behaviour by numerically calculating the waveguide
dispersion relation (see Methods) and comparing the results to s-SNOM measurements taken at different
frequencies (Figure 3h). Figure 3i shows a cross-section of a guided mode obtained from numerical
simulation, illustrating how polaritons are confined both vertically and laterally.
Metasurfaces have recently emerged as a novel and versatile method for engineering light propagation by
using arrays of discrete elements, which locally alter the phase of transmitted light. By changing the size
and shape of these elements, arbitrary predetermined phase profiles can be implemented30. Figure 4 shows
the adaptation of this concept for polaritons and its implementation in an hBN-GST heterostructure. Our
approach allows designing one-dimensional metalenses, which focus polaritons that propagate in two
dimensions. The metalens elements are truncated c-GST waveguides defined in a-GST environment and
create the hyperbolic phase profile31:
8
𝜙(𝑦) = −
2𝜋
𝜆eff,a
(√𝑦2 + 𝑓2 − 𝑓)
(2)
where y is the element position in y-direction and f is the focal length (Figure 4a). We build a phase library
for elements of varying lengths (Metalens 1, periodicity of unit cell: 1.8 µm) and widths (Metalens 2,
periodicity of unit cell: 1.2 µm) (Figure 4b and c), and subsequently incorporate the required phase profile
by choosing the corresponding elements (Figure 4d) 30. First, we demonstrate a metalens based on length
changes, then we erase and replace it with a metalens where only the width of the individual elements varies.
Figure 4 Reconfigurable Metalenses. a, Adaptation of metalenses to the 2D case. Each discrete element
changes the local phase of light, so the wavefronts converge to a focal spot. b-c, Local phase (in degrees)
of each element as a function of the element parameter and the frequency. b refers to unit cells with period
1.8 µm, 1 µm width and variable lengths. c refers to unit cells with period 1.2 µm, 9 µm length and variable
widths. d, Designed phase profile of the two metalenses (see Supplementary Information for additional
information). The design frequency was 1452 cm-1 and 1448 cm-1 for Metalens 1 and Metalens 2
respectively. e, Optical images of the written metalenses. Metalens 1 was written, characterised, erased and
subsequently the same area was reconfigured into Metalens 2. f, s-SNOM image of Metalens 1 showing
focusing of polaritons at 1452 cm-1. g, s-SNOM image of Metalens 2 at 1445 cm-1. h, Intensity profile
(square of s-SNOM signal) at the cross-section of the diffraction-limited focused polariton beam for both
lenses (dashed vertical lines in panels f and g). Scale bars are 5 µm.
9
s-SNOM characterisation was carried out after each step and clearly reveals the focusing effect of the two
designed lenses (Figures 4f and 4g). Figure 4h shows the confinement of polaritons at the focal spot.
Quantitative analysis (see Supplementary Information) of both metalenses confirms that we were able to
achieve spherical aberration free, diffraction-limited focusing using both approaches: the lateral size of the
focal spots are 1.6 𝜇m and 2 𝜇m respectively, which is in good agreement with the respective Abbe limits
(1.66 𝜇m and 1.90 𝜇m respectively). The focal spots are 23 % and 29 % of the free space wavelength
respectively.
In summary, our results clearly establish that the hBN-GST heterostructure used in this work can serve as a
versatile platform to arbitrarily control polaritons at the nanoscale to achieve freeform, transformation, and
meta optics18. While we chose to integrate hBN with GST, it can be readily combined with other polaritonic
vdW materials3 and other phase change materials31, thereby enabling a whole range of deeply
subwavelength polaritonic devices, from visible to infrared spectral regimes. Fully-fledged polaritonic
circuits can be cheaply fabricated without the need for traditional photolithography, allowing the low-cost
realisation of biosensors15, and high-density optical storage, which benefit from the extreme volume
confinement that can be achieved with polaritons. Additionally, the reconfigurability offered by GST-vdW
heterostructures and the possibility of electrically switching GST -- vdW materials heterostructures will pave
the way to applications such as modulators, photo-detectors and, more generally, programmable optical
devices as optical counterparts to field programmable gate arrays.
Methods
Sample Fabrication: A 55 nm GST-326 film (with 15 nm of ZnS:SiO2 protection layer) was sputter-coated
onto a 1 mm thick CaF2 substrate. Lithography was performed to define alignment markers (positive tone
photoresist S1813 spin coated at 3000 RPM and baked at 115 °C for 90 seconds) followed by Pt sputtering
(30 nm) and lift-off at room temperature. Isotopically pure h11BN flakes were mechanically exfoliated onto
10
the substrate after plasma activation (5 minutes of O2 plasma at 100 W) using a standard Scotch tape process.
We removed traces of the glue from the Scotch tape by placing the sample in acetone for 10 minutes
followed by an isopropyl alcohol rinse for 5 minutes and drying with nitrogen. Afterwards, samples were
further cleaned with O2 plasma (10 minutes of O2 plasma at 100 W). The thicknesses of the flakes and of
GST were confirmed through AFM measures (Cypher AFM from Asylum Research).
Reconfigurable Pattern Writing: see Supplementary Method 3 for details on the technique used to write
and erase patterns in GST.
Lens and metalens parameters: see Supplementary Method 2 for a summary of the fabricated lens
parameters.
Numerical Simulations: Numerical simulations were performed using Lumerical Mode solutions and
Lumerical FDTD. 1D simulations in Mode Solutions were used to calculate the effective indices 𝑛eff,a,
𝑛eff,c. A mesh size of 1 nm was used to compute the fundamental mode profile and effective indices of the
hBN-GST-326 heterostructures in the RS2 band of h11BN. Here, hBN was modelled as an anisotropic
dielectric with its permittivity values obtained from the Lorentz model (Supplementary Method 1). The
effective indices of the waveguides were calculated via 2D simulations of their cross-sections, also done in
Modes Solutions. Metalenses were designed by first simulating numerically each element in a periodic
environment with full wave 3D simulations performed in FDTD. The result was a library of elements, used
to implement the required phase profile. The final metalens was also completely simulated to verify the
focusing behaviour. For 3D simulations we used a mesh size smaller than 50 nm along x- and y-axes and 5
nm along the z-axis. A series of dipole sources were used to excite the polariton modes in hBN in the RS2
band.
s-SNOM Measurements: The near field scans were obtained using a commercial system from NeaSpec.
Tapping-mode AFM is used (tapping amplitude of 130 nm, Pt-Ir coated tips with resonant frequency of
~300 kHz, tip diameter of ~20 nm). A quantum cascade laser (QCL) array from Daylight Solutions was
11
used as a tuneable mid-IR source for imaging. The phase-amplitude images are obtained using a pseudo-
heterodyne demodulation. The images of lenses and metalenses were processed using the technique shown
in the Supplementary Method 4 to isolate the polaritons focused by the structure. In all cases polaritons are
self-launched by the edges32. See Supplementary method 5 for additional measurements.
References
[1] Ma, W., et al. In-plane anisotropic and ultra-low-loss polaritons in a natural van der Waals crystal.
Nature 562, 557 (2018).
[2] Alonso-González, P., et al. Acoustic terahertz graphene plasmons revealed by photocurrent nanoscopy.
Nat. Nanotech. 12, 31 (2017).
[3] Basov, D., Fogler, M. & de Abajo, F. G. Polaritons in van der Waals materials. Science 354, aag1992
(2016).
[4] Low, T. et al. Polaritons in layered two-dimensional materials. Nat. Mater. 16, 182 -- 194 (2017).
[5] Giles, A. J. et al. Ultralow-loss polaritons in isotopically pure boron nitride. Nat. Mater. 17, 134 -- 139
(2018).
[6] Dai, S., et al. Tunable phonon polaritons in atomically thin van der Waals crystals of boron nitride.
Science 343, 1125-1129 (2014).
[7] Li, P. et al. Infrared hyperbolic metasurface based on nanostructured van der Waals materials. Science
359, 892 -- 896 (2018).
[8] Ambrosio, A. et al. Selective excitation and imaging of ultraslow phonon polaritons in thin hexagonal
boron nitride crystals, Light Sci. App. 7, 27 (2018).
[9] Caldwell, J. D. et al. Low-loss, infrared and terahertz nanophotonics using surface phonon polaritons.
Nanophotonics 4, 44 -- 68 (2015).
[10] Yoxall, E. et al. Direct observation of ultraslow hyperbolic polariton propagation with negative phase
velocity. Nat. Photon. 9, 674 -- 678 (2015).
[11] Kumar, A., Low, T., Fung, K. H., Avouris, P. & Fang, N. X. Tunable light-matter interaction and the
role of hyperbolicity in graphene-hbn system. Nano Lett. 15, 3172 -- 3180 (2015).
[12] Tamagnone, M. et al. Ultra-confined mid-infrared resonant phonon polaritons in van der Waals
nanostructures. Sci. Adv. 4, 4 -- 10 (2018).
[13] Grigorenko, A. N. et al. Graphene plasmonics. Nat. Photon. 6, 749-758 (2012).
[14] Hillenbrand, R., T. Taubner, and F. Keilmann. Phonon-enhanced light -- matter interaction at the
nanometre scale. Nature 418, 159 (2002).
[15] Rodrigo, D., et al. Mid-infrared plasmonic biosensing with graphene. Science 349 165-168 (2015).
[16] Chen, J., et al. Optical nano-imaging of gate-tunable graphene plasmons. Nature 487, 77 (2012).
[17] Chaudhary, K. et al. Engineering Hyperbolic Phonon Polaritons in van der Waals Heterostructure
Waveguides to Enhance In-plane Optical Anisotropy, arXiv:1807.03339, (2018).
[18] Ashkan, V., and Engheta, N. Transformation optics using graphene. Science 332, 1291-1294 (2011).
[19] Wuttig, M., Bhaskaran, H. and Taubner, T. Phase-change materials for non-volatile photonic
applications. Nat. Photon. 11, 465 -- 476 (2017).
[20] Lin, H. et al. Chalcogenide glass-on-graphene photonics. Nature Photon. 11, 798 -- 805 (2017).
[21] Wuttig, M. & Yamada, N. Phase-change materials for rewriteable data storage. Nat. Mater. 6, 824 --
832 (2007).
[22] Tuma, T., et al. Stochastic phase-change neurons. Nat. Nanotech. 11, 693 -- 699 (2016).
[23] Ríos, C. et al. Integrated all-photonic non-volatile multi-level memory. Nat. Photon. 9, 725 -- 732
(2015).
12
[24] Hosseini, P., Wright, C. D. and Bhaskaran, H. An optoelectronic framework enabled by low-
dimensional phase-change films. Nature 511, 206 -- 211 (2014).
[25] Yin, X. et al. Beam switching and bifocal zoom lensing using active plasmonic metasurfaces. Light:
Science & Applications 6, e17016 (2017).
[26] Li, P. et al. Reversible optical switching of highly confined phonon -- polaritons with an ultrathin
phase-change material. Nat. Mater. 15, 870 -- 875 (2016).
[27] Folland, T. G. et al. Reconfigurable infrared hyperbolic metasurfaces using phase change materials.
Nat. Commun. 9, (2018).
[28] Michel, A.-K. U. et al. Using Low-Loss Phase-Change Materials for Mid-Infrared Antenna
Resonance Tuning. Nano Lett. 13, 3470 -- 3475 (2013).
[29] Yariv, A. & Yeh, P. Photonics. (Oxford University Press, USA, 2007).
[30] Khorasaninejad, M. et al. Metalenses at visible wavelengths: Diffraction-limited focusing and
subwavelength resolution imaging. Science 352, 1190 -- 1194 (2016).
[31] Raty, J.‐Y., et al. A Quantum‐Mechanical Map for Bonding and Properties in Solids. Advanced
Materials, 2018 1806280 (2018).
[32] Gilburd, L., et al. Hexagonal boron nitride self-launches hyperbolic phonon polaritons. The journal of
physical chemistry letters, 8 2158-2162 (2017).
Acknowledgements
Funding: This work was supported by the NSF EFRI, award no. 1542807. This work was performed in part
at the Center for Nanoscale Systems (CNS), a member of the National Nanotechnology Coordinated
Infrastructure Network (NNCI), which is supported by the National Science Foundation under NSF award
no. 1541959. M.T. acknowledges the support of the Swiss National Science Foundation (SNSF) grant no.
168545 and 177836. S.L.O. acknowledges "Fondazione Angelo Della Riccia", and the program for
"International Mobility of Researchers" of the University of Naples "Federico II" (Italy), for financial
support. The h11BN crystal growth was supported by the National Science Foundation, award number
1538127. K.W. and T.T. acknowledge support from the Elemental Strategy Initiative conducted by the
MEXT, Japan and and the CREST (JPMJCR15F3), JST.
Author Contributions
K.C., M.T. and X.Y. conceived the project. M.T., K.C., X.Y., C.S., A.A. and F.C. devised experiments.
X.Y., C.S. designed and implemented the setup for writing patterns on GST-326 with help from R.L. and
N.A.R.. J.L. and J.E. provided isotopically pure h11BN for final devices. C.P. and M.W. optimized and
13
deposited the GST layer on the substrate. K.W., T.T., L.A.J. and P.K. provided natural abundance hBN for
initial tests. X.Y. and C.S. wrote and erased patterns on h11BN/GST-326 heterostructures. K.C., M.T.,
S.L.O., C.S. and A.A., performed s-SNOM and AFM scans. K.C., M.T., X.Y. and C.S. performed FDTD
simulations. M.T. and K.C. analysed the experimental data. F.C. led the project. All authors contributed to
discussions and manuscript preparation.
Author Information
The Authors declare no competing financial interests. Please contact M.T. for any request of materials and
additional data.
14
|
1711.06034 | 1 | 1711 | 2017-11-16T11:29:00 | An Inkjet Printed Chipless RFID Sensor for Wireless Humidity Monitoring | [
"physics.app-ph"
] | A novel chipless RFID humidity sensor based on a finite Artificial Impedance Surface (AIS) is presented. The unit cell of the AIS is composed of three concentric loops thus obtaining three deep and high Q nulls in the electromagnetic response of the tag. The wireless sensor is fabricated using low-cost inkjet printing technology on a thin sheet of commercial coated paper. The patterned surface is placed on a metal backed cardboard layer. The relative humidity information is encoded in the frequency shift of the resonance peaks. Varying the relative humidity level from 50% to 90%, the frequency shift has proven to be up to 270MHz. The position of the resonance peaks has been correlated to the relative humidity level of the environment on the basis of a high number of measurements performed in a climatic chamber, specifically designed for RF measurements of the sensor. A very low error probability of the proposed sensor is demonstrated when the device is used with a 10% RH humidity level discrimination. | physics.app-ph | physics | M. Borgese, F. A. Dicandia, F. Costa, S. Genovesi, G. Manara "An Inkjet Printed
Chipless RFID Sensor for Wireless Humidity Monitoring" IEEE Sensors Journal
(Volume: 17, Issue: 15, Aug.1, 1 2017)
doi: 10.1109/JSEN.2017.2712190
URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7938611
© © 2017 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.
>IEEE SENSORS JOURNAL SUBMITTED<
1
An Inkjet Printed Chipless RFID Sensor for
Wireless Humidity Monitoring
Michele Borgese, Student Member, IEEE, Francesco Alessio Dicandia, Student Member, IEEE,
Filippo Costa, Member, IEEE, Simone Genovesi, Member, IEEE, and Giuliano Manara, Fellow, IEEE
Abstract- A novel chipless RFID humidity sensor based on a
finite Artificial Impedance Surface (AIS) is presented. The unit
cell of the AIS is composed of three concentric loops thus obtaining
three deep and high-Q nulls in the electromagnetic response of the
tag. The wireless sensor is fabricated using low-cost inkjet printing
technology on a thin sheet of commercial coated paper. The
patterned surface is placed on a metal-backed cardboard layer.
The relative humidity information is encoded in the frequency
shift of the resonance peaks. Varying the relative humidity level
from 50% to 90%, the frequency shift has proven to be up to
270MHz. The position of the resonance peaks has been correlated
to the relative humidity level of the environment on the basis of a
high number of measurements performed in a climatic chamber,
specifically designed for RF measurements of the sensor. A very
low error probability of the proposed sensor is demonstrated when
the device is used with a 10% RH humidity level discrimination.
Index Terms-Radio Frequency Identification (RFID), Chipless
RFID, Sensor, Artificial Impedance Surface (AIS).
I. INTRODUCTION
T
he research of methods used to transform a chipless RFID
tag into a chipless RFID sensor has attracted increasing
attention in recent years due to the growing interest in
proposing low-cost sensors for realizing the pervasive networks
required by the Internet of Things paradigm. At the basis of all
proposed solutions there is the use of a material acting as a
transducer of the physical parameter that has to be sensed into
a change of the RF response of the tag [1], [2]. Different
environmental parameters can be monitored depending on the
physical properties of the transducer. Although wired sensors
are the most popular and well-established ones [1], [3], [4], in
recent years wireless sensors based on direct radio-frequency
interrogation have been intensively investigated [5], mainly to
reduce the price of sensors nodes and deployment cost. One of
the first examples is provided in [6], where a chipless RFID
temperature sensor is realized by using three functional layers
of magnetic materials. An alternative solution is based on
Surface Acoustic Wave (SAW) technology. In [7], a wireless
Manuscript received October, 11th 2016.
M. Borgese, F.A. Dicandia, F. Costa, S. Genovesi, and G. Manara are with
Dipartimento di Ingegneria dell'Informazione, University of Pisa,
56122 – Pisa. E-mail: michele.borgese(alessio.dicandia)@for.unipi.it,
filippo.costa(simone.genovesi, g.manara)@iet.unipi.it.
temperature and pressure sensor working around 433 MHz is
designed by the synergic use of two SAW resonators. In [8], a
chipless RFID sensor for detection of CO2 is realized by
exploiting the properties of a polymer/single-walled carbon
nanotube ink that is used to load one resonator, whilst a second
one is used for calibration. Both solutions are fabricated with a
Dimatix printer [9] for material deposition. Furthermore,
metamaterials have been proposed in [10], [11] for realizing
temperature as well as strain sensors. Chipless RFID sensors
exploiting the difference between the backscattered signal
determined by the structural mode and the antenna mode have
been proposed in [12] for monitoring ethylene gas. A sensor
based on group-delay and RCS changes has been presented in
[13] for humidity monitoring. The Relative Humidity (RH)
variation is translated into an alteration of the frequency
response by employing silicon nanowires deposited on a PCB
chipless RFID tag. In [14], a chipless RFID humidity sensor is
realized by adding polyvinyl-alcohol (PVA) on top of a single
electric inductive-capacitive resonator. The PVA polymer
needs to be dissolved and magnetically stirred before being
applied with a droplet on the resonator. In [14], the two
antennas are required because the sensing of the humidity level
is based on the detection of the transmission coefficient. The
shift proves to be up to 607 MHz around 6.87 GHz. As shown
in [15], inductive coupling can be used to read chipless RFID
humidity sensors. The humidity estimation is based on a
frequency shift of 150MHz of the frequency response. The
same principle is employed in [16] where a dual-polarized
chipless tag is printed with silver nanoparticle-based ink on HP
photo paper. This moisture sensor relies on the frequency shift
of the peaks encoded in the radar cross section (RCS) of the tag
and requires a bandwidth of 13.5 GHz and a dynamic of the
RCS within (-80dB,-30dB). Concerning
the fabrication
methods of printed sensors, standard photolithographic process
can be used [17], [18]. An alternative approach is based on the
direct printing of the metal pattern with inkjet-printing
technology [4], [19]–[23]. The former solution could be
problematic since potential air gaps between the resonating
particles and the sensitive materials could reduce the sensitivity
of measurement. Inkjet-printed sensors are usually fabricated
by using a Dimatix material printer, which is suitable for
different sensitive materials. The fabrication method is versatile
and precise but requires expensive hardware.
>IEEE SENSORS JOURNAL SUBMITTED<
2
The sensor proposed in this paper is instead realized with
commercial desktop inkjet printer and the ink does not require
any curing. The employed conductive ink is applied to a special
paper which acts as a transducer of the humidity level of the
external environment into a variation in the frequency response
of the chipless RFID tag. The frequency shift has been
demonstrated to be as high as 270 MHz without requiring any
precise recovering of the RCS value. The sensor is provided
with a ground plane (GP), which improves the quality factor of
the employed resonator. In addition, the GP allows the sensor
to be placed on objects, even metallic ones, without any
detuning effect of the frequency response of the tag. Finally, the
reading system comprises only a single antenna operating in
linear polarization within the bandwidth 2-8 GHz.
The paper is organized as follows. In Section II the periodic
configuration of the proposed chipless RFID sensor and the
working principle are addressed. Section III aims to properly
correlate the electromagnetic response of the tag with the level
of humidity of the environment as well as to illustrate the
ad-hoc setup designed for the tests. The fabricated prototype
and measurements are proposed in Section IV. A statistical
analysis of the measured data is also proposed in Section V to
assess the reliability and robustness of the chipless sensor.
Finally, the conclusions are reported in Section VI.
II. CHIPLESS RFID SENSOR
The proposed chipless RFID sensor tag is based on a finite
Artificial Impedance Surface (AIS). The AIS comprises a finite
Frequency Selective Surface (FSS), formed by only a few unit
cells, accommodated on a cardboard substrate backed by a
metallic ground plane. The FSS is inkjet-printed on a thin sheet
of NB-TP-3GU100 Mitsubishi paper. The employed paper is
coated with a thin layer composed of polyvinyl alcohol and
Aluminum Oxide in order to allow the ink to be correctly
deposited on it. The silver nanoparticle ink is deposited with a
conventional piezoelectric
inkjet printer (Brother DCP-
J152W). It is worth highlighting that the conductive pattern is
obtained without any heating or sintering of the ink thus
obtaining a fast and cost-effective design process [24]. The
advantage of this configuration is that the resonators are directly
in contact with the substrate and no air gaps are present.
Typically, the air gaps invalidate this type of measurements.
A graphic representation of the proposed tag is reported in
Fig. 1. The unit cell (P = 15 mm) consists of a three loop FSS
printed on a thin paper sheet (dm = 150 µm) and placed on a
thick grounded cardboard layer (dc = 3 mm) characterized by a
permittivity εr = 2.4 - j0.2.
Fig. 1 – Stack-up of the chipless sensor and layout of the resonant element.
A. Periodic Configuration
The FSS and
together
the FSS, which
the ground plane
form a
subwavelength resonant cavity. The number of resonances of
the cavity is proportional to the resonances of the FSS. If the
tag is used on a metallic surface, the ground plane is already
available and the tag can be simply realized by printing the FSS
on the Mitsubishi paper and gluing it on a cardboard layer,
which is then placed on the metal surface. It is worth noting that
the reading of a chipless tag composed by a single FSS without
the ground plane might be critical. In fact, the presence of
objects close to the backside of the tag could affect the
reflection response of
is a partially
transmitting/reflecting surface (PRS). Moreover, the quality
factor of the resonances is much higher in the case of
metal-backed FSS than for a freestanding FSS. This can be
explained either in the frequency domain by using a simple
equivalent circuit [25] or in time domain by using the
interference theory commonly exploited in the analysis of
Fabry-Perot resonators [26]. The latter approach consists of
considering the FSS as a partially reflecting mirror. The total
reflection coefficient of the cavity is the result of an infinite set
of contributions. The first one is due to the first reflection of the
FSS, the second one is due to the part of the signal crossing the
first FSS that is reflected back by the ground plane, and so on
and so forth. A sketch of this time-domain representation is
reported in Fig. 2(a). The geometric series converges to the
global reflection coefficient of the cavity:
(1)
In (1) T1 represents the transmission coefficient of the FSS, R1a
and R1b are the reflection coefficients of the FSS for the two
different directions of incidence, which are respectively from
the AIS to the GP (AISGP) and from the GP to the AIS
(GPAIS). The reflection coefficients R1a is equal to R1b if the
FSS is sandwiched between two identical dielectric layers. The
propagation constant and the thickness of the cardboard spacer
are respectively indicated with β and d. As an example, we have
simulated the stack-up shown in Fig. 1. In this case, the
dielectrics enclosing the FSS are different (air above and
photo-paper and thick cardboard below) and consequently R1a
and R1b are different. By using those transmission/reflection
coefficients, the response of the whole tag including the ground
plane can be computed by using relation (1) without repeating
the fullwave simulation of the entire structure. In doing that, the
phases of the transmission/reflection coefficients have to be
preemptively de-embedded to the position of the FSS [27].
The comparison between the interference theory approach and
the fullwave simulation of the entire structure preformed with a
periodic MoM code (PMM) is reported in Fig. 2(b). The plot
shows a good agreement between the two approaches since the
FSS is sufficiently distanced from the ground plane and its
impedance is not perturbed [28]. If this decoupling condition is
not satisfied, the impedance of the FSS is altered by the
presence of the ground plane. In this case, the equivalent circuit
2211211jdajdbTeRRReCardboardGround planeMitsubishi paperdcdmMitsubishi inkP >IEEE SENSORS JOURNAL SUBMITTED<
3
model would be more accurate than the interference method
approach.
(a)
(b)
Fig. 2 – (a) Pictorial representation of the reflection coefficient of an AIS;
(b) Reflection coefficient of the sensor tag formed by a three loops FSS
placed on a 4mm thick cardboard layer backed by a metallic ground plane.
The increasing or decreasing number of unit cells does not
improve the coding capacity but it allows increasing or
decreasing the level of the RCS and thus improving the tag
detection probability for a given reading distance. The effect of
the periodic surface on the RCS for a chipless RFID tag based
on 5 nested loops printed on top of a 1.6-mm-thick FR4 layer,
is shown in Fig. 3. In particular, the RCS as a function of the
frequency for a 1x1, 2x2 and 3x3 unit cells is reported. The full
wave simulation of the finite structures, performed with a
commercial Finite Element Method code (HFSS), shows that
by increasing the number of unit cells it is possible to increase
both the RCS and the Q-factor of the tag response. Indeed, by
comparing the curves reported in Fig. 3, it is possible to observe
that the absorption peaks of the 3x3 tag are deeper and sharper
with respect to the 1x1 and 2x2 tag. In addition, it is apparent
that in the case of the tag composed by a single unit cell, the
low frequency peaks between 2GHz and 5GHz are barely
recognizable because of the small size of the tag. For
completeness, the simulation of the infinite periodic structure
performed with the PMM code is also reported in in Fig. 3.
highlighting that this is the first time that the commercial
Mitsubishi ink has been proved to provide humidity sensitivity
when deposited on the Mitsubishi paper substrate. This
structure exhibits a copolar frequency response with a certain
number of deep nulls depending on the topology of the AIS. A
variation of the RH level of the surrounding environment
produces a variation of the electric permittivity of the CIM with
a consequent downshift of the deep nulls of the tag response in
the frequency spectrum. The sensing mechanism of the tag is
based on the translation of the deep nulls shift into a variation
of the RH level in the environment. Even if a single frequency
peak is sufficient in principle, the estimation of the RH level at
different frequencies with multiple frequency peaks may be
beneficial. Indeed, it provides a certain level of redundancy of
the relative humidity variations, thus reducing the probability
of false reading. Although the cardboard substrate is a moisture
sensitive material, the Mitsubishi paper is the only CIM of the
sensor because the cardboard layer is not directly exposed to the
external environment (and consequently to the moisture). The
Mitsubishi paper, the cardboard and the metallic ground plane
are bonded together with a plastic adhesive tape, which is
placed on the edge of each side of the tag. In addition, the tape
isolates the cardboard from the external environment. It is
important to highlight the effect of the distance between the
resonator elements (the nested rings) and the CIM. If we
describe the AIS as a lumped LC circuit, the observed
downshift of the resonance frequency is caused by the increase
in the aforementioned capacitance value determined by an
increased permittivity of the material that is closely interacting
with the resonators of the AIS. If the resonators are directly
printed on the CIM, there is no air gap and the amount of
frequency shift is maximum. On the contrary, it is apparent that
even a small air gap can deteriorate the sensor response since it
limits the amount of shift achieved with the permittivity change
of the CIM. This is clearly reported in Fig. 4 where a
double-ring chipless RFID tag with the CIM placed at a
distance named 'air gap' from the resonators is simulated. The
stack-up of the analysed structure is shown in the inset of Fig.
4(a). The CIM permittivity varies within [2 - 2.9] and
air gap within
the CIM
permittivity leads a remarkable frequency downshift when
air gap = 0 mm (Fig. 4(a)) but a limited shift is observed when
air gap = 0.2 mm (Fig. 4(b)).
[0 - 0.1] mm. The change of
Fig. 3 – Radar Cross Section as a function of the frequency for the chipless
RFID tag composed by 1x1 (20 mm x 20 mm), 2x2 (30mm x 30 mm) and
3x3 unit cells (45 mm x45 mm) and for the infinite periodic structure.
B. Working Principle of the sensor
The paper substrate with the coating layer acts as a chemical
interactive material (CIM) because it is sensitive to humidity
variations. For this reason, it has been employed for designing
the proposed chipless RFID sensor. The stack-up of the sensor
comprises the FSS printed on the Mitsubishi paper which is
placed on a metal-backed cardboard substrate. It is worth
(a)
-50-40-30-20-100234567891x12x23x3InfiniteRCS (dBm2)Frequency (GHz)Fr4air gapSensitive material
>IEEE SENSORS JOURNAL SUBMITTED<
4
mist-maker, the fan and the commercial humidity sensor, is able
to automatically control the RH level inside the box according
to a humidity profile chosen by the user. The humidity profile
is defined by a table in which each entry contains the value and
the duration of the desired RH level. During the RH controlled
cycle, the electromagnetic response of the chipless tag is
collected at the predetermined rate. Subsequently, having
completed the measurements, the electromagnetic response of
the tag is correlated to the RH level with a post-processing
algorithm. For the single measure, the correlation is performed
on the basis of the timestamp registered by the VNA and the
time stamp registered by the Matlab code, which controls the
climatic chamber. At the end of this stage, all the data provided
by the commercial humidity sensor is associated with the single
RF measure.
IV. PROTOTYPE AND MEASUREMENT RESULTS
The fabricated prototype consists of 3x3 FSS elements
placed on a grounded cardboard substrate with thickness equal
to 4 mm. The stack-up and the unit cell of the tag are shown
respectively in Fig.1 (a) and (b). The periodic pattern has been
printed with a water-based silver nanoparticle ink, which has
been printed on a single side coated transparency PET film
(NB-TP-3GU100 Mitsubishi paper). A stereomicroscope image
of the printed chipless tag with a scale bar of 200 µm is shown
in Fig. 6a. An enlarged version of this picture is reported in Fig.
6b in which the scale bar is 500 m. Despite the use of a
commercial printer, the Fig. 6(b) shows a very high accuracy of
the printing process. The area between one loop and the other
is clean and there is no presence of satellite drops.
(a)
(b)
Fig. 6 – Stereomicroscope image of the printed chipless tag. The scale bar
is 200 µm in (a) and 500 µm in (b). The microscope adopted in this work is
a Leica M165C.
After carrying out the measurements and the post processing
stage, it is possible to link the variation of the RH level in the
climatic chamber to the variation of the electromagnetic
response of the proposed chipless RFID tag. The frequency
responses of the tag measured for RH levels, equals to 50%,
70%, 80% and 90% are shown in Fig. 7. It is evident that the
tag response exhibits 3 deep nulls at the frequencies of
3.076 GHz, 5.888 GHz and 7.275 GHz when RH=50%
(minimum RH level achievable in the climatic chamber). By
increasing the RH level inside the box, a downshift of the
resonance peaks for the three frequencies is visible. This is due
to the fact that the paper absorbs the moisture inside the climatic
chamber and increases its electrical permittivity. Consequently,
the resonant frequencies of the printed tag decrease. The higher
the RH level the greater the downshift of the frequency response
of the tag. The cardboard does not suffer from any degradation
when the RH level is in the range 50% to 90% at room
(b)
Fig. 4 – Frequency response for a varying CIM permittivity within [2 - 2.9]
and air gap = 0 (a) air gap = 0.2mm (b). Stack-up of the chipless RFID tag
with the CIM at distance air gap from the resonators in inset (a).
III. MEASUREMENT SETUP
In order to properly correlate the electromagnetic response of
the tag with the level of humidity of the environment, a high
number of measurements is required. Consequently, it is
important to use a reliable setup that allows the collection of a
large amount of data during a certain interval of time. The
measurement setup is sketched in Fig. 5.
Fig. 5 – Pictorial sketch of measurement setup. The dimension are not in
scale.
The chipless tag is attached inside a plastic box of dimensions
77 x 58 x 43cm at a height of 22cm from the bottom of the box.
A dual-polarized horn (Flann DP240) is placed in front of the
tag at a distance R of 25cm and it is connected to a two-ports
VNA (Keysight - E5071C). The VNA is connected to a laptop
via an USB cable. The electromagnetic response of the tag is
measured at the desired time intervals by using a Matlab code.
At the same time, it is possible to regulate the level of humidity
inside the box with a feedback system controlled by Matlab. To
perform this operation, an 8-channel relay (KMTronic USB
Relay) driven by the laptop computer is connected to a
mist-maker placed inside the box. The relay is also connected
to a fan placed inside of the box. The mist-maker can increase
the relative humidity (RH) whereas the fan removes the moist
air inside the box with a consequent reduction of the RH level.
A commercial humidity sensor (RS Pro 1365 Data Logger),
connected to the laptop computer, monitors the RH level inside
the box. When the RH level exceeds the value defined by the
predetermined humidity profile, the fan is activated. In this
level. This
way,
Matlab-controlled system composed of
relay,
the
the desired
the
the RH
is reduced
to
>IEEE SENSORS JOURNAL SUBMITTED<
5
temperature. In addition, as mentioned in Section II, the
cardboard layer is not directly exposed to the external
environment. The Mitsubishi paper employed
the
fabrication of the tag is a PET film which does not degrade with
the humidity of the external environment. According to the
datasheet [29], the ink exhibits a good solubility in water.
for
Fig. 7 – Normalized reflection coefficient measurement of the chipless tag
for different humidity levels.
In order to verify whether the performance of the sensor is
compromised when it is employed in extreme conditions of
humidity and if the process is reversible, the tag has been
repeatedly treated with nebulized water. The tag response as a
function of the frequency for different condition is shown in
Fig. 8.
Fig. 8 – Tag response as a function of the frequency for different conditions
of the tag. Several layers of nebulized water are deposited on the surface of
the tag. Subsequently, the tag is dried with warm air.
After the deposition of two layers of nebulized water, a down
shift and a reduction of the depth of the absorption peaks is
observed. Subsequently, once
layers of
nebulized water are deposited, the frequency response of the tag
is completely degraded. Indeed, the high frequency peak is still
visible and a further downshift of the peak is present.
Conversely, the other two peaks are not recognizable. It is
two additional
worthwhile to point out that this extreme test condition is
outside the considered working range and much more severe
than 90% humidity. However, it is interesting to observe that,
once the tag is dried with warm air, the frequency response of
the tag returns to its initial state. As shown in Fig. 8, the curves
of the dry and dried tag are superimposed. From this result, it is
possible two conclude that the tag is able to absorb and release
humidity with a reversible process. In addition, although the ink
exhibits a good solubility in water, the direct application layers
of water on the surface of the tag, does not permanently
jeopardize the function of the sensor.
In order to realize a humidity sensor with the proposed
chipless tag, it is necessary to correlate the position of the
resonance peaks to the RH level. To perform this task, the
frequency response of the tag was monitored when the RH level
within the climate chamber was increased from 60% to 90% in
steps of 10%. To test the moisture absorption time of the tag,
each humidity level was kept constant for 15 minutes.
Furthermore, to the purpose of verifying whether the paper was
able to release the absorbed moisture, after reaching the 90%,
the RH level was decreased to 60% in steps of 10%, similar to
the previous process. In this way, a stepped RH profile with a
total observation equal to 105 minutes is obtained. The
measurements of the tags were carried out at intervals of 10
seconds, thus obtaining 630 measures in the observation time.
A short measurement interval has been chosen in order to
appreciate the reaction time of the tag to the variation of the RH
level.
With the aim of better displaying the shift of the resonance
peaks with the variation of the RH level, the resonant
frequencies as a function of the observation time has been
plotted. The curves show the behavior of all three working
frequencies of the tag (Fig. 9 (a), (b), (c)). The level of humidity
inside the climatic chamber as a function of the observation
time is superimposed in the same graphs. In all three of the
plots, it is evident that in correspondence to a rapid variation of
the RH curve, the resonance frequencies vary rapidly as well.
On the contrary, when the RH level is constant, the resonance
frequencies are almost constant. In addition, these graphs
confirm that the moisture absorption of the tag is a reversible
phenomenon. In fact, when the RH level is decreased, the
frequency of the three deep nulls increases, thus obtaining the
symmetric RH profile shown in Fig. 9 (a), (b), (c).
In order to compare the magnitude of the frequency shift for
the three bands, it is convenient to consider the normalized
resonance frequency shift.
(a)
(b)
(c)
Fig. 9 – RH level and resonance frequency as a function of the observation time for the three working bands of the chipless tag: Band 1 (a), Band 2 (b) and Band 3 (c).
-30-25-20-15-10-502.533.555.566.577.58RH=50%RH=70%RH=80%RH=90%Normalized Tag Reflection (dB)Frequency (GHz)Band 1Band 2Band 3-15-10-502.533.545.566.577.58DRY2 LAYERS5 LAYERSDRIEDNormalized Tag Reflection (dB)Frequency (GHz)505560657075808590952.922.942.962.9833.023.043.063.080:0020:0040:0060:0080:00100:00RHBand 1RH (%)Resonance Frequency (GHz)Time (min)505560657075808590955.65.655.75.755.85.855.90:0020:0040:0060:0080:00100:00RHBand 2RH (%)Resonance Frequency (GHz)Time (min)505560657075808590956.856.96.9577.057.17.157.27.250:0020:0040:0060:0080:00100:00RHBand 3RH (%)Resonance Frequency (GHz)Time (min) >IEEE SENSORS JOURNAL SUBMITTED<
6
It is defined as the value of the resonance frequency at a
certain time, f0(t), divided by the value of the resonance
frequency registered at the time t0 when the minimum level of
RH is present in the chamber:
(2)
In this way, it is possible to calculate the percentage frequency
shift with respect to the resonance frequency measured in
normal humidity conditions. The RH level and the normalized
frequency shift as a function of the observation time for the
three working bands of the chipless tag is reported in Fig. 10.
Fig. 10 – RH level and the normalized resonance frequency shift as a
function of the observation time for the three working bands of the chipless
tag.
It is interesting to note that the tag exhibits almost the same
relative shift in the three bands. This property is very useful
because it increases the redundancy of the information retrieved
by the interrogation of the chipless tag with a consequent
reduction of the false reading probability of the RH level.
V. STATISTICAL ANALYSIS
In order to analyse the reliability and robustness of the
proposed sensor, a statistical analysis of the measured data has
been performed. The resonant frequencies collected from the
electromagnetic measures can be considered as a set of random
variables with a certain probability density function (PdF). In
particular, the PdF for different RH levels for the three working
bands have been calculated and are respectively shown in Fig.
11 (a), (b), (c). The measured frequency response of the tag is
affected both by the oscillations of the RH level inside the
climatic chamber and by some inaccuracies of the designed
sensor. Ideally, in absence of errors introduced by the
environment and equipment, the PdF for all resonance
frequencies would be a Dirac delta function. However, because
of measurement uncertainties, the statistical distribution of the
measurement is not a Dirac delta function. It is demonstrated
that the noise introduced by the environment, the equipment
and the sensor can be modelled as Additive White Gaussian
Noise (AWGN). To this purpose, in Fig. 11 the Normal PdF
with the same mean value and variance calculated from
measured data have been reported. The figure shows a good
agreement between the measured data and the Normal
distributions. In the graphs of the measurements, some
fluctuations around the desired value in the stepped RH profile
are visible. These fluctuations are due to the feedback
mechanism used to stabilize the humidity in the climatic
chamber. In addition, these fluctuations produce additional
residual uncertainty in the measurements with a consequent
increase of the variance of the PdF. The data collected from
these graphs has been reported in Table I. In the table, the mean
value (), the variance (2) and the maximum distance of the
resonance frequency with respect to the mean value (max), have
been reported for different RH levels and for the three working
bands of the chipless tag. Based on the Gaussian model of the
noise, it is interesting to evaluate the performance of the tag
when it is employed as a threshold humidity sensor. In
particular, the error probability (P(e)) in the estimation of RH
level by using a single interrogation of the tag can be calculated
as follows:
(3)
where k represents the set of the RH levels (sample space of the
variable RH) and N is the number of RH levels assumed by k.
P(RH = RHk(n)) represents the a-priori probability that RH
assumes the value RHk(n) and P(e RH= RHk(n)) are the
conditional probabilities. In this model, the RH level is
assimilated to a discrete random variable with uniform
distribution with the a-priori probabilities equal to 1/N. In this
case, the optimum thresholds (n) are fixed in the middle of the
mean values (Fig. 12) according to the maximum a-posteriori
probability (MAP).
(a)
(b)
(c)
Fig. 11 – Comparison between the PdF of the resonance frequency and the Normal PdF for different RH levels (60%, 70%, 80%, 90%) for the three working
bands of the chipless tag: Band 1 (a), Band 2 (b) and Band 3 (c).
000ftNormFreqtft50607080900.940.950.960.970.980.991020406080100120140160180200RHBand 1Band 2Band 3RH (%)NormFreq (%)Time (min)()()1,60,70,80,90NknknnPePeRHRHPRHRHk0204060801002.852.92.9533.053.1RH=60%RH=60% (Normal)RH=70%RH=70% (Normal)RH=80%RH=80% (Normal)RH=90%RH=90% (Normal)PdFFrequency (GHz)010203040505.555.65.655.75.755.85.855.9RH=60%RH=60% (Normal)RH=70%RH=70% (Normal)RH=80%RH=80% (Normal)RH=90%RH=90% (Normal)PdFFrequency (GHz)02040606.856.96.9577.057.17.157.27.25RH=60%RH=60% (Normal)RH=70%RH=70% (Normal)RH=80% RH=80% (Normal)RH=90%RH=90% (Normal)PdFFrequency (GHz)
>IEEE SENSORS JOURNAL SUBMITTED<
7
Table I - Statistical analysis of the performance of the climate chamber.
VI. CONCLUSION
RH=60%
GHz
max GHz
GHz
RH=70%
max GHz
GHz
RH=80%
max GHz
GHz
RH=90%
maxGHz
Band 1
3.047
1.83e-5
0.010
3.020
2.28e-5
0.014
2.977
1.37e-4
0.028
2.935
1.63e-4
0.028
Band 2
5.831
1.50e-4
0.024
5.780
7.69e-5
0.026
5.724
8.69e-5
0.029
5.661
3.95e-4
0.053
Band 3
7.186
5.02e-5
0.023
7.132
1.49e-4
0.036
7.036
6.32e-4
0.064
6.942
13e-4
0.074
Consequently, in order to evaluate the P(e), the following
conditional probabilities are needed:
(4)
Once the PdF shown in Fig. 11 are known, it is possible to
calculate the error probability P(e) in the estimation of RH level
with a single interrogation of the tag for each working
frequency band (Table II). The overall error probability of the
sensor could be further reduced with a thorough processing of
the received signal. This approach is widely used in Multiple
Input Multiple Output (MIMO) applications in which the
presence of multiple transmitting and receiving antennas allows
the reliability of the system to be improved. In these
applications, the spatial diversity provides a redundancy of the
received information, which can be exploited by techniques
such as the Maximum Ratio Combining (MRC) [30] to reduce
the error probability of the MIMO link. It is important to remark
that in the case of the presented sensor, the redundancy of the
information is provided by the frequency diversity.
Fig. 12 – Graphical representation of the optimal thresholds for the MAP
decision criterion in the case of Gaussian distribution.
Table II – P(e) in the estimation of RH level with a single interrogation of the
tag for each working band.
P(e)
Band 1
0.031
Band 2
0.019
Band 3
0.042
A novel chipless RFID humidity sensor based on an Artificial
Impedance Surface chipless tag has been presented. The tag
comprises a cardboard substrate and a thin sheet of Mitsubishi paper
with the FSS resonators printed on top. The conductive ink has been
printed with a commercial piezoelectric printer without any curing
or sintering of the ink thus obtaining a very fast and cost effective
printing process. In order to correlate the RH level of the external
environment, a high number of measurements of the sensor have
been performed in a Matlab-driven climatic chamber designed
specifically. The estimation of the RH level of the environment is
based on the normalized frequency shift of the AIS resonance
peaks. The results have shown a high sensitivity of the chipless tag
to the small variation of the RH level. Indeed, changing the relative
humidity level from 50% to 90%, the frequency shift has proven to
be up to 270 MHz. In addition, the statistical analysis of the
robustness and reliability of the proposed chipless RFID tag when
it is employed as a threshold sensor has been presented. The
statistical analysis performed on the collected data showed that the
tag exhibits a low error probability when employed as a threshold
sensor.
VII. REFERENCES
[1] Z. Chen and C. Lu, "Humidity Sensors: A Review of Materials and
Mechanisms," Sensor Letters, vol. 3, no. 4, pp. 274–295, Dec. 2005.
[2] C.-Y. Lee and G.-B. Lee, "Humidity Sensors: A Review," Sensor Letters,
vol. 3, no. 1, pp. 1–15, Jan. 2005.
[3] V. Lakafosis, A. Rida, R. Vyas, L. Yang, S. Nikolaou, and M. M. Tentzeris,
"Progress Towards the First Wireless Sensor Networks Consisting of Inkjet-
Printed, Paper-Based RFID-Enabled Sensor Tags," Proceedings of the IEEE,
vol. 9, no. 98, pp. 1601–1609, 2010.
[4] B. Ando and S. Baglio, "Inkjet-printed sensors: a useful approach for low
cost, rapid prototyping," IEEE Instrum. Meas. Mag., vol. 14, no. 5, pp. 36–
40, Oct. 2011.
[5] G. Marrocco, "Pervasive electromagnetics: sensing paradigms by passive
RFID technology," IEEE Wireless Commun., vol. 17, no. 6, pp. 10–17, Dec.
2010.
[6] R. R. Fletcher and N. A. Gershenfeld, "Remotely interrogated temperature
sensors based on magnetic materials," IEEE Trans. Magn., vol. 36, no. 5, pp.
2794–2795, Sep. 2000.
[7] W. Buff, S. Klett, M. Rusko, J. Ehrenpfordt, and M. Goroli, "Passive remote
sensing for temperature and pressure using SAW resonator devices," IEEE
Trans. Ultrason., Ferroelect., Freq. Control, vol. 45, no. 5, pp. 1388–1392,
Sep. 1998.
[8] A. Vena, L. Sydänheimo, M. M. Tentzeris, and L. Ukkonen, "A Fully Inkjet-
Printed Wireless and Chipless Sensor for CO2 and Temperature Detection,"
IEEE Sensors J., vol. 15, no. 1, pp. 89–99, Jan. 2015.
[9] Fujifilm Dimatix , Inc., "www.dimatix.com." accessed: 09-Feb-2017
[10] M. Schueler, C. Mandel, M. Puentes, and R. Jakoby, "Metamaterial
Inspired Microwave Sensors," IEEE Microwave Magazine, vol. 13, no. 2,
pp. 57–68, Mar. 2012.
[11] T. Chen, S. Li, and H. Sun, "Metamaterials Application in Sensing,"
Sensors, vol. 12, no. 3, pp. 2742–2765, Feb. 2012.
[12] S. Shrestha, M. Balachandran, M. Agarwal, V. V. Phoha, and K.
Varahramyan, "A Chipless RFID Sensor System for Cyber Centric
Monitoring Applications," IEEE Transactions on Microwave Theory and
Techniques, vol. 57, no. 5, pp. 1303–1309, May 2009.
[13] R. S. Nair, E. Perret, S. Tedjini, and T. Baron, "A Group-Delay-Based
Chipless RFID Humidity Tag Sensor Using Silicon Nanowires," IEEE
Antennas Wireless Propag. Lett., vol. 12, pp. 729–732, 2013.
[14] E. M. Amin and N. C. Karmakar, "Development of a low cost printable
humidity sensor for chipless RFID technology," in 2012 IEEE RFID-TA,
2012, pp. 165–170.
[15] Y. Feng, L. Xie, Q. Chen, and L.-R. Zheng, "Low-Cost Printed Chipless
RFID Humidity Sensor Tag for Intelligent Packaging," IEEE Sensors J.,
vol. 15, no. 6, pp. 3201–3208, Jun. 2015.
32312160%60%70%70%70%80%80%80%90%90%PeRHfxRHdxPeRHfxRHdxfxRHdxPeRHfxRHdxfxRHdxPeRHfxRHdx >IEEE SENSORS JOURNAL SUBMITTED<
8
[16] N. Javed, A. Habib, Y. Amin, J. Loo, A. Akram, and H. Tenhunen,
"Directly Printable Moisture Sensor Tag for Intelligent Packaging," IEEE
Sensors J., vol. 16, no. 16, pp. 6147–6148, Aug. 2016.
[17] E. M. Amin, J. K. Saha, and N. C. Karmakar, "Smart Sensing Materials for
Low-Cost Chipless RFID Sensor," IEEE Sensors J., vol. 14, no. 7, pp.
2198–2207, Jul. 2014.
[18] A. Vena, E. Perret, D. Kaddour, and T. Baron, "Toward a Reliable Chipless
RFID Humidity Sensor Tag Based on Silicon Nanowires," IEEE Trans.
Microw. Theory Techn., vol. 64, no. 9, pp. 2977–2985, Sep. 2016.
[19] H. Andersson, A. Manuilskiy, T. Unander, C. Lidenmark, S. Forsberg, and
H. E. Nilsson, "Inkjet Printed Silver Nanoparticle Humidity Sensor With
Memory Effect on Paper," IEEE Sensors J., vol. 12, no. 6, pp. 1901–1905,
Jun. 2012.
[20] J. G. D. Hester and M. M. Tentzeris, "Inkjet-Printed Flexible mm-Wave
Van-Atta Reflectarrays: A Solution for Ultralong-Range Dense Multitag
and Multisensing Chipless RFID Implementations for IoT Smart Skins,"
IEEE Trans. Microw. Theory Techn., vol. 64, no. 12, pp. 4763–4773, Dec.
2016.
[21] S. Kim et al., "Inkjet-printed antennas, sensors and circuits on paper
substrate," Antennas Propagation IET Microwaves, vol. 7, no. 10, pp. 858–
868, Jul. 2013.
[22] F. Molina-Lopez, D. Briand, and N. F. de Rooij, "All additive inkjet printed
humidity sensors on plastic substrate," Sensors and Actuators B: Chemical,
vol. 166–167, pp. 212–222, Maggio 2012.
[23] J. Virtanen, L. Ukkonen, T. Bjorninen, A. Z. Elsherbeni, and L.
Sydänheimo, "Inkjet-Printed Humidity Sensor for Passive UHF RFID
Systems," IEEE Trans. Instrum. Meas., vol. 60, no. 8, pp. 2768–2777, Aug.
2011.
[24] http://www.mitsubishiimaging.com/digital-imaging-diamond-jet-
NANOINK.html." [Accessed: 07-Feb-2017].
[25] F. Costa, S. Genovesi, and A. Monorchio, "A Chipless RFID Based on
IEEE Transactions on
Multiresonant High-Impedance Surfaces,"
Microwave Theory and Techniques, vol. 61, no. 1, pp. 146–153, Jan. 2013.
[26] R. Sauleau, "Fabry–Perot Resonators," in Encyclopedia of RF and
Microwave Engineering, John Wiley & Sons, Inc., 2005.
[27] D. M. Pozar, Microwave engineering. John Wiley & Sons, 2009.
[28] F. Costa, A. Monorchio, and G. Manara, "An equivalent-circuit modeling
of high impedance surfaces employing arbitrarily shaped FSS," in ICEAA,
2009, pp. 852–855.
[29] http://www.k-mpm.com/agnanoen/agnano_ink.html. [Accessed: 09-Feb-
2017].
[30] T. K. Y. Lo, "Maximum ratio transmission," IEEE Transactions on
Commun., vol. 47, no. 10, pp. 1458–1461, Oct. 1999.
Michele Borgese received B.E. and M.E. degrees in
telecommunications engineering from the University of
Pisa, Italy, in 2010 and 2013, respectively. In 2017, he
received the PhD degree in Ingegneria dell'Informazione
from the University of Pisa where he is currently a Post-
Doctoral Researcher at the Microwave and Radiation
Laboratory. Currently, he is involved in the design of
multiband reflactarrays and chipless RFID sensors.
Francesco Alessio Dicandia received the Bachelor's and
Master's degrees in telecommunications engineering
from the University of Pisa, Pisa, Italy, in 2012 and 2014,
respectively, where he is currently pursuing the Ph.D.
degree. His
include
reconfigurable antennas, multiple-input and multiple-
output antennas, non-Foster matching network, and
characteristic modes analysis.
interests
current
research
received
the M.Sc. degree
Filippo Costa
in
telecommunication engineering and the Ph.D. degree in
applied electromagnetism from the University of Pisa,
Pisa, Italy, in 2006 and 2010, respectively. In 2009, he
was a Visiting Researcher at the Department of Radio
Science and Engineering, Helsinki University of
Technology, TKK (now Aalto University), Finland.
During the period 2015-2017, he was several times a
short-term Visiting Researcher at Grenoble Institute of
Technology, Valance, France and at University Rovira I
Virgili, Tarragona, Spain.
He is currently an Assistant Professor at the University of Pisa. His research
interests include metamaterials, metasurfaces, antennas and Radio Frequency
Identification (RFID).
He serves as an Associate Editor of the IEEE SENSORS LETTERS. In 2015
and 2016, he was appointed as outstanding
IEEE
TRANSACTIONS ON ANTENNAS AND PROPAGATION. He was recipient
of the Young Scientist Award of the URSI International Symposium on
Electromagnetic Theory, URSI General Assembly and URSI AT-RASC in
2013, 2014 and 2015, respectively.
reviewer of
Simone Genovesi (S'99-M'07) received the Laurea
degree in telecommunication engineering and the
Ph.D. degree in information engineering from the
University of Pisa, Pisa, Italy, in 2003 and 2007,
respectively. Since 2003 he has been collaborating with
the Electromagnetic Communication Laboratory,
Pennsylvania State University (Penn State), University
Park. From 2004 to 2006 he has been a research
associate at the ISTI institute of the National Research
Council of Italy (ISTI-CNR) in Pisa. From 2010 to
2012 he has been also a postdoctoral research associate at the Institute for
Microelectronics and Microsystems of the National Research Council of Italy
(IMM-CNR).
He is currently an Assistant Professor at the Microwave and Radiation
Laboratory, University of Pisa. He is also a member of RaSS National
Laboratory of CNIT
le
Telecomunicazioni). Current research topics focus on metamaterials, radio
frequency
identification (RFID) systems, optimization algorithms and
reconfigurable antennas.
He was the recipient of a grant from the Massachusetts Institute of Technology
in the framework of the MIT International Science and Technology Initiatives
(MISTI).
(Consorzio Nazionale
Interuniversitario per
the Bachelor Program
Giuliano Manara received the Laurea (Doctor)
degree in electronics engineering (summa cum laude)
from the University of Florence, Italy, in 1979.
Currently, he is a Professor of the University of Pisa,
Italy. From 2000 to 2010 and since 2013, he has been
serving as the President of the Bachelor and the Master
Programs in Telecommunications Engineering at the
same University. Since 2010, he has been serving as
the President of
in
Telecommunications Engineering at the Italian Navy
Accademy in Livorno, Italy. Since 1980, he has been collaborating with the
Department of Electrical Engineering of the Ohio State University, Columbus,
Ohio, USA, where, in the summer and fall of 1987, he was involved in research
at the ElectroScience Laboratory.
His research interests have centered mainly on the asymptotic solution of
radiation and scattering problems. He has also been engaged in research on
numerical, analytical and hybrid techniques, frequency selective surfaces (FSS)
and electromagnetic compatibility. More recently, his research has also been
focused on antenna design and on Radio Frequency Identification (RFID).
Prof. Manara was elected an IEEE (Institute of Electrical and Electronic
Engineers) Fellow in 2004 for "contributions to the uniform geometrical theory
of diffraction and its applications.". He served as the International Chair of
URSI Commission B for the triennium 2011-2014. Prof. Manara has served as
the General Chair of the International Symposium on Electromagnetic Theory
(EMTS 2013), held in Hiroshima, Japan, May 20-24, 2013.
Prof. Manara
(Consortium UBIquitous
Technologies) S.C.A.R.L., a consortium created by the Dipartimento di
Ingegneria dell'Informazione of the University of Pisa, Polo Navacchio S.p.A.
(Navacchio, Cascina) and some highly innovative Italian companies, with the
aim of defining a new knowledge transfer model from university to industry.
the President of CUBIT
is
|
1705.10275 | 1 | 1705 | 2017-05-26T07:29:25 | Experimental demonstration of an ultra-compact on-chip polarization controlling structure | [
"physics.app-ph"
] | We demonstrated a novel on-chip polarization controlling structure, fabricated by standard 0.18-um foundry technology. It achieved polarization rotation with a size of 0.726 um * 5.27 um and can be easily extended into dynamic polarization controllers. | physics.app-ph | physics | Experimental demonstration of an ultra-compact
on-chip polarization controlling structure
Qingzhong Deng, Lu Liu, Zhiping Zhou*
State Key Laboratory of Advanced Optical Communication Systems and Networks,
School of Electronics Engineering and Computer Science, Peking University, Beijing ,China 100871
*[email protected]
Abstract: We demonstrated a novel on-chip polarization controlling structure, fabricated by
standard 0.18-m foundry technology. It achieved polarization rotation with a size of 0.726 m ×
5.27 m and can be easily extended into dynamic polarization controllers.
OCIS codes: (130.0130) Integrated optics; (130.3120) Integrated optics devices; (230.5440) Polarization-selective devices.
1. Introduction
Silicon on insulator (SOI) is a prevailing platform for its CMOS compatibility and integration compactness.
Simultaneously, strong polarization dependence occurs which makes polarization control essential. A great deal of
effort has been made but on-chip polarization control is still on the way mainly due to the fabrication problems,
either too small features size (60 nm) [1], stringent fabrication accuracy requirement [2, 3], large device size (longer
than 100 m) [4] or incompatibility with CMOS fabrication process [5].
In this paper, we propose and experimentally demonstrate a practical polarization controlling structure with
compact size, wide bandwidth and low insertion loss. Moreover, the structure directly links dynamic polarization
control with phase manipulation and it can be easily fabricated by the mature 0.18-m CMOS technology.
2. Polarization rotation (Static polarization control)
A partially etched slot waveguide is introduced to realize the polarization rotation [Fig. 1(a)]. H=220 nm,
Slot=200 nm and Slab=90 nm are chosen according to the design rules of the fabrication foundry, while W=726 nm
is optimized for polarization rotation. As shown in Fig. 1(b), it supports three eigenmodes. The mode with neff
slot, is quasi-TE polarized since the major transverse fields are Ex and Hy. The other two
=2.222, marked as TE0
slot and EM2
slot, are hybridized in polarizations.
modes, EM1
Fig. 1. (a) Schematic and (b) Eigenmodes of partially etched slot waveguide. All simulations are done by 3D full vector finite-
element- method (FEM) in this paper. The refractive indices of Si and SiO2 are set to 3.48 and 1.45, respectively.
When a fundamental TM mode in strip waveguide (TM0
slot
slot
slot will be excited since it does not support TM0
slot and EM2
strip) is injected into such a partially etched slot
slot and EM2
mode. The effective refractive index
waveguide, EM1
modes will cause beat and W has been optimized to achieve polarization
(neff) difference between EM1
, as shown in Fig. 2(a). Then, two strip waveguides are utilized to
rotation at half-beat length
strip is
symmetrically decouple the light into two separate TE0
strip modes with
injected, it will be converted to TE0
a phase difference of This polarization rotation scheme has been verified with the measured optical images in Fig.
strip into this structure simultaneously, interference occurs in the two output ports and a
2(b). Inputting TE0
strip
wavelength shift of FSR/2 is observed [Fig. 2(c)]. It can be explained as: in a reference waveguide (WGref), TE0
strip modes will not interfere with each other since they are orthogonal. However, if the injected TM0
strip
and TM0
strip mode, interference
mode gradually transforms into TE0
will take place and result in two complementary output transmission spectra. From this perspective, the observed
interference and wavelength shift are also strong evidences of polarization rotation.
slot due to the field similarity and decoupled into two separate TE0
strip mode and it has a phase delay with the original TE0
strip modes with a phase difference of . When TE0
strip and TM0
Separating the incident optical power in TE0
strip modes is always wanted. A polarization rotator and
splitter (PRS) is proposed by attaching a π/2 phase-shifter and a directional coupler (DC) [Fig. 3(a)]. The measured
results in Fig. 3(b) and 3(c) indicates that the insertion loss (IL) is -0.90 dB (-2.28 dB) and the extinction ratio (ER)
strip) incidence. The insets in Fig. 3(b) and 3(c)
is 14.56 dB (11.77 dB) at 1550 nm wavelength for TE0
indicates that the output light is dominated by TE polarization but have not been fully coupled into one waveguide.
Therefore, the performance will be better if the DC is optimized properly. Moreover, simulations proved that using
strip and TM0
strip (TM0
/(2)5.27 μmeffLn strip-slot mode converter [6] can further elevate the performance: the IL is -0.7 dB (-0.9 dB) and the ER is 22 dB
(34 dB) for TE0
strip) incidence.
strip (TM0
Fig. 2. (a) Optical field evolution (Hy) and Micro micrographs of the proposed polarization rotator (PR); (b) optical images captured from the
output waveguides; (c) The measured transmission spectra for hybrid TE0 and TM0 incidence.
Fig. 3. (a) Optical field evolution (Hy) and Micro micrographs of the proposed polarization rotator and splitter (PRS); the measured transmission
spectra and optical images (Insets): (b) TM0 incident and (c) TE0 incident.
3. Dynamic polarization control
As discussed above, the proposed polarization rotation structure associates the phase difference with polarization
state: TE0 corresponds to two light beams with phase difference of 0 and TM0 corresponds to π. Here we show an
example of dynamic polarization controllers by integrating the structure with phase tuning elements (Fig. 4).
Fig. 4. Schematics of dynamic polarization controller.
4. Conclusion
We have proposed and designed an untra-compact polarization rotation structure using partially etched slot
waveguide which is just 0.726 m × 5.27 m in size and can be easily fabricated. The function of rotation is
experimentally verified with a PR and PRS which are fabricated by standard 0.18-m foundry technology.
Moreover, the structure associates polarization states with phase difference which enables on-chip dynamic
polarization control by mature phase manipulation in silicon photonics.
This work was supported by the Natural Science Foundation of China (Grant No. 61120106012).
Reference:
[1] A. V. Velasco, M. L. Calvo, and P. Cheben, et. al., "Ultracompact
polarization converter with a dual subwavelength trench built in a
silicon-on-insulator waveguide," Opt. Lett. 37, 365 (2012).
[2] L. Liu, Y. Ding, K. Yvind, and J. Hvam, "Efficient and compact
TE-TM polarization converter built on silicon-on-insulator platform
with a simple fabrication process," Opt. Lett. 36, 1059-1061 (2011).
[3] L. Gao, Y. Huo, and K. Zang, et. al., "On-chip plasmonic
waveguide optical waveplate," Sci. Rep.-UK 5, 15794 (2015).
[4] W. D. Sacher, T. Barwicz, B. J. F. Taylor, and J. K. S. Poon,
"Polarization rotator-splitters in standard active silicon photonics
platforms," Opt. Express 22, 3777-3786 (2014).
[5] H. Deng, D. Yevick, C. Brooks, et. al., "Design Rules for Slanted-
Angle Polarization Rotators," J. Lightwave Technol. 23, 432 (2005).
[6] Q. Deng, Q. Yan, and L. Liu, et. al., "Robust polarization-
insensitive strip-slot waveguide mode converter based on symmetric
multimode interference," Opt. Express 24, 7347-7355 (2016).
|
1712.05436 | 1 | 1712 | 2017-12-14T20:08:41 | Real Time Visualization of Dynamic Magnetic Fields with a Nanomagnetic FerroLens | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Due to advancements in nanomagnetism and latest nanomagnetic materials and devices, a new potential field has been opened up for research and applications which was not possible before. We herein propose a new research field and application for nanomagnetism for the visualization of dynamic magnetic fields in real-time. In short, Nano Magnetic Vision. A new methodology, technique and apparatus were invented and prototyped in order to demonstrate and test this new application. As an application example the visualization of the dynamic magnetic field on a transmitting antenna was chosen. Never seen before high-resolution, photos and real-time color video revealing the actual dynamic magnetic field inside a transmitting radio antenna rod has been captured for the first time. The antenna rod is fed with six hundred volts, orthogonal pulses. This unipolar signal is in the very low frequency (i.e. VLF) range. The signal combined with an extremely short electrical length of the rod, ensures the generation of a relatively strong fluctuating magnetic field, analogue to the signal transmitted, along and inside the antenna. This field is induced into a ferrolens and becomes visible in real-time within the normal human eyes frequency spectrum. The name we have given to the new observation apparatus is, SPIONs Superparamagnetic Ferrolens Microscope (SSFM), a powerful passive scientific observation tool with many other potential applications in the near future. | physics.app-ph | physics | MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000
Real Time Visualization of Dynamic Magnetic Fields with a Nanomagnetic FerroLens
Emmanouil Markoulakisa*, Iraklis Rigakisa, John Chatzakisa, Antonios Konstantarasa, Emmanuel
Antonidakisa
aTechnological Educational Institute of Crete, ComputerTechnology, Informatics & Electronic Devices Laboratory, Romanou 3, Chania, 73133, Greece
A R T I C L E I N F O
A B S T R A C T
Article history:
Received 05 June 2017 initial
submission MAGMA_2017_1507.
Invited Resubmission from 23 June
2017 MAGMA_2017_1658.
Revised submission at 05/Dec/2017.
Accepted 00 xx 00
Keywords:
Ferrofluid film
Lens
Visualization
Dynamic Magnetic fields
Superparamagnetism
Quantum effects
Highlights:
Passive Visualization
Method of Radio Magnetic
Waves
Prototype Microscope
Apparatus using a
nanomagnetic optical lens
Experimental Results and
Theoretical Analysis
1. Introduction
Fig.1 First Time Recorded in History EM Radio Wave inside Antenna (Video Link)
4
Due to advancements in nanomagnetism and latest nanomagnetic materials and devices , a new potential
field has been opened up for research and applications which was not possible before. We herein propose
a new research field and application for nanomagnetism for the visualization of dynamic magnetic fields
in real-time. In short, Nano Magnetic Vision. A new methodology, technique and apparatus were
invented and prototyped in order to demonstrate and test this new application. As an application example
the visualization of the dynamic magnetic field on a transmitting antenna was chosen. Never seen before
high-resolution, photos and real-time color video (Fig.1) revealing the actual dynamic magnetic field
inside a transmitting radio antenna rod has been captured for the first time. The antenna rod is fed
with six hundred volts, orthogonal pulses. This unipolar signal is in the very low frequency (i.e.
VLF) range. The signal combined with an extremely short electrical length of the rod, ensures the
generation of a relatively strong fluctuating magnetic field, analogue to the signal transmitted, along and
inside the antenna. This field is induced into a ferrolens and becomes visible in real-time within the
normal human eyes frequency spectrum. The name we have given to the new observation apparatus is,
SPIONs Superparamagnetic Ferrolens Microscope (SSFM), a powerful passive scientific observation
tool with many other potential applications in the near future.
© 2017 xxxxxxxx. Hosting by Elsevier B.V. All rights reserved.
migrating birds [1] use, as researchers have indicated lately, that these
birds may actually see the Earth's magnetic field in the sky propagating
from geographic north, since
their eyes act similarly
to nano
ferrohydrodynamic lenses. This magnetic sight apparatus, because our
brain and eyes are not well suited for processing this kind of information,
must be able to transform in real-time the magnetic flux field into an
observable EM spectrum and correctly and adequately visualize all of its
dynamic ( i.e. intensity, spatial and temporal) characteristics.
Ever wondered what we would see if we could actually look with the
naked eye at the dynamic magnetic field, part of the electromagnetic (EM)
wave signal, on a transmitting radio antenna? In order to achieve this, we
would have to invent an artificial magnetic sight apparatus and
methodology that would enable us to observe the magnetic flux stream of
the field under investigation. Similar to the natural mechanism the
* Corresponding author. Tel.:+30-28210-23035
E-mail address: [email protected]
Peer review under responsibility of xxxxx.
Hosting by Elsevier B.V. All rights reserved.
doi: https://doi.org/10.1016/j.jmmm.2017.12.023
Hyperlinks:
1. http://www.ferrocell.us/intro.html
2. http://www.ferrocell.us/references.html
3. N42 Ring Magnet Data sheet:
https://www.supermagnete.de/eng/data_sheet_R-27-16-05-N.pdf
4. New Discovery: First Time Recorded in History EM Radio Wave Video:
https://www.youtube.com/watch?v=fGcvh4Rb0G4
5. Side by Side Comparison MIT Motion Amplification Algorithm video:
https://www.youtube.com/watch?v=enTUapltGsA
6. Ferrofluid data sheet:
https://ferrofluid.ferrotec.com/products/ferrofluid-educational-fluid/efh/efh1/
doi: https://doi.org/10.1016/j.jmmm.2017.12.023
2
MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000
2. Materials and Methods
2.1. Basics
The ferrolens we use in our invention proposal is Timm Vandereli's
patented and trademarked FerrocellTM1,2. The basic operation principle of
the ferrolens is illustrated in fig.2a,b,c,d. This nano Ferrohydrodynamic
[2] Lens consists of a special mixture made utilizing, SPIONs [3] (i.e.
superparamagnetic Iron Oxide Nanoparticles), to create a ferrofluid lens,
a 62mm in diameter by 2mm thick, optical – magnetic device. Between
the two 1mm thick round optical grade glass disks, is a 50μm thin active
layer of superparamagnetic [4] ferrofluid. The glass and ferrofluid inside
are sealed with an optical cement around the periphery to prevent the
mixture from drying out, making up the final Ferrocell lens. See fig3.a
(orange disk).
transparent and practically invisible i.e. invisibility cloak using nano
ferrofluid [8]. The color of the magnetic flux lines depends on the emitted
light spectrum from the light source we use and the strength of the
magnetic field. Generally, intense magnetic flux lines tend to attract more
SPIONs creating thicker dipolar chains which absorb colors in the red
and yellow end of the spectrum, making them to appear blue-green and
vice versa, smaller-sized SPION chains absorb blues and greens, resulting
in a red appearance. Also, most importantly, absence of magnetic flux in a
region of the magnetic field is depicted and displayed by the ferrolens as a
dark region.
2.2 Dynamic Magnetic Field Operation
In the previous paragraphs we covered the basic operation of the
superparamagnetic ferrolens under static conditions, whereas permanent
magnets are used to induce a magnetic field, we then observed using the
ferrolens and experimentally proved that the nano-sized SPIONs can
effectively visualize the intensity and spatial information of magnetic
fields, so far so good. But, what will the results be when we try to
observe a dynamic magnetic field, such as, the fluctuating magnetic
field on a transmitting radio antenna. This has never been attempted
before. Can the ferrolens adequately visualize and give us in real-time the
temporal information of a dynamic, fluctuating, magnetic field?
For this lab experiment, a special prototype apparatus has been invented
namely the SPIONs Superparamagnetic Ferrolens Microscope in short
SSFM© described in fig3.(a)&(b)
Figure 2 (a) eye looking through the Superparamagnetic 50μm thin
SPIONs layer, placed inside the ferrolens, toward magnetic field of
antenna (b) electronic microscope picture of the 10 nanometer (nm) in
size SPIONs forming dipole chains, times 400 enlarged. SPIONs absorb
light in the visual spectrum thus are appearing colored and due to their
superparamagnetic nature align with the magnetic flux lines of the
antenna field effectively bending light and aligning it with the flux lines
of the magnetic field therefore making it visible to the naked eye. (c)(d)
Static magnetic field visualizations of permanent neodymium magnets as
shown by the Ferrocell (i.e. ferrolens).
……………………………………………………………………………
In both cases, fig.2 (c)&(d), a light source is placed under the ferrolens,
(c) a RGB LED strip peripherally to the lens, (d) a small incandescent
lamp. The ferrolens apparatus is then placed on top of the magnetic object
its field we like to observe, in the case of fig.2(c)(d) the magnetostatic
field of neodymium permanent magnets, a cube magnet (c) and (d) a
cylindrical magnet. Because the 10 nanometers (nm) average in size
SPIONs of the ferrofluid superparamagnetic properties i.e. dipolar chains
aligning to the magnetic field flux lines [5] and also their superabsorbing
light properties [6,7], the magnetic field becomes visible. Notice as well,
the object itself under magnetic viewing due to light bending becomes
Figure 3 (a) column Programmable high voltage Pulse generator with the
ferrolens assembly sitting on top and switched off. Orange-colored
ferrolens placed above a 3D printer constructed lighting cone with a 24
RGB LED strip emitting white light attached to its inner bottom periphery
controlled via infrared remote control. Black telescopic whip antenna rod
under investigation, passing through the cone via holes and just 1mm
under the ferrolens without any surface contact (b) column Ferrolens
Microscope switched on and ready for use. An adjustable focus
magnifying lens apparatus is placed on the top surface of the ferrolens
with an N42 neodymium ring magnet in between, fastened under the
magnifier and facing down the ferrolens center looking at the antenna rod
behind the ferrolens.
To reach the sensitivity threshold and improve responsiveness to
magnetic field fluctuations of the SSFM Microscope, a very electrically
MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000 3
small [9], antenna rod is feed with a high voltage orthogonal unipolar 600
volts RMS signal using a programmable pulse generator fig.3(a)(b).
Thus,
(1)
Ideally from (1) above, we calculate for 600 volts RMS, Vp = 900 volts
with a signal duty cycle equal to D = 0.5 with D = t1/T . Additionally, in
order to overcome mass inertia on the SPIONs in the ferrofluid, thus,
initially excite the ferrolens to improve sensitivity and further increase
responsiveness of the superparamagnetic microscope, a N42 neodymium
ring magnet is attached at the center of the ferrolens, top view,
sandwiched between the ferrolens and the x2.4 magnifying lens sitting on
top of the microscope, fig. 3(b) . Dimensions of the ring magnet3 used,
and fastened under the magnifying lens apparatus with exact axial
alignment, are 26.75mm outer diameter x16mm inner diameter x5mm
height, rated at Br = 1.3 Tesla of residual magnetism. Therefore, the
dynamic magnetic field strength generated by the signal applied at the
transmitting antenna rod, is fluctuating at the microscope region on top of
the static field induced by the ring magnet and is direct analogue to the
signal and electromagnetic wave transmitted by the antenna. Later on, we
will also discuss the effect of the period T of the signal applied, thus, the
effect of the signal's fundamental frequency fs on the SSFM microscope's
sensitivity and responsiveness and prove the main criterion for choosing
the operation frequency range of the proposed SSFM, nanotechnology
SPIONs Superparamagnetic Ferrolens Microscope.
3. Results
3.1 A 'MRI' Scan of an Antenna
Microscopic Photography and video data
information evidence resembling a MRI
scan, were collected in real-time, of the
actual magnetic field inside and around a
transmitting antenna rod, i.e. whip antenna.
A small sample of it is shown in fig.4 and
fig.5(a)&(b) . All the recorded visual data
evidence shown here, were taken looking
through the 16mm inner diameter of the ring
magnet attached above the ferrolens at its
center with
the aid of an additional
magnification optical lens with adjustable
focus and x2.4 zoom in. The telescopic
whip antenna we use, has a maximum of
8mm thickness at the lower end, is made out
is coated with black
of bronze and
under
Figure 4 Antenna rod is
transparent
the
powered up ferrolens (A
LED strip lighting is under
the antenna rod)
ferromagnetic metallic paint. The later is very essential for the operation
of the SSFM microscope since bronze is a diamagnetic [10] material
and it would not be possible to look at the existing magnetic field inside
the antenna rod unless it was coated with a ferromagnetic paint. Thus,
the antenna rod body becomes transparent (i.e looks like glass) while
under magnetic viewing under the ferrolens shown at fig.4 and only the
magnetic field inside and around of transmitting antenna is observed. In
photographs fig.4 & fig.5 we can clearly observe (located at the center of
the antenna body, in the middle), a black strip of accumulated unpolarized
SPIONs nanoparticles population surrounding the Bloch domain wall
region of the antenna's magnetic dipole field which is located across the
width of the rod, and with the Bloch region [11] squeezed in the middle.
A region where the magnetic field strength drops from nearly zero to zero
across the few atoms thick Bloch region (shown also previously in fig2.d),
a quantum effect shown by the ferrolens at microscopic level. Therefore
we specifically can say that the observed black strip shown by the
ferrolens of a transmitting antenna is effectively, the part of its magnetic
field centered around the Bloch region. This is a region with zero
magnetism (as discussed above, ferrolens colors zero magnetism as
black). This is a novel scientific observation and proof of its existence
in antennas. To our knowledge this effect has never been reported or
shown before, hereby demonstrating the potential capability of the
SSFM microscope.
Figure 5. (a)(b) Transmitting antenna rod under the ferrolens. Field
inside (red glow with black wall Bloch region) and outside the antenna
rod (black and white magnetic field rays). A 'MRI' of an Antenna Rod.
(c)(d) Fields at the thinner part of the whip (rod) antenna and the tip.
……………………………………………………………………………
Continuing our analysis and observation of the field, in fig.5 (a)&(b) we
observe two red-orange glowing halo bands, left and right of the centre
part of the magnetic field (i.e. black strip in the middle) running along the
total length of the antenna rod as we have observed. These reddish bands
are more apparent when shown in fig.5 (c)&(d) as two reddish lines
running in parallel along the antenna rod. These regions are in fact the
dipolar magnetic , North-South pole, fields of the transmitting antenna.
Notice also here the 'black sun' effect, distinct black finger rays separated
by faint white rays, popping out from all directions around the antenna rod
body, like the classical iron fillings experiment, specially noticeable
around the periphery, these are the radiation flux lines of the antenna
contained inside our ring magnet viewing window. Specially, fig.5(d) is
an amazing, first time, photograph showing off the antenna's tip
magnetic dipolar field appearing like an upside down U-shaped
horseshoe with a radiating magnetic black sun at the background.
The whole antenna structure resembles a magnetically vibrating tuning
fork.
doi: https://doi.org/10.1016/j.jmmm.2017.12.023
4
MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000
3.2 A Dissection Analysis of the Field
Figure 6 Graphical analysis of the field
In Figure 6(a) the classical graphical representation of Electromagnetic
(EM) wave propagation [12] in space is shown, where S is the directional
propagation Poynting vector of the wave and B and E the amplitude
vectors of the wave's magnetic and electric fields respectively. Figure
6(b) shows a superimposed graphics photograph version we have taken
using the SSFM microscope of the magnetic field of a transmitting
telescopic antenna rod (i.e. whip). By comparison of (a) and (b) where
the antenna rod length axis is pointing to the wave's propagation S
Poynting vector through space, we clearly can distinguish and identify in
fig.6(b) the red halo bands above and below as the magnetic field, part of
the Electromagnetic wave transmitted by the antenna, perpendicular to the
EM wave propagation axis. To demonstrate this effect more vividly
imagine the whole antenna rod length (e.g. about 1 meter) as a giant bar
magnet with very large elongated North and South poles, similar to
placing side by side in parallel multiple individual smaller magnets
adding their magnetic flux lines coming from their North and South
poles to form altogether the large in size magnetic dipole field of the
transmitting, radiating antenna. Interestingly also, Fig6.(b) shows how all
this Bloch wall regions of the many individual small magnets add up to
form the total Bloch region of the magnetic field (i.e. black strip in the
middle) of the antenna rod. Furthermore observation upon fig6(b), reveals
that the red flux lines comprising the magnetic field appear more faint at
the centre of our viewing window of the microscope than they appear at
the periphery where magnetic field looks more intense. This is due the
interaction of the field of the antenna with the static field of the SSFM
microscope's ring magnet we used. The dynamic magnetic dipole field of
the transmitting antenna is contained inside the static field of the ring
magnet of the microscope, and gradually spatially squeezes the field of
the antenna more and more until the antenna's flux becomes V-shaped
close to the ring magnet where interaction is stronger fig6(b). Normally,
without the microscope apparatus the magnetic dipole field of the antenna
is extended outside its physical dimensions.
3.3 Dynamic Video Analysis of the Field and Motion Amplification
Untill now in this paper we have presented a static analysis, by means of
microscopic photography to record the observations we obtained through
the ferrolens, of the composite dynamic magnetic field of a transmitting
antenna rod. However, in order to experimentally prove the SSFM
microscope can efficiently display the dynamical temporal information
data of the antenna field, a dynamic visual information time recording
method was needed, thus, real-time high definition video recording is
used. We succeeded to record the first time ever video of the
electromagnetic wave footprint on a transmitting antenna's dipolar
magnetic field, Fig.1 Video Link4 . In short, a magnetic wave fig 6(b).
The same result came out from all of our recordings. All of the original
videos were recorder at high-definition 1080p without any video
artifacts present. Although the SSFM microscope cannot visualize and
display electric field vectors but only the magnetic part, it succeeded to
register and display in real-time an EM wave signature of the signal
transmitted by the antenna. Specifically, a periodic sinusoidal transverse
ripple effect was observed upon the surface areas (i.e. up and down
surface) of the antenna's magnetic field centre (i.e. black strip in the
middle of antenna), projected clearly by its dynamic dipolar field (i.e. red
halo surrounding the centre field) due to mainly its envelope [13]
amplitude fluctuations. Manifested in the ferrofluid, we speculate, as
quantum effect phonons [14] or else called as ripplons by Rosensweig [15]
et al. or Ferrofluid flows in AC and traveling wave magnetic fields by
Markus Zahn [16] et al. and warrants further investigation. These ripples
were constantly present along the whole length of the antenna rod with a
frequency varying as measured from 1 to 5 Hertz approximately, Fig.1
Video Link4, fig.6(b) and fig.7.
Figure 7 Animated gif of ripple effect (click)
https://drive.google.com/file/d/0B0A8uTBvEiQRTUlRYUFVckZRRHc/view
MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000 5
At first, this ripple motion of the field (because its microscopic nature),
could not be detected by naked eye although there was a hint that
something was in motion. Things changed dramatically after we
processed the videos with MIT Motion Amplification [17] Algorithm,
using the Eulerian amplification method. Motion amplification was set to
x20 at the frequency range 0.1 to 5 Hertz. A high definition side by side
before and after the algorithm was applied, was also
comparison video
5
produced.
4. Discussion
At the end of section 2.2 of the paper, we have referred to the limitations
in the operation of the ferrolens apparatus due to the maximum signal
frequency fs at which the induced magnetic field, can be successfully
displayed by the ferrolens, in phase (i.e. in real-time) with the applied
signal. Thus, beyond and above this specific maximum frequency of
operation, the ferrolens visual information display becomes progressively
out of phase with the signal under observation, until for relative high field
frequencies the SPIONs in the ferrofluid become totally unresponsive to
the field fluctuations. Simply put, the SPIONs in the ferrofluid cannot
keep up with the speed of the amplitude fluctuations of rapidly-changing
dynamic magnetic field. A criterion must be established for choosing
the maximum safe operational frequency for the ferrolens fsmax .
Rosensweig et al. [2,18–20] explains that the two dominant mechanisms
through which the ferrofluid particle magnetic moment (i.e. magnetic
dipole movement ) may align with the applied magnetic field are
Brownian motion, which is the physical rotation of the particle into
alignment with the field, and Néel relaxation, which is characterized by
the movement of the particle magnetic moment relative to the crystal
structure axis of the ferrofluid ferromagnetic particles.
Relaxation times for each are,
(2)
(3)
for which ηo , fo , and K are the carrier fluid dynamic viscosity, the
frequency constant of Néel relaxation, and the anisotropy constant of the
particle, respectively. The two particle volumes VB and VN are given by,
(4)
(5)
In (4) δ represents the thickness of the adsorbed surfactant layer, and R =
d/2 is the magnetic particle radius. The relaxation times defined in (2) and
(3) are typically on the order of hundreds of milliseconds to nanoseconds
[2,18]. The effective relaxation time for ferrofluid particles can be derived
by considering
that both the Brownian and Néel processes act
simultaneously. When both mechanisms play a role in the relaxation
process the effective time constant is,
(6)
A plot of the three relaxation times as a function of particle diameter
[16,19] is shown in Figure 8, which indicates that the smallest time
constant dominates the physical process of relaxation. For small particles
Néel relaxation is faster than Brownian, and so the Néel time constant
dominates τeff. For large particles the Brownian relaxation is faster than
Néel relaxation, and so the Brownian time constant dominates τeff .
Figure 8 [2,20] The Brownian, Néel, and effective relaxation time constants as
a function of spherical magnetic particle diameter. The plots correspond to
EFH1 hydrocarbon-based ferrofluid, with the corresponding fluid parameters:
the fluid mass density ρ = 1169 kg/m3, the dynamic viscosity ηo = 10 cP, the
anisotropy constant K = 23, 000 J/m for magnetite, the temperature T = 300 K,
and the frequency constant, fo = 109 Hz. The Brownian plot assumes zero
surfactant thickness, δ = 0.
the cell
Our ferrofluid (e.g. data sheet6) inside the ferrolens used in the
experiments, is a type EFH1 hydrocarbon-based ferrofluid with an
average of magnetite SPIONs size, of 10 nanometers, and with very
similar specifications to the ones of the plot in fig.8. Therefore, the plot
in fig.8 is also true for our case and applies directly and we can
securely say that our SSFM microscope can operate up to the tens or
hundreds of Megahertz (MHz) frequency range. Any other movements
of the SPIONs inside the lens due to different phenomena other than the
previously described are ineffective or negligible because the relatively
strong existing Van der Waals force [21] within the ferrolens. The
encapsulated thin layer of ferrofluid inside the ferrolens in this
state, does not flow anymore, but exists in a balanced state of
is oriented. The
equilibrium no matter what position
nanoparticles inside the ferrolens do not settle with gravity. More in
detail, the anionic surfactant coating [22] on the nanoparticles keeps the
particles from touching each other (i.e. clumping or agglomeration) in the
free state when there is no external magnetic field present. Notice here
that the generated Van der Waals forces in the ferrofluid are not attractive
but due to steric repulsion [2], results to stabilization.
Nevertheless, in our proposal we derive a maximum safe operational
frequency fsmax for the SSFM microscope where magnetic susceptibility
x(ω), of the SPIONs nanoparticles of the ferrofluid inside the ferrolens
system, does not start to fall dangerously with frequency, in which case
magnetic moment of the SPIONs, is in phase with the induced external
magnetic field by the signal in order to have a real-time display of the
field . The later dictates us that we are interested in finding a maximum
frequency where the real part (i.e. in phase) of the susceptibility of the
system is stable before it is starting to drop with increasing frequency.
According to Debye theory [23,24] [Debye 1929] for low-field regime
the real and imaginary components of the complex susceptibility
doi: https://doi.org/10.1016/j.jmmm.2017.12.023
6
MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000
are thus given respectively by,
(7)
(8)
where τ is effective relaxation time [24] of the system given by (2),(3)
and (6) and x0 is the initial value [25] of susceptibility of our system given
by, x0 = Msat/(3kBT) the initial value [26] of susceptibility (9). Msat is the
magnetic saturation value of the ferrofluid in A/m units which is given for
equivalently to 35 KA/m units, kB is the
our system, in SI units, at 44 mT
Boltzmann constant 1.38 · 10−23 J/K and T the temperature. For room
temperature at T = 300 Kelvin the initial susceptibility of the ferrofluid of
6
6
our system is given by the manufacturer at x0 = 2.62 in SI units
.
Figure 9 Susceptibility vs. frequency for various values of τeff system
relaxation time. https://www.desmos.com/calculator/ha5n6qafbw
Using the information from (2),(3),(6),(7) equations the information from
the data in fig.8 and the data from the ferrofluid manufacturer
6
we plotted
in fig.9 the frequency dependency of our ferrolens's
magnetic
susceptibility for three worse case scenarios values of τeff system
relaxation times in order to draw our conclusions. From the above plot
and considering the values for real-time operation requirement of the
SSFM microscope (i.e. there is no hysteresis in the display of the dynamic
magnetic field fluctuations on the microscope) and also considering as an
acceptable drop for real-time operation, in the initial maximum value 2.62
(see fig.9) of the susceptibility, a minus 10%, we then calculate, fsmax = 40
KHz maximum safe real-time operational frequency for 1μs (blue)
relaxation time, 120 KHz for 500 ns relaxation (green) and fsmax = 0.5
MHz for 100 ns (red) relaxation time.
However, we must stress here that the above calculations are really
worse case scenarios. In reality 10nm magnetite SPIONs EHF1 type
ferrofluids encapsulated in the ferrolens glass structure (i.e. vacuum
sealed μm thin layer of ferrofluid) are at the 10 ns relaxation time range
[27] and depending on the concentration percentage in the ferrofluid by
the manufacturer, which gives us a safe real-time operating frequency as
calculated
we
5 MHz
https://www.desmos.com/calculator/pv1fzk6qm8
observed
experimentally. Furthermore, when real-time is not a problem and
as we
defined
have
it,
around
and
examining single frequency fields and not composite harmonical rich
fields, with todays high initial [25] susceptibility valued x0 manufactured
ferrofluids, we can happily operate the ferrolens at the tens and even
hundreds of megahertz depending the size of the nanoparticles.
Regarding, increasing the field strength sensitivity of the ferrolens, few
tens [28] of mT static magnetic (see section 2.2) field should be applied
depending
ferrofluid
saturation magnetization value given by
6
(i.e. in our case it was 44mT) before a signal induced field
manufacturer
is applied. In this way we ensure that all nanoparticles in the ferrofluid
are activated and polarized with the external field. Also the signal
amplitude applied should be at the hundreds of volts range with a low
current. In our experiments we used a custom programmable high voltage
pulse generator 600 Volts rms mostly at the VLF frequency range. All
photographs and videos in this paper were taken from a 7 KHz pulsed,
signal induced field. As a suggestion for future experimentation
regarding our subject is the use of a 100K frames per second or more
high speed camera. In this way someone could video record and then
playback in slow motion the full display effect of the on-off field
transitions, induced by the pulsed signal.
5. Conclusion
A new application for nanomagnetism is proposed. A new cost
effective nanotechnology optical-magnetic visualization, passive
technique, methodology and apparatus (i.e. SSFM microscope), was
invented, introduced and demonstrated using the superparamagnetic
properties of a commercially available ferrolens. A micron thin film of
ferrofluid encapsulated in a vacuum inside a lens and applied to a
custom-made optical microscope namely, "SPIONs Superparamagnetic
Ferrolens Microscope" or in short SSFM for the observation and
research of dynamic magnetic fields, which was described and prototyped
in a lab while the operational parameters and limitations were analyzed
and proven theoretically and by experiment. In order to demonstrate our
results we applied the above apparatus and method, for visualizing the
dynamic field inside and around a transmitting Radio Antenna rod and to
our knowledge this has never be done before in this particular way. It
successfully resolved and visualized all the spatial and temporal
information of the dynamic field induced by the transmitter in the radio
antenna signal with sensitivity factor down to nT scale and with real-time
responsiveness up to 5 MHz observed by lab experiments i.e. limitations
of our testing instrumentation prevented us from going higher in
frequency. It was proven that in active thin layer only microns thick films
of ferrofluid when introduced inside the middle of an optical lens and
illuminated by a artificial light source, under the right conditions, have
enhanced superparamagnetic and magnetic viewing properties compared
to the free state (i.e. un-enclosed) ferrofluid, produces detailed visual
information of the field in real-time and in color. Also it has been shown
theoretically that with modern ferrofluid products, when real-time display
is not an issue and for single frequency fields with little to no harmonics,
the ferrolens can be used up to the hundreds of MHz range for SPIONs
nanoparticles less than 10 nm in size. In addition the special optical
properties of this kind of optical magnetic lenses were discussed and an
invisibility cloaking as a side effect was presented (fig.4, fig2c). As a final
remark, we believe that due the rapid scale-down of nanotechnology over
the last years, nanomagnetism [29] will lead to more and more advanced
ferrohydrodynamic products and methods with new potentials that before
were not possible. This enables us to open up the view of our physical
world from the macroscopic into the microscopic quantum effects scale.
Additional
further development and novel
applications of these advanced new nanomagnetic viewing devices.
research shall yield
MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000 7
Acknowledgements
Foremost, we thank Timm Vanderelli of FERROCELL.USA, inventor of
the patented FerrocellTM, for his resourceful technical help and support.
Special thanks for Mr. Zisis Makris (research assistant) for his lab
assistance in the experiments undertaken and also for assisting in the
shooting of all the photographic and video recordings at the experiments.
Thanks also to Dr. Stellios Kouridakis academic staff member and Mr.
Vaggelis Tzavopoulos technical staff member for their scientific and
engineering support.
Appendix
Supplemental sample of experimental data information and visual
material:
I.
Inside google drive 560MB
https://drive.google.com/drive/folders/0B0A8uTBvEiQRT2p0dHRVa2VjWUE?usp=sharing
II.
At https://ferrocellmicroscope.blogspot.gr/
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
D. Heyers, M. Manns, H. Luksch, O. Güntürkün, H.
Mouritsen, A visual pathway links brain structures active
during magnetic compass orientation in migratory birds,
PLoS One. 2 (2007). doi:10.1371/journal.pone.0000937.
R.E. Rosensweig, Rosensweig, Ronald E.
Ferrohydrodynamics. Courier Corporation, in: Courier
Corporation, 2013: pp. 46–50.
https://www.google.com/books?hl=en&lr=&id=ng_DAg
AAQBAJ&oi=fnd&pg=PP1&ots=Fnew3xnZ5R&sig=p
mQwgiT88T_R4TkvkmSZDH7xPqU.
M. Mahmoudi, S. Sant, B. Wang, S. Laurent, T. Sen,
Superparamagnetic iron oxide nanoparticles (SPIONs):
Development, surface modification and applications in
chemotherapy, Adv. Drug Deliv. Rev. 63 (2011) 24–46.
doi:10.1016/j.addr.2010.05.006.
Manuel Benz, "Superparamagnetism: Theory and
Applications." Discussion, (2012).
https://drive.google.com/file/d/0B0A8uTBvEiQRcG9PU
kZPd2VuSmM/view?usp=sharing (accessed December
4, 2017).
K. Butter, P.H.H. Bomans, P.M. Frederik, G.J. Vroege,
A.P. Philipse, Direct observation of dipolar chains in
iron ferrofluids by cryogenic electron microscopy, Nat.
Mater. 2 (2003) 88–91. doi:10.1038/nmat811.
K. Ladutenko, P. Belov, O. Peña-Rodríguez, A. Mirzaei,
A.E. Miroshnichenko, I. V. Shadrivov, Superabsorption
of light by nanoparticles, Nanoscale. 7 (2015) 18897–
18901. doi:10.1039/C5NR05468K.
Alberto Tufaile, Timm A. Vanderelli, and Adriana
Pedrosa Biscaia Tufaile, "Light polarization using
ferrofluids and magnetic fields." Advances in Condensed
Matter Physics, Condens. Matter Phys. 2017. (2017).
doi:10.1155/2017/2583717.
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Y. Gao, J.P. Huang, Y.M. Liu, L.X. Gao, K.W. Yu, X.
Zhang, Optical negative refraction in ferrofluids with
magnetocontrollability, Phys. Rev. Lett. 104 (2010).
doi:10.1103/PhysRevLett.104.034501.
R.C. Hansen, Fundamental Limitations in Antennas,
Proc. IEEE. 69 (1981) 170–182.
doi:10.1109/PROC.1981.11950.
J. Tyndall, "Diamagnetism and magne-crystallic action,
including the question of diamagnetic polarity. New
ed..," D.APPLETON AND COMPANY, 1888.
https://archive.org/details/researchesondiam00tynduoft
(accessed December 4, 2017).
S. Weinberg, The Quantum Theory of Fields, vol. 2,
1995. doi:10.1126/science.269.5231.1742.
J.C. Maxwell, A dynamical theory of the electromagnetic
field, Philos. Trans. R. Soc. London. 155 (1865) 459–
512. doi:10.1098/rstl.1865.0008.
P.A. Tipler, G. Mosca, Physics for Scientists and
Engineers, in: 6th ed., W.H. Freeman, 2007: p. 538.
https://books.google.gr/books?id=BMVR37-
8Jh0C&pg=PA538&redir_esc=y.
A.M. Martin et al., Vortices and vortex lattices in
quantum ferrofluids, Condens. Matter 29.10 103004.
(2017). doi:10.1088/1361-648X/aa53a6.
V.G. Bashtovoi, R.E. Rosensweig, Excitation and study
of subcritical waves on a magnetic fluid surface, J.
Magn. Magn. Mater. 122 (1993) 234–240.
doi:10.1016/0304-8853(93)91082-I.
[16] M. Zahn, L.L. Pioch, Ferrofluid flows in AC and
[17]
[18]
[19]
[20]
[21]
traveling wave magnetic fields with effective positive,
zero or negative dynamic viscosity, J. Magn. Magn.
Mater. 201 (1999) 144–148. doi:10.1016/S0304-
8853(99)00099-2.
H.-Y. Wu, M. Rubinstein, E. Shih, J. Guttag, F. Durand,
W. Freeman, Eulerian video magnification for revealing
subtle changes in the world, ACM Trans. Graph. 31
(2012) 1–8. doi:10.1145/2185520.2335416.
R.E. Rosensweig, M. Zahn, R. Shumovich, Labyrinthine
instability in magnetic and dielectric fluids, J. Magn.
Magn. Mater. 39 (1983) 127–132. doi:10.1016/0304-
8853(83)90416-X.
S. Van Berkum, J.T. Dee, A.P. Philipse, B.H. Erné,
Frequency-dependent magnetic susceptibility of
magnetite and cobalt ferrite nanoparticles embedded in
PAA hydrogel, Int. J. Mol. Sci. 14 (2013) 10162–10177.
doi:10.3390/ijms140510162.
T. Franklin, Ferrofluid Flow Phenomena (Thesis),
Massachussetts Inst. Technol. (2003).
http://pages.csam.montclair.edu/~yecko/ferro/papers/Fer
roReviewsTheses/Franklin_FerrofluidFlow_Thesis.pdf.
C.J. Van Oss, D.R. Absolom, A.W. Neumann,
Applications of net repulsive van der Waals forces
between different particles, macromolecules, or
biological cells in liquids, Colloids and Surfaces. 1
(1980) 45–56. doi:10.1016/0166-6622(80)80037-0.
doi: https://doi.org/10.1016/j.jmmm.2017.12.023
MAGNETISM AND MAGNETIC MATERIALS JOURNAL 00 (2017) 000–000
S.H. Ebaadi, Van der Waals interaction between
surfactant-coated and bare colloidal particles, Colloids
and Surfaces. 2 (1981) 155–168. doi:10.1016/0166-
6622(81)80005-4.
P. Debye, Polar Molecules, J. Soc. Chem. Ind. 48 (1929)
1036–1037. doi:10.1002/jctb.5000484320.
J. Connolly, T.G. St Pierre, Proposed biosensors based
on time-dependent properties of magnetic fluids, J.
Magn. Magn. Mater. 225 (2001) 156–160.
doi:10.1016/S0304-8853(00)01245-2.
K.H.J. Buschow, Handbook of Magnetic Materials,
Volume 16, in: 1st ed., Elsevier, 2006: pp. 139–140.
https://www.elsevier.com/books/handbook-of-magnetic-
materials/buschow/978-0-444-51850-7.
S. May, Rotational relaxation dynamics of ferrofluid
systems. Diss., University of Hull, 2011.
http://dcf.ds.mpg.de/fileadmin/mitarbeiter/Sonia_Uterma
nn/publications/050131_Rotational_relaxation_dynamics
_of_ferrofluid_systems.pdf.
I. Torres-Díaz, C. Rinaldi, Recent progress in ferrofluids
research: novel applications of magnetically controllable
and tunable fluids: Fig.1 at p.8587, Soft Matter. 10
(2014) 8584–8602. doi:10.1039/C4SM01308E.
K.H.J. Buschow, Handbook of Magnetic Materials,
Volume 16, in: 1st ed., Elsevier, 2006: p. 147.
https://www.elsevier.com/books/handbook-of-magnetic-
materials/buschow/978-0-444-51850-7.
H.B. Braun, Topological effects in nanomagnetism:
From superparamagnetism to chiral quantum solitons,
Adv. Phys. 61 (2012) 1–116.
doi:10.1080/00018732.2012.663070.
Please cite this article as: E. Markoulakis, I. Rigakis, J. Chatzakis, A. Konstantaras, E. Antonidakis,
Real Time Visualization of Dynamic Magnetic Fields with a Nanomagnetic FerroLens, Journal of
Magnetism and Magnetic Materials (2017), doi: https://doi.org/10.1016/j.jmmm.2017.12.023
NonCommercial-NoDerivatives 4.0 International License
This work is licensed under Creative Commons Attribution-
8
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
.
|
1812.05559 | 3 | 1812 | 2019-04-23T13:44:24 | Measuring thickness in thin NbN films for superconducting devices | [
"physics.app-ph",
"cond-mat.supr-con"
] | We present the use of a commercially available fixed-angle multi-wavelength ellipsometer for quickly measuring the thickness of NbN thin films for the fabrication and performance improvement of superconducting nanowire single photon detectors. The process can determine the optical constants of absorbing thin films, removing the need for inaccurate approximations. The tool can be used to observe oxidation growth and allows thickness measurements to be integrated into the characterization of various fabrication processes. | physics.app-ph | physics | Measuring thickness in thin NbN films
Measuring thickness in thin NbN films for superconducting devices
O. Medeiros,1, a) M. Colangelo,1 I. Charaev,1 and K. K. Berggren1, b)
Massachusetts Institute of Technology, Department of Electrical
Engineering and Computer Science, Cambridge, MA 02139,
United States of America
(Dated: 24 April 2019)
We present the use of a commercially available fixed-angle multi-wavelength ellip-
someter for quickly measuring the thickness of NbN thin films for the fabrication
and performance improvement of superconducting nanowire single photon detectors.
The process can determine the optical constants of absorbing thin films, removing
the need for inaccurate approximations. The tool can be used to observe oxidation
growth and allows thickness measurements to be integrated into the characterization
of various fabrication processes.
9
1
0
2
r
p
A
3
2
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
3
v
9
5
5
5
0
.
2
1
8
1
:
v
i
X
r
a
a)Also at Department of Interdisciplinary Engineering, Wentworth Institute of Technology
b)Electronic mail: [email protected]
1
Measuring thickness in thin NbN films
I.
INTRODUCTION
Niobium nitride NbN is widely used for the fabrication of superconducting devices such as:
superconducting nanowire single photon detectors (SNSPDs)1, microwave kinetic inductance
detectors (MKIDs)2, superconducting electronic devices3,4, nanowire memory devices5, and
hot electron bolometers (HEBs)6. The working principle of these devices is intrinsically
dependent on the kinetic inductance (Lk) which can be related to thickness (d) by Lk ∝ d−1 7.
Measuring the thickness of thin films is therefore critical for the characterization, fabrication,
understanding, and improvement of superconducting devices.
The thickness of a thin film can be measured by a number of mechanical, electrical, and
optical methods8. In each of these categories, the methods rely on some assumption about
the material properties. Mechanical methods based on accurate mass determination, such as
weighing, use the thin film's average density which can be much less than the bulk density
and is often difficult to determine8. Similarly, electrical properties are also highly dependent
on the deposition conditions, making any resistance-based thickness determination equally
unreliable8. These resistance measurements have also been shown to change over time as a
function of surface oxide formation, adding an additional uncertainty source9. Optically, an
infrared transmissiometer has been shown to be a simple and fast method for determining
the thickness of NbN thin films10. Unfortunately, this technique cannot determine the opti-
cal properties (complex index of refraction, optical constants, refractive index and extinction
coefficient, n and k) of the deposited NbN, leaving a degree of uncertainty in the accuracy
of the measurement. Other optical methods, such as x-ray reflectivity (XRR), are subject to
long alignment and measurement times which makes the measurement difficult to integrate
into most fabrication processes. Alternatively, the thickness can be approximated by the
deposition time. However, these approximations are generally inaccurate as the deposition
rate can vary depending on the deposition parameters and target erosion11. Topographical
methods, such as atomic force microscopy or profilometry, can produce high-resolution mea-
surements but require additional fabrication (e.g. etch or lift-off) when determining film
thickness. The preferred method for measuring a thin film in-process would thus be one
that simultaneously determines the refractive index, extinction coefficient and thickness.
Ellipsometry is a well-established method of thin film metrology applicable to plasmonic12,
semiconductor13, and biosensor14 applications. The thickness and optical constants of trans-
2
Measuring thickness in thin NbN films
parent thin films, such as SiO2, can generally be determined by ellipsometry with high
precision, but due to the absorbing nature of NbN these properties cannot be determined
from a single-angle, single-wavelength ellipsometry measurement15. However, it is possible
to determine these properties by analyzing multiple samples16. This method is referred
to as the multi-sample analysis method and requires a set of measurements from samples
with identical optical properties and varying thickness. The measured values will follow a
unique curve which is determined by the materials optical constants as the thickness of the
film increases. This approach is significantly less involved than the use of variable-angle
spectroscopic ellipsometry17 thanks to the fixed angle and limited number of wavelengths
necessary for applying this method. For that reason, the thickness and optical constants of
NbN are best determined by the ellipsometry based multi-sample analysis method.
In this paper, we determined the thickness and optical constants of NbN thin films using
a fixed-angle spectral ellipsometer (Film Sense FS-1 Multi-Wavelength Ellipsometer). Tra-
ditional ellipsometer parameters, Ψ and ∆, relating to the change in magnitude and phase
of the reflected polarized beam, were measured15. They were then used to fit generated Ψ,
∆ curves for each wavelength as a function of the film's thickness and optical constants.
The generated data was produced using OpenFilters, an open-source software package for
simulating thin film optical models18 and was fit to the measured data using the Marquardt-
Levenberg algorithm19,20. The thicknesses determined by the fitting were compared to x-ray
reflectivity and sheet resistance measurements and the optical properties were used to track
changes in the film over time.
II. METHODS
The thicknesses of NbN films measured by ellipsometry, x-ray reflectivity, and sheet
resistance measurements were obtained to assess ability to measure thin films. The result of
each technique depends on the technical details of the method. In this section we describe
how samples were prepared, how each method was conducted, and how the ellipsometer can
record changes in NbN and surface oxide thicknesses over time.
3
Measuring thickness in thin NbN films
A. NbN Deposition
To produce samples for multi-sample analysis, thin films of NbN were deposited on 220 nm
of thermal oxide SiO2 on Si (100) substrates cut into 1 cm by 1 cm samples from a 4 in wafer.
The depositions were conducted at room temperature by DC reactive magnetron sputtering
using an AJA International Inc Orion series system. The deposition conditions were deter-
mined by Dane et al to produce NbN films suitable for SNSPDs at room temperature21.
These conditions were controlled by creating an automated process in the AJA Phase IIJ
software and were constant for all samples while the deposition time was varied to produce
a range of film thicknesses. The deposition times ranged from 20 s to 200 s in increments
of 20 s, and also included 400 s and 600 s depositions. These times were chosen to produce
thicknesses between 1 nm and 30 nm.
B. Spectroscopic Ellipsometry
To determine the optical constants and thickness of the NbN thin films, each sample was
first measured to produce Ψ and ∆. To measure a sample, an alignment was performed to
be within ± 0.01◦ and the acquisition time was set to 1 s. Longer acquisition times did not
show an increase in accuracy. The alignment and measurement of a sample was completed
in less than 30 s using the ellipsometer. Data from each measurement was then assembled
in the ellipsometer's software and evaluated using an optical model. The optical model
was composed of optical layers that were representative of the measured sample. Here, the
optical model included a n and k layer for our undefined NbN on top of a thermally grown
SiO2 layer on a Si substrate. In multi-sample analysis it was assumed that each sample has
identical layer parameters other than the thickness of a single layer. For this reason, a Nb
surface oxide layer was not included during the multi-sample analysis and the SiO2 layer was
fit as a single thickness for all samples. The multi-sample analysis simultaneously produced
thicknesses for each sample and the n and k values of our NbN.
Having defined the optical properties of our NbN, a predefined Nb2O5 layer22 was added
to the optical model to investigate the oxidation of NbN. NbN exhibits poor oxidation
resistance and the formation of a surface oxide presents some doubt in our understanding
of the film's material properties which can be problematic for fabrication processes like
4
Measuring thickness in thin NbN films
reactive ion etching23. The product of NbN oxidation has been shown to be Nb2O5 without
the formation of lower valent oxides in earlier studies by Gallagher et al 24,25. By adding a
Nb2O5 layer to the optical model, fixed angle ellipsometry was able to track the oxidation
NbN. In this experiment, a 200 s deposition (8 nm) of NbN was transported in reduced
atmosphere (∼0.5 atm) box after being removed from the deposition chamber so the exposure
to atmospheric conditions before the initial measurement would be minimized. The sample
was measured over increasing time intervals to track the progress of oxidation while stored
in atmospheric conditions. The measurement assumed the SiO2 thickness was constant for
the full duration of the experiment. The oxidation of NbN was observed by fitting the NbN
and Nb2O5 layers in the optical model.
C. Characterization
The thickness measurements from the ellipsometer were validated by two additional mea-
surement methods. In the first method, x-ray measurements were performed using a Rigaku
Smartlab X-Ray Defractometer conducting parallel-beam x-ray reflectometry which pro-
duced x-rays at a wavelength of 0.1541 nm from a copper target. The XRR measurements
took more than 40 minutes per sample. In the second method, the resistance of each film
was measured after deposition at room temperature using a four point probe. The measured
resistance (R) can be used to calculate the sheet resistance (Rs) by Rs = Raπ(ln(2))−1 26,
where a is a geometric correction factor. The thickness can then be related to Rs by d = ρR−1
s
where ρ is the electrical resistivity and d is the thickness.
III. RESULTS
In this work, we obtained thickness measurements from each technique and tracked a
sample over a period of 100 days. These results are necessary for validating the ellipsometer's
ability to measure the thickness of NbN thin films. The details provided in this section
compare the ellipsometer and XRR measurements and present changes in NbN and Nb2O5
thicknesses over 100 days.
5
Measuring thickness in thin NbN films
A. Optical Model Validation
The Ψ, ∆ curves in Fig. 1 shows the locus of increasing thickness for the determined
complex index of refraction by fitting to the measured points. The experimental data from
a single sample produced four points, one for each wavelength. The curves were fit to the
experimental data with a mean square error of 0.011. The mean square error, or fit dif-
ference, calculated by the ellipsometer ranged from 0.055 for the thinnest sample to 0.0156
for the thickest sample. The values in Table I are the optical constants of NbN determined
by multi-sample analysis. These values are specific to the NbN produced under the condi-
tions described but are comparable to values previously reported27. The model was further
validated using the built-in validation software.
The thickness measurements from the ellipsometer are shown as a function of the sheet
conductance in Fig. 2. The slope of this linear fit corresponds to a resistivity of 245 µΩcm
and an adjusted R-square value of 0.9991. The fit is expected to have a y-intercept at
zero but has an intercept equal to 0.88 nm.
It can be expected to be zero because zero
conductive material should equate to zero thickness. The resulting intercept implies that
resistivity increases with reduced film thickness and this dependency is shown in the Fig.
2 inset. This increase has been observed by other groups as well and could be attributed
to films where the mean free path is less than the bulk materials mean free path and the
charge carrier density is reduced28,29.
A comparison between the measured thickness of each thin film by ellipsometry and XRR
as a function of deposition time is shown in Fig. 3. In this figure, the XRR measurements for
deposition times shorter than 100 s, shown as open circles, deviated from the ellipsometer
measurements, closed circles. As the deposition time approaches zero, the thickness of
the deposited material should also approach zero, which was not observed in the XRR
measurements. However, these measurements on the XRR did not produce interference
fringes, likely reducing the accuracy of the fit. For film deposition times longer than 100 s
the ellipsometer thicknesses fell within 0.5 nm of the XRR measurements.
6
Measuring thickness in thin NbN films
FIG. 1. The inferred Ψ, ∆ values for each sample are plotted as closed shapes. The legend denotes
the corresponding wavelength and its respective color. The curves show simulated data using the
optical constants determined by multi-sample analysis. The curves increase from zero thickness
(open circles) to 60 nm in 10 nm increments (tick marks).
B. Oxide Formation
The time-dependence of thicknesses of NbN and Nb2O5 according to the ellipsometry
measurements are shown in Fig. 4a. After 100 days, the measured thickness of the NbN
FIG. 2. Thickness determined by ellipsometry measurements plotted as a function of sheet con-
ductance. The linear fit shown has a slope of 245 µΩcm. The inset shows the resistivity of the
NbN films as a function of thickness.
7
Measuring thickness in thin NbN films
TABLE I. Optical constants of NbN for each wavelength.
λ (nm)
466
522
598
638
n
2.456
2.554
2.643
2.822
k
2.487
2.536
3.047
3.197
FIG. 3. Measured thickness of NbN as a function of deposition time according to XRR and
ellipsometer measurements. The XRR measurements are shown as open circles and the ellipsometer
measurements are shown as closed circles. As deposition time approaches zero the thickness should
approach zero. The XRR measurements for deposition times under 100 s deviated from the expected
linear trend.
film decreased by 8.7%. The optical model for these measurements used the thickness of
Nb2O5 and NbN as fitting parameters while the SiO2 thickness remained constant. The
Nb2O5 thicknesses, shown in Fig. 4a and isolated in Fig. 4b, fit the line dox = 0.78t0.17
where dox is the Nb2O5 thickness and t is the elapsed time in days, with sum of squares
error equal to 0.04. The oxide formation agrees with the general relationship defined by the
Deal-Grove diffusion model30.
8
Measuring thickness in thin NbN films
m = 0.17
FIG. 4. a. Stacked thickness of Nb2O5 on NbN over a duration of 100 days. The ellipsometer
measurements show a decrease in NbN thickness and an increase in Nb2O5 thickness. The Nb2O5
is shown in white on top of the NbN shown in gray. b. The same Nb2O5 thickness measured by
the ellipsometer at different intervals after being removed from the deposition chamber is shown
with a linear fit.
IV. DISCUSSION
The optical method evaluated in this study is a convenient method for simultaneously
characterizing the thickness and optical constants of an unknown absorbing film. Measure-
ments performed using the ellipsometer were, in total, faster than a single measurement
performed using the XRR while being more accurate for depositions under 100 s. This ap-
proach differs from the use of the sheet resistance to thickness relation ρ = Rsd because
the ellipsometer is able to provide a qualitative metric for the accuracy of the measurement
without relying on an assumption of the material's resistivity. By assuming a constant re-
sistivity, it is unknown whether a change in thickness or deposition parameter resulted in a
change in a materials sheet resistance. Defining the material's optical properties with the
ellipsometer can resolve this uncertainty.
Understandably, there are certain scenarios where this method is not applicable. The first
scenario applies to films where there is a strong correlation between the optical constants
and thickness. This dependency is true for many metallic thin films, including NbN27,31.
9
Measuring thickness in thin NbN films
The correlation between thickness and optical constants can depend on thickness as well as
the wavelength and crystalline orientation. Decorrelating these parameters can be achieved
by increasing the number of interference-oscillations. This interference enhancement can be
accomplished by adding a transparent film below the absorbing layer and simultaneously
analyzing multiple samples at various wavelengths or angles of incidence16,32. The second
scenario occurs when deposition conditions are not constant. Changes in the deposition
conditions could result in optical properties that differ from those defined by the multi-
sample analysis, producing inaccurate results. This constraint could be used to determine
what effect the deposition conditions have on the structure of the film by observing changes
in the measurement's fit difference. The third scenario is when the thickness of the film
is greater than the film's absorbing limit. As thickness of an absorbing film increases the
incident beam becomes attenuated and the ellipsometer becomes inaccurate. The attenuated
beam likely causes the increase in fit difference across our measurements. To determine this
limit the ellipsometer is able to calculate Ψ, ∆ values of an optical model as a function of
thickness. As thickness increases, the resulting Ψ, ∆ coordinate begins to converge to a
single value and the measurement is no longer accurate. For these NbN films, a conservative
upper limit of the measurement is approximately 40 nm.
V. CONCLUSION
We presented a fast, nondestructive and accurate method for determining the film thick-
ness of NbN thin films. The results show promise for tracking time or process dependent
changes in ultra thin films by determining the optical properties of an unknown absorbing
film using a fixed angle multi-wavelength ellipsometer. Applications that rely on kinetic in-
ductance or other thickness-dependent electrical properties would benefit from this method.
By accurately determining the film thickness and optical constants, we construct a bet-
ter understanding of the material we are creating. This understanding can lead to further
investigation of the effect that deposition parameters have on optical constants.
10
Measuring thickness in thin NbN films
ACKNOWLEDGMENTS
The authors would like to acknowledge Blaine Johs for his support with the use of the
Film Sense FS-1. They would also like to thank James Daley and Mark Mondol of the MIT
Nanostructures Laboratory and Charles Settens of the MIT Center for Material Science and
Engineering for their technical support. Support for investigation and characterization of
the measurement method itself was sponsored by the Army Research Office (ARO) and was
accomplished under Cooperative Agreement Number W911NF-16-2- 0192. The views and
conclusions contained in this document are those of the authors and should not be interpreted
as representing the official policies, either expressed or implied, of the Army Research Office
or the U.S. Government. The U.S. Government is authorized to reproduce and distribute
reprints for Government purposes notwithstanding any copyright notation herein. Support
for the study of oxide growth vs time was provided by Skoltech under Research Agreement
No. 1921/R.
REFERENCES
1G. Gol'tsman, O. Okunev, G. Chulkova, A. Lipatov, A. Semenov, K. Smirnov, B. Voronov,
A. Dzardanov, C. Williams,
and R. Sobolewski, "Picosecond superconducting single-
photon optical detector," Applied physics letters 79, 705 -- 707 (2001).
2P. K. Day, H. G. LeDuc, B. A. Mazin, A. Vayonakis, and J. Zmuidzinas, "A broadband
superconducting detector suitable for use in large arrays," Nature 425, 817 (2003).
3A. N. McCaughan and K. K. Berggren, "A superconducting-nanowire three-terminal elec-
trothermal device," Nano letters 14, 5748 -- 5753 (2014).
4A. N. McCaughan, N. S. Abebe, Q.-Y. Zhao, and K. K. Berggren, "Using geometry to
sense current," Nano letters 16, 7626 -- 7631 (2016).
5Q.-Y. Zhao, E. A. Toomey, B. A. Butters, A. N. McCaughan, A. E. Dane, S.-W. Nam,
and K. K. Berggren, "A compact superconducting nanowire memory element operated by
nanowire cryotrons," Superconductor Science and Technology 31, 035009 (2018).
6P. Khosropanah, J. Gao, W. Laauwen, M. Hajenius, and T. Klapwijk, "Low noise nbn
hot electron bolometer mixer at 4.3 thz," Applied Physics Letters 91, 221111 (2007).
11
Measuring thickness in thin NbN films
7A. J. Annunziata, D. F. Santavicca, L. Frunzio, G. Catelani, M. J. Rooks, A. Frydman,
and D. E. Prober, "Tunable superconducting nanoinductors," Nanotechnology 21, 445202
(2010).
8O. S. Heavens, Optical properties of thin solid films (Courier Corporation, 1991).
9D. F. Santavicca and D. E. Prober, "Aging of ultra-thin niobium films," IEEE Transactions
on Applied Superconductivity 25, 1 -- 4 (2015).
10K. A. Sunter, A. E. Dane, C. I. Lang, and K. K. Berggren, "Infrared transmissometer to
measure the thickness of nbn thin films," Applied optics 54, 5743 -- 5749 (2015).
11P. J. Kelly and R. D. Arnell, "Magnetron sputtering: a review of recent developments and
applications," Vacuum 56, 159 -- 172 (2000).
12T. Oates, H. Wormeester, and H. Arwin, "Characterization of plasmonic effects in thin
films and metamaterials using spectroscopic ellipsometry," Progress in Surface Science 86,
328 -- 376 (2011).
13S.-G. Lim, S. Kriventsov, T. N. Jackson, J. Haeni, D. Schlom, A. Balbashov, R. Uecker,
P. Reiche, J. Freeouf, and G. Lucovsky, "Dielectric functions and optical bandgaps of
high-k dielectrics for metal-oxide-semiconductor field-effect transistors by far ultraviolet
spectroscopic ellipsometry," Journal of applied physics 91, 4500 -- 4505 (2002).
14C. Striebel, A. Brecht, and G. Gauglitz, "Characterization of biomembranes by spec-
tral ellipsometry, surface plasmon resonance and interferometry with regard to biosensor
application," Biosensors and Bioelectronics 9, 139 -- 146 (1994).
15R. Archer, "Determination of the properties of films on silicon by the method of ellipsom-
etry," JOSA 52, 970 -- 977 (1962).
16W. McGahan, B. Johs, and J. Woollam, "Techniques for ellipsometric measurement of
the thickness and optical constants of thin absorbing films," Thin solid films 234, 443 -- 446
(1993).
17J. A. Woollam, B. D. Johs, C. M. Herzinger, J. N. Hilfiker, R. A. Synowicki, and C. L.
Bungay, "Overview of variable-angle spectroscopic ellipsometry (vase): I. basic theory and
typical applications," in Optical Metrology: A Critical Review, Vol. 10294 (International
Society for Optics and Photonics, 1999) p. 1029402.
18S. Larouche and L. Martinu, "Openfilters: open-source software for the design, optimiza-
tion, and synthesis of optical filters," Applied optics 47, C219 -- C230 (2008).
12
Measuring thickness in thin NbN films
19D. W. Marquardt, "An algorithm for least-squares estimation of nonlinear parameters,"
Journal of the society for Industrial and Applied Mathematics 11, 431 -- 441 (1963).
20F. S. M.-W. Ellipsometers, FS-1 Manual (2017).
21A. E. Dane, A. N. McCaughan, D. Zhu, Q. Zhao, C.-S. Kim, N. Calandri, A. Agarwal,
F. Bellei, and K. K. Berggren, "Bias sputtered nbn and superconducting nanowire de-
vices," Applied Physics Letters 111, 122601 (2017).
22M. N. Polyanskiy, "Refractive index database," .
23E. Toomey, M. Colangelo, N. Abedzadeh, and K. K. Berggren, "Influence of tetram-
ethylammonium hydroxide on niobium nitride thin films," Journal of Vacuum Science &
Technology B, Nanotechnology and Microelectronics: Materials, Processing, Measurement,
and Phenomena 36, 06JC01 (2018).
24P. Gallagher, W. Sinclair, D. Bacon, and G. Kammlott, "Oxidation of sputtered niobium
nitride films," Journal of The Electrochemical Society 130, 2054 -- 2056 (1983).
25P. K. Gallagher and W. R. Sinclair, "Oxidation of polycrystalline niobium nitride," Israel
Journal of Chemistry 22 (1982).
26L. J. Van der Pauw, "A method of measuring specific resistivity and hall effect of discs of
arbitrary shape," Philips Res. Rep. 13, 1 -- 9 (1958).
27A. Banerjee, R. M. Heath, D. Morozov, D. Hemakumara, U. Nasti, I. Thayne,
and
R. H. Hadfield, "Optical properties of refractory metal based thin films," Optical Materials
Express 8, 2072 -- 2088 (2018).
28S. Chockalingam, M. Chand, J. Jesudasan, V. Tripathi, and P. Raychaudhuri, "Super-
conducting properties and hall effect of epitaxial nbn thin films," Physical Review B 77,
214503 (2008).
29A. Kamlapure, M. Mondal, M. Chand, A. Mishra, J. Jesudasan, V. Bagwe, L. Benfatto,
V. Tripathi, and P. Raychaudhuri, "Measurement of magnetic penetration depth and
superconducting energy gap in very thin epitaxial nbn films," Applied Physics Letters 96,
072509 (2010).
30B. E. Deal and A. Grove, "General relationship for the thermal oxidation of silicon,"
Journal of Applied Physics 36, 3770 -- 3778 (1965).
31A. Semenov, B. Gunther, U. Bottger, H.-W. Hubers, H. Bartolf, A. Engel, A. Schilling,
K. Ilin, M. Siegel, R. Schneider, et al., "Optical and transport properties of ultrathin nbn
films and nanostructures," Physical Review B 80, 054510 (2009).
13
Measuring thickness in thin NbN films
32J. N. Hilfiker, N. Singh, T. Tiwald, D. Convey, S. M. Smith, J. H. Baker, and H. G.
Tompkins, "Survey of methods to characterize thin absorbing films with spectroscopic
ellipsometry," Thin Solid Films 516, 7979 -- 7989 (2008).
14
|
1801.05131 | 2 | 1801 | 2018-08-08T08:13:51 | Spin-orbit-torque and magnetic damping in tailored ferromagnetic bilayers | [
"physics.app-ph",
"cond-mat.mes-hall",
"cond-mat.mtrl-sci"
] | We study spin-orbit-torque-driven ferromagnetic resonance (FMR) in ferromagnetic (FM) bilayers, consisting of Co and permalloy (Py), sandwiched between Pt and MgO layers. We find that the FM layer in contact with the Pt layers dominantly determines that spin Hall angle, which is consistent with the spin-transparency model. By contrast, the FMR linewidths are considerably influenced not only by the spin-pumping effect across the Pt|FM in terface but also by the spin relaxation such as two-magnon scattering at the FMMgO interface.The CoMgO interface leads to notably increased FMR linewidths, while the Py|MgO interface has less effect. This different contribution of each interface to the spin Hall angel and dissipation parameter suggests that the stack configuration of Pt|Co|Py|MgO requires less writing energy than Pt|Py|Co|MgO in spin-orbit-torque-driven magnetic switching. Our approach offers a promising method to optimize material parameters by engineering either interfaces in contact with the heavy-metal or the oxide layer. | physics.app-ph | physics | Spin-orbit-torque and magnetic damping in tailored ferromagnetic bilayers
DongJoon Lee1,2, JongHyuk Kim1,3, HeeGyum Park1,4, Kyung-Jin Lee2,5, Byeong-Kwon Ju3,
Hyun Cheol Koo1,2, Byoung-Chul Min1,4, and OukJae Lee1*
1Center for Spintronics, Korea Institute of Science and Technology, Seoul 02792, Korea
2KU-KIST Graduate School of Converging Science and Technology, Korea University, Seoul 02841, Korea
3Department of Electrical Engineering, Korea University, Seoul 02841, Korea
4Division of Nano and Information Technology, KIST school, Korea University of Science and Technology,
Seoul, 02792, Korea
5Department of Materials Science and Engineering, Korea University, Seoul 02841, Korea
We study spin-orbit-torque-driven ferromagnetic resonance (FMR) in ferromagnetic
(FM) bilayers, consisting of Co and permalloy (Py), sandwiched between Pt and MgO
layers. We find that the FM layer in contact with the Pt layer dominantly determines the
spin Hall angle, which is consistent with the spin-transparency model. By contrast, the
FMR linewidths are considerably influenced not only by the spin-pumping effect across
the PtFM interface but also by the spin relaxation such as two-magnon scattering at the
FMMgO interface. The CoMgO interface leads to notably increased FMR linewidths,
while the PyMgO interface has less effect. This different contribution of each interface
to the spin Hall angle and dissipation parameter suggests that the stack configuration of
PtCoPyMgO requires less writing energy than PtPyCoMgO in spin-orbit-torque-
driven magnetic switching. Our approach offers a practical tactic to optimize material
parameters by engineering either interfaces in contact with the heavy-metal or the oxide
layer.
* Email: [email protected]
1
I. INTRODUCTION
Electrical manipulation of magnetization [1,2,3,4] has been of great importance because
of its potential application in low-power and high-speed spintronic devices as well as
remarkable scientific findings. More recently, a number of studies have demonstrated that in-
plane charge currents generate spin-orbit-torques (SOTs) that can directly switch the
magnetization in hetero-structures with a strong spin-orbit coupling (SOC) [2,3,4,5,6]. For the
practical use of SOT for magnetic random-access memory (MRAM) applications, it is required
to reduce the writing current density (Jc) not only for low power consumption and CMOS
compatibility [7] but also for enhanced endurance and reliability. The application of a relatively
high Jc inevitably generates a substantial amount of Joule heating or electro-migration that
could deteriorate magnetic and transport properties in each memory cell, thereby limiting the
lifetime of SOT-based MRAMs.
The zero-temperature critical current density (Jc0) depends on the characteristics of
magnetic switching processes as well as many material parameters. For an in-plane
magnetization switching (in the rigid-domain approximation) driven by an in-plane anti-
damping SOT [1,8],
𝐽𝑐0 ≈
2𝑒
ℏ
𝛼
𝐼𝑃 (
𝜃𝑆𝐻
4𝜋𝑀𝑒𝑓𝑓
2
) 𝑀𝑠𝑡𝐹𝑀, (1)
where α is the magnetic damping, 𝜃SH
𝐼𝑃 is the conversion efficiency from charge currents to
spin currents (having in-plane spin-polarization), Ms is the saturation magnetization, tFM is the
magnet thickness, 4𝜋𝑀eff is the effective out-of-plane demagnetization field (= 4𝜋𝑀𝑠 −
2𝐾𝑠/𝑀𝑠𝑡FM > 0) for an in-plane-magnetized film. For a perpendicular magnetic reversal via
coherent rotation by the in-plane anti-damping SOT [9],
2
𝐽𝑐0 ≈
2𝑒
ℏ
1
𝐼𝑃 (
𝜃𝑆𝐻
𝑒𝑓𝑓
𝐻𝑘
2
) 𝑀𝑠𝑡𝐹𝑀, (2)
where 𝐻𝑘
𝑒𝑓𝑓
is the effective out-of-plane field (= 2𝐾𝑠/𝑀𝑠𝑡FM − 4𝜋𝑀𝑠 > 0). In order to
achieve a uniform switching of an out-of-plane magnetized film, the lateral dimension of a
magnet is limited to less than 30 nm. Otherwise, the magnetic reversal is achieved by the
nucleation of reversed domains followed by expansion via domain-wall depinning and
propagation [10,11]. Another interesting case is an anti-damping SOT with out-of-plane spin
polarization as recently reported [12] in semi-metal WTe2 whose crystal structure has broken
lateral mirror symmetry. This type of SOT [12,13] provides a much more efficient pathway for
switching of a perpendicular magnet as
𝐽𝑐0 =
2𝑒
ℏ
𝛼
𝑂𝑃 (𝐻𝑘
𝜃𝑆𝐻
𝑒𝑓𝑓)𝑀𝑠𝑡𝐹𝑀, (3)
where 𝜃SH
𝑂𝑃 is the conversion efficiency from charge currents to spin currents having out-of-
plane spin-polarization. A much lower Jc0 is expected in this case, since 𝛼𝐻𝑘
𝑒𝑓𝑓
<< 𝐻𝑘
𝑒𝑓𝑓/2
with a typical magnetic damping constant of metallic ferromagnets (α ≈ 0.005 − 0.03).
In order to minimize 𝐽c, hence, it is important to employ a multilayer system exhibiting a
large θSH
eff, a small α and a low 𝐻𝑘
𝑒𝑓𝑓
. A low resistivity of HM (ρHM) is also desirable to supply
more electrons, given the same voltage bias. Unfortunately, there exist trade-offs among those
material parameters. Several studies [2,4,14] have demonstrated that a strong SOC material
tends to exhibit a larger θSH
eff as its ρHM increases. Furthermore, the recent spin-transparency
model [15,16] suggests that the simultaneous achievement of a large θSH
eff and a low α may be
improbable. In the model, both the effective magnitude of θSH
eff and the increased amount of
magnetic damping (Δαsp) owing to the spin-pumping effect [17] are influenced concurrently by
the same parameter, namely the interface transparency (Tsp) of the spin current between a heavy
3
metal (HM) and a FM layer. For instance, Ref. [15] showed θSH
eff(PtCo) ≈ 0.11 vs θSH
eff(PtPy)
≈ 0.05, but Δαsp(PtCo) ≈ 2·Δαsp(PtPy). This indicates that the increase of θSH
eff with a higher
Tsp will be compensated by an increase of α so that a variation of Tsp would not be very helpful
for lowering 𝐽c.
In this paper, we suggest a practical strategy to relieve this issue, using tailored FM bilayers
instead of a single FM layer. The bilayers consisting of Co and Py (=Ni79Fe21) are sandwiched
between Pt and MgO layers and in-plane magnetized. Using spin-torque ferromagnetic
resonance (ST-FMR), we investigated how the stacking order and relative thicknesses influence
the effective spin Hall angle (θSH
eff), magnetic damping constant (α), inhomogeneous
linewidth-broadening (𝛥𝐻0), and effective out-of-plane demagnetization field (4𝜋𝑀eff). Our
results show that (i) the θSH
eff is mostly determined by the interface material in contact with
the Pt layer, and (ii) the α and 𝛥𝐻0 are considerably influenced by both interfaces in contact
with the HM (Pt) and the oxide layer (MgO).
II. SAMPLES AND METHODS
We used DC/RF magnetron sputtering to deposit two series of multilayer films with
different stacking orders, PtCoPyMgO or PtPyCoMgO, on thermally oxidized Si substrates
at room temperature (RT). The multilayer structures consist of, from the substrate side, (CoPy)-
series: Ta(1)Pt(5)Co(tCo)Py(5‒tCo)MgO(2)Ta(2) and (PyCo)-series: Ta(1)Pt(5)Py(5‒
tCo)Co(tCo)MgO(2)Ta(2) (nominal thickness in nm). The thickness of the Co layer (tCo) was
varied from 0 -- 5 nm while the total thickness of the FM bilayers was fixed to 5 nm for both
series, as illustrated in Fig. 1(a). The magnetic bilayer of CoPy or PyCo can be regarded as a
single FM unit in our analysis because the total thickness of the bilayer (5 nm) is comparable
4
to or smaller than the exchange length of each Co or Py layer (≈ 5 − 10 nm). The Ta (1 nm)
seed layer was employed as a wetting layer, and the MgO(2)Ta(2) capping layer protects its
under-layers and is expected to be fully air-oxidized. The base pressure of the chamber was
maintained less than 5 × 10-8 Torr and the deposition rates were kept lower than 0.5 Å /s. For
the ST-FMR measurement, we patterned the multilayer films into rectangular strips with 15
μm-width (𝑤) and 50 μm-length (𝑙), using optical lithography and Ar ion milling [Fig. 1(b)].
In a subsequent process step, a waveguide contact made of Ti(10 nm)Au(100 nm) was defined
on top of the samples to apply a microwave electric current to the devices. The samples were
not exposed to high temperature (> 120 C) during the fabrication process as no post-annealing
was carried out.
The first step is to characterize the stacking-order dependence of Ms and Hs
op (out-of-plane
saturation field) for the un-patterned films of the (CoPy)- and (PyCo)-series. The M -- H loop
measured by a vibrating sample magnetometer [the inset of Fig. 1(d)] shows that the two FM
layers are coupled via a strong exchange energy. Figure 1(c) shows that the Ms of both (CoPy)-
and (PyCo)- series monotonously increases with increasing tCo because the Ms of Co (≈
1140 emu/𝑐𝑚3) is larger than that of Py (≈ 640 emu/𝑐𝑚3). The Ms of both (CoPy)- and
(PyCo)-series are almost identical to each other, thereby indicating that the stacking order of
CoPy or PyCo on top of the Pt layer has a minor effect on Ms. By contrast, as shown in Fig.
1(d), the Hs
op has a clear dependence on the stacking order: the Hs
op of the (CoPy)- stack is
lower than that of the (PyCo)-series with the same tCo. With the measured Ms and Hs
op, the net
difference of 𝐾𝑠 between two stacks is estimated as ∆𝐾𝑠 = 𝐾𝑠(𝑃𝑡𝐶𝑜𝑃𝑦𝑀𝑔𝑂) −
𝐾𝑠(𝑃𝑡𝑃𝑦𝐶𝑜𝑀𝑔𝑂) ≈ 0.6 ± 0.07 erg/cm2, which is close to the previously reported [16,18]
𝐾𝑠 ≈ 0.8 − 1.1 erg/cm2 in PtCoMgO. Therefore, the net difference is mostly because of the
5
PtCo interface that has a strong 5d-3d hybridization, resulting in a higher surface anisotropy
energy (𝐾𝑠 ). The CoNi and CoMgO interfaces are known to have non-negligible surface
energies, but the contribution to the total anisotropy energy is not that significant in our films.
Next, we systematically investigated their effective spin-Hall angles and magnetic
damping constants using ST-FMR [12,14,15,16,19,20] method. The circuit diagram of the ST-
FMR measurements is illustrated in Fig. 1(b). A pulsed microwave signal in the range of 4 -- 14
GHz with a nominal output power of 10 dBm was applied to the samples. In the meantime, an
external magnetic field-sweep (from -1.8 to +1.8 kOe) was conducted at an angle of 45o within
the sample plane. The applied RF current (Irf) generates two different types of oscillating SOTs,
anti-damping torque ( 𝜏DL ) ∝ 𝑚 × (𝑦 × 𝑚) and field-like torque ( 𝜏FL ) ∝ 𝑚 × 𝑦 , to the
magnetization of the adjacent FM bilayer, as well as an Oersted field torque. These torques
excite the magnetic precession if frequency and external field satisfy the resonance condition,
thereby producing a net oscillation in the anisotropic and spin-Hall magnetoresistances. The
mixing of an oscillatory resistance and Irf passing through the FM bilayer generates a finite DC
voltage (Vmix), which is simultaneously detected with a lock-in amplifier connected to the DC
port of a bias-tee.
Figures 2(a) -- (d) show the representative spectra for the ST-FMR devices with Py(5), Co(5),
Py(4)Co(1), and Co(1)Py(4) at 8 GHz, exhibiting different resonant positions and linewidths.
The single resonance peak observed for all samples indicates that the two FM layers are
strongly coupled in-phase. All measured curves are in good agreement with fit to an equation
of ST-FMR signal (red curve) consisting of symmetric and asymmetric Lorentzian functions
according to:
6
𝑉mix(𝐻) = 𝑆
𝛥𝐻2
(𝐻−𝐻𝑟𝑒𝑠)2+(𝛥𝐻)2 + 𝐴
(𝐻−𝐻𝑟𝑒𝑠)𝛥𝐻
(𝐻−𝐻𝑟𝑒𝑠)2+𝛥𝐻2 (4)
where S (A) is the voltage amplitude of the symmetric (asymmetric) Lorentzian function, 𝐻res
is the resonance field, and 𝛥𝐻 is the half linewidth at the half maximum. In Figs. 2(a)-(d), the
symmetric (green) and anti-symmetric (blue) parts of the signals are also plotted for
comparison. From the ST-FMR theory, the symmetric part (S) of the signal is proportional to
damping-like SOT acting on the FM magnetization, while the anti-symmetric part (A)
originates from the sum of field-like SOT and Oersted field torques.
The ST-FMR spectra provide important parameters such as Hres, ΔH, S and A that will be
𝑒𝑓𝑓
used to quantify the magnitudes of 4𝜋𝑀eff, α, 𝛥𝐻0 and 𝜃𝑆𝐻
. The center frequency of the
resonance peak (f) follows the Kittel equation, 𝑓 = (𝛾/2𝜋)√𝐻res(𝐻res + 4𝜋𝑀eff), where 𝛾
is the gyromagnetic ratio (Fig. 2(e)) and the 4πMeff was extracted from a fit to the Kittel
equation. The α and 𝛥𝐻0 of each sample were derived from ∆𝐻 = ∆𝐻0 + 𝛼(2𝜋/𝛾)𝑓 (Fig.
𝑒𝑓𝑓
2(f)). The magnitude of 𝜃𝑆𝐻
was quantified using the voltage ratio, S/A; 𝜃𝑆𝐻
𝑒𝑓𝑓 = (𝑆/
𝐴) (𝑒4𝜋𝑀𝑠𝑡Pt𝑡FM/ℏ)√1 + 4𝜋𝑀eff/𝐻res, where 𝑡Pt is the thickness of Pt (5 nm) and 𝑡FM is
𝑒𝑓𝑓
the total thickness of the FM bilayers (= 5 nm). This voltage ratio analysis to obtain 𝜃𝑆𝐻
should be cautiously applied, since the analysis gives incorrect values if there is a significant
magnitude of non-negligible field-like SOT [3,16] or a self-induced Oersted field torque from
a non-uniform distribution of current density inside the FM bilayer. We crosschecked that the
𝑒𝑓𝑓
𝜃𝑆𝐻
𝑒𝑓𝑓
obtained in this study coincides with the 𝜃𝑆𝐻
extracted by the different analyses, and
falls into the category of a proper analysis.
III. RESULTS: effect of FM stacking order
7
𝑒𝑓𝑓
The effect of stack order on 4𝜋𝑀eff, α, 𝛥𝐻0 and 𝜃𝑆𝐻
has been investigated as a
function of Co and Py thicknesses. Several interesting results are illustrated in Fig. 3(a) -- (d).
Firstly, the magnitude of 4𝜋𝑀eff is in good agreement with the value of Hs
op in the M -- H
hysteresis; the 4𝜋𝑀eff of the (CoPy)-series is lower than that of the (PyCo)-series. The
difference of interface magnetic anisotropy between PtCoPyMgO and PtPyCoMgO stack is
estimated
from
the measured Ms and 4𝜋𝑀eff of ST-FMR
samples, ∆𝐾𝑠 =
𝐾𝑠(𝑃𝑡𝐶𝑜𝑃𝑦𝑀𝑔𝑂) − 𝐾𝑠(𝑃𝑡𝑃𝑦𝐶𝑜𝑀𝑔𝑂) ≈ 0.55 ± 0.03 erg/cm2, which is close to the
∆𝐾𝑠 ≈ 0.6 erg/cm2 obtained with the un-patterned films. The results assure that the ST-FMR
samples were not damaged during the device fabrication and reconfirm that the 𝐾𝑠 in the
PtCoPyMgO stacks are stronger than that in the PtPyCoMgO stacks because of the strong
𝐾𝑠 at the PtCo interface. This difference gives rise to an important consequence in the in-plane
magnetization switching driven by the anti-damping SOT, since the 4πMeff is also one of the
key material parameters in determining Jc0 [see Eq. (1)]. Secondly, for a single FM layer (𝑡Co =
0 or 5 nm), 𝜃𝑆𝐻
𝑒𝑓𝑓(PtCoMgO) ≈ 0.14 > 𝜃𝑆𝐻
𝑒𝑓𝑓(PtPyMgO) ≈ 0.06 while α(PtCoMgO) ≈
0.026 > α(PtPyMgO) ≈ 0.018 as shown in Figs. 3(b) and (d). This result is consistent with
that expected from the spin-transparency model [15], i.e. a higher spin transparency
accompanied with an enhanced magnetic damping.
𝑒𝑓𝑓
Thirdly, for FM bilayers (1 nm ≤ 𝑡Co ≤ 4 nm), the measured 𝜃𝑆𝐻
are still in agreement
with the spin-transparency model in which the interface between the FM and Pt layer mostly
𝑒𝑓𝑓
determines 𝜃𝑆𝐻
. By contrast, the measured α of PtFM-bilayers shows some inconsistency
with
the spin
transparency model. Figure 3(b) shows
that
the α of Pt(5)Py(5-
tCo)Co(tCo)MgO(2) series is more or less the same as the α of Pt(5)Co(5)MgO(2), ≈ 0.026,
8
and the α of Pt(5)Co(tCo)Py(5- tCo)MgO(2) is very close to the Pt(5)Py(5)MgO(2), ≈ 0.018.
Both results show a weak tCo-dependence, indicating that the magnitude of α is more
significantly influenced by the material adjacent to MgO or FMMgO interface than the
material adjacent to Pt or the PtFM interface. This is in discord with the spin-transparency
model where the enhanced damping is mainly determined by the spin-pumping effect occurring
through the PtFM interface. Instead, we found that the magnitude of α has a relation to the
magnitude of 𝛥𝐻0 , which is another magnetic relaxation parameter associated with the
magnetic inhomogeneity of a sample. Fig. 3(c) shows that ∆𝐻0 of the bilayers also
significantly depends on the FM material next to the MgO layer rather than on the FM material
interfacing the Pt layer. The magnitude of ∆𝐻𝑜 was four to five times higher in the (PyCo)-
devices than that of the (CoPy)-devices since we observed ∆𝐻𝑜 (PtPyCoMgO) ≈ 62 ±
4 𝑂𝑒 > ∆𝐻𝑜 (PtCoPyMgO) ≈ 13 ± 6 𝑂𝑒 for 1 ≤ 𝑡𝐶𝑜 ≤ 4 𝑛𝑚.
The results shown above clearly suggest that the FMMgO interface considerably
contributes to the magnitudes of α and 𝛥𝐻0. A remaining question is how large portion of
magnetic damping in the FM bilayers originates from the FMMgO and the PtFM interface
respectively. In order to extract the contribution of the PtFM interface, we have conducted an
additional ST-FMR experiment for almost identical multilayer stacks but without a Ta(1)Pt(5)
buffer layer. The measured α and 𝛥𝐻0 are also plotted in Figs. 3 (b) and (c). The ST-FMR
signal in the devices was barely measurable, and the resonant peaks were distinguishable only
when an inhomogeneous current density existed in the magnetic bilayer that can be excited by
itself. Furthermore, some of the devices did not exhibit any discernible spectra within our
measurement resolution. Hence, the result that we were able to obtain is the averaged
magnitude of α, α(CoPyMgO) ≈ 0.01 and α(PyCoMgO) ≈ 0.012. These are somewhat
9
lower than α(PtCoPyMgO) ≈ 0.018 and α(PtPyCoMgO) ≈ 0.03 respectively, thereby
confirming the increase of the damping due to the PtFM interface as well.
If this enhanced damping solely arises from the spin-pumping effect, the Pt(CoPy)
interface has the effective spin-mixing conductance: 𝑔eff
↑↓ =
4𝜋𝑀𝑠𝑡FM
𝛾ℏ
[α(𝑃𝑡𝐶𝑜𝑃𝑦𝑀𝑔𝑂) −
α(𝐶𝑜𝑃𝑦𝑀𝑔𝑂)] ≈ 21 ± 7 𝑛𝑚−2 . Thus, the spin-transparency is Tsp≈
↑↓
2(𝑒2/ℎ)𝑔eff
1/𝜆Pt𝜌Pt
≈ 0.72 ±
0.24, where 𝜆Pt is the spin-diffusion length in Pt (if using 𝜆Pt ≈ 1 𝑛𝑚 [15,16]) and 𝜌𝑃𝑡 is
the electrical resistivity of Pt (≈ 45 μΩ ∙ cm) on top of the Ta(1) buffer layer. The calculated
Tsp(Pt(CoPy)) gives the intrinsic magnitude of the spin-Hall angle (𝜃SH
𝑖𝑛𝑡) in Pt as large as
0.19 (∵ 𝜃𝑆𝐻
𝑖𝑛𝑡 are in good agreement with the
𝑒𝑓𝑓 = 𝑇𝑠𝑝 ∙ 𝜃SH
𝑖𝑛𝑡) , where our Tsp and 𝜃SH
previously reported [15] Tsp (≈ 0.65) at the interface of PtCo and 𝜃SH
𝑖𝑛𝑡 of Pt (≈ 0.17) although
we had quantitatively different values in 𝑔eff
↑↓ and 𝜌Pt. The same analysis is applied to obtain
of 𝑔eff
↑↓ and 𝑇sp of the Pt(PyCo) interface, but this gives rise to an unphysical consequence:
↑↓ ≈ 32 ± 10 𝑛𝑚−2 and consequently 𝑇sp ≈ 1.1 ± 0.36 . The obtained 𝑔eff
𝑔eff
↑↓ remains
consistently in the range of previous reports: 𝑔eff
↑↓ (PtPy) ≈ 20 − 40 𝑛𝑚−2, but the Tsp larger
than unity is an unphysical fallout.
We note that the estimation of Tsp is uncertain at the moment, mainly due to the variation
in the reported values of λPt by more than one order of magnitude, ranging from 1 to 11 nm, so
that the used value (𝜆Pt ≈ 1 nm) above is possibly to be incorrect. A recent work [21] has
reported an important relation, λPt∝ρPt
-1, in order to reconcile such discrepancies, with the
assumption that Elliott-Yafet (EY) scattering mechanism dominates the spin relaxation in Pt at
RT. The scenario is quite reasonable because another work [ 22 ] has confirmed that EY
10
scattering is dominant in Pt at RT while D'yakonov-Perel' (DP) relaxation dominates at
cryogenic temperatures. The reported spin-resistance (rs,Pt = λPt·ρPt) of Pt is still scattered over
a wide range of magnitude: rs,Pt ≈ 0.6 fΩ·m2 [23], rs,Pt ≈ 0.77 fΩ·m2 [21], and rs,Pt ≈ 2.6 fΩ·m2
[24]. Given the relation λPt∝ρPt
-1, the expected λPt from our Pt films could be, respectively, 1.3
nm, 1.7 nm, and 5.8 nm, resulting in the TSP (Pt(CoPy)) ≈ 0.94, 1.2, and 4.2. Even for the case
of TSP ≈ 0.94, the corrected TSP would be larger than one if taking account of spin-memory-loss
(SML) at the interface [23]. The estimated values of TSP are unphysical regardless of its
variation, since it is greater than unity for all of the cases.
We could understand such unphysical results in two different ways. The first case
happens if the spin current is generated at the PtFM interface rather than in bulk Pt [25,26].
The origin of spin current is still under significant debate in the spintronic community so its
conclusion is beyond our scope. The other is the case when the magnitude of 𝑔eff
↑↓ is
overestimated for PtFM bilayers. By now, most of experimental works for 𝑔eff
↑↓ have assumed
that the enhanced damping with the Pt-interface is solely due to the spin-pumping effect, and
thereby calculated the value of 𝑔eff
↑↓ from the difference in the magnetic damping between two
stacks with and without Pt-interface. However, there might exist additional dissipation
channels for the spin-dynamics such as interfacial spin-flip scattering [27,28] or two-magnon
scattering (TMS) [29,30], originating from the PtFM interface. Ref. [23] has studied the SML
effect at the PtCo interface in which the transmission of spin-current is reduced at the interface
due to the interfacial spin-flip scattering. The origin of SML effect has been attributed to the
magnetic proximity effect in Pt [23] or to the development of magnons due to the interfacial
(non-collinear) Dzyaloshinskii-Moriya interaction [31]. In the same way, such effects should
influence on the dissipation of spin dynamics as well, resulting in the enhancement of magnetic
11
damping.
The possibility of additional relaxation channels at the CoMgO interface are also revealed
by the increased inhomogeneity in Fig. 3(c), where 𝛥𝐻0 (PtPyCoMgO) > 𝛥𝐻0
(PtCoPyMgO). The Co in contact with MgO leads to broadening of resonance peak and, in
other words, additional spin relaxations. Moreover, the (CoPy)-devices without a Ta(1)Pt(5)
buffer, where the Co layer directly interfaces to the SiOx layer, exhibits a larger 𝛥𝐻0 than that
in the (CoPy)-devices with the Ta/Pt buffer layer (see Fig. 3(c)). This implies that the presence
of Co adjacent to oxides or a possible formation of interfacial Co-oxide gives rise to increased
magnetic inhomogeneity and consequently enhanced magnetic damping as well.
An important consequence of our results described so far is that, even with the same
material combinations of Co and Py, the stacking order influences the 4𝜋𝑀eff, α, 𝛥𝐻0 and
𝜃SH, which in turn affect the writing energy required for switching the Co and Py magnetization.
Combining the measured parameters in Figs. 3(a) -- (d), we used Eq. (1) to calculate the
magnitude of Jc0 as a function of tCo for the (CoPy)- and (PyCo)-series (see Fig. 3(e)). The
(CoPy)-series have consistently smaller 𝐽c0's than the (PyCo)-series for the same tCo because
𝑒𝑓𝑓
of combined effects: the former group has smaller 𝛼 (and 𝛥𝐻0) and 4π𝑀eff but larger 𝜃𝑆𝐻
than the latter. The result suggests a possibility that can lead to a reduction of 𝐽c0 by
optimizing the magnetic stack order and the interfaces in contact with the HM and the oxide
layer.
We note that our bilayer approach can be applicable to a perpendicularly magnetized
system as well. Most of experimental studies have utilized magnetic layers of 1-2 nm thickness
12
for SOT-driven magnetic switching, because in the thickness ranges the interfacial energy
becomes sufficient to compensate the volume demagnetization energy, so as to make the system
perpendicularly magnetized or to lower the 4𝜋𝑀eff for an in-plane magnetized system.
Furthermore, the magnitude of SOT becomes enough to reverse the polarity of magnet since it
is proportional to (MS× tFM)-1. Naturally one can control 4𝜋𝑀eff or 𝐻𝑘
𝑒𝑓𝑓
of the CoPy
bilayers by optimizing their relative thicknesses within total 1-2 nm, while minimizing the α
𝑒𝑓𝑓
and maximizing the 𝜃𝑆𝐻
from the stacking order and interface engineering. We also expect
that the interfacial contributions will be more significant as the total thickness decreases and
the FM bilayer becomes perpendicularly magnetized.
IV. DISCUSSION: contribution of FMMgO interface on magnetic damping
As mentioned previously, we presume that there exists a strong correlation between α and
𝛥𝐻0. To further clarify our speculation, we have conducted additional ST-FMR experiments
for the stacks comprising a FM dusting layer at the interface adjacent to the MgO layer. The
devices
in
this study have FM multilayers consisting of:
(Co2Py2Co)-series:
Ta(1)Pt(5)Co(2)Py(2)Co(tCo)MgO(2)Ta(2); (Py2Co2Py)-series: Ta(1)Pt(5)Py(2)Co(2)
Py(tPy)MgO(2)Ta(2); (Co4Py)-series: Ta(1)Pt(5)Co(4)Py(tPy)MgO(2)/Ta(2); and (Py4Co)-
series: Ta(1)Pt(5)Py(4)Co(tCo)MgO(2)Ta(2), between Pt and MgO layers. Both tCo and tPy
were varied from 0 -- 1 nm and the thickness of the FM single or bilayer, which is adjacent to Pt,
was fixed to 4 nm, as illustrated in Figs. 4(a) and 4(b).
Figures 4 (c) -- (f) show the obtained 𝛼 and 𝛥𝐻0, as functions of tCo or tPy, in which one can
see apparent effects of the FM dusting layer. First, the insertion of Py dusting layer in between
13
Co and MgO leads to a reduction of 𝛼 and 𝛥𝐻0. For example, as we increase the thickness of
Py dusting layer (tPy) from 0 nm to 1 nm in the devices of the (Py2Co2Py)- and (Co4Py)-
series (solid shape in the figures), the 𝛼 is gradually reduced from 0.034 to 0.016 for the
Py2Co2Py(tPy) series, and from 0.029 to 0.015 for the Co4Py(tPy) series (Fig. 4(c) and (d));
the 𝛥𝐻0 is reduced from 90 Oe to 10 Oe for the Py2Co2Py(tPy) series and from 70 Oe to 20
Oe for the Co4Py(tPy) series (Fig. 4(e) and (f)). Second, the insertion of Co dusting layer in
between Py and MgO leads to an enhancement of 𝛼 and 𝛥𝐻0. For instance, as we increase the
thickness of Co dusting layer (tCo) from 0 nm to 1 nm in the (Co2Py2Co)- and (Py4Co)-series
(cross shape in Figs. 4(a) and (b)), the 𝛼 is gradually enhanced from 0.02 to 0.029 for the
Co2Py2Co(tCo) series, and from 0.022 to 0.029 for the Py4Co(tCo) series (Fig. 4(c) and (d));
the 𝛥𝐻0 is increased from 20 Oe to 30 Oe in the Co2Py2Co(tCo) series, and from 2 Oe to 60
Oe in the Py4Co(tCo) series (Fig. 4(e) and (f)).
Our experimental results described above clearly demonstrate that α and 𝛥𝐻0 are
strongly correlated. Figure. 5(a) shows the plot of 𝛼 vs 𝛥𝐻0 determined from all our ST-FMR
devices having FM bi- or tri-layers. For the samples with a PyMgO interface (solid symbols),
𝛥𝐻0 is mostly distributed below 20 Oe, and the 𝛼 remains as small as 0.02. For the samples
with a CoMgO interface (open symbols with a cross in the figure), the 𝛥𝐻0 is distributed in
a wide range from 0 to 60 Oe, and the 𝛼 is enhanced quasi-linearly with increasing 𝛥𝐻0. The
increase of 𝛥𝐻0 has been known to be related to the samples' (magnetic) inhomogeneity. The
enhanced 𝛼 associates with the increased 𝛥𝐻0 definitely indicating that a new magnetic
relaxation channel is developed at the CoMgO interface.
14
To the best of our knowledge, there is no theoretical model that directly connects ∆𝐻0 to
α. An important clue observed from our experiment is fact that both ∆𝐻0 and α are increased
with decreasing thickness (tFM) of an ultrathin FM film; i.e., α and ∆𝐻0 ∝ 1/𝑡FM
𝑛 in general.
In order to understand this, we have checked several mechanisms such as spin-pumping, spin
scattering, and TMS that might be account for the increased ∆𝐻0 and α. First, the spin-
pumping effect [17] occurs when a FM is interfacing to a strong spin-scatterer (e.g., Pt), but
this is not the case for our Co oxide interface. Second, the spin scattering can be increased with
decreasing tFM because the surface contribution [27,28] significantly increases the electron
scattering rate (1/τs) and consequently the spin-flip scattering rate (1/τsf). However, this
mechanism is not in accordance with our observation in which the measured resistivities of FM
layers from all ST-FMR devices exhibit no clear correlation to α (see Fig. 5(b)). Third, a
probable mechanism that might be related to our experiment is the TMS [29,30], which
contributes to the magnetic damping of ultrathin ferromagnetic films in which the uniform
mode (k = 0) is excited by ST-FMR scatters into degenerated magnons (𝑘 ≠ 0) due to surface
roughness or defects, which is strongly related to the magnetic disorder.
The antiferromagnetic (AF) formation of interfacial magnetic oxides (CoO, NiO and Fe2O3)
can be attributed to the source of dynamic non-uniformity and enhanced magnetic damping via
TMS process. The interfacial AF layer can change the magnetic behavior of the FM with the
introduction of extra anisotropies via exchange-bias effects. A distribution of grains in the AF-
oxide induces anisotropy fluctuation in random directions. Thereby the AF-oxide can open an
additional relaxation pathway and spatially non-uniform dynamics during the magnetic
precession through the exchange coupling to the fluctuating spins of AF-oxide and by the slow
dragging of AF-oxide domains.
15
We presume, however, that it is unlikely that possible interfacial oxides in our stacks have
an antiferromagnetic order at RT. The exchange bias becomes effective when the AF-oxide
becomes block magnetically, i.e. when the temperature is below the blocking temperature (TB).
The TB's of bulk CoO and NiO are known to be ≈ 290 K and ≈ 470 K respectively [32], but the
ones in ultrathin film are often very different from the bulk values for AF-oxides [33]. For
instance, 2.5 nm-thick NiO has TB ≈ 200 K [34] and in general the TB decreases with decreasing
the thickness of the AF-oxide. Our interfacial CoO, NiO, and Fe2O3 layers should have much
lower TB's than RT because their effective thicknesses are expectedly ~0.6-0.8 nm, estimated
from the observation in Fig. 4 (c)-(d) at which the damping becomes saturated. Moreover, the
TB of CoO is expected to be even lower than the one of NiO, so that the Py-MgO interface
should have more substantial interfacial AF-oxide than the Co-MgO interface. This is in
opposition to our observation.
Nevertheless, it would be useful study FMR-linewidths as functions of measurement and
annealing temperatures. The strength of exchange bias from the interfacial AF-oxide is strongly
dependent on the microstructural properties of both FM and MgO such as grain size distribution,
structural defect, interfacial roughness, enthalpy of formation and variation in chemical
composition. Advanced interface chemical analysis such as X-ray photoemission spectroscopy
(XPS), electron energy loss spectroscopy or polarized-neutron reflectometry can be useful tools
to investigate the microstructural properties of both FM and MgO as well. Furthermore, in
order to confirm the development of TMS at the CoMgO interface, further experiments are
required such as the measurement of 𝛥𝐻 as functions of wide ranges of angle and frequency
[ 35 ].The results will provide better explanation about the physical origins of linewidth
16
broadening at the CoMgO interface.
The bilayer approach might give one possible strategy that can reduce the switching current
in the STT/SOT-MRAM technology. In a typical MRAM configuration, CoFeB-alloy has been
utilized as a magnetic free layer that interfaces to MgO-tunnel barrier. We note that Py with
MgO tunnel barriers does not provide a good tunnel magneto-resistance (TMR), which makes
the proposed configuration be ineffective in the electrical readout of the magnetic change. If
the oxidation of Co plays an important role in the interfacial enhancement of damping, the
damping could be reduced by the insertion of a magnetic dusting layer, for instance Fe or FeB,
between CoFeB and MgO layers. This may be an alternative way for lowering the write energy
as long as the interface engineering does not damage other parameters, such as TMR, thermal
stability and spin-polarization. We believe our results suggest a practical strategy in the
optimization of SOT and magnetic damping by engineering both interfaces in contact with the
HM and the tunnel-barrier layer, for achieving more energy-efficient MRAM.
V. SUMMARY
In summary, we utilized Co/Py FM-bilayers, which are in-plane magnetized and
sandwiched between Pt and MgO layers, as test-beds for understanding the roles of interfaces
in the spin transparency and understanding the interfacial contributions on the magnetic
damping. Our results demonstrated that the magnitude of θSH
eff is mostly determined by the
interfacial material in contact with the Pt layer, which is consistent with the spin transparency
model. By contrast, the magnetic relaxation parameters, α and 𝛥𝐻0 , are substantially
17
influenced not only by the PtFM interface but also by the spin relaxation such as TMS at the
FMMgO interface. Both α and 𝛥𝐻0 are significantly increased at the CoMgO interface
probably via increased TMS processes whereas such increases are negligible at the PyMgO
interface. In order to achieve a low writing current density in SOT-driven magnetic switching,
the multilayer configuration of PtCoPyMgO has more preferable material parameters than the
stack of PtPyCoMgO.
Acknowledgements
This work was supported by the National Research Council of Science & Technology (NST)
grant (No. CAP-16-01-KIST) and the KIST Institutional Program (2E28190). K. -J. L. was
supported by the National Research Foundation of Korea (NRF) [NRF-2017R1A2B2006119]
and KU-KIST School Project. K. -J. L. acknowledges the KIST Institutional Program (Project
No. 2V05750)
18
Figure Captions
FIG. 1. (a) Multilayer structures of (CoPy)- and (PyCo)-series. The Co layer thickness (tCo)
was varied from 0 to 5 nm while the total thickness of the FM bilayers was fixed to 5 nm. (b)
Schematic of ST-FMR measurement. (c) Saturated magnetization (MS) and (d) Hs
op (out-of-
plane saturation field) for un-patterned films of (CoPy)- and (PyCo)-series.
19
FIG. 2. (a)-(d) Representative spectra for ST-FMR devices with Py(5), Co(5), Py(4)Co(1),
and Co(1)Py(4) at 8 GHz, exhibiting different resonant positions and linewidths. Red curves
are fit to Eq. (1) (red curve). The symmetric (green) and anti-symmetric (blue) parts of the
signals are also plotted. (e) Obtained 𝐻res and (f) ∆𝐻 as a functions of f along with fitting
curves.
20
FIG. 3. Measurement results of (a) 4𝜋𝑀eff, (b) α, (c) 𝛥𝐻0 and (d) 𝜃𝑆𝐻
𝑒𝑓𝑓 as functions of
stacking order and thickness of Co and Py. (e) Calculated Jc0 as a function of tCo for (CoPy)-
and (PyCo)-series by using Eq. (1) with the measured parameters in (a) -- (d).
21
FIG. 4. Multilayer structure of (a) (Co2Py2Co)-, (Py2Co2Py)-, (b) (Co4Py)-, and (Py4Co)-
series. Additional FM dusting layer was inserted at the interface of FMMgO and its thickness
was varied from 0 to 1 nm. The thickness of the initial FM layer was fixed to 4 nm. (c) -- (f)
Obtained 𝛼 and 𝛥𝐻0 as functions of tCo or tPy.
22
FIG. 5. (a) Plot of 𝛼 vs 𝛥𝐻0 of devices having FM bi- or tri-layers. Data for the samples with
the CoMgO interface are represented by cross shapes, and data for devices with the PyMgO
interface is by solid shapes. (b) Plot of ρ (the resistivity of FM bi- or tri-layers) vs 𝛼.
23
References
1 D. C. Ralph and M. D. Stiles, Spin transfer torques, J. Magn. Magn. Mater. 320, 1190
(2008).
2 J. Sinova, Sergio O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, Spin Hall
effects, Rev. Mod. Phys. 87, 1213 (2015).
3 F. Hellman et al., Interface-induced phenomena in magnetism, Rev. Mod. Phys. 89, 025006
(2017).
4 A. Hoffmann, Spin Hall Effects in Metals, IEEE Trans. Magn. 49, 5172 -- 5193 (2013).
5 I. M. Miron et al., Perpendicular switching of a single ferromagnetic layer induced by in-
plane current injection, Nature 476, 189 (2011).
6 L. Liu, O. J. Lee, T. J. Gudmundsen, D. C. Ralph, and R. A. Buhrman, Current-Induced
Switching of Perpendicularly Magnetized Magnetic Layers Using Spin Torque from the Spin
Hall Effect, Phys. Rev. Lett. 109, 096602 (2012).
7 A. Brataas, A. D. Kent, and H. Ohno, Current-induced torques in magnetic materials, Nat.
Mater. 11, 372 -- 381 (2012).
8 P. M. Haney, R. A. Duine, A. S. Nunez, and A. H. MacDonald, Current-induced torques in
magnetic metals: Beyond spin-transfer, J. Magn. Magn. Mater. 320, 1300 (2008).
9 K.-S. Lee, S.-W. Lee, B.-C. Min, and K.-J. Lee, Threshold current for switching of a
perpendicular magnetic layer induced by spin Hall effect, Appl. Phys. Lett. 102, 112410 (2013).
10 O. J. Lee, L. Q. Liu, C. F. Pai, Y. Li, H. W. Tseng, P. G. Gowtham, J. P. Park, D. C. Ralph,
R. A. Buhrman, Central role of domain wall depinning for perpendicular magnetization
switching driven by spin torque from the spin Hall effect, Phys. Rev. B 89, 024418 (2014).
11 M. Baumgartner et al., Spatially and time-resolved magnetization dynamics driven by
spin -- orbit torques, Nat. Nano. 12, 980 (2017).
12 D. MacNeill, G. M. Stiehl, M. H. D. Guimaraes, R. A. Buhrman, J. Park and D. C. Ralph,
Control of spin -- orbit torques through crystal symmetry in WTe2/ferromagnet bilayers, Nat.
Phys. 13, 300 (2017).
13 S. C. Baek, V. P. Amin, Y. W. Oh, G. Go, S. J. Lee, M. D. Stiles, B. G. Park, K. J. Lee,
Spin-orbit torques induced by interface-generated spin currents, arXiv:1708.06864 (2017).
14 K. Demasius, T. Phung, W. Zhang, B. P. Hughes, S.-H Yang, A. Kellock, W. Han, A.
Pushp, and S. S. P. Parkin, Enhanced spin -- orbit torques by oxygen incorporation in tungsten
films, Nat. Comm. 7, 10644 (2016).
15 W. Zhang, W. Han, X. Jiang, S.-H. Yang and S. S. P. Parkin, Role of transparency of
platinum -- ferromagnet interfaces in determining the intrinsic magnitude of the spin Hall effect,
Nat. Phys. 11, 496 (2015).
24
16 C.-F. Pai, Y. Ou, L. H. Vilela-Leão, D. C. Ralph, and R. A. Buhrman, Dependence of the
efficiency of spin Hall torque on the transparency of Pt/ferromagnetic layer interfaces, Phys.
Rev. B 92, 064426 (2015).
17 Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer, Spin pumping and magnetization
dynamics in metallic multilayers, Phys. Rev. B 66, 224403 (2002).
18 B. Dieny and M. Chshiev, Perpendicular magnetic anisotropy at transition metal/oxide
interfaces and applications, Rev. Mod. Phys. 89, 025008 (2017).
19 L. Liu, T. Moriyama, D. C. Ralph, and R. A. Buhrman, Spin-Torque Ferromagnetic
Resonance Induced by the Spin Hall Effect, Phys. Rev. Lett. 106, 036601 (2011).
20 J. H. Kim, D. J. Lee, K.-J. Lee, B.-K. Ju, H. C. Koo, B.-C. Min and O. J. Lee, Spin-orbit
torques associated with ferrimagnetic order in Pt/GdFeCo/MgO layers, Sci. Rep. 8, 6017
(2018).
21 M.-H. Nguyen, D. C. Ralph, and R. A. Buhrman, Spin Torque Study of the Spin Hall
Conductivity and Spin Diffusion Length in Platinum Thin Films with Varying Resistivity,
Phys. Rev. Lett. 116, 126601 (2016).
22 R. Freeman, A. Zholud, Z. Dun, H. Zhou, and S. Urazhdin, Evidence for Dyakonov-Perel-
like Spin Relaxation in Pt, Phys. Rev. Lett. 120, 067204 (2018).
23 J.-C. Rojas-Sánchez et al., Spin Pumping and Inverse Spin Hall Effect in Platinum: The
Essential Role of Spin-Memory Loss at Metallic Interfaces, Phys. Rev. Lett. 112, 106602
(2014).
24 C. Stamm, C. Murer, M. Berritta, J. Feng, M. Gabureac, P. M. Oppeneer, and P.
Gambardella, Magneto-Optical Detection of the Spin Hall Effect in Pt and W Thin Films,
Phys. Rev. Lett. 119, 087203 (2017).
25 L. Wang, R. J. H. Wesselink, Y. Liu, Z. Yuan, K. Xia, and P. J. Kelly, Giant Room
Temperature Interface Spin Hall and Inverse Spin Hall Effects, Phys. Rev. Lett. 116, 196602
(2016).
26 V. P. Amin and M. D. Stiles, Spin transport at interfaces with spin-orbit coupling:
Formalism, Phys. Rev. B 94, 104419 (2016).
27 F. Warkusz, Size Effects in Metallic Films, Electrocomponent Sci. Technol. 5, 99 (1978).
28 S. Ingvarsson, L. Ritchie, X. Y. Liu, Gang Xiao, J. C. Slonczewski, P. L. Trouilloud, and
R. H. Koch, Role of electron scattering in the magnetization relaxation of thin Ni81Fe19
films, Phys. Rev. B 66, 214416 (2002).
29 R. Arias and D. L. Mills, Extrinsic contributions to the ferromagnetic resonance response
of ultrathin films, Phys. Rev. B 60, 7395 (1999).
30 K. Lenz, H. Wende, W. Kuch, K. Baberschke, K. Nagy, and A. Jánossy, Two-magnon
scattering and viscous Gilbert damping in ultrathin ferromagnets, Phys. Rev. B 73, 144424
(2006).
25
31 J. H. Oh, I.-S. Hong, and K.-J. Lee, Dzyaloshinskii-Moriya interaction induced extrinsic
linewidth broadening of ferromagnetic resonance, Phys. Rev. B 97, 014430 (2018).
32 P. Grünberg, Layered magnetic structures: facts, figures, future, J. Phys.: Condens. Matter
13 7691 (2001).
33 M. Molina-Ruiz, A. F. Lopeandía, F. Pi, D. Givord, O. Bourgeois, and J. Rodríguez-Viejo,
Evidence of finite-size effect on the Néel temperature in ultrathin layers of CoO nanograins,
Phys. Rev. B 83, 140407(R) (2011).
34 O. Ozatay et al., Sidewall oxide effects on spin-torque- and magnetic-field-induced
reversal characteristics of thin-film nanomagnets, Nat. Mat. 7, 567 (2008).
35 Kh. Zakeri et al., Spin dynamics in ferromagnets: Gilbert damping and two-magnon
scattering, Phys. Rev. B 76, 104416 (2007).
26
|
1906.09396 | 1 | 1906 | 2019-06-22T06:17:50 | Universal coupling between the photonics and phononics in a 3D graphene sponge | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Photon-phonon coupling holds strong potential for sound and temperature control with light, opening new horizons in detector technology, remote sound generation and signal broadcasting. Here, we report on a novel stereoscopic ultralight converter based on a three dimensional graphene structure 3G-sponge, which exhibits very high absorption, near-to-air density, low inertia, and negligible effective heat capacity. We studied the heat and sound generation under the excitation of electromagnetic waves. 3G-sponge shows exceptional photon to heat and sound transduction efficiency over an enormous frequency range from MHz to PHz. As an application, we present an audio receiver based on a 3G-sponge amplitude demodulation. Our results will lead to a wide range of applications from light-controlled sound sources to broadband high-frequency graphene electronics. | physics.app-ph | physics | Title: Universal coupling between the photonics and phononics in a 3D
graphene sponge
M. Shalaby1,2,†, C. Vicario3, F. Giorgianni3,4, M. A. Gaspar3, P. Craievich3, Y. Chen5, B. Kan5, S.
Lupi6 and C. P. Hauri3
•
•
•
•
•
•
1Beijing Advanced Innovation Center for Imaging Technology and Key Laboratory of Terahertz
Optoelectronics, CNU, Beijing 100048, China
2Swiss Terahertz Research-Zurich, Technopark, 8005 Zurich, Switzerland and Park Innovaare,
5234 Villigen, Switzerland
3Paul Scherrer Institute, 5232 Villigen, Switzerland
4Center for Life Nano Science@Sapienza, Istituto Italiano di Tecnologia, V.le Regina Elena 291,
I-00186, Roma, Italy; and Department of Physics, Sapienza University of Rome, Piazzale Aldo
Moro 2, I-00185 Roma, Italy.
5State Key Laboratory and Institute of Elementog-Organic Chemistry, Nankai University, Tianjin
300071, China
6Graphene Labs, Istituto Italiano di Tecnologia, Via Morego 30, 16163 Genova, Italy
† Corresponding authors: [email protected]
Abstract: Photon-phonon coupling holds strong potential for sound and temperature control with
light, opening new horizons in detector technology, remote sound generation and signal
broadcasting. Here, we report on a novel stereoscopic ultralight converter based on a three
dimensional graphene structure 3G-sponge, which exhibits very high absorption, near-to-air
density, low inertia, and negligible effective heat capacity. We studied the heat and sound
generation under the excitation of electromagnetic waves. 3G-sponge shows exceptional photon
to heat and sound transduction efficiency over an enormous frequency range from MHz to PHz.
As an application, we present an audio receiver based on a 3G-sponge amplitude demodulation.
Our results will lead to a wide range of applications from light-controlled sound sources to
broadband high-frequency graphene electronics.
One Sentence Summary: We present a universal photon-phonon coupling in a 3D graphene
sponge for efficient heat and sound emission.
Main Text:
Introduction: Photons and phonons are both energy quanta that can be associated with waves
oscillating in time and space at various frequencies. However, their generation, propagation, and
interaction with matter are fundamentally different, respectively constituting photonic
(electromagnetic, EM) and phononic spectra. While the phonons carry sound through mechanical
vibrations and require the presence of a medium, the EM waves propagate even in vacuum.
Photon-phonon coupling control is an exciting but rather an unexplored route to manipulate heat
and sound emission with light. The process involves the physical transfer of energy through
vibrations. This makes sound waves propagate much further than heat waves. Light-controlled
phononics is foreseen to open up a broad range of novel applications including electromagnetic
detection, remote sound generation technologies and broadband signal processing [1].
The efficiency of energy transfer from EM waves to sound and heat (photon-phonon coupling) is
the present main obstacle for these applications. Sound and heat can be coupled and the high
frequency heat energy (THz range) can be converted into sound energy (kHz range). Such
coupling is complex. Elasticity and inertia are the most relevant medium properties for sound
and mechanical wave generation.
Here we present the 3D graphene sponge (3G-sponge) as a distinguished material for a universal
coupling between photons and phonons. 3G-sponge is a newly-developed material [2, 3] that
shows intriguing properties such as super-compressive elasticity and near zero Poisson's ratio
[4]. These properties are prerequisites for a highly efficient photon-phonon coupling [5]. Recent
studies of mono-layer (i.e. 2D) graphene have reported the outstanding molecular properties in
view of thermal conductivity [6], charge carrier mobility and Young' modulus [7, 8, 9]. But, so
far, little attention has been paid to the equivalent macroscopic properties of 3D graphene, as
stacking graphene sheets to form a bulk has always compromised the interesting properties of 2D
graphene. It is only very recently that Wu et al. succeeded in developing the graphene foam [4], a
bulk 3D free-standing graphene sponge structure - the lightest solid material on earth.
Results: By exploiting its unique properties, we present the 3G-sponge as a universal coupler
between light, sound and heat with exceptional efficiency. The proposed concept of photon-
phonon coupling is shown in Fig. 1. We explored two different excitation schemes of the 3G-
sponge with a time-continuous EM wave (CW) and amplitude-modulated (AM) stimuli. Due to
the high absorption capability [10] and the negligible heat capacity of the 3G-sponge [3-4] the
excitation with a CW wave leads to a rapid rise in temperature.
To prove this fundamental coupling concept experimentally, we started our investigation with
the CW microwave radiation source shown in Fig. 1A. To reduce free-space losses the 3G-
sponge is directly attached to the coaxial cable and is irradiated by an EM sine wave with an
average power of up to 1 W and GHz frequency. Upon EM excitation, we observed a fast rise of
the 3G-sponge temperature, which rapidly reaches a steady state. The final temperature was
studied for different excitation wave frequencies (0.5- 4 GHz, Fig. 1B). We found linear
dependence on the average power that the rise in temperature increases as the excitation
frequency decreases (Fig. 1B). For the lowest excitation frequency (0.5 GHz) the 3G-sponge
heats up beyond 150°C, which represents the maximum temperature limit of our thermographic
camera. The fast rise in temperature originates from the remarkable absorption properties of the
3G-sponge to the used microwave [10].
While the CW stimulus gives rise to extensive heat emission of the 3D-sponge, no
emission of sound could be observed. The physical scenario changed when the CW stimulus was
modulated in intensity. For this, the CW GHz source was modulated in amplitude with a 1-50
kHz sine wave (experimental setup shown in Fig. 1C). The sponge illuminated with such an AM
stimulus reproduces a thermal wave modulated at the same frequency. However, direct
recording of such a fast heat transient is technically challenging. Instead, we experimentally
observed an audible acoustic signal from the sponge caused by the modulated heat wave. Indeed,
using a conventional microphone, we could measure a powerful acoustic signal emitted from the
3G-sponge at precisely 5 kHz frequency (Fig. 1D), in correspondence to the 5 kHz EM stimulus
modulation frequency. This surprising result demonstrates that the sound generation is directly
linked to photon absorption, leading to temperature transients on a fast, microsecond timescale.
The corresponding emitted sound wave is shown in black in Fig. 1D) together with the
modulating signal (blue). Sound emission occurs exactly at a p/2 phase shift with respect to the
EM stimulus. This is in line with the expectations that the level of sound is a function of the
derivative of the EM intensity signal. From these data, it is evident that the emitted sound wave
follows the temporal dynamics of the modulated EM stimulus. It is worth mentioning that this
efficient coupling mechanism enabled loud sound generation which was easily perceptible by the
human ear.
Graphene-based materials are known for their extremely wide spectral absorption [11].
Therefore, we expanded our study on sound emission and heating from the microwave (GHz) to
the terahertz (THz) [12, 13] and optical (PHz) frequency ranges. For this, we used a pulsed laser
stimulus. The pulsed THz source (see Material and Methods section) provides a sub-picosecond
single-cycle pulse with main spectral contents located in the 1-12 THz range at a repetition rate
of 100 Hz [12]. To study spectrally- dependent response, we used different THz low pass filters
(LPFs) with cut-off frequencies at 2, 3, 4.2, 6, 10 and 20 THz, respectively. In this measurement,
the 3G-sponge is placed at the focus of the THz beam [12]. The audio signal level scale linearly
with the THz pulse energy and it is independent of the THz spectrum. The microphone output
signals obtained with different THz spectra clearly show the pulsed time structure of the THz
source (Fig. 2A lower graph). Similar emission of sound and heat is also triggered by a pulsed
femtosecond laser source operating at carrier frequencies in the optical regime (nph=200 THz,
Eph=0.83 eV) as shown in Fig. 2B and by a microwave source (Fig. 2C). The thermal image of
the 3G-sponge irradiated by the THz beam , Fig. 2E, shows that the heat emission and
consequently the sound generation are localized within the sub-mm THz focus where the
temperature increases by 3.1 °C. Similar to the microwave measurements, we observed a linear
dependence of the sound signal and heating on the THz and PHz pulse energy impinging on the
3G-sponge.
A summary of the acoustic spectral output for different stimuli is given in Fig. 2D.
Independent of the laser stimulus central frequency, our experiment shows the emission of
broadband sound spectra (0.1-50 kHz) with minor qualitative differences. This suggests that the
coupling between the EM waves to sound in the 3D-sponge is nearly independent of the pulsed
stimulus central frequency. This establishes graphene sponge as a universal coupling medium
between phonons and photons across an extremely large frequency range from the far infrared to
the visible. It allows for efficient heat and sound generation by simple illumination with an
amplitude-modulated light beam. While the reported phenomenon of sound emission relies on
transient photon absorption and heating, we stress that the sound generation mechanism observed
in our experiments depends mainly on the pulse energy rather than the peak power. Therefore, a
high repetition rate pulsed laser or a modulated continuous wave THz source are better suited for
sound emission applications.
Discussion: The physical origin of the ultra-broadband coupling mechanism depends on the
stimulus' center frequency. 3G-sponge can be considered as interlaced carbon sheets. Optical
phonons in graphene occur at high frequency (~6 µm, 50 THz) [11,14-17]. This optical phonon
which is not IR active in single layer carbon (not dipole active at the Gamma point) turns active
in bilayer and multilayered carbon (shows net dipole at the Gamma point). Due to the high
excitation frequency there is no direct path of excitation of such a phonon with a microwave and
far infrared stimulus. The photons in this low-frequency THz range are rather absorbed by Dirac
fermions, i.e. high mobility electrons. After absorption the electron-phonon scattering process
during which the energy gained by electrons is transferred rapidly to phonons takes place on the
order of 10-100 ps [15]. In graphene, this electron-phonon scattering process is efficient. It was
recently shown that at high THz electric field, significant fraction of carriers can go beyond the
(Dirac point) phonon energy (0.2 eV) [15]. This leads to a corresponding increase in the optical
phonon population which consequently results in a reduction in electron mobility, THz
absorption, and sound generation [17]. For a stimulus at optical and near infrared frequency, the
3G-sponge still shows perfect absorption but the underlying energy transfer mechanism is
different. At photon energies 𝐸"# > 2 𝜇'=400 𝑚𝑒V (𝜇': the chemical potential) electronic
electrons and absorption from interband transitions which is already happening at 𝜔>2𝜇'
(Pauli blocking). In 3G-sponge, one expects that 𝐸2 is nearly zero, implying a strong reduction
of free electron absorption and interband transitions already happening around 𝜔 close to DC. In
this regard, while 2𝜇' = 400 meV seems to be a good approximation, its exact value depends on
interband transitions are excited. The absorbed energy is dissipated by the relaxation of the
excited carriers and emission of optical phonons [16, 17]. The photon-electron absorption in
graphene depends on the Fermi energy. For a doped system, one finds absorption from free Dirac
the specific doping of graphene. We stress that the above-mentioned picture is based on the
studies of doped graphene layers. 3G-sponge is a complex disordered insulator which implies
modified and perhaps different thermodynamics. Yet, the most probable interpretation is that the
microwave absorption depends mainly on the free charge carriers already present in the systems
[4]. The photon-phonon interaction at higher frequencies (such as n-IR) is mediated by electrons.
During the sound generation, the thermal barrier is not expected to play a major role as the sound
generation occurs on a much slower time scale (microsecond) than the heating (picoseconds).
Despite the difference in the underlying photon absorption mechanisms in 3G-sponge
below and above the 3G-sponge bandgap our measurements confirm the universality of the heat
and sound generation across the extremely large electromagnetic frequency range from GHz to
PHz. We note that carbon nanotubes have been used in the past for ultrasound generation and
that graphene sheets have been demonstrated as thermo-acoustic transducers [18]. The advance
of our work is based on the unique physical properties of the 3G-sponge structure, which enables
universal absorption and coupling of the EM radiation to phonon and sound generation with
unprecedented high efficiency. From the application point of view, the characteristic time scale
of light conversion into sound is an important parameter. First, light-heat conversion is mediated
by electron-phonon scattering occurring on the sub-picosecond time scale [15-17]. Second, heat-
sound generation occurs through modulation of the air pressure inside the sponge taking place. In
our measurements, the maximum sound transient time was around 10 microseconds limited by
the spectral response of our microphone. Although it is likely shorter than that, we believe that
this is close to the physical limits due to the inherent mechanical characteristics of the process.
We measured the conversion efficiencies our sample to be 3.3x10-4, 6.9x10-4, and 6.0x10-4 in the
GHz, THz and optical regimes (1.5 µm), respectively. This is 41 times larger than that of the
state-of-the-art MWCNT (1.7x10-5) light-sound converter [19, 20].
As a first application towards a useful device, we employed the properties of the 3G-
sponge for the realization of an audio receiver based on an amplitude demodulation at
microwave frequencies. The standard communication setup used for this experiment is shown in
Fig. 3A and is described in details the Materials and Methods section. Surprisingly, the 3G-
sponge was able to perform the demodulation process and generate audible sound. For a
demodulator, it is useful to show the characteristic current -- voltage (I-V) characteristic (Fig. 3B).
The I-V for the 3G-sponge shows a perfect symmetry around the zero crossing point. This
characteristic is advantageous over the conventional electronic diodes where the I-V diagram is
highly asymmetric, requiring a bias voltage to be applied. Second, the I-V correlation shows a
linear component for lower voltages and a cubic term for higher voltages while the power-
voltage characteristic (Fig 3B lower plot) is quadratic. In the experiment the modulated signals
create thermal oscillations at the 3G-sponge-air interface, which are then converted to sound
waves. The demodulated signal (acoustic wave) successfully preserves the quality and
information of the modulating one. Due to the symmetric even power-voltage characteristic
combined with the demodulation scheme, the 3G-sponge-based demodulator can effectively
suppress odd order intermodulation. This leads to very high fidelity in the demodulation since
spurious components are suppressed. Figure 3C demonstrates that the original signal Vsignal(t)
and the demodulated acoustic one Vsound(t) present identical spectral characteristics. It is worth
noting that the presented innovative concept can be applied for the realization of universal and
simple demodulation and detection devices. The results of the experiments performed both at
optical and THz/microwave frequencies show that this demodulation scheme is independent of
the frequency of the carrier. Moreover it can be applied to high frequencies without dedicated
advanced electronics. This makes the 3G-sponge a promising material for next generation
graphene electronics.
In conclusion our results show a direct and efficient (nearly 41 times larger efficiency than
present photo-thermo-acoustic converters) universal coupling between the photonic and
phononic spectra in a novel 3D graphene sponge which offers unique elasticity and
compressibility. This makes the graphene sponge a very good EM absorber and broadband
detector from microwave to optical frequencies. The remarkable coupling in 3G-sponge between
light, heat and sound is well-suited for remote heat and sound emission by light. Such properties
allowed us to demonstrate a novel heat-mediated frequency demodulator scheme in the
microwave range. These unique properties open new opportunities for next-generation graphene
electronics and trigger novel research activities for EM wave detection, as well as efficient sound
and heat devices controlled by an amplitude-modulated light beam.
Materials and Methods:
THz source: Terahertz pulses at 100 Hz repetition rate were produced by optical rectification of
an intense mid-infrared femtosecond laser source in an organic crystal. See ref. [12,13] for
further details. The THz low pass filters are multi mesh and commercially available.
Setup of the standard communication system (Fig. 3A): We set up a standard communication
system where the lower frequency signal 𝑠(𝑡), corresponding to the information to be
transmitted, is carried on a higher frequency carrier 𝑓7 allowing for long distance propagation.
The carrier is also modulated in amplitude with a square wave 𝑠(𝑡) at 40 Hz with a modulation
where we superimposed another high frequency carrier of slightly different frequencies 𝑓7 + ∆𝑓
to the carrier at 𝑓7. ∆𝑓 is in the useful part of the frequency spectrum for the high-frequency to
index of 70% and a duty cycle of 25%. In our experiment, we adapted a demodulation scheme
phononic conversion. The schematic of the transmitter and receiver circuits used for producing
the amplitude modulation (AM) and the demodulation are shown in Fig. 3A. All the devices are
connected with coaxial cables and the sample of the 3G-sponge was placed in the middle of an
N-type connector. The position of the 3G-sponge was adjusted to assure an adequate matching of
the forward microwave and to guarantee nearly complete absorption of the EM energy.
Transmitter oscillator and analog AM modulator: Rohde&Schwarz SMA 100 A, CW signals
with a frequency range from 9 kHz to 6 GHz. We used the frequency of 1 GHz for the carrier f0.
Function generator: TTi TG1010A. We used it to generate the square wave s(t) at 40 Hz with a
modulation index of 70% and a duty cycle of 25%.
Local oscillator: Rohde&Schwarz SMA 100 A, frequency range from 9kHz to 6GHz. We used
the frequency of 1 GHz to demodulate the carrier f0 and a Df in the range between 0.1 kHz to 100
kHz.
Graphene sponge production: The graphene 3D sponge was fabricated by an in situ solvothermal
process using ethanol solvent for low concentrations of Graphene Oxide (GO) sheets (~20 to
50 μm lateral dimension). The low concentration GO ethanol solution (0.20 -- 5.00 mg ml−1) was
then solvothermally treated in a Teflon-lined autoclave at 180 °C for 12 h to form an
intermediate solid, having about 1/3 to 1/2 of the former GO solution volume. After the
solvothermal reaction, the ethanol-filled sponge was removed from autoclave and completely
immersed in a mixture of acetone and ethanol (1:1 in volume). Then water was added to the
system in order to substitute the ethanol with water. This step was then repeated about 15 -- 20
times taking ~6 h for each cycle and up to 5 -- 7 days for the whole process. After the solvent
exchange process, the water-filled sponge was freeze-dried to remove the remaining water
absorbed. Finally, the sample was annealed at 400 °C for an hour in argon to obtain the final 3D
graphene sponge as measured in the photoacoustic experiments.
Thermal camera is a FLIR T420. The camera provides resolution of 320x240 pixel. Its accuracy
is ±2% or 2°C -IR -- The thermal Sensitivity <0.045°C over the temperature range between -20°C
and 150°C.
The microphone is an MK202 from Microtech Gefell and is calibrated in frequency from 20 Hz
to 40 kHz. The microphone is mounted on the MV210 preamplifier from the same company
(Microtech Gefell). The signal is digitalized in a PXI National Instruments acquisition card.
References and Notes:
1. M. Maldova, Sound and heat revolutions in phononics, Nature 503, 209 (2013).
2. J. Zhao, W. Ren, and H. -- M. Cheng, J. Mater. Chem., 22, 20197 (2012).
3. T. Zhang et al., Nat. Photonics 9, 471 (2015).
4. Y. Wu et al., Nat. Commun. 6, 7141 (2015).
5. F. Giorgianni, et al. Adv. Funct. Mater. 2017, 28, 1702652.
https://doi.org/10.1002/adfm.201702652
6. A. Balandin, S. Ghosh, W. Bao, I Calizo, D. Teweldebrhan, F. Miao, and C. N. Lau, Nano
Lett. 8, 902 (2008).
7. A. H. C. Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys.
81, 109 (2009).
8. A. K. Geim and K. S. Novoselov, Nat. Materials 6, 183 (2007).
9. A. K. Geim, Science, 324, 1530 (2009).
10. Y. Wang, X. Wu, and W. Zhang, Mat. Lett. 165, 71 (2016).
11. F. Maka, L. Ju, F. Wang, and T. F. Heinz, Solid State Commun. 152, 1341 (2012).
12. M. Shalaby and C. P. Hauri, Nat. Commun. 6, 8439 (2015).
13. C. Vicario, B. Monoszlai, & C. P. Hauri, Phys. Rev. Lett. 112, 213901 (2014).
14. R. E. Kelly, Am. J. Phys. 49, 714 (1981).
15. Z. Mics, et al., Nat. Commun. 6, 7655 (2015).
16. I. Gierz et al., Phys. Rev. Lett. 114, 125503 (2015).
17. I. Gierz et al., Nature. Materials 12, 1119 (2013).
18. L. Xiao et al., Nano Lett. 8, 4539 (2008).
19. K. Nobuyoshi, et al., Appl. Phys. Lett. 102, 123504 (2013).
20. A. Aliev, Y. N. Gartstein, and R. H. Baughman, Nanotechnology 24, 235501 (2013).
Acknowledgments:
We are grateful to Dominique Zehnder and Michael Eichenberger for assistance with the
measurements. We thank Sergejs Dementjevs for the loan of the acoustic measuring system.
Funding: We acknowledge financial support from the Swiss National Science Foundation
(SNSF) (Grant No. 200021_146769 and No. IZLRZ2-164051). MS acknowledges partial
funding from the European Community's Seventh Framework Programme (FP7/2007-2013)
under grant agreement no. 290605 (PSI-FELLOW/COFUND). CPH acknowledges association to
NCCR-MUST.
Competing interest: The authors declare no competing interests.
Data and material availability: All data needed to evaluate the conclusions in the paper are
present in the paper and/or the Supplementary Materials.
Author Contributions: The reported phenomenon was accidentally discovered by M.S. and
F.G.; M. G. proposed the sponge modulator. M.G., P.C., C.V. and M.S. performed the
microwave measurements. THz and optical measurements were done by M.S. and
C.V. C.V and P.C. prepared the figures. Y. C. and B. K. prepared the sample. M.S.
proposed the paper coupling concept and wrote the draft. S.L. and C.P.H. coordinated the
collaboration. All the authors contributed to the discussions and interpretation of the
results.
Fig. 1. Photon-phonon coupling in 3D sponge for heat and sound generation. As electromagnetic
excitation we used a microwave sine wave generator at frequency varying between 500 MHz to
4 GHz in CW and amplitude modulation mode. Under the effect of the CW microwave stimulus
the graphene sponge heats up rapidly but no sound emission is detectable. The sponge
temperature is measured with a thermographic camera (yellow region in (A)). The record-high
absorption capability and the low thermal inertia of the 3G-sponge leads to rapid and large rise of
the temperature beyond 150°C. (B) shows the temperature rise as function of the driving
microwave power and frequency inside the sample. (C) Under the excitation of 1 GHz wave
modulated in amplitude by a 5 kHz sine signal, the strong photon-phonon coupling in 3D-sponge
results in time varying temperature and pressure wave with consequent sound generation at the
modulation frequency show in (D). The emitted sound (black) reproduces excellently the
modulating signal (blue) and demonstrates the capability of the 3D-sponge for a high-fidelity
loudspeaker.
Fig. 2. Pulsed optical excitation of 3G-sponge in the Terahertz and optical range. (A) The sound
amplitude shows a linear dependence on the THz energy at frequency between 1 and 1 THz.
Sound emission from 3G-sponge occurs at the repetition rate of the laser (lower graph). Very
similar properties are measured at (B) optical and (C) microwave frequencies. The sound
amplitude is linearly dependent on the femtosecond laser pulse energy. (D) The recorded sound
spectra generated by a pulsed stimulus centered at frequencies across 15 octaves (THz to PHz)
are qualitatively comparable. (E) The thermal picture of the sponge indicates localized heating at
the THz focus. We note that together with a thermal detector the 3G-sponge could thus serve as a
2-dimensional THz intensity profiler.
Fig. 3. Application of 3G sponge as a demodulator using a GHz carrier. (A) Experimental
transmitter/receiver communication setup based on the 3G-sponge demodulator. (B) Intensity
voltage (I -- V) characteristics of 3G-sponge. A fit with a pure cubic polynomial is also shown in
the plot and corroborates the outstanding symmetry of the 3G-sponge demodulator. The
corresponding power-voltage (P-V) characteristic of 3G-sponge is also shown. The nonlinear
characteristics usually show a very significant quartic component which makes 3G-sponge an
excellent RF demodulator. (C) Output spectrum of the acoustic signal Vsound(f) measured from
the 3G-sponge after its demodulation and carrier suppression corresponds to the transmitted
signal spectrum Vsignal(f) .
|
1705.00905 | 1 | 1705 | 2017-05-02T11:02:57 | Analysis of the current-driven domain wall motion in a ratchet ferromagnetic strip | [
"physics.app-ph",
"cond-mat.other"
] | The current-driven domain wall motion in a ratchet memory due to spin-orbit torques is studied from both full micromagnetic simulations and the one dimensional model. Within the framework of this model, the integration of the anisotropy energy contribution leads to a new term in the well known q-$\Phi$ equations, being this contribution responsible for driving the domain wall to an equilibrium position. The comparison between the results drawn by the one dimensional model and full micromagnetic simulations proves the utility of such a model in order to predict the current-driven domain wall motion in the ratchet memory. Additionally, since current pulses are applied, the paper shows how the proper working of such a device requires the adequate balance of excitation and relaxation times, being the latter longer than the former. Finally, the current-driven regime of a ratchet memory is compared to the field-driven regime described elsewhere, then highlighting the advantages of this current-driven regime. | physics.app-ph | physics |
Analysis of the current-driven domain wall motion in a ratchet ferromagnetic strip
L. S´anchez-Tejerina,1, ∗ ´O. Alejos,1 E. Mart´ınez,2 and V. Raposo2
1Dpto. Electricidad y Electr´onica. Facultad de Ciencias,
Universidad de Valladolid, 47011 Valladolid, Spain
2Dpto. Fsica Aplicada, Facultad de Ciencias, Universidad de Salamanca, 37011 Salamanca, Spain
The current-driven domain wall motion in a ratchet memory due to spin-orbit torques is studied
from both full micromagnetic simulations and the one dimensional model. Within the framework
of this model, the integration of the anisotropy energy contribution leads to a new term in the
well known q-Φ equations, being this contribution responsible for driving the domain wall to an
equilibrium position. The comparison between the results drawn by the one dimensional model
and full micromagnetic simulations proves the utility of such a model in order to predict the
current-driven domain wall motion in the ratchet memory. Additionally, since current pulses are
applied, the paper shows how the proper working of such a device requires the adequate balance of
excitation and relaxation times, being the latter longer than the former. Finally, the current-driven
regime of a ratchet memory is compared to the field-driven regime described elsewhere, then
highlighting the advantages of this current-driven regime.
Keywords: Domain Wall motion, Magnetic memory, Ratchet memory, Magnetocristalline Anisotropy.
INTRODUCTION
Domain Walls (DW) are defined as the boundaries be-
tween regions of different magnetization (domains) in a
magnetic ordered system. DWs in ultrathin films, where
high perpendicular anisotropy appears due to this ul-
tra low thickness,[1] have been recurrently proposed as
promising devices for data storage based on DW motion
caused by applied fields [2, 3] or currents. [4, 5] In partic-
ular, the motion of these DWs under the solely applica-
tion of a magnetic field Hz parallel to the magnetization
of the domains is a well-known issue. [6] Under such an
excitation, it is known that the velocity of the DW in-
creases linearly until a well defined field, known as Walker
field HW , is reached. Above this field the DW velocity
falls down, reaches a minimum, and then starts increas-
ing, eventually overcoming the maximum speed achieved
at HW . The most characteristic difference between both
regimes, below (Hz < HW ) and above (Hz > HW ) the
Walker field, is that the former regime is characterized
by a constant orientation of the inner magnetization of
the DW, while the latter imposes that this magnetiza-
tion precesses around the applied magnetic field. This
behavior, together with the fact that applied fields drive
DWs of different types in opposite directions, constitutes
a serious handicap for the implementation of DW-based
memory elements.
However, recent studies focused on multilayer systems
with inversion asymmetry, where a ferromagnetic layer
(FM) is sandwiched between a heavy metal (HM) and
[7 -- 9] have open new promising ways
an oxide (Ox),
of efficient DW-driving.
the presence
of
interfacial Dzyaloshinskii-Moriya interaction
(DMI) at the HM/FM interface makes DWs adopt an
Particularly,
the
homochiral N´eel configuration, which is important for
such an efficient driving, since DWs can be pushed by
the torque exerted by an injected current through the
HM, due to the spin Hall effect (SHE). The conclusions
drawn by these studies have shown that there is not
an analogous Walker current, and the DW velocity
continuously increases up to a value which is reached
as the orientation of the magnetization within the DW
approaches the direction perpendicular to the current,
due to the exerted torque.
in
Once the suitable procedure for DW-driving is pro-
vided, a pinning system is required to exactly control the
DW position in the FM strip. [10, 11] This problem has
been solved in most cases by introducing notches along
the strip. However, this tactic was designed for systems
with in-plane magnetization, while in PMA systems this
effect is weaker, and induce deformations of the DW.
For these reasons, some alternatives have been proposed,
such as the application of a voltage in an epitaxial mag-
netic tunnel junction (MTJ), [12] or a strain-mediated
coupling
bilayer
structures.
[13] The control of the pinning (and DW
nucleation) by means of a tailored PMA has also been
proposed. [10] The sample is irradiated with heavy ions
so as to create an anisotropy landscape characterized by
a sawtooth profile. This idea led to a new proposal for
a magnetic memory, known as ratchet memory, which
was studied under the field-driven regime, in particular,
the effect of an alternate applied magnetic field with
fixed orientation. [14] The sawtooth was then meant to
both fix DW positions (and so define the bit size), and
establish one direction of bit shifting by avoiding DWs
backward movement due to the applied field.
piezoelectric/magnetostrictive
∗ Email address:[email protected]
This work goes deeper into this idea, but proposes
a current-driven mechanism to ensure the proper bit
shifting along the FM strip, which constitutes a much
more interesting alternative from the technological point
of view to the field-driven basis, since current both
promotes the dynamics of all DWs in the same direction
and does not lead to precessional DW dynamics, as
magnetic field does. Both facts contribute to reduce bit
sizes (and so increase bit densities), and speed up bit
shifting. The work has been carried out with the help
of micromagnetic (µMag) simulations as to mimic as
much as possible realistic conditions and to explore the
feasibility of the proposed mechanisms. Additionally,
a one dimensional model (1DM), updated so as to
include an effective field accounting for the anisotropy
landscape, has been used in order to clarify some aspects
of the DW dynamics and can be of help in the further
development of the proposed system. According to this,
the work is structured as follows. Section I presents the
system under study and describes the 1DM. In Section II
full µMag simulations at T = 300 are presented and
compared to those provided by the 1DM. This section
focuses on the tuning of the performances of the system
by means of the adequate current amplitude, pulse time
and relaxation time, being divided in three subsections.
While in section II A 50% duty cycle pulses have been
used, section II B shows how larger relaxation times
improve the system response, by reducing the duty cycle
to 33%. A short discussion is made in Section II C about
the effect of not considering a constant DMI parameter,
since ion irradiation may also alter the characteristics
of this interaction. Additionally, the results shown for
the duty cycle of 33% have been used to be compared
with those obtained for the field-driven ratchet memory.
[14] The main conclusions of the study are drawn in
Section III.
I. GEOMETRY AND MODELS
A. µMag Simulations
A typical ratchet FM strip with high perpendicular
magnetic anisotropy (PMA) sandwiched between a HM
and an Ox is considered along the text. The ratchet
anisotropy profile is obtained by tuning the out-of-plane
anisotropy as done elsewhere.
[10] The central column
of figure 1 depicts five periods of this anisotropy profile
along the ratchet. Here the values of the highest and
u = 1 MJ
lowest anisotropy values, K +
m3 ,
respectively, are presented. A ratchet period d = 128nm
has been taken, as indicated also on the figure. The
left column in figure 1 sketches the behaviour of a
DW when short,
intermediate, and large currents are
applied. For low currents, the DW is not able to skip
the teeth, labeled as point B in the upper subfigures,
and then returns to its starting position A. Intermediate
currents, however, lead the DW to skip one teeth at each
pulse, i.e., a single DW jump, as it is depicted in the
middle-right image. Finally, sufficiently large currents
m3 and K−
u = 1.27 MJ
can make the DW skip two or more tooth at each pulse.
2
m2 has been considered.
In order to check the operation range of this unidi-
rectional ratchet device, material parameters commonly
found in the literature [3] have been used for the present
study: saturation magnetization Ms of 1.1 MA
m , exchange
constant A of 16 pJ
m and a Gilbert damping parameter
α = 0.5. An interfacial DMI with a DMI parameter
D = 1 mJ
It must be noted that
such an interfacial DMI is a constant, although the irradi-
ation procedure is likely to affect not only the anisotropy,
but also the DMI. However, to the best of our knowl-
edge, there are no studies which link the change of the
anisotropy constant with a change of the DMI value.
Nevertheless, section II C has been included to show that
a linear change of the DMI with a similar ratio to that
considered for the PMA does not dramatically modify the
results. The proposed DMI parameter is sufficiently high
as to induce N´eel DWs and so, to allow efficient current-
driven DW movement by means of the SHE. Such an elec-
tric current is applied through the HM, being θSH = 0.1
the considered spin Hall angle. A FM strip with a cross
section Ly × Lz of 128nm × 0.6nm has been taken. The
evolution in time of the system normalized magnetiza-
tion (cid:126)m is described by the Landau-Lifshitz-Gilbert (LLG)
equations augmented by the spin-orbit torques (SOT)
and thermal fluctuations:
d (cid:126)m
dt
= −γ0 (cid:126)m ×(cid:16)
−α (cid:126)m × d (cid:126)m
dt
(cid:17)−
(cid:126)Hef f + (cid:126)Hth
− γ0 (cid:126)m ×(cid:16)
(cid:126)m × (cid:126)HSH
(cid:17)
.
(1)
where γ0 and α are the gyromagnetic ratio and the
damping parameter, (cid:126)Hef f is the effective field, includ-
ing exchange, anisotropy, and magnetostatic interactions,
along with the DMI, and (cid:126)HSH and (cid:126)Hth are, respectively,
the effective field associated to the SHE and the thermal
field, the latter included as a gaussian-distributed ran-
dom field. [15 -- 18] The study of the motion of a DW in
such a system requires solving this equation by means
of full micromagnetic simulations (µMag). These µMag
simulations have been performed with the help of the
Mumax3 package. [19]
B. 1DM
The one dimensional model (1DM) was originally de-
veloped to describe the behaviour of one DW in an in-
finite wire, where magnetization changes only along the
longitudinal coordinate. [3, 20, 21] Within the framework
of this model, two parameters account for the DW state,
named q and Φ. The former determines the DW position,
while the latter is, in our context, related to the relative
orientation of the in-plane component of the magnetiza-
tion with respect to the strip longitudinal direction. A
further development of the 1DM [10] contributed to ad-
equately describe the behavior of one DW in an infinite
3
Figure 1. DW motion under applied current in the FM ratchet considered. The sawtooth profile locates the DW in well
defined areas as represented in the magnetization snapshots on the right. The final position of the DW depends on the current
amplitude. If this amplitude is rather small, the DW does not cross any teeth and come back to its starting position A or an
intermediate position A', as in the upper case. If the amplitude is excessively high, multiple teeth may be skipped at once, as
in the lower case. The proper performance of the device is obtained when the DW skips one teeth at one current pulse.
strip with an anisotropy energy profile characterized by a
linear variation in a region centred at x = 0, and constant
values of the anisotropy at the extremes of such a region.
Here, a sawtooth profile, as that presented in figure 1, is
to be considered, i.e., a periodical repetition of a linear
anisotropy profile with a period given by the previously
defined distance d. Two mutually dependent equations
can be derived for the DW dynamic behaviour in such a
system:
q =
+
Φ =
+
γ0∆
1 + α2 [(HD−Hk cos Φ) sin Φ] +
γ0∆
1 + α2 [α (HSH cos Φ + Hth + Hrr(q))] ,
1 + α2 [α (Hk cos Φ−HD) sin Φ] +
1 + α2 [HSH cos Φ + Hth + Hrr(q)] .
γ0
γ0
(2a)
(2b)
The terms HD, Hk, HSH stand for the DMI, magneto-
static interaction, and SHE, respectively, and their def-
inition can be found elsewhere.
[3, 21] Similarly, some
bibliography can be posed where the thermal term Hth
is adequately discussed.
[15 -- 18] Finally, the anisotropy
profile introduces the terms Hr = K+
, and
r(q) =
−
1
(cid:3) +
cosh2(cid:2)(cid:0)1 − { q
d}(cid:1) d
(cid:1)
d sinh(cid:0) q
(cid:20) (1−{ q
cosh(cid:0){ q
(cid:1) cosh
d} d
∆
∆
∆
∆
∆
(cid:1)−
d} d
∆
1
u −K
−
u
2µ0Ms
cosh2(cid:0){ q
(cid:21) ,
d})d
(3)
where the braces stand for the fractional part function. It
must be pointed out here that well-defined DW equilib-
rium positions can be derived from (2), which are found
close to the edges of the sawtooth profile, where the
anisotropy minima are located. More details are given
in Appendix A.
II. RESULTS
A. Comparison between the 1DM and full µMag
simulations at 300K for a 50% duty cycle
The working principle for the device presented in
section I is the following. Each region of length d can
pin one DW, i.e., the region size defines the bit size.
When a pulse of current is applied to the system, every
DW moves forward during the pulse time te. When
the current stops, the anisotropy interaction drives the
DW during a relaxation time tr towards the closest
anisotropy minimum, where the DW is more stable. If
4
not skip the anisotropy tooth and reverses during the
time tr.
If tr is sufficiently long, the DW recovers its
starting position A at equilibrium. However, the time
tr is in this case rather short, and the DW acquires an
intermediate position A' when the current is switched
on again. An eventual bit shifting may occur during the
application of the second current pulse, since the DW
may skip the anisotropy tooth from such an intermediate
position. This is, in general, an undesirable behavior.
Contrarily, if the amplitude of the pulse is adequately
high as to displace the DW a distance larger than d
in a time te, the DW skips the anisotropy tooth and
does not return to its starting position, but to the
subsequent equilibrium position, a distance d away from
its starting position, if the relaxation time is sufficiently
long. This can be seen in figure 2(b), which corresponds
to the central case of figure 1, being this situation
the desired behavior, since the information is shifted
only one bit at a time, i.e., a single DW jump occurs.
Finally,
if the current amplitude is rather high, the
DW may reach an intermediate position prior to the
application of the next pulse. This can cause the DW to
eventually advance two bits at a time, instead of only
one, when subsequent pulses are applied. What is more,
for sufficiently high currents, the DW runs a distance
of two bits at once instead of only one, as figure 2(c)
depicts, which corresponds to the bottom case in fig 1.
The proper range of operation of the device is the range
of values of the current amplitude Ja for a given pair of
times te and tr that promote single DW jumps after the
application of one current pulse. As an example, Figure 3
depicts the probability of one single DW jump after the
application of one pulse as a function of the current, ob-
tained from both µMag simulations and the 1DM, when
te = tr, that is, a duty cycle of 50% for the injected
current. Three different pulse times, te = 1ns, 1.5ns
and te = 2ns, are considered. The probability has been
statistically obtained by evaluating twenty different real-
izations for each current amplitude and pulse length. It
must be noticed that the proper range of operation for a
duty cycle of 50% increases with decreasing pulse lengths.
Alternatively, next subsection shows that a more efficient
way to increase the range of operation is to increase only
the relaxation time, then varying the duty cycle. In any
case, it can be checked from the plots in this graph that
the results provided by full µMag simulations and the
1DM are in a rather good agreement.
Figure 2. DW position as a function of time for a 2-ns pulse
length with a 2-ns relaxation time. Three different current
amplitudes are considered: (a) Ja = 0.4 TA
m2 , (b) Ja = 0.6 TA
m2 ,
and (c) Ja = 1.1 TA
m2 . The graphs also compare the results
drawn by both µMag simulations (solid lines) and the 1DM
(dashed lines). The vertical black dashed lines indicate the
time at which HM current is switched on and off, while the
horizontal black dashed lines mark the subsequent positions
of each teeth (−256, −128, 0 and 128, 256nm, for d = 128nm).
the amplitude of the pulse is rather low, the DW does
not skip any tooth of the anisotropy landscape, and
no bit shifting occurs, as the image on top of figure 1
depicts. A certain correspondence between this figure
and the graph labelled as figure 2(a) can be checked.
Actually, the graph in this figure reflects the case when
the DW is firstly driven by the current for a time te a
distance shorter than d, up to a point B. The DW does
B. Comparison between one dimensional model
and full µMag simulations at 300K for 33% duty cycle
With the aim of extending the 100% probability range
of proper working of this device, the following strategy
can be proposed. Since the DW requires a minimal relax-
ation time to reach an equilibrium position, this time can
be lengthened, then reducing the current duty cycle. As
-2-101201234567800.20.40.60.81time(ns)DWposition/dAppliedcurrent(TA/m2)(c)-2-101200.20.40.60.81(b)-2-101200.20.40.60.81(a)Ja1DMµMagJa1DMµMagJa1DMµMag 5
Figure 3. Probability of single jumps of one DW after the
application of one pulse with a duty cycle of 50% as a function
of the current amplitude for pulses with te = 1ns, 1.5ns, and
2ns, computed from full µMag simulations (filled symbols)
and the 1DM (open symbols).
an example, the following results refer to a ratchet mem-
ory initially driven by an injected current with a time
pulse te = 1ns, and further relaxing for tr = 2ns prior to
the application of the subsequent pulse, that is, the duty
cycle is reduced to 33%.
Figure 4 compares the probability of single jumps of a
DW after one current pulse (one single jump), and after
four pulses (four consecutive single DW jumps) calcu-
lated by means of µMag simulations and the 1DM. Two
duty cycles, 50% and 33%, are considered.
As in the preceding case, twenty different realizations
have been carried out in order to compute statistics.
The results for current pulses with a 50% duty cycle
(see top graph) reveal that the range of proper working
of the device notably reduces as the number of pulses is
increased from one to four (the latter may be regarded
as the fourth power of the after-one-pulse results). This
fact makes the system much less reliable when long bit
shifting is needed, or when a long array of bits must
be displaced. However, the bottom graph in the figure
reveals that, even though the range of proper operation
also reduces as the number of pulses is increased, this
range is still rather wide in the case of a 33% duty cycle.
Thus,
it is possible to improve the reliability of the
current-driven ratchet memory by lowering the current
duty cycle.
Figure 5 depicts the magnetization of an array of eight
bits in a ratchet memory with a bit length given by the
period d = 128nm. Three 33% duty cycle current pulses
of width te = 1ns and relaxation times of tr = 2ns are
applied to the device. The figure demonstrates the cor-
rect operation of the system, which drastically improves
the performances of the field-driven ratchet memory. [14]
In particular the bit size is reduced to an approximate
Figure 4. Probability of one single jump of a DW promoted
by one pulse current with 50% (top graph) and 33% (bottom
graph) duty cycles (solid line for the results provided by the
1DM and squares for those obtained from µMag simulations),
and probability of four single jumps after four current pulses
(dashed line for the results provided by the 1DM and triangles
for those obtained from µMag simulations). The pulse time is
te = 1ns in both cases, while the relaxation times are tr = 1ns
for the 50% duty cycle and tr = 2ns for the 33% duty cycle.
factor of two.
In fact, the field-driven system requires
the length of two periods of the anisotropy landscape
to store one bit, while one period is needed to trap
every DW in the case of the current-driven device, then
reducing the bit size to the length of only one period.
Furthermore, Walker field limits the highest DW speed
in the case of the field-driven system. The required time
to shift one bit is in that case as long as 16ns. However,
3ns are sufficient to carry out one correct bit shifting in
the case of the current-driven element here proposed.
C. Effect of a variable DMI
Finally, it must be reminded that a constant DMI pa-
rameter has been considered along this work. Although
this assumption might not seem sufficiently realistic,
to the best of our knowledge there are no studies on
how the DMI changes when the HM-FM interface is
00.20.40.60.810.40.60.811.21.41.61.822.2ProbabilityAppliedcurrent(TA/m2)1DM:1ns1DM:1.5ns1DM:2ns1ns1.5ns2ns00.20.40.60.8111.522.53Appliedcurrent(TA/m2)Probability0.20.40.60.8150%DutyCycle33%DutyCycle1pulseµMag1pulse1DM4pulsesµMag4pulses1DM1pulseµMag1pulse1DM4pulsesµMag4pulses1DM 6
Figure 6. Probability of single jumps of one DW for the model
with constant DMI parameter (solid line) and with a linear
variation of the DMI parameter (dashed line).
respect to the case of constant DMI, and the predictions
about the proper range of operation are not highly
affected.
Figure 5. The operation of a current-driven ratchet memory
containing eight bits is studied by means of µMag simulations.
Three 33% duty cycle current pulses of width te = 1ns and
relaxation times tr = 2ns are applied to the device. The
current amplitude is fixed to Ja = 1.2 TA
m2 . The intermediate
figure displays the anisotropy landscape, which defines the bit
size, as a color map. The storage of every bit then requires a
surface of d × w = 128 × 128nm2, to complete a total length
of the ratchet of L = 1024nm. Snapshots at the end of every
subsequent interval of time either te or tr are sketched to
illustrate the dynamical behaviour.
irradiated with heavy ions so as to tailor the PMA. As
a first approach to the effect of this irradiation on the
DMI, a linear variation of the DMI parameter, analogous
to that of the PMA, can be proposed.
In this way,
µMag simulations have been carried out to compare the
results obtained both on the constant DMI basis and
from the proposed linear variation of the DMI. In the
latter case, the DMI is considered to change between
D− = 0.8 mJ
m2 . 33% duty cycle current
pulses are applied with an amplitude Ja = 1.2 TA
m2 , and
the probability of single jumps is analysed in both cases,
then drawing the results plotted in figure 6. No drastic
changes are found in the case of variable DMI with
m2 and D+ = 1 mJ
III. CONCLUSIONS
The work presents a study of the current-driven
ratchet memory as a DW-based magnetic memory.
This study has been carried out by means of full µMag
simulations along with the results provided by a 1DM,
showing both approaches a rather good agreement.
It
was necessary in the latter model to introduce a new
term in the time derivatives of the DW position q and
the magnetization angle Φ taking into account the space
dependent anisotropy profile which makes it possible
this one-way working principle. The work shows how
bit density and shifting speeds are notably increased if
the memory device works on this current-driven basis,
as compared with the field-driven scheme. A fine tuning
of the pulse and relaxation times may even improve its
performances. In this way, the best results are obtained
when the combined effect of the times of both the
applied current pulse and the relaxation interval leads
every DW from an equilibrium state to another one
at the contiguous tooth of the anisotropy landscape,
the former time by promoting only one single jump
and the latter by allowing the DW to recover this new
equilibrium position.
00.20.40.60.8111.522.53ProbabilityAppliedcurrent(TA/m2)constantDMIvariableDMI Appendix A: Derivation of the one dimensional
model
In order to derive the 1DM the usual assumption that
magnetization only changes appreciably in one direction
is considered. The well known Walker ansatz is taken in
order to describe the magnetization along the FM strip
where a DW is formed. The use of variational principles
[6, 20] leads in the case of a strip with constant anisotropy
to a couple of well established equations for the DW po-
sition and the magnetization orientation within the DW.
In our case, and due to the additive character of the en-
ergy accounting for the considered interactions, the only
term in these equations that must be explicitly rewritten
is the one resulting from the integration of the anisotropy
landscape along the longitudinal direction. Accordingly,
the expression of the magnetocristalline anisotropy den-
sity energy uanis = Ku sin2 θ is to be considered along
this section, where θ is the angle between the magnetiza-
tion and the out-of-plane direction (z). As to draw some
preliminary results, the case of a single anisotropy slope
can be recalled.
[10] In such a case Ku is no longer a
constant, and is defined as:
K−
+ K+
K +
u
x < − d
2 ,
2 < x < d
2 ,
x > d
2 ,
x if − d
Ku (x) =
u −K
(A.1)
where K +
u is the uniaxial anisotropy constant at the right
side of the strip while K−
u is the uniaxial anisotropy
constant at the left side, d being the distance of vari-
ation of the anisotropy. If the Walker ansatz is consid-
ered to define the magnetization along the FM strip, i.e.,
(cid:1)(cid:3), q being the DW position and
θ = 2 arctan(cid:2)exp(cid:0) x−q
u +K
K+
−
u
−
u
if
if
u
d
2
∆ the DW width, taken as a constant value, then the
anisotropy free energy density per unit surface σanis is
given by:
∆
−∞
(cid:90) ∞
(cid:90) − d
(cid:18) K +
(cid:90) d
(cid:90) ∞
−∞
− d
K
2
2
2
σanis =
=
+
+
K sin2 θdx =
−
u sin2 θdx+
u + K−
u
+
2
K +
u sin2 θdx,
d
2
(cid:19)
dx+
u − K−
K +
u
d
x sin2 θ
(A.2)
I2 = ∆
tanh
+ tanh
and the second integral is:
u
2
u + K−
K +
(cid:34)
(cid:34)
u − K−
K +
(cid:1) ∆2
u − K
−
u
2
u
tanh
ln
d
∆
(cid:32) d
(cid:33)
2 − q
(cid:32) d
(cid:33)
2 − q
(cid:16) d
cosh
(cid:16) d
∆
cosh
+ ∆
+(cid:0)K +
− tanh
(cid:17)
(cid:17)
.
2 +q
∆
2 −q
∆
(cid:32) d
(cid:32) d
2 + q
∆
2 + q
∆
7
(cid:33)(cid:35)
(cid:33)(cid:35)
+
+
(A.4)
By adding up this results, (A.2) can be worked out as:
σanis = ∆(cid:0)K +
(cid:0)K +
+
∆2
d
u + K
−
u
u − K
−
u
(cid:1) +
(cid:1) ln
cosh
cosh
(cid:16) d
(cid:16) d
2 +q
∆
2 −q
∆
(cid:17)
(cid:17)
.
(A.5)
which leads to the following interaction term as due to
the local variation of the anisotropy:
(cid:0)K +
u − K
−
u
∂σanis
∂q
=
∆
d
anisotropy areas(cid:0)q > d
2
2 sinh(cid:0) d
(cid:1)
(cid:1) + cosh(cid:0) 2q
(cid:1)
cosh(cid:0) d
(cid:1)), and drives the wall towards
(A.6)
(cid:1)
∆
∆
∆
The effect of this term vanishes for the constant
lower anisotropy areas when a local variation of this
interaction appears, as expected. Accordingly, for an
anisotropy variation following a sawtooth profile, i.e., an
anisotropy slope periodically repeated as shown in fig-
ure 1, the local anisotropy minima define equilibrium
positions for the DWs. This can be analytically demon-
strated by taken an anisotropy parameter Ku (x) defined
by the formula:
Ku (x) = K
−
u +
u − K−
K +
u
d
x − (cid:98) x
d
(cid:99)d
,
(A.7)
where the brackets (cid:98) (cid:99) stand for the floor function. For
the sake of simplicity, Ku (x) has been considered to be
discontinuous between the points of maximal and mini-
mal anisotropy. σanis can be worked out as:
(cid:104)
(cid:105)
(cid:90) L
σanis =
0
N−1(cid:88)
n=0
where the first and third integral can be obtained as:
Ku (x) sin2 θdx =
In,
(A.8)
(cid:34)
(cid:34)
I1 = ∆K
−
u
I3 = ∆K +
u
1 − tanh
1 − tanh
(cid:32) d
(cid:32) d
2 − q
2 + q
∆
(cid:33)(cid:35)
(cid:33)(cid:35)
∆
,
,
(A.3)
L being the total length of the ratchet, then containing
N slopes, that is, L = N d. In is the result of the integral
of the anisotropy along the n-th slope, n enumerating the
subsequent slopes, and ranging from 0 to N − 1. In then
reads:
∆
−
(cid:34)(cid:0){ q
(cid:35)
d} + n(cid:1) d
(cid:35)
(cid:34)(cid:0){ q
d} + n−1(cid:1) d
(cid:20)
cosh
(cid:1) ln
(cid:20)
−
u
∆
cosh
(cid:21)
(cid:21)
(A.9)
.
+
({ q
d}+n)d
({ q
d}+n−1)d
∆
∆
In = ∆K
−
u tanh
− ∆K +
u tanh
(cid:0)K +
u − K
+
∆2
d
−
u + K
u
−
u + K
u
−
u + K
u
σanis = ∆(cid:0)K +
+ ∆(cid:0)K +
− ∆(cid:0)K +
(cid:0)K +
+
∆2
d
u + K
(cid:1) +
(cid:20)(cid:16)
}(cid:17) d
(cid:1) tanh
(cid:18)
(cid:19)
(cid:1) tanh
cosh(cid:0){ q
(cid:1) ln
(cid:20)
1−{ q
d
} d
∆
{ q
d
+
−
u
∆
d} d
d})d
(1−{ q
∆
(cid:1)
cosh
∆
8
(cid:21)
−
(A.10)
.
(cid:21)
It must be reminded here that the braces stand for the
fractional part function. Under the assumption that the
DW width ∆ is negligibly small if compared with the
length d, the sum in (A.8) adds up to:
This result, together with the other contributions to
the energy density per unit surface, as long as the thermal
field,[15 -- 18] allows us to finally write the 1DM equations
(2).
[1] M. T. Johnson, P. J. H. Bloemen, F. J. A. den Broeder,
and J. J. de Vries, Reports on Progress in Physics 59,
1409 (1996).
[2] R. P. Cowburn, J. Ferr, S. J. Gray, and J. A. C. Bland,
Applied Physics Letters 74, 1018 (1999).
[3] A. Thiaville, S. Rohart, E. Ju´e, V. Cros, and A. Fert,
Europhysics Letters 100, 57002 (2012).
[4] J. Vandermeulen, B. V. de Wiele, A. Vansteenkiste, B. V.
Waeyenberge, and L. Dupr´e, Journal of Physics D: Ap-
plied Physics 48, 035001 (2015).
[5] T. A. Moore, I. M. Miron, G. Gaudin, G. Serret, S. Auf-
fret, B. Rodmacq, A. Schuhl, S. Pizzini, J. Vogel, and
M. Bonfim, Applied Physics Letters 93, 262504 (2008).
[6] N. L. Schryer and L. R. Walker, Journal of Applied
Physics 45, 5406 (1974).
[11] S. S. P. Parkin, M. Hayashi, and L. Thomas, Science
320, 190 (2008).
[12] Y. Shiota, S. Murakami, F. Bonell, T. Nozaki, T. Shinjo,
and Y. Suzuki, Applied Physics Express 4, 043005 (2011).
[13] N. Lei, T. Devolder, G. Agnus, P. Aubert, L. Daniel,
J. Kim, W. Zhao, T. Trypiniotis, R. P. Cowburn,
C. Chappert, D. Ravelosona, and P. Lecoeur, Nature
Communications 4, 1378 (2013).
[14] J. H. Franken, H. J. M. Swagten, and B. Koopmans,
Nature Nanothecnology 7, 499 (2012).
[15] J. W. F. Brown, Physical Review 130, 1677 (1963).
[16] J. L. Garc´ıa-Palacios and F. J. L´azaro, Physical Review
B 58, 14 937 (1998).
[17] E. Martinez, L. Lopez-Diaz, L. Torres, C. Tristan, and
O. Alejos, Physical Review B 75, 174409 (2007).
[7] E. Martinez, S. Emori, and G. S. D. Beach, Applied
[18] R. A. Duine, A. S. N´unez, and A. H. MacDonald, Phys-
Physics Letters 103, 072406 (2013).
ical Review Letters 98, 056605 (2007).
[8] S. Emori, U. Bauer, S. Ahn, E. Martinez, and G. S. D.
Beach, Nature Materials 12, 611 (2013).
[9] E. Martinez, S. Emori, N. Perez, L. Torres, and G. S. D.
[19] A. Vansteenkiste, J. Leliaert, M. Dvornik, M. Helsen,
F. Garcia-Sanchez, and B. V. Waeyenberge, AIP Ad-
vances 4, 107133 (2014).
Beach, Journal of Applied Physics 115, 213909 (2014).
[20] A. Thiaville, J. M. Garc´ıa, and J. Miltat, Journal of Mag-
[10] J. H. Franken, M. Hoeijmakers, R. Lavrijsen,
and
H. J. M. Swagten, Journal of Physics: Condensed Matter
24, 024216 (2012).
netism and Magnetic Materials 242-245, 1061 (2002).
[21] L. S´anchez-Tejerina, O. Alejos, E. Mart´ınez, and J. M.
Munoz, J. Magn. Magn. Mater. 409, 155 (2016).
|
1807.07664 | 1 | 1807 | 2018-07-19T23:49:34 | Enhanced dynamic nuclear polarization via swept microwave frequency combs | [
"physics.app-ph",
"physics.atom-ph",
"physics.chem-ph",
"quant-ph"
] | Dynamic Nuclear Polarization (DNP) has enabled enormous gains in magnetic resonance signals and led to vastly accelerated NMR/MRI imaging and spectroscopy. Unlike conventional cw-techniques, DNP methods that exploit the full electron spectrum are appealing since they allow direct participation of all electrons in the hyperpolarization process. Such methods typically entail sweeps of microwave radiation over the broad electron linewidth to excite DNP, but are often inefficient because the sweeps, constrained by adiabaticity requirements, are slow. In this paper we develop a technique to overcome the DNP bottlenecks set by the slow sweeps, employing a swept microwave frequency comb that increases the effective number of polarization transfer events while respecting adiabaticity constraints. This allows a multiplicative gain in DNP enhancement, scaling with the number of comb frequencies and limited only by the hyperfine-mediated electron linewidth. We demonstrate the technique for the optical hyperpolarization of 13C nuclei in powdered microdiamonds at low fields, increasing the DNP enhancement from 30 to 100 measured with respect to the thermal signal at 7T. For low concentrations of broad linewidth electron radicals, e.g. TEMPO, these multiplicative gains could exceed an order of magnitude. | physics.app-ph | physics | Enhanced dynamic nuclear polarization via swept microwave frequency combs
A. Ajoy,1, ∗ R. Nazaryan,1 K. Liu,1 X. Lv,1 B. Safvati,2 G. Wang,1
E. Druga,1 J. A. Reimer,3 D. Suter,4 C. Ramanathan,5 C. A. Meriles,6, 7 and A. Pines1
1Department of Chemistry, University of California Berkeley, and Materials Science
Division Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA.
2Department of Physics, University of California Berkeley, Berkeley, California 94720, USA.
3Department of Chemical and Biomolecular Engineering, and Materials Science Division
Lawrence Berkeley National Laboratory University of California, Berkeley, California 94720, USA.
4Fakultat Physik, Technische Universitat Dortmund, D-44221 Dortmund, Germany.
5Department of Physics and Astronomy, Dartmouth College, Hanover, New Hampshire 03755, USA.
6Department of Physics, CUNY-City College of New York, New York, NY 10031, USA.
7CUNY-Graduate Center, New York, NY 10016, USA.
8
1
0
2
l
u
J
9
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
4
6
6
7
0
.
7
0
8
1
:
v
i
X
r
a
Dynamic Nuclear Polarization (DNP) has enabled enormous gains in magnetic resonance signals and led to
vastly accelerated NMR/MRI imaging and spectroscopy. Unlike conventional cw-techniques, DNP methods
that exploit the full electron spectrum are appealing since they allow direct participation of all electrons in the
hyperpolarization process. Such methods typically entail sweeps of microwave radiation over the broad electron
linewidth to excite DNP, but are often inefficient because the sweeps, constrained by adiabaticity requirements,
are slow. In this paper we develop a technique to overcome the DNP bottlenecks set by the slow sweeps, em-
ploying a swept microwave frequency comb that increases the effective number of polarization transfer events
while respecting adiabaticity constraints. This allows a multiplicative gain in DNP enhancement, scaling with
the number of comb frequencies and limited only by the hyperfine-mediated electron linewidth. We demon-
strate the technique for the optical hyperpolarization of 13C nuclei in powdered microdiamonds at low fields,
increasing the DNP enhancement from 30 to 100 measured with respect to the thermal signal at 7T. For low
concentrations of broad linewidth electron radicals, e.g. TEMPO, these multiplicative gains could exceed an
order of magnitude.
Introduction: – Dynamic nuclear polarization (DNP) – the
process of polarizing (cooling) nuclear spins to a spin tempera-
ture far lower than the lattice temperature [1, 2] – has emerged as
a technological breakthrough that serves as the starting point for
a wide-range of applications, including signal enhanced spec-
troscopy [3, 4] and imaging [5] and for state initialization in
quantum information processing and metrology [6, 7]. Indeed,
magnetic resonance (NMR and MRI) signals from hyperpolar-
ized nuclear spins can be enhanced by several orders of magni-
tude allowing enormous gains, even approaching a million-fold,
in experimental averaging time. This has opened up avenues for
the sensitive probing of phenomena, species and surfaces [8],
whose detection would otherwise have remained intractable.
In its simplest manifestation DNP involves the use of elec-
trons whose polarization is transferred to the nuclear spins via
microwave irradiation [10], allowing a polarization enhance-
ment ε (cid:46) γe/γn, where γe,n are the gyromagnetic ratios of
the electron and nuclear spins respectively. Resonant polariza-
tion transfer between electron and nuclear spin is achieved via
microwave excitation. Depending on the concentrations of the
electron and nuclear spins in the insulating solid, the transfer
can be mediated by thermal mixing, the cross effect, the solid ef-
fect and even the Overhauser effect. However several common
(e.g. nitroxide based) electron polarizing agents have large g-
anisotropy and severely inhomogeneously broadened electronic
linewidths that scale rapidly with field and can be as broad as
0.5GHz at high fields (>3T) [9, 11–13]. This broadening lim-
its the number of spins contributing to the resonant energy ex-
change at a particular microwave frequency. Similar problems
can exist even at low fields for some systems. For instance, Ni-
trogen Vacancy (NV) center defects in diamond [14, 15] have
∗ [email protected]
garnered much attention as optical hyperpolarizing agents be-
cause the NV electrons can be fully optically polarized at room
temperature [16], opening the possibility for DNP enhancements
larger than traditional bounds set by the gyromagnetic ratios,
without the need for cyrogens. Interest has been particularly fo-
cused on "hyperpolarized nanodiamonds", because their inher-
ently high surface area makes them attractive for the optical hy-
perpolarization of liquids brought in contact with them [17, 18].
While this has been a long-sought goal, technical challenges
presented by the NV electrons make the production of hyper-
polarized nanodiamonds challenging. In particular, the spin-1
NV centers have significant broadening on account of differ-
ent crystallite orientations having different frequencies, giving
rise to spectra broadened by >1GHz even at modest (30mT)
fields. Unsurprisingly, precise energy matching to the nuclei in
all these situations is challenging to achieve. Indeed DNP tra-
ditionally has relied largely on cw-microwave techniques (solid
and cross-effects) where a single frequency is saturated [19, 20],
and consequently for static samples only a small fraction of the
broad electron spectrum directly contributes to the obtained en-
hancement.
In principle however, significant gains in polarization en-
hancements can be achieved by exploiting the full broad elec-
tron linewidth for DNP via more sophisticated quantum control
on the electron spins, wherein every electron "packet" directly
contributes to the DNP process [21].
In this paper we shall
demonstrate a strategy to achieve this for the case of strongly
anisotropic radicals in the limit of low (dilute) concentrations,
where inter-electron couplings can be neglected – a situation
pertinent for a wide class of nitroxide radicals, and endoge-
nously radicals native to several systems [22, 23]. Since savings
in experimental time scale ∝ ε2, methods to increase hyperpo-
larization efficiency will directly translate to accelerated spec-
troscopy and imaging. Indeed, a surge in recent interest in DNP
2
Figure 2. 13C Hyperpolarization in diamond powder. (A) Sequence
of events. Room temperature 13C DNP from optically pumped NV
centers is achieved via microwave sweeps at low field B=1-30mT un-
der continuous 532nm optical illumination. Bulk polarization is in-
ductively detected by sample shuttling to 7T. (B) Sweep rate. DNP
dependence on MW sweep rate shows an optimal ω set by adiabaticity
constraints. Inset: Dependence on Rabi frequency Ωe. (C) DNP Mech-
anism. Energy levels of a NV electron and a single 13C nuclear spin
hyperfine-coupled with A (grey box), in the low field regime where
nuclear Larmor frequency ωL (cid:46) A. For simplicity the ms = +1
manifold is not shown. Swept microwaves lead to sequential excitation
of LZ crossings and consequent 13C hyperpolarization. For an NV cen-
ter coupled to multiple 13C nuclei one obtains a broadened ESR line by
∆f. Sweeping over any window (shaded) leads to hyperpolarization
with the signal proportional to the local density of states. (D) Diamond
lattice in a microparticle. NV axes are randomly oriented with respect
to the field B.
on the DNP process, since an increasing bandwidth B leads to a
longer period T . For instance, for the typical case of TEMPO at
3.35T and 50K, B ≈ 0.5GHz and considering Ωe=1MHz [9, 35],
T =500ms, which far exceeds the inherent repolarization time,
T (cid:38) B/Ω2
e (cid:29) T1e ≈ 1ms [9, 20].
In this paper, we demonstrate a simple method to overcome
this bottleneck, increasing the effective number of polarization
transfer events while maintaining the optimal adiabatic sweep
rates set by Landau-Zener conditions. Our method involves a
swept microwave frequency-comb, that coherently and simul-
taneously sweeps the entire electron linewidth B at
ω, while
maintaining adiabaticity for each sweep over an individual elec-
tron packet (see Fig. 1B). This allows repeated polarization
transfer from each successive sweep of the comb, allowing
one to gain a multiplicative DNP enhancement boost.
Intu-
itively, the individual comb teeth can be as close as the elec-
tron linewidth ∆f = 1/T2e in frequency, and can sweep each
electron packet as often as trepol, allowing an enhancement gain
ε → N ε. Since we work in the dilute electron limit, this elec-
tron packet linewidth predominantly arises due to hyperfine in-
teractions with the surrounding nuclei. For the case of 10mM
TEMPO for instance, this corresponds to a comb teeth sepa-
ration of 1/TM = 66kHz [38], where TM is the phase mem-
ory time. Note that TM serves here as a lower bound for the
Figure 1. Frequency comb enhanced DNP. (A) Hyperpolarization
processes via frequency/field swept techniques are effectively Landau-
Zener traversals of level anti-crossings in a dressed electron-nuclear
basis. The electron is repolarized every trepol, and sweeps with an adi-
abatic scan rate ω lead to polarization transfer. Eg refers to the energy
gap. Levels shown here are for the NV center - 13C spin system that
(B) Principle. Microwave
we shall consider in detail in this paper.
frequency comb sweeping the entire inhomogeneously broadened elec-
tron spectrum (with linewidth B) allows a repeated polarization transfer
event with every successive comb frequency, and produces a multiplica-
tive boost in DNP enhancement. Red shaded area shows the spectrum
for NV center electrons in diamond powder at 35.9mT. The N comb
frequencies can be as close as ∆f, the hyperfine mediated linewidth
Inset:
(shaded), and sweep every electron packet as often as trepol.
similar method could be applied to broad line electron radicals like
TEMPO, shown here 3.35T with the spectrum centered at 95GHz [9].
Lower panel: time domain implementation through multiple cascaded
frequency sweepers, illustrating the ability to maintain the adiabatic
rate ω while increasing the effective number of sweeps by N.
control techniques has been fueled by advances in instrumen-
tation (sources [24, 25] and synthesizers [26]) that enable the
rapid and coherent manipulation of electrons at high fields [27–
29]. Particularly attractive amongst them is the use of frequency
or field swept techniques, e.g. integrated solid effect (ISE) [30–
32], that are suited to exploiting the wide electron bandwidth
while only requiring modest microwave power.
Principle: – The DNP process underlying these techniques
can be simplistically described as traversals of a level anti-
crossing (LAC) in an electron-nuclear dressed basis (Fig. 1).
Polarization transfer occurs via Landau-Zener (LZ) tunnel-
ing [33], the onus of thermal contact being placed on main-
taining adiabaticity during the sweep [34]. The DNP transfer
efficiency, governed by the tunneling probability, is given by
ε ∝ exp(−E2
g / ω), where ω is the sweep rate and the Eg is
the effective energy gap, and depends on several parameters in-
cluding the electron Rabi frequency Ωe [30], hyperfine coupling
to the target nucleus, and orientation. Despite harnessing the full
electron linewidth, the frequency sweeps are often slow, and the
requirement of adiabaticity sets bounds on the rate of polariza-
tion transfer. To illuminate this in more detail, let us assume an
inhomogeneous electron linewidth B, leading to a single traver-
sal time T = B/ ω. Each electron frequency packet, however,
has repolarized within a time trepol ≤ T1e (cid:28) T , and is available
again for DNP transfer, but instead has to wait the full period T
when the subsequent sweep leads to the next polarization trans-
fer event. Since the nuclear polarization is proportional to the
total number of sweeps T1n/T , the slow sweeps set a bottleneck
ω·trepolABHyperpolarizedTT1nωFrequencyTime (ms)····MW Sweepω·/NΔfΔfFrequency (MHz)-2000200TEMPO···BθNH3CH3CCH3CH3OFrequency/FieldRot. Frame Energy Offset-1e,β0e,α0e,α-1e,βEgFrequency (GHz)~2 GHz at 35.9 mT2345SweptComb/N050100150200(MHz/ms)020406080Signal Enhancement over 7T~40MHz/msNVtrepol∆f� AωLms=0ms=-1∆ - γeBcosθ0,α0,α-1,β -1,β648 msSampleFieldShuttlingTime MW SweepPolarizationTransfer LaserLow Field 1-30 mT7TNMR n ~3000 cycles ~20 ms/MHzω·ABCSweep WindowMultiple Coupled 13CSingle Coupled 13CVVBNNθDiamond Lattice in Particle13C13CDNV axisMW~Enhancement Ωe (a.u.)10-310-210-1104080 3
Figure 4. Multiplicative DNP gains by frequency combs. (A) Indi-
rect mapping of ESR lineshape. Exemplary NV center powder pattern
at B=27.7mT indirectly obtained via 13C DNP on a 100MHz window
swept across the ESR line. Obtained signal is a convolution of the un-
derlying electron spectrum with the sweep window. The ESR spectrum
is orientationally broadened to B ∼1.8GHz. Sign of hyperpolarization
is identical for the ms = ±1 manifolds and depends only on the direc-
tion of the MW sweep. (B) Enhanced DNP gains. Upper panel shows
the multiplicative boost in the DNP enhancements employing a cascade
of upto 3 microwave sweepers, over a 700MHz bandwidth at B=13mT.
Lower panel: obtained hyperpolarized 13C spectra at 7T after 30 av-
erages. (C) Bandwidth dependence. We characterize the multiplicative
gain factor employing (upto) 4 cascaded sweepers, normalized by the
use of a single one over the same band. Sweep bandwidths are centered
at 2.8GHz in these experiments.
NV center axis in each crystallite, allowing the hyperpolariza-
tion of the entire high surface area powder [36]. We evaluate
the obtained hyperpolarization by benchmarking the polariza-
tion enhancement against the room temperature thermal equilib-
rium signal at 7T by sample shuttling (see Fig. 2A). We note that
the sample shuttling time (≈648ms) is longer than the electron
T1e, and to a good approximation therefore the NV electrons are
unpolarized during NMR readout. The shuttling time is how-
ever small compared to the nuclear 13C lifetime, T1n >120s at
B >100mT. Fig. 2B details the typical dependence on MW
sweep rate ω and Rabi frequency Ωe, both of which have opti-
mal values set by adiabaticity constraints of the underlying mi-
croscopic DNP mechanism.
To intuitively understand the main features of the DNP mech-
anism, let us first consider the energy level structure of an NV
center coupled to a single 13C nuclear spin (Fig. 2C and Fig.
3). We work at low fields where ωL < A, and the domi-
nant nuclear quantization axis in the ms = ±1 manifolds are
set by the hyperfine coupling, referred to as β↑,↓ in Fig. 2C.
The ms = 0 state is magnetically silent, and in that man-
ifold the nuclear eigenstates α↑,↓ are dominated (for weakly
coupled 13C) by the Zeeman field with a second order correc-
∆−γeB cos ϑ, where
∆=2.87GHz is the zero-field splitting and ϑ is the angle from the
applied field to the N-V axis (Fig. 2D). Given our detected 13C
linewidths (<1kHz), it is these relatively weakly coupled 13C
tion from the hyperfine field, (cid:101)ωL ≈ ωL + γeBA sin ϑ
Figure 3. Mechanism of polarization transfer. Calculated energy di-
agram in the rotating frame for an NV electron spin hyperfine-coupled
to a single 13C nuclear spin (grey box in Fig. 2C), and corresponding
to the mS = 0 ↔ mS = −1 subset of transitions [36, 37]. (A) and (B)
panels assume a hyperfine coupling Azz = +0.5MHz and −0.5MHz
respectively, and B=10 mT, ϑ=45 deg., and use a transverse hyperfine
constant Azx = 0.3Azz. Colored solid circles denote populations at
different stages during a sweep in the direction of the arrow, and faint
dashed circles indicate the narrower LACs where population transfer
takes place. Panel illustrate that sweeping from low-to-high frequen-
cies in the ms = −1 manifold results in buildup of hyperpolarization
in a direction aligned with the magnetic field for positive Azz (and anal-
ogous for ms = +1 manifold and negative Azz).
electronic T2e, since phenomena such as spectral and instanta-
neous diffusion reduce TM relative to T2e [39]. Without hav-
ing to take into account specific details of the DNP mechanism
in operation, one could bound the maximum enhancement gain
N ≤ min(cid:8)B/∆f,B/ ωtrepol ≈ B/Ω2
(cid:9). The payoffs in hy-
eT1e
perpolarization enhancements stemming from this multiplica-
tive boost can be significant – for TEMPO it could exceed an or-
der of magnitude. More importantly, since the microwave power
for each sweep remains Ωe, the technique can be relatively eas-
ily implemented with existing technology – the frequency comb
being constructed by time-cascading sweeps from N separate
low-power amplifiers (Fig. 1B).
While more generally employable, in this paper we demon-
strate its application to 13C hyperpolarization in diamond parti-
cles via optically polarized electron spins associated with Nitro-
gen Vacancy (NV) center defects. We have recently developed
a method for optical 13C DNP in powdered diamond at room
temperature [36], employing a combination of laser and swept
microwave irradiation at low magnetic fields (B ∼ 1-30mT).
The DNP mechanism itself is a low-field complement to ISE,
working in the regime where ωL < A, where ωL = γnB is the
nuclear Larmor frequency. The NV centers are inhomogenously
broadened to a powder pattern with bandwidth B ≈ 2γeB, and
here too the slow rate of microwave sweeps over B limit the
overall achievable nuclear polarization – a challenge we over-
come by the use of frequency combs.
Low-field 13C DNP in diamond powder: – To be more con-
crete, Fig. 2A presents the hyperpolarization sequence. Laser ir-
radiation polarizes the NV centers to the ms = 0 state, a feature
that occurs independent of field. We estimate the resulting NV
electron polarization to be close to 100% in our experiments. Si-
multaneously applied swept microwave (MW) irradiation (with
Rabi frequency Ωe (cid:46) ωL) causes the transfer of polarization to
13C nuclei in the surrounding lattice. As described in Fig. 2A
the frequency-swept MW is applied in a sawtooth pattern for
∼60s. The DNP occurs independent of the orientation of the
MW Frequency (MHz)-0.8-0.400.40.8Eigen-energies (MHz)A-0.8-0.400.40.8B0,α0,α0,α0,α-1,β-1,β-1,β-1,β0,α0,α0,α0,αInitialAfter LZ crossingsAfter NV- optical repumpingPopulations-1,β-1,β-1,β-1,β266826722676266826722676Norm. Enhancement GainSweep Bandwidth (MHz)No. of Sweepers128003460044003200201CMultiplicative Gain3x1 sweeper2 sweepers3 sweepers060120Signal (a.u.)2060100Enhancement over 7TBAFrequency (GHz)2 GHzMultiplicative Gainms= -1922 Hz3.775GHzSignal (a.u.)22.533.54020406080100120140100 MHzSweepWindowms= +1 4
While considering the more realistic scenario of multiple 13C
nuclei coupled to the NV center, one obtains a continuum of
levels stemming from the hierarchy of the hyperfine interac-
tions [41], the closeby 13C's dominating the spectral widths.
The density of states reflect the underlying hyerfine-broadened
electron linewidth (see Fig. 6A). Sweeping over any small spec-
tral window in the broadened line (Fig. 2C) still leads to hyper-
polarization the sign of which depends on the direction of sweep
(see Fig. 4A).
Even with this brief description, it is already apparent why
the hyperpolarization is inefficient with a single sweeper. The
electron resonance frequencies ∆ ± γeB cos ϑ are orientation
dependent and in a randomly oriented powder the ESR spectrum
is broadened to B = 2γeB ≈ 1.12GHz at 20mT. This is shown
in Fig. 4A for example, where we indirectly map the NV center
ESR spectrum at 27.7mT from the 13C hyperpolarization en-
hancement by performing DNP over small (100MHz) windows
swept across in frequency space. The obtained spectrum is a
convolution of the ESR spectrum with the employed sweep win-
dow, the two extremities of the spectrum correspond to the zero
degree orientations. The experiment in Fig. 4A was performed
on a collection of ≈300 diamond microparticles (Element6) of
200µm size containing a natural abundance (1.1%) 13C and
≈1ppm of NV centers. Since both the ms = −1 and ms = +1
manifolds contain all the NV center electron packets, it is suffi-
cient to just sweep over one of them to obtain the optimal hyper-
polarization on the 13C nuclei. However, the sweep widths re-
quired in the ms = −1 manifold, spanning the 0◦ (at frequency
f0 = ∆ ∓ γeB) and 90◦ (f90 = 1
center orientations are still rather large (614MHz at 20mT). Due
to fixed sweep rates ω constrained by adiabaticity, the large B
leads to a long MW sweep time T = B/ ω ≈16ms> trepol that
far exceeds the repolarization time and bottlenecks the DNP en-
hancement. Given the laser power employed in our experiments
≈80mW/mm2, we estimate trepol ∼1ms, on the same order as
T1e [42]. The exact trepol is challenging to measure especially
on account of optical scattering, total internal reflections, and
NV center charge dynamics.
2 [∆ +(cid:112)∆2 + (2γeB)2]) NV
Swept frequency combs for multiplicative DNP gain: – Fre-
quency combs provide an elegant means to overcome these bot-
tlenecks, decoupling the rate at which the NV centers are swept
over, and the effective rates at which the LZ anti-crossings are
traversed for polarization transfer to the 13C nuclei. Indeed, a
swept microwave frequency comb can maintain the adiabaticity
constraints for a single sweep while increasing the cumulative
number of sweeps in the total DNP period bounded by nuclear
relaxation time T1n. Moreover, in experiments where the NV
repolarization rate trepol ∼ T1e, the swept frequency comb can
ensure that the NV electrons are swept over sufficiently slowly
so as to maximize the NV electron polarization at every sweep
event.
Microwave frequency combs can be constructed by semicon-
ductor lasers under negative optoelectronic feedback [43, 44]
and nonlinear mixing in tunneling junctions [45].
In this
paper we follow a more brute-force approach instead, time-
cascading MW sweeps generated by N voltage controlled oscil-
lator (VCO) sources [40]. Fig. 4B shows the effect of employing
a frequency comb for DNP in the NV-13C system. The DNP en-
hancement gains are significant, scaling linearly with N, and al-
lowing a multiplicative boost to the DNP enhancement over 7T
from 30x to 100x. This constitutes an order of magnitude de-
Figure 5. Characterizing limits to multiplicative DNP gains by deter-
mining the optimal frequency separation for an N sweeper comb over
a B=400MHz bandwidth (ms=-1 manifold) and fixed optimal sweep
rate ω =39MHz/ms. Inset: shaded region denotes sweep bandwidth
over the experimentally obtained powder pattern at 12mT. Sweepers
are time cascaded with the ramps in Fig. 1B shifted by a phase ϕ (up-
per axis), translating to a frequency separation of the comb teeth by
f = (ϕ/2π)B (lower axis). We study 13C DNP enhancements for
(A) 2 and (B) 3 sweepers, with phases set to {0, ϕ} and {−ϕ, 0, ϕ}
respectively. Insets show them explicitly as phasors and in frequency
domain. Data demonstrates that time ramps should be optimally phase
shifted by ϕopt = 2π/N for maximum DNP gains. As expected when
ϕ = 0 there is no gain in using N sweepers over a single one (lower
dashed line), and for ϕ = 180◦ two and three sweepers provide similar
enhancement (upper dashed line). Solid lines are guides to the eye.
nuclei, 300kHz (cid:46) A (cid:46) 1MHz that participate most strongly in
the hyperpolarization process [36] (see also Supplemental Infor-
mation [40]).
Crucially, the large separation between the nuclear eigen-
states in the ms = ±1 manifolds and the low MW powers
employed ensure that the swept MWs sequentially excite a set
of transitions that drive the polarization transfer. This mani-
fests as a pair of LZ crossings in the rotating frame. Fig. 3
(reproduced from [36]) shows this for this in the ms=-1 mani-
fold considering positive and negative Azz hyperfine couplings,
where crossings occur between the states 0, α↑(cid:105) ↔ −1, β↑(cid:105)
and 0, α↓(cid:105) ↔ −1, β↓(cid:105). Under the condition that one is adi-
abatic with respect to the larger energy gap and positive Azz
(see Fig. 3A), traversal through the level-anticrossings leads to
a complete (bifurcated) transfer of population starting from the
(cid:11), causing a bias in the system that hypepolarizes
states(cid:12)(cid:12)0, α↑(↓)
the nuclei to the state α↑. This simple model also captures why
experimentally we find that the 13C DNP sign depends only on
the direction of the microwave sweep, hyperpolarized aligned
(anti-aligned) to B under microwave sweeps from low-to-high
(high-to-low) frequencies [36]. While the laser is applied simul-
taneously with the MW sweep, it is of sufficiently low power
that the optical repolarization of the NV takes place far away
from the LZ events, predominantly during the longer intervals
separating successive sweeps - the laser serving just to reset
the NV center to the ms = 0 state. The optimal sweep rates
(see Fig. 2B) are set by adiabaticity constraints that maximize
the differential LZ transfer probability between the two pairs of
level-anticrossings, and depend both on the Rabi frequency as
well as on the hyperfine coupling and orientation (see [40]). A
more detailed description of the exact energy gaps, and numeri-
cal evaluation of the adiabaticity requirements will be presented
in a forthcoming publication [37].
Phase Ф (deg)Signal (a.u.)045901351800100200300045901351800Ф≡1sweeper≡ 2 sweepers≡ 3 sweepers≡ 3 sweepers≡ 2 sweepers≡1 sweeper050100150200050100150200Comb Teeth Seperation (MHz)Frequency (GHz)Signal (a.u)2.42.83.2050100(Ф/2π)·-ФФFrequency (GHz)Signal (a.u)2.42.83.2050100(Ф/2π)·0 5
The cascaded sweeps entail an increase of the total microwave
power seen by the sample. For DNP mechanisms (eg.
ISE)
where the energy gap (see Fig. 1) is predominantly deter-
mined by the electron Rabi frequency, employing a higher MW
power leads to a faster ω, and the same gains in principle can
be achieved by the use of a single sweeper with higher power.
However even in this case, there are several technological ad-
vantages of using swept frequency combs for DNP. The costs of
MW sources and amplifiers scale rapidly (≈ quadratically) with
power [46, 47], but employing a cascade of N low-power ampli-
fiers leads to only a linear cost scaling. Moreover it is easier to
directly synthesize slower frequency sweeps [24], for instance
using inexpensive AWGs and mixers (see Supplemental Infor-
mation [40]). Our frequency comb method allows one to harness
several slow, low-power, sweeps to gain the advantages of more
expensive high-power platforms, an advantage especially perti-
nent at high fields. Moreover, the technique highlights the in-
herent merits of frequency swept modalities as opposed to field
swept ones; while they are equivalent for a single sweep [30],
when cascaded into swept frequency combs the former can pro-
vide multifold DNP gains.
Limits of multiplicative DNP gains: – Let us finally evaluate
the factors affecting the ultimate limits to the multiplicative en-
hancement gain. In Fig. 4C, for a fixed ω, we vary the sweep
bandwidth, equivalent to bringing the frequency comb sweeps
closer in frequency and time. We measure the multiplicative
gain by normalizing the signal of N sweepers against that from
a single one. We observe that when the frequency comb teeth
are separated by under ≈50MHz, there is first a saturation in the
DNP boosts and subsequent drop. We ascribe this to the inher-
ent limit set by hyperfine mediated electron broadening ∆f in
the powder pattern (see also Fig. 6). ∆f here being the width of
each individual NV center electron packet. When two sweeps
occur simultaneously on different parts of the ESR line corre-
sponding to a single NV center (Fig. 2C), there is interference
between them and consequently lower efficiency in the hyper-
polarization transfer. While we do expect that the comb teeth
separation is ultimately also limited by Rabi frequency, in our
experiments Ωe ≈430kHz, and it does not play a significant role
in setting the limits in Fig. 4C.
Similar experiments allow us to quantify the optimal spacing
between successive comb teeth. In Fig. 5 we change the phase ϕ
between the time-cascaded ramps (e.g. in Fig. 1B) that generate
the swept frequency combs. More intuitively, this phase directly
corresponds to the frequency separation between the successive
comb teeth as indicated in the lower panel of Fig. 5. Note that
the phase of the MWs are arbitrary, here we refer to the phase of
the sawtooth patterns generating the frequency sweeps. Chang-
ing phase has the effect of varying the frequency comb teeth
separation over the fixed sweep bandwidth by ϕB/(2π) and the
time period between successive electron sweeps by ϕB/(2π ω).
Intuitively, one would expect that the trepol and ∆f limits would
require that the comb teeth be maximally separated in both fre-
quency and time, entailing a frequency separation of B/N, and
phase separation ϕopt = 2π/N. Fig. 5 confirms this simple pic-
ture. Interestingly, it also demonstrates how the enhancement
gains arise from the use of multiple sweepers. When all ramps
have the same phase, the enhancement from the comb is identi-
cal to that employing a single sweeper (dashed line in Fig. 5),
increasing as the ramps are phase-shifted, with the expected op-
timal DNP gains at phase separation ϕopt. The plateaus in Fig.
Figure 6. ESR linewidth limits to enhancement gain elucidated by
performing 13C DNP on single crystals with [100] axis placed parallel
to the field B. All NV axes are then at the magic angle ϑM =54.7◦ to
B (inset). (A) ESR lineshape of a 10% enriched 13C crystal mapped
indirectly via DNP on a 25MHz window at B=10.5mT. Blue (yellow)
points show obtained DNP enhancements sweeping MWs from low-
to-high (high-to-low) frequency with ω = 21MHz/ms (insets show
schematic ramps for an exemplary window). Hyperpolarization sign
depends on the direction of MW sweep, and the mirror symmetry of
the lineshapes reflect the local density of states (Fig. 2C). Black line
is the difference signal and faithfully represents the hyperfine broad-
ened NV center ESR spectrum with ∆f dominated by coupling to first
shell 13C nuclei, hyperfine coupled by ∼ 130MHz. (B) Effects of 13C
enrichment. Indirectly mapped ESR spectra of 1% , 3% and 10% 13C
enriched single crystals under DNP with low-to-high frequency sweeps,
showing increasing ∆f (legend: FWHM) with enrichment. (C) Cas-
caded sweeps over a hyperfine broadened line. For the 10% 13C single
crystal in (A), we perform DNP with one and two cascaded sweepers
over a varying bandwidth centered at 2.688 GHz (peak of (a) in A).
Results indicate that frequency comb teeth have to be separated beyond
∆f to provide enhancement gains. Inset: Optimal sweep bandwidths
while employing one and two cascaded sweepers for crystals of differ-
ent 13C enrichment.
crease in averaging time for the same SNR. In the experiments,
all the sources sweep the entire bandwidth B, and the frequency
ramps are time-shifted by B/(N ω) so as to maximize the period
between successive sweeps (Fig. 1B). While one could consider
an alternate scenario where B is partitioned into N sub-bands
which are swept by individual sources, the current strategy per-
forms better since the electrons at the boundaries between the
frequency partitions are also swept across completely, as re-
quired for optimal LZ population transfer. Certain implementa-
tional aspects deserve brief mention. The individual VCOs have
slightly different frequency-voltage characteristics, and to cas-
cade them effectively we match their exact sweep bandwidths to
within ∆f < 1MHz via a fast-feedback algorithm (see Supple-
mentary Information [40]). The sources are then power com-
bined and amplified, with the amplifier operating well below
compression to prevent inter-modulation distortion (IMD) arti-
facts [40].
Sweep Bandwidth (MHz)Signal (a.u)01002003004000200400600800100012001 sweeper2 sweepers10%137MHzCEnrichment(%)Optimal Bandwidth(MHz)131050150250262MHz-200-1000100200Frequency Deviation (MHz)∆f-0.200.20.40.60.81BNormalized Signal ∆f = 38 MHz ∆f = 62 MHz ∆f = 130MHz10%3%1%Increasing∆fSignal (a.u)2.52.62.72.82.9-400-2000200400600AB[1 0 0]NV axis[1 1 1]25 MHz10% single crystal2.7552.78MW (GHz)1.17 ms 2.7552.78MW (GHz)1.17 ms TimeTimea.b.a-bab54.7°Frequency (GHz)130MHz 5 indicate that the enhancement gains are achievable as long as
the comb teeth are separated beyond ∆f.
To make this more concrete, in Fig. 6 we perform DNP on
single crystals with different 13C enrichment. The crystals have
≈1ppm NV centers, and since we are in the dilute electron limit,
the electron packet linewidths are dominated by couplings to the
13C nuclei. Moreover, the crystals are oriented parallel to the
[100] direction such that all N-V axes are equivalent and at the
magic angle to the polarizing field B and have the same fre-
quency, hence eliminating inhomogeneous broadening. This is
most evident in Fig. 6A that demonstrates the electron spectrum
mapped via 13C DNP, evidenced by the mirror symmetry in the
obtained DNP signals with opposite sweep directions. We note
that while we had considered the theory of the hypepolariza-
tion mechanism in the context of weakly coupled 13C nuclei that
contribute predominantly to the bulk NMR signal that we mea-
sure, Fig. 6A also provides direct insight into strongly coupled
first shell nuclei. The asymmetry in the obtained ESR spectra
directly reports on the polarization of the first shell 13C spins.
Indeed, the difference signal obtained from alternate sweep di-
rections (black line in Fig. 6A) shows the characteristic ESR
spectrum with satellites from first-shell 13C nuclei strongly hy-
perfine coupled by ∼130MHz [48, 49].
The fact
that hyperfine couplings dominate the ESR
linewidths (Fig. 2C) are most evident in Fig. 6B, where we
measure ∆f with increasing 13C enrichment. This is in con-
trast to Fig. 4A where the ESR spectrum was inhomogenously
broadened due to different orientations of the NV centers in a
random powder. In Fig. 6C we study the DNP enhancements
with one and two cascaded sweepers for varying sweep band-
widths and fixed ω over these hyperfine broadened lines, choos-
ing as a representative example the 10% enriched sample stud-
ied in Fig. 6A. The sweep bandwidths in these experiments are
centered at the peak of the ESR spectrum. Let us first consider
the case of a single sweeper (blue line in Fig. 6C). The en-
hancement increases with sweep bandwidth, reaching an opti-
mal value when B ≈ ∆f, corresponding to the MWs being ap-
plied most efficiently over the electron spectrum. While employ-
ing two sweepers on the other hand, there are no DNP enhance-
ment gains when the comb frequencies are closer than ∆f. Note
that in these experiments we space the comb teeth (optimally) by
half the bandwidth. The maximum enhancement occurs when
the comb separation is ∆f, corresponding to a total sweep band-
width of 2∆f, a strong indication that the two sweepers interfere
with each other when simultaneously employed on the hyperfine
broadened electron line. Performing similar experiments on the
samples with different 13C enrichment allows us to quantify the
sweep bandwidths at which two sweepers perform better than a
single one (see Supplemental Information Fig. S2). The optimal
sweep bandwidths for one and two sweepers are elucidated in
the inset of Fig. 6C, and they closely match the intrinsic ∆f
linewidths in Fig. 6B,scaling with 13C enrichment. Overall Fig.
5 and Fig. 6 demonstrate the inherent constraints of the tech-
nique, and in the ultimate limit, the frequency combs approach
an excitation of all ∆f-wide electron packets at once, sweeping
6
them as often as trepol – approaching the efficiency of a pulsed
DNP experiment over the entire electron bandwidth.
Applications to conventional DNP: – Let us now describe
how the current method can be applied in the context of DNP
with electron radicals. The most direct applications are for
DNP in systems with endogenous radicals (eg. Si, diamond sur-
faces [50]) or where they can be optically excited [51], since
the linewidths of such radicals are hard to control. For radi-
cals with a large g-anisotropy (e.g. TEMPO, Galvinoxyl) and
low-concentrations, the large inhomogeneous broadening leads
to a inefficient transfer and lower DNP enhancement due to the
(differential) solid effect. Indeed, experiments are typically per-
formed at higher radical concentrations (>20mM [38]), where
DNP can occur via the cross effect [52], with significantly faster
growth times. However the higher radical concentrations lead
to a broadening of the observed NMR lines [53] and are a chal-
lenge for high-resolution spectroscopy applications [54]. The
use of swept frequency combs can dramatically improve DNP
enhancement at low radical concentrations by enacting a transi-
tion to the integrated solid effect [55]. Since ISE is bottlenecked
by similar factors, the demonstrated gains in Fig. 4 should be
directly transferable to ISE. The use of frequency modulation to
implement ISE was recently demonstrated at X-band [29], and
we anticipate large gains with frequency combs. While the in-
creased reliance on spin diffusion will increase the growth time
of the DNP signal, frequency combs offer the possibility of ob-
taining high DNP enhancements without concomitant NMR line
broadening.
Conclusions and outlook: – We have proposed and experi-
mentally demonstrated a simple and scalable technique to ob-
tain multiplicative enhancement gains in dynamic nuclear po-
larization. The method entails a swept frequency comb to ex-
cite the entire inhomogenously broadened electron bandwidth
for polarization transfer. It can be implemented by cascading
N sweeps from individual low-power sources/amplifiers to ob-
tain a DNP enhancement boost ∝ N, with ultimate limits set
by the hyperfine-mediated electron linewidth and lifetime T1e.
As such the technique affirms the notion the electron spin con-
trol can significantly enhance DNP by harnessing the full power
of the electron spectrum. We demonstrated its utility for the
hyperpolarization of 13C nuclei in diamond microparticles via
optically pumped NV centers at room temperature, obtaining a
300% boost in DNP efficiency. When employed for conven-
tional polarizing radicals at high fields, the technique promises
to yield DNP enhancement boosts in excess of one order of mag-
nitude, with a relatively simple implementation employing ex-
isting technology and only modest cost overheads.
Acknowledgments: – We gratefully acknowledge discussions
with J.P. King, P. R. Zangara and S. Dhomkar. CAM acknowl-
edges support from the National Science Foundation through
grants NSF-1401632 and NSF-1547830 and from Research Cor-
poration for Science Advancement through a FRED Award. CR
acknowledges support from the National Science Foundation
through grant CHE-1410504.
[1] T. R. Carver and C. P. Slichter, Physical Review 92, 212 (1953).
[2] A. Abragam and M. Goldman, Reports on Progress in Physics 41,
395 (1978).
[3] R. Wind, M. Duijvestijn, C. Van Der Lugt, A. Manenschijn, and
J. Vriend, Progress in Nuclear Magnetic Resonance Spectroscopy
17, 33 (1985).
7
[4] T. Maly, G. T. Debelouchina, V. S. Bajaj, K.-N. Hu, C.-G. Joo,
M. L. MakJurkauskas, J. R. Sirigiri, P. C. A. van der Wel,
J. Herzfeld, R. J. Temkin, and R. G. Griffin, The Journal of Chem-
ical Physics 128, 052211 (2008).
[5] E. R. McCarney, B. D. Armstrong, M. D. Lingwood, and S. Han,
Proceedings of the National Academy of Sciences 104, 1754
(2007).
[6] B. E. Kane, Nature 393, 133 (1998).
[7] S. Foletti, H. Bluhm, D. Mahalu, V. Umansky, and A. Yacoby,
Nature Physics 5, 903 (2009).
[8] A. Lesage, M. Lelli, D. Gajan, M. A. Caporini, V. Vitzthum,
P. Mi´eville, J. Alauzun, A. Roussey, C. Thieuleux, A. Mehdi,
et al., J. Am. Chem. Soc 132, 15459 (2010).
[9] A. Feintuch, D. Shimon, Y. Hovav, D. Banerjee, I. Kaminker,
Y. Lipkin, K. Zibzener, B. Epel, S. Vega, and D. Goldfarb, Journal
of Magnetic Resonance 209, 136 (2011).
[10] M. Goldman, Spin Temperature and NMR in Solids. (Clarendon
[11] P. Song, Z. Zhao, X. Xu, P. Deng, and J. Xu, Optical Materials
Press, Oxford, 1970).
29, 471 (2006).
[12] K.-N. Hu, V. S. Bajaj, M. Rosay, and R. G. Griffin, The Journal
of chemical physics 126, 044512 (2007).
111, 7527 (2014).
[33] C. Zener, Proc. R. Soc. Lond. A 137, 696 (1932).
[34] J. R. Rubbmark, M. M. Kash, M. G. Littman, and D. Kleppner,
Physical Review A 23, 3107 (1981).
[35] T. A. Siaw, A. Leavesley, A. Lund, I. Kaminker, and S. Han,
Journal of Magnetic Resonance 264, 131 (2016).
[36] A. Ajoy, K. Liu, R. Nazaryan, X. Lv, P. R. Zangara, B. Safvati,
G. Wang, D. Arnold, G. Li, A. Lin, et al., Science Advances 4,
eaar5492 (2018).
[37] P. Zangara, S. Dhomkar, A. Ajoy, K. Liu, R. Nazaryan,
D. Pagliero, D. Suter, J. Reimer, A. Pines, and C. Meriles, (to
be published) (2018).
[38] T. A. Siaw, M. Fehr, A. Lund, A. Latimer, S. A. Walker, D. T.
Edwards, and S.-I. Han, Physical Chemistry Chemical Physics
16, 18694 (2014).
[39] A. Schweiger and G. Jeschke, Principles of pulse electron param-
agnetic resonance (Oxford University Press on Demand, 2001).
[40] See supplementary online material.
[41] A. Dr´eau, P. Jamonneau, O. Gazzano, S. Kosen, J.-F. Roch,
J. Maze, and V. Jacques, Physical review letters 113, 137601
(2014).
[42] A. Jarmola, V. Acosta, K. Jensen, S. Chemerisov, and D. Budker,
[13] B. D. Armstrong and S. Han, The Journal of chemical physics
Physical review letters 108, 197601 (2012).
127, 104508 (2007).
[14] J. Wrachtrup and F. Jelezko, J. Phys.: Condens. Matter 18, S807
[43] Y.-S. Juan and F.-Y. Lin, Optics letters 34, 1636 (2009).
[44] S.-C. Chan, G.-Q. Xia, and J.-M. Liu, Optics letters 32, 1917
(2006).
shoni,
(2013).
[15] M. W. Doherty, F. Dolde, H. Fedder, F. Jelezko, J. Wrachtrup,
N. B. Manson, and L. C. L. Hollenberg, Phys. Rev. B 85, 205203
(2012).
[16] R. Fischer, C. O. Bretschneider, P. London, D. Budker, D. Ger-
and L. Frydman, Physical review letters 111, 057601
[17] Q. Chen, I. Schwarz, F. Jelezko, A. Retzker, and M. Plenio, Phys-
ical Review B 92, 184420 (2015).
[18] P. Andrich, B. J. Aleman, J. C. Lee, K. Ohno, C. F. de las Casas,
F. J. Heremans, E. L. Hu, and D. D. Awschalom, Nano letters 14,
4959 (2014).
[19] Y. Hovav, A. Feintuch, and S. Vega, Journal of Magnetic Reso-
nance 207, 176 (2010).
[20] D. Shimon, Y. Hovav, A. Feintuch, D. Goldfarb, and S. Vega,
Physical Chemistry Chemical Physics 14, 5729 (2012).
(2007).
[45] M. J. Hagmann, A. Efimov, A. J. Taylor, and D. A. Yarotski,
Applied Physics Letters 99, 011112 (2011).
[46] C. D. Joye, R. G. Griffin, M. K. Hornstein, K.-N. Hu, K. E.
Kreischer, M. Rosay, M. A. Shapiro, J. R. Sirigiri, R. J. Temkin,
and P. P. Woskov, IEEE Transactions on Plasma Science 34, 518
(2006).
[47] V. Denysenkov, M. Prandolini, A. Krahn, M. Gafurov, B. En-
deward, and T. Prisner, Applied Magnetic Resonance 34, 289
(2008).
[48] K. R. K. Rao and D. Suter, Physical Review B 94, 060101 (2016).
[49] A. J. Parker, K. Jeong, C. E. Avalos, B. J. Hausmann, C. C. Vas-
siliou, A. Pines, and J. P. King, arXiv preprint arXiv:1708.00561
(2017).
[50] M. Cassidy, C. Ramanathan, D. Cory, J. Ager, and C. M. Marcus,
Physical Review B 87, 161306 (2013).
[21] Y. Hovav, A. Feintuch, S. Vega, and D. Goldfarb, Journal of Mag-
[51] A. Capozzi, T. Cheng, G. Boero, C. Roussel, and A. Comment,
netic Resonance 238, 94 (2014).
Nature Communications 8 (2017).
[22] M. L. Guy, K. J. van Schooten, L. Zhu, and C. Ramanathan, The
[52] C. F. Hwang and D. A. Hill, Physical Review Letters 19, 1011
Journal of Physical Chemistry C 121, 2748 (2017).
[23] E. Rej, T. Gaebel, D. E. Waddington, and D. J. Reilly, J. Am.
(1967).
[24] M. L. Guy, L. Zhu, and C. Ramanathan, Journal of Magnetic
Chem. Soc 139, 193 (2017).
Resonance 261, 11 (2015).
[25] D. Yoon, M. Soundararajan, P. Cuanillon, F. Braunmueller, S. Al-
berti, and J.-P. Ansermet, Journal of Magnetic Resonance 262, 62
(2016).
[26] I. Kaminker, R. Barnes, and S. Han, Journal of Magnetic Reso-
nance 279, 81 (2017).
[27] D. E. Hoff, B. J. Albert, E. P. Saliba, F. J. Scott, E. J. Choi, M. Mar-
dini, and A. B. Barnes, Solid state nuclear magnetic resonance 72,
79 (2015).
[28] T. Can, J. Walish, T. Swager, and R. Griffin, The Journal of chem-
ical physics 143, 054201 (2015).
[29] T. V. Can, R. T. Weber, J. J. Walish, T. M. Swager, and R. G. Grif-
fin, Angewandte Chemie International Edition 56, 6744 (2017).
[30] A. Henstra, P. Dirksen, and W. T. Wenckebach, Physics Letters A
134, 134 (1988).
[31] A. Henstra, T. S. Lin, J. Schmidt, and W. T. Wenckebach, Chem-
ical Physics Letters 165, 6 (1990).
[32] K. Tateishi, M. Negoro, S. Nishida, A. Kagawa, Y. Morita, and
M. Kitagawa, Proceedings of the National Academy of Sciences
[53] S. Lange, A. H. Linden, U. Akbey, W. T. Franks, N. M. Loen-
ing, B.-J. van Rossum, and H. Oschkinat, Journal of Magnetic
Resonance 216, 209 (2012).
[54] E. P. Saliba, E. L. Sesti, F. J. Scott, B. J. Albert, E. J. Choi,
N. Alaniva, C. Gao, and A. B. Barnes, Journal of the American
Chemical Society 139, 6310 (2017).
[55] P. Dirksen, A. Henstra,
and W. T. Wenckebach, Journal of
Physics: Condensed Matter 1, 8535 (1989).
[56] G. A.
´Alvarez, C. O. Bretschneider, R. Fischer, P. London,
H. Kanda, S. Onoda, J. Isoya, D. Gershoni, and L. Frydman,
Nature communications 6 (2015).
[57] A. Henstra and W. T. Wenckebach, Molecular Physics 112, 1761
(2014).
[58] A. Ajoy, X. Lv, K. Druga, E.and Liu, B. Safvati, A. Morabe,
M. Fenton, R. Nazaryan, S. Patel, T. Sjolander, D. Pagliero, J. A.
Reimer, D. Sakellariou, and A. Meriles, C.and Pines, to be pub-
lished (2017).
Enhanced dynamic nuclear polarization via swept microwave frequency combs
Supplementary Information:
8
CONTENTS
References
I. Linewidth limits of frequency comb DNP
II. Hyperpolarized signal and buildup
III. Hyperpolarization Mechanism
IV. Integrated Solid Effect
V. Experimental design
VI. Electronics for swept MW frequency combs
A. Setup
B. Generating swept frequency combs
C. Intermodulation Distortion limits
D. MW frequency combs at high field
VII. Data Analysis
6
8
8
9
10
10
11
11
11
12
13
13
I. LINEWIDTH LIMITS OF FREQUENCY COMB DNP
In order to highlight the linewidth limitations when employ-
ing multiple cascaded sweepers, we performed detailed exper-
iments mapping the NV center ESR spectrum via 13C DNP in
a narrow 12.5MHz sweep window for samples of different 13C
enrichments (see Fig. S1). Similar to Fig. 6 in the main pa-
per, the crystals are placed with the NV axes at the magic an-
gle to the polarizing field, allowing the obtained spectra to be
broadened predominantly by hyperfine couplings to 13C nuclei.
We ascribe the slight asymmetry in the lineshape of Fig. S1B
to crystal misalignment from the magic angle. The linewidth of
the highly enriched samples are dominated by the hyperfine cou-
pling to closeby 13C. A systematic discussion of the changes in
lineshape is beyond the scope of this work and will be addressed
in a forthcoming publication.
It is evident that the two cascaded sweepers provide a gain
over a single one (right panels in Fig. S1) only when the comb
teeth are separated by approximately the hyperfine mediated
electron linewidth, which increases with 13C enrichment. This
data was also used to report the exact optimal sweep bandwidths
for one and two sweepers, plotted in the inset of Fig. 6C of the
main paper.
Fig. S2 studies a similar bandwidth dependence of DNP
gains with multiple cascaded sweepers for a 1% (natural abun-
dance) microdiamond sample. The sweep bandwidth is centered
at the center of the powder pattern (inset of Fig. S2). Once
again, the enhancement gains are a strong function of the sweep
bandwidth, decreasing sharply when the comb teeth spacing ap-
proaches the hyperfine mediated electron linewidth.
Figure S1. Linewidth limits of multiplicative DNP gains. Left pan-
els: NV center electronic spectrum mapped indirectly via 13C DNP in
a 12.5MHz window for single crystals of (A) 1% (B) 3% and (C) 10%
13C enrichment. Crystals are placed parallel to the [100] axis (inset),
so that all families of NV centers have overlapping spectra. Blue data
indicate the obtained 13C enhancements sweeping microwaves from
low-to-high frequency (inset). Solid lines are guides to the eye. Elec-
tron linewidth ∆f (FWHMs are marked) increases with enrichment
and the sign of 13C hyperpolarization only depends on the direction of
MW sweep. Right panels: Obtained DNP enhancements employing
one (blue) and two (green) sweepers for varying bandwidths over the
electron spectrum. Sweep bandwidths are centered on the spectra in the
left panels. Multiple sweeper combs provide an advantage over a single
sweeper when the comb teeth are separated by more than the electron
linewidth.
II. HYPERPOLARIZED SIGNAL AND BUILDUP
Fig. S3 exhibits essential results from Ref.
[36], demon-
strated for a typical example of 200µm microparticles with nat-
ural abundance 13C containing about 1ppm of NV centers (Fig.
S3C). We obtain 13C hyperpolarization over 116 times that of
the 7T Boltzmann level (Fig. S3A) with a completely randomly
oriented powder. For clarity the signals in Fig. S3A have their
010020030040000.20.40.60.8100.20.40.60.810100200300400010020030040000.20.40.60.81Frequency Deviation (MHz)Sweep Bandwidth (MHz)B[1 0 0]NV axis[1 1 1]54.7°-200-1000100200-200-100010020000.5100.51-200-100010020000.51Signal (a.u.)1%singlecrystalABC3%singlecrystal10%singlecrystalff+0.0125MW (GHz)0.58 ms Time1 sweeper2 sweepers130MHz62MHz38MHz 9
III. HYPERPOLARIZATION MECHANISM
We now briefly describe the low field DNP mechanism that
governs the polarization transfer in our experiments. For more
details, and experimental characterization of the mechanism, we
point the reader to Ref.
[36]. Consider for simplicity a NV
center coupled to a single 13C nuclear spin. The Hamiltonian of
the system is,
H = ∆S2
z − γe (cid:126)B · (cid:126)S − γn (cid:126)B · (cid:126)I + AzzSzIz
+ AyySyIy + AxxSxIx + AxzSxIz + AzxSzIx
(1)
where (cid:126)S and (cid:126)I respectively denote the NV and 13C vector spin
operators, and (cid:126)B is the magnetic field (10-30 mT) at angle ϑ
(ϕ) to the NV axis. Within the ms = ±1 states, the hyperfine
coupling produces a 13C splitting,
C =(cid:112)(Azz ∓ γnB cos ϑ)2 + A2
ω(±1)
zx
(2)
For the ms = 0 manifold, second-order perturbation theory
leads to the approximate formula [56],
(cid:101)ωL ≈ γnB
+ 2
(cid:16) γeB
(cid:17)
∆
sin ϑ(cid:0)(cid:112)A2
zx cos2 ϕ + Ayy sin2 ϕ(cid:1)
xx + A2
(3)
From Eqs. 2 and 3 we conclude that each manifold (includ-
ing the ms = 0 manifold) has its own, distinct quantization
axis which might be different from the direction of the applied
magnetic field. In particular, the second term in Eq. 3 can be
dominant for hyperfine couplings as low as 1 MHz (correspond-
ing to nuclei beyond the first two shells around the NV) if ϑ is
sufficiently large, implying that, in general, 13C spins coupled
to NVs misaligned with the external magnetic field experience
a large frequency mismatch with bulk carbons, even if optical
excitation makes ms = 0 the preferred NV spin state.
Assuming fields in the range 10-30 mT, it follows that 13C
spins moderately coupled to the NV (300 kHz (cid:46) Azz (cid:46) 1
MHz) are dominant in the hyperpolarization process because
they more easily spin diffuse into the bulk and contribute most
strongly to the observed NMR signal at 7T. For sweep rates near
the optimum (∼ 40 MHz/ms), the time necessary to traverse the
set of transitions connecting ms = 0 with either the ms = −1
or ms = +1 manifolds is relatively short(cid:0)(cid:46) 30 µs for weakly
coupled carbons(cid:1) meaning that optical repolarization of the NV
preferentially takes place during the longer intervals separating
two consecutive sweeps, as modeled in Fig. 3.
Nuclear spin polarization can be understood as arising from
the Landau-Zener crossings in Fig. 3. Efficient polarization
transfer takes place when the narrower LZ crossings connect
branches with different electron and nuclear spin quantum num-
bers, precisely the case in the ms = 0 ↔ ms = −1 (ms =
0 ↔ ms = +1) subset of transitions when the hyperfine cou-
pling is positive (negative). When probing ensembles, both sets
of transitions behave in the same way, i.e., 13C spins polarize
positive in one direction, negative in the other. A more detailed
exposition of the hyperpolarization mechanism and simulations
are presented in Ref. [36].
Figure S2. Bandwidth dependence of multiplicative DNP gains in
microdiamond powder. For varying sweep bandwidths centered on the
NV center powder pattern, we obtain the 13C DNP signal employing
a swept frequency comb of upto three cascaded sweepers at ≈20mT.
Inset: Experimentally determined
Solid lines are guides to the eye.
powder pattern. Shaded region denotes an example 600MHz sweep
bandwidth. Panel demonstrates electron linewidth limits on the ob-
tained multiplicative DNP gain (see Fig. S1).
noise unit-normalized, and a single shot DNP signal has about
10 times the signal-to-noise (SNR) of the 7T thermal signal ob-
tained after ≈7hr of averaging – a time gain of over 5 orders of
magnitude to get the identical SNR at 7T with thermal signal.
The polarization within the diamond particles builds up to
≈0.1% polarization level in about 40s of optical pumping (Fig.
S3B). This is reflective of the efficiency of the low field mecha-
nism, being able to hyperpolarize a large number of 13C nuclei
(not just in the first shell). The pumping time is limited by the
nuclear lifetime T1n ≈ 20s at low field (≈ 20mT). The polar-
ization buildup curve exhibits a linear ramp at long times due to
slow spin diffusion away from the directly polarized 13C spins.
Increasing 13C enrichment leads to more rapid spin diffusion
and consequently faster DNP buildup.
In our experiments, the sign of the polarization only depends
on the direction of the microwave sweeps (Fig. S3C). Sweep-
ing the microwaves in a ramp fashion from low-to-high fre-
quency leads to nuclear polarization aligned to the polarizing
B which we term positive polarization. Anti-alignment can be
achieved accordingly by sweeping from high-to-low-frequency.
This allows on-demand control of the sign of polarization. As
expected, a triangular sweep pattern with equal amounts of high-
to-low and low-to-high frequency sweeps leads to destructive in-
terference in alternate periods, and no net polarization buildup.
Indeed, increasing the number of cascaded sweepers N main-
tains the same optimal sweep rate set by adiabaticity constraints,
and hyperpolarization sign dependence on the direction of the
sweep (see Fig. S4).
Bandwidth (MHz)Signal (a.u)020040060080010000501001502002503001 sweeper2 sweepers3 sweepersFrequency (GHz)Signal (a.u)2.42.62.833.23.4050100 10
Figure S3. Optical hyperpolarization in diamond microparticles. Hyperpolarization experiments were performed on 200µm HPHT parti-
cles [36]. Solid fit lines are depicted over experimental data points. (A) Signal gain by DNP under optimized conditions. Green line shows the
13C NMR signal due to Boltzmann polarization at 7T, averaged 120 times over 7 hours. Blue line is a single shot DNP signal obtained with 40s
of optical pumping, enhanced by 116 over the 7T thermal signal (enhanced 101500 times at polarizing field B=8mT). The signals have their noise
unit-normalized for clarity. Hyperpolarization thus leads to over 5 orders of magnitude gains in averaging time (inset). (B) Buildup curve showing
rapid growth of bulk 13C polarization. Slow rise at longer times is reflective of 13C spin diffusion. (C)SEM micrographs (Hitachi S5000) of two
individual e6 HPHT diamond particles. The particles have a uniform size distribution (edge length 87 ± 3.9 µm), and a truncated octahedral
shape set by particle growth conditions. (D) Hyperpolarization sign is controlled by MW sweep direction across the NV center powder pattern.
Continuous family of sawtooth-sweeps demonstrating the concept, varying the duty cycle of upward ramps. Extremal points represent low-to-high
frequency MW sweeps and vice-versa. Inset: 13C signal undergoes near-perfect sign inversion upon reversal of the sweep direction. Sweeping in a
symmetric fashion leads to net cancellation, and no buildup of hyperpolarization.
the electron and nuclei. In the rotating frame,
H = ∆ωSz+ΩeSx−ωIz+SzAzzIz+
SzAz−I+
(5)
where ∆ω = ωe−ω, I± = Ix±Iy, Az± = Azx±Azy. Moving
to a tilted reference frame via a rotation about y-axis, exp(iϑSy)
so that the effective frequency vector (Ωe, 0 , ∆ω) is along the
new zt-axis gives,
SzAz+I−+
1
2
1
2
Figure S4. Hyperpolarization gains with cascaded sweeps occurs
while maintaining the same sweep rate and polarization direction char-
acteristics of the DNP mechanism. Figure illustrates the (A) sweep rate
dependence and (B) sweep direction dependence on the obtained 13C
enhancement while employing upto three cascaded sweepers. The pan-
els indicate that (A) the swept frequency comb provides multiplicative
DNP gains while maintaining approximately the same optimal sweep
rate, and (B) the hyperpolarization sign that only depends on the direc-
tion of the overall microwave sweeps (insets and Fig. S3).
IV.
INTEGRATED SOLID EFFECT
For completeness, here we review the principles of Integrated
Solid Effect (ISE) (see [30, 57]) and the adiabaticity conditions
for optimal polarization transfer (see Fig. 1 of main paper). The
DNP mechanism we employ for the NV-13C system shares sev-
eral implementational similarities with ISE. Under swept MW
irradation with frequency ω, the Hamiltonian of a coupled elec-
tron nuclear spin system is,
H = ωeSz − ωLIz + 2ΩeSx cos(ωt) + S · A · I.
(4)
The first two terms are electron and nuclear Zeeman term re-
spectively, and the the last term is hyperfine coupling between
Ht = Z t + V† sin ϑ
z−S+I i
z+S−I i− + Ai
(6)
where Z t = ωeffSzt − ωLI i
z is the Zeeman part and V † =
− 1
+), assuming that the components of
4 (Ai
the hyperfine tensors are small. In this tilted rotating frame set
s(cid:105)mI(cid:105), V† sin ϑ contributes to non-diagonal term
by basis mt
and induces states transitions. There are then effectively two
level anti-crossings (LACs) at which polarization polarization
transfer from the electron to nuclei can be affected, given by
(cid:112)(∆ω)2 + Ω2 − ωL ≈ 0. Assuming the rate for frequency
e
sweep is ω, the probability of electron spin staying at the same
state equals is exp(− πΩ2
2 ω ). Thus this results in the adiabatic
condition for the frequency sweep πΩ2
2 ω (cid:29) 1.
e
V. EXPERIMENTAL DESIGN
During our DNP process, 1W laser (520nm LasterTack) light
is applied continuously for a fixed time (∼60s) to polarize the
NV centers. Simultaneously applied swept microwave (MW)
irradiation (MiniCircuits ZHL16W-43S+ 16W) across the NV
center spectrum at 1-30mT transfers the polarization to 13C nu-
clei (see Fig. 2A in the main paper). A mechanical field cy-
cler [58] then rapidly carries the sample from the low field mag-
netic shield (NETIC S3-6 alloy 0.062" thick, Magnetic Shield
0204060DNP Time (s)04080120Signal Enhancement over 7TBADNP7T1051041031021011061075 orders of magnitudeTime for equal SNR (s)120 shots120Thermal( )Hyperpolarized (1 shot)at Bpol ~ 8 mTNoise=1-10010Frequency (kHz)02550Signal (a.u.)Enhancement over 7 Tesla = 11600.250.500.751Proportion of High-to-Low Ramp (t/T)-100-50050Signal Enhancement over 7T100Positive PolarizationZero PolarizationNegative Polarization0153045Time (ms)2.63.2MW (GHz)t=15T=150153045Time (ms)2.63.2MW (GHz)T=150153045Time (ms)2.63.2MW (GHz)t=7.5t=0 T=15Signal (a.u)DC50 μm87 μm edgeCSweep Rate (Hz)Enhancement over 7T010020030040002040608000.2500.751Proportion of High-to-Low Ramp-100-50050100Enhancement over 7T1 sweeper2 sweepers3 sweepersABFrequencyTime (ms)···FrequencyTime (ms)··· 11
are delivered by means of a stub antenna (loop) employed below
the tube. The motion and subsequent detection is controlled and
sequenced with pulse generator (SpinCore PulseBlaster USB
100 MHz) using a high voltage MOSFET switch (Williamette
MHVSW-001V-036V). For more details on the device construc-
tion and performance we point the reader to Ref. [58].
VI. ELECTRONICS FOR SWEPT MW FREQUENCY COMBS
A. Setup
Microwave (MW) sweeps are applied across the NV cen-
ter powder pattern to drive the DNP process (see Fig. 2A
in the main paper). For the experiments in this work, we
employed voltage controlled oscillator (VCO) (Minicircuits
ZX95-3800A+ (1.9-3.7 GHz)) sources to generate the frequency
sweeps, by employing DC shifted ramp input voltages to pro-
duce the sweeps (see Fig. S6). The ramp is produced using
a programmable power supply (Circuit Specialists) that gener-
ates a DC voltage Vdc that is combined with an AC sawtooth
ramp with peak to peak value Vpp from an arbitrary waveform
generator (Rigol 1022A), in a high-pass configuration with a
∼ 1Hz cutoff frequency. The input ramps are programmed to
carefully tune the VCO outputs to the target sweep bandwidths
corresponding to the NV center powder pattern [36]. Lastly, af-
ter being power combiened (MiniCircuits ZN4PD1-63HP+), a
16W amplifier (MiniCircuits ZHL16W-43S+) transmits the mi-
crowaves to a stub antenna matched to the diameter of the tube
containing the sample [36] to excite the 13C hyperpolarization.
B. Generating swept frequency combs
Cascaded sweeps utilizing multiple VCOs (NVCO) are gen-
erated by using input voltage ramps that are phase shifted by
2π/NVCO. The VCO output frequency f (V ) and input voltage
V has a approximate linear relationship, f (V ) = b · V + F ,
where b is a constant coefficient and F ≈1.9GHz is the fre-
quency when V = 0. However, the VCOs have slightly differing
f-V characteristics, even when of the same family, due to inter-
device variation and temperature dependence (see Fig. S7). In
order to match all the VCOs to sweep the target band, and to
generate the equally spaced frequency comb, we implemented a
gradient descent feedback algorithm employing a fast spectrum
analyzer (SignalHound USB-SA44B).
to be Vi
Let us define the DC and AC voltage inputs to the VCOs for
dc. We define
the ith iteration (i=1, 2, 3...)
center of the spectrum fi for the ith iteration, and bandwidth of
the spectrum (cid:52)fi, while the target spectrum center and width are
f0, (cid:52)f0 respectively. Given the linearity of the VCO response,
Vdc and Vpp are predominantly related to f and (cid:52)f respectively.
The following equations are applied to update Vpp and Vdc level
pp and Vi
for each iteration:
V i+1
pp =
bi =
pp
(cid:52)f0(cid:52)f i · V i
f i − F
V i
dc
V i+1
dc = Vi +
f0 − f i
bi
(7)
Figure S5. Hyperpolarization setup. Hyperpolarization is carried out
at low field (1-30mT) and the bulk 13C polarization measured at 7T
by rapid sample shuttling. (A) A mechanical shuttler is constructed at
the top of a superconducting magnet (7T) on X and Y adjustable rails
for alignment. Hyperpolarization is performed in a low field shield,
which is secured on sliding rails. An actuator shuttles a carbon fiber rod
with sample tube attached to the end along a conveyer belt for 1630mm
travel distance. (B) Diamond particles are distributed in water, and car-
ried in a glass tube. A plunger holds the contained volume firmly to
ensure samples stay in position during shuttling. A mirror was em-
ployed to concentrate light that is applied from below. (C) Illustration
of the MW coil used for exciting the swept frequency combs, consist-
ing of a MW stub antenna. (D) DNP setup inside the low field shield.
Laser irradiation is applied from the bottom of the NMR tube and the
tube is positioned above the microwave coil.
Corp) where hyperpolarization is excited to a 7T superconduct-
ing magnet (see Fig. S5A) in a total travel time of 648±2.6ms.
Inductive detection of the 13C NMR signal starts immediately
after the sample is in position at high field. The entire hyper-
polarization procedure is relatively easy to conduct on account
of the low laser and MW powers employed, as well as abso-
lutely no requirement for alignment of diamond samples to the
magnetic field. The field cycler consists of a high-precision ac-
tuator (Parker HMRB08) with a twin carriage mount carrying
a carbon fiber rod (8mm, Rockwest composites) into which the
NMR tube (8 mm, Wilmad) containing the sample is pressure
fit. Single crystal or powder samples are immersed in water
(Fig. S5B), and a plunger firmly holds the sample solution to
prevent changes in sample orientation and position during shut-
tling (Fig. S5C). Single crystals have the NV axes oriented at
magic angle (see Fig. 6) to the polarizing field. Fig. S5D details
the hyperpolarization setup in the low field shield. The laser
beam is collimated to a ≈4mm diameter and irradiated at the
bottom of the NMR tube carrying the sample. The microwaves
Thin Wall NMR TubeDiamond SampleMicrowave CoilTube HolderLaserMirror7T DBMWCMW LoopDiamond ParticlesMirrorPlunger1630 mm Travel LengthServo MotorLaserCarbon Fiber RodLow Field Shielding7 TeslaTo DNP CoilXY Adjustable FrameSliding Rails1 - 10 mTA20cmxyzBDiamond Particlesin Water 12
Figure S6. Schematic circuit for DNP excitation. Low field DNP from NV centers to 13C is excited by microwave sweeps produced by employing
voltage controlled oscillators (VCOs) with ramp generator inputs (see Fig. S7). Multiple VCOs are employed in a cascade to increase polarization
transfer efficiency. A spectrum analyzer is used to implement a feedback algorithm that exactly matches the VCO bandwidths to <2 MHz (see Fig.
S8). The microwaves are finally amplified by a 16W amplifier into a stub antenna that produces either longitudinal or transverse fields.
Feedback matching of VCO sweep bandwidths.
Figure S7.
Frequency-voltage characteristics of two Minicircuits ZX95-3800A+
VCO frequency sources measured with an in-situ spectrum analyzer
(Fig. S6), showing dissimilarity in frequency output with tuning volt-
age by ≈ 3MHz/V (inset A). (B) Feedback is now implemented to
match their swept bandwidths to 500± 1 MHz, shown in the insets for
2kHz sweep frequency (see Fig. S8). This allows the VCOs to be cas-
caded to simultaneously sweep over the NV center powder pattern to
enhance DNP efficiency.
During each the feedback loop, Vpp is adjusted based on the
assumption that bandwidth (cid:52)f is approximately proportional to
Vdc, and Vdc shifted to approach the target band center. To en-
sure VCO receiving reasonable input, initial values are set to be
Vpp=2V and Vdc=6V. based on empirical b and F . The pre-set
deviation we typically use is 2MHz, which is approximately the
VCO output linewidth when input is a single constant voltage.
The efficiency of the algorithm is highlighted in Fig. S8.
Figure S8. Demonstration of VCO matching process, to a sweep
bandwidth of 2.4-2.7GHz, typical for single crystal DNP experiments
in a 12mT polarizing field field. Only two iterations were taken to
match the VCO output to targeted band within deviation of 1MHz. (A)
Spectrum of VCO output (2kHz sweep rate) for the individual itera-
tions. (B) Errors to the left and right boundaries of the target sweep
band with matching iterations. (C) Band edges (left and right) converge
to the target band in 2 iterations.
C.
Intermodulation Distortion limits
The multiplicative DNP gains of our microwave frequency
comb technique will be limited by hyperfine mediated electron
broadening, and also practical constraints such as non-linear dis-
tortions set by amplifier intermodulation distortion (IMD). Here
we highlight the latter and discuss its effect on our hyperpolar-
ization technique.
IMD appears in a wide range of RF and microwave systems,
and particularly affects our experiments in the power amplifier
stage Fig. S6. When input a two-tone signal, an ideal linear
amplifier would produce an output signal of the amplified two
tones at exactly the same frequencies as the input signal. A re-
alistic amplifier, however, will produce additional signal content
at frequencies other than the two input tones. As one can tell
from Fig. S9, two fundamental tones, f1 and f2, injected into
the amplifier mix to produce interfering signals with the most
notable interference given by third order products, which are
PasternackE1V31012ZASWA-2-50DR+Pulse TriggerSpinCore Pulseblaster USBVCOVCOVCOVCO123421DCS3421DC Power SourceDC Shifted RampSplitter or Combiner MiniCircuitsANNE 50+50 Ω TerminatorAmplifierRamp GeneratorMW SwitchKey:DCPPS2116ARIGOL DG1022670 Ω 470 µFSpectrum Analyzer:SIGNAL HOUNDSA44BZN4PD1-63HP+ZAPDQ-4-SComputerSS32 Channels100 MHzMiniCircuitsZHL-16W-43SLow Field Shield (8 mT) 1.8-4.2GHz, 16WStub AntennaFeedback MatchingOutCascaded InputsCombinerDC Shifted Ramp678910VCO Input Voltage (V)2.62.72.82.93.03.13.2Output Frequency (GHz)VCO1 VCO22.63.1Output Frequency (GHz)500MHzf(cid:1)<2MHzVCO1VCO2ABSlope = 135 MHz/V∆Slope = 3 MHz/VInput Voltage (V)Frequency (GHz)VCO Output (a.u)Error (MHz)Band Edge (GHz)2.22.42.62.83-60-40-2002.22.42.62.83-60-40-200Frequency (GHz)Signal (dBm)Iteration NumbersSignal (dBm)Signal (dBm)Iteration Numbers2.22.42.62.83-60-40-2000120515250122.32.42.52.6Left boundaryRight boundaryLeft boundaryRight boundaryACB 13
increases linearly with the number of comb teeth, the ability to
combine several low-power amplifiers to obtain multiplicative
gains in DNP enhancements also has serious advantages in over-
all cost of the electronic infrastructure required.
D. MW frequency combs at high field
Let us now describe how the swept frequency combs can be
constructed in the context of high field DNP with radicals. The
availability of arbitrary waveform generators with bandwidths
that reach into the microwave range, combined with solid-state
millimeter wave mixers, now permits arbitrary pulse shaping up
to Terahertz frequencies. There are two approaches possible.
At frequencies below about 100GHz, it is possible to directly
mix a high frequency carrier with a modulated microwave sig-
nal, filter it, and feed the resulting modulated millimeter waves
into an amplifier. Solid-state amplifiers with up to 1W of power
are commercially available below 100GHz. At even higher
millimeter wave frequencies, an active multiplier chain (AMC-
Virginia Diodes) is used to generate the millimeter wave signal,
with a microwave input to the AMC in the 10-20GHz range.
These AMCs are available with up to about 100mW of power
at 270GHz, dropping to about 10mW at 500GHz and 1mW at
1THz. By modulating the input to the AMC, it is possible to
create arbitrarily modulated millimeter waves. Note that in this
case it is necessary to scale the desired modulation down by the
multiplication factor of the AMC.
The use of solid-sources at higher frequencies is typically
constrained by the available power and the significant increase
in cost with increased power output. At liquid helium temper-
atures, the electron spin relaxation times become much longer,
and it is possible to excite DNP at lower millimeter wave power.
At liquid nitrogen temperatures typically used for DNP-MAS
experiments, the electron spin relaxation times are short, and
high millimeter wave powers are needed to ensure good DNP.
Gyrotrons are used to generate sufficient millimeter wave power
for DNP in this regime. While gyrotrons have typically been
narrow band (resonant devices), voltage-tunable gyrotrons that
enable frequency modulation are being developed [27].
VII. DATA ANALYSIS
The DNP enhancement in our experiments is quantified by
scaling the hyperpolarized 13C signal to the corresponding ther-
mal signal at 7T for each sample. The spectra are all phased,
baseline corrected, and scaled to have an average noise of 1.
This allows comparison between signals taken with a different
number of averages. The areas of each peak area was calcu-
lated, and the ratio between them determines the enhancement
factor. Zero-order phase correction is applied by multiplying the
spectrum by a phase value that maximizes peak height.
A fitted absorptive Lorentzian curve identifies the peak in
Standard
each spectra with the real portion of the data.
Lorentzian formulas were used to calculate the area under the
fitted curve, and peak limits were designated such that the area
between the limits encompassed 90 percent of the entire spec-
trum area. The portion outside of the peak limits was defined as
noise. To flatten the baseline, a 12th order polynomial was fitted
through these parts, and subtracted from the spectrum. For com-
Intermodulation distortion limits to cascaded sweeps.
Figure S9.
Here we demonstrate amplifier nonlinearity effects on employing mul-
tiple cascaded sweeps on a single microwave amplifier (here Minicir-
cuits ZHL-16W-43S+). We operate here in the linear region of the am-
plifier. (A)-(C) Amplifier output spectrum with input of 2 frequency
tones f1, f2 (insets) with decreasing separation. The third order IMD
harmonics are clearly discernible, but are several orders of magnitude
lower in power than the primary tones, and do not excite hyperpolar-
ization. The nonlinearity however can cause severe distortion when the
tones are <10MHz apart, or the amplifier is operating at saturation.
(D) Amplifier output spectrum with 3 input frequency tones (insets).
Several additional IMD harmonics are visible, although still at much
reduced powers.
2f1−f2 and 2f2−f1. The power in these intermodulation prod-
ucts depends on how close the amplifier is to saturation. After
saturation, the gain becomes nonlinear and enters a compression
regime where the output power becomes independent of the in-
put power. In this compression regime, intermodulation prod-
ucts and distorted signals arise from the mixing of fundamental
signals which can adversely alter the signals of the amplified
bandwidth. Since in our DNP mechanism the transfer efficiency
falls rapidly with MW power, as long as one operates far below
the amplifier compression point, these harmonics do not play
a significant role in hyperpolarization process. However, when
the two tones approach a frequency separation <10MHz (Fig.
S9C), the nonlinearity of the amplifier substantially distorts the
output signal and deleteriously affects the DNP efficiency. In
general, cascading more frequency sweeps lead to a larger num-
ber of spurious IMD harmonics, all of which take away MW
power from the main frequency comb that drives the hyperpo-
larization process. This is evident for instance in Fig. S9D,
where we consider three frequency tones.
This technical obstacle can be overcome by employing multi-
ple cascaded amplifiers, each amplifying a component of the fre-
quency comb, which are then subsequently combined. This ex-
ploits the fact that power splitters have significantly lower non-
linearity, are not prone to IMD, and can yield a distortion-free
combination of frequency tones. While the overall MW power
2.42.62.833.2-5002.62.642.682.722.762.8-5002.662.682.72.72-5002.22.633.4-5002f1-f2=2.484GHz1f=2.691GHzf2=2.898GHz2f2-f1=3.105GHzFrequency(GHz)Frequency(GHz)Frequency(GHz)Frequency(GHz)ADCB207MHz6MHz72MHz135MHzSignal (dBm)Signal (dBm)Signal (dBm)Signal (dBm)FrequencyFrequencyFrequencyFrequency5MHzAmpInputAmpInputAmpInputAmpInput47MHz parison between DNP and thermal spectra, the average noise of
both spectra was scaled to 1 by dividing the noise section by
its standard deviation. The area of each peak was then obtained
through a Riemann sum across the peak limits, and the DNP
enhancement factor given by the equation:
ε =
SNRDNP
SNRThermal
(cid:114) NDNP
NThermal
14
(8)
|
1712.00419 | 2 | 1712 | 2017-12-04T17:35:15 | Significant improvement in estimation of the Young's modulus of the metal foam based on the Timoshenko's bend theory | [
"physics.app-ph"
] | A computational and experimental approach based on a natural vibration of a free prismatic thick beam with square cross-section is suggested. Three variants of the beam sample were used (one with skin and two without skin). From 13 to 16 lowest resonant frequencies of longitudinal, torsional and flexural vibration of each beam were analyzed. A rule for dependence of Young's modulus on the average foam density was derived from the sample without skin. It is shown that the skin presence causes known anisotropy that the stiffness in transverse direction is about 50% greater than that in the longitudinal direction. It has also been shown that the seeming frequency dependence of the Young's module can be explained by non-uniform distribution of mass density in the sample. Agreement among experimental and numerical data is excellent in most cases. The rule is also verified on solution of two lowest resonant frequencies of the free foam plates with various mass densities (from 400 to 1350 kg/m3). Agreement among experimental and numerical frequencies is acceptable. | physics.app-ph | physics | Significant improvement in estimation of the Young's modulus
of the metal foam based on the Timoshenko's bend theory
Tibor Mazúch
Nábrežná 13/23, SK - 03861 Vrútky, Slovak republic
E-mail: [email protected]
A computational and experimental approach based on a natural vibration of a free prismatic thick
beam with square cross-section is suggested. Three variants of the beam sample were used (one with
skin and two without skin). From 12 to 16 lowest resonant frequencies of longitudinal, torsional and
flexural vibration of each beam were analyzed. A rule for dependence of Young's modulus on the
average foam density was derived from the sample without skin. It is shown that the skin presence
causes known anisotropy that the stiffness in transverse direction is about 50% greater than that in the
longitudinal direction. It has also been shown that the seeming frequency dependence of the Young's
module can be explained by non-uniform distribution of mass density in the sample. Agreement
among experimental and numerical data is excellent in most cases. The rule is also verified on solution
of two lowest resonant frequencies of the free foam plates with various mass densities (from 400 to
1350 kg/m3). Agreement among experimental and numerical frequencies is acceptable.
Key words: Aluminium foam with closed cells; natural vibration of free beam; longitudinal, torsional
and flexural mode shapes; finite element method vs engineering theories; estimations of
Young's modulus; shear modulus; foam anisotropy
_____________
Note: Short version of the paper was published in the Proceedings of Interaction and Feedbacks
'2006, XIII-th National seminar with international participation. IT AS CR Prague, Czech Republic,
November 28-29, 2006, p. 53-60.
2
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
1. Introduction
One from important factors influencing the expansion of the usage of the aluminium foam
in the technical praxis is unquestionably the ability for sufficiently reliably prediction of the
behavior of foamed component parts in the operating mode. Besides undisputable interesting
properties of this perspective material, is also well known, that the modelling of the response
of the metal foam with closed cells on the mechanical forcing is very difficult. Analytical
approaches known from courses of the technical (or advanced) mechanics are usable only for
few simplest constructional shapes. From several numerical methods the finite element
method (FEM) was mostly expanded in this field, see Fig. 1 for an illustration. It is
consequence not only of the possibility of modelling complicated shapes and complicated
boundary conditions, but also the possibility of the considering various material properties in
various positions inside the foam bodies. Reliability of the modelling by the FEM strongly
depends on the input data accuracy, including also material constants. For simpler types of
forcing and shapes of component parts also approaches based on estimations of moments of
inertia are sufficient alternative with respect to finite methods [4-6].
Therefore the aim of this paper is the suggestion and the verification of a reliable approach
for the stating Young's modulus in dependence on the foam density. This basic material
characteristic is determined from natural vibration of rectangular sample of the foam.
Fig. 1 A real element of the aluminium foam and its FE idealisation
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
3
2. Suggested approach
From available samples we chose a prismatic beam with square cross-section. Besides
good manipulation with the sample during the experiment due to dimensions and the weight
(see table 1), is valuable also occurrence of longitudinal, torsional and also flexural mode
shapes (see figs. 2 – 4) in chosen frequency range. This allow us multiple verification of
results. Due to square cross-section are all flexural resonant frequencies twice multiple, so
frequency spectrum in chosen finite interval is (with respect to the rectangle cross-section)
less dense. Furthermore, possible significant split of twice multiple resonant frequencies can
indicate major defects of inner structure of the sample.
st1
st1
Fig. 2 Longitudinal mode shapes
nd2
nd2
rd3
Fig. 3 Torsional mode shapes
4
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
st1
nd2
rd3
th4
Fig. 4 Flexural mode shapes
st1
nd2
rd3
th4
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
5
Chosen prismatic beam from aluminium foam has following parameters:
Table 1 Basic parameters of three variants of the sample.
Parameter
Cross-section side length
Beam length
ap
l
Beam weight
m
Average mass density ρ
Beam with the skin
Beam without the skin
1st variant
2nd variant
3rd variant
49 mm
448 mm
0.620 kg
46.45mm
44 mm
444 mm
440 mm
0.4531 kg
0.3684 kg
576 kg/m3
472 kg/m3
432.5 kg/m3
The experimental modal analysis was based on the evaluation of the frequency response
functions measured by means of a Bruel & Kjaer 2034 FFT Analyzer. Used experimental
approaches, including a simulation of boundary conditions, the exciter and the transducer
positions, are evident from Figures 5 and 6. The need for a torsional exciter which we have
not in disposition, was avoided as follows. With asymmetric excitation and scanning (see
Fig. 6), the frequency characteristic for simultaneously excited flexural and torsional vibration
was obtained. Then, searched torsional resonant frequencies were identified by using the
comparison with characteristic for ‚pure' flexure.
Several configurations of the measurement with various positions of the exciter and the
transducer were attempted. The reproducibility of attempts was excellent. The same values of
resonant frequencies were mostly obtained. Occasionally the difference between them was
equal to the analyzer resolution limit (8Hz) and rarely twice (16 Hz).
Measurements were performed on 3 variants of the sample, see the Table 1. Firstly were
measured frequencies on the beam with the skin, then on the beam after 1st machining and
finally, on the beam after 2nd machining.
To avoid a significant distortion of the results in determining the Young's modulus, it
was necessary use experimental results of the sample without skin. The 2nd variant was
used for the reason, because for this configuration all 11 theoretical lowest resonant
frequencies (2 longitudinal, 4 torsional and 5 flexural) were excited and identified at the
chosen interval (0, 6400 Hz). It was mentioned, that all resonant frequencies of the flexural
vibrations are twice multiple with independent natural modes (in two mutually perpendicular
planes). So, finally, 16 lowest frequencies were measured for that configuration.
6
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
Fig. 5 Experimental set up to obtain longitudinal resonant frequencies
LEGEND: Up – Simulation of free boundary conditions, and exciter vs transducer
positions.
Down – The frequency response function for obtaining first two longitudinal
natural frequencies.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
7
Fig. 6 Experimental set up to obtain torsional and flexural resonant frequencies
LEGEND: Up – Simulation of free boundary conditions, and exciter vs transducer positions.
• Left – for the excitation of flexural mode shapes, only.
• Right – asymmetric (diagonal set up) for simultaneous excitation of
torsion and bending shapes.
Down – The frequency response functions (FRP)
• Full line – FRP of flexural vibration, only.
• Dashed line – FRP of simultaneous torsional and flexural vibration.
8
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
3. Young's modulus in dependence on the mass density
Based on the hypothesis about the quadratic law of the dependence of the foam Young's
modulus on the foam mass density was derived several relationships between them. Tobolka
and Kováčik derived in [2] a relation from 1st (or lowest) eigenfrequency of free square plate
TE
,1
=
38.7
10
⋅
3
2
ρ
⋅
+
86.2
10
⋅
6
ρ
+⋅
77.2
10
⋅
9
,
(1a)
which was used for the estimation of Young's modulus for described beam with the skin in
[1]. Based on its use an excellent approximation of the 1st and very good approximation of the
2nd resonant frequency of the flexural vibration with respect to the experimental results were
obtained. But this relation has also certain disadvantage. For the case of ρ = 0 (zero weight),
nonzero and furthermore relatively great Young's modulus E = 2.77 GPa is obtained. It
restricts the equation validity for the foam with lower density. It should be noted that from 2nd
eigenfrequency free square plate Tobolka and Kováčik derived another equation
TE
,2
=
70.7
10
⋅
3
2
ρ
⋅
+
00.2
10
⋅
6
+⋅
ρ
65.2
10
⋅
9
,
(1b)
which gives less estimation of the Young's modulus for the same mass density ρ …
Fleck, Ashby and co-workers suggested in [3] another equation
EE
/
0
⋅=
30
Φ
2
(
ρρ
0
/
)
2
−+
1(
Φ
)(
/
ρρ
0
);
Φ
=
7.0
,
where E0 and ρ0 are constants of the pure aluminium. Hence
0
⋅=EE
/
0
147
(
ρρ
0
/
)
2
⋅+
(30
/
ρρ
0
)
.
(2)
(3)
For the case of pure aluminium (not foam), i.e.
0ρρ
/
1=
we obtain less than half of the value
of the Young's modulus for aluminium
the foam with higher density.
0
⋅=EE
/
0
447
. It restricts the equation validity for
It is evident, that for satisfaction of the condition
=ρE
(
)0
=
0
the quadratic equation can
sustain from quadratic and linear member, only.
E
(
ρρρ
=)
+
b
a
2
(4)
In that case for the unequivocal derivation of the function
= ρ(EE
)
will suffice reliably
stated Young's modulus only for one value of the foam density
E
1
= ρE
1
(1
)
. Needed
coefficients a and b by solution of an equation
ρρ
0
ρρ
1
2
0
2
1
a
b
=
E
0
E
1
,
(5)
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
9
can be obtained, where E0 and ρ0 are values of the pure aluminium, i.e E0=69 GPa and
ρ0=2700 kg/m3. Then
a
b
=
ρρ
0
ρρ
1
2
0
2
1
1
−
E
0
E
1
.
(6)
From the basic course of the strength theory for the case of linear, isotropic and elastic
material is known a relation
or
G
=
E
ν1(2
)+
E
=
ν1(2G
)+
,
where G is the shear modulus and ν is the Poisson's ratio. Then holds
ν
=
E
2
G
−
1
.
(7)
(8)
(9)
Table 2 Computed material characteristics from measured resonant frequencies sample
without the skin with mass density ρ = 472 kg/m3
Mode shape
description
1st longitudinal
2nd longitudinal
1st torsional
2nd torsional
3rd torsional
4th torsional
1st flexural
1st flexural
2nd flexural
2nd flexural
3rd flexural
3rd flexural
4th flexural
4th flexural
5th flexural
5th flexural
Experiment
Shear modulus
Young's modulus
f [Hz]
G [GPa]
E [GPa]
2976
5376
1716
3036
4680
6184
637
662
1549
1549
2720
2720
4130
4130
5552
5552
1.276
1.017
1.074
1.055
3.237
2.641
3.368
2.641
2.792
2.792
3.521
3.803
3.105
3.105
2.904
2.904
2.792
2.792
2.544
2.544
The shear modules from resonant frequencies of torsional mode shapes were calculated by
the relation (A-4) (see Appendix). Resonant frequencies of longitudinal mode shapes by
analytical relation (A-1) do not depend on the Poisson's ratio ν. Based on 3D FE modeling
this experience was verified and negligible influence was finded in the study [1]. Carefully we
10
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
investigated the influence of Poisson's number on the accuracy of the Young's modulus
calculated from the flexural resonance frequency, see (A-7) in the Appendix. We have
calculated the modules from the 5 lowest frequencies for Poisson numbers from the
interval
∈ν
0.29
4
;0.3
. The largest relative difference occurred at 5th frequency and was less
than 1.23%. The largest relative difference at the 4th frequency was less than 0.93 %, etc. So
we can say that the influence of Poisson's number on the precision of determination of the
Young's module is negligible.
Ashby et al., in their monographs [9], indicate the range of Poisson's number from 0.31 to
0.34 for all commercial aluminium foams. Therefore, we have decided to use the value for
pure aluminum ν=0.32, which is almost at the center of the mentioned interval.
]
a
P
G
[
s
e
l
u
d
o
M
4
3
2
1
0
E = 8.3155 f - 0,129
R² = 0,7655
G = 3.6852f - 0,147
R² = 0,5657
0
2000
4000
6000
8000
Frequency f [Hz]
4
3,5
]
a
P
G
[
E
3
2,5
2
E = 8.501 f -0,132
R² = 0,9032
0
2000
4000
6000
8000
Frequency f [Hz]
Fig. 7 The apparent frequency dependence of the sample modules
The Table 2 shows an acceptable agreement among Young's modules calculated from
resonant frequencies corresponding with simplest mode shapes of longitudinal, torsional and
flexural vibration (with except of the 2nd copy of the 1st flexural mode shape). However they
are higher than modules calculated from frequencies, which correspond with higher mode
shapes. It looks like the foam Young's modulus, is frequency-dependent. It decreases, when
frequency of oscillation increases, so it is not constant. Reality is different. This phenomenon
could be explained by non-uniform distribution of the sample density. Consider, for example,
the largest deformations in 1st bending mode shapes, see Fig. 4. If the density of the sample is
greater in the central region then logically there is also a greater bending stiffness (resistance
against bending deformations). Largest deformations in next bending modes are in places
with less density. We will discuss the problem in more detail when verifying the proposed
procedure on other experimental data.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
11
ρ
1.15 ρ
0.8 ρ
0.8 ρ
1.15 ρ
Fig. 8 Roughly estimated density distribution in the sample for anisotropic model
Furthermore, it is known fact that boundary conditions (in the experiment they are not
totally equal with conditions considered in calculations) most considerable affect those
resonant frequencies, which correspond with the simplest mode shapes (1st in this case).
Therefore it is probable that more realistic will be coefficients obtained from frequencies
which correspond with all measured resonant frequencies.
Three values of the foam Young's modulus E1 for obtaining coefficients of the equation
(4) were considered:
a) E1,a = 3.51 GPa – average value of Young's modules calculated from resonant
frequencies corresponding with the simplest (1st) mode shapes (highlighted in the
Table 2). It should be noted that most of approaches for obtaining the function E=E(ρ)
is based on measuring the lowest frequencies of the samples. Then
aE
,1
(
ρ
=)
10
6
.0(
00813
2
ρ
+
60.3
)
ρ
.
(10)
b) E1,b = 3.028 GPa – average value of Young's modules calculated from all measured
resonant frequencies
(
ρ
=)
6
.0(10
00859
2
ρ
+
36.2
)
ρ
bE
,1
(11a)
or in dimesionless form
2
+
(
ρρ
0
E b
,1
.0
908
=
/
E
0
/
)
.0
092
/
(
)
ρρ
,
0
(11b)
where E0 and ρ0 are constants of the pure aluminium.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
c) Only for an illustration we introduce next equation obtained by using measured
resonant frequencies corresponding with the 1st (or the simplest) mode shapes of the
flexural vibration of the sample with the skin (1st configuration from the table 1). Then
for the average value E1,c = 6.052 GPa
(
ρ
=)
cE
,1
6
.0(10
00645
2
ρ
+
14.8
.
)
ρ
(12)
12
Graphical interpretations of relations (1a), (10-12) are in Fig. 9.
Fig. 9 Dependence of the Young's modulus on the foam mass density
LEGEND:
Full line – recommended values, by equation (11),
dashed line – values by equation (10),
dotted line – values by equation (12) and
dot-dashed line – values by equation (1a).
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
13
4. Verification of the approach on experimental results of beams without skin
Derived relationships E=E(ρ) can be verified by using the comparison calculated and
experimental resonant frequencies. Consider now the sample with the mass density ρ = 432.5
kg/m3 (see the Table 1). Computed mode shapes are in Figures 2 ÷ 4. Results are in the Table
3. It should be noted that all measured frequencies of bending oscillation are split here,
but the maximum difference between two split frequencies is about 5%. All here used FEM
models consisted from quadratic 3D finite elements of serendipity type.
Fig. 10 Detail from the figure 9 for quick estimation of Young's modulus
LEGEND:
of the beam without the skin
Full line – recommended values, by equation (11),
dashed line – values by equation (10),
dotted line – values by equation (12) and
dot-dashed line – values by equation (1a).
14
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
As expected, only 1st flexural, longitudinal and torsional FE frequencies good approximate
experimental results for E=3.14 GPa, which was estimated by using eqn. (10). The other
approximations (except of 2nd copy of 2nd flexural frequency) are poor. Better agreement
among FE and experimental results was reached for the recommended value E=2.63 GPa
given by eqn. (11), see also the solid line in Figs 9 and 10.
Table 3 Measured and computed resonant frequencies of the free beam without skin
with ρ = 432.5 kg/m3
Mode
number
Mode shape
description
1 ÷ 6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
as rigid body
1st flexural
1st flexural
2nd flexural
2nd flexural
1st torsional
3rd flexural
3rd flexural
1st longitudinal
2nd torsional
4th flexural
4th flexural
3rd torsional
2nd longitudinal
4th torsional
Average error
FEM Anisotropic
model with ρ
distribution by Fig. 8
εm33
[%]
fm33[Hz]
0
608
608
1469
1469
1746
2621
2621
3085
3104
3980
3980
4646
5471
6036
2.7
2.6
2.6
1.8
4.4
3.7
2.2
4.5
0.3
0.7
1.7
3.3
0.3
2.4
Experiment
Giba
FEM
FEM
Isotropic model with
Isotropic model with
E=3.14 GPa
E=2.628 GPa
fg3 [Hz]
fm31[Hz]
-----
592
624
1432
1496
1672
2528
2680
2952
-----
3968
4008
4568
5296
6056
0
608
608
1584
1584
1732
2893
2893
3059
3463
4420
4420
5195
6103
6926
εm31
[%]
2.7
2.5
10.6
5.9
3.6
14.4
7.9
3.6
11.4
10.3
13.7
15.2
14.4
fm32[Hz]
0
557
557
1451
1451
1586
2650
2650
2803
3173
4049
4049
4759
5590
6344
9.0
εm32
[%]
5.9
10.7
1.3
3.0
5.1
4.8
1.1
5.1
2.1
1.0
4.2
5.6
4.8
4.2
Used anisotropic model was composed from five isotropic regions depicted in Fig. 8. For the
density values from the figure, Young's modules were calculated according to Equation (11),
see also solid line in Fig. 10. The average error value 2.4 % indicates that this model provided
the best results. A similar situation also occurred with a higher density sample, see Table 4.
Comparison of experimental resonance frequencies of flexural oscillation with the
calculated frequencies from the anisotropic FEM model is shown also in Fig. 11. The
calculated frequencies for the less dense sample are found between the split experimental
frequencies. Their trend lines are almost covered. The frequencies calculated for the denser
sample are in very good agreements with experimental ones. These agreements indicate that
the chosen anisotropic model very well reflects the reality. In other words, this confirms our
hypothesis that there is no frequency dependence of the foam modules, but there is an uneven
distribution of foam density in the sample.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
15
Table 4 Measured and computed resonant frequencies of the free beam without skin
with ρ = 472kg/m3
FEM
FEM
FEM
Mode
number
Mode shape
description
Experiment
Isotropic model
Isotropic model
Anisotropic model
Giba
with
with
with ρ distribution by
E=3.513 GPa
E=3.028 GPa
Fig. 8
1 ÷ 6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
fg2 [Hz]
fm21 [Hz]
fm22 [Hz]
εm21
[%]
as rigid body
1st flexural
1st flexural
2nd flexural
2nd flexural
1st torsional
3rd flexural
3rd flexural
1st longitudinal
2nd torsional
4th flexural
4th flexural
3rd torsional
2nd longitudinal
5th flexural
5th flexural
4th torsional
Average error
-----
637
662
1549
1549
1716
2720
2720
2976
3036
4130
4130
4680
5376
5552
5552
6184
0
637
637
1650
1650
1738
3000
3000
3070
3475
4563
4563
5213
6122
6259
6259
6949
0.1
3.8
6.5
6.5
1.3
10.3
10.3
3.1
14.5
10.5
10.5
11.4
13.9
12.7
12.7
12.4
0
590
590
1530
1530
1611
2782
2782
2847
3223
4232
4232
4834
5677
5804
5804
6445
8.8
εm22
[%]
7.3
10.8
1.2
1.2
6.1
2.3
2.3
4.3
6.2
2.5
2.5
3.3
5.6
4.5
4.5
4.2
3.8
fm23 [Hz]
εm32
[%]
0
635
635
1552
1552
1719
2761
2761
3055
3138
4185
4185
4703
5495
5723
5723
6142
0.3
4.0
0.2
0.2
0.1
1.5
1.5
2.7
3.3
1.3
1.3
0.5
2.2
3.1
3.1
0.7
1.6
Fig. 11 Anisotropic FEM vs experimental flexural frequencies of beams without skin
LEGEND: Left – sample with ρ = 432.5 kg/m3; Right – sample with ρ = 472kg/m3;
■ – experimental frequencies; - numerical frequencies
full trend line – experimental frequencies; dashed trend line – numerical frequencies
16
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
5. Verification of the approach on experimental results of the beam with skin
It can be assumed, that the skin of the foam sample is also the foam, but with much higher
density. Therefore, for next verification of derived relations E=E(ρ), the results from the
experiment of the beam with the skin need be used. Results are in the Table 5. All here used
FEM models consisted from quadratic 3D finite elements of serendipity type.
In the classical approach (without separate model of the skin), for the average mass
density ρ = 576 kg/m3 the Young's modulus Ec = 6.83 GPa by the eqn. (12) was estimated.
Fig. 12 Detail from the figure 9 for quick estimation of Young's modulus of the foam skin
LEGEND:
Full line – recommended values, by equation (11),
dashed line – values by equation (10),
dotted line – values by equation (12) and
dot-dashed line – values by equation (1a).
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
17
We noted that Tobolka - Kováčik's relationship (1a) gave almost the same value – E1T =
6.85 GPa (see also the Fig. 10). This totally isotropic model gives good aproximations for
both copies of the 1st flexural frequency and the 2nd copy of the 2nd flexural frequency, only.
The other approximations are poor.
Two variants of anistropic FEM model were considered. The mass density of the skin on
opposite square faces was ρs1 = 1692 kg/m3 with Es1 = 28.6 GPa. The average mass density of
the skin on all other parts of the beam was ρs2 = 1424 kg/m3 with Es2 = 20.8 GPa. The inner
foam (or the core) of considered sample had average mass density ρf = 472 kg/m3 and
Young's modulus Ef = 3.03 GPa. It is evident from the table, that this model yields to results
in much more greater accuracy than the classical approach.
The 2nd anisotropic model is more complicated. It consists from the core with ρ
distribution by the Fig. 8 and two isotropic skins. By other words all anisotropic models used
in these analyzes were ‚per partes' isotropic. The most accurate results were obtained for the
2nd anisotropic FEM model, see two last columns in Table 5.
Table 5 Measured and computed resonant frequencies of the free beam with the skin
Mode
number
Mode shape
description
Experiment
Giba
FEM
Isotropic model with
E=6.83 GPa
FEM
Anisotropic model
composed from
isotropic core and
isotropic skin
fg1 [Hz]
fm11 [Hz]
εm11
[%]
fm12 [Hz]
εm12
[%]
as rigid body
1st flexural
1st flexural
2nd flexural
2nd flexural
1st torsional
1st longitudinal
3rd flexural
3rd flexural
2nd torsional
4th flexural
4th flexural
3rd torsional
2nd longitudinal
Average error
1 ÷ 6
7
8
9
10
11
12
13
14
15
16
17
18
19
-----
784
856
1992
2072
-----
3295
3552
3552
3768
5176
5176
5640
6176
0
830
830
2141
2141
2174
3839
3873
3873
4347
5864
5864
6520
7655
5.8
3.1
7.5
3.3
16.5
9.0
9.0
15.4
13.3
13.3
15.6
23.9
12.3
0
799
799
2023
2023
1953
3237
3589
3589
3905
5335
5335
5853
6431
1.9
6.7
1.5
2.4
1.8
1.0
1.0
3.6
3.1
3.1
3.8
4.1
3.1
FEM
Anisotropic model
with considering
ρ distribution by Fig. 8
in core and isotropic
skin
fm13 [Hz]
0
807
807
1995
1995
2032
3406
3521
3521
3787
5239
5239
5741
6152
εm13
[%]
2.9
5.7
0.1
3.7
3.4
0.9
0.9
0.5
1.2
1.2
1.8
0.4
1.9
18
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
One interesting experience is founded from three presented sets of experimental results
(see Tables 5, 4 and 3). It is evident from the Table 6, that for the beam with the skin was
calculated from 2nd copy of split 1st flexural resonant frequency the Young's modulus
E=7.430 GPa and from the 1st longitudinal resonant frequency E = 5.021 GPa. The relative
difference between them was about 48 %. Fleck, Ashby with co-workers in [3] described
similar experience that strength and stiffness of some foams in the transverse direction is
about 50% greater than longitudinal and through thickness directions. They justify it by
"shoebox" shaped cells of the foam. After removing of the skin of the sample (for the foam
mass density ρ = 472.0 kg/m3) this difference decreased approximately to 15 %. Then after
removing next foam layer from all sides of the sample (see last three columns in the Table 5)
the difference decreased to 11 %.
Table 6 Foam anisotropy properties of three variants of considered sample
Beam with the skin
Beam without the skin
Mode
ρ = 576.0 kg/m3
ρ = 472.0 kg/m3
ρ = 432.5 kg/m3
f
E
Difference
f
E
Difference
f
E
Difference
[Hz]
[GPa]
[%]
[Hz]
[GPa]
1st flexural
856
7.43
1st flexural
784
6.23
1st
longitudinal
3295
5.02
48
24
-
662
3.80
637
3.52
2976
3.30
[%]
15
7
-
[Hz]
[GPa]
624
3.31
529
2.98
2952
2.96
[%]
11
1
-
It follows that in the case of here-considered sample this strong anisotropy effect was
caused above all by the existence of the skin on the sample. Another significant factor was the
non-uniform distribution of density in tested sample.
Conformity of numerical calculations with experimental data (see Tables 3, 4 and 5) is
limited by the fact that bending frequencies have been split. This could be improved by
considering of so-called bimodular behaviour in transverse directions. It is possible to
simulate this by using a different distribution of density in transversal directions. However,
this improvement is not the subject of this publication.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
19
6. Verification of the approach on the Tobolka – Kováčik's plates
Here proposed rule to determine the Young's modulus from average foam density (11) we
have also tested on the task with a larger range of foam density values.
Fig. 13 Tobolka – Kováčik's plate
LEGEND: Left – cross section of the plate
Right – Roughly estimated density distribution in the sample for anisotropic FE model.
Highlighted parts have considered density 1.15 ρ and the other parts have density 0.85ρ.
Considered plate [2] has dimensions 137 x 137 x 8 mm and for Young's modulus
determination was there used the Leissa's relationship [10]. Note that due to the apparent
frequency dependence of the module, there were determined 2 different equations to
determine the modulus of elasticity in [2], see (1a) and (1b).
f
n
=
B
n
2
ρπ
a
2
−
Eh
(
4
1
with
,
B
1
=
08.4
and
B
2
=
.91.5
(13)
2
ν
)
By using FEM model (consisted from 3D quadratic elements), we have tested the accuracy
of the relationship for the plate from pure iron and aluminium. Error on the 1st frequency was
7.6% and on the 2nd frequency about 6% in both cases. Tobolka – Kováčik's experimental
data (not affected by this inaccuracy) can simply be reconstructed by substituting the relation
(1a) for the 1st frequency and the relation (1b) for 2nd frequency into (13).
According to the sample from Fig. 13 we have estimated the surface skin thickness to
about 0.3 mm on both sides of the plate. The density of the skin (ρ =1424 kg/m3) obtained
from previously tested beam was used also in these FEM calculations (for all considered
average densities of the plate). Computed mode shapes are on Fig. 14 and the comparison of
numerical and experimental resonant frequencies shows Fig. 15.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
st1
nd2
20
Fig. 14 Natural modes corresponding with two lowest resonant frequencies of the plate
Fig. 15 Comparison of numerical and experimental results on 2 lowest frequencies
in dependence on average plate density
LEGEND:
Full lines – Tobolka – Kováčik'experimental data
Discrete points – FEM results with considering only one equation
(11) for both lowest resonant frequencies.
FEM simulations were performed for a plate mass density in the range of 400 to 1350
kg/m3. Largest relative differences between numerical and experimental data less than 3 % on
1st frequency and less than 4.8 % on 2nd frequency were occurred. We noted that very good
agreement among these FE results and the experimental data should be treated with some
reserve, because the parameters of the skin were only estimated.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
21
7. Concluding remarks
Based on the original numerical-experimental way, we obtained a rule for the Young's
module's dependence on the density of the aluminum foam, see equations (11a, 11b). The rule
was derived from natural vibration of thick prismatic beam with constant square cross-
section. Longitudinal, torsional and flexural vibrations were taken into account. The
properties of the derived rule are predetermined by the extremely accurate Timoshenko's bend
theory and hypothesis about the quadratic law of the module dependence on the foam density.
Validity of here-derived relationship was verified by comparisons of FEM calculations
with experimental results on the natural vibration of thick beam. Very good agreements were
obtained by using the isotropic FEM models of beams without skin. Anisotropic models
(composed of several isotropic regions) gave an excellent agreement with experimental data
in most cases. The anisotropic model gave excellent approximations also for the beam with
the skin. Finally, we verified the validity of the rule on natural vibration of the free thinner
foamed plate with the skin on both sides. The agreement among experimental and numerical
data (2 lowest resonant frequencies) of square foamed plates with varying mass densities
(from 400 to 1350 kg/m3) was very good, too. It has been shown that even a rough estimate of
density distribution can significantly increase the accuracy of frequency approximations.
It was demonstrated, that strong anisotropic foam properties were caused mainly by the
existence of the skin and non-uniform distribution of the mass density in the foam sample.
The derived relationship (11) could be useful in improving the accuracy of numerical or
analytical modeling of mechanical behavior of foam (in various sandwich or functionally
graded material approaches, e.g.). The relationship and conclusions are valid for aluminum
foams of "closed cells". It would be useful if someone verified the validity of the
dimensionless equation (11b) also for other (not aluminum) metal foams.
Acknowledgement and dedication
The author would like to dedicate this article to the memory of Dr. V. Giba. His perfect
realization of here proposed experiment is gratefully acknowledged.
22
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
REFERENCES
[1] MAZÚCH, T.: On finite element modelling of aluminium foam. Pilot study. Report
IMMM SAS, SHW-FEM 3402/03, 2003.
[2] TOBOLKA, P., KOVÁČIK, J.: Young's modulus and loss factor of aluminium foams
from free vibrations of a square plate (in Slovak). Metallic materials 34, 1996, p. 20-24.
[3] FLECK, N. A., ASHBY, M. F., LU, LIU, J. S., GUY, G.: Mechanical properties of
metallic foams. M.U.R.I at C.U.E.D. Presentation.
[4] SIMANČÍK, F., KOVÁČIK, J., MINÁRIKOVÁ, N.: Bending properties of foamed
aluminum panels and sandwiches. In MRS Online Proceedings Library Archive, Vol.
521, 1998, p. 91-96.
[5] TIMOSHENKO, S. P.: On the correction for shear of the differential equation for
transverse vibrations of bars of uniform cross-section. Philosophical Magazine 41,
1921, 744-746.
[6] TIMOSHENKO, S. P.: On the transverse vibrations of bars of uniform cross-section.
Philosophical Magazine 43, 1922, 125-131.
[7] HAN, S. M., BENAROYA, H., WEI, T.: Dynamics of transversely vibrating beams
using four engineering theories. Journal of Sound and Vibration 225, 1999, p. 935-988.
[8] BLEVINS, R. D.: Formulas for natural frequency and mode shape. Van Nostrand
Reinhold Co., New York, 1979.
[9] ASHBY, M. F., EVANS, T., FLECK, N. A., HUTCHINSON, J. W., WADLEY, H. N.
G., & GIBSON, L. J.: Metal foams: a design guide. Elsevier, 2000.
[10] LEISSA, A. W.: Free vibration of rectangular plates. J. Sound Vibr. 31, 1973, p. 257-
293.
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
23
Appendix
A-1 Nomenclature
Young's modulus
shear modulus
mass density
Poisson's number
area of the cross-section
area moment of inertia of the cross-section about the neutral axis
polar moment of inertia of the cross-section about the torsional axis
beam length
shape factor of the cross-section
torsional constant of the cross-section
n-th resonant frequency (for all three types of natural vibration)
n-th dimensionless circular frequency (for Timoshenko's theory of natural vibration,
only)
coversheet of the foam
E
G
ρ
ν
A
J
Jp
l
k′
c
fn
nω
skin
A-2 Used analytical relations for natural vibration of prismatic beam
Here are used relations for resonant frequencies of isotropic free prismatic beam and
following relations for Young' modulus.
Longitudinal vibration [8]
By elementary theory for resonant frequencies holds
n
E
2
ρπ
l
f n
=
,
n
=
1,2,...
(A-1)
(A-2)
and hence
E
=
2
n ρ
f
.
2
24
l
n
Torsional vibration [8]
For resonant frequencies of the prismatic beam with constant square cross-section holds
=
f
n
n
2
l
cG
ρ
J
p
where
,
c
=
1406
.0
,
JA
p
=
A
6
,
n
=
,...2,1
(A-3)
and hence
G
=
(
2
f
22
)
n ρ
J
l
2
cn
p
.
(A-4)
24
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
Flexural vibration – Euler-Bernoulli's theory [8]
and hence
f
n
=
2
(
)
β
l
EJ
n
2
ρπ
A
,
n
=
1,2,...
For free beam on both ends holds
=
E
(
2
π
f
J
n
42
)
ρ
Al
)4
(
β
l
n
.
(A-5)
(
β
l
)
1
=
.4
730
(
,
β
l
)
2
=
.7
,853
(
β
l
)
2
=
.10
996
(
a
β
l
)
n
=
(
2
n
+
π
)
1
2
;
n
>
.3
Flexural vibration – Timoshenko's theory [5-7]
By Timoshenko's theory for free beam with given geometrical a material properties holds
f
n
=
,
n
=
,...2,1
(A-6)
2
1
−
b
n
)
2
2
γρπ
(
2
aE
n
(
1
+
l
)
where
with
an
=
B
1
+
B
2
+
(
B
1
−
B
2
2
)
+
B
3
,
bn
−=
B
1
−
B
2
+
(
B
1
−
B
2
2
)
+
B
3
,
B
1
=
2
ωρ
n
J
2
,
B
2
=
γγ
B
1
,
=
)
ν
,
(
12
k
+
′
B
3
=
For a rectangle (and also for a square) holds
J
=
2
A
12
,
k
=′
+
(
)
110
ν
ν
11+12
Α
2
ωρ
n
2
L
.
and 5 lowest dimensionless circular frequencies for the beam with the skin and also for two
configurations without the skin are in the next table.
Table A-1 Lowest dimensionless circular frequencies of three variants of the foam sample for
Poisson's number ν = 0.32
Configurations
of the beam
by Table 1
1st variant
2nd variant
3rd variant
1ω
2ω
3ω
4ω
5ω
8.17972
21.0813
38.0873
57.6022
78.611
9.47838
24.5463
44.5671
67.7106
92.7728
10.3911
27.0339
49.3209
75.2725
103.5430
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
25
These values were solved from a transcendental equation
(
a
2
n
)(
ab
2
n
2
n
−
+
2
b
n
2
2
γ
+
ba
nn
(
2
+
bba
n
nn
)(
−
aba
)(
2
2
γ
a
n
nn
2
a
n
2
n
+
2
+
−
b
n
)
22
γ
b
n
2
γ
ba
nn
+
ba
nn
)
sin
a
n
sinh
b
n
−
cos
a
n
cosh
b
n
−=
1
.
Young's modulus can be obtained by
E
=
(
2
π
f
(
n
a
)
l
2
n
2
)
γρ
(
1
+
)2
n
−
b
.
(A-7)
It should be noted, that described relations hold for considered configurations of the beam.
For the other parameters these can be more complicated.
640 elements
80 elements
Fig. A-1 Used FE meshes for the test of covergence
Beam dimensions: 0.049m × 0.049m × 0.448m.
Material constants are taken from [1]:
E = 6.85 GPa
Young's modulus
ρ = 576 kg/m3
Mass density
Poisson's number
ν = 0.3
In the table A-2 are results from three FE models (composed from 80 to 640 3D quadratic
20-noded solids of serendipity type) and also from analytical theories. It is evident that the FE
results are only little varied with respect to the size of the FE model. It follows that, for
sufficiently accurate modelling of the problem, even the simplest model is enough.
26
T. Mazúch: Young's modulus of metal foam. arXiv, December 4, 2017
The agreement of the analytical approach [8] with our FE results for torsional resonant
frequencies is excellent. Euler-Bernoulli's (elementary) theory gives acceptable agreement
only for the lowest flexural resonant frequency. For higher frequencies it is falling down. On
the opposite side the Timoshenko's theory [5-7] (results are in the column for engineering
theories) gives an excellent agreement (relative difference is less than 1%) also for the 5th
resonant
frequency. Agreement
among
analytical
estimations
for
longitudinal
eigenfrequencies with our FE solutions is also excellent (relative difference is less than 0.3%).
A-3 FEM convergence
Table A-2 Convergence of FE model and the comparison with analytical solutions
Mode
description
6x rigid body mode
1st flexural
1st flexural
2nd flexural
2nd flexural
1st torsional
1st longitudinal
3rd flexural
3rd flexural
2nd torsional
4th flexural
4th flexural
3rd torsional
2nd longitudinal
5th flexural
5th flexural
4th torsional
FEM
FEM
FEM
Engineering
Euler-Bernoulli
80 elem.
270 elem.
640 elem.
teories
bend theory
fM1[Hz]
fM1[Hz]
fM2[Hz]
fet[Hz]
feb[Hz]
0
833
833
2151
2151
2212
3852
3895
3895
4425
5906
5906
6640
7684
8082
8082
8857
0
832
832
2149
2149
2199
3852
3889
3889
4389
5893
5893
6597
7681
8057
8057
8796
0
832
832
2148
2148
2197
3851
3889
3889
4393
5891
5891
6589
7681
8053
8053
8784
831
831
2142
2142
2195
3854
3872
3872
4390
5860
5860
6584
7709
8002
8002
8780
867
867
2389
2389
4683
4684
7742
7742
11565
11565
It can be said, that analytical theories described in Appendix A-2 (with exception of the
Euler-Bernoulli theory) are sufficiently exact and therefore they are sufficient for
estimates of material constants from resonant frequencies of the beam considered.
|
1812.04921 | 1 | 1812 | 2018-12-12T12:49:46 | Brewster effect when approaching exceptional points of degeneracy: Epsilon-near-zero behavior | [
"physics.app-ph"
] | We reveal that the phenomenon of full transmission without phase accumulation commonly associated with epsilon-near-zero (ENZ) materials for a plane-wave does not require vanishing of permittivity. We theoretically connect the phenomenon with condition of the Brewster effect satisfied at the edges of stop bands (so called exceptional points of degeneracy) and show that the full transmission without phase accumulation can be observed in various one-dimensional periodic structures. Particularly, exploiting the manifold of exceptional points of degeneracy in one-dimensional all-dielectric periodic lattices, we demonstrate that these structures not only offer a lossless and extremely simple, CMOS compatible alternative for some applications of ENZ media, but exhibit new properties of all-angle full transmission with zero phase delay. | physics.app-ph | physics |
Brewster effect when approaching exceptional points of degeneracy:
Epsilon-near-zero behavior
Vladislav Popov∗
SONDRA, CentraleSup´elec, Universit´e Paris-Saclay, F-91190, Gif-sur-Yvette, France
Aalto University, P.O. 15500, FI-00076 Aalto, Finland
Sergei Tretyakov†
Andrey Novitsky‡
DTU Fotonik, Technical University of Denmark,
Ørsteds Plads 343, DK-2800 Kongens Lyngby, Denmark and
Department of Theoretical Physics and Astrophysics,
Belarusian State University, Nezavisimosti av. 4, 220030 Minsk, Belarus
We reveal that the phenomenon of full transmission without phase accumulation commonly as-
sociated with epsilon-near-zero (ENZ) materials for a plane-wave does not require vanishing of
permittivity. We theoretically connect the phenomenon with condition of the Brewster effect satis-
fied at the edges of stop bands (so called exceptional points of degeneracy) and show that the full
transmission without phase accumulation can be observed in various one-dimensional periodic struc-
tures. Particularly, exploiting the manifold of exceptional points of degeneracy in one-dimensional
all-dielectric periodic lattices, we demonstrate that these structures not only offer a lossless and
extremely simple, CMOS compatible alternative for some applications of ENZ media, but exhibit
new properties of all-angle full transmission with zero phase delay.
Metamaterials promise amazing possibilities in manip-
ulation of electromagnetic fields, which are not available
with natural materials (see, e.g., [1 -- 5]). Extreme prop-
erties and unique effects leading to novel functionalities
require artificial materials with extreme and singular val-
ues of material parameters. Recently, much attention
has been attracted by exotic properties of materials with
permittivity ε near zero. Both natural substances and
metamaterials possessing ε ≈ 0 are called epsilon-near-
zero (ENZ) media. From the physical point of view, this
special value of permittivity corresponds to the topolog-
ical transition between metals and dielectrics, as can be
illustrated by tuning parameters of a hyperbolic meta-
material, whose wave dispersion changes from hyperbolic
to elliptic type [6, 7] at the ENZ point. From the ap-
plications point of view, properties of ENZ media can
be exploited for energy tunneling through subwavelength
channels [8, 9], improvement of antenna directivity [10],
phase matching due to zero phase advance in ENZ me-
dia [11], enhancement of nonlinear effects [12, 13], etc.
[14 -- 16].
Conditions for ε ≈ 0 have been the subject of scrutiny
for more than a decade [15, 17]. Unfortunately, natural
ENZ materials are rather lossy, and this may suppress or
even ruin their useful properties [18]. To circumvent this
difficulty one may exploit the accidental degeneracy near
the Γ-point in all-dielectric photonic crystals, where the
zero-refractive-index properties are available [19, 20]. On
∗ [email protected]
† [email protected]
‡ [email protected]
FIG. 1. (a) Reflection and transmission through an isotropic
slab and (b) 1D photonic crystal composed of alternating lay-
ers of isotropic dielectrics. (c) Three-layer (having an inver-
sion center) unit cells of a 1D photonic crystal.
the other hand, superlattices constructed of positive and
negative index photonic crystals can also be used [21, 22].
However, it is important to keep in mind that all artificial
ENZ materials are complex 2D or 3D lattices that can
be characterized by effective permittivity only under the
homogenization condition (period is much smaller than
the operating wavelength).
In this letter, we show that a physical mechanism be-
hind properties of ENZ media does not require vanishing
of permittivity. We theoretically connect the effect of
complete wave tunneling without phase accumulation to
the condition of the Brewster effect satisfied when ap-
proaching exceptional points of degeneracy. This the-
ory opens up a possibility to exploit the manifold of ex-
ceptional points of degeneracy in all-dielectric periodic
lattices in order to emulate properties of ENZ media.
Particularly, we demonstrate that one-dimensional (1D)
periodic structures can be designed in order to exhibit
ENZ behavior. We validate the theory by demonstrat-
ing the complete wave tunneling without phase advance
through simple layered structures composed of conven-
tional dielectrics. We also show that there is no phase
accumulation throughout the whole thickness of a struc-
ture.
To explain the analogy between existence of excep-
tional points of degeneracy, the Brewster effect, and the
properties of ENZ media, let us first consider the well-
known analytical solution for the plane wave transmission
through an isotropic slab of thickness ds, permittivity
εs, and permeability µs, as illustrated in Fig. 1(a). The
transmission coefficient for an obliquely incident plane-
wave of arbitrary polarization can be written as [10]
T =(cid:20)cos(kzsds) − i
0 − k2
Z 2
s + Z 2
a
2ZsZa
sin(kzsds)(cid:21)−1
.
(1)
Here, kzs = pεsµsk2
t is the normal component of
the wave vector inside the slab, k0 = ω/c is the vacuum
wavenumber (ω is the frequency, c is the speed of light in
vacuum), kt represents the tangential component of the
wave vector, Zs and Za are the wave impedances inside
and outside the slab, respectively. The impedances are
defined as the ratios of the tangential to the slab plane-
wave field components. Assuming absence of losses, the
reflectivity 1 − T2 vanishes provided Zs = Za. Except
for the trivial case of the same materials of the slab and
the ambient medium, the equal wave impedances are re-
alized under conditions of Brewster's law. In this case the
transmission coefficient T = eikzsds describes the phase
shift of the fully transmitted wave. When the full trans-
mission is complemented by zero phase shift kzsds = 0,
a unique phenomenon of complete wave tunneling with-
out phase accumulation is realized. Originally, it was as-
sociated with the ENZ/mu-near-zero(MNZ)/epsilon-mu-
near-zero(EMNZ) materials.
Noteworthy, a Fabry-Perot resonance occurring in any
dielectric slab under the conditions kzsds = 2πm (m =
1, 2, . . .) is also characterized by the full transmission
and zero phase shift. However, the Brewster effect is
fundamentally different from the Fabry-Perot resonance.
It exists at any thickness of the slab ds. Moreover, a
plane wave does not accumulate phase when it prop-
agates across the slab under conditions Zs = Za and
kzs = 0. On the output the phase accumulation is 0, but
not a multiple of 2π.
Condition kzs = 0 defines the critical angle θc of the
total internal reflection (TIR). Here we utilize only the
critical angle condition kzs = 0 while the phenomenon of
TIR itself is not observed for the considered slabs, be-
ing available only for very thick layers. Since we need
the two conditions Zs = Za and kzs = 0 to be satis-
fied simultaneously for full transmission without phase
2
FIG. 2. (a) Transmission coefficient T through a dielectric
slab for TM polarization versus incidence angle. (b) Zoom
of the boxed area in (a). (c) The phase distribution inside
the slab under the critical angle incidence. The green line
describes the phase evolution of a plane-wave propagating
through vacuum within the distance ds. In all figures λ repre-
sents the vacuum wavelength. Parameters: εa = 20, µa = 1,
εs = 1 and εs = 1. The Brewster angle and the critical angle
of TIR equal 12.6◦ and 12.9◦, respectively.
accumulation, the Brewster angle and the critical an-
gle of total internal reflection should be close.
Indeed,
as illustrated by Fig. 2, the TM-polarized wave is fully
transmitted with no phase accumulation for conventional
dielectric materials (εa = 20 and εs = 1). Even though
the Brewster angle (θB = 12.6o) and the critical angle of
TIR (θc = 12.9o) do not coincide exactly, the system is
highly transparent at θ = θc, see Fig. 2 (b). Similarly
to an ENZ case the phase accumulation across the slab
is insignificant unlike the unbounded plane-wave propa-
gation in vacuum as shown in Fig. 2 (c). Interestingly,
the investigated phenomenon does not suffer from fre-
quency dispersion, because the critical angle of TIR and
Brewster's angle depend only on the practically disper-
sionless material parameters of the dielectrics, but not
on the frequency. It is demonstrated in Fig. 2 (b) with
the help of two curves corresponding to different thick-
nesses ds of the slab. Thus, we have obtained surprising
results demonstrating that the permittivity should not
necessarily vanish to achieve high transmission without
phase accumulation. Instead, under specific conditions,
conventional dielectrics can be exploited.
Generally speaking, the critical angle of TIR corre-
sponds to an exceptional point of degeneracy (EPD) since
the condition kzs = −kzs = 0 means that the propaga-
tion constants of the oppositely directed waves coincide
(degenerate), while the eigenvectors and, hence, eigen-
3
FIG. 3. Band-gap structure for (a) TE and (b) TM polarizations of a 1D photonic crystal depicted in Fig 1 (b). The black
dashed curves correspond to Brewster's law. (c) -- (e) Absolute value and phase of the transmission coefficient through a single
unit cell of the photonic crystal versus the normalized frequency: (c) normal incidence, (d) TE polarization, θ = 60◦, (e) TM
polarization, θ = 60◦. In all figures shaded regions represent propagation bands and red and blue curves and lines represent
the edges of stop bands where q = π/d and q = 0, respectively. (f) Spatial evolution of the electric field phase over periods of
the 4-cell 1D photonic crystal (circles and cubes) compared to the phase accumulation of a plane wave propagating through a
vacuum region of equivalent thickness (solid lines). Used parameters are ε1 = 2, ε2 = 16 and f = 0.1.
waves' impedances are indistinguishable. Thus, the phe-
nomenon of full transmission without phase accumula-
tion can be found for many different structures. Indeed,
let us consider a simple (for analytical derivations) ex-
ample of a 1D periodic structure which can be described
by an ABCD matrix [23] (or,
in other words, trans-
fer matrix) and has a symmetric unit cell (in this case
A = D [23]) of geometric size d. The asymmetric case
(A 6= D) is discussed in the Supplementary Material [24].
Bloch waves characterized by a wavenumber q and wave
impedance ZB propagate through the periodic structure.
Keeping in mind the unimodularity of the ABCD ma-
trix (A2 − BC = 1) for the non-dissipative structure,
one can write the Bloch wavenumber q = ±d−1 cos−1(A)
and wave impedance ZB = ±pC/B solving the eigen-
value and eigenvector problem for the ABCD matrix [23].
Here signs plus and minus correspond to forward and
backward Bloch waves. They cannot be distinguished at
the edges between stop and pass bands (i.e. when q = 0
and π/d), where the wave impedance diverges or van-
ishes by means of B = 0 or C = 0. Thus, the edges of
stop bands can be always treated as EPDs. Transmis-
sion coefficient through a N -cell 1D periodic structure
is given by the same Eq. (1), but the wave impedance of
the Bloch wave ZB, Bloch wavenumber q, and N d should
substitute respectively Zs, ksz, and ds. When the Brew-
ster condition ZB = Za is satisfied close to a band gap
edge, the transmission coefficient T = exp(iqN d) is able
to approach unity arbitrarily close. As in the case of a
homogeneous ENZ slab, the wave does not accumulate
phase when propagating. Thus, a discrete analogue of
the wave propagation phenomenon in homogeneous ENZ
media can be achieved in periodic structures. To that
end one needs to appropriately design a unit cell to get a
required ABCD matrix, e.g., see Ref. [23], where a design
of microwave networks is discussed.
As a concrete example we consider a 1D photonic
crystal represented by a periodic multilayer composed
of three-layer unit cells with an inversion center [25], as
shown in Figs. 1(b) and (c). EPDs are well studied in
such structures, see, e.g., [26 -- 28]. Complete wave tunnel-
ing without phase accumulation in 1D photonic crystals
is available for any incidence angle θ and any polariza-
tion, if the boundaries of the stop bands are almost flat.
The required band structure can be achieved for alter-
nating low- and high-permittivity dielectric layers with
a small filling fraction f = d1/d of the low-permittivity
dielectric. Then the stop bands become narrower, while
their edges are almost flat for both TE [Fig. 3 (a)] and
TM polarizations [Fig. 3 (b)]. The full transmission
T = 1 independent of the total thickness N d (Brew-
ster's resonance) appears at ZB = Za. The values of the
incidence angles and frequencies at which Brewster's law
in periodic multilayers is satisfied are depicted as dashed
lines in Figs. 3 (a) and (b).
It should be noticed that
Bloch impedance ZB is frequency dependent what gives
rise to frequency dispersion. For small filling fractions f ,
the curves ZB = Za shift to the top edges of the stop
bands, thus, providing both full transmission and zero
phase accumulation. For a discussion on EPDs in the
considered 1D photonic crystals see Supplementary Ma-
terial [24].
Transmission coefficient through a single unit cell (N =
1) as a function of the normalized frequency k0d = ωd/c
is demonstrated in Figs. 3 (c) -- (e). All maxima in Figs. 3
(c) -- (e) correspond to Brewster's resonances which are
4
tial evolution of the phase.
Realistic losses and random fluctuations of geometrical
parameters of the dielectric structures do not affect much
the overall performance (see Supplementary Material [24]
for details) in bright contrast to the ENZ media [18]. Ad-
ditionally, it is worth to note that we have considered only
propagation of monochromatic plane waves, however, an
information carrying signal would consists of a spectrum
of such waves. From Fig. 3 one can get information about
the frequency response and, particularly, see that the
group delay ∂Arg(T )/∂ω (an important characteristic in
signal processing [29]) increases when approaching the
frequencies of the top edges of stop bands (in coherence
with the well-known slow light phenomenon [30]). This
observation brings us to conclusion that only spectrally
narrow signals can propagate without distortion through
the considered photonic crystal when showing ENZ prop-
erties. See Supplementary Material for more details.
The polarization insensitive illusion effect represents
an impressive demonstration of the discussed phe-
nomenon. An ideal device making an illusion is invisi-
ble; therefore, its realization using a 1D photonic crystal,
which completely tunnels the wave without a phase ad-
vance in a vast range of incident angles, is natural. In
Fig. 4, a source is placed below the photonic crystal and
radiates at the frequencies specified by the top edge of
the stop band. Then an observer sees the source at a
shorter distance, the displacement towards the observer
being equal to the thickness of the photonic crystal slab.
Such an illusion is demonstrated in Fig. 4 for magnetic
and electric line sources, i.e.
for TE and TM polariza-
tions. It is worth noticing that an illusion phenomenon
was reported in Refs. [31 -- 34] on the base of transforma-
tion optics.
To summarize, we have identified the physical mech-
anism behind the properties of ENZ media, namely, the
Brewster effect occurring next to exceptional points of
degeneracy of an electromagnetic structure. This has al-
lowed us to demonstrate that vanishing of permittivity
is not required for achieving full transmission without
phase accumulation and can be realized with many dif-
ferent 1D periodic structures. Particularly, we demon-
strated the phenomenon using simple layered structures
composed of conventional dielectrics. These structures
not only offer a lossless and extremely simple, CMOS
compatible alternative for some applications of ENZ me-
dia, but exhibit new properties of all-angle polarization
insensitive full transmission with zero phase accumula-
tion. Weak sensitivity of the wave tunneling property
to variations of electromagnetic and geometrical param-
eters of the structure makes this system attractive for re-
alization of such extreme effects as source-displacement
illusion. Importantly, since exceptional points of degen-
eracy are ubiquitous, we expect that one can also realize
ENZ-like properties with more complex 2D and 3D non-
homogenizable structures.
FIG. 4. (a) Distribution of the magnetic field when a magnetic
line source illuminates a 1D photonic crystal. (b) Distribution
of the electric field when an electric line source illuminates
the photonic crystal. Normalized frequency is k0d = 0.86, the
distance to the photonic crystal is λ, the 1D photonic crystal
has N = 4 three-layer unit cells and the length of 20λ. The
eye represents an observer who sees an illusion (top star) of
the source (bottom star).
B + Z 2
independent of the total thickness N d. There are no
Fabry-Perot resonances at the band gap edges q = 0 and
q = π/d in Figs. 3 (c) -- (e), since ZB goes either to zero
or infinity, while the term (Z 2
a)/(2ZBZa) sin(qN d)
takes a non-zero value and, therefore, T < 1 according
to Eq. (1). On the contrary, when the Brewster condition
ZB = Za is satisfied next to stop band edges, transmis-
sion coefficient can approach unity arbitrarily close when
sufficiently small f is chosen (of course, f cannot be ex-
actly zero as there would be no band gap). Except for
the case of normal incidence θ = 0, the multilayer struc-
ture is polarization-sensitive. When the incidence angle
of the TE(TM) wave increases, the stop bands get wider
(narrower) [see Figs 3 (a) and (b)], while the widths of
the transmission resonances shown in Figs 3 (c)-(f) de-
crease (increase). Transmission coefficients for the waves
at the top edges of stop bands q = π/d and q = 0 possess
the phases respectively π or 0 as it is shown by dashed
curves in Figs. 3 (c) -- (e).
Figure 3 (f) demonstrates the phase of the electric field
at the points multiple to the period of the 4-cell multi-
layer. At the top edge of the first stop band q = π/d the
phases are close for even and odd periods, and there is
no phase accumulation at discrete points zm = 2md and
zm = (2m + 1)d (m is an integer number). At the top
edge of the second stop band q = 0 the phases are nearly
equal after every period of the multilayer, i.e. there is
no phase accumulation at discrete points zm = md. To
compare, plane waves propagating in unbounded vacuum
of the same thickness experiences significant phase accu-
mulation as shown by the solid lines in Fig. 3 (f). We
should stress that the phase within each unit cell chan-
des significantly, albeit the phase accumulation between
the periodically arranged points is almost absent. See
Supplementary Material [24] for more details on the spa-
SUPPLEMENTARY MATERIAL
S1: ENZ BEHAVIOR IN 1D PERIODIC
STRUCTURES WITH ASYMMETRIC UNIT
CELL
Electromagnetic wave propagation through one-
dimensional periodic structures can be described by
means of the ABCD matrix approach [23]. In the main
text, we discuss conditions of full transmission without
phase accumulation for waves penetrating periodic struc-
tures with a symmetric unit cell. Here we focus on
the distinct case of an asymmetric unit cell resulting in
A 6= D. Then, a Bloch wavenumber q is found to be [23]
q = ±
1
d
cos−1(cid:20) A + D
2
(cid:21) .
(2)
Meanwhile, wave impedances of the eigenwaves are given
by the equation
Z ±
B =
C
A−D
2 ∓ i sin(qd)
.
(3)
The top and bottom sings correspond to the waves prop-
agating along and against z-axis, respectively. From the
formula (3) it follows that Z +
B in this general
case. Furthermore, even in the lossless scenario the wave
impedance Z ±
B is complex-valued contrary to the case of
symmetric unit cells when the impedance is either real
(passband band) or imaginary (stop band). Then, the
formula (1) for the transmission coefficient takes the fol-
lowing, more general, form
B 6= −Z −
T =(cid:20)cos(qN d) − i
a − Z +
Z 2
Za(Z +
B Z −
B
B − Z −
B )
sin(qN d)(cid:21)
−1
.
(4)
Let us consider the behavior of the transmission coeffi-
cient at a stop band edge. It may seem that the transmis-
sion coefficient equals 1 when, e.g., q = 0. However, when
approaching a boundary of a stop band, sin(qd) → 0 and
accordingly Z +
B . That is, wave impedances at a
B → Z −
stop band boundary are not independent while the ex-
pression for the difference (Z +
B ) can be reduced as
follows:
B − Z −
5
C
A−D
2 − i sin(qd) −
C
A−D
2 + i sin(qd)
.
(5)
B − Z −
Z +
B =
=
2 (cid:1)2
(cid:0) A−D
2Ci sin(qd)
+ sin(qd)2
B Z −
B )/[Za(Z +
B − Z −
As a result, the denominator of the fraction (Z 2
a −
Z +
B )] goes to zero while the numer-
B − Z −
ator has a nonzero limit. Since the difference (Z +
B )
is proportional to sin(qd) [when sin(qd) is close to zero,
of course], sin(qN d)/(Z +
B ) also has a nonzero limit
proportional to N . Eventually, the transmission coeffi-
cient does not go to 1 at a stop band boundary. However,
B goes to Za when approaching a stop band
boundary one would observe full transmission without
phase accumulation.
if qZ +
B − Z −
B Z −
S2: ADDITIONAL DETAILS ON THE ENZ
BEHAVIOR OF 1D PERIODIC STRUCTURES
WITH A SYMMETRIC UNIT CELL
In this section, we provide additional details (on the ex-
ample of a 1D photonic crystal) on the exceptional points
of degeneracy in periodic structures having a symmet-
ric unit cell. After that we show a limiting behavior of
the transmission coefficient at stop band edges and make
some comments on a spatial evolution of the field passed
through the periodic structure.
We start with considering characteristics of Bloch
waves propagation through a periodic multilayer, whose
unit cell has an inversion center. The unit cell is formed
by three isotropic slabs, as illustrated by Fig. 1(c) of
the main text. Spatial evolution of eigenmode fields over
a period of the structure is described for each polariza-
tion by 2× 2 ABCD matrix P acting on the field column
(H, E)T [where H and E are the tangential components
of the magnetic and electric fields, respectively]. In case
of the 1D photonic crystal the components of the ABCD
matrix are as follows [25]
i
1
+
Z2
B =
A = cos(kz1f d) cos(kz2[1 − f ]d) −
2(cid:18) Z1
Z1 (cid:18)sin (kz1f d) cos (kz2[1 − f ]d) +
C = iZ1(cid:18)sin (kz1f d) cos (kz2[1 − f ]d) −
Z1(cid:19) sin(kz1f d) sin(kz2[1 − f ]d),
2(cid:20)(cid:18) Z1
Z2 −
2(cid:20)(cid:18) Z1
Z2 −
Z1(cid:19) +(cid:18) Z1
Z1(cid:19) −(cid:18) Z1
Z2
1
Z2
Z2
Z2
Z2
1
+
+
Z2
Z1(cid:19) cos (kz1f d)(cid:21) sin (kz2[1 − f ]d)(cid:19) ,
Z1(cid:19) cos (kz1f d)(cid:21) sin (kz2[1 − f ]d)(cid:19) .
Z2
(6)
Z1 and Z2 are the wave impedances of either TE- or TM-
polarized plane-waves in dielectric slabs of permittivities
ε1 and ε2, respectively, f is the fill fraction of ε1, d is
t , kt =
the thickness of the unit cell, kz1,2 =pε1,2k2
0 − k2
k0 sin θ, and θ is the angle of incidence.
A Bloch wave of the given polarization can be char-
acterized by a wavenumber q and wave impedance ZB.
They can be found by solving an eigenvalue problem for
6
FIG. 6. Frequency dependence of (a) the absolute value and
(b) phase of the transmission coefficient through a 1D pho-
tonic crystal having N unit cells. (c) Dependence of the field
phase on discrete period's number n of the 1D photonic crys-
tal having N = 4 unit cells compared to the phase accumula-
tion of a plane wave propagating through a vacuum region of
equivalent thickness. Parameters: normal incidence, εa = 1,
ε1 = 2(1+i0.01), ε2 = 16(1+i0.01) and f = 0.1. Calculations
were conducted in the assumption of 5% random errors in the
thicknesses of each dielectric layer.
Keeping in mind that sin(qd) = √−BC, we find the
transmission coefficient in the two limiting cases:
T →(cid:26) (1 − N C/(2Za))−1, ZB → ∞
(1 − N BZa/2)−1, ZB → 0.
(11)
Albeit none of these limits equal unity, the transmission
coefficient can approach 1 arbitrarily close at a band gap
edge, if the Brewster law is satisfied next to it (as in
case of the examples demonstrated in the main text and
Section S5 below).
Let us now discuss the spatial evolution of the field
FIG. 5. Frequency behaviour of Bloch wavenumber q, wave
impedance ZB and functions B and C. Parameters: θ = 0,
ε1 = 2, ε2 = 16 and f = 0.1.
the transfer matrix P . The Bloch wavenumber q sat-
isfies the dispersion equation cos[qd] = A or, equiva-
lently, sin[qd] = √−BC, since P is unimodular, that
is, A2 − BC = 1. The following explicit form of the dis-
persion equation can be found in the literature, e.g., in
Ref. [35]:
1
d
cos−1(cid:20)cos(kz1d1) cos(kz2d2) −
q = ±
× sin(kz1d1) sin(kz2d2)] .
1
2(cid:18) Z1
Z2
+
Z2
Z1(cid:19)
(7)
When eigenvalues of P are known, the wave impedances
of Bloch waves can be found by solving the eigenvector
equation
C A (cid:19)(cid:18) 1
(cid:18) A B
ZB (cid:19) = eiqd(cid:18) 1
ZB (cid:19) .
(8)
After some algebra one arrives at a simple expression for
the wave impedances:
ZB = ±r C
B
.
(9)
Signs plus and minus in Eqs. (7) and (9) correspond to
forward and backward Bloch waves, respectively. Edges
of stop bands correspond to zeros of either function B
or C, where wave impedance diverges or vanishes (see
Eq. (9) and Fig. 5). Thus, forward and backward Bloch
waves are indistinguishable at stop band edges which can
be treated as surfaces (owing to the axial symmetry) of
degeneracy.
In the general case, at a band gap edge, the transmis-
sion coefficient through N unit cells of a 1D photonic
crystal is not equal to 1. It follows from Eq. (1) in the
main text which can be rewritten as
T ="cos(qN d) − i
C
a
B + Z 2
2Za r B
C
sin(qN d)#−1
.
(10)
7
over periods of the structure. For the sake of brevity we
omit the phrase "over periods of the structure" in what
follows but implicitly assume it. There are two princi-
pal situations corresponding to the incident plane-wave
exactly matched with (i) the Brewster condition or (ii)
a stop band edge. In the first case, the incident wave is
impedance matched with the forward Bloch wave, which
spatial evolution through the structure is then given by
the phase factor exp[iqN d]. Phases of both magnetic and
electric fields vary as qN d. In the other situation one has
to turn to the ABCD matrix which at a stop band edge
has A = 1 and either B = 0 (C 6= 0) or C = 0 (B 6= 0).
The spatial evolution is given by the ABCD matrix acting
on the field column (1+R, Za(1−R))T at the input of the
structure (R is the reflection coefficient). When B = 0
the magnetic field does not change throughout the struc-
ture and has the phase Arg(H) = Arg(1+R). Meanwhile,
the phase profile of the electric field is given by the equa-
tion Arg(E) = Arg(N C[1 + R] + Za[1 − R]). If C is zero
and B is not, the electric field is constant with the phase
Arg(E) = Arg(1− R) and the phase of the magnetic field
changes as Arg(H) = Arg(N BZa[1 − R] + 1 + R). Both
situations become indistinguishable, when the Brewster
condition occurs at a stop band edge. It should be noted
that exactly the same is true for a homogeneous ENZ
slab which is treated by means of the similar analytical
technique and has the similar physics.
To conclude we would like to note that the main results
of this section are obviously applicable for any periodic
structure with a symmetric unit cell, not only the multi-
layer.
S3: ANALYSIS OF TOLERANCES
In this section we discuss tolerances with respect to
small changes in the frequency and structure's geometry
related to the 1D photonic crystal considered in the main
text. Requirements to the design of the unit cells are
quite flexible as one does not have to satisfy exact math-
ematical conditions. However, there is a general rule that
a photonic crystal should be formed of alternating high-
and low-permittivity dielectric layers, with the fill frac-
tion of the low-permittivity material being small. Influ-
ence of the dielectric losses and moderate (5%) random
errors of the layers thicknesses is demonstrated in Fig. 6.
The losses do not destroy the effect of zero phase delay,
but obviously reduce the transmission amplitude at the
resonance frequencies as clearly seen from comparison of
Figs. 3(c) -- (e) with Figs. 6(a), (b). The widths and po-
sitions of the peaks only slightly change compared with
the lossless multilayer. The random errors in addition to
the losses affect mainly the phase accumulation, but the
latter is still quite small, as shown in Fig. 6(c).
FIG. 7. Frequency dependence of normalized group delay.
(a), (b) Through a single unit cell for different incidence angles
in case of (a) TE and (b) TM polarizations. (c) Through 1D
photonic crystal having different number of unit cells N ver-
sus normalized frequency in case of normal incidence. Other
parameters: εa = 1, ε1 = 2, ε2 = 16 and f = 0.1.
S4: ANALYSIS OF GROUP DELAY
In the main text we deal only with monochromatic
plane waves. Generally, a spectral composition of such
waves (signal) may propagate with significant distortions
due to a complex frequency response of the propagation
medium. From Figs. 3 (c) -- (e) one can see that the
transfer function of the considered 1D photonic crystal
(namely, the transmission coefficient T ) significantly de-
pends on the frequency.
In order to estimate possible
signal distortion due to the nonlinear (with respect to
the frequency) phase response we consider group delay
τg defined as ∂Arg[T ]/∂ω. Group delay at the frequency
ω can be understood as the time it takes for a signal with
8
FIG. 8. (a), (c) Band gap structures of a photonic crystal illustrated in Fig. 1 (b): (a) TM polarization and (c) TE polarization.
(b), (d) Zooming of the boxed areas in Figures (a) and (c), respectively. The Brewster condition is satisfied at the black dashed
curves. Shaded regions represent propagation bands, red and blue curves represent the edges of stop bands where q = π/d and
q = 0, respectively. (e), (f) Angular dependence of the transmission coefficient through slabs of photonic crystals having 10
and 30 unit cells. (g), (h) Frequency dependence of the transmission coefficient through the 30 unit cells photonic crystal. In
all figures the material parameters are ε1 = 2, ε2 = 16, and the volume fraction of the low-index material is f = 0.9.
slow light phenomenon [30]), where we expect to observe
full transmission without phase delay. Notably, in case
of TE polarization the magnitude of maxima increases
with the incidence angle, while it decreases for TM po-
larization. Group delay increases and additional maxima
appear, when one adds unit cells to the photonic crystal
slab, see Fig. 7 (c). Overall, we can conclude that only
signals of small bandwidth can be used for transferring
information through such a photonic crystal possessing
ENZ properties.
S5: ANOTHER EXAMPLE WITH 1D PHOTONIC
ENZ-LIKE SPATIAL FREQUENCY FILTERING
CRYSTAL:
A thick epsilon-near-zero slab acts as an effective fil-
ter of spatial frequencies transmitting only normally in-
cident waves. Although in case of 1D photonic crystals
it is difficult to get such a functionality for the normally
incident wave, sharp transmission resonances are well ac-
cessible for grazing incident waves, when the wavevector
of the incident wave is almost parallel to the photonic
crystal interfaces. To that end, we study a multilayer
composed by alternating low- and high-permittivity di-
electric layers keeping the filling fraction f = d1/d of
the low-permittivity dielectric high. Figures 8(a) and (c)
depict the band gap structure for TM and TE polariza-
tions. The Brewster condition is met at the black dashed
curves. Blue and red curves correspond to the edges of
FIG. 9. Distribution of the field phase throughout the pho-
tonic crystal having 30 unit cells, cases of TM (k0d = 1.9595)
and TE (k0d = 4.0528) polarizations. The points represent
the field phase at the interfaces of unit cells. The structure
parameters are ε1 = 2, ε2 = 16, and f = 0.9.
a narrow spectrum (centered around ω) to traverse the
structure [29]. Figure 7 shows the normalized group delay
through the 1D photonic crystal considered in the main
text as a function of frequency. Results for different po-
larizations and incidence angles are presented by Figs. 7
(a) and (b) for a single unit cell. The maxima of the
group delay (except the one at ω = 0) correspond to the
top edges of band gaps (in coherence with the well-known
9
stop bands q = 0 and q = π/d, respectively. By magni-
fying the red boxed areas we clearly see from Figs. 8(b)
and (d) that the Brewster effect occurs at a stop band
edge for both polarizations (but at different frequencies),
when the incidence angle θ approaches 90 degrees.
In
Figs. 8(e) and (f) we plot the angular dependence of the
transmission coefficient at the corresponding frequencies
for photonic crystals composed of 10 (solid curve) and
30 (dashed curve) unit cells. Since the whole range of
the incidence angles apart from a vicinity of 90 degrees
corresponds to a stop band, highly selective transmission
is observed. The transmission resonances do not depend
on the total thickness N d of the multilayer that is they
are Brewster's resonances. When the Brewster condition
is satisfied at a stop band edge, the phase of the trans-
mission coefficient vanishes. It can be seen from the fre-
quency dependence of the transmission coefficient around
the stop band edge frequency (the incidence angle is 89.9
degrees), shown in Figs. 8(g) and (h), where one can ac-
tually recognize both the Brewster and the Fabry-Perot
resonances. The Brewster resonances appear right at the
stop band edge.
Full power transmission occurs also at Fabry-Perot res-
onances, where qN d = πm (m is an integer number), in
this case the transmission coefficient given by Eq. (1) in
the main text is T = 1/ cos(πm). The Fabry-Perot reso-
nances are seen in Figs. 8(g) and (h) at frequencies below
the Brewster peak at the stop band edge. Interestingly,
for even m we have T = 1 but the phase accumulation is
2πm. As it is noted above, the Brewster effect is indepen-
dent on N d, and there is no phase accumulation in the
full transmission regime, which makes it fundamentally
different from the Fabry-Perot resonances.
Absence of the phase accumulation over the full length
of the photonic crystal with 30 unit cells is demonstrated
by Fig. 9. Field phases at the interfaces of unit cells are
shown at the incidence angle of 89.9 degrees.
In case
of the TM polarization the stop band edge of interest
corresponds to q = π/d (k0d ≈ 1.96), and there is no
phase accumulation between discrete points zm = 2md or
zm = (2m+1)d (m is an integer number). For the TE po-
larization the Brewster condition is satisfied at the stop
band edge q = 0, when the frequency is approximately
k0d = 4.05, and zero phase accumulation is observed af-
ter every period.
In order to conclude this section let us note that the
results represented by Figs. 8 and 9 look nicer than those
demonstrated in the main text. It is due to the fact that
the wave impedances of incident plane-waves diverge for
TE polarization and vanish in case of TM one at ninety-
degrees incidence. Meanwhile, the wave impedances of
Bloch waves behave the same at stop band edges (see
Section S2). Thus, the Brewster condition is satisfied
exactly at the considered stop band edges.
[1] S. Jahani and Z. Jacob, "All-dielectric metamaterials,"
Nat. Nanotech. 11, 23 -- 36 (2016).
[2] S. A. Cummer,
J. Christensen,
and A. Al`u,
sound with acoustic metamaterials,"
"Controlling
Nat. Rev. Mat. 1, 16001 (2016).
[3] S. B. Glybovski, S. A. Tretyakov, P. A. Belov, Y. S.
Kivshar, and C. R. Simovski, "Metasurfaces: From mi-
crowaves to visible," Physics Reports 634, 1 -- 72 (2016).
[4] K. Sun, R. Fan, X. Zhang, Z. Zhang, Z. Shi, N. Wang,
P. Xie, Z. Wang, G. Fan, H. Liu, C. Liu, T. Li,
C. Yan,
and Z. Guo, "An overview of metamateri-
als and their achievements in wireless power transfer,"
J. Mater. Chem. C 6, 2925 -- 2943 (2018).
[5] X.C. Tong, Functional Metamaterials and Metadevices
(Springer, 2018).
[6] A. Poddubny, I. Iorsh, P. Belov, and Y. Kivshar, "Hyper-
bolic metamaterials," Nat. Photon. 7, 948 -- 957 (2013).
[7] V.P. Drachev, V.A. Podolskiy, and A.V. Kildishev, "Hy-
perbolic metamaterials: new physics behind a classical
problem," Opt. Express 21, 15048 -- 15064 (2013).
[8] R. Liu, Q. Cheng, T. Hand,
J.J. Mock, T.J.
and D.R. Smith, "Experimental
Cui, S.A. Cummer,
demonstration of electromagnetic tunneling through an
epsilon-near-zero metamaterial at microwave frequen-
cies," Phys. Rev. Lett. 100, 023903 (2008).
[9] B. Edwards, A. Al`u, M.E. Young, M. Silveir-
inha,
and N. Engheta, "Experimental verifica-
tion of epsilon-near-zero metamaterial coupling and
energy squeezing using
a microwave waveguide,"
Phys. Rev. Lett. 100, 033903 (2008).
[10] A. Al`u, M. Silveirinha, A. Salandrino,
and N. En-
gheta, "Epsilon-near-zero metamaterials and electromag-
netic sources: Tailoring the radiation phase pattern,"
Phys. Rev. B 75, 155410 (2007).
[11] C. Argyropoulos, G. D'Aguanno, and A. Al`u, "Giant
second-harmonic generation efficiency and ideal phase
matching with a double ǫ-near-zero cross-slit metama-
terial," Phys. Rev. B 89, 235401 (2014).
[12] M. A. Vincenti, D. de Ceglia, A. Ciattoni,
and
M. Scalora, "Singularity-driven second- and third-
harmonic generation at ǫ-near-zero crossing points,"
Phys. Rev. A 84, 063826 (2011).
[13] C. Argyropoulos,
N. Engheta,
linearities
Phys. Rev. B 85, 045129 (2012).
in
P.-Y. Chen, G. D'Aguanno,
and A. Al`u, "Boosting optical non-
channels,"
ǫ-near-zero
plasmonic
[14] S. Feng, "Loss-induced omnidirectional bending to the
normal in ǫ-near-zero metamaterials," Phys. Rev. Lett.
108, 193904 (2012).
[15] I. Liberal and N. Engheta, "Near-zero refractive index
photonics," Nat. Photon. 11, 149 -- 159 (2017).
[16] I. Liberal, A.M. Mahmoud, Y. Li, B. Edwards,
and
N. Engheta, "Photonic doping of epsilon-near-zero me-
dia," Science 355, 1058 -- 1062 (2017).
[17] N. Garcia, E.V. Ponizovskaya, and J.Q. Xiao, "Zero per-
mittivity materials: Band gaps at the visible," Applied
Physics Letters 80, 1120 -- 1122 (2002).
[18] M.H. Javani and M.I. Stockman, "Real and imaginary
properties of epsilon-near-zero materials," Phys. Rev.
Lett. 117, 107404 (2016).
[19] X. Huang, Y. Lai, Z.H. Hang, H. Zheng, and C.T. Chan,
"Dirac cones induced by accidental degeneracy in pho-
tonic crystals and zero-refractive-index materials," Nat.
Mat. 10, 582 (2011).
[20] P. Moitra, Y. Yang, Z. Anderson, I.I. Kravchenko, D.P.
Briggs, and J. Valentine, "Realization of an all-dielectric
zero-index optical metamaterial," Nat. Photon. 7, 791
(2013).
[21] N.C. Panoiu, R.M. Jr. Osgood, S. Zhang, and S.R.J.
Brueck, "Zero-n bandgap in photonic crystal superlat-
tices," J. Opt. Soc. Am. B 23, 506 -- 513 (2006).
[22] S. Kocaman, M.S. Aras, P. Hsieh, J.F. McMillan, C.G.
Biris, N.C. Panoiu, M.B. Yu, D.L. Kwong, A. Stein, and
C.W. Wong, "Zero phase delay in negative-refractive-
index photonic crystal superlattices," Nat. Photon. 5,
499 (2011).
[23] David M Pozar, Microwave engineering (John Wiley &
Sons, 2009).
[24] Supplementary Material.
[25] V. Popov, A. V. Lavrinenko, and A. Novitsky, "Sur-
face waves on multilayer hyperbolic metamaterials: Op-
erator approach to effective medium approximation,"
Phys. Rev. B 97, 125428 (2018).
[26] J.M. Bendickson, J.P. Dowling, and M. Scalora, "Ana-
lytic expressions for the electromagnetic mode density in
finite, one-dimensional, photonic band-gap structures,"
Phys. Rev. E 53, 4107 -- 4121 (1996).
I. Vitebskiy,
[27] A. Figotin
and
"Electromagnetic
crystals,"
photonic
unidirectionality
in magnetic
Phys. Rev. B 67, 165210 (2003).
[28] A. Figotin and I. Vitebskiy, "Gigantic transmission band-
10
edge resonance in periodic stacks of anisotropic layers,"
Phys. Rev. E 72, 036619 (2005).
[29] A. V. Oppenheim, A. S. Willsky, and S. H. Nawab, Sig-
nals & Systems (2Nd Ed.) (Prentice-Hall, Inc., Upper
Saddle River, NJ, USA, 1996).
Scalora, M.
J. Bloemer,
"The photonic band edge
A new approach to gain enhancement,"
[30] J. P. Dowling, M.
and C. M. Bowden,
laser:
Journal of Applied Physics 75, 1896 -- 1899 (1994),
https://doi.org/10.1063/1.356336.
[31] Y. Lai, J. Ng, H.-Y. Chen, D.-Z. Han, J.-J. Xiao, Z.-
Q. Zhang, and C. T. Chan, "Illusion optics: The op-
tical transformation of an object into another object,"
Phys. Rev. Lett. 102, 253902 (2009).
[32] W. X.
J. Cui,
Jiang, H. F. Ma, Q. Cheng,
and
vir-
metamateri-
Applied Physics Letters 96, 121910 (2010),
T.
tual
als,"
https://doi.org/10.1063/1.3371716.
"Illusion media:
Generating
realizable
objects
using
[33] P. H. Tichit, S. N. Burokur, J. Yi,
"Transformation electromagnetics
trac,
nas with an illusion on the
IEEE Antennas and Wireless Propagation Letters 13, 1796 -- 1799 (2014).
and A. de Lus-
anten-
radiation pattern,"
for
J. Luo,
[34] Z. Yao,
via
crystals with
tics
tonic
Opt. Express 25, 30931 -- 30938 (2017).
one-dimensional
and Y. Lai,
shifted
spatial
ultratransparent
"Illusion op-
pho-
dispersions,"
[35] P. Yeh, A. Yariv,
and C.-S. Hong, "Electromagnetic
propagation in periodic stratified media. I. General the-
ory*," J. Opt. Soc. Am. 67, 423 -- 438 (1977).
|
1811.10386 | 1 | 1811 | 2018-10-19T20:02:16 | CMUT on Glass Substrate for Next-Generation Medical Imaging and Beyond | [
"physics.app-ph"
] | CMUT technology was invented and has been widely researched on silicon substrates. In recent years, fabrication of CMUT on glass substrates has raised significant interests because a number of advantages it can offer. First, parasitic capacitance could be easily reduced since glass is an insulating material. Second, insulation steps required in the silicon-based CMUT fabrication process could be avoided, leading to a reduced fabrication complexity. Third, anodic bonding can be used when the substrate is borosilicate glass, which is a low-temperature bonding technique and has a high tolerance to bonding surface area and roughness in addition to the advantages of silicon wafer bonding. Moreover, glass transparency can enable novel applications beyond pulse-echo ultrasound medical imaging and enlarge the potential markets of CMUTs. | physics.app-ph | physics | ABSTRACT
ZHANG, XIAO. Capacitive Micromachined Ultrasonic Transducers (CMUTs) on Glass
Substrates for Next-Generation Medical Imaging and Beyond. (Under the direction of Dr. Ömer
Oralkan).
Capacitive Micromachined Ultrasonic Transducers (CMUTs) have been widely researched
for the last two decades. One of the most immediate applications is ultrasound medical imaging.
Comparing to the traditional medical ultrasound transducer technology based on piezoelectric
materials, CMUT technology offers benefits including fabrication of large transducer arrays, close
integration with electronics, and broad bandwidth in immersion. CMUT technology is especially
competitive for fabrication of 2D transducer arrays and their close integration with electronics for
real-time 3D imaging and for fabrication of high-frequency broadband transducer arrays for high-
resolution imaging.
CMUT technology was invented and has been widely researched on silicon substrates. In
recent years, fabrication of CMUT on glass substrates has raised significant interests because a
number of advantages it can offer. First, parasitic capacitance could be easily reduced since glass
is an insulating material. Second, insulation steps required in the silicon-based CMUT fabrication
process could be avoided, leading to a reduced fabrication complexity. Third, anodic bonding can
be used when the substrate is borosilicate glass, which is a low-temperature bonding technique and
has a high tolerance to bonding surface area and roughness in addition to the advantages of silicon
wafer bonding. Moreover, glass transparency can enable novel applications beyond pulse-echo
ultrasound medical imaging and enlarge the potential markets of CMUTs.
This dissertation investigates the design and fabrication of several CMUT-based key
components on glass substrates to provide a tool kit for the development of the next-generation
ultrasound system frontends. On one side, a densely populated 2D transducer array with through-
wafer interconnects and a MEMS transmit/receive (T/R) switch were developed to overcome the
transducer integration challenges. On the other side, by taking the advantage of glass substrate
transparency, a transparent CMUT was developed for both medical applications and consumer
electronics applications.
We first developed a platform process to fabricate vacuum-sealed CMUTs on a glass substrate
with anodic bonding. The CMUT performance and the array uniformity were characterized. Low
parasitics, good process control, and good array uniformity were achieved by this platform
technology. We further demonstrated a high-frequency (30-MHz), broadband (100%) 1D CMUT
array based on this process.
Furthermore, we extended the platform process and built a 16×16 2D CMUT array by
incorporating through-glass-via (TGV) interconnects. The parasitic capacitance of a via pair with
a 250-µm pitch was measured as 21 fF. The resistance of a single via plus via-to-electrode contact
resistance was 2
. The 2D CMUT array element showed a low parasitic capacitance.
In addition, we developed a MEMS switch co-fabricated with a CMUT. Static and dynamic
characterization was performed in air and in immersion, respectively. The MEMS switch showed
a dc switching voltage of 68 V and could be operated using a control voltage of 2.5 V when biased
at 67 V. The optimum switching time is 1.34 µs and the optimum release time is 80 ns.
Last but not least, we demonstrated a CMUT with improved transparency for backward-mode
PAI and discussed the effect of silicon plate absorption. Second, a fully transparent CMUT was
fabricated with two ITO coated glass wafers by adhesive bonding. The fabricated device shows
60%-80% optical transmittance in the full visible wavelength range.
© Copyright 2017 by Xiao Zhang
All Rights Reserved
Capacitive Micromachined Ultrasonic Transducers (CMUTs) on Glass Substrates for
Next-Generation Medical Imaging and Beyond
by
Xiao Zhang
A dissertation submitted to the Graduate Faculty of
North Carolina State University
in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy
Electrical and Computer Engineering
Raleigh, North Carolina
2017
APPROVED BY:
_______________________________
Dr. Ömer Oralkan
Committee Chair
_______________________________
Dr. Troy Nagle
_______________________________
Dr. Alper Bozkurt
_______________________________
Dr. Yun Jing
ii
BIOGRAPHY
Xiao Zhang received the B.S. degree from Xi'an Jiaotong University, Xi'an, China, in 2012. He
was then admitted to the direct doctoral program in the Department of Electrical and Computer
Engineering at North Carolina State University, Raleigh, NC.
His research focuses on design and fabrication of 1D and 2D capacitive micromachined ultrasonic
transducer (CMUT) arrays and associated MEMS transmit/receive switches on glass substrates to
overcome the transducer integration challenges with the front-end electronics. He also developed
transparent CMUTs for medical applications and consumer electronics applications.
He has authored or co-authored over 20 scientific publications. He received student travel awards
in the 2015, 2016, and 2017 IEEE International Ultrasonics Symposia. He was a student paper
competition finalist in the 2016 and 2017 IEEE International Ultrasonics Symposia. He won the
first place in 2017 NCSU ECE Research Symposium and he was also a recipient of the UGSA
Conference Award at NCSU in 2015.
iii
ACKNOWLEDGMENTS
First of all, I would like to express my sincere gratitude to my Ph.D. advisor Dr. Ömer
Oralkan. I was fortunate to be the first student in his research group and also the first Ph.D. student
to graduate. Being the first is a big challenge but it is also the most rewarding experience through
my five years of study. During the years I worked with Dr. Oralkan, he has been not only my
academic advisor but also my mentor. What I learned from him is not only the knowledge about
ultrasound and MEMS but also the approach and attitude to address a research problem with little
or no references. He not only helped me to make breakthroughs on my research projects but also
guided me to be a successful researcher and engineer who could be positive and do not quit even
facing the most challenging technical problems.
I would also like to acknowledge the other professors in my Ph.D. committee, Dr. Alper
Bozkurt, Dr. Yun Jing, and Dr. Troy Nagle. The knowledge I acquired on their courses about
human-machine interfaces, bio-electronics, and acoustics is very helpful for my research and my
understanding of electrical and computer engineering. I would also like to thank them for working
with me through all the milestones in my Ph.D. program from qualification exam, preliminary
exam, and to the final defense.
During my years of study at NC State, I'm very glad to know and become friends with my lab
mates and colleagues. Dr. Feysel Yalcin Yamaner joined the group on the same day as me. He
taught me the basics of CMUT, finite-element modeling, and helped me make my first CMUT in
the cleanroom. Xun, I enjoyed our useful discussions towards the medical imaging algorithms and
iv
also our fun chats about Japanese music and animation. Jeanne, thank you for correcting my
grammar and pronunciation before my conference presentations. Marzana, thanks for inviting us
to your baby's events and I wish your daughter happiness every day. Femi, I won't forget the
countless hours we spent together in the cleanroom. CK, thank you for praying for me during my
difficult times. Besides, I would also like acknowledge other colleagues and close friends: Dr.
Xiaoning Jiang, Dr. Sibo Li, Dr. Chen Shen, Mr. Weiyi Chang, Dr. Zhenhao Li, Dr. Szuheng Ho,
Dr. Xu Zhang, Dr. Murat Yokus, Dr. Lujun Huang, Mr. Peter Sotory, Mr. Rupak Roy, Mr. Mohit
Kumar, Dr. Bhoj Gautam, Mr. Robert Younts, Ms. Feiyan Lin, Mr. Dingjie Suo, Mr, Steve Lipa,
Mr. Joe Matthews, etc.
In addition, I would like to thank all the current and previous NNF staff members (Marcio,
Nicole, Henry, Bruce, and Jeff) and AIF staff members (Chuck and Roberto) for their detailed
instructions on microfabrication tools and microscopy techniques, as well as the fun chats we had
while waiting for the process to complete.
Finally, I want to express my deepest gratitude to my family and my beloved wife. To my
parents and my elder brother: Your unconditional love started from the very beginning of time.
You made everything possible for me and I'm blessed to be born in this family. To my wife Hui:
Meeting, dating, marrying you were the most exciting moments in my life. Thank you for always
standing by my side and encouraging me even during our most difficult times.
This dissertation is dedicated to my parents, my elder brother, and my wife.
v
TABLE OF CONTENTS
LIST OF TABLES .................................................................................................................. vii
LIST OF FIGURES ............................................................................................................... viii
Chapter 1 INTRODUCTION ................................................................................................... 1
1.1 Overview of Ultrasound Transducers ..........................................................................1
1.2 Capacitive Micromachined Ultrasonic Transducers (CMUTs) ...................................2
1.3 Summary of Contributions ...........................................................................................6
1.4 Thesis Organization ...................................................................................................10
Chapter 2 CAPACITIVE MICROMACHINED ULTRASONIC TRANSDUCERS
(CMUTs)
2.1 CMUT Structure and Operation Principle .................................................................11
2.2 CMUT Performance Attributes..................................................................................13
2.3 CMUT Fabrication Process........................................................................................26
2.4 CMUT Integration with Electronics ..........................................................................31
2.5 Summary ....................................................................................................................34
Chapter 3 PLATFORM TECHNOLOGY: VACUUM-SEALED CMUTS FABRICATED
ON GLASS SUBSTRATE USING ANODIC BONDING .................................................... 36
3.1
Introduction ................................................................................................................36
3.2 Fabrication Process Development..............................................................................37
3.3 Characterization .........................................................................................................43
3.4 High-Frequency Broadband 1D CMUT array ...........................................................47
3.5 Chapter Conclusions and Future Work ......................................................................50
Chapter 4 2D CMUT ARRAYS WITH TGV INTERCONNECTS ................................... ...51
4.1
Introduction ................................................................................................................51
4.2 Through wafer interconnects .....................................................................................52
4.3 Fabrication of 2D CMUT with TGV interconnect ....................................................54
4.4 Device Characterization .............................................................................................65
4.5 Chapter Conclusions and Future Work ......................................................................71
Chapter 5 A MEMS T/R SWITCH EMBEDDED IN CMUT STRUCTURE ...................... 73
5.1 Transmit/Receive Switching in a Ultrasound Imaging System .................................74
5.2 MEMS Switches ........................................................................................................77
vi
5.3 Design of a MEMS Switch Embedded in CMUT Structures ....................................80
5.4 Fabrication Process ....................................................................................................87
5.5 Device characterization ..............................................................................................94
5.6 Chapter Conclusions and Future Work ....................................................................102
Chapter 6 TRANSPARENT CMUTS ON GLASS SUBSTRATES ................................... 104
6.1 Applications of Transparent Transducers ................................................................104
6.2 CMUT with ITO Bottom Electrodes for Improved Transparency ..........................105
6.3 Backward-mode Photoacoustic Imaging .................................................................111
6.4 An Optically Transparent CMUT Fabricated Using SU-8 or BCB Adhesive Wafer
Bonding ..............................................................................................................................121
6.5 Chapter Conclusions and Future Work ....................................................................128
Chapter 7 Summary ............................................................................................................. 130
vii
LIST OF TABLES
Table 1-1: Comparison of transducer technologies for ultrasound imaging applications .............. 5
Table 4-1: Physical dimension of the fabricated 2D CMUT array element. ................................ 64
Table 4-2: Average via-to-via capacitance: Measurement and Simulation Results ..................... 69
Table 4-3: Material Properties Used in Simulations. .................................................................... 71
Table 4-4: Simulation Versus measured device performance in air ............................................. 71
Table 5-1: Important dimensional parameters for the Rockwell MEMS switch [75]. ................. 79
Table 5-2: Examples of the calculated switching time ................................................................. 84
Table 5-3: Material properties used in FEM simulations ............................................................. 85
Table 5-4: Physical parameters of the tested switch ..................................................................... 93
Table 6-1: Physical Dimensions of the Fabricated Transparent CMUTs ................................... 125
viii
LIST OF FIGURES
Figure 1-1: Examples of ultrasound applications at various frequencies. ...................................... 2
Figure 1-2: A limited overview of CMUT technology progress .................................................... 4
Figure 2-1: Different implementations of a CMUT cell. .............................................................. 11
Figure 2-2: CMUT arrays of different geometries fabricated on the same wafer ........................ 12
Figure 2-3: Equivalent circuit model ............................................................................................ 15
Figure 2-4: CMUT with extended insulation layer. ...................................................................... 29
Figure 2-5: CMUT with a thick buried oxide layer. ..................................................................... 29
Figure 2-6： CMUT fabrication technologies categorized by substrate types............................. 31
Figure 3-1： 3D cross-sectional model of a completed CMUT cell. ........................................... 38
Figure 3-2: Fabrication process flow ............................................................................................ 42
Figure 3-3: SEM cross-sectional image of an accomplished CMUT cell. ................................... 43
Figure 3-4: Zygo interfereometer measurement ........................................................................... 44
Figure 3-5: Electrical input impedance measurement .................................................................. 45
Figure 3-6: Immersion measurement using a calibrated hydrophone ........................................... 46
Figure 3-7: (a) A 66-element 1D CMUT array fabricated on glass substrate using anodic
bonding. (b) Array uniformity measurement. ............................................................................... 47
Figure 3-8: Gap height can be control as small as 50 nm through the fabrication process .......... 48
Figure 3-9: (a) Experimental received signal by the hydrophone. (b)Corresponding frequency
spectrum. ....................................................................................................................................... 49
Figure 4-1: A schematic cross-section of a completed CMUT element with TGV interconnects
flip-chip bonded on an IC. ............................................................................................................ 54
Figure 4-2: Initial borosilicate substrate. ...................................................................................... 55
Figure 4-3: Define patterned through-wafer holes by laser drilling. ............................................ 56
Figure 4-4: Fill the through holes with copper paste. ................................................................... 56
Figure 4-5: SEM cross-section of the TGV embedded in the glass substrate (left). AFM image of
the TGV surface profile (right). .................................................................................................... 56
Figure 4-6: Glass and copper etch to define CMUT cavities........................................................ 58
Figure 4-7: Evaporation and lift-off the bottom electrode and forming electrical connection to the
dedicated TGV. ............................................................................................................................. 58
Figure 4-8: SEM images of the completed glass substrate with TGVs that is ready for anodic
bonding. ........................................................................................................................................ 58
Figure 4-9: Deposition of silicon nitride insulation layer on SOI device layer. Anodic bonding in
vacuum. ......................................................................................................................................... 60
Figure 4-10: Handle wafer and BOX layer removal. .................................................................... 60
Figure 4-11: Silicon/silicon nitride etch for gas evacuation, array separation, and reaching top
electrode connection. .................................................................................................................... 61
Figure 4-12: Conformal PECVD silicon nitride deposition for sealing. ...................................... 61
Figure 4-13: Silicon nitride etch to reach the plate and top electrode TGV. ................................ 63
Figure 4-14: Sputter top electrode and build connection to dedicated TGV. ............................... 63
ix
Figure 4-15: Evaporation and lift-off for backside contact pad and testing grid. ......................... 63
Figure 4-16: Completed device. .................................................................................................... 64
Figure 4-17: Measured and simulated deflection profile of the plate after the device fabrication
was completed. .............................................................................................................................. 65
Figure 4-18: Via test structures (left panel), via resistance measurement setup (right panel top),
and via-to-via capacitance measurement setup (right panel bottom). ........................................... 67
Figure 4-19: Measured results for resistance test structures. ........................................................ 67
Figure 4-20: Measured results for capacitance test structures. ..................................................... 69
Figure 4-21: Impedance measurements (Vdc = 15 V). ................................................................. 70
Figure 5-1: Top-level block diagram of a single TX/RX channel for ultrasound frontends ........ 75
Figure 5-2: Current implementation of T/R switches using diodes pair (single channel) ............ 75
Figure 5-3: A RF MEMS switch example: Rockwell Scientific MEMS series switch [74] ........ 78
Figure 5-4: (a) MEMS switch structure modified from a single CMUT cell. (b) T/R switch in an
ultrasound imaging system. .......................................................................................................... 81
Figure 5-5: (a) Two-switch configuration where TX and RX paths are totally isolated. (b) Three-
switch configuration for seamless integration of low-voltage frontend IC with a standard imaging
system. .......................................................................................................................................... 83
Figure 5-6: Plate structure model and created elements for static simulation. ............................ 86
Figure 5-7: Fabrication process flow ............................................................................................ 88
Figure 5-8: AFM image of the center region of the switch cell (left), and the corresponding metal
contact. .......................................................................................................................................... 89
Figure 5-9: AFM images of the switch structure and the co-fabricated CMUT structure ............ 89
Figure 5-10: Optical images of the critical processing steps ........................................................ 91
Figure 5-11: Cross-sectional SEM image of the finished switch ................................................. 92
Figure 5-12: A switch die wire bonded on the PCB. .................................................................... 94
Figure 5-13: Static deflection at different bias voltages. (a) Wyko measured results. (b) FEM
static simulation results. ................................................................................................................ 96
Figure 5-14: Steady-state dc measurement. .................................................................................. 96
Figure 5-15: Dynamic characterization with a control signal rise time of (a) 100 ns; (b) 200 ns;
and (c) 300 ns. Figure (d)-(f) shows the corresponded plate center displacement. ...................... 98
Figure 5-16: Switch switching time versus control signal rise time. ............................................ 99
Figure 5-17: Dynamic characterization with a control signal fall time of (a) 100 ns; (b) 200 ns;
and (c) 300 ns, respectively. Figure (d)-(f) shows the corresponded center displacement. ........ 100
Figure 5-18: Switch releasing time versus control signal fall time. ........................................... 100
Figure 5-19: (a) 1-MHz, 300-mVpp CW input signal turned on and off by a 1-kHz control signal.
(b) 1-MHz, 5-Vpp pulse signal turned on and off by a 1-kHz control signal. ............................. 101
Figure 6-1: Simplified fabrication process flow. ........................................................................ 106
Figure 6-2: Optical image after the ITO bottom electrode definition (left). AFM image showing
the profile of the glass post and ITO bottom electrode (right). .................................................. 107
Figure 6-3: Atmospheric deflection measurement of a finished CMUT element. The inset shows
a finished CMUT element. .......................................................................................................... 107
x
Figure 6-4: (a) Six CMUT elements with ITO bottom electrodes show the improved transparency
of the bottom electrodes. (b) Six CMUT elements with Cr/Au bottom electrodes presented for
comparison. ................................................................................................................................. 108
Figure 6-5: (a) Transmission measurement. Transmission of the bottom. (b) Transmission of the
final devices. ............................................................................................................................... 109
Figure 6-6: (a) Electrical input impedance measurement in air. (b) Pulse-echo measurement in
vegetable oil. ............................................................................................................................... 110
Figure 6-7: A schematic diagram of the CMUT with improved transparency ........................... 116
Figure 6-8: (a) Bottom view of the PCB with CMUT (inset graph indicates the relative location
of the laser output and CMUT elements). (b) Side view of PCB attached on the holder with the
optical fiber bundle in the back. .................................................................................................. 116
Figure 6-9: Pencil rise cross-sectional PA imaging results: (a) A sample A-scan at X=0 mm and
its Fourier transform after applying a Gaussian window; (b) Signal paths of the four signals on
the A-scan; (c) Reconstructed image. ......................................................................................... 118
Figure 6-10: (a) Received PA signal using 1-mJ laser power at 12-mm travel distance: one-way
pencil rise signal (PAtarget) and two-way silicon plate signal (PAcmut). (b) Equivalent electrical
excitation amplitude for 1 mJ laser excitation. ........................................................................... 119
Figure 6-11: ICG tube phantom (a) Photograph of the ICG-filled polyethylene tube. (b) 3D
rendered image of the ICG-filled polyethylene tube. ................................................................. 120
Figure 6-12: Fabrication of fully transparent CMUT using adhesive bonding. ......................... 124
Figure 6-13: (a) Profilometer measurement of the formed cavities with ITO bottom electrodes on
a glass substrate. (b) Atmospheric deflection of a fabricated CMUT cell. ................................. 125
Figure 6-14: Fabricated fully transparent CMUT. Optical photos (top), transmittance
measurement (bottom). (a) SU-8 bonded wafer pair. (b) BCB bonded wafer pair. ................... 126
Figure 6-15: Input electrical impedance measurement. Real part (top). Imaginary part (bottom).
(a) SU-8 bonded wafer pair. (b) BCB bonded wafer pair. .......................................................... 127
1
Chapter 1 INTRODUCTION
1.1 Overview of Ultrasound Transducers
Ultrasound is defined as sound waves that have a frequency above 20 kHz. An ultrasound
transducer is a device that could transmit ultrasound by converting electrical energy into acoustical
energy, and could also receive ultrasound by converting acoustic energy into electrical energy. The
history of modern ultrasound can be traced back to the discovery of piezoelectric effect in 1880 by
French physicists Jacques and Pierre Curie [1]. Piezoelectric ceramics were first introduced as
ultrasonic transducers during the 1940s. Later, piezoelectric materials, including polyvinylidene
fluoride (PVDF), were introduced in the 1960s, piezo-composites in 1980s, and piezoelectric
single crystals in the 1990s.
Since the late 1960s, the exponential growth of the microelectronic industry has largely
accelerated the application of ultrasonic devices in different areas, such as ultrasound medical
imaging, nondestructive testing (NDT), acoustic microscopy, flow metering, surface acoustic
wave (SAW), and bulk acoustic wave (BAW) devices. Figure 1-1 summarizes the ultrasound
applications at different frequency ranges. The application market for ultrasound still grows
steadily. More recently, ultrasound fingerprint sensors, ultrasound wireless power transfer, etc. are
being widely investigated and pushed towards commercialization.
2
Figure 1-1: Examples of ultrasound applications at various frequencies.
1.2 Capacitive Micromachined Ultrasonic Transducers (CMUTs)
CMUTs were made practical in the mid-1990s [2], although the idea of capacitive ultrasound
transducers is as old as early piezoelectric transducers. The reason why capacitive transducers had
not been popular is that a high electric field strength of the order of a million volt per centimeter
(106 V/cm) is required [3], which is critical for delivering a reasonable transducer efficiency. The
explosive growth of microelectronics and MEMS technology in the late 1960s and 1970s has
enabled the definition of a device dimension with an unprecedented accuracy. With the advanced
micro-fabrication and micromachining techniques, a very shallow (sub-micron) cavity could be
defined in a capacitive transducer, which translates to a relatively low voltage (<100 V) to reach
the required electric field strength, making electrostatic ultrasound transducers possible.
3
The first generation CMUTs were built on silicon wafers with no vacuum-sealing for air-
coupled applications [2]. Since then, extensive research effort has been conducted to improve the
device performance and reliability as well as to exploit the potential applications.
On the device aspect, various design and fabrication approaches have been investigated,
including vacuum sealing of the cavity, dual-electrode designs, various plate geometries, reduction
of electric charging, etc. In terms of device fabrication, CMUT was originally invented on silicon
substrates using the sacrificial release process. Later in 2003, the CMUT fabrication was
demonstrated by direct wafer-to-wafer fusion bonding technique [4], which offers better
dimensional and stress control compared to surface micromachining. During the same period, 2D
CMUT arrays with through-silicon-via (TSV) interconnects were developed for 3D real-time
ultrasound imaging [5]. Later trench-isolated through-wafer interconnects were developed which
eliminate the complicated process to make TSVs and are also compatible with wafer bonding [6].
High-frequency CMUT arrays were also demonstrated for high-resolution imaging where deep
penetration is not required [7]. Later in 2009, row-column addressed CMUT (RC-CMUT) arrays
were demonstrated and investigated for 3D ultrasound imaging [8]. In recent years, CMUT has
been made on insulating substrates in order to further reduce the device parasitics and the process
complexity [9] -- [11]. Also, adhesive bonding has been proposed as an alternative way to build
CMUTs [12]. More recently, transparent CMUT has been investigated for applications where
optics and acoustics need to be combined [13]. An overview of the above-described progress
regarding CMUT fabrication is depicted below.
4
Figure 1-2: A limited overview of CMUT technology progress. The underlined items indicate
author's contribution.
In terms of CMUT application, although first developed for air-coupled applications, CMUT
later found its most promising application in ultrasound medical imaging. The tremendous interest
of using CMUT technology as an alternative or complement for piezoelectric medical transducers
stems from the advantages offered by CMUT, such as ease of fabricating large arrays and
integration with electronics, as well as the wide bandwidth in immersion. CMUT is especially
attractive for high-frequency ultrasound imaging and 3D real-time ultrasound imaging, where
transducer arrays are difficult to fabricate and integrate with electronics using the traditional
5
piezoelectric technology and its associated fabrication approach. Table 1-1 is a summary of the
important advantages of CMUT technology compared to the piezoelectric transducer technology
for medical imaging applications.
Table 1-1: Comparison of transducer technologies for ultrasound imaging applications
Piezoelectric transducers
CMUT
Fabrication method
Ceramic technology
MEMS technology
Array fabrication
Difficult for 2D and high-
through-wafer interconnects.
frequency arrays
Enable high-frequency
2D arrays enabled by
broadband transducer.
Moderate, matching layer
Wide, no matching layer
Transducer bandwidth
Array uniformity
required
Moderate
Thermal stability
Low
IC integration
Difficult
needed
High
High
Monolithic or hybrid
integration
Output pressure
High
Relatively low but improving
Researchers have progressed from early demonstration of the CMUT operation to the full
integration of 2D [14], 3D [15], [16], intravascular [17] -- [19], and photoacoustic imaging systems
[20], [21]. CMUT for medical imaging has reached the commercial market in 2009 by Hitachi,
Ltd. [22] and the first practical CMUT 1D linear array probe was presented in 2015 by KOLO
6
Medical Inc. [23], which is originated from the Stanford research group where CMUT was
invented.
In addition, CMUTs have been demonstrated for therapeutic ultrasound, CMUTs can provide
MR-compatible HIFU [24], [25]. Apart from medical ultrasound, CMUT has also been widely
researched for applications such as chemical sensors [26] -- [28], parametric speaker [29], and
biometric applications such as ultrasonic fingerprint sensor, ultrasonic touchless interactive panels
[30] -- [32], etc.
1.3 Summary of Contributions
The first objective of this dissertation is to develop a new approach to fabricate CMUT 1D
and 2D arrays on glass substrates. The CMUTs fabricated on glass substrates simplifies the
fabrication flow and reduce the device parasitics and thus improve the transducer efficiency. The
second objective is to facilitate the CMUT integration with electronics. On one side, we
demonstrate the CMUT 2D arrays with through-glass-via interconnects that are feasible for hybrid
integration. On the other side, a MEMS switch that could be embedded in a CMUT structure is
developed. The significance of the MEMS switch is that it could eliminate the high-voltage process
required to implement the front-end electronics that therefore could improve the performance of
the front-end electronics of the system. The third objective is to develop transparent CMUTs on
glass substrates and demonstrate its potential for applications where acoustics and optics are
combined.
7
The contributions of this dissertation are listed below.
1. This dissertation will demonstrate a novel CMUT fabrication method that is versatile to make
different types of CMUT arrays and enable novel applications.
2. A 16×16 CMUT 2D array on a glass substrate with through-glass-via (TGV) interconnects
was fabricated. A single via resistance plus the contact resistance is 2 Ω. The capacitance of a
250-µm pitch via pair was measured as 21 fF. A 2D array element was characterized in the
air.
3. A finite-element model was developed for a MEMS switch that could be embedded and co-
fabricated with CMUT on glass substrates using anodic bonding. The switching mechanism
and switching speed were studied using the model and verified by characterizing the actual
fabricated device.
4.
The MEMS switch was fabricated and characterized by static characterization in the air and
dynamic characterization under oil. The actual device performance agrees well with the FEM
model. A switching time of 1.34 µs and a releasing time of 80 ns were demonstrated. The
control voltage of the switch is 2.5 V.
5. A CMUT element with improved transparency was presented by using ITO as the bottom
electrode. The fabricated CMUT was functional and shows 30%-70% optical transmission in
600-800 nm wavelength range, which is suitable for photoacoustic imaging applications.
8
6. The CMUT with improved transparency was used for backward photoacoustic imaging. A
graphite target and a tube filled with ICG solution were used as two imaging targets. The effect
of silicon plate absorption was not significant (30 dB less than the graphite target signal at 830
nm wavelength).
7. An optically transparent CMUT is developed for integrating ultrasound with flat panel
displays. The process only requires two masks and the maximum processing temperature is
250ºC. The fabricated device has an optical transparency of 60%-80% in the full visible
wavelength range.
Most of the works related to this dissertation have been previously published or presented.
The publications related to Chapter 3 are:
[1] F. Y. Yamaner, X. Zhang, and Ö. Oralkan, "A three-mask process for fabricating vacuum-sealed capacitive
micromachined ultrasonic transducers using anodic bonding," IEEE Trans. Ultrason. Ferroelect., Freq. Contr.,
vol. 62, no. 5, pp. 972 -- 982, 2015.
[2] F. Y. Yamaner, X. Zhang, and Ö. Oralkan, "Fabrication of anodically bonded capacitive micromachined
ultrasonic transducers with vacuum-sealed cavities," Proc. IEEE Ultrason. Symp., 2014, pp. 604 -- 607.
[3] X. Zhang, F. Y. Yamaner, O. Adelegan, and Ö. Oralkan, "Design of high-frequency broadband CMUT
arrays," in Proc. IEEE Ultrason. Symp., 2015, pp. 1 -- 4.
9
The publications related to Chapter 4 are:
[1] X. Zhang, F. Y. Yamaner, and Ö. Oralkan, "Fabrication of Vacuum-Sealed Capacitive Micromachined
Ultrasonic Transducers With Through-Glass-Via Interconnects Using Anodic Bonding," J. Microelectromech.
Syst., vol. 26, no. 1, pp. 226 -- 234, 2017.
[2] X. Zhang, F. Y. Yamaner, and Ö. Oralkan, "Fabrication of capacitive micromachined ultrasonic
transducers with through-glass-via interconnects," Proc. IEEE Ultrason. Symp., 2015, pp. 1 -- 4.
The publications related to Chapter 5 are:
[1] X. Zhang, F. Y. Yamaner, and Ö. Oralkan, "A Fast-Switching (1.35-µs) Low-Control-Voltage (2.5-V) MEMS
T/R Switch Monolithically Integrated With a Capacitive Micromachined Ultrasonic Transducer (CMUT)," J.
Microelectromech. Syst. (Accepted)
[2] X. Zhang, A. Zeshan, O. J. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "A MEMS T/R switch embedded in
CMUT structure for ultrasound imaging frontends," Proc. IEEE Ultrason. Symp., pp. 1 -- 4, 2016.
The publications related to Chapter 6 are:
[1] X. Zhang, X. Wu, O. J. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "Backward-mode photoacoustic imaging
using illumination through a CMUT with improved transparency," IEEE Trans. Ultrason., Ferroelect., Freq.
Contr. DOI 10.1109/TUFFC.2017.2774283 (IEEE early access article available).
[2] X. Zhang, O. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "CMUTs on the glass with ITO bottom electrodes
for improved transparency," Proc. IEEE Ultrason. Symp., pp. 1 -- 4, 2016.
[3] X. Zhang, O. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "An Optically Transparent Air-Coupled
Capacitive Micromachined Ultrasonic Transducer (CMUT) Fabricated Using Adhesive Bonding" pp. 1 -- 4,
2017.
10
1.4 Thesis Organization
Chapter 2 introduces the background and modeling of CMUTs. Chapter 3 explains the
platform process, which requires only three masks and three photolithographic steps to fabricate
CMUT single transducers and 1D arrays on glass substrates using anodic bonding.
In Chapter 4, we incorporate through-glass-via (TGV) interconnects into the platform process
to make 2D CMUT arrays. The TGV parasitic resistance and capacitance were measured and an
element of a 16×16 2D CMUT array was characterized in the air.
Chapter 5 introduces a MEMS switch that can be embedded and co-fabricated in the CMUT
structure that is fabricated on glass substrates using anodic bonding. The MEMS switch is
characterized in air for static performance and characterized under oil for dynamic performance.
The results demonstrated that the co-fabricated MEMS switch is suitable for the front-end
electronics of a medical ultrasound imaging system.
In Chapter 6, we first demonstrate CMUT with improved transparency that could be used for
backward-mode photoacoustic imaging applications. Then we replace the thin silicon plate with a
glass plate. By using adhesive bonding technique, we demonstrate a 2-mask process to make fully
transparent air-coupled CMUT that could be integrated with a flat-panel display. Such applications
include a parametric array for generating uni-directional sound, fingerprint sensors, and ultrasound
gesture sensors, etc.
Chapter 7 is the summary of the dissertation.
11
Chapter 2 CAPACITIVE MICROMACHINED
ULTRASONIC TRANSDUCERS (CMUTs)
2.1 CMUT Structure and Operation Principle
The basic building block of a CMUT array is a capacitor cell consisting of a movable plate
suspended above a vacuum gap. An insulation layer is incorporated between the two electrodes to
prevent the two electrodes shorting in case of contact during the CMUT operation. The insulation
layer could be implemented either under the top electrode or on the bottom electrode. The rule of
thumb for CMUT design is to maximize the electric field strength in the cavity with a practical dc
bias voltage so that the CMUT could be operated with sufficient efficiency. Therefore it is desired
to have the top electrode at the bottom of the plate.
The substrate for fabricating CMUT can be a conductive substrate, such as a heavily doped
silicon substrate, or a dielectric substrate, such as a glass substrate. CMUT plate can be
implemented either using a conductive material, such as conductive silicon, or a dielectric material.
Figure 2-1: Different implementations of a CMUT cell.
A single CMUT element consists of many capacitor cells connected in parallel, and multiple
elements form an array. Arrays of different geometries can be fabricated on the same wafer
simultaneously (Figure 2-2).
12
Figure 2-2: CMUT arrays of different geometries fabricated on the same wafer. The four pictures
shows the CMUTs at wafer level, array level, element level, and cell level.
During CMUT operation, a dc voltage is applied to an element and the movable top plates are
attracted towards the bottom electrodes due to electrostatic force, which is resisted by the
mechanical restoring force due to the stiffness of the plate. The plate reaches a steady-state when
the two force balances. Driving the plate with an alternating voltage (ac) modulates the
electrostatic force and therefore generates ultrasound.
On the receive side, if the biased plate is subjected to ultrasound pressure, the stored electric
charge on the electrodes will change due to the capacitance change under the constant bias voltage.
(2.1)
dcdQdCVdtdt
Therefore, a current is produced with an amplitude that is a function of the bias voltage
,
the frequency of the incident wave
, the steady-state capacitance of the device
, and the
fractional displacement
due to the incident wave, which can be mathematically expressed as:
13
(2.2)
2.2 CMUT Performance Attributes
2.2.1 CMUT Modeling
In order to analyze the CMUT performance attributes, we need to understand the CMUT
modeling. There are mainly two approaches used to simulate the performance attributes of a
CMUT, which are the analytical electrical equivalent circuit model (ECM), and the finite element
model (FEM). In this chapter, we will review the equivalent circuit model. The finite-element
model will be discussed in Chapter 5.
Mason's equivalent circuit model is shown in Figure 2-3, with an inset showing the CMUT
parallel plate capacitor model (spring-mass-damper model) to derive some important parameters
of CMUT that is needed to understand the equivalent circuit model.
In this parallel plate capacitor model, the CMUT top plate is constrained so that it moves like
a piston, which forms a parallel plate capacitor with the fixed bottom electrode. The piston with a
mass
is suspended over an effective electrostatic gap
by a spring constant
, which derives
dcV00C0xx000acdcxIVCxm0gk 14
from the stiffness of the plate and results in the mechanical restoring force during the CMUT
operation. The damper with a damping factor
represents the mechanical loss during the piston
movement.
is the voltage applied between the top and bottom electrode to introduce electrostatic
force. It should be noted that this model deviates from the reality since CMUT plate is edge-
clamped with a maximum displacement in the center. In addition, the piston-like motion also
neglects the higher order vibration modes of the plate that are inherent in the actual devices.
Furthermore, the assumption of a linear spring differs from the actual restoring force especially
when the displacement is significant. Despite the limitations, the model can still provide a good
prediction of important CMUT parameters such as pull-in voltage, center frequency, fractional
bandwidth, etc.
The Mason's equivalent circuit model is a small-signal model and therefore is only suitable
for the analysis of the receiver, or when the ac signal is small compared to the dc bias during
transmit. The model can be considered as a two-port network composed of an electrical domain
and a mechanical domain. On the electrical port, the input of the circuit is a voltage source with
an internal resistance of
.
is the clamped device capacitance at the bias voltage.
is the
parasitic capacitance. On the mechanical port of the circuit, voltage corresponds to force and
current corresponds to the plate average velocity with a transformer ratio of
, which is also
related to the electromechanical coupling coefficient (
).
represents the mechanical
impedance of the CMUT plate, which could be modeled using the spring constant
and the mass
of the plate in the spring-mass-damper model.
is the radiation impedance of the
bVsR0CpCn2KtplateZkmmediumZ
surrounding medium. The spring softening behavior is modeled using a spring softening capacitor
15
.
The equation of motion is expressed as:
(2.3)
where
is the displacement of the plate, m is the mass of the piston,
is the damping coefficient,
is the spring constant, and
is the electrostatic force.
Figure 2-3: Equivalent circuit model of CMUT that relates voltage and current (V and I) to force
and velocity (F and U). (For TX, Fs=0; For RX, Vs=0)
02Cn220edxdxmbkxfdtdtxbkef
2.2.2 Center frequency (fc), fractional bandwidth (FBW), and quality
16
factor (Q)
2.2.2.1 Center Frequency
When the CMUT operates in air or vacuum, the center frequency of a mass-spring-damper
system is determined by the equivalent mass
and the equivalent spring constant
. Since the
medium damping is low it can be ignored in such case. The center frequency is equivalent to the
natural resonant frequency
of the plate and can be written as:
.
(2.4)
where
,
,and denote Young's modulus, density, and Poisson's ratio of the plate material,
respectively.
represents the radius of the plate and
is the plate thickness [33].
When the CMUT operates in immersion, the center frequency will also be influenced by the
complex damping factor
, the resonant frequency of the system becomes:
(2.5)
eqmeqkrf222.95(1)eqcrpeqkhEamEpahribbb222.95(1)10.67eqceqmphEkamah 17
where
is the density of the medium. The imaginary part of the loading impedance of the
medium contributes to the additional mass to the plate while the spring constant of the plate
remains the same. As a result, the center frequency shifts down. This equation is a more general
form of the center frequency, which indicates the center frequency shift down with the presence
of medium damping (
in the air and
in immersion).
2.2.2.2 Fractional Bandwidth (FBW) and Quality Factor (Q)
As seen in Figure 2-3, the second order spring-mass-damper model can be regarded as a
second-order RLC resonant network as an electrical equivalent. Therefore, one can identify the
lower 3-dB and higher 3-dB cutoff frequencies which can be written as the following two equations
respectively [34]:
(2.6)
(2.7)
where
is the quality factor of the system, which is proportional to the ratio
of the stored energy over dissipated energy at the operating frequency. The fractional bandwidth
can therefore be obtained as:
(2.8)
m112011(()1)22HQQ2011(()1)22HQQ()iLrkmbXQRb01()HLribFBWQkmb
The complex damping factor
is contributed by the mechanical loss within the CMUT structure
as well as the energy dissipated into the medium. The medium loss can be expressed as:
(2.9)
18
where
is the wave number. It can be seen that
needs to be large so that a wider FBW
could be obtained.
2.2.2.3 Discussion
The operating frequency and the bandwidth of the transducer are mainly determined by the
dimensions, the shape, and the mechanical properties of the thin plate. A wide bandwidth is desired
for better axial resolution in ultrasound imaging. One of the major advantages of CMUTs
compared to piezoelectric transducers is the wide bandwidth when operated in immersion.
Compared to a PZT transducer, the vibrating plate of a CMUT is substantially thinner and therefore
the stiffness or the spring constant of the plate is low. The transducer's plate mechanical impedance
is much smaller than the loading impedance over a large frequency range. As a result, a CMUT
experiences a much higher damping which leads to a wider bandwidth, which translates into less
ringing in the time domain signal and therefore leads to a higher axial resolution in ultrasound
imaging. The higher cutoff frequency is determined by the plate mass and the mass loading of the
immersion medium as well as the higher order resonant modes of the plate. The lower cutoff
b220221mediumKajKaZZKa2KKa 19
frequency is determined by the spring constant. As the spring constant increases, the bandwidth of
the transducer decreases.
2.2.3 Pull-in Voltage (
), Coupling Coefficient (
), T/R Sensitivity (
,
)
2.2.3.1 Pull-in Voltage (Vpull-in)
Pull-in of the plate occurs when electrostatic force overcomes the mechanical restoring force,
and the plate abruptly collapses down to the bottom electrode.
Before the plate pulls in, the equilibrium exists between the electrostatic force
and the
mechanical restoring force
:
(2.10)
where
,
,
is the piston electrode area, and
is the permittivity
of the free space. Then the pull-in voltage can be obtained when electrostatic force gradient
overcomes the gradient of the mechanical restoring force:
(2.11)
It can be seen that
is 2/3 of the effective electrostatic gap:
pullinV2TkTXSRXSeFmF0meFFF()meffFkgg2022dceAVFgA023030,()0pipipipipipigVFAVkgkVggApig
Therefore, the pull-in voltage can be expressed as:
20
(2.12)
(2.13)
2.2.3.2 Electromechanical Coupling Coefficient (
)
The electromechanical coupling coefficient is an important figure of merit of ultrasonic
transducers. The coupling coefficient is, by definition, the ratio of delivered mechanical energy to
the stored total energy in the transducer, as expressed by equation,
(2.14)
where
. For the parallel plate structure, the coupling coefficient is
(2.15)
Initially,
is zero and increases as the displacement increases. When the displacement
equals to one-third of the initial gap distance the electrostatic force gradient is larger than that of
mechanical force and the top plate collapses on the bottom electrode; at this point
is equal to
1 [3].
2200322223pipipipieffeffpieffpipiAVAVgkggggggkg30827effpikgVA2Tk211mechTelectotalmechEkEEEtotalmechelecEEE202Txkgx2Tk2Tk 21
It should be understood that the coupling coefficient has a direct impact on the device
efficiency for piezoelectric transducers. However, this is not the case for CMUTs. Although the
can be as high as 0.85 when CMUT is biased close to
[35], it is not desired in actual
CMUT as a transmitter because in such case there will not be enough gap for the plate to vibrate.
Typically, a CMUT device is biased approximately 80% of the
for the optimum balance
between output pressure and coupling coefficient.
2.2.3.3 Transmit Sensitivity (
)
During the transmission, the input is the electrical actuation voltage and the output is the
acoustic pressure delivered to the surrounding medium. The transmission efficiency (
), is
expressed as the ratio of the output pressure delivered to the medium to the electrical actuation
voltage. The stored electric energy inside the CMUT is:
(2.16)
where
is the device capacitance,
is the voltage difference between the top and bottom
electrodes,
is the area of the piston, and
is the permittivity of the medium between the
electrodes. Therefore, the electrostatic force can be represented as:
(2.17)
2TkpullinVpullinVTXSTxS2222()CVAVWdxCVA222()eWAVfxdx 22
where
. For the fundamental frequency,
is the only term
contributes to the output pressure. Therefore, at a given dc operation point, the acoustic output
pressure could be expressed as:
(2.18)
where
and
are the capacitance of the device and the electrical field strength
in the gap at the dc operating point, respectively. As a result, the transmission efficiency can be
represented as:
(2.19)
The above equation indicates that CMUTs with larger capacitance and stronger electrical field (i.e.
a smaller gap and larger dc bias) have better coupling efficiency between electrical and acoustic
domains during transmission.
The maximum output pressure at a frequency of
can be written as:
(2.20)
where x is the amplitude of the acoustic wave, which is equivalent to the CMUT plate
displacement, which is determined by the gap height and the dc bias. As a result, a larger electrode
222()2()dcacdcacVVVtVVt2()dcacVVt2()()()()()2()dcacacacAVVtptAFtCEVtnVtdxACdxdcVEdx()(/)()TxacPtnSPaVVtARe()Re()mmdxPZZxdt
gap is desired for higher output pressure, while a smaller gap is required for a better transmit
efficiency at a certain dc bias. Therefore the transmitting efficiency and the maximum output
23
pressure is a trade-off.
2.2.3.4 Receive Sensitivity (
):
On the receive side, the input of the CMUT is the acoustic pressure from the surrounding
medium and the output of the device is a current which resulted from the device capacitance change
caused by plate vibration under a constant dc bias voltage.
(2.21)
Therefore, a current is produced with an amplitude which is a function of the bias voltage,
, the frequency of the incident wave
, the steady-state capacitance of the device
, and the
fractional displacement
due to the incident wave, which can be mathematically expressed as:
(2.22)
By plugging in the
,
, and
into above equation, we can rewrite
the received current as:
(2.23)
RXSacdcQCIVttdcV00C0xx000acdcxIVCx0ACx0dcVExRe()mxPZtxICEt 24
Therefore, the receive sensitivity can be expressed as:
(2.24)
It is seen that both
and
are proportional to the electromechanical transformer ratio
.
We can also define the two-way sensitivity,
, with the unit A/V by the product of
and
:
(2.25)
We can derive the transfer function of the CMUT in TX mode by setting
and in RX
mode by setting the
. The voltage across
represents the force on the top plate. Assuming
the parasitics and
and
can be ignored, the magnitude of TX and RX transfer function can
be expressed as:
(2.26)
(2.27)
RXmmICEnASAPZZTXSRXSn/TXRXSTXSRXS2/TXRXmnSZ0F0VmRlossRmm12222021()(1)TXmmPnSVAR12222021()(1)RXmmmIAnSVRR 25
where
,
,
. It can be concluded that
the maximum sensitivity in TX and RX both occur at the center frequency
, where the TX,
RX, and two-way sensitivity can be expressed as:
2.2.3.5 Discussion
(2.28)
(2.29)
(2.30)
Overall, it is observed that the vacuum gap height design and control is one of the most critical
aspects to achieve a balance between transmit and receive performances. A larger gap is desired
for higher output pressure, while a smaller gap is desired for optimum sensitivity at a practical dc
bias voltage.
Parasitics can heavily affect CMUT efficiency. In the Mason's equivalent circuit model, the
parasitics derive from the following aspects. First, the parallel parasitic capacitance
is the most
dominant parasitic component because it sinks the current in the electrical port and reduces the
available current that could be converted into the mechanical domain. Essentially,
exists where
the actuation voltage is applied but there is no plate movement, which is where not electrically
active. It includes the overlapped post region, cable capacitance, top-to-bottom via capacitance
0kmpsoftkkk20softnkCpistonmediummmm0,maxTXnSA,maxRXmnASR2/,maxTXRXmnSRpCpC 26
when TSVs are incorporated, and
also exists because the plate average movement deviates
from the ideal piston movement. Having an insulating substrate is beneficial for reducing the
parallel capacitance resulting from the CMUT structure.
Another parasitic capacitance is the series parasitic capacitance
, which comes from the
insulating layer embedded inside the CMUT structure. A high-k dielectric insulation layer and a
thinner dielectric layer thickness are desired to reduce the effect of the series parasitic capacitance.
2.3 CMUT Fabrication Process
2.3.1 CMUT on Silicon Substrates
2.3.1.1 Sacrificial-Release Process
CMUT technology was invented on silicon substrates and extensive research has been
conducted to improve the device performance and reliability in terms of the device structure design
and fabrication process flow.
Surface micromachining with sacrificial release process is the first technology used to
fabricate CMUTs. The basic principle of sacrificial release process is to first deposit a sacrificial
layer and then selectively remove it using an appropriate etchant after depositing the plate layer
on top. An extension of this process includes through-silicon via interconnects from the front side
of the substrate to the backside which can be used for making 2D CMUT arrays. A general method
for making CMUTs through surface micromachining method, or sacrificial releasing method is
pCsC 27
described as following. (a) First, an insulation is deposited; it also acts as an etch stop layer. (b)
Next, a sacrificial layer is deposited. (c) Channels are etched define etch channels. (d) Membrane
material is deposited. (e) Release holes are etched. (f) Membrane is released by wet-etching of the
sacrificial layer. (g) Etch holes are filled. (h) Electrodes are deposited.
There are several disadvantages associated with this process. First of all, the fabrication
process is relatively complex. Second, the deposited layers have poor uniformity and could
introduce stress. Finally, it introduces additional parasitics because the silicon substrate is a
conductive material and requires additional insulation steps.
2.3.1.2 Wafer Bonding Process
The second method, typically referred to as wafer bonding fabrication, involves fabricating
the cavities/bottom electrode on one wafer and the membrane/top electrode on a second wafer
[3.6]. The wafers are then bonded together with high force and elevated temperature. The
fabrication process is described as following. (a) First, an insulation layer is grown on a highly
doped substrate. (b) Cavities are then etched on the insulation layer. (c) A separate silicon-on-
insulator (SOI) wafer is bonded together with the (b) wafer. (d) An etch-back process is done to
release the substrate and buried-oxide (BOX) layer of the SOI wafer. (e) A two-step etching
process is done to expose the bottom substrate. (f) Metal electrodes are deposited. Because the
cavities and membranes can be fabricated separately, it decouples design constraints so each
structure can be optimized individually. With wafer bonding, sacrificial layers, and thus holes, are
not required hence the fill-factor is improved. Also, without the constraints of releasing, cavity
28
depths can be made shallower which further improves electrical stability as actuation voltages can
be reduced and insulating thickness can be increased instead. Furthermore, fabrication
repeatability and controllability is generally better with the wafer bonding process as the number
of lithography steps and masks are typically fewer.
Wafer bonding method is demonstrated to overcome the above shortcomings. Wafer bonding
process begins with two wafers: a substrate wafer and a silicon-on-insulator (SOI) wafer. For
silicon substrates, the cavities are defined on the substrate wafer. Then the SOI wafer and the
substrate are brought together. Wafer bonding has also been implemented using eutectic bonding
and more recently adhesive bonding [12] to make CMUTs on silicon substrates.
2.3.1.3 Insulation on Silicon Substrate
In CMUT operation, it is desired to have the electric field applied only where it is required,
which is the vacuum gap. However, the commonly used silicon substrate is a semiconductor and
therefore result in the parallel parasitic capacitance. Several strategies have been investigated to
address the problem on the silicon substrate.
One way is to use an extended insulation layer structure in the post area to address the low
breakdown voltage and high parasitic capacitance using a LOCOS process [36][37]. However, this
process is not suitable for 2D arrays.
29
Figure 2-4: CMUT with an extended insulation layer.
Another structure is to isolate the CMUT bottom electrode only to the region under the gap
where a high electric field is desired so that the possibility of dielectric breakdown and parasitic
capacitance in the post region can be minimized [38][37]. This structure also includes TSV
interconnects for accessing each element from the backside. The key component in this approach
is an SOI wafer with a thick buried oxide layer to completely insulate silicon bottom electrodes
below the active plate region in each CMUT cell. The hot electrode is provided through an opening
in the thick buried oxide layer filled with TSV.
Figure 2-5: CMUT with a thick buried oxide layer.
30
2.3.2 CMUT on Insulating Substrates
Building MEMS devices on insulating substrates have raised tremendous research interest
[39] -- [41]. Using an electrically insulating substrate to fabricate CMUTs can reduce the device
parasitics and simplify the fabrication flow by eliminating the complicated isolation steps.
Furthermore, insulating substrates such as glass, fused silica, quartz, can provide excellent optical
transparency and therefore can enable novel applications where optics and acoustics are combined.
Fabrication of CMUT on insulating substrates is relatively new and is attracting more and more
attention.
It is desired to fabricate CMUTs on insulating substrates using the wafer bonding technique,
although it has also been shown that sacrificial release process could be employed and result in a
reduced parasitic capacitance [9]. Anodic bonding comes across as an attractive method to
combine the benefit of wafer bonding and an insulating substrate. The bottom electrode could be
patterned and deposited in the etched glass cavities and a silicon or SOI wafer could be anodically
bonded as the plate to realize a CMUT structure. However, one challenge with anodic bonding is
the outgassing during bonding. In the previously demonstrated CMUTs on glass substrates with
anodic bonding, either the cavities were pressurized with trapped oxygen gas under a thick plate
[42], or the cavities were exposed to outside for gas evacuation, thus making the transducer not
suitable for immersion operation [10]. This question needs to be addressed to fully unleash the
potential to make CMUTs on glass substrates using anodic bonding. Apart from anodic bonding,
adhesive bonding (with SU8, BCB) or eutectic bonding can also be combined with insulating
substrates.
Below is a summary of the CMUT fabrication technologies categorized by the substrate type.
The topics in the black box will be discussed in this dissertation.
31
Figure 2-6： CMUT fabrication technologies categorized by substrate types.
2.4 CMUT Integration with Electronics
Close integration of CMUT and the electronics are highly desired, especially for transducer
arrays with small elements, thus small capacitance, such as 2D arrays and the arrays designed for
use at the end of a catheter. In the current conventional ultrasonic imaging systems, the array is
located in a hand-held probe, which is connected to the main processing unit via a cable bundle.
Transmit and receive electronics are located in the main processing unit. In order to avoid the
additional parasitic capacitance introduced by the cable degrading the signal quality, the receiver
electronics must be close to the transducer array. Furthermore, multiplexing and beamforming
circuitry can also be integrated with the array to minimize the number of active electronic channels
and the number of physical connections between the probe and the backend system.
For integrating CMUT arrays with electronic circuits, mainly two categories of methods
32
have been adapted.
2.4.1 Monolithic Integration
The first approach is monolithic integration, which is to build CMUTs and CMOS
concurrently (co-processing), or build the CMUTs on top of the finished electronic wafer (post-
processing).
2.4.1.1 Co-processing
One method to monolithically integrate CMUTs and electronics is to fabricate both
components concurrently using a standard or minimally modified process. A BiCMOS process
using 16 masks has been used with only minor modifications including an additional
photolithography step and sacrificial layer etching to fabricate CMUTs side-by-side with
electronic circuits on the same substrate [43]. Although this method offers a cost-effective solution
for electronics integration, it suffers from two major limitations: (1) the transducer area is shared
by electronic circuits or interconnections. (2) The device layer thickness are limited by the process
used for electronic circuits.
33
2.4.1.2 Post-process (CMUT-on-CMOS)
The second monolithic integration technique is to fabricate the electronics first using a
standard foundry process and then build the CMUT on top of the finished electronics. This post-
processing is augmenting a standard foundry process only with two blanket post-process steps for
sacrificial etching and cavity sealing [44]. This approach has good area utilization and more control
over device dimensions, but it still suffers from the limitations on the device dimensions in the
vertical direction due to the layers available in the standard foundry process.
Low-temperature wafer bonding can also be used in post-process. It brings the advantages
of wafer bonding process such as control over the plate thickness, process simplicity, and provides
a monolithic integration solution without going through the complexity of sacrificial release-
process [45]. In this low-temperature bonding process, a thin titanium layer is used for electrical
via-contact to CMUTs substrate as well as an adhesion layer for wafer bonding. The gap height is
set by the total thickness of the titanium adhesion layer and the passivation layer on the CMOS
wafer.
2.4.2 Hybrid Integration
2.4.2.1 Chip-to-chip bonding
The fabrication processes for both CMUT and CMOS could be optimized if the two
components can be fabricated on separate substrates. In this case, CMUTs need through-wafer-
interconnections from the front side of the substrate to the backside. The CMUT chip can then be
34
directly bonded on the electronics using the well-established flip-chip bonding process [46]. Direct
bonding requires that the electronic die area be greater than the CMUT die area so that the
peripheral pads on the electronic die will be accessible to provide connections between the front-
end circuits and the backend system.
2.4.2.2 Bonding Through Intermediate Substrates
Using an intermediate substrate will make the size of the CMUT and the IC be independent
and it is desirable in the following cases: (1) To implement a very large CMUT array, electronics
or both electronics and CMUT array should be implemented by tiling several unit blocks. This
approach also helps improve the overall yield one can achieve [47]. (2) Using a flexible
intermediate substrate, the overall form factor of an integrated ultrasound probe can be minimized.
This approach has been used to demonstrate a 64-element CMUT ring array for forward-looking
catheter-based intracardiac imaging [48].
2.5 Summary
In this chapter, we reviewed the CMUT basics and the device modeling (parallel plate
capacitor model and equivalent circuit model). The critical performance attributes are derived and
the design trade-offs are discussed. CMUTs have a wider bandwidth immersion derived from the
thin plate and low plate mechanical impedance. Therefore it holds great potential for ultrasound
medical imaging as a complement to piezoelectric transducers. A good device design and
minimized parasitics are required for optimized CMUT device efficiency.
35
The CMUT fabrication process and integration methods are also discussed. We categorized
the CMUT fabrication technology based on substrate types. CMUTs are traditionally fabricated
on silicon substrates. However, fabricating CMUT on insulating substrates can reduce device
parasitics and simplify the fabrication process flow. Furthermore, insulating substrates such as
glass can offer the benefit of optical transparency and help innovate for new applications.
36
Chapter 3 PLATFORM TECHNOLOGY:
VACUUM-SEALED CMUTS FABRICATED ON
GLASS SUBSTRATE USING ANODIC BONDING
3.1
Introduction
CMUT technology has demonstrated great promise for next-generation ultrasound
applications. Wafer-bonding technology has largely simplified the fabrication of CMUTs by
eliminating the requirement for a sacrificial layer and increases control over device parameters.
Fabrication of CMUTs on glass substrates using anodic bonding has many advantages over other
bonding methods, such as low-temperature compatibility, high bond strength, high tolerance to
particle contamination and surface roughness, and cost savings. Furthermore, the glass substrates
lower the parasitic capacitance and improve reliability. The major drawback is the trapped gas
inside the cavities, which occurs during bonding. Earlier CMUT fabrication efforts using anodic
bonding failed to demonstrate a vacuum-sealed cavity.
In this Chapter, we developed a fabrication scheme to overcome this issue and demonstrated
vacuum-backed CMUTs using anodic bonding. This new approach also simplifies the overall
fabrication process for CMUTs. We demonstrated a CMUT fabrication process with three
lithography steps. A vibrating plate is formed by bonding the device layer of a silicon-on-insulator
(SOI) wafer on top of submicron cavities defined on a borosilicate glass wafer. The cavities and
37
the bottom electrodes are created on the borosilicate glass wafer with a single lithography step.
The recessed bottom metal layer over the glass surface allows bonding the plate directly on glass
posts and therefore helps reduce the parasitic capacitance and improve the breakdown reliability.
3.2 Fabrication Process Development
The realized CMUT structure consists of a thin single-crystal silicon plate with a thin silicon
nitride insulation layer and a patterned metal bottom electrode deposited inside a vacuum-sealed,
sub-micrometer cavity (Figure 3-1). The vibrating plate is formed by depositing a layer of silicon
nitride on the device layer of an SOI wafer before bonding. The silicon nitride layer is mainly to
prevent electrical shorting when the plate comes in contact with the metal bottom electrode
following pull-in and also acts as an intermediate bonding layer. Anodic bonding of borosilicate
glass to thin-film coated silicon wafers was demonstrated previously [49] -- [51]. The device layer
is chosen to be highly doped conductive silicon as it is used as the top electrode. The bottom
electrode is deposited on the surface of the cavities so that the overall parasitic capacitance of the
device is reduced and the dielectric reliability is improved as the post region in this approach does
not experience any significant electric field. During the formation of bond pads to provide
electrical access to bottom electrodes of individual transducer elements, a metal layer is also
deposited on top of the silicon plate to further increase the conductivity of the top electrode and to
provide a suitable layer for wire bonding.
38
Figure 3-1： 3D cross-sectional model of a completed CMUT cell.
The initial substrate was a standard 0.7-mm-thick, 100-mm-diameter borosilicate glass wafer
(Borofloat33, Schott AG, Jena, Germany) that has a high surface quality with an RMS roughness
(Rq) of 0.7 nm and a good flatness with a warp that is less than 0.05%. The thermal expansion
coefficient of the borosilicate glass substrate is 3.25 ppm/°C, close to that of silicon (3.2 ppm/°C)
preventing stress in the silicon plate after anodic bonding. The SOI wafer that we used for
fabrication has a 2 ± 0.5-µm-thick, n-type device layer with 0.001 to 0.005 Ω·cm resistivity, a 0.5-
µm-thick BOX layer, and a 500-µm-thick handle wafer with 1 to 10 Ω·cm resistivity. Before the
process, the borosilicate glass substrate was cleaned for 15 minutes in a Piranha solution for
removal of organics and other gross particle contaminants from the surface. The cavity pattern was
defined using 2-µm-thick negative photoresist (AZ-5214E IR, Clariant, Wiesbaden, Germany),
which is suitable for lift-off [Figure 3-2 (a)]. The patterned wafer was hard-baked for 2 h at an
elevated temperature of 125°C [Figure 3-2 (b)]. This step promotes the adhesion between the
photoresist and the substrate and makes the photoresist a better mask for the etching. The cavities
39
were created in 10:1 buffered oxide etch (BOE) solution [Figure 3-2 (c)]. Wet etching was
preferred for a uniform etching and minimal surface roughness in the cavities. We measured the
BOE lateral etch as 10 times faster than its vertical etch. It has been reported that a water-rich
interface layer between the wafer and resist causes the etchant to penetrate very fast laterally [21].
As a result, faster lateral than vertical etching is often seen in isotropic etching. This lateral etch
must be considered in the mask design to achieve the target cavity size after etching. 230-nm-deep
cavities were etched in 15-min total time with 5 cycles of BOE etching of 3 min each. The
photoresist was hard-baked for 10 min between each cycle to prevent peeling. After the cavity
etching was completed, the wafer was transferred to the evaporation chamber without removing
the resist. The gap height of the CMUTs was defined by the difference of the etch depth and the
thickness of the metal deposited in the cavities. Thus a metal stack that consists of 20 nm of
chromium as an adhesion layer and 90 nm of gold was deposited to obtain the 120-nm gap height
[Figure 3-2 (d)]. The undercut that was formed during the wet etch helps to confine the metal
electrode to the bottom surface of the cavity and also makes the lift-off process easier.
Prior to bonding, we deposited 200-nm silicon nitride on top of the device layer of the SOI
wafer by using plasma-enhanced, chemical-vapor deposition (PECVD) at 1000-mTorr chamber
pressure and 350°C temperature. This silicon nitride layer serves as an insulation layer between
the conductive silicon plate (top electrode) and the metal in the cavity (bottom electrode) during
device operation. The glass wafer and the SOI wafers were cleaned using solvents and Piranha
solution, respectively. The borosilicate glass surface and the nitride surface were anodically
bonded together at 350°C under 2.5-kN down force in vacuum (10−4 Torr) in a semi-automatic
40
bonding system [model EVG510, EVG Group, St. Florian, Austria; Figure 3-2 (e)]. Typically, it
is recommended to limit the current during the bonding [22]. The setup that we use does not have
a current limited bonding option. Thus, the voltage was ramped up to its final value at a rate of 20
V/min not to cause a breakdown in the silicon nitride layer and kept at the target value for 30 min.
The metal is exposed to a high electrical field during bonding, thus various bonding voltages were
evaluated to maintain high bond yield without damaging the floating bottom electrode inside the
cavities. The bonding was tested at 1000, 700, 600, and 500 V. A high bonding yield with no
damage on the bottom electrodes was observed at 600 V for the presented process and wafer
parameters. After bonding, the handle wafer was ground down to 100 µm. We used a heated
tetramethylammonium hydroxide (TMAH) solution (10% TMAH at 80°C) to selectively remove
the remaining handle wafer over the BOX layer, which was subsequently removed using 10:1 BOE
solution [Figure 3-2 (f)].
The borosilicate glass substrate is exposed to a high electrostatic field, which causes
outgassing during bonding [52]. We proposed to evacuate the gas inside the cavities and seal them
in vacuum. To access the bottom electrode for forming bond pads, the plate at the pad location has
to be etched. When the plate over the metal pad region is etched, the channel over the metal surface
is exposed and allows the gas to escape [Figure 3-2 (g)]. We used reactive ion etching with SF6
gas to etch silicon. Different arrays implemented on the same wafer are also separated during this
step by etching the conductive silicon between different arrays. After evacuating the trapped gas
in the cavities, oxygen plasma is used to remove the photoresist. Because there is no wet cleaning
at this step, no liquid reaches inside the cavities. Avoiding wet processing is important at this step
41
as it can lead to stiction and consequently collapsed cells in the drying stage. To seal the cavities
we deposited PECVD silicon nitride [Figure 3-2 (h)]. The thickness of the silicon nitride was
chosen to be more than three times the cavity height for a proper sealing [9].
After the sealing step, the wafer surface is completely covered by silicon nitride. To create
electrical contacts, the silicon nitride layer on the bond pads has to be removed. At this point, the
silicon nitride deposited on the conductive silicon plate is also removed leaving the silicon nitride
only on the locations where sealing is required. For etching the silicon nitride we used reactive ion
etching where the AZ5214E IR photoresist was used as a mask. The photoresist was hard-baked
for 5 min at 125°C before etching. After removing the nitride layer on the pads and on the silicon
plates [Figure 3-2 (i)], 20-nm chromium and 130-nm gold were deposited. The chromium-gold
metal stack was then lifted off in N-methyl-2-pyrrolidone (NMP) solvent [Figure 3-2 (j)]. At this
step, the device fabrication is completed.
42
(b)
(d)
(f)
(h)
(j)
(a)
(c)
(e)
(g)
(i)
Figure 3-2: Fabrication process flow
43
Figure 3-3: SEM cross-sectional image of a completed CMUT cell.
3.3 Characterization
3.3.1 Characterization in Air
The success of the sealing process was confirmed by measuring the deflection profile of the
plate after the completion of the entire process (Figure 3-4). The upward plate deflection due to
gas trapped inside an array of cells of design #1 was measured using an optical surface profilometer
(model NewView 5000, Zygo Corporation, Middlefield, CT, USA). After evacuating the gas and
sealing the channels, the deflection under atmospheric pressure was measured as 28 nm. Finite
element analysis (FEA; ANSYS v.14, ANSYS Inc., Canonsburg, PA, USA) for this cell predicts
the atmospheric deflection as 26.8 nm, which also proves that there is no significant stress due to
anodic bonding and the cavity is under vacuum.
44
Figure 3-4: Zygo interferometer measurement. Before gas evacuation (left). After gas evacuation
and re-seal (right).
The device was tested in air using a network analyzer (model E5061B, Agilent Technologies
Inc., Santa Clara, CA, USA) with an internal dc supply available up to 40 V. The measured real
and imaginary parts of electrical input impedance in air are shown for bias voltages of 10, 15, and
20 V in Figure 3-5. The baseline in the real part corresponds to the series resistance of the device,
which is measured as 21 Ω. To measure the collapse voltage, the resonant frequency is observed
while increasing the bias voltage by 1-V steps. The collapse voltage is determined by the sudden
resonant frequency jump at the collapse.
The collapse voltages and the resonant frequencies of these three designs were also simulated
using FEA. TRANS 126, electromechanical transducer elements are used for the direct coupling
of electrostatic and structural domains. The element is capable of handling the spring softening
effect in the simulations. First, the static analysis was performed to find the collapse voltages.
45
Second, the prestressed harmonic analysis was carried out to find the resonant frequency at the
70% of the collapse voltage. The results show that the fabricated CMUTs operate as predicted by
the finite element model. We have measured the electromechanical coupling coefficient (kT
2) as
0.1 at 15-V dc bias (75% of the collapse voltage) and 0.3 at 20-V dc bias (90% of the collapse
voltage), which are consistent with results reported earlier [35]. The single cell capacitance of the
same design under atmospheric pressure is calculated as 0.23 pF by using FEA, which corresponds
to a total capacitance of 64.17 pF when multiplied by the total number of cells. We measured the
total capacitance of the device as 67.84 pF, indicating that the external parasitic capacitance is less
than 6%.
Figure 3-5: Electrical input impedance measurement. Real part (left). Imaginary part (right).
3.3.2 Characterization in Immersion
For immersion tests, vegetable oil was used because the transducer surface was not electrically
insulated. A small tank was built over the transducer element that was wire-bonded to a chip
46
carrier. A calibrated hydrophone (model HGL-0200, Onda Corporation, Sunnyvale, CA, USA)
connected to a preamplifier (model AH-2010, Onda Corporation) was placed at 14-mm distance
from the transducer surface on the central axis of the transducer. A 20-V unipolar pulse was
superimposed on the dc bias voltage through a bias-T circuit. The conductive plate layer was
grounded and the bottom electrode was used as the active electrode. A 12-V dc voltage was applied
and the element was driven by a 110-ns wide pulse. The signal received by the hydrophone is
shown in Figure 3-6. The notches at 3.8 MHz and its higher harmonics correspond to the ringing
in the substrate [53] as it was also evident in the time domain data as a tail following the main
pulse. Substrate ringing can be pushed out of the frequency band of interest by choosing the
substrate thickness accordingly.
(a) (b)
Figure 3-6: Immersion measurement using a calibrated hydrophone. (a) Received time-domain
signal. (b) Fourier transform.
47
3.3.3 Array Characterization
We checked the resonant frequency of each individual element in an imaging array that we
fabricated (Figure 3-7). The results show that the standard deviation is 0.11 MHz with a mean
value of 12.5 MHz in resonant frequency for a 66-element array.
(a) (b)
Figure 3-7: (a) A 66-element 1D CMUT array fabricated on a glass substrate using anodic bonding.
(b) Array uniformity measurement.
3.4 High-Frequency Broadband 1D CMUT array
As a special design for the platform process, we demonstrate a high-frequency (29-MHz)
broadband (100% FBW) CMUT 1D array. The devices are fabricated using anodic bonding with
only three photolithography steps. We also discuss the design guidelines for high-frequency
broadband CMUTs using the simulations. A high fill factor and a thin plate are important for the
broadband design. Small cell size is required for the increased center frequency. To improve the
48
transducer sensitivity and to keep the collapse voltage low, the gap height should be small and a
high-k dielectric insulation layer should be employed. The fabrication steps we report in this paper
have good potential to meet the high-frequency broadband CMUT design requirements. So far we
have demonstrated that we can define a 50-nm gap, bond to a post as narrow as 2 µm, and pattern
a high-k dielectric layer on the bottom electrode.
(a) (b)
Figure 3-8: Gap height can be control as small as 50 nm through the fabrication process. (a)
Different regions measured on the wafer. (b) AFM image showing 50-nm gap before bonding (top).
SEM image showing 50-nm gap after bonding (bottom).
The immersion test was performed in a tank filled with vegetable oil. A 1D array was diced
and wire bonded to a chip carrier. The frequency response of the element in transmitting was
49
measured using a hydrophone (Model HGL0200, Onda Corporation, Sunnyvale, CA) at a distance
of 1.8 mm. The element was biased at 50-V DC and excited using a pulser/receiver (Model
5073PR, Olympus Corporation, Waltham, MA. Pulse repetition frequency (PRF): 200 Hz; Energy
level: 2; Damping level: 1). The signal received by the hydrophone is shown in Figure 3-9(a) in
the time domain. The corresponding frequency spectrum is shown in Figure 3-9(b). The frequency
spectrum shows the transducer center frequency is 29 MHz and the 3-dB FBW is 100% after
correcting for the pulse spectrum. The hydrophone calibration is available up to 40 MHz at the
time. Therefore the spectrum is not corrected for the hydrophone response.
(a) (b)
Figure 3-9: (a) Experimental received signal by the hydrophone. (b) Corresponding frequency
spectrum.
50
3.5 Chapter Conclusions and Future Work
We have presented a fabrication process for CMUTs based on anodic bonding [54] [11]. This
process offers the well-known advantages of wafer bonding such as good control over the
thickness and mechanical properties of the plate and overall reduced process complexity. In
addition to these general advantages of wafer bonding process, anodic bonding has the specific
advantage of being more tolerant to roughness on the bonding surface. Furthermore, the narrower
post structures are feasible with anodic bonding to maximize the fill factor, which is especially
critical to achieve wide bandwidth at high frequencies. The maximum processing temperature in
the presented approach is 350°C, which allows a patterned metal bottom electrode for reducing
device series resistance. Use of the glass substrate helps reduce the parasitic capacitance and
improves the dielectric reliability as the top plate and the bottom electrode mainly overlap on the
active transducer area and not on the posts. Using the same process, we further demonstrated a
high-frequency broadband 1D CMUT array using only three photolithography steps in the
fabrication [55]. The fabricated CMUT was characterized in immersion. The CMUT shows 29-
MHz center frequency and 100% FBW after correcting for pulse shape.
Chapter 4 2D CMUT ARRAYS WITH TGV
51
INTERCONNECTS
4.1
Introduction
Close integration of ultrasonic transducer arrays and front-end integrated circuits is critical
for the overall ultrasound system efficiency and also a compact form factor. For 2D arrays and
arrays used in ultrasound imaging catheters where the element area is small, the receiver
electronics must be closely integrated with the transducer array to avoid additional parasitic
capacitance introduced by the cables to preserve signal quality. CMUT technology has attracted a
great deal of attention because of advantages such as ease of fabricating arrays and integrating
them with supporting electronics, as well as wide bandwidth [14]. Two main methods have been
developed for integrating CMUTs with supporting circuits: monolithic integration and hybrid
integration [37]. For monolithic integration, one method is to co-process the CMUTs and the
electronics side by side [56]. In this method, not only the substrate area is shared by CMUTs and
electronic circuits, but also the properties and vertical dimensions of the layers used in the CMUT
structure are limited by the CMOS process materials and film thicknesses. The other method is to
fabricate the electronic circuit first and then build the CMUTs on top by post-processing [19], [44],
[57], [58]. This method has a good area utilization and more dimensional control, but the CMUT
process is still limited because of the temperature constraints set by the existing metal lines. The
complexity of the overall process also increases.
52
4.2 Through-Wafer Interconnects
4.2.1 2D CMUT with TSV interconnects
Hybrid integration allows the optimization of CMUTs and electronics independently. In
hybrid integration, through-wafer interconnects are established as part of CMUT arrays to enable
a close connection with supporting electronics by chip-to-chip bonding or through an intermediate
layer such as an interposer or a flex circuit [15], [59]. The current implementation of this hybrid
integration approach is based on using through-silicon-via [46]. Since silicon is a semiconductor,
a reverse biased PN junction or a metal-insulator-semiconductor (MIS) structure has to be used to
isolate individual connections to each element in an array and to reduce the parasitic capacitance
[60]. The implementation of an isolation structure complicates the fabrication process. Using
polysilicon as a via and electrode material degrades the surface quality and wafer flatness, affecting
subsequent processing steps. Last, TSVs can only be used for surface micromachining because the
complicated process of making TSV degrades the wafer surface quality and make it not suitable
for wafer bonding.
4.2.2 2D CMUT with Trench-Isolated Interconnects
An alternative interconnection method used for CMUTs is to create deep isolation trenches in
a highly conducting silicon substrate, where the resulting pillar underneath each element serves as
the interconnect from the front to the backside of the wafer [6]. This process is also complicated
53
and creates reliability problems due to the exposed gaps between top and bottom electrodes of the
CMUT array elements on the top surface.
4.2.3 2D CMUT with Through-Glass-Via Interconnects
Through-glass-vias (TGVs) have been used for advanced electronic packaging, especially for
RF applications where parasitics are critical [61]. Anodic bonding is widely used for wafer-level
hermetic MEMS packaging [62]. In recent years, fabrication of CMUTs on a glass substrate has
aroused significant interest [10], [63], [64]. We have recently reported a process for fabricating
vacuum-sealed CMUTs on a borosilicate glass substrate using anodic bonding [11]. This process
benefits from all the advantages of wafer bonding, including process simplicity, control over plate
thickness and properties, high fill factor, and ability to implement large vibrating cells.
Additionally, it reduces the parasitic capacitance and series resistance benefiting from an insulating
substrate and a metal bottom electrode, respectively.
The targeted implementation is illustrated in Figure 4-1, where the CMUTs are fabricated on
a TGV substrate and directly flip-chip bonded to an integrated circuit (IC). The vibrating plate
consists of a thin single-crystal silicon layer embedded between a silicon nitride insulation layer
at the bottom and a metal electrode on top. The bottom electrodes are formed in the etched glass
cavities and connected to the dedicated TGV interconnects giving each element electrical access
from the backside. The metal layer deposited on top of the silicon plate is common to all elements
in an array and is connected to a TGV for backside access. A stacked metal layer is deposited and
patterned on the backside to form the bond pads.
54
Figure 4-1: A schematic cross-section of a completed CMUT element with TGV interconnects flip-
chip bonded on an IC.
4.3 Fabrication of 2D CMUT with TGV Interconnect
4.3.1 Formation of TGV Substrate
Five masks are used for the fabrication process presented in this section. The via on the left in
the cross-sectional drawings and the via in the back in the 3D drawings are for the top electrode
connection. The via on the right in the cross-sectional drawings and the via on the front in the 3D
drawings are for the bottom electrode connection.
We use the same starting substrate and SOI wafer as in the platform process described in the
last chapter. The vias are designed on the 100-mm wafer area for the both bottom and top electrode
connections for different array geometries. The alignment marks are also formed by TGVs for
aligning the subsequent layers to the via pattern. Through-wafer channels are created first in the
0.7-mm borosilicate glass substrate at the designed via locations by laser drilling (Figure 4-3). The
through channels are then metalized by using the conductive copper paste technology (Triton
Microtechnologies, Oro Valley, AZ) [65] (Figure 4-4). A benefit of this approach is that the paste
55
has a thermal coefficient of expansion (TCE) matched to that of the glass substrate. Therefore,
potential mechanical stress that can be caused by further heating steps and possible cracks around
the vias can be minimized. Then, the wafer is sintered and polished. By selecting the appropriate
coating material on the substrate before polishing and using a slurry loaded with composite
particles as abrasive during the chemical-mechanical polishing (CMP) process, a smooth glass
surface and a good copper-to-glass surface co-planarity are obtained [66], [67]. For a completed
TGV wafer, the via location is in the range of ±5 µm from the designed location in both X and Y
directions, which is the accuracy of the laser drilling. The diameter of the via is 70 µm at the laser
entry side and 50 µm at the laser exit side. We build the CMUTs on the laseFr exit side and use
the laser entry side for the backside pad formation. With 33207 vias formed on the 100-mm wafer
area, the average wafer warp increased from ∼10 µm to ∼50 µm. Figure 4-5 shows the SEM cross-
sectional image of a TGV embedded in the glass substrate and AFM image of the via surface on
the laser exit side. At the via-glass boundary on the surface, copper and glass are at the same level
in most regions around each via, which makes a reliable electrical connection between the CMUT
electrodes and the TGVs possible.
Figure 4-2: Initial borosilicate substrate.
Figure 4-3: Define patterned through-wafer holes by laser drilling.
Figure 4-4: Fill the through holes with copper paste.
56
Figure 4-5: SEM cross-section of the TGV embedded in the glass substrate (left). AFM image of the
TGV surface profile (right).
57
4.3.2 Forming Cavities and Bottom Electrodes Connected to the TGVs
The TGV substrate was cleaned using a heated NMP@70◦C solution to remove organics and
contaminants from the surface. Acid-based cleaning solutions cannot be used because of the
copper in the via. After cleaning, the cavities were patterned using a negative photoresist with the
alignment of each element to the dedicated TGV. We used RIE to etch glass instead of buffered
oxide etchant (BOE) in order to avoid peeling of the photoresist and not to damage copper vias.
Dry etching also helps achieve a deeper cavity, which is desired for a thicker bottom electrode to
build a reliable connection to the TGV. We performed RIE with SF6 gas to realize a glass cavity
depth of 320 nm (Figure 4-6). After removing the photoresist, the bottom electrode was patterned
by a second photolithography step using a 2-µm-thick negative photoresist (AZ-5214E IR,
Clariant, Wiesbaden, Germany), which is suitable for lift-off. Prior to the bottom metal deposition,
we wet etched the copper vias also 320 nm using a copper etchant (Copper Etchant 49-1, Transene
Company, Inc., Danvers, MA) in order to keep the glass surface at the bottom of the cavity level
with the copper via surface. This selective wet etchant does not attack the photoresist and helps
remove the copper oxide on the via surface and consequently improve the electrical connectivity
between the bottom electrode and the TGV. Then a stacked metal that consists of 20-nm chromium
as an adhesion layer and 130-nm gold was deposited into the cavity by evaporation and defined by
lift-off as the bottom electrode to obtain a 170-nm gap height and to build the electrical connection
from the TGV to the bottom electrode (Figure 4-7). Figure 4-8 shows the top view of the processed
substrate that is ready for anodic bonding.
Figure 4-6: Glass and copper etch to define CMUT cavities.
58
Figure 4-7: Evaporation and lift-off the bottom electrode and forming an electrical connection to
the dedicated TGV.
Figure 4-8: SEM images of the completed glass substrate with TGVs that is ready for anodic
bonding.
59
4.3.3 Anodic Bonding
A 200-nm low-stress silicon nitride insulation layer was deposited on the device layer of the
SOI wafer by plasma-enhanced chemical vapor deposition (PECVD) prior to bonding. The TGV
wafer was cleaned with a solvent-based solution and the SOI wafer was cleaned using Piranha
solution. The borosilicate glass surface and the nitride insulation layer surface were brought
together in vacuum (10−4 Torr) and then bonded at 350◦C under 2.5-kN down force in a semi-
automatic bonding system (Model EVG510, EVG Group, St. Florian, Austria) (Figure 4-9). It is
important to keep the wafer in vacuum before increasing the temperature to prevent copper
oxidation. The handle layer of the SOI wafer was then ground down to 100 µm. At this step, a
protection layer (ProTEK B3 alkaline protective coating, Brewer Science, Rolla, MO) was coated
and cured on the backside in order to protect the vias during the handle wafer removal process.
This protection layer also prevents any liquid from flowing into the bonded cavities through the
via locations in case there is a leakage. A heated TMAH solution was used to selectively etch the
remaining handle layer over the BOX layer. The silicon plate was released after removing the BOX
layer in 10:1 BOE solution (Figure 4-10).
60
Figure 4-9: Deposition of the silicon nitride insulation layer on SOI device layer. Anodic bonding in
vacuum.
Figure 4-10: Handle wafer and BOX layer removal.
4.3.4 Reaching Vias and Sealing
We performed a third photolithography step in order to evacuate the gas generated during
anodic bonding, and also to reach the vias for top electrode connection. The lines to guide the final
dicing of the arrays were defined at the same time. For the gas evacuation, we chose to open the
plate over the TGV for bottom metal connection (Figure 4-11). In this way, we could later seal this
location so that the via is mechanically isolated from the active area but electrically connected to
the bottom electrode. This improves the reliability of the process because CMUT operation will
be independent of the TGV condition in case of failure of hermetic sealing around the via, which
61
would break the CMUT vacuum and degrade the CMUT performance. This was observed in the
first-generation devices, mainly because of some micro-cracking around the via boundary [68].
Prior to sealing, the photoresist on the front side was removed by oxygen plasma and the protection
layer on the backside was removed by RIE using CF4 gas. The wafer was then sealed under
vacuum with 1-µm conformal PECVD silicon nitride at 1000-mTorr chamber pressure and 350◦C
temperature (Figure 4-12). In order to avoid copper oxidation, the wafer needs to stay in vacuum
when chamber temperature is above 100◦C.
Figure 4-11: Silicon/silicon nitride etch for gas evacuation, array separation, and reaching top
electrode connection.
Figure 4-12: Conformal PECVD silicon nitride deposition for sealing.
62
4.3.5 Forming Electrical Contacts
To create the top electrode connections, the silicon nitride layer on the top electrode
connection vias had to be removed. At this point, the silicon nitride deposited on the conductive
silicon plate was also removed and the sealing layer was only left covering the cell that encloses
bottom electrode TGV (Figure 4-13). After removing the negative photoresist using oxygen
plasma, 20-nm chromium and 180-nm gold were deposited by dc sputtering over the entire wafer.
The silicon nitride etching mask was used again, this time with positive photoresist to remove the
metal on the sealing region in order to eliminate the parasitic capacitance at this location (Figure
4-14). The last photolithography was done on the backside to define the bottom electrode pads and
a rectangular grid of top electrode connections to facilitate probing of elements. 20-nm chromium
and 180-nm gold were deposited and then lifted off on the backside (Figure 4-15). At this step, the
device fabrication was completed. Figure 4-16 shows completed device. On the left is the SEM
cross-section of a completed CMUT cell with the dedicated TGV for bottom electrode connection.
On the right are the optical image of the front side and backside of the fabricated 16×16-element
2D CMUT array. The physical parameters of the fabricated transducer elements are shown in Table
4-1
63
Figure 4-13: Silicon nitride etch to reach the plate and top electrode TGV.
Figure 4-14: Sputter top electrode and build a connection to dedicated TGV.
Figure 4-15: Evaporation and lift-off for backside contact pad and testing grid.
64
Figure 4-16: Completed device. SEM (Left). Optical photos (Right)
Table 4-1: The physical dimensions of the fabricated 2D CMUT array element.
Shape of the cell
Cell width
Cell-to-cell distance
Top metal thickness
Silicon layer thickness
Insulating layer thickness
Gap height
Bottom metal thickness
Substrate thickness
Number of cells per element
Length of an element
Width of an element
Element pitch
Circular
78 µm
4 µm
0.2 µm
1.5 µm
0.2 µm
0.17 µm
0.15 µm
700 µm
8
243
243
250
65
4.4 Device Characterization
4.4.1 Static Deflection Under Atmospheric Pressure
One of the main features of the 2D CMUT with TGVs is that a reliable vacuum sealing of the
active CMUT cells could be achieved by mechanically isolating the through-glass via from the
active CMUT cells. The achievement of vacuum sealing can be confirmed by measuring the plate
deflection under atmospheric pressure. We used a stylus surface profilometer (Dektak 150, Veeco
Instruments Inc, Plainview, NY) and measured a maximum deflection of 80 nm in the center of a
circular CMUT cell (Figure 4-17). The finite element model (ANSYS v.15, ANSYS, Inc.,
Canonsburg, PA) predicts that the atmospheric deflection is 78 nm, confirms the sealing, and also
proves that there is no significant stress generated on the plate during the fabrication.
Figure 4-17: Measured and simulated deflection profile of the plate after the device fabrication was
completed.
66
4.4.2 Parasitic Resistance and Capacitance of the TGVs
One of the main motivations for using copper through-glass vias for interconnection is the
reduced parasitic resistance and parasitic capacitance. We designed via test structures with
different pitch (125 µm, 190 µm, and 250 µm) to characterize the via resistance and via-to-via
capacitance as shown in the left panel of Figure 4-18. Both resistance and capacitance
measurements were performed by accessing the vias from the backside of the wafer after the
backside metal pad formation. The resistance test vias are connected on the front side by a metal
layer that is formed at the step of bottom metal deposition. The measurement setups are shown in
the right panel of Figure 4-18. The via resistance test structures include two vias connected with a
metal line. Therefore the resistance includes two TGV resistances in series with the resistance of
the 20-µm-wide metal line between the vias. We measured 10 test structures for each pitch using
a multimeter (U1272A Handheld Digital Multimeter, Agilent, Santa Clara, CA) connected to two
needle probes in a probe station. The resistance distribution is shown in Figure 4-19 and the
average resistance of structures with 125-µm, 190-µm, and 250-µm pitch are 6.8 Ω, 9.7 Ω, and
13.1 Ω, respectively. As a result, the resistance of a single via including the contact resistance is
calculated as approximately 2 Ω. The bottom gold sheet resistance is approximately 1 Ω /sq, which
matches the sheet resistance of 130-nm-thick gold that has been reported in the literature [69].
67
Figure 4-18: Via test structures (left panel), via resistance measurement setup (right panel top), and
via-to-via capacitance measurement setup (right panel bottom).
Figure 4-19: Measured results for resistance test structures.
The via-to-via capacitance is mainly contributed by the two vias as electrodes with the glass
as a dielectric between them. The metal pads on the via will only contribute a negligible amount
of capacitance. The capacitance measurement is performed by using a 125-µm pitch coplanar
68
microwave probe (ModelACP40-GSG-125, CascadeMicrotech, Beaverton, OR) connected to a
network analyzer (Model E5061B, Agilent Technologies, Inc., Santa Clara, CA). The calibration
was carefully done in a tight frequency range from 10 MHz to 10.000005 MHz in order to measure
the femtofarad-level capacitance. The measured capacitance distribution is shown in Figure 4-20
and the average capacitance of structures with 125-µm, 190-µm, and 250-µm pitch are 47.8 fF,
34.8 fF, and 21.0 fF, respectively. The measured values are confirmed by the finite element model
and analytical calculation using Eq.1. In the equation, 2d is the center-to-center pitch of two vias;
R is the via radius; l is the substrate thickness, i.e, via length. Some variance between the model
and the measurement could be due to the fact that the actual via is tapered while the model assumes
the via to be cylindrical. The comparison of the measurements and models is summarized in Table
4-2. One should also note that the parasitic capacitance is between the signal electrode (bottom
electrode) and ground (top electrode) for each element. Hence the measured via parasitic
capacitance represents the worst case. To that end, the measured via-to-via capacitance is a more
accurate representation for element-to-element electrical coupling. In any case, given the CMUT
device capacitance is usually on the order of picofarads, the parasitic capacitance introduced by
the TGV interconnects is negligible.
69
Figure 4-20: Measured results for capacitance test structures.
(2.31)
Table 4-2: Average via-to-via capacitance: Measurement and Simulation Results
Via pitch
Measured
FEM
Analytical
125 µm
47.8 fF
51.7 fF
54.1 fF
190 µm
34.8 fF
37.9 fF
38.6 fF
250 µm
21 fF
30.5 fF
32.8 fF
4.4.3 Electrical Input Impedance in Air
We tested the fabricated elements in air using a network analyzer (Model E5061B, Agilent
Technologies, Inc., Santa Clara, CA) with an internal dc voltage source available up to 40 V. We
probed the elements from the backside of the wafer. The input electrical input impedance are
02ln[()1]rlCddRR 70
measured in air (Figure 4-21). The open-circuit resonance frequency of a 2D CMUT array element
was measured as 3.32 MHz at 15-V dc voltage, which is approximately 80% of the pull-in voltage.
The baseline in the real part corresponds to the series resistance of the device that includes via
resistance as well as the resistance of the bottom electrode, which is measured as 21 Ω at 5 MHz
from Figure 4-21(a). The device capacitance is calculated as 2274 fF at 5 MHz from Figure
4-21(b). The resonance frequency, collapse voltage, and device capacitance were simulated using
the equivalent circuit model [70] and by finite element model using TRANS 126 electromechanical
transducer elements for direct coupling of electrostatic and structural domains. The material
properties used in the simulations are listed in Table 4-3. The simulations and the measurements
are compared in Table 4-4. The results show that the fabricated CMUTs with TGV interconnects
have a small parasitic capacitance (approximately 200 fF) and operate as predicted by the models.
(a) (b)
Figure 4-21: Impedance measurements (Vdc = 15 V) (a) Real part of the electrical input impedance;
(b) Imaginary part of the electrical input impedance.
71
Table 4-3: Material Properties Used in Simulations.
Glass
Silicon
Silicon Nitride
Gold
Young's Modulus, GPa
Density, kg/m3
Poisson ratio
148
260
70
2328
3100
3300
0.17
0.27
0.33
Relative permittivity
4.6
11.7
5.7
Table 4-4: Simulation Versus measured device performance in air
Measured
FEM
Analytical
Resonance Frequency
3.32 MHz
3.47 MHz
3.34 MHz
Collapse Voltage
17.5 V
15.4 V
17 V
Device Capacitance
2274 fF
2001 fF
2054 fF
4.5 Chapter Conclusions and Future Work
We presented a CMUT fabrication process that integrates TGV interconnects and anodic
bonding [68], [71]. The process is low-temperature and eliminates the need for an insulating lining
for making through-wafer interconnects. Anodic bonding has the specific advantage of being more
tolerant to roughness on the bonding surface compared to commonly used fusion bonding
technique. By opening the plate over the bottom electrode TGV and then re-sealing, a reliable
72
vacuum seal can be achieved for all the elements. The use of glass as substrate and metal for
interconnects and electrodes reduce the parasitic capacitance and the series resistance of the
CMUTs. The resistance of a single via is measured as 2 Ω. The via-to-via capacitance of a 250-
µm-pitch via pair is measured as 21 fF. The impedance measurements demonstrate the fabricated
device has low parasitics and operates as the models predict.
In the presented approach, by decreasing the substrate thickness, a smaller via diameter could
be realized and thus more space and flexibility in array design can be achieved. 40-µm vias with
80-µm pitch have already been demonstrated in a 0.65-mm-thick borosilicate glass wafer [22]. 20-
µm vias are possible in 0.3-mm-thick borosilicate glass. Several aspects of the process flow could
be further improved. First, the insulation silicon nitride could be replaced by an ALD HfO2 layer
defined in the CMUT cavities by a lift-off process. We demonstrated the feasibility of this
approach in [55]. Also, for a better plate thickness control, an etch-stop could be added after the
handle removal process to avoid the silicon plate over-etch in sealing nitride etching step. First
and foremost, this paper demonstrates the process feasibility for incorporating TGVs in 2D CMUT
array fabrication using anodic bonding. The fabricated arrays need to be flip-chip bonded to the
electronics for immersion measurements, which is outside the scope of the presented work but will
be investigated in the future.
73
Chapter 5 A MEMS T/R SWITCH EMBEDDED IN
CMUT STRUCTURE
Ultrasound imaging systems require high-voltage transmit pulses (>100 Vpp) in the transmit
mode and require low-noise amplification of the received echo signals in the receive mode.
Transmit/receive (T/R) switches are critical components in an ultrasound imaging system as the
system needs to frequently switch back and forth between the transmit mode and receive mode
during imaging. An ideal T/R switch should act like an open circuit in the off-state and short circuit
in on-state.
Traditionally, the T/R switches are implemented based on electronic components such as
diodes or field-effect transistors (FETs) [72][73], which is not ideal as they introduce noise and
distortion to the received signals and fail to completely isolate the receiving paths from the transmit
pulse causing some extended dead zone in the near field. A micro-electromechanical (MEMS)
switch can significantly help as it provides low insertion loss in closed state and a high-isolation
in open state. In this chapter, we present a fast-switching (switching on in 1.25 µs, switching off
in 80 ns), low control voltage (2.5 V), high isolation (<-50 dB), DC-contact mode MEMS switch
that could be integrated and co-processed into a CMUT element. The significance of this work is
to improve the CMUT based ultrasound imaging system efficiency and ease the high-voltage
requirements of the front-end circuits.
74
5.1 Transmit/Receive Switching in an Ultrasound Imaging System
In many current ultrasonic imaging systems, the transducer array located in a hand-held probe
is connected to the main processing unit via a bundle of cables. The transmit circuitry and receive
circuitry are all located in the main processing unit. A close integration of the transducer array and
the electronics is highly desired to minimize the effect of cable losses and parasitic capacitance,
which could degrade the noise performance of the system, and limit the achievable bandwidth.
Furthermore, integration helps to decrease the number of connections needed to the main
processing unit by multiplexing several channels. For some applications such as intraoperative
navigation and intravascular diagnostic imaging, the size of the probe and the number of the cables
connected to the probe are even more important.
A significant challenge of the ultrasound analog front-end electronics is the requirement for
high-voltage excitation signals. The front-end integrated circuit (IC) needs to be capable of passing
through the high-voltage excitation pulse and also amplifying the received small echo signal.
Therefore, an important task is to facilitate the use of high-voltage excitation pulse while protecting
the low-voltage amplifiers.
75
Figure 5-1: Top-level block diagram of a single TX/RX channel for ultrasound frontends
The common implementations of the T/R switch in ultrasound frontends are based on using a
diode bridge or cross-coupled diode pair [59]. The top-level block diagram of a single T/R channel
is shown in Figure 5-2.
Figure 5-2: Current implementation of T/R switches using diodes pairs (single channel)
The circuit architecture can be divided into two paths. In the transmit paths on top, the diode
expander circuitry D3 allows the transmit pulse to pass through during transmit cycle, while it
76
presents an open circuit to the transducer during receive cycle. In the IC implementation, each
diode in the expander circuitry was implemented in an electrically isolated region of the n-type
epi-layer. To make sure that the parasitic bipolar transistors do not adversely affect the operation
of the diodes, the n-type epi region was connected to the p-side of the diode. In the receive path at
the bottom, a trans-impedance preamplifier (TIA) is followed by a buffer. Both the input and the
output of the receiving path is protected from the high-voltage pulses using diode limiters, which
composed of a resistor and two back-to-back diodes (D1 and D2).
Although this circuit architecture provides an automatic T/R switching mechanism, it suffers
from several disadvantages [72]. First, the resistor R1 and R2 plays an important role of the trade-
off between power losses and thermal noise. During the pulsed excitation, both the transducer and
the two resistors load the driving circuit. To limit the power losses, high resistor values are desired.
However, this results in additional noise to the received signals. Also, the diode-based switches
could introduce some distortion to the signals due to nonlinearity. Third, this implementation fails
to completely isolate the receiver from the transmit pulse and therefore the input of the TIA on the
receiving paths is charged by the transmit pulse. During the receiving cycle, the discharge has to
occur first in order to bring the TIA back to its operating point, which causes some extended dead
zone in the near field on the reconstructed image.
To overcome the above shortcomings, the development for a high isolation, low insertion loss,
fast-speed T/R switch that could be integrated with an ultrasound transducer array is highly needed.
In addition to excellent T/R isolation, a MEMS switch implementation could also benefit the
system noise performance in two ways: first, a low-voltage process can be used for the amplifier,
which is optimized for low-noise performance. Second, it will eliminate the high-voltage
protection circuitry and reduce the overall electronic noise. Besides, a MEMS switch will also help
77
improve the linearity of the front end.
5.2 MEMS Switches
MEMS switches have been widely studied for decades, mainly for RF applications, such as
radar systems, satellite communication systems, and wireless communication systems. MEMS
switches have some advantages that make it closer to an ideal T/R switch required in an ultrasound
imaging system front-end:
• Very high isolation in the off-state and low insertion loss in the on-state. MEMS
switches are fabricated with air gaps, therefore have a very low off-state capacitance
(2-4 fF) and result in excellent isolation.
• Near-zero power consumption. Although tens of volts electrostatic actuation voltage
is needed, MEMS switches have near-zero power dissipation because they do not
consume any current.
• Good linearity. MEMS switches are very linear devices and therefore can minimize
signal distortion or intermodulation.
• Low cost and ease of fabrication. MEMS devices are fabricated by micromachining
techniques on substrates such as silicon, glass, GaAs, or low-temperature cofired
ceramics (LTCC).
On the other hand, the currently available MEMS switches have some well-known problems:
78
• Relatively low-speed. The switching speed of most RF MEMS switches that has been
reported is around 2-40 µs.
• Power handling. Most RF MEMS switches cannot handle power more than 20-50 mW
with good reliability.
• High-voltage drive. Electrostatic MEMS switches require 20-80 V actuation voltage
for reliable operation.
• Reliability. Many systems require switches with 20-200 billion cycles. However, the
current mature RF MEMS switches can switch 0.1-10 billion cycles.
• Packaging. Packaging costs are high as MEMS switches need to be packaged in inert
atmospheres such as nitrogen and argon.
An example of an RF MEMS switch is shown in Figure 5-3. Typically, RF MEMS switches
have a lateral dimension of hundreds of microns, and an air gap of several microns.
Figure 5-3: A RF MEMS switch example: Rockwell Scientific MEMS series switch [74]
Table 5-1: Important dimensional parameters for the Rockwell MEMS switch [75].
79
Horizontal dimensions
Total Length
Total Width
Control electrode length
Control electrode width
RF line length
RF line width
Vertical dimensions
Gap height
Bottom metal thickness
Bump thickness
Contact area
Cantilever thickness
250 µm
150 µm
75 µm
75 µm
200 µm
20-40 µm
2-2.5 µm
2 µm
1 µm
400 µm2
2 µm
However, the MEMS switches designed for RF applications usually cannot be directly applied
to an ultrasound imaging system mainly due to the following two reasons. First, most of the RF
MEMS switches do not meet the switching speed requirement for an ultrasound imaging system
front-end [76]. Also, the previously reported MEMS switches do not have a compatible process
flow and cannot be conveniently integrated with ultrasound transducers.
CMUTs and MEMS switches are both electrostatically actuated capacitors with a moving
plate and share similar fabrication technologies. Although the integration of these two components
80
seems a natural decision, there has been no demonstration of CMUTs with embedded MEMS T/R
switches. Integrating a MEMS switch on the same substrate with the transducers will not only
improve the system performance contributed by its high isolation and low insertion loss but also
it can enable implementation of the LNA circuitry in a standard low-voltage CMOS process. This
front-end amplification is essential to preserve the bandwidth and signal integrity, especially when
small transducer elements are loaded by a long cable.
We propose to design and fabricate a fast-switching, low-control-voltage, DC-contact-mode
series MEMS switch that can be integrated and co-fabricated with CMUTs on the same substrate
using our previously reported anodic-bonding-based process. A highly integrated CMUT based
ultrasound system could improve the system performance, ease the high-voltage requirement of
the front-end electronics and therefore enable new imaging, diagnostic, and therapeutic
applications.
5.3 Design of a MEMS Switch Embedded in the CMUT Structure
Since the basic building block of a CMUT is a parallel plate capacitor that pulls in due to
mechanical instability at a certain electrical field, an electromechanical switch can be easily
embedded in the CMUT structure. The electrostatic force that pulls down CMUT plate is also what
makes a MEMS switch work. Therefore integration of these two components on the same substrate
is highly feasible.
81
5.3.1 Switch Structure Design
We need to make some modifications to a single CMUT cell in order to build a switch as
shown in Figure 5-4. In the proposed MEMS switch structure, we have an interrupted transmission
line crossing the midline of the device, so the switch is normally in OFF-state. On the two sides of
this transmission line, there are control electrodes to apply the electrostatic force to pull the plate
down. A metal contact is defined under the silicon nitride insulation layer at the bottom side of the
plate so that it closes the circuit in the interrupted transmission line when the plate is pulled down,
which is the switch ON-state. The insulation layer is needed so that the RF signal does not interfere
with the control signal. This approach will only require the distribution of digital control signals
for the switches and analog signals lines.
(a) (b)
Figure 5-4: (a) MEMS switch structure modified from a single CMUT cell. (b) T/R switch in an
ultrasound imaging system.
82
The designed MEMS switches could be integrated into a CMUT 1D array in the following
two ways as shown in Figure 5-5. In one design [Figure 5-5(a)], one switch will be placed on both
sides of a 1D CMUT array element. The control signal for the switch on each side will be 180º out
of phase. Therefore, when the top switch is on and bottom switch is off, the element will be
connected to the high-voltage transmitting circuitry and disconnected from the receivers. On the
contrary, when the control signal is reversed, the transmitter will be disconnected and the receiver
will be connected to the element. All the T/R switching will take place at the same time, therefore
only one control signal and its complement are required for the whole array. In the actual imaging
case, the "on time" for TX switches can be set according to the maximum delay required for
transmit beamforming. The second configuration [Figure 5-5(b)] is more suitable for a direct
connection to a standard imaging system. As the additional switch provides protection for the
output of the LNA, and hence only one cable is needed to pass through the high-voltage transmit
pulse and received the amplified echo signal to the system.
83
(a)
(b)
Figure 5-5: (a) Two-switch configuration where TX and RX paths are totally isolated. (b) Three-
switch configuration for seamless integration of low-voltage frontend IC with a standard imaging
system.
5.3.2 Analytical Calculation of Switching Time:
It has been reported that the switching time is mainly determined by the device mechanical
resonant frequency
. For an inertia-limited system (with a small damping coefficient and Q≥2),
the switching time
depends on the device resonant frequency
, pull-in voltage
, and
actuation voltage
, as shown in Eq. (5.1)
0fst0fpVsV
84
(5.1)
The vacuum-backed membrane structure with small dimensions gives CMUTs the ability to
make devices with MHz resonant frequencies which translate to fast switching. Considering that
the CMUT resonant frequency for medical applications is usually in the range of several MHz,
pull-in voltage is in the range of tens of volts, Table I presents several examples of fast switching
time assuming typical CMUT operation parameters.
Table 5-2: Examples of the calculated switching time
1 MHz
3 MHz
5 MHz
10 V
15 V
20 V
13 V
20 V
26 V
450 ns
146 ns
90 ns
5.3.3 FEM Modeling:
In order to get the device dimensions and verify the analytic calculation results, a finite-
element model (FEM) was developed in ANSYS APDL (ANSYS v.17.2, ANSYS Inc.,
Canonsburg, PA, USA). CMUT structure has been widely studied using FEM models [77][35].
Although the switch structure is similar to a CMUT, the operating principle and the performance
specifications are different. Also, the metal contact, RF line, and the control electrodes need to be
defined in the switch structure.
03.672pssVtVf0fpVsVst 85
A full 3D model is developed in order to get accurate results. We start the modeling with the
plate geometry definition. The plate is composed of a silicon plate layer, a stacked Cr/Au layer top
electrode on the plate, a silicon nitride insulation layer underneath, and a Cr/Au metal contact at
the bottom of the insulation layer. The material properties of each layer are listed in Table 5-3.
SOLID45 elements are suitable for the plate modeling because it has plasticity, creep, swelling,
stress stiffening, large deflection, and large strain capabilities. The plate is fixed by the 10-µm
outer boundary at the bottom of the silicon nitride layer, which represents the post region formed
by anodic bonding.
Table 5-3: Material properties used in FEM simulations
Silicon
Silicon nitride
Gold
Immersion medium
Young's modulus (GPa)
148
296
79
Density (kg/m3)
Poisson ratio
2328
0.17
Relative permittivity
Speed of sound (m/s)
3187
19300
0.27
5.7
0.44
1000
1500
The control electrodes and the metal contact are differentiated using two types of
TRANS126 electromechanical transducer elements according to their gap distances. For the
control electrodes, the gap height is the effective gap height
,
(5.2)
SiNeffgaprTTT
and the close gap height is 0. For the metal contact, the gap height is
, and the close gap height
is the metal contact height
. We first perform the static simulation in air to model device
steady-state behavior, including the atmospheric deflection and the plate deflection with DC bias.
86
Figure 5-6: Plate structure model and created elements for static simulation.
A more important aspect of the switch is the dynamic behavior, which is the switching
speed. The switch is next to the CMUTs and will operate in immersion, which is beneficial to the
switching speed because the medium will increase the damping and reduce the plate oscillation.
We model the immersion medium using a sphere medium using FLUID30 element. The fluid-
structure interface is specified using proper fluid-structure interaction flags. FLUID130 element is
used for an absorbing boundary to extend the fluid domain to infinity. Transient simulation with
the full method is used to including the all the non-linearities. The time step is set to be smaller
gapTcontactT
than
to capture the mode frequency at
, which is first determined by performing a harmonic
analysis. The simulated results are compared and discussed with the actual device measurement
87
results in section 5.5.
5.4 Fabrication Process
The fabrication process flow is shown in Figure 5-7. The first cell on the left of the cross-
sectional drawing and in the front of 3D drawing represent the switch. The other cells represent
the co-fabricated CMUT element. The initial glass substrate was the same wafer that was used in
the platform process [Figure 5-7(a)]. First, the substrate was cleaned using a piranha solution for
removal of organics and other contaminants. Then the cavities of the switch and the CMUT were
patterned and etched to 350-nm deep simultaneously [Figure 5-7(b)]. The photoresist was then
removed and the bottom metal pattern was defined by a second photolithography step using the
same photoresist. After hard baking the photoresist for 5 minutes at 125ºC, a 1-min BOE etch was
performed to smooth the bottom of the cavity and also to enlarge the photoresist undercut to
facilitate lift-off. A stacked metal layer that consists of 20-nm chromium and 100-nm gold was
formed in the etched cavities by evaporation and liftoff [Figure 5-7(c)]. The metal layer serves as
the bottom electrode for the CMUT and also control electrodes and RF line for the switch. Figure
5-10(a) shows the optical image of the switch and the CMUT after the completion of process on
the glass substrate. Figure 5-8(a) shows the AFM image of the center region of the switch cell.
120ff 88
(b)
(d)
(f)
(h)
(j)
(a)
(c)
(e)
(g)
(i)
Figure 5-7: Fabrication process flow
89
Figure 5-8: AFM image of the center region of the switch cell (left), and the corresponding metal
contact.
Figure 5-9: AFM images of the switch structure and the co-fabricated CMUT structure
Next, we deposited a 200-nm PECVD silicon nitride insulation layer on the device layer of
the SOI wafer. The bump can then be formed on the insulation layer by evaporating and lifting off
a stacked metal layer of 20-nm chromium and 50-nm gold [Figure 5-7(d)]. The insulation layer is
needed for both the switch and the CMUT: it prevents the short circuit when the CMUT pulls in,
and isolate the metal contact from the top plate electrode for the switch.
90
The processed glass and SOI wafers were first aligned and then anodically bonded at 350◦C
under 2.5-kN down force in vacuum (10−4 Torr) using an in-situ aligned wafer bonding system
(Model AML-AWB-04, Applied Microengineering Ltd, Oxfordshire, United Kingdom) [Figure
5-7(e)]. To release the plate, the handle layer of the SOI wafer was ground down to 100 µm and
the remaining handle layer and the BOX layer were removed using a 10% TMAH at 80ºC and a
10:1 BOE solution, respectively [Figure 5-7(f)]. Figure 5-10(b) displays the optical image that
shows the bonded SOI device layer as the plate and a close-up view of the aligned metal bump
over the RF line.
In the next step, we dry-etched the silicon/silicon nitride plate [Figure 5-7(g)]. This step serves
the following purposes. First, it evacuates the generated gas during anodic bonding in the switch
and CMUT cells, so that the cells could be later sealed in vacuum. Second, it isolates the top
electrode plate of the switch and the CMUTs to enable independent biasing of the CMUT and the
switch cell. Third, it allows accessing the bottom pads of the switch and CMUTs. Additionally,
this step defines the dicing lines for device separation. After the etching, the photoresist was
removed using oxygen plasma. Figure 5-10c shows the optical image after this step.
We deposited 800-nm conformal PECVD silicon nitride on the entire wafer surface to seal the
cavities [Figure 5-7(h)]. After that, the silicon nitride needs to be etched in order to access the top
plates and also the bottom electrodes for both the switch and the CMUT. The sealing silicon nitride
was only left at the places where sealing is needed [Figure 5-7(i)]. Then we transferred the wafer
to the evaporation chamber without removing the photoresist. Finally, a stacked layer of 20-nm
chromium and 180-nm gold was evaporated and lifted off to form electrical contacts [Figure
5-7(j)]. Figure 5-10(d) shows the optical image of the final device.
91
(b) Aligned anodic bonding
(a) Bottom metal formation
(c) Top plate etch
(d) Completed devices
Figure 5-10: Optical images of the critical processing steps
92
Figure 5-11: Cross-sectional SEM image of the finished switch
Table 5-4: Physical parameters of the tested switch
Horizontal dimensions
Shape of the cell
Cell width, Wcell
Control electrode width, Wcontrol
Control electrode length, Lcontrol
Bump width, Wbump
Bump length, Lbump
RF line width, WRF_line
Ctrl electrode to RFline gap,Gcontrol
RF line gap, GRF
Vertical dimensions
Substrate thickness,
Glass etching depth,
Bottom metal thickness,
Bump thickness,
Insulation layer thickness,
Silicon plate thickness,
Top metal thickness,
93
Square
80 µm
29 µm
74 µm
6 µm
13 µm
10 µm
3 µm
3 µm
700 µm
0.35 µm
0.15 µm
0.07 µm
0.2 µm
2 µm
0.2 µm
94
5.5 Device Characterization
5.5.1 Experimental Setup
The switch die was diced from the wafer and the switch pads were wire bonded to a chip
carrier. During the wire bonding and test, electrostatic discharge (ESD) should be prevented to
avoid damaging the switches due to stiction or dielectric breakdown [78]. A printed
circuit board (PCB) was designed for the test to reduce the cable parasitics.
Figure 5-12: A switch die wire bonded on the PCB.
5.5.2 Static Characterization
The PCB with the wire bonded MEMS switch was first characterized in air to study the steady-
state behavior of the MEMS switch. A dc power source (Model PS310, Stanford Research
Systems, Sunnyvale, CA) was connected through the PCB to the top plate of the switch and the
95
control electrodes were terminated to ground. A multimeter (Model U1272A, Agilent
Technologies, Santa Clara, CA) was connected between the RFin and RFout to measure the
resistance of the switch. In the meanwhile, a Wyko surface optical profiler (Wyko NT1100, Veeco
Instruments Inc., Plainview, NY) was used to measure the plate surface profile. The dc bias was
increased from 0 V to 90 V with a step of 1 V and a dwell time of 10 s to let the switch reach the
steady state. The measured switch surface center deflection at different dc bias voltages are shown
in Figure 5-13(a). The FEM static simulation results are shown in Figure 5-13(b). A good
agreement between the model and the measurement was observed, confirming that the MEMS
switch is vacuum sealed and a good dimensional control and stress control was achieved. From
the steady-state measurement, it was observed that the switch turned on and off both at 68 V. The
on-resistance was measured as approximately 75 Ω. Given the RFin to RFout metal path resistance
is 25 Ω, the switch contact resistance is approximately 50 Ω, which could be reduced by optimizing
the switch geometry. Figure 5-14 shows the conductivity between the RFin and RFout from 40 V
bias voltage to 90 V bias voltage. The inset graphs are the switch surface profile for the off-state
(65 V) and on-state (70 V).
96
Figure 5-13: Static deflection at different bias voltages. (a) Wyko measured results. (b) FEM static
simulation results.
Figure 5-14: Steady-state dc measurement.
97
5.5.3 Dynamic Characterization
For medical ultrasound imaging applications, the MEMS T/R switches integrated next to the
CMUT elements need to work in immersion, which is also beneficial for the switching speed of
the MEMS switch because the medium damping will reduce the plate oscillation. A low-voltage
control signal is desired for an ultrasound imaging system front end to ease the requirement of
circuit design. It is also required for switch operation in order to minimize the oscillation amplitude
during the switching. Therefore, we biased the switch at 67 V, which is close enough to its
switching voltage (68 V). Then we increased the control signal amplitude from 0-V with a 0.1-V
step. We found that the switch could be operated at a control signal amplitude of 2.5 V. Another
important factor related to the switching speed is the rise time and the fall time of the control
signal. We investigated this behavior using a single-point laser Doppler vibrometer (OFV-534
sensor head, OFV-5000 controller, Polytec GmbH, Waldbronn, Germany). A 10 microscope
lens was used to concentrate the laser beam to a 3-µm diameter spot on the top plate over the metal
bump. The DD-300 decoder with a frequency range from 30 kHz to 24 MHz was
used to filter out the low-frequency noise. We used vegetable oil as the immersion medium due to
the exposed pads and bonding wires. The laser was focused through the oil on the
plate surface. A dual-channel waveform generator (33522A, Agilent Technologies, Santa Clara,
CA) was used to deliver the control signal and also the RF signal to the MEMS switch.
The control signal was generated using the pulse mode in order to enable modification of the edge
speed. For the rising edge, the rise time was changed from 20 ns to 1 µs. The control
signal and the RFout signal were recorded using an oscilloscope (MSO-X 3024A, Agilent
Technologies, Santa Clara, CA). In the meanwhile, the vibration of the plate was measured using
the vibrometer. Figure 5-15 shows
the recorded RF signal and
the corresponding
plate displacement for a rise times of 100-ns, 200-ns, and 300-ns. The measured displacement
results were compared to FEM simulation result.
98
Figure 5-15: Dynamic characterization with a control signal rise time of (a) 100 ns; (b) 200 ns; and
(c) 300 ns. Figure (d)-(f) shows the corresponding plate center displacement.
It can be observed that the switching time is first dominated by the metal bump rebound after
the first contact. As the rise time increases, the rebound becomes less significant and reduces the
switching time, which is due to the center frequency of the rising edge deviate from the plate
mechanical resonant frequency. However, as the rise time further increases, the switching time is
dominated by the prolonged rise time and therefore increase again as shown in Figure 5-16. For
the switch under measurement, the optimum switching time is 1.34 µs and the corresponding rise
time is 300 ns.
99
Figure 5-16: Switch switching time versus control signal rise time.
For the release time, the speed is only determined by the fall time of the control signal. Figure
5-17 shows the recorded RF signal and its corresponded displacement compared to the FEM
simulation for the fall time of 100 ns, 200 ns, and 300 ns, respectively. Figure 5-18 shows the
relationship between the fall time and the release time. It can be observed that the fall time and the
release time have a linear relationship. Therefore, a fast fall time is desired for a fast switching off.
For the switch under test, the optimum release time is 80 ns while the fall time is 20 ns.
100
Figure 5-17: Dynamic characterization with a control signal fall time of (a) 100 ns; (b) 200 ns; and
(c) 300 ns, respectively. Figure (d)-(f) shows the corresponding center displacement.
Figure 5-18: Switch release time versus control signal fall time.
101
Finally, we used a 2.5-Vpp, 1-kHz square wave as the control signal with the optimum rise
time and fall time. A 1-MHz, 300-mVpp continuous sinusoidal wave was first used as the RF
input. Figure 5-19(a) shows the measured output signal. Then we used a 1-MHz, 5-Vpp unipolar
pulse wave as the input and the output is shown in Figure 5-19(b). The results demonstrate that the
fabricated MEMS switch is suitable for ultrasound imaging system operating frequency.
(a) (b)
Figure 5-19: (a) 1-MHz, 300-mVpp CW input signal turned on and off by a 1-kHz control signal. (b)
500-ns, 5-Vpp pulse signal turned on and off by a 1-kHz control signal.
102
5.6 Chapter Conclusions and Future Work
We designed and fabricated a MEMS T/R switch that could be co-fabricated and closely
integrated with a CMUT element on the same glass substrate using anodic bonding [79], [80]. A
switch test structure was characterized. The static deflection measurement agrees well with the
FEM simulation results, confirmed that fabricated switch is vacuum-sealed and has a good
dimension and stress control. The static measurement demonstrates the switch has a switching
voltage of 68 V and a contact resistance of 50 Ω, which could be reduced by
optimizing the switch geometry. We further investigated into the switching mechanism using a
dynamic characterization. The result shows the switching time is first dominated by metal
bump rebound and then dominated by the control signal rise time. An optimum rise time exists for
the fastest on-switching. For the switch under test, the best switching time is 1.34 µs with 300-ns
control signal rise time. The release time is only dominated by the fall time. A shorter fall time is
desired for a fast release. A release time of 80 ns has been demonstrated with 20-ns
control signal fall time. We also demonstrated that the MEMS switch could be operated by a 2.5-
V, 1-kHz control signal
to conduct and block a 1-MHz, 300-mVpp continuous
sinusoidal wave as an input signal, and 500-ns , 5-Vpp unipolar pulse wave as an input signal. This
work demonstrates that the MEMS T/R switch can meet the operating frequency requirement in
an ultrasound imaging system.
The fabricated switch needs to be tested with the co-fabricated CMUTs to demonstrate its
capability in an ultrasound imaging system, which will be investigated in the future. There are
103
several aspects of the MEMS switch that could be further improved. First, a thicker metal bump
could be used to reduce the series resistance of the switch, and thus improve the power handling
capability, which is especially important for transmitting high-pressure ultrasound and also
applications where a high-power continuous wave is needed such as HIFU. Also, some other metal,
such as Pt and Pd, could be employed to improve the power handling and reliability of a MEMS
switch [81] -- [84]. Second, the gap height should be reduced so that the dc bias voltage can be
minimized, which will also reduce the plate oscillation amplitude and make the switch faster.
Besides, the switch cell could be designed at a higher frequency for an even faster-switching speed
and integrate with high-frequency CMUT arrays as reported in Chapter 3. Furthermore, the
proposed MEMS switch structure is compatible with 2D CMUT array fabrication process that we
reported in Chapter 4 and could enable reconfigurable transducer array [85].
Chapter 6 TRANSPARENT CMUTS ON GLASS
104
SUBSTRATES
6.1 Applications of Transparent Transducers
The combination of ultrasonics and optics is desired in many applications such as integrating
ultrasound sensing with flat panel displays, embedded optical vibrometry, and photoacoustic
imaging [86][32]. The opacity of materials used in conventional piezoelectric transducer
constructs has severely restricted such applications. Lithium niobate with indium tin oxide (ITO)
electrodes has been investigated to build a transparent piezoelectric transducer [87]. However,
transparent CMUTs are desired because of the benefits such as ease of fabrication and integration,
and broad bandwidth. Our platform process enables fabrication of CMUT on a glass substrate that
has a high optical transmittance. A transparent CMUT structure can be realized by replacing the
metal electrode with transparent electrodes.
Indium tin oxide (ITO) has high optical transparency and electrical conductivity and hence
has been used as a transparent conductor in devices such as light emitting diodes (LED) and liquid
crystal displays (LCD) [88]. In this work, we use ITO as the bottom electrode of CMUTs to
improve the optical transparency.
The CMUTs with ITO bottom electrodes are fabricated based on our anodic bonding process.
We had also applied the process to make CMUTs that has silicon nitride instead of metal on the
105
silicon plate [55]. Glass substrate and silicon nitride have good transparency. The thin silicon plate
has some limited transparency in the red to NIR wavelength range. The major limitation in the
total transmission through the CMUT structure we have reported in our previous work was the
metal (Cr/Au) bottom electrodes. In this work, we use ITO as bottom electrode material instead of
Cr/Au. In the following section, the fabrication process and the optical and acoustical
characterization results are first presented. Then, we use the device for experiment backward-mode
photoacoustic imaging. In section 6.4, we further improved the transparency by replacing the
silicon plate with a glass plate. The device is designed for applications where optics and acoustics
need to be combined, such as integrating an ultrasound sensor under a flat panel display.
6.2 CMUT with ITO Bottom Electrodes for Improved Transparency
6.2.1 Device Fabrication:
The simplified fabrication process flow is shown in Figure 6-1, which is similar to the platform
process reported in Chapter 3 with replacing the metal bottom electrode to ITO electrode. CMUT
cavities are patterned after cleaning the wafer. The wafer was baked at 125ºC between the wet
etching cycles to prevent photoresist from peeling off. The BOE etching process has a lateral-to-
vertical etch rate ratio of 10:1. Therefore a 3.5-µm undercut can be achieved during the cavity
formation, which is beneficial for the later ITO lift-off step Figure 6-1(a). The wafer was then
transferred into an RF sputtering system without removing the photoresist. An ITO film with a
thickness of 170 nm was sputtered over the wafer in ambient temperature and then lifted off in a
heated NMP solution. Then the wafer was annealed at 450ºC for 5 minutes in a rapid thermal
106
annealing (RTA) system to improve the ITO conductivity and transparency [Figure 6-1(c)]. Figure
6-2 shows optical image and AFM image of the ITO bottom electrodes defined in the glass cavities.
The SOI wafer and the processed glass substrate were bonded together by anodic bonding.
The top plate was formed after the handle wafer and BOX layer removal [Figure 6-1(d)]. The
silicon plate was etched at the bottom pad location to evacuate the gas generated during anodic
bonding [Figure 6-1(e)]. The device was sealed using a 1-µm conformal PECVD silicon nitride
[Figure 6-1(f)]. The sealing nitride was etched to reach the conductive top plate silicon and the
bottom electrode to form electrical contacts Figure 6-1(g). The device fabrication was completed
after evaporating and lifting off a stacked layer of 20-nm chromium and 180-nm gold as the bond
pads [Figure 6-1(h)].
Figure 6-1: Simplified fabrication process flow.
107
Figure 6-2: Optical image after the ITO bottom electrode definition (left). AFM image showing the
profile of the glass post and ITO bottom electrode (right).
Figure 6-3 shows the optical image of a finished CMUT element with ITO bottom electrode
and circular cells with a diameter of 78 µm and a plate thickness of 2.5 µm. The atmospheric
deflection measurement in the center of a cell confirmed the devices are vacuum-sealed.
Figure 6-3: Atmospheric deflection measurement of a finished CMUT element. The inset shows a
finished CMUT element.
108
6.2.2
Device Characterization:
Figure 6-4 is the optical images of the CMUTs fabricated with ITO bottom electrodes (left)
and metal electrodes (right) under a microscope with backside illumination. The die with six
CMUT elements was placed above an "NC STATE UNIVERSITY" logo. From the optical image,
it is clear that the device with ITO bottom electrode has an improved transparency in visible light
range.
Figure 6-4: (a) Six CMUT elements with ITO bottom electrodes show the improved transparency of
the bottom electrodes. (b) Six CMUT elements with Cr/Au bottom electrodes presented for
comparison.
To further characterize the transmission coefficient, the ITO and Cr/Au bottom electrodes and
the finished devices were measured using a spectrophotometer (Cary 60 UV-Vis, Agilent
Technologies, Santa Clara, CA) from 400 nm to 1000 nm wavelength range. Figure 6-5(a) shows
the transmission through the 150-nm ITO bottom electrode on the glass before and after annealing,
in comparison to 150-nm Cr/Au bottom electrode on the glass. Figure 6-5(b) shows the optical
transmission through the final device with ITO bottom electrodes and Cr/Au electrodes. It is clear
that the CMUTs with ITO bottom electrodes have a significant transmission improvement in the
measured wavelength range. However, the 2-µm silicon plate is the main hurdle for the
transmission in the shorter wavelength range.
109
(a) (b)
Figure 6-5: (a) Transmission measurement. Transmission of the bottom. (b) Transmission of the
final devices.
The real and imaginary parts of the electrical input impedance of the fabricated CMUT was
measured in air using a network analyzer (Model E5061B, Agilent Technologies, Inc., Santa Clara,
CA) [Figure 6-6(a)] . The open circuit resonance frequency of the CMUT element was measured
as 3.62 MHz at 18-V dc voltage, which is approximately 75% of the pull-in voltage. The 1-kΩ
baseline in the real part corresponds to the series resistance of the device, which is mainly
contributed by the resistance of the patterned ITO bottom electrode.
110
This resistance could be reduced by depositing a thicker ITO layer or using parallel
connections to the pads. For the PAI experiment, we wire-bonded two connections to two pads
reaching the ITO bottom electrode to reduce the series resistance.
A pulse-echo measurement was performed in vegetable oil to characterize the small-signal
bandwidth in immersion and also to help quantify the effects of optical absorption in the silicon
plate on the generation of spurious transmit signals. The CMUT was placed 1.2 cm away from a
plane reflector and was biased at 18-V dc voltage (75% Vpull-in). A 1-V, 250-ns pulse was used to
excite the device. The received echo signal and its Fourier transform are shown in Figure 6-6(b).
The center frequency of the CMUT is 1.4 MHz with a 6-dB fractional bandwidth of 105%.
Figure 6-6: (a) Electrical input impedance measurement in air. (b) Pulse-echo measurement in
vegetable oil.
111
6.3 Backward-mode Photoacoustic Imaging
6.3.1 Background of Photoacoustic Imaging:
Photoacoustic imaging (PAI) is emerging as an attractive hybrid imaging modality that
combines the physics of optical imaging and ultrasound imaging [89]. It is a complement to the
traditional pulse-echo ultrasound imaging since it can provide optical contrast information.
Compared to optical imaging techniques, PAI provides excellent spatial resolution with deeper
penetration primarily determined by ultrasound wavelength [21].
An important challenge in PAI is the arrangement of the laser source and the ultrasound
transducer. Various approaches have been demonstrated for different applications [90]. One of the
commonly used arrangement is to have the light source illuminate the target from the single side
or two sides at a right angle to the acoustic path [91]. Another implementation is to have the light
source integrated as two fiber bundles along the length direction of a 1D transducer array [92],
[93].
However, this method does not illuminate the surface area under the transducer array and
results in a blind spot in front of the transducer. Also, it is difficult to use this approach with 2D
arrays, which would result in a larger area under the 2D transducer array not being illuminated.
This approach also limits the footprint of the imaging probe as the fiber bundle and the closure
required to place it next to the ultrasound array occupy extra space. The small footprint is
especially important for intracavital probes. The light illumination has also been distributed using
112
a spherical mirror [94]. Another approach a is a face-to-face arrangement which is also referred to
as forward-mode [95] -- [97], which is not practical for many clinical applications.
In some diagnostic imaging and image-guided therapeutic applications, it is desirable to have
the light source integrated behind the ultrasound transducer for implementing the so-called
backward-mode PAI. First of all, this approach would allow a more uniform illumination in the
near field. Also, it could lead to a compact form factor, which is important for therapy monitoring
applications using a catheter-based imaging system. One approach to implement the backside
illumination is to use a ring array or annular array and introduce an optical fiber bundle into the
central lumen [93], [98], or place a forward-viewing ultrasound catheter in the central lumen of an
annular light ring [99]. However, the former arrangement limits array geometry and illuminated
area and the latter results in a larger catheter size. The ideal implementation is to have a transparent
transducer which allows the laser illumination to go through with minimal self-absorption. Some
transparent piezoelectric materials such as polyvinylidene fluoride (PVDF) and lead lanthanum
zirconate titanate (PLZT) have been proposed for such purpose [100], [101]. A PVDF annular
array has been adopted with through-hole illumination for PAI [102]. Also, transparent Fabry-
Perot polymer film optical ultrasound sensors and optical micro-ring resonator (MRR) have been
investigated for implementation of backward-mode PAI [103], [104]. However, there are only
limited reports on using CMUTs for such an operation. An earlier attempt was based on CMUTs
fabricated on a silicon substrate [86], where transmission is limited and the wavelength of the light
source has to be above 1000 nm to reduce the silicon substrate absorption.
113
The above-described CMUT on a glass substrate with indium tin oxide (ITO) bottom electrode
improved the device transmission to 30% to 70% in the wavelength range from 700 nm to 900 nm,
which is commonly used for in-vivo photoacoustic imaging. We will use this device in immersion
and have laser illumination through the fabricated CMUT to demonstrate preliminary
photoacoustic imaging results.
6.3.2 Experimental Setup:
The diagram of the experimental setup for photoacoustic imaging is shown in Figure 6-7. The
CMUT die was mounted on a PCB that has a bias-T circuit and switches to select individual CMUT
elements for testing. The PCB was designed with a rectangular cutout in the center to allow light
to pass through the CMUT die, as shown in the inset of Figure 6-7(b). After wire bonding, the
CMUT and the laser output were fixed using a 3D-printed holder and mounted on a 3-axis linear
stage (model PRO165, Aerotech Inc., Pittsburgh, PA, USA) to enable mechanical scanning. The
3D-printed holder was used to ensure the relative position of the CMUT and the laser output does
not change during the experiment. A dc power supply (model E3631A, Agilent Technologies,
Santa Clara, CA) was connected to the PCB and the signal received by the CMUT was filtered and
amplified by a receiver (model 5072PR, Olympus Corporation, Tokyo, Japan). The filtered and
amplified signal was recorded by a PC-controlled digitizer (model NI PCI-5124, National
Instruments, Austin, TX). The excitation laser source is a fiber-coupled optical parametric
oscillator (OPO) pumped by a Q-Switched Nd: YAG laser (model Phocus Mobile, Opotek Inc.,
Carlsbad, CA) with a wavelength range from 690 nm to 950 nm. The laser pulses had a pulse-
width of 4.5 ns and a repetition rate of 20 Hz. The output energy of the laser was calibrated using
114
a pyroelectric energy detector (model: QE25, Gentec Inc., Quebec City, Canada). The output of
the laser was coupled to the backside of the CMUT die using a fiber bundle with a diameter of 5
mm. The target and the coupling medium were placed in a glass container under the PCB. The
bottom view of the PCB is shown in Figure 6-8(a) with the inset graph indicating the relative
location of the laser beam and the CMUT elements. We chose to use the number 2 CMUT element
on the die because the light illuminated through this entire device. The side view of the fiber
bundle, the PCB, and the holder is shown in Figure 6-8(b).
Two different targets were imaged. The first imaging target was a 0.7-mm diameter pencil
lead. The pencil lead was suspended 2 cm above the bottom surface of the glass container.
Vegetable oil was used as the medium instead of water as the transducer surface and bond wires
were not electrically insulated. We filled the vegetable oil up to approximately 1.5 cm above the
pencil lead. Then we lowered the holder until the CMUT surface touched the oil. The laser beam
output from the fiber bundle into the CMUT chip had a wavelength of 830 nm and a fluence of 12
mJ/cm2. The CMUT was biased at 18-V dc voltage. We set the receiver gain at 20 dB and the
cutoff frequency of the low-pass filter at 10 MHz. The transducer was scanned across the pencil
lead and the received signals at each location were sampled at a rate of 200 MSa/s, digitized, and
averaged 16 times to improve the SNR before recording.
The second target was designed to better mimic the condition of biological tissues. We looped
a polyethylene tube with an inner diameter of 2.3 mm and an outer diameter of 3.6 mm and filled
it with an indocyanine green (ICG) solution (50-µM), which is commonly used as a contrast
enhancement agent in PAI. The tube was then suspended using fishing lines inside the glass
115
container. Then we filled up the container with a mixed solution of 1% Agar and 1% Intralipid
(20% intravenous fat emulsion) in DI water to build the photoacoustic imaging phantom. After the
phantom was solidified, we added a 5-mm oil layer on top of the solid phantom for acoustic
coupling. The CMUT was again biased at 18-V dc voltage. This time the received signals were
amplified with 40-dB gain. The laser wavelength was chosen as 790 nm to match the maximum
absorption of the ICG solution. In order to get a stronger PA signal, this time we used laser output
fluence of 20 mJ/cm2 into the CMUT chip. By mechanically scanning in x and y directions,
volumetric data was recorded at a sampling rate of 200 MSa/s and by averaging of each scan line
16 times.
Photoacoustic images were reconstructed using the standard delay-and-sum (DAS)
beamforming algorithms [105] along with a coherence factor (CF) weighting [106]. Prior to image
reconstruction, every A-scan S(t) was processed as in [107]:
(6.1)
(6.2)
This preprocessing suppresses the low-frequency component in the signal. Then, the A-scans
were filtered by a 0.15-MHz 4.5-MHz band-pass filter to eliminate out-of-band noise. After that,
DAS receive-only beamforming was applied to form the PA image. Considering the radiation
pattern of the CMUT and to maximize the image SNR, a threshold value of 14º was chosen and
the contribution from an element was not taken into consideration if the angle from its normal to
()()(),processedStStSttt()()(()(1)),processedSiSiiSiSi
the pixel location was larger than the threshold. Finally, envelope detection was performed and the
image was multiplied by the coherence factor map. Logarithmic compression was performed
before displaying the PA image.
116
(a) (b)
Figure 6-7: A schematic diagram of the CMUT with improved transparency
(a) (b)
Figure 6-8: (a) Bottom view of the PCB with CMUT (inset graph indicates the relative location of
the laser output and CMUT elements). (b) Side view of PCB attached on the holder with the optical
fiber bundle in the back.
117
6.3.3 Backward-mode PAI Results:
6.3.3.1 Graphite Target
The experimental results of imaging of the pencil lead in oil are shown in Figure 6-9. A sample
A-scan at X=0 mm is shown in Figure 6-9(a). Four signals (S1, S2, S3, S4) are marked on the A-
scan with the signal paths shown in Figure 6-9(b). As the 4.5-ns wide laser pulse shines through
the CMUT, some of the optical energy is absorbed in the silicon plate and converted to heat.
Thermoelastic expansion of the silicon plate caused by rapid heating and cooling will set the plate
into vibration at its natural frequency in oil, which results in the Sl on the A-scan. S2 is the received
PA signal generated by the pencil lead. The tail signal after S2 is because of the substrate ringing
of the device and the reverberations in the pencil lead. S3 is the pulse-echo signal transmitted due
to S1 and reflected by the pencil lead, and therefore occurred at double the distance compared to
S2. S4 is the PA signal generated by the pencil lead reflected by the silicon plate and then reflected
back by the pencil lead. Therefore it appeared at three times the target depth. The reconstructed B-
scan image of the cross section of the pencil lead is shown in Figure 6-9(c) with 40-dB dynamic
range. The pencil lead was seen at the depth of approximately 12 mm. Substrate ringing and the
reverberations in pencil lead can be observed after the main signal. At the distance of 24 mm, a
weaker signal (34 dB lower than the pencil lead) was detected, which is due to the pulse-echo
signal generated by the silicon plate absorption and corresponds to S3 on the A-scan.
118
Figure 6-9: Pencil lead cross-sectional PA imaging results: (a) A sample A-scan at X=0 mm and its
Fourier transform after applying a Gaussian window; (b) Signal paths of the four signals on the A-
scan; (c) Reconstructed image.
6.3.3.2 Silicon Plate Absorption Spurious Signal
In order to further evaluate the effect of the silicon plate absorption, we designed the following
two experiments. First, we compared the photoacoustic signals generated by the light absorption
in the pencil lead (referred as PAtarget in the following context) and the pulse-echo signals generated
by photoacoustically induced vibration of the silicon plate (referred as PAcmut in the following
context) for the same travel distance in oil across the 690 nm to 950 nm wavelength range. PAtarget
was first recorded as described for the imaging experiments. To record PAcmut also at 12-mm travel
distance, we placed the CMUT 6-mm away from the glass container bottom, which served as a
plane reflector without generating interfering PA signals. The laser wavelength was scanned from
690 nm to 950 nm wavelength range with a step of 10 nm. The received signals were normalized
to 1-mJ/cm2 laser energy through the CMUT. The results are plotted in Figure 6-10(a) with curve
119
fitting. It can be seen that for the same travel distance, PAcmut is much smaller than PAtarget
(approximately 30 dB lower at the wavelength of 830 nm).
In the second experiment, we compare the PAcmut to a pulse-echo signal generated by the
electrical excitation (PEcmut) for the same travel distance. The aim of this experiment is to find an
equivalent excitation voltage for the CMUT that would generate a PEcmut equals to PAcmut. We use
a 250-ns, 1-V unipolar pulse to perform a regular pulse-echo test. The received echo signal
amplitude was 1.5 mVpp. Thus the equivalent electrical excitation amplitude can be calculated for
1-mJ/cm2 laser excitation through the CMUT [Figure 6-10(b)]. At the wavelength of 830 nm, the
pulse-echo signal generated by photoacoustically induced vibration of the silicon plate using 1-mJ
laser power is equivalent to that generated by the CMUT using 0.29-V electrical excitation.
(a) (b)
Figure 6-10: (a) Received PA signal using 1-mJ laser power at 12-mm travel distance: one-way
pencil lead signal (PAtarget) and two-way silicon plate signal (PAcmut). (b) Equivalent electrical
excitation amplitude for 1 mJ laser excitation.
120
6.3.3.3 ICG Solution Target
The photograph of the phantom is shown in Figure 6-11(a), where a looped polyethylene tube
filled with the ICG solution was embedded in the tissue mimicking material and suspended using
fishing lines. 3D image reconstruction was performed and then the volumetric data was rendered
using a medical image viewing software (Osirix, Pixmeo SARL, Bernex, Switzerland) [108] and
displayed by using maximum intensity projection (MIP) [Figure 6-11(b)]. The ICG tube and the
fishing line node could be seen in the rendered 3D image.
(a) (b)
Figure 6-11: ICG tube phantom (a) Photograph of the ICG-filled polyethylene tube. (b) 3D
rendered an image of the ICG-filled polyethylene tube.
121
6.4 An Optically Transparent CMUT Fabricated Using SU-8 or BCB
Adhesive Wafer Bonding
Fully transparent ultrasonic transducers are desired in consumer electronics such as
integrating large-area ultrasound sensing or directed sound generation integrated with an electronic
display. Transparent piezoelectric transducers have been reported using transparent piezoelectric
materials, such as PVDF [100], LNO [87], and transparent PZT thin film [109]. Also, AlN-based
PMUTs have been fabricated on a transparent substrate [110]. Besides, optical ultrasound sensors
can be used as a transparent ultrasound receiver [111], [112].
Transparent capacitive micromachined ultrasonic transducers (CMUTs) are desired to take
advantage of the wide bandwidth, ease of fabrication, and broad selection of processing materials.
Transparent CMUTs with polymer-based vibrating plates have been fabricated using a sacrificial
release process [32] and a roll-lamination process [113]. Wafer bonding technique has various
benefits for fabrication of CMUTs. We have fabricated CMUTs on glass substrates using anodic
bonding. We have also demonstrated a CMUT with indium tin oxide (ITO) bottom electrode for
improved transparency for the backward-mode photoacoustic imaging application.
Here we have designed and fabricated an air-coupled CMUT to achieve high optical
transparency in the full visible wavelength range by replacing the silicon plate with a glass plate.
The transparent CMUT structure was realized with two ITO coated glass wafers bonded together
using adhesive wafer bonding. Adhesive wafer bonding with photosensitive benzocyclobutene
(photo-BCB) has been demonstrated for CMUT fabrication on a silicon substrate [114]. Here we
122
investigate both SU-8 photoresist and photo-BCB as the adhesive bonding layer, which are also
used to form the cavities and the insulation layer in the CMUT structure. SU-8 has the benefits of
low cost and ease of processing. BCB has the benefits of high dielectric breakdown voltage and
high bonding strength.
The transparent CMUT was designed as a single transducer on a 100-mm-diameter wafer,
specifically for display-based air-coupled applications. The fabrication process flow is depicted in
Figure 6-12 and described below.
The starting substrate was a 1.1-mm-thick, 100-mm-diameter glass wafer (HQ-Float Glass,
Präzisions Glas & Optik GmbH) with a flat cut, coated with a 200-nm ITO film on top of a thin
SiO2 barrier passivation layer, which is suitable for display technology [Figure 6-12(a0)]. A
patterned photoresist was used as a mask for etching the ITO bottom electrode in a heated mixture
of hydrochloric acid (35% HCl), nitric acid (65% HNO3), and deionized (DI) water with a mixing
ratio of 1:0.08:1 [Figure 6-12(a1)].
To form the cavities, on one wafer SU-8-5 was spun at a rate of 3000 rpm to achieve a uniform
layer and sequentially patterned to create cavities aligned to the bottom ITO electrodes. On another
wafer, we used a commercially available bisbenzocyclobutene (BCB) electronic resin
(CYCLOTENE 4022-35, Dow Chemical Company, Midland, MI) to form cavities. An adhesion
promoter (AP3000, Dow Chemical Company, Midland, MI) was first applied and the photo-BCB
was spun at 2500 rpm. After exposure and puddle development, cavities aligned to the bottom ITO
electrode were realized Figure 6-13(a). The cavities are electrically connected to the bottom pad
123
but physically isolated from the atmosphere so that vacuum sealing of the device could be achieved
after the wafer bonding [Figure 6-12(a2)].
The plate wafer was implemented using a 175-µm-thick, 100-mm-diameter glass (D263-T
Glass, Präzisions Glas & Optik GmbH) wafer with a flat cut, coated with a 200-nm ITO film on top
of a thin SiO2 barrier passivation layer [Figure 6-12(b0)]. ITO etching was performed to remove
part of the top ITO electrode to avoid electrical field being applied to the polymer [Figure 6-12(b1)].
For SU-8 bonding process, SU-8 2 was span at 3000 rpm to form a 1.5-µm layer on the ITO film.
For BCB bonding process, BCB resin (CYCLOTENE 4022-35, Dow Chemical Company, Midland,
MI) was spun at a rate of 6000 rpm to form a 2-µm layer on the ITO film. The polymer layer serves
as the insulation layer for the CMUT and also the bonding material. We patterned the polymer layer
so that the ITO top electrode could be exposed through the flat cut on the substrate for electrical
contact. At this step, the top plate wafer was also ready for wafer bonding [Figure 6-12(b2)].
Adhesive bonding was performed in a wafer bonder (AML-AWB-04, Applied Microengineering
Ltd, Oxfordshire, United Kingdom). The top and bottom wafers were manually aligned so that the
ITO contact pads could be accessed through the wafer flat cuts on both wafers. The wafer pair was
then brought into contact with 0.3-MPa bonding pressure applied over the 100-mm wafer area. For
SU-8 bonding, the wafer pair under pressure was heated up to 120°C and maintained for 1 h. For
BCB bonding, the wafer pair under pressure was first heated up to 150°C, maintained for 15
minutes, and then heated up 250°C and maintained for 1 h, which finished the fabrication process.
The physical dimensions of the fabricated CMUTs are shown in Table 6-1.
124
Figure 6-12: Fabrication of fully transparent CMUT using adhesive bonding.
125
Table 6-1: Physical Dimensions of the Fabricated Transparent CMUTs
SU-8 bonding
BCB bonding
Shape of cells
Cell radius
Cell-to-cell distance
Cavity depth
Bottom ITO thickness
Insulation layer thickness
Top ITO thickness
Plate thickness
Circular
2.8 mm
400 µm
4.4 µm
200 nm
1 µm
200 nm
175 µm
Circular
2.8 mm
400 µm
3.6 µm
200 nm
2 µm
200 nm
175 µm
(a)
(b)
Figure 6-13: (a) Profilometer measurement of the formed cavities with ITO bottom electrodes on a
glass substrate. (b) Atmospheric deflection of a fabricated CMUT cell.
Figure 6-14(a) shows the optical image of the single CMUT element fabricated using the SU-
8 bonding process. The wafer was placed on an "NC STATE UNIVERSITY" logo to show the
transparency. In comparison, Figure 6-14(b) shows the CMUT element fabricated using the BCB
bonding process. The optical picture is shown on the top and the measured transmittance was
shown at the bottom. The average transmittance was then calculated according to the fill factor of
the CMUT active area, which is shown as the red solid curves on the plots, indicating that both
devices have 70%-80% optical transmittance in the full visible wavelength range.
126
The electrical connections to an SMA connector were established using silver epoxy on ITO
contact pads. The electrical input impedance of the fabricated CMUT elements was then measured
in air using a calibrated network analyzer (Model: E5061B, Agilent Technologies, Santa Clara,
CA). Both CMUT elements showed a resonance frequency of approximately 62 kHz and a series
resistance of approximately 30 Ω. The real and imaginary parts of the electrical input impedance
are shown in Figure 6-15(a) for the SU-8 process and in Figure 6-15(b) for the BCB process.
(a)
(b)
Figure 6-14: Fabricated fully transparent CMUT. Optical photos (top), transmittance measurement
(bottom). (a) SU-8 bonded wafer pair. (b) BCB bonded wafer pair.
127
(a) (b)
Figure 6-15: Input electrical impedance measurement. Real part (top). Imaginary part (bottom). (a)
SU-8 bonded wafer pair. (b) BCB bonded wafer pair.
Although the CMUTs reported here were designed with cavity depth of several microns, the
thinner gap can be realized by modifying the polymer coating process. A CMUT fabricated using
photo-BCB with several hundreds of nanometers gap has been reported before and SU-8 as thin
as 0.5 µm is commercially available. Besides, the plate could be implemented using a thinner glass
wafer to extend the design flexibility. The electrical contacts in the demonstrated CMUT were
realized by exposing the contact pad using the wafer flat cut. In order to facilitate integration with
electronics, through-glass-vias could be incorporated to access the ITO electrodes from the
128
backside of the wafer. Furthermore, CMUTs can be implemented with a vacuum-sealed gap or
with open cavities.
6.5 Chapter Conclusions and Future Work
In this chapter, we reported a CMUT fabricated on a glass substrate with ITO bottom
electrodes for improved transparency [13]. The CMUT plate was formed by anodic bonding a 2-
µm thick SOI silicon device layer on to borosilicate glass. The completed device has 40% to 70%
transmission in the wavelength range from 700 nm to 900 nm. The resonant frequency was 3.62
MHz in air. Center frequency in immersion was 1.4 MHz and the fractional bandwidth was 105%.
The CMUT was mounted on a custom-designed PCB with a cutout in the center to allow the laser
illumination from the backside. A pencil lead cross section was first imaged and the effect of the
silicon plate absorption was investigated. The artifact introduced by the spurious transmit signal
caused by the light absorption in the silicon plate is 34 dB lower compared to the photoacoustic
image intensity of the pencil lead at a wavelength of 830 nm. The second target, a polyethylene
tube filled with 50-µM ICG solution was embedded in a tissue-mimicking material. The
volumetric image was reconstructed by mechanically scanning the CMUT element with the fiber
bundle on the backside in x and y directions. The two preliminary PA experiments demonstrated
that CMUTs with the proposed fabrication approach is promising for backward-mode
photoacoustic imaging [115].
We are currently working on designing and fabricating 1D and 2D CMUT arrays with ITO
bottom electrodes as well as further improving the plate transparency to reduce them light
129
absorption in the device. By doing that a faster data acquisition rate and a better image quality can
be realized. In the long term, we will implement the through-illumination approach in
intravascular/intracardiac imaging, endoscopy, laparoscopy, and other intracavital applications
that require compact integration of optical and acoustic components for photoacoustic imaging.
We also demonstrated a fully transparent CMUT with glass plate, glass substrate, and ITO
electrodes fabricated using SU-8/BCB-based adhesive bonding [116]. The preliminary
measurements show the functionality of the device and indicated the transmittance of ~80% in the
visible wavelength range.
130
Chapter 7 Summary
The realization of densely populated CMUT arrays with TGV interconnects and MEMS T/R
switches on glass substrates are important for simplified transducer fabrication and the front-end
electronics integration. Also, parasitics are significantly reduced benefiting from the insulating
glass substrate, which could significantly improve the efficiency of the transducer. Furthermore, a
transparent CMUT was fabricated by taking the advantage of glass substrate transparency, which
can find its application in both medical applications and consumer electronics. The above
components were described in detail in this dissertation to establish a tool kit for the development
of the next-generation ultrasound transducers.
We use borosilicate glass as an initial substrate and use anodic bonding to implement the
movable plate of CMUTs. We first developed a platform process to fabricate vacuum-sealed
CMUTs on a glass substrate using a three-mask process. The fabricated CMUT was characterized
in air and also in immersion. The measured performance agrees well with finite-element modeling
results (FEM), which indicates low parasitics and good dimensional control was achieved. A high-
frequency (30-MHz), broadband (100%) CMUT array was further demonstrated using the same
process flow.
2D CMUT array was then developed on a glass substrate by incorporating through-glass-via
(TGV) interconnects. The fabrication flow to fabricate 2D CMUT array with vacuum-backing is
presented. The parasitic capacitance of a via pair of a 250-µm pitch was measured as 21 fF. The
131
resistance of a single via plus via-to-electrode contact resistance was 2
. A CMUT element from
a 16 16 2D array was characterized by measuring its input electrical impedance.
A MEMS switch is also enabled by the platform process. We designed the MEMS switch
using FEM and fabricated the MEMS switch and CMUT side by side. Static characterization was
performed in air and dynamic characterization was performed in immersion. The MEMS switch
has a dc switching voltage of 68 V and could be operated using a control voltage of 2.5 V. The
optimum switching time is 1.34 µs and release time is 80 ns.
Glass substrate transparency enables applications where acoustics and optics are combined. A
CMUT element with improved transparency was first fabricated and used for backward-mode
photoacoustic imaging application. A graphite target and a tube filled with ICG solution were
imaged and the effect of silicon plate absorption was discussed. Besides, CMUTs for consumer
electronics applications
is appearing as another
important aspect
for
the CMUT
commercialization. We developed a fully transparent CMUT by adhesively bonding two ITO
coated glass substrates. The device shows 60%-80% optical transmission across the visible
wavelength range and the input electrical impedance measurements demonstrated the functionality
of the fabricated device. Such a device can be used for integrating ultrasound with flat panel
display, including embedding parametric speakers under a TV screen, and integrating finger
printer sensors under a smartphone display panel.
132
BIBILOGRAPHY
[1] K. F. Graff, "Ultrasonics: historical aspects," Proc. IEEE Ultrason. Symp., 1977.
[2] M. I. Haller and B. T. Khuri-Yakub, "A surface micromachined electrostatic ultrasonic air
transducer," IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 43, no. 1, pp. 1 -- 6,
1996.
[3]
Frederick V. Hunt, Electroacoustics: The Analysis of Transduction, and Its Historical
Background. 1954.
[4] Y. Huang, A. S. Ergun, E. Haeggström, and M. H. Badi, "Fabricating Capacitive
Micromachined Ultrasonic Transducers With Wafer-Bonding Technology," J.
Microelectromech. Syst., vol. 12, no. 2, pp. 128 -- 137, 2003.
[5] C. H. Cheng, E. M. Chow, X. Jin, S. Ergun, and B. T. Khuri-Yakub, "An efficient
electrical addressing method using through-wafer vias for two-dimensional ultrasonic
arrays," in Proc. IEEE Ultrason. Symp., 2000, pp. 1179 -- 1182.
[6] X. Zhuang, A. S. Ergun, Y. Huang, I. O. Wygant, Ö. Oralkan, and B. T. Khuri-Yakub,
"Integration of trench-isolated through-wafer interconnects with 2D capacitive
micromachined ultrasonic transducer arrays.," Sens. Actuators. A, Phys., vol. 138, no. 1,
pp. 221 -- 229, 2007.
[7] D. T. Yeh, Ö. Oralkan, I. O. Wygant, A. Sanli Ergun, J. H. Wong, and B. T. Khuri-Yakub,
133
"High-resolution imaging with high-frequency 1-D linear CMUT arrays," in Proc. IEEE
Ultrason. Symp., 2005, vol. 1, no. c, pp. 665 -- 668.
[8] A. S. Logan, L. L. Wong, and J. T. W. Yeow, "2-D CMUT Wafer Bonded Imaging Arrays
with a Row-Column Addressing Scheme," in Proc. IEEE Ultrason. Symp., 2009, pp. 984 --
987.
[9]
J. Knight, J. McLean, and F. L. Degertekin, "Low temperature fabrication of immersion
capacitive micromachined ultrasonic transducers on silicon and dielectric substrates,"
IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 51, no. 10, pp. 1324 -- 1333, Oct.
2004.
[10] M. Bellaredj, G. Bourbon, V. Walter, P. Le Moal, and M. Berthillier, "Anodic bonding
using SOI wafer for fabrication of capacitive micromachined ultrasonic transducers," J.
Micromech. Microeng., vol. 24, no. 2, p. 25009, 2014.
[11] F. Y. Yamaner, X. Zhang, and Ö. Oralkan, "A three-mask process for fabricating vacuum-
sealed capacitive micromachined ultrasonic transducers using anodic bonding," IEEE
Trans. Ultrason. Ferroelect., Freq. Contr., vol. 62, no. 5, pp. 972 -- 982, 2015.
[12] Z. Li, L. L. P. Wong, A. I. H. Chen, S. Na, J. Sun, and J. T. W. Yeow, "Fabrication of
capacitive micromachined ultrasonic transducers based on adhesive wafer bonding
technique," J. Micromech. Microeng., vol. 26, no. 11, p. 115019, 2016.
[13] X. Zhang, O. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "CMUTs on glass with ITO
134
bottom electrodes for improved transparency," in Proc. IEEE Ultrason. Symp., 2016, pp.
1 -- 4.
[14] Ö. Oralkan, M. Karaman, U. Demirci, K. Kaviani, T. H. Lee, and B. T. Khuri-yakub,
"Capacitive micromachined ultrasonic transducers: Next-generation arrays for acoustic
imaging?," IEEE Trans. Ultrason. Ferroelect., Freq. Contr., vol. 49, pp. 1596 -- 1610,
2002.
[15] A. Bhuyan et al., "Integrated circuits for volumetric ultrasound imaging with 2-D CMUT
arrays.," IEEE Trans. Biomed. Circuits Syst., vol. 7, no. 6, pp. 796 -- 804, Dec. 2013.
[16]
I. O. Wygant et al., "Integration of 2D CMUT arrays with front-end electronics for
volumetric ultrasound imaging.," IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol.
55, no. 2, pp. 327 -- 42, Feb. 2008.
[17] D. T. Yeh, O. Oralkan, I. O. Wygant, M. O'Donnell, and B. T. Khuri-Yakub, "3-D
ultrasound imaging using a forward-looking CMUT ring array for
intravascular/intracardiac applications," IEEE Trans. Ultrason. Ferroelectr. Freq.
Control, vol. 53, no. 6, pp. 1202 -- 1211, Jun. 2006.
[18] G. Gurun et al., "Single-chip CMUT-on-CMOS front-end system for real-time volumetric
IVUS and ICE imaging.," IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 61, no.
2, pp. 239 -- 50, Feb. 2014.
[19] J. Zahorian et al., "Monolithic CMUT-on-CMOS integration for intravascular ultrasound
135
applications," IEEE Trans. Ultrason. Ferroelect., Freq. Contr., vol. 58, no. 12, pp. 2659 --
2667, 2011.
[20] S. Vaithilingam et al., "Three-dimensional photoacoustic imaging using a two-
dimensional CMUT array.," IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 56,
no. 11, pp. 2411 -- 9, Nov. 2009.
[21] S. Kothapalli, T.-J. Ma, S. Vaithilingam, Ö. Oralkan, B. T. Khuri-Yakub, and S. S.
Gambhir, "Deep Tissue Photoacoustic Imaging Using a Miniaturized 2-D Capacitive
Micromachined Ultrasonic Transducer Array," IEEE Tran. Biomed. Eng., vol. 59, no. 5,
pp. 1199 -- 1204, 2012.
[22] F. O. R. I. Release, "www.hitachi.com/New/cnews/month/2017/03/170302a.pdf," 2017.
[23] D. Zhao, S. Zhuang, and R. Daigle, "A Commercialized High Frequency CMUT Probe for
Medical Ultrasound Imaging," in Proc. IEEE Ultrason. Symp., 2015, pp. 3 -- 6.
[24] S. H. Wong, M. Kupnik, R. D. Watkins, K. Butts-pauly, and B. T. P. Khuri-yakub,
"Capacitive Micromachined Ultrasonic Transducers for Therapeutic Ultrasound
Applications," IEEE Trans Biomed Eng., vol. 57, no. 1, pp. 114 -- 123, 2010.
[25] J. H. Jang et al., "Dual-Mode Integrated Circuit for Imaging and HIFU With 2-D CMUT
Arrays," in Proc. IEEE Ultrason. Symp., 2015, pp. 3 -- 6.
[26] H. J. Lee, K. K. Park, and M. Kupnik, "Chemical Vapor Detection Using a Capacitive
136
Micromachined," Anal. Chem., pp. 9314 -- 9320, 2011.
[27] Q. Stedman, K. Park, and B. T. Khuri-yakub, "Distinguishing Chemicals Using CMUT
Chemical Sensor Array and Artificial Neural Networks," in Proc. IEEE Ultrason. Symp.,
2014, pp. 162 -- 165.
[28] M. M. Mahmud et al., "A low-power gas sensor for environmental monitoring using a
capacitive micromachined ultrasonic transducer," in Proc. IEEE Sensors Conf., 2014, pp.
677 -- 680.
[29]
I. O. Wygant et al., "50 kHz capacitive micromachined ultrasonic transducers for
generation of highly directional sound with parametric arrays," IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, vol. 56, no. 1, pp. 193 -- 203, 2009.
[30] N. Lamberti, G. Caliano, A. Iula, and A. Stuart, "Sensors and Actuators A : Physical A
high frequency cMUT probe for ultrasound imaging of fingerprints," Sensors Actuators A.
Phys., vol. 172, no. 2, pp. 561 -- 569, 2011.
[31] P. Fesenko, "Capacitive micromachined ultrasonic transducer ( cMUT ) for biometric
applications Capacitive micromachined ultrasonic transducer ( cMUT ) for biometric
applications," Master Thesis, 2012.
[32] T. I. Chiu, H. C. Deng, S. Y. Chang, and S. B. Luo, "Implementation of ultrasonic
touchless interactive panel using the polymer-based CMUT array," in Proc. IEEE Sensors
Conf., 2009, pp. 625 -- 630.
137
[33] W. H. Peake, "The lowest resonant frequency of a water-loaded circular plate," J. Acoust.
Soc. Am., vol. 26, pp. 166 -- 168, 1954.
[34] J. D. Irwin, Basic Engineering Circuit Analysis. 1996.
[35] G. G. Yaralioglu, A. S. Ergun, B. Bayram, E. Haeggstrom, and B. T. Khuri-Yakub,
"Calculation and measurement of electromechanical coupling coefficient of capacitive
micromachined ultrasonic transducers," IEEE Trans. Ultrason. Ferroelect., Freq. Contr.,
vol. 50, no. 4, pp. 449 -- 456, Apr. 2003.
[36] K. K. Park, H. Lee, M. Kupnik, and B. T. Khuri-yakub, "Fabrication of Capacitive
Micromachined Ultrasonic Transducers via Local Oxidation and Direct Wafer Bonding,"
Microelectromechanical Syst. J., vol. 20, no. 1, pp. 95 -- 103, 2011.
[37] B. T. Khuri-Yakub and Ö. Oralkan, "Capacitive micromachined ultrasonic transducers for
medical imaging and therapy.," J. Micromech. Microeng., vol. 21, no. 5, pp. 54004 --
54014, 2011.
[38] M. Kupnik, S. Vaithilingam, K. Torashima, I. O. Wygant, and B. T. Khuri-yakub,
"CMUT Fabrication Based on a Thick Buried Oxide Layer," in Proc. IEEE Ultrason.
Symp., 2010, pp. 547 -- 550.
[39] S. E. Alper and T. Akin, "A Single-Crystal Silicon Symmetrical and Decoupled MEMS
Gyroscope on an Insulating Substrate," J. Microelectromech. Syst., vol. 14, no. 4, pp.
707 -- 717, 2005.
138
[40] S. E. Alper and T. Akin, "Symmetrical and decoupled nickel microgyroscope on
insulating substrate," in The 17th European Conference on Solid-State Transducers, 2004,
vol. 115, pp. 336 -- 350.
[41] R. C. Meier, J. Höfflin, and V. Badilita, "Resonance A fully MEMS-compatible process
for 3D high aspect ratio micro coils obtained with an automatic wire bonder," J.
Micromech. Microeng., vol. 20, no. 15021, 2009.
[42] F. Y. Yamaner, S. Olçum, H. K. Oğuz, A. Bozkurt, H. Köymen, and A. Atalar, "High-
power CMUTs: design and experimental verification.," IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, vol. 59, no. 6, pp. 1276 -- 84, Jun. 2012.
[43] P. C. Eccardt, K. Niederer, and B. Fischer, "Micromachined transducers for ultrasound
applications," 1997 IEEE Ultrason. Symp. Proceedings. An Int. Symp. (Cat.
No.97CH36118), vol. 2, pp. 1609 -- 1618, 1997.
[44] D. F. Lemmerhirt et al., "A 32 x 32 capacitive micromachined ultrasonic transducer array
manufactured in standard CMOS.," IEEE Trans. Ultrason. Ferroelect., Freq. Contr., vol.
59, no. 7, pp. 1521 -- 36, 2012.
[45] Y. Tsuji, M. Kupnik, B. T. Khuri-yakub, and A. P. Flow, "Low Temperature Process for
CMUT Fabrication with Wafer Bonding Technique," in Proc. IEEE Ultrason. Symp.,
2011, pp. 1 -- 4.
[46]
I. O. Wygant et al., "Integration of 2D CMUT arrays with front-end electronics for
139
volumetric ultrasound imaging.," IEEE Trans. Ultrason. Ferroelect., Freq. Contr., vol.
55, no. 2, pp. 327 -- 42, Feb. 2008.
[47] R. Wodnicki et al., "Multi-row linear cMUT array using cMUTs and multiplexing
electronics," in Proc. IEEE Ultrason. Symp., 2009, pp. 2696 -- 2699.
[48] A. Nikoozadeh et al., "Forward-Looking Intracardiac Imaging Catheters Using Fully
Integrated CMUT Arrays," in Proc. IEEE Ultrason. Symp., 2010, vol. 2, pp. 770 -- 773.
[49] S. Weichel et al., "Low-temperature anodic bonding to silicon nitride," Sensors Actuators,
A Phys., vol. 82, no. 1, pp. 249 -- 253, 2000.
[50] G. W. Hsieh, C. H. Tsai, and W. C. Lin, "Anodic bonding of glass and silicon wafers with
an intermediate silicon nitride film and its application to batch fabrication of SPM tip
arrays," Microelectronics J., vol. 36, no. 7, pp. 678 -- 682, 2005.
[51] T. M. H. Lee, D. H. Y. Lee, C. Y. N. Liaw, A. I. K. Lao, and I. M. Hsing, "Detailed
characterization of anodic bonding process between glass and thin-film coated silicon
substrates," Sensors Actuators, A Phys., vol. 86, no. 1 -- 2, pp. 103 -- 107, 2000.
[52] S. Mack, H. Baumann, U. Gösele, H. Werner, and R. Schlögl, "Analysis of Bonding-
Related Gas Enclosure in Micromachined Cavities Sealed by Silicon Wafer Bonding," J.
Electrochem. Soc., vol. 144, no. 3, pp. 1106 -- 1111, 1997.
[53]
I. Ladabaum et al., "Silicon substrate ringing in microfabricated ultrasonic transducers,"
140
in Proc. IEEE Ultrason. Symp., 2000, no. C.
[54] F. Y. Yamaner, X. Zhang, and O. Oralkan, "Fabrication of anodically bonded capacitive
micromachined ultrasonic transducers with vacuum-sealed cavities," in IEEE
International Ultrasonics Symposium, 2014, pp. 604 -- 607.
[55] X. Zhang, F. Y. Yamaner, O. Adelegan, and Ö. Oralkan, "Design of high-frequency
broadband CMUT arrays," in Proc. IEEE Ultrason. Symp., 2015, pp. 1 -- 4.
[56] P. C. Eccardt, K. Niederer, and B. Fischer, "Micromachined transducers for ultrasound
applications," in Proc. IEEE Ultrason. Symp., 1997, vol. 2, pp. 1609 -- 1618.
[57] C. Daft, S. Calmes, D. da Graca, K. Patel, P. Wagner, and I. Ladabaum, "Microfabricated
ultrasonic transducers monolithically integrated with high voltage electronics," in Proc.
IEEE Ultrason. Symp., 2004, vol. 1, pp. 493 -- 496.
[58] R. A. Noble et al., "Low-temperature micromachined cMUTs with fully-integrated
analogue front-end electronics," in Proc. IEEE Ultrason. Symp., 2002, vol. 0, no. c, pp.
1045 -- 1050.
[59] A. Nikoozadeh et al., "Forward-looking volumetric intracardiac imaging using a fully
integrated CMUT ring array," in Proc. IEEE Ultrason. Symp., 2009, pp. 511 -- 514.
[60] C. H. Cheng, A. S. Ergun, and B. T. Khuri-Yakub, "Electrical through-wafer
interconnects with sub-picofarad parasitic capacitance," in Proc. IEEE Int. Conf. MEMS,
141
2001, pp. 18 -- 21.
[61] M. Topper et al., "3-D Thin film interposer based on TGV (Through Glass Vias): An
alternative to Si-interposer," in Proc. IEEE Electr. Comp. Tech. Conf., 2010, pp. 66 -- 73.
[62] M. Esashi, "Wafer level packaging of MEMS," J. Micromech. Microeng., vol. 18, no. 7,
p. 73001, 2008.
[63] N. Toan, S. Hahng, Y. Song, and T. Ono, "Fabrication of vacuum-sealed capacitive
micromachined ultrasonic transducer arrays using glass reflow process," Micromachines,
vol. 7, no. 5, p. 76, 2016.
[64] R. Mukhiya et al., "Fabrication of capacitive micromachined ultrasonic transducer arrays
with isolation-trenches using anodic wafer bonding," IEEE Sensors J., vol. 15, no. 9, pp.
5177 -- 5184, 2015.
[65] S. Takahashi, K. Horiuchi, K. Tatsukoshi, M. Ono, N. Imajo, and T. Mobley,
"Development of through glass via (TGV) formation technology using electrical
discharging for 2.5/3D integrated packaging," in Proc. IEEE Electr. Comp. Tech. Conf.,
2013, pp. 348 -- 352.
[66] Q. Yan, H. Lei, Y. Chen, Y. Zhu, and Research, "Preparation and polishing performances
of α-Al2O3 /Fe2O3 composite particles," in Proc. CIST2008 & ITS-IFToMM2008, 2007,
vol. 14, no. 5, pp. 945 -- 950.
142
[67] R. Keusseyan and T. Mobley, "Hermetically sealed glass packages," in Proc. IMAPS Int.
Symp. on Microelectronics, 2015, pp. 375 -- 378.
[68] X. Zhang, F. Y. Yamaner, and Ö. Oralkan, "Fabrication of capacitive micromachined
ultrasonic transducers with through-glass-via interconnects," in Proc. IEEE Ultrason.
Symp., 2015, pp. 1 -- 4.
[69] S. Enderling et al., "Sheet resistance measurement of non-standard cleanroom materials
using suspended Greek cross test structures," IEEE Trans. Semicond. Manuf., vol. 19, no.
1, pp. 2 -- 9, 2006.
[70] G. G. Yaralioglu, S. A. Ergun, and B. T. Khuri-Yakub, "Finite-element analysis of
capacitive micromachined ultrasonic transducers," IEEE Trans. Ultrason. Ferroelect.,
Freq. Contr., vol. 52, no. 12, pp. 2185 -- 2198, 2005.
[71] X. Zhang, F. Y. Yamaner, and Ö. Oralkan, "Fabrication of Vacuum-Sealed Capacitive
Micromachined Ultrasonic Transducers With Through-Glass-Via Interconnects Using
Anodic Bonding," J. Microelectromechanical Syst., vol. 26, no. 1, pp. 226 -- 234, Feb.
2017.
[72] J. Camacho and C. Fritsch, "Protection circuits for ultrasound applications," IEEE Trans.
Ultrason. Ferroelect., Freq. Contr., vol. 55, no. 5, pp. 1160 -- 1164, 2008.
[73] N. C. Chaggares, R. K. Tang, A. N. Sinclair, F. S. Foster, K. Haraieciwz, and B.
Starkoski, "Protection circuitry and time resolution in high frequency ultrasonic NDE,"
143
Proc. IEEE Ultrason. Symp., vol. 1, pp. 819 -- 822, 1999.
[74] R. E. Mihailovich et al., "MEM relay for reconfigurable RF circuits," IEEE Microw.
Wirel. Components Lett., vol. 11, no. 2, pp. 53 -- 55, 2001.
[75] J. J. Yao and M. F. Chang, "A Surface Micromachined Miniature Switch For
Telecommunications Applications With Signal Frequencies From DC Up To 4 Ghz,"
Proc. Int. Solid-State Sensors Actuators Conf. - TRANSDUCERS '95, vol. 2, pp. 384 -- 387,
1995.
[76] G. M. Rebeiz and J. B. Muldavin, "RF MEMS switches and switch circuits," IEEE
Microw. Mag., vol. 2, no. 4, pp. 59 -- 71, 2001.
[77] F. Y. Yamaner, "Finite element and equivalent circuit modeling of capacitive
micromachined ultrasonic transducer (cMUT)," Master Thesis, 2006.
[78] W. D. Greason, "Analysis of the effect of ESD on the operation of MEMS," IEEE Trans.
Ind. Appl., vol. 45, no. 6, pp. 2185 -- 2191, 2009.
[79] X. Zhang, X. Wu, O. J. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "A MEMS T/R switch
embedded in CMUT structure for ultrasound imaging frontends," IEEE Trans. Ultrason.
Ferroelect., Freq. Contr., pp. 1 -- 4, 2017.
[80] X. Zhang, O. J. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "A fast-switching (1.35-µs),
low-control-voltage (2.5-V) MEMS T/R switch monolithically integrated with a
144
capacitive micromachined ultrasonic transducer (CMUT)," J. Microelectromechanical
Syst., accepted, 2017.
[81] A. Fortini, M. I. Mendelev, S. Buldyrev, and D. Srolovitz, "Asperity contacts at the
nanoscale: Comparison of Ru and Au," J. Appl. Phys., vol. 104, no. 7, 2008.
[82] H. Kwon et al., "Investigation of the electrical contact behaviors in Au-to-Au thin-film
contacts for RF MEMS switches," J. Micromech. Microeng., vol. 18, p. 105010, 2008.
[83] Z. Yang, D. J. Lichtenwalner, A. S. Morris, J. Krim, and A. I. Kingon, "Comparison of Au
and Au-Ni alloys as contact materials for MEMS switches," J. Microelectromech. Syst.,
vol. 18, no. 2, pp. 287 -- 295, 2009.
[84] J. Dhennin, F. Courtade, C. Dieppedal, P. Pons, X. Lafontan, and R. Plana,
"Characterization of gold/gold, gold/ruthenium, and ruthenium/ruthenium ohmic contacts
in MEMS switches improved by a novel methodology," J. Micro/Nanolithography,
MEMS, MOEMS, vol. 9, no. 4, p. 41102, 2010.
[85] R. Fisher et al., "Reconfigurable arrays for portable ultrasound," in Proc. IEEE Ultrason.
Symp., 2005, vol. 1, no. c, pp. 495 -- 499.
[86] J. Chen, M. Wang, J. C. Cheng, Y. H. Wang, P. C. Li, and X. Cheng, "A photoacoustic
imager with light illumination through an infrared-transparent silicon CMUT array," IEEE
Trans. Ultrason. Ferroelect., Freq. Contr., vol. 59, no. 4, pp. 766 -- 775, 2012.
145
[87] G. Brodie, Y. Qiu, S. Cochran, G. Spalding, and M. MacDonald, "Optically transparent
piezoelectric transducer for ultrasonic particle manipulation," IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, vol. 61, no. 3, pp. 389 -- 391, 2014.
[88] H. Kim et al., "Electrical, optical, and structural properties of indium -- tin -- oxide thin films
for organic light-emitting devices," J. Appl. Phys., vol. 86, no. 11, p. 6451, 1999.
[89] M. Xu and L. V Wang, "Photoacoustic imaging in biomedicine," Rev. Sci. Instrum., vol.
77, no. 41101, 2006.
[90] L. V Wang and J. Yao, "A practical guide to photoacoustic tomography in the life
sciences," Nat. Methods, vol. 13, no. 8, pp. 627 -- 638, 2016.
[91] S. Vaithilingam et al., "Three-dimensional photoacoustic imaging using a two-
dimensional CMUT array," IEEE Trans. Ultrason. Ferroelect., Freq. Contr., vol. 56, no.
11, pp. 2411 -- 2419, 2009.
[92] X. Wu, J. Lunsford Sanders, Ö. Oralkan, and D. N. Stephens, "Photoacoustic-Imaging-
Based Temperature Monitoring for High-Intensity Focused Ultrasound Therapy," in Proc
IEEE Eng. Med. Biol. Soc, 2016, pp. 3235 -- 3238.
[93] S. A. Ermilov et al., "Development of laser optoacoustic and ultrasonic imaging system
for breast cancer utilizing handheld array probes," in Proc. SPIE, 2009, vol. 7177, p.
717703.
146
[94] H. F. Zhang, K. Maslov, G. Stoica, and L. V. Wang, "Functional photoacoustic
microscopy for high-resolution and noninvasive in vivo imaging," Nat. Biotechnol., vol.
24, no. 7, pp. 848 -- 851, 2006.
[95] Y. Asao et al., "Photoacoustic mammography capable of simultaneously acquiring
photoacoustic and ultrasound images," J. Biomed. Opt., vol. 21, no. 11, p. 116009, 2016.
[96] S. Manohar, A. Kharine, J. C. G. van Hespen, W. Steenbergen, and T. G. van Leeuwen,
"The Twente Photoacoustic Mammoscope: system overview and performance.," Phys.
Med. Biol., vol. 50, no. 11, pp. 2543 -- 57, 2005.
[97] B. Chen, F. Chu, X. Liu, Y. Li, J. Rong, and H. Jiang, "AlN-based piezoelectric
micromachined ultrasonic transducer for photoacoustic imaging," Appl. Phys. Lett., vol.
103, no. 3, pp. 1 -- 4, 2013.
[98] A. Nikoozadeh et al., "Photoacoustic imaging using a 9F microLinear CMUT ICE
catheter," in Proc. IEEE Ultrason. Symp., 2012, pp. 24 -- 27.
[99] A. Nikoozadeh et al., "An integrated Ring CMUT array for endoscopic ultrasound and
photoacoustic imaging," in Proc. IEEE Ultrason. Symp., 2013, pp. 1178 -- 1181.
[100] J. J. Niederhauser, M. Jaeger, M. Hejazi, H. Keppner, and M. Frenz, "Transparent ITO
coated PVDF transducer for optoacoustic depth profiling," Opt. Commun., vol. 253, no.
4 -- 6, pp. 401 -- 406, 2005.
147
[101] X. Chen, R. Chen, Z. Chen, J. Chen, K. K. Shung, and Q. Zhou, "Transparent lead
lanthanum zirconate titanate (PLZT) ceramic fibers for high-frequency ultrasonic
transducer applications," Ceram. Int., vol. 42, no. 16, pp. 18554 -- 18559, 2016.
[102] P. V. Chitnis, O. Aristizábal, E. Filoux, A. Sampathkumar, J. Mamou, and J. Ketterling,
"Coherence-Weighted Synthetic Focusing Applied to Photoacoustic Imaging Using a
High-Frequency Annular-Array Transducer.," Ultras. Imaging, vol. 38, no. 1, pp. 32 -- 43,
2016.
[103] E. Zhang, J. Laufer, and P. Beard, "Backward-mode multiwavelength photoacoustic
scanner using a planar Fabry Perot polymer film ultrasound sensor for high resolution
three-dimensional imaging of biological tissues," Appl. Opt., vol. 47, no. 4, pp. 561 -- 577,
2008.
[104] E. Zhang and J. Laufer, "In vivo high-resolution 3D photoacoustic imaging of superficial
vascular anatomy," Phys. Med. Biol., vol. 54, 2009.
[105] M. Karaman, L. Pai-Chi, and M. O'Donnell, "Synthetic aperture imaging for small scale
systems," IEEE Trans. Ultrason. Ferroelect., Freq. Contr., vol. 42, no. 3, pp. 429 -- 442,
May 1995.
[106] P. C. Li and M. L. Li, "Adaptive imaging using the generalized coherence factor," IEEE
Trans. Ultrason. Ferroelect., Freq. Contr., vol. 50, no. 2, pp. 128 -- 141, 2003.
[107] M. Xu and L. V. Wang, "Universal back-projection algorithm for photoacoustic computed
148
tomography," Phys. Rev. E, vol. 71, no. 1, pp. 1 -- 7, 2005.
[108] O. Ratib, "OSIRIX: An Open Source Platform for Advanced Multimodality Medical
Imaging," in 4th Int. Conf. Information and Communications Technology, 2006, pp. 1 -- 2.
[109] D. Sette et al., "Transparent piezoelectric transducers for large area ultrasonic actuators,"
in Proc. IEEE Int. Conf. MEMS, 2017, pp. 793 -- 796.
[110] G. Luo et al., "High fill factor piezoelectric micromachined ultrasonic transducers on
transparent substrates," TRANSDUCERS - Int. Solid-State Sensors, Actuators
Microsystems Conf., pp. 1053 -- 1056, 2017.
[111] S. Chen, "Optical Microring Resonators for Photoacoustic Imaging and Detection," PhD
Thesis, Univ. Michigan, 2012.
[112] E. Z. Zhang, "2D backward-mode photoacoustic imaging system for NIR (650-1200nm)
spectroscopic biomedical applications," in Proc. SPIE, 2006, vol. 6086, p. 60860H --
60860H -- 8.
[113] D.-C. Pang and Y.-H. Chiang, "A transparent capacitive micromachined ultrasonic
transducer (CMUT) array for finger hover-sensing dial pads," TRANSDUCERS - Int.
Solid-State Sensors, Actuators Microsystems Conf., pp. 2171 -- 2174, 2017.
[114] A. I. Chen, L. L. P. Wong, S. Na, Z. Li, M. Macecek, and J. T. W. Yeow, "Fabrication of
a Curved Row -- Column Addressed Capacitive Micromachined Ultrasonic Transducer
149
Array," J. Microelectromechanical Syst., vol. 25, no. 4, pp. 675 -- 682, 2016.
[115] X. Zhang, X. Wu, O. J. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "Photoacoustic
Imaging With Backside Illumination Through a CMUT With Improved Transparency,"
IEEE Trans. Ultrason. Ferroelectr. Freq. Control, DOI 10.1109/TUFFC.2017.2774283,
2017. (IEEE early access article available)
[116] X. Zhang, O. J. Adelegan, F. Y. Yamaner, and Ö. Oralkan, "An optically transparent
capacitive micromachined ultrasonic transducer (CMUT) fabricated using SU-8 or BCB
adhesive wafer bonding," in Proc. IEEE Ultrason. Symp., 2017, pp. 1 -- 4.
APPENDICES
150
Appendix A: Finite element modeling of the MEMS switch (ANSYS 17.2)
Static Analysis
151
152
153
Transient Analysis (with a control signal rising edge of 300 ns and fall edge of 300 ns)
154
155
156
157
158
159
Appendix B: Image reconstruction code for backward-PAI with universal back-projection
and coherence factor weighting for graphite target (MATLAB 2017a)
160
clear;
close all;
fileName = 'C: \M2.csv';
ascan = csvread(fileName);
fc = 1.2709e6;
flow = 0.15e6;
fcutoff = 4.5e6;
fs = 50e6;
c = 1490;
lamda = c/fc;
stepSize = 0.3e-3;
scanDistance = 12.9e-3;
numberOfElements = round(scanDistance/stepSize+1);
elementPosition = -0.5*scanDistance+(0:(numberOfElements-1))*stepSize;
imagingStartDepth = 0e-3;
imagingEndDepth = 32e-3;
imagingWidth = scanDistance;
delta = lamda/16;
numberOfPixelsX = ceil(imagingWidth/delta);
numberOfPixelsZ = ceil((imagingEndDepth-imagingStartDepth)/delta);
pixelPositionsX = -0.5*imagingWidth + (0:(numberOfPixelsX-1))*delta;
pixelPositionsZ = imagingStartDepth + (0:(numberOfPixelsZ-1))*delta;
numberOfSamples = size(ascan,2);
ascanBackProj = ascan;
for m = 1:numberOfElements;
for n = 2:numberOfSamples
ascanBackProj(m,n) = ascan(m,n) - (n-1)*(ascan(m,n)-ascan(m,n-1));
end
end
numberOfZeros = 50;
lpf = fir1(2*numberOfZeros,[flow/(fs/2),fcutoff/(fs/2)]);
filteredAscan = zeros(size(ascan,1),size(ascan,2)+numberOfZeros*2);
161
filteredAscanBackProj = zeros(size(ascan,1),size(ascan,2)+numberOfZeros*2);
for m = 1:numberOfElements;
temp = conv(lpf,ascan(m,:));
filteredAscan(m,:) = hilbert(temp);
temp = conv(lpf,ascanBackProj(m,:));
filteredAscanBackProj(m,:) = hilbert(temp);
end
hahahaha = 15*pi/2/99;
threshold = cos(hahahaha);
pic = zeros(numberOfPixelsZ,numberOfPixelsX);
coherence = zeros(numberOfPixelsZ,numberOfPixelsX);
picBackProj = zeros(numberOfPixelsZ,numberOfPixelsX);
coherenceBackProj = zeros(numberOfPixelsZ,numberOfPixelsX);
for m = 1:numberOfPixelsX
for n = 1:numberOfPixelsZ
x = pixelPositionsX(m);
z = pixelPositionsZ(n);
pixelValue = 0;
b = 0;
ha = 0;
pixelValueBackProj = 0;
bBackProj = 0;
for l = 1:numberOfElements
elementPos = elementPosition(l);
distance = norm([x,z]-[elementPos,0]);
cosValue = z/distance;
if(cosValue<threshold)continue;end
index = round(distance/c*fs)+numberOfZeros;
if(index<=size(filteredAscan,2))
pixelValue = pixelValue + filteredAscan(l,index);
b = b + abs(filteredAscan(l,index))^2;
pixelValueBackProj = pixelValueBackProj + filteredAscanBackProj(l,index);
bBackProj = bBackProj + abs(filteredAscanBackProj(l,index))^2;
ha = ha+1;
end
end
pic(n,m) = pixelValue;
coherence(n,m) = abs(pixelValue)^2/b/ha;
162
picBackProj(n,m) = pixelValueBackProj;
coherenceBackProj(n,m) = abs(pixelValueBackProj)^2/bBackProj/ha;
end
end
dyn = 40;
envRealData = abs(pic);
envRealDataCF = envRealData.*coherence;
envRealDatadB = 20*log10(envRealData/max(max(envRealData)));
envRealDatadB(envRealDatadB<-dyn) = -dyn;
envRealDataCFdB = 20*log10(envRealDataCF/max(max(envRealDataCF)));
envRealDataCFdB(envRealDataCFdB<-dyn) = -dyn;
envRealDataBackProj = abs(picBackProj);
envRealDataBackProjCF = envRealDataBackProj.*coherenceBackProj;
envRealDataBackProjdB = 20*log10(envRealDataBackProj/max(max(envRealDataBackProj)));
envRealDataBackProjdB(envRealDataBackProjdB<-dyn) = -dyn;
envRealDataBackProjCFdB = 20*log10(envRealDataBackProjCF/max(max(envRealDataBackProjCF)));
envRealDataBackProjCFdB(envRealDataBackProjCFdB<-dyn) = -dyn;
[m1,n1] = find(envRealDatadB==0);
disp(['In real data, the 0dB pixel is at X:',num2str(pixelPositionsX(n1)*1000),'mm,
Z:',num2str(pixelPositionsZ(m1)*1000),'mm']);
[m2,n2] = find(envRealDataCFdB==0);
disp(['In real data with coherence factor, the 0dB pixel is at
X:',num2str(pixelPositionsX(n2)*1000),'mm, Z:',num2str(pixelPositionsZ(m2)*1000),'mm']);
[m3,n3] = find(envRealDataBackProjdB==0);
disp(['In real data with back projection, the 0dB pixel is at
X:',num2str(pixelPositionsX(n3)*1000),'mm, Z:',num2str(pixelPositionsZ(m3)*1000),'mm']);
[m4,n4] = find(envRealDataBackProjCFdB==0);
disp(['In real data with back projection with coherence factor, the 0dB pixel is at
X:',num2str(pixelPositionsX(n4)*1000),'mm, Z:',num2str(pixelPositionsZ(m4)*1000),'mm']);
figure(1)
imagesc(pixelPositionsX*10^3,pixelPositionsZ*10^3,envRealDatadB,[-dyn,0]);
xlabel('mm','fontsize',12,'fontweight','bold');
ylabel('mm','fontsize',12,'fontweight','bold');
title(['Real data: angle:
',num2str(hahahaha/pi*180),'^{\circ}'],'fontsize',12,'fontweight','bold');
set(gca,'linewidth',1.5,'fontsize',12,'fontweight','bold');
163
colormap('gray');
h = colorbar;
set(h,'fontsize',12,'fontweight','bold');
truesize;
figure(2)
imagesc(pixelPositionsX*10^3,pixelPositionsZ*10^3,envRealDataCFdB,[-dyn,0]);
xlabel('mm','fontsize',12,'fontweight','bold');
ylabel('mm','fontsize',12,'fontweight','bold');
title(['Real data with coherence factor: angle:
',num2str(hahahaha/pi*180),'^{\circ}'],'fontsize',12,'fontweight','bold');
set(gca,'linewidth',1.5,'fontsize',12,'fontweight','bold');
colormap('gray');
h = colorbar;
set(h,'fontsize',12,'fontweight','bold');
truesize;
figure(3)
imagesc(pixelPositionsX*10^3,pixelPositionsZ*10^3,envRealDataBackProjdB,[-dyn,0]);
xlabel('mm','fontsize',12,'fontweight','bold');
ylabel('mm','fontsize',12,'fontweight','bold');
title(['Real data with back projection: angle:
',num2str(hahahaha/pi*180),'^{\circ}'],'fontsize',12,'fontweight','bold');
set(gca,'linewidth',1.5,'fontsize',12,'fontweight','bold');
colormap('gray');
h = colorbar;
set(h,'fontsize',12,'fontweight','bold');
truesize;
figure(4)
imagesc(pixelPositionsX*10^3,pixelPositionsZ*10^3,envRealDataBackProjCFdB,[-dyn,0]);
xlabel('mm','fontsize',12,'fontweight','bold');
ylabel('mm','fontsize',12,'fontweight','bold');
title(['Real data with back projection and coherence factor: angle:
',num2str(hahahaha/pi*180),'^{\circ}'],'fontsize',12,'fontweight','bold');
set(gca,'linewidth',1.5,'fontsize',12,'fontweight','bold');
colormap('gray');
h = colorbar;
set(h,'fontsize',12,'fontweight','bold');
truesize;
|
1808.01312 | 1 | 1808 | 2018-06-07T18:51:30 | Optimization approach for optical absorption in three-dimensional structures including solar cells | [
"physics.app-ph",
"physics.optics"
] | The rigorous coupled-wave approach (RCWA) and the differential evolution algorithm (DEA) were coupled in a practicable approach to maximize absorption in optical structures with three-dimensional morphology. As a model problem, optimal values of four geometric parameters and the bandgaps of three i-layers were found for an amorphous-silicon, multi-terminal, thin-film tandem solar cell comprising three p-i-n junctions with a metallic hexagonally corrugated back-reflector. When the optical short-circuit current density was chosen as the figure of merit to be maximized, only the bandgap of the topmost i-layer was significant and the remaining six parameters played minor roles. While this configuration would absorb light very well, it would have poor electrical performance. This is because the optimization problem allows for the thicknesses and bandgaps of the semiconductor layers to change. We therefore devised another figure of merit that takes into account bandgap changes by estimating the open-circuit voltage. The resulting configuration was found to be optimal with respect to all seven variable parameters. The RCWA+DEA optimization approach is applicable to other types of photovoltaic solar cells as well as optical absorbers, with the choice of the figure of merit being vital to a successful outcome. | physics.app-ph | physics |
Optimization approach for optical absorption in three-dimensional structures including solar
Benjamin J. Civilettia, Tom H. Andersona, Faiz Ahmadb, Peter B. Monka, and Akhlesh Lakhtakiab
aUniversity of Delaware, Department of Mathematical Sciences, 501 Ewing Hall, Newark, DE 19716, USA
bPennsylvania State University, Department of Engineering Science and Mechanics,
NanoMM -- Nanoengineered Metamaterials Group, University Park, PA 16802, USA
cells
Abstract
The rigorous coupled-wave approach (RCWA) and the differential evolution algorithm (DEA) were
coupled in a practicable approach to maximize absorption in optical structures with three-dimensional
morphology. As a model problem, optimal values of four geometric parameters and the bandgaps of three
i-layers were found for an amorphous-silicon, multi-terminal, thin-film tandem solar cell comprising three
p-i-n junctions with a metallic hexagonally corrugated back-reflector. When the optical short-circuit
current density was chosen as the figure of merit to be maximized, only the bandgap of the topmost
i-layer was significant and the remaining six parameters played minor roles. While this configuration
would absorb light very well, it would have poor electrical performance. This is because the optimization
problem allows for the thicknesses and bandgaps of the semiconductor layers to change. We therefore
devised another figure of merit that takes into account bandgap changes by estimating the open-circuit
voltage. The resulting configuration was found to be optimal with respect to all seven variable parameters.
The RCWA+DEA optimization approach is applicable to other types of photovoltaic solar cells as well
as optical absorbers, with the choice of the figure of merit being vital to a successful outcome.
1 INTRODUCTION
Three items are needed to numerically optimize the design of an optical absorber such as a thin-film solar cell.
The first item is a fast solver that can predict the performance of the device in a variety of configurations.
The second item is an optimization code that can mitigate the effect of local minima without excessive
computational time. The third item is a figure of merit that adequately captures the desired performance
characteristics of the device so that a good design will emerge. Some of these choices are explored in this
paper.
The rigorous coupled-wave approach (RCWA) [1, 2] can be used to model the optical performance of a
thin-film optical absorbers, as has been shown for solar cells with metallic back-reflectors that are periodically
corrugated along one direction [3, 4, 5]. Indeed, the RCWA provides accurate results with high computational
speed for boundary-value problems involving structures that are invariant only along, say, the y axis and
therefore are quasi-two-dimensional [6, 7]. Furthermore, the RCWA can be coupled with the differential
evolution algorithm (DEA) [8, 9] for optimization [5]. However, the computational requirements of RCWA
increase significantly when the back-reflector is periodically corrugated in two directions, i.e., the boundary-
value problem is fully three-dimensional (3D) in nature [10, 11]. Additional design parameters enter the
optimization process thereby to increase the computational burden further.
The dimensions of the unit cell of an optical absorber with a PCBR directly affect optical absorption
[5]. Optimization of thin-film solar cells with two-dimensionally corrugated back-reflectors for maximum
absorption has not been reported heretofore, to our knowledge. As a preliminary study showed that it
is becoming a practicable proposition with commonly available computational resources, we decided to
implement the RCWA+DEA approach to optimize a fully 3D absorbing structure [12].
To demonstrate this approach, we report here the maximization of optical absorption in an idealized thin-
film tandem solar cell fabricated over a periodically corrugated back-reflector (PCBR) with hillock-shaped
corrugations arranged on a hexagonal lattice. The active region of the chosen solar cell comprises three
electrically isolated p-i-n junctions. The semiconductor layers were taken to have the bandgap-dependent
optical properties of amorphous silicon.[13] Silver [14] is a good choice for the PCBR because its plasmonic
nature can be harnessed to launch surface-plasmon-polariton (SPP) waves inside the device and thereby
enhance the optical electric field and optical absorption [15, 16, 17]. With the foregoing choices, our results
indicate that maximization of the optical short-circuit current density, the standard figure of merit,[18, 19, 20]
1
does not result in a desirable design. Instead, we found that the maximum power density is a better figure
of merit.
The plan of this paper is as follows. The optical boundary-value problem that is solved to determine
the spectrally integrated number of absorbed photons per unit volume per unit time Nph is presented in
Sec. 2. The numerical techniques adopted for this work are presented in Sec. 3: the three-dimensional
implementation of the RCWA is briefly described in Sec. 3.1, Sec. 3.2 contains the diagonalization of a
matrix that emerges in the RCWA implementation, and Sec. 3.3 briefly describes the DEA. Numerical
results are provided in Sec. 4, Sec. 4.3 discusses the convergence of the numerical methods, while Sec. 4.4
briefly compares these numerical results to a thin-film tandem solar cell with a bi-sinusoidal PCBR. Closing
remarks are presented in Sec. 5.
k0 = 2π/λ0, ω = k0c0, and η0 = (cid:112)µ0/ε0, respectively, where λ0 is the free-space wave length, µ0 is the
√
ε0µ0 is the speed of light in free
permeability of free space, ε0 is the permittivity of free space, and c0 = 1/
space. Vectors are underlined, column vectors and matrices associated with the RCWA are in boldface with
breve notation, and the Cartesian unit vectors are identified as x, y, and z. The imaginary unit is denoted
by i =
The free-space wavenumber, angular frequency, and intrinsic impedance of free space are denoted by
√−1.
2 MODEL BOUNDARY-VALUE PROBLEM
We considered the boundary-value problem shown schematically in Fig. 1, which also defines the thicknesses
Ld, Lg, and Lm. The device occupies the region
X : {(x, y, z) − ∞ < x < ∞,−∞ < y < ∞, 0 < z < Ld + Lg + Lm} ,
(1)
with the half spaces z < 0 and z > Ld + Lg + Lm 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 < Ld + Lg + Lm}, the back-reflector (which
also functions as an electrode in a solar cell) being doubly periodic with period Lx along the x axis and
period Ly along the y axis.
The region 0 < z < Ld comprises an antireflection window and three p-i-n junctions which are electrically
isolated from each other by two windows, as shown in Fig. 1(a). The relative permittivity εd(x, y, z, λ0) of
this multilayered material depends on λ0. The layers are identified in the figure. All windows are made of a
material of relative permittivity εw(λ0). The λ0-dependent relative permittivity of each semiconductor layer
depends on the bandgap chosen for that layer.
The region Ld + Lg < z < Ld + Lg + Lm is occupied by a metal with relative 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. The unit cell in the xy plane was chosen to
form a two-dimensional rectangular lattice that is equivalent to a hexagonal lattice. If the side of the regular
hexagons in this lattice is denoted by Lh, then Lx = Lh and Ly =
3Lh for the rectangular lattice.
√
The grating region is defined by hillocks arranged as in Fig. 1(b). Each hillock is a frustum of a sphere
of radius Rsph. The base of the hillock is a circle of radius Rg and the height of the hillock equals Lg, as
shown in Fig. 1(a); accordingly,
(cid:19)
(cid:18) R2
g
Lg
1
2
Rsph =
+ Lg
.
(2)
(3)
(4)
The intersection of the plane z = zg ∈ [Ld, Ld + Lg] and a hillock is a circle of radius
(cid:113)
R =
(zg − Ld)[2Rsph − (zg − Ld)] .
The relative permittivity εg(x, y, z, λ0) = εg(x ± Lx, y ± Ly, z, λ0) in the grating region is
εg(x, y, z, λ0) = εm(λ0) − [εm(λ0) − εw(λ0)] U(x, y, z)
2
(a)
(b)
Figure 1: (a) Schematic of the model boundary-value problem in the plane y = 0. The n-type semiconductor
layers are blue, the p-type semiconductor layers are red, and the i-type are gray. White regions are occupied
by a material with real relative permittivity εw(λ0).
(b) Schematic of the grating region in the plane
z = zg ∈ [Ld, Ld + Lg].
with
(cid:26) 1, D ≥ R
0, D < R
U(x, y, z) =
(5)
where D is the minimum distance between the point (x, y, z) and the centers (0, 0, zg), (Lx, 0, zg), (Lx/2, Ly/2, zg),
(Lx, Ly, zg) and (0, Ly, zg) of the circle and four quarter circles in Fig. 1(b).
3 Numerical Techniques Used
3.1 Rigorous coupled-wave approach
The RCWA was used the calculate the electric field phasor everywhere inside the chosen device as a result
of illumination by a monochromatic plane wave normally incident on the plane z = 0 from the half space
z < 0. The electric field phasor of the incident plane wave was taken as
√
where Eo = 4
15π V m−1.
Einc(x, y, z, λ0) = Eo
x + y√
2
exp (ik0z) ,
(6)
As a result of the PCBR being doubly periodic, the x- and y-dependences of the electric and magnetic
field phasors have to be represented everywhere by Fourier series as
(cid:104)
(cid:16)
(cid:17)(cid:105)
E(x, y, z, λ0) =
i
x x + k(n)
k(m)
y y
(7)
m=∞(cid:88)
n=∞(cid:88)
m=−∞
n=−∞
e(m,n)(z, λ0) exp
3
and
m=∞(cid:88)
n=∞(cid:88)
m=−∞
n=−∞
(cid:16)
(cid:104)
i
x x + k(n)
k(m)
y y
(cid:17)(cid:105)
,
h(m,n)(z, λ0) exp
H(x, y, z, λ0) =
(8)
x = m(2π/Lx), k(n)
where k(m)
h(m,n)
x
where is represented by the Fourier series
x + h(m,n)
y + h(m,n)
y
z
y = n(2π/Ly), and e(m,n) = e(m,n)
z as well as h(m,n) =
z are Fourier coefficients. Likewise, the relative permittivity εrel(x, y, z, λ0) every-
x + e(m,n)
y + e(m,n)
x
y
z
εrel(x, y, z, λ0) =
ε(m,n)
rel
(z, λ0) exp
i
x x + k(n)
k(m)
y y
,
(9)
m=∞(cid:88)
n=∞(cid:88)
m=−∞
n=−∞
(cid:104)
(cid:16)
(cid:17)(cid:105)
where ε(m,n)
curl postulates yields the matrix ordinary differential equation [21, Chap. 3]
(z, λ0) are Fourier coefficients. Substitution of Eqs. (7) -- (9) in the frequency-domain Maxwell
rel
d
dz
f (z, λ0) = i P(z, λ0) • f (z, λ0),
(10)
where the column vector f (z, λ0) contains the x- and y-directed components of the Fourier coefficients of the
electric and magnetic field phasors.
Detailed descriptions of the algorithm to solve Eq. (10) are available elsewhere [12, 21]. In brief, com-
putational tractability requires the foregoing expansions to be truncated to include only m ∈ {−Mt, ..., Mt}
and n ∈ {−Nt, ..., Nt}, with Mt ≥ 0 and Nt ≥ 0. The region R is partitioned into a sufficiently large number
of slices along the z axis. Each slice is taken to be homogeneous along the z axis but may be periodically
nonhomogeneous along the x and y axes. Thus, the matrix P(z, λ0) is assumed to be piecewise uniform
in z. Boundary conditions are enforced on the planes z = 0 and z = Ld + Lg + Lm to match the fields
to the incident, reflected, and transmitted fields, as appropriate. A stable marching algorithm is then used
to determine the Fourier coefficients of the electric and magnetic field phasors in each slice[21]. Finally,
the z-directed components of the electric and magnetic field phasors in the device can be obtained through
algebraic equations arising during the derivation of Eq. (10). Thus, the electric field phasor E(x, y, z, λ0)
can be determined throughout the solar cell.
The spectrally integrated number of absorbed photons per unit volume per unit time is given by
(cid:90) λ0max
λ0min
(cid:12)(cid:12)(cid:12)(cid:12) E(x, y, z, λ0)
Eo
(cid:12)(cid:12)(cid:12)(cid:12)2
Nph(x, y, z) =
1
c0
Im{εrel(x, y, z, λ0)}
S(λ0) dλ0 ,
(11)
where is the reduced Planck constant, S(λ0) is the AM1.5G solar spectrum [22], λ0min = 400 nm, and
λ0max = 1100 nm. With the assumption that the absorption of every photon in a semiconductor layer
releases an electron-hole pair, the charge-carrier-generation rate G(x, y, z) equals Nph(x, y, z), which can be
determined at any location in the nine semiconductor layers. The integral on the right side of Eq. (11) was
approximated using the trapezoidal rule [23] with the integrand uniformly sampled every 10 nm.
λ0 ∈ (cid:2)λ0min , λ0max
(cid:3) [21, Sec. 3.8]. As the relative permittivity is not uniform in any slice in the grating
3.2 Diagonalization of P(z, λ0)
The numerically stable RCWA algorithm requires that P(z, λ0) be diagonalizable [24] in each slice for every
region, the built-in function eig in Matlab R(cid:13) (version R2016b) was used to compute the eigenvalues and
eignenvectors of P. In all other slices, the eigenvalues and eigenvectors of P were determined analytically,
to increase the computational speed.
A superindex
τ = m(2Nt + 1) + n , m ∈ [−Mt, Mt] , n ∈ [−Nt, Nt] ,
(12)
4
0,··· , 0,(cid:0)g±
τ
(cid:1)−1(cid:26)
v1±
τ
=
(cid:17)2(cid:27)
(cid:16)k(τ )
x
ωµ0 −
1
ωε0εrel
τ ∈ {−τt,··· , τt} .
, 0,··· , 0,− k(τ )
x k(τ )
ωε0εrelg±
y
τ
, 0,··· , 0, 1, 0,··· , 0
(cid:34)
(cid:34)
,
(14)
(cid:35)T
(cid:35)T
is defined for convenience, where τt = 2MtNt + Mt + Nt.[21] In any slice in which the relative permittivity
εrel(x, y, z, λ0) is independent of x and y, the 2(2τt + 1) distinct eigenvalues of P are given by
(cid:113)
τ = ±
g±
0 εrel −(cid:0)k(τ )
x
(cid:1)2 −(cid:0)k(τ )
y
(cid:1)2
k2
τ ∈ {−τt,··· , τt} ,
,
(13)
with each eigenvalue having an geometric multiplicity of 2, k(τ )
eigenvectors are
x = k(m)
x
, and k(τ )
y = k(n)
y . Half of the 4(2τt +1)
In the column vector on the right side of Eq. (14), the non-zero entries occur in the following locations
(counting from the top): τ , τ + 2τt + 1, and τ + 6τt + 3. The remaining 2(2τt + 1) eigenvectors are
, 0,··· , 0,−(cid:0)g±
τ
(cid:1)−1(cid:26)
(cid:17)2(cid:27)
(cid:16)k(τ )
y
ωµ0 −
1
ωε0εrel
, 0,··· , 0, 1, 0,··· , 0
v2±
τ
=
0,··· , 0,
k(τ )
x k(τ )
ωε0εrelg±
y
τ
τ ∈ {−τt,··· , τt} .
(15)
The non-zero entries occur in the column vector on the right side of Eq. (15) in the following locations
(counting from the top): τ , τ + 2τt + 1, and τ + 4τt + 2.
3.3 Differential evolution algorithm
To maximize the figure of merit C : S ⊂ R ¯N → R, we wish to find an optimal point V opt ≡ {v1, v2,··· , v ¯N} ∈
We employed the DEA to maximize a figure of merit over a variety of optical and electrical parameters
numbering ¯N . The DEA has been used previously[5] for seeking optimal designs of PCBRs that are invariant
along the y axis, as the algorithm is well-suited to search a large space of candidate solutions. The number
of candidate solutions depends on the number of parameters and discretization of parameter ranges for the
optimization.
S, where S is a search space of all possible parameter combinations.
We note that the DEA requires that all parameter ranges be discretized, so the search is conducted over a
finite number of possible outcomes. The DEA requires specification of the crossover probability CR ∈ (0, 1),
a differential weight α ∈ (0, 2) and the number of random points NP . Details of the algorithm have been
provided elsewhere[5, 8, 9].
The DEA is very useful for solving complicated optimization problems, but it does not guarantee con-
vergence to a global extremum [25]. However, stochastic sampling of the search space helps to avoid local
maxima.
4 NUMERICAL RESULTS AND DISCUSSION
In this paper we have chosen to maximize optical absorption. The figures of merit defined later in this section
take into account all optical effects such as the excitation of SPP waves and waveguide modes [26]. This
allows a tradeoff between the various optical phenomena without prejudicing one mechanism over another.
However, the parameter space is chosen so that the excitation of SPP waves and waveguide modes can be
supported.
For all numerical results in this paper, the window layers were chosen to be made of alumimum-doped
zinc oxide (AZO). For the two windows between p-i-n junctions, the thicknesses were fixed so that d2d =
d1d = 20 nm. The relative permittivity εw(λ0) of AZO was taken from a standard source[27]. The minimum
5
thickness of the PCBR was fixed at Lm = 150 nm. The metal was chosen to be silver, whose relative
permittivity εm also depends on λ0[14].
The bandgaps E(cid:96)i, (cid:96) ∈ {1, 2, 3}, of the i-layers in the triple-junction solar cell were kept variable in the
range [1.3, 1.95] eV, but their thicknesses d1i = d2i = d3i = 200 nm were kept fixed. The thicknesses of
all three n-layers and all three p-layers were also kept fixed: d(cid:96)n = d(cid:96)p = 20 nm. The bandgaps of all
three n-layers were fixed as E(cid:96)n = 1.8 eV, (cid:96) ∈ {1, 2, 3}. The bandgaps of the p-layers were fixed as follows:
E1p = 1.8 eV and E2p = E3p = 1.95 eV.
By introducing C or Ge into the lattice, a new material is formed, but the i-layers are still of the a-
Si:H GeC family. This process changes the bandgap, where the λ0-dependent relative permittivity of the
material is obtained by an analytical model.[17, 13]. The electrical properties of the material also change,
and can be found by applying Vegard's law to known values.[28] Since the electrical properties have no effect
on our optical model, they have no role in this study.
Furthermore, the lattice parameter Lh ∈ [200, 800] nm, the antireflection window's thickness dw ∈
[10, 130] nm, the base radius Rg ∈ [10, 400] nm, and the corrugation height Lg ∈ [0, 300] nm were al-
lowed to vary. Thus, the dimension of the search space S was ¯N = 7, and we sought an optimal design over
a candidate space of 9 × 1011 possible configurations. We used parameter values CR = 0.7, α = 0.8, and
NP = 70 for optimization.
4.1 Optimization for optical short-circuit current density
The figure of merit C for the DEA optimization was initially chosen to be standard figure of merit for optical
modeling of solar cells[29]: the optical short-circuit current density
(cid:90)(cid:90)(cid:90)
J Opt
SC =
qe
LxLy
Nph(x, y, z) dx dy dz ,
Rsc
(16)
Plots of J Opt
where Rsc is the portion of the reference unit cell R occupied by the nine semiconductor layers in the solar
cell. and qe = 1.6 × 10−19 C is the elementary charge. This figure of merit will maximize the number of
photons absorbed in the solar cell, but disregards all electrical properties. In this section we proceed to show
that this results in a poor design.
SC against the bandgaps E(cid:96)i, (cid:96) ∈ {1, 2, 3}, for points sampled by the optimization exercise are
presented in Fig. 2. In this figure as well as in Fig. 3, the data points from DEA are projected onto the plane
formed by the variable being investigated and the figure of merit. The bandgap of the topmost i -- layer is
the most influential parameter that controls J Opt
SC , the DEA minimized
the bandgap in this layer, and we see that many parameter sets with Eopt
3i = 1.3 eV at the boundary of the
constraint set were evaluated. As E3i increases from 1.3 eV to 1.95 eV, J Opt
SC decreases throughout most of
this interval as shown in Fig. 2(a).
SC showed that minimizing E3i was the most important factor compared with the
remaining ¯N − 1 variable parameters (i.e., E1i, E2i, Lh, dw, Rg, and Lg). This phenomenon is clearly
exemplified in Fig. 2(a), where for any fixed value of E3i, the remaining ¯N − 1 variable parameters contribute
only to a ±2 mA cm−2 variation in J Opt
SC . Unfortunately, such a configuration is likely to be electrically
inefficient: with a narrow bandgap, more charge carriers are excited, but the operating voltage of the solar
cell will be reduced. [30] Contrast this to Figs. 2(b) and (c) wherein the variations of J Opt
SC with the bandgaps
E2i and E1i, respectively, are shown. We see that while J Opt
SC is maximum when E1i = E2i = 1.35 eV, there
SC ranging from 18 mA cm−2 to 27 mA cm−2.
are values of J Opt
SC . While attempting to maximize J Opt
Maximization of J Opt
Parenthetically, when the thicknesses of the i-layers were included as variables in an optimization exercise,
the DEA simply focused on the maximization of those thicknesses. The resulting configuration would also
have poor electrical performance. While this type of solar cell would absorb more light, an excited charge
carrier would have to travel further to reach an electrode, thereby increasing recombination and decreasing
efficiency. We note that Fig. 2 demonstrates the choice of figure of merit is very important. We only
included these results to contrast them with numerical results in Fig. 5, since all the chosen parameters in
the optimization exercise should affect the solar-cell performance.
6
(b)
(a)
(c)
Figure 2: J Opt
by DEA as the algorithm progresses. Larger values of J Opt
SC in relation to (a) E3i, (b) E2i, and (c) E1i. Each marker (·) represents a choice of parameters
SC are desirable.
7
4.2 Optimization for maximum power density
In order to improve the optimal design of the chosen solar cell without including a full electrical model, we
devised a new figure of merit for the DEA which penalizes the effect of minimizing the bandgap of any of
the i-layers. We defined the power density
3(cid:88)
(cid:90)(cid:90)(cid:90)
1
Psup =
E(cid:96)i
LxLy
(cid:96)=1
Nph(x, y, z) dx dy dz
R(cid:96)
(17)
as the new figure of merit, with R(cid:96) ⊂ R being the region occupied by the (cid:96)-th p-i-n junction. Let us note that
Psup is a theoretical upper bound on the maximum extractable power density of the solar cell. Furthermore,
the summation over the index (cid:96) indicates that the tandem solar cell is to be configured in the multi-terminal
format. Note that this power density is computed solely from the absorption of photons and the material
bandgap. This estimates the maximum electrical power density but does not involve any electrical modeling
(e.g., recombination and mobility of electrons and holes). A useful extension of our approach would be to
include an electrical model, but that extension lies beyond the scope of this paper.
Figures 3 and 4 show the results of the DEA optimization for the variable geometric and bandgap
parameters of the solar cell -- namely, the lattice parameter Lh, the antireflection-window thickness dw, the
duty cycle ζ = Rg/Lh, the corrugation height Lg, and the bandgaps E1i, E2i, and E3i. All seven of these
parameters influence the figure of merit Psup defined by Eq. (17). Indeed, a steady increase in Psup is seen
in Fig. 3(a) on the interval 200 ≤ Lh ≤ 600 nm; a very sharp increase in Psup is evident in Fig. 3(b) as
dw increases from 10 nm to 80 nm; Psup increases steeply in Fig. 3(c) as ζ approaches 0.35 with a drop off
thereafter; and Psup peaks in the neighborhood of Lg = 200 nm in Fig. 3(d). Also shown is the hillock base
radius Rg in Fig. 4(d). We note the contrast between the behavior seen in Fig. 2(a) and Fig. 4(c). In the
latter optimization exercise, the effect of the other parameters amounts to a variation of ±5 mW cm−2.
The optimal values found are as follows: E1i = 1.35 eV, E2i = 1.95 eV, E3i = 1.65 eV, Lh = 642 nm,
dw = 89 nm, ζ = 0.36, and Lg = 231 nm. These values yielded a maximum Psup = 43.66 mW cm−2. These
optimal parameters, found by maximizing Psup, contrast sharply with our findings in Sec. 4.1, wherein
SC was dominated by the minimization of E3i with the remaining ¯N − 1 parameters
maximization of J Opt
having very little effect.
The optimization exercise yielded two distinct categories of unit cells with relatively high Psup values.
The first comprises configurations for which Lh lies in the interval [200, 300] nm, and the second comprises
configurations for which Lh ∈ [550, 650] nm. For the first category, the optimal value of Lg is approximately
40 nm with a base radius Rg of 50 nm. The second type of configuration has much deeper corrugations, with
Psup maximized when Lg takes values near 200 nm and Rg around 230 nm. This phenomenon is evidenced
by two distinct peaks in Fig. 3(a) -- (d) and Fig. 4(d).
4.3 Convergence of RCWA and DEA
To ensure convergence of the optical short-circuit density, a representative configuration for the unit cell was
used to determine an appropriate choice of Nt and Mt. We let Nt vary in the set {2, 3, 4, 5, 6}, and defined
Mt = (cid:100)√
3Nt(cid:101), where (cid:100)·(cid:101) is the ceiling function. After determining that Nph changed by ≤ 1% for two
successive values of Nt and Mt, the number of Fourier modes were fixed for all numerical results reported in
this paper, and taken to be Nt = 2, Mt = 4. Since Eq. (16) has the charge-carrier generation rate integrated
over the nine semiconductor layers, the susceptibility of J Opt
SC to the effect of Gibbs' phenomenon[6] on the
electric field in the region z ∈ [0, Ld] is negligible. Hence, the electric field converges everywhere in the
semiconductor layers, even for relatively small values of Nt and Mt. We performed 50 DEA iterations in
the optimization exercise. Figure 5 shows the convergence of Psup with the number of DEA iterations. We
see that the best value of Psup does not change after the first 32 DEA iterations. As with any stochastic
optimization method, it is always possible that further iteration would result in improvement of the computed
maximum power density.
8
(a)
(b)
(c)
(d)
Figure 3: Psup in relation to (a) Lh, (b) dw, (c) ζ = Rg/Lh, and Lg. Each marker (·) represents a choice
of parameters by DEA as the algorithm progresses. The red marker indicates the maximum value of Psup
achieved.
9
(a)
(b)
(c)
(d)
Figure 4: Psup in relation to (a) E1i, (b) E2i, (c) E3i, and Rg. Each marker (·) represents a choice of
parameters by DEA as the algorithm progresses. The red marker indicates the maximum value of Psup
achieved.
10
Figure 5: The best value of Psup versus number of DEA iterations when optimizing for maximum power
density as in Sec. 4.2.
4.4 Comparison to bi-sinusoidal PCBR
A third DEA-based optimization exercise was performed with the PCBR taken to be bi-sinusoidally corru-
gated [11]. The relative permittivity εg(x, y, z, λ0) = εg(x ± Lx, y ± Ly, z, λ0) in the grating region is
εg(x, y, z, λ0) = εm(λ0) − [εm(λ0) − εw(λ0)] U(z − g1(x)) U(z − g2(x)),
with the corrugation-shape functions
(cid:2)1 − sin(2π
(cid:2)1 − sin(2π
)(cid:3),
)(cid:3).
x
Lx
y
Ly
g1(x) = Ld + Lg
g2(x) = Ld + Lg
(18)
(19)
(20)
The DEA parameters were kept the same as in Sec. 4, and we set Lx = Ly for the computation. The
optimization exercise yielded an optimal configuration with dw = 87 nm, Lg = 110 nm, Lx = 607 nm,
Rg = 212 nm, E3i = 1.7 eV, E2i = 1.55 eV and E1i = 1.3 eV. The maximal power density achieved
was Psup = 42.84 mW cm−2, which is slightly lower than 43.66 mW cm−2 obtained for the hexagonally
corrugated PCBR in Sec. 4.2. However, the average Psup over all configurations visited by the DEA for the
bi-sinusoidally corrugated PCBR is 38.17 mW cm−2 but 35.06 mW cm−2 for the hexagonally corrugated
PCBR. There were many more configurations with relatively poor Psup values for the hexagonal case.
5 Closing Remarks
As a model problem to demonstrate the practicability of the RCWA+DEA approach to design efficient
optically absorbing 3D structures, optimal values of four geometric parameters and the bandgaps of three i-
layers were found for an idealized, multi-terminal, thin-film tandem solar cell comprising three p-i-n junctions
with a silver PCBR with hillock-shaped corrugations arranged on a hexagonal lattice. The figure of merit
for the DEA was either (i) the optical short-circuit current density in Sec. 4.1 or (ii) the power density in
Sec. 4.2. Thus, two different optimization exercises using the RCWA+DEA approach where performed.
As the optical short-circuit current density takes into account only the optical constitutive properties of
the solar cell, maximization of that quantity resulted in a poor design. In particular, we determined that
only the bandgap of the topmost i-layer (i.e., E3i) was significant to the optimization of the optical short-
circuit current density, and the remaining six parameters played minor roles. While photon absorption in the
11
topmost p-i-n junction was maximized thereby, this configuration would have poor electrical performance.
This is because, when the thicknesses of the i-layers were included in the optimization, the DEA simply
focused on the maximization of those thicknesses. Although increasing those thicknesses can enhance light
absorption, the electrical performance may be sacrificed, thereby reducing efficiency. The design of thin-film
solar cells must balance optical and electrical performances [31, 32].
In order to avoid configurations with potentially poor electrical performance, we next used the power
density -- thereby weighting the optical short-circuit current density -- as a new figure of merit. The resulting
configuration was optimal with respect to all seven design parameters. Another optimization exercise was
then performed on a similar tandem solar cell, but with a bi-sinusoidally corrugated PCBR. In this case the
configurationally averaged power density tested by the DEA was about 10% higher than with the hexagonally
corrugated PCBR. In the future, we plan to supplement the optical model by an electrical drift-diffusion
model [32] and then optimize the overall electrical performance of the solar cell.
In closing, let us emphasize that the triple-junction tandem solar cell was chosen as a model problem to
show here the capabilities of the RCWA+DEA approach developed for 3D optically absorbing structures.
Our approach can be extended not only to other types of photovoltaic solar cells [33, 34] but also to optical
absorbers [35, 36] with 3D morphology.
Note. This paper is substantially based on a paper titled, "Optimization of charge-carrier generation in
amorphous-silicon thin-film tandem solar cell backed by two-dimensional metallic surface-relief grating,"
presented at the SPIE Optics and Photonics conference Next Generation Technologies for Solar Energy Con-
version VIII, held August 5 -- 11, 2017 in San Diego, California, United States.
Acknowledgments. The research of B. J. Civiletti, T. H. Anderson, and P. B. Monk is partially supported
by the US National Science Foundation (NSF) under grant number DMS-1619904. The research of F. Ahmed
and A. Lakhtakia is partially supported by the US NSF under grant number DMS-1619901. A. Lakhtakia
thanks the Charles Godfrey Binder Endowment at the Pennsylvania State University for ongoing support of
his research.
References
[1] M. G. Moharam et al., "Formulation for stable and efficient implementation of the rigorous coupled-wave
analysis of binary gratings," J. Opt. Soc. Am. A 12, 1068 -- 1076 (1995).
[2] N. Chateau and J.-P. Hugonin, "Algorithm for the rigorous coupled-wave analysis of grating diffraction,"
J. Opt. Soc. Am. A 11, 1321 -- 1331 (1994).
[3] P. Sheng, A. N. Bloch, and R. S. Stepleman, "Wavelength-selective absorption enhancement in thin-film
solar cells," Appl. Phys. Lett. 43, 579 -- 581 (1983).
[4] C. Heine and R. H. Morf, "Submicrometer gratings for solar energy applications," Appl. Opt. 34,
2476 -- 2482 (1995).
[5] M. Solano et al., "Optimization of the absorption efficiency of an amorphous-silicon thin-film tandem
solar cell backed by a metallic surface-relief grating," Appl. Opt. 52, 966 -- 979 (2013); erratum: 54,
398 -- 399 (2015).
[6] M. Weismann, D. F. G. Gallagher, and N. C. Panoiu, "Accurate near-field calculation in the rigorous
coupled-wave analysis method," J. Opt. (Bristol) 17, 125612 (2015).
[7] M. V. Shuba et al., "Adequacy of the rigorous coupled-wave approach for thin-film silicon solar cells with
periodically corrugated metallic backreflectors: spectral analysis," J. Opt. Soc. Am. A 32, 1222 -- 1230
(2015).
[8] R. Storn and K. Price, "Differential evolution -- a simple and efficient heuristic for global optimization
over continuous spaces," J. Global Optim. 11, 341 -- 359 (1997).
12
[9] K. Price, R. M. Storn, and J. A. Lampinen, Differential Evolution: A Practical Approach to Global
Optimization, Springer, New York, NY, USA (2005).
[10] 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, 1747 -- 1758 (2011).
[11] F. Ahmad et al., "On optical-absorption peaks in a nonhomogeneous thin-film solar cell with a two-
dimensional periodically corrugated metallic backreflector," J. Nanophoton. 12(1), 016017 (2018).
[12] B. J. Civiletti et al., "Optimization of charge-carrier generation in amorphous-silicon thin-film tandem
solar cell backed by two-dimensional metallic surface-relief grating," Proc. SPIE 10368, 1036809 (2017).
[13] A. S. Ferlauto et al., "Analytical model for the optical functions of amorphous semiconductors from
the near-infrared to ultraviolet: Applications in thin film photovoltaics," J. Appl. Phys. 92, 2424 -- 2436
(2002).
[14] E. D. Palik (ed.), Handbook of Optical Constants of Solids, Academic Press, Boston, MA, USA (1985).
[15] L. M. Anderson, "Harnessing surface plasmons for solar energy conversion," Proc. SPIE 408, 172 -- 178
(1983).
[16] M. A. Green and S. Pillai, "Harnessing plasmonics for solar cells," Nat. Photonics 6, 130 -- 132 (2012).
[17] 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, 087106 (2013); errata: 53, 129801 (2014).
[18] S. Zhou et al., "Optimizing two-level hierarchical particles for thin-film solar cells," Opt. Express 21,
A285 -- A294 (2013).
[19] A. Naqavi et al., "Enhanced light trapping in realistic thin film solar cells using one-dimensional grat-
ings," Proc. SPIE 8065, 80650A (2012).
[20] L.C. Andreani, A. Bozzola, and M. Liscidini, "Light trapping in thin-film solar cells: towards the
Lambertian limit," Proc. SPIE 8438, 884380C (2012).
[21] J. A. Polo Jr., T. G. Mackay, and A. Lakhtakia, Electromagnetic Surface Waves: A Modern Perspective,
Elsevier, Waltham, MA, USA (2013).
[22] National Renewable Energy Laboratory, Reference Solar Spectral Irradiance: Air Mass 1.5.
[23] Y. Jaluria, Computer Methods for Engineering, Taylor & Francis, Washington, DC, USA (1996).
[24] H. Hochstadt, Differential Equations: A Modern Approach, Dover Press, New York, NY, USA (1975).
[25] R. Knobloch, J. Mlynek, and R. Srb, "Improving convergence properties of a differential evolution
algorithm," AIP Conf. Proc. 1789, 030005 (2016).
[26] L. Liu et al., "Planar light concentration in micro-Si solar cells enabled by a metallic grating -- photonic
crystal architecture," ACS Photonics 3(4), 604 -- 610 (2016).
[27] 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, 710 -- 714 (2010).
[28] L. Vegard. "Die Rontgenstrahlen im Dienste der Erforschung der Materie," Z. Kristallographie 67(1-6),
239 -- 259 (1928).
[29] N. Anttu et al., "Absorption and transmission of light in III -- V nanowire arrays for tandem solar cell
applications," Nanotechnology 28, 205203 (2017).
13
[30] J.-P. Colinge and C. A. Colinge, Physics of Semiconductor Devices, Kluwer Academic, New York, NY,
USA (2002).
[31] M. Faryad et al., "Optical and electrical modeling of an amorphous-silicon tandem solar cell with
nonhomogeneous intrinsic layers and a periodically corrugated back-reflector," Proc. SPIE 8823, 882306
(2013).
[32] T. H. Anderson et al., "Combined optical-electrical finite-element simulations of thin-film solar cells
with homogeneous and nonhomogeneous intrinsic layers," J. Photon. Energy 6, 025502 (2016).
[33] J. Zhu et al., "Nanodome solar cells with efficient light management and self-cleaning," Nano Lett. 10,
2342 -- 2348 (2010).
[34] R. Kapadia et al., "Nanopillar photovoltaics: Materials, processes, and devices," Nano Energy 1, 132 --
144 (2011).
[35] N. Liu et al., "Infrared perfect absorber and its application as plasmonic sensor," Nano Lett. 10, 1979 --
1984 (2010).
[36] S. Butun and K. Aydin, "Structurally tunable resonant absorption bands in ultrathin broadband plas-
monic absorbers," Opt. Express 22, 19457 -- 19468 (2014).
14
|
1712.09497 | 1 | 1712 | 2017-12-27T05:23:06 | Stable Self-Assembled Atomic-Switch Networks for Neuromorphic Applications | [
"physics.app-ph",
"cond-mat.dis-nn",
"cs.ET"
] | Nature inspired neuromorphic architectures are being explored as an alternative to imminent limitations of conventional complementary metal-oxide semiconductor (CMOS) architectures. Utilization of such architectures for practical applications like advanced pattern recognition tasks will require synaptic connections that are both reconfigurable and stable. Here, we report realization of stable atomic-switch networks (ASN), with inherent complex connectivity, self-assembled from percolating metal nanoparticles (NPs). The device conductance reflects the configuration of synapses which can be modulated via voltage stimulus. By controlling Relative Humidity (RH) and oxygen partial-pressure during NP deposition we obtain stochastic conductance switching that is stable over several months. Detailed characterization reveals signatures of electric-field induced atomic-wire formation within the tunnel-gaps of the oxidized percolating network. Finally we show that the synaptic structure can be reconfigured by stimulating at different repetition rates, which can be utilized as short-term to long-term memory conversion. This demonstration of stable stochastic switching in ASNs provides a promising route to hardware implementation of biological neuronal models and, as an example, we highlight possible applications in Reservoir Computing (RC). | physics.app-ph | physics | Stable Self-Assembled Atomic-Switch Networks for
Neuromorphic Applications
Saurabh K. Bose, Joshua B. Mallinson, Rodrigo M. Gazoni, and Simon A. Brown
2
Abstract
Nature inspired neuromorphic architectures are being explored as an alternative to imminent limitations of
conventional complementary metal-oxide semiconductor (CMOS) architectures. Utilization of such architectures
for practical applications like advanced pattern recognition tasks will require synaptic connections that are both
reconfigurable and stable. Here, we report realization of stable atomic-switch networks (ASN), with inherent
complex connectivity, self-assembled from percolating metal nanoparticles (NPs). The device conductance reflects
the configuration of synapses which can be modulated via voltage stimulus. By controlling Relative Humidity (RH)
and oxygen partial-pressure during NP deposition we obtain stochastic conductance switching that is stable over
several months. Detailed characterization reveals signatures of electric-field induced atomic-wire formation within
the tunnel-gaps of the oxidized percolating network. Finally we show that the synaptic structure can be reconfigured
by stimulating at different repetition rates, which can be utilized as short-term to long-term memory conversion.
This demonstration of stable stochastic switching in ASNs provides a promising route to hardware implementation
of biological neuronal models and, as an example, we highlight possible applications in Reservoir Computing (RC).
Atomic switch networks, Clusters, Neuromorphic architecture
Index Terms
I. INTRODUCTION
T HE astounding success of the von Neumann architecture for computers [1], as encapsulated in
Moore's Law, is now meeting with fundamental limitations (physical transistor dimensions are ap-
proaching classical limits) and practical limitations (the exponential increase in research and development
costs for every new process line) [2], [3]. Natural information processing systems, like the biological
brain, on the other hand, can perform highly complex computational tasks like navigation, recognition
and decision-making with remarkable ease and with very low energy consumption [4]. This natural
computation, processing the useful data (patterns) from a multitude of sensory information, is immediate
and cannot be matched by even the most-advanced supercomputers [5], [6]. Nature inspired architectures
[7]–[13] are therefore currently being pursued as a disruptive alternative to the von Neumann architecture.
A recent review on Neuromorphic architecture [14] and implementations can be found in Ref. [15].
The alternative brain-inspired hardware approach must address three key issues simultaneously: mimic
the complex biological network of neurons, replicate synaptic structures and allow implementation of
standard computational algorithms [16]. Achieving all of these goals is obviously enormously challenging
and will require long-term research. Nevertheless significant progress has been made towards solving
several different problems, using a variety of architectures. Proposals for non-CMOS approaches include
those based on networks of memristors [17]–[19], atomic switches [10], [11] and Metal Oxide Resistive
Random Access Memory (RRAM) [20], [21]. There have been interesting demonstrations of memristor-
based neural networks [22], associative memory [23], synaptic emulators [24], conditional programming
[25], reconfigurable logic [26], solving mazes [27], pattern recognition [28]–[31], and reservoir computing
The authors are with The MacDiarmid Institute for Advanced Materials and Nanotechnology, Department of Physics and Astronomy,
University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand (email:[email protected])
The authors gratefully acknowledge financial support from the Marsden Fund, New Zealand, and the MacDiarmid Institute for Advanced
Materials and Nanotechnology.
This is a post-peer-review, pre-copyedit version of an article published in IEEE Trans. Elect. Dev. 2017. The final authenticated version is
available online at: http://dx.doi.org/10.1109/TED.2017.2766063.
7
1
0
2
c
e
D
7
2
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
7
9
4
9
0
.
2
1
7
1
:
v
i
X
r
a
3
Fig. 1.
(a) Schematic depiction (top and side view) of the nanoparticle network between electrical contacts of our two-electrode devices
showing tunnelling gaps in the Sn NP network. The outer shaded region on the NPs depicts the thin oxide layer. (b) Variation of the onset
of conductance with increase in partial pressure Pdep (over 2 orders of magnitude) during NP deposition. (c) This pressure variation results
in shorter conduction onset times t0 (left scale) with longer onset width (cid:52)t0 (right scale). (d) Scanning electron micrograph of samples
prepared at lowest ambient air pressure (Pdep ∼10 µTorr) shows more coalescence and larger grain-size in comparison to the highest pressure
(600 µTorr) prepared samples shown in (e). (f) Use of dry synthetic air (Pdep ∼10 µTorr) produces a different microstructure with reduced
sample stability as compared with ambient air. The white scale bars are 100 nm.
(RC) [32]–[34]. Even in the most heavily explored architectures (regular cross-bar arrays of memris-
tors) [35] there remain unsolved challenges in regard to realisation of the required properties of both
individual switching elements and networks of these elements.
RC is simpler to implement than many other unconventional computation schemes since synapses do
not need to be addressed individually and can be seen as an important step towards achieving other
types brain-like computation. In RC a 'reservoir' comprising a complex network of switching elements
allows the transformation of input signals into a higher dimensional space. [32]–[34] Training of a
single 'output layer' then allows implementation of various time series prediction, pattern recognition and
classification tasks. [36]–[38] Randomly assembled atomic switch networks (ASNs) based on sulphidised
Ag nanowires [10], [11] and percolating films of nanoparticles [39], [40] are immediately ammenable to
RC. ASNs are also an appealing alternative to regular arrays of devices because they allow realisation
of complexity similar to that of the brain and fabrication via self-assembly immediately circumvent the
limitations of lithographic processing. Ag-based ASNs have recently been used to demonstrate a form of
RC in which the non-linear properties of the reservoir allows generation of target waveforms, and a clear
roadmap towards further implementation has been mapped out [38].
Systems of inorganic synapses are however in the early stages of development with improvements re-
quired in production methods, reliability and actual functionality [35]. While robust switching over 10,000
cycles has been reported for Ag-AgS nanowire systems [10] a different, and a particularly important, issue
for any real-world applications is long-term device stability, which is the main focus of the present work.
Such stability has not been reported previously in either Ag-AgS nanowire [10], [11], [34], [38], [41] or
percolating ASNs [39].
Here we report a straightforward fabrication procedure for realization of randomly connected ASNs
within a percolating network of metal nanoparticles (NPs). We show that deliberate introduction of oxygen
and moisture during NP deposition leads to long term device stability. Despite the presence of oxides,
the switching mechanisms associated with increases(decreases) in device conductance G↑(G↓) are shown
to be formation (destruction) of atomic scale wires in tunnel gaps in the network. The two-terminal
device conductance (G) quantifies the input-output electrode connectivity [40] and reflects the synaptic
(a) (d) Ambient air (10 µTorr)Substrate Ambient air (10 µTorr) (e) Ambient air (600 µTorr)(c)(b)Ambient air (600 µTorr)(c) (b) (f) Ambient air (600 µTorr) Synthetic air (10 µTorr)Synthetic air (10 µTorr) 4
Fig. 2. Stochastic and stable switching behaviour for sample prepared with Pdep = 10 µTorr and high humidity (RH ∼ 80%) ambient
air conditions, showing multi-level conductance switching, induced by application of both triangular and pulsed voltage stimulus. On the
1st day, immediately after sample fabrication, we use voltage-sweeps in order to check the voltage threshold for switching, as discussed in
section IV. On subsequent days we utilize controlled voltage pulses. As we tested the samples in a variety of ways at different times over
a period of several months, the applied voltage in panel (d) is slightly different to that in the other panels. The sample exhibits qualitatively
similar switching for several months of device operation.
configuration of the network which can be reconfigured by voltage stimulation. Finally, we discuss the
observed synaptic stochasticity and why it is useful for implementation of hardware analogues of the
biological brain [42].
II. EXPERIMENTAL
The nano-cluster deposition system used in this study is based on magnetron sputtering to generate
a vapour of the metal of interest and gas aggregation to condense the vapour into particles, and has
been described in detail in previous publications [43]–[45]. The deposition scheme provides a narrow
cluster size distribution [43] and allows precise control over the NP surface coverage near the percolation
threshold [46], so that the system is poised near criticality [47]. Sn NPs with mean diameter ∼ 8.5 nm
are deposited between 50 nm thick Au/NiCr electrodes on Si3N4 substrates, with active area of 100 µm ×
300 µm. The two-contact devices allow for a demonstration of network stability and associated dynamics,
but samples with multiple contacts will be required for demonstration of RC.
The Sn NPs are deposited at room temperature which means that ordinarily the surface atoms have
sufficient mobility to allow coalescence [48]. For samples poised near the percolation threshold, the
coalescence can lead to the loss of conducting pathways through the film, because neighboring particles
that are initially joined by a fragile connection are pulled apart as they coalesce with other neighbor
NPs, thus contributing to the short life-span (∼ hours) of previous devices [39]. In the present work, the
coalescence of the Sn NPs is controlled by partial oxidation (during NP deposition) via a controlled leak
of air with a needle-valve. As will be shown, the controlled oxide formation leads to reduced coalescence
and enhanced device stability. We emphasize that by 'stability' we do not mean that the device has a fixed
conductance, but that the device is in a state in which it continues to exhibit multiple switching events in
response to voltage stimuli.
5
Fig. 3. The normalized distribution of the change in conductance for the switching events (∆G) on day 10 and day 40, for long periods
of time, shows the stability of the switching dynamics. The solid lines are Gaussian fits to the data.
A. Self-assembly of ASNs
III. SAMPLE FABRICATION
We have self-assembled interconnected and active network of atomic-switches as depicted in the
schematic shown in Fig. 1(a). Deposition of the Sn NPs at Base Pressure (BP, ∼ 6µTorr) led to initial
observation of a non-zero conductance (time t0) at around 800s with a sharp onset behaviour ((cid:52)t0) ∼ 10s,
as shown in Fig. 1(b). Here, we define the width of the onset (cid:52)t0 as the time after t0 required to reach a
conductance of 6G0 (G0 = 2e2/h is the quantum of conductance [49]). The cluster deposition is stopped
[arrows in Fig. 1(b)] when G ∼8G0, which represents a nanoparticle surface coverage slightly greater
than the percolation threshold, and has been found experimentally to yield optimal switching behaviour
when Pdep is in the range ∼ 10− 50µTorr, as described below . As can be seen in Fig. 1(c) the increase in
deposition pressure (Pdep) leads to a monotonic decrease in t0 coupled with an increase in (cid:52)t0. The smooth
conduction onset and longer (cid:52)t0 is in contrast to the step-wise conduction onset of samples deposited at BP
(minimum oxidation) in Ref. [39]; those samples were dominated by few quantized conduction pathways
and therefore lacked the large-scale distributed synaptic network essential for neuromorphic applications.
The post-deposition scanning electron micrographs (SEM) shown in Fig. 1(d-e) reveal markedly smaller
NP sizes for devices prepared at higher Pdep. This can be understood in the framework of diffusion of the
tin NPs on the substrate being inhibited by the formation of tin-oxide-shell, leading to reduced coalescence
at higher Pdep. Such formation of oxide-shell is known to inhibit grain-rotation-induced grain coalescence
(GRIGC) [50] thus favoring smaller grain-sizes. This reduced coalescence means that the percolation
threshold is reached more quickly (smaller t0) and the conductance then increases more slowly (larger
(cid:52)t0). After the deposition is stopped [arrows in Fig. 1(b)] the slow conductance change is primarily due
a small amount of further coalescence of the NPs, decreasing G.
B. Optimization of Pressure and Humidity
The samples fabricated with high Relative Humidity (RH ∼ 80%) ambient air have been stimulated
with voltage sweeps and square voltage pulses over several months. Data for a typical sample are shown
in Fig. 2. The four panels show representative snapshots of the switching events on the 1st, 10th, 15th
and 40th day. As described below, the detailed switching behaviour of the network is a complex function
01201(b)Day 10(a)01201Count (Normalized)Day 40G (G0) 6
Fig. 4. Temporal retention of the network conductance tested over more than one week. Left: the device shows switching in response to 4V
pulses before a constant 'read' voltage (0.1V) was applied for 5 days. Right: the conductance showed no measurable change and switching
re-commenced on application of 4V pulses again.
of applied voltage, pulse widths (τp) and history of the inputs, but qualitatively similar conductance
switching is observed for long periods of time. Application of voltage sweep or pulses induces Electric
Field Induced Evaporation (EFIE) and Electric Field Induced Surface Diffusion (EFISD) of the surface
atoms [51], resulting in atomic-wire formation in tunnel gaps in the percolating network (resulting in G↑).
The electronic flow causes electromigration induced opening of the previously connected atomic-wires
[52] resulting in G↓. The conductance thus switches between multiple conductance states with G ∼ 1–
3G0. G → 0 at multiple points, but the electric field induces reconnections and results in the network
configuration returning to the connected regime of non-zero conductance. Samples fabricated in high
humidity conditions (RH ∼ 80%) with ambient air have been tested in this way for periods of several
months without the sample becoming permanently open circuit. The key point is that the samples prepared
with high RH oxidation exhibit stable switching behaviour without significant performance degradation.
For example, Fig. 3 shows the distribution of the change in conductance for switching events (∆G) on
day 10 and 40 (see Fig. 2). Both the mean and variance of the distributions are very similar, indicating
that the average switching behavior is the same. Obviously it is an onerous task to test such samples for
much longer periods and further testing is required to determine the ultimate lifetime of the samples.Both
the mean and variance of the distributions are very similar. Obviously it is an onerous task to test such
samples for much longer periods and further testing is required to determine the ultimate lifetime of
the samples. The same kind of stable switching behaviour is observed for voltages upto (≥ 7-8V) but
application of very high voltages (≥ 10V) causes irreversible breakdown in the devices.
The oxidation of pure Sn into tin-oxides [SnO, SnO2] is well known to be accelerated under humid
conditions. [53] Studies of oxidation of Sn in conditions similar to the present ones show that only partial
surface oxides are formed. [54], [55] Ex-situ analysis of the present oxide-structure is obviously not
feasible and so instead we have investigated in-situ the device stability over several weeks. Fig. 4 shows
that the device conductance did not change measurably when the device was left for 5 days with 0.1V
applied. The conductance switching is resumed when voltage pulses were applied again, which clearly
indicates that the oxide structure did not evolve significantly in this period.
The final microstructure and the associated device stability achieved with NP oxidation depends strongly
on the relative humidity during NP deposition. Therefore, in order to develop a reliable fabrication process,
mitigating day-to-day variation in RH of the ambient air and to build understanding of the critical RH
7
Fig. 5. Samples prepared with Pdep = 10µTorr and sub-optimum moisture content (RH ∼ 55%) can be partially stabilized with a current-
surge protection in the form of a 1kΩ in-line resistor. The top panels in (a) and (b) show the samples prepared without oxidation or moisture
are unstable even with the current surge protection. The samples become open circuit (G → 0) on application of 0.1V (no resistor) and 1V
(with resistor). The middle panel in (a) shows that samples prepared with dry oxygen (no moisture) are only slightly more stable as G → 0
at 1V (no resistor). The equivalent sample with in-line resistor showed stability in the data shown here but G → 0 in the next measurement
(not shown here). The bottom panels show that samples prepared in oxygen and partial humidity (∼ 55 %) are more stable than those in
the middle panels (no moisture). The sample prepared with oxygen + moisture and measured with 1kΩ resistor showing G switching for
several weeks.
necessary for stabilization, a new set of samples were prepared in a more controlled environment using
commercial dry synthetic air coupled with custom-built humidifier (bubbler). The deposition with dry
synthetic air resulted in unstable samples with the corresponding SEM micrographs [shown in 1(f)]
indicating a slightly more coalesced morphology in comparison to the RH ∼ 80% ambient air in 1(d).
Although the microstructure has only subtle differences, these differences become very important as the
NP system is poised near the percolation threshold and nanoscale changes can promote(inhibit) inter-NP
atomic-switch formation.
C. Current surge protection
The variation in conductance during the first voltage sweeps applied to the new series of samples are
shown in Fig. 5. The top panel of Fig. 5(a) shows that the atomic switch networks prepared at BP with
no oxygen and no moisture are disconnected on application of very small voltages of 0.1V. Introduction
of dry air at Pdep = 10µTorr, no moisture [Fig. 5(a) middle panel] leads to samples that can only sustain
small voltage sweeps. Samples prepared with the same Pdep = 10µTorr and higher RH ∼ 55% are more
stable but still do not show sustained reconnections (G↑) and become disconnected at ∼2V, indicating
that these samples are only partially stabilized. This provides an opportunity to demonstrate an additional
method for stabilizing the switching behaviour in these devices.
Fig. 5(b) shows that the devices show more stable conductance switching when measured with a current-
limiting resistor (1kΩ) in series with the device. The series resistor limits the maximum allowed current
flowing through the percolating network and hence prevents the destruction of the key connections via
electromigration. The sample prepared with Pdep = 10µTorr but without moisture and measured with
the series resistor survived the voltage sweeps [Fig. 5(b) middle panel], but got disconnected in the next
measurement (not shown here). In contrast, the sample prepared with oxidation with RH ∼ 55% and
measured with the in-line resistor was stable for several weeks. Further samples prepared in synthetic
air with a higher RH (∼60%) exhibit stable (i.e. for months) switching behaviour as in Fig. 2, without
a current-limiting resistor. This indicates that oxidation with a critical amount of moisture RH (≥60%)
creates a microstructure which incorporates a robust current-limiting resistor backbone, and thus do not
require additional in-series resistor protection.
(a) (b) 8
Fig. 6. Further conductance data of the sample described in Fig. 2 showing that switching in these NP assemblies requires application of a
minimum voltage-stimulus. The 3V pulses generate very few switching events, whereas 4V pulses trigger multitude of switching events.
D. Fabrication summary
As discussed above, the crucial fabrication parameters for ASN stability are Pdep and relative humidity
RH%. Pdep was varied between BP (6 µTorr) and 600 µTorr, and RH% was varied from completely dry
(0%) to nearly saturated (80%). The optimal fabrication parameters for stability of these Sn cluster devices
are Pdep = 10-50 µTorr and relative humidity RH = 60-80% when an in-series resistor of 1kΩ is used
for current surge protection.
IV. SWITCHING MECHANISM AND DYNAMICS
To understand the physical process underlying the switching mechanism, we present further voltage
and time dependent studies. Fig. 6 presents a segment of data acquired during the long sequence of
measurements on the sample used to obtain the data in Fig. 2 (ambient air, Pdep = 10µTorr, RH ∼
80%) showing that a critical voltage (or equivalently, electric field) is required to activate the switching
process. The switching dynamics is voltage polarity independent, with negative V pulses (not shown here)
showing exactly the same switching dynamics as positive V pulses. This polarity-independence allows
us to eliminate other possible switching mechanisms such as Coulomb charging and electrochemical
redox reactions [10], and further substantiates the electric-field and current induced switching mechanism
described here. As shown in Fig. 6, the pulses with amplitude 3V cause almost no switching events,
whereas 4V pulses induce multiple stochastic switching events. The inherently probabilistic nature of
the synaptic connections are clearly visible in the snapshots shown in Fig. 2, where stimulus near
the threshold voltage induces less than one switching event per pulse. Such stochastic or probabilistic
dynamics of the synapses are integral to the functioning of the biological brain: e.g. the opening and
closing of synaptic ion channels and associated transmission of neurotransmitter molecules is inherently
stochastic [42] and is understood to be critical for noise-filtering [56], signal transmission [7] and reward-
modulated Hebbian learning [57]. The existence of a critical stimulus strength (electric field here) for
the stochastic formation (annihilation) of atomic-scale wires in tunnel gaps in the network is consistent
with EFIE/EFISD (electromigration) mechanisms and provides a unique global control over the synaptic
network reconfiguration.
To further validate the model and estimate the effective tunnel barrier parameters associated with the
gap in which the atomic wires are formed, pulsed voltage measurements were stopped when the device
9
(a–b) Slow bipolar voltage sweeps were applied to the sample described in Fig. 2 when G = 0. The tunneling behaviour is evident
Fig. 7.
from the non-linear current-voltage I(V) characteristics e.g. at points A and B. Successive voltage-sweeps increase G by two orders of
magnitude before the jump-to-contact and G → 2G0 at ∼22 min. This newly formed connection is then stable under further voltage sweeps
and shows the expected linear I(V) characteristics. The solid lines in panel (b) are fits to Simmons' tunneling model (1). (c) The calculated
tunnel gap dgap decreases monotonically before an ohmic connection is formed at sweep 12. The I(V) is ohmic after that (e.g. position C).
(d) Schematic depicting the gradual formation of atomic wire in one of the tunnel barrier.
became open circuit (i.e. G < 10−5G0). A series of slow bipolar voltage sweeps were then applied which
showed non-linear current-voltage I(V) characteristics as in Fig. 7(a–b). On the 1st voltage sweep, G jumps
from < 10−5G0 to ∼ 10−2G0, which corresponds to the formation of a tunnel gap which is sufficiently
small to allow a measurable current to flow. The corresponding tunneling current through a non-ideal
potential barrier with height ΦB and width d is [58]:
I ∝ (ΦB − eV /2) exp
−(ΦB + eV /2) exp
(cid:104) − 2(2m)1/2
(cid:104) − 2(2m)1/2
(cid:105)
(cid:105)
α(ΦB − eV /2)1/2d
α(ΦB + eV /2)1/2d
(1)
with m being the free electron mass and α being an adjustable parameter representing the non-
ideal character of the tunneling barrier and effective electron mass. Fig. 7(b) shows representative I(V)
characteristics for the voltage sweeps marked A, B and C in Fig. 7(a) along with the fits to (1) (solid
lines). The associated barrier width d, with calculated barrier height ΦB ∼ 2 eV, decreases monotonically
with the sweep number #, when either an ideal (α = 1) or highly non-ideal (α = 0.5) barrier is assumed.
Interestingly the resultant electric field exceeds the ∼ 1 Vnm−1 threshold for EFISD but remains lower
than the ∼ 25 Vnm−1 required for EFIE [39], [51]. As shown schematically in Fig. 7(d), the narrowing
of the tunnel gap under the influence of the electric field continues, until after about 20 minutes an
atomic scale wire closes the tunnel gap, i.e. a "jump to contact" occurs [59] leading to a conductance
G = 2G0. These atomic-scale wires are similar to those formed in mechanically controlled break junctions
(MCBJs) [59]. The wire breaks and re-forms a couple of times [Fig. 7(a)] and then is observed to be
completely stable when subjected to further voltage sweeps. The Ohmic conductance is marked by linear
I(V) behavior depicted in curve C of Fig. 7(b). Such conductance modulation with successive stimulus
(electric-field) is one of the key requirements for synaptic learning capability in neuromorphic systems
[40], [60], and is similar to the sensory memory reported in Ref. [61].
10
Fig. 8.
(a) Synaptic plasticity dependence on stimulus frequency is depicted in switching behaviour in response to a sequence of voltage
pulses with fixed Vp=4V and variable pulse widths τp (1 – 30s). The conductance remains unaltered for read voltages of 0.1 V as seen in
the flat section (OFF) in middle of the sequence. (b) Schematic depiction of representative synaptic pathways with only very few pathways
shown for clarity. The real device ASN is much more complex. Shorter pulse widths leads to electric-field induced connections as described
in Fig. 7. Formation of additional atomic-wire connections (one depicted here) cause more potentiated synaptic pathways and thus higher
conductance as seen in panel (a).
Realization of neuromorphic behavior in these ASNs also requires a scheme to modify the density of
potentiated synaptic pathways [40]. In Fig. 8 we show such stimulus frequency dependent potentiation.
Square voltage pulses with fixed Vp=4V (just above threshold voltage) with various pulse widths τp (1,
2, 5, 10, 20, and 30 s) are applied successively for 30 mins each (the green lines depict the pulse-
width). Longer τp (slow pulses) leads to lower conductance whereas shorter τp (faster pulses) leads to
additional formation of synaptic pathways as schematically depicted in Fig. 8(b), resulting in higher G.
This variation can be understood in the light of the electric field-induced reconnections dominating over
the electromigration induced disconnection of atomic wires for shorter τp. Such stimulus rate dependent
reconfiguration of network connectivity can be modeled as short-term to long-term memory conversion
[12], [62].
V. CONCLUSION
In summary, we have demonstrated a unique approach for realization of self-assembled atomic switch
networks with stimulus induced control of the synaptic configuration reflected in the device conductance.
By controlling oxidation and humidity during NP deposition, nanoparticle coalescence is inhibited resulting
in stochastic switching that is stable over several months. The atomic-wire formation in these oxidised
nanostructures is very surprising and detailed modeling [63] of the atomistic mechanisms [39], [51], [52]
in the presence of oxides [54], [55] is required, as is atomic scale modelling of the effect of humidity [53]
on the oxidation process.
We have also highlighted the stochastic nature of the switching mechanism together with the stability
of the distribution of switching events - these reflect inherently complex network dynamics that are a
key requirement for neuromorphic applications. The next stage of this research will be to build devices
with multiple contacts and to demonstrate that the networks exhibit the required network dynamics. The
(a)(b)Electrodes (a) (b) Electrodes ShorterPotentiated pathwayAvailable pathwayShorter pulse widths(cid:1) HigherPotentiated pathwayAvailable pathway pulse widths Higher G Potentiated pathway precise requirements are different for different applications [15], [35], but for RC include distributed
spatio-temporal dynamics, recurrency, higher harmonic generation, switching speed, network size and
type of connectivity [10], [32], [33], [38], [64], [65].
More generally, these complex percolating structures mimic some of the features of biological neural
networks and synaptic structures and so could provide a foundation for a range of future neuromorphic
architectures. Future work will focus on utilizing these structures to implement previously suggested
algorithms [32], [33] and hence to demonstrate utility in key applications.
11
The technical support of G. MacDonald and G. Graham are acknowledged.
ACKNOWLEDGMENT
REFERENCES
[1] H. H. Goldstine and Neumann Von. J., "On the principles of large scale computing machines," John von Neumann Collect. Work.,
vol. 5, pp. 1–32, 1946.
[2] Y. Taur, D. A. Buchanan, W. Chen, D. J. Frank, K. E. Ismail, L. Shih-Hsien, G. A. Sai-Halasz, R. G. Viswanathan, H. J. C. Wann,
S. J. Wind, and H. S. Wong, "CMOS scaling into the nanometer regime," Proc. IEEE, vol. 85, no. 4, pp. 486–503, 1997.
[3] D. J. Frank, "Power-constrained CMOS scaling limits," IBM J. Res. Dev., vol. 46, no. 2.3, pp. 235–244, 2002.
[4] E. R. Kandel, J. H. Schwartz, and J. M. Jessell, Principles of Neural Science 4th edn. McGraw-Hill, 2000.
[5] T. Roska, "Cellular Wave Computers for Brain-Like Spatial-Temporal Sensory Computing," IEEE Circuits Syst. Mag., vol. 5, pp. 5–19,
[6] T. M. Wong, R. Preissl, P. Datta, M. D. Flickner, R. Singh, S. K. Esser, E. McQuinn, R. Appuswamy, W. P. Risk, H. D. Simon, and
D. S. Modha, "IBM Research Report 1014," IBM Res. Rep. 1014, vol. 10502, pp. 1–3, 2012.
[7] T. Tuma, A. Pantazi, M. Le Gallo, A. Sebastian, and E. Eleftheriou, "Stochastic phase-change neurons," Nat. Nanotechnol., vol. 11,
2005.
no. 8, pp. 693–699, 2016.
[8] S. K. Bose, C. P. Lawrence, Z. Liu, K. S. Makarenko, R. M. J. van Damme, H. J. Broersma, and W. G. van der Wiel, "Evolution of
a designless nanoparticle network into reconfigurable Boolean logic," Nat. Nanotechnol., vol. 10, no. 12, pp. 1048–1052, 2015.
[9] R. F. Service, "The brain chip," Science, vol. 345, no. 6197, pp. 614–616, 2014.
[10] A. V. Avizienis, H. O. Sillin, C. Martin-Olmos, H. H. Shieh, M. Aono, A. Z. Stieg, and J. K. Gimzewski, "Neuromorphic atomic
switch networks." PLoS One, vol. 7, no. 8, p. e42772, 2012.
[11] A. Z. Stieg, A. V. Avizienis, H. O. Sillin, C. Martin-Olmos, M. Aono, and J. K. Gimzewski, "Emergent criticality in complex turing
B-type atomic switch networks," Adv. Mater., vol. 24, pp. 286–293, 2012.
[12] 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." Nat. Mater., vol. 10, no. 8, pp. 591–595, 2011.
[13] M. Suri, D. Querlioz, O. Bichler, G. Palma, E. Vianello, D. Vuillaume, C. Gamrat, and B. Desalvo, "Bio-inspired stochastic computing
using binary CBRAM synapses," IEEE Trans. Electron Devices, vol. 60, no. 7, pp. 2402–2409, 2013.
[14] C. Mead, "Neuromorphic Electronic Systems," Proc. IEEE, vol. 78, no. 10, pp. 1629–1636, 1990.
[15] R. A. Nawrocki, R. M. Voyles, and S. E. Shaheen, "A Mini Review of Neuromorphic Architectures and Implementations," IEEE Trans.
Electron Devices, vol. 63, no. 10, pp. 3819–3829, 2016.
[16] D. Sterratt, B. P. Graham, A. Gillies, and D. J. Willshaw, Principles of Computational Modelling in Neuroscience. Cambridge Univ.
Press, 2011.
[17] G. Indiveri, B. Linares-Barranco, R. Legenstein, G. Deligeorgis, and T. Prodromakis, "Integration of nanoscale memristor synapses in
neuromorphic computing architectures." Nanotechnology, vol. 24, no. 38, p. 384010, 2013.
[18] D. Querlioz, O. Bichler, P. Dollfus, and C. Gamrat, "Immunity to device variations in a spiking neural network with memristive
nanodevices," IEEE Trans. Nanotechnol., vol. 12, no. 3, pp. 288–295, 2013.
[19] G. C. Adam, B. D. Hoskins, M. Prezioso, F. Merrikh-Bayat, B. Chakrabarti, and D. B. Strukov, "3-D Memristor Crossbars for Analog
and Neuromorphic Computing Applications," IEEE Trans. Electron Devices, vol. 64, no. 1, p. 312, 2017.
[20] S. Yu, Y. Wu, R. Jeyasingh, D. Kuzum, and H. S. P. Wong, "An electronic synapse device based on metal oxide resistive switching
memory for neuromorphic computation," IEEE Trans. Electron Devices, vol. 58, no. 8, pp. 2729–2737, 2011.
[21] J. Park, M. Kwak, K. Moon, J. Woo, D. Lee, and H. Hwang, "TiOx-Based RRAM Synapse With 64-Levels of Conductance and
Symmetric Conductance Change by Adopting a Hybrid Pulse Scheme for Neuromorphic Computing," IEEE Electron Device Lett.,
vol. 37, no. 12, pp. 1559–1562, 2016.
[22] A. Thomas, "Memristor-based neural networks," J. Phys. D Appl. Phys., vol. 46, p. 093001, 2013.
[23] Y. V. Pershin and M. Di Ventra, "Experimental demonstration of associative memory with memristive neural networks," Neural Networks,
vol. 23, no. 7, pp. 881–886, 2010.
[24] Z. Wang, S. Joshi, S. E. Savel'ev, H. Jiang, R. Midya, P. Lin, M. Hu, N. Ge, J. P. Strachan, Z. Li, Q. Wu, M. Barnell, G.-L. Li, H. L.
Xin, R. S. Williams, Q. Xia, and J. J. Yang, "Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing,"
Nat. Mater., vol. 16, pp. 101–108, 2017.
[25] J. Borghetti, Z. Li, J. Straznicky, X. Li, D. A. A. Ohlberg, W. Wu, D. R. Stewart, and R. S. Williams, "A hybrid nanomemristor/transistor
logic circuit capable of self-programming." Proc. Natl. Acad. Sci. U. S. A., vol. 106, no. 6, pp. 1699–1703, 2009.
12
[26] Q. Xia, W. Robinett, M. W. Cumbie, N. Banerjee, T. J. Cardinali, J. J. Yang, W. Wu, X. Li, W. M. Tong, D. B. Strukov, G. S. Snider,
G. Medeiros-Ribeiro, and R. S. Williams, "Memristor-CMOS hybrid integrated circuits for reconfigurable logic," Nano Lett., vol. 9,
no. 10, pp. 3640–3645, 2009.
[27] Y. V. Pershin and M. Di Ventra, "Solving mazes with memristors: A massively parallel approach," Phys. Rev. E - Stat. Nonlinear, Soft
Matter Phys., vol. 84, no. 4, pp. 1–6, 2011.
[28] G. S. Snider, "Self-organized computation with unreliable, memristive nanodevices," Nanotechnology, vol. 18, p. 365202, 2007.
[29] F. Alibart, E. Zamanidoost, and D. B. Strukov, "Pattern classification by memristive crossbar circuits using ex situ and in situ training."
Nat. Commun., vol. 4, no. May, p. 2072, 2013.
[30] M. Chu, B. Kim, S. Park, H. Hwang, M. Jeon, B. H. Lee, and B. G. Lee, "Neuromorphic Hardware System for Visual Pattern
Recognition with Memristor Array and CMOS Neuron," IEEE Trans. Ind. Electron., vol. 62, no. 4, pp. 2410–2419, 2015.
[31] P. Sheridan, W. Ma, and W. Lu, "Pattern recognition with memristor networks," Proc. - IEEE Int. Symp. Circuits Syst., pp. 1078–1081,
[32] Z. Konkoli and G. Wendin, "On Information Processing with Networks of Nano-Scale Switching Elements," Int. Journ. Unconv.
Comput., vol. 10, pp. 405–428, 2014.
[33] M. S. Kulkarni and C. Teuscher, "Memristor-based Reservoir Computing," in IEEE/ACM Inter- Natl. Symp. Nanoscale Archit., 2012,
2014.
p. 226.
[34] H. O. Sillin, R. Aguilera, H.-H. Shieh, A. V. Avizienis, M. Aono, A. Z. Stieg, and J. K. Gimzewski, "A theoretical and experimental
study of neuromorphic atomic switch networks for reservoir computing." Nanotechnology, vol. 24, p. 384004, 2013.
[35] G. W. Burr, R. M. Shelby, A. Sebastian, S. Kim, S. Kim, S. Sidler, K. Virwani, M. Ishii, P. Narayanan, A. Fumarola, L. L. Sanches,
I. Boybat, M. Le Gallo, K. Moon, J. Woo, H. Hwang, and Y. Leblebici, "Neuromorphic computing using non-volatile memory," Adv.
Phys. X, vol. 2, no. 1, pp. 89–124, 2017.
[36] C. Fernando and S. Sojakka, "Pattern Recognition in a Bucket." Springer, Berlin, Heidelberg, 2003, pp. 588–597.
[37] S. Choi, P. Sheridan, and W. D. Lu, "Data Clustering using Memristor Networks," Sci. Rep., vol. 5, no. 1, p. 10492, 2015.
[38] E. C. Demis, R. Aguilera, K. Scharnhorst, M. Aono, A. Z. Stieg, and J. K. Gimzewski, "Nanoarchitectonic atomic switch networks
for unconventional computing," Jpn. J. Appl. Phys., vol. 55, no. 11, p. 1102B2, 2016.
[39] A. Sattar, S. Fostner, and S. A. Brown, "Quantized conductance and switching in percolating nanoparticle films," Phys. Rev. Lett., vol.
111, no. 13, p. 136808, 2013.
[40] S. Fostner and S. A. Brown, "Neuromorphic behavior in percolating nanoparticle films," Phys. Rev. E, vol. 92, no. 5, p. 052134, 2015.
[41] A. Z. Stieg, A. V. Avizienis, H. O. Sillin, C. Martin-Olmos, M. L. Lam, M. Aono, and J. K. Gimzewski, "Self-organized atomic switch
networks," Jpn. J. Appl. Phys., vol. 53, p. 01AA02, 2014.
[42] E. T. Rolls and G. Deco, The Noisy Brain: Stochastic dynamics as a principle of brain function. Oxford University Press, 2010.
[43] R. Reichel, J. G. Partridge, A. D. F. Dunbar, S. A. Brown, O. Caughley, and A. Ayesh, "Construction and Application of a UHV
Compatible Cluster Deposition System," J. Nanoparticle Res., vol. 8, no. 3-4, pp. 405–416, 2006.
[44] A. D. F. Dunbar, J. G. Partridge, M. Schulze, and S. A. Brown, "Morphological differences between Bi, Ag and Sb nano-particles and
how they affect the percolation of current through nano-particle networks," Eur. Phys. J. D, vol. 39, no. 3, pp. 415–422, 2006.
[45] A. I. Ayesh, A. Lassesson, S. A. Brown, A. D. F. Dunbar, M. Kaufmann, J. G. Partridge, R. Reichel, and J. van Lith, "Experimental
and simulational study of the operation conditions for a high transmission mass filter," Rev. Sci. Instrum., vol. 78, no. 5, p. 053906,
2007.
[46] D. Stauffer and A. Aharony, Introduction to percolation theory, 2nd ed. New York: CRC Press, 1992.
[47] D. R. Chialvo, "Emergent complex neural dynamics," Nat. Phys., vol. 6, no. 10, pp. 744–750, 2010.
[48] X. Yu, P. M. Duxbury, G. Jeffers, and M. A. Dubson, "Coalescence and percolation in thin metal films," Phys. Rev. B, vol. 44, no. 23,
[49] B. J. van Wees, H. van Houten, C. W. J. Beenakker, J. G. Williamson, L. P. Kouwenhoven, D. van der Marel, and C. T. Foxon,
"Quantized conductance of point contacts in a two-dimensional electron gas," Phys. Rev. Lett., vol. 60, no. 9, pp. 848–850, 1988.
[50] D. Moldovan, V. Yamakov, D. Wolf, and S. R. Phillpot, "Scaling behavior of grain-rotation-induced grain growth." Phys. Rev. Lett.,
[51] M. Olsen, M. Hummelgard, and H. Olin, "Surface modifications by field induced diffusion," PLoS One, vol. 7, no. 1, p. e30106, 2012.
[52] C. Xiang, J. Y. Kim, and R. M. Penner, "Reconnectable sub-5 nm nanogaps in ultralong gold nanowires," Nano Lett., vol. 9, no. 5,
pp. 13 163–13 166, 1991.
vol. 89, no. 20, p. 206101, 2002.
pp. 2133–2138, 2009.
[53] S. Cho, J. Yu, S. K. Kang, and D.-Y. Shih, "Oxidation study of pure tin and its alloys via electrochemical reduction analysis," J.
Electron. Mater., vol. 34, no. 5, pp. 635–642, 2005.
[54] E. Sutter, F. Ivars-Barcelo, and P. Sutter, "Size-dependent room temperature oxidation of tin particles," Part. Part. Syst. Charact.,
vol. 31, no. 8, pp. 879–885, 2014.
B, vol. 58, no. 7, pp. 4156–4165, 1998.
[55] A. F. Lee and R. M. Lambert, "Oxidation of Sn overlayers and the structure and stability of Sn oxide films on Pd(111)," Phys. Rev.
[56] A. A. Faisal, L. P. J. Selen, and D. M. Wolpert, "Noise in the nervous system." Nat. Rev. Neurosci., vol. 9, pp. 292–303, 2008.
[57] G. M. Hoerzer, R. Legenstein, and W. Maass, "Emergence of complex computational structures from chaotic neural networks through
reward-modulated hebbian learning," Cereb. Cortex, vol. 24, no. 3, pp. 677–690, 2014.
[58] J. G. Simmons, "Generalized Formula for the Electric Tunnel Effect between Similar Electrodes Separated by a Thin Insulating Film,"
J. Appl. Phys., vol. 34, no. 6, pp. 1793–1803, 1963.
[59] N. Agraıt, A. L. Yeyati, and J. M. van Ruitenbeek, "Quantum properties of atomic-sized conductors," Phys. Rep., vol. 377, no. 2-3,
pp. 81–279, 2003.
[60] T. Hasegawa, T. Ohno, K. Terabe, T. Tsuruoka, T. Nakayama, J. K. Gimzewski, and M. Aono, "Learning abilities achieved by a single
solid-state atomic switch," Adv. Mater., vol. 22, no. 16, pp. 1831–1834, 2010.
[61] T. Ohno, T. Hasegawa, A. Nayak, T. Tsuruoka, J. K. Gimzewski, and M. Aono, "Sensory and short-term memory formations observed
in a Ag2S gap-type atomic switch," Appl. Phys. Lett., vol. 99, no. 20, p. 203108, 2011.
[62] T. Chang, S. H. Jo, and W. Lu, "Short-term memory to long-term memory transition in a nanoscale memristor," ACS Nano, vol. 5,
13
[65] L. Busing, B. Schrauwen, and R. Legenstein, "Connectivity, dynamics, and memory in reservoir computing with binary and analog
neurons." Neural Comput., vol. 22, no. 5, pp. 1272–1311, 2010.
[63] N. Onofrio, D. Guzman, and A. Strachan, "Atomic origin of ultrafast resistance switching in nanoscale electrometallization cells," Nat.
[64] M. Lukosevicius and H. Jaeger, "Reservoir computing approaches to recurrent neural network training," Comput. Sci. Rev., vol. 3,
no. 9, pp. 7669–7676, 2011.
Mater., vol. 14, no. 4, pp. 440–446, 2015.
no. 3, pp. 127–149, 2009.
|
1810.12968 | 1 | 1810 | 2018-10-30T19:07:50 | Efficient Wave Optics Modeling of Nanowire Solar Cells Using Rigorous Coupled Wave Analysis | [
"physics.app-ph"
] | We investigate the accuracy of rigorous coupled wave analysis (RCWA) for near-field computations within cylindrical GaAs nanowire solar cells and discover excellent accuracy with low computational cost at long incident wavelengths, but poor accuracy at short incident wavelengths. These near fields give the carrier generation rate, and their accurate determination is essential for device modeling. We implement two techniques for increasing the accuracy of the near fields generated by RCWA, and give some guidance on parameters required for convergence along with an estimate of their associated computation times. The first improvement removes Gibbs phenomenon artifacts from the RCWA fields, and the second uses the extremely well-converged far field absorption to rescale the local fields. These improvements allow a computational speedup between 30 and 1000 times for spectrally integrated calculations, depending on the density of the near fields desired. Some spectrally resolved quantities, especially at short wavelengths, remain expensive, but RCWA is still an excellent method for performing those calculations. These improvements open up the possibility of using RCWA for low cost optical modeling in a full optoelectronic device model of nanowire solar cells. | physics.app-ph | physics |
Efficient Wave Optics Modeling of
Nanowire Solar Cells Using
Rigorous Coupled Wave Analysis
KYLE W. ROBERTSON,1 RAY R. LAPIERRE,2 AND
JACOB J. KRICH1,3
1Department of Physics, University of Ottawa, Ottawa, ON, K1N 6N5, Canada
2Department of Engineering Physics, McMaster University, Hamilton, Ontario, L8S 4L7, Canada
3School of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, Ontario, K1N 6N5,
Canada
[email protected]
Abstract: We investigate the accuracy of rigorous coupled wave analysis (RCWA) for near-field
computations within cylindrical GaAs nanowire solar cells and discover excellent accuracy
with low computational cost at long incident wavelengths, but poor accuracy at short incident
wavelengths. These near fields give the carrier generation rate, and their accurate determination
is essential for device modeling. We implement two techniques for increasing the accuracy
of the near fields generated by RCWA, and give some guidance on parameters required for
convergence along with an estimate of their associated computation times. The first improvement
removes Gibbs phenomenon artifacts from the RCWA fields, and the second uses the extremely
well-converged far field absorption to rescale the local fields. These improvements allow
a computational speedup between 30 and 1000 times for spectrally integrated calculations,
depending on the density of the near fields desired. Some spectrally resolved quantities, especially
at short wavelengths, remain expensive, but RCWA is still an excellent method for performing
those calculations. These improvements open up the possibility of using RCWA for low cost
optical modeling in a full optoelectronic device model of nanowire solar cells.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
References
1. D. Wu, X. Tang, K. Wang, and X. Li, "An Analytic Approach for Optimal Geometrical Design of GaAs Nanowires
for Maximal Light Harvesting in Photovoltaic Cells," Sci. Reports 7, 46504 (2017).
2. B. C. P. Sturmberg, K. B. Dossou, L. C. Botten, A. A. Asatryan, C. G. Poulton, R. C. McPhedran, and C. M.
de Sterke, "Optimizing Photovoltaic Charge Generation of Nanowire Arrays: A Simple Semi-Analytic Approach,"
ACS Photonics 1, 683 -- 689 (2014).
J. Wallentin, N. Anttu, D. Asoli, M. Huffman, I. Aberg, M. H. Magnusson, G. Siefer, P. Fuss-Kailuweit, F. Dimroth,
B. Witzigmann, H. Q. Xu, L. Samuelson, K. Deppert, and M. T. Borgström, "InP Nanowire Array Solar Cells
Achieving 13.8% Efficiency by Exceeding the Ray Optics Limit," Science 339, 1057 -- 1060 (2013).
J. Kupec, R. L. Stoop, and B. Witzigmann, "Light absorption and emission in nanowire array solar cells," Opt.
Express 18, 27589 -- 27605 (2010).
5. O. M. Ghahfarokhi, N. Anttu, L. Samuelson, and I. Åberg, "Performance of GaAs Nanowire Array Solar Cells for
3.
4.
Varying Incidence Angles," IEEE J. Photovoltaics 6, 1502 -- 1508 (2016).
6. K. L. Kavanagh, "Misfit dislocations in nanowire heterostructures," Semicond. Sci. Technol. 25, 024006 (2010).
7. M. T. Borgström, M. H. Magnusson, F. Dimroth, G. Siefer, O. Höhn, H. Riel, H. Schmid, S. Wirths, M. Björk,
I. Åberg, W. Peijnenburg, M. Vijver, M. Tchernycheva, V. Piazza, and L. Samuelson, "Towards Nanowire Tandem
Junction Solar Cells on Silicon," IEEE J. Photovoltaics 8, 733 -- 740 (2018).
8. K. W. Robertson, R. R. LaPierre, and J. J. Krich, "Optical Optimization of Passivated GaAs Nanowire Solar Cells,"
in Proc. IEEE 44th Photovoltaic Spec. Conf., (Washington, D.C, 2017), p. 5.
9. Y. Hu, R. R. LaPierre, M. Li, K. Chen, and J.-J. He, "Optical characteristics of GaAs nanowire solar cells," J. Appl.
10. K. M. Azizur-Rahman and R. R. LaPierre, "Wavelength-selective absorptance in GaAs, InP and InAs nanowire
11. A. H. Trojnar, C. E. Valdivia, R. R. LaPierre, K. Hinzer, and J. J. Krich, "Optimizations of GaAs Nanowire Solar
Phys. 112, 104311 (2012).
arrays," Nanotechnology 26, 295202 (2015).
Cells," IEEE J. Photovoltaics 6, 1494 -- 1501 (2016).
12. K. T. Fountaine, W. S. Whitney, and H. A. Atwater, "Resonant absorption in semiconductor nanowires and nanowire
arrays: Relating leaky waveguide modes to Bloch photonic crystal modes," J. Appl. Phys. 116, 153106 (2014).
13. Q. G. Du, C. H. Kam, H. V. Demir, H. Y. Yu, and X. W. Sun, "Broadband absorption enhancement in randomly
positioned silicon nanowire arrays for solar cell applications," Opt. Lett. 36, 1884 -- 1886 (2011).
14. M. G. Moharam, T. K. Gaylord, E. B. Grann, and D. A. Pommet, "Formulation for stable and efficient implementation
of the rigorous coupled-wave analysis of binary gratings," JOSA A 12, 1068 -- 1076 (1995).
15. M. Weismann, D. F. Gallagher, and N. C. Panoiu, "Accurate near-field calculation in the rigorous coupled-wave
analysis method," J. Opt. 17, 125612 (2015).
16. P. Lalanne and M. P. Jurek, "Computation of the near-field pattern with the coupled-wave method for transverse
magnetic polarization," J. Mod. Opt. 45, 1357 -- 1374 (1998).
17. K.-H. Brenner, "Aspects for calculating local absorption with the rigorous coupled-wave method," Opt. Express 18,
10369 -- 10376 (2010).
18. E. A. Bezus and L. L. Doskolovich, "Stable algorithm for the computation of the electromagnetic field distribution of
eigenmodes of periodic diffraction structures," JOSA A 29, 2307 -- 2313 (2012).
19. V. Liu and S. Fan, "S4 : A free electromagnetic solver for layered periodic structures," Comput. Phys. Commun. 183,
20. L. Li, "Formulation and comparison of two recursive matrix algorithms for modeling layered diffraction gratings," J.
21. D. M. Whittaker and I. S. Culshaw, "Scattering-matrix treatment of patterned multilayer photonic structures," Phys.
2233 -- 2244 (2012).
Opt. Soc. Am. A 13, 1024 (1996).
Rev. B 60, 2610 -- 2618 (1999).
22. M. G. Moharam and A. B. Greenwell, "Rigorous analysis of field distribution and power flow in grating coupler of
finite length," in Diffractive Optics and Micro-Optics, (Optical Society of America, 2004), p. DMC5.
23. H. Kim, I.-M. Lee, and B. Lee, "Extended scattering-matrix method for efficient full parallel implementation of
rigorous coupled-wave analysis," J. Opt. Soc. Am. A 24, 2313 (2007).
24. R. C. Rumpf, "Improved formulation of scattering matrices for semi-analytical methods that is consistent with
convention," Prog. In Electromagn. Res. B 35, 241 -- 261 (2011).
25. S. Adachi, Optical Constants of Crystalline and Amorphous Semiconductors: Numerical Data and Graphical
Information (Springer US, 1999), 1st ed.
26. P. Götz, T. Schuster, K. Frenner, S. Rafler, and W. Osten, "Normal vector method for the RCWA with automated
vector field generation," Opt. express 16, 17295 -- 17301 (2008).
27. L. Li, "Use of Fourier series in the analysis of discontinuous periodic structures," JOSA A 13, 1870 -- 1876 (1996).
28. "Standard Tables for Reference Solar Spectral Irradiances: Direct Normal and Hemispherical on 37° Tilted Surface,"
Tech. rep., ASTM International (2012).
29. O. Tange, GNU Parallel 2018 (Ole Tange, 2018).
Introduction
1.
Nanowire solar cells (NWSC) are a new solar cell technology with the potential to improve upon
existing solar cell devices. Their potential stems from their ability to effectively absorb incident
light while using less semiconductor material than planar solar cells.
The optimal nanowire solar cell arrays consists of nanowires that are a few microns in
height and with diameters and periodicities comparable to the wavelengths present in the solar
spectrum [1,2]. These small sizes require full wave optics simulations to accurately model their
optical properties, unlike in standard planar devices [3, 4]. Both experimental measurements
and modeling have shown high levels of absorption with low sensitivity to the incident angle of
light [5]. Additionally, the finite in-plane dimensions of nanowires can accommodate strain due
to growth on lattice-mismatched substrates without introducing dislocation faults in the crystal
lattice [6]. This capability opens up the possibility for III-V tandem cells grown on silicon [7].
The larger design parameter space of NWSC relative to planar solar cells requires careful
optimization of geometric parameters to maximize device performance [8]. There is a need for
fast, accurate modeling tools to enable rapid exploration and optimization of nanowire designs.
Conventionally, finite element [4,9 -- 11] and finite difference methods [1,12,13] have been used
in optical models of NWSC. While these techniques are highly accurate, they are computationally
expensive, limiting their usefulness in a closed-loop global device optimization. Rigorous
coupled wave analysis (RCWA) is another wave-optics modeling technique that lacks the memory
and computational requirements of competing techniques [14]. RCWA is a Fourier domain
technique ideally suited to periodic arrays. It is promising for its speed and is highly accurate
when computing far-field quantities such as total absorptance, reflectance, and transmittance.
RCWA simulations become more accurate as the number of plane waves NG increases, and the
computational cost scales as N3
G. However, naive implementations lack accuracy at reasonable NG
when computing near-fields internal to the device due to the well-known Gibbs phenomenon [15].
Such near-fields are required to compute carrier generation rates and are thus an essential
component of a fully-coupled optoelectronic device model.
In this work, we assess the accuracy of RCWA for use in optical modeling of nanowire
solar cells. We examine a test device (see Fig. 1 and Table 1), indicate where RCWA lacks
a desirable level of accuracy, and provide two techniques for increasing accuracy of the near
fields. The first is an implementation of an already published technique for introducing proper
discontinuities in the near fields and mitigating the Gibbs phenomenon [15 -- 18]. The second
is a new rescaling technique that increases the accuracy of device simulations while keeping
computational cost reasonable when computing spectrally integrated quantities. We show that
even with our two improvements, some spectrally resolved quantities continue to require more
expensive calculations. Using our improvements, RCWA shows promise as an effective technique
for rapid optical modeling of nanowire solar cells.
Fig. 1. The test device used for assessment of RCWA. Left: A single unit cell in a square
nanowire array containing a cylindrical GasAs nanowire passivated by an AlInP shell on
a GaAs substrate, planarized by a cyclotene dielectric, and top-contacted with a layer of
indium tin oxide. A thin layer of SiO2, surrounding the GaAs core but lacking the AlInP
shell, exists between the cyclotene and substrate. Right: A top down view of the unit cell,
demonstrating the piecewise constant material parameters in the plane.
2. RCWA
RCWA is a Fourier-space method for solving the source-free frequency domain Maxwell's
equations:
∇ × H = −iωE
∇ × E = iωµH
∇ · H = 0
∇ · E = 0
(1)
(2)
(3)
(4)
Substrate (GaAs)ITOCycloteneAlInP ShellCore radiusShell thicknessArray periodSiO2 Parameter
NW Core Length
NW Shell Length
SiO2 Thickness
Substrate Thickness
ITO Thickness
Array Period
Core Radius
Shell Thickness
Value
1.3 µm
1.27 µm
30 nm
1 µm
300 nm
250 nm
60 nm
20 nm
Table 1. Numerical values for all geometric parameters in the test device.
where H is the magnetic field, E is the electric field, ω is the oscillation frequency, is the electric
permitivitty, and µ is the magnetic permeability. RCWA relies on two critical assumptions about
the geometry of the system. First, the device must be composed of discrete, axially-invariant
layers such that at a given x-y point within a layer, the material parameters along the z-direction
remain constant. Second, the device must be decomposable into fundamental unit cells that are
2D periodic in the plane. If these conditions are satisfied, then the longitudinal and transverse
dimensions are separable and the fields in a single layer can be written as:
HG(z)ei(k+G)·r,
(5)
H(r, z) =
G
where G is one of NG in-plane reciprocal lattice vectors, k is the in-plane component of the
excitation, and r = xx + yy. Note the G is generally chosen to be an array of reciprocal lattice
points with a circular truncation, keeping all G with G less than some constant, which maintains
symmetry in Fourier space [19]. The in-plane dielectric profile (r) may depend on the material,
allowing it to have piecewise-constant dependence on the transverse spatial coordinates. Vertical
nanowire arrays (see Fig. 1) satisfy these geometric constraints.
The essential part of RCWA is determining HG(z) in Eq. (5) for a given set of reciprocal lattice
vectors G. One can assume the coefficients in Eq. (5) take the form [19]
φG, x x + φG,y y − (kx + Gx)φG, x + (ky + Gy)φG,y
HG(z) =
q
(cid:21)
z
eiqz,
(6)
(cid:20)
where the φ are expansion coefficients and the z-component has been chosen to satisfy the
∇ · H = 0 condition. This form of the fields illustrates one of the key advantages of RCWA over
competing techniques, namely the analytic dependence on the z coordinate. By inserting Eq.
(6) into Eq. (1), one arrives at an eigenvalue equation for determining the set of eigenvalues
q and the components of the eigenvectors φ for a single layer. Once the eigenmodes of each
layer have been determined, multilayer structures are joined together by introducing propagation
amplitudes for the eigenmodes and using the scattering matrix method to join solutions at layer
interfaces [20 -- 24]. Results increase in accuracy with NG. Our work is an extension to S4,
an open-source implementation of RCWA built on the scattering matrix method [19]. In the
remainder of the manuscript, we refer to S4 as the standard RCWA method, but it has included a
significant number of improvements from the original RCWA methods; for details, see Ref. 19.
For optoelectronic device modeling, we are most concerned with determining the local carrier
generation rate, which is determined from the local electric field strength in each material. RCWA
expresses the fields using the Fourier series in Eq. (5). Any finite Fourier series representation
is always continuous, even across in-plane material interfaces, as between the core and shell
of a nanowire. In an exact solution, the normal components of E should be discontinuous
across material boundaries, but a Fourier reconstruction requires an intractable number of
terms to accurately model such a discontinuity, even though far-field quantities such as the total
absorptance may be well converged. For any finite NG, standard RCWA-produced fields have
spurious oscillations, especially near material interfaces.
To assess the convergence of RCWA with NG, we define two methods for computing the
absorptance of a layer of the device. The first method relies only on the power exiting from the
top of the layer and the power transmitting through the bottom of the layer. These powers can
be computed entirely in Fourier space [19], and do not suffer from convergence issues in the
reconstruction of the near fields. These emitted powers of layer i are defined as:
Þ
Þ
Sz(ω)dA
top
up(ω) =
Pi
down(ω) =
Pi
Sz(ω)dA,
bottom
(7)
(8)
where Sz is the z-component of the Poynting vector and the integration is over the top or bottom
surface of the unit cell, with the appropriate sign for emitted power. Considering the top (layer 1)
and bottom (layer n) together, the total reflectance and transmittance are
R(ω) =
T(ω) =
up(ω)
P1
Pin(ω)
down(ω)
Pn
Pin(ω) ,
(9)
(10)
where Pin is the input power of the incident plane wave. Then total absorptance is:
Afar field(ω) = 1 − R(ω) − T(ω).
(11)
The contribution of a single layer to the device absorptance can be calculated similarly. We
consider the test structure detailed in Table 1 and use S4 [19] to perform RCWA calculations
with normally-incident circularly polarized light. We consider 60 equally spaced frequencies
corresponding to wavelengths from 300 nm to 900 nm, just beyond the GaAs absorption edge
of 871 nm. Figure 2 shows that the far-field absorptance spectrum of the full device converges
rapidly with basis terms, and is self-converged within 0.5% with NG = 75.
Equation (11) expresses the power absorbed in a layer in terms of the fluxes into and out of the
layer. The divergence theorem and Maxwell's equations allow rewriting that power in terms of
the local fields, instead. The absorbed power can then be written,
Þ
.
Pabs(ω) = 0ω
Pabs
Pin
Anear field =
n(x, y, z; ω)k(x, y, z; ω)E(x, y, z; ω)2dV
(12)
(13)
The complex dielectric at each frequency is constructed from tabulated real n and imaginary k
parts of the index of refraction in each material [9,25].
Fig. 2. Absorptance of the entire device calculated using the far field fluxes, Eqs. (7) - (11).
The markers for all values of NG lie nearly on top of one another, indicating convergence at
low numbers of basis terms.
Fig. 3. Absorptance calculated from near fields using Eqs. (12) - (13) (circles). Far field
absorptance at NG = 997 (triangles). a) S4 implementation of RCWA. b) Continuous
variable formulation. Gray background shows the AM1.5 solar spectrum. The CVF shows
significant improvement, especially at short wavelengths.
We calculate Anear field by extracting E(r) on a cubic mesh with 1 nm spacing in the plane for
all layers. We use 3 nm spacing along the z-direction in the ITO layer and 3.5 nm spacing in
the nanowire layer. A sparser mesh of 16 nm spacing is used in the substrate due to the weak
absorption there. This choice of mesh is sufficiently dense to converge the result better than 1%
using a simple trapezoidal rule integration. Figure 3a shows the convergence of Anear field with
NG. Though the near fields are well converged for λ > 450 nm, they are not converged at short
wavelengths even for NG = 997.
In the following sections, we provide two techniques for improving the accuracy of the
near fields in RCWA. The first is an implementation of an existing technique, which we
call the continuous variable formulation (CVF), which mitigates the Gibbs phenomenon and
ensures proper discontinuities at interfaces by modifying the field computations such that only
quantities that are continuous in real space are reconstructed from their Fourier components.
The discontinuities across in-plane material boundaries are then handled in real space. The
second technique uses the well-converged, highly accurate far-field computation of each layer's
absorption to rescale the near fields, ensuring correct total generation within a device layer.
3. Continuous Variable Formulation
The CVF is a modification to RCWA that only Fourier reconstructs quantities which are continuous
across material interfaces in real space. These quantities are the components of the displacement
field D that are normal to, and the components of the electric field E that are tangential to, a
material interface. Using these real-space continuous quantities, one can determine the full
electric field everywhere by using the constitutive relationship
D = E
(14)
with a discontinuous real-space . S4 already uses a related technique for calculating the Fourier
modes, but it does not use this method when extracting real-space quantities.
To compute the normal and tangential components of any electromagnetic (EM) field, one
must construct a locally-defined vector field that is both tangent to all material interfaces in the
unit cell and periodic in the in-plane coordinates, which can be generated automatically, as is
done by S4 [19,26]. This vector field induces an associated projection operator T that can be used
to project the Cartesian components of the EM fields onto this local coordinate system such that:
where N = 1 − T, ET is the component of E along the tangential vector field and DN is the
component of D perpendicular to the tangential vector field. The total field satisfies
By taking the Fourier transform of T(r), the projection onto the tangential vector field can
also be done in Fourier space. Mirroring the notation of Ref. 15, we denote discrete real-space
quantities with upper case letters (as in Ex to represent the vector Ex(ri) for many points ri),
vectors of Fourier coefficients with lower case letters surrounded by single brackets (as in [ex]),
Ey(r)
Dy(r)
ET,y(r)
Ex(r)
= T(r)
ET, x(r)
DN, x(r)
= N(r)
Dx(r)
,
Ex(r)
ET, x(r)
=
EN, x(r)
+
DN,y(r)
EN,y(r)
ET,y(r)
Ey(r)
(15)
(16)
(17)
.
and Fourier space matrix operators with double brackets (as in (cid:110)T(cid:111)). Using this notation, the
Fourier transform of Eq. 15 is given by [15]:
(18)
where (cid:110)T(cid:111) is the Fourier convolution matrix [15,19]. That is, one calculates the Fourier transform
T(G) of T(r), and the (G, G(cid:48)) element of (cid:110)T(cid:111) is T(G − G(cid:48)).
We extract [ex] and [ey] from a standard RCWA implementation and then construct [dN, x] and
[dN,y]. Since (r) and EN(r) are both discontinuous, the proper Fourier factorization takes [27]
eT,y
ey
(19)
where (cid:110)1/(cid:111)−1 is the 2NG × 2NG block diagonal matrix whose upper-left and lower-right blocks
are the inverse of the NG × NG Fourier convolution matrix of 1/(r). Reference 15 showed that
,
= (cid:110)T(cid:111)ex
eT, x
(cid:22)
(cid:23)−1 [dN] = [eN],
(cid:22)
(cid:22)
(cid:23)−1 (cid:110)N(cid:111)
(cid:23)−1
(cid:110)N(cid:111)
1
+
(cid:32)
1
1
(cid:33)ex
ey
,
the symmetric formulation,dN, x
dN,y
=
1
2
(20)
converges well and conserves power for lossless structures.
After finding dN, x and dN,y we reconstruct the real space electric field
(21)
where F −1 indicates the inverse Fourier transform. This EN(r) has correct discontinuities at
material interfaces where r jumps. Finally, the real space electric fields in Cartesian coordinates
can be recovered using:
EN(r) =
F −1(dN)
0r(r) ,
Ey(r) =
Ex(r) =
F −1(dN,y)
0r(r) + F −1(eT,y)
F −1(dN, x)
0r(r) + F −1(eT, x).
(22)
(23)
Figure 4 shows the norm-squared components of the electric fields computed using unmodified
RCWA and the CVF on a line cut along the x-direction through the center of the nanowire. In
this cut, Ex is normal to the interface and should therefore be discontinuous, while Ey should be
continuous. Note the Gibbs oscillations in the standard result for Ex, while the CVF result has
introduced discontinuities at the boundaries and significantly reduced the amplitude of the Gibbs
oscillations.
Figure 3b shows the improved agreement between the CVF absorptance and the well converged
far field absorptance. Figure 5 shows the relative difference between the far and near field
absorptances calculated with and without the CVF. It is clear that the CVF significantly improves
the agreement at all wavelengths, but the disagreement is still significant for wavelengths shorter
than 450 nm. Figure 5 indicates the AM1.5G spectrum, which shows that the CVF-based Anear field
agrees well with the far field results through the most important parts of the solar spectrum. In
the next section, we introduce a simple rescaling technique to increase accuracy of the near fields
at all incident wavelengths.
Fig. 4. Line cuts of Ex2 (left) and Ey2 (right) along the x-direction through the center of
the nanowire 101 nm from the top of the nanowire with and without use of the CVF with an
incident wavelength of 453 nm and NG = 997. The CVF formulation reduces the Gibbs
oscillations in Ex and introduces proper discontinuities while maintaining the continuity of
Ey.
Fig. 5. Relative difference between the far field and near field calculations of A with
NG = 997. Orange line uses the unmodified fields in Eq. 13, blue line uses the CVF fields.
Gray background shows the AM1.5G solar spectrum, which is strongest in the region of
good convergence. Black dashed line indicates the 1% mark.
4. Rescaling Technique
In device simulations, the total optical generation rate must be determined accurately. The exact
position where generation occurs is somewhat less important, as the carriers drift and diffuse,
and deviations on the scale of a few nanometers are rarely significant. We can ensure that the
total generation in each layer is calculated correctly, even with inexpensive RCWA calculations
that have not fully converged the local fields. To achieve this goal, we use the well-converged far
field results (as shown in Fig. 2) to rescale the components of the near fields in each layer such
that Afar field and Anear field agree exactly. This rescaling can be done by defining a rescaling factor
F for each layer i and frequency ω:
Then, the components of E may be rescaled such that:
Fi(ω) =
far field(ω)
Ai
near field(ω) .
Ai
Erescaled(ω) =(cid:112)Fi(ω)E(ω)
(24)
(25)
for fields in the appropriate layer.
This rescaling technique allows accurate determination of spectrally-integrated generation
rates with small numbers of basis terms. Figures 6 and 7 show line cuts through the test structure
at three representative wavelengths and spectrally integrated under AM1.5G illumination [28],
calculated with 60 equally spaced frequencies. At 487 nm, the fields are quantitatively converged
at small NG, while the Gibbs oscillations are not entirely removed either at shorter or longer
wavelengths. Calculations dependent on spectrally resolved local fields, such as external quantum
efficiency (EQE), thus require relatively large NG at some wavelengths. When the fields are
spectrally integrated, however, the essentially random phases of the oscillations average away, and
the spectrally-integrated fields are quantitatively converged by NG = 197. Figure 8 compares the
rescaled spectrally-integrated generation rates along a plane through the center of the nanowire
at NG = 997 and 197, showing the excellent agreement that rescaling permits, even at low NG.
This spectrally integrated generation rate is sufficient for optoelectronic modeling while reducing
the requirements for NG by a factor of 5.
The computational cost of the RCWA method scales as N3
G, so reducing NG by a factor of 5
(from 1000 to 200) theoretically reduces the runtime by a factor of 125, and reducing to NG = 100
can reduce the runtime by a factor of 1000. Extracting the electric fields on a dense mesh of
points, however, also has a computational cost, and for sufficiently small NG, this electric-field
extraction limits the runtime. Figure 9 shows an estimate of the simulation run times for a
single incident wavelength. Each desired wavelength must be calculated separately, and they
all take approximately the same amount of processor time. Simulations were run on a single
core of an Intel Xeon E5-2640 v4 CPU with a 2.40GHz clock speed. The figure shows both
the simulation time for RCWA to determine the field amplitudes, in Fourier space, and for the
extraction of those fields on the dense real-space mesh described above. The CVF method does
not significantly change the run times. Efforts were made to minimize data input/output time and
resource contention in all benchmarks. These results show that running at NG = 197 has a cost
25 times less than at NG = 997, and that cost is dominated by the field calculation and export.
The field calculation and export time can possibly be optimized further and would certainly
be reduced if a coarser mesh were requested. Decreasing that cost could allow computation
times to be reduced by an additional factor of 5. Simulations at each frequency are completely
independent, so computation time for a full spectral sweep can be easily reduced by running all
frequencies in parallel [29]. With a sufficient number of CPU cores, a full spectral sweep can be
run in the same clock time as a single simulation.
Fig. 6. Rescaled generation rate on a line cut along the x direction through the center of the
nanowire 83 nm from the top of the nanowire layer. The spectrally integrated and λ = 487
nm case are clearly converged even at NG = 197, while the longer and shorter wavelengths
need high NG to remove all the Gibbs oscillations.
Fig. 7. Rescaled generation rate on a line cut along the z direction through the center of the
nanowire core.
Fig. 8. Left: Spectrally integrated generation rate with AM1.5G spectrum along a cut through
the middle of the nanowire using the rescaled fields with NG = 197. Right: Absolute
difference between the generation rate shown at left and the well-converged, rescaled
generation rate at NG = 997. The deviations between the two generation maps are small.
White regions are areas of vacuum and SiO2, where the generation rate is zero. Area within
the solid contour indicates the location of peak generation, greater than 2.75× 1022 cm−3s−1,
while the differences there are much smaller. Dashed lines indicate line cuts shown in Figs.
6 and 7.
Fig. 9. Run time of a single simulation as a function of basis terms with and without
computation of the local fields. Field computations dominate the runtime at small NG. A
least-squares fit to the blue line yields a slope of 3.06, consistent with the N3
G scaling of the
QR algorithm for solving eigenvalue problems. Dashed lines are guides to the eye.
5. Conclusion
In this work, we investigate the accuracy of RCWA for optical modeling of nanowire solar
cells. We find excellent accuracy with low computational cost at long incident wavelengths,
but poor accuracy at short incident wavelengths. To increase the accuracy of RCWA we extend
the open-source library S4 [19] to include an already published technique for improving near
field computations in RCWA [15]. Our implementation mitigates the Gibbs phenomenon and
introduces physically expected discontinuities in the fields at material interfaces, improving
convergence of the near fields at all incident wavelengths. To bring convergence within a 1%
tolerance, we introduce a simple rescaling technique that uses the well converged far field
quantities to rescale the near fields on a per layer basis. These improvements open up the
possibility of using RCWA as a low cost optical modeling technique in a full optoelectronic
device model of nanowire solar cells.
6. Acknowledgements
We thank Anna Trojnar for helpful conversations and acknowledge funding from the Natural
Sciences and Engineering Research Council of Canada.
|
1808.04462 | 2 | 1808 | 2018-08-20T22:57:01 | Emerging applications of integrated optical microcombs for analogue RF and microwave photonic signal processing | [
"physics.app-ph",
"physics.optics"
] | We review new applications of integrated microcombs in RF and microwave photonic systems. We demonstrate a wide range of powerful functions including a photonic intensity high order and fractional differentiators, optical true time delays, advanced filters, RF channelizer and other functions, based on a Kerr optical comb generated by a compact integrated microring resonator, or microcomb. The microcomb is CMOS compatible and contains a large number of comb lines, which can serve as a high performance multiwavelength source for the transversal filter, thus greatly reduce the cost, size, and complexity of the system. The operation principle of these functions is theoretically analyzed, and experimental demonstrations are presented. | physics.app-ph | physics | Emerging applications of integrated optical micro-combs for analogue
RF and microwave photonic signal processing
Xingyuan Xu,1 Jiayang Wu,1 Sai T. Chu,3 Brent E. Little,4 Roberto Morandotti,5
Arnan Mitchell,2 and David J. Moss1,*
1Centre for Micro-Photonics, Swinburne University of Technology, Hawthorn, VIC 3122 Australia
2School of Engineering, RMIT University, Melbourne, VIC 3000, Australia
3Department of Physics and Material Science, City University of Hong Kong, Hong Kong, China.
4Xi'an Institute of Optics and Precision Mechanics Precision Mechanics of CAS, Xi'an, China.
5INRS -- Énergie, Matériaux et Télécommunications, Varennes, Québec, J3X 1S2, Canada.
*[email protected]
Abstract
We review new applications of integrated micro-combs in RF and microwave photonic systems. We demonstrate a wide
range of powerful functions including a photonic intensity high order and fractional differentiators, optical true time delays,
advanced filters, RF channelizer and other functions, based on a Kerr optical comb generated by a compact integrated
micro-ring resonator, or "micro-comb". The micro-comb is CMOS-compatible and contains a large number of comb lines,
which can serve as a high-performance multi-wavelength source for the transversal filter, thus greatly reduce the cost, size,
and complexity of the system. The operation principle of these functions is theoretically analyzed, and experimental
demonstrations are presented.
Keywords: Optical frequency combs, micro-ring resonator, optical signal processing.
1. INTRODUCTION
Nonlinear optics for all-optical signal processing has proven to be an extremely powerful approach, particularly when
implemented in photonic integrated circuits based on highly nonlinear materials such as silicon [1, 2]. Functions that have
been demonstrated include all-optical logic [3], demultiplexing from 160Gb/s [4] to over 1Tb/s [5], to optical performance
monitoring using slow light at speeds of up to 640Gb/s [6-7], all-optical regeneration [8,9], and many others [10-16].
CMOS compatible platforms are centrosymmetric, and so nonlinear devices have been based on a range of third order
nonlinear processes including third harmonic generation [11,17-21] and, more commonly, the Kerr nonlinearity (n2) [1,2].
The efficiency of Kerr based all-optical devices depends on the waveguide nonlinear parameter, γ = n2 / c Aeff where n2
is the Kerr nonlinearity and Aeff is the waveguide effective area). Although silicon can achieve extremely high values of
γ, it suffers from high nonlinear losses due to two-photon absorption (TPA) and the resulting free carriers [2]. Even if the
free carriers are eliminated by p-i-n junctions, silicon's poor intrinsic nonlinear figure of merit (FOM = n2 / (β λ), where β
is the TPA) of around 0.3 in the telecom band is far too low to achieve high performance. While TPA can be turned to
advantage for some all-optical functions [22-24], for the most part silicon's low FOM in the telecom band is a limitation.
This has motivated research on a range of alternate nonlinear platforms highlighted by perhaps by the chalcogenide
glasses[25-35]. While offering many advantages, these platforms however do not offer the compatibility with the silicon
computer chip industry -- CMOS (complementary metal oxide semiconductor).
In 2008 to 2010, new CMOS compatible platforms for nonlinear optics, including silicon nitride [35,36] and Hydex [37-
47] were introduced that exhibit negligible nonlinear absorption in the telecom band, a moderate nonlinear parameter and
extremely high nonlinear figure of merit, which are ideal for micro-comb generation [47]. Following the first report of
Kerr frequency comb sources in 2007 [48], the first integrated CMOS compatible integrated optical parametric oscillators
were reported in 2010 [46, 47], and since then this field has exploded [47]. Many cutting-edge applications have been
demonstrated based on CMOS-compatible micro-combs, ranging from filter-driven mode-locked lasers [49-52] to
quantum physics [53-58]. Most recently, Kerr micro-combs have demonstrated their enormous potential for sources for
ultrahigh bandwidth coherent optical fiber communications[59], optical frequency synthesis [60] and many other
applications [61-68]. The success of these new CMOS platforms as well as other CMOS platforms such as amorphous
silicon [69] arises from a combination of their low linear loss, moderate to high nonlinearity and low or even negligible
nonlinear loss (TPA).
Updated 3/20/14
An emerging and very powerful application of micro-ring resonator based Kerr frequency combs - "microcombs" [47] -
has been in the area of radio frequency (RF) and microwave photonics that bring together the worlds of radiofrequency
engineering and optoelectronics [70-90]. RF and microwave photonics exploit the potential of optical technologies and
offers many benefits for RF systems including high speed, broad operation bandwidth, low loss, and strong immunity to
electromagnetic interference. A diverse range of photonic approaches to RF signal generation, transmission, processing,
and sensing have been proposed and widely employed in RF systems and communication networks[91-117]. However,
most RF systems are composed of discrete components, which impose certain drawbacks in terms of cost, power
consumption and reliability, thus holding RF photonic systems from reaching maturity and replacing traditional RF
solutions. As one of the most powerful tools in RF photonic systems, optical frequency combs offer an alternative as multi-
wavelength sources for multiple RF channels, and thus can greatly increase the capacity for transmission and performance
for transversal processers. This is particularly true of optical micro-combs [47, 118-129].
Here, we review our recent progress in applications of optical micro-combs to RF and microwave photonics [118]. In
particular, we demonstrate a reconfigurable RF photonic intensity differentiator [124] based on CMOS-compatible micro-
combs. By employing an on-chip nonlinear micro-ring resonator (MRR), we generate a broadband Kerr comb with a large
number of comb lines and use it as a high-quality multi-wavelength source for a transversal differentiator. The large
frequency spacing of the integrated Kerr comb yields a potential operation bandwidth of over 100 GHz, well beyond the
processing bandwidth of electronic devices. By programming and shaping the power of individual comb lines according
to corresponding tap weights, reconfigurable intensity differentiators with variable differentiation orders can be achieved.
Detailed analyses of the operation principle and experimental demonstrations of fractional-, first-, second-, and third-order
intensity differentiations are performed. We also present results on optical true time delays for radar applications [129],
programmable RF filters [122] and other advanced functions [122]. Optical micro-combs offer powerful solutions to
implement RF and microwave photonic signal processing functions.
2. OPERATION PRINCIPLES
Based on the theory of signals and systems, the spectral transfer function of an N-th order temporal differentiator can be
expressed as
HH(ω) ∝ (jω)N, (1)
where j = √−1, ω is the angular frequency, and N is the differentiation order. According to the above transfer function,
the amplitude response of a temporal differentiator is proportional to ωN, while the phase response has a linear profile,
with a zero and π jump at zero frequency for N even and odd, respectively. The ideal RF amplitude and phase responses
of first-, second-, and third-order microwave differentiators are shown in Figs. 1(a) -- (c), respectively.
We employ [114] a versatile approach towards the implementation of microwave photonic differentiators based
on transversal filters, where a finite set of delayed and weighted replicas of the input RF signal are produced in the optical
domain and combined upon detection. The transfer function of a typical transversal filter can be described as
HH(ω) = ∑
M-1
n=0
an
e-jωnT, (2)
where M is the number of taps, an is the tap coefficient of the nth tap, and T is the time delay between adjacent taps. It
should be noted that the differentiator designed based on Eq. (2) is an intensity differentiator, i.e., the output RF signal
after being combined upon detection yields an exact differentiation of the input RF signal, in contrast to field differentiators
that yield the derivative of a complex optical field [92-96]. Figure 1(f) depicts the difference between field differentiation
and intensity differentiation.
Fig. 1. Simulated RF amplitude and phase responses of the (a) first-, (b) second-, and (c) third-order temporal differentiators. The
amplitude and phase responses of the differentiators designed based on Eq. (2) with different number of taps are also shown
accordingly. (d) RMSEs between calculated and ideal RF amplitude responses of the first-, second-, and third-order intensity
differentiators as a function of the number of taps. Simulated (e) input Gaussian pulse and its corresponding first-order
differentiation results.
Fig. 2. Schematic illustration of the proposed reconfigurable microwave photonic intensity differentiator. TLS: tunable laser
source. EDFA: erbium-doped fibre amplifier. PC: polarization controller. BPF: optical bandpass filter. TCS: temperature control
stage. MZM: Mach-Zehnder modulator. SMF: single mode fibre. OC: optical coupler. PD: photo-detector. OSA: optical spectrum
analyzer. VNA: vector network analyzer. AWG: arbitrary waveform generator.
To implement the temporal differentiator in Eq. (1), we calculate the tap coefficients in Eq. (2) based on the Remez
algorithm [114].The corresponding amplitude and phase responses of the first-, second-, and third-order differentiators as
a function of the numbers of taps are also plot in Figs. 1(a) -- (c). When the number of taps is increased, it is clear that the
discrepancies between the amplitude response of the transversal filters and the ideal differentiators are improved for all
three orders, whereas the phase response of the transversal filters is identical to that of the ideal differentiators regardless
of the number of taps. To quantitatively analyze the discrepancies in the amplitude response, we further calculate the root
mean square errors (RMSEs) for the first-, second-, and third-order differentiators as a function of the number of taps, as
shown in Fig. 1(d). One can see that the RMSE is inversely proportional to the number of taps, as reasonably expected. In
particular, we note that when the number of taps increases, the RMSE decreases dramatically for a small number of taps,
and then decreases more gradually as the number of taps becomes larger.
Figure 2 shows a schematic illustration of the reconfigurable microwave photonic intensity differentiator [124]. It consists
of two main blocks: one is a Kerr comb generation module based on a nonlinear MRR and the other is a transversal filter
module for reconfigurable intensity differentiation. In the first module, the continuous-wave (CW) light from a tunable
laser source is amplified by an erbium-doped fibre amplifier (EDFA), followed by a tunable optical bandpass filter to
suppress the amplified spontaneous emission noise. A polarization controller is inserted before the nonlinear MRR to make
sure that the polarization state matches the desired coupled mode. When the wavelength of the CW light is tuned to a
resonance of the nonlinear MRR and the pump power is high enough for sufficient parametric gain, the optical parametric
oscillation (OPO) process in the nonlinear MRR is initiated, which generates a Kerr optical comb with nearly equal line
spacing. The nonlinear MRR is mounted on a highly precise temperature control stage to avoid resonance drifts and
maintain the wavelength alignment of the resonance to the CW light. Owing to the compact size and ultra-high quality
factor of the nonlinear MRR, the generated Kerr comb provides a large number of wavelength channels with narrow
linewidths for the subsequent transversal filter module. As compared to conventional intensity differentiators based on
laser diode arrays, the cost, size and complexity can be greatly reduced. After being amplified by another EDFA, the
generated Kerr comb then is directed to the second module. The Kerr comb is processed by a waveshaper to get weighted
taps according to the coefficients calculated by means of the Remez algorithm. Considering that the generated Kerr comb
is not flat, a real-time feedback control path is introduced to read and shape the comb lines' power accurately. A 2×2
balanced Mach-Zehnder modulator (MZM) is employed to generate replicas of the input RF signal. When the MZM is
quadrature-biased, it can simultaneously modulate the input RF signal on both positive and negative slopes, thus yielding
modulated signals with opposite phases and tap coefficients with opposite algebraic signs. After being modulated, the
tapped signals from one output of the MZM are delayed by a dispersive fibre. The time delay between adjacent taps is
determined jointly by the frequency spacing of the employed comb source and the dispersion accumulated in the fibre.
Finally, the weighted and delayed taps are combined upon detection and converted back into RF signals to form the
differentiation output.
Due to the intrinsic advantages of transversal filters, this scheme features a high degree of reconfigurability in terms of
processing functions and operation bandwidth, thus offering a reconfigurable platform for diverse microwave photonic
computing functions. By programming the waveshaper to shape the comb lines according to the corresponding tap
coefficients, this scheme can also apply to other computing functions such as Hilbert transforms[123]. The operation
bandwidth can also be changed by adjusting the time delay between adjacent taps or employing different tap coefficients.
An increased operation bandwidth can be achieved by simply employing a dispersive fibre with shorter length. The
operation bandwidth is fundamentally limited by the Nyquist zone, which is determined by the comb spacing. In our case,
the frequency spacing of the Kerr comb generated by the nonlinear MRR reaches 200 GHz, thus leading to a potential
operation bandwidth of over 100 GHz.
Fig. 3. (a) Schematic illustration of MRR. (b) SEM image of the cross-section of the MRR before depositing the SiO2 upper cladding.
3. EXPERIMENTAL RESULTS
In our experiment, as Fig. 3(a) shows, the nonlinear MRR used to generate the Kerr comb was fabricated on a high-index
doped silica glass platform using CMOS compatible fabrication processes [28]. First, high-index (n = 1.7) doped silica
glass films were deposited using standard plasma enhanced chemical vapour deposition (PECVD), then photolithography
and reactive ion etching (RIE) were employed to form waveguides with exceptionally low surface roughness. Finally, the
waveguides were buried in a lower index upper cladding. The radius of the Hydex MRR is ~135 μm. The compact
integrated MRR has a large free spectral range (FSR) of ~1.6 nm, i.e., ~200 GHz. Such a large FSR enables an increased
Nyquist zone of ~100 GHz, which is challenging for mode-locked lasers and externally-modulated comb sources. The
advantages of the platform for integrated nonlinear optics include ultra-low linear loss (~0.06 dB‧ cm−1), a moderate
nonlinearity parameter (~233 W−1‧ km−1), and in particular a negligible nonlinear loss up to extremely high intensities
(~25 GW‧ cm−2). After packaging the input and output ports of the device with fibre pigtails, the total insertion loss is
~3.5 dB. A scanning electron microscope (SEM) image of the cross-section of the MRR before depositing the SiO2 upper
cladding is shown in Fig. 3(b). By boosting the power of the CW light from the TLS via an EDFA and adjusting the
polarization state, multiple mode-spaced combs were first generated, in which the primary spacing was determined by the
parametric gain. When the parametric gain lobes became broad enough, secondary comb lines with a spacing equal to the
FSR of the MRR were generated via either degenerate or non-degenerate four wave mixing (FWM). In our experiment,
the power threshold for the generation of secondary comb lines was ~500 mW.
Fig. 3. (a) Schematic illustration of MRR. (b) SEM image of the cross-section of the MRR before depositing the SiO2 upper
Fig. 4. (a) Optical spectrum of the generated Kerr comb in a 300-nm wavelength range. Inset shows a zoom-in spectrum with a span
cladding.
of ~32 nm. (b) -- (d) Measured optical spectra (red solid) of the shaped optical combs and ideal tap weights (green crossing) for the
first-, second-, and third-order intensity differentiators.
The resulting Type II Kerr optical comb [37] as shown in Fig. 4(a) is over 200-nm wide, and flat over ~32 nm. Since the
generated comb only served as a multi-wavelength source for the subsequent transversal filter in which the optical power
from different taps was detected incoherently by the photo-detector, the coherence of the comb was not crucial and the
proposed differentiator was able to work under relatively incoherent conditions, although by necessity avoiding the chaotic
regime [37]. In the experiment, the numbers of taps used for fractional-, first-, second-, and third-order differentiation
demonstrations were 7, 8, 6, and 6, respectively. The choice of these numbers was made mainly by considering the power
dynamic range of the comb lines, i.e., the difference between the maximum power of the generated comb lines and the
power associated with the noise floor. The power dynamic range was determined by the EDFA before waveshaping, and
in our case, it was ~30 dB. An increased number of taps requires a broader power dynamic range, which can be achieved
by using an EDFA with a lower noise floor. As analysed in section 2, more taps are needed when the differentiation order
increases, and for a fixed number of taps, increasing the order of differentiation also increases the required power dynamic
range. In order to get better performance with a limited number of taps, we decreased the operation bandwidth of the
second- and third-order differentiators to half of the transversal filter's Nyquist frequency when designing the response
function with the Remez algorithm. It should be noted that the actual bandwidth of the differentiator is not limited by this
design since the FSR of the transversal filter can be enlarged. The calculated tap coefficients for fractional-, first-, second-
, and third-order differentiations are listed in Table I. The selected comb lines of the generated optical comb were processed
by the waveshaper based on these coefficients. Considering that the generated Kerr comb was not flat, we adopted a real-
time feedback control path to increase the accuracy of comb shaping. The comb lines' power was first detected by an
optical spectrum analyzer (OSA) and compared with the ideal tap weights, and then an error signal was generated and fed
back into the waveshaper to calibrate the system and achieve accurate comb processing. The shaped optical combs are
shown in Figs. 4(b) -- (d). A good match between the measured comb lines' power (red solid line) and the calculated ideal
tap weights (green crossing) was achieved, indicating that the comb lines were successfully shaped. They were then divided
into two parts according to the algebraic sign of the tap coefficients and fed into the 2×2 balanced MZM biased at
quadrature. The modulated signal after the MZM was propagated through ~2.122-km single mode (dispersive) fibre
(SMF). The dispersion of the SMF is ~17.4 ps/(nm‧km), which corresponds to a time delay of ~59 ps between adjacent
taps and yields an effective FSR of ~16.9 GHz in the RF response spectra.
Fig. 5. Measured and simulated RF amplitude and phase responses of (a-i) -- (a-ii) the fractional-order, (b-i) -- (b-ii) the first-order, (c-i) --
(c-ii) second-order, and (d-i) -- (d-ii) third-order intensity differentiators. The simulated amplitude and phase responses after
incorporating the tap error, chirp, and the third-order dispersion (TOD) are also shown accordingly.
Table 1.
Tap coefficients for the fractional-, first-, second-, and third-order differentiations.
Order of
Number of taps
Tap coefficients
differentiation
Fractional-order
First-order
Second-order
7
8
6
[0.0578, -0.1567, 0.3834, 1, -0.8288, 0.0985, -0.0892]
[−0.0226, 0.0523, −0.1152, 1, −1, 0.1152, −0.052, 0.0226]
[0.0241, −0.1107, 0.0881, 0.0881, −0.1107, 0.0241]
Third-order
6
[0.0450, −0.4076, 1, −1, 0.4076, −0.0450]
After the weighted and delayed taps were combined upon detection, the RF responses for different differentiation orders
were characterized by a vector network analyser. Figures. 5(a-i), (b-i), (c-i) and (d-i) show the measured and simulated
amplitude frequency responses of the first-, second-, and third-order intensity differentiators, respectively. The
corresponding phase responses are depicted in Figs. 5(a-ii), (b-ii), (c-ii) and (d-ii). It can be seen that all three differentiators
feature responses close to that expected from ideal differentiations. The FSR of the RF response spectra is ~16.9 GHz,
which is consistent with the time delay between adjacent taps, as calculated before measurement. Note that by adjusting
the FSR of the proposed transversal filter through the dispersive fibre or by programming the tap coefficients, a variable
operation bandwidth for the intensity differentiator can be achieved, which is advantageous for meeting diverse
requirements in operation bandwidth.
Fig. 6. Theoretical (red dashed) and experimental (blue solid) responses of the (a) fractional- (b) first-, (b) second-, and (c) third-
order intensity differentiators.
We also performed [124] system demonstrations of real-time signal differentiations for Gaussian input pulses with a full-
width at half maximum (FWHM) of ~0.12 ns, generated by an arbitrary waveform generator (AWG, KEYSIGHT
M9505A). The waveform of the output signals after differentiation are shown in Figs. 6(a) -- (d) (blue solid curves). They
were recorded by means of a high-speed real-time oscilloscope (KEYSIGHT DSOZ504A Infiniium). For comparison, we
also depict the ideal differentiation results, as shown in Figs. 6(a) -- (d) (red dashed curves). The practical Gaussian pulse
was used as the input RF signal for the simulation. One can see that the measured curves closely match their theoretical
counterparts, indicating good agreement between experimental results and theory. Unlike the field differentiators, temporal
derivatives of intensity profiles can be observed, indicating that intensity differentiation was successfully achieved. For
the first-, second-, and third-order differentiators, the calculated RMSE between the measured and the theoretical curves
are ~4.15%, ~6.38%, and ~7.24%, respectively.
4. RF PHOTONIC TRUE TIME DELAY LINES (TTDLS)
In modern radar and communications systems, RF photonic true time delay lines (TTDLs), which introduce multiple
progressive time delays by means of RF photonics technologies, are basic building blocks with wide applications in phased
array antennas (PAAs), RF photonic filters, analog-to-digital or digital-to-analog conversion, and arbitrary waveform
generation [110, 130, 131]. For RF photonic TTDLs, increasing the number of delay channels can lead to significantly
improved performance. For instance, in PAAs, the number of radiating elements determines the beamwidth, and so an
improved angular resolution can be achieved by increasing the channel number [132]. Consequently, RF photonic TTDLs
with a large number of channels are highly desirable for PAAs with high angular resolution. In conventional methods,
discrete lasers arrays [110, 133] or fiber Bragg grating (FBG) arrays [132] are employed for implementing multiple TTDL
channels, and so system cost and complexity significantly increase with the rising channel number, thus greatly limiting
the number of available channels in practical systems. In contrast, for a broadband Kerr optical micro-comb generated by
a single micro-resonator, a large number of high-quality wavelength channels can be simultaneously provided for the RF
photonic TTDL, thus greatly reducing the size, cost, and complexity of the system [134].
Figure 7(a) shows a schematic diagram of a multi-channel RF photonic TTDL with an optical micro-comb source.
The TTDL is constructed with three modules. The first module generates the optical micro-comb with a large number of
wavelength channels using a micro-resonator. The second module shapes the optical micro-comb and directs it to a Mach-
Zehnder modulator (MZM), where replicas of the input RF signal are generated on each wavelength to establish multiple
RF channels. Time delays were then introduced between the RF channels by dispersive elements in the third module. The
measured time delay responses of a 21-channel RF photonic TTDL based on the optical micro-comb generated by a Hydex
MRR with a FSR of ~200 GHz are shown in Fig. 7(b). We used two different lengths of dispersive single mode fibres
(SMFs), yielding two different delay steps of ~118 ps/channel and ~59 ps/channel. As shown in Fig. 7(c), we also measured
the time delay responses of an 81-channel RF photonic TTDL based on the optical micro-comb generated by another
Hydex MRR with a FSR of ~49 GHz and got two delay steps of ~29.6 ps/channel and ~14.8 ps/channel for different
lengths of the dispersive SMF. The time delay steps were jointly determined by the frequency spacing of the generated
optical comb and the accumulated dispersion of the SMF, yielding the four different delay steps shown in Figs. 7(b) and
(c). The optical spectra of the Kerr combs generated by the Hydex MRRs are shown in Figs. 7(d-i) and (d-ii) accordingly.
One of the important applications of an RF photonic TTDL is in phased array antenna (PAA) systems, where optical
signals on selected channels are separately converted into the electrical domain and then fed to an antenna array to form
radiating elements. Considering that the antenna array is a uniformly spaced linear array with an element spacing of dPAA,
the steering angle θ0 of the PAA can be given as [111]
θ0= sin−1 c∙τ
dPAA
, (3)
where c is the speed of light in vacuum, and τ = mT (m = 1, 2, 3, …) is the time delay difference between adjacent radiating
elements, with T denoting the time delay difference between adjacent delay channels. The steering angle can be tuned by
adjusting τ, i.e., by changing the length of the dispersive medium or simply selecting every mth channel as radiating
elements. The corresponding array factor AF of the PAA can be expressed as [111]
AF(θ, λ)=
sin2[Mπ(dPAA/λ)(sinθ−c∙mT dPAA
)]
M 2sin2[π(dPAA/λ)(sinθ−c∙mT dPAA
⁄
⁄
)]
(4)
where θ is the radiation angle, M is the number of radiating elements, and λ is the wavelength of the RF signals. The
angular resolution of the PAA is the minimum angular separation at which two equal targets at the same range can be
separated. It is determined by the 3-dB beamwidth of the PAA, which can be approximated as θ3dB = 102 / M [112], thus
indicating that the angular resolution would greatly increase with a larger number of radiating elements, and this can be
realised via optical micro-combs that can provide numerous wavelength channels for beam steering.
Fig. 7. RF photonic TTDLs based on Hydex optical micro-comb sources. (a) Schematic diagram. MRR: micro-ring resonator. MZM:
Mach-Zehnder modulator. (b) Measured time delay responses of a 21-channel TTDL with (i) 4-km-long dispersive SMF and (ii) 2-km-
long dispersive SMF. (c) Measured time delay responses of an 81-channel TTDL with (i) 4-km-long dispersive SMF and (ii) 2-km-long
dispersive SMF. Insets in (b) and (c) show the flat delays over a wide RF frequency range. (d) Optical spectra of the micro-combs
generated by (i) the Hydex MRR with a FSR of ~200 GHz and (ii) the Hydex MRR with a FSR of ~49 GHz.
Figure 8(a) depicts the relationship between the number of radiating elements M and the 3-dB beamwidth θ3dB of the
PAA, where smaller beamwidths can be obtained when M increases, therefore suggesting, given the large number of
available channels provided by the optical micro-comb based RF photonic TTDL, that the angular resolution for the PAA
can be greatly enhanced. The improved angular resolution for increased M can also be observed in Fig. 8(b), which shows
the AF of the PAA for different M calculated based on Eq. (4). To achieve a tunable beam steering angle, we selected
every mth (m = 1, 2, 3, …) wavelength of the TTDL by using the waveshaper, in such a way that the time delay (τ) between
the radiating elements could be varied with a step size of T. Figures. 8(c) and (d) show the calculated AF for m (m =
Channel number / M) varying from 1 to 7 based on a 200-GHz-FSR Hydex optical micro-comb and varying from 1 to 27
based on a 49-GHz-FSR Hydex optical micro-comb, respectively. As can be seen, the 49-GHz-FSR Kerr comb enables a
much larger M as m varies, thus leading to a much smaller beamwidth and greatly enhanced angular resolution, as well as
finer tuning steps and a large tuning range of the beam steering angle because of the larger number of channels (m).
Moreover, PAAs based on RF photonic TTDLs can also achieve wide instantaneous bandwidths without beam squint (the
variation of beam steering angle as a function of RF frequency). As indicated in Figs. 8(e) and (f), the beam steering angle
remains the same while the RF frequency varies. These results confirm that PAAs based on RF photonic TTDLs with an
optical micro-comb source feature greatly improved angular resolution, a wide tunable range in terms of beam steering
angle, and a large instantaneous bandwidth. By using tunable dispersive elements to replace the SMF, allowing a tunable
adjustment in time delay [136], optical micro-comb based two-dimensional PAAs with continuously-tunable time delays
can be further realised.
Fig. 8. (a) Relationship between the number of radiating elements M and the 3dB beamwidth θ3dB of the PAA. (b) Calculated AF of the
PAA with M varying from 4 to 81 based on a 49-GHz-FSR Hydex optical micro-comb. (c) Calculated AF of the PAA with m varying
from 1 to 7 based on a 200-GHz-FSR Hydex optical micro-comb. (d) Calculated AF of the PAA with m varying from 1 to 27 based on
a 49-GHz-FSR Hydex micro-comb. (e) Calculated AF of the PAA with various RF frequencies based on a 200-GHz-FSR Hydex optical
micro-comb. (f) Calculated AF of the PAA with various RF frequencies based on a 49-GHz-FSR Hydex optical micro-comb.
5. CONCLUSION
Optical micro-combs represent an exceptionally active field of research that has only developed in the last ten years. This
young and fast-growing field has its roots in the pioneering development of high-Q micro-cavity technologies, starting in
the late 1980s and experiencing significant recent advances, both for bulk and integrated resonators. We review our recent
progress in using optical micro-combs in the context of RF and microwave photonic signal processing. We demonstrate a
reconfigurable microwave photonic intensity differentiator based on an integrated Kerr comb source. The Kerr optical
comb is produced via a CMOS-compatible nonlinear MRR, which greatly increases the processing bandwidth and has the
potential for a low cost, size, and complexity processing system. By programming and shaping the individual comb lines'
power according to the calculated tap weights, we successfully demonstrate the first-, second-, and third-order intensity
differentiations of the RF signal. The RF amplitude and phase responses of the proposed differentiator are characterized,
and system demonstrations of real-time differentiations are performed for Gaussian input pulses. We achieve good
agreement between theory and experimental results, thus verifying the effectiveness of our method. We also demonstrate
optical true time delays for advanced RF radar applications. Our approach of using microcombs for RF and microwave
signal processing provides a new way to implement microwave photonic signal processing functions with compact device
footprint, high processing bandwidth, and high reconfigurability, thus holding great promise for future ultra-high-speed
computing and information processing.
6. ACKNOWLEDGMENTS
This work was supported by the Australian Research Council Discovery Projects Program (No. DP150104327). RM
acknowledges support by Natural Sciences and Engineering Research Council of Canada (NSERC) through the Strategic,
Discovery and Acceleration Grants Schemes, by the MESI PSR-SIIRI Initiative in Quebec, and by the Canada Research
Chair Program. He also acknowledges additional support by the Government of the Russian Federation through the ITMO
Fellowship and Professorship Program (grant 074-U 01) by the 1000 Talents Sichuan Program. BEL was supported by the
Strategic Priority Research Program of the Chinese Academy of Sciences, Grant No. XDB24030000.
References
[1] B.J. Eggleton, S.Radic, and D.J. Moss, Chapter 20, p759-828 in Optical Fiber Telecommunications V: Components
and Sub-systems, Edited by Ivan P. Kaminow, Tingye Li, and Alan E. Willner, Academic Press, Oxford, UK, February
(2008).
[2] J. Leuthold, C. Koos, and W. Freude, Nat. Photon. 4 (8), 535 (2010).
[3] Li, F. et al., Optics Express 19, 20364-20371 (2011). DOI: 10.1364/OE.19.020364.
[4] Li, F. et al., Optics Express 18, 3905-3910 (2010). DOI: 10.1364/OE.18.003905.
[5] H.Ji, et.al., Photonics Technology Letters 22 1762 (2010).
[6] Monat, C. et al., Optics Express 18 (7), 6831-6840 (2010). DOI: 10.1364/OE.18.006831.
[7] Corcoran, B., et. al., Optics Express 18, (8) 7770-7781 (2010). DOI: 10.1364/OE.18.007770.
[8] V.G. Ta'eed, et al., Optics Letters 30 2900 (2005). DOI: 10.1364/OL.30.002900.
[9] M. Rochette, J. Kutz, J. Blows, D. Moss, J. T. Mok, and B. J. Eggleton, IEEE Photonics Technology Letters 17 908
(2005).
[10] M.Ferrera et al., Optics Express 22 (18) 21488 - 21498 (2014). DOI: 10.1364/OE.22.021488.
[11] C.Monat et al., Nature Communications 5 Article:3246 (2014). doi:10.1038/ncomms4246.
[12] F. Li, et al., Optics Express 19, (23) 22410-22416 (2011).
[13] T.D. Vo, et al., IEEE Journal of Lightwave Technology 29 (12) 1790-1796 (2011).
[14] M.Ferrera, et al., Optics Express 19 (23) 23153-23161 (2011).
[15] B. Corcoran, et al., Optics Express 18 20190 (2010). DOI: 10.1364/OE.18.020190.
[16] L.Caspani, et al., J. of
the Optical Society of America B, JOSA B28
DOI:10.1364/JOSAB.28.000A67
(12) A67 -- A82
(2011).
[17] B.Corcoran, et al., Nature Photonics 3 (4) 206 (2009). doi:10.1038/nphoton.2009.28.
[18] D. J. Moss, H. M. van Driel, and J. E. Sipe, Opt. Lett. 14 (1), 57 (1989).
[19] J. E. Sipe, D. J. Moss, and H. M. van Driel, Phys. Rev. B 35 (3), 1129 (1987).
[20] D. J. Moss, E. Ghahramani, J. E. Sipe, and H. M. van Driel, Phys. Rev. B 41 (3), 1542 (1990).
[21] D. J. Moss, H. M. van Driel, and J. E. Sipe, Appl. Phy. Lett. 48 (17) 1150 (1986).
[22] D.J.Moss, et al., Electronics Letters 41 320 (2005). DOI:10.1049/el:20058051
[23] M.R.E. Lamont, et al., Photonics Technology Letters 18 1185 (2006). DOI:10.1109/LPT.2006.874718.
[24] A.Tuniz, G.Brawley, D.J.Moss, B.J.Eggleton,, Optics Express 16 18524 (2008). DOI: 10.1364/OE.16.018524.
[25] BJ Eggleton, B Luther-Davies, K Richardson, "Chalcogenide photonics", Nature Photonics 5 (3), 141 (2011).
[26] M.Lee, et al., Optics Express 15 1277-1285 (2007). DOI:10.1364/OE.15.001277
[27] S. Tomljenovic-Hanic, M.J. Steel, C. Martijn de Sterke and D. J. Moss, "High-Q cavities in photosensitive photonic
crystals", Optics Letters 32 (5) 542-544 (2007). DOI:10.1364/OL.32.000542
[28] C.Grillet, et al., Optics Express 15 (3) 1267-1276 (2007). DOI:10.1364/OE.15.001267
[29] V.G. Ta'eed, N. Baker, L. Fu, K. Finsterbusch, M.R.E. Lamont, H.Nguyen, D.J. Moss, and B.J. Eggleton, Y. Choi,
S. Madden, B. Luther-Davies "Ultrafast all-optical chalcogenide glass photonic circuits", Optics Express 15 9205
(2007).
[30] D.Freeman, C.Grillet, M.Lee, B.Luther-Davies, CLC Smith, Y Ruan, A Rode, D. J. Moss, and B. J. Eggleton,
"Chalcogenide Glass Photonic Crystal Devices", Photonic and Electromagnetic Crystal Structures, Photonics and
Nanostructures-Fundamentals and Applications, Science Direct Elsevier Publishing 6 (1) Pages: 3-11 (2008).
doi:10.1016/j.photonics.2007.11.001.
[31] C.Grillet, D.Freeman, B.Luther-Davies, S.Madden, R.McPhedran, D.J. Moss, M.J. Steel, and B.J.Eggleton,
"Characterization and modeling of Fano resonances in chalcogenide photonic crystal membranes", Optics Express 14
369 (2006).
[32] V.G. Ta'eed, M. Shokooh-Saremi, L.B Fu, D.J. Moss, M.Rochette, I.C.M. Littler, B.J. Eggleton, Y.Ruan and
B.Luther-Davies, "Self-phase modulation based integrated optical regeneration in chalcogenide waveguides", IEEE
Journal of Selected Topics in Quantum Electronics 12 360 (2006).
[33] M.Shokooh-Saremi, V.G. Ta'eed, I.Littler, D.J. Moss and B.J. Eggleton, Y.Ruan and B.Luther-Davies, "High
performance Bragg gratings in chalcogenide rib waveguides written with a modified Sagnac interferometer:
experiment and modeling", Journal of the Optical Society of America B (JOSA B) 23 1323 (2006).
[34] M.R.E. Lamont, V.G. Ta'eed, M.A.F. Roelens, D.J. Moss, B.J. Eggleton, D-Y. Choy, S. Madden and B.
Luther-Davies, "Error-free wavelength conversion via cross phase modulation in 5 cm of As2S3 chalcogenide glass
rib waveguide", Electronics Letters 43 945 (2007).
[35] Ikeda, K., Saperstein, R. E., Alic, N. & Fainman, Y., "Thermal and Kerr nonlinear properties of plasma-deposited
silicon nitride/silicon dioxide waveguides", Opt. Express 16, 12987 -- 12994 (2008).
[36] J. S. Levy, A. Gondarenko, M. A. Foster et al., "CMOS-compatible multiple-wavelength oscillator for on-chip optical
interconnects," Nature Photonics 4 (1), 37-40 (2010).
[37] L. Razzari, D. Duchesne, M. Ferrera et al., "CMOS-compatible integrated optical hyper-parametric oscillator,"
Nature Photonics 4 (1), 41-45 (2010).
[38] D. J. Moss, R. Morandotti, A. L. Gaeta et al., "New CMOS-compatible platforms based on silicon nitride and Hydex
for nonlinear optics," Nature Photonics 7 (8), 597-607 (2013).
[39] M. Ferrera, L. Razzari, D. Duchesne et al., "Low-power continuous-wave nonlinear optics in doped silica glass
integrated waveguide structures," Nature Photonics, 2 (12), 737-740 (2008).
[40] M. Ferrera, D. Duchesne, L. Razzari et al., "Low power four wave mixing in an integrated, micro-ring resonator with
Q=1.2 million," Optics Express, 17 (16), 14098-14103 (2009).
[41] D. Duchesne, M. Peccianti, M. R. E. Lamont et al., "Supercontinuum generation in a high index doped silica glass
spiral waveguide," Optics Express, 18 (2), 923-930 (2010).
[42] M. Ferrera, et al., "On-chip CMOS-compatible all-optical integrator", Nature Communications, 1 Article 29 (2010).
DOI:10.1038/ncomms1028
[43] A. Pasquazi, R. Ahmad, M. Rochette et al., "All-optical wavelength conversion in an integrated ring resonator,"
Optics Express, 18 (4), 3858-3863 (2010).
[44] A. Pasquazi, Y. Park, J. Azana et al., "Efficient wavelength conversion and net parametric gain via Four Wave
Mixing in a high index doped silica waveguide," Optics Express, 18 (8), 7634-7641 (2010).
[45] M. Peccianti, M. Ferrera, L. Razzari et al., "Subpicosecond optical pulse compression via an integrated nonlinear
chirper," Optics Express, 18 (8), 7625-7633 (2010).
[46] D. Duchesne, M. Ferrera, L. Razzari et al., "Efficient self-phase modulation in low loss, high index doped silica glass
integrated waveguides," Optics Express, 17 (3), 1865-1870 (2009).
[47] Pasquazi, M. Peccianti, L. Razzari, D. J. Moss, S. Coen, M. Erkintalo, Y. K. Chembo, T. Hansson, S. Wabnitz, P.
Del'Haye, X. Xue, A. M. Weiner, and R. Morandotti, Physics Reports 729 (1), 1 (2018).
[48] P. Del'Haye, A. Schliesser, O. Arcizet, T. Wilken, R. Holzwarth, T.J. Kippenberg, "Optical frequency comb
generation from a monolithic microresonator", Nature 450 1214 -- 1217 (2007).
[49] M.Peccianti, A.Pasquazi, Y.Park, B.E Little, S.Chu, D.J Moss, and R.Morandotti, "Demonstration of an ultrafast
nonlinear microcavity modelocked laser", Nature Communications, 3 765 (2012). DOI:10.1038/ncomms1762
[50] M.Kues, et. al., "Passively modelocked laser with an ultra-narrow spectral width", Nature Photonics, 11 (3) 159
(2017). DOI:10.1038/nphoton.2016.271
[51] A. Pasquazi, L. Caspani, M. Peccianti et al., "Self-locked optical parametric oscillation in a CMOS compatible
microring resonator: a route to robust optical frequency comb generation on a chip," Optics Express, 21 (11), 13333-
13341 (2013).
[52] A. Pasquazi, M. Peccianti, B. E. Little et al., "Stable, dual mode, high repetition rate mode-locked laser based on a
microring resonator," Optics Express, 20 (24), 27355-27362 (2012).
[53] C. Reimer, L. Caspani, M. Clerici et al., "Integrated frequency comb source of heralded single photons," Optics
Express, 22 (6), 6535-6546 (2014).
[54] C.Reimer, et al., "Cross-polarized photon-pair generation and bi-chromatically pumped optical parametric oscillation
on a chip", Nature Communications, 6 Article 8236 (2015). DOI: 10.1038/ncomms9236
[55] L. Caspani, C. Reimer, M. Kues et al., "Multifrequency sources of quantum correlated photon pairs on-chip: a path
toward integrated Quantum Frequency Combs," Nanophotonics, 5 (2), 351-362 (2016).
[56] C. Reimer, M. Kues, P. Roztocki et al., "Generation of multiphoton entangled quantum states by means of integrated
frequency combs," Science, 351 (6278), 1176-1180 (2016).
[57] M.Kues, et al., "On-chip generation of high-dimensional entangled quantum states and their coherent control",
Nature, 546 (7660) 622-626 (2017).
[58] P. Roztocki, M. Kues, C. Reimer et al., "Practical system for the generation of pulsed quantum frequency combs,"
Optics Express, 25 (16), 18940-18949 (2017).
[59] P.Marin-Palomo, et al., "Microresonator-based solitons for massively parallel coherent optical communications",
Nature 546 (7657) 274 (2017).
[60] DT Spencer et al., "An optical-frequency synthesizer using integrated photonics", Nature 557 (7703) 81-85 (2018).
[61] J Pfeifle, V Brasch, M Lauermann, Y Yu, D Wegner, T Herr, K Hartinger, et al., "Coherent terabit communications
with microresonator Kerr frequency combs", Nature Photonics 8 (5), 375-380 (2014).
[62] P Del'Haye, T Herr, E Gavartin, ML Gorodetsky, R Holzwarth, et al., "Octave spanning tunable frequency comb
from a microresonator", Physical Review Letters 107 (6), 063901 (2011).
[63] TJ Kippenberg, R Holzwarth, SA Diddams, "Microresonator-based optical frequency combs", Science 332 (6029),
555-559 (2011).
[64] T Herr, V Brasch, JD Jost, CY Wang, NM Kondratiev, ML Gorodetsky, et al., "Temporal solitons in optical
microresonators", Nature Photonics 8 (2), 145-152 (2014).
[65] F Ferdous, H Miao, DE Leaird, K Srinivasan, J Wang, L Chen, ... "Spectral line-by-line pulse shaping of on-chip
microresonator frequency combs", Nature Photonics 5 (12), 770 (2011).
[66] X Xue, PH Wang, Y Xuan, M Qi, AM Weiner, "Microresonator Kerr frequency combs with high conversion
efficiency", Laser & Photonics Reviews 11 (1) (2017).
[67]Y Xuan, Y Liu, LT Varghese, AJ Metcalf, X Xue, PH Wang, K Han, ... "High-Q silicon nitride microresonators
exhibiting low-power frequency comb initiation", Optica 3 (11), 1171-1180 (2016).
[68] X Xue, M Qi, AM Weiner, "Normal-dispersion microresonator Kerr frequency combs", Nanophotonics 5 (2), 244-
262 (2016).
[69] C.Grillet, et al., "Amorphous Silicon Nanowires with Record High Nonlinearity, FOM, and Optical Stability", Optics
Express 20 (20) 22609-22615 (2012). DOI: 10.1364/OE.20.022609.
[70] J. Capmany, and D. Novak, "Microwave photonics combines two worlds," Nature Photonics, 1(6), 319-330 (2007).
[71] J. Capmany, J. Mora, I. Gasulla, J. Sancho, J. Lloret, and S. Sales, "Microwave Photonic Signal Processing," Journal
of Lightwave Technology, 31, (4) pp. 571-586 (2013).
[72] A. J. Seeds, "Microwave photonics," IEEE Transactions on Microwave Theory and Techniques, 50 (3), 877-887
(2002).
[73] R. C. Williamson, and R. D. Esman, "RF photonics," Journal of Lightwave Technology, 26 (9-12), 1145-1153 (2008).
[74] J. P. Yao, "Microwave Photonics," IEEE Journal of Lightwave Technology, 27 (1-4), 314-335 (2009).
[75] V. R. Supradeepa, C.M. Long, R.Wu, F.Ferdous, E.Hamidi, D.E. Leaird, and A.M. Weiner, "Comb-based
radiofrequency photonic filters with rapid tunability and high selectivity," Nature Photonics 6, 186 -- 194 (2012).
[76] V. R. Supradeepa, C. M. Long, R. Wu, F. Ferdous, E. Hamidi, D. E. Leaird, and A. M. Weiner, "Comb-based
radiofrequency photonic filters with rapid tunability and high selectivity," Nature Photonics, 6, (3) 186-194 (2012).
[77] MH Khan, H Shen, Y Xuan, L Zhao, S Xiao, DE Leaird, AM Weiner, M Qi, "Ultrabroad-bandwidth arbitrary
radiofrequency waveform generation with a silicon photonic chip-based spectral shaper", Nature Photonics 4 (2),
117 (2010).
[78] A Fülöp, M Mazur, A Lorences-Riesgo, TA Eriksson, PH Wang, Y Xuan, ... "Long-haul coherent communications
using microresonator-based frequency combs", Optics Express 25 (22), 26678-26688 (2017).
[79] V. Torres-Company and A.M. Weiner, "Optical frequency comb technology for ultra-broadband radio-frequency
photonics," Laser & Photonics Reviews 8 (3) 368-393 (2014).
[80] D. Marpaung, M. Pagani, B. Morrison et al., "Nonlinear Integrated Microwave Photonics," IEEE Journal of
Lightwave Technology, 32 (20), 3421-3427 (2014).
[81] Q. Z. Cen, Y. T. Dai, F. F. Yin et al., "Rapidly and continuously frequency-scanning opto-electronic oscillator,"
Optics Express, 25 (2), 635-643 (2017).
[82] J. Dai, X. Y. Xu, J. L. Ke et al., "Self-Oscillating Triangular Pulse Generator Based on 90 degrees Photonic-Assisted
Phase Shifter," IEEE Photonics Technology Letters, 29 (3), 271-274 (2017).
[83] J. Dai, X. Y. Xu, Z. L. Wu et al., "Self-oscillating optical frequency comb generator based on an optoelectronic
oscillator employing cascaded modulators," Optics Express, 23 (23), 30014-30019 (2015).
[84] K. Xu, R. X. Wang, Y. T. Dai et al., "Microwave photonics: radio-over-fiber links, systems, and applications,"
Photonics Research, 2 (4), B54-B63 (2014).
[85] X. Y. Xu, J. Dai, Y. T. Dai et al., "Broadband and wide-range feedback tuning scheme for phase-locked loop
stabilization of tunable optoelectronic oscillators," Optics Letters, 40 (24), 5858-5861 (2015).
[86] A. X. Zhang, Y. T. Dai, F. F. Yin et al., "Stable radio-frequency delivery by lambda dispersion-induced optical
tunable delay," Optics Letters, 38 (14), 2419-2421 (2013).
[87] S. Mansoori, and A. Mitchell, "RF transversal filter using an AOTF," IEEE Photonics Technology Letters, 16 (3),
879-881 (2004).
[88] H. Emami, N. Sarkhosh, L. A. Bui et al., "Wideband RF photonic in-phase and quadrature-phase generation," Optics
Letters, 33 (2), 98-100 (2008).
[89] S. L. Pan, and J. P. Yao, "Optical generation of polarity- and shape-switchable ultrawideband pulses using a chirped
intensity modulator and a first-order asymmetric Mach-Zehnder interferometer," Optics Letters, 34 (9), 1312-1314
(2009).
[90] D. Marpaung, C. Roeloffzen, R. Heideman et al., "Integrated microwave photonics," Laser & Photonics Reviews, 7
(4), 506-538 (2013).
[91] E. Hamidi, D. E. Leaird, and A. M. Weiner, "Tunable Programmable Microwave Photonic Filters Based on an Optical
Frequency Comb," IEEE Transactions on Microwave Theory and Techniques, 58 (11), 3269-3278 (2010).
[92] J. Y. Wu, P. Cao, X. F. Hu et al., "Compact tunable silicon photonic differential-equation solver for general linear
time-invariant systems," Optics Express, 22 (21), 26254-26264 (2014).
[93] J. Y. Wu, P. Cao, X. F. Hu et al., "Nested Configuration of Silicon Microring Resonator With Multiple Coupling
Regimes," IEEE Photonics Technology Letters, 25 (6), 580-583 (2013).
[94] J. Y. Wu, P. Cao, T. Pan et al., "Compact on-chip 1 x 2 wavelength selective switch based on silicon microring
resonator with nested pairs of subrings," Photonics Research, 3 (1), 9-14 (2015).
[95] J. Y. Wu, B. Y. Liu, J. Z. Peng et al., "On-Chip Tunable Second-Order Differential-Equation Solver Based on a
Silicon Photonic Mode-Split Microresonator," IEEE Journal of Lightwave Technology, 33 (17), 3542-3549 (2015).
[96] J. Y. Wu, J. Z. Peng, B. Y. Liu et al., "Passive silicon photonic devices for microwave photonic signal processing,"
Optics Communications, 373, 44-52 (2016).
[97] J.Wu, T.Moein, X.Xu, G.Ren, A. Mitchell, and D.J. Moss, "Micro-ring resonator quality factor enhancement via an
integrated Fabry-Perot cavity", Applied Physics Letters (APL) Photonics, 2 056103 (2017).
[98] O. F. Yilmaz, L. Yaron, S. Khaleghi et al., "True time delays using conversion/dispersion with flat magnitude
response for wideband analog RF signals," Optics Express, 20 (8), 8219-8227 (2012).
[99] R. Wu, V. Torres-Company, D. E. Leaird et al., "Supercontinuum-based 10-GHz flat-topped optical frequency comb
generation," Optics Express, 21 (5), 6045-6052 (2013).
[100] A. J. Metcalf, V. Torres-Company, D. E. Leaird et al., "High-Power Broadly Tunable Electrooptic Frequency Comb
Generator," IEEE Journal of Selected Topics in Quantum Electronics, 19 (6), (2013).
[101] Y. C. Han, Z. Li, and J. P. Yao, "A Microwave Bandpass Differentiator Implemented Based on a Nonuniformly-
Spaced Photonic Microwave Delay-Line Filter," IEEE Journal of Lightwave Technology, 29 (22), 3470-3475
(2011).
[102] J. Azana, C. Madsen, K. Takiguchi et al., "Guest editorial - Optical signal processing," IEEE Journal of Lightwave
Technology, 24 (7), 2484-2486 (2006).
[103] R. Slavik, Y. Park, M. Kulishov et al., "Ultrafast all-optical differentiators," Optics Express, 14 (22), 10699-10707
(2006).
[104] J. Azana, "Ultrafast Analog All-Optical Signal Processors Based on Fiber-Grating Devices," IEEE Photonics
Journal, 2 (3), 359-386 (2010).
[105] R. Ashrafi, M. H. Asghari, and J. Azana, "Ultrafast Optical Arbitrary-Order Differentiators Based on Apodized
Long-Period Gratings," IEEE Photonics Journal, 3 (3), 353-364 (2011).
[106] R. Ashrafi, and J. Azana, "Figure of merit for photonic differentiators," Optics Express, 20 (3), 2626-2639 (2012).
[107] J. McClellan, T. Parks, and L. Rabiner, "A computer program for designing optimum FIR linear phase digital
filters," IEEE Transactions on Audio and Electroacoustics, 21 (6), 506-526 (1973).
[108] T. Herr, K. Hartinger, J. Riemensberger et al., "Universal formation dynamics and noise of Kerr-frequency combs
in microresonators," Nature Photonics, 6 (7), 480-487 (2012).
[109] L.Di Lauro, J.Li, D.J. Moss, R.Morandotti, S.T. Chu, M.Peccianti, A.Pasquazi, "Parametric Control of Thermal
Self-Pulsation in Microcavity", Optics Letters 42 (17) 3407 -- 3410 (2017).
[110] J. J. Zhang, and J. P. Yao, "Photonic True-Time Delay Beamforming Using a Switch-Controlled Wavelength-
Dependent Recirculating Loop," Journal of Lightwave Technology, 34 (16), 3923-3929 (2016).
[111] M. Longbrake, "True Time-Delay Beamsteering for Radar," Proceedings of the 2012 IEEE National Aerospace
and Electronics Conference (Naecon), pp. 246-249, 2012.
[112] M. I. Skolnik, "Introduction to radar systems," 3rd ed. New York, NY: McGraw-Hill, 2001.
[113] J. Capmany, B. Ortega, and D. Pastor, "A tutorial on microwave photonic filters," Journal of Lightwave
Technology, 24, (1) 201-229 (2006).
[114] J. Capmany, B. Ortega, D. Pastor, and S. Sales, "Discrete-time optical processing of microwave signals," Journal
of Lightwave Technology, 23 (2) 702-723 (2005).
[115] A. Ortigosa-Blanch, J. Mora, J. Capmany, B. Ortega, and D. Pastor, "Tunable radio-frequency photonic filter based
on an actively mode-locked fiber laser," Optics Letters, 31, (6) 709-711 (2006).
[116] W. L. Liu, M. Li, R. S. Guzzon, E. J. Norberg, J. S. Parker, M. Z. Lu, L. A. Coldren, and J. P. Yao, "A fully
reconfigurable photonic integrated signal processor," Nature Photonics, 10 (3) 190 (2016).
[117] HJ Kim, DE Leaird, AM Weiner, "Rapidly tunable dual-comb RF photonic filter for ultrabroadband RF spread
spectrum applications", IEEE Transactions on Microwave Theory and Techniques 64 (10), 3351-3362 (2016).
[118]J.Wu, X.Xu, R., and D. J. Moss, "RF photonics: An optical micro-combs' perspective", IEEE Journal of Selected
Topics in Quantum Electronics 24 (4) 1-20 (2018). DOI: 10.1109/JSTQE.2018.2805814.
[119]X. X. Xue, Y. Xuan, H. J. Kim, J. Wang, D. E. Leaird, M. H. Qi, and A. M. Weiner, "Programmable Single-Bandpass
Photonic RF Filter Based on Kerr Comb from a Microring," IEEE Journal of Lightwave Technology, 32 (20) 3557-
3565 (2014).
[120]X.X.Xue and A.M. Weiner, "Microwave photonics connected with microresonator frequency combs", Frontiers of
Optoelectronics 9 (2) 238-248 (2016).
[121]X Xue, Y Xuan, C Bao, S Li, X Zheng, B Zhou, M Qi, AM Weiner, "Microcomb-based true-time-delay network for
microwave beamforming with arbitrary beam pattern control", Journal of Lightwave Technology 36 (12), 2312-2321
(2018).
[122]X.Xu, J.Wu, M.Shoeiby, T.G. Nguyen, S.T. Chu, B.E. Little, R. Morandotti, A.Mitchell, and D.J. Moss, "Advanced
RF and microwave functions based on an integrated optical frequency comb source", Optics Express 26 (3), 2569-
2583 (2018).
[123]T. G. Nguyen, M. Shoeiby, S. T. Chu et al., "Integrated frequency comb source based Hilbert transformer for
wideband microwave photonic phase analysis," Optics Express, 23 (17), 22087-22097 (2015).
[124]X.Xu, et al., "Reconfigurable broadband microwave photonic intensity differentiator based on an integrated optical
frequency comb source", Applied Physics Letters (APL) Photonics, 2 (9) 096104 (2017). DOI: 10.1063/1.4989871.
[125]X. X. Xue, Y. Xuan, H. J. Kim et al., "Programmable Single-Bandpass Photonic RF Filter Based on a Kerr Comb
from a Microring," IEEE Journal of Lightwave Technology, 32 (20), 3557-3565 (2014).
[126]X.Xu, J.Wu, T.G. Nguyen, S.T. Chu, B.E. Little, R.Morandotti, A.Mitchell, and D. J. Moss, "Continuously tunable
orthogonally polarized optical RF single sideband generator and equalizer based on an integrated micro-ring
resonator", submitted, IEEE Journal of Lightwave Technology (JLT) 36 (12) p (2018).
[127] X. Y. Xu, J. Y. Wu, T. G. Nguyen, S. T. Chu, B. E. Little, R. Morandotti, A. Mitchell, and D. J. Moss, "Integrated
optical power equalizer based on a dual-polarization micro-ring resonator," in CLEO: QELS_Fundamental Science,
San Jose, CA, May 2018, Paper JW2A.22.
[128]X.Xu, J.Wu, T.G. Nguyen, S.T. Chu, B.E. Little, R.Morandotti, A.Mitchell, and D.J. Moss, "Broadband RF
Channelizer based on an Integrated Optical Frequency Comb Source", IEEE Journal of Lightwave Technology (JLT)
Special Issue on Microwave Photonics 36 (12) p (2018). DOI: 10.1109/JLT.2018.2819172
[129]X.Xu, J.Wu, T.G. Nguyen, T.Moein, S.T. Chu, B. E. Little, R.Morandotti, A.Mitchell, and D.J. Moss, "Photonic
microwave true time delays for phased array antennas using a 49GHz FSR integrated optical micro-comb source",
OSA Photonics Research Journal, Special Issue on Integrated Nonlinear Photonics 6 (5) B30-B36 (2018).
DOI:10.1364/PRJ.6.000B30
[130] D. H. Yang andW. P. Lin, "Phased-array beam steering using optical true time delay technique," Opt. Commun.,
350 90 -- 96 (2015).
[131] X. W. Ye, F. Z. Zhang, and S. L. Pan, "Optical true time delay unit for multi-beamforming," Opt. Express, 23, (8)
10002 -- 10008 (2015).
[132] Y. Q. Liu, J. P. Yao, and J. L. Yang, "Wideband true-time-delay unit for phased array beamforming using discrete-
chirped fiber grating prism," Opt. Commun., 207, (1 -- 6) 177 -- 187, (2002).
[133] S. Chin et al., "Broadband true time delay for microwave signal processing, using slow light based on stimulated
Brillouin scattering in optical fibers," Opt. Express, 18, (21) 22599 -- 22613 (2010).
[134] X. Xue et al., "Microcomb-based microwave photonics," in Proc. Asia Commun. Photon. Conf., Guangzhou, China,
2017, Paper Su3E.2.
[135] M. Longbrake, "True time-delay beamsteering for radar," in Proc. 2012 IEEE Nat. Aerosp. Electron. Conf., 2012,
pp. 246 -- 249.
[136] X. W. Ye, F. Z. Zhang, and S. L. Pan, "Compact optical true time delay beamformer for a 2D phased array antenna
using tunable dispersive elements," Opt. Lett., 41, (17) 3956 -- 3959 (2016).
|
1912.13479 | 1 | 1912 | 2019-11-27T16:13:03 | Alignment Tolerant Broadband Compact Taper for Low-Loss Coupling to a Silicon-on-Insulator Photonic Wire Waveguide | [
"physics.app-ph",
"physics.optics"
] | We experimentally demonstrate a broadband, fabrication tolerant, CMOS compatible compact silicon waveguide taper (34.2 um) in silicon-on-insulator wire waveguides. The taper works on multi-mode interference along the length of the taper. A single taper design has a broadband operation with coupling efficiency >70% over 700 nm that can be used in O, C and L-band. The compact taper is highly tolerant to fabrication variations; >100 nm change in the taper and end waveguide width varies the taper transmission by <5%. The footprint of the device i.e. taper along with the linear gratings is ~ 442 m2; i.e. 11.5X smaller than the adiabatic taper. The taper with linear gratings provides comparable coupling efficiency as standardly used focusing gratings. We have also compared the translational and rotational alignment tolerance of the focusing grating with the linear grating. | physics.app-ph | physics | Alignment Tolerant Broadband Compact Taper
for Low-Loss Coupling to a Silicon-on-
Insulator Photonic Wire Waveguide
PURNIMA SETHI AND SHANKAR KUMAR SELVARAJA
1Centre for Nano Science and Engineering (CeNSE), Indian Institute of Science Bangalore, India.
*[email protected], [email protected]
Abstract: We experimentally demonstrate a broadband, fabrication tolerant compact silicon
waveguide taper (34.2 µm) in silicon-on-insulator. The taper works on multi-mode interference
along the length of the taper. A single taper design has a broadband operation with coupling
efficiency >70% over 700 nm that can be used in O, C and L-band. The compact taper is highly
tolerant to fabrication variations; ±100 nm change in the taper and end waveguide width varies
the taper transmission by <5%. The footprint of the device i.e. taper along with the linear
gratings is ~ 442 m2; 11.5X smaller than the adiabatic taper. The taper with linear gratings
provides comparable coupling efficiency as standardly used focusing gratings. We have also
compared the translational and rotational alignment tolerance of the focusing grating with the
linear grating.
1.
Introduction
Photonics in silicon-on-insulator (SOI) platform is a tremendously promising technology due
to its comprehensive applications owing to its compactness and CMOS-compatibility [1, 2].
Consequently, it has been the focus of a considerable research suited for enabling the
integration of highly complex optical circuits for making compact devices. The strong light
confinement in high index-contrast waveguide platform ensues dense optical integration with
sub-micron dimensions and low bending losses which brings in new challenges in circuit design
and routing [3-5].
The building block of an optical device/circuit is an optical waveguide which enables low-
loss light propagation and is thereby, used to connect components and devices. Waveguides are
generally designed with different cross-sections to realize various integrated photonic device
such as, arrayed-waveguide gratings, spot-size converters, multimode interference couplers,
grating couplers (GCs) as well as crossings [6-10]. The devices with different waveguide width
should be connected through a low-loss interface. When footprint of these waveguide
transitions. Since the taper length depends predominantly on the starting and ending waveguide
width and the effective refractive index, the transition between a GC and a single-mode
photonic waveguide in a SOI platform is substantial [11-15].
A grating footprint of 10 µm ×10 µm is typically chosen to mode-match the grating field
with an optical fiber. The grating is then coupled to a waveguide through an adiabatic/non-
adiabatic taper [16-25]. The function of the taper is to change the optical mode size and shape
to achieve high coupling efficiency between the two waveguides of different cross-sections. In
an adiabatic taper, the local first-order mode of the waveguide should propagate through the
taper without coupling to higher-order modes and radiatiing modes. The adiabatic tapers in SOI
wire waveguides are generally 300-500 µm long. Several designs of GC based adiabatic tapers
have been proposed for SOI-based photonic devices, including linear [6, 16], exponential [9,
17], parabolic [18], and Gaussian [10]. However, the footprint of the spot-size converters based
on linear GCs is limited by the length of the taper.
To reduce the footprint of the coupler, a compact focusing grating is commonly used which
allows an 8X length reduction in the footprint (~18.5 µm × 28 µm) without performance
penalty compared to a linear GC with an adiabatic taper [26]. However, focusing gratings
require accurate fiber alignment, are bandwidth limited and suffer from reflection [27].
Thus, it would be exceedingly beneficial to use a linear GC with a short taper with low-
insertion loss, low-reflection, broadband, alignment tolerant as well as robust to fabrication
imperfections for a compact light-chip coupling scheme. Therefore, designing an improved
waveguide taper for obtaining an efficient coupling between two different optical waveguides
is essentially indispensable.
In the previous paper, we have shown an ultra-compact taper between a linear GC and single
mode silicon waveguide using a quadratic sinusoidal function, merely 15 µm long with an
insertion loss as low as 0.22 dB at 1550 nm and a bandwidth >150 nm [28]. However, the tapers
were designed for shallow-etched waveguides. Wire waveguides allow low-loss sharp bends
and thus, ultra-dense photonic circuits. In this paper, we propose a taper for wire waveguide
with the combined advantages of broadband operation as well as compactness.
Fig. 1. Schematic of the compact taper structure for wire waveguide on SOI platform.
2. Compact Taper: Design and Simulation
The schematic of the taper structure along with GCs is shown in Fig. 1. The proposed taper
works on self-imaging principle in multi-mode waveguides along the propagating length [28,
29]. The length and width of the taper are optimized to obtain interference progressively
between the resonance modes along the taper resulting in maximum coupling to the
fundamental waveguide mode. The interpolation formula used to define the proposed taper to
connect a broad waveguide to a submicron waveguide section is [28, 29],
𝑋 = 𝑎 [𝑏𝑧2 + (1 − 𝑏)𝑧] + (1 − 𝑎)[sin2 (
𝑐𝜋
2
𝑧)] (1)
where a lies between 0 to 1, b (used for fine tuning the optimal response) lies between -1 to 1,
c is any odd integer 3 (c = 1 creates the trivial case of half a sinusoidal oscillation). This formula
meets the following boundary conditions: X (z = 0) = 0 and X (z = 1) = 1 where z is the relative
length of the taper. All four (a, b, c, z) design parameters allow one to design an appropriate
taper profile for maximum transmission between the waveguides. The iterative feedback-based
approach allows for lower design cycles as finer parameter spacing is required only near the
optimum. The approach also greatly reduces the number of simulations, thereby reducing
design time.
Table 1. Optimized parameters for the proposed taper
Length of the Taper
14.2 µm + 20 µm
'a' variable
'b' variable
'c' variable
Width of Initial Waveguide
Width of Final Waveguide
Wavelength
Efficiency
0.4
0.5
7
10 µm
500 nm
1.55 µm
94.7%
Fig. 2. Optical intensity profiles for the optimized compact taper at 1550 nm TE polarization
(b) 2-D contour plot profile along the length (L) of the compact taper.
The simulations were carried out using Eigenmode Expansion (EME) propagation algorithm
and overlap between modes is computed using Finite Difference Method (FDM). The compact
taper was designed in an SOI wafer with 220 silicon on 2000 nm buffer oxide. Table 1 shows
the optimized values of the relevant taper design parameters. A maximum coupling efficiency
of 94.7% is achieved for a taper length of 34.2 µm shown by the optical intensity profile in
Figure 2(a). Figure 2(b) illustrates the evolution of modes along the length (L = 34.2 µm) of
the proposed taper.
Figure 3(a) shows the effect of end waveguide width variation from an optimized width of
500 nm on the transmission. The first four modes propagating in the waveguide are also shown
in the inset. As is evident, > 75% transmission is achieved for a variation of ±200 nm. However,
in practice one can expect a linewidth variation of < 10% which corresponds to a width change
of ±25 nm. A variation in this range would result in transmission degradation by < 2% (0.08
dB), which shows the resilience of the proposed taper.
Figure 3(b) depicts the spectral response of the compact taper based on linear GCs. The
inset shows the higher order modes for a wavelength span of 1000 nm. The proposed taper has
a broadband operation with the 3dB bandwidth > 900 nm covering O, C, L-band and beyond.
Furthermore, the effect of dimensional variation on the transmission performance was also
calculated to take fabrication tolerances into account. Figure 3(c) shows the effect of the total
taper width variation on the coupling efficiency obtained by varying the optimized a value. As
is evident, the tapers are very resilient (> 80% efficiency for ± 500 nm shift in optimized taper
width). During fabrication, a width variation of ±25 nm (25%) may occur, which results in a
transmission degradation by merely < 1%. Figure 3(d) shows the effect of the total taper width
variation on the coupling efficiency obtained by varying the optimized b value. The proposed
structures have high manufacturing tolerances (> 80% efficiency with shift in optimized taper
width of ± 200 nm).
Fig. 3. (a) Variation in taper's efficiency with waveguide width. Inset shows the variation for
higher order modes, (b) Spectral response of the proposed compact taper in the C & L-band
(1480 nm -- 1640 nm) and beyond. Inset shows the broadband 1000 nm range for higher order
modes; Tolerance of the proposed taper i.e. effect of compact taper width variation with (c) 'a'
and (d) 'b'.
Fig. 4. Scanning Electron Microscope (SEM) of the proposed compact taper structure along an
Adiabatic taper shoulder of length (a) 20 µm, (b) 2 µm.
3. Experimental Results and Discussion
To compare the proposed fiber-to-waveguide taper performance with the existing designs, three
combinations of tapers were fabricated; (i) linear GCs with adiabatic taper, (ii) focused GCs
and (iii) the proposed compact taper with linear GCs. The test structures were designed with an
input GC coupling into a 500 nm wire waveguide and tapering-out to an identical output coupler
configuration. All the GCs were designed for TE-polarized 1550 nm with grating period of 630
nm and 50% fill-factor [30].
The devices were fabricated using electron-beam lithography and Inductively Coupled
Plasma-Reactive Ion Etching (ICP-RIE) process. Pattering was done on a standard SOI
substrate with a 220 nm thick device layer on a 2 µm BOX layer. Figure 4 shows the scanning
electron microscope (SEM) image of the proposed compact taper. Figure 4(a) shows the
proposed structure along with a 20 µm long adiabatic taper. The taper shoulders along the wire
waveguide aids in confining the mode. Figure 4(b) shows the proposed structure with a 2 µm
long adiabatic taper. However, this configuration was less efficient, since the taper is short and
hence, confinement is poor.
The fabricated devices were characterized using a tunable laser source (1510-1630 nm) and
a photodetector. The polarization of the light from the laser source is controlled using
polarization wheels before the input GC. The transmitted light is detected by an InGaAs
photodetector. Figure 5 and Table 2 shows the summary of the characterization results. In order
to see the tolerance to grating period variation, 5 set of devices were fabricated with period of
590, 610, 630, 650, and 670 nm.
Figure 5(a) compares the performance of the compact taper with the long tapers and focused
GCs for different grating periods. The performance of the compact taper is marginally better
than the long adiabatic taper. Although, focusing gratings are more efficient, their tolerance to
fabrication imperfections is less. The efficiency of the focusing gratings, linear grating and
compact taper degrades by 2.07/5.17, 1.98/3.58, 2.3/2.6 dB per coupler for a period shift of
±40 nm. The 3-dB bandwidth which is another important performance metric for a GC is ~ 5
nm higher for compact taper compared to a focusing GC for 630 nm period. The insertion loss
per coupler is 6.2 dB, 6.32 dB and 5.73 dB for GC with compact taper, GC with adiabatic taper
and focusing GC respectively. The insertion loss of the adiabatic long taper is slightly higher,
which we can attribute to the waveguide loss in the adiabatic section.
Fig. 5(b) shows the response of the proposed taper and adiabatic taper by subtracting the
patch response. Using the patch waveguides, after deducting the coupler loss, we observe
insertion loss <0.7 dB and <0.8 dB per taper for a compact taper and adiabatic taper
respectively. The performance of the proposed taper is marginally better than the adiabatic with
93% reduction in length.
Fig. 5. Coupling efficiency of the various configuration of the GCs with variation in the period,
(b) Insertion loss of the taper alone by neglecting the loss of the GC through a patch structure.
Table 2. Characterization result of the proposed devices with different periods
Long
Taper
Focusing
Grating
Compact
Taper
Period
(nm)
Coupling
3 dB
Footprint
Efficiency per
Bandwidth
Coupler
590
610
630
650
670
590
610
630
650
670
590
610
630
650
670
9.9
5.9
6.32
6.9
8.3
10.9
6.6
5.73
6.49
7.8
8.8
6.5
6.2
6.9
8.5
210
40
44.2
50
52
54
52
56
50
50
53
50
52
51
56
58
54
56
Table 3. Change in efficiency with Rotational and Translational Misalignment for linear and focusing gratings
Rotational
Insertion
3-dB
Translational
Linear
Gratings
Focusing
Gratings
Misalignment
-100
-8
-6
-4
-2
0
2
4
6
8
10
-100
-8
-6
-4
-2
0
2
4
6
8
10
Loss
(dB)
13.9
12.8
11.9
11.7
7.87
5.43
6.12
7.45
11.47
12.4
12.1
19.1
21.0
18.1
13.2
7.61
5.29
6.1
10.56
16.9
19.2
18.3
Bandwidth
Shift (pm)
51 nm
50 nm
61 nm
57 nm
62 nm
62 nm
65 nm
60.8 nm
58.6 nm
60.8 nm
65 nm
53 nm
65 nm
68 nm
54 nm
58 nm
60.1 nm
58 nm
56 nm
42 nm
48.7 nm
52 nm
-800
-600
-400
-200
0
200
400
600
800
-800
-600
-400
-200
0
200
400
600
800
Insertion
Loss
(dB)
15.6
12.4
11.6
9.34
9.0
9.87
10.2
11.57
14.37
19.07
14.4
10.97
8.7
8.1
8.89
10.9
14.26
18.03
3-dB
Bandwidth
60 nm
40 nm
40 nm
40 nm
41 nm
40 nm
45 nm
51 nm
50 nm
70 nm
58 nm
44 nm
40.6 nm
42 nm
41 nm
44 nm
51 nm
61 nm
Table 3 compares the alignment tolerance of the focusing and the linear GCs. To obtain the
rotational alignment tolerance, the angle of the GCs were fabricated with a ∆ shift of 20
clockwise and anticlockwise (Fig. 6(a) and (b) shows a 20 anticlockwise shift in grating
placement). As is evident from Fig. 6(c), linear GCs are more tolerant with a roll-off of 2.613
dB/degree and 1.32 dB/degree (linear part) for focusing and linear gratings respectively.
Fig. 6. (a) and (b) SEM image of the rotational misalignment in the focusing and linear gratings
by 20, Change in coupling efficiency with (c) Rotational misalignment and (d) Translational
misalignment of the focusing and linear gratings.
Table 4. Various adiabatic and non-adiabatic grating assisted tapers on SOI proposed in literature
Taper Designs
Length
(µm)
Initial Width (µm)
-Final Width (µm)
Coupling
Efficiency
Remarks
Adiabatic
Linear
20-200
10 -- 0.5
44.9-98.5%
Exponential (Positive)
20-200
10 -- 0.5
14-99.5%
Exponential
(Negative)
Parabolic
Efficient Adiabatic19
Adiabatic Taper
Based on Thin Flat
Focusing Lenses20
200
200
120
22.5
10 -- 0.5
48.5-97.4%
10 -- 0.5
12 -- 0.5
10 -- 0.5
32.4-98.6%
98.3%
95.4%
Hollow tapered spot-
60
15 -- 0.3
72%
Non-
Adiabatic
size converter21
Segmented22
Segmented- Stepwise
New23
15.4
20
10 -- 0.56
12 -- 0.5
Lens-assisted24
20
10 -- 0.45
Discontinuous25
10 -- 0.45
Proposed Taper
34.2
10 -- 0.5
98.3%
92.1%
1 dB (TE),
5 dB (TM)
(Experimental)
90%
95%
Trade-off between the
taper length and coupling
efficiency due to the
adiabatic transition
Silicon subwavelength
slits are fabrication
alignment intolerant
Complexity in fabrication
Robust to Fabrication
Errors, Broadband,
Efficient
Vermeulen et. al. have shown tilted GCs to minimize Fresnel back-reflections by designing the
grating teeth such that reflections are directed away from the aperture of the focusing GC [31].
However, we have arbitrarily varied the angle as in the case of overlay misalignment to inspect
the robustness of the gratings. Fig. 6(d) compares the translational alignment tolerance of the
focusing and linear GC where one of the fibers is aligned from the optimum position and the
change in efficiency is measured. As is evident, focusing gratings are less tolerant to fiber
alignment as compared to linear gratings. The roll-off for the linear part (since the graph is
parabolic in nature) is 13.5 dB/fm and 8.3 dB/fm for focusing and linear GC respectively. The
devices based on proposed compact tapers are 11 times smaller (~ 442 µm2) in comparison to
linear GC-based couplers (5100 µm2). Table 4 summarizes the various configurations of
adiabatic/non-adiabatic grating assisted tapers proposed in literature. The proposed tapers have
been optimized for the TE polarization only. However, similar structures would work for TM
coupling as well.
The compact lateral waveguide tapers are necessary to realize coupling between devices of
varying dimensions. We have designed and demonstrated a compact tapered spot-size converter
to couple light to a single mode waveguide from a 10 µm wide waveguide. By using the taper
with a linear GC, we have experimentally shown no degradation in coupling efficiency
compared to standard focusing GC. We have also shown that the proposed structure achieves
an improved 3 dB bandwidth of ~58 nm against ~53 nm (focusing GC) in the 1550 nm band.
The device shows 11X reduction in the footprint of a single device based on linear GCs using
adiabatic tapers. We have also shown the fabrication tolerance of the compact taper by varying
various parameters. Moreover, translation as well as rotational alignment tolerance of the
focusing and linear GC are also compared.
References
1.
S. Jahani, S. Kim, J. Atkinson, J. C. Wirth, F. Kalhor, A. A. Noman, W. D. Newman et al., "Controlling
evanescent waves using silicon photonic all-dielectric metamaterials for dense integration," Nature
communications 9(1), 1893-1902 (2018).
J. Witzens, "Silicon photonics: Modulators make efficiency leap," Nature Photonics 11(8), 459-462 (2017).
2.
3. D. Pérez, I. Gasulla, L. Crudgington, D. J. Thomson, A. Z. Khokhar, K. Li, W. Cao, G. Z. Mashanovich, and J.
Capmany, "Multipurpose silicon photonics signal processor core," Nature communications 8(1), 636-645
(2017).
4. W. Bogaerts, Wim, Martin Fiers, and Pieter Dumon, "Design challenges in silicon photonics," IEEE Journal of
Selected Topics in Quantum Electronics 20(4), 1-8 (2014).
5. T. Komljenovic, M. Davenport, J. Hulme, A. Y. Liu, C. T. Santis, A. Spott, S. Srinivasan, E. J. Stanton, C.
Zhang, and J. E. Bowers, "Heterogeneous silicon photonic integrated circuits," J. Lightw. Technol. 34(1), 20-
35(2016).
S. Pathak, D. V. Thourhout, and W. Bogaerts, "Design trade-offs for silicon-on-insulator-based AWGs for
(de)multiplexer applications," Opt. Lett. 38(16), 2961 -- 2964 (2013).
6.
7. H. F. Zhou, J. F. Song, C. Li, H. J. Zhang, and P. G. Lo, "A library of ultra-compact multimode interference
8.
optical couplers on SOI," IEEE Photon. Technol. Lett. 25(12), 1149 -- 1152 (2013).
S. Nambiar, P. Sethi, S. K. Selvaraja, "Grating-Assisted Fiber to Chip Coupling for SOI Photonic Circuits"
Applied Sciences 8, 11421164 (2018).
9. C. H. Chen and C. H. Chiu, "Taper-integrated multimode-interference based waveguide crossing design," IEEE
J. Quantum Electron. 46(11), 1656 -- 1661 (2010).
10. W. Bogaerts, P. Dumon, D. V. Thourhout, and R. Baets, "Low-loss, lowcross-talk crossings for silicon-on-
insulator nanophotonic waveguides," Opt. lett., 32(19), 2801-2803 (2007).
11. B. J. Offrein, G-L Bona, R. Germann, I. Massarek and D. Erni "A very short planar silica spot-size converter
using a nonperiodic segmented waveguide," J. Lightw. Technol. 16(9), 1680-1684 (1998).
12. B. Luyssaert, P. Bienstman, P. Vandersteegen, P. Dumon, and R. Baets, "Efficient non-adiabatic planar
waveguide tapers," IEEE J. of Lightwave Technol. 23 (8), 2462-2468 (2005).
13. D. Chen, X. Xiao, L. Wang, Y. Yu, W. Liu and Q. Yang, "Low-loss and fabrication tolerant silicon mode-order
converters based on novel compact tapers," Opt. Exp. 23 (9), 11152-11159 (2015).
14. K. Shiraishi, H. Yoda, A. Ohshima, H. Ikedo, and C. S. Tsai, "A silicon based spot-size converter between
single-mode fibers and Si-wire waveguides using cascaded tapers," Appl. Phys. Lett. 91 (14), 141120-22
(2007).
15. M. Yangjin, Y. Zhang, S. Yang, A. Novack, R. Ding, A. E-J Lim, G-Q Lo, T. B. Jones, and M. Hochberg,
"Ultralow loss single layer submicron silicon waveguide crossing for SOI optical interconnect," Opt. Express
21(24), 29374-29382 (2013).
16. F. V. Laere, G. Roelkens, M. Ayre, J. Schrauwen, D. Taillaert, D. V. Thourhout, T. F. Krauss, and R. Baets,
"Compact and highly efficient grating couplers between optical fiber and nanophotonic waveguides," J. Lightw.
Technol. 25 (1), 151-156 (2007).
17. J. Wu, B. Shi, and M. Kong, "Exponentially tapered multi-mode interference couplers," Chinese Opt. Lett.,
4(3), 167-169 (2006).
18. T. Ye, Y. Fu, L. Qiao, and T. Chu, "Low-crosstalk Si arrayed waveguide grating with parabolic tapers," Opt.
Exp., 22 (26), 31899-31906 (2014).
19. Y. Fu, T. Ye, W. Tang, and T. Chu, "Efficient adiabatic silicon-on insulator waveguide taper," Photon. Res. 2,
A41-A44 (2014).
20. H. X. Yang, J. Huang, D. Chen, and Z. Zhaojian, "Ultrashort and efficient adiabatic waveguide taper based on
thin flat focusing lenses," Opt. Express 25(17), 19894-19903 (2017).
21. Md. Asaduzzaman, M. Bakaul, E. Skafidas, and Md. Rezwanul Haque Khandokar. "A compact silicon grating
coupler based on hollow tapered spot-size converter," Scientific reports 8(1), 2540 (2018).
22. B. Luyssaert, P. Vandersteegen, W. Bogaerts, P. Dumon, P. Sanchis, J. Marti, and R. Baets, "A versatile optical
spot-size converter design," in Proceedings of European Conference on Optical Communication. (Stockholm,
2004), 468 -- 469.
23. J. Zou, Y. Yu, M. Ye, L. Liu, S. Deng, X. Xu, and X. Zhang, "Short and efficient mode-size converter designed
by segmented-stepwise method," Opt. Lett. 39(21), 6273-6276 (2014).
24. K. Van Acoleyen and R. Baets, "Compact lens-assisted focusing tapers fabricated on silicon-on-insulator," in
Proceedings of IEEE Conference on Group IV Photonics (2011), 157 -- 159.
25. D. Vermeulen, K. Van Acoleyen, S. Ghosh, S. Selvaraja, W. A. D. de Cort, N. A. Yebo, E. Hallynck, K. de
Vos, P. P. P. Debackere, P. Dumon, G. Roelkens, D. Van Thourhout, and R. Baets, "Efficient tapering to the
fundamental quasi-TM mode in asymmetrical waveguides," in Proceedings of European Conference on
Integrated Optics, Cambridge, 2010, paper WeP16.
26. F. V. Laere, T. Claes, J. Schrauwen, S. Scheerlinck, W. Bogaerts, D. Taillaert, L. O'Faolain, D. V. Thourhout,
and R. Baets, "Compact focusing grating couplers for silicon-on-insulator integrated circuits," IEEE Photonics
Technol. Lett. 19(23), 1919 -- 1921 (2007).
27. S. R. Garcia, B. Shen, F. Merget, B. Marzban, and J. Witzens, "Alignment tolerant couplers for silicon
photonics," IEEE J. Sel. Top. Quantum Electron. 21(6), 765 -- 778 (2015).
28. P. Sethi, A. Haldar, and S.K. Selvaraja, "Ultra-compact low-loss broadband waveguide taper in silicon-on-
insulator," Opt. Express, 25(9), 10196-10203 (2017).
29. P. Sethi, R. Kallega, A. Haldar, and S. K. Selvaraja. "Compact broadband low-loss taper for coupling to a
silicon nitride photonic wire," Opt. Lett. 43(14), 3433-3436 (2018).
30. http://www. europractice-ic.com
31. D. Vermeulen, Y. De Koninck, Y. Li, E. Lambert, W. Bogaerts, R. Baets, and G. Roelkens. "Reflectionless
grating couplers for Silicon-on-Insulator photonic integrated circuits," Opt. Express 20(20), 22278-22283
(2012).
|
1905.04948 | 1 | 1905 | 2019-05-13T10:34:06 | Top-down fabrication of ordered arrays of GaN nanowires by selective area sublimation | [
"physics.app-ph",
"cond-mat.mes-hall"
] | We demonstrate the top-down fabrication of ordered arrays of GaN nanowires by selective area sublimation of pre-patterned GaN(0001) layers grown by hydride vapor phase epitaxy on Al$_{2}$O$_{3}$. Arrays with nanowire diameters and spacings ranging from 50 to 90 nm and 0.1 to 0.7 $\mu$m, respectively, are simultaneously produced under identical conditions. The sublimation process, carried out under high vacuum conditions, is analyzed \emph{in situ} by reflection high-energy electron diffraction and line-of-sight quadrupole mass spectromety. During the sublimation process, the GaN(0001) surface vanishes, giving way to the formation of semi-polar $\lbrace1\bar{1}03\rbrace$ facets which decompose congruently following an Arrhenius temperature dependence with an activation energy of ($3.54 \pm 0.07$) eV and an exponential prefactor of $1.58\times10^{31}$ atoms cm$^{-2}$ s$^{-1}$. The analysis of the samples by low-temperature cathodoluminescence spectroscopy reveals that, in contrast to dry etching, the sublimation process does not introduce nonradiative recombination centers at the nanowire sidewalls. This technique is suitable for the top-down fabrication of a variety of ordered nanostructures, and could possibly be extended to other material systems with similar crystallographic properties such as ZnO. | physics.app-ph | physics |
Cite this: DOI: 10.1039/C8NA00369F
Top-down fabrication of ordered arrays of GaN
nanowires by selective area sublimation
Sergio Fernández-Garrido,ࢩ Thomas Auzelle, Jonas Lähnemann, Kilian Wimmer,
Abbes Tahraoui, and Oliver Brandt
We demonstrate the top-down fabrication of ordered arrays of GaN nanowires by selective area
sublimation of pre-patterned GaN(0001) layers grown by hydride vapor phase epitaxy on Al2O3.
Arrays with nanowire diameters and spacings ranging from 50 to 90 nm and 0.1 to 0.7 µm, re-
spectively, are simultaneously produced under identical conditions. The sublimation process,
carried out under high vacuum conditions, is analyzed in situ by reflection high-energy electron
diffraction and line-of-sight quadrupole mass spectromety. During the sublimation process, the
GaN(0001) surface vanishes, giving way to the formation of semi-polar {1İ103} facets which de-
compose congruently following an Arrhenius temperature dependence with an activation energy
of (3.54 ± 0.07) eV and an exponential prefactor of 1.58 × 1031 atoms cm−2 s−1. The analysis of
the samples by low-temperature cathodoluminescence spectroscopy reveals that, in contrast to
dry etching, the sublimation process does not introduce nonradiative recombination centers at
the nanowire sidewalls. This technique is suitable for the top-down fabrication of a variety of
ordered nanostructures, and could possibly be extended to other material systems with similar
crystallographic properties such as ZnO.
1 Introduction
Since the first reports on the self-assembled formation of single-
crystalline GaN nanowires (NWs) in the late nineties, 1,2 increas-
ing efforts have been devoted to master the bottom-up fabrica-
tion of random and ordered arrays of group-III nitride NW het-
erostructures by metal-organic chemical vapor deposition and
molecular beam epitaxy. 3 The research in this field is motivated
by the technological relevance of GaN, the material that has en-
abled the commercialization of solid-state lighting for general il-
lumination, and the potential advantages promised by the NW ar-
chitecture, specifically: (i) the absence of threading dislocations
(TDs) in heteroepitaxial growth, (ii) the possibility of combining
lattice mismatched compounds without introducing extended de-
fects, (iii) the absence (reduction) of the quantum-confined Stark
effect in radial (axial) NW heterostructures, (iv) the improved
light extraction efficiency in comparison to as-grown planar de-
vices, and (v) the opportunity to increase the size of the active re-
gion per area unit of the substrate when fabricating devices with
a core-shell geometry. 3 -- 8 However, despite all these advantages
Paul-Drude-Institut für Festkörperelektronik, Leibniz-Institut im Forschungsver-
bund Berlin e.V., Hausvogteiplatz 5 -- 7, 10117 Berlin, Germany; E-mail:
ser-
[email protected]
Grupo de Electrónica y Semiconductores, Dpto. Física Aplicada, Universidad
Autónoma de Madrid, C/ Francisco Tomás y Valiente 7, 28049 Madrid, Spain
† Electronic Supplementary Information (ESI) available: Analysis of GaN facets
formed during thermal sublimation by reflection high-energy electron diffraction.
See DOI: 10.1039/C8NA00369F
and worldwide research activities, the performance of NW based
light-emitting diodes (LEDs) lags behind that of their state-of-the-
art two-dimensional counterparts fabricated on Al2O3 and Si sub-
strates, which nowadays exhibit wall-plug and external quantum
efficiencies above 80 % in the blue spectral range. 9,10 The inferior
performance of NW-based LEDs is mainly caused by the limita-
tions and complexities inherent to their formation using bottom-
up methods. Particularly, the nonideal growth conditions often
required to promote either uniaxial or radial growth may favor
the incorporation of higher concentrations of impurities and de-
fects. 3,11 Furthermore, the different chemical and physical prop-
erties of the precursors, which are simultaneously deposited on
different crystal facets, can result in the formation of detrimental
compositional and structural inhomogenities. 11 -- 13
Given that the epitaxial growth of group-III nitrides has
reached a high level of maturity, it is appealing to combine this
well-developed technology with some of the advantages offered
by the NW architecture using a top-down approach. 7,14 -- 17 It is
important to note that top-down methods are also suitable to ob-
tain arrays of largely dislocation free NWs despite the presence of
a high density of TDs in the layer used for the top-down process.
In fact, the average number of TDs per NW is determined by the
product of the TD density of the initial GaN layer and the cross-
sectional area of one NW. 15 For instance, for a substrate with a
TD density on the order of 108 cm−2, typical for GaN layers grown
on Al2O3, and a NW diameter of 100 nm, less than 1% of these
objects will contain TDs. 18
1 -- 9 1
Journal Name The top-down fabrication of NW arrays typically relies on dry
etching of a lithographically patterned two-dimensional layer.
This fabrication method has, however, not achieved much popu-
larity within the NW community because, in addition to the diffi-
culties in achieving vertical NW sidewalls, the dry etching process
inevitably creates point defects acting as nonradiative recombina-
tion centers at the NW sidewalls. Significant progress has been
made in recent years by introducing an additional anisotropic
wet chemical etching step after the creation of NWs by induc-
tively coupled plasma reactive ion etching (ICP-RIE). This ad-
ditional step removes the damaged material and facilitates the
formation of smooth and straight NW sidewalls. 15,16 Using this
method, the fabrication of NW LEDs 15 and optically pumped NW
lasers has been demonstrated. 16 As an alternative to this two-step
etching approach, a new top-down fabrication method coined as
selective area sublimation (SAS) was recently demonstrated by
Damilano et al. 17,19 This approach avoids any damage to the NW
sidewalls as it is based not on chemical etching but on material
sublimation in vacuum, a process used before by different groups
to decrease the diameter of as-grown NWs. 20 -- 22 Damilano et al
utilized this simple method to obtain random arrays of GaN NWs
and (In,Ga)N/GaN NW heterostructures. 17,19
For device applications, it is almost mandatory that the NWs
do not have a random arrangement, but form an ordered array
with precisely tunable spatial arrangements, NW diameters, and
NW-to-NW spacings.
In the present work, we explore the fab-
rication of ordered arrays of Ga-polar 23 GaN NWs by SAS. We
demonstrate that this method is suitable to simultaneously fabri-
cate ordered arrays of NWs with various diameters and spacings
on a single wafer. The sublimation process, carried out under ul-
tra high-vacuum conditions, is further analyzed in situ by reflec-
tion high-energy electron diffraction (RHEED) and line-of-sight
quadrupole mass spectrometry (QMS). A detailed analysis of the
decomposition process allows us to assess the temperature de-
pendence of the thermal etching rate, a crucial factor to control
the formation by SAS of NWs as well as other types of nanos-
tructures. Finally, we investigate the excitonic transitions from an
ordered array of GaN NWs by low-temperature cathodolumines-
cence spectroscopy (CL). The emission from the array is found to
be more intense than that from the original layer, confirming that
the sublimation process does not introduce any nonradiative de-
fects at the NW sidewalls, and revealing an enhanced extraction
efficiency of the NW array.
2 Experimental
As substrates for SAS, we use commercially available 2߰߰
GaN(0001) layers grown by hydride vapor phase epitaxy on
Al2O3 (purchased from Suzhou Nanowin Science and Technol-
ogy). The thickness of the GaN layer is (4.5 ± 0.5) µm and its TD
density 2×108 cm−2. The backside of the substrates is coated with
a 1.3 µm thick Ti layer for efficient thermal coupling during the
sublimation process. To produce the mask for SAS, a 20 nm thick
SiN layer is first sputtered on the substrate surface. Afterwards,
the SiN layer is spin-coated with a 100 nm thick positive resist.
Subsequently, the pattern is written by electron beam lithography
(EBL) in a RAITH system. The pattern comprises 100 × 100 µm2
2
1 -- 9
fields containing hexagonal arrays of circles. Upon writing the
pattern, the resist is developed before depositing 25 nm of Ni. Af-
ter a lift-off step, the substrates are diced in 10 × 10 mm2 pieces
and the SiN layer is etched at the regions not protected by Ni us-
ing RIE with a CF4/O2 gas mixture. Finally, the pieces are further
etched by ICP-RIE using a mixture of Ar, Cl2 and BCl3 gases. This
final etching step is used to eliminate residual surface contami-
nants created after etching SiN as well as upon the chemical
removal of the remaining Ni layer.
The sublimation of the GaN layer is carried out under ul-
tra high-vacuum conditions in a molecular beam epitaxy sys-
tem where the process can be monitored in situ by RHEED and
QMS. The substrate temperature is measured with an emissivity
corrected infrared optical pyrometer from Laytec. Customized
Python routines are used to extract the angular intensity pro-
files around selected GaN diffraction spots of RHEED patterns.
Because GaN decomposes congruently in vacuum 24, we employ
QMS to directly assess the decomposition rate by measuring the
desorbing Ga flux. The QMS system response to the Ga69 signal is
calibrated in units of atoms/cm2 s using the procedure described
in Ref. 25. Regardless of the surface morphology, we assume that
Ga atoms desorb isotropically.
After the sublimation process, the samples are investigated by
scanning electron microscopy (SEM) and low-temperature (14 K)
CL spectroscopy and imaging. Scanning electron micrographs are
acquired using either a Hitachi S-4800 or a Zeiss Ultra55 field-
emission microscope. CL measurements are performed at accel-
eration voltages of 5 kV using a Gatan MonoCL4 system mounted
to the Zeiss Ultra55 microscope. The system is equipped with a
parabolic mirror for light collection and with both a photomulti-
plier and a charge-coupled device for detection.
3 Results and discussion
3.1 Top-down fabrication of ordered arrays of GaN
nanowires
Ordered arrays of GaN NWs with different diameters and spac-
ings in between NWs are produced simultaneously in an indi-
vidual sample using a single sublimation step. Figure 1 shows
scanning electron micrographs of an exemplary NW array pro-
duced by annealing a patterned substrate at 825 C for 20 min.
The SiN patches used as a mask for the sublimation process
can be distinguished at the top of the GaN NWs by their darker
contrast. The nominal NW diameter and spacing (distance be-
tween adjacent patches) values for this particular array are 90
and 100 nm, respectively. The final average NW diameter, which
roughly corresponds to the actual size of the SiN patches, is
about 80 nm. As can be observed in Fig. 1(a), the NW array is
homogeneous on a large scale. Inside the patterned region, the
NWs have rather vertical sidewalls [see Fig. 1(b)]. The NW side
facets are, however, not as flat and well defined as in the case
of self-assembled GaN NWs produced by conventional bottom-up
growth approaches. 3,26 -- 31 Specifically, we do not observe pro-
nounced M-plane sidewall facets. The NW sidewalls are instead
rather roundish, most likely because the SiN patches do not
have a hexagonal but a circular shape (a representative scanning
Fig. 1 (a) and (b) Bird's eye-view scanning electron micrographs with different magnifications of a GaN NW array produced by selective area
sublimation. The nominal values of the NW diameter (ࣼ) and the spacing (L) are 90 nm and 0.1 µm, respectively. The micrograph shown in (b) is
acquired at one of the edges of the patterned region.
electron micrograph of the SiN patches before the thermal sub-
limation process is provided as supplemental material).
Figure 2 presents scanning electron micrographs of four addi-
tional NW arrays produced at the same time as the one shown in
Fig. 1. These arrays differ from the previous one in either the NW
diameter or the spacing. In all cases we observe the formation of
homogeneous NW arrays with a very high yield. The yield only
decreases due to the presence of TDs. The latter are easily rec-
ognized after the sublimation process because of the formation
of hexagonal pits which are randomly distributed with a density
of about 0.5 × 108 cm−2, lower than the nominal TD density of
the parent GaN layer (2 × 108 cm−2). This discrepancy originates
likely from the fact that the sublimation process, analogously to
chemical etching, 32 preferentially etches screw TDs. Note that ac-
cording to the nominal TD density of the GaN layer, the average
number of TDs per NW (estimated as the product of the TD den-
sity of the original GaN layer and the NW cross-sectional area 15)
amounts to 0.004 to 0.012 when the NW diameter is varied from
50 to 90 nm. Consequently, even for the array with the larger
NW diameter, approximately 99% of the NWs are expected to
be free of TDs. 18 The results shown in Figs. 1 and 2 demonstrate
that homogeneous arrays of GaN NWs with various diameters and
spacings can be obtained with identical sublimation parameters.
As can be seen in Figs. 2(c) and 2(d), the morphology of the
initially flat GaN layer in between the SiN patches is clearly
altered during the sublimation process. Figure 3(a) presents a
highly magnified scanning electron micrograph of an unpatterned
GaN(0001) layer decomposed for 20 min at 825 C . This micro-
graph reveals that the original (0001) surface gives way to the
formation of three-dimensional islands with a six-fold symme-
try and well-defined semi-polar facets. This result is in appar-
ent contrast to the step-edge and layer-by-layer decomposition
mechanisms reported in Refs. 33 and 34, respectively. However,
in these previous studies, where the decomposition process was
analyzed at different temperatures, the GaN layer was not contin-
uously decomposed (as in the present case), but a smooth (0001)
surface was recovered prior to each (comparatively brief) decom-
position step by depositing a thin GaN layer under conditions
favoring step-flow growth. As a matter of fact, in Ref. 33 it is
also noted that continuous decomposition can result in surface
faceting, a phenomenon that was tentatively ascribed to an en-
hanced sublimation rate near defects or/and grain boundaries. In
the present study, the faceting under continuous decomposition is
also detected in situ by RHEED. Figures 3(b) and 3(c) present the
RHEED patterns recorded along the [11İ20] and [1İ100] azimuths,
respectively. In both azimuths, a transmission pattern is observed,
accompanied by pronounced chevrons. These features, caused by
the refraction and transmission diffraction of electrons entering
and exiting crystal facets, can be used to derive the shape of the
three-dimensional objects from which they originate. 35 -- 40 As dis-
cussed in detail in the supplemental material, the analysis of the
vertex angles allows us to conclusively conclude that, in agree-
ment with the results reported by Damilano et al in Ref. 17, the
facets of the islands seen in Fig. 3 are formed by {1İ103} planes.
3.2 Thermal decomposition of GaN{1İ103} facets: tempera-
ture dependence
To properly control the fabrication of nanostructures by SAS, it is
essential to know the precise temperature dependence of the GaN
decomposition rate in vacuum. The decomposition of GaN(0001)
layers has been measured as a function of the temperature by dif-
1 -- 9 3
200 nm(a)1 μm (b)Ø= 90 nmL = 0.1μm Fig. 2 (a) -- (d) Bird's eye view scanning electron micrographs of NW arrays produced by selective area sublimation in the same run with varying
nominal NW diameter (ࣼ) and spacing (L) values, as indicated in their corresponding labels. The magnification is the same for all micrographs and the
scale bar is indicated in (d).
4
1 -- 9
1 μm (a)Ø= 50 nm L = 0.1 μm Ø= 70 nmL = 0.1μmØ= 90 nmL = 0.2μmØ= 90 nmL = 0.7μm(b)(c)(d) Fig. 3 (a) Bird's eye view scanning electron micrograph of a GaN layer after 20 min of annealing in vacuum at 825 C . RHEED patterns acquired
along the (b) [11İ20] and (c) [1İ100] azimuths after the thermal decomposition process.
ferent groups. 33,34,41 -- 44 However, as discussed in section 3.1, the
(0001) surface is unstable against the formation of {1İ103} facets,
which are expected to decompose with a different rate. Hence,
we next analyze the temperature dependence of the decomposi-
tion rate of {1İ103} facets by measuring in situ the desorbing Ga
flux at different substrate temperatures.
Figure 4(a) shows the desorbing Ga flux as measured by
QMS during the congruent thermal decomposition of an unpat-
terned 2߰߰ GaN wafer in vacuum. As shown in the graph, the
substrate temperature is first increased up to 870 C with a
rate of 20 K/min. Afterwards, the temperature is decreased in
steps down to 786 C . To assess the desorbing Ga flux under-
steady state conditions, we wait for the stabilization of the sub-
strate temperature after every temperature step. The measure-
ments evidence that the desorbing Ga flux steadily decreases
from 4.4 × 1015 cm−2 s−1 (ࣈ 60 nm/min) at 870 C down to 0.3 ×
1015 cm−2 s−1 (ࣈ 4 nm/min) at 786 C . The Arrhenius plot of the
steady-steady state values of the decomposition rate of {1İ103}
facets () is presented in Fig. 4(b). A fit of the data by
= exp(−ࢧ ),
(1)
with the Boltzmann constant yields a prefactor A of 1.8 ×
1031 atoms cm−2 s−1 and an activation energy E of (3.54 ±
0.07) eV. When taking into account the increased surface area
associated to the faceting, we find that the actual value of the
exponential prefactor describing the thermal decomposition rate
of {1İ103} facets is reduced by a factor (tan()2 + 1)1ࢧ2, where is
the angle between the normal vectors of the {1İ103} and (0001)
planes (32.0). The actual value of the exponential prefactor is
thus 1.58 × 1031 atoms cm−2 s−1. The corresponding values of E
and A measured by QMS in the case of a GaN(0001) plane are
(3.1 ± 0.1) eV and 1.17 × 1029 atoms cm−2 s−1. 34 Consequently,
even though the energy barrier for thermal decomposition is
slightly higher in the case of the {1İ103} facets, in the temperature
range of interest, these facets decompose faster than the (0001)
one due to the much higher exponential prefactor. These results
are consistent with the idea that, while during growth the mor-
phology of the crystal is governed by slow growing facets, dur-
ing dissolution/thermal decomposition the crystal shape is domi-
nated by fast desolving/decomposing crystal facets. 45,46
3.3 Luminescence from ordered arrays of GaN nanowires
produced by selective area sublimation
The luminescence from ordered arrays of GaN NWs is investi-
gated by CL spectroscopy at 14 K. Figure 5(a) presents the near
band-edge CL spectrum of the NW array shown in Fig. 1. For com-
parison, we have also included a second CL spectrum recorded on
an unpatterned region of the same sample. Both spectra are dom-
inated by a high-energy line originating from the radiative decay
of free excitons (FX), and its first- and second-order longitudinal-
optical phonon replica (LO) at lower energies. The dominance of
free over bound exciton recombination in this specific experiment
is due to both the comparatively high excitation density (on the
order of several 1017 cm−3) and the resulting high effective carrier
temperature [amounting to about 60 K for the spectra in Fig.5(a),
as determined by a fit to the high-energy slope of the free exciton
line]. Two observations are worth to be stressed. First, the lu-
minous intensity measured from the NW array is notably higher
1 -- 9 5
250 nm(a)(b)(c) Fig. 4 (Color online) (a) Change of substrate temperature and the resulting temporal evolution of the desorbing Ga flux (red and blue lines, respec-
tively) during the decomposition of a GaN layer.
(b) Arrhenius representation of the temperature dependence of the desorbing Ga flux during the
decomposition of a {1İ103}-faceted GaN layer. The line shows the best fit that yields an activation energy of = (3.54 ± 0.07) eV.
than that from the adjacent parent layer. This observation shows,
most importantly, that the sublimation process does not degrade
the internal quantum efficiency of the structure created, and fur-
thermore, that the extraction efficiency of light is enhanced signif-
icantly over that of the planar reference. 7,8 Second, the spectrum
of the NW array is rigidly red-shifted by 3 -- 4 meV in compari-
son to the planar reference, as shown in the inset of Fig. 5(a)
for the FX transition. This redshift results from the elastic relax-
ation of the residual compressive strain in the parent GaN(0001)
layer grown on Al2O3. The same phenomenon, benefiting from
the large surface-to-volume ratio of these nanostructures, was ob-
served by different groups for pure GaN NWs as well as in the case
of compressively strained (In,Ga)N quantum wells embedded into
GaN NWs. 14,17,19,47,48
The CL spectra shown in Fig. 5 were acquired immediately after
exposing the sample to the electron beam. After prolonged expo-
sure, the CL intensity in the patterned areas invariably quenches
strongly. As an example, Fig. 5(b) shows a panchromatic CL
map superimposed with its corresponding bird's eye view scan-
ning electron micrograph. The CL from the patterned region ap-
pears to be inhomogeneous with several NWs being apparently
not emitting at all. However, this quenching of the CL intensiy
upon irradation is solely the result of carbonaceous depositions
that introduce nonradiative recombination channels at the NW
sidewalls, as discussed in detail in Ref. 49.
6
1 -- 9
4 Summary and conclusions
We have demonstrated the fabrication of ordered arrays of GaN
NWs by SAS of pre-patterned GaN(0001) layers grown on Al2O3.
In a single sample, we simultaneously produced different arrays
with NW diameters and spacings ranging from 50 to 90 nm and
0.1 to 0.7 µm, respectively. The resulting NW sidewalls are verti-
cal, but do not exhibit well-defined M-plane facets. The roundish
shape of the NWs is attributed to the use of SiN patches with a
circular shape as a mask for SAS. According to Ref. 17, we expect
the formation of M-plane sidewall facets when employing patches
with a hexagonal shape properly oriented with respect to the GaN
template underneath. During the sublimation process, the (0001)
surface vanishes giving way to the formation of fast decompos-
ing semi-polar {1İ103} facets. The stability of these facets deter-
mines the thermal etching rate. We found that the {1İ103} facets
decompose following an Arrhenius-like temperature dependence
with an activation energy of (3.54 ± 0.07) eV and an exponential
prefactor of 9.46 × 1032 atoms cm−2 s−1. Low-temperature CL ex-
periments reveal a higher luminous intensity from the NW array
thanks to an improved light extraction efficiency. The emission is
red-shifted with respect to the one of the GaN layer because the
large NW aspect ratio facilitates the elastic relaxation of residual
strain. Although our CL experiments indicate that the sublimation
process does not generate nonradiative recombination centers at
the NW sidewalls, more conclusive results in this respect could be
obtained by time-resolved photoluminescence spectroscopy. Such
0204060801001200123450.880.900.920.9433343536Desorbing Ga flux (1015 at/cm2s)Time (min)400500600700800900Temperature (°C)(a)Ln[desorbing Ga flux (at/cm2s)]1/T (103/K)ED = (3.54±0.07) eV(b)880860840820800780Temperature (°C) Fig. 5 (Color online) (a) Near band-edge CL spectrum recorded at 14 K in bird's eye view geometry of the GaN NW array with a nominal NW diameter of
90 nm and a spacing of 0.1 µm. For comparison, the CL spectrum from the GaN layer surrounding the NW array (acquired under identical conditions)
is also included. The emission from free excitons and its corresponding first and second order phonon replicas are labeled as FX, LO1, and LO2,
respectively. The inset presents the normalized CL spectra of the NWs and the surrounding layer on a linear scale. (b) Superposition of a panchromatic
CL intensity image recorded at 14 K (yellow) with the corresponding bird's eye view scanning electron micrograph for the same NW array as in (a).
an analysis would be possible for GaN NWs containing (In,Ga)N
quantum wells, for which the luminescence signals from the NWs
and the GaN layer underneath would not spectrally overlap.
Selective area sublimation is, therefore, a suitable top-down
approach to produce ordered arrays of GaN NWs with high lumi-
nous efficiency without the need of any elaborate chemical treat-
ment. This fabrication method could be readily extended to other
types of micro- and nanostructures as well as to additional mate-
rial systems provided that they decompose congruently (to avoid
the accumulation of constituent elements on the sample surface)
and exhibit a marked anisotropy in the stability of its crystallo-
graphic planes. A prominent candidate for such experiments is
ZnO, another wide-bandgap semiconductor of interest for optical
applications that shares many properties with GaN.
Conflicts of interest
There are no conflicts to declare.
Acknowledgements
We thank Katrin Morgenroth for her support during the prepa-
ration and characterization of the samples as well as for her
dedicated maintenance of the molecular beam epitaxy system
together with Carsten Stemmler and Hans-Peter Schönherr, Se-
bastian Meister and Sander Rauwerdink for patterning the sub-
strates, and Anne-Kathrin Bluhm for her help during the acquisi-
tion of scanning electron micrographs. We are indebted to David
van Treeck for numerous discussions on SAS, and to Vladimir Ka-
ganer for discussions on the probability to find dislocations in
NWs. Special thanks are due to Lutz Geelhaar for his contin-
uous encouragement and support and a critical reading of the
manuscript. Funding from the Bundesministerium für Bildung
und Forschung through project FKZ:13N13662 is gratefully ac-
knowledged. Sergio Fernández-Garrido acknowledges the partial
financial support received through the Spanish program Ramón
y Cajal (co-financed by the European Social Fund) under grant
RYC-2016-19509 from Ministerio de Ciencia, Innovación y Uni-
versidades.
Notes and references
1 M. Yoshizawa, A. Kikuchi, M. Mori, N. Fujita and K. Kishino,
Jpn. J. Appl. Phys., 1997, 36, L459.
2 M. Sánchez-García, E. Calleja, E. Monroy, F. Sánchez, F. Calle,
E. Muñoz and R. Beresford, J. Cryst. Growth, 1998, 183, 23.
3 S. Li and A. Waag, J. Appl. Phys., 2012, 111, 071101.
4 D. Zubia and S. D. Hersee, J. Appl. Phys., 1999, 85, 6492.
5 S. D. Hersee, A. K. Rishinaramangalam, M. N. Fairchild,
L. Zhang and P. Varangis, J. Mater. Res., 2011, 26, 2293.
6 Q. Yue, K. Li, F. Kong, J. Zhao and W. Li, IEEE J. Quant. Elec-
tron., 2013, 49, 697.
7 N. P. Reddy, S. Naureen, S. Mokkapati, K. Vora, N. Shahid,
F. Karouta, H. H. Tan and J. Chennupati, Nanotechnology,
2016, 27, 065304.
8 C. Hauswald, I. Giuntoni, T. Flissikowski, T. Gotschke, H. T.
Grahn, L. Geelhaar and O. Brandt, ACS Photonics, 2017, 4,
52.
9 S. Nakagawa, T. Tabata, Y. Honda, M. Yamaguchi and
H. Amano, Jpn. J. Appl. Phys., 2013, 52, 08JE07.
1 -- 9 7
10 S. Kimura, H. Yoshida, K. Uesugi, T. Ito, A. Okada, S. Nunoue,
S. Kimura, H. Yoshida, K. Uesugi, T. Ito and A. Okada, J. Appl.
Phys., 2016, 120, 113104.
11 M. S. Mohajerani, M. Müller, J. Hartmann, H. Zhou, H.-H.
Wehmann, P. Veit, F. Bertram, J. Christen and A. Waag, Jpn.
J. Appl. Phys., 2016, 55, 05FJ09.
12 S. Albert, A. Bengoechea-Encabo, J. Ledig, T. Schimpke, M. A.
Sánchez-García, M. Strassburg, A. Waag and E. Calleja, Cryst.
Growth Des., 2015, 15, 3661.
13 M. Müller, P. Veit, F. F. Krause, T. Schimpke, S. Metzner,
F. Bertram, T. Mehrtens, K. Müller-Caspary, A. Avramescu,
M. Strassburg, A. Rosenauer and J. Christen, Nano Lett., 2016,
16, 5340.
14 C. H. Chiu, T. C. Lu, H. W. Huang, C. F. Lai, C. C. Kao, J. T.
Chu, C. C. Yu, H. C. Kuo, S. C. Wang, C. F. Lin and T. H.
Hsueh, Nanotechnology, 2007, 18, 445201.
15 Q. Li, K. R. Westlake, M. H. Crawford, S. R. Lee, D. D. Koleske,
J. J. Figiel, K. C. Cross, S. Fathololoumi, Z. Mi and G. T. Wang,
Opt. Commun., 2011, 19, 25528.
16 Q. Li, J. B. Wright, W. W. Chow, T. S. Luk, I. Brener, L. F.
Lester and G. T. Wang, Opt. Express, 2012, 20, 17873.
J. Lähnemann, A. Trampert, F. S, S. F. Li, H.-H. Wehmann and
A. Waag, Nanotechnology, 2010, 21, 305201.
29 K. A. Bertness, S. Member, N. A. Sanford and A. V. Davydov,
IEEE J. Sel. Topics in Quantum Electron., 2011, 17, 847.
30 L. Geelhaar, C. Chèze, B. Jenichen, O. Brandt, C. Pfüller,
S. Münch, R. Rothemund, S. Reitzenstein, A. Forchel, T. Ke-
hagias, P. Komninou, G. P. Dimitrakopulos, T. Karakostas,
L. Lari, P. R. Chalker, H. G. Mhairi and H. Riechert, IEEE J.
Sel. Topics in Quantum Electron., 2011, 17, 878.
31 O. Brandt, S. Fernández-Garrido, J. K. Zettler, E. Luna,
U. Jahn, C. Chèze and V. M. Kaganer, Cryst. Growth Des.,
2014, 14, 2246.
32 J. Weyher, S. Lazar, L. Macht, Z. Liliental-Weber, R. Molnar,
S. Müller, V. Sivel, G. Nowak and I. Grzegory, J. Cryst. Growth,
2007, 305, 384.
33 N. Grandjean, J. Massies, F. Semond, S. Y. Karpov and R. A.
Talalaev, Appl. Phys. Lett., 1999, 74, 1854.
34 S. Fernández-Garrido, G. Koblmüller, E. Calleja and J. S.
Speck, J. Appl. Phys., 2008, 104, 033541.
35 T. Hanada, B.-H. Koo, H. Totsuka and T. Yao, Phys. Rev. B,
2001, 64, 165307.
17 B. Damilano, S. Vézian, J. Brault, B. Alloing and J. Massies,
36 H. R. Gutiérrez, M. A. Cotta and M. M. G. de Carvalho, Appl.
Nano Lett., 2016, 16, 1863.
Phys. Lett., 2001, 79, 3854.
18 G. T. Wang, Q. Li, J. J. Wierer, D. D. Koleske and J. J. Figiel,
Phys. Status Solidi (a), 2014, 211, 748.
19 B. Damilano, V. S. Vézian, M. Portail, B. Alloing, J. Brault,
A. Courville, B. V. Brändli, M. Leroux and J. Massies, J. Cryst.
Growth, 2017, 477, 262.
37 D. Pashley, J. Neave and B. Joyce, Surf. Sci., 2001, 476, 35.
38 C. Gaire, F. Tang and G.-C. Wang, Thin Solid Films, 2009, 517,
4509.
39 J. Lee, D. Schuh, M. Bichler and G. Abstreiter, Appl. Surf. Sci.,
2004, 228, 306.
20 L. Brockway, C. Pendyala, J. Jasinski, M. K. Sunkara and
40 I. Ayahiko and P. I. Cohen, Reflection High-Energy Electron
S. Vaddiraju, Crys. Growth Des., 2011, 11, 4559.
Diffraction, Cambridge University Press, 2004.
21 B. Loitsch, D. Rudolph, S. Morkötter, M. Döblinger,
G. Grimaldi, L. Hanschke, S. Matich, E. Parzinger, U. Wurst-
bauer, G. Abstreiter, J. J. Finley and G. Koblmüller, Adv. Mater.,
2015, 27, 2195.
22 J. K. Zettler, P. Corfdir, C. Hauswald, E. Luna, U. Jahn, T. Flis-
sikowski, E. Schmidt, C. Ronning, A. Trampert, L. Geelhaar,
H. T. Grahn, O. Brandt and S. Fernández-Garrido, Nano Lett.,
2016, 16, 973.
23 J. Zúñiga-Pérez, V. Consonni, L. Lymperakis, X. Kong,
A. Trampert, S. Fernández-Garrido, O. Brandt, H. Renevier,
S. Keller, K. Hestroffer, M. R. Wagner, J. S. Reparaz, F. Akyol,
S. Rajan, S. Rennesson, T. Palacios and G. Feuillet, Appl. Phys.
Rev., 2016, 3, 041303.
24 N. Newman, J. Cryst. Growth, 1997, 178, 102 -- 112.
25 S. Fernández-Garrido, J. K. Zettler, L. Geelhaar and O. Brandt,
Nano Lett., 2015, 15, 1930.
26 A. Trampert, J. Risti´c, U. Jahn, E. Calleja and K. H. Ploog,
Proceedings of the 13th International Conference on Microscopy
of Semiconducting Materials, IOP Conf. Ser. No., 2003, 180,
167.
27 L. Largeau, D. L. Dheeraj, M. Tchernycheva, G. E. Cirlin and
J. C. Harmand, Nanotechnology, 2008, 19, 155704.
28 W. Bergbauer, M. Strassburg, C. H. Kölper, N. Linder, C. Roder,
41 R. Groh, L. Bartha and J. I. Yankove, Phys. Status Solidi A,
1974, 26, 353.
42 R. Held, D. E. Crawford, A. M. Johnston, A. M. Dabiran and
P. I. Cohen, Surf. Rev. Lett., 1998, 5, 913.
43 O. Ambacher, J. Vac. Sci. Technol., 1996, 14, 3532.
44 H. W. Choi, M. A. Rana, S. J. Chua, T. Osipowicz and J. S.
Pan, Semicond. Sci. Technol., 2002, 17, 1223.
45 Ryan C. Snyder and Michael F. Doherty, Materials, interfaces,
and electrochemical phenomena, 2007, 53, 1337 -- 1348.
46 M. R. Singh, N. Nere, H.-H. Tung, S. Mukherjee, S. Bor-
dawekar and D. Ramkrishna, Crys. Growth Des., 2014, 14,
5647.
47 M.-Y. Hsieh, C.-Y. Wang, L.-Y. Chen, T.-P. Lin, M.-Y. Ke, Y.-W.
Cheng, Y.-C. Yu, C. P. Chen, D.-M. Yeh, C.-F. Lu, C.-F. Huang,
C. C. Yang and J. J. Huang, IEEE Electron Device Lett., 2008,
29, 658.
48 V. Ramesh, A. Kikuchi, K. Kishino, M. Funato, Y. Kawakami,
V. Ramesh, A. Kikuchi, K. Kishino, M. Funato and
Y. Kawakami, J. Appl. Phys., 2010, 107, 114303.
49 J. Lähnemann, T. Flissikowski, M. Wölz, L. Geelhaar, H. T.
Grahn, O. Brandt and U. Jahn, Nanotechnology, 2016, 27,
455706.
8
1 -- 9
Table of contents entry
Table of contents entry We demonstrate the top-down fabrication of
ordered arrays of GaN nanowires by selective area sublimation of pre-
patterned GaN(0001) layers.
1 -- 9 9
200 nm1 m |
1811.00309 | 1 | 1811 | 2018-11-01T10:58:12 | Reservoir computing with the frequency, phase and amplitude of spin-torque nano-oscillators | [
"physics.app-ph",
"cond-mat.mes-hall",
"physics.comp-ph"
] | Spin-torque nano-oscillators can emulate neurons at the nanoscale. Recent works show that the non-linearity of their oscillation amplitude can be leveraged to achieve waveform classification for an input signal encoded in the amplitude of the input voltage. Here we show that the frequency and the phase of the oscillator can also be used to recognize waveforms. For this purpose, we phase-lock the oscillator to the input waveform, which carries information in its modulated frequency. In this way we considerably decrease amplitude, phase and frequency noise. We show that this method allows classifying sine and square waveforms with an accuracy above 99% when decoding the output from the oscillator amplitude, phase or frequency. We find that recognition rates are directly related to the noise and non-linearity of each variable. These results prove that spin-torque nano-oscillators offer an interesting platform to implement different computing schemes leveraging their rich dynamical features. | physics.app-ph | physics |
Reservoir computing with the frequency, phase and amplitude of spin-torque
nano-oscillators
D.Markovi´c,1, a) N.Leroux,1 M. Riou,1 F. Abreu Araujo,2 J. Torrejon,1 D. Querlioz,3 A. Fukushima,4 S. Yuasa,4 J.
Trastoy,1 P. Bortolotti,1 and J. Grollier1
1)Unit´e Mixte de Physique, CNRS, Thales, Univ. Paris-Sud, Universit´e Paris-Saclay, 91767 Palaiseau,
France
2)Institute of Condensed Matter and Nanosciences, UCLouvain, Place Croix du Sud 1, 1348 Louvain-la-Neuve,
Belgium
3)Centre de Nanosciences et de Nanotechnologies, CNRS, Univ. Paris-Sud, Universit´e Paris-Saclay,
91405 Orsay France
4)National Institute of Advanced Industrial Science and Technology (AIST), Spintronics Research Center, Tsukuba,
Ibaraki 305-8568, Japan
Spin-torque nano-oscillators can emulate neurons at the nanoscale. Recent works show that
the non-linearity of their oscillation amplitude can be leveraged to achieve waveform classifi-
cation for an input signal encoded in the amplitude of the input voltage. Here we show that
the frequency and the phase of the oscillator can also be used to recognize waveforms. For
this purpose, we phase-lock the oscillator to the input waveform, which carries information
in its modulated frequency. In this way we considerably decrease amplitude, phase and fre-
quency noise. We show that this method allows classifying sine and square waveforms with
an accuracy above 99% when decoding the output from the oscillator amplitude, phase or
frequency. We find that recognition rates are directly related to the noise and non-linearity
of each variable. These results prove that spin-torque nano-oscillators offer an interesting
platform to implement different computing schemes leveraging their rich dynamical features.
Spin-torque nano-oscillators are promising for neuro-
morphic computing1 -- 5. These magnetic tunnel junctions
can indeed emulate important properties of artificial neu-
rons through the non-linearity and relaxation properties
of current-induced magnetization dynamics. It has been
shown recently that a time-multiplexed, single oscillating
junction can enable or improve classification of different
waveforms, distinguishing sines from squares, and even
spoken digits6,7. In these experiments, the input wave-
form was encoded in the amplitude of the input voltage
and the quantity used for computing was the amplitude
of voltage oscillations across the junction. Other dynam-
ical variables can potentially be leveraged for computing,
such as the frequency or the phase of the oscillators, of-
fering a compelling platform to implement different neu-
romorphic computing approaches. However, all of these
variables tend to be highly noisy8 -- 10, which has been
shown to be detrimental to pattern classification Torre-
jon et al. 7. Magnetization dynamics indeed takes place
in nanoscale magnetic volumes, which makes them sen-
sitive to thermal fluctuations. In addition, phase noise
is enhanced by amplitude noise due to the inherent cou-
pling between the phase and amplitude of magnetization
oscillations11.
In this work, we show that these issues
can be circumvented by working in a regime where the
oscillator is synchronized to the input waveform that it
has to process which considerably reduces magnetization
fluctuations12. For this purpose, we use a sinusoidal in-
put waveform, that carries information encoded in its
modulated frequency, chosen close to the spin-torque os-
cillator frequency.
We first explain in detail our computing strategy and
a)Author to whom correspondance should be addressed. Electronic
mail : [email protected]
FIG. 1.
Schematic of the measurement set-up. The spin-
torque nano-oscillator is composed of two magnetic layers,
of fixed magnetization M (gray) and free magnetization m
(blue), separated by a thin insulating layer. At an external
magnetic field of H0 = 2000 Oe, a direct current Idc = 5
mA is injected in order to induce magnetization precessions.
The microwave signal encoding the input data in its frequency
(blue) is injected into a strip line above the oscillator, thus
generating a microwave magnetic field interacting with the
free layer. The microwave voltage V (t) emitted by the oscil-
lator is added to a microwave signal (subtraction waveform)
that compensates for the residual input signal and then is
measured with an oscilloscope.
describe the experimental set-up used to implement it.
We then show experimentally that sine and square wave-
forms can be classified by exploiting the frequency, phase
or amplitude of the oscillations. We highlight the corre-
lation between recognition rates and the non-linearity of
these dynamical variables as a function of the input sig-
nal to classify. Our work shows that it is possible to
take full advantage of magnetization dynamics by com-
puting through all the dynamical variables describing a
spin-torque nano-oscillator. In addition, since the input
ArbitraryWaveform GeneratorCH1CH2input waveformsubtraction waveformoscilloscope 2
(a) Frequency fosc, (b) phase ∆Φ and (c) Amplitude V of the oscillator as a function of the frequency fRF of the
FIG. 2.
injected microwave signal. The phase is determined with respect to that of the input waveform. Measurement uncertainties,
determined on 5 µs time intervals on which the mean is calculated, are shown in lighter color shaded area. Yellow and green
shaded areas designate respectively the locking range and the frequency pulling range.
waveform and the oscillator output are sinusoidal wave-
forms with close frequencies, our scheme should allow
chain-connecting the oscillators to build large neural net-
works.
Spin-torque nano-oscillators13,14 are composed of two
ferromagnets separated by a thin non-magnetic layer.
The magnetization of the bottom ferromagnet is pinned
whereas that of the top one is free. The spin-torque os-
cillator used in this experiment is a nano-pillar of 350 nm
diameter, composed of a 1.6 nm thick CoFeB layer with a
pinned magnetization, a 1 nm thick MgO insulating bar-
rier, and a 4 nm thick FeB layer whose ground state is
a magnetic vortex. Such nano-pillars can be fabricated
with diameters down to 10 nm15, which is adapted for
building large scale neural networks. When a direct cur-
rent is injected into this magnetic tunnel junction in the
presence of a magnetic field perpendicular to the layers
stack, it induces magnetization precessions in the free
layer through the effect of spin torque. Magnetoresis-
tance effects convert magnetic oscillations into resistance
oscillations, such that a microwave voltage is emitted by
the oscillator and can be detected using an oscilloscope.
The experimental set-up is shown in Figure 1.
Spin-torque nano-oscillators have the ability to syn-
chronize their voltage oscillations
to external mi-
crowave signals at frequencies close to their natural
frequency16 -- 19. The frequency and phase of the oscillator
lock to the external signal frequency, and its amplitude
is modified. Importantly, noise is reduced in frequency
and phase so that these variables are well defined in this
regime. We choose to work in this regime where the in-
put signal synchronizes the oscillator in order to reduce
the noise. The range of input microwave frequencies that
synchronize the oscillator is called the injection locking
range.
In the following we also take advantage of the
frequency pulling regime, by setting the frequency of the
input signal just outside of the locking range, such that
the oscillator does not get phase locked, but its frequency
gets pulled towards the frequency of the input signal.
We apply a perpendicular magnetic field H = 2000 Oe
to the oscillator and inject a direct current Idc = 5 mA,
which induces voltage oscillations of amplitude V = 13
mV at a frequency of 232.1 MHz. We choose these field
and current bias parameters in order to have a large lock-
ing range which is important for the frequency encoding
and to minimize the linewidth and maximize the out-
put signal. We use an Arbitrary Waveform Generator
(AWG) to generate microwave signals that we inject into
a strip line patterned 350 nm above the oscillator. This
signal induces a microwave magnetic field on the oscil-
lator, as well as a microwave current in the oscillator
due to capacitive coupling with the strip line. In order
to synchronize the oscillator, amplitudes of ≈ 350 mV
of the injected signal need to be applied, such that the
total voltage detected by the oscilloscope is dominated
by a residual capacitive microwave tone rather then the
oscillator voltage. We compensate for this residual tone
by adding the output voltage in a power combiner to an
exactly opposed microwave signal waveform (subtraction
signal in Figure 1) delayed by the time t0 that the in-
put signal takes to travel through the lines and that we
calibrate prior to the measurement.
In order to characterize the synchronization of the
spin-torque oscillator with an external source, we send
5 µs long waveforms modulated at different frequencies
in a 20 MHz range within the natural frequency of the
oscillator. We apply the Hilbert transform20,21 on the
detected voltage in order to extract frequency, amplitude
and phase relative to that of the injected microwave sig-
nal, that we average over the entire 5 µs waveform. The
oscillator frequency, phase and amplitude as a function of
the frequency of the injected microwave signal are shown
in Figure 2(a-c). As the injection signal frequency ap-
proaches the natural oscillator frequency, oscillator fre-
quency first gets pulled towards it and then becomes
identical to it in the locking range. Noise is reduced in all
three variables in the locking range. Due to the subtrac-
tion of the residual microwave signal performed using a
power combiner, the detected amplitude of the oscillator
voltage is divided by two. This results in low signal-to-
noise ratio even in the locking range [note large error bars
in Figure 2(c)]. The locking range, highlighted in yellow
in Figure 2, is experimentally determined from the stan-
dard deviation of the phase that strongly decreases in
this range and is found to be 7.2 MHz. As expected, the
measured frequency of the oscillator is equal to the in-
jected frequency in the locking range [Figure 2(a)]. The
phase difference between the oscillator and the input sig-
nal roughly follows the arcsin dependence on the input
frequency predicted by theory11 [Figure 2(b)].
An oscillator can only achieve good performance at
neuromorphic computing if it transforms the input sig-
22022523023524024523023223423623822022523023524024534567fRF (MHz)fRF (MHz)fRF (MHz)fosc (MHz)V(mV)220225230235240245-1-0.500.51/ (a)(b)(c) 3
Half of this data is used for training and the other half
for testing the performance. Each waveform is discretized
into 8 points [see Figure 3(a)] and the task consists
in determining which of the two waveform types each
point belongs to. This is a non-linearly separable task
and thus represents a good benchmark for neuromorphic
computing6,7,23.
Time multiplexing is achieved by preprocessing the in-
put data as illustrated in Figure 3(b). The detailed pro-
cedure can be found in previous work7. Each input point
is multiplied by the same binary valued sequence called
mask, whose length N determines the size of the emu-
lated reservoir. Figure 3(b) is an illustrative schematic
for a reservoir containing N = 6 neurons, whereas in our
experiment we have used N = 25 virtual neurons. The
output of the neural network for each input time step
is a weighted sum of the outputs of each virtual neuron
corresponding to this input,
N(cid:88)
i=1
y =
WifNL(xi),
(1)
FIG. 3. (a) The input data is a sequence of random sine and
square waves of equal periods and different amplitudes dis-
cretized in 8 points. (b) Pre-processed data corresponding to
half a sine wave followed by half a square one.
In this ex-
ample, the mask maps the problem to six virtual neurons. Y
axis corresponds to one example of encoding frequencies.(c)
Sketch of the input voltage corresponding to four neuron en-
tries for a sine wave. Different input values are represented
in different colors. The waveform amplitudes are encoded in
the frequency of the microwave voltage that is then injected
into the strip line for 150 ns for each data point.
nal in a non-linear manner7,22 -- 24. In the pulling regime
(green in Figure 2), the oscillator frequency, phase and
amplitude are all highly non-linear. The oscillator fre-
quency is linear over the entire locking range, whereas
the phase difference and the oscillator amplitude are non-
linear at the edges [Figure 2(b)-(c)].
The fact that frequency, amplitude and phase are all
non-linear functions of input frequency, enables us to use
them to compute as with an artificial neuron. We now
demonstrate this capability on a task that consists in
classification of sine and square waves of equal periods
but different amplitudes. For this we use a method called
single node reservoir computing6,7,22,23. This method
uses time multiplexing in order to emulate a reservoir
with a single nano-oscillator that plays a role of a differ-
ent effective virtual neuron at each time step.
The input data encoding procedure is shown in Fig-
ure 3. The input data is a sequence of 100 waveforms
randomly chosen between sines and squares of equal fre-
quency and different amplitudes : the amplitude of the
square wave is 50% larger than that of the sine wave.
where N = 25 is the number of neurons, Wi is the weight
matrix element corresponding to the ith neuron, fNL is
the non-linear function implemented by the nanodevice
and xi is the input of the ith neuron, that is the cor-
responding microwave frequency. The weight matrix is
calculated on a computer in order to match the target(cid:101)y
vector (cid:101)Y containing targets (cid:101)y for all the training exam-
ples, the weight matrix is calculated as W = (cid:101)Y F †, where
= 0 or 1 respectively for sines or squares. For a target
F † is the Moore-Penrose pseudo-inverse of the matrix
F containing outputs fNL(xi) of all neurons and for all
training examples22.
Classification performance is highly dependent on the
frequency window chosen for input data encoding. We
choose this window in a partial or full overlap with the
oscillator locking range. We fix the window width such
that sine and square waves always take values in a range
of 4 MHz and 6 MHz respectively. We repeat the encod-
ing and measurement procedure for center frequencies of
the encoding window varied between 225 MHz and 241
MHz and we calculate the success rate. Recognition rates
obtained when decoding neuron outputs from frequency,
phase and amplitude are shown in Figure 4((a)-(c)) as a
function of the center frequency of the sliding window.
Changing the center frequency has a double impact
on output variables, that is the presence of noise, and
the non-linear dependence on the input frequency. Noise
is minimized in the middle of the locking range but the
output in this regime is a linear function of the input
[see Figure 2], which results in a disability of the neural
network to learn the task.
Indeed, as can be seen in
Figure 4((a)-(c)), success rate for the frequency encoding
window centered in the middle of the locking range is
50% for all the three output variables, which for this
task corresponds to random choice. The linear regime is
larger for frequency than for amplitude and phase, which
is reflected in the bad performance for a larger number
of center frequencies in the middle of the locking range.
We find the best performance for the center frequency
on the edge of the locking range, with some of the fre-
(a)(b)(c)time stepInputPre-processed inputtime steptime Input voltage233231229Encoding frequency (MHz)227235233231229Encoding frequency (MHz)2272350.900.751.051.201.35 4
FIG. 4.
Success rates obtained when decoding from frequency, phase and amplitude of the oscillator, as a function of the
center of the frequency range chosen for encoding the input data. The frequency range used for encoding is indicated by a
blue double arrow for two measurement points. Yellow and green shaded areas designate respectively the locking range and
the frequency pulling range.
quencies used for encoding laying in the highly non-linear
frequency pulling regime. The best recognition rates
are obtained when neuron outputs are decoded from the
phase of the oscillations (99.75%, Figure 4(b)), as phase
is both more non-linear than frequency (best recognition
rate of 99.5%, Figure 4(a)) and less noisy then amplitude
(best recognition rate of 99%, Figure 4(c)).
These high classification rates have been obtained by
using relatively large input microwave amplitudes to
drive the oscillator and reduce its noise. In this regime,
the magnetization relaxation time, which decreases with
excitation amplitude in the locking range25, is very short,
smaller than 4 ns in our case. Therefore, the emulated
neural network performs best at tasks that do not re-
quire a memory, such as classification of different inputs.
When the waveforms to separate have identical input val-
ues that can only be recognized by keeping memory of
past inputs, as is the case for sine and square waves with
the same amplitude, the network performance is lower
(82% recognition rate at maximum). In the future, it will
be interesting to study the network intrinsic memory as a
function of drive amplitude and oscillator noise. In addi-
tion, an external memory can be added to the system by
using a time-delayed feedback loop and re-injecting the
signal emitted by the oscillator together with the input
data22 -- 24.
As a conclusion, we have shown that spin-torque nano-
oscillators synchronized to microwave signals can emulate
artificial neural networks. The frequency, phase and am-
plitude of the voltage emitted by the oscillator are all
non-linear functions of the frequency of the input mi-
crowave signal and can be used as outputs of the network.
Working with synchronized neurons has the advantage of
decreasing the frequency and phase noise, which will be
of particular importance when scaling down the size of
nano-pillars.
In addition, frequency encoding is a sim-
ple way to use the output of an oscillator to drive the
input of another, thus paving the path towards neural
networks composed of chain-connected spin-torque nano-
oscillators.
This work was supported by the European Research
Council ERC under Grant bioSPINspired 682955 and the
French ministry of defense (DGA).
and K. Roy, IEEE Trans. Magn. 51,
1F. Maci`a, A. D. Kent, and F. C. Hoppensteadt, Nanotechnology
22, 095301 (2011).
2M. R. Pufall, W. H. Rippard, G. Csaba, D. E. Nikonov, G. I.
Bourianoff, and W. Porod, IEEE J. Explor. Solid-State Comput.
Devices Circuits 1, 76 (2015).
3D. E. Nikonov, G. Csaba, W. Porod, T. Shibata, D. Voils,
D. Hammerstrom, I. A. Young, and G. I. Bourianoff, IEEE J.
Explor. Solid-State Comput. Devices Circuits 1, 85 (2015).
4K. Yogendra, D. Fan,
4003909 (2015).
5J. Grollier, D. Querlioz, and M. D. Stiles, Proc. IEEE 104, 2024
(2016).
6M. Riou, F. A. Araujo, J. Torrejon, S. Tsunegi, G. Khalsa,
D. Querlioz, P. Bortolotti, V. Cros, K. Yakushiji, A. Fukushima,
H. Kubota, S. Yuasa, M. D. Stiles,
and J. Grollier, in 2017
IEEE International Electron Devices Meeting (IEDM) (2017)
pp. 36.3.1 -- 36.3.4.
7J. Torrejon, M. Riou, F. A. Araujo, S. Tsunegi, G. Khalsa,
D. Querlioz, P. Bortolotti, V. Cros, K. Yakushiji, A. Fukushima,
H. Kubota, S. Yuasa, M. D. Stiles, and J. Grollier, Nature 547,
428 (2017).
8M. Quinsat, D. Gusakova, J. F. Sierra, J. P. Michel, D. Hous-
sameddine, B. Delaet, M. C. Cyrille, U. Ebels, B. Dieny, L. D.
Buda-Prejbeanu, J. A. Katine, D. Mauri, A. Zeltser, M. Prigent,
J. C. Nallatamby, and R. Sommet, Applied Physics Letters 97,
182507 (2010).
9M. W. Keller, M. R. Pufall, W. H. Rippard, and T. J. Silva,
Physical Review B - Condensed Matter and Materials Physics
82, 054416 (2010), 1005.5372.
10E. Grimaldi, A. Dussaux, P. Bortolotti, J. Grollier, G. Pillet,
A. Fukushima, H. Kubota, K. Yakushiji, S. Yuasa, and V. Cros,
Physical Review B - Condensed Matter and Materials Physics
89, 104404 (2014).
11A. Slavin and V. Tiberkevich, IEEE Trans. Magn. 45, 1875
(2009).
12R. Lebrun, A. Jenkins, A. Dussaux, N. Locatelli, S. Tsunegi,
E. Grimaldi, H. Kubota, P. Bortolotti, K. Yakushiji, J. Grollier,
A. Fukushima, S. Yuasa, and V. Cros, Physical Review Letters
115, 017201 (2015).
13S. I. Kiselev, J. C. Sankey, I. N. Krivorotov, N. C. Emley, R. J.
Schoelkopf, R. A. Buhrman, and D. C. Ralph, Nature 425, 380
(2003).
14W. H. Rippard, M. R. Pufall, S. Kaka, S. E. Russek, and T. J.
Silva, Physical Review Letters 92, 027201 (2004).
15H. Sato, E. C. Enobio, M. Yamanouchi, S. Ikeda, S. Fukami,
S. Kanai, F. Matsukura, and H. Ohno, Applied Physics Letters
105, 062403 (2014).
16W. H. Rippard, M. R. Pufall, S. Kaka, T. J. Silva, S. E. Russek,
and J. A. Katine, Physical Review Letters 95, 067203 (2005).
17M. Quinsat, J. F. Sierra, I. Firastrau, V. Tiberkevich, A. Slavin,
D. Gusakova, L. D. Buda-Prejbeanu, M. Zarudniev, J. P. Michel,
U. Ebels, B. Dieny, M. C. Cyrille, J. A. Katine, D. Mauri, and
A. Zeltser, Applied Physics Letters 98, 182503 (2011).
224228232236240center frequency (MHz)405060708090100Success Rate (%)output from phase 224228232236240center frequency (MHz)Success Rate (%)224228232236240center frequency (MHz)405060708090100Success Rate (%)output from frequencyencoding frequency range405060708090100output from amplitude (a)(b)(c) 18A. Houshang, E. Iacocca, P. Durrenfeld, S. Sani, J. Akerman,
and R. Dumas, Nat. Nanotechnol. 11, 280 (2016).
19M. Romera, P. Talatchian, S. Tsunegi, F. A. Araujo, V. Cros,
P. Bortolotti, K. Yakushiji, A. Fukushima, H. Kubota, S. Yuasa,
D. Vodenicarevic, N. Locatelli, D. Querlioz,
and J. Grollier,
ArXiv e-prints (2017), arXiv:1711.02704.
20B. Picinbono, IEEE Transactions on Signal Processing 45, 552
(1997).
21L. Bianchini, S. Cornelissen, J. V. Kim, T. Devolder, W. Van
Roy, L. Lagae, and C. Chappert, Applied Physics Letters 97,
032502 (2010).
22L. Appeltant, M. C. Soriano, G. Van Der Sande, J. Danckaert,
5
S. Massar, J. Dambre, B. Schrauwen, C. R. Mirasso, and I. Fis-
cher, Nature Communications 2, 468 (2011).
23Y. Paquot, F. Duport, A. Smerieri, J. Dambre, B. Schrauwen,
M. Haelterman, and S. Massar, Scientific Reports 2, 468 (2012),
1111.7219.
24L. Larger, A. Bayln-Fuentes, R. Martinenghi, V. S. Udaltsov,
Y. K. Chembo, and M. Jacquot, Physical Review X 7, 011015
(2017).
25W. Rippard, M. Pufall, and A. Kos, Applied Physics Letters
103, 182403 (2013).
|
1906.10653 | 1 | 1906 | 2019-03-24T08:08:06 | Temperature dependence of normalized sensitivity of Love wave sensor with unidirectional carbon fiber epoxy composite/Mn-doped 0.24PIN-0.46PMN-0.30PT ternary single crystal configuration | [
"physics.app-ph",
"physics.ins-det"
] | We have derived a general formula for sensitivity optimization of gravimetric sensors and use it to design a high precision and high sensitivity gravimetric sensor using unidirectional carbon fiber epoxy composite (CFEC) guiding layer on single crystal Mn-doped yPb(In1/2Nb1/2)O3-(1-x-y)Pb(Mg1/3Nb2/3)O3-xPbTiO3 (Mn: PIN-PMN-PT) piezoelectric substrate. The normalized maximum sensitivity exhibits a decreasing tendency with temperature up to 55 degrees Celsius. For the CFEC-on-Mn: PIN-PMN-PT sensor configuration with wavelength 24 {mu}m at 25 degrees Celsius, the maximum sensitivity can reach as high as 760.88 cm2/g, which is nearly twice that of traditional SiO2/ST quartz configuration gravimetric sensor. | physics.app-ph | physics | sensor with unidirectional carbon
Temperature dependence of normalized sensitivity of Love
fiber epoxy
wave
composite/Mn-doped 0.24PIN-0.46PMN-0.30PT ternary single
crystal configuration
Ziqing Luo,1 Yujiao Ma,1 Xiaopeng Wang,1 Naixing Huang,1,a) Xudong Qi,2 Enwei
Sun,3,b) Rui Zhang,3 Bin Yang,3 Tianquan Lü,3 Jian Liu,3 and Wenwu Cao3,4,c)
1Department of Physics, Northeast Petroleum University, Daqing, Heilongjiang, 163318,
China
2Condensed Matter Science and Technology Institute and Department of Physics, Harbin Institute
of Technology, Harbin 150080, China
3The School of Instrumentation Science and Engineering, Harbin Institute of Technology, Harbin
150080, China
4Department of Mathematics and Materials Research Institute, The Pennsylvania State University,
University Park, PA 16802, USA
We have derived a general formula for sensitivity optimization of gravimetric
sensors and use it to design a high precision and high sensitivity gravimetric sensor
using unidirectional carbon fiber epoxy composite (CFEC) guiding layer on single
crystal
Mn-doped
yPb(In1/2Nb1/2)O3-(1-x-y)Pb(Mg1/3Nb2/3)O3-xPbTiO3
(Mn:PIN-PMN-PT) piezoelectric substrate. The normalized maximum sensitivity
(
mS exhibits a decreasing tendency with temperature up to 55℃. For the
f
)max
CFEC-on-Mn:PIN-PMN-PT sensor configuration with = 24 m at 25℃, the
maximum sensitivity
f
mS
max
can reach as high as 760.88 cm2/g, which is nearly
twice that of traditional SiO2/ST quartz configuration gravimetric sensor.
Keywords: piezoelectric materials, surface acoustic wave sensors, acoustic properties
1
Corresponding authors. E-mail address: a)[email protected] (N. Huang), b)[email protected] (E. Sun),
c)[email protected] (W. Cao).
Surface acoustic wave sensors have been employed in liquids and gases for
immunoassay formats, gas detections, etc.1-3 It is known that Love waves exist only
when a thin layer is attached on a substrate, and the velocity of the shear wave in the
layer must be slower than that in the substrate.4 By analyzing the gravimetric sensors
made of fused quartz, polymethylmethacrylate (PMMA), ZnO, SiO2, etc., as the
waveguide layers, a general guideline for the improvement of the mass sensitivity has
been summarized: the waveguide layer materials should have low shear velocity, low
density, and low acoustic attenuation.5-6 The Love wave device with polymer guiding
layers provides not only excellent mass loading sensitivity, but also good temperature
stability.7 In addition, the elastic property of carbon fiber epoxy composites (CFEC) is
almost invariable from 25℃ to 55℃.8 And the CFEC provides acoustic impedance
comparable to that of aluminum, which means low acoustic loss.9 Hence, the
unidirectional carbon fiber epoxy composite (CFEC) is selected as the waveguide
material in this work. In a Love mode sensor, the insertion loss can be reduced by
selecting a substrate with a larger electromechanical coupling coefficient.10 In resent
years, Mn-doped
rhombohedral
phase
ternary
single
crystal
yPb(In1/2Nb1/2)O3-(1-x-y)Pb(Mg1/3Nb2/3)O3-xPbTiO3 (Mn:PIN-PMN-PT) has attracted
considerable attention for its outstanding piezoelectric properties and high mechanical
quality factor.11 It was report that the ternary PIN-PMN-PT single crystals have
higher phase transition temperatures (TRT and TC), which are advantageous for higher
temperature applications.12-15 The Mn:PIN-PMN-PT ternary single crystal has a lower
pyroelectric coefficient, lower relative permittivity, and smaller dielectric loss
compared to the binary ones.16 In addition, the Mn:PIN-PMN-PT single crystals
2
possess high piezoelectric constants and electromechanical coefficients,17 which help
enhance the efficiency of a transducer in converting electrical to mechanical energy
(and
vice
versa).
For
these
reasons,
this work
uses Mn-doped
0.24PIN-0.46PMN-0.30PT ternary single crystal as the substrate material of the
gravimetric sensor. In order to design high precision and high sensitivity gravimetric
sensors with temperature compensation, it is important to know the relationship
between the normalized sensitivity and environmental temperature.
Partial wave theory is the most commonly used method for analyzing acoustic
wave propagation in anisotropic media.18-21 In this work, we derived the dispersion
equation by employing partial wave theory for a specific sensor structure using
Mn-doped PIN-PMN-PT ternary single crystal poled along [001]c pseudo-cubic
direction as the substrate. Based on modal analysis, optimal parameters have been
derived for designing a gravimetric sensor with a CFEC-on-Mn:PIN-PMN-PT
configuration. This letter reports the characterization of the temperature dependence
for the normalized sensitivity from 25℃ to 55℃ and optimal design for the
normalized waveguide thickness of the sensor.
The wave guiding layer and piezoelectric substrate are bonded as shown in
Fig.1. It is a basic structure supporting Love waves without loading layer. The
guiding layer is unidirectional CFEC with the fibers are parallel to the x2-axis, which
is perpendicular to the wave propagation direction. The substrate is Mn-doped
0.24PIN-0.46PMN-0.30PT ternary single crystal poled along [001]c pseudo-cubic
direction. It is assumed that the [100]c pseudo-cubic direction of the single crystal is
along the x1 direction. The elastic wave equations in the media are given by
3
j
2
u
2
t
c
ijkl
2
u
k
x x
i
l
e
kij
2
x x
i
k
=
0
, (1a)
e
ikl
2
u
k
x x
i
l
ik
2
x x
i
k
=
0
i j k l
( , , ,
=
1,2,3)
, (1b)
where ρ is the density, cijkl is the elastic stiffness tensor component at constant electric
field, ekij is the piezoelectric coefficient tensor component at constant strain, and εik is
the dielectric permittivity tensor component at constant strain. The particle
displacements and the electric potential in the media can be written as
u
=j
j
exp(i
kbx
3
k x
)exp[i (
1
t
v
)]
, (2a)
4
=
exp(i
kbx
3
k x
)exp[i (
1
t
v
)]
, (2b)
where v is the phase velocity, k is the magnitude of wave vector k, b is the decay
coefficient along x3, α's give the relative amplitudes. Since the guiding layer thickness
h is finite and the substrate is considered infinite half-space (the thickness of substrate
is much greater than the wavelength), the general solutions of displacements and the
electric potential can be written as22
u
j
=
n
=
n
C
( )
n
n j
exp(i
kb
( )
n
x
3
) exp[i (
k x
1
vt
)]
(
, n =1,2,,8
), (3a)
C
( )
n
4 exp(i
n
( )
n
kb x
3
) exp[i (
k x
1
vt
)]
(
, n =1,2,,8
), (3b)
u
j
=
m
C
)
(
m
m j
exp(i
kb
(
m
)
x
3
) exp[i (
k x
1
vt
)]
, (
, , ,
m a b c d
=
)
, (3c)
=
m
C
)
(
m
4 exp(i
m
kb
(
m
)
x
3
) exp[i (
k x
1
vt
)]
, (
, , ,
m a b c d
=
)
, (3d)
where, Cn and Cm are weighting factors of these partial waves in the guiding layer
4
and substrate, respectively. The symbols marked with '^' are parameters of the
waveguide layer to distinguish them from that of piezoelectric substrate.
The particle displacements and the three components of stress must be
continuous across the interface (x3 = 0) assuming no slippery interface. The electrical
boundary conditions involve the continuity of the electric potential and the normal
component of the electric displacement across both the surface and the interface. At
the surface (x3 = h) of the basic structure, the traction must vanish. In the coordinate
system shown in Fig. 1, the boundary conditions of the problem are as follows:
u
1
u ,
1=
u
2
u , 3
2=
u
u , 31
T
3=
31=
T , 32
T
32=
T , 33
T
33=
T ,
= ,
3D D=
3
, ( 3
0=x
), (4a)
31 0=T
0=T
, 32
, 33 0=T
,
D k=
3
0
, ( 3 =x
h ). (4b)
The traction stresses are expressed by
T
3
j
=
c
3
jkl
(
u
k
/
x
l
)
e
k j
3
(
x
)
k
/
, (5)
and the normal component of the electric displacement is given by
D e
=
3
3
kl
(
u
k
/
x
l
)
/
3
(
k
x
k
)
. (6)
When being poled along [001]c, the Mn-doped 0.24PIN-0.46PMN-0.30PT
ternary
single
crystal
has
a
4mm macroscopic
symmetry.
The
CFEC-on-Mn:PIN-PMN-PT structure fits the NP53 symmetry.22 The [001]c poled
Mn-doped 0.24PIN-0.46PMN-0.30PT ternary single crystal is used as the substrate,
and the wave propagation is along the x1 direction in our design. It should be pointed
out that the unidirectional CFEC is not elastically isotropic. Substituting Eq. (2) and
material constants from Table I & Ref. [24] into Eq. (1), and using Eqs. (3) -- (6), we
can derive the dispersion relation of Love waves:
5
tan
kh
(
v v
s
2
)
1 2
1
=
1
(
(
v v
2
)
v v
s
)
s
2
1
1 2
, (7)
where
v
s
v v
s (for exact real solutions of k),
=
c
44
44
c
,
(
= c
66
c
44
)1 2
,
(
= c
55
c
44
)1 2
,
v
=s
(
c
44
)1 2
,
sv
(
=
c
44
1 2
)
,
k p =
2
. Using Eqs. (1) -- (7), the
distributions of Love mode displacements, which are normalized to the values at the
surface (x3 = h), can be expressed as:
(1)
xU
(
3
)
=
cos
k
(
v v
s
2
)
1 2
1
h
exp
k
1
(
v v
s
2
)
1 2
x
3
,
x
3(
0)
, (8)
(2)
xU
(
)
3
=
cos
k
(
v v
s
2
)
1 2
1
(
x
3
h
)
,
(0
x
3
h
)
. (9)
In practical device configurations, a thin loading layer (density 3, shear
modulus 3, shear wave velocity vS3 and thickness d) is generally loaded on the
surface of the guiding layer. Based on the perturbation theory,25 the relative frequency
sensitivity definition for a gravimetric sensor is given by4-5,9
S
f
m
=
1
f
0
lim
m
s
0
f
m
s
=
f
1
d f
3
0
=
(1
)
2
2
v
v
S
0
3
2
h
( )
, (10)
where
f
=
f
f
0
, f and f0 are the oscillation frequencies at perturbed and
unperturbed cases, respectively,
sm is the mass per unit area, and
( )h
is given by
h
( )
=
0
h
U
(2)
x
(
3
)
2
dx
3
0
U
(1)
x
(
3
)
2
dx
3
. (11)
When the Rayleigh hypothesis is applicable, i.e., when the elasticity of the
loading layer can be ignored
(
2
v
S =
3
3
3
=
0)
, the sensitivity equation becomes:
6
f
mS
=
1
2
h
( )
. (12)
Substituting Eqs. (8)-(9) into Eq. (11) then into Eq. (12) gives the final sensitivity
equation:
S
f
m
=
1 1
h
sin(
2
2
h
)cos(
2
h
1
h
)
h
)
2
cos (
2
h
1
, (13)
with
k
1
1
=
(
v v
s
2
)
1 2
,
k
2
=
(
v v
s
2
)
1 2
1
, and k = 2p= 2pf v .
The dispersion curves of Love waves for an unidirectional CFEC guiding layer
on Mn:PIN-PMN-PT ternary single crystal piezoelectric substrate basic structure at
25℃ are shown in Fig. 2. Each dispersion curve starts at the shear horizontal (SH)
wave velocity (2809.9 m/s) of the Mn:PIN-PMN-PT ternary single crystal. At this
speed, all higher order Love modes have low-frequency cutoffs. At high frequencies,
the phase velocities of all Love modes approach asymptotically to the shear wave
velocity 1916 m/s in CFEC. At other temperatures (T = 30, 35, 40, 45, 50, 55℃), we
can obtain similar dispersion curves. However, it must be pointed out that the SH
wave velocity (2761.1 m/s at 55℃) in the substrate decreases with temperature,
while the low-frequency cutoffs of higher order Love modes have the rising trend.
Fig. 3 shows the normalized sensitivity
f
mS for the first four modes of the
CFEC/Mn:PIN-PMN-PT configuration at 25℃. We find that the first-order mode
possesses the highest (
mS . For the first-order mode, the optimal ratio of the
f
)max
waveguide thickness to the wavelength is (h/)opt = 0.2691 the normalized
maximum sensitivity is (
mS = 1.8261 (10-3m3/kg) which is nearly twice that
f
)max
7
of traditional fused quartz/ST quartz structure sensor.4 Since the wavelength used in
Ref. [26] was 24 m, for the CFEC/Mn:PIN-PMN-PT configuration at 25℃, the
optimal design parameters lead to: h = 6.46 m and the maximum sensitivity of
f
mS
max
= 760.88 (cm2/g) which is a much higher sensitivity than 400 cm2/g for a
traditional structure.
Fig. 4 shows the normalized sensitivity
f
mS for the first-order mode at
different temperatures. The peak data from Fig. 4 are listed in Table Ⅱ, which reveal
that the normalized maximum sensitivity (
mS decreases with temperature,
f
)max
meanwhile, the optimal ratio (h/)opt exhibits a slow increasing tendency, while both
elastic constants
Ec ＆ 66
Ec
44
exhibit a decreasing tendency.
The data points shown in Fig. 4 are the simulation results of the normalized
maximum sensitivity and optimal ratio of normalized layer thickness at different
temperatures. The solid lines in Fig. 5 are fitted results, which can be described by
the following equations:
(
f
mS = 1.85722-7.11431×10-4T-1.90938×10-5T2 (10-3m3/kg) (14)
)max
(h/)opt = 0.25978+4.18431×10-4T-2.27476×10-6T2 (15)
From Figs. 4-5 & Table Ⅱ, one can find that the changing trend of normalized
maximum sensitivity (
mS is strongly dependent on the change of elastic
f
)max
constants
Ec ＆ 66
Ec
44
of the piezoelectric substrate with temperature. Though the
optimal design (h/)opt has a slight increasing trend, it is approximately invariant
from 25℃ to 55℃. At the optimal design points (h/)opt, the velocity values (vopt) of
8
the first-order mode Love waves have a decreasing tendency with increasing
temperature as shown in Table Ⅱ. And for a given thickness h, the sensitivity can
be always raised by increasing the operation frequency in a specified sensor
configuration,4 obviously for a fixed wavelength (e.g. 24 μm), the operation
frequency values at the optimal points have a decreasing trend with temperature.
Because the unidirectional CFEC can be fabricated by different carbon proportions,
some changes of the elastic constants, density, acoustic impedance and thermal
stability could be beneficial for improving the performance of the sensor. In addition,
the design changes of the fibers direction may bring some benefits. Therefore, the
CFEC is a very promising waveguide material and the high electromechanical
coefficient of Mn:PIN-PMN-PT ternary single crystals with good temperature
stability certainly make it an ideal substrate.
This research was supported in part by the NSFC under Grant No. 11304061 and
51572056.
9
References
1H. Wohltijon and R. Ressy, Anal. Chem. 51, 1458 (1979).
2M. Puiu, A. M. Gurban, L. Rotariu, S. Brajnicov, C. Viespe, and C. Bala, Sensors 15,
10511 (2015).
3W. Wang, S. Y. Fan, Y. Liang, S. T. He, Y. Pan, C. H. Zhang, and C. Dong, Sensors
18, 3247 (2018).
4B. Jakoby and M. J. Vellekoop, Smart Mater. Struct. 6, 668 (1997).
5Z. Wang, J. D. N. Cheeke, and C. K. Jen, Appl. Phys. Lett. 64, 2940 (1994).
6J. Du and G. L. Harding, Sens. Actuators A. 65, 152 (1998).
7W. Wang, X. Xie, J. L. Hou, and S. T. He, IEEE Sensors, Taipei, Taiwan, pp.1-4,
(2012).
8S. Sathish, J. Welter, R. Reibel, and C. Buynak, AIP Conf. Proc. 820, 1015 (2006).
9N. X. Huang, T. Q. Lü, R. Zhang, and W. W. Cao, Appl. Phys. Lett. 103, 053507
(2013).
10J. S. Liu and S. T. He, J. Appl. Phys. 107, 073511 (2010).
11W. C. Ou, S. Y. Li, W. W. Cao, and M. Yang, J. Electroceram. 37, 121 (2016).
12Y. M. Zhou, Q. Li, F. P. Zhuo, Q. F. Yan, Y. L. Zhang, and X. C. Chu, Ceram. Int. 44,
9045 (2018).
13Y. Chen, K. H. Lam, D. Zhou, Q. W. Yue, Y. X. Yu, J. C. Wu, W. B. Qiu, L. Sun, C.
Zhang, H. S. Luo, H. L. W. Chan, and J. Y. Dai, Sensors 14, 13730 (2014).
14S. J. Zhang and F. Li, J. Appl. Phys. 111, 031301 (2012).
10
15S. J. Zhang, F. Li, N. P. Sherlock, J. Luo, H. J. Lee, R. Xia, R. J. Meyer, Jr., W.
Hackenberger, and T. R. Shrout, J. Cryst. Growth 318, 846 (2011).
16Y. Li, Y. X. Tang, J. W. Chen, X. Y. Zhao, L. R. Yang, F. F. Wang, Z. Zeng, and H. S.
Luo, Appl. Phys. Lett. 112, 172901 (2018).
17X. Q. Huo, S. J. Zhang, G. Liu, R. Zhang, J. Luo, R. Sahul, W. W. Cao, and T. R.
Shrout, J. Appl. Phys. 113, 074106 (2013).
18C. W. Chen, R. Zhang, H. Chen, and W. W. Cao, Appl. Phys. Lett. 91, 02907
(2007).
19A. H. Nayfeh and H. T. Chien, J. Acoust. Soc. Am. 91, 1250 (1992).
20C. H. Yang and D. E. Chimenti, Appl. Phys. Lett. 63, 1328 (1993).
21C. H. Yang and D. E. Chimenti, J. Acoust. Soc. Am. 97, 2103 (1995).
22G. W. Farnell and E. L. Adler, in Physical Acoustics, edited by W. P. Mason and R.
N. Thurston (Academic, New York, 1972) Vol. IX, p. 35.
23N. X. Huang, T. Q. Lü, R. Zhang, and W. W. Cao, Chin. Phys. B., 23 (11), 117704
(2014).
24L. G. Tang, H. Tian, Y. Zhang, and W. W. Cao, Appl. Phys. Lett. 108, 082901
(2016).
25B. A. Auld, Acoustic Fields and Waves in Solids (Wiley, New York, 1976), Vol. 2,
Chap. 12, p. 277.
26F. Moreira, M. E. Hakiki, F. Sarry, L. L. Brizoual, O. Elmazria, and P. Alnot, IEEE
Sensors J. 7, 336 (2007).
11
TABLE I. Material constants of the unidirectional carbon fiber epoxy composite
(CFEC) for the fibers are parallel to the x2-axis.
Elastic constants (1010 N/m2)
12c
0.67
22c
0.71
23c
12.6
44c
0.58
55c
0.33
66c
0.58
Density (kg/m3)
1580
11c
1.37
aRef. [23]
12
TABLE Ⅱ. Normalized maximum sensitivity (
mS and optimal design of
f
)max
normalized layer thickness (h/) for the first-order mode device with unidirectional
CFEC/Mn:doped 0.24PIN-0.46PMN-0.30PT configuration at different temperatures.
Temperature (℃)
25
30
35
40
45
50
55
aRef. [24]
(
)max
mS
f
(10-3m3/kg)
1.8261
1.8204
1.8099
1.7981
1.7858
1.7722
1.7618
(
h
)opt
0.2691
0.2701
0.2710
0.2731
0.2744
0.2752
0.2757
vopt
(m/s)
2492.0
2488.2
2484.0
2476.5
2471.1
2467.1
2464.2
Ec
66
Ec
44
(1010 N/m2)a
(1010 N/m2)a
6.409
6.390
6.353
6.312
6.269
6.222
6.188
6.171
6.137
6.101
6.063
6.023
5.982
5.930
13
Figure captions
Fig. 1. Illustration of the sensor design and the coordinate system.
Fig. 2. Dispersion curves of Love waves at 25
.℃
Fig. 3. Normalized sensitivity
f
mS for the first four modes at 25
.℃
Fig. 4. Normalized sensitivity
f
mS for the first-order mode device with
unidirectional CFEC/Mn-doped 0.24PIN-0.46PMN-0.30PT configuration at different
temperatures.
Fig. 5. Fitted curves of normalized maximum sensitivity (
design of normalized layer thickness (h/) for the first-order mode device with
f
mS and optimal
)max
unidirectional CFEC/Mn-doped 0.24PIN-0.46PMN-0.30PT configuration from 25
℃
to 55
.℃
14
Fig. 1.
15
Fig. 2.
16
Fig. 3.
17
Fig. 4.
18
Fig. 5.
19
|
1802.01656 | 1 | 1802 | 2017-12-18T19:12:49 | Innovative thermo-solar air heater | [
"physics.app-ph"
] | In the present work we elaborate the innovative design of the solar air heater and justify it by a Computational Fluid Dynamics (CFD) simulation, implementing and experimentally testing a sample. We propose to use this device for maintenance of constant ambient conditions for thermal comfort and low energy consumption for indoor environments, inside greenhouses, passive houses, and to protect buildings against temperature fluctuations. We tested the functionality of our sample of the solar air heater for 50 weeks and obtained an agreement between the results of the numerical simulation, implemented using OpenFOAM (an open source numerical CFD software) and the experimental results. | physics.app-ph | physics |
Innovative Thermo-Solar Air Heater
M. Cuzminschia,d, R. Gherasima,b, V. Girleanua, A. Zubareva,c,∗, I. Stamatina
aUniversity of Bucharest, Faculty of Physics, Atomistilor 405, P.O 38, Bucharest-Magurele,
Romania, 077125
bCenter of Technology and Engineering for Nuclear Projects, 409, Atomistilor Street,
cNational Institute for Laser, Plasma and Radiation Physics, P.O. Box MG-36, Magurele,
Magurele, Judet Ilfov, Romania
dDepartment of Theoretical Physics, IFIN-HH, Magurele, Romania
Bucharest, Romania
Abstract
In the present work we elaborate the innovative design of the solar air heater
and justify it by a Computational Fluid Dynamics (CFD) simulation, imple-
menting and experimentally testing a sample. We propose to use this device for
maintenance of constant ambient conditions for thermal comfort and low en-
ergy consumption for indoor environments, inside greenhouses, passive houses,
and to protect buildings against temperature fluctuations. We tested the func-
tionality of our sample of the solar air heater for 50 weeks and obtained an
agreement between the results of the numerical simulation, implemented using
OpenFOAM (an open source numerical CFD software) and the experimental
results.
Keywords: air heating, solar energy, passive houses, thermal comfort
1. Introduction
Heating, ventilation and air-conditioning (HVAC) systems are mainly de-
signed for the building sector aiming to ensure the comfort living standards for
various climatic zones. The building sector accounts for more than 39% of the
primary energy requirements [1] and is a main contributor to carbon emission.
∗Corresponding author
Email address: [email protected] (A. Zubarev)
Preprint submitted to Journal of LATEX Templates
February 7, 2018
The Solar-HVAC has been considered as an alternative to reduce contribution
of the primary energy and in this respect, are developed solutions based on the
direct transformation of the solar energy in internal energy of the transport
medium [2, 3, 4, 5].
This basic principle is successfully applied to the solar air heaters in pas-
sive houses keeping a minimal comfortable temperature of 15oC [6] during cold
seasons.
One niche where the solar-air heaters can bring an input of additional heat
during cold season is the greenhouse growers. The cost of fuel is an increasingly
significant production expense for greenhouse growers in temperate climates.
High heating costs motivate growers to improve the efficiency of crop production
to minimize energy inputs. The two parameters influencing plant development
are: mean daily/night temperature (MDNT) and photosynthetic daily light
integral (DLI)[7].
MDNT usually must stay in a close range of 15−20oC for optimal conditions
in the greenhouse environment. During warm season the growers are faced with
high energy consumption for seeds and fruits drying, the intermediate stage
in preservation. Several experimental demonstrations and reviews show that
the electricity from primary energy sources can be replaced with solar energy
(thermal and photovoltaic [8, 9, 10] conversion followed by storage in heat and
electricity)[11, 12, 13].
To date, the key technologies applied to greenhouses are focused mainly
on transforming diurnal solar energy into heat storage complemented with a
smart insulation. The proposed solutions of solar energy utilization and the
reduction of heating power via various combined systems (external thermosolar
systems coupled with a storage tank, heating pumps and photovoltaics) [11]
show promising solutions but the initial investment with return of investment
are visible in the price of vegetables that can be higher than when using classical
electricity heating.
Another key issue until now not taken into account is the solar energy con-
version, during daylight, to internal energy of the air from greenhouse with free
2
convective circulation. Given how it is not appropriate to use forced convec-
tion in a greenhouse because the airstreams have a negative effect on the plant
growth [14], the only solution remains the free convection for thermal energy
transfer.
A second requirement is to maximize the heat transfer from a blackened
flat surface under sunlight irradiation to the backside cavity. In this respect,
the solution proposed with this device is to increase the backside surface of the
solar absorber [15] from A to nA by one set of decorative elements uniformly
distributed along the air stream direction. If in an air-heater with forced con-
vection such elements can induce a transition from laminar to turbulent flow
decreasing the efficiency of the thermal energy transfer [16, 17] in the air-heater
with free convection such elements have a minimum effect only in the hydrody-
namic resistance [18, 19]. The maximum heat transfer in the free convective air
heaters is dependent only on the residence time with the hot surface [20].
Taking into account the above considerations we propose one cost-effective
solution for one solar air-heater to supply thermal energy by free convection of
the air into a greenhouse as well as for other farming application [21], as for a
supplementary advantage of the air heaters which work upon the principle of
natural convection is their independence from external energy sources.
By comparison with other types of solar air systems glazed [22, 23], unglazed
[24, 25], and with double glazing and double pass air circulation [26] for a higher
yield and relative simple production was chosen the simple glazing design with
an effective back cavity. To ensure maximal conversion of solar radiation and
efficient heat transfer to air flow the solar absorber was made from blackened
aluminum. This decision was justified by the fact that, usually, most energy
losses occur through the front cover [26], other parts being insulated.
One of the advantages of portable air heaters is the possibility to flexibly vary
the power of the heating in respect to the consumer's preferences and weather.
In this paper, we propose a new model of a portable, light-weighted and modular
air heater based on solar air heating and natural convection inside of the device
shown in Figure 1, which is suitable to be integrated in greenhouses, passive
3
houses and office buildings, and to dry fruits, seeds, and nuts [27, 28] during
warmer seasons.
2. Device description. OpenFOAM simulation
Solar air heater design. The most effective and flexible air heaters are made
as self-contained devices, meaning that they can be attached to an exterior
wall of a building [29, 30]. A vertical installation is considered. The unit
consists of an insulated frame (implemented out of extruded polystyrene foam),
a solar absorbent (presented by a blackened aluminum metal board inside of
the thermo-insulated case), a front plexiglass glazing, an inlet for the incoming
air recirculation, and an outlet. At the outlet of the heater is placed a hood in
order to collect hot out-coming air (Figure 2 a).
We propose an innovative way of leaving a cavity between the back insulating
part and the solar absorbent in order to decrease dissipation of thermal energy.
This can lead to an efficiency increase by 30% in comparison to the glazed
single pass solar collector, especially in case of air recirculation. In other solar
air heaters a part of the thermal energy losses occur due to the contact between
hot air and cold front glazing, which we have successfully prevented in our
device. For better efficiency, we designed the solar absorbent part to have a flat
side that faces the glazing, and the radiator-like surface facing the insulating
back part. At the back part of the blackened aluminum board with a thickness
of 0.8 mm U-shape profiles are made with the dimensions of 7.5 mm to increase
the contact surface between heated air and the metal (Figure 2 b).
Due to the fact that the heat transfer is three times larger at the back side
of the metal board in comparison with the front side area [31, 15], we have:
Qback/Qf ront = Aback/Af ront ≈ 3.
Theoretical model. the functionality of the solar air heater is achieved by the
buoyancy phenomena [32, 33] and greenhouse effect [34]. Solar rays enter the
plexiglass glazing and are captured by the solar absorbent. Due to that, the
air inside of the installation is heated up. Because of the thermal transfer the
4
air inside of the back cavity is heated up. Temperature differences between the
exterior air and the air inside of the heater lead to the appearance of buoyancy
force, by which the incoming air is pushed up through the heater.
The decreasing air density because of the heating results in a positive buoy-
ancy force. The greater the thermal difference and the height of the structure,
the greater the buoyancy force. At the same time, the maximum value of the
velocity is limited by the air friction. The hot air exits the device at the outlet
and can be collected for future consumption by a hood.
The airflow used to calculate the efficiency of the air heater can be calculated
by the formula:
φ = ρoutuoutS
Where S is the surface area of the hood and uout and ρout = ρin
Tin
Tout
are
velocity and density at the output of the air heater, Tout and Tin are absolute
temperatures at the outlet and inlet of the device respectively.
The thermal performance of the air heater can be calculated by the formula:
η =
φcair∆t
Wsolar cos θ
where cair is specific heat of air, ∆t is the difference between inlet and outlet
temperatures, Wsolar is solar flux measured by pyranometer and θ is incidence
angle.
OpenFOAM numerical simulation. The processes inside the solar converter are
strongly nonlinear and cannot be evaluated analytically. For the estimation
of the main output parameters we implemented a numerical simulation us-
ing OpenFOAM (Open Field Operation and Manipulation). OpenFOAM is an
open-source application for computation fluid dynamics (CFD) problems [35, 36]
with a vast utilization in microclimate research, especially for heat transfer in
the air [37].
The solar air heater presented in Figure 2 a was approximated by a 2D
model, that is perfectly justified by the width of the air heater being much
greater than its thickness. The ratios between the dimensions of the simulated
5
model correspond to the existing assembled sample of the air heater. Due to the
U-shaped profiles at the back part of the solar absorbent, the heat flow is three
times higher in comparison with the front side. The external back wall was
treated as adiabatic due to the insulation of the air heater. For the transparent
window was used constant temperature condition. The radiator is heated up
by the sun and correspondingly heats up the air, and the device was studied in
the steady state case, therefore the heat flux transfered from the front and back
sides of the blackened aluminum board was taken as constant (fixedGradient
temperature condition).
The heat produced by the installation depends on the incidence angle (60o
on average) and the absorber efficiency of 90% (due to reflection). The effective
total heat flux was considered 600W/m2 and the inlet temperature of 27oC,
which corresponds to the fruit drying process during the warm season. Our
tests showed that the inlet temperature is not too relevant for the temperature
difference obtained. The discretization was designed by an orthogonal mesh
presented in Figure 3. The y+ value was calculated to be between 30 and
150 during the whole simulation period. As the initial values for the sample
simulations we used the averaged experimental data. For simulation of industrial
device, we gradually increased the height of the device and have taken into
account the results of previous simulation. Were used k, omega, epsilon and
alphat standard wall functions recommended for Buoyant Simple Foam solver.
For the simulation (k − ω) RANS (Reynolds Averaged NervierStokes) tur-
bulence model and Buoyant Simple Foam solver were used, which according to
[38] gives satisfactory results. It is a steady state solver, which uses ideal gas
approximation for air convection for buoyant and turbulent flow of compressible
fluids for ventilation and heat transfer.
The equations used by Buoyant Simple Foam solver are:
6
∂ρ
∂t + ∇ · (ρu) = 0
∂t (ρu) + ∇ · (ρu) − ∇[µef f (∇u + ∇uT ) − 2
∂t (ρh) + ∇ · (ρuh) − ∇(αef f∇h) = u · ∇p
ρ = p
RT
∂
∂
3 µef f∇ · (uT )] = −∇pd − (∇ρ)g
The h is the specific enthalpy; u is velocity, α is air thermal diffusivity, and
µ is dynamic viscosity.
The simulation was stopped when the residuals became constant with values
less than 10−6 for temperature and pressure and 10−5 for velocity.
The simulation was repeated for different heights of the heater within pa-
rameters from 1 to 20 meters, and for three different values of solar fluxes, that
correspond to a sunny day (900W/m2), an average day (600W/m2) and a cloudy
day (300W/m2).
3. Results and discussions
Results of the tested sample. Aiming to obtain the experimental proof regarding
the functionality of the device, a testing sample was built. The thickness of the
sample was the same as in the description; the width of 0.56 m and height of 1
meter.
During the test period (from 28th of September 2015 to 11th of September
2016) daily experimental data was collected for input and output temperatures,
and output air velocity. The solar air heater was protected from wind influence
for its optimal functionality. The OMEGA HHF141 rotating vane anemometer
was used to measure the output velocity value. The solar radiation intensity was
recorded by the Yanishevskii pyrometer. Two temperature sensors (AS6200)
were installed near the inlet and outlet of the device. All the data was collected
automatically every 7.5 minutes throughout the testing period. Temperatures
were also periodically measured at the bottom and top parts of hot the radia-
tor place. The radiator temperature was on average 12oC higher than the air
7
temperature in the nearest flow.
The average experimental data for each week is presented in Figure 4. It was
observed that the output temperature depends on the input air temperature
and, on average the solar air heater was heating up the air by 23.5oC. The
mean thermal performance of air heater for the entire testing period was of
η = 60.4%, which was calculated for the mean value of the velocity registered
by an anemometer and solar flux being measured by a pyranometer for the sun
peak hours.
The variation of the air heater's efficiency (b) and the intensity of the mea-
sured solar radiation (a) through the day for the randomly selected colder days
of the months is presented in Figure 5. While the 15th of November and the 20th
of December turned out to be some rather clear days, the 15th of January and
the 20th of March were partially cloudy days. Can be observed that the curves
for the thermal performance of the testing sample are smoother in comparison
to the radiation intensity curves, which is caused by the release of the stored
energy by the solar air heater 10 to 20 minutes after the receipt. Otherwise, the
efficiency of the heater is directly correlated with the received solar energy.
The implemented model of the device, as can be seen in Figure 5 is more
efficient during sunny days.
It can be a great supplementary energy source
during October, November, February, March, and April. However, it is not
efficient enough during the months of December and January. The better effect
of the solar air heater can be observed in the second part of the day, when the
optimal thermal conditions are the most needed in residence buildings.
Simulation study. CFD numerical simulations offer the detailed image on phys-
ical parameters inside of the air heater. This gives us a better understanding of
heat transfer and buoyancy that occur in the installation.
Initially, we built our version of the device in order to establish if there is
an agreement of theoretical results with the experimental ones.
In Figure 6
is presented the distribution of the temperature inside the air heater. We can
observe that the temperature grows almost linearly along the metal board. A
8
thin layer of air near the metal board appears with a higher temperature in
comparison with the average temperature at the same hight. This is due to
the fact that the aluminum solar absorbent reaches a temperature 12oC greater
than the average air temperature. According to the CFD numerical simulation,
the air is heated up by 22oC, which is in good agreement with the experimental
results, where we obtained heating up by 23.5oC.
In Figure 7 is represented the distribution of velocities inside the air heater.
During our work, we obtained good results, the velocity of air being important
because it increases the efficiency of the heater, less energy being dissipated in
contact with front glazing area. The velocity is almost zero near the metal board
and insulated walls, and reaches its maximum value due to the air viscosity
between the solar absorbent and the insulated back side. The velocity value
increases in the hood area and reaches its maximum at the outlet, due to reduced
outlet area in comparison with the inlet one.
Key values of experimental and theoretical results are carried out in the
tabular form: 1
After obtaining correspondence of the theoretical and experimental results
we decided to run our simulation for different heights of the solar air heater. We
propose that the industrial models of the device will have the same thickness,
but a larger height and width adopted to building dimensions. In Figure 8 is
presented the dependence of the output velocity and in Figure 9 of inlet-outlet
temperature difference depending on the height of the heater. We can observe
that the output velocity and temperature difference increase with the height
of the device. The temperature difference increases slower than the velocity,
because thermal dissipation becomes considerable.
The simulation results show that in case of an average day the temperature
difference increases from 22oC for 1 meter to 53.8oC for 20 meters height of
the device. Also we can observe that for heights larger than 12 meters the
temperature increases slower and strives to a limit, which it reaches at about
16 meters height. That being the optimal height of the device for tall buildings.
The velocity increases from 0.33 m/s for elaborated sample dimensions to 2.03
9
m/s for 20 meters height of the device. For large values of height the velocity
increases almost linearly.
4. Conclusions
The analysis of numerical simulations and experimental data lead us to con-
clude that this product can be used as an air heating system for residential
and office buildings. It is also very suitable for green-houses heating, due to it
working upon the principle of natural convection, to avoid airstreams and their
negative effect on plant growth. For thermal comfort of the indoor environment
the device can be used as a heating and ventilation system depending on the
needs of the user. When outdoor temperatures are high and there is no need
for the heating system, the solar air heater can be used to dry fruits, nuts and
seeds.
Our solar air heater can replace the heating system during the autumn-spring
period when the outdoor temperature is above −10oC. The average thermal
performance of the designed air heater is of η = 60.4% for the sun peak hours,
and the device can significantly reduce heating costs and protect the environ-
ment. The heater can produce a flow of around 59 l/s of air that will be 23oC
hotter then outside temperature. Also, the solar air heater protects walls against
temperature fluctuations, provides thermal insulation, and acoustic insulation,
which lead to comfort increase.
Acknowledgements
This work was partially supported by the Romanian Ministry of National
Education by the contract PN 16 47 0101 with UEFISCDI.
This work was partially supported by the Romanian Ministry of National
Education by the contract : PN-II-PT-PCCA-2013-4-1102 with FC-FARM 46-
2014
Raluca Gherasim was supported by the strategic grant POSDRU/159/1.5/S/133652,
Integrated system to improve the quality of doctoral and postdoctoral research
10
in Romania and promotion of the role of science in society co financed by the
European Social Found within the Sectorial Operational Program Human Re-
sources Development 2007 - 2013.
References
References
[1] K. K. Wan, D. H. Li, D. Liu, J. C. Lam, Future trends of building heating
and cooling loads and energy consumption in different climates, Building
and Environment 46 (1) (2011) 223–234.
[2] S. A. Kalogirou, Solar thermal collectors and applications, Progress in en-
ergy and combustion science 30 (3) (2004) 231–295.
[3] Q. Li, C. Zheng, A. Shirazi, O. B. Mousa, F. Moscia, J. A. Scott, R. A.
Taylor, Design and analysis of a medium-temperature, concentrated so-
lar thermal collector for air-conditioning applications, Applied Energy 190
(2017) 1159–1173.
[4] R. W. McCullough, T. A. Hewett, Solar air heater, uS Patent 4,262,657
(Apr. 21 1981).
[5] W. Fan, G. Kokogiannakis, Z. Ma, P. Cooper, Development of a dynamic
model for a hybrid photovoltaic thermal collector–solar air heater with fins,
Renewable Energy 101 (2017) 816–834.
[6] W. Fan, G. Kokogiannakis, Z. Ma, P. Cooper, Development of a dynamic
model for a hybrid photovoltaic thermal collector–solar air heater with fins,
Renewable Energy 101 (2017) 816–834.
[7] J. Garcıa, S. De la Plaza, L. Navas, R. Benavente, L. Luna, Evaluation of
the feasibility of alternative energy sources for greenhouse heating, Journal
of Agricultural Engineering Research 69 (2) (1998) 107–114.
11
[8] M. Debbarma, K. Sudhakar, P. Baredar, Thermal modeling, exergy analy-
sis, performance of bipv and bipvt: A review, Renewable and Sustainable
Energy Reviews 73 (2017) 1276–1288.
[9] M. Debbarma, K. Sudhakar, P. Baredar, Comparison of bipv and bipvt: A
review, Resource-Efficient Technologies.
[10] A. K. Shukla, K. Sudhakar, P. Baredar, A comprehensive review on de-
sign of building integrated photovoltaic system, Energy and Buildings 128
(2016) 99–110.
[11] M. Esen, T. Yuksel, Experimental evaluation of using various renewable
energy sources for heating a greenhouse, Energy and Buildings 65 (2013)
340–351.
[12] V. Sethi, S. Sharma, Survey and evaluation of heating technologies for
worldwide agricultural greenhouse applications, Solar energy 82 (9) (2008)
832–859.
[13] V. Sethi, S. Sharma, Survey of cooling technologies for worldwide agricul-
tural greenhouse applications, Solar Energy 81 (12) (2007) 1447–1459.
[14] M. G. Blanchard, E. S. Runkle, The influence of day and night temper-
ature fluctuations on growth and flowering of annual bedding plants and
greenhouse heating cost predictions, Hortscience 46 (4) (2011) 599–603.
[15] J. A. Myhren, S. Holmberg, Design considerations with ventilation-
radiators:
comparisons to traditional two-panel radiators, Energy and
buildings 41 (1) (2009) 92–100.
[16] M. Mohanraj, P. Chandrasekar, Performance of a forced convection solar
drier integrated with gravel as heat storage material for chili drying, Journal
of Engineering Science and Technology 4 (3) (2009) 305–314.
[17] M. Mohanraj, P. Chandrasekar, Drying of copra in a forced convection solar
drier, Biosystems Engineering 99 (4) (2008) 604–607.
12
[18] N. Hatami, M. Bahadorinejad, Experimental determination of natural con-
vection heat transfer coefficient in a vertical flat-plate solar air heater, Solar
Energy 82 (10) (2008) 903–910.
[19] W. Gao, W. Lin, E. Lu, Numerical study on natural convection inside the
channel between the flat-plate cover and sine-wave absorber of a cross-
corrugated solar air heater, Energy Conversion and Management 41 (2)
(2000) 145–151.
[20] B. Prasad, J. Saini, Effect of artificial roughness on heat transfer and fric-
tion factor in a solar air heater, Solar energy 41 (6) (1988) 555–560.
[21] D. R. Pangavhane, R. Sawhney, P. Sarsavadia, Design, development and
performance testing of a new natural convection solar dryer, Energy 27 (6)
(2002) 579–590.
[22] R. Yang, P.-L. Wang, A simulation study of performance evaluation of
single-glazed and double-glazed collectors/regenerators for an open-cycle
absorption solar cooling system, Solar energy 71 (4) (2001) 263–268.
[23] S. A. Kalogirou, Solar thermal collectors and applications, Progress in en-
ergy and combustion science 30 (3) (2004) 231–295.
[24] S. Medved, C. Arkar, B. Cerne, A large-panel unglazed roof-integrated
liquid solar collector--energy and economic evaluation, Solar Energy 75 (6)
(2003) 455–467.
[25] J. C. Hollick, Unglazed solar wall air heaters, Renewable Energy 5 (1)
(1994) 415–421.
[26] A. Omojaro, L. Aldabbagh, Experimental performance of single and double
pass solar air heater with fins and steel wire mesh as absorber, Applied
energy 87 (12) (2010) 3759–3765.
[27] S. Dhanushkodi, V. H. Wilson, K. Sudhakar, Thermal performance evalu-
ation of indirect forced cabinet solar dryer for cashew drying, American-
13
Eurasian Journal of Agricultural and Environmental Science 14 (11) (2014)
1248–1254.
[28] S. Dhanuskodi, V. Wilson, S. Kumarasamy, Design and thermal perfor-
mance of the solar biomass hybrid dryer for cashew drying, Facta Univer-
sitatis, Series: Mechanical Engineering 12 (3) (2014) 277–288.
[29] J. Tonui, Y. Tripanagnostopoulos, Improved pv/t solar collectors with heat
extraction by forced or natural air circulation, Renewable energy 32 (4)
(2007) 623–637.
[30] T. Yang, A. K. Athienitis, A study of design options for a building inte-
grated photovoltaic/thermal (bipv/t) system with glazed air collector and
multiple inlets, Solar Energy 104 (2014) 82–92.
[31] F. Kreith, R. M. Manglik, M. S. Bohn, Principles of heat transfer, Cengage
learning, 2012.
[32] J. Tanny, V. Haslavsky, M. Teitel, Airflow and heat flux through the vertical
opening of buoyancy-induced naturally ventilated enclosures, Energy and
Buildings 40 (4) (2008) 637–646.
[33] S. Sinha, R. Arora, S. Roy, Numerical simulation of two-dimensional room
air flow with and without buoyancy, Energy and Buildings 32 (1) (2000)
121–129.
[34] R. L. Bradley, R. W. Fulmer, Energy: The master resource, Kendall/Hunt,
Dubuque.
[35] H. Weller, Openfoam, URL http://www. openfoam. org.
[36] H. G. Weller, G. Tabor, H. Jasak, C. Fureby, A tensorial approach to com-
putational continuum mechanics using object-oriented techniques, Com-
puters in physics 12 (6) (1998) 620–631.
14
[37] A. A. Aliabadi, E. S. Krayenhoff, N. Nazarian, L. W. Chew, P. R. Arm-
strong, A. Afshari, L. K. Norford, Effects of roof-edge roughness on air tem-
perature and pollutant concentration in urban canyons, Boundary-Layer
Meteorology (2017) 1–31.
[38] A. Limane, H. Fellouah, N. Galanis, Thermo-ventilation study by openfoam
of the airflow in a cavity with heated floor, in: Building Simulation, Vol. 8,
Springer, 2015, pp. 271–283.
15
Figure 1: Experimental sample of the device.
Data Experimental
Simulation
∆T
η
23.5oC
60.4%
22oC
63.9%
Table 1: Key values for air heater
16
Figure 2: Design of the sample of the device. (a) General preview. 1 - insulated case of the
device, 2 - plexiglass front glazing, 3 - radiator-like absorbent. (b) Inset shows the detailed
U-shape form of profiles of the absorbent part.
17
123(a)(b) Figure 3: Discretization scheme of the air-heater model.
18
Figure 4: Experimental values of input (tin) and output (tout) temperatures and output
velocity (uout) .
19
-10 0 10 20 30 40 50 6001/10/1501/12/1501/02/1601/04/1601/06/1601/08/16 0.15 0.2 0.25 0.3 0.35 0.4t(oC)u(m/s)datetintoutuout Figure 5: (a) The intensity of direct radiation in kW/m2 throughout the day for the dates
of the 15th of November, the 20th of December, the 15th of February and the 20th of march
recorded by the pyranometer. (b) The efficiency of our model of solar air heater throught the
corresponding days
20
0 0.4 0.8 6 8 10 12 14 16 18 20(a)P (kW/m2)time (h) 0.3 0.5 0.7 6 8 10 12 14 16 18 20sunset(b)η15 Nov20 Dec15 Feb20 Mar Figure 6: Temperature distribution inside of the sample.
21
Figure 7: Air velocity distribution inside of the sample. The inset with the plot shows the
velocity distribution at the outlet along x and y directions and its average value of 0.33 m/s.
22
Figure 8: Dependence of the difference in temperatures at the inlet and outlet areas on the
height of the building for various solar fluxes.
23
15 20 25 30 35 40 45 50 55 60 65 0 2 4 6 8 10 12 14 16 18 20∆t(oC)Z(m)Sunny DayAverage DayCloudy Day Figure 9: Dependence of the output velocity on the height of the building for various solar
fluxes.
24
0.15 0.55 0.95 1.35 1.75 2.15 0 2 4 6 8 10 12 14 16 18 20uout(m/s)Z(m)Sunny DayAverage DayCloudy Day |
1808.09844 | 1 | 1808 | 2018-08-29T14:15:30 | An electrically pumped phonon-polariton laser | [
"physics.app-ph"
] | We report a device that provides coherent emission of phonon polaritons, a mixed state between photons and optical phonons in an ionic crystal. An electrically pumped GaInAs/AlInAs quantum cascade structure provides intersubband gain into the polariton mode at = 26.3 \mu m, allowing self-oscillations close to the longitudinal optical phonon energy of AlAs. Because of the large computed phonon fraction of the polariton of 65%, the emission appears directly on a Raman spectrum measurement exhibiting a Stokes and anti-Stokes component with the expected shift of 48 meV. | physics.app-ph | physics | An electrically pumped phonon-polariton laser
Keita Ohtani, Bo Meng, Martin Franckié*, Lorenzo Bosco, Camille Ndebeka-Bandou, Mattias
Beck, Jérôme Faist
Affiliations:
Institute for Quantum Electronics, ETH Zurich, August-Piccard-Hof 1, 8093 Zurich, Switzerland
*Correspondence to: [email protected]
Abstract: We report a device that provides coherent emission of phonon polaritons, a mixed state
between photons and optical phonons in an ionic crystal. An electrically pumped GaInAs/AlInAs
quantum cascade structure provides intersubband gain into the polariton mode at (cid:79) = 26.3 (cid:80)m,
allowing self-oscillations close to the longitudinal optical phonon energy of AlAs. Because of the
large computed phonon fraction of the polariton of 65%, the emission appears directly on a Raman
spectrum measurement exhibiting a Stokes and anti-Stokes component with the expected shift of
48 meV.
One Sentence Summary: We report the direct observations of coherently emitted phonon
polaritons via their photon, phonon, and polariton signatures.
Main Text: The polariton, a mixed state between a photon and an electronic excitation in solid-
state matter, has recently attracted much attention because of its rich physics that includes
superfluidity, quantized vortices and Bose-Einstein condensation (1, 2). In addition, the possibility
of engineering both the optical and matter part of this quasi-particle can be exploited for the
creation of polaritonic devices with enhanced non-linear properties (3-5). An exciton-polariton
laser created using a semiconductor microcavity is an elegant example since it reaches threshold
by stimulated scattering of the polaritons (6, 7) instead of population inversion between conduction
and valence band states. Other functional polaritonic devices are also investigated using other
material elementary excitations (8-11).
The phonon is the elementary excitation of lattice vibrations in a solid; in an ionic crystal
its optical mode is strongly coupled with light, leading to the formation of a phonon polariton (12,
13). As vibration frequencies of many polar materials lay in the terahertz (THz) spectral region,
the excitation and emission of such polaritons from a near-infrared source via a non-linear
interaction has attracted much interest (14, 15), including its use in THz spectroscopy applications
(15). In addition, the phonon polariton displays very peculiar features of coherent emission and
propagation (16, 17), in its wave-guiding (18), coupling to metamaterial resonators (19) and
tunable resonant energy scaled down to a few atomic layers in two-dimensional materials (20).
Control and manipulation of such properties open up the possibilities to build a platform based on
solid state materials to explore THz nano-photonics and -phononics.
In contrast to optical pumping of phonon polaritons, electrical pumping, desirable for
numerous applications, remains a challenging task. Here we report the demonstration of an
electrically pumped phonon-polariton semiconductor laser. A suitably engineered semiconductor
1
quantum well heterostructure is electrically excited and provides gain mainly into the photon
fraction of the phonon polariton. As a result, phonon polariton lasing action at an energy close to
the longitudinal optical (LO) phonon energy (ħωLO) with a phonon fraction of 65% is achieved.
As shown in Fig. 1A, phonon polaritons are generated in our structure in the quantum
cascade active region based on an InGaAs/AlInAs heterostructure (21) (the computed electron
band structure is shown in Fig. 1S in the supplementary material). In this work we targeted the
AlAs transverse optical phonon (TO) in the AlInAs barrier layers at ħωTO ≈ 42 meV because it is
energetically well separated from the frequency of the other TO phonons (InAs: ħωTO ≈ 29 meV
and GaAs: ħωTO ≈ 31 meV) allowing it to be selectively addressed optically. The active region,
based on a bound-to-continuum transition scheme (22), is designed to provide gain mainly into the
photon fraction of the polariton at ħωTO of AlAs by our density matrix simulator (23). Additionally,
recent theoretical results show that, depending on frequency, the phonon fraction will also
experience a gain of approximatively 1-10% of the one experienced by the photon fraction (24).
This has been confirmed for the present design by computing the phonon polariton emission rate,
shown in Fig 1C, using a non-equilibrium Green's function model (25). Although the inter-
subband laser transition energy is 42 meV, due to the polaritonic nature of the emitted radiation
the laser is expected to operate at 48.2 meV. This is lower than the peak of the emission rate since
the photon loss rate is greater than the phonon one. The cavity thus favors a greater phonon
component of the polariton, resulting in a non-photonic fraction of the emission rate of 7% (at the
experimental lasing energy of 47.2 meV it is 10%). The simulations also show that the polariton
laser threshold is lower than that expected from the purely optical emission (see the supplementary
section S5).
A 4.1 μm thick structure comprised of 60 repetitions of the active region was grown by a
molecular beam epitaxy on InP substrates. For comparison, we also grew six additional structures
in which the gain is designed to peak at different energies higher than ħωTO of AlAs. Waveguiding
of the phonon polaritons is done in a hybrid way, as the phonon excitation is naturally confined
within the AlInAs layers, while the photonic part is guided along the plane of the layers by two
metallic contact layers (26). As shown by the computation of the mode displayed in Fig. 1A, the
squared electric field is concentrated in the barrier material in a deep subwavelength dimension.
The contribution of the mechanical energy Umechanical to the total energy Utotal as a function of
photon energy and position in the active region is shown in Fig. 1B.
Results
To provide an unambiguous probe of the phonon-polaritonic nature of the emission, we
investigated the latter using both a direct measurement of the emission as well as using Raman
scattering with a green light excitation. First, the subthreshold emission spectra (T = 10 K) of a
30(cid:80)m wide and 250(cid:80)m long device were recorded using a home-build vacuum Fourier transform
infrared spectrometer equipped with a He-cooled Si bolometer. As seen in Fig. 2A, the emission
spectra display an asymmetric shape, with two sharp maxima, separated by the energy gap formed
between the two AlAs LO and TO optical phonon energies, which were determined by Raman
scattering (Fig. 2B) and far-infrared transmission measurements (Fig. 2C), respectively. Lasing
occurs at a photon energy of 47.2 meV, which is slightly higher than ħωLO of AlAs. The light
intensity versus injected current curve (shown in Fig. 2D) exhibits a single threshold behavior. We
attribute the absence of a double threshold, seen usually in exciton polariton lasers, to the different
nature of the material excitations: In the case of exciton polaritons, the second threshold occurs
2
when excitons are ionized to free electrons/holes for a large density whereas phonons are not
ionized (6, 7, 27, 28).
To probe the polariton, we measured the low temperature (T = 5 K), micro Raman
scattering spectra of a 300 (cid:80)m long and 30 (cid:80)m wide quantum cascade structure under operation.
To limit the dissipation, the device was driven at a 10% duty cycle. A polarized continuous-wave
green laser light with a wavelength of (cid:79) = 532 nm was focused on the cleaved facet of the laser
and the back scattered light was measured though a polarizer. As depicted in Fig. 3A, for Raman
shifts below 40 meV, the spectrum was consistent with the phonon peaks associated with the
InGaAs/AlInAs layers. In particular the peaks at 28 meV and 32 meV were assigned to the InAs
and GaAs TO phonons, respectively. In this geometry, an AlAs TO phonon peak was not visible
on the Raman spectra due to the smaller volume concentration ratio of AlAs. The peak at 35 meV
denoted by "x" is an artefact attributed to the Raman scattering inside the optics used in the setup.
In this spectrum, for a driving current of I = 1050 mA a peak appears both in the Stokes and anti-
Stokes sides at an energy corresponding exactly to the energy of the phonon-polariton. As
expected, this peak disappeared when the device was driven at a current below the threshold (I =
670 mA). While for the TO phonon peaks, the ratio of the Anti-Stokes to Stokes sideband intensity
is 0.42 and corresponds to a thermal phonon population at a temperature of ~150 K mainly caused
by localized heating from the P = 20 mW green laser spot (assuming a constant Raman cross-
section and thaking the InAs TO phonon). The same ratio is close to unity for the polaritonic peak.
This is expected as the occupancy of the phonon polariton is well above unity above threshold. In
fact, in the driving conditions considered in this experiment, we estimated a population of the order
of approximately 105 polaritons. As shown in Fig. 3B, the peak exhibits the selection rules
expected for nonlinear optical interaction with an electric field along the growth direction. The
Raman signal originating from the phonon polaritons will only be visible for the horizontal (along
the plane of the layers)-horizontal polarization direction combination, while the other phonon
peaks are visible for the cross-polarization combinations (vertical (perpendicular to the plane of
the layers)-horizontal and horizontal-vertical).
Shown in Fig. 3C is the Raman backscattering spectra obtained for a laser where the active
region consisted of a reference laser heterostructure where the AlInAs barriers were substituted by
GaAsSb ones. These devices do not contain any AlInAs material, but exhibit a similar
intersubband gain coefficient (29). Although the laser emitted at roughly the same frequency, in
the Raman shift region where the phonon-polariton signal is expected, only a faint peak is visible
just above the noise. This behavior is expected because the coupling of the photons to the phonon
has been reduced by the much larger detuning of the photon emission to the closest TO phonon
line of GaAs at 32 meV. Finally, Fig. 3D compares the intensity of the Raman peak and the light
power as a function of the injected current for these two devices. In the device with the AlInAs
barrier layers, the Raman peak follows linearly the intensity of the measured laser signal, in
agreement with the fact that both emissions originated from the same phonon polariton. Note,
however, that the Raman emission in backscattering is normally forbidden by the momentum
selection rule, because of the very small momentum carried by the phonon polariton. Emission is
still detected because of the very short penetration length of the pump laser. As a result, it is
actually difficult to use the intensity of the Raman sideband as a true quantitative measure of the
phonon weight of the polariton.
In order to further quantify the polaritonic nature of the emission, high resolution ((cid:39)(cid:81) =
0.015 meV) optical spectra of various Fabry Pérot (FP) lasers of fixed cavity lengths (1 mm),
designed to have their optical gain peak between 53 meV and 47 meV, are reported in Fig. 4A. As
3
evident from the data, the longitudinal mode spacing rapidly shrinks in a very small photon energy
range (≈ 5 meV). The group refractive index ng, experimentally retrieved from the angular
frequency spacing Δω of the longitudinal modes of the FP cavity using ng = c/vg = πc / (LΔω)
where c is the light velocity and vg is the group velocity, is plotted by red circles in Fig. 4B. The
group index exhibits a very strong frequency dependence, changing by a factor of two within 2
meV. Such highly dispersive ng is expected for a phonon polariton, arising from the "slowing
down" of the group velocity as the upper polariton branch departs from the light line and converges
to the pure phonon excitation. As a contrast, the blue triangles in Fig. 4B that report the group
index ng of the reference quantum cascade laser with GaAsSb barriers show a flat dispersion. The
comparison between these two characteristics shows clearly that the strong dispersion observed in
the group index ng must be predominantly attributed to the presence of the AlAs optical phonon
and that the contribution of the intersubband gain is negligible.
In contrast to the case of the exciton polaritons, where the dispersion (cid:90)(k) of the polaritons
is measured directly by angle-dependent photoluminescence (30), our approach of retrieving ng((cid:90))
= c (∂k/∂ω) is a measure of its inverse derivative. In Fig. 5B, we compare the experimental results
to the prediction of a theoretical model (described in the section S4 in the supplementary
materials), in which the light-matter coupling is studied in the dipole gauge, and the quadratic term
of the vector potential appears as a polarization self-interaction (24, 31). In this model, the Rabi
coupling energy ΩR between the polarization arising from the phonon excitation and the
fundamental cavity mode, responsible for the anti-crossing between them, can be written at
resonance as
ΩR = (ωp/2) fp1/2, (1)
where ωp is a plasma frequency associated with the phonon excitation having a resonance energy
of ħωo, and fp is their filling factor of the phonon-containing material in the cavity. The polariton
dispersion is obtained by solving the following equation:
(ω2 (ωo2+ωp2))(ω2 ωc2) = fpωc2ωp2, (2)
where ωc the cavity photon frequency. We consider the AlAs TO phonon excitation as providing
a mechanical resonance at ωo = ωTO. The effective plasma frequency of the Al and As ions is given
by ωp = (ωLO2 ωTO2)1/2. fp is here defined as the volume fraction of AlAs as used in a dielectric
function model under the effective medium approximation. The measured AlAs ħωLO = 45.8 meV
and ħωTO = 43.2 meV, together with fp = 0.067, give ΩR =2.0 meV.
Equation (2) is solved using the following approach. At an energy ħωA = 47.7 meV chosen
to be much larger than the anticrossing point of the phonon and photon excitations, the slope of
the dispersion is determined mostly by interband transitions (12). In addition, at this point, the
contribution of the polarizations of InAs and GaAs is large enough to neglect that one of AlAs
because of their higher volume ratios (41% for GaAs and 53% for InAs) as compared to the one
of AlAs (6%). Hence the waveguide group index ng and effective index neff are approximately
determined by the reference QCLs in which all the AlInAs barriers are replaced by GaAsSb ones.
While ng (= 4.3) is retrieved from the longitudinal mode spacing of the FP cavity laser of one of
the InGaAs/GaAsSb reference structures, the effective index neff (= 3.3) is derived from the grating
period dependence on the emission wavelength of first order distributed feedback lasers based on
4
the same reference structure (see in Figs. 2S in the Supplement material). Thus, the bare cavity
frequency ωc is given as a function of k by:
ωc ωA = (c/ng) (k ωAneff/c). (3)
The computation of the polaritonic branches by solving the equations (2) and (3) enables us to
compute the group index of ng = c (∂k/∂ω) as a function of ω and compare it to the measured ng as
shown by the red line in Fig. 4B. As a comparison, we also report by a dashed line in Fig. 4B the
predictions of a classical model that treats the active region as a quasi-bulk material, and whose
details are reported in a supplemental material section S3 (32). This approach completely fails to
correctly predict the experimental data. The excellent agreement between the computed dispersion
of ng and the measured one is further evidence of the polaritonic nature of the emission.
In Fig. 4C, the measured emission energies are plotted onto the computed upper branch of
the phonon polariton dispersion. From their wave vectors, the phonon and photon fraction of the
polaritons can be derived from the Hopfield coefficients αphoton and αphonon (with αphoton + αphonon =
1) (31, 33), as shown in Fig. 4D, a phonon fraction as high as αphonon = 65% is inferred.
Discussion
The fact that the lasing is observed on the Raman scattering spectra is a strong indication of the
phonon-polariton nature of the emitted radiation. Indeed, while the creation of sidebands on a near
infrared carrier has been reported previously, the effect was either based on the bulk GaAs χ(2)
non-linearity using long interaction length and a careful phase matching (34) or based on a resonant
near-infrared excitonic non-linearity (35). In contrast, in our case, the interaction length is very
short (< 0.5 (cid:80)m) and the laser is detuned. The Raman scattering is a consequence of the large
phonon component of the polaritons and the large resulting optical non-linearity at the polariton
energy. The large phonon component of the polariton is also responsible for the very steep
dispersion of the emission which translates into a very strong energy dependence of the group
index.
In our polariton laser, the threshold is reached when the overall polariton gain, sum of its
photon and phonon components (24), equals the losses. The polariton lifetime can be written in
terms of the weighted content of the photon and phonon part: 1/τpolariton = αphoton/τcavity +
αphonon/τphonon. Considering the values for the device operating at an energy of 47.2 meV with the
maximum phonon fraction of 0.65, we evaluate for the cavity a value τcavity = 1.7 ps, taking into
account absorption losses by the metal and the active region. The width of the TO phonon Raman
peak strongly depends on the angle of the extracted light (36), and therefore does not reflect the
lifetime broadening in our setup. Taking only the intrinsic broadening (36), we obtain τpolariton =
3.2 ps, corresponding to an energy broadening of 0.2 meV. The latter is one order of magnitude
smaller than the Rabi coupling energy (ΩR = 2.0 meV), justifying the existence of the polariton
description even in the absence of gain, in contrast to our previous results reported in (37). In order
to lower the threshold of phonon-polariton lasing, one possibility is to use binary AlAs barriers
without disorder atomic distributions as a much narrower phonon linewidth is theoretically
predicted in that case (38).
In contrast to the upper polaritonic branch, we did not observe laser action in the lower
polariton branch. We attribute this behavior to the larger polariton losses, arising from the
proximity of the InAs and GaAs phonons.
5
In conclusion, we have demonstrated the first electrically pumped phonon polariton laser.
Our claim is supported by the direct observation of the photon, phonon, and polariton signatures
of the emission. Specifically, we have measured the photon emission into a laser mode close to the
AlAs LO phonon frequency, a highly non-thermal phonon population at the laser frequency above,
but not below, laser threshold, as well as a highly frequency-dependent group index, which
matches very well the one predicted from a phonon-polariton dispersion and strongly deviates
from that of a classical model. Finally, our simulations of the phonon-polartion emission rate show
a peak at a frequency close to the observed laser frequency, and a similar threshold current density
to the experimental one. A unique feature of this laser is that a large fraction of the energy of the
emitted "light" is carried in the mechanical motion of the atoms, and this device can therefore be
seen as a version of a phonon laser. Indeed, because of their bosonic nature, it is legitimate to
consider a laser-like process in which a coherent population of phonons with an occupation number
much larger than one is created and maintained by pumping. The first implementation of such an
idea used as phonon modes the mechanical excitations of the Mg+ ion in a trap potential (39).
Using an opto-mechanical platform, a phonon laser operating at a phonon frequency of 21 MHz
was demonstrated by optical pumping. A pure mechanical phonon laser operating at 100 kHz has
also been achieved recently using a piezo-electric excitation of a micromechanical resonator (40).
Much higher frequencies, achieved using optical phonons of solids, have also been considered (41,
42). However, as was pointed out by Chen and Khurgin (41), the very large optical phonon density
of states makes the realization of such optical phonon lasers difficult as all the modes of the
resonator within the energy range of the gain must be populated until threshold is reached for the
mode with the large gain and lower losses. In the work presented here, the use of phonon polaritons
alleviates this problem as the coupling of the phonon to the photon effectively decreases the density
of states close to the anticrossing point, thus reducing the threshold current density. The effect is
similar to the one observed in microcavities, where the coupling of the exciton to the cavity mode
reduces their mass and enables their condensation. Finally, because the energy is partially stored
under the form of phonons that have a relatively long lifetime and are naturally confined within
nanometer thick layers, phonon polariton lasers could be operated with extremely small cavity
sizes that would maintain higher quality factors than the one possible with plasmonics (43). As a
result, the design concept explored here would be very well suited for applications based on two-
dimensional Van der Waals materials, where low-loss Boron Nitride phonons could provide the
phonon-polariton mode (44).
I. Carusotto, C. Ciuti, Quantum fluids of light. Rev. Mod. Phys. 85, 299-366 (2013).
References and Notes:
1.
2. J. Kasprzak, M. Richard, S. Kundermann, A. Bass, P. Jeambrun, J. M. J. Keeling, F. M.
Marchetti, M. H. Szymańska, R, André, J. L. Staehli, V. Sanova, P. B. Littlewood, B.
Deveaud, Le Si Dang, Bose-Einstein condensation of exciton polaritons. Nature 443, 409-414
(2006).
3. Daniele Sanvitto, Stéphane Kéna-Cohen, The road towards polaritonic devices. Nat. Mater.
15, 1061-1073 (2016).
4. D. Ballarini, M. De Giorgi, E. Cancellieri, R. Houdré, E. Giacobino, R. Cingolani, A. Bramati,
G. Gigli, D. Sanvitto, All-optical polariton transistor. Nat. Commun. 4, 1778 (2013).
6
5. C. Sturm, D. Tanese, H. S. Nguyen, H. Flayac, E. Galopin, A. Lemaître, I. Sagnes, D.
Solnyshkov, A. Amo, G. Malpuech, J. Bloch, All-optical phase modulation in a cavity-
polariton Mach-Zehnder interferometer. Nat. Commun. 5, 3278 (2014).
6. A. Imamoglu, R. J. Ram, S. Pau, Y. Yamamoto, Nonequilibrium condensates and lasers
without inversion: exciton-polatiron laser. Phys. Rev. A 53, 4250-4253 (1996).
7. C. Schneider, A. Rahimi -- Iman, N. Y. Kim, J. Fischer, I. G. Savenko, M. Amthor, M. Lermer,
A. Wolf, L. Worschech, V. D. Kulakovskii, I. A. Shelykh, M. Kamp, S. Reitzenstein, A.
Forchel, Y. Yamamoto, S. Hӧfling, An electrically pumped polariton laser. Nature 497, 348-
352 (2013).
8. Raffaele Colombelli, Jean-Michel Manceau, Perspective for intersubband polariton laser.
Phys. Rev. X 5, 011031 (2015).
9. P. Li, X. Yang, T. W. W. Maβ, J. Hanss, M. Lewin, A. -- K. Michel, M. Wuttig, T. Taubner.
Reversible optical switching of highly confined phonon-polaritons with an ultrathin phase-
change material. Nat. Mater. 15, 870-876 (2016).
10. H. Huebl, C. W. Zollitsch, J. Lotze, F. Hocke, M. Greifenstein, A. Marx, R. Gross, S. T. B.
Goennenwein, High cooperativity in coupled microwave resonator ferromagnetic insulator
hybrids. Phys. Rev. Lett. 111, 127003 (2013).
11. Y. Tabuchi, S. Ishino, T. Ishikawa, R. Yamazaki, K. Usami, Y. Nakamura, Hybridizing
ferromagnetic magnons and microwave photons in the quantum limit. Phys. Rev. Lett. 113,
083603(2014).
12. D. L. Mills, E. Burstein, Polaritons: the electromagnetic modes of media. Rep. Prog. Phys.
37, 817-926 (1974).
(1965).
13. C. H. Henry, J. J. Hopfield, Raman scattering by polaritons. Phys. Rev. Lett. 15, 964-966
14. J. M. Yarborough, S. S. Sussman, H. E. Purhoff, R. H. Pantell, B. C. Johnson, Efficient,
tunable optical emission from LiNbO3 without a resonator. Appl. Phys. Lett. 15, 102-105
(1969).
15. K. Kawase, M. Sato, T. Taniuchi, H. Ito, Coherent tunable THz wave generation from LiNbO3
with monolithic grating coupler. Appl. Phys. Lett. 68, 2483 (1996).
16. J. J. Greffet, R. Carminati, K. Joulain, J. P. Mulet, S. Mainguy, Y. Chen, Coherent emission
of light by thermal sources. Nature 416, 61-64 (2002).
17. T. Feurer, J. C. Vaughan, K. A. Nelson, Spatiotemporal coherent control of lattice vibrational
waves. Science 299, 374-377 (2003).
18. N. S. Stoyanov, D. W. Ward, T. Feurer, K. A. Nelson, Terahertz polariton propagation in
patterned materials. Nat. Mater. 1, 95-98 (2002).
19. D. J. Shelton, I. Brener, J. C. Ginn, M. B. Sinclair, D. W. Peters, K. R. Coffey, G. D. Boreman,
Strong coupling between nanoscale metamaterials and phonons. Nano Lett. 11, 2104-2108
(2011).
20. S. Dai, Z. Fei, Q. Ma, A. S. Rodin, M. Wagner, A. S. McLeod, M. K. Liu, W. Gannett, W.
Regan, K. Watanabe, T. Taniguchi, M. Thiemens, G. Dominguez, A. H. Castro Neto, A. Zettl,
F. Keilmann, P. Jarillo-Herrero, M. M. Fogler, D. N. Basov, Tunable phonon polaritons in
atomically thin van der Waals crystals of Boron Nitride. Science 343, 1125-1129 (2014).
21. J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, A. Y. Cho, Quantum cascade
laser. Science 264, 553-556 (1994).
22. J. Faist, M. Beck, T. Aellen, E. Gini, Quantum-cascade lasers based on a bound-to-continuum
transition. Appl. Phys. Lett. 78, 147-149 (2001).
7
23. R. Terazzi, J. Faist, A density matrix model of transport and radiation in quantum cascade
lasers. New J. Phys. 12, 033045 (2010).
24. M. Franckié, C. Ndebeka-Bandou, K. Ohtani, and J. Faist, Phys. Rev. B 97, 075402 (2018).
25. A. Wacker, M. Lindskog, D.O. Winge, Nonequilibrium Green's function model for simulation
of quantum cascade laser devices under operating conditions. IEEE J. Sel. Top. Quant.
Electron. 19 1200611-1 -- 1200611-11 (2013)
26. K. Unterrainer, R. Colombelli, C. Gmachl, F. Capasso, H. Y. Hwang, A. Sergent, A. Michael,
D. L. Sivco, A. Y.Cho, Quantum cascade lasers with double metal-semiconductor waveguide
resonators. Appl. Phys. Lett. 80, 3060-3062 (2002).
27. H. Ding, G. Weihs, C. Santori, J. Bloch, Y. Yamamoto, Condensation of semiconductor
microcavity exciton polaritons. Science 298, 199-202 (2002).
28. H. Ding, G. Weihs, D. Snoke, J. Bloch, Y. Yamamoto, Polariton lasing vs. photon lasing in a
semiconductor microcavity. Proc. Natl. Acad. Sci. USA 100, 15318-15323 (2003).
29. K. Ohtani, M. Beck, M. J. Süess, J. Faist, A. M. Andrews, T. Zederbauer, H. Detz, W. Schrenk,
G. Strasser, Far-infrared quantum cascade lasers operating in AlAs phonon reststrahlen band.
ACS Photonics 3, 2280-2284 (2016).
30. R. Houdré, C. Weisbuch, R. P. Stanley, U. Oesterle, P. Pellandini, M. Ilegems, Measurement
of cavity-polariton dispersion curve from angle-resolved photoluminescence experiments.
Phys. Rev. Lett. 73, 2043-2046 (1994).
31. Y. Todorov, C. Sirtori, Intersubband polaritons in the electrical dipole gauge. Phys. Rev. B 85,
045304 (2012).
32. H. Chu, Y. -C. Chang, Phonon-polariton modes in superlattices: The effect of spatial
dispersion. Phys. Rev. B 38, 12369-12376 (1988).
33. J. J. Hopfield, Theory of the contribution of excitons to the complex dielectric constant of
crystals. Phys. Rev. 112, 1555-1567 (1958).
34. S. S. Dhillon, C. Sirtori, J. Alton, S. Barbieri, A. De Rossi, H. E. Beere, and D. A. Ritchie,
Terahertz transfer onto a telecom optical carrier. Nat. Photonics 1, 411 -- 415, (2007).
35. S. Houver, P. Cavalié, M. R. St-Jean, M. I. Amanti, C. Sirtori, L. H. Li, A. G. Davies, E. H.
Linfield, T. A. S. Pereira, A. Lebreton, J. Tignon, and S. S. Dhillon, Optical sideband
generation up to room temperature with mid-infrared quantum cascade lasers. Opt. Express,
23, 4012 -- 4019 (2015).
36. M. Giehler, E. Jahna, K. Ploog, Linewidth of phonon modes in infrared transmission spectra
of GaAs/AlAs superlattices. J. Appl. Phys. 77, 3566-3568 (1995)
37. K. Ohtani, C. Ndebeka-Bandou, L. Bosco, M. Beck, J. Faist, arXiv:1610.00963.
38. A. Debernardi, Phonon
in III-V semiconductors from density-functional
linewidth
perturbation theory. Phys. Rev. B 57, 12847-12858 (1998).
39. K. Vahala, M. Herrmann, S. Knünz, V. Batteiger, G. Saathoff, T. W. Hänsch, Th. Udem, A
phonon laser. Nat. Phys. 5, 682-686 (2009).
40. I. Mahboob, K. Nishiguchi, A. Fujiwara, H. Yamaguchi, Phonon
lasing
in an
electromechanical resonator. Phys. Rev. Lett. 110, 127202 (2013).
41. J. Chen, J. B. Khurgin, Feasibility analysis of phonon lasers. IEEE J. Quantum Electron. 39,
600-607 (2003).
42. J. B. Khurgin, S. Bajaj, S. Rajan, Amplified spontaneous emission of phonons as a likely
mechanism for density-dependent velocity saturation in GaN transistor. Appl. Phys. Exp. 9,
094101 (2016).
8
43. J. D. Caldwell, L. Lindsay, V. Giannini, I. Vurgaftman, T. L. Reinecke, S. A. Maier, O. J.
Glembocki. Low-loss, infrared and terahertz nanophotonics using surface phonon polaritons.
Nanophotonics 4, 1-26 (2014).
44. A. J. Giles, S. Dai, I. Vurgaftman, T. Hoffman, S. Liu, L. Lindsay, C. T. Ellis, N. Assefa, I.
Chatzakis, T. L. Reinecke, J. G. Tischler, M. M. Fogler, J. H. Edgar, D. N. Basov, and J. D.
Caldwell, Ultra-low loss polaritons in isotopically pure boron nitride. Nat. Mat. 17, 134-139
(2018).
45. C. Deutsch, M. Krall, M. Brandstetter, H. Detz, A. M. Andrews, P. Klang, W. Schrenk, G.
Strasser, K. Unterrainer. High performance InGaAs/GaAsSb terahertz quantum cascade lasers
operating up to 142 K. Appl. Phys. Lett. 101, 211117 (2012).
46. G. Scamarcio, L. Tapfer, W. Kӧnig, A. Fischer, K. Ploog, E. Molinari, S. Baroni, P.
Giannozzi, and D. de Gironcoli, Infrared reflectivity by transverse-optical phonons in
(GaAs)m/(AlAs)n ultrathin-layer superlattices. Phys. Rev. B 43, 14754-14757 (1991).
47. O. Madelung, U. Rӧssler, and M. Schulz, eds., "Indium arsenide (InAs) dielectric constants"
in Group IV elements, IV-IV and III-V compounds. Part a -- lattice properties (Springer Berlin
Heidelberg, Berlin, Heidelberg, 2001) pp. 1-3.
48. D. J. Lockwood, G. Yu, and N. L. Rowell, Optical phonon frequencies and damping in AlAs,
GaP, GaAs, InP, InAs and InSb studied by oblique incidence infrared spectroscopy. Sol. Stat.
Comm. 136, 404-409 (2005).
49. O. Madelung, U. Rӧssler, and M. Schulz, eds., "Aluminum arsenide (AlAs) dielectric
constants" in Group IV elements, IV-IV and III-V compounds. Part a - lattice properties
(Springer Berlin Heidelberg, Berlin, Heidelberg, 2001) pp. 1-4.
50. M. A. Ordal, R. J. Bell, R. W. Alexander, Jr., L. L. Long, and M. R. Querry, Optical properties
of fourteen metals in the infrared and far infrared: Al, Co, Cu, Au, Fe, Pb, Mo, Ni, Pd, Pt, Ag,
Ti, V and W, Appl. Opt. 24, 4493 (1985).
Acknowledgments:
We acknowledge very fruitful discussions with Angela Vasanelli, Simone de Liberato, and Jacob
B. Khurgin, and also thank Aaron Maxwell Andrews, Tobias Zederbauer, Hermann Detz, Werner
Schrenk, and Gottfried Strasser for supporting MBE growth of InGaAs/GaAsSb material at early
stage of this work. Funding: The presented work is supported in part by the ERC Advanced grant
Quantum Metamaterials in the Ultra Strong Coupling Regime (MUSiC) with the ERC Grant
340975, as well as by the Swiss National Science Foundation (SNF) through the National Center
of Competence in Reasearch Quantum Science and Technology (NCCR QSIT). Author
contributions: KO did the design of all structures, the MBE growth of Sb-based material, the
computation of the theoretical polariton dispersion and dielectric function calculations, the device
characteristics, and the low temperature Raman scattering experiment under biasing. BM has
conducted recent Raman experiments and LB processed the grown structures into laser devices.
MF, CNB, and JF have developed the theoretical framework and software for the phonon-polariton
gain calculations. MB performed MBE growth of the Sb-free structures. KO and JF had the
original idea and have drafted the initial manuscript, and MF and BM have done the reviewing and
editing. Finally, JF has been supervising this work. Competing interests: Authors declare no
competing interests. Data and materials availability: All data is available in the main text or the
supplementary materials. The measured samples and the codes used to generate the theoretical
data are available at ETH Zürich.
9
Supplementary Materials:
Materials and Methods
Figures S1-S3
Table S1
References (45-50)
A
B
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
C
Emission rate
Loss rate
46
48
50
52
56
54
58
Energy (meV)
60
62
64
)
1
-
s
p
(
e
t
a
R
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
66
o
i
t
a
r
s
s
o
l
/
i
n
o
s
s
m
E
i
Fig. 1. Schematics and design of the device structure. (A) Schematic of the phonon-polariton
laser. In the waveguide the Au metal layers sandwich the quantum cascade active region and guide
the photon part of the phonon polariton with a confinement factor of ≈ 1. In the magnified view, a
part of the computed heterostructure potential with the relevant electron wavefunctions in the
active region (depicted by white lines) is overlaid on the computed squared electric field of the
fundamental cavity mode. Being associated with the AlAs vibrational mode, the phonon part of
10
the phonon polariton is confined in the AlInAs barriers. As shown by the displayed mode, the
squared electric field is concentrated in the barrier material in a deep subwavelength dimension.
(B) The contribution of the mechanical energy Umechanical to the total one Utotal as a function of
photon energy and position in the active region. The upper figure shows the ratio of Umechanical/Utotal
at z = 5.3 nm. The dotted vertical line indicates the observed lasing frequency of the device. The
right figure shows the corresponding position in the active region with the relevant wavefunctions
overlaid. (C) The polariton emission and loss rates (left axis), as well as the ration between the
emission and loss rates (right axis) computed at an electric field of 19 kV/cm and current density
of 10.4 kA/cm2. The gain is computed by first solving the transport in a non-equilibrium Green's
function model and using these results to obtain the polariton coupling constants (see section S5
of the supplementary materials). Since the calculated photon lifetime τphot. = 1.7 ps is shorter than
the phonon lifetime τphon. = 6 ps, the laser will operate at a lower frequency (48.2 meV) than that
of peak polartion emission (49.6 meV), where the phonon component of the polariton is larger.
11
Fig. 2. Device characteristics. (A) Measured subthreshold emission spectra (T = 10 K). Two peaks separated by the
AlAs reststrahlen band were observed. Those peaks stem from emission from the lower and upper phonon polariton
X Y Z X geometry. (C) Room
branches. (B) Measured Raman spectrum taken from the active region in the
temperature direct transmission spectrum using a room temperature Deuterated Tri Glycine Sulfate (DTGS) detector.
(D) Light output versus current characteristic. A single threshold current (= 880 mA) was observed. The inset shows
the transport curve of the device. The ridge width of the device was 30 μm and the length was 250 μm.
)
(
,
12
Polariton
laser
Reference
laser
Fig. 3 (A) Low temperature (T = 5.5 K) polarized Raman scattering spectra of the facet of the
phonon-polariton laser under operation. The peaks at 28 meV and 32 meV are attributed to the
InAs and GaAs phonons of the active region, respectively. The phonon-polariton peak at 48 meV
is seen when the device is driven above threshold (1060 mA) and disappears when the device is
below threshold (670 mA). (B) Polarization selection rules for the phonon-polariton peak, which
appears only for the horizontal-horizontal polarization combination. (C) Raman spectrum of the
reference sample at a current of 835 mA, which is close to the roll-over current of the device. (D)
Light and Raman side band intensity of the phonon-polariton and the reference devices as function
of the injected current.
13
)
.
u
.
a
(
y
t
i
s
n
e
n
t
I
6
5
4
3
2
1
A
0
-0.75
-0.50
B
wA
A
11
10
)
g
9
w
y
8
c
n
e
7
u
q
e
r
6
F
r
a
u
5
g
n
A
4
l
n
(
x
e
d
n
I
e
v
i
t
c
a
r
f
e
R
p
u
o
r
G
3
46
47
wA(1- neff(wA)/ng(wA))
(a) E = 47.2 meV
(b) E = 47.9 meV
(c) E = 48.6 meV
(d) E = 51.0 meV
(e) E = 53.0 meV
0.50
0.75
-0.25
0.00
Energy (meV)
0.25
Red: with AlAs phonon
Blue: without AlAs phonon
Classical dielectric function
Linear slope: ng(wA)
model
A (neff(wA)wA/c, wA)
Computed polariton dispersion
w = (c/ng(wA))k
+ wA(1- neff(wA)/ng(wA))
48
50
49
51
Energy (meV)
Wavevector k
52
53
54
60
55
50
45
40
35
30
1.0
0.8
0.6
0.4
0.2
0.0
0.2
)
V
e
m
(
y
g
r
e
n
E
n
o
i
t
c
a
r
F
n
o
n
o
h
P
/
n
o
t
o
h
P
C
D
aphonon = 0.65
Photon
Phonon
aphonon = 0.06
0.6
1.0
1.2
Wavevector (µm-1)
0.8
1.4
0.4
Figure 4
Fig. 4. Phonon polariton laser emission spectra. (A) High resolution emission spectra of the
Fabry-Pérot (FP) devices. The cavity length of all the devices was fixed at 1.0 mm. The
longitudinal mode spacing is strongly reduced when the laser emission energy approaches to the
AlAs optical phonon energy. The spectral measurements were done at T = 15 K. (B) Group
14
refractive index (ng) retrieved from the longitudinal mode spacing of the FP devices. The red solid
points represent ng of the FP devices containing the AlAs phonon oscillators while the group index
ng of the FP devices without the AlAs phonon oscillators are depicted by the blue solid points. The
red solid line shows ng derived from the computed phonon polariton dispersion. (C) Computed
dispersion of the two polariton branches, including the contribution from the two phonon modes
(AlAs, InAs and GaAs). The point A is the energy at which the values of the effective and group
indices are experimentally measured. (D) Computed phonon polariton dispersion for our laser
devices. The thick red lines represent the range of emission energies of the measured devices. (C)
Computed Hopfield coefficients. In this model, a phonon fraction of 0.65 was obtained at the
lowest emission energy (47.2 meV).
15
Keita Ohtani, Bo Meng, Martin Franckié, Lorenzo Bosco, Camille Ndebeka-Bandou,
Mattias Beck, Jérôme Faist
1
Supplementary Materials for
An electrically pumped phonon polariton laser
This PDF file includes:
Materials and Methods
Supplementary Text
Figs. S1 to S3
Table S1
Materials and Methods
Materials
by
60
times:
in
repeated
(EV2128) was
The samples were grown by molecular beam epitaxy on Fe-doped InP (001) substrates,
starting with a 60 nm thick In0.53Ga0.47As buffer layer. Then a 4.1 μm thick active region
was deposited and the growth was completed by a 15 nm thick Si-doped n-type
Ga0.53In0.47As contact layer (n = 5.0 × 1018 cm-3). In the active region, the following layer
structure
nanometers,
3.0/6.0/0.3/9.3/0.4/8.9/0.4/8.8/0.4/7.7/0.5/6.3/0.8/6.6/1.4/7.5, where Al0.48In0.52As barrier
layers are in bold, In0.53Ga0.47As well layers are in roman, and Si-doped In0.53Ga0.47As layer
(n = 4.1 × 1017 cm-3) is underlined. All the layer structures presented here are summarized
in Table S1. The corresponding conduction band diagram of the one period of the active
region with moduli-squared relevant wave functions is also shown in Fig. S1.
Methods
For laser device fabrication, a bottom metal contact layer composed of a 10 nm thick
Titanium (Ti) and a 500 nm thick gold (Au) was first deposited on the grown wafer. It was
wafer-bonded onto the Au evaporated n-InP (100) carrier substrate by thermo-compression
bonding technique. Then, the Fe-doped InP substrate used for the epitaxial growth was
removed by mechanical polishing and selective wet chemical etching. After evaporation of
a 260 nm thick Ti/Au top contact, the ridge laser structures were defined by a wet chemical
etching. After device fabrication, the processed wafer was cleaved into 0.25-1.0 mm long
Fabry-Pérot lasers with a ridge width of 30 μm, and then mounted on copper mounts. The
laser devices were placed in a temperature controlled He flow cryostat with KRS5 vacuum
windows.
For the low temperature (T = 5 K) micro-Raman scattering experiment, a continuous
wave frequency doubled Nd:YAG (wavelength: 532.1 nm) laser was focused through the
cryostat window into a 2.5 μm diameter spot by a ×50 objective lens with a numerical
aperture of 0.4. Backward scattered light was collected and sent to the single grating
monochrometer equipped with a LN2 cooled Si CCD detector. According to the Raman
(
)
scattering selection rule for a zinc blende crystal,
(cid:99) geometries
Y X Z Y
,
(cid:99)
(cid:99)
were used for LO and TO phonons, respectively. Here, X denotes the (cid:62)
(cid:64)
100 axis, which is
parallel to the InP substrate surface (100), and Y (Z) is along the direction of (cid:62)
(cid:64)001
). Also, Y (cid:99) and Z(cid:99) denote the (cid:62)
(cid:64)011 and the 011
(cid:188) axes, repsectively. The power of the
(cid:186)
laser was varied from 5 mW to a maximum of 20 mW, limited by local heating at the focal
point.
X Y Z X and
To determine the AlAs TO phonon energy, a normal incident light transmission was
measured. The active region film was glued on a Si substrate. To compensate the
temperature difference of the two experiments, the transmission spectrum was shifted by
0.8 meV, which is consistent with the temperature-induced energy shift of the LO phonon
energy as measured by Raman scattering experiment.
The sub-threshold electroluminescence measurements (T = 10 K) were performed
using a home-made vacuum Fourier Transform Infrared (FT-IR) spectrometer equipped
with a calibrated Si bolometer. A current pulse train composed of 100 ns wide pulses with
a repetition rate of 1 MHz (a duty cycle of 10%) were modulated at 450 Hz to match the
frequency response of the calibrated Si bolometer. The light output power was also
(cid:64)010 ((cid:62)
(
,
)
(cid:170)
(cid:172)
2
measured by the Si bolometer. For high-resolution spectral measurements, a Bruker
vacuum FT-IR was used. To minimize the broadening of the spectra, the length of the
electrical pulses was kept below 50 ns. The spectral resolution used in this study was 0.125
cm-1.
Supplementary Text
S1. Layer structures and emission characteristics
In Table S1, the grown layer structures including thickness and chemical composition
are summarized. The QCL structures are based on the In0.53Ga0.47As/Al0.48In0.52As material
while the reference ones are composed of the AlAs-free In0.53Ga0.47As/GaAs0.51Sb0.49
material. Here, Al0.48In0.52As barrier layers are in bold, In0.53Ga0.47As well layers are in
roman, GaAs0.51Sb0.49 barrier lays are in italic, and Si-doped In0.53Ga0.47As layer (n = 4.1 ×
1017 cm-3) is underlined. All the samples used here were grown on Fe-doped (001) InP
substrates by molecular beam epitaxy. The computed conduction band diagrams together
with the relevant electron wavefunctions of (A) EV2128 and (B) EP1561 are depicted in
Fig. S1. Here, an electric field of 19.5 kV/cm is applied to align subbands. The band
calculation was done by self-consistent Poisson and Schrödinger solver. The well layers
are composed of lattice matched In0.53Ga0.47As to InP, while the barrier layers are lattice
matched Al0.48In0.52As (A) or GaAs0.51Sb0.49 layers (B). Effective masses of 0.043 m0 for
In0.53Ga0.47As, 0.076 m0 for Al0.48In0.52As, and 0.045 m0 for GaAs0.51Sb0.49 (m0 is a free
electron mass) are used. The conduction band offset energy is 0.52 eV for (A) Al0.48In0.52As
and 0.36 eV for (B) GaAs0.51Sb0.49 (45). Other structures have similar band structures. Also
low temperature threshold current densities and emission wavelength are summarized in
Table S1.
S2. Laser characteristics of the first order distributed feedback laser based on
InGaAs/GaAsSb
First order distributed feedback (DFB) laser structures equipped with a lateral grating
and a quarter wave shift, as depicted in Fig. S2 A, were used for the measurements of the
effective refractive index neff. The wafer used was EP1562 listed in Table S1. The DFBs
show single mode emission with a threshold current density of ≈ 350 mA (see Figs. S2 B
and C). As seen in Fig. S2 D, neff = 3.3 was obtained from a relationship between the lasing
wavelengths and the grating periods, which is in good agreement with the simulated results
by the commercial software (COMSOL Multiphysics 5.2a). A group index of ng = 4.3 is
measured by the laser mode spacing of the Fabry-Pérot laser as depicted in Fig. 2S E.
S3. Computed results by a classical dielectric function
We consider the structure as a uniaxial crystal with an average dielectric constant
and write the dielectric constant (εz) for a direction (z) perpendicular to the layers in the
framework of the effective media approximation (32, 46):
(1)
1
total
well
1
(cid:72)
z
=
(cid:167)
(cid:168)
(cid:168)
(cid:169)
d
total
d
(cid:72)
(cid:183)(cid:167)
(cid:184)(cid:168)
(cid:184)(cid:168)
(cid:185)(cid:169)
well
+
total
barrier
d
(cid:72)
barrier
(cid:183)
(cid:184)
(cid:184)
(cid:185)
,
)
(
d
total
total
barrier
welld
where
is the total thickness of the In0.53Ga0.47As well (Al0.48In0.52As barrier)
layers, εwell (εbarrier) is the dielectric constant of the In0.53Ga0.47As well (Al0.48In0.52As
barrier) layer, dtotal is the total thickness of wells and barriers (
. εwell and
d
εbarrier of the alloy materials are given by the following formulae:
total
barrier
total
well
=
d
+
d
)
total
3
(cid:72) (cid:90) (cid:72)
=
well
(cid:102)
InGaAs
(
)
+
x
(cid:72)
(cid:102)
InAs
In
−
)
(
2
LO InAs
(
(cid:90)
(
(cid:90)
2
TO InAs
(
)
)
)
+
x
(cid:102)
(cid:72)
Ga GaAs
2
TO InAs
(
(cid:90)
−
2
(cid:90)
)
−
)
(
2
LO GaAs
(
(cid:90)
(
(cid:90)
2
TO GaAs
(
)
)
)
(
x
In
+
x
Ga
=
)
1
, (2)
(cid:90)
−
2
TO GaAs
(
2
(cid:90)
)
(cid:72) (cid:90) (cid:72)
=
barrier
(cid:102)
AlInAs
+
x
(cid:72)
(cid:102)
InAs
In
(cid:99)
(
)
−
)
(
2
LO InAs
(
(cid:90)
(
(cid:90)
2
TO InAs
(
)
2
TO InAs
(
(cid:90)
−
2
(cid:90)
)
)
)
+
x
(cid:72)
(cid:102)
AlAs
Al
−
)
(
2
LO AlAs
(
(cid:90)
(
(cid:90)
2
TO AlAs
(
)
2
TO AlAs
(
(cid:90)
−
2
(cid:90)
)
)
)
(
x
In
(cid:99)
+
x
Al
=
)
1
, (3)
, (1)
)
(
)
(
)
(
)
(
)
(
)
(
(cid:102)
(cid:102)
In
Al
In
)
x
x
x
+
+
=
(cid:72)
(cid:72)
(cid:102)
InAs
(cid:102)
AlAs
(cid:102)
InAs
(cid:72)
(cid:72)
Ga GaAs
(cid:102)
AlInAs
(
(cid:72)
10.9
x
(cid:99)=
InGaAs(cid:72)(cid:102)
AlAs(cid:72)(cid:102) =
GaAs(cid:72)(cid:102) =
(47),
( AlInAs(cid:72)(cid:102)
(cid:72)
(48), and
where xi expresses the group III atom compositional ratio (in this case, xIn = 0.53, xGa =
0.47, xIn' = 0.52, xAl = 0.48),
) is the high frequency dielectric constant of
In0.53Ga0.47As (Al0.48In0.52As) defined as InGaAs
. The
longitudinal and transverse optical phonon energies of InAs, GaAs, and AlAs of the QCL
active region were determined by low temperature (T = 5 K) polarized Raman scattering
measurements. Because those energies did not show structure dependence, the following
energies were used:
TO GaAs(cid:90) = 31.3 meV,
LO AlAs(cid:90) = 45.8 meV. For high-frequency
LO GaAs(cid:90) = 33.2 meV,
dielectric constants, the literature values of
12.3
8.2
(49) were used. A dispersion of the Au dielectric function is also taken into account (50).
In Fig. S3, the red dotted straight line represents the computed ng using
of the
literature. The computed result based on the classical dielectric model does not reproduce
the steep dispersion of ng experimentally observed using the experimental parameters.
S4. Phonon-photon interaction Hamiltonian in the dipole gauge
TO AlAs(cid:90) = 43.2 meV, and
InAs(cid:72)(cid:102) =
LO InAs(cid:90) = 28.4 meV,
TO InAs(cid:90) = 27.5 meV,
r
( )
z(cid:72)(cid:72)
Here we briefly describe how to derive the phonon-photon coupling energy from the
theory developed for an intersubband polariton in a dipole gauge by Y. Todorov and C.
H in the
Sirtori (31). Details can be found in Ref. (24). The interaction Hamiltonian
electrical dipole gauge is written, neglecting the magnetic interaction,
(cid:170)
(cid:171)
(cid:172)
P r is the polarization density operator of the
where ( )D r is the displacement field,
( )
( )z(cid:72) is the background dielectric function in the z-direction (crystal growth
phonon, and
direction). Due to a selection rule of the intersubband transitions, the polarization of the
photonic part is Transverse Magnetic (TM). The two metal plates, which are separated by
the active region (of thickness cavL ) and have an area S, guide the fundamental mode (TM0)
with a field confinement factor of ≈ 1. The mode profile function is defined as
z , and
( )
pf
( )
z dz L
z = ,
( ) 1
( )
r
( )
r
( )
r
GaAs AlAs InAs
1
2
(cid:72)(cid:102)
D
H
(cid:186)
(cid:187)
(cid:188)
P
P
pf
=
=
−
+
d
(cid:179)
cav
f
(cid:152)
int
int
p
3
2
0
,
,
in the case of perfect metallic boundaries, the TM0 mode satisfies
. The displacement field operator is thus expressed as
(cid:72)(cid:72) (cid:90) (cid:152)
(
e
2
D z r
,
( )
z
(cid:166)
SL
−
=
a
a
)
)
(
q r
i
i
f
†
−
cq
p
q
q
q
0
cav
(cid:102)
(cid:179)
−(cid:102)
cq(cid:90) is an angular frequency of the guided mode with an in-plane wavevector of q,
where
and
a a are photon creation and annihilation operators. The polarization is given by
†,q
q
4
, (2)
(
P z r
,
)
=
*
e
SM
2
(cid:166)
q
(cid:68)
( )
z
(cid:91)
(cid:90)
TO
q r
i
(cid:152)
e
(
d
†
(cid:68)
−
q
+
d
q
(cid:68)
)
. (6)
where dαq is the TO phonon annihilation operator, e* is the effective charge of the ions, M
is their reduced mass, and the phonon microcurrent is defined by
via the ground and first excited states of the harmonic oscillator potential:
(cid:92)
=
(cid:79)
(cid:92)
−
(cid:80)
(cid:91)
(cid:68)
=
( )
z
e
d
(cid:92)
(cid:79)
dz
2 1
2
(cid:83)(cid:86)
d
(cid:92)
(cid:80)
dz
2
(cid:167)
−(cid:168)
(cid:169)
z
(cid:86)
(cid:183)
(cid:184)
(cid:185)
. (5)
(cid:79)(cid:92)
(cid:167)
= (cid:168)
(cid:169)
1
2
(cid:86)(cid:83)
1 4
(cid:183)
(cid:184)
(cid:185)
2
z
−
e (cid:86)
2
2
,
(cid:80)(cid:92)
(cid:167)
= (cid:168)
(cid:169)
1
2
2
2
(cid:86)(cid:83) (cid:86)
1 4
(cid:183)
(cid:184)
(cid:185)
2
z
−
ze (cid:86)
2
2
. (3)
Here, σ measures the extent of the polarization due to the presence of the phonon modes
inside a material layer. We relate this quantity to the thickness d of the layer by equating
the filling factor fa0 (Eq. (27) of Ref. (24)) with d/Lcav, which results in σ = d/(cid:151)(2π). After
a Bogoliubov transformation for diagonalizing the quadratic Hamiltonian, we finally
obtain
(12)
D Pdr
a
a
p
a
p
p
=
+
,
(cid:152)
cq
P
(
(cid:58) −
a
q
)(
†
−
q
†
(cid:68)
−
q
)
q
(cid:68)
0
3
(cid:179)
=
(cid:166)
1
(cid:72)(cid:72)
(cid:90)(cid:90)
2
(cid:90)
)
where pαq is the LO phonon annihilation operator and
(cid:58) =
)(
†
(cid:68)
−
−
+
p
0
(cid:68)
(
f
q
(cid:68)
LO
†
−
cq
q
q
q
q
o
f(cid:68)
(cid:166)
q
(13)
P
(cid:90)(cid:90)
2
(cid:90)
LO
o
R
P
.
2
p
−
=
f(cid:68)
2
LO
(cid:58) =
(cid:90)
2
is the light-matter coupling constant, which is similar to the expression for Ω of the
intersubband polariton in the dipole gauge (31). Here,
is the phonon plasma
frequency and is deduced from the measured phonon frequencies. Hence, the Rabi coupling
energy (Eq. (1) in the main text), expressing the minimum splitting of the two blanches, is
given by
S5. Transport and gain simulations
(14)
2
(cid:90) (cid:90) (cid:90)
TO
The structure EV2128 has been simulated using a non-equilibrium Green's function
(NEGF) model (25), and the resulting current-bias is shown in Fig. S3 A. There is a good
agreement with the experimental data, considering the simulated optical gain is starting to
surpass the calculated losses for a current density of 10.4 kA/cm2, which is close to the
observed threshold current density of 10.3 kA/cm2 (which in addition includes a phononic
contribution to the gain). The gain spectra are also consistent with the EL spectra in Fig. 2
in the main text considering the large phonon optical absorption between 42-46 meV
photon energy, and which also show a blue shift with increasing injected current. We then
take the results from the NEGF transport simulation to calculate the phonon-polariton
emission rate as in Ref. (24), resulting in the emission rate shown in Fig. 1 C in the main
text.
5
Fig. S1 Computed conduction band diagram with relevant electron wavefunctions of
a part of the active region (A) EV2128 and (B) EP1561. An electric field of 19.5 kV/cm
is applied to align the subbands.
6
B
D
T = 15 K
10 μm
E
A
C
Fig. S2 Characteristics of the first order distributed feedback (DFB) lasers and the
Fabry-Perot (FP) lasers of the reference InGaAs/GaAsSb structure. (A) SEM image
of the fabricated first order DFB. (B) Voltage-current-light output characteristic of the 1st
order DFB laser (T = 10 K). (C) Single mode laser emission of the 1st order DFB laser (T
= 15 K) (D) Emission wavelength versus grating period of the 1st order DFB laser. neff =
3.3 is derived. The red, blue, and green solid circles are the simulated results by COMSOL
Multiphysics 5.2a. (E) Laser emission spectrum of the FP laser. ng of 4.3 around an energy
of 47.7 meV is determined.
7
•
In0.53Ga0.47As/Al0.48In0.52As QCL layer structures
ID
Layer structure in nanometer
EV2036
EV2039
EV2091
EV2092
EV2106
EV2107
EV2128
3.0/5.4/0.4/8.6/0.5/8.2/0.5/8.1/0.6/7.1/0.7/6.1/1.1/6.4/2.0/7.2
3.0/5.5/0.4/8.8/0.5/8.4/0.5/8.3/0.6/7.2/0.7/6.2/1.1/6.5/2.0/7.3
3.0/5.7/0.4/9.1/0.5/8.7/0.5/8.6/0.6/7.5/0.7/6.5/1.1/6.8/2.0/7.6
3.0/5.8/0.4/9.3/0.5/8.9/0.5/8.7/0.6/7.7/0.7/6.6/1.1/6.9/2.0/7.8
3.0/5.5/0.3/8.7/0.4/8.3/0.4/8.2/0.5/7.1/0.6/6.1/1.1/6.6/2.0/7.0
3.0/5.9/0.3/9.3/0.4/8.9/0.4/8.8/0.5/7.7/0.6/6.6/1.1/7.0/1.6/8.0
3.0/6.0/0.3/9.3/0.4/8.9/0.4/8.8/0.4/7.7/0.5/6.3/0.8/6.6/1.4/7.5
Jth & λ
(Low Temp)
6.2 kA/cm2
23.6 μm
7.6 kA/cm2
24.4 μm
5.7 kA/cm2
25.1 μm
5.8 kA/cm2
25.3 μm
6.7 kA/cm2
26.0 μm
9.0 kA/cm2
25.6 μm
10.3 kA/cm2
26.3 μm
•
In0.53Ga0.47As/GaAs0.51Sb0.49 reference QCL layer structures
ID
Layer structure
EP1561
EP1562
4.8/5.4/0.7/8.6/0.9/8.2/0.9/8.1/1.0/7.1/1.2/6.1/1.6/6.4/3.0/7.2
4.8/5.4/0.6/8.6/0.8/8.2/0.8/8.1/0.9/7.1/1.1/6.1/1.6/6.4/3.0/7.2
Jth & λ
(Low Temp)
3.7 kA/cm2
25.5 μm
5.3 kA/cm2
26.6 μm
Table S1. Layer structures and properties of all the grown samples. Threshold current
densities (Jth) and emission wavelengths (λ) were measured around T = 50 K.
8
)
2
m
c
/
A
k
(
y
t
i
s
n
e
d
t
n
e
r
r
u
C
16
14
12
10
8
6
4
2
0
)
m
c
/
1
(
n
a
G
i
600
400
200
0
-200
-400
-600
A
0
5
B
10
20
Electric field (kV/cm)
15
25
19.0 kV/cm
23.4 kV/cm
30 35 40 45 50 55 60 65 70
Energy (meV)
Figure S3. Non-equilibrium Green's function simulation results. (A) Simulated current
density at a lattice temperature of 100 K. The maximum current density is close to the
experimental one (13 kA/cm2). (B) Simulated optical gain at two applied electric fields;
close to the experimental threshold current (green) and close to the maximum current
density (red). For electric fields between these two, the gain peak lies in the region of
anomalous dispersion between the TO and LO phonon energies. At the lower electric field,
the optical gain at an energy of 48 meV is 65/cm, just below the calculated optical losses
of 68/cm.
9
|
1902.06716 | 2 | 1902 | 2019-05-28T17:13:40 | Holograms to focus arbitrary ultrasonic fields through the skull | [
"physics.app-ph"
] | We report 3D-printed acoustic holographic lenses for the formation of ultrasonic fields of complex spatial distribution inside the skull. Using holographic lenses, we experimentally, numerically and theoretically produce acoustic beams whose spatial distribution matches target structures of the central nervous system. In particular, we produce three types of targets of increasing complexity. First, a set of points are selected at the center of both right and left human hippocampi. Experiments using a skull phantom and 3D printed acoustic holographic lenses show that the corresponding bifocal lens simultaneously focuses acoustic energy at the target foci, with good agreement between theory and simulations. Second, an arbitrary curve is set as the target inside the skull phantom. Using time-reversal methods the holographic beam bends following the target path, in a similar way as self-bending beams do in free space. Finally, the right human hippocampus is selected as a target volume. The focus of the corresponding holographic lens overlaps with the target volume in excellent agreement between theory in free-media, and experiments and simulations including the skull phantom. The precise control of focused ultrasound into the central nervous system is mainly limited due to the strong phase aberrations produced by refraction and attenuation of the skull. Using the present method, the ultrasonic beam can be focused not only at a single point but overlapping one or various target structures simultaneously using low-cost 3D-printed acoustic holographic lens. The results open new paths to spread incoming biomedical ultrasound applications including blood-brain barrier opening or neuromodulation. | physics.app-ph | physics |
Holograms to focus arbitrary ultrasonic fields through the skull
Sergio Jim´enez-Gamb´ın, No´e Jim´enez,∗ Jos´e Mar´ıa Benlloch, and Francisco Camarena
Instituto de Instrumentaci´on para Imagen Molecular, Consejo Superior de Investigaciones Cient´ıficas,
Universitat Polit`ecnica de Val`encia. Camino de vera s/n 46022 Val`encia, Spain
(Dated: May 29, 2019)
We report 3D-printed acoustic holographic lenses for the formation of ultrasonic fields of complex
spatial distribution inside the skull. Using holographic lenses, we experimentally, numerically and
theoretically produce acoustic beams whose spatial distribution matches target structures of the
central nervous system. In particular, we produce three types of targets of increasing complexity.
First, a set of points are selected at the center of both right and left human hippocampi. Experiments
using a skull phantom and 3D printed acoustic holographic lenses show that the corresponding
bifocal lens simultaneously focuses acoustic energy at the target foci, with good agreement between
theory and simulations. Second, an arbitrary curve is set as the target inside the skull phantom.
Using time-reversal methods the holographic beam bends following the target path, in a similar
way as self-bending beams do in free space. Finally, the right human hippocampus is selected as
a target volume. The focus of the corresponding holographic lens overlaps with the target volume
in excellent agreement between theory in free-media, and experiments and simulations including
the skull phantom. The precise control of focused ultrasound into the central nervous system is
mainly limited due to the strong phase aberrations produced by refraction and attenuation of the
skull. Using the present method, the ultrasonic beam can be focused not only at a single point
but overlapping one or various target structures simultaneously using low-cost 3D-printed acoustic
holographic lens. The results open new paths to spread incoming biomedical ultrasound applications
including blood-brain barrier opening or neuromodulation.
I.
INTRODUCTION
Holographic plates are surfaces that when illuminated
by a wave, typically light, modify the phase of the trans-
mitted or reflected wavefront in such a manner that a
complex image can be formed [1 -- 3]. In recent years, sub-
wavelength thickness holographic metasurfaces have been
designed using structured materials with subwavelength
resonances, namely metamaterials [4, 5]. Analogously,
in acoustics, a broad range of locally-resonant structures
have been proposed to obtain a precise control of the
wavefront at a subwavelength scale [6, 7], including ef-
fective negative mass density [8] and/or bulk modulus
metamaterials [9, 10]. Acoustic metamaterials allow an
accurate control of the reflected [11 -- 15] or transmitted
wavefronts [16 -- 18]. The use of these structures has been
exploited to design negative-refraction superlenses [19] or
hyperbolic dispersion-relation hyperlenses [20] that ex-
hibit subwavelength focusing properties in the near field.
Holographic lenses have also been reported in acoustics to
generate complex acoustic fields [21 -- 24]. Multi-frequency
holograms have been also reported [25]. Equivalently,
using phased-array sources it has been reported the gen-
eration of complex beam patterns [26], self-bending and
bottle beams[27], or vortex beams for particle levitation
and manipulation [28]. Mixed approaches between meta-
materials and phased arrays have been also presented
[29].
In these applications, holographic lenses have demon-
strated the ability to manipulate acoustic waves in free
∗ [email protected]
media, i.e., without inhomogeneities. However, when us-
ing ultrasound in biomedical engineering applications,
acoustic beams encounter in their path multiple tissue
layers of complex geometry with non-homogeneous prop-
erties. For instance, an accurate control of the focused
beam is at the basis of focused ultrasound therapy tech-
niques, e.g., as in high intensity focused ultrasound hy-
perthermia, thermal ablation or histotripsy, or in extra-
corporeal shockwave lithotripsy [30, 31]. Focusing di-
rectly into human soft-tissues can efficiently be achieved
by using conventional systems as ultrasound beam aber-
rations are typically small in these media [32]. However,
when the target tissue lays behind high-impedance tis-
sues, e.g., soft-tissue surrounded by bones, the beam ex-
periences strong aberrations due to refraction, reflection
and absorption processes [33]. Some applications make
use of existing acoustic windows by targeting tissues from
specific locations. Nevertheless, in the case of transcra-
nial propagation skull bones are always present in the
path towards the central nervous system (CNS). In this
way, the precise control of acoustic focus into the CNS is
mainly limited due to the strong phase aberrations pro-
duced by the refraction and attenuation of the skull [34].
To overcome these limitations, minimally-invasive
techniques were developed in the past to design active
focusing systems using the time-reversal invariance of the
acoustic propagation [35] or phase conjugation methods
[36]. In minimally-invasive techniques, a small acoustic
source is introduced into the skull, together with a biopsy
catheter. When the catheter reaches the target tissue it
radiates a short ultrasonic pulse that travels outwards
and it is recorded by a hemispherical multi-element ar-
ray surrounding the patient's head. Then, the elements
of the phased-array are set to re-emit the time-reversed
recorded waveforms (or phase conjugated harmonic sig-
nals). Due to spatial reciprocity and time-reversal in-
variance of the acoustic media, the generated wavefront
focuses at the catheter location, i.e., at target tissue [35].
Later, it was demonstrated that non-invasive versions of
these techniques can be obtained using numerical simu-
lations [37, 38]. In these techniques a tomographic image
is previously obtained from patient's head to extract the
geometry of the skull and its acoustic properties [38].
Using full-wave simulations the time-reversed wavefront
is calculated by exciting the simulation with a virtual
source at the desired focal spot. Then, a physical hemi-
spherical phased-array is excited with the synthetic time-
reversed waveforms [39] or phase conjugated signals [37],
and sharp focusing through skull-aberration layers is re-
trieved. Other techniques include the optimization of
phase-arrays using magnetic resonance imaging (MRI) to
maximize acoustic-radiation-force induced displacements
into the target focus [40]. However, up-to-date phased-
array systems are restricted to a limited number of chan-
nels, e.g., 1024 for the Exablate R(cid:13) Model 4000 (InSightec,
Ltd)[41], that can be insufficient to fully record the re-
quired holographic information in order to conform a
complex beam pattern.
In addition, phased-arrays are
effective, but due to its high economical cost it is desir-
able to use passive structures to control acoustic beams.
Only few theoretical works have tackled the problem
of beam focusing through aberrating layers using meta-
materials [42] or phase plates [43, 44]. In Ref. [42] a 2D
configuration was proposed theoretically using a meta-
surface based on membranes. Recently, the use of phase
plates to generate simple focused sources have been re-
ported to avoid beam aberrations in transcranial propa-
gation [43]. However, the technique was limited to focus
the beam into a single focal spot at the near field of the
source. Besides, in some non-thermal transcranial ultra-
sound applications such as blood-brain barrier opening
[45] or neuromodulation [46] the ultrasound beam might
be set to fully-cover a geometrically complex CNS struc-
ture rather than focusing over a small focal spot.
In this work, we propose the use of 3D-printed holo-
graphic phase plates to produce ultrasonic fields of arbi-
trary shape into the human brain. The holographic lenses
designed in this work allow the reconstruction of complex
diffraction-limited acoustic images including the compen-
sation of the aberrations produced by a skull phantom.
In particular, we theoretically, numerically and exper-
imentally demonstrate the generation of several holo-
graphic patterns, of increasing complexity, all with di-
rect practical application to biomedical ultrasound: an
arbitrary set of points, an arbitrary curved line, and
an arbitrary volume. First, we provide the conditions
to generate a simple holographic pattern, i.e., a set of
diffraction-limited focal points, as sketched in Fig. 1 (b).
In particular, we extend the use of holographic lenses
to generate bifocal beams, matching both foci simulta-
neously the location of left and right human hippocam-
pus. Second, we demonstrate that ultrasonic beams with
2
FIG. 1.
(a) Scheme of the holographic lens focusing over
a target CNS structure. (b) Focusing on a set of arbitrary
points (bifocal holographic lens), (c) Focusing over arbitrary
line (self-bending holographic lens), (d) Focusing over an ar-
bitrary volume (volumetric holographic lens).
curved trajectory along the internal CNS tissues can also
be produced, as Fig. 1 (c) shows. In this way, the acoustic
beam can be bent following arbitrary paths producing a
self-bending beam inside the CNS. Finally, we report the
generation of a beam pattern that overlaps with the vol-
ume of a specific CNS structure, as shown in Fig. 1 (d),
in particular we target the right human hippocampus.
II. METHODS
The process of hologram generation is composed of
four steps. First, we extract the geometry and acous-
tic properties of a human skull from X-ray CT images,
as shown in Fig. 2 (a), and from MRI tomographic im-
ages we identify the target tissue structure, e.g., the right
human hippocampus as shown Fig. 2 (b). Second, a
back-propagation method is used to calculate the acous-
tic wavefront generated from a set of virtual sources and
impinging on a holographic surface located outside the
SourceUltrasonic beamHolographic plate(a)Target tissue(arbitrary shape)(b) Multiple point Multiple p(c) Line(c) Line(d) Volumetric) Volume 3
FIG. 2. Hologram generation process. (a) CT+MRI tomographic images. (b) Selected target (red volume) acting as a virtual
acoustic source and holographic recording surface (blue area), (c) Lens design using the TR back-propagated field. (d) Forward
propagation from the holographic lens (red area) to the target tissue (blue volume).
skull phantom, as shown in Fig. 2 (b). Third, the phase-
plate lens is generated by using the phase and amplitude
of the recorded wavefront at the holographic surface, as
shown in Fig. 2 (c). Finally, the lens is excited with a flat
and uniform ultrasonic transducer and the target acous-
tic image is reconstructed by either theoretical, numerical
forward-propagation or experimental methods, as shown
in Fig. 2 (d).
A. Tomographic image acquisition
First, in order to model the skull geometry, we used
the CT Datasets of a female human head with an
isotropic resolution of 1 mm (interpolated to 0.22 mm for
the numerical simulation) from the National Library of
Medicine's Visible Human Project available for general
use by the University of Iowa. Experiments were con-
ducted in a 3D printed skull phantom, while, in addition,
we included full-wave simulations using the acoustical
properties of the skull bones. Thus, for the skull phan-
tom simulations we used homogeneous acoustical param-
eters matching those of the 3D printing material, while
for the realistic skull simulations we used the same geom-
etry but the inhomogeneous acoustical parameters of the
skull were derived using the same the CT data, convert-
ing the apparent density tomographic data in Hounsfield
units to density and sound speed distributions using the
linear-piecewise polynomials proposed in Refs. [47, 48].
After, we used a human atlas made publicly avail-
able by the International Consortium for Brain Mapping
(ICBM) from the Laboratory of Neuro Imaging [49]. This
atlas provided us T1 weighted MRI data that was used to
identify the shape and location of the human hippocam-
pus.
In particular we used for segmentation the ITK-
SNAP software [50] to obtain the shape and location of
the left and right hippocampi.
B. Calculation methods
We use two methods, of increasing complexity, to
estimate the back-and-forward acoustic fields: a semi-
analytical method using Rayleigh-Sommerfeld diffraction
integral and a pseudo-spectral time-domain simulation
method.
On the one hand, for theoretical calculations in homo-
geneous media, i.e, in water without the skull phantom,
the acoustic pressure field given by p(r) at point r, gen-
erated by a moving surface S of arbitrary shape located
at coordinates r0 and vibrating with a complex parti-
cle velocity v0(r0) normal to the surface, is given by the
Rayleigh-Sommerfeld diffraction integral [51]:
v0(r0) exp (−k0 r − r0)
(cid:90)
dS,
(1)
p(r, ω) =
iωρ0
2π
S
r − r0
where ω = 2πf ; k0 = ω/c0, c0 and ρ0 are the wavenum-
ber, sound speed and density of water. Note that in
Eq. (1) diffraction is captured exactly as compared with
angular spectrum methods, so it can be applied to high-
aperture sources.
On the other hand, for calculations including aberra-
tion layers we use a pseudo-spectral simulation method
with k-space dispersion correction to numerically inte-
grate the linearized constitutive relations of acoustics
In an inhomogeneous and absorbing media,
[52, 53].
the governing equations,
i.e., the continuity equation,
the momentum conservation equation and the pressure-
density relation, can be written as three-coupled first-
order partial differential equations as:
∂ρ
∂t
∂u
∂t
= −ρ0∇ · u − u · ∇ρ0,
= − 1
ρ0
0 (ρ + d · ∇ρ0 − Lρ) ,
∇p,
p = c2
(2)
(3)
(4)
where u is the acoustic particle velocity, d is the acous-
tic particle displacement, p is the acoustic pressure, ρ is
the acoustic density, ρ0 is the ambient (or equilibrium)
050100150200x (mm)050100150200y (mm)050100150200x (mm)050100150200050100150200050100150200holographic lenszxy(a) CT + MRI imagesVirtual sourcesHolographic surfaceSkull phantomTarget tissue (acoustic image)Holographic lensSkull phantom(b) Backward propagation(d) Forward propagation(c) Lens designphase (x,y)magnitude (x,y)-20-1001020x (mm)-20-1001020y (mm)00.20.40.60.81-20-1001020x (mm)-20-1001020y (mm)--/20/2 4
arrival of the wavefront. Finally, for the volumetric holo-
gram, as sketched in Fig. 1 (c), a set of virtual sources
were spatially distributed with a separation of λ/6 (to
match the numerical grid used) over a sagittal plane of
the right human hippocampus. The recorded field was
captured at a given surface, i.e, at a holographic surface,
outside the skull phantom.
Second, the recorded conjugated pressure distribution
at the working frequency was used to design the physical
lens. The lens surface was divided in squared pixels of
different height, h(x, y) and uniform width, ∆h, as shown
in Fig. 3. We assume each elastic column to vibrate lon-
gitudinally as a Fabry-P´erot resonator. For each column,
the field at the holographic plane located at x0 = (x, y, d)
is given by the complex transmission coefficient[55]:
T (x0) =
2Ze−ik0[d−h(x0)]
2Z cos [kLh(x0)] + i (Z 2 + 1) sin [kLh(x0)]
,
(6)
where d is the distance from the bottom of the lens (z =
0) to the holographic surface, the normalized impedance
is given by Z = ZL/Z0, and Z0 = ρ0c0 is the impedance
of water and ZL = ρLcL, kL = ω/cL, ρL and cL, are
the impedance, wavenumber, density and sound speed of
the lens material. In order to obtain the height of each
pixel of the lens, an analytic inversion of Eq. (6) is not
possible due to the trigonometric terms. Instead, we first
numerically evaluate the expression for a broad range of
pixel heights ranging from a minimum height that was set
to hmin = 5 mm to guarantee structural consistency, to
a given height that provides a phase of the transmission
coefficient 2π greater than for hmin, i.e., d = 15 mm,
and using steps of 1 µm, well below the printer accuracy.
Finally, we performed interpolation using a cubic-spline
method to obtain the heigh of the pixel as a function of
the required phase. In this way, by tuning the height of
each Fabry-P´erot resonator the phase at the output of
each pixel can be tailored to that of a target holographic
surface.
However, using this kind of lenses the degree of freedom
to modify the magnitude of the field at the holographic
surface is limited. Iterative methods were employed in
the past to obtain equivalent lenses only with phase dis-
tributions [21]. In this work, iterative methods are pro-
hibitive: the 3D simulations including aberration layers
involve long calculation times, e.g., 20 hours in a Intel R(cid:13)
Xeon R(cid:13) CPU E5-2680 v2 2.80GHz, 256 GB RAM, us-
ing a CPU parallel implementation of the code. Instead,
we use a direct method to estimate an equivalent holo-
graphic lens of uniform field magnitude [56]. The basis of
this direct method is the sequential scanning of the pix-
els to modify the complex transmission coefficient. The
method work as follows: First, the odd and even rows
are scanned from opposite directions, and a bidirectional
error of the diffusion process is calculated. The magni-
tude of each visited pixel is forced to be a constant value
while the exact phase value is preserved. The resulting
FIG. 3. Geometry of the holographic lens. The lens, of aper-
ture 2a is subdivided in pixels of height h(x, y). The source
is located at z = 0, while the holographic plane is located at
z = d.
density, c0 is the sound speed, and L is a linear opera-
tor introducing the frequency-dependent absorption and
dispersion [52]. Tissue absorption following a power-law
on frequency given by α(ω) = α0ωγ, where α0 is the ab-
sorption coefficient and γ is the exponent of the frequency
power law, together with its corresponding physical dis-
persion are included by the integro-differential operator
as:
+ η(cid:0)−∇2(cid:1) γ+1
2 −1
,
(5)
(cid:0)−∇2(cid:1) γ
2 −1
L = τ
∂
∂t
0
and η = 2α0cγ
where τ = −2α0cγ−1
0 tan (πγ/2) and the
absorption and dispersion proportionality coefficients.
This operator is solved efficiently using the fractional
Laplacian in the k-space. This simulation method was
selected as it provides low numerical dispersion as com-
pared with finite-differences methods [54]. We used a
numerical grid with a spatial step of ∆x = ∆y = ∆z =
223 µm and a numerical temporal step of ∆t = 19.1
ns, leading to a Courant-Friedrichs-Lewy number [52] of
0.13 in water and an spatial sampling of 6 grid points
per wavelength in water for a frequency of 1 MHz. These
parameters were fixed to all simulations in this paper.
C. Lens design
Under the assumption of reciprocity, time-invariance
and linearity of the system, a time-reversal (TR) tech-
nique together with a direct method was used to design
the only-phase holographic lens.
First, we set some virtual sources inside the skull phan-
tom and the back-propagated field was estimated at a
given surface outside the skull phantom. For the bifo-
cal lens, two virtual sources were set as monopoles with
same phase and amplitude, located at the center of mass
of the two hippocampi (right and left), as sketched in
Fig. 1 (a). For the self-bending beam, a set of 50 vir-
tual sources were located following an arbitrary curve as
sketched in Fig. 1 (b), each source compensated by a
phase factor of exp(ikzz) accounting for the direction of
zxyh(x,y)¢hz=d,z=0,2aholographic planesource plane error is diffused to the neighboring pixels. Finally, the re-
sult gives a surface with a modified phase depending on
the bidirectional error diffusion process [56]. The main
limitation of this method is that if the pixel width is
small it will appear areas with isolated long pixels, i.e.,
columns, that can experience bending modes. Note this
do not imply that a lens cannot be designed, but the
theory presented here only apply to longitudinal modes
on each pixel. The size of the pixels used in this work,
5/6 times the wavelength, is thick enough to ensure that
the resonance frequency of the first bending mode is far
away from the first longitudinal Fabry-P´erot resonance
frequency.
Third, lens with an aperture of 2a = 50 mm were
manufactured using 3D printing techniques. On the one
hand, the bifocal holographic lens was manufactured by
additive 3D printing techniques using Ultimaker 3 Ex-
tended (Ultimaker B.V., The Netherlands) with a res-
olution of 100 µm in both, lateral and axial directions
and PLA material. As, in general, the height profile of
the lens is smooth, we set the square pixel resolution to
∆h = 0.22 mm. The acoustical properties of PLA ma-
terial were obtained experimentally using a pulse-echo
technique in a test cylinder, resulting in a measured
sound speed of cL = 1818 m/s and a density of ρL = 1127
kg/m3, matching with those reported in existing litera-
ture [57], and the absorption was set to α = 13.72 dB/cm
at 1.112 MHz [57]. On the other hand, the self-bending
and volumetric holographic lenses, which needed a more
accurate printing technique for their complex pattern,
were 3D printed using Polyjet techniques with an Ob-
jet30 printer (Stratasys, USA), with a resolution of 100
µm and 28 µm in lateral and axial directions respectively,
and using a photo-resistive polymer (Veroclear R(cid:13), Strata-
sys, USA). As a result of the direct method to obtain the
equivalent holographic lens of uniform field magnitude
[56], the height distribution presents high spatial modu-
lations (see Fig. 2 (c)). Thus, the pixel resolution was
increased to ∆h = 1 mm to ensure each column vibrates
as a longitudinal Fabry-P´erot resonator avoiding bending
modes around the working frequency for the volumetric
hologram. For this material we experimentally estimated
cL = 2312 m/s and ρL = 1191 kg/m3 and α = 3.06
dB/cm at 1.112 MHz, matching the values reported in
existing literature [21].
D. Skull phantom
The geometry of the skull phantom was extracted
from the 3D CT images as described previously. The
sound speed and density distributions were first esti-
mated from the apparent density given by the CT images
in Hounsfield units [47, 48]. Then, as the 3D printing
technique results in homogeneous material, the acous-
tic properties, including the absorption were considered
uniform along the skull bone volume [58 -- 60]. The skull
phantom was manufactured by additive 3D printing tech-
5
niques using Ultimaker 3 Extended (Ultimaker B.V., The
Netherlands) with resolution of 100 µm in both, lateral
and axial directions and using PLA material. The acous-
tic parameters for the 3D printed phantom are the same
than for the PLA lenses.
Finally, for the simulations using a realistic skull, the
acoustical parameters were derived using the CT data,
converting the apparent density in Hounsfield units to
density and sound speed distributions using the linear-
piecewise polynomials proposed in Refs. [47, 48]. The
density data ranges between ρ0 = 1000 kg/m3 (water)
and ρmax = 2206 kg/m3 (bone), the sound speed values
range between c0 = 1500 m/s and cmax = 3117 m/s,
matching those reported in literature [31, 61] and the
bone absorption was set to 12.6 dB/cm at 1.112 MHz
[62].
The details about the measurement system can be
found in Appendix A.
III. MULTIPLE-POINT HOLOGRAMS
We start with the bifocal holographic lens. First, two
points located at the center of mass of both left and
right human hippocampi are selected. Second, we set
this pair of points as the location of virtual sources for
the TR method. For the lens designs in free media,
i.e., without the skull, we make use of the Rayleigh-
Sommerfeld diffraction integral (see Methods section for
further details). For the lens designs of holographic sur-
faces including the skull-aberration layers we make use
of low-numerical-dispersion simulations based on pseudo-
spectral methods [52].
In this way the simultaneous
back-propagation of the fields irradiated by both virtual
monopoles can be calculated at the holographic surface
which is located at the rear part of the skull. The phase-
plate lens is designed using the conjugated complex field
recorded at the surface. Then, the lens is placed at the
location of the holographic surface as shown in Fig. 4 (a),
and a forward-propagation calculation is carried out to
test the quality of the reconstructed acoustic image.
The field produced by the bifocal lens propagating
through a human occipital/parietal skull phantom in-
cluding the compensation for the aberrations of the skull
is shown in Figs. 4 (a-f). First, Figs. 4 (a,b) show the
axial field cross-section, p(x, y = 0, z), using the pseudo-
spectral simulation method and measured experimen-
tally, respectively. We observe that the reconstructed
field accurately matches the target foci, and the exper-
imental results agree with the simulations. The corre-
sponding transverse field distributions at z = 105 mm
are shown in Figs. 4 (c,d) where sharp focusing is ob-
served. The focal spots present larger dimensions in the
axial (z) direction than in transverse ones, as expected
from limited-aperture holographic lenses, where the spa-
tial spectral components in axial direction are limited
by the finite-aperture source[63]. Axial (measured at
x = 25 mm and y = 0 mm) and transversal (measured
6
FIG. 4. Axial field cross-section obtained for the bifocal lens using simulations (a) and experiments (b). (c-d) Corresponding
transversal pressure field distributions. Colorbar in units normalized to the peak pressure. (e,f) Simulated and experimental
normalized axial and transversal field cross-sections, respectively.
at z = 105 mm and y = 0 mm) cross-sections are shown
in Figs. 4 (e,f), respectively. Excellent agreement is ob-
served between simulation and experiment for the axial
field profile at the focal region. A small secondary lobe
located before the main one appears experimentally. The
transverse profile shows a small lateral shift of ± 0.5 mm
in both experimental foci towards the x-axis origin.
Note that, due to diffraction, the geometrical focus of
a geometrically focused source do not correspond to the
acoustic focus of the source [63]. In our case, the target
location was set to z = 105.5 mm, the acoustical focus
of an equivalent focused source of same frequency and
aperture in water peaks at z = 99.8 mm and the focus
of the lens peak at z = 100.4 mm and z = 100.1 mm
in simulations and experiments including the skull phan-
tom, respectively. These shifts corresponds to errors of
0.6% and 0.3%, respectively, showing the accuracy of the
focusing performance of the holographic lenses.
IV. SELF-BENDING BEAMS
tory are, in principle, not available [27]. Instead, we make
use of a TR method: a set of virtual sources were placed
along this trajectory and their back-propagated field were
calculated. We set a factor of (z/zmax) exp(izkz) to com-
pensate for the amplitude and phase of each source to
set the main direction of propagation, being kz the ax-
ial component of the wave-vector and zmax the distance
to the farthest virtual source (45 mm in this example).
A sketch of the target trajectory is shown in Fig. 5 (a).
The axial and transversal cross-sections of the forward
propagated field in water are shown in Figs. 5 (a,b), re-
spectively. We observe that using TR method the self-
bending beams can be obtained, and the beam accurately
follows the target trajectory. Using simulation and a lens
made of elastic material a similar result is obtained, as
shown in Figs. 5 (c,d). The experimental tests show a
similar pressure field distribution in comparison with the-
ory using the Rayleigh-Sommerfeld integral and simula-
tions using pseudo-spectral methods. A lateral shift of
the peak pressure location of 0.3 mm in the x direction
is observed at z = 30 mm and y = 0 mm in the experi-
ments.
The previous results show that holographic phase
plates can retain phase information of multiple foci. Us-
ing this idea, we can set more complex targets follow-
ing the shape of functional structures found in the CNS.
Here, we set the target holographic pattern to a beam
following a curved trajectory as those reported in ho-
mogeneous media without aberrating layers using active
sources or metamaterials [17, 27, 64]. As the aberration
layers will be present in the real application, known an-
alytical methods to calculate the phase of the 3D trajec-
Finally, when the aberration layer of the skull phantom
is included the corresponding holographic lens also recon-
structs the target acoustic image with curved trajectory,
as shown in Figs. 5 (e,f). A similar lateral shift of the
peak pressure location in the experiments, of 0.25 mm
in the x direction is observed at z = 30 mm and y = 0
mm. The measured pressure field inside the skull phan-
tom agree the simulation. Note that the transversal size
of the curved beam at z = 30 mm is 1.11, 1.07 and 1.19
times the wavelength in water for the theoretical calcula-
050100150z (mm)00.51SimulationExperiment-50050x (mm)050100150z (mm)00.51-40-200 20 40 x (mm)-10010y (mm)-40-2002040-10010y (mm)-40-2002040x (mm)00.51-40-200 20 40 x (mm)90100110120130140z (mm)ExperimentExperimentSimulationz = 105 mmz = 105 mmy = 0 mmSimulation(a)(e)(f)(b)(d)(c)Skull phantom 7
FIG. 5. Theoretical (a) axial and (b) transverse pressure field distribution for the self-bending beam in water. Simulated (c)
axial and (d) transverse pressure field distribution for the self-bending beam in water. Corresponding experimental results
are shown in the insets in (c) and (d).
(e,f) Simulated axial and transversal pressure field including the skull phantom.
Corresponding experimental results are shown in the insets in (e) and (f).
tion, and for the simulations in water and including the
skull phantom, respectively. Both results demonstrate
that using TR methods self-bending beams following a
target curve can be obtained inside the skull phantom
using acoustic holographic lenses.
V. VOLUMETRIC HOLOGRAMS
OVERLAPPING CNS STRUCTURES
Going further, we designed a holographic lens which
produces an acoustic image that fits the right human hip-
pocampus volume. The holographic surface was placed
near the occipital/parietal bones to adapt the acoustic
image to the elongated geometry of the hippocampus.
However, we locate the lens at the center of the skull
symmetry plane in order to demonstrate the steering ca-
pabilities of this holographic lens. The lens generation
process is based on the TR method with multiple vir-
tual sources covering the target area (see Methods sec-
tion for further details). Figure 6 summarizes the results
for both, water and including the aberration layer of the
skull phantom.
On the one hand, Figs. 6 (a,b,c) show the forward-
propagation field distribution of the holographic lens de-
signed for water obtained using the theoretical, experi-
mental and simulation results, respectively. First, we ob-
serve a good agreement between experiments, simulation
and theory in both axial (Figs. 6 (a-c)) and transversal
field distributions (Figs. 6 (f-h)). The beam is steered in
the correct direction and a broad focal spot is generated.
The transversal and axial field cross-sections are shown
in Figs. 6 (d,e). The diffraction-limited image is recon-
structed and the field is enhanced mainly at the target
volume. To quantify the performance, we define the over-
lapping volume as the overlapping volumes of the target
region and the region of the acoustic pressure field under
a threshold corresponding to the half of the peak ampli-
tude.
In particular, using this lens we obtain in water
an overlapping volume of 29.7%, 20.1% and 19.0% for
the theoretical calculation, simulation and experiment,
respectively.
On the other hand, the field distribution produced by
acoustic holographic lenses including the skull phantom
is shown in Figs. 6 (i-n). The experimental forward-
propagation field distribution overlaps a similar volume
in comparison with simulation result. Both holographic
images present the same qualitative performance and
provide a similar overall covering of the interest zone.
In particular, an overlapping volume of of 21.1% and
23.2% was obtained in simulation and experiment, re-
spectively.
In addition, both axial (Fig. 6 (i,j)) and
-20020x (mm)01020304050z (mm)00.51lens(c)-20020x (mm)-30-20-100102030y (mm)00.51Source aperturetarget(d)024-202024-202-20020x (mm)01020304050z (mm)00.51lens(e)-20020x (mm)01020304050z (mm)00.51target trajectory(a)holographic plane-20020x (mm)-30-20-100102030y (mm)00.51Source aperturetarget(b)-20020x (mm)-30-20-100102030y (mm)00.51Source aperturetarget(f)z = 30 mm024-202024-202SimulationExperimentSimulationExperimentExperimentSimulationExperimentSimulationSkull phantom 8
FIG. 6. Volumetric hologram results. (a,b,c) Theoretical, experimental and simulated axial pressure distribution in water. (d,e)
Transversal and axial field cross-sections in water. (f,g,h) Simulated, experimental and theoretical transversal field distribution
in water.
(k,l) Corresponding
transversal pressure distribution and (m,n) transversal and axial cross-section, respectively.
(i,j) Simulated and experimental axial pressure distribution including the skull phantom.
transversal (Figs. 6 (k,l)) field distributions are similar
of those produced in water without the skull phantom,
showing that, first, limited-diffraction holographic vol-
umes can be reconstructed and, second, the aberrations
produced by the skull phantom on these complex beams
can be compensated at the source plane by the acous-
tic holographic lenses. Finally, the transversal and axial
cross-sections, shown in Figs.6 (m,n), show that the ex-
perimental and simulated acoustic holographic lens pro-
duces a field enhancement that matches the target dis-
tribution.
Note that the spatial bandwidth of the image is lim-
ited by the diffraction limit and the spatial bandwidth
of the acoustic holographic lens [21].
In this case, the
holographic lens focuses at z ≈ 74.1λ (100 mm), and its
limited-aperture is only a = 18.5λ (25 mm). Therefore,
the transversal components of the wave-vector are band-
limited and the performance of the holographic lens at
this distance is restricted. Using lenses with larger aper-
ture will improve the quality of the holographic acoustic
image, and, therefore, the total overlapping volumes.
VI. HOLOGRAM SIMULATION USING A
REALISTIC SKULL
It is worth noting here that the impedance of the avail-
able 3D printing material used to manufacture the skull
phantom is soft compared with the skull bone. In this
way, the phase aberrations produced by a real skull will
be stronger than the ones observed in the previous ex-
periments. To demonstrate the focusing performance of
the proposed lenses in a realistic situation a set of sim-
ulations were performed using the acoustical parameters
of skull bones. The parameters were derived using the
same the CT data and were listed in Section II D.
First, the results of the bifocal lens simulation using a
realistic skull are summarized in Fig 7. First, the sagittal
cross-section of the absolute value of the pressure field
at y = 0 mm is shown in Fig. 7 (a). We can see that
the lens focuses at two clear spots, almost at the tar-
get distance. The corresponding traversal cross-section
is shown in Fig. 7 (b) measured at z = 100 mm. In fact,
good agreement is found between the simulations using
a realistic skull and the calculations using the Rayleigh-
Sommerfeld integral considering homogeneous water me-
dia. These two focal spots are generated together with
small amplitude secondary lobes. The amplitude of the
side lobes is -8.86 dB below the peak pressure in the the-
ory in water and -5.16 dB in the simulation including
the skull. To quantify the focusing performance of the
lens, we show in Fig. 7 (c) the transversal cross-section
at z = 100 mm and y = 0 mm. The lateral shift of the
left focus is -26.3 mm and -26.0 mm for the theoretical
prediction and for the simulation, respectively. A relative
error of 1.1% was committed. These small lateral shifts
050100150z (mm)00.51-40-20x (mm)80100120z (mm)-50050-20020y (mm)00.51050100150200z (mm)00.51TheorySimulationExperiment-50050x (mm)00.51TheorySimulationExperiment-40-20x (mm)80100120z (mm)-40-20x (mm)80100120z (mm)050100150z (mm)00.51-10010-40-20x (mm)-10010y (mm)-10010y (mm)-40-20x (mm)-50050x (mm)00.51SimulationExperiment-40-20x (mm)-10010-50050x (mm)-20020y (mm)00.5105010015020000.51SimulationExperiment(d)(e)(m)(n)(i)(j)(b)(a)(c)(f)(h)(g)(l)(k)Simulationtarget volumetarget volumeSimulationSimulationSimulationExperimentExperimentExperimentExperimentTheoryTheorytarget volumetarget volumetarget volumeTargetTargetTargetTargetTheorySimulationExperimentSimulationExperimentSkull phantomz (mm)x (mm) 9
FIG. 7. Simulation results for the bifocal holographic lens designed for a realistic skull. (a) Axial cross-section of the pressure
distribution at y = 0 mm. (b) Transversal cross-section at z = 100 mm. (c) Lateral cross-section at z = 100 mm and y = 0
mm. (d) Lateral cross-section at z = 100 mm and x = −26 = 0 mm. (e) Axial cross-section at x = −26 = 0 mm and y = 0
mm.
are of the order of the experimental test shown previously
with the 3D printed phantom. Moreover, the amplitude
of the side lobes in the simulation using a realistic skull
are 0.3 times the peak pressure. These side-lobes present
higher amplitude in the lateral cross-section joining both
foci (Fig 7 (c)) than in the lateral cross-section measured
at x = 0 mm, as shown in Fig 7 (d). Finally, Fig. 7 (e)
shows the axial pressure distribution measured at the lo-
cation of the right hippocampus. The axial peak location
of the simulation including a realistic skull (z = 99.3 mm)
matches the location of the corresponding peak pressure
using the theory in water (z = 99.8 mm). A relative
error of 0.5% is obtained, showing that the aberrations
of a real skull can be mitigated using holographic lenses
even when the target acoustic field presents a complex
structure.
Second, the results for the self-bending beam simula-
tion inside a realistic skull are shown in Fig. 8. The sagit-
tal cross-section is shown in Fig. 8 (a), measured at y = 0
mm, together with the location of the target marked in
red dashed line. In this case the performance of the lens
to produce such a complex beam is reduced as compared
with the previous cases, as can be seen by the presence of
secondary lobes. This is mainly caused by the generation
of strong stationary waves between the skull bone and
the lens. However, the peak pressure follows the target
trajectory and the location of the peak pressure matches
the center of the curve. A clearer picture is given in
Fig. 8 (b), that shows the transversal field distribution
measured at z = 35 mm. Here, a sharp focal spot is visi-
ble and location of the peak pressure almost matches the
location of the target focus. A lateral shift of +0.22 mm,
that corresponds to one numerical grid step in the sim-
ulation was found in the x direction. The corresponding
transversal field distributions are shown in Fig. 8 (c) for
the x direction and in Fig. 8 (d) for the y direction, re-
spectively. Here, the reference calculations using theoret-
ical methods in a homogeneous medium (water) are also
FIG. 8. Simulation results for the self-bending holographic
beam designed for a realistic skull. (a) Axial cross-section of
the pressure distribution at y = 0 mm. (b) Transversal cross-
section at z = 35 mm. (c) Lateral cross-section at z = 35 mm
and y = 0 mm. (d) Lateral cross-section at z = 35 mm and
x = 2 mm.
-20020x (mm)-20-1001020y (mm)00.51-50050x (mm)00.20.40.60.81SimulationTheory-20020y (mm)00.20.40.60.81SimulationTheory50100150z (mm)00.20.40.60.81SimulationTheory-20020x (mm)020406080100120140z (mm)00.51(a)(c)(d)(e)(b)(c)(d)(c)(e)SimulationSimulationSkulllens(water)(water)(water)-30-20-100102030x (mm)-30-20-100102030y (mm)00.51-30-20-100102030x (mm)01020304050z (mm)00.51Skull-30-20-100102030y (mm)00.51SimulationTheory-30-20-100102030x (mm)00.51SimulationTheorytarget trajectoryLens aperturetargetLens(b)(a)(c)(d)SimulationSimulation(water)(water) 10
FIG. 9. Simulation results for the volumetric hologram designed for a realistic skull. (a) Axial cross-section of the pressure
distribution at y = 0 mm. (b) Transversal cross-section at z = 95 mm. (c) Lateral cross-section at z = 95 mm and y = 0 mm.
(d) Lateral cross-section at z = 95 mm and x = −26 = 0 mm. (e) Axial cross-section at x = −26 = 0 mm and y = 0 mm.
shown for comparison. The location of the focal spots
observed in both lateral directions for the simulations in-
cluding the realistic skull are in excellent agreement with
the corresponding focal spots in water. The width of the
focal spot obtained using both calculation methods also
is in agreement. The main discrepancy is the presence of
secondary lobes in the simulated field, presenting a peak
amplitude of 0.36 times the pressure at the focus. The
amplitude of these lobes is 1.7 times larger in direction
in which the beam is bent.
Third, a holographic lens was designed with the tar-
get of producing a volumetric hologram overlapping with
the right hippocampus volume, and in this case including
the acoustic properties of a realistic skull. The resulting
forward-simulated pressure field is shown in Figs. 9 (a-
e). First, the sagittal cross-section of the magnitude of
the pressure field at y = 0 mm is shown in Fig. 9 (a).
The produced field focuses around the target volume,
shown in dashed white lines. The beam is steered in the
direction of the right hippocampus while the transducer
axis remains normal to the skull surface. The transversal
cross-section at z = 95 mm is shown in Fig. 9 (b). While
the acoustic field is focused into the target volume, there
exists areas not covered by the beam, mainly in the outer-
most regions away from the transducer source. To quan-
tify the focusing performance, field cross-sections along
the corresponding dashed lines are given in Figs. 9 (c-e).
The lateral cross-section at z = 95 mm and y = 0 mm
is shown in Fig. 9 (c). As a comparison, we also plot
the corresponding cross-section of a holographic lens de-
signed to produce the same hologram in water using the
theoretical Rayleigh-Sommerfeld integration. The sim-
ulated beam using the realist skull and the theoretical
prediction in water mostly overlaps, being the energy of
the beam concentrated into the target volume. Small
side-lobes with an amplitude 5.3 times smaller than the
peak pressure are observed in the simulations including
the realistic skull. Note that the corresponding acoustic
intensity of the side lobes is about 30 times smaller than
the intensity at the focus. The volume of the beam, de-
fined as the total volume of the beam under a threshold
of 0.5 times the peak pressure, roughly overlaps with the
target volume. The overlapping volume between the tar-
get and the volumetric acoustic hologram is 29.7% for the
theoretical calculation in water and 21.4% for the simu-
lation including the skull. The transversal cross-section
along the y axis is given in Fig. 9 (d), where excellent
agreement is found between the two configurations. Fi-
nally, the axial cross-section along the z axis is shown
in Fig. 9 (e).
In this case, the field presents remark-
able side-lobes before and after the target region, but its
amplitude is lower than half of the maximum pressure.
Note this direction corresponds to the beam axis and the
pressure distribution of the corresponding focused beam
presents an elongated shape due to the limited aperture
of the source. In this case, a good agreement is also found
between the axial pressure distribution of the simulation
including the realistic skull and the theoretical prediction
in water.
VII. CONCLUSIONS
We have shown that using 3D printed acoustic holo-
grams it is possible to conform diffraction-limited ultra-
sonic fields of arbitrary shape compensating the aberra-
-20020y (mm)00.51SimulationTheory50100150z (mm)00.51SimulationTheorySimulation-50050x (mm)00.51SimulationTheory-40-2002040x (mm)-20020y (mm)00.51-40-2002040x (mm)050100150z (mm)00.51(a)(c)(e)(d)(b)(c)(d)(e)(c)TargetTargetSimulationSimulationSkull(water)(water)(water) tions of a the human skull. In particular, experimental
tests using a 3D printed skull phantom and numerical
simulations using a realistic skull were performed to ac-
curately generate multiple focal holograms, self-bending
beams and volumetric holographic fields overlapping a
target CNS structure. The proposed approach using
holographic lenses represents a step forward when com-
pared with the existing solutions using phase arrays, since
it opens new venues to develop reliable and cost reduced
ultrasonic applications.
The quality of the reconstructed acoustic images is re-
lated to the diffraction limit and the spatial bandwidth of
the holographic lens, which depends on the spatial aper-
ture of the lens, the number of pixels of the lens, and
the frequency of the beam [21]. In this work, we target
a human hippocampus using a single holographic lens of
only 50 mm aperture and operating at a frequency of 1.1
MHz. The reported experimental results inside a skull
phantom are in good agreement with theory and simula-
tions. Only small shifts, of the order of one wavelength
(1.4 mm in water), were found between the target lo-
cation and the field produced by the holographic lens.
These shifts can be caused by experimental reasons that
include small positioning error between the lens and the
curved surface of the phantom and can be corrected us-
ing optimization methods during lens design. It is worth
noting here that the phantom used in the experiments
presents a smaller acoustic impedance than a real skull.
However, full-wave simulations performed using the den-
sity, sound speed and attenuation values of skull-bone
show that these arbitrary fields can also be produced in
a realistic situation. The generated acoustic fields inside
the skull were in good agreement with those produced in
water. This shows that the aberrations produced by the
skull can be mitigated by using holographic lenses even
when a complex field is required.
Moreover, using the proposed methodology diffraction
is captured exactly as compared with Fraunhofer or an-
gular spectrum methods, leading to a better accuracy
of the generated acoustic fields.
In addition, the holo-
graphic lens design is based on resonating slabs, which
include not only the refraction over a curved plate, but
the resonating waves inside the lens.
Phased-arrays are efficient but their high-cost can be
prohibitive to spread out some of the incoming ultrasonic
transcranial therapy treatments. Using phased arrays
the ultrasonic beams can be adjusted in real time and
monitored using MRI, obtaining a precise location of the
acoustic focus into de CNS. Nevertheless, the number of
elements of the phased-array systems can be insufficient
to produce a complex volumetric ultrasonic field that
matches a specific CNS structure. The use of holographic
lenses present several advantages to produce complex vol-
umetric patterns. First, the cost of a 3D printed lens
is low as compared with phased-array systems. Second,
each pixel in a holographic lens acts as an element of the
phased-array with fixed phase. Due to the high number
of passive sources in a holographic lens, more than 4000
11
for the small lenses considered here, complex patterns
can be generated. However, once the lens is designed its
focal distance and the spatial features of the holographic
image remains fixed and, in principle, it is not possible
to steer the ultrasonic beam in real time with accuracy.
For this reason, the technique is specially relevant for the
treatment using a single sonification of structures or in
the sonification of large volumes.
The concept shown in this paper opens new doors to
optimize and widespread incoming therapy treatments
such as ultrasound-assisted blood-brain barrier opening
for drug delivery and neuromodulation, or ultrasonic
imaging of the central nervous system using low-cost de-
vices. Considering the emergence of metamaterials and
their huge flexibility, we also advance incoming biomed-
ical applications of active holographic metasurfaces for
the generation of complex fields in the central nervous
system.
ACKNOWLEDGMENTS
This work was supported by the Spanish Ministry of
Economy and Innovation (MINECO) through Project
TEC2016-80976-R. NJ and SJ acknowledge financial
support from Generalitat Valenciana through grants
APOSTD/2017/042, ACIF/2017/045 and GV/2018/11.
FC acknowledges financial support from Ag`encia Valen-
ciana de la Innovaci´o through grant INNCON00/18/9
and European Regional Development Fund (ID-
IFEDER/2018/022).
Appendix A: Measurement setup
The experiments were conducted inside a 1×0.75×0, 5
m water tank filled with degassed and distilled water at
26◦. The ultrasonic transducer was composed by a sin-
gle element circular piezoceramic crystal (PZT26, Fer-
roperm Piezoceramics, Denmark) mounted in a custom
designed stainless-steel housing with aperture 2a = 50
mm as shown in Fig. 10 (e). The transducer was driven
with a 50 cycles sinusoidal pulse burst at a frequency
of f = 1.112 MHz by a signal generator (14 bits, 100
MS/s, model PXI5412, National Instruments) and am-
plified by a linear RF amplifier (ENI 1040L, 400 W, 55
dB, ENI, Rochester, NY). The pressure field was mea-
sured by a needle hydrophone with a 500 µm active di-
ameter (149.6 mV/MPa sensitivity at 1.112 MHz, Model
HNR-500, Onda) calibrated from 1 MHz to 20 MHz. The
source amplitude was set low enough to avoid any non-
linear effects in the propagation, we measure 1.8 kPa at
the focus for the bifocal lens. The hydrophone signals
were digitized at a sampling rate of 64 MHz by a dig-
itizer (model PXI5620, National Instruments) averaged
100 times to increase the signal to noise ratio.
An 3D micro-positioning system (OWIS GmbH) was
used to move the hydrophone in three orthogonal direc-
12
tions with an accuracy of 10 µm. For the bifocal lens
experiment the scanning an area for the sagittal cross-
sections, p(x, z), covers from -40 mm to 40 mm in the
x direction and from 82 mm to 143 mm in the z direc-
tion, using a step of 0.5 mm in both directions, for the
transversal cross-section planes, p(x, y), covers from -40
to 40 mm in the x direction and from -10 mm to 10 mm
in the y direction, using a step of 0.3 mm. For the self-
bending lens, the scanned area covers from 20 to 50 mm
in the z direction, from -5 to 5 mm in the x direction, and
from -5 to 5 mm in the y direction, using the same spatial
steps. Finally, for the volumetric holograms the scanned
area covers from 65 to 135 mm in the z direction, from
-45 to -5 mm in the x direction, and from -10 to 10 mm in
the y direction, using the same spatial steps. All the sig-
nal generation and acquisition processes were based on a
NI8176 National Instruments PXI-Technology controller,
which also controlled the micro-positioning system. Tem-
perature measurements were performed throughout the
whole process to ensure no temperature changes of 0.5◦
C.
based on ultrathin planar acoustic metasurfaces. Sci-
entific Reports, 3:2546, 2013.
[12] Yangbo Xie, Wenqi Wang, Huanyang Chen, Adam Kon-
neker, Bogdan-Ioan Popa, and Steven A Cummer. Wave-
front modulation and subwavelength diffractive acoustics
with an acoustic metasurface. Nature Communications,
5:5553, 2014.
[13] No´e Jim´enez, Trevor J Cox, Vicent Romero-Garc´ıa,
Deep-
Scientific Reports,
and Jean-Philippe Groby.
subwavelength sound diffusers.
7(1):5389, 2017.
Metadiffusers:
[14] No´e Jim´enez, Vicent Romero-Garc´ıa, Vincent Pagneux,
and Jean-Philippe Groby. Rainbow-trapping absorbers:
Broadband, perfect and asymmetric sound absorption by
subwavelength panels for transmission problems. Scien-
tific Reports, 7(1):13595, 2017.
[15] Shuibao Qi, Yong Li, and Badreddine Assouar. Acous-
tic focusing and energy confinement based on multilat-
eral metasurfaces. Physical Review Applied, 7(5):054006,
2017.
[16] Eun Bok, Jong Jin Park, Haejin Choi, Chung Kyu Han,
Oliver B Wright, and Sam H Lee. Metasurface for
water-to-air sound transmission. Physical review letters,
120(4):044302, 2018.
[17] Yong Li, Xue Jiang, Bin Liang, Jian-chun Cheng, and
Likun Zhang. Metascreen-based acoustic passive phased
array. Physical Review Applied, 4(2):024003, 2015.
[18] Yong Li and M Badreddine Assouar. Three-dimensional
collimated self-accelerating beam through acoustic
metascreen. Scientific reports, 5:17612, 2015.
[19] Nad`ege Kaina, Fabrice Lemoult, Mathias Fink, and Ge-
offroy Lerosey. Negative refractive index and acoustic su-
perlens from multiple scattering in single negative meta-
materials. Nature, 525(7567):77, 2015.
[20] Jensen Li, Lee Fok, Xiaobo Yin, Guy Bartal, and Xi-
FIG. 10. Experimental setup showing the block diagram and
skull phantom inside the water tank with the ultrasonic source
and the acoustic hologram at the bottom.
[1] Dennis Gabor. A new microscopic principle. Nature,
161:777 -- 778, 1948.
[2] Dionys Gabor et al. Microscopy by reconstructed wave-
fronts. Proc. R. Soc. Lond. A, 197(1051):454 -- 487, 1949.
Recon-
structed wavefronts and communication theory. JOSA,
52(10):1123 -- 1130, 1962.
[3] Emmett N Leith and Juris Upatnieks.
[4] Xingjie Ni, Alexander V Kildishev, and Vladimir M Sha-
laev. Metasurface holograms for visible light. Nature
Communications, 4:2807, 2013.
[5] Lingling Huang, Xianzhong Chen, Holger Muhlenbernd,
Hao Zhang, Shumei Chen, Benfeng Bai, Qiaofeng Tan,
Guofan Jin, Kok-Wai Cheah, Cheng-Wei Qiu, et al.
Three-dimensional optical holography using a plasmonic
metasurface. Nature Communications, 4:2808, 2013.
[6] Guancong Ma and Ping Sheng. Acoustic metamateri-
als: From local resonances to broad horizons. Science
advances, 2(2):e1501595, 2016.
[7] Steven A Cummer, Johan Christensen, and Andrea Al`u.
Controlling sound with acoustic metamaterials. Nature
Reviews Materials, 1:16001, 2016.
[8] Zhengyou Liu, Xixiang Zhang, Yiwei Mao, YY Zhu,
Zhiyu Yang, Che Ting Chan, and Ping Sheng. Locally
resonant sonic materials. Science, 289(5485):1734 -- 1736,
2000.
[9] Nicholas Fang, Dongjuan Xi, Jianyi Xu, Muralidhar
Ambati, Werayut Srituravanich, Cheng Sun, and Xiang
Zhang. Ultrasonic metamaterials with negative modulus.
Nature Materials, 5(6):452, 2006.
[10] Min Yang, Guancong Ma, Zhiyu Yang, and Ping
Sheng. Coupled membranes with doubly negative mass
density and bulk modulus. Physical Review Letters,
110(13):134301, 2013.
[11] Yong Li, Bin Liang, Zhong-ming Gu, Xin-ye Zou, and
Jian-chun Cheng. Reflected wavefront manipulation
HologramTransducerHydrophoneSkull phantomzxy3D MicropositioningSystemDigitizerPXI5620PXI ControllerNI8176Function GeneratorPXI5412AmplifierENI 1040L ang Zhang. Experimental demonstration of an acoustic
magnifying hyperlens. Nature Materials, 8(12):931, 2009.
[21] Kai Melde, Andrew G Mark, Tian Qiu, and Peer Fischer.
Holograms for acoustics. Nature, 537(7621):518, 2016.
[22] Yangbo Xie, Chen Shen, Wenqi Wang, Junfei Li, Dingjie
Suo, Bogdan-Ioan Popa, Yun Jing, and Steven A
Cummer. Acoustic holographic rendering with two-
dimensional metamaterial-based passive phased array.
Scientific Reports, 6:35437, 2016.
[23] Yifan Zhu, Jie Hu, Xudong Fan, Jing Yang, Bin Liang,
Xuefeng Zhu, and Jianchun Cheng. Fine manipulation
of sound via lossy metamaterials with independent and
arbitrary reflection amplitude and phase. Nature com-
munications, 9(1):1632, 2018.
[24] Gianluca Memoli, Mihai Caleap, Michihiro Asakawa,
Deepak R Sahoo, Bruce W Drinkwater, and Sriram Sub-
ramanian. Metamaterial bricks and quantization of meta-
surfaces. Nature communications, 8:14608, 2017.
[25] Michael D Brown, Ben T Cox, and Bradley E Treeby.
Design of multi-frequency acoustic kinoforms. Applied
Physics Letters, 111(24):244101, 2017.
[26] Y Hertzberg and G Navon. Bypassing absorbing objects
in focused ultrasound using computer generated holo-
graphic technique. Medical Physics, 38(12):6407 -- 6415,
2011.
[27] Peng Zhang, Tongcang Li, Jie Zhu, Xuefeng Zhu, Sui
Yang, Yuan Wang, Xiaobo Yin, and Xiang Zhang. Gener-
ation of acoustic self-bending and bottle beams by phase
engineering. Nature Communications, 5:4316, 2014.
[28] Asier Marzo, Sue Ann Seah, Bruce W Drinkwater,
Deepak Ranjan Sahoo, Benjamin Long, and Sriram Sub-
ramanian. Holographic acoustic elements for manip-
ulation of levitated objects. Nature Communications,
6:8661, 2015.
[29] Mohd Adili Norasikin, Diego Martinez Plasencia, Spy-
ros Polychronopoulos, Gianluca Memoli, Yutaka Tokuda,
and Sriram Subramanian. Soundbender: dynamic acous-
tic control behind obstacles. In The 31st Annual ACM
Symposium on User Interface Software and Technology,
pages 247 -- 259. ACM, 2018.
[30] Gail ter Haar and Constantin Coussios. High intensity
focused ultrasound: physical principles and devices. In-
ternational journal of hyperthermia, 23(2):89 -- 104, 2007.
[31] Ayache Bouakaz and Jean-Michel Escoffre. Therapeutic
ultrasound: From biophysics concepts to clinical applica-
tions, volume 880 of Advances in Experimental Medicine
and Biology. Springer International Publishing, 2016.
[32] Thomas L Szabo. Diagnostic ultrasound imaging: inside
out. Academic Press, 2004.
[33] P G´elat, G Ter Haar, and N Saffari. A comparison of
methods for focusing the field of a hifu array transducer
through human ribs. Physics in Medicine & Biology,
59(12):3139, 2014.
[34] FJ Fry and JE Barger. Acoustical properties of the
human skull. The Journal of the Acoustical Society of
America, 63(5):1576 -- 1590, 1978.
[35] J-L Thomas and Mathias A Fink. Ultrasonic beam focus-
ing through tissue inhomogeneities with a time reversal
mirror: application to transskull therapy. IEEE Transac-
tions on Ultrasonics, Ferroelectrics, and Frequency Con-
trol, 43(6):1122 -- 1129, 1996.
[36] Kullervo Hynynen and Ferenc A Jolesz. Demonstration of
potential noninvasive ultrasound brain therapy through
an intact skull. Ultrasound in Medicine and Biology,
13
24(2):275 -- 283, 1998.
[37] Jie Sun and Kullervo Hynynen. Focusing of thera-
peutic ultrasound through a human skull: a numerical
study. The Journal of the Acoustical Society of America,
104(3):1705 -- 1715, 1998.
[38] J-F Aubry, M Tanter, M Pernot, J-L Thomas, and
M Fink. Experimental demonstration of noninvasive
transskull adaptive focusing based on prior computed to-
mography scans. The Journal of the Acoustical Society
of America, 113(1):84 -- 93, 2003.
[39] Mickael Tanter, Jean-Louis Thomas, and Mathias Fink.
Focusing and steering through absorbing and aberrating
layers: Application to ultrasonic propagation through the
skull. The Journal of the Acoustical Society of America,
103(5):2403 -- 2410, 1998.
[40] Y Hertzberg, A Volovick, Y Zur, Y Medan, S Vitek,
and G Navon. Ultrasound focusing using magnetic res-
onance acoustic radiation force imaging:
application
to ultrasound transcranial therapy. Medical Physics,
37(6Part1):2934 -- 2942, 2010.
[41] Ferenc A Jolesz. Intraoperative imaging and image-guided
therapy. Springer Science & Business Media, 2014.
[42] Chen Shen, Jun Xu, Nicholas X Fang, and Yun Jing.
Anisotropic complementary acoustic metamaterial for
canceling out aberrating layers. Physical Review X,
4(4):041033, 2014.
[43] Guillaume Maimbourg, Alexandre Houdouin, Thomas
Deffieux, Mickael Tanter, and Jean-Fran¸cois Aubry. 3d-
printed adaptive acoustic lens as a disruptive technol-
ogy for transcranial ultrasound therapy using single-
element transducers. Physics in Medicine & Biology,
63(2):025026, 2018.
[44] Marcelino Ferri, Jos M. Bravo, Javier Redondo, and
Juan V. Snchez-Prez. Enhanced numerical method for
the design of 3-d-printed holographic acoustic lenses for
aberration correction of single-element transcranial fo-
cused ultrasound. Ultrasound in Medicine & Biology,
45(3):867 -- 884, 2019.
[45] Kullervo Hynynen, Nathan McDannold, Natalia
Vykhodtseva, and Ferenc A Jolesz. Noninvasive mr
imaging -- guided focal opening of the blood-brain barrier
in rabbits. Radiology, 220(3):640 -- 646, 2001.
[46] William J Tyler, Yusuf Tufail, Michael Finsterwald,
Monica L Tauchmann, Emily J Olson, and Cassondra
Majestic. Remote excitation of neuronal circuits us-
ing low-intensity, low-frequency ultrasound. PloS one,
3(10):e3511, 2008.
[47] Uwe Schneider, Eros Pedroni, and Antony Lomax. The
calibration of ct hounsfield units for radiotherapy treat-
ment planning. Physics in Medicine & Biology, 41(1):111,
1996.
[48] T Douglas Mast. Empirical relationships between acous-
tic parameters in human soft tissues. Acoustics Research
Letters Online, 1(2):37 -- 42, 2000.
[49] John C Mazziotta, Arthur W Toga, Alan Evans, Peter
Fox, Jack Lancaster, et al. A probabilistic atlas of the
human brain: theory and rationale for its development.
Neuroimage, 2(2):89 -- 101, 1995.
[50] Paul A Yushkevich, Joseph Piven, Heather Cody Hazlett,
Rachel Gimpel Smith, Sean Ho, James C Gee, and Guido
Gerig. User-guided 3d active contour segmentation of
anatomical structures: significantly improved efficiency
and reliability. Neuroimage, 31(3):1116 -- 1128, 2006.
[51] David T. Blackstock. Fundamentals of physical acoustics.
John Wiley & Sons, 2000.
[52] Bradley E Treeby and BT Cox. Modeling power law
absorption and dispersion for acoustic propagation using
the fractional laplacian. The Journal of the Acoustical
Society of America, 127(5):2741 -- 2748, 2010.
[53] Bradley E Treeby, Jiri Jaros, Alistair P Rendell, and
BT Cox. Modeling nonlinear ultrasound propagation
in heterogeneous media with power law absorption us-
ing ak-space pseudospectral method. The Journal of the
Acoustical Society of America, 131(6):4324 -- 4336, 2012.
[54] No´e Jim´enez, Francisco Camarena, Javier Redondo,
V´ıctor S´anchez-Morcillo, Yi Hou, and Elisa E Konofagou.
Time-domain simulation of ultrasound propagation in a
tissue-like medium based on the resolution of the nonlin-
ear acoustic constitutive relations. Acta Acustica united
with Acustica, 102(5):876 -- 892, 2016.
[55] No´e Jim´enez, Vicent Romero-Garc´ıa, Vincent Pagneux,
and Jean-Philippe Groby. Quasiperfect absorption by
subwavelength acoustic panels in transmission using ac-
cumulation of resonances due to slow sound. Physical
Review B, 95(1):014205, 2017.
[56] Peter Wai Ming Tsang and T-C Poon. Novel method for
converting digital fresnel hologram to phase-only holo-
gram based on bidirectional error diffusion. Optics Ex-
press, 21(20):23680 -- 23686, 2013.
[57] Robert Lirette and Joel Mobley. Focal zone characteris-
tics of stepped fresnel and axicon acoustic lenses. Pro-
ceedings of Meetings on Acoustics, 31(1):045001, 2017.
14
[58] Matteo Gatto, Gianluca Memoli, Adam Shaw, Neelaksh
Sadhoo, Pierre Gelat, and Russell A Harris. Three-
dimensional printing (3dp) of neonatal head phantom for
ultrasound: Thermocouple embedding and simulation of
bone. Medical Engineering and Physics, 34(7):929 -- 937,
2012.
[59] James Robertson, Eleanor Martin, Ben Cox, and
Bradley E Treeby. Sensitivity of simulated transcra-
nial ultrasound fields to acoustic medium property maps.
Physics in Medicine & Biology, 62(7):2559, 2017.
[60] Chen Bai, Meiling Ji, Ayache Bouakaz, Yujin Zong, and
Mingxi Wan. Design and characterization of an acousti-
cally and structurally matched 3d-printed model for tran-
scranial ultrasound imaging. IEEE Transactions on Ul-
trasonics, Ferroelectrics, and Frequency Control, 2018.
[61] Christopher Rowland Hill,
and
GR Ter Haar. Physical principles of medical ultrasonics.
John Wiley & Sons New YorkChichesterBrisbane-
Toronto, 2004.
JC Bamber,
[62] Richard SC Cobbold. Foundations of biomedical ultra-
sound. Oxford University Press, 2006.
[63] HoT O'Neil. Theory of focusing radiators. The Journal of
the Acoustical Society of America, 21(5):516 -- 526, 1949.
[64] Di-Chao Chen, Xing-Feng Zhu, Qi Wei, Da-Jian Wu, and
Xiao-Jun Liu. Broadband acoustic focusing by airy-like
beams based on acoustic metasurfaces. Journal of Ap-
plied Physics, 123(4):044503, 2018.
|
1705.08254 | 1 | 1705 | 2017-05-16T05:20:40 | Origin of Operating Voltage Increase in InGaN-based Light-emitting Diodes under High Injection: Phase Space Filling Effect on Forward Voltage Characteristics | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | As an attempt to further elucidate the operating voltage increase in InGaN-based light-emitting diodes (LEDs), the radiative and nonradiative current components are separately analyzed in combination with the Shockley diode equation. Through the analyses, we have shown that the increase in operating voltage is caused by phase space filling effect in high injection. We have also shown that the classical Shockley diode equation is insufficient to comprehensively explain the I-V curve of the LED devices since the transport and recombination characteristics of respective current components are basically different. Hence, we have proposed a modified Shockley equation suitable for modern LED devices. Our analysis gives a new insight on the cause of the wall-plug-efficiency drop influenced by such factors as the efficiency droop and the high operating voltage in InGaN LEDs. | physics.app-ph | physics | Origin of Operating Voltage Increase in InGaN-based Light-
emitting Diodes under High Injection: Phase Space Filling Effect
on Forward Voltage Characteristics
Dong-Pyo Han1, 2, a), Jong-In Shim2, Dong-Soo Shin3
1Faculty of Science and Technology, Meijo University, 1-501 Shiogamaguchi, Tempaku-ku,
Nagoya 468-8502, Japan
2Dept. of Electronics and Communication Engineering, Hanyang University, ERICA Campus,
Ansan, Gyeonggi-do 426-791, Korea
3Dept. of Applied Physics and Dept. of Bionanotechnology, Hanyang University, ERICA
Campus, Ansan, Gyeonggi-do 426-791, Korea
Abstract
As an attempt to further elucidate the operating voltage increase in InGaN-based light-
emitting diodes (LEDs), the radiative and nonradiative current components are separately
analyzed in combination with the Shockley diode equation. Through the analyses, we have
shown that the increase in operating voltage is caused by phase space filling effect in high
injection. We have also shown that the classical Shockley diode equation is insufficient to
comprehensively explain the I-V curve of the LED devices since the transport and
recombination characteristics of respective current components are basically different. Hence,
we have proposed a modified Shockley equation suitable for modern LED devices. Our
analysis gives a new insight on the cause of the wall-plug-efficiency drop influenced by such
factors as the efficiency droop and the high operating voltage in InGaN LEDs.
a) E-mail: [email protected]
1
Recently, the InGaN-based light-emitting diodes (LEDs) covering the visible and
ultraviolet spectral regions have been hailed as one of the great inventions of the twentieth
century with the Nobel prize awarded to the three principle innovators, I. Akasaki, H. Amano,
and S. Nakamura.1 Owing to the progress, general lighting based on LEDs has become a
reality with the advantages of brightness, energy efficiency, and environment friendliness.2
Although the performances of InGaN-based LEDs have significantly improved so far,
higher light output power and lower electrical input power, namely higher wall-plug
efficiency (WPE) are still required for further replacement of conventional lighting source.
However, at high driving currents required for general lighting, the WPE of InGaN-based
LEDs is limited by well-known degradation mechanisms such as the efficiency droop and
high operating voltage.3-4 Many physical mechanisms have been proposed to explain the
origin of efficiency droop including the hot electrons to the vacuum level by Auger
recombination,5 carrier leakage caused by asymmetry of carrier concentration and mobility,6
carrier overflow assisted by the piezoelectric field,7 and phase space filling effect triggering
the carrier spill-over.8 These researches have pointed out that the carrier overflow or leakage
from the active region is responsible for the efficiency droop. On the other hand, the origin of
high operating voltage of InGaN-based LEDs is typically explained by high series resistance
caused by the high activation energy of acceptors in GaN,9 current crowding,10 and the
contact between metal and semiconductor of the LED device.11 Still, the physical origin of the
degradation mechanisms of WPE is not clarified yet.
It is generally known that the total driving current vs. applied voltage (I-V) curve of the
pn-junction semiconductor device is the most important electrical characteristics and is well
explained by the Shockley diode equation.12-13 The parameters in the Shockley equation such
as the ideality factor, the reverse saturation current, and the series resistance represent the
physical mechanisms including the carrier transport and recombination in the LED device. In
other words, we can infer the problem and remedy it from the information of extracted
parameters in the Shockley equation fitted with experimental results. The theory of the
Shockley equation was first developed for homojunction pn devices based on Si and Ge. In
particular, it is deeply related to the diffusion current component in the pn junction and is
therefore usually referred to as the "diffusion current theory".13 However, unlike in the
homojunction devices, various current transport and recombination mechanisms, not just the
2
diffusion, play important roles in heterojunction-based modern LED devices.14-16 Hence, the
Shockley diode equation is often insufficient to comprehensively explain the I-V curves of
modern LED devices employing multiple quantum wells (MQWs) as active layers.
In the light-generating LED devices, unlike other semiconductor devices such as
photodiodes and solar cells, the injected current can be separated into two components, the
radiative current,
, and the nonradiative current,
, while the Shockley diode equation
only explains the diffusion current. While there have been frequent investigations and
analyses of total driving current I vs. applied voltage V by the Shockley equation,16-18 the data
of
and
vs. V still requires a deeper understanding for the LED device.
In this letter, we aim to understand the origin of WPE degradation mechanisms in
InGaN-based LED, the characteristics of
and
as a function of V is precisely
investigated. To separate the current components, we utilize the information of the I-V curve
combined with the IQE characteristics of the LED device. We then carefully analyze each
current component as a function of V with the Shockley equation. Based on these analyses,
we give an explanation on the characteristics of
and
as a function of V including
physical mechanisms of carrier transport and recombination. We also show the close
interrelationship between efficiency droop and high operating voltage in InGaN-based LED
device.
For experiments, a commercial blue LED sample with InGaN MQWs was utilized. The
peak wavelength and the chip size of the sample were 450 nm and 600×1250 μm2,
respectively. The sample had been fabricated into a chip with lateral electrodes and mounted
on a surface-mounted-device (SMD) package without epoxy dome. The I-V characteristics
were measured by using a Keithley semiconductor parameter analyzer under the pulsed-
voltage condition. The light output power was collected by a Si photodiode under the pulsed-
current driving condition (pulse period: 100 μs, duty cycle: 1%) to avoid the self-heating
effect. The IQE and light-extraction efficiency (LEE) data were obtained by the conventional
temperature-dependent electroluminescence (TDEL) measurement, which is widely used in
various research works.19-21
Figures 1 (a) and (b) depict the IQE values of the sample under investigation as a
function of driving current on linear and semi-log scales at cryogenic and room temperature,
respectively. The peak value of IQE is continuously increased as decrease in operating
3
RINRIRINRIRINRIRINRI temperature from 300K to 25K. Thus, we can think that the measured IQE value at room
temperature is reliable. The IQE of the sample reaches a maximum value of ~84% at a low
current density of ~3 A/cm2 and decreases. The LEE is estimated as ~64% by using the
experimental data of the IQE and the external quantum efficiency (EQE). These are
considered as typical IQE, LEE, and EQE characteristics obtained from commercial InGaN-
based blue LEDs.20-22
Now, from the definition of the IQE,
and
can be separated from the total
current I as expressed in eqs. (1) and (2), respectively: 23
(1)
(2)
where
is the IQE.
Shown in Fig. 2 is the ideality factors
and
corresponding to
and
as a
function of V, obtained by differentiating
and
as expressed in eqs. (3) and (4),
respectively:24
(3)
(4)
where q is the elementary charge,
is the Boltzmann constant, and T is the absolute
temperature. It is seen that
remains at 1 in the low voltage < ~2.45 V and then increases
with increasing driving current. On the other hand,
remains at 2 in the similar low
voltage region and then increases with increasing driving current. The increase in ideality
factors at high voltages is caused by the additional potential drop unaccounted for in eqs. (3)
and (4). The ideality factor of
has been shown to be 1 for the direct band-to-band
recombination in MQWs of diffusion carrier.23 On the other hand, the ideality factor of 2 is
predicted to be resulted from two mechanisms, namely, the Sah-Noyce-Shockley (SNS)
current and the current under high-level injection condition.25 In our case,
of 2 is a result
from the SNS current since it is observed in the low-bias region. Thus,
is the result of
the nonradiative recombination via defects. The value of
greater than 2 in the lower
4
RINRIRIQEIINRIQEI1IIQERnNRnRINRIRINRIln1RRBIqnkTVln1NRNRBIqnkTVBkRnNRnRINRnNRINRn voltage region (< ~2.25 V) is considered to be caused by the tunneling mechanism resulting
from the threading dislocations introduced during the epitaxial growth, which seem to saturate
in the early stage of driving current.26
The additional potential drop can be obtained by comparing the experimental results
with the ideal diode behavior.27 To find the potential drop for each current component, we
show
and
vs. V plotted on semi-log scales in Fig. 3 (a) and (b), respectively. The
ideal diode behaviors for
and
, which are plotted as solid lines in Fig. 3 (a) and (b),
are expressed in eqs. (5) and (6):
(5)
(6)
where
(
) is the reverse saturation current corresponding to
(
). Note the
ideality factors of 1 for
and 2 for
are from the result in Fig.2. As mentioned before,
the discrepancy between the ideal curve and experimental data of
in the low-bias region
(< 2.25 V) is thought to be caused by the tunneling current.26 The additional potential drop
(
,
) between the ideal curves and the experimental results of
and
can
indicate the different current conduction mechanisms for respective current components.
In aspect of
, from the Shockley theory,13
vs.
characteristic in eq. (5) is
the current flowing under carrier recombination rate is greater than carrier diffusion
(injection) rate, namely,
is limited by carrier diffusion rate. The carrier diffusion rate
exponentially increase as a function of
, while radiative recombination rate is saturated due
to phase space filling effect under high injection.28,29 In this case, carrier injection rate exceed
the carrier recombination rate and therefore,
is limited by radiative recombination rate.
In rate equation model, the relation between
and carrier density can be written as,28
(7)
where
, B, n, and p are effective active volume, bimolecular radiative recombination
coefficient, electron and hole density in active region, respectively. B is divided by factor,
to include phase space filling effect in high current density.28,29 Under high-level
injection condition, n and p has similar quantity, i.e.,
. In low injection, i.e.,
<< 1,
5
RINRIRINRI,,expRidealSRBqVIIkT,,expNRidealSNRBqVII2kT,SRI,SNRIRINRIRINRINRIRVNRVRINRIRI,RidealIV,RidealIVRIRIReffIqVBnp1neffV1nnpn eq. (7) can be rewritten since n and p depend on quasi-Fermi level as following:23
(8)
where
is the intrinsic carrier density.
in eq. (8) seems to have identical voltage
dependence with eq. (5). Thus,
can be considered as diffusion limited current under low
current injection. However, under high injection, the
can be rewritten as following:
(9)
Eq. (9) implies that the current flowing is decreased due to phase space filling effect,
namely additional voltage drop caused by phase space filling effect (
) should be
considered to account for
vs. V curve. Since n is proportional to
13 and
composed of
and conventional ohmic potential drop, i.e.,
relation between
and
can be expressed as following:
is
, the
(10)
where
is the parameter corresponding to
. Thus, it seems that
is
significantly influenced by the degree of
, namely higher value of
lead to higher
operating voltage. It is note that the
is the parameter representing reduced recombination
probability of e-h pair in k-space due to difference in effective mass between hole and
electron as well as carrier accumulation in reduced effective volume caused by quantum-
confined stark effect (QCSE), asymmetry carrier distribution, and
local potential
fluctuation.28,30,31 In Fig. 4,
vs.
curve is plotted and data is fitted with eq.
(10). The extraction method of value of
is discussed in next section in detail. It is seems
that data is fitted very well with theoretical expectation, implying that the additional potential
drop
is described by
. Hence, we can conclude that phase space filling
effect lead to additional potential drop,
, which is comparable to potential drop caused
by constant series resistance under high injection.
In aspect of
, the non-radiative recombination rate in active region is saturated due
to such factors as the saturation of the carrier capture rate in defect32 and phase-space filling
effect on Auger recombination rate, i.e.,
under high injection currents.28, 29
Consequently, it is induced the condition of carrier injection rate exceeding the recombination
6
exp2ReffiBqVIqVBnkTinRIRIRIexp2ReffiBqV1IqVBnkT1nPSFVRIRIRVPSFVRPSFSVVIRPSFVRIexp'11PSFRV1I'RVexpPSFVRISRRVPSFSVIRPSFVNRI/3Cn1n rate in the active region. Under this condition, carriers begin to accumulate and transport to
retain charge neutrality as a form of drift current. The relation between current and voltage in
such drift current has been predicted as the Lampert-Rose law under the name of so-called
"double-injection current" in semiconductors:33,34
,
(11)
where
(
) is the equilibrium electron (hole) concentration,
is the average lifetime of
injected carriers,
(
) is the electrons (holes) mobility, and L is the anode-cathode
spacing. In particular, double-injection current has quadratic dependence on the applied
voltage. Thus, we can think that
is composed of conventional ohmic potential drop
and potential drop by double injection current, i.e.,
. In Fig. 5, the
experimental data of
is plotted with
and data is fitted with
. It is
seems that data is fitted very well with theoretical expectation, implying that the additional
potential drop
can be described by
. In other words, the
has a
quadratic dependence on the
and
is limited by the double injection current,
which appears as an additional potential drop in the LED device. In the case of InGaN-based
LEDs, it is frequently pointed out that carrier spill-over and overflow from the active region
initiated by such factor as the saturation of the recombination rate caused by phase-space
filling in the small effective active volume as a form of drift current is as origin of efficiency
droop.6,16,35 We believe that this double-injection current behavior can be considered as
another indication that the carrier overflow and spill-over from the active region does occur
under high injection. Therefore, we can conclude that the low degree of carrier recombination
rate due to phase space filling compared to injection rate can lead to both the efficiency droop
and the high operating voltage in InGaN-based LED devices.
Now, we can represent
and
by using the above information in the Shockley
equations as expressed below:
,
(12)
,
(13)
7
200np39VJqnp8L0n0pnpNRVSNRDIIRIDNRVNRISNRDIIRIDNRVNRDISVVIRNRINRVNRIRINRI,ln'expSRRSRBqVIR1IIIkT,expSNRDINRSNRBqVIRIDII2kT
Note that the unit of
and
are
while that of
is
A0.5. Figure 6 shows
and
data fitted by eqs. (12) and (13). It is seen that the modified Shockley equation [eq.
(12) and (13)] fits the experimental data perfectly well for both current components. The
fitting parameters are
=1.3×10-45A,
=2.3×10-24A,
=2.6
,
=4.9
, and
=1.8 A0.5, respectively. The inset in Fig. 6 is the schematic illustration of the equivalent
circuit. The equivalent circuit basically consists of an ideal diode and a resistor. As shown in
equivalent circuit, additional potential drop for each current component such as phase space
filling and double injection current should be considered to account for I-V curve in InGaN-
based LEDs.
In conclusion, to clarify the origin of high operating voltage in InGaN-based LED, we
separately analyze the
and
in combination with the Shockley diode equation,
carefully. Through the analyses, we have shown that the increase in operating voltage is
caused by phase space filling effect in high injection. We have also shown that the classical
Shockley diode equation is insufficient to comprehensively explain the I-V curve of the LED
devices since
the
transport and recombination characteristics of respective current
components are basically different. Hence, we have proposed a modified Shockley equation
suitable for modern LED devices. In particular, it is considered that the low degree of carrier
recombination rate can lead to both the efficiency droop and the high operating voltage in
InGaN-based LED devices. To remedy this, the radiative recombination rate influenced by
reduced effective volume, QCSE, asymmetry carrier distribution in MQWs, and local
potential fluctuation should be increased further, which will enhance the WPE in InGaN-
based LEDs. We believe that characteristics of
and
pointed out in this letter give a
new insight into the electrical and optical characteristics of LED devices.
8
SR'DIDRINRI,SRI,SNRISR'DIDRINRIRINRI References
1. http://www.nobelprize.org/nobel_prizes/physics/laureates/2014/press.html
2. E. F. Schubert and J. K. Kim, Science 308, 1274 (2005).
3. G. Verzellesi, D. Saguatti, M. Meneghini, F. Bertazzi, M. Goano, G. Meneghesso, and E.
Zanoni, J. Appl. Phys. 114, 071101 (2013).
4. D.-P. Han, J.-I. Shim, D.-S. Shin, and K.-S. Kim, Appl. Phys. Express 9, 081002 (2016).
5. J. Iveland, L. Martinelli, J. Peretti, J. S. Speck, and C. Weisbuch, Phys. Rev. Lett. 110,
177406 (2013).
6. D. S. Meyaard, G.-B. Lin, J. Cho, E. F. Schubert, H. Shim, S.-H. Han, M.-H. Kim, C.
Sone, Y. S. Kim, Appl. Phys. Lett. 102, 251114 (2013).
7. M.-H. Kim, M. F. Schubert, Q. Dai, J. K. Kim, E. F. Schubert, J. Piprek, and Y. Park,
Appl. Phys. Lett. 91, 183507 (2007).
8. D.-P. Han, M.-G. Kang, C.-H. Oh, H. Kim, K.-S. Kim, D.-S. Shin, and J.-I. Shim,
Phys. Status Solidi A, 210, 2204 (2013).
9. D. S. Meyaard, G.-B. Lin, Q. Shan, J. Cho, E. F. Schubert, H. Shim, M.-H. Kim, and C.
Sone, Appl. Phys. Lett. 99, 251115 (2011).
10. V. K. Malyutenko, S. S. Bolgov, and A. D. Podoltsev, Appl. Phys. Lett. 97, 251110
(2010).
11. R.-H. Horng, D.-S. Wuu, Y.-C. Lien, and W.-H. Lan, Appl. Phys. Lett. 79, 2925 (2001).
12. W. Shockley, The Bell System Technical Journal 28, 435 (1949).
13. S. M. Sze and Kwok K. Ng, Physics of Semiconductor Device, 3rd edition (Wiley-
Interscience, New Jersey, 2006), Ch. 2.
14. Y. C. Shen, G. O. Mueller, S. Watanabe, N. F. Gardner, A. Munkholm, and M. R.
Krames, Appl. Phys. Lett., 91, 141101 (2007).
15. G.-B. Lin, D. Meyaard, J. Cho, E. F. Schubert, H. Shim, and C Sone, Appl. Phys. Lett.,
100, 161106 (2012).
16. D.-P. Han, H. Kim, J.-I. Shim, D.-S. Shin, and K.-S. Kim, Appl. Phys. Lett., 105,
191114 (2014).
17. E. F. Schubert, Light-Emitting Diode, 2nd ed. (Cambridge University Press, New York,
2006), Chapter. 4.
9
18. S. W. Lee, D. C. Oh, H. Goto, J. S. Ha, H. J. Lee, T. Hanada, M. W. Cho, T. Yao, S. K.
Hong, H. Y. Lee,S. R. Cho, J. W. Choi, J. H. Choi, J. H. Jang, J. E. Shin, and J. S. Lee,
Appl. Phys. Lett. 89, 132117 (2006).
19. J. Hader, J. V. Moloney, and S. W. Koch, Appl. Phys. Lett. 99, 181127 (2011).
20. G.-B. Lin, Q. Shan, Y. Wang, T. Li, and E. F. Schubert, Appl. Phys. Lett. 105, 221116
(2014).
21. I. E. Titkov, S. Y. Karpov, A. Yadav, V. L. Zerova, M. Zulonas, B. Galler, M. Strassburg,
I. Pietzonka, H.-J. Lugauer, and E. U. Rafailov, IEEE, J. Quantum Electron., 50, 911
(2014).
22. T.-Y. Seong, J. Han, H. Amano, and H. Morkoç, III-Nitride Based Light Emitting
Diodes and Applications (Springer, Heindelberg, 2013) Chapter. 9-10.
23. G. W. Lee, J.-I. Shim, and D.-S. Shin, Appl. Phys. Lett. 109, 031104 (2016).
24. D.-P. Han, C.-H. Oh, D.-G. Zheng, H. Kim, J.-I. Shim, K.-S. Kim, and D.-S. Shin, Jpn. J.
Appl. Phys. 54, 02BA01 (2015).
25. C.-T. Sah, Fundamentals of Solid State Electronics, (World Scientific, London, 1991),
Chapter. 5.
26. D.-P. Han, C.-H. Oh, H. Kim, J.-I. Shim, K.-S. Kim, and D.-S. Shin, IEEE Trans.
Electron Dev. 62, 587 (2015).
27. R. F. Pierret, Semiconductor device fundamentals (Addison-Wesley publishing,
Massachusetts, 1996), Chapter. 6.
28. H.-Y. Ryu, D.-S. Shin, and J.-I. Shim, Appl. Phys. Lett. 100, 131109 (2012).
29. A. David, and M. J. Grundmann, Appl. Phys. Lett. 96, 103504 (2010).
30. J.-I. Shim, H. Kim, D.-S. Shin, and H.-Y. Ryu, J. Korean Phys. Soc. 58, 503 (2011).
31. R. Olshansky, C. B. Su, J. Manning, and W. Powazinik, IEEE, J. Quantum Electron., 20,
838 (1984).
32. R. K. Ahrenkiel, B. M. Keyes, and D. J. Dunlavy, J. Appl. Phys. 70, 225 (1991).
33. M. A. Lampert and A. Rose, Phys. Rev. 121, 26 (1961).
34. M. A. Lampert and P. Mark, Current injection in solids (Academic press, New York,
1970), Chapter. 12.
35. I. Prudaev, O. Tolbanov, and S. Khludkov, Phys. Status Solidi A 212, 930 (2015).
10
Figures Captions
FIG. 1. IQE characteristics of the InGaN LED sample under investigation plotted with (a)
linear and (b) semi-log scale at cryogenic and room temperature, respectively.
FIG. 2. Ideality factors corresponding to
and
as a function of applied voltage.
FIG. 3. (a)
and (b)
as a function of applied voltage with ideal diode curves
(
,
) and additional potential drops (
,
), respectively.
FIG. 4. (a)
as a function of
and fitting curve of theoretical approach.
FIG. 5.
as a function of
and fitting curve of theoretical approach.
FIG. 6.
and
vs. V fitted by the modified Shockley equations for an InGaN LED
sample under investigation. . The inset shows the equivalent circuit of an InGaN-based
LED.
11
RINRIRINRI,RidealI,NRidealIRVNRVPSFVeRINRVNRIRINRI
Fig. 1
12
Fig. 2
13
Fig. 3
14
Fig. 4
15
Fig. 5
16
Fig. 6
17
|
1705.07719 | 1 | 1705 | 2017-05-07T19:40:05 | Tunable Optical Bistability and Optical Switching by Nonlinear Metamaterials | [
"physics.app-ph"
] | We demonstrate a nonlinear metamaterial in microwave frequency regime with hysteresis effect and bistable states, which can be utilized as a remotely controllable micro second switching device. A varactor loaded split-ring resonator (SRR) design which exhibits power and frequency dependent broadband tunability of the resonance frequency for an external control signal is used. More importantly, the SRR shows bistability with distinct transmission levels. The transition between bi-states is controlled by impulses of an external pump signal. Furthermore, we experimentally demonstrate that transition rate is in the order of microseconds by using a varactor loaded double split-ring resonator (DSRR) design composed of two concentric rings. | physics.app-ph | physics | Tunable Optical Bistability and Optical Switching by Nonlinear
Metamaterials
Sinhara R Silva, Alexander D Shields, Jiangfeng Zhou
University of South Florida
Abstract
We demonstrate a nonlinear metamaterial in microwave frequency regime with hysteresis effect
and bistable states, which can be utilized as a remotely controllable micro second switching device.
A varactor loaded split-ring resonator (SRR) design which exhibits power and frequency
dependent broadband tunability of the resonance frequency for an external control signal is used.
More importantly, the SRR shows bistability with distinct transmission levels. The transition
between bi-states is controlled by impulses of an external pump signal. Furthermore, we
experimentally demonstrate that transition rate is in the order of microseconds by using a varactor
loaded double split-ring resonator (DSRR) design composed of two concentric rings.
Introduction
Electromagnetic Metamaterials are artificial structures periodically arranged to exhibit fascinating
electromagnetic properties, not existing in nature [1]. One example is a synthetic homogeneous
media that consists of element structures termed meta-atoms which are engineered to have both a
negative permittivity and a negative permeability [2-4]. These structures introduce peculiar
behaviors such as negative index of refraction [5, 6], super-resolution or perfect lensing [7], and
electromagnetic cloaking [8, 9].
A great deal of research in the field of metamaterial was conducted in a linear regime, where the
electromagnetic responses are independent of the external electric or magnetic fields.
Unfortunately, in the linear regime the desired properties of metamaterials have only been
achieved within a narrow bandwidth around a fixed frequency. There are a few ways that passive
tunability in metamaterials can be achieved. Some methods include geometric modification,
insertion of variable resistors, and insertion of variable capacitors. The introduction of additional
controllable media [14-16] such as nonlinear elements to the structure has led to extensive studies
on nonlinear metamaterials [16-18]. In turn, nonlinear metamaterials have given way to broadband
resonance responses in metamaterials. Amongst various nonlinear elements, introducing a varactor
diode to the structure has been the most preferred and more effective method for obtaining
broadband responses in the microwave and optical regime. It has been demonstrated that applying
a DC bias voltage [19-21], externally applied electromagnetic fields [15, 22, 23], and even heat
[10, 24] can be used to achieve non-linearity and tunability in varactor loaded metamaterials.
Among basic metamaterial structures split ring resonators are known as the most common and best
characterized [15, 25] metamaterial with geometrically scalable meta-atoms which can be
translated to operability in many decades of frequencies. In this paper, we present a nonlinear
tunable metamaterial operating at microwave frequencies. We demonstrate that an external pump
signal can be used to remotely tune the SRR resonance frequency. Furthermore, we investigate the
bistability and hysteresis behavior of the SRR under treatment of an external pump. We also show
that the external pump can be used to remotely trigger the SRR between bistable resonance states.
The SRR sample has nonlinear behavior that was achieved by mounting a hyper-abrupt tuning
varactor diode [26, 27] as a bridge for one of the gaps in the SRR.
Design and the experimental set up
The resonator structures are fabricated using copper on a FR4 substrate of thickness 0.2mm. A
PNA-X Network Analyzer is used to measure the S-parameters of the SRR sample while a
N5181A MXG Analog Signal Generator is used as the pump signal. The experiments were
performed with both the pump and the probe signals in such a way as to have the electric field
polarization parallel to the vacant gap of the SRR structure.
Figure 1: Experimental design of the SRR structure
a) Experimental configuration of the metamaterial behavior as a switching device b) Geometry
and dimensions of the SRR structure. H=40mm, h=20mm, G=0.5mm, W=1mm, where (D) is a
Hyper abrupt varactor diode c) Surface current density obtained by finite element simulations
using CST.
Figure 1(a) illustrates the SRR being controlled by external signals and being used as a switchable
device in the microwave frequency regime. The SRR structure was exposed to an external signal
for both increasing and decreasing pump frequency with a constant power as well as both
increasing and decreasing pump power at constant frequencies. The behavior of the SRR's
transmission was observed using the PNA-X Network Analyzer.
Hysteresis and Bistability
Using the PNA-X Network Analyzer, S-parameters of the sample are measured under varying
conditions. To clearly show bistability as well as a hysteresis effect, measurements of the SRR
under increasing pump frequency and decreasing pump frequency were compared as shown in fig
2.
Figure 2. Resonance frequency dependence for continuously increasing (red) and decreasing
(blue) external signal frequency with constant power Ppump=-25dBm
a) Hysteresis behavior of the frequency dependent resonance frequency b), c), d), Frequency
dependant transmitted intensity for pump frequency fpump = 782MHz(region 1), fpump =
796MHz(region 2) , fpump = 809MHz(region 3), respectively e), f), frequency dependent 2D
transmitted intensity behavior.
The Varactor loaded SRR structure displays a dynamic tuning range of 60MHz for an external
pump signal of a constant power of -25dBm. As shown in Figure 2(a), the width of the hysteresis
loop is more than 20MHz for that particular intensity. Figures 2(b), 2(c), and 2(d) display the
transmission behavior for the corresponding regions indicated in figure 2(a). In Regions 1 and 3
the frequency dependent transmitted intensity for decreasing and increasing pump frequencies are
similar. But in region 2, it follows two different paths demonstrating bi-stability of the
metamaterial structure. We observed the sudden jump between two resonance frequencies when
the external pump frequency coincides with the resonance frequency of the structure.
The hysteresis-type behavior and bistability[26] of the resonance frequency can be seen not only
for the external signal frequency as shown in figure 3(a), but also for external signal intensities
with a constant frequency as shown in figure 3(b).
Figure 3. Resonance frequency dependence for external signal
Resonance frequency dependence for continuously increasing and decreasing a) Pump frequency
b) Pump power c) Resonance frequency dependence for continuously and periodically changing
pump frequency at Ppump = -25dBm d) resonance frequency dependence with the angle of the E
filed (Black : E field along the gap of the structure, Green : E field is perpendicular to the structure)
For very low pump powers and frequencies, hysteresis behavior vanishes and we are no longer
able to observe the jump between two states. In addition, the width of the hysteresis loop widens
and narrows with the increase and decrease of both the external pump signal power and frequency
indicating an increase in nonlinearity. Furthermore, the direction of the sudden jump in the
resonance frequency is different for varying power and varying frequency. When increasing pump
power, the resonance frequency jumps from a lower frequency to a higher frequency while the
opposite is true for increasing the pump frequency. We found success when modeling our structure
as an RLC circuit. With the increase of the pump power, the voltage across the diode increases,
which results in a decrease in the total capacitance of the structure and an increase in the resonance
frequency. But, sudden transition of the resonance frequency for external frequency dependence,
behaves opposite to external signal intensity dependence. Furthermore, in order to observe the
hysteresis effect external pump should be applied continuously as shown in fig 3c otherwise jump
is observed but not hysteresis effect. Polarization angle of the E filed with respect to the gap of the
SRR has an impact on the behavior of the structure as shown in fig 3d. When the E filed is parallel
to the gap of the structure we observe the bistable states but with the change of the angle for the
same power level the magnitude of the e filed component along the gap reduces and shrink the
hysteresis gap size.
Switching device and transition time
The SRR structure is improved by inserting a 1M resistor in parallel to the nonlinear varactor
diode. This modification of the structure decreases the transition time between the states and
improves its applicability, but minimum step size of the time was limited by the delay time of the
instruments used in the experiment. Furthermore, by using a pulse train of appropriate powers, the
hysteresis in figure 4(d) was observed which clearly indicated two states in the transmission.
Figure 4. Trigger transitions between the two states
a) Transition between states for power dependent transmitted intensity b) S parameter plots of the
two states c) Trigger signal d) Bistable states of transmission signal
To explore the transition of the sample under a greater resolution, different measurement
techniques must be utilized. In order to capture the transition, we designed a double split ring
resonator (DSRR) that could be pumped at the resonant frequency (1200MHz) of inner ring and
probed at the frequency of outer ring. The power of the pump signal is about 10 dBm greater in
magnitude than the signal used to probe. Figure 5a shows the new design which was used to capture
the transition times in microsecond speed.
Figure 5. a) DSRR structure and S parameter plot b) transition between two states c)
spectrogram of a transition d) Trigger signal of from the pump e) and f) transition time
dependence with the resistance
As shown in fig 5b by inserting a resistor with parallel to the varactor diode of the structure we
can increase the transition time by allowing a path to discharge the accumulated charge across the
diode. We experimentally demonstrate that by choosing a particular resistor value jump up
transition and jump down transition time are different (Fig 5b) and the transition time depends on
the resistor value (fig 5e and fig 5f). In order to make a switching device with a switching speed
in microseconds we need to choose a correct resistor which optimizes the both up and down
transition times.
Currently, metamaterials with frequency tunability are particular of interest due to the flexibility
of frequency control and has become an essential part of metamaterial devices. It makes material
devices more versatile, adaptive to varying external disturbances and changes its effective
parameters accordingly. Nonlinear metamaterials provide the basis for manipulating and
controlling of the characteristics of a media. Therefore, the functionality demonstrated above will
benefit to better understanding of the nonlinear behavior of the metamaterials. Furthermore,
difference between power dependent and frequency dependent hysteresis behavior, its increased
tunability and usefulness of bistable states as a switchable device in microwave. Finally,
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
V. G. Veselago, "Electrodynamics of Substances with Simultaneously Negative Values of Sigma and
Mu," Soviet Physics Uspekhi‐Ussr, vol. 10, pp. 509‐&, 1968.
D. R. Smith, S. Schultz, P. Markos, and C. M. Soukoulis, "Determination of effective permittivity
and permeability of metamaterials from reflection and transmission coefficients," Physical Review
B, vol. 65, May 15 2002.
J. B. Pendry, A. J. Holden, W. J. Stewart, and I. Youngs, "Extremely low frequency plasmons in
metallic mesostructures," Physical Review Letters, vol. 76, pp. 4773‐4776, Jun 17 1996.
J. B. Pendry, A. J. Holden, D. J. Robbins, and W. J. Stewart, "Magnetism from conductors and
enhanced nonlinear phenomena," Ieee Transactions on Microwave Theory and Techniques, vol.
47, pp. 2075‐2084, Nov 1999.
R. A. Shelby, D. R. Smith, and S. Schultz, "Experimental verification of a negative index of
refraction," Science, vol. 292, pp. 77‐79, Apr 6 2001.
D. R. Smith, W. J. Padilla, D. C. Vier, S. C. Nemat‐Nasser, and S. Schultz, "Composite medium with
simultaneously negative permeability and permittivity," Physical Review Letters, vol. 84, pp. 4184‐
4187, May 1 2000.
J. B. Pendry, "Negative refraction makes a perfect lens," Physical Review Letters, vol. 85, pp. 3966‐
3969, Oct 30 2000.
J. B. Pendry, D. Schurig, and D. R. Smith, "Controlling electromagnetic fields," Science, vol. 312,
pp. 1780‐1782, Jun 23 2006.
U. Leonhardt, "Optical conformal mapping," Science, vol. 312, pp. 1777‐1780, Jun 23 2006.
J. Han and A. Lakhtakia, "Semiconductor split‐ring resonators for thermally tunable terahertz
metamaterials," Journal of Modern Optics, vol. 56, pp. 554‐557, 2009.
Y. X. He, P. He, S. D. Yoon, P. V. Parimi, F. J. Rachford, V. G. Harris, et al., "Tunable negative index
metamaterial using yttrium iron garnet," Journal of Magnetism and Magnetic Materials, vol. 313,
pp. 187‐191, Jun 2007.
L. Kang, Q. Zhao, H. J. Zhao, and J. Zhou, "Ferrite‐based magnetically tunable left‐handed
metamaterial composed of SRRs and wires," Optics Express, vol. 16, pp. 17269‐17275, Oct 27
2008.
H. J. Zhao, L. Kang, J. Zhou, Q. Zhao, L. Y. Li, L. Peng, et al., "Experimental demonstration of tunable
negative phase velocity and negative refraction in a ferromagnetic/ferroelectric composite
metamaterial," Applied Physics Letters, vol. 93, Nov 17 2008.
K. Aydin and E. Ozbay, "Capacitor‐loaded split ring resonators as tunable metamaterial
components," Journal of Applied Physics, vol. 101, Jan 15 2007.
J. Carbonell, V. E. Boria, and D. Lippens, "Nonlinear effects in split ring resonators loaded with
heterostructure barrier varactors," Microwave and Optical Technology Letters, vol. 50, pp. 474‐
479, Feb 2008.
[16] M. Lapine, M. Gorkunov, and K. H. Ringhofer, "Nonlinearity of a metamaterial arising from diode
[15]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
insertions into resonant conductive elements," Physical Review E, vol. 67, Jun 2003.
A. A. Zharov, I. V. Shadrivov, and Y. S. Kivshar, "Nonlinear properties of left‐handed
metamaterials," Physical Review Letters, vol. 91, Jul 18 2003.
H. T. Chen, W. J. Padilla, J. M. O. Zide, A. C. Gossard, A. J. Taylor, and R. D. Averitt, "Active terahertz
metamaterial devices," Nature, vol. 444, pp. 597‐600, Nov 30 2006.
I. V. Shadrivov, S. K. Morrison, and Y. S. Kivshar, "Tunable split‐ring resonators for nonlinear
negative‐index metamaterials," Optics Express, vol. 14, pp. 9344‐9349, Oct 2 2006.
D. X. Wang, L. X. Ran, H. S. Chen, M. K. Mu, J. A. Kong, and B. I. Wu, "Active left‐handed material
collaborated with microwave varactors," Applied Physics Letters, vol. 91, Oct 15 2007.
I. Gil, F. Martin, X. Rottenberg, and W. De Raedt, "Tunable stop‐band filter at Q‐band based on
RF‐MEMS metamaterials," Electronics Letters, vol. 43, pp. 1153‐1154, Oct 8 2007.
D. A. Powell, I. V. Shadrivov, Y. S. Kivshar, and M. V. Gorkunov, "Self‐tuning mechanisms of
nonlinear split‐ring resonators," Applied Physics Letters, vol. 91, Oct 1 2007.
A. Degiron, J. J. Mock, and D. R. Smith, "Modulating and tuning the response of metamaterials at
the unit cell level," Optics Express, vol. 15, pp. 1115‐1127, Feb 5 2007.
T. Driscoll, S. Palit, M. M. Qazilbash, M. Brehm, F. Keilmann, B. G. Chae, et al., "Dynamic tuning of
an infrared hybrid‐metamaterial resonance using vanadium dioxide," Applied Physics Letters, vol.
93, Jul 14 2008.
K. Aydin, I. Bulu, K. Guven, M. Kafesaki, C. M. Soukoulis, and E. Ozbay, "Investigation of magnetic
resonances for different split‐ring resonator parameters and designs," New Journal of Physics, vol.
7, Aug 8 2005.
B. N. Wang, J. F. Zhou, T. Koschny, and C. M. Soukoulis, "Nonlinear properties of split‐ring
resonators," Optics Express, vol. 16, pp. 16058‐16063, Sep 29 2008.
E. Poutrina, D. Huang, and D. R. Smith, "Analysis of nonlinear electromagnetic metamaterials,"
New Journal of Physics, vol. 12, Sep 6 2010.
|
1712.09137 | 1 | 1712 | 2017-12-25T22:25:06 | Probing frontier orbital energies of (Co9(P2W15)3) polyoxometalate clusters at molecule-metal and molecule-water interfaces | [
"physics.app-ph",
"physics.chem-ph"
] | Functionalization of polyoxotungstates with organoarsonate co-ligands enabling surface decoration was explored for the triangular cluster architectures of the composition [CoII9(H2O)6(OH)3(p-RC6H4AsVO3)2({\alpha}-PV2WVI15O56)3]25-({Co9(P2W15)3}, R = H or NH2), isolated as Na25[Co9(OH)3(H2O)6(C6H5AsO3)2(P2W15O56)3]86H2O (Na-1) and Na25[Co9(OH)3(H2O)6(H2NC6H4AsO3)2(P2W15O56)3]86H2O (Na-2). The axially oriented para-aminophenyl groups in 2 facilitate the formation of self-assembled monolayers on gold surfaces, and thus provide a viable molecular platform for charge transport studies of magnetically functionalized polyoxometalates. The title systems were isolated and characterized in the solid state and in aqueous solutions, and on metal surfaces. Using conducting tip atomic force microscopy (C-AFM), the energies of {Co9(P2W15)3} frontier molecular orbitals in the surface-bound state were found to directly correlate with cyclic voltammetry data in aqueous solution. | physics.app-ph | physics | Probing
frontier
orbital
energies
of
{Co9(P2W15)3}
polyoxometalate
clusters
at
molecule–metal
and
molecule–water
interfaces
Xiaofeng
Yi,†,‡
Natalya
V.
Izarova,†
Maria
Stuckart,†,‡
David
Guérin,§
Louis
Thomas,§
Stéphane
Lenfant,§
Dominique
Vuillaume,§,*
Jan
van
Leusen,‡
Tomáš
Duchoň,
Slavomír
Nemšák,†,¶
Svenja
D.
M.
Bourone,‡
Sebastian
Schmitz‡
and
Paul
Kögerler†,‡,*
†
Jülich--‐Aachen
Research
Alliance
(JARA--‐FIT)
and
Peter
Grünberg
Institute
6,
Forschungszentrum
Jülich,
D--‐52425
Jülich,
Germany
‡
Institute
of
Inorganic
Chemistry,
RWTH
Aachen
University,
D--‐52074
Aachen,
Germany
§
Institute
of
Electronics,
Microelectronics
and
Nanotechnology,
CNRS,
University
of
Lille,
59652
Villeneuve
d'Ascq,
France
Faculty
of
Mathematics
and
Physics,
Charles
University,
18000
Prague,
Czech
Republic
¶
BESSY--‐II,
Helmholtz--‐Zentrum
Berlin,
D--‐12489
Berlin,
Germany
ABSTRACT:
Functionalization
of
polyoxotungstates
with
organoarsonate
co--‐ligands
enabling
surface
decoration
was
ex--‐
plored
for
the
triangular
cluster
architectures
of
the
composition
[CoII9(H2O)6(OH)3(p--‐RC6H4AsVO3)2(α--‐PV2WVI15O56)3]25–
({Co9(P2W15)3},
R
=
H
or
NH2),
isolated
as
Na25[Co9(OH)3(H2O)6(C6H5AsO3)2(P2W15O56)3]·∙86H2O
(Na--‐1;
triclinic,
P−1,
a
=
25.8088(3)
Å,
b
=
25.8336(3)
Å,
c
=
27.1598(3)
Å,
α
=
78.1282(11)°,
β
=
61.7276(14)°,
γ
=
60.6220(14)°,
V
=
13888.9(3)
Å,
Z
=
2)
and
Na25[Co9(OH)3(H2O)6(H2NC6H4AsO3)2(P2W15O56)3]·∙86H2O
(Na--‐2;
triclinic,
P−1,
a
=
14.2262(2)
Å,
b
=
24.8597(4)
Å,
c
=
37.9388(4)
Å,
α
=
81.9672(10)°,
β
=
87.8161(10)°,
γ
=
76.5409(12)°,
V
=
12920.6(3)
Å,
Z
=
2).
The
axially
oriented
para--‐
aminophenyl
groups
in
2
facilitate
the
formation
of
self--‐assembled
monolayers
on
gold
surfaces,
and
thus
provide
a
viable
molecular
platform
for
charge
transport
studies
of
magnetically
functionalized
polyoxometalates.
The
title
systems
were
isolated
and
characterized
in
the
solid
state
and
in
aqueous
solutions,
and
on
metal
surfaces.
Using
conducting
tip
atomic
force
microscopy
(C--‐AFM),
the
energies
of
{Co9(P2W15)3}
frontier
molecular
orbitals
in
the
surface--‐bound
state
were
found
to
directly
correlate
with
cyclic
voltammetry
data
in
aqueous
solution.
INTRODUCTION
Highly
energy--‐efficient
molecular
spintronics
is
emerging
as
a
rapidly
growing
field
due
to
the
prospects
of
the
com--‐
bined
exploitation
of
molecular
charge
and
spin
states.[1]
Polyoxometalates
(POMs)
as
discrete
nanoscale
metal--‐oxo
clusters
able
to
incorporate
magnetic
centers
in
their
structures
offer
a
number
of
advantages
for
creating
mole--‐
cular
spintronic
devices,
such
as
thermal
and
redox
stabili--‐
ty
coupled
with
large
structural
diversity
and
tunability
of
the
magnetic
properties.[2]
One
of
the
key
challenges
to
fabricate
molecular
devices
is
control
over
molecular
an--‐
choring
on
the
surfaces
of
metallic
electrodes
in
terms
of
spatial
molecular
orientation
as
well
as
the
degree
of
elec--‐
tronic
interactions
between
the
metallic
surface
states
and
the
molecular
orbitals.[3]
As
such,
understanding
the
elec--‐
tronic
consequences
caused
by
the
interactions
of
magnet--‐
ically
functionalized
POMs
with
metal
substrates
is
essen--‐
tial
to
their
eventual
use
in
molecular
spintronics.
We
here
elaborate
a
POM--‐based
model
system
that
allows
us
to
assess
and
compare
the
molecular
frontier
orbitals
that
are
accessible
in
molecular
charge--‐transport
measurements
for
a
surface--‐adsorbed
POM
via
independent
methods,
namely
transient
voltage
spectroscopy
and
cyclic
voltam--‐
metry.
A
possible
step
towards
this
goal
relies
on
the
pre--‐func--‐
tionalization
of
a
magnetic
molecule
with
"glue
groups"
(e.
g.
–SH,
–N2+,
–NH2
etc.)
that
covalently
bind,
or
chemisorb,
to
a
specific
surface.
In
recent
years
it
was
demonstrated
that
such
groups
can
be
introduced
into
POM
species
by
attachment
of
various
organic
ligands,
e.
g.
alkoxides,
siloxanes,
organo(bis)phosphonates
and
--‐
arsonates,[4--‐6]
although
examples
of
functionalized
mag--‐
netic
POMs
are
still
scarce,[5]
despite
the
prospect
that
POM
functionalization
with
glue
groups
has
already
en--‐
abled
well--‐ordered
patterning
of
various
surfaces.[6]
Here
we
explore
a
novel
surface
anchoring
mode
(organoamino
group–Au
surface)
in
an
approach
to
render
magnetically
functionalized
POMs
accessible
to
charge
transport
experiments
in
distinct
environments,
in
solu--‐
tion
as
well
as
in
surface--‐adsorbed
monolayers.
For
the
design
of
our
target
molecules
we
exploited
the
fact
that
organoarsonates
on
one
hand
can
provide
a
robust
tetra--‐
hedral
arsonate
site
that
often
can
be
readily
integrated
as
a
part
of
a
magnetic
core
of
transition
metal
ions,
e.
g.
to
replace
phosphate
groups
possessing
a
terminal
oxygen
(see
[7]
for
some
examples
of
POMs
containing
such
HPO4–
groups).
On
the
other
hand,
starting
with
pheny--‐
larsonate,
as
an
easily
accessible
model
ligand,
one
can
further
introduce
various
functional
substituents
to
the
phenyl
ring
(e.
g.
–NH2
groups
in
meta--‐
or
para--‐positions).
Until
now
no
magnetic
organoarsonate--‐polyoxotungstate
derivatives
have
been
reported,
although
several
magnetic
bisphosphonates--‐containing
POMs
are
known,[5]
e.
g.
[{(B--‐
α--‐PW9O34)Co3(OH)(H2O)2(Ale)2}2Co]14–
(H5Ale
=
(+H3N(CH2)3)(OH)C(PO3H2)2)
that
was
shown
to
exhibit
single--‐molecule
magnet
features
and
is
composed
of
two
{B--‐α--‐PW9O34Co3}
subunits
connected
via
an
additional
CoII
center
as
well
as
two
alendronate
ligands.[5a]
Our
efforts
resulted
in
isolation
of
two
novel
species
with
general
formula
[CoII9(H2O)6(OH)3(p--‐RC6H4AsO3)2(α--‐
P2W15O56)3]25–,
where
R
is
either
H
(1)
or
NH2
(2),
which
we r e
c r y s t a l l i z e d
a s
h yd r a t e d
s o d i u m
s a l t s
N a 2 5 [ C o 9 ( H 2 O ) 6 ( O H ) 3 ( p --‐ R C 6 H 4 A s O 3 ) 2 ( α --‐
P2W15O56)3]·∙86H2O
(Na--‐1
and
Na--‐2,
respectively)
and
characterized
in
the
solid
state
and
aqueous
solutions.
We
also
carried
out
surface
deposition
studies
on
bare
gold
surfaces
for
the
amino--‐terminated
polyanion
2,
which
are
of
particular
interest
as
to
date
the
amine--‐containing
mol--‐
ecule
self--‐assembled
monolayers
(SAMs)
on
Au
were
less
studied
in
comparison
with
thiol--‐containing
SAMs.[8]
RESULTS
AND
DISCUSSION
Synthesis.
The
polyanions
were
prepared
by
reacting
of
CoCl2,
phenylarsonic
(for
1)
or
p--‐arsanilic
(for
2)
acid
and
the
tri--‐lacunary
Wells--‐Dawson
type
polyoxotungstate
salt
Na12[α--‐P2W15O56]·∙24H2O[9]
in
0.66
M
CH3COOH
/
CH3--‐
COONa
buffer
solution
(pH
5.2)
at
60
°C
for
four
days.
Plate--‐like
brown
crystals
of
Na16–xHx[(H2O)2Co4(α--‐
P2W15O56)2]·∙nH2O
(Na--‐3),
based
on
the
well--‐known
sand--‐
wich--‐type
polyanion
[(H2O)2Co4(α--‐P2W15O56)2]16–
(3),[10]
form
as
the
first
product
during
evaporation
of
the
reac--‐
tion
solution
and
should
be
repeatedly
removed
by
filtra--‐
tion.
Further
evaporation
of
the
obtained
filtrate
within
several
days
leads
to
pink
needle--‐like
(or
elongated
plate--‐
like)
crystals
of
Na--‐1
or
Na--‐2.
The
crystalline
materials
of
Na--‐1
and
Na--‐2
should
be
collected
within
1
to
2
days
after
formation
to
prevent
their
contamination
with
Na--‐3
side--‐
product.
The
use
of
any
larger
amounts
of
CoII
ions
(than
specified
in
the
synthetic
procedures)
during
the
synthesis
of
Na--‐1
or
Na--‐2
leads
to
the
presence
of
CoII
as
counterca--‐
tions
and
thus
should
be
prevented.
The
influence
of
other
synthetic
reaction
parameters
on
the
formation
of
1
and
2
is
discussed
after
the
structural
description.
Crystal
structures.
Single--‐crystal
X--‐ray
structural
analy--‐
sis
of
Na--‐1
and
Na--‐2
revealed
that
polyanions
1
and
2
dis--‐
play
a
similar
molecular
structure
based
on
a
{CoII9(H2O)6(OH)3(p--‐RC6H4AsO3)2}
core
(Co9L2,
where
L
=
C6H5AsO32–
for
1
and
p--‐H2NC6H4AsO32–
for
2,
Fig.
1c),
sta--‐
bilized
by
three
[α--‐P2W15O56]12–
POT
moieties
(Fig.
1a,
b).
Alternatively,
the
core
structure
of
1
and
2
can
be
viewed
as
a
C3h--‐symmetric
trimer
of
{Co3(H2O)2P2W15O56}
units
(that
can
be
considered
as
tri--‐substituted
{M18P2O56}
Wells--‐
Dawson--‐type
phosphotungstates)
linked
together
by
three
OH
groups
and
two
phenylarsonate
or
p--‐arsanilate
lig--‐
ands.
1
and
2
represent
equilateral
triangular
structures
with
side
lengths
of
ca.
2.2
nm
and
a
thickness
(maximum
extension
of
the
organoarsonates
groups)
of
ca.
1.7
nm.
Each
CoII
ion
in
the
{Co3(H2O)2P2W15O56}
building
block
resides
in
an
octahedral
coordination
environment
and
coordinates
to
one
µ4--‐O
(3Co,
P)
of
the
inner
phosphate
group
(Co–O:
2.166(14)
–
2.268(13)
Å
for
1,
2.156(14)
–
2.239(14)
Å
for
2)
and
two
µ2--‐O
(Co,
W)
atoms
(Co–O:
2.014(14)
–
2.109(12)
Å
for
1,
2.014(14)
–
2.100(14)
Å
for
2)
of
a
{P2W15}
subunit.
One
of
the
three
CoII
ions
additionally
coordinates
two
µ3--‐O,
linking
it
with
two
other
CoII
centers
and
the
AsV
center
of
the
phenylarsonate
or
p--‐arsanilate
ligand
(Co–O:
2.038(13)
–
2.119(13)
Å
for
1,
2.048(14)
–
2.093(14)
Å
for
2)
as
well
as
a
µ3--‐OH
group
that
connects
it
to
two
CoII
ions
of
the
neighboring
{Co3(H2O)2P2W15O56}
unit
(Co–O:
2.041(13)
–
2.146(13)
Å
for
1,
2.048(14)
–
2.093(14)
Å
for
2).
The
other
two
CoII
ions
bind
to
a
terminal
aqua
ligand
(Co–O:
2.111(15)
–
2.135(14)
Å
for
1,
2.087(16)
–
2.157(15)
Å
for
2),
one
above--‐mentioned
µ3--‐oxygen
of
the
phenylar--‐
sonate/p--‐arsanilate
and
a
µ3--‐OH
group
linking
them
to
each
other
and
to
a
CoII
of
the
third
{Co3(H2O)2P2W15O56}
moiety
(Fig.
1a).
The
protonation
sites
(bridging
OH
and
terminal
H2O
ligands)
are
confirmed
by
bond
valence
sum
calculations
(Tables
S1,
S2).
The
P–O
and
W–O
bonds
in
1
and
2
are
typical
for
POTs.
As–O
bonds
amount
to
1.659(12)
to
1.685(12)
Å
for
1
and
1.680(13)
to
1.704(15)
Å
for
2,
while
As–C
bonds
are
1.89(2)
−
1.918(13)
Å
(1)
and
1.93(2)
Å
(2).
The
two
phenylarsonate/p--‐arsanilate
ligands
in
1/2
a r e
l o c a t e d
o n
t h e
o p p o s i t e
s i d e s
o f
t h e
{Co9(H2O)6(OH)3(α--‐P2W15O56)3}
assembly,
with
their
(p--‐
amino)phenyl
groups
oriented
along
the
C3
axis
of
the
inorganic
core
structure
(Fig.
1b,c).
These
organic
groups
also
eliminate
all
symmetry
elements
of
polyanions
1
and
2;
the
dihedral
angle
between
the
two
phenyl
rings
is
81.8(7)°
in
1
and
39.4(7)°
in
2.
Overall,
the
structure
of
the
{Co9(H2O)6(OH)3L2(α--‐P2W15O56)3}
assemblies
in
1
and
2
is
reminiscent
of
[Co9(H2O)6(OH)3(HPO4)2(α--‐P2W15O56)3]25–
(4)
reported
several
years
ago
by
the
Cronin
group,[7e]
where
two
phosphate
groups
are
present
instead
of
pheny--‐
l a r s o n a t e
o r
p --‐ a r s a n i l a t e .
T h e
c e n t r a l
{Co9(H2O)6(OH)3(HPO4)2}
core
is
also
present
in
[Co9(H2O)6(OH)3(HPO4)2(PW9O34)3]16−
polyanions[7a--‐c]
that
were
shown
to
catalyze
heterogeneous
water
oxida--‐
tion[11a--‐c]
while
stabilized
by
Keggin--‐type
{PW9O34}--‐type
POTs,
along
with
several
other
Co(II)--‐based
POTs.[11d]
We
note
that
already
back
in
1984,
Weakley
predicted
the
possibility
to
replace
terminal
oxygen
in
each
of
the
two
external
phosphate
groups
by
alkyl
or
aryl
groups
for
POM
functionalization.[7a]
Our
findings
support
a
general
strat--‐
egy
that
uses
organoarsonate
ligands
for
2
As
expected,
IR
spectra
of
Na--‐1
and
Na--‐2
(Fig.
S1)
are
simi--‐
lar
and
exhibit
a
set
of
three
peaks
characteristic
for
P–O
vibrations
at
1084,
1042
and
1011
cm–1
for
Na--‐1
and
1086,
1042
and
1009
cm–1
for
Na--‐2,
comparable
to
the
peaks
in
the
IR
spectrum
of
Na12[α--‐P2W15O56]·∙24H2O
at
1130,
1086
and
1009
cm–1.
The
disappearance
of
the
band
at
1130
cm–1
and
appearance
the
band
at
1042
cm–1
are
in
agreement
with
coordination
of
the
O
atom
of
PO4
group
at
the
lacu--‐
nary
site
of
{P2W15}
ligand
by
three
CoII
ions
in
1
and
2.
The
bands
at
933
(Na--‐1)
and
932
cm–1
(Na--‐2)
correspond
to
W=O
vibrations.
Peaks
at
880,
806,
731,
598,
525
and
459
cm−1
(Na--‐1)
and
881,
806,
725,
600,
521
and
457
cm–1
(Na--‐2)
are
associated
with
W–O–W,
W–O–Co
and
W–O–P
bond
vibrations.
The
remarkable
shift
of
these
bands
compared
to
W–O–W
bands
in
non--‐coordinated
{P2W15}
(see
Fig.
S1)
is
consistent
with
the
coordination
of
CoII
to
{P2W15}
in
Na--‐1
and
Na--‐2.
The
characteristic
As–O
bands
(800
−
815
cm–1)[13]
overlap
with
W–O–W/Co/P
modes
(strong
band
at
806
cm–1).
Overlap
also
affects
As–C,
C–C
and
C–H
vi--‐
brations
of
the
organic
moieties
in
Na--‐1
and
Na--‐2
in
the
POT
region.
However,
C–C
and
C–H
vibrations
result
in
weak
bands
between
1600
and
1100
cm–1
(Fig.
S1).
Charac--‐
teristic
to
Na--‐1
is
the
sharp
C–N
band
at
1352
cm–1.
We
have
additionally
performed
ATR
FT--‐IR
measurements
on
saturated
solution
of
Na--‐2
in
H2O
in
comparison
to
solid
Na--‐2
sample
in
KBr.
The
good
match
of
the
spectra
(Fig.
S3)
suggests
solution
stability
of
polyanions
2
in
aqueous
medium,
at
least
within
the
duration
of
the
mea--‐
surement.
Thermogravimetrical
analysis.
The
thermal
stability
of
Na--‐1
and
Na--‐2
was
investigated
in
the
range
of
25
–
900
°C
under
N2
atmosphere.
The
TGA
curves
of
both
compounds
are
similar
(Figs.
S4,
S5)
and
exhibit
a
major
mass
loss
in
several
consecutive
non--‐resolved
steps
up
to
300
°C
due
to
the
release
of
86
lattice
water
molecules
per
formula
unit
(10.6
%
observed
vs.
10.8
%
calcd.
for
Na--‐1,
10.3
%
obs.
vs.
10.8
%
calcd.
for
Na--‐2).
Several
additional
steps
in
the
300
–
690
°C
range
are
attributed
to
the
loss
of
six
coordinated
water
molecules
and
three
hydroxo
ligands,
combined
with
decomposition
and
removal
of
phenyl
(Na--‐1)
or
p--‐
aminophenyl
(Na--‐2)
groups
of
the
p--‐RC6H4AsO32–
ligands
as
well
as
with
O2
release
due
to
reduction
of
AsV
ions
(2.5
%
obs.
vs.
1.7
%
calc.
for
Na--‐1
and
2.6
%
obs.
vs.
1.8
%
calc.
for
Na--‐2).
Additional
decrease
in
mass
between
700
–
800
°C
may
stem
from
loss
of
volatile
arsenic
oxide
(e.g.
as
in--‐
complete
release
of
0.5
As4O6
per
formula
unit:
0.8
%
exp.
vs.
1.4
%
calc.
(Na--‐1)
and
0.6
%
obs.
vs.
1.4
%
calc.
(Na--‐2)).
The
total
mass
loss
at
900
°C
is
14.0
%
for
Na--‐1
and
13.8
%
for
Na--‐2.
Magnetochemical
analysis.
The
magnetic
data
of
Na--‐1
and
Na--‐2
are
shown
as
χmT
vs.
T
and
Mm
vs.
B
curves
in
Fig.
2
(molar
magnetic
susceptibility
χm,
molar
magnetiza--‐
tion
Mm,
temperature
T,
and
magnetic
field
B).
At
290
K,
the
χmT
value
of
Na--‐1
is
25.98
cm3
K
mol–1
at
0.1
T,
i.e.
in
the
range
of
20.81–30.43
cm3
K
mol–1
expected
for
nine
non--‐interacting,
octahedrally
coordinated
high--‐spin
CoII
centers.[14]
Upon
lowering
T,
χmT
decreases,
initially
slowly
(T
≥
180
K)
and
subsequently
rapidly,
down
to
3.37
cm3
K
mol–1
at
2.0
K.
At
this
temperature,
the
molar
magnetiza--‐
3
Figure
1.
Structure
of
polyanion
2
in
front
(a;
H
atoms
omit--‐
ted
for
clarity)
and
side
(b)
views.
(c)
Structure
of
the
{Co9(p--‐
ars)2}
core.
Color
code:
WO6:
red
octahedra,
PO4:
yellow
tetrahedral;
As:
green,
Co:
purple,
C:
black,
N:
blue,
H:
gray,
O:
red
spheres;
O
sites
belonging
to
O3As
groups
are
shown
in
pink,
bonds
in
the
phenylarsonate
or
p--‐arsanilate
groups
are
in
gray,
Co–O
bonds
in
orange.
functionalization
of
magnetic
polyanions
incorporating
tetrahedral
phosphate
or
arsenate
groups
with
terminal
oxo/hydroxo
groups.
We
also
explored
similar
reactions
with
organophosphonates
(e.
g.
phenylphosphonate),
this
however
only
resulted
in
the
formation
of
the
hydrated
sodium
salts
of
the
known
phosphate--‐based
POMs
4,
NaxH25–x[Co9(H2O)6(OH)3(HPO4)2(α--‐P2W15O56)3]·∙nH2O
(Na--‐4).[12]
Formation
of
Na--‐4
was
also
observed
by
react--‐
ing
CoCl2,
phenylarsonate
(or
p--‐arsanilate)
and
{P2W15}
in
1M
LiCl
at
higher
pH
values
(8−10)
or
in
Na2CO3
/
NaHCO3
buffer
solution
(pH
9.4).
We
hypothesize
that
high
pH
of
the
reaction
medium
leads
to
partial
{P2W15}
decomposi--‐
tion
and
the
release
of
free
phosphate
required
for
forma--‐
tion
of
4.
Also
addition
of
dimethylammonium
chloride
(DMACl)
or
CsCl
to
our
reaction
mixtures
for
preparation
of
1
and
2
results
in
crystallization
of
polyanions
4
rather
than
DMA
or
Cs
salts
of
the
desired
products.
Thus,
pH
and
counterion
size
play
a
crucial
role
in
the
isolation
of
1
and
2
over
non--‐functionalized
4.
tion
Mm
is
almost
linear
with
the
applied
field
up
to
ca.
2
T.
For
higher
fields,
the
magnetization
subsequently
increas--‐
es
with
continuously
decreasing
slope
yielding
8.1
NA
μB
at
5.0
T,
well
below
the
expected
saturation
value
of
30.0–
36.3
NA
μB
for
nine
non--‐interacting
CoII
centers.
Both
curves
thus
reveal
predominant
antiferromagnetic
ex--‐
change
interactions
between
the
nine
CoII
centers
in
1.
While
the
rapid
decrease
of
χmT
upon
cooling
below
≈
100
K,
is,
for
the
most
part,
due
to
these
exchange
interac--‐
tions,
the
χmT
vs.
T
curve
is
also
effected
by
the
ligand
field
of
each
single
CoII
center
causing
a
similar
contribution
that
distinctly
deviates
from
the
spin--‐only
behavior
in
the
range
2.0–180
K.
This
is
due
to
the
thermal
depopulation
of
the
energy
states
originating
from
the
4T1g
ground
mul--‐
tiplet
of
the
4F
ground
term
that
is
further
split
by
spin--‐
orbit
coupling
contributions,
in
particular
due
to
mixing
with
the
states
originating
from
the
exited
4T1g(4P)
multi--‐
plet.[15]
Figure
2.
Temperature
dependence
of
χmT
at
0.1
Tesla
of
Na--‐1
(back
circles)
and
Na--‐2
(red
triangles).
Insets:
molar
magne--‐
tization
Mm
vs.
applied
field
B
at
2.0
K.
The
magnetic
data
of
Na--‐2
and
of
Na--‐1
match
almost
per--‐
fectly.
Taking
into
account
the
structure
of
both
com--‐
pounds,
the
slight
differences
of
the
characteristic
values
(Na--‐2:
at
0.1
T,
χmT
=
25.75
cm3
K
mol–1
at
290
K,
and
2.91
cm3
K
mol–1
at
2.0
K;
Mm
=
7.6
NA
μB
at
5.0
T
and
2.0
K)
are
potentially
due
to
the
slightly
different
exchange
coupling
mediated
by
the
organoarsonate
ligands,
but
the
differences
might
also
be
caused
by
minor
paramagnetic
impurities.
Nevertheless,
the
conclusions
drawn
from
the
magnetochemical
analysis
of
Na--‐1
are
the
same
for
Na--‐2.
A
quick
comparison
of
the
magnetic
data
of
Na--‐1
and
Na--‐2
to
Na--‐4
[7e]
reveals
extensive
resemblance.
The
shape
of
the
χmT
vs.
T
and
Mm
vs.
B
curves
are
very
similar,
while
their
magnitudes
are
different.
This
may
be
caused
by
the
dif--‐
ferent
ligands
(organoarsonate
(Na--‐1,
Na--‐2)
vs.
phosphate
groups
(Na--‐4)).
The
magnitude
of
deviation
(about
25
%)
is,
however,
surprisingly
large
which
rather
points
to
an
uncertainty
in
the
molar
mass
(as
scaling
factor),
i.e.
a
dif--‐
ferent
amount
of
a
diamagnetic
component
such
as
crystal
water
within
the
measured
sample.
The
samples
used
for
magnetic
measurements
thus
were
subsequently
charac--‐
terized
by
TGA,
however,
the
results
are
the
same
as
pre--‐
sented
above.
In
addition,
compounds
Na--‐1,
Na--‐2
and
Na--‐
4
do
not
show
out--‐of--‐phase
ac
susceptibility
signals
down
to
1.8
K
and
up
to
1000
Hz.
UV--‐Vis
spectroscopy.
Aqueous
solutions
of
Na--‐1
and
Na--‐
2
(Figs.
S6,
S8)
show
similar
absorption
spectra
that
exhib--‐
it
a
strong
maximum
at
around
200
nm
(215
nm
for
Na--‐1
and
195
nm
for
Na--‐2)
followed
by
an
absorption
maximum
at
around
260
nm
(which
is
well--‐resolved
for
Na--‐1
and
is
overlapped
with
the
first
absorption
peak
for
Na--‐2)
in
the
UV
light
area
and
a
less
intense
absorption
maximum
at
532
nm
in
the
visible
light
area.
The
spectrum
remains
unchanged
for
at
least
18
h
confirming
the
short--‐term
sta--‐
bility
of
polyanions
in
aqueous
solutions
(Fig.
S9)
in
line
with
the
conclusions
obtained
from
the
Diamond
ATR--‐
FTIR
measurements.
Electrochemical
studies.
We
have
recorded
cyclic
voltammograms
for
0.7
mM
Na--‐1,
Na--‐2,
Na--‐3
and
Na--‐4
solutions
in
0.5
M
CH3COONa
buffer
(pH
4.8).
The
elec--‐
trochemical
behavior
of
the
three
Co9--‐based
POMs
(1,
2
and
4)
is
very
similar.
The
cyclic
voltammograms
for
these
species
exhibit
four
redox
couples
attributed
to
reduction
and
re--‐oxidation
of
the
WVI
centers
of
the
POT
ligands
between
–0.50
and
–1.05
V
vs.
Ag
/
AgCl
(Figs.
3,
S10,
Table
1).
At
potentials
below
–1.05
V
a
reduction
of
solvent
oc--‐
curs,
coupled
with
formation
of
a
film
on
the
glassy
carbon
electrode.
In
comparison
to
the
electrochemical
activity
of
the
sandwich--‐type
polyanions
3
in
the
same
medium,
ex--‐
hibiting
three
well--‐defined
redox
waves
before
hydrogen
evolution
(Table
1,
Fig.
S11),
reduction
of
WVI
centers
in
the
Co9--‐based
species
takes
place
at
more
negative
potentials
showing
higher
redox
stability
of
the
latter.
The
peak
cur--‐
rents
for
the
redox
processes
in
1
−
4
are
proportional
to
the
square
root
of
the
scanning
rate,
which
is
characteris--‐
tic
for
diffusion--‐controlled
electrode
reactions.
At
higher
pH
(6.4)
all
redox
waves
are
shifted
towards
more
negative
potentials
(Fig.
S12)
indicating
that
the
reduction
of
tung--‐
sten(VI)
ions
is
coupled
with
proton
transfer,
as
it
is
com--‐
mon
for
POTs.[16]
Correspondingly,
only
three
redox
waves
are
accessible
for
2
in
the
pH
6.4
medium
before
hydrogen
evolution
No
reversible
redox
waves
associated
with
CoII
oxidation
in
the
{Co9L2}
core
could
be
observed
in
the
pos--‐
itive
potential
range
at
both
pH
4.8
and
6.4.
The
CV
curves
of
1,
2
and
4
remain
unchanged
for
several
hours,
however
after
one
day
an
additional
shoulder
cen--‐
tered
at
around
–0.50
V
appears,
while
there
are
also
other
slight
changes
in
relative
intensities
and
shape
of
the
other
redox
waves.
The
shoulder
at
–0.50
V
looks
very
similar
to
the
first
redox
wave
observed
for
3
in
the
same
medium
(Fig.
S13).
These
changes
are
accompanied
by
a
color
change
of
the
POMs
solutions
from
pink
to
light--‐brown.
These
observations
imply
partial
decomposition
of
the
{Co9L2(P2W15)3}
POMs
in
0.5
M
sodium
acetate
solutions
(pH
4.8)
following
by
formation
of
the
{CoxNa4–x(P2W15)2}
species,
this
is
also
in
agreement
with
the
presence
of
Na--‐3
as
a
common
side
product
during
the
synthesis
of
Na--‐1
and
Na--‐2.
4
RMS
roughness
of
0.31
nm
(root
mean
square
roughness
is
defined
here
as
the
square
root
of
the
arithmetic
mean
of
the
squares
of
each
z
values
measured
on
an
AFM
image)
obtained
by
AFM
is
in
a
good
agreement
with
a
standard
deviation
of
0.2
nm
observed
by
ellipsometry.
This
value
is
also
well
comparable
with
the
RMS
roughness
of
the
tem--‐
plate--‐stripped
(TS)
Au
surface
before
grafting
molecules
measured
at
the
same
conditions,
which
is
around
0.5
nm.
However,
we
note
that
AFM
images
reveal
some
pinholes
with
a
diameter
of
few
tens
of
nm
and
apparent
depth
of
about
1
nm
(about
half
the
total
SAM
thickness,
albeit
the
exact
value
cannot
be
determined
due
to
AFM
tip
convolu--‐
tion
effect).
The
obtained
POM--‐covered
surface
was
also
characterized
by
CV.
The
three--‐electrode
cell
setup
was
used
with
a
POM--‐covered
gold
surface
as
the
working
electrode,
Pt
wire
as
a
counter
electrode,
and
Ag/AgCl
(saturated
KCl)
as
a
reference
electrode
(Fig.
S16).
The
characteristic
wave
of
the
successive
one--‐electron
reduction
process
attributed
to
2
on
the
gold
surface
(black
line,
Fig.
S16)
was
identified
at
ca.
–0.38
V
vs.
Ag/AgCl
(saturated
KCl),
but
it
is
not
well
defined
and
appears
irreversible.
We
also
perform
CV
measurements
of
the
Na--‐2
in
the
same
electrolyte
solution
we
used
for
CV
studies
of
the
POM--‐containing
surface
for
comparison
(red
line,
Fig.
S16).
For
this
measurements
Pt
wires
acted
as
working
and
counter
electrodes,
whereas
Ag/AgCl
(saturated
KCl)
was
used
as
a
reference
electrode.
The
XPS
experiments
of
the
Na--‐2
SAM
on
gold
(Fig.
S17)
show
distinctive
photoemission
lines
of
W4f7/2
(36.2
eV)
and
W4f5/2
(38.4),
As3d
(42.0
and
45.5
eV),
C1s
(284.8
eV)
and
Co2p
(~782
eV,
almost
invisible).
The
results
are
con--‐
firmed
by
XPS
measurements
of
the
Na--‐2
powder
sample,
which
also
reveal
photoemission
peaks
of
the
other
ele--‐
ments
of
interest
(Fig.
S18).
Due
to
its
low
content
and
relatively
low
photoionization
cross--‐sections,
a
very
weak
photoemission
peak
of
N
1s
(~401
eV)
is
present
just
at
the
detection
limit
of
the
experiment
and
hints
at
a
broaden--‐
ing
and
splitting
indicative
of
attachment
of
the
amino
group
to
the
Au
substrate.
Doublets
of
P
2p
(~136
eV)
and
Co
2p
(~781
eV)
for
the
powder
sample
are
significantly
stronger
comparing
to
the
Na--‐2
on
Au,
which
is
aligned
with
our
expectations.
A
semi--‐quantitative
analysis
of
the
photoemission
lines
belonging
to
Co,
Na,
P,
W
and
C
was
performed
using
the
SESSA
simulation
package.[18]
The
analysis
confirms
the
powder
retains
its
stoichiometry.
Photoelectron
cross--‐sections
values
were
taken
from
ref.
19,
inelastic--‐mean
free
path
was
calculated
using
the
TPP--‐
2
formula.[20]
Interestingly,
the
Na
1s
peak
was
not
ob--‐
served
for
Na--‐2
SAM
on
gold,
in
contrast
to
the
XPS
data
for
bulk
material
(~1073
eV
for
Na--‐2
powder).
Additionally,
a
thin
layer
of
Na--‐2
immobilized
on
an
Au
surface
from
an
aqueous
solution
was
characterized
by
FT--‐
IRRAS
(Fig.
S19).
The
measurement
reveals
several
charac--‐
teristic
peaks
that
further
supports
uniform
Na--‐2
immobi--‐
lization
on
the
Au
surface.
Thus,
the
bands
at
821
cm−1
as
well
as
at
919
cm−1
are
most
likely
vibrations
of
W–O–W,
Co–O–W,
P–O–W
bonds,
while
that
at
1108
cm−1
corre--‐
sponds
to
vibrations
of
the
P–O
bonds.
They
are
slightly
shifted
as
compared
to
the
peaks
detected
by
FT--‐IR
mea--‐
5
Figure
3.
Room--‐temperature
cyclic
voltammograms
of
0.7
mM
solution
of
Na--‐1
in
0.5
M
CH3COONa
buffer
(pH
4.8)
with
different
negative
potential
limits
(–0.625
V,
–0.75
V,
–
0.95
V
and
–1.025
V)
at
scan
rate
20
mV/s.
Table
1.
Electrochemical
data
for
the
polyanions
1
−
4
in
0.5
M
CH3COONa
buffer
(pH
4.8).
The
potentials
are
given
vs.
Ag/AgCl
reference
electrode
at
20
mV/s
scan
rate.
POM
E1/2I,
V
1
2
4
3
−0.57±0.0
4
−0.56±0.0
7
−0.58±0.0
7
−0.50±0.0
8
E1/2II,
V
E1/2III,
V
−0.67±0.03 −0.82±0.0
3
−0.67±0.04 −0.82±0.0
3
−0.67±0.05 −0.80±0.0
2
−0.70±0.06 −0.89±0.0
6
E1/2IV,
V
−0.94±0.0
3
−0.94±0.0
3
−0.92±0.0
3
–
Formation
and
characterization
of
self--‐assembled
monolayer
(SAM).
We
investigated
the
electron
transport
properties
of
a
molecular
device
based
on
the
amine--‐ter--‐
minated
polyanions
2.
For
this
purpose,
we
prepared
SAMs
of
2
on
flat
bare
template--‐stripped
gold
surface
(AuTS)
[17]
and
characterized
it
using
ellipsometry,
FT--‐IRRAS,
X--‐ray
photoemission
spectroscopy
(XPS),
cyclic
voltammetry
(CV)
and
C--‐AFM
measurements.
It
was
found
that
the
incubation
of
the
freshly
prepared
gold
surface
in
the
aqueous
solution
of
Na--‐2
(10–3
M
in
DI
water)
during
24
hours
leads
to
the
formation
of
a
layer
with
the
thickness
between
1.4
±
0.2
and
2.0
±
0.2
nm
(measured
on
two
batches,
see
experimental
section),
which
is
consistent
with
the
expected
thickness
of
the
polyoxoanion
monolayer
in
the
most
probable
orientation
on
the
gold
surface
shown
on
Fig.
S14
(~1.7
nm
according
to
the
XRD
data).
The
thickness
of
the
obtained
layer
did
not
change
after
ultrasonic
cleaning
of
the
sample
in
water,
indicating
a
high
stability
of
the
POM
SAM.
The
results
of
topographic
AFM
in
contact
mode
(Fig.
S15)
further
suggest
the
homogeneity
of
the
obtained
layer.
The
surements
in
transmission
of
Na--‐2
powder
(806,
881
cm−1;
and
1086
cm−1,
respectively).
This
shift
is
expected
due
to
the
different
sample
forms
and
measurement
methods
that
lead
to
variant
vibration
frequencies.[21]
The
numerous
rotational--‐vibrational
transitions
observed
between
2000
cm−1
and
1200
cm−1
should
be
owing
to
the
hydrophilic
nature
of
Na--‐2,
the
high
amount
of
crystal
water
present
in
this
compound
as
well
as
the
preparation
of
the
sample
from
ultrapure
water.
We
also
analyze
the
I--‐V
curves
with
the
TVS
(transient
voltage
spectroscopy)
method[26]
to
estimate
the
energy
position
(with
respect
to
the
Fermi
energy
of
the
elec--‐
trodes)
of
the
molecular
orbital
(HOMO
or
LUMO)
in--‐
volved
in
the
charge--‐transport
process
(Fig.
4b).
From
the
minima
of
the
TVS
plots
(Fig.
S21),
we
deduced
(see
Exper--‐
imental
Part)
a
LUMO
at
ca.
(0.56±0.17)
eV
above
the
elec--‐
trode
Fermi
energy
(energy
diagram
Fig.
4b),
or
at
(4.24±0.17)
eV
below
the
vacuum
level
(considering
a
work
function
of
AuTS
at
4.8
eV).
This
later
value
is
consistent
with
the
LUMO
position
determined
from
CV
on
the
SAM,
the
reduction
peak
at
−0.38
V
giving
a
LUMO
at
–4.2
eV
below
vacuum
level
(see
equation
in
the
Experimental
Part).
This
consistency
between
I--‐V
and
CV
measurements
indicates
a
weak
or
moderate
electronic
coupling
between
the
molecules
and
the
Au
surface
(through
the
phenyl
amine
groups)
on
one
side
and
through
the
C--‐AFM
tip
mechanical
contact
on
the
other
side.
Figure
4.
(a)
2D
current
histogram
of
325
I--‐V
curves
mea--‐
sured
by
C--‐AFM
(at
a
loading
force
of
30
nN)
on
the
SAM
of
2
chemically
grafted
on
ultra--‐flat
template
stripped
gold
elec--‐
trode
(AuTS).
Voltages
applied
on
the
AuTS
electrode
(C--‐AFM
grounded).
(b)
Schematic
energy
diagram
of
the
AuTS
/
POM
/
PtIr
C--‐AFM
tip
as
deduced
from
both
the
I--‐V
and
CV
measurements.
Current--‐voltage
(I--‐V)
curves
were
measured
by
electrically
contacting
the
SAM
by
C--‐AFM.
Figure
4a
shows
the
2D
histograms
of
around
three
hundred
of
I--‐V
curves
(see
experimental
part).
The
I--‐V
curves
are
dispersed
over
about
three
decades
as
shown
by
current
histograms
at
a
given
bias
(e.
g.
+0.9V
and
−0.9
V,
Fig.
S20).
This
disper--‐
sion
is
large
when
compared
to
similar
C--‐AFM
measure--‐
ments
of
alkylthiol
SAMs
on
Au
(here
about
3
decades
vs.
about
one
[22])
or
π--‐conjugated
molecules.[23]
The
I--‐V
curves
are
symmetric
with
respect
of
the
voltage
polarity.
The
current
histograms
are
fitted
by
two
Gaussian
peaks
(Fig.
S20),
the
mean
values
of
the
current
histograms
(Fig.
S20)
are
ca.
1.82×10–11
A
/
1.6×10–10
A
(two
peaks)
at
0.9
V
and
ca.
1.9×10–11
A
/
2.1×10–10
A
at
–0.9
V.
The
results
have
been
reproduced
in
a
second
batch
of
SAMs
(see
Figs.
S22--‐
S24).
We
can
infer
several
possible
origins
for
this
large
disper--‐
sion:
(1)
Disorder
in
the
SAMs.
For
example,
in
some
cases
mul--‐
ti--‐peaks
and
consequently
a
larger
dispersion
can
be
ob--‐
served
for
disordered
SAMs
over
two
decades
of
current.[24]
This
large
dispersion
can
also
be
due
to
the
presence
of
"defects"
(such
as
pinholes,
see
above
and
Fig.
S15)
with
reduced
thickness
and
thus
larger
current
through
the
SAMs.
Thus,
the
two
peaks
in
the
current
histograms
can
be
ascribed
to
zones
of
different
molecular
organizations
in
the
SAMs.
(2)
Intermolecular
interaction
(π--‐π
interactions)
between
neighboring
molecules
which
can
broaden
the
conduc--‐
tance
histograms.[25]
EXPERIMENTAL
PART
General
methods
and
materials.
Reagents
were
used
as
purchased
without
further
purification.
Na12[α--‐
P2W15O56]·∙24H2O
was
obtained
according
to
the
reported
procedure
[9]
starting
from
K6[α--‐P2W18O62]·∙14H2O.[27]
Ele--‐
mental
analysis
results
(ICP--‐OES
and
C,
H,
N)
were
ob--‐
tained
from
Central
Institute
for
Engineering,
Electronics
and
Analytics
(ZEA--‐3),
Forschungszentrum
Jülich
GmbH
(D--‐52425
Jülich,
Germany).
TGA/DTA
measurements
were
carried
out
with
a
Mettler
Toledo
TGA/SDTA
851
in
dry
N2
(60
mL
min–1)
at
a
heating
rate
of
5
K
min–1.
Vibrational
spectra
were
recorded
on
a
Bruker
VERTEX
70
FT--‐IR
spec--‐
trometer
coupled
with
a
RAM
II
FT--‐Raman
module
(1064
nm
Nd:YAG
laser)
on
KBr
disks
for
the
FT--‐IR
and
the
solid
material
for
the
Diamond
ATR--‐FTIR
and
Raman
mea--‐
surements.
Liquid--‐phase
ATR--‐FTIR
spectra
were
obtained
on
saturated
Na--‐2
solution
in
H2O.
UV--‐Vis
spectra
were
measured
using
10
mm
quartz
cuvettes
on
Analytik
Jena
Specord
S600
spectrophotometer.
Synthesis
of
Na25[Co9(OH)3(H2O)6(C6H5AsO3)2--‐
(P2W15O56)3]·∙86H2O
(Na--‐1).
Na12[α--‐P2W15O56]·∙24H2O
(0.221
g,
0.05
mmol),
CoCl2
(0.023
g,
0.175
mmol),
and
C6H5AsO3H2
(0.010
g,
0.05
mmol)
were
dissolved
in
6
mL
of
0.66
M
CH3COONa
buffer
(pH
5.2)
[28]
under
vigorous
stirring
in
an
oil
bath
at
60
°C
for
4
days.
The
reaction
mix--‐
ture
was
then
filtered,
divided
into
several
vials
and
left
for
evaporation
at
room
temperature.
Brown
plate--‐like
crys--‐
t a l s
N a 1 6 –
xHx[(H2O)2Co4(P2W15O56)2]·∙nH2O
(Na--‐3)
start
to
appear
immediately
after
cooling
the
reaction
mixture
to
room
temperature.
They
have
to
be
removed
by
filtration,
and
filtration
of
Na--‐3
has
to
be
repeated
in
the
next
few
days.
Pure
dark--‐pink
crystals
of
Na--‐1
form
after
2
weeks.
The
crystals
were
collected
by
filtration,
washed
with
0.66
M
CH3COONa
buffer
and
dried
in
air.
Yield:
0.016
mg
(6.6
%
based
on
{P2W15}).
Elemental
analysis:
calculated
for
C12H197Co9Na25O269P6As2W45
(found):
C,
1.00
(1.04);
H,
1.38
(1.38);
Co,
3.69
(3.70);
Na,
4.00
(4.00);
P,
1.29
(1.29);
As,
1.04
(1.03);
W,
57.61
(55.7)
%.
IR
(KBr
pellet),
cm–1:
3454
(s,
br);
1632
(s);
1439
(w);
1418
(w);
1385
(w);
1352
(w);
1084
(s);
6
p r o d u c t
o f
t h e
s i d e
1042
(m);
1011
(m);
933
(s);
912
(s);
880
(s);
806
(s);
731
(s);
598
(m);
561
(m);
525
(m);
492
(m);
459
(m);.
Raman
(in
solid),
cm–1:
977
(s);
956
(s);
880
(m);
816
(m);
523
(w);
370
(m);
121
(m).
UV--‐Vis
(H2O):
λ
=
215
nm,
ε
=
2.45×107
M–
1cm–1;
λ
=
265
nm,
ε
=
1.06×107
M–1cm–1;
λ
=
532
nm,
ε
=
3.67×105
M–1cm–1.
UV--‐Vis
(0.5
M
CH3COONa
solution,
pH
5.1):
λ
=
235
nm,
ε
=
3.21×109
M–1cm–1;
λ
=
531
nm,
ε
=
3.88×105
M–1cm–1.
Synthesis
of
Na25[Co9(OH)3(H2O)6(H2NC6H4AsO3)2
(P2W15O56)3]·∙86H2O
(Na--‐2).
Na12[α--‐P2W15O56]·∙24H2O
(0.221
g,
0.05
mmol),
CoCl2
(0.023
g,
0.175
mmol),
and
H2NC6H4AsO3H2
(0.011
g,
0.05
mmol)
were
dissolved
in
6
mL
of
0.66
M
CH3COONa
buffer
(pH
5.2)
[28]
under
vigor--‐
ous
stirring
in
an
oil
bath
at
60
°C
for
4
days.
Then
the
resulting
reaction
mixture
was
filtered,
divided
into
several
vials
and
left
for
evaporation
at
room
temperature.
Brown
plate --‐like
cr ystals
of
the
side
product
Na 16–
xHx[(H2O)2Co4(P2W15O56)2]·∙nH2O
(Na--‐3)
start
to
form
immediately
after
cooling
the
reaction
mixture
and
have
to
be
removed
by
filtration
after
the
reaction
and
on
the
next
day.
Pink
needle--‐like
crystals
of
Na--‐2
form
after
2
days.
They
were
collected
by
filtration,
washed
with
0.66
M
CH3COONa
buffer
and
dried
in
air.
Yield:
0.0075
mg
(3.1
%
based
on
{P2W15}).
Elemental
analysis:
calculated
for
C12H199N2Co9Na25O269P6As2W45
(found):
C,
1.00
(1.09);
H,
1.39
(1.37);
N,
0.19
(0.19);
Co,
3.69
(3.73);
Na,
3.99
(3.93);
P,
1.29
(1.33);
As,
1.04
(1.09);
W,
57.49
(57.67)%.
IR
(KBr
pel--‐
let),
cm–1:
3449
(s,
br);
1626
(s);
1504
(w);
1470
(w);
1383
(w);
1354
(w);
1321
(w);
1292
(w);
1264
(w),
1194
(w);
1142
(m);
1086
(s);
1042
(m);
1009
(m);
932
(s);
910
(s),
881
(s);
806
(s);
725
(s);
600
(m);
561
(m);
521
(s);
457
(m).
Raman
(in
solid),
cm–1:
978
(s);
957
(s);
916
(m),
905
(m),
893
(m),
883
(m);
839
(m);
523
(w).
UV--‐Vis
(H2O):
λ
=
195
nm,
ε
=
1.23×108
M–1cm–1;
λ
=
255
nm,
ε
=
3.01×107
M–1cm–1;
λ
=
532
nm,
ε
=
6.60×104
M–1cm–1.
X--‐ray
crystallography.
Single--‐crystal
diffraction
data
for
Na--‐1
and
Na--‐2
were
collected
on
a
SuperNova
(Agilent
Technologies)
diffractometer
with
MoKα
radiation
(λ
=
0.71073
Å)
at
120
K.
Crystals
were
mounted
in
a
Hampton
cryoloop
with
Paratone--‐N
oil
to
prevent
water
loss.
Ab--‐
sorption
corrections
were
applied
numerically
based
on
multifaceted
crystal
model
using
CrysAlis
software.[29]
The
SHELXTL
software
package
[30]
was
used
to
solve
and
refine
the
structure.
The
structures
were
solved
by
direct
meth--‐
ods
and
refined
by
full--‐matrix
least--‐squares
method
against
F2
with
anisotropic
thermal
parameters
for
all
heavy
atoms
(W,
P,
Co,
Na).
The
hydrogen
atoms
of
the
phenyl
rings
in
Na--‐1
and
Na--‐2
and
the
amino
groups
in
Na--‐2
were
placed
in
geometrical--‐
ly
calculated
positions.
The
hydrogens
of
the
water
mole--‐
cules
and
OH
groups
were
not
located.
The
relative
site
occupancy
factors
for
the
disordered
positions
of
Na+
countercations
and
oxygen
atoms
of
co--‐crystallized
sol--‐
vent
water
molecules
were
first
refined
in
an
isotropic
ap--‐
proximation
with
Uiso=
0.05
and
then
fixed
at
the
obtained
values
and
refined
without
the
thermal
parameters
restric--‐
tions.
One
of
the
P2W15
ligands
in
Na--‐2
structure
was
disordered
with
95:5
%
ratio
between
two
components.
The
disorder
was
modeled
using
combination
of
PART
and
EADP
in--‐
structions.
The
rather
high
value
still
remaining
for
the
second
weighting
term
in
(650
for
Na--‐1
and
1420
for
Na--‐2)
most
likely
reflects
the
large
volume
occupied
by
highly
disordered
solvate
and
is
consistent
with
large
solvent--‐ac--‐
cessible
volume
remaining
in
the
structures.
Severe
disor--‐
der
did
not
allow
localizing
the
positions
for
all
the
Na
cations
and
O
atoms
of
crystal
waters
in
Na--‐1.
Thus,
only
16.5
Na+
ions
and
60
co--‐crystallized
H2O
molecules
have
been
found
from
the
X--‐Ray
data
for
this
compound,
while
25
Na+
and
86
crystal
waters
are
present
as
based
on
ele--‐
mental
and
thermogravimetrical
analyses.
For
the
overall
consistency,
the
final
formula
in
the
CIF
file
of
Na--‐1
corre--‐
spond
to
the
composition
of
the
bulk
materials
deter--‐
mined
by
elemental
analysis
and
TGA,
as
all
further
stud--‐
ies
were
performed
on
the
isolated
well--‐dried
bulk
materi--‐
als
of
Na--‐1.
The
number
of
counterions
and
co--‐crystallized
solvent
molecules
found
in
the
crystal
structure
for
Na--‐2
was
consistent
with
those
determined
by
other
analytical
techniques.
Additional
crystallographic
data
are
summarized
in
Table
2.
Further
details
on
the
crystal
structures
investigation
can
be
obtained,
free
of
charge,
on
application
to
CCDC,
12
Union
Road,
Cambridge
CB2
1EZ,
UK:
http://www.ccdc.--‐
cam.ac.uk/,
e--‐mail:
[email protected],
or
fax:
+441223
336033
upon
quoting
1552113
(Na--‐1),
and
1552114
(Na--‐2)
numbers.
Magnetic
measurements
were
performed
using
a
Quan--‐
tum
Design
MPMS--‐5XL
SQUID
magnetometer.
The
poly--‐
crystalline
samples
were
compacted
and
immobilized
into
cylindrical
PTFE
capsules.
In
static
magnetic
field,
data
were
acquired
as
a
function
of
the
magnetic
field
(0.1–5.0
T
at
2.0
K)
and
temperature
(2.0–290
K
at
0.1
T).
Dynamical
field
(ac)
data
were
collected
in
the
absence
of
a
static
bias
field
in
the
frequency
range
1–1500
Hz
(T
=
2.0–50
K,
Bac
=
3
G),
however,
no
out--‐of--‐phase
signals
were
detected.
Data
were
corrected
for
the
diamagnetic
contributions
of
sam--‐
ple
holder
and
compound
(Na--‐1:
χdia
=
–7.18×10–3 cm3 mol–
1,
Na--‐2:
χdia
=
–7.20×10–3 cm3 mol–1).
Table
2.
Crystal
data
and
structure
refinement
for
Na--‐1
and
Na--‐2
Sample
Na--‐1
Na--‐2
Empirical
formula
C12H197As2Co9Na25O269P6W45
C12H199As2Co9N2Na25O269P6W45
Formula
weight,
g
mol–1
14360.73
14390.76
7
Crystal
system
Space
group
a,
Å
b,
Å
c,
Å
α
β
γ
Volume,
Å3
Z
Dcalc,
g/cm3
Absorption
coefficient,
mm–1
F(000)
Crystal
size,
mm3
Theta
range
for
data
collection
Completeness
to
Θmax
Index
ranges
Reflections
collected
Independent
reflections
Rint
Observed
(I
>
2σ(I))
Absorption
correction
Tmin
/
Tmax
Data
/
restraints
/
parameters
Goodness--‐of--‐fit
on
F2
R1,
wR2
(I
>
2σ(I))
R1,
wR2
(all
data)
Triclinic
P–1
25.8088(3)
25.8336(3)
27.1598(3)
78.1282(11)
°
61.7276(14)
°
60.6220(14)
°
13888.9(3)
2
3.434
19.496
12850
0.08×
0.23
×
0.81
4.08°
–
25.35°
99.5%
–31
≤
h
≤
30,
–31
≤
k
≤
31,
–32
≤
l
≤
32
259838
50595
0.0981
36674
Triclinic
P–1
14.2262(2)
24.8597(4)
37.9388(4)
81.9672(10)°
87.8161(10)°
76.5409(12)°
12920.6(3)
2
3.699
20.958
12882
0.11×
0.16
×
0.51
4.08°
–
25.68°
99.5%
–17
≤
h
≤
17,
–30
≤
k
≤
30,
–45
≤
l
≤
46
248406
48866
0.0799
41474
analytical
using
a
multifaceted
crystal
model
0.0201
/
0.3081
50595
/
60
/
1941
1.057
R1
=
0.0614,
wR2
=
0.1450
R1
=
0.0923,
wR2
=
0.1698
0.0159
/
0.2120
48866
/
54
/
2189
1.142
R1
=
0.0643,
wR2
=
0.1374
R1
=
0.0771,
wR2
=
0.1443
Largest
diff.
peak
and
hole,
e
Å–3
5.061
and
–4.591
4.614
and
–4.239
Cyclic
voltammograms
of
1
–
4
in
aqueous
media
were
recorded
using
a
Bio
Logic
SP--‐150
potentiostat
controlled
via
EC--‐Lab
software.
The
conventional
three--‐electrode
electrochemical
cell
included
a
glassy
carbon
working
electrode
with
the
diameter
of
3
mm,
a
platinum
wire
counter
electrode
and
an
aqueous
Ag/AgCl
(3M
NaCl)
reference
electrode
(0.196
V
vs.
SHE
determined
by
mea--‐
suring
[Fe(CN)6]3–/4–
as
an
internal
standard).
The
solu--‐
tions
were
thoroughly
deaerated
with
pure
argon
and
kept
under
a
positive
Ar
pressure
during
the
experiments.
Alu--‐
mina
powder
was
used
for
the
cleaning
of
the
working
electrode
which
was
then
thoroughly
rinsed
with
deion--‐
ized
water.
Redox
potentials
were
determined
from
the
average
values
of
the
anodic
and
cathodic
peak
potentials
and
reported
vs.
Ag/AgCl
(3M
NaCl)
reference
electrode.
8
Template--‐stripped
Au
substrates.
Very
flat
AuTS
sur--‐
faces
were
prepared
according
to
the
method
reported
by
the
Whiteside's
group.[17]
In
brief,
a
300−500
nm
thick
Au
film
is
evaporated
on
a
very
flat
silicon
wafer
covered
by
its
native
SiO2
(RMS
roughness
of
0.4
nm)
which
was
previ--‐
ously
carefully
cleaned
by
piranha
solution
(30
min
in
2
:
1
H2SO4
:
H2O2
(v/v).
Caution:
Piranha
solution
is
exother--‐
mic
and
strongly
reacts
with
organics),
rinsed
with
deion--‐
ized
(DI)
water
and
dried
under
a
stream
of
nitrogen.
A
clean
glass
piece
(ultrasonicated
in
acetone
for
5
min,
ul--‐
trasonicated
in
isopropanol
for
5
min
and
UV--‐irradiated
in
ozone
for
10
min)
is
glued
(UV
polymerizable
glue)
on
the
evaporated
Au
film
and
mechanically
stripped
with
the
Au
film
attached
on
the
glass
piece
(Au
film
is
cut
with
a
razor
blade
around
the
glass
piece).
This
very
flat
(RMS
rough--‐
ness
of
0.4
nm
same
as
the
SiO2
surface
used
as
template)
and
clean
template--‐stripped
AuTS
surface
is
immediately
used
for
the
formation
of
the
molecule
self--‐assembled
monolayer.
Spectroscopic
ellipsometry.
We
recorded
spectroscopic
ellipsometry
data
in
the
visible
range
using
an
UVISEL
(Jobin
Yvon
Horiba)
Spectroscopic
Ellipsometer
equipped
with
DeltaPsi
2
data
analysis
software.
The
system
ac--‐
quired
a
spectrum
ranging
from
2
to
4.5
eV
(corresponding
to
300
to
750
nm)
with
intervals
of
0.1
eV
(or
15
nm).
Data
were
taken
at
an
angle
of
incidence
of
70°,
and
the
com--‐
pensator
was
set
at
45.0°.
Data
were
fitted
by
a
regression
analysis
to
a
film--‐on--‐substrate
model
as
described
by
their
thickness
and
their
complex
refractive
indexes.
First,
a
background
before
monolayer
deposition
for
the
gold
coated
substrate
was
recorded.
Secondly,
after
the
mono--‐
layer
deposition,
we
used
a
2--‐layer
model
(substrate/SAM)
to
fit
the
measured
data
and
to
determine
the
SAM
thick--‐
ness.
We
employed
the
previously
measured
optical
prop--‐
erties
of
the
gold
coated
substrate
(background),
and
we
fixed
the
refractive
index
of
the
organic
monolayer
at
1.50.
The
usual
values
in
the
literature
for
the
refractive
index
of
organic
monolayers
are
in
the
range
1.45−1.50.[31]
We
can
notice
that
a
change
from
1.50
to
1.55
would
result
in
less
than
1
Å
error
for
a
thickness
less
than
30
Å.
We
estimated
the
accuracy
of
the
SAM
thickness
measurements
at
±
2
Å.
IRRAS
measurements
were
performed
on
a
Bruker
Ver--‐
tex
70
FT--‐IR
spectrometer
equipped
with
a
high--‐sensitivity
Hg--‐Cd--‐Te
(MCT)
detector
and
an
A513/Q
variable
angle
reflection
accessory
including
an
automatic
rotational
holder
for
MIR
polarizer.
The
IR
beam
was
polarized
with
a
KRS--‐5
polarizer
with
99
%
degree
of
polarization.
Dou--‐
ble--‐sided
interferograms
were
collected
with
a
sample
fre--‐
quency
of
20
kHz,
an
aperture
of
1.5
mm
and
a
nominal
spectral
resolution
of
4
cm−1.
The
interferograms
were
apodized
by
a
Blackmann--‐Harris
3--‐term
apodization
and
zero--‐filled
with
a
zero--‐filling
factor
of
2.
The
angle
of
inci--‐
dence
was
set
to
80°,
and
p--‐polarized
IR
radiation
was
used
to
record
the
spectra.
For
the
background
measure--‐
ments,
the
sample
chamber
was
purged
with
argon
for
5
min,
then
1024
scans
were
collected
while
continuing
to
purge.
For
the
sample
measurements,
argon
purging
was
started
at
the
moment
the
first
scan
was
recorded.
The
scans
were
averaged
until
the
peaks
arising
from
the
water
vapor
in
the
sample
chamber
were
compensated,
for
what
typically
800−1500
scans
were
necessary.
Where
necessary,
scattering
correction
was
applied
to
the
spectra.
General
procedure
for
the
preparation
of
Au
substrates
for
IRRAS:
Au
substrates
were
fabricated
by
sputtering
a
10
nm
adhesive
film
of
Ti
and
a
100
nm
thick
layer
of
Au
on
<100>
oriented
silicon
wafers
with
a
native
SiO2
layer.
The
freshly
prepared
Au
substrates
were
cleaned
in
oxygen
plasma
[p(O2)
=
0.4
mbar,
f
=
40
kHz
and
P
=
75
W]
for
4
min
immediately
prior
to
the
Na--‐2
deposition.
For
the
deposition,
a
low
concentrated
solution
(~
0.4
mmol)
of
the
Na--‐2
sample
was
prepared
using
ultrapure
water
with
a
conductivity
of
<
55
nS
cm−1.
The
Au
substrate
was
stored
for
24
h
in
this
solution,
then
washed
with
a
small
amount
of
ultrapure
water
and
dried
for
24
h
in
a
desiccator.
Cyclic
voltammograms
of
2
on
gold.
CV
experiments
were
performed
with
a
Modulab
potentiostat
from
So--‐
lartron
Analytical
and
a
classical
three--‐electrode
electro--‐
chemical
cell.
The
SAM
covered
AuTS
electrode
was
used
as
the
working
electrode
(WE).
The
counter
electrode
(CE)
was
a
platinum
wire
(0.5
mm)
and
Ag/AgCl
(saturated
KCl)
was
used
as
a
reference
electrode
(REF).
CV
curve
was
recorded
at
a
scan
rate
100
mV/s.
The
energy
position
(with
respect
to
vacuum)
of
the
LUMO
of
the
SAM
was
estimated
from
the
first
reduction
peak
Ered
by
ELUMO
=
–
(Ered
+
EREF/SHE)
–
4.24[32]
in
eV
with
EREF/SHE
=
0.196
V
for
the
Ag/AgCl
reference
electrode.
XPS
experiments
of
2
on
gold
were
performed
to
analyze
the
chemical
composition
of
the
obtained
layer
on
gold
and
to
detect
any
unremoved
contaminant.
We
used
a
Physical
Electronics
5600
spectrometer
fitted
in
an
UHV
chamber
with
a
residual
pressure
of
2×10–10
Torr.
High
res--‐
olution
spectra
were
recorded
with
a
monochromatic
Al
Kα
X--‐ray
source
(hν
=
1486.6
eV),
a
detection
angle
of
45°
as
referenced
to
the
sample
surface,
an
analyzer
entrance
slit
width
of
400
µm
and
with
an
analyzer
pass
energy
of
12
eV.
In
these
conditions,
the
overall
resolution
as
measured
from
the
full--‐width
half--‐maximum
(FWHM)
of
the
Ag
3d5/2
line
is
0.55
eV.
The
spectra
were
recalibrated
with
respect
to
the
C
1s
peak
at
284.8
eV.
Semi--‐quantitative
analysis
were
completed
after
standard
background
sub--‐
traction
according
to
Shirley's
method.[33]
Peaks
were
de--‐
composed
by
using
Voigt
functions
and
a
least--‐square
minimization
procedure
and
by
keeping
constant
the
Gaussian
and
Lorentzian
broadenings
for
each
component
of
a
given
peak.
XPS
experiments
of
the
reference
powder
sample
of
Na--‐2
were
carried
out
using
the
Specs
Phoibos--‐150
energy
analyzer
and
a
non--‐monochromatized
Al
Kα
X--‐ray
source,
with
the
overall
energy
resolution
of
0.9
eV.
The
energy
calibration
of
XPS
measurements
was
done
by
aligning
of
C1s
core
level
peak
to
284.8
eV.
The
spectra
were
analyzed
after
Shirley
background
subtraction,
with
exceptions
of
Co2p,
N1s
and
As3p+P2p
regions,
where
linear
background
was
used
instead.
Spectra
were
decomposed
using
Voigt--‐
function
shapes
and
fitting
employed
the
KolXPD
analysis
package.
Small
charging
effects
were
observed;
C1s
core--‐
level
exhibits
an
asymmetry
and
it
was
decomposed
into
9
two
symmetrical
Voigt
components
(287.1
and
288.5
eV).
The
energy
axis
of
measurements
was
left
unmodified.
C--‐AFM
measurements
and
TVS
analysis.
We
performed
current--‐voltage
measurements
by
C--‐AFM
in
ambient
air
(ICON,
Bruker),
using
a
PtIr--‐coated
tip
(tip
radius
of
cur--‐
vature
less
than
25
nm,
force
constant
in
the
range
0.17−0.2
N/m).
Placing
the
conducting
tips
at
a
stationary
point
contact
formed
nano--‐junctions.
A
square
grid
of
10×10
is
defined
with
a
lateral
step
of
2
nm.
At
each
point,
10
I--‐V
curves
are
acquired
(back
and
forth)
leading
to
the
mea--‐
surements
of
1000
I--‐V
traces.
Out
of
these
1000
I--‐V
traces,
some
were
removed
(main
causes:
no
current
--‐
bad
tip
contact,
noise
larger
than
average
current
--‐
tip
contact
fluctuations,
too
high
current,
e.g.
short--‐circuit
or
pin--‐hole
in
the
SAM,
inducing
saturation
of
the
current
preamplifi--‐
er)
leading
to
about
300
useful
I--‐V
traces
(exact
number
indicated
in
the
related
figures).
The
load
force
was
ad--‐
justed
in
the
range
20−30
nN
and
measured
by
force--‐dis--‐
tance
curves
with
the
controlling
software
of
the
ICON.
The
bias
was
applied
on
the
AuTS
substrate
and
the
tip
was
grounded
through
the
input
of
the
current
amplifier.
The
voltage
sweeps
(back
and
forth)
were
applied
from
0
to
1
V
and
then
from
0
to
−1
V.
The
I--‐V
curves
are
analyzed
by
the
TVS
(transient
voltage
spectroscopy)
method.[26]
In
brief,
the
I--‐V
data
are
plotted
as
ln(I/V2)
vs.
1/V.
A
minimum
in
this
curve
corresponds
to
a
transition
from
a
direct
tunneling
electron
transport
through
the
molecules
and
a
resonant
tunneling
via
a
fron--‐
tier
molecular
orbital
(LUMO
or
HOMO).
The
energy
po--‐
sition
ε0
of
the
orbital
involved
in
the
transport
mecha--‐
nism
with
respect
to
the
Fermi
energy
of
the
metal
elec--‐
trode
is
given
by:
where
e
is
the
electron
charge,
VT+
and
VT−
are
the
voltage
of
the
minima
of
the
TVS
plot
at
positive
and
negative
voltages,
respectively.[34]
CONCLUSIONS
In
summary,
we
suggest
a
general
strategy
for
functional--‐
ization
of
polyoxometalates
by
integration
of
organoar--‐
sonates
as
prosthetic
co--‐ligands
to
polyoxotungstate
units,
which
can
also
stabilize
polynuclear
magnetic
cores.
De--‐
pending
on
the
terminal
organoarsonate
residue,
this
can
enable
their
attachment
to
metallic
electrode
surfaces,
which
we
intend
to
explore
in
the
context
of
molecular
electronics
and
spintronics.
Two
corresponding
polyanions
with
external
phenyl
and
para--‐aminophenyl
groups
have
been
prepared
and
characterized
in
the
solid
state
and
in
solution.
The
derivative
comprising
terminal
amino
groups
was
anchored
to
an
Au
surface
and
the
thus--‐obtained
SAM
was
extensively
characterized
via
ellipsometry,
FT--‐IRRAS,
XPS,
cyclic
voltammetry,
and
C--‐AFM
measurements.
This
study
thereby
proves
the
recently
suggested
suitability
of
the
–NH2
functional
group
for
direct
binding
of
molecules
to
noble
metal
surfaces
(e.g.
Au0)
without
a
mercaptocar--‐
boxylate
link
commonly
used
for
this
purpose.
Electron
transport
measurements
by
C--‐AFM
show
a
relatively
large
dispersion
of
the
current
though
the
molecular
junctions,
corresponding
to
the
polyoxotungstate--‐centered
LUMO
orbitals
located
between
ca.
0.4
and
0.7
eV
above
the
Fer--‐
mi
energy
of
the
Au
electrode.
The
presence
of
the
second
amino
group
not
bound
to
the
metal
substrate
in
principle
can
also
be
used
for
post--‐functionalization
of
the
formed
SAM
that
we
plan
to
explore
in
follow--‐up
work.
ASSOCIATED
CONTENT
Supporting
Information.
Bond
valence
sum
values;
(ATR)--‐
IR,
Raman,
UV--‐Vis,
XPS
and
IRRAS
spectra;
TGA
curves,
fur--‐
ther
CV
details,
AFM
image,
current
histograms
and
TVS
plots
for
SAM
of
2
on
Au
surface,
packing
diagrams
and
crys--‐
tallographic
data
for
Na--‐1
and
Na--‐2
in
CIF
format.
This
mate--‐
rial
is
available
free
of
charge
via
the
Internet
at
http://pub--‐
s.acs.org
AUTHOR
INFORMATION
Corresponding
Authors
*
[email protected]‐aachen.de
*
[email protected]
Author
Contributions
The
manuscript
was
written
through
contributions
of
all
au--‐
thors.
All
authors
have
given
approval
to
the
final
version
of
the
manuscript.
Funding
Sources
EU
ERC
Starting
Grant
MOLSPINTRON,
no.
308051;
COST
Action
CM
1203
for
STSM
at
IEMN
Notes
The
authors
declare
no
competing
financial
interests.
ACKNOWLEDGMENT
We
gratefully
acknowledge
financial
support
by
Forschungszentrum
Jülich,
EU
ERC
Starting
Grant
308051
–
MOLSPINTRON
(P.K.)
and
COST
Action
CM
1203
(M.
S.).
We
thank
Brigitte
Jansen
for
TGA
measurements
and
Dr.
Volkmar
Hess
for
fruitful
discussions.
REFERENCES
(1)
See
for
example:
(a)
Sanvito,
S.;
Rocha,
A.
R.
J.
Comput.
Theor.
Nanosci.
2006,
3,
624--‐642;
(b)
Bogani,
L.;
Wernsdorfer,
W.
Nat.
Mater.
2008,
7,
179--‐188;
(c)
Osorio,
E.
A.;
Bjørnholm,
T.;
Lehn,
J.--‐M.;
Ruben,
M.;
van
der
Zant,
H.
S.
J.
J.
Phys.
Condens.
Matter
2008,
20,
374121
/
1--‐14;
(d)
Sanvito,
S.
Chem.
Soc.
Rev.
2011,
40,
3336--‐3355;
(e)
Fahrendorf,
S.;
Atodiresei,
N.;
Besson,
C.;
Ca--‐
ciuc,
V.;
Matthes,
F.;
Blügel,
S.;
Kögerler,
P.;
Bürgler,
D.
E.;
Schneider,
C.
M.
Nat.
Commun.
2013,
4,
2425/1--‐6;
(f)
Perrin,
M.
L.;
Burzurí,
E.;
van
der
Zant,
H.
S.
J.
Chem.
Soc.
Rev.
2015,
44,
902--‐919,
and
references
therein.
(2)
See
for
example:
(a)
Lehmann,
J.;
Gaita--‐Ariño,
A.;
Corona--‐
do,
E.;
Loss,
D.
Nat.
Nanotechnol.
2007,
2,
312--‐317;
(b)
Fang,
X.;
Kögerler,
P.
Chem.
Commun.
2008,
3396--‐3398;
(c)
Kortz,
U.;
Mül--‐
ler,
A.;
van
Slageren,
J.;
Schnack,
J.;
Dalal,
N.
S.;
Dressel,
M.
Co--‐
ord.
Chem.
Rev.
2009,
253,
2315--‐2327;
(d)
Kögerler,
P.;
Tsukerblat,
B.;
Müller,
A.
Dalton
Trans.
2010,
39,
21--‐36;
(e)
Clemente--‐Juan,
J.
M.;
Coronado,
E.;
Gaita--‐Ariño,
A.
Chem.
Soc.
Rev.
2012,
41,
7464--‐7478;
(f)
Song,
Y.
F.;
Tsunashima,
R.
Chem.
Soc.
Rev.
2012,
41,
7384--‐7402;
(g)
Izarova,
N.
V.;
Kögerler
P.
in
Trends
in
Polyoxo--‐
metalates
Research;
Ruhlmann,
L.,
Schaming,
D.,
Eds.;
Nova
Science
Publishers:
Hauppauge,
2015,
121--‐149;
(h)
Ji,
Y.
C.;
Huang,
10
L.
J.;
Hu,
J.;
Streb,
C.;
Song,
Y.
F.
Energy
Environ.
Sci.
2015,
8,
776--‐
789;
(i)
Busche,
C.;
Vilà--‐Nadal,
L.;
Yan,
J.;
Miras,
H.
N.;
Long,
D.--‐
L.;
Georgiev,
V.
P.;
Asenov,
A.;
Pedersen,
R.
H.;
Gadegaard,
N.;
Mirza,
M.
M.;
Paul,
D.
J.;
Poblet,
J.
M.;
Cronin,
L.
Nature
2014,
515,
545--‐549;
(j)
Palii,
A.;
Tsukerblat,
B.;
Clemente--‐Juan,
J.
M.;
Coronado,
E.
J.
Phys.
Chem.
C
2016,
120,
16994--‐17005;
(k)
Shiddiq,
M.;
Komijani,
D.;
Duan,
Y.;
Gaita--‐Ariño,
A.;
Coronado,
E.;
Hill,
S.
Nature
2016,
531,
348--‐351;
(l)
Linnenberg,
O.;
Moors,
M.;
Solé--‐
Daura,
A.;
López,
X.;
Bäumer,
C.;
Kentzinger,
E.;
Pyckhout--‐Hint--‐
zen,
W.;
Monakhov,
K.
Yu.
J.
Phys.
Chem.
C,
2017,
121,
10419–
10429.
(3)
(a)
Lissel,
F.;
Schwarz,
F.;
Blacque,
O.;
Riel,
H.;
Lörtscher,
E.;
Venkatesan,
K.;
Berke,
H.
J.
Am.
Chem.
Soc.
2014,
136,
14560--‐14569;
(b)
Monakhov,
K.
Yu.;
Moors,
M.;
Kögerler,
P.
Adv.
Inorg.
Chem.
2017,
69,
251–286,
and
references
therein.
(4)
See
for
example:
(a)
Dolbecq,
A.;
Dumas,
E.;
Mayer,
C.
R.;
Mialane,
P.
Chem.
Rev.
2010,
110,
6009--‐6048;
(b)
Proust,
A.;
Matt,
B.;
Villanneau,
R.;
Guillemot,
G.;
Gouzerh,
P.;
Izzet,
G.
Chem.
Soc.
Rev.
2012,
41,
7605--‐7622;
(c)
Santonia,
M.--‐P.;
Hanana,
G.
S.;
Hasenknopf,
B.
Coord.
Chem.
Rev.
2014,
281,
64--‐85,
and
refer--‐
ences
therein.
(5)
See
for
example:
(a)
El
Moll,
H.;
Dolbecq,
A.;
Marrot,
J.;
Rousseau,
G.;
Haouas,
M.;
Taulelle,
F.;
Rogez,
G.;
Wernsdorfer,
W.;
Keita,
B.;
Mialane,
P.
Chem.
Eur.
J.
2012,
18,
3845--‐3849;
(b)
El
Moll,
H.;
Zhu,
W.;
Oldfield,
E.;
Rodriguez--‐Albelo,
M.;
Mialane,
P.;
Marrot,
J.;
Vila,
N.;
Mbomekallé,
I.
M.;
Rivière,
E.;
Duboc,
C.;
Dolbecq,
A.
Inorg.
Chem.
2012,
51,
7921--‐7931;
(c)
Rousseau,
G.;
Rivière,
E.;
Dolbecq,
A.;
Marrot,
J.;
Oms,
O.;
Mialane,
P.
Eur.
J.
Inorg.
Chem.
2013,
1793--‐1798;
(d)
Saad,
A.;
Zhu,
W.;
Rousseau,
G.;
Mialane,
P.;
Marrot,
J.;
Haouas,
M.;
Taulelle,
F.;
Dessapt,
R.;
Se--‐
rier--‐Brault,
H.;
Rivière,
E.;
Kubo,
T.;
Oldfield,
E.;
Dolbecq,
A.
Chem.
Eur.
J.
2015,
21,
1--‐12;
(e)
Saad,
A.;
Anwar,
N.;
Rousseau,
G.;
Mialane,
P.;
Marrot,
J.;
Haouas,
M.;
Taulelle,
F.;
Mc
Cormac,
T.;
Dolbecq,
A.
Eur.
J.
Inorg.
Chem.
2015,
4775--‐4782;
(f)
Xue,
H.;
Zhao,
J.--‐W.;
Pan,
R.;
Yang,
B.--‐F.;
Yang,
G.--‐Y.;
Liu,
H.--‐S.
Chem.
Eur.
J.
2016,
22,
12322--‐12331;
(g)
Ban,
R.;
Sun,
X.;
Wang,
J.;
Ma,
P.;
Zhang,
C.;
Niu,
J.;
Wang.,
J.
Dalton
Trans.
2017,
46,
5856--‐5863.
(6)
See
for
example:
(a)
Joo,
N.;
Renaudineau,
S.;
Delapierre,
G.;
Bidan,
G.;
Chamoreau,
L.
M.;
Thouvenot,
R.;
Gouzerh,
P.;
Proust,
A.
Chem.
Eur.
J.
2010,
16,
5043--‐5051;
(b)
Musumeci,
C.;
Luzio,
A.;
Pradeep,
C.
P.;
Miras,
H.
N.;
Rosnes,
M.
H.;
Song,
Y.--‐F.;
Long,
D.--‐L.;
Cronin,
L.;
Pignataro,
B.
J.
Phys.
Chem.
C
2011,
115,
4446--‐4455;
(c)
Mercier,
D.;
Boujday,
S.;
Annabi,
C.;
Villanneau,
R.;
Pradier,
C.--‐M.;
Proust,
A.
J.
Phys.
Chem.
C
2012,
116,
13217--‐13224;
(d)
Rinfray,
C.;
Izzet,
G.;
Pinson,
J.;
Gam
Derouich,
S.;
Combellas,
C.;
Kanoufi,
F.;
Proust,
A.
Chem.
Eur.
J.
2013,
19,
13838--‐13846;
(e)
Derouich,
S.
G.;
Rinfray,
C.;
Izzet,
G.;
Pinson,
J.;
Gallet,
J.--‐J.;
Kanoufi,
F.;
Proust,
A.;
Combellas,
C.
Langmuir
2014,
30,
2287--‐2296;
(f)
Yvon,
C.;
Surman,
A.
J.;
Hutin,
M.;
Alex,
J.;
Smith,
B.
O.;
Long,
D.--‐L.;
Cronin,
L.
Angew.
Chem.
Int.
Ed.
2014,
126,
3404--‐3409;
(g)
Volatron,
F.;
Noël,
J.--‐M.;
Rinfray,
C.;
Decorse,
P.;
Combellas,
C.;
Kanoufi,
F.;
Proust,
A.
J.
Mater.
Chem.
C
2015,
3,
6266--‐6275;
(h)
Lombana,
A.;
Rinfray,
C.;
Volatron,
F.;
Izzet,
G.;
Battaglini,
N.;
Alves,
S.;
Decorse,
P.;
Lang,
P.;
Proust,
A.
J.
Phys.
Chem.
C
2016,
120,
2837--‐2845.
(7)
See
for
example:
(a)
Weakley,
T.
J.
R.
J.
Chem.
Soc.
Chem.
Commun.
1984,
1406--‐1407;
(b)
Gálán--‐Mascarós,
J.
R.;
Gómez--‐
García,
C.
J.;
Borrás--‐Almenar,
J.
J.;
Coronado,
E.
Adv.
Mater.
1994,
6,
221--‐223;
(c)
Clemente--‐Juan,
J.
M.;
Coronado,
E.;
Gálán--‐Mas--‐
carós,
J.
R.;
Gómez--‐García,
C.
J.
Inorg.
Chem.
1999,
38,
55--‐63;
(d)
Ritchie,
C.;
Boyd,
T.;
Long,
D.--‐L.;
Ditzel,
E.;
Cronin,
L.
Dalton
Trans.
2009,
1587--‐1592;
(e)
Lydon,
C.;
Sabi,
M.
M.;
Symes,
M.
D.;
Long,
D.--‐L.;
Murrie,
M.;
Yoshii,
S.;
Nojirib,
H.;
Cronin,
L.
Chem.
Commun.
2012,
48,
9819--‐9821.
(8)
(a)
Quek,
S.
Y.;
Venkataraman,
L.;
Choi,
H.
J.;
Louie,
S.
G.;
Hybertsen,
M.
S.;
Neaton,
J.
B.
Nano
Lett.
2007,
7,
3477--‐3482;
(b)
de
la
Llave,
E.;
Clarenc,
R.;
Schiffrin,
D.
J.;
Williams,
F.
J.
J.
Phys.
Chem.
C
2014,
118,
468--‐475;
(c)
Koo,
K.
M.;
Sina,
A.
A.
I.;
Carras--‐
cosa,
L.
G.;
Shiddiky,
M.
J.
A.;
Trau,
M.
Anal.
Methods
2015,
7,
7042--‐7054.
(9)
Contant,
R.
Inorg.
Synth.
1990,
27,
106--‐111.
(10)
Finke
R.
G.;
Droege,
M.
W.
Inorg.
Chem.
1983,
22,
1006--‐1008.
(11)
(a)
Goberna--‐Ferrón,
S.;
Vigara,
L.;
Soriano--‐López,
J.;
Gálán--‐Mascarós,
J.
R.
Inorg.
Chem.
2012,
51,
11707--‐11715;
(b)
Soria--‐
no--‐López,
J.;
Goberna--‐Ferrón,
S.;
Vigara,
L.;
Carbó,
J.
J.;
Poblet,
J.
M.;
Galán--‐Mascarós,
J.
R.
Inorg.
Chem.
2013,
52,
4753--‐4755;
(c)
Goberna--‐Ferrón,
S.;
Soriano--‐López,
J.;
Gálán--‐Mascarós,
J.
R.;
Nyman,
M.
Eur.
J.
Inorg.
Chem.
2015,
2833--‐2840;
(d)
Yin,
Q.;
Tan,
J.
M.;
Besson,
C.;
Geletii,
Y.
V.;
Musaev,
D.
G.;
Kuznetsov,
A.
E.;
Luo,
Z.;
Hardcastle,
K.
I.;
Hill,
C.
L.
Science
2010,
328,
342--‐345;
(e)
Zhu,
G.;
Geletii,
Y.
V.;
Kögerler,
P.;
Schilder,
H.;
Song,
J.;
Lense,
S.;
Zhao,
C.;
Hardcastle,
K.
I.;
Musaev,
D.
G.;
Hill,
C.
L.
Dalton
Trans.
2012,
41,
2084--‐2090;
(f)
Lv,
H.;
Song,
J.;
Geletii,
Y.
V.;
Vickers,
J.
W.;
Sumliner,
J.
M.;
Musaev,
D.
G.;
Kögerler,
P.;
Zhuk,
P.;
Bacsa,
J.;
Zhu,
G.;
Hill,
C.
L.
J.
Am.
Chem.
Soc.
2014,
136,
9268--‐9271;
(g)
Soriano--‐López,
J.;
Musaev,
D.
G.;
Hill,
C.
L.;
Galán--‐Mascarós,
J.
R.;
Carbo,
J.
J.;
Poblet,
J.
M.
J.
Catal.
2017,
350,
56--‐63;
(h)
Song,
F.;
Ding,
Y.;
Ma,
B.;
Wang,
C.;
Wang,
Q.;
Du,
X.;
Fua,
S.;
Song,
J.
Energy
Environ.
Sci.
2013,
6,
1170--‐1184.
(12)
Unit
cell
for
NaxH25–y[Co9(H2O)6(OH)3(HPO4)2(α--‐
P2W15O56)3]·∙nH2O
(Na--‐4):
Triclinic,
P–1,
a
=
13.9486(4)
Å,
b
=
29.2148(7)
Å,
c
=
31.7155(8)
Å,
α
=
74.603(2)°,
β
=
81.760(2)°,
γ
=
89.949(2)°,
V
=
12320.2(5)
Å3,
Z
=
2.
(13)
Barsukova,
M.;
Izarova,
N.
V.;
Ngo
Biboum,
R.;
Keita,
B.;
Nadjo,
L.;
Ramachandran,
V.;
Dalal,
N.
S.;
Antonova,
N.
S.;
Carbó,
J.
J.;
Poblet,
J.
M.;
Kortz,
U.
Chem.
Eur.
J.
2010,
16,
9076--‐9085.
(14)
Lueken,
H.
Magnetochemie,
Teubner,
Stuttgart,
1999.
(15)
(a)
Ballhausen,
C.
Introduction
to
Ligand--‐Field
Theory,
McGraw--‐Hill,
New
York,
1962;
(b)
Figgs,
N.
B.;
Hitchman,
M.
A.
Ligand--‐Field
Theory
and
its
Applications,
Wiley--‐VCH,
New
York,
2000.
(16)
See
for
example:
(a)
Ruhlmann,
L.;
Nadjo,
L.;
Canny,
J.;
Contant,
R.;
Thouvenot,
R.
Eur.
J.
Inorg.
Chem.
2002,
975--‐986;
(b)
Lisnard,
L.;
Mialane,
P.;
Dolbecq,
A.;
Marrot,
J.;
Clemente--‐Juan,
J.
M.;
Coronado,
E.;
Keita,
B.;
de
Oliveira,
P.;
Nadjo,
L.;
Sécheresse,
F.
Chem.
Eur.
J.
2007,
13,
3525--‐3536;
(c)
Ruhlmann,
L.;
Schaming,
D.;
Ahmed,
I.;
Courville,
A.;
Canny,
J.;
Thouvenot,
R.
Inorg.
Chem.
2012,
51,
8202--‐8211;
(d)
Duan,
Y.;
Clemente--‐Juan,
J.
M.;
Giménez--‐
Saiz,
C.;
Coronado,
E.
Inorg.
Chem.
2016,
55,
925--‐938.
(17)
(a)
Weiss,
E.;
Chiechi,
R.;
Kaufman,
G.;
Kriebel,
J.;
Li,
Z.;
Duati,
M.;
Rampi,
M.;
Whitesides,
G.
J.
Am.
Chem.
Soc.
2007,
129,
4336--‐4349.
(18)
Smekal,
W.;
Werner,
W.
S.
M.;
Powell,
C.
J.
Surf.
Interf.
Anal.
2005,
37,
1059.
(19)
Yeh
J.
J.;
Lindau,
I.
At.
Data
Nucl,
Data
Tables,
1985,
32,
1--‐155.
1994,
21,
165.
11,
1215--‐1220.
(20)
Tanuma,
S.;
Powell,
C.
J.;
Penn,
D.
R.
Surf.
Interf.
Anal.
(21)
Allara,
D.
L.;
Baca,
A.;
Pryde,
C.
A.
Macromolecules
1978,
(22)
(a)
Engelkes,
V.
B.;
Beebe,
J.
M.;
Frisbie,
C.
D.
J.
Phys.
Chem.
B
2005,
109,
16801--‐16810;
(b)
Kim,
T.--‐W.;
Wang,
G.;
Lee,
H.;
Lee,
T.
Nanotechnology
2007,
18,
315204.
(23)
Smaali,
K.;
Lenfant,
S.;
Karpe,
S.;
Oçafrain,
M.;
Blanchard,
P.;
Deresmes,
D.;
Godey,
S.;
Rochefort,
A.;
Roncali,
J.;
Vuillaume,
D.
ACS
Nano
2010,
4,
2411--‐2421.
(24)
Smaali,
K.;
Clément,
N.;
Patriarche,
G.;
Vuillaume,
D.
ACS
Nano
2012,
6,
4639--‐4647.
(25)
(a)
Reuter,
M.
G.;
Hersam,
M.
C.;
Seideman,
T.;
Ratner,
M.
A.
Nano
Lett.
2012,
12,
2243--‐2248;
(b)
Trasobares,
J.;
Rech,
J.;
Jon--‐
ckheere,
T.;
Martin,
T.;
Aleveque,
O.;
Levillain,
E.;
Diez--‐Cabanes,
V.;
Olivier,
Y.;
Cornil,
J.;
Nys,
J.
P.;
Sivakumarasamy,
R.;
Smaali,
K.;
Leclere,
P.;
Fujiwara,
A.;
Théron,
D.;
Vuillaume,
D.;
Clément,
N.
Nano
Lett.
2017,
17,
3215–3224.
(26)
(a)
Beebe,
J.
M.;
Kim,
B.;
Gadzuk,
J.
W.;
Frisbie,
C.
D.;
Kushmerick,
J.
G.
Phys.
Rev.
Lett.
2006,
97,
026801;
(b)
Ricoeur,
G.;
Lenfant,
S.;
Guérin,
D.;
Vuillaume,
D.
J.
Phys.
Chem.
C
2012,
116,
20722--‐20730.
(27)
Contant,
R.
Inorg.
Synth.
1990,
27,
105--‐106.
11
(28)
The
0.66
M
CH3COONa
buffer
(pH
5.2)
was
prepared
by
dissolving
41
g
of
CH3COONa
and
9.14
mL
of
glacial
CH3COOH
in
1000
mL
of
H2O.
(29)
CrysAlisPro,
Agilent
Technologies,
1.171.36.28
(release
G.
C.
Adv.
Mater.
2011,
23,
2367--‐2371.
(33)
Shirley,
D.
A.
Phys.
Rev.
B
1972,
5,
4709--‐4714.
(34)
Bâldea,
I.
Phys.
Rev.
B
2012,
85,
035442.
01--‐02--‐2013
CrysAlis171
.NET).
(30)
Sheldrick,
G.
M.
Acta
Cryst.
2008,
A64,
112--‐122.
(31)
(a)
Ulman,
A.
An
Introduction
to
Ultrathin
Organic
Films:
From
Langmuir--‐Blodgett
to
Self--‐assembly,
Academic
Press,
Bos--‐
ton,
1991;
(b)
Parikh,
A.
N.;
Allara,
D.
L.;
Ben
Azouz,
I.;
Rondelez,
F.
J.
Phys.
Chem.
1994,
98,
7577--‐7590.
(32)
Cardona,
C.
M.;
Li,
W.;
Kaifer,
A.
E.;
Stockdale,
D.;
Bazan,
12
13
SYNOPSIS
TOC
14
SUPPLEMENTARY INFORMATION
Probing frontier orbital energies of {Co9(P2W15)3}
polyoxometalate clusters at molecule–metal and
molecule–water interfaces
Xiaofeng Yi,†,‡ Natalya V. Izarova,† Maria Stuckart,†,‡ David Guérin,§ Louis Thomas,§
Stéphane Lenfant,§ Dominique Vuillaume,§,* Jan van Leusen,‡ Tomáš Duchoň, Slavomír
Nemšák,†,¶ Svenja D. M. Bourone,‡ Sebastian Schmitz ‡ and Paul Kögerler †,‡,*
† Jülich-Aachen Research Alliance (JARA-FIT) and Peter Grünberg Institute 6,
Forschungszentrum Jülich, D-52425 Jülich, Germany
‡ Institute of Inorganic Chemistry, RWTH Aachen University, D-52074 Aachen, Germany
§ Institute of Electronics, Microelectronics and Nanotechnology, CNRS, University of Lille,
59652 Villeneuve d'Ascq, France
Faculty of Mathematics and Physics, Charles University, 18000 Prague, Czech Republic
¶ BESSY-II, Helmholtz Zentrum Berlin, D-12489 Berlin, Germany
S1
Content
I. Bond valence sum calculations …………………………………………………….…….p. S3
II. Vibrational spectra ………………………………………………………….………...... p. S9
III. Thermogravimetrical analysis ………………………………………………………..... p. S10
IV. UV-Vis spectroscopy measurements ………………………………………….……..... p. S11
V. Cyclic voltammetry…………………………………......……………………………..... p. S13
VI. Self-assembled monolayer experiments………………………………………………. p. S16
VII. Crystal packing of 1 in compound Na-1……………...…………………………….. p. S23
VIII. Crystal packing of 2 in compound Na-2……………...…………………………….. p. S26
S2
I. Bond valence sum calculations
Table S1. Bond valence sum values for different atoms in Na-1
W, P, Co, and As centers W, P, Co, and As centers
Terminal oxygens of
W1
W2
W3
W4
W5
W6
W7
W8
W9
W10
W11
W12
W13
W14
W15
W16
W17
W18
W19
W20
W21
W22
W23
W24
W25
W26
W27
W28
W29
W30
W31
W32
W33
W34
W35
W36
W37
W38
W39
W40
6.32
6.35
6.28
6.27
6.29
6.27
6.23
6.31
6.34
6.30
6.33
6.11
6.26
6.22
6.17
6.24
6.29
6.14
6.23
6.22
6.30
6.49
6.37
6.13
6.06
6.35
6.04
6.17
6.21
6.33
6.28
6.10
6.28
6.23
6.13
6.23
6.34
6.24
6.37
6.18
W41
W42
W43
W44
W45
P1
P2
P3
P4
P5
P6
Co1
Co2
Co3
Co4
Co5
Co6
Co7
Co8
Co9
As1
As2
6.26
6.22
6.03
6.17
6.21
4.71
4.73
4.63
4.85
4.72
4.73
1.93
2.00
2.03
1.99
1.95
2.04
2.04
1.97
2.00
5.11
5.26
Terminal oxygens of
W=O
O1T
O2T
O3T
O4T
O5T
O6T
O7T
O8T
O9T
O10T
O11T
O12T
O13T
O14T
O15T
O16T
–1.77
–1.64
–1.82
–1.63
–1.69
–1.66
–1.72
–1.91
–1.77
–1.97
–1.71
–1.66
–1.87
–1.63
–1.66
–1.73
W=O
O17T
O18T
O19T
O20T
O21T
O22T
O23T
O24T
O25T
O26T
O27T
O28T
O29T
O30T
O31T
O32T
O33T
O34T
O35T
O36T
O37T
O38T
O39T
O40T
O41T
O42T
O43T
O44T
O45T
–1.77
–1.77
–1.69
–1.71
–1.90
–1.84
–1.91
–1.66
–1.60
–1.81
–1.60
–1.96
–1.73
–1.76
–1.73
–1.92
–1.67
–1.76
–1.77
–1.90
–1.67
–1.76
–1.74
–1.72
–1.66
–1.79
–1.62
–1.65
–1.63
Terminal aqua–ligands
of Co–OH2O
O2C
O3C
O4C
O5C
O7C
O9C
–0.36
–0.30
–0.30
–0.36
–0.35
–0.35
S3
Table S1. Bond valence sum values for different atoms in Na-1 (continuation)
µµµµ2-O (W–O–Co)
µµµµ2-O (W–O–W)
µµµµ2-O (W–O–W)
O0C1
O5C1
O1C2
O2C2
O3C3
O4C3
O0C4
O5C4
O6C5
O7C5
O8C6
O9C6
O0C7
O1C7
O3C8
O2C8
O4C9
O5C9
–1.86
–1.95
–2.03
–1.92
–1.86
–1.93
–1.98
–1.81
–1.90
–1.90
–1.89
–1.90
–1.92
–2.06
–1.88
–1.92
–1.89
–1.91
µµµµ2-O (W–O–W)
O12
O13
O14
O19
O23
O25
O26
O37
O38
O45
O49
O56
O67
O78
O89
O101
O105
O112
O123
O134
O145
−2.01
−1.95
−2.07
−1.95
−1.88
−1.99
−1.96
−1.93
−1.96
−1.98
−2.19
−2.20
−1.99
−2.21
−2.03
−1.92
−2.12
−2.07
−1.97
−2.16
−1.95
O164
O167
O168
O169
O170
O171
O178
O182
O183
O190
O194
O195
O201
O206
O212
O217
O223
O228
O234
O239
O240
O250
O256
O267
O278
O289
O290
O312
O313
O314
O315
O323
O326
O327
O338
O339
O340
O345
O349
O351
−1.95
−1.89
−1.88
−1.98
−1.96
−1.95
−1.90
−2.01
−1.96
−2.00
−2.20
−2.03
−2.22
−2.05
−1.95
−2.03
−2.18
−2.05
−1.99
−2.09
−2.07
−2.07
−1.95
−2.13
−1.93
−2.07
−2.00
−2.00
−1.92
−2.01
−2.07
−1.94
−2.06
−2.01
−2.05
−1.99
−2.06
−2.21
−2.01
−2.05
O356
O362
O367
O373
O378
O384
O389
O395
O401
O405
O410
O412
O423
O434
O445
O511
O612
O713
O814
O915
−2.05
−1.95
−2.22
−2.02
−2.09
−1.99
−2.22
−2.07
−2.07
−1.92
−2.05
−1.85
−2.02
−1.96
−2.16
−2.09
−2.00
−2.05
−2.05
−2.09
µµµµ3-O (2 W + P)
O1P2
O1P4
O1P6
O2P1
O2P2
O2P3
O2P4
O2P5
O2P6
O3P1
O3P2
O3P3
O3P5
O3P4
O3P6
O4P1
O4P3
O4P5
−1.86
−1.87
−1.88
−1.88
−1.85
−1.89
−1.86
−1.81
−1.82
−1.81
−1.84
−1.82
−1.89
−1.90
−1.81
−1.86
−1.82
−1.84
S4
Table S1. Bond valence sum values for different atoms in Na-1 (continuation)
µµµµ3-O (2 Co + As)
O1A1
O1A2
O2A1
O2A2
O3A1
O3A2
−2.15
−2.03
−2.06
−1.99
−2.00
−2.06
µµµµ3-O (3 Co)
O1
O2
O3
µµµµ4-O (3 Co + P)
−1.00
−1.06
−1.05
O6P4
O4P4
O4P2
−1.83
−1.83
−1.82
µµµµ4-O (3 W + P)
O1P1
−1.91
O1P3
−1.88
O1P5
−1.89
S5
Table S2 Bond valence sum values for different atoms in Na-2
W, P, Co, and As centers W, P, Co, and As centers
Terminal oxygens of
W1
W2
W3
W4
W5
W6
W7
W8
W9
W10
W11
W12
W13
W14
W15
W16
W17
W18
W19
W20
W21
W22
W23
W24
W25
W26
W27
W28
W29
W30
W31
W32
W33
W34
W35
W36
W37
W38
W39
W40
6.05
6.12
6.16
6.19
6.29
6.16
6.20
6.31
6.15
6.29
6.30
6.19
6.26
6.18
6.18
6.26
6.20
6.22
6.25
6.26
6.19
6.24
6.32
6.32
6.19
6.17
6.20
6.36
6.06
6.23
6.12
6.09
6.31
6.15
6.29
6.42
6.16
6.31
6.38
6.12
W42
W43
W44
W45
W42
P1
P2
P3
P4
P5
P6
Co1
Co2
Co3
Co4
Co5
Co6
Co7
Co8
Co9
As1
As2
6.08
6.13
6.09
6.32
6.08
4.79
4.66
4.74
4.60
4.78
4.56
1.95
2.07
1.96
1.93
1.96
2.01
1.97
2.02
2.05
4.93
4.99
Terminal oxygens of
W=O
O1T
O2T
O3T
O4T
O5T
O6T
O7T
O8T
O9T
O10T
O11T
O12T
O13T
O14T
O15T
O16T
−1.87
−1.89
−1.97
−1.69
−1.79
−1.78
−1.96
−1.69
−1.70
−1.67
−1.77
−1.65
−1.62
−1.72
−1.77
−1.63
W=O
O17T
O18T
O19T
O20T
O21T
O22T
O23T
O24T
O25T
O26T
O27T
O28T
O29T
O30T
O31T
O32T
O33T
O34T
O35T
O36T
O37T
O38T
O39T
O40T
O41T
O42T
O43T
O44T
O45T
−1.70
−1.72
−1.66
−1.64
−1.57
−1.95
−1.84
−1.76
−1.61
−1.84
−1.77
−1.98
−1.66
−1.64
−1.75
−1.62
−1.94
−1.73
−2.06
−1.74
−1.66
−1.89
−1.74
−1.87
−1.59
−1.86
−1.84
−1.57
−1.72
Terminal aqua–ligands
of Co–OH2O
O1CH
O3CH
O5CH
O6CH
O8CH
−0.36
−0.33
−0.31
−0.33
−0.35
O9CH
−0.33
S6
Table S2 Bond valence sum values for different atoms in Na-2 (continuation)
µµµµ2-O (W–O–Co)
µµµµ2-O (W–O–W)
µµµµ2-O (W–O–W)
O0C1
O1C1
O2C2
O3C2
O4C3
O5C3
O5C4
O0C4
O6C5
O7C5
O8C6
O9C6
O5C7
O0C7
O2C8
O1C8
O3C9
O4C9
−1.92
−1.86
−1.96
−1.90
−1.85
−1.92
−2.01
−2.06
−2.21
−2.13
−2.15
−2.17
−1.89
−1.82
−1.84
−1.93
−1.93
−1.90
µµµµ2-O (W–O–W)
O167
O168
O178
O12
O13
O14
O15
O23
O26
O27
O38
O39
O45
O49
O56
O67
O78
O89
O101
O105
O112
−1.88
−1.81
−1.80
−1.88
−1.90
−2.05
−1.96
−1.99
−1.95
−2.00
−2.04
−2.04
−2.17
−2.04
−2.13
−2.19
−2.05
−2.21
−2.11
−1.94
−2.00
O123
O134
O145
O164
O169
O170
O171
O182
O183
O190
O194
O195
O201
O206
O212
O217
O223
O228
O234
O239
O240
O250
O256
O267
O278
O289
O290
O312
O313
O314
O319
O323
O325
O326
O337
O338
O340
O345
O349
O351
−2.13
−2.01
−2.10
−1.95
−1.99
−1.87
−2.57
−2.55
−1.91
−1.97
−2.16
−2.23
−2.27
−2.15
−1.93
−2.47
−2.22
−2.47
−1.99
−2.29
−2.20
−2.07
−1.95
−2.07
−2.01
−2.03
−1.88
−1.95
−1.93
−2.02
−2.02
−1.90
−2.00
−1.93
−1.97
−1.89
−2.02
−2.17
−2.15
−2.01
O356
O362
O367
O373
O378
O384
O389
O395
O401
O405
O410
O412
O423
O434
O445
O511
O612
O713
O814
O915
−2.21
−2.06
−1.97
−2.06
−2.22
−2.01
−2.04
−2.07
−1.96
−2.09
−2.00
−2.04
−1.93
−2.07
−2.05
−2.08
−1.98
−2.02
−2.01
−1.99
µµµµ3-O (2 W + P)
O1P2
O1P4
O1P6
O2P1
O2P2
O2P3
O2P4
O2P5
O2P6
O3P1
O3P2
O3P3
O3P4
O3P5
O3P6
O4P1
O4P3
O4P5
−1.85
−1.85
−1.83
−1.82
−1.83
−1.86
−1.80
−1.86
−1.81
−1.87
−1.79
−1.85
−1.83
−1.86
−1.83
−1.84
−1.86
−1.84
S7
Table S2 Bond valence sum values for different atoms in Na-2 (continuation)
µµµµ3-O (2 Co + As)
O1A1
O2A1
O3A1
O1A2
O3A2
O2A2
−2.02
−2.08
−1.98
−2.10
−2.00
−2.08
µµµµ3-O (3 Co)
O1
O2
O3
µµµµ4-O (3 Co + P)
−1.05
−1.03
−1.03
O4P2
O4P4
O4P6
−1.82
−1.76
−1.78
µµµµ4-O (3 W + P)
O1P1
−1.91
O1P3
−1.99
O1P5
−1.91
S8
Figure S1. FT-IR spectra of Na-1 (purple) and Na-2 (blue) in comparison to that of the Na12[α-
P2W15O56]·24H2O precursor (black).
Figure S2. Raman spectra of Na-1 (purple) and Na-2 (blue) in comparison to that of the Na12[α-
P2W15O56]·24H2O precursor (black).
S9
Figure S3 ATR-FTIR spectra of Na-2 in the solid state (black) and saturated aqueous solution
(blue).
S10
Figure S4. TGA (blue), differential TGA (green) and DTA (black) curves for Na-1 from room
temperature to 900 °C under N2 atmosphere.
Figure S5. TGA (blue), differential TGA (green) and DTA (black) curves for Na-2 from room
temperature to 900 °C under N2 atmosphere.
S11
Figure S6. Room-temperature UV-Vis spectrum of Na-1 solution in H2O (ε values are averaged
from the spectra of the solutions with concentrations between 6.4⋅10–6 M and 9.6⋅10–6 M for the
UV region and 2.8⋅10–4 M for the visible light region).
Figure S7. Room-temperature UV-Vis spectrum of Na-1 solution in 0.5 M NaCH3COO aqueous
medium at pH 5.1 (ε values are averaged from the spectra of the solutions with concentrations
between 2.3⋅10–8 M and 5.4⋅10–7 M for the UV region and between 6.7⋅10–5 M and 2.8⋅10–4 M for
the visible light region).
S12
Figure S8. Room-temperature UV-Vis spectrum of Na-2 solution in H2O (ε values are averaged
from the spectra of the solutions with concentrations between 1.1⋅10–6 M and 2.2⋅10–6 M).
Figure S9. Time-dependent room-temperature UV-Vis spectrum of a 2.8⋅10–4 M Na-1 solution in
H2O.
S13
Figure S10. Comparison of cyclic voltammograms of 0.7 mM Na-1 (black), Na-2 (red) and Na-4
(blue) solutions in 0.5 M CH3COONa buffer (pH 4.8), scan rate 20 mV / s.
Figure S11. Comparison of cyclic voltammograms of Na-1 (black) and Na-3 (blue) 0.7 – 1 mM
solutions in 0.5 M CH3COONa buffer (pH 4.8), scan rate 20 mV / s.
S14
Figure S12. Cyclic voltammograms of 0.7 mM Na-2 solutions in aqueous 0.5 M CH3COONa
media with pH 4.8 (black) and pH 6.4 (blue); scan rate 20 mV / s.
Figure S13. Cyclic voltammograms of 0.7 mM Na-1 solutions in 0.5 M CH3COONa buffer with
pH 4.8: freshly prepared solution (black) and the solution after one day (red). A comparison with
the CV of Na-3 solution in the same medium.
S15
Figure S14. Representation of possible orientation of 2 on the gold surface.
Figure S15. AFM image (a) and height profile corresponding to the blue line in the Fig. S15a (b)
of Na-2 on the gold surface.
Figure S16. CV curves of SAM of 2 on gold surface (black line) with 0.66 M NaOAc/HOAc
solution with pH 5.2 solution as an electrolyte and Na-2 solution in the same medium at 100
mV/s.
S16
Figure S17. XPS spectra of Na-2 SAM on a template stripped gold surface (black line; 10–3 M in
DI water, 24h) and of a film of Na-2 crystals on Au (red line; a drop of a concentrated solution of
Na-2 in deionized water was evaporated on a gold surface).
S17
y
t
i
s
n
e
t
n
I
S
P
X
d
e
z
i
l
a
m
r
o
N
y
t
i
s
n
e
n
t
I
S
P
X
d
e
z
i
l
a
m
r
o
N
y
t
i
s
n
e
n
t
I
S
P
X
d
e
z
i
l
a
m
r
o
N
O1s
W4p1/2
W4p3/2
As LMM
C1s
W4d
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0.00
y
t
i
s
n
e
n
t
I
S
P
X
d
e
z
As3p + P2p
W4f
i
l
a
m
r
o
N
Na1s
1.0
0.8
0.6
0.4
0.2
0.0
600
500
400
300
200
100
0
1080
1078
1076
1074
1072
1070
Binding Energy (eV)
Binding Energy (eV)
250
Co2p
200
150
100
50
0
805 800 795 790 785 780 775
y
t
i
s
n
e
n
t
I
S
P
X
d
e
z
i
l
a
m
r
o
N
O1s
1.0
0.8
0.6
0.4
0.2
0.0
540
538
536
534
532
530
528
Binding Energy (eV)
Binding Energy (eV)
-3
22x10
y
t
i
s
n
e
n
t
I
S
P
X
d
e
z
i
l
a
m
r
o
N
As3p
P2p
P2s
21
20
19
18
17
16
15
C1s
1.0
0.8
0.6
0.4
0.2
0.0
y
t
i
s
n
e
n
t
I
S
P
X
d
e
z
i
l
a
m
r
o
N
y
t
i
s
n
e
n
t
I
S
P
X
d
e
z
i
l
a
m
r
o
N
N1s
408
404
400
396
Binding Energy (eV)
W4f
1.0
0.8
0.6
0.4
0.2
0.0
1.0
0.8
0.6
0.4
0.2
0.0
292
288
284
Binding Energy (eV)
200
180
160
140
120
46
44
42
40
38
36
34
32
Binding Energy (eV)
Binding Energy (eV)
Figure S18. XPS spectra of Na-2 powder.
S18
Figure S19. FT-IRRAS spectrum of layer obtained after overnight exposure of an Au plate to an
aqueous solution of Na-2.
Figure S20. Current histograms at a given bias: +0.9V (a) and –0.9 V (b), as determined from I-
V data in Fig. 4 (main text). The histograms are fitted by two log normal distributions with the
following parameters: mean current/standard deviation –10.74 (i.e. 1.82⋅10–11 A) /0.37 and –9.8
(1.6⋅10–10 A)/0.49 at 0.9 V, –10.72 (1.9⋅10–11 A)/0.41 and –9.67 (2.1⋅10–10 A) /0.33 at –0.9V.
S19
Figure S21. Typical TVS plots of the I-V curves at positive (a) and negative (b) voltages (red
lines are guides to the eye) and histograms of the transition voltage VT+ (67 counts) and VT- (38
counts) at positive (c) and negative (d) voltages, respectively (I-V data from Fig. 4, the counts for
TVS is lower than the number of I-V curves because some I-V curves did not clearly show any
minima in the TVS plot, too noisy curves). The VT absolute values are dispersed between ca. 0.4
and 0.9 V, centered at ca. 0.65 V given a position of the LUMO (see Eq. in Experimental part,
main text) at 0.56 eV ± 0.17 eV. VT histograms are fitted with a Gaussian peak with the
parameters: mean voltage/standard deviation of 0.64 V/0.15 V and –0.7 V/0.10 V, at positive and
negative voltages, respectively.
S20
Figure S22. 2D current histogram of 872 I-V curves measured by C-AFM (at a loading force of
30 nN) on the SAM of 2 (second batch) chemically grafted on ultra-flat template stripped gold
electrode (AuTS). Voltages are applied on the AuTS electrode (C-AFM grounded).
Figure S23. Current histograms at a given bias: +0.9V (a) and –0.9 V (b), from I-V data in Fig.
S22. The histograms are fitted by two log normal distributions with the following parameters:
mean current/standard deviation –9.6 (i.e. 2.5⋅10–10 A) /0.6 and –8.32 (4.8⋅10–9 A)/0.46 at 0.9 V, ⋅
9.64 (2.3⋅10–10 A)/0.77 and –8.34 (4.6⋅10–9 A) /0.14 at –0.9V.
S21
Figure S24. Histograms of the transition voltage VT+ (90 counts) and VT- (52 counts) at positive
(a) and negative (b) voltages from data Fig. S22, respectively. VT histograms are fitted with a
Gaussian peak with the parameters: mean voltage/standard deviation of 0.75 V/0.11 V and –0.79
V/0.16 V, at positive and negative voltages, respectively.
S22
VII. Crystal packing of polyanions 1 in Na-1
Figure S25. Packing of polyanions 1 in the crystal lattice of Na-1. View along the
crystallographic a axis. The WO6 and PO4 polyhedra of the neighboring polyanions 1 alternate in
color. Na and O atoms of crystal water are omitted for clarity. Color code for the other atoms: Co
purple, O red, As green, C black, H gray spheres.
S23
Figure S26. Packing of polyanions 1 in the crystal lattice of Na-1. View along b. Color code as
in Fig. S25.
S24
Figure S27. Packing of polyanions 1 in the crystal lattice of Na-1. View along c. Color code as in
Fig. S25.
S25
VIII. Crystal packing of polyanions 2 in compound Na-2
Figure S28. Packing of polyanions 2 in the crystal lattice of Na-2. View along a. The WO6 and
PO4 polyhedra of the neighboring polyanions 2 alternate in color. Na and O atoms of crystal
water are omitted for clarity. Color code for the other atoms: Co purple, O red, As green, C black,
N blue, H gray spheres.
S26
Figure S29. Packing of polyanions 2 in the crystal lattice of Na-2. View along b. Color code as
in Fig. S28.
S27
Figure S30. Packing of polyanions 2 in the crystal lattice of Na-2. View along c. Color code as in
Fig. S28.
S28
|
1805.04683 | 2 | 1805 | 2019-05-15T15:13:48 | Delta H = Delta B region in volume defect-dominating superconductor | [
"physics.app-ph",
"cond-mat.supr-con"
] | According to Ginzburg-Landau theory, it has been generally accepted that the diamagnetic property decreases after the lower critical field. However, we found that (Fe, Ti) particle doped MgB2 specimens reveal the Delta H = Delta B section in the magnetization curves, which are not following the theory. We present whether this phenomenon appears to be only confined to (Fe, Ti) particle doped Magnesium diborid superconductor, whether there is a theoretical basis and why it does not appear in other superconductors. We have understood that the cause of the Delta H = Delta B section is the pinning phenomenon of defects in the superconductor and it only occurs in volume defect dominating superconductors. The width of the Delta H = Delta B section along the number of defects and Hc2 was estimated assuming that defects are in the ideal state, and compared with experimental results. We hypothesized that pinned fluxes have to be picked out from the defect and move into an inside of a superconductor regardless free energy depth of the defect if the distance between fluxes pinned at the defect is equal to the one of upper critical field. It is considered that the reason that this phenomenon has not been reported yet is the flux jump of the volume defect dominating superconductor. The section means that the fluxes that have penetrated into a inside of a superconductor in which volume defects exist are preferentially pinned on them over the entire specimen before Ginzburg-Landau behavior. If the size of volume defects is uniform in some extent, the influence of the planar and line defects is small and the flux jump does not occur, we believe that the section must be observed in any superconductor. It is because this is one of the basic natures of pinning phenomenon in the volume defect dominating superconductor. | physics.app-ph | physics | ∆H = ∆B region in a Volume Defect-Dominating
Superconductor
H. B. Lee, G. C. Kim, H. J. Park, D. Ahmad, and Y. C. Kim∗
Department of Physics, Pusan National University, Busan 46241, Korea
Abstract
It has been generally accepted that the diamagnetic property of type II superconductor decreases
after Hc1. On the other hand, we found that (Fe, Ti) particle-doped MgB2 specimens have a ∆H
= ∆B region in the M-H curves, which is the region that the increase of a magnetic induction is
as much as the increase of an applied magnetic field. Here we study whether this phenomenon
was only confined to (Fe, Ti) particle-doped MgB2 superconductor, whether there is a theoretical
basis, and why it does not appear in other superconductors. The cause of the ∆H = ∆B region
was the pinning phenomenon of defects in the superconductor and it only occurs in volume defect-
dominating superconductors. Widths of the ∆H = ∆B region along the number of defects and
Hc2 were calculated, and compared with the experimental results. We hypothesized that pinned
fluxes have to be depinned from the defect and move into an inside of a superconductor regardless
∆Gdef ect if the distance between fluxes pinned at the volume defect is equal to that of Hc2. The
region means that the fluxes that have penetrated into the inside of a superconductor are pinned
preferentially on the volume defects over the entire specimen before the general behavior.
9
1
0
2
y
a
M
5
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
2
v
3
8
6
4
0
.
5
0
8
1
:
v
i
X
r
a
∗Electronic address: [email protected]; Fax: +82-51-513-7664
1
Although it is clear that superconductors have a flux pinning effect, the exact mechanism
is not completely understood [1 -- 4]. All superconductors have flux pinning effects because
they have defects even if defects are few. Most of type II superconductors have shown that
the diamagnetic property of the superconductor decreases gradually after H(cid:48)
c1 (not Hc1),
which is defined as the field showing the maximum diamagnetic property. On the other
hand, we observed a phenomenon in (Fe, Ti) particle-doped MgB2 specimens that do not
show the general behavior after H(cid:48)
c1, which is the existence of a ∆H = ∆B region in the
magnetization-applied field (M-H) curves. This is an unusual phenomenon that has not
reported in other superconductors.
The flux pinning phenomenon is caused mainly by defects in the superconductor. Gener-
ally, defects in the superconductor contain volume defects (such as general volume defects,
precipitators, inclusions and columnar defects, etc.), planar ones (such as grain boundaries,
twin boundaries and stacking faults plane, etc.), and line ones (such as dislocations). Al-
though they all belong to a family of defects, a role difference for flux pinning is considerable.
In the case of volume defects, pinned fluxes are difficult to escape from defects except when
a force balance (Fpinning = Fpickout) is broken; hence, they are called strong pinning sites.
On the other hand, weak pinning sites, such as planar defects and line defects, are entirely
different from the strong pinning sites in the flux pinning mechanism. The grain boundaries
(GBs), which have relatively lower pinning energy caused by planar characteristic, are con-
nected to each other in the entire specimen as a planar defect. Therefore, fluxes pinned on
the GB move easily along the GBs. In addition, because the total area as a defect is large,
they have significant importance in the overall flux pinning effects. A superconductor dom-
inated by planar defects in the flux pinning effect can be called a planar defect-dominating
superconductor. High Tc superconductor (HTSC) bulks are associated in this category [6 -- 9].
The depinning, which means the phenomenon that magnetic fluxes escape from the de-
fect, will be considered by two ways. The one is pick-out depinning, which is the depinning
that fluxes pinned on the volume defect are depinned together. The other is leak-out de-
pinning, which is the depinning that fluxes pinned on the volume defect are depinned one
by one. The former is the depinning by the force balance of fluxes pinned at the defect and
the latter is depinning through grain boundaries connected on volume defect because grain
boundaries does not only pin the fluxes but also leak out fluxes pinned at volume defects.
Regarding dislocations, the penetration of fluxes through them is not too difficult as planar
2
defect-dominating superconductors do because dislocations as a line defect are also inter-
connected throughout the specimen,. Worked NbTi superconducting wires are associated in
this category [11, 12]. Therefore, planar and line-defect dominating superconductors appear
superficially to follow the general behavior.
MgB2, which was made by a synthetic method at high temperatures also has grain bound-
aries, but most of them are low angle ones due to their fabricating characteristic [13 -- 17].
Hence it has significantly fewer weak links than HTSC bulks produced by a solid state
reaction method. Therefore, it can be called a volume defect-dominating superconductor.
MgB2 has been known as a superconductor which field dependence is weak, but a definite
effect can be obtained by doping artificial defects because it is a volume defect-dominating
superconductor [14, 18 -- 21].
Pure MgB2 and (Fe, Ti) particle-doped MgB2 specimens for this study were synthesized
using a non-special atmosphere synthesis (NAS) method [19]. All specimens which had been
synthesized at 920oC for 1 hour were cooled in air, but 5 wt.% (Fe, Ti) particle-doped MgB2
specimen, which had shown prominent results, underwent two different cooling processes.
One was cooled in air and the other was quenched in water. Figure 1 (a) presents the NAS
method for MgB2 and Fig. 1 (b) shows a photograph of the method. Figure 1 (c) is a
photograph of (Fe, Ti) particles, which are slightly far from sphere and Fig. 1 (d) shows
(Fe, Ti) particles present in MgB2. The radius of the particles is rather irregular, and the
average radius of them is 163 nm.
I. RESULTS
A. A diamagnetic property increase and the confirmation of the ∆H = ∆B region
in experiments
Fluxes would penetrate into the superconductor in flux quantum form over Hc1 [22].
The diamagnetic property of the superconductor decreases gradually after the maximum
property and this phenomenon continues to Hc2. This is true if there are no defects, which
are pinning sites in the superconductor. However, real superconductors which have defects
behave differently. Fluxes, having penetrated into the superconductor, are pinned at the
defects near the surface and the diamagnetic property increases rather than that of Hc1.
3
We call it H(cid:48)
superconductor.
c1, which represents the field of the maximum diamagnetic property in a real
In planar defect-dominating superconductors and line defect-dominating superconduc-
tors, the diamagnetic property of H(cid:48)
c1 did not make a large difference from that of Hc1 if
there are no volume defects. It is caused by the fact that the small volume of an individual
defect induces a weak pinning force. Hence, the increase of the diamagnetic property at H(cid:48)
is small. In particular, they are interconnected; thus, they allow well for flux penetration.
c1
Therefore, it appears to follow the general behavior superficially and there is no the ∆H =
∆B region in the M-H curves.
On the other hand, volume defect-dominating superconductors show distinctly different
behavior. The pinning effect is strong due to their relatively larger volume, and the most
important thing is that they are not interconnected with each other. Therefore, they continue
to pin fluxes until their pinning limits. They would act as another barrier to prevent the
fluxes from penetrating into the superconductor over Hc1. Thus, the diamagnetic property
of the volume defect-dominating superconductors certainly increases. As shown in all the
M-H curves except for pure MgB2 in Fig. 2, a linear region ends about 600 Oe, which means
perfect diamagnetism. After that, they show a slight decrease in slope. This behavior means
that the fluxes penetrated into the superconductor are pinned at defects near the surface and
cannot move easily into the specimen. Therefore, the diamagnetic property of the specimen
continues to increase even though Hc1 has passed.
Figure 2 (a) presents M-H curves of pure MgB2 and 5 wt.% (Fe, Ti) particle-doped MgB2
that were air-cooled and measured at 5 K. The M-H curve of pure MgB2 used as reference.
It is clear that the ∆H = ∆B region is observed after H(cid:48)
c1 in 5 wt.% (Fe, Ti) particle-doped
MgB2. The width of the region can be disputed, but it is definite that the M-H curve of the
specimen forms a ∆H = ∆B region from the H(cid:48)
c1 to 8 kOe. After flux jump, it continues
to show the ∆H = ∆B region up to 15 kOe. And it shows gradual decrease of diamagnetic
properties, which are the ∆H > ∆B region over 15 kOe.
Figure 2 (b) presents the M-H curve of 5 wt.% (Fe, Ti) particle-doped MgB2 that was
water-quenched and measured at 5 K. Generally, the water-quenching method is used to
refine the grains by impeding the growth rate of grains or to induce a rapid phase transfor-
mation (e.g.: martensite transformation). In current experiments, it was used to increase
the angle between the grains of MgB2 and refine the grains due to the rapid cooling rate.
4
This treatment has the purpose of providing further opportunities for the fluxes that have
been pinned at volume defects to leak out through the grain boundary. Thus, this procedure
can reduce the stress of the concentration of fluxes on the volume defects.
Therefore, the ∆H = ∆B region in the figure is formed up to 20 kOe in a wide view, even
though there was a small flux jump. As shown in these two figures, it is reasonable that
the ∆H = ∆B region of the 5 wt.% (Fe, Ti) particle-doped MgB2 specimen is from H(cid:48)
c1 to
a point between 15 kOe and 20 kOe. Figure 2 (c) presents the M-H curve measured at 10
K on the air-cooled 5 wt.% (Fe, Ti) particle-doped MgB2 specimen and (d), (e) and (f) in
Fig. 2 are ones measured at 5 K with different doping concentrations. It is clear that the
∆H = ∆B region is observed in all specimens except for the 1 wt.% doped specimen.
B. Pinned fluxes movement and the basis of ∆H = ∆B region
A previous study reported that the flux quanta pinned at a defect move with a bundle
and hop from one pinning site to another [23, 24].
If the distance between the volume
defects is wide enough, the fluxes that are pinned at the defect move when the force balance
is broken (Fpinning < Fpickout), which is based on ∆Gdef ect and a repulsive force between the
flux quanta. On the other hand, when the distance between the volume defects is short,
fluxes pinned at the defect would move into an inside of the superconductor by a different
mechanism. This means that when a volume defect of the superconductor reach the limit
value of pinned fluxes, they have to be depinned from the defect and move into an inside of
the superconductor regardless of ∆Gdef ect
If a volume defect existing near the surface of the superconductor pins fluxes and they
are blocked from moving into an inside of the superconductor until defect's pinning limit,
the free energy density of a spherical defect in the superconductor can be expressed as
∆Gsuper − ∆Gnor =
H 2
8π
⇒ ∆Gdef ect = −H 2
8π
× 4
3
πr3
(1)
where H is the applied field and r is the radius of the defect. According to the equation,
∆Gdef ect is dependent on the external field H when r is constant. The flux quanta pinned at
the defect can move into an inside of the superconductor when they are at Fpinning < Fpickout
state, and they will be pinned again at another defect in front of them.
It is necessary
to increase H in order for the fluxes to be depinned from the defect and move into an
5
inside of the superconductor. If H is increased, ∆Gdef ect in the superconductor becomes
larger. Therefore, a stronger H will be needed for penetrating fluxes into an inside of the
superconductor.
Although a diamagnetic property of the superconductor increases due to the pinning
phenomenon, it does not increase continuously. There must be a limit of the pinned fluxes
that brakes this premise as shown in the experiment. We considered the basis of this limit
to be the minimum distance between the pinned fluxes at the defect. This is because the
neighborhood around the defect which have pinned fluxes is no longer a superconducting
state when the minimum distance between the fluxes pinned at the defect is less than that
of Hc2; thus, there is no pinning effect anymore.
The reason for creating a ∆H = ∆B region in the M-H curve is originated from the flux
pinning limit of a volume defect in the superconductor. The ∆H = ∆B region is formed
in the M-H curve because the defects are filled with flux quanta step by step from the
surface to the center of the superconductor when the defects have a pinning limit of flux
quanta according to their radius. Figure 3 (a) and (b) show the flux pinning limit of defects
for having the same and different radius, respectively. Both are superconductors with four
defects along y-axis and nine defects along x-axis. The difference between the two is the
particle size of defects, which is uniform in (a) and the average particle size in (b) is the
same as that of (a). When the flux quantum lies in y-axis and moves along x-axis direction,
it is natural that the ∆H = ∆B region is formed in (a) because the fluxes coming from the
outside of the defect are pinned at the defect and move if a defect exceeds its flux pinning
limit.
On the other hand, Figure 3 (b) shows slightly different behavior. Fluxes pinned at a
small pinning site moves first because the flux pinning limit is low, and fluxes pinned at
a larger defect move later. When many fluxes that have been pinned at larger defect are
depinned from the defect, they move together; thus, there is a high possibility of flux jump.
The 5 wt.% (Fe, Ti) particle-doped MgB2 is an example of unevenness of defects, as shown
in Fig. 2 (a). It has approximately 80003 defects in 1 cm3 of MgB2, of which the radius is
163 nm on average. In this situation, a single quantum flux in a superconductor would be
simultaneously pinned at 8000 defects on average. Although there are some fluxes pinned
at a defect that move first and some fluxes at another defect that move later, the ability to
pin fluxes on average is similar to the counterpart when it is observed as a whole specimen.
6
Therefore, there is no problem in forming the ∆H = ∆B region.
C. Calculations for a flux pinning limit of a defect and the width of ∆H = ∆B
region
Assuming that volume defects are spherical, their size is constant, and they are arranged
regularly in a superconductor, a superconductor of 1 cm3 has m3 volume defects. The
maximum number of flux quanta that can be accommodated at a spherical defect of radius
r in a static state is
n2 =
πr2
π( d
2 )2
× P = (
)2 × P
2r
d
(2)
where r, d and P is the radius of defects, a distance between quantum fluxes and filling rate
which is π/4 when they have square structure, respectively (see Fig. 3 (c) and (d)). If the
radius of a defect is 163 nm, the maximum number of quantum fluxes that can be pinned
by the defect is approximately 452 at 0 K in the static state because the distance (d) is 6.43
nm when Hc2 is 50 T (Hc2 = Φ0/d2) [26]. We thought that quantum fluxes had a square
structure rather than a triangular one when they were pinned at the defect [27].
Therefore, the magnetic induction B can be expressed as
B = n2mcpsmΦ0
(3)
where n2, mcps, m, and Φ0 are the number of quantum fluxes pinned at a defect, the number
of defects which are in the vertically closed packed state, the number of defects with pinned
fluxes from the surface to the center of the superconductor, and flux quantum, respectively.
mcps is explained in Fig. 3 (e) and (f). mcps is the minimum number of defects when
the penetrated fluxes into the superconductor are completely pinned. This conversion was
introduced to calculate the number of flux quanta which are pinned on defects of a plane
because the fluxes between defects can penetrate into the superconductor without pinning if
defects are arranged regularly like a lattice as shown in Fig. 3 (e). The conversion is much
closer to reality because defects are arranged randomly in a real superconductor. If 80003
defects are in 1 cm3 superconductor, as described in the experiment, there are approximately
80002 defects in a plane. Therefore, there are almost no penetrating fluxes without pinning.
Thus, the total number of flux quanta pinned on the defects of a plane perpendicular to the
flux-moving direction are n2mcps.
7
Hence, the magnetization M is
B = H + 4πM ⇒ M =
B − H
4π
=
n2mcpsmΦ0 − H
4π
Therefore, a width of the ∆H = ∆B region is
∆H = H − H(cid:48)
c1 = n2mcpsmΦ0 − 4πM − H(cid:48)
c1
(4)
(5)
where H(cid:48)
c1 is the field showing the first maximum diamagnetic property in the supercon-
ductor. If the radius of defects is fixed, n2 and mcps are also fixed. Therefore, the width
of the ∆H = ∆B region is dependent only on the m. A calculated width of the ∆H = ∆B
region along a number of defect is shown in Fig. 4 (a) when the radius of a defect is 163 nm
and Hc2 is 50 T. As shown in the figure, the width of the ∆H = ∆B region increases with
increasing number of pinning sites except for over-doping.
One of the important factors calculating the width of the ∆H = ∆B region is what is
Hc2 of a superconductor. Hc2 is a fundamental property according to the material of a
superconductor, but it is inferred from the indirect method at a low temperature because it
has difficulty in being measured directly. For example, Hc2 of MgB2 varies from a theoretical
value of 64 T to experimental one of approximately 20 T [25, 29 -- 31]. The calculated width
of the ∆H = ∆B region along Hc2 variation is shown in Fig. 4 (b) when the radius of a
defect is 163 nm and there are 80003 defects in a 1 cm3 superconductor, which is equivalent
to 5 wt.% (Fe, Ti) particle-doped MgB2.
When the width of the ∆H = ∆B region was calculated with Hc2 = 50 T, it reasonably
matches the experimental results as shown in Fig. 4 (a) and (b), which were converted from
5 K to 0 K (the width of the ∆H = ∆B region was conservatively determined to be 1.3
T at 5 K in the 5 wt.% specimen, thus it will be 5.2 T if expanded by 1 cm because the
thickness of measured specimen is 2.5 mm). The experimental result is rather higher than
the theoritical one in the figures, low purity of boron (96.6 %) caused volume defects of
which radius is 1µm on average (SMFig. 7). It is determined that the cowork of (Fe, Ti)
particles with them make the result. Figure 4 (c) shows the flux penetration method based
on the general behavior [28] and (d), (e) and (f) show the flux penetration method based
on the existence of the ∆H = ∆B region. They indicated that the fluxes penetrated into
the superconductor are pinned preferentially on the volume defects over the entire specimen
before the general behavior. The width of the ∆H = ∆B region increases with increasing
8
number of volume defects, and the width of the region is narrow if the number of volume
defects are few or too many.
II. DISCUSSION
The presence of ∆H = ∆B region is of great importance in practical applications of the
superconductor. Consider, for example, the case of using superconductors in magnetic levi-
tation train. Superconductors showing the general behavior are difficult to use diamagnetic
property up to the maximum. When the train levitates and moves, there will be up and
down vibrations, which will bring in more magnetic fields on the superconductor. When the
magnetic field is applied beyond the field which produces the maximum diamagnetic prop-
erty, there is a fear that the train may fall to the bottom because diamagnetic property will
be reduced. On the other hand, superconductors with ∆H = ∆B region have no problem
even when using the maximum diamagnetic property because the maximum diamagnetic
property is maintained in a considerable region if there is no flux jump.
To solve a weak magnetic field dependence of MgB2, many researchers have doped a
variety of materials and achieved considerable results [32 -- 34]. However, despite the improved
field dependence in high field, there were still a lot of flux jumps in low field. Therefore, it
was easy that the ∆H = ∆B region in MgB2 specimens was overlooked. In our experiments,
we didn't recognize the region owing to flux jump in 5% doped specimen. However, we
suspected the diamagnetization point after flux jump (the point of 1.5 T in Fig. 2 (a)),
which was too much higher, and confirmed the region after quenching the specimen in
water, which lowered the flux jump (Fig. 2 (b)).
The essence of this communication is as follows. Fluxes that have been penetrated into
the superconductor are pinned preferentially on volume defects over the entire specimen
if it is a volume defect-dominating superconductor. This is because fluxes pinned on the
defects are bent like a bow; thus, unpinned ones are difficult to exist without pinning on
defects due to the repulsive force between fluxes and the irregular distribution of defects.
Since the volume defect reach its pinning limits when the external field exceeded H(cid:48)
c1, the
internal fluxes (B) increase as much as the external field (H) increases. Therefore, when the
superconductor is is dominated by volume defects, ∆H = ∆B region is first formed after H(cid:48)
and the ∆H > ∆B region is formed later in the M-H curve.
c1
9
On the other hand, it might be hard to accept that flux pinning on defects cause a larger
diamagnetic property than that of Hc1. However, this is a common phenomenon because
there is no material having no defects. It is rather natural to explain that planar and line
defect-dominating superconductors follow the general behavior is due to the interconnectivity
of the defects. A typical example of increasing the diamagnetic property by flux pinning
is the fishtail effect. The fishtail effect is often observed in superconducting single crystal
(SC), particularly in HTSC SCs. There are many opinions about the cause of the fishtail
effect, but there is some consensus that it is due to the pinning phenomenon [35, 36].
One of the important features of volume defect-dominating superconductor (VDS) is the
flux jump. If pinned fluxes on volume defects do not leak out through grain boundaries, the
volume defects will pin fluxes to their pinning limit. In addition, they move together when
they are picked out from the defect; thus, flux jump can occur if they are many. Moreover,
because diamagnetic property of the VDS is always higher than that of the pure state of
superconductor and the superconductor with volume defects are pinned from the surface,
the fluxes pinned on defects are always under pressure that they may penetrate into an
inside of the superconductor. This is the reason that the flux jump occurs well in MgB2
synthesized at high temperature.
The main reason that a ∆H = ∆B region has not been reported so far is considered to
be low density of volume defects and the lack of a proper VDS like MgB2. In addition, the
thickness of the measured specimen also influence considerable effect on the width of ∆H
= ∆B region. Though the density of volume defect is meaningful, the region cannot be
observed owing to its thickness if a measured specimen is thin. The number of pinning sites
decreases as the thickness of the specimen becomes thinner, thus the region do not appear
to be distinguished level.
As shown in Fig. 2, the ∆H = ∆B region does not appear when the density of the volume
defect is low (Fig. 2 (d)) and the region is too short to recognize it when the density of
the volume defect is high (Fig. 2 (f)). The important thing in the ∆H = ∆B region is
the number of volume defects as well as the density of them. Under the same density of
defects, it is difficult to observe the region if the volume defects are large and a few, whereas
observation of the region is possible if they are small and many. In our experiments, we
have compared the width of the regions according to the density of volume defect using of
which radius is 163 nm on average, showing that 5 wt.% doped specimen had the widest
10
region and width of the region tends to decrease if the volume defects are denser or sparser.
Other examples of VDS are unworked NbTi and melt-texture growth (MTG) specimens
of HTSC. In the case of NbTi, it is difficult to find an experiment of a correlation with the
volume defect because the focus for the increase of the pinning effect was more on the line
defects. However, unworked NbTi can be classified as a VDS owing to its flux jump [37]. On
the other hand, melt-texture growth (MTG) was introduced to eliminate the weak links of
HTSCs. One of the distinct features of MTG is that flux jump occurs frequently like MgB2
[38]. The fact that the flux jump, which was not observed in the HTSC specimen prepared
by the solid-phase reaction method, frequently occurs in a HTSC specimen prepared by
the MTG method means that the dominating flux pinning mechanism has changed from
planar defects to volume defects. Because MTG has a higher concentration of impurities,
an extensive literature search on the M-H curves of MTG was performed and two papers
were found [39, 40]. The considerable width of the ∆H = ∆B region has been formed after
H(cid:48)
c1 in these papers. Therefore, we had clear confirmation that the phenomenon that ∆H
= ∆B region appears is not to be confined to MgB2, especially this experiment, but to be
common in volume defect-dominating superconductors.
The ∆H = ∆B region was demonstrated by experiments of (Fe, Ti) particle-doped MgB2
specimens that are a volume defect-dominating superconductor. And we represented a
theoretical base of the phenomenon and compared with experiment results. Moreover, we
found a superconductor in a literature that show a ∆H = ∆B region in MTG HTSC,
and confirmed the generality of the phenomenon.
It is considered that the behavior of
the superconductor is based on the flux pinning limit of the volume defects regardless of
∆Gdef ect. In addition, it was also emphasized that superconductors should be classified not as
materials but as defects in order to understand the flux pinning phenomena properly. There
is no defect-free material, and it is proper to interpret the phenomenon of superconductivity
based on this point. The ∆H = ∆B region has the basis of the flux pinning phenomenon,
which appears ahead of the general behavior in a volume defect-dominating superconductor.
III. METHOD
The starting materials were Mg (99.9% powder), B (96.6% amorphous powder) and (Fe,
Ti) particles. The mixed Mg and B stoichiometry, and (Fe, Ti) particles were added by
11
weight. They were finely ground and pressed into 10 mm diameter pellets. The (Fe, Ti)
particles were ball-milled for several days, and the average radius of the (Fe, Ti) particles
was approximately 0.163 µm. On the other hand, an 8 m-long stainless- steel (304) tube was
cut into 10 cm pieces. One side of the 10 cm-long tube was forged and welded. The pellets
and excess Mg were placed in the stainless-steel tube. The pellets were annealed at 300oC for
1 hour to make them hard before inserting them into the stainless-steel tube. The other side
of the stainless-steel tube was also forged. High-purity Ar gas was put into the stainless-steel
tube, and which was then welded. All of the specimens were synthesized at 920oC for 1 hour.
The field and temperature dependence of magnetization were measured using a MPMS-7
(Quantum Design). During the measurement, sweeping rates of all specimens were equal
for the same flux-penetrating conditions.
[1] Thomas Schuster et al. Flux motion in thin superconductors with inhomogeneous pinning.
Phys. Rev. B 50 16684 (1994).
[2] S. R. Ghorbani et al. Flux-pinning mechanism in silicone-oil-doped MgB2: Evidence for charge-
carrier mean free path fluctuation pinning. Phys. Rev. B 78 184502 (2008).
[3] A. K. Geim et al. Non-quantized penetration of magnetic field in the vortex state of super-
conductors. Nature 407 55 (2000).
[4] Ming Xu et al. Comparison of superconducting flux pinning due to ion irradiation and fishtail
effects. Phys. Rev. B 53 5815 (1996).
[5] Aleksandra Petkovi´c, Thorsten Emig and Thomas Nattermann, Pinning of flux lines by planar
defects. Phys. Rev. B 79, 224512 (2009).
[6] E. Shimizu and D. Ito, Critical current density obtained from particle-size dependence of
magnetization in YBa2Cu307−δ powders. Phys. Rev. B 39 2921 (1989).
[7] J. E. Tkaczyk et al., Critical-state scaling and weak links in Ag-sheathed Bi2Sr2Ca2Cu30z.
Phys. Rev. B 45 12506 (1992).
[8] S. Senoussi, M. Oussena and S. Hadjoud, On the critical fields and current densities of
YBa2Cu3O7 and La1.85Sr0.15CuO4 superconductors. J. Appl. Phys. 63 4176 (1988).
[9] R Suryanarayanan et al. Magnetization of YBaCuO and Pb-substituted BiSrCaCuO. Super-
cond. Sci. Technol. 2 261 (1989).
12
[10] K Tachikawa, T Watanabe and T lnoue, Synthesis of Bi-base high-Tc oxides through a diffusion
reaction. Supercond. Sci. Technol. 3 180 (1990).
[11] Charles P. Bean and Roland W. Schmitt, The Physics of High-Field Superconductors. Science
140 26 (1963).
[12] A. K. Ghosh and W. B. Sampson, Anomalous low field magnetization fine filament NbTi
conductors. IEEE Transactions on Magents Mag − 23 1724 (1987).
[13] W.N. Kang, H. J. Kim, E. M. Choi, C. U. Jung and S. I. Lee, MgB2 Superconducting Thin
Films with a Transition Temperature of 39 Kelvin. Science 292 1521 (2001).
[14] S. X. Dou et al. Flux jumping and a bulk-to-granular transition in the magnetization of a
compacted and sintered MgB2 superconductor. Physica C 361 79 (2001).
[15] C. B. Eom, L. D. Cooley, A. Polyanskii, E. E. Hellstrom, and M. A. Hayward, High critical
current density and enhanced irreversibility field in superconducting MgB2 thin films. N ature
411 558 (2001).
[16] S. Jin, H. Mavoori, C. Bower, and R. B. van Dover, High critical currents in iron-clad super-
conducting MgB2 wires. N ature 411 563 (2001).
[17] P. C. Canfield et al. Superconductivity in Dense MgB2 Wires. Phys. Rev. Lett. 86 2423 (2001).
[18] W K Yeoh, J H Kim, J Horvat, S X Dou and P Munroe, Improving flux pinning of MgB2 by
carbon nanotube doping and ultrasonication. Supercond. Sci. Technol. 19 L5 (2006).
[19] H. B. Lee, Y. C. Kim and D. Y. Jeong, Non-special atmosphere synthesis for MgB2. J. Kor.
Phys. Soc. 48 279 (2006).
[20] H. B. Lee, G. C. Kim, Y. C. Kim, D. Ahmad and Yong Seung Kwon, Flux jump behaviors
and mechanism of MgB2 synthesized by the non-special atmosphere synthesis. J. of Supercond.
and Nov. Magn. 28 2663 (2015).
[21] H. B. Lee, G. C. Kim, Y. C. Kim and D. Ahmad, Flux jump behaviors and mechanism of
FeTi doped MgB2 at 5 K, Physica C 515 31 (2015).
[22] W. A. Littler and R. D. Park, Observation of quantum periodicity in the transition tempera-
ture of a superconducting cylinder. Phys. Rev. Lett. 9 2423 (1962).
[23] D. J. Van Ooljkn and G. J. Van Gurp, Measurement of noise in the resistive state of type II
superconductor. Phys. Lett. 17 230 (1965).
[24] J. E. Bonevich et al. Electron Holography Observation of Vortex Lattices in a Superconductor.
Phys. Rev. Lett. 70 2952 (1993).
13
[25] Cristina Buzea and Tsutomu Yamashita, Review of the superconducting properties of MgB2.
Supercond. Sci. Technol. 14 R115 (2001).
[26] Michael Tinkham. ITRODUCTION TO SUPERCONDUCTIVITY 2nd Dover pub. Mineola
146-147 (2004)
[27] A. A. Abrikosov, On the Magnetic Properties of Superconductors of the Second Group. Sov.
Phys.-JETP 5 1174 (1957).
[28] Charles P. Bean, Magnetization of High-Field Superconductors. Rev. of Mor. Phy. Jan. 31
(1964).
[29] Sung Hoon Lee, Soon-Gul Lee and Won Nam Kang, Superconducting Transition Properties
of Grain Boundaries in MgB2 Films. J. Kor. Phys. Soc. 66 7 (2015).
[30] A. V. Sologubenko, J. Jun, S. M. Kazakov, J. Karpinski, and H. R. Ott, Temperature depen-
dence and anisotropy of the bulk upper critical field Hc2 of MgB2. Phys. Rev. B 65 180505(R)
(2002),
[31] Charles P. Poole, Jr., Horacio A. Farach, Richard J. Creswick, SUPERCONDUCTIVITY the
first 337 (Academic Press, 1995).
[32] Z. X. Shi et al. Doping effect and flux pinning mechanism of nano-SiC additions in MgB2
strands. Supercond. Sci. Technol. 24 065015 (2011).
[33] K.S.B. De Silva et al. Flux pinning mechanisms in graphene-doped MgB2 superconductors.
Scripta Materialia 65 634 (2011).
[34] Chengduo Wang et al. Improved Jc-B properties of MgB2 multifilamentary wires and tapes.
Supercond. Sci. Technol. 25 125001 (2011).
[35] Y. Radzyner et al. Anisotropic order-disorder vortex transition in La2−xAxSrxCuO4. Phys.
Rev. B 65 214525 (2002)
[36] M. Daeumling, J. M. Seuntjens and D. C. Larbalestier, Oxygen-defect flux pinning, anomalous
magnetization and intra-grain granularity in YB2Cu3O7−δ. Nature 346 332 (1990).
[37] V. V. Chabanenko et al. Oscillation mode in the screening properties of NbTi plate as a result
of flux jumps. Physica C 369 77 (2002)
[38] Y. T. Xing et al. Flux jump in Y1−xNdxBa2Cu3O7−y superconductors prepared by the MTG
method. Physica C 337 200 (2000).
[39] S. W. Hsu, K. Chen and W. H. Lee, Temperature and Field-Sweeping Rate Dependence of
Flux Jumps in A Melt-Textured YBa2Cu3O7−x Superconductor. Solid State Communications
14
75 799 (1990).
[40] K. H. Muller and C. Andrikidis, Flux jumps in melt-textured Y-Ba-Cu-O, Phys. Rev. B 49,
1294-1307 (2003).
Acknowledgements
The authors would like to thank Dr. B. J. Kim of PNU for careful discussion and Dr. L.
K. Ko and Dr. D. Y. Jeong of KERI for experimental supports.
Author Contributions
This communication was conceived by H. B. Lee, experimented by H. B. Lee and G. C Kim,
written by H. B. Lee, G. C. Kim, D. Ahmad and Y. C. Kim. H. J. Park did mathematical
calculation.
Competing Interests
The authors declare no competing interests.
15
FIG. 1: Non-special atmosphere synthesis (NAS) method for MgB2 and (Fe, Ti) particles for the
experiment. (a): Schematic representation of the non-special atmosphere synthesis (NAS) method.
(b): Photograph of the specimen for the NAS method. (c): Photograph of (Fe, Ti) particles before
doped in MgB2, which were ball-milled for several days. (d): A photograph of 25 wt.% (Fe, Ti)
particle-doped MgB2, which was taken by field emission scanning electron microscope (FE-SEM).
The white bright ones in the MgB2 base are doped (Fe, Ti) particles.
16
FIG. 2: Field dependence of magnetization for pure MgB2 and (Fe, Ti) particle-doped MgB2 (M-H
curves). (a): Field dependence of magnetization for pure MgB2 and 5 wt.% (Fe, Ti) particle-doped
MgB2. Specimens were air-cooled and measured at 5 K. (b): Field dependence of magnetization
for 5 wt.% (Fe, Ti) particle-doped MgB2, which was water-quenched and measured at 5 K. (c):
Field dependence of magnetization for 5 wt.% (Fe, Ti) particle-doped MgB2, which was air-cooled
but measured at 10 K. (d): Field dependence of magnetization for 1 wt.% (Fe, Ti) particle-doped
MgB2, which was air-cooled and measured at 5 K. (e): Field dependence of magnetization for
10 wt.% (Fe, Ti) particle-doped MgB2, which was air-cooled and measured at 5 K. (f): Field
dependence of magnetization for 25 wt.% (Fe, Ti) particle-doped MgB2, which was air-cooled and
measured at 5 K. Full versions are shown in Supplementary Materials
17
FIG. 3: Flux pinning limit of defects, filling rate calculation and the definition of mcps. (a): Flux
pinning limit of defects when they have same radius and regular arrangement.
(b): The flux
pinning limit of defects when they have different radii and a regular arrangement. (c): Shape of
the maximum fluxes pinned at the defect, which is cut off the center of the defect and assumed to
be spherical. (d): The definition of d (e): Ideal arrangement of defects. There is a possibility that
fluxes are not pinned at defects if the fluxes lie on the y axis and move along the x axis. (f): The
definition of mcps. The mcps is the number of defects which are a vertically closed packed state
of defects. The defect arrangement in (e) needs to change to that in the (f) for calculating B in
the superconductor for not having any flux quantum penetrating into the superconductor without
pinning.
18
FIG. 4: Calculated width of the ∆H = ∆B region and flux penetration method compared to Bean
Model. (a): Calculated width of the ∆H = ∆B region along the number of defects in a super-
conductor. (b): Calculated width of the ∆H = ∆B region along the upper critical field (Hc2) of
a superconductor. (c): Flux penetration method which are based on the general behavior [28].
(d): Flux penetration method when the superconductor has a good pinning condition in a volume
defect-dominating superconductor. Fluxes penetrated into the superconductor are pinned on vol-
ume defects from existing ones around the surface of the superconductor. (e): Flux penetration
method when the superconductor has a proper pinning condition in a volume defect-dominating
superconductor. (f): Flux penetration method when the superconductor has a poor pinning con-
dition in a volume defect-dominating superconductor.
19
|
1908.05379 | 1 | 1908 | 2019-08-15T00:35:58 | Macroscopic electron-hole distribution in silicon and cubic silicon carbide | [
"physics.app-ph"
] | Electron excitations at silicon and 3C-SiC surfaces caused by an intense femtosecond laser pulse can be calculated by solving the time-dependent density functional theory and the Maxwell's equation simultaneously. The energy absorption, carrier density, and electron-hole quasi-temperatures decrease exponentially in 100 nm from the surface. The electron and hole quasi-temperatures have finite values even at large distances from the surface because of a specific photo-absorption channel. Although the quasi-temperature in the silicone shows smooth exponential descrease, 3C-SiC shows stepwise decrease because of the change of concerning bands. The quasi-temperature depends not only on the excitation process, i.e., tunnel and multi-photon absorption, but also on the band structure significantly. | physics.app-ph | physics |
Macroscopic electron-hole distribution in silicon and cubic silicon carbide
T. Otobe1
Kansai Photon Science Institute, National Institutes for Quantum and Radiological Science and Technology (QST),
Kyoto 619-0215, Japan
Electron excitations at silicon and 3C-SiC surfaces caused by an intense femtosecond laser pulse can be
calculated by solving the time-dependent density functional theory and the Maxwell's equation simultaneously.
The energy absorption, carrier density, and electron-hole quasi-temperatures decrease exponentially in 100
nm from the surface. The electron and hole quasi-temperatures have finite values even at large distances from
the surface because of a specific photo-absorption channel. Although the quasi-temperature in the silicone
shows smooth exponential descrease, 3C-SiC shows stepwise decrease because of the change of concerning
bands. The quasi-temperature depends not only on the excitation process, i.e., tunnel and multi-photon
absorption, but also on the band structure significantly.
I.
INTRODUCTION
Processing of solid materials using femtosecond laser
pulses has attracted considerable interest for potential
application to high-precision processing technology.1 -- 12
Because a femtosecond laser pulse can deposit large
amounts of energy into solid materials within a much
shorter time than the conventional spatial diffusion of
thermal energy to the exterior of the irradiated spot, we
can process materials with small thermal denaturation
outside of the irradiated volume.10,11
The precise description of the electron-hole distribu-
tion and their quasi-temperatures are important to un-
derstand the initial stage of laser processing. The multi-
photon absorption and tunnel excitation processes are
the crucial electron excitation processes in semiconduc-
tors under the femtosecond laser pulse. Since these pro-
cesses are nonlinear and/or nonperturbative, we have to
treat the dynamics of the electron and electromagnetic
fields simultaneously13,14.
The two temperature model15,16 (TTM) is a common
approach to describe the energy flow between an electron
and phonon in metals. The TTM is applied to semicon-
ductors and dielectrics by combining electron excitation
and scattering models17,18. Although the electron exci-
tation process and electron-hole distribution is compli-
cated, the excitation process and the estimation of the
electron temperature assuming simple model. To un-
derstand the electron-hole state induced by a laser field
around a surface, the dynamics of the electrons and elec-
tromagnetic field should be treated directly.
Recently, we developed a first-principle numerical pro-
gram SALMON, which combines time-dependent den-
sity functional theory (TDDFT) and the Maxwell's
equation19.
In the present work, we would like to cal-
culate the macroscopic carrier and hole distribution at
the surface of silicon and cubic silicon carbide (3C-SiC).
II. COMPUTATIONAL METHOD
The theory and its implementation in the present cal-
culation have been described elsewhere19 -- 21; therefore,
we describe it briefly. The laser pulse that enters from the
vacuum and attenuates in the medium varies on a scale
of micrometers, while the electron dynamics take place
on a subnanometer scale. To overcome these conflicting
spatial scales, we have developed a multiscale implemen-
tation by introducing two coordinate systems: macro-
scopic coordinate X for the laser pulse propagation and
the microscopic coordinate r for local electron dynam-
ics. The laser pulse is described by the vector potential
~AX (t), which satisfies
1
c2
∂2 ~AX (t)
∂t2 −
∂2 ~AX (t)
∂X 2 = −
4πe2
c
~JX (t).
(1)
At each point X, we consider lattice-periodic elec-
tron dynamics driven by the electric field EX (t) =
− 1
c dAX (t)/dt. They are described by the electron or-
bitals ψi,X (~r, t), which satisfy the time-dependent Kohn --
Sham equation at each point X 19.
i
∂
∂t
ψi,X (~r, t) =" 1
2m(cid:16)−i∇r +
e
c
~AX (t)(cid:17)2
+ µxc,X(~r, t)#ψi,X (~r, t),
− φX (~r, t)
(2)
where the potential φX (~r, t), which includes Hartree and
ionic contributions, and the exchange-correlation poten-
tial µxc,X(~r, t) are periodic in the lattice. In this work, we
use local density approximation (LDA)22 for exchange-
correlation potential in adiabatic approximation.
The electric current at X, JX (t), is provided from elec-
tron orbitals as
e
JX (t) = −
mV ZV
+ JX,N L(t),
d~rXi
Reψ∗
i,X(cid:16)~p +
e
c
~AX (t)(cid:17) ψi,X
(3)
where V is the volume of a unit cell. JX,N L(t) is the
current caused by the nonlocality of pseudopotential.
We solved Eqs. (1) -- (3) simultaneously as an initial
value problem, where the incident laser pulse is prepared
in a vacuum region in front of the surface, while all the
Kohn -- Sham orbitals are set to their ground state.
The incident laser field in vacuum, Ein(X, t), is de-
scribed as,
Ein(X, t) =(E0 sin2(cid:16)π tX
0
Tp(cid:17) cos(ω0tX ) 0 < tX < Tp
Tp < tX < Te,
(4)
!"#
#"$
#"#
#"$
!"#
!$
!"#
#"$
#"#
#"$
:
@
?
>
8
0
=
<
7
-
;
!"#
!$
!"#
#"$
#"#
#"$
!"#
!$
85:
$F!#
!(
0G?63
(
0;-7<=05.0DE#001,
0;-7<=05.0DE'%01,
0A/72BC
!#
$
8H:
!F!#
!)
0G?63
(
!#
$
86:
(F!#
!)
0G?63
#
#
(
!#
$
#
*+,-.-+/012+30,4215670893:
!#
%
&
'
(
#
)#
($
(#
!$
!#
$
#
A
/
7
2
B
C
0
8
7
>
?
@
)
:
!##
%#
&#
'#
(#
#
9
6
+
8
/
7
6
/
0
1
/
5
1
3
/
+
4
3
/
2
1
,
0
/
.
-
,
+
*
)
"
$
!
$
&
"
$
!
$
&
!'!
!'
!'%
!'(
!'$
!'#
!'"
'!
'%
!
!'%
!'!
&!'%
8@9/
A< !
%/
%
/B?C2
"
#
$
%
!
&%
&$
&#
!
'%
'%
!
< !!
'!
'!
!'!
!'
!'%
!'(
!'$
!'!
!'
!'%
!'(
!'$
!'#
!'#
!'"
!'"
!'%
!'!
&!'%
8D9/
< !
(/
%
/B?C2
!'!
!'
!'%
!'(
!'$
!'!
!'
!'%
!'(
!'$
!'#
!'#
!'"
!'"
< !!
'!
'!
'%
'%
"
$
!
$
&
!'$
!'!
&!'$
8C9/
%< !
(/
%
/B?C2
< !!
!'!
!'
!'%
!'(
!'$
!'#
!'"
'!
'%
:/8;29
2
!
"
#
$
%
!
&%
&$
&#
!
"
#
$
%
!
!
&%
&$
&#
!
!
"
#
$
%
!
&%
&$
&#
!
!
=1>/8 ?+69
"
#
$
%
!
&%
&$
&#
!
!
"
#
$
%
!
&%
&$
&#
FIG. 1. Energy density and light field after the laser field
shine on the silicon surface with an intensity of (a) 5 × 1012,
(b) 1 × 1013, and (c) 2 × 1013 W/cm2. The red line represents
the initial field (0 fs), and the blue lines represent the field at
48 fs. The energy of densities are represented by the green
filled lines.
where E0 is the electric field amplitude at the peak, ω0 is
the laser frequency, tX = t−X/c describes the space-time
dependence of the field. The pulse length Tp is set to be
10.81 fs, and the computation is terminated at Te = 48.38
fs.
III. RESULTS
A. Silicon
FIG. 2. Change of the electron occupation from the initial
state. The right panels show the density of states of silicon,
for reference.
Eex = Etot(Te) − Etot(0). The electron-hole density (Ne)
is defined by the projection of the time-dependent wave-
function at t = Te onto the initial state23.
Individual projections to the initial states at each
macro point X, given by,
~k
X,j(t) =
O
hΦ
~k+ e
c
X,j
~A(t)
~k
X,i′ (t)i2,
u
(5)
1
V Xi′=occ
give the occupation of state j. We prepare adequate un-
occupied states in the conduction band, typically com-
prising 100 states for each ~k, to calculate the overall elec-
tron distribution.
We first assume a typical silicon semiconductor. Be-
cause we use the LDA exchange-correlation potential, the
calculated direct band gap (2.4 eV) is smaller than the
experimental one (3.1 eV) Figure 1 shows the initial laser
field (red dashed line in (a)), reflected and transmitted
field (blue solid lines), and the energy density (green filled
lines). We define the energy of the ground state as 0
eV/A3.
The cubic unit cell containing eight carbon atoms was
discretized into grids of 163. The Bloch k-space was also
discretized into 163 grid points. The macroscopic mesh
size for the Maxwell Eq. (Eq. (1)) is 13 nm. The time
step is set to 0.04 atomic unit (0.97 attosecond).
The absorbed energy Eex is defined as the difference
of Etot(t) between its initial and final values, where
Figure 2 shows the change in the electron distribution,
~k
~k
~k
δO
X,i = O
X,i(Te) − O
X,i(0),
(6)
in the form of the density of states (DoS). Red color indi-
cates the excited electron in the conduction band (CB),
whereas the blue indicates the hole in the valence band
(VB). The color map in the area deeper than 0.3 µm is
100 times enhanced.
We can estimate the quasi-temperatures of the electron
and hole with the QTM, and the temperature of the total
system with the STM24. The reduced internal energies
in the conduction and valence bands, given by UX,c and
UX,v, respectively, can be defined as
UX,v(c) = X~k,i=v(c)
~k
~k
X,i,
X,i(Te)ǫ
O
(7)
!"#$%"
%
%!&
%!"
"!&
"!"
%!"#$%"
9
8
7
6
5
%
4
#
3
2
-
1
0
.
/
#
,
.
-
,
,
+
*
"!'
"!(
"!)
"!
"!"
%!"#$%"
"!'
"!(
"!)
"!
"!"
4+9
&$%"
%
#A567
#*+,,-.,#/.01-23
#:.
#:?
#:2@2
"!
"!)
"!(
"!'
4B9
%$%"
%8
#A567
"!
"!)
"!(
"!'
469
$%"
%8
#A567
"!
"!)
"!(
"!'
=#4>79
!&
!"
%!&
%!"
"!&
"!"
%!"
!&
:
.
7
;
!
#
4
.
<
9
!"
%!&
%!"
"!&
"!"
%!"
!&
!"
%!&
%!"
"!&
"!"
%!"
FIG. 3. Position dependence of the carrier density (filled red),
electron (blue), and hole (green) temperatures in the quasi-
temperature model (QTM). The electron temperature in the
single-temperature model (STM) is presented by the black
dashed lines.
where v(c) represents the states in the valence (conduc-
tion) band, and ǫ~k
X,i is the energy eigenvalue of the i-th
state at X. We can assume the quasi-temperatures of
carrier (Te) and hole (Th) at each X from UX,v(c), and
the temperature (Ttot) from the UX,tot = UX,v + UX,c.
Figure 3 shows the carrier density, quasi-temperatures,
and temperature as functions of position from the sili-
con surface. The color map of the density is 100 times
enhanced in regions deeper than X > 0.3 µm. At the
surface, the electron (red) and hole (blue) spread into a
wide energy range as the laser intensity increases. How-
ever, in the region X > 0.1 µm, the electron and the hole
states have specific peaks whose energy positions do not
depend on the position and laser intensity.
The position-independent behavior at X > 0.1 µm and
laser intensity dependence at the surface are presented in
Fig. 2, and the position-dependent quasi-temperatures
are shown in Fig. 3. Around the surface (X < 0.1µm),
the quasi-temperatures show an exponential decrease as
X increases. However, in the deeper region, the electron
and hole show approximately same quasi-temperatures
around 1.0 eV. It should be noted that the carrier density
and the electron temperature in the STM show a mono-
tonic decrease, as predicted by the previous models. We
can also see a small dip in the hole quasi-temperature
around X = 0.05 µm in Fig. 2 (b) and (c), which is the
1
<
;
,
4
/
)
:
5
8
9
)
-
8
7
6
-
5
3
4
7
5
6
)
5
.
0
7
0
A
@
?
>
:
5
=
4
/01
&'!
&'"
&'#
&'$
&'%
&'&
&'&
/21
!
"
#
$
%
&
&'&
)!*%&%$)+,-.$
)%*%&%#
)$*%&%#
3
&'$
&'"
&'
&'(
%'&
&'$
&'"
&'
&'(
%'&
)B)/C.1
FIG. 4. Position dependence of the (a) maximum field inten-
sity and (b) Keldysh parameter with each initial laser inten-
sities.
transient region between the exponential decrease and
the position-independent region. This dip in the hole
quasi-temperature corresponds to the intense single peak
in hole density.
The dip in the hole quasi-temperature indicates the
excitation process dependence in the electron-hole dis-
tribution.
In general, the photo-excitation process can
be attributed to regime, that is, multi-photon absorption
and tunneling process. Since an intense laser field in-
duces the tunneling process preferentially, the tunneling
process occurs around the surface. The Keldysh parame-
ter γ = ωpm∗Egap/eE 25 is a good parameter to classify
the excitation process. Here, m∗ is the effective mass,
Egap is the band gap, and E is the electric field. The
tunneling process (multi-photon absorption) is dominant
for γ ≪ 1 (γ ≫ 1).
Figure 4 shows the (a) maximum field and (b) γ at
X. While the weakest laser intensity (5 × 1012 W/cm2)
shows γ > 1 for all the X points, γ across the one around
X = 0.05 µm with a laser intensity of 1×1013 and 2×1013
W/cm2. The positions of γ ∼ 1 (X = 0.05 µm) corre-
spond to the dip in the hole quasi-temperature, as shown
in Fig. 3. This result indicates that the hole and/or
electron distribution change by the excitation process.
The tunneling process gives a broader distribution of the
electron-hole pairs, which correspond to a higher quasi-
temperature. However, the multiphoton process gives
specific peaks of electron-hole pairs, which correspond to
the laser intensity -- independent lower quasi-temperature.
Although the hole quasi-temperature shows a complex
X-dependence, the electron quasi-temperature (Te) may
be approximated by the simple double exponential func-
tion,
Tc(X) ∼ T 0 + T1 exp(cid:20)−
X
τ1(cid:21) + T2 exp(cid:20)−
X
τ2(cid:21) ,
(8)
*7-
-
,
+
*
)
(
'
*<-
-
,
+
*
)
+
6
9
8
7
6
5
2
+
'
!
!"
#!$
#!%
#!&
#!
#!"
#!
#!"
"!$
"!%
"!&
"!
)'"
)'#
)'
) !".#"
)#!4.#"
)#!".#"
)"!4.#"
#/)
#/)
#/)
01(2
)))
))))))))
))))))))
#/)
)#!3.#"
)#! .#"
)"!3.#"
#/)
#/)
#/)
!&
!%
!$
!#
!"
'()*+,
*(-
4
&
/
-
2
;
*
)
:
"!#
%
4
&
/
):#
):
"!"
"
-
.
/
0
&$
&
"!"#
%
"
-
.
/
0
&$
&
%
123425637()89:+
%
,
%
,
123425637()89:+
FIG. 5. (a) Te(X) and the fitted function assuming double
exponential (Eq. (8)). (b) Fitting parameters for the temper-
atures, T0, T1, and T2. (c) Damping factors for the surface
(τ1) and deeper region (τ2).
because there are two different excitation processes.
Here, T1 and τ1 are the peak temperature and the damp-
ing factor at the surface, T2 and τ2 are the parameters in
a deeper region, T0 is the parameter for saturated tem-
perature. Figure5 (a) shows the computational results
(solid lines) and fitted function (dashed lines). The fitted
function well reproduces Te(X). The fitting parameters
are shown in Fig. 5 (b) and (c).
T0 does not depend on the laser intensity, which is
consistent with the results in Fig. 3. T0 is the dominant
parameter at a deeper region, because T2 is small. T1,
which corresponds to Te(X = 0), increases monotonically
as the function of laser intensity. T1 and T2 are close to
each other at 5 × 1012 and 7 × 1012 W/cm2. This sim-
ilarity indicates that the multiphoton absorption is the
dominant excitation process at such intensities. How-
ever, above 1 × 1013 W/cm2, the deviation between T1
and T1 indicates that a change in the excitation process
occurs at the surface, as shown in Fig. 4. The damping
parameter τ1 decreases as the laser intensity increases,
which corresponds to the steep decrease of Te around the
surface.
B.
3C-SiC
SiC is attracting interest as a foundation base in the
next generation, because of its wide band gap (2 ∼ 3
eV), temperature resistance, good thermal conductivity,
as well as impact resistance26 -- 29. However, for its hard-
ness and chemical and mechanical stability, SiC is dif-
ficult to process. The processing by the femtosecond
laser pulse enables the processing of SiC by the nonlinear
processes30.
Figure 6 shows the dielectric function of 3C-SiC with
,
*
)
+
)
*
*
)
(
'
&
%
!
"!
#!
!
$#!
"
4
37&89:
3;(89:
/
.
-
#!
01&'2,34&56
FIG. 6. Dielectric function of 3C-SiC.
!G3==H7=!87<IHA?
!J7
!JB
!JA@A
284
$&*
$&)
$&%
#&
#&(
#&'
274
$&*
$&)
$&%
#&
#&(
#&'
264
L!@00KE3AH@<!2#/7F4
,
,(!"#$
$
(
234
(-#$
#*
!./01
+$
*$
$
#$
$
,#$
$&$
$&*
$&)
$&%
#&
#&(
#&'
'+$
254
#+
#-#$
!./01
+$
*$
$
#$
$
,#$
$&$
$&*
$&)
$&%
#&
#&(
#&'
'+$
204
#+
-#$
!./01
+$
*$
$
#$
$
,#$
4
F
7
2
!
E
@
A
!
7
0
<
7
3
C
!
D
7
B
A
!
1
@
=
6
!
?
>
=
7
<
;
*
4
1
0
/
#
2
!
?
A
H
I
<
7
8
!
=
7
H
=
=
3
G
%
#
$
#
"
!
#
$
%
#
$
#
"
!
$&$
#
$
%
#
$&$
$
#
"
!
+
$
$&$
$&*
$&)
$&%
#&
#&(
#&'
'+$
$&$
$&*
$&)
$&%
#&
#&(
#&'
9!2:14
M@N!2#/7F4
9!2:14
#$
)
+
#
)
+
$&#
#$
)
+
#
)
+
$&#
#$
)
+
#
)
+
$&#
J
7
1
E
7
=
3
A
K
=
7
!
2
7
F
4
FIG. 7. (a) -- (c) The change of electron occupation in 3C-SiC
as a function of the position from the surface together with
the DoS. (d) -- (f) The carrier density, quasi-temperatures (Te
and Th), and temperature (Ttot).
LDA potential. From the imaginary part of the dielectric
function (Im[ε]), the optical band gap is 4.2 eV.
In the macroscopic calculation, we discretized a cubic
cell, including 4 carbon and 4 silicon atoms, into 20 ×
20 × 20 grids. We used 8 × 8 × 8 K-sampling grids for the
Bloch phase space. Eighty-four conduction bands were
prepared to calculate the excited electron distribution.
7
shows
Figure
(a) -- (c)
the position-dependent
electron-hole distribution at three different laser inten-
sities. The hole density has a peak at -3 eV and the elec-
tron density has a peak at 3 eV at a deeper region, whose
energy difference (6 eV) corresponds to four-photon ab-
sorption.
Although the quasi-temperatures of silicon show a
smooth decrease with increasing X, 3C-SiC shows a step-
#5
#>
,$:#"
,#:#"
, :#"
#>
,;<=0
,;<=0
,;<=0
!"
#!$
#!"
"!$
/
&
1
&
0
.
/
.
-
,
+
*
)
(
&
%
'
"!"
"!"
"!5
"!4
"!3
#!
#!$
#!2
6,7809
FIG. 8. Position dependence of the Keldysh parameter in
3C-SiC.
wise structure in Fig. 6 (d) -- (f). The abrupt change in Te
and Th corresponds to the change of the number of con-
tributing bands. The stepwise structure indicates that
the band structure affects the electron-hole distribution
and temperature. However, the carrier density and Ttot
show monotonic behavior as X increases. At all inten-
sities, Th becomes position-independent at X > 0.3 µm,
which corresponds to the behavior of the Keldysh param-
eter, as shown in Fig. 8.
While Te is much higher than Th in silicon, Th of 3C-
SiC becomes much higher than Te. Although Th at a
deeper region is 4 eV at all laser intensities, which cor-
responds to hole creation at same energy point, Te de-
creases as X increases. From Fig. 7 (a)-(c), the elec-
tron at 5 eV survives even at a relatively deeper position
(weak field intensity). These results indicate that the
quasi-temperatures depend strongly on the material, in
other words, band structure.
IV. SUMMARY
In summary, we studied the spatial dependence of
the quasi-temperatures in silicon and 3C-SiC by em-
ploying the time-dependent Kohn -- Sham equation and
Maxwell equation. We found that the quasi-temperature
and the electron-hole distribution depend on the exci-
tation process. While the quasi-temperatures do not
show any spatial dependence where multiphoton absorp-
tion is dominant, they increase at the surface where the
tunneling excitation process is important. Although we
found that the carrier quasi-temperature can be approxi-
mated by a simple double exponential function in silicon,
3C-SiC shows a stepwise spatial distribution in quasi-
temperatures. These results indicate that the estimation
of the quasi-temperatures from the quantum mechanical
simulation is important to employ electron-lattice mod-
els, such as the two-temperature model, to understand
the initial state of the laser processing and/or laser dam-
age.
5
field dynamics without any artificial parameters. How-
ever, we should contain lattice dynamics induced by the
electron excitation, which occurs in longer time scale,
to simulate the laser processing. We can understand
the early stage of the laser processing by combining our
method and classical molecular dynamics or fluid dynam-
ics in the future.
ACKNOWLEDGEMENT
This work was supported by MEXT Quantum Leap
Flagship Program (MEXT Q-LEAP) Grant Number JP-
MXS0118067246, JST-CREST under grant number JP-
MJCR16N5, and by JSPS KAKENHI Japan Grant Num-
bers JP17H03525. The numerical calculations were per-
formed on the supercomputer SGI ICE X at the Japan
Atomic Energy Agency (JAEA). We would like to thank
Editage (www.editage.com) for English language editing.
D.
G.
Du,
and
Liu,
Mourou,
1B. N. Chichkov, C. Momma, S. Nolte, F. von Alvensleben, and
A. Tunnermann, Applied Physics A 63, 109 (1996).
2B. C. Stuart, M. D. Feit, S. Herman, A. M. Rubenchik, B. W.
Shore, and M. D. Perry, Phys. Rev. B 53, 1749 (1996).
3X.
IEEE Journal of Quantum Electronics 33, 1706 (1997).
4M. Lenzner, J. Kruger, S. Sartania, Z. Cheng, C. Spiel-
mann, G. Mourou, W. Kautek,
and F. Krausz,
Phys. Rev. Lett. 80, 4076 (1998).
5M. Geissler, G. Tempea, A. Scrinzi, M. Schnurer, F. Krausz, and
T. Brabec, Phys. Rev. Lett. 83, 2930 (1999).
6M. Lenzner, F. Krausz, J. Kruger,
Applied Surface Science 154-155, 11 (2000).
7L. Sudrie, A. Couairon, M. Franco, B. Lamouroux,
B.
and A. Mysyrowicz,
Phys. Rev. Lett. 89, 186601 (2002).
8G. Doumy, F. Qu´er´e, O. Gobert, M. Perdrix, P. Martin, P. Au-
debert, J. C. Gauthier, J.-P. Geindre,
and T. Wittmann,
Phys. Rev. E 69, 026402 (2004).
9S. Amoruso, G. Ausanio, R. Bruzzese, M. Vitiello, and X. Wang,
Phys. Rev. B 71, 033406 (2005).
and W. Kautek,
S. Tzortzakis,
Prade,
10R.
R.
Gattass
and
E.
Mazur,
Nature Photonics 2, 219 EP (2008).
11E. Gamaly, Physics Reports 508, 91 (2011).
12B. Chimier, O. Ut´eza, N. Sanner, M. Sentis, T.
P. Lassonde, F. L´egar´e, F. Vidal,
Phys. Rev. B 84, 094104 (2011).
Itina,
and J. C. Kieffer,
13H. Li and H. Ki, Journal of Applied Physics 100, 104907 (2006),
https://doi.org/10.1063/1.2388853.
14B. Rethfeld, D. S. Ivanov, M. E. Garcia, and S. I. Anisimov,
Journal of Physics D: Applied Physics 50, 193001 (2017).
15S. I. Snishimov, B. L. Kapeliovicj, and T. L. Perel'man, Sov.
Phys.-JETP 39, 375 (1975).
16P. B. Allen, Phys. Rev. Lett. 59, 1460 (1987).
17E.
Silaeva,
Vella,
A.
P.
G. Martel,
New Journal of Physics 14, 113026 (2012).
B. Deconihout,
N.
and T.
Sevelin-Radiguet,
Itina,
E.
18L. Gallais, D.-B. Douti, M. Commandr´e, G. Batavici¯ute,
E. Pupka, M. Sciuka, L. Smalakys, V. Sirutkaitis,
and
A. Melninkaitis, Journal of Applied Physics 117, 223103 (2015),
https://doi.org/10.1063/1.4922353.
19M. Noda, S. A. Sato, Y. Hirokawa, M. Uemoto, T. Takeuchi,
S. Yamada, A. Yamada, Y. Shinohara, M. Yamaguchi,
K.
Ishimura,
T. Boku, G. F. Bertsch, K. Nobusada,
and K. Yabana,
Computer Physics Communications (2018), https://doi.org/10.1016/j.cpc.2018.09.018.
I. Floss, T. Otobe, K.-M. Lee, K.
Iida,
20G. F. Bertsch, J.-I.
Iwata, A. Rubio,
and K. Yabana,
Our method treats the electron and electromagnetic
Phys. Rev. B 62, 7998 (2000).
21K. Yabana, T. Sugiyama, Y. Shinohara, T. Otobe, and G. F.
Bertsch, Phys. Rev. B 85, 045134 (2012).
22J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981).
23T. Otobe, M. Yamagiwa, J.-I. Iwata, K. Yabana, T. Nakatsukasa,
and G. F. Bertsch, Phys. Rev. B 77, 165104 (2008).
24T.
Otobe,
T.
Hayashi,
and M.
Nishikino,
Applied Physics Letters 111, 171107 (2017),
https://doi.org/10.1063/1.4997363.
25L. V. Keldysh, Sov. Phys.-JETP 20, 1307 (1965).
26C. Persson and U. Lindefelt, Phys. Rev. B 54, 10257 (1996).
27H. Morko¸c, S. Strite, G. B. Gao, M. E. Lin, B. Sverdlov,
Journal of Applied Physics 76, 1363 (1994),
and M. Burns,
https://doi.org/10.1063/1.358463.
6
28H.
Matsunami
Kimoto,
Materials Science and Engineering: R: Reports 20, 125 (1997),
r20.
and
T.
29N. T. Son, P. N. Hai, W. M. Chen, C. Hallin, B. Monemar, and
E. Janz´en, Phys. Rev. B 61, R10544 (2000).
30I. Choi, H. Y. Jeong, H. Shin, G. Kang, M. Byun, H. Kim,
A. M. Chitu, J. S. Im, R. S. Ruoff, S.-Y. Choi, and K. J. Lee,
Nature Communications 7, 13562 EP (2016).
|
1907.00195 | 1 | 1907 | 2019-06-29T12:25:48 | Modeling the Generic Breakthrough Curve for Adsorption Process | [
"physics.app-ph"
] | This work is aimed at understanding the basic principles of adsorption process in great details as adsorptive separation process has broad applications in the industry. To this end, a simple mathematical model has been used to describe transient fixed bed physical adsorption process. Governing equations are solved numerically to obtain breakthrough curves for single component and multi-component monolayer adsorption. Desorption of a saturated bed by an inert fluid is also considered. A full parametric study is performed to analyze the effects of different parameters such as bed length, velocity, diffusivity, particle radius and isotherm properties on the nature of the breakthrough curve. Analysis of these results led to the development of the generic breakthrough curve for a single component monolayer adsorption which will enable us to tell the nature of breakthrough curve for different process parameters without recourse to the numerical simulation or experiment. Thus this study will be of great interest in the industrial separation process. | physics.app-ph | physics |
Modeling the Generic Breakthrough Curve for
Adsorption Process
Saikat Roya,∗, Arun S. Moharirb
aDept of Chemical Physics, The Weizmann Institute of Science, Israel
bIndian Institute of Technology-Bombay, India
Abstract
This work is aimed at understanding the basic principles of adsorption pro-
cess in great details as adsorptive separation process has broad applications
in the industry. To this end, a simple mathematical model has been used to
describe transient fixed bed physical adsorption process. Governing equations
are solved numerically to obtain breakthrough curves for single component and
multi-component monolayer adsorption. Desorption of a saturated bed by an
inert fluid is also considered. A full parametric study is performed to analyze
the effects of different parameters such as bed length, velocity, diffusivity, par-
ticle radius and isotherm properties on the nature of the breakthrough curve.
Analysis of these results led to the development of the generic breakthrough
curve for a single component monolayer adsorption which will enable us to tell
the nature of breakthrough curve for different process parameters without re-
course to the numerical simulation or experiment. Thus this study will be of
great interest in the industrial separation process.
Keywords: Adsorption, Generic Breakthrough curve, Mathematical
model, Linear driving force
1. Introduction
Every solid surface has a discontinuous structure where unsaturated forces
act. When the solid is exposed to a fluid, the fluid molecules get attached to
the surface by forming bonds. This phenomenon is known as adsorption. In
the past two decades adsorption has become a key process for fluid separation
in the chemical and petrochemical industries following the developments of new
adsorbents and new process cycles. Invention of synthetic zeolites or molecular
sieve carbon marked the beginning of a new era in separation process by ad-
sorption. These types of adsorbents preferentially adsorb nitrogen over oxygen
by an approximate factor of three. So this can be a viable alternative to the
∗Corresponding author
Email address: [email protected] (Saikat Roy)
Preprint submitted to Journal of LATEX Templates
July 2, 2019
conventional cryogenic air separation process with the help of an efficient pro-
cess cycle. Regeneration of the process is also important for the reuse of the bed
and recovery of the adsorbed fluid. Regeneration by heating is not so efficient
as heating-cooling cycle is time consuming. So this led to the development of
efficient pressure swing adsorption cycles [1, 2]. This is a very energy efficient
cycle. Regeneration can take place within minutes or seconds by lowering the
pressure. This allows higher throughput of gas to be separated. Cryogenic air
separation requires liquefaction of air followed by distillation. Although this is
the most frequently used process for large scale air separation, it is not so energy
efficient as a low temperature and a high pressure is required for liquefaction.
Following criteria describe where adsorption can be used over distillation.
• The relative volatility between the key components to be separated is
in the order of 1.2 to 1.5 or less. Separation of isomers is an example.
Adsorption separation factor for separation of isomers by zeolite is infinite.
• The two groups of components to be separated have overlapping boiling
point. This will require several distillation columns for the desired sepa-
ration. The said separation can efficiently be done by adsorption if the
two groups contain chemically or geometrically dissimilar molecules.
• Major cost of a pressure swing adsorption is the compressor costs. If the
feed is available at an elevated pressure, cost will be drastically reduced.
It has been observed that it is better to opt for the adsorption separation over
distillation for small to medium throughput and when high purity products are
not required. As the development of new adsorbent and modification of process
cycles takes place, pressure swing adsorption will compete with distillation at
high throughput[3]. Based on the method of adsorbent regeneration, adsorptive
separation can be categorized in the following process cycles
• Temperature swing adsorption (TSA).
• Pressure swing adsorption (PSA).
• Inert purge cycle.
• Displacement purge cycle.
Adsorptive separation has been used extensively in industry for different pur-
poses like drying of cracked gas, ethylene; n-Paraffin recovery from naphtha
and kerosene; aromatic separation; oxygen and nitrogen production from air;
high purity hydrogen production from steam reformer products, solid liquid
separation and many more. Zeolite, activated carbon, molecular sieve carbon,
activated alumina are generally used as adsorbents in industry for the aforemen-
tioned purposes. Most of the adsorptive separation takes place in a fixed bed
packed with adsorbent particles. In order to find out the fluid phase adsorbate
concentration profile within the bed and also to find out the effluent concen-
tration history i.e. breakthrough curve a model is required to account for the
2
physical processes happening in the bed. In this report a mathematical model
[3, 4]has been used with proper boundary and initial conditions to predict the
breakthrough nature for single component monolayer adsorption onto an adsor-
bent. It has also been extended for the multi-component case. And also the
breakthrough curve for desorption by an inert is predicted through the same
model with different boundary and initial conditions. Although many experi-
mental and simulation studies exist in the literature to find the breakthrough
nature for the adsorption process [5, 6, 7, 8, 9] , but most of the studies are
confined to specific set of cases. In this work, we propose a model to find out a
generic breakthrough curve which will predict breakthrough nature of adsorp-
tion for different values of process variables without recourse to the numerical
simulation or experiment.
2. Breakthrough nature for single component adsorption
2.1. Mathematical modelling
In the present work following assumptions are made to model the fixed bed
adsorption process:
1. The system operates under isothermal condition.
2. Negligible pressure drop through the adsorbent bed.
3. Velocity is assumed to be constant throughout the bed.
4. Langmuir isotherm is valid for the system.
5. Ideal plug flow is assumed; i.e there is no axial or radial dispersion.
6. The mass transfer rate is represented by a linear driving force expression
[10].
7. The bed is clean initially.
Based on the above said assumptions, for the control volume A × dz and for
the limit z → 0, the net rate of accumulation is given as,
u
∂c
∂z
+
∂c
∂t
+
1 − ǫ
ǫ
ρp
∂q
∂t
= 0
(1)
whereas c is the concentration in fluid phase (kg/m3), ǫ is the bed porosity , ρp
is adsorbent density (kg/m3), q is adsorbed phase concentration (kg adsorbed/
kg of adsorbent).
The mass transfer kinetics is modeled using the LDF, Linear driving force
approximations, based on the simplifications of Fick's second law of diffusion,
∂q
∂t
=
15De
R2
p
(q∗ − q)
(2)
whereas De is intracrystalline diffusivity, Rp is adsorbent particle radius, q∗
is equilibrium concentration of adsorbed phase (kg adsorbed/ kg of adsorbent).
The adsorption isotherm is described by Langmuir,
q∗ =
qmbc
1 + bc
3
(3)
Table 1
Parameter
L,Bed length,m
ǫ
u, Velocity ,m/s
15De
p
R2
b
qm
co, inlet conc. of the adsorbate in the feed , kg/m3
ρp
Value
0.5
0.4
0.01
0.5
0.3
0.04
1
1000
whereas qm is maximum adsorption capacity (kg adsorbed / kg of adsorbent)
and b is Langmuir isotherm constant (m3/kg). The following initial conditions
are considered,
c = co
q = 0
c = 0
z = 0, t = 0
0 < z ≤ L, t = 0
0 < z ≤ L, t = 0
The boundary conditions are given as ,
c = co
z = 0, t > 0
(4)
(5)
We now substitue Eqn. 2 in Eqn. 1 and on using expression for isotherm(Eqn. 3)
we get the following coupled PDEs,
u
∂c
∂z
+
∂c
∂t
+
1 − ǫ
ǫ
ρp
15De
R2
1 + bc
p (cid:18) qmbc
− q(cid:19)
− q(cid:19) = 0
(6)
(7)
∂q
∂t
=
15De
p (cid:18) qmbc
1 + bc
R2
Simulation technique:. Preceding set of partial differential equations is initially
discretized by finite difference method to obtain a set of algebraic equations.
Then it is solved by explicit Euler method to get the concentration profile for
both in fluid and solid phase as a function of time and space and also to obtain
outlet concentration profile with respect to time. A mathematical algorithm to
solve these coupled equations is developed and implemented into a computer
program using MATLAB software.
4
1
0.8
0.6
0.4
0.2
o
c
/
c
0
0
200
400
600
t
800
1000
1200
Figure 1: Breakthrough curve for single component adsorption
Results:. The breakthrough curve obtained from the simulation is shown in Fig.
1. The values of the parameter used in the simulation are given in Table 1. From
the breakthrough curve we can find out at what time 90% or 100% breakthrough
is taking place and accordingly feed will be switched to another bed until the
saturated bed gets regenerated. Fluid phase concentration profile of adsorbate
within the bed at a particular time is also shown in Fig. 2
3. Single component desorption by an inert fluid
Mathematical modeling. Governing equations for desorption is same as that
of adsorption. Only initial conditions and boundary conditions are different.
So the same model described earlier is used. Modified boundary and initial
conditions are given as follows,
Boundary condition:
c = 0
z = 0, t > 0
Initial conditions:
q = q∗
o
c = co
0 < z ≤ L, t = 0
0 < z ≤ L, t = 0
5
(8)
(9)
o
c
/
c
1
0.8
0.6
0.4
0.2
0
0
Time=300 s
0.2
0.4
0.6
0.8
1
z/L
Figure 2: Fluid phase concentration profile in the bed at a particular time
Results. The breakthrough curve for desorption of single component by an inert
fluid is shown in Fig 3. Same values of the parameters given in Table1 are used
for simulation whereas q∗
o is given as,
q∗
o =
qmbco
1 + bco
= 0.0092
kg adsorbed
kg adsorbent
(10)
If we plot both adsorption and desorption breakthrough curve (see Fig 4) on
the same graph, then we can see that the curves are asymmetric for nonlinear
isotherm. As the value of b increases, asymmetry of the curves increases. But
for linear isotherm curves are symmetric. For linear isotherm ordinate value of
point of intersection between adsorption and desorption breakthrough curves
is 0.5. But for nonlinear isotherm, it is always below 0.5. As the value of b
increases it dips further below 0.5.
4. Multi-component adsorption breakthrough curve
Mathematical modeling:. Model principles and assumption will remain same
for this case, but here governing equation are solved for all the components.
Binary component adsorption is considered here. We consider a case where
linear driving force of both the components are equal with different isotherm
properties. Governing equations are as follows:
u
∂c1
∂z
+
∂c1
∂t
+
ρp
∂q1
∂t
= 0
1 − ǫ
ǫ
6
(11)
1
0.8
0.6
0.4
0.2
o
c
/
c
0
0
200
400
800
1000
1200
600
t
Figure 3: Desorption breakthrough curve for single component
1
0.8
0.6
0.4
0.2
o
c
/
c
Desorption
Adsorption
0
0
200
400
800
1000
1200
600
t
Figure 4: Adsorption and Desorption breakthrough curve for single component
7
u
∂c2
∂z
+
∂c2
∂t
+
1 − ǫ
ǫ
ρp
∂q2
∂t
= 0
∂q1
∂t
∂q2
∂t
= (LDF )1 (q∗
1 − q1)
= (LDF )2 (q∗
2 − q2)
(12)
(13)
(14)
whereas adsorption isotherms are described by extended Langmuir isotherm,
q∗
1 =
q∗
2 =
qm1 b1c1
1 + b1c1 + b2c2
qm2 b2c2
1 + b1c1 + b2c2
(15)
(16)
The following initial conditions are considered,
c1 = c10
c2 = c20
q1 = 0
q2 = 0
c1 = 0
c2 = 0
z = 0, t = 0
z = 0, t = 0
0 < z ≤ L, t = 0
0 < z ≤ L, t = 0
0 < z ≤ L, t = 0
0 < z ≤ L, t = 0
Boundary conditions for the problem is given as,
c1 = c10
c2 = c20
z = 0, t > 0
z = 0, t > 0
Governing equations are solved along with the adsorption isotherms and
boundary conditions as well as initial conditions to yield the breakthrough curve
for two components competitive adsorption.
Results:. Value of the parameters for simulation is given in Table 2. And the
breakthrough curves obtained from simulation is shown in Fig 5. From the
plot it is found that for one component effluent concentration has exceeded feed
concentration. This phenomenon is known as roll up or roll over. For this case
roll up is caused by the displacement of a weaker adsorbate by a stronger one.
5. Generic breakthrough curve
In order to determine the breakthrough nature for adsorption process under
different operating conditions we need to solve the model equations for each
case. But if we can capture what change in the breakthrough curve is brought
about by the change in the value of an operating variable, then it is possible
to say what would be the nature of a new breakthrough curve for the changed
8
Table 2
Parameter
L,Bed length,m
ǫ
u, Velocity ,m/s
(LDF )1 (sec)−1
(LDF )2 (sec)−1
b1 m3/kg
b2 m3/kg
qm1 kg adsorbed/kg adsorbent
qm2 kg adsorbed/kg adsorbent
c10, inlet conc. of the component 1 in the feed , kg/m3
c20, inlet conc. of the component 1 in the feed , kg/m3
ρp kg/m3
Value
0.3
0.4
0.01
1.5
1.5
0.4
0.3
0.04
0.03
0.75
0.5
800
o
c
/
c
1.2
1
0.8
0.6
0.4
0.2
0
0
Component 1
Component 2
200
400
600
800
1000
t
Figure 5: Multicomponent adsorption breakthrough curve
9
value of the variable with reference to a known breakthrough curve without
doing numerical simulation. This can be done for different variables and then
the results obtained for each case can be clubbed together to accommodate the
change in the breakthrough curve if all the variables change simultaneously. So
initially a parametric study is performed to find out the influence of different
parameters on the breakthrough curve and the curves are properly rescaled to
obtain a generic breakthrough curve. To accomplish this objective, the same
mathematical model as described earlier has been used to describe transient
adsorption process in a fixed bed adsorber for a single component adsorption
onto a monolayer adsorbent. After nondimensionalization of model equations, it
has been solved numerically with proper boundary and initial conditions. Then
a parametric study is done to generate the the universal breakthrough curve.
Mathematical modeling. Same model described earlier is used here. But here
governing equations, boundary conditions and initial conditions are nondimen-
sionalized. Parameters used for nondimensionalization and nondimensionalized
equations, boundary conditions and initial conditions are as follows,
Dimensionless parameters:
c′ =
c
c0
; t′ =
15Det
R2
p
; z ′ =
15De
p (cid:16) z
u(cid:17)
R2
ρ′
p = (cid:18) 1 − ǫ
ǫ (cid:19) ρp
c0
; b′ = bc0
This leads to the following nondimensionalized equations
∂c′
∂z ′ +
∂c′
∂t′ + ρ′
∂q
1 + b′c′
p(cid:18) qmb′c′
∂t′ = (cid:18) qmb′c′
− q(cid:19) = 0
− q(cid:19)
1 + b′c′
Modified initial and boundary conditions are as follows ,
Initial conditions:
c′ = 1
z ′ = 0, t′ = 0
q = 0
0 < z ′ ≤
c′ = 0
0 < z ′ ≤
15De
R2
p (cid:18) L
p (cid:18) L
u(cid:19) , t′ = 0
u(cid:19) , t′ = 0
15De
R2
Boundary conditions are given as;
c′ = 1
z ′ = 0, t′ > 0
10
(17)
(18)
(19)
(20)
′
c
1
0.8
0.6
0.4
0.2
0
0
50
100
150
200
250
t′
Figure 6: Reference breakthrough curve
Table 3
Parameter
L,Bed length,m
ǫ
u, Velocity ,m/s
(LDF ) (sec)−1
b m3/kg
qm kg adsorbed/kg adsorbent
c0, inlet conc. of the adsorbate in the feed , kg/m3
ρp kg/m3
Value
0.3
0.4
0.005
0.5
0.1
0.02
1.5
1100
11
′
c
1
0.8
0.6
0.4
0.2
0
0
L=0.2
L=0.3
L=0.4
50
100
150
200
250
t′
Figure 7: Effect of bed length on the breakthrough curve
Results and analysis. At the outset the model equations are solved for some
chosen values of different parameters. Now as breakthrough curve is highly
sensitive to different parameters like bed length, superficial velocity, diffusivity,
adsorbent particle radius, isotherm properties of adsorbent etc., we will initially
see how breakthrough pattern changes with change in the value of these parame-
ters separately and a function is developed in each case to predict breakthrough
pattern with reasonable accuracy for different scenarios. Then all the param-
eters are varied simultaneously and accordingly rules are set up to predict the
nature of the breakthrough for different cases. First a reference breakthrough
curve is produced and values of the parameters for the reference breakthrough
curve are given in Table 3. The reference breakthroiugh curve is shown in Fig 6.
Effect of bed length, L. Fig 7 shows how the bed length affects the breakthrough
pattern. It is seen from the plot that as the bed length increases breakthrough
point also shifts towards right along the time scale. Smaller bed length corre-
sponds to lesser amount of adsorbent. Consequently a smaller capacity for the
bed to adsorb and the bed gets saturated in less time.
Effect of LDF (cid:0)15De/Rp2(cid:1). Fig 8 upper panel shows how LDF affects break-
through nature. It is seen from the plot that as LDF increases breakthrough
point shifts towards right along the time scale. As LDF is the measure of the up-
take rate of the adsorbent particle, when its value is lower the bed gets saturated
faster than the higher value of LDF.
Effect of velocity,u. Fig 8 lower panel shows that as the velocity increases break-
through point shifts towards the left and also curve becomes much steeper. This
12
1
0.8
0.6
0.4
0.2
′
c
0
0
1
′
c
0.8
0.6
0.4
0.2
0
0
ldf=1
ldf=0.5
ldf=0.25
100
200
300
400
500
t′
u=0.0025
u=0.005
u=0.01
100
200
300
400
500
t′
Figure 8: Upper panel: Effect of LDF (cid:0)15De/Rp2(cid:1), Lower panel: Effect of velocity on the
breakthrough curve
13
u
u
(LDF )×L =0.017
(LDF )×L =0.033
(LDF )×L =0.07
u
10
20
u
(LDF )×L
t′
30
40
Figure 9: Collapse of breakthrough curve for different value of
u
LDF ×L
qm =0.08
qm =0.06
qm =0.04
qm =0.02
′
c
1
0.8
0.6
0.4
0.2
0
0
1
0.8
0.6
0.4
0.2
′
c
0
0
200
400
t′
600
800
Figure 10: Effect of maximum adsorption capacity, qm on the breakthrough curve
14
1
0.8
0.6
0.4
0.2
′
c
0
0
0.5
1
t′q−0.82
m
1.5
2
×10 4
Figure 11: Data collapse of breakthrough curves for different values of maximum adsorption
capacity, qm
1
0.8
0.6
0.4
0.2
′
c
b=0.01
b=0.03
b=0.06
b=0.10
0
0
50
100
t′
150
200
Figure 12: Effect of Langmuir isotherm constant, b on the breakthrough curve
15
1
0.8
0.6
0.4
0.2
′
c
0
0
1000
2000
t′b−0.6
3000
4000
Figure 13: Data collapse of breakthrough curves for different values of Langmuir isotherm
constant, b
is because of the residence time of the solute in the column, which is not long
enough for adsorption equilibrium to be reached at high velocity. So at high
velocity the adsorbate solution leaves the column before equilibrium occurs.
Furthermore, a fixed saturation capacity of bed based on the same driving force
gives rise to a shorter time for saturation at higher velocity.
The effect of velocity, linear driving force constant and the bed length can
be clubbed into a non-dimensional number
LDF ×L and when this number is
multiplied with the dimensionless time scale, then the breakthrough curve col-
lapses onto a single curve for different values of this number i.e different values
of L, u and LDF . The same is shown in Fig. 9
u
Effect of maximum adsorption capacity ,qm. Fig 10 shows that as qm increases
breakthrough point shifts towards the right of the dimensionless time scale.
As qm increases, equilibrium adsorption capacity of the adsorption particle in-
creases. So it takes more time for the bed to get saturated for higher value
of qm. The breakthrough curves for different qm can be collapsed onto a sin-
gle curve for the values considered if the dimensionless time is multiplied with
q−0.82
m
. Such scaling is shown in Fig 11.
Effect of Langmuir isotherm constant,b. Fig 12 shows that as value of b in-
creases, breakthrough point shifts towards the right of the dimensionless time
scale. As the value of b increases, equilibrium adsorption capacity of the adsor-
bent particle increases. So it takes longer time for the bed to get saturated at
higher value of b. The breakthrough curves for different b can be collapsed onto
a master curve for the values considered if the dimensionless time is multiplied
with b−0.6 and re-plotted. The scaling is shown Fig 13. The scaling seems to
deviate as the value of b increases.
16
6. Conclusions
The present work is a building block to the understanding of more com-
plicated pressure swing adsorption process. Basic principles of adsorption and
adsorption process are understood through a simple mathematical model. Solu-
tion of the model yielded breakthrough curves for single component and multi-
component adsorption. Also fluid phase concentration profile of adsorbate is
obtained from simulation. Breakthrough curve for desorption of a saturated bed
by an inert fluid is also obtained after solving model equations with different
initial and boundary condition compared to adsorption. A detailed paramet-
ric study is performed to get an insight on the effects of different influencing
and crucial parameters like bed length, velocity, diffusivity, particle radius and
isotherm properties on the nature of the breakthrough curve for adsorption
process. These results can be very useful in designing of adsorption columns.
Analysis of these results led to the development of generic breakthrough curve
that will enable us to tell the nature of breakthrough curve for different process
parameters without recourse to the numerical simulation for a single component
monolayer adsorption. Thus we can save on computational/experimental time.
Future work will involve more realistic mathematical model and determining
generic breakthrough curve for the same.
7. References
References
[1] S. Jain, A. Moharir, P. Li, G. Wozny, Heuristic design of pressure swing
adsorption: a preliminary study, Separation and Purification Technology
33 (1) (2003) 25 -- 43.
[2] M. Chahbani, D. Tondeur, Mass transfer kinetics in pressure swing adsorp-
tion, Separation and Purification Technology 20 (2-3) (2000) 185 -- 196.
[3] D. M. Ruthven, Principles of adsorption and adsorption processes, John
Wiley & Sons, 1984.
[4] R. T. Yang, Gas separation by adsorption processes, Butterworth-
Heinemann, 2013.
[5] N. Casas, J. Schell, R. Pini, M. Mazzotti, Fixed bed adsorption of co 2/h
2 mixtures on activated carbon: experiments and modeling, Adsorption
18 (2) (2012) 143 -- 161.
[6] R. P. Ribeiro, T. P. Sauer, F. V. Lopes, R. F. Moreira, C. A. Grande, A. E.
Rodrigues, Adsorption of co2, ch4, and n2 in activated carbon honeycomb
monolith, Journal of Chemical & Engineering Data 53 (10) (2008) 2311 --
2317.
17
[7] A. Poursaeidesfahani, E. Andres-Garcia, M. de Lange, A. Torres-Knoop,
M. Rigutto, N. Nair, F. Kapteijn, J. Gascon, D. Dubbeldam, T. J. Vlugt,
Prediction of adsorption isotherms from breakthrough curves, Microporous
and Mesoporous Materials 277 (2019) 237 -- 244.
[8] R. Serna-Guerrero, A. Sayari, Modeling adsorption of co2 on amine-
functionalized mesoporous silica. 2: Kinetics and breakthrough curves,
Chemical Engineering Journal 161 (1-2) (2010) 182 -- 190.
[9] M. K. Al Mesfer, M. Danish, Breakthrough adsorption study of activated
carbons for co2 separation from flue gas, Journal of Environmental Chem-
ical Engineering 6 (4) (2018) 4514 -- 4524.
[10] E. Alpay, D. Scott, The linear driving force model for fast-cycle adsorption
and desorption in a spherical particle, Chemical Engineering Science 47 (2)
(1992) 499 -- 502.
18
|
1910.00625 | 1 | 1910 | 2019-10-01T19:24:23 | Sub-nanosecond spin-torque switching of perpendicular magnetic tunnel junction nanopillars at cryogenic temperatures | [
"physics.app-ph",
"cond-mat.mes-hall"
] | Spin-transfer magnetic random access memory is of significant interest for cryogenic applications where a persistent, fast, low-energy consumption and high device density is needed. Here we report the low-temperature nanosecond duration spin-transfer switching characteristics of perpendicular magnetic tunnel junction (pMTJ) nanopillar devices (40 to 60 nm in diameter) and contrast them to their room temperature properties. Interestingly, at fixed pulse voltage overdrive the characteristic switching time decreases with temperature, in contrast to macrospin model predictions, with the largest reduction in switching time occurring between room temperature and 150 K. The switching energy increases with decreasing temperature, but still compares very favorably to other types of spin-transfer devices at 4 K, with < 300 fJ required per switch. Write error rate (WER) measurements show highly reliable (WER <= 5x10^-5 with 4 ns pulses at 4 K) demonstrating the promise of pMTJ devices for cryogenic applications and routes to further device optimization. | physics.app-ph | physics | Sub-nanosecond spin-torque switching of perpendicular magnetic tunnel junction
nanopillars at cryogenic temperatures
L. Rehm,1, ∗ G. Wolf,2 B. Kardasz,2 M. Pinarbasi,2 and A. D. Kent1, †
1Center for Quantum Phenomena, Department of Physics,
New York University, New York, NY 10003, USA
2Spin Memory Inc., Fremont, CA 94538, USA
(Dated: October 3, 2019)
Spin-transfer magnetic random access memory is of significant interest for cryogenic applications
where a persistent, fast, low-energy consumption and high device density is needed. Here we report
the low-temperature nanosecond duration spin-transfer switching characteristics of perpendicular
magnetic tunnel junction (pMTJ) nanopillar devices (40 to 60 nm in diameter) and contrast them
to their room temperature properties. Interestingly, at fixed pulse voltage overdrive the characteris-
tic switching time decreases with temperature, in contrast to macrospin model predictions, with the
largest reduction in switching time occurring between room temperature and 150 K. The switching
energy increases with decreasing temperature, but still compares very favorably to other types of
spin-transfer devices at 4 K, with < 300 fJ required per switch. Write error rate (WER) measure-
ments show highly reliable (WER ≤ 5×10-5 with 4 ns pulses at 4 K) demonstrating the promise of
pMTJ devices for cryogenic applications and routes to further device optimization.
Spin-transfer torque (STT) magnetic memory elements
are interesting for cryogenic applications, such as com-
puting systems based on superconducting circuits [1], be-
cause they are fast, energy efficient, have a small foot-
print and offer non-volatile data storage [2, 3]. STT
memory devices typically consist of two thin ferromag-
netic layers, one with a magnetization free to reorient
and the other with a fixed magnetization direction both
with perpendicular anisotropy separated by a thin insu-
lating barrier. The memory states are layer magnetiza-
tions aligned either parallel (P) or antiparallel (AP). In-
tense commercial interest has led to the optimization of
perpendicular magnetic tunnel junction (pMTJ) devices
and materials that function near and even above room
temperature [4]. However, pMTJ device characteristics
have not been studied in detail at low temperature.
Recently,
three-terminal cryogenic spin-Hall-based
memory devices have been demonstrated [5]. While these
devices were optimized for low temperature operation
and integration with superconducting circuitry, a two-
terminal pMTJ device has advantages in terms of the
integration density and simplicity of fabrication. Dif-
ferent two-terminal STT all-metallic magnetic memory
elements [6, 7] have also been investigated at low tem-
perature. They have a lower impedance, but they do
not simultaneously offer high switching probabilities and
large readout signals, i.e. large magnetoresistance. In ad-
dition to the advantages already mentioned, pMTJs offer
long-term data storage even for nanopillar junctions just
10 nanometers in diameter and large tunnel magnetore-
sistance (TMR) [8 -- 10].
In conventional STT-MRAM devices operating at or
above room temperature, the angle between the magneti-
zation of layers is always non-zero. This reduces the time
∗ [email protected]
† [email protected]
required to reverse the magnetization and hence reduces
write errors. In other words, elevated temperature helps
the write process but at the same time reduces the data
retention time. For cryogenic memory devices, on the
other hand, a simple macrospin model predicts that in
absence of temperature the switching time would increase
substantially [11]. But this prediction has not been tested
in state-of-the-art perpendicularly magnetized magnetic
tunnel junctions.
In this letter we report the low-temperature high-
speed spin-transfer switching characteristics of pMTJs
and compare them to those at room temperature. We
find that at a fixed pulse voltage overdrive the character-
istic switching time decreases with temperature, in con-
trast to macrospin model predictions. The largest reduc-
tion in switching time occurs between room temperature
and 150 K. Further, at low temperatures there is a fac-
tor two increase in the device magnetoresistance, provid-
ing a much large readout signal. Remarkably, the write
energies (103 fJ, AP→P and 286 fJ, P→AP at 4 K) are
much lower than devices with a metallic write channel,
and thus a lower impedance [5]. Results on nanopillars as
small as 40 nm in diameter are presented, including write
error rate (WER) measurements showing highly reliable
(WER ≤ 5×10-5 with 4 ns pulses at 4 K) demonstrating
the promise of state-of-the-art pMTJ devices for cryo-
genic applications.
We studied pMTJ nanopillars with a perpendicu-
larly magnetized CoFeB composite free layer (FL) con-
sisting of CoFeB layer with a thin W insertion layer,
CoFeB(1.5)/W(0.3)/CoFeB(0.8), where the numbers are
the layer thicknesses in nm. The W insertion layer
increases the perpendicular magnetic anisotropy and
therefore enhances the thermal stability of the de-
vice [12, 13]. This FL is one of the electrodes of
a MgO tunnel
The other electrode is a
CoFeB(0.9) reference layer (RL), which is ferromag-
netically coupled to a first synthetic antiferromagnetic
junction.
9
1
0
2
t
c
O
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
5
2
6
0
0
.
0
1
9
1
:
v
i
X
r
a
2
FIG. 1. a) Schematic of a pMTJ device and the pulse and
readout measurement circuit. Nanosecond duration write
pulses are applied through the capacitive port of a bias tee
while the DC port is used for device read out. b) Resistance
versus perpendicular field free layer hysteresis loop of a 40 nm
diameter device at 4 K. The TMR ratio is 203%. c) Voltage-
induced switching with long duration (100 ms) pulses of the
same device at 4 K in zero applied field. The junction resis-
tance for the data in panels (b) and (c) is measured with a
30 mV DC bias, a bias much less than the switching voltage.
layer (SAF1) (see Fig. 1(a)). The synthetic antiferro-
magnetic layers (SAF) incorporate two antiferromagneti-
cally coupled perpendicularly magnetized layers: (SAF1)
[Pt(0.4)/Co(0.6)]×2 and (SAF2) [Pt(0.4)/Co(0.6)]×7; the
full stack is SAF/RL(0.9)/MgO(1)/FL(2.6). Following
deposition, the wafer was annealed at 400◦C for 25 min.
The annealed wafer was then pattered into circular-
shaped nanopillars of 40, 50, and 60 nm diameter using
a combination of electron beam lithography and Ar ion
beam milling.
The devices are first characterized by measuring their
field and current pulse resistance hysteresis loops. Fig-
ure 1(b) shows the minor hysteresis loops of a 40 nm di-
ameter pMTJ device measured in an applied perpendic-
ular field at 4 K. We observe sharp switching from P to
AP states and vice versa with a field offset of 56 mT, re-
flecting the fringe field from the SAF acting on the free
layer [14]. This sample exhibits a tunnel magnetoresis-
tance (TMR) ratio of 203% and an average coercive field
of 283 mT. Figure 1(c) shows voltage-induced switching
of the same 40 nm diameter device in zero field with
100 ms duration voltage pulses. We observe a bistable
region around zero applied voltage and voltage-induced
switching with pulse amplitudes of 405 mV for AP→P
switching and -358 mV for P→AP switching. Table I
shows the TMR values extracted from the pulsed voltage
FIG. 2. Nanosecond pulsed current switching results at 4 and
295 K. Switching phase diagrams of a 40 nm diameter pMTJ
at 4 K, a) AP→P and b) P→AP, and 295 K, c) AP→P and d)
P→AP. The color in the plot represents the switching prob-
ability, where red corresponds to 0% and black is 100%. The
blue points represent the 50% switching probability and the
solid cyan line shows the fit to the macrospin model described
in the main text.
loops from 4 to 295 K. We observe almost a factor of two
increase of the TMR at 4 K compared to its value at room
temperature, consistent with earlier studies [15, 16].
High speed spin-torque switching was studied by ap-
plying current pulses less than 5 ns in duration using
a pulse generator (Picosecond Pulse Labs 10,070A). A
DAC board (National Instruments PCIe-6353) was used
to apply longer (10 µs) duration pulses to set and reset
the magnetization direction of the free layer. The state
of the device is again determined by applying a small
voltage (30 mV) with the DAQ board and measuring the
resulting junction current. We use a bias-tee (Picosec-
ond Pulse Labs 5575A) to combine low-frequency mea-
surements with the DAQ with nanosecond pulses (see
Fig. 1(a)). All measurements are performed in a cryo-
genic probe station where the sample stage can be heated
up to 150 K. Room temperature measurements are per-
formed in the same setup with the cryostat cold head
turned off.
The measurement procedure thus consists of applying
two square pulses -- reset and write pulses -- with opposite
pulse polarities and reading the junction resistance and
thus the junction state (P or AP) after each pulse. We
start by applying a reset pulse to bring the device to a
known state, either P or AP. We then verified the desired
state by measuring the device resistance. The subsequent
write pulse is applied by the pulse generator and the end
state is determined by measuring the device resistance.
The whole procedure is repeated about 100 times for each
write pulse amplitude and duration combination to deter-
mine the switching probability. We systematically vary
the pulse amplitude and duration to create the phase di-
agrams shown in Fig. 2. We focused our measurements
on the most information rich area of the phase diagram,
the vicinity of the 50% switching probability boundary,
by employing an adaptive measuring strategy [17]. We
1Fig. 1: Measurement setup and characterizationDCBias TeePulseFLSAFMgO-400-20002004003456789Resistance (kW)Field (mT)-900-600-30003006009003456789Resistance (kW)Voltage (mV)a)b)c)RLSAF1SAF22Fig. 2: Ballistic spin-torque switching of 40 nm devicea)b)c)d)700Amplitude (mV)600500400200300800Amplitude (mV)7006005003004000.01.00.80.60.40.2ProbabilityAP→P4 KAP→P295 KP→AP4 KP→AP295 K observed in the 50 and 60 nm diameter pMTJ nanopillars.
The threshold voltage in the macrospin model is given
by:
3
2αeARtµ0MsHk,eff
,
P
Vc =
(2)
where A is the disk area, R is a device resistance, t is
the free layer thickness, P is the spin polarization and
Hk,eff is the effective perpendicular magnetic anisotropy.
Hk,eff = 2K/(µ0Ms) − Ms, the perpendicular anisotropy
associated with spin-orbit coupling K minus the demag-
netization field Ms.
In our fits to the data we find that Vc increases with
decreasing temperature and saturates at temperatures
less than about 150 K for both AP→P and P→AP tran-
sitions (see Table I). Eq. 2 shows that Vc depends on
several material parameters that can vary with temper-
ature, notably, α, Ms, Hk,eff and P [19 -- 22]. Specifically,
the magnetization and magnetic anisotropy both increase
with decreasing temperature. The increase in TMR at
low temperature also suggests that the spin polarization
increases with decreasing temperature. The increase in
the spin-polarisation therefore counteracts the increase
in magnetization and magnetic anisotropy and this, at
least qualitatively, can explain the saturation of Vc be-
low 150 K.
The characteristic switching time scale τ0 is given by
τ0 =
1 + α2
αγµ0Hk,eff
ln(2/θ0),
(3)
where α is the damping, γ is the gyromagnetic ratio, µ0
is the permittivity of free space, and θ0 is the average
magnetization initial angle. θ0 is related to the temper-
ature and the energy barrier to magnetization switching
by θ0 ≈ 1/(2
π∆), where ∆ is the ratio of the energy
barrier to reversal to the thermal energy, ∆ = Eb/(kBT ).
Hence τ0 is, again, related to material parameters that
depend on temperature:
√
ln(4(cid:112)πEb/(kBT )).
(4)
τ0 =
1 + α2
αγµ0Hk,eff
As Vc is independent of temperature below about 150 K
it seems reasonable to assume that the relevant junction
material parameters, like Eb and α, are also indepen-
dent of temperature below 150 K. Eq. 4 then predicts
that the characteristic switching time would increase by
about 20% at 4 K relative to its value at 150 K. This
predicted behavior reflects the decrease in magnetiza-
tion fluctuations as the temperature decreases. This be-
havior is clearly not seen experimentally and shows that
the macrospin model cannot explain the low temperature
data trends.
In addition to characterizing the temperature depen-
dence of the characteristic switching time and switching
threshold we have measured write error rates for nanosec-
ond current pulses. Figure 4 shows the WER for 4 ns du-
ration pulses at 4 and 295 K for the same 40 nm diameter
FIG. 3. 50% switching probability boundary of the same
40 nm diameter device for a) AP→P and c) P→AP switching
directions for 4 and 295 K. b) and d) At fixed overdrive V/Vc
the device switches faster at 4 K than at room temperature
for both switching polarities. The lines show the fit to the
macrospin model described in the main text.
performed pulse measurements at 4, 75, 150, and 295 K;
the 4 and the 295 K phase diagrams are shown in Fig. 2.
Figure 2 shows the switching phase diagrams for
AP→P (left panels) and P→AP transitions (right pan-
els) for a 40 nm diameter pMTJ at 4 K (Figs. 2(a) and
(b)) and 295 K (Fig. 2(c) and (d)) in zero applied field.
We observe high switching probability for pulse durations
less than 1 ns from room temperature to 4 K. For ∼5 ns
pulse durations, switching of the pMTJ at 4 K occurs for
higher pulse amplitudes compared to that at room tem-
perature as shown in Figs. 3(a) and (c), especially for the
AP→P transition (Fig. 3(a)).
In order to characterize the data trends we consider
a macrospin model, a simple model that provides ana-
lytic expressions for the switching times in the ballistic
limit and how they vary with material and device pa-
rameters [11, 18]. In this model the threshold voltage for
spin-transfer switching is given by:
(cid:17)
(cid:16)
1 +
τ0
τ
V = Vc
,
(1)
where τ is the pulse duration, τ0 is the characteristic time
for switching and Vc is the switching voltage in the long
pulse duration limit. The fits of the switching boundary
(i.e. the 50% switching probability versus pulse duration)
to Eq. 1 are displayed as blue lines in Fig. 2 and the
corresponding fit parameters are listed in Table I. The
fit parameters for 75 and 150 K are also shown in Table
I.
In Figs. 3(b) and (d) we compare the data at 4 and
295 K directly by plotting the normalized pulse voltages
versus pulse duration. We thus see that at fixed pulse
voltage overdrive, V /Vc, the switching time has decreased
at 4 K relative to that at 295 K. The same behavior was
123452003004005006007008009001000 P®AP4 K295 KAmplitude (mV)Pulse Duration (ns)123452003004005006007008009001000Amplitude (mV)Pulse Duration (ns) AP®P4 K295 Ka)c)b)d)123451.01.52.02.53.03.5V/VcPulse Duration (ns) AP®P4 K295 K123451.01.52.02.53.03.5 P®AP4 K295 KV/VcPulse Duration (ns) TABLE I. TMR and fit parameters from the pulsed switching measurements for various temperatures and the corresponding
optimal write energies.
4
T (K)
TMR (%)
4
75
150
295
200
193
182
117
AP→P
399
393
381
225
P→AP
421
416
403
305
AP→P
0.94
0.94
0.96
1.48
V c (mV)
τ 0 (ns)
E (fJ)
P→AP
1.03
1.05
1.10
1.38
AP→P
103
98
94
51
P→AP
286
283
287
195
associated with the increase in the threshold current for
switching. As expected, we observe a reduction of the
optimal write energies with increasing temperature; that
is, thermal energy reduces the device switching energy
(see Table I).
These results clearly show an advantageous scaling of
the switching energy with device diameter, the switch-
ing energy decreases as the device size decreases. These
write energies are also comparable to write energies of all
metal spin-valves with a Permalloy free layer [7] and or-
thogonal spin-transfer spin valve devices [6], which have
much lower resistances but significantly larger switching
currents. Further, two terminal pMTJ switching ener-
gies are lower than spin-Hall-based devices [5], because
of their lower switching currents.
The low energy consumption of pMTJ devices at 4 K
as well as their increased switching speed makes them
very interesting as a low-energy cryogenic data storage
solution. Foremost, the increased device magnetoresis-
tance at low temperatures makes it possible to further
reduce the resistance area product of the magnetic tunnel
junction (reducing the device resistance and therefore the
write energy) while maintaining a fast readout. It is also
possible to significantly reduce the switching energy. The
most straightforward way would be by further reducing
the perpendicular magnetic anisotropy (see Eq. 2), as the
FL in these studies are thermally stable at room temper-
ature (∆ > 26, determined by using a read disturb rate
method described in Ref. [24]), meaning their magnetic
anisotropy can be further reduced while still maintaining
stable magnetic states at 4 K. The FL magnetic moment
can also be reduced to decrease the switching current. In
addition, lower damping FL materials are also desirable
for this application. In summary, two-terminal pMTJs
are very promising for cryogenic applications and there
are straightforward paths to further device optimization.
Note added: We are aware of related research by Dr. Li
Ye at the Shanghai Institute of Microsystem and Infor-
mation Technology, Chinese Academy of Sciences. Their
work focuses on the low-temperature switching charac-
teristics of pMTJs at longer time scales and the relation
between the magnetic anisotropy, the magnetization and
the switching voltages at low temperatures. Their paper
will also be posted on the arXivs.
FIG. 4. Write error rates versus pulse amplitude for 4 ns
pulses at 4 and 295 K of the identical 40 nm diameter pMTJ
device.
pMTJ nanopillar. As already seen in the phase diagrams,
the switching voltages are larger at 4 K than at room
temperature. Limited only be the measurement time, we
found WER as low as 5×10-5 at 4 K (AP→P) and no
noticeable change in the slope of the WER curves versus
pulse amplitude between 4 and 295 K. This is an impor-
tant result that highlights the fact that WER characteris-
tics are not significantly dependent on temperature; that
temperature simply rescales the pulse amplitude needed
to achieve a desired WER performance.
Bases on these results we can determine the write ener-
gies and compare them to other types of cryogenic mag-
netic memory devices. The optimal (i.e.
lowest) write
energy is for pulse durations at the characteristics time
τ0 [23] and given by E = V 2τ0/R with V = 2Vc and
R the device resistance at the switching voltage [15, 16].
We find 103 fJ for AP→P and 286 fJ for P→AP at 4 K
for 40 nm diameter devices. For larger diameter devices
the optimal switching energy increases: for 50 nm diame-
ter devices, AP→P 167 and P→ AP 451 fJ and for 60 nm
diameter devices AP→P 226 and P→AP 610 fJ. The in-
creased write energies for the larger devices is mostly
4Fig. 4: Write Error Rates of 40 nm device-800-600-4002004006008001E-51E-40.0010.010.11Write Error RatePulse Amplitude (mV) 4 K 295 KP→APAP→P ACKNOWLEDGMENTS
This research is supported in part by Spin Memory Inc.
It is also based on work partially supported by the Office
of the Director of National Intelligence (ODNI), Intel-
ligence Advanced Research Projects Activity (IARPA),
via contract W911NF-14-C0089. The views and conclu-
sions contained herein are those of the authors and should
not be interpreted as necessarily representing the official
policies or endorsements, either expressed or implied, of
the ODNI, IARPA, or the U.S. Government. The U.S.
Government is authorized to reproduce and distribute
reprints for Governmental purposes notwithstanding any
copyright annotation thereon. This document does not
contain technology or technical data controlled under ei-
ther the U.S. International Traffic in Arms Regulations
or the U.S. Export Administration Regulations.
5
[1] S. Holmes, A. L. Ripple, and M. A. Manheimer, IEEE
Transactions on Applied Superconductivity 23, 1701610
(2013).
[2] J. C. Slonczewski, Journal of Magnetism and Magnetic
Materials 159, L1 (1996).
[3] L. Berger, Physical Review B 54, 9353 (1996).
[4] A. D. Kent and D. C. Worledge, Nature Nanotechnology
10, 187 (2015).
[5] M.-H. Nguyen, G. J. Ribeill, M. Gustafsson, S. Shi,
S. V. Aradhya, A. P. Wagner, L. M. Ranzani, L. Zhu,
R. Baghdadi, B. Butters, E. Toomey, M. Colan-
gelo, P. A. Truitt, A. Jafari-Salim, D. McAllister,
D. Yohannes, S. R. Cheng, R. Lazarus, O. Mukhanov,
K. K. Berggren, R. A. Buhrman, G. E. Rowlands, and
T. A. Ohki, arXiv preprint arXiv:1907.00942 (2019).
[6] G. E. Rowlands, C. A. Ryan, L. Ye, L. Rehm, D. Pinna,
A. D. Kent, and T. A. Ohki, Scientific Reports 9, 803
(2019).
[7] L. Rehm, V. Sluka, G. E. Rowlands, M.-H. Nguyen,
T. A. Ohki, and A. D. Kent, Applied Physics Letters
114, 212402 (2019).
[8] J.-H. Park, Y. Kim, W. C. Lim, Y.-J. Lee, J. H. Kim,
S. H. Park, J. H. Kim, W. Kim, K. W. Kim, J. H. Jeong,
K. S. Kim, H. Kim, Y. J. Lee, S. C. Oh, J. E. Lee,
S. O. Park, S. Watts, D. Apalkov, V. Nikitin, M. Krounbi,
S. Jeong, S. Choi, H. K. Kang, and C. Chung, Proc.
Symp. VLSI Technol. Dig. Tech. Papers, 57-58,
IEEE
(2012).
[9] G. Jan, L. Thomas, S. Le, Y.-J. Lee, H. Liu, J. Zhu,
R.-Y. Tong, K. Pi, Y.-J. Wang, D. Shen, R. He, J. Haq,
J. Teng, V. Lam, K. Huang, T. Zhong, T. Torng, and P.-
K. Wang, Proc. Symp. VLSI Technol. Dig. Tech. Papers,
1-2 IEEE (2014).
[10] S. Ikeda, K. Miura, H. Yamamoto, K. Mizunuma,
H. D. Gan, M. Endo, S. Kanai, J. Hayakawa, F. Mat-
sukura,
and H. Ohno, Nature Materials 9, 721-724
(2010).
[11] J. Z. Sun, Physical Review B 62, 570 (2000).
[12] H. Sato, M. Yamanouchi, S. Ikeda, S. Fukami, F. Mat-
and H. Ohno, Applied Physics Letters 101,
sukura,
022414 (2012).
[13] J.-H. Kim, J.-B. Lee, G.-G. An, S.-M. Yang, W.-
S. Chung, H.-S. Park, and J.-P. Hong, Scientific Reports
5, 16903 (2015).
[14] D. B. Gopman, D. Bedau, S. Mangin, C. H. Lambert,
E. E. Fullerton, J. A. Katine, and A. D. Kent, Applied
Physics Letters 100, 062404 (2012).
[15] Y. Lu, X. W. Li, G. Xiao, R. A. Altman, W. J. Gallagher,
A. Marley, K. Roche, and S. Parkin, Applied Physics
Letters 83, 6515 (1998).
[16] K. Cao, H. Li, W. Cai, J. Wei, L. Wang, Y. Hu, Q. Jiang,
H. Cui, C. Zhao, and W. Zhao, IEEE Transactions on
Magnetics 55, 3400304 (2018).
[17] M.-H. Nguyen, S. Shi, G. E. Rowlands, S. V. Aradhya,
C. L. Jermain, D. C. Ralph, and R. A. Buhrman, Applied
Physics Letters 112, 062404 (2018).
[18] R. H. Koch, J. A. Katine, and J. Z. Sun, Physical Review
Letters 92, 088302 (2004).
[19] H. Yu, R. Huber, T. Schwarze, F. Brandl, T. Rapp,
P. Berberich, G. Duerr, and D. Ohno, Applied Physics
Letters 100, 262412 (2012).
[20] J. M. Iwata-Harms, G. Jan, H. Lui, S. Serrano-Guisan,
and P.-
J. Zhu, L. Thomas, R.-Y. Tong, V. Sundar,
K. Wang, Scientific Reports 8, 14409 (2018).
[21] H. Sato, P. Chureemart, F. Matsukura, R. W. Chantrell,
H. Ohno, and R. F. L. Evans, Physical Review B 98,
214428 (2018).
[22] J. Beik Mohammadi, B. Kardasz, G. Wolf, Y. Chen,
M. Pinarbasi, and A. D. Kent, arXiv preprint
arXiv:1905.09329 (2019).
[23] H. Liu, D. Bedau, J. S. Sun, S. Mangin, E. E. Fullerton,
J. A. Katine, and A. D. Kent, Journal of Magnetism and
Magnetic Materials 358-359, 233-258 (2014).
[24] R. Heindl, W. H. Rippard, S. E. Russek, M. R. Pu-
fall, and A. B. Kos, Applied Physics Letters 109, 073910
(2011).
|
1909.00113 | 1 | 1909 | 2019-08-31T02:56:58 | Domain Wall Enabled Hysteresis-Free Steep Slope Switching in MoS$_2$ Transistors | [
"physics.app-ph",
"cond-mat.mes-hall",
"cond-mat.mtrl-sci"
] | The device concept of operating ferroelectric field effect transistors (FETs) in the negative capacitance (NC) regime offers a promising route for achieving energy-efficient logic applications that can outperform the conventional CMOS technology, while the viable mechanisms for stabilizing the NC mode remain a central topic of debate. In this work, we report hysteresis-free steep slope switching in few-layer and bilayer MoS$_2$ transistors back-gated by single layer polycrystalline PbZr$_{0.35}$Ti$_{0.65}$O$_3$ films. The devices exhibit current on/off ratios up to 8$\times$10$^6$ within an ultra-low gate voltage window of V$_g$ = $\pm$0.5 V and subthreshold swing as low as 9.7 mV/decade at room temperature, transcending the 60 mV/decade Boltzmann limit. Unlike previous studies, the quasi-static NC mode is realized in a ferroelectric without involving an additional dielectric layer. Theoretical modeling reveals the dominant role of the metastable polar states within ferroelectric domain walls in enabling the NC mode in the MoS$_2$ transistors. Our findings shed light into a new mechanism for NC operation, providing a simple yet effective material strategy for developing high speed, low-power 2D nanoelectronics. | physics.app-ph | physics | Domain Wall Enabled Hysteresis-Free Steep Slope Switching in MoS2
Transistors
Jingfeng Song,1 Yubo Qi,2 Zhiyong Xiao,1 Seung-Hyun Kim,3 Angus I. Kingon,3
Andrew M. Rappe,2 and Xia Hong1*
1 Department of Physics and Astronomy, University of Nebraska-Lincoln, Lincoln, NE 68588,
USA
2 Department of Chemistry, University of Pennsylvania, Philadelphia, PA 19104-6323, USA
3 School of Engineering, Brown University, Providence, RI 02912, USA
*email: [email protected]
Abstract
The device concept of operating ferroelectric field effect transistors (FETs) in the negative
capacitance (NC) regime offers a promising route for achieving energy-efficient logic
applications that can outperform the conventional CMOS technology, while the viable
mechanisms for stabilizing the NC mode remain a central topic of debate. In this work, we report
hysteresis-free steep slope switching in few-layer and bilayer MoS2 transistors back-gated by
single layer polycrystalline PbZr0.35Ti0.65O3 films. The devices exhibit current on/off ratios up to
8×106 within an ultra-low gate voltage window of 𝑉(cid:2917)(cid:3404)(cid:3399)0.5 V and subthreshold swing as low as
9.7 mV/decade at room temperature, transcending the 60 mV/decade Boltzmann limit. Unlike
previous studies, the quasi-static NC mode is realized in a ferroelectric without involving an
additional dielectric layer. Theoretical modeling reveals the dominant role of the metastable
polar states within ferroelectric domain walls in enabling the NC mode in the MoS2 transistors.
Our findings shed light into a new mechanism for NC operation, providing a simple yet effective
material strategy for developing high speed, low-power 2D nanoelectronics.
Introduction
dec), is determined by Boltzmann statistics:
While the ever-growing thermal power becomes a central challenge faced by information
technology in the post-Moore's law era1, ferroelectric-gated field effect transistors (FeFETs)
operating in the negative capacitance (NC) mode provides a promising route for developing
energy-efficient logic applications that can transcend the classic thermal limit for turning on the
devices2,3. For conventional transistors, the subthreshold swing (SS), defined as the gate voltage
(cid:4666)𝑉(cid:2917)(cid:4667) required to change the channel source-drain current 𝐼(cid:2914) by one order of magnitude (decade,
𝑆𝑆≡
which imposes a fundamental limit of 𝑆𝑆(cid:3406)60 mV/dec at 300 K2. Here ψ(cid:2929) is the surface
potential of the channel, 𝐶(cid:2913)(cid:2918) is the channel capacitance, and 𝐶(cid:2917) is the gate capacitance. It has
𝐶(cid:2917), which in turn reduces 𝑆𝑆 below the thermal limit (Eq. 1), known as steep slope switching2.
been proposed that by replacing the gate dielectric with a ferroelectric layer coupled with proper
capacitance matching, it is possible to stabilize the device in the regime with a negative effective
(cid:2986)(cid:4666)(cid:2922)(cid:2925)(cid:2917)(cid:3117)(cid:3116)(cid:3010)(cid:3162)(cid:4667)(cid:3404)(cid:3436)1(cid:3397)(cid:3004)(cid:3161)(cid:3166)(cid:3004)(cid:3165)(cid:3440)(cid:3038)(cid:3134)(cid:3021)(cid:3044) ln10 ,
(cid:2986)(cid:2984)(cid:3177)
(cid:2986)(cid:4666)(cid:2922)(cid:2925)(cid:2917)(cid:3117)(cid:3116)(cid:3010)(cid:3162)(cid:4667)(cid:3404)(cid:2986)(cid:3023)(cid:3165)(cid:2986)(cid:2984)(cid:3177)⋅
(cid:2986)(cid:3023)(cid:3165)
(1)
The key to accessing the intrinsic NC regime of ferroelectrics relies on the instability of the
spontaneous polarization2,4, which has been identified experimentally either in single layer
ferroelectric capacitors in transient measurements during polarization switching5, or in
ferroelectric/dielectric stacks6-15, exploiting the dielectric layer to stabilize the quasi-static NC
mode. Since polarization switching is a first-order physical process, a hysteresis loop in the 𝐼(cid:2914) vs.
𝑉(cid:2917) curve is inevitable, which means that the on and off switching must be operated at different
voltages. Such an operation voltage span lowers the operation speed, increases energy
consumption, and compromises the reliability of the device performance, and thus is not desired.
Alternative scenarios proposed to harness the NC effect include charge trapping16 and
polarization rotation effects17. While the underlying mechanism for the NC-FETs remains a
central topic of debate, the technological implementation of this device concept calls for
hysteresis-free operation3.
Since the initial proposal of the NC-FET, a wide variety of material systems have been
investigated theoretically or experimentally as channel materials for NC-FET. Compared with
conventional
transition metal
semiconductors6,7,
two-dimensional
layered
(2D)
the
2
8-14,18 offer an intrinsic advantage in terms of
dichalcogenides (TMDC) such as MoS2 and MoSe2
size scaling19,20. 2D MoS2 is a semiconductor with band gap of 1.2−1.8 eV, and has been widely
investigated for building high-performance logic applications19, where high current on/off
ratio21,22, high mobility23, and high breakdown field24 have been demonstrated using
20-24. Interfacing TMDC with ferroelectric
conventional dielectric gates, such as SiO2 and HfO2
oxides9-13,25-27 and polymers8,14,18,28,29 further introduces new functionalities into the 2D channel,
including nonvolatile memories, programmable junctions, and steep-slope transistors30.
In this work, we report hysteresis-free steep slope switching in few-layer and bilayer MoS2
transistors back-gated by single layer polycrystalline PbZr0.35Ti0.65O3 (PZT) films. These devices
exhibit current ratios up to 8×106 within an ultra-low gate voltage window of 𝑉(cid:2917)(cid:3404)(cid:3399)0.5 V and
SS as low as 9.7 mV/dec at room temperature. Unlike the widely pursued device structure with a
ferroelectric/dielectric stack gate, no dielectric layer is employed to stabilize the NC mode of the
polar layer. Instead, our theoretical modeling reveals that the steep slope switching originates
from the metastable polar state within the domain walls (DWs) in the polycrystalline PZT gate,
where a sudden boost of surface potential can be induced at an electric field well below the
ferroelectric coercive field. Compared with conventional NC mechanisms that involve
polarization switching, this mechanism is intrinsically low power and high speed. Our study thus
provides new insights into the viable mechanism for the NC operation, and points to a novel and
simple material scheme for achieving hysteresis-free steep slope transistors with reduced
fabrication complexity.
Results
Characterization of polycrystalline PZT thin films
We work with 300 nm thick polycrystalline PbZr0.35Ti0.65O3 films deposited on Pt/Ti/SiO2/Si
substrates (see Supplementary Information for deposition details). Figure 1a shows the x-ray
diffraction spectrum of the PZT film, which reveals predominant (001) and (111) growth with a
small fraction of (110) grains. We estimate the average crystallite size from the full-width-half-
maximum (L) of the Bragg peaks using the Scherrer Equation, 𝐵(cid:4666)2θ(cid:4667)(cid:3404) (cid:3012)(cid:2971)(cid:3013)(cid:2913)(cid:2925)(cid:2929)(cid:2968), where K = 1 is the
Scherrer constant, θ is the Bragg angle, and λ(cid:3404)1.5406 Å. The averaged grain sizes are 27.8 ±
0.8 nm and 27.7 ± 0.5 nm for the (001) and (111) oriented grains, respectively. Atomic force
3
microscopy (AFM) measurements show that these films possess smooth surface morphology
(Fig. 1b), with a typical root mean squared roughness of 1-2 nm.
We have characterized the distribution and orientation of the PZT polarization using
piezoresponse force microscopy (PFM). Figures 1c-f show the PFM measurements conducted in
both vertical (V-PFM) and lateral (L-PFM) modes on the same region of a PZT film. We
observed domains with up to 180° phase contrast (Figs. 1c, e) and large amplitude variations
(Figs. 1d, f) in both vertical and lateral PFM, indicating a wide distribution of polarization
orientation in the as-grown state of the film. The domains range in size from 20 nm to more than
100 nm, pointing to the presence of a high density of DWs. There is no clear correlation between
the domain distribution and the surface topography, and the average size of the polar domains is
much larger than that of the polycrystalline grains. These results suggest that the domain
formation is not confined by the grain boundaries, consistent with previous phase-field
simulation results31.
Figure 1g shows the polarization P versus bias voltage (Vbias) measured in a capacitance
structure, which exhibits robust switching hysteresis with remanent polarization of about 0.3
C/m2 and coercive voltages of +1.3 V and -1.1 V. The hysteresis becomes negligibly small at the
small bias voltage range of ±0.5 V (Fig. 1h insert). Within the hysteresis-free regime, we
extracted a dielectric constant of 630−650, which is one to two orders of magnitude higher than
those of conventional dielectrics such as SiO2 and HfO2. The dielectric constant shows little
variation in this Vbias range, and can yield highly efficient doping in the 2D channel30.
Characterization of high-performance MoS2 FET gated by PZT
We mechanically exfoliate few-layer and bi-layer MoS2 flakes on PZT, and fabricate them
into PZT back-gated transistor devices (Fig. 2a, Methods). Figure 2b shows the AFM topography
image of a five-layer MoS2 device (Methods), which conforms well with the PZT surface
morphology. We first investigate the transfer characteristic of the device (Id vs. Vg) at 300 K and
source-drain voltage Vd = 0.1 V within the hysteresis-free regime (Fig. 2c). Within an ultra-low
voltage range of 𝑉(cid:2917)(cid:3404)(cid:3399)0.5 V, the device exhibits a high current switching ratio (Ion/Ioff) of about
correction (Supplementary Information). From the quasi-linear region of the 𝐼(cid:2914)(cid:3398)𝑉(cid:2917) curves, we
8×106 in the forward Vg-sweep, which clearly reflects the high doping efficiency of the PZT gate.
Figure 2d shows the transfer curves of the device from 290 K to 320 K after pyroelectric
4
Hysteresis-free steep slope switching in few-layer and bilayer MoS2 transistors
Figure 2f shows the point-by-point SS of the MoS2 FET calculated from the inverse slope
extracted the field effect mobility μ(cid:2890)(cid:2889)(cid:3404) (cid:2869)(cid:3004)(cid:3148)(cid:3158)(cid:3152)(cid:3013)(cid:3024)(cid:3031)(cid:3008)(cid:3031)(cid:3023)(cid:3165), where 𝐶(cid:2900)(cid:2910)(cid:2904) is the areal capacitance for 300
nm PZT, 𝐺(cid:3404)𝐼(cid:2914)/𝑉(cid:2914) is the channel conductance, and L (W) is the channel length (width). At 300
K, μ(cid:2890)(cid:2889)(cid:3404)7.8 cm(cid:2870)/V∙s, comparable with previously reported values for MoS2 FETs interfaced
with ferroelectrics26,28,29. In this temperature range, μ(cid:2890)(cid:2889) decreases with increasing temperature,
following a power law T-dependence of ~𝑇(cid:2879)(cid:2869).(cid:2877) (Fig. 2e), which can be attributed to phonon
scattering23. The exponent α(cid:3404)1.9 is between the theoretically predicted values of for single
layer MoS2 (α(cid:3404)1.52)32 and bulk MoS2 crystals (α(cid:3404)2.6)33.
(𝜕log(cid:4666)𝐼(cid:2914)(cid:4667)/𝜕𝑉(cid:2917)) of the transfer curves in Fig. 2c. For the forward 𝑉(cid:2917)-sweep, we have achieved a
minimum subthreshold swing of SSmin ≈ 37 mV/dec as the device starts to turn on at 𝐼(cid:2914)(cid:3406)1.6(cid:3400)
10(cid:2879)(cid:2869)(cid:2871) A. The SS remains below the 60 mV/dec thermal limit over four decades of channel
current (10(cid:2879)(cid:2869)(cid:2871)(cid:3398)10(cid:2879)(cid:2877) A). In the reverse scan (Fig. 2g), SS is close to 60 mV/dec in the channel
current range of 10(cid:2879)(cid:2869)(cid:2869)(cid:3398)10(cid:2879)(cid:2876) A, higher than the SS value observed in the forward scan. This
this point, we also show the leakage current (𝐼(cid:2922)(cid:2915)(cid:2911)(cid:2921) ) measured between the gate and drain
electrodes, as shown in the lower panel of Fig. 2c. It is clear that the current floor of 𝐼(cid:2914) is in
higher 𝐼(cid:2922)(cid:2915)(cid:2911)(cid:2921) is observed in the reverse scan within the negative 𝑉(cid:2917) regime. Once 𝐼(cid:2914) exceeds the
two 𝑉(cid:2917) scan directions, agreeing well with the dielectric measurement of the PZT gate (Fig. 1h
shown in Fig. 3b, a current switching ratio of 5(cid:3400)10(cid:2874) is achieved in the device within a small
voltage window ΔVg of 0.76 V (-0.26 V to 0.5 V) in the forward 𝑉(cid:2917)-sweep. Compared with the
characteristic at low channel current, with an SSmin of 9.7 mV/dec at 𝐼(cid:2914) (cid:3406)10(cid:2879)(cid:2869)(cid:2870) A (Fig. 3c). The
discrepancy is likely due to the artifact associated with increased leakage current through the
PZT gate rather than an intrinsic property of the device. As our measurement utilizes a two-point
device geometry (Methods), Id is inevitably influenced by this current contribution. To illustrate
few-layer device, the bilayer channel exhibits a much steeper slope in the initial turn-on
qualitative agreement with the leakage current level in both forward and reverse scans, and a
leakage contribution, the transfer curve of the device is essentially free of hysteresis between the
inset).
Similar switching characteristics have been observed in a bilayer MoS2 device (Fig. 3a). As
5
with the dielectric layer thickness, the bilayer MoS2 possesses a larger capacitance in the
weight for the NC term in Eq. 1, thus significantly reducing the initial SS value. On the other
shift the Fermi energy to the conduction band edge. Once the device reaches the on state, or
sharp contrast to the moderate Vg-dependence of SS observed in the few-layer device at this
current range. The tradeoff between the steepness of the initial turn-on behavior and the current
range of low SS value can be attributed to the competing effects of the channel capacitance
SS then increases quickly with 𝐼(cid:2914), reaching about 60 mV/dec at 𝐼(cid:2914)(cid:3406)2(cid:3400)10(cid:2879)(cid:2869)(cid:2869) A. This is in
𝐶(cid:2897)(cid:2925)(cid:2903)(cid:3118) and doping efficiency. As shown in Eq. 1, 𝐶(cid:2913)(cid:2918) not only plays a critical role in stabilizing
the NC mode, but also tailors the fractional weight of the second term. As 𝐶(cid:2897)(cid:2925)(cid:2903)(cid:3118) scales inversely
depletion state compared with the few-layer device. For a given 𝐶(cid:2917), it yields a larger fractional
hand, a thinner channel also corresponds to a lower 2D density of states, requiring a lower 𝑉(cid:2917) to
𝐶(cid:2897)(cid:2925)(cid:2903)(cid:3118) exceeds 𝐶(cid:2900)(cid:2910)(cid:2904), the quasi-static NC mode is no longer energetically favorable, and SS of the
resolved in 𝐼(cid:3039)(cid:2915)(cid:2911)(cid:2921) .
the forward sweep in the Id range of 10(cid:2879)(cid:2869)(cid:2871)(cid:3398)10(cid:2879)(cid:2869)(cid:2869) A. For all devices, SSavg is consistently
below the classical thermal limit of 𝑘(cid:2886)𝑇ln10/𝑞 (Eq. 1) over the entire temperature range
device returns to the classical operation regime. Similar to the few-layer device, the reverse scan
exhibits much higher SS that exceeds 60 mV/dec (Fig. 3d). Due to the high noise level in the
measured leakage current, no clear difference between the forward and reverse scans can be
The hysteresis-free steep slope switching is a robust effect observed in five MoS2 FET
devices, as summarized in Fig. 4a. For consistency, we plot the average SS values extracted from
investigated. Figure 4b plots the Ion/Ioff vs. ΔVg result obtained on the few-layer device (Fig. 2c)
in comparison with those previously reported for MoS2 NC-FETs6-14,18 and classical
FeFETs25,26,28, which highlights the superb performance combination of ultra-low supply voltage
and high current on/off ratio in our devices. Within the noise level of the measurements (1-2
mV), all devices exhibit essentially zero-hysteresis switching at 𝐼(cid:2914)(cid:3408)𝐼(cid:2922)(cid:2915)(cid:2911)(cid:2921). Despite the simple
material scheme, the minimum SS of 9.7 mV/dec observed in our bilayer MoS2 device is
comparable with the best result reported in hysteresis-free NC FETs (5.6 mV/dec in Ref. [13])
using ferroelectric/dielectric stack gates (Fig. 4c).
Theoretical modeling of the DW enabled NC effect
6
channel exceed the applied voltage, or
Unlike previous experimental studies of NC-FETs based on ferroelectric/dielectric stack
gates6-13, the sub-60 mV/dec SS acquired in our devices in the hysteresis-free region of PZT
suggests the existence of a quasi-static NC mode without the presence of an additional dielectric
layer and hence the associated capacitance matching. The switching occurs at a gate bias
significantly smaller than the coercive field (Ec) of the ferroelectric gate, further ruling out the
contribution of polarization reversal to the steep slope switching. As shown in Eq. 1, the key to
accessing the NC mode is to have the gate-induced surface potential change in the semiconductor
Gibbs free energy to be negative, which can be realized in ferroelectrics below TC near the
centrosymmetric transition state during polarization reversal, as shown in Fig. 5a. The initial
proposal of the NC FET device concept thus builds on this polarization switching region2. Close
to Ec, a relatively small change in Vg can cause polarization reversal by going through the
(cid:2986)(cid:3023)(cid:3165)(cid:2986)(cid:2984)(cid:3177)(cid:3407)1. This requires the second-order derivative of the
negative capacitance state, resulting in a sudden boost in polarization, surface potential ψ(cid:2929) and
surface charge density 𝑄 in the semiconducting channel, and hence 𝐼(cid:2914)34. Figure 5b illustrates
how a polarization reversal in PZT (𝑃(cid:3404)0.5 C/m2) can cause a jump in the surface potential of
∆ψ(cid:2929) = 1.54 V, assuming 𝑄(cid:3406)𝑃 (Supplementary Information). Besides polarization reversal,
theoretically predicted that the sudden change of ψ(cid:2929) in the semiconductor channel can also be
which is a first-order process that inevitably leads to switching hysteresis, it has been
achieved through ferroelectric polarization rotation from the in-plane to out-of-plane orientation,
which can lead to hysteresis-free operation with higher speed and lower energy consumption17.
In our experiments, the SS falls below the Boltzmann limit within the hysteresis-free region
at an applied field well below Ec, suggesting that it is not driven by polarization reversal. To
understand the underlying mechanism, we carried out three-dimensional (3D) force field
simulations based on the Landau-Ginzburg-Devonshire (LGD) theory35, in which the
thermodynamic potential (Gibbs free energy) 𝐹 can be expressed as:
𝐹(cid:3404) (cid:1516) (cid:3435)𝑓(cid:2912)(cid:2931)(cid:2922)(cid:2921)(cid:3397)𝑓(cid:2915)(cid:2922)(cid:2911)(cid:2929)(cid:3397)𝑓(cid:2917)(cid:2928)(cid:2911)(cid:2914)(cid:3397)𝑓(cid:2915)(cid:2922)(cid:2915)(cid:2913)(cid:3439)𝑑𝑉
(cid:3023)
where 𝑓(cid:2912)(cid:2931)(cid:2922)(cid:2921), 𝑓(cid:2915)(cid:2922)(cid:2911)(cid:2929), 𝑓(cid:2917)(cid:2928)(cid:2911)(cid:2914), and 𝑓(cid:2915)(cid:2922)(cid:2915)(cid:2913) are the energy densities associated with the thermodynamic
,
(2)
potential of a PZT single crystal, elastic energy, dipole gradient, and electrostatic energy,
respectively (see Supplementary Information for modeling details). Fig. 5c shows the simulation
result for an equilibrated multiple-domain structure in PZT. There are equal volumes of up and
7
chiral dipole structures at the domain walls, which is consistent with previous experimental
depolarization field36. Near the surfaces, the dipoles mostly lie in the plane to satisfy the
down polarization domains, which minimizes the electrostatic energy (𝑓(cid:2915)(cid:2922)(cid:2915)(cid:2913)) cost induced by
continuity of electric displacement, minimizing 𝑓(cid:2917)(cid:2928)(cid:2911)(cid:2914). This leads to vortex or anti-vortex like
observation and theoretical simulations15,37,38. We then impose a gate voltage 𝑉(cid:2917) on this
equilibrated model in a sweeping sequence of 0 → 0.5 → 0 → -0.5 → 0 V, and calculated the
evolution of local dipoles (Fig. 5d) and associated channel current with respect to 𝑉(cid:2917). Fig. 5d
shows the simulated profile for the polarization change (∆𝑃) upon sweeping 𝑉(cid:2917) across -0.25 V,
where an abrupt increase in the polarization occurs only at the DWs. This local boost of ∆𝑃 can
be well correlated with a sudden jump in 𝐼(cid:2914), as clearly shown in the simulated transfer curves
(Fig. 5e). The corresponding SS reaches the minimum value of about 27 mV/dec at 𝐼(cid:2914)(cid:3406)3(cid:3400)
10(cid:2879)(cid:2869)(cid:2870) A, and remains below the 60 mV/dec limit for more than three decades of 𝐼(cid:2914). Without
considering the leakage contribution, the forward and backward gate sweeps overlap with each
other. The simulated transfer characteristics thus well capture the main features of the
experimental observation.
With the simulation results, we attribute the experimentally observed hysteresis-free sub-60
mV/dec switching to the negative capacitance states at the DWs, which are abundant in
polycrystalline PZT (Figs. 1c-f)31. In this scenario, the DW regime possesses significantly
suppressed local polarization due to their neighbors with antiparallel dipole orientations. These
metastable polar states are delicate and have the tendency to collapse into a polar state upon
external perturbation, which is manifested as the enhanced dielectric susceptibility observed in
the DWs15,38. An external electric field can thus induce a much larger increase in the dipoles at
the DW compared with the dipoles inside the uniformly polarized domains. The simulated jump
in the polarization is on the order of 0.004 C/m2, on the same order of magnitude as that
measured in our PZT films at Vbias below the coercive voltage (Fig. 1h). Even though this
polarization value is much smaller than the remanent polarization of bulk PZT, it is comparable
with that of (Hf,Zr)O2
indicated in Fig. 5b. Besides, the electrostatic energy required for modulating the DW regions
between the non-polar and NC states is much smaller than that for reorienting the polarization
13 and large enough to induce a significant boost in ψ(cid:2929) (0.76 V), as
8
approximately zero hysteresis window in the 𝐼(cid:2914) vs. 𝑉(cid:2917) curve.
within
the uniformly polarized domains (Supplementary Information),
leading
to an
The responsiveness of DWs also naturally explains the discrepancy in the transfer
characteristics between the forward and reverse Vg-scans at the initial turn-on region of MoS2
device (Figs. 2c and 3b). Previous conductive-probe AFM studies have revealed thermally
activated, diode-like conduction at DWs in PZT39. In the polycrystalline PZT film, a high density
of DWs is present at zero bias, which accounts for a higher leakage current in the reverse scan to
the negative 𝑉(cid:2917) regime. Once the metastable states are polarized by the gate bias, the volume
fraction of the DWs in PZT is reduced, leading to lower leakage current. We then incorporated
the leakage contribution in the model (Supplementary Information), and this can successfully
reproduce the softening of the turn-on behavior of the transfer curve (Fig. 5e).
Discussion
It is worth noting that utilizing the DW enabled NC state to construct the steep slope FETs
has distinct advantages in terms of device performance compared with the extensively studied
mechanisms based on polarization switching. It builds upon the small perturbation to a
metastable polarization-suppressed state within the DWs, where a sudden boost in polarization
can be induced by an electric field well below Ec. It is thus intrinsically hysteresis-free and low
energy. As the process does not involve dipole reorientation, it also promises high operation
speed. The fact that it does not require an additional dielectric layer further reduces the
fabrication complexity. This mechanism is expected to be general applicable to DW-rich
ferroelectric systems, such as polycrystalline thin films and films deposited along a crystalline
orientation off the major polar axis. In fact, sub-60 mV/dec SS has previously been observed in
MoS2 FETs with a single layer ferroelectric polymer gate14,18 (Fig. 4c), which naturally host
abundant DWs. While previous study attributes the NC effect to the possible existence of an
interfacial dead layer18, it is likely driven instead by the DWs. On the other hand, engineering
low leakage current in the DW-rich ferroelectrics is critical for practical implementation of DW-
based NC-FETs.
In terms of the MoS2 channel, even though we have achieved similar current on/off ratios in
the few-layer and bilayer devices, they exhibit distinct turn-on behaviors. The few-layer MoS2
shows steep-slope switching over four decades of channel current, with only moderate Vg-
9
dependence of SS. The bilayer device, in sharp contrast, possesses a much steeper initial turn on
with an SSmin that is only 26% of the value for the few-layer device, while the sub-60 mV/dec SS
only persists for two decades of current. The key parameter that determines the transfer
characteristics is the layer thickness, as the stabilization of the NC mode depends on the relative
length scales of film thickness and depletion length of MoS2. Regarding the optimal thickness for
MoS2, it is contingent upon the channel mobility and required operation current level for the
specific applications.
In summary, we demonstrated hysteresis-free steep slope MoS2 FETs utilizing a single
ferroelectric layer without additional dielectric matching. These devices exhibit current on/off
ratios up to 8×106 within an ultra-low gate voltage window of 𝑉(cid:2917)(cid:3404)(cid:3399)0.5 V and a minimum SS of
9.7 mV/dec at room temperature, well below the Boltzmann thermal limit for conventional FETs.
Theoretical modeling points to a new operation mechanism for NC-FETs, where the NC mode is
enabled by the metastable polar states within ferroelectric domain walls. With solution-processed,
easy-to-fabricate polycrystalline ferroelectric thin films, single layer gate geometry, sub-60
mV/dec SS and ultra-low working voltages, our work points to a novel cost effective material
strategy for developing high performance low power 2D nanoelectronics.
Methods
Characterization of PZT thin films. The structural properties of the polycrystalline PZT films
were characterized using a Rigaku SmartLab Diffractometer with Cu Kα radiation (λ = 1.5406 Å).
The surface and PFM characterizations of the PZT films were carried out on a Bruker Multimode
8 AFM. The PFM measurements were conducted in V-PFM and L-PFM modes with Bruker
SCM-PIT tip at around 300 kHz and 660 kHz, respectively. For the dielectric/ferroelectric
characterizations, we deposited Pt or Au top electrodes on PZT. The dielectric constant of PZT
was extracted from the C-V measurements conducted with an HP 4291A RF Impedance
Analyzer between ±0.5 V at 1 kHz. The low voltage P-V loops were measured with triangular
waves using an aixACCT TF analyzer 2000 between ±0.5 V at 1 kHz. The high voltage P-V
hysteresis loops were measured at a range of ±5 V with Precision Premier II Ferroelectric Tester
(Radiant Technologies, USA) at 1 kHz.
10
Fabrication and characterization of MoS2 devices. We mechanically exfoliated MoS2 flakes
on elastomeric films (Gel-Film® WF×4 1.5mil from Gel-Pak) from bulk single crystals. Few-
layer flakes were identified using optical microscopy and Raman spectroscopy and transferred
onto PZT. For the device shown in Fig. 2b, the frequency difference ∆𝜔 between the Raman 𝐸(cid:2870)(cid:2917)(cid:2869)
and 𝐴(cid:2917)(cid:2869) modes is 24.4 cm-1, corresponding to about five-layer thickness. For the bilayer device
shown in Fig. 3a, ∆ω is 22.0 cm-1. We then fabricated the MoS2 samples into two-point devices
using e-beam lithography followed by evaporation of 5 nm Ti/50 nm Au electrodes. The variable
temperature electrical characterizations of the MoS2 FETs were performed on either the
Quantum Design PPMS or the Lakeshore TTP4 probe station. For measurements taken on the
PPMS, Id was measured between the source and drain contacts using Keithley 6430 Sub-
Femtoamp Remote SourceMeter, Vg (Ileak) was applied (measured) between the gate and drain
contacts via a Keithley 2400 SourceMeter. For measurements carried on the probe station, the
transfer curves were taken using Keysight 1500A Semiconductor Analyzer, where Id was
measured via the high precision port and Vg (Ileak) was applied (measured) via the medium
precision port. The transfer curves were taken at Vd = 0.1-0.2 V with Vg sweep at a step size of
10 mV.
Data availability. The data that support the plots within this paper and other findings of this
study are available from the corresponding author upon reasonable request.
11
References
1 Waldrop, M. M. The semiconuctor industry will soon abandon its pursuit of Moore's law.
Now things could get a lot more interesting. Nature 530, 144-147 (2016).
2 Salahuddin, S. & Datta, S. Use of Negative Capacitance to Provide Voltage Amplification
for Low Power Nanoscale Devices. Nano Letters 8, 405-410 (2008).
3 Alam, M. A., Si, M. W. & Ye, P. D. D. A critical review of recent progress on negative
capacitance field-effect transistors. Applied Physics Letters 114, 090401 (2019).
4 Chang, S. C., Avci, U. E., Nikonov, D. E., Manipatruni, S. & Young, I. A. Physical Origin
of Transient Negative Capacitance in a Ferroelectric Capacitor. Physical Review Applied 9,
014010 (2018).
5 Khan, A. I. et al. Negative capacitance in a ferroelectric capacitor. Nature Materials 14,
182-186 (2015).
6 Dasgupta, S. et al. Sub-kT/q Switching in Strong Inversion in PbZr0.52Ti0.48O3 Gated
Negative Capacitance FETs. IEEE Journal on Exploratory Solid-State Computational
Devices and Circuits 1, 43-48 (2015).
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).
7
8 McGuire, F. A., Cheng, Z. H., Price, K. & Franklin, A. D. Sub-60 mV/decade switching in
2D negative capacitance field-effect transistors with integrated ferroelectric polymer.
Applied Physics Letters 109, 093101 (2016).
9 McGuire, F. A. et al. Sustained Sub-60 mV/decade Switching via the Negative Capacitance
Effect in MoS2 Transistors. Nano Letters 17, 4801-4806 (2017).
10 Nourbakhsh, A., Zubair, A., Joglekar, S., Dresselhaus, M. & Palacios, T. Subthreshold
swing improvement in MoS2 transistors by the negative-capacitance effect in a ferroelectric
Al-doped-HfO2/HfO2 gate dielectric stack. Nanoscale 9, 6122-6127 (2017).
11 Yu, Z. et al. in 2017 IEEE International Electron Devices Meeting (IEDM), pp. 23.6.1-
23.6.4.
12 Si, M. et al. in 2017 IEEE International Electron Devices Meeting (IEDM), pp. 23.5.1-
23.5.4.
13 Si, M. W. et al. Steep-slope hysteresis-free negative capacitance MoS2 transistors. Nature
Nanotechnology 13, 24-28 (2018).
12
14 Liu, X. Q. et al. MoS2 Negative-Capacitance Field-Effect Transistors with Subthreshold
Swing below the Physics Limit. Advanced Materials 30, 1800932 (2018).
15 Yadav, A. K. et al. Spatially resolved steady-state negative capacitance. Nature 565, 468-
471 (2019).
16 Daus, A. et al. Charge Trapping Mechanism Leading to Sub-60-mV/decade-Swing FETs.
IEEE Transactions on Electron Devices 64, 2789-2796 (2017).
17 Qi, Y. B. & Rappe, A. M. Designing Ferroelectric Field-Effect Transistors Based on the
Polarization-Rotation Effect for Low Operating Voltage and Fast Switching. Physical
Review Applied 4, 044014 (2015).
18 Wang, X. et al. Two-dimensional negative capacitance transistor with polyvinylidene
fluoride-based ferroelectric polymer gating. NPJ 2D Materials and Applications 1, 38
(2017).
19 Wang, Q. H., Kalantar-Zadeh, K., Kis, A., Coleman, J. N. & Strano, M. S. Electronics and
dichalcogenides. Nature
transition metal
of
optoelectronics
Nanotechnology 7, 699-712 (2012).
two-dimensional
20 Liu, H., Neal, A. T. & Ye, P. D. Channel Length Scaling of MoS2 MOSFETs. ACS Nano 6,
8563-8569 (2012).
21 Radisavljevic, B., Radenovic, A., Brivio, J., Giacometti, V. & Kis, A. Single-layer MoS2
transistors. Nature Nanotechnology 6, 147 (2011).
22 Yoon, Y., Ganapathi, K. & Salahuddin, S. How Good Can Monolayer MoS2 Transistors Be?
Nano Letters 11, 3768-3773 (2011).
23 Radisavljevic, B. & Kis, A. Mobility engineering and a metal-insulator transition in
monolayer MoS2. Nature Materials 12, 815-820 (2013).
24 Lembke, D. & Kis, A. Breakdown of High-Performance Monolayer MoS2 Transistors. ACS
Nano 6, 10070-10075 (2012).
25 Lu, Z. Y. et al. Nonvolatile MoS2 field effect transistors directly gated by single crystalline
epitaxial ferroelectric. Applied Physics Letters 111, 023104 (2017).
26 Zhou, C. J. et al. Low voltage and high ON/OFF ratio field-effect transistors based on CVD
MoS2 and ultra high-k gate dielectric PZT. Nanoscale 7, 8695-8700 (2015).
27 Lipatov, A., Sharma, P., Gruverman, A. & Sinitskii, A. Optoelectrical Molybdenum
Disulfide (MoS2)-Ferroelectric Memories. ACS Nano 9, 8089-8098 (2015).
13
28 Lee, H. S. et al. MoS2 Nanosheets for Top-Gate Nonvolatile Memory Transistor Channel.
Small 8, 3111-3115 (2012).
29 Xiao, Z., Song, J., Ferry, D. K., Ducharme, S. & Hong, X. Ferroelectric-Domain-Patterning-
Controlled Schottky Junction State in Monolayer MoS2. Physical Review Letters 118,
236801 (2017).
30 Hong, X. Emerging ferroelectric transistors with nanoscale channel materials: the
possibilities, the limitations. Journal of Physics: Condensed Matter 28, 103003 (2016).
31 Choudhury, S., Li, Y. L., Krill, C. E. & Chen, L. Q. Phase-field simulation of polarization
switching and domain evolution in ferroelectric polycrystals. Acta Materialia 53, 5313-5321
(2005).
32 Kaasbjerg, K., Thygesen, K. S. & Jacobsen, K. W. Phonon-limited mobility in n-type
single-layer MoS2 from first principles. Physical Review B 85, 115317 (2012).
33 Fivaz, R. & Mooser, E. Mobility of Charge Carriers in Semiconducting Layer Structures.
Physical Review 163, 743 (1967).
34 Taur, Y. & Ning, T. H. Fundamentals of modern VLSI devices. 2nd edn, (Cambridge
University Press, 2009).
35 Koukhar, V. G., Pertsev, N. A. & Waser, R. Thermodynamic theory of epitaxial
ferroelectric thin films with dense domain structures. Physical Review B 64, 214103 (2001).
36 Batra, I. P., Wurfel, P. & Silverman, B. D. Phase Transition, Stability, and Depolarization
Field in Ferroelectric Thin Films. Physical Review B 8, 3257-3265 (1973).
37 Jia, C. L., Urban, K. W., Alexe, M., Hesse, D. & Vrejoiu, I. Direct Observation of
Continuous Electric Dipole Rotation in Flux-Closure Domains in Ferroelectric Pb(Zr,Ti)O3.
Science 331, 1420-1423 (2011).
38 Zubko, P. et al. Negative capacitance in multidomain ferroelectric superlattices. Nature 534,
524 (2016).
39 Guyonnet, J., Gaponenko, I., Gariglio, S. & Paruch, P. Conduction at Domain Walls in
Insulating Pb(Zr0.2Ti0.8)O3 Thin Films. Advanced Materials 23, 5377 (2011).
14
Acknowledgments
We would like to thank Patty Niemoth, Dawei Li, and Pratyush Buragohain for technical
assistance, Yongfeng Lu and Alexei Gruverman for providing equipment access, and Zoran
Krivokapic for inspiring discussions. This work was primarily supported by the U.S. Department
of Energy (DOE), Office of Science, Basic Energy Sciences (BES), under Award No. DE-
SC0016153. Additional support is provided by the National Science Foundation (NSF) Grant No.
OIA-1538893 and Semiconductor Research Corporation (SRC) under GRC Task Number
2831.001. A.M.R. and Y.Q. acknowledge the support from the U.S. Department of Energy
(DOE), Office of Science, Basic Energy Sciences (BES), under Award No. DE-FG02-
07ER46431 and computational support from the National Energy Research Scientific Computing
Center (NERSC) of the DOE. The research was performed in part in the Nebraska Nanoscale
Facility: National Nanotechnology Coordinated Infrastructure and the Nebraska Center for
Materials and Nanoscience, which are supported by NSF under Award ECCS: 1542182, and the
Nebraska Research Initiative.
Author contributions
X.H. conceived and supervised the project. S.-H.K. and A.I.K. prepared the PZT films and
performed the ferroelectric studies. J.S. performed the structural characterization and PFM
studies of the PZT films. J.S. and Z.X. fabricated the MoS2 FETs and carried out the electrical
characterizations. Y.Q. and A.M.R. performed the modeling of the MoS2 FETs and DWs. J.S.,
Y.Q. and X.H. wrote the manuscript. All authors discussed the results and contributed to the
manuscript preparation.
Competing interests
The authors declare no competing interests.
Additional information
Correspondence and requests for materials should be addressed to Xia Hong.
15
Figures
Fig. 1 Characterization of polycrystalline PZT films. a, X-ray -2 scan taken on a 300 nm
PZT film. b, AFM topography, and c-f PFM images of the same area on a PZT film. c, V-PFM
phase and d, amplitude images. e, L-PFM phase and f, amplitude images. The lower panels show
the signal profiles along the dashed lines. g, P vs. Vbias hysteresis taken on a PZT film. h,
Dielectric constant of the film vs. Vbias, with Vbias well below the coercive voltage. Inset: P vs.
Vbias taken at this voltage range.
16
Fig. 2 Characterization of a few-layer MoS2 FET. a, Device schematic. b, AFM topography
of a few-layer MoS2 device with the height profile along the dashed line (lower panel). The
dotted lines outline the MoS2 flake. c, Transfer characteristics (upper panel) and gate leakage
current (lower panel) of the MoS2 FET at 300 K in both forward and reverse Vg-sweeps. d,
Transfer characteristics of the device from 290 K to 320 K, and e, the corresponding μFET vs. T
with a fit to 𝑇(cid:2879)(cid:2869).(cid:2877). f-g, Point-to-point SS vs. Id extracted from (c) for (f) the forward and (g)
reverse Vg-sweeps. The dashed lines depict the thermal limit for SS at 300 K.
17
Fig. 3 Characterization of a bilayer MoS2 FET. a, AFM topography of a bilayer MoS2 device
with the height profile averaged over the entire channel (lower panel). b, Transfer characteristics
(upper panel) and gate leakage current (lower panel) of the MoS2 FET at 300 K in both forward
and reverse Vg-sweeps. c-d, Point-to-point SS vs. Id extracted from (c) the forward and (d)
reverse Vg-sweeps. The dashed lines depict the thermal limit for SS at 300 K.
18
Fig. 4 Performance of the MoS2 NC-FETs. a, SSavg within the 𝐼(cid:3031) range of 10(cid:2879)(cid:2869)(cid:2871)(cid:3398)10(cid:2879)(cid:2869)(cid:2869)A vs.
T taken on three few-layer (denoted as FL D1, D2, and D3) and two bilayer (denoted as 2L D1
and D2) MoS2 FETs, including the data shown in Fig. 2 (FL D1) and Fig. 3 (2L D2). The red
dashed line depicts the theoretical Boltzmann limit of SS. b, Current on/off ratio vs. required ΔVg
and c, SSmin vs. average hysteresis voltage window taken from the current work (solid symbols)
and those from literature (open symbols)6-14,18,25,26,28. The error bar is given by the noise level of
the measurement.
19
Fig. 5 3D force field simulation results. a, Schematic double-well Gibbs free energy profile of
a ferroelectric material showing the negative capacitance region (circled). b, 2D charge density Q
vs. ψ(cid:2929) curve for MoS2. For a ferroelectric gate, the continuity of electric displacement yields Q ≈
P. The red arrow indicates the change of ψ(cid:3046) during a polarization flipping (from 𝑄(cid:3404)
(cid:3398)0.5 C/m(cid:2870),ψ(cid:3046)(cid:3404)(cid:3398)0.53 V to 𝑄(cid:3404)(cid:3397)0.5 C/m(cid:2870),ψ(cid:3046)(cid:3404)1.01 V ). The blue arrow indicates the
change of ψ(cid:2929) during a polarization increase (from 𝑄(cid:3404)0 C/m(cid:2870),ψ(cid:2929)(cid:3404)0 V to 𝑄(cid:3404)(cid:3397)0.0043 C/
m(cid:2870),ψ(cid:2929)(cid:3404)0.76 V). c, Simulated multiple-domain structure in PZT with two inequivalent DWs.
The left DW hosts a polar vortex, and the right one hosts an anti-vortex. d, Simulated change of
polarization after and before the Id jump around Vg ≈ -0.25 V. e, Simulated Id vs. Vg curves
without (red) and with (black) the gate leakage current taken into account.
20
|
1805.05468 | 1 | 1805 | 2018-04-28T09:06:49 | A simplified superheating Rankine pump with possible application in irrigation | [
"physics.app-ph"
] | In this paper a simplified superheating Rankine pump is presented. For an easy understanding and study of its functioning principle, the pump has been experimentally implemented, with superheating obtained by an electrical resistance. The pump can be adapted to serve as a reliable and cheap irrigation pump, having reduced maintenance and operating costs, and consuming clean energy (concentrate solar energy). | physics.app-ph | physics |
A simplified superheating Rankine pump with
possible application in irrigation
M.-F. Danca
September 23, 2018
Abstract
In this paper a simplified superheating Rankine pump is presented. For
an easy understanding and study of its functioning principle, the pump
has been experimentally implemented, with superheating obtained by an
electrical resistance. The pump can be adapted to serve as a reliable and
cheap irrigation pump, having reduced maintenance and operating costs,
and consuming clean energy (concentrate solar energy).
keywords Superheating, Rankine cycle, irrigation system
1
Introduction
As is well known, superheating appears when a liquid is heated under a tem-
perature above its boiling point without boiling (vaporization) (see e.g.
[9, 1,
2, 3, 4, 5, 6, 7, 8]). The superheating can determine a dramatic increase of
vapor volume. For example, by super heating one gram of pure water from
100◦C to 100.26◦C, one obtains 1.3 l of vapor. Generally speaking, a substance
undergoes a phase change from the liquid state to the gaseous state while it is
heated to its boiling point. As is well known, pure water boils at 100◦C under
standard atmospheric pressure. However, if one considers a container with a
smooth surface, such as a glass, the relatively static heating environment inside
a microwave oven is unfavorable for the formation of steam bubbles. Even it
is heated to or above its boiling point, the water is prevented from converting
into steam and thus the boiling process is delayed. The water is said to be in a
superheated state.
On the other side, there are several hydraulic pumps that use solar energy.
Generally, these solutions are based on a solar panel that generates electricity
and a control system used to drive a submersible water pump. The disadvantage
is the high cost of installation due to system complexity and low efficiency due
to successive conversion of solar energy into electricity-mechanical-hydraulic.
Another more complicated system, using solar energy, consists of a concentrator
solar radiation that radiates some horizontal water pipes, resulting saturated
1
Figure 1: (a) Sketch of a standard Rankine cycle; (b) Sketch of the proposed
simplified Rankine pump.
vapor passing through a solar superheater which is formed by another set of
horizontal pipes [10]. Transforming solar energy into steam can be achieved by
a system composed of a central receiver and a superheater. Solar radiation is
concentrated by a heliostat radiation on exposed surfaces of the evaporator and
superheater, resulting in superheated steam [11]. These systems have certain
disadvantages in the required irrigation water pumping operations. Also, most
irrigation systems are gravitational [13, 12].
In this paper, a simplified traditional superheating Rankine-like pump is
proposed (see e.g.
[1, 7, 14] and references therein). The pump uses water
as the working fluid, which might transport water at higher destination than
the level of the water source. Compared to the existing heating pumps, which
usually have closed water and vapor circuits, the exhaust end of the proposed
pump is open and, therefore, the system works under the exterior pressure.
However, as shown in Section 3, it can be considered as a closed-loop system,
where a working fluid repeatedly circulates through its components.
The paper is organized as follows: Section 2 describes the pump concept,
Section 3 presents the underlying related phenomena, Section 4 proposes the
possible adaptation for irrigation, and Conclusion section ends the investigation.
2 The pump
The proposed system is a simplified superheated Rankine-like cycle, the funda-
mental operating cycle, where an operating fluid is continuously evaporated and
condensed (see e.g., [7]), with a "liquid piston".
A classical Rankine superheating cycle, where the working fluid undergos the
phase change from a liquid to vapor phase, and vice versa, is composed of four
main devices: pump, boiler, turbine and condenser (Fig. 1 (a)). Superheating
2
Figure 2:
- The proposed
1(cid:13):
sealing sys-
pump.
tem; 2(cid:13): electrical resistor;
3(cid:13), 5(cid:13): pipes;
4(cid:13) plastic
box; 6(cid:13): valves system; 7(cid:13):
suction pipe; 8(cid:13): water to
be transported; 9(cid:13): trans-
ported water.
is important because it increases the thermal efficiency of the Rankine cycle.
In this paper, for simplicity, by pump one understands the proposed simpli-
fied superheating Rankine cycle system.
The system is composed of only three devices: evaporator (boiler), condenser
and a valves subsystem situated at a hight h from evaporator (Fig. 1 (b)). The
role of the pump in the classical Rankine system is replaced here by the influence
of gravitational force applied to the column of water with weight h (gravitational
water). Actually, one can consider that the proposed system still has a pump-
like (the water column in the exhaust will pump, via the liquid piston, getting
the water back in evaporator). Also, there exists a turbine-like (water transfer
within the valves subsystem) structure.
The pump was realized experimentally by the author, composed of a cylin-
drical evaporator heated by a resistor, a spiral condenser of length L, connected
with a hose to the exhaust (Fig. 2). The system is primed by filling these tubes
with water. In the top of the evaporator there remains a small volume of air
(vapor) with height l0. The experiment reveals that this vapor volume remains
about the same after every admission phase. In the exhaust, the level of the
water column, must be higher than that in the evaporator. Due to the heating
and pressure of the water column of height h, at some moment, the superheating
3
Figure 3: Pump phases and cycles time. (a) Heating; (b) Expansion; (c) Cool-
ing; (d) Admission.
produces a large amount of superheated saturated vapor which, via the liquid
piston, produces work in exhaust. The obtained amount of superheated vapor,
once arrived in the condenser, cools and then reduces suddenly its volume1 and
the cycle repeats.
A system of valves 6(cid:13) allows the water suction and evacuation. In this way,
the water can be absorbed from the source 8(cid:13) and transported to destination
9(cid:13). All tubes are 2 cm (3/4(cid:48)(cid:48)) in diameter. The evaporator, is made from glass
in order to observe the phenomena. The resistor 2(cid:13), submerged in water, is
connected to 12 V AC (or DC) source, which has an electrical resistance of,
about 100−150 Ω. The condenser is made from copper (copper alloy), to have
a quick heat dissipation. The evaporator and the condenser are placed within
a sealed plastic box filled with water. The exhaust and suction tubes, 5(cid:13) and
1Due to the short steam cooling time, the vapor condensation is negligible.
4
7(cid:13) respectively, can be plastic pipes. The tubes are coupled with the sealing
system 1(cid:13).
With the pump device dimensions indicated in Fig. 2, the optimum hight h
of the exhaust was about 150 cm.
Note that a too high length (water pressure) of the water column in exhaust
obstructs the superheating appearance, while a too small value destabilizes the
system, in the sense that the volume of water produces a too small pressure
to restart the cycle. Also, the length L of the condenser must be long enough
to include the entire huge vapor volume obtained by superheating (more than
1 m). A too small value of L cannot ensure the reducing of the vapor volume
created by superheating and, after the superheating phase, the pump stops.
3 Phenomenological analysis
Variables and notations:
t[min];
Pressure: p[mmHg] or p[N m−2];
Temperature: T [◦C] or T [K];
Time:
Length: L, h,
r: [m] or
Volume: V [m3] or V [l];
R = 8.31 N m mol−1 K−1
c = 4.18 K J kg−1 K−1.
[mm]
Other variables:
Energy: W , Q measured in [J];
Electrical power: P [W];
Electrical resistance: R[Ω];
Voltage: V [V];
Electrical direct source: DC;
Assume the following:
-Pressure losses are neglected;
-The outlet (pushed) water through exhaust and the inlet water through suction
tube are equal;
-Compression and expansion of the working medium are adiabatic reversible
(isentropic) processes2;
-Heat losses and pressure drops in particular elements of the system are negli-
gible;
-After the cooling phase, the water returns into vaporizer, with the same initial
temperature T1.
The proposed pump can be considered as a four-cycle system (Fig. 3):
for a relative long period of time (the longest,
compared to the other times), tI, temperature increases from T1 (point 1(cid:13) in
Fig. 4) to T4 (superheating temperature at point 4(cid:13); the process of heating of
the working medium can be divided into three stages: heating, evaporation and
overheating;
I: heating (Fig. 3 (a)):
II: expansion (Fig. 3 (b)):
for a short period of time, tII, temperature
remains constant, T4, and the work is done;
2Even the system is not truly isentropic, one considers that the cycle works with ideal gas,
where isentropic assumptions are applicable.
5
III:cooling (Fig. 3 (c)): for a short period of time, tIII, temperature de-
creases drastically from superheating T4 to T1 and the vapor volume decreases;
IV: water admission (Fig. 3 (d)): for a short period of time, tIV, the water
enters into the evaporator and the new water is sucked.
Note that the longest time cycle, tI , corresponds to the heating time (see
Fig. 3 (a)). This relatively long time (in order of several minutes), represents an
disadvantage, which can be compensated by the quantity of transported water,
or by the efficiency improvement (see Section 4).
The entire process is reversed by cooling the vapor, and the water from
exhaust will go back to vaporizer, retracing the same path. During this process,
the amount of heat released is considered to exactly match the amount of heat
added through the heating.
Because of a careful choice of lengths of pump elements (evaporator, con-
denser, exhaust and connection tube 3(cid:13) in Fig. 2), the water circulating within
vaporizer-condenser3 can be considered to be the same; therefore, the pump is a
reversible closed system. Actually, the liquid piston can be considered as being
composed by the water inside the evaporator and condenser, which pushes the
water from the exhaust to the valves subsystem.
The system consists of two isobars and two isentropics processes, and is
characterized by the following transformations (see the T − s diagram in Fig.
4):
1(cid:13)- 2(cid:13): Reversible isentropic (adiabatic) compression of saturated liquid at
the pressure of water column of hight h in the exhaust; a reduced work is
produced. The entropy can be reasonably considered as constant (no transfer
of heat between the system and the surroundings takes place);
2(cid:13)- 2'(cid:13)- 3(cid:13): heat addiction: the system receives Qin and an isobaric expan-
sion of steam is produced. At point 2(cid:13) the first saturated vapor appears and
the system enters the isobar 2(cid:13)- 4(cid:13). On this path, the system meets the point
2'(cid:13), where boiling water under the boiling pressure takes place. Next, on the
segment 2'(cid:13)- 3(cid:13), the system produces a binary mixture liquid-vapor, at constant
temperature and pressure. Further heating increment causes evaporation of the
liquid until it is fully converted to saturated steam (point 3(cid:13)). At the point 3(cid:13),
there exists saturated steam under the boiling pressure;
3(cid:13)- 4(cid:13): Once the isobar 2(cid:13)- 4(cid:13) crosses the saturated vapor line (point 3(cid:13)),
the system enters into the superheating region. Due to further temperature
increasing, the isobar ascend to 4(cid:13), where superheating can produce steam by
boiling pressure. Further transfer of heat results in an increase in both the
temperature and the specific volume;
4(cid:13)- 5(cid:13): The highly unstable superheating generates a huge volume of vapor
5(cid:13)- 1(cid:13): constant-pressure and temperature (isobar) heat rejection in the con-
denser; the system delivers Qout. The wet steam condenses completely along
the isobar.
and a work is produced (reversible isentropic expansion);
For clarity, all important parameters of water and steam, used in this paper,
3Only the exhaust water is (periodically) replaced.
6
are tabulated in so called "Steam Tables" [15, 16, 17, 18, 19, 20].
Suppose that T1 = 20 ◦C. The pressure at the point 1(cid:13) is composed by the
pressure in the top of the evaporator i.e. the atmosphere pressure pa and the
hydrostatic pressure ph of the (open) exhaust column of the water of height h
(Fig. 3 (a)):
p1 = pa + ph = 760 mm HG + ρgh = 780.13 mm HG.
(1)
Since at point 1(cid:13) the pressure is given by (1), the necessary temperature to
boil the water at the saturated line (point 2'(cid:13)) is: T (cid:48)
2 = 103.8 ◦C [15], and the
pressure is: p2(cid:48) = 868.61 mm Hg [16]. Because the point is on an isobar, the
same pressure will be at the point 2(cid:13). Next, one can determine the isentropic
1(cid:13)- 2(cid:13), by using the isentropic relation for an ideal gas (see also [17]):
wherefrom
T2 = T1
T2
T1
=
(cid:18) p2(cid:48)
p1
(cid:19) γ−1
γ
(cid:18) p2(cid:48)
(cid:19) γ−1
p1
γ
,
= 29.13 ◦C.
From interpolations given by online calculators and Steam Tables (see e.g.
[18],[19], or [20]), if one give a temperature slightly above value T3, e.g. T4 =
104.5 ◦C (in [18], the degrees of superheat for temperature above 103.8 ◦C is
about 0.2 ◦C), one obtains the specific superheated vapor volume V4 ≈ 1.5 l4.
4Along 2'(cid:13)- 3(cid:13), the concentration of air in the mixture is about 45% [19].
Figure 4: T − s diagram of the pump.
7
Note that, even the ideal gas law pV = nRT cannot be used for the isentropic
transformation, if one considers 1 g of water participating to this cycle, by this
formula, the steam volume is similar to those given by Steam Tables, i.e.,
V4 =
nRT4
p(cid:48)
2
=
0.055 mol × 8.31 N m mol−1 K−1 × 377.15 K
115 803.80 N m−2
≈ 1.49 l
On the other side, the volume of the exhaust, with r = 19 mm(3/4(cid:48)(cid:48)) and h =
1.5 m, is Vext = πr2h ≈ 1.7 l.
The necessary mechanical work to pump the 1.5 l of water at 1.5 m is
W = Gh = ρV gh = 1000 kg m−3×0.0015 m3×9.81 m s−21.5 m ≈ 22.1 kg m2 s−2 = 22.1 J.
On the other side, to transform one gram of water into steam between T1 =
20 ◦C = 293.15 K and T4 = 104.5 ◦C = 377.65 K, one needs the following energy:
Qin = mc∆T = 0.001 kg × 4.18 kJ kg−1 K−1 × (377.65 K − 293.15 K) ≈ 350 J,
where the specific heat of water is considered as c = 4.18 K J kg−1 K−1.
The underlying electrical power, in about 5 min necessary for the heating
cycle, is P = W/5 min = 1.17 W, which can be obtained with 12 V DC power
and electrical resistance of R = U 2/P = 144 V2/1.17 W ≈ 125 Ω.
The following efficiency can be determined: η1 = 1 − W/Qin ≈ 94 %.
4 On possible irrigation utilization
It is easy to understand that, under the sun temperature, larger evaporators
can produce superheat and the pump can serve as an irrigation system.
Thus, the evaporator, horizontal and blackened, should be situated in the
focus of a concave reflector (concentrator) mirror (Fig. 5), which is a concen-
trated solar power (see e.g. [21]) and the condenser should be placed within the
water source (river). In order to increase the system efficiency, several pumps
can be primed into a battery such that, function on evaporator temperature,
the cycles of each pump intercalate (start at different time moments) and the
entire system work pumps the water continuously-like. A possible intercalation
procedure could be as follows: every consecutive pump will start (primed) at
the end (or after a fraction) of the previous pump. In this way, because of the
periodicity of pump cycles, the delay between two successive expansions (of two
nearest pumps) can be reduced substantially and the low efficiency of the pump
can be compensated.
5 Conclusion
In this paper, a simplified superheating Rankine-like system, without the clas-
sical pump and turbine, is presented. Compared to other heating pumps, the
exhaust of the proposed pump communicates with the atmosphere pressure.
8
Figure 5: Schematic utilization of the proposed pump in irrigation. Solar energy
is directed by the concave reflector towards the heating blackened cylinder. The
metal (copper) condenser is cooled by the river water. The river water is directed
via a pipe system to destination.
The experimental model has be heated with an electrical resistance.
It was
proved both experimentally and analytically that the superheated volume of
vapor is large enough to pump water. By the liquid piston of water situated in
the exhaust, the vapor produces a mechanical work through pumping water. It
is shown that the pump can be utilized as irrigation pump. Moreover, due to
its extremely simple structure, the mechanical stresses is drastically reduced.
To increase the efficiency, several pumps can be primed at different instants,
so as to work sequentially. Compared to the existing irrigation systems, the
advantages of the proposed pump are: cheap with low construction costs (e.g.
the evaporator and condenser can be made of metal while the exhaust and pipes
can be made of plastic materials), good reliability and reduced maintenance and
operating costs.
9
References
[1] C. Sprouse III, C. Depcik, Appl. Therm. Eng. 51, 711-722 (2013). DOI:
10.1016/j.applthermaleng.2012.10.017
[2] C. A. Angell, M. Oguni and W. J. Sichina, J. Chem Phys, 86(6),
998-1002 (1982). DOI: 10.1021/j100395a032
[3] S. Mutair and Y. Ikegami, Int. J. Therm. Sci. 57, 37-44 (2012).
DOI:10.1016/j.ijthermalsci.2011.10.013.
[4] V. N. Chukanov, B. A. Korobitsyn, J. Eng. Thermophys., 16, 192-199,
2007. DOI: 10.1134/S1810232807030125
[5] P. Mirabel and J.L. Katz, J. Chem Phys., 67(4), 1697-1704, 1977. DOI:
10.1063/1.435004
[6] W. Zima, S. Gr¸adziel and A. Cebula, Archives of Thermodynamics 31(3),
19-36 (2010). DOI: 10.2478/v10173-010-0012-y.
[7] Wikipedia, "Superheated water", https://goo.gl/p1sQMm;
"Superheating", https://en.wikipedia.org/wiki/Superheating. "Rankine
cycle", https://goo.gl/WPPxG2
[8] Superheating and microwave ovens, PHYSCLIPS, UNSV, School of
Physics, Sydney, Australia.
http://www.animations.physics.unsw.edu.au/jw/superheating.htm.
[9] I. Dincer, H. Al-Muslim, Int. J. Energy Res. 2001; 25:727-739. DOI:
10.1002/er.717
[10] US2004035111- Method and device for producing steam by means of solar
energy.
[11] DE 10248068- System for solar thermal steam generation has at least one
heliostat that concentrates solar radiation onto radiation exposure surfaces
of evaporator stage and superheater stage of central receiver.
[12] F. Merk at Bucuresti, Ianuarie 2011, Strategia Investitiilor in Sectorul
Irigatiilor, Romania, Ministerul Agriculturii si Dezvoltarii Rurale,
Proiectul de Reabilitare si Reforma a Sectorului de Irigati, p.48.
http://old.madr.ro/pages/strategie/strategie-investitii-irigatii.pdf
[13] T. Arthey BSc (Hons), The Legal Structure of Irrigation In Romania
(MRICS, January 2013), Mintridge International Farmland Investment
Specialist.
[14] V.A. Kirillin, V.V. Sychev A.E. Sheindlin, Engineering Thermodynamics,
chapter 11.2, MPEI Publishers (2011).
[15] Planetcalc Online Calculatros https://planetcalc.com/275/
10
[16] Steam System Modeler Tool (SSMT), US Department of Energy,
http://goo.gl/moKVB2
[17] fxSOLVER https://www.fxsolver.com/solve/
[18] Sprax Sarco https://goo.gl/SVB4wJ
[19] TLV A Steam Specialist Company, Engineering Calculator
https://www.tlv.com/global/TI/calculator/
[20] GESTRA AG, http://www.gestra.com/service-support/calcuquick-
calculate-parameters.html?calc=23
[21] Concentrating Solar Power: Energy from Mirrors, Energy Efficiency and
Renewable Energy, DOE/GO-102001-1147 FS 128 March 2001.
11
|
1909.06625 | 1 | 1909 | 2019-09-14T16:24:32 | Self-powered, flexible and room temperature operated solution processed hybrid metal halide p-type sensing element for efficient hydrogen detection | [
"physics.app-ph"
] | Hydrogen (H2) is a well-known reduction gas and for safety reasons is very important to be detected. The most common systems employed along its detection are metal oxide-based elements. However, the latter demand complex and expensive manufacturing techniques, while they also need high temperatures or UV light to operate effectively. In this work, we first report a solution processed hybrid mixed halide spin coated perovskite films that have been successfully applied as portable, flexible, self-powered, fast and sensitive hydrogen sensing elements, operating at room temperature. The minimum concentrations of H2 gas that could be detected was down to 10 ppm. This work provides a new pathway on gases interaction with perovskite materials, launches new questions that must be addressed regarding the sensing mechanisms involved due to the utilization of halide perovskite sensing elements while also demonstrates the potential that these materials have on beyond solar cell applications. | physics.app-ph | physics | Self-powered, flexible and room temperature operated solution processed hybrid metal
halide p-type sensing element for efficient hydrogen detection
E. Gagaoudakis1,2, †, A. Panagiotopoulos3,5 †, T. Maksudov3,5 †, M. Moschogiannaki2,3, D.
Katerinopoulou1,2, G. Kakavelakis3,4, G. Kiriakidis1,2, V. Binas1,2, E. Kymakis5 and K. Petridis6, *
1Department of Physics, University of Crete, Heraklion, Greece
2Institute of Electronic Structure and Laser, Foundation for Research and Technology Hellas, 100 N. Plastira str.,
Vassilika Vouton, 70013 Heraklion, Crete, Greece
3Department of Materials Science and Technology, University of Crete, GR-71003, Heraklion, Greece
4Cambridge Graphene Centre, University of Cambridge, 9 JJ Thomson Avenue, Cambridge CB3 0FA, UK
5Department of Electrical & Computer Engineering, Hellenic Mediterranean University, Estavromenos P.B 1939,
Heraklion, GR-71 004, Crete, Greece
6Department of Electronic Engineering, Hellenic Mediterranean University, Romanou 3, Chalepa, 73100, Chania,
Crete, Greece
†these authors contributed equally to the preparation of the manuscript
KEYWORDS. lead halide perovskites, self-powered hydrogen sensing element, Solution
Processed, Room Temperature Sensing, p-type sensing elements, sensitivity, flexible
substrate
Abstract
Hydrogen (H2) is a well-known reduction gas and for safety reasons is very important to be
detected. The most common systems employed along its detection are metal oxide-based
elements. However, the latter demand complex and expensive manufacturing techniques,
while they also need high temperatures or UV light to operate effectively. In this work, we
first report a solution processed hybrid mixed halide spin coated perovskite films
(CH3NH3PbI3−xClx) that have been successfully applied as portable, flexible, self-powered, fast
and sensitive hydrogen sensing elements, operating at room temperature. The minimum
concentrations of H2 gas that could be detected was down to 10 ppm. This work provides a
new pathway on gases interaction with perovskite materials, launches new questions that
must be addressed regarding the sensing mechanisms involved due to the utilization of halide
perovskite sensing elements while also demonstrates the potential that these materials have
on beyond solar cell applications.
Introduction
Hydrogen (H2) gas is expected to be a green (no emissions) and renewable energy source
(with high heat combustion 142 kJ/g, minimum ignition energy 0.017 mJ and high flammable
range up to 75%) for many applications such as glassmaking, semiconductor processing,
biomedical applications, seismic surveillance, fuel cells, automobiles, power generators and
aerospace (liquid H2 already been used for rocket fuels) [1,2]. In the near future it could be
used as a city gas or as a fuel to power cars in the same way as natural gas is leveraged.
However, H2 is an extremely dangerous gas since it is odourless, colourless, and highly
flammable, with high burning velocities and its leakage poses explosion hazards (a lower
explosion limit (LEL) at 40000 ppm) [3]. So, it is essential (see Figure 1) to be detected reliably
and fast, in low concentrations preferably below 100 ppm, in order to monitor possible
leakage during storage and transport, for safety protection reasons [4-7].
A H2 sensor could be a transducer that converts a variation of physical or chemical
characteristic
into an electrical current. A number of applications,
including gas
chromatography, mass spectrometry, thermal conductivity,
laser-induced breakdown
spectroscopy, scanning photoelectrochemical microscopy, and gas sensors, have been
employed to detect hydrogen gas [8-14]. Among them, the hydrogen gas semiconductor-
based sensors are being studied for their small size; low power consumption; high accuracy,
reliability; fast response; and reliable / low cost fabrication processes [15]. Subject to the
signal monitored, the H2 sensing elements can be divided into (a) resistance based; (b) work
function based; (c) optical; and (d) acoustic elements. Figures of merit of an ideal sensing
element are (a) the sensitivity to various targeted gas agents; (b) the selectivity between
various targeted gases; (c) the fast response when exposed in the environment of the
targeted gas; (d) reversibility to its initial pristine stage before the exposure to the targeted
agent; (e) the efficient detection of the signal generated as a result of the interaction of the
sensing element with the targeted gas; (f) the low cost and facile fabrication process; (g)
stability and long life time (h) the operation at low temperature (ideally at room
temperature), without the need of an external trigger (temperature or UV light) to provide
sensing abilities.
In this paper, we will be focused on the resistance-based sensing element technique and
demonstrate a very promising performance towards H2 gas sensing. The motivation
originated from the fact that the most common employed H2 sensing materials (metal oxides
-- SnO2, ZnO, TiO2, Nb2O5, In2O3, FeO, NiO, Fe2O3, Ga2O3, MoO3, V2O5, WO3) do not contain all
the figures of merit of an ideal sensing element, despite their appealing characteristics: small
size, high sensitivity, high repeatability and simplicity to use. Often, they suffer from some
serious drawbacks such as (a) high operation temperatures (operation at room temperature
has shown weak response to low concentration H2 or have long response-recovery time.
Moreover operating at high temperatures is very risky especially when the targeted gas is an
explosive gas such as H2); (b) they need a pre-treatment with UV light in order to become
conductive; and (c) their fabrication is complicated and expensive (e.g. rf sputtering, dc
sputtering, high vacuum facilities, pulsed laser deposition [16-21].
Hybrid metal halide perovskite materials are the new star materials in the solar cell
technological field and not only. They are of the type ABX3 where, A is an organic cation
usually methylammonium (MA+), formamidinium (FA+) etc. or an inorganic cation such as Cs+,
Rb+ and K+, B is a metal such as Pb2+, Sn 2+, while the anion X is a halogen such as I-, Br-, Cl- or
a mixture thereof. The very first report on semiconductor halide perovskite was via observing
photoconductivity in the all inorganic CsPbX3 systems at late 1950'by Moller [22]. Later on,
perovskite was used as absorbing material in photovoltaic applications by Kojima et al. in 2009
[23]. In the last decade, the efficiency of perovskite based solar cells have been skyrocketed
from 3.8% to higher than 25% [24]. The hybrid lead halide perovskites, such as the mixed
halide one studied here (CH3NH3PbI3−xClx), have recently attracted the scientific communities'
attention as one of the most promising solar light energy harvesting materials due to their
direct bang gap, long diffusion charge carrier lengths, large absorption coefficients, long
carrier lifetime and large carrier mobility [25-27]. The impressive impact of perovskite in solar
cell technology, have attracted an intensive research interests towards the applications of
perovskites beyond solar cells such as in lasers [28] in light emitting diodes [29,30] and photo-
detectors [31,32]. However, one of the key issues to be solved in order to boost their
commercialization is related to their stability. Halide perovskites are very sensitive to polar
gases and vapours, as, exposure to such elements deteriorates substantially and very fast the
perovskite devices' performance. The conversion of this disadvantage to an advantage
reflects the introduction of perovskite-based gas sensing elements. The sensitivity of
perovskite to environmental gases is an opportunity to convert a drawback to an opportunity
[33]. Recently, halide perovskites have been explored and demonstrate their competence
over the existed technologies, as potential sensing elements from gas molecules to X-ray
photons [34-38]. Increasingly new publications regarding inorganic perovskites applications
in H2 sensing, have started to appear [39,40] (Figure 1) showing the potential of this family of
materials towards hydrogen sensing. However, all reported inorganic perovskite materials
need to operate at very high temperatures (of few hundred Celsius) in order to function as
hydrogen sensors. In this paper, we introduce for the first time a solution processed hybrid
metal halide perovskite film (CH3NH3PbI3−xClx), that requires much simpler fabrication and
deposition techniques & facilities than the previously reported for other H2 sensing element
materials. Moreover, the performance characteristics of the demonstrated sensing element
are not inferior to many of figures of merit of the other technologies reported. The perovskite
based sensing element presented here, demonstrated a very promising performance
characteristics towards hydrogen sensing, i.e.: (a) operation at room temperature; (b) no
requirement of an external optical signal to be switched on prior to its exposure into a
hydrogen environment; (c) fast response time (few secs); (d) detection sensitivity with
resolution down to 10 ppm; and (e) compatibility with flexible substrates. The
aforementioned features make hybrid mixed halide perovskite semiconductors competitive
(and beyond the simpler and of lower cost fabrication and deposition techniques) candidates
to other already demonstrated in hydrogen sensing materials such as metal oxides and metals
[4]. More particular compared to metal oxides (e.g. SnO2, ZnO, TiO2, FeO, Fe2O3, NiO, Ga2O3,
In2O3, Sb2O5, MoO3 and WO3) that request high temperatures to operate as hydrogen sensing
elements, the demonstrated hybrid perovskite film operate in room temperature reducing a
lot the consuming power. To highlight that the performance of both systems regarding the
reaction times (of the order of few seconds) and minimum detection limits are similar (of the
order of 10 ppm). Regarding their advantages compared to metallic resistors (e.g. palladium
& platinum metals) (a) request simpler deposition techniques (e.g. spin coating) compared to
more complicated and expensive techniques the metallic resistors request: vacuum
evaporation, electrode position, sputtering and pulsed laser deposition; (b) do not suffer from
mechanical degradation when exposed to hydrogen environment and thus do not request
complicated metallic alloys to address the issue of the mechanical degradation. To add that
hybrid perovskite semiconductors, show similar minimum detection limits as metallic
resistors.
The studied sensing films demonstrated a p-type semiconductor behaviour (attributed to its
stoichiometry) and thus under H2 gas (reducing gas) exposure their electrical resistance was
enhanced upon H2 exposure. However, its pristine electrical properties were restored very
fast (within few secs) after the removal of the H2 gas. This work is believed to set the
framework for research of low temperature operated, efficient and of low cost new
conductometric (resistance measurement), compatible with flexible electronic industry
halide perovskite H2 sensing elements and systems. This is essential for various applications
ranging from the energy sector to aerospace industry.
Figure 1: Number of publications regarding H2 sensing and applications of perovskites as gas sensors and H2
sensors (Source Scopus)
Methodology
To fabricate the CH3NH3PbI3−xClx precursor solution, methylammonium iodide (MAI,
purchased by Dyesol) with lead (II) chloride (PbCl2, 99.999% purchased by Sigma Aldrich) were
mixed in a molar ratio of 3:1 within a anhydrous N,Ndimethylformamide (DMF) solvent. The
resulted solution (with 40 wt% concentration) was continuously stirred on a hot plate for 12
hrs at 70 °C. Afterwards, the precursor solution was cooled down at room temperature and
deposited (approximately 20 μl filtered precursor solution) onto the sensing element
electrodes template (purchased from DropSens). The latter and prior to the spreading of the
solution precursor, was UV ozone cleaned to remove any hazards (e.g. organic contaminants)
that can lower its hydrophilicity. The electrode substrate was glass made and on top of the
glass two interdigitated platinum electrodes were pre-patterned. The distance between the
electrodes was 5 μm. The deposited solution was spin-coated at 4000 rpm for 45 seconds.
Afterward, the spin-coated precursor solution over the electrodes was thermally annealed at
100 °C for 75 minutes to form the CH3NH3PbI3−xClx hybrid perovskite semiconductor. The
entire processing was conducted inside a nitrogen-filled glove box with O2 and H2O
concentrations below 0.1 ppm. Prior to the testing of the sensing properties of the
glass/Pt/CH3NH3PbI3−xClx sandwich elements, the latter were fully characterized as a function
of their electronic, morphological, structural properties. Prior to the sensing testing the
glass/Pt/CH3NH3PbI3−xClx system was exposed under simulated solar light (A.M 1.5G at
100mW/cm2) in ambient conditions (~45% relative humidity). The film roughness was
examined using Atomic Force Microscopy (AFM), by employing a Park XE-7 instrument in
tapping mode. The total scan area was set to 50 μm x 50 μm and the scan rate was fixed at
0.3 Hz. The successful crystallization of the hybrid CH3NH3PbI3−xClx perovskite semiconductor
was assessed using a D/MAX-2000 X-Ray diffractometer under monochromatic Cu Kα
irradiation (λ=1.5418 Å) at a scan rate of 4° min−1. Whereas the grain size of the crystallized
perovskite film was evaluated through Scanning Electron Microscope (SEM, using the JEOL
JSM-7000F) measurements.
All the hydrogen sensing measurements occurred within a homemade gas test chamber
under dark conditions in order to lower the rates of photoexcitation that contributes to dark
current. Prior to the sensing test measurements that occurred, the electrical conductivity of
our elements was tested. A potential difference of one Volt was applied across the two
platinum electrodes of glass/Pt/CH3NH3PbI3−xClx systems and the induced generated current
was measured using a Keithley 6517A multi-meter set-up. All these initial measurements
were taken at room temperature and under a pressure of 1.6 mbar.
After the baseline of the conductivity of our sensing elements was measured, the
CH3NH3PbI3−xClx sensors were exposed for five minutes to hydrogen gas at a constant flow of
500 sccm (standard cubic centimetres per minute), while the pressure in the chamber was
kept constant at 120 mbar, leading to a decrease of current. The hydrogen concentrations
that the detector was exposed were 100, 75, 50, 25, and 10 ppm. The sensing performance
of our sensing elements could not be assessed for hydrogen concentrations higher than 100
ppm for safety reasons. The CH3NH3PbI3−xClx sensor was exposed for time intervals of five
minutes at each concentration, while another five minutes was given to the sensor to relax to
its steady-state conditions.
Results
1. Structural, Morphological and Optical Analysis
A 300 nm thick CH3NH3PbI3-xClx film was fabricated onto a glass substrate with prepatterned
electrodes; the interdigitated electrodes made of platinum had distance between them of 5
μm. It must be highlighted at this point that the majority carriers within the perovskite
semiconducting film are subject to the stoichiometry of the PbI2/MA+I precursor ratio; in our
case this ratio was less than one and the film demonstrated a p-type semiconducting
behaviour [41]. The entire fabrication process was performed inside a glovebox under
nitrogen atmosphere.
The film's surface morphology controls the number of the provided interaction pathways that
the targeted gas molecules can interact on with the sensing element; increased roughness,
facilitates the gas molecules adsorption by providing longer diffusion lengths within the active
material. The surface film's morphology was revealed using Atomic Force Microscopy (AFM)
measurements (Figure 2). AFM patents were taken before (Figure 2a) and after the exposure
to H2 (Figure 2b). A striking change in the AFM images was observed, with the roughness to
reduce almost to its half value after the exposure to 100 ppm of hydrogen gas. The reason of
this morphological change is thought to relate to the interaction of H-molecules with the
perovskite surface species; however, the exact cause needs further investigation before it can
be attributed to a specific factor. Subsequently, a slight decrease in the measured current (of
the order of few nA during the measurement cycles) was attributed to the resulted smoother
surfaces; the latter provide shorter percolations paths to allow the H2 molecules to interact
with the perovskite platform.
(a) (b)
Figure 2: AFM image of the surface morphology of the MHP film; the RMS value was 69.27 nm before the H2
exposure (a) whereas after prolonged exposure the morphology has changed to 36.70 nm (b)
The Scanning Electron Microscopy (SEM) images of the employed mixed halide films reveal
the well-shaped formed grains on its surface and the film's porosity. The existence of a porous
surface on the film, facilitates the penetration and the evacuation paths of the hydrogen
atoms through it (Figure 3). The good conductivity of the film is very much linked with the
grain size; using the Scherrer formula their size was measured of the order of 250 nm. The
SEM images before and after the exposure to the H2 gas revealed no substantial changes on
the grain size.
Figure 3: SEM image of the surface of the MHP employed as sensing element for hydrogen before (left image)
and after (right image) exposure to hydrogen. No changes in the morphology were observed. Using the Scherrer
equation the grain size was calculated to be of the order of 250 nm
The film's crystal structure was studied using the X-ray diffraction (XRD) technique. The
successful crystallization of the spin coated perovskite film is depicted in the acquired XRD
image (Figure 4). The crystal directions of the first (110) and the second (220) crystallographic
plane can be seen, at ~14.2o and ~28.5o respectively, confirming the, as expected, cubic
perovskite phase [42]. No compositional changes were observed before and after the H2 to
hydrogen gas molecules (at 100 ppm for more than one hour), showing the sensing potential
and stability of the fabricated films. This is a clear indication that the interacted H2 molecules
just adsorbed and de-adsorbed through the perovskite-based template.
Figure 4: XRD images of the mixed halide perovskite films before and after its exposure to 100 ppm hydrogen
gas. The results showed no structural changes after the interaction of the sensing element with the hydrogen.
The XRD of the non-exposed halide perovskite film was taken on a glass substrate without electrodes
2. Sensing Properties of the CH3NH3PbI3-xClx thin films
The sensing functionality of the films on H2 was examined by electrical measurements at room
temperature, carried out in a home-made set-up. We exposed the aforementioned perovskite
films to various H2 gas concentrations up to the maximum of 100 ppm for safety reasons.
Interruption the H2 gas flow, restored the films' conductivity to their initial levels. The
excellent electrical properties (long diffusion lengths, high charge mobility and excellent
mobility to charge lifetime product), the interdigitated electrodes (Pt electrodes with distance
of the order of 10 μm) setup and the application of a forward bias across the electrodes (0.5
and 1 V) allowed us to register (a) the current across the film (of the order of μΑ); and (b) its
modulations when the film was exposed to H2 environment. The report of such high currents
(in μΑ range) through the film is an indication of its excellent electrical properties as a result
of the long grain's size and high carrier's diffusion length of the CH3NH3PbI3−xClx perovskite
layer employed as sensing material. The demonstrated sensing elements were self-powered
thus there was no the need of any external assistance such as UV irradiation or heating in high
temperatures. H2 molecules operated as a reducing gas, and since the exhibited perovskite
film is a p-type semiconductor, lowering of the current was observed, as expected, caused by
the adsorption of the H2 molecules (the opposite behaviour than n-type semiconductors).
Figure 5 depicts the current modulation for (a) various exposure times (one and five minutes);
and (b) under different forward biases 0.1, 0.5 and 1 Volt. The O.1 V acquired measurements
were ignored since provided very low currents and low sensitivities.
Figure 5: Dependence of the sensitivity (with the external bias) and demonstration of the reproducibility &
stability of the solution processed CH3NH3PbI3-xClx sensing elements as a function of the operational time (under
exposure at multiple cycles of synthetic air-100 ppm H2) and the applied external bias (1V, 0.5V and 0.1 V)
During the sensing measurement process, the modulation of the current flowing through the
film was recorded, as the H2 gas in the synthetic air was inserted and stopped, following
several cycles. The film demonstrated sensing reversibility as this is illustrated in Figure 6.
Interruption H2 gas flow lead to the recovery of the prior sensing element electrical resistance.
The reproducibility of the acquired results, supports the credibility of our sensing element
performance and reliability. The sensing film demonstrated a reversible sensing behaviour
under various hydrogen gas concentrations: 100 ppm, 75 ppm, 50 ppm, 25 ppm and 10 ppm.
As a result, the reducing gas causes an increase in the film resistance, reconfirming its p-type
character.
010203040810121490105120320340360380air100 ppm H2 t (min) Vin = 0.1 V5 min air5 min H21 min air1 min H2 I (nA) Vin = 0.6 V Vin = 1.0 V
Figure 6: Reversible current modulation through 300 nm thick mixed halide perovskite films under exposure (for
five minutes) in various hydrogen concentrations. As was expected the sensitivity drops with the concentration
of the hydrogen atoms within the chamber. The minimum detection limit was 10 ppm of H2
As expected, our perovskite sensing platform demonstrated different sensitivity at different
targeted gas concentrations (Figure 6). This, in combination with the abrupt increase of the
resistance, was attributed to the H2 molecules adsorption into the perovskite surface.
Furthermore, as the need for low temperature gas sensors is becoming a necessity and
moreover for safety reasons in the case of the highly flammable H2 gas, all the measurements
were taken at room temperature although sacrificing part of the response signal obtained at
elevated temperatures. All the measurements were recorded until the current through the
sample had reached its lowest value. This process duration was lasted for approximately five
minutes after which no further substantial changes were observed. The sensing measurement
was repeated under each hydrogen gas concentration for five times (five cycles) -- Figure 6,
showing excellent reproducibility. The sensing ability was assessed with the measurement of
(a) the sensitivity (S); and (b) the response & recovery times of the film. The sensitivity
parameter is calculated using the following formula (1) [43]
𝑆(%) =
𝐼𝑎𝑖𝑟−𝐼𝑔𝑎𝑠
𝐼𝑎𝑖𝑟
100% (1)
where Iair denotes the electrical current of the sensor before the exposure into the H2
environment, Igas the resistance of the sensor after ten minutes to hydrogen. The response
(tresp.) and the recovery (trecov.) times were calculated as the times take the measured
current to reach the 10% of its maximum value under H2 exposure and 90% of its
maximum value under synthetic air environment (H2 gas flow has been interrupted),
respectively. The response and recovery times demonstrated almost identical average
values, 45 and 35 seconds, respectively. The dependence of sensitivity with the H2
concentration is depicted on Figure 7.
Figure 7: Sensitivity dependence on the concentration of the H2 atoms
Table 1 presents all the calculated values, extracted from the measurements depicted on
Figure 7, under the different H2 gas concentrations taken. All these values have been plotted
and illustrated on Figure 8.
H2 concentration
(ppm)
100
75
50
25
10
Iair
(nA)
343.6
339.6
336.8
334.8
332.6
Igas
(nA)
ΔI
(nA)
325.6
328.6
331.4
332.4
331.6
18
11
5.4
2.4
1
S
(%)
5.2
3.2
1.6
0.7
0.3
tresp
(s)
58.50
44.58
34.56
49.56
38.52
mean
tresp
(s)
45.14
trec
(s)
20.40
23.04
42.48
30.00
61.08
mean
trec
(s)
35.40
Table 1: Calculated values of the sensitivity (S), response and recovery times of the demonstrated perovskite
films under various H2 gas concentration exposures
Figure 8: Mean values of electrical current after exposure to H2 (Igas) and synthetic air (Iair) and response of sensor
as function of H2 concentration. Sensitivity increased with the concentration of the H2 molecules
The fast sensing response of the exhibited films allowed the distinction of the various H2
concentrations. You can see in Figure 10 below, that the maximum exposure time was 10s.
Figure 9: Normalized current with time of exposure at different concentrations of H2
The absence of any phase transformation and impurity formation, upon exposure to a H2
atmosphere, shows the non-chemical interaction of the hydrogen with the halide perovskite
platform. This is also supported from the reversibility of the electrical properties of the
perovskite-based film exhibited after the removal of the H2 gas molecules from the chamber.
It is crystal clear that the H2 molecules are adsorbed through the porous of the perovskite film
and bond loosely with its crystal template close to the surface; this can be interpreted by the
small response and recovery times the film showed. The increase of the films' resistance
under the H2 exposure (a reduction gas) indicate the introduction of electrons to the halide
perovskite platform. These electrons recombine with the holes (majority charge carriers in
the p-type semiconductor) and result in the lowering of the current through the film. We
attribute the donation of the electrons to the perovskite platform to the following
mechanism: Firstly, neutral oxygen molecules (the most probable from the synthetic air
existed within the vacuum chamber) are adsorbed into the perovskite platform and convert
into oxygen ions: O2
−, O− and O2− ions by attracting electrons from the semiconductor valence
band. Subsequently, the hydrogen atoms interact with these oxygen ions, producing water
with the simultaneous release of electrons. The latter recombine with the perovskite holes in
the valence band and result in the observed enhancement of the film's resistance. When
hydrogen is desorbed, the surface of the material continues to adsorb oxygen from the
environment to generate oxygen ions. At this time, the resistance of the material returns to
the base value. However, the XRD measurements showed no compositional changes (the
formation of water molecules should lead to a severe degradation and decomposition of the
halide perovskite). This does not occur even though the films were exposed to H2
environment for more than an hour. This is an open question. We conclude that further
studies and theoretical modelling should be done to support further the aforementioned
operational mechanism for the H2 sensing25.
It is noted that after a number of successive cycles, the current did not fully return to its initial
value. This is attributed possibly to residual adsorbed H2 molecules due to slight changes in
the film's morphology as the AFM measurements revealed. The observation of no
compositional change after the exposure to H2 molecules is attributed to the good crystallinity
of the perovskite film that reflects film's good stability. Moreover, the sensing properties in
successive hydrogen/synthetic air switching cycles of the mixed halide films was tested after
three weeks of storage under vacuum within the measurement chamber demonstrating
remarkable sensing element durability. The results were encouraging, with the sensing
element to demonstrate a good stability (see Figure 10), as the sensitivity and the detected
currents remained almost the same after three weeks of storage.
Figure 10: Electrical current variation of MHP sensor after exposure at 100 ppm H2 and synthetic air at input
voltage of 1.0 V. The measurements were taken for as deposited samples as well as after three weeks
The ability to distinguish different gases was also demonstrated by the fabricated films. The
same film exposed to the H2 gas, was exposed to ozone molecules (Figure 11a and b). The
latter lowered the resistance of the films. The opposite result in the electronic properties of
the same sensing platform the two different gases produced, accompanied with the different
calculated sensitivities (in the case of ozone exposure the perovskite film demonstrated much
better sensitivities), revealed the ability of the mixed halide perovskite films to distinguish
two different gases.
(a)
(b)
Figure 11: The same film measured ozone molecules; the latter operated as oxidizing gas with the perovskite
platform showing the selectivity of our sensing element compared to the reductive behaviour to the same
platform of the H2 molecules. (a) The modulation of electrical current under exposure to various ozone
concentrations. (b) The impact of the various ozone concentrations to the sensing times of the perovskite film
010203040500,40,60,811,21,41,61,822,2airairairairairairair3 ppb11 ppb50 ppb175 ppb520 ppb830 ppb1800 ppb I (mA)t (min)2400 ppbMHP-O3-AIR-RT-1 V-Concentrations050010001500200025008090100110120130140 tresp trecO3 concentration (ppb)tresp (s)115120125130135140145 trec (s) The solution processability and the crystallization of the employed perovskite in low
temperatures, permitted the deposition of sensing active onto flexible e.g. PET substrates
where similar electrode patterns with these films on rigid substrates have been printed. The
film was exposed to H2 environment under (a) no bending; and (b) bending conditions. In
Figure 12a-c is depicted the sensing performance under bending conditions (Fig. 12b); and
the two extreme cases of no bending and bending close to 180o (Fig. 12c); in both cases the
modulation of current due to H2 interaction with the perovskite template could be detected
and be measured. This is the first time that a report on the flexibility of the tested material as
sensor is made and these results are very promising. The sensitivity recorded was comparable
(around 7% under bending and exposed to 100 ppm of hydrogen molecules) with the one of
the mixed halides films deposited onto glass substrates.
(a)
(b)
(c)
Figure 12: (a) Demonstration of hydrogen sensing ability onto PET based substrate under 10 Volts biasing voltage
for 100 ppm H2 molecule concentration under bending angle of 60o; (b), the sensing element bended up to 180o
(c)
0240480202122232425V=10 VoltP=700 mbarS(%)=6.6MHP/plastic-H2-BENDEDsyntheticairsyntheticairsyntheticair100 ppmH2100 ppmH2I (nA)t (s)100 ppmH202468101215202530354045502nd cycle1st cycle2nd cycleI (nA)t (min) 0o 180o100 ppm H2synthetic air1st cycle 3. Conclusions
We report here for the first time, solution processed hybrid mixed halide perovskite films
(CH3NH3PbI3-xClx) as gas sensor elements. The films were prepared and spin coated onto rigid
(glass) and flexible (PET), prepatterned with interdigitated electrodes, substrates. All films
operated as self-powered sensing elements for H2 at room temperature. Their sensing
properties were based on modulations of its electrical resistance. H2 sensing measurements
revealed very promising results regarding the potential of this material as H2 sensing element;
the p-type semiconductor characterised by maximum sensitivities of the order of 5-7%, with
very fast reaction times of the order of few seconds and the capability to be able to distinguish
between two different tested gases (H2 and O3). The compatibility of the demonstrated
sensing element with flexible substrates was also confirmed. Ageing measurements showed
that the devices retained their sensing abilities even after three weeks of storage. However,
despite the first promising results obtained, further work is planned be done in order to
improve mixed halide perovskite film's sensing performance towards H2 gas molecules. First
of all, the operational mechanism should be confirmed and studied further. Second, doped
hybrid perovskite mixed halide films or nanocomposites with graphene-based materials that
exhibit higher conductivities, should be tested. The impact of higher conductivities to the
sensing performance must be linked. This is expected to enhance the acquired sensitivity.
Another idea, is to try to apply 2D solution processed perovskite materials since the lowering
of dimension leads to higher surface to volume ratio and thus to the enhancement of all the
figures of merit of a sensor: (a) higher sensitivities; (b) broader limit of detection; (c) higher
stabilities. The better engineering of mixed halide perovskite probably will provide more
tolerant to ion migration problem and will allow the application of higher applied voltages.
This is expected to elevate the sensitivity of the particular sensing elements. Stimulated
emission Raman Spectroscopy principles will be employed as a tool to be able to distinguish
the various gases in situ.
Based in the above, it may be concluded that the hybrid mixed halide perovskite films could
be a promising self-powered p-type sensing material for reducing gases as H2 at room
temperature.
Acknowledgements
Part of this work was supported by the project "National Research Infrastructure on nanotechnology,
advanced materials and micro/nanoelectronics" (MIS 5002772), as well as by the action "QUALITY of
LIFE" (MIS 5002464) both of which are implemented under the "Action for the Strategic Development
on the Research and Technological Sector" funded by the Operational Programme "Competitiveness,
Entrepreneurship and Innovation" (NSRF 2014-2020) and co-financed by Greece and the European
Union (European Regional Development Fund).
Corresponding Author
*Address correspondence to [email protected]
Notes
The authors declare no competing financial interest.
Bibliography
1. A. Sanger, A. Kumar, R. Chandra, Highly sensitive and selective hydrogen gas sensor
using sputtered grown Pd decorated MnO2 nanowalls, Sensors and Actuators B:
Chemical 2016, 8, 234
2. K.J Yoon, S.I. Lee, H. An, J. Kim, J.W. Son, J.H. Lee, H.J. Je, H.W. Lee, B.K. Kim, Gas transport
in hydrogen electrode of solid oxide regenerative fuel cells for power generation and
hydrogen production, Int. J. Hydrogen Energy 2014, 39, 3868-3878
3. E.S. Smotkin, J. Jiang, A. Nayar, R. Liu, High-throughput screening of fuel cells
electrocatalysts, Applied Surface Science 2006, 252, 2573-2579
4. T. Hubert, L. Boon -- Brett, G. Black, U. Banach, Hydrogen Sensors -- A review, Sens. Actuat.
B 2011, 157, 329-352
5. W.J. Buttner, M.B. Post, R. Burgess, C. Rivkin, An overview of hydrogen safety sensors and
requirements, Int. J. Hydrog. Energy 2011, 36, 2462 -- 2470
6. J.S. Aroutiounian, Hydrogen detectors, Int. Sci. J. Altern. Energy Ecol., 2005, 3, 21-31
7. J.V. Lith, A. Lassesson, S.A. Brown, M. Schulze, J.G. Partridge, A. Ayesh, A hydrogen sensor
based on tunnelling between palladium clusters, Appl. Phys. Lett. 2007, 91, 181910
8. I. Simon, M. Arndt, Thermal and gas-sensing properties of a micromachined thermal
conductivity sensor for the detection of hydrogen in automotive applications, Sens
Actuators A Phys 2002, 97-98, 104-8
9. T. Hübert, L. Boon-Brett, G. Black, U. Banach, Hydrogen sensors -- A review, Sens Actuators
B Chem 2011, 157, 329-52
10. F. Zhao, F. Conzuelo, V. Hartmann, H. Li, S. Stapf, M.M. Nowaczyk, et al., A novel versatile
microbiosensor for local hydrogen detection by means of scanning photoelectrochemical
microscopy, Biosens Bioelectron 2017, 94, 433-7
11. M. Fisser, R.A. Badcock, P.D. Teal, A. Hunze, Optimizing the sensitivity of palladium-based
hydrogen sensors, Sens Actuators B Chem 2018, 259, 10
12. V. Ganzha, K. Ivshin, P. Kammel, P. Kravchenko, P. Kravtsov, C. Petitjean, et al.,
Measurement of trace impurities in ultra-pure hydrogen and deuterium at the parts-per-
billion level using gas chromatography, Nucl Instrum Methods Phys Res A 2018, 880, 181-
7.
13. B. Podlepetsky, M. Nikiforova, A. Kovalenko, Chip temperature
influence on
characteristics of MISFET hydrogen sensors, Sens Actuators B Chem 2018, 254, 1200-5
14. Ytsma, C.R.; Dyar, M.D. Effects of univariate and multivariate regression on the accuracy
of hydrogen quantification with laser-induced breakdown spectroscopy, Spectrochim Acta
Part B At Spectrosc 2018, 139, 27-37
15. J. Moon, H.-P. Hedman, M. Kemell, A. Tuominen, R. Punkkinen, Hydrogen sensor of Pd-
decorated tubular TiO2 layer prepared by anodization with patterned electrodes on
SiO2/Si substrate, Sens Actuators B Chem 2016, 222, 190-7.
16. M. Stamataki, D. Tsamakis, N. Brilis, I. Fasaki A. Giannoudakos and M. Kompitsas,
Hydrogen gas sensors based on PLD grown NiO thin film structures phys. stat. sol. 2008,
205, 2064 -- 2068
17. M. A. Haija, A. I. Ayesh, A. Sadiqa, M. S. Katsiotis, Selective hydrogen gas sensor using
CuFe2O4 nanoparticle based thin film, Surf.Sci., 2016, 369, 443
18. D. Zhang, Y. Sun, C. Jiang, Y. Zhang, Room temperature hydrogen gas sensor based on
palladium decorated tin oxide/molybdenum disulfide ternary hybrid via hydrothermal
route, Sensors and Actuators B: Chemical 2017, 242, 15
19. C.-H. Wu, Z. Zhu, S.-Y. Huang, and R.-J. Wu, Preparation of palladium-doped mesoporous
WO3 for hydrogen gas sensors, J. Alloys Compd. 2019, 776, 965 -- 973
20. E. Gagaoudakis, G. Michail, V. Kampylafka, K. Tsagaraki, E. Aperathitis, K. Moschovis, V.
Binas, and G. Kiriakidis, Room Temperature p-type NiO Nanostructure Thin Film Sensor
for Hydrogen and Methane Detection, Sensor Lett. 2017, 15, 1-5
21. M. Weber, J.-Y. Kim, J.-H. Lee, J.-H. Kim, I. Iatsunskyi, E. Coy, P. Miele, M. Bechelany, S.S.
Kim, Highly efficient hydrogen sensors based on Pd nanoparticles supported on boron
nitride coated ZnO nanowires, J. Mater. Chem. A, 2019,7, 8107-8116
22. C.K. Moller, Crystal Structure and Photoconductivity of Caesium Plumbohalides, Nature
1958, 182, 1436
23. A. Kojima, K. Teshima, Y. Shirai, T. Miyasaka, Organometal Halide Perovskites as Visible-
Light Sensitizers for Photovoltaic Cells, J. Am. Chem. Soc. 2009, 131, 176050-6051
24. https://www.nrel.gov/pv/assets/pdfs/best-research-cell-efficiencies.20190802.pdf
25. W. Zhang, G.E. Eperon, H.J. Snaith, Metal halide perovskites for energy applications. Nat.
Energy 2016, 1, 16048
26. S.D. Stranks, H.J. Snaith, Metal-halide perovskites for photovoltaic and light-emitting
devices. Nat. Nanotechnol. 2014,10, 391-402
27. T.M. Brenner, D.A. Egger, L. Kronik, G. Hodes, D. Cahen, Hybrid organic -- inorganic
perovskites: low-cost semiconductors with intriguing charge-transport properties. Nature
Reviews Materials 2016, 1, 15007
28. M.M. Stylianakis, T. Maksudov, A. Panagiotopoulos, G. Kakavelakis, and K. Petridis,
Inorganic and Hybrid Perovskite Based Laser Devices: A Review, Materials 2019, 12, 859
29. W. Xu, Q. Hu, S. Bai, C. Bao, Y. Miao, Z. Yuan, T. Borzda, A.J. Barker, E. Tyukalova, Z. Hu,
M. Kawecki, H. Wang, Z. Yan, X. Liu, X. Shi, K. Uvdal, M. Fahlman, W. Zhang, M. Duchamp,
J.-M. Liu, A. Petrozza, J. Wang, L.-M. Liu, W. Huang, F. Gao, Rational molecular passivation
for high-performance perovskite light-emitting diodes, Nature Photonics, 2019, DOI:
10.1038/s41566-019-0390-x
30. M. Yuan, L.N. Quan, R. Comin, G. Walters, R. Sabatini, O. Voznyy, S. Hoogland, Y. Zhao,
E.M. Beauregard, P. Kanjanaboos, Z. Lu, D.H. Kim, E.H. Sargent, Perovskite energy funnels
for efficient light-emitting diodes, Nat. Nanotechnol. 2016, 11, 872
31. F. Mei, D. Sun, S. Mei, J. Feng, Y. Zhou J. Xu and X. Xiao, Recent progress in perovskite-
based photodetectors: the design of materials and structures, ADVANCES IN PHYSICS: X
2019, 4, 1592709
32. Y. Fang, Q. Dong, Y. Shao, Y. Yuan, J. Huang, Highly narrowband perovskite single-crystal
photodetectors enabled by surface-charge recombination. Nat. Photonics 2015, 9, 679
33. Z. Zhu, Q. Sun, Z Zhang, J. Dai, G. Xing, S. Li, X. Huang and W. Huang, Metal halide
perovskites: stability and sensing-ability, J. Mater. Chem. C,2018, 6, 10121-10137
34. G. Kakavelakis, E. Gagaoudakis, K. Petridis, V. Petromichelaki, V. Binas, G. Kiriakidis and E.
Kymakis, Solution Processed CH3NH3PbI3−xClx Perovskite Based Self-Powered Ozone
Sensing Element Operated at Room Temperature, ACS Sens. 2018, 3, 135−142
35. K. Brintakis, E. Gagaoudakis, A. Kostopoulou, V. Faka, A. Argyrou, V. Binas, G. Kiriakidis and
E. Stratakis, Ligand-free all-inorganic metal halide nanocubes for fast, ultra-sensitive and
self-powered ozone sensors, Nanoscale Adv., 2019, 1, 2699-2706
36. H. Wei, Y. Fang, P. Mulligan, W. Chuirazzi, H. H. Fang, C. Wang, B. R. Ecker, Y. Gao, M. A.
Loi, L. Cao, J. Huang, Sensitive X-ray detectors made of methylammonium lead tribromide
perovskite single crystals. Nat. Photonics 2016, 10, 333−339
37. Z. Zhu, Q. Sun, Z. Zhang, J. Dai, G. Xing, S. Li, X. Huang and W. Huang, Metal Halide
perovskites: stability and sensing ability, J. Mater. Chem. C 2018,6, 10121-10137
38. E. Kymakis, A. Panagiotopoulos, M.M. Stylianakis and K. Petridis, Organometallic Hybrid
Perovskites for humidity and gas sensing applications, submitted book chapter in 2Dl
Nanomaterials for Energy Applications, 2019, (under publication)
39. A. Bala, S.B. Majumder, M. Dewan, A.R. Chaudhuri, Hydrogen sensing characteristics of
perovskite-based calcium doped BiFeO3 thin films, International Journal of Hydrogen
Energy 2019, 44, 18648 -- 18656
40. W. Huang, Y. Li, Y. Ding and Li, H., Preparation and conductive properties of double
perovskite Ba3Sr1+xTa2-xO9-δ and application for hydrogen sensor, Journal of Alloys and
Compounds, 2019, 792, 759 -- 769
41. P. Cui, D. Wei, J. Ji, H. Huang E. Jia, S. Dou, M. Li, Planar p -- n homojunction perovskite solar
cells with efficiency exceeding 21.3%. Nature Energy, 2019, 4, 150 -- 159
42. Y. Li, W. Sun, W. Yan, S. Ye, H. Peng, Z. Liu, Z. Bian, C. Huang, High Performance Planar
Solar Cells Based on CH3NH3PbI3-xClx Perovskites with Determined Chlorine Mole
Fraction. Adv. Funct. Mater. 2015, 25, 4867
43. H. Steinebach, S. Kannan, L. Rieth and F. Solzbacher, H2 gas sensor performance of NiO at
high temperatures in gas mixtures, Sensors and Actuators B 2010, 151, 162
|
2001.09939 | 1 | 2001 | 2019-10-21T02:50:47 | Hydrogel Leclanche Cell: Construction and Characterization | [
"physics.app-ph",
"physics.chem-ph"
] | A liquid-to-gel based Leclanch\'e cell has been designed, constructed, and characterized for use in implantable medical devices and other applications where battery access is limited. This well-established chemistry will provide reliable electrochemical potential over a wide range of applications and the novel construction provides a solution for the re-charging of electrodes in hard to access areas such as an internal pacemaker. The traditional Leclanch\'e cell comprised of zinc (anode) and manganese dioxide (cathode), conductive carbon powder (acetylene black or graphite), and aqueous electrolyte hydrogel (NH4Cl and ZnCl2) has been suspended in an agar hydrogel to simplify construction while maintaining electrochemical performance. Agar hydrogel, saturated with electrolyte, serves as the cell support and separator allowing for the discharged battery suspension to be easily replaced once exhausted. | physics.app-ph | physics | Hydrogel Leclanché Cell: Construction and Characterization
G. Jenson1,*, G.Singh2, A. Ratner2, J. Bhama1†
1 University of Iowa, Department of Surgery, Iowa City, Iowa, 52242
2 University of Iowa, Department of Mechanical Engineering, Iowa City, Iowa, 52242
†Bhama-Ratner Artificial Heart and MCS Advancement Lab (BRAHMA) Lab, Department of Surgery,
Carver College of Medicine, University of Iowa, Iowa City, Iowa, 52242
*E-mail: [email protected]
A liquid-to-gel based Leclanché cell has been designed, constructed, and characterized for use in
implantable medical devices and other applications where battery access is limited. This well-
established chemistry will provide reliable electrochemical potential over a wide range of applications
and the novel construction provides a solution for the re-charging of electrodes in hard to access areas
such as an internal pacemaker. The traditional Leclanché cell comprised of zinc (anode) and
manganese dioxide (cathode), conductive carbon powder (acetylene black or graphite), and aqueous
electrolyte hydrogel (NH4Cl and ZnCl2) has been suspended in an agar hydrogel to simplify
construction while maintaining electrochemical performance. Agar hydrogel, saturated with
electrolyte, serves as the cell support and separator allowing for the discharged battery suspension to
be easily replaced once exhausted.
1.INTRODUCTION
The first clinical cardiac pacemaker was implanted in 1958 by Ake Senning. However, the
electrochemical power supply only lasted a few hours. This groundbreaking advancement inspired
collaborative effort to overcome the challenge to produce a reliable power supply for medical implants
[1]. When considering the design of batteries for implantable medical devices, additional challenges
arise when choosing chemical components, electric cell design, and most importantly how to recharge
the electrochemical cell. Today, various lithium ion technologies have significantly extended the
battery lifetime however, replacing the battery requires a surgical procedure that may result in further
complications [2, 3]. We propose an alternative approach to recharge the battery through the re-
development of Leclanché electrodes. These battery components may be injected through a needle,
simplifying the medical procedure and decreasing the risk to the patient.
In pursuit of this idea we have used the Leclanché type chemistry to demonstrate the proof of
concept of our liquid battery because this well characterized chemistry boasts high energy density and
reliability [4, 5]. In addition to favorable electrical characteristics, we have also evaluated the current
state of battery and fuel cell technology. We believe that repurposing established battery technology
exemplifies creativity, broadens the landscape for electrochemical possibilities, and serves to establish
2
new methods to further electro-mobility [6-8]. This well-established chemistry has served as a reliable
power source for more than a century and high performance commercial batteries have been utilized in
a variety of applications since the 1970's [9]. Recent studies regarding electrolyte composition,
conductive additives, and new applications, such as flow batteries, demonstrate the relevance and
applicability of the original Leclanché cell [10-12].
The first zinc carbon primary is known as the Leclanché cell and was developed over a century
ago by a French telegraph engineer, Georges Leclanché [4]. Originally, these primary cells were
formulated with zinc anode and manganese oxide cathode in combination with various conductive
carbon powders saturated with an aqueous electrolyte. Later, Carl Gassner reconstructed these
batteries with powdered or paste electrodes to develop the first "dry cell" battery. Dry cells are ideal
for mobile applications that require intermittent discharge, such as flashlights or radios, where leakage
of liquid cells is detrimental. Today, the commercial Leclanché is housed in a cylindrical zinc
container that encapsulates the separator, manganese based cathode and a carbon rod current collector.
Often the mobile electrolytes, ammonium chloride and zinc chloride, are suspended in a separator such
as: porcelain, paper, or gelling agents such as starch. Modest cost, simplicity of production, and
favorable discharge characteristics have made this chemistry desirable for a variety of applications and
has proven to occupy a significant portion of the market [9,13].
In this research, the original Leclanché chemistry and battery construction has been simplified
to be used in internal medical devices. The electrodes have been suspended in an agar hydrogel to
simplify the construction and replacement/recharging procedure of spent batteries. While under
electrochemical load the hydrogel provides mechanical support allowing the battery to sample many
spatial confirmations to maintain energy density. Hydrogels are characterized by polymeric networks
than can absorb a significant amount of water and become swollen granting new physical properties,
such as physical stability and ionic conductivity [14]. Hydrogels as energy storage materials have seen
recent interest and have been used in a variety of electronic devices such as capacitors, sensors, and
scaffolds for catalysis [15-18]. Excitingly, there has been some research and characterization of the
zinc MnO2 alkaline cell in gel form, but the application and reproducibility has yet to be demonstrated
[19]. Additionally, our hydrogel battery features a simple construction due to the fact that the hydrogel
also serves as the separator allowing the entire battery hydrogel to be constructed in a simple
container. Gel based separators are desirable because of high ionic conductivity, ability to form a
variety of shapes, and ease of fabrication [20]. Various carbohydrates binders and electro-spun
modified agar have been reported for use as a battery separator however, we are not aware of literature
reports using unmodified agar as a battery separator [21]. Due the instability of agar, we demonstrate
the recharging of these electrochemical cells by treating the exhausted agar electrode matrix with
warm aqueous acid to remove and replace the cell with new active material [22].
2. EXPERIMENTAL
2.1 Design of Experiment
The experimental setup has been designed to calculate the discharge rate of a given battery. The
system can acquire analog data from a maximum of 16 data inputs in parallel and log it appropriately in
a comma-separated values (.CSV) format (Figure 1).
3
Figure 1: A single-battery test setup
To calculate the discharge rate, a given battery is connected to a 10 kΩ resistor and is discharged
across it until empty. Voltage across the resistor is measured using an acquisition card (more details in
next section) at 30-minute intervals. Plotting the voltage values of the battery at different points of time
shows the required discharge trend line and helps compare different batteries.
Once set, the experiment can simultaneously log data from a variety of batteries for weeks on
end without intervention or supervision. The .CSV file generated by the system can then be used in a
standard program like MS Excel or MATLAB to generate data plots.
2.2 Data Acquisition System
The data acquisition system consists of an acquisition card and a data logger. The acquisition
card used here is an Arduino ATMega 2560, which is an electronic microcontroller that can be
programmed to acquire analog data from a maximum of 16 channels at any given interval. Signal range
is from 0-5 V, and resolution is 4.9 mV [23].
The acquired signal is sent via USB to a data logger, which is a Raspberry Pi 3 Model B running
Raspbian. It is a Linux-based, single-board mini-computer. It can provide a GUI interface output to a
computer screen via HDMI, and peripherals like a keyboard and mouse can be plugged into it (Figure
2).
The data logger runs a Linux-based version of Arduino. The acquisition card conditions the
signal into a series of comma-separated values, which can be logged on to the Arduino Serial Monitor
screen. These can be saved directly as a .CSV file.
4
Figure 2: Schematic of data acquisition system
2.3 Gel Leclanché Cell Construction
An Agar-Leclanché-graphite cell is constructed in a 20 ml vial by mass or volume of
components. The electrolyte hydrogel is used at various points in the method and consists of the
following by weight: 26% NH4Cl (RPI, 99.5%), 8.8% ZnCl2 (Sigma, 98%), and 65.2% DI H2O. Anode
hydrogel is constructed by heating 4 g Zn powder (Sigma, 10 𝝻m, 98%), 2 g graphite powder (Sigma,
20 𝝻m), and 6 ml electrolyte hydrogel until dissolved. The hydrogel is vortex mixed, and 3 g of the
anode hydrogel is transferred to a new vial with 3 g of a warm 1% w/v agar (RPI) electrolyte hydrogel.
The warm agar separator hydrogel (3 g) was layered on the anode hydrogel followed by 3 g of cathode
hydrogel, which consists of MnO2 8 g (Sigma, 99%), graphite 4 g, and 15 ml electrolyte hydrogel. 5 mm
carbon electrodes (Eisco) are coated in paraffin (Gulf Wax) to span the cathode hydrogel and are
submerged in the fuel cell for characterization (Figure 3).
5
Figure 3: Gel Leclanché-graphite cell constructed in a 20 ml vial.
2.4 Gel Leclanché Cells with Various Conductive Additives
After the initial characteristics and experimental setup of the Zn agar cells were defined, the
quantities of Zn and conductive additive were varied using a similar cell design. The anode hydrogel
consists of 2 g saturated conductive additive electrolyte hydrogel (1.5 g acetylene black or graphite and
20 ml electrolyte hydrogel), 1 g agar electrolyte hydrogel (2.5% w/v) and 0.25 g, 0.5 g, or 0.75 g Zn
powder. Warm agar electrolyte separator (3 g), as discussed previously, is layered on the anode and
allowed to cool. The warm cathode hydrogel is layered on the separator and consists of 2 g of the
following mixture: MnO2 3 g, 6 g acetylene black electrolyte hydrogel, and 3 g 2.5% w/v agar electrolyte
hydrogel. 0.5 mm carbon electrodes coated in paraffin to span the cathode are used as current collectors
(Figure 3).
2.5 Closed Leclanché Gel Cell Construction and Purging
The gel Leclanché cell is constructed in an 11 ml glass chamber equipped with purge/fill ports
and current collection ports (Figure 4). The gel anode and cathode consist of 0.5% w/v agar dissolved
in 3 ml aqueous electrolyte and ethylene glycol hydrogel (Sigma) (1:1) with 0.1 g acetylene black and
0.25 g MnO2 or Zn. The gel separator consists of 0.25% w/v agar dissolved in aqueous electrolyte and
ethylene glycol hydrogel (1:1). The cathode hydrogel was layered first, followed by the separator and,
finally, the anode hydrogel. The original gel hydrogel was removed by treating the cell with warm acidic
water (pH ~1) until clean, and the next Leclanché gel components were layered as previously described.
6
Figure 4: Closed gel Leclanché cell construction in an 11 ml glass chamber with ports for
purge/filling and ports for current collection.
3. RESULTS AND DISCUSSION
3.1 Gel Leclanché Cell: Proof of Concept
Initially, the standard Leclanché chemistry was suspended in an aqueous agar hydrogel to
explore the feasibility of the cell design and characterize the overall cell electrochemistry. Discharge
characteristics of three trials are shown in Figure 5. Each cell was characterized by recording the
voltage under a 10 kΩ load every 30 min until the cell potential reached a constant reading (~0.1V) .
This resistor was chosen because it allowed the experiments to be conducted on a reasonable timeline
that yielded data which compares some of the important battery components. Many battery standards
use resistors in the ohm range, however in the interest of time our battery needed a larger resistor to
generate complete discharge plots for each varied component [24].
The average open circuit voltage (OCV) of three electrochemical cells on construction (~1.4V)
was larger than initial closed circuit voltage (CCV) (~0.75V) which is due to the electrochemical load
imposed by the resistor. Typically the CCV approaches the OCV and can be optimized based on the
electrolyte, cell construction, and separator which will be discussed later. The general discharge
characteristics displayed by the Zn-graphite cells are similar to previous reports in the literature, however
the timescales may differ depending on the load tested [25-26]. There is an initial rapid drop in potential
(0-50000 seconds) followed by a period of stabilization (50000-240000 seconds) until the final rapid
discharge (Figure 5).
7
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
V
/
e
g
a
t
l
o
V
Zn-GR
0
200000
400000
600000
Time / s
Figure 5: Gel Leclanché cell continuous discharge data across a 10 kΩ resistor with GR as the
conductive additive. Average of three trials.
3.2 Gel Leclanché Cell: Different Conductive Additives
Zinc agar batteries constructed with acetylene black as the conducting additive display slightly
enhanced characteristics under continuous discharge. Initial OCV's, irrespective of the zinc quantity, in
each cell is greater in cells constructed with acetylene black than cells constructed with graphite
(~1.15V vs. ~0.75V). Additionally, the acetylene black cell produces a nearly constant CCV (~1.05V)
from time zero until about 500000 seconds in cells constructed with 0.5g and 0.75g zinc powder.
Beyond 0.5g zinc powder there is not a significant difference in the cell lifetime however, increasing
the MnO2 in combination with increasing zinc may deliver longer cell lifetimes. It has been previously
reported that the specific sources of MnO2 can have a profound effect on the cell performance which
will be further optimized [26-7]
8
Zn 0.25g AB
Zn .5 AB
Zn 0.75g AB
V
e
g
a
t
l
o
V
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
5000
10000
Time s
15000
20000
Figure 6: Gel Leclanché cell continuous discharge data across a 10 kΩ resistor with AB as the
conductive additive, 0.5 g MnO2, and variable zinc 0.25 g (
) 0.5 g ( ) 0.75 g (
).
Variable Zinc-GR cells are constructed in a similar manner to the variable Zn-AB cells and
display similar discharge characteristics to the initial Zinc-GR primary cells. On construction OCV's
are dependent on the quantity of Zinc. Above 0.5g Zn powder, it appears that the OCV is not affected
when the MnO2 is held constant (0.5 g) and approach ~1.2V (Figure 7). Discharge characteristics of
GR cells do not appear to rely heavily on the quantity of Zn and display a more rapid discharge rate
when compared to AB cells. The terminal voltage appears to increase with the amount of Zn used in
construction and is approached more gradually when compared to the AB cells. It must also be noted
that the potential is more variable in GR based cells than the AB based cells.
9
Zn .25g_GR
Zn .5g_GR
Zn .75g_GR
V
e
g
a
t
l
o
V
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
2000
4000
6000
8000
10000
Time s
Figure 7: Gel Leclanché cell continuous discharge data across a 10 kΩ resistor with GR as the
conductive additive, 0.5 g MnO2, and variable zinc 0.25 g (
) 0.5 g ( ) 0.75 g (
).
AB based cells display greater voltage stability and maintains higher voltage for prolonged
time. This suggests that AB cells may better serve situations where higher voltage is required over
greater time spans, while GR cells are better suited for rapid discharge over shorter time spans. It has
previously been reported that the conductive additive can be used to tune Leclanché type cell discharge
for specific applications and our data showcases the reproducibility of the agar based system [28-29].
The conductive additive also influences the energy density (JL^-1, total mAh, and the
coulombic capacity (Ah) of cells constructed with different conductive additives. Although the
quantity of active material does not change and therefore, theoretically the energy density should
remain the same regardless of the conductive additive. We observed differences in the energy density
as a function of the conductive additive and quantity of zinc (Figure 8). As expected, the energy
density increased with increasing zinc.
1
-
^
L
J
y
t
i
s
n
e
D
y
g
r
e
n
E
5000
4000
3000
2000
1000
0
0
10
Zn GR
Zn AB
0.2
0.4
Zinc g
0.6
0.8
Figure 8. Energy density of cells made with variable amounts of zinc and different conductive
additives (AB and GR)
However, the conductive additive also appears to influence the total mAh (Figure 9). At 0.25g
zinc cells with AB and GR have similar energy densities (AB 102.14 JL^-1, GR 101.25 JL^-1) (Figure
9). Beyond 0.5 g zinc the difference in energy densities between cells made with AB and GR is large
and appear to be great in cells made with AB.
11
h
A
m
18
16
14
12
10
8
6
4
2
0
Zn GR
Zn AB
0
0.2
0.4
0.6
0.8
Zinc g
Figure 9. Total milliamp hours of cells made with variable amounts of zinc and different conductive
additives (AB and GR)
With regards to mAh, cells constructed with variable quantities of zinc and either AB or GR a
similar trend to energy density is observed, greater zinc increases mAh with AB performing better than
GR at larger quantities of zinc (Figure 9). It must be noted that the total mAh of cells made with 0.25g
zinc and GR (1.24 mAh) is larger than cells made with AB (0.52 mAh). Further studies should be done
to determine the validity of this finding however, it may suggest that, for applications where the size of
the battery must be extremely small, GR should be used as the conductive additive.
Total Ah were calculated for theoretical cells and compared to cells made with AB and GR at
different quantities of Zn (Figure 10). Coulombic capacity is a way to relate the potential
electrochemical energy of a material to the harvestable electrochemical energy of the system as
reported in reference 9. Here we show that the constructed cells are able to perform near 1% of their
theoretical coulombic efficiency. It is well known that primary cells do not reach the theoretical
coulombic efficiency, but can be improved upon with different conductive additives, cell design, and
separators.
12
AB CELLS
GR CELLS
h
A
y
t
i
c
a
p
a
C
l
a
c
i
t
e
r
o
e
h
T
%
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
0.2
0.4
0.6
0.8
Zinc / g
Figure 10. Coulombic efficiency as a % of theoretical coulombic capacity.
The ideal conductive additive should simply serve as a medium for electron transportation
without influencing the chemical reactions of the cell. There are obvious trends in the OCV, CCV,
discharge rates, energy density, and mAh depending on what conductive additive is used. There is
considerable literature reporting physical characteristics of conducting carbon powders when used in
their compressed form and limited literature regarding their use in other specialized applications
[30,31]. It is generally agreed that the particle size and size range, morphology, source, and pressure
used to compress powder influence the conductivity of the material and in general, smaller and more
tightly packed particles conduct electrons better [32-34]. The GR particles used were <20µm in
diameter and the AB particles were 25-45nm in diameter and it that the considerably smaller size of
AB particles is responsible for the greater electrical performance of cells. Further studies cannot
discount the influence of other physical characteristics such as adsorption, solubility, and chemical
structure of the conductive additive to influence the electrochemical performance of the cell.
When comparing our battery to commercially available batteries and other reports in the
literature we find that the OCV general discharge trends are similar and that notable differences are
obvious. Differences displayed by our batteries include high internal resistance (Mohm range) as
compared to other systems that boast ohm range. We expect to optimize our hydrogel as to decrease
internal resistance and increase battery performance. Our energy density, and coulombic capacity is
about (0.5-1%) of other reports, however we have not saturated our hydrogel with active material and
13
believe this will be improved upon in the near future [35,36]. Our batteries, which contain the same
active chemicals as other zinc-MnO2 systems, are performing similar chemical reactions to those
already established, however in a very different construction. We can be confident that our novel
construction has not significantly altered the feasibility of this chemistry.
3.3 Gel Leclanché Cell Purge and Refill
To demonstrate the applicability of this method a closed glass cell was constructed to record
fill and purge voltage of gel Leclanché cell. Ethylene glycol was added to make the purge and refill
process easier and the influence on electrochemical performance in under investigation. Discharge data
for gel based Leclanché cell under 10kΩ continuous load (Figure 11). Upon construction a potential of
0.15V is observed, a peak potential observed at 0.35V after 125000 seconds, and a stabilized potential
of 0.29V is seen from 150000 seconds to 350000 seconds. At 400000 seconds the cell was purged and
refilled and the potential again peaked a 0.28V.
Fill 1
Fill 2
V
l
a
i
t
n
e
t
o
P
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0
100000 200000 300000 400000 500000 600000
Figure 11. Fill and purge voltage of a closed gel Leclanché cell under 10 kΩ load with AB conductive
additive. The cell was purged and refilled at 400000 seconds.
Time s
4. CONCLUSIONS
The traditional chemistry of the Leclanché cell has been re-constructed in aqueous hydrogel for
use in implantable medical devices or other hard access areas. In this work the standard Leclanché
chemistry has been suspended in agar hydrogel to immobilize the anode and cathode as to prevent
14
spilling in addition to serving as the battery separator. These cells are easily constructed and utilize
components (carbon electrode, glass vials, agar separators) that are inexpensive and common in many
research laboratories. Construction of these fuel cells in a single vial exemplifies the feasibility of this
method for hard to access applications, such as a pacemaker. These implantable cells will be tuned to
increase cell lifetime, voltage, and discharge characteristics that meet the energetic demands of a specific
device. The role of the conductive additive plays a significant role in cell potential in addition to the
discharge characteristics and will be exploited to suit particular energy demands. Fine tuning the removal
and reconstruction of the spent agar electrode hydrogel is being refined. This preliminary data highlights
the potential of this methodology as a technically simplify the procedure to replace exhausted
electroactive material in devices where access is limited.
ACKNOWLEDGMENTS
The authors would like to acknowledge Dr. Ronald Weigel, Head of Department of Surgery and
Professor of Biochemistry, for his support in securing lab space and providing financial assistance. This
support has been instrumental in making this research a reality.
This research is funded, in part, by the Mid-America Transportation Center via a grant from the U.S.
Department of Transportation's University Transportation Centers Program, and this support is
gratefully acknowledged. The contents reflect the views of the authors, who are responsible for the facts
and the accuracy of the information presented herein, and are not necessarily representative of the
sponsoring agencies.
References
1. W. Greatbatch, J.Cardiovasc. Nurs., 5 (1991) 80.
2. V. Sarma Mallela, V. Ilankurnaran, N. Srinivasa Rao, Indian Pacing and Electrophysiol. J., 4
(2004) 201.
3. M. Kotsakou, I. Kioumis, G. Lazaridis, G. Pitsiou, S. Lampaki, A. Papaiwannou, A. Karavergou, K.
Tsakiridis, N. Katsikogiannis, I. Karapantzos, C. Karapantzou, S. Baka, I. Mpoukovinas, V.
Karavasilis, A. Rapti, G. Trakada, A. Zissimopoulos, K. Zarogoulidis, P. Zaro. Goulidis, Ann. Transl.
Med., 3 (2015) 45.
4. G.Leclanché, Notes sur l'emploi des piles électriques en télégraphie, pile constante au peroxyde de
manganèse à un seul liquide, Impr. de Hennuyer et fils, (1867) Paris, France
5. J. Larcin, Ph.D.Thesis: Chemical and electrochemical studies of Leclanché cells, (1991) Middlesex
University, London, United Kingdom.
6. U. Stimming, P. Bele, J. Fuel Cells, 16 (2016) 2.
7. U. Stimming, P. Bele, J. Fuel Cells, 18 (2018) 3.
8. U. Stimming, P. Bele, J. Fuel Cells, 17 (2017) 2.
9. T. Reddy, D. Linden, Linden's Handbook of Batteries, McGraw Hill, (2010) New York, United
States.
10. N.H. Khalid, Y.M. Baba Ismail, A.A. Mohamad, J. Power Sources, 176 (2008) 393.
11. L. Ank T.S. Zhao, X.L. Zhou, X.H. Yan, C.Y. Jung, J. Power Sources, 275 ( 2015) 831.
12. G. Li, M.A. Mezaal, R. Zhang, K. Zhang, L. Lei, J. Fuel Cells, 3 (2016) 395.
15
13. The Leclanché Cell, can be found under
http://www.che.uc.edu/jensen/w.%20b.%20jensen/Museum
%20Notes/24.%20The%20Leclanch%C3%A9%20Cell.pdf (2014).
14. E. Ahmed, J. Adv. Res., 6 (2013) 105.
15. R. Sing, B. Veer, Chemistry Select, 3 (2018) 1309.
16. B. Crulhas, N. Ramos, C. Basso, V. Costa, G. Castro, V. Pedrosa, Int. J. Electrochem. Sci., 9
(2014) 7596.
17. M. Chen, R. Wang, S. Cai, P. Mei, X. Yan, Y. Jiang, Y. Zhang, W. Xiao, H. Tang, Int. J.
Electrochem. Sci., 13 (2018) 2401.
18. M. Rosi, F. Iskandar, M. Abdullah, and Khairurrijal, Int. J. Electrochem. Sci., 9 (2014) 4251.
19. Y.M.B. Ismail, H.Haliman, A.A. Mohamad, Int. J. Electrochem. Sci., 7 (2012) 3555.
20. A. Mainar, E. Iruin, L. Colmenares, A. Kvasha, A. Kvasha, I. de Meatze, M. Bengoechea, O.
Leonet, I. Boyano, Z. Zhang, J. Alberto Blazquez, J.of Energy Storage, 15 (2018) 304.
21. J-M. Kim, C. Kim, S. Yoo, J-H. Kim, J-H. Kim, J-M. Lim, S. Park, S-Y. Lee. J. Mater. Chem.,3
(2015) 10687.
22. D. McHugh, Food and Agriculture Organization Of The United Nations, 288 (1987) 1.
23. AnalogRead(), can be found under
https://www.arduino.cc/reference/en/language/functions/analog-io/analogread/, (2017).
24. S. Wicelinski, M. Babiak, R. Runkles. American National Standard For Portable Primary Cells
and Batteries With Aqueous Electrolyte, Part 1 (2001) 1.
25. Z. Rogulski, A. Czerwiński, J. Solid State Electrochem., 7 (2001), 118.
26. J. O. Bockris, B. Conway, E. Yeager, R. White. Comprehensive Treatise of Electrochemistry,
Plenum Press, (1981) New York, United States.
27. G. Li, M.A. Mezaal, R. Zhang, K. Zhang, L. Lei, J. Fuel Cells, 3 (2016) 395.
28. G. Vinal, Primary Batteries, Wiley, (1950) New York, United States.
29. G. Benson, J. Gluck, C. Kaufmann, J. Electrochem. Soc., 90 (1946) 441.
30. E. B. Sebok, R. L. Taylor, Encyclopedia of Materials: Science and Technology, Second Edition
(2001) 902
31. G. Singh, M. Esmaeilpour, A. Ratner, Fuel, 246 (2019) 108
32. A. Celzard, J.F. Mareche, F. Payot, G. Furdin, Carbon, 40 (2002) 2801.
33. X. Zhang, Y. Cui, Z. Lv, M. Li, S. Ma, Z. Cui, Q. Kong, Int. J. Electrochem. Sci., 6 (2011) 6063.
34. B. Marinho, M. Ghislandi, E. Tkalya, C. Koning, G. de With, Powder Technology, 221 (2012) 351
35. Energizer, Alkaline Manganese Dioxide Handbook and Application Manual,
http://data.energizer.com/pdfs/alkaline_appman.pdf, (2018)
36. G.Ghiurcan, C-C. Liu, A. Webber, F. Freddrix, J. Electrochem. Sci., 150 (2003) 922.
Figure Captions
Figure 1: A single-battery test setup
Figure 2: Schematic of data acquisition system
Figure 3: Gel Leclanché-graphite cell constructed in a 20 ml vial.
16
) 0.5 g ( ) 0.75 g (
Figure 4: Closed gel Leclanché cell construction in an 11 ml glass chamber with ports for purging/filling
and ports for current collection.
Figure 5: Gel Leclanché cell continuous discharge data across 10 kΩ resistor with GR as conductive
additive. Average of three trials.
Figure 6: Gel Leclanché cell continuous discharge data across a 10 kΩ resistor with AB as the conductive
additive, 0.5 g MnO2, and variable zinc 0.25 g (
Figure 7: Gel Leclanché cell continuous discharge data across a 10 kΩ resistor with GR as the conductive
additive, 0.5 g MnO2, and variable zinc 0.25 g (
Figure 8. Energy density of cells made with variable amounts of zinc and different conductive additives
(AB and GR).
Figure 9. Total milliamp hours of cells made with variable amounts of zinc and different conductive
additives (AB and GR)
Figure 10. Coulombic efficiency as a % of theoretical coulombic capacity.
Figure 11. Fill and purge voltage of a closed gel Leclanché cell under 10 kΩ load with AB conductive
additive. The cell was purged and refilled at 400000 seconds.
).
) 0.5 g ( ) 0.75 g (
).
|
1906.01089 | 1 | 1906 | 2019-03-03T05:39:53 | Screen-printed and spray coated graphene-based RFID transponders | [
"physics.app-ph"
] | We report Ultra-High-Frequency (UHF, 800MHz-1GHz) Radio Frequency Identification (RFID) transponders consisting of printed dipole antennas combined with RFID microchips. These are fabricated on Kapton via screen printing and on paper via spray coating, using inks obtained via microfluidization of graphite. We introduce a hybrid antenna structure, where an Al loop (small compared to the overall size of the antenna) is connected to a microchip with the double function of matching the impedances of antenna and microchip and avoiding bonding between exfoliated graphite and chip. The transponders have reading distance~11m at UHF RFID frequencies, larger than previously reported for graphene-based RFID and comparable with commercial transponders based on metallic antennas. | physics.app-ph | physics |
Screen-printed and spray coated graphene-based RFID transponders
K. Jaakkola1, V. Ermolov1, P. G. Karagiannidis2, S. A. Hodge2, L. Lombardi2,
X. Zhang2, R. Grenman1, H. Sandberg1, A. Lombardo2, A. C. Ferrari2
1VTT Technical Research Centre of Finland, Espoo 02044, Finland and
2Cambridge Graphene Centre, 9 JJ Thomson Avenue,
University of Cambridge, Cambridge, CB3 0FA, UK
We report Ultra-High-Frequency (UHF, 800MHz-1GHz) Radio Frequency Identification (RFID)
transponders consisting of printed dipole antennas combined with RFID microchips. These are
fabricated on Kapton via screen printing and on paper via spray coating, using inks obtained via
microfluidization of graphite. We introduce a hybrid antenna structure, where an Al loop (small
compared to the overall size of the antenna) is connected to a microchip with the double function
of matching the impedances of antenna and microchip and avoiding bonding between exfoliated
graphite and chip. The transponders have reading distance∼11m at UHF RFID frequencies, larger
than previously reported for graphene-based RFID and comparable with commercial transponders
based on metallic antennas.
INTRODUCTION
Radio frequency identification (RFID) is a ubiquitous
technology[1], with applications in access control[1], con-
tactless payment[2], electronic passports[1], supply chain
management[3], healthcare[4], food packaging[5] and an-
imal identification[6].
It is also the cornerstone of the
so-called "Internet of Things" (IoT)[7], where people and
devices are seamlessly integrated in a decentralized com-
mon set of resources, creating a convergence of the phys-
ical realm with human-made virtual environments[8].
Within IoT, every "thing" is connected[8], and the al-
ready widespread RFID technology is likely to become
even more ubiquitous, combining additional functionali-
ties such as sensing[9, 10] and energy harvesting[11, 12].
The basic elements of a typical RFID system are tags[1]
and readers[1], which exchange information via radio
waves[1]. Tags comprise integrated circuits containing
a memory to store the tag identity (ID) and the read-
ing/writing circuitry[1]. Tags communicate with the
reader via a suitable antenna, which typically has the
double role of drawing energy from the reader to ener-
gize the integrated circuit[1], and exchange data with
the reader[1]. RFID offers advantages over other iden-
tification technologies, such as barcodes[1], since an RF
tag does not need to be in sight of the reader and can,
therefore, be embedded in objects[1]. Also, RFID allows
simultaneous reading of several tags[1], making the iden-
tification process very fast, typically a few milliseconds
for passive (i.e. powered by the reader through the an-
tenna) RFID tags[1] and even shorter for active ones (i.e.
powered by a battery)[1].
RFID tags should combine mechanical robustness (e.g.
to tolerate vibrations)[1],
light weight (typically<10g)
[1], compact dimensions (∼cm)[1], reliability[1] and low
cost(<0.05$)[13]. Mechanical flexibility might also be
required (especially for IoT[14]), adding specific chal-
lenges not present on rigid systems, such as shifts in res-
onant frequency[14], and return loss (i.e. reflected power
loss caused by antenna input impedance mismatch)[14]
and changes in effective capacitance (i.e.
the ratio of
the change in charge to the corresponding change in
potential)[14], radiation pattern distortion[14] and gain
degradation[14]. Different operational scenarios also in-
troduce additional complexity, e.g. proximity to tissues
in wearable devices[14].
Large volume (several millions of units)[13] and low
cost (<0.05$ per unit)[13] manufacturability is essen-
tial, as it is expected that over one trillion IoT de-
vices will be deployed by 2025[15, 16]. The most com-
mon RFID tags, consisting of a planar electric dipole
antenna[17 -- 19], are fabricated from a metallized plas-
tic foil by acidic etching[1]. However, this process re-
sults in metal waste[20], which is also environmentally
harmful[20].
Printing is a promising alternative[14], as it combines
high volume production (e.g. an industrial screen printer
can print areas> 3m×6m in a single pass[21]) and, at the
same time, avoids chemical etching and material wastage.
Ag inks are typically used for printed RFID[22, 23], since
they have high conductivity∼ 106S/m[24]. However, the
Ag cost is very high (∼800-1000$/kg)[25]. Printed Ag
films have limited flexibility, breaking∼ 75% strain[26]
and resistance increase up to∼ 15% upon bending[26].
Moreover, they can be toxic and carcinogenic[27].
Printed graphene
layers
and mechanical
can be dispersed in solvents
can be an alternative
to printed metals[28] as graphene combines good
robustness[28].
conductivity[28]
Graphene
(such as
NMP[29] or water[29]), doped[29] or functionalized[29].
The surface resistivity of single layer graphene (SLG) at
radio (300KHz to 300MHz) and microwave (300MHz to
300GHz) frequencies is higher than metals[30], resulting
in losses[30] that prevent its use in antennas with high
(> 90%) efficiency (i.e. ratio between power irradiated
by the antenna and power supplied)[30]. The SLG
conductivity can be tuned by field effect[31]. However,
the changes are mostly in the real part[30], while in the
imaginary part these are small up to∼100GHz[32, 33],
resulting in limited reconfigurability (i.e.
tunability of
radiation frequency, pattern or polarization)[34].
Thick (> 1µm) exfoliated graphite films, consisting
of few-layer graphene (FLG) flakes, can overcome such
limitations, showing sheet resistances RS < 2Ω/(cid:3))[35],
corresponding to conductivities>104S/m[35]. These can
also be deposited over large (m2) areas by screen printing
or spray coating.
Screen printing is a common industrial technique for
roll-to-roll patterned deposition[21]. Typical formula-
tions of screen inks contain a conductive filler, such as
Ag particles[26], and insulating additives (e.g. stabilizers
and binders)[36], at a total concentration>100g/L[36].
Of this,>60g/L consists of the conductive filler needed
to achieve sufficiently high (> 106S/m) conductivities[26,
37].
Spray coating is also suitable for roll-to-roll
production[38]. To the best of our knowledge, there
are no reports on spray coated graphene-based antennas.
However, spray coated FLG films with similar specifica-
tions to those needed for RFID antennas (RS ∼6Ω/(cid:3) and
thickness∼8µm) were reported for use in Electromagnetic
Interference (EMI) shielding[39].
Their
A number of antennas based on solution-processed
FLG films have been reported[40 -- 45].
re-
duced performance in gain and radiation efficiency com-
pared to metallic antennas (typically over one order of
magnitude[34]) is compensated by other functionalities,
such as mechanical flexibility[45]. RFID transponders,
based on FLG film antennas combined with RFID inte-
grated circuits, were demonstrated[41, 42, 45], showing
typical reading distance up to∼9m[44]. This is smaller
than commercial RFIDs, providing>10m[41, 42, 45].
The input impedance of a typical RFID microchip
at operating frequencies (865-868 MHz in Europe and
915MHz in US[1]) is capacitive[1, 48], with a real part
lower than the absolute value of the reactance[1, 48].
Thus, to match the impedance conjugately, i.e. to ensure
that both microchip and antenna are electrically compat-
ible with each other, the impedance of the antenna should
be the complex conjugate to that of the microchip at the
frequency of operation[22]. A two-branch dipole antenna
might not have such a point on its impedance curve be-
cause of design[47], dimensions[47] or materials used[47].
The conjugate impedance match between microchip and
antenna can be achieved by forming a loop inductor par-
allel to the feeding point on the antenna conductor[19].
Here, we report RFID transponders consisting of
graphitic antennas either screen printed on Kapton or
sprayed on paper, coupled with RFID chip through Al
inductive loops, ensuring impedance matching, i.e. that
the impedance of the antenna is the complex conjugate
impedance of the microchip at the frequency of opera-
tion. The Al loop is significantly smaller than the overall
antennas size, therefore minimizing use of metal and not
compromising the flexibility of the overall transponder.
RFID microchip
2
Jsurf [A/m]
Jsurf [A/m]
5
3.2
2.1
1.4
0.9
0.6
0.4
0.3
0.2
0.1
FIG. 1. a)Antenna with FLG inductor. b) Simulated current
distribution. Jsurf is the surface current density in A/m
Al loop
RFID microchip
Jsurf [A/m]
Jsurf [A/m]
5
3.2
2.1
1.4
0.9
0.6
0.4
0.3
0.2
0.1
FIG. 2. (a) Hybrid antenna with metal inductor. The larger
structure is the printed FLG, while the inner loop is the Al
inductor. b) Simulated current distribution. Jsurf is the sur-
face current density in A/m.
These have reading distances up to∼11m in the relevant
UHF RFID bands: 865.6-867.6MHz (Europe) and 902-
928 MHz (USA and Japan), larger than graphene-based
RFID tag previously reported[41, 42, 44, 45] and compa-
rable with commercial RFID transponders[46].
RESULTS AND DISCUSSION
The antennas are designed using the electromagnetic
simulation software High Frequency Structure Simula-
tor (HFSS) 15 (Ansys Inc. USA), assuming RS ∼
3Ω/(cid:3), as typical
for dried FLG films produced by
microfluidizaton[35]. The two main parameters of a
transponder antenna are the input impedance[1], to
match the antenna with the transponder microchip, and
the radiation efficiency, defined as the ratio of power ra-
diated by the antenna and power supplied[47].
We use a Impinj Monza R6 UHF RFID microchip, with
a 96 bits memory. This employs unregulated codes and
is compatible with a wide range of tag form factors[48].
The input impedance is 16-j139Ω at 915MHz[48]. This
is prevalently capacitive, with a real part lower than
the absolute value of the reactance. Thus, to match
the impedances conjugately the antenna should have an
impedance Zant=16+j139Ω at the same frequency, i.e.
the antenna should be sufficiently inductive with a low
real part of the impedance. In order to achieve this, a
parallel inductor in the dipole antenna is implemented as
a opening on the conductor[41, 45].
We also introduce a hybrid structure in which we com-
bine the printed FLG antenna with an Al inductive loop
for impedance matching. The Al loop is significantly
smaller than the overall size of the transponder, therefore
minimizes the use of metals and does not compromise
flexibility. The loop forms inductive coupling between
microchip and the antenna FLG conductor. Thus, no
direct connection of microchip to FLG film is required.
We design and simulate FLG antennas using both FLG
inductors and Al inductive loops. Both designs are made
for the same FLG RS ∼ 3Ω/(cid:3). The optimized outer di-
mensions of the antenna to work at 915 MHz with the
FLG inductive loop, shown in Fig. 1, are 114mm×34mm
and the dimensions of the opening are 13.3mm×10.1mm.
The outer dimensions of the hybrid antenna, Fig.2, are
the same. The dimensions of the upper opening of the
antenna are 18.3mm×6mm, and those of the lower open-
ing are 18.3mm×20mm.
The main tunable parameters of the antennas, opti-
mized by simulations, are the circumference of the loop
and the length of the antenna. The first determines the
input reactance of the antenna[41, 45], while the latter
determines the radiation resistance, i.e.
the resistance
caused by the radiation of electromagnetic waves from
the antenna[49].
In the hybrid antenna, a rectangular
opening is added, rather than a loop, in order to mini-
mize Eddy currents induced by the inductive loop, since
these would increase losses and decrease radiation effi-
ciency. Shape and dimensions of the opening are cho-
sen to minimize Eddy currents without significantly af-
fecting antenna conductivity. The inductive loop, with
14mm×6mm outer dimensions, is made of 0.8mm wide
and 9µm thick Al and is shown in Fig.2.
Table I summarizes the simulated parameters at
915MHz:
input impedance Zant, attenuation due to
impedance mismatch LZ, radiation efficiency η, direc-
tivity (i.e. the ration between the maximum radiation
intensity in the main beam and the average radiation in-
tensity over all space) Dtag and calculated read range, i.e
the calculated maximum distance that the RFID tag can
be read, Rread. The attenuation due to the impedance
mismatch between antenna and microchip is be calcu-
lated from the impedances as[47]:
LZ = 1 − (Zant − Z ∗
IC)/(Zant + ZIC)2
(1)
where ZIC is the complex impedance of the microchip.
The forward-link (i.e. the transmission from the reader
)
.
u
a
(
y
t
i
s
n
e
n
t
.
I
G
D
D'
1250
1500
2000
1750
2500
Raman shift (cm -1)
2250
3
2D
2750
3000
FIG. 3. Representative Raman spectrum at 514nm for flakes
processed for 70 cycles.
Transponder
Zant (Ohms)
LZ
(dB)
η
(dB)
Dtag
(dB)
Rread
(m)
Antenna with FLG
inductive loop
Hybrid antenna
77.5 + j138
-2.3
-5.4
17.2 + j136
-0.0
-4.0
3.2
3.0
8.9
13.1
TABLE I. Simulated parameters of the two tag antennas in
Figs.1,2 at 915MHz
to the tag[50]) read range can be calculated as[50, 51]:
Rread = (c/4πf ) × (PtxEIRP DtagηtagLZ/PICsens)1/2
(2)
where c is the speed of light, f is the frequency, PtxEIRP is
the equivalent isotropically radiated power (i.e. the mea-
sured radiated power in a single direction) of the reader
device and PICsens is the read sensitivity of the microchip
(i.e. the minimum power required to activate the chip).
PtxEIRP =3.28W is the maximum allowed radiated power
of a UHF RFID reader as defined by the European regula-
tory environment for radio equipment and spectrum[52].
PICsens=-20dBm, as specified for the Impinj Monza R6
microchip by the manufacturer[48].
Table I indicates that the transponder with a hybrid
antenna has a longer read range (13.1m). This is due
to both better impedance match between antenna and
microchip, and higher radiation efficiency.
Based on the design optimized by simulations, FLG
antennas are fabricated either by screen printing or
spray coating. Two inks suitable for screen print-
ing and spray coating are formulated by adding dif-
ferent amounts of rheology modifiers after exfoliation
of graphite to tune the ink viscosity. Graphite flakes
(Timrex KS25) are added to deionized (DI) water
at a concentration∼100g/L and sodium deoxycholate
(∼5g/L). The mixture is processed using a microfluidizer
(a)
(b)
FIG. 4. SEM images of a)SP film on Kapton; b) SC film on
paper
)
m
(
e
g
n
a
r
d
a
e
R
9.5
9.0
8.5
8.0
7.5
7.0
6.5
6.0
5.5
800
Loop of FLG (simulated)
Loop of FLG (printed)
825
850
900
925
875
Frequency (MHz)
950
975
1000
FIG. 5. Simulated and measured read range as a function of
frequency for antenna with FLG inductive loop
(M-110P) at 207MPa for 70 cycles. One cycle is defined
as one pass of the liquid mixture through the interaction
chamber, where high shear rate (∼108s−1) is applied[35].
The exfoliated graphite flakes have a lateral size distri-
bution peaked at∼1µm and thickness∼12nm[35]. Fig.3
plots a representative Raman spectrum, acquired by a
)
m
(
e
g
n
a
r
d
a
e
R
4
14
13
12
11
10
9
8
7
6
5
800
Hybrid antenna (simulated)
Hybrid antenna (printed)
Hybrid antenna (sprayed)
825
850
875
900
925
950
975
1000
Frequency (MHz)
FIG. 6. Simulated and measured read range as a function of
frequency for hybrid antenna
Renishaw inVia at 514nm excitation, of the processed
material after microfluidization. The 2D peak consists
of two components (2D2, 2D1). Their intensity ra-
tio I(2D2)/I(2D1), changes from∼1.5 for the starting
graphite to∼1.2,
indicating exfoliation, but not com-
plete to SLG[35, 53]. Following microfluidization, car-
boxymethylcellulose (CMC) sodium salt is added at a
concentration∼10g/L to prepare a screen printable (SP)
ink and∼5g/L for the spray coating (SC) one. CMC
acts as rheology modifier imparting to the SP-ink a vis-
cosity ranging from∼570mPa s at 100s−1 to∼140mPa s
at 1000s−1, and to the SC one∼220mPA s at 100s−1 to
60mPa s at 1000s−1.
The SP-ink is used to form FLG films both for anten-
nas with FLG inductor and hybrid antennas on Kapton
using a screen printer (Kippax KPX-2012) equipped with
a 90 mesh per inch screen. These are then annealed at
265◦C for 10 minutes to remove the binder and increase
conductivity. RS of the printed antennas measured using
a four-point probe is∼5Ω/(cid:3), reduced to∼3Ω/(cid:3) after an-
nealing at 265◦C for 10mins. Fig.4a is a scanning electron
microscope (SEM) image of the printed film after anneal-
ing. Annealing at higher temperatures or for longer times
further reduces RS, however it causes delamination from
Kapton, making the antenna not usable.
The SC ink is used for hybrid antennas and
sprayed onto 3 substrates: 1) Polyethylene Naphthalate
(PEN), Q65HA-125µm); 2) multicoated matt art paper
(Lumisilk-120µm); 3) uncoated printing paper (Tesorp).
The substrate is cut into the shape of the simulated an-
tenna. SC is performed using a hand held manual spray
pen for∼5s, while moving over the antenna area, so that
ink covers the whole substrate, resulting in a self-standing
antenna. Air pressure is kept constant and the spraying
distance is∼20cm. The dry thickness of one pass is∼15-
18µm. A SEM image of a FLG film on paper is in Fig.4b.
The uncoated paper completely absorbs the water from
the ink and the samples are dried and flattened using a
hot press at∼130 ◦C. The samples are then calendered
using a cylinder press with one steel roller and one hard
rubber roller, generating a pressure∼80bar (∼36kN/m).
5
FIG. 7. Measured (solid line) and simulated (dashed line) directivity of antenna with FLG inductive loop in azimuth and
elevation plane
FIG. 8. Measured (solid line) and simulated (dashed line) directivity of hybrid antenna in azimuth and elevation plane
The compression is performed at 2m/min and up to 3
times. The adhesion of the dry ink on plastic and multi-
coated paper is not optimal, so this process is only done
for uncoated printing paper where the ink is more eas-
ily absorbed deep into the substrate. RS is measured by
four-probe close to the centre of the antenna, where the
highest conductivity is required, as shown in the simula-
tions in Figs.1,2. RS saturates at∼3.6Ω/(cid:3) after 2 spray
passes. Further calendaring or additional coating do not
further reduce RS. The reason is that paper fibres limit
the conducting pathways available for the FLG flakes as
the ink is absorbed into the substrate before it can dry,
due to the FLG concentration and the evaporation of wa-
ter. SEM images of SP films on Kapton and SC on paper
are shown in Figs.1a,b.
For the transponder with FLG inductor, the microchip
is glued directly to the antenna using Ag paste. For the
hybrid system, the Al inductive loop is fabricated simi-
larly to conventional dipole transponders[1], i.e. by etch-
ing Al on polyethylene terephthalate (PET)[1]. The mi-
crochip is subsequently attached onto the Al loop using
anisotropic conducting adhesive (ACA)[54] and the loop
is attached on the antenna with adhesive tape.
inductor
Using an Al
loop not only improves
impedance matching in terms of conjugate impedance,
but also reduces signal attenuation between antenna and
microchip. Indeed, forming a contact between FLG an-
tennas and microchip is challenging, especially consider-
ing the small (∼400µm×250µm) contact pads of an RFID
microchip. ACA, typically used with metallic tags[54],
does not necessarily work on FLG, due to the temper-
ature and pressure required by the bonding process[55].
Therefore, similar to Ref.[56], for the antenna wit FLG
inductive loop we use Ag paint to establish an electri-
cal contact between FLG films and RFID chip, Fig.1a.
Conversely, in our hybrid design, the printed FLG an-
tenna and the RFID chip are connected through the Al
loop and no bonding or Ag paint is required between loop
and FLG antenna. Therefore, conventional ACA can be
used to bond the RFID chip to the Al loop.
Fig.1 shows image and simulated current distribution
of the antenna with parallel inductor implemented as an
opening on the FLG conductor. The current is concen-
trated around the opening or the loop inductor of the
transponder, Fig.1b. The hybrid antenna is shown in
Fig.2a. Fig.2b is the corresponding simulated current dis-
tribution. The highest density of current is in the metal
conductor, thus maximizing power transfer to the mi-
crochip, therefore improving the reading range. The an-
tennas are measured with a TagformanceTM UHF RFID
measurement system[57] in its anechoic cabinet. The
evaluation is based on measuring the activation level of
the transponder in a fixed and known setup[50, 58]. The
transponders are attached on a piece of Styrofoam, acting
as radiation-transparent support. The measured activa-
tion level is then used to calculate the theoretical reading
range (i.e. the maximum range) in Figs.5,6. The simu-
lated reading ranges are also included for comparison.
For all antennas, the measured read range is shorter
than simulations. However, for the hybrid antenna the
discrepancy is smaller. A possible cause for this is the
roughness of the edges in the SP antennas. Fig.1 indi-
cates that the current concentrates on the edges of the
opening or the loop in the middle. Thus any added re-
sistivity there has a significant impact on losses. This
also explains why the difference between simulations and
measurements is greater for antennas with FLG induc-
tive loop. These also use Ag paste as the conductor be-
tween antenna and microchip. The connections between
Ag paste and FLG, as well as between Ag paste and mi-
crochip contact pads, are likely to introduce additional
contact resistance, hence signal attenuation.
Fig.6 shows that the reading range of SP and SC an-
tennas are almost identical. Only below∼880MHz the
distance of SP antennas is∼10% smaller than SC, show-
ing how both deposition methods are suitable for the
6
realization of FLG antennas.
The radiation patterns are also measured with the
TagformanceTM system. Figs.7,8 plot the measured di-
rectivities (solid red lines) as compared to simulations
(dashed blue lines). As the absolute directivity is diffi-
cult to measure, the measured radiation patters are nor-
malized to the simulated ones at φ=0, θ=0.
Radiation patterns, both simulated and measured, re-
veal a small difference compared to an ideal dipole an-
tenna. The radiation pattern is not perfectly round on
the azimuth plane. The difference in directivity between
0 and 180o is 2.8dB for the FLG inductive loop antenna
and 1.7dB for the hybrid one. This can be seen also
on the elevation plane and as the maximum directiv-
ity values in Table 1 being above the theoretical one of
a dipole antenna, 2.15dBi (decibels relative to isotropic
radiator)[47]. This can be attributed to the asymmetry
of the transponders combined with the FLG RS.
CONCLUSIONS
UHF RFID transponders with screen-printed and
sprayed FLG antennas were designed,
fabricated and
tested. Read ranges∼6.7 and 11.1m were measured for
antennas with FLG inductive loop and hybrid anten-
nas, respectively. The transponders operate at the fre-
quency bands reserved for UHF RFID: 865.6-867.6 MHz
(Europe) and 902-928 MHz (USA, Japan). The hybrid
antenna has reading performance superior to previously
reported graphene-based RFID tags[41, 42, 44, 45] and
comparable with commercial ones[46]. It also avoids the
need for a direct contact between FLG film and mi-
crochip, making the fabrication of FLG antennas com-
patible with existing industrial processes.
ACKNOWLEDGEMENTS
We acknowledge funding from EU Graphene Flag-
ship, the French RENATECH network, ERC Grant Het-
ero2D, EPSRC Grants EP/K01711X/1, EP/K017144/1,
EP/N010345/1, and EP/L016087/1.
[1] K. Finkenzeller, RFID Handbook: Radio-Frequency
Identification Fundamentals and Applications (Wiley,
New York, 2004) 2nd ed.
[2] I. Lacmanovi, B. Radulovi, and D. Lacmanovi, Contact-
less Payment Systems Based on RFID Technology (The
33rd International Convention MIPRO, Opatija, 2010)
1114.
[3] A. Musa and A. Dabo, Global J. Flexible Syst. Manage.
17, 189 (2016).
[4] S. Fosso Wamba, A. Anand, and L. Carter, Int. J. Inf.
Manage. 33, 875 (2013).
[5] K. B. Biji, C. N. Ravishankar, C. O. Mohan, and T. K.
S. Gopal, J. Food Sci. Tech. Mys. 52, 6125 (2015).
[6] R. Geers, B. Puers, V. Goedseels, and P. Wouters, Elec-
tronic identification, monitoring and tracking of animals
(CAB INTERNATIONAL, Wallingford, 1997).
[7] K. Ashton, RFID Journal, (2009).
[8] R. Buyya and A. V. Dastjerdi, Internet of things princi-
ples and paradigms, (Elsevier, 2016).
[9] R. Nair, E. Perret, S. Tedjini, and T. Barron, A hu-
midity sensor for passive chipless RFID applications
(IEEE International Conference on RFID-Technologies
and Applications(RFID-TA), Nice, 2012) 29-33.
[10] S. Manzari, S. Pettinari, and G. Marrocco, Miniatur-
ized and tunable wearable RFID tag for body-centric
applications (IEEE International Conference on RFID-
Technologies and Applications (RFID-TA), Nice, 2012)
239-243.
[11] A. N. Parks and J. R. Smith, Sifting through the air-
waves: Efficient and scalable multiband RF harvesting
(IEEE International Conference on RFID (IEEE RFID),
Orlando, 2014) 74-81.
[12] R. M. Ferdous, A. W. Reza, and M. F. Siddiqui, Renew.
Sust. Energ. Rev. 58, 1114 (2016).
[13] G. Swamy and S. Sarma, Manufacturing Cost Simula-
tions for Low Cost RFID, Auto-ID Center, MIT, 2003.
[14] H. Kaleel, Innovation in Wearable and Flexible Antennas
(WIT Press, Southampton, 2015).
[15] J. Manyika, R. Dobbs, M. Chui, J. Bughin, P. Bisson,
and J. Woetzel, The IoT: mapping the value beyond the
hype, McKinsey Global Institute, 2015.
[16] M. Romagnoli, V. Sorianello, M. Midrio, F. H. L. Kop-
pens, C. Huyghebaert, D. Neumaier, P. Galli, W. Templ,
A. D'Errico, A. C. Ferrari; Nature Rev. Mater. 3, 392
(2018)
[17] K. V. S. Rao, P. V. Nikitin, and S. F. Lam, IEEE T.
Antenn. Propag. 53, 3870 (2005).
[18] G. Marrocco, IEEE Antenn. Propag. M. 50, 66 (2008).
[19] K. Finkenzeller, RFID Handbook: Fundamentals and
Applications in Contactless Smart Cards, Radio Fre-
quency Identification and near-Field Communication
(Wiley, Chichester, 2010).
[20] Z. Cui, Printed Electronics: Materials, Technologies and
Applications (John Wiley & Sons Singapore Pte. Ltd,
Singapore, 2016).
[21] https://www.hgkippax.co.uk.
[22] P. V. Nikitin, S. Lam, and K. V. S. Rao, Low cost sil-
ver ink RFID tag antennas (IEEE Antennas and Prop-
agation Society International Symposium, Washington,
2005) 2B, 353.
[23] P. Pongpaibool, A study of cost-effective conductive ink
for inkjet-printed RFID application (International Sym-
posium on Antennas and Propagation (ISAP), Nagoys,
2012) 1248-1251.
[24] A. L. Dearden, P.J. Smith, D.Y. Shin, N. Reis, B. Derby,
and P. OBrien, Macromol. Rapid Commun. 26, 315
(2005).
7
[29] F. Bonaccorso, A. Lombardo, T. Hasan, Z. Sun, L.
Colombo, and A. C.Ferrari, Mater. Today 15, 564-589
(2012).
[30] J. Perruisseau-Carrier, Graphene for antenna applica-
tions: Opportunities and challenges from microwaves to
THz (Loughborough Antennas & Propagation Confer-
ence (Lapc), Loughborough, 2012).
[31] K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, Y.
Zhang, S.V. Dubonos, G. IV, and A.A. Firsov, Science
306, 666-669 (2004).
[32] S.A. Awan, A. Lombardo, A. Colli, G. Privitera, T.S.
Kulmala, J.M. Kivioja, M. Koshino, and A.C. Ferrari,
2D Mater. 3, 015010 (2016).
[33] J. Horng, C.F. Chen, B.S. Geng, C. Girit, Y.B. Zhang,
Z. Hao, H.A. Bechtel, M. Martin, A. Zettl, M.F. Crom-
mie, Y.R. Shen, and F. Wang, Phys. Rev. B 83, 165113
(2011).
[34] J. Perruisseau-Carrier, M. Tamagnone, J. S. Gomez-
Diaz, and E. Carrasco, Graphene antennas: Can integra-
tion and reconfigurability compensate for the loss (Euro-
pean Microwave Conference (EuMC), Nuremberg, 2013)
369-372.
[35] P. G. Karagiannidis, et al., ACS Nano 11, 2742 (2017)
[36] J. W. Birkenshaw, The Printing Ink Manual, R. H.
Leach, R. J. Pierce, E. P. Hickman, M. J. Mackenzie,
and H. G. Smith (Springer, Dordrecht, 2007) 5th ed. 2,
14-85.
[37] W. J. Hyun, S. Lim, B. Y. Ahn, J. A. Lewis, C. D. Fris-
bie, and L. F. Francis, ACS Appl. Mater. & Interfaces 7,
12619 (2015).
[38] A. Moridi, S.M. Hassani-Gangaraj, M. Guagliano, and
M. Dao, Surf. Eng. 30, 369 (2014).
[39] C. Acquarelli, A. Rinaldi, A. Tamburrano, G. D. Bel-
lis, A. G. D. Aloia, and M. S. Sarto, Graphene-based
EMI shield obtained via spray deposition technique (In-
ternational Symposium on Electromagnetic Compatibil-
ity, Gothenburg, 2014) 488-493.
[40] X. J. Huang, T. Leng, X. Zhang, J. C. Chen, K. H.
Chang, A. K. Geim, K. S. Novoselov, and Z. R. Hu, Appl.
Phys. Lett. 106, 203105 (2015).
[41] M. Akbari, M. W. A. Khan, M. Hasani, T. Bjorninen,
L. Sydanheimo, and L. Ukkonen, IEEE Antennas Wirel.
Propag. Lett. 15, 1569 (2016).
[42] P. Kopyt, B. Salski, M. Olszewska-Placha, D. Janczak,
M. Sloma, T. Kurkus, M. Jakubowska, and W. Gwarek,
IEEE T. Antenn. Propag. 64, 2862 (2016).
[43] A. Lamminen, K. Arapov, G. de With, S. Haque, H.
G. O. Sandberg, H. Friedrich, and V. Ermolov, IEEE
Antennas Wirel. Propag. Lett. 16, 1883 (2017).
[44] K.W. Pan, Y.Y. Fan, T. Leng, J.S. Li, Z.Y. Xin, J.W.
Zhang, L. Hao, J. Gallop, K.S. Novoselov, and Z.R. Hu,
Nat. Commun. 9, 5197 (2018).
[45] K. Arapov, K. Jaakkola, V. Ermolov, G. Bex, E. Rub-
ingh, S. Haque, H. Sandberg, R. Abbel, G. de With, H.
Friedrich, Physica Status Solidi 10, 812 (2016)
[46] https://www.smartrac-group.com/files/content/Products
[25] A. Louwen, W. van Sark, R. Schropp, and A. Faaij, Sol.
Services/PDF/0028 SMARTRAC DOGBONE.pdf.
Energ. Mat. Sol. C 147, 295-314 (2016).
[47] D. M. Pozar, Microwave Engineering (Wiley, New York,
[26] S. Merilampi, T. Laine-Ma, and P. Ruuskanen, Micro-
2011).
electron. Reliab. 49, 782 (2009).
[27] R. R. Søndergaard, N. Espinosa, M. Jorgensen, and F.
C. Krebs, Energ. Environ. Sci. 7, 1006 (2014).
[48] Impinj Monza R6TM UHF RFID microchip,
https://support.impinj.com/hc/en-
datasheet.
us/articles/202765328-Monza-R6-Product-Datasheet.
[28] A.C. Ferrari, et al., Nanoscale 7, 4598-4810 (2015).
[49] T. A. Milligan, Modern Antenna Design (McGraw Hill,
New York, 1985).
[50] P. Nikitin, K. V. S. Rao, and S. Lam, UHF RFID
tag characterization: Overview and state-of-the-art (An-
tenna measurement techniques association symposium
(AMTA), Seattle, 2012).
[51] Z. Su, S. C Cheung, K. T. Chu, Investigation of radio
link budget for UHF RFID systems, In Proc. 2010 IEEE
International Conference on RFID-Technology and Ap-
plications, (Guangzhou, 2010)
[52] The European regulatory environment for radio equip-
ment and spectrum, https://cept.org/ (2017).
[53] 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).
[54] Y.C. Lin and J. Zhong, J. Mater. Sci. 43, 3072 (2008).
[55] J. Liu, ACA bonding technology for low cost electron-
ics packaging applications-current status and remain-
ing challenges (4th International Conference on Adhesive
Joining and Coating Technology in Electronics Manufac-
turing, Espoo, 2000).
[56] X. Huang, et al., Sci. Rep. 5, 18298 (2015).
[57] http://www.voyantic.com.
[58] P. Pursula, M. Hirvonen, K. Jaakkola, and T. Varpula,
IEEE T. Antenn. Propag. 55, 2836 (2007).
8
|
1803.08914 | 1 | 1803 | 2018-03-23T17:57:49 | Long-term deep supercooling of large-volume water via surface sealing with immiscible liquids | [
"physics.app-ph",
"cond-mat.soft"
] | Supercooling of aqueous solutions below their melting point without any crystallization is a fundamentally and practically important physical phenomenon with numerous applications in biopreservation and beyond. Under normal conditions, heterogeneous nucleation mechanisms critically prohibit the simultaneous long-term (> 1 week), large volume (> 1 ml) and low temperatures (< -10 oC) supercooling of aqueous solutions. Here, in order to overcome this bottleneck and enable novel and practical supercooling applications, we report on the use of surface sealing of water by an oil phase to drastically diminish the primary heterogeneous nucleation at the water/air interface. Using this approach, we have achieved deep supercooling (as low as -20 oC) of large-volumes of water (up to 100 ml) for long periods (as long as 100 days) simultaneously. Since oils are mixtures of various hydrocarbons we also report on the use of pure alkanes and primary alcohols of various lengths to achieve the same. All alcohols and some of the longer alkane chains we studied show high capacity to inhibit freezing. The relationship of this capacity with the chain length, however, shows opposite trends for alcohols and alkanes due to their drastically different interfacial structures with the water molecules. We find that the deeply supercooled water (at -20 oC) can withstand vibrational and thermal disturbances with all sealing liquids used, and even an extreme disturbance, ultrasonication, when alcohols are used as the sealing phase. The deep supercooling approach we developed here, for large samples and long periods, is expected to enable novel applications of supercooling in a variety of areas including biopreservation and food storage among others. | physics.app-ph | physics | Long-term deep supercooling of large-volume water via surface sealing with immiscible
liquids
Haishui Huang1, Martin L. Yarmush1,2*, O. Berk Usta1*
1 Center for Engineering in Medicine, Massachusetts General Hospital, Harvard Medical School
and Shriners Hospitals for Children, Boston, Massachusetts 02114, United States
2 Department of Biomedical Engineering, Rutgers University, Piscataway, New Jersey 08854,
United States
* Correspondence should be addressed to:
O. Berk Usta, PhD
Center for Engineering in Medicine (CEM)
Massachusetts General Hospital
Harvard Medical School
Shriners Hospitals for Children
51 Blossom Street
Boston, MA 02114
E-mail: [email protected]
Abstract:
Martin L. Yarmush, MD, PhD
Center for Engineering in Medicine (CEM)
Massachusetts General Hospital
Harvard Medical School
Shriners Hospitals for Children
51 Blossom Street
Boston, MA 02114
E-mail: [email protected]
Supercooling of aqueous solutions below their melting point without any crystallization is a
fundamentally and practically important physical phenomenon with numerous applications in
biopreservation and beyond. Under normal conditions, heterogeneous nucleation mechanisms
critically prohibit the simultaneous long-term (> 1 week), large volume (> 1 ml) and low
temperatures (< -10 oC) supercooling of aqueous solutions. Here, in order to overcome this
bottleneck and enable novel and practical supercooling applications, we report on the use of
surface sealing of water by an oil phase to drastically diminish the primary heterogeneous
nucleation at the water/air interface. Using this approach, we have achieved deep supercooling
(as low as -20 oC) of large-volumes of water (up to 100 ml) for long periods (as long as 100 days)
simultaneously. Since oils are mixtures of various hydrocarbons we also report on the use of pure
alkanes and primary alcohols of various lengths to achieve the same. All alcohols and some of
the longer alkane chains we studied show high capacity to inhibit freezing. The relationship of this
capacity with the chain length, however, shows opposite trends for alcohols and alkanes due to
their drastically different interfacial structures with the water molecules. We find that the deeply
supercooled water (at -20 oC) can withstand vibrational and thermal disturbances with all sealing
liquids used, and even an extreme disturbance, ultrasonication, when alcohols are used as the
sealing phase. The deep supercooling approach we developed here, for large samples and long
periods, is expected to enable novel applications of supercooling in a variety of areas including
biopreservation and food storage among others.
Water is a seemingly simple yet a practically complex liquid with extraordinary phase behavior,
which enables many of life's intricacies. While water is possibly the most studied liquid, there
remain many areas where its behavior is still largely mysterious (1). A prime example for this is
the freezing and the supercooling of water, one of the most important yet least understood
phenomena in our daily lives and scientific research (2, 3). Ice formation and the preceding
supercooled state of microdroplets in atmospheric clouds is a crucial element for precipitation and
reflection of solar radiation (4, 5). Furthermore, chilling, freezing, freeze avoidance, and
supercooling are important strategies to combat cold weather for ectothermic animals (6, 7) , treat
malignant diseases via cryotherapy (8), and preserve food and various biological samples, such
as cells, tissues, and organs (9, 10).
Recent advances have shown that supercooling can be a promising alternative approach for the
preservation of cells, tissues, and especially organs (11). Nevertheless, an important hurdle for
supercooling preservation, as well as other applications of supercooling, is that simultaneous low
temperature (< -10 oC), large volume (> 1ml), and long period (> 1 week) of supercooling for
aqueous solutions cannot be readily achieved (12-14). While high pressure based approaches
have provided supercooled states of water down to -92 oC briefly (1), according to the water phase
diagram.They are, however, expensive, might further complicate preservation of biological
samples, and their long-term fate is unknown. Few experiments have unstably supercooled large
volumes, several hundred milliliters, of water to -12 oC (15), albeit also for periods on the order of
seconds. Similarly, in Dorsey's classical work on freezing of supercooled water, he was able to
achieve a temperature of -19 oC for a few milliliters of water very briefly during his constant cooling
experiments (16). A method that overcomes these hurdles and enable long-term supercooling of
large aqueous samples at low temperatures could enable applications in biopreservation as well
as many other areas which have previously been practically prohibited.
Under normal atmospheric conditions, ice transitions into water at 0 °C, i.e. the melting point or
the equilibrium temperature (𝑇𝑒). Nevertheless, the observed freezing temperature (𝑇𝑓) for pure
water is usually below the equilibrium temperature (𝑇𝑒). Water, in the liquid phase, below the
equilibrium temperature is said to be "supercooled" where ∆𝑇 = 𝑇𝑒 − 𝑇𝑓 measures the degree of
supercooling. Supercooled water is intrinsically metastable and can spontaneously transform to
lower-energy-level ice crystals through the formation of ice nuclei, which can be readily achieved
by ice seeding (17), ultrasonicating (18), or presenting ice-nucleating agents (19). On the contrary,
it is very difficult to maintain supercooled water unfrozen, especially for a large volume, under a
high degree of supercooling (∆𝑇), or for a long period, as each of these increases the possibility
of ice nucleation and water freezing (Supplemental Information (SI)). For instance, ∆𝑇 of a water
droplet decreases logarithmically with increasing volume under a constant cooling rate (20).
Similarly, supercooling frequency (𝑓𝑠, 𝑓𝑠 = number of unfrozen droplets / number of total droplets)
of an ensemble of droplets decreases exponentially with increasing droplet volume, storage time,
and nucleation rate (𝐽) (21, 22), while 𝐽 itself increases exponentially with ∆𝑇 (23). Consequently,
simultaneous long-term (> 1 week), large-volume (> 1 ml), and deep supercooling (DSC, ∆𝑇 > 10
°C) of water has not yet been achieved. In what follows, we describe a novel and unexpected
method based on sealing of the water surface by an immiscible hydrocarbon based liquid, such
as oils, pure alkanes and pure primary alcohols. This method, as we demonstrate through a series
of experiments, enables stable supercooling of large volumes of water for long periods at
temperatures well below -10 °C.
two general
ice nucleation mechanisms, homogenous and heterogeneous
There are
crystallization. Homogeneous crystallization occurs due to random aggregation of interior water
molecules to create a critically large nucleus of ice crystal, which could only be achieved and
observed below -20 °C (24). Heterogeneous crystallization, on the other hand, stems from ice
nucleus formation catalyzed by a substrate and/or with the aid of foreign objects at much higher
temperatures (25). Consequently, water freezing is generally initiated by heterogeneous
nucleation, and especially, the water/air interface is the primary nucleation site as revealed in
theoretical (26, 27), experimental (14, 28), and numerical (29, 30) studies. When water molecules
aggregate on the water surface (water/air interface) to form an ice nucleus, they need to overcome
an energy barrier 𝛾𝑖𝑎 − 𝛾𝑤𝑎 (𝛾: interfacial tension, superscripts 𝑖, 𝑤, 𝑜, and 𝑎 refer to ice, water,
oil, and air, respectively) per unit area as the ice/air interface replaces original water/air interface.
In comparison, the energy barrier for homogeneous ice nucleation within bulk water is proportional
to the water/ice interfacial tension, 𝛾𝑤𝑖. This interfacial tension can be expressed via the Young
equation as 𝛾𝑤𝑖 = 𝛾𝑖𝑎 − 𝛾𝑤𝑎𝑐𝑜𝑠𝜃𝑖𝑤𝑎 ≥ 𝛾𝑖𝑎 − 𝛾𝑤𝑎 (𝜃𝑖𝑤𝑎: water contact angle on ice/water/air
interface, Fig. S1(A)). This inequality indicates that heterogeneous ice nucleation on the surface
is thermodynamically more favorable than homogeneous nucleation in bulk as complete wetting
(𝜃𝑖𝑤𝑎 = 0°) is not generally observed (27), and receding contact angles of 12o have been reported
(31). Therefore, if the water surface is sealed by an oil phase, the energy barrier of ice nucleation
at the water-oil interface would be 𝛾𝑖𝑜 − 𝛾𝑤𝑜. Similarly, the homogenous nucleation energy barrier
can be now expressed in terms of another triple interface, namely the oil/water/ice as 𝛾𝑤𝑖 = 𝛾𝑖𝑜 −
𝛾𝑤𝑜𝑐𝑜𝑠𝜃𝑖𝑤𝑜 where 𝜃𝑖𝑤𝑜, for many oils can be nearly 0° (Fig. S1(B)) (28, 32). In the case of
𝜃𝑖𝑤𝑜 ≅ 0, the energy barrier approaches the limiting case 𝛾𝑖𝑜 − 𝛾𝑤𝑜 ≅ 𝛾𝑤𝑖 . This analysis
indicates that the energy barrier of heterogeneous crystallization at the surface is elevated almost
to the level of homogeneous one when the water/air interface is replaced by an oil-water interface.
Accordingly, we hypothesized that surface sealing of water with an appropriate oil phase could
suppress primary heterogeneous ice nucleation at the surface and enable extended storage of
deeply supercooled water.
First, we cooled a large ensemble of polystyrene tubes containing 1ml of ultra-pure water to -13
°C (Fig. 1(A-B)). This resulted in > 90% of samples to be frozen after 24 hours and nearly all
samples to be frozen after 5 days. In contrast, the ultra-pure water samples could be kept in the
liquid phase for a week, at the same temperature, if their surfaces were sealed by various types
of immiscible oils, such as light mineral oil (MO), olive oil (OO), heavy paraffin oil (PO), and
nutmeg oil (NO). Interestingly, the curdling of OO during DSC does not trigger water freezing,
though the cumulative freezing frequency (𝑓𝑓, 𝑓𝑓 = 1 − 𝑓𝑠 ) increases significantly compared to
water sealed by other oils (Fig. 1(A)). In supplementary experiments, we have observed that the
water degassed by vacuuming for 24 hours, has similar 𝑓𝑓 as normal water, with or without oil
sealing (Fig. S2). These experiments indicate that air dissolved in the water does not play a major
role in ice nucleation in our experiments. Given this result and the consistent efficacy of surface
sealing by different oils on freezing reduction, we infer that the air-water interface is the primary
nucleation site.
We also examined the influence of water volume on the efficacy of oil sealing for freezing
inhibition. We studied the two most promising oils, MO and PO, at -13 and -16 oC for ultra-pure
water ranging from 100 105 µl (Fig. 1(C-D), Fig. S3). We found that MO sealing can effectively
suppress water freezing for water volumes up to 104 µl at -13 and -16 °C. PO sealing was even
more effective with a low 𝑓𝑓 throughout the entire volume range at -13 °C, and only 45.8% of
samples frozen at -16 °C for the 105 µl samples. In addition, 8 out of 35 (22.8%) samples of 105
µl water were kept in the supercooled state at -16 °C for 100 days without any freezing event after
Day-3 (Fig. S3(B)). While further investigations might be necessary, these observations are
incompatible with conventional stochastic freezing processes (SI), which implies exponential
decrease of 𝑓𝑠 with time (21, 22). Alternatively, the freezing of DSC water sealed by oil could be
depicted as "case-specific" that some of sealed water samples are more susceptible to
crystallization than others due to their unique and slightly different microstructures on the
interface, even though all of them possess higher capability to resist freezing compared to the
water without sealing. A reconciliation of these two cases might lie in the fact that those samples
that don't' freeze within our observation period have much fewer impurities and thus a much
smaller exponential for the decay of 𝑓𝑠 than those that freeze within 3 days.
In order to further support our hypothesis that the water/air interface plays a dominant role in ice
nucleation and subsequent freezing, we measured water freezing frequencies under differential
degrees of surface sealing by MO, ranging from a) unsealed (0 oil), b) ring sealed along the
contact line between water and tube wall (0.01 ml), c) partially sealed with partial exposure to air
(0.1 ml), d) critically sealed with water surface just completely covered (0.15 ml), e) normally
sealed (0.5 ml), and f) over sealed with excessive oil mounted on water surface (3.5 ml) (Fig. 2(A-
B) and Fig. S4). The results indicate that the capacity of freezing inhibition increases with the
degree of sealing, with a statistically maximum plateau achieved by critical sealing (Fig. 2(A)).
Ring sealing (0.01 ml) that nullifies the triple solid/water/air contact line has a mild effect on
freezing inhibition at high temperatures (-10 °C) but is not effective below -13 °C. Taken together
with partial sealing results (0.1 ml), this result implies that the contact nucleation at the air-water-
solid triple interface is not as dominant as that at water/air interface especially at low
temperatures. Considering the crystallization efficiency depends on the integration of nucleation
probability 𝐽 and nucleation length (or area), the triple contact line of short length would provide
smaller crystallization efficiency compared than the air/water interface even though it has higher
𝐽 (33, 34). Overall, we confirmed that the water/air interface is the primary ice nucleation site for
DSC water, and surface oil sealing that removes the water/air interface can effectively inhibit ice
nucleation and water freezing.
We also observed that additional oil beyond the critical sealing has a statistically negligible effect
on freezing suppression. This indicates that additional pressure and dampening effects,
associated with a long-column of viscous oil phase, have a negligible effect on freezing inhibition.
In order to further test this, we examined the effects of viscosity of the sealing agents where we
used hydroxy (-OH) terminated polydimethylsiloxane (PDMS) of different chain lengths (Fig.
2(C)). In a similar fashion, we did not observe statistically significant differences in the capacity of
freezing inhibition of PDMS with a viscosity range of 1 - 5x105 cP, with the exception of 3500 cP
PDMS which resulted in almost no freezing suppression effect. We hypothesize that this odd
behavior is likely due to the formation of an ordered structure between water and this particular
PDMS on the interface through hydrogen bonding, which closely matches the lattice of hexagonal
ice (35).
Most oils are complex mixtures of alkanes, saturated cyclic alkanes, alkylated aromatic groups,
and fatty acids among other hydrocarbon compounds. In an effort to more systematically study
the observed freezing inhibition effect of supercooled water sealed with an immiscible
hydrocarbon phase, we studied two prototypical families of hydrocarbons: linear alkanes and their
corresponding primary alcohols of different lengths (Fig. 3). Specifically, we have studied alkanes
(CmH2m+2, denoted Cm, m = 5 - 11) and primary alcohols (CmH2m+1OH, denoted CmOH, m = 4 - 8)
as the sealing agents for DSC water at -20 °C. Since linear alkanes have very low polarity, they
have weak interaction with polar water molecules. On the other hand, the primary alcohols, which
are amphipathic, can form strong hydrogen bonds with water through their hydroxyl group
(hydrophilic end) and even stable ordered interfacial structures.
We found that 𝑓𝑓 of DSC water, at -20 °C, sealed with alkanes decreases monotonically with
increasing carbon number m and chain length 𝑙 (Fig. 3(A)). The capacity of alkanes in freezing
inhibition matches that of MO (Fig. 2(A) at -20 °C) at m > 9. Given that mineral oils tend to have
carbon chain lengths above 10, this result is expected. We posit that this trend, correlated with
the alkane chain length, might also explain the capacity differences between PO and MO in Fig.
1(D); PO likely consists of higher carbon chain alkanes than MO based on their densities (PO ~
0.855-0.88 vs MO ~ 0.838 g/ml). On a molecular level, the mechanism for this trend might lie in
the structure of the alkane/water interface. It has been observed that an interfacial electron
depletion layer with a thickness 𝛿 exists between water and hydrophobic alkane chains by both
X-ray reflectivity (XR) measurements (36, 37) and atomistic molecular dynamics (MD) simulations
(38, 39). The few water molecules in the depletion layer (electron density < 40% that of bulk water
(40)) can buckle in the intermolecular space near the ends of alkane molecules (Fig. 3(B)), and
create a template for the formation of an ice nucleus (29). The alkane chains adjacent to the water
molecules preferentially have their longest axis parallel to the water interface with a tilt angle 𝛽
(36). This tilt angle increases with m and 𝑙, resulting in a more parallel orientation for longer
alkanes (36). Accordingly, longer alkane chains are expected reduce the corrugation and
roughness of the interface on the side of alkanes. This, consequently, is expected to decrease
the number of buckled water molecules and nucleus templates, and thus lower the probability of
heterogeneous ice nucleation on that layer (29). These expectations are in line with our
observations of decreasing freezing frequencies for longer alkane chain lengths. From the
perspective of thermodynamics, longer and flatter-oriented alkanes results in fewer and sparser
buckled water molecules in the interface serving as nucleation template, which implies smaller
contact region between ice embryo and sealing alkanes, smaller 𝜃𝑖𝑤𝑜(even though they are
already much smaller than 𝜃𝑖𝑤𝑎), and higher energy barrier for heterogeneous ice nucleation (Fig.
S1(B)).
On the other hand, 𝑓𝑓 of DSC water, at -20 °C, sealed with alcohols increases with m and 𝑙. This
opposite trend is likely due to the starkly different structures of the alcohol/water interface
compared to that of alkane/water interface (Fig. 3(C)). Unlike the alkanes which prefer a parallel
orientation, the primary alcohols orient perpendicularly to the interface with a small 𝛽. The primary
alcohols align their hydroxyl (-OH) heads toward the interface to form hydrogen bonds with water
molecules. Accordingly, no depletion layer of interfacial water exists as in the alkane/water
interface. The 2D layer of interfacial water molecules are strongly hydrogen-bonded to the
hydroxyl groups, with their H atoms pointing toward alcohol as revealed by heterodyne-detected
vibrational sum frequency spectroscopy (SFG) (41). Therefore, structures and dimensions of the
contacting layer of amphilic alcohols essentially determine the distribution and arrangement of
interfacial water molecules, and the formation of heterogeneous ice nucleus (35, 42).
Experimental and computational investigations of ice nucleation in droplets under monolayers of
long primary alcohol chains with 16 ≤ 𝑚 ≤ 31, revealed a very low tilt angle 𝛽 (~ 7.5 - 10 °) and a
very good lattice match between hexagonal ice and the alcohol structure for 29 ≤ m ≤ 31 (43).
For these longest chains ice nucleation occurs at temperatures as high as -1 °C. As m and 𝑙
decrease, the tilt angle 𝛽 increases up to ~ 19° for m = 16 (44). In conjunction, a greater lattice
mismatch between hexagonal ice lattice and ordered alcohol layer at the interface along with a
lower ice nucleation efficiency and freezing temperature were observed (35, 43-45). Extrapolating
this information and trend to the shorter alcohols (4 ≤ m ≤ 8) in this study, we expect a larger tilt
angle 𝛽 (e.g. 𝛽 = 28° for m = 6 and 𝛽 = 30° for m = 5 (44)) and a greater lattice mismatch between
hexagonal ice and ordered alcohol structure given the general structural similarity of primary
alcohols. Compared to longer alcohol chains, the interfacial -OH groups anchored to smaller
alcohols have stronger in- and out-of-plane fluctuations at the same temperature. We, therefore,
expect that the large lattice mismatch along with the –OH group fluctuations can destabilize any
ordered domaine of crystalline water and impede the formation of ice nucleus of critical size (45).
Given that both effects are larger with smaller chain lengths, we expect that higher nucleation
inhibition can be achieved by smaller primary alcohols, in line with our experimental observations.
From the perspective of thermodynamics, greater lattice mismatch and interface fluctuation
associated with shorter alcohol molecules directly reduce the probability of the formation of icing
template of critical size for successful nucleation, which indicates smaller stable contact area
between ice nucleus and sealing alcohols and thus, higher free energy barrier for heterogeneous
ice nucleation at the interfac. Once again, we observed that there is no significant difference of 𝑓𝑓
between 1-day and 7-day storage when sealed by either alkanes or alcohols. This further
suggests the case-specific, rather than stochastic, nature of water freezing with oil sealing that
we have previously discussed.
Having established the efficacy of the deep supercooling approach using either oils or pure alkane
and alcohol phases, we then studied its stability under vibrational, thermal, and ultrasonic
disturbances. Vibrational disturbances were introduced by placing DSC water onto a shaking
plate with various shaking speeds and frequencies (SI). When the DSC water (-20 °) is sealed by
MO, its 𝑓𝑓 is 0% and 5.6%, respectively, under 0.84 g and 2.1 g centrifugal acceleration (Fig.
4(A)), which are are much higher than ac/deceleration forces of a commercial airliner (0.2-0.4 g)
during potential transporation. Thermal distrubances were induced by putting the DSC samples
into 37 °C incubator or plungig them into 37 °C water bath with warming rate of 100 C/min (heated
by natural convection in air) or 102 C/min (heated by forced convection in water), respectively.
Very few (0% for gas warming, 2.5% for water warming) of the samples freeze under these
thermal fluctuations. In contrast, these samples can not endure ultrasonication in 40 kHz
ultrasonic water bath (Fig. 4(A) and Video S1), with 𝑓𝑓 of ~ 84%. This is probably due to the
vigorous collapse of cavitation bubbles in water during ultrasonication (18), which would cause
ultrahigh local pressure (> 1 GPa) (48), and therefore, significantly increase equilibrium
temperature 𝑇𝑒 and the degree of supercooling ∆𝑇.
Upon the instability of DSC water sealed by MO under ultrasonication, we further tested its
stability sealed by pure alkanes and pimary alcohols. DSC water sealed by alkanes freeze
immediately upon being ultrasonicated (Fig. 4(B-C) and Video S2), which is consistent with
previous observation of MO sealed water since MO has a high content of various alkanes. On the
contrary, none of the samples freeze upon ultrasonification if they were sealed by any of the
primary alcohols (Fig. 4(B)). Instead, the sealing alcohols would be emulsified with supercooled
water, starting from the interface and then evolving toward supercooled water (Fig. 4(C) and Video
S3). The exact mechanism of the freezing resistance of DSC water sealed by alcohols to
ultrasonic disturbance is still unknown, and one hypothesis would be that ultrasound
preferrentially transduces its energy into joint molecular motion at interface due to the hydrogen
bonding between water and amphilic alcohols to form nanoemulsion (49), rather than cavitation
bubbles for ice nucleation in viscous DSC water.
In the preceding, we demonstrated a seemingly counterintuitive and novel approach to achieve
long-term deep supercooling of large-volume water by using a hydrocarbon-based immiscible
phase to seal the water surface. Our initial observations with laboratory grade oils demonstrated
that replacing the water/air interface, which is the primary ice nucleation site, with a water/oil
interface dramatically inhibits stochastic freezing processes. The seemingly time independent
nature of the freezing frequency of oil-sealed water suggests that its freezing might be case-
specific rather than stochastic. Our studies with linear alkanes and primary alcohols suggest that
freezing inhibition can be achieved by surface sealing with starkly different interfacial structures
and microscopic mechanisms, which results in opposite trends of inhibition capacity correlated to
the chain length. While all sealed DSC water show great stability under vibrational and thermal
disturbances simulating normal storage and transportation conditions, only the primary alcohol
sealed supercooled water can withstand ultrasonication. We note that while we have
hypothesized about possible macroscopic (thermodynamic) and microscopic mechanisms that
might explain our observations, further studies are warranted to test, confirm, and improve upon
them. Especially, since most existing literature focuses on longer alkane and alcohol chains at
the water interfaces and the resulting molecular structures, computational and experimental
studies with short chains might prove useful. Similarly, careful measurements of interfacial
properties and structures at low temperatures with mixed and pure hydrocarbons can shed further
light on why some oils are more effective than others.
Given that monolayers of long alcohol chains have been historically used to initiate ice nucleation,
our results with the short chains to prevent nucleation expand the use of alcohols to provide a
robust control mechanism over the temperature at which nucleation can be achieved in an
aqueous solution. Further studies with different families of hydrocarbons and their mixtures will
be aimed at expanding this robust control of supercooling degree to allow novel applications.
Beyond its fundamental implications, deep supercooling of large volumes of aqueous solutions
can enable previously prohibitive applications, and provide new biopreservation methodologies
for cell, tissue, and organ engineering and transplantation, as well as other areas, such as food
preservation. Given our prior experience and interest in both organ and cell preservation using
supercooling and the limitations we have previously encountered in terms of temperatures,
volumes and durations for preservation, we believe that the deep supercooling via the surface
sealing with immiscible phases will be vital in advancing these applications forward. The
immediate goal is to translate this approach to preservation of large number of cells that are
amenable to preservation at the supercooled temperatures we can achieve here, and then
translate such results to the clinic. We will then explore tissue and organ preservation with deep
supercooling approaches.
ACKNOWLEDGEMENTS
We are grateful to Prof. Mehmet Toner for the very helpful discussions and his suggestions. We
would also like to thank the NIH for funding this work through grants no. 5P41EB002503
(BioMEMS Resource Center), 1R21EB020192, 5R01EB023812.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
H. Kanno, R. J. Speedy, C. A. Angell, Supercooling of Water to -92°C Under Pressure. Science 189,
880-881 (1975).
T. Bartels-Rausch, Chemistry: Ten things we need to know about ice and snow. Nature 494, 27-
29 (2013).
R. E. Lee Jr, G. J. Warren, "Biological ice nucleation and its applications," (1995).
J. D. Atkinson et al., The importance of feldspar for ice nucleation by mineral dust in mixed-
phase clouds. Nature 498, 355-358 (2013).
U. Burkhardt, B. Kärcher, Global radiative forcing from contrail cirrus. Nature climate change 1,
54 (2011).
J. P. Costanzo, M. C. do Amaral, A. J. Rosendale, R. E. Lee, Jr., Hibernation physiology, freezing
adaptation and extreme freeze tolerance in a northern population of the wood frog. J Exp Biol
216, 3461-3473 (2013).
J. S. Bale, S. A. Hayward, Insect overwintering in a changing climate. J Exp Biol 213, 980-994
(2010).
W. K. Yiu, M. T. Basco, J. E. Aruny, S. W. Cheng, B. E. Sumpio, Cryosurgery: A review. Int J Angiol
16, 1-6 (2007).
G. J. Morris, E. Acton, Controlled ice nucleation in cryopreservation--a review. Cryobiology 66,
85-92 (2013).
C. James, G. Purnell, S. J. James, A review of novel and innovative food freezing technologies.
Food and bioprocess technology 8, 1616-1634 (2015).
T. A. Berendsen et al., Supercooling enables long-term transplantation survival following 4 days
of liver preservation. Nat Med 20, 790-793 (2014).
A. Gholaminejad, R. Hosseini, A study of water supercooling. Journal of Electronics Cooling and
Thermal Control 3, 1 (2013).
O. B. Usta et al., Supercooling as a viable non-freezing cell preservation method of rat
hepatocytes. PloS one 8, e69334 (2013).
R. A. Shaw, A. J. Durant, Y. Mi, Heterogeneous surface crystallization observed in undercooled
water. J Phys Chem B 109, 9865-9868 (2005).
T. Inada, X. Zhang, A. Yabe, Y. Kozawa, Active control of phase change from supercooled water
to ice by ultrasonic vibration 1. Control of freezing temperature. International Journal of Heat
and Mass Transfer 44, 4523-4531 (2001).
N. E. Dorsey, The Freezing of Supercooled Water. Transactions of the American Philosophical
Society 38, 247-328 (1948).
H. Huang et al., Pre-dehydration and Ice Seeding in the Presence of Trehalose Enable Cell
Cryopreservation. ACS Biomaterials Science & Engineering, (2017).
R. Hickling, Nucleation of freezing by cavity collapse and its relation to cavitation damage.
Nature 206, 915-917 (1965).
B. Nagare, C. Marcolli, A. Welti, O. Stetzer, U. Lohmann, Comparing contact and immersion
freezing from continuous flow diffusion chambers. Atmospheric Chemistry and Physics 16, 8899-
8914 (2016).
E. Langham, B. Mason, in Proceedings of the Royal Society of London A: Mathematical, Physical
and Engineering Sciences. (The Royal Society, 1958), vol. 247, pp. 493-504.
B. Krämer et al., Homogeneous nucleation rates of supercooled water measured in single
levitated microdroplets. The Journal of Chemical Physics 111, 6521-6527 (1999).
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
T. Koop, B. Luo, U. M. Biermann, P. J. Crutzen, T. Peter, Freezing of HNO3/H2SO4/H2O solutions
at stratospheric temperatures: Nucleation statistics and experiments. The Journal of Physical
Chemistry A 101, 1117-1133 (1997).
B. J. Murray et al., Kinetics of the homogeneous freezing of water. Physical chemistry chemical
physics : PCCP 12, 10380-10387 (2010).
E. Sanz et al., Homogeneous ice nucleation at moderate supercooling from molecular
simulation. Journal of the American Chemical Society 135, 15008-15017 (2013).
C. Hoose, O. Möhler, Heterogeneous ice nucleation on atmospheric aerosols: a review of results
from laboratory experiments. Atmospheric Chemistry and Physics 12, 9817 (2012).
J. W. Cahn, Critical point wetting. The Journal of Chemical Physics 66, 3667-3672 (1977).
Y. Djikaev, A. Tabazadeh, P. Hamill, H. Reiss, Thermodynamic conditions for the surface-
stimulated crystallization of atmospheric droplets. The Journal of Physical Chemistry A 106,
10247-10253 (2002).
A. Tabazadeh, Y. S. Djikaev, H. Reiss, Surface crystallization of supercooled water in clouds. Proc
Natl Acad Sci U S A 99, 15873-15878 (2002).
M. Fitzner, G. C. Sosso, S. J. Cox, A. Michaelides, The Many Faces of Heterogeneous Ice
Nucleation: Interplay Between Surface Morphology and Hydrophobicity. J Am Chem Soc 137,
13658-13669 (2015).
S. J. Cox, S. M. Kathmann, B. Slater, A. Michaelides, Molecular simulations of heterogeneous ice
nucleation. I. Controlling ice nucleation through surface hydrophilicity. J Chem Phys 142, 184704
(2015).
C. A. Knight, The contact angle of water on ice. Journal of Colloid and Interface Science 25, 280-
284 (1967).
A. Konno, K. Izumiyama, in Proceedings of the 17th International Symposium on Okhotsk Sea and
Sea Ice. (2002), pp. 275-282.
S. Suzuki et al., Freezing of water droplets on silicon surfaces coated with various silanes.
Chemical Physics Letters 445, 37-41 (2007).
R. P. Sear, Nucleation at contact lines where fluid–fluid interfaces meet solid surfaces. Journal of
Physics: Condensed Matter 19, 466106 (2007).
M. Gavish, R. Popovitz-Biro, M. Lahav, L. Leiserowitz, Ice nucleation by alcohols arranged in
monolayers at the surface of water drops. Science 250, 973-975 (1990).
M. Fukuto et al., Nanoscale Structure of the Oil-Water Interface. Phys Rev Lett 117, 256102
(2016).
T. R. Jensen et al., Water in contact with extended hydrophobic surfaces: direct evidence of
weak dewetting. Physical review letters 90, 086101 (2003).
C. D. Wick, T.-M. Chang, J. A. Slocum, O. T. Cummings, Computational investigation of the n-
alkane/water interface with many-body potentials: the effect of chain length and ion
distributions. The Journal of Physical Chemistry C 116, 783-790 (2011).
Y. Qiu, V. Molinero, Strength of Alkane–Fluid Attraction Determines the Interfacial Orientation
of Liquid Alkanes and Their Crystallization through Heterogeneous or Homogeneous
Mechanisms. Crystals 7, 86 (2017).
A. Poynor et al., How water meets a hydrophobic surface. Physical review letters 97, 266101
(2006).
J. A. Mondal, V. Namboodiri, P. Mathi, A. K. Singh, Alkyl Chain Length Dependent Structural and
Orientational Transformations of Water at Alcohol-Water Interfaces and Its Relevance to
Atmospheric Aerosols. J Phys Chem Lett 8, 1637-1644 (2017).
L. H. Seeley, G. T. Seidler, Two-dimensional nucleation of ice from supercooled water. Phys Rev
Lett 87, 055702 (2001).
43.
44.
45.
46.
47.
48.
49.
Y. Dai, J. S. Evans, Molecular Dynamics Simulations of Template-Assisted Nucleation: Alcohol
Monolayers at the Air− Water Interface and Ice Formation. The Journal of Physical Chemistry B
105, 10831-10837 (2001).
R. Popovitz-Biro et al., Induced freezing of supercooled water into ice by self-assembled
crystalline monolayers of amphiphilic alcohols at the air-water interface. Journal of the
American Chemical Society 116, 1179-1191 (1994).
Y. Qiu et al., Ice Nucleation Efficiency of Hydroxylated Organic Surfaces Is Controlled by Their
Structural Fluctuations and Mismatch to Ice. J Am Chem Soc 139, 3052-3064 (2017).
A. Abdelmonem et al., Surface charge-induced orientation of interfacial water suppresses
heterogeneous ice nucleation on α-alumina (0001).
D. Ehre, E. Lavert, M. Lahav, I. Lubomirsky, Water freezes differently on positively and negatively
charged surfaces of pyroelectric materials. Science 327, 672-675 (2010).
R. Hickling, Transient, high-pressure solidification associated with cavitation in water. Phys Rev
Lett 73, 2853-2856 (1994).
A. Gupta, H. B. Eral, T. A. Hatton, P. S. Doyle, Nanoemulsions: formation, properties and
applications. Soft Matter 12, 2826-2841 (2016).
Figures
Figure 1. Deep supercooling (DSC) of pure water enabled by surface sealing with various types
of oils. (A) Cumulative freezing frequency (𝑓𝑓) for 1 ml water at -13 °C over 7-day DSC, without
sealing (W/O seal), with surface sealing by light mineral oil (MO), olive oil (OO), heavy paraffin oil
(PO), and nutmeg oil (NO). Number of independent experiments n = 6, number of total tested
samples for each case N = 56. NS: p > 0 .05; *: 0.005 < p < 0.05; **: 1.0 × 10-6 < p < 0.005, ***: p
< 1.0 × 10-6. Error bar represents standard deviation. (B) Corresponding samples of (A) post 1-
day storage. (C) 𝑓𝑓 of DSC water of various volumes post 1-day storage at -13 and -16 °C. n = 7,
N = 272, 145, 336, 123, and 125 for 3, 30, 200, 1000, and 10000 μl water, respectively. (D) 𝑓𝑓 of
100000 μl water with different sealing oils and temperatures post 1-day storage, n = 7, N = 35.
Figure 2. Dependence of freezing efficiency of DSC water on the volume and viscosity of sealing
agent. (A) Effect of sealing oil (MO) volume on 𝑓𝑓 post 1-day DSC at different temperatures. n =
6, N = 70. (B) Side view of corresponding samples of (A). MO includes Oil Red O for staining and
imaging. 0 oil, 0.01 ml, 0.1ml, 0.15 ml, 0.5 ml, and 3.5 ml indicate no seal, ring seal, partial seal,
critical seal (just complete surface seal), standard seal, and over seal by MO, respectively. (C)
Effect of viscosity of sealing agents on 𝑓𝑓 post 1-day DSC at -16 °C. The sealing agents are
hydroxy (OH) terminated polydimethylsiloxane (PDMS) of different chain lengths and viscosities.
n = 5, N = 56.
Figure 3. Freezing efficiency of 1 ml water sealed with linear alkanes and primary alcohols at -20
°C. n = 7, N = 87. When m > 11 for linear alkanes and m > 8 for primary alcohols, the sealing
agents are frozen at -20 °C and cause DSC water frozen. When m < 5, the linear alkanes are
gaseous under atmospheric condition and not suitable for sealing. When m < 4, the primary
alcohols are miscible with water and not suitable for sealing either. (B-C) Schematic
configurations of alkane/water (B), and alcohol/water interface (C), respectively. The alkane and
alcohol molecules are displayed without aliphatic hydrogen atoms and colored in light green. The
O and H atoms in hydroxyl group of alcohol and water are shown in red and white dots,
respectively.
Figure 4. Stability tests for -20 °C DSC water. (A) 𝑓𝑓 of DSC water sealed by MO under various
disturbances. Vibrational disturbance was imposed by shaking plate with different shaking
frequencies and centrifugal forces (i.e. 0.84 g or 2.1 g). Thermal disturbance was imposed by
placing or plunging the DSC tubes into 37 °C incubator (37 °C gas) or water bath (37 °C water).
Ultrasonic disturbance was introduced by putting the DSC tubes into 40 kHz ultrasonic water bath.
n = 6, N = 48. (B) 𝑓𝑓 of DSC water sealed by linear alkanes and primary alcohols under 40 kHz
ultrasonic disturbance. n = 3, N = 24 (except for C5, N = 8). (C) Representative image sequences
of ultrasonication tests for DSC water sealed by linear alkanes or primary alcohols.
Figure 1
Figure 2
Figure 3
Figure 4
Materials and Methods
Supplemental Information (SI)
For all experiments in this study, DNase/RNase-free distilled water (Life Technologies/Thermo-
Fisher Scientific, USA) was used to minimize potential pollutants or ice-nucleating agents, except
DSC trials of 100 ml water where deionized (DI) water (resistivity 𝑅 = 18.2 MΩ) produced by a
deionizing water system (METTLER TOLEDO Thornton, USA) was used. All water containers
(dishes, 96-well plates, round-bottomed tubes, and bottles, Corning, USA) were made of
polystyrene, and clean and sterile before experiments. The purity of all oil phases used for water
surface sealing is at least 99% as specified by the vendor (Sigma-Aldrich, USA).
The loading of water into containers was performed in a chemical hood to avoid contamination of
the samples by pollutants or dust particles in the air. Water of small volume (< 1 ml) was loaded
into containers (dishes or 96-well plates) using clean and sterile tips (Thermo Fisher Scientific,
USA) and calibrated pipets (PIPETMAN, Gilson, USA), while that of large volume (≥ 1 ml) was
loaded into containers (round-bottomed tubes or bottles) using serological pipets (Thermo Fisher
Scientific, USA) by pipette filler (Drummond Scientific, USA). We note that as water droplets
smaller than 100 µl are subject to significant evaporation during long-term deep supercooling
(DSC) experiments, and those bigger than 105 µl (100 ml) beyond the volume capacity of the
freezing chamber, they were not investigated in this study. After loading water samples into the
containers, oil phase was gently added onto the water surface using serological pipets, trickling
down along the wall of containers to avoid splashing or trapping air bubbles at the interface. The
water-laden containers (with or without sealing oil) were transferred into portable temperature
controlled freezers (Engel MHD-13, Engel, USA) that were placed in 4 °C cold room to minimize
temperature fluctuations, or stored in -20 °C freezer (Thermo Fisher Scientific, USA). The
temperatures within these freezers were verified by Toluene-filled low-temperature thermometer
(Sigma, USA).
To examine the effects of dissolved air in water on ice nucleation and water freezing, the water
was vacuumed at a pressure below 10-4 atmosphere for 24 hours to extract dissolved air
molecules. The degassed water was, then, gently pipetted into tubes and sealed with mineral oil
(MO) for supercooling tests at -16 °C. The air content of the degassed water is significantly lower
than that of normal water without degassing, as no air bubbles emerge from the degassed water
(second row of Fig. S2(B)) under vacuum. The same procedure was carried out for normal water
for comparative purposes, and several big air bubbles can be observed after 3-hour degassing
(first row of Fig. S2(B))
To test the stability of DSC water sealed by oil phase, three types of disturbances, vibrational,
thermal, and ultrasonic disturbances, were studied. For vibrational disturbance test, DSC tubes
were placed on shaking plate (Labline 4625 titer shaker, Marshall Scientific, USA) with shaking
speed 500 and 800 rpm for 30 seconds, which give rise to the centrifugal acceleration of 0.84 and
2.1 g (g is gravitational acceleration), respectively. To prevent heat transfer, the tubes were
wrapped with thick tissue paper in tube racks, all of which had been previously cooled to -20 °C
in freezer. The temperature of the DSC water would not change noticeably during experiments
given the brief shaking period and thick insulation layer. For thermal disturbance test, the DSC
tubes were put into 37 °C incubator (warmed by air) or plunged into 37 °C water bath (warmed by
water). Therefore, DSC water would experience different warming rates and temperature
gradient. For ultrasonic disturbance test, DSC tubes would be plunged into 4 °C ultrasonic water
bath. The sonicator (Branson B-200, TMC Inductries, USA) generates 40 kHz ultrasonic wave
with power 30 W. The freezing of the DSC water can be evidenced by the change of sample
transparency (from transparent to opaque).
All data were organized and reported as the mean ± standard deviation from at least three
independent runs of experiments (n > 3); further information on sample numbers are disclosed in
figure captions. The statistical significance of mean values between two groups was determined
by Microsoft Excel based on Student's two-tailed t-test, assuming equal variance. Although a p-
value less than 0.05 is generally regarded as statistically significant, different ranges of p-value
(NS: 0.05 < p, *: 0.005 < p < 0.05, **: 0.005 < p < 10-6, ***: p < 10-6) were provided to show different
degrees of significance.
Stochastic process of ice nucleation and freezing
The formation of a critical ice embryo, i.e. a successful nucleation, in metastable supercooled
water is generally regarded as a stochastic process that does not depend on the number of
previous nucleation trials or correlate to other nucleation events during the same period (1, 2). In
addition, heterogenous nucleation is the major type of crystallization in this study since
homogeneous nucleation in water occurs at much lower temperatures (around - 40 °C). As a
result, the heterogeneous ice nucleation on water surfaces/interfaces would follow Poisson
statistics
𝐼𝑛 (1 − 𝑓𝑓(𝑡)) = −𝐽(𝑇) ∙ 𝑆 ∙ 𝑡
where 𝑓𝑓(𝑡) is the freezing frequency after supercooling of a period 𝑡, 𝐽(𝑇) is the nucleation rate
at temperature 𝑇, and 𝑆 is the area of heterogeneous nucleation sites. Therefore, for water
samples of the same volume and shape under a constant temperature, the non-frozen
(supercooled) fraction is expected to decline exponentially with time.
However, in our experiments we found that 𝑓𝑓(𝑡) of DSC water with oil sealing does not change
significantly after Day 3 as shown in Fig. 1, Fig. 3, Fig. S2, and Fig. S3. These results indicate
that the heterogeneous ice nucleation in DSC water sealed by oil phase does not follow the
conventional theory of stochastic nucleation processes at the interface. Particularly, since 45.8%
of 100 ml DSC water (-16 °C) sealed by PO are frozen post 1-day storage (Fig.
1(D)), 𝐼𝑛 (1 − 0.458) = −𝐽(−16 °𝐶) ∙ 𝑆 ∙ 1, which gives rise to 𝐽(−16 °𝐶) ∙ 𝑆 = 0.612. Therefore, the
expected fraction of unfrozen samples, 1 − 𝑓𝑓(𝑡) = 𝑒−0.612𝑡, would decrease exponentially with
storage time for DSC water of the same volume and shape at -16 °C. As a result, the fraction of
unfrozen samples would be 0.22% and 2.5×10-25 % on Day-10 and Day-100, respectively, which
implies almost all the sample would be frozen after 10-day storage. However, we observed that
22.9% (8 out of 35) of samples were still unfrozen post 100-day storage in our experiments.
Moreover, no freezing event occurred between Day-3 and Day-100; that is all the samples that
were unfrozen on Day-3 remained unfrozen till the end of our experiments on Day-100. These
observations strongly demonstrated a non-stochastic process of ice nucleation in our DSC water-
oil phase systems.
Using similar heuristics, our experimental results also suggest that the freezing we observed is
not due to homogeneous ice nucleation either, where the formation of critical ice embryo is caused
by spontaneous aggregation of water molecules via random translational, rotational, and
vibrational movements that would conform stochastic process (3). The exact kinetics and statistics
of this heterogeneous ice nucleation in DSC water sealed by oil phase is still unknown and
detailed future investigations are certainly warranted.
Freezing point depression due to oil-water mixing
When a water sample is sealed by an oil phase (i.e mixed oils, and pure alkanes and alcohols)
for DSC, the "immiscible" oil might slightly dissolve in supercooled water to decrease the
equilibrium melting temperature 𝑇𝑒 below 0 °C, the equilibrium melting point of pure water under
atmospheric conditions. We, therefore, quantified the potential depression of freezing point due
to this effect and assessed whether it's comparable to the high degree of supercooling we
observed in our experiments. According to the Bladgen's Law, the extent of freezing point
depression ∆𝑇𝐹 can be calculated by
∆𝑇𝐹 = 𝑖𝑏𝐾𝐹
where 𝑖 is the Van't Hoff factor (𝑖 = 1 for nonelectrolytes or oil phase in this study), 𝑏 is the molality
of oil phase in water, and 𝐾𝐹 is the cryoscopic constant (𝐾𝐹 = 1.85 K∙kg/mol for water). Therefore,
∆𝑇𝐹 can be determined by the solubility of sealing oils in water at DSC temperatures. Solubility of
oils in metastable water at DSC temperatures are not readily available; as such we have assessed
∆𝑇𝐹 using the available solubility data of oils in water under room temperature. This approach,
likely, leads to an overestimation since solubility of oils in water typically increases with
temperature (4).
For oil mixtures (MO, OO, PO, NO, and PDMS) and linear alkanes (C5 ~ C11) utilized in this study,
the maximum solubility is 0.04 g/L (or 0.55 mM) (C5 in water), and the corresponding estimate for
∆𝑇𝐹 is less than 1.03 × 10-3 °C, which is negligible compared to the degree of supercooling ∆𝑇 (10
to 20 °C) achieved using these oil phases as sealing agents.
For alcohols used in this study, the maximum solubility is 73 g/L (or 0.98 M) (C4OH in water at
room temperature) and the corresponding estimate for ∆𝑇𝐹 is less than 1.82 °C. This likely
overestimated freezing point depression accounts for about 9.1% of ∆𝑇 (20 °C) enabled by alcohol
sealing. Moreover, the DSC water and sealing alcohols are likely not mixed altogether. The stable
contact interface with strong hydrogen bonding on the head and a long hydrophobic tail of
alcohols, low molecular mobility, and viscous water at - 20 °C would significantly impede the
diffusion of alcohol molecules into water. We, therefore, conclude that the depression of freezing
temperature due to oil-water mixing does not play a significant role in achieving the observed high
degree of supercooling in our experiments.
1.
2.
3.
4.
T. Koop, B. Luo, U. M. Biermann, P. J. Crutzen, T. Peter, Freezing of HNO3/H2SO4/H2O solutions
at stratospheric temperatures: Nucleation statistics and experiments. The Journal of Physical
Chemistry A 101, 1117-1133 (1997).
D. Niedermeier et al., Heterogeneous ice nucleation: exploring the transition from stochastic to
singular freezing behavior. Atmospheric Chemistry and Physics 11, 8767-8775 (2011).
B. Krämer et al., Homogeneous nucleation rates of supercooled water measured in single
levitated microdroplets. The Journal of Chemical Physics 111, 6521-6527 (1999).
S. Paasimaa, Factors affecting water solubility in oils. Vaisala News 169, 24-25 (2005).
Supplemental figures
Figure S1. Schematics for ice/water/air (A) and ice/water/oil (B) contacts. θiwa and θiwo are water
contact angles on ice/water/air and ice/water/oil interfaces, respectively.
Figure S2. Comparison of cumulative freezing frequencies between normal and degassed DSC
water at - 16 °C. "D_W/O seal", "D_MO", and "D_PO" represent degassed water without sealing,
surface sealing with MO, and PO, respectively. n = 6, N = 60, NS: p > 0.05. (B) Degassing images
for normal and degassed water with or without MO sealing. The red dash circles indicate air
bubbles precipitated under vacuum.
Figure S3. Representative images of DSC water of various volumes. (A) 30 μl droplets with or
without MO sealing post 1-day DSC at - 16 °C. (B) 100 ml deionized (DI) water sealed by 16 ml
PO post 100-day DSC at - 16 °C. All possible freezing events occur before Day-3, and 8 out of
total 35 (n = 7, N = 35) bottles of water maintain unfrozen after 100-day storage.
Figure S4. Vertical view of water surface in round-bottomed tubes sealed by MO of various
volumes. The oil was stained by Oil Red O for enhanced contrast. Newton's rings were observed
due to the curvature of sealing oil near the tube wall.
Supplemental videos
Video S1. Ultrasonication for 1 ml DSC water sealed by MO. The water is supercooled at - 20 °C
for 1 day and the ultrasonic frequency is 40 kHz.
Video S2. Ultrasonication for 1 ml DSC water sealed by undecane (C11). The water is supercooled
at - 20 °C for 1 day and the ultrasonic frequency is 40 kHz.
Video S3. Ultrasonication for 1 ml DSC water sealed by 1-butanol (C4OH). The water is
supercooled at - 20 °C for 1 day and the ultrasonic frequency is 40 kHz.
Figure S1
Figure S2
Figure S3
Figure S4
|
1912.10916 | 1 | 1912 | 2019-10-02T17:46:31 | Advances in Numerical Simulation of High-Speed Impact Welding | [
"physics.app-ph",
"nlin.PS"
] | Numerical simulations of high-speed forming and welding are of significant interest to industry, but are challenging due to the coupled physics and dynamic nature of the processes. With the advancement in hardware and computational capabilities, the next generation of computational methods, so called meshless methods, have received significant attention. Among all meshless methods, smoothed particle hydrodynamics (SPH) has received major consideration. The main advantage of the SPH method is to bypass the requirement for a numerical grid to calculate spatial derivatives. This avoids severe problems associated with mesh tangling and distortion which usually occur in Lagrangian analysis involving high-strain-rate events. In this study to better understand the effects of oxide layer, coating and diffused materials on weldability, a novel hybrid SPH platform was developed. Then, the high-speed impact between Steel/Steel, Copper/Titanium, and Aluminum/Steel were simulated. To experimentally validate the numerical efforts, results were compared to vaporizing foil actuator welding and explosive welding tests. Good agreement between the numerical simulations and experimental results provided confidence in the numerical modeling. | physics.app-ph | physics | Proceedings of NUMIFORM 2019: The 13th International Conference on Numerical Methods in Industrial Forming Processes
Editors: Yannis Korkolis, Brad Kinsey, Marko Knezevic, and Nikhil Padhye
Advances in Numerical Simulation of High-Speed Impact Welding
Ali Nassiri 1*, Tim Abke 2, and Glenn Daehn 3
1Simulation Innovation and Modeling Center (SIMCenter), The Ohio State University,
Smith Laboratory Room 3076, 174 West 18th Avenue, Columbus, OH 43210, USA
2Honda R&D Americas, Inc., 21001 OH-739, Raymond, OH 43067, USA
3 Department of Materials Science and Engineering, The Ohio State University, 116 W
19th Avenue, Room 141 Fontana Labs, Columbus, OH 43210, USA
*[email protected]
Keywords: Impact welding, Smoothed particle hydrodynamics, Meshless method
Abstract
Numerical simulations of high-speed forming and welding are of significant interest to
industry, but are challenging due to the coupled physics and dynamic nature of the
processes. With the advancement in hardware and computational capabilities, the next
generation of computational methods, so called meshless methods, have received
significant attention. Among all meshless methods, smoothed particle hydrodynamics
(SPH) has received major consideration. The main advantage of the SPH method is to
bypass the requirement for a numerical grid to calculate spatial derivatives. This avoids
severe problems associated with mesh tangling and distortion which usually occur
in Lagrangian analysis involving high-strain-rate events. In this study to better understand
the effects of oxide layer, coating and diffused materials on weldability, a novel hybrid
SPH platform was developed. Then, the high-speed impact between Steel/Steel,
Copper/Titanium, and Aluminum/Steel were simulated. To experimentally validate the
numerical efforts, results were compared to vaporizing foil actuator welding and explosive
welding tests. Good agreement between the numerical simulations and experimental
results provided confidence in the numerical modeling.
1. Introduction
In the automotive, aerospace and various other industries, there is a primary need for
lightweight structures. This goal can be achieved by using multi-material assemblies. But
owing to the disparate melting temperatures of the materials and concerns about brittle
intermetallic compounds formation, traditional fusion welding processes cannot be used.
One means to join dissimilar metals is through high-speed impact welding (HSIW).
Typically, the HSIW process involves a high-speed, oblique collision between two metals
arranged in relatively simple geometries. A striking characteristic of metals joined using
HSIW is the emergence of a distinctive wavy pattern at the interface of two welded
workpieces which is assumed to be necessary for creation of a quality impact weld (see
Figure 1). To this end, numerical approaches are considered to be necessary, especially
in predicting the shape and temperature distribution at the interface.
647
Figure 1. A wavy morphology between a) Steel/Steel [1], Copper/Brass [2], C)
Copper/Titanium [3], and d) Copper/Aluminum [4]
However, finite element simulations of the HSIW process is very challenging, especially
near the welding zone where large deformations occur. Thus, a traditional Lagrangian
method where the mesh is fixed to the workpiece geometry fails as a result of excessive
element distortion. Smoothed particle hydrodynamics (SPH) is a viable method to
simulate the highly dynamic events. SPH is a Lagrangian technique where the
coordinates move with the objects. The SPH particles are interpolation points from which
values of functions and their derivatives can be estimated at discrete points in the
continuum. The function values and their derivatives are found by a kernel approximation
instead of being constructed from a grid. Since in SPH there is no connectivity
between the particles, they are able to move relative to each other in the domain
of simulation. Hence, metal jet emission can be simulated which was previously
impossible using Lagrangian mesh-based numerical methods. Jetting is believed to be
as a prerequisite for welding to occur. This phenomenon cleans nascent oxide layers,
contaminants, and coating off the mating surfaces, and this allows solid state bonding
without extensive melting and long range diffusion. In this study, in order to further
understand the science behind the HSIW process and find a relationship between the
composition of the jetted materials and weldability, a new hybrid computational platform
was developed using a combination of SPH and traditional Lagrangian mesh-based
methods. Then, the high-speed impact between Steel/Steel, Copper/Titanium, and
Aluminum/Steel were simulated using this platform. To experimentally validate the
numerical efforts, results were compared to vaporizing foil actuator welding (VFA) and
explosive welding experimental tests.
2. Numerical Simulation Model
The numerical simulations were carried out in LS-DYNA software package. Figure 2a
shows a schematic illustration of the traditional SPH oblique impact model. A flyer plate
is inclined β degree with respect to a target plate and given the initial impact velocity of
Vf through the flyer plate. To model a typical impact between two plates with the same
size (e.g., a=c=2.5mm and b=d=2.5mm) over 2,000,000 particles with a size of 4µm
needs to be generated. In the new proposed platform, SPH region was set to be 250µm
in the thickness where the plates suffer heavy deformation by collision. A Lagrangian
mesh-based platform was used for the rest of the model (See Figure 2b). The boundary
between the SPH and Lagrangian regions was connected by defining a joint function. The
joint function ties the two regions. In this model, the normal and shear stresses and their
failure values are used as controlling parameters that define the strength of the tie. In the
new platform less than 90,000 particles were generated. Another advantage of the hybrid
platform is that different layers of materials can be added to the SPH section in which
648
each layer has its own thermo-mechanical properties and particle size (see Figure 2c). In
this study, the material behavior was modeled using the Johnson-Cook constitutive
relationship. The pressure at the contact point, which is incorporated in the stress tensor,
was computed using the Grϋneisen state equation. Also, the experimentally measured
impact velocity was used as an initial condition and input to the numerical code. See Ref
[5,6] for more information about the governing equations and the material parameters.
Figure 2. Schematic of the simulation setup; a) traditional SPH, b) hybrid platform, c)
hybrid platform with coating and oxide layers
3. Results and Discussion
Figure 3 shows a feature-by-feature comparison between the numerical simulations and
experimental tests in Steel/Steel and Copper/Titanium bimetallic systems. As is clear, the
new platform was able to capture both wavy morphology and the swirling motion of the
materials. Figure 4a shows the simulation result between an aluminum (Al6111-T4) plate
impacting a dual phase high-strength steel (JAC980) plate.
Figure 3. Comparison between the numerical simulations and experimental tests; Left
column- impact between Steel/Steel using explosive welding [6]; right column- impact
between Copper/Titanium using VFA
In JAC980, a galvanneal steel, to avoid corrosion, a layer of Zn coating is added to the
base metal. As is shown, the hybrid platform was captured the jetted materials. The result
shows that the ejected particles contain aluminum oxide, base aluminum, and zinc. Also,
649
there is no indication of the base steel among the jetted materials which implies that
pressure at the contact point was not enough to penetrate through the target plate and
remove the coating. Therefore, no bond between the two base materials was cretaed.
This can be confirmed by comparing the experimental result and numerical simulation at
the end of the process which shows a failed welded sample and simulation, respectively
(Figure 4b,c). In term of computational time, the suggested hybrid platforms are
approximately 4 times faster than the traditional SPH platform.
Figure 4. a) Simulation result between Steel/Aluminum, b) failed sample, c) simulation
result as the end of the process showing the interface is separated
To validate the accuracy of the new platform, in the future, the composition of the jetted
materials from the model will be compared with the jetted materials that will be collected
in experiments by placing a witness block perpendicular to the direction of weld. The new
platform can be utilized to reduce the number of trial-and-error iteration during the
experimental tests by providing the numerically captured process parameters (e.g. impact
angle and impact velocity).
4. References
[1] Mendes R, Ribeiro JB, Loureiro A (2013) Effect of explosive characteristics on the explosive
welding of stainless steel to carbon steel in cylindrical configuration. Mater. Des., 51, 182-192.
[2] Faes K et al. (2010) Joining of Copper to Brass using magnetic pulse welding. In: Proceedings of
the 4th International conference on High Speed Forming, Columbus, OH, 84 -- 96
[3] Vivek A, Hansen SR, Liu B, and Daehn GS (2013) Vaporizing foil actuator: A tool for collision
welding. J. Mater. Process. Technol. 213 (12), 2304-2311.
[4] Raoelison RN, Racine D, Zhang Z, Buiron N, Marceau D, Rachik M (2014) Magnetic pulse welding:
Interface of Al/Cu joint and investigation of intermetallic formation effect on the weld features.
Journal of Manufacturing Processes, 6(4): 427 -- 434.
[5] Nassiri A, Vivek A, Abke T, Liu B, Lee T, Daehn G (2017) Depiction of interfacial morphology in
impact welded Ti/Cu bimetallic systems using smoothed particle hydrodynamics. Applied Physics
Letters, 110(23), 231601.
[6] Li XJ, Mo F, Wang XH, Wang B, Liu KX (2012) Numerical study on mechanism of explosive welding.
Science and Technology of Welding and Joining, 17(1), 36-41.
650
|
1908.10442 | 1 | 1908 | 2019-08-02T14:54:28 | Time-resolved mid-infrared dual-comb spectroscopy | [
"physics.app-ph",
"physics.ins-det",
"physics.optics",
"physics.plasm-ph"
] | Dual-comb spectroscopy can provide broad spectral bandwidth and high spectral resolution in a short acquisition time, enabling time-resolved measurements. Specifically, spectroscopy in the mid-infrared wavelength range is of particular interest, since most of the molecules have their strongest rotational-vibrational transitions in this "fingerprint" region. Here we report time-resolved mid-infrared dual-comb spectroscopy for the first time, covering ~300 nm bandwidth around 3.3 {\mu}m with 6 GHz spectral resolution and 20 {\mu}s temporal resolution. As a demonstration, we study a CH4/He gas mixture in an electric discharge, while the discharge is modulated between dark and glow regimes. We simultaneously monitor the production of C2H6 and the vibrational excitation of CH4 molecules, observing the dynamics of both processes. This approach to broadband, high-resolution, and time-resolved mid-infrared spectroscopy provides a new tool for monitoring the kinetics of fast chemical reactions, with potential applications in various fields such as physical chemistry and plasma/combustion analysis. | physics.app-ph | physics | Preprint
August 2, 2019
Time-resolved mid-infrared dual-comb spectroscopy
Muhammad A. Abbas, Qing Pan, Julien Mandon, Simona M. Cristescu, Frans J. M.
Harren & Amir Khodabakhsh*
Trace Gas Research Group, Department of Molecular and Laser Physics, Institute for Molecules and Materials,
Radboud University, 6525 AJ Nijmegen, The Netherlands
Abstract
Dual-comb spectroscopy can provide broad spectral bandwidth and high spectral resolution in
a short acquisition time, enabling time-resolved measurements. Specifically, spectroscopy in the
mid-infrared wavelength range is of particular interest, since most of the molecules have their
strongest rotational-vibrational transitions in this "fingerprint" region. Here we report time-
resolved mid-infrared dual-comb spectroscopy for the first time, covering ~300 nm bandwidth
around 3.3 μm with 6 GHz spectral resolution and 20 μs temporal resolution. As a
demonstration, we study a CH4/He gas mixture in an electric discharge, while the discharge is
modulated between dark and glow regimes. We simultaneously monitor the production of C2H6
and the vibrational excitation of CH4 molecules, observing the dynamics of both processes. This
approach to broadband, high-resolution, and time-resolved mid-infrared spectroscopy provides a
new tool for monitoring the kinetics of fast chemical reactions, with potential applications in
various fields such as physical chemistry and plasma/combustion analysis.
1
Preprint
August 2, 2019
Time-domain monitoring of fast chemical reactions is of particular interest in several
fundamental and applied scientific fields, including physical chemistry, plasma/combustion
analysis, biology, and environmental studies. Broadband, time-resolved absorption spectroscopy
can provide the possibility to simultaneously monitor time-dependent parameters of the chemical
reactions, such as concentrations of intermediate/final chemical products, transient free radicals
and ions, as well as branching ratios, reaction rate coefficients, temperature and number densities
of molecular excited-states. Generally, the main challenge is to obtain a broadband spectrum
with high spectral resolution and high detection sensitivity in a short measurement time.
Continuous-wave (cw) laser absorption spectroscopy can provide time-resolved measurements
for a single chemical species with a high detection sensitivity. However, for a broad spectral
coverage the laser source needs to be scanned over the spectral range, inevitably reducing the
measurement speed. Alternatively, one can use broadband time-resolved absorption spectroscopy
techniques, which are traditionally based on incoherent light sources. They can provide an ultra-
broadband time-resolved spectrum, but they need a long averaging time to achieve a high signal-
to-noise ratio (SNR) and detection sensitivity. Two widely used methods are step-scan
mechanical Fourier transform spectroscopy (FTS) 1-3 and dispersion-based detection 4-7. The
former exhibits very long measurement times due to the step-scanning, while the latter yields
shorter measurement times, but usually has a coarse spectral resolution.
In contrast to these traditional broadband methods, optical frequency comb spectroscopy
(OFCS) simultaneously provides a broad spectral coverage and a high spectral resolution. It can
also yield a high SNR within a short measurement time, due to the coherency and high spectral
brightness of optical frequency comb sources. Specifically, OFCS in the mid-infrared (mid-IR)
wavelength range (2-20 µm) has been of particular interest, since almost all molecules have their
2
Preprint
August 2, 2019
fundamental rotational-vibrational transitions in this region with distinct absorption patterns (i.e.
fingerprints). Various OFCS techniques have been utilized in the mid-IR wavelength region; e.g.
combining an optical frequency comb with a mechanical FTS 8, dual-comb spectroscopy (DCS)
9-11, and dispersion-based methods 12-15. A comprehensive review of these spectroscopic methods
can be found elsewhere 16.
Time-domain/time-resolved spectroscopy using optical frequency combs has emerged
strongly in the last decade. In a first demonstration, DCS was used for measuring molecular free
induction decay in the near-infrared (near-IR) wavelength range using two Er:fiber mode-locked
lasers 17. A few other works have been reported in near-IR region using Ti:sapphire mode-locked
lasers including multidimensional dual-comb spectroscopy (M-DCS2) able to differentiate and
assign the Doppler-broadened features of two naturally occurring isotopes of Rb 18, dual
frequency comb-based transient absorption (DFC-TA) spectroscopy for measurement of the
relaxation processes of dye molecules in solution from femtosecond to nanosecond timescales 19,
and DCS for the study of laser-induced plasma from a solid sample, simultaneously measuring
trace amounts of Rb and K in a laser ablation 20. In the visible range (~530 nm), cavity-enhanced
transient absorption spectroscopy (CE-TAS) has been demonstrated for study of the ultrafast
dynamics of I2 in a molecular beam 21, and more recently, time-resolved dual-comb spectroscopy
has been reported for measurement of number density and temperature in a laser-induced plasma
by monitoring three excited-state transitions of Fe 22. In the mid-IR region, the demonstration has
been limited to cavity-enhanced time-resolved frequency comb spectroscopy (TRFCS) for
monitoring of transient free radicals and kinetics of the OD + CO → DOCO reaction by 2D
cross-dispersion of the spectrum on a liquid N2 cooled camera using a virtually imaged phase
array (VIPA) etalon in combination with a conventional grating 23-25.
3
Preprint
August 2, 2019
Here, we report time-resolved dual-comb spectroscopy (TRDCS) in the mid-IR wavelength
region, for the first time to the best of our knowledge, and demonstrate its application for
monitoring fast chemical dynamics. For this, we study the vibrational excitation/de-excitation of
CH4 in an electrical discharge and the concentration of the reaction product C2H6, at millisecond
and microsecond time scales, while the discharge is modulated between dark and glow regimes.
Results
Dual-comb mid-IR spectrometer. The dual-comb spectrometer (Fig. 1) is based on a
femtosecond singly-resonant optical parametric oscillator (OPO) containing two nonlinear
crystals in a single cavity. The crystals are synchronously pumped by two Yb:fiber mode-locked
lasers (counter propagating and cross polarized) with a stabilized repetition rate (frep) difference
of ∆frep ≈ 250 Hz and free running carrier-envelope offset frequencies (fceo) 10,26. The two mid-IR
idler beams (~3.3 μm) are combined on a beam splitter producing two pairs of beams. One pair is
transmitted through a ~50 cm-long discharge tube (diameter 3 mm) containing CH4 diluted in
He, and focused onto a fast photodetector. The discharge tube has a continuous gas flow (4
normal liter/hour, 4 Nl/h) and is connected to a gas handling system. The second comb pair is
sent through a reference absorption cell (filled with CH4 at low pressure) and dispersed by a
diffraction grating. A part of the spectrum is focused on a second photodetector to monitor a
single well-defined absorption line of the reference sample. The time-domain interferograms in
the output of the two photodetectors are digitized, a Blackman apodization function is applied to
the both (sample and reference) interferograms, and followed by a Fourier transformation to
yield the corresponding absorption spectra. The apodization function is used to minimize the
ringing effect around the (narrow) absorption lines 27, which also limits the effective
measurement time-window of each interferogram to ~120 μs around the central burst. Therefore,
4
Preprint
August 2, 2019
each individual interferogram is measured in 120 μs with a repetition period of 1/∆frep (≈ 4 ms).
The absorption line in the spectrum of the reference gas is used to correct for the frequency jitter
in the sample spectrum, which is mainly due to the free running fceo values; it also provides an
absolute optical frequency calibration. The free-running fceo of the two combs makes the
experimental setup much less complex compared to the state-of-the-art, fully stabilized, mid-IR
dual-comb spectrometers 28,29 . The explanation of the data acquisition and signal processing of
the optical referencing method can be found elsewhere 27 and a more detailed description of the
experimental setup is presented in the Methods section.
Fig. 1. Experimental setup. Two femtosecond Yb:fiber lasers with stabilized and
slightly different frep (and free running fceo), synchronously pump two MgO:PPLN
crystals in a single OPO cavity, providing two mid-IR idler beams. The two
combined idler beams are sent through the sample (discharge) cell and a reference
gas cell. The latter yields the dual-comb spectrum of a single well-defined
5
Preprint
August 2, 2019
absorption line, which is used to correct for the free running fceo, as well as
absolute frequency calibration of the sample spectrum. M flat mirror, M1-4
concave dielectric mirror, CM flat chirped dielectric mirror, DM dichroic mirror,
BD beam dump, BS beam splitter, λ⁄2 half-waveplate, L lens, PD photodetector,
LPF low pass filter.
To demonstrate the performance of the spectrometer, we filled the discharge cell with 50% CH4
diluted in He, at 25 mbar total pressure and a total flow rate of 4 Nl/h, and measured the
transmission spectrum without the discharge. Figure 2a shows the normalized transmission
spectrum of the fundamental CH4 transition from the ground state to the ν3 vibrational state (in
black, 500 averages, ~2 s measurement time). The spectrum is normalized to an averaged
background spectrum, which was recorded when the sample absorption cell was filled with pure
He at 25 mbar. We compare this experimental absorption spectrum to the theoretical model
spectrum of CH4 (in blue) developed based on the corresponding parameters from the HITRAN
database 30. We used a Voigt profile and convoluted the model spectrum with a Blackman
instrument line-shape function, corresponding to the applied apodization function. Note that the
simulated spectrum is inverted and the two measured and simulated spectra are offset for clarity.
We fit the developed model spectrum to the measured spectrum (with CH4 concentration as the
fitting parameter) and also take into account the remaining baseline and etalon fringes by
including a sum of a low order polynomial and few low frequency sinewave functions in the fit.
The retrieved CH4 concentration is 49.7(9)%, where the error is the standard deviation of 10
consecutive measurements. The spectral resolution of the measured spectrum is ~6 GHz (~0.2
cm-1), and the precision of the frequency calibration is ~120 MHz (~0.004 cm-1). The residual of
6
Preprint
August 2, 2019
the fit is shown in Fig. 2b, indicating the good agreement between the measurement and the
model.
Fig. 2. Measured spectrum without discharge. (a) Normalized transmission
spectrum of 50% CH4 diluted in He at 25 mbar total pressure and room
temperature (black, 500 averages), along with a fit model spectrum of CH4 (in
blue, inverted and offset for clarity) using HITRAN database parameters. (b)
Residual of the fit.
Methane spectrum in a static discharge. We measured the normalized transmission spectrum
of the aforementioned gas sample (50% CH4 in He, 25 mbar) in a static discharge. We applied a
DC voltage of 10 kV and a stabilized current of 10 mA (current density 1.4 mA/mm2) to the
discharge tube and kept a constant sample gas flow of 4 Nl/h through the cell. Figure 3a shows
the measured normalized spectrum of the sample with the discharge (in red, 500 averages, ~2 s
measurement time) compared to the measured normalized spectrum of the sample without the
discharge (in black, 500 averages, ~2 s measurement time). Note that, neither a baseline
7
Preprint
August 2, 2019
correction nor an etalon-fringes removal process is applied to these spectra. In the discharge the
CH4 absorption is reduced, which indicated a lower population in the vibrational ground state,
due to vibrational excitation, ionization and molecular dissociations. This is accompanied by the
appearance of two groups of additional absorption lines in the spectrum with the discharge,
enlarged in Fig. 3b and Fig. 3c. The additional absorption lines in Fig. 3b are due to the produced
C2H6 in the discharge. Since the C2H6 model in the HITRAN database is incomplete, we
measured the room temperature absorption spectrum of 10% C2H6 diluted in He (25 mbar total
pressure) in order to obtain a proper reference spectrum for C2H6. We fit the retrieved reference
spectrum to the recorded discharge spectrum, with the concentration of C2H6 as the fitting
parameter. This fitted spectrum is shown in Fig. 3b (in blue, inverted for clarity), and the
retrieved C2H6 concentration from the fit is 6.52(11)%. Note that, we excluded few C2H6
absorption lines from the broadband fitting routine, since they showed deviation from the
corresponding absorption lines in the reference spectrum, most probably due to differences in
number densities at these levels caused by the discharge. The second group of additional
absorption lines, indicated by "*" in Fig. 3c, are due to vibrational hot band transitions in CH4,
since CH4 molecules are excited in the discharge. Many more (weaker) hot band absorption lines
appear in all three P-, Q-, and R- branches, but are not highlighted in the figure. The first
vibrational excited state of CH4 is ν4 (energy level ~1310 cm-1) followed by ν2 (energy level
~1533 cm-1). The majority of the detected hot band absorption lines undergo a vibrational ν3+ν4
← ν4 transition and a smaller portion originate from the ν3+ν2 ← ν2 transition. All hot band ro-
vibrational transitions can be assigned using the HITRAN database.
8
Preprint
August 2, 2019
9
Preprint
August 2, 2019
Fig. 3. Methane spectrum in the discharge. (a) Normalized spectrum with the
discharge (in red, 500 averages) compared to the normalized spectrum without the
discharge (in black, 500 averages). (b) An enlargement to the absorption lines of
C2H6 produced in the discharge compared to a fitted room temperature spectrum
of C2H6 (in blue, inverted). (c) An enlargement to a number of stronger rotational
lines (indicated by '*') within the hot band transitions of CH4.
Time-resolved measurement at milliseconds time scale. To perform time-resolved
measurements, we modulated the current of the discharge, with a square-wave function in an
"off" (dark) and "on" (glow) regime (20% duty cycle for "on" time). The current modulation was
synchronized with the repetition rate difference (∆frep ≈ 250 Hz) of the dual-comb spectrometer,
and the modulation frequency, fmod, was chosen to be equal to ∆frep divided by 100, i.e. fmod =
∆frep/100 (≈ 2.5 Hz). Therefore, each period of the discharge modulation was recorded by a set of
100 consecutive interferograms, each at its own time-bin. This allowed averaging of the spectra
(after Fourier transform of the interferograms) for each corresponding time-bin of different
discharge cycles to achieve high SNR averaged spectra. Therefore, we monitored each period of
the discharge modulation with a time resolution equal to the inverse of ∆frep, i.e. Tres = 1/∆frep (≈
4 ms). We obtained the normalized spectra of the gas sample (50% CH4 in He, 25 mbar, flow
rate 4 Nl/h) over the entire period of the modulation (400 ms) with a time-resolution of 4 ms.
Figure 4a shows the retrieved concentrations of the generated C2H6, and Fig. 4b demonstrates the
absorbance of R(7) line at ~3082.2 cm-1 corresponding to the vibrational ν3+ν4 ← ν4 hot band
transition of CH4, as an indicator of the number of excited CH4 molecules. The discharge was
turned on and off at t = 0 ms and t = 80 ms, respectively. Each data point is averaged for 250
10
Preprint
August 2, 2019
times (recording time of 1 s) and the total measurement time for the entire data set is ~100 s.
Note that we began to record the data after the modulation was applied for a few minutes, in
order to avoid any possible transient conditions from a static to a dynamic regime.
As shown in Fig. 4a, after turning the discharge on at t = 0 ms, the concentration of C2H6
increases from 0.71% to 2.4% in ~20 ms and reaches to a pseudo-plateau region. The absorbance
of the hot band line abruptly appears at t = 0 ms, as shown in Fig. 4b, not resolvable with the
current (4 ms) time resolution. After this abrupt appearance, the absorbance demonstrates a
decrease to half of its initial value in ~20 ms and reaches to a pseudo-plateau region. The
comparable time scale in the increase of the C2H6 concentration and decrease of the hot band
absorbance suggests that these two processes are likely to be correlated. After the discharge is
turned off at t = 80 ms, the C2H6 concentration increases instantaneously from 2.5% to 6.7%,
while the hot band transition disappears. None of these two processes can be resolved with 4 ms
time resolution. After the sudden increase, the C2H6 concentration decreases linearly, within the
gas refresh time of the long and narrow discharge cell, before it reaches a pseudo-plateau region
slowly decreasing from 1.0% to 0.71% by the end of the cycle. The latter demonstrates the
remaining C2H6 molecules in the discharge cell most probably due to purging inhomogeneity.
11
Preprint
August 2, 2019
Fig. 4. Time-resolved measurement at milliseconds time scale. (a) Concentrations
of generated C2H6 and (b) absorbance of the R(7) rotational line at ~3082.2 cm-1
in the vibrational ν3+ν4 ← ν4 hot band of CH4, with 4 ms time resolution, recorded
while the discharge is switched between dark and glow regimes. Different states
of the discharge are separated by red dashed lines and are indicated by red "On"
and "Off" labels.
Time-resolved measurement at microseconds time scale. To perform time-resolved
measurement at microseconds time scale, the square-wave current modulation was synchronized
with ∆frep, with an equal frequency fmod = ∆frep (≈ 250 Hz), while the current modulation could be
deliberately time-delayed (phase-shifted) with respect to ∆frep. For each particular time-delay the
consecutive interferograms were recorded and directly averaged after Fourier transform to yield
a high SNR spectrum. To cover the period of the current modulation, the time-delay was step-
scanned and an averaged spectrum was measured for each step. In this configuration the time
12
Preprint
August 2, 2019
resolution is equal to the time-delay steps (e.g. 20 μs); however, for rapid changes happening
faster than the measurement time of each individual interferogram (120 μs), the measurement
results would be smoothened by a moving average. We obtained the normalized spectra of 50%
CH4 in He (25 mbar, flow rate 4 Nl/h) over the entire discharge modulation period of 4 ms
(square-wave function, ~10% duty cycle for "on" time). The step size was 20 μs near the
switching events and 50 μs far from the switching events. Figure 5a shows the generated C2H6
concentrations and Fig. 5b demonstrates the absorbance of the same CH4 hot band line of R(7) at
~3082.2 cm-1, that was previously monitored in the millisecond time scale. Data are shown for a
time span of 1 ms around the 400 μs period that the discharge was on. Each data point is
averaged for 250 times (recording time of 1s) and the total measurement time for the shown data
set is ~42 s, excluding the standby time for varying the time-delay.
In contrast to the results obtained in the milliseconds time scale, no discontinuity of the
monitored parameters is observed in the microseconds time scale measurements. Note that after
applying the current modulation to the discharge, the system operated for a few minutes before
recording the data, to avoid any possible transient conditions from a static to a dynamic regime.
Due to the high frequency of the discharge modulation the gas flow is insufficient to purge the
generated C2H6 on each modulation cycle, which leads to an accumulation of C2H6 up to a
concentration of 7.7% just before the discharge is turned on, as shown in Fig. 5a. When the
discharge is turned on at t = 0 μs, the C2H6 concentration initially decreases rapidly to 3.0% in
~100 μs, after which it increases slightly to 3.2% in the next 300 μs. This slight increase reflects
early-time dynamics, observed at the milliseconds time scale measurements right after turning
the discharge on. In Fig. 5b, the absorbance of the monitored hot band line demonstrates a rapid
increase to a comparable amplitude that has been observed with the milliseconds time scale
13
Preprint
August 2, 2019
measurements. The rise time is ~60 μs, which is faster than the observed fall time of the C2H6
concentration. After the rapid formation of the hot band line, the absorbance amplitude is slowly
decreased from ~7.5×10-3 cm-1 to ~6.5×10-3 cm-1, right before the discharge is turned off. This is
in agreement with the observed decay dynamics in the milliseconds time scale measurement,
following the initial abrupt increase. After the discharge is turned off, the C2H6 concentration
increases from 3.2% to 7.6 % in ~500 μs. Meanwhile, the absorbance of the monitored hot band
line decreases to zero in ~200 μs. The dynamics of the vibrational excited CH4 and the generated
C2H6 are opposing each other, with the former slightly faster than the latter. The dynamics of
C2H6 implies that when the discharge is turned on, it dissociates the C2H6 into free radicals, such
as C2H5 and CH3, while after the discharge is turned off these radicals recombine, and amongst
others, form C2H6. This interpretation also explains the abrupt increase of C2H6 concentration
after the discharge was turned off in the milliseconds time scale measurements (Fig 4b at t = 80
ms).
At these pressures in the discharge, we can also observe that the dissociation (formation) of
C2H6, to (by recombination of) free radicals is slower than vibrational excitation (de-excitation)
of CH4 molecules. The vibrational de-excitation of methane is mainly due to the collisions. We
measured the collisional de-excitation decay time of different rotational lines in the hot bands, by
fitting exponential functions to the absorbance of the hot band lines (after the discharge is turned
off).
14
Preprint
August 2, 2019
Fig. 5. Time-resolved measurement at microseconds time scale. (a)
Concentrations of generated C2H6 and (b) absorbance of the R(7) rotational line at
~3082.2 cm-1 in the vibrational ν3+ν4 ← ν4 hot band of CH4, with 20 μs time
resolution, recorded while the discharge is switched between dark and glow
regime. Different states of the discharge are separated by red dashed lines and are
indicated by red "On" and "Off" labels.
Figure 6 shows the corresponding measurements and fits for absorbance of three rotational hot
band lines, R(2) at ~3043.9 cm-1 and R(7) at ~3082.2 cm-1 corresponding to the vibrational ν3+ν4
← ν4 hot band transition, and R(8) at ~3091.4 cm-1 corresponding to the vibrational ν3+ν2 ← ν2
hot band transition. The hot band ro-vibrational transitions were assigned using the HITRAN
database. The collisional de-excitation decay times retrieved from the fits are 57(5) μs, 52(3) μs,
and 54(2) μs, respectively. Although the decay times are shorter than the measurement time of
each individual interferogram (120 μs), they are not affected by the moving average, since the
15
Preprint
August 2, 2019
shape and decay rate of an exponential function will not be affected by integration. This is also
evident by the good agreement between the measurements and the fit exponential functions.
Fig. 6. Collisional de-excitation of three hot band lines. The rotational lines are
R(2) at ~3043.9 cm-1 (red squares) and R(7) at ~3082.2 cm-1 (blue diamonds) in
the vibrational transition of ν3+ν4 ← ν4, as well as R(8) at ~3091.4 cm-1 (black
circles) in the vibrational transition of ν3+ν2 ← ν2 of CH4. The time-resolved
absorbance is demonstrated along with an exponential fit to each data set. The
retrieved decay times form the fits are shown in the figure legend.
Conclusion
We demonstrated, for the first time, the capabilities of time-resolved dual-comb spectroscopy
(TRDCS) in the mid-IR wavelength range. For this, we studied the vibrational excitation and de-
excitation of CH4 in a modulated electrical discharge and its reaction product C2H6 at
milliseconds and microseconds time scales. The total acquisition time for each measurement was
16
Preprint
August 2, 2019
in the order of tens of seconds. The spectrometer covered a wavelength range from ~3.1 μm to
~3.4 μm with a spectral resolution of 6 GHz and a single shot acquisition time of ~120 μs. By
tuning the OPO light source, it is possible to monitor the other wavelength ranges from 2.7 μm to
4.7 μm. In this initial demonstration of TRDCS, we observed the products at %-levels due to the
relative short interaction path length. The detection sensitivity of the spectrometer can be
enhanced by using a multipass arrangement or an enhancement resonant cavity. Higher detection
sensitivity, in combination with the broad spectral tunability of the OPO, provides the possibility
to monitor less abundant interesting species in the discharge process in different wavelength
ranges, e.g. free radicals, transient species and ions. In the dynamics, a wealth of information
becomes available, which we only scratched the surface in this first demonstration. A
comprehensive time-resolved study of a discharge process for the kinetics of reactions and
branching next to temperature information could be feasible. The system can also be used for
studying any fast chemical reaction that can be periodically triggered with an acceptable level of
reproducibility. Recently, TRDCS in the near-IR wavelength range has been utilized to monitor a
fast, single shot reaction with millisecond time scale resolution 31. The results are promising,
although they are limited to high pressure spectroscopy for the time being. Finally, it should be
noted that DCS has an inherent trade-off between the measurement time and the spectral
resolution, which provides flexibility to choose a proper combination for each particular
application. The minimum measurement time can be deliberately selected to be the shortest time
with which the spectral resolution is still sufficient for detecting the desired species and/or
resolving the target spectral features.
17
Preprint
Methods
August 2, 2019
Optical setup. The two mid-IR frequency combs are generated from a singly-resonant optical
parametric oscillator (OPO) based on two MgO:PPLN crystals (Covesion Ltd., UK) positioned
in a common OPO cavity. Each crystal is synchronously pumped by a Yb:fiber mode-locked
laser (Menlo Systems, Germany) emitting around 1040 nm. The pump beams are counter
propagating in the ~3.3 m-long OPO cavity and are perpendicularly polarized. The repetition
rates (frep) of the mode-locked lasers are ~90 MHz and stabilized to a common reference clock
but are slightly different (∆frep ≈ 250 Hz), yielding the repetition rate difference in the generated
idlers combs. The carrier-envelop offset frequency (fceo) of the two pump mode-locked lasers as
well as the OPO cavity length are not stabilized. Each idler beam can provide up to 200 mW of
average power, covering a wavelength range of around 350 nm, and is tunable from 2.7 to 4.7
μm using different poling periods in the nonlinear crystals. The two idler beams (after
polarization adjustment) are combined by a beam splitter to produce two pair of beams on
reflection and transmission. One pair is transmitted through the discharge tube, filled with CH4
diluted in He at a total pressure of 25 mbar. The transmitted beams are focused on a fast (50
MHz) thermoelectrically cooled HgCdTe photodetector (PVI-4TE, Vigo System, Poland),
detecting the down-converted RF interferogram. The second pair is transmitted through a
reference absorption cell, containing pure CH4 at 100 mbar, and is dispersed by a diffraction
grating. A small part of the dispersed spectrum, containing a single well-known absorption line
of CH4 (at 3038.498 cm-1), is focused on a second HgCdTe photodetector (PVI-4TE, Vigo
System, Poland). The outputs of the two detectors are recorded by a two channel 125 MSam/s,
16 bits, analog-to-digital convertor (NI-5762, National Instruments, US) and saved on a
18
Preprint
August 2, 2019
computer for data processing. A common 10 MHz clock is used to synchronize the dual-comb
spectrometer, the data acquisition, and the modulation of the discharge current.
Data processing and averaging. Each of the recorded interferograms is Fourier transformed to
reconstruct the spectrum. The free running carrier envelope offset frequencies and unstabilized
OPO cavity length reduce the complexity of the experimental setup, but they cause a changing
frequency shift in consecutive recorded spectra. Since the fluctuations of the offset frequencies
and the OPO cavity length are negligible in the measurement time of a single interferogram (120
μs), i.e. the two idler combs are coherent in this time scale, a linear frequency shift is sufficient
to correct for these changes in each single measurement. To address this frequency shift, we use
the frequency position of the known reference absorption line and correct the frequency shift of
each individual spectrum before averaging, which also yields an absolute optical frequency
calibrated spectrum. The spectra are averaged after the shift correction.
Discharge setup. The Pyrex discharge tube is ~50 cm long, with an internal diameter of 3 mm
and it is water cooled. The hollow cathodes are at the two ends of the tube and the anode is at the
center. A current-stabilized high-voltage (HV) power supply (Haefely Hipotronics, US,
providing up to 25 kV, 40 mA) is used for generating a DC discharge in the tube. The power
supply is able to limit the current overshoots during switching/modulation of the discharge. The
discharge can be switched on (glow) and off (dark) by modulating the current of HV power
supply using a signal generator, whose clock is synchronized with the repetition rate different
(∆frep) of the dual-comb spectrometer. The modulation signal is a square-wave whose frequency,
duty cycle, and time-delay (with respect to the ∆frep) can be adjusted independently. Therefore,
the data acquisition and discharge process are synchronized and can also be programmed to have
a time delay with respect to each other.
19
Preprint
Acknowledgments
August 2, 2019
This work was financially supported by Dutch Technology Foundation (NWO, 11830) and EU
H2020-ICT29 (FLAIR project, 732968). The authors thank David H. Parker and Giel Berden for
useful comments on the manuscript.
Correspondence
Correspondence should be addressed to Amir Khodabakhsh ([email protected])
References
Murphy, R. E., Cook, F. H. & Sakai, H. Time-resolved Fourier Spectroscopy. J. Opt. Soc. Am. 65,
600-604, (1975).
Uhmann, W., Becker, A., Taran, C. & Siebert, F. Time-resolved FT-IR absorption-spectroscopy
using a step-scan interferometer. Appl. Spectrosc. 45, 390-397, (1991).
Jiang, E. Y. Phase-, time-, and space-resolved step-scan FT-IR spectroscopy - Principles and
applications to dynamic and heterogeneous systems. Spectroscopy 17, 22-34, (2002).
Iwata, K. & Hamaguchi, H. O. Construction of a versatile microsecond time-resolved infrared
spectrometer. Appl. Spectrosc. 44, 1431-1437, (1990).
Yuzawa, T., Kato, C., George, M. W. & Hamaguchi, H. O. Nanosecond time-resolved infrared-
spectroscopy with a dispersive scanning spectrometer. Appl. Spectrosc. 48, 684-690, (1994).
Cunge, G., Vempaire, D., Touzeau, M. & Sadeghi, N. Broadband and time-resolved absorption
spectroscopy with light emitting diodes: Application to etching plasma monitoring. Appl. Phys.
Lett. 91, 231503, (2007).
Matsugi, A., Shiina, H., Oguchi, T. & Takahashi, K. Time-resolved broadband cavity-enhanced
absorption spectroscopy behind shock waves. J. Phys. Chem. A 120, 2070-2077, (2016).
Adler, F. et al. Mid-infrared Fourier transform spectroscopy with a broadband frequency comb.
Opt. Express 18, 21861-21872, (2010).
Zhang, Z. W., Gardiner, T. & Reid, D. T. Mid-infrared dual-comb spectroscopy with an optical
parametric oscillator. Opt. Lett. 38, 3148-3150, (2013).
Jin, Y. W., Cristescu, S. M., Harren, F. J. M. & Mandon, J. Two-crystal mid-infrared optical
parametric oscillator for absorption and dispersion dual-comb spectroscopy. Opt. Lett. 39, 3270-
3273, (2014).
Baumann, E. et al. Spectroscopy of the methane ν3 band with an accurate midinfrared coherent
dual-comb spectrometer. Phys. Rev. A 84, 9, (2011).
Nugent-Glandorf, L. et al. Mid-infrared virtually imaged phased array spectrometer for rapid and
broadband trace gas detection. Opt. Lett. 37, 3285-3287, (2012).
Galli, I. et al. Mid-infrared frequency comb for broadband high precision and sensitivity
molecular spectroscopy. Opt. Lett. 39, 5050-5053, (2014).
Khodabakhsh, A., Rutkowski, L., Morville, J. & Foltynowicz, A. Mid-infrared continuous-filtering
Vernier spectroscopy using a doubly resonant optical parametric oscillator. Appl. Phys. B-Lasers
Opt. 123, 12, (2017).
20
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Preprint
August 2, 2019
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Iwakuni, K., Bui, T. Q., Niedermeyer, J. F., Sukegawa, T. & Ye, J. Comb-resolved spectroscopy
with immersion grating in long-wave infrared. Opt. Express 27, 1911-1921, (2019).
Weichman, M. L. et al. Broadband molecular spectroscopy with optical frequency combs. J. Mol.
Spectrosc. 355, 66-78, (2019).
Coddington, I., Swann, W. C. & Newbury, N. R. Time-domain spectroscopy of molecular free-
induction decay in the infrared. Opt. Lett. 35, 1395-1397, (2010).
Lomsadze, B. & Cundiff, S. T. Frequency combs enable rapid and high-resolution
multidimensional coherent spectroscopy. Science 357, 1389-1391, (2017).
Kim, J., Cho, B., Yoon, T. H. & Cho, M. Dual-frequency comb transient absorption: broad dynamic
range measurement of femtosecond to nanosecond relaxation processes. J. Phys. Chem. Lett. 9,
1866-1871, (2018).
Bergevin, J. et al. Dual-comb spectroscopy of laser-induced plasmas. Nat. Commun. 9, 6, (2018).
Reber, M. A. R., Chen, Y. N. & Allison, T. K. Cavity-enhanced ultrafast spectroscopy: ultrafast
meets ultrasensitive. Optica 3, 311-317, (2016).
Zhang, Y. et al. Time-resolved dual-comb measurement of number density and temperature in a
laser-induced plasma. Opt. Lett. 44, 3458-3461, (2019).
Fleisher, A. J. et al. Mid-infrared time-resolved frequency comb spectroscopy of transient free
radicals. J. Phys. Chem. Lett. 5, 2241-2246, (2014).
Bjork, B. J. et al. Direct frequency comb measurement of OD + CO -> DOCO kinetics. Science 354,
444-448, (2016).
Bui, T. Q. et al. Direct measurements of DOCO isomers in the kinetics of OD + CO. Sci. Adv. 4, 8,
(2018).
Jin, Y. W., Cristescu, S. M., Harren, F. J. M. & Mandon, J. Femtosecond optical parametric
oscillators toward real-time dual-comb spectroscopy. Appl. Phys. B-Lasers Opt. 119, 65-74,
(2015).
Abbas, M. A. et al. Mid-infrared dual-comb spectroscopy with absolute frequency calibration
using a passive optical reference. Opt. Express 27, 19282-19291, (2019).
Ycas, G. et al. High-coherence mid-infrared dual-comb spectroscopy spanning 2.6 to 5.2 mu m.
Nat. Photonics 12, 202-208, (2018).
Muraviev, A. V., Smolski, V. O., Loparo, Z. E. & Vodopyanov, K. L. Massively parallel sensing of
trace molecules and their isotopologues with broadband subharmonic mid-infrared frequency
combs. Nat. Photonics 12, 209-214, (2018).
Gordon, I. E. et al. The HITRAN2016 molecular spectroscopic database. J. Quant. Spectrosc.
Radiat. Transf. 203, 3-69, (2017).
Draper, A. D. et al. Broadband dual-frequency comb spectroscopy in a rapid compression
machine. Opt. Express 27, 10814-10825, (2019).
21
|
1710.05769 | 1 | 1710 | 2017-10-16T15:06:44 | Electonic transport properties of nitrate-doped carbon nanotube networks | [
"physics.app-ph",
"cond-mat.mes-hall"
] | The conductivity of carbon nanotube (CNT) networks can be improved markedly by doping with nitric acid. In the present work, CNTs and junctions of CNTs functionalized with NO$_3$ molecules are investigated to understand the microscopic mechanism of nitric acid doping. According to our density functional theory band structure calculations, there is charge transfer from the CNT to adsorbed molecules indicating p-type doping. The average doping efficiency of the NO$_3$ molecules is higher if the NO$_3$ molecules form complexes with water molecules. In addition to electron transport along individual CNTs, we have also studied electron transport between different types (metallic, semiconducting) of CNTs. Reflecting the differences in the electronic structures of semiconducting and metallic CNTs, we have found that besides turning semiconducting CNTs metallic, doping further increases electron transport most efficiently along semiconducting CNTs as well as through a junction between them. | physics.app-ph | physics |
Electronic transport properties of nitrate-doped carbon nanotube networks
COMP, Department of Applied Physics, Aalto University, P.O. Box 11100, FI-00076 Aalto, Finland
T. Ketolainen,∗ V. Havu, E. O. J´onsson, and M. J. Puska
(Dated: July 24, 2018)
The conductivity of carbon nanotube (CNT) networks can be improved markedly by doping with
nitric acid. In the present work, CNTs and junctions of CNTs functionalized with NO3 molecules
are investigated to understand the microscopic mechanism of nitric acid doping. According to our
density functional theory band structure calculations, there is charge transfer from the CNT to
adsorbed molecules indicating p-type doping. The average doping efficiency of the NO3 molecules is
higher if the NO3 molecules form complexes with water molecules. In addition to electron transport
along individual CNTs, we have also studied electron transport between different types (metallic,
semiconducting) of CNTs. Reflecting the differences in the electronic structures of semiconducting
and metallic CNTs, we have found that besides turning semiconducting CNTs metallic, doping
further increases electron transport most efficiently along semiconducting CNTs as well as through
a junction between them.
I.
INTRODUCTION
Flexible carbon nanotube (CNT) thin films are ma-
terials, where randomly oriented CNTs form a network
structure1. The fabrication methods of CNT thin films
have improved remarkably and several novel applications
of the CNT networks have been introduced. It may be
even possible to replace the commonly used indium tin
oxide films by CNT thin films. Advantages of the CNT
thin films are their low absorption in a broad range of
optical wavelengths2 and a possibility to bend the films
without lowering the conductivity significantly3.
In applications, CNT thin films with high electrical
conductivity and optical transparency are required. For
this purpose, a significant enhancement of the conduc-
tivity of the CNT thin films has been found, for exam-
ple, in the case of AuCl3 doping4. Computational stud-
ies have revealed that formation of AuCl4 molecules or
anions on semiconducting CNTs leads to a p-type dop-
ing effect, i.e. to electron transfer to acceptor molecules
leaving behind holes in the substrate CNTs5. Further-
more, in our recent study we have found a significant
enhancement of the electron transport through a junc-
tion of semiconducting CNTs due to the AuCl4 doping
and the improvement has been shown to be robust with
respect to the molecular concentration provided that it
is large enough6. Recently, iodine monochloride and io-
dine monobromide have also been used to dope CNT thin
films, indicating a remarkable reduction of their electrical
resistivity7. Another example of improving the conduc-
tivity of CNT thin films is doping them with I chains
or CuCl as presented experimentally in Ref.8. More-
over, previous computational studies of CNT junctions
with transition metal linker atoms have shown enhanced
junction conductances9.
A common method to improve the conductivity of
CNT thin films is to dope them with nitric acid (see, for
example, Refs.10,11, and12). Experimentally, the nitric-
acid doping has been shown to reduce the junction re-
sistances as well as increase the conductivity of indi-
vidual CNTs in the network13,14. However, the mech-
anism behind the nitric-acid doping of CNTs is not well-
known.
It is assumed to be related to nitrogen oxide
molecules bound to CNTs. The adsorption of various
NOx molecules on CNTs has been studied using density
functional theory (DFT) in order to understand their de-
tection in gas sensing applications15. Computationally,
it has also been found that NO2 molecules can form NO3
molecules and the binding energy is higher in the case
of molecular pairs16. According to previous experimen-
tal and computational studies, the strength and type of
the adsorption of nitrogen oxides are also affected by the
metallicity of the CNT17.
In addition to the chirality,
recent DFT calculations of the nitric-acid treated CNTs
have shown that the charge state of the CNT also influ-
ences the adsorption process18.
The electronic structure and transport properties of
CNT junctions need to be considered in more detail to
understand the effect of nitric acid on the conductiv-
ity of CNT networks. Earlier experimental studies have
shown that the resistance of the CNT junction depends
largely on the chiralities of the CNTs and the resistance
may be especially high when a metallic and a semicon-
ducting CNT form a junction with a Schottky barrier19.
Furthermore, CNT junctions with physisorbed O2 and
N2 molecules have been investigated previously experi-
mentally and theoretically20. According to the compu-
tational part of that work, molecules physisorbed in the
vicinity of the junctions lead to an improvement of the
conductance of the junction, which has been explained to
result from improved tunneling probability due to hop-
ping via molecular orbitals20.
In this work, we consider the influence of nitric acid
on CNT networks by studying CNTs with adsorbed NO3
molecules and junctions of NO3-doped CNTs. On the
basis of simple electron counting, NO3 molecules are ex-
pected to work as acceptors (or anions). More specifi-
cally, optimal geometries, band structures, and electronic
transmission functions for NO3-doped CNTs are deter-
mined using DFT. The DFT calculations reveal that the
physisorbed NO3 molecules receive electrons from the
CNT resulting in a downshift of the CNT Fermi level
and ensuing p-type doping of the CNT. Intratube as well
as intertube conductances achieved in doping depend on
the chiralities of the CNTs. Semiconducting CNTs be-
come conductive and the electron transport can be fur-
ther efficiently increased by lowering the Fermi level be-
low van Hove singularities when they are doped with NO3
molecules. Moreover, junctions of two NO3-doped semi-
conducting CNTs with a large number of molecules show
a clearly improved conductance so that the improvement
can be a factor of ten or even more. The doping of metal-
lic CNTs lowers the intratube conductances a little de-
pending on the molecular concentration. However, the
doping does not lower the Fermi level as efficiently to-
ward the van Hove singularities of the CNT electronic
structure as in the case of semiconducting CNTs. This
diminishes the effect of doping on the intratube and in-
tertube conductances. The clearly dissimilar electronic
structures of semiconducting and metallic CNTs hinder
to enhance the conductance of their junctions by dop-
ing. Similarly to the predictions for complex molecular
systems21, we have found that water molecules coordi-
nated to NO3 molecules enhance the doping effect both
in semiconducting and in metallic CNTs.
The structure of this article is as follows. The main
aspects of the methods and investigated systems are pre-
sented in Sec. II. Then the geometries and band struc-
tures with the ensuing doping efficiencies are discussed
at the beginning of Sec. III. The essential topics at the
end of this section are electron transport in NO3-doped
CNTs and CNT junctions. A brief summary of the re-
sults is given in Sec. IV.
II. METHODOLOGY AND SYSTEMS
The calculations presented are carried out using two
different electronic-structure codes, FHI-aims22 -- 24 and
GPAW25,26. We use the FHI-aims code to perform
most of the calculations apart from calculations including
the Perdew-Zunger self-interaction correction (SIC)27.
The latter calculations are carried out with the GPAW
electronic-structure code package. The FHI-aims code
package is based on numeric atom-centered orbitals
and also provides tools for studying transport proper-
ties of materials. Two exchange-correlation functionals
are used, namely the generalized-gradient approximation
functional PBE28 and the hybrid functional HSE0629.
Since van der Waals interactions can be significant in
nanocarbon systems, the Tkatchenko-Scheffler van der
Waals correction is applied together with PBE in all
atomic-structure optimizations in this work30. Charge
transfer in NO3-doped CNT systems is investigated also
with GPAW by applying the Perdew-Zunger SIC to DFT
calculations27,31. In this case, the exchange-correlation
functional is chosen to be PW9132. The PW91 energy
functional is corrected by subtracting a SIC term that is
scaled by a factor of 1/2 (for details, see Ref.31).
The effect of nitric-acid doping is considered in
2
infinitely long NO3-doped semiconducting (10,0) and
metallic (8,8) CNTs by computing band structures for
the computational unit cells depicted in Figs. 1(a) -- (d).
Our basic doped CNT system has only one NO3 molecule
in the computational unit cell of either a (10,0) or an (8,8)
CNT. Higher doping concentrations are investigated by
increasing the number of NO3 molecules in the basic sys-
tem. The concentration of NO3 molecules in the inves-
tigated systems varies between 0.06 and 0.47 molecules
per A. Further, the influence of water on the doping ef-
ficiency is studied using the systems shown in Figs. 1(b)
and 1(d), where a NO3 molecule lying on the CNT is
coordinated to three H2O molecules.
FIG. 1. Computational unit cells of (a) a semiconducting
(10,0) zigzag CNT with one NO3 molecule, (b) a NO3-doped
(10,0) CNT with three H2O molecules, (c) a metallic (8,8)
armchair CNT with one NO3 molecule, and (d) an (8,8) CNT
with one NO3 molecule coordinated to three H2O molecules.
The computational (10,0) unit cells comprise either two or
four primitive unit cells. The computational (8,8) unit cell
contains seven primitive unit cells.
The geometries of the primitive unit cells of (10,0) and
(8,8) CNTs are first optimized with the FHI-aims code.
The initial positions of the carbon atoms are computed
by using the TubeGen 3.4 nanotube generator33 and the
whole primitive unit cell including the lattice vectors is
relaxed.
In addition, the structures of the molecules
added on the CNT are relaxed separately. Thereafter, a
NO3-doped CNT system is constructed with the help of
the relaxed molecule and primitive CNT unit cells. The
lengths of the computational unit cells of the (8,8) and
(10,0) systems are 17.3 and 17.1 A, respectively. This
means that the computational unit cells of the (8,8) and
(10,0) systems comprise seven and four primitive CNT
unit cells, respectively. In the case of the doped (10,0)
system with H2O molecules, the computational unit cell
is shorter, around 8.5 A. This system has to have a
smaller number of atoms than in the other CNT struc-
tures so that Perdew-Zunger SIC calculations can be car-
ried out. Periodic boundary conditions are used and the
distance between the axes of the CNTs in the neighbor-
ing unit cells is 30.0 A. The relaxation is performed with
PBE and stopped when all the force components have
decreased below 10−2 eV/A. A 1 × 1 × 27 k-point grid
is used when the computational unit cells of the doped
systems are relaxed. The calculations of semiconducting
(10,0) CNTs with NO3 molecules can include spin since
in some cases it gives a ground state with a lower energy
than a non-spin-polarized system.
The binding energy Eb for a NO3 molecule on a CNT
can be expressed as
Eb = ET[CNT-NO3] − ET[CNT] − ET[NO3],
(1)
where ET[CNT-NO3], ET[CNT], and ET[NO3] are the
computational unit cell total energies of the CNT with
one NO3 molecule, the pristine CNT, and the isolated
NO3 molecule, respectively. If the binding energy is de-
termined for a doped system with water molecules, the
total energy of the NO3 molecule in Eq. (1) has to be
replaced by that of the molecular complex.
Electron transport in infinitely long homogeneously
doped CNTs and in junctions of CNTs at the zero-bias
limit is investigated using the transport module of the
FHI-aims code. Two- or four-terminal transport sys-
tems can be divided into the central scattering region
and semi-infinite leads. First, the electronic structure
of the transport system relaxed with PBE and includ-
ing the van der Waals correction has to be determined.
Then one computes the electronic transmission function
by solving the Green's function consistent with the PBE
exchange-correlation potential for the CNT system34. A
significant challenge is aligning the energy levels of the
semi-infinite leads and their counterparts in the computa-
tional transport unit cell. This alignment is discussed in
Refs.6 and35. The k-point grid used for finding the elec-
tronic structure in the transport calculations is 1 × 1 × 27
for individual metallic (8,8) CNTs. A sparser grid is used
for the transport calculations of semiconducting CNTs
but the calculations are converged despite the different
grid. In the case of a CNT junction, the k-point mesh is
1 × 3 × 3.
III. RESULTS AND DISCUSSION
A. Semiconducting CNTs with NO3 molecules
Relaxation of the computational unit cell of a semicon-
ducting (10,0) CNT with one NO3 molecule is carried out
first with spin. After the relaxation, the center of the
NO3 molecule (the N atom) lies above a carbon atom.
Our results are consistent with those of Peng et al.15.
Their results have been computed using the local spin-
density approximation (LSDA). In a more recent study
by Kroes et al.36, the optimal geometry is different so
that the NO3 molecule has moved a bit from the top
configuration.
The distance between the CNT and the NO3 molecule
dCNT-NO3
is defined as a distance between the nitrogen
atom and the carbon atom to which the NO3 molecule
3
binds. With the van der Waals correction, the distance
between the NO3 molecule and the (10,0) CNT is 3.09 A.
This value is significantly larger than the distance 2.87
A obtained in an LSDA calculation. The overbinding
of the LSDA is a known issue and is largely due to the
rapid (exponential) decay of the LSDA potential. On
the other hand, the CNT-NO3 distance is close to the
value 3.08 A determined by Kroes et al. by using van
der Waals corrected DFT36. The binding energy for one
NO3 molecule on a (10,0) CNT calculated using Eq. (1)
is −0.80 eV in good agreement with the value found in
Ref.36. Semiconducting (10,0) CNTs doped with NO3-
H2O complexes are also examined but the computational
unit cell (see Fig. 1(b)) is shorter than that used for (10,0)
CNTs with only NO3 molecules. A detailed description
of these systems is given in the appendix.
The band structures of pristine and NO3-doped (10,0)
CNTs are shown in Figs. 2(a) -- (c) along with the densi-
ties of states (DOSs) for the pristine (Fig. 2(a)) and the
most doped (Fig. 2(c)) CNTs. The system with either six
or eight NO3 molecules is investigated without relaxing
the atomic structure except for the CNT-NO3 distance
because the molecules do not affect the CNT structure
remarkably and the interaction between the molecules is
small. The pristine (10,0) CNT has a direct band gap of
0.77 eV at the Γ point in the case of PBE as shown in
Fig. 2(a).
A downshift of the Fermi level in the band structures
of NO3-doped (10,0) CNTs in Figs. 2(b) and 2(c) is ob-
served.
In addition, there is a dispersionless molecu-
lar state just below the Fermi level in the band struc-
tures shown in Figs. 2(b) and 2(c). A previous study
of graphene doped with nitric acid has proposed that
the work function of the half-filled orbital of the NO3
molecule is greater than that of graphene and therefore
the molecular state lies below the Fermi energy37. Fur-
thermore, the Fermi level in the highly doped system in
Fig. 2(c) is pinned to a van Hove singularity and the flat
molecular state. This is similar to our previous observa-
tion in the case of AuCl4 doping and it is an important
ingredient for establishing a robust enhanced intertube
conduction6.
We estimate the charge transfer in the NO3-doped sys-
tems by using the method described in Ref.6.
In this
method, the crossing points of the Fermi level and the
valence bands are first searched for. These points divide
the bands into occupied and unoccupied parts. There-
after, by measuring the width of the unoccupied regions
and dividing this value by the maximum kz wave vector,
an estimate for the magnitude of the charge transfer can
be obtained when the orbital and spin degeneracies are
taken into account.
The doping efficiency for one NO3 molecule in the ba-
sic computational unit cell,
the electron transfer
from the CNT to NO3 molecules, is 0.6 electrons per
molecule when the band structure is computed with PBE.
Band structures for computational unit cells consisting of
(10,0) CNTs with two NO3 molecules are shown in the
i.e.
(a)
(10,0) CNT
PBE
2.0
1.5
1.0
0.5
Ev
−0.5
−1.0
−1.5
)
V
e
(
y
g
r
e
n
E
−2.0
0.000 0.075 0.150
0
kz (1/Å)
25
DOS
(arbitr. units)
4
(b)
(c)
(10,0) CNT + 6xNO3
(10,0) CNT + 8xNO3
PBE
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
PBE
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
−2.0
0.000 0.075 0.150
50
−2.0
0.000 0.075 0.150
0
100 200
kz (1/Å)
kz (1/Å)
DOS
(arbitr. units)
FIG. 2. Band structures of (a) a pristine (10,0) CNT and (b) -- (c) NO3-doped (10,0) CNTs. The DOSs for the systems studied
are also shown in (a) and (c). The DOS peaks in (a) are van Hove singularities and the additional high DOS peaks in (c) are
due to localized molecular states. The red dashed line represents the valence band maximum Ev of the pristine (10,0) CNT or
the Fermi level of the NO3-doped (10,0) CNT.
appendix and the charge transfer per one molecule in this
system is 0.5 electrons. The band structures for highly
doped (10,0) CNTs in Figs. 2(b) and 2(c) show signifi-
cant downshifts of the Fermi level but the average doping
efficiencies of the NO3 molecules are smaller than those
determined for the two-molecule system (see Table I).
Therefore, the doping efficiency decreases as the dopant
density increases.
In the case of two molecules in the
computational unit cell, the total energy of the system is
0.15 eV lower when spin is included in the calculation. No
significant differences between the band structures of the
NO3-doped CNTs with and without spin-polarization are
found but a clear (spin) splitting of the molecular state
close to the Fermi level occurs in the spin-polarized case
(see Fig. 12(b) in the appendix). Otherwise almost all
of the bands in the −2.0 to 2.0 eV range, relative to the
Fermi level, overlap with each other in the spin-polarized
band structure presented in Fig. 12(b). The doping effi-
ciency determined for a computational unit cell with two
NO3 molecules taking the spin into account is approxi-
mately 0.4 electrons per one molecule.
A recent DFT study has revealed that the NO3
molecule on a (10,0) CNT obtains a charge of 0.5 -- 0.6
e according to a Bader charge analysis36. The charge
transfer found in the present work is rather close to that
result. For comparison, we have determined the Hirshfeld
charge for a single NO3 molecule and obtained a value
0.4 e. This is also in agreement with the Bader charge
although slightly smaller than that given by the band
structure analysis.
TABLE I. Doping efficiency in NO3-doped (10,0) CNTs per
one molecule. The values are given per molecule as a function
of the number of NO3 molecules (NNO3
) in the computational
unit cell. The results have been computed with PBE.
NNO3 Includes spin
Doping efficiency
per NO3 molecule (e)
1
2
2
6
8
No
No
Yes
No
No
0.6
0.5
0.4
0.4
0.3
B. Metallic CNTs with NO3 molecules
The number of NO3 molecules in the computational
unit cell of a metallic (8,8) CNT is varied and the
atomic structure of each doped system is optimized sep-
arately. After the relaxation, the nitrogen atom of the
NO3 molecule resides, as in the case of the semicon-
ducting (10,0) CNT, on top of a carbon atom and the
oxygen atoms are located near the centers of the carbon
hexagons. The CNT-NO3 distance in a NO3-doped (8,8)
CNT shown in Fig. 1(c) is 3.06 A. The binding energy
given by Eq. (1) for a NO3 molecule on an (8,8) CNT
is −1.31 eV, indicating a more stable structure than the
(10,0) CNT, which has a binding energy of −0.80 eV.
The trend that the binding energy is higher for metal-
lic CNTs is in accordance with experimental findings
for semiconducting and metallic CNTs17.
It is due to
the stronger polarization of the electron gas in metal-
lic CNTs. The CNT-molecule distance does not change
significantly when the number of NO3 molecules in the
(a)
(8,8) CNT
PBE
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
)
V
e
(
y
g
r
e
n
E
−2.0
0.000 0.075 0.150
0
kz (1/Å)
25
DOS
(arbitr. units)
5
(b)
(c)
(8,8) CNT + 4xNO3
(8,8) CNT + 8xNO3
PBE
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
PBE
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
−2.0
0.000 0.075 0.150
50
−2.0
0.000 0.075 0.150
0
100 200
kz (1/Å)
kz (1/Å)
DOS
(arbitr. units)
FIG. 3. Band structures of (a) a pristine (8,8) CNT and (b) -- (c) NO3-doped (8,8) CNTs. The DOSs for the systems studied
are also shown in (a) and (c). The red dashed lines denote the Fermi level. The origin of the DOS peaks is the same as in
Fig. 2.
computational unit cell of the CNT is increased from one
to eight. The values of the average CNT-molecule dis-
tances and other details are given in the appendix.
level
The band structures of a pristine (8,8) CNT and NO3-
doped (8,8) CNTs are displayed in Figs. 3(a) -- (c). An
(8,8) CNT without doping has a band structure of a
metallic system and the Fermi
is located at a
point where two bands cross each other. Placing NO3
molecules on top of an (8,8) CNT results in a relative
downshift of the Fermi level as can be seen from Fig. 3(b).
This downshift can be regarded as a p-type doping effect
since there is charge transfer from the CNT states to
the molecular ones. In addition, similarly to the case of
the NO3-doped semiconducting (10,0) CNTs, there is a
flat band very close to the Fermi level that is a molecu-
lar state. A similar state has been observed in another
study38 but it was found lower below the Fermi level com-
pared with this work. The downshift of the Fermi level
can be enhanced by increasing the number of molecules
in the computational unit cell. The Fermi level, however,
does not reach the first van Hove singularity although the
molecular concentration is increased to the same level as
that of the highly doped semiconducting (10,0) CNT.
This can be seen in Fig. 3(c). Thus, obtaining a stable
doping effect via the Fermi level pinning at the van Hove
singularity can be challenging6.
The values for the charge transfer in NO3-doped (8,8)
CNTs are given in Table II. The doping efficiency of
one NO3 molecule on an (8,8) CNT per computational
unit cell is close to 0.7 electrons. A significant finding
is also that the doping efficiency of NO3 molecules de-
creases when the molecular concentration on the CNT is
increased. If the number of NO3 molecules around the
(8,8) CNT is doubled, the doping efficiency decreases ap-
proximately by 0.1 electrons per molecule. In the case of
a computational unit cell with eight NO3 molecules, the
TABLE II. Doping efficiency in NO3-doped (8,8) CNTs. The
values are given per molecule as a function of the number of
NO3 molecules (NNO3
) in the computational unit cell and
have been computed with PBE.
NNO3
Doping efficiency
per NO3 molecule (e)
1
2
3
4
5
8
0.7
0.6
0.6
0.5
0.5
0.4
value of the charge transfer has decreased to 0.4 electrons
per molecule.
C. Doping efficiencies determined with HSE06 and
SIC
The spurious self-interaction error inherent to prac-
tical implementations of DFT, particularly at the GGA
level of theory, makes charge transfer systems notoriously
difficult to describe39. This is due to incomplete cancel-
lation of the self-Coulomb term (i.e. an electron feels its
own Coulomb repulsion) in the exchange-correlation po-
tential. Electrons tend to delocalize in space in order to
minimize the self-repulsion and as a result the observed
charge transfer, or effective doping, is lowered. This can
be remedied by including (partial) exact-exchange -- as
in hybrid exchange-correlation functionals -- or by ex-
plicitly subtracting the self-Coulomb term from the to-
tal energy functional. Therefore, we compare our PBE
results shown above with those obtained by using the
6
HSE06 functional and the Perdew-Zunger SIC approach.
Computing the band structure for a NO3-doped (10,0)
CNT with HSE06 shows an enhancement of the down-
shift of the Fermi level with respect to the PBE results
given above. This indicates a higher doping efficiency.
Moreover, the molecular levels move toward lower ener-
gies with respect to the CNT bands. Besides the FHI-
aims code, the other code package GPAW is used to com-
pute charge transfer in a short computational unit cell of
a NO3-doped (10,0) CNT using the SIC. Perdew-Zunger
SIC calculations have been successful in describing chal-
lenging systems where the commonly used exchange-
correlation functionals of DFT are in error. This in-
cludes a localized defect state in a crystal31, and charged
localized states of a molecule40.
In the case of NO3-
doped (10,0) CNTs, ground state calculations including
the Perdew-Zunger SIC are performed first. Thereafter,
we carry out a Bader charge analysis41 and determine
the partial charges of the NO3 molecule. If there is only
one NO3 molecule in the short computational unit cell
and the SIC is included in the calculation, the value of
the charge transfer is 0.9 electrons. Therefore, the SIC
method results in clearly higher doping efficiency for the
NO3 molecule than the PBE calculations.
efficiencies of NO3-doped semiconducting and metallic
CNTs with different molecular concentrations are given
in Table III. The Perdew-Zunger SIC value has been com-
puted using a smaller computational unit cell than that
of the HSE06 calculations. The functional in our trans-
port calculations is PBE and the comparison between
the PBE and HSE06 band structures indicates that our
transport calculations will give qualitative results. How-
ever, they will predict phenomena taking place with dop-
ing and, thereby, they give insight into the electron trans-
port mechanisms along individual CNTs and across CNT
junctions in particular.
TABLE III. Doping efficiency determined with HSE06 or the
Perdew-Zunger SIC in (10,0) and (8,8) CNTs doped with one
or two (NNO3
) NO3 molecules per computational unit cell.
Chirality NNO3 Functional
Doping efficiency
per NO3 molecule
(10,0)
(10,0)
(8,8)
(8,8)
1
1
1
2
HSE06
PW91 (SIC)
HSE06
HSE06
0.8
0.9
0.9
0.8
(a)
(8,8) CNT
HSE06
(b)
(8,8) CNT + 1xNO3
HSE06
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
)
V
e
(
y
g
r
e
n
E
−2.0
0.000 0.075 0.150
−2.0
0.000 0.075 0.150
kz (1/Å)
kz (1/Å)
FIG. 4. Band structures computed with HSE06 for (a) a pris-
tine (8,8) CNT and (b) an (8,8) CNT with one NO3 molecule
in the computational unit cell.
The band structures calculated with HSE06 for pristine
and NO3-doped (8,8) CNTs are presented in Figs. 4(a)
and 4(b). Compared to the PBE results in Fig. 3(a),
a shift of the bands below the Fermi level toward lower
energies is observed in the band structure for a pristine
(8,8) CNT (see Fig. 4(a)). Similarly to the case of a (10,0)
CNT with one NO3 molecule in the computational unit
cell, the downshift of the Fermi level in Fig. 4(b) be-
comes more remarkable than in PBE calculations giving
a higher doping efficiency when the calculation is carried
out with the HSE06 functional. The doping efficiency in-
creases to over 0.9 electrons per molecule. The doping
D. Carbon nanotubes with NO3-H2O complexes
The charge transfer (or doping efficiency) can be en-
hanced by water molecules which coordinate to anionic
species in molecular systems21. Due to strong dielec-
tric screening, water molecules stabilize charged atoms
or molecules, facilitating a more chemically robust charge
transfer, in the sense that a unit, or near unit, of charge
(a)
(short) (10,0) CNT
+ 1xNO3
HSE06
(b)
(short) (10,0) CNT
+ 1xNO3 + 3xH2O
HSE06
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
)
V
e
(
y
g
r
e
n
E
−2.0
0.000 0.150 0.300
−2.0
0.000 0.150 0.300
kz (1/Å)
kz (1/Å)
FIG. 5. Band structures computed with HSE06 for the short
computational unit cells (see Fig. 1(b)) of NO3-doped (10,0)
CNTs (a) without and (b) with three H2O molecules. The
Fermi level is denoted by a red dashed line.
(a)
(b)
(c)
7
(8,8) CNT + 2xNO3
(8,8) CNT + 3xNO3
+ 6xH2O
HSE06
+ 9xH2O
HSE06
(8,8) CNT + 1xNO3
+ 3xH2O
HSE06
2.0
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
)
V
e
(
y
g
r
e
n
E
2.0
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
2.0
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
0.000 0.075 0.150
−2.0
0.000 0.075 0.150
−2.0
0.000 0.075 0.150
kz (1/Å)
kz (1/Å)
kz (1/Å)
FIG. 6. Band structures computed with HSE06 for (8,8) CNTs with NO3-H2O molecular complexes. The number of molecular
complexes on the CNT is (a) one, (b) two, or (c) three per computational unit cell. The Fermi level is marked with a red
dashed line.
is transferred between the donor and acceptor. Inspired
by this observation, we have studied NO3-H2O complexes
on semiconducting (10,0) and metallic (8,8) CNTs. After
structure optimizations with the PBE functional and the
van der Waals correction, band structures are calculated
using the HSE06 functional.
For the (10,0) CNT, the calculations are performed
with the short computational unit cell depicted in
Fig. 1(b) in order to facilitate a comparison with Perdew-
Zunger SIC results. The HSE06 band structures for one
NO3 molecule without or with three H2O molecules in
the computational unit cell are presented in Fig. 5. The
comparison shows that water is capable of enhancing the
charge transfer from the CNT. Without water, the charge
transfer from the CNT to the NO3 molecule is 0.6 elec-
trons, which is slightly larger than the PBE result of 0.5
electrons for the same dopant concentration (see Table I).
With the three coordinated H2O molecules, the doping
efficiency is 0.9 electrons per NO3-H2O complex. Per-
forming the same calculation with the Perdew-Zunger
SIC method gives a doping efficiency of 1.1 electrons.
Therefore, both the Perdew-Zunger SIC and HSE06 cal-
culations predict a practically complete charge transfer
of one electron from the CNT to the NO3-H2O complex.
In the case of the metallic (8,8) CNT, we have exam-
ined systems of several NO3-H2O complexes per com-
putational unit cell. The HSE06 band structures for
one, two, and three NO3 molecules each coordinated to
three H2O molecules in the computational unit cell (see
Fig. 1(d)) are presented in Fig. 6. The remarkable qual-
itative effect of H2O molecules is the pushing of molec-
ular states clearly below the Fermi level as can be seen
by comparing figures. As a result, the calculated doping
efficiencies indicate in all the cases studied a complete
charge transfer of one electron per NO3-H2O complex.
This is in contrast to the efficiencies of 0.9 and 0.8 elec-
trons per bare NO3 molecule for one and two molecules in
the computational unit cell, respectively (see Table III)
since the doping efficiency decreases as the number of
molecules in the system is increased. Figure 6 shows also
the increase of dispersionless molecular states in number
due to coordinated H2O molecules. Moreover, when the
concentration of the molecular complexes increases, the
splitting of the molecular states originating from the in-
teractions between the large complexes enhances. At the
same time, also part of the dispersive CNT states are
split indicating some hybridization between the states of
the CNT and the NO3-H2O molecular complexes. In con-
clusion, the H2O molecules enhance the doping efficiency,
but the increased interactions between the molecular and
CNT states will increase the electron scattering and are
harmful for the electron transport.
E. Electron transport in doped individual (10,0)
and (8,8) CNTs
Studying the conductivity of single NO3-doped CNTs
is done by constructing a two-terminal transport sys-
tem by using the computational unit cells depicted in
Figs. 1(a) -- (d). Then the system contains three similar
computational unit cells of which the first and third ones
belong to the leads that are also doped. The middle cell
forms the scattering region. The electronic transmission
functions for individual semiconducting (10,0) and metal-
lic (8,8) CNTs are shown in Fig. 7. In addition to the
transmission functions for pristine CNTs, the transport
curves for NO3-doped CNTs are presented in Fig. 7.
In the case of a pristine (10,0) CNT, the transmission
function possesses several steps and has a gap indicat-
ing semiconducting behavior.
In contrast, there is no
gap in the transmission function for the pristine (8,8)
8
(10,0) CNT
(10,0) CNT + 8xNO3
20.0
15.0
10.0
5.0
(8,8) CNT
(8,8) CNT + 8xNO3
20.0
15.0
10.0
5.0
0.0
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
0.0
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
Energy (eV)
Energy (eV)
i
i
n
o
s
s
m
s
n
a
r
T
FIG. 7. Electronic transmission functions for individual pristine (black solid lines) and NO3-doped (red dashed lines) CNTs.
The systems in the left panel are a pristine semiconducting (10,0) CNT and a (10,0) CNT with eight NO3 molecules in the
computational unit cell. The right panel presents the transmission functions for a pristine metallic (8,8) CNT and for a metallic
(8,8) CNT doped with eight NO3 molecules in the computational unit cell. The energy zero denotes the top of the valence
band of the pristine semiconducting (10,0) CNT or the Fermi level of the metallic systems.
CNT. Therefore, this system is metallic. Corresponding
to the changes in the band structures, in the case of both
CNTs the transmission curves are shifted upwards in en-
ergy when the number of NO3 molecules on the CNT is
increased. Importantly, in the case of the doped semicon-
ducting CNT the transmission at the Fermi level is re-
markable, i.e., the doped CNT is metallic. Interestingly,
a significant dip appears near the Fermi level in the trans-
mission function for the NO3-doped (8,8) CNT. The posi-
tion of the dip corresponds to the energy of the molecular
state in Figs. 3(b)-(c). The decrease in the electronic
transmission function at the Fermi energy can be at-
tributed to Fano antiresonances in nanostructures42, ef-
fects found also, e.g., in graphene nanoribbon systems43.
A decrease in the conductance has been observed in a
computational study where a metallic (8,8) CNT with
a NO2 molecule has been placed between two Au (111)
electrodes44 so that our finding of the lowering of the
transport in the NO3 doping process is in agreement with
the previous work.
F. Junctions of doped CNTs
The computational unit cells of doped CNTs inves-
tigated in the previous sections can also be used to
form CNT junctions.
In particular, we consider junc-
tions of two perpendicular doped semiconducting (10,0)
or metallic (8,8) CNTs. These junctions consist of CNTs
with eight NO3 molecules in the computational unit cell.
Further, another junction between a NO3-doped metal-
lic (8,8) and a NO3-doped semiconducting (10,0) CNT
is investigated.
In this junction, the numbers of NO3
molecules in the computational unit cells of (8,8) and
(10,0) CNTs are eight and six, respectively.
The distance between the doped CNTs is optimized be-
fore performing the transport calculation. When the van
der Waals correction is taken into account, the distance
between the (10,0) CNTs with NO3 molecules is 3.21 A.
Thus, the distance is larger than 2.5 − 2.6 A obtained in
similar calculations for a junction of (8,8) CNTs without
doping9,35, reflecting the effect of electron charge trans-
fer from the CNTs to the molecules leaving the CNTs
slightly electron deficient and resulting in Coulomb repul-
sion between the CNTs. The distance between the CNTs
of the two other junctions are also optimized with the van
der Waals correction. In the junction between two NO3-
doped (8,8) CNTs, the CNT-CNT distance is 3.16 A. The
distance in the third junction between a metallic (8,8)
and a semiconducting (10,0) CNT is 3.11 A. Thus, the
doping increases the CNT-CNT distance but distances
do not vary significantly when changing the chiralities of
the CNTs.
The intratube and intertube transmission functions for
a junction of NO3-doped (10,0) CNTs are presented in
Fig. 8. The intratube transport in Fig. 8 is rather good
although the CNTs are doped with a large number of
molecules. There are, however, small dips near the en-
ergy zero that are related to Fano antiresonances due to
localized molecular states. The logarithmic plot in the
bottom panel of Fig. 8 shows an increase of about one
decade in transmission when crossing a single van Hove
singularity below the Fermi level. A similar increase can
be seen in the transmission between two pristine (10,0)
CNTs over the same singularity (published in Fig. 6 in
our previous work6). This increase is clearly larger than
the corresponding increase in DOS reflecting the fact that
also changes in the wavefunctions and not only the num-
ber of states available affect the intertube transmission.
Besides the change in the average transmission over
the Fermi level, the intertube transmission function in
Fig. 8 increases strongly just at the Fermi level, where
there is a van Hove singularity and a molecular state.
Thus, the increased DOS of the CNTs enhances the elec-
tron tunneling between the CNTs. Moreover, when com-
paring the intertube transmission with that between two
pristine (10,0) CNTs (Fig. 6 in Ref.6) we notice a clear
broadening of the van Hove singularity derived peak just
Intratube
-1.0
0.0
1.0
2.0
Intertube
-1.0
0.0
1.0
2.0
Intertube
9
Intratube
-1.0
0.0
1.0
2.0
Intertube
-1.0
0.0
1.0
2.0
Intertube
20.0
15.0
10.0
5.0
0.0
-2.0
1.0
0.8
0.6
0.4
0.2
0.0
-2.0
i
i
n
o
s
s
m
s
n
a
r
T
1.0e+00
1.0e-01
1.0e-02
1.0e-03
1.0e-04
20.0
15.0
10.0
5.0
0.0
-2.0
1.0
0.8
0.6
0.4
0.2
0.0
-2.0
i
i
n
o
s
s
m
s
n
a
r
T
1.0e+00
1.0e-01
1.0e-02
1.0e-03
1.0e-04
1.0e-05
-1.6 -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 1.6
1.0e-05
-1.6 -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 1.6
Energy (eV)
Energy (eV)
FIG. 8.
Intratube (upper plot) and intertube (middle and
lower plots) electronic transmission functions for a junction
of two (10,0) CNTs with eight NO3 molecules in the compu-
tational unit cell. The energy zero is at the Fermi level.
FIG. 9.
Intratube (upper plot) and intertube (middle and
lower plots) electronic transmission functions for a junction
of two (8,8) CNTs with eight NO3 molecules in the computa-
tional unit cell. The energy zero is at the Fermi level.
below the Fermi level. This means that the CNT and
molecular levels are hybridized increasing the extent of
the wavefunctions and the ensuing electron transmission.
Because the Fermi level can be pinned to regions of high
DOS and because the electric current is determined by
integrating the transmission function over a Fermi-level
centered bias window the total current through the junc-
tion will be increased. We should also note that for this
reason the total current will increase due to doping al-
though the intertube distance increases.
The results from the four-terminal transport calcula-
tions for a junction of NO3-doped (8,8) CNTs are dis-
played in Fig. 9. In this case, electron transport along
individual CNTs (see the upper plot in Fig. 9) remains al-
most as good as in pristine metallic (8,8) CNTs. The val-
ues of the intertube transmission function increase over
the van Hove singularity nearly by one decade and the
singularity causes a sharp peak. There is also a sharp
and high peak close to the Fermi energy. This peak,
which is due to tunneling through localized molecular
states, is higher than a similar one found in the inter-
tube function in Fig. 8. When we compare the inter-
tube transmission in Fig. 9 with that between two pris-
tine (8,8) CNTs (see Fig. 3b in our previous article9),
we notice that the transmission peaks corresponding to
van Hove singularities below the Fermi level have similar
widths. These observations about the peak shapes mean
that the hybridization between the CNT and molecular
states is minimal, the extents of the wavefunctions do
not increase, and a doping-induced increase in the total
intertube current is expected to remain lower than for
the junction of two doped semiconducting (10,0) CNTs.
The enhancement of the electron transport through
the junction of NO3-doped (10,0) CNTs relative to that
of NO3-doped (8,8) CNTs follows from enhanced hy-
bridization of the CNT and molecular states as shown by
the charge densities of the highest occupied eigenstates
of the CNT-NO3 systems (see Figs. 10(a) and 10(b)).
The spreading of the wavefunctions and the hybridiza-
tion between the CNT and molecular orbitals is remark-
ably larger in the NO3-doped (10,0) CNTs than in the
NO3-doped (8,8) CNTs, which leads to spreading of the
eigenstates in energy and an ensuing increase in the in-
tertube transmission over wide energy regions between
broadened transmission peaks at van Hove singularities.
Performing a four-terminal transport calculation for a
junction consisting of a doped metallic (8,8) and a doped
semiconducting (10,0) CNT shows that the intratube
transmission curves (see the upper panel of Fig. 11) of
this junction are similar to those presented in Figs. 8
and 9. The intertube transmission function in the two
lowest plots of Fig. 11 indicates that the transmission
10
resemble those computed for a similar CNT junction of
two AuCl4-doped (10,0) CNTs6. The intratube trans-
mission function for the NO3-doped system, however, is
smoother, i.e. has less peaks per energy unit, than that
of the CNT junction with AuCl4 molecules. This indi-
cates that there are fewer molecular states in the NO3
molecules and they do not hybridize with those of the
CNT so strongly causing less electron scattering than
the AuCl4 molecules on the same CNT. Correspondingly,
the number of peaks in the vicinity of the Fermi level
in the intertube transmission function for (10,0) CNTs
with NO3 molecules is smaller than in the same region
in the intertube function for AuCl4-doped (10,0) CNTs
and doping with NO3 is expected to be less efficient than
doping with AuCl4.
Increases in the conductances of the junctions of doped
CNTs are related to the Fermi level crossing van Hove
singularities and the high and wide peaks of the inter-
tube transmission functions. Similarly to the junctions
of AuCl4-doped (10,0) CNTs6, the conductances of junc-
tions of NO3-doped semiconducting (10,0) or metallic
(8,8) CNTs enhance due to doping when the Fermi level
shifts to a region, where the DOS is high. The Fermi level
can be pinned to the van Hove singularities or localized
molecular states leading to a robust mechanism for the
improvement of the conductivity of CNT networks6. The
pinning of the Fermi level at regions of high DOSs can
also increase the conductances between dissimilar CNTs
such as in the case of semiconducting (10,0) and metallic
(8,8) CNTs discussed above.
Using atomic force microscopy, Znidarsic13 et al. mea-
sured resistances of single-wall CNTs and their junctions
forming CNT networks before and after nitric acid treat-
ments. They analyzed the measured data for junction re-
sistances as a function of the diameters of the two CNTs
forming X- or Y-shape junctions and as a function of
the angle between the CNTs. The smallest contact re-
sistances found were 29 kΩ. The contact resistances de-
creased with increasing CNT radii and decreasing the
contact angle. Y-junctions showed smaller resistances
than X-junctions. Because the unity transmission corre-
sponds to the conductance quantum 2e2/h, the junction
resistance of 29 kΩ would correspond to a transmission
of about 0.5. According to our results in Figs. 8, 9, and
11, these kinds of transmission values are not plausible
for the rather small radii (10,0) and (8,8) CNTs forming
a perpendicular junction outside the transmission sin-
gularity peak regions unless the Fermi level has crossed
three van Hove singularities. According to Ref.13, the
nitrogen acid treatment decreased the average junction
resistance by a factor of around three, which may signal-
ize the lowering of the Fermi level in CNTs and also its
pinning around the van Hove singularities as a function
of increased doping.
In a recent experimental work, Tsebro at al.8 mea-
sured, as a function of temperature, sheet resistances of
CNT networks before and after iodine or CuCl doping.
The CNT radii were rather large, around 2 nm, and the
FIG. 10. Charge density of the highest occupied eigenstate
of (a) a (10,0) CNT and (b) an (8,8) CNT with one NO3
molecule in the computational unit cell. The value of the
electron density of the isosurface is the same for both systems.
Intratube 1
Intratube 2
-1.0
0.0
1.0
2.0
Intertube
-1.0
0.0
1.0
2.0
Intertube
20.0
15.0
10.0
5.0
0.0
-2.0
1.0
0.8
0.6
0.4
0.2
0.0
-2.0
i
i
n
o
s
s
m
s
n
a
r
T
1.0e+00
1.0e-01
1.0e-02
1.0e-03
1.0e-04
1.0e-05
-1.6 -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 1.6
Energy (eV)
FIG. 11. Intratube (upper plot) and intertube (middle and
lower plots) electronic transmission functions for a junction
of a NO3-doped (8,8) and a NO3-doped (10,0) CNT. The
numbers of NO3 molecules in the computational unit cells of
the (8,8) and (10,0) CNTs are eight and six, respectively. The
energy zero is located at the Fermi level.
through the junction below the band gap of the (10,0)
CNT is rather high in comparison with the symmet-
ric junctions discussed above. Moreover, the molecular
state at the Fermi level causes a very narrow transmission
peak. However, clear steps and wider regions of higher
transmission are absent until rather low energies. This
is due to the fact that the regions of high DOS due to
van Hove singularities and molecular states in the two
CNTs do not match at the same energies. Therefore, the
conductances of junctions between doped semiconduct-
ing and metallic CNTs are expected to be low.
Intra- and intertube transmission functions in Fig. 8
dopants filled the CNTs. Tsebro et al. analyzed their
data in terms of intra- and interbundle contributions cor-
responding to phonon scattering and fluctuation-assisted
tunneling, respectively. With the help of calculated
phonon dispersion relations and electronic band struc-
tures, they were able to conclude that iodine and CuCl
doping lower the Fermi levels by 0.6 and 0.9 eV, respec-
tively. The latter figure means that the Fermi level has
crossed in semiconducting (metallic) CNTs three (one)
van Hove singularities (singularity). The sheet resis-
tances of the CNT networks decreased almost by a decade
in the CuCl doping and the effect did not decay indicat-
ing the the internal doping of the CNT is stable. Our
results show that the crossing of the Fermi level below
a van Hove singularity causes a decrease of the junction
resistance, which is of the same order of magnitude. Re-
lated to this, we also note that Tsebro et al. proposed
that the main cause for the sheet resistance of the CNT
network is the junction resistance and that doping mainly
decreases this contribution.
IV. CONCLUSIONS
The influence of NO3 molecules on the band structures
and electronic transmission functions of semiconducting
and metallic CNTs has been explored to understand the
nitric-acid doping process used to improve the conduc-
tivity of CNT thin films. In the present work, (10,0) and
(8,8) CNTs are used in our density functional modeling
to represent semiconducting and metallic CNTs, respec-
tively. Our results show that electrons are transferred
from the CNT to the NO3 molecule and this results
in p-type doping of the CNT. The average doping effi-
ciency per NO3 molecule decreases when the number of
molecules on the CNT is increased. However, the doping
efficiency enhances considerably when the NO3 molecules
form complexes with H2O molecules.
Electron transport calculations of
individual NO3-
doped semiconducting CNTs reveal that doping converts
them metallic although sharp dips in the transmission
function due to scattering off localized molecular states
appear. The transmission functions increase stepwise
over the van Hove singularities of the CNT electronic
structures. Reflecting qualitatively different band struc-
tures, shifting the Fermi level toward the next singularity
requires less doping in the case of semiconducting CNTs
in comparison with metallic CNTs. This would favor
semiconducting CNTs in CNT networks.
The intertube transmission functions between two
semiconducting or two metallic CNTs increase around
one decade over a van Hove singularity. This increases
directly the conductance of the CNT junction. Moreover,
the pinning of the Fermi level at the van Hove singularity
and partially filled molecular states with the hybridiza-
tion of the molecular and CNT states there can further
increase the junction conductance and the conductivity
of the CNT network. The hybridization of the molecu-
11
lar and CNT states is stronger for semiconducting than
metallic CNTs, which would also favor the former in CNT
networks. The energy mismatch between the high DOS
regions for the clearly dissimilar electronic structures of
metallic and semiconducting CNTs hinders to enhance
the conductance of their junctions.
APPENDIX
The geometries of metallic (8,8) and semiconducting
(10,0) CNTs with NO3 molecules are optimized using the
FHI-aims code package. The number of NO3 molecules
on the CNTs is varied and the atomic structures of indi-
vidual doped CNTs are relaxed. The relaxation is started
with the light settings for the basis functions. After com-
pleting the relaxation with the light settings, the relax-
ation is continued with the tight settings. The relaxation
calculations are stopped when the maximum force com-
ponent falls below the tolerance values as explained in
the article text. The optimal geometries of doped metal-
lic and doped semiconducting CNTs are considered in
the following sections. Some systems also have H2O
molecules near the NO3 molecule. Moreover, this ma-
terial contains a few figures of band structures for NO3-
doped CNTs and for these systems with H2O molecules.
The k-point grids of the band structure calculations for
the PBE and HSE06 functionals are 1 × 1 × 27 and
1 × 1 × 15, respectively.
Optimizing the geometry of the computational unit
cell of an (8,8) CNT with one NO3 molecule shows that
the distance between the NO3 molecule and the CNT
decreases by 0.16 A when the Tkatchenko-Scheffler van
der Waals correction is taken into account. The CNT-
molecule distances for NO3-doped (8,8) CNTs are listed
in Table IV. The distances are averages over all CNT-
molecule distances in the system and have been calcu-
lated with the van der Waals correction.
TABLE IV. Average CNT-molecule distances for NO3-doped
(8,8) CNTs when taking the van der Waals correction into
account.
(A)
NNO3
1
2
3
4
5
8
dCNT-NO3
3.06
3.05
3.07
3.07
3.07
3.07
The average CNT-molecule distance does not depend
on the number of NO3 molecules remarkably. Only a
small increase occurs in the distances when the dopant
concentration is increased. This change, however, is neg-
ligible.
Placing water molecules around the NO3 dopant af-
fects the CNT-molecule distance a little. The distance
between the (8,8) CNT and the NO3 molecule increases
to 3.29 A if the calculation includes the van der Waals
correction. Neglecting this correction leads to an even
larger CNT-molecule distance, 3.55 A.
(a)
(10,0) CNT
+ 2xNO3
PBE
(b)
(10,0) CNT
+ 2xNO3
PBE (spin)
spin up
spin down
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
2.0
1.5
1.0
0.5
EF
−0.5
−1.0
−1.5
)
V
e
(
y
g
r
e
n
E
−2.0
0.000 0.075 0.150
−2.0
0.000 0.075 0.150
kz (1/Å)
kz (1/Å)
FIG. 12. Band structures computed (a) without and (b) with
spin for the computational unit cells of (10,0) CNTs doped
with two NO3 molecules.
Optimization of the geometries of (10,0) CNTs doped
with NO3 molecules is performed with spin when there
is only one molecule on a (10,0) CNT. The (10,0) CNTs
with two NO3 molecules are relaxed both with and with-
out spin.
In the case of (10,0) CNTs with NO3-H2O
complexes, spin is omitted in the relaxation calculations.
After optimizing the short computational unit cell of a
∗ [email protected]
1 L.
Yu,
C.
Shearer,
and
J.
Shapter,
Chem. Rev. 116, 13413 (2016).
2 L. Hu,
D.
S. Hecht,
Appl. Phys. Lett. 94, 081103 (2009).
and G. Gruner,
3 C. Wang,
J.-C. Chien, K. Takei,
T. Taka-
and A. Javey,
hashi, J. Nah, A. M. Niknejad,
Nano Lett. 12, 1527 (2012).
J.
4 K. K. Kim,
J. Bae, H. K. Park,
S. M.
Kim, H.-Z. Geng, K. A. Park, H.-J. Shin, S.-M.
Yoon, A. Benayad, J.-Y. Choi,
and Y. H. Lee,
J. Am. Chem. Soc. 130, 12757 (2008).
5 A. Murat, I. Rungger, C. Jin, S. Sanvito, and U. Schwin-
genschlogl, J. Phys. Chem. C 118, 3319 (2014).
6 T. Ketolainen, V. Havu,
and M.
J. Puska,
J. Phys. Chem. C 121, 4627 (2017).
7 D. Janas, K. Z. Milowska, P. D. Bristowe, and K. K. K.
Koziol, Nanoscale 9, 3212 (2017).
8 V. I. Tsebro, A. A. Tonkikh, D. V. Rybkovskiy, E. A.
and E. D. Obraztsova,
Obraztsova, E. I. Kauppinen,
Phys. Rev. B 94, 245438 (2016).
9 T. Ketolainen, V. Havu,
and M.
J. Puska,
J. Chem. Phys. 142, 054705 (2015).
12
NO3-doped (10,0) CNT without water, the distance be-
is 3.05 A.
tween the CNT and the molecule dCNT-NO3
Thus, the molecule moves closer to the CNT compared
with the geometry of a larger computational unit cell.
The movement probably occurs because the charge of
the CNT becomes larger on average when the molecular
concentration is increased. As a result, the electrostatic
interaction in the system pulls the molecules closer to the
CNT. Adding H2O molecules to the NO3-doped (10,0)
CNT system makes the molecular complex move further
from the CNT surface. The distance dCNT-NO3
is 3.16
A and the alignment of the molecular system changes
because of the water molecules.
The downshift of the Fermi level occurs in the band
structure for a (10,0) CNT with one NO3 molecule in
the computational unit cell. A larger downshift can be
achieved by placing another molecule on the CNT in ad-
dition to the first one (see Fig. 12(a)). Carrying out a
spin-polarized band structure calculation results in split-
ting of the energy levels near the Fermi energy region as
shown in Fig. 12(b). The average doping efficiencies for
the NO3 molecules in this system can be found in the
article text.
ACKNOWLEDGMENTS
The computational resources provided by CSC - IT
Center for Science Ltd. have been of great importance.
We would like to thank Prof. Esko I. Kauppinen for in-
spiring discussions. This work was supported by the
Academy of Finland through its Centres of Excellence
Programme (2012-2017) under Project No. 251748.
In
addition, the work was funded by the AEF Aalto Energy
Efficiency research programme.
10 R. Graupner, J. Abraham, A. Vencelov´a, T. Seyller,
and L. Ley,
F. Hennrich, M. M. Kappes, A. Hirsch,
Phys. Chem. Chem. Phys. 5, 5472 (2003).
11 W. Zhou, J. Vavro, N. M. Nemes, J. E. Fischer,
and D. B. Tanner,
F. Borondics, K. Kamar´as,
Phys. Rev. B 71, 205423 (2005).
12 A. Kaskela, A. G. Nasibulin, M. Y. Timmermans,
B. Aitchison, A. Papadimitratos, Y. Tian, Z. Zhu,
H. Jiang, D. P. Brown, A. Zakhidov, and E. I. Kauppinen,
Nano Lett. 10, 4349 (2010).
13 A. Znidarsic, A. Kaskela, P. Laiho, M. Gaberscek,
Y. Ohno, A. G. Nasibulin, E. I. Kauppinen, and A. Has-
sanien, J. Phys. Chem. C 117, 13324 (2013).
14 H.
Jeong
and
J.-Y.
Park,
15 S. Peng, K. Cho,
J. Phys. Chem. C 119, 9665 (2015).
P. Qi,
Chem. Phys. Lett. 387, 271 (2004).
16 J.
Dai,
P.
Giannozzi,
Surf. Sci. 603, 3234 (2009).
and H. Dai,
and
J.
Yuan,
17 G. Ruiz-Soria, A. P´erez Paz, M. Sauer, D. J. Mow-
bray, P. Lacovig, M. Dalmiglio, S. Lizzit, K. Yanagi,
A. Rubio, A. Goldoni, P. Ayala,
and T. Pichler,
ACS Nano 8, 1375 (2014).
18 Y. Kanai, V. R. Khalap, P. G. Collins, and J. C. Gross-
Scuseria, J. Chem. Phys. 125, 224106 (2006).
man, Phys. Rev. Lett. 104, 066401 (2010).
19 M. S. Fuhrer, J. Nygard, L. Shih, M. Forero, Y.-G. Yoon,
M. S. C. Mazzoni, H. J. Choi, J. Ihm, S. G. Louie, A. Zettl,
and P. L. McEuen, Science 288, 494 (2000).
20 D. J. Mowbray, C. Morgan,
and K. S. Thygesen,
Phys. Rev. B 79, 195431 (2009).
21 E. O. J´onsson, K. S. Thygesen, J. Ulstrup, and K. W.
30 A.
Tkatchenko
and
M.
Scheffler,
Phys. Rev. Lett. 102, 073005 (2009).
31 H. Gudmundsd´ottir, E. O. J´onsson,
New J. Phys. 17, 083006 (2015).
and H. J´onsson,
32 J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jack-
and C. Fiolhais,
son, M. R. Pederson, D. J. Singh,
Phys. Rev. B 46, 6671 (1992).
13
Jacobsen, J. Phys. Chem. B 115, 9410 (2011).
33 J. T. Frey
and D.
J. Doren,
"TubeGen
F. Hanke,
P. Havu,
and M. Scheffler,
University
http://turin.nss.udel.edu/research/tubegenonline.html.
Delaware,
Newark
DE,
of
3.4,
2011,"
22 V. Blum,
R. Gehrke,
V. Havu, X. Ren, K. Reuter,
Comput. Phys. Commun. 180, 2175 (2009).
23 S. V. Levchenko, X. Ren,
hanni, P. Rinke, V. Blum,
Comput. Phys. Commun. 192, 60 (2015).
J. Wieferink, R. Jo-
and M. Scheffler,
24 A. C.
Ihrig, J. Wieferink,
X. Ren, P. Rinke, M. Scheffler,
New J. Phys. 17, 093020 (2015).
I. Y. Zhang, M. Ropo,
and V. Blum,
25 J. Enkovaara, C. Rostgaard, J. J. Mortensen, J. Chen,
M. Du lak, L. Ferrighi, J. Gavnholt, C. Glinsvad,
V. Haikola, H. A. Hansen, H. H. Kristoffersen, M. Kuisma,
A. H. Larsen, L. Lehtovaara, M. Ljungberg, O. Lopez-
Acevedo, P. G. Moses, J. Ojanen, T. Olsen, V. Pet-
zold, N. A. Romero, J. Stausholm-Møller, M. Strange,
G. A. Tritsaris, M. Vanin, M. Walter, B. Ham-
mer, H. Hakkinen, G. K. H. Madsen, R. M. Niem-
inen,
J. K. Nørskov, M. Puska, T. T. Rantala,
J. Schiøtz, K. S. Thygesen,
and K. W. Jacobsen,
J. Phys.: Condens. Matter 22, 253202 (2010).
34 J. Ferrer, C. J. Lambert, V. M. Garc´ıa-Su´arez, D. Z. Man-
rique, D. Visontai, L. Oroszlany, R. Rodr´ıguez-Ferrad´as,
I. Grace, S. W. D. Bailey, K. Gillemot, H. Sadeghi, and
L. A. Algharagholy, New J. Phys. 16, 093029 (2014).
35 P.
E.
J. Phys.: Condens. Matter 23, 112203 (2011).
Hashemi,
and
M.
R. M.
Havu,
T.
Seppala,
M.
J.
Kaukonen,
Nieminen,
36 J. M. H. Kroes,
C. Roman,
Appl. Phys. Lett. 108, 033111 (2016).
and
C. Hierold,
Guo,
Lu,
Y.
37 H.
F. Pietrucci, K. Chikkadi,
and W. Andreoni,
J.
Robertson,
J. Appl. Phys. 121, 224304 (2017).
38 S. C. Guerini, D. L. Azevedo, M. C. A. Lima, I. Zanella,
and J. M. Filho, Cent. Eur. J. Phys. 6, 105 (2008).
39 A. Ruzsinszky, J. P. Perdew, G. I. Csonka, O. A. Vydrov,
and G. E. Scuseria, J. Chem. Phys. 125, 194112 (2006).
40 X. Cheng, Y. Zhang, E. J´onsson, H. J´onsson, and P. M.
Weber, Nat. Commun. 7, 11013 (2016).
26 J. J. Mortensen, L. B. Hansen,
Phys. Rev. B 71, 035109 (2005).
and K. W. Jacobsen,
41 G. Henkelman, A. Arnaldsson,
and H. J´onsson,
Comput. Mater. Sci. 36, 354 (2006).
27 J.
P.
Perdew
and
A.
Zunger,
Phys. Rev. B 23, 5048 (1981).
28 J. P. Perdew, K. Burke,
Phys. Rev. Lett. 77, 3865 (1996).
and M. Ernzerhof,
L. Karkkainen, M. J. Puska,
Phys. Rev. B 83, 205125 (2011).
42 C. J. Lambert, Chem. Soc. Rev. 44, 875 (2015).
43 K.
Saloriutta,
Hancock,
Y.
A. Karkkainen,
and A.-P. Jauho,
29 A. V. Krukau, O. A. Vydrov, A. F. Izmaylov, and G. E.
44 S. Sivasathya, D. J. Thiruvadigal,
and S. M. Jaya,
Chem. Phys. Lett. 609, 76 (2014).
|
1810.07722 | 1 | 1810 | 2018-10-17T18:18:14 | Hot electron heatsinks for microwave attenuators below 100 mK | [
"physics.app-ph"
] | We demonstrate improvements to the cooling power of broad bandwidth (10 GHz) microwave attenuators designed for operation at temperatures below 100 mK. By interleaving 9-$\mu$m thick conducting copper heatsinks in between 10-$\mu$m long, 70-nm thick resistive nichrome elements, the electrical heat generated in the nichrome elements is conducted more readily into the heatsinks, effectively decreasing the thermal resistance between the hot electrons and cold phonons. For a 20 dB attenuator mounted at 20 mK, a minimum noise temperature of $T_{n} \sim$ 50 mK was obtained for small dissipated powers ($P_d <$ 1 nW) in the attenuator. For higher dissipated powers we find $T_n \propto P_{d}^{(1/4.4)}$, with $P_{d} =$ 100 nW corresponding to a noise temperature of 90 mK. This is in good agreement with thermal modeling of the system and represents nearly a factor of 20 improvement in cooling power, or a factor of 1.8 reduction in $T_n$ for the same dissipated power, when compared to a previous design without interleaved heatsinks. | physics.app-ph | physics | Hot electron heatsinks for microwave attenuators below 100 mK
Jen-Hao Yeh,1,2,a) Rui Zhang,1,2 Shavindra Premaratne,1,2 Jay LeFebvre,3,b) F. C.
Wellstood,2,4 and B. S. Palmer1,2
1Laboratory for Physical Sciences, 8050 Greenmead Drive, College Park, Maryland, 20740, USA
2Department of Physics, University of Maryland, College Park, Maryland, 20742, USA
3Department of Physics and Astronomy, University of California, Riverside, California, 92521, USA
4Joint Quantum Institute and Center for Nanophysics and Advanced Materials, University of Maryland,
College Park, Maryland, 20742, USA
We demonstrate improvements to the cooling power of broad bandwidth (10 GHz) microwave
attenuators designed for operation at temperatures below 100 mK. By interleaving 9-m thick
conducting copper heatsinks in between 10-m long, 70-nm thick resistive nichrome elements,
the electrical heat generated in the nichrome elements is conducted more readily into the
heatsinks, effectively decreasing the thermal resistance between the hot electrons and cold
phonons. For a 20 dB attenuator mounted at 20 mK, a minimum noise temperature of 𝑇n ~ 50 mK
was obtained for small dissipated powers (𝑃d < 1 nW) in the attenuator. For higher dissipated
1/4.4, with 𝑃d = 100 nW corresponding to a noise temperature of 90 mK.
powers we find 𝑇n ∝ 𝑃d
This is in good agreement with thermal modeling of the system and represents nearly a factor of
20 improvement in cooling power, or a factor of 1.8 reduction in 𝑇n for the same dissipated
power, when compared to a previous design without interleaved heatsinks.
a) Electronic mail: [email protected]
b) This research was performed while Jay LeFebvre was at Department of Physics, University of Maryland, College Park,
Maryland, 20742, USA.
The field of superconducting quantum computing
has advanced to a point where systems with multiple
qubits are now available.1,2 Essential advances that
have enabled the development of multi-qubit systems
include improvements in the energy relaxation time 𝑇1
and coherence time 𝑇2, which have risen by five orders
of magnitude over the past two decades.3,4 As 𝑇1
increases, 𝑇2 becomes increasingly sensitive to the
presence of dephasing.5 -- 7 A common source of
dephasing in superconducting qubits that use a circuit
quantum electrodynamics (QED) architecture8
is
fluctuations in the number of photons in the cavity
used to either measure the state of the qubit or act as a
quantum bus.6,9 -- 11 Since remarkably small fluctuations
in photon number can cause significant dephasing
rates, it is essential to isolate the cavity from sources
of thermally generated photons in the input and output
control lines.12
To
lines
connected to a superconducting circuit QED cavity, a
series of microwave attenuators or directional couplers
are commonly installed at the low-temperature stages
in the measurement apparatus. These components
reduce thermal noise from higher temperature stages,
the microwave control
thermalize
1
5 − 𝑇p
5)
𝑃e−p = 𝑉𝛴(𝑇e
but when microwave signals are applied to control or
read-out the device, electrical power will be dissipated
in the resistive elements.13 At millikelvin temperatures,
a surprisingly small level of dissipated power can
cause the temperature 𝑇e of the electrons in a thin-film
normal metal element to increase well above the
temperature 𝑇p of the phonons.14,15 In the steady state,
this "hot-electron" effect can be described by15
(1)
where 𝑃e−p is the rate at which heat flows from the
electrons to the phonons, 𝑉 is the volume of the
normal metal, and 𝛴 is the electron-phonon coupling
constant, which is a material-dependent parameter.
Since 𝑃e−p results in 𝑇e increasing above 𝑇p , a
resistive element that dissipates electrical power will
produce more Johnson-Nyquist thermal noise, and
cause more dephasing, than if it were not dissipating
power. Hence, when designing an attenuator for
operation at millikelvin temperatures, it is important to
achieve a large effective coupling (𝑉𝛴) between the
electrons and phonons, such that 𝑇e is close to 𝑇p .12
Since 𝛴 is fixed by the choice of material, this is most
easily done by increasing the volume 𝑉of the normal
metal that is dissipating power.
recently
reported
results on heating
FIG. 1. (a) Circuit diagram for a single T-pad attenuator
cell with 10 dB power attenuation and 50 Ω characteristic
impedance. (b) Schematic drawing of a 10 dB cell for the non-
interleaved design and (c) for the design with resistive elements
interleaved by normal metal heatsinks. Red areas are resistive
elements, and blue areas are conducting normal metal. Note: (b)
and (c) are not drawn to scale, and (c) does not show the exact
number of resistive elements or heatsinks. (d) False-colored
scanning electron microscope (SEM) image of the Cu (blue)
heatsinks on top of the NiCr (red) resistor and quartz substrate
(gray). There is a 20-μm long Cu heatsink for each 𝐿R = 10 μm
long resistor. (e) SEM image of electroplated-Cu heatsinks with a
thickness of 9 μm.
We
in
broadband microwave attenuators at cryogenic
temperatures.12 These attenuators were constructed
from 10 dB attenuation cells with a T-pad
configuration (see Fig. 1(a)) and had a characteristic
impedance 𝑍0 = 50 Ω and a flat frequency response
up to 10 GHz. The resistive elements in each cell were
embedded in a coplanar waveguide geometry (Fig.
1(b)) and patterned from a sputtered 70 ± 5 nm thick
nichrome (NiCr) film on a single-crystal quartz
substrate, giving a sheet resistance 𝑅s = 25 ± 5 Ω/□.
To connect the resistors together, a patterned highly
conducting thin film of Ag was used. By using a
transmon-cavity system as a sensitive average-photon-
number thermometer,16 a minimum noise temperature
𝑇n ≤ 50 mK was obtained at low dissipated powers
𝑃d . However, for 𝑃d > 0.1 nW, the output noise
1/5.4
temperature of the attenuator increased as 𝑇n ∝ 𝑃d
.
Both this power law and finite element simulations of
the thermal response of the attenuator suggested that
the
thermal resistance between
the heat was not efficiently conducted out of the
nichrome, resulting in the electrons in the NiCr
resistors being driven out of equilibrium with the
phonons. If
the
electrons and the phonons could have been decreased,
a lower noise temperature would have been achieved
when dissipating power. In this article, we demonstrate
how the cooling power of the attenuator can be
improved by an order of magnitude by modifying the
layout to increase the effectiveness of the hot-electron
heatsinks.
First, to better conduct heat out of the NiCr resistor
elements and into the cold conducting heatsinks,15,17,18
we have divided the resistor elements in the attenuator
circuit (see Fig. 1(b)) into many small sections (see
Fig. 1(c)). The 𝐿R = 10 μm length of the individual
NiCr sections was chosen based on a detailed finite
element simulation. We can understand what sets this
length scale by examining the heat flow in a simplified
geometry. In particular, consider two cold normal-
metal heatsinks on each end of a resistor with length
𝐿R and cross-sectional area 𝐴R. If all of the power 𝑃R
dissipated
to and
distributed uniformly in the heatsinks, the temperature
of the electrons in the heatsinks will increase to
is conducted
the resistor
in
𝑇H ≲ (𝑇p
5 +
1
5
)
,
𝑃R
𝑉H𝛴H
(2)
where 𝑉H is the total volume of the two heatsinks and
𝛴H is the electron-phonon coupling constant for the
material used to construct the heatsinks. If we ignore
the emission of phonons by the electrons in the
nichrome (see supplementary material for a more
complete discussion), then heat generated in the
nichrome will only flow into the heatsinks by thermal
conduction via the electrons, and the maximum
electron temperature 𝑇R
will be in the center of the
resistor, furthest from the two heatsinks. For good
cooling, 𝑇R
should not be much larger than 𝑇H ,
and we choose the specific temperature criterion
(max) < √2𝑇H for convenience. Using this criterion
𝑇R
and Eq. (2), one finds an upper bound on the resistor
length
(max)
(max)
𝐿R <
(𝑇p
5 +
4𝜅e𝐴R
𝑃R
2
5
)
𝑃R
𝑉H𝛴H
,
(3)
where 𝜅e𝑇e is the thermal conductivity of the resistor
due to the electrons. For design values 𝑇p = 20 mK,
𝑃R = 100 nW, 𝜅e = 0.02 Wm-1K-2, 𝐴R = 56 μm2 ,
𝑉H = 1.4 × 105 μm3 , and 𝛴H = 3 × 107 Wm−3K−5 ,
we find 𝐿R < 10 μm. The implication is that heat can
2
be effectively removed from the NiCr electrons at low
temperatures if the resistor sections have a length less
than 𝐿R . On the other hand, we require that the
resistors in our T-pad attenuator circuit have specific
values (see Fig. 1(a)). To achieve this, for 𝑅1 and 𝑅2
we used 83 sections of 10-μm long resistive NiCr
elements interleaved with highly-conductive large
volume heatsinks (see Fig. 1(d)), while 𝑅3 and 𝑅4 each
had 208 interleaved resistor elements.
FIG. 2. (a) Photo of 20 dB attenuator packaged in a Cu
box (lid not shown). (b) Magnitude of the scattering parameters
𝑆112 (red), 𝑆212 (black), and 𝑆222 (blue) for the 20 dB
attenuator.
The second change to the attenuator involved
increasing the volume 𝑉H of the hot-electron heatsinks.
Unfortunately, from Eq. (2) one can see that the
electron temperature depends very weakly on the
volume. To greatly increase the heatsink volume, thick
Cu heatsinks were grown via electroplating.19 Before
performing the electroplating, the individual NiCr
elements were covered with a patterned layer of
photoresist. A 300 nm thick Cu seed layer was then
electron-beam evaporated over
the entire wafer,
providing good electrical conduction to the electrodes.
Next, an 11 µm thick photoresist layer was patterned
to define the areas for Cu electroplating. After
electroplating,
layers were
removed and the exposed regions of the Cu seed layer
were etched away. Each resulting Cu heatsink had an
two photoresist
the
average thickness of 9 m and a lateral length of 20
m between the NiCr elements (see Fig. 1(e)).
the
silver
epoxy. Room
transmission 𝑆212 of
For testing, a 20 dB attenuator chip was
packaged in a copper box (see Fig. 2(a)). The back of
the chip and the two ground planes were secured to the
box using
temperature
measurements of
the
packaged device showed an attenuation of (20 ± 2)
dB from 0 to 10 GHz, with low levels of reflection
𝑆112 and 𝑆222 from the input and output ports (see
Fig. 2(b)).
To measure the effective noise temperature of the
attenuator at millikelvin temperatures, we connected
the attenuator to the input line of an Al microwave
cavity that contained a transmon qubit.20 The transmon
had a transition frequency 𝜔q/2𝜋 = 3.67 GHz and
consisted of a single Al/AlOx/Al Josephson junction
on a sapphire substrate. The
three-dimensional
superconducting aluminum cavity had a resonance
frequency 𝜔c/2𝜋 = 7.96 GHz. The transmon-cavity
system and 20 dB attenuator were mounted on the
mixing chamber stage of a dilution refrigerator with a
20 mK base temperature. To thermalize the signals
from the 300 K stage to the tested attenuator, an
additional 20 dB attenuator was mounted on the 3 K
stage, and a 30 dB attenuator on the 70 mK stage (see
Fig. 3(a)).12 In a separate test for higher dissipated
power (discussed below), the 30 dB attenuator was
replaced with a 20 dB one. We extracted the qubit
dephasing rate 𝛤φ and then used 𝛤φ to estimate the
average photon number 𝑛th stored in the cavity.11,12,16
To increase the sensitivity of the qubit to thermal noise
from the 20 dB attenuator, we made the coupling
strength of the input cavity port larger than the output
port (i.e. 𝑄in = 1.8 × 104 ≪ 𝑄out = 7.9 × 104).
To find 𝛤φ, we measured the qubit relaxation
time 𝑇1 and spin-echo coherence time 𝑇2 . The qubit
dephasing rate 𝛤φ was then extracted using 𝛤φ =
−1 − (2𝑇1)−1 . When no additional heating was
𝑇2
applied to the attenuator, we measured mean values of
〈𝑇1〉 = 14.2 ± 0.1 μs and 〈𝑇2〉 = 26.7 ± 0.8 μs, where
the error bars are the standard deviations of the mean
of the 5 measurements. These results led to 〈𝛤φ〉 =
(2 ± 1) × 103 s−1. By using the relation11
2
)
2𝑖𝜒
𝜅
+
8𝑖𝑛th𝜒
𝜅
𝛤φ =
Re [√(1 +
𝜅
2
− 1]
(4)
along with measurements of the dispersive shift
𝜒/2𝜋 = −0.34 MHz and
the cavity decay rate
𝜅/2𝜋 = 0.7 MHz, we found that in this case the
3
average number of photons in the read-out cavity was
𝑛th = (1.0 ± 0.7) × 10−3 . Given the total 70 dB
attenuation in the input line, 𝑛th has a lower bound of
~10−4. Our result of 𝑛th is about 5 times larger than
that recently reported in a system in which the
microwave signal to a one-port cavity was filtered by a
bandpass cavity attenuator.21 For an attenuator with
temperature 𝑇n , we expect 𝑛th =
output noise
−1
and find 𝑇n ≤ 55 mK, similar to
[exp (
our previous design.12 We note that fluctuations in our
measured 𝑇1 and 𝑇2 produce significant uncertainty in
𝛤φ and the relatively larger fluctuations in 𝑇2 suggest
that other sources of dephasing may be present.
ℏ𝜔c
𝑘𝐵𝑇n
) − 1]
FIG. 3: (a) Schematic of attenuators along the input line
and the typical temperatures of the stages where they are mounted.
In separate cool-downs, a 20 dB or 30 dB attenuator was used at
the 70 mK stage. (b) Effective noise temperature 𝑇n and average
photon number 𝑛th in the cavity versus dissipated power 𝑃d in a 20
dB attenuator. Red and blue triangles are data obtained when a 20
dB or 30 dB attenuator, respectively, was used at the 70 mK stage.
Black circles are data from an attenuator with the previous
design.12 Red, blue, and black curves are from finite-element
thermal simulations for the corresponding attenuator and input line
layout. Green filled squares are data obtained on a commercial
cryogenic
from XMA
Corporation).
To evaluate the cooling power of the attenuator, we
extracted the effective noise temperature 𝑇n while
continuously dissipating power in the attenuator by
applying a 1 GHz drive. Figure 3(b) shows 𝑇n versus
the dissipated power 𝑃d for the 20 dB attenuator in
(2082-6243-20-CRYO
attenuator
4
improvement between our
In our previous design without
two different cool-downs (blue and red triangles), our
previous 20 dB cryogenic attenuator12 (black circles),
and a 20 dB commercial cryogenic attenuator (2082-
6243-20-CRYO) from XMA Corporation (green
squares). To measure
the attenuator at higher
dissipated powers (red triangles), we reduced the total
attenuation in the refrigerator by replacing a 30 dB
attenuator on the 70 mK cold plate with a 20 dB one.
We also note that a traveling wave parametric
amplifier22 was used in the output line to amplify the
cavity read-out signals while collecting the low power
data set (blue triangles). Examining Fig. 3(b), one sees
that it takes about 100 nW of dissipated power to drive
the new attenuator to 100 mK. This power is about 10
times larger than the power required to drive the old
version (black circles) to 100 mK, and 50 times larger
than that of the commercial attenuator (green squares).
The amount of
two
attenuators is mainly due to the total volume 𝑉H of the
new heatsinks being 260 larger than that of the
previous design.
Finally, we can use the data in Fig. 3(b) to extract
the electron-phonon coupling strengths of different
materials.
the
interleaved heatsinks, simulations showed that the heat
flow out of the NiCr resistors was the major bottleneck
in cooling
the output noise
temperature was sensitive to the electron-phonon
and 𝛴NiCr = 5 ×
coupling
108 Wm−3K−5 was found (see the black curve in Fig.
2(b)).12 For the new design, where we anticipate more
of the dissipated heat generated in the NiCr resistors to
conduct into the heatsinks, we are more sensitive to
the coupling of the electrons with the phonons in the
Cu heatsinks. Setting the electron-phonon coupling
strength of Cu to 𝛴Cu = 3 × 107 Wm−3K−5 yields
good
finite-element
simulations and the data (see blue and red curves in
Fig. 3(b)). We note that this value of 𝛴Cu is
𝛴Cu ≈
comparable
108 Wm−3K−5 .15 This suggests
that additional
increases in the cooling power can be achieved by
further enlarging the volume of the Cu heatsinks.
In summary, we have developed microwave
attenuators with increased cooling power that are
suitable for use with quantum devices below 100 mK.
This improvement was achieved by decreasing the
length that dissipated heat has to travel in each resistor
element and interleaving large-volume hot-electron
heatsinks. These simple components and related
techniques for improving thermalization of signals at
strength of NiCr,
the electrons, so
previous
agreement
between
our
to
reports,
10 P. Bertet, I. Chiorescu, G. Burkard, K. Semba, C. J.
P. M. Harmans, D. P. DiVincenzo, and J. E. Mooij,
Phys. Rev. Lett. 95, 257002 (2005).
11 A. A. Clerk and D. W. Utami, Phys. Rev. A 75,
042302 (2007).
12 J.-H. Yeh, J. LeFebvre, S. Premaratne, F. C.
Wellstood, and B. S. Palmer, J. Appl. Phys. 121,
224501 (2017).
13 B. Suri, Z. K. Keane, R. Ruskov, L. S. Bishop, C.
Tahan, S. Novikov, J. E. Robinson, F. C. Wellstood,
and B. S. Palmer, New J. Phys. 15, 125007 (2013).
14 M. L. Roukes, M. R. Freeman, R. S. Germain, R. C.
Richardson, M. B. Ketchen, and S. State, Phys. Rev.
Lett. 55, 422 (1985).
15 F. C. Wellstood, C. Urbina, and J. Clarke, Phys.
Rev. B 49, 5942 (1994).
16 R. J. Schoelkopf, A. A. Clerk, S. M. Girvin, K. W.
Lehnert, and M. H. Devoret, in Quantum Noise
Mesoscopic Phys., edited by Y. V. Nazarov (Springer
Netherlands, 2003), pp. 175 -- 203.
17 R. C. Ramos, M. A. Gubrud, A. J. Berkley, J. R.
Anderson, C. J. Lobb, and F. C. Wellstood, IEEE
Trans. Appiled Supercond. 11, 998 (2001).
18 J. Pleikies, O. Usenko, R. Stolz, L. Fritzsch, G.
Frossati, and J. Flokstra, Supercond. Sci. Technol. 22,
114007 (2009).
19 J. Gobet, F. Cardot, J. Bergqvist, and F. Rudolf, J.
Micromechanics Microengineering 3, 123 (1993).
20 H. Paik, D. I. Schuster, L. S. Bishop, G. Kirchmair,
G. Catelani, A. P. Sears, B. R. Johnson, M. J. Reagor,
L. Frunzio, L. I. Glazman, S. M. Girvin, M. H.
Devoret, and R. J. Schoelkopf, Phys. Rev. Lett. 107,
240501 (2011).
21 Z. Wang, S. Shankar, Z. K. Minev, P. Campagne-
Ibarcq, A. Narla, and M. H. Devoret, "Cavity
Attenuators
Superconducting Qubits,"
submitted to Phys. Rev. Applied.
22 C. Macklin, K. O'Brien, D. Hover, M. E. Schwartz,
V. Bolkhovsky, X. Zhang, W. D. Oliver, and I.
Siddiqi, Science 350, 307 (2015).
23 P. K. Day, H. G. LeDuc, B. A. Mazin, A.
Vayonakis, and J. Zmuidzinas, Nature 425, 817
(2003).
24 C. Enss, Cryogenic Particle Detection (Springer,
2005).
25 F. Giazotto, T. T. Heikkilä, A. Luukanen, A. M.
Savin, and J. P. Pekola, Rev. Mod. Phys. 78, 217
(2006).
for
millikelvin temperatures will be useful for increasing
the dephasing times of superconducting qubits, and
possibly improving the performance of other types of
millikelvin devices.23 -- 25
SUPPLEMENTARY MATERIAL
See supplementary material for the analysis of heat
flow in a simplified geometry in order to estimate the
length 𝐿R of the resistor sections.
ACKNOWLEDGMENTS
The authors gratefully acknowledge Warren Berk
for assisting with fabrication of the Cu heatsinks, and
MIT Lincoln Laboratory (Greg Calusine and W. D.
Oliver) for providing a traveling wave parametric
amplifier. F.C.W. acknowledges support from the
Joint Quantum Institute and the State of Maryland
through the Center for Nanophysics and Advanced
Materials.
REFERENCES
1 G. Wendin, Reports Prog. Phys. 80, 106001 (2017).
2 J. M. Gambetta, J. M. Chow, and M. Steffen, NPJ
Quantum Inf. 3, 2 (2017).
3 W. D. Oliver and P. B. Welander, MRS Bull. 38, 816
(2013).
4 M. Reagor, W. Pfaff, C. Axline, R. W. Heeres, N.
Ofek, K. Sliwa, E. Holland, C. Wang, J. Blumoff, K.
Chou, M. J. Hatridge, L. Frunzio, M. H. Devoret, L.
Jiang, and R. J. Schoelkopf, Phys. Rev. B 94, 014506
(2016).
5 C. Rigetti, J. M. Gambetta, S. Poletto, B. L. T.
Plourde, J. M. Chow, A. D. Córcoles, J. A. Smolin, S.
T. Merkel, J. R. Rozen, G. A. Keefe, M. B. Rothwell,
M. B. Ketchen, and M. Steffen, Phys. Rev. B 86,
100506(R) (2012).
6 A. P. Sears, A. Petrenko, G. Catelani, L. Sun, H.
Paik, G. Kirchmair, L. Frunzio, L. I. Glazman, S. M.
Girvin, and R. J. Schoelkopf, Phys. Rev. B 86,
180504(R) (2012).
7 F. Yan, S. Gustavsson, A. Kamal, J. Birenbaum, A.
P. Sears, D. Hover, T. J. Gudmundsen, D. Rosenberg,
G. Samach, S. Weber, J. L. Yoder, T. P. Orlando, J.
Clarke, A. J. Kerman, and W. D. Oliver, Nat.
Commun. 7, 12964 (2016).
8 A. Blais, R.-S. Huang, A. Wallraff, S. M. Girvin, and
R. J. Schoelkopf, Phys. Rev. A 69, 062320 (2004).
9 D. I. Schuster, A. Wallraff, A. Blais, L. Frunzio, R.-
S. Huang, J. Majer, S. M. Girvin, and R. J.
Schoelkopf, Phys. Rev. Lett. 94, 123602 (2005).
5
|
1903.11954 | 1 | 1903 | 2019-03-28T13:23:34 | Solar Energy Conversion and the Shockley-Queisser Model, a Guide for the Perplexed | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | The Shockley-Queisser model is a landmark in photovoltaic device analysis by defining an ideal situation as reference for actual solar cells. However, the model and its implications are easily misunderstood. Thus, we present a guide to help understand and avoid misinterpreting it. Focusing on the five assumptions, underlying the model, we define figures of merit to quantify how close real solar cells approach each of these assumptions. | physics.app-ph | physics | S-Q Guide for the Perplexed 10-10-2018
Solar Energy Conversion and the Shockley-Queisser Model,
a Guide for the Perplexed
Jean-Francois Guillemoles1, Thomas Kirchartz2,3, David Cahen4, and Uwe Rau2
1CNRS, UMR 9006, Institut Photovoltaique d'Ile de France (IPVF), Palaiseau, France
2IEK5-Photovoltaik, Forschungszentrum Jülich, 52425 Jülich, Germany
3Fac. of Engineering and CENIDE, Univ. of Duisburg-Essen, Carl-Benz-Str. 199, 47057 Duisburg, Germany
4Department of Materials and Interfaces, Weizmann Institute of Science, Rehovoth 76100, Israel
e-mails for correspondence:
[email protected] ; [email protected] ; [email protected]; [email protected]
Abstract
The Shockley-Queisser model is a landmark in photovoltaic device analysis by
defining an ideal situation as reference for actual solar cells. However, the model and its
implications are easily misunderstood. Thus, we present a guide to help understand and avoid
misinterpreting it. Focusing on the five assumptions, underlying the model, we define figures
of merit to quantify how close real solar cells approach each of these assumptions.
Introduction
In 1961 Shockley and Queisser1 (SQ) analyzed the limits of photovoltaic energy conversion,
using the basic thermodynamic principle of detailed balance instead of phenomenological
approaches, used earlier.2-4 The final result of their analysis is commonly referred to as the 'SQ-
limit'. While arguably the most important theoretical contribution to photovoltaic energy
conversion, the paper also relies on a highly idealized model for solar cells, using substantially
simplifying assumptions. Therefore, only within the assumptions of their model (denoted the
SQ-model in the following) does the term 'SQ-limit' make sense. In view of the emergence of
promising new photovoltaic absorber materials and devices with very high efficiencies5 with
various claims of 'exceeding or approaching the SQ-limit',6,7 we will critically discuss the
connection of the SQ-model to real world solar cells and will explain what 'close' to the SQ-
model means.
First, we briefly describe the SQ-model in its initial form by illustrating its three
fundamental steps, noting the energy losses associated with each of these. We then describe the
five assumptions that are the essence of the model (Table 1). Subsequently, we examine how
each of these assumptions compares to more realistic situations, discuss experimentally
1
S-Q Guide for the Perplexed 10-10-2018
measurable figures of merit (FoMs) and quantify how real-world solar cells differ from the
ideal model, thereby providing guidelines for effective use of the SQ-model.
Description of SQ-model and its assumptions
Figure 1a shows the setting of the SQ-model in its original form. The solar cell interacts with
the surroundings by exchanging light particles (photons) with the sun and with the ambient.
Furthermore, the cell exchanges electrons with the external electrical circuit and heat with a
temperature reservoir to maintain the cell temperature Tcell constant and equal to the ambient
temperature Tamb, such that without solar irradiation cell and ambient are in thermodynamic
equilibrium. We are not considering light concentration or a restriction of the angle of optical
interaction of the cell with the ambient. With this setting we aim at a simple picture that,
however, is compatible with more elaborate, thermodynamic descriptions of the solar cell
working principles.8-10
The photovoltaic absorber in the SQ-model is a semiconductor, described by two groups
of electronic energy levels that extend throughout the material (cf. Fig. 1b). The lower one --
called valence band, VB -- is filled with electrons, while the upper one -- the conduction band,
CB -- is initially empty. The energy difference between the edges EC and EV of the conduction
and valence bands is the band gap energy Eg (Fig. 1b). If sufficient energy is supplied, an
electron can be promoted from VB to CB, leaving behind an empty state in the VB. The electron
in the CB and the empty state in the VB (called hole) now behave as "free" charge carriers that
can move in their respective bands (cf. Fig 1b).
The solar radiation is described by photons with a distribution of energies, the solar
spectrum (NB: while differing situations may require different solar spectra, the present
analysis remains valid for any). Depending on their energy, these photons may have enough
energy to create free electrons and holes. These electron-hole pairs, generated by photon
absorption, can also annihilate themselves by releasing their energy in the form of photons
(radiative recombination).
The interaction of the photon with the semiconductor and the photovoltaic action in the
solar cell proceed in three stages (1-3, as sketched in Fig. 1b), with relevant time-scales shown
in Fig. 1c and Tab. 1.
(A -- optical) absorption of a photon with creation of an electron-hole pair,
(B -- thermal) relaxation of this electron-hole pair towards EC and EV,
2
S-Q Guide for the Perplexed 10-10-2018
(C -- electronic) extraction of the electron and hole at two different contacts, or their radiative
recombination followed by emission of a photon.
Note that the extraction of charge carriers at different contacts requires a built-in
asymmetry that separates electrons from holes and that may be achieved electrostatically via a
pn-junction. However, while the pn-junction is featured in the title of the original SQ paper and
used for illustration purposes in Fig. 1b, it is by no means a necessary requirement for an
efficient solar cell.8,11
Within the SQ-model, the three stages are defined by 5 assumptions:
OPTICAL: It is assumed that for impinging photons with photon energy E > Eg, the photon is
absorbed, whereas for E < Eg photons do not interact with the solar cell at all (assumption 1),
i.e., the absorptivity A(E), a measure of the material's ability to absorb radiation, is a step
function, 0 for E < Eg and 1 for E > Eg. Hence, the first energy loss is due to the solar cell's
transparency for E < Eg. Also, absorption of a photon with E > Eg generates precisely one
electron-hole pair that contributes to the short-circuit current JSC (assumption 2).
THERMAL: The electron-hole pair loses (to the absorber material's lattice) all excess energy
above Eg, i.e., the pair relaxes in sub ps timescales to the average energy of a thermalized
electron-hole pair (in thermal equilibrium with the cell at
). The underlying assumption 3
is that in the solar cell all relaxed electron-hole pairs are at the same temperature
.
ELECTRONIC: At this point, one of two things can happen to the electron-hole pair: Either the
electron and the hole are collected at their respective contacts or they recombine radiatively by
emission of a photon (this is the only allowed recombination mechanism, assumption 4). Note
that for an actual recombination, the photon must be emitted to the ambient because
reabsorption of the photon in the solar cell creates a new electron-hole pair and restarts the
whole process from the beginning (photon recycling).12 Both phenomena (recombination and
extraction) occur typically on ns to µs timescales -- collection usually being faster than
recombination. Emission of photons from radiative recombination causes a photon flux,
described by the product of the absorptivity and the black body spectrum of the solar cell. The
connection between absorption and emission of a semiconductor is a result of the principle of
detailed balance, which states that every microscopic process must have the same rate as its
inverse process in thermal equilibrium.13 Otherwise, thermal equilibrium could not be reached.
3
cellTcellT S-Q Guide for the Perplexed 10-10-2018
The SQ-theory, assumes that the rate constants derived from the principle of detailed balance
for thermal equilibrium are also valid in the non-equilibrium situation.8
The collection of photogenerated carriers implies that the solar cell has two different
external contacts that can support an external voltage V and carry a current J. Assumption 5
states that each contact is ideal because each exchanges only one carrier type (electrons or
holes) with the absorber (selective contact) and because it has negligible resistance.
Nevertheless, collection of the electron-hole pair implies that its total energy is reduced from
the band gap energy Eg to qV, the electrical work of transferring a charge carrier between the
contacts. We denote this loss of potential energy 'isothermal dissipation' as it generates heat in
the solar cell without a change of the temperature of electrons and holes (unlike during
thermalization) and the carriers do not recombine (unlike during radiative recombination). Note
that this loss can be subdivided further into different reversible and non-reversible
contributions.10,14 We have listed four energy loss mechanisms in the SQ-model in Table I and
illustrated in Fig. 2a which also shows the maximum efficiency of 30 % (referring to a 5800 K
black body solar spectrum, likewise a value of 33 % would hold for the more complex terrestrial
spectrum) at an optimum band gap energy. Fig. 2b shows the specific share of the four energy
losses for a specific band gap energy. It also illustrates how the output power is maximized by
the proper choice of the voltage minimizing the sum of emission and isothermal dissipation
losses.
The efficiencies depicted in Fig. 2a based on the SQ-model represent a separation line
between the so-called third generation photovoltaics15 (where higher efficiencies can be
achieved by bypassing at least one of the 5 assumptions in a way to reduce power losses) and
the 'normal' single junction solar cells with efficiencies below the SQ-case (for which the 5
assumptions are approached but not reached resulting in higher power losses than implied by
the SQ model, as described by the FoMs, below).
Accounting for losses in real, single absorber solar cells
We now describe departures from the ideal SQ-model for real-world single junction solar
cells by successively relaxing the 5 assumptions. Given that quantitatively accounting for the
losses in a physically meaningful way becomes quite detailed and technical in parts we
encourage the casual reader to jump directly to the "Consequences" section. For readers
preferring to quantitatively understand the consequences of these departures from the SQ-
4
S-Q Guide for the Perplexed 10-10-2018
model, we express the current versus voltage (JV) characteristics of the solar cell in terms of a
simple balance equation (Eq. (1) in the Box) expressing that the net total electrical current J
extracted from the cell equals the short circuit photocurrent 𝐽SC, (a gain) minus the
recombination or diode current (a loss). The diode current is a product of a prefactor (the
saturation current 𝐽0 that measures the recombination loss) and a rectifying term varying
exponentially with voltage (see Fig. 2b and c). From equation 1 all further information like the
open-circuit voltage VOC, the maximum output power 𝑃max, and the efficiency 𝜂 are derived by
simple mathematics; the Box gives the key equations. The rectifying character of the JV-
characteristic, essential for an efficient photovoltaic device, is measured by the fill factor FF,
the ratio of the electrical power Pmax at the maximum power point divided by the product
JSCVOC. As long as equation 1 is valid, the fill factor - at a given temperature 𝑇cell -- is a well-
defined, increasing, function of 𝑉OC, i.e. 𝐹𝐹 = 𝐹𝐹0(𝑉OC) (equation 6, Box).16
The SQ-model with assumptions 1-5 yields a short circuit current 𝐽SC
SQ(𝐸g) and a saturation
SQ(𝐸g) that only depend on 𝐸g (as material property). These two values determine the
current 𝐽0
SQ
output power 𝑃max
, for a given Tcell and a given solar spectrum, solely by 𝐸g, as shown in Fig.
2a). As discussed in the following, relaxing assumptions 1-5 leads to power losses by changes
in one or more of the three parameters 𝐽SC, 𝐽0 , Tcell and also to changes of the overall J-V shape
as discussed in the following.
(A) OPTICAL: No real solar cell can fulfill assumption 1, i.e, have a step function absorptivity
A(E), e.g. because of broadened CB and VB edges, called band tails17 resulting from (static)
structural disorder,18 due to the existence of charge transfer states in organic semiconductors,19
or simply because of a finite cell thickness and finite absorption coefficients. An obvious
SQ
problem in this context is to define the band gap to use in the SQ model to determine 𝐽SC
SQ
𝐽0
that may serve as reference values. Especially for semiconductors with (static) disorder the
and
exact definition or method to determine the band gap is problematic. To be consistent with the
SQ-model we recently proposed to use the derivative of the external photovoltaic quantum
efficiency with respect to photon energy as measure of photovoltaic band gap.20 In violation of
assumption 2, parasitic absorption of photons in contact layers or by free carriers (electrons) in
the optical absorber reduces the average number of photogenerated electron-hole pairs per
absorbed photon to < 1.
5
S-Q Guide for the Perplexed 10-10-2018
The consequence of violating assumptions 1 and 2 of the SQ-model are best understood in
terms of the external photovoltaic quantum efficiency 𝑄e
PV(𝐸), i.e., the probability that a photon
of energy E, impinging on the cell, generates an electron-hole pair that, under short-circuit
conditions, is extracted at the contacts. In general we have 𝑄e
PV(𝐸) < 𝐴(𝐸) and it is easily
understood that also this leads to reduction of the short circuit current.
Thus, violating assumptions 1 and 2 decreases the short circuit current from the ideal value
SQ
𝐽SC
QE
to the real value 𝐽SC
𝑆𝑄
PV(𝐸) and the electroluminescence of a solar cell,21 𝐽0
, determined by 𝑄e
𝑄e
PV(𝐸). Because of the reciprocity between
also changes to 𝐽0
QE
(which still
describes radiative loss to the ambient). It should be noted that the reciprocity applies to all
optical absorbers, nanophotonic or not, so that this analysis is generally valid. This point is
discussed in detail in ref. 22.
Summarizing the loss resulting from assumptions 1 and 2, we use the ratios of short-circuit
QE(= 𝐹em) ≤ 1 as
(= 𝐹SC) ≤ 1 and radiative emission loss currents 𝐽0
QE 𝐽SC
SQ⁄
SQ/𝐽0
currents 𝐽SC
two FoMs that describe departures from the ideal SQ-case. The best experimental values of FSC
are > 90% for top laboratory cells of all commercial types (95% for best c-Si cells),5 while
typical 𝐹em values range from ≈ 0.1 − 0.5 in GaAs or c-Si cells to ≪ 10−3 in amorphous or
organic cells.20 Note the different weight of 𝐹em entering only logarithmically in the open circuit
voltage and in the efficiency equation (Eqs. 3, 5; Box) as compared to 𝐹SC, which has an
approximately linear effect on the solar cell efficiency. Thus, a 10 % loss in 𝐹SC implies a little
bit more than 10 % loss in efficiency whereas a 90 % loss in 𝐹em involves a loss in VOC of
𝑘𝑇cell q⁄ × ln(10) ≈ 60 mV, i.e., < 10 % loss, if VOC > 600 mV.
(B) THERMAL: The most likely consequence of violating assumption 3 (electron-hole pairs,
absorber and contacts at the same temperature) is that the operation temperature 𝑇cell of real
solar modules outdoors23 is > 300 K (as assumed in the SQ-model) or than the standard-testing-
conditions (STC) temperature 𝑇cell
STC = 25 °C, used to rate efficiency. Because any recombination
current (radiative and non-radiative, expressed by the respective 𝐽0 -values) is thermally
activated, the corresponding recombination loss increases exponentially with increasing
temperature. Thus, the major effect of elevated temperature is reducing the open-circuit voltage
(see Eq. (3), Box), with second order effects on other parameters. The reduction of the annual
energy yield of modules outdoors is 2 -- 10 % depending on technology and location.23
6
S-Q Guide for the Perplexed 10-10-2018
(C) ELECTRICAL: An equivalent circuit model for the solar cell (Fig. 1d) helps to understand
the electrical losses. As long as assumptions 4 and 5 are valid, the circuit consists of a diode
where the saturation current 𝐽0
QE
is given by radiative emission. This implies that we are still in
the "radiative recombination only" situation, while in practical solar cells non-radiative
recombination of electron-hole pairs in the bulk and at interfaces (violation of assumption 4)
occurs and represents a major additional loss channel. With the notable exception of the GaAs
record cells, recombination losses in solar cells are entirely dominated by non-radiative
recombination, Thus, violation of assumption 4 requires replacing 𝐽0
real, the sum of non-radiative recombination and the still unavoidable radiative one. This
𝐽0
by the saturation current
QE
change generally implies an increase of recombination losses by orders of magnitude and a
corresponding decrease of the open-circuit voltage (Eq. 3, Box). A common FoM, the external
luminescence quantum efficiency 𝑄e
real has been extensively analyzed for various
lum = 𝐽0
QE/𝐽0
solar cell types with peak values of 𝑄e
lead-halide perovskite cells under operating conditions.25 But many solar cells, including the
lum= 20 % for the record GaAs devices,24 and ~ 5% for
ones based on crystalline Si, have 𝑄e
lum ≤ 1 % implying >120 mV VOC losses with respect to
the radiative limit. We note that in the present analysis we have assumed that the so-called diode
ideality factor 𝑛𝑖𝑑 for non-radiative recombination is unity (as for radiative recombination).
Violations of assumption 5, stating that electrons and holes can move freely everywhere in
the absorber towards/from their electron (hole) contact are, in the simplest case, described by
additional resistive elements in Fig. 1d, that change the shape of the JV-curve while hardly
affecting 𝐽SC and 𝑉OC. As illustrated in Fig. 2c, violating assumption 5 (e.g. increased series
resistance) will decrease FF below 𝐹𝐹0 and reduce the solar cell efficiency accordingly. Thus,
real). As presented in table II,
an especially simple figure of merit is 𝐹FF
res = 𝐹𝐹real/𝐹𝐹0(𝑉OC
res values are > 97% for record Si and GaAs solar cells, and range between over 90%
typical 𝐹FF
for polycrystalline thin film and near 85 % for polymer-based devices.26
Consequences
Returning to our initial question, we can now use the FoMs 𝐹SC, 𝐹em, 𝑄e
lum and 𝐹FF to
measure how close a solar cell is to the SQ-model in the dimensions associated with 4 out of
the 5 assumptions of the SQ-model (𝑇cell is kept at 300 K). This procedure goes beyond a simple
comparison of the photovoltaic parameters 𝐽SC, 𝑉OC, and 𝐹𝐹 with their respective SQ-reference
7
S-Q Guide for the Perplexed 10-10-2018
values.27 We stress that 𝐹SC, 𝐹em 𝑄e
lum , and 𝐹FF
rec should be experimentally determined on
complete devices. The SQ-situation is approached if all four quantities are close to unity.
Optimizing devices with respect to only one of the FoMs cannot be associated with the SQ-
limit because such optimization often goes at the expense of the others. This issue becomes
even more critical if measurements are performed on incomplete devices or on photovoltaic
absorbers only. For absorber materials, the external luminescence efficiency 𝑄e
frequently measured by means of photoluminescence,28 thus replacing electrical excitation as
lum is
in the proper experiment on completed devices, by optical excitation. However, optimizing
lum , e.g. by applying well-passivating (but possibly current-blocking) surface layers, may
𝑄e
guide us in the wrong direction. Instead of maximizing one FoM against all others, we need to
optimize all of them simultaneously and need to understand the trade-offs that prevent to put
all FoMs close to unity at the same time. Notably, a part of the FoMs can be extracted from
standard illuminated current-voltage and quantum efficiency measurements that are, e.g.,
tabulated in Ref. 5 for a set of recent record cells from different technologies. Table II and Fig.
3 show the FoMs 𝐹𝑆𝐶 and 𝐹FF
res and the values 𝑉OC
SQ
real/𝑉OC
and 𝐹𝐹0(𝑉OC
real)/ 𝐹𝐹0(𝑉OC
SQ) that are
used in Eq. (5) (Box) to calculate the efficiency ratio 𝜂real/ 𝜂0. Note that the FoMs 𝑄e
lum and
𝐹𝑒𝑚 cannot be obtained separately without an analysis of the luminescence emission of the
device.20 However, a presentation of different loss terms based on an incomplete set of FoMs
like Fig. 3 is already useful to identify strengths and weaknesses of different technologies with
a precise reference to the SQ-case.
Acknowledgements:
JFG thanks the French program of "investment for the future" (ANR-IEED-002-0). DC thanks
the Inst. PV d'Ile de France for a visiting professorship and the Ullmann family foundation (via
the Weizmann Inst) for support.
8
S-Q Guide for the Perplexed 10-10-2018
Table I: The three steps (A-C) of the SQ-model, their time scales and energy losses, associated with them, and
real solar cell departures from the five SQ-assumptions (Fig.1), quantified by five FoMs (see Box for symbols):
1. + 2. Relaxing assumption 1 towards a non-step function-like absorptivity A(E), combined with relaxing
assumption 2 (not all absorbed photons lead to electron-hole pairs that are collected) has two consequences: Firstly,
QE
the short circuit current 𝐽SC
, the SQ-model value. Secondly, the radiative
loss current 𝐽0
op
but < 1 if assumptions 1 and 2. are not fulfilled. 3. Deviation of the operating cell's temperature, 𝑇cell
SQ
𝑇cell
recombination (violation of assumption 4), is described by 𝑄e
the total recombination current 𝐽0
low charge-carrier mobilities and finite parallel resistances, in violation of assumption 5.
. Accordingly, we define two FoMs Fem and FSC, unity in the SQ-limit,
, and
=300K assumed in the SQ-model is expressed by 𝐹T = 𝑇cell
QE
lum, the ratio between the emitted radiation, 𝐽0
and
res describes fill factor losses due to non-zero series resistance,
SQ
of a real cell becomes smaller than 𝐽SC
SQ
becomes larger than 𝐽0
. 4. The occurrence of non-radiative
real. 5. The ratio 𝐹FF
𝑆𝑄 /𝑇cell
QE
op
Table 1: Stages and Assumptions of the SQ-model:
Stages A-C of SQ-model
Assumptions 1-5 of
Changes of
Figures of merit
(Time scales)
- Energy losses
A. OPTICAL
(1-10 fs)
- Loss of photons that are
not absorbed
B. THERMAL
(0.1-10 ps)
- Loss of excess kinetic
energy
C. ELECTRONIC
(0.1-1000 ns)
- Loss by emission of
photons
- Isothermal dissipation
loss during carrier
collection
SQ-model
diode
parameters
1. At Eg absorptivity
of photons in
absorber switches
from 0 to 1
2. Exactly one
electron-hole pair
per absorbed photon.
Each pair is collected
at short circuit.
3. Heat extraction
from the carrier
system such that the
carrier temperature
equals cell and
ambient temperature
4. Electron-hole
recombination is
only radiative
(emission of
radiation)
5. No Ohmic losses,
contacts are perfectly
selective
1. 𝐽SC
2. 𝐽0
3. 𝑇cell
QE
SQ → 𝐽SC
𝑆𝑄 → 𝐽0
𝑆𝑄 → 𝑇cell
𝑄𝐸
𝑜𝑝
𝑄𝐸 → 𝐽0
real
4. 𝐽0
5. 𝐹𝐹0(𝑉OC
real) →
𝐹𝐹cell
1. 𝐹SC = 𝐽SC
2. 𝐹em = 𝐽0
3. 𝐹T = 𝑇𝑐𝑒𝑙𝑙
QE
QE/𝐽SC
SQ/𝐽0
𝑆𝑄 /𝑇cell
𝑜𝑝
SQ
𝑄e
lum = 𝐽0
𝑄𝐸/𝐽0
real
res = 𝐹𝐹cell/𝐹𝐹0(𝑉OC
real)
4. 𝐹FF
9
S-Q Guide for the Perplexed 10-10-2018
Table II: The figures of merit discussed in table I for a range of record cells listed in supplementary table 1 and 2
in ref. 5. As a reference band gap, we use the photovoltaic band gap extracted from solar cell quantum efficiency
res as well as
data as described in ref. 20. From the typically published data on solar cells, the FoMs as 𝐹SC and 𝐹FF
lum and 𝐹em requires the
the product 𝑄e
determination of 𝐽0
by, e.g., an electroluminescence measurement, which is often unavailable for the top-
SQ) which result from the FoMs
performing cells. Also tabulated are the values 𝑉OC
and are used in Eq. (5) (Box) to calculate the efficiency ratio 𝜂real/ 𝜂0. Note that for the calculations of the SQ
reference values, we used an AM1.5G spectrum and we assumed nid = 1 for the calculation of FF0.
real are easily determined whereas separation of 𝑄e
real)/ 𝐹𝐹0(𝑉OC
lum𝐹em = 𝐽0
SQ/𝐽0
QE
SQ
real/𝑉OC
and 𝐹𝐹0(𝑉OC
𝑭𝑭𝟎(𝑽𝐎𝐂
⁄
𝐫𝐞𝐚𝐥)
𝐒𝐐)
𝑭𝑭𝟎(𝑽𝐎𝐂
𝑭𝑭𝟎(𝑽𝐎𝐂
𝐒𝐐)
𝐫𝐞𝐚𝐥) 𝑭𝑭𝟎(𝑽𝐎𝐂
89.3 %
89.5 %
99.8 %
85.2 %
86.8 %
98.2 %
85.2 %
87.0 %
98.0 %
89.2 %
90.3 %
98.8 %
87.0 %
89.5 %
97.3 %
89.5 %
91.3 %
98.0 %
85.8 %
90.6 %
94.7 %
Absorber
Band gap
𝐏𝐕
𝑬𝐠
GaAs
1.42 eV
c-Si
1.10 eV
Cu(In,Ga)Se2
1.12 eV
ABX3
1.55 eV
CdTe1 -- xSex
1.42 eV
QD
1.77 eV
OPV
1.62 eV
𝜼𝐫𝐞𝐚𝐥
𝜼𝐒𝐐
𝐐𝐄
𝑱𝐒𝐂
𝐒𝐐
𝑱𝐒𝐂
( mA/cm²)
( mA/cm²)
𝜼𝐫𝐞𝐚𝐥
⁄
𝜼𝐒𝐐
29.1 %
33.2 %
87.7 %
26.7 %
33.0 %
80.9 %
22.9 %
33.4 %
68.6 %
20.9 %
31.5 %
66.4 %
21.0 %
33.2 %
63.3 %
13.4 %
27.8 %
48.2 %
11.2 %
30.3 %
37.0 %
𝑭𝐒𝐂
29.8
32.1
92.8 %
42.6
44.3
96.2 %
38.8
43.9
88.4 %
24.9
27.3
91.2 %
30.2
32.1
94.1 %
15.2
20.5
74.2 %
19.3
24.9
77.5 %
𝐫𝐞𝐚𝐥)
𝑭𝑭𝐜𝐞𝐥𝐥
𝑭𝑭𝟎(𝑽𝐎𝐂
𝐫𝐞𝐬
𝑭𝐅𝐅
86.7 %
89.3 %
97.1 %
84.9 %
85.2 %
99.7 %
79.5 %
85.2 %
93.3 %
74.5 %
89.2 %
83.5 %
79.4 %
87.0 %
91.3 %
76.6 %
89.5 %
85.6 %
74.2 %
85.8 %
86.5 %
𝐫𝐞𝐚𝐥(mV)
𝑽𝐎𝐂
𝐒𝐐
𝑽𝐎𝐂
(mV)
𝐒𝐐⁄
𝐫𝐞𝐚𝐥 𝑽𝐎𝐂
𝑽𝐎𝐂
1130
1157
97.7 %
740
858
86.3 %
740
877
84.4 %
1120
1278
87.6 %
880
1157
76.1 %
1160
1484
78.2 %
780
1343
58.1 %
𝑭𝐞𝐦𝑸𝐞
𝐥𝐮𝐦
38.7 %
1.1 %
0.57 %
0.24 %
2.4 × 10-3
%
4.9 × 10-4
%
4.5 × 10-8
%
10
Box: Key equations in the SQ model:
Equation
Explanation
S-Q Guide for the Perplexed 10-10-2018
𝑱 = 𝑱𝐒𝐂−𝑱𝟎 [𝐞𝐱𝐩 (
𝒒𝑽
𝒏𝐢𝐝𝒌𝑻𝐜𝐞𝐥𝐥
) − 𝟏] (1)
𝑽𝐎𝐂 =
𝒏𝐢𝐝𝒌𝑻𝐜𝐞𝐥𝐥
𝒒
𝐥𝐧 (
𝑱𝐒𝐂
𝑱𝟎
+ 𝟏) (2)
𝐫𝐞𝐚𝐥 − 𝑽𝐎𝐂
𝑽𝐎𝐂
𝒌𝑻𝐜𝐞𝐥𝐥
𝐒𝐐 =
𝐥𝐧(𝑭𝐒𝐂 𝑭𝐞𝐦𝑸𝐞
𝐥𝐮𝐦) (3)
𝒒
𝜼 =
𝑷𝐦𝐚𝐱
𝑷𝐒𝐮𝐧
=
𝑱𝐒𝐂𝑽𝐎𝐂𝑭𝑭(𝑽𝐎𝐂,… )
(4)
𝑷𝐒𝐮𝐧
𝜼𝐫𝐞𝐚𝐥
𝜼𝐒𝐐
= 𝑭𝐒𝐂
𝐫𝐞𝐚𝐥 𝑭𝑭𝟎(𝑽𝐎𝐂
𝐫𝐞𝐚𝐥)
𝑽𝐎𝐂
𝐒𝐐)
𝐒𝐐 𝑭𝑭𝟎(𝑽𝐎𝐂
𝑽𝐎𝐂
𝒓𝒆𝒔 (5)
𝑭𝑭𝑭
𝑭𝑭𝟎
=
𝒒𝑽𝐎𝐂
𝒏𝐢𝐝𝒌𝑻𝐜𝐞𝐥𝐥
− 𝐥𝐧 (
𝒒𝑽𝐎𝐂
𝒏𝐢𝐝𝒌𝑻𝐜𝐞𝐥𝐥
𝒒𝑽𝐎𝐂
𝒏𝐢𝐝𝒌𝑻𝐜𝐞𝐥𝐥
+ 𝟏
+ 𝟎. 𝟕𝟐)
The current-voltage, JV, characteristics of a solar cell are given by a diode
current that is increased by the short circuit current 𝐽SC. Upon relaxing
assumptions 1-4, 𝐽0 , 𝐽SC, and 𝑇cell change from their SQ-values to the real cell
values. Here kTcell is thermal energy at the temperature of the solar cell. Note
that the ideality factor nid = 1 in the SQ-model but may deviate from one in real
cells. If solar concentration or emission angle restrictions29 are considered, it
can be taken into account by considering the appropriate values for J° and JSC.
The open-circuit voltage results from solving eqn (1) for J = 0. With successive
changes from the SQ-case to a real situation, the gain term 𝐽𝑆𝐶 decreases and
the loss term 𝐽0 increases, such that 𝑉OC decreases from the SQ- to its real solar
cell-value. Here, q is the elementary charge.
The loss of open-circuit voltage between that of the real device and the SQ-
lum
model, is described by the product of three FoMs (cf. Table 1) 𝐹SC. 𝐹em. 𝑄e
with values < 1. 𝑇cell is taken to be that in the SQ-model.
The efficiency of the solar cell is the maximum output power 𝑃max = (𝐽𝑉)max
, related to the power input 𝑃sun. Usually, 𝑃max is factorized by the product
𝐽SC𝑉OC𝐹𝐹, where the fill factor FF depends on VOC and various parameters that
change the general shape of the diode equation like resistive losses, and nid.
The ratio between the real efficiency and the SQ-value is defined with the help
of four FoMs. Note that 𝐹SC enters linearly as well as logarithmically via the
real (cf. Eq. (3)). The equation also considers the change
open circuit voltage 𝑉OC
of 𝐹𝐹0 from the SQ-case to the real situation as well as the change from
real) to 𝐹𝐹real, induced by resistive losses with the FoM 𝐹FF
res =
𝐹𝐹0(𝑉OC
𝐹𝐹real
𝐹𝐹0(𝑉OC
⁄
There is no analytical solution for the fill factor FF even in the simplest
situation of an ideal diode. However, very precise approximate equations16
provide a valid relation between FF0, the value without resistive losses, and
VOC.
real)
.
(6)
11
S-Q Guide for the Perplexed 10-10-2018
Fig. 1: a Schematic of the world of the SQ-model, with the sun at Tsun illuminating the solar cell at Tcell = 300 K,
also the temperature of the ambient (Tamb = Tcell), and the external circuit of the solar cell. b The three essential
steps of photovoltaic power conversion illustrated within an energy band diagram of a p/n-junction solar cell; p(n)
semiconductor, on right(left). The conversion process consists of three steps, (A) light absorption, (B) local
thermalization, directly after photogeneration, and (C) charge collection with further thermalization within the
semiconductor and contacts. Selectivity is illustrated as the flow of each of the two types of charge carriers
(electrons, blue, and holes, red) into a different contact. c Typical time scales of these energy losses, , for
illustrative purposes (NB, in Si solar cells, collection and recombination times could be much longer). Non
absorption loss, refers to photon whose energy is <Eg . d Simple equivalent (electrical engineering) circuit of a
solar cell. The SQ-model only requires the current source and the diode (the red photon represents radiative
recombination with the current Jem). Real solar cells are typically described by the addition a second diode,
representing non-radiative recombination with current Jnonr (indicated by the blue springs, representing heat
dissipation), a parallel or shunt resistance Rp and a series resistance Rs.
12
S-Q Guide for the Perplexed 10-10-2018
Fig. 2: a Illustration of within the SQ-model as function of band gap (always at maximum power point) using for
the solar spectrum a 5800 K black body spectrum normalized to 100mW/cm². b Energy losses for a given band
gap energy depicted as a function of photon flux vs. photon energy (adapted from ref. 10). By dividing the photon
flux by the elementary chare q and multiplying the energy with q, the axes can be also read as current density vs.
voltage. The black curve (JV) denotes the current vs. voltage curve, and the maximum output power is obtained
for a maximum area of the white rectangle (Power out), likewise a minimum area for the recombination and
isothermal dissipation losses. c Current-voltage curves of a solar cell and the power losses occurring upon relaxing
the SQ-assumptions 1+2 (combined), 4, and 5. The maximum output power (illustrated by the rectangles) reduces
𝑆𝑄
SQ
real. The cell temperature 𝑇cell is kept at the SQ-value 𝑇cell
stepwise from the SQ-value 𝑃max
such that assumption 3 is still valid. Violations of assumptions 1-5 are kept at a level such that the real device is
still a useful solar cell.
to the real value 𝑃max
13
1230.00.20.40.60.81.0fraction of powerband gap Eg (eV)Power outnot absorbedThermalizationisothermalRecombinationacurrent densityvoltage(1,2)(4)(5)Power outphoton fluxphoton energyEgEg/qnot absorbedbThermalizationPower outRecombinationisothermalJSQSCJSQSC/qJQESCVSQOCV radOCVOCJVqVSQOCc
S-Q Guide for the Perplexed 10-10-2018
Fig. 3: Visualization of potential improvement of top-performing solar cells, based on their band gaps, relative to
the ideal, SQ-case, using partitioning of the efficiency losses, according to Eq. (5) (Box). The following figures of
SQ). Only the two mature
merit (FoMs) are used and shown: 𝐹SC and 𝐹FF
technologies GaAs and Si achieve 𝜂real/ 𝜂0 > 80 %. The poly-crystalline thin-film technologies Cu(In,Ga)Se2
(CIGS), metal-halide-perovskite (ABX3), and Cd(Te,Se) are still below 70 %. Quantum-dot (QD) and organic
solar cells (OPV) are below 50 %. Note that for the calculation of FF0, nid = 1 was assumed, because nid for record
cells is generally not known.
real)/ 𝐹𝐹0(𝑉OC
and 𝐹𝐹0(𝑉OC
res, 𝑉OC
SQ
real/𝑉OC
14
0.30.40.50.60.70.80.91VrealOC/VSQOCFF0(VrealOC)/FF0(VSQOC)FresFFpercentage of SQ efficiencyGaAs c-Si CIGS ABX3 Cd(Te,Se) QD OPVFSC S-Q Guide for the Perplexed 10-10-2018
Reference List
1 Shockley, W. and Queisser, H. J., "Detailed balance limit of efficiency of pn-junction solar cells,"
J. Appl. Phys. 32(3), 510 (1961).
2 Prince, M. B., "Silicon Solar Energy Converters," J. Appl. Phys. 26(5), 534 (1955).
3 Loferski, J. J., "Theoretical Considerations Governing the Choice of the Optimum Semiconductor
for Photovoltaic Solar Energy Conversion," J. Appl. Phys. 27(7), 777 (1956).
4 Wolf, M., "Limitations and Possibilities for Improvement of Photovoltaic Solar Energy Converters:
Part I: Considerations for Earth's Surface Operation," Proceedings of the IRE 48(7), 1246 (1960).
5 Nayak, P. K., Mahek, S., Snaith, H. J., and Cahen, D., "Photovoltaic solar cell technologies:
Analysing the state of the art," Nat. Rev. Mater. doi: 10.1038/s41578-019-0097-0 (2019).
6 Krogstrup, P., et al., "Single-nanowire solar cells beyond the Shockley-Queisser limit," Nat.
Photon. 7(4), 306 (2013).
7 Stolterfoht, M., et al., "Approaching the fill factor Shockley-Queisser limit in stable, dopant-free
triple cation perovskite solar cells," Energ. Environ. Sci. 10(6), 1530 (2017).
8 P. Würfel, Physics of solar cells: From basic principles to advanced concepts, 2nd ed. (Wiley-
VCH, Weinheim, 2009).
9 Araujo, G. L. and Marti, A., "Absolute Limiting Efficiencies for Photovoltaic Energy-Conversion,"
Sol. Energy Mater. Sol. Cells 33(2), 213 (1994).
10 Hirst, L. C. and Ekins-Daukes, N. J., "Fundamental losses in solar cells," Prog. Photovolt. : Res.
Applic. 19(3), 286 (2011).
11 Würfel, U., Cuevas, A., and Würfel, P., "Charge Carrier Separation in Solar Cells," IEEE J. Photov.
5(1), 461 (2015).
12 Asbeck, P., "Self-Absorption Effects on Radiative Lifetime
in GaAs-GaAlAs Double
Heterostructures," J. Appl. Phys. 48(2), 820 (1977).
13 Bridgman, P. W., "Note on the Principle of Detailed Balancing," Phys. Rev. 31, 101 (1928).
14 Markvart, T., "Solar cell as a heat engine: energy-entropy analysis of photovoltaic conversion,"
Phys. Stat. Sol. A 205(12), 2752 (2008).
15 Green, M. A., "Third generation photovoltaics: Ultra-high conversion efficiency at low cost," Prog.
Photovolt. : Res. Applic. 9(2), 123 (2001).
16 Green, M. A., "Solar cell fill factors: General graph and empirical expressions," Solid State
Electron 24(8), 788 (1981).
17 Tiedje, T., Cebulka, J. M., Morel, D. L., and Abeles, B., "Evidence for Exponential Band Tails in
Amorphous-Silicon Hydride," Phys. Rev. Lett. 46(21), 1425 (1981).
18 Nayak, P. K., et al., "Photovoltaic efficiency limits and material disorder," Energ. Environ. Sci.
5(3), 6022 (2012).
19 Vandewal, K., et al., "Efficient charge generation by relaxed charge-transfer states at organic
interfaces," Nat. Mater. 13, 63 (2014).
20 Rau, U., Blank, B., Müller, T. C. M., and Kirchartz, T., "Efficiency Potential of Photovoltaic
Materials and Devices Unveiled by Detailed-Balance Analysis," Phys. Rev. Applied 7(4), 044016
(2017).
21 Rau, U., "Reciprocity relation between photovoltaic quantum efficiency and electroluminescent
emission of solar cells," Phys. Rev. B 76(8), 085303 (2007).
22 Xu, Y., Gong, T., and Munday, J. N., "The generalized Shockley-Queisser limit for nanostructured
solar cells," Sci. Rep. 5, 13536 (2015).
23 Schweiger, M., Herrmann, W., Gerber, A., and Rau, U., "Understanding the energy yield of
photovoltaic modules in different climates by linear performance loss analysis of the module
performance ratio," IET Renewable Power Generation 11(5), 558 (2017).
24 Green, M. A., "Radiative efficiency of state-of-the-art photovoltaic cells," Prog. Photovolt. : Res.
Applic. 20(4), 472 (2012).
15
S-Q Guide for the Perplexed 10-10-2018
25 Liu, Z., et al., "Open-Circuit Voltages Exceeding 1.26 V in Planar Methylammonium Lead Iodide
Perovskite Solar Cells," ACS Energy Lett. 4, 110 (2019).
26 Green, M. A., et al., "Solar cell efficiency tables (version 51)," Prog. Photovolt: Res. Appl. 26(1),
3 (2018).
27 Polman, A., et al., "Photovoltaic materials: Present efficiencies and future challenges," Science
352(6283) (2016).
28 Braly, I. L., et al., "Hybrid perovskite films approaching the radiative limit with over 90%
photoluminescence quantum efficiency," Nat. Photon. 12, 355 (2018).
29 Marti, A., Balenzategui, J. L., and Reyna, R. F., "Photon recycling and Shockley's diode equation,"
J. Appl. Phys. 82(8), 4067 (1997).
16
|
1710.00630 | 2 | 1710 | 2017-10-03T10:41:16 | Highly efficient rubrene-graphene charge transfer interfaces as phototransistors in the visible regime | [
"physics.app-ph",
"cond-mat.mes-hall",
"physics.ins-det"
] | Atomically thin materials such as graphene are uniquely responsive to charge transfer from adjacent materials, making them ideal charge transport layers in phototransistor devices. Effective implementation of organic semiconductors as a photoactive layer would open up a multitude of applications in biomimetic circuitry and ultra-broadband imaging but polycrystalline and amorphous thin films have shown inferior performance compared to inorganic semiconductors. Here, we utilize the long-range order in rubrene single crystals to engineer organic semiconductor-graphene phototransistors surpassing previously reported photo-gating efficiencies by one order of magnitude. Phototransistors based upon these interfaces are spectrally selective to visible wavelengths and, through photoconductive gain mechanisms, achieve responsivity as large as 10^7 A/W and a detectivity of 1.5 10^9 Jones at room temperature. These findings point towards implementing low-cost, flexible materials for amplified imaging at ultra-low light levels. | physics.app-ph | physics |
Highly efficient rubrene-graphene charge transfer interfaces as phototransistors in the
visible regime
Gareth F. Jones 1 , Rui M. Pinto 2 , Adolfo De Sanctis1 , V. Karthik Nagareddy 1 , C. David
Wright 1 , Helena Alves 3 , Monica F. Craciun 1 , Saverio Russo 1*
1 Centre for Graphene Science, College of Engineering, Mathematics and Physical Sciences,
University of Exeter, Exeter EX4 4QF, United Kingdom.
2 INESC MN and IN, Rua Alves Redol No. 9, 1000-029 Lisboa, Portugal.
3CICECO - Aveiro Institute of Materials, Physics Department, University of Aveiro, 3810
Aveiro, Portugal
* Correspondence: S. Russo, Email: [email protected]
Keywords: rubrene, graphene phototransistor, organic single crystal, high quantum efficiency,
photodetector
Atomically thin materials such as graphene are uniquely responsive to charge transfer from
adjacent materials, making them ideal charge transport layers in phototransistor devices.
Effective implementation of organic semiconductors as a photoactive layer would open up a
multitude of applications in biomimetic circuitry and ultra-broadband imaging but
polycrystalline and amorphous thin films have shown inferior performance compared to
inorganic semiconductors. Here, we utilize the long-range order in rubrene single crystals to
engineer organic semiconductor-graphene phototransistors surpassing previously reported
photo-gating efficiencies by one order of magnitude. Phototransistors based upon these
interfaces are spectrally selective to visible wavelengths and, through photoconductive gain
mechanisms, achieve responsivity as large as 𝟏𝟎𝟕𝑨𝑾−𝟏 and a detectivity of 𝟗 × 𝟏𝟎𝟏𝟏 𝑱𝒐𝒏𝒆𝒔
at room temperature. These findings point towards implementing low-cost, flexible materials
for amplified imaging at ultra-low light levels.
1
The planar interfaces formed between monolayer graphene and semiconductor materials present
unique opportunities for amplified detection of weak light signals. In these systems, electron-hole
pairs are excited in the semiconductor layer by an incident flux of photons (𝜙) with energy equal
to or greater than the optical band gap. Charge carriers of one polarity are then transferred from
the semiconductor into graphene according to the electrochemical potential gradient at the
interface, modulating the carrier concentration in graphene by 𝛥𝑛. Charge carriers of the opposite
polarity remain in the semiconductor layer, often in localized trap states, and recombine after an
average lifetime 𝜏𝐿. The associated probability encompassing the photo-excitation of charges and
their transfer to graphene is the photo-gating quantum efficiency (PGQE), 𝜂𝑃𝐺 = 𝛥𝑛/𝜙𝜏𝐿. If an
external voltage (𝑉𝐷𝑆) is applied along the length (𝐿) of a graphene-semiconductor interface
(Figure 1a), charge carriers transferred into graphene will drift between source and drain
electrodes over an average time 𝜏𝑡𝑟 = 𝐿2/𝜇𝑉𝐷𝑆 where 𝜇 is the charge carrier mobility of graphene.
For channels that are just a few microns long, trapping lifetimes are typically up to nine orders of
magnitude longer than transit time-scales, resulting in a net photoconductive gain 𝐺 = 𝜏𝐿/𝜏𝑡𝑟. The
external quantum efficiency (EQE, 𝜂𝐸𝑄𝐸 = 𝐺𝜂𝑃𝐺 ) of these phototransistors can therefore far
exceed 100%, particularly if the PGQE is optimal, allowing electrical detection of femtowatt light
signals at room temperature in micron-scale devices. Optimizing the EQE of graphene
phototransistors through exploration of various material combinations is now a highly active field
of research, with previous studies focusing on hybrid structures of graphene interfaced with
colloidal quantum dots, [1-5] transition metal dichalcogenides, [6-8] III-VI semiconductors, [9] metal
oxides, [10] perovskites, [11] chlorophyll, [12], organometallic complexes [13] and organic
semiconductors thin films [14-17]. So far, room temperature EQE as large as 108 electrons per
photon has been reported in graphene/colloidal quantum dot phototransistors, with operational
speeds suitable for video-rate imaging. [2].
2
Phototransistors that combine graphene with organic semiconductors are particularly desirable
owing to the gamut of complementary properties found in these systems. The spectral selectivity
of π-conjugated semiconductors can be tailored through chemical or structural modification to
emulate the trichromatic response of cone cells in mammalian retina [18] or exhibit ultra-broadband
UV-to-NIR sensitivity. [19] Additionally, organic semiconductors have an intrinsic affinity to
biological systems that is vital for developing innovative healthcare sensors. However, organic
semiconductor-graphene phototransistors [14-17] have shown radically inferior PGQE compared to
inorganic semiconductor-graphene phototransistors, where 𝜂𝑃𝐺~25% using colloidal PbS
quantum dots [2]. Short exciton diffusion lengths (~10nm) [20] and inhomogeneity [21] present in
organic semiconductor films are likely to play a critical role in limiting quantum efficiencies.
Amorphous films of P3HT [16] have been shown to exhibit a 𝜂𝑃𝐺~0.002%, whereas 𝜂𝑃𝐺~0.6%
has been achieved using polycrystalline films of epitaxially grown C8-BTBT [17] (see Table S1),
suggesting that the disorder of the crystal structure in graphene/organic semiconductor
phototransistors plays a pivotal adverse role. Furthermore, the current operational speed of these
hybrid devices is also far from ideal. Response times lasting many seconds make them too slow
for imaging applications, yet not sufficiently stable to function as optical memories. Here, we
address both of these challenges for the first time using a single crystal organic semiconductor,
rubrene, as the light-absorbing layer in a graphene phototransistor. Long-range herringbone
stacking of rubrene molecules in a single crystal facilitates exciton diffusion over several microns.
[22, 23] We exploit this to achieve both efficient light absorption and efficient extraction of photo-
excited charge carriers, resulting in an external and internal PGQE as high as 1% and 5%
respectively. These organic single crystal-graphene phototransistors exhibit responsivity as large
as 107𝐴𝑊−1 at room temperature and a specific detectivity of 9 × 1011 𝐽𝑜𝑛𝑒𝑠.
The device structure consists of a rubrene single crystal grown by physical vapor transport (see
Methods) laminated onto a pre-fabricated graphene transistor (Figure 1b). After device fabrication,
3
we used the well-established used method of cross-polarized optical microscopy [24] to demonstrate
macroscopic molecular ordering and absence of polycrystalline domains across the rubrene single
crystals. Figures 1c and 1d show uniform brightness across the entirety of the interface, with the
magnitude of brightness dependent on the angle between the crystal's long axis and the
polarization plane of incident light. Polar plots of the average brightness over three distinct
interface regions (Figure 1e) revealed identical birefringence, which can only be present in a
structurally pristine single crystal of rubrene. Figure 1f shows photoluminescence (PL) spectra
from another rubrene crystal measured at two locations, one with and one without an underlying
sheet of graphene. In both cases, the PL spectra fit well with two sets of equidistant Voigt functions
representing the vibronic progression of radiative transitions polarized along the L/N (< 2.05𝑒𝑉)
and M (> 2.05𝑒𝑉) axes of rubrene molecules (Figure 1g). [25] Although M-polarized emission is
10-20 times stronger, these peaks are suppressed in Figure 1f. This indicates that our axis of
illumination/detection is oriented parallel to the M axes of rubrene molecules and, therefore, is
normal to the ab crystal plane (Figure 1h). We confirm this crystallographic orientation via
polarized Raman spectroscopy, shown in Figure S5 [27]. A PL band located at 1.91𝑒𝑉, which is a
signature of photo-oxidation [25] and deep trap states [28], was absent from all measured samples
confirming the high purity of these crystals. Comparing the two PL spectra, the presence of
graphene underneath rubrene reduces the PL intensity by approximately 25%. This PL quenching
suggests that a substantial fraction of excitons dissociate across the rubrene-graphene interface,
although Förster resonance energy transfer could also cause PL quenching [29] and would be
detrimental to the PGQE. In order to determine the efficiency with which electron-hole pairs
dissociate at the rubrene-graphene interface, we proceeded to study the electrical response of a
channel segment to flood illumination (𝜆 = 500𝑛𝑚).
Figure 1i shows that photocurrent (𝐼𝑃𝐻) measured from a rubrene-graphene channel segment has
a linear dependence on source-drain voltage. This linearity is expected for systems exhibiting
4
photoconductive gain (𝐺 =
𝜇𝜏𝐿𝑉𝐷𝑆
𝐿2 ∝ 𝑉𝐷𝑆) and suggests that excitons in rubrene are dissociated at
the interface with graphene. For a source-drain bias voltage of 30𝑚𝑉 and a measured field-effect
mobility of 1300𝑐𝑚2𝑉−1𝑠−1 (see Figure S8), we estimate the transit time to be 𝜏𝑡𝑟 = 10𝑛𝑠. In
Figure 2a, we show the resistance of the same rubrene-graphene channel as a function of applied
gate voltage under a variety of optical power densities (𝑃). In all cases, charge transport along the
interface is clearly dominated by the ambipolar behavior of monolayer graphene with illumination
inducing an up-shift of the charge neutrality point (∆𝑉𝐶𝑁𝑃). This up-shift is indicative of photo-
excited holes being transferred from rubrene into graphene whilst electrons remain confined to the
rubrene crystal. [2] A plot of the upward shift of the charge neutrality point as a function of power
density in Figure 2b reveals that the photo-gating effect saturates for 𝑃 > 5𝑊𝑚−2 . Such an
observation could originate from screening of the built-in field at the rubrene-graphene interface
[2] or an increased probability of bimolecular recombination in rubrene [23] at high photo-excited
charge carrier densities. To characterize the gain in these phototransistors, we focus on the
transient response of this interface to weak light signals, where 𝑃 < 300𝜇𝑊𝑚−2, in the inset of
Figure 2b. Whilst the rise time of the detector is relatively fast, taking approximately 100𝑚𝑠 to
reach steady-state conditions under illumination, the transition back to dark current levels lasts for
tens of seconds and is indicative of the average lifetime of electrons localized in rubrene. [2] The
photocurrent after illumination was fit with a bi-exponential decay function (pink dashed line),
suggesting that at least two distinct lifetimes exist for electrons in rubrene [12]. Taking a weighted
average of the two decay constants, we calculate an average lifetime of 𝜏𝐿 = 24 ± 3 seconds and
gain of 𝐺 ≈ 109. The photoresponse of these devices is tunable with applied gate voltage and we
observe a responsivity (𝛾 = 𝐼𝑃𝐻𝐿𝑊/𝑃) as large as 1.4 × 105𝐴𝑊−1 and EQE= 3.4 × 107%
under light levels equivalent to moonlit conditions.
5
In Figure 2e, we compare the spectral response of a rubrene-graphene interface with an equivalent
sample solely comprised of rubrene. Identical voltages were applied to each device (𝑉𝐷𝑆 = 30𝑚𝑉,
𝑉𝐺 = 0𝑉) and channel geometries were kept consistent (𝐿 = 5𝜇𝑚, 𝑊 = 90 − 100𝜇𝑚). Dashed
lines show the simulated absorbance of the isolated crystal to be approximately twice that of the
crystal in contact with graphene due to differences in thickness (405𝑛𝑚 and 202𝑛𝑚 respectively).
The two transistors produced responsivity spectra with very similar shapes, confirming that
photocurrent signals in rubrene-graphene transistors arise purely from light absorption in the
organic single crystal and that no new chemical species form at the interface. A comparison of the
magnitude of responsivity in each device in the inset of Figure 2e shows that graphene/rubrene
phototransistors exhibit values of 𝛾 up to six orders of magnitude larger than in isolated rubrene.
This observation demonstrates that the high charge carrier mobility in graphene plays an essential
role for efficient transport of photo-excited holes between source and drain electrodes whereas the
surface photoconductivity of rubrene [23] does not significantly contribute towards read-out signals.
Channel geometry is a significant, but extraneous, factor that is largely responsible for the large
variations of responsivity amongst previously reported graphene-based phototransistors [2-17].
Indeed, a proportionality of 𝛾 ∝ 𝐺𝜂𝑃𝐺 ∝ 𝐿−2 is expected, given that 𝐺 ∝ 𝐿−2 whilst 𝜂𝑃𝐺 is
independent of channel length. However, large lateral electric fields and possible exciton
quenching effects at metallic source/drain electrodes could significantly affect the PGQE in shorter
channels. To exclude these spurious effects, we have conducted for the first time a scaling
experiment of 𝛾 as a function of channel length in Figure 3a. We find that the responsivity shows
the expected 𝐿−2 dependence when normalized to the charge carrier mobility and potential
difference (𝑉𝑐ℎ) across each segment in order to account for contact resistance and doping
inhomogeneity (Figure S8). This demonstrates that the active area of the phototransistors
comprises the whole rubrene-graphene interface between the source and drain electrodes. Hence,
6
a more meaningful comparison of 𝛾 for any graphene-based phototransistor can be achieved by
accounting for the inverse square dependence on channel length, provided the PGQE is
independent of L.
As previously mentioned, the PGQE of graphene-based phototransistors generally increases at
lower absorbed photon densities. In Figure 3b, we explore the limit of this effect by measuring the
non-linear power dependence of responsivity in a rubrene-graphene interface exposed to ultra-
weak light signals. A maximum of 𝛾𝑚𝑎𝑥 ≈ 1 × 107𝐴𝑊−1 is reached for the lowest measured
optical power densities. This marks the first report of responsivity comparable to the record room-
temperature performance of inorganic semiconductor-graphene phototransistors [2, 6] from an
entirely organic equivalent. Analogous to previous phototransistor studies, [2, 4, 6] we fit this non-
linear power dependence with the function
𝛾 =
𝛾𝑚𝑎𝑥
1+(𝑃 𝑃0⁄ )𝑛
(1)
where 𝑃0 marks a threshold power below which the responsivity saturates, and 𝑛 is an exponent
which dictates the decline in responsivity above this threshold. A best fit of Equation (1) (blue
dashed line) yields 𝑃0 ≈ 1.1𝜇𝑊𝑚−2 and 𝑛 = 0.70 ± 0.04. Re-arranging the expanded expression
for responsivity, 𝛾 = (𝑒𝜂𝑃𝐺 ⁄ ℎ𝑣)(𝜇𝑉𝐷𝑆𝜏𝐿 ⁄ 𝐿2 ), we are able to calculate the PGQE (blue data)
shown on the right y-axis in Figure 3b. The simulated absorbance (𝐴) of the rubrene crystal,
shown in Figure 2e, is then used to calculate the internal photo-gating quantum efficiency (charges
transferred to graphene per absorbed photon, orange data) as 𝜂𝑖𝑃𝐺 = 𝜂𝑃𝐺 ⁄ 𝐴. For power densities
equivalent to sub-femtowatt incident signals we calculate 𝜂𝑃𝐺 ≈ 1% and 𝜂𝑖𝑃𝐺 ≈ 5% .
Comparatively, this value of PGQE is four orders of magnitude greater than a previous study that
combined amorphous films of P3HT [16] with graphene and one order of magnitude higher than
epitaxially grown polycrystalline films of C8-BTBT grown on graphene [17]. We attribute the
superior PGQE in rubrene-graphene interfaces to a combination of two factors. Firstly, a rubrene
7
single crystal serves as an ideal light-absorbing layer due to the extremely low density of charge
traps in the bulk of the crystal [30] and the large intermolecular overlap of pi orbitals which
facilitates Dexter-type diffusion of triplet excitons over several microns. [22, 23] A far larger number
of excitons are therefore able to diffuse to the graphene interface and dissociate. Secondly, this is
the first study to examine the PGQE of organic semiconductor-graphene phototransistors at
extremely low absorbed photon densities where bimolecular recombination and triplet-triplet
fusion are not significant loss mechanisms. [23]
Previous studies of hybrid graphene phototransistors have attributed a decline in responsivity with
increasing optical power to the saturation of available charge trap in the light-absorbing
semiconductor layer. [4] Assuming this to be true in the case of rubrene-graphene interfaces, we
use the threshold power density to estimate the density of trap states in rubrene available for photo-
gating processes as 𝑁𝑡 = 𝜂𝑖𝑃𝐺𝜏𝐿 𝑃0 ℎ𝑣⁄ ≈ 5 × 108𝑐𝑚−2 . Even if graphene screens an
overwhelming proportion of the traps otherwise present at the rubrene-SiO2 interface
(~1012𝑐𝑚−2), [30] our estimate of 𝑁𝑡 is too low to be physically plausible. Hence, the density of
available interface trap states does not govern the non-linear responsivity. Instead, we note that
responsivity follows a power exponent of approximately −2/3 for 𝑃 ≫ 𝑃0 , which closely
correlates with the signature of triplet-charge recombination from surface photoconductivity
experiments on rubrene. [23] The onset of these interactions occurs at higher absorbed photon
densities (> 1015𝑐𝑚−3𝑠−1) in isolated crystals but it is reasonable to expect a lower threshold
considering the additional population of charge carriers from physical contact with graphene. This
finding should help to inform future strategies of interface modification.
In Figure 3c, we operate a 5𝜇𝑚 rubrene-graphene channel at a gate voltage of 𝑉𝐺 = 10𝑉 such that
the Fermi level of graphene lies within the conduction band. With successive cycles of illumination,
the drain current gradually drifts away from its original dark value due to the fall time of the
detector exceeding the time under dark conditions. By applying a gate voltage pulse when the light
8
source is extinguished, we momentarily reduce the built-in field across the interface which
otherwise limits recombination of photo-excited electrons in rubrene [2] (Figure 3d). Using this
technique, we surpass the bandwidth limitations that previous graphene-organic semiconductor
phototransistors have suffered from resulting in a maximum detectivity of 9 × 1011𝐽𝑜𝑛𝑒𝑠 (see
Supplementary Section S6).
A comparison of the responsivity measured in published state-of-the-art organic semiconductor-
graphene phototransistors (Figure 4a) demonstrates that our devices attain record high values of
responsivity at unprecedentedly low incident photon flux, comparable to that of inorganic
equivalents (see Figure S11). However, since responsivity depends on extraneous parameters such
as channel length, source-drain voltage and the charge carrier mobility of graphene, it is difficult
to gain accurate insight as to the relative performance of each organic material solely from this
figure of merit. Indeed, all of these extraneous parameters vary significantly amongst studies and
influence photoconductive gain rather than the intrinsic photo-gating quantum efficiency of each
material interface. Hence, we define 𝜂𝐸𝑄𝐸𝐿2 𝜇𝑉𝐷𝑆
⁄
= 𝜂𝑃𝐺𝜏 as a more appropriate figure of merit
which is independent of 𝜇, 𝑉𝐷𝑆 and 𝐿 (Figure 4b). This quantity reflects the maximum achievable
EQE. Specific detectivity could also serve as an informative figure of merit but is often not
reported and sometimes overestimated by assuming Shot-noise limited performance. The
comparative plots of Figure 4 conclusively demonstrate that rubrene single crystal-graphene
phototransistors are uniquely suited for amplified detection of extremely weak light signals in all-
organic electronics, where the long-range diffusion of excitons in rubrene facilitates both high
absorbance and efficient extraction of photo-generated charge carriers. Two caveats when
considering possible applications of graphene-based phototransistors are the operational
bandwidth and noise-equivalent power of each phototransistor. Whilst bandwidth can be improved
through gate voltage modulation or screening of deep trap states, for example with ionic polymer
gates, [8] it is the 1/f dark current noise that limits the noise-equivalent power of graphene detectors
for bandwidths below 100𝑘𝐻𝑧. [31] This limitation can be addressed to some degree by using single
9
crystal graphene films [32] and one-dimensional electrode contacts. [33]
Overall, the parallel efforts
to develop optimal light-absorbing layers, improve operational bandwidth and reduce 1/f noise
could enable graphene-based phototransistors to reach detectivity values rivalling those of single
photon detectors.
In conclusion, interfaces of monolayer graphene and rubrene single crystals are promising systems
for ultra-sensitive detection of visible light. Long-range order in rubrene crystals facilitates
effective transfer of photo-generated charges to graphene with an external and internal efficiency
of 1% and 5% respectively. Utilizing these interfaces as phototransistors, responsivity as high as
107𝐴𝑊−1 can be achieved for sub-femtowatt incident signals, comparable to the record
performance of graphene-quantum dot detectors. Finally, we emphasize the importance of
distinguishing between the contributions of internal gain, photo-gating quantum efficiency and
carrier lifetime towards the responsivity of phototransistors. Following this procedure, accurate
conclusions can be made as to which combination of materials warrant further research and how
to continue improving the performance of this novel class of high-gain, micro-scale photodetectors.
Experimental Section
Device Fabrication: Monolayer graphene and rubrene single crystals were grown by chemical
vapor deposition [34] and physical vapor transport [35] respectively. Full details of material growth
and device fabrication are provided in Supplementary Section S1 and S2.
Photocurrent and PL Measurements: Phototransistor devices were housed in a vacuum probe
station (10−3𝑚𝑏𝑎𝑟) with a fused silica viewport for photocurrent measurements. A xenon lamp
and monochromator with variable low-pass filters (Newport TLS300X) provided spectrally
tunable, collimated light incident over the entire sample. We adjusted optical power levels using a
series of neutral density filters. A mechanical shutter (Thorlabs SHB1T) modulated light signals
and power densities were calibrated using a photodiode (Thorlabs S130CV) before and after each
10
dataset run. Excluding spectral scans, 𝜆 = 500𝑛𝑚 for all measurements. PL spectra were excited
in atmospheric conditions using a 532nm laser through a microscope objective (numerical aperture
= 0.5). Rubrene-graphene channel dimensions are 𝐿 = 5µ𝑚, 𝑊 = 91𝜇𝑚, except in Figure 3a.
Absorbance Calculations: From Figure 2e, 𝐴 = 0.207 for the rubrene-graphene transistor if 𝜆 =
500𝑛𝑚. The methodology of our absorbance calculations is shown in Figure S4 and S6.
11
Supporting Information
Supporting Information is available from
http://onlinelibrary.wiley.com/doi/10.1002/adma.201702993/full
Acknowledgements
S. Russo and M.F. Craciun acknowledge financial support from EPSRC (Grant 464 no.
EP/J000396/1, EP/K017160/1, EP/K010050/1, EP/G036101/1, EP/M001024/1,
465
EP/M002438/1), from Royal Society international Exchanges Scheme 2012/R3 and 466 2013/R2
and from European Commission (FP7-ICT-2013-613024-GRASP). The authors would like to
thank Paul Wilkins for technical assistance in designing and building the vacuum chamber probe
station used for all photocurrent measurements and Dr Dominique J. Wehenkel for assisting with
initial photocurrent measurements and for suggesting to measure the length-dependence of
responsivity.
References
[1]
A. V. Klekachev, M Cantoro, M. H. van der Veen, A. L. Stesmans, M. M. Heyns, S. De
Gendt, Physica. E. 2011, 43, 1046.
[2]
Konstantatos G,
Badioli M, Gaudreau L, Osmond J, Bernechea M, F. P. G. D.
Arquer, F. Gatti, F. H. L. Koppens, Nat. Nanotechnol. 2012, 7, 363.
[3]
Z. Sun, Z. Liu, J. Li, G. A. Tai, S. P. Lau, F. Yan, Adv. Mater. 2012, 24, 5878.
[4] W. Guo, S. Xu, Z. Wu, N. Wang, M. M. T. Loy, S.Du, Small. 2013, 9, 3031.
[5]
S. H. Cheng, T. M. Weng, M. L. Lu, W. C. Tan, J. Y. Chen, Y. F. Chen, Sci. Rep. 2013,
3, 2694.
[6]
K. Roy, M. Padmanabhan, S. Goswami, T. P. Sai, G. Ramalingam, S. Raghavan, A.
Ghosh, Nat. Nanotechnol. 2013, 8, 826.
[7]
X. Li, J. Wu, N. Mao, J. Zhang, Z. Lei, Z. Liu, H. Xu, Carbon. 2015, 92, 126.
[8]
J. D. Mehew, S. Unal, E. Torres Alonso, G. F. Jones, S. F. Ramadhan, M. F. Craciun, S.
Russo, Adv. Mater. 2017, 1700222.
[9]
R. Lu, J. Liu, H. Luo, V. Chikan, J. Z. Wu, Sci. Rep. 2016, 6, 19161.
[10] K. Zheng, F. Meng, L. Jiang, Q. Yan, H. H. Hng, X. Chen, Small. 2013, 9, 2076.
12
[11] Y. Lee, J. Kwon, E. Hwang, C. H. Ra, W. J. Yoo, J. H. Ahn, J. H. Park, J. H. Cho, Adv.
Mater. 2015, 27, 41.
[12] S. Y. Chen, Y. Y. Lu, F. Y. Shih, P. H. Ho, Y. F. Chen, C. W. Chen, Y. T. Chen, W. H.
Wang, Carbon. 2013, 63, 23.
[13] X. Liu, E. K. Lee, J. H. Oh, Small. 2014, 10, 3700.
[14] E. K. Jeon, C. S. Yang, Y. Shen, T. Nakanishi, D. S. Jeong, J. J. Kim, K. S. Ahn, K. S.
Kong, J. O. Lee, Nanotechnology. 2012, 23, 455202.
[15] S. Jang, E. Hwang, Y. Lee, S. Lee, J. H. Cho, Nano Lett. 2015, 15, 2542.
[16] E. H. Huisman, A. G. Shulga, P. J. Zomer, N. Tombros, D. Bartesaghi, S. Z. Bisri, M. A.
Loi, L. J. A. Koster, B. J. van Wees, ACS Appl. Mater. Interfaces. 2015, 7, 11083.
[17] X. Liu, X. Luo, H. Nan, H. Guo, P. Wang, L. Zhang, M. Zhou, Z. Yang, Y. Shi, W. Hu,
Z. Ni, T. Qiu, Z. Yu, J. B. Xu, X. Wang, Advanced Materials. 2016, 28, 5200.
[18] N. Martino, D. Ghezzi, F. Benfenati, G. Lanzani, M. R. Antognazza, J. Mater. Chem. B.
2013, 1, 3768.
[19] N. K. S. Davis, A. L. Thompson, H. L. Anderson, J. Am. Chem. Soc. 2011, 133, 30.
[20] M. Sim, J. Shin, C. Shim, M. Kim, S. B. Jo, J. H. Kim, K. Cho, J. Phys. Chem. C. 2014,
7, 11083.
[21] W. L. Kalb, S. Haas, C. Krellner, T. Mathis, B. Batlogg, Phys. Rev. B. 2010, 81, 155315.
[22]
a) H. Najafov, B. Lee, Q. Zhou, L. C. Feldman, V. Podzorov, Nat. Mater. 2010, 9, 938. b)
P. Irkhin, I. Biaggio, Phys. Rev. Lett. 2011, 107, 017402.
[23] P. Irkhin, H. Najafov, V. Podzorov. Sci. Rep. 2015, 5, 15323.
[24] J. Vrijmoeth, R. W. Stok, R. Veldman, W. A. Schoonveld, T. M. Klapwijk, J. Appl. Phys.
1998, 83, 3816.
[25] P. Irkhin, A. Ryasnyanskiy, M. Koehler, I. Biaggio, Phys. Rev. B. 2012, 96, 085143.
[26]
a) O. D. Jurchescu, A. Meetsma, T. T. M. Palstra. Acta Cryst. 2006, B62:330. b) C. R.
Groom, I. J. Bruno, M. P. Lightfoot, S. C. Ward, Acta Cryst. 2016; B72, 171.
13
[27] E. Venuti, I. Bilotti, R. Guido, D. Valle, A. Brillante, P. Ranzieri, M. Masino, A.
Girlando, J. Phys. Chem. C. 2008, 112, 17416.
[28] X. Song, L. Wang, Q. Fan, Y. Wu, H. Wang, C. Liu, N. Liu, J. Zhu, D. Qi, X. Gao, A. T.
S. Wee. Appl. Phys. Lett. 2010, 97, 032106.
[29]
a) L. Gaudreau, K. J. Tielrooij, G. E. D. K. Prawiroatmodjo, F. J. G. de Abajo, F. H. L.
Koppens. Nano Lett. 2013, 13, 2030. b) R. S. Swathi, K. L. Sebastian, J. Chem. Phys. 2009, 130,
086101.
[30] C. Goldmann, C. Krellner, K. P. Pernstich, S. Haas, D. J. Gundlach, B. Batlogg, J. Appl.
Phys. 2006, 99, 034507.
[31] A. A. Balandin, Nat. Nanotechnol. 2013, 8, 549.
[32] V. Kochat, C. S. Tiwary, T. Biswas, G. Ramalingam, K. Hsieh, K. Chattopadhyay, S.
Raghavan, M. Jain, A. Ghosh, Nano Lett. 2016, 16, 562.
[33] M. A. Stolyarov, G. Liu, S. L. Rumyantsev, M. Shur, A. A. Balandin, Appl. Phys. Lett.
2015, 107, 023106.
[34] T. H. Bointon, M. D. Barnes, S. Russo, M. F. Craciun, Adv. Mater. 2015, 27, 4200.
[35] R. A. Laudise, C. Kloc, P. G. Simpkins, T. Siegrist, J. Cryst. Growth. 1998, 201, 449.
14
Figure 1. Characterization of rubrene-graphene interfaces. a) Schematic of a rubrene-graphene
phototransistor on SiO2/Si with a measurement circuit diagram and inset denoting crystallographic
axes. b) Un-polarized and c), d) cross-polarized optical micrographs of a rubrene-graphene
interface (Scale bars: 200µm). Colored squares denote the regions analyzed in e) polar plots of the
greyscale brightness. f) Photoluminescence spectra of a rubrene single crystal at regions with
(blue) and without (red) underlying graphene. Dashed peaks are Voigt functions fit to the blue
spectra. Inset: Micrograph image showing the location of each PL scan (circles) and boundary
between rubrene and rubrene-graphene (dashes). Scale bar: 25μm. g) Molecular and h) crystalline
structure of rubrene at room temperature (CSD-QQQCIG08). [26] Axis notation conforms to charge
transport and photoluminescence studies. [22, 25] i) Photocurrent as a function of source-drain
voltage for various illumination intensities (𝑉𝐺 = 0𝑉). Inset: Resistance vs gate voltage sweep of
the same channel in dark conditions.
15
Figure 2. Photo-gating in rubrene-graphene interfaces. a) Channel resistance vs gate voltage in
dark conditions and under various optical power densities. Inset: Schematic of charge transfer at
the rubrene-graphene interface. b) Shift in the charge neutrality point of graphene as a function of
optical power density. Inset: 20-run average of the transient response of a 5μm rubrene-graphene
channel to 60 seconds of illumination. Dashed lines denote 10% and 90% thresholds of the steady-
state shift in current (orange) and a bi-exponential decay fit of the return to dark conditions (pink).
c) Photocurrent and d) responsivity as a function of gate voltage. e) Responsivity spectra of rubrene
(red) and rubrene-graphene (blue) transistors. Spectra are normalized to the maximum of each
dataset (smoothed via adjacent averaging, solid lines). Dashed lines are the simulated net
absorbance of each rubrene crystal (see Supplementary Section S3). Inset: Responsivity spectra
for rubrene and rubrene-graphene transistors.
16
Figure 3. Photo-response in rubrene-graphene phototransistors. a) Length scaling of the photo-
response in a single rubrene-graphene interface normalized with respect to the charge carrier
mobility and potential difference of/across each channel segment (Figure S8 and S9). b)
Responsivity vs optical power density and absorbed photon density (blue points, left y-axis) from
the averaged response of a 5μm channel to 20 illumination cycles. External (blue points, right y-
axis) and internal (orange points, right y-axis) photo-gating quantum efficiencies are calculated
from the same dataset. c) Transient photo-response of a rubrene-graphene phototransistor relative
to dark current levels (dashes) under light modulated at 0.5Hz with (red) and without (blue)
application of gate voltage pulses. Current spikes due to gate pulsing are readily removed with
filtering circuitry. d) Schematic band diagrams illustrate the charge transfer dynamics across at
each stage of the light modulation cycle.
17
Figure 4. Performance metrics of organic semiconductor-graphene phototransistors. Plots
catalogue the power dependence of rubrene single crystal-graphene phototransistors with all
other relevant studies using a) responsivity and b) the external quantum efficiency normalized to
extraneous parameters as figures of merit. Solid lines connect data taken from a single device,
marker colors denote the excitation wavelength used and data from this work is highlighted
(pink). Table S2 provides detailed citations, Figure S11 shows plots that include inorganic photo-
active layers.
Keywords: graphene; high quantum efficiency; phototransistor; rubrene single crystal; visible
wavelength.
G. F. Jones, R. M. Pinto , A. De Sanctis, V. K. Nagareddy, C. D. Wright, H. Alves, M. F.
Craciun , S. Russo*
Title Highly efficient rubrene-graphene charge transfer interfaces as phototransistors in the
visible regime
18
|
1805.05159 | 2 | 1805 | 2018-05-15T11:14:41 | Mapping the absolute magnetic field and evaluating the quadratic Zeeman effect induced systematic error in an atom interferometer gravimeter | [
"physics.app-ph",
"physics.atom-ph",
"quant-ph"
] | Precisely evaluating the systematic error induced by the quadratic Zeeman effect is important for developing atom interferometer gravimeters aiming at an accuracy in the regime ( ). This paper reports on the experimental investigation of Raman spectroscopy-based magnetic field measurements and the evaluation of the systematic error in the Gravimetric Atom Interferometer (GAIN) due to quadratic Zeeman effect. We discuss Raman duration and frequency step size dependent magnetic field measurement uncertainty, present vector light shift (VLS) and tensor light shift (TLS) induced magnetic field measurement offset, and map the absolute magnetic field inside the interferometer chamber of GAIN with an uncertainty of 0.72 nT and a spatial resolution of 12.8 mm. We evaluate the quadratic Zeeman effect induced gravity measurement error in GAIN as . The methods shown in this paper are important for precisely mapping the absolute magnetic field in vacuum and reducing the quadratic Zeeman effect induced systematic error in Raman transition-based precision measurements, such as atomic interferometer gravimeters. | physics.app-ph | physics | Mapping the absolute magnetic field and evaluating the quadratic Zeeman effect induced
systematic error in an atom interferometer gravimeter
Qing-Qing Hu1,2†, Christian Freier1, Bastian Leykauf1, Vladimir Schkolnik1, Jun Yang2, Markus Krutzik1, Achim
2Interdisciplinary Center for Quantum Information, National University of Defense Technology, Changsha,
1Institut für Physik, Humboldt-Universität zu Berlin, 12489 Berlin, Germany
Peters1
Hunan, 410073, China
Abstract
Precisely evaluating the systematic error induced by the quadratic Zeeman effect is important for developing
atom interferometer gravimeters aiming at an accuracy in the
regime (
). This paper
reports on the experimental investigation of Raman spectroscopy-based magnetic field measurements and the
evaluation of the systematic error in the Gravimetric Atom Interferometer (GAIN) due to quadratic Zeeman effect.
We discuss Raman duration and frequency step size dependent magnetic field measurement uncertainty, present vector
light shift (VLS) and tensor light shift (TLS) induced magnetic field measurement offset, and map the absolute
magnetic field inside the interferometer chamber of GAIN with an uncertainty of 0.72 nT and a spatial resolution of
12.8 mm. We evaluate the quadratic Zeeman effect induced gravity measurement error in GAIN as
. The
methods shown in this paper are important for precisely mapping the absolute magnetic field in vacuum and reducing
the quadratic Zeeman effect induced systematic error in Raman transition-based precision measurements, such as
atomic interferometer gravimeters.
I.
Introduction
Thanks to the development of laser cooling and trapping techniques [1], precision measurements based on cold
atom interferometers (AIs) have been demonstrated remarkable prospects, which stretch from atomic gravimeters
[2,3], gravity gradiometers [4,5], gyroscopes [6,7] and atomic clocks [8,9], to the measurement of physical constants,
such as fine structure constant [10,11], gravitational constant [5,12], and applications in fundamental physics such as
quantum tests of the weak equivalence principle [13].To avoid systematic error induced by the first-order Zeeman
effect, the atoms are usually prepared in the
sublevel before entering the experiment zone. The quadratic
Zeeman effect, however, still leads to a non-negligible systematic error [14]. For example, the error caused by the
quadratic Zeeman shift is the largest error among all the systematic effect in rubidium fountain clocks [15] and a
challenge for developing Raman transition-based atomic gravimeters aiming at an accuracy in the
regime
[16,17]. Though this error can be alleviated by the Raman wave vector reversing method [4], it cannot be canceled
completely due to the spatial non-overlap of the two interference paths [16], especially in the cases of large momentum
transfers [18-20] and long pulse intervals [21]. Therefore, mapping the absolute magnetic field intensity in the
interference region, and evaluating the corresponding error is a more accurate method for laboratory research and field
applications [14,22].
Magnetically-sensitive atom interferometers [23] and double fountain-based simultaneous differential atom
interferometers [17] have been proposed to map the gradient of the magnetic field inside a vacuum chamber. However,
because the resonance frequencies of the magnetically-sensitive transitions are sensitive to the magnetic field with a
scale of 14 Hz/nT [24], a common magnetic field inhomogeneity of 200 nT [23] will cause a frequency detuning of
2.8 kHz, which could change the transition probability of each Raman pulse and result in a non-negligible error.
Besides, the absolute magnetic field map is needed in order to precisely evaluate the Zeeman shift induced systematic
† Corresponding author. E-mail: [email protected]
μGal8291μGal=10ms10g2.04μGal0FmμGal error. Therefore, one more process of distinguishing the sign of the magnetic field gradient and calculating its spatial
integral is needed. Other methods, such as the weak magnetically-sensitive Zeeman splitting [25] and the Bragg
interferometers using the three magnetic states simultaneously [26], require experimental conditions such as 30 ms
polarized Raman pulse [25] or Bose-Einstein condensate (BEC) atomic sources [26], which are impractical
for GAIN and other systems of the same kind [4,27,28].
Raman spectroscopy-based magnetic field mapping was previously reported with a measurement uncertainty of
20 nT [23] and 0.28 nT [29], respectively. In the latter case, in order to achieve this measurement uncertainty and
ensure a good spatial resolution simultaneously, the frequency step from shot to shot is as small as 10 Hz, and a 12 ms
Raman π pulse is applied when the atoms reaching their apogee. Therefore, the launch velocity and detection time of
atomic cloud, as well as the moment of irradiation by Raman pulses need to be adjusted manually for each launch
height at which one intends to measure the magnetic field. Besides, taking the parameters of GAIN for example
(quantization magnetic field ~5
, interferometer chamber length ~68 cm), the 10 Hz frequency step means a time
consumption of ~17 days for mapping the magnetic field inside the interferometer chamber with a spatial resolution
of 1 cm, and more time is needed for a better spatial resolution. Apparently, this method takes too long a time and
might suffer from the problem of magnetic field drift within one measurement. However, this time consumption can
be shortened by one order of magnitude if a larger frequency step of 100 Hz is used. Besides, the shorter pulse enables
a better spatial resolution and more sampling points in one transition peak (Fig. 3 (a)-(c)) and this might improve the
magnetic field measurement uncertainty. Consequently, there has been some interest in investigating the influences of
Raman pulse duration and frequency step size on the Raman spectroscopy-based magnetic field measurement. On the
other hand, the vector and tensor ac stark light shifts, which influence the measurement accuracy of Raman
spectroscopy-based magnetic field measurement, haven't been discussed before.
In this paper, we report on the experimental investigation of Raman spectroscopy-based magnetic field mapping
and the evaluation of quadratic Zeeman effect induced systematic error in GAIN. This paper is structured as follows.
Section II briefly introduces the measurement principle, experimental setup and procedure. Section III presents (1):
the relationship between measurement uncertainty and Raman pulse duration for different frequency step size; (2) the
influence of the vector light shift (VLS) and tensor light shift (TLS); (3) the absolute magnetic field map inside the
interferometer chamber of GAIN and its time stability; (4) the quadratic Zeeman shift induced systematic error and
the uncertainty of this error. Section IV discusses the principle of nulling the ac stark effect in atom interferometer.
Section V summarizes our main results and provides an outlook. The appendix shows the calculation of the
polarizabilities, the scalar, vector and tensor light shifts in Raman transitions.
II.
Experimental principle and apparatus
A. Experimental principle
ground state magnetic sublevels and Raman transition configurations are shown in Fig. 1. The
ground state magnetic sublevel will be shifted by
when a static magnetic field
exists, where
is atomic total angular momentum,
are the projections of total angular momentum on the
quantization axis,
is the Bohr Magneton, the Landé g factor
equals 1/2 for the
state and equals -1/2
for the
state. If the magnetic quantization axis is absent, seven transition peaks (dashed black lines in Fig.1)
will be observed by sweeping the relative frequency of the two Raman lasers. However, if a quantization magnetic
field parallel to the propagating direction of the Raman beams is applied, and the Raman beams are
or
polarized, according to the electric dipole transition selection rules, only three transition peaks
(
,
) can be observed (green lines with double arrows in Fig.1) and the magnetic
linlinμT87Rb,FFmBFFEgmBBF0,1,FmFBFg2F1F2,1,FFFmFm0,1Fm quantum numbers
remains constant before and after Raman transitions. In this case, the resonance frequencies of
the
transition, denoted as
(
), can be achieved by fitting the Raman
spectrum with [30]:
,
(1)
where
is the effective Rabi frequency and
is the Rabi frequency of the
( =1, 2)
Raman beam,
is the Doppler frequency shift,
is the differential ac Stark light shift,
is the ac Stark light shift of the
Raman beam,
is the single photon detuning, and
is the
Raman pulse duration. After achieving the three resonance transition frequencies, the magnetic field intensity at the
atom-laser interaction position can be inferred from the frequency difference between
and
, denoted as
,
or from the frequency difference between
and
, denoted as
. Both magnetic field results write:
,
(2)
where
is the first order Zeeman coefficient [24].
Fig. 1.
ground state magnetic sublevels and Raman transition configurations. The green lines with double arrows are
possible transitions when a quantization magnetic field parallel to the Raman beams is applied, and the black dashed lines are
possible transitions if the quantization axis is absent.
is the energy level shift induced by first order Zeeman effect, and
(
) are the resonance frequencies of the
transitions.
B. Experimental apparatus
The schematic diagram of the experimental setup is shown in Fig. 2 and a detailed description of the whole
system can be found in Ref. [31,32]. In this experiment, approximately
atoms are first cooled and trapped
in the magnetically shielded Magneto-Optical Trap (MOT) chamber within 0.6 s, and then launched vertically by
moving molasses technique, achieving a temperature of ~2
and an initial velocity of ~4.4 m/s in 3 ms. The
repumping laser, whose frequency is near resonance with
transition frequency, is switched off 1
ms later than the cooling laser to ensure all atoms are in the
ground state. The collimated Raman π pulse with
an
diameter of 29.5 mm are irradiated from the top vacuum window while the atoms are moving in the
Fm2,1,FFFmFm,FFmm0,1Fm2222122sin()()effeffDACeffDACPEE12/4eff22iisIIthiiD12ACACAC2/4ACiithi1,10,01B1,10,0-1B11,10,0110,01,11()/()/BB12π14Hz/nTF = 1F = 25 2S1/25 2P3/2Δ δ (2π×7 Hz/nT)δ (-2π×7 Hz/nT)F'-10+1+2mF = -2σ1-σ1+ σ2- σ2+ω-1,-1ω1,1ω0,087RbE,FFmm0,1Fm2,1,FFmm91087RbμK12FF=2F2e magnetically shielded interferometer chamber. As shown in Fig.2, the magnetic shield consists of three equally spaced
concentric cylinders made from 0.75 mm thick sheets of high-permeability nickel-iron-molybdenum alloy. The
innermost cylinder has a length of 68 cm and a diameter of 8 cm while the outermost is 72 cm and 13 cm with the
middle and outer cylinder caped to the open diameter of the inner cylinder. This shield provides an attenuation factor
of roughly 1000 for background magnetic field. A solenoid, driven by a precision laser diode current driver whose
root of mean square (RMS) current noise is ~1
, is wound precisely inside the inner mu-metal shield to create a
highly homogeneous quantization magnetic field in vertical direction. When the atoms fall down through the detection
chamber, a normalized fluorescence detection process is applied. Specifically, an 18 mm
diameter detection pulse
whose frequency is near resonance with the
transition frequency is applied for 320
in
horizontal plane and a photomultiplier tube is applied in orthogonal direction to detect the atom population
in
state. Afterwards, a 60
repumping beam and a second 320
detection pulse are applied orderly to
measure the total atom number
. The Raman transition probability can be inferred by
.
The whole launch-detection experimental cycle time is ~1.5 s.
Fig. 2. Schematic diagram of the experimental setup.
III.
Experimental results
A. The Raman pulse duration and frequency step dependent magnetic field measurement uncertainty
To investigate the influences of Raman pulse duration
and frequency step size
on the Raman
spectroscopy-based magnetic field measurement, we first measured the magnetic field at one fixed position (the same
timing for Raman pulse) with different Raman pulse duration and frequency step size. The obtained Raman spectra is
shown in Fig. 3, in which Fig. 3(a)-(c) show the influence of frequency step size for the same Raman pulse duration,
Fig. 3(a) and 3(d) (Fig. 3(b) and 3(e)) show the influence of Raman pulse duration for the same frequency step. The
three resonance frequencies
(
) of each Raman spectrum are obtained by fitting it with a combined
Raman transition function of Eq. (1) (red lines in Fig. 3). The RMS fitting uncertainty of the three resonance
frequencies divided by
is defined as the magnetic field measurement uncertainty. This fitting uncertainty is also
confirmed by the Bootstrap method[33].
μA2e23FFμs2N=2Fμsμs12NN12121PNNNfFm0,1Fm1
Fig. 3. Raman spectra obtained at one fixed height with different Raman duration
and frequency step
. Black points
are measured data and red lines are fitting results with a combined function of Eq. (1).
The dependencies of measurement uncertainty on Raman pulse duration and frequency step size are shown in
Fig. 4. When the frequency domain sampling theorem is fulfilled (
), the measurement uncertainty
improves with decreasing frequency step size
approximatively as
, which matches the relationship of
, considering the number of sampling points
. As a result of fewer sampling points in the
transition peaks when increasing the Raman pulse duration with the same frequency step size, the steepness of the
measurement uncertainty is smaller than
(black dotted line with crosses in Fig. 4). Alternatively, by choosing a
suitable scanning region, i.e., taking data only within the position of the Raman peaks, one could improve these results,
getting closer to the expected 1/τ behavior. It's worth mentioning that when
,
, the achieved
magnetic field measurement uncertainty is 0.157 nT, which is better than the best Raman spectroscopy-based magnetic
field measurement result of 0.28 nT we found in Ref. [29]. An even lower measurement uncertainty can be achieved
by using a smaller
together with a longer
. However, a longer
corresponds to a lower spatial resolution
(assuming a fixed launch velocity), and a smaller
increases the time needed for a complete scan of the spectrum
and thus the time needed to map the magnetic field inside the whole interferometer chamber. Therefore, a compromise
should be made between the measurement uncertainty, the spatial resolution and the time consumption on the choice
of the experimental parameters for mapping the magnetic field of the whole interferometer chamber, especially when
the measurement time is limited, such as in field applications.
f1/(2)ffBUf1BUN1Nf125Hzf4msff Fig. 4. Magnetic field measurement uncertainty as a function of Raman pulse duration τ for different frequency step
.
When the frequency domain sampling theorem is fulfilled (
), the measurement uncertainty improves with decreasing
approximatively as
while improves with increasing τ with a smaller steepness than
. The measured data are
represented by the points, and the lines are simply to guide the eye. The data marked with (a)-(f) represent the achieved
measurement uncertainties corresponding to the spectrums shown in Fig. 3(a)-3(f).
B. The vector and tensor light shift induced magnetic field measurement offset
From Eq. (1) and (2) we can see that the magnetic field measurement accuracy is mainly affected by the Doppler
shift and ac stark shift. In copropagating Raman configuration, the Doppler effect is very small (223.7 mHz during
the 1 ms Raman pulse) compared to the 1st order Zeeman frequency shift (~70 kHz), and is canceled further by
calculating the difference of the adjacent transition peaks because they are identical for the three transition peaks (Fig.
3). The ac Stark energy shift on the hyperfine-structure state
induced by laser field
is [9,34-36]
,
(3)
where the superscripts
,
, and
distinguish the scalar, vector, and tensor parts of the polarizability
.
and
are unit vectors along the laser wave vector and quantization magnetic field, respectively.
is the degree of
circular polarization of the laser [37]:
for
laser and
for a linearly polarized laser (the vector light
shift drops out in this case).
is the complex polarization vector of the laser and may be expressed as
, where
is a real, the real unit vectors
and
(
) align with the
semi-major axis and semi-minor axis of the ellipse which swept out by the electric field vector of laser in one period,
respectively, and
, defined by
, is directly related to the degree of circular polarization
.
More information can be find in Figure 3.3 of Ref. [37].
represents the higher order terms.
From Eq. (2) and (3), it's clear that the scalar light shifts (SLS) are identical for all the three magnetic states
(
) thus is canceled out by calculating the frequency difference of the adjacent transition peaks (Fig. 3).
When the Raman laser is circular polarized (
), the vector light shift (VLS) will be opposite for the
and
(magnetically-sensitive) states [25] but equals 0 for the
(magnetically-insensitive) state, thus
will manifest as a "fictitious" magnetic field
. Assuming the intensity ratio of the two Raman lasers is
,
calculating the frequency shift in Raman transition caused by VLS using Eq. (7.471) of Ref. [35], and taking the
Zeeman splitting
into consideration, the VLS induced "fictitious" magnetic field in
Raman transition can be written as:
,
(4)
where
is a constant determined by the frequencies and intensity ratio of the two Raman lasers (see Eq. (A11) in
Appendix). Eq. (4) shows that a polarization dependent "fictitious" magnetic field
is created in the circular
f1/(2)ffBUf1FFm()..LitkrLtεeccE222431()()()31()22221FFACSVTLFFmFLFLFLmFFmEkBBFFFASVT()FkBΑ1Α0Amajmin=cos+sinieimajminmajmin=kminmajtanA40,1Fm1Α1Fm1Fm0FmvlsB12/IIqFBFmBFFEgmB0Fm22FVmvlsLBFEBmAvlsB polarized Raman configuration (
), and the amplitudes of
is inversely proportional to the Raman duration
as
. The measured magnetic field intensities
are shown as points in Fig. 5, in which the
superscript
(
) represents the polarization of Raman laser and the subscript
represents the magnetic
field inferred from the resonance frequency of the
magnetic state, respectively. The fitting results show the
absolute magnetic field is 5651.7 nT and the VLS induced offset is 26.8 nT when the circular polarized (
)
Raman pulse duration is 1 ms.
From Eq. (2) and (3), we can see that the tensor light shifts (TLS) are identical for
and
states
but different for
state, thus will lead to a difference between
and
. This difference is proportional to
the intensity of Raman laser but equal for both left- and right-handed circularly polarized Raman laser beams in our
experiments. The tensor polarizability
calculated from Eq. (7.471) of Ref. [35] is about one order of magnitude
smaller than the vector polarizability and tends to zero when the laser is far-detuned (see Eq. (12)-(13)), thus the effect
of TLS (the difference between magnetic fields inferred from different magnetic states, i.e.,
and
) is one
order of magnitude smaller than the effect of VLS (the difference between magnetic fields measured with different
Raman laser polarization, i.e.,
and
), as shown in Fig.5. By calculating the average value of the measured
results for both
and
polarization configuration (orange stars), the polarization dependent VLS can be
canceled and the effect of TLS can be extracted. The fitting result (orange line) shows that the TLS induced magnetic
field offsets are 2.2 nT when the Raman pulse duration is 1 ms and 44.3 nT when the Raman pulse duration is 50
,
respectively.
Fig. 5. The influence of vector and tensor light shift on measured magnetic field intensity. The measured results
are
shown as points, in which the superscript
(
) represents the polarization of Raman laser, and the subscript
represents the result inferred from the resonance frequency of the
transition peak, respectively. The
amplitude of the "fictitious" magnetic field induced by VLS is inversely proportional to the Raman duration as
. The
effect of TLS is one order of magnitude smaller than the effect of VLS.
C. Magnetic field map inside the interferometer chamber and its stability
Here we choose the parameters of 1 ms pulse length and 400 Hz frequency step size to map the magnetic field
inside the 68-cm-high interferometer chamber of GAIN. These parameters correspond to a time consumption of ~12.5
minutes (200 kHz frequency sweep range) and a measurement uncertainty of 0.72 nT (a sensitivity of 19.7
)
1ΑvlsB22πeffLFmB,FmFm1Α1Fm1Fm0Fm1B1B()TF1B1B1B1BμsFmB,Fm2,1,FFFmFm1vlsBnTHz for each measurement. In order to map the magnetic field inside the whole interferometer chamber, one has to irradiate
the atoms with Raman pulse at different times on the atom's trajectory. We choose a time delay of 4 ms for Raman
pulse irradiation between measurements, resulting in 69 measurement heights (~14 hours) in total. The whole magnetic
field mapping process is implemented automatically by making a 2-Dimension scan (Raman laser frequency scan and
Raman laser irradiation time scan). In order to obtain the absolute magnetic field intensity and identify the source of
the field inhomogeneity, we implemented the above mapping process four times with nominal (13 mA) and half (6.5
mA) solenoid currents for
and
polarization configurations, respectively.
The achieved absolute magnetic field maps of the interferometer chamber of GAIN for nominal and half solenoid
currents are shown in Fig. 6 (a) and (b), in which the height is referred to the center of the MOT chamber (Fig.2). The
spatial resolution of the magnetic field maps is determined by three factors: atom's flight distance during the 1 ms
Raman pulse, atom's flight distance during the 4 ms Raman pulse delay, as well as atomic cloud's diameter at the time
of Raman pulse. We here take atom's largest flight distance of 12.8 mm during the 4 ms Raman pulse delay time as a
lower limit of the spatial resolution.
Considering the magnetic field in vertical direction consists of solenoid magnetic field (
) and background
magnetic field (
) (the horizontal component of
is very small and omitted here), the measured magnetic fields
for nominal (
) and half (
) solenoid current can be written as
,
.
(5)
(6)
where the offset of the current driver, which may result in some offset in quantization magnetic field and need to be
analyzed later in detail, is omitted here. Therefore,
and
can be inferred from
,
.
(7)
(8)
The inferred solenoid magnetic field
is very homogeneous with a standard deviation (SD) of 6.24 nT (Fig.
6(c)), while the inferred background magnetic field
has a similar fluctuation of about 100 nT (Fig. 6(d)) as the
measured magnetic fields
and
(Fig. 6(a) and 6(b)). Therefore, the magnetic field inhomogeneity in the
interferometer zone is attributed to the residual background magnetic field.
snBbgBbgBNBHBNsnbgBBB0.5HsnbgBBBsnBbgB2()snNHBBB2bgHNBBBsnBbgBNBHB Fig. 6. Measured magnetic field map for the (a) nominal (13 mA) and (b) half (6.5 mA) solenoid current; Inferred field map
for the (c) solenoid and (d) background magnetic field. The data are represented by the points, and the lines are simply to guide the
eye. Relevant heights for our standard atom interferometer configuration are indicated in (a).
In order to evaluate the stability of the background magnetic field and current driver, we measure continuously
at one fixed position with nominal solenoid current for 12.5 hours (the experimental parameters are same as the
mapping process, namely 1 ms Raman pulse duration and 400 Hz frequency step). The Allan Deviation of the
measured magnetic field decreases to ~0.4 nT after
seconds and decreases again for longer averaging time
after a small increase, indicating the stability of the current driver and background magnetic field, as well as the
measurement precision of this method.
D. Quadratic Zeeman effect induced systematic error in GAIN
With the knowledge of the magnetic field intensity the atoms experienced during flight, denoted as
, the
quadratic Zeeman effect induced phase shift
and gravity error
in atom interferometer gravimeters can be
inferred from
,
(9)
,
(10)
where
is the quadratic Zeeman coefficient,
is the sensitivity function [38] of the
Mach-Zehnder (M-Z) atom interferometer, and
is the time interval between pulses.
As shown in Eq. (9), in order to decrease the influence of the quadratic Zeeman effect, GAIN is implemented in
fountain configuration in which situation the flight trajectory of the atoms during the first half and second half of the
interferometer path of the Mach-Zehnder (M-Z) atom interferometer is almost symmetric (see Fig. 6(a)), with a small
discrepancy from the fountain apex to ensure the atoms have a non-negligible velocity for Doppler-sensitive Raman
3210()Btzeemang22()()TzeemansTgtBtdt2zeemaneffgkT222π0.0575Hz/nT()sgtT transition. With typical experimental parameters of
,
,
, (
,
,
,
, see Fig. 6(a)), the gravity offset
inferred by the interpolation
integral of the magnetic field map for the nominal solenoid current is
. The uncertainty of the gravity
offset due to the uncertainty of the magnetic field, inferred from
, is
. Therefore, the
experimental parameters of 1 ms pulse duration and 400 Hz frequency step size are sufficient for subtracting the
quadratic Zeeman effect related systematic error to an uncertainty of
level.
For comparison, if this atom interferometer would be implemented in a free fall configuration (i.e., releasing the
atoms from a MOT at the top of the instrument), then the geometric symmetry of atom's moving trajectory will be
lost and the interval time between the Raman pulses will be decreased to
, the gravity error caused by the
magnetic field inhomogeneity will be enlarged to
with an enlarged uncertainty of
.
IV.
Discussion
For atom interferometer gravimeter, the intensity ratio
of the two Raman lasers is usually set to a particular
value in order to cancel the influence of the light shift [39]. Taking the experimental parameters of GAIN as example,
in which
,
, the differential energy shifts induced by the scalar, vector and
tensor light shifts of the
Raman transitions (see Eq. (A8) - (A10) in Appendix) can be
simplified to:
,
,
.
(11)
(12)
(13)
where the propagation direction of the Raman laser is assumed parallel to the quantization magnetic field, namely
in Eq. (A9).
Usually, the polarization (propagation) direction of the Raman laser is perpendicular (parallel) to the magnetic
quantization axis, corresponding to
in Eq. (13), thus the total differential light shift of the
Raman transition can be inferred from Eq.(11)-(13) as:
.
(14)
Furthermore, the Raman lasers of the atom interferometers are usually in
polarization configurations
[39], correspond to
, and the atoms are prepared in the magnetically-insensitive state, correspond to
.
According to Eq. (11)-(14), the vector light shift equals zero, and the total (including the scalar and tensor) light shift
can be canceled by setting the intensity ratio of the two Raman lasers to
. It's worth mentioning that this
result of
is more closer to the experimental result of
shown in Fig. 5.7 of Ref. [39] than the result
4.1mslaunchvπ20.73st0.26sTπ245.1cmhπ85.7cmhπ260.0cmhApex86.1cmhg2.04μGalg2gBUgUB0.52nGalnGal0.14sT12.47μGalg3.20nGalgUq112700MHz216.834GHz1,2,FFFmFm228.327-14.8142SLEhq221.069+2.1232FVLmFEmhqA222231-0.168-0.207+0.245+0.1052FTLmFEBhqmq=1kB20B1,2,FFFmFm222-8.534-14.646+(2.123+1.069)-0.245+0.1052FTotalLmFFEhqmmqAlinlin0A0Fm1.718q1.718q1.72q of
obtained when neglecting the tensor polarizability.
Eq. (11) – (14) also show that in order to cancel the influence of light shift, the Raman laser intensity ratio
for
atomic interferometers with atoms in magnetically-sensitive states (
) is different from the intensity ratio
for atomic interferometers with atoms in magnetically-insensitive state (
) considering the contribution of the
tensor light shift. Furthermore, for atom interferometer with magnetically-sensitive atom states (
), the
contribution of the vector light shift should be considered as well if the Raman laser is circular polarized (
).
V.
Conclusion and outlook
In this paper, we reported on the experimental investigation of Raman spectroscopy-based magnetic field
mapping method and the evaluation of quadratic Zeeman effect induced systematic offset in the Gravimetric Atom
Interferometer (GAIN). We show both Raman pulse duration and frequency step size dependent measurement
uncertainty, investigated the influence of vector light shift (VLS) and tensor light shift (TLS), and presented a method
to extract the absolute magnetic field intensity and the TLS. We mapped the absolute magnetic field inside the
interferometer chamber of GAIN automatically with 1 ms Raman π pulse and 400 Hz frequency step, achieving a
magnetic field measurement uncertainty of 0.72 nT and a spatial resolution of lower than 12.8 mm. We attributed the
magnetic field inhomogeneity of ~100 nT to the residual background magnetic field which can be decreased further
by improving the magnetic shield. The quadratic Zeeman effect induced gravity measurement offset in GAIN is
evaluated as
, in which the offset of the current driver and other error sources still need to be analyzed later
in detail. The methods shown in this paper can be used for precisely mapping the absolute magnetic field in vacuum
and reducing the systematic error budget in Raman transition-based precision measurements, such as atomic
interferometer gravimeters.
Acknowledgments
This material is based on work funded by the European Commission (FINAQS, Contr. No. 012986-2 NEST), by
ESA (SAI, Contr. No. 20578/07/NL/VJ) and by ESF/DFG (EuroQUASAR-IQS, DFG grant PE 904/2-1 and PE 904/4-
1). Qingqing Hu would like to thank the support of the National Natural Science Foundation of China under Grant No.
51275523, Specialized Research Fund for the Doctoral Program of Higher Education of China under Grant No.
20134307110009, and the Graduate Innovative Research Fund of Hunan Province under Grant No. CX2014A002.
Appendix: Calculation of the polarizabilities and light shifts in Raman transition
From Eq.(7.471) of Ref. [35], the scalar, vector, and tensor polarizabilities of atom in ground state
are:
,
,
(A1)
(A2)
,
(A3)
where
represents the resonant laser frequency of ground state
to excited state
,
is the frequency of
1.779qq1Fmq0Fm1Fm1Α2.04μGalF22223FFSFFFFFFd212262111111FFFFVFFFFFFFFFFFFd22240212111213123FFFFTFFFFFFFFFFFFFFdFFFF is the reduced matrix element [24],
and
are the Wigner 6-j symbols
laser,
[35].
By calculating the Wigner 6-j symbols, and taking the reduced matrix elements and the resonant laser frequencies
of
dipole transitions from Ref. [24], the scalar, vector, and tensor polarizabilities of the 87Rb atom in
and
ground states can be simplified to:
,
,
,
(A4)
(A5)
(A6)
where
is the reduced D2 transition dipole matrix element of 87Rb.
In the case of two-photon Raman transition, there are two laser fields,
and
. According to the ac
Stark energy level shift of hyperfine state
induced by laser field
, Eq.(3) in the text,
the differential energy level shifts on Raman transition caused by scalar, vector, and tensor light shifts can be inferred
by first summing the light shifts on
state and
state induced by the two Raman lasers
and
with frequencies of
and
(define intensity ratio
) respectively, and then calculating the differential
light shifts between
state and
state, which are written as:
,
(A7)
,
,
(A8)
(A9)
FFd111FFF112FFFFFFF1F2F2011121212222220111212321222222222221222325591818730615SSdd20111212122222201112123212222222222212223255624247201215VVdd2011121212222220111212321222222222221222325+-936362+-30615TTdd21232dJeJr1tE2tEFFm()..LitkrLtεeccE222431()()()31()22221FFACSVTLFFmFLFLFLmFFmEkBBFFFA1F2F1tE2tE1212/qII1F2F2221221112-2SSSSSLEqq222122111211()-242FVVVVVLmFEkBmqqA ,
(A10)
where
, calculated from Eq. (A4) - (A6), represent the scalar
, vector
, and tensor
polarizabilities
of 87Rb atom's
and
ground states induced by Raman lasers
and
with frequencies of
and
. In our experiment,
,
.
Considering the Zeeman splitting
, where
is the Bohr Magneton, the Landé g factor
equals 1/2 for the
state and equals -1/2 for the
state, the VLS induced "fictitious" magnetic field in
Raman transition can be written as:
where
is a constant determined by the frequencies and
intensity ratio of the two Raman lasers.
References
,
(A11)
C. E. Wieman, D. E. Pritchard, and D. J. Wineland, Rev. Mod. Phys. 71, S253 (1999).
A. Peters, K. Y. Chung, and S. Chu, Metrologia 38, 25 (2001).
N. Poli, F. Y. Wang, M. G. Tarallo, A. Alberti, M. Prevedelli, and G. M. Tino, Phys. Rev. Lett. 106, 038501
J. M. McGuirk, G. T. Foster, J. B. Fixler, M. J. Snadden, and M. A. Kasevich, Phys. Rev. A 65 (2002).
F. Sorrentino, Y. H. Lien, G. Rosi, L. Cacciapuoti, M. Prevedelli, and G. M. Tino, New J Phys 12, 095009
A. Derevianko and H. Katori, Review of Modern Physics 83, 2015 (2010).
P. Clade, E. de Mirandes, M. Cadoret, S. Guellati-Khelifa, C. Schwob, F. Nez, L. Julien, and F. Biraben, Phys.
J. K. Stockton, K. Takase, and M. A. Kasevich, Phys. Rev. Lett. 107, 133001 (2011).
B. Dubetsky and M. A. Kasevich, Phys. Rev. A 74 (2006).
N. Huntemann, M. Okhapkin, B. Lipphardt, S. Weyers, C. Tamm, and E. Peik, Phys. Rev. Lett. 108, 186
[1]
[2]
[3]
(2011).
[4]
[5]
(2010).
[6]
[7]
[8]
(2012).
[9]
[10]
Rev. Lett. 96, 033001 (2006).
[11] M. Cadoret, E. de Mirandes, P. Clade, S. Guellati-Khelifa, C. Schwob, F. Nez, L. Julien, and F. Biraben, Phys.
Rev. Lett. 101, 230801 (2008).
[12]
[13]
E. M. Rasel, Phys. Rev. Lett. 112, 203002 (2014).
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
(2013).
[23] M.-K. Zhou, Z.-K. Hu, X.-C. Duan, B.-L. Sun, J.-B. Zhao, and J. Luo, Phys. Rev. A 82 (2010).
[24]
[25]
B. Wu, Z. Wang, B. Cheng, Q. Wang, A. Xu, and Q. Lin, Metrologia 51, 452 (2014).
Y. Ovchinnikov and G. Marra, Metrologia 48, 87 (2011).
P. Gillot, B. Cheng, S. Merlet, and D. S. Pereira, F, Phys. Rev. A 93 (2016).
Z.-K. Hu, X.-C. Duan, M.-K. Zhou, B.-L. Sun, J.-B. Zhao, M.-M. Huang, and J. Luo, Phys. Rev. A 84 (2011).
H. Muller, S. W. Chiow, Q. Long, S. Herrmann, and S. Chu, Phys. Rev. Lett. 100, 180405 (2008).
P. Clade, S. Guellati-Khelifa, F. Nez, and F. Biraben, Phys. Rev. Lett. 102, 240402 (2009).
S. S. Szigeti, J. E. Debs, J. J. Hope, N. P. Robins, and J. D. Close, New J Phys 14, 023009 (2012).
L. Zhou et al., Gen. Relativ. Gravit. 43, 1931 (2011).
Y. Bidel, O. Carraz, R. e. Charriere, M. Cadoret, N. Zahzam, and A. Bresson, Appl. Phys. Lett. 102, 144107
G. Rosi, F. Sorrentino, L. Cacciapuoti, M. Prevedelli, and G. M. Tino, Nature 510, 518 (2014).
D. Schlippert, J. Hartwig, H. Albers, L. L. Richardson, C. Schubert, A. Roura, W. P. Schleich, W. Ertmer, and
D. A. Steck, Rubidium 87 D Line Data 2015).
R. Dong, R. Wei, Y. Du, F. Zou, J. Lin, and Y. Wang, Appl. Phys. Lett. 106, 152402 (2015).
2222221221112363231-2122FTTTTTLFFmmmEBqqlFjlSlVlT1F2F1tE2tE12112700MHz216.834GHzFBFmBFFEgmBBFg2F1F0Fm222122111222216FVVVVVmvlsLLBFBqqEBmAA21221112216VVVVBqq K. S. Hardman et al., Phys. Rev. Lett. 117, 138501 (2016).
A. Wood, L. Turner, and R. Anderson, Phys. Rev. A 94, 052503 (2016).
K. Beloy, Theory of the ac Stark effect on the atomic hyperfine structure and applications to microwave
P. Cheinet, B. Canuel, F. P. D. Santos, and A. Gauguet, IEEE Transactions on Instrumentation & Measurement
R. Charriere, M. Cadoret, N. Zahzam, Y. Bidel, and A. Bresson, Phys. Rev. A 85 (2012).
Z.-C. Zhou, R. Wei, C.-Y. Shi, T. Li, and Y.-Z. Wang, Chinese Physics B 20, 034206 (2011).
K. Moler, D. S. Weiss, M. Kasevich, and S. Chu, Phys. Rev. A 45, 342 (1992).
B. Efron and R. Tibshirani, Statistical science, 54 (1986).
F. Le Kien, P. Schneeweiss, and A. Rauschenbeutel, Eur Phys J D 67, 92 (2013).
D. A. Steck, Quantum and Atom Optics (Daniel A. Steck, Quantum and Atom Optics, available online at
[26]
[27] M.-K. Zhou, X.-C. Duan, L.-L. Chen, Q. Luo, Y.-Y. Xu, and Z.-K. Hu, Chinese Physics B 24, 050401 (2015).
[28]
[29]
[30]
[31] M. Hauth, C. Freier, V. Schkolnik, A. Senger, M. Schmidt, and A. Peters, Appl. Phys. B 113, 49 (2013).
[32] M. Schmidt, M. Prevedelli, A. Giorgini, G. M. Tino, and A. Peters, Appl. Phys. B 102, 11 (2010).
[33]
[34]
[35]
http://steck.us/teaching (revision 0.11.5, 27 November 2016). 2016).
[36]
[37]
atomic clocks, PhD theis, University of Nevada, (2009).
[38]
57, 1141 (2008).
[39] M. Hauth, A mobile, high-precision atom-interferometer and itsapplication to gravity observations, PhD theis,
Humboldt-Universität zu Berlin, (2015).
|
1905.02695 | 1 | 1905 | 2019-05-07T17:11:37 | Temporal pattern recognition with delayed feedback spin-torque nano-oscillators | [
"physics.app-ph"
] | The recent demonstration of neuromorphic computing with spin-torque nano-oscillators has opened a path to energy efficient data processing. The success of this demonstration hinged on the intrinsic short-term memory of the oscillators. In this study, we extend the memory of the spin-torque nano-oscillators through time-delayed feedback. We leverage this extrinsic memory to increase the efficiency of solving pattern recognition tasks that require memory to discriminate different inputs. The large tunability of these non-linear oscillators allows us to control and optimize the delayed feedback memory using different operating conditions of applied current and magnetic field. | physics.app-ph | physics | Temporal pattern recognition with delayed feedback spin-
torque nano-oscillators
M. Riou,1 J. Torrejon,1 B. Garitaine,1 F. Abreu Araujo,1 P. Bortolotti1, V. Cros,1 S. Tsunegi,3 K.
Yakushiji, 3 A. Fukushima,3 H. Kubota,3 S. Yuasa,3 D. Querlioz,4 M. D. Stiles,2 J. Grollier1
1Unité Mixte de Physique CNRS, Thales,Université Paris-Sud, Université Paris-Saclay,
91767 Palaiseau, France
2Center for Nanoscale Science and Technology, National Institute of Standards and Technology,
3National Institute of Advanced Industrial Science and Technology (AIST), Spintronic Research Center,
Gaithersburg, Maryland 20899-6202, USA.
4Center for Nanoscience and Nanotechnology, CNRS, Université Paris-Sud, Université Paris-Saclay,
Tsukuba, Ibaraki 305-8568, Japan
91405, Orsay, France
The recent demonstration of neuromorphic computing with spin-torque nano-oscillators has opened a
path to energy efficient data processing. The success of this demonstration hinged on the intrinsic
short-term memory of the oscillators. In this study, we extend the memory of the spin-torque nano-
oscillators through time-delayed feedback. We leverage this extrinsic memory to increase the
efficiency of solving pattern recognition tasks that require memory to discriminate different inputs.
The large tunability of these non-linear oscillators allows us to control and optimize the delayed
feedback memory using different operating conditions of applied current and magnetic field.
I.
INTRODUCTION
Recurrence can play a powerful role in information processing. It is thought to provide a source of
memory in the brain and allows recurrent artificial neural networks to process sequences [1], [2].
When a network is recurrent, input data can remain present in the network for an extended time
creating dynamical memory. The state of the network not only depends on the current input value, but
also on past values. For neuromorphic computing, recurrence has been adapted in hardware by
combining a feedback loop with the physical component that plays the role of the neuron. This
approach has been tested in different systems (electronic, optical, photonic) for tasks that require
memory such as chaotic series prediction [3] -- [11].
Here, we implement recurrence in lower power nano-devices called spin-torque nano-oscillators,
[12], [13] which are composed of two ferromagnetic layers separated by a non-magnetic material. In
these devices, magnetization dynamics is driven by spin-polarized current through an effect known as
spin-transfer torque. When current flows through the magnetic multilayer, the current becomes spin
polarized and transfers angular momentum between the magnetic layers, resulting in a torque on the
magnetization. For high enough current density, this torque can destabilize the magnetization of the
free layer, which is then set in a sustained precessional state. These magnetic oscillations are converted
into resistance oscillations due to magnetoresistance effects: the device then functions as an electrical
nano-oscillator. The frequency of these oscillators can be tuned from a few hundreds of megahertz to
tens of gigahertz by varying the applied electrical current and magnetic field.
This control makes spin-torque nano-oscillators promising as nanoscale microwave sources.
Moreover these oscillators have the all essential features needed as hardware neurons because they
have nanometer size, they can interact through their emitted magnetic fields [14], [15] or electrical
currents [16] to emulate synaptic coupling, the amplitude of the oscillation depends non-linearly on
the input current, and they have an intrinsic memory related to the relaxation of magnetization. Using
these features, neuromorphic computing with spin-torque oscillators was recently demonstrated [17] --
[20]. However the intrinsic memory of these oscillators derived solely from their relaxation time,
which is quite short: tens to hundreds of nanoseconds for vortex gyration [17], [18] and even less for
other types of magneto-dynamical modes such as uniform precession, bullet modes, or spin waves.
This short time limits the number of cognitive tasks that can be performed. In particular, sequence
problems such as chaotic series prediction or natural language processing require longer memory to
recognize temporal patterns. Delayed feedback spin torque oscillators have been studied both
theoretically [21], [22] and experimentally [23], [24]. But the delay in these studies was much smaller
(tens of nanoseconds) and applications were mainly related to the reduction of phase noise as well as
the optimization of emitted power and linewidth.
In this study, we use a delayed-feedback spin-torque nano-oscillator to remember and recognize
patterns. In Sec. II, we show that adding delayed-feedback to the oscillator can enhance the range of
the memory from one relaxation time (≈ 200 ns) to tens of relaxation times. In Sec. III, we emulate a
reservoir computer with a single, time-multiplexed oscillator [25], [26], and demonstrate the
performance of the extrinsic memory by recognizing temporal patterns composed of random sine and
square waveforms. Finally, in Sec. IV, we vary the current and magnetic field applied to the nano-
oscillator to find the optimal operating regime of the oscillator for pattern recognition.
FIG 1: Experimental set up. The spin-torque oscillator is subjected to DC current and perpendicular
magnetic field which set the operating point. It emits an oscillating voltage 𝐕𝐨𝐬𝐜(𝐭). The time-varying
input is generated by an arbitrary waveform generator. A diode allows measuring directly the
amplitude of the oscillations 𝐱(𝐭), which is used for computation. The feedback loop consists of an
electronic delay line (𝛕 =4.3 µs) and an amplifier (the total amplification in the line is +20 dB). The
signals are added with power splitters.
FIG 2: (a) Input spike (in magenta) sent by the arbitrary waveform generator to the spin-torque
oscillator. (b) Blue curve: variation in the amplitude of the delayed feedback oscillator response 𝐱(𝐭).
The shaded areas in blue and orange indicate the intrinsic and feedback memory respectively. The
operating point is 600 mT and -6.5 mA.
II.
EXPERIMENTAL IMPLEMENTATION OF DELAYED FEEDBACK SPIN
TORQUE OSCILLATOR
The experimental set up is illustrated in Fig 1. The nano-oscillator is a cylindrical magnetic tunnel
junction of diameter 375 nm. The pinned layer is a CoFeB-based synthetic antiferromagnet, the tunnel
barrier is MgO and the free layer is a 3 nm thick FeB layer. For this aspect ratio of the free layer, the
AmplifierDelay lineIDCArbitrary Waveform GeneratorVosc(t)HCoFeBFeBMgOInputDiode+375nmFeedback loop ( )IntrinsicmemoryFeedback memory(a) (b) vortex magnetization is the stable configuration in the ground state. Such oscillators exhibit a high
power (a few microwatts) over noise ratio (≈ 100), low phase noise [27], and show optimal
microwave properties for neuromorphic computing [17] -- [20]. The oscillator is connected to a DC
current source and is subjected to an external magnetic field to induce gyrotropic motion of vortex
core via spin transfer torque mechanism. These two tunable parameters set the regime where the
oscillator operates.
The classification process used in the next section starts by encoding the different patterns in time-
varying electrical input signals generated by an arbitrary waveform generator. This electrical input
signal moves the oscillator out of its operating point, thus changing the amplitude and phase of the
oscillator voltage response 𝑉osc( ) (in grey in Fig. 1). For simplicity, only the amplitude ( ) (in blue
in Fig. 1(a)) of the oscillations is used for computing. The amplitude is measured by placing a diode
after the oscillator to capture the envelope of the oscillating signal. The delayed feedback loop is
composed of an electronic delay line with a delay = 4.3 µs and a total amplification of about 20 dB.
The addition of the delayed feedback to the input is made with a power splitter. The emitted voltage
of the oscillator (10 mV to 15 mV) is reinjected by the delay line. The reinjected voltage level depends
on the signal amplitude emitted by the oscillator itself ( ) and thus depends on the operating point
(magnetic field and current). For the highest emitted voltage (at low magnetic field), the reinjected
amplitude can reach approximately 50 % of the amplitude of the input signal (250 mV peak to peak).
Finally, the oscillation amplitude is recorded with an oscilloscope.
The intrinsic memory of the oscillator and the feedback memory are illustrated in Fig. 2. A current
pulse of 200 ns duration is sent to the oscillator (in magenta in Fig. 2(a)) and the voltage amplitude of
oscillator ( ) is recorded by an oscilloscope (in blue in Fig. 2(b)) during a much longer time (20 µs)
to observe the reinjection effects and the memory induced by the delay line. The input spike creates a
perturbation in the oscillator and modifies the orbit of gyrotropic motion of the vortex core. This
changes the amplitude of the signal emitted by the oscillator ( ). After the spike, the vortex core
returns to the orbit defined by the fixed magnetic field and DC current: the amplitude of the
oscillations ( ) returns progressively to its initial level. During this time, which corresponds to the
relaxation, the oscillator still remembers the input because the oscillation amplitude has not returned to
its initial level. The relaxation time of magnetization dynamics, roughly defined by the magnetic
damping and the frequency of oscillator (𝑇𝑟𝑒𝑙𝑎𝑥~
1
𝛼𝑓
), is around 200 ns, which corresponds to the
range of the intrinsic memory of the oscillator (except when the oscillator operates in the immediate
proximity of the critical current, where the relaxation time increases but the emission level is very
low). This intrinsic memory, highlighted in blue in Fig. 2(b), has a duration of a few hundreds of
nanoseconds.
The feedback is implemented by propagating the perturbation of the oscillation amplitude in the
delay line for = 4.3 µs and reinjecting it into the oscillator. This injection induces new variations in
the amplitude of the oscillator response ( ). Indeed, echoes in the oscillator response are observed
every τ after the end of the input signal in Fig. 1(c). These echoes are the manifestation of the external
memory provided by the delayed feedback. Since after each τ the echo is more and more attenuated,
the feedback memory is a fading memory. The fact that memory fades is important to process
temporal sequence for which only the recent history is important [3]. For a magnetic field of 600 mT
and a DC current of -6.5 mA, we observe variations of the oscillator output even after 13 µs
(approximately 60 relaxation times). The range of the feedback memory depends of the delay in the
line. Here we observe a memory of 13 µs, which corresponds to three times the delay of the line, but
choosing a longer delay in the line would have resulted in a longer range of memory. Moreover, the
feedback memory is sparse, that is, the information that the oscillator received as input in the past is
only accessible every time τ (4.3 µs in our case) in the areas highlighted in orange in Fig. 2(b). In
between, this information cannot be extracted from the measured trace. If the input is discrete, the time
step of the input and the delay in the line are often chosen to be equal so the oscillator remembers the
input at previous time steps.
𝒊𝒏 to a
FIG 3: Principle of reservoir computing. The input 𝒖(𝒕) is connected via fixed connections 𝒘𝒊
recurrent neural network, called a reservoir. The reservoir maps the input 𝒖(𝒕) into a higher
dimensional state 𝑿(𝒕), that is, each neuron output is a coordinate of the projected input. The output
𝒚(𝒕) is obtained by combining the neuron responses with trained connections 𝑾, 𝒃.
Random fixconnections Recurrent neural network ( )Trainedconnections , Input Output ( )(a)(b) Preprocessed input ( )Input ( )Oscillation amplitude time traces ( )Offline linearcombination Output ( )Higher dimensionMapping ( )(c)(d)(e)(f)(g)Randomfixedconnections
FIG 4: Implementation of reservoir computing with a single oscillator. (a) The patterns to recognize
are discretized sine and square periods. (b) The discrete input 𝒖(𝒌) is a time sequence of these sine
and square periods, randomly arranged. (c) Preprocessed input 𝑱(𝒕) obtained by multiplying 𝒖(𝒌) with
𝒊𝒏
a fast varying sequence called a mask. This mask varies every θ and represents the coupling 𝒘𝒊
between the input and each virtual neuron of the reservoir. (d) 𝒙(𝒕) is the oscillator response in
oscillation amplitude to the preprocessed input 𝑱(𝒕). The oscillator plays the role of all virtual neurons
one after the other, every θ. (e) The higher dimension mapping 𝑿(𝒌) of the input 𝒖(𝒌) can be
retrieved in the different sequences of 𝒙(𝒕) of duration τ. In our experiment the oscillator emulates
𝒏 = 𝟐𝟒 neurons, 𝜽 = 𝟏𝟖𝟎 𝐧𝐬, 𝝉 = 𝟒. 𝟑 𝛍𝐬. (f) Then the output 𝒚(𝒌) is reconstructed offline on a
computer by combining linearly the responses of the virtual neurons 𝒚(𝒌) = 𝑾. 𝑿(𝒌) 𝒃 . (g) If the
output 𝒚(𝒌) is less than 0.5, the input 𝒖(𝒌) is classified as a sine, otherwise it is classified as a square.
III.
NEUROMORPHIC COMPUTING APPROACH: RESERVOIR COMPUTING
To test the efficiency of the extrinsic memory induced by reinjection for computation, we perform
pattern recognition of sine and square waveforms. To perform this task, we have adapted the delayed
feedback spin torque oscillator to a recurrent neural network approach, called reservoir computing,
which has been identified to be suitable for different hardware implementations: optical [3], [4],
photonic [5], [11] and spintronic devices [17], [18], [20]. Fig. 3 shows the architecture of a general
reservoir computing scheme. The input ( ) (in black in Fig. 3) is coupled with fixed connections
𝐼
(in magenta in Fig. 3) to a recurrent neural network referred to as reservoir (in blue in Fig. 3) where
the internal connections between neurons are chosen to be fixed and random. The reservoir maps the
(b) Preprocessed input ( )Input ( )Oscillation amplitude time traces ( )Offline linearcombination Output ( )Higher dimensionMapping ( )(c)(d)(e)(f)(g)(a)Patterns to recognize input ( ) in a higher dimensional state referred as reservoir state 𝑿( ), where the output of each
neuron represents a transformation of the projected input ( ) (note that the dimension of 𝐗( ) is
equal to the number of neurons 𝑛). If the reservoir is complex enough (which means that it is on the
one hand of high enough dimension, with sufficient non-linearity and on the other it has sufficient
memory), the transformed problem becomes linearly separable in the reservoir state. The output ( )
is obtained by combining the neuron outputs ( ( ) values) with trained connections 𝐖, .
The experimental implementation of reservoir computing with multiple interconnected devices is
challenging, but a simplified approach has been successfully demonstrated [3] based on constructing a
reservoir from a time-multiplexed single device. Here, the entire recurrent network is replaced by a
single non-linear node which serves as each virtual neuron one after the other. In our case, the non-
linear node is a spin-torque oscillator. The input is projected to higher dimension in time instead of in
multiple devices (see figures Fig. 4(a-g)). The input stream is an aggregation of discretized sine and
square periods (patterns in Fig. 4(a)). The goal is to return an output 0 if the input value ( ) belongs
to a sine and an output of 1 if the input value ( ) belongs to a square. The system processes a
discretized version of the input stream ( ) with each input value ( ) (in black in Fig. 4(b)) being
applied n times into the reservoir (where n is the number of virtual neurons) with a time step . The
varying connections between the input and the virtual nodes of the reservoir are captured by
multiplying ( ) by a sequence that varies every time step and repeats every = 𝑛 , the time
between different input values. We choose the connection between the input and the virtual neurons to
be either +1 or -1. The resulting preprocessed input ( ) (see Fig. 4(c)) is applied to the oscillator.
The oscillator then emits a transient amplitude response ( ) (Fig. 4(d)). The responses of the
𝑛 neurons ( ) in a spatial reservoir are here replaced by the response of the single oscillator ( )
sampled 𝑛 times over intervals of length during a time (Fig. 4(e)). If the time is shorter than the
relaxation time, ( ) depends on ( − ). This situation is analogous to a neural network consisting
of multiple connected devices. Here the coupling between the devices is provided by the influence of
previous states of the oscillator on its current state. If a delayed feedback loop with a delay = 𝑛 is
added to the oscillator, the oscillator also receives the signal ( − ) at time . The virtual neurons
therefore receive their own past output, so they are recurrently coupled to themselves. The equivalent
multiple-device neuron network architecture is represented in blue in Fig. 3. Dashed arrows are the
connections emulated by the relaxation and solid line feedback arrows are the connections emulated
by the delayed feedback loop. 𝐗( ) (dimension (n,1)) represents the mapping ( ) into a higher
dimensional space that is obtained from discrete values sampled every in each interval of length =
𝑛 in the time trace ( ) (highlighted by circles in Fig. 3(d)). The output value is obtained by the
linear combination ( ) = 𝐖 ∙ 𝐗( ) (Fig. 4(f)) (with 𝐖 of dimension (1, 𝑛)). If the output ( )
is less than 0.5, the input ( ) is classified as a sine. Otherwise, it is classified as a square (Fig. 4(g)).
In our experiment, we use 𝑛 = 24 virtual neurons. The oscillator acts as each virtual neuron every
= 180 ns time interval. To emphasize the role of the memory effect brought by delayed feedback, we
choose a larger time (comparable with the relaxation time) than that used in the previous spintronic
implementation without delayed feedback [17], [18]. The longer interval attenuates the intrinsic
oscillator memory effect so as to focus on the memory provided by the feedback. The input ( ) is
processed with a time step = 4.3 µs, which corresponds to the delay time in the feedback circuit. The
preprocessed input ( ) is created by the arbitrary waveform generator. The oscillator amplitude
response ( ) is recorded with an oscilloscope and the higher dimension mappings 𝑿( ) are extracted
from ( ). The optimal output weights 𝐖, , which are determined during the learning phase, and the
output ( ) are computed offline on a computer.
FIG 5 : (a) Pattern to recognize: the time-varying input 𝒖(𝒌) used to evaluate the delayed feedback
spin torque oscillator is composed of sine and square sequences. The 8 inputs from sine are designated
as si1-8 and the 8 inputs of a square are designated as sq1-8. (b) Comparison of the average
experimental time traces of 𝒙(𝒕) without feedback for si7 (represented in orange) and sq5-8
(c)(b)(a) (represented in cool colors, blues and greens). The shaded regions indicate twice the standard
deviation of 𝒙(𝒕) for sq5, sq8, and si7. The operating point is 300mT and -6.5mA. (c) Similar
comparison with feedback.
FIG 6: (a) 2D visualization of the higher dimension mapping 𝐗(𝒌) obtained from experiment for all
the different inputs 𝒖(𝒌) without feedback. The operating point is 300 mT and -6.5 mA. The
separation between sine and square cases is the vertical black solid x = 0.5. For a perfect recognition,
points corresponding to sine input should be left to this line and points corresponding to square input
should be to the right. Without feedback, the points for si7 (orange) and si3 (yellow) are mixed up
with the corresponding square cases. The error rate is 10.8 % in the test phase (69 errors). Five clusters
are observed corresponding to the five different input values. These clusters are highlighted by grey
ellipses and denominated by the letters a to e. (b) 2D visualization of the higher dimension mapping
𝐗(𝒌) obtained from experiment for all the different inputs 𝒖(𝒌) with feedback. The operating point is
300 mT and -6.5 mA. Thirteen clusters are observed confirming that feedback enables the oscillator to
separate inputs by remembering sequences of two consecutive inputs. These clusters highlighted by
grey ellipses are denominated by the numbers 1 to 13. The points for si7 and si3 are different from the
square cases and fall to the left of the vertical separation line. The final classification is almost perfect
with only 0.16 % error rate (one error) during the test phase.
12345678910111213abcde(a)(b)SineSquareSineSquare
IV.
PATTERN RECOGNITION USING DELAYED FEEDBACK SPIN TORQUE
NANO-OSCILLATOR
We choose to classify sine and square waveforms, as done in previous studies [4], [10], [17], [18],
to evaluate the performance of our delayed-feedback reservoir computing based on a spin torque
oscillator. Each period of the waveforms is discretized in 8 points giving 16 different inputs (Fig.
5(a)). We refer to the 8 points of the sine waveform as si1-8, which are represented in warm colors,
and the 8 points of the square waveform as sq1-8, which are represented in cool colors in Fig. 5(a).
Because of degeneracies, the inputs take 5 different values in a sine waveform and two different values
in a square waveform. To return the same output value for 5 different input values of a sine (or 2 in the
case of the square), the reservoir must be non-linear. In the sine period, the 3rd and the 7th point
(referred as si3 and si7 see Fig. 5(a)) have the values +1 and -1, which are the same values as in the
square waveform. In the absence of memory, it is impossible to classify the input value of +1 or -1 as
belonging either to a sine or to square waveform. Therefore, this temporal pattern recognition task
needs both non-linearity and memory in the neural network. The input ( ) is composed of 1280
points (160 periods of sine or square randomly arranged). The first half of the points is used for
training (to find optimum output weights 𝐖 and ) and the second half for testing. The operating
point of the oscillator is 300 mT and -6.5 mA.
Figures 5(b) and 5(c) show the average ( ) response over all the cases where input is the same
and twice the standard deviation for the si7, sq5, and sq8 cases with and without feedback. First,
without feedback, we can see that time traces ( ) for si7 overlap within the two-standard-deviation
uncertainties with time traces ( ) for sq5 to sq8 (Fig. 5(b)). When the feedback memory is included
(Fig. 5(c)), the time trace si7 differs from the respective sq time traces. The feedback also breaks the
symmetry between the sq points with the same input values. The traces for sq5 becomes significantly
different from the traces for sq6 to sq8 (see Fig. 5(c)).
To analyze the performance of the oscillator in the absence or presence of delayed feedback, we
plot functions of 𝐗( ) in a 2D map in Fig. 6(a) and (b). As mentioned in the previous section, each
long time trace corresponds to the mapping in 24 dimensions 𝐗( ) of an input value ( ). To
visualize the separation, the mappings 𝐗( ) are projected linearly in two dimensions in (Fig. 6(a) and
(b)). To see the separation between the sine and the square, the first coordinate (x axis) of the 2D
representation is given by the linear combination 𝐖 ∙ 𝐗 . It can be seen geometrically as a
projection of the data along the weight vector 𝐖. In this 2D projection, the separation between sine
and square is defined by the vertical line at x = 0.5 in Fig 6(a-b). Ideally the sine cases (si1-8) should
all fall on the left of this line and square cases (sq1-8) on the right, respectively.
The second coordinate (y axis) of the projected data points is the first component from a principal
component analysis (PCA) of the data reduced in the space orthogonal to 𝐖. The PCA separates the
clusters (collections of points that are neighbors in the 24-dimension reservoir state) in the data. The
PCA is performed in the space orthogonal to 𝐖 to project the data along two orthogonal vectors.
Without feedback, 5 clusters denominated a to e are observed in the 2D map. They correspond
respectively to the 5 different values taken by the input: 0 (si1 and si5), 0.71 (si2 and si4), 1 (si3 and
sq1-sq4), -0.71 (si6 and si8) and -1 (si7 and sq5-sq8). The ambiguous cases corresponding to si3 and
si7 are completely mixed up with sq1-4 and sq5-8, respectively. Since the time step has been chosen
close to the relaxation time of the oscillator, the intrinsic memory is too small to remember a two-point
sequence. With no feedback, the error rate for the identification of sine and square is 10.8 % for this
value of . All the 69 errors during test phase are due to bad classification between si3 and sq1-4 or
si7 and sq5-8. These results are in excellent agreement with the prediction and the qualitative analysis
of the time traces.
Without memory, we expect five clusters, because the input takes only five different values. With
memory, new clusters appear and the number of clusters depends of the range of the memory. If the
reservoir remembers one time step, the clusters in the higher dimension mapping should correspond to
the different sequences of two consecutive input points at time k and k-1. We expect nine different
clusters for the sine and five different clusters for the square. Similarly, if the reservoir can remember
two time steps in the past, eighteen clusters are expected.
With delayed feedback to provide memory, the 2D map shows 13 clusters instead of the 5 found
without delayed feedback. This symmetry breaking can be explained by a memory of one time step in
the past. We observed 13 (and not 14 clusters as expected) on the 2D projection because points for si7
and si8 overlap (cluster 8 in Fig. 6(b)). The first component of the PCA may have not conserved all
the clusters present in the initial 24-dimension reservoir state. However, all the other 13 expected
clusters are clearly separated in the 2D projection, confirming that the delayed feedback spin torque
nano-oscillator remembers sequence of two consecutive inputs. With delayed feedback, si3 and si7
become different from the square traces and a linear separation can be found with a very small error
rate (0.16 % errors on the test set which corresponds to only one error). Adding delayed feedback to
the oscillator allows it to distinguish si3 and si7 from square points because the feedback adds a
memory of the previous input point. For the optimal operating point (300 mT and -6.5 mA), the
feedback is very efficient and suppresses 98.6 % of the errors (Fig 6(a-b), Fig 9), demonstrating the
efficiency of the feedback memory for computation.
FIG 7: Reduction of errors on the ambiguous cases si3 and si7 due to feedback depending of the
operating point, Eq. (1): this reduction of the error is evaluated on the training set. The number of
suppressed errors is renormalized by the total number of errors on the training set without feedback.
Brighter colors indicate high reduction of the error. The magnetic field is swept from 200 mT to 600
mT and the DC current is swept from -2 mA to -7 mA. In 90 % of operating points, feedback helps
distinguishing si3 and si7 during the training phase. Error reduction is high for high DC current (larger
than -4 mA) and magnetic field between 250 mT and 400 mT
FIG 8: Normalized noise level as a function of the operating point: the magnetic field is swept from
200 mT to 600 mT and the DC current is swept from -2 mA to -7 mA. At low field (200 mT to 400
mT) and low current (-2 mA to -5 mA) the noise level in the oscillation amplitude response is high.
Areas of high noise (brighter color) correspond to areas where the feedback is not efficient at
suppressing errors.
FIG 9: Proportion of si3 and si7 misclassified on training set without feedback, Eq. (2): the magnetic
field is swept from 200 mT to 600 mT and the DC current is swept from -2 mA to -7 mA. For 600 mT
and -5 mA, si3 and si7 are well classified even without feedback (only 17 % error). Other operating
points exhibit good performance (orange and red areas).
FIG 10: Global error reduction on the training set due to the feedback as a function of the operating
point, Eq. (3): the number of suppressed errors is renormalized by the total number of errors on the
training set without feedback. The magnetic field is swept from 200 mT to 600 mT and the DC current
is swept from -2 mA to -7 mA. Feedback reduces the error rate on the training set only in 66 % of the
cases. The feedback thus generates new errors in at least 24 % of the bias points. Brighter colors
indicate high reduction of the error. Error reduction is high for high DC current (larger than -4mA) and
fields between 250 mT and 400 mT.
FIG 11: Global error rate reduction due to feedback on the test set as a function of the operating point,
Eq. (4): the number of suppressed errors is renormalized by the total number of errors on the test set
without feedback. The magnetic field is swept from 200 mT to 600 mT and the DC current is swept
from -2 mA to -7 mA. Global error rate reduction during test phase is in good agreement with the
improvement during training phase. Feedback improves the result in 60 % of the cases during the test
phase. Brighter colors indicate high reduction of the error. Error reduction is high for high DC current
(larger than -4 mA) and fields between 250 mT and 400 mT.
Both the magnetic field and the DC current change the non-linear dependence of the voltage
oscillation amplitude with the input current (see figure 3 in Ref. 1) with the DC current acting like an
offset for the input current. The operating point fixed by these two parameters changes the
effectiveness of the feedback memory. As seen previously the reduction of the errors classifying the
si3 and si7 cases shows that the memory of the oscillator was enhanced. This improvement is
quantified by the proportion
𝑃𝑟1 =
,
(1)
𝑓𝑏 +𝐸𝑠𝑖7
𝑓𝑏)
(𝐸𝑠𝑖3
𝑛𝑜 𝑓𝑏+𝐸𝑠𝑖7
2(𝑁𝑠𝑖3
𝑛𝑜 𝑓𝑏)−(𝐸𝑠𝑖3
𝑡𝑟𝑎𝑖𝑛+𝑁𝑠𝑖7
𝑡𝑟𝑎𝑖𝑛)
𝑜 𝑓𝑏
, where 𝐸𝑠 3
𝑜 𝑓𝑏
and 𝐸𝑠 7
are respectively the number of misclassified si3 and si7 examples during
𝑓𝑏
training in the case without feedback, 𝐸𝑠 3
𝑓𝑏
and 𝐸𝑠 7
are respectively the number of misclassified si3
and si7 examples during training phase in the case with feedback and 𝑁𝑠 3
𝑡𝑟𝑎 and 𝑁𝑠 7
𝑡𝑟𝑎 are the total
number of si3 and si7 examples during training phase. Fig. 7 is a 2D map of this quantity as a function
of the operating point, where bright colors indicate high improvements and dark colors low
improvement or, in the case where 𝑃𝑟1 is negative, that more cases si3 and si7 are misclassified with
feedback. In 90% of the operating point conditions the feedback improves the recognition of the si3
and si7 cases (see Fig. 7), showing the memory it brings breaks the degeneracies between inputs from
square and sine periods. The low improvements are mainly at low magnetic fields: 200 mT and
between 250 mT and 400 mT and between -2.0 mA to -5.0 mA. This area correlates well with regions
where the amplitude noise of the oscillator is high compared to the signal amplitude (see Fig 8, bright
region).
Some other cases where 𝑃𝑟1 is negative or positive but low are observed on isolated operating
points such as 600 mT and -4.5 mA, 600 mT and -5.0 mA, 400 mT and -5.0 mA, 450 mT and -5.0
mA. Fig. 9 shows
𝑃𝑟2 =
(𝐸𝑠𝑖3
(𝑁𝑠𝑖3
𝑛𝑜 𝑓𝑏+𝐸𝑠𝑖7
𝑡𝑟𝑎𝑖𝑛+𝑁𝑠𝑖7
𝑛𝑜 𝑓𝑏)
,
𝑡𝑟𝑎𝑖𝑛)
(2)
which is the proportion of si3 and si7 inputs misclassified during the training phase in the case without
feedback. For these operating points, where higher relaxation time are measured, using the oscillator
without feedback already leads to partly good classification of the si3 and si7 cases (see Fig. 9).
Indeed with a 180 ns θ time step, the intrinsic memory is not completely negligible close to these
operating points, resulting in good classification even without feedback. In particular, at 600 mT and -
5.0 mA, 87% of the si3 and si7 cases are well classified without feedback. For 400 mT and -5.0 mA,
the feedback generates new mistakes on the si3 and si7 cases. In this particular case the feedback
works against the intrinsic memory of the oscillator.
The
total
error
improvement
for
the
training
phase
is
computed
as
Δ𝜖𝑡𝑟𝑎 =
𝜖𝑛𝑜 𝑓𝑏−𝜖𝑓𝑏
𝜖𝑛𝑜 𝑓𝑏
,
(3)
where 𝜖 𝑜 𝑓𝑏 and 𝜖𝑓𝑏 are the total error rates respectively without feedback and with feedback, such
that Δ𝜖𝑡𝑟𝑎 > 0 corresponds to an improvement of classification due to the feedback. For the training
phase, the feedback improves the result in 66% of the operating point conditions (Fig. 10). Areas
where the feedback is detrimental (Δ𝜖𝑡𝑟𝑎 < 0) correspond to areas where 𝑃𝑟1 was low or negative.
If, in the large majority of cases, feedback improves the classification of si3 and si7 cases, it may also
generate new errors that in some operating point conditions, overcome the benefit of external memory.
These new errors could be due to the increase of dispersion in the reservoir state. Indeed, if feedback
suppresses unwanted degeneracies between inputs from square and inputs from sine, it also generates
new clusters as seen at the beginning of this section. This effect is reinforced in areas where dispersion
is high due to stochastic behavior at threshold current and field for auto-oscillation (for 200 mT and
between 250 mT and 400 mT and -2 mA to -5 mA).
Computing the error gain on the testing data
Δ𝜖𝑡𝑒𝑠𝑡 =
𝜖𝑛𝑜 𝑓𝑏−𝜖𝑓𝑏
𝜖𝑛𝑜 𝑓𝑏
,
(4)
where 𝜖 𝑜 𝑓𝑏 and 𝜖𝑓𝑏 are the error rates for the testing data respectively in the case without feedback
and with feedback, we found that the dependency with the operating point is similar to Δ𝜖𝑡𝑟𝑎 with
the feedback improving in 60% of the cases (Fig. 11). Some values of Δ𝜖𝑡𝑒𝑠𝑡 are worse than Δ𝜖𝑡𝑟𝑎
because of bad generalization such as observed for instance at 400 mT and -5.0 mA. The feedback
brings memory in a vast majority of the operating point conditions and even though it may bring new
types of errors due to an increased dispersion of the data in the reservoir state, it still improves the
error rate in a large range of operating point conditions. Best improvements are obtained in areas of
low noise when compared to the amplitude and of low intrinsic memory.
CONCLUSION
Spin-torque nano-oscillators demonstrate clear memory effects when feedback is added. The
feedback oscillator can remember the effect of the input signal tens of times longer than it can just
with the intrinsic memory that is defined by its relaxation time. This feedback can be adapted to the
time steps of the input by tuning the delay line. We evaluate the efficiency of feedback as a memory
using a single spin torque oscillator with feedback as a reservoir computing recurrent network. The
single oscillator with time multiplexing projects the initial problem in a higher dimensional state,
turning it into a linearly separable problem. We tested this computation scheme on a temporal pattern
recognition task, sine and square waveform classification, which requires memory and non-linearity.
Choosing the optimal operating point, the error rate drops from 10.8 % without feedback to 0.16%
with feedback. Analyzing the different clusters that appear in the data with a 2D projection for the
optimal operating point shows that the delayed feedback remembers one time step in the past allowing
a nearly perfect separation of the data. We identify the optimal operating point to be when the
amplitude of the reinjected level is high and the noise of the oscillator is low (300 mT and -6.5 mA).
While we demonstrated the value of adding delayed feedback to a single node reservoir computing
scheme, delayed feedback can also be used as memory for a collection of many such oscillators. In this
work, we overcome the intrinsic memory limitation of spin-torque oscillators, opening the path to
solve complex sequence recognition with networks of spin-torque oscillators.
Acknowledgements
This work was supported by the European Research Council ERC under Grant bioSPINspired
682955. F.A.A. is a Research Fellow of the F.R.S.-FNRS.
[1] S. Hochreiter et J. Schmidhuber, « Long Short-term Memory », Neural Comput., vol. 9, p.
1735‑80, déc. 1997.
[2] M. Schuster et K. K. Paliwal, « Bidirectional recurrent neural networks », IEEE Trans. Signal
Process., vol. 45, no 11, p. 2673‑2681, nov. 1997.
[3] L. Appeltant et al., « Information processing using a single dynamical node as complex
system », Nat. Commun., vol. 2, p. 468, sept. 2011.
[4] Y. Paquot et al., « Optoelectronic Reservoir Computing », Sci. Rep., vol. 2, p. 287, févr. 2012.
[5] D. Brunner, M. C. Soriano, C. R. Mirasso, et I. Fischer, « Parallel photonic information
processing at gigabyte per second data rates using transient states », Nat. Commun., vol. 4, p.
1364, janv. 2013.
[6] A. Dejonckheere et al., « All-optical reservoir computer based on saturation of absorption »,
Opt. Express, vol. 22, no 9, p. 10868‑10881, mai 2014.
[7] Q. Vinckier et al., « High-performance photonic reservoir computer based on a coherently driven
passive cavity », Optica, vol. 2, no 5, p. 438‑446, mai 2015.
[8] F. Duport, B. Schneider, A. Smerieri, M. Haelterman, et S. Massar, « All-optical reservoir
computing », Opt. Express, vol. 20, no 20, p. 22783‑22795, sept. 2012.
[9] F. Duport, A. Smerieri, A. Akrout, M. Haelterman, et S. Massar, « Fully analogue photonic
reservoir computer », Sci. Rep., vol. 6, p. 22381, mars 2016.
[10] K. Vandoorne et al., « Toward optical signal processing using Photonic Reservoir Computing »,
Opt. Express, vol. 16, no 15, p. 11182‑11192, juill. 2008.
[11] K. Vandoorne et al., « Experimental demonstration of reservoir computing on a silicon photonics
chip », Nat. Commun., vol. 5, p. 3541, mars 2014.
[12] S. I. Kiselev et al., « Microwave oscillations of a nanomagnet driven by a spin-polarized
current », Nature, vol. 425, no 6956, p. 380‑383, sept. 2003.
[13] W. H. Rippard, M. R. Pufall, S. Kaka, S. E. Russek, et T. J. Silva, « Direct-Current Induced
Dynamics in Co90Fe10/Ni80Fe20 Point Contacts », Phys. Rev. Lett., vol. 92, no 2, janv. 2004.
[14] A. D. Belanovsky et al., « Numerical and analytical investigation of the synchronization of
dipolarly coupled vortex spin-torque nano-oscillators », Appl. Phys. Lett., vol. 103, no 12, p.
122405, sept. 2013.
[15] A. A. Awad et al., « Long-range mutual synchronization of spin Hall nano-oscillators », Nat.
Phys., vol. 13, no 3, p. 292‑299, mars 2017.
[16] R. Lebrun et al., « Mutual synchronization of spin torque nano-oscillators through a long-range
and tunable electrical coupling scheme », Nat. Commun., vol. 8, p. 15825, juin 2017.
[17] J. Torrejon et al., « Neuromorphic computing with nanoscale spintronic oscillators », Nature,
vol. 547, no 7664, p. 428‑431, juill. 2017.
[18] M. Riou et al., « Neuromorphic computing through time-multiplexing with a spin-torque nano-
oscillator », in 2017 IEEE International Electron Devices Meeting (IEDM), 2017, p. 36.3.1-
36.3.4.
[19] M. Romera et al., « Vowel recognition with four coupled spin-torque nano-oscillators », Nature,
vol. 563, no 7730, p. 230, nov. 2018.
[20] D. Marković et al., « Reservoir computing with the frequency, phase, and amplitude of spin-
torque nano-oscillators », Appl. Phys. Lett., vol. 114, no 1, p. 012409, janv. 2019.
[21] G. Khalsa, M. D. Stiles, et J. Grollier, « Critical current and linewidth reduction in spin-torque
nano-oscillators by delayed self-injection », Appl. Phys. Lett., vol. 106, no 24, p. 242402, juin
2015.
[22] V. S. Tiberkevich, R. S. Khymyn, H. X. Tang, et A. N. Slavin, « Sensitivity to external signals
and synchronization properties of a non-isochronous auto-oscillator with delayed feedback »,
Sci. Rep., vol. 4, p. 3873, janv. 2014.
[23] S. Tsunegi et al., « Self-Injection Locking of a Vortex Spin Torque Oscillator by Delayed
Feedback », Sci. Rep., vol. 6, p. 26849, mai 2016.
[24] D. Kumar et al., « Coherent microwave generation by spintronic feedback oscillator », Sci. Rep.,
vol. 6, p. 30747, août 2016.
[25] W. Maass, T. Natschläger, et H. Markram, « Real-time computing without stable states: a new
framework for neural computation based on perturbations », Neural Comput., vol. 14, no 11, p.
2531‑2560, nov. 2002.
[26] H. Jaeger et H. Haas, « Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy
in Wireless Communication », Science, vol. 304, no 5667, p. 78‑80, avr. 2004.
[27] S. Tsunegi, K. Yakushiji, A. Fukushima, S. Yuasa, et H. Kubota, « Microwave emission power
exceeding 10 μW in spin torque vortex oscillator », Appl. Phys. Lett., vol. 109, no 25, p. 252402,
déc. 2016.
|
1805.03752 | 1 | 1805 | 2018-05-09T22:52:16 | Active control of a plasmonic metamaterial for quantum state engineering | [
"physics.app-ph",
"quant-ph"
] | We experimentally demonstrate the active control of a plasmonic metamaterial operating in the quantum regime. A two-dimensional metamaterial consisting of unit cells made from gold nanorods is investigated. Using an external laser we control the temperature of the metamaterial and carry out quantum process tomography on single-photon polarization-encoded qubits sent through, characterizing the metamaterial as a variable quantum channel. The overall polarization response can be tuned by up to 33% for particular nanorod dimensions. To explain the results, we develop a theoretical model and find that the experimental results match the predicted behavior well. This work goes beyond the use of simple passive quantum plasmonic systems and shows that external control of plasmonic elements enables a flexible device that can be used for quantum state engineering. | physics.app-ph | physics | Active control of a plasmonic metamaterial for quantum state engineering
S. A. Uriri,1 T. Tashima,1 X. Zhang,1 M. Asano,2 M. Bechu,3, 4
D. O. Guney,5 T. Yamamoto,2 S¸. K. Ozdemir,6 M. Wegener,3, 4 and M. S. Tame1, ∗
1School of Chemistry and Physics, University of KwaZulu-Natal, Durban 4001, South Africa
2Department of Material Engineering Science, Graduate School of Engineering Science, Osaka University, Osaka 560-8531, Japan
3Institute of Applied Physics, Karlsruhe Institute of Technology, 76128 Karlsruhe, Germany
4Institute of Nanotechnology, Karlsruhe Institute of Technology, 76128 Karlsruhe, Germany
5Department of Electrical and Computer Engineering, Michigan Technological University, Houghton, MI 49931, USA
6 Department of Engineering Science and Mechanics, Pennsylvania State University, University Park, Pennsylvania 16802, USA
(Dated: May 11, 2018)
8
1
0
2
y
a
M
9
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
2
5
7
3
0
.
5
0
8
1
:
v
i
X
r
a
We experimentally demonstrate the active control of a plasmonic metamaterial operating in the quantum
regime. A two-dimensional metamaterial consisting of unit cells made from gold nanorods is investigated. Using
an external laser we control the temperature of the metamaterial and carry out quantum process tomography on
single-photon polarization-encoded qubits sent through, characterizing the metamaterial as a variable quantum
channel. The overall polarization response can be tuned by up to 33% for particular nanorod dimensions. To
explain the results, we develop a theoretical model and find that the experimental results match the predicted
behavior well. This work goes beyond the use of simple passive quantum plasmonic systems and shows that
external control of plasmonic elements enables a flexible device that can be used for quantum state engineering.
Introduction.- Metamaterials are artificially engineered
materials that provide optical, mechanical or thermal re-
sponses beyond what can be achieved by conventional mate-
rials [1–4]. In optics, metamaterials have traditionally been
made from metallic nanostructures much smaller than the
wavelength of interest, with the collective behavior of many
nanostructures giving rise to the bulk response of the mate-
rial [5]. The use of metal in the design of optical metama-
terials is mainly due to the ability to modify the electric and
magnetic resonances of the nanostructures in the optical band
easily using plasmonic techniques [6]. However, the use of di-
electric metamaterials has also been considered for achieving
similar resonance behaviour [7]. In recent years, researchers
have studied in great detail the electric and magnetic res-
onances of plasmonic nanostructures, using them to design
metamaterials for a wide range of applications, including new
types of lenses and imaging devices [8], transformation optics
components [9], sensing platforms [10], and many others [11].
Most recently, studies have started to probe optical meta-
materials in the quantum regime [12–22], with applications
in quantum state engineering, such as entanglement distil-
lation [23, 24], remote quantum interference [25], quantum
state generation [26, 27] and dissipative-based quantum state
control [28, 29]. However, a highly desirable behavior of
metamaterials is the ability to control their response. Much
work has been done in the classical regime in demonstrating
metamaterials that can be actively controlled [30–32]. Tech-
niques used include adding phase change materials [33–36],
embedding liquid crystals [37–39], using mechanical defor-
mation [40–43], electrical stimulation [32, 44] and exploiting
thermal effects [46–48]. In the quantum regime, there have
not yet been any studies showing the active control of a meta-
material. Given the many quantum applications of metama-
terials already demonstrated, it is important to investigate the
possibility of active control and develop flexible components
for quantum state engineering tasks.
In this work we experimentally demonstrate the active con-
trol of a metamaterial in the quantum regime. We investi-
gate a two-dimensional metamaterial, a metasurface [49–53],
that is comprised of unit cells made from gold nanorods, and
whose usefulness was recently shown in an experiment per-
forming entanglement distillation of partially entangled two-
photon states [23]. Here, we go beyond this work to show how
the response of this metasurface can be actively controlled.
Such control is useful in a real-world application in quan-
tum communication [54], where a single metasurface whose
properties can be tuned to cover a wide operating range is
desired in order to avoid the need to physically change the
metasurface being used. To provide this tunability, we control
the metasurface via its thermal response. We use an external
laser to vary the temperature optically and carry out quantum
process tomography on single photons sent through, with the
goal of characterizing the metasurface as a variable quantum
channel in the polarization basis. The process tomography
method provides full information about how the metamaterial
responds in the quantum regime under an external stimulus
and enables us to assess the quality of its quantum operation.
We find that with the correct nanorod dimensions one can
achieve tuneability of the metasurface's polarization response
by up to 33%. To help explain the results, a theoretical model
is developed and used. We find the experimental results match
well the predicted behavior. Overall, the result is rather sur-
prising given that we are increasing the amount of loss in the
nanorods by heating – one would expect the loss to have a neg-
ative impact on the operation of the metasurface. However, it
is exactly the extra loss that changes the resonance properties
of the nanorods and modifies the bulk response of the meta-
surface. Our work goes beyond previous studies using simple
passive plasmonic systems in the quantum regime and shows
that external control of plasmonic elements provides a versa-
tile metamaterial that can be used to carry out quantum state
engineering.
2
FIG. 1: Overview and experimental setup for demonstrating active control of a metamaterial in the quantum regime. (a) Pictorial representation
of one of the metamaterials used with a single photon (red) and an active control laser beam (white) sent through. The spot size of the control
and single-photon beams are the same in the experiment, however the control beam is shown as smaller for pictorial purposes. The inset
shows a 3D figure of the nanorods in each unit cell (dimensions considered are given in the main text). (b) The experimental setup, where a
nonlinear BiBO crystal is pumped at 405 nm, producing pairs of photons at 810 nm via spontaneous parametric down-conversion. One photon
is detected at detector DA and heralds the presence of a single photon in the other arm. Here, H is a half-wave plate, Q is a quarter-wave plate,
L is a plano-convex lens ( f = 25 mm), PBS is a polarizing beamsplitter, IF is an interference filter (λ = 810 nm and ∆λ = 10 nm), and DA and
DB are single-photon detectors. (c) Temperature dependence of the permittivity, d of the silica substrate (i), and that of the gold used for the
nanorods, Re[m] (ii) and Im[m] (iii).
Overview.- A diagram of the scenario for demonstrating
the active control of a metamaterial in the quantum regime is
shown in Fig. 1 (a). Here, a single-photon (red beam) and
an external control laser for heating (white beam) are inci-
dent on the metamaterial. The inset shows the geometry of
the nanorods in each unit cell. In Fig. 1 (b) the optical setup
used is shown. Pairs of photons are generated via parametric
down-conversion using a pump laser [55, 56], with one pho-
ton of the pair acting as a heralding photon (top path) and the
other as a probe photon (bottom path). Quantum information
is encoded in the polarization degree of freedom of the probe
single photons, with each photon representing a qubit. The
qubits are encoded in different states and then sent through
the metamaterial. The metamaterial is then characterized as a
quantum channel as the temperature is changed via the con-
trol laser. Further details of the setup are given later in the
experimental setup section.
Theoretical model.- Before introducing the experimental
setup and analysing the results, we briefly summarize our
theoretical model for the temperature dependence of a given
metamaterial. The transmission response of the type of plas-
monic metamaterial fabricated and shown in Fig. 1 (a) can
be modelled as a periodic array of nanoparticles in a planar
rectangular lattice with periods dx and dy. In the dipole ap-
proximation, each nanoparticle representing a unit cell of the
metamaterial is modelled by a dipole with a polarizability ten-
sor α, which relates the dipole moment P to the local electric
field E0, i.e. P = αE0 [57–60]. The plasmonic nanoparti-
cles fabricated are rod-like in shape and are well described as
an ellipsoid with semi-axes a, b and c. This gives a diagonal
polarizability tensor with non-zero elements [60]
m − d
αi = 4π0abc
3d + 3Li(m − d) ,
(1)
where 0 is the free-space permittivity, Li (i = x, y, z) is the ge-
ometrical factor [60], m is the relative permittivity of gold and
d is the relative permittivity of the surrounding medium (sil-
ica). To relate the semi-axes to the geometry of our nanorods
we set a = w/2, b = t/2 and c = l/2, where w, t, and l are the
nanorod width, thickness and length, respectively, as shown
in the inset of Fig. 1 (a).
The reflection and transmission through a given metamate-
rial nanorod array can be related to the polarizability tensor
of the nanorod unit cell and the interaction dyadic of the ar-
ray, as described in detail in Ref. [57]. For light with normal
incidence to the array and polarized in direction k we have
iµ0π f c
dxdy
αk
Tk = 1 +
1 − βkαk
(2)
and Rk = Tk − 1, where µ0 is the free-space permeability, f
is the frequency of the propagating electromagnetic wave, c is
the speed of light in vacuum and βk is the interaction dyadic.
Here, we set βk = 0 due to the large nanorod spacing con-
sidered in our experiment. Note that in the special case of no
absorption in the array the relation R2 + T2 = 1 is satisfied.
The final step is to introduce the temperature dependence of
the permittivities d and m, after which we are in a position
to model transmission through the metamaterial of qubits en-
coded into single photons as horizontal and vertical polariza-
tion states. We choose the vertical axis of the single photons
to be oriented along the long (length) axis of the nanorods.
3
0.1162414, B3 = 9.896161 and λ = 0.81 µm is the wavelength
of interest [62]. We then have dn
= (GR + HR2)/2n(Tr, λ),
dT
with R = λ2/(λ2 − λ2
ig), λig = 0.109 µm is the band-gap
wavelength of silica, G = −1.6548 × 10−6 K−1 and H =
31.7794 × 10−6 K−1 [61]. The temperature dependence of d
is shown in Fig. 1 (c) (i) for T = 300 to 350 K.
For the gold nanorods, the temperature dependence is de-
scribed using a modified Drude model, valid below the inter-
band transition frequency 2.4 eV (λ (cid:38) 520 nm) [63–65]
m(T) = ∞ −
ω2
p(T)
ω(ω + iωc(T)) ,
(4)
T
and ωe−ph(T) = ω0[2/5 +4(T/θD)5(cid:82) θD
where ω is the angular frequency of the electromagnetic
field, ∞ is the high-frequency permittivity of the metal, and
ωp(T) and ωc(T) represent the temperature-dependent plas-
mon frequency and collision frequency of the free electrons,
respectively. The plasmon frequency is given by ωp(T) =
ωp(Tr)/[1 + 3γ(T − Tr)]1/2, where ωp(Tr) is the plasmon fre-
quency at the reference temperature and γ = 14.2 × 10−6
K−1 is the thermal linear expansion coefficient. The colli-
sion frequency results from a combination of electron-electron
and electron-phonon scattering, with ωc(T) = ωe−e(T) +
ωe−ph(T), where ωe−e(T) = π3Γ∆[(KBT)2+(ω/2π)2]/12EF
z4(ez−1)−1dz]. Here,
kB is the Boltzmann constant, is Plank's constant, θD is the
Debye temperature, EF is the Fermi-level energy for gold, Γ
is the Fermi-surface average of scattering probability, ∆ is the
fractional Umklapp scattering coefficient and ω0 is a constant.
The following parameters are used for the above equations:
θD = 185 K, EF = 5.5 eV, Γ = 0.55 and ∆ = 0.77 [65].
Furthermore, the following parameters are obtained by fitting
the experimental data for gold from Ref. [63] at the reference
temperature (λ (cid:38) 600 nm) to Eq. (4): ω0 = 0.346 eV, ∞ = 8
and ωp(Tr) = 53.41 eV. The temperature dependence of m
is shown in Fig. 1 (c) (ii) for Re[m] and (iii) for Im[m], for
T = 300 to 350 K.
0
Using Eqs. (3) and (4) in Eq. (2) we are able to model the
temperature-dependent response of the metamaterial trans-
mission. As an example, in Fig. 2 (a) we show the transmis-
sion T2 for horizontal and vertical polarized light over the
wavelength range 600-1000 nm for a metamaterial at two dif-
ferent temperatures (T = 300 K and T = 340 K). The dimen-
sions used for the simulation are chosen based on the size of
the nanorods fabricated (see experimental section) and given
by dx = dy = 200 nm for the period, t = 30 nm for the thick-
ness, w = 46 nm for the width and l = 130 nm for the length.
One can clearly see the change in the transmission for verti-
cally polarized light as the temperature changes (lower solid
and dotted curves), whereas the transmission for horizontally
polarized light is not affected significantly (upper solid line).
This contrast is due to the dependence of the vertical transmis-
sion coefficient on the plasmon resonance along the length of
the nanorod, which is relatively strong and can change signifi-
cantly depending on the value of the permittivity of the metal.
On the other hand, for horizontally polarized light, the plas-
FIG. 2: Temperature-dependent transmission response of metamate-
rials with different nanorod dimensions (theory). In panels (a)-(f), the
period is fixed at dx = dy = 200 nm and the thickness is t = 30 nm.
In panels (a)-(c), the lower solid resonance curve is for vertical trans-
mission at T = 300 K and the lower dotted resonance curve is for
vertical transmission at T = 340 K. The horizontal solid line is for
horizontal transmission. The dimensions of the nanorods used are:
(a) width w = 46 nm and length l = 130 nm, (b) w = 47 nm and
l = 140 nm, and (c) w = 48 nm and l = 144 nm. Panels (d)-(f) show
the corresponding temperature dependence over a range of 50 K at
λ = 810 nm with the nanorod dimensions chosen as those used in
panels (a)-(c), respectively.
(g) Transmission response of a meta-
material with nanorod dimensions corresponding to panel (a) in the
middle, and with +/−2 nm added to the length, width and thickness.
To model the temperature dependence of the silica substrate
we use the wavelength-dependent thermo-optic coefficient dn
dT ,
where n is the refractive index (d = n2) and T is the temper-
ature [61, 62]. The refractive index is related to the thermo-
optic coefficient by the relation
n(T) = n(Tr) + (T − Tr)
dn
dT ,
(3)
where n(T) is the temperature dependent refractive index,
which is also wavelength dependent, and Tr = 300 K is a
reference temperature. It is known that the refractive index
of fused silica at Tr can be well described by the Sellmeier
equation n(Tr, λ) = [1 + A1λ2/(λ2 − β2
2) +
A3λ2/(λ2−β2
3)]1/2, where the coefficients are A1 = 0.6961663,
A2 = 0.4079426, A3 = 0.8974794, B1 = 0.0684043, B2 =
1) + A2λ2/(λ2 − β2
monic resonance is weak along the width of the rod and so
changes in the permittivity do not have a significant effect. In
Figs. 2 (a)-(c) and (g), a vertical line marks the wavelength
of interest for our experiment (λ = 810 nm). In Fig. 2 (d)
we show the temperature dependence of the transmission for
λ = 810 nm over the range 300-350 K. In order to under-
stand further how the transmission changes depending on the
nanorod dimensions we show two more examples of meta-
materials in Fig. 2 (b), (c), (e) and (f). The dimensions used
are the same as the previous example, but with w = 47 nm
and l = 140 nm for Fig. 2 (b) and (e), and w = 48 nm and
l = 144 nm for Fig. 2 (c) and (f). One can see that depending
on the nanorod dimensions the value of the transmission for
vertically polarized light can vary significantly as the temper-
ature is modified.
In Fig. 2 (g) we show how deviations in the nanorod di-
mensions (±2 nm for w, t and l) affect the transmission of
vertically polarized light through the metamaterial at a fixed
temperature of 300 K. One can see that with only a small de-
viation of 2 nm the transmission curve plotted as a function of
the wavelength of the incident light is shifted considerably to
the left (+2 nm) or to the right (-2 nm). This provides useful
information about how a realistic metamaterial might respond,
as consistency of nanorod dimensions across the array is hard
to achieve during fabrication. Based on the behavior seen in
Fig. 2 (g), the result of this would be a linewidth broadening
and a shift of the wavelength where the transmission becomes
minimum (the resonance wavelength).
Experimental setup.- In Fig. 1 (b) the optical setup used
in our experiment is shown. Here, single photons are gener-
ated using Type-1 spontaneous parametric down-conversion.
Pairs of single photons at λ = 810 nm are produced at angles
±3 degrees when a photon from a pump laser at 405 nm is in-
cident on a nonlinear BiBO crystal [55, 56]. The pump laser
(200 mW) is rotated to vertical polarization by a half-wave
plate (HWP) and incident on the BiBO crystal (0.5 mm thick-
ness). A single photon from the pump produces two 'twin'
(idler and signal) horizontally polarized photons. One pho-
ton is produced in arm A and the other in arm B. The detec-
tion of a single photon in arm A using a single-photon de-
tector (Excelitas SPCM-AQRH-15) heralds the presence of a
single photon in arm B within an 8 ns coincidence window.
A qubit is encoded onto the single photon in arm B using a
quarter-wave plate (QWP) and a HWP. Here, the polarization
states H(cid:105) and V(cid:105) are used as the orthogonal basis states of
the qubit. This qubit is then sent through the plasmonic meta-
material, after which the state of the qubit is measured via a
projective measurement using a QWP, a HWP and a single-
photon detector [66]. A broadband external control laser (Fia-
nium WhiteLase micro) is used to vary the temperature of the
metamaterial by heating it with a range of laser powers.
In order to quantify the performance of the metamaterial
as a quantum channel at different temperatures, quantum state
tomography is carried out on the output states of four differ-
(H(cid:105) +
ent polarization-encoded qubits {H(cid:105) ,V(cid:105) ,+(cid:105) = 1√
2
4
FIG. 3: Temperature-dependent transmission response of metama-
terials with different nanorod dimensions (experiment). Panels (a),
(b) and (c) are atomic force microscope images of three metamateri-
als used in the experiment. The period and thickness of the nanorods
are approximately equal, whereas the width and length increase from
(a) to (c). See main text for dimensions. Panels (d), (e) and (f) show
classical transmission spectra of the metamaterials at T0 for horizon-
tal (squares) and vertical (circles) polarized light. Panels (g), (h) and
(i) are transmission probabilities in the quantum regime for single
qubits encoded into the vertical polarization of single photons as V(cid:105)
and sent through the metamaterials as the temperature is changed.
The five different temperature settings are T0 = 295 K, T1 = 303 K,
T2 = 319 K, T3 = 331 K and T4 = 347 K, corresponding to values
consistent with the range used in the theory model. The values are
determined by the laser power used and are spaced apart by approxi-
mately 10 K.
2
(H(cid:105) + iV(cid:105))} sent through, with the density
V(cid:105)),R(cid:105) = 1√
matrices reconstructed via projective measurements [66]. The
output of a given projective measurement is sent to a single-
photon detector and a coincidence between the detector in the
heralding arm A and the detector in arm B is measured. In-
terference filters (810 ± 5 nm) are placed in front of each
detector to cut out photons of higher and lower frequencies
corresponding to unwanted down-conversion processes and
the pump beam, leading to ∼ 1000 coincidences per second
(for H(cid:105) encoded and H(cid:105) measured). The density matrices
obtained from quantum state tomography of the four probe
states are then used to reconstruct the quantum channel of the
metamaterial using quantum process tomography [67, 68], the
details of which are discussed later.
Three different plasmonic metamaterials were used in this
study, each with a specific set of dimensions for the gold
nanorod unit cells. The metamaterials were fabricated on
an indium tin oxide (ITO)-coated fused silica substrate by
electron-beam lithography. A 5 nm thin layer of ITO was
deposited on a 5 mm × 5 mm silica substrate by electron-
5
sions for the unit cells, as shown in the atomic force micro-
scope (AFM) images in Fig. 3 (a)-(c). The general trend in
dimensions is the same as that used in the theoretical model,
i.e. the length and width increase when going from (a) to (c).
Due to the background dielectric material not completely en-
compassing the nanorods, as well as the presence of the ITO
bonding layer and differences in the permittivity of gold, it
is not possible to obtain an exact fit of our theoretical model
to the experimental transmission data. However, the general
trend seen in the experimental classical transmission results of
Fig. 3 (d)-(f) matches well that seen in the theoretical model
of Fig. 2 (a)-(c) at 300 K, also taking into account broaden-
ing due to the fabrication process. The dimensions measured
by the AFM are w (cid:39) 50 nm and l (cid:39) 100 nm for Fig. 3 (a),
w (cid:39) 60 nm and l (cid:39) 110 nm for (b), and w (cid:39) 70 nm and
l (cid:39) 130 nm for (c). The thickness of the nanorods is 30 nm.
In Fig. 3 (g), (h) and (i) we show the transmission results of
probing the metamaterials with single photons in the state V(cid:105)
in arm B as the temperature is increased. Here, the transmis-
sion is given by the ratio of heralded detection counts (coin-
cidences) when the state V(cid:105) is sent through the metasurface
and counts when it is sent through the substrate only (no meta-
surface). It represents the relative probability for a photon in
the state V(cid:105) to be transmitted through the metamaterial. The
temperature is changed by increasing the power of the con-
trol laser in 4 steps, from 0 mW to 200 mW, which heats up
the metamaterial. The time between the control laser acti-
vation and the start of measurements is 480 s for each tem-
perature, however a steady state response is reached within
240 s. For quantum applications such as entanglement dis-
tillation [23, 24], this response time is practical as it is much
shorter than the time scale on which birefringent fluctuations
would occur in a realistic optical fiber quantum network [45].
The response time could be made shorter, if needed for a given
application, using alternative heating methods [46–48]. Con-
trol via the laser power gives five different temperature set-
tings: T0 = 295 K, T1 = 303 K, T2 = 319 K, T3 = 331 K
and T4 = 347 K, consistent with the range used in the theo-
retical model. The values are spaced apart by approximately
10 K and are determined by the power set by the control laser
software. They are measured using a point-probe temperature
sensor placed close to the laser beam on the metasurface sam-
ple. Measurements are carried out at the steady state response
time and the error in the values is < 1 K, consistent with am-
bient temperature fluctutations.
At T = T0, one can see in Fig. 3 (g), (h) and (i) that the
photon transmission slightly deviates from that of the classi-
cal transmission measured using vertically polarized light at
λ = 810 nm, as shown in Fig. 3 (d), (e) and (f). This de-
viation can be attributed to the spot size of the beam – in
the classical case the spot is smaller and easier to align on
the metasurface using a CCD camera, whereas in the single-
photon case the spot size is comparable to the metasurface
and alignment is achieved by optimising single-photon detec-
tion counts. As a result there is some non-ideal overlap of the
beam and the metamaterial. Regardless of this, the trend of
FIG. 4: Quantum process tomography χ matrices for the first meta-
material at two different temperatures, T0 and T4. Panels (a) and (c)
show the real and imaginary parts of the experimental χ matrix at
T0. Panels (b) and (d) show the real and imaginary parts of an ideal
partial polarizer χ matrix with TV = 0.476. Panels (e) and (g) show
the real and imaginary parts of the experimental χ matrix at T4. Pan-
els (f) and (h) show the real and imaginary parts of an ideal partial
polarizer χ matrix with TV = 0.324.
beam evaporation and then a 200 nm thick film of polymethyl-
methacrylate photoresist (MicroChem) was spin-coated on
top of the ITO. Using electron beam-writing (Raith e-line), the
photoresist was patterned and then developed, leaving a mask.
Subsequent gold evaporation and lift-off produced the gold
nanorod antenna arrays for the different metamaterials, each
with an area of 100 µm × 100 µm. The nanorods have a period
of 200 nm, a thickness of 30 nm, a width between 50 nm and
70 nm, and a length between 100 nm and 130 nm. Specific di-
mensions of a given metamaterial are provided later. The full
5 mm × 5 mm metamaterial sample consisting of an array of
metamaterials with different nanorod sizes is placed inside a
telescope system designed in such a way that the beam before
and after the lenses (planoconvex, f = 25 mm) is collimated
and the beam between the lenses is focused to a spot size with
diameter (cid:46) 100 µm.
Experimental results.- We now analyse our experimental
results in light of the theoretical model described. Here, each
of the three metamaterials used has different nanorod dimen-
6
FP
M1
TH
M2
TH
M3
TH
TV
TV
FP
TV
FP
0.935 ± 0.007 1 0.476 ± 0.008 0.909 ± 0.012 1 0.560 ± 0.009 0.895 ± 0.009 1 0.634 ± 0.008
0.899 ± 0.006 1 0.433 ± 0.008 0.935 ± 0.006 1 0.540 ± 0.009 0.922 ± 0.007 1 0.624 ± 0.011
0.911 ± 0.006 1 0.397 ± 0.007 0.934 ± 0.008 1 0.513 ± 0.009 0.942 ± 0.009 1 0.610 ± 0.009
0.922 ± 0.005 1 0.362 ± 0.007 0.948 ± 0.008 1 0.491 ± 0.008 0.970 ± 0.006 1 0.602 ± 0.010
0.897 ± 0.005 1 0.324 ± 0.006 0.912 ± 0.006 1 0.481 ± 0.008 0.939 ± 0.009 1 0.604 ± 0.010
Temp
T0
T1
T2
T3
T4
TABLE I: Process fidelities for the three metamaterials investigated as the temperature is changed, as well as horizontal and vertical transmis-
sion probabilities TH and TV extracted from maximizing the process fidelity, respectively.
the single-photon transmission at T0 matches that of the clas-
sical case as the nanorod dimensions increase. Moreover, as
the temperature increases one can see the effect on the trans-
mission of V(cid:105) states for the three metamaterials considered.
The largest change is seen for the first metamaterial, shown in
Fig. 3 (g), where the transmission changes from 0.48 to 0.32,
corresponding to a percentage change of 33 %. The percent-
age changes for the other two metamaterials are 14% and 5%.
We also measured the transmission of H(cid:105) states through the
metamaterials as the temperature was changed and found that
the transmission remained roughly the same as when the states
were sent through the substrate only. The exact transmission
values of the H(cid:105) state, as well as those of the additional probe
states D(cid:105) and R(cid:105) are combined with the values obtained for
the V(cid:105) state to obtain a full characterization of the metamate-
rial as a variable single-qubit quantum channel. The transmis-
sion values are part of a larger set of projective measurements
which we use for quantum process tomography [67, 68] at the
five different temperatures and discussed in detail next.
The four probe states sent through the metamaterial in the
quantum process tomography are H(cid:105), V(cid:105), D(cid:105) and R(cid:105). From
projective measurements on the outputs of these states in the
bases H/V(cid:105), D/A(cid:105) and R/L(cid:105), we reconstructed their den-
sity matrices using quantum state tomography [66]. Using
the density matrices we then obtained the quantum process
matrices, or χ matrices, for the three different metamaterials
in our investigation [67]. The polarization response of the
metamaterials is such that they act as partial polarizers and
are well represented by a single Kraus operator K0 = H(cid:105)(cid:104)H
TV V(cid:105)(cid:104)V corresponding to a non-trace preserving chan-
nel [23], i.e. ρ → K0ρK†
0 , where ρ is the qubit of the input
single-photon state in the polarization basis. This channel is
i j χi jEiρE†
j ,
where the single-qubit Pauli operators, Ei = I, X, Y and Z, pro-
vide a complete basis for the Hilbert space and the elements
of the χ matrix are set to values that allow the general channel
to completely match the action of K0 [23].
equivalent to the general quantum channel ρ →(cid:80)
√
+
The χ matrix obtained for the first metamaterial at T0 is
shown in Fig. 4 (a) and (c). Fig. 4 (a) shows the real part
and (c) shows the imaginary part. The real and imaginary
parts of an ideal partial polarizer matrix χid with TV = 0.476
are shown in Fig. 4 (b) and (d), respectively. The value
of TV has been found using the process fidelity, FP(TV) =
χ)2/Tr(χ)Tr(χid), which quantifies how close
the experimental channel is to an ideal channel of a partial
Tr((cid:112)√
χχid
√
polarizer. We find a maximum of FP(TV) = 0.935 ± 0.008 at
TV = 0.476±0.008, which shows that the metamaterial repre-
sents well a partial polarizer for single photons with a TV value
consistent with the single-photon transmission measured pre-
viously (see Fig. 3 (g)). The χ matrix for the first metamaterial
at T4 is shown in Fig. 4 (e) and (g). Fig. 4 (e) shows the real
part and (g) shows the imaginary part. The real and imagi-
nary parts of an ideal partial polarizer matrix with TV = 0.324
are shown in Fig. 4 (b) and (d). The value of TV has again
been found by maximising the process fidelity, with a value of
FP = 0.897 ± 0.005. The process fidelities and correspond-
ing TV values extracted for all three metamaterials at all tem-
peratures are given in Tab. I. All process fidelities are above
89%, with TV consistent with the values measured previously
(see Fig. 3 (g), (h) and (i)), showing the metamaterials act as
variable partial polarizers in the quantum regime. As a re-
sult, they can be used to induce a temperature-controlled col-
lective polarization-dependent loss at the single-photon level
for quantum information tasks, such as entanglement distilla-
tion [23].
Summary.- We investigated the active control of a plas-
monic metamaterial in the quantum regime via its thermal
response. Metamaterials with unit cells made from gold
nanorods were probed with qubits encoded into single pho-
tons. Using an external laser we controlled the temperature
of the nanorods and substrate. We then carried out quantum
process tomography, characterizing the metamaterials as vari-
able quantum channels. It was found that the overall polar-
ization response of the metamaterials can be tuned by up to
33% for particular nanorod dimensions. We used a theoretical
model to describe the thermal response of the metamaterials
and found that our experimental results matched the predicted
behavior well. Our work goes beyond previous studies of
simple passive plasmonic systems in the quantum regime and
shows that external control of plasmonic elements provides
variable metamaterials that can be used for quantum state en-
gineering tasks.
the UKZN Nanotechnology Platform,
Acknowledgments.- This research was supported by the
South African National Research Foundation, the National
Laser Centre,
the
South African National Institute for Theoretical Physics,
MEXT/JSPS KAKENHI Grant Number No.
JP15KK0164
and No. JP18H04291, the Helmholtz program Science and
Technology of Nanosystems (STN) and the Karlsruhe School
of Optics & Photonics (KSOP). S. K. O is supported by ARO
grant No. W911NF-18-1-0043 and Pennsylvania State Uni-
versity Materials Research Institute (MRI).
∗ Electronic address: [email protected]
[1] J. H. Lee, J. P. Singer and E. L. Thomas, Micro-nanostructured
mechanical metamaterials, Adv. Mater. 24, 4782-4810 (2012).
[2] J. Christensen, M. Kadic, M. Wegener and O. Kraft, Vibrant
times for mechanical metamaterials, MRS Commun. 5, 453-
462 (2015).
[3] S. A. Cummer, J. Christensen and A. Al`u, Controlling sound
with acoustic metamaterials, Nature Rev. Mater. 1, 16001
(2016).
[4] W. Cai and V. Shalaev, Optical Metamaterials: Fundamentals
and applications (Springer, Dordrecht, 2010).
[5] C. M. Soukoulis and M. Wegener, Past achievements and future
challenges in the development of three-dimensional photonic
metamaterials, Nature Photonics 5, 523 (2011).
[6] S. A. Maier, Plasmonics: Fundamentals and Applications,
[7] S. Jahani and Z. Jacob, All-dielectric metamaterials, Nature
(Springer, New York, 2007).
Nanotech. 11, 23-36 (2016).
[8] X. Zhang and Z. Liu, Superlenses to overcome the diffraction
limit, Nature Mater. 7, 435-441 (2008).
[9] H. Chen, C. T. Chan and P. Sheng, Transformation optics and
metamaterials, Nature Mater. 9, 387 (2010).
[10] T. Chen, S. Li and H. Sun, Metamaterials Application in Sens-
ing, Sensors 12, 2742-2765 (2012).
[11] L. Billings, Metamaterial world, Nature 500, 138 (2013).
[12] S. M. Wang, S. Y. Mu, C. Zhu, Y. X. Gong, P. Xu, H. Liu, T. Li,
S. N. Zhu and X. Zhang, Hong-Ou-Mandel interference me-
diated by the magnetic plasmon waves in a three-dimensional
optical metamaterial. Opt. Exp. 20, 5213 (2012).
[13] Z.-Y. Zhou, D.-S. Ding, B.-S. Shi, X.-B. Zou and G. C. Guo,
Characterizing dispersion and absorption parameters of meta-
material using entangled photons, Phys. Rev. A 85, 023841
(2012).
[14] A. Chipouline, S. Sugavanam, V. A. Fedotov and A. E. Niko-
laenko, Analytical model for active metamaterials with quan-
tum ingredients, J. Opt. 14, 114005 (2012).
[15] C. L. Cortes, W. Newman, S. Molesky and Z. Jacob, Quan-
tum nanophotonics using hyperbolic metamaterials, J. Opt. 16,
129501 (2014).
[16] J. A. Scholl, A. Garcia-Etxarri, G. Aguirregabiria, R. Esteban,
T. C. Narayan, A. L. Koh, J. Aizpurua and J. A. Dionne, Evo-
lution of Plasmonic Metamolecule Modes in the Quantum Tun-
neling Regime, ACS Nano 10, 1346-1354 (2016).
[17] K. McEnery, M. S. Tame, S. A. Maier and M. S. Kim, Tun-
able negative permeability in a quantum plasmonic metamate-
rial, Phys. Rev. A 89, 013822 (2014).
[18] S. A. Moiseev, A. Kamli and B. C. Sanders, Low-loss nonlinear
polaritonics, Phys. Rev. A. 81, 033839 (2010).
[19] M. Siomau, A. Kamli, S. A. Moiseev and B. C. Sanders, Entan-
glement creation with negative index metamaterials, Phys. Rev.
A 85, 050303 (2012).
[20] M. Zhou, J. Liu, M. Kats and Z. Yu, Atomic metasurfaces for
manipulation of single photons, ACS Photonics 4, 1279 (2017).
[21] M. S. Tame, K. R. McEnery, S¸. K. Ozdemir, S. A. Maier and M.
S. Kim, Quantum Plasmonics, Nature Physics 9, 329 (2013).
[22] A. M. Zagoskin, D. Felbacq and E. Rousseau, Quantum meta-
materials in the microwave and optical ranges, EPJ Quant. Tech.
7
3, 2 (2016).
[23] M. Asano, M. Bechu, M. Tame, S. K. Ozdemir, R. Ikuta, D.
O. Guney, T. Yamamoto, L. Yang, M. Wegener and N. Imoto,
Distillation of photon entanglement using a plasmonic metama-
terial, Sci. Rep. 5, 18313 (2015).
[24] Md Abdullah al Farooqui, J. Breeland, M. I. Aslam, M. Sadat-
gol, S¸. K. Ozdemir, M. S. Tame, L. Yang and D. O. Guney,
Quantum entanglement distillation with metamaterials, Opt.
Exp. 23, 17941-17954 (2015).
[25] P. K. Jha, X. Ni, C. Wu, Y. Wang and X. Zhang, Metasur-
face enabled remote quantum interference, Phys. Rev. Lett. 115,
025501 (2015).
[26] N. Gheeraert, S. Bera and S. Florens, Spontaneous emission
of many-body Schrdinger cats in metamaterials with large fine
structure constant, New J. Phys. 19, 023036 (2017).
[27] A. N. Poddubny, I. V. Iorsh and A. A. Sukhorukov, Genera-
tion of photon-plasmon quantum states in nonlinear hyperbolic
metamaterials, Phys. Rev. Lett. 117, 123901 (2016).
[28] T. Roger, S. Vezzoli, E. Bolduc, J. Valente, J. J. F. Heitz, J.
Jeffers, C. Soci, J. Leach, C. Couteau, N. I. Zheludev and D.
Faccio, Coherent perfect absorption in deeply subwavelength
films in the single-photon, Nature Commun. 6, 7031 (2015).
[29] C. Altuzarra, S. Vezzoli, J. Valente, W. Gao, C. Soci, D. Faccio
and C. Couteau, Nonlocal control of dissipation with entangled
photons, arXiv:1701.05357 (2017).
[30] A. D. Boardman, V. V. Grimalsky, Y. S. Kivshar, S. V. Koshe-
vaya, M. Lapine, N. M. Litchinitser, V. N. Malnev, M. Noginov,
Y. G. Rapoport and V. M. Shalaev, Active and tunable metama-
terials, Las. Phot. Rev. 5, 287-307 (2011).
[31] O. Hess, J. B. Pendry, S. A. Maier, R. F. Oulton, J. M. Hamm
and K. L. Tsakmakidis, Active nanoplasmonic metamaterials,
Nature Mat. 11, 573-584 (2012).
[32] N. I. Zheludev and Y. S. Kivshar, From metamaterials to
metadevices, Nature Mat. 11, 917-924 (2012).
[33] A. Gavinit and C. C. Y. Kwan, Optical Properties of Semicon-
ducting VO2 Films, Phys. Rev. B 5, 3138 (1972).
[34] P. Uhd Jepsen, B. M. Fischer, A. Thoman, H. Helm, J. Y. Suh,
R. Lopez and R. F. Haglund, Jr., Metal-insulator phase transi-
tion in a VO2 thin film observed with terahertz spectroscopy,
Phys. Rev. B 74, 205103 (2006).
[35] S. Prayakarao, B. Mendoza, A. Devine, C. Kyaw, R. B. van
Dover, V. Liberman and M. A. Noginov, Tunable VO2/Au hy-
perbolic metamaterial, App. Phys. Lett. 109, 061105 (2016).
[36] 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 Photon. 10, 60 (2015).
[37] D. H. Werner, D.-H. Kwon and I.-C. Khoo, A. V. Kildishev and
V. M. Shalaev, Liquid crystal clad near-infrared metamaterials
with tunable negative-zero-positive refractive indices, Opt. Exp.
15, 3342 (2007).
[38] I. C. Khoo, A. Diaz, J. Liou, M. V. Stinger, J. Huang and Y.
Ma, Liquid Crystals Tunable Optical Metamaterials, IEEE J.
Sel. Top. Quant. Elec. 16, 410 (2010).
[39] D. Shrekenhamer, W.-C. Chen and W. J. Padilla, Liquid Crystal
Tunable Metamaterial Absorber, Phys. Rev. Lett. 110, 177403
(2013).
[40] P. Gutruf, C. Zou, W. Withayachumnankul, M. Bhaskaran, S.
Sriram and C. Fumeaux, Mechanically Tunable Dielectric Res-
onator Metasurfaces at Visible Frequencies, ACS Nano 10, 133
(2016).
[41] H.-S. Ee and R. Agarwal, Tunable Metasurface and Flat Optical
Zoom Lens on a Stretchable Substrate, Nano Lett. 16, 2818-
2823 (2016).
[42] W. Liu, Y. Shen, G. Xiao, X. She, J. Wang and C. Jin, Mechan-
ically tunable sub-10nm metal gap by stretching PDMS sub-
strate, Nanotechnology 28, 075301 (2017).
[43] S. M. Kamali, E. Arbabi, A. Arbabi, Y. Horie and A. Faraon,
Highly tunable elastic dielectric metasurface lenses, Laser Pho-
ton. Rev. 10, 1002-1008 (2016).
[44] E. Feigenbaum, K. Diest and H. A. Atwater, Unity-Order Index
Change in Transparent Conducting Oxides at Visible Frequen-
cies, Nano Lett. 10, 2111-2116 (2010).
[45] A. Poppe, A. Fedrizzi, T. Loruenser, O. Maurhardt, R. Ursin,
H. R. Boehm, M. Peev, M. Suda, C. Kurtsiefer, H. Weinfurter,
T. Jennewein and A. Zeilinger, Practical Quantum Key Distri-
bution with Polarization-Entangled Photons, Opt. Express 12,
3865-3871 (2004).
[46] T. Driscoll, S. Palit, M. M. Qazilbash, M. Brehm, F. Keilmann,
B.-G. Chae, S.-J. Yun, H.-T. Kim, S. Y. Cho, N. M. Jokerst,
D. R. Smith and D. N. Basov, Dynamic tuning of an infrared
hybrid-metamaterial resonance using vanadium dioxide, App.
Phys. Lett. 93, 024101 (2008).
[47] X. Liu and W. J. Padilla, Thermochromic Infrared Metamateri-
als, Adv. Mater. 28, 871 (2016).
[48] W. Lewandowski, M. Fruhnert, J. Mieczkowski, C. Rockstuhl
and E. G´orecka1, Dynamically self-assembled silver nanoparti-
cles as a thermally tunable metamaterial, Nat. Comm. 6, 6590
(2015).
[49] A. V. Kildishev, A. Boltasseva and V. M. Shalaev, Planar Pho-
tonics with Metasurfaces, Science 339, 1232009 (2013).
[50] N. Yu and F. Capasso, Flat optics with designer metasurfaces,
Nature Mater. 13, 139 (2014).
[51] N. Meinzer, W. L. Barnes and I. R. Hooper, Plasmonic meta-
atoms and metasurfaces, Nature Photon. 8, 889 (2014).
[52] F. Xia, H. Wang, D. Xiao, M. Dubey and A. Ramasubrama-
niam, Two-dimensional material nanophotonics, Nature Pho-
ton. 8, 899 (2014).
[53] D. Lin, P. Fan, E. Hasman and M. L. Brongersma, Dielec-
tric gradient metasurface optical elements, Science 345, 298
(2014).
[54] N. Gisin and R. Thew, Quantum communication, Nature Pho-
ton. 1, 165-171 (2007).
[55] D. C. Burnham and D. L. Weinberg, Observation of simultane-
8
ity in parametric production of optical photon pairs, Phys. Rev.
Lett 25, 84-87 (1970).
[56] C. K. Hong and L. Mandel, Experimental realization of a local-
ized one-photon state, Phys. Rev. Lett 56, 58-60 (1986).
[57] A. Al`u and N. Engheta, Ch. 3 in Structured surfaces as opti-
cal metamaterials, Ed. A. A. Maradudin, Cambridge University
Press (2011).
[58] Y. Zhao, N. Engheta and A. Al`u, Homogenization of plasmonic
metasurfaces modeled as transmission-line loads, Metamateri-
als 5, 90 (2011).
[59] Y. Zhao and A. Al`u, Tailoring the Dispersion of Plasmonic
Nanorods To Realize Broadband Optical Meta-Waveplates,
Nano Lett. 13, 1086 (2013).
[60] C. F. Bohren and D. R. Huffman, Absorption and scattering of
light by small particles, Wiley (1983).
[61] K.-Q. Lin, L.-M. Wei, D.-G. Zhang, R.-S. Zheng, P. Wang, Y.-
H. Lu and H. Ming, Temperature effects on prism-nased surface
plasmon resonance sensor, Chin. Phys. Lett. 24, 3081 (2007).
[62] D. B. Leviton and B. J. Frey, Temperature-dependent absolute
refractive index measurements of synthetic fused silica, Proc.
SPIE 6273 June (2008).
[63] A. D. Raki´c, A. B. Djurisi´c, J. M. Elazar and M. L. Majewski,
Optical properties of metallic films for vertical-cavity optoelec-
tronic devices, Appl. Opt. 37, 5271 (1998).
[64] S¸. K. Ozdemir and G. Turhan-Sayan, Temperature Effects on
Surface Plasmon Resonance: Design Considerations for an Op-
tical Temperature Sensor, J. Light. Tech. 21, 805 (2003).
[65] J.-S. G. Bouillard, W. Dickson, D. P. O'Connor, G. A. Wurtz
and A. V. Zayats, Low-Temperature Plasmonics of Metallic
Nanostructures, Nano Lett. 12, 1561 (2012).
[66] D. F. V. James, P. G. Kwiat, W. J. Munro, and A. G. White.
Measurement of qubits. Phys. Rev. A 64, 052312 (2001).
[67] I. L. Chuang and M. A. Nielsen. Prescription for experimental
determination of the dynamics of a quantum black box. J. Mod.
Opt. 44, 2455 (1997).
[68] M. A. Nielsen and I. L. Chuang, Quantum Computation and
Quantum Information, Cambridge University Press, Cambridge
(2000).
|
1710.05630 | 1 | 1710 | 2017-10-16T11:32:27 | Activation of Microwave Fields in a Spin-Torque Nano-Oscillator by Neuronal Action Potentials | [
"physics.app-ph"
] | Action potentials are the basic unit of information in the nervous system and their reliable detection and decoding holds the key to understanding how the brain generates complex thought and behavior. Transducing these signals into microwave field oscillations can enable wireless sensors that report on brain activity through magnetic induction. In the present work we demonstrate that action potentials from crayfish lateral giant neuron can trigger microwave oscillations in spin-torque nano-oscillators. These nanoscale devices take as input small currents and convert them to microwave current oscillations that can wirelessly broadcast neuronal activity, opening up the possibility for compact neuro-sensors. We show that action potentials activate microwave oscillations in spin-torque nano-oscillators with an amplitude that follows the action potential signal, demonstrating that the device has both the sensitivity and temporal resolution to respond to action potentials from a single neuron. The activation of magnetic oscillations by action potentials, together with the small footprint and the high frequency tunability, makes these devices promising candidates for high resolution sensing of bioelectric signals from neural tissues. These device attributes may be useful for design of high-throughput bi-directional brain-machine interfaces. | physics.app-ph | physics | Activation of Microwave Fields in a Spin-Torque Nano-Oscillator by Neuronal Action Potentials
J. M. Algarin1, B. Ramaswamy2, L. Venuti3,4, M. E. Swierzbinski3,4, J. Baker-McKee7, I. N. Weinberg7,
Y.J. Chen8, I. N. Krivorotov8, J. A. Katine9, J. Herberholz3,4, R. C. Araneda4,5, B. Shapiro2,6 and E. Waks1,a)
1Institute for Research in Electronics and Applied Physics (IREAP), 2Fischell Department of
Bioengineering, 3Department of Psychology, 4Neuroscience and Cognitive Science Program, 5Department
of Biology, 6Institute for Systems Research (ISR), University of Maryland, College Park, Maryland, 20742,
United States.
7Weinberg Medical Physics Inc., Bethesda, Maryland, 20852, United States.
8Department of Physics and Astronomy, University of California, Irvine, California, 92697, United States
9HGST Research Center, San Jose, California, 95135, United States.
Abstract
Action potentials are the basic unit of information in the nervous system and their reliable detection and
decoding holds the key to understanding how the brain generates complex thought and behavior.
Transducing these signals into microwave field oscillations can enable wireless sensors that report on brain
activity through magnetic induction. In the present work we demonstrate that action potentials from
crayfish lateral giant neuron can trigger microwave oscillations in spin-torque nano-oscillators. These nano-
scale devices take as input small currents and convert them to microwave current oscillations that can
wirelessly broadcast neuronal activity, opening up the possibility for compact neuro-sensors. We show that
action potentials activate microwave oscillations in spin-torque nano-oscillators with an amplitude that
follows the action potential signal, demonstrating that the device has both the sensitivity and temporal
resolution to respond to action potentials from a single neuron. The activation of magnetic oscillations by
action potentials, together with the small footprint and the high frequency tunability, makes these devices
promising candidates for high resolution sensing of bioelectric signals from neural tissues. These device
attributes may be useful for design of high-throughput bi-directional brain-machine interfaces.
Introduction
At its core, the brain is a complex network of neurons connected by synapses. Action potentials are the
fundamental units of communication between neurons that form the basic building blocks for thought and
behavior [1]–[3]. Detecting these action potentials wirelessly with high spatial and temporal resolution is
highly useful to understand how the brain processes information and thought [1], as well as to diagnose
and treat neurological diseases [4]–[6].
A number of different techniques exist for wirelessly measuring human brain activity. For example,
functional magnetic resonance imaging provides wireless measurements with spatial resolutions on the
order of millimeters [7]. However, this technique only measures brain activity indirectly through
hemodynamic effects. Furthermore, it does not have the spatial and temporal resolution to isolate single
neurons or small clusters and read out individual action potentials [8], [9]. Other methods such as
magnetoencephalography provide excellent temporal resolution in the milliseconds range but exhibit very
imaging with
poor
electroencephalography or magnetoencephalography could potential yield
temporal
functional magnetic
resonance
improved
spatial
resolution
[10]. Combining
a)Author to whom correspondence should be addressed. Electronic mail: [email protected]
resolution[11], but spatial resolution still remains in the millimeters range. Currently, the most advanced
method for performing highly localized measurements of neuronal action potentials in humans and other
primates involves surgical implantation of electrodes to target areas of the brain [12]. Dongjin et al.
demonstrated an ultrasonic backscatter system that enables communication with such implanted
bioelectronics in the peripheral nervous system [13]. Other techniques like optical methods based on
voltage sensitive contrast agents (dyes, quantum dots) and optogenetics have also been demonstrated [14]–
[17]. But focused optical beams cannot penetrate the skull or deep tissue, and thus cannot access deep-brain
regions [18]. Currently, there is great need in neuroscience for new methods to transduce biological activity
to wireless signals that can penetrate through deep tissue.
The application of spintronics to biological sensing remains a relatively unexplored area that has potential
to resolve some of the difficult challenges inherent to wireless signal detection. Recent work incorporated
giant magnetoresistors into electrode arrays to perform magnetoencephalography [19], but this technique
is not wireless. Another compelling spintronic device is the spin-torque nano-oscillator (STNO), which
takes as an input small direct currents and converts them to microwave oscillations [20]–[24] that can report
wirelessly to a receiver by electromagnetic coupling [25]. The STNO is nanoscale in dimensions and can
operate at microwave frequencies ranging from 0.1 – 10 GHz. This property resolves the long-standing
challenge inherent to oscillators based on electrical LC circuits that are difficult to scale down to small
dimensions. A standard LC circuit of 10 m dimensions typically exhibits oscillation frequencies exceeding
100 GHz due to limits in achievable values of inductance and capacitance [13]. But these frequencies are
incompatible with biological tissues, which become highly absorbing above 5 GHz. Spin-torque oscillators
can operate at biologically compatible frequencies while maintaining nanoscale dimensions. Furthermore,
they can operate with small input currents, on the order of a micro-amp, which may be sufficiently low to
be directly driven by neurons without the need for amplifiers. Finally, the oscillation frequency of the
device shifts in the presence of an external magnetic field [22], [26], enabling the precession frequency to
encode spatial information by applying a magnetic field gradient, analogous to conventional magnetic
resonance imaging. These properties make STNOs promising candidates for detecting weak bioelectric
signals with high spatial precision, potentially, up to single cell resolution. But the ability of STNOs to
transduce biological signals to microwave fields remains an unexplored area.
Here, we demonstrate that a STNO can transduce a biological signal to microwave field oscillations. We
drive the device with action potentials from crayfish neurons. Crayfish possess giant neurons that generate
voltages on the order of a few millivolts when measured with extracellular recording electrodes, making
them an ideal system to study spintronic devices. We utilized the extracellular voltage produced by the
lateral giant neuron to drive the device and observed a clear microwave signal whose temporal envelope
accurately reproduced the action potential waveform. This result shows that spintronic devices could
potentially serve as nanoscale sensors for bioelectric signals with high spatial resolution and sufficient
bandwidth to temporally resolve neuronal action potentials.
The lateral giant escape circuit of crayfish is one of the best understood neuron circuits in the animal
kingdom [27], [28]. The key element is a pair of lateral giant neurons. The lateral giants receive
mechanosensory inputs in all abdominal segments and produce single action potentials that propagate along
the entire ventral nerve cord, the caudal part of the crayfish nervous system, to activate flexor motor neurons
[29], [30]. In freely behaving animals this leads to a rapid flexion of the tail and a stereotyped forward "tail-
flip" that thrusts the animal away from an attacking predator [31]. The lateral giants are the largest neurons
in the ventral nerve cord with axon diameters of up to 200 µm in adult crayfish and can be readily stimulated
with extracellular silver wire electrodes both in intact animals and in isolated nerve cords [32]. The
extracellular (field) potential generated by the lateral giant spike is large enough to be recorded outside the
animals during a naturally evoked tail-flip [33]. These large extracellular fields make the lateral giant
neurons ideal biological models to be interfaced with STNOs and produce a microwave signal.
Materials and methods
The STNO that we employ in this work is an elliptical magnetic tunnel junction nanopillar with lateral
dimensions 50 nm 190 nm. Fig. 1.a shows the complete layer structure for the device, with thicknesses
(in nanometers) indicated in parentheses. We deposited all layers using magnetron sputtering in a Singulus
TIMARIS system, and patterned the magnetic tunnel junctions using electron beam lithography followed
by ion milling. The synthetic antiferromagnet is PtMn/Co70Fe30/Ru/Co40Fe40B20 with the Co70Fe30 pinned
layer and the Co40Fe40B20 reference layer antiferromagnetically coupled by the tuned thickness of Ru. Prior
to patterning, we anneal the multilayer for 2 hours at 300 °C in a 1 T in-plane field to set the pinned layer
exchange bias direction parallel to the long axis of the nanopillars.
We obtained adult crayfish (Procambarus clarkii) of both sexes from a commercial supplier and kept
them in large communal tanks before the experiments. Individual animals (total body lengths 7-10 cm,
measured from rostrum to telson) were anaesthetized on ice for several minutes until immobility. We
separated the abdomen from the anterior part of the body and pinned it down in a petri dish. We removed
the membrane covering the ventral nerve cord and muscles, cut all ganglionic nerves, and dissected out the
ventral nerve cord. Next, we firmly pinned down the ventral nerve cord dorsal side up in a round petri dish
lined with silicone elastomer (Sylgard) and filled with fresh crayfish saline (Fig. 1.b). The saline in the dish
maintained a constant temperature of 20-21°C throughout the experiments. Only preparations that appeared
healthy were used, which allowed continuous measurements for several hours after the dissection.
We placed a pair of silver wire electrodes on the upper side surface of the ventral nerve cord to stimulate
the lateral giant neuron and a second identical pair of electrodes near the frontal end of the nerve cord to
record the lateral giant action potential (Fig. 1.b). To evoke lateral giant action potentials, we applied
voltage pulses with amplitudes of 5-10 V and pulse durations of 0.2-0.5 ms to the ventral nerve cord. We
stimulated using a data acquisition board (NI USB-6211) controlled by a LabVIEW (National Instruments)
program. We used a differential amplifier (A-M Systems, Model 1700) for stimulations and recordings. A
stimulus isolation unit (Grass, Model SIU5) applied a constant voltage stimulus. We used an amplifier with
1000x of gain to amplify the recorded signals, and then measured the signal using an oscilloscope (Lecroy
HRO 64Zi).
To drive the STNO with extracted signal from the crayfish, we utilized the experimental configuration
shown in Fig. 1.c. We placed the STNO in a home-built probe station and connected to the input and output
leads using a non-magnetic picoprobe (10-50/30-125-BeCu-2-R-200, GGB industries). An electromagnet
applied a magnetic field along the in-plane minor axis of the device to produce precession of the magnetic
free layer [34]. We connected the silver electrodes from the crayfish neuron to the input port of the device.
A bias tee separated out the direct electrical signal from the neuron from the induced microwave field
oscillations in the STNO. This technique provides access to both signals and enables us to compare the
direct neuron activity to the device microwave response. We measured the electrical action potential from
the lateral giant neuron using an oscilloscope (Lecroy HRO 64Zi). We measured the microwave signal
using a low noise amplifier (Pasternack PE15A1013) and a spectrum analyzer (Agilent 8564 EC). In order
to improve signal-to-noise, we averaged the output microwave signal over 1000-6000 stimulation pulses at
frequencies between 0.5-20 Hz. The lateral giant action potential can be easily identified by its characteristic
shape, large amplitude, fast conduction velocity, and low firing threshold. However, depending on electrode
placement, recordings can vary across preparations. This variability was more prominent when using the
STNO because it required using one recording electrode as a ground electrode; among some other minor
effects, this substantially reduced the amplitude of the recorded action potential. We stimulated the crayfish
neuron at subthreshold level as controls to confirm that recording experiments evoked neural activity. Since
thousands of stimuli at high frequency can lead to occasional lateral giant spike failure, we obtained best
results with longer inter-stimulus intervals. In total, we were able to successfully record three examples of
neural activity in the isolated ventral nerve cord from three different preparations using the STNO.
Results and discussion
We first characterized the STNO properties to determinate the strength of the external magnetic field that
produced the maximum output. We drove the device with an external power supply and monitored the
microwave response. Fig. 2 shows an average of 100 acquisitions of the power spectral density of the STNO
output for different direct voltages as well as the measurements for the optimal external magnetic field of
10 mT that produced the maximum microwave power output. These measurements show the device
oscillates with a frequency of 1.2 GHz and with amplitudes ranging from 0.05 fW/MHz to
0.42 fW/MHz for the given voltages from 0.25 mV to 0.75 mV, respectively. From the measured resistance
of the device (600 Ω), the voltage range corresponds to a peak input current in the range of 0.4 µA to 1.25
µA. Based on these currents, we conclude that the device works in the sub-threshold regime where the
applied current is below the critical current for zero-temperature onset of self-oscillations. In this case, the
observed microwave signal arises from temperature-induced precession of magnetization of the free layer
[35].
Using the optimized magnetic field, we applied repeated electrical stimuli to the lateral giant neuron to
evoke action potentials. Fig. 3.a shows the direct electrical signal from a crayfish neuron that we extracted
from the inductive port of the bias tee. The black trace shows the voltage that we measured with the
oscilloscope at the recording electrodes. In this specific case, the stimulus was a square pulse with an
amplitude of 10 V and a duration of 0.2 ms. The initial spike (before 0.5 ms) in the voltage trace is the
stimulus artifact due to direct coupling of the electrical signal from the stimulus electrode to the recording
electrode. At approximately 1 ms after the stimulus, we observe a second voltage pulse that corresponds to
an action potential, which reaches a maximum amplitude of 0.23 mV.
We next drove the STNO device directly with the output voltage of the neuron. The red trace in Fig. 3.a
shows the microwave power versus time at the frequency of 1.2 GHz with a bandwidth of 2 MHz. To
increase the signal-to-noise ratio we recorded an average of 3,000 sweeps (i.e., 3000 stimulations of the
neuron). The instantaneous microwave output power follows the voltage waveform of the action potential.
The large voltage at the beginning is due to direct activation of the STNO by the stimulus artifact. Following
the stimulus artifact, we recorded another peak power 1 ms later, which matches the action potential. We
found that the peak power that coincides with the action potential had a magnitude of 0.08 fW.
To determine the sensitivity of the device and assess the strength of the input signal, we generated artificial
action potentials using an analog voltage waveform generator. We stimulated the STNO with input voltage
pulses with temporal waveforms identical to those produced by a crayfish neuron (Fig. 3.b inbox blue line).
We artificially generated the action potential with the data acquisition board and the Labview software and
inputted it directly to the STNO by the inductive port of the bias tee. We employed action potentials with
amplitudes of 0.25 mV, 0.6 mV and 1.2 mV. Fig. 3.b shows the microwave power output from the STNO
versus time for artificial action potentials with a peak voltage range from 0.25 mV to 1.2 mV at a frequency
of 1.2 GHz, with a bandwidth of 2 MHz. To discriminate the signal from the noise, we needed to average
1,000 acquisitions for the 0.25 mV action potential, 100 acquisitions for the 0.6 mV and 10 acquisitions for
1.2 mV. We observed peak power of 0.1 fW, 0.8 fW and 2.8 fW for action potential amplitudes of 0.25
mV, 0.6 mV and 1.2 mV, respectively. The power that we obtained for 0.25 mV with the artificial action
potential agrees well with the result shown in Fig. 3.a. This result shows that increasing the peak voltage
by 4.8 times produced a 28-fold increase in the peak power making the detection easier. Taking into account
the device resistance (600 Ω) input currents on the order of microamps could provide single action potential
detection.
Conclusion
In summary, we demonstrated that action potentials from crayfish isolated nerve cords can generate
microwave signals in STNOs. The current device produced a peak power of 0.08 fW, which required
averaging 3,000 action potentials from a single neuron. A number of approaches could significantly
increase this power to the regime, which would allow us to measure a single action potential. Devices with
lower threshold currents could significantly increase the output power [36]. Devices with large-amplitude
magnetization precession [34], [37], reduced phase noise [38], oscillators with large volume of the free
magnetic layer [39] or Phase locked oscillators [40], [41] could further improve the sensing by emitting
more power in a narrower bandwidth. Another possibility is to stimulate simultaneously many spintronic
devices attached to a single neuron. According to the microwave output detected in the present work, an
array of 100 oscillators, fitting in an area over 1 µm2, could produce sufficient microwave power to detect
single action potentials. Improved electrodes with reduced contact resistance could also potentially increase
the input current to the STNO, thereby significantly increasing the generated microwave signal.
Alternatively, glucose fuel cells can bias an amplifier able to increase the input current to the STNO [42].
Ultimately, our results open up a new approach for high resolution sensing of bioelectric signals using
spintronic devices. STNOs occupy a small device footprint, potentially in the nanoscale, and operate at low
input currents, opening up the possibility for extremely dense low-power wireless sensor arrays.
Furthermore, the oscillation frequency of these devices is highly tunable through the external magnetic field
[22], [26]. In the presence of a strong magnetic field gradient, this property could enable STNOs to encode
their position in the oscillation frequency in an analogous way to magnetic resonance imaging. Furthermore,
the small size of these devices opens up the possibility to introduce them intravenously. Previous studies
showed that magnetic particles of similar dimensions can cross the blood-brain-barrier and reach targets in
the brain without disrupting the barrier in rat models[43], [44]. In addition to neuronal sensing, spintronic
sensors could be useful for detecting electrical signals from other tissue such as heart, or other muscles.
These properties could significantly enhance and extend current biological sensing capabilities.
ACKNOWLEDGEMENT
We thank Dr. John Rodger and Bisrat Adissie for providing access to the microwave equipment. We also
thank Pablo Villar del Rio who provided comments on this work. This work was supported by a seed grant
from the Brain and Behavior Initiative (BBI) at the University of Maryland, College Park. We gratefully
acknowledge support from a NSF BRAIN EAGER grant (grant number DBI1450921) as part of the BRAIN
initiative. The work of Yu-Jin Chen and Ilya Krivorotov on sample design and characterization was
supported as part of the SHINES, and Energy Frontier Research Center funded by the U.S. Department of
Energy, Office of Science, Basic Energy sciences under Award # SC0012670.
References
[1] C. Koch and I. Segev, "The role of single neurons in information processing," Nat Neurosci, vol. 3, pp.
[2] W. Gerstner and W. M. Kistler, Spiking Neuron Models: Single Neurons, Populations, Plasticity.
[3] D. Mareschal, Neuroconstructivism: How the brain constructs cognition. Oxford University Press,
1171–1177, Nov. 2000.
Cambridge University Press, 2002.
2007.
[4] M. Mantegazza, G. Curia, G. Biagini, D. S. Ragsdale, and M. Avoli, "Voltage‐gated sodium channels as
therapeutic targets in epilepsy and other neurological disorders," The Lancet Neurology, vol. 9, no.
4, pp. 413–424, Apr. 2010.
[5] M. H. Meisler and J. A. Kearney, "Sodium channel mutations in epilepsy and other neurological
disorders," J Clin Invest, vol. 115, no. 8, pp. 2010–2017, Aug. 2005.
[6] "The GABA Excitatory/Inhibitory Shift in Brain Maturation and Neurological DisordersThe
Neuroscientist ‐ Yehezkel Ben‐Ari, Ilgam Khalilov, Kristopher T. Kahle, Enrico Cherubini, 2012."
[Online]. Available: http://journals.sagepub.com/doi/pdf/10.1177/1073858412438697.
[7] P. M. Matthews and P. Jezzard, "Functional magnetic resonance imaging," Journal of Neurology,
Neurosurgery & Psychiatry, vol. 75, no. 1, pp. 6–12, Jan. 2004.
[8] J. W. BELLIVEAU et al., "Magnetic Resonance Imaging Mapping of Brain Function Human Visual
Cortex," Invest Radiol, vol. 27, no. 0 2, pp. S59–S65, Dec. 1992.
[9] J. Frahm, K.‐D. Merboldt, and W. Hänicke, "Functional MRI of human brain activation at high spatial
resolution," Magnetic Resonance in Medicine, vol. 29, no. 1, pp. 139–144, Jan. 1993.
[10] M. Haemaelaeinen, R. Hari, R. J. Ilmoniemi, J. Knuutila, and O. V. Lounasmaa,
"Magnetoencephalography‐‐‐theory, instrumentation, and applications to noninvasive studies of
the working human brain," Reviews of Modern Physics, vol. 65, no. 2, pp. 413–498, 1993.
[11] A. M. Dale and M. I. Sereno, "Improved Localizadon of Cortical Activity by Combining EEG and MEG
with MRI Cortical Surface Reconstruction: A Linear Approach," Journal of Cognitive Neuroscience,
vol. 5, no. 2, pp. 162–176, Apr. 1993.
[12] R. M. Rothschild, "Neuroengineering Tools/Applications for Bidirectional Interfaces, Brain–
Computer Interfaces, and Neuroprosthetic Implants – A Review of Recent Progress," Front
Neuroengineering, vol. 3, Oct. 2010.
[13] D. Seo et al., "Wireless Recording in the Peripheral Nervous System with Ultrasonic Neural Dust,"
Neuron, vol. 91, no. 3, pp. 529–539, Aug. 2016.
[14] A. C. Millard, L. Jin, M. Wei, J. P. Wuskell, A. Lewis, and L. M. Loew, "Sensitivity of Second Harmonic
Generation from Styryl Dyes to Transmembrane Potential," Biophys J, vol. 86, no. 2, pp. 1169–1176,
Feb. 2004.
[15] M. Canepari et al., "Imaging Submillisecond Membrane Potential Changes from Individual Regions
of Single Axons, Dendrites and Spines," in Membrane Potential Imaging in the Nervous System, M.
Canepari and D. Zecevic, Eds. Springer New York, 2010, pp. 25–41.
[16] G. Aston‐Jones and K. Deisseroth, "Recent advances in optogenetics and pharmacogenetics," Brain
Research, vol. 1511, pp. 1–5, May 2013.
[17] L. Fenno, O. Yizhar, and K. Deisseroth, "The Development and Application of Optogenetics," Annual
Review of Neuroscience, vol. 34, no. 1, pp. 389–412, 2011.
[18] A. Abdo and M. Sahin, "NIR Light Penetration Depth in the Rat Peripheral Nerve and Brain Cortex,"
Conf Proc IEEE Eng Med Biol Soc, vol. 2007, pp. 1723–1725, 2007.
[19] "In vivo magnetic recording of neuronal activity bioRxiv." [Online]. Available:
http://www.biorxiv.org/content/early/2016/12/09/092569. [Accessed: 18‐Sep‐2017].
[20] W. H. Rippard, M. R. Pufall, S. Kaka, S. E. Russek, and T. J. Silva, "Direct‐current induced dynamics in
Co90Fe10/Ni80Fe20 point contacts," Physical review letters, vol. 92, no. 2, p. 027201.1‐027201.4,
2004.
[21] A. M. Deac et al., "Bias‐driven high‐power microwave emission from MgO‐based tunnel
magnetoresistance devices," Nat Phys, vol. 4, no. 10, pp. 803–809, Oct. 2008.
[22] S. I. Kiselev et al., "Microwave oscillations of a nanomagnet driven by a spin‐polarized current,"
Nature, vol. 425, no. 6956, pp. 380–383, Sep. 2003.
[23] D. Houssameddine et al., "Spin transfer induced coherent microwave emission with large power
from nanoscale MgO tunnel junctions," Applied Physics Letters, vol. 93, no. 2, p. 022505, Jul. 2008.
[30] J. Herberholz, B. L. Antonsen, and D. H. Edwards, "A Lateral Excitatory Network in the Escape Circuit
of Crayfish," J. Neurosci., vol. 22, no. 20, pp. 9078–9085, Oct. 2002.
[31] J. Herberholz, M. M. Sen, and D. H. Edwards, "Escape behavior and escape circuit activation in
juvenile crayfish during prey–predator interactions," Journal of Experimental Biology, vol. 207, no.
11, pp. 1855–1863, May 2004.
[32] M. E. Swierzbinski, A. R. Lazarchik, and J. Herberholz, "Prior social experience affects the behavioral
and neural responses to acute alcohol in juvenile crayfish," Journal of Experimental Biology, vol.
220, no. 8, pp. 1516–1523, Apr. 2017.
[33] J. Herberholz, F. A. Issa, and D. H. Edwards, "Patterns of Neural Circuit Activation and Behavior
during Dominance Hierarchy Formation in Freely Behaving Crayfish," J. Neurosci., vol. 21, no. 8, pp.
2759–2767, Apr. 2001.
[34] G. E. Rowlands and I. N. Krivorotov, "Magnetization dynamics in a dual free‐layer spin‐torque nano‐
oscillator," Phys. Rev. B, vol. 86, no. 9, p. 094425, Sep. 2012.
[35] S. Petit et al., "Influence of spin‐transfer torque on thermally activated ferromagnetic resonance
excitations in magnetic tunnel junctions," Phys. Rev. B, vol. 78, no. 18, p. 184420, Nov. 2008.
[36] Z. Zeng et al., "Ultralow‐current‐density and bias‐field‐free spin‐transfer nano‐oscillator," Scientific
Reports, vol. 3, p. srep01426, Mar. 2013.
[37] "Large Emission Power over 2 µW with High Q Factor Obtained from Nanocontact Magnetic‐Tunnel‐
Junction‐Based Spin Torque Oscillator," Appl. Phys. Express, vol. 6, no. 11, p. 113005, Oct. 2013.
[38] L. Yang et al., "Reduction of phase noise in nanowire spin orbit torque oscillators," Sci Rep, vol. 5,
Nov. 2015.
[24] Y. Zhou, C. L. Zha, S. Bonetti, J. Persson, and J. Åkerman, "Spin‐torque oscillator with tilted fixed
layer magnetization," Applied Physics Letters, vol. 92, no. 26, p. 262508, Jun. 2008.
[25] B. Ramaswamy et al., "Wireless current sensing by near field induction from a spin transfer torque
nano‐oscillator," Applied Physics Letters, vol. 108, no. 24, p. 242403, 2016.
[26] A. Slavin and V. Tiberkevich, "Nonlinear Auto‐Oscillator Theory of Microwave Generation by Spin‐
Polarized Current," IEEE Transactions on Magnetics, vol. 45, no. 4, pp. 1875–1918, Apr. 2009.
[27] D. H. Edwards, W. J. Heitler, and F. B. Krasne, "Fifty years of a command neuron: the neurobiology
of escape behavior in the crayfish," Trends in Neurosciences, vol. 22, no. 4, pp. 153–161, Apr. 1999.
[28] J. Herberholz and G. D. Marquart, "Decision Making and Behavioral Choice during Predator
Avoidance," Front Neurosci, vol. 6, Aug. 2012.
[29] J.J. Wine and F.B. Krasne, "The Cellular Organization of Crayfish Escape Behavior," The Biology of
Crustacea, vol. 4, pp. 241–292.
[39] Z. Duan et al., "Nanowire spin torque oscillator driven by spin orbit torques," Nature
Communications, vol. 5, p. ncomms6616, Dec. 2014.
[40] "Mutual Phase‐Locking of Microwave Spin Torque Nano‐Oscillators ‐ IEEE Conference Publication."
[Online]. Available: http://ieeexplore.ieee.org/abstract/document/4261436/. [Accessed: 07‐Sep‐
2017].
[41] S. Tamaru, H. Kubota, K. Yakushiji, S. Yuasa, and A. Fukushima, "Extremely Coherent Microwave
Emission from Spin Torque Oscillator Stabilized by Phase Locked Loop," Sci Rep, vol. 5, Dec. 2015.
[42] S. Kerzenmacher, J. Ducrée, R. Zengerle, and F. von Stetten, "Energy harvesting by implantable
abiotically catalyzed glucose fuel cells," Journal of Power Sources, vol. 182, no. 1, pp. 1–17, Jul. 2008.
[43] "Magnetic targeting of nanoparticles across the intact blood–brain barrier ‐ ScienceDirect." [Online].
Available: http://www.sciencedirect.com/science/article/pii/S0168365912007092. [Accessed: 12‐
Jun‐2017].
[44] R. Sensenig, Y. Sapir, C. MacDonald, S. Cohen, and B. Polyak, "Magnetic nanoparticle‐based
approaches to locally target therapy and enhance tissue regeneration in vivo," Nanomedicine, vol. 7,
no. 9, pp. 1425–1442, Sep. 2012.
Figure 1. (a) Schematic of the nanopillar spin-torque nano-oscillator device. The numbers in parentheses
are the layer thicknesses in nanometers (b) Picture of experimental setup used for crayfish neuron
stimulation and recording. (c) A schematic of the circuit to trigger the spin-torque nano-oscillators with
action potentials from crayfish neuron.
Figure 2. The power spectral density of the microwave signal measured from the spin-torque nano-
oscillator for different input direct voltages with an in-plane magnetic field of 10 mT along the minor
axis.
Figure 3. (a) Action potential recorded from crayfish neurons using silver electrodes (black) and the
corresponding microwave power measured from the spin-torque nano-oscillator (red). (b) Microwave
power from the spin-torque nano-oscillator when excited with an artificial action potential (inbox blue
curve) of different amplitudes.
|
1806.03930 | 1 | 1806 | 2018-06-11T12:00:41 | High Efficiency and Low Distortion Photoacoustic Effect in 3D Graphene Sponge | [
"physics.app-ph",
"cond-mat.mes-hall"
] | The conversion of light in sound plays a crucial role in spectroscopy, applied physics, and technology. In this paper, light sound conversion in 3D graphene sponge through a photothermoacoustic mechanism is reported. It is shown that the unique combination of mechanical, optical, and thermodynamic properties of graphene assembled in a 3D sponge structure allows an unprecedented high efficiency conversion independent of light wavelength from infrared to ultraviolet. As a first application of this effect, a photothermal based graphene sponge loudspeaker is demonstrated, providing a full digital operation for frequencies from acoustic to ultrasound. The present results suggest a new pathway for light generation and control of sound and ultrasound signals potentially usable in a variety of new technological applications from high fidelity loudspeaker and radiation detectors to medical devices. | physics.app-ph | physics | Light and Sound in three-Dimensional Graphene Sponge
Flavio Giorgianni,1, 2 Carlo Vicario,2 Mostafa Shalaby,2 Lorenzo Donato Tenuzzo,1 Augusto
Marcelli,3, 4, 5 Tengfei Zhang,6 Kai Zhao,6 Yongsheng Chen*,6 Christoph Hauri,2 and Stefano Lupi*1
1Department of Physics, University of Rome La Sapienza, P.le A. Moro 2, 00185, Rome, Italy
2Paul Scherrer Institute, SwissFEL, 5232 Villigen-PSI, Switzerland
3INFN-LNF, via E. Fermi 40, 00044 Frascati, Italy
4RICMASS, Rome International Center for Materials Science Superstripes, Via dei Sabelli 119A, 00185 Rome, Italy
5CNR - Istituto Struttura della Materia and Elettra-Sincrotrone Trieste,
Basovizza Area Science Park 34149 Trieste, Italy
6State Key Laboratory and Institute of Elemento-Organic Chemistry,
Collaborative Innovation Center of Chemical Science and Engineering (Tianjin),
Key Laboratory of Functional Polymer Materials and the Centre of Nanoscale Science and Technology,
Institute of Polymer Chemistry, College of Chemistry, Nankai University, Tianjin, 300071, China
(Dated: June 13, 2018)
Light modulation plays a key role in modern data transfer technologies providing many advan-
tages, like low attenuation, large bandwidth and electric noise reduction.
In three-dimensional
(3D) graphene sponge, we show that intensity modulated light can be transduced in acoustic waves
through a highly efficient photo thermal-acoustic mechanism. As a first application of this effect,
which is independent of light wavelength from infrared to ultraviolet, we demonstrate a photo-
thermal based 3D graphene loudspeaker, permitting a full digital operation for frequencies from
acoustic to ultrasound. The present results suggest a new pathway for light generation and control
of sound and ultrasound signals potentially usable in a variety of new technological applications
from high-fidelity loudspeaker and radiation detectors to medical devices.
PACS numbers: Valid PACS appear here
INTRODUCTION
Sound generation has been explored for millennia for
communications, enjoyment and cultural reasons. Clas-
sical examples being drumheads and whistles for rel-
atively long-distance exchange of information, and all
kinds of musical instruments for religious and entertaine-
ment activities [1]. In contemporary society, sound gen-
eration and recording is even more important and effi-
cient small-scale audio transduction systems like portable
loudspeakers or wireless communication devices repre-
sent a cutting-edge technology for daily life.
For an efficient human audibility, an ideal speaker
should generate a uniform sound pressure level (SPL)
in the range 20 Hz-20 kHz. Most of
loudspeakers
commercially avalaible today, are based on a thin
membrane connected with a voice coil nested in a
permanent magnet. When alternating current passes
through the coil, mechanical oscillations of the mem-
brane are produced, leading adiabatic compressions and
expansions of the surrounding air, i.e. sound. However,
the electro-mechanical resonances underlying the sound
generation have an inherently narrow frequency response
[2]. This represents the major limitation for high fidelity
reproduction of a real sound. As a matter of fact,
it is not possible to cover with a single mechanical
loudspeaker a wide spectrum from acoustic region to
ultrasound.
A first non-mechanical sound emission mechanism has
been discovered in 1880 by A. G. Bell [3–6]. Bell found
that when a pulsed light beam shines both solids, liq-
uids, and gases, an audible sound is generated. His work
generated a flurry of interest until recent years where
photoacustic emission is used as a powerful spectroscopic
technique in condensed matter physics [7].
Another sound emission mechanism has been investi-
gated by Arnold and Crandall [8]. Here, sound is pro-
duced through a conversion of heat energy in pressure
waves and is called thermo-acoustic effect. More specifi-
cally, when an alternating current passes through a ma-
terial a time dependent Joule heat is generated and then
converted in sound which may extend beyond the acous-
tic region.
Although thermoacustic and photoacustic devices and
their combination were conceived a long time ago, only
recently the application of nanofabrication techniques
allowed to produce acoustic devices based on ultrathin
conductive structures [9–12], which permit to increase
their conversion efficiency. However, the discovery of
new, highly performant, photo-thermo materials could
overcome the actual technical
limits representing a
cutting-edge technology for daily life.
Graphene, the two dimensional structure of carbon,
reaches several records in term of mechanical strength
[13], electron mobility [14], and thermal conductivity
[15]. For these remarkable properties this material has
been the subject of fundamental researches and it has
used for new photonics and plasmonics devices [16, 17].
8
1
0
2
n
u
J
1
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
0
3
9
3
0
.
6
0
8
1
:
v
i
X
r
a
light because it shows:
Graphene is the ideal material for the photo thermo-
acoustic conversion of
i) A
low heat capacity necessary to achieve large thermal
gradients;
ii) A high thermal conductivity to deliver
rapidly heat to the surrounding gas; iii) An absorption
coefficient, due to Dirac-particle-like electronic states
[22], practically indipendent of radiation wavelength
from terahertz (THz) to ultraviolet (UV). Those un-
precedented combination of optical, mechanical and
thermal properties candidates graphene as one of the
most useful material to transduce light in acoustic waves.
The first graphene thermo-acoustic device was based
on a single layer graphene transferred on a substrate
to achieve a good mechanical stability [18].
In this
case, however, the major part of heat relaxed from
the absorbed light, is dissipated in the substrate thus
reducing the conversion efficiency [18]. This leakage
effect can be reduced by using patterned substrate or
nano-struttured graphene systems [19, 20]. Beside these
constraints, single layer graphene based thermo-acoustic
devices are able to operate in a wide frequency range up
to tens of kHz.
Recently,
it has been observed that the properties of
2-dimensional graphene can be extended on a macro-
scopic scale arranging individual graphene sheets in
3-dimensional (3D) monolithic structures [21].
In par-
ticular, graphene 3D sponges (G-sponge), while keeping
the peculiar characteristics of single-layer graphene, add
further interesting properties like an extremely lightness
associated to a robust mechanical strength, and a highly
repeatable compression and complete volume recovery
in a wide temperature range both in air and in liquid
without a substantial structure degradation [23, 24].
In this paper, we report on the first light-driven
loudspeaker based on photo thermo-acoustic effect in
graphene-sponges. We found that a G-sponge sample
is a highly efficient magnet-free, metal-free and con-
tactless loudspeaker driven by light. In particular, our
experiment reveals sound emission linearly dependent
on light intensity and independent of light wavelength
in a broad spectral range from infrared (IR) to UV.
The graphene-sponge light-driven loudspeaker covers an
acoustic range from 100 Hz to 20 kHz and, ultimately, it
plays music from modulated light.
RESULTS
Acoustic waves in graphene sponge induced by
modulated light.
3D G-sponges were grown by an in situ solvothermal
process of graphene oxide (GO) sheets in ethanol (see
2
methods). These sponges have an average transversal
size of 1.5 cm and a thickness not less than 1 cm (see
Fig.
1b). According to previous works [23, 24], the
2D graphene electronic behavior due to a Dirac-like
band structure is essentially preserved for the individual
graphene sheets forming the G-sponge, while a weak
band gap opens in proximity of the sheet interconnection
regions [23, 24]. The G-sponge samples here investigated
show a flat reflectivity from IR to UV of about 5 %, and
a density of 0.92 kg m−3.
The scheme of the photo-acoustic setup is reported in
Fig. 1a.
Intensity modulated light was generated by
LED sources powered by an electric signal properly
amplified by a driver circuit. Emitted light enters in
an anechoic chamber through an optical lens. This lens
allows to focus the radiation onto the G-sponge sample
with a spot size much smaller than the sponge surface
(see Fig. 1b). The anechoic chamber was used both
to acoustically isolate from environmental noise the
experimental apparatous and to avoid internal sound
reflection interference.
The emitted sound is finally acquired using a calibrated
microphone (Samson-C02) placed behind the G-sponge
at a fixed distance of 2.5 cm in the far-field region.
We investigated the sound emission as a function of
excitation light wavelength covering the range from
IR to UV by means of different nearly-monochromatic
LEDs emitting at 1720 nm, 1050 nm, 780 nm, 280
nm, respectively, and through a broadband white LED
emitting in the visible range from 400 nm to 750 nm.
LED emission spectra are reported in the supplementary
material.
We first discuss the results of acoustic generation in
G-sponge by a sinusoidal modulated optical waveform.
Fig. 1d shows the emitted sound driven by a sinusoidally
modulated white LED at different modulation frequen-
cies having a root-mean-square (rms) optical power of 20
mW. The excitation optical waveform is instead reported
in Fig. 1a at 1 kHz. As one can observe, the acoustic
wave is emitted at the modulating frequency of the opti-
cal excitation and its pressure amplitude increases with
increasing frequency. This behavior is better shown in
Fig. 1e, where the emitted sound pressure level (in dB),
is reported by red dots as a function of the modulation
frequency for an optical power of 20 mW. SPL monotoni-
cally increases versus the modulation frequency showing,
for frequencies above 10 KHz, a saturation behavior.
Fig. 1f-g show the acoustic pressure vs. the optical power
q0 at two different modulation frequency: 1 kHz and 10
kHz. As one can clearly see from those panels, the acous-
tic pressure linearly depends on the input power and it
is practically independent of the excitation wavelengths.
This is also true for a broadband white LED. All these
data indicate the same sound generation efficiency for
wavelengths from 1720 nm (IR) to 280 nm (UV).
The wavelength independent acoustic behavior previ-
ously observed is determined by two different factors: i)
The photo-acoustic effect investigated in this paper is re-
lated to intercone transitions in the Dirac band structure
of graphene, sketched in Fig. 1c. This results in a photon
wavelength independent absorption coefficient [22]; ii) At
these photon wavelengths the dominant electronic cool-
ing channel is the non-radiative electron-phonon scatter-
ing [24]. This implies, approximately, that the whole
optical energy is converted into Joule heating.
For higher photon energies, however, other non-radiative
mechanisms take place, such as the emission of electrons
by Auger-like effect [24] that can affect the efficiency of
photo-thermal conversion.
The increase of lattice temperature due to the electron re-
laxation towards phonons leads to a thermal wave inside
the G-sponge. Due to the low heat capacity of graphene
this thermal energy is rapidly dissipated towards the air
layers adjacent to the graphene interfaces. This gener-
ates a quick expansion of the air layers which act as an
acoustic piston on the rest of surrounding gas column
producing a pressure wave, i.e. a sound wave.
As a result, light driven acoustic emission mechanism
in G-sponge can be described in terms of a combina-
tion of two distinct processes characterized by different
timescales: A photo-thermal and a thermo-acoustic pro-
cess. The energy relaxation in the photo-thermal mecha-
nism is driven mainly by the electron-phonon scattering
and occurs on a fast time scale of tens of picoseconds
[25, 26]. Through this effect, light energy absorbed by
electronic intercone transitions, is mainly transduced in
an out-of-equilibrium phonon populations. The thermal
energy here accumulated is then transfered over a mi-
crosecond scale to the surroinding air. Assuming that
the light energy absorbed is completely converted in ther-
mal heating (see supplementary material for the ther-
mal properties of the G-sponge under illumination), the
acoustic generation can be described through a model
proposed by Hu et al. [28], which calculates the thermo-
acoustic emission from a solid.
αG = 2.25 · 10−6m2s−1 is the graphene-sponge thermal
diffusivity [29] and F is the acoustic frequency, represents
the maximum spatial extension of the thermal waves in
the G-sponge. For the 3D G-sponge investigated in this
paper and for acoustic frequency higher than a few of Hz,
λG is less than the sample thickness. In this limit, the
sound pressure (rms value), according with the model of
Hu et al., can then be expressed as [28]:
The effective thermal length λG = (cid:112)αG/(πF ), where
p =
γ − 1
vg
R0
r0
eg
M eg + eG
(1 − R)I0,
(1)
where eg and γ are the thermal effusivity (i.e. the rate
at which a specific material can exchange heat with
the surrounding environment), and the heat capacity
3
√
ratio of the surrounding gas, respectively. M is a
frequency dependent factor which approaches 1 at high
frequency.
vg is the sound velocity in the gas, and
eG =
κρGCG is the thermal effusivity of the G-sponge.
Here, κ = αGρGCG is the sponge thermal conductivity,
ρG its mass density and CG = 690 Jkg−1K−1 its heat
[29].
capacity whose value has been measured in Ref.
R ∼ 0.05,
is the G-sponge reflectivity, taking into
account reflection loss.
I0 = q0/A is the input light
intensity where q0 the optical input power and A the
illuminated sponge surface.
In Eq.1, R0/r0, where R0 = F A/vg is the Rayleigh
distance, and r0 is the distance between the sound
source and the microphone, provides the right far-field
limit where the experiments have been performed.
As shown in Fig. 1e by a blue dashed line, Eq. 1 quan-
titatively reproduces the sound pressure as a function
of frequency without the use of free parameters. Let us
notice, that the overestimate at high pressure of theory
with respect to experimental data, can be associated
mainly to a reduced efficiency of the microphone.
Moreover, the calculated acoustic pressure has a linearly
dependence on the optical power q0 as experimentally
observed. The slope calculated by Eq. 1 is 0.07 (0.75)
Pa/W at 1 kHz (10 kHz) which is very close to 0.08
(0.81) Pa/W experimentally obtained (see Fig. 1f and
g). The small difference between data and calculation
are related to the incertitudes in the G-sponge thermal
parameters entering in Eq. 1.
Acoustic waves emission from compressed G-
sponge
A crucial parameter of the photo thermo-acoustic effect
is the thermal effusivity e.
Indeed, as observed in Eq.
1, in order to obtain an efficient photo thermo-acoustic
effect the effusivity of a material must be comparable
to that of the surrounding gas.
The effusivity in
conventional bulk conductors is very high thus they
are not suitable for thermo-acoustic applications. This
limitation has been partially overcame by fabricating
thin and nanostructured materials in which the mass
density and the heat capacity per unit area (HCPUA)
are strongly reduced [11, 12, 30, 31].
In order to investigate the dependence of the acoustic
emission on the G-sponge morphology, we have applied
a huge compressive strain to the G-sponge sample. SEM
data of an uncompressed and compressed G-sponge over
99% of applied strain are represented in Fig. 2. The as-
grown sponge consists of a frame of interconnected micro-
scale voids, resulting in an open-cell structure. With
an applied compressive strain larger than 99%, the G-
sponge enters in a plastic deformation regime, where the
void space shrinks and the sample becomes denser as
shown in Fig. 2b. Notably, the graphene-microvoid walls
originally randomly distributed, stack in nearly parallel-
aligned arrays perpendicular to the compressive strain
direction. Ultimately, a 99% compression, results in an
increased mass density of nearly a factor 10 in compari-
son with the uncompressed sample.
Fig. 2c shows the sound pressure level (blue points),
versus frequency, generated from a compressed G-sponge
illuminated by a sinusoidally modulated white LED at
20 mW. SPL, as for the uncompressed G-sponge sample,
increases with frequency. However, the compressed
G-sponge shows lower sound pressure (about a factor 2,
blue points) than the uncompressed one (black line), for
the same input optical power. This means a reduction in
the photoacoustic conversion efficiency of about a factor
4. A reduced efficiency is also observed in Fig. 2d,e
which show the acoustic pressure vs the input optical
power at two different frequencies (1 kHz and 10 kHz),
compared with the uncompressed results (colored points
and continous blue line, respectively).
At a very high compressive strain, anisotropy is
induced into the sponge, and a growing alignment of the
graphene sheets perpendicular to the direction of com-
pression occurs (see Fig. 2b). This leads to a final plastic
deformation of the sample determining a long-range
(millimeter) order, keeping intact,
instead, the short
spatial structure (micrometer) of the graphene sponge
network, as confirmed by SEM image in the insets in
Fig. 2a and Fig. 2b.
In particular, the macroscopic
order does not affect the thermal and optical properties
of the individual sheets. Therefore, the reduction of the
sound emission efficiency is mainly due to the increase of
the G-sponge density which reduces the sponge effusivity
eG. This behavior is further confirmed by putting the
density of compressed G-sponge in Eq. 1. The numerical
results, shown through a red curve in Fig.
2c, well
reproduce the experimental data.
Coherent acoustic wave emission
As previously described, a sound wave is generated when
graphene-sponge samples are shined by a modulated
light. The sound power is linearly dependent on the light
intensity and independent of the light wavelength from
infrared to ultraviolet.
In order to investigate the time scale formation of the
sound wave,
in this subsection we study the acoustic
generation as a function of the rise time of a square-wave
optical-excitation. The square wave from a white LED
at different rise times are reported in Fig. 3a by red
lines, while the corresponding sound waves are shown
by colored lines. A narrow acoustic pulse appears
when the optical rise time is less than a few tens of µs.
For these time-scales a broadband pulse is generated
in the frequency domain, covering the whole acoustic
region, as shown in Fig. 3b. An increase of the rise
time determines a temporal broadening of the acoustic
pulses, which reflects in a reduction of their intensity
4
and spectral distribution in the frequency domain.
The acoustic wave emitted, in the same optical condi-
tions, for a rise time less than 1 µs and a repetition rate
frequency of 100 Hz, is shown in Fig. 3c as a function of
time. Each period has two optical edges, corresponding
to the rising and falling of the light wavefront, which
are labeled in Fig. 3c, by a sign (+) and (-), respec-
tively. An expansion pressure wave is then generated
in correspondence of the optical rise time. This is due
to a fast heating of the graphene sponge above room
temperature (the thermal properties of graphene sponge
under illumination have been reported in SI). At the
optical falling time, a fast cooling process towards room
temperature is instead produced inducing a compression
pressure wave (Fig. 3c). The expansion and compression
waves actually present an exact specular shape.
Increasing the repetition rate of the optical square wave
for a fixed optical power, the time dependence of the
acoustic waveform changes significantly (see Fig. 3d).
This effect can be ascribed to a progressive temporal
superposition of the expansion and compression waves.
In particular, the heating-cooling waves are construc-
tively overlapped when the frequency approaches 10 kHz
resulting in an effective increase of the acoustic pressure.
For higher frequencies instead, heating-cooling waves
begin to distructively interact decreasing the emitted
acoustic pressure as observed, for instance, at 15 kHz in
Fig. 3d.
In order to simulate this overlapping effect we per-
formed a numerical calculation. In this simulation (see
Fig. 3e), the acoustic pulses at 100 Hz (see Fig. 3c),
which can be considered as non-interacting, have been
superimposed with alternating signs. By increasing the
modulation frequency up to 15 KHz i.e. by reducing
the period among two consecutive pulses, the simulation
accurately reproduces both the shapes and the intensi-
ties of the measured acoustic waves. This superposition
effect clearly influences the sound spectral distribution
as observed in Fig. 3f. From a broadband spectrum
generated by a single acoustic pulse (at 100 Hz), the
sound spectral distribution evolves in a comb spectrum
of odd harmonics of the modulation frequency (from 1
kHz to 15 kHz). This agrees to the fact that the Fourier
expansion of a square wave contains only odd harmonics.
Light-driven loudspeaker
The first light-driven loudspeaker was proposed in 1983
by W. F. Rush et al which, following the Bell's idea,
generated sound through an intensity modulated laser
beam in a gas [32]. This device was able to overcome the
mechanical limitations of the conventional loudspeaker
achieving high fidelity sound reproduction. However, due
to the very weak gas photoacoustic efficiency and there-
fore the need of a powerful laser system, the spread of
this device was limited.The light-sound transducer based
on G-sponges investigated in this manuscript has instead,
due to its high efficiency and fidelity, a huge potential-
ity for new applications, in particular as loudspeaker and
broadband light detector.
As discussed above, LED sources are very fast but their
current-voltage response shows a weak nonlinear behav-
ior which leads to a small distortion of the light output
waveform. This represents a drawback in the generation
of acoustic waves using the light intensity modulation
scheme. As a result, one generates both the fundamental
modulation frequency and high order harmonics which
may distort consequently the acoustic output (see Fig.
4a).
In order to generate a pure tone and achieving a high fi-
delity in sound reproduction, G-sponge loudspeaker can
be driven by a light pulse density modulation (PDM)
method [33]. In PDM mode, an analog input signal is
encoded in sequential pulse trains with constant height
and a relative temporal pulse density that corresponds
to the analog signal's amplitude. A sinusoidal analog
waveform (5 kHz), is encoded by PDM in a digital signal
which drives the LED (Fig. 4b). The generated light
pulse trains measured by a photodiode, where a single
light pulse has a FWHM of 1 µs, are shown in Fig. 4c.
The acoustic output signal is instead shown in Fig. 4d.
We can see that from a light pulse density modulated
signal the original analog signal is well reproduced.
The generated acoustic wave in frequency domain is re-
ported in Fig. 4e. The first harmonic intensity, which,
in the intensity modulation scheme, is nearly 10 % of the
fundamental one, decreases to about 1 % in the PDM
mode. Moreover, the distortion in the light driven loud-
speaker based on G-sponge is lower than in a commercial
moving-coil loudspeaker as shown in Fig. 4e. Even bet-
ter results can be obtained by reducing the duration of
the single light pulse and increasing the digital sampling
of the analog input waveform.
Fig. 4f, g show the time-frequency plots for linear swept-
frequency sine input signal from 200 Hz to 20 kHz by
using light intensity modulation and PDM mode, respec-
tively. Let us further observe that high order harmonics
are strongly suppressed using the PDM method.
Therefore, G-sponge loudspeaker is absolutely compati-
ble with the PDM sampling technique, permitting a full
digital operation integrable with other electronic devices
and providing an ultra-high sound fidelity with frequen-
cies from acoustic to ultrasound. The capability of a
G-sponge light speaker in reproducing a real audio file is
shown in Supplementary Video 1 and 2. Here, G-sponge
based-speaker is able to play songs by intensity modu-
lated light with a 20 mW rms power generated from a
commercial with LED.
In this work we have presented an unprecedented con-
cept of photo-acoustic speaker taking advantage of the
thermodynamic, mechanical and optical properties of a
free-standing three-dimensional graphene sponge struc-
5
ture. The use of graphene sponges, which have a density
similar to air, provides an unparalleled lightness and me-
chanical stability of the loudspeaker while the absence of
the substrate prevents thermal leakage. The light-driven
acoustic generation allows to emit sound from far dis-
tances without any type of physical connection. More-
over, the heat distribution on the sponges can be fully
controlled by the spatial distribution of the light giving
rise to a scalable device with an acoustic power linearly
dependent on the illuminated surface. The present re-
sults suggest that the light-driven acoustic generation
in graphene sponge could be widely used in a variety
of new technological applications from sound and ultra-
sound production, radiation detectors and medical de-
vices.
AUTHOR CONTRIBUTIONS
Tengfei Zhang, Kai Zhao, Yongsheng Chen fabricated
and characterized 3D graphene sponges. F. Giorgianni,
S. Lupi, A. Marcelli, M. Shalaby, L. D. Tenuzzo, C.
Vicario, carried out the infrared, visible and ultravio-
let experiments. Data analysis has been performed by
F. Giorgianni and L. D. Tenuzzo. F. Giorgianni and S.
Lupi planned and managed the project with inputs from
all the co-authors. F. Giorgianni and S. Lupi wrote the
manuscript. All authors extensively discussed the results.
ADDITIONAL INFORMATION
The authors declare no competing financial interests.
Correspondence and requests for materials should be ad-
dressed to S.L. ([email protected]) and Yong-
sheng Chen ([email protected]).
REFERENCES
[1] Sachs C. "The history of musical instruments." (Courier
Dover Publications, 2006).
[2] Borwick, John "Loudspeaker and headphone handbook."
(CRC Press, 2012).
[3] Bell, Alexander Graham. "On the production and repro-
duction of sound by light." American Journal of Science
118 (1880): 305-324.
[4] Tyndall, John. "Action of an intermittent beam of radi-
ant heat upon gaseous matter." Proceedings of the Royal
Society of London 31.206-211 (1880): 307-317.
[5] Rntgen, Wilhelm C. "On tones produced by the inter-
mittent irradiation of a gas." (1881): 308-311.
[6] Bell A. G. "LXVIII. Upon the production of sound by
radiant energy." Philos. Mag. 11, 510 (1881).
[7] Ball David W. "Photoacoustic Spectroscopy Spec-
troscopy," Volume 21, Issue 9, Sep 1, 2006
6
[8] Arnold, H. D., and I. B. Crandall. "The thermophone as
a precision source of sound." Physical review 10.1 (1917):
22.
[9] Xiao, Lin, et al. "Flexible, stretchable, transparent car-
bon nanotube thin film loudspeakers." Nano letters 8.12
(2008): 4539-4545.
[10] Wei, Yang, et al. "Thermoacoustic chips with carbon
nanotube thin yarn arrays." Nano letters 13.10 (2013):
4795-4801.
[11] Dutta, Rajen, et al. "Gold Nanowire Thermophones."
The Journal of Physical Chemistry C 118.50 (2014):
29101-29107.
[12] Tian, He, et al. "Transparent, flexible, ultrathin sound
source devices using indium tin oxide films." Applied
Physics Letters 99.4 (2011): 043503.
[13] Lee, Changgu, et al. "Measurement of the elastic prop-
erties and intrinsic strength of monolayer graphene." sci-
ence 321.5887 (2008): 385-388.
[14] Neto, AH Castro, et al. "The electronic properties of
graphene." Reviews of modern physics 81.1 (2009): 109.
[15] Balandin, Alexander A., et al. "Superior thermal conduc-
tivity of single-layer graphene." Nano letters 8.3 (2008):
902-907.
[22] Mak, Kin Fai, et al. "Optical spectroscopy of graphene:
from the far infrared to the ultraviolet." Solid State Com-
munications 152.15 (2012): 1341-1349.
[23] Wu, Yingpeng, et al. "Three-dimensionally bonded
spongy graphene material with super compressive elas-
ticity and near-zero Poissons ratio." Nature communica-
tions 6 (2015).
[24] Zhang, Tengfei, et al. "Macroscopic and direct light
propulsion of bulk graphene material." Nature Photonics
9.7 (2015): 471-476.
[25] Wang, Haining, et al. "Ultrafast relaxation dynamics of
hot optical phonons in graphene." Applied Physics Let-
ters 96.8 (2010): 081917.
[26] Gierz, Isabella, et al. "Snapshots of non-equilibrium
Dirac carrier distributions in graphene." Nature mate-
rials 12.12 (2013): 1119-1124.
[27] Tian, Yichao, et al. "Coherent generation of photo-
thermo-acoustic wave from graphene sheets." Scientific
reports 5 (2015).
[28] Hu, Hanping, Tao Zhu, and Jun Xu. "Model for ther-
moacoustic emission from solids." Applied Physics Let-
ters 96.21 (2010): 214101.
[29] Aliev, Ali E., et al. "Alternative Nanostructures for Ther-
[16] Geim, Andre K., and Konstantin S. Novoselov. "The rise
mophones." ACS nano 9.5 (2015): 4743-4756.
of graphene." Nature materials 6.3 (2007): 183-191.
[17] Geim, Andre Konstantin. "Graphene:
status and
prospects." science 324.5934 (2009): 1530-1534.
[18] Tian, He, et al. "Graphene-on-paper sound source de-
vices." ACS nano 5.6 (2011): 4878-4885.
[19] Suk, Ji Won, et al. "Thermoacoustic sound generation
from monolayer graphene for transparent and flexible
sound sources." Advanced Materials 24.47 (2012): 6342-
6347.
[20] Tian, He, et al. "Single-layer graphene sound-emitting
experiments and modeling." Nanoscale 4.7
devices:
(2012): 2272-2277.
[21] Jiang, Lili, and Zhuangjun Fan. "Design of advanced
porous graphene materials: from graphene nanomesh to
3D architectures." Nanoscale 6.4 (2014): 1922-1945.
[30] MLA Niskanen, Antti O., et al. "Suspended metal wire
array as a thermoacoustic sound source." Applied Physics
Letters 95.16 (2009): 163102.
[31] Tian, He, et al. "Flexible, ultrathin, and transparent
sound-emitting devices using silver nanowires film." Ap-
plied Physics Letters 99.25 (2011): 253507.
[32] Rush, William F., James E. Huebler, and Peter Lysenko.
"Photoacoustic speaker and method." U.S. Patent No.
4,641,377. 3 Feb. 1987.
[33] K. Nobuyoshi et al. "Characteristics of thermally induced
acoustic emission from nanoporous silicon device under
full digital operation." Applied Physics Letters 102.12
(2013): 123504.
7
FIG. 1: Sound generation from Graphene Sponge a) Scheme of experimental setup for acoustic generation measurements:
Intensity modulated light at different wavelengths and modulation frequencies is emitted by a LED powered by a driver circuit
coupled with a waveform generator. The light then enters through a lens inside the anechoic chamber irradiating the G-sponge.
b) View of the G-sponge sound emitter. c) Linear (Dirac) electronic band dispersion in graphene: The arrows represent the
inter-band transitions under an optical excitation. The electrons involved relax by electron-phonon interaction inducing a
lattice heating on fast (ps) time-scale. The fast-rise thermal gradient in the G-sponge network generates acoustic waves. d)
Acoustic emission in time domain at different frequencies of a sinusoidally modulated white light with a rms power of 20 mW
(vertical offset added). e) The generated sound pressure level as a function frequency with at 20 mW input power of sinusoidal
light. Red dots represent the experimental data and the blue solid line corresponds to a model of thermo-acoustic emission,
which is described in the main text. f)-g) Generated acoustic pressure vs rms optical power at 1 kHz and 10 kHz, respectively.
Dots with different colors indicate different wavelengths of excitation light: 1710 nm, 1050 nm, 780 nm, 280 nm, and white
light. Dashed line is a linear fit. The clear linear dependence of the acoustic pressure vs. the optical power is independent of
the light wavelength across the visible range.
8
FIG. 2: Effect of compressive strain on the sound generation efficiency in G-sponge. SEM images: a) G-sponge
surface b) G-sponge surface parallel to the compressive direction for a strain over 99 % (plastic regime). Insets in a) and b)
show magnified SEM images for G-sponge and compressed G-sponge, respectively. At short-range the network morphology and
the physical parameters of graphene sheets remain approximatively unchanged. c) Generated SPL from compressed G-sponge
by a sinusoidal optical waveform with 20 mW of rms power: Blue dots are experimental data, red solid line results from
the thermo-acoustic model for compressed G-sponge (see main text), while red solid line corresponds to the thermo-acoustic
emission for the uncompressed G-sponge. d)-e) Generated acoustic pressure vs. the rms optical power for the compressed
G-sponge at 1 kHz and 10 kHz, respectively. Dots with different colors indicate different wavelengths of excitation light: 780
nm, 280 nm, white line. Dashed line is a linear fit. As a result, the conversion efficiency decreases with respect to uncompressed
G-sponge (blue line).
FIG. 3: Broadband acoustic generation from optical square-wave in G-sponge: time and frequency domain. a)
Temporal profile of the acoustic pulse generate in G-sponge as a function of rise time of an optical square-wave which is shown
by a red curve (vertical offset added). b) Related acoustic spectra for different rise time. c) Generated acoustic wave (blue
curve) by an optical square-wave (red curve) with 100 Hz modulating frequency. d) Time domain signal of G-sponge sound at
different modulation frequencies of the optical square wave. A vertical offset has been applied to the curves for a clearer view.
e) Numerical simulation of the temporal overlapped acoustic waves. f) Evolution of the emission spectrum vs. the amodulation
frequency. All measurements were performed with a white LED optical waveform with a rms power of 20 mW.
9
FIG. 4: Light Pulse Density Modulation: a) Sinusoidally intensity modulated light in frequency domain emitted from a
white LED. Modulation frequency is 5 kHz. The LED response induces a distortion of emitted light waveform and consequently
a first harmonic is observed at 10 KHz. b) Waveform of an analog input signal (5 kHz). c) The corresponding light pulse trains
from PDM-digitized analog signal. d) Measured acoustic output signal. This full digital operation mode very well reproduces
the analog input signal. e) Acoustic wave emitted in the G-sponge through the intensity modulation mode (red curve), PDM
mode (blue curve). The corresponding acoustic wave emitted from a commercial moving-coil loudspeaker at 5 kHz (green
curve).
f) Time-frequency plot of generated acoustic waves of linear swept-frequency input signal by intensity modulation
mode (see main manuscript) g) Corresponding generation acoustic waves by PDM mode. As one can observe the harmonic
distortion is strongly suppressed.
|
1811.04814 | 2 | 1811 | 2019-01-15T15:12:43 | Valley based splitting of topologically protected helical waves in elastic plates | [
"physics.app-ph"
] | Topological protection offers unprecedented opportunities for wave manipulation and energy transport in various fields of physics, including elasticity, acoustics, quantum mechanics and electromagnetism. Distinct classes of topological waves have been investigated by establishing analogues with the quantum, spin and valley Hall effects. We here propose and experimentally demonstrate the possibility of supporting multiple classes of topological modes within a single platform. Starting from a patterned elastic plate featuring a double Dirac cone, we create distinct topological interfaces by lifting such degeneracy through selective breaking of symmetries across the thickness and in the plane of the plate. We observe the propagation of a new class of heterogeneous helical-valley edge waves capable of isolating modes on the basis of their distinct polarization. Our results show the onset of wave splitting resulting from the interaction of multiple topological equal-frequency wave modes, which may have significance in applications involving elastic beam-splitters, switches, and filters. | physics.app-ph | physics | Valley based splitting of topologically protected helical waves in elastic plates
M. Miniaci∗,1, 2 R. K. Pal,1 R. Manna,1 and M. Ruzzene1, 3
1)School of Aerospace Engineering, Georgia Institute of Technology,
GA 30332 Atlanta, USA
2)EMPA, Laboratory of Acoustics and Noise Control, Uberlandstrasse 129,
8600 Dubendorf, Switzerland
3)School of Mechanical Engineering, Georgia Institute of Technology,
GA 30332 Atlanta, USAa)
(Dated: 16 January 2019)
9
1
0
2
n
a
J
5
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
2
v
4
1
8
4
0
.
1
1
8
1
:
v
i
X
r
a
a)Electronic mail: [email protected], [email protected]
1
Topological protection1 offers unprecedented opportunities for wave manipu-
lation and energy transport in various fields of physics, including elasticity2 -- 5,
acoustics6,7, quantum mechanics8 and electromagnetism9 -- 11. Distinct classes of
topological waves have been investigated by establishing analogues with the
quantum12, spin13 and valley Hall14 effects. We here propose and experimentally
demonstrate the possibility of supporting multiple classes of topological modes
within a single platform. Starting from a patterned elastic plate featuring a
double Dirac cone, we create distinct topological interfaces by lifting such de-
generacy through selective breaking of symmetries across the thickness and in
the plane of the plate. We observe the propagation of a new class of heteroge-
neous helical-valley edge waves capable of isolating modes on the basis of their
distinct polarization. Our results show the onset of wave splitting resulting from
the interaction of multiple topological equal-frequency wave modes, which may
have significance in applications involving elastic beam-splitters, switches, and
filters.
Interfaces between distinct topological phases of matter support exotic localized wave
modes that allow defect-immune, lossless energy transport3,4. Distinct classes of topological
phases exist depending on the dimension and the symmetries associated with different in-
terface modes11. Examples in two dimensions include analogues of the quantum Hall, spin
Hall and valley Hall effects, supporting chiral, helical and valley modes, respectively4,15 -- 23.
While chiral modes require the breaking of time-reversal symmetry, helical and valley modes
involve solely passive components and arise from the breaking of geometrical symmetries in
lattices whose reciprocal space is characterized by singularities such as double Dirac cones24
and Weyl points25. Recent studies have indicated that novel physical phenomena may arise
from the interaction of distinct classes of topological modes26,27. Indeed, while structures
supporting chiral, helical and valley modes separately have been broadly investigated, the
implementation of a single platform supporting multiple classes of such modes has not been
illustrated yet. This is a particularly challenging task for mechanical substrates as in the
case of elastic plates, due to the presence of multiple guided wave modes and their tendency
to hybridize at interfaces and free boundaries28. We here report on an elastic plate capable
of hosting purely helical and heterogeneous helical-valley modes. Numerical models and
experimental implementations investigate the interaction of helical edge waves at interfaces
2
between configurations that are topologically distinct. Through this platform, we demon-
strate the ability to split equal-frequency helical edge waves differing on the basis of their
polarization when they impinge on distinct interfaces at a common junction.
We consider an elastic plate patterned with a periodic array of through-the-thickness
circular and triangular holes4, as shown in Fig. 1a. The plate is periodic along the directions
defined by the a1 and a2 vectors. Its band structure exhibits an isolated double Dirac cone
at the K point as illustrated by the dispersion curves denoted by the red circles in Fig. 1d.
The Dirac cones arise as a result of the D3h symmetry of the structure, i.e. consisting of
C3 (three fold rotational) symmetry, σh symmetry (or reflection symmetry about the mid-
plane of the plate), and σv symmetry (or inversion symmetry about a plane normal to the
mid-plane of the plate and along the lattice vectors).
Starting from this configuration, geometric perturbations are introduced so to break the
σh and σv symmetries, and produce nontrivial bandgaps that respectively support helical
and valley modes in a common frequency range. Specifically, we break the σh symmetry
by replacing the through holes with blind holes of height h, as shown in Fig. 1b. We
denote the configuration with the blind holes on the top (bottom) surface as H + (H−).
This geometric perturbation causes modes spanning the two Dirac cones to hybridize in
analogy with the spin-orbital coupling interaction in QSHE, which breaks the degeneracy
and opens a topological bandgap (Fig. 1e). The interface between H + and H−, here denoted
as I(H +, H−), separates phases that are inverted (σh-transformed) copies of each other, and
supports two helical edge modes spanning the gap with positive (Φ+) and negative (Φ−)
group velocity, respectively (Fig. 2a).
Next, we break the σv while preserving C3 and σh symmetries, by considering holes in
each unit cell of different radii, namely r and R. This leads to two distinct phases, denoted
as V r and V R (Fig. 1c). Contrary to the previous case, an interface that separates two σv-
transformed copies of the structure supports a single valley mode, with positive or negative
group velocity, depending on the type of interface, namely I(V r, V r) or I(V R, V R) with
two adjacent holes of diameter r or R, respectively. The existence of these edge modes
is a consequence of the bulk-boundary correspondence principle27 and can be predicted by
computing the valley Chern numbers. Although the total Chern number is zero in each
band, the Chern number computed around the K and K(cid:48) points will have non-zero values29.
Based on these assumptions, we can infer that an interface between structures supporting
3
helical and valley modes will still support a single hybrid edge mode, named helical-valley
(HV) mode hereafter, with either positive (Ψ+), as in the case of I(H +, V R) (Fig. 2c), or
negative (Ψ−) group velocity for I(H−, V r) (Fig. 2e).
The existence of the above mentioned hybrid HV edge modes is verified through the
computation of the band structure of finite strips including a total of 20 × 1 unit cells,
with periodicity conditions imposed along the a1 direction and free boundaries along a2 (see
Methods for details on computations). Results are reported in Figs. 2a, 2c and 2e, where
the bulk modes are shaded in gray, while the edge states are denoted by the black, blue
and red circles for the I(H +, H−), I(H +, V R) and I(H−, V r), respectively. The additional
notation of the indexes +/− for the modes keeps track of their different group velocity with
respect to the direction of propagation. As noted above, two edge modes Φ+,− are supported
by the I(H +, H−) interface, while a single mode exist at the HV interfaces: Ψ+ and Ψ− for
I(H +, V R) and I(H−, V r), respectively.
A domain wall formed according to each of the three interfaces considered, i.e. I(H +, H−),
I(H +, V R) and I(H−, V r), separates two phases in the middle of the strip. Let us consider
the two edge waves that initially propagate along the I(H +, H−) interface and subsequently
encounter two I(H, V ) interfaces, each supporting a single HV mode with distinct polar-
ization. At the y-junction, each wave follows the interface that matches its polarization,
thus causing the two wave modes to split. This is possible under the condition that the
frequencies for the 3 edge states match. To ensure this, the bandgaps of the H and V
lattices are designed to occur in a common range of frequencies highlighted by the gray
rectangles in Figs. 1e,f which is achieved by properly selecting the geometric perturbations
that produce the distinct topological phases. Specifically, the results for the H phase (Fig.
1e) correspond to a blind hole depth h = 0.91H, with H denoting the plate thickness, while
the V phase results (Fig. 1f) are obtained for r = 0.51R.
Numerical evaluations of the mode shapes at point C (Fig. 2c) and D (Fig. 2e), shown
in Figs. 2d,f respectively for the two interfaces, confirm the localized nature of the modes
and reveal their distinct distribution of the displacement magnitude and phase along the
interface (see the zoomed-in plots). The magnitude displacement and phase plots suggest
the possibility of selective modal excitation by applying an input at the locations shown in
the figures, which highlight the spatial separation of the maximum amplitude points for the
two modes. For example, preferential excitation of mode Ψ+ (Ψ−) could be achieved by
4
injecting a perturbation at locations where the motion of the interface unit cells is high, and
where the displacement for the other mode Ψ− (Ψ+) is small (see zoomed-in views in Figs.
2d,f).
To confirm the splitting of the topologically protected helical waves, we designed and
fabricated a waveguide made of 35 (in the a1 direction) × 25 (in the a2 direction) unit
cells hosting the three different domains H + (green boundary), H− (blue boundary) and
V (red boundary), as shown in Fig. 3a. These domains are separated by three interfaces:
I(H +, H−), I(H +, V R) and I(H−, V r). Such an arrangement is chosen to illustrate the
ability of the waveguide to split the two helical waves (Φ+ and Φ−) at the y-shaped junction.
The plate is made of aluminum and the unit cell lattice parameter is a = 20.5 mm.
First, numerical simulations are conducted to evaluate the distinct propagation patterns
followed by the edge modes along the interfaces, depending on the selective mode excita-
tion. A Finite Element (FE) model for the finite plate shown in Fig. 3a is implemented in
ABAQUS. Calculations are conducted in the frequency domain. Elastic waves are excited
by imposing an out-of-plane harmonic excitation at half of the I(H +, H−) interface (white
dot in Figs. 3c and 3d) according to the Ψ+ and Ψ− configurations presented in Figs. 2d
and 2f. The frequency content of the excitation is set to 98 kHz, so to prevent the excitation
of bulk modes. The resulting distribution of the von Mises stress fields, reported in Figs.
3c,d, clearly show that when the wave reaches the y-shaped junction it follows either the
I(H +, V R) or I(H−, V r) interface depending on the initial type of input. In both cases,
weak penetration inside the bulk region is observed. Refer to SM30 for additional transient
dynamic simulations.
The splitting of these modes is then demonstrated experimentally testing the plate
(Fig. 3a) by means of a Scanning laser Doppler Vibrometer (SLDV). The SLDV measures
the out-of-plane velocity component of the motion of the plate surface produced by a surface
bonded piezoelectric transducer, measuring 12 mm in diameter. The excitation is applied
along the I(H +, H−) interface at the location denoted by the yellow dot in Fig. 3b, and
consists of a 51-cycle sine burst modulated by a Hanning window. The center frequency is
98 kHz which falls inside the bulk bandgap and excites both Φ+ and Φ− waves4. First, one
dimensional (1D) line scans of a spatial step of 0.2 mm are conducted along the interfaces
I(H +, V R) and I(H−, V r) (the locations of line scan measurements are shown as dotted
black lines in Fig. 3b). A temporal window of 800 µs is applied to the recorded signals to
5
eliminate reflections from the plate edges. Next, the recorded signals are represented in the
frequency/wavenumber domain by performing a temporal/spatial Fourier transform (2D-
FT), whose magnitude is superimposed in Figs. 4a to the numerical dispersion predictions
(white square dots) for the I(H +, V R) and I(H−, V r) finite strips. The 2D-FT magnitude
colormaps clearly confirm the numerical edge state predictions along the two interfaces, and
show how the two energy spots are associated to different modes in the dispersion diagrams,
confirming the wave splitting.
To fully unveil the distinct nature of the Ψ+ and Ψ− modes, two fine scans are conducted
over the two 2 × 2 unit cell areas shown in red in Fig. 3b. The velocity distributions at
specific representative time instants t = 842 µs and t = 980 µs are shown in Fig. 4b,
where 1 V in the colorbar corresponds to a velocity of 20 mm/s. The two modes feature
opposite spins (clockwise/anti-clockwise) of the velocity field across the interface, which is
highlighted by the black arrows drawn on the basis of the phase evolution of the measured
wavefield. These representations provide further evidence of mode splitting occurring at
the y junction. Clear visualization of the opposite spins of the two modes along the two
interfaces is obtained from the measurements time animations provided in the SM30.
Finally, the 2D wavefield recorded over the region highlighted by the blue dots, and labeled
as "2D scan region" in Fig. 3b, illustrates the I(H +, H−) interface bounded propagation
along with the splitting occurring at the y-junction (Fig. 4f). Specifically, the measured
out-of-plane velocity distribution at an instant of time after the wave splitting, i.e.
for
t = 1120 µs from the excitation, is reported. The wavefield at the considered instant of
time is then represented in the wavenumber domain by performing a spatial/spatial 2D-FT,
which effectively illustrates the modal content of the wavefield in the reciprocal space kx, ky.
The results of this analysis shown in the 2D-FT amplitude contours of Fig. 4c, illustrates
the presence of 2 pairs of diffraction peaks, each corresponding to two distinct modes that
co-exist at the excitation frequency, and are characterized by two distinct wavenumbers,
k1 = 60 rad/m and k2 = 80 rad/m. These wavenumbers are highlighted by the red and
black circles of different radii in Fig. 4c, and correspond to the two wavenumber values
associated with modes Ψ+ and Ψ−, respectively. The contribution of the two modes to the
wavefield of Fig. 4f can be effectively separated through wavenumber filtering in reciprocal
space31. To this end, the 2D-FT for the wavefield is masked by 2D Gaussian windows (see
Methods) centered at k1 and k2, whose application leads to the filtered 2D-FT in Figs. 4d,e
6
showing the two separated modes. Inverse 2D-FT transformation in physical space provides
the decoupled contributions to the wavefield shown in Figs. 4g,h. From these figures it clearly
emerges that when the two rightward-propagating helical modes Φ+ and Φ− (Fig. 2a) reach
the y-shaped junction, they split and respectively follow the I(H +, V R) and I(H−, V r)
interface as Ψ+ and Ψ− modes on the basis of their polarization.
In conclusion, for the first time we proposed and experimentally tested a platform that
supports multiple classes of topological modes. In the proposed configuration, implemented
on a patterned plate, topologically non-trivial gaps are obtained by creating interfaces be-
tween material phases that selectively break spatial inversion symmetries. Through engi-
neering of the nontrivial gaps, the considered system is capable of splitting purely topological
protected helical edge waves into heterogeneous helical-valley modes on the basis of the ini-
tial polarization. The results presented herein, both numerical and experimental, provide
fundamental insights in the behavior of topologically protected edge modes in elastic sys-
tems, and suggest new avenues for topologically protected wave transmission that may be
extended to other physical domains, such as acoustics, and photonics. The findings of this
study have direct implications for applications where selective waveguiding, or the isolation
and control of vibrations are ultimate goals, as in civil, mechanical and aerospace engi-
neering structures. Also, the wave mode selective capabilities of the considered interfaces
and y-junction may be of significance for the transmission of information through elastic or
acoustic waves as, for example, in the case of surface acoustic wave (SAW) devices.
Acknowledgments
M.M. has received funding from the European Union's Horizon 2020 research and inno-
vation programme under the Marie Sk(cid:32)lodowska-Curie grant agreement N. 754364. R.K.P.
and M.R. acknowledge the support of the EFRI Award 1741685 from the National Science
Foundation.
REFERENCES
1M. Z. Hasan and C. L. Kane, "Colloquium: topological insulators," Rev. Mod. Phys. 82,
3045 (2010).
7
2R. Susstrunk and S. D. Huber, "Observation of phononic helical edge states in a mechanical
topological insulator," Science 349, 47 -- 50 (2015).
3S. H. Mousavi, A. B. Khanikaev, and Z. Wang, "Topologically protected elastic waves in
phononic metamaterials," Nat. Comm. 6, 8682 (2015).
4M. Miniaci, R. K. Pal, B. Morvan, and M. Ruzzene, "Experimental observation of topo-
logically protected helical edge modes in patterned elastic plates," Phys. Rev. X 8, 031074
(2018).
5R. Susstrunk and S. D. Huber, "Classification of topological phonons in linear mechanical
metamaterials," Proc. Natl. Acad. Sci. USA 113, E4767 -- E4775 (2016).
6Z. Yang, F. Gao, X. Shi, X. Lin, Z. Gao, Y. Chong, and B. Zhang, "Topological acoustics,"
Phys. Rev. Lett. 114, 114301 (2015).
7S. Wang, G. Ma, and C. T. Chan, "Topological transport of sound mediated by spin-
redirection geometric phase," Science Advances 4 (2018).
8C.-K. Chiu, J. C. Y. Teo, A. P. Schnyder, and S. Ryu, "Classification of topological
quantum matter with symmetries," Rev. Mod. Phys. 88, 035005 (2016).
9A. B. Khanikaev, S. H. Mousavi, W.-K. Tse, M. Kargarian, A. H. MacDonald,
and
G. Shvets, "Photonic topological insulators," Nat. Mat. 12, 233 (2013).
10L. Lu, J. D. Joannopoulos, and M. Soljaci´c, "Topological photonics," Nat. Phot. 8, 821
(2014).
11T. Ozawa, H. M. Price, A. Amo, N. Goldman, M. Hafezi, L. Lu, M. Rechtsman, D. Schus-
ter, J. Simon, O. Zilberberg, and I. Carusotto, "Topological photonics," arXiv:1802.04173
(2018).
12L. M. Nash, D. Kleckner, A. Read, V. Vitelli, A. M. Turner, and W. T. Irvine, "Topological
mechanics of gyroscopic metamaterials," Proc. Natl. Acad. Sci. USA 112, 14495 -- 14500
(2015).
13C. He, X. Ni, H. Ge, X.-C. Sun, Y.-B. Chen, M.-H. Lu, X.-P. Liu, and Y.-F. Chen,
"Acoustic topological insulator and robust one-way sound transport," Nat. Phys. 12, 1124
(2016).
14J. Lu, C. Qiu, L. Ye, X. Fan, M. Ke, F. Zhang, and Z. Liu, "Observation of topological
valley transport of sound in sonic crystals," Nat. Phys. 13, 369 (2017).
15R. 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).
8
16A. Souslov, B. C. van Zuiden, D. Bartolo, and V. Vitelli, "Topological sound in active-
liquid metamaterials," Nat. Phys. 13, 1091 (2017).
17S. D. Huber, "Topological mechanics," Nat. Phys. 12, 621 (2016).
18J. Ningyuan, C. Owens, A. Sommer, D. Schuster, and J. Simon, "Time-and site-resolved
dynamics in a topological circuit," Phys. Rev. X 5, 021031 (2015).
19R. K. Pal and M. Ruzzene, "Edge waves in plates with resonators: an elastic analogue of
the quantum valley hall effect," New J. Phys. 19, 025001 (2017).
20R. K. Pal, M. Schaeffer, and M. Ruzzene, "Helical edge states and topological phase
transitions in phononic systems using bi-layered lattices," J. Appl. Phys. 119, 084305
(2016).
21C. Brendel, V. Peano, O. J. Painter, and F. Marquardt, "Pseudomagnetic fields for sound
at the nanoscale," Proc. Natl. Acad. Sci. USA 114, E3390 -- E3395 (2017).
22F. Li, X. Huang, J. Lu, J. Ma, and Z. Liu, "Weyl points and fermi arcs in a chiral phononic
crystal," Nat. Phys. 14, 30 -- 34 (2017).
23H. Zhu, T.-W. Liu, and F. Semperlotti, "Design and experimental observation of valley-
hall edge states in diatomic-graphene-like elastic waveguides," Physical Review B 97,
174301 (2018).
24K. Sakoda, "Double dirac cones in triangular-lattice metamaterials," Optics express 20,
9925 -- 9939 (2012).
25L. Lu, Z. Wang, D. Ye, L. Ran, L. Fu, J. D. Joannopoulos, and M. Soljaci´c, "Experimental
observation of weyl points," Science 349, 622 -- 624 (2015).
26X. Ni, D. Purtseladze, D. A. Smirnova, A. Slobozhanyuk, A. Al`u, and A. B. Khanikaev,
"Spin- and valley-polarized one-way klein tunneling in photonic topological insulators,"
Science Advances 4 (2018).
27K. Lai, Y. Yu, Y. Han, Y. Gao, B. Zhang, and G. Shvets, "Sorting and routing topologi-
cally protected edge states by their valleys," arXiv:1712.04589 (2017).
28K. F. Graff, Wave motion in elastic solids (Courier Corporation, 2012).
29J. Vila, R. K. Pal, and M. Ruzzene, "Observation of topological valley modes in an elastic
hexagonal lattice," Phys. Rev. B 96, 134307 (2017).
30"Supplemental material,".
31M. Ruzzene, "Frequency -- wavenumber domain filtering for improved damage visualiza-
tion," Smart Materials and Structures 16, 2116 (2007).
9
32M. Miniaci, A. Marzani, N. Testoni, and L. De Marchi, "Complete band gaps in a polyvinyl
chloride (pvc) phononic plate with cross-like holes: numerical design and experimental
verification," Ultrasonics 56, 251 -- 259 (2015).
Methods
Simulations. Dispersion diagrams and mode shapes presented in Figs. 1d-f and Fig. 2
are computed using Bloch-Floquet theory in full 3D FEM simulations carried out via the Finite
Element solver COMSOL Multiphysics. Full 3D models are implemented to capture all possible
wave modes supported by the plate structure. A linear elastic constitutive law is adopted and the
following mechanical parameters used for the plate material (aluminum): density ρ = 2700 kg/m3,
Young modulus E = 70 GPa, and Poisson ratio ν = 0.33. The elastic domain is meshed by means
of 8-node hexaedral elements of maximum size LF E = 0.5 mm, which is found to provide accurate
eigen solutions up to the frequency of interest32.
The band structures shown in Fig. 1d-f are obtained assuming periodic conditions along the lattice
vectors a1 and a2. Dispersion diagrams shown in Figs. 2 are computed instead considering a 20× 1
a1-periodic strip. The resulting eigenvalue problem (K − ω2M)u = 0 is solved by varying the
non-dimensional wavevector k along the boundaries of the irreducible Brillouin zone [Γ, M, K] for
dispersion diagrams in Fig. 1d-f and within [−π, π] for band structures presented in Figs. 2.
The distribution of the von Mises stress fields reported in Figs. 3c,d are conducted in the frequency
domain via the Finite Element solver ABAQUS. Free boundary conditions are applied at the edges
of waveguide.
Experimental measurements and data processing. The plate, consisting of 35 (in the a1
direction) × 25 (in the a2 direction) unit cells, is fabricated through a two-step machining process.
First, the triangular holes are obtained through water-jet cutting. Circular blind and through holes
are then obtained via a computer assisted drilling process. The specimen is made of aluminum
6082 T6, with the following nominal properties: density ρ = 2700 kg/m3, Young modulus E = 70
GPa, and Poisson ratio ν = 0.33. The plate dimensions and key geometrical parameters are as
follows: a = 20.5 mm, H = 5.9 mm, h = 0.5 mm, R = 1.75 and r = 0.8 mm. Elastic waves are
excited through a piezoelectric disk (12 mm diameter) bonded to the top surface of the plate at
10
location shown in Fig. 3b. Ultrasonic pulses consisting of 51 sine cycles modulated by a Hanning
window of central frequency of 98 kHz are used as the excitation signals.
The experimental wavefields shown in Figs. 4b and 4f are recorded by a scanning laser Doppler
vibrometer (SLDV) that measures the out-of-plane velocity of points belonging to a predefined
grid over the structure. The spatial resolution of the grid is approximately 0.2 mm for the 2D local
scan represented by the red dotted area in Fig. 3b and 0.6 mm for the 2D local scan in the blue
dotted area also in Fig. 3b.
The frequency/wavenumber representation of the edge modes presented in Figs. 4a are obtained
by performing a temporal and spatial Fourier transform (2D-FT) of the signals detected along
the 1D-scan lines reported as black dotted lines in Fig. 3b. The wavenumber content of the 2D
scan wavefield shown in Fig. 4f-h are obtained by performing spatial 2D-FT of the acquired data
interpolated over a regular square grid. Filtering in the wavenumber domain31 for modal separation
relies on the application of 2D Gaussian windowing functions centered at wavenumber ki, which
ca be expressed as follows:
Hi(kx, ky) = e
−(k−ki)2
2σ2
where i = 1, 2, with k1 = 60 [rad/m], k2 = 80 [rad/m], k =
(cid:113)
k2
x + k2
y, σ2 = 50. The wavefields
corresponding to the separated modal contributions shown in Figs. 4g,h are obtained through an
inverse 2D-FT of the filtered wavenumber representations shown in Figs. 4d,e.
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.
Contributions
All authors contributed extensively to the work presented in this paper.
Competing interests
The authors declare no competing financial interests.
11
FIG. 1. Design of the unit cells leading to distinct topological phases and their disper-
sion properties. a, In-plane and cross-sectional view of the unit cell with through holes. The
holes have equal diameter R = 0.0875a, where a = 20.5 mm is the magnitude of the lattice vectors
(a = a1 = a2) and H = 5.9 mm denotes the plate thickness. The inset shows the irreducible
Brillouin zone and the high symmetry points Γ, K and M . b, Perspective and cross-sectional view
of the unit cells (H + and H−) emulating spin orbital coupling in Quantum spin Hall effect with σh
broken symmetry (blind holes). c, Perspective and cross-sectional views of the unit cells (V r and
V R) emulating the Quantum valley Hall effect with σv broken symmetry (through holes of radii
r (cid:54)= R). d-f, Calculated phononic band structure for the plate with through holes, and for the
plates composed of H + (H−) and V r (V R) unit cells, respectively. The plate with through holes
is characterized by a double degenerate Dirac point visible in (d), while the cases of H + (H−) (e)
and V r (V R) (f) feature a complete bandgap (light gray rectangle) centered at approximately 102
kHz. The widths and center frequency of the bandgaps are matched by selecting the partial depth
of the blind holes h in the H + (H−) configuration (h = 0.91H for the band diagram shown), and
the radii r and R of the through holes in the V r (V R) case (r = 0.51R for the diagrams shown).
Refer to the Methods section Simulations for details on band structure calculations.
12
M K M Plate unit cella)Helical Phononic Crystal (H)b)Valley Phononic Crystal (V)c)HhHhHd)e)f)a1a2rRrRRRRRRRFrequencyReduced Wavenumber k*M K ΓM1201101009080Reduced Wavenumber k*M K ΓM1201101009080Reduced Wavenumber k*M K ΓM1201101009080KMΓ FIG. 2. Non-trivial interfaces: band structure and edge states a,c,e, Dispersion diagrams
for the non-trivial waveguides defined by I(H +, H−), I(H +, V R), and I(H−, V r) interfaces. The
band structures are computed considering a 20×1 a1-periodic strip (10 unit cells on each side of the
domain wall). The bulk modes are reported as gray dots while the interface modes in black, blue
and red dotted lines, respectively. The edge modes are denoted by the index + (−) according to
the positive (negative) group velocity relative to the propagation direction. b,d,f, Corresponding
eigenvectors (colors represent magnitudes of the absolute normalized displacement, varying from
zero (blue) to maximum (red)) show mode localization at the interface (the deformation for only 6
cells is reported for the clarity of representation). c,d Close-ups for the Ψ+ and Ψ− modes highlight
the different displacement distribution and phases (positive in blue and negative in red) of the
modes at the interface, which suggests the possibility of selective mode excitation. The preferential
mode excitation for Ψ+ (Ψ−) can be achieved by applying an excitation at the maximum amplitude
point highlighted in the insets.
13
/a /a /a 0 /a C 0 Ψ+a)Ψ -- De)c)Φ -- Φ+-π0 πReduced Wavenumber10510095Frequency [kHz]10510095Frequency [kHz]-π0 πReduced Wavenumber-π0 πReduced Wavenumber10510095Frequency [kHz]f)d)Φ -- (point A)Φ+(point B)Ψ+(point C)Ψ -- (point D)b)ABMax amplitude pointsMax amplitude pointPhasePhase FIG. 3. Configuration of the finite structure and numerical simulations showing se-
lective mode waveguiding. a, Schematic representation and experimental implementation of
the non-trivial waveguide hosting H + (in red), H− (in green) and V (in blue) phases giving rise
to 3 interfaces: I(H +, H−) (blue-green), I(H +, V R) (blue-red), and I(H−, V r) (red-green). b,
Waveguide schematic showing the locations of the excitation and 1D and 2D scan points/regions
considered in the experiments (1D scans are the black dotted lines, local 2D scans are the red
dotted areas, while large 2D scan of the y-junction region is the blue dotted area). c,d, Numerical
distribution of the von Mises stress field resulting from harmonic excitation at 98 kHz, i.e. within
the bulk gap. The excitation is applied at the location shown by the white dot as an out-of-plane
displacement distribution. Specific displacement distribution of the surface stress is applied ac-
cording to the modal content of the modes shown in Fig. 2d,f in order to selectively induce mode
Ψ+ (c) and Ψ− (d), respectively. The calculations clearly illustrate the possibility to preferentially
excite one of the two modes and to remotely select the interface along which the wave propagates.
Colors indicate the von Mises stress magnitude, ranging from zero displacement (blue) to maximum
displacement (red). Refer to SM30 for additional transient dynamic simulations.
14
d)c)a)b) FIG. 4. Experimental observation. a, Spatio/temporal 2D-FT of 1D line scans along interfaces
I(H +, V R) and I(H−, V r). Colormaps show normalized 2D-FT amplitudes, superimposed to the
numerically predicted band structure (white square dots) highlighting the Ψ+ (left panel) and Ψ−
(right panel) modes. b, Measured wavefield for the Ψ+ and Ψ− propagating edge modes displaying
opposite spins profiles, as highlighted by the superimposed black arrows. The measurements corre-
spond to the area denoted as "2D local scan regions" in Fig. 2b. Time animations of the measured
wavefields are provided in the SM30. c, Spatial 2D-FT for a representative snapshot (t = 1120µ)
of the wavefield recorded over the "2D scan region" in Fig. 3b and shown in Fig. 4f. The 2D-FT
highlights the presence of a pair of diffraction peaks defined by the concentration at contour levels
in the reciprocal space kx, ky, which are associated with wavenumbers k1 = 60 rad/m (red circle)
and k2 = 80 (black circle) rad/m, corresponding to the distinct modes of propagation Ψ+ and Ψ−.
d,e Filtered 2D-FTs with isolated modes and g,h corresponding propagation in physical space
showing the decoupled wavefields and distinction propagation paths for the two separated modes.
15
00110.70.4[a.u.] 0010.40-0.4[mV]t = 975.7813 s-0.03-0.02-0.0100.010.02x [mm]0.020.030.040.05y [mm]x [cm]15 16 17 18 19 20 21 10987y [cm]t = 842.9688 s-0.06-0.05-0.04-0.03-0.02-0.01x [mm]0.020.030.040.05y [mm]x [cm]22 23 24 25 26 27 28 19181716y [cm] 001[mV] 00.51[a.u.]0.40-0.4Reduced wavenumberReduced wavenumbert = 842 μst = 975 μsΨ -- Ψ+a)c)d)e)b)f)g)h)-- |
1912.12061 | 1 | 1912 | 2019-12-27T11:42:06 | Strongly Absorbing Nanoscale Infrared Domains within Graphene Bubbles | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Graphene has shown great potential for modulating infrared (IR) light in devices as small as 350 nm. At these length scales, nanoscale features of devices, and their interaction with light, can be expected to play a significant role in device performance. Bubbles in van der Waals heterostructures are one such feature, which have recently attracted considerable attention thanks to their ability to modify the optoelectronic properties of 2D materials through strain. Here we use scattering-type scanning near-field optical microscopy (sSNOM) to measure the nanoscale IR response from a network of variously shaped bubbles in hexagonal boron nitride (hBN)-encapsulated graphene. We show that within individual bubbles there are distinct domains with strongly enhanced IR absorption. We correlate this with strain in the graphene, found with confocal Raman microscopy and vector decomposition analysis. This reveals intricate and varied strain configurations, in which bubbles of different shape induce more bi- or uniaxial strain configurations. Ridges in the bubbles, seen by atomic force microscopy (AFM), coincide with the domain boundaries, which leads us to attribute the domains to nanoscale strain differences in the graphene. This reveals pathways towards future strain-based graphene IR devices. | physics.app-ph | physics | Strongly Absorbing Nanoscale Infrared
Domains within Graphene Bubbles
Tom Vincent,†,‡ Matthew Hamer,§,║ Irina Grigorieva,§,║ Vladimir Antonov,¶,‡
Alexander Tzalenchuk†,‡ and Olga Kazakova*,†
†National Physical Laboratory, Hampton Road, Teddington TW11 0LW, U.K.
‡Department of Physics, Royal Holloway University of London, Egham TW20 0EX, U.K.
§School of Physics and Astronomy, University of Manchester, Manchester M13 9PL, U.K.
║National Graphene Institute, University of Manchester, Manchester M13 9PL, U.K.
¶Skolkovo Institute of Science and Technology, Nobel str. 3, Moscow, 143026, Russia
ABSTRACT: Graphene has shown great potential for modulating infrared (IR) light in devices
as small as 350 nm. At these length scales, nanoscale features of devices, and their interaction
with light, can be expected to play a significant role in device performance. Bubbles in van der
Waals heterostructures are one such feature, which have recently attracted considerable
attention thanks to their ability to modify the optoelectronic properties of 2D materials
through strain. Here we use scattering-type scanning near-field optical microscopy (sSNOM)
to measure the nanoscale IR response from a network of variously shaped bubbles in
hexagonal boron nitride (hBN)-encapsulated graphene. We show that within individual
bubbles there are distinct domains with strongly enhanced IR absorption. We correlate this
with strain
in the graphene, found with confocal Raman microscopy and vector
decomposition analysis. This reveals intricate and varied strain configurations, in which
bubbles of different shape induce more bi- or uniaxial strain configurations. Ridges in the
bubbles, seen by atomic force microscopy (AFM), coincide with the domain boundaries, which
leads us to attribute the domains to nanoscale strain differences in the graphene. This reveals
pathways towards future strain-based graphene IR devices.
KEYWORDS: Graphene, bubbles, infrared, absorption, strain, SNOM, Raman
Graphene and hexagonal boron nitride (hBN)'s unique optoelectronic properties make
them ideally suited for a variety of applications in the infrared (IR) range.1,2 In graphene,
surface plasmon polaritons (SPPs; coupled oscillations of light and free charge carriers) have
been shown to have wavelengths many orders of magnitude smaller than the diffraction limit,
long lifetimes and exceptional electronic tunability via gating.3 -- 10 Similarly in hBN, hyperbolic
phonon polaritons (HPhPs; coupled oscillations of light and optical phonons) have short
wavelengths and long lifetimes, as well as negative phase velocities.11 -- 14 When the two
materials are layered together in van der Waals heterostructures they support hybrid
phonon-plasmon polaritons,15 as well as even longer polariton lifetimes16 and moiré
modulated polariton dispersions.17 These properties have been utilised in a range of
optoelectronic devices, including photodetectors operating in the IR18 and THz19 regions, and
optical modulators with footprints as small as 350 nm.20,21
Recently, a great deal of attention has been paid to the effects of strain on 2D materials.
Compared to bulk materials, their electronic and optical properties are especially susceptible
to modification by strain. This, combined with their significant pliability, has led to a new field
of research dubbed straintronics.22 Examples in the infrared regime include demonstrations
that the dispersion of HPhPs is altered by the presence of strain in hBN,23 and that wrinkles in
graphene may act as scattering sites for SPPs.24
Bubbles in van der Waals heterostructures have emerged as an interesting platform to
study the effects of strains on 2D materials.25 In transition metal dichalcogenides, bubbles
have been shown to act as highly localised photoluminescent emitters, with strain-dependent
peak energies.26 In graphene, nanoscale bubbles have been shown to act as localised
plasmonic hotspots,27 and to sustain high-Tesla pseudomagnetic fields.28
These bubbles are formed due to competition between van der Waals and elastic potential
energies, in the presence of interlayer contamination, which may be formed of adsorbed
hydrocarbons and water vapour.29 This effectively squeezes the contamination into pockets,
leaving micron-scale areas with atomically sharp interlayer interfaces. This process, referred
to as self-cleaning, promotes interlayer adhesion enabling large-area van der Waals
heterostructures to be realised.29,30
A benefit of using bubbles to strain 2D materials is that the stochastic nature of their
formation means they provide a wide range of strain values and configurations. This means
that, with a suitable method to measure local strain, a single sample can be used to correlate
strain with induced effects, without the need for external sources of strain variation. These
strain-induced effects have great potential for exploitation in novel devices, which may be
based on bubbles themselves, or, perhaps more feasibly, on more controllable methods of
straining 2D materials, such as transfer onto patterned substrates.31
From a different perspective, the ubiquity of bubbles means it is also important to
understand any unintended effects they may have on devices, for quality control purposes.
Particularly because the small sizes of modern devices are comparable to those of typical
bubbles in van der Waals heterostructures. For large-scale production of graphene-based
optical modulation devices to become viable, knowledge of the role that bubbles may play in
device performance is vital.
In this work, we use scattering-type scanning near-field optical microscopy (sSNOM) to
probe the nanoscale IR response of a network of variously shaped bubbles in an hBN-
encapsulated graphene heterostructure. This reveals distinct domains with significantly
enhanced absorption within individual bubbles. The boundaries of these domains correlate
with ridges in the shape of the bubbles, which leads us to attribute them to nanoscale
variations in strain configuration. We investigate this further by using confocal Raman
spectroscopy, along with vector decomposition analysis,32 -- 34 to create spatial maps of strain
and doping variations from the same heterostructure. We demonstrate that networks of
bubbles induce mixed and intricate strain configurations, with localised areas of mostly
uniaxial or biaxial strain, and that there is a pronounced increase in hole doping induced by
the contaminants in the bubbles.
The encapsulated graphene heterostructure in this work was fabricated on 290 nm SiO2
using mechanically exfoliated crystals of hBN and graphene. The heterostructure consists of
~220 nm-thick lower hBN, single layer graphene (SLG), and ~1.2 nm-thick upper hBN and was
made using the now standard dry peel transfer technique35 with a bespoke
micromanipulation setup36 (see Methods). The optical image in Figure 1a shows a large
amount of bubbles, of varied size and shape, in the encapsulated region of the
heterostructure. Zooming in on an area of the heterostructure using peak-force tapping
atomic force microscopy (AFM) reveals that these bubbles are connected by thin wrinkles or
filaments (Figure 1b). These filaments are related to the shape of the bubbles, with the
number of filaments connected to each bubble corresponding to the number of sharp corners
in its footprint. This implies that the proximity of the bubbles in this heterostructure may have
led to their exerting a collective influence on each other's shape during formation. It is also
revealed that there are variations in shape within individual bubbles. The bubbles shown have
ridges in their sides, which separate regions of different curvature. There are also smaller
bubbles, which are more circular in shape and are not connected by filaments.
The inset to Figure 1a shows a schematic cross-section of a bubble in this heterostructure,
with the contamination between the lower hBN and graphene. It is also possible for bubbles
to exist between the graphene and upper hBN. It would be difficult to distinguish these types
of bubble using AFM alone, but we show below that the bubbles in this work strain the
graphene, so we conclude that the contamination is beneath the graphene layer.
Figure 1. Bubbles in an hBN-graphene-hBN heterostructure. (a) An optical microscope image of the heterostructure showing
bubbles between the layers. The ~220 nm lower hBN layer fills the entire view; the known edges of the single layer graphene
(solid yellow line, determined by confocal Raman) and the ~1.2 nm upper hBN layer (dashed blue line, determined by AFM)
are indicated. The darker purple areas are multilayer graphene. The region focused on in the rest of this work is indicated by
a dashed white rectangle. Inset: side view schematic of a bubble, showing contamination (in red) trapped beneath graphene
and hBN. (b) A 3D surface topography map of the area indicated in (a). The scale of the z-coordinates has been exaggerated
to show topographical features more clearly.
It is important to understand the nanoscale IR properties of graphene and hBN, so they
may be exploited in devices. To this end, we use sSNOM to probe how the presence of bubbles
affects the IR response of the heterostructure. In this technique, IR light is focused onto a
metallised AFM tip, exciting a tightly confined near-field around the tip's apex. The scattered
light from the near-field interacting with the sample can then be measured. This allows
imaging of complex light-matter interactions at a resolution many orders of magnitude below
the diffraction limit, typically ~30 nm (see Methods).37
We show below that within graphene bubbles there exist nanoscale domains whose
optical absorption is significantly altered at incident wavenumbers, k, of around 1000 cm-1.
These domains are not reproduced at 1362 cm-1 which shows the effect is wavelength
dependent.
Figure 2. sSNOM images of nanoscale optical domains within hBN-encapsulated graphene bubbles. (a, b) Third harmonic
sSNOM amplitude (s3) maps, at k=1000 and 1362 cm-1 respectively. Values are normalised so that the median value in each
map is equal to 1. Inset of (a) shows zoom on a single bubble with domains outlined. (c, d) Third harmonic sSNOM phase (ϕ3)
maps. Values are normalised so that the median value in each map is equal to 0°. Scale bar in (a) shared by all images.
Figure 2 shows the third harmonic near-field scattering from the area under study (see
Figure 1), taken at k=1000 and 1362 cm-1. Figures 2a and 2b show the scattering amplitude,
s3, normalised to the background graphene value by dividing by the median amplitude for
each map, so that the background amplitude is ~1. At both 1000 and 1362 cm-1, the bubbles
and filaments have a reduced scattering amplitude, however the reduction is less significant
at 1362 cm-1. At 1000 cm-1, there are two distinct levels of contrast observed within the larger
bubbles, with one level at an amplitude of ~0.6 and the other at ~0.25 times the background
amplitude (shown by red and white outlines, respectively, in the zoomed portion of
Figure 2a). These are most visible in the two bubbles at the left of the image. They form
domains whose boundaries correlate well with the topographic ridges of Figure 1b. The
amplitude within the bubbles at 1362 cm-1 is more homogeneous and the same pronounced
domains are not observed.
Figures 2c and 2d show the corresponding complex phase of the scattered light, ϕ3,
normalised to the background graphene by subtracting the median phase for each map. At
k=1000 cm-1 the domains seen in Figure 2a are well reproduced, with multiple domains clearly
visible in all large bubbles, again correlating with the geometrical ridges seen in Figure 1b.
The parts of the bubbles with the greatest reduction in s3 have a significant phase shift of
~90°, which indicates that those parts of the bubbles strongly absorb light at 1000 cm-1. The
remaining parts of the same bubbles are characterised by a slight negative phase shift of ~5°.
Conversely, the phase shifts within bubbles at 1362 cm-1 are much smaller, at ~30°, and the
same pronounced domains are not observed.
To investigate the wavelength dependence in more detail, a single bubble (from the lower
left corner of the image) was imaged repeatedly, while changing k from 960 to 1040 cm-1.
Figures 3a-e show the resulting third harmonic near-field scattering amplitude maps from the
bubble and the surrounding area.
Figure 2. Spectroscopic dependence of domains within a bubble. (a-e) Third harmonic sSNOM amplitude (s3) maps from a
bubble, at wavenumbers between 960 and 1040 cm-1. Left: Domains returned by fitting a two-component Gaussian mixture
model (GMM) at 960 cm-1. (f) Histograms showing the distribution of s3 values at each wavenumber, overlaid with error bars
showing the means and standard deviations returned by fitting the GMM to the distribution. Lines are coloured to match
the domains shown left of (a-e). (g-k) Corresponding sSNOM phase (ϕ3) maps. Left: Domains returned by fitting a three-
component GMM at 960 cm-1. (l) Histograms showing the distribution of ϕ3 values at each wavenumber, overlaid with error
bars showing the means and standard deviations returned by fitting the GMM to the distribution. Lines are coloured to
match the domains shown left of (g-k). Scale bar in (a) shared by all images.
The bubble is again split into two domains. To illustrate these internal domains more
clearly, and to remove the influence of surface adsorbants, which were present on the flat
area, the s3 values from within the bubble were isolated (see Methods). These values are
shown for each k value by histograms in Figure 3f.
To gain a more quantitative measure of the evolution of these domains, a Gaussian mixture
model (GMM) was fit to each distribution. This assumes that a dataset is composed of
N normally distributed clusters, then uses the expectation-maximisation algorithm38 to
determine the parameters of the N Gaussian peaks that best describe these clusters. Here N
was set to match the number of experimentally observed domains.
For the s3 maps we observed two domains experimentally, so used N=2 for the GMM. The
means and standard deviations returned by the GMM are overlaid as error bars in Figure 3f.
To verify that the components returned are the same as the experimentally observed
domains, and to serve as a visual key, the s3 points were evaluated according to which
Gaussian component they are most likely to belong to and coloured accordingly. This is
displayed, using k=960 cm-1 as an example, to the left of the plot.
Across the range of wavenumbers studied, the complex amplitudes of each domain do not
vary significantly, with the left side remaining at ~0.6 and the right-hand side at ~0.25 times
the background amplitude, as seen in Figure 2a.
Figures 3g-k show the corresponding third harmonic near-field scattering phase maps. In
these images (captured at higher resolution than in Figure 2) there are three, rather than two,
distinct levels of phase contrast within this bubble. The left side has a small negative phase
shift (relative to the zero-normalised background), and the right-hand side has a high shift in
the centre and reduced shift at the edges. The phase shifts in the right-hand side reduce with
increasing k.
The same process of isolating values from the bubble, plotting the distribution and fitting
to the distribution with a GMM was repeated for the phase images, this time with N=3. The
resulting data is plotted in Figure 3l. The domain with a slight negative phase shift of around
5° is unchanged by k. However the domains with high phase shift are wavenumber
dependent, with the centre right domain shifting from ~150° to ~90° and the edges of the
right-hand side shifting from ~75° to ~30°, with a k increase from 960 to 1040 cm-1. This may
indicate that this range of wavenumbers is on the side of an absorption peak for this area of
the bubble.
These domains are separated by ridges in the bubbles' shape, and the shapes of bubbles
are known to result from competition between van der Waals and elastic potential energies.
For this reason it is probable that the domains will have different strain configurations. To
investigate this further, we used Raman spectroscopy to visualise the strain variations in this
heterostructure.
Raman is routinely used as an indicator of the quality of graphene.39 The solid blue line in
Figure 4a shows a Raman spectrum taken from a flat area of the heterostructure, free from
bubbles. It displays the characteristic G and 2D peaks of graphene, at ~1580 and ~2680 cm-1
respectively. The 2D to G height ratio of ~3 and 2D peak full width at half maximum (FWHM)
of ~20 cm-1 are characteristic of high-quality hBN-encapsulated SLG. The graphene D peak at
~1362 cm-1 is not apparent, which is another indicator of defect-free graphene.
Analysis of these Raman peaks can provide a wealth of information about the strain and
doping of graphene. To produce spatial maps of these quantities, a Raman datacube was
collected from the area of the sample shown above, and Lorentzians were fit to the G and 2D
peaks. The extracted positions for the peak centres, ωG and ω2D, are correlated in a scatterplot
in Figure 4b.
Figure 4. Graphene strain and doping analysis using vector decomposition model. (a) Raman spectra from a flat area (solid
blue, location: triangle in (c)) and a bubble (dashed purple, location: circle in (c)) in the hBN-encapsulated graphene
heterostructure. (b) G and 2D position scatterplot from the region under study, showing the model used to separate the
effects of hydrostatic strain (εh) and hole concentration (n). The colour of the points indicates the local density in the
scatterplot, where bright yellow corresponds to a higher density of points. (c) AFM topography map. (d) Median normalised
map of εh changes. (e, f) Line profiles showing height and strain across a more biaxially and uniaxially strained bubble,
respectively (locations shown in (d)). (g) Map of 2D splitting, which correlates with shear strain. (h) Median normalised map
of doping changes. Scale bar in (c) shared by (d), (g) and (h).
Both hole doping, n, and strain, ε, in graphene cause a shift of the G and 2D peaks, so
additional analysis is needed to determine the separate ε and n contributions. This can be
achieved by correlating the shift of both peaks using vector decomposition, and comparing
them to empirical measurements.33,40 The result of this is that the evolution of a point in an
ωG-ω2D correlation plot under changing ε or n can be approximated by a straight line with a
known gradient.32 This vector decomposition analysis is illustrated by the additional axes
shown in Figure 4b.
Different species of strain, for example biaxial and uniaxial, result in different gradients for
the straight line associated with strain changes. In samples with unknown or mixed strains
this can be accounted for by choosing to use the gradient associated with the hydrostatic
strain, εh. This is a component of the full biaxial strain tensor, εbi, which can be described by
two components: εh and shear strain, εs.33
𝜀𝑏𝑖 = (
𝜀𝑥𝑥
𝜀𝑦𝑥
𝜀𝑥𝑦
𝜀𝑦𝑦
)
𝜀ℎ = 𝜀𝑥𝑥 + 𝜀𝑦𝑦
𝜀𝑠 = √(𝜀𝑥𝑥 + 𝜀𝑦𝑦)
2
2
+ 4𝜀𝑥𝑦
(Assuming 𝜀𝑥𝑦 = 𝜀𝑦𝑥)
(1)
(2)
(3)
A qualitative explanation of these relationships is that εh corresponds to an isotropic
change in size of the unit cell, while εs corresponds to a change in the shape of the unit cell,
which leaves its area unchanged.
For ease of comparison with the colocalised Raman-acquired maps, the AFM topography
image from Figure 1b is shown again as a 2D image in Figure 4c.
Figure 4d shows the median-normalised hydrostatic strain distribution around the same
area of bubbles. The greatest values of tensile (positive) strain are localised at the centres of
bubbles, coinciding with the areas of greatest height. This is consistent with expectations for
bubbles with contamination beneath the graphene layer. Interestingly the areas of graphene
and hBN in the vicinity of the bubbles are not free from strain. The areas between bubbles,
close to the filaments (Figure 1b), show small increases in tensile strain relative to the
background.
Additionally, at the sides of some bubbles, particularly the narrow, elongated bubble in the
centre of the image, there are small areas of more compressive strain. These are due to
Poisson contraction, a phenomenon associated with uniaxial strain configurations.41 This is
illustrated more clearly by the AFM and εh line profiles taken from a more biaxially (Figure 4e)
and a more uniaxially (Figure 4f) strained bubble, as indicated in Figure 4d.
Uniaxial strains cause an anisotropic deformation to the lattice, and therefore correlate
with εs. This anisotropy causes a polarisation dependent splitting of the peaks about their
centres.33 The dashed purple line in Figure 4a shows a Raman spectrum taken from the
central bubble in the heterostructure (indicated in Figure 4c). The presence of εs has caused
splitting of the 2D peak. To minimise the polarisation dependence, we used circularly
polarised light for the incident Raman laser. However there is a small residual polarisation
dependence, introduced by the diffraction grating, which explains the asymmetry of the split
2D peak.
By fitting two Lorentzians to a peak, it is possible to obtain a measure of this peak splitting.
split, is shown in Figure 4g.
We performed this for the 2D peak, and the resulting splitting, ω2D
split
As discussed above, this should be proportional to εs. Indeed the greatest values of ω2D
correlate well with the areas of Poisson contraction in Figure 4d, providing further evidence
that these areas have a more uniaxial strain configuration.
The change in hole concentration of the graphene is also returned by the vector analysis.
Figure 4h shows the median-normalised doping distribution. The concentration correlates
well with the topography of the bubbles, showing an increase of ~4×1012 cm-2 relative to the
background at the bubble locations. The influence of bubbles on the carrier concentration is
much more localised than on εh. Doping is seen even from the small, more rounded bubbles,
not connected by filaments.
Compared to the strain distribution, the doping is more tightly confined to the bubble
locations. This confirms that the primary doping source in these bubbles is the contaminant
that fills them, and that the doping and strain here are independent. The graphene between
the bubbles is shown to be dopant-free, which supports the perception that bubble formation
is an effective self-cleaning mechanism.29,30
We do not observe features that correlate with the domains seen from sSNOM in any of
the Raman-acquired maps. However this is to be expected, as the spot size for our Raman
measurements (~450 nm) is comparable to the lateral sizes of the domains (~500 nm).
The vector decomposition model relies on a few assumptions about the nature of the
graphene, namely that it is single layer, relatively defect free and primarily p-type doped.32
The validity of the first two assumptions for our heterostructure is verified by Raman
spectroscopy, as shown above, but the type of doping (n- or p-) cannot be determined from
the peak shift alone. However bubble-free graphene encapsulated in hBN is known have an
intrinsic doping close to charge neutrality, due to a lack of dangling bonds in the hBN and
screening of the graphene from charged impurities and atmospheric dopants.42 -- 45 Typical
hydrocarbon contaminants in bubbles induce hole doping in graphene,29,30,43 so we assume
that the primary doping mechanism is p-type.
The strain and doping maps presented above are both normalised so that their median
value is equal to 0. This is to compensate for dielectric screening of the graphene Kohn
anomaly, caused by hBN, which adds a constant offset to both ωG and ω2D.46,47 This results in
expected behaviour for encapsulated graphene in the Δn map, with flat areas close to charge
neutrality and increased p-type doping at bubble locations, as discussed above, so we may
make the approximation that Δn≈n.
It is known that variations in in Fermi level, EF, can change the IR absorption of graphene.
We used the above approximation to calculate the Fermi level, according to the following
equation. 48
𝐸𝐹 =
ℎ
2𝜋
𝑣𝐹√𝜋𝑛
(4)
Here h is Planck's constant, and vF (≈106 m s-1) is the graphene Fermi velocity. Applying
this to the Raman-acquired n values yields a map of EF, shown in Figure 5a.
The absorption of graphene is proportional to the real part of its frequency dependent
optical conductivity, σ(ω). At high photon energy, Eph, interband transitions dominate, leading
to a universal conductance value of σ0=πe2/2h, where e is the elementary charge. This results
in a flat absorption of ~2.3%. But for photon energies below 2EF, these interband transitions
are prevented by Pauli blocking. This leads to a drop in the absorption and a Drude-type
response to light. As 𝐸𝐹 ∝ √𝑛, the position of the Pauli blocking transition, at Eph=2EF, is
dependent on the level to which the graphene is doped.4
The optical conductivity of graphene as a function of EF, calculated using the local random
phase approximation4 (see Methods) at a temperature, T=0 K, is shown for wavenumbers of
1000 and 1362 cm-1 in Figure 5b. The onset of the Pauli blocked regime occurs at EF=62 and
84 meV respectively. To more easily compare this calculation with the experimentally
determined EF values, the values from Figure 5a are shown as a histogram in Figure 5b.
At both wavenumbers, EF is below the Pauli transition for the areas of flat graphene
(corresponding to the prominent peak in the histogram), which indicates that they should be
in the high absorption, universal conductance regime. The doping at the bubbles shifts EF into
the Pauli blocked regime, which should be accompanied by a reduced σ and absorption, as
well as difference in σ between 1000 and 1362 cm-1. This may explain why the strongly
absorbing domains appear only at the lower wavenumber. However, the fact that we see an
increase, rather than the predicted decrease, in absorption at the bubble locations tells us
that we cannot explain all the variations from EF alone, and that other factors need to be
considered.
Figure 5. Graphene Fermi level and onset of Pauli blocking. (a) Raman-acquired map of EF, calculated from the data in
Figure 4h. (b) Real part of graphene optical conductivity, normalised to the universal conductance value, σ0. Calculated using
the local random phase approximation at T=0 K.4 Background: normalised kernel density histogram of EF values in (a) (linear
y-scale).
Graphene-hBN heterostructures, doped above the Pauli transition are known to support
SPPs in the wavenumber region around 1000 cm-1.1,2,4,16,17 This is the case for the bubbles in
this heterostructure. We do not observe the plasmonic standing waves typical of sSNOM
measurements of SPPs in graphene,3 -- 10 however this may be due to the lack of low-loss
reflection sites for SPPs, such as graphene edges. Nevertheless, it is possible that the origin of
the domains we observe at 1000 cm-1 could be plasmonic in nature.
Theoretical simulations of self-assembled 3D graphene nanostructures, such as pyramids
and polyhedrons, have suggested that vertexes and edges in such structures could have
strong plasmonic field enhancement due to combined in- and out-of-plane coupling of
plasmonic modes 49,50. We do not see evidence of this enhancement at the edges and vertexes
in the bubbles in this work. This may be because the bubbles we studied (which have heights
~0.1 × their footprints) are much flatter than the 3D self-assembled structures (which have
comparable heights to their footprints). This would reduce the effect of the mode coupling.
Nanoscale strain differences induced in the graphene by bubbles are a likely candidate for
the origin of the optical domains. Though the Raman spot size prevents us from visualising
strain variations within individual bubbles, we have shown that, at a larger scale, the strain
configuration from a network of closely spaced bubbles is intricate and varied, and not simply
related to height. The distributions we have measured, featuring Poisson contraction and
increased tensile strain between bubbles, are similar to theoretical calculations of stress in
graphene sheets deposited over networks of trapped particles.51 Knowing that strain
minimisation is one of the main factors governing a bubble's shape,25,51 the topographic
ridges separating the optical domains lead us to conclude that the different domains must
have different strain configurations.
The bandstructure of graphene is significantly altered by strain,22,52 -- 56 which may explain
the different absorption in different domains. If the origin of the domains is indeed plasmonic,
then they may be caused by an alteration of the plasmonic dispersion brought about by
bandstructure changes between differently strained areas.17,27,57 -- 60
Strain-induced areas of enhanced light absorption have great potential for the design of
future graphene-based IR devices. Networks of bubbles like the one shown here are generally
created stochastically during material transfer and are therefore difficult to control. But it
may be possible to create similar strain patterns in more reproducible ways, for example by
inflating graphene bubbles by pumping pressurised gas through networks of holes in a
substrate,61 or by depositing graphene onto an array of nanopillars.31,62 In the latter example,
one could imagine creating a photonic crystal, or metamaterial, with spatially modulated
absorption, simply by depositing graphene onto a prepatterned substrate.
In conclusion, we have shown that within bubbles in closely spaced networks in graphene-
hBN heterostructures there exist nanoscale domains with strongly enhanced absorption of IR
light at 1000 cm-1. We demonstrated with Raman spectroscopy and vector decomposition
analysis that these networks induce intricate and varied strain configurations in graphene.
Furthermore, we deduced that, due to shape variations seen in AFM, the IR domains must
also have nanoscale differences in strain configuration. This led us to attribute the origin of
the domains to these strain differences, which could alter the plasmonic dispersion of the
graphene. This result will have profound consequences for the design and quality control of
future graphene-based IR devices.
Methods. Sample fabrication. Graphite (obtained from HQ Graphene) was mechanically
exfoliated by peeling with Nitto Denko BT-150E-CM tape before being pressed onto a
Si/SiO2-290 nm wafer (heated to 60°C to improve adhesion). Suitable monolayer graphene
was then identified using optical microscopy and picked up with few-layer hBN using the
poly(methyl-methacrylate) (PMMA) dry peel transfer technique35 using a bespoke
micromanipulation setup.36 Following this, the sample on the membrane was stamped onto
a bulk (> 30 nm) crystal of laterally large (> 50 × 50 µm) hBN which had previously been
exfoliated onto a Si/SiO2-290 nm wafer. The completed structure therefore had the form
hBN/Gr/hBN/SiO2/Si.
After full encapsulation, the sample was annealed at 300°C for 3 hours in a high vacuum
environment (> 10-7 mbar) to remove polymer residue on the surface of the heterostructure
and promote the self-cleaning mechanism29 and subsequent bubble formation.25
Atomic Force Microscopy. The AFM image shown was taken on a Bruker Dimension Icon
SPM, using silicon nitride AFM tips with a resonant frequency of ~300 kHz. The
heterostructure was imaged using peak-force tapping mode, with a peak force setpoint of
500 pN to ensure that the tapping was gentle and didn't disturb the bubbles.
sSNOM. The sSNOM measurements were performed on an Anasys Instruments NanoIR-2s
AFM, with excitation provided by a quantum cascade laser (QCL). PtIr coated AFM tips with a
resonant frequency of ~285 kHz were used. IR light from the QCL was focused onto the AFM
tip using a parabolic mirror, exciting a tightly confined near-field around the tip apex. The
same mirror was used to collect the light scattered from the tip and sample, which was
measured with a mercury cadmium telluride detector. To isolate the component of the total
scattered light which is caused by the near-field interacting with the sample, the AFM was
operated in tapping mode and a lock-in amplifier was used to demodulate the detected light
at the third harmonic of the tapping frequency.37
To capture the full complex information, the light was collected via a Michelson
interferometer. The sample was imaged twice, with the interferometer reference mirror fixed
at two orthogonal phases, and the images were combined to yield the complex amplitude
and phase. For each image, an additional pass was taken with the interferometer reference
arm blocked, which was then subtracted from the image to account for any self-homodyne
background.37,63
To combine consecutive images, any lateral thermal drift of the sample position was
determined using cross correlation of the topography images, then the images were
translated and cropped to show the same area of the sample (to the nearest pixel).
To isolate the s3 and ϕ3 values of the bubble from the flat areas in Figure 3, we used only
values whose corresponding height, determined by AFM, was greater than 15 nm. We
attribute the variations in s3 and ϕ3, seen on the flat areas, to loose material on the sample
surface, which we could see in the AFM topography. We did not observe this contamination
on the bubbles themselves (likely due to the steeper sides), meaning the influence of this
material was effectively removed by selecting only values from within the bubble.
To verify that the observed domains were not the result of tip shadowing, or other
direction sensitive effects, bubbles were imaged in multiple orientations. The same domains
were observed consistently.
Raman Spectroscopy. Raman mapping was performed using a Renishaw inVia confocal
Raman microscope. We used a 532 nm excitation laser focused through a 100× objective, with
a numerical aperture of 0.85. The laser power incident on the sample was ~0.5 mW. A 1800
line mm−1 diffraction grating was used. The measurements were taken in high confocality
mode, resulting in an estimated spot size of ~450 nm in FWHM.
A quarter-wave plate was used to create circularly polarised light in the incident beam to
reduce the polarisation dependency of the measurements and therefore better resolve the
splitting of the 2D peak.33 There is still a small residual polarisation dependence from the
spectrometer, typically less than ~10%.
For extracting strain and doping information from the shifts of the G and 2D peaks, single
Lorentzians were fit to each peak to extract the positions. The gradients used to distinguish
between carrier concentration and hydrostatic strain, as well as the Grueneisen parameter
used to convert peak shift in cm-1 to percentage strain were taken from a 2017 work by
Mueller et al.33 The empirical measurements used to convert peak shift in cm-1 to hole
concentration in cm-2 were taken from a 2015 work by Froehlicher and Berciaud.40
Optical Conductivity Calculations. The optical conductivity of graphene was calculated
using the local random phase approximation in the limit where T=0 K, as described in the
Supporting Information of ref. 4. We used an estimated carrier scattering time of τ=10-13 s,
taken from the same work.
AUTHOR INFORMATION
Corresponding Author
*Email: [email protected]
ORCID
Tom Vincent: 0000-0001-5974-9137
Matthew Hamer: 0000-0003-3121-6536
Irina Grigorieva: 0000-0001-5991-7778
Vladimir Antonov: 0000-0002-0415-5267
Alexander Tzalenchuk: 0000-0001-7199-9331
Olga Kazakova: 0000-0002-8473-2414
Notes
The authors declare no competing financial interest
ACKNOWLEDGEMENTS
This project has received funding from the European Union's Horizon 2020 research and
innovation programme under grant agreement GrapheneCore2 785219 number. The work
has also been financially supported by the Department for Business, Energy and Industrial
Strategy though NMS funding (2D Materials Cross-team project).
REFERENCES
(1)
(2)
Basov, D. N.; Fogler, M. M.; Garcia de Abajo, F. J. Polaritons in van Der Waals Materials.
Science (80-. ). 2016, 354 (6309), aag1992 -- aag1992.
Low, T.; Chaves, A.; Caldwell, J. D.; Kumar, A.; Fang, N. X.; Avouris, P.; Heinz, T. F.;
Guinea, F.; Martin-Moreno, L.; Koppens, F. Polaritons in Layered Two-Dimensional
Materials. Nat. Mater. 2017, 16 (2), 182 -- 194.
(3) Grigorenko, A. N.; Polini, M.; Novoselov, K. S. Graphene Plasmonics. Nat. Photonics
2012, 6 (11), 749 -- 758.
Koppens, F. H. L.; Chang, D. E.; Garcia de Abajo, F. J. Graphene Plasmonics: A Platform
for Strong Light -- Matter Interactions. Nano Lett. 2011, 11 (8), 3370 -- 3377.
Fei, Z.; Rodin, A. S.; Andreev, G. O.; Bao, W.; McLeod, A. S.; Wagner, M.; Zhang, L. M.;
Zhao, Z.; Thiemens, M.; Dominguez, G.; et al. Gate-Tuning of Graphene Plasmons
Revealed by Infrared Nano-Imaging. Nature 2012, 486 (7405), 82 -- 85.
Chen, J.; Nesterov, M. L.; Nikitin, A. Y.; Thongrattanasiri, S.; Alonso-González, P.;
(4)
(5)
(6)
(7)
Slipchenko, T. M.; Speck, F.; Ostler, M.; Seyller, T.; Crassee, I.; et al. Strong Plasmon
Reflection at Nanometer-Size Gaps in Monolayer Graphene on SiC. Nano Lett. 2013, 13
(12), 6210 -- 6215.
Fei, Z.; Rodin, A. S.; Gannett, W.; Dai, S.; Regan, W.; Wagner, M.; Liu, M. K.; McLeod, A.
S.; Dominguez, G.; Thiemens, M.; et al. Electronic and Plasmonic Phenomena at
Graphene Grain Boundaries. Nat. Nanotechnol. 2013, 8 (11), 821 -- 825.
(8) Nikitin, A. Y.; Alonso-González, P.; Vélez, S.; Mastel, S.; Centeno, A.; Pesquera, A.;
Zurutuza, A.; Casanova, F.; Hueso, L. E.; Koppens, F. H. L.; et al. Real-Space Mapping of
Tailored Sheet and Edge Plasmons in Graphene Nanoresonators. Nat. Photonics 2016,
10 (4), 239 -- 243.
Fei, Z.; Goldflam, M. D.; Wu, J. S.; Dai, S.; Wagner, M.; McLeod, A. S.; Liu, M. K.; Post, K.
W.; Zhu, S.; Janssen, G. C. A. M.; et al. Edge and Surface Plasmons in Graphene
Nanoribbons. Nano Lett. 2015, 15 (12), 8271 -- 8276.
(9)
(10) Gerber, J. A.; Berweger, S.; O'Callahan, B. T.; Raschke, M. B. Phase-Resolved Surface
Plasmon Interferometry of Graphene. Phys. Rev. Lett. 2014, 113 (5), 1 -- 5.
(11) Dai, S.; Fei, Z.; Ma, Q.; Rodin, A. S.; Wagner, M.; McLeod, A. S.; Liu, M. K.; Gannett, W.;
Regan, W.; Watanabe, K.; et al. Tunable Phonon Polaritons in Atomically Thin van Der
Waals Crystals of Boron Nitride. Science (80-. ). 2014, 343 (6175), 1125 -- 1129.
(12) Dai, S.; Ma, Q.; Yang, Y.; Rosenfeld, J.; Goldflam, M.; McLeod, A.; Sun, Z.; Andersen, T.;
Fei, Z.; Liu, M.; et al. Efficiency of Launching Highly Confined Polaritons by Infrared Light
Incident on a Hyperbolic Material. Nano Lett. 2017, acs.nanolett.7b01587.
(13) Yoxall, E.; Schnell, M.; Nikitin, A. Y.; Txoperena, O.; Woessner, A.; Lundeberg, M. B.;
Casanova, F.; Hueso, L. E.; Koppens, F. H. L.; Hillenbrand, R. Direct Observation of
Ultraslow Hyperbolic Polariton Propagation with Negative Phase Velocity. Nat.
Photonics 2015, 9 (10), 674 -- 678.
(14) Li, P.; Lewin, M.; Kretinin, A. V.; Caldwell, J. D.; Novoselov, K. S.; Taniguchi, T.;
Watanabe, K.; Gaussmann, F.; Taubner, T. Hyperbolic Phonon-Polaritons in Boron
Nitride for near-Field Optical Imaging and Focusing. Nat. Commun. 2015, 6 (May), 1 -- 9.
(15) Dai, S.; Ma, Q.; Liu, M. K.; Andersen, T.; Fei, Z.; Goldflam, M. D.; Wagner, M.; Watanabe,
K.; Taniguchi, T.; Thiemens, M.; et al. Graphene on Hexagonal Boron Nitride as a
Tunable Hyperbolic Metamaterial. Nat. Nanotechnol. 2015, 10 (8), 682 -- 686.
(16) Woessner, A.; Lundeberg, M. B.; Gao, Y.; Principi, A.; Alonso-González, P.; Carrega, M.;
Watanabe, K.; Taniguchi, T.; Vignale, G.; Polini, M.; et al. Highly Confined Low-Loss
Plasmons in Graphene -- Boron Nitride Heterostructures. Nat. Mater. 2015, 14 (4), 421 --
425.
(17) Ni, G. X.; Wang, H.; Wu, J. S.; Fei, Z.; Goldflam, M. D.; Keilmann, F.; Özyilmaz, B.; Castro
Neto, A. H.; Xie, X. M.; Fogler, M. M.; et al. Plasmons in Graphene Moiré Superlattices.
Nat. Mater. 2015, 14 (12), 1217 -- 1222.
(18) Woessner, A.; Parret, R.; Davydovskaya, D.; Gao, Y.; Wu, J.-S.; Lundeberg, M. B.; Nanot,
S.; Alonso-González, P.; Watanabe, K.; Taniguchi, T.; et al. Electrical Detection of
Hyperbolic Phonon-Polaritons in Heterostructures of Graphene and Boron Nitride. npj
2D Mater. Appl. 2017, No. January, 1 -- 5.
(19) Wang, L.; Chen, X.; Lu, W. Intrinsic Photo-Conductance Triggered by the Plasmonic
Effect in Graphene for Terahertz Detection. Nanotechnology 2016, 27 (3), 035205.
(20) Lundeberg, M. B.; Gao, Y.; Woessner, A.; Tan, C.; Alonso-González, P.; Watanabe, K.;
Taniguchi, T.; Hone, J.; Hillenbrand, R.; Koppens, F. H. L. Thermoelectric Detection and
Imaging of Propagating Graphene Plasmons. Nat. Mater. 2017, 16 (2), 204 -- 207.
(21) Woessner, A.; Gao, Y.; Torre, I.; Lundeberg, M. B.; Tan, C.; Watanabe, K.; Taniguchi, T.;
Hillenbrand, R.; Hone, J.; Polini, M.; et al. Electrical 2π Phase Control of Infrared Light
in a 350-Nm Footprint Using Graphene Plasmons. Nat. Photonics 2017, 11 (7), 421 --
424.
(22) Bissett, M. A.; Tsuji, M.; Ago, H. Strain Engineering the Properties of Graphene and
Other Two-Dimensional Crystals. Phys. Chem. Chem. Phys. 2014, 16 (23), 11124 -- 11138.
(23) Lyu, B.; Li, H.; Jiang, L.; Shan, W.; Hu, C.; Deng, A.; Ying, Z.; Wang, L.; Zhang, Y.; Bechtel,
H. A.; et al. Phonon Polariton-Assisted Infrared Nanoimaging of Local Strain in
Hexagonal Boron Nitride. Nano Lett. 2019, acs.nanolett.8b05166.
(24) Slipchenko, T. M.; Nesterov, M. L.; Hillenbrand, R.; Nikitin, A. Y.; Martín-Moreno, L.
Graphene Plasmon Reflection by Corrugations. ACS Photonics 2017, 4 (12), 3081 -- 3088.
(25) Khestanova, E.; Guinea, F.; Fumagalli, L.; Geim, A. K.; Grigorieva, I. V. Universal Shape
and Pressure inside Bubbles Appearing in van Der Waals Heterostructures. Nat.
Commun. 2016, 7, 1 -- 10.
(26) Tyurnina, A. V.; Bandurin, D. A.; Khestanova, E.; Kravets, V. G.; Koperski, M.; Guinea, F.;
Grigorenko, A. N.; Geim, A. K.; Grigorieva, I. V. Strained Bubbles in van Der Waals
Heterostructures as Local Emitters of Photoluminescence with Adjustable Wavelength.
ACS Photonics 2019, 6 (2), 516 -- 524.
(27) Fei, Z.; Foley, J. J.; Gannett, W.; Liu, M. K.; Dai, S.; Ni, G. X.; Zettl, A.; Fogler, M. M.;
Wiederrecht, G. P.; Gray, S. K.; et al. Ultraconfined Plasmonic Hotspots Inside Graphene
Nanobubbles. Nano Lett. 2016, 16 (12), 7842 -- 7848.
(28) Levy, N.; Burke, S. A.; Meaker, K. L.; Panlasigui, M.; Zettl, A.; Guinea, F.; Neto, A. H. C.;
Crommie, M. F. Strain-Induced Pseudo-Magnetic Fields Greater Than 300 Tesla in
Graphene Nanobubbles. Science (80-. ). 2010, 329 (5991), 544 -- 547.
(29) Haigh, S. J.; Gholinia, A.; Jalil, R.; Romani, S.; Britnell, L.; Elias, D. C.; Novoselov, K. S.;
Ponomarenko, L. A.; Geim, A. K.; Gorbachev, R. Cross-Sectional Imaging of Individual
Layers and Buried Interfaces of Graphene-Based Heterostructures and Superlattices.
Nat. Mater. 2012, 11 (9), 764 -- 767.
(30) Geim, A. K.; Grigorieva, I. V. Van Der Waals Heterostructures. Nature 2013, 499 (7459),
419 -- 425.
(31) Chaste, J.; Missaoui, A.; Huang, S.; Henck, H.; Ben Aziza, Z.; Ferlazzo, L.; Naylor, C.;
Balan, A.; Johnson, A. T. C.; Braive, R.; et al. Intrinsic Properties of Suspended MoS 2 on
SiO 2 /Si Pillar Arrays for Nanomechanics and Optics. ACS Nano 2018, 12 (4), 3235 --
3242.
(32) Lee, J. E.; Ahn, G.; Shim, J.; Lee, Y. S.; Ryu, S. Optical Separation of Mechanical Strain
from Charge Doping in Graphene. Nat. Commun. 2012, 3 (May), 1024.
(33) Mueller, N. S.; Heeg, S.; Alvarez, M. P.; Kusch, P.; Wasserroth, S.; Clark, N.; Schedin, F.;
Parthenios, J.; Papagelis, K.; Galiotis, C.; et al. Evaluating Arbitrary Strain Configurations
and Doping in Graphene with Raman Spectroscopy. 2D Mater. 2017, 5 (1), 015016.
(34) Vincent, T.; Panchal, V.; Booth, T.; Power, S. R.; Jauho, A.; Antonov, V.; Kazakova, O.
Probing the Nanoscale Origin of Strain and Doping in Graphene-HBN Heterostructures.
2D Mater. 2018, 6 (1), 015022.
(35) Frisenda, R.; Navarro-Moratalla, E.; Gant, P.; Pérez De Lara, D.; 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 (1), 53 -- 68.
(36) Cao, Y.; Mishchenko, A.; Yu, G. L.; Khestanova, E.; Rooney, A. P.; Prestat, E.; Kretinin, A.
V.; Blake, P.; Shalom, M. B.; Woods, C.; et al. Quality Heterostructures from Two-
Dimensional Crystals Unstable in Air by Their Assembly in Inert Atmosphere. Nano Lett.
2015, 15 (8), 4914 -- 4921.
(37) Keilmann, F.; Hillenbrand, R. Near-Field Microscopy by Elastic Light Scattering from a
Tip. Philos. Trans. R. Soc. London. Ser. A Math. Phys. Eng. Sci. 2004, 362 (1817), 787 --
805.
(38) Dempster, A. P.; Laird, N. M.; Rubin, D. B. Maximum Likelihood from Incomplete Data
Via the EM Algorithm. J. R. Stat. Soc. Ser. B 1977, 39 (1), 1 -- 22.
(39) Ferrari, A. C.; Basko, D. M. Raman Spectroscopy as a Versatile Tool for Studying the
Properties of Graphene. Nature Nanotechnology. 2013, pp 235 -- 246.
(40) Froehlicher, G.; Berciaud, S. Raman Spectroscopy of Electrochemically Gated Graphene
Transistors: Geometrical Capacitance, Electron-Phonon, Electron-Electron, and
Electron-Defect Scattering. Phys. Rev. B - Condens. Matter Mater. Phys. 2015, 91 (20),
1 -- 17.
(41) Mihai, L. A.; Goriely, A. How to Characterize a Nonlinear Elastic Material? A Review on
Nonlinear Constitutive Parameters in Isotropic Finite Elasticity. Proc. R. Soc. A Math.
Phys. Eng. Sci. 2017, 473 (2207), 20170607.
(42) Yankowitz, M.; Ma, Q.; Jarillo-Herrero, P.; LeRoy, B. J. Van Der Waals Heterostructures
Combining Graphene and Hexagonal Boron Nitride. Nat. Rev. Phys. 2019, 1 (2), 112 --
125.
(43) Kretinin, A. V.; Cao, Y.; Tu, J. S.; Yu, G. L.; Jalil, R.; Novoselov, K. S.; Haigh, S. J.; Gholinia,
A.; Mishchenko, A.; Lozada, M.; et al. Electronic Properties of Graphene Encapsulated
with Different Two-Dimensional Atomic Crystals. Nano Lett. 2014, 14 (6), 3270 -- 3276.
(44) Dean, C. R.; Young, A. F.; Meric, I.; Lee, C.; Wang, L.; Sorgenfrei, S.; Watanabe, K.;
Taniguchi, T.; Kim, P.; Shepard, K. L.; et al. Boron Nitride Substrates for High-Quality
Graphene Electronics. Nat. Nanotechnol. 2010, 5 (10), 722 -- 726.
(45) Yamoah, M. A.; Yang, W.; Pop, E.; Goldhaber-Gordon, D. High Velocity Saturation in
Graphene Encapsulated by Hexagonal Boron Nitride. ACS Nano 2017,
acsnano.7b03878.
(46) Forster, F.; Molina-Sanchez, A.; Engels, S.; Epping, A.; Watanabe, K.; Taniguchi, T.;
Wirtz, L.; Stampfer, C. Dielectric Screening of the Kohn Anomaly of Graphene on
Hexagonal Boron Nitride. Phys. Rev. B - Condens. Matter Mater. Phys. 2013, 88 (8), 1 --
7.
(47) Banszerus, L.; Janssen, H.; Otto, M.; Epping, A.; Taniguchi, T.; Watanabe, K.; Beschoten,
B.; Neumaier, D.; Stampfer, C. Identifying Suitable Substrates for High-Quality
Graphene-Based Heterostructures. 2D Mater. 2017, 4 (2), 025030.
(48) Castro Neto, A. H.; Guinea, F.; Peres, N. M. R.; Novoselov, K. S.; Geim, A. K. The
Electronic Properties of Graphene. Rev. Mod. Phys. 2009, 81 (1), 109 -- 162.
(49) Agarwal, K.; Dai, C.; Joung, D.; Cho, J. Nano-Architecture Driven Plasmonic Field
Enhancement in 3D Graphene Structures. ACS Nano 2019, acsnano.8b08145.
(50) Joung, D.; Nemilentsau, A.; Agarwal, K.; Dai, C.; Liu, C.; Su, Q.; Li, J.; Low, T.; Koester, S.
J.; Cho, J.-H. Self-Assembled Three-Dimensional Graphene-Based Polyhedrons Inducing
Volumetric Light Confinement. Nano Lett. 2017, 17 (3), 1987 -- 1994.
(51) Budrikis, Z.; Sellerio, A. L.; Bertalan, Z.; Zapperi, S. Wrinkle Motifs in Thin Films. Sci. Rep.
2015, 5 (1), 8938.
(52) Shioya, H.; Russo, S.; Yamamoto, M.; Craciun, M. F.; Tarucha, S. Electron States of
Uniaxially Strained Graphene. Nano Lett. 2015, 15 (12), 7943 -- 7948.
(53) Dong, J.; Liu, S.; Fu, Y.; Wang, Q. Investigation of Strain-Induced Modulation on
Electronic Properties of Graphene Field Effect Transistor. Phys. Lett. A 2017, 381 (4),
292 -- 297.
(54) Huang, M.; Yan, H.; Heinz, T. F.; Hone, J. Probing Strain-Induced Electronic Structure
Change in Graphene by Raman Spectroscopy. Nano Lett. 2010, 10 (10), 4074 -- 4079.
(55) Settnes, M.; Power, S. R.; Brandbyge, M.; Jauho, A.-P. Graphene Nanobubbles as Valley
Filters and Beam Splitters. Phys. Rev. Lett. 2016, 117 (27), 276801.
(56) Huang, S.; Yankowitz, M.; Chattrakun, K.; Sandhu, A.; LeRoy, B. J. Evolution of the
Electronic Band Structure of Twisted Bilayer Graphene upon Doping. Sci. Rep. 2017, 7
(1), 7611.
(57) Stauber, T.; Kohler, H. Quasi-Flat Plasmonic Bands in Twisted Bilayer Graphene. Nano
Lett. 2016, 16 (11), 6844 -- 6849.
(58) Stauber, T.; San-Jose, P.; Brey, L. Optical Conductivity, Drude Weight and Plasmons in
Twisted Graphene Bilayers. New J. Phys. 2013, 15 (11), 113050.
(59) Pellegrino, F. M. D.; Angilella, G. G. N.; Pucci, R. Dynamical Polarization of Graphene
under Strain. Phys. Rev. B 2010, 82 (11), 115434.
(60) Dugaev, V. K.; Katsnelson, M. I. Graphene in Periodic Deformation Fields: Dielectric
Screening and Plasmons. Phys. Rev. B 2012, 86 (11), 115405.
(61) Zabel, J.; Nair, R. R.; Ott, A.; Georgiou, T.; Geim, A. K.; Novoselov, K. S.; Casiraghi, C.
Raman Spectroscopy of Graphene and Bilayer under Biaxial Strain: Bubbles and
Balloons. Nano Lett. 2012, 12 (2), 617 -- 621.
(62) Milovanović, S. P.; Covaci, L.; Peeters, F. M. Strain Fields in Graphene Induced by
Nanopillar Mesh. J. Appl. Phys. 2019, 125 (8).
(63) Aubert, S.; Bruyant, A.; Blaize, S.; Bachelot, R.; Lerondel, G.; Hudlet, S.; Royer, P.
Analysis of the Interferometric Effect of the Background Light in Apertureless Scanning
Near-Field Optical Microscopy. J. Opt. Soc. Am. B 2003, 20 (10), 2117.
TABLE OF CONTENTS ONLY
|
1806.10765 | 1 | 1806 | 2018-06-28T04:26:05 | Tribological Properties of Ultrananocrystalline Diamond Nanowire Thin Film: Influence of Sliding Ball Counterbodies | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Ultrananocrystalline Diamond Nanowire (UNCD NW) thin film was deposited on mirror polished silicon substrate (100) using Microwave Plasma Enhanced Chemical Vapor Deposition (MPECVD) System with optimized deposition parameters in CH4 (6%)/N2 plasma media. The film exhibited wire like morphology with randomly oriented and homogeneously distributed ultranano diamond grains separated by an interphase boundary of graphitic and amorphous carbon (a-C) phases. Micro-tribological studies of film were carried out against Al2O3, SiC and steel balls in ambient atmospheric conditions. Initially, the friction coefficient was found to be high for UNCD NW/SiC and UNCD NW/Steel sliding pairs which gradually decreased to low value. While, in UNCD NW/Al2O3 sliding combination, the ultralow value of friction coefficient was maintained throughout the whole sliding process. High wear resistant properties of the film were observed in UNCD NW/SiC and UNCD NW/Steel pairs. In UNCD NW/Al2O3 case, ball counterbody showed negligible wear dimension. Such kind of tribological behavior was attributed to the different type of mechanical and chemical interactions of ball counterbodies with UNCD NW thin film. | physics.app-ph | physics | Tribological Properties of Ultrananocrystalline
Diamond Nanowire Thin Film: Influence of
Sliding Ball Counterbodies
Revati Rani1, N. Kumar1* and I-Nan Lin2
1Materials Science Group, Indira Gandhi Centre for Atomic Research, HBNI Kalpakkam 603102, INDIA
2Department of Physics, Tamkamg University, Tamsui 25137, TAIWAN
substrate
Abstract- Ultrananocrystalline Diamond Nanowire
(UNCD NW) thin film was deposited on mirror polished
(100) using Microwave Plasma
silicon
Enhanced Chemical Vapor Deposition
(MPECVD)
System with optimized deposition parameters in CH4
(6%)/N2 plasma media. The film exhibited wire like
morphology with randomly oriented and homogeneously
distributed ultranano diamond grains separated by an
interphase boundary of graphitic and amorphous carbon
(a-C) phases. Micro-tribological studies of film were
carried out against Al2O3, SiC and steel balls in ambient
atmospheric conditions. Initially, the friction coefficient
was found to be high for UNCD NW/SiC and UNCD
NW/Steel sliding pairs which gradually decreased to low
value. While, in UNCD NW/Al2O3 sliding combination,
the ultralow value of friction coefficient was maintained
throughout the whole sliding process. High wear resistant
properties of the film were observed in UNCD NW/SiC
and UNCD NW/Steel pairs. In UNCD NW/Al2O3 case,
ball counterbody showed negligible wear dimension.
Such kind of tribological behavior was attributed to the
different type of mechanical and chemical interactions of
ball counterbodies with UNCD NW thin film.
I.
INTRODUCTION
Low friction and high wear resistance of thin films
are demanding for sustainable and energy efficient
applications. Carbon based materials are beneficial in
this regard due to their tunable microstructure and
morphology and therefore, it is easy to manipulate film
surface
compatible.
Nanocrystalline/ultrananocrystalline diamond (NCD/
UNCD) thin films show nearly vanishing friction and
high wear resistant properties [1-3].
tribologically
to make
it
There are several factors which may influence the
tribological response of thin films. One among them is
sliding counterbodies [3]. In real tribology applications,
thin films can undergo alternate sliding against
metallic and ceramic counterpart. In this condition, the
wear and chemical structure of sliding interface
becomes
affect overall
tribological efficiency.
complex which may
Therefore, it is important to evaluate the tribological
performance of this film under various counterbody
interactions.
In this study, metallic and ceramic balls were used
to quantify the tribological properties of UNCD NW
film.
II.
EXPERIMENTAL SECTION
A. Film Deposition and Characterization
UNCD NW film was grown on mirror finished
silicon (100) substrate using MPECVD system (2.45
GHz 6'' IPLAS-CYRANNUS). Before the deposition
process, substrate was ultrasonicated in a methanol
solution containing a mixture of nanodiamond powder
(~5 nm) and titanium powder (325 nm) for 45 minutes
to facilitate
the generation of surface sites for
nucleation process. The UNCD NW film was
deposited using CH4
(6%)/N2 plasma with a
microwave power of 1200 W, a pressure of 50 Torr
and a substrate
temperature of 700ºC. Detailed
deposition process is reported elsewhere [1].
examined
The surface topography and roughness of film were
analyzed using an atomic force microscope (AFM,
Park XE-100). The local chemical structure of film
surface was
using Micro-Raman
spectrometer (Andor SR-500i-C-R, wavelength 532
nm). The morphology of film was obtained using field
emission scanning electron microscope (FE-SEM,
Zeiss Supra 55). The hardness and elastic modulus of
film were evaluated by nanoindentation (Triboindenter
TI 950, USA) coupled with Berkovich diamond
indenter with a tip curvature of 150 nm. A maximum
load of 6 mN and a loading-unloading rate of 1.5 mN/
minute were used. To avoid any substrate effect, the
maximum penetration depth of indentation was kept
well below the 1/10th of the film thickness. Rutherford
Back Scattering technique with α particle as projectile
(energy 3800 keV) was used to have elemental
analysis in UNCD NW film. The scattering angle was
kept as 165º.
B. TRIBOLOGICAL TESTING
Tribological testing on UNCD NW thin film has
been carried out using ball-on-disc Micro-tribometer
(CSM-Switzerland) working in linear reciprocating
mode. The tribological properties of UNCD NW film
has been measured while sliding against three different
standard ball counterbodies, i.e. Al2O3, SiC and
100Cr6 steel with diameter 6 mm. The standardized
hardness value of these balls is 18 GPa, 26 GPa and 5
GPa, respectively.
The normal constant load of 2 N, sliding speed of 4
cm/sec was used in each experiment. All tests were
being run for total sliding distance of 500 m which
equals to 62,500 numbers of sliding cycles. The data
acquisition rate of 10 Hz and stroke length of 4 mm
were used for all experiments. The tests were carried
out in ambient (dry and unlubricated conditions).
The wear tracks and ball scars developed after
tribology testing were examined using an optical
microscope. Three-dimensional wear profiles in tracks
were also obtained using 3D optical profilometer
(Taylor Hobson).
III.
RESULTS AND DISCUSSIONS
the diamond
film
Topography of
showed
homogeneous distribution of diamond particles (Fig.
1a). In the AFM image, agglomeration of diamond
particles is not observed. The roughness value of film is
~13 nm. In Raman spectra, a broad peak
is
deconvoluted into five segments; three of them are
related to TPA (ν1, ν2 and ν3). These chemical entities
are the fingerprint of UNCD structure (Fig. 1b) [4]. G
and D band correspond to graphitic and amorphous
diamond phase occupying the grain boundary region.
Nanowire feature of film is indicated in the FESEM
image having high aspect ratio with no agglomeration
(Fig 1c). RBS analysis clearly showed that there are two
distinct RBS lines related to C and Si peak (Fig. 1d). No
other elemental signal was observed which indicates
the compositional purity of film. The elastic Modulus
and hardness values of film are 165 GPa and 14.5 GPa,
respectively.
Fig. 1. AFM image of UNCD NW thin film (a), Raman spectra (b), FESEM image (c), and RBS spectra (d).
The friction results of UNCD NW film sliding
against various counterbody balls are shown in Fig. 2.
Tribotest parameters are mentioned in the inset. In
Fig. 3, two-dimensional (2D) wear track and ball scar
optical images are shown. In addition, 3D optical
image and 2D wear profile of wear track are also
shown in Fig. 3.
Al2O3 ball is chemically inert and thus having weak
affinity towards UNCD NW film during sliding
process. Hence,
the ultralow value of friction
coefficient in the range of 0.02-0.06 is maintained
throughout the whole sliding process of 500 m (Fig.
2). In the response, ball was also less worn out as
shown in Fig. 3a.
observed which gradually decreased to ~0.05 with
sliding distance. The average value of friction
coefficient in this case is 0.13. In the beginning of the
sliding, high friction caused severe wear of ball, scar
size ~200 μm (Fig. 3b). Wear debris are clearly
observed along edge side of the wear track in the
optical image. Conversely, only a few nm of wear
depth is seen in the film.
In the case of UNCD NW/steel sliding combination,
high friction was observed in the beginning of the
sliding cycles which gradually decreased to 0.12. Steel
ball is mechanically soft with the hardness value of 5
GPa compared to film having hardness of 14.5 GPa.
Therefore, due to soft-hard sliding combination, wear
takes place from soft steel ball which develops a
transfer layer on film surface as shown in 3D profile of
Fig 3c. The wear debris can be observed clearly
scattered along edge side in the optical image of wear
track (Fig 3c). Contrastingly, in this case, only few
nanometer of wear loss was observed from the film.
Fig. 2. Friction coefficient of UNCD NW film against Al2O3,
SiC and steel ball.
On the other side, SiC ball is also inert but there is a
possibility of C-C strong bond formation across sliding
interface with UNCD NW film during the sliding
process. Therefore, initially high friction value was
Fig. 3. Wear track and ball scar optical images along with 3D profile of wear track against (a) Al2O3 (b) SiC and (c) Steel ball, respectively.
From the above results, wear resistance of the film is
superior in UNCD NW/SiC and UNCD NW/Steel
sliding combination. However, wear loss from the ball
was significant due to high frictional energy in the
run-in regime. In contrast, UNCD NW/Al2O3 sliding
pair showed negligible wear from the ball counterbody.
This can be possibly due to low friction value in such
sliding combination from the beginning itself. But, deep
wear track was developed in this case that could be
related to mechanical properties of the ball.
IV.
CONCLUSIONS
UNCD NW thin film was deposited on silicon
substrate. Tribological studies were carried out against
three different balls. In the beginning of sliding process,
film friction coefficient against SiC and steel ball
showed high value and it was gradually decreased to
low value 0.05 and 0.12, respectively. However,
ultralow friction value was observed in the sliding
combination of UNCD NW/Al2O3 ball. The wear
resistance of film was high in the case of UNCD
NW/SiC and UNCD NW/Steel sliding pair. However,
wear of the ball was negligible in the case of UNCD
NW/Al2O3 sliding combination.
ACKNOWLEDGMENT
Polaki
The authors would like to acknowledge Mr. Syamala
Rao
Srivastava,
IGCAR/Kalpakkam; Dr. D. Dinesh Kumar, Sathyabama
and Mr. Pankaj Kr. Das,
University/Chennai
and Dr.
S. K.
NIT/Agartala for various experimental help and fruitful
discussions.
REFERENCES
[1] Revati Rani, N. Kumar, A.T. Kozakov, K.A. Googlev, K.J.
Sankaran, P.K. Das, S. Dash, A.K. Tyagi and I-Nan Lin,
"Superlubrication properties of ultrananocrystalline diamond
film sliding against a zirconia ball" RSC Adv., vol. 5, pp.
100663-100673, 2015.
[2] N. Kumar, K. Panda, S. Dash, J.P. Reithmaier, B.K.
Panigrahi, A.K. Tyagi and B. Raj, "Tribological properties of
nanocrystalline diamond films deposited by hot filament
chemical vapor deposition" AIP Adv., vol. 2, pp.
032164-032156, 2012.
[3] K. Panda, N. Kumar, K.J. Sankaran, B.K. Panigrahi, S. Dash,
H-C. Chen, I-Nan Lin, N-H. Tai and A.K. Tyagi,
"Tribological properties of ultrananocrystalline diamond and
diamond nanorod films" Surf & Coat Technol., vol. 207, pp.
535-545, 2012.
[4] R. Pfeiffer, H. Kuzmany, P. Knoll, S. Bokova, N. Salk and B.
Gunther,
in
nano-crystalline diamond films" Diamond and Related
Mater., vol. 12, pp. 268-271, 2003.
trans-polyacetylene
"Evidence
for
E-mail of the author(s): *[email protected],
[email protected]
|
1910.02024 | 1 | 1910 | 2019-10-04T16:30:28 | Improvement of temperature uniformity of induction-heated T-shape susceptor for high-temperature MOVPE | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | The induction heating is a common method applied in metalorganic vapor phase epitaxy (MOVPE) especially for higher-temperature growth conditions. However, compared to the susceptor heated by the multiple-zone resistant heater, the inductive-heated susceptor could suffer from severe thermal non-uniformity issue. In this simulation study, we propose to employ a T-shape susceptor design with various geometric modifications to significantly improve the substrate temperature uniformity by manipulating thermal transfer. Specifically, the thermal profile can be tailored by horizontal expansion and vertical elongation of the susceptor, or forming a cylindrical hollow structure at the susceptor bottom cylinder. Three optimized designs are shown with different temperature uniformity as well as various induction heating efficiencies. The temperature variation of the entire substrate surface can be less than 5 {\deg}C at ~1900 {\deg}C with high induction heating efficiency after applying the proposed techniques. | physics.app-ph | physics | K.-H. Li et al.
Improvement of temperature uniformity of induction-heated
T-shape susceptor for high-temperature MOVPE
Kuang-Hui Li, Hamad S. Alotaibi, Xiaohang Li*
King Abdullah University of Science and Technology (KAUST)
Advanced Semiconductor Laboratory
Thuwal, 23955-6900, Saudi Arabia
*Corresponding Author: [email protected]
ABSTRACT
The induction heating is a common method applied in metalorganic vapor phase epitaxy (MOVPE)
especially for higher-temperature growth conditions. However, compared to the susceptor heated
by the multiple-zone resistant heater, the inductive-heated susceptor could suffer from severe
thermal non-uniformity issue. In this simulation study, we propose to employ a T-shape susceptor
design with various geometric modifications to significantly improve the substrate temperature
uniformity by manipulating thermal transfer. Specifically, the thermal profile can be tailored by
horizontal expansion and vertical elongation of the susceptor, or forming a cylindrical hollow
structure at the susceptor bottom cylinder. Three optimized designs are shown with different
temperature uniformity as well as various induction heating efficiencies. The temperature variation
of the entire substrate surface can be less than 5 °C at ~1900 °C with high induction heating
efficiency after applying the proposed techniques.
Keywords: A1. Computer Simulation; A1. Heat Transfer; A3. Metalorganic chemical vapor
deposition processes; B1. Nitrides; B2. Semiconducting aluminum compounds
1. Introduction
Ultra-wide bandgap III-nitride materials including AlN, BN, and their alloys with other
group-III elements are promising for optoelectronics and power electronics applications.1,2 High
material quality of these alloys is essential for device performance and investigation of material
properties. However, it has been challenging to realize it especially on commercially-viable
K.-H. Li et al.
foreign substrates including sapphire and silicon primary due to large lattice mismatch. The
metalorganic vapor phase epitaxy (MOVPE) is the most common method for growing III-nitride
materials. To improve quality of the ultra-wide bandgap III-nitride materials such as AlN, MOVPE
growers have employed various methods including precursor pulsing to enhance adatom
movement and patterned substrates to leverage lateral coalescence.3 Another notable method is to
apply extremely high temperature (EHT) ( >1600 °C) to enhance adatom mobility and suppress
undesirable impurity incorporation, which has led to greatly improved material quality.4,5
However, existing commercial MOVPE systems with the resistant heater may not be
suitable for long-term and low-cost EHT operation. The filament of the resistant heater is usually
made of refractory metals such as tantalum (Ta), tungsten (W), rhenium (Re), or their alloys which
can sustain high temperature while emitting thermal radiation. Ideally, the heated susceptor can
absorb all the thermal radiation and reach thermal equilibrium temperature according to its
emissivity and absorptivity ratio according to the Kirchhoff's law of thermal radiation.6 However,
in reality the susceptor both absorbs and reflects thermal radiation. The susceptor reflects even
more thermal radiation at higher temperatures. Therefore, the filament usually needs to be several
hundred degrees higher than the susceptor target temperature; otherwise, the susceptor could not
reach the target temperature. At higher temperatures, the filament has thermal expansion and can
cause warpage and possibly short circuit. Moreover, though refractory metals have high melting
point, extreme heating-and-cooling cycles can cause thermal fracture and may break the filament
eventually.7
Compared to the resistant heater, the induction heater has several advantages due to a
different working principle. The induction coil generates alternating magnetic field and the
susceptor induces Eddy current accordingly, i.e. inductive coupling. The Eddy current causes the
Joule heating effect on the susceptor and heats up the susceptor. Unlike the resistant heater, such
energy transfer mechanism is independent of temperature, meaning there is no thermal radiation
reflection, warpage, or lifetime issue. That's the reason for the induction heater to have good
heating efficiency and higher reliability than the resistant heater. Nevertheless, the induction heater
also has drawbacks. The inductive coupling efficiency between the induction coil and the susceptor
is affected by induction coil geometry, susceptor geometry, susceptor material, and frequency of
AC power load. 8 Furthermore, the induction heater could result in severe temperature non-
K.-H. Li et al.
uniformity for conventional column-shape susceptors as compared to the resistant heater. Large
temperature non-uniformity can cause problems because it affects metalorganic compound
pyrolysis efficiency, material composition, growth rate, adatom mobility, and wafer curvature.
The resistant heater users can apply the multi-zone technique 9,10 to tune the substrate
temperature uniformity. However, it is difficult to apply the multi-zone technique for the induction
heater. There have been studies that propose techniques to improve substrate temperature
uniformity of the induction-heated susceptor. But these techniques are often complicated or not
applicable for EHT.11-16 In the previous report, an MOVPE reactor design was proposed by having
the induction coil placed around the bottom cylinder under the top plate of a T-shape susceptor as
shown in Figure 1(a).17 Hence, the magnetic field is nearly fully coupled to the susceptor to greatly
improve induction heater efficiency at EHT and allow the use of small susceptor-gas inlet distance
because of the magnetic shielding effect of the T-shape susceptor.17 Thus, the proposed reactor
could reach higher temperatures and possess lower parasitic reaction rates for Al- and B-containing
metalorganic precursors which are desirable for MOVPE processes of the ultra-wide bandgap III-
nitride materials. Despite these technical advantages, the T-shape susceptor also suffers from the
temperature non-uniformity issue that ought to be addressed for growing high quality and uniform
epitaxial wafers.
In this work, the substrate temperature uniformity of the T-shape susceptor has been
studied. The substrate temperature uniformity can be improved via controlling the heat transfer
path by vertically elongating the susceptor, horizontally expanding the susceptor, or forming a
cylindrical hollow structure on the bottom of the susceptor, as shown in Figure 1(b). The induction
heating efficiency has also been investigated.
K.-H. Li et al.
Figure 1 (a) The reference T-shape susceptor with set dimensions of Ao = 1.8 cm, Bo = 1 cm, Co =
5 cm, and Fo = 2.5 cm (one inch). (b) The optimized T-shape susceptor with geometric variables.
Both susceptors can accommodate a two-inch susceptor. Variables R and D in (b) are the radius
and depth of the cylindrical hollow structure, respectively. Variable E is the radius increase of the
susceptor and variable L is the height increase of the bottom cylinder. The black arrows indicate
the heat transfer path.
2. Materials and Methods
The heat transfer study was carried out using the cylindrical symmetry due to the geometric
of the susceptors. The susceptors were assumed to accommodate a two-inch substrate. Similar
works can be conducted on the larger susceptors and thus they are not included in this report. The
reference T-shape susceptor in Figure 1(a) is a solid piece comprising a top plate and a bottom
cylinder with set dimensions where the bottom cylinder is directly below the two-inch substrate
pocket. The optimized T-shape susceptor in Figure 1(b) is similar to the reference T-shape
susceptor in terms of the overall shape but have four geometric dimension variables: R and D are
radius and height of the cylindrical hollow structure within the bottom cylinder; E is the radial
increase of the top plate and the bottom cylinder; and L is the height increase of the bottom cylinder.
The position of induction coil can influence the induction coupling. For the reference and
(a)AoBoCoAo+EBoCo+LFo+ED(b)RSusceptor earFoFoFoTop plateBottomcylinder K.-H. Li et al.
optimized T-shape susceptors, the edge of the lowest coil always align with the bottom surface of
the bottom cylinder.
The numerical analysis were conducted by using the finite element analysis of the
COMSOL Multiphysics 4.3a. There were139,656 triangular elements and 191,043 degrees of
freedom included in the mesh. Heat transfer by the induction heating, conduction, and thermal
radiation was calculated by the build-in models.18 In this study, the frequency was fixed at 10 kHz.
Most EHT susceptors are made of (crystalline or amorphous) graphite coated by silicon carbide
(SiC) or tantalum carbide (TaC). However, in this work, the proposed T-shape susceptor is made
of tungsten (W) due to its low cost, high melting point, and superior isotropic electric conductivity
and thermal conductivity. TaC is an excellent material for induction-heated high temperature
applications but TaC has higher cost than graphite and W. The crystalline graphite has in-plane
and out-of-plane lattice planes, which makes it an anisotropic material.19,20,21 The in-plane electric
and thermal conductivities are similar to the tungsten's, depending on the quality of graphite.
However, the out-of-plane electric and thermal conductivities are inferior to the tungsten's. Such
anisotropic properties affect inductive coupling efficiency and heat transfer. The amorphous
graphite is a porous material with poor electric and thermal conductivity. It can lead to poor
inductive coupling efficiency that is detrimental for heating efficiency at EHT. Hence, tungsten is
a good candidate for inductively-heated high-temperature and low-cost susceptors. All the physical
quantities required in the simulation can be found in the previous report17 and CRC Handbook of
Chemistry and Physics. 22 For tungsten susceptor, the parameters of resistivity are 𝜌𝑊 =
𝜌𝑊0[1 + 𝛼𝑊(𝑇 − 𝑇0)], where T0 = 273 K, W = 5.7×10-3 K-1, and W0 = 4.63×10-8 Ω-m; the
parameters of thermal conductivity are 𝑘𝑊 =
1
𝐴𝑊+𝐵𝑊 (𝑇−𝑇0)
, where AW = 6.2×10-3 m-K/W and BW
= 3×10-6 m/W. For copper coil, the parameters of resistivity are 𝜌𝐶𝑢 = 𝜌𝐶𝑢0[1 + 𝛼𝐶𝑢(𝑇 − 𝑇0)],
where T0 = 273 K, Cu = 4.68×10-3 K-1, and Cu0 = 1.52×10-8 Ω-m; the parameters of thermal
conductivity are 𝑘𝐶𝑢 =
𝐴𝐶𝑢+𝐵𝐶𝑢 (𝑇−𝑇0)
, where ACu = 2.5×10-3 m-K/W and BCu = 5×10-7 m/W. For
molybdenum supporter, the parameters of resistivity are 𝜌𝑀𝑜 = 𝜌𝑀𝑜0[1 + 𝛼𝑀𝑜(𝑇 − 𝑇0)], where
T0 = 273 K, Mo = 5.42×10-3 K-1, and Mo0 = 4.78×10-8 Ω-m; the parameters of thermal
1
1
conductivity are 𝑘𝑀𝑜 =
𝐴𝑀𝑜+𝐵𝑀𝑜 (𝑇−𝑇0)
, where AMo = 7.4×10-3 m-K/W and BMo = 2×10-6 m/W. For
Stainless steel showerhead and bottom flange, the parameters of resistivity are 𝜌𝑆𝑆 =
K.-H. Li et al.
𝜌𝑆𝑆0[1 + 𝛼𝑆𝑆(𝑇 − 𝑇0)], where T0 = 273 K, SS = 5.84×10-4 K-1, and SS0 = 7.5×10-7 Ω-m; the
parameters of thermal conductivity are 𝑘𝑆𝑆 =
𝐴𝑆𝑆+𝐵𝑆𝑆 (𝑇−𝑇0)+𝐶𝑆𝑆 (𝑇−𝑇0)2, where ASS = 6.8×10-2 m-
K/W, BSS = −5×10-5 m/W, and CSS = 2×10-8 m/W-K. For the zirconium oxide thermal insulator,
1
the parameters of thermal conductivity are 𝑘𝑍𝑟𝑂2 =
𝐴𝑍𝑟𝑂2+𝐵𝑍𝑟𝑂2 (𝑇−𝑇0)+𝐶𝑍𝑟𝑂2 (𝑇−𝑇0)2, where AZrO2
= 1.38×10-1 m-K/W, BZrO2 = 2×10-4 m/W, and CZrO2 = −6×10-8 m/W-K. The emissivity of polished
1
metal is around 0.01 to 0.05 and zirconium oxide thermal insulator is 0.95 at room temperature;
however, the emissivity of the metals rises to 0.2 at high temperature. For simplicity, the emissivity
of the metals is fixed at 0.2 in the simulation. Sapphire is chosen as the substrate in the simulation.
Sapphire has the lowest thermal conductivity (~25 W/m-K) among a few common high-melting-
point substrates. High thermal conductivity substrate such as SiC (~360 W/m-K) has better
temperature uniformity than sapphire. Sapphire is the worst-case scenario for temperature
uniformity. Other substrate won't have temperature uniformity problem if sapphire can achieve
temperature uniformity by the techniques in this study. The reactor pressure is kept at 50 Torr
close to the ones used to grow AlN in a common MOVPE process today. The gas flow is not
considered due to negligible impact on the susceptor temperature. The T-shape susceptor rotation
is not included in the simulation since the T-shape susceptor and the induction coil are both
cylindrically symmetric. Rotation neither affects the induction coupling efficiency nor changes the
substrate temperature profile.
The temperature distribution on the induction-heated susceptor depends on heat transfer.
For the induction heating modeling, the governing equation is:
[∇2 + 𝜇0𝜇𝑟(𝜖0𝜖𝑟𝜔2 − 𝑖𝜎𝜔)]𝐴⃗ = 𝜇0𝜇𝑟(𝜎 + 𝑖𝜖0𝜖𝑟𝜔) 𝑉𝑐𝑜𝑖𝑙
2𝜋𝑅
𝜙,
(1)
where 𝑖 is imaginary number, 𝜎 is the electrical conductivity, 𝜔 is angular frequency of alternating
current, 𝜂 is resistivity of material, 𝜖0 is electrical permeability at free space, 𝜖𝑟 is relative
electrical permittivity, 𝐴⃗ is magnetic vector potential, 𝜇0 is magnetic permeability at free space,
and 𝜇𝑟 is relative magnetic permeability. The induction coil was modeled as torus shape; therefore,
𝜙, where 𝑅, 𝑉𝑐𝑜𝑖𝑙, and 𝜙 are the radius of the
the electrical field of the induction coil is ∇⃗⃗⃗𝑉 =
𝑉𝑐𝑜𝑖𝑙
2𝜋𝑅
induction coil, the electric potential, and the unit vector, respectively.
For thermal conduction modeling, the governing equation is:
K.-H. Li et al.
𝜌𝐶𝑃
𝜕𝑇
𝜕𝑡
+ 𝜌𝐶𝑃𝑢⃗⃗ ∙ 𝛻⃗⃗𝑇 = 𝛻⃗⃗ ∙ (𝑘𝛻⃗⃗𝑇) + 𝑄,
(2)
where 𝜌 is density, 𝐶𝑃 is the specific heat capacity at a constant pressure, 𝑇 is absolute temperature,
𝑡 is time, 𝑢⃗⃗ is velocity vector, 𝑘 is thermal conductivity, and 𝑄 =
Re(𝐽⃗ ∙ 𝐸⃗⃗) is the power
1
2
generated by the Eddy current.
For thermal radiation modeling, the governing equations are:
𝐸𝑏(𝑇) = 𝜀𝜎𝑇4,
(3)
(1 − 𝜀)𝐺 = 𝐽 − 𝐸𝑏(𝑇),
(4)
−𝑛⃗⃗ ∙ 𝑞⃗ = 𝐺 − 𝐽,
(5)
where 𝜎 is the Stefan-Boltzmann constant, 𝐸𝑏(𝑇) is the blackbody hemispherical total emissive
power, 𝜀 is the emissivity of the material, 𝐺 is incoming radiative heat flux, 𝐽 is the total outgoing
radiative heat flux, 𝑛⃗⃗ is the normal unit vector on the boundary, and 𝑞⃗ is the radiation heat flux
vector.
From the simulation results, the radius and the depth of the hollow structure, and horizontal
expansion and vertical elongation of the susceptor can significantly influence the substrate
temperature uniformity. The mechanism and optimized parameters for the T-shape susceptor will
be discussed thoroughly in the next sections.
3. Results and discussion
To quantify the substrate temperature uniformity, we define a value called the Uniformity
Length (UL) as the distance from the substrate center to the farthest point within which the
substrate surface temperature variation is equal or less than 5 °C. Thus, the larger the UL, the
better the substrate temperature uniformity is. For a two-inch substrate, the maximum UL in the
ideal situation is one inch where the temperature difference of the entire substrate is less than 5 °C.
Although temperature variation on a wafer in a state-of-the-art MOVPE reactor could be less than
5 °C at lower growth temperatures such as ~1000 °C for InGaN light emitters, the threshold of 5
°C was reasonable because the target is EHT in this study. In other words, a 5 °C difference
represented a very small, if not negligible temperature non-uniformity budget at EHT. Figure 2(a)
and 2(b) include a convex and a camel-back temperature line profile, respectively, where the UL's
K.-H. Li et al.
are shown as examples. In the following sections, the simulation results related to Figure 3, Figure
5, and Figure 6 have the same induction heating power of 6.5 kW, but the substrate average
temperature is not the same. It is because heat transfer path and induction heating coupling are
geometry dependent.
Figure 2 (a) and (b) show the UL's with convex and camel-back substrate surface temperature line
profiles, respectively. The zero on the x-axis represents the substrate center.
3.1 Temperature profile of the reference T-shape susceptor
The cross-sectional temperature profile of the reference T-shape susceptor in the reactor is
shown in Figure 3(a). Due to the reactor's axial symmetry, only half of the cross section is shown.
The detail reactor configuration can be find elsewhere.17 The bottom cylinder temperature is higher
than the top plate temperature, because the heat transfer is mainly from the bottom cylinder to the
top plate. To keep such heat transfer path, the heat transferred downward to the susceptor supporter
(made of Molybdenum) and heat released by thermal radiation have to be reduced. Otherwise,
these heat sinks will affect the heating efficiency of the susceptor. To reduce the heat sink, several
thermal insulators (made of zirconia)23 are placed on the lateral and bottom sides of the bottom
cylinder. The lateral thermal insulator not only reduces thermal radiation, but also protects the
induction coil from thermal radiation. The bottom thermal insulator blocks the heat transferring
downward to the susceptor supporter.
18401845185018551860-2.5-2-1.5-1-0.500.511.522.5Temperature ( C)2" substrate (cm)ΔT ~ 5 CUniformity Length18451850185518601865187018751880-2.5-2-1.5-1-0.500.511.522.5Temperature ( C)2" substrate (cm)ΔT ~ 5 CUniformity Length(a)(b) K.-H. Li et al.
Figure 3 (a) The cross-sectional temperature profile of half of the reference T-shape susceptor in
the reactor. (b) The substrate line temperature line profile of the two-inch substrate measured
from the blue dash line in (a) with Tcenter of 1987 °C.
The substrate temperature line profile is shown in Figure 3(b). The average substrate
temperature (Taverage) is 1972 °C with the standard deviation (𝜎) of 13.5 °C. The temperature
difference (ΔT = Tcenter − Tedge) between the susceptor center (Tcenter) and the susceptor edge (Tedge)
is as large as 45 °C which is not acceptable. The red curve of Figure 4 shows that Figure 4ΔT a
quadratic function of Tcenter. When Tcenter is 1000 C, ΔT is ~5 C which is still acceptable. However,
ΔT rapidly increases to over 25 C above the EHT, suggested that the reference T-shape susceptor
design be modified to be applicable for acceptable uniformity at EHT.
To develop techniques improving the substrate temperature uniformity, understanding the
induction heating mechanism and the heat transfer in the T-shape susceptor is important. Based on
classical electrodynamics, EM waves only reach a certain depth below a conductor surface and the
depth is defined as the skin depth (𝛿), which can be calculated by the following formula,24, 25, 26
𝛿 = √
𝜌𝜂
𝜋𝑓𝜇𝑜𝜇𝑟
√√1 + (2𝜋𝑓𝜌𝜂𝜖0𝜖𝑟)
2
+ 2𝜋𝑓𝜌𝜂𝜖0𝜖𝑟, (6)
where 𝑓 is the frequency of the alternating current, 𝜖0 is the electrical permeability at free space,
𝜖𝑟 is the relative electrical permittivity, 𝜇0 is the magnetic permeability at free space, 𝜇𝑟 is the
relative magnetic permeability, and 𝜌𝜂 is the resistivity of the conductor at temperature 𝜂. Because
200015001000500Induction coilT-shaped susceptorThermal insulatorThermal insulatorGas inletDouble-walled quartzSusceptor supporterEmpty(a)(°C)0510152025303540451943194819531958196319681973197819831988-2.5-2-1.5-1-0.500.511.522.5ΔT ( C)Substrate temperature ( C)2" substrate (cm)2" substrate temperature profile(b)2.5 cmSubstrate temperature line profile K.-H. Li et al.
2𝜋𝑓𝜌𝜂𝜖0𝜖𝑟 is a small quantity for common induction heaters frequency and metal, Equation (6)
can be further reduced to:
𝛿(𝜂) ≈ √
𝜌𝜂
𝜋𝑓𝜇0𝜇𝑟
≈ 503.29√
𝜌0[1+𝛼(𝜂−𝜂0)]
𝑓
,
(7)
where 𝜌0 is the reference resistivity of the conductor at temperature 𝜂0 (300K) and 𝛼 is the
temperature coefficient of resistivity. Tungsten is a paramagnetic material which can be consider
as low-level magnetization, and paramagnetic property can be described by the Curie's Law.
Therefore without large deviation, 𝜇𝑟 can be approximately considered as one. From Equation (7),
𝛿 depends on the electrical properties of the conductor, the frequency of the induction heater, and
the temperature of the susceptor.
Figure 4 (Red curve) The temperature difference (ΔT) between Tcenter and Tedge of the reference T-
shape susceptor as a function of Taverage. (Black curve) Skin depth as a function of the temperature
η.
012345051015202530354045025050075010001250150017502000Skin depth δ(η) (mm)ΔT(Tcenter) ( C)Tcenteror η( C)ΔT(Taverage) ( C)Taverageor η( C) K.-H. Li et al.
The skin depth vs temperature 𝜂 was further calculated and shown in Figure 4. The skin
depth gradually increases from 1.1 to 4 mm when temperature increases from 100 to 2000 °C. The
magnetic field intensity decays exponentially when the magnetic field penetrates into the T-shape
susceptor (~𝑒 𝑥/𝛿 ).25,26 When the magnetic field penetrates one skin depth distance from the
surface, the intensity decays to 36.7% (~𝑒−1), and decays to 13.5% (~𝑒−2) and 4.9% (~𝑒−3) when
the penetration distances are two and three skin depth, respectively. Therefore within the distance
of three skin depths from the surface, the bottom cylinder will induce most of the Eddy current
and generate an internal magnetic field against the external magnetic field by the Faraday-Lenz
law of induction.27 The Eddy current encircles the bottom cylinder and generates heat by the Joule-
Lenz law. The skin depth at 1900 °C is 4 mm from Figure 4, which means that from the bottom
cylinder surface to 1.2 cm below, the Eddy current will be induced to generate heat. Once the
generated heat transfers to the top plate surface, the temperature of the outer region of the bottom
cylinder near the surface is lower than that of the inner region. It is because the outer region will
release heat by radiating thermal radiation and conducting to the thermal insulator. Furthermore,
when the heat approaches the top plate surface, part of the heat goes to the susceptor ear (Figure
1), making the outer region of the bottom cylinder release more heat.
The heat transfer behaviors explain that the two-inch substrate has higher Tcenter and lower
Tedge [Figure 3(b)], and the temperature difference between the center and the edge increases as
the average temperature goes higher. The substrate temperature line profile [Figure 3(b)] has a UL
of 0.92 cm, which corresponding to 13.5% temperature uniformity on the substrate surface
[ (
0.92 𝑐𝑚
2.5 𝑐𝑚
2
)
~13.5% ]. Such uniformity is not acceptable. However, it can be improved by
geometric modification techniques in Section 3.2.
3.2 Impacts of geometric options on substrate temperature uniformity
In this section, out of the geometric options, i.e. the formation of the hollow structure (R
and D), the radius increases of the top plate and the bottom cylinder (E), and the bottom cylinder
elongation (L), only one is implemented at one time while others are the same as the reference
susceptor. After the impact of each option is known, it helps further improving the substrate
temperature uniformity when multiple variables are involved (Section 3.3). Figure 5(b) presents
the substrate temperature line profile at different R and D values, while keeps E and L zero. Figure
K.-H. Li et al.
5(c) exhibits the substrate temperature line profile evolution at various E values, while R, D, and
L are zero. Figure 5(d) shows the impact of L while R, D, and E are zero. The three options are
found to significantly impact the heat transfer path and the substrate temperature uniformity.
In Figure 5(b), the substrate temperature line profile shifts downward when R and D
increase. Meanwhile, the substrate temperature uniformity gradually improves. For instance, when
R is 2 cm and D is 5 cm (blue curve) shows superior substrate temperature uniformity to red and
green curves. The explanation is that when there is a hollow structure in the bottom cylinder, the
heat transfer is not simply from the entire bottom cylinder to the top plate, since the inner region
of the bottom cylinder is empty, as shown in Figure 1(b). When R and D increase, more heat starts
to transfer from the side of the bottom cylinder to the center of the bottom cylinder. Such a shift
of the heat transfer results in the substrate temperature drop at the center. Since the substrate
temperature at the center was higher than the substrate temperature on the edge, the substrate
temperature uniformity can be improved by forming the hollow structure.
In Figure 5(c), the substrate temperature line profile evolves with different values of E. The
substrate temperature line profile shifts downward with increasing E, while the temperature
difference between the center and the edge reduces and achieves acceptable substrate temperature
uniformity when E is 1.5 cm. In Section 3.1, the substrate temperature difference between the
substrate center and edge can be explained by different heat transfer paths between the bottom
cylinder inner and outer regions. Here, a larger diameter keeps the bottom cylinder inner region
away from the outer region. Hence, the heat transferring to the top plate surface is mainly from the
bottom cylinder inner region. Since the temperature in the bottom cylinder inner region is much
uniform than that in the bottom cylinder outer region, the temperature uniformity is improved.
In Figure 5(d), interestingly, the substrate temperature line profile shifts downward without
alternating its shape. Also, the profile drops almost linearly (about 30 °C for every 0.5 cm increase
in L). This is because the vertical elongation does not change the heat transfer path. The bottom
cylinder outer region still has a faster heat lost rate than that of the bottom cylinder inner region.
Furthermore, since the induction coil also shifts downward amid the vertical elongation, the heat
generated in the bottom cylinder is even farther away from the top plate surface, which makes the
heat transfer path longer. Therefore, the substrate temperature line profile shifts downward without
improvement or deterioration of the substrate temperature uniformity. By studying how the
K.-H. Li et al.
temperature line profile is affected by R, D, E, and L, it is obvious that R, D, and E can improve
the uniformity greatly if appropriate R, D, and E are chosen. On the other hand, L is not useful.
Figure 5 (a) The cross-sectional temperature profile of half of T-shape susceptor in the reactor.
The vertical yellow dash line shows the original radius of the bottom cylinder. The horizontal
yellow dash line shows the original bottom edge of the bottom cylinder before elongation. The red
solid line on the up-right corner is a scale bar. (b), (c), and (d) show different substrate temperature
line profiles by adjusting R and D, E, and L, respectively, while fixing the other variables as shown
on the top of each figure.
The current section (3.2) discusses the impacts of the three geometric options. But they are
not fully optimized even though some examples in Figure 5 show better substrate temperature
uniformity. To further improve the substrate temperature uniformity, R, D, and E have to be
optimized according to the temperature standard deviation, the UL, and the susceptor volume. In
1820183018401850186018701880189019001910-2.5-2-1.5-1-0.500.511.522.5Substrate temperature ( C)2" substrate (cm)2" substrate temperature profileR = 1 cm D = 3 cmR = 1.5 cm D = 4 cmR = 2 cm D = 5 cm190019051910191519201925-2.5-2-1.5-1-0.500.511.522.5Substrate temperature ( C)2" substrate (cm)2" substrate temperature profileE = 0.5 cmE = 1 cmE = 1.5 cm200015001000500(°C)ERD2.5 cmF0C0L17601780180018201840186018801900-2.5-2-1.5-1-0.500.511.522.5Substrate temperature ( C)2" substrate (cm)2" substrate temperature profileL = 0.5 cmL = 1 cmL = 1.5 cm(a)(b)(c)(d)R = 1 cm D = 3 cmR = 1.5 cm D = 4 cmR = 2 cm D = 5 cmL = 0.5 cmL = 1 cmL = 1.5 cmE = 0.5 cmE = 1 cmE = 1.5 cmE = 0 cm L = 0cmR = 0 cm D = 0 cm E = 0 cmR = 0 cm D = 0 cm L = 0 cm K.-H. Li et al.
Section 3.3, three optimized designs (Designs 1-3) are shown and discussed with their own benefits
and drawbacks.
3.3 Comparison between three optimized designs
The optimization follows one major rule: it has to keep the UL as large as possible. The
maximum is 2.5 cm. On top of that, if there are multiple optimized setups which leads to the same
UL, the susceptor volume is preferably smaller, which is related to heating efficiency at the EHT.
It is important to note that the difference in the required induction power may not be large in this
study between the two-inch substrate susceptors. But it would be expectedly significant for larger
susceptors particularly the ones used for production. The substrate temperature line profile of
Design 1 is shown in Figure 6(b). It is apparent that the substrate temperature uniformity matches
the criteria (ΔT ≤ 5 °C). Its substrate average temperature is 1907 °C, the temperature standard
deviation is 1.0 °C, and the UL is 2.5 cm covering the entire substrate. For Design 2 [Figure 6(d)],
its substrate temperature uniformity is not as good as Design 1. Design 2 has average substrate
temperature of 1835 °C, the temperature standard deviation of 3.1 °C, and the UL of 2.1 cm
covering 71% of the substrate surface. The reduced UL is caused by an unfavorable substrate
temperature drop (~ 13 °C) near the substrate edge. For Design 3 [Figure 6(e)], it has the substrate
average temperature of 1837 °C, the temperature standard deviation of 1.4 °C, and the UL of 2.5
cm covering the entire substrate.
Through optimizing horizontal expansion only, Design 1 seems to be better than Design 2
optimizing the hollow structure only. However, there is a drawback. The substrate temperature
uniformity is improved by increasing the radius versus the reference substrate. This causes the
volume of Design 1 is roughly twice larger than that of the reference substrate. Thus, Design 1
requires more induction power. For instance, Design 1 requires 6.4 kW at 1750 °C while the
reference susceptor needs only 4.4 kW [Figure 7(a)]. The uniformity of Design 2 is poorer.
However, it has the closest heating efficiency to that of the reference substrate. For Design 1 and
Design 2, there is a tradeoff between the temperature uniformity and the heating efficiency. It is
possible to have a compromised design, which means excellent temperature uniformity and high
heating efficiency, i.e. Design 3. Design 3 leads to the same UL as Design 1. Since Design 3 has
larger standard deviation than Design 1's, Design 1's temperature uniformity is better. But Design
3 has considerably higher heating efficiency than that of Design 1, as shown in Figure 7(b).
K.-H. Li et al.
Figure 6 (a), (c), and (e) are the temperature profiles of the optimized T-shape susceptors Design
1 -- 3, respectively. The unit of the numbers is cm. (b), (d), and (f) are the substrate temperature line
profiles of Design 1 -- 3, respectively.
181418191824182918341839184418491854-2.5-2-1.5-1-0.500.511.522.5Substrate temperature ( C)2" substrate (cm)Design 2 temperature profile200015001000500(°C)1.25200015001000500(°C)24.75200015001000500(°C)24.3750.5(a)(c)(e)191719221927193219371942194719521957-2.5-2-1.5-1-0.500.511.522.5Substrate temperature ( C)2" substrate (cm)Design 3 temperature profile2.5 cm2.5 cm2.5 cmDesign 1Design 2Design 3(d)181718221827183218371842184718521857-2.5-2-1.5-1-0.500.511.522.5Substrate temperature ( C)2" substrate (cm)Design 3 temperature profile(f)R = 2 cm D = 4.75 cmE = 0 cm L = 0 cmR = 2 cm D = 4.375 cmE = 0.5 cm L = 0 cm188718921897190219071912191719221927-2.5-2-1.5-1-0.500.511.522.5Substrate temperature ( C)2" substrate (cm)Design 1 temperature profile(b)R = 0 cm D = 0 cmE = 1.25 cm L = 0 cm K.-H. Li et al.
Figure 7 (a) The substrate average temperature of different designs as a function of the induction
power, indicating various heating efficiencies. (b) The UL and the standard deviation of the
substrate temperature as a function of the substrate average temperature.
The UL and the standard deviation of the substrate temperature as a function of the
substrate average temperature of the three designs are shown in Figure 7(a). For Design 1 and
Design 3, the UL is 2.5 cm amid the entire temperature range indicating their excellent candidacy
for the temperature uniformity. For Design 2, the UL is 2.5 cm until reaching temperatures over
~1500 °C, which means that it is perfect for lower temperatures but not good for the EHT. The UL
of the reference susceptor starts to decrease at 900 °C which is even lower than the conventional
growth temperature of GaN (~1000 °C). The standard deviation increases quadratically versus the
average temperature and largely reflects the same phenomena as the UL does. There is a correlation
between the UL and the temperature standard deviation: once the substrate temperature standard
deviation goes beyond ~1.6 °C, the UL starts to drop. The explanation is that if taking substrate
temperature line profile as a Laplace-Gauss distribution, 3σ covers 99.7% of the data points. To
match acceptable substrate temperature uniformity, 3σ should be equal to or less than 5 °C (3σ ≤
5 °C), which gives the result of σ ≤ 1.67 °C.
4 Conclusion
In summary, the T-shape susceptor is a candidate for high temperature MOVPE processes but can
suffer severe temperature non-uniformity issues. In this study, it is found that the modifications of
0250500750100012501500175020000123456789Substrate temperature ( C)Induction power (kW)Design 0Design 1Design 2Design 3(a)Reference0250500750100012501500175020000123456789Substrate average temperature ( C)Induction power (kW)Reference Design 1 Design 2 Design3(b)0246810121400.511.522.53025050075010001250150017502000Standard deviation σ( C)Uniformity Length (cm)Substrate average temperature ( C)ULmax=2.5 cmReference:No hollow structure [Fig. 1(a)]Design 1:R 0 cm D 0 cm E 1.25 cm L 0 cmDesign 2:R 2 cm D 4.75 cm E 0 cm L 0 cmDesign 3:R 2 cm D 4.375 cm E 0.5 cm L 0 cm K.-H. Li et al.
the susceptor geometric can significantly impact the temperature profile and improve uniformity.
Specifically, the radius increase of the susceptor and the formation of the hollow structure of the
susceptor bottom cylinder can greatly improve temperature uniformity through manipulating the
thermal transfer, while the length increase of the susceptor bottom cylinder can only shift the
temperature profile. The geometric modification also causes change in the induction heating
efficiency. With the proposed techniques, the T-shape susceptor can exhibit excellent temperature
uniformity with temperature variation less than 5 °C at ~1900 °C and high induction heating
efficiency.
5 Acknowledgement
The authors would like to acknowledge the support of KAUST Equipment Fund BAS/1/1664-01-
08, KAUST Baseline BAS/1/1664-01-01, Competitive Research Grant URF/1/3437-01-01, and
GCC Research Council REP/1/3189-01-01. In addition, we thank Dr. Gary Tompa from Structured
Materials Industries (SMI) for fruitful discussion of the substrate temperature uniformity and the
MOVPE designs with induction heating and resistant heating.
Reference
1 A. Khan, K. Balakrishnan, and T. Katona, Nat. Photon. 2, 77(2008).
2 K. Watanabe, T. Taniguchi, T. Niiyama, K. Miya, and M. Taniguchi, Nat. Photon. 3, 591 (2009).
3 J. Zhang, H. Wang, W. Sun, V. Adivarahan, S. Wu, A. Chitnis, C. Chen, M. Shatalov, E. Kuokstis, J.
Yang, A. Khan, J. Electron. Mater. 32, 364 (2003).
4 A. Rice, A. Andrew, C. Mary, B. Thomas, O. Taisuke, S. Catalin, F. Jeffrey, and S. Michael, J. Cryst.
Growth 485, 90 (2018).
5 N. Fujimoto, T. Kitano, G. Narita, N. Okada, K. Balakrishnan, M. Iwaya, S. Kamiyama, H. Amano, I.
Akasaki, K. Shimono, T. Noro, T. Takagi, and A. Bandoh, Phys. Status Solidi C 3, 1617 (2006).
6 G. Kirchhoff, Monatsberichte der Akademie der Wissenschaften zu Berlin, 1859, 783 (1860).
7 C. Li, D. Zhu, X. Li, B. Wang, J. Chen, Nucl. Mater. and Ener. 13, 68 (2017).
8 S. Hu, Q. Wu, J. Li, H. Cao, Y. Zhang, and Z. Li, Mater. Sci. Eng. 322, 022006 (2018).
9 Y. Qu, B. Wang, S. Hu, X. Wu, Z. Li, Z. Tang, J. Li, and Y. Hu, J. Cent. South Univ. 21, 3518 (2014).
10 M. Tsai, C. Fang, and L. Lee, Chem. Eng. Process. 81, 48 (2014).
11 Z. Li, Y. Hao, J. Zhang, L. Yang, S. Xu, Y. Chang, Z. Bi, X. Zhou, and J. Ni, J. Cryst. Growth 311,
4679 (2009).
K.-H. Li et al.
12 Z. Li, H. Li, J. Zhang, J. Li, H. Jiang, X. Fu, Y. Han, Y. Xia, Y. Huang, J. Yin, L. Zhang, and S. Hu,
Appl. Therm. Eng. 67, 423 (2014).
13 Z. Li, J. Zhang, J. Li, H. Jiang, X. Fu, Y. Han, Y. Xia, Y. Huang, J. Yin, L. Zhang, and Y. Hao, J.
Cryst. Growth 402, 175 (2014).
14 C. Kim, J. Hong, J. Shim, Y. Won, Y. Kwon, IEEE 11th International Conference on EuroSimE, pp. 1-
7 (2010).
15 Z. Li, H. Li, X. Gan, H. Jiang, J. Li, X. Fu, Y. Han, Y. Xia, J. Yin, Y. Huang, and S. Hu, J. Semicond.
35, 092003 (2014).
16 Z. Li, H. Li, J. Zhang, J. Li, H. Jiang, X. Fu, Y. Han, Y. Xia, Y. Huang, J. Yin, L. Zhang, S. Hu, Int. J.
Heat Mass Transfer 75, 410 (2014).
17 K.-H. Li, H. Alotaibi, H. Sun, R. Lin, W. Guo, C. Torres-Castanedo, K. Liu, S. Valdes-Galán, and X.
Li, J. Cryst. Growth 488, 16 (2018).
18 COMSOL Multiphysics® User's Guide VERSION 4.3
19 P. Wallace, Phys. Rev. 71.9, 622 (1947).
20 L. Wang, Z. Tamainot-Telto, S. Metcalf, R. Critoph, R. Wang, Appl. Therm. Eng. 30, 1805 (2010).
21 G. Slack, Phys. Rev. 127, 694 (1962).
22 W. Haynes, ed. CRC handbook of chemistry and physics. CRC press, 2014.
23 L. Hu, C. Wang, and Y. Huang., J. Mater. Sci. 45, 3242 (2010).
24 V. Rudnev, D. Loveless, R. L. Cook, and M. Black (2002). Handbook of Induction Heating. CRC
Press.
25 D. J. Griffiths (1999). Reed College, Introduction to Electrodynamics, Vol. 3.
26 D. J. John (1999). Classical Electrodynamics.
27 E. Lenz, Annalen der Physik und Chemie 107, 483 (1843).
|
1810.00937 | 1 | 1810 | 2018-08-23T12:39:04 | Sodium-Ion-Conducting Polymer Nanocomposite Electrolyte of TiO2/PEO/PAN Complexed with NaPF6 | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | A free standing transparent film of solid state polymer electrolyte based on PEO/PAN + NaPF6 with different compositions of nano sized TiO2 in weight percent (x = 0, 1, 2, 5, 10, 15, 20 ) is synthesized by using standard solution cast technique. The homogeneous surface of above polymer composition is examined by FESEM. The microscopic interaction among polymer, salt and nanoceramic filler has been analyzed by Fourier Transformed Infra-Red (FTIR) spectroscopy. The reduction of ion pair formation in polymeric separator is clearly observed on addition of nanofiller in the polymer salt complex film. Electrical conductivity has been recorded of the prepared polymeric separator which is of the order of ~10-3 Scm-1 after addition of nanofiller (15% wt/wt) which support the FTIR results. Electrochemical potential window has been observed of the order of ~6V by the cyclic voltammetry results. The observed data of the prepared separator are at par with the desirable value for device applications. | physics.app-ph | physics | Sodium-Ion-Conducting Polymer Nanocomposite Electrolyte
of TiO2/PEO/PAN Complexed with NaPF6
Chandni Bhatt, Ram Swaroop, Parul Kumar Sharma and A. L. Sharma*
Centre for Physical Sciences, Central University of Punjab, Mansa Road Bathinda-151001, Punjab, India
*[email protected]
Abstract.A free standing transparent film of solid state polymer electrolyte based on PEO/PAN + NaPF6 with different
compositions of nano sized TiO2 in weight percent (x = 0, 1, 2, 5, 10, 15, 20 ) is synthesized by using standard solution cast
technique. The homogeneous surface of above polymer composition is examined by FESEM. The microscopic interaction among
polymer, salt and nanoceramic filler has been analyzed by Fourier Transformed Infra-Red (FTIR) spectroscopy. The reduction of
ion pair formation in polymeric separator is clearly observed on addition of nanofiller in the polymer salt complex film. Electrical
conductivity has been recorded of the prepared polymeric separator which is of the order of ~10-3 Scm-1 after addition of
nanofiller (15% wt/wt) which support the FTIR results. Electrochemical potential window has been observed of the order of ~6V
by the cyclic voltammetry results. The observed data of the prepared separator are at par with the desirable value for device
applications.
INTRODUCTION
Sodium is light alkali metal with single electron in outermost shell. This metal exhibit unexpected complexity
properties [1]. Sodium is more environmentally friendly than Li, less expensive, more abundant, easily distributed
and easier to extract than lithium [2]. Sodium is fourth most abundant element in earth crust and easy to recycle,
used for sodium ion batteries. Sodium batteries provide low cost energy storage device, they operate at ambient
temperature [3]. Polymer electrolyte act as electrolyte cum separator material in batteries, their main purpose is to
separate anode and cathode material in batteries and act as ions transport medium for the conduction of ions. In most
system conductivity much lower than the desirable values for the device application under surrounding conditions.
The factor responsible for that are: low ambient ionic conduction, concentration polarization, poor stability (thermal,
mechanical and electrochemical) [4]. To increase ionic conductivity various method employed such as, addition of
nano-sized inorganic ceramic fillers particles such as Al2O3, SiO2, TiO2, ZrO2 etc. in host polymer and other with the
addition of low molecular weight plasticizers such as ethylene carbonate (EC), propylene carbonate (PC), dimethyl
carbonate (DMC), polyethyleneglycol (PEG) [5]. In this work, investigation of solid state polymer electrolyte cum
separator is fabricated by taking various composition of TiO2 nano-filler adding in PEO/PAN polymer blend
complexed with anatomic ratio of (O/Na = 20) NaPF6 salt.
EXPERIMENTAL WORK
In experiment the pure polymer blend of composition (PEO + PAN) films complexed with NaPF6 salt prepared by
solution cast technique. The weight percent ratio such as 50 weight percent of PEO, PAN, and salt of O/Na ratio is
20 mixed by a solution casting technique using DMF as solvent. Whereas in another samples the same polymer
blend doped with different composition of nano-filler (TiO2) was prepared in different proportion such as 1, 2, 7, 10,
15 and 20 weight percent. The solution of above composition mixed for several hours by means of stirrer, until the
solution mixed properly. The obtained polymer blend nanocomposites were cast onto polypropylene dishes and left
for evaporation at room temperature. The final product of polymer blend nanocomposites was dried under vacuum at
10-3 mbar pressure for 2-3 days. Finally we got free standing film of pure and doped polymer electrolyte.
Impedance spectroscopy was carried out for conductivity measurement with CH Instruments electrochemical
workstation over the frequency range 1MHz, the sample sandwiched between stainless steel electrodes. FTIR
analysis of the sample was done to detect the presence of various functional group. Surface morphology of the
sample was observed by scanning electron microscopy. Working voltage range evaluated by cyclic voltammetry.
RESULT AND DISCUSSION
Complex Impedance Spectroscopy
Figure 1 represents complex impedance plot of prepared thin polymer nanocomposite films comprising of film
(PEO/PAN)20NaPF6+xwt.%TiO2. From fig1, it is observed that the higher frequency semicircle and low frequency
spike present in both the sample. The higher frequency semicircle represents bulk contribution of the materials and
spike due to the capacitance developed at electrode electrolyte interface. A nonlinear least square fitting of
impedance response of all samples agrees well with electrical equivalent circuit model comprising a series
combination of constant phase element (CPE) with another constant phase element (CPE) and resistance (Rb), which
is parallel with one another, further they are in series combination with another resistance (R1).
FIGURE. 1. Nyquist plot of thin PNC films comprising of (PEO/PAN)20NaPF6+xwt.%TiO2 (a) x=0 (b) x=15
The conductivity (σ) of the polymer electrolyte was calculated from the following formula. σ = L /RbA, where L is
thickness of polymer electrolyte film (cm), and A is the area of crossection of electrode (cm2) [6]. The electrical
conductivities values for blend polymer and TiO2 doped blend polymer films are recorded in the table 1.
Table 1. Conductivities values for polymer blend PEO/PAN complexed with NaPF6 with different concentration of
nano-filler TiO2 (0, 1, 2, 5, 10, 15, 20) for sodium ion batteries.
Table1 clearly indicates that electrical conductivity has improved by an order value on after addition of 15wt.%
TiO2 in pure blend polymeric films.
Conductivity (S/cm)
Polymer Electrolyte
(PEO/PAN + NaPF6)
(PEO/PAN + NaPF6)
Nano-filler TiO2 (in wt.
percent)
0
15
Bulk Resistance (Rb)
In (Ω)
6331
1134
3.2 x 10-5
1.8 x 10-4
Electrochemical Analysis
For electrochemical devices such as batteries, capacitors and electrochromic devices the electrochemical stability
window curve is prime requirement. The cyclic voltammetry has been performed of the prepared cathode materials.
This process was observed in the potential range of ~6V at current density of 0.00004 to -0.00025 through blocking
electrode of stainless steel. Electrochemical stability window curve is obtained in the range from about ~6V, which
is an acceptable voltage range for device application [7].
FIGURE. 2. Cyclic voltammeter images of (a) pure (PEO/PAN+NaPF6) (b) doped15wt. % TiO2 + ((PEO/PAN+NaPF6)
polymer electrolyte.
Fourier Transform Infrared (FTIR) analysis
Fourier Transfer infrared microscopy analysis of the sample was done to detect the presence of various functional
groups. FTIR spectrum was observed in the range of 600-4000 cm-1 as shown in fig.3. Also shift in peak position
with addition of salt was clearly seen infig.3 (a). The peak in the wavenumber range 800 cm-1 to 900 cm-1
corresponds to C-O-C of PEO and salt
−). The characteristic absorption peak observed in the spectral pattern
at the wavenumbers ~690, ~840-920, ~1088, ~ 1247, ~1670, ~ 2242, ~2856 and 2932 cm-1 are attributed to
def(CH2),
(CH2) respectively. Two clear CH2
vibrational modes appear due to PEO near 1458 cm-1 which, correspond to asymmetric CH2 bending (δ(CH2)a) and
near 1356 cm-1 which, corresponds to symmetric CH2 wagging and some C- C stretching (w(CH2)s + ν (C―C)).
The FTIR peaks at 1958 cm-1 are related to asymmetric stretching of CH2 in PEO. The C≡N stretching band in the
IR spectrum appearing at 2245 cm-1is the most characteristic feature of nitrile group in pure PAN. The vibrational
peak at 2935 cm-1, 2938 cm−1, 2921 cm−1 and 2923 cm−1 are assigned to asymmetric stretching of CH2. As NaPF6
salt concentration increases, there is shift in peak position of pure PEO/PAN+ NaPF6 sample related to other TiO2
(PF6−), t(CH2)s, t(CH2)as, ʋ(C=O) , ʋ(C≡N),
(CH2) and
(PF6
doped sample as shown in fig. (b), this is due to the additional amount of Na+ in the complexes. Pure PEO shows a
large, broad band of CH2 stretching between 2950 and 2840 cm-1. The FTIR analysis gives us the information of
role of salt in the polymer and salt interaction in the polymer film.
Figure. 3. FTIR spectrum for (a) different composition for x wt. % of TiO2 (b) Variation of PEO peak at 1600 cm-1.
CONCLUSION
Free standing film was prepared by using PEO/PAN as polymer blend complexed with NaPF6 salt with different
concentration of TiO2 (0, 1, 2, 5, 10, 15, 20) as nano filler. From impedance spectroscopy the value for maximum
conductivities is 1.8 Х 10-4 Scm-1 for 15wt. % of TiO2 concentration which further decrease with the more addition
of nano-filler. Electrochemical stability for our film is obtained in the potential range from ~6V which is good for
device application. FTIR spectroscopy confirms the salt, polymer, and nano-filler interaction.
ACKNOWLEDGEMENT
Author is thankful to Central University of Punjab for providing material, instrumental and Laboratory facility.
REFERENCES
1. Y.Q. Yang, Z. Chang, M.X. Li, X.W. Wang, and Y.P. Wu, Solid State Ionics 269, 1-7(2015).
2. Ho KhacHieu, Vacuum 120, 13-16(2015).
3. J. Serra Moreno, M. Armand, M.B Berman, S.G. Greenbaum, B. Scrosati, and S. Panero, Jouranal of Power
Sources 248, 695-702 (2014)
4. A.L.Sharma, Awalendra K. Thakur, Journal of Material Science 46, 1916-1931 (2011).
5. Harinder Pal Singh Missan, Boor Singh Lalia, Kunal Karan, and Andersan Maxwell, Material Sciences and
Engineering B 175, 143-149(2010).
6. YatimLailunNi'mah, Ming-Yao Cheng, Ju Hsiang Cheng, John Rick, and Bing-Joe Hwang, Journal of Power
Sources 278, 375-381(2014).
7. Chengying Cao, Haibin Wang, Weiweiliu, Xiaozhen Liao, and Lei Li, international Journal of Hydrogen energy
39, 16110-16115(2014).
8. F. A. Amarai, R.M. Sousa, L. C. T. Morais, R.G. Rocha. I.O. Campos, W.S. Fagundes, C.N.P. Fonseca, and S.C.
Canobre, J ApplElectrochem45, 809-820 (2015).
|
1805.03246 | 1 | 1805 | 2018-04-23T13:06:15 | Strong PUFs from arrays of resonant tunnelling diodes | [
"physics.app-ph"
] | In this work, we design and implement a strong physical uncloneable function from an array of individual resonant tunnelling diodes that were previously described to have a unique response when challenged. The system demonstrates the exponential scalability of its responses when compared to the number of devices present in the system, with an expected large set of responses while retaining a 1:1 relationship with challenges. Using a relatively small set of 16 devices, 256 responses are shown to have promising levels of distinctness and repeatability through multiple measurements. | physics.app-ph | physics | This paper is a preprint of a paper submitted to IET Electronic Letters and is subject to Institution of Engineering and Technology
Copyright. If accepted, the copy of record will be available at IET Digital Library
Strong PUFs from arrays of resonant
tunnelling diodes
B. Astbury, I.E. Bagci, T. McGrath, J. Sexton, M. Missous,
U. Roedig, R. Bernardo-Gavito, and R.J. Young
shift in conduction mechanism (Fig. 1) serves as the unique characteristic
that is different for each device.
In this work, we design and implement a strong physical uncloneable
function from an array of individual resonant tunnelling diodes that were
previously described to have a unique response when challenged. The
system demonstrates the exponential scalability of its responses when
compared to the number of devices present in the system, with an
expected large set of responses while retaining a 1:1 relationship with
challenges. Using a relatively small set of 16 devices, 256 responses are
shown to have promising levels of distinctness and repeatability through
multiple measurements.
Introduction: With the advent of interconnected societies in a modern
world with a large-scale Internet of Things (IoT), our reliance on
software, hardware and networks has increased dramatically. Globally,
the dependence on the security of these systems further increases with
our adoption of these networks, where the market is expected to continue
to expand rapidly in the coming years [1]. The trust we place in the
security of these systems may be unfounded, with reports of attacks using
hundreds of thousands of compromised IoT devices [2]. With an increase
in the number of attacks comes an expected increase in success.
One solution to address the security and authentication of users is to
exploit physically uniqueness from a system in the growing field of
physical uncloneable functions (PUFs). PUFs, as opposed to digital keys,
boast uniqueness that is uncloneable. Various implementations exist
ranging from optical to electronic-based devices. These can be
categorised by their inherent security and implementation method as
being either weak or strong [3].
One type of PUF proposed recently uses a quantum effect in resonant
tunnelling diodes (RTDs) to create a single unique response from
uncontrollable atom-scale differences in semiconductor manufacture.
The quantum-confinement PUF (QC-PUF) [4] has a single challenge-
response pair (CRP) per device, derived from a peak in its current-voltage
trace (see Fig. 1a). While this weak PUF can be used to generate keys, its
limited number of CRPs leaves it exposed to attackers and thus requires
protected access (a protected housing, for example). For a lightweight
authentication system, a larger set of CRPs, which increase exponentially
with some system parameter, is required to ensure each response, when
used, is not compromised.
The solution proposed herein consists of combining RTDs in arrays
such that each set of RTDs can output a much larger CRP set.
Theory: RTDs provide a unique fingerprint from naturally-occurring
atom-scale variations in the semiconductor manufacturing process.
RTDs were previously used by Roberts et al. as a weak PUF [4]. They
are a physical manifestation of a finite quantum well (QW) with potential
barriers in a conducting channel. The devices are created by sandwiching
a thin, narrow band-gap semiconductor between two wide band-gap
semiconductors to act as the main QW structure. Beyond the tunnelling
barriers lies highly-doped narrow band-gap regions of semiconductor
which form the electron source and sink [5].
These devices operate under two electron transport regimes, depending
on the voltage that is applied across them. The transition between these
regimes provides their characteristic electronic behaviour, which is
illustrated in Fig. 1a. At low voltages, quantum resonant tunnelling
through the classically forbidden region dominates, for which the device
finds its name. The conduction level rises as a single confined energy
level comes into resonance with electrons that have the required energy
to pass through [6]. A shift in transport mechanism happens when the
energy level passes out of resonance, resulting in a shift to classical
conduction. This is known as thermionic emission, where electrons with
the most thermal energy pass over the top of the tunnelling barrier [6].
The transition between these regimes is a unique response in the
current-voltage characteristic of every RTD, which is defined by the first
allowed energy level within the QW structure. This is highly sensitive to
small variations, such as well width. The single peak that results from the
1
To create a large database of responses, characterisation of how two
devices interact in a single circuit must show that the properties of one
device acts on that of the second device. The impedance of devices, for
example, results in a combined shift in peak from that of a single device.
It is seen that different devices cause different shifts due to slight
variations in resistances at different points in a single current-voltage
response of the devices. The result is that each combination of devices is
unique to the devices within them, and the base RTDs are unique due to
variations in manufacture.
Methodology: Networks of diodes were connected with switches,
allowing the current path to be controlled. This is illustrated in Fig 1c. It
allows the uniqueness of the convoluted response of multiple devices to
be measured, to ascertain how the CRPs for the system grows with the
number of elements within it. Rectangular arrays of devices were built,
with n columns containing m RTDs. The aim is to increase the number of
responses by more than the simple product, m × 𝑛, to 𝑚𝑛. For an array
with 3 × 4 elements, for example, this would provide 81 permutations
(i.e. chains of 4 RTDs in series), with each current-voltage characteristic
containing 4 peaks in each permutation (see Fig. 1b).
Fig. 1 Current-voltage characteristics of single and arrayed RTDs.
a - A trace showing the current-voltage curve for a single RTD,
sectioned to show the prominent transport mechanisms and the negative
differential resistance (NDR). b - Current-voltage curves for two
different linear chains of 4 RTD devices connected in series. c - A 3x4
array of individually switched RTDs. The circuit representation shows
two possible current paths (blue or red) through the system, for different
switch configurations.
The total number of permutations of current paths through RTDs
linked in series through switched arrays with different array dimensions
is illustrated in Fig. 2. The parameters of the array can be tuned to
maximise the outputs and the points by which authentication can be made
(peak positions). As can be seen in Fig. 2, the number of permutations
increases much faster as the number of devices per column increase. This
corresponds to having more devices to swap and more peaks per
permutation. Due to the exponential nature of the increase, the optimal
value of rows is the natural logarithm, 𝑒. In terms of a physical number
of devices per column, this corresponds to 2 or 3 rows. For example, 100
devices can be configured in a 10 × 10 array, giving 1 × 1010 responses,
or a 2 × 50 array, giving far more (1.125 × 1015) responses. Optimising
the array's configuration makes a significant difference to the total
number of responses therefore, increasing the time in which it takes an
adversary to characterise all the responses of a single system.
A 4 x 4 array of RTDs was constructed where each diode is in series
with a switch. Each RTD was individually wire-bonded using 15 µm gold
wire to the circuit, with 16 switches between the diodes controlled by an
Arduino Due. RTD characteristics were measured using a Keithley
2602B SMU. Permutations were run consecutively with multiple-
subsequent measurements of each to test similarity.
Fig. 2 The number of permutations of an array (colour scale) as a
function the rows and columns it contains. The black contours show the
total number of RTDs in the array.
Results: The axes of the current-voltage curve were divided into
equidistant bins with a permutation's response characterised by the bin/s
containing peak/s. This system of bins was used to determine distinctness
and similarity of the permutations denoted by uniqueness and robustness.
Uniqueness is defined as how distinct each permutation is from one
another and uses the expected output of a permutation as its comparison
point.
A 1 is given if a compared peak is found in the same bin and 0 if
different, the term is denoted by equation (1). The uniqueness is given by
equation (2) as a sum of difference metrics and divided by the number of
summed terms. Robustness is given as the complement of metric given
from (2) using repeat measurements as 𝑅𝑖, 𝑅𝑗. Both metrics are created as
probabilities that the next measurement will conform to the expected
output.
𝑑𝑖𝑓𝑓(𝑅𝑖, 𝑅𝑗) = {
0, 𝑅𝑖 = 𝑅𝑗
1, 𝑅𝑖 ≠ 𝑅𝑗
𝑘
∑
𝑖,𝑗=1,𝑖≠𝑗
𝑑𝑖𝑓𝑓(𝑅𝑖, 𝑅𝑗)
1
𝑘
(1)
(2)
Table 1 gives results for a 4x4 array with 256 permutations having 256
bins. For each peak position, on average a high percentage, unique
measurement is seen with 3 out of 4 peaks registering greater than 90%
uniqueness. This is coupled with a similar robustness for each of the peak
positions of ~80%. Peak 1 was found to have the lowest uniqueness, due
to the near-linear dependence between current and voltage in the resonant
tunnelling regime.
Comparing each 4-peak permutation to a subsequent permutation, if
one or more peaks are distinct then the permutation is distinct. This gives
a uniqueness measure that convolutes all four peak positions. The same
process can be repeated for subsequent measurements of the same
permutation to ascertain the robustness of 4-peak measurements. As
expected, an increase in overall uniqueness and a decrease in robustness
is observed in Table 1. This finding was expected, as convoluting peak
positions provide more points for a distinct signature, but measurement
variations also increase the likelihood of failure of measuring true-
positive results.
Table 1: The result of uniqueness and robustness measurements
performed on a 4 x 4 array with 256 permutations and 4 peaks. Results
are first shown for each of the four individual peaks, then their average
and finally for using all peaks simultaneously.
Array Size Bins
Identifier Uniqueness (%) Robustness (%)
4x4
256 Peak 1
Peak 2
Peak 3
Peak 4
Average
4-Peak
76.8
95.0
94.4
92.3
85.6
99.7
83.9 ± 12.6
78.4 ± 10.4
82.8 ± 10.9
81.0 ± 14.8
81.5 ± 12.3
60.7± 𝟏𝟕. 𝟐
the system
implementation
in
to reduce noise
Discussion: Overall, the system uniqueness is high, with a 4-peak value
over 99%. Real implementations of a system like this in cryptographic
applications would require higher values that could be achieved by
scaling the array size and increasing the number of bins used. The
relatively low robustness of 60% would necessitate measuring the
response to challenges multiple times to reduce false-negative rates.
Modifying
the
measurement process would inherently increase the robustness of the
measurements, and also allow the bin size to be decreased, in turn
increasing the uniqueness of the responses. The implications of results
provided herein are evocative of a system useful in providing increased
security to IoT devices with low power, cost-effective results.
Conclusion: We have demonstrated an exponentially growing design for
a strong PUF system using resonant tunnelling diodes. The number and
form of the responses can be tailored to the security requirement of an
application by maximising the unique outputs of the system. This system
shows promising results in its ability to unique, repeatable responses in
scalable devices. The system is comprised of nanoscale elements that can
be implemented in CMOS technology [7], with a relatively small gate
count for the entire device. We believe the small scale, low power and
cost requirements make the implementation of this technology well
suited to IoT devices and other applications of small embedded systems.
Acknowledgements: RJY acknowledges support by the Royal Society
through a University Research Fellowship (UF110555 and UF160721).
This material is based upon work supported by the Air Force Office of
Scientific Research under award number FA9550-16-1-0276. This work
was also supported by grants from The Engineering and Physical
Sciences Research Council
(EP/K50421X/1 and
EP/L01548X/1) and the Royal Society through a Brian Mercer award.
[Copyright statement goes here
Submitted:
doi:
One or more of the Figures in this Letter are available in colour online]
the UK
in
B. J. Astbury, R. Bernardo-Gavito, T. McGrath and R. J. Young (Physics
Department, Lancaster University, Lancaster, United Kingdom, LA1
4YB)
E-mail: [email protected] or [email protected]
I.E. Bagci, U. Roedig (School of Computing and Communications,
InfoLab21, Lancaster University, Lancaster, United Kingdom, LA14WA)
J. Sexton, M. Missous, (School of Electrical and Electronic Engineering,
The University of Manchester, Manchester, United Kingdom M13 9PL)
References
1 Lund, D., Morales, M.: 'Worldwide and Regional Internet of Things
(IoT) 2014 – 2020 Forecast : A Virtuous Circle of Proven Value and
Demand', IDC Anal. Futur., 2014, (May), p. 29.
2 Antonakakis, M., April, T., Bailey, M., et al.: 'Understanding the
Mirai Botnet', Proc. 26th USENIX Secur. Symp., 2017, pp. 1093–1110.
3 Gao, Y., Ranasinghe, D.C., Al-Sarawi, S.F., Kavehei, O., Abbott, D.:
'Emerging Physical Unclonable Functions with Nanotechnology', IEEE
Access, 2016, 4, pp. 61–80.
4 Roberts, J., Bagci, I.E., Zawawi, M.A.M., et al.: 'Using Quantum
Confinement to Uniquely Identify Devices', Sci. Rep., 2015, 5, (1), p.
16456.
5 Zawawi, M.A.M., Ian, K.W., Sexton, J., Missous, M.: 'Fabrication
of submicrometer INGAAS/ALAS resonant tunneling diode using a
trilayer soft reflow technique with excellent scalability', IEEE Trans.
Electron Devices, 2014, 61, (7), pp. 2338–2342.
6 Ricco, B., Azbel, M.Y: 'Physics of resonant tunneling. The one-
dimensional double-barrier case', Phys. Rev. B, 1984, 29, (4), p. 1970-
1981.
7 Lavieville, R., Triozon, F., et al: 'Quantum dot made in metal oxide
silicon-nanowire field effect transistor working at room tempterature',
Nano. Lett., 2015, 15, (5), p. 2958-2964.
2
|
1903.05560 | 2 | 1903 | 2019-09-27T13:50:08 | Femtosecond laser-induced electron emission from nanodiamond-coated tungsten needle tips | [
"physics.app-ph",
"cond-mat.mes-hall"
] | We present femtosecond laser-induced electron emission from nanodiamond-coated tungsten tips. Based on the shortness of the femtosecond laser pulses, electrons can be photo-excited for wavelengths from the infrared (1932 nm) to the ultraviolet (235 nm) because multi-photon excitation becomes efficient over the entire spectral range. Depending on the laser wavelength, we find different dominant emission channels identified by the number of photons needed to emit electrons. Based on the band alignment between tungsten and nanodiamond, the relevant emission channels can be identified as specific transitions in diamond and its graphitic boundaries. It is the combination of the character of initial and final states (i.e. bulk or surface-near, direct or indirect excitation in the diamond band structure), the number of photons providing the excitation energy and the peak intensity of the laser pulses that determines the dominant excitation channel for photoemission. A specific feature of the hydrogen-terminated nanodiamond coating is its negative electron affinity that significantly lowers the work function and enables efficient emission from the conduction band minimum into vacuum without energy barrier. The properties of these tips are encouraging for their use as laser-triggered electron sources in applications such as ultrafast electron microscopy and diffraction and novel photonics-based laser accelerators. Emission is stable for bunch charges up to 400 electrons per laser pulse. We infer a normalized emittance of < 0.20~nm~rad and a normalized brightness of > $1.2 \cdot 10^{12} \text{A} \text{ m}^{-2} \text{ sr}^{-1}$. | physics.app-ph | physics |
Femtosecond laser-induced electron emission from nanodiamond-coated tungsten needle tips
Department of Physics, Friedrich-Alexander-Universität Erlangen-Nürnberg, Staudtstrasse 1, D-91058 Erlangen, Germany
A. Tafel,∗ S. Meier, J. Ristein, and P. Hommelhoff
(Dated: September 30, 2019)
We present femtosecond laser-induced electron emission from nanodiamond-coated tungsten tips. Based on
the shortness of the femtosecond laser pulses, electrons can be photo-excited for wavelengths from the infrared
(1932 nm) to the ultraviolet (235 nm) because multi-photon excitation becomes efficient over the entire spectral
range. Depending on the laser wavelength, we find different dominant emission channels identified by the
number of photons needed to emit electrons. Based on the band alignment between tungsten and nanodiamond,
the relevant emission channels can be identified as specific transitions in diamond and its graphitic boundaries. It
is the combination of the character of initial and final states (i.e. bulk or surface-near, direct or indirect excitation
in the diamond band structure), the number of photons providing the excitation energy and the peak intensity
of the laser pulses that determines the dominant excitation channel for photo-emission. A specific feature of
the hydrogen-terminated nanodiamond coating is its negative electron affinity that significantly lowers the work
function and enables efficient emission from the conduction band minimum into vacuum without energy barrier.
Emission is stable for bunch charges up to 400 electrons per laser pulse. We infer a normalized emittance of
< 0.20 nm rad and a normalized peak brightness of > 1.2· 1012 A m−2 sr−1. The properties of these tips are
encouraging for their use as laser-triggered electron sources in applications such as ultrafast electron microscopy
and diffraction and novel photonics-based laser accelerators.
− 10−9 Pa regime to minimize bombardment with ion-
Tip-shaped cathodes are amongst the most commonly used
electron sources in electron microscopy due to their ability
to provide a high quality beam. Typical materials are zirco-
nia in common Schottky type emitters and lanthanum hex-
aboride because of their low work function, and tungsten
due to the easy fabrication of sharp tips ideally suited for
(cold) field emission [1]. Most of the commonly used emit-
ters are operated under ultra-high vacuum conditions in the
10−8
ized gas and adsorption on the emitter surface. Furthermore,
they are heated for thermal enhancement of the emission or to
achieve stable operation due to reduced adsorption.
Over the last decades, ultrafast electron microscopy has
emerged [2 -- 4]. Until today, the same emitters are used in
DC and ultrafast mode.
In the latter case, the cathode is
typically triggered by femtosecond laser pulses resulting in
femto- to picosecond electron pulses [5 -- 10]. One of the ma-
jor drawbacks of these laser-triggered electron sources is the
continuous decrease of emission current over time [5, 9, 10],
which is attributed to laser-induced changes at the emitter sur-
face. Femtosecond laser-induced photo-emission from tip-
shaped cathodes has been extensively studied for the materials
tungsten [11 -- 16], gold [17 -- 19], silver [20], hafnium carbide
[21] and carbon nanotubes [22, 23]. Pulsed photo-emission
from single crystal diamond tips has been investigated with
nanosecond pulses [24]. Femtosecond photo-emission from
tip-shaped heterostructures offers promising opportunities yet
to be discovered.
Diamond is one of the most robust materials due to its ex-
ceptional chemical inertness, mechanical strength and thermal
conductivity. Nanocrystalline diamond (NCD) is a good elec-
tron emitter, especially if the surface exhibits negative elec-
tron affinity (NEA) [25]. The graphitic grain boundaries in
this composite material provide electrical conductivity, and
the low work function of the diamond matrix that goes along
with the NEA also lowers the surface energy barrier for the
electrons even if they originate from the graphitic parts [25].
NEA is also known to boost the photo-electron yield due
to fundamental absorption, i.e. optical excitation across the
band gap: electrons photo-excited into the diamond conduc-
tion band can be emitted into vacuum without any barrier af-
ter migration to the surface [26, 27]. The electron affinity of
hydrogen-terminated diamond is as low as -1.3 eV for both
main crystallographic surfaces (100) [28] and (111) [29].
The combination of high beam quality from tip-shaped pho-
tocathodes with the mechanical strength and low work func-
tion of hydrogen-terminated diamond promises a robust and
high-brightness photocathode. Here we present first photo-
emission results from a tip-shaped semiconductor/metal het-
erostructure - diamond-coated tungsten tips - triggered with
femtosecond laser pulses and characterize the underlying
photo-emission physics by identifying various emission chan-
nels. We define an emission channel as the combination
of photon energy and energy states involved in the photo-
emission process of electrons.
To obtain the nanodiamond-coated tips, 100 µm diameter
tungsten wire is electrochemically etched resulting in a tip
with a typical apex radius of roughly 10 nm. The freshly
etched tip is dip-seeded in nanodiamond suspension and dry-
blown with pressurized nitrogen. NCD is grown on the seeded
tips with microwave-enhanced chemical vapor deposition re-
sulting in a dense film of hydrogen-terminated nanocrystalline
diamond with negative electron affinity covering the tungsten
surface (Fig. 1 inset). A thin layer of tungsten carbide (WC) is
expected to be formed at the diamond/tungsten interface [30].
Samples used in this work have apex radii between 60 nm and
200 nm including the diamond coating. Details of the fabri-
cation process and a structural characterization of the tips are
published elsewhere [31].
The so-fabricated tips are mounted in an ultra-high vacuum
chamber with a base pressure of 1 · 10−7 Pa. Femtosecond
laser pulses are focused at the tip with the help of a 51 mm
2
(a)
(b)
e
s
l
u
p
r
e
p
s
n
o
r
t
c
e
l
E
100
10
1
0.1
0.01
e
s
l
u
p
100
r
e
p
s
n
o
r
t
c
e
l
E
10
1
5 1 2 n m / 2 . 4 e V
1 9 3 2 n m
(
(
0 . 6 4 e V
/
0.5
1.0
1.5
2.0
2.5
Peak Intensity (1011 W/cm2)
4 . 8 e V
/
2 5 6 n m
1
t
i
f
T o t a l
2
=
n
10
Pulse energy (nJ)
n = 1
50
FIG. 2. (a) Photo-emission at 1932 nm (red diamonds) and 512 nm
(green squares), with slopes 5.0 (red dashed line, last three data
points not included due to potential saturation effects) and 3.4 (green
dashed line).
(b) Transition from one- to two-photon emission at
256 nm. The grey dashed lines are the corresponding contributions,
the solid black line is the sum of the two contributions. Note that
we used pulse energy instead of peak intensity in (b) as we could not
measure the laser spot size in the UV.
(a)
d
e
z
i
l
a
m
r
o
N
(b)
)
.
u
.
a
(
t
n
e
r
r
u
c
o
t
o
h
p
y
t
i
r
a
e
n
i
l
n
o
n
e
v
i
t
c
e
f
f
E
1
0.1
0.01
0.5
2 3 5 n m
2 6 0 n m
3 5 0 n m
1
Pulse energy (nJ)
5
350nm
260nm
235nm
2
1
3.5
3.75
4.0
4.5
4.25
4.75
Photon energy (eV)
n
r
e
d
r
o
n
o
t
o
h
p
5.0
5.25
5.5
FIG. 3. Power scaling in the UV with transition from one- to two-
photon emission as the dominant channel.
(a) Data at 235 (black
squares), 260 (violet circles) and 350 nm (blue triangles) with effec-
tive nonlinearities 1.1, 1.4 and 2.0, respectively. (b) Effective nonlin-
earity vs photon energy with a gradual transition from n=1 to n=2 at
roughly 4.8 eV.
with comparable strength, the linearized slope is non-integer
and is called effective nonlinearity. Depending on the photon
energy and laser intensity, different emission channels can be-
come dominant. We show the power dependence of the photo-
electron current at a wavelength of 1932, 512 and 256 nm in
Fig. 2. For 1932 nm, we find an integer slope of 5.0 indi-
cating a single dominating emission channel with 5 photons.
At 512 nm, the plot shows an effective nonlinearity of 3.4.
This is indicative for two channels with photon order 3 and 4
contributing. At 256 nm, we observe a transition from photon
order one at low intensities to photon order two at high inten-
sities.
FIG. 1. Experimental setup. The laser is focused onto the diamond-
coated tungsten tip with an off-axis parabolic mirror (OAP). The in-
set shows a transmission electron micrograph of the nanodiamond-
coated tungsten tip. A voltage clearly below the DC field emission
threshold is applied between tip and microchannel plate (MCP) to
accelerate electrons towards the MCP. See text for details.
diameter off-axis parabolic mirror with 152 mm focal length
outside of the vacuum chamber, resulting in a measured spot
radius of 3.8 µm at 512 nm (1/e2 intensity radius). The em-
ployed commercial laser system consists of a regeneratively
amplified Ti:Sa oscillator (1 kHz repetition rate, 80 fs pulse
duration), an optical parametric amplifier and a stage for sec-
ond harmonic and sum-frequency generation. Additionally, a
Ti:Sa oscillator (780 nm, 80 MHz, 6 fs) is used for long-term
stability measurements. We apply a negative voltage below 50
% of the DC field emission threshold (400-2000 V depending
on the individual tip). Due to the dielectric surface with a
small work function of 2.8 eV (equation 2), the Schottky re-
duction is lower compared to metal tips and is neglected here.
The DC field was chosen low enough, so that photon-assisted
field emission does not occur, only multi-photon emission.
The laser pulses are linearly polarized parallel to the tip axis.
Photo-emitted electrons are detected with a multichannel plate
(MCP) with grounded front plate. For bunch charges below
one electron per laser pulse, we count detection events on the
MCP, above one electron per pulse we measure the calibrated
MCP screen current and for large average currents at high rep-
etition rates we are able to additionally measure the current
through the tip.
In order to identify the different contributions to the photo-
electron current J, we have measured its dependence on the
peak intensity Ip. Due to the high Ip of the femtosecond laser
pulses, optical excitation is not limited to one-photon absorp-
tion processes as multi-photon absorption becomes efficient.
The dependence of the photo-electron current J on Ip is ex-
pected to be a sum of power law contributions:
J =
∞
∑
n
an · In
p
(1)
where n reflects the number of photons necessary to pro-
vide the excitation energy and an the corresponding coeffi-
cient for the specific emission channel. Often, one channel
is dominant, hence the slope of log(J) vs log(Ip) directly re-
veals the photon order n. If more than one channel is involved
(a)
(b)
(c)
Γ15
(direct E )
g
1.6 eV
CBM
5.5 eV
E along line
A
c
-1.3 eV vacuum level
e-
2.8 eV
E
F
W / WC
1.2 eV
1.4 eV
VBM
diamond
W / WC
diamond
graphite
1
.
0
2
.
0
2
.
4
2
.
5
2
.
7
x
e-
A
e-
e-
3 . 9
4 .3
4.6
B
E along line
B
vacuum level
4.7 eV
W / WC
graphite
x
FIG. 4. Sketch of the relevant energy levels and optical excitation
paths in the nanodiamond needle tip coating, including the graphitic
boundaries. The concatenated arrows indicate excitation channels
we identify as relevant here. Length and color of the arrows in (a)
and (c) represent the photon energies 0.64 (near infrared), 2.4 (green)
and 4.8 eV (near ultraviolet). With these photo-excitation channels,
we can explain the observed laser power and wavelength dependence
discussed around Figs. 2 and 3. Intriguingly, this emission channel
identification, except for the assignment of the one photon process
in the UV, seems to result in a unique attribution in spite of the in-
tricate level structure. The work function of diamond is 2.8 eV for a
negative electron affinity of χ = -1.3 eV. In the diamond bulk, elec-
trons can be excited across the indirect (5.5 eV) or direct bandgap
(7.1 eV), with two ultraviolet or three green photons, migrate to the
surface and cross the surface into vacuum. Alternatively, electrons
can be excited into vacuum directly by one ultraviolet or five infrared
photons. Even if the electrons originate from graphite close to the di-
amond interface, they effectively feel the work function of diamond
as indicated by their trajectories across the equipotential lines in (b).
See text for details.
In the UV (235-350 nm) we have investigated the wavelength
dependence of the effective nonlinearity in more detail. Fig. 3
(a) shows the log(J) vs log(Ip) plot for 235, 260 and 350 nm.
We find effective nonlinearities of 1.1, 1.4 and 2.0, respec-
tively. Again, this reflects the transition of the dominant emis-
sion channel from first to second order. Note that we do not
observe a transition in the power dependence directly in con-
trast to Fig. 2 (b). This is due to the restricted pulse energy
range in Fig. 3. The effective nonlinearities for all wave-
lengths in the UV are summarized in Fig. 3 (b), confirming
the transition mentioned above.
For the interpretation of the data, we sketch the energy
states relevant for this work in Fig. 4. Five junctions between
W/WC, diamond, vacuum and the graphitic grain boundaries
(called graphite in Fig. 4) are formed. Diamond forms Schot-
tky junctions with graphite and W/WC with Schottky barri-
ers EB,G=1.4 eV [32] and EB,W /WC=1.2 eV [33], respectively.
As the sample surface only consists of diamond grains and
3
their graphitic boundaries, the junctions diamond/vacuum and
graphite/vacuum are the relevant ones for electron emission
into vacuum. In a hetero system involving metallic (W/WC
and in a good approximation also the half metal graphite) and
semiconducting (diamond) components, the Fermi level in the
semiconductor relative to the valence band maximum (VBM)
is identical to the Schottky barrier height as long as the dimen-
sions of the semiconducting parts are much below the Debye
length of the semiconductor. This is certainly the case for
the diamond grains. We expect EB,G to dominate at the di-
amond surface as the average grain size (approx. 20 nm) is
smaller than the thickness of the diamond film. Consequently,
the Fermi level is EB,W /WC=1.2 eV above the VBM at the back
contact and EB,G =1.4 eV above the VBM at the free surface
(see Fig. 4(a)). The work function Φ is defined as the energy
difference between the vacuum level and the surface Fermi
level. Graphite has a work function of 4.7 eV [34], while the
work function of diamond depends on the electron affinity χ
and EB,G and results in
Φdia = Eg − EB,G + χ = (5.5 − 1.4 − 1.3) eV = 2.8 eV, (2)
where we inserted -1.3 eV for the electron affinity of a fully
hydrogen-terminated diamond surface [29, 35]. The diamond
work function also constitutes the low energy threshold for
electrons originating from graphite (see Fig. 4(b)).
Based on this band diagram, we can identify electron emis-
sion channels with different energy thresholds as indicated in
Fig. 4(a). For diamond with negative electron affinity, the en-
ergy barrier between the conduction band minimum (CBM)
at the surface does not exist: Electrons can be excited into the
conduction band across the indirect (5.5 eV [36]) and direct
bandgap (7.1 eV [37, 38]), migrate to the surface and escape
straight into vacuum even if they have thermalized to CBM.
Alternatively, direct optical excitation from electronic states
at the surface to the plane wave like states in vacuum can lead
to photoelectrons as well.
The emission probabilities of the different channels are
complex functions of the densities of initial and final states,
the number of photons necessary to provide the transition
energy and the laser intensity. We discuss them by referring
to their signature in the log(J) vs log(Ip) plots in Figs. 2 and
3. Excitation with 1932 nm (¯hω = 0.64 eV, red arrows in
Fig. 4(a)) and an observed photon order 5 can be identified as
transitions at the surface from the Fermi level to the vacuum
level. For clearer presentation, we have sketched the red
arrows only in Fig. 4(a) although the initial states at the
Fermi level can be assigned either to defects in the diamond
or, more likely, to the graphitic grain boundaries [27].
At 512 nm (¯hω = 2.4 eV, green arrows in Fig. 4) the effective
nonlinearity equals 3.4, which we attribute to excitation
across the direct band gap by three or by four photons.
With UV excitation (λ < 350 nm, ¯hω > 3.5 eV, violet
arrows in Fig. 4), we observe one- and two-photon processes
(Fig. 2(b) and 3). We assign the one photon process at low
intensities to excitation at the surface from the Fermi level or
the diamond VBM to the vacuum level. Evaluating energy
differences only, the two-photon process (∆E > 7.2 eV for
λ < 350 nm) could be assigned to all transitions in the band
diagram of Fig. 4. We suggest for this process the transition
across the direct band gap of diamond: The spatial overlap
of the wave functions, the direct nature of the transition and
the large excitation volume make this process by far the most
likely. This argument is supported also by the nonlinearity
of 3.4 which we observe for 512 nm (¯hω = 2.4 eV: Two
photons of that energy would suffice to excite electrons from
the VBM directly into vacuum. Nevertheless, this channel is
not observed as the dominant one. The situation is different
for 1932 nm (¯hω = 0.64 eV): With this wavelength excitation
across the indirect bandgap would require 9 photons and 11
photons across the direct bandgap. These extremely high
order processes are so unlikely that we observe the fifth
order process at the surface as the dominant channel instead.
For the spectral range investigated, we find no evidence
of absorption across the indirect bandgap as the dominant
mechanism.
Last, we characterize the photo-emission stability over time
at different bunch charges and estimate the normalized peak
brightness Bp,norm. For best comparison of Bp,norm with ex-
isting literature on ultrafast tip-shaped electron sources [8, 9],
we calculate all quantities as normalized root-mean-squared
(rms) values and use the following definition:
Bp,norm =
Jp
4 · π2 · εx,norm · εy,norm
,
(3)
where Jp is the peak current, εi are the transverse emit-
tances and the subscript norm indicates normalized values.
As an upper bound for the transverse emittances, we mea-
sure the emission angles αi and assume homogeneous emis-
sion across the geometrical radius of the emitter (r=170 nm,
rrms = r
). Note that the effective source size and there-
√3
fore the emittance of tip-shaped emitters can be an order
of magnitude smaller as the curved surface induces corre-
lations between origin and transverse momentum [16, 39].
Photo-emission at 1932 nm and 40 eV electron energy yields
αx = 0.16(6) rad, αy = 0.15(9) rad, εx,norm = 0.20 nm rad
and εy,norm = 0.19 nm rad. Assuming that the emission dura-
tion matches the laser pulse duration, we calculate the normal-
ized rms peak brightness Bp,norm = 1.2 · 1012 A m−2 sr−1
for one electron per pulse, comparable to a femtosecond cold
field emitter at 15 electrons per pulse [9]. Because we use the
geometrical and not the effective source size and because we
consider currents of one electron per pulse, we consider this
peak brightness a lower bound.
The photo-emitted current is stable over a time scale of at
least half an hour at 256, 512 and 1932 nm with bunch charges
of 55, 32 and 0.75 electrons per pulse, respectively (Fig. 5).
With a stable 80 MHz Ti:Sa oscillator, the photo-current is sta-
ble over more than 12 hours and trillions of pulses. In contrast,
the photo-emission from an uncoated monocrystalline [310]-
oriented tungsten tip decays over time (Fig. 5); a comparable
e
s
l
u
p
r
e
p
s
n
o
r
t
c
e
l
E
60
50
35
30
0.8
0.7
0
5
a)
b)
4
256 nm
512 nm
1932 nm
15
10
Time (min)
20
25
Laser realigned
780 nm
80 MHz
NCD coated W
1
0.8
0.6
Bare W
d
e
z
i
l
a
m
r
o
N
t
n
e
r
r
u
c
o
t
o
h
p
0
2
4
6
8
10
12
14
Time (hours)
FIG. 5. a) Photo-emission at 1 kHz repetition rate is stable at 256,
512 and 1932 nm from the UV to the infrared both below and above
one electron per pulse. The noise is caused by laser power and point-
ing fluctuations.
b) Long-term photo-emission at 80 MHz and 780 nm is stable from a
diamond-coated tungsten tip (blue) and unstable from a bare tungsten
tip (red). The short term fluctuations (over 1 min) from the coated
and uncoated tip are 3% and 5%, the bunch charges at t=0 are 25 and
6.5 electrons per pulse, respectively.
behavior with even stronger decay was observed in a trans-
mission electron microscope (¯hω = 2.4 eV, p = 1 · 10−9 Pa
[9]). Schottky emitters in scanning electron microscopes
(¯hω = 3.6 eV, reduced barrier height Φeff = 1.6 eV [5] and
¯hω =4.7 eV, Φeff = 2.8-3 eV, p < 4· 10−8 Pa [10]) show a sim-
ilar behavior. Hence, nanodiamond-coated tungsten tips are
more stable than these emitters, especially at low photon ener-
gies. (Working with low photon energies can be advantageous
as the field enhancement at the apex [40] in combination with
the nonlinearity enhances forward emission.)
In DC field emission occasional jumps occur, typical for cold
field emission. The angular distribution in this emission mode
is even smaller compared to laser-induced emission.
We did not observe a change in laser-induced emission be-
haviour during our experiments with a laser fluence up to
30 mJ
cm2 peak intensity. Hence we find these
as lower bounds of the damage threshold for diamond-coated
cm2 , we
have measured 400 electrons per pulse. At these large bunch
charges, pulse broadening due to Coulomb repulsion is ex-
pected to be severe [10, 41], which is why we have focused
on smaller bunch charges.
tungsten tips. With 1932 nm pulses at 3.4 · 1011 W
cm2 and 3.4 · 1011 W
In conclusion, we have presented femtosecond laser-
induced electron emission from diamond-coated tungsten tips
at 235-350 nm, 512 nm, 780 nm and 1932 nm. Based on
the involved junctions between tungsten, diamond and the
graphitic grain boundaries, we have proposed an emission
model which explains our experimental data well. Individual
emission channels can be selected by proper choice of laser in-
tensity and wavelength. These channels are identified by the
number of photons needed to emit an electron. Stable photo-
electron current and the high brightness of the emitted elec-
trons are encouraging to further investigate diamond-coated
tungsten tips as an ultrafast electron source.
Before resubmission of this manuscript, we became aware of
new and related work [42].
ACKNOWLEDGEMENT
The authors acknowledge Mingjian Wu and Erdmann
Spiecker for the transmission electron micrograph, funding
from the Deutsche Forschungsgemeinschaft via grant SFB
953 "Synthetic Carbon Allotropes", from the European Re-
search Council through grant "Near Field Atto" and the Gor-
don and Betty Moore Foundation via Grant GBMF4744 "Ac-
celerator on a Chip International Program -- ACHIP".
∗ [email protected]
[1] J. C. H. Spence, High-Resolution Electron Microscopy, 4th ed.
(Oxford University Press, 2013).
5
[17] C. Ropers, D. R. Solli, C. P. Schulz, C. Lienau, and T. El-
saesser, Physical Review Letters 98 (2007).
[18] R. Bormann, M. Gulde, A. Weismann, S. V. Yalunin,
and
C. Ropers, Physical Review Letters 105 (2010).
[19] L. Wimmer, G. Herink, D. R. Solli, S. V. Yalunin, K. E.
Echternkamp, and C. Ropers, Nature Physics 10, 432 (2014).
[20] M. R. Bionta, S. J. Weber, I. Blum, J. Mauchain, B. Chatel, and
B. Chalopin, New Journal of Physics 18, 103010 (2016).
[21] C. Kealhofer, S. M. Foreman, S. Gerlich, and M. A. Kasevich,
Physical Review B 86 (2012).
[22] M. Bionta, B. Chalopin, A. Masseboeuf,
and B. Chatel,
Ultramicroscopy 159, 152 (2015).
[23] C. Li, X. Zhou, F. Zhai, Z. Li, F. Yao, R. Qiao, K. Chen,
and Q. Dai,
M. T. Cole, D. Yu, Z. Sun, K. Liu,
Advanced Materials 29, 1701580 (2017).
[24] V. Porshyn, V.
I. Kleshch, E. A. Obraztsova, A. L.
and A. N. Obraztsov,
Chuvilin, D. Lützenkirchen-Hecht,
Applied Physics Letters 110, 182101 (2017).
J. Ristein,
Journal of Applied Physics 88, 3667 (2000).
[25] J. B. Cui, M. Stammler,
and L. Ley,
[26] F. Himpsel, J. Knapp, J. VanVechten, and D. Eastman, Phys.
Rev. B 20, 624 (1979).
J.
Cui,
B.
[27] J.
Ristein,
and
L.
Ley,
Physical Review B 60, 16135 (1999).
[2] O. Bostanjoglo, R. Elschner, Z. Mao, T. Nink, and M. We-
[28] F.
Maier,
J.
Ristein,
and
L.
Ley,
ingärtner, Ultramicroscopy 81, 141 (2000).
Physical Review B 64, 165411 (2001).
[3] A. H. Zewail, Science 328, 187 (2010).
[4] A. Arbouet, G. M. Caruso,
in
Advances in Imaging and Electron Physics (Elsevier, 2018) pp.
1 -- 72.
and F. Houdellier,
[29] J.
B.
Cui,
J.
Ristein,
and
L.
Ley,
Physical Review Letters 81, 429 (1998).
[30] C. F. Davidson, G. B. Alexander,
and M. E. Wadsworth,
Metallurgical Transactions B 9, 553 (1978).
[5] D.-S. Yang, O. F. Mohammed,
and A. H. Zewail,
[31] A. Tafel, M. Wu, E. Spiecker, P. Hommelhoff, and J. Ristein,
Proceedings of the National Academy of Sciences 107, 14993 (2010).
Diamond and Related Materials 97, 107446 (2019).
[6] D. A. Plemmons, P. K. Suri,
and D. J. Flannigan,
Chemistry of Materials 27, 3178 (2015).
[32] J.
B.
Ristein,
Physical Review B 59, 5847 (1999).
Cui,
J.
and
L.
Ley,
[7] J. Sun, V. A. Melnikov, J. I. Khan, and O. F. Mohammed,
[33] H. Shiomi, H. Nakahata, T. Imai, Y. Nishibayashi, and N. Fu-
The Journal of Physical Chemistry Letters 6, 3884 (2015).
[8] A. Feist, N. Bach, N. R. da Silva, T. Danz, M. Möller, K. E.
Priebe, T. Domröse, J. G. Gatzmann, S. Rost, J. Schauss,
S. Strauch, R. Bormann, M. Sivis, S. Schäfer, and C. Ropers,
Ultramicroscopy 176, 63 (2017).
[9] F. Houdellier, G. Caruso, S. Weber, M. Kociak, and A. Arbouet,
jimori, Japanese Journal of Applied Physics 28, 758 (1989).
[34] R.
F. Willis,
B.
Fitton,
and G.
S.
Painter,
Physical Review B 9, 1926 (1974).
[35] J. Cui, J. Ristein, M. Stammler, K. Janischowsky, G. Kleber,
and L. Ley, Diamond and Related Materials 9, 1143 (2000).
[36] P. Dean, E. Lightowlers, and D. Wight, Physical review 140,
Ultramicroscopy 186, 128 (2018).
A352 (1965).
[10] M. Kozák, J. McNeur, N. Schönenberger, J. Illmer, A. Li,
and P. Hommelhoff,
A. Tafel, P. Yousefi, T. Eckstein,
Journal of Applied Physics 124, 023104 (2018).
[11] P. Hommelhoff, Y. Sortais, A. Aghajani-Talesh, and M. A. Ka-
[37] F. Giustino, S. G. Louie, and M. L. Cohen, Physical Review
Letters 105 (2010).
[38] S. Logothetidis, J. Petalas, H. M. Polatoglou, and D. Fuchs,
Physical Review B 46, 4483 (1992).
sevich, Physical Review Letters 96 (2006).
[39] D. Ehberger,
[12] P. Hommelhoff, C. Kealhofer, and M. A. Kasevich, Physical
Review Letters 97 (2006).
[13] B. Barwick, C. Corder, J. Strohaber, N. Chandler-Smith,
C. Uiterwaal, and H. Batelaan, New Journal of Physics 9, 142
(2007).
[14] M. Krüger, M.
Schenk,
and
P. Hommelhoff,
Nature 475, 78 (2011).
[15] M. Bionta, B. Chalopin,
A. Masseboeuf, P. Moretto-Capelle,
Journal of Modern Optics 61, 833 (2013).
J. Champeaux,
S. Faure,
and B. Chatel,
[16] S. Meier, T. Higuchi, M. Nutz, A. Högele, and P. Hommelhoff,
Applied Physics Letters 113, 143101 (2018).
Noe,
J.
P.
Physical Review Letters 114, 227601 (2015).
Högele,
A.
J. Hammer, M. Eisele, M. Krüger,
Hommelhoff,
and
[40] S. Thomas, G. Wachter, C. Lemell, J. Burgdörfer, and P. Hom-
melhoff, New Journal of Physics 17, 063010 (2015).
[41] B. Cook and P. Kruit, Applied Physics Letters 109, 151901 (2016).
[42] M. Borz, M. H. Mammez,
S.
I. Blum, J. Houard, G. D.
Idlahcen, A. Haboucha,
and A. Vella,
F. Delaroche,
Costa,
A. Hideur, V. I. Kleshch, A. N. Obraztsov,
Nanoscale 11, 6852 (2019).
|
1810.05920 | 1 | 1810 | 2018-10-13T20:01:29 | Quick-cast: A method for fast and precise scalable production of fluid-driven elastomeric soft actuators | [
"physics.app-ph",
"cond-mat.soft"
] | Fluid-driven elastomeric actuators (FEAs) are among the most popular actuators in the emerging field of soft robotics. Intrinsically compliant, with continuum of motion, large strokes, little friction, and high power-to-weight ratio, they are very similar to biological muscles, and have enabled new applications in automation, architecture, medicine, and human-robot interaction. To foster future applications of FEAs, in this paper we present a new manufacturing method for fast and precise scalable production of complex FEAs of high quality (leak-free, single-body form, with <0.2 mm precision). The method is based on 3d moulding and supports elastomers with a wide range of viscosity, pot life, and Young's modulus. We developed this process for two different settings: one in laboratory conditions for fast prototyping with 3d printed moulds and using multi-component liquid elastomers, and the other process in an industrial setting with 3d moulds micromachined in metal and applying compression moulding. We demonstrate these methods in fabrication of up to several tens of two-axis, three-chambered soft actuators, with two types of chamber walls: cylindrical and corrugated. The actuators are then applied as motion drivers in kinetic photovoltaic building envelopes. | physics.app-ph | physics | Quick--cast: A method for fast and precise scalable production of fluid--
driven elastomeric soft actuators
Bratislav Svetozarevic1*, Moritz Begle1*, Stefan Caranovic1, Zoltan Nagy2, Arno Schlueter1
1 Architecture and Building Systems, Department of Architecture, ETH Zurich, Switzerland
2 Intelligent Environments Laboratory, Department of Civil, Architectural and Environmental
Engineering, The University of Texas at Austin, TX 78712, USA
*These authors contributed equally to the work.
Competing interests: B.S., M.B., S.C., and A.S. are inventors on a patent application (no. EP
17201677.6) submitted by ETH Zurich that covers this manufacturing method. The remaining authors
declare that they have no competing interests.
Abstract
Fluid--driven elastomeric actuators (FEAs) are among the most popular actuators in the emerging field
of soft robotics. Intrinsically compliant, with continuum of motion, large strokes, little friction, and high
power--to--weight ratio, they are very similar to biological muscles, and have enabled new applications
in automation, architecture, medicine, and human--robot interaction. To foster future applications of
FEAs, in this paper we present a new manufacturing method for fast and precise scalable production
of complex FEAs of high quality (leak--free, single--body form, with <0.2 mm precision). The method is
based on 3d moulding and supports elastomers with a wide range of viscosity, pot life, and Young's
modulus. We developed this process for two different settings: one in laboratory conditions for fast
prototyping with 3d printed moulds and using multi--component liquid elastomers, and the other process
in an industrial setting with 3d moulds micromachined in metal and applying compression moulding.
We demonstrate these methods in fabrication of up to several tens of two--axis, three--chambered soft
actuators, with two types of chamber walls -- cylindrical and corrugated. The actuators are then applied
as motion drivers in kinetic photovoltaic building envelopes.
Keywords: Fluidic elastomer actuators, soft robotics, scalable production, two--axis, corrugated
1
Introduction
Soft robotics is a rapidly growing field offering novel actuators [1], end effectors [2], skins [3], and
even entire robots [4] made of compliant material. They carry several important advantages compared
to classical rigid--bodied robots, such as intrinsic compliance, large power--to--weight ratio, low friction,
simpler control, and cheaper fabrication, and have enabled solutions to challenging problems in
automation (e.g. universal grasping [5]), architecture (e.g. adaptive solar façade [6]), medicine (e.g.
minimally invasive surgery [7]), and human--robot interaction (e.g. safe continuum manipulation [8]). As
contrary to rigid--bodied robots, where components are selected from the standard set of components
(e.g. electromotor, gearbox, spring) and designs follow rigid--body kinematics principles (rigid links
connected by discrete joints), soft robotics offers much more design freedom, with a plethora of soft
materials available (e.g. intrinsically compliant -- elastomers, and extrinsically compliant, such as wires
1
of shape memory alloy [9], thermoplastics [10], and jamming particles [5]), a range of actuation
principles (fluidic--, thermal--, humidity--, pH--, magnetic--, and electric--driven), and freedom in defining
components shape, structure, and working principle. Consequently, the main challenge in working with
soft components is in their design and manufacturing. Compared to design, manufacturing techniques
are directly linked to the performance of the actuators, which often brought major advances in the past
[11] -- [14].
Among soft actuators, one of most popular types are fluid--driven elastomeric actuators (FEAs).
These actuators contain fluidic pathways within their soft bodies. When they are filled up with
pressurised fluid (gas or liquid), the surrounding elastic material expands, leading to a change of the
outer shape of the actuator. Depending on the geometry of fluidic pathways and shape of the actuator,
as well as on the pressurisation method (inflation or deflation), different output motion paths can be
obtained, such as expansion, contraction, bending, or twisting [15]. Besides motion diversity, FEAs
exhibit advanced actuation features, similar to those of biological muscles, such as continuum of
motion, high power--to--weight ratio, large strokes, and little friction. Also, due to use of elastomers, FEAs
are intrinsically compliant and can passively mitigate external disturbances (acting as a spring--damper
system), allowing for reduction of control complexity and safe interaction with the environment and
humans. For existing reviews of FEAs, please see [12], [16].
Even though several methods for manufacturing of FEAs exists, they are of limited suitability for
scalable production due to either high process complexity, long total fabrication time, difficulties in
achieving leak--free actuators, or limited support for different types of elastomers. To strengthen future
applications of soft robotics (e.g. soft--robotic driven adaptive solar facades [17]), a manufacturing
method suitable for scalable production of FEAs is necessary.
In this paper we present a new manufacturing method for soft actuators that is suitable for fast and
precise scalable production of complex FEAs of high quality (leak--free, single--body, with <1 mm
precision). This method is based on 3d moulding, where outer walls of an FEA and internal voids (fluidic
pathways) are formed in a single casting process. The outer mould consists of several parts that can
be easily disassembled, while internal voids are created by pulling out the mould parts. In that, the
elasticity of elastomers enables pulling out of the three internal mould parts.
The advantages of this method are: (i) speed -- a single casting method with the total production
time of approximately one hour per actuator, (ii) quality of the actuators -- single--body (no gluing or
bonding required), leak--free, and precise (<0.2 mm) actuators, and (iii) a wide range of supported
elastomers in terms of viscosity, pot life, and Youngs' modulus. We call the method shortly Quick--cast.
We developed this process for two different settings: one for fast prototyping in laboratory conditions
with 3d printed moulds and using multi--component pourable elastomers, and the other process in an
industrial setting with 3d moulds micromachined in metal and using compression moulding. We
demonstrate these methods in fabrication of up to several tens of two--axis, three--chambered FEAs,
with two types of chamber walls -- cylindrical and corrugated. We tested them in terms of pressure--
deflection characteristics and motion repeatability. The actuators are then used as motion drivers in two
real--world building--scale prototypes of kinetic photovoltaic envelopes, with 50 modules at ETH Zurich
Hoengerberg Campus [6] (Fig. 1) and with 30 modules at EMPA, Duebendorf [17], both in Switzerland.
2
These novel dynamic building envelopes are very lightweight, resilient to weather condition, robust to
wind loads, with low self--power consumption, and can improve a building's net energy demand through
adaptive shading and electricity generation [17], [18]. Besides the three--chambered actuator design
reported here, the Quick--cast method is suitable for other single or multi--chamber designs.
2 Methods and materials
2.1 Actuator design
We developed the Quick--cast manufacturing method from the need to fabricate a large number
(>100) of two--axis three--chambered single--body FEAs, called SoRo--Tracks (Fig. 1h) [19], in order to
apply them as motion drivers in adaptive photovoltaic façades [6], [17]. In terms of motion capabilities,
we decided for two--axis actuation instead of single--axis. Two--axis allow for both vertical (altitude angle)
and horizontal positioning (azimuth angle) of façade elements, which increases the yearly PV energy
output by 10--15% compared to single--axis trackers [20]. Furthermore, the possibility to move each of
the façade elements in two--axis provides architects with the maximum design freedom when it comes
to aesthetic expressions. In terms of actuator design requirements, we wanted to reduce its complexity
as much as possible, as well as to improve its visual appearance and allow for easier cleaning by having
a single--body actuator, instead of, for example, multiple single--axis soft actuators.
We considered two geometries of internal voids: cylindrical and corrugated. The cylindrical is
simpler in terms of design complexity and it was the very first intuitive design to test the method. The
corrugated design has more complex geometry, offers much more design parameters (e.g. ribs
geometry, walls thicknesses), and therefore requires more effort to obtain a well performing actuator.
In the design of bending actuators, it is important to maximise the forces acting on bottom and top discs
of the actuator and minimise the radial expansion of rubber. This would maximise the bending moment
of the actuator. Also, it is preferable that the actuator stays in the linear region of rubber elasticity during
inflation as then it allows for implementation of open--loop (sensorless) control, instead of sensor--based
closed loop control. In our case, the ribs are very precisely designed in 3d in such a way that the inner
ribs have thinner walls than the outer ones, and therefore they unfold to the outside, increasing the
distance between the outer ribs. The outer ribs have a rather high thickness preventing overall radial
expansion of the chamber, and therefore maximising the bending moment. In this way, we avoid large
expansions of the rubber and the corrugated SoRo--Track actuator functions in the linear region (Fig.
1g).
2.2 State--of--the--art manufacturing methods
We defined the following process properties as important for the scalable production in an
industrial setting: (i) achieving leak--free actuators, (ii) number of manufacturing steps, (iii) steps
complexity, (iv) required gluing / bonding step, (v) total fabrication time, (vi) precision, (vii) supporting
wide range of elastomers, and (ix) possibility to base it on an already established industrial process. In
the following text we review the state--of--the--art methods in terms of these properties.
At macro scale (cm to m), moulding and direct 3d printing of soft materials are the most popular
manufacturing techniques, due to the fact that elastomers can be processed as liquids -- multi--
3
component elastomers that harden over time, with the pot life from min to hours. A typical moulding
technique is a 2d moulding process (a layup process) where a part of the actuator containing the fluid
pathways is moulded first, and then it is glued to another, stiffer elastomer or fabric, forming networks
of pneumatic channels, PneuNets [13]. This method is very popular in the community, due to the
availability of 3d printers for fabrication of plastic moulds (ABS, PLA, etc). The drawback of this process
is in the required gluing or bonding of separately produced elastomer pieces, which might result in weak
spots prone to delamination and leaking at higher pressures. Also, the process consists of several
steps, requiring several hours for fabrication of a single actuator.
To overcome the drawbacks of the 2d moulding process, a 3d moulding technique using wax cores
has been proposed [14]. In this process, first, the wax cores are casted, and then they are used in the
second casting step for making fluidic pathways. This process enables fabrication of very complex
fluidic pathways in a single--body actuator form (avoiding gluing), with uniform material properties across
the actuator. This process is, however, more time consuming than 2d moulding, as it consists of two
castings and, in addition, curing (cross--linking of elastomer chains) cannot be made faster by heating
up the mould, due to the low melting point of the beeswax of 62--64ºC. Also, the wax needs to be
removed and cleaned from the fluidic pathways afterwards, which is an additional step.
Besides moulding techniques, direct 3d printing of soft materials offers advanced possibilities, such
as varying elastomer parameters across the actuator volume and achieving higher geometrical
complexity [21] -- [23]. However, due to difficulties in obtaining cross--links between polymer chains during
the layering process, obtaining leak--tight actuators is often challenging. There are multiple parameters
of the 3d printing process (e.g. layer height, nozzle size, and extrusion temperature) that need to be
properly tuned for each material, and the 3d printed actuators need to be post--processed to achieve
polymer cross--linking [24]. Moreover, current soft--material 3d printing techniques have lower precision
than moulding processes and fabrication time per single actuator is much longer.
In terms of manufacturing methods suitable for scalable production, only one method has been
proposed in the literature to the best of our knowledge. This process is based on a standard industrial
process, rotational casting, where uncured elastomer is casted as a hollow structure in a closed mould
[25]. The limitation of this process is that it supports elastomers with limited range of viscosity, pot life,
and Youngs' modulus (uncured elastomer needs to be of a certain viscosity in order to be processed in
rotational casting). Also, the precision of internal voids depends on how good the models of the
rotational casting process and elastomer curing process are, which might be additional limiting factor
for the process applicability, in particular for achieving complex and precise fluid pathways.
Furthermore, it is a multi--step process, where multiple individual balloons (chambers) obtained in the
rotational casting step are then used to produce the final actuator by casting the surrounding elastomer
body. In that, firm positioning of the individual balloons is required, which typically needs additional
engineering effort.
4
Fig. 1. Quick--cast method applied to fabrication of FEAs with cylindrical (a) and corrugated (b) walls. See--through
images of cylindrical (c) and corrugated (d) two--axis, three--chambered SoRo--Track actuators made of semi--
translucent ELASTOSIL® VARIO from Wacker Chemie AG. Corrugated SoRo--Tracks made of (e) ELASTOSIL®
VARIO in laboratory conditions with 3d printed moulds. (f) Corrugated SoRo--Track industrially manufactured from
Neoprene rubber. (g) Pressure--deflection characteristics of cylindrical and corrugated SoRo--Tracks. (h) A single
module of Adaptive Solar Façade driven by SoRo--Track. Outside view (i) and view from inside (j) of Adaptive Solar
Façade with 50 cylindrical SoRo--Tracks at ETH Zurich, Hönggerberg Campus, Switzerland.
5
2.3 Quick--cast manufacturing method in laboratory conditions
We aimed for developing a manufacturing method based on 3d printed moulds as it allows for quick
prototyping and testing of soft actuators. After the soft actuator and its mould was modelled in
Rhinoceros® (Step A), it was 3d printed from Nylon on a laser sintering system EOS P396 (Step B).
The details of the mould are presented in Fig. A.2. After assembling the mould, a multi--component liquid
elastomer ELASTOSIL® VARIO from Wacker Chemie AG is poured (step 1). VARIO is a two--component
silicone rubber, with Shore A hardness 15 and 40, respectively, which allow for achieving any Shore A
hardness between 15 and 40. The influence of the mixing ratio on rubber parameters (tensile strength,
elongation at break, and tear strength) are given in Fig. A.1. We mixed the two elastomers in a 3:1 ratio
and added 15% of the catalyst (resulting Shore A index 30). After pouring, the mixture was degassed
to remove the trapped air--bubbles (Step 2). Then, the curing of elastomer was done in an oven at 75ºC
for about 20 min (Step 3). This is much faster than curing at the room temperature, which takes several
hours. Finally, the mould is disassembled in two steps (Step 4). First, the two parts forming the outer
shape of the actuator are taken apart. Second, the inner cores are pulled out of the soft actuator (Fig.
A.3 and A.5). Pulling out of the hard cores is possible due to the large elongation at break of elastomers
(in this case it is around 500%;; may be up to 1000%, e.g. of EcoflexÔ 00--50 from Smooth--On, Inc.). To
allow for easer pulling out of the inner cores, we were slightly inflating the chambers. The total
fabrication time of a single actuator, including the mould assembling, takes approximately one hour. To
produce the next actuator, one needs to repeat steps 1 to 4. Obtained actuators may be then tested for
performance (Step C) and the information fed back to the modelling software to inform the change of
the design. The images of this manufacturing process for cylindrical SoRo--Track and using VARIO are
given in Fig. A.3. Also, we show the manufacturing of the same actuator with ELASTOSIL® M4601 (Fig.
A.4.).
6
Fig. 2. Quick--cast manufacturing method for fast prototyping of FEAs in laboratory conditions with 3d printed
moulds and using multi--component liquid elastomers.
2.4 Quick--cast manufacturing method in industry
The Quick--cast method in laboratory conditions was adapted to industrial setting, in order to obtain
the industry--quality actuators, with repeatable characteristics over a large number of samples. We found
compression moulding to be a suitable process, which allows for both multi--component elastomers (e.g.
ELASTOSIL® M4601 and ELASTOSIL® VARIO 15/40) and rubbers (e.g. Neoprene and EPDM) to be
used. We only tested this method with rubbers. Because of that, the mould is heated up to 350ºC and
the uncured rubber is injected under pressure into the metal mould (Step 1). After curing at this elevated
temperature for 20 minutes (Step 2), the mould is cooled down and taken apart (Step 3). The
disassembling of the mould happens in the same way as in the process in laboratory conditions. First,
the outer moulds are separated, and then the inner cores are pulled out. As Neoprene has higher
hardness (Shore A 35) compared to the hardness of the ELASTOSIL® VARIO mixture (Shore A 30)
that we used in the laboratory conditions, it was much harder to remove the cores from the corrugated
actuator. Therefore, we decided to take them out on the other side through a slightly larger opening
(radius 4 mm on the top, rather than 1.75 mm on the bottom), and then those small holes were closed
at the end by gluing conical plugs. The total production time for one actuator is about 1 hour. The images
of the industrial process are shown in Fig. A.5.
Fig. 3. Quick--cast manufacturing method for scalable production of FEAs in industrial setting
3 Results and discussion
We tested two batches of actuators with 10 actuators in each batch. The actuators are randomly
selected from larger batches of about 50 actuators. The first batch consists of cylindrical SoRo--Track
7
CNC - milling of the metal mould for Quick-castingUnmoldingElastic material allows to pull out the centercoreCuring at an elevated temperature23Injection moulding1~20 min~10 minproduction time: ~1 hourtwo-axis soft (FEA) actuator~15 minrubber(Neoprene) actuators made in laboratory from ELASTOSIL® VARIO 15/40 with the mixture ration 3:1 (resulting
hardness Shore A 30). The second batch consists of corrugated actuators manufactured under
industrial conditions from Neoprene rubber (hardness Shore A 35). Our goal with these tests is to
compare the actuators within each batch among each other to see what degree of performance
similarity can be achieved. All the tests were done using the pneumatic control system and inertial
measurement unit as orientation sensor, as described in [19].
3.5 Pressure--deflection characterisation
The pressure--deflection behaviour of cylindrical actuators manufactured in laboratory conditions is
shown in Fig. 4a and 4b, for roll and pitch angles, respectively. Pressure--deflection curves for each of
the three chambers are presented. The measurements are done with the SoRo--Track actuator mounted
at 45º roll (vertical angle), resembling the mounting angle on the façade. In that case, one chamber is
below the other two chambers. Therefore, in Fig. 4a we can see that inflation of chamber 1 moves the
panel vertically up, thus increasing the roll angle, while the inflations of chambers 2 and 3 cause a
bending of the panel towards ground, thus decreasing the roll angle. In azimuth angle (pitch angle), the
inflation of chamber 2 decreases the pitch angle, the inflation of chamber 3 increases the pitch angle,
while chamber 1 does not have any influence on the pitch angle. This can be also seen on roll--pitch
graph shown in Fig. 4c, where the main directions of actuation nicely follow the indicated 120º angle
difference.
Regarding the differences among the actuators within the same batch, we can see that there is a
significant spread in the obtained pressure-- deflection curves (22%--26% difference in pressure change
for 7%--23% in achieved angles). One reason for this is in manual preparation of ELASTOSIL® VARIO
15/40 mixture for each actuator before pouring. The second reason is in the insufficient precision of the
3d printed nylon mould, hence small offsets in the fixation of the inner cores during mould assembly,
as well as a slight bending of the metal rods on which these cores are mounted. These metal rods are
rather thin (3 mm stainless steel rod) and they bend during the manual pulling out of the inner cores
(Fig. A.3).
The pressure--deflection curves for corrugated SoRo--Track actuators made from Neoprene rubber
and manufactured industrially are shown in Fig. 4e--g. We can see that the pressure--deflection curves
overlap for all 10 of tested actuators is very good (1% in pressure change for 12%--16% in achieved
angles). As the mould is micromachined from stainless steel, there are no geometrical differences
between these actuators. The only source of uncertainty is in the preparation of the rubber mix, which,
due to the early development phase of establishing of this process, is still done manually for each
actuator separately. However, the absolute angle error is about +/-- 3º, which is acceptable for controlling
the system in feed--forward mode, i.e. only using pre--calculated control inputs, without the need for
feedback signals from sensors, thus simplifying the control and reducing the actuator implementation
costs.
8
3.6 Repeatability of motion
One of the actuators from each batch is randomly chosen for testing the repeatability of motion (Fig.
4d and 4h). Chamber 1 was inflated and then deflated 50 times. Both actuators show very good
repeatability over number of cycles.
Fig. 4. Pressure--deflection (a--c and e--g) and repeatability characterisation (d and h) of cylindrical
two--axis three--chamberd (SoRo--Track) actuator made of ELASTOSIL® VARIO in laboratory conditions
and of corrugated actuator made of Neoprene in industrial conditions.
3.7 Comparison with state--of--the--art methods
We tested the manufacturing of SoRo--Track actuator using some the state--of--the--art methods
(PneuNets [13], Wax--cores [14], and Rotation casting [25]). We did not have access to any of the direct
3d printing of soft materials techniques [21] -- [23]. It turned out that manufacturing of SoRo--Track was
9
challenging for some of the above methods, due to its specific shape, where the maximum radius of
fluidic pathways is several times (5 times) larger than the radius of the fluidic pathway entrance. The
actuators fabricated using PneuNets approach [13] started to delaminate after certain time. We did not
have major difficulties with Wax--cores [14], but due to the softness of wax and very complex 3d
geometry of SoRo--Track, we were not always sure that all parts of the wax cores stayed intact during
demoulding them and afterward while pouring of elastomers over them in the main fabrication step. In
terms of using the rotational casting method [25], we could not tune it for ELASTOSIL® M4601 to obtain
the precise geometry of the inner cores. However, this could have been due to our lack of expertise in
this method. The total time for manufacturing of a single SoRo--Track actuator using the above methods
was between 6h to 10h, after basic preparations were done. We are confident in these numbers, as all
of the above methods require multiple steps (between 3 and 4).
Based on our review of the state--of--the--art methods provided in Section 2.2 and based on the results
of our Quick--cast method, we provide the summary of the comparison in Table 1. With the table fields
shaded in grey we emphasize which methods show the best performance for each feature analysed.
Overall, our method shows comparable or advanced performance across all features besides one
(geometrical complexity of fluidic pathways), where it shows moderate performance. Our method shows
advance performance in terms of number of steps (reduction to a single step), precision, and total
fabrication time (reduction from one working day to one hour).
In terms of the geometrical complexity of fluidic pathways, our process has certain limitations defined
by the materials properties and certain geometrical parameters of fluidic pathways. The difference
between the radius of the entrance of the fluidic pathways and the maximum radius of the fluidic
pathways at any point, should not be larger than the elongation at break of a soft material. Given that
the elongation at break of soft materials is in the range of 100% to 1000%, this may not be a critical
limitation. The elongation at break is inversely correlated with the elastomer hardness. This means that
for stronger soft actuators (with higher elastomer hardness), the elongation at break is getting smaller.
A designer of a soft actuator would need to check if the design satisfies this limitation before fabricating
the actuator.
10
Features
PneuNets [13] Wax--cores
[14]
Rotation
casting [25]
3d printing of
soft materials
[21] -- [23]
Quick--cast
Leak--free actuator
Number of steps
(yes)
Requires
bonding
3
Steps complexity
low
yes
yes
4
low / moderate moderate /
3
high
(yes)
1
yes
1
moderate / high
low
Additional engineering
effort required to tune
the process for another
elastomer
Single--body form (no
bonding / gluing step)
Total fabrication time
for SoRo--Track
actuators
Precision
Supported elastomers
Geometrical complexity
of fluidic pathways
Based on standard
industrial process
no
no
6--8h
low
yes
6--8h
(< 1 mm)
< 1 mm
wide range
wide range
low / moderate moderate /
high
moderate /
high
moderate / high
no
yes
10h
yes
(est.) 12h
yes
1h
unknown
~ 1 mm
limited range unknown
low /
moderate
moderate / high moderate
< 0.2 mm
wide range
no
no
yes
no
yes
Table 1. Comparison of Quick--cast manufacturing method with state--of--the--art methods. The fields
highlighted in grey indicate the best performing method for that feature.
4 Conclusions
In summary, we demonstrated a new manufacturing method for scalable production of complex,
industrial grade FEAs. Compared to the state--of--the--art methods, Quick--cast offers advantages in terms
of: (i) speed (total fabrication time of one hour, compared to typical fabrication time of one working day,
8h, of other methods), (ii) quality of the actuator (leak--free, single--body, with < 1 mm precision), and (iii)
the range of supported elastomers in terms of viscosity, pot life, and Young's modulus. We developed
this process for two different settings: for laboratory conditions with 3d printing moulds for fast
prototyping and using multi--component rubbers, and for industrial setting with moulds micromachined
in metal and using compression moulding. We showed the application of those methods in the scalable
manufacturing of the two--axis, three--chamber FEA actuator SoRo--Track. These actuators are then
applied as motion drivers in adaptive solar building facades. We tested two batches of two types of
actuators, with cylindrical and corrugated chamber walls, in terms of pressure--deflection and
repeatability of motion. The industrial process showed advantages over the laboratory process in terms
of repeatable performance over a large number of samples -- the pressure--deflection curves overlap
very well for 10 actuators, randomly selected from a batch of more than 50.
11
The only potential limitation of our approach is in the maximal achievable geometrical complexity of
the fluidic pathways. The difference between the radius of the entrance of the fluidic pathways and the
maximum radius of the fluidic pathways at any point, should not be larger than the elongation at break
of a soft material. The elongation at break is inversely correlated with the elastomer hardness, which
means that for a stronger soft actuator (with higher elastomer hardness), the elongation at break is
smaller. Given that the elongation at break of soft materials is in the range of 100% to 1000%, this may
not be a critical limitation.
In the paper we demonstrate the scalable production of two--axis three--chambered FEA called SoRo--
Track. The short manufacturing time of the Quick--cast method here allowed for rapid scalable
manufacturing of large number (>100) of these actuators and has enabled the construction of several
real--world building--scale prototypes of Adaptive Solar Façades [17], [6]. The Quick--cast, however, is
also suitable for other single or multi--chamber designs. The method is also suitable for larger scale
actuators. We believe, that this novel manufacturing method for fast and precise scalable production of
complex, high quality FEAs will foster and enable novel application of soft robotics in the future.
5 Acknowledgements
We acknowledge support from the Building Technologies Accelerator program of Climate--KIC. We
acknowledge the support from our industrial partner Maagtechnic AG, Duebendorf, Switzerland, on
establishing the industrial manufacturing process at their facilities.
6 References
[1] C. Majidi, "Soft Robotics: A Perspective -- Current Trends and Prospects for the Future," Soft
Robot., vol. 1, no. 1, pp. 5 -- 11, Mar. 2014.
[2] R. Deimel and O. Brock, "A novel type of compliant and underactuated robotic hand for dexterous
grasping," Int. J. Robot. Res., vol. 35, no. 1 -- 3, pp. 161 -- 185, Jan. 2016.
[3] S. A. Morin, R. F. Shepherd, S. W. Kwok, A. A. Stokes, A. Nemiroski, and G. M. Whitesides,
"Camouflage and Display for Soft Machines," Science, vol. 337, no. 6096, pp. 828 -- 832, Aug.
2012.
[4] M. Wehner et al., "An integrated design and fabrication strategy for entirely soft, autonomous
robots," Nature, vol. 536, no. 7617, pp. 451 -- 455, Aug. 2016.
[5] E. Brown et al., "Universal robotic gripper based on the jamming of granular material," Proc. Natl.
Acad. Sci., vol. 107, no. 44, pp. 18809 -- 18814, Nov. 2010.
[6] Z. Nagy et al., "The Adaptive Solar Facade: From concept to prototypes," Front. Archit. Res., vol.
5, no. 2, pp. 143 -- 156, Jun. 2016.
[7] M. Cianchetti et al., "Soft Robotics Technologies to Address Shortcomings in Today's Minimally
Invasive Surgery: The STIFF--FLOP Approach," Soft Robot., vol. 1, no. 2, pp. 122 -- 131, Jun. 2014.
[8] S. Sanan and C. G. Atkeson, "A continuum approach to safe robots for physical human
interaction," in In Int'l Symposium on Quality of Life Technology, 2011.
[9] H.--T. Lin, G. G. Leisk, and B. Trimmer, "GoQBot: a caterpillar--inspired soft--bodied rolling robot,"
Bioinspir. Biomim., vol. 6, no. 2, p. 026007, 2011.
12
[10] N. Cheng et al., "Design and analysis of a soft mobile robot composed of multiple thermally
activated joints driven by a single actuator," in 2010 IEEE International Conference on Robotics
and Automation, 2010, pp. 5207 -- 5212.
[11] D. Rus and M. T. Tolley, "Design, fabrication and control of soft robots," Nature, vol. 521, no.
7553, pp. 467 -- 475, May 2015.
[12] P. Polygerinos et al., "Soft Robotics: Review of Fluid--Driven Intrinsically Soft Devices;;
Manufacturing, Sensing, Control, and Applications in Human--Robot Interaction: Review of Fluid--
Driven Intrinsically Soft Robots," Adv. Eng. Mater., vol. 19, no. 12, p. 1700016, Dec. 2017.
[13] F. Ilievski, A. D. Mazzeo, R. F. Shepherd, X. Chen, and G. M. Whitesides, "Soft Robotics for
Chemists," Angew. Chem. Int. Ed., vol. 50, no. 8, pp. 1890 -- 1895, Feb. 2011.
[14] A. D. Marchese, R. K. Katzschmann, and D. Rus, "A Recipe for Soft Fluidic Elastomer Robots,"
Soft Robot., vol. 2, no. 1, pp. 7 -- 25, Mar. 2015.
[15] R. V. Martinez, C. R. Fish, X. Chen, and G. M. Whitesides, "Elastomeric Origami: Programmable
Paper--‐‑Elastomer Composites as Pneumatic Actuators," Adv. Funct. Mater., vol. 22, no. 7, pp.
1376 -- 1384, Apr. 2012.
[16] B. Gorissen, D. Reynaerts, S. Konishi, K. Yoshida, J.--W. Kim, and M. De Volder, "Elastic Inflatable
Actuators for Soft Robotic Applications," Adv. Mater., vol. 29, no. 43, p. n/a--n/a, Nov. 2017.
[17] B. Svetozarevic et al., "Soft robotic driven kinetic photovoltaic building envelope for adaptive
energy and comfort management," Nat. Energy Rev., 2018.
[18] P. Jayathissa, M. Luzzatto, J. Schmidli, J. Hofer, Z. Nagy, and A. Schlueter, "Optimising building
net energy demand with dynamic BIPV shading," Appl. Energy, vol. 202, pp. 726 -- 735, Sep. 2017.
[19] B. Svetozarevic et al., "SoRo--Track: a two--axis soft robotic platform for solar tracking and building--
integrated photovoltaic applications," in Robotics and Automation (ICRA), 2016 IEEE International
Conference on, 2016, pp. 4945 -- 4950.
[20] H. Mousazadeh, A. Keyhani, A. Javadi, H. Mobli, K. Abrinia, and A. Sharifi, "A review of principle
and sun--tracking methods for maximizing solar systems output," Renew. Sustain. Energy Rev.,
vol. 13, no. 8, pp. 1800 -- 1818, Oct. 2009.
[21] B. N. Peele, T. J. Wallin, H. Zhao, and R. F. Shepherd, "3D printing antagonistic systems of
artificial muscle using projection stereolithography," Bioinspir. Biomim., vol. 10, no. 5, p. 055003,
2015.
[22] H. K. Yap, H. Y. Ng, and C.--H. Yeow, "High--Force Soft Printable Pneumatics for Soft Robotic
Applications," Soft Robot., vol. 3, no. 3, pp. 144 -- 158, Sep. 2016.
[23] H. Rodrigue, B. Bhandari, W. Wang, and S.--H. Ahn, "3D soft lithography: A fabrication process
for thermocurable polymers," J. Mater. Process. Technol., vol. 217, pp. 302 -- 309, Mar. 2015.
[24] A. Zolfagharian, A. Z. Kouzani, S. Y. Khoo, A. A. A. Moghadam, I. Gibson, and A. Kaynak,
"Evolution of 3D printed soft actuators," Sens. Actuators Phys., vol. 250, pp. 258 -- 272, Oct. 2016.
[25] H. Zhao, Y. Li, A. Elsamadisi, and R. Shepherd, "Scalable manufacturing of high force wearable
soft actuators," Extreme Mech. Lett., vol. 3, pp. 89 -- 104, Jun. 2015.
13
Appendix A. Supplementary data
Fig. A.1. ELASTOSIL® VARIO is a two--component silicone rubber with variable Shoare A hardness.
Influence of blending ratio of VARIO 15 : VARIO 40 on rubber parameters.
Fig. A.2. Exploded view of the nylon mould for the two--axis three--chambered soft actuators (SoRo--
Track) with cylindrical walls. For corrugated walls, the outer moulds are the same, but the "bullets"
forming the internal voids are different (see Fig. A.5).
14
Fig. A.3. Manufacturing of cylindrical two--axis, three--chambered FEA actuator (SoRo--Track) in
laboratory conditions with 3d printed Nylon mould from ELASTOSIL® VARIO 15/40. a) Process steps.
b) Mould geometry. c) Images of fabrication steps.
Fig. A.4. Manufacturing of cylindrical SoRo--Track actuator in laboratory conditions using
ELASTOSIL® M4601.
15
Fig. A.5. Manufacturing of corrugated two--axis three--chamebred FEA (SoRo--Track) actuator in
laboratory conditions with 3d printed Nylon mould using ELASTOSIL® VARIO 15/40. a) Process steps.
b) Mould geometry. c) Images of fabrication steps.
Fig. A.6. Quick--cast industrial manufacturing of corrugated two--axis three--chambered FEA (SoRo--
Track) based on compression moulding and chloroprene rubber.
16
|
1806.09159 | 1 | 1806 | 2018-06-24T14:44:33 | Structural phase transition, grain growth and optical properties of uncompensated Ga-V co-doped TiO2 | [
"physics.app-ph"
] | Effect of uncompensated Ga-V co-doping on structural phase transition, grain growth process and optical properties of TiO2 is reported here. Inhibition of phase transition due to co-doping is confirmed by X-ray diffraction measurement. Activation energy of phase transition increases from 120KJ/mol (x=0) to 140 KJ/mol (x=0.046) due to Ga-V co-doping. In anatase phase, lattice constants increase by the effect of Ga3+ interstitials. This results in inhibition of phase transition. Anatase phase becomes stable up to ~650 C in co-doped sample whereas for pure TiO2 phase transition starts in between 450-500 C. In anatase phase, strain increases due to co-doping which reduces crystallite size. In rutile phase, grain growth process is enhanced due to co-doping and particles show a rod-like structure with majority 110 facets. Bandgap decreases in both phases and reduced to a visible light region. BET analysis shows that surface area increases from 4.55 m2/g (x=0) to 96.53 m2/g (x=0.046) by Ga-V incorporation which provide a large number of active site for photocatalytic activity. Hence, co-doped anatase nanoparticle can be used as a promising candidate for photocatalytic applications using visible light up to a higher temperature ~650 C. | physics.app-ph | physics | Structural phase transition, grain growth and optical properties of uncompensated Ga-V
co-doped TiO2
Nasima Khatun1, Saurabh Tiwari2, Jayanti Lal3, Chuan-Ming Tseng4, Shun Wei Liu5,
Sajal Biring6 and Somaditya Sen1,2*
1Department of Physics, Indian Institute of Technology Indore, Simrol Campus, Khandwa Road,
Indore 453552, India
2Metallurgy Engineering and Material Sciences, Indian Institute of Technology Indore, Simrol
Campus, Khandwa Road, Indore 453552, India
3Department of chemistry, Vels Institute of Science, Technology & Advanced Studies, Chennai,
Tamil Nadu – 600117, India
4Department of Materials Engineering, Ming Chi University of Technology, New Taipei City -
24301, Taiwan
5Organic Electronic Research Center, Ming Chi University of Technology, New Taipei City -
24301, Taiwan
6Electronic Engg., Ming Chi University of Technology, New Taipei City -24301, Taiwan
Abstract
Effect of uncompensated Ga-V co-doping (0≤x≤0.046) on structural phase transition, grain
growth process and optical properties of TiO2 is reported here. Inhibition of phase transition due
to co-doping is confirmed by X-ray diffraction measurement. Activation energy of phase
transition increases from 120KJ/mol (x=0) to 140 KJ/mol (x=0.046) due to Ga-V co-doping. In
anatase phase, lattice constants increase by the effect of Ga3+ interstitials. This results in
inhibition of phase transition. Anatase phase becomes stable up to ~650 C in co-doped sample
whereas for pure TiO2 phase transition starts in between 450-500 ⁰C. In anatase phase, strain
increases due to co-doping which reduces crystallite size. In rutile phase, grain growth process is
enhanced due to co-doping and particles show a rod-like structure with majority {110} facets.
Bandgap decreases in both phases and reduced to a visible light region. BET analysis shows that
surface area increases from 4.55 m2/g (x=0) to 96.53 m2/g (x=0.046) by Ga-V incorporation
which provide a large number of active site for photocatalytic activity. Hence, co-doped anatase
1
nanoparticle can be used as a promising candidate for photocatalytic applications using visible
light up to a higher temperature ~650 C.
Introduction:
TiO2 is of continual interest due to its multifunctional properties. Different crystal structure and
corresponding electronic band structure facilitate its applications in different fields such as in
opto-electronic devices1, 2, self-cleaning glass coating materials3, photocatalyst 4, 5, fuel cell6,
dye-sensitized solar cell7, 8, opacifier and white pigment9, 10, etc.. It facilitates environmentally
beneficial reactions through photocatalytic activity by splitting of water to generate hydrogen
and treatment of polluted air and water 11. Low cost, nontoxicity, and high chemical stability add
a special importance for application.
TiO2 has three naturally occurring polymorph12. In order of abundance, these are rutile (R),
anatase (A), and brookite (B)13. At lower temperature anatase is the most stable phase due to its
low surface free energy 14, 15. Pure brookite phase is not available at normal ambient condition
due to its complex crystal structure. Both anatase and brookite are metastable phases. With
increasing temperature ( 750 C) both the phases are irreversibly transformed into stable rutile
phase16. Density functional theory (DFT) calculation showed that effective mass of electrons and
holes is smaller in anatase phase compared to brookite and rutile phases17. This facilitates the
migration of carriers and enhances photocatalytic activity. Due to lighter effective mass, smaller
particle size (highly stable 10-15 nm)12, 18 and longer lifetime of photogenerated charge
carriers17, anatase is an active polymorph for photocatalytic applications than brookite and
rutile19-21.
In spite of these important properties of anatase TiO2, there are few restraints. Anatase TiO2 has
lower thermal stability ( 450-500 C) and wide bandgap (3.2 eV)22. Due to its wide bandgap, it
only absorbs ~5% radiation of the solar spectrum. The entire visible region (~45%) remains
unutilized. From a practical application point of view, utilization of visible light is beneficial.
Therefore, people have made significant efforts to shift its phase transition temperature to a
higher temperature region and tune the bandgap of TiO2 in visible light range. Different
processes have been adapted for this purpose. These include synthesis methods 23-25, strain26, 27,
doping with different elements (Fe, Mo, V, Ru, Cu, Fe, Cr etc.) 14, 28-32, etc.
2
Among all these processes, doping is the easiest way to control phase transition and thereby
properties (tune the bandgap). From the literature, it was observed that Ga doping inhibits the
phase transition and play a robust role in photocatalytic activity (PCA) in UV region33, 34.
Vanadium considerably reduces the bandgap but promotes phase transition 16, 35, 36. Hence, in this
context uncompensated Gallium and Vanadium co-doping have been chosen to overcome both
the problem. Here in this work, effect of uncompensated Ga-V co-doping on structural phase
transition, grain growth process and optical properties of TiO2 has been discussed.
Experimental
Ga and V co-doped TiO2 (Ti(1-x)( Ga0.8V0.2)xO2: TGV) nanoparticles (with x=0.00 (TGV0), 0.015
(TGV1), 0.031 (TGV3) and 0.046 (TGV4)) are prepared by modified sol-gel synthesis. The Ti-
solution is prepared by mixing required amount of dihydoxy-bis titanium (TALH: C6H18N2O8Ti)
in deionized (DI) water at room temperature. An appropriate amount of Ga(NO3)3 is dissolved in
DI water in one beaker. In another beaker, V2O5 is also dissolved in DI water by adding little
amount of NH4OH while stirring. Both the solutions are added dropwise into a Ti-solution.
After 1h of mixing, citric acid and ethylene glycol is added to it. This mixture is stirred for
another 1h for homogeneous mixing. Thereafter it is slowly heated. Temperature of this solution
is maintained at 80 ⁰C for 4-5h to get the thick gel. The gel is burnt on a hot plate at 100 ⁰C in
normal ambient condition resulting in a black dry powder. This powder is denitrified and
decarburized at 450 ⁰C in an air atmosphere for 6h to get desired nanoparticles. The collected
powder is subsequently heated at eight different temperatures in 50 ⁰C steps from 450 ⁰C to 800
⁰C, stabilizing for 6h at each temperature.
Thermal gravimetric analysis (TGA) was performed to estimate the crystallization temperature
of the samples by METTLER TOLEDO (TGA/DSC 1) system using the STARe software system
up to 800 ⁰C in an air atmosphere with a heating rate of 5 ⁰C min-1. Structural analysis was
studied by powder X-ray diffraction (XRD) patterns using Bruker D2 phaser diffractometer with
Cu-Kα radiation (λ=1.5418 Å). Morphology and particle size were investigated using high-
resolution transmission electron microscope (HRTEM) (JEOL JEM-2100 LaB6, accelerating
voltage - 200 kV) and field emission scanning electron microscopy (Supra55 Zeiss- FESEM).
Surface area of the samples was calculated by Brunauer–Emmett–Teller (BET) measurement.
3
Diffuse reflectance spectroscopy (DRS) measurements were carried out using Bentham TMc300
Monochromator to estimate the changes in the bandgap.
Results and Discussion
TGA measurement on dry gel powder (TGV0) is performed from room temperature (RT-27 C)
to 800 C. Weight loss of 1.04%, from RT to ~110 °C, is observed in pure TiO2 (TGV0) (Fig. 1).
This is due to elimination of physically adsorbed water37. In the temperature regime, ~110 to 315
°C a sharper weight loss of 2.52% is observed. This may be attributed to the rupture of a
polymeric chain of black powder and removal of ethylene glycol units38. A final very sharp
weight loss of 5.61% is observed in the temperature regime ~315 to 430 C. This drastic loss is
due to decomposition of the organic compounds into carbon dioxide and nitrogen dioxide,
desorption of chemisorbed water molecules39, 40. This temperature is well enough to consume
unreacted precursor and thereby form well crystalline samples. Note that beyond ~430-450 C,
there is almost no change in weight loss. Thus, 450 ⁰C is selected as the optimum calcination
temperature which is high enough to achieve crystallization, and optimum to reduce the thermal
growth of the particles to maintain nano-scale features in the calcined powder. A minor weight
loss of 0.48%, in between 550 to 650 °C can be ascribed to the removal of the surface hydroxyls
present in the samples41.
Fig. 1 TGA curve of the TGV0 sample at a temperature range of 27-800 C.
HRTEM is a very powerful tool to investigate a particle in a very small range (around 1-2 nm). It
gives crystallographic information and clear morphology of the nanoparticles. Fig. 2 (a and c)
shows the TEM images of TGV0 and TGV3 samples. Almost spherical shape particles are
observed for both the samples. Particles size of both the sample has been calculated using Image
4
J software. Histogram of TGV0 and TGV3 sample (inset of Fig. 2(a) and (c)) shows average
particles size are in the range of ~12-15 nm and ~8-10 nm respectively. It is observed that
particle size reduces due to co-doping. In V doped TiO2, it was observed that crystallite size
reduced with doping 35, 42. Ga doping also reduces crystallite size21, 33. Hence, a combination of V
and Ga co-doping is supposed to reduce crystallite size. TEM results confirm the same. In most
cases, strain increases upon doping of foreign elements into TiO2 and this strain hinders the grain
growth process of nanoparticles. From HRTEM images, it is observed that d spacing of lattice
fringes of both the samples are ~0.35 nm which corresponds to 101 planes of anatase TiO2 (Fig.
2 (b: TGV0 and d:TGV3). Clarity of the fringes signifies both the samples are well crystalline.
Such a good crystallinity at a low temperature ~450 C, is possible due to proper choice of
specific reagents (ethylene glycol and citric acid)43, 44 used in this synthesis methods. The ring-
like SAED patterns (inset of Fig. 2 (b: TGV0 and d: TGV3) reveals the polycrystalline nature
and confirms anatase phase of TiO2 of both the samples.
5
Fig. 2: TEM images of TGV0 (a) and TGV3 (c) and inset shows the histogram of particle size
distribution of corresponding samples. (b and d) HRTEM images of TGV0 and TGV3 and insets
show SAED pattern of corresponding samples.
Table 1. BET surface area, pore diameter, pore volume of Ga-V co-doped TiO2 samples.
Sample
name
BET surface area
(m2/g)
TGV0
4.55
TGV1
53.95
TGV3
85.25
TGV4
96.53
Pore diameter
(nm)
3.819
3.823
3.826
3.829
Pore volume
(cm3/g)
0.009
0.044
0.060
0.061
6
Nitrogen adsorption/desorption isotherms (Fig. 3) of all the samples display type IV isotherms
according to IUPAC classification. Hysteresis loop of the isotherms are of typical H2(a) type45.
BET surface area increases from 4.55 m2/g (TGV0) to 96.53 m2/g (TGV4). Surface area depends
on size and morphology of nanoparticles. Smaller the size, larger is the surface area of
nanoparticles. Hence, as BET surface area increases with doping a reduction in crystallite size is
expected. Pore size distribution is calculated from BJH method on the desorption isotherms (Fig.
3(e-h)). Pores sizes for all samples are <4 nm. Mesoporous materials have pore diameters
ranging from 2-nm to 50nm 45. Hence, these samples can be classified as mesoporous materials
based on the pore diameter and nature of hysteresis loop. A larger surface area due to Ga-V
incorporation provides a large number of active sites which makes the materials better for PCA.
Fig. 3 (a-d) Nitrogen adsorption/desorption isotherms of TGV samples (450 ⁰C) and (e-h)
Barrett-Joyner Halenda (BJH) pore size distribution curve of the samples.
XRD pattern for all samples (TGV0, TGV1, TGV3, and TGV4), heated at temperatures ~450 C,
500 C, 550 C, 600 C, 650 C, 700 C, 750 C, and 800 C, are shown in Fig. 4. XRD pattern
for all TGV samples heated at 450 ⁰C for 6h (Fig. 4(a)) matches well with COD ID-9015929
which is of tetragonal anatase phase of TiO2 having space group I41/amd. Hence, all the samples
are in pure anatase phase. There are no traces of any rutile phase at this temperature. Also, there
is no evidence of any simple or complex metal oxide phases related to Ti, Ga, and V. With
increasing temperature the anatase phase of TGV samples gradually starts to convert to a rutile
phase and forms a mixed phase. Further heating at a higher temperature (~800 ⁰C), all the TGV
7
samples are converted into an entire rutile phase. XRD patterns of TGV samples (800 ⁰C),
matches well with COD ID-9009083 which is of tetragonal rutile phase of TiO2 having space
group P42/mnm. The samples heat treated between 450-800 ⁰C shows mixed phase of anatase
and rutile.
Fig. 4 XRD patterns of all TGV samples at eight different temperatures (450-800 C) in the
range of 2θ=20-80.
For pure TiO2 (TGV0), AR phase transformation starts in between ~450-500 C. Phase
transition at this particular temperature happens due to the choice of specific reagent (ethylene
glycol and citric acid)43 used in this method. Complete transformation into rutile phase is
observed at ~750 C. In case of co-doped (Ga-V) samples, no trace of rutile phase has been
detected below ~550 C. For TGV1 and TGV3, AR phase transition starts ~550-600 C. For
TGV4, AR phase transition starts at ~650-700 C. A complete conversion into rutile phase
happens at ~800 C. It is observed from these XRD spectra that the appearance of rutile phase
and complete conversion into rutile phase both are shifted to higher temperature with increasing
8
doping concentration. Hence, Ga and V co-doping into TiO2 inhibits the phase transition or
stabilize the anatase phase to a higher temperature (for TGV1 and TGV3 up to ~550 C while for
TGV4 up to ~650 C).
Vigilant investigation on XRD patterns of the samples at rutile phase (800 ⁰C) shows small
appearance of -Ga2O3 phase for TGV3 and TGV4 samples (provided in supplementary file Fig.
S1) which matches with COD ID-2004987 (-Ga2O3). However, in anatase phase, such type of
impurity has not been detected. Anatase phase has some inherent empty space inside crystal
structure46. Therefore Ga and V easily incorporated into TiO2 lattice and occupy the position of
interstitials and substitutional sites. Density () of rutile phase (4.25 gm/cm3) is higher than
anatase (3.89 gm/cm3)14. Hence, rutile phase has less empty space compared to anatase phase. As
Ga3+ (0.76Å) ion has slightly bigger ionic radius compared to both Ti4+ (0.745Å) and V5+/4+
(0.68Å/0.72Å), therefore at higher temperature due to thermal instability and less space, Ga3+
ions move out from TiO2 lattice and segregate on the surface of the particles. These Ga ions at
higher temperature react with environment oxygen and form -Ga2O3 which are highly dispersed
on the surface of particles. -Ga2O3 is a stable crystalline form of gallium oxide at a higher
temperature (650 C)33.
Rutile phase fraction (fR) in the mixed phases is estimated at different temperatures using Spurr
and Mayers equation 47. Temperature dependence of fR (Fig. 5) for co-doped samples ensures an
inhibition of phase transformation with increasing doping concentration. All processing
parameters (like heating/cooling rates, environment of calcination, etc.) are kept constant. Hence,
this inhibition of phase transformation entirely depends on the concentration of Ga and V co-
doping. In general, oxygen vacancy results in lattice contraction and promotes AR phase
transition. On the other hand, interstitials expand the lattice and thereby inhibits the phase
transition48. Ga ion has slightly bigger ionic (VI-0.76Å) radius and lesser charge +3 compared to
Ti4+ (VI-0.745Å), while V has variable charge states (3+,4+,5+) with ionic radius (V3+ (VI-
0.78Å), V4+ (VI-0.72Å), and V5+ (VI-0.68Å)). From literature, it was observed that charge states
and ionic radius are very sensitive to accelerate and delay the AR phase transition14. Ga and V
ions have different charge states. Total charge compensation can only happen if amount of Ga
and V are equal and the entire V-population is in V5+ state. Ga: V ratio in all the samples is 4:1.
Hence, for charge compensation, it either creates oxygen vacancies or form interstitials. As
9
discussed above, interstitials are responsible for inhibition of phase transitions and XRD results
show inhibition of phase transition due to co-doping. This hints that effect of interstitials is more
prominent compared to oxygen vacancies.
Fig. 5 XRD pattern of all TGV samples at eight different temperature (~450-800 C) in the range
of 2θ=24.5-28. (b) Fraction of rutile phase (fR) at different temperature.
Activation energy (Ea) is the minimum energy required to overcome the energy barrier for AR
phase transition between the two phases. It was also observed from literature49 that Ea decreases
due to oxygen vacancies whereas interstitials are responsible for the increase of Ea. Ea is
calculated using Arrhenius equation: 𝑙𝑛(𝑓𝑅) = -
𝐸𝑎
𝑅𝑇
; where, 𝑓𝑅 is the fraction of rutile phase
present in a sample, R is universal gas constant and T is the temperature in Kelvin. Linear fits of
ln(fR) vs 1/T gives Ea (Fig. 6(a, b, c, and d)). It is observed that there is a drastic increase in Ea
from pure TiO2 (120 KJ/mol) to TGV3 (243 KJ/mol). For TGV4, Ea decreases slightly (240
KJ/mol) from TGV3 but remains higher compared to TGV0 and TGV1. This increasing trend of
Ea (Fig. 6(e)) support that the effect of interstitials is more prominent than oxygen vacancies
which expands the lattice in Ga-V co-doped samples. This expansion of lattice results in
inhibition of phase transition and is consistent with XRD results.
10
Fig. 6 Fits of ln(fR) vs 1/T ((a): TGV0; (b): TGV1; (c): TGV3 and (d): TGV4). (b) Variation of
activation energy with doping concentration (solid line is just a guide to the eye).
Samples in anatase phase, when heated to a higher temperature (500 C) leads to rearrangement
of Ti-O bonds as a result unit cell volume contracts and phase transformation (AR) occurs. In
anatase phase, lattice constant 'a=b' (3.785 Å) is smaller and 'c' (9.514 Å) is larger compared to
lattice constants of rutile phase (a=b=4.59 4Å and c=2.958 Å)14. Hence, unit cell volume of
anatase phase is larger (136.3 Å3) compared to rutile phase (62.4 Å3). For phase transition, 'a'
always increase and 'c' decrease. Hence, this delay of phase transition can be explained in terms
of change in lattice constants.
Fig. 7(a) shows the Rietveld refinement of TGV samples in pure anatase phase (450 C). It is
observed that all the three lattice constant increases with increasing doping concentration (Fig.
7(b)). Unit cell volume also follows the similar trend as observed in lattice constants (Fig. 7(c)).
Cr3+ has a comparable ionic radius (0.755 Å) as Ga3+ and has same charge state. Zhu et al.50 from
their DFT calculation showed that anatase phase formation energy is low when Cr3+ occupies
interstitials sites than substitutional sites. With increasing doping concentration, Cr3+ going from
interstitial to substitutional sites was observed to vary. Hence, Ga3+ ions too may have the same
tendency to go more into interstitial sites than substitutional sites. Banerjee et al.51
experimentally showed that Ga3+ ions occupy more interstitial sites than substitutional sites in
TiO2. These interstitial sites are responsible for the expansion of lattice and inhibit the phase
transition. Depero et al.34 experimentally proved that Ga doping inhibits phase transition. It was
also observed that formation energy of anatase TiO2 is low when V occupies the substitutional
11
sites rather than interstitial sites52. Hence, theoretically and experimentally it was proved that V
occupies substitutional sites in TiO2 35, 53, 54 and thereby decrease all the three lattice constants.
This is because V4+/5+ have smaller ionic radius compared to Ti4+. Vittadini et al.55 reported that
V5+ is more likely the major surface species where V4+ is stable inside balk. From TEM results it
was observed that particle is in a spherical shape and in nano size. With increasing doping
concentration, particle size decreases which result to increase the surface area to volume ratio of
the samples. BET measurement shows surface area increases with increasing doping
concentration. Hence at anatase phase, all the Vanadium ions are mostly in 5+ oxidation states42,
54. Hence, contraction of lattice constants and thereby unit cell volume by V incorporation
promoted the AR phase transition16. In all the co-doped samples as Ga content is more
compared to V (Ga:V=4:1), hence the effect of Ga interstitial play a significant role over V
substitution and oxygen vacancies which expands the lattice. Rietveld refinement on anatase
phase shows this expansion of lattice and results in inhibition of phase transition.
Fig. 7(a) Rietveld refinement of anatase TGV samples (450 C). Change of lattice constants (b),
and unit cell volume (c) with Ga-V doping concentration.
12
From Rietveld refinement on rutile phase (800 ⁰C), it is observed that lattice constant 'a' and 'b'
increases with increasing doping concentration. However, lattice constant 'c' nominally
increases for TGV1 and thereafter decreases rapidly for TGV3 and TGV4. Unit cell volume also
increases for TGV1 and thereafter decreases (for TGV3 and TGV4). As mentioned above that
for TGV3 and TGV4, due to thermal instability and less space, few Ga ions move out from
lattice structure. Hence, relative percentage of V4+/5+ ions compared to Ga3+ ions increases from
targeted values (4:1). At rutile phase, due to high temperature (800 ⁰C) particle size increases for
all TGV samples and with increasing doping content grain growth process enhanced (discussed
later at Fig. 9). Hence surface area to volume ratio decreases which results to increase in V4+
species in the samples as discussed above that V4+ is more likely stable into bulk. In our earlier
report16, it was observed that V ions are in mixed valence states of V5+/V4+ and with increasing
doping concentration presence of V4+ ions increases. In rutile phase, both the V5+ and V4+ ions
occupy the substitutional sites in TiO2 lattice. V5+/4+ ions have smaller ionic radius compared to
Ti4+ and Ga3+ which results in a decrease of lattice constants 'c' as well as unit cell volume. Ga3+
ions occupy more interstitials sites than substitutional sites in TiO2 discussed above. At lower
doping (TGV1), as all the Ga ions are inside the crystal structure and due to significant role of
this Ga3+ interstitials unit cell volume increased. However at higher doping (TGV3 and TGV4),
due to substitutional V4+/5+ ions and oxygen vacancies unit cell volume decreased.
Fig. 8 Change in lattice constants (a) and unit cell volume (b) with co-doping concentration at
rutile phase (800 C).
13
To investigate the effect of Ga and V co-doping on grain growth process, crystallite size (at 450
C) is calculated using Scherrer equation. It is observed that crystallite size decreases from 14.5
(pure TiO2) nm to 8.9 nm (TGV4) by Ga-V incorporation (Fig. 9(e)). Hence co-doping restrains
the grain growth process of anatase nanoparticles which is consistent with TEM results. In most
metal oxides this restrains grain growth is due to increasing strain in the nanoparticles.
Such increase of strain in lattice due to Ga-V co-doping has been verified by the shape and peak
positions of pure anatase TGV samples (at 450 C). Usually, crystallites in polycrystalline
aggregates are in a state of compression or tension by its neighboring crystallites which produce
uniform or non-uniform strain in the lattice. From literature, it was observed that shifting of
diffraction peak creates uniform strain whereas peak broadening without changing peak position
creates non-uniform strain 51. Careful inspection reveals that the 101 peak become broad but
position remain almost same in TGV samples (provided in supplementary file Fig. S2). Hence,
incorporation of Ga3+/V4+/5+ at lattice site and as well as in interstitial site may be responsible for
such nonuniform stain. Williamson Hall plot is used to calculate quantitative changes in strain
due to Ga-V co-doping. Slop of linear fits of cos() vs 4sin() gives strain; where is the
FWHM of corresponding peaks of XRD spectra. It is observed that strain increases with doping
concentration (Fig. 9(e)). This increasing strain due to Ga-V incorporation retards the grain
growth process of anatase nanoparticle.
Fig. 9 Linear fits of cos(θ) vs 4sin(θ) of TGV samples in anatase phase (450 C). (e) change of
strain and crystallite size with doping concentration.
14
Fig. 10 shows the FESEM images of TGV samples at rutile phase (800 ⁰C). Oliver et al. 56, from
their DFT calculations on rutile phase, reported that {110} surface has the lowest surface energy
(1.78 J/m2) whereas {100} surface perpendicular to {110} surface has the highest surface energy
(2.08 J/m2). During crystal growth, the low energy surface {110} grows fastest and high energy
surface {100} tend to decrease its surface area to minimize the total energy per crystal57. As
mentioned above, particles of anatase TGV samples (450 ⁰C) are almost in a spherical shape.
With increasing temperature anatase phase converted into mixed phase and with further heating
transform into an entirely rutile phase. Similarly, particle shape and size also changed with
temperature. With increasing temperature spherical anatase crystals enlarge its size and become
elongate spherical to a rod-like structure. For pure TiO2 (TGV0), particles are in irregular
spherical shape or distorted rod-like structure and average particle size is ~200 nm (calculated
using Image J software). From these images (all the images are in same magnification), it clearly
observed that particles have prominent rod-like structure and grain growth process enhanced
with increasing doping concentration. It was observed that Ga doping restrain the rutile grain
growth process34. Whereas in earlier reports it was observed that V enhanced the grain growth
process16. Effect of Vanadium is more sensitive in grain growth process of rutile particle than
Gallium. Hence, the faster grain growth process in co-doped samples is mainly due to the effect
of vanadium. According to the nature of surface edges and area at par with reported literature,
{110} surfaces are the most prominent surfaces of rutile particle56, 58, 59. Some portions of
FESEM images of TGV samples have been zoomed to show these surfaces. For TGV0 samples,
the zoomed view is shown in the inset of Fig. 10(a). The zoomed views of the co-doped samples
are shown as ② (TGV1), ③ (TGV3), and ④ (TGV4).
15
Fig. 10 FESEM images of TGV samples ((a): TGV0; (b): TGV1; (c): TGV3 and (d): TGV4)
heated at 800 ⁰C.
TEM images of TGV1 sample heated at 800 ⁰C are shown in Fig. 11((a) and (c)). SAED patterns
of the particle are shown as insets of corresponding images and it belongs to the rutile phase of
TiO2. The d-spacing of HRTEM images are also belonging to the same crystallographic planes
(particle #1, d110~0.33 nm and d101~0.25nm (Fig. 11(b)) while in particle #2, d110~0.33 nm and
d001~0.29 nm (Fig. 11 (d))). From these images, it observed that {110} facets are the major
surface of the particle. The size of the particles seems to be in the range of ~ 100 nm and beyond.
16
This is smaller than the average size obtained from FESEM studies. A possible reason may be
the sample preparation process for TEM measurements. Disperse solution of samples are
prepared in ethyl alcohol and a droplet is dropped on TEM grids and dried. In most cases, only
the lighter and smaller particles get selected in this process.
Fig. 11 TEM image of TGV1 sample ((a) and (c)) heated at 800 ⁰C and inset show the
corresponding SAED patterns. HRTEM images ((b) and (d)) show lattice fringes of the particle.
Ga-V co-doping also affects the optical properties of TiO2. Room temperature DRS
measurement has been carried out to investigate the bandgap of the samples (Fig. 12(a) and (b)).
Bandgap is calculated using Tauc plot (F(R))hʋ=A(h-Eg)n, where, R is the reflectance, A is a
17
constant, h is the frequency of illumination, Eg is the bandgap and n is a unitless parameter with
value ½ or 2 for direct or indirect bandgap semiconductor respectively. From literature, it was
observed that anatase is an indirect (n=2) bandgap whereas rutile is direct (n=1/2) bandgap
semiconductor17. It is observed that bandgap decreases due to co-doping for both the phases. At
anatase phase, bandgap decreases from 3.14 eV (TGV0) to 2.86 eV (TGV4) (Fig. 12(c)) and at
rutile phase, bandgap decreases from 3.06 (TGV0) to 2.84 eV (TGV4) (Fig. 12(d)). O 2p and
Ti3d hybridization (p-d) form strong bonding states which are responsible to form valence band
(VB) in TiO2. On the other hand, antibonding states due to p-d hybridization between O 2p, Ti
3d, and Ti 2p form conduction band (CB)17. Due to Ga doping, the hybridization between O 2p,
Ti 3d, and Ga 3d becomes weaker and form empty states in the bandgap of TiO2 which therefore
widen the conduction band33, 60. This results in increasing the bandgap. From literature, it was
also observed that effect of Ga for enhancement of bandgap is not much pronounced21, 33, 51, 61. In
case of V doped TiO2, p-d hybridization of O 2p, Ti 3d, and V 3d form impurity energy levels
(or donor levels) inside the bandgap. Due to these energy levels bandgap decreases42. It was also
observed that effect of V is more sensitive compared to Ga for bandgap change. Hence, due to
combined effect of both V and Ga, bandgap decreases in co-doped samples.
In case of rutile phase, bandgap gradually shifted to lower values from 3.06 eV (TGV0) to 2.84
eV (TGV4) with increasing doping concentration. For TGV1, bandgap shows slight red shift
because V concentration is very low in the samples (Ga: V ~ 4:1). For TGV3 and TGV4, it is
observed from XRD data that some amount of Ga3+ ions move out of the lattice and forms -
Ga2O3 phase. Hence, relative amount of V4+/5+ to Ga3+ is increased (Ga:V=4:1; ratio decreased)
compared to intended values. It was discussed that V4+ ion is more effective than V5+ ion in
reducing the bandgap62 of TiO2: (Eg(V4+)Eg(V5+)). This may also be a reason, for a gradual shift
of bandgap in rutile phase.
Structural modifications are inevitable when foreign elements are incorporated into any lattice. It
is known that electronic band structure is strongly correlated with a lattice structure. Urbach
energy (EU) is a measure of lattice distortion in the samples which affects electronic band
structure 63, 64. In most semiconductors, it was observed that bandgap decreases if EU increases65,
66. EU is calculated from linear fits of "lnF(R)-hʋ" plots just below the absorption edge of DRS
54. In anatase phase, EU decrease from 145 meV for TGV0
data. Reciprocal of the slope gives EU
18
to 506 meV for TGV4 (Fig. 12(c)). Whereas for rutile phase; EU decrease from 63 meV for
TGV0 to 256 meV for TGV4 (Fig. 12(d)). This increase in EU signifies a more distorted lattice
due to co-doping and a band tailing (Urbach tail) just below absorption edge.
Fig. 12 Room temperature DRS data of co-doped (Ga-V) samples in anatase (a) and rutile phase
(b). Inset shows the change of bandgap with co-doping concentration for corresponding samples.
As mentioned above for pure TiO2, phase transition (AR) starts in between 450-500 ⁰C. Upon
co-doping, anatase phase becomes stable up to ~650 ⁰C. Surface area plays an important role for
PCA. In the anatase phase, nano-sized spherical particles (8-15 nm) provide a large surface area
and BET analysis confirms this increasing trend of surface area with doping concentration. This
provides a larger number of active sites for PCA. V decreases the bandgap to visible regions
while Ga inhibits phase transition, thereby making the materials a promising candidate for "high
temperature visible light photocatalytic application."
19
Conclusions
Uncompensated Ga-V co-doped TiO2 has been prepared successfully by modified sol-gel
process. Inhibition of phase transition due to co-doping is confirmed by XRD spectra. Activation
energy of phase transition increases from 120 KJ/mol (x=0) to 240 KJ/mol (x=0.046) by Ga-V
incorporation which also reveals this inhibition of phase transition. Ga ions occupy more
interstitial sites than substitutional sites whereas V ions occupy substitutional sites in TiO2. In
anatase phase, lattice constant increases by the effect of Ga3+ interstitials, as Ga content is more
than V content (Ga:V:=4:1). This expansion of lattice results in inhibition of phase transition.
Hence, anatase phase becomes stable up to ~650 C in co-doped sample. In anatase phase, grain
growth process restrained as strain increases by the effect of co-doping and thereby reduces
crystallite size. In rutile phase, grain growth process in co-doped sample is enhanced mainly by
the effect of Vanadium. Bandgap decreases in both phases and reduces to the visible light region.
BET analysis shows that surface area increases from 4.55 m2/g (x=0) to 96.53 m2/g (x=0.046) by
Ga-V incorporation. Hence, co-doped anatase nanoparticles can be used as a good photocatalyst
using visible light up to a higher temperature ~650 C.
Acknowledgment
The authors are sincerely thanking IIT Indore for providing funds and all research related
facilities. The authors also thank Sophisticated Instrument Centre (SIC) of IIT Indore for TGA,
BET, and FESEM studies. One of the authors (Dr. Sajal Biring) acknowledges support from
Ministry of Science and Technology, Taiwan (MOST 105-2218-E-131-003 and 106-2221-E-
131-027).
References
M. Zukalova, A. Zukal, L. Kavan, M. K. Nazeeruddin, P. Liska and M. Gratzel, Nano. Lett. 5 (9),
B. E. Hardin, E. T. Hoke, P. B. Armstrong, J.-H. Yum, P. Comte, T. Torres, J. M. J. Fréchet, M.
1.
1789-1792 (2005).
2.
K. Nazeeruddin, M. Grätzel and M. D. McGehee, Nat. Photonics 3 (7), 406-411 (2009).
3.
4.
5.
6.
7.
Sol. Cells 144, 194-209 (2016).
8.
N. Abidi, L. Cabrales and E. Hequet, ACS Appl. Mater. Interfaces 1 (10), 2141-2146 (2009).
C. Fan, C. Chen, J. Wang, X. Fu, Z. Ren, G. Qian and Z. Wang, Sci. Rep. 5, 11712 (2015).
R. Ren, Z. Wen, S. Cui, Y. Hou, X. Guo and J. Chen, Sci. Rep. 5, 10714 (2015).
B. Park and E. J. Cairns, Electrochem. Comm. 13 (1), 75-77 (2011).
T. K. Das, P. Ilaiyaraja, P. S. V. Mocherla, G. M. Bhalerao and C. Sudakar, Sol. Energy Mater.
J.-Y. Liao, J.-W. He, H. Xu, D.-B. Kuang and C.-Y. Su, J. Mater. Chem. 22 (16), 7910 (2012).
20
M. J. A. Ruszala, N. A. Rowson, L. M. Grover and R. A. Choudhery, Int. J. Chem. Eng. Appl. 6
Y. Wang, J. Li, L. Wang, T. Qi, D. Chen and W. Wang, Chem. Eng.Technol. 34 (6), 905-913
T. Kamegawa, J. Sonoda, K. Sugimura, K. Mori and H. Yamashita, J. Alloys. Compound. 486 (1-
N. Khatun, Anita, P. Rajput, D. Bhattacharya, S. N. Jha, S. Biring and S. Sen, Ceram. Int. 43
J. Zhang, P. Zhou, J. Liu and J. Yu, Phys. Chem. Chem. Phys. 16 (38), 20382-20386 (2014).
N. Satoh, T. Nakashima and K. Yamamoto, Sci. Rep. 3, 1959 (2013).
Q. L. XiaoBo Li, XiaoYing Jiang, Jianhua Huang, Int. J. Electrochem. Sci. 7, 11519-11527
A. Sclafani and J. M. Herrmann, J. Phys. Chem. 100 (32), 13655-13661 (1996).
A. K. Chandiran, F. d. r. Sauvage, L. Etgar and M. Graetzel, J. Phys. Chem. C 115 (18), 9232-
B. Buchholcz, H. Haspel, Á. Kukovecz and Z. Kónya, CrystEngComm 16 (32), 7486 (2014).
C. Huang, X. Liu, L. Kong, W. Lan, Q. Su and Y. Wang, Appl. Phys. A 87 (4), 781-786 (2007).
A. Zaleska, Recent Patents on Engineering 2 (3), 157-164 (2008).
C. Rath, P. Mohanty, A. C. Pandey and N. C. Mishra, J. Phys. D: Appl. Phys. 42 (20), 205101
V. N. Koparde and P. T. Cummings, ACS Nano 2 (8), 1620-1624 (2008).
J. Muscat, V. Swamy and N. M. Harrison, Phys. Rev. B 65 (22) (2002).
D. A. H. Hanaor and C. C. Sorrell, J. Mater. Sci. 46 (4), 855-874 (2010).
K. Ding, Z. Miao, B. Hu, G. An, Z. Sun, B. Han and Z. Liu, Langmuir 26 (12), 10294-10302
9.
(5), 331-340 (2015).
10.
(2011).
11.
2), 685-688 (2009).
12.
13.
14.
15.
(2010).
16.
(16), 14128-14134 (2017).
17.
18.
19.
(2012).
20.
21.
9240 (2011).
22.
23.
24.
25.
(2009).
26.
27.
Mater. Electron. (2017).
28. W. Choi, A. Termin and M. R. Hoffmann, J. Phys. Chem. 98 (51), 13669-13679 (1994).
B. Choudhury, A. Choudhury and D. Borah, J. Alloys. Compound. 646, 692-698 (2015).
29.
Y. Zhang, Y. Shen, F. Gu, M. Wu, Y. Xie and J. Zhang, Appl. Surf. Sci. 256 (1), 85-89 (2009).
30.
N. Khatun, R. Amin, Anita and S. Sen, AIP Conf. Proc. 1953, 040028 (2018).
31.
32.
G. Hassnain Jaffari, A. Tahir, N. Z. Ali, A. Ali and U. S. Qurashi, J. Appl. Phys. 123 (16),
161541 (2018).
33.
W. Xu, D. E. Polyansky, J. A. Rodriguez and S. D. Senanayake, Phys. Chem. Chem. Phys. 20 (3), 2104-
2112 (2018).
34.
Mater. Res. 15 (10), 2080-2086 (2011).
35. W. Avansi, R. Arenal, V. R. de Mendonça, C. Ribeiro and E. Longo, CrystEngComm 16 (23),
5021 (2014).
36.
37.
38.
39.
(9), 2629-2636 (2010).
40. M. V. Swapna and K. R. Haridas, J. Exp. Nanosci. 11 (7), 540-549 (2015).
41.
Sci. Technol. 63 (1), 16-22 (2012).
42. W. Zhou, Q. Liu, Z. Zhu and J. Zhang, J. Phys. D: Appl. Phys. 43 (3), 035301 (2010).
S. Klosek and D. Raftery, J. Phys. Chem. B 105 (14), 2815-2819 (2001).
S. Shang, X. Jiao and D. Chen, ACS Appl. Mater. Interfaces 4 (2), 860-865 (2012).
X. Jiang, T. Herricks and Y. Xia, Adv. Mater. 15 (14), 1205-1209 (2003).
H. Yaghoubi, N. Taghavinia, E. K. Alamdari and A. A. Volinsky, ACS Appl. Mater. Interfaces 2
G. Rajender and P. K. Giri, J. Alloys. Compound. 676, 591-600 (2016).
Anita, A. K. Yadav, N. Khatun, S. Kumar, C.-M. Tseng, S. Biring and S. Sen, J. Mater. Sci.:
S. Luo, T. D. Nguyen-Phan, D. Vovchok, I. Waluyo, R. M. Palomino, A. D. Gamalski, L. Barrio,
L. E. Depero, A. Marino, B. Allieri, E. Bontempi, L. Sangaletti, C. Casale and M. Notaro, J.
V. Caratto, L. Setti, S. Campodonico, M. M. Carnasciali, R. Botter and M. Ferretti, J. Sol-Gel
21
K. Farhadian Azizi and M. M. Bagheri-Mohagheghi, J. Sol-Gel Sci. Technol. 65 (3), 329-335
R. A. Spurr and H. Myers, Anal.l Chem. 29 (5), 760-762 (1957).
R. D. Shannon and J. A. Pask, J. Am. Ceram. Soc. 48 (8), 391-398 (1965).
43.
(2012).
44.
A. B.-N. A. Shalaby, R. Iordanova, Y. Dimitriev, J. Chem. Technol. Metall. 48, 585-590 (2013).
45. M. Thommes, K. Kaneko, A. V. Neimark, J. P. Olivier, F. Rodriguez-Reinoso, J. Rouquerol and
K. S. W. Sing, Pure Appl. Chem. 87 (9-10) (2015).
46. W.-J. Yin, S. Chen, J.-H. Yang, X.-G. Gong, Y. Yan and S.-H. Wei, Appl. Phys. Lett. 96 (22),
221901 (2010).
47.
48.
49. M. A. Malati and W. K. Wong, Surf. Technol. 22 (4), 305-322 (1984).
50. W. Zhu, X. Qiu, V. Iancu, X. Q. Chen, H. Pan, W. Wang, N. M. Dimitrijevic, T. Rajh, H. M.
Meyer, 3rd, M. P. Paranthaman, G. M. Stocks, H. H. Weitering, B. Gu, G. Eres and Z. Zhang, Phys. Rev.
Lett. 103 (22), 226401 (2009).
51.
52.
53. M. Amini, H. Naslhajian and S. M. F. Farnia, New J. Chem. 38 (4), 1581 (2014).
54.
7-13 (2016).
55.
56.
(1997).
57.
58.
59.
60.
Chem. Mater. 27 (11), 3936-3945 (2015).
61.
62.
63.
64.
(14), 141913 (2007).
65.
66.
A. S. Barnard and P. Zapol, Phys. Rev. B 70 (23) (2004).
Z. Lai, F. Peng, H. Wang, H. Yu, S. Zhang and H. Zhao, J. Mater. Chem. A 1 (13), 4182 (2013).
T. Ohno, K. Sarukawa and M. Matsumura, New J. Chem. 26 (9), 1167-1170 (2002).
C. Gionco, S. Livraghi, S. Maurelli, E. Giamello, S. Tosoni, C. Di Valentin and G. Pacchioni,
S. Sudou, K. Kado and K. Nakamura, Trans. Mater. Res. Soc. Japan 35 (1), 171-174 (2010).
A. Y. Choi and C.-H. Han, J. Nanosci.Nanotechnol. 14 (10), 8070-8073 (2014).
S. J. Ikhmayies and R. N. Ahmad-Bitar, J. Mater. Res.Technol. 2 (3), 221-227 (2013).
A. Paleari, F. Meinardi, A. Lauria, R. Lorenzi, N. Chiodini and S. Brovelli, Appl. Phys. Lett. 91
A. S. Hassanien and A. A. Akl, J. Alloys. Compound. 648, 280-290 (2015).
S. Benramache, B. Benhaoua and F. Chabane, J. Semicon. 33 (9), 093001 (2012).
A. N. Banerjee, S. W. Joo and B.-K. Min, J. Nanomater. 2012, 1-14 (2012).
R. Long and N. J. English, Chem. Phys. Chem. 11 (12), 2606-2611 (2010).
N. Khatun, E. G. Rini, P. Shirage, P. Rajput, S. N. Jha and S. Sen, Mater. Sci. Semicon. Proc. 50,
A. Vittadini, M. Casarin and A. Selloni, J Phys Chem B 109 (5), 1652-1655 (2005).
P. M. Oliver, G. W. Watson, E. T. Kelsey and S. C. Parker, J. Mater. Chem. 7 (3), 563-568
22
|
1909.06011 | 1 | 1909 | 2019-09-13T03:16:30 | Thermal Transport in 3D Nanostructures | [
"physics.app-ph",
"cond-mat.mes-hall",
"cond-mat.mtrl-sci"
] | This work summarizes recent progress on the thermal transport properties of three-dimensional (3D) nanostructures, with an emphasis on experimental results. Depending on the applications, different 3D nanostructures can be prepared or designed to either achieve a low thermal conductivity for thermal insulation or thermoelectric devices, or a high thermal conductivity for thermal interface materials used in the continuing miniaturization of electronics. A broad range of 3D nanostructures have been discussed, ranging from colloidal crystals/assemblies, array structures, holey structures, hierarchical structures, 3D nanostructured fillers for metal matrix composites and polymer composites. Different factors that impact the thermal conductivity of these 3D structures are compared and analyzed. This work provides an overall understanding of the thermal transport properties of various 3D nanostructures, which will shed light on the thermal management at nanoscale. | physics.app-ph | physics | Thermal Transport in 3D Nanostructures
Haifei Zhan, Yihan Nie, Yongnan Chen*, John M. Bell, and Yuantong Gu*
Dr. H.F. Zhan, Mr. Y.H. Nie, Prof. J.M. Bell, Prof. Y.T. Gu
School of Chemistry, Physics and Mechanical Engineering, Queensland University of
Technology, Brisbane QLD 4001, Australia
E-mail: [email protected]
Prof. Y.N. Chen
School of Materials science and Engineering, Chang'an University, Xi'an 710064, China
E-mail: [email protected]
Keywords: thermal conductivity; nanostructures; carbon nanotube; nanocomposite; metal-
matrix composites.
This work summarizes recent progress on the thermal transport properties of three-
dimensional (3D) nanostructures, with an emphasis on experimental results. Depending on the
applications, different 3D nanostructures can be prepared or designed to either achieve a low
thermal conductivity for thermal insulation or thermoelectric devices, or a high thermal
conductivity for thermal interface materials used in the continuing miniaturization of
electronics. A broad range of 3D nanostructures have been discussed, ranging from colloidal
crystals/assemblies, array structures, holey structures, hierarchical structures, 3D
nanostructured fillers for metal matrix composites and polymer composites. Different factors
that impact the thermal conductivity of these 3D structures are compared and analyzed. This
work provides an overall understanding of the thermal transport properties of various 3D
nanostructures, which will shed light on the thermal management at nanoscale.
1. Introduction
Thermal transport is one of the fundamental characteristics that determine the applications of
materials. Depending on the application, materials are required to have either a high thermal
conductivity, or a strongly suppressed thermal conductivity.[1] For energy saving in both
residential and commercial buildings, there has been a continuing search for high-
1
performance, light weight, and mechanically strong, thermally insulating materials, where a
low thermal conductivity is required (Figure 1a).[2] Good thermal insulation is also required
for the electrical, optical, and space applications in order to tightly regulate heat transfer
during operation. Another major application that can be affected dramatically by the thermal
transport properties is the direct energy conversion from heat to electricity using
thermoelectric materials,[3] such as bismuth telluride based-electronics for low temperature
applications.[4] The performance of the thermoelectric devices is measured by the figure of
merit (ZT) as calculated from 𝑍𝑇=𝑆%𝜅'𝜅()*𝑇, where S is the Seebeck coefficient; 𝜅' and 𝜅(
are the electrical and thermal conductivity, respectively; and T is the average temperature
within the material. To obtain a high ZT with optimum efficiency, there are generally two
strategies, including phonon engineering and band engineering. The purpose of phonon
engineering is to reduce lattice thermal conductivity without degrading the electrical
properties.[3b] Thermoelectric devices possess a vast potential for power generation from solar,
automobile, industrial heat sources, and even human body heat (Figure 1b).[5]
On the other hand, the continuing miniaturization of electronic devices/systems (such
as light-emitting diodes, integrated circuits and microprocessors) has inevitably increased
their power density (~ 100 Wcm2) and led to dramatically increased heat generation.[6] To
ensure reliable performance and lifetime, effective and efficient heat removal is desired,
which relies heavily on the high thermal conductivity of the packaging substrates and thermal
interface materials (Figure 1c).[7] The thermal interface materials are normally based on
polymers, whose intrinsic thermal conductivity is only about 0.2-0.5 Wm-1K-1, being
restricted by the strong inherent phonon scattering between chain ends, entanglements and
impurities.
2
Figure 1. The palette of 3D nanostructures for thermal management. (a) Example applications
for thermal insulation, including the house (1 -- Ref.[8]), the active window (2 -- Ref.[9]), the
active wall (3 -- Ref.[9]), and the aerogel above a blow torch (4 -- Ref.[10]). (b) Example
application of thermoelectric devices, including a thermoelectric module sketch (5 -- Ref.[11]),
a wearable thermoelectric generator (6 -- Ref.[12]), a body-powered wireless pulse oximeter (7
-- Ref.[13]), a thermoelectric generator powered flat bulk (8 -- Ref.[14]). (c) Electronics,
including a printed circuit board (9 -- Ref.[15]), a schematic view of the electronic packing (10
-- Ref. [16]). (d) Different low-dimensional nanomaterials. (e) Different 3D nanoarchitectures,
including colloidal assemblies (11 -- Ref.[17]), silica hollow sphere colloidal crystals (12 -
Ref.[18]), Cu inverse opals (13 -- Ref.[19]), vertically aligned carbon nanotube array (14 --
Ref.[20]), aligned Cu nanowire arrays (15 -- Ref.[21]), the atomic model of a pillared graphene
(16 -- Ref.[22]), a carbon nanolattice (17 -- Ref.[23]), the atomic model of a carbon honeycomb
3
structure (18 -- Ref.[24]), and a holey silicon (19 -- Ref. [25]). (f) Different metal matrix
composites, including Cu matrix with carbon nanotube and reduced graphene oxide network
(20 -- Ref.[26]), Cu matrix with embedded carbon nanotubes (21 -Ref.[27]). (g) Polymer
composites with 3D nanostructures, including melamine foam with BN nanosheets assemblies
(22 -- Ref.[28]), 3D BN cellular architecture (23 -- Ref.[29]), polystyrene composite with 3D
segregated double networks (24 -- Ref.[30]), epoxy composite with 3D BN network (25 --
Ref.[31]), and 3D-BN nanosheets aerogels (26 -- Ref.[32]). 1- reproduced with permission.[8]
Copyright (2004), Elsevier. 2 and 3- reproduced with permission.[9] Copyright (2015),
Elsevier. 4- reproduced with permission.[10] Copyright (2017), Elsevier. 5- reproduced with
permission.[11] Copyright (2012), RSC publishing. 6- reproduced with permission.[12]
Copyright (2018), Elsevier. 7- reproduced with permission.[13] Copyright (2009), AIP
publishing. 8- reproduced with permission.[14] Copyright (2017), Wiley. 9- reproduced from
reference.[15] 10- reproduced with permission.[16] Copyright (2012), Elsevier. 11- reproduced
with permission.[17] Copyright (2015), American Chemical Society. 12- reproduced with
permission.[18] Copyright (2017), Wiley. 13- reproduced with permission.[19] Copyright
(2016), American Chemical Society. 14- reproduced from reference.[20] 15- reproduced with
permission.[21] Copyright (2015), American Chemical Society. 16- reproduced with
permission.[22] Copyright (2008), American Chemical Society. 17- reproduced with
permission.[23] Copyright (2016), Springer Nature. 18- reproduced with permission.[24]
Copyright (2016), American Physical Society. 19- reproduced with permission. [25] Copyright
(2016), American Chemical Society. 20- reproduced with permission.[26] Copyright (2019),
Elsevier. 21- reproduced with permission.[27] Copyright (2016), Elsevier. 22- reproduced with
permission.[28] Copyright (2018), Elsevier. 23- reproduced with permission.[29] Copyright
(2017), American Chemical Society. 24- reproduced with permission.[30] Copyright (2017),
4
American Chemical Society. 25- reproduced with permission.[31] Copyright (2017), American
Chemical Society. 26- reproduced with permission.[32] Copyright (2015), Wiley.
To facilitate various thermal management requirements, significant research efforts
have been devoted to either screen the materials with desired thermal transport properties or
design novel materials/structures that satisfy different requirements.[33] The advance of
nanotechnology enables the construction of novel materials from the bottom up, based on
low-dimensional nanomaterials (Figure 1d), including zero-dimensional (e.g., fullerene and
nanoparticles), one-dimensional (e.g., nanotubes, nanowires, and nanothreads),[34] and two-
dimensional (e.g., nanoribbons or nanosheets)[35] nanomaterials. This work brings together an
overview of the recent progress on the thermal transport properties of different three-
dimension (3D) nanostructures with a focus on experimental studies, ranging from 3D
nanoarchitectures (Figure 1e), metal-matrix composites with nanostructured fillers (Figure
1f), and the polymer composites with nanostructured fillers (Figure 1g). The discussions are
conceptually divided into three major parts, i.e., 3D nanostructures for low thermal
conductivity, 3D nanostructures for high thermal conductivity, and polymer composites with
nanostructured fillers for high thermal conductivity. Sections in each part are organized based
on the complexity of the nanostructures. The preparation or fabrication method of the
corresponding nanostructures will be briefly introduced before discussing their thermal
transport properties.
2. Three-dimensional Nanostructures for Low Thermal Conductivity
There is a plethora of 3D nanostructures being synthesized or theoretically predicted to meet
low thermal conductivity requirements, which are normally constructed from low-
dimensional nanomaterials (e.g., 0D fullerene, 1D nanowire or nanotube, and 2D nanosheet or
5
nanoribbon). In subsequent sections, the 3D nanostructures with low thermal conductivity are
critically discussed and generally divided into two categories considering their applications,
i.e., for thermal isolation and thermoelectric devices.
2.1. Low Thermal Conductivity for Thermal Isolation
Thermal insulating materials have been widely used in our daily lives, such as insulating
gloves, pipes and buildings.[36] To facilitate various advanced thermal insulating usages,
several different types of 3D nanostructures with a low thermal conductivity have been
synthesized or prepared. Based on their constituent components or building blocks, these 3D
nanostructures can be categorized as: colloidal crystals and assemblies based on 0D
nanostructures; 3D hierarchical structures/networks constructed from a combination of low-
dimensional nanomaterials; and highly porous nanostructures.
2.1.1 Polymer Colloidal Crystals and Assemblies
Colloidal crystals are typically prepared through the bottom-up approach with a highly
ordered hierarchical structures of colloidal particles.[17] They can be assembled into 2D films
and 3D structures with either a single type of particles or multiple types of particles, and
close-packed or non-close-packed with controlled defects. Different techniques have been
developed to fabricate 3D colloidal crystals, such as sedimentation,[37] electrodeposition,[38]
shear alignment,[39] and filtration.[40] Their hierarchical nature endows several unique features,
including wide variability of material composition, scalability, tunable symmetry (amorphous
or ordered crystals), and a large number of interfaces.[41] Extensive research efforts have been
devoted to colloidal crystals over the past 30 years, covering the constituent materials,
particle-particle interactions, applications (e.g., sensors, waveguides), and physical properties
6
(e.g., the opalescent color) of the colloidal crystals.[17, 42] However, their thermal transport
properties have rarely been investigated.
Figure 2a shows the types of structural hierarchies with colloidal particles in two- and
three-dimensions. Generally, monodispersed spheres favor close-packed face-center-cubic
(fcc) lattice geometry with particle volume fraction of 74% and 26% of interstitial space. The
large number of interfaces always induce strong interface phonon scattering and lead to low
thermal conductivity (k). This feature makes them ideal for thermal insulating applications.
For instance, a glass substrate with a hollow silicate particles coating has been reported to be a
thermal insulator, with a reduction of around 30% in k compared to the uncoated substrate.[43]
Colloidal particles can be synthesized as a solid sphere, core-shell sphere, or a hollow sphere
with different compositions. Polystyrene (PS) and poly(methyl methacrylate-co-n-butyl
acrylate) (PMMA) are the common materials for solid particle colloidal crystals (Figure 2b),
and the silica colloidal crystals are normally constructed from core-shell particles. The
thermal conductivity of colloidal crystals is calculated from 𝜅=𝛼𝐶-𝜌, where 𝐶- and r are
the heat capacity and density; a is the thermal diffusivity as determined from laser flash
analysis. In general, the polymer solid particle-based colloidal crystals have been reported
with low thermal conductivity, e.g., ~ 51 ± 6 mWm-1K-1 for a PS colloidal crystal at 25 °C,[44]
and 84 ± 2 mWm-1K-1 for PMMA (20vol% n-butyl acrylate (nBA)) colloidal crystal at 25
°C.[45]
7
Figure 2. The colloidal crystals/assemblies. (a) Structural hierarchies of colloidal particles in
two- and three-dimensions. Reproduced with permission.[17] Copyright 2015, American
Chemical Society. (b) Optical and Scanning Electron Microscope (SEM) images of PMMA
colloidal crystals. Reproduced from Ref.[46]
Temperature dependence: Due to the softening of polymer particles above their glass
transition temperature (𝑇/), the thermal conductivity of the colloidal crystals has a unique
continuous film when the temperature exceeds 𝑇/ (around 105 °C), and a significant thickness
crystals with an initial thickness of 1094 µm.[44] The initial particulate system transfers to a
temperature dependence. Figure 3a compares the thermal conductivity of PS colloidal
drop around 84%-89% is observed. As illustrated in Figure 3a (during the first measuring
cycle with increasing temperature from 25 °C), the thermal conductivity of the native
colloidal crystals, which is around 50 mWm-1K-1, has a weak relationship with the
temperature below 𝑇/. Due to the sudden thickness drop above 𝑇/, the thermal conductivity
increases to around 150 mWm-1K-1, which is retained in the following cooling cycle from 𝑇/
to 25 °C. Similar results are observed from the PMMA colloidal crystals as illustrated in
Figure 3b,[47] the thermal conductivity experiences a step-like, irreversible increase from
around 120 mWm-1K-1 to ~ 200 mWm-1K-1 at the transition temperature (around 74 °C). It is
observed that the initial contact points between colloidal particles are enlarged after reaching
the transition temperature (Figure 3c), which reduces the thermal interfacial resistance and
leads to a remarkable increase to their thermal conductivity, e.g., up to ~ 200% increase.
8
Figure 3. Thermal transport in polymer colloidal crystals. (a) The thermal conductivity of PS
colloidal crystals as a function temperature. The particle has a diameter of 366 nm. The closed
and open markers denote the heating and the cooling cycles, respectively. Reproduced with
permission.[44] Copyright (2015), Elsevier. (b) The thermal conductivity of PMMA colloidal
crystals with 20vo% of nBA. The particle has a diameter of 214 ± 7 nm. (c) Schematic
illustration of the change of interfacial contacts between polymer colloidal crystals after
reaching glass transition temperature. (b) and (c) are reproduced from Ref.[47]
The intrinsic step-like temperature dependence of the thermal conductivity of polymer
colloidal crystals opens avenue path to fabricate colloidal superstructures with geometric
constriction-controlled thermal transport properties. Figure 4a shows the fabrication of a
multilayer colloidal crystals with each of three layers containing different PMMA particles
Although the heat transfer will be enhanced for the individual layer after reaching its
(containing varying volume percentage of nBA) with a same diameter of ~ 420 nm but
different transition temperatures (i.e., 𝑇/,*=61 °C, 𝑇/,%=103 °C, 𝑇/,5=124 °C).[46]
corresponding 𝑇/, the thermal resistances or contacts between colloidal particles are
conductivity (Figure 4b). Each time the temperature exceeds the 𝑇/ of a given layer, the
unchanged in the unmolten layers and therefore the material retains a low effective thermal
effective thermal conductivity of the sample will experience a stepwise increase. With this
9
approach, colloidal crystals with specified number of stepwise increases in its thermal
conductivity can be designed, and the temperature range of the transition. By tuning the
thickness of the layers, the degree of transition change in each step (or the increment of the
thermal conductivity) can also be effectively controlled.
Figure 4. Thermal transport in colloidal crystals exhibiting multiple stepwise increase feature.
(a) Schematic illustration of a colloidal monolith comprised of different PMMA particle
layers and each layer has its own 𝑇/. (b) The corresponding stepwise increase of the thermal
conductivity due to the increasing temperature. Closed and open markers represent the heating
and cooling cycle. Reproduced from Ref.[46]
Particle size dependence: Besides temperature, the colloidal assemblies can also be prepared
from colloidal particles with different sizes to modify their thermal transport properties.
Figure 5a shows how the thermal conductivity of a binary PS-based colloidal assemblies can
be tailored by varying the percentage of the larger particles.[48] The two particles possess a
diameter of 𝑑9= 243 nm and 𝑑;= 306 nm, respectively, with a size ratio 𝜂= of 0.79 (here,
𝜂==𝑑9/𝑑;). It is shown that the effective k of the colloidal assembly is very sensitive to the
mixture ratio when the large particle volume ratio 𝜂?; is below 20% or above 80%. Here, 𝜂?;=
𝑉;/(𝑉9+𝑉;), and 𝑉9 and 𝑉; represent the volume of the small and large particles, respectively.
10
For intermediate mixing ratios 0.2 < 𝜂?; < 0.8, the effective k drops to around 80% compared
with that of the homo-particle colloidal assemblies.
Figure 5. Thermal transport in colloidal assemblies with binary PS particle mixtures. (a) The
normalized thermal conductivity (in relative to the sample with mono-large particles) as a
function of the large particle volume ratio. (b) Schematic comparison of the heat flux
streamline length in an ordered crystal and a disordered assembly. (c) The normalized thermal
conductivity as a function of the size ratio at 𝜂?; ~ 0.2. (d) The heat flux densities of particle
assemblies with varying size ratios. (e) Histograms of the streamline length at the size ratio of
0.54 (upper panel) and 0.80 (bottom panel). Reproduced with permission. [48] Copyright
(2018), Wiley.
Two mechanisms are responsible for the suppressed thermal conductivity in a
disordered colloidal assembly. One is the reduced density of the structure due to the non-
close-packing, as the binary assembles can have as much as a 90% density reduction
compared to the mono-particle crystals. The other is the increased thermal pathway or
11
streamlines for the heat transport resulting from the size mismatch between adjacent particles.
As illustrated in Figure 5b, the streamline is very straight and unperturbed in an ordered
assemble, but is strongly bent and perturbed with the introduction of disorder (or different
sized particles). According to the finite element method (FEM) modeling, the thermal
transport pathways in the colloidal crystal are uniform and quite straight (𝑑9 𝑑;⁄ =1.0 in
Figure 5d), whereas the disordered colloidal assemblies possess highly distorted pathways. As
illustrated in Figure 5e, the colloidal assembly with smaller size ratio contains higher number
of long pathways, and their lengths can reach around 140% in relative to the length of the
simulation box. Such observations suggest that the bigger difference between the particle
diameters, the more likely it is that the heat flux streamlines will be distorted and elongated.
As evidenced in Figure 5c, both experimental measurements and FEM simulations show a
systematic increase of the thermal conductivity when the particle size mismatch diminishes,
and a significant reduction of about 50% is observed for the scenario with a diameter ratio of
about 0.54.
2.1.2 Hollow and Core-shell Particle-based Colloidal Crystals and Assemblies
Hollow particles synthesized from silica or silicate are also being frequently utilized to
fabricate colloidal crystals. Comparing with the solid polymer particle, the hollow particle
colloidal crystals also provide effective opportunities to reach lower thermal conductivity. For
instance, the hollow silica colloidal crystal shows a low thermal conductivity of 35 mWm-1K-
1,[18] which is comparable with polymer foam-based thermal insulation materials.[36] Studies
show that the particle geometry, the packing density and symmetry, and the interparticle
bonding strength play an important role on the thermal transport of the hollow particle
colloidal crystals.[36, 49]
12
Four different heat transport mechanisms have been proposed for the hollow particle
colloidal crystals as illustrated in the Figure 6a,[41] including solid conduction, open- and
closed-pore volume gaseous conduction, radiative transport, and convection. The solid
conduction and the open- and closed-pore volume gaseous conduction are the major
contributors. According to Figure 6b, the colloidal crystals with larger particles possess
smaller thermal conductivity. For instance, the silica colloidal crystal with a particle size of
266 nm shows a thermal conductivity of about 80 mWm-1K-1 (in vacuum condition), which is
more than two times higher than the counterpart with a particle size of 469 nm (~ 25 mWm-
1K-1). The thermal conductivity also increases with the shell thickness, e.g., about 43% of
increase (from 46 to 82 mWm-1K-1) in vacuum condition when the thickness increases from
14 to 40 nm (with a similar particle size of 266 nm). Such results suggest that the colloidal
crystals can be built with a low thermal conductivity with larger particles but thinner shells at
a price of stiffness reduction.[50] The influence on the heat transport from the open-pore
volume is evidenced from the different thermal conductivity as measured in different
environments. As shown in Figure 6b, the colloidal crystals exhibit a much lower thermal
conductivity in vacuum condition than that under gaseous environment.
Figure 6. Thermal transport in colloidal crystal comprised of silica hollow sphere. (a)
Schematic view of the thermal transport pathways. Reproduced with permission.[18] Copyright
(2017), Wiley. (b) The thermal conductivity as a function of the diameter and shell thickness
13
of the hollow particle at 25 °C. The gaseous pressure is ~ 1000 mbar, and the vacuum is ~
0.05 mbar. Reproduced with permission.[41] Copyright (2017), Wiley.
Considering the dominant solid conduction mechanism (Figure 6a), the thermal
transport in the colloidal crystals can be further modified by changing the interparticle
interactions, including altering the interaction strength and the number of contact points.
Through calcination, strong covalent bonds can be introduced between adjacent hollow
spheres, which reduces interfacial phonon scatting and leads to higher thermal conductivity.
In the extreme scenario where there is only weak van der Waals (vdW) interactions, the
colloidal crystal yields to the lowest thermal conductivity of only about 8 mWm-1K-1 in the
vacuum condition (with a particle size and shell thickness of 469 nm and 17 nm,
respectively).[41] Altering the interparticle interactions will also influence the temperature
dependence of the thermal conductivity. For instance, a change in the temperature dependence
from 𝜅~𝑇G.H to 𝜅~𝑇G.I is observed for the silica hollow sphere colloidal crystal after 500 °C
and 950 °C calcination respectively.[49]
The hollow nature of the silica sphere also brings the opportunity to fabricate colloidal
crystals with core-shell particles, which brings a phase change characteristic to the silica
colloidal crystals. Figure 7a compares the thermal conductivity of colloidal crystals
constructed with the diameter of PS-silica core-shell particles ranging from 269 nm to 479 nm
(but an identical silica shell thickness of ~ 15 nm).[41] As is seen, the thermal conductivity
increases significantly during the first heating cycle, which remains at the higher level during
the cooling circle. Such results are analogous to that observed from the polymer particle
colloidal crystals (Figure 3). Scanning Electron Microscope (SEM) images reveal that the PS
leaks through the silica shells at higher temperature, which endows the thermal conductivity
of the colloidal crystal a stepwise characteristic. As shown in Figure 7c, necks between
14
adjacent particles are formed due to the PS leakage after the first heating cycle. Experimental
results show that the PS leakage can be prevented by increasing the silica shell thickness. As
compared in Figure 7b, the core-shell particle colloidal crystal with a shell thickness of 42 nm
exhibits an almost linearly increasing thermal conductivity. According to the SEM images,
there are no necks formed during the first heating circle.
Figure 7. The influence of particle geometrical sizes on the thermal conductivity of the core-
shell particles colloidal crystals. (a) The stepwise increase of the thermal conductivity when
the temperature increases for the particle diameter of 269 nm, 392 nm, and 479 nm,
respectively. All particles have an identical shell thickness of ~ 15 nm. (b) The temperature
influence on the thermal conductivity of the core-shell particles colloidal crystals with a
particle shell thickness of 42 nm. The diameter of the core-shell particle is 270 nm. A
comparison of the SEM images of the core-shell particles after heating up to 200 °C for: (c)
the particle with a diameter of 392 nm and shell thickness of 15 nm, and (d) the particle with a
dimeter of 270 nm and shell thickness of 42 nm. Reproduced with permission. [41] Copyright
(2017), Wiley.
In summary, investigations of the thermal transport properties of colloidal assemblies
are still in their infancy. The colloidal particles can be fabricated using different materials,
such as polymers, ceramics, metals, and semiconductors, and various shapes from solid
15
sphere, hollow sphere, core-shell, to non-spherical shapes.[17] These are also known to possess
inherent functional properties such as photonic and phononic properties, and with polymer
materials, the phase transition of the constituent materials gives the colloidal assemblies
unique temperature-dependent thermal conductivity. Specifically, different external stimuli
may be used to trigger the transition, such as pH, solvents, light, electric or magnetic fields.[46]
These varieties bring a huge possibility to construct multiphysical and multifunctional heat
management materials based on colloidal assemblies (that can be integrated into thermal
switches, transistors, or diodes), which is a fertile field for more research efforts in the near
future.
2.1.3 Three-dimensional Foam
Comparing with the colloidal crystals/assemblies, an analogous type of polymer-based 3D
porous structure -- polymer foam (also called nanocellular foam), has been extensively
studied. The extremely low thermal conductivity (typically in the range of 30 -- 40 mW/m-1K-
1) of the polymer foams makes them ideal for thermal insulation applications.[51] The polymer
nanocellular foams are normally prepared with nanoconfined voids (or channels with trapped
gas) in a stable solid matrix, producing open-cell foams or closed-cell foams.[51a]
Similar as the colloidal crystals in Figure 6a, there are four heat transfer mechanisms
involved in the nanocellular foams, including the contribution from radiation (𝜅J), convection
(𝜅K), gas phase conduction (𝜅/) and solid phase conduction (𝜅9). According to the Grashof
number (Gr), convection of air is enabled when (𝐺𝑟 ≥1000), which corresponds to the void
from 𝜅J=*O5PQRSTU
'(T)V . Here, 𝑛X and 𝜌 are the mean index of refraction and the apparent density
of the specimen, respectively; 𝑇 is the temperature; and 𝜎 is the Stefan-Boltzmann constant.
contribution to the heat transfer.[52] Meanwhile, the radiation contribution can be calculated
diameter larger than 10 mm. Thus, gas convection in the nanocellular foams makes no
16
constituent materials and affects the overall heat transfer of the nanocellular foams.[52] The
𝑒(𝑇) denotes the specific Rossland mean extinction coefficient, which dependents on the
kinetic theory of gases, i.e., 𝜅9=*5𝐶-𝑣/Λ, where 𝐶- and 𝑣/ are the specific heat capacity and
the mean group velocity of phonons; and Λ is the phonon mean free path (MFP).[1a] The gas
solid phase conduction depends on the phonon transport, which can be described by the
phase conduction originates from the collisions of gas molecules between themselves and
with the solid walls, which is described by the kinetic theory model with a correction factor
that accounts for the influence of viscosity.[53]
It is reported that the gas phase conduction contributes significantly to the effective
heat transfer of the polymer foam (up to 70%),[54] which can be calculated from 𝜅/=
𝜅G/(1+𝐾P). Here 𝜅G is the thermal conductivity of a bulk gas, and 𝐾P=Λ//𝑑 is the
Knudsen number (with Λ^ as the MFP of gas molecules and d as characteristic dimension of
the voids). It is clear from above relationship that the gas phase conduction decreases when
the Knudsen number increases. In other words, the dimension of the voids can effectively
change the thermal conductivity of the polymer foam. Using this concept, recent research has
focused on the fabrication of polymer nanocellular foams with gas-filled pore (or void) sizes
down to 100 nm, which can on the one hand prevent structural failure but on the other hand
turn the gas into Knudsen flow.[55] In theory, the gas molecules will enter the ballistic regime
and the gas phase conduction will be significantly suppressed when the pore sizes are on the
order of the MFP of gas molecules (e.g., around 70 nm for air at ambient condition).
Extensive results have shown an ultra-low thermal conductivity of the polymer nanocellular
foam with pore sizes down to 100 nm. For instance, Rizvi et al successfully fabricated
polypropylene foams using industrial-scale foam molding technique to achieve a low thermal
conductivity of 29 mWm-1K-1 (for air) at a density of 0.074 gcm-3.[56]
17
Overall, researchers have investigated the thermal transport of the polymer
nanocellular foams with various constituent polymer matrices, different pore sizes, and
different densities. Since there are already several recent reviews devoted to the thermal
conductivity of nanocellular foams,[51] we will not focus on this group of structures in this
review.
2.1.4 Nanolattice
Besides the colloidal crystals/assemblies and 3D polymer foams, a recently reported 3D
hierarchical structure -- the nanolattice, also shows promising applications for thermal
isolation. Nanolattices, also known as mechanical metamaterials, are prepared by high-
precision additive manufacturing techniques,[57] whose certain mechanical properties are
defined by their geometry rather than their composition.[58] Due to the size effect at nanoscale,
the nanoscale mechanical metamaterials, opens up a new material-property design domain.
Different techniques have been developed to fabricate nanolattices,[59] such as direct laser
writing (DLW), self-assembly, genetic engineering. Nanolattices fabricated from different
materials have been demonstrated, including carbon (Figure 8a),[23, 60] ceramic (Figure 8b),[61]
metals or alloys,[62] and metallic glass.[63] Extensive research has already explored the
mechanical performance of nanolattices, with excellent mechanical properties found,
including high deformability, high recoverability, and ultrahigh stiffness.[64] For instance, the
alumina nanolattices exhibit an excellent energy-absorbing capability, and the full recovery is
observed even with over 50% compressive strain.[61a] The deformation mode of the hollow-
tube Cu60Zr40 metallic glass nanolattices are found to be tailorable through the tube-wall
thickness and temperature, i.e., shifting from a full shape recovery characteristic to ductile,
and also brittle characteristics.[65]
18
Figure 8. Thermal transport in nanolattice. (a) A carbon nanolattice fabricated by DLW and
subsequent pyrolysis. Reproduced with permission. [23] Copyright (2016), Springer Nature. (b)
The hollow-beam alumina lattice showing the octet-truss architecture. (c) The influence of
density on the thermal conductivity of the alumina nanolattice. (d) The thermal conductivity
of different alumina nanolattices at varying temperature. The nanolattice has a wall thickness
of 36 nm, 81 nm, and 182 nm, respectively. (b)-(d) are reproduced with permission. [66]
Copyright (2018), American Chemical Society.
Though significant understanding of the mechanical properties of nanolattices has
been established, only one experimental work has probed the thermal transport properties of
the alumina nanolattice. The relative density and the structural imperfections (such as
thickness nonuniformities and beam waviness) are found to have a marked influence on the
heat transfer in alumina nanolattices.[66] Figure 8c compares the experimentally measured
thermal conductivity of the hollow-beam alumina nanolattices with the predictions based on
the resistor[67] and cellular models[68] (both models are only dependent on the relative density
of the structure). It is found that the measured thermal conductivity agrees well with the FEM
simulation results, which are based on the classical effective medium theories, implying the
dominant diffusive heat transport in the alumina nanolattice.
19
Figure 8d plots the temperature dependency of the thermal conductivity, from which
we see that k increases as the temperature increases. This observation is consistent with that
previously reported for amorphous oxide thin films.[69] The reasonable agreement between the
FEM simulations and experimental measurements further suggest diffusion heat transport is
dominant in the nanolattice. Overall, a low thermal conductivity, as low as 2 mWm-1K-1 is
measured in the alumina nanolattice. It is expected that the thermal conductivity of the
nanolattice can be further suppressed by creating multilayered walls (to promote interfacial
phonon scattering). Given the rapid advances in additive manufacturing technology and the
excellent mechanical performance of the nanolattices, it is of great interest to further exploit
their thermal transport properties for thermal isolating applications.
2.2. Low Thermal Conductivity for Thermoelectric Device
Another application that demands low thermal conductivity is the thermoelectric (TE) device,
which possesses great technological potentials in the energy sector. As mentioned previously,
efficient thermoelectric devices require a high figure of merit (ZT), which is calculated from
𝑍𝑇=𝑆%𝜅'𝜅()*𝑇. Here, S is the Seebeck coefficient; 𝜅' and 𝜅( are the electrical and thermal
conductivity, respectively; and T is the average temperature within the material. To achieve a
high ZT, significant work has been conducted to prepare or design materials with suppressed
thermal conductivity (but high electrical conductivity), such as core-shell NWs, phononic
crystals, GeTe-based nanomaterials,[70] and SnSe-based nanomaterials.[71] There are extensive
comprehensive reviews devoted to TE materials,[5] here we will only discuss the thermal
conductivity of some representative 3D nanostructures, including nanowire arrays, inverse
opals structures, and holey nanostructures or phononic crystals.
20
2.2.1 Nanowire Arrays
Si nanowire (NW) arrays are one of the nanostructures being frequently discussed for TE
device with a low thermal conductivity (but a high electrical conductivity).[72] Besides Si NW
arrays, metallic NW arrays have also been prepared, but targeting the usage as thermal
interface materials with a high thermal conductivity. For their utilization in thermal interface
materials, the NWs act as an expressway for the heat transport,[73] which are normally
infiltrated with a polymer to form a composite structure. We will discuss the thermal
conductivity of NW array-based composite in the following Sec. 4.4. NW arrays can be
prepared via electrodeposition (Figure 9a)[21] or bio-template mask (Figure 9b).[74]
Bulk Si has been shown to have a poor ZT ~ 0.01 due to its high thermal conductivity
(around 150 Wm-1K-1 at room temperature). Boukai et al reported that by varying the size and
impurity doping levels of Si NWs, the ZT value of the Si NW can reach ~ 1 at 200 K, which
is approximately 100-fold improvement.[75] These results demonstrate the promising
applications of Si NWs in TE devices, and greatly promote the studies on the thermal
transport properties of Si NW arrays.
Studies have shown that the thermal conductivity of Si NW arrays is highly dependent
on their geometrical parameters (e.g., length and diameter), pore size, the polymer filler, and
surface doping.[76] For instance, the discrete surface doping of Ge on Si NW arrays can induce
23% reduction in the thermal conductivity at room temperature compared to the undoped
counterpart.[76d] As illustrated in Figure 9c, the thermal conductivity of the Si NW array
decreases continuously with the increase of pore diameter and a low k around 1.68 Wm-1K-1
is observed at 300 K with a mean pore size around 12.35 nm.[77] The decreasing k results from
enhanced phonon scattering at boundaries and nanopores.[78] Through the bio-template mask,
recent studies show a much smaller k of the Si NW arrays around 0.5 ± 0.1 Wm-1K-1 along
the axial direction over a temperature range of 300-350 K. These arrays have a high density
21
and ordered structure (with a height of 30 nm).[74a] When the NW length (material thickness)
increases to 100 nm, k increases to 1.8 ± 0.3 Wm-1K-1 (see Figure 9d). The increasing thermal
conductivity with the length is commonly observed in 1D nanostructures[34a, 79] and 2D
nanoribbons,[80] which can be explained from the perspective of the kinetic theory.[81] By
embedding the same NW array into SiGe0.3 matrix, a similar low thermal conductivity of 3.5
± 0.3 Wm-1K-1 is measured in the composite structure in the temperature range of 300 -- 350
K.[74b] Such low thermal conductivity of the Si NW array results from the enhanced phonon
boundary scattering and the phonon confinement effect.[82]
Figure 9. Thermal transport in NW arrays. (a) SEM images of the Cu NW array. Reproduced
with permission.[21] Copyright (2015), American Chemical Society. (b) SEM image of Si NW
array. Reproduced with permission.[74a] Copyright (2017), AIP Publishing. (c) The axial
thermal conductivity of etched Si NW array as a function of the mean pore diameter at 300
K.[77] The diameter of the NW varies from 20 nm to 200 nm. (d) The axial thermal
conductivity of the Si NW array as a function of the sample thickness in the temperature
range of 300-350 K. The Si NW has a dimeter of 10 nm. Reproduced with permission. [74a]
Copyright (2017), AIP Publishing.
Overall, the Si NW array has been shown with significantly suppressed thermal
conductivity but unchanged electronic properties,[72b] and their thermal transport properties
22
have been extensively studied, targeting TE applications.[83] Besides NW arrays, a lot of
research has focused on the thermal conductivity of individual Si NW by investigating the
impacts from a wide range of factors, such as surface roughness, geometrical size (diameter
and length), temperature, defects (e.g., grain boundaries or stacking faults), and doping.[34a, 72b,
84] In addition to that, extensive studies have also been carried out on different Si crystal
nanostructures, such as the porous Si nanomeshes (as discussed in the following Sec. 2.2.3),
and the nanocrystalline bulk Si.[72c]
2.2.2 Inverse Opals Structures
The structural periodicity in the inverse opals (IOs) structures provides an effective avenue to
tailor their heat transport properties for TE devices. Analogous to the colloidal crystals, the
intrinsic optical properties of IOs enabled the fabrication of sensors based on color responses.
Their highly porous nature with a large surface-to-volume ratio also make them ideal building
blocks for the Li-ion batteries (e.g., the anodes or cathodes).[85] Generally, IOs are fabricated
by first depositing the inverse opals materials into the interstitial spaces of a sacrificial opal
template, and then removing/dissolving the opal template, which will thus form a 3D structure
with interconnected spherical pores.
The heat transport path in the IOs can be mimicked by the interconnected hard spheres
that occupy the interstitial sizes as schematically shown in Figure 10a,[19] and two controlling
mechanisms are proposed for the thermal transport.[86] At the continuum length scale (the
diffusive heat transfer), the porosity distorts the heat transfer pathway, which induces strong
thermal resistance (similar to the example in Figure 5b for the colloidal crystals). At the
micro/nanoscale, strong phonon scattering occurs at surfaces and grain boundaries, which
remarkably suppresses the thermal conductivity of the material. Figure 10b compares the
thermal conductivity of Si inverse opals with varying grain size (s) and shell thickness (h).
23
The Si inverse opals are fabricated from the 3D fcc silica opal template that is comprised of
particle's diameter in the range of 300 to 640 nm.[86-87] As is shown, a low thermal
conductivity (~ 1 Wm-1K-1) is observed for all examined Si inverse opals. Specifically, the
temperature dependence of the thermal conductivity can be approximated as 𝜅~𝑇*._ in the
low temperature regime, which deviates from the well-known ~𝑇5 dependence of the heat
capacity (for bulk). Such deviation is believed to result from coherent phonon scattering at
grain boundaries that is frequency-dependent, similar to what is observed in nanocrystalline
Si.[88]
Figure 10. Thermal transport in inverse opals structures. (a) Schematic view of a fcc inverse
opal unit cell (left). The octahedral and tetrahedral interstitial sites are highlighted by the
space-filling spheres. The interconnected networks between the space-filling spheres (red) and
the linkages (blue) represent the heat transfer pathway (red arrows). Reproduced with
permission.[19] American Chemical Society. (b) The influence of temperature on the thermal
conductivity of the inverse opals. The structures have different grain size (s) and silicon shell
thickness (h). Reproduced with permission.[86] Copyright (2013), American Chemical Society.
(c) The thermal conductivity of Cu and Ni close-packed inverse opals as a function of the
pore diameter at room temperature. Reproduced with permission.[19] Copyright (2016),
American Chemical Society.
24
The thermal conductivity of the inverse opals is highly related with the pore size, as
expected from the primary mechanisms governing heat transfer discussed above. Figure 10c
compares the thermal conductivity of Cu and Ni inverse opals measured at room temperature
when the pore diameter changes from 100 nm to 1000 nm.[19] Basically, the thermal
conductivity increases when the pore diameter increases, which saturates to the diffusive
limit. In particular, the Cu inverse opals with the smallest pore diameter (~ 100 nm) is only
about 43% of its counterpart with a dimeter of 1000 nm. According to the kinetic theory, the
thermal conductivity is linearly related with the MFP of the energy carriers under the gray
approximation, which is reasonable as the metals have a narrow electron MFP distribution
with energy (e.g., 90% of electron thermal conductivity comes from electrons with MFPs
⁄
+1 Λ9bJ'cK'
d)*
between 22 nm and 41 nm for Au).[89] Since both surface scattering and internal scattering
(e.g., electron-phonon, electron-grain boundary) affect the heat transfer in inverse opals, the
effective electron MFP (Λ''') can be calculated based on Matthiessen's rule from Λ'''=
⁄
a1 Λ=X''
. For the fcc inverse opals, the surface-limited MFP Λ9bJ'cK'≈
0.233𝐷 (the ballistic limit), with D as the diameter of the pore.[19, 90] Thus, the thermal
conductivity of the inverse opals can be derived as a 𝜅gh=𝜅gh,=X''×0.233𝐷/(0.233𝐷+
Λ=X''). According to the Drude-Sommerfeld model, the diffusive electron MFP is 39 nm for
Cu and 6 nm for Ni at 293 K. Evidently, the thermal conductivity of the Cu and Ni inverse
opals align with the diffusive limit as the effective MFP is close to the diffusive MFP for large
pore diameter. As such, Ni inverse opals exhibit a diffusive-like thermal conduction for a
diameter of around 100 nm as they have a short electron MFP.
Like the colloidal assemblies, the thermal transport properties of the inverse opals
structures are still lacking of investigation. Given the large variety of materials (e.g., metals
and semiconductors), there are great opportunities to design novel inverse opals-based heat
management materials. Besides TE devices, inverse opals structures also exhibit promising
25
applications for phase change materials used for thermal storage. For instance, by infiltrating
Cu inverse opals with paraffin wax, the temperature rise in kWcm-2 hotspots is suppressed by
~ 10% compared with the copper thin film heat spreaders.[91] In the thermofluidic system,
metal inverse opals can be used as microfluidic heat exchanger with a high thermal
conductivity. In general, inverse opals represent an important category of porous
morphologies for enhanced interfacial transport in applications like electrochemical surfaces
and microscale heat exchangers or suppressed heat transfer for TE devices, while their
thermal conduction principles (e.g., the heat transfer across the interface) are still requiring
extensive studies.
2.2.3 Holey Nanostructures
In recent years, the 3D nanostructure with regular vertically etched holes -- holey Si structure
or phononic crystals (as acoustic analogs of photonic crystals), have attracted extensive
experimental and theoretical efforts. The periodic holes are usually introduced to the sample
by etching.[92] In theory, the periodic holes will remarkably increase the surface-to-volume
ratio and enhance surface phonon scattering, which will thus significantly suppress the
thermal conductivity and benefit the applications in TE devices.[93] The majority of current
studies have focused on the Si phononic crystals being prepared as 2D nanomeshes or films.
A low thermal conductivity of around 1.9 Wm-1K-1 was reported for a holey silicon film,
which is more than 20% reduction compared to the Si NW array device.[93] As illustrated in
Figure 11a, a holey silicon nanostructure can be described by three geometrical parameters,
i.e., the neck size (n), pitch length (p) and thickness (L). The hole diameter can be calculated
from 𝑑=𝑝−𝑛.
The holey nature of the structure endows the phononic crystal with anisotropic thermal
conductivity. Specifically, the cross-plane thermal conductivity can be tuned by the sample
26
thickness and the in-plane thermal conductivity can be tailored by the hole diameter (or the
increasing tendency with the sample thickness. Such size-dependence can be described by the
neck size). As illustrated in Figure 11b, the cross-plane thermal conductivity exhibits an
spectral scaling model that accounts for the phonon frequency dependence, i.e., 𝜅l=
𝜅ln(𝜔,𝑛)×p1+qrn(s,P)
. Here, 𝜔 and 𝜔w are the phonon frequency and the
∫
svG
t %⁄
Debye cut-off frequency, respectively; and 𝜅ln(𝜔,𝑛) and 𝜆ln(𝜔,𝑛) are the thermal
u)*𝑑𝜔
conductivity and phonon MFP, respectively.[94] Similarly, the increase of the neck size results
in a larger thermal conductivity (Figure 11c).[25, 92]
Figure 11. Thermal transport in holey silicon. (a) Schematic illustration of the holey silicon
nanostructure. (b) The thermal conductivity of the holey silicon at different sample lengths.
(a) and b are reproduced with permission.[94b] Copyright (2015), American Chemical Society.
(c) The increasing thermal conductivity with the neck size. Reproduced with permission.[25]
Copyright (2016), American Chemical Society.
Besides the dependence on the geometrical parameters, the thermal conductivity of the
holey silicon exhibits a temperature dependence (𝜅~𝑇%) different from the classical
temperature dependence (𝜅~𝑇5) in the temperature range of 20-60 K.[94b] Two mechanisms
are proposed for such observation, one is the quantum confinement effect, which alters the
phonon dispersion relation.[95] Another one is the enhanced role of the surface disorder and
27
the boundary scattering, which is experimentally observed at room temperature.[96] The
phonon confinement effect has been argued in literature, some studies show that the
confinement effect is only significant for low temperature, e.g., around and below 6.4 K for Si
NWs with a diameter of 10.8 nm.[97] While, other studies show that the phonon confinement
occur even at room temperature.[98]
Overall, low thermal conductivity, sufficient electrical properties and good mechanical
strength make the holey Si nanostructures or Si phononic crystals an ideal candidate for TE
applications. Though it is widely accepted that the nanoscopic holes suppress the thermal
conductivity, the underlying mechanism is still under argument. The phonon dispersion
relations are potentially modified due to the presence of periodic holes in the nanostructure.
From the perspective of coherent phonon transport, the thermal conductivity would be
reduced due to the resulting phononic bandgaps and the reduced group velocities.[99] Several
works have argued the role of coherent phonons on the thermal transport of holey
nanostructures,[25, 96, 100] and the range of occurrence of the phonon interference is still an open
question. In a recent work, Maire et al show the thermal conduction controlled by coherence
in Si holey nanostructures over a relatively large temperature range, until the transition to
purely diffusive heat conduction was observed at 10 K.[100a] With recent advances of
nanofabrication technology, it is expected that the working temperature range of phononic
crystals can be further broadened.
3. Three-dimensional Nanostructures for High Thermal Conductivity
At the opposite extreme of TE devices or thermal isolation, many electronic applications
require materials to possess a high thermal conductivity for efficient heat removal during
operation. In this regard, extensive works have been conducted to explore the nanomaterials
with high thermal conductivity, especially for the thermal interface material (TIM) as used in
28
modern electronic packing. Here we will discuss the recent progress on three categories of
nanostructures pursuing high thermal conductivity, including the 3D architectures, metal-
matrix composites with nanostructured fillers, and polymer composites with 3D
nanostructured fillers. The thermal transport properties of polymer composites will be
discussed separately in the following section due the large volume of relevant works in
literature.
3.1. Hierarchical Structures for Thermal Management
3.1.1 Nanotube Arrays
Besides low thermal conductivity, the assembled or packed hierarchical nanostructures have
also been prepared or proposed for high thermal conductivity applications, among which, the
array structure is one of the simplest scenarios. Vertically aligned carbon nanotube (VACNT)
arrays have received extensive attention. Due to the high thermal conductivity of individual
CNTs, the VACNT arrays possess a high effective thermal conductivity parallel to the CNT
axial direction, which make them ideal candidate as thermal interface material. Researchers
have prepared TIMs based on either pure VACNT arrays or VACNT array-based composites
(embedded in polymer). Almost all studies have shown that VACNT composites exhibit a
decreased performance compared with their pristine array counterparts,[20] which is caused by
the degraded CNT alignment, damping of phonon modes, and poor contact between CNTs
and substrates. In this regard, a large volume of work has focused on the heat transport of
pristine VACNT arrays. Typically, the total thermal resistance of a TIM can be determined
⁄
from 𝑅Tgz=𝑅K*+ℎ;( 𝜅XP(
+𝑅K%, which correspond to three components (see Figure 12a),
including the thermal contact resistances of the interfaces at the two ends (𝑅K* and 𝑅K%), and
⁄
thermal resistance of the TIM (ℎ;( 𝜅XP(
).[20] Here, ℎ;( and 𝜅XP( are the thickness and the
intrinsic thermal conductivity of the TIM, respectively. Therefore, to ensure effective heat
29
removal, it is essential to minimize the thermal contact resistances at the two interfaces and
enhance the intrinsic thermal conductivity of VACNT arrays.
Figure 12. Thermal transport in vertically aligned CNT arrays. (a) Schematic illustration of a
pure CNT array-based TIM. Reproduced with permission. [101] Copyright (2013), American
Physical Society. (b) The influence of volume fraction on the intrinsic thermal conductivity of
VACNT arrays, including multi-walled CNT (MWCNT) arrays (1 -- Ref.[102], 2 -- Ref.[103], 3 --
Ref.[104], 4 -- Ref.[105], 5 -- Ref.[106], 6 -- Ref.[107], 7 -- Ref.[108], 8 -- Ref.[109], 9 -- Ref.[110], 10 --
Ref.[111], 11 -- Ref.[112], 12 -- Ref.[113], 13 -- Ref.[114], 14 -- Ref.[115], 15 -- Ref. [116], 16 -- Ref.
[117]), and single-walled CNT (SWCNT) arrays (17 -- Ref.[118], 18 -- Ref.[119]). Error-bars
represent the range of the thermal conductivity measured under different conditions.
Reproduced with permission.[101] Copyright (2013), American Physical Society.
The intrinsic thermal conductivity of VACNT arrays is highly dependent on the
quality of the constituent CNTs and their packing density. Catalytic chemical vapor
deposition (CVD) has been widely used to grow VACNT arrays from supported or floating
catalysts, and many factors have been identified that affect their quality, such as the growth
temperature, pressure, precursors, and the active catalyst.[120] The as-produced high-quality
VACNT arrays exhibit a thermal conductivity in the order of 10 Wm-1K-1, which is better than
most commercially available TIMs.[20] Various post-growth treatments have been proposed to
30
enhance the heat transport in VACNT arrays, such as high-temperature annealing, microwave
annealing, and coating. An earlier dedicated review comprehensively compared the intrinsic
thermal conductivity of VACNT arrays reported before 2013.[101] Thus, this section will be
limited to extend some of the results reported recently. Figure 12b summarizes the intrinsic
thermal conductivity of VACNT arrays at different volume fractions. The dashed/solid lines
represent the predicted intrinsic thermal conductivity of VCNAT array based on 𝜅XP(=
𝜙𝜅KP(.[101] Here, 𝜙 is the volume fraction, and 𝜅KP( is the thermal conductivity of an
individual CNT. According to Figure 9b, some VCANT arrays exhibit an equivalent thermal
conductivity of the individual CNT less than 30 Wm-1K-1, signifying other factors other than
the CNT density degrade the thermal transport in the array structure. Overall, all reported
thermal conductivities of the array fall well below the predictions based on the individual
CNT with a thermal conductivity of 3000 Wm-1K-1.[121]
To date, considerable work has been carried out to investigate the thermal transport
properties of VACNT arrays, which have focused on their intrinsic thermal conductivity. To
facilitate their engineering applications, more studies are expected to investigate the total
thermal resistance of the VACNT array-based TIMs, especially the interface heat transfer
between VACNT array and the mating surface. Currently, it is still challenging to fabricate
and transfer high quality VACNT arrays, which greatly affect the interfacial heat transfer.[20]
It is shown that different agents (e.g., In, Cu, Au) can be used to bond CNT tips with mating
surfaces to reduce the thermal contact resistance,[116, 122] however, it is unclear how reliable
the bonding structure is under long-time thermal cycling.
3.1.2 Pillared Structures
Along with the CNT arrays, the high thermal conductivity of graphene has also enabled it as
appealing building block for 3D hierarchical structures.[123] Pillared graphene is one of
31
hierarchical structures that received extensive attentions recently, in which the graphene layer
is supported by CNT pillars (Figure 13a and b). The CNT pillared graphene structures were
initially proposed for the H2 storage application,[22] and subsequently their electronic transport
properties[124] and mechanical properties[125] have been studied. Although several papers have
reported the successful synthesis of the pillared graphene,[126] their properties have rarely been
studied experimentally, primarily due to the preparation and measurement difficulties. To
date, only one study has investigated the thermal transport in CNT pillared graphene,[127]
which however only measured the thermal resistance (9 ´ 10-10 m2KW-1) at the CNT-
graphene junction and demonstrated the high heat dissipation capability of the structure. In
comparison, several papers have investigated the thermal conductivity (either in-plane or
cross-plane) of the pillared graphene through molecular dynamics (MD) simulations,[128] from
which the major factors influencing their thermal transport properties have been identified,
such as the junction configurations,[129] and the CNT length and density (or inter-CNT
distance).[130]
Figure 13. Thermal transport in pillared graphene. (a) Atomic configuration of a CNT
pillared graphene. Reproduced with permission.[131] Copyright (2014), Elsevier. (b) Simulated
STEM image of CNT and graphene junction areas. Reproduced with permission. [126b]
Copyright (2012), Springer Nature. (c) The thermal conductivity of 3D carbon nanostructures
32
as a function of the hot-pressing pressure. CNT/EGB represents 3D CNT/exfoliated graphite
block,[132] and CNT/GCC represents 3D CNT/graphite consolidated composite.[133]
Apart from the hybrid structures of the CNT-pillared graphene, there are also some
other 3D carbon-based hierarchical structures being reported, as listed in Table 1, including
the CNT/graphene oxide(GO) hierarchical structure, CNT/graphene hierarchical structure
(annealed from CNT/GO at 1000 °C);[134] the 3D CNT network;[135] the CNT/ultrathin
graphite foams;[136] and the 3D CNT intercalated graphite.[132-133] For the 3D CNT/GO
hierarchical structured film with 15wt% of CNTs, an ultrahigh in-plane thermal conductivity
of 1388.7 Wm-1K-1 was reported at 1000 °C, which decreases at either higher or lower weight
fraction of CNTs.[134] In comparison, the 3D CNT network exhibits a cross-plane k of about
19.8 Wm-1K-1, which decreases with increasing transverse interconnected CNTs.[135]
Table 1. The thermal transport properties of 3D CNT based hierarchical structures.
3D hierarchical
structures
CNT/graphite foam
(1.8wt% graphite and
0.8wt% CNT)
CNT/exfoliated
graphite block (EGB)
CNT/graphite
consolidated
composite (GCC)
3D CNT/GO (15wt%
CNTs, at 25 °C)
3D CNT/graphene
(15wt% CNTs, at
1000 °C)
Preparation method
k (Wm-1K-1) R (cm2KW-1) Year
CVD and hot-pressing
CVD and hot-pressing
CVD and hot-pressing
Vacuum filtration
Vacuum filtration and
annealing
4.1 ± 0.3
38 (𝜅~);
290 (𝜅∥)
24.3 (𝜅~);
211.5 (𝜅∥)
~ 3 (𝜅∥)
1388.7 (𝜅∥)
-
-
-
-
-
33
2015[136]
2016[132]
2014[133]
2017[134]
2017[134]
3D CNT network
CNT pillared
graphene
CVD
CVD
~19.8 (𝜅~)
0.67~0.78
2018[135]
-
9 ´ 10-6
2018[127]
For the 3D hierarchical structures prepared through a hot-processing technique, their
thermal conductivities are highly dependent on the growth time, growth density, hot-pressing
temperature, and pressure.[132-133] A higher hot-pressing temperature leads to a higher cross-
plane and in-plane thermal conductivity.[132] Figure 13c shows the cross-plane (𝜅~) and in-
plane (𝜅∥) thermal conductivity under different hot-pressing pressure for the 3D CNT/GCC
and the 3D CNT/EGB. Due to the high in-plane thermal conductivity of the graphite plate, the
in-plane thermal conductivity of the CNT/GCC or CNT/EGB is nearly one order of
magnitude greater than the cross-plane thermal conductivity. In general, the in-plane and
cross-plane thermal conductivities increase as the hot-pressing pressure increases, which
results from a denser structure and fewer voids at higher pressure. The decreasing of 𝜅~ for
CNT/EGB (from 5 MPa to 15 MPa) is likely to be a result of the interplay between interlayer
thermal contact resistances and in-plane thermal conduction in the graphite plate.[132] In
particular, the graphite plates are expected as not aligned completely perpendicular in the
pressing direction under low pressure, and thus their high in-plane thermal conductivity
contributes to a higher overall thermal conductivity. At higher pressure, the graphite plates are
better aligned, which counterbalances the reduced thermal resistance between graphite layers
and leads to a temporary reduction to 𝜅~.
3.1.3 Honeycomb
It is worthy to mention that researchers have also proposed to construct 3D nanostructures
based on 2D nanomaterials such as graphene[137] and BN,[137a] which form a regular
34
honeycomb structure. The initial honeycomb structure was proposed by Karfunkel and
Dressel in 1992.[138] Thereafter, there has been a sustained research on honeycomb structures,
covering the structural stability and varieties,[139] mechanical properties,[140] electronic
properties,[141] thermal transport properties,[139a, 141-142] and gas storage applications.[24]
However, majority of these works are based on theoretical calculations or MD simulations,
and the successful synthesis of the honeycomb structure was not reported until 2016 (Figure
14a).[24] Besides the carbon 3D honeycomb structures, recent studies also discussed the boron
nitride-based honeycomb.[143] To the best of our knowledge, the thermal conductivity of
honeycomb structure has not been experimentally measured. According to the theoretical
calculations or MD simulations, the carbon honeycomb exhibits anisotropic thermal transport
properties, i.e., a larger axial thermal conductivity than the transverse thermal conductivity.
These in silico studies have also shown marked influences on the thermal conductivity of the
honeycomb structures from various factors, such as the cell structures (or morphologies),[141,
142c] mechanical strain,[142b] and temperature.[142a] Figure 14b illustrates the specific axial and
lateral thermal conductivity of a carbon honeycomb structure with varying density as obtained
from density functional theory (DFT) calculations. Comparing with the thermal conductivity
of graphene (~ 4000 Wm-1K-1), the honeycomb structure has a considerably smaller k by a
factor of 15 to 45.[139a]
35
Figure 14. Thermal transport in carbon honeycomb structure. (a) Transmission electron
microscopy (TEM) images of a carbon film showing the honeycomb structure. Reproduced
with permission. [24] Copyright (2016), American Chemical Society. (b) The DFT calculated
thermal conductivity of a carbon honeycomb structure,[139a] and the experimental
measurements for 3D graphene and 3D BN foam (denoted as 3D-C and 3D-BN,
respectively).[137a]
A few works have also studied the thermal conductivity of 3D foam structures
constructed from 2D nanostructures. One recent work shows that the 3D-C and 3D-BN (as
prepared via CVD method) exhibit an increasing cross-plane thermal conductivity (𝜅~) when
the density of the structure increases (Figure 14b). Comparing with the 𝜅~ of graphene and
hBN paper (between 1-5 Wm-1K-1),[144] the 3D-C and 3D-BN exhibit a significantly high 𝜅~
of 86 ± 10 Wm-1K-1 and 62 ± 10 Wm-1K-1, respectively. Such remarkable enhancement is
resulted from the covalent connections at the interfaces, which enables efficient phonon
transport.[137a]
Overall, there are extensive works being conducted on CNT arrays, however, current
studies on the thermal transport properties of other complex 3D hierarchical nanostructures
(such as pillared graphene, honeycomb) are still very limited. For the CNT/graphite
composites, the introduction of CNTs are found to enhance the cross-plane thermal
conductivity, whereas they will impair the heat transport in the in-plane direction. In terms of
the pillared structures and honeycomb stuctures, their thermal conductivity is still yet to be
measured experimentally. There are several difficulties around these hierarchical structures,
for instance, it is very challenging to fabricate and characterize 3D hierarchical
nanostructures. The experimental measurement is also facing extensive challenges due to the
structural complexities.
36
3.2. Metal-matrix Composites with Nano-structured Fillers
Metal-matrix composites (MMCs), comprising a contiguous metal matrix and reinforcements
(filler materials), are widely used as large-scale structural materials.[145] Within the MMC, the
metal matrix acts as a compliant support for the reinforcement fillers, and commonly used
metals include Al, Ag, Cu, Mg, and Ti. The reinforcement fillers are used to tailor the
properties of interest. Conventional MMCs are reinforced by ceramic fibers and particles in
order to achieve a high-temperature resistance, high mechanical performance (e.g., specific
strength and specific elastic modulus), and a low coefficient of thermal expansion (CTE).
Besides structural materials, MMCs also have broad applications in the thermal management
field for miniaturized electronic devices. Compared to their monolithic metal counterparts,
MMCs are expected to possess a high thermal conductivity and a small CTE. Benefiting from
their ultra-high thermal conductivity and low CTE, carbonaceous nanomaterials (e.g., CNT or
graphene) have been widely utilized as a filler for MMCs. To illustrate, at room temperature,
the multi-walled CNTs (MWCNTs) exhibit a thermal conductivity over 3000 Wm-1K-1 and a
CTE of -2.5 ppmK-1.[1a, 146] Whereas, pure Cu has a k of around 400 Wm-1K-1 and a CTE of
17 ppmK-1.[147] Several papers have comprehensively reviewed the fabrication/processing
approaches, mechanical properties, tribological properties, corrosion properties, electrical
properties, and applications of various MMCs.[145, 147-148] This section will outline the thermal
conductivity of various MMCs as reinforced by CNTs.
3.2.1. Copper-matrix
Cu-matrix composites are the most extensively studied MMCs due to their wide applications
in electrical and electronic applications. However, current studies show negligible
enhancement or even degradation of the thermal conductivity of the Cu-MMCs when CNT
37
fillers are used, compared to the thermal conductivity of pure Cu.[149] Several factors have
been proposed to be responsible for such observation, including the intrinsic compatibility (or
wettability) between CNTs and Cu, interfacial bonding, and dispersion.[150] These factors are
closely related to the fabrication or processing techniques. Most of the current studies have
discussed the volume (or weight) fraction dependence of the thermal conductivity, which
however show divergent results.
Figure 15a compares some results that report a decreasing thermal conductivity when
the volume fraction of CNT increases,[151] and some others that reported an initial increasing
tendency of k within a small volume fraction.[149, 152] Several factors may be responsible for
such divergent observations, such as processing techniques; the dispersion/agglomeration and
alignment of CNTs (CNT has a very low thermal conductivity in the transverse direction,
~5.5 Wm-1K-1);[153] interfacial bonding at CNT/Cu interface; and the geometrical parameters
(length and diameter) and defects of CNT.[151c] The CNT/matrix interfaces act as a thermal
barrier that causes strong lattice phonon scattering, resulting in degradation of the overall
k.[149] It is observed that there are extensive defects (including defects in CNT and grain
boundaries in Cu matrix) and voids formed in the Cu-MMCs, and the high aspect ratio of
CNT also leads to severe entanglement and kinks (Figure 15b-c).[151a] These facts promote
phonon scattering and lead to a reduction in k. Several studies have reported an improved
thermal conductivity with surface treatments (e.g., with Mo-coating[154] or chemical
treatment[152]), whereas the enhancement is very limited. It is shown from theoretical
calculations (based on the mixture rule) that the super-aligned CNTs can aid the heat transfer
in the Cu-MMC, however it is still lacking experimental validation.[155]
38
Figure 15. Thermal transport in Cu-MMCs reinforced by CNTs at room temperature. (a) The
thermal conductivity of different Cu-MMCs at varying volume fraction of CNT, including the
Cu matrix with pristine MWCNT and chemically treated MWCNT (denoted as MWCNT/Cu
and c-MWCNT/Cu, respectively),[152] with Mo-coated MWCNT (M-MWCNT/Cu),[154] and
with super-aligned MWCNT (s-MWCNT/Cu, theoretical calculation).[155] (b) TEM images of
sintered Cu. (c) TEM images of pristine MWCNT/Cu (left), the interface between MWCNT
and Cu (middle), and the schematic view of the interface. (d) TEM images of chemically
treated MWCNT/Cu (left), the interface between MWCNT/Cu (middle), and the schematic
view of the interface. (b)-(d) are reproduced with permission. [152] Copyright (2012), Elsevier.
Very few works have explored how the thermal conductivity of Cu-MMCs will
change under different temperatures. One early work shows that the thermal conductivity
fluctuates around 250 Wm-1K-1 for a CNT volume fraction of 5 vol.% in the temperature
range of 298 K to 550 K,[149] indicating a relatively temperature independent behavior.
However, a recent work shows that the thermal conductivity decreases from 310.5 Wm-1K-1
to 296.7 Wm-1K-1 in the temperature range of 20 °C to 500 °C.[27] Overall, to achieve a high
thermal conductivity, the Cu-MMCs are supposed to contain well-aligned and well dispersed
CNT fillers. The interfacial bonding of CNT/matrix and the structural damage of CNTs are
also crucial influential factors, which are mainly unknown in the current studies.
39
3.2.2. Other Metal-matrices
In addition to Cu, CNTs have also been introduced to other metal matrices such as Ag, Al, Ni,
Ti, Mg and alloys, but the composite thermal conductivities have not been fully investigated.
Similar to Cu-MMCs, both enhancement and degradation effects are observed in the heat
transfer due to the introduction of CNTs. For example, k of the Ag-MMC with chemically
treated MWCNTs deceases significantly from 430 Wm-1K-1 to ~ 175 Wm-1K-1 when the
volume fraction of CNT increases from zero to 6%,[156] which is most likely as a result of the
covalent bonds between CNT and the matrix that impair the electron conduction. By
comparison, the Ag-MMC with non-covalent functionalized MWCNT (that improves the
wettability of CNT with the matrix) exhibits an increasing k with the CNT content (Figure
16a). Some early work shows that the CNT/Al MMC exhibits an increasing k at low volume
fraction which then decreases at higher volume fraction.[157] Interestingly, the experimental
results show that the alloy-matrix MMCs normally show an increasing thermal conductivity
with increasing CNT volume fraction, such as the Cu-Cr composite,[158] Cu-Ti composite,[159]
and 70Sn-30Bi alloy composite (with Ni-coated CNTs).[160] Such enhancement is explained
from the perspective of the wettability of CNT, i.e., a thin interface layer that reduces the
Kapitza resistance formed in the alloy matrix.
Figure 16. Thermal conductivity of various MCCs reinforced by CNTs at room temperature.
(a) The thermal conductivity exhibiting a divergent relationship with the volume fraction of
CNT for CNT/Cu-Cr,[158] CNT/Cu-Ti,[159] CNT/Ag (with covalent functionalization),[156] N-
40
CNT/Ag (with non-covalent functionalization),[156] MWCNT/Al (produced by spark plasma
sintering, the weight fraction is converted to volume fraction based on a density of 1.8 gcm3
for MWCNTs and 2.7 gcm3 for Al).[157] (b) The thermal conductivity of different MMCs
reinforced by CNTs, including Al-MMCs (1 - Ref.[161], 2 - Ref.[157], 3 - Ref.[162], 4 -- Ref.[163]),
Cu-MMCs (5 -- Ref.[149], 6 -- Ref.[151a], 7- Ref.[154], 8 -- Ref.[152], 9 -- Ref.[155], 10 -- Ref.[164], 11
-- Ref.[150], 12 -- Ref.[165], 13 -- Ref.[27], 14 -- Ref. [166]), Ag-MCCs (15 -- Ref.[156], 16 -- Ref.[167]),
and alloy-MCCs (17 - Ref.[158], 18 -- Ref.[159], 19 -- Ref.[160], 20 -- Ref.[147]). Dashed lines
denote the thermal conductivity of monocrystalline metals. The error bars represent the
variation of the thermal conductivity in the studied weight/volume fraction range. The solid
markers represent a decreasing thermal conductivity within the examined weight/volume
fraction, and the open markers represent an initial increasing and then decreasing thermal
conductivity.
Figure 16b summarizes the reported thermal conductivity of MMCs reinforced by
CNTs with their monolithic metal counterparts. Since the measurement methods vary in
literature, and the MMCs contain different CNT volume or weight fractions, their thermal
conductivities can only be qualitative compared. Compared to their monolithic metal
counterparts, almost all types of CNT reinforced MMCs show a smaller thermal conductivity.
In the MMC, the heat transfer relies on both electrons and phonons (i.e., 𝜅=𝜅'+𝜅-), and
the electron heat transfer plays the dominant role. Experimental results show that the addition
of CNTs significantly disturb and block the electron transportation path, and this results in a
decreasing k when the filler loading increases.[157] In all, the CNT fillers are found to reduce
the CTE for the MMCs, while high thermal conductivity has not yet been achieved.
The investigation of the thermal transport properties of MMCs is still at a very early stage.
The thermal conductivity of the MMC depends on many factors, such as the intrinsic heat
41
transfer in the metal matrix and the nanofiller, and the interfacial thermal resistance.
However, it is challenging to characterize the interfacial bonding between the metal matrix
and the nanofiller, and the structural damage of the nanofiller during sample preparation is
hard to be identified. It is also difficult to control the dispersion and alignment of nanofillers
during sample preparation. Extensive experimental, theoretical and computational effort is
still required to understand how the nanofiller can be used to enhance the heat transfer in
MMCs.
4. Polymer Composites with Nanostructured Fillers for High Thermal Conductivity
As the common thermal interface materials (TIMs) in the electronic packing industry,
polymer composites have received extensive interests from both scientific and engineering
communities.[168] They also have broad applications as underfill materials, organic substrate
materials, and in flexible electronic devices. To ensure efficient and effective thermal
management, polymer composites are expected to possess a high thermal conductivity. As
such, a large amount of research has focused on preparation of polymer composites with a
high thermal conductivity by adding different types of highly (thermally) conductive fillers.
Theoretically, there are many factors that constrain the thermal conductivity of the polymer
composite, including the intrinsic thermal conductivity of the filler, the filler/filler and
filler/matrix interface resistances, the dispersion and alignment of the filler, and the
geometrical parameters of the filler (e.g., size, shape, and surface area).[169] Owing to the
absence of heterogeneous interface, 3D nanostructured filler is the ideal option to reduce the
filler/filler interfacial thermal resistance with a relative low filler loading. This section is
intended to outline the recent progress in using different types of 3D nanostructured fillers in
polymer composites.
42
4.1. Carbon Nanotube-based 3D Nanostructured Fillers
CNT has been reported with a thermal conductivity of ~3,000-3,500 Wm-1K-1 at room
temperature along the axial direction.[121, 170] Extensive efforts have been devoted to
incorporate CNT into polymers to improve their thermal conductivity. Researchers have
reported a thermal conductivity of CNT/polymer composite reaching 10 Wm-1K-1 at a CNT
loading less than 5wt%.[171] The common problem with CNT fillers is that they will improve
the electric conductivity of the composites, which can be a problem when electrical insulation
is required. To resolve this issue, different surface modifications have been proposed, such as
fluorination, which could on the one hand achieve similar enhancement to k compared with
the pristine CNT while retaining a good electrical insulation characteristic of the
composite.[172] Several earlier papers already contain a wealth of information on the thermal
transport properties of CNT reinforced polymer composites, covering the influential factors
like the CNT geometries, alignment, dispersion, weight/volume fraction, and surface
modifications.[173] As such, this section will only limited to the recent studies on the thermal
transport properties of polymer composites with 3D nanostructures/networks based on CNT.
The implementation of 3D network filler can be dated back to 2005 when Zhang et
al.[174] demonstrated a 3D network structure from the in situ growth of CNTs from the
interlayer of clay. Thereafter, such method has been widely accepted and applied to construct
3D hybrid network structures using two or multiple fillers, such as graphene (or graphite)
nanoplatelets/CNTs,[175] graphene oxide/CNTs,[176] boron nitride (BN)/CNTs,[177]
SiC/CNTs,[178] and BN/AlN.[179] Based on the nature of the bond between CNT and other
fillers, the following discussions will first discuss the thermal transport properties of polymer
composites with weakly bonded 3D nanostructures (a common synergistic network), followed
by the polymer composites with 3D nanostructured (a hetero-structure) fillers with chemically
bonded interfaces.
43
4.1.1 Synergistic Network
A synergistic effect in a polymer composite with two or more fillers arises when the increase
in thermal conductivity of the composite is greater than the sum of the increase in thermal
conductivity for each individual component. Benefiting from their ultra-high aspect ratio
(length over diameter) and thermal conductivity, CNTs have been popularly adopted to form
synergistic networks with other fillers, in which the synergistic networks are normally
generated through melting and mixing binary or more type of fillers directly with the polymer.
The synergistic effect (𝑓) can be quantified as 𝑓=
((cid:129)(cid:130))(cid:129)(cid:132))(cid:133)((cid:129)(cid:131))(cid:129)(cid:132)),[180] where 𝜅(cid:134) and 𝜅(cid:135) are
(cid:129)(cid:130)(cid:131))(cid:129)(cid:132)
the thermal conductivity of the polymer composite with filler A and B, respectively; 𝜅- and
𝜅(cid:134)(cid:135) are the thermal conductivity of the pristine polymer and the composite with the AB
synergistic network, respectively. The effective synergistic network is formed when f is larger
than 1, and greater f suggests stronger synergistic effect. Two-dimensional nanomaterials,
such as BN nanosheet (BNNS) and graphene (nanosheet or nanoplatelets), have been
commonly used to form the synergistic networks with CNT.[177, 180-181] For BNNS, its
electrical insulation characteristic can interrupt the electrically conductive network of CNTs
in the polymer composites, and retain the electrical insulation characteristic of the
polymer.[182]
To optimally use the high intrinsic thermal conductivity of the filler in the polymer
matrix, a percolation network of the filler is necessary, which relies heavily on the dispersion,
aspect ratio, and content (volume fraction) of the filler.[183] Various additional factors are
responsible for the synergistic effect of the 3D structured networks, such as the alignment and
content of each filler, and the percolation threshold of individual fillers.[177, 181, 184] Figure 17a
plots the k as a function of the filler loading for high-density polyethylene (HDPE)
composites containing different types of fillers.[184] For the HDPE with mono-fillers, i.e., only
44
expanded graphite (EG) filler, or only CNT filler, k exhibits a nearly linear increase tendency
with the increase of filler loading. Such relationship is generally described by the Nan model,
i.e.,(cid:129)(cid:136)(cid:129)(cid:137)=1+(cid:138)(cid:139)(cid:134)(cid:140)5
(cid:129)(cid:141)/(cid:129)(cid:137)(cid:134)(cid:140)(cid:133)(cid:142) ,[185] which is derived from the Maxwell-Eucken model[186] with the
(cid:129)(cid:141)(cid:129)(cid:137); 𝜅' and 𝜅(cid:145) are the
consideration of interfacial thermal resistance. Here, 𝑃=%(cid:140)(cid:144)(cid:129)(cid:137)=
of polymer matrix, respectively; 𝐴𝑅 and Φ' are the aspect ratio and volume fraction of the
filler, respectively; 𝑅(cid:129) is the interface thermal resistance; 𝜅K is the thermal conductivity of the
filler; and 𝑑 is the diameter of the filler. From Figure 17a, the k of the HDPE composite with
effective thermal conductivity of the polymer composite and the intrinsic thermal conductivity
10wt% of EG increases linearly with the CNT content, similar as that observed from the
EG/HDPE with increasing EG content. Such results indicate that the synergistic effect is not
activated. In comparison, the addition of ~ 0.5wt% of CNTs for the 15wt%EG/HDPE and
20wt%EG/HDPE composites triggers a sudden increase to the effective thermal conductivity,
signifying the synergistic effect.
Figure 17. Thermal transport in HDPE composites with CNT-based synergistic networks. (a)
The thermal conductivity of HDEP composites with single CNT fillers, single EG fillers, and
a binary CNT and EG fillers. Reproduced with permission.[184] Copyright (2017), Elsevier. (b)
The thermal conductivity of HDPE composites prepared by different techniques. R and P
represent the hot rolling and hot-pressing, respectively. The samples contain single BN fillers,
and a binary BN and CNT fillers. The dispersion states of the HDPE composites (25wt%BN
45
and 3wt%CNT) prepared by: (c) the rolling technique, and (d) the hot-pressing technique. (b)-
(d) are reproduced with permission.[177] Copyright (2018), Elsevier.
The fabrication technique is found to have a profound effect on the synergistic
network. Figure 17b illustrates the content dependence of the k for HDPE composites molded
by hot-pressing (P) and hot rolling (R) (including BN and CNT fillers).[177] As is seen, the
samples molded by hot rolling exhibit a much higher thermal conductivity, and the synergistic
effect is activated at a much lower content of BN (~ 5 wt%).[177] According to the SEM
images (Figure 17c and 17d), the hot rolling is able to arrange BN and CNTs along the rolling
direction, which results in better filler alignment and more thermally conductive paths.
4.1.2 Segregated Double Network
Since the synergistic networks are normally generated through melting and mixing/blending
binary or more types of fillers directly with the polymer, the enhancement to the heat transfer
is still limited. To obtain interconnected networks with a lower filler loading, researchers
proposed the fabrication of segregated double networks in the composites,[30, 187] i.e., the
polymer particles are embedded with one conductive filler before being coated with another
conductive filler. Figure 18a compares the thermal conductivity of PS composites that are
prepared through three different procedures with a binary filler of CNT and 3.5vol% graphene
nanoplatelets (GNPs), including a randomly dispersed network, the segregated network, and
the segregated double network (Figure 18b).[30] As is seen, the thermal conductivity of the PS
composites behaves differently with the volume fraction. For the sample with randomly
dispersed network, k shows a gradual increase with no sharp change as the MWCNT content
increase and no synergistic effect is observed. For the sample with segregated network, a
decreasing k is observed, which is resulted from the interrupted contacts of GNPs/GNPs due
46
to the introduction of CNTs. In comparison, the sample with segregated double network
shows a significant increase in k with a small content of CNT (~ 1.5vol%). Such observation
suggests that the synergistic double network can significantly enhance the heat transfer in the
polymer composite compared with other 3D networks.
Figure 18. Thermal transport in PS composites with synergistic networks. (a) The distinguish
changing tendency of the thermal conductivity of the PS composites (containing
3.5wt%GNPs) as prepared with a randomly dispersed network (RD), a segregated network
(SN), and a segregated double network (SDN).[30] (b) The corresponding schematic
illustration of the different polystyrene composites. (a) and (b) are reproduced with
permission.[30] Copyright (2017), American Chemical Society.
4.1.3 Hetero-structure
In contrast to the weak interactions between fillers in the synergistic network, extensive
efforts have also been attributed to establish 3D hetero-structures based on CNTs, within
which the filler/filler interfaces are chemically bonded. The covalent bonds that bridge the
multi-filler interfaces are expected to further reduce phonon scattering and achieve a more
efficient heat transfer in the polymer composites. The hetero-structures are normally prepared
by growing or grafting CNTs on the surfaces of other structures (e.g., carbon fibers[188]) or use
CNT as skeletons to grow other low-dimensional nanostructures (e.g., MoS2, graphene,[189]
47
and BN nanosheet[190]). Different from the synergistic network, the hetero-structure allows a
good control of the alignment of fillers, which endows the polymer composites with a
controllable anisotropic thermal conductivity, i.e., a high cross-plane thermal conductivity
along the CNT axis, and a low in-plane thermal conductivity in the transverse direction.
Figure 19a depicts the variation of the thermal conductivity of epoxy (EP) composites
with different filler contents, including single CNT filler, hetero-structured MoS2/CNT filler,
and hetero-structured MoS2/graphene/CNT filler. The hetero-structure is prepared by growing
MoS2 and graphene on surface of CNTs via a hydrothermal method.[189] As is seen, for the
filler loading less than around 20wt%, k increases almost linearly with the filler content.
According to the Foygel model,[191] the effective thermal conductivity of the composites can
be estimated from 𝜅'=𝜅GaΦ'−ΦKd((c)
, where 𝜅G is the effective thermal conductivity of
the filler network; 𝑡(𝑎) depends on the aspect ratio of the filler; and ΦK is the volume fraction
measurements, the thermal contact resistance 𝑅K is then calculated from 𝑅K=
p𝜅G𝐿ΦK((c)u)*
.[192] It is estimated that the 𝑅K for CNT/EP and MoS2/CNT/EP are around 2.17-
of the percolation threshold. Note that the Foygel model assumes percolating networks in the
matrix with random distributed fillers. With the fitted values from the experimental
3.98 ´ 107 KW-1 and 1.9 ´ 107 KW-1, respectively, which are much higher than the
MoS2/graphene/CNT/EP (8.3 ´ 106 KW-1). This enables higher heat transfer efficiency in the
MoS2/graphene/CNT/EP composite than its counterparts.
48
Figure 19. Thermal transport in the polymer composites with hetero-structured fillers. (a)
Cross-plane thermal conductivity of EP composites with different fillers (including CNT
filler, MoS2/CNT filler, and MoS2/graphene/CNT filler), and the silastic (SI) composites with
carbon fibers (CFs) with 35 µm long aligned CNTs grown on their surface. (b) The highest
thermal conductivity and corresponding weight fraction using CNT hetero structure fillers or
networks (1-Ref.[188]; 2- Ref.[187b]; 3- Ref.[193]; 4- Ref.[176]; 5- Ref.[194]; 6- Ref.[195]; 7- Ref.[181];
8- Ref.[196]; 9- Ref.[182b] ; 10- Ref.[190]; 11- Ref.[189]; 12- Ref.[184];13- Ref.[177]1;14-
Ref.[182a] ;15- Ref.[197]). The marker shapes present the polymer base type, and the colors
present the filler structures: the blue is the hetero-structured fillers (HS); the red is the random
synergistic network (RN); the magenta is segregate network (SN); and the black is the
skeleton network (Skeleton).
Other larger scale skeletons have also been used to grow CNTs on their surfaces to
prepare hetero-structured fillers. For instance, the vertically aligned carbon fibers (CFs) with
CNTs grown on their surfaces can increase the k of silastic (SI) composite to 8.14 Wm-1K-1
with a total carbon load of only 2.86wt%.[188] A recent paper reported the growth of CNT on a
Cu foam, which successfully bridges the holes in the foam, and enables a higher k of 3.49
Wm-1K-1 in the paraffin composite.[197] Comparing with the synergistic networks, the
preparation of hetero-structures is more complicated for industrial scale-up.
49
Figure 19b summarizes the recent works on the polymer composites with CNT-based
3D networks. For comparison, the polymer composites with randomly dispersed pre-modified
CNT fillers are also listed, including the polymer functionalization,[193, 198] AlN doping,[199]
and metal nanoparticles (NPs) deposition.[195] Surface functionalization are introduced in the
purpose of reducing the thermal resistance at the CNT/polymer matrix interface. From Figure
19b, although the percolation networks are presented in the composites, majority of the
prepared composites still exhibit a k lower than 2 Wm-1K-1, and usually a large filler loading
(over 20wt%) is required to obtain a higher thermal conductivity.
It is noted that nearly all studies have emphasized the relationship between the thermal
conductivity and the filler content. Unfortunately, it is challenging to establish a uniform
theoretical model that can be used to describe and predict the thermal conductivity of the
polymer composites. The difficulties arise from the complexities of the heat transfer in the
polymer composites, which is related with the filler network, filler structure or morphology,
filler dispersion, intrinsic thermal conductivity of the filler, and the interfacial thermal
resistance (or phonon scattering) at the filler/filler interface and filler/matrix interface.
Although different structures/networks have been explored, there is still a significant gap
between the effective thermal conductivity of the polymer composites and the intrinsic
thermal conductivity of nanostructured filler. Thus, a fundamental challenge that deserves
further investigation is how to minimize the interfacial thermal resistance. Different
functionalization strategies have been proposed for the nanofiller, which however normally
degrades the intrinsic thermal conductivity of the filler.
4.2. hBN-based 3D Nanostructured Fillers
The high thermal conductivity, electrical insulation and low coefficient of thermal expansion,
make inorganic ceramics another popular type of filler for insulator polymer nanocomposites.
50
Hexagonal boron nitride (hBN) is one of the most extensively investigated ceramic fillers for
polymer composites. There are four common structures of hBN, including 0D NPs, 1D
nanotubes (NTs), and 2D nanosheets (NSs), and 3D nanostructures (e.g., foam and network).
Xu et al.[169] briefly reviewed the thermal conductivities of polymer composites reinforced by
these BN nanostructures. A very recent review also concisely discussed the heat transport in
the polymer composites with 2D BNNSs.[200] As such, this section will be limited to the
thermal conductivity of polymer composites with 3D BN nanostructures or networks.
4.2.1 hBN Nanotubes
Similar to CNT, hBN nanotubes (BNNTs) possess high thermal conductivity (~ 350 Wm-1K-
1)[201] and high aspect ratio, which make them ideal for building percolation networks in
polymer composites. Extensive research has investigated the thermal conductivity of BNNTs
reinforced polymer composites, with an emphasis on the influences from the volume (or
weight) fraction and temperature. The big gap between the axial and lateral thermal
conductivity of the BNNT lead to the polymer composites (with aligned BNNTs) exhibiting
anisotropic thermal transport properties, i.e., a higher thermal conductivity along the tube axis
than that across the tube axis.
The temperature dependency of the thermal conductivity varies with the polymer
matrix, which is related with the glass transition temperature (𝑇/) of the polymer matrix. For
disordered nanostructures, the thermal conductivity increases with temperature when 𝑇<𝑇/,
which is caused by the enhanced phonon transmission at the interfaces.[202] For the crystalline
phase, k decreases with temperature due to the Umklapp phonon scattering. As illustrated in
Figure 20a, the in-plane thermal conductivity of the cellulose nanofiber (CNF) composites
increases with temperature before reaching ~ 60 °C,[201, 203] suggesting that the thermal
51
transport of the amorphous polymer dominates the heat transport. In comparison, only a slight
increase is observed for the epoxy composite with hBNNTs.[204]
Figure 20. Thermal conductivity of polymer composite reinforced by hBNNTs. (a) The
temperature dependency of the cross-plane thermal conductivity (𝜅~) for the 25wt%
BNNT/CNF,[201] 25wt%Ag-BNNT/CNF,[203] and 30wt%BNNT/epoxy composites.[204] (b)
The thermal conductivity of different polymer composites reinforced by BNNTs, including
thermoplastic polyurethane (TPU, 1 -- Ref. [205]), polyacrylic acid (PAA)/polyvinyl alcohol
(PVA) polymers (PAA/PVA, 2 -- Ref. [206]), CNF (3 -- Ref. [203], 4 -- Ref. [201], 5 -- Ref. [172]),
polyvinylpyrrolidone (PVP, 6 -- Ref. [207]), EP (7 - Ref. [204], 8 -- Ref. [201] ), PVA (9 - Ref. [208],
10 -- Ref. [209], 11 - [201]), polymethyl methacrylate (PMMA, 12 - Ref. [210]), polyvinyl butyral
(PVB, 13 - Ref. [210] ), polystyrene (PS, 14 - Ref. [210]), polyethylene vinyl alcohol (PEVA, 15
- Ref. [210]), and polyvinyl formal (PVF, 16 - Ref. [209]). The open and solid markers represent
the in-plane and out-of-plane thermal conductivity, respectively.
The weight (or volume) fraction exerts a marked influence on the thermal conductivity
of the polymer composite. Both in-plane and cross-plane thermal conductivity are found to
increase with the increase of weight fraction.[204-205] However, for the CNF composite, the in-
plane thermal conductivity decreases with the weight fraction when the BNNTs content is less
than 25wt% due to the lack of effective connections (or percolation networks) at low content
of CNF.[201, 203] Different models have been developed to interpret the relationship between
52
the thermal conductivity of the composite and the filler loading, e.g., the effective medium
theory model that incorporates the interface resistance,[211] the Lewis-Nielsen model,[212] and
the nonlinear model developed by Foygel et al.[191] Assuming the percolation critical power
law, the best fitted cross-plane thermal conductivity of the CNT composite is around 60.2
Wm-1K-1, which is comparable to the range of 55-170 Wm-1K-1 for vertically aligned BNNT
films. Such observation suggests the crucial role of the thermal resistance at the BNNT/matrix
interface.
Figure 20b compares the thermal conductivity of various polymer composites
reinforced by BNNTs. It is seen that the thermal conductivity of the polymer composite
exhibits a large variation even the filler loading is the same. Such discrepancy primarily
originates from the different alignment of the BNNTs and also the interfacial resistance of
BNNT/polymer as determined by the preparation approach. For instance, using the hot-
pressing technique, the BNNT/PMMA achieves a k of 3.16 Wm-1K-1 at 24 wt% of
BNNTs,[210] which is 5-fold higher than the BNNT/PVP(polyvinylpyrrolidone) fibers
fabricated from electrospun (about 0.54 Wm-1K- at 30 wt% of BNNTs).[207] According to Fu
et al.,[203] a low weight fraction of 0.199% of Ag NPs reduces the interfacial resistance of
25wt% BNNT/CNF from 1.7 ´ 10-9 m2KW-1 to 0.73 ´ 10-9 m2KW-1, which increases the k
from ~ 12.9 Wm-1K-1 to 20.89 Wm-1K-1. Overall, high thermal conductivity (> 5 Wm-1K-1) of
polymer composites with BNNTs have been reported, which however normally requires a
high filler loading (over 20wt%).
4.2.2 Three-dimensional BN Networks
As aforementioned, 3D interconnected network is an effective way to enhance the heat
transfer in polymer composites while keeping a relative low filler loading, which is benefited
from the minimized interfacial phonon scattering between fillers. Different
53
approaches/methods have been developed to construct 3D nanostructures based on BNNS or
BNNT.[213] For instance, Zeng et al developed an ice-templated fabrication method,[32] which
produces an ordered 3D honeycomb-like network (Figure 21a). Wang and Wu proposed to
fabricate 3D BNNS network via multiple layer-by-layer assembly using 3D melamine foam
(MF) (Figure 21b).[28] Through a two-step CVD method, Xue et al. synthesized a BN cellular
architecture made of interconnective nanotubular hBN (Figure 21c).[29]
The temperature dependence of k for polymer composites varies with both the
polymer matrix and the 3D BN networks. Take the epoxy composite with nacre-mimetic 3D
BN network as an example, its thermal conductivity in the parallel direction to the lamellar
layer (cross-plane) appears sensitive to the temperature, which increases from 6.07 Wm-1K-1
to 8.41 Wm-1K-1 when the temperature increases from 20 °C to 100 °C. However, in the
perpendicular direction (in-plane), k is insensitive to temperature.[214] Comparing with the
composite with randomly dispersed BNNS, a stronger temperature dependency of k is
observed for the composite with 3D BN networks. Though some works report a slight
reduction of the thermal conductivity at higher temperature, such as the polydimethylsiloxane
(PDMS) composite with aligned and interconnected BNNS.[215] A general increasing thermal
conductivity with increasing temperature is observed for the 3D BN networks reinforced
polymer composites, which is resulted from the enhanced interfacial phonon transmission at
the filler/matrix interface. Similar as the composites with BNNTs (Figure 20a), k exhibits a
declination after the temperature is over a certain threshold value, e.g., around 70 °C for
epoxy composite.[31, 216] Such observation is explained from the perspective of the aggravated
phonon Umklapp scattering of the crystalline BN filler, which outweighs the decreased
Kapitza resistance, and thus impairs the thermal transport in the composite.[1a]
54
Figure 21. Thermal transport in polymer composites with 3D hBN networks. (a) SEM images
showing the 3D-BNNS aerogels prepared from ice template. Reproduced with permission.
[32] Copyright (2015), Wiley. (b) SEM images of MF foam with BNNSs assemblies.
Reproduced with permission. [28] Copyright (2018), Elsevier. (c) SEM images of 3D BN
cellular architecture made of interconnective nano-tubular hBN. Reproduced with permission.
[29] Copyright (2017), American Chemical Society. (d) The thermal conductivity of epoxy
composite reinforced by 3D BN networks, including the randomly dispersed BNNS (r-
BNNS/EP, 9.29 vol% BNNS),[32] ice-templated assembly (ice-BNNS/EP, 9.29 vol%
BNNS),[32] CNF supported (CNF-BNNS/EP, 9.6 vol% BNNS),[217] nacre-mimetic (nm-
BNNS/EP, 15 vol% BNNS),[214] and hierarchically ordered (h-BNNS/EP, 31.8 vol%
BNNS).[31] (e) The thermal conductivity of various polymer composites reinforced by 3D BN
networks, including PDMS (1 -- Ref. [215], 1 -- Ref. [218], 3 -- Ref.[219] , 4 -- Ref. [220]),
polyamideimide (PAI) (5 -- Ref. [213]), EP (6 -- Ref. [214], 7 -- Ref. [32], 8 -- Ref. [217], 9 -- Ref. [7a],
10 -- Ref. [28], 11 -- Ref. [216], 12 -- Ref. [31]), polyimide (PI) (13 -- Ref. [221], 14 -- Ref. [222], 15 --
55
Ref. [223], 16 -- Ref. [224]). NS represents the BNNS-based 3D network, and PL/P represents
BN platelets- or particles-based 3D structure.
Comparing with the randomly dispersed BNNSs, the 3D BN network shows much
stronger enhancement to the heat transfer in the polymer composite.[32, 217-218] Depending on
the fabrication approach, the 3D BN network can also endow the polymer composite with a
tailorable anisotropic thermal conductivity.[214] Figure 21e summarizes the thermal
conductivity of different polymer composites reinforced by 3D BN networks. It is found that
3D BN network brings a significant enhancement to the thermal transport of the polymer
composites compared to the random dispersed BNNS or BN platelets/particles. Based on the
Foygel model,[191] the 3D BN network is found to possesses a much smaller filler/filler
interfacial resistance compared with the composite with randomly dispersed BNNSs, which
leads to a higher thermal conductivity. According to Figure 21e, very few polymer composites
have been reported with a k over 5 Wm-1K-1 with a filler loading within 20vol%. Although
some works reported a promising enhancement with a very low volume fraction (e.g., a k of
1.17 Wm-1K-1 for the 2vol% BNNS/PAI (polyamideimide) composite), it is still unclear
whether such high enhancement can be realized in a larger volume fraction. A few works
reported a high k over 10 Wm-1K-1, which however requires a very high filler loading and the
3D structure is based on BN platelets.
Like the CNT-based 3D nanostructured fillers, most of current studies have
emphasized the thermal conductivity of polymer composites containing various filler content,
under different temperature, or with varying processing parameters/settings. A wide range of
factors are shown to affect the heat transfer in the polymer matrix, such as the filler alignment
and orientation, filler size and geometry, filler content. These results suggest several key
challenges that need to be addressed for engineering application, including how to fabricate
56
and characterize high-quality hBN-based 3D fillers; how to effectively and efficiently control
the homogeneous dispersion of the fillers. More importantly, how to minimize the thermal
interfacial resistance, especially between the BN filler and the polymer matrix. In addition, to
facilitate the engineering applications, it is necessary to investigate how the thermal
conductivity of the polymer composites will change under cyclic thermal or mechanical
loadings, which still requires great research efforts. An early work show that the thermal
conductivity of 3D BNNS/PU (polyurethane) composite maintains a high thermal
conductivity after 100 times of bending and twisting tests, which decreases significantly when
the tensile strain exceeds ~ 20%.[225]
4.3. Ceramic Fillers
In addition to hBN, other high conductivity inorganic ceramics have also been employed to
reinforce polymer composites, such as SiC, Si3N4, Al2O3, and AlN.[226] They are normally in
the form of NW or nanoparticle (NP), and being randomly dispersed in the polymer
composites via solution mixing or milling. Due to the strong interfacial thermal resistance, the
enhancement on the heat transfer from the random dispersion is limited. Increasing the filler
loading can help establish a percolation network and further enhance the heat transfer in the
composite, while it will generally degrade the mechanical properties of the composite
remarkably. In this regard, recent works have focused on the ceramic fillers based on 3D
aligned NWs or networks.
SiC is one of the popular ceramic fillers being used to in the polymer composites.
Different methods have been developed to construct 3D SiC networks based on SiC NWs,[227]
or Si NWs and graphene NS.[228] The intrinsically different thermal conductivities of the 1D
NW (along the axial and lateral directions) brings the opportunities to prepare the polymer
composite with anisotropic thermal conductivities.[227] Figure 22a compares the thermal
57
conductivity of the EP composite at different filler loadings. Despite different types of SiC
fillers, the EP composites show a similar k less than 0.4 Wm-1K-1 in the low filler loading
regime (< 2wt%). With the further increasing of the filler loading, the 3D SiC NW network is
found to induce a significant enhancement to the thermal conductivity, e.g., a k of about 1.67
Wm-1K-1 at the weight fraction of 5.6%. In comparison, the EP composites with randomly
dispersed SiC NWs and NPs show a much smaller k, e.g., k is ~ 0.36 Wm-1K-1 at 10 wt% of
SiC NP. Such observation is reasonable as the NPs are not able to form effective pathways for
the thermal transport compared with that of the NWs. At low filler content, the filler/matrix
interface thermal resistance dominantly affect the heat transfer in the polymer composites.
Whereas, at higher filler loading (with effective percolation network), the filler/filler interface
thermal resistance dominantly affect the heat conduction.[229] Therefore, for 3D networks with
the presence of percolation network, the filler/filler interfacial thermal resistance is the main
factor that impair effective heat transfer. Based on the Foygel model,[230] the 3D
interconnected network possesses a much smaller contact resistance compared with that of the
random dispersed NW network.[231] To note that, excessive loading would weaken the heat
transfer by creating voids or defects in the polymer matrix. Such structural defects will induce
strong phonon scattering and degrade the thermal conductivity in the polymer composites.[232]
Figure 22. The thermal conductivity of polymer composites reinforced by ceramic fillers. (a)
The thermal conductivity of the epoxy composite as a function of the weight fraction of SiC
58
NPs (SiCNP/EP),[232] randomly dispersed SiC NWs (r-SiCNW/EP),[233] and 3D SiC NWs
(3D-SiCNW/EP).[227] The volume fraction is converted to the weight fraction based on a
density of 1.175 g·cm3 and 3.2 g·cm3 for the epoxy matrix and SiC NW, respectively.[228a]
The thermal conductivity of the epoxy matrix is 0.2 Wm-1K-1. 𝜅∥ and 𝜅~ represent the in-plane
and out-of-plane thermal conductivity. (b) A summary of the thermal conductivity of the
polymer composited reinforced by ceramic fillers, including EP composite (1 -- Ref.[234], 2 --
Ref. [233], 3 -- Ref. [232], 4 -- Ref. [227], 5 -- Ref. [235], 6 -- Ref. [236], 7 -- Ref.[231], 8 -- Ref. [237], 9 --
Ref. [238], 10 -- Ref. [239]), PI (9 -- Ref. [228b]), Polypropylene (PP, 10 -- Ref. [240]), phenolic resin
(PR, 11 -- Ref. [241]), and high density polyethylene (HDPE, 12 -- Ref. [242]).
Comparing with the carbon-based and hBN-based fillers, the investigation on the
thermal transport in the polymer composites reinforced by other ceramic fillers are still
limited. The filler content is the most frequently discussed influential factor, and only a very
few studies have explored the temperature influence. A recent work shows that the k of
epoxy composites containing 3wt% of SiC NWs increases with temperature.[233] Figure 22b
summaries the thermal conductivity of polymer composites reinforced by different ceramic
nano-fillers. Comparing with the hBN-based filler (in Figure 21b), the SiC, Al2O3, and AlN
fillers are still lacking of investigation, and only a few studies have reported the polymer
composites possessing a thermal conductivity larger than 1 Wm-1K-1. Most of the studies are
based on epoxy composites. The highest reported k is 10.10 Wm-1K-1 with 5wt% highly
aligned SiC NWs,[232] which is benefited from the good alignment of the NWs and their high
aspect ratio. Overall, although the ceramic fillers are normally low-cost, to reach a 𝜅 in the
range of ~ 1 -- 5 Wm-1K-1 at room temperature normally requires a high-volume fraction (up
to 70 vol%).[243] Unfortunately, high filler content will not only seriously deteriorate the
mechanical properties of the composites,[244] but also deteriorate their dielectric properties.[204]
59
4.4. Metallic Fillers
Besides the carbon-based and ceramic-based fillers, metallic nanofillers are also appealing
candidates to enhance the heat transfer in polymer composites. They are often in forms of
NP[245] and NW[244, 246] with a high aspect ratio (AR, length over diameter). As
aforementioned, NWs are preferable for TIMs as they can form effective percolation network
at a low filler content.[191] The Au[247] or Ag NWs-based fillers are frequently adopted due to
their excellent thermal conductivity and resistance to surface oxidation, whereas their
industrial applications are limited due to the high cost. On the other hand, Cu NWs-based
fillers are attracting increasing attention in recent years majorly because of their much lower
cost compared to Au or Ag (while they also have a high thermal conductivity ~ 400 Wm-1K-
1). However, the drawback of Cu NWs is that they suffer from oxidation even in composites,
which will result in detrimental influence on the long-term performance of the composites.
Due to the high thermal conductivity of metallic NW filler, the heat transfer in the
polymer composites can be remarkably enhanced at a very low filler loading. For instance,
with a low volume fraction of 0.9% of Cu NWs (AR = 100 ~ 1000), polyacrylate (PA)
composite is reported to achieve a high k of 2.46 Wm-1K-1.[246b] Together with the high
thermal conductivity, metallic NWs also possess a high electrical conductivity, which usually
ruins the electrical insulating properties of the polymer composites. For example, the PA
composites with 0.9vol% of Cu NWs is measured with an electrical conductivity of 0.04 Scm-
1, which are no longer electrical insulting.[246b] To overcome this issue, different surface
modifications have been proposed for metallic NWs, e.g., grow a nano-coating of SiO2[248] or
polydopamine (PDA)[244] on the NW surface. For the Cu NW, the coating can not only be
used to resume the electrical insulating feature of the polymer, but also function as surface
protector for the NWs from oxidation.
60
Figure 23a shows the dependency of the thermal conductivity of polymer composites
on the loading of Cu NW fillers. In general, k increases when the filler loading increases,
which is commonly in a nonlinear fashion. The enhancement can be understood from the
perspective of the percolation threshold theory.[249] At a low filler content regime, the NWs
can barely form a connected network, which greatly hinder the electronic and phonon thermal
conduction and thus k exhibits a gradual increasing tendency with the filler content. When an
effective percolation network is formed at higher filler loading, the high thermal conductivity
of the NWs will significantly enhance the heat transfer, as is seen from the PA composite with
long Cu NWs in Figure 23a. Since long NWs are beneficial for the conductive network
formation, the composites exhibit a much high k reinforced by long NWs in comparison with
the counterpart reinforced by short NWs (under a same filler content).[246b] Experimental
results reveal that a proper coating can not only resume a high electrical resistivity, but also
enhance the phonon propagation across the NW/matrix interface and promote a further
enhance to the heat transfer in the composites.[244, 250] Specifically, the coating layer should
have a proper thickness to achieve the highest k, and too thick a coating layer will lead to
smaller k.[244] Different theoretical models have been adopted to interpret the relationship
between k and the filler volume fraction, e.g., the Maxwell model,[186] and the Agari
model.[251] It is found that the Nan model (that includes interfacial thermal resistances)[185] can
provide a reasonable description for the relationship between k and the filler content.
61
Figure 23. Thermal transport in polymer composites with metallic NW fillers. (a) The
changing tendency of the thermal conductivity at different Cu NW filler contents for PA
composite with short and long Cu NW (denoted as s-CuNW/PA and l-CuNW/PA,
respectively),[246b] and epoxy (EP) composite without and with 25 nm and 100 nm PDA
coated Cu NW (denoted as CuNW/EP, CuNW@PDA25/EP, and CuNW@PDA100/EP,
respectively).[244] (b) A summary of the thermal conductivity for metallic NWs reinforced
polymer composites, including EP composite (1 -- Ref. [244], 2 -- Ref. [250b], 3 -- Ref. [252], 4 --
Ref. [250a]), PDMS (5 -- Ref. [253], 6 -- Ref. [247], 7 -- Ref. [254]), polyacrylate (PA, 8 -- Ref. [246b]),
PU (9 -- Ref. [253]), EcoFlex (EF, 10 -- Ref. [255]), epoxy-terminated dimethylsiloxane (ETDS,
11 -- Ref. [248]), polyether ether ketone (PEEK, 12 -- Ref. [256]), and polyvinylidene fluoride
(PVDF, 13 -- Ref. [254]). The open markers represent the NWs with coatings.
Since the thermal conduction is dominant by electrons in the metallic NWs, the
thermal conductivity of the polymer composites normally decreases with increasing
temperature.[244, 246b, 255] Figure 23b summarizes the thermal conductivity of various polymer
composites reinforced by metallic NWs. It is seen that long NW fillers are preferable due to
the easy formation of effective conductive networks in the polymer composites.[252] The heat
transfer can be further enhanced by reducing the filler/filler resistance. For instance, through
microwave radiation treatment, the thermal conductivity of the EcoFlex (a platinum-catalyzed
addition-type silicone elastomer) composite with 1.8vol% Cu NWs can be significantly
improved from 0.78 Wm-1K-1 to 3.1 Wm-1K-1.[255] It is worthy to mention that besides NWs,
other metallic multiscale fillers have also been used to enhance the thermal transport in
polymer composites. For instance, for the EP composite with 3D microscale Ag flakes
(around 36.7% vol%), an addition of around 2.3 vol% of multi-wall CNTs (decorated with Ag
NPs) can results in a significant enhance to the thermal conductivity from 1.64 Wm-1K-1 to
62
160 Wm-1K-1.[257] Such remarkable enhancement is supposed as resulted from the efficient
phonon transport pathways between Ag flakes that are bridged by the CNT (with the aid of
Ag NPs).
Overall, current efforts have been mainly devoted to identify the relationship between
the thermal conductivity of the polymer composites and the filler loading. Some works also
discussed the temperature influence on the heat transfer. Comparing with the carbon-based
and hBN-based fillers, the metallic NW fillers are normally randomly dispersed in the
polymer matrix. Although the heat transfer in the polymer matrix with pre-prepared 3D
metallic networks are still unexplored, the challenges regarding the filler preparation, filler
characterization, and dispersion are still expected for the metallic nanofillers. Furthermore,
the combination of electron and phonon transport is expected to bring more uncertainties to
the bottlenecks that hinder the effective heat transfer, i.e., the interfacial thermal resistance
between the metallic filler and the polymer matrix.
5. Conclusion and Perspectives
Diverse 3D nanostructures have been fabricated recently to meet specific thermal
management purposes, ranging from nanoarchitectures to nanostructured fillers for polymers
and metals. Here, we have outlined the recent progress on the thermal conductivity of those
nanostructures with a focus on experimental results.
5.1. Low thermal conductivity
To pursue low thermal conductivity for thermal isolation application, different types of 3D
nanoarchitectures with a high porosity have been prepared based on low-dimensional
nanostructures. Recent years have witnessed a growing interest in preparing colloidal
assemblies to achieve low thermal conductivity, which can be fabricated from single or
multiple colloidal particles (mostly based on polymers). Studies have identified several
63
influential factors that affect the thermal transport of the colloidal assemblies, including
particle size, composition, and arrangement or packing. The intrinsic phase change
characteristic of the polymer can endow colloidal assemblies with a unique temperature
dependence, which enables emerging heat management applications in phase change
materials. More sophisticated core-shell particles have also been prepared to fabricate the
colloidal assemblies with improved mechanical performance while maintaining a low heat
transfer. The majority of current studies have focused on spherical particles, it is still
unknown how the particle shape would impact the thermal transport performance. In
considering the actual usage, significant efforts are still desired to correlate the thermal
transport properties of the colloidal assemblies with other application-specified properties
(e.g., mechanical properties or transparency) while tailoring their structures.
In addition to colloidal crystals, many other 3D nanostructures have also been
prepared for thermal isolation, such as nanocellular foams and nanolattice. The polymer-based
nanocellular foams have been extensively explored, however the investigation on the thermal
transport properties of the nanolattice, is still rare. Given the rapidly developing additive
manufacturing techniques, it is promising to design and fabricate nanolattice with desired
thermal transport properties together with excellent mechanical performance for thermal
isolation applications.
In terms of the thermoelectric devices, researchers have considered the array
assembles of nanowires, the inverse opals structures and holey nanostructures. The thermal
transport properties of the inverse opals structures are still lacking investigation. While,
considering the large variety of materials (e.g., metals and semiconductors) and structures
from the sacrificial opal template, there are great potentials to design novel heat management
materials with suppressed heat transfer for TE devices, or efficient heat transfer for thermal
storage and heat exchangers. The holey 3D Si nanostructures or Si phononic crystals have
64
been shown with a good control over the phonon propagation, which have attracted increasing
attention in recent years. The enhanced phonon boundary and surface scattering in the porous
structure are proposed to suppress the thermal conductivity. Relationships between the
geometrical factors and the thermal conductivity of the holey nanostructures have been
extensively investigated, such as the diameter and arrangement. However, although it is
widely accepted that the nanoscopic holes in the phononic crystals suppress the thermal
conductivity, the underlying mechanism is still under argument. Besides, using the phonon
coherence to tune the thermal transport in phononic crystals is still limited at low temperature.
However, it is believed that with the advancing of nanofabrication technology, the working
temperature range of phononic crystals can be further broadened like the way that wave optics
revolutionized the manipulations of light.
To summarize, the porous or hollow structures are main players for thermal insulation
applications, which are commonly prepared from materials with inherently low thermal
conductivity such as polymers and silica. The heat transfer in the hollow structures contains
the contribution from radiation, convection, gas phase conduction and solid phase conduction,
among which the gas phase conduction and solid phase conduction are the main contributors.
Therefore, suppress the contributions from the gas phase and solid phase are the main avenues
to obtain a low thermal conductivity. It has been shown that the gas phase conduction can be
significantly suppressed when the pore sizes in the hollow structures are on the order of mean
free path of gas molecules, in which circumstance the gas changes into Knudsen flow. Such
concept has been successfully implemented and proved in the 3D polymer-based foams or
nanocellular foams. On the other hand, phonons are the main heat carrier for solid phase
conduction, and the factors that influence the phonon transport are all useful means to
engineer the thermal conductivity of the porous structures. For instance, in the colloidal
65
crystals/assemblies, the interface phonon scattering, the distorted and elongated heat transfer
pathways, and the porosity are all effective ways to restrain heat transfer in the structure.
There are generally two strategies to attain high efficiency in TE devices, i.e., phonon
engineering to reduce the lattice thermal conductivity and the band engineering to boost the
electrical power factor. Nanostructuring has been shown as an effect avenue to suppress
phonon transport by enhancing phonon scattering (including Umklapp phonon-phonon
scattering, phonon-impurity scattering, phonon-boundary scattering) or introducing phonon
confinement effect. Recent years, great interests have been attracted on the manipulation of
the so-called wave-like coherent phonon transport, which can be realized through the presence
of a secondary periodicity in the nanostructure. The phonon coherence has already been
demonstrated to result in remarkably reduced thermal conductivity in phononic crystals,
which opens up a new and promising avenue to construct high-performance TE devices.[99]
5.2. High thermal conductivity
High thermal conductivity is desirable for the thermal interface materials (TIMs) and phase
change materials (for thermal energy storage), and recent years have witnessed a significant
research focus on TIMs. One promising TIM candidate is the CNT array, which have been
shown with excellent thermal conductivity. The emerging challenge is how to minimize the
thermal contact resistance between CNT arrays and the mating surface, and how to fabricate
well-aligned high-quality CNT arrays.
Plenty of in silico studies have been carried out on 3D nanoarchitectures (mainly
constructed from CNT or graphene), such as pillared structures and honeycomb, while the
experimental synthetization, characterization and measurement are still very challenging.
Several works reported the thermal conductivity for some of the hierarchical structures
(containing graphene/graphite and CNT), which however exhibit a large variance. Given the
66
vital role of the influence of the structure on the energy carrier, it is hard to compare the
results between different structures. Compared with the CNT arrays, these 3D
nanoarchitectures are still far away from real applications.
Another extensively explored TIM candidate is the polymer composites, and various
types of highly conductive nanostructures have been utilized as fillers to enhance their
thermal conductivity, such as nanowire and 2D nanosheet (ranging from carbon, ceramic, to
metallic nanomaterials). Compared with the intrinsically low thermal conductivity of the
polymer (~ 0.2 Wm-1K-1), remarkable enhancement has been reported. However, the thermal
conductivity for most of the reported polymer composites are still well below 5 Wm-1K-1. To
attain higher thermal conductivity, the filler content needs to be high, which inevitably brings
negative impacts on the mechanical performance or the electrical insulation characteristic of
the composites. Specifically, compared with the intrinsic thermal conductivity of the
nanofillers, there is still a significant gap between the effective thermal conductivity of the
composite and expectations (e.g., ~ 3000 for Wm-1K-1 CNTs). There is a tendency to use
multiple (two or more) fillers to reach a high thermal conductivity at a relatively lower filler
content utilizing the synergistic effect of the fillers. Again however, the improvement is still
far below expectation.
A wide range of factors have been shown to significantly affect the heat transfer in the
polymer composites, such as the filler content, filler/matrix interface, filler/filler interface,
filler geometry and size, filler alignment and orientation, and the fabrication/processing
techniques. It is widely accepted that a percolation network of fillers is desirable in the
polymer matrix to achieve effective heat transfer. The dispersion of filler in a polymer matrix
plays a crucial role for the heat transfer in the polymer. Similar as the CNT arrays, the
bottleneck to obtain highly conductive polymer composites is to minimize the thermal
interfacial resistance within the matrix.
67
There is also increasing interest in adding nanostructured fillers to metal matrix in the
past decade, in the purpose of reducing the coefficient of thermal expansion and enhancing
their thermal conductivity. According to present studies, the coefficient of thermal expansion
can be reduced by the nanofillers, which however commonly bring negative influence to the
heat transfer in metal matrix. Currently, it is challenge to prepare metal matrix composites
with desired nanofillers and to identify the bonds at the interface. As such great efforts are
still expected in the future to explore the influential factors on the heat transfer in the metal
matrix, such as temperature, filler shape and geometry, and filler dispersion or network.
Overall, there is an increasingly growing demand for high thermal conductivity
materials due to the rapidly advancing electronic industry, especially the thermal interface
materials for electronic packing. Majority works have been focused on CNT-based or polymer
composite-based TIMs, where the phonon is the dominant heat carrier. Therefore, a key
feature for TIMs is having efficient heat transfer channel to spread undesired heat. CNT
arrays are good example with excellent thermal conductivity. For polymer composites, adding
highly thermal conductive fillers has been shown as an effective way to promote the heat
transfer. However, it is urgently needed to find out how to minimize the thermal interfacial
resistance that introduces strong phonon scattering.
To conclude, the use of nanostructuring of materials is an effective strategy to modify
the thermal conductivity of materials (either by phonon or electron transport). Studies have
shown that a broad range of factors, such as structure (or morphology), geometrical size, and
structural defects, will affect the heat carrier. To facilitate engineering applications, it is vital
to explore and develop cost-effective and efficient approaches to prepare high-quality
nanostructures. In addition, it is still an open question whether these
nanostructures/nanomaterials can retain their efficient heat transfer capability under repetitive
thermal or mechanical loadings during service, which is an important area for future research.
68
Acknowledgements
H.Z., J.B. and Y.G. would like to acknowledge the supports by the ARC Discovery Project
(DP170102861). Y.C. would like to acknowledge the financial support by Key projects of
Shaanxi Natural Science Foundation (2019JZ-27), Shaanxi Natural Science Basic Research
Program-Shaanxi Coal (2019JLM-47) and Fundamental Research Funds for the Central
Universities CHD (300102319304).
Received: ((will be filled in by the editorial staff))
Revised: ((will be filled in by the editorial staff))
Published online: ((will be filled in by the editorial staff))
References
[1]
a) A. A. Balandin, Nat Mater. 2011, 10, 569; b) Y. Xu, Z. Li, W. Duan, Small 2014,
10, 2182.
[2]
T. Li, J. Song, X. Zhao, Z. Yang, G. Pastel, S. Xu, C. Jia, J. Dai, C. Chen, A. Gong, F.
Jiang, Y. Yao, T. Fan, B. Yang, L. Wågberg, R. Yang, L. Hu, Sci Adv. 2018, 4,
eaar3724.
[3]
a) L. Hu, T. Zhu, X. Liu, X. Zhao, Adv. Funct. Mater. 2014, 24, 5211; b) L. Hu, H.
Wu, T. Zhu, C. Fu, J. He, P. Ying, X. Zhao, Adv. Energy Mater. 2015, 5, 1500411; c)
Q. Zhang, X. Ai, L. Wang, Y. Chang, W. Luo, W. Jiang, L. Chen, Adv. Funct. Mater.
2015, 25, 966; d) B. Wölfing, C. Kloc, J. Teubner, E. Bucher, Phys. Rev. Lett. 2001,
86, 4350.
[4]
D.-Y. Chung, T. Hogan, P. Brazis, M. Rocci-Lane, C. Kannewurf, M. Bastea, C. Uher,
M. G. Kanatzidis, Science 2000, 287, 1024.
[5]
a) L. Yang, Z.-G. Chen, M. S. Dargusch, J. Zou, Adv. Energy Mater. 2018, 8,
1701797; b) M. Hong, J. Zou, Z. G. Chen, Adv. Mater. 2019, 31, 1807071.
[6]
[7]
X. Wang, X. Gan, Chin. Phys. B 2017, 26, 034203.
a) C. Chen, Y. Xue, Z. Li, Y. Wen, X. Li, F. Wu, X. Li, D. Shi, Z. Xue, X. Xie, Chem.
Eng. J. 2019, 369, 1150; b) K. M. Hoogeboom-Pot, J. N. Hernandez-Charpak, X. Gu,
69
T. D. Frazer, E. H. Anderson, W. Chao, R. W. Falcone, R. Yang, M. M. Murnane, H.
C. Kapteyn, D. Nardi, Proc. Natl. Acad. Sci. U.S.A. 2015, 112, 4846; c) Y. Chen, S.
Liu, Y. Zhao, Q. Liu, L. Zhu, X. Song, Y. Zhang, J. Hao, Vacuum 2017, 143, 150.
[8]
S. Maneewan, J. Hirunlabh, J. Khedari, B. Zeghmati, S. Teekasap, Solar Energy 2005,
78, 495.
[9]
Z. Liu, L. Zhang, G. Gong, H. Li, G. Tang, Energy Build. 2015, 102, 207.
[10] L. Aditya, T. Mahlia, B. Rismanchi, H. Ng, M. Hasan, H. Metselaar, O. Muraza, H.
Aditiya, Renew. Sust. Energ. Rev. 2017, 73, 1352.
[11] O. Bubnova, X. Crispin, Energy Environ. Sci. 2012, 5, 9345.
[12] Y. Wang, Y. Shi, D. Mei, Z. Chen, Appl. Energy 2018, 215, 690.
[13] V. Leonov, R. J. M. Vullers, J. Renew. Sustain. Energy 2009, 1, 062701.
[14] R. He, G. Schierning, K. Nielsch, Adv. Mater. Technol. 2018, 3, 1700256.
[15] L. Carroll, J.-S. Lee, C. Scarcella, K. Gradkowski, M. Duperron, H. Lu, Y. Zhao, C.
Eason, P. Morrissey, M. Rensing, S. Collins, H. Y. Hwang, P. O'Brien, Appl. Sci.
2016, 6, 426.
[16] K. M. F. Shahil, A. A. Balandin, Solid State Commun. 2012, 152, 1331.
[17] N. Vogel, M. Retsch, C.-A. Fustin, A. del Campo, U. Jonas, Chem. Rev. 2015, 115,
6265.
[18] P. Ruckdeschel, A. Philipp, M. Retsch, Adv. Funct. Mater. 2017, 27, 1702256.
[19] M. T. Barako, A. Sood, C. Zhang, J. Wang, T. Kodama, M. Asheghi, X. Zheng, P. V.
Braun, K. E. Goodson, Nano Lett. 2016, 16, 2754.
[20] L. Ping, P.-X. Hou, C. Liu, H.-M. Cheng, APL Mater. 2019, 7, 020902.
[21] M. T. Barako, S. Roy-Panzer, T. S. English, T. Kodama, M. Asheghi, T. W. Kenny,
K. E. Goodson, ACS Appl. Mater. Interfaces 2015, 7, 19251.
[22] G. K. Dimitrakakis, E. Tylianakis, G. E. Froudakis, Nano Lett. 2008, 8, 3166.
70
[23]
J. Bauer, A. Schroer, R. Schwaiger, O. Kraft, Nat Mater. 2016, 15, 438.
[24] N. V. Krainyukova, E. N. Zubarev, Phys. Rev. Lett. 2016, 116, 055501.
[25]
J. Lim, H.-T. Wang, J. Tang, S. C. Andrews, H. So, J. Lee, D. H. Lee, T. P. Russell, P.
Yang, ACS Nano 2016, 10, 124.
[26] X. Chen, J. Tao, Y. Liu, R. Bao, F. Li, C. Li, J. Yi, Carbon 2019, 146, 736.
[27] S. Zhao, Z. Zheng, Z. Huang, S. Dong, P. Luo, Z. Zhang, Y. Wang, Mater. Sci. Eng.,
A 2016, 675, 82.
[28] X. Wang, P. Wu, Chem. Eng. J. 2018, 348, 723.
[29] Y. Xue, P. Dai, M. Zhou, X. Wang, A. Pakdel, C. Zhang, Q. Weng, T. Takei, X. Fu, Z.
I. Popov, P. B. Sorokin, C. Tang, K. Shimamura, Y. Bando, D. Golberg, ACS Nano
2017, 11, 558.
[30] K. Wu, C. Lei, R. Huang, W. Yang, S. Chai, C. Geng, F. Chen, Q. Fu, ACS Appl.
Mater. Interfaces 2017, 9, 7637.
[31]
J. Hu, Y. Huang, Y. Yao, G. Pan, J. Sun, X. Zeng, R. Sun, J.-B. Xu, B. Song, C.-P.
Wong, ACS Appl. Mater. Interfaces 2017, 9, 13544.
[32] X. Zeng, Y. Yao, Z. Gong, F. Wang, R. Sun, J. Xu, C.-P. Wong, Small 2015, 11,
6205.
[33] N. Yang, X. Xu, G. Zhang, B. Li, AIP Adv. 2012, 2, 041410.
[34]
a) H. F. Zhan, Y. Y. Zhang, J. M. Bell, Y. T. Gu, J. Phys. D: Appl. Phys. 2014, 47,
015303; b) J. Chen, G. Zhang, B. Li, Nano Lett. 2012, 12, 2826; c) G.-J. Hu, B.-Y.
Cao, Chin. Phys. B 2014, 23, 096501; d) Y. Feng, J. Zhu, D.-W. Tang, Chin. Phys. B
2014, 23, 083101; e) Z.-X. Guo, D. Zhang, X.-G. Gong, Phys. Rev. B 2011, 84,
075470; f) X. Chen, Y. Xu, X. Zou, B.-L. Gu, W. Duan, Phys. Rev. B 2013, 87,
155438; g) X. Liu, G. Zhang, Y.-W. Zhang, Nano Lett. 2016, 16, 4954; h) G. Zhang,
B. Li, J. Chem. Phys. 2005, 123, 114714; i) J. Chen, G. Zhang, B. Li, Nano Lett. 2010,
71
10, 3978; j) H. Zhan, G. Zhang, V. B. Tan, Y. Cheng, J. M. Bell, Y. W. Zhang, Y. Gu,
Adv. Funct. Mater. 2016, 26, 5279.
[35]
a) Z. Gang, Z. Yong-Wei, Chin. Phys. B 2017, 26, 034401; b) Z.-X. Guo, J. W. Ding,
X.-G. Gong, Phys. Rev. B 2012, 85, 235429; c) H.-Y. Cao, Z.-X. Guo, H. Xiang, X.-
G. Gong, Phys. Lett. A 2012, 376, 525; d) H.-S. Zhang, Z.-X. Guo, X.-G. Gong, J.-X.
Cao, J. Appl. Phys. 2012, 112, 123508; e) H.-Y. Cao, H. Xiang, X.-G. Gong, Solid
State Commun. 2012, 152, 1807; f) Z. Guo, D. Zhang, X.-G. Gong, Appl. Phys. Lett.
2009, 95, 163103; g) A. A. Balandin, S. Ghosh, W. Bao, I. Calizo, D. Teweldebrhan,
F. Miao, C. N. Lau, Nano Lett. 2008, 8, 902; h) W. Xu, L. Zhu, Y. Cai, G. Zhang, B.
Li, J. Appl. Phys. 2015, 117, 214308; i) Y. Cai, Q. Ke, G. Zhang, Y. P. Feng, V. B.
Shenoy, Y. W. Zhang, Adv. Funct. Mater. 2015, 25, 2230.
[36] B. P. Jelle, Energy Build. 2011, 43, 2549.
[37]
J. Zhu, M. Li, R. Rogers, W. Meyer, R. Ottewill, W. Russel, P. Chaikin, Nature 1997,
387, 883.
[38] L. Cui, J. Zhang, X. Zhang, L. Huang, Z. Wang, Y. Li, H. Gao, S. Zhu, T. Wang, B.
Yang, ACS Appl. Mater. Interfaces 2012, 4, 2775.
[39] R. Amos, J. Rarity, P. Tapster, T. Shepherd, S. Kitson, Phys. Rev. E 2000, 61, 2929.
[40] O. Velev, T. Jede, R. Lobo, A. Lenhoff, Nature 1997, 389, 447.
[41] P. Ruckdeschel, M. Retsch, Adv. Mater. Interfaces 2017, 4, 1700963.
[42] Y. Xia, B. Gates, Y. Yin, Y. Lu, Adv. Mater. 2000, 12, 693.
[43] R. V. Rivera Virtudazo, Y. Lin, R. T. Wu, RSC Adv. 2015, 5, 104408.
[44] F. A. Nutz, P. Ruckdeschel, M. Retsch, J. Colloid Interface Sci. 2015, 457, 96.
[45] P. Ruckdeschel, A. Philipp, B. A. F. Kopera, F. Bitterlich, M. Dulle, N. W. Pech-May,
M. Retsch, Phys. Rev. E 2018, 97, 022612.
[46] F. A. Nutz, M. Retsch, Sci Adv. 2017, 3, eaao5238.
72
[47] F. A. Nutz, M. Retsch, Phys. Chem. Chem. Phys. 2017, 19, 16124.
[48] F. A. Nutz, A. Philipp, B. A. Kopera, M. Dulle, M. Retsch, Adv. Mater. 2018, 30,
1704910.
[49] P. Ruckdeschel, T. W. Kemnitzer, F. A. Nutz, J. Senker, M. Retsch, Nanoscale 2015,
7, 10059.
[50]
J. Yin, M. Retsch, E. L. Thomas, M. C. Boyce, Langmuir 2012, 28, 5580.
[51]
a) S. Liu, J. Duvigneau, G. J. Vancso, Eur. Polym. J. 2015, 65, 33; b) C. Forest, P.
Chaumont, P. Cassagnau, B. Swoboda, P. Sonntag, Prog. Polym. Sci. 2015, 41, 122;
c) G. Wang, C. Wang, J. Zhao, G. Wang, C. B. Park, G. Zhao, Nanoscale 2017, 9,
5996.
[52] F. Hu, S. Wu, Y. Sun, Adv. Mater. 2018, 0, 1801001.
[53] S. N. Schiffres, K. H. Kim, L. Hu, A. J. McGaughey, M. F. Islam, J. A. Malen, Adv.
Funct. Mater. 2012, 22, 5251.
[54] C. V. Vo, F. Bunge, J. Duffy, L. Hood, Cell. Polym. 2011, 30, 137.
[55] B. Notario, J. Pinto, E. Solorzano, J. A. De Saja, M. Dumon, M. Rodríguez-Pérez,
Polymer 2015, 56, 57.
[56] A. Rizvi, R. K. M. Chu, C. B. Park, ACS Appl. Mater. Interfaces 2018, 10, 38410.
[57]
a) L. C. Montemayor, L. R. Meza, J. R. Greer, Adv. Eng. Mater. 2014, 16, 184; b) M.
Kaur, S. M. Han, W. S. Kim, MRS Commun. 2017, 7, 8.
[58] 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. Spadaccini,
Science 2014, 344, 1373.
[59]
J. Bauer, L. R. Meza, T. A. Schaedler, R. Schwaiger, X. Zheng, L. Valdevit, Adv.
Mater. 2017, 29, 1701850.
73
[60] X. Zhang, A. Vyatskikh, H. Gao, J. R. Greer, X. Li, Proc. Natl. Acad. Sci. U.S.A.
2019, 116, 6665.
[61]
a) L. R. Meza, S. Das, J. R. Greer, Science 2014, 345, 1322; b) A. J. Mateos, W.
Huang, Y.-W. Zhang, J. R. Greer, Adv. Funct. Mater. 2019, 29, 1806772.
[62]
a) A. Vyatskikh, S. Delalande, A. Kudo, X. Zhang, C. M. Portela, J. R. Greer, Nat.
Commun. 2018, 9, 593; b) L. Gao, J. Song, Z. Jiao, W. Liao, J. Luan, J. U. Surjadi, J.
Li, H. Zhang, D. Sun, C. T. Liu, Y. Lu, Adv. Eng. Mater. 2018, 20, 1700625.
[63] R. Liontas, J. R. Greer, Acta Mater. 2017, 133, 393.
[64]
a) Z. Liu, P. Liu, W. Huang, W. H. Wong, A. L. Commillus, Y.-W. Zhang, Mater.
Des. 2018, 160, 496; b) Z. D. Sha, C. M. She, G. K. Xu, Q. X. Pei, Z. S. Liu, T. J.
Wang, H. J. Gao, Mater. Today 2017, 20, 569; c) M. L. Lifson, M.-W. Kim, J. R.
Greer, B.-J. Kim, Nano Lett. 2017, 17, 7737; d) B. B. Lewis, B. A. Mound, B.
Srijanto, J. D. Fowlkes, G. M. Pharr, P. D. Rack, Nanoscale 2017, 9, 16349.
[65] S.-W. Lee, M. Jafary-Zadeh, D. Z. Chen, Y.-W. Zhang, J. R. Greer, Nano Lett. 2015,
15, 5673.
[66] N. G. Dou, R. A. Jagt, C. M. Portela, J. R. Greer, A. J. Minnich, Nano Lett. 2018, 18,
4755.
[67] N. G. Dou, A. J. Minnich, Appl. Phys. Lett. 2016, 108, 011902.
[68] M. F. Ashby, Philos. Trans. Royal Soc. A 2006, 364, 15.
[69] S. M. Lee, D. G. Cahill, T. H. Allen, Phys. Rev. B 1995, 52, 253.
[70]
a) M. Hong, Y. Wang, W. Liu, S. Matsumura, H. Wang, J. Zou, Z.-G. Chen, Adv.
Energy Mater. 2018, 8, 1801837; b) M. Hong, Z.-G. Chen, L. Yang, Y.-C. Zou, M. S.
Dargusch, H. Wang, J. Zou, Adv. Mater. 2018, 30, 1705942; c) M. Hong, Z.-G. Chen,
L. Yang, Z.-M. Liao, Y.-C. Zou, Y.-H. Chen, S. Matsumura, J. Zou, Adv. Energy
Mater. 2018, 8, 1702333.
74
[71]
a) X. Shi, A. Wu, T. Feng, K. Zheng, W. Liu, Q. Sun, M. Hong, S. T. Pantelides, Z.-
G. Chen, J. Zou, Adv. Energy Mater. 2019, 9, 1803242; b) X.-L. Shi, K. Zheng, W.-D.
Liu, Y. Wang, Y.-Z. Yang, Z.-G. Chen, J. Zou, Adv. Energy Mater. 2018, 8, 1800775.
[72]
a) A. Stranz, A. Waag, E. Peiner, J. Mater. Res. 2011, 26, 1958; b) A. I. Hochbaum,
R. Chen, R. D. Delgado, W. Liang, E. C. Garnett, M. Najarian, A. Majumdar, P. Yang,
Nature 2008, 451, 163; c) G. Schierning, physica status solidi (a) 2014, 211, 1235; d)
E. Krali, Z. A. K. Durrani, Appl. Phys. Lett. 2013, 102, 143102.
[73] C. Yan, T. Yu, C. Ji, X. Zeng, J. Lu, R. Sun, C.-P. Wong, J. Appl. Polym. Sci. 2019,
136, 47054.
[74]
a) A. Kikuchi, A. Yao, I. Mori, T. Ono, S. Samukawa, Appl. Phys. Lett. 2017, 110,
091908; b) A. Kikuchi, A. Yao, I. Mori, T. Ono, S. Samukawa, J. Appl. Phys. 2017,
122, 165302.
[75] A. I. Boukai, Y. Bunimovich, J. Tahir-Kheli, J.-K. Yu, W. A. Goddard Iii, J. R. Heath,
Nature 2008, 451, 168.
[76]
a) B. M. Curtin, E. W. Fang, J. E. Bowers, J. Electron. Mater. 2012, 41, 887; b) B.
Xu, W. Khouri, K. Fobelets, IEEE Electron Device Letters 2014, 35, 596; c) J. M.
Weisse, A. M. Marconnet, D. R. Kim, P. M. Rao, M. A. Panzer, K. E. Goodson, X.
Zheng, Nanoscale Res. Lett. 2012, 7, 554; d) Y. Pan, G. Hong, S. N. Raja, S.
Zimmermann, M. K. Tiwari, D. Poulikakos, Appl. Phys. Lett. 2015, 106, 093102.
[77] T. Zhang, S. Wu, J. Xu, R. Zheng, G. Cheng, Nano Energy 2015, 13, 433.
[78] T. Zhang, S.-l. Wu, R.-t. Zheng, G.-a. Cheng, Nanotechnology 2013, 24, 505718.
[79] H. Zhan, G. Zhang, Y. Zhang, V. B. C. Tan, J. M. Bell, Y. Gu, Carbon 2016, 98, 232.
[80] X. Liu, G. Zhang, Q.-X. Pei, Y.-W. Zhang, Appl. Phys. Lett. 2013, 103, 133113.
[81] P. G. Klemens, Int. J. Thermophys. 2001, 22, 265.
75
[82] D. Nika, A. Cocemasov, C. Isacova, A. Balandin, V. Fomin, O. Schmidt, Phys. Rev. B
2012, 85, 205439.
[83]
a) D. Dávila, A. Tarancón, C. Calaza, M. Salleras, M. Fernández-Regúlez, A. San
Paulo, L. Fonseca, Nano Energy 2012, 1, 812; b) C. Calaza, M. Salleras, D. Dávila, A.
Tarancón, A. Morata, J. D. Santos, G. Gadea, L. Fonseca, Materials Today:
Proceedings 2015, 2, 675.
[84]
J. Lee, W. Lee, J. Lim, Y. Yu, Q. Kong, J. J. Urban, P. Yang, Nano Lett. 2016, 16,
4133.
[85] A. Stein, B. E. Wilson, S. G. Rudisill, Chem. Soc. Rev. 2013, 42, 2763.
[86]
J. Ma, B. R. Parajuli, M. G. Ghossoub, A. Mihi, J. Sadhu, P. V. Braun, S. Sinha, Nano
Lett. 2013, 13, 618.
[87] A. Blanco, E. Chomski, S. Grabtchak, M. Ibisate, S. John, S. W. Leonard, C. Lopez,
F. Meseguer, H. Miguez, J. P. Mondia, Nature 2000, 405, 437.
[88] Z. Wang, J. E. Alaniz, W. Jang, J. E. Garay, C. Dames, Nano Lett. 2011, 11, 2206.
[89] A. Jain, A. J. McGaughey, Phys. Rev. B 2016, 93, 081206.
[90] D. Hung, Z. Liu, N. Shah, Y. Hao, P. C. Searson, J. Phys. Chem. C 2007, 111, 3308.
[91] M. T. Barako, S. Lingamneni, J. S. Katz, T. Liu, K. E. Goodson, J. Tice, J. Appl. Phys.
2018, 124, 145103.
[92]
J. Tang, H.-T. Wang, D. H. Lee, M. Fardy, Z. Huo, T. P. Russell, P. Yang, Nano Lett.
2010, 10, 4279.
[93]
J.-K. Yu, S. Mitrovic, D. Tham, J. Varghese, J. R. Heath, Nat. Nanotechnol. 2010, 5,
718.
[94]
a) Z. Ren, J. Lee, Nanotechnology 2017, 29, 045404; b) J. Lee, J. Lim, P. Yang, Nano
Lett. 2015, 15, 3273.
76
[95] P. E. Hopkins, C. M. Reinke, M. F. Su, R. H. Olsson, E. A. Shaner, Z. C. Leseman, J.
R. Serrano, L. M. Phinney, I. El-Kady, Nano Lett. 2011, 11, 107.
[96] S. Alaie, D. F. Goettler, M. Su, Z. C. Leseman, C. M. Reinke, I. El-Kady, Nat.
Commun. 2015, 6, 7228.
[97] R. Prasher, T. Tong, A. Majumdar, Nano Lett. 2008, 8, 99.
[98]
a) F. Kargar, B. Debnath, J.-P. Kakko, A. Säynätjoki, H. Lipsanen, D. L. Nika, R. K.
Lake, A. A. Balandin, Nat. Commun. 2016, 7, 13400; b) F. Kargar, S. Ramirez, B.
Debnath, H. Malekpour, R. K. Lake, A. A. Balandin, Appl. Phys. Lett. 2015, 107,
171904.
[99] G. Xie, D. Ding, G. Zhang, Advances in Physics: X 2018, 3, 1480417.
[100] a) J. Maire, R. Anufriev, R. Yanagisawa, A. Ramiere, S. Volz, M. Nomura, Sci Adv.
2017, 3, e1700027; b) G. Xie, Z. Ju, K. Zhou, X. Wei, Z. Guo, Y. Cai, G. Zhang, Npj
Comput. Mater. 2018, 4, 21; c) M. R. Wagner, B. Graczykowski, J. S. Reparaz, A. El
Sachat, M. Sledzinska, F. Alzina, C. M. Sotomayor Torres, Nano Lett. 2016, 16, 5661;
d) J. Lee, W. Lee, G. Wehmeyer, S. Dhuey, D. L. Olynick, S. Cabrini, C. Dames, J. J.
Urban, P. Yang, Nat. Commun. 2017, 8, 14054.
[101] A. M. Marconnet, M. A. Panzer, K. E. Goodson, Rev. Mod. Phys. 2013, 85, 1295.
[102] X. J. Hu, A. A. Padilla, J. Xu, T. S. Fisher, K. E. Goodson, J. Heat Transfer 2005,
128, 1109.
[103] D. J. Yang, S. G. Wang, Q. Zhang, P. J. Sellin, G. Chen, Phys. Lett. A 2004, 329, 207.
[104] T. Tong, Y. Zhao, L. Delzeit, A. Kashani, M. Meyyappan, A. Majumdar, IEEE Trans.
Compon. Packag. Technol. 2007, 30, 92.
[105] S. K. Pal, Y. Son, T. Borca-Tasciuc, D.-A. Borca-Tasciuc, S. Kar, R. Vajtai, P. M.
Ajayan, J. Mater. Res. 2008, 23, 2099.
77
[106] Y. Son, S. K. Pal, T. Borca-Tasciuc, P. M. Ajayan, R. W. Siegel, J. Appl. Phys. 2008,
103, 024911.
[107] X. Wang, Z. Zhong, J. Xu, J. Appl. Phys. 2005, 97, 064302.
[108] I. Ivanov, A. Puretzky, G. Eres, H. Wang, Z. Pan, H. Cui, R. Jin, J. Howe, D. B.
Geohegan, Appl. Phys. Lett. 2006, 89, 223110.
[109] H. Xie, A. Cai, X. Wang, Phys. Lett. A 2007, 369, 120.
[110] A. Okamoto, I. Gunjishima, T. Inoue, M. Akoshima, H. Miyagawa, T. Nakano, T.
Baba, M. Tanemura, G. Oomi, Carbon 2011, 49, 294.
[111] M. B. Jakubinek, M. A. White, G. Li, C. Jayasinghe, W. Cho, M. J. Schulz, V.
Shanov, Carbon 2010, 48, 3947.
[112] Y. Xu, Y. Zhang, E. Suhir, X. Wang, J. Appl. Phys. 2006, 100, 074302.
[113] W. Lin, J. Shang, W. Gu, C. P. Wong, Carbon 2012, 50, 1591.
[114] C. Silvestri, M. Riccio, R. H. Poelma, A. Jovic, B. Morana, S. Vollebregt, A. Irace, G.
Q. Zhang, P. M. Sarro, Small 2018, 14, 1800614.
[115] L. Qiu, K. Scheider, S. A. Radwan, L. S. Larkin, C. B. Saltonstall, Y. Feng, X. Zhang,
P. M. Norris, Carbon 2017, 120, 128.
[116] L. Ping, P.-X. Hou, C. Liu, J. Li, Y. Zhao, F. Zhang, C. Ma, K. Tai, H. Cong, H.-M.
Cheng, Nanoscale 2017, 9, 8213.
[117] C. Silvestri, M. Riccio, R. H. Poelma, B. Morana, S. Vollebregt, F. Santagata, A.
Irace, G. Q. Zhang, P. M. Sarro, Nanoscale 2016, 8, 8266.
[118] M. A. Panzer, G. Zhang, D. Mann, X. Hu, E. Pop, H. Dai, K. E. Goodson, J. Heat
Transfer 2008, 130, 052401.
[119] M. Akoshima, K. Hata, D. N. Futaba, K. Mizuno, T. Baba, M. Yumura, Jpn. J. Appl.
Phys. 2009, 48, 05EC07.
78
[120] G. Eres, C. M. Rouleau, A. A. Puretzky, D. B. Geohegan, H. Wang, Phys. Rev. Appl.
2018, 10, 024010.
[121] P. Kim, L. Shi, A. Majumdar, P. McEuen, Phys. Rev. Lett. 2001, 87, 215502.
[122] a) Y. Yao, J. N. Tey, Z. Li, J. Wei, K. Bennett, A. McNamara, Y. Joshi, R. L. S. Tan,
S. N. M. Ling, C. Wong, IEEE Transactions on Components, Packaging and
Manufacturing Technology 2014, 4, 232; b) M. A. Peacock, C. K. Roy, M. C.
Hamilton, R. Wayne Johnson, R. W. Knight, D. K. Harris, Int. J. Heat Mass Transfer
2016, 97, 94.
[123] B. Kumanek, D. Janas, J. Mater. Sci. 2019, 54, 7397.
[124] F. D. Novaes, R. Rurali, P. Ordejón, ACS Nano 2010, 4, 7596.
[125] a) C.-H. Wang, T.-H. Fang, W.-L. Sun, J. Phys. D: Appl. Phys. 2014, 47, 405302; b)
K. Xia, H. Zhan, Y. Wei, Y. Gu, Beilstein J. Nanotechnol. 2014, 5, 329; c) K. Xia, H.
Zhan, Y. Gu, Procedia IUTAM 2017, 21, 94.
[126] a) R. K. Paul, M. Ghazinejad, M. Penchev, J. Lin, M. Ozkan, C. S. Ozkan, Small
2010, 6, 2309; b) Y. Zhu, L. Li, C. Zhang, G. Casillas, Z. Sun, Z. Yan, G. Ruan, Z.
Peng, A.-R. O. Raji, C. Kittrell, Nat. Commun. 2012, 3, 1225; c) W. Zhang, H. Xie, R.
Zhang, M. Jian, C. Wang, Q. Zheng, F. Wei, Y. Zhang, Carbon 2015, 86, 358; d) C.
Tang, Q. Zhang, M.-Q. Zhao, J.-Q. Huang, X.-B. Cheng, G.-L. Tian, H.-J. Peng, F.
Wei, Adv. Mater. 2014, 26, 6100.
[127] S. Sun, M. K. Samani, Y. Fu, T. Xu, L. Ye, M. Satwara, K. Jeppson, T. Nilsson, L.
Sun, J. Liu, Adv. Mater. Interfaces 2018, 5, 1800318.
[128] H. F. Zhan, Y. T. Gu, Chin. Phys. B 2018, 27, 38103.
[129] N. Sakhavand, R. Shahsavari, ACS Appl. Mater. Interfaces 2017, 9, 39122.
79
[130] a) J. Chen, J. H. Walther, P. Koumoutsakos, Adv. Funct. Mater. 2015, 25, 7539; b) V.
Varshney, S. S. Patnaik, A. K. Roy, G. Froudakis, B. L. Farmer, ACS Nano 2010, 4,
1153; c) J. Park, V. Prakash, J. Mater. Res. 2013, 28, 940.
[131] C. Tang, Q. Zhang, M.-Q. Zhao, G.-L. Tian, F. Wei, Nano Energy 2014, 7, 161.
[132] M. Qin, Y. Feng, T. Ji, W. Feng, Carbon 2016, 104, 157.
[133] W. Feng, M. Qin, P. Lv, J. Li, Y. Feng, Carbon 2014, 77, 1054.
[134] H. Lu, J. Zhang, J. Luo, W. Gong, C. Li, Q. Li, K. Zhang, M. Hu, Y. Yao, Compos.
Part A Appl. Sci. Manuf. 2017, 102, 1.
[135] Q. Kong, L. Qiu, Y. D. Lim, C. W. Tan, K. Liang, C. Lu, B. K. Tay, Surf. Coat.
Technol. 2018, 345, 105.
[136] I. Kholmanov, J. Kim, E. Ou, R. S. Ruoff, L. Shi, ACS nano 2015, 9, 11699.
[137] a) M. Loeblein, S. H. Tsang, M. Pawlik, E. J. R. Phua, H. Yong, X. W. Zhang, C. L.
Gan, E. H. T. Teo, ACS Nano 2017, 11, 2033; b) M. T. Pettes, H. Ji, R. S. Ruoff, L.
Shi, Nano Lett. 2012, 12, 2959.
[138] H. R. Karfunkel, T. Dressler, JACS 1992, 114, 2285.
[139] a) Z. Pang, X. Gu, Y. Wei, R. Yang, M. S. Dresselhaus, Nano Lett. 2017, 17, 179; b)
S. Wang, D. Wu, B. Yang, E. Ruckenstein, H. Chen, Nanoscale 2018, 10, 2748.
[140] a) X. Gu, Z. Pang, Y. Wei, R. Yang, Carbon 2017, 119, 278; b) Z. Zhang, A. Kutana,
Y. Yang, N. V. Krainyukova, E. S. Penev, B. I. Yakobson, Carbon 2017, 113, 26.
[141] Y. Gao, Y. Chen, C. Zhong, Z. Zhang, Y. Xie, S. Zhang, Nanoscale 2016, 8, 12863.
[142] a) X.-K. Chen, J. Liu, D. Du, Z.-X. Xie, K.-Q. Chen, J. Phys.: Condens. Matter 2018,
30, 155702; b) Y. Han, J.-Y. Yang, M. Hu, Nanoscale 2018, 10, 5229; c) J. Zhang,
Carbon 2018, 131, 127.
[143] J. Zhang, Nano Energy 2017, 41, 460.
[144] H. Wu, L. T. Drzal, Carbon 2012, 50, 1135.
80
[145] S. C. Tjong, Mater. Sci. Eng. R Rep. 2013, 74, 281.
[146] Alamusi, N. Hu, B. Jia, M. Arai, C. Yan, J. Li, Y. Liu, S. Atobe, H. Fukunaga,
Comput. Mater. Sci. 2012, 54, 249.
[147] A. Miranda, N. Barekar, B. J. McKay, J. Alloys Compd. 2019, 774, 820.
[148] a) M. Sundaram Rajyashree, A. Sekiguchi, M. Sekiya, T. Yamada, K. Hata, Royal
Soc. Open Sci. 2018, 5, 180814; b) W. A. D. M. Jayathilaka, A. Chinnappan, S.
Ramakrishna, J. Mater. Chem. C 2017, 5, 9209; c) K. S. Munir, P. Kingshott, C. Wen,
Crit. Rev. Solid State Mater. Sci. 2015, 40, 38; d) T. Wejrzanowski, M. Grybczuk, M.
Chmielewski, K. Pietrzak, K. J. Kurzydlowski, A. Strojny-Nedza, Mater. Des. 2016,
99, 163.
[149] K. T. Kim, J. Eckert, G. Liu, J. M. Park, B. K. Lim, S. H. Hong, Scripta Mater. 2011,
64, 181.
[150] C. Subramaniam, Y. Yasuda, S. Takeya, S. Ata, A. Nishizawa, D. Futaba, T. Yamada,
K. Hata, Nanoscale 2014, 6, 2669.
[151] a) R. Vignesh Babu, S. Kanagaraj, J. Mater. Process. Technol. 2018, 258, 296; b) S.
Cho, K. Kikuchi, T. Miyazaki, K. Takagi, A. Kawasaki, T. Tsukada, Scripta Mater.
2010, 63, 375; c) P. G. Koppad, H. R. A. Ram, C. S. Ramesh, K. T. Kashyap, R. G.
Koppad, J. Alloys Compd. 2013, 580, 527.
[152] S. Cho, K. Kikuchi, A. Kawasaki, Acta Mater. 2012, 60, 726.
[153] C. Jianwei, Ç. Tahir, A. G. William, III, Nanotechnology 2000, 11, 65.
[154] J. Nie, C. Jia, X. Jia, Y. Zhang, N. Shi, Y. Li, Rare Metals 2011, 30, 401.
[155] J. Shuai, L. Xiong, L. Zhu, W. Li, Compos. Part A Appl. Sci. Manuf. 2016, 88, 148.
[156] H. Pal, V. Sharma, T. Nonferr. Metal. Soc. 2015, 25, 154.
[157] J. Wu, H. Zhang, Y. Zhang, X. Wang, Mater. Des. 2012, 41, 344.
[158] J. Kong, C.-y. Zhang, X. Cheng, Appl. Phys. A 2013, 112, 631.
81
[159] K. Chu, C.-c. Jia, W.-s. Li, Appl. Phys. A 2013, 110, 269.
[160] M. M. Billah, Q. Chen, Compos. B Eng. 2017, 129, 162.
[161] S. E. Shin, H. J. Choi, D. H. Bae, J. Compos. Mater. 2012, 47, 2249.
[162] B. Chen, S. Li, H. Imai, L. Jia, J. Umeda, M. Takahashi, K. Kondoh, J. Alloys Compd.
2015, 651, 608.
[163] F. Ogawa, S. Yamamoto, C. Masuda, Mater. Sci. Eng., A 2018, 711, 460.
[164] K. Chu, H. Guo, C. Jia, F. Yin, X. Zhang, X. Liang, H. Chen, Nanoscale Res. Lett.
2010, 5, 868.
[165] S. Zhou, C. Wu, T. Zhang, Z. Zhang, Scripta Mater. 2014, 76, 25.
[166] K. Chu, Q. Wu, C. Jia, X. Liang, J. Nie, W. Tian, G. Gai, H. Guo, Compos. Sci.
Technol. 2010, 70, 298.
[167] C. Edtmaier, T. Steck, R. C. Hula, L. Pambaguian, F. Hepp, Compos. Sci. Technol.
2010, 70, 783.
[168] a) A. L. Moore, L. Shi, Mater. Today 2014, 17, 163; b) X. Huang, T. Iizuka, P. Jiang,
Y. Ohki, T. Tanaka, J. Phys. Chem. C 2012, 116, 13629.
[169] C. Xu, M. Miao, X. Jiang, X. Wang, Compos. Commun. 2018, 10, 103.
[170] a) M. Fujii, X. Zhang, H. Xie, H. Ago, K. Takahashi, T. Ikuta, H. Abe, T. Shimizu,
Phys. Rev. Lett. 2005, 95, 065502; b) S. Berber, Y.-K. Kwon, D. Tománek, Phys. Rev.
Lett. 2000, 84, 4613; c) E. Pop, D. Mann, Q. Wang, K. Goodson, H. Dai, Nano Lett.
2006, 6, 96.
[171] F. Zhang, Y. Feng, M. Qin, T. Ji, F. Lv, Z. Li, L. Gao, P. Long, F. Zhao, W. Feng,
Carbon 2019, 145, 378.
[172] X. Wang, P. Wu, ACS Appl. Mater. Interfaces 2018, 10, 34311.
[173] a) N. Burger, A. Laachachi, M. Ferriol, M. Lutz, V. Toniazzo, D. Ruch, Prog. Polym.
Sci. 2016, 61, 1; b) H. Chen, V. V. Ginzburg, J. Yang, Y. Yang, W. Liu, Y. Huang, L.
82
Du, B. Chen, Prog. Polym. Sci. 2016, 59, 41; c) C. Huang, X. Qian, R. Yang, Mater.
Sci. Eng. R Rep. 2018, 132, 1; d) J. Hansson, T. M. J. Nilsson, L. Ye, J. Liu, Int.
Mater. Rev. 2018, 63, 22; e) Z. Han, A. Fina, Prog. Polym. Sci. 2011, 36, 914.
[174] W. D. Zhang, I. Y. Phang, T. X. Liu, Adv. Mater. 2006, 18, 73.
[175] a) S.-Y. Yang, W.-N. Lin, Y.-L. Huang, H.-W. Tien, J.-Y. Wang, C.-C. M. Ma, S.-M.
Li, Y.-S. Wang, Carbon 2011, 49, 793; b) J. Li, P.-S. Wong, J.-K. Kim, Mater. Sci.
Eng., A 2008, 483-484, 660.
[176] W.-b. Zhang, Z.-x. Zhang, J.-h. Yang, T. Huang, N. Zhang, X.-t. Zheng, Y. Wang, Z.-
w. Zhou, Carbon 2015, 90, 242.
[177] J. Che, M. Jing, D. Liu, K. Wang, Q. Fu, Compos. Part A Appl. Sci. Manuf. 2018, 112,
32.
[178] T. Zhou, X. Wang, X. Liu, D. Xiong, Carbon 2010, 48, 1171.
[179] Z.-G. Wang, F. Gong, W.-C. Yu, Y.-F. Huang, L. Zhu, J. Lei, J.-Z. Xu, Z.-M. Li,
Compos. Sci. Technol. 2018, 162, 7.
[180] Y.-j. Xiao, W.-y. Wang, X.-j. Chen, T. Lin, Y.-t. Zhang, J.-h. Yang, Y. Wang, Z.-w.
Zhou, Compos. Part A Appl. Sci. Manuf. 2016, 90, 614.
[181] K. Wu, Y. Xue, W. Yang, S. Chai, F. Chen, Q. Fu, Compos. Sci. Technol. 2016, 130,
28.
[182] a) K. Kim, J. Kim, Compos. B Eng. 2018, 140, 9; b) Y.-j. Xiao, W.-y. Wang, T. Lin,
X.-j. Chen, Y.-t. Zhang, J.-h. Yang, Y. Wang, Z.-w. Zhou, J. Phys. Chem. C 2016,
120, 6344; c) Y. Xue, X. Li, H. Wang, D. Zhang, Y. Chen, J. Appl. Polym. Sci. 2019,
136, 46929.
[183] M. Shtein, R. Nadiv, M. Buzaglo, K. Kahil, O. Regev, Chem. Mater. 2015, 27, 2100.
[184] J. Che, K. Wu, Y. Lin, K. Wang, Q. Fu, Compos. Part A Appl. Sci. Manuf. 2017, 99,
32.
83
[185] C.-W. Nan, R. Birringer, D. R. Clarke, H. Gleiter, J. Appl. Phys. 1997, 81, 6692.
[186] R. C. Progelhof, J. L. Throne, R. R. Ruetsch, Polym. Eng. Sci. 1976, 16, 615.
[187] a) K. Wu, Y. Li, R. Huang, S. Chai, F. Chen, Q. Fu, Compos. Sci. Technol. 2017, 151,
193; b) G. He, X. Li, Y. Dai, Z. Yang, C. Zeng, C. Lin, S. He, Compos. B Eng. 2019,
162, 678; c) Z.-G. Wang, Y.-F. Huang, G.-Q. Zhang, H.-Q. Wang, J.-Z. Xu, J. Lei, L.
Zhu, F. Gong, Z.-M. Li, Ind. Eng. Chem. Res. 2018, 57, 10391.
[188] T. Ji, Y. Feng, M. Qin, S. Li, F. Zhang, F. Lv, W. Feng, Carbon 2018, 131, 149.
[189] C. Ji, C. Yan, Y. Wang, S. Xiong, F. Zhou, Y. Li, R. Sun, C.-P. Wong, Compos. B
Eng. 2019, 163, 363.
[190] Y. Li, X. Tian, W. Yang, Q. Li, L. Hou, Z. Zhu, Y. Tang, M. Wang, B. Zhang, T. Pan,
Y. Li, Chem. Eng. J. 2019, 358, 718.
[191] P. Bonnet, D. Sireude, B. Garnier, O. Chauvet, Appl. Phys. Lett. 2007, 91, 201910.
[192] F. Wang, X. Zeng, Y. Yao, R. Sun, J. Xu, C.-P. Wong, Sci. Rep. 2016, 6, 19394.
[193] S. Song, Y. Zhang, Carbon 2017, 123, 158.
[194] P. Zhang, X. Ding, Y. Wang, Y. Gong, K. Zheng, L. Chen, X. Tian, X. Zhang,
Compos. Part A Appl. Sci. Manuf. 2019, 117, 56.
[195] L. Wang, H. Qiu, C. Liang, P. Song, Y. Han, Y. Han, J. Gu, J. Kong, D. Pan, Z. Guo,
Carbon 2019, 141, 506.
[196] M. C. Vu, T. S. Tran, Y. H. Bae, M. J. Yu, V. C. Doan, J. H. Lee, T. K. An, S.-R.
Kim, Macromol. Res. 2018, 26, 521.
[197] W. Zhu, N. Hu, Q. Wei, L. Zhang, H. Li, J. Luo, C.-T. Lin, L. Ma, K. Zhou, Z. Yu,
Mater. Des. 2019, 172, 107709.
[198] M. Cao, C. Du, H. Guo, S. Song, X. Li, B. Li, Compos. Sci. Technol. 2019, 173, 33.
[199] S. Choi, H. Im, J. Kim, Nanotechnology 2012, 23, 065303.
[200] V. Guerra, C. Wan, T. McNally, Prog. Mater Sci. 2019, 100, 170.
84
[201] X. Zeng, J. Sun, Y. Yao, R. Sun, J.-B. Xu, C.-P. Wong, ACS Nano 2017, 11, 5167.
[202] V. Goyal, A. A. Balandin, Appl. Phys. Lett. 2012, 100, 073113.
[203] C. Fu, Q. Li, J. Lu, S. Mateti, Q. Cai, X. Zeng, G. Du, R. Sun, Y. Chen, J. Xu, C.-P.
Wong, Compos. Sci. Technol. 2018, 165, 322.
[204] X. Huang, C. Zhi, P. Jiang, D. Golberg, Y. Bando, T. Tanaka, Adv. Funct. Mater.
2013, 23, 1824.
[205] L. Wang, D. Han, J. Luo, T. Li, Z. Lin, Y. Yao, J. Phys. Chem. C 2018, 122, 1867.
[206] D. Kim, S. Ha, H. K. Choi, J. Yu, Y. A. Kim, RSC Adv. 2018, 8, 4426.
[207] D. Kim, M. You, J. H. Seol, S. Ha, Y. A. Kim, J. Phys. Chem. C 2017, 121, 7025.
[208] T. Terao, C. Zhi, Y. Bando, M. Mitome, C. Tang, D. Golberg, J. Phys. Chem. C 2010,
114, 4340.
[209] T. Terao, Y. Bando, M. Mitome, C. Zhi, C. Tang, D. Golberg, J. Phys. Chem. C 2009,
113, 13605.
[210] C. Zhi, Y. Bando, T. Terao, C. Tang, H. Kuwahara, D. Golberg, Adv. Funct. Mater.
2009, 19, 1857.
[211] C.-W. Nan, G. Liu, Y. Lin, M. Li, Appl. Phys. Lett. 2004, 85, 3549.
[212] L. E. Nielsen, Ind. Eng. Chem. Fund. 1974, 13, 17.
[213] F. Jiang, S. Cui, N. Song, L. Shi, P. Ding, ACS Appl. Mater. Interfaces 2018, 10,
16812.
[214] J. Han, G. Du, W. Gao, H. Bai, Adv. Funct. Mater. 2019, 1900412.
[215] J. Chen, X. Huang, B. Sun, Y. Wang, Y. Zhu, P. Jiang, ACS Appl. Mater. Interfaces
2017, 9, 30909.
[216] J. Hu, Y. Huang, X. Zeng, Q. Li, L. Ren, R. Sun, J.-B. Xu, C.-P. Wong, Compos. Sci.
Technol. 2018, 160, 127.
[217] J. Chen, X. Huang, Y. Zhu, P. Jiang, Adv. Funct. Mater. 2017, 27, 1604754.
85
[218] X. Hou, Y. Chen, L. Lv, W. Dai, S. Zhao, Z. Wang, L. Fu, C.-T. Lin, N. Jiang, J. Yu,
ACS Appl. Nano Mater. 2019, 2, 360.
[219] H. Fang, S.-L. Bai, C. P. Wong, Compos. Part A Appl. Sci. Manuf. 2017, 100, 71.
[220] H. Shen, C. Cai, J. Guo, Z. Qian, N. Zhao, J. Xu, RSC Adv. 2016, 6, 16489.
[221] D. Ding, H. Wang, Z. Wu, Y. Chen, Q. Zhang, Macromol. Rapid Commun. 2019, 0,
1800805.
[222] M. Tanimoto, T. Yamagata, K. Miyata, S. Ando, ACS Appl. Mater. Interfaces 2013, 5,
4374.
[223] S. Diaham, F. Saysouk, M.-L. Locatelli, B. Belkerk, Y. Scudeller, R. Chiriac, F.
Toche, V. Salles, J. Appl. Polym. Sci. 2015, 132, 42461.
[224] K. Sato, H. Horibe, T. Shirai, Y. Hotta, H. Nakano, H. Nagai, K. Mitsuishi, K. Watari,
J. Mater. Chem. 2010, 20, 2749.
[225] H.-J. Hong, S. M. Kwan, D. S. Lee, S. M. Kim, Y. H. Kim, J. S. Lim, J. Y. Hwang, H.
S. Jeong, Compos. Sci. Technol. 2017, 152, 94.
[226] H. Hong, J. Kim, T.-i. Kim, Polymers 2017, 9, 413.
[227] Y. Yao, X. Zhu, X. Zeng, R. Sun, J.-B. Xu, C.-P. Wong, ACS Appl. Mater. Interfaces
2018, 10, 9669.
[228] a) J. He, H. Wang, Z. Su, Y. Guo, X. Tian, Q. Qu, Y.-L. Lin, Compos. Part A Appl.
Sci. Manuf. 2019, 117, 287; b) W. Dai, J. Yu, Y. Wang, Y. Song, F. E. Alam, K.
Nishimura, C.-T. Lin, N. Jiang, J. Mater. Chem. A 2015, 3, 4884.
[229] A. J. McNamara, Y. Joshi, Z. M. Zhang, Int. J. Therm. Sci. 2012, 62, 2.
[230] M. Foygel, R. D. Morris, D. Anez, S. French, V. L. Sobolev, Phys. Rev. B 2005, 71,
104201.
[231] C. Xiao, L. Chen, Y. Tang, X. Zhang, K. Zheng, X. Tian, Compos. Part A Appl. Sci.
Manuf. 2019, 116, 98.
86
[232] Y. Huang, J. Hu, Y. Yao, X. Zeng, J. Sun, G. Pan, R. Sun, J.-B. Xu, C.-P. Wong, Adv.
Mater. Interfaces 2017, 4, 1700446.
[233] D. Shen, Z. Zhan, Z. Liu, Y. Cao, L. Zhou, Y. Liu, W. Dai, K. Nishimura, C. Li, C.-T.
Lin, N. Jiang, J. Yu, Sci. Rep. 2017, 7, 2606.
[234] D. Shen, M. Wang, Y. Wu, Z. Liu, Y. Cao, T. Wang, X. Wu, Q. Shi, K. W. A. Chee,
W. Dai, H. Bai, D. Dai, J. Lyu, N. Jiang, C.-T. Lin, J. Yu, High Voltage 2017, 2, 154.
[235] Z. Gao, L. Zhao, Mater. Des. 2015, 66, 176.
[236] W. Bian, T. Yao, M. Chen, C. Zhang, T. Shao, Y. Yang, Compos. Sci. Technol. 2018,
168, 420.
[237] W. Sun Lee, J. Yu, Diamond Relat. Mater. 2005, 14, 1647.
[238] P. Anithambigai, M. K. Dheepan Chakravarthii, D. Mutharasu, L. H. Huong, T.
Zahner, D. Lacey, I. Kamarulazizi, J. Mater. Sci. Mater. Electron. 2017, 28, 856.
[239] G. Pan, Y. Yao, X. Zeng, J. Sun, J. Hu, R. Sun, J.-B. Xu, C.-P. Wong, ACS Appl.
Mater. Interfaces 2017, 9, 33001.
[240] M. Hu, J. Feng, K. M. Ng, Compos. Sci. Technol. 2015, 110, 26.
[241] Y. Ouyang, G. Hou, L. Bai, B. Li, F. Yuan, Compos. Sci. Technol. 2018, 165, 307.
[242] S. Zhang, Y. Ke, X. Cao, Y. Ma, F. Wang, J. Appl. Polym. Sci. 2012, 124, 4874.
[243] Y. Gao, J. Liu, Appl. Phys. A 2012, 107, 701.
[244] H. Yuan, Y. Wang, T. Li, P. Ma, S. Zhang, M. Du, M. Chen, W. Dong, W. Ming,
Compos. Sci. Technol. 2018, 164, 153.
[245] a) Y. Zhou, L. Wang, H. Zhang, Y. Bai, Y. Niu, H. Wang, Appl. Phys. Lett. 2012, 101,
012903; b) D. Yorifuji, S. Ando, Macromol. Chem. Phys. 2010, 211, 2118; c) J. Li, X.
Li, Y. Zheng, Z. Liu, Q. Tian, X. Liu, J. Mater. Sci. 2019, 54, 6258.
[246] a) Y. Zhou, S. Yu, H. Niu, F. Liu, Polymers 2018, 10, 1412; b) S. Wang, Y. Cheng, R.
Wang, J. Sun, L. Gao, ACS Appl. Mater. Interfaces 2014, 6, 6481.
87
[247] N. Balachander, I. Seshadri, R. J. Mehta, L. S. Schadler, T. Borca-Tasciuc, P.
Keblinski, G. Ramanath, Appl. Phys. Lett. 2013, 102, 093117.
[248] K. Kim, K. Ahn, H. Ju, J. Kim, Ind. Eng. Chem. Res. 2016, 55, 2713.
[249] L. Gao, X. Zhou, Y. Ding, Chem. Phys. Lett. 2007, 434, 297.
[250] a) K. Ahn, K. Kim, J. Kim, Polymer 2015, 76, 313; b) C. Chen, Y. Tang, Y. S. Ye, Z.
Xue, Y. Xue, X. Xie, Y.-W. Mai, Compos. Sci. Technol. 2014, 105, 80.
[251] Y. Agari, T. Uno, J. Appl. Polym. Sci. 1985, 30, 2225.
[252] W. Chen, Z. Wang, C. Zhi, W. Zhang, Compos. Sci. Technol. 2016, 130, 63.
[253] A. Rai, A. L. Moore, Compos. Sci. Technol. 2017, 144, 70.
[254] D. Zhu, W. Yu, H. Du, L. Chen, Y. Li, H. Xie, J Nanomater. 2016, 2016, 8.
[255] S. Bhanushali, P. C. Ghosh, G. P. Simon, W. Cheng, Adv. Mater. Interfaces 2017, 4,
1700387.
[256] L. Rivière, A. Lonjon, E. Dantras, C. Lacabanne, P. Olivier, N. R. Gleizes, Eur.
Polym. J. 2016, 85, 115.
[257] D. Suh, C. M. Moon, D. Kim, S. Baik, Adv. Mater. 2016, 28, 7220.
88
89
Author Bio-data:
Haifei Zhan is currently a Lecturer of Mechanical Engineering at
the Queensland University of Technology. He received his Ph.D.
degree in mechanical engineering from there in 2013, and worked as
Postdoc Research Fellow until 2016. In 2017, he worked as a
lecturer in Western Sydney University and moved back to
Queensland University of Technology in 2018. His research focuses
on nanoscale thermal transport properties, and mechanical properties of low-dimensional
nanomaterials, nanocomposites, and nanofibers.
Yongnan Chen is currently a Professor of Material Engineering at
Chang'an University. He received his Ph.D. degree in material
science from Xi'an Jiaotong University in 2009, and then moved to
Chang'an University since 2010. His research focuses on the
synthetization, characterization and commercialization of high-
performance metals/alloys and metallic composites, and wear and
corrosion protection.
Yuantong Gu is currently a Professor of Mechanical Engineering at
the Queensland University of Technology. He received his Ph.D.
degree in mechanical engineering from National University of
Singapore in 2003 and worked as a Research Fellow in the
University of California (Irvine) in 2004 and The University of
Sydney from 2005. In 2007, he moved to Queensland University of
Technology. His research focuses on the computational mechanics, nanomechanics, and
biomechanics.
90
Table of Contents
Keyword: Thermal conductivity, nanostructures, carbon nanotube, nanocomposite, metal
matrix composites
H.F. Zhan, Y.H. Nie, Y.N. Chen*, J. M. Bell, Y.T. Gu*
Title: Thermal Transport of 3D Nanostructures
ToC figure:
ToC entry: Diverse 3D nanostructures have been fabricated to meet specific thermal
management purposes, including nanoarchitectures (like colloidal assemblies, carbon
nanotube arrays), and nanocomposites from polymer matrix or metal matrix. It is shown that
the use of nanostructuring of materials is an effective strategy to modify the thermal
conductivity of materials.
91
|
1810.08254 | 1 | 1810 | 2018-10-18T19:34:03 | Atomically-thin quantum dots integrated with lithium niobate photonic chips | [
"physics.app-ph",
"quant-ph"
] | The electro-optic, acousto-optic and nonlinear properties of lithium niobate make it a highly versatile material platform for integrated quantum photonic circuits. A prerequisite for quantum technology applications is the ability to efficiently integrate single photon sources, and to guide the generated photons through ad-hoc circuits. Here we report the integration of quantum dots in monolayer WSe2 into a Ti in-diffused lithium niobate directional coupler. We investigate the coupling of individual quantum dots to the waveguide mode, their spatial overlap, and the overall efficiency of the hybrid-integrated photonic circuit. | physics.app-ph | physics | Atomically-thin quantum dots integrated with lithium niobate photonic chips
Daniel White1, Artur Branny1, Robert J. Chapman2, Raphaël Picard1, Mauro Brotons-Gisbert1, Andreas Boes2,
Alberto Peruzzo2, Cristian Bonato1, Brian D. Gerardot1,*
1Institute of Photonics and Quantum Sciences, SUPA, Heriot-Watt University, Edinburgh EH14 4AS, UK
2 Quantum Photonics Laboratory and Centre for Quantum Computation and Communication Technology,
School of Engineering, RMIT University, Melbourne, Victoria 3000, Australia
*[email protected]
The electro-optic, acousto-optic and nonlinear properties of lithium niobate make it a highly
versatile material platform for integrated quantum photonic circuits. A prerequisite for quantum
technology applications is the ability to efficiently integrate single photon sources, and to guide the
generated photons through ad-hoc circuits. Here we report the integration of quantum dots in
monolayer WSe2 into a Ti in-diffused lithium niobate directional coupler. We investigate the coupling
of individual quantum dots to the waveguide mode, their spatial overlap, and the overall efficiency of
the hybrid-integrated photonic circuit.
INTRODUCTION
Solid-state quantum emitters have emerged in the past
few years as prominent systems for quantum technology.
Single-photon sources based on semiconductor quantum
dots [1], defects [2] and impurities [3] have reached
extremely high purity [4] and efficiency [1], making them
competitive for future quantum applications such as
quantum communication [5] or linear optical quantum
computing [6]. Additionally, spin-active quantum emitters
provide an excellent platform for quantum networking,
enabling efficient interfacing between photons to share
quantum states over long distances, and spins, which can
store and process quantum information locally [7], [8].
required
robustness
When implementing either purely optical or hybrid spin-
photon schemes, a large number of optical components are
required to construct the required photonic quantum gates.
For increasing number of qubits, approaches based on bulk
optics become quickly impractical, due to the size and the
inherent instability of these individual components. Only
photonic integration can provide the stability, repeatability
and
for high-visibility quantum
interference and scalability [9]. For these reasons a large
research effort in the past decade has been dedicated to the
development of integrated quantum photonics circuits with
embedded quantum emitters [10] -- [13]. However, several
challenges are still outstanding, such as total internal
reflection in bulk high-index materials that severely limits
optical collection efficiency and the in-situ manipulation of
the phase in interferometer arms. Furthermore, materials
such as diamond, while featuring excellent spin properties,
are notoriously hard to grow and fabricate, posing a serious
challenge to the development of a large-scale integrated
platform.
In this respect, quantum dots in two-dimensional layered
materials such as transition metal dichalcogenide (TMDs)
exhibit promising properties. Bulk TMDs form layered
structures with weak interlayer van-der-Waals interactions
which can be micro-mechanically exfoliated to obtain
monolayer crystals [14]. One of the most prominent
examples is WSe2 which, at the monolayer level, is a direct
band-gap semiconductor featuring bright localized excitons
[15] or bi-excitons [16] at cryogenic temperatures. Once
exfoliated, the TMD material can be stacked to create layer-
by-layer van der Waals heterostructures with designer
functionality [17], [18] or placed onto arbitrary substrates
[19], including complex photonic circuits, using a variety of
transfer methods [20]. As quantum emission is embedded
in the single atomic layer, optical extraction is not limited
by total internal reflection, an enormous challenge for bulk
materials such as GaAs, diamond or SiC. In addition, the
monolayer can be transferred to any photonic structure
fabricated in any material platform, removing any need for
lattice matching or complex bonding procedures. Finally,
the quantum emitter can be deterministically created at a
pre-determined position by strain engineering [21] -- [23].
An attractive photonic platform to integrate quantum
emitters is lithium niobate. Lithium niobate is the industry
standard for the photonic telecommunication industry,
possessing the ideal properties for integrated photonics,
including low-loss transmission in a broad wavelength
range, from UV
to mid-IR, and ultrafast switching
capability. Crystalline lithium niobate uniquely combines
nonlinear, piezoelectric, photorefractive and electro-optic
effects in the same platform [24], making it an attractive
system for quantum technology. Low-loss waveguides and
other passive components, such as beam-splitters and ring
resonators [25], have been demonstrated, both by Ti in-
diffusion or
Its electro-optic
properties have been exploited to build fast electrically-
ridge structures
[26].
controlled phase modulators at gigahertz operating speeds
[27], [28], with recent work reporting 210 Gbit s-1 devices
[29]. Integration with superconducting nanowire detectors,
which offer single photon detection with the performance
required for quantum photonics application, has already
been demonstrated [30].
to an
integrated
Here we demonstrate the coupling of quantum dots in
monolayer WSe2
lithium niobate
directional coupler, a fundamental building block for
quantum interference experiments. The integration of 2D
materials hosting quantum emitters in this mature photonic
platform holds the promise to realise fully-integrated
quantum circuits for future quantum technologies.
DEVICE DESIGN, FABRICATION, AND
EXPERIMENTAL SET-UP
The
fabricated
directional
coupler was
by
photolithographically patterning the waveguide structure (4
µm width) into a Ti thin film (70 nm thickness) on the
surface of an X-cut LiNbO3 wafer. The titanium was
diffused into LiNbO3 in a wet oxygen atmosphere at
1010°C for several hours. After the diffusion, the wafer was
diced into a 2.5 cm long chip and the chip end faces were
polished to optical grade. To enable stable fiber coupling
inside the cryostat, a fiber array was aligned to the
waveguides and glued to one facet of the chip using
cryogenic compatible epoxy glue. The coupling was found
to remain constant after several cryogenic cooling cycles.
The waveguide propagation loss is estimated to be 0.5
dB/cm, while the coupling efficiency from fiber
to
waveguide is estimated to be 3.5 dB. This is confirmed by a
total loss from output fiber to input facet of 4 dB, with the
reflectivity of the directional coupler measured to be 70%.
The facet was confocally imaged with a 760 nm laser
Fig. 1. (a) Spatial profile of the waveguide mode at the facet of the WSe2 flake, measured by illuminating a fiber-
coupled port with a 750 nm laser and detecting light through the Confocal Port. The white dash-dot line indicates the 1/e2
mode profile, the color scale shows the normalized intensity. (b) Optical microscope image of WSe2 transferred onto the
facet of the waveguide. The 20 µm monolayer region posseses a wrinkle with localized strain. The illuminated waveguide
mode (red spot) is used to align the monolayer area of the flake with respect to the optical mode. (c) Schematic of Ti in-
diffused lithium niobate directional coupler with a WSe2 flake at the input facet. PL was measured by exciting the emitters
in a confocal microscope, while emission was detected in confocal geometry (Confocal Port) and through the two fiber
output ports (Port 1 and Port 2).
placed into one of the fiber output ports to illuminate the
input mode. The image (see Fig. 1(a)) shows an elliptical
mode with major and minor axes of 10 µm and 6 µm,
respectively (1/e2). The numerical aperture (NA) of the
waveguide is estimated to be 0.07.
Micromechanical exfoliation was used
to obtain
monolayer WSe2 flakes, identified by optical microscopy.
We select a monolayer featuring a wrinkle as it has been
shown that the associated localized strain induces single
photon emitters [21]. The flake was transferred onto the
waveguide using
stamping
procedure [31]. To illuminate the waveguide optical mode
position, a HeNe (633nm) laser was sent from the opposite
end of the directional coupler. This allowed deterministic
transfer of flakes with wrinkles directly onto the optical
mode. One example is shown in Fig. 1(b).
the all-dry viscoelastic
Photoluminescence (PL) measurements were performed
in a confocal microscope setup with the sample mounted
onto a 3-axis nanopositioner and cooled to 4K. The sample
was excited off-resonantly with 1 µW of continuous wave
532 nm laser, focused to a 1µm spot through an objective
of numerical aperture 0.82. The experimental setup, shown
in Fig. 1(c), enables PL to be collected confocally from the
WSe2 flake or from the two waveguide collection ports.
RESULTS AND DISCUSSION
The confocal image produced in Fig. 2(a) shows PL in
the spectral range 720 -- 735 nm from a flake transferred
onto the waveguide facet. The marked areas reveal a
region (solid white line) closest to the
monolayer WSe2
mode center and a bilayer region (dotted white line). In the
bilayer region we find the characteristic indirect optical
transition at around 800 nm (not shown), while the
monolayer has local regions of bright emission, as shown in
Fig. 2(a). In the areas of strongly localized emission, sharp
spectral lines are observed over a distribution of emission
wavelengths. These regions feature quantum dots and are
attributed to the local strain induced by the wrinkle and the
edges of the flake. To demonstrate the coupling of the
quantum dot emission to the directional coupler, we
spatially excite at the localized bright regions (marked by
Roman numerals in Fig. 3(a, b)) and collect PL from the
two fiber output ports. We find identical spectra at each
output port which match the emission measured confocally,
Fig. 3 (c, d). This confirms the directional coupler acts as
an on-chip beam-splitter. The device operates as a 70:30
beam-splitter around the peak of the PL ensemble (720 to
760 nm). Whilst no single photon emitters were located
directly at the centre of the mode (which would maximize
the coupling efficiency) for this particular flake, numerous
isolated emitters were still coupled to the waveguide with
PL from centres visible at a displacement of up to 3.5µm.
An estimate of the coupling efficiency can be obtained by
calculating the dipole emission collected by a waveguide
Fig. 2. (a) Confocal PL image of the WSe2 flake on the waveguide facet in the spectral region of 720-735 nm, measured at
4K. The dotted white line indicates the edge of the few-layer region while the solid line marks the edges of the monolayer.
Several areas of strong localized PL, corresponding to individual quantum dots, are present in the region surrounding the
waveguide mode (marked by the dash-dot line). (b) Spatial map of localized PL from monolayer WSe2 with individual single
photon emitters marked by Roman numerals I-VII. Color bars show the normalized intensity.
Fig. 3. (a) Spatially-resolved PL spectrally filtered at 745 nm and 725 nm (b). Color scale represents the normalized
intensity. (c) PL spectra measured through the waveguide fiber output Ports 1 and 2 and confocally when the position of
Emitter I is excited. (d) Spectra measured at Ports 1 and 2 and confocally when Emitter II is spatially excited. The spectra
observed in confocal geometry is replicated at both output ports, albeit with a reduced photon count rate.
with NA = 0.07, in the ray optics approximation.
Using the transfer matrix approach, around 10% of
emission from a dipole at an air-lithium niobate (n = 2.2)
interface is collected into an objective of numerical aperture
0.82. Accounting for optical and fiber coupling losses,
approximately 3% of the total emission is collected in the
confocal geometry. In contrast, a fraction of 0.7% of the
dipole emission is collected by a lithium niobate waveguide
with numerical aperture NA = 0.07. Assuming 40%
coupling efficiency from the waveguide into single-mode
fibers, we expect 0.3% of the dipole emission at the fiber
ports. Therefore, the expected ratio between the collection
efficiency through the fibers (Ports 1 & 2) and through the
confocal port is approximately 10% for a dipole located at
the mode center.
We confirm these values through finite-difference time-
domain (FDTD) simulations, modelling a dipole with
varying displacement from the mode centre. For a dipole
positioned at the mode centre we expect a collection
efficiency of 0.3% (including fiber coupling losses), which
reduces to 0.12% when the dipole is located 1.1 μm away
from the mode centre. Taking the ratio of the counts
experimentally measured through both fiber output ports
and confocally, we achieve a collection efficiency of 0.11 ±
0.02% for an emitter at position I (Fig. 3 (a)), which agrees
well with FDTD and transfer matrix simulations. The
measured and
simulated collection efficiencies are
displayed in Table 1 for emitters I-VII (as shown in Fig. 3).
The low overall coupling efficiency of the device is
associated with the small index contrast of the Ti in-
diffused region and, consequently, with the small numerical
aperture (0.07) of the device. This problem can be
addressed by using ridge waveguides etched in thin film
lithium niobate-on-insulator (LNOI) [25] -- [28], [32], [33]
which feature much larger NA.
Table 1. Comparison of measured and simulated
emitter
efficiency as a
collection
displacement
function of
Emitter Wavelength
(nm)
745.5
731.2
724.8
733.4
795
734.5
751.6
I
II
III
IV
V
VI
VII
Displacement
(µm)
1.1
1.3
1.8
2.5
3.2
3.4
3.5
Collection
efficiency (%)
Simulated
efficiency (%)
0.11±0.02
0.1±0.03
0.02±0.01
0.02±0.01
0.01±0.01
0.01±0.01
0.01±0.01
0.12
0.12
0.06
CONCLUSION
the operation of
two-dimensional materials
In conclusion, we have achieved the integration of
quantum emitters in atomically thin WSe2 onto a Ti in-
diffused lithium niobate waveguide with a fiber coupled,
on-chip beam-splitter. PL from the quantum dots was
measured confocally at the waveguide facet and identical
emission was collected through each output port of the
device, demonstrating
the on-chip
Hanbury-Brown and Twiss interferometer. We expect the
integration of
lithium
niobate elements will be a key platform in the development
of integrated quantum photonic circuits. In particular, the
high
film LNOI
waveguides can provide improved optical coupling between
the quantum dots and the waveguide and local strain-
inducing features (such as nanopillars) can be incorporated
on the facet edge. The hybrid quantum emitter-LNOI
platform can be extended with in-situ strain tuning
capability via the piezoelectric effect as well as access to
passive and active optical elements, including grating
couplers, compact
integrated beam-splitters and ring
resonators, and efficient reconfigurable interferometers.
into
refractive
index contrast of
thin
FUNDING
for Quantum Computation
AP acknowledges funding from Australian Research
Council Centre
and
Communication Technology CE170100012; Australian
Research Council Discovery Early Career Researcher
Award, Project No. DE140101700; RMIT University Vice-
Chancellors Senior Research Fellowship. Work at Heriot-
Watt
the EPSRC (grant numbers
EP/L015110/1 and EP/P029892/1) and the ERC (number
725920).
is supported by
ACKNOWLEDGEMENTS
This work was performed in part at the Micro
Nano Research Facility (MNRF) at RMIT University,
Melbourne, part of the ANFF, a company established under
the National Collaborative Research Infrastructure Strategy
to provide nano-and microfabrication
for
Australia's researchers. BDG thanks the Royal Society for a
Wolfson Merit Award and
the Royal Academy of
Engineering for a Chair in Emerging Technology.
facilities
APPENDIX
[1]
[2]
[3]
[4]
[5]
single-photon
P. Senellart, G. Solomon, and A. White, "High-performance
sources," Nature
semiconductor quantum-dot
Nanotechnology. 2017.
A. Sipahigil et al., "Indistinguishable photons from separated silicon-
vacancy centers in diamond," Phys. Rev. Lett., 2014.
R. Kolesov et al., "Optical detection of a single rare-earth ion in a
crystal," Nat. Commun., 2012.
L. Schweickert et al., "On-demand generation of background-free
single photons from a solid-state source," Appl. Phys. Lett., 2018.
K. Takemoto et al., "Quantum key distribution over 120 km using
ultrahigh purity single-photon source and superconducting single-
photon detectors," Sci. Rep., 2015.
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
E. Knill, R. Laflamme, and G. J. Milburn, "A scheme for efficient
quantum computation with linear optics," Nature, 2001.
P. C. Humphreys et al., "Deterministic delivery of remote
entanglement on a quantum network," Nature, 2018.
W. B. Gao, A. Imamoglu, H. Bernien, and R. Hanson, "Coherent
manipulation, measurement and entanglement of individual solid-
state spins using optical fields," Nat. Photonics, 2015.
S. Bogdanov, M. Y. Shalaginov, A. Boltasseva, and V. M. Shalaev,
"Material platforms for integrated quantum photonics," Opt. Mater.
Express, 2017.
N. Prtljaga et al., "Monolithic integration of a quantum emitter with
a compact on-chip beam-splitter," Appl. Phys. Lett., 2014.
A. D. Greentree, B. A. Fairchild, F. M. Hossain, and S. Prawer,
"Diamond integrated quantum photonics," Materials Today. 2008.
M. Radulaski et al., "Scalable Quantum Photonics with Single Color
Centers in Silicon Carbide," Nano Lett., vol. 17, no. 3, pp. 1782 --
1786, 2017.
T. Zhong, J. M. Kindem, E. Miyazono, and A. Faraon,
"Nanophotonic coherent light-matter interfaces based on rare-earth-
doped crystals," Nat. Commun., 2015.
R. Lv et al., "Transition metal dichalcogenides and beyond:
Synthesis, properties, and applications of single- and few-layer
nanosheets," Acc. Chem. Res., 2015.
P. Tonndorf et al., "Single-photon emission from localized excitons
in an atomically thin semiconductor," Optica, 2015.
Y. M. He et al., "Cascaded emission of single photons from the
biexciton in monolayered WSe 2," Nat. Commun., 2016.
C. Palacios-Berraquero et al., "Atomically thin quantum light-
emitting diodes," Nat. Commun., 2016.
G. B. Brotons-Gisbert M, Branny A, Kumar S, Picard R, Proux R,
Gray M, Burch K, Watanabe K, Taniguchi T, "Coulomb blockade in
an atomically thin quantum dot strongly coupled to a tunable Fermi
reservoir," ArXiv, 2018.
P. Tonndorf et al., "On-Chip Waveguide Coupling of a Layered
Semiconductor Single-Photon Source," Nano Lett., 2017.
R. Frisenda et al., "Recent progress in the assembly of nanodevices
and van der Waals heterostructures by deterministic placement of 2D
materials," Chemical Society Reviews. 2018.
S. Kumar, A. Kaczmarczyk, and B. D. Gerardot, "Strain-Induced
Spatial and Spectral Isolation of Quantum Emitters in Mono- and
Bilayer WSe2," Nano Lett., 2015.
A. Branny, S. Kumar, R. Proux, and B. D. Gerardot, "Deterministic
strain-induced arrays of quantum emitters in a two-dimensional
semiconductor," Nat. Commun., 2017.
C. Palacios-Berraquero et al., "Large-scale quantum-emitter arrays
in atomically thin semiconductors," Nat. Commun., 2017.
M. Bazzan and C. Sada, "Optical waveguides in lithium niobate:
Recent developments and applications," Applied Physics Reviews.
2015.
A. P. Inna Krasnokutska, Jean-Luc J. Tambasco, "Large free spectral
range microring resonators in lithium niobate on insulator," ArXiv,
no. arXiv:1807.06531, 2018.
H. Hu, R. Ricken, W. Sohler, and R. B. Wehrspohn, "Lithium
niobate ridge waveguides fabricated by wet etching," IEEE
Photonics Technol. Lett., 2007.
W. Sohler et al., "Integrated Optical Devices in Lithium Niobate,"
Opt. Photonics News, 2008.
H. Herrmann, K. D. Büchter, R. Ricken, and W. Sohler, "Tunable
integrated electro-optic wavelength
filter with programmable
spectral response," J. Light. Technol., 2010.
C. Wang et al., "Integrated lithium niobate electro-optic modulators
operating at CMOS-compatible voltages," Nature, vol. 562, no.
7725, pp. 101 -- 104, 2018.
M. G. Tanner, L. S. E. Alvarez, W. Jiang, R. J. Warburton, Z. H.
Barber, and R. H. Hadfield, "A superconducting nanowire single
photon detector on lithium niobate," Nanotechnology, 2012.
A. Castellanos-Gomez et al., "Deterministic transfer of two-
dimensional materials by all-dry viscoelastic stamping," 2D Mater.,
2014.
I. Krasnokutska, J.-L. J. Tambasco, X. Li, and A. Peruzzo, "Ultra-
low loss photonic circuits in litium nibate on insulator," Opt.
Express, 2018.
M. Zhang et al., "Ultra-High Bandwidth Integrated Lithium Niobate
Modulators with Record-Low Vπ," in Optical Fiber Communication
Conference Postdeadline Papers, 2018.
|
1901.07206 | 1 | 1901 | 2019-01-22T08:38:58 | A novel design for cruiser type motorcycle silencer based on micro perforated elements | [
"physics.app-ph"
] | ...Micro perforated elements are innovative acoustic solutions,which silencing effect is based on the dissipation of the acoustic wave energy in a pattern of sub-millimeter apertures. Similarly to fibrous materials the micro-perforated materials have been proved to provide effective sound absorption in a wide frequency range. Additionally, the silencer is designed as a two-stage system that provides an optimal solution for a variety of exploitation conditions. In this paper a novel design for a cruiser type motorcycle silencer, based on micro-perforated elements, is presented. It has been demonstrated that the micro-perforated elements can successfully be used to achieve high attenuation of IC-engine noise in strictly limited circumstances. A technical description of the design and manufacturing of the prototype silencer is given and technological issues are discussed. The acoustical and fluid-dynamical performance of the silencer is characterized by transmission loss and pressure drop data. The influence of the two-stage system valve operation has been analyzed by studying the acoustics data and engine output characteristics. In addition to the experimental investigations, numerical 1-D models were developed for the optimization of the silencer geometry and the results are compared in a number of operating conditions. The studies have resulted in development of a silencer system for a small series cruiser type motorcycle. The first silencer prototypes have been tested on the motorcycle. While maintaining acceptable pressure drop characteristics, it has proven to comply with standard noise criteria without incorporating fibrous materials.... | physics.app-ph | physics | Preprint of the paper:
Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements". SAE Technical Paper 2012-32-
0109, 2012. https://doi.org/10.4271/2012-32-0109
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
A Novel Design for Cruiser Type Motorcycle Silencer Based on
Micro-Perforated Elements
Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J.
Tallinn University of Technology, Tallinn, Estonia
ABSTRACT
Regulations stipulating the design of motorcycle silencers are strict, especially when the unit incorporates fibrous absorbing
materials. Therefore, innovative designs substituting such materials while still preserving acceptable level of characteristic sound
are currently of interest.
Micro perforated elements are innovative acoustic solutions, which silencing effect is based on the dissipation of the acoustic wave
energy in a pattern of sub-millimeter apertures. Similarly to fibrous materials the micro perforated materials have been proved to
provide effective sound absorption in a wide frequency range. Additionally, the silencer is designed as a two-stage system that
provides an optimal solution for a variety of exploitation conditions.
In this paper a novel design for a cruiser type motorcycle silencer, based on micro-perforated elements, is presented. It has been
demonstrated that the micro perforated elements can successfully be used to achieve high attenuation of IC-engine noise in strictly
limited circumstances. A technical description of the design and manufacturing of the prototype silencer is given and technological
issues are discussed. The acoustical and aerodynamical performance of the silencer is characterized by transmission loss and pressure
drop data. The influence of the two-stage system valve operation has been analyzed by studying the acoustics data and engine output
characteristics.
In addition to the experimental investigations, numerical 1-D models were developed for the optimization of the silencer geometry
and the results are compared in a number of operating conditions.
The studies have resulted in development of a silencer system for a small series cruiser type motorcycle. The first silencer prototypes
have been tested on the motorcycle. While maintaining acceptable pressure drop characteristics, it has proven to comply with
standard noise criteria without incorporating fibrous materials.
The radiated motorcycle sound, as one of the key features of successful design, has been evaluated. The sound design has been
recognized as well suitable for the product.
INTRODUCTION
A motorcycle design is a complex task integrating challenges of engineering analyzes with marketing goals while complying with
a number of regulations. The exhaust system of this type of motorcycle (see Fig. 1) has to provide not only adequate engine noise
cancellation but also a pleasant characteristic sound while preserving acceptable level of exhaust gas flow restriction.
The present work is focusing on the exhaust system design for exclusive small series motorcycle. Traditionally, this type of
motorcycle is equipped with a relatively large displacement internal combustion engine (see Table 1) designed for a moderate rpm
range.
Page 2 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
Figure 1 -- A side view of the motorcycle prototype [1]. The location of the silencer unit (orange rectangle) and the direction of
exhausting gas flow (dashed red arrow) are illustrated.
Table 1 -- Characteristic parameters of the motorcycle engine [1].
layout
total displacement
geom. compression ratio
bore
stroke
V2, 90°
1326cm3
10,8
102mm
81,2mm
max. torque
134Nm (5600rpm)
max. power
123hp (7100rpm)
engine management
electronical (EFI, Euro4)
cooling system
water cooling
In order to introduce this type of vehicle for initial registration in European Union (EU) according to Directive 2002/24/EC [2] the
type approval evaluation has to be performed. Procedures for the evaluation of exhaust systems are stipulated in Directive 97/24/EC
[3]. Accordingly the maximum by-pass (vehicle in motion) noise level for this type of motorcycle is limited to 80dB(A).
Typically, a motorcycle silencer incorporates fibrous materials. To avoid negative side effects, e.g. deterioration of performance due
to relocation and blow out of the fibers, challenging demands have been stipulated in Directive 97/24/EC [3] on the implementation
of the fibrous materials. A recent investigation [4] on the use of micro-perforated (MP) elements in flow-duct silencers has proven
that the fibrous materials can be successfully substituted by the MP elements.
As a well-known issue the noise attenuation ability typically compromises the pressure drop of the silencer. As the pressure drop is
one of the key parameters affecting the engine performance and fuel consumption, innovative technical solutions (e.g. micro
perforations and meta-materials) providing satisfactory noise cancellation while preserving low pressure drop are of great interest.
The motorcycle exhaust system treated in this paper was designed taking into account the following technical constraints,
concentrated into Table 2. In the present work a technical description of the silencer is given together with experimentally and
numerically obtained characteristic parameters, relevant to successfully fulfill the technical constraints listed in Table 2.
Page 3 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
Table 2 -- Technical constraints for the exhaust system design.
No.
Constraint
1
2
3
4
5
6
The geometry of the silencer is restricted by the
overall design of the motorcycle components (See
Fig. 1).
The motorcycle equipped with the silencer must
satisfy the stipulated noise limits [3].
The silencer design should provide an option for
less flow restrictive "straight flow" configuration.
The fibrous acoustic materials should be avoided
[3] inside the silencer.
The silencer should resist corrosive environments
The mass of the complete silencer system should
be minimal.
SILENCER DESIGN
OVERVIEW OF DESIGN PROCEDURES
The research and development of the silencer system including experimental testing and 1-D computer simulations were carried out
in cooperation with technical acoustics laboratory at Tallinn University of Technology. A dedicated hot flow test facility presented
in [4, 5 and 6] was implemented for the experimental acoustic characterization of the complete silencer and silencer components.
During the product development project the following design procedures were performed:
1. Analysis on the positioning of the silencer. Determination of the space available, the location of the mounting structures
and the inlet and outlet tubes;
2. Manufacturing and acoustic characterization of micro-perforated tubular elements (40 test samples);
3. Set up of acoustic simulation models in 1-D analysis software followed by preliminary analyses;
4. Manufacturing and acoustic characterization of the first geometrically simplified silencer prototype;
5. On-vehicle testing of the silencer prototype in a variety of operating conditions (stationary: idle, 3000rpm, 6000rpm and
by-pass tests [3]);
6. Development and road testing [3] of extra noise control guide valves;
7. Motorcycle testing on rolling road and analysis for the engine equipped with the 2-stage silencer prototype;
8. Acoustic 2-port testing and characterization of noise control guide valves;
9. Manufacturing of the complete silencer system for the motorcycle.
ACOUSTICAL DESIGN
Since the dominant noise radiation typically originates from the exhaust gas pressure pulsations related to the first harmonics of the
firing frequencies, the engine (see Table 1) tuned for medium crankshaft RPM range can be regarded as a low frequency source.
Hence, an effective noise cancellation for such engine is technically obtained by the maximization of the acoustic wave reflections.
In order to provide high reflections the cross section area ratio at the sudden exhaust system area discontinuities (e.g. in expansion
chambers) should be maximized. Therefore, it is natural to utilize all the limited volume available. A CAD model of the silencer
presented (Fig. 2) illustrates an effective use of the space available underneath the motorcycle (see Fig. 1).
Page 4 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
Figure 2 -- A CAD model of the geometry available for the exhaust system.
Acoustically the silencer (see Fig. 3) has been designed to incorporate a combined noise cancellation principle, where the exhausting
pressure pulsations are reflected backwards to the source by the three sequential reactive expansion chambers (a, b and c) and
attenuated by the dissipative micro-perforated tubes (1 and 2). Both the engine cylinders are designed to exhaust via autonomous
primary pipes and tailpipes, which are coupled inside the silencer housing through the micro-perforation. In order to satisfy the
described noise limits the perforated tubes inside the silencer have been equipped with guide valves (see Fig. 4), positioned in the
middle of the largest expansion chamber. The function of the guide valves is to direct the pulsating exhausted gas flow through the
micro-perforated elements before terminating from the tailpipe. The dissipative acoustic effect of the micro-perforated elements has
been found to be remarkably dependent on the viscous losses introduced inside the perforated apertures [6, 7 and 8] and the sound
dissipation typically improves in higher flow velocity conditions. The reduction of propagating sound pressure amplitudes can be
explained by the increase in acoustic losses originating from the flow induced vortex shedding in this region. Hereby, this technical
solution has acoustically and aerodynamically been proved to successfully enhance the noise cancellation while preserving
acceptable pressure drop (see results section).
Figure 3 -- A geometrical layout of silencer unit revealing the three sequential expansion chambers: 1, 2 -- micro-perforated
pipes and a, b, c -- expansion chambers.
Page 5 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
TECHNICAL DESCRIPTION
The technical layout of the silencer incorporating three sequential expansion chambers and micro-perforated tubes is described in
Figs. 2-4 and the characteristic data of the micro-perforated tubular elements is presented in Table 3.
253
47
76
62
0
3
1
Figure 4 -- A side view of the silencer unit exhibiting removable guide valve position (green line) and illustrating the gas flow
path along the ducts and through the micro-perforated walls (dashed red arrows).
Table 3 -- Characteristic data of the micro-perforation used in the silencer.
Photo
Aperture
sketch
Porosity
PA ratio
10%
13.4
TECHNOLOGY AND MANUFACTURING
An overview of the materials and technological procedures used to manufacture the silencer system components is composed in
Table 2.
Page 6 of 12
0.1520 Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
Table 4 -- Material specifications and technological procedures used for the manufacturing of the silencer components.
Silencer
component
Quantity Parts, material
Technological
procedures
Pipe with a bend, stainless
steel AISI316
(45x1.25-R100)
conical element, stainless
steel AISI316
Cut,
machined,
welded*
(D45-D38x1.25-L44.1)
2 flanges, V-clamp
Pipe with a bend, stainless
steel AISI316
(38x1.25-R57)
Pipe, stainless steel
AISI316 (38x1.25)
Laser
perforated
Pipe, stainless steel
AISI316 (38x1.25)
Cut, welded*
Sheet metal, stainless steel
AISI304 (1.5mm)
Laser cut,
formed,
welded*
Inlet ducts 2
Perforated
duct
Outlet
ducts
Silencer
housing
2
2
1
Baffles
3
Sheet metal, stainless steel
AISI304 (1.5mm)
Laser cut
Guide
valve
2
Rod, stainless steel
AISI316 (D40)
Machined
*TIG welding
EXPERIMENTS
In order to evaluate the overall performance of the silencer a variety of common engineering parameters were determined including
acoustic transmission loss spectra, the vehicle in motion noise emission level and the aerodynamic pressure loss (see the results
section). A symmetrical geometry of the silencer unit (see Fig. 3) is expected to offer equal TL and PD characteristics for both
exhaust passages (sides). To simplify the experimental procedures, according to the assumption, the characteristic data were
determined only for one side of the silencer.
Page 7 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
For the acoustic characterization the silencer unit was treated as an acoustic two-port [9]. The acoustic power was determined at the
inlet and outlet cross-section of the silencer by using the classical two-microphone wave decomposition method [10]. The TL as the
measure of the acoustic power reduction across the silencer was calculated [11]:
TL = 10 log (
𝑊in
𝑊out
) =
= 10 log [
𝐴in ∙ 𝜌out ∙ 𝑐out ∙ (1 + 𝑀in)2
𝑇2 ∙ 𝐴out ∙ 𝜌in ∙ 𝑐in ∙ (1 + 𝑀out)2], (1)
where W denotes the acoustic power, A is the cross sectional area of the flow-duct, ρ is the density of the flowing media, c is the
speed of sound, M is a Mach number and T is the transmission coefficient [11]. The subscripts denote the respective sides of the
two-port. The transmission coefficient T originates from the scattering matrix formulation [10] of the two-port and is defined as the
ratio between incident and transmitted complex acoustic wave amplitudes. In [5] a detailed description about the determination of
the transmission coefficient T has been treated by the authors together with the overview of the aero-acoustic hot flow facility used
for the experimental investigations in this paper.
The maximum by-pass (vehicle in motion) noise level was measured following the Directive 97/24/EC [3] and the aerodynamic PD
was obtained by measuring the static pressure difference across the inlet and outlet of the silencer.
MODELLING
The 1D analysis of the muffler has been performed by implementing commercial software packages Gamma Technologies GT-
PowerTM [12] and SIDLABTM [13]. GT-PowerTM is a widely employed tool for the analyses of internal combustion engines, and the
SIDLABTM is a specific tool for 1D acoustic simulations focusing on flow duct applications. In GT-PowerTM the 1D non-linear flow
equations of mass, momentum and energy conservation [14] are solved in the time domain for each component of the complete
engine (air box, intake, exhaust system, muffler, etc.).
In the specific case of a silencer, the chambers are schematized by sets of quasi-3D elements whose ports are oriented along the
three spatial directions. In this way the radial interactions between the perforated tubes and the chambers are accounted. Straight
tubes are represented by 1D elements. Eventually, perforated tubes are modeled by a set of quasi-3D elements connected to each
other along the direction of the tube and linked, through a proper number of orifices, to the quasi-3D elements constituting the
chambers (see Fig. 5).
The alternative software package SIDLAB uses a 1D linear frequency domain approach, which is very fast and especially suitable
for flow duct acoustic applications. The computations are based on the well-known acoustic plane wave equation [15]. In GT-
PowerTM the porosity is accounted considering a respective friction coefficient in the energy equation. Whereas in Sidlab the
perforated tubes are divided into a number of discrete lumped impedance two-port elements which are separated by hard segments
on both sides [16]. The two-microphone random-excitation technique proposed by Seybert and Ross [17] is used in GT-PowerTM in
order to calculate the transmission loss of the silencer.
Figure 5 -- The two-microphone technique virtually implemented for the TL analysis in 1D GTPower simulations.
Page 8 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
RESULTS
EXPERIMENTAL RESULTS
Experimentally determined results for the acoustic transmission loss spectra with and without the guide valves, measured in the
presence of 40m/s room temperature mean flow, are presented in Fig. 7. The aerodynamic pressure loss characteristics measured in
0-40m/s flow velocity range for both the silencer settings are shown in Fig. 8. The respective output torque curves measured in
rolling road test facility for the motorcycle engine equipped with the silencer are exhibited in Fig. 9.
Figure 6 -- Transmission loss spectrums of the silencer measured in the presence of 40m/s mean flow velocity and presented for
straight flow (grey dashed line) and high attenuation (green solid line) setup.
Figure 7 -- Experimentally determined pressure drop of the silencer presented for straight flow (grey dashed line) and high
attenuation (green solid line) setup.
Page 9 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
Figure 8 -- The output torque curves measured in rolling road test facility for the motorcycle engine equipped with the silencer
in straight flow (grey dashed line) and high attenuation (green solid line) setup.
MODELLING RESULTS
In Fig. 9 the 1-D modeling results by using GT-PowerTM and SIDLABTM for the acoustic TL of the silencer in straight flow
configuration (in the absence of the guide valves) are compared to the experimental ones in no-flow conditions.
It is clearly noticeable that the entire plane wave region shown is well represented by the 1D simulation. The result demonstrates
that the use of the quasi 3-D elements for the chamber modeling allows to provide reliable results almost up to 1000 Hz.
Figure 9 -- The acoustic TL of the silencer in straight flow configuration and in the absence of mean flow: experimental results
(red solid line), 1D simulation results with GT-PowerTM (blue dashed line) and SIDLAB (black dots).
A comparison of the TL results for 40m/s mean flow velocity through the silencer in straight flow configuration (Fig. 10) and in
case of the extra attenuation guide valves (Fig. 11) is presented. Due to the viscous effects generated by the flow passing the
apertures, both the experimental and the numerical results obtained by GT-PowerTM demonstrate an enhancement of the attenuation
performance of the silencer at low frequencies. This effect is even more pronounced when the guide valves are equipped (see Fig.
11).
Page 10 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
Figure 10 -- The acoustic TL of the silencer in straight flow configuration and in the 40m/s mean flow condition:
experimental results (magenta solid line), 1D simulation results with GT-PowerTM (orange dashed line).
Figure 11 -- The acoustic TL of the silencer equipped with the guide valves and in the 40m/s mean flow condition:
experimental results (magenta solid line), 1D simulation results with GT-PowerTM (orange dashed line).
The PD calculated by GT-PowerTM simulations for two silencer configurations are 31mbar and 135 mbar. The simulated PD values
agree well with the experimentally determined ones: 35mbar and 118mbar respectively (see Fig. 7). The relatively small difference
between the simulated and measured PD of the silencer indicates well-captured viscous losses that occur in the presence of mean
flow.
CONCLUSIONS
A novel design for a cruiser type motorcycle silencer incorporating custom micro-perforated elements has been presented in this
paper. The silencer represents an effective engine noise cancellation solution in constricted conditions without the implementation
of the traditional fibrous materials.
The performance of the silencer as well as the engine output characteristics are presented for two different silencer configurations.
The numerical 1-D simulation models developed for the optimization procedures have exhibited a good agreement with the
experimentally determined data.
While maintaining acceptable pressure drop characteristics and providing pleasant engine sound, the silencer has proven to comply
with standard noise criteria.
Page 11 of 12
Preprint of the paper: Kabral, R.; Rämmal, H.; Auriemma, F.; Luppin, J.; Koiv, R.; Tiikoja, H.; Lavrentjev, J. "A Novel Design for
Cruiser Type Motorcycle Silencer Based on Micro-Perforated Elements".
SAE Technical Paper 2012-32-0109, 2012. https://doi.org/10.4271/2012-32-0109
REFERENCES
1. http://www.renardmotorcycles.com.
2. Directive 2002/24/EC of the European Parliament and of the Council of 18 March 2002 relating to the type-approval of two or
three-wheel motor vehicles and repealing Council Directive 92/61/EEC, Official Journal of the European Communities L124
(2002).
3. Directive 97/24/EC of the European Parliament and of the Council of 17 June 1997 on certain components and characteristics
of two or three-wheel motor vehicles, Official Journal of the European Communities (1997).
4. J. Lavrentjev, H. Rämmal, H. Tiikoja, The passive acoustic effect of automotive catalytic converters, SAE Technical Paper
2011-24-0219 (2011).
5. R. Kabral, H. Rämmal, J. Lavrentjev, F. Auriemma, Acoustic Studies on Small Engine Silencer Elements, SAE Technical Paper
20119514 (2011).
6. R. Kabral, H. Rämmal, J. Lavrentjev, Acoustic Studies of Micro-Perforates for Small Engine Silencers, SAE Technical Paper
2012320030 (2012).
7. D.Y. Maa, Theory and design of Microperforated-panel sound-absorbing construction. Sci Sin, XVIII (1975), pp. 55 -- 71.
8. S. Allam, M. Abom, A New Type of Muffler Based on Microperforated Tubes, J. Vib. Acoust., (2011), 133 (3), 8 pp.
9. H. Bodén, M. Åbom, Modeling of fluid machines as sources of sound in duct and pipe systems, Acta Acustica 3 (1995) 549-
560.
10. M. Åbom, Measurement of the scattering-matrix of acoustical two-ports, Mechanical Systems and Signal Processing 5 (1991).
11. M. Åbom, An Introduction to Flow Acoustics, 164 (2004).
12. Gamma Technologies, GT-SUITE Flow Theory Manual Version 7.0, the GT-SUITE Interactive Simulation Environment
(2009).
13. SIDLAB Acoustics User Manual, version 2.6.
14. D.E. Winterbone, R.J. Pearson, Theory of engine manifold design, Wave action methods for IC engines, Professional
engineering publishing (2000).
15. H.P. Wallin, U. Carlsson, M. Abom, H. Boden, Sound and vibration, KTH MWL, ISBN 978-91-7415-553-2.
16. T. Elnady, M. Abom, S. Allam, Modelling perforates in muffler using two ports, Journal of Vibration and Acoustics, December
Vol. 132 / 061010-3.
17. A.F. Seybert and D.F. Ross, Experimental determination of acoustic properties using a two-microphone random-excitation
technique, Journal of the Acoustical Society of America, 61, 1362-1370, 1977.
CONTACT INFORMATION
Raimo Kabral: The Marcus Wallenberg Laboratory, The Royal Institute of Technology, Teknikringen 8, Stockholm, SE-10044,
Sweden, [email protected], phone: +372 50 24 992;
Dr. Hans Rämmal: Department of Automotive Engineering, Tallinn University of Technology, Ehitajate tee 5, Tallinn, 19086,
Estonia, [email protected], phone: +372 56 465 738;
ACKNOWLEDGMENTS
The authors would like to acknowledge technical consultancy companies Lettore and Triple Seven for successful co-operation
including financial, technical and technological support. In addition, the support from Estonian Science Foundation (Grant No.
7913), Marie Curie European doctoral student program and "DoRa" program of Archimedes Foundation are acknowledged.
DEFINITIONS/ABBREVIATIONS
micro-perforated
transmission loss
perimeter to area ratio
pressure drop
rotations per minute
MP
TL
PA Ratio
PD
RPM
Page 12 of 12
|
1803.06275 | 1 | 1803 | 2018-03-16T15:41:53 | Growth monitoring with sub-monolayer sensitivity via real time thermal conductance measurements | [
"physics.app-ph"
] | Growth monitoring during the early stages of film formation is of prime importance to understand the growth process, the microstructure and thus the overall layer properties. In this work, we demonstrate that phonons can be used as sensitive probes to monitor real time evolution of film microstructure during growth, from incipient clustering to continuous film formation. For that purpose, a silicon nitride membrane-based sensor has been fabricated to measure in-plane thermal conductivity of thin film samples. Operating with the 3{\omega}-V\"olklein method at low frequencies, the sensor shows an exceptional resolution down to {\Delta}({\kappa}*t)=0.065 nm*W/(m*K), enabling accurate measurements. Validation of the sensor performance is done with organic and metallic thin films. In both cases, at early stages of growth, we observe an initial reduction of the effective thermal conductance of the supporting amorphous membrane, K, related with the surface phonon scattering enhanced by the incipient nanoclusters formation. As clusters develop, K reaches a minimum at the percolation threshold. Subsequent island percolation produces a sharp increase of the conductance and once the surface coverage is completed K increases linearly with thickness The thermal conductivity of the deposited films is obtained from the variation of K with thickness. | physics.app-ph | physics | Growth monitoring with sub-monolayer sensitivity via real
time thermal conductance measurements.
P. Ferrando-Villalba1, D. Takegami1, Ll. Abad2, J. Ràfols-Ribé1, A. Lopeandía1*, G.
Garcia1, J. Rodríguez-Viejo1
1Grup de Nanomaterials i Microsistemes, Dep. Física, Universitat Autònoma de
Barcelona, 08193 Bellaterra, Spain.
2 Institut de Microelectrònica de Barcelona- Centro Nacional de Microelectrònica − CSIC,
Cerdanyola del Vallès, 08193, Spain
Abstract
Growth monitoring during the early stages of film formation is of prime importance to
understand the growth process, the microstructure and thus the overall layer properties.
In this work, we demonstrate that phonons can be used as sensitive probes to monitor real
time evolution of film microstructure during growth, from incipient clustering to
continuous film formation. For that purpose, a silicon nitride membrane-based sensor has
been fabricated to measure in-plane thermal conductivity of thin film samples. Operating
with the 3𝜔-Völklein method at low frequencies, the sensor shows an exceptional
resolution down to 𝛥(𝜅 · 𝑡) = 0.065
𝑊
𝑚·𝐾
· 𝑛𝑚, enabling accurate measurements.
Validation of the sensor performance is done with organic and metallic thin films. In both
cases, at early stages of growth, we observe an initial reduction of the effective thermal
conductance of the supporting amorphous membrane, 𝐾, related with the surface phonon
scattering enhanced by the incipient nanoclusters formation. As clusters develop, 𝐾
reaches a minimum at the percolation threshold. Subsequent island percolation produces
a sharp increase of the conductance and once the surface coverage is completed 𝐾
increases linearly with thickness The thermal conductivity of the deposited films is
obtained from the variation of 𝐾 with thickness.
email: [email protected]
Introduction
Monitoring the first stages of thin film growth is of key importance to understand and
thus tune the properties of grown layers. Critical microstructure features such as grain
size, morphology, crystal orientation, nature of grain boundaries and surface morphology,
are defined during the early growth process. Real-time measurements have proven their
potentiality to understand the growth dynamics, either for thin films or nanoparticles
deposited on surfaces. Actually, in-situ diagnostics during growth with monolayer
sensitivity have already been performed by a variety of techniques such as wafer
curvature measurements mapping
the stress evolution1, ellipsometry2, X-ray
reflectivity3,4 and resistance-based measurements. Low or medium energy electron
diffraction (LEED, MEED) are also reliable tools to monitor 2D ordering during epitaxial
growth5,6. Among all, electrical measurements are very powerful since i) the electrical
resistance in metallic thin films7 may vary orders of magnitude above the percolation
threshold , and ii) drastic electrical conductivity changes during the initial growth stages
can identify phase transformations, such as amorphous-to-crystal transition in Mo films8.
Unfortunately, although simple and accessible, this approach is limited to metallic or
highly conductive layers, precluding the analysis of organic or insulating materials. In
contrast, phonons are a more generic probe, extremely sensitive to film structure thanks
to their larger mean free path compared to electrons. However, real-time thermal
conductance measurements during film growth are much scarcer, mainly due to the
technical challenges associated.
Additionally, the potential application of nanomaterials, thin films and nanostructures in
heat management and efficient thermoelectric devices has boosted the necessity to
perform accurate the thermal conductivity measurements at the nanoscale. In particular,
phonon engineering in low-dimensional materials has appeared as the most effective
approach to enhance the thermoelectric figure of merit9–11 through the reduction of
thermal conductivity. The implementation of novel nanomaterials designs has encouraged
the development of new sensors and methodologies, enabling accurate determination of
thermal conductivity in low dimensional architectures. Whether based on optical12 or
electrical13 signals, these novel thermal sensors and associated methodologies have
allowed in-plane14,15 and out-of-plane16,17 thermal conductivity measurements, of
nanowires and thin films, with outstanding nanometre spatial resolution18,19. A
remarkable contribution to the field was achieved by Völklein et al. in 1990 who
developed a suspended membrane-based sensor using a long and thin Pt electrical
transductor operated in DC to measure in-plane thermal conductivity of thin films15. More
recently, Sikora et al. went a step further in improving this technology by combining the
Völklein geometry with the AC 3𝜔-method, reaching exceptional thermal conductance
sensitivity,
Δ𝐾
𝐾
≅ 10−320,21.
In parallel, ex-situ thermal probe studies performed on thin films22,23 have shown that the
thermal conductivity of a grown material is conditioned by the thermal conductance loss
of the substrate induced both by interfacial scattering of phonons in the in-plane
measurements22 and by thermal boundary resistance in the out-of-plane measurements23.
Accordingly, the thermal conductivity of the thin layer cannot be simply calculated via
the differential measurement of the thermal conductance of the whole sample (film +
membrane/substrate) and a reference (only membrane/substrate), but must be calculated
using a set of thermal conductance measurements performed at different film thicknesses.
Up to date, most of these measurements have only been performed ex-situ, evaluating the
temperature dependence of the thermal conductivity 𝑘 for each singular thickness. To our
knowledge, real-time studies during growth at early stages taking into account the impact
of the microstructure on phonon scattering have not been previously reported. It is worth
noting that although Völklein and Starz24 already demonstrated in 1997, that a Völklein-
like sensor, operating in DC mode, could perform in-situ measurement of thermal
conductance in thin films, those measurements were limited to metallic films thicker than
1 µm, on the contrary to the sensor presented here.
In this paper we present i) an improved 3𝜔-Völklein technique allowing highly sensitive
conductance measurements (Δ𝐾/𝐾 < 10−3), ii) the sensor optimization through FEM
(Finite Element Modelling) thermal analysis, iii) a novel study of the thermal evolution
of the sensor during growth and iv) real-time measurements of the thermal conductance
during growth of both highly insulating and conductive (electrically and thermally)
materials with sub-monolayer sensitivity. Proof-of-concept is achieved by analysing two
different materials: a metal, such as In (indium), and an organic conductor, N,N′-Bis(3-
methylphenyl)-N,N′-diphenylbenzidine (TPD), often used as hole injector in OLEDs.
Complementary characterization techniques (SEM, AFM and electrical measurements)
are used to correlate conductance features with sample morphology throughout the
growth.
Experimental Section
Sensor Design and Simulation
The sensor developed here consists in a long and thin Pt line deposited on a suspended
SiNx membrane and connected in a 4-wire configuration (Fig. 1a,b), along with an
external Pt line. The thermal conductance of the whole membrane is determined with the
centred Pt line (normal operation), while both lines, centred and external, are needed to
measure exclusively the thermal conductance of the membrane volume portion between
them. Details of the sensor microfabrication can be found in the Supplementary
Information (SI), in Fig. S1.
In normal operation, the thermal conductance from the central Pt line to the substrate is
calculated by using the 1D Fourier law and assuming that the external line is much more
conductive than the SiNx beneath, which yields equation 1
𝐾 = 𝑘𝑆𝑖𝑁𝑥𝑡𝑆𝑖𝑁𝑥 (
𝐿
𝑙
+
𝐿
𝑙−𝑤
)
(1)
where 𝑡𝑆𝑖𝑁𝑥 and 𝑘𝑆𝑖𝑁𝑥 are the thickness and the thermal conductivity of the silicon nitride
respectively, 𝐿 is the length of the Pt strip between the voltage probes, 𝑤 = 4.5 μm is the
width of the Pt strip and 𝑙 = 0.002m is the distance between the Pt strip and the substrate.
Fig. 1a) Scheme of the sensor (side and top views), consisting on a suspended membrane with a
Pt strip on it. b) Optical micrograph of the sensor. c) Longitudinal and transversal temperature
distributions of the sensor in a FEM DC study. d) Surface temperature distribution of the sensor
(brighter is hotter) in a FEM DC study.
We have modelled several sensor geometries using Finite Element Model (FEM) with
COMSOL software in order to ensure temperature homogeneity between the voltage
probes of the central strip. In the first place, we have introduced voltage probes to measure
the temperature oscillations of the central part of the strip, which has a flatter temperature
profile than the borders. This simulation reproduces a steady state by feeding the Pt strip
with a DC current. The optimum sensor (considering the resolution limits of the available
photolithographic system and the structural stability of the SiNx membrane) consists in a
long and narrow SiNx membrane (3mm x 250 m x 180 nm) that supports the two Pt
sensing strips (3mm x 5 m x 100 nm) with voltage probes separated by 2 mm. A key
point in this design is the close proximity of the voltage probes and the central Pt strip,
minimizing heat leakage and any temperature drop along the strip. The simulated profile
of the longitudinal temperature (Fig. 1c) shows a large and flat central plateau where the
inhomogeneity is lower than 3% of the total temperature rise. We notice two tiny
temperature depressions at ±1000 m, coinciding with the voltage probe locations. The
transversal profile exhibits a constant slope only perturbed by minimal flat segments
a)b)-2000-1000010002000292294296298300 Temperature (K)Longitudinal position (m)-1000100Transversal position (m)c)d) corresponding with the position of the Pt strips. With this sensor, the thermal conductivity
of the SiNx layer can be measured with an accuracy of 0.1% in the absence of other
uncertainty sources. Main parameters used and more information about the modelling can
be found in the SI (Fig. S2, S3).
Thermal analysis during film deposition
Based on Sikora et al. developments20,21, we deduced a mathematical expression to extract
the thermal conductivity from the measured thermal conductance at different thicknesses.
When the central Pt line is fed with an oscillating current 𝐼 = 𝐼0 sin 𝜔𝑡, the amplitude of
the temperature oscillations induced can be calculated with the 1D heat equation
(assuming that the sensor is infinitely long), which after some approximations (See SI)
yields:
∆𝑇2𝜔 =
𝑃0
𝐾√1+𝜔2(4𝜏2+
2𝑙4
3𝐷2+
4𝜏𝑙2
3𝐷
(2)
)
where 𝑃0 =
2𝑅
𝐼𝑜
2
, 𝐷 =
𝑘𝑆𝑖𝑁𝑥
𝜌𝑆𝑖𝑁𝑥·𝑐𝑆𝑖𝑁𝑥
, 𝜏 =
𝐶𝑝
𝐾
and 𝐶𝑝 = 𝑤𝐿(𝜌𝑆𝑖𝑁𝑥 · 𝑐𝑆𝑖𝑁𝑥 · 𝑡𝑆𝑖𝑁𝑥 + 𝜌𝑃𝑡 · 𝑐𝑃𝑡 ·
𝑡𝑃𝑡). Here 𝜌𝑆𝑖𝑁𝑥,𝑃𝑡 is the density of the SiNx membrane and the Pt strip, 𝑐𝑆𝑖𝑁𝑥,𝑃𝑡 is their
specific heat capacity, 𝐶𝑝 is the heat capacity of the strip and the portion of SiNx beneath
it, 𝜏 is the thermal time constant of the sensor, 𝐷 is the thermal diffusivity of SiNx, 𝑅 is
the Pt strip electrical resistance between the voltage probes and 𝐼0 is the current amplitude.
From equation (2), the apparent thermal conductance can be calculated as 𝐾2𝜔 =
𝑃0/∆𝑇2𝜔, which resembles 𝐾 at low frequencies.
When a thin-film sample grows on the SiNx membrane, the parameters in equation (2)
may vary as 𝑘, 𝑐 and 𝜌 will no longer correspond solely to the SiNx membrane but to the
combination of the deposited film (from now on called "sample") and the substrate
membrane. In this case, an effective value for these magnitudes can be calculated
(assuming a vertical growth of the film) by pondering the different values as a function
of the film thickness, 𝑡𝑠𝑚𝑝:
𝑘𝑒𝑓𝑓 =
𝑘𝑆𝑖𝑁𝑥·𝑡𝑆𝑖𝑁𝑥+𝑘𝑠𝑚𝑝·𝑡𝑠𝑚𝑝
𝑡𝑆𝑖𝑁𝑥+𝑡𝑠𝑚𝑝
(4)
𝑐𝑒𝑓𝑓 =
𝑐𝑆𝑖𝑁𝑥·𝑡𝑆𝑖𝑁𝑥+𝑐𝑠𝑚𝑝·𝑡𝑠𝑚𝑝
𝑡𝑆𝑖𝑁𝑥+𝑡𝑠𝑚𝑝
(5)
𝜌𝑒𝑓𝑓 =
𝜌𝑆𝑖𝑁𝑥·𝑡𝑆𝑖𝑁𝑥+𝜌𝑠𝑚𝑝·𝑡𝑠𝑚𝑝
𝑡𝑆𝑖𝑁𝑥+𝑡𝑠𝑚𝑝
(6)
𝐷𝑒𝑓𝑓 =
𝑘𝑒𝑓𝑓
𝜌𝑒𝑓𝑓∙𝑐𝑒𝑓𝑓
(7)
Here, 𝑘𝑠𝑚𝑝, 𝑐𝑠𝑚𝑝 and 𝜌𝑠𝑚𝑝 are the sample thermal conductivity, specific heat capacity
and density. In the same way, the extrinsic values 𝐾 and 𝐶𝑝 will vary, as well as the
thermal time constant 𝜏, as:
𝐾(𝑡𝑠𝑚𝑝) = (𝑘𝑆𝑖𝑁𝑥 ∙ 𝑡𝑆𝑖𝑁𝑥 + 𝑘𝑠𝑚𝑝 ∙ 𝑡𝑠𝑚𝑝) (
𝐿
𝑙
+
𝐿
𝑙−𝑤
)
(8)
𝐶𝑝(𝑡𝑠𝑚𝑝) = 𝑤𝐿(𝜌𝑆𝑖𝑁𝑥 ∙ 𝑐𝑆𝑖𝑁𝑥 ∙ 𝑡𝑆𝑖𝑁𝑥 + 𝜌𝑃𝑡 ∙ 𝑐𝑃𝑡 ∙ 𝑡𝑃𝑡 + 𝜌𝑠𝑚𝑝 ∙ 𝑐𝑠𝑚𝑝 ∙ 𝑡𝑠𝑚𝑝)
(9)
𝜏′ =
𝐶𝑝(𝑡𝑠𝑚𝑝)
𝐾(𝑡𝑠𝑚𝑝)
(10)
If the measuring frequency is low (𝜔 ≪
3𝐷𝑒𝑓𝑓
8𝑙2 ), then the measured 𝐾2𝜔 resembles 𝐾 and
the derivative of equation (8) can be used to calculate the thermal conductivity of the
sample film by measuring in real-time the thermal conductance during growth
(𝐾2𝜔(𝑡𝑠𝑚𝑝)), as shown in equation (11).
𝑑𝐾
𝑑𝑡𝑠𝑚𝑝
= 𝑘𝑠𝑚𝑝 (
𝐿
𝑙
+
𝐿
𝑙−𝑤
) → 𝑘𝑠𝑚𝑝 =
𝑑𝐾
𝑑𝑡𝑠𝑚𝑝
𝐿
𝐿
+
𝑙
𝑙−𝑤
(
)
(11)
However, if a higher current angular frequency 𝜔 is used, 𝑘𝑠𝑚𝑝 can be extracted by fitting
the measured 𝐾2𝜔with equation (12) using the thickness-dependent parameters (equations
(7) - (10)):
𝐾2𝜔(𝑡𝑠𝑚𝑝) = 𝐾(𝑡𝑠𝑚𝑝)√1 + 𝜔2 (4𝜏′2 +
2𝑙4
2 +
3𝐷𝑒𝑓𝑓
4𝜏′𝑙2
3𝐷𝑒𝑓𝑓
)
(12)
In Fig. 2a the calculated 𝐾2𝜔 is plotted for current frequencies of 0 Hz (DC), 1 Hz and 3
Hz (the parameters used are listed in the figure caption). If the conductance is monitored
with a current at 1 Hz, the apparent thermal conductance 𝐾2𝜔 is very similar to 𝐾
throughout the deposition. However, at 3 Hz, there is an evident difference in the slope
of the curves: Although the absolute value of the apparent thermal conductance only
varies from 97% to 93% of 𝐾 (Fig.2b), the slope of 𝐾2𝜔 is up to 40% higher than the one
of 𝐾(Fig.2a). Generally, measuring with higher frequencies increases the dependence of
𝐾2𝜔with intrinsic properties of the sample that may not be well known (like 𝑐𝑠𝑚𝑝 or
𝜌𝑠𝑚𝑝), hindering the determination of 𝑘𝑠𝑚𝑝.
Fig.2 a) Calculated real thermal conductance (black) and apparent thermal conductance modelled
at 1 Hz (red) and 3 Hz (blue) during the deposition of a layer. b) Calculated frequency dependence
of the temperature oscillations in the sensor before and after depositing a 250 nm film (the inset
is a zoom out).The material properties used in both plots are the following: 𝑘𝑆𝑖𝑁𝑥 = 2.65Wm-1K-
1,𝑐𝑆𝑖𝑁𝑥 = 0.7 J·K-1·Kg-1, 𝜌𝑆𝑖𝑁𝑥 = 3.18g/cm3, 𝑘𝑃𝑡 = 33Wm-1K-1, 𝑐𝑃𝑡 = 0.133J·K-1·Kg-1, 𝜌𝑃𝑡 =
21.45 g/cm3, 𝑘𝑠𝑚𝑝 = 0.21Wm-1K-1, 𝑐𝑠𝑚𝑝 = 1.05J·K-1·Kg-1and𝜌𝑠𝑚𝑝 = 1.08 g/cm3.Structural
parameters are𝑡𝑆𝑖𝑁𝑥 = 180 nm, 𝑡𝑃𝑡 = 110 nm,𝑙 = 123m,𝑤 = 5m and 𝐿 = 2000 m.
Experimental setup
The sensor is introduced in a high vacuum chamber equipped with an effusion cell that
enables a good control of the evaporation rate. A previously calibrated quartz crystal
microbalance is located nearby the sensor to monitor the layer growth rate with a
precision of 0.01 Å/s. The temperature of the sample was controlled with a custom-made
PID system that reads the temperature of a Pt100 and provides heat through a Kapton
heater, yielding a temperature control with fluctuations smaller than 0.003K, from 77 K
up to 400 K.
The experiment is performed by feeding two sensors (sample and reference) with a
current wave of a given amplitude and frequency, generating a voltage drop in each
sensor. The voltage signals from both sensors, as well as the differential voltage between
123450.900.920.940.960.981.000.05.0x10-81.0x10-71.5x10-72.0x10-72.5x10-715161718T2/TDC Frequency (Hz) Before deposition After deposition110100100001b) T2/TDCFrequency (Hz)a) Thermal conductance (W/K)Sample thickness (m) Real conductance Apparent conductance 1 Hz Apparent conductance 3 Hz them, were subtracted using the low-noise amplifiers with gains 𝐺𝑠𝑚𝑝, 𝐺𝑟𝑒𝑓 and 𝐺𝑑𝑖𝑓𝑓,
respectively. The reference sensor is a twin (equal to the sample one) non-suspended
sensor which produces no self-heating, and is used to subtract the 1𝜔 component of the
sample sensor. Thus, the 3𝜔 component of the differential voltage is only produced in
the sample sensor, and owing to the cancellation of the 1𝜔 voltage, it could be amplified
with a gain 𝐺𝑑𝑖𝑓𝑓 = 75. The main benefit of using a twin sensor (instead of a variable
resistance) as a reference for the differential measurement is that, if the temperature of
the sample holder varies, the resistance of both sensors will change hand-in-hand, making
unnecessary to build a control system for the cancellation of the 1𝜔 voltage. The exact
electronics used in the measurement of the different voltage signals are detailed in the SI
(Fig. S3).
From the measured voltage signals, the resistance 𝑅𝑠𝑚𝑝 and the temperature oscillations
Δ𝑇2𝜔 can be calculated as:
𝑅𝑠𝑚𝑝 =
𝑉1𝜔
𝐼0∙𝐺𝑠𝑚𝑝
(13)
Δ𝑇2𝜔 =
2𝑉3𝜔
𝐼0
𝑑𝑅𝑠𝑚𝑝
𝑑𝑇
∙𝐺𝑑𝑖𝑓𝑓∙𝐺𝑠𝑚𝑝
(14)
where 𝑉1𝜔 and 𝑉3𝜔are the 1𝜔and 3𝜔voltage components measured in the sample sensor,
and
𝑑𝑅𝑠𝑚𝑝
𝑑𝑇
is the slope of the sample resistance as a function of the temperature.
Results
Sensor performance test
Initially, the self-heating of the Pt sensor is determined by measuring both the 3𝜔 voltage
and the variation of the 1𝜔 voltage with 𝜔 = 2𝜋 𝑟𝑎𝑑/𝑠 (Fig.3a). Since the frequency is
very low, both signals yield identical self-heating, but as can be seen in Fig. 3a, the self-
heating calculated with the 3𝜔 voltage (Δ𝑇2𝜔 )is less noisy than the one calculated with
the 1𝜔 voltage (Δ𝑇𝐷𝐶). Also, the slope of log Δ𝑇2𝜔 vs log 𝐼0 has a value very close to 2,
demonstrating that the self-heating depends on the square of the current.
Δ𝑇2𝜔 has been measured in a wide frequency range (1 Hz to 2000 Hz) and compared with
finite element simulations in a time-dependent study (See SI), as can be seen in Fig. 3b.
The high coincidence between both datasets suggests that the behaviour of the sensor is
purely driven by heat transport physics. This is an important difference from the device
presented by Sikora et al.20, where the use of a NbN strip sensor allowed measurements
at very low temperatures, but produced non-negligible electrical effects due to the high
electrical impedance of that material.
Fig.3a) Self-heating of the sensor measured with the 3𝜔 and 1𝜔 voltages (𝜔 = 1Hz). b)
Temperature oscillations as a function of the frequency. The measured values are compared with
the simulated ones using the COMSOL model. c) Measurement of the uncertainty in the thermal
conductance Δ𝐾 refers to the standard deviation of the data. Red circles and black squares
correspond to data obtained with input currents of 500 and 300 µA respectively, with 𝜔 = 1Hz.
Finally, as shown in Fig.3c, the uncertainty of the conductance measurement has been
determined at two different current intensities, 300 A and 500 A, which generated
temperature amplitudes of about 2 K and 6 K and produced a standard deviation of data
equal to 8 nW/K and 1.3 nW/K respectively. This accuracy in the thermal conductance
provides an extremely high resolution in the product of the thermal conductivity and the
thickness, Δ(𝑘 · 𝑡) = 0.065
W
m·K
· nm. Although there is a decrease in the data dispersion
for 500A, there is also a noticeable reduction in the average value of the thermal
10-410-30.1110TDC (K)T2 (K)T (K) Current (A)0.11101001000100000.00.20.40.60.81.0T2/TDC Frequency (Hz) Measured COMSOL Model16.84µ16.86µ16.88µ16.90µ16.92µ16.94µG = 1.3 nW/K I0= 300 A I0= 500 AG2 (W/K)# measurementG = 8 nW/Ka)b)c) conductance, from 16.91 W/K to 16.85 W/K that may be justified by a small reduction
of the sensor dR/dT due to the higher average heater temperature.
The thermal conductivity of the SiNx membrane was evaluated for temperatures between
80 K and 230 K, as shown in Fig. 4 (black squares). The results are very similar to the
values found in the out-of-plane direction using the 3𝜔 method on a similar SiNx
membrane (red circles in Fig. 4). The latter values were obtained through a differential
measurement of two samples with thicknesses of 180 nm and 450 nm, granting that the
thermal boundary resistances between the film and the substrate are cancelled. Thus, the
similarity of both the in-plane and the out-of-plane values confirm that there are not
substantial phonon size effects, or anisotropy, in our layers. In both cases, the variation
of 𝑘 with temperature show also a similar tendency than data presented by Sikora et al20,21
(continuous line in Figure 4). The discrepancy in the absolute values may account for
density or stoichiometry variations related to the different growth film characteristics in
each work. Huge differences in 𝑘 variation with temperature are observed when
comparing our data with Sultan et al.25 work (blue triangles in Fig.4). We suggest, as will
be discussed later, that this difference is probably related to fact that they use a
nanocrystalline membrane.
The measurements performed in this work have an extremely low variability owing to the
high sensitivity of the method. The main source of uncertainty in the in-plane
measurement is the precise determination of SiNx membrane thickness, which was
measured throughout the wafer with a contactless optical profilometer, yielding a
standard deviation of 1% of the 180 nm averaged thickness. However, wet etchings at the
final steps of microfabrication process could slightly reduce this thickness up to 10 nm,
inevitably increasing the uncertainty up to 5% .
Fig. 4 Thermal conductivity values measured at different temperatures with the 3𝜔-Völklein
Method (in-plane direction) and the 3𝜔 method (out-of-plane direction). Data from Sultan et al.25
and Sikora et al.21is also included for comparison.
Real-time measurement of thin-film growth: sensor proof of concept
Organic thin-film layers
We initially measure the thermal conductance of the membrane as a function of the TPD
thickness deposited at 266±2K, as shown in Fig.5a. The temperature uncertainty comes
from the thermal oscillations produced by the current wave of amplitude 300 A. The
evaporation was carried out below 10-7 mbar, by heating up an effusion cell up to 200ºC,
which yielded a stable deposition rate of 0.29nm/s. From the slope of the curve
𝐾2𝜔(𝑡𝑠𝑚𝑝) and applying equation (9), we calculate the thermal conductivity of the TPD
layer, resulting
in 𝑘𝑠𝑚𝑝 = (0.153 ± 0.001)Wm-1K-1. However,
if
the
thermal
conductivity is calculated by fitting equation (12) we obtain 𝑘𝑠𝑚𝑝 = (0.145 ± 0.001)
Wm-1K-1. As discussed previously both results are comparable due to the low frequency
used in the measurement. The low value of the thermal conductivity is an indication of
the glassy character of the TPD layers, as confirmed by calorimetric measurements
showing a clear signature of the glass transition temperature (SI, Fig.S6).
501001502002503003501.52.02.53.03.5 3 Völklein 3 Out-of-plane Sultan et al.25 Sikora et al.21k (Wm-1K-1)Temperature (K) As indicates Fig. 5a, at the early stages of the deposition, i.e. very low thicknesses, we
observe a decrease in the thermal conductance of around 1.2 % of its initial value, while
afterwards 𝐾2𝜔 roughly increases linearly with thickness. In this particular measurement,
film growth was stopped at 340 nm, yielding a constant value of the thermal conductance
after this particular point.
Fig.5b,c show in more detail the initial stages of the evolution of 𝐾2𝜔 versus thickness
for two TPD films deposited at 267 K and 304 K, both with a low growth rate of 0.02
nm/s. In both cases, we identify four different regions where thermal conductance follows
different trends with thickness. In region I, the overall conductance decreases abruptly
following an exponential behaviour, while it decreases slowly in Region II until a
minimum point is reached. Region III covers the thickness range where 𝐾2𝜔 increases
slowly up to a point where a linear regime of the conductance with thickness is attained
and labelled as Region IV. The extent of each of these regions is represented in Fig. 5b,c
by dashed red lines, and as is clearly seen, vary from one sample to another, confirming
the importance of deposition temperature as will be discussed later.
To understand and interpret the variation of 𝐾2𝜔 with thickness, we also performed ex-
situ analysis of the film morphology either by SEM or by AFM described in Fig. S4a of
the SI. Complementary electrical characterizations were also performed, as shown in Fig.
5b inset, to corroborate the hypothesis described later on. Surface morphology evolution
with thickness showed that TPD do not wet the SiNx membrane surface, as it grows in
3D islands in the early growth stages. For films deposited at 267K and according to the
conductance evolution, island coalescence seems to happen at a nominal film thickness
of 2.6 nm while complete surface coverage seems to take place at 14 nm. Ex-situ analysis
of growth regimes is precluded by the evidence of dewetting of TPD at temperatures well
below the glass transition temperature26. The fast film dynamics at temperatures below
the glass transition temperature makes in-situ analysis an indispensable tool to gain a
better understanding of film formation in molecular glasses.
Fig. 5 Thermal conductance vs thickness during TPD growth: a) 𝑇𝑑𝑒𝑝=267±2 K. Growth rate =
0.29 nm/s. b) 𝑇𝑑𝑒𝑝=267±2 K. Growth rate = 0.02 nm/s. c) 𝑇𝑑𝑒𝑝 = 304±2 K. Growth rate = 0.02
nm/s. In graphs b) and c) data is box averaged with 10 points/box. The main regions of film
growth are separated by dashed red lines. Region I located between 0 nm and the first vertical
line correspond to nucleation and isolated island formation; region II to island coalescence; region
III to percolation across the layer and region IV to vertical growth of a continuous layer. The
slight difference in the initial conductance between all the graphs is mainly due to the use of
different sensors/membranes for the experiments. The black downward arrow marks the
percolation threshold (separation between regions II and III). The inset of b) shows the abrupt
variation of the electrical conductance at the percolation threshold (nominal thickness of 2.6 nm)
of the TPD film. This value coincides with the minimum of the thermal conductance (black arrow)
in graph b).
Before relating the large drop observed in region I with film microstructure, we performed
few tests to rule out any potential experimental artefact that could have led to temperature
variations of the sensor. As extra heat originated by condensation of molecules on the
substrate is not modulated at angular frequency 𝜔 or any of its harmonics and therefore,
does not affect the 3𝜔 component of the voltage, the decrease of conductance cannot be
05010015020025030035015.215.616.016.416.804812162024283217.117.217.3K2 (W/K)K2 (W/K) K2 (W/K)Sample growinga)2.5 nmIVIVIIIIIIIc) 14 nm0.5 nmb)0102030405013.513.613.713.86 nmIIIII1.1 nm Thickness (nm)0123123 Thickness (nm)G (n-1)2.6 nm attributed to a sensor temperature variation due to this phenomenon. Also, we ruled out
any effect of the sample emissivity variations during TPD growth on the membrane, since
a measurement at 81 K, where radiation effects should be substantially lower, showed a
similar drop in thermal conductance (Fig. S5 in SI). The reduction of thermal conductance
observed in the early stages of growth seems thus to be due to phonon-related phenomena,
although covering the surface with TPD is not expected to affect the thermal conductance
of the SiNx membrane since it is already a disordered structure with an average phonon
mean free path of the order of the interatomic distance.
Nevertheless, a previous work by Sultan et al.22,25 already showed similar behaviour (𝐾
decrease at early growth stages with a subsequent increase for larger thicknesses) by
performing ex-situ measurements of thermal conductance versus thickness of metal and
alumina deposited on SiNx. Nonetheless, our data differs from this previous work in
several important aspects: i) the magnitude of the conductance drop is much lower and
ii), our measurements, are carried out in-situ with a much higher conductance sensitivity,
providing a much clear signature of the various growth regimes.
Our thermal conductance initial drop, although much higher than the uncertainty of the
measurement, only amounts 1.2%, compared to 5-10% in Sultan's work. These authors
estimated that 40%-50% of the total 𝐾 was due to long-wavelength with long mean free
path phonons. The estimated average λ-value for these phonons was around 4.5 nm, much
higher than thermal phonons at room-temperature that have λ~0.2nm. We thus believe
that the main difference between both results arises from a different microstructure of the
used SiNx in each case. The thermal conductivity value of our SiNx, slightly below Sultan
et al.22,25values combined with the temperature dependence shown in Fig. 4, clearly
indicates that our nitride is fully amorphous while the one used in Sultan et al work was
probably nanocrystalline. Therefore, a lower drop of the thermal conductance in fully
amorphous nitride membrane is consistent with a scenario where the contribution of long-
λ phonons is reduced with respect to previous nanocrystalline nitride membranes. Even
though, the conductance drop due to enhanced surface scattering requires that phonons
slightly larger than the average interatomic distance at room temperature have to be
involved in heat transport along the nitride membrane. We can thus tentatively attribute
the initial sharp reduction of the thermal conductance to the formation of TPD isolated
clusters on top of the nitride membrane surface, modifying the interfacial phonon
scattering and thus leading to a decrease of the thermal conductance.
Although we currently lack a complete understanding of the microscopic processes
occurring at the interfaces, we believe that the growth of new material on top of the
membrane changes the specularity of the surface, resulting on an effective increase in the
phonon scattering rate. We foresee that the future use of crystalline membranes, such as
single crystalline Si, will provide a convenient platform to investigate nucleation and
island growth dynamics during the early stages of film growth with even higher
conductance sensitivity.
In Region II the thermal conductance still decreases but with a lower slope. We believe
that in this region, clusters enlarge and start to coalesce, providing additional paths for
heat transfer which partially compensate the interface scattering until a minimum is
reached at the end of the present region. As island coalescence continues, percolation
builds up new channels across the layer structure (Region III), providing additional heat
flow paths that start to exceed the contribution of the interface scattering. The coincidence
of the minimum in thermal conductance with a percolation phenomenon threshold was
demonstrated in a separate experiment where electrical conductance was measured as a
function of thickness (inset in Fig. 5b), showing a sharp variation of the slope at 2.6 nm,
due to electrical continuity through the TPD film.
As percolation persists, the thermal conductance increases reaching a linear regime that
we identify with the formation of a continuous layer with complete coverage, marking
the onset of Region IV. Thus, Region IV corresponds to the vertical growth of the
continuous film. In this regime the increase in thermal conductance is linear and
proportional to the thickness of the growing layer. Compared to the end of region III,
there is a small reduction in the slope of the conductance vs thickness since the islands
are no longer forming new conductive channels.
Differences in growth dynamics with the deposition temperature were also studied, as
shown in Fig. 5c, where TPD film was deposited at 304K. Although the four regions
appear in both cases, remarkable differences in region limits appears compared with
sample deposited at 267 K (Fig. 5b). The TPD sample grown at 𝑇𝑑𝑒𝑝=267 K shows values
of the percolation threshold (black downward arrows) and film continuity at 2.5 nm and
14 nm, respectively. However, sample grown at 304 K showed higher percolation
threshold, 6 nm (probably due to higher molecular mobility taking place at a higher
deposition temperature), and the thickness value for film continuity is not clearly resolved
from the data since the reduction in the slope after Region III is not observed. Recent
work by Fakhraai and coworkers26 in TPD films grown at 315K have shown that film
continuity was reached for film thickness above 20 nm, which is consistent with our
results.
Although similar behaviour of 𝐾2𝜔 was observed for both samples, final values of in-
plane conductivity determined form the slope in Regions IV of Fig. 5a and c, differ
considerably: 𝑘=0.145 Wm-1K-1 for films deposited at 𝑇𝑑𝑒𝑝=267 K and 𝑘= 0.132 Wm-
1K-1 for those deposited at 𝑇𝑑𝑒𝑝=304 K. While the difference only amounts to 10%, it is
substantially larger than the measurement uncertainty. Both films were expected to be
amorphous at those deposition temperatures, as was confirmed by clear glass transition
signatures shown in the calorimetric traces of Fig. S6 (SI). We thus believe that the
variation in thermal properties has to be related to the diverse characteristics of vapour-
deposited glasses, in particular density and molecular orientation, which strongly depend
on the deposition temperature. It has already been demonstrated that vapour-deposited
thin-film organic glasses grown at deposition temperatures slightly below their glass
transition develop enhanced kinetic and thermodynamic stability with a maximum at 𝑇𝑑𝑒𝑝
in the vicinity of 0.85𝑇𝑔
27,28. Glasses grown in these conditions are coined ultrastable
glasses. The glass transition temperature of a conventional TPD glass (this is, a glass
cooled from the liquid at 10 K/min) is 333 K. Glasses grown in the region 0.80-0.90𝑇𝑔,
which is the case for film deposited at 267 K, are stable glasses, as evidenced by the
higher onset of their glass transition temperature upon heating(Fig. S6 in SI). However,
films grown above 0.90𝑇𝑔, which is the case for film grown at 304K, are less stable.
Stability can be directly translated to density, meaning that the sample grown at 267 K
(0.80𝑇𝑔) has a slightly higher density that the one vapor-deposited at 304 K (0.91𝑇𝑔).
According to Dalal et al.29, the difference in density between the 2 samples should be
around 0.3%. Besides, stable glasses embedded with higher density also exhibit higher
values of the sound velocity up to 10%30–32, therefore it is reasonable to expect that more
stable glasses will also exhibit an enhancement of the thermal conductivity.
In the same way, vapour-deposited stable glasses show anisotropic molecular packing
with a molecular orientation that depends on the deposition temperature. TPD films
grown at 0.80𝑇𝑔 have molecules partially aligned parallel to the substrate while those
grown at 0.91𝑇𝑔 are mostly randomly oriented29. Molecular anisotropy can also play a
role in heat transport since it could be slightly favoured in the direction of molecular
alignment. More studies of the thermal conductivity variation as a function of deposition
temperature are under way to disentangle the effects of density and molecular orientation.
Metallic thin-film layers
To complete the proof of concept of the sensor, we also analyse growth kinetics of a
metallic indium layer.Fig.6 shows the real-time in-situ thermal conductance measurement
during growth, also for two deposition temperatures, 315 K and 260 K. Conductance
values (Fig. 6a,b)follow a similar pattern to the one observed in TPD films but with much
larger thickness scales. The conductance regions can be identified and discussed, as in
the TPD case, in view of microstructure evolution with thickness (see Fig. S4 for a
complete set of microscopy images). As shown in the inset of Fig. 6a, the fast decrease
of conductance at the very early stages of deposition is also present. In this thickness
range (up to 0.65 nm), tiny isolated clusters are expected to develop on the surface of the
SiNx membrane. The observed conductance drop of 1.5% is again an indication that
phonons with mean-free-paths slightly larger than those typically accounted for in
disordered solids are being scattered by the In/SiNx interface. Nevertheless, the minimum
conductance found in this case cannot be correlated with a percolation threshold, as we
considered in TPD, since individual In islands are much more conductive than the TPD
ones and they can contribute significantly to the thermal conduction, even though they
are still physically disconnected at this stage.
Interestingly, neither percolation threshold nor complete coverage of In grown on SiNx
are reached for metallic layer as thick as 120 nm (nominal thickness), as clearly evidenced
in the SEM images. As shown in Fig.6b, where growth characterization is performed up
to 450 nm, percolation starts to play a significant role at thicknesses around 200 nm,
where the conductance sharply increases due to continuous channels formation.
Subsequently, the thermal conductance increase slows down becoming linear with
thickness. The corresponding fitting line (red line in Fig. 6b) is proportional to the
thickness, meaning that the In layer is already continuous and the thermal conductivity
can be evaluated from the slope of the conductance versus thickness. The value of 𝑘 for
the continuous In layer deposited at 260K is 47 Wm-1K-1, substantially lower than the
tabulated value for In, 𝑘=83 Wm-1K-1, probably due to smaller grain size and enhanced
boundary scattering. From the images taken by SEM of the In layer at different
thicknesses (insets of Fig.6a and SI Fig.S4b), we approximate the coverage ratio and the
mean island area, as specified in SI.
As indicated in Fig. 6a,b we can differentiate four growth regimes: I: nucleation and initial
small island growth. II: growth of islands, divided in IIa (with small isolated circular
islands) and IIb (with larger islands irregularly shaped forming a bimodal island size
distribution). III: island percolation forming continuous channels and IV: vertical growth
of a continuous film.
Since the conductance evolution is slightly different (region II is divided in two stages)
than the one previously observed for TPD layers, we conduct finite element modelling
using the structural information provided by the electron microscopy images. We use a
simplified representation of our sample by building an array of 9x9 In square islands on
a 180 nm thick SiNx film. Changing the nominal thickness of the layer implies modulating
the size and separation of the islands to match the island size and coverage ratio observed
by SEM (Table S2 in SI). The thermal conductance was monitored by imposing a heat
flow and measuring the temperature difference arising on the simulated structure. The
results of the simulation are shown in Fig.6c. The simulation closely predicts the increase
of the slope of the curve 𝐾2𝜔(𝑡𝑠𝑚𝑝) deposited at 315 K in a thickness range around 30-
50 nm. In this thickness range and up to the percolation threshold above 200 nm In islands
are still isolated from one another and the increased conductance is due to the formation
of larger islands as In evaporation proceeds. At around 200 nm the sudden increase in the
slope is related to island percolation (Region III starts). At the end of the percolation
regime a continuous film forms and the conductance is heavily dominated by the In film
(Region IV). Then, the conductance increases linearly with a slope given by the thermal
conductivity of the film.
Fig.6 a) Thermal conductance vs thickness during deposition of In films: a) 𝑇𝑑𝑒𝑝=315 K. The
electron microscopy images of films of varying nominal thickness show the microstructural
evolution of the In layer. SEM images were acquired with a magnification of 100k, except the
thickest one recorded at 50k. The 1 m scale is provided inside the graph. The inset of a)
highlights the conductance drop during the early growth stages. b) 𝑇𝑑𝑒𝑝=260 K. Thickness range
is extended to attain complete percolation and total coverage. c) FEM simulation: normalized
conductance vs. In nominal thickness for representative structures with isolated islands of
different sizes and percolated islands, finally forming a continuous layer. The different growth
regimes are shown in Roman letters. Inset of c): Image of the 3D model, consisting on an array
of 9x9 square In islands (grey) on a SiNx thin membrane(green).
Conclusions
We have developed a universal sensor with extreme accuracy and methodology able to
perform real-time thermal conductance measurements during film growth. The technique
is adapted for any kind of material and thickness. Large sensitivity allows disentangling
the evolution of film microstructure with thicknesses. By analysing changes in the
apparent thermal conductance versus thickness, differentiated growth stages could be
identified giving rise to a complete comprehension of the growth process for several
materials. At early stages, the fast drop of the thermal conductance was related to
nucleation and isolated clusters formation. This step is followed by a regime where
clusters grow to form islands through coalescence and absorption of atoms/molecules
0204060801001200204060801000100200300400010020030040050060070080001002003004005000102030405060708090c)a)b)IIbIIbIIaIIa G (W/K)Tdep=315 K1 m024681014.714.814.915.0 0.65 nmISimulationIIaTdep=260 K Thickness (nm)IVIIIIVIIIIIb G/G0IIa from the gas phase, which dominates the variation of the heat conductance with thickness.
In an intermediated stage the percolation threshold was revealed by a conductance rise
while final mode, where the thermal conductance changes linearly with thickness,
corresponded to the formation and growth of a continuous film. In this regime, the thermal
conductivity of the film can be directly derived from the slope of the conductance versus
thickness plot.
The methodology ad-hoc, presented here, is easily extensible to devices with other
substrate materials compatibles with epitaxial growth. Using single crystalline
membranes, the conductance reduction within the first stages of growth will be enhanced
increasing the sensitivity. The extreme sensitivity will pave the way to apply the
technique to interesting phenomena such as phase changes during growth, size effects or
molecular orientation and density in organic glasses films, among others.
.
Acknowledgements
We acknowledge financial support from the Ministerio de Economía y Competitividad
through Grants CSD2010-00044 (ConsoliderNANOTHERM), FIS2013-50304-EXP and
MAT2016-79579-R. P. Ferrando-Villalba and J. Ràfols-Ribé were in receipt of an FPU
grant from the Spanish Ministry of Education, Culture and Sports. Ll. Abad was
supported by the "Ramón y Cajal" program from Spanish Government. The authors
gratefully acknowledge the ICTS-IMB-CNM clean room for the chips microfabrication.
References
1.
Floro, J. A., Chason, E. & Lee, S. R. Real Time Measurement of Epilayer Strain
Using a Simplified Wafer Curvature Technique. MRS Proc. 406, 491 (1995).
2.
An, I., Nguyen, H. V., Nguyen, N. V. & Collins, R. W. Microstructural evolution
of ultrathin amorphous silicon films by real-time spectroscopic ellipsometry. Phys.
Rev. Lett. 65, 2274–2277 (1990).
3.
You, H., Chiarello, R. P., Kim, H. K. & Vandervoort, K. G. X-ray reflectivity and
scanning-tunneling-microscope study of kinetic roughening of sputter-deposited
gold films during growth. Phys. Rev. Lett. 70, 2900–2903 (1993).
4.
Dendooven, J. et al. Atomic layer deposition-based tuning of the pore size in
mesoporous thin films studied by in situ grazing incidence small angle X-ray
scattering. Nanoscale 6, 14991–14998 (2014).
5.
Yeom, H. W. et al. Initial stage growth of In and A1 on a single-domain Si(001)2
× 1 surface. Surf. Sci. 341, 328–334 (1995).
6.
Schwartzkopf, M. et al. From atoms to layers: in situ gold cluster growth kinetics
during sputter deposition. Nanoscale 5, 5053 (2013).
7.
Schuisky, M., Elam, J. W. & George, S. M. In situ resistivity measurements during
the atomic layer deposition of ZnO and W thin films. Appl. Phys. Lett. 81, 180
(2002).
8.
Colin, J. J. et al. A load-lock compatible system for in situ electrical resistivity
measurements during thin film growth. Rev. Sci. Instrum. 87, 023902 (2016).
9.
Pei, Y., Wang, H. & Snyder, G. J. Band engineering of thermoelectric materials.
Adv. Mater. 24, 6125–35 (2012).
10. Kim, J., Lee, S., Brovman, Y. M., Kim, P. & Lee, W. Diameter-dependent
thermoelectric figure of merit in single-crystalline Bi nanowires. Nanoscale 7,
5053–9 (2015).
11. Hong, M., Chen, Z.-G., Yang, L. & Zou, J. Enhancing thermoelectric performance
of Bi 2 Te 3 -based nanostructures through rational structure design. Nanoscale 8,
8681–8686 (2016).
12. Cahill, D. G. Analysis of heat flow in layered structures for time-domain
thermoreflectance. Rev. Sci. Instrum. 75, 5119–5122 (2004).
13. Muñoz-Rojo, M., Caballero-Calero, O., Lopeandia, A. F., Rodriguez-Viejo, J. &
Martin-Gonzalez, M. Review on measurement techniques of transport properties
of nanowires. Nanoscale 5, 11526–11544 (2013).
14.
Shi, L. et al. Measuring thermal and thermoelectric properties of one-dimensional
nanostructures using a microfabricated device. J. Heat Transfer 125, 881–888
(2003).
15. Völklein, F. Thermal conductivity and diffusivity of a thin film SiO2/Si3N4
sandwich system. Thin Solid Films 188, 27–33 (1990).
16. Cahill, D. G. Thermal conductivity measurement from 30 to 750 K: the 3ω method.
Rev. Sci. Instrum. 61, 802 (1990).
17. Koh, Y. K. et al. Comparison of
the 3ω method and
time-domain
thermoreflectance for measurements of the cross-plane thermal conductivity of
epitaxial semiconductors. J. Appl. Phys. 105, 054303 (2009).
18. Muñoz Rojo, M. et al. Fabrication of Bi2Te3 nanowire arrays and thermal
conductivity measurement by 3ω-scanning thermal microscopy. J. Appl. Phys.
113, 054308 (2013).
19. Grauby, S., Puyoo, E., Rampnoux, J.-M., Rouvière, E. & Dilhaire, S. Si and SiGe
Nanowires: Fabrication Process and Thermal Conductivity Measurement by 3ω-
Scanning Thermal Microscopy. J. Phys. Chem. C 117, 9025–9034 (2013).
20.
Sikora, A. et al. Highly sensitive thermal conductivity measurements of suspended
membranes (SiN and diamond) using a 3ω-Vo lklein method. Rev. Sci. Instrum. 83,
1–17 (2012).
21.
Sikora, a. et al. Erratum: 'Highly sensitive thermal conductivity measurements of
suspended membranes (SiN and diamond) using a 3ω-Vo lklein method' [Rev. Sci.
Instrum. 83, 054902 (2012)]. Rev. Sci. Instrum. 84, 029901 (2013).
22.
Sultan, R. et al. Heat transport by long mean free path vibrations in amorphous
silicon nitride near room temperature. Phys. Rev. B 87, 214305 (2013).
23. Lee, S.-M. & Cahill, D. G. Heat transport in thin dielectric films. J. Appl. Phys. 81,
2590 (1997).
24. Völklein, F. & Starz, T. Thermal conductivity of thin films-experimental methods
and theoretical interpretation. XVI ICT '97. Proc. ICT'97. 16th Int. Conf.
Thermoelectr. (Cat. No.97TH8291) (1997). doi:10.1109/ICT.1997.667630
25.
Sultan, R., Avery, A. D., Stiehl, G. & Zink, B. L. Thermal conductivity of
micromachined low-stress silicon-nitride beams from 77 to 325 K. J. Appl. Phys.
105, 043501 (2009).
26. Zhang, Y. et al. Long-range correlated dynamics in ultra-thin molecular glass
films. J. Chem. Phys. 145, 114502 (2016).
27.
Swallen, S. F. et al. Organic glasses with exceptional thermodynamic and kinetic
stability. Science 315, 353–356 (2007).
28. Leo n-Gutierrez, E. et al. In situ nanocalorimetry of thin glassy organic films. J.
Chem. Phys. 129, 181101 (2008).
29. Dalal, S. S., Walters, D. M., Lyubimov, I., de Pablo, J. J. & Ediger, M. D. Tunable
molecular orientation and elevated thermal stability of vapor-deposited organic
semiconductors. Proc. Natl. Acad. Sci. 112, 4227–4232 (2015).
30.
Pogna, E. A. A. et al. Probing equilibrium glass flow up to exapoise viscosities.
Proc. Natl. Acad. Sci. 2331–36 (2015). doi:10.1073/pnas.1423435112
31. Kearns, K. L., Still, T., Fytas, G. & Ediger, M. D. High-modulus organic glasses
prepared by physical vapor deposition. Adv. Mater. 22, 39–42 (2010).
32.
Pogna, E. A., Rodriguez-Tinoco, C., Krisch, M., Rodriguez-Viejo, J. & Scopigno,
T. Acoustic-like dynamics of amorphous drugs in the THz regime. Sci Rep 3, 1–5
(2013).
Additional information
Author Contributions Statement
PFV drive most of the experiments, prepared the figures and contributed to the writing of
the main manuscript text. DT and LlA, designed and microfabricated the membrane-
based sensors used in the experiments. JRR carry out the experiments dealing with
organic molecules based thin films. ALF conceived the experiments and coordinated all
the team. ALF, JRV and GG wrote the main manuscript text. All Authors reviewed the
manuscript.
Competing financial interests.
The author(s) declare no competing financial interests.
|
1705.10278 | 1 | 1705 | 2017-05-26T00:27:08 | Materials processing with intense pulsed ion beams and masked targets | [
"physics.app-ph",
"cond-mat.mtrl-sci"
] | Intense, pulsed ion beams locally heat materials and deliver dense electronic excitations that can induce materials modifications and phase transitions. Materials properties can potentially be stabilized by rapid quenching. Pulsed ion beams with (sub-) ns pulse lengths have recently become available for materials processing. Here, we optimize mask geometries for local modification of materials by intense ion pulses. The goal is to rapidly excite targets volumetrically to the point where a phase transition or local lattice reconstruction is induced followed by rapid cooling that stabilizes desired materials properties fast enough before the target is altered or damaged by e. g. hydrodynamic expansion. We performed HYDRA simulations that calculate peak temperatures for a series of excitation conditions and cooling rates of silicon targets with micro-structured masks and compare these to a simple analytical model. The model gives scaling laws that can guide the design of targets over a wide range of pulsed ion beam parameters. | physics.app-ph | physics | 1
Materials processing with intense pulsed ion beams and masked targets
John J. Barnard1,2 and Thomas Schenkel2
1Lawrence Livermore National Laboratory, Livermore, CA
2Lawrence Berkeley National Laboratory, Berkeley, CA
Abstract
Intense, pulsed ion beams locally heat materials and deliver dense electronic
excitations that can induce materials modifications and phase transitions. Materials
properties can potentially be stabilized by rapid quenching. Pulsed ion beams with
(sub-) ns pulse lengths have recently become available for materials processing.
Here, we optimize mask geometries for local modification of materials by intense
ion pulses. The goal is to rapidly excite targets volumetrically to the point where a
phase transition or local lattice reconstruction is induced followed by rapid cooling
that stabilizes desired materials properties fast enough before the target is altered
or damaged by e. g. hydrodynamic expansion. We performed HYDRA simulations
that calculate peak temperatures for a series of excitation conditions and cooling
rates of silicon targets with micro-structured masks and compare these to a simple
analytical model. The model gives scaling laws that can guide the design of targets
over a wide range of pulsed ion beam parameters.
I. Introduction
Beams of energetic ions have been used to alter materials properties for many
years, with doping of semiconductors by ion implantation as one leading application
[1, 2]. Dynamic annealing processes induced by energetic ions can aid damage
repair [3, 4] but targets, such as semiconductor wafers, are generally annealed
thermally following ion implantation in order to repair lattice damage and to ensure
substitutional incorporation of dopants in the host crystal matrix. In the early
1980s short intense ion pulses with pulse lengths of 50 to 100 ns became available
where the pulse energy was high enough, 1 to 10 J/cm2, to simultaneously implant
and anneal materials [5]. Ion beam modification of materials with intense pulsed
ion beams continues to be applied in areas where the characteristics of ion-solid
interactions provide complementary advantages vs. other sources of directed
energy for materials processing (such as laser or electron beams) [2, 6]. More
recently, intense ion beams with much shorter pulse lengths of tens of ps to a few ns
have become available for materials studies. These ion beams are based on laser-
plasma acceleration of ions [7] or based on induction accelerator technology [8, 9].
The much shorter pulse lengths can now enable access to new materials processing
regimes and we explore this with simulations of the thermal response of materials
to intense, short ion pulses. In masked targets, the heating and cooling rate can be
tuned by the mask geometry together with the ion pulse parameters. We
investigate process parameters for intense ion pulses for applications where the
characteristics of ion-solid interaction are advantageous vs. complementary
processing schemes with (ultra-) short pulse lasers [2, 7], where a key differences
for ion vs. laser pulses is the uniform energy deposition that can be achieved with
energetic ions over many microns for any material.
We view intense ion pulses with masked targets as a processing opportunity that
combines aspects of two more common cases. One is the interaction of single, swift
heavy ions with materials, where local phase transitions and materials
modifications can be induced by the very high rate of electronic energy loss (e. g. 20
keV/nm for 1 GeV gold ions in silicon). Local heating and electronic excitation on a
nanometer scale can drive materials well above the melting point for a few
picoseconds followed by very rapid cooling and quenching of the intense, ultrafast
electronic excitation [10, 11], driving complex materials dynamics such as track
formation and defect annealing [12, 13]. When combined with external pressure,
novel phases can be induced and stabilized [14] and color centers, such as nitrogen
vacancy centers in diamond can be formed locally and without need for subsequent
thermal annealing [15]. The other common scenario is to simply use a broad beam
of intense ions [6]. Now heat conduction is mostly limited to the depth of the target
and materials simply evaporate or explode due to hydrodynamic expansion on a
time scale roughly given by the ratio of the ion range and the speed of sound in the
material, e. g. a few ns for excitation of solid targets over a depth of a few microns
[16, 17]. Here, we show simulations and an analytical model that inform mask
design for local heating of materials by intense ion pulses with optimized heating
and cooling rates. We illustrate the three regimes of pulsed ion beam processing of
materials in Figure 1.
Figure 1. Illustration of processing regimes. a) a single, swift heavy ion (nm and ps
scale heating and cooling), b) intense, short (ps to ns) ion pulse with (sub-) micron
masks to optimize heating and cooling rates, and c) ion beam pulse without
masking, where cooling is limited to one dimension.
2
Our article is motivated by the intriguing prospects of extreme chemistry and phase
engineering where one could drive materials such as silicon [19], diamond [10, 15]
or complex oxides [14] to extreme conditions of electronic excitation density,
temperature and pressure and then stabilize novel materials properties by rapid
quenching. By using a mask, the longitudinal dimension can be large compared to
the transverse dimension, allowing the possibility of rapid transverse cooling. We
explore the tradeoffs between mask hole size, beam pulse duration, and beam
intensity such that target temperature and cooling rate (after heating) are both
maximized.
II. Model
Figure 2 illustrates the basic geometrical configuration that we are considering. For
concreteness, we use 1.2 MeV helium ion beam pulses (as routinely available at
NDCX-II, an induction accelerator a Berkeley Lab [8, 9]), a gold mask, and a silicon
target. The ion beam illuminates the high density, metallic mask that has small holes
(0.1 to 1 µm radius) which allow the ion beam to pass and reach the target. The
mask is sufficiently thick (~ 3 µm of gold) so that it can absorb the beam pulse
energy and so that the ions are stopped in the mask before reaching the target. The
beam then heats a nearly uniform cylinder of target material that has a length that is
given by the ion energy deposition profile, (ca. 4 µm for 1.2 MeV He+ in silicon) and
that is much longer than the radius that is defined by the mask. We neglect ion
straggling here, which will broaden this radius slightly.
R
L
Target (Si)
Mask (Au)
Ion beam (He)
Figure 2. Geometry of ion beam, mask, and target. The (He+) ion beam (full beam
radius ~1 mm) impinges upon a (gold) mask absorbing most of the beam energy,
but allowing a ~0.1 µ to ~1 µ radius ion beam to pass to the silicon target. For a 1.2
MeV He ion beam the gold mask must be ca. 2 to 5 microns thick.
3
We note that geometries for the mask that we consider do not have to be restricted
to single beam openings. Multiple channels can be used as well as long as the
spacing between channels is sufficient so that cooling from one channel is not
affected by adjacent channels, thus spacings should be larger than the hole
diameters. Our analysis can also be extended to arbitrary mask geometries.
III. Simulation Result
We have carried out simulations using the hydrodynamics code HYDRA [18]. In
addition to the Au mask (with 0.2 µ hole) and Si target, the hole and region facing
the mask was filled with a low density argon gas for simulation convenience. The
gas was low enough density (density ~ 10-9 g/cm3) so that it did not impact the
dynamics. The ion beam had a parabolic temporal profile with full width of 0.8 ns,
and total integrated fluence of 0.4 J/cm2. Figure 3 a) shows the target density
profiles at the end of an 0.8 ns ion pulse where the gold mask material had
expanded slightly into the hole volume, reducing the fluence delivered to the un-
masked silicon target area by about 20%. Figure 3 b) shows a snapshot of the
corresponding temperature profile 0.5 ns into the ion pulse. Here, we see that the
gold foil prevented heating of the silicon target as expected at large radius and that a
warm region (~0.036 eV, or 690 K) was created in the silicon volume that had not
been covered by the gold mask.. Figure 4 summarizes the time history of the un-
masked, central silicon temperature for four different mask hole radii of 0.1, 0.2, 0.4,
and 1.0 µm.
4
(a)
(b)
Figure 3. (a) False color map of density of a silicon target (light blue[top]), gold
mask (red), with 0.2 micron hole filled with very low density Ar gas heated by 1.2
MeV He ion beam with parabolic intensity profile in time, illuminating the target
from below at the end of the heating pulse that lasted 0.8 ns. (b) temperature
profiles (in eV) near peak temperature at 0.5 ns.
5
Figure 4. Evolution of central temperature in the silicon target vs. time for four
different hole radii. For all four curves, the ion beam (1.2 MeV He+) had a pulse
duration of 0.8 ns and a fluence on target of 0.4 J/cm2.
6
Figure 5. Application of simple analytic model, discussed in section IV to
parameters of figure 4. Differences between results in the two figures are due to
hole closing not being considered in the analytical model, ; the ion range, L, was now
estimated from SRIM [20] rather than HYDRA and a simplified constant specific heat
and conductivity model were used.
IV. Analytic Model Results
In order to quickly understand the scaling of the achievable target temperatures and
cooling rates obtainable from these mask geometries, we derive a simplified model
that incorporates the principal physics of these masked targets. Since the heated
portion of the target is primarily cylindrical (with radius R and length L), with L > R,
we estimate the evolution of the temperature, by equating the rate of change of the
internal energy by the heating rate from the ion beam, less the cooling rate from
conductive losses through the boundaries.
(1)
7
,
Here R = radius of mask hole, L = ion stopping length, ∆t = pulse duration of the ion
beam (full width of parabolic pulse), kBT = target (silicon or diamond) temperature
(in eV, with kB Boltzmann constant), ρ, m = target (silicon or diamond) density,
mass, κ = target thermal conductivity, and F = incident ion fluence (energy/area).
Note that if R is large then the left hand side in eq 1 is balanced by heating and the
maximum target temperature occurs at the end of the beam pulse when t= ∆t:
(2).
assuming
However, when R is small, heating is always just balanced by cooling and the
temperature follows the instantaneous beam intensity during the ion beam pulse,
which for the case of a parabolic pulse occurs at t t= ∆t/2:
(3).
assuming
The simple model has the form:
and can be solved exactly, with solutions during and after the ion pulse:
(4).
Here
is the cooling time, and from the definitions
, where we introduce the ratio of a given mask radius, R, to a critical
mask radius, Rc, to highlight the connection of mask geometry to the time structure,
given by the ration of the cooling time, τ, and the ion pulse duration, ∆t.
In eq. (4) the normalized temperature, T(t)/Tmax1, depends only on time relative to
the pulse duration, t/∆t, and on the ratio of the cooling time to the pulse
duration, τ/∆t. In order to visualize this scaling, we now plot the normalized
temperature and its normalized time derivative in figures 6 and 7. We also plot the
time of maximum temperature and the values of maximum temperatures in figures
8a and 8b as a function of τ/∆t (the ratio of the cooling time to the pule duration).
We further show the maximum normalized time derivative of the normalized
temperature (also as a function of τ/∆t ) in figure 9.
8
Figure 6. Evolution of normalized target temperature (∆kBT/kBTmax1) for nine
values of τ/∆t.
Figure 6 shows that, as expected, for very low cooling rates, cooling times, τ, are
very long compared to the pulse length, ∆t, and a high steady state temperature is
reached for τ/∆t>10. At the other extreme heating is always balanced by very high
cooling rates and short cooling times, τ/∆t<0.3,and the temperature hardly rises.
The most interesting case is intermediary, where heating and cooling are balanced
during an ion pulse and a distinct peak temperature is reached for τ/∆t ~1.
Figure 7. Evolution of derivative of normalized target temperature d(∆kBT/kBTmax1)
/d(t/∆t) for nine values of τ/∆t, the ration of the cooling time to the pulse duration.
Note that the maximum cooling rate occurs for all curves at t/∆t =1 (at the end of
the heating pulse).
9
In figure 7 we show the time derivative of the relative temperature change from
Figure 6, thus the rate of temperature change as a function of time during the ion
pulse, t/∆t. Here, positive values of the derivative indicate heating and negative
values show net cooling. This derivative is useful in highlighting a maximum cooling
rate for a ratio of cooling time to pulse duration of τ/∆t=0.3.
(a)
(b)
Figure 8. (a) Time of maximum normalized temperature and (b) value of maximum
normalized temperature as a function of τ/∆t. Note that at small τ/∆t (small hole
radii R) the heating balances cooling, so that the temperature follows the ion beam
intensity, which is maximum at t/∆t =1/2, whereas at large τ/∆t (large hole radii R)
cooling is negligible and the maximum temperature just depends on the
accumulated fluence, and so the temperature is maximum at the end of the pulse
t/∆t =1. In the latter case the maximum temperature is just Tmax1.
10
Figure 9. Maximum normalized time derivative (at t/∆t =1) of the normalized
temperature as function as a function of τ/∆t . The maximum cooling rate occurs at
τ/∆t = 0.37, and the maximum normalized temperature ∆kBT/kBTmax1 = 0.41 for this
value of τ/∆t , at a time t/ ∆t= 0.76.
V. Discussion
Implementation of rapid local heating followed by rapid quenching of a material
using masked targets requires a choice of hole radius and pulse duration that
maximizes both the temperature achieved during the ion pulse and the cooling rate
after the ions have been delivered.
As an example in figures 10 and 11 we use the simple model (eq. 4) to explore these
tradeoffs for silicon targets. In these two figures, we fix the desired target
temperature at 0.5 eV (about 5800 K). We then calculate what fluence is required
(figure 10) as a function of hole radius, for three pulse durations, for the
combination of a Si target (κ = 1.49 W/(cm deg C, ρ = 2.3 g/cm3) and a 1.2 MeV He
ion beam (L = 4.2 µm). We further determine the cooling rate that is achieved at the
needed fluence as a function of hole radius (figure 11). We also note that HYDRA
simulations predict that pressure increases to about 7 GPa are associated with
lattice temperatures rises to 0.5 eV driven by these ion pulse conditions [17].
11
Figure 10. Required fluence to reach a maximum target temperature of 0.5 eV as a
function of hole radius, for three ion pulse durations using the model of equation (4)
and a Silicon target illuminated by a 1.2 MeV ion beam.
Figure 11. Cooling rate at the end of the beam pulse for a target reaching a
maximum temperature of 0.5 eV as a function of hole radius, for three ion pulse
durations using the model of equation (4) and a Silicon target illuminated by a 1.2
MeV ion beam.
12
As can be seen from figures 10 and 11 (and 9) if R << Rc (i.e. the cooling time is much
shorter than the ion pulse length, τ/∆t << 1), then the fluence required to reach the
desired temperature goal is much larger than in the opposite extreme when R >> Rc
(i. e. the cooling time is much longer than the pulse length, τ/∆t >>1). However, as
R/Rc increases the cooling rate decreases. It is likely that the desirable operating
point would be in the range where τ/∆t ~ 0.4 -1 near the knee of the curve in Figure
9 where cooling rates are large due to relatively small hole radii and where there is
not too much of a penalty yet in the required fluence. Importantly, we also see that
shorter ion pulses optimize both maximum temperature and maximum post-heating
cooling rate.
We may now compare heating and beam parameters for two example ion beams: an
accelerator produced He+ ion pulse from NDCX-II (the neutralized drift compression
experiment at Berkley Lab) [8, 9] and a laser generated proton pulse from a
petawatt laser such as Bella-i at Berkeley Lab [7, 21]. NDCX-II can produce ion
energy fluences of 0.4 to 0.8 J/cm2 in pulse durations of order 2 ns, and we expect
Bella-i will be able to produce much higher fluences of order 100 J/cm2 on similar
time scales of a few ns after ballistic de-bunching of the ion pulse with broad energy
distribution. Ion pulses are initially not much longer than the driving laser pulse
[22], but ion pulse de-bunch quickly due to large ion energy distributions and sub-
ns pulse durations can be restored by re-bunching [23]. Because the ion energies
are higher in Bella-i and ion ranges increase in proportion to the increased ion
energy the expected temperatures are only modestly higher, but the decreased
pulse duration (with compression) will allow an order of magnitude higher cooling
rate, and exploration of lattice relaxation on the 10 to 100 ps time scale.
Considerably higher lattice temperatures than for proton pulses are obtainable
using higher mass ions such as carbon. With our two approaches to ion pulse
generation and this target geometry, lattice relaxation can be studied (both in situ
and ex-situ) with both heating and cooling times of order the pulse duration
(between 10 ps and a few ns).
Ion beam driven heating of a target lattice results from the deposition of kinetic
energy through relatively well known elastic and inelastic scattering processes. For
1.2 MeV He ions, electronic stopping processes account for about 99% of energy loss
processes. Energetic electrons cascade and couple to the atomic lattice through
elastic and inelastic collisions and effective electron – phonon coupling dynamics,
resulting in target lattice heating [11]. We have explored how we can control local
heating through the balance of heating and cooling with a simple mask geometry for
short, intense ion pulses. We have found that this is indeed promising because the
characteristic length scale for cooling during (sub)-ns – scale pulses is of order 0.1
to 1 micron, i. e. in a range readily accessible by standard lithography techniques.
The range of delta electrons (energetic secondary electrons from inelastic
scattering) can be much shorter than mask openings. For 1.2 MeV He+, the cut-off
energy for delta electrons is about 600 eV with a range of only about 10 nm in
13
silicon [25]. The rate of local electronic excitation is thus not affected by masking.
For 10 MeV protons, the cut-off delta-electron energy is about 20 keV with a range
of about 1 micron in silicon, now comparable to or larger than possible mask
dimensions. For high energy protons with ranges in excess of 50 micron in e. g.
gold layers, fabrication of high aspect ratio mask structures with ratios of mask
thickness to mask opening >50 also becomes challenging. Mask aspect ratio limits
favor a few MeV protons or helium ions. Clearly, the balance of beam induced
damage to the tailoring of desired properties will have to be optimized for any ion
pulse condition. Further, mixed excitation by broad proton beams (which
penetrate a mask) combined with masked heavy ion beams can be considered.
We note that the peak temperature of 690 K in the example of Figure 2 is well below
the temperature for thermal re-crystallization of silicon of about 850 K [1].
However, light ion beam enhanced annealing effects [3] can enhance damage repair
and re-crystallization at much lower sample temperatures. Masked exposures of
materials with intense, pulsed ion beams now enable us to control the rate of local
electronic excitations partially decoupled from beam induced lattice heating. This is
exciting, because it can enable e. g. local damage repair and activation of dopants
that had been implanted in a prior processing step with minimal thermal budget
and thus minimal diffusion or segregation to a nearby interface. Further, repeated
and iterative steps of local processing such as addition of vacancies and local
excitation can enable the formation of ordered arrays of dopants and color centers,
where current formation efficiencies are too low for the formation of networks of
coupled spins [25, 26].
We have used HYDRA to simulate a silicon target heated by ion beam deposition
through a gold mask of various hole sizes, and with ion beams of various pulse
durations and beam fluences. The simulations and a simple heating/cooling model
show that when the hole radius is significantly greater than a critical mask radius,
Rc, (corresponding to the cooling time τ being significantly greater than the pulse
duration ∆t) the central temperature is close to Tmax1 (the temperature determined
by the beam fluence alone, without any masking). To maximize the cooling rate, the
radius can be chosen such that τ ~ 0.37 ∆t (corresponding to R ~ 0.75 Rc) and this
yields a maximum temperature kBTmax~ 0.41 kBTmax1. Intermediate choices such that
τ ~ ∆t (where the cooling time equals the pulse duration) may be made, if
minimizing fluence is weighed more heavily than maximizing cooling rate. For very
small radii, hole closing reduces the effective fluence on the target and must be
included in the calculation. This work is of interest when rapid cooling is desired to
quench and stabilize a change in phase or lattice configuration of a material that was
induced by intense ion beam pulses.
14
VI. Conclusion
Acknowledgments:
This work was supported by the US DOE under contracts DE-AC02-05CH11231
(LBNL) and DE-AC52- 07NA27344 (LLNL).
References
[1] Michael Nastasi and James W. Mayer, "Ion Implantation and Synthesis of Materials",
Springer Science & Business Media, 2007
[2] Yves Pauleau (ed.), "Materials Surface Processing by Directed Energy Techniques",
European Materials Research Society Series, ISBN: 978-0-08-044496-3
[3] F. Priolo, E. Rimini, Mat. Sci. Rep. 5, 319 (1990)
[4] M. Adel, R. Kalish, V. Richter, J. of Mat. Res. 1, 503 (1986)
[5] R. T. Hodgson, et al., Appl. Phys. Lett. 37, 187 (1980)
[6] T. J. Renk, et al., Proc. IEEE 92, 1057 (2004)
[7] J. Schreiber, P. R. Bolton, and K. Parodi, Rev. Sci. Instr. 87, 071101 (2016)
[8] P. A. Seidl., et al., Nucl. Instr. Meth. A 800, 98 (2015)
[9] P. A. Seidl, et al., https://arxiv.org/abs/1703.05697
[10] D. Schwen, E. Bringa, Nucl. Instrum. Methods Phys. Res., Sect. B 256, 187 (2007)
[11] N. A. Medvedev, R. A. Rymzhanov, A. E. Volkov, J. Phys. D: Appl. Phys. 48
(2015) 355303
[12] Y. W. Zhang, et al., Nat. Comm. 6, 8049 (2015)
[13] M.C. Ridgway, F. Djurabekova, K. Nordlund, Current Opinion in Solid State and
Materials Science 19, 29 (2015)
[14] M. Lang, et al., Nature Materials 8, 793 (2009)
[15] J. Schwartz, et al., J. Appl. Phys. 116, 214107 (2014)
[16] P. K. Patel, et al., , Phys. Rev. Lett. 91, 125004 (2003)
15
[17] J. J. Barnard, et al., Nucl. Instr. Meth. A 733, 45 (2014)
[18] M.M. Marinak, et al Phys. Plasmas 8 2275 (2001)
[19] B. Haberl, T. A. Strobel, J. E. Bradby, Appl. Phys. Rev. 3 040808 (2016)
[20] J. F. Ziegler, J. P. Biersack, M. D. Ziegler, Nucl. Instr. Meth. Phys. Res. Sec. B 268,
1818 (2010)
[21] Q. Ji, et al., Proc. International Particle Accelerator Conference, IPAC 2016,
THPMR004; S. Steinke, Q. Ji. S. S. Bulanov, in preparation
[22] B. Dromey, et al., Nat. Comm. 7, 10642 (2016)
[23] S. Busold, et al., Sci. rep. 5, 12459 (2015)
[24] D. Drouin, et al., Scanning 29, 92 (2007)
[25] A. Bienfait, et al., Nature 531, 74 (2016)
[26] T. Schröder, et al., J. Opt. Soc. America B – Opt. Phys. 33, B65 (2016)
16
|
1803.01824 | 3 | 1803 | 2018-07-05T17:01:08 | Origin of the anapole condition as revealed by a simple expansion beyond the toroidal multipole | [
"physics.app-ph"
] | Toroidal multipoles are a topic of increasing interest in the nanophotonics and metamaterials communities. In this paper, we separate out the toroidal multipole components of multipole expansions in polar coordinates (two- and three-dimensional) by expanding the Bessel or spherical Bessel functions. We discuss the formation of the lowest order of magnetic anapoles from the interaction between the magnetic toroidal dipole and the magnetic dipole. Our method also reveals that there are higher order current configurations other than the electric toroidal multipole that have the same radiation characteristics as the pure electric dipole. Furthermore, we find that the anapole condition requires that there is a perfect cancellation of all higher order current configurations. | physics.app-ph | physics | A simple expansion method for understanding toroidal multipoles
and anapoles in light-matter interactions
Shi-Qiang Li1, ∗ and Kenneth B. Crozier1, 2, †
1Department of Electrical and Electronic Engineering
University of Melbourne, Victoria 3010
Australia
2School of Physics
University of Melbourne, Victoria 3010
Australia ‡
(Dated: July 6, 2018)
Abstract
Toroidal multipoles are a topic of increasing interest in the nanophotonics and metamaterials
communities. In this work, we separate out the toroidal multipole components of multipole ex-
pansions in polar coordinates (two- and three-dimensional) by expanding the Bessel or spherical
Bessel functions. We discuss the formation of magnetic anapoles from the interaction between the
magnetic toroidal dipole and the magnetic dipole. Our method also reveals that there are higher
order current configurations other than the toroidal electric multipole that have the same radiation
characteristics as the pure electric dipole.
8
1
0
2
l
u
J
5
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
3
v
4
2
8
1
0
.
3
0
8
1
:
v
i
X
r
a
1
I.
INTRODUCTION
The toroidal multipole is a class of complex current configurations. The fundamental
order of the toroidal multipole is a toroidal dipole, whose current flow is similar to current
flowing inside a solenoid coil bent into the shape of a torus. The radiation pattern of a
toroidal dipole is the same as an electric dipole, and paired strengths of the two result in
a total cancellation of radiation. This condition is a non-radiating but non-trivial current
configuration that is called an anapole. It was first proposed in atomic physics by Zeldovich1
and observed later in cesium2. The toroidal multipole has also been found responsible for
some unusual phenomena in condensed matter physics3–6 and is currently the topic of much
interest in the nanophotonics and metamaterials communities7–9.
One of the most prominent characteristics of the toroidal multipoles is that each of the
multipoles has the same radiation pattern as its Cartesian electric multipole counterpart,
but has a very different current configuration and scales with a different order of kr, where
k is the wavenumber and r is the radius of the scatterer9–13. The toroidal dipole moment
can be picked out from the multipole expansion in Cartesian coordinates14,15. However, the
complexity of the extraction makes it difficult to obtain higher order toroidal moments (i.e.
beyond dipoles). By contrast, the classical multipole expansion in spherical coordinates was
discovered more than one century ago16. It expands an arbitrary radiation pattern as an
incoherent summation of orthogonal radiation patterns. Therefore, the Cartesian electric
multipole and the toroidal electric multipole17, which have the same radiation pattern, will
be contained in the same term of the expansion.
In this work, we separate out the toroidal multipole in the classical multipole expansion
of the scattering current. We expect that our method will be useful for any application
that involves toroidal multipoles including magnetoelectricity18–21, metamaterials7,9,22, and
nanophotonics23–25. The structure of our article is as follows. We first derive the two-
dimensional (2-D) multipole expansion in polar coordinates, which is simple to perform and
understand. We then extend the method to three-dimensional (3-D) spherical coordinates.
Following this, we present examples of application of this method to 2-D (nanotube with
both electric and magnetic anapole excitations) and 3-D (sphere with anapole excitation)
cases.
2
II. MULTIPOLE EXPANSION IN 2-D POLAR AND 3-D SPHERICAL COORDI-
NATES
Scattering from objects can be treated as radiation from scattering currents. In bounded
regions with scattering current sources, we only need to consider the source electric cur-
rent, as long as the relative permeability µ is 1, which is valid for non-magnetic materials.
Following Jackson's formulation26, we obtain:
(∇2 + k2) (cid:126)E(cid:48) = − i
k η∇ × ∇ × (cid:126)J,
∇· (cid:126)E(cid:48) =
(∇2 + k2) (cid:126)H =
∇· (cid:126)H =
0,
− ∇ × (cid:126)J,
, where k is the wavenumber, η is the wave impedance (cid:112)µ0/(0h), (cid:126)J is the scattering
current, defined as (cid:126)J = −iω0 ((cid:126)r − h) (cid:126)E15, and (cid:126)E(cid:48) = (cid:126)E + i
of the wave under consideration, (cid:126)r is the relative permittivity at the position (cid:126)r and h is the
(cid:126)J. ω is the angular frequency
(1d)
ω0h
0
relative permittivity of the host medium. µ0 and 0 are the permeability and the permittivity
of the free space respectively. This technique is also referred to as the volume-current method
for solving scattering problems27,28.
In 2-D polar coordinates, Eq 1a to Eq 1d can be solved separately for the TE wave
(electric field (cid:126)E = Ez z) and the TM wave ( (cid:126)H = Hz z) to obtain the following multipole
coefficients
(1a)
(1b)
(1c)
(2a)
(2b)
(2c)
(cid:82) eimψ
r
Am =
Bm = i
4
Ez =
[(kr) JψJm+1 + m (iJr − Jψ) Jm] ds
ηk
4
(cid:82) Jzeimψ [−Jm] ds,
(cid:80)m=+∞
(cid:80)m=+∞
m=−∞ AmH (1)
m=−∞ BmH (1)
m eimψ
, where (cid:126)r = (cid:112)(x2 + y2) (cos(ψ)x + sin(ψ)y), (cid:126)ψ = cos(ψ)x + sin(ψ)y, Hm is the Hankel
m eimψ
Hz =
(2d)
function of the first kind with argument kr, and Jm is the Bessel function of the first kind
with argument kr and order m. We use the boldface letter J to represent the scattering
current, with a subscript that indicates which component of the three principal axes of
cylindrical coordinates is being referred to. A detailed derivation is presented in Section 1
3
of the Appendix. With these equations, we can calculate the partial scattering cross-section
from each individual multipole m for any current distributions.
Recall that the Cartesian multipole expansion is expanded in terms of the order of kr,
while the Mie expansion is performed with respect to orthogonal radiation patterns15. Since
the Cartesian toroidal multipole has the same radiation pattern as its corresponding electric
multipole but scales with two more orders of kr9, a natural way to separate the Cartesian
toroidal multipole from the spherical electric multipole is then to group the terms in the
integrands on the right-hand sides of Eq 2a and 2b by their order of kr. However, care must
be taken here because the Bessel function is dependent on kr, thus it is necessary to expand
the Bessel function Jm in powers of kr by the method of Frobenius29,
(cid:0) kr
2
(3)
(cid:1)m+2 as Jm,1.
(m+1)!
(4a)
(4b)
We denote the leading term 1
m!
2
This enables direct comparison with the results of the Cartesian multipole expansion9,11,13,
by keeping the two leading terms in the expansion, as shown in Eq 4a and Eq 4b:
Jm =
1
m!
2
1
+ ...
(m + 1)!
2
(cid:18) kr
(cid:19)m+2
(cid:18) kr
(cid:19)m −
(cid:1)m as Jm,0 and the second term − 1
(cid:0) kr
(cid:90)
Jzeimψ(cid:104)−(cid:16)
(cid:90) eimψ
(cid:16)
[(kr) JψJm+1,0
Jm,0 + Jm,1
(cid:17)(cid:105)
(cid:17)
ds,
Jm,0 + Jm,1
]ds.
Am ≈ηk
4
Bm ≈ i
4
+m (iJr − Jψ)
r
In Eq 4a and Eq 4b, the underlined terms have a dependence on kr, which is two orders
higher kr (i.e. by (kr)2) than the terms without underlines. We can now write these
coefficients down explicitly for m=0 and m=1, in a manner that separates the pure dipole
term from the toroidal term. This is done as Eqs 5a and 5b (for TE mode) and Eqs 5c and
5b (for TM mode). The top and bottom rows of each equation are the pure dipole term
and toroidal term, respectively. Note that we need to treat B0 as a special case, as the
second term in Eq 4b vanishes when m = 0. We therefore include both J1,0 and J1,1 in the
expansion. We observe that the terms in the bottom rows of Eq 5a-d are two orders of kr
higher than those in the top rows, and they have the same radiation pattern. These match
the characteristics of the known toroidal terms. Based on the order (i.e. kr dependence)
and the polarization of the radiated field, we can further identify Eq. 5a and Eq. 5d as
being the electric dipoles/toroidal dipoles and Eq. 5b and Eq. 5c as being the magnetic
4
dipoles/toroidal dipoles.
A0 =
A1 =
B0 =
B1 =
ηk
4
ηk
4
ik
4
ik
4
4
− (kr)2
Jzds
(cid:90) −1
Jzeiψds
(cid:90) − kr
(cid:1)3
(cid:0) kr
Jψds
(cid:90) kr
(cid:90)
(iJr − Jψ)
(iJr − 3/2Jψ)
− (kr)3
− (kr)2
+ 1
2
16
2
2
2
2
(5a)
(5b)
(5c)
(5d)
ds
eiψ
2
In general, the separated terms found by the expansion of Bessel functions are not or-
thogonal. This is different from the original spherical multipole expansion, in which the
incoherent summation of the partial scattering cross-sections from different multipoles gives
the radiated power. A cross-term of the electric and toroidal multipole may be used to
account for the interaction between the terms with the same radiation pattern13. That is
the origin of the anapole condition8,30,31, for which there is destructive interference between
the electric dipole and toroidal dipole.
Within the same framework, we find that the 3-D spherical expansion can also be ex-
pressed in a similar manner as follows (readers interested in the derivation should refer to
the Appendix Section 3),
aE,d (l, m) =
(−i)l−1 k2ηOlm
[π (2l + 1)]1/2
(cid:90)
v
(cid:26)
eimψ (l + 1)
kr
jl,0
lP m
l (cos θ) Jr + τlm (θ) Jθ + iπlm (θ) Jψ
(cid:27)
dv
(6a)
aE,t (l, m) =
(−i)l−1 k2ηOlm
[π (2l + 1)]1/2
(cid:90)
v
eimψ −kr
4l + 6
jl,0
(cid:26)
(l + 1) P m
l (cos θ) Jr
(cid:27)
+(l + 3) [τlm (θ) Jθ + iπlm (θ) Jψ]
dv (6b)
, where P m
l (cos θ) are the associated Legendre polynomials26. aE,d (l, m) is the term corre-
sponding to the Cartesian electric multipole and aE,t (l, m) corresponds to the toroidal mul-
tipole. The other three terms in the equation are: Olm = (2l+1)(l−m)!
4π[l(l+1)](l+m), τlm = d
l (cos θ),
dθ P m
5
and πlm = m
sin θ P m
tion as jl,0 = (2kr)ll!
(2l+1)! .
l (cos θ). We denote the small argument limit of the spherical Bessel func-
III. EXAMPLES
We next verify our method by applying the above equations to two example light scatter-
ing problems. We use the commercial finite element method solver COMSOL for two pur-
poses. First, the scattering current distribution is found by numerically solving Maxwell's
equations. Second, our theory is used to find the partial scattering cross section from the
scattering current. A two-dimensional problem and a three-dimensional problem are con-
sidered, both involving anapoles.
To demonstrate the application of our method in two dimensions and the anapole con-
dition, we choose to study a hollow cylinder that we term a nanotube (Fig. 1). We plot the
partial scattering cross-section with m ranging from 0 to 4 (shown as the solid line plots)
with Am and Bm found from Eq 2a and Eq 2b, in Fig. 1 (a) and (b). We also plot the result
calculated from the definition of scattering cross-section (gray solid line) , i.e.
integrating
the Poynting vector of the scattered fields over a contour enclosing the nanotube. It is in
complete agreement with the result obtained by summing the multipoles from m = 0 to 4
(shown as the square markers in Fig. 1). Note that due to the circular symmetry of the
nanotube, the positive and negative orders of m are degenerate. For both the TE and the
TM polarizations, there are three very distinct peaks and one distinct dip in the scattering
spectra. The dips are denoted with black dashed arrows and labeled Ta and Tb in the figure.
It can be seen that the electric field distribution at each scattering-cross-section peak has
the number of nodes that would be expected from the order of the corresponding multipole
mode. For example, in panel b we have two nodes for the dipole (m = 1), though the field
pattern is distorted due to the interference with the other modes. It is worth noting that
the quality factor increases significantly for modes with higher orders. For instance, the
hexapole mode (m = 3) with the TM polarization has a full-width-half-maximum FWHM
of 10 nm and is located at a center wavelength of 500 nm.
In Fig. 1 (a) and (b), the minima Ta and Tb are interesting because they are almost
scattering-less, a characteristic that is associated with anapoles. With Eq 4a, Eq 4b, and
Eq 5a to Eq 5d, we can calculate the partial scattering cross-sections of toroidal multipoles.
6
We find that only the dipole terms (i.e. m = 0 and m = 1) are non-trivial at Ta and Tb.
The partial scattering cross-sections of the expanded terms according to Eq 5a to Eq 5d
are plotted in Fig. 1 (c) and (d). In the figure legends, we use subscript t to denote the
toroidal components of the multipole coefficients separated from either the magnetic dipole
or the electric dipole. These correspond to the bottom rows of Eqs 5a-5d. For brevity, we
henceforth use the acronyms MTD and ETD to denote the magnetic toroidal dipole and
electric toroidal dipole, respectively.
In the figure legends, we use the subscripts m and
e to denote the dipole coefficients associated with the Cartesian magnetic (i.e. A1,m and
B0,m) and electric (i.e. A0,e and B1,e) dipoles, respectively. For brevity, we also henceforth
use the acronyms CMD and CED to denote the Cartesian magnetic and electric dipoles,
respectively.
As discussed, from Fig. 1 (a) and (b), it can be seen that coefficients A1 and B0 exhibit
dips around Ta and Tb, respectively. Fig. 1 (c) and (d) provide insight into this phenomenon.
From Fig. 1 (c), it can be seen that the MTD component A1,t has the same magnitude as its
CMD counterpart A1,m at Ta. This allows destructive interference, provided that the phases
differ by a factor π (which is indeed the case at Ta). Similarly, Fig. 1(d) reveals that this is
also true for the TM case, with the MTD compnent B0,t having the same magnitude as the
CMD term B0,m at Tb. Destructive interference is again possible because the phases differ
by a factor of π at Tb. We thus conclude that magnetic anapoles occur at wavelengths Ta
and Tb. It can be seen from Fig. 1(c) that there is another wavelength (other than Ta) at
which A1,m and A1,t are of equal magnitude. Similarly, there is another wavelength (other
than Tb) at which B0,m and B0,t match. A1 and B0 do not show zeros at these wavelengths,
however, as the phase condition is not met.
We now consider the ETD and CEM components. It can be seen that A0 displays a dip at
approx 0.63 µm. From Fig. 1(c), it can be seen that the magnitude of the ETD component
A0,t matches that of the CED component A0,e. Destructive interference occurs because the
phase condition is met. We also note that while A0,t and A0,e have appreciable magnitudes
at shorter wavelengths, A0 is relatively small. This is because the phases of A0,t and A0,e
in this spectral range are such that destructive interference occurs. From Fig. 1(b), it can
be seen that B1 is small for wavelengths shorter than 0.6 µm. This is due to destructive
interference between B1,e and B1,t. The terms have comparable magnitudes and appropriate
phase differences in this wavelength range. It can also be seen that B1,e and B1,t intersect
7
at a wavelength of approx. 0.76 µm. B1 is not zero at that wavelength however, due to the
fact that the phase condition is not met.
Although the destructive interference of the far-field radiation from the dipole pairs (i.e.
(CED, ETD) and (CMD, MTD)) leads to the scattering dips at Ta and Tb, there are strong
field perturbations locally. These can be seen from the field profiles we present as Fig. 2. The
distribution of electric/magnetic field strengths in Fig. 2 reveals an intriguing phenomenon.
The roles of the magnetic field and the electric field are swapped. In the TE case (Fig. 2a),
the toroidal moment can be easily identified by the toroidal distribution of the magnetic
field vector, whose double curl defines the direction of the toroidal moment (denoted by the
boldface symobol Ta). On the other hand, for the TM case, it is the double curl of the electric
field that defines the direction of the toroidal moment ( denoted by the boldface symbol Tb).
Although magnetic anapoles form at wavelengths Ta and Tb due to destructive interference
between the CMD and MTD components shown in Fig. 1, we note that for the TE case, at
Ta, the strengths of the fields associated with the CED and ETD are around three times
higher than for the CMD and MTD. For the TM case, at Tb, the fields associated with the
CED and ETD are comparable to those of the CMD and MTD. In both cases, therefore,
the field patterns of Fig. 2 have contributions from all components (CED, CMD, ETD and
MTD). The field patterns of Fig. 2 are thus not purely magnetic or electric anapole.
We now consider a three-dimensional problem. We compare the results of applying our
method to light scattering by a sphere (with anapole excitation) with the results reported by
Miroshinichenko et al8 using Cartesian expansions. The expansion results from our spherical
expansion (Eq 6a and Eq 6b) are plotted in Fig. 3a. They are in agreement with the results
calculated and plotted in Fig. 2 of Miroshinichenko et al (the curves with diamond markers
in Fig. 3).
To consider this in further detail, we first recall the equations for the Cartesian electric
dipole and the toroidal dipole from Miroshinichenko et al8,
(cid:126)p =
(cid:126)t =
i
ω
1
10
(cid:126)Jdv,
(cid:104)(cid:16)
(cid:126)r· (cid:126)J
(cid:17)
(cid:105)
dv.
(cid:126)r − 2r2(cid:126)J
(7)
(8)
(cid:90)
(cid:90)
v
v
8
Comparing these equations with Eq 6a and 6b, we get:
px =− 3πi
ωk2η
py =− 3π
√
ωk2η
3
2πi
ωk2η
pz =
[aE,d (1, 1) − aE,d (1,−1)] ,
[aE,d (1, 1) + aE,d (1,−1)] ,
aE,d (1, 0) ,
tx =− 3π
k4η
ty =− 3π
√
k4ηi
3
2π
k4η
tz =
[aE,t (1, 1) − aE,t (1,−1)] ,
[aE,t (1, 1) + aE,t (1,−1)] ,
aE,t (1, 0) .
(9a)
(9b)
(9c)
(10a)
(10b)
(10c)
The detailed derivation can be found in Section 4 of the Appendix. Note that in Eq 6a
and 6b, the spherical Bessel function jl(kr) is expanded only up to order (kr)3, but this fully
accounts for the toroidal moments. We have plotted the next higher order moment having the
order of (kr)5 in Fig. 3a as the solid purple curve. We denote it as aE,h(1, 1). Although this
coefficient is quite small compared to aE,t(1, 1) and aE,d(1, 1), its presence is ultimately the
reason why the anapole condition does not coincide with the perfect cancellation of aE,t(1, 1)
and aE,d(1, 1). This can been from careful examination of Fig. 3, i.e. the perfect cancellation
occurs at slightly longer wavelength than the wavelength at which aE(1, 1) takes a value of
zero. The same shift can also be observed from Fig. 2 of the work by Miroshinichenko
et al8. The field distribution of aE,h(1, 1), however, cannot be identified due to the small
strength in comparison to the lower order poles. Nevertheless, we point out that, for larger
particles, retaining additional expansion terms might reveal the existence of moments other
than the toroidal moment that have even more complicated current configurations but the
same radiation pattern.
Next, we show that the Eq 6a and the Eq 6b can be used for obtaining any higher order
toroidal multipole. We set l to 2 and m to 1 and we solve for electrical quadrupoles. We
plot the calculated quadrupole coefficients in Fig. 3b. Surprisingly, quadrupolar toroid and
anapole show up for a sphere with radius on 30 nm larger than for which the dipole toroid
and anapole occur. From the field patterns, we can clearly see the vortices of electric field,
similar to the toroidal dipole shown in Fig. 3a. Furthermore, there are additional nodes in
the field variation, which is due to the quadrupolar nature of the moments.
9
Similarly, the toroidal moments of any arbitrary order l and degree m can be found, using
the same set of equations (Eq 6a and Eq 6b ). The results are trivial in the region of interest
studied here thus we do not present them here.
We note that the magnetic-type anapole was discussed recently by Luk'yanchuk et al.32.
This anapole was uncovered by comparing the zeros of the Mie coefficients with those of the
Cartesian magnetic dipole and the Cartesian magnetic-type toroidal dipole. As shown in
Section 5 of the Appendix, the direct expansion of the spherical Bessel function used in our
work can also be applied to the spherical magnetic multipole terms, giving the counterpart
of the toroidal moment from the magnetic multipole13,33,34. We find this term for the dipole
(cid:90)
v
(cid:126)r × (cid:126)r ×(cid:16)
(cid:126)r × (cid:126)J
(cid:17)
dv =
k2
20
(cid:90)
v
can be defined by:
(cid:126)tm =
k2
20
r2(cid:126)r × (cid:126)Jdv.
(11)
This is exactly the mean square radius of magnetic dipole defined in the literature.13,33,34
This term has the same radiation pattern as the Cartesian magnetic dipole, while it is
composed of toroidal magnetic current distribution defined by the double cross-product of
the magnetic moment with position vector (cid:126)r (see the middle expression of Eq 11). It has a
dependence of position in the integration of order (kr)3. Note that we have used kr instead
of r, which is used in most references. The introduction of wavenumber k ensures that the
dipole moments can be substituted into formulae (e.g. scattering cross-sections) developed
for electric and magnetic dipoles without any scaling factors. In Table 1, we summarize the
four multipoles discussed in this paper.
IV. CONCLUSIONS
In conclusion, we demonstrated a simple but general method to separate higher order
terms from curvilinear coordinate systems, including the classical Mie expansion (3D) and
the cylindrical expansion (2D). By expanding each electric multipole term with respect
to kr, we found that the leading term is the Cartesian electric multipole, while the next
term is the toroidal multipole. Our expansion approach reveals the existence of terms
of order even higher (> (kr)3) than toroidal multipoles. There might be situations (e.g.
larger particles) for which these higher order terms provide valuable physical insights. In
addition, we discovered a means to readily separate toroidal terms from magnetic multipole
10
TABLE I. A summary of the four types of multipoles discussed in this paper.
Category
Order of kr
Cartesian Electric Multipole
Cartesian Magnetic Multipole
Electric Toroidal Multipole
Magnetic Toroidal Multipole
l
l+1
l+2
l+3
(cid:126)m = 1
2
(cid:104)(cid:16)
(cid:82)
(cid:126)te = k2
10
v
(cid:126)tm = − k2
20
Dipole Expression
(cid:126)d = i
k
v
(cid:82)
(cid:82)
(cid:126)Jdv
(cid:17)
(cid:105)
v (cid:126)r × (cid:126)Jdv
(cid:82)
(cid:126)r − 2r2(cid:126)J
(cid:126)r· (cid:126)J
v r2(cid:126)r × (cid:126)Jdv
Multipole Expression
Eq. 6a
Eq. A.46
dv
Eq. 6b
Eq. A.47
expansions. This represents a counterpart to the separation of toroidal terms from electric
multipole expansions.
∗ [email protected]
† [email protected]
‡ This work was supported by the Australian Research Council (DP150103736 and FT140100577)
and by the Victorian Endowment for Science, Knowledge and Innovation (VESKI). The authors
thank Dr. Patrick Grahn. from COMSOL Inc. and Prof. John B. Ketterson from Northwestern
University for helpful discussion.
1 I. B. Zel'Dovich, Sov. J. Exp. Theo. Phys. 6, 1184 (1958).
2 C. Wood, S. Bennett, D. Cho, B. Masterson, J. Roberts, C. Tanner, and C. E. Wieman, Science
275, 1759 (1997).
3 B. B. Van Aken, J.-P. Rivera, H. Schmid, and M. Fiebig, Nature 449, 702 (2007).
4 I. I. Naumov, L. Bellaiche, and H. Fu, Nature 432, 737 (2004).
5 C. M. Ho and R. J. Scherrer, Phys. Lett. B 722, 341 (2013).
6 V. Scagnoli, U. Staub, Y. Bodenthin, R. De Souza, M. Garca-Fernndez, M. Garganourakis,
A. Boothroyd, D. Prabhakaran, and S. Lovesey, Science 332, 696 (2011).
7 T. Kaelberer, V. Fedotov, N. Papasimakis, D. Tsai, and N. Zheludev, Science 330, 1510 (2010).
8 A. E. Miroshnichenko, A. B. Evlyukhin, Y. F. Yu, R. M. Bakker, A. Chipouline, A. I. Kuznetsov,
B. Lukyanchuk, B. N. Chichkov, and Y. S. Kivshar, Nature Commun. 6 (2015).
9 N. Papasimakis, V. Fedotov, V. Savinov, T. Raybould, and N. Zheludev, Nature Mater. 15,
263 (2016).
10 I. Fernandez-Corbaton, S. Nanz, and C. Rockstuhl, arXiv preprint arXiv:1507.00755 (2015).
11
11 S. Nanz, Toroidal Multipole Moments in Classical Electrodynamics: An Analysis of Their Emer-
gence and Physical Significance (Springer, 2016).
12 C. Vrejoiu, J. Phys. A. Math. Gen. 35, 9911 (2002).
13 E. Radescu and G. Vaman, Phys. Rev. E 65, 046609 (2002).
14 V. I. Dubovik and A. A. Cheshkov, Soviet Physics JETP 24 (1967).
15 P. Grahn, A. Shevchenko, and M. Kaivola, New J. Phys. 14, 093033 (2012).
16 G. Mie, Annalen der physik 330, 377 (1908).
17 A. B. Evlyukhin, T. Fischer, C. Reinhardt, and B. N. Chichkov, Phys. Rev. B 94, 205434
(2016).
18 A. Gorbatsevich and Y. V. Kopaev, Ferroelectrics 161, 321 (1994).
19 W. Eerenstein, N. Mathur, and J. F. Scott, Nature 442, 759 (2006).
20 A. S. Zimmermann, D. Meier, and M. Fiebig, Nature Commun. 5 (2014).
21 S. Hayami, H. Kusunose, and Y. Motome, Phys. Rev. B 90, 024432 (2014).
22 A. A. Basharin, V. Chuguevsky, N. Volsky, M. Kafesaki, and E. N. Economou, Phys. Rev. B
95, 035104 (2017).
23 T. Raybould, V. Fedotov, N. Papasimakis, I. Kuprov, I. Youngs, W. Chen, D. Tsai, and
N. Zheludev, Phys. Rev. B 94, 035119 (2016).
24 Z. Liu, S. Du, A. Cui, Z. Li, Y. Fan, S. Chen, W. Li, J. Li, and C. Gu, Adv. Mater. (2017).
25 A. Mirzaei, A. E. Miroshnichenko, I. V. Shadrivov, and Y. S. Kivshar, Phys. Rev. Lett. 115,
215501 (2015).
26 J. D. Jackson, Classical electrodynamics (Wiley, 1999).
27 S. G. Johnson, S. Fan, A. Mekis, and J. Joannopoulos, Appl. Phys. Lett. 78, 3388 (2001).
28 A. Snyder, IEEE Trans. Microw. Theory Tech. 18, 608 (1970).
29 G. Teschl, Ordinary differential equations and dynamical systems, Vol. 140 (American Mathe-
matical Society Providence, 2012).
30 G. Afanasiev and Y. P. Stepanovsky, J. Phys. A Math. Gen. 28, 4565 (1995).
31 R. Wang and L. Dal Negro, Opt. Express 24, 19048 (2016).
32 B. Luk'yanchuk, R. Paniagua-Domnguez, A. I. Kuznetsov, A. E. Miroshnichenko, and Y. S.
Kivshar, Physical Review A 95, 063820 (2017).
33 V. Dubovik and V. Tugushev, Phys. Rep. 187, 145 (1990).
34 I. Fernandez-Corbaton, S. Nanz, R. Alaee, and C. Rockstuhl, Optics express 23, 33044 (2015).
12
FIG. 1. Decomposition of the scattering cross-section of a nanotube in the 2D multipole expansion.
The incident plane wave has a wavevector along x direction. TE wave case is shown as panel (a)
and TM wave case is shown as panel (b).
Inset of (a) shows a schematic of a cross-section of
the nanotube, whose inner r and outer R radii are 50 nm and 150 nm, respectively. The host
medium is air with n = 1 and the tube has a refractive index of 3.5. Insets in panels (a) and (b)
are mode profiles of various orders (amplitude of the electric field), with details explained in the
text. The expanded toroidal terms and Cartesian dipole terms are plotted in Figure (c) and (d)
for TE and TM cases respectively. These are partial scattering cross-sections calculated based on
the separation shown in Eq. 5.
13
Ta(a)(b)Tb(c)TaTb(c)(d) FIG. 2.
(a). Instantaneous electric (Ez, colour map) and magnetic (Hx and Hy, quiver plot) fields
around nanotube for TE mode. It can be seen that two loops are formed in the magnetic field
pattern. Fields are calculated for illumination wavelength, denoted by Ta, in Fig. 1b. Direction
of toroidal moment is indicated in this panel by vector also denoted by Ta.(b).
Instantaneous
magnetic (Hz, colour map) and electric (Ex and Ey, quiver plot) fields around nanotube for TM
mode. Calculation is performed for illumination wavelength, denoted by Tb, in Fig. 1a. Vector
also denoted by Tb in this panel gives direction of toroidal moment.
14
FIG. 3. (a) Spherical multipole coefficients for the dipole moments of order 1 and degree 1. The
dipole moments with (solid red line) and without (solid blue line) toroidal moment separation are
calculated with Eq 6a. The toroidal moment (found using Eq 6b) is plotted as the solid yellow
curve . The curve with red diamond markers is the coefficient converted from Eq 7 and that with
yellow markers is the coefficient converted from Eq 8. The insets are the field patterns of the
electric field strength cut at the center of the nano-sphere at two representative wavelengths (the
pure toroidal condition and the anapole condition). The solid purple curve plots the next higher
order term ((kr)5). (b) The spherical multipole coefficients for the quadrupole moments of order 2
and degree 1. The quadrupole moments with (solid red line) and without (solid blue line) toroidal
quadrupole moment separation are calculated with Eq 6a. The toroidal quadrupole moment (found
using Eq 6b)is plotted as the solid yellow curve. The insets are the field patterns of the electric field
strength cut at the center of the nano-sphere at the toroidal condition and the anapole condition.
15
(a)(b)ToroidalAnapoleAnapoleToroidal |
1906.05549 | 2 | 1906 | 2019-08-15T17:19:32 | Designing modular 3D printed reinforcement of wound composite hollow beams with semidefinite programming | [
"physics.app-ph"
] | Fueled by their excellent stiffness-to-weight ratio and the availability of mature manufacturing technologies, filament wound carbon fiber reinforced polymers represent ideal materials for thin-walled laminate structures. However, their strong anisotropy reduces structural resistance to wall instabilities under shear and buckling. Increasing laminate thickness degrades weight and structural efficiencies and the application of a dense internal core is often uneconomical and labor-intensive. In this contribution, we introduce a convex linear semidefinite programming formulation for truss topology optimization to design an efficient non-uniform lattice-like internal structure. The internal structure not only reduces the effect of wall instabilities, mirrored in the increase of the fundamental free-vibration eigenfrequency, but also keeps weight low, secures manufacturability using conventional three-dimensional printers, and withstands the loads induced during the production process. We showcase a fully-automatic procedure in detail for the design, prototype manufacturing, and verification of a simply-supported composite machine tool component, including validation with roving hammer tests. The results confirm that the 3D-printed optimized internal structure almost doubles the fundamental free-vibration eigenfrequency, allowing to increase working frequency of the machine tool, even though the ratio between elastic properties of the carbon composite and the ABS polymer used for 3D printing exceeds two orders of magnitude. | physics.app-ph | physics | Designing modular 3D printed reinforcement of wound composite hollow beams with
semidefinite programming
M. Tybureca,b,∗, J. Zemana, J. Nov´akb, M. Lepsa, T. Plach´ya, R. Poulc
aDepartment of Mechanics, Faculty of Civil Engineering, Czech Technical University in Prague, Th´akurova 2077/7, Prague 6, Czech
bExperimental Center, Faculty of Civil Engineering, Czech Technical University in Prague, Th´akurova 2077/7, Prague 6, Czech Republic
cCompo Tech PLUS, s.r.o., Nov´a 1316, Susice, Czech Republic
Republic
9
1
0
2
g
u
A
5
1
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
2
v
9
4
5
5
0
.
6
0
9
1
:
v
i
X
r
a
Abstract
Fueled by their excellent stiffness-to-weight ratio and the availability of mature manufacturing technologies, filament
wound carbon fiber reinforced polymers represent ideal materials for thin-walled laminate structures. However, their
strong anisotropy reduces structural resistance to wall instabilities under shear and buckling.
Increasing laminate
thickness degrades weight and structural efficiencies and the application of a dense internal core is often uneconomical
and labor-intensive. In this contribution, we introduce a convex linear semidefinite programming formulation for truss
topology optimization to design an efficient non-uniform lattice-like internal structure. The internal structure not
only reduces the effect of wall instabilities, mirrored in the increase of the fundamental free-vibration eigenfrequency,
but also keeps weight low, secures manufacturability using conventional three-dimensional printers, and withstands
the loads induced during the production process. We showcase a fully-automatic procedure in detail for the design,
prototype manufacturing, and verification of a simply-supported composite machine tool component, including validation
with roving hammer tests. The results confirm that the 3D-printed optimized internal structure almost doubles the
fundamental free-vibration eigenfrequency, allowing to increase working frequency of the machine tool, even though the
ratio between elastic properties of the carbon composite and the ABS polymer used for 3D printing exceeds two orders
of magnitude.
Keywords: Topology optimization, internal structure, semidefinite programming, additive manufacturing, elastic
instabilities, experimental validation
1. Introduction
Offering excellent stiffness-to-weight ratios, high damp-
ing, and a low sensitivity to fatigue and corrosion, car-
bon fiber reinforced polymers (CFRPs) are employed in
high-tech applications, including bodies of racing cars in
automotive [1], propulsors and turbines in naval [2, 3],
wing boxes and ailerons in aerospace [4], and rocket bod-
ies in the space industry [5]. Considerable attention has
been therefore paid to methods for optimizing structural
performance of these laminates, particularly the laminate
layup [6, 7]. To concurrently maximize bending stiffness
and keep weight low, the outer dimensions of these struc-
tures tend to be maximized, while the wall thickness is
minimized. The "thin-walledness" of the resulting struc-
tures, combined with their anisotropy, renders them highly
sensitive to shear and wall buckling instabilities manifested
in low fundamental free-vibration eigenfrequencies.
Below, we review common approaches to topology op-
timization that reduce wall instabilities by designing an
internal structure, Section 1.1. Section 1.2 provides a brief
∗Corresponding author
Preprint submitted to arXiv.org
introduction to semidefinite programming and highlights
several applications in structural optimization. Finally,
Section 1.3 reveals the merits of designing internal struc-
tures using semidefinite programming.
1.1. Topology optimization
Topology optimization techniques
[8] provide the
means for reducing wall instabilities when designing suf-
ficiently stiff yet lightweight structures.
In the simplest
setting -- beam cross section optimization -- we search an
optimal two-dimensional cross-sectional shape or a stiff-
ening of structures whose outer shape is predefined [9].
Blasques [10],
for example, maximized the fundamen-
tal free-vibration eigenfrequency while accounting for the
mass and shear center position constraints; Nguyen et al.
[11] optimized cross-sections of prismatic beams to maxi-
mize their buckling loads.
The design of optimal core sandwich structures, whose
skins are stiffened by a thick core is a related challenge.
For honeycomb, solid, truss, and foam rectangular panels
under in-plane compression or shear loads, optimal pe-
riodic topologies can be found analytically by consider-
ing the optimality criterium of all failure modes occur-
August 16, 2019
ring simultaneously [12]. For complex boundary condi-
tions, parametric shape-optimization studies are usually
performed. Wang and McDowell [13] studied the geometry
of a metal honeycomb sandwich beam core under torsion
and bending and Xu and Qiu [14] optimized the lattice
core of a composite sandwich panel to increase the funda-
mental eigenfrequency while accounting for uncertainties
in the model. They concluded that bending eigenfrequen-
cies increase with increasing strut thicknesses, with an in-
crease in the elastic and shear modulus of the composite,
and with a decrease in density. Although Daynes et al.
[15] optimized spatially-graded lattice structures within
a single sandwich panel domain, surprisingly, almost no
prior research seems to have stepped beyond parametric
intuition-based designs [16, 17], the rare exception be-
ing the multi-scale topology optimization approach inves-
tigated by Coelho and Rodrigues [18].
Questioning whether, where, and how to stiffen already
engineered designs in order to further improve their struc-
tural performance constitutes the central question of the
reinforcement problem [19, 20], superseding the former di-
mensional reduction and periodicity assumptions. Initial
studies in this area have considered maximization of the
fundamental eigenfrequency [20] and improving the struc-
tural frequency response of plane elastic structures [21] us-
ing the homogenization and optimality criteria methods,
respectively.
Using the ground structure approach for topology op-
timization of truss structures, Bendsøe et al. [22] fixed
cross-sectional areas of a set of bars and searched for
their stiffest truss reinforcement, a (non-smooth) convex
quadratic programming formulation. Alternatively, the ef-
fect of a fixed boundary structure has been approximated
by an appropriate application of nodal forces to the ground
structure [23, 24], but this choice influences, however, the
optimized design.
In the setting of continuous topology optimization,
Luo and Gea [25] developed a systematic optimization ap-
proach for the topology and orientation design of compos-
ite stiffeners of plates and shells in both static and dynamic
settings, and Wang et al. [26] optimized the overall struc-
tural rigidity of an automobile body through a maximiza-
tion of the fundamental eigenfrequency. In aerospace ap-
plications, Maute and Allen [27] optimized a wing's inter-
nal structure, subjected to fluid-surface interactions; Aage
et al. [28] performed an extremely large-scale optimization
of the internal structure of a Boeing 777 wing, while avoid-
ing the traditional rib and spar designs [29]. In military
applications, topology optimization was the basis for the
design of additively-manufactured lattice-reinforced pene-
trative warheads [30] and for optimizing the layout weight
of stiffeners in composite submarines subjected to nonsym-
metric wave slap loads [31].
Other methods relevant to internal structure design
have arisen in conjunction with recently introduced coat-
ing and infill optimization problems. Clausen et al. [32] de-
veloped a formulation for the optimization of (uniformly)
2
coated structures, wherein a base material, infill, was sur-
rounded by another material at the interfaces, finding
a porous, complex infill significantly improves both struc-
tural buckling resistance and robustness to local pertur-
bations when compared to optimized solid structures of
equal weight and similar stiffnesses [33, 34]. In three di-
mensions, optimized designs further exploit the merits of
closed shell surfaces through the sandwich effect [34].
Inspired by natural, bone-like microstructures, Wu
et al. [35] optimized a spatially non-uniform porous infill,
Wang et al. [36] developed a sequential approach for gen-
erating graded lattice mesostructures, and Zhu et al. [37]
introduced a novel asymptotic-analysis-based homogeniza-
tion approach. All these methods automatically design
stiff yet porous infills for additive manufacturing prod-
ucts while superseding the traditional pattern-based de-
signs [38]. Finally, Wu et al. [39] extended their approach
to the ultimate setting of a concurrent optimization of
coated structures and porous infills, and Groen et al. [40]
have developed a homogenization-based method to accel-
erate solutions.
1.2. Semidefinite programming
It has been shown in recent decades that several struc-
tural optimization problems can be modeled as semidefi-
nite programs. Linear semidefinite programming (SDP) is
a subset of convex optimization of the form
cTx
min
x
s.t. X = F0 +
X (cid:23) 0,
m(cid:88)
i=1
xiFi,
(1a)
(1b)
(1c)
and involves minimization of a linear function (1a) over
a spectrahedron, which is an intersection of an affine space
(1b) with the cone of symmetric positive semidefinite ma-
trices (1c). In (1c), the notation "(cid:23) 0" enforces positive
semidefiniteness of the left hand side. Due to the linear
dependence of X on x (1b), (1c) is commonly referred to
as a linear matrix inequality (LMI).
Applications of semidefinite programming to struc-
tural design were pioneered by Ben-Tal and Nemirovski
[41], de Klerk et al. [42], and Vandenberghe and Boyd
[43] who developed formulations for minimum-compliance
and weight truss topology optimizations.
The main
added value of SDP lies in its ability to effectively avoid
the non-differentiability of multiple eigenvalues for free-
vibrations [44, 45] and buckling [46, 47], robust optimiza-
tion [48], and bounds improvement for optimization prob-
lems in a discrete setting [49]. Semidefinite programming
has also found applications in optimal materials design,
the Free Material Optimization approach [50], or in the
limit analyses [51].
8 0 m m
8 0 m m
z
y
x
A
A '
1000 mm
(a)
80 mm
2.22 mm
(b)
8
0
m
m
200 kN/m2
(c)
Figure 1: Case study setup. (a) Outer dimensions and simply supported boundary conditions, (b) prismatic cross-section, and (c) compression
molding load case.
1.3. Aims and novelty
In this contribution, we consider an industrial problem
of designing the least-weight internal structure of a thin-
walled filament-wound composite machine tool component
prone to shear and buckling wall instabilities. The beam
laminate was designed for bearing dynamic loads, allowing
us to describe the wall instabilities naturally in terms of
free-vibrations eigenfrequencies.
In current production process, the wall instabilities are
reduced by inserting a uniform foam core structure into the
beam interior, an uneconomical and labor-intensive pro-
cess. Conversely, we have aimed to automatically design
a structurally-efficient internal structure which can easily
be manufactured using conventional low-cost 3D printers.
To this goal, we extended the convex (linear) semidefi-
nite programming formulation introduced by Ohsaki et al.
[44] and Ben-Tal and Nemirovski [48] to design globally-
optimal least-weight lattice-like internal structures and ap-
ply it to increasing the fundamental eigenfrequency and
decreasing the compression-molding compliance of a thin-
walled composite beam prototype. Note that Achtziger
and Kocvara [45] avoided prescribed structural elements
but allowed for a non-structural mass and Ohsaki et al.
[44] did not consider prescribed mass or stiffness.
After introducing the case study of a simply-supported
CFRP beam design in Section 2, we develop its finite el-
ement representation in Section 3.1. For this represen-
tation, a semidefinite programming formulation for truss
topology optimization of internal structures is developed
in Section 3.2. Having designed the optimal internal struc-
ture, we post-process the optimization outputs and export,
in a fully-automated way, the internal structure for ad-
ditive manufacturing in Section 3.3. During manufactur-
ing, the internal structure serves as the support for carbon
fibers in the filament-winding production phase, and a pro-
totype is created. Section 4 describes verification and ex-
perimental validation of the prototype and concludes that
its response agreed well with the model prediction.
tion consists of several steps, in which a supporting struc-
ture made of manually processed high-density foam is
wound biaxially with a combination of ultra high modulus
(UHM) and high modulus (HM) carbon fibers saturated
with epoxy resin. The supporting structure prevents cross-
section distortions induced by compression-molding loads
as shown in Fig. 1c. Subsequently, the beam is cured, the
supporting structure is pulled out, and the beam outer
surface is finalized.
The final product is exposed primarily to loads that in-
duce bending. For this purpose, most of the carbon fibers
are aligned with the beam's longitudinal axis (layer 2 in
Table 1), denoted by x in Fig. 1a, whereas the remain-
ing layers reduce the susceptibility to delamination. See
Table 1, where all layers are listed by their orientations
relative to the beam's longitudinal axis, θ. This layered
composition reliably transmits the design forces to the sup-
ports, and is thus fully sufficient in this sense.
Attributed to transversely isotropic material proper-
ties, the beam's walls are, however, prone to elastic wall
instabilities under shear and buckling, which also mani-
fests in free-vibration modes and frequencies of the non-
reinforced beam. Figure 2 confirms that the first fun-
damental eigenmode with a frequency of 128.5 Hz cor-
responds to shear wall instabilities, whereas the second
eigenmode combines bending with buckling; all higher
eigenmodes (not shown) exhibit similar wall instabilities.
Because the fundamental eigenfrequency limits the maxi-
mum working frequency of the machine part, its increase
is of considerable interest.
Although the effect of these instabilities can be re-
duced by additional laminate layers or by also keeping
the uniform foam structure
for operational loads, the
2. Case study
As the basic structure, we consider a prismatic, lami-
nated composite beam 1, 000 mm long, with a 80× 80 mm
thin-walled cross-section 2.2 mm thick, Fig. 1b. Accord-
ing to current manufacturing technology, beam produc-
(a) First eigenmode with a fre-
quency of 128.5 Hz.
(b) Second eigenmode with a fre-
quency of 403.1 Hz.
Figure 2: Axonometric and front view on the (a) first and (b) second
eigenmodes of the composite beam predicted by the finite element
model.
3
Table 1: Material properties of the wound composite beam laminae. E1 and E2 stand for the Young moduli in the fiber and transverse
directions, respectively; G12 denotes the shear modulus, ν12 and ν23 are Poisson's ratios. θ constitutes the angle between the 1-direction and
x, rotating around the beam surface normals. Finally, ρ and t denote the density and thickness of the plies.
Layer
1
2
3
4
5
6 (casing)
E1 [GPa]
128.2
421.9
130.9
130.9
130.9
2.0
E2 [GPa] G12 [GPa]
3.4
3.2
3.4
3.4
3.4
0.7
5.0
3.7
5.0
5.0
5.0
2.0
ν12 [-]
0.34
0.37
0.34
0.34
0.34
0.37
ν23 [-]
0.35
0.35
0.35
0.35
0.35
0.37
θ [deg]
89.3
0.0
26.9
−26.9
26.9
0.0
ρ [kg/m3]
1, 428
1, 680
1, 458
1, 458
1, 458
1, 040
t [mm]
0.25
1.25
0.18
0.36
0.18
0.80
added weight, decrease in the bending eigenfrequencies,
and labor-intensive production process render these ap-
proaches both time-inefficient and uneconomical.
3. Optimal design of internal structure
The aim of this section is to cast the optimal internal
structure design problem in the form of a linear semidef-
inite program (1). The internal structure has to with-
stand compression molding loads with a maximum deflec-
tion bound, while the internal structure is temporarily
supported by a steel mandrel passing through the beam
interior, Fig. 1c. Most importantly, the internal structure
is supposed to increase the beam fundamental eigenfre-
quency via reduction of wall instabilities.
In this section, we first describe the finite element
model of the composite beam. This finite element model
serves then as the basis for establishing the optimization
problem formulation, yielding an optimal internal struc-
ture design. The section is concluded by discussing post-
processing steps necessary to maintain manufacturability
of the design.
3.1. Finite element model
The outer composite beam surface is discretized with
shell elements which are supplied with the material proper-
ties from Table 1. The beam internal structure is modeled
by bar (truss) elements, with the isotropic Acrylonitrile
Butadiene Styrene (ABS) material properties [52]: elastic
modulus EABS = 2 GPa, Poisson ratio νABS = 0.37, and
density ρABS = 1, 040 kg/m3.
Special care needs to be paid to establishing a rigid
connection between the internal structure and the car-
bon composite. The so-called casing, see Fig. 3, which
is a 0.8 mm thin layer of printed beam walls, further pre-
vents leaking of the epoxy resin into the beam's interior.
Casing is modeled as the bottom layer of the laminate
composition, recall Table 1.
The finite element model for the optimization part was
developed in Matlab. In this model, the outer laminate
was modeled with four-node Mitc4 elements [53]. The
composite beam interior was discretized into the ground
structure [54], a set of admissible truss1 elements, whose
cross-sections we search in the optimization part. The
ground structure was constructed from 47 × 4 × 4 modu-
lar building blocks shown in Fig. 4, to guarantee manufac-
turability of the entire internal structure with 3D printing.
Note that the bars placed within the location of the steel
mandrel were removed from the ground structure and that
the shell element nodes coincided with the ground struc-
ture nodes, resulting in a rather coarse discretization of
the outer layer.
3.2. Formulation of the optimization problem
3.2.1. Non-convex formulation
Composite beam, CFRP
Adopting the previously described discretization, our
goal is to find the cross-sectional areas a of nb bars in the
Internal structure, ABS
Casing, ABS
1Based on comparative simulations (not shown), modeling inter-
nal structure with trusses or beams leads to an insignificant difference
in the structural response which enabled us to employ truss topology
optimization approaches in Section 3.2.
Figure 3: The entire structure of considered composite beam design:
internal structure (used for the reduction of wall instabilities and
for increase of the lowest free-vibration frequency); casing of the
internal beam structure (to allow for wounding the final composite
layer); composite layers, which transmits working load applied to the
beam.
4
z
y
x
21.3
m
m
19.5 mm
m
m
1 9.5
Figure 4: Ground structure building block, which fill in the entire
internal volume of the composite beam to represent a (to be opti-
mized) internal beam structure. Cross-sectional areas of individual
trusses are design variables of the optimization problem (2).
nb(cid:88)
nb(cid:88)
minimum-weight (or volume) ground structure, such that
the fundamental eigenfrequency exceeds the user-defined
lower threshold f , taken as 300 Hz in what follows, while
exhibiting limit displacements u of the reinforced structure
during the compression molding load case. This leads to
the following optimization problem
min
a,ucm,u
(cid:96)Ta
inf
(MIS
fv (a)+MC
fv)u(cid:54)=0
s.t.
with
uT(cid:0)KIS
uT(cid:0)MIS
(cid:0)KIS
fv (a) + KC
fv
fv (a) + MC
fv
(cid:1) u
(cid:1) u
(cid:1) ucm = fcm,
≥ λ,
cm(a) + KC
cm
−u1 ≤ ucm,disp ≤ u1,
0 ≤ a ≤ a1,
λ = 4π2f
2
.
(2a)
(2b)
(2c)
(2d)
(2e)
(3)
fv, and mass, MC
In this formulation, the vector (cid:96) appearing in the objective
function (2a) collects the bar lengths in the truss ground
structure. The Rayleigh quotient in (2b) involves stiffness,
KC
fv, matrices of the outer shell structure
for free-vibration analysis together with stiffness, KIS
fv (a),
and mass, MIS
fv (a), matrices of the internal structure. The
design-dependent contributions of the internal structure
are obtained as
KIS
fv (a) =
KIS
fv,eae, MIS
fv (a) =
MIS
fv,eae,
(4)
e=1
e=1
fv,e and MIS
where KIS
fv,e stand for the stiffness and mass
matrix of individual bars in the free-vibration (fv) setting,
respectively; ae is the e-th component of a and λ the limit
fundamental free-vibrations eigenvalue.
cm, and design-dependent, KIS
The constraints (2c) and (2d) address the compression-
molding (cm) load case, recall Fig. 1c. Specifically, (2c)
introduces the generalized nodal displacements ucm in re-
sponse to the generalized load vector fcm corresponding
to the compressive load, and ucm,disp denotes the dis-
placement components of ucm. The stiffness matrix cor-
responding to this load case consists again of the design-
independent, KC
cm(a), parts;
the latter is obtained as in (4). The symbol 1 denotes a
column vector of all ones. Notice that the stiffness matri-
ces in (2b) and (2c) differ because of different boundary
conditions in the operational, Fig. 1a, and manufacturing,
Fig. 1c, load cases. The constraint (2d) requires the dis-
placement components of ucm to remain smaller than the
user-defined limit value u, considered to be 0.5 mm in this
study. Finally, (2e) requires the cross-sectional areas of the
bars to be non-negative and smaller than a = 200 mm2,
a value set by the additive manufacturing constraints.
A closer comparison of the optimization problem of
Eq. (2) and that of Eq. (1) reveals that the problem of
Eq. (2) lacks the structure of a semidefinite program.
Namely, the objective function (2a) and the matrices in
the constraints depend affinely on the design variables, a.
However, the constraints (2b) and (2c) are non-convex as
the stiffness and mass matrices may become singular when
the zero lower-bound for cross-sectional areas is attained
in (2e). Moreover, (2b) might become non-differentiable
when an eigenvalue with multiplicity higher than one is
encountered. Altogether, this renders the problem (2) ex-
tremely difficult to solve in its original form. In the follow-
ing section, we show how to re-cast the problem of Eq. (2)
as a linear semidefinite programming problem.
3.2.2. Convex semidefinite program
Similar eigenvalue constraints such as (2b) have al-
ready been studied in detail by Ohsaki et al. [44] and
Achtziger and Kocvara [45]. Their results allow us to
rewrite (2b) equivalently as a convex LMI
KIS
fv (a) + KC
fv − 4π2f
fv (a) + MC
fv
(5)
2(cid:0)MIS
(cid:1) (cid:23) 0,
where the left hand side expression is a linear function of a.
This constraint also avoids the non-differtiability of mul-
tiple eigenvalues, see, e.g., [45], and effectively eliminates
the kinematic variables u from the problem formulation.
To attain convexity of the final formulation, the com-
pression molding constraints (2c) -- (2d) must be enforced
only approximately in the form of the LMI [42, 43, 48]:
−f T
cm
−fcm KIS
cm(a) + KC
cm
(cid:23) 0,
(6)
(cid:18) ccm
(cid:19)
in which ccm denotes a prescribed upper bound on com-
pliance (work done by external forces) of the compression
molding load case. As found from parametric studies (not
shown), an appropriate value of the bound is provided as
ccm = ccm,0
u
max{ucm,disp} ,
(7)
where ccm,0 stands for the compliance of the non-reinforced
structure:
ccm,0 = f T
cm
fcm.
(8)
(cid:17)−1
(cid:16) KC
cm
cm and fcm are constructed from KC
Here, KC
cm and fcm, re-
spectively, by application of appropriate boundary condi-
tions. For this particular problem, this compliance bound
resulted in a maximum deflection of 0.4 mm.
The final linear semidefinite programming formulation
eventually reads as
min
(cid:96)Ta
a
s.t. KIS
fv (a) + KC
2(cid:0)MIS
fv − 4π2f
(cid:18) ccm
−fcm KIS
(cid:1) (cid:23) 0,
(cid:19)
fv (a) + MC
fv
−f T
cm
cm(a) + KC
cm
(cid:23) 0,
1a ≥ a ≥ 0.
(9a)
(9b)
(9c)
(9d)
This formulation now possesses the structure of the linear
semidefinite program introduced in Section 1.2, and thus
5
Figure 5: Symmetric half of the beam as cut off by the xz plane. The top shell surface is hidden to reveal the internal structure.
can be solved efficiently via modern interior-point meth-
ods.
For numerical solution, we adopted the state-of-the-
art industrial optimizer Mosek [55]. After discretiza-
tion, the problem in Eq. (9) has 10, 216 admissible bars
in total, with the corresponding sizes of the linear matrix
inequalities 5, 154 × 5, 154 (free-vibration, Eq. (9b)) and
4, 608 × 4, 608 (compliance, Eq. (9c)). After tweaking the
optimization problem with the steps outlined in the fol-
lowing subsection, the optimization process itself required
13 GB of memory, and terminated after 5.75 core hours
running on Intel R(cid:13) Xeon R(cid:13) Gold 6130 processors at the
MetaCentrum2 virtual organization cluster. The resulting
distribution of the optimal internal structure is shown in
Fig. 5. Note that the internal structure increased the orig-
inal weight of the beam, 1, 094 g, by an additional 488 g
(280 g of casing and 208 g of reinforcing bars).
Improving solver performance. To reduce the number of
iterations and time per iteration to solve problem (9), we
rescale the cross-sectional areas to obtain the optimal val-
ues of the order of 1.0 mm. Second, to improve both the
numerical stability and convergence of the algorithm con-
siderably, we rescale Eqs. (9b) and (9c) with the square
root of the Frobenius norm estimates of KC
fv (Eq. (9b)),
and KC
cm (Eq. (9c)). Finally, using the static condensa-
tion, Appendix A, and Schur complement, Appendix B,
decomposition techniques, the sizes of LMIs reduce to
3, 426× 3, 426 (free-vibration, Eq. (9b)) and 2, 880× 2, 880
(compliance, Eq. (9c)). Consequently, memory usage was
decreased from 21 GB to 13 GB, and the solution process
was accelerated by 71% (from 19.5 to 5.75 core hours).
3.3. Post-processing
Manufacturing of the optimal design is preceded by
three preprocessing steps addressing individual bars, seg-
2https://metavo.metacentrum.cz/
mentation into modules, and conversion to a solid model.
Note that we checked that none of the steps led to the
constraint violation and have a rather negligible impact on
the objective function, i.e., after all post-processing steps,
the internal structure volume increased from 168.3 cm3 to
175 cm3.
Bars post-processing. In the initial step of module post-
processing, we assign square cross-sections to each bar with
√
the square side length according to the optimal area, de =
ae. Next, we check potential intersection of bars and
place a node at each intersection, which subdivides them
into two and defines the new element lengths. Third, for
each bar, we set the cross-sectional size de to at least le/40,
because more slender bars are difficult to manufacture with
the Prusa 3D printers used in this study. In addition to
the optimized bars, the internal structure is extended with
short L-shaped beams that ensure mechanical interaction
between the internal structure and the steel mandrel, thus
defining an empty 20.05× 20.05 mm space along the beam
longitudinal axis x for its insertion, see Figs. 9 and 8a.
Segmentation. To enable parallel manufacturing with con-
ventional 3D printers, we split the optimized internal
structure into 48 segments of approximately 20 mm in
Figure 6: Segmentation of the beam internal structure.
6
re-alignment
−→
(a)
re-alignment
−→
left end
center
right end
left end
center
right end
(b)
Figure 7: Illustration of bar cross-sections re-alignment along the
beam (a) yz section, and (b) longitudinal axis x.
length, see Fig. 6 where ten selected segments are shown,
to be assembled later on the steel mandrel. Such seg-
mentation requires re-alignment of bars within each beam
cross-section and along the beam longitudinal axis to en-
sure the correct external beam dimension and a clearly
defined interface among adjacent modules, see Fig. 7 for
an illustration. Note that segment production does not
require any supporting material when printed along the
beam longitudinal axis x, which would be impossible when
printing the internal structure as a single-piece product.
Solid conversion. Axial model conversion is performed in-
dependently and in parallel for each node of the ground
Figure 9: Solid models of typical topologies of segments.
structure. We determine first all bars attached to the con-
sidered node, elongate them by one half of their cross-
sectional side lengths at both of their ends, and cut the
more distant half of each of these bars off. These half-
bars are then modeled by a mesh-based representation.
Geometries of individual nodes then result from the mesh-
boolean operations performed with the Cork3 library. Fi-
nally, the overall segment geometry consists of the union
of all nodal geometries, see Fig. 9 for typical topologies of
post-processed segments, and can be readily exported to
patch-based Stl file format, for example.
4. Results
4.1. Manufacturing
After the automated export of the optimized internal
structure into Stl format, the part was additively man-
ufactured using the Fused Deposition Modeling method
with Prusa i3 MK3 printers. Printed segments were in-
serted on a 20 × 20 × 1, 200 mm steel mandrel of 1.5 mm
wall thickness, with its surface lubricated with Vaseline to
simplify the pull-out process, and connected with acetone
etching and a thin layer of epoxy glue.
The prototype beam was produced by CompoTech Plus
company using the filament winding technology with axial
3https://github.com/gilbo/cork
a
d
b
e
(c)
(d)
(e)
(a)
(b)
Figure 8: Endoscope camera photographs (a), (b), (d) and (e) as captured in the manufactured beam interior (c) showing that the 3D-printed
internal structure successfully withstood the compression-molding loads.
7
Figure 10: Manufactured prototype of a composite beam with opti-
mized stiffening internal structure.
fiber placement. This technology relies on the position-
ing of the tows of carbon fibers impregnated by the epoxy
resin on the casing, placed in specified directions and spec-
ified quantity to reach expected dimensions and mechan-
ical properties of the final product. The casing defines
the internal shape of the beam and acts as an internal
mold. After the fiber placement operation, the product
(with still liquid resin) is placed into the press, the outer
shape is formed, and the composite is consolidated. In the
press, the product hardens at the room temperature. Fi-
nally, the prototype, Fig. 10, is postcured at the elevated
temperature of 90◦C.
The successful manufacturing process was followed by
inspection of the prototype using an endoscope camera.
Video and photograph sequences, see Fig. 8, revealed that
the internal structure successfully withstood the compres-
sion molding pressure without any significant visual de-
fects. The minor deviations from the assumed model re-
side in a small amount of Vaseline residue and slight leak-
age of the epoxy resin through casing interfaces. Another
difference appeared in the increased outer dimensions of
the beam, 0.32 mm on average, caused by an insufficiently
closed press cover. The total prototype weight of 1.768 kg
was therefore 186 g higher than the model predictions due
to the additional epoxy resin.
4.2. Verification
Recall that in the optimization we required increas-
ing the fundamental free-vibration eigenfrequency above
300 Hz. To check this value, we employed an independent
model in Ansys. Compared to the model used for op-
timization, this model employs the dimensions measured
in-situ, more refined discretization of the outer shells (ele-
ment type Shell181), and models the internal structure
with beam elements (Beam188) instead of trusses. Be-
sides, the composite shells are supplemented with an ad-
ditional layer of epoxy resin to account for the increased
epoxy content. As a result, the model predicts that the
beam fundamental eigenfrequency was increased by 92%
from 128.5 Hz to 246.7 Hz, compare Figs. 2 and 11. The
effect of wall instabilities was reduced jointly in all the
remaining eigenmodes (not shown).
Even though the fundamental eigenfrequency did not
exceed the limit value, we find these results satisfactory be-
cause of two reasons: First, we attribute this discrepancy
(a) First eigenmode with a fre-
quency of 246.7 Hz.
(b) Second eigenmode with a fre-
quency of 347.0 Hz.
Figure 11: Axonometric and front view on the (a) first and (b) sec-
ond eigenmodes of the reinforced composite beam predicted by the
refined finite element model.
mainly to the manufacturing imperfections, which can be
attributed to the prototype character of the manufactur-
ing process and can be easily resolved in serial production.
Second, the constraint violation is comparable to the dif-
ference between numerics and experiments as shown in the
next section.
4.3. Validation
Dynamic response was validated with the roving ham-
mer test in the free-free-vibration setting because it elim-
inates the need to reproduce the simply supported kine-
matic boundary condition in the experiment. To this goal,
the beam was suspended at one of its ends, three piezoelec-
tric acceleration transducers Type 4507B005 Bruel&Kjaer
were placed on the beam's outer surface, two of which
were located in the middle of adjacent sides of the beam's
cross-section at one-eighth of the beam's length, and the
third one was placed at the corner of the beam, Fig. 12.
Two adjacent sides of the beam surface were marked with
21
3
Figure 12: Free-free-vibration validation setup. Locations of 54 im-
pact points are indicated by gray squares and positions of 3 ac-
celerometers are marked by white circles.
8
a regularly spaced grid of 54 points, 27 on each side. These
points then served as the excitation points for the impact
hammer Type 8206 Bruel&Kjaer equipped with a force
transducer.
Measurement was realized using data acquisition front-
end hardware Type 3560B Bruel&Kjaer.
The fre-
quency response functions (FRFs) were evaluated from the
recorded response (acceleration) and excitation (force) us-
ing the Fast Fourier Transform for all 54 points. The nat-
ural frequencies and mode shapes were evaluated from the
FRFs with MEscope software developed by the Vibrant
Technology company.
Experimentally determined natural modes and the val-
ues of natural eigenfrequencies, Fig. 13 top, were compared
with the results of numerical simulations, Fig 13 bottom.
Direct comparison in Table 2 reveals sufficient agreement
of up to 9% for eigenfrequencies of shear, bending, and
torsional eigenmodes.
In the case of buckling, we failed
to measure the first and second buckling natural modes,
and for the higher eigenmodes, the model predictions un-
derestimate the natural frequencies by more than 20%.
We attribute these deviations to the overall difficulty of
measuring the buckling natural modes and to the manu-
facturing defects discussed in the previous section.
Our methodology was verified by designing and pro-
ducing the simply-supported CFRP beam prototype. Op-
timization yielded an internal structure of 488 g which in-
creased the fundamental eigenfrequency by 92% and lim-
ited the effect of wall instabilities. Moreover, the deflec-
tions within the compression-molding load case were lim-
ited to ±0.5 mm.
After a successful prototype production, the structural
response was validated using the roving hammer test,
which showed that bending, torsional, and shear eigen-
modes exhibited good agreement with model predictions.
For the wall buckling eigenmodes, however, the finite el-
ement model underestimated the natural frequencies by
almost 22%. We attribute this to difficulties in measuring
these natural modes and to manufacturing defects associ-
ated with compression-molding deformations of the casing.
Improving the structural response with a material more
than two orders of magnitude more compliant when com-
pared to CFRP suggests concentrating on substituting
ABS with high-stiffness continuous carbon fiber in fu-
ture studies. Another essential future enhancement resides
in accelerating the optimization algorithm by exploiting
the range-space sparsity [56] associated with the segment-
based internal-structure decomposition.
5. Summary and outlook
Acknowledgments
This contribution introduces and investigates a unique,
fully-automatized procedure from an idea to prototyp-
ing, with applications to the manufacturing of thin-walled
structural composite hollow beams. In particular, the con-
sidered prototype product is stiffened with a low weight
internal structure designed by an efficient convex lin-
ear semidefinite programming formulation. This formu-
lation increased the fundamental free-vibration eigenfre-
quency above a specified threshold value while avoiding
the traditional issue of non-differentiability of multiple
eigenvalues [45], and limited structural compliance of a
compression-molding load case. The optimization output
of the non-uniformly distributed lattice-like internal struc-
ture was further automatically post-processed and con-
verted into a solid model ready for support-less additive
manufacturing.
Table 2: Comparison of model prediction of eigenfrequencies fFEM
and measured natural frequencies fEXP using the roving hammer
test. Accuracy of individual measurements is denoted by A and the
deviation of the model from the experiment by D.
Eigenmode
First shear
First bending y
First bending z
Third buckling
First torsion
Fourth buckling
Fifth buckling
Sixth buckling
Second bending z
fFEM [Hz]
600.1
747.0
748.2
682.3
833.7
708.9
741.4
790.6
1020.0
fEXP [Hz] A [Hz] D [%]
−8.8
+4.6
+3.3
−19.3
−3.5
−20.9
−21.3
−21.3
−7.4
658
714
724
846
864
896
942
1004
1102
2
2
2
4
4
4
6
6
6
We thank Edita Dvor´akov´a for providing us with her
implementation of the Mitc4 shell elements [57], and
Ondrej Rokos and Stephanie Krueger for a critical review
of the initial versions of this manuscript.
The work of Jan Nov´ak and Robin Poul was supported
by the Technology Agency of the Czech Republic, through
the project TA CR TH02020420. Marek Tyburec, Jan
Zeman, and Matej Leps acknowledge the support of the
Czech Science Foundation project No. 19-26143X.
Access to computing and storage facilities owned by
parties and projects contributing to the National Grid In-
frastructure MetaCentrum provided under the program
"Projects of Large Research, Development, and Innova-
tions Infrastructures" (CESNET LM2015042), is greatly
appreciated.
Data availability
The raw/processed data required to reproduce these
findings cannot be shared at this time due to legal or eth-
ical reasons.
Appendix A. Static condensation of static LMI
Consider the equilibrium equation
K(a)u = f
(10)
9
(a) 658 Hz
(b) 714 Hz
(c) 724 Hz
(d) 846 Hz
(e) 864 Hz
(f) 896 Hz
(g) 942 Hz
(h) 1102 Hz
(a) 600.057 Hz
(b) 747.011 Hz
(c) 748.241 Hz
(d) 682.283 Hz
(e) 833.719 Hz
(f) 708.852 Hz
(g) 741.389 Hz
(h) 1019.972 Hz
Figure 13: Selected experimentally determined natural frequencies and mode shapes, (a) -- (h) top, and finite element model predictions of
eigenmodes and eigenfrequencies, (a) -- (h) bottom.
split into two sets of equations
(cid:18)Ka(a) Kb
(cid:19)(cid:18)ua
KT
b
Kc
ub
(cid:19)
(cid:18)fa
(cid:19)
fb
,
=
(11)
such that only the principal submatrix Ka(a) depends
affinely on a4. Assuming that the system (10) is solvable
uniquely for some a, i.e., it holds that ∃a ≥ 0 : K(a) (cid:31) 0,
where "(cid:31) 0" denotes positive definiteness of the left hand
side. Note that a = 1 is sufficient for verification that no
rigid movement within the structure can occur. Because
Kc is therefore invertible, the degrees of freedom ub can
be expressed from the second row in terms of ua
−1 fb − (Kc)
and inserted back into the first row,
ub = (Kc)
−1 KT
b ua
(12)
Ka(a) − Kb (Kc)
−1 KT
b
ua = fa − Kb (Kc)
−1 fb.
(13)
(cid:105)
(cid:104)
Structural compliance (work done by external forces) is
expressed as
c = uT
a fa + uT
b fb.
After inserting (12) and acknowledging that K−1
mitian, we obtain
c
(14)
is Her-
(cid:104)
(cid:105)
i.e., compliance of the condensed problem (13) and a con-
stant term. Because the compliance of the condensed prob-
lem is positive by definition, the constant term represents
a non-negative lower bound on compliances achievable by
the internal structure design.
(cid:19)
−f T
−f K(a)
Finally, the LMI(cid:18) c
(cid:18) c − f T
b (Kc)
−fa + Kb (Kc)
is equivalent to a smaller LMI
−1 fb
b (Kc)
−1 fb Ka(a) − Kb (Kc)
a + f T
−f T
−1 KT
−1 KT
b
b
(cid:19)
(cid:23) 0. (17)
(cid:23) 0
(16)
b (Kc)
−1 fb is a prescribed constant (i.e.,
Further, if c > f T
not a variable), then (17) is further reducible, using the
Schur complement lemma, e.g., [58, Proposition 16.1], to
a yet smaller LMI
Ka(a) − Kb (Kc)
b (Kc)
b −(cid:16)−f T
(cid:17)−1(cid:16)−fa + Kb (Kc)
a + f T
−1 KT
−1 fb
(cid:17)
(cid:17) (cid:23) 0.
−1 KT
−1 fb
c − f T
(cid:16)
(18)
b (Kc)
b
c = uT
a
fa − Kb (Kc)
−1 fb
+ f T
b (Kc)
−1 fb,
(15)
Appendix B. Reducing size of free-vibration LMI
4In the context of this article, the matrix Ka(a) comprises the
degrees of freedom of the truss ground structure, Kc contains the
remaining (rotational) degrees of freedom, and Kb is the coupling
term.
In the case of the free-vibration constraint, we need to
directly apply the (generalized) Schur complement lemma.
Beginning with reordering of rows and columns, we split
the symmetric LMI (9b) such that only the Ka(a) and
10
(cid:32)
Ma(a) matrices are functions of a, and the other blocks
are constant,
Ka(a) − 4π2f
b − 4π2f
KT
2
2
Ma(a) Kb − 4π2f
Kc − 4π2f
MT
b
2
2
Mb
Mc
(cid:23) 0.
(19)
2
For the (standard) Schur complement trick we require
Mc (cid:31) 0 [58, Proposition 16.1]. Since Kc (cid:31) 0
Kc − 4π2f
(boundary conditions exclude rigid motions), and Mc (cid:31) 0
by definition, we only need to secure that the fundamental
eigenfrequency f0 of the generalized eigenvalue problem
Kcub − λMcub = 0,
(20)
with λ = 4π2f 2, is strictly greater than f .
Let us therefore first assume that 0 ≤ f < f0. Then,
the inverse of Kc−4π2f
Mc exists and (19) can be rewrit-
ten equivalently using the Schur complement lemma into
a smaller-sized LMI
2
Ma(a) −(cid:16)
(cid:17)−1(cid:16)
2
Ka(a) − 4π2f
2
(cid:16)
Kc − 4π2f
Mc
2
Kb − 4π2f
b − 4π2f
KT
Mb
2
MT
b
(cid:17) (cid:23) 0.
(21)
(cid:33)
(cid:17)
Second, consider that f0 < f . Because the matrix
Kc− 4π2(f0 + ε)2Mc is indefinite for any ε > 0, which ren-
ders the original LMI (19) infeasible, the eigenfrequency
f0 constitutes an upper bound for achievable fundamen-
tal eigenfrequencies of the reinforced structure. From the
mechanical point of view, the eigenmodes ub associated
with f0 excite degrees of freedom not reinforced by the
internal structure, and therefore the associated eigenfre-
quencies can not be increased by any admissible internal
structure design (given the specific discretization).
In the case f = f0, reduction of (19) relies on the
generalized Schur complement lemma [58, Theorem 16.1],
so that (19) is equivalent to
2
(cid:16)
(cid:20)
I −(cid:16)
Ka(a) − 4π2f
Kc − 4π2f
Kc − 4π2f
Ma(a) −(cid:16)
(cid:17)†(cid:16)
(cid:17)(cid:16)
(cid:16)
Mc
Mc
2
2
(cid:17)
(cid:17) (cid:23) 0,
(cid:17)†(cid:21)
(cid:17)
= 0,
2
Kb − 4π2f
b − 4π2f
KT
Kc − 4π2f
b − 4π2f
KT
2
2
2
Mb
MT
b
Mc
MT
b
(22a)
(22b)
†
where (•)
denotes the Moore-Penrose pseudo-inverse of •,
and I is the identity matrix. The second condition (22b)
holds iff the columns of KT
b are in the image
of Kc − 4π2f
Mc. Indeed, (22b) can then be rewritten to
b − 4π2f
MT
2
2
(cid:104)(cid:16)
(cid:16)
(cid:17) −(cid:16)
(cid:17)†(cid:16)
Kc − 4π2f
2
Mc
Kc − 4π2f
2
Mc
2
Kc − 4π2f
Kc − 4π2f
Mc
2
Mc
(cid:17)
(cid:17)(cid:21)
(23)
C = 0.
11
with the columns of C being the coefficients of linear com-
binations of the columns of Kc − 4π2f
Mc, making the
term in the square brackets vanish [58, Lemma 14.1].
2
Because Im(Kc − 4π2f
2
Mc) = Ker(Kc − 4π2f
2
Mc)⊥
by [58, Lemma 13.1], it is spanned by
(cid:110)
(cid:16)
(cid:17)
(cid:111)⊥
span
ub :
Kc − 4π2f
2
Mc
ub = 0
.
(24)
2
MT
b −4π2f
Clearly, f = f0 might be achieved iff the columns of the
coupling term KT
b are orthogonal to the eigen-
modes occurring in (20) at f0. From the mechanical point
of view,
induction of these eigenmodes would result in
a decrease of the associated eigenfrequencies. Note that
in practice, equation (22b) can be verified numerically,
but it does not guarantee a feasible solution to (22a),
because other (higher) eigenfrequencies associated with
eigenmodes of (20) may decrease below f0 due to the cou-
pling term.
References
[1] K. Friedrich, A. A. Almajid, Manufacturing aspects of advanced
polymer composites for automotive applications, Applied Com-
posite Materials 20 (2) (2013) 107 -- 128, ISSN 0929189X, doi:
10.1007/s10443-012-9258-7.
[2] K. Challis, P. Burchill, A. P. Mouritz, E. Gellert, Review
of advanced composite structures for naval ships and sub-
marines, Composite Structures 53 (1) (2001) 21 -- 42, doi:10.
1016/S0263-8223(00)00175-6.
[3] Y. L. Young, M. R. Motley, R. Barber, E. J. Chae, N. Garg,
Adaptive Composite Marine Propulsors and Turbines: Progress
and Challenges, Applied Mechanics Reviews 68 (6) (2016)
060803, ISSN 0003-6900, doi:10.1115/1.4034659.
[4] E. Irving, Polymer composites in the aerospace industry, Wood-
head Publishing, Kidlington, UK, ISBN 978-0-85709-523-7, doi:
10.1016/C2013-0-16303-9, 2015.
[5] V. V. Vasiliev, V. A. Barynin, A. F. Razin, Anisogrid compos-
ite lattice structures - Development and aerospace applications,
Composite Structures 94 (3) (2012) 1117 -- 1127, ISSN 02638223,
doi:10.1016/j.compstruct.2011.10.023.
[6] H. Ghiasi, D. Pasini, L. Lessard, Optimum stacking sequence
design of composite materials Part I: Constant stiffness design,
Composite Structures 90 (1) (2009) 1 -- 11, ISSN 02638223, doi:
10.1016/j.compstruct.2009.01.006.
[7] H. Ghiasi, K. Fayazbakhsh, D. Pasini, L. Lessard, Optimum
stacking sequence design of composite materials Part II: Vari-
able stiffness design, Composite Structures 93 (1) (2010) 1 -- 13,
ISSN 02638223, doi:10.1016/j.compstruct.2010.06.001.
[8] M. P. Bendsøe, O. Sigmund, Topology optimization: The-
ory, methods, and applications, Springer Berlin Heidel-
berg, Berlin, Heidelberg, ISBN 978-3-642-07698-5, doi:10.1007/
978-3-662-05086-6, 2003.
[9] Y. Y. Kim, T. S. Kim, Topology optimization of beam cross
sections, International Journal of Solids and Structures 37 (3)
(2000) 477 -- 493, ISSN 00207683, doi:10.1016/S0020-7683(99)
00015-3.
[10] J. P. Blasques, Multi-material topology optimization of lami-
nated composite beams with eigenfrequency constraints, Com-
posite Structures 111 (2014) 45 -- 55, ISSN 02638223, doi:10.
1016/j.compstruct.2013.12.021.
[11] H.-D. Nguyen, G.-W. Jang, D.-M. Kim, Y. Y. Kim, Finite
prism method based topology optimization of beam cross sec-
tion for buckling load maximization, Structural and Multidis-
ciplinary Optimization 57 (1) (2018) 55 -- 70, ISSN 1615-147X,
doi:10.1007/s00158-017-1860-8.
[12] J. R. Vinson, Structural optimization to obtain minimum weight
sandwich panels, in: Plate and Panel Structures of Isotropic,
Composite and Piezoelectric Materials, Including Sandwich
Construction, vol. 120 of Solid Mechanics and Its Applica-
tions, chap. 17, Springer, Dordrecht, 345 -- 377, doi:10.1007/
1-4020-3111-4 17, 2005.
[13] A.-J. Wang, D. McDowell, Optimization of a metal honeycomb
sandwich beam-bar subjected to torsion and bending, Interna-
tional Journal of Solids and Structures 40 (9) (2003) 2085 -- 2099,
ISSN 00207683, doi:10.1016/S0020-7683(03)00033-7.
[14] M. Xu, Z. Qiu, Free vibration analysis and optimization of com-
posite lattice truss core sandwich beams with interval param-
eters, Composite Structures 106 (2013) 85 -- 95, ISSN 02638223,
doi:10.1016/j.compstruct.2013.05.048.
[15] S. Daynes, S. Feih, W. F. Lu, J. Wei, Optimisation of func-
tionally graded lattice structures using isostatic lines, Ma-
terials & Design 127 (2017) 215 -- 223, ISSN 02641275, doi:
10.1016/j.matdes.2017.04.082.
[16] V. Birman, G. A. Kardomateas, Review of current trends in
research and applications of sandwich structures, Composites
Part B: Engineering 142 (2018) 221 -- 240, ISSN 13598368, doi:
10.1016/j.compositesb.2018.01.027.
[17] M. Helou, S. Kara, Design, analysis and manufacturing of lattice
structures: An overview, International Journal of Computer In-
tegrated Manufacturing 31 (3) (2018) 243 -- 261, ISSN 13623052,
doi:10.1080/0951192X.2017.1407456.
[18] P. G. Coelho, H. C. Rodrigues, Hierarchical topology opti-
mization addressing material design constraints and applica-
tion to sandwich-type structures, Structural and Multidisci-
plinary Optimization 52 (1) (2015) 91 -- 104, ISSN 1615-147X,
doi:10.1007/s00158-014-1220-x.
[19] N. Olhoff, J. E. Taylor, On structural optimization, Journal of
Applied Mechanics 50 (4b) (1983) 1139, ISSN 00218936, doi:
10.1115/1.3167196.
[20] A. R. D´ıaaz, N. Kikuchi, Solutions to shape and topology eigen-
value optimization problems using a homogenization method,
International Journal for Numerical Methods in Engineering
35 (7) (1992) 1487 -- 1502, ISSN 0029-5981, doi:10.1002/nme.
1620350707.
[21] Z. D. Ma, N. Kikuchi, I. Hagiwara, Structural topology and
shape optimization for a frequency response problem, Compu-
tational Mechanics 13 (3) (1993) 157 -- 174, ISSN 0178-7675, doi:
10.1007/BF00370133.
[22] M. P. Bendsøe, A. Ben-Tal, J. Zowe, Optimization methods for
truss geometry and topology design, Structural and Multidis-
ciplinary Optimization 7 (3) (1994) 141 -- 159, ISSN 0934-4373,
doi:10.1007/bf01742459.
[23] V. Balabanov, R. T. Haftka, Topology optimization of transport
wing internal structure, Journal of Aircraft 33 (1) (1996) 232 --
233, ISSN 0021-8669, doi:10.2514/3.46926.
[24] M. M. Opgenoord, K. E. Willcox, Aeroelastic tailoring using
additively manufactured lattice structures, in: 2018 Multidisci-
plinary Analysis and Optimization Conference, American Insti-
tute of Aeronautics and Astronautics, Reston, Virginia, ISBN
978-1-62410-550-0, doi:10.2514/6.2018-4055, 2018.
[25] J. Luo, H. C. Gea, A systematic topology optimization approach
for optimal stiffener design, Structural Optimization 16 (4)
(1998) 280 -- 288, ISSN 0934-4373, doi:10.1007/BF01271435.
[26] L. Wang, P. K. Basu, J. P. Leiva, Automobile body reinforce-
ment by finite element optimization, Finite Elements in Anal-
ysis and Design 40 (8) (2004) 879 -- 893, ISSN 0168874X, doi:
10.1016/S0168-874X(03)00118-5.
[27] K. Maute, M. Allen, Conceptual design of aeroelastic struc-
tures by topology optimization, Structural and Multidisci-
plinary Optimization 27 (1-2) (2004) 27 -- 42, ISSN 1615-147X,
doi:10.1007/s00158-003-0362-z.
[28] N. Aage, E. Andreassen, B. S. Lazarov, O. Sigmund, Giga-
voxel computational morphogenesis for structural design, Na-
ture 550 (7674) (2017) 84 -- 86, doi:10.1038/nature23911.
[29] B. K. Stanford, P. D. Dunning, Optimal topology of air-
craft rib and spar structures under aeroelastic loads, Jour-
nal of Aircraft 52 (4) (2015) 1298 -- 1311, ISSN 0021-8669, doi:
10.2514/1.C032913.
[30] Z. A. Provchy, A. N. Palazotto, P. J. Flater, Topology op-
timization for projectile design, Journal of Dynamic Behav-
ior of Materials 4 (1) (2018) 129 -- 137, ISSN 2199-7446, doi:
10.1007/s40870-018-0143-9.
[31] M. Rais-Rohani, J. Lokits, Reinforcement layout and sizing op-
timization of composite submarine sail structures, Structural
and Multidisciplinary Optimization 34 (1) (2007) 75 -- 90, ISSN
1615-147X, doi:10.1007/s00158-006-0066-2.
[32] A. Clausen, N. Aage, O. Sigmund, Topology optimization of
coated structures and material interface problems, Computer
Methods in Applied Mechanics and Engineering 290 (2015) 524 --
541, ISSN 00457825, doi:10.1016/j.cma.2015.02.011.
[33] A. Clausen, N. Aage, O. Sigmund, Exploiting additive manu-
facturing infill in topology optimization for improved buckling
load, Engineering 2 (2) (2016) 250 -- 257, ISSN 20958099, doi:
10.1016/J.ENG.2016.02.006.
[34] A. Clausen, E. Andreassen, O. Sigmund, Topology optimiza-
tion of 3D shell structures with porous infill, Acta Mechan-
ica Sinica 33 (4) (2017) 778 -- 791, ISSN 0567-7718, doi:10.1007/
s10409-017-0679-2.
[35] J. Wu, A. Clausen, O. Sigmund, Minimum compliance topology
optimization of shell-infill composites for additive manufactur-
ing, Computer Methods in Applied Mechanics and Engineering
326 (2017) 358 -- 375, ISSN 00457825, doi:10.1016/j.cma.2017.08.
018.
[36] Y. Wang, L. Zhang, S. Daynes, H. Zhang, S. Feih, M. Y. Wang,
Design of graded lattice structure with optimized mesostruc-
tures for additive manufacturing, Materials & Design 142 (2018)
114 -- 123, ISSN 02641275, doi:10.1016/j.matdes.2018.01.011.
[37] Y. Zhu, S. Li, Z. Du, C. Liu, X. Guo, W. Zhang, A novel
asymptotic-analysis-based homogenisation approach towards
fast design of infill graded microstructures, Journal of the
Mechanics and Physics of Solids 124 (2019) 612 -- 633, ISSN
00225096, doi:10.1016/j.jmps.2018.11.008.
[38] M. Livesu, S. Ellero, J. Mart´ınez, S. Lefebvre, M. Attene,
From 3D models to 3D prints: An overview of the process-
ing pipeline, Computer Graphics Forum 36 (2) (2017) 537 -- 564,
ISSN 14678659, doi:10.1111/cgf.13147.
[39] J. Wu, N. Aage, R. Westermann, O. Sigmund, Infill optimiza-
tion for additive manufacturing -- Approaching bone-like porous
structures, IEEE Transactions on Visualization and Computer
Graphics 24 (2) (2018) 1127 -- 1140, ISSN 1077-2626, doi:10.
1109/TVCG.2017.2655523.
[40] J. P. Groen, J. Wu, O. Sigmund, Homogenization-based stiff-
ness optimization and projection of 2D coated structures with
orthotropic infill, Computer Methods in Applied Mechanics
and Engineering 349 (2019) 722 -- 742, doi:10.1016/j.cma.2019.
02.031.
[41] A. Ben-Tal, A. Nemirovski, Structural Design, in: H. Wolkow-
icz, R. Saigal, L. Vandenberghe (Eds.), Handbook of semidef-
inite programming, chap. 15, Springer, Boston, MA, 443 -- 467,
doi:10.1007/978-1-4615-4381-7 15, 2000.
[42] E. de Klerk, C. Roos, T. Terlaky, Semi-definite prob-
topology
95-
https://pdfs.semanticscholar.org/dd6b/
lems
128),
8a14a816adba3aa31c6c8553ef51f2219c6f.pdf.
optimization
(Report
in
truss
URL
[43] L. Vandenberghe, S. Boyd, Semidefinite programming, SIAM
Review 38 (1) (1996) 49 -- 95, ISSN 0036-1445, doi:10.1137/
1038003.
[44] M. Ohsaki, K. Fujisawa, N. Katoh, Y. Kanno, Semi-definite pro-
gramming for topology optimization of trusses under multiple
eigenvalue constraints, Computer Methods in Applied Mechan-
ics and Engineering 180 (1-2) (1999) 203 -- 217, ISSN 00457825,
doi:10.1016/S0045-7825(99)00056-0.
[45] W. Achtziger, M. Kocvara, On the maximization of the fun-
damental eigenvalue in topology optimization, Structural and
Multidisciplinary Optimization 34 (3) (2007) 181 -- 195, ISSN
1615147X, doi:10.1007/s00158-007-0117-3.
[46] A. Ben-Tal, F. Jarre, M. Kocvara, A. Nemirovski, J. Zowe, Op-
12
timal design of trusses under a nonconvex global buckling con-
straint, Optimization and Engineering 1 (2) (2000) 189 -- 213,
doi:10.1023/A:1010091831812.
[47] M. Kocvara, On the modelling and solving of the truss de-
sign problem with global stability constraints, Structural and
Multidisciplinary Optimization 23 (3) (2002) 189 -- 203, ISSN
1615147X, doi:10.1007/s00158-002-0177-3.
[48] A. Ben-Tal, A. Nemirovski, Robust truss topology design
via semidefinite programming, SIAM Journal on Optimiza-
tion 7 (4) (1997) 991 -- 1016,
ISSN 1052-6234, doi:10.1137/
S1052623495291951.
[49] A. Cerveira, F. Bastos, V. Real, Semidefinite relaxations and
Lagrangean duality in truss topology design problem, Interna-
tional Journal of Mathematics and Statistics 9 (2011) 12 -- 25,
URL http://www.ceser.in/ceserp/index.php/ijms/article/
view/2759.
[50] A. Ben-Tal, M. Kocvara, A. Nemirovski, J. Zowe, Free material
design via semidefinite programming: The multiload case with
contact conditions, SIAM Journal on Optimization 9 (4) (1999)
813 -- 832, ISSN 1052-6234, doi:10.1137/S1052623497327994.
[51] C. D. Bisbos, P. M. Pardalos, Second-order cone and semidefi-
nite representations of material failure criteria, Journal of Opti-
mization Theory and Applications 134 (2) (2007) 275 -- 301, ISSN
0022-3239, doi:10.1007/s10957-007-9243-8.
[52] J. T. Cantrell, S. Rohde, D. Damiani, R. Gurnani, L. DiSandro,
J. Anton, A. Young, A. Jerez, D. Steinbach, C. Kroese, P. G.
Ifju, Experimental characterization of the mechanical properties
of 3D-printed ABS and polycarbonate parts, Rapid Prototyping
Journal 23 (4) (2017) 811 -- 824, ISSN 1355-2546, doi:10.1108/
RPJ-03-2016-0042.
[53] E. N. Dvorkin, K.-J. Bathe, A continuum mechanics based four-
node shell element for general non-linear analysis, Engineering
Computations 1 (1) (1984) 77 -- 88, ISSN 0264-4401, doi:10.1108/
eb023562.
[54] W. S. Dorn, Automatic design of optimal structures, Journal
de Mecanique 3 (1964) 25 -- 52.
[55] MOSEK ApS, The MOSEK optimization toolbox for MAT-
LAB manual. Version 8.1., URL http://docs.mosek.com/8.1/
toolbox/index.html, 2017.
[56] S. Kim, M. Kojima, M. Mevissen, M. Yamashita, Exploit-
ing sparsity in linear and nonlinear matrix inequalities via
positive semidefinite matrix completion, Mathematical Pro-
gramming 129 (1) (2011) 33 -- 68, ISSN 0025-5610, doi:10.1007/
s10107-010-0402-6.
[57] E. Dvor´akov´a, Finite elements for analysis of plates and
shells, Master's thesis, Czech Technical University in Prague,
Prague, Czech Republic, URL https://mech.fsv.cvut.cz/
wiki/images/f/f3/DP_Dvorakova.pdf, 2015.
[58] J. Gallier, Geometric Methods and Applications, Springer New
York, doi:10.1007/978-1-4419-9961-0, 2011.
13
|
1907.07183 | 1 | 1907 | 2019-07-16T17:27:41 | Magnetic, Structural and cation distribution studies on $FeO \cdot Fe_{(2-x)}Nd_{x}O_{3}$ ($x=0.00, 0.02, 0.04, 0.06 \text{ and } 0.1$) nanoparticles | [
"physics.app-ph",
"cond-mat.soft"
] | We synthesized and characterized the colloidal suspensions of $FeO \cdot Fe_{(2-x)}Nd_{x}O_{3}$ nanoparticles with $x=0.00, 0.02, 0.04, 0.06 \text{ and }0.1.$ The effect of the $Fe^{3+}$ ion replacement by $Nd^{3+}$ on the crystal structure is in-depth studied, through X-ray diffraction (XRD) and the obtained cation distribution. The magnetic properties of the synthesized $FeO \cdot Fe_{(2-x)}Nd_{x}O_{3}$ nanoparticles also were investigated and corroborated by other physical methods. A remarkable saturation magnetization of 105 $Am^{2}/kg$ was achieved for $x=0.06$. | physics.app-ph | physics | Magnetic, Structural and cation distribution studies on FeO·Fe(2-x)NdxO3
(x=0.00, 0.02, 0.04, 0.06 and 0.1) nanoparticles.
W. W. R. Araujo*, C. L. P. Oliveira, G. E. S. Brito, A. M. Figueiredo Neto
Institute of Physics, University of São Paulo, Rua do Matão 1371, 05508-090, São Paulo, SP,
Brazil.
Department of Physics, Pontifical Catholic University of Rio de Janeiro, RJ, Brazil.
J. F. D. F. Araujo
Abstract
We synthesized and characterized the colloidal suspensions of FeO·Fe(2-x)NdxO3
nanoparticles with x=0.00, 0.02, 0.04, 0.06 and 0.1. The effect of the Fe3+ ion replacement by Nd3+
on the crystal structure is in-depth studied, through X-ray diffraction (XRD) and the obtained cation
distribution. The magnetic properties of the synthesized FeO·Fe(2-x)NdxO3 nanoparticles also were
investigated and corroborated by other physical methods. A remarkable saturation magnetization of
105 Am2/kg was achieved for x=0.06.
Keywords: Magnetic nanoparticles; Characterization; Synthesis; Magnetite; Iron oxide; ZFC-FC;
SAXS
PACS number(s): 82.70.Dd, 75.50.Mm, 82.20.Wt, 75.75.−c, 75.60.−d, 61.10.Eq
I. Introduction
There has been an enormous research effort recently in colloidal magnetic nanocrystals, so
that hundreds of papers are published per year in this field of scientific research [1]. These
nanocrystals can be obtained with controlled size, shape and composition [2, 3]. Colloidal magnetic
nanocrystals (CMNs) attract increasing interest both in fundamental sciences and in technological
applications [4]. Colloidal magnetic nanocrystals (CMNs) are suspensions of magnetic
nanoparticles (MNPs) dispersed in a liquid carrier [4, 5]. When the particle dimension is small
(typically ~10 nm), the particle presents a single magnetic domain, with a large magnetic moment
called superspin [6-8]. The surface of MNPs can be modified by several stabilizing agents or
functional groups. Once, the surface of the magnetic nanoparticles is modified, they become highly
functional materials. Some of the MNPs physical properties can be controlled by external magnetic
fields or magnetic-field gradients [9, 10], that is, they are stimuli-responsive systems. Some
applications of ferrofluids are separation media [1], heat-conduction media [4], gas fluidized beds
[1], sealants [2, 7, 11] and hydraulic car suspensions [7]. MNPs can also be used for drug delivery,
medical diagnosis, and cell destruction [4]. One of the peculiarities of a ferrofluid is how its
nanoscopic organization is affected by an applied external magnetic field [10]. In the field of
1
magneto-optical devices some applications can be found, for integrated optics [9, 11], optical fibers
[9, 11] and tunable beam splitter [11].
Magnetic nanoparticles are free to move when dispersed in a liquid carrier medium, and
different physical event may occur as a function of an applied magnetic field [12]. The blocking
temperature (TB) defines when the system of MNPs passes from blocked to superparamagnetic
(SPM) state [12, 13]. The Zero-field cooled (ZFC) and Field cooled (FC) technique is widely used
to study granular materials. ZFC-FC curves allows the determination of the blocking temperature of
the system TB [12, 13]. The Small Angle X-ray Scattering is used to study structural parameters of
fluid samples from the analysis of the experimental scattering pattern [2, 9]. For polydisperse
systems, the experimental scattering pattern corresponds to an average that is performed over the
particles in the solution. Thus, the experimentally reached values correspond to an average over the
entire ensemble of particles rather than a single particle [2, 9]. To tackle this problem, ZFC-FC
curves complemented by SAXS scattered intensities could supply several insights about the MNPs
investigated.
‒
‒
Rare-Earth doping significantly alters the nucleation and growth of nanoferrites, which
facilitate magnetic spin orientation [2, 14]. It is known that the magnetic behavior of spinel ferrite
compounds is mostly due to the interaction between the iron atoms [14]. Rare-Earth ions are more
favorable to enter in octahedral sites of the spinel structure; this causes 4f-3d interactions that
promotes structural distortion, lattice strain and changes in saturation magnetization [14]. The
literature reports the use of Neodymium (Nd) for doping copper nanoferrites [14]. Mixed
manganese neodymium copper (Mn -- Nd -- Cu) nanoferrites was produced by sonochemical method
[14]. Aslam and co-workers performed a co-doping of Nd3+ and Pr3+ on lithium nanoferrite and
reported the effects on the magnetic and structural properties of the system [15]. The effect of Nd+3
doping on Mn-Zn ferrite was reported in the literature [16, 17], and an enhancement of the
saturation magnetization, due to the new cation distribution imposed by Nd doping was reported.
Jain and co-workers studied the influence of rare earth ions on structural, magnetic and optical
properties of magnetite nanoparticles [18]. Jain also reported that there is a variation in the
saturation magnetization maximum value, directly proportional to the number of unpaired 4f
electrons in the dopant element [18]. Huan and co-workers found similar results for RE3+-doped
Fe3O4 samples (RE3+=Ln3+, Eu3+ and Dy3+) [19].
In the present study we report on the synthesis and characterization of colloidal suspensions
of FeO·Fe(2-x)NdxO3 nanoparticles with x=0.00, 0.02, 0.04, 0.06 and 0.1. The samples were
synthesized by co-precipitation method. The characterization was done by: X-ray diffraction
(XRD), Transmission Electron Microscopy (TEM), Small Angle X-ray Scattering (SAXS), Optical
bandgap and Zero-field and Field Cooling magnetization (ZFC-FC). Therefore, the effect of the Fe3+
2
ion replacement by Nd3+ on the crystal structure is in-depth studied, through X-ray diffraction
(XRD) and the obtained cation distribution. The magnetic properties of the synthesized FeO·Fe(2-
x)NdxO3 nanoparticles also were investigated and corroborated by other physical methods.
II. Material and methods
A. Materials reagents
The materials used to obtain the MNPs were: FeCl3·6H2O, ≥99%; FeCl2·4H2O, ≥99%;
NdCl3·6H2O, ≥99%; cis-9-Octadecenoic acid (Oleic acid), ≥99%; NaOH, ≥99%; and kerosene. All
primary materials were acquired from Sigma-Aldrich and used as received.
B. Synthesis procedure by co-precipitation method
The magnetic nanoparticles were synthesized by co-precipitation of an aqueous mixture of
FeCl3 (ferric chloride) and FeCl2 (ferrous chloride) salts and stabilized at pH~12 [11, 20-24].
Briefly, ferric chloride (4.00 mmol) and ferrous chloride (2.00 mmol) in a 2:1 molar ratio and 8 mL
of oleic acid are mixed in 25 mL of deionized water. The solution was heated up to 80 °C and
magnetically stirred for 30 min. Then, 30 mL of NaOH was added to the solution to precipitate the
particles at room conditions, under vigorously stirred for more 30 min. When the pH reaches ~ 11, a
severe reaction occurs and the solution becomes dark brownish. Thereafter, the resultant solution
was cooled to room temperature. At last, the MNPs were precipitated with a permanent magnet and
then washed ten times with deionized water to remove residual unreacted salts. The procedure
described above was used to produce magnetite nanoparticles with x=0.00 and that samples was
called as FF-REF.
The samples FF-ND1, FF-ND2, FF-ND3 and FF-ND5 are labeled according to molar
percentage amount of substitution of Fe3+ ions by Nd3+ ions. These percentages are 1 ,2, 3 and 5%
(x=0.02, 0.04, 0.06 and 0.10), respectively. FF-ND1, FF-ND2, FF-ND3 and FF-ND5 were prepared
with an aqueous solution composed of FeCl3·6H2O ((4 - i) mmol), NdCl3·6H2O (i=(0.04, 0.08, 0.12
and 0.20) mmol), FeCl2·4H2O (2.00 mmol) and 8 mL cis-9-octadecene. Then, 30 mL of NaOH was
added to the solution to precipitate the particles at room conditions, under vigorously stirring for 30
min. The remaining procedures were the same used to prepare the FF-REF sample.
Hereafter, the doped concentration (x=0.02, 0.04, 0.06 and 0.10) of Nd3+ in magnetite
(x=0.00, Fe3O4) will be denoted as FF-ND1, FF-ND2, FF-ND3 and FF-ND5, respectively. All
samples were sterically stabilized with a single oleic acid layer, chemisorbed on the particles'
surfaces. MNPs are dispersed in kerosene.
C. X-ray Diffraction
3
Powder X-ray Diffraction (XRD) patterns were obtained to investigate nanoparticles'
crystalline structure. XRD was done in a Brucker-AXS D8 series 2 diffractometer, set to a Bragg
Brentano Parafocussing Geometry. A Cu Ka source (λ = 1.5414 Å) generated X-rays at room
temperature. The difractometer was operated at 40 kV, 30 mA. The experimental pattern data were
registered in continuous scan mode, scattering angle 2θ from 15 to 80º, in steps of 0.02º.
D. Transmission Electron Microscopy
The overall form (morphology) and size distribution of the MNPs were examined by
transmission electron microscopy (TEM). TEM was performed on a JEOL 1010 (Japan) microscope
with an acceleration voltage of 80 kV. TEM micrographs were acquired by a Gatan Bioscan 782
CDD camera of 1K x 1K pixels. The colloidal suspension of MNPs were prepared by diluting the
original colloidal suspensions 100 times and maintained in an ultrasonic bath for 30 min. A drop of
the colloidal suspension was placed on a Formvar™ coated 200 mesh copper grid. The residual
excessed fluid was blotted and dried until the time that the grids was examined into microscope.
MNPs number-weighted-size distributions were obtained by measuring about ~500 MNPs with the
ImageJ freeware [25]. The number-weighted size-distribution data were fitted to a log-normal
distribution function given by [2]:
(1)
where Dm and s are fitting parameters. The number-weighted mean diameter,
deviation of particle size, , are written as [2]:
and standard
(2)
(3)
E. Small Angle X-ray Scattering
The Small Angle X-ray Scattering (SAXS) patterns of the solutions were investigated. The
setup used to data acquisition was a Xeuss (XenocsTM). The Xeuss is equipped with a microfocus x-
ray source Genix, with radiation of λ = 1.5414 Å (Cu). The system uses two scatterless slits for
beam collimation and it reaches the sample with a square cross section of 0.4 x 0.4 mm 2. The
4
primary and scattered beams remain in a vacuum (10-2 mbar) chamber to avoid scattering by the air.
Each sample were mounted in a cylindrical Mark-tubes of quartz glass capillary (Hilgenberg, 1.5
mm outer diameter) with principal axis in the vertical direction. The measured 2D scattered data
were recorded by a Pilatus (Dectris) 300 K 20 Hz 2D detector. The exposition time was 600 s and
all measurements were performed at room temperature (~ 22 °C).
The 1D scattering intensity versus scattering vector module, I(q) and q, defined by q = (4π
sinθ)/λ was obtained by averaging the data over a 20° slice in horizontal and vertical directions. The
data treatment, blank subtraction, and data normalization were performed with the software
SUPERSAXS (C.L.P. Oliveira and J. S. Pedersen, unpublished). The contribution of blank
(kerosene) was independently measured and subtracted from the sample data. To obtain the
scattered intensity data in absolute-scale units, cm-1, water was used as standard. The sample-
detector distance was 839 cm, which allows measurements with q in the range 0.01 < q < 0.35 Å−1
[2, 9, 12].
F. Zero-field cooling and field-cooling (ZFC-FC)
The magnetic properties of the MNPs were measured in a home-made Hall-effect
magnetometer [26, 27]. The magnetization curves were performed at room temperature and low
temperatures (6K) and under different applied magnetic fields from −2.0 to +2.0 T. The DC
magnetization as a function of temperature was measured in both zero-field cooled (ZFC) and field
cooled (FC) regimes. ZFC-FC protocol was performed in a temperature range from 5 to 300 K,
applying a 5.0 mT magnetic field [26, 27].
III. Results and Discussion
A. X-ray Diffraction (XRD)
XRD patterns of the samples investigated are shown in FIG. 1. These results reveal a single-
phase cubic spinel structure corresponding to the Fd3m space group [28]. The XRD pattern show
diffraction peaks corresponding to the (220), (311), (400), (422), (511), (440), (620) and (533)
crystallographic planes. These results agree with the XRD pattern of the Fe3O4.
The instrumental broadening (βhkl) was corrected, using Warren's relation
[29]. The average crystallite size (CS) of all the samples has been
estimated using full width at half maximum (FWHM) of each diffraction peak and the Debye-
Scherrer formula CS= 0.9 λ
βhkl cos(θ)
[ 28-31], where λ is the X-ray wavelength (Cu Kα, λ = 1.5414
Å), 2θ is the Bragg angle and βhkl is the FWHM of the diffraction peaks. The lattice parameter was
5
[28, 29]. The observed XRD patterns of all the
investigated using the equation
samples were analyzed by Rietveld method [32], using the MAUD 2.80 software [33] to get the
refinement parameters [28-32]. The peak shape was fitted with a pseudo-Voigt (pV) function in the
refinement procedure [30]. The cation distribution was evaluated by refining the changes in the
diffraction intensities, while locating the cation in an appropriate position. The background of each
pattern was fitted by a polynomial function of order 5 [30]. The densities of the samples were
estimated by XRD, using the relation
, where 8 refers to the number of atoms per
unit cell of the spinel structure, MW is the molecular weight of the sample. The quantities Na and
"a" are the Avogadro's number and the lattice parameter of the sample, respectively. The fitting
parameters, as well as the lattice parameter (aexp), unit cell volume V (Å3), crystallite size (CS) and
evaluated density ρXRD (g/cm3) are reported in Table I.
(a)
(b)
(c)
(d)
6
(e)
FIG. 1. XRD patterns of the magnetic nanoparticles investigated. (a) x=0.00, (b) x=0.02, (c) x=0.04, (d)
x=0.06 and (e) x=0.10. Solid lines are best fits of a pseudo-Voigt (pV) function [33].
In the XRD pattern of doped samples (FF-ND1, FF-ND2, FF-ND3 and FF-ND5) we did not
observe any peaks corresponding to NdFeO3. The ReFeO3 (RE3+=Rare-Earth) peak was observed
elsewhere due to rare earth doped ferrite [29, 30].
Table I
Average lattice parameter (aexp), unit cell volume V (Å3), crystallite size (CS), evaluated density ρXRD
(g/cm3) and results of Rietveld analysis for FeO·Fe(2-x)NdxO3 nanoparticles.
Sample
aexp (Å)
V (Å3)
CS (nm)
Comp.
(x)
0.00
FF-REF
0.02
FF-ND1
0.04
FF-ND2
0.06
FF-ND3
0.10
FF-ND5
8.39
(0.09)
8.41
(0.02)
8.42
(0.03)
8.44
(0.03)
8.46
(0.03)
590.59
(1.31)
594.82
(0.29)
596.94
(0.44)
601.21
(0.44)
605.49
(0.44)
23.9
(8.7)
18.3
(2.7)
15.6
(2.2)
16.2
(4.6)
14.9
(6.4)
ρXRD
(g/cm3)
5.18
(0.01)
5.24
(0.01)
5.27
(0.01)
5.39
(0.01)
5.56
(0.01)
Rwp(%) Rexp(%) GoF
2.23
2.26
2.60
1.58
1.83
1.29
1.73
1.24
1.81
1.26
2.06
1.32
1.19
1.23
1.48
The ionic radius of Nd3+ is 0.098 nm while the ionic radius of Fe3+ is 0.067 nm, which is
about 1.46 times smaller than the dopant radius. Therefore, replacement of Fe3+ by larger Nd3+ ions
causes an expansion of unit cell [30]. This causes an increase of the lattice parameter (aexp), as
shown in Table I, for FF-ND samples as compared with FF-REF sample. The robustness of fit
(GoF), the weighted profile factor (Rwp) and expected weighted profile factor (Rexp), assure the
reliability of the fits, since, low values of GoF were obtained [34,35].
B. XRD Cation Distribution
Several physical properties of a crystal can be accessed through the knowledge of the cation
distribution. Experimental Techniques such as: X-ray diffraction (XRD) pattern [30, 36-38], X-ray
magnetic circular dichroism (XMCD) [39] and X-ray absorption spectroscopy [39] can be used to
7
estimate the cation distribution for spinel ferrite materials. The cation distribution in the present work,
was obtained from X-ray diffraction pattern analysis. Experimental intensity ratios were compared
with the calculated intensity ratios, according to Bertaut et al.[40] method. In this method, pairs of
reflections are selected according to the expression [40].
(4)
and
where
are the experimental and calculated intensities for reflections (hkl),
respectively. We used the intensity ratios corresponding to the planes (220), (400), (440), which are
known to be sensitive to the cation distribution [30, 36-38]. In order to obtain the best-
simulated/evaluated structure, the R factor was defined, according to Eq. (5).
(5)
The determination of the structure is attained by varying the cation distribution in the
calculated intensity in such a way that the R factor will be minimized [30, 36-38].
The relative integrated intensity of the XRD lines can be calculated using Eq. (6):
where Ihkl corresponds to the relative integrated intensity. The quantity Fhkl is the structure factor,
while P is the multiplicity factor for the plane (hkl), and Lp is a Lorentz polarization factor (Eq. (7)),
and it will be a function of the Bragg diffraction angle. The multiplicity factor was obtained from
the literature [41]:
(6)
The structure factor of the spinel ferrite has 24 divalent and trivalent cations and 32 oxygen
anions [45]. The structural factors were calculated by using the equation proposed by Furuhashi et
al.[46], Eq. (8):
(7)
where
and
are related to crystal planes
and can be determined with Eq. (9) and (10):
8
(8)
and
(9)
(10)
For evaluation of the atomic scattering factor we used values reported in the International
Tables for X-ray Crystallography [42]. The temperature and absorption factors were neglected in
our evaluation because at room temperature these factors do not affect the relative XRD intensity
calculations [36]. In general, spinel structures have a high melting temperature. So, small thermo-
vibrational effect of spinel on XRD patterns is expected [43].
In the present evaluation all possible cation configurations were considered with 0.01
stoichiometric sensitivities that Nd3+ and Fe3+ ions can site in both tetrahedral and octahedral sites,
according with Eq. (11) [30]:
(Fe3+
(1-γ)Nd3+
γ]BO4
(1-δ)Nd3+
δ)A[ Fe2+Fe3+
(11)
and x are the molar stoichiometric amount of replacement of Fe3+ ions by Nd3+. The
where
closest correspondence with the actual sample structure was achieved by varying the cation
distribution of the calculated intensity, which will provide a minimum R factor. (Eq. (5)) [30]. The
cation distribution, the corresponding relative intensities of experimental and calculated XRD lines
are given in Table II.
Table II
Cation distribution of FeO·Fe(2-x)NdxO3 nanoparticles, experimental and calculated ratios between
peak intensities.
Comp. (x)
I220/I400
I220/I440
A-site
B-site
0.00
0.02
0.04
0.06
0.10
(Fe3+
(Fe3+
(Fe3+
(Fe3+
(Fe3+
1.00)
1.00)
1.00)
0.95Nd3+
0.94Nd3+
0.05)
0.06)
[Fe2+
[Fe2+
[Fe2+
[Fe2+
[Fe2+
1.00Fe3+
1.00Fe3+
1.00Fe3+
1.00Fe3+
1.00Fe3+
1.00]
0.98Nd3+
0.96Nd3+
0.99Nd3+
0.915Nd3+
0.02]
0.04]
0.01]
0.05]
Exp.
0.54
0.70
0.73
0.65
0.89
Calc
0.52
0.66
0.71
0.61
0.83
Exp.
1.15
1.50
1.38
1.62
2.04
Calc
1.03
1.33
1.29
1.57
2.03
The mean ionic radii in the tetrahedral (rA) and octahedral (rB) sites were calculated by
using Eqs. (S1) and (S2) from Electronic Supplementary Information (ESI) file. The values of the
ionic radius for rNd3+,rFe3+ and rFe3+were taken from the literature [44], 0.98 Å, 0.67 Å, and 0.49 Å,
9
respectively. The value of the oxygen positional parameter u can be determined with Eq. (S3)(ESI),
where RO is the radius of oxygen ion (1.32 Å) [25,31,39]. The theoretical lattice constant (ath) is
calculated by using Eq. (S4)(ESI) [30,36,45].
The structural parameters: tetrahedral bond length (dAL); octahedral bond length (dBL);
tetrahedral edge length (dAE); shared (dBE ) and unshared (dBEU) octahedral edge length; the
tetrahedral and octahedral jump length (LA and LB ) were calculated using Eqs. (S5)-(S11)(ESI).
The results are given in Table III.
Table III
Theoretical parameters based on the proposed cation distribution. See text for the symbols.
Comp
(x)
0.00
0.02
0.04
0.06
0.10
dBEU
(Å)
2.969
2.976
2.980
2.987
2.997
u (Å) dAL (Å) dBL (Å) dAE (Å) dBE (Å)
0.384
0.384
0.384
0.385
0.384
0.560
0.564
0.567
0.562
0.573
1.950
1.950
1.950
1.968
1.971
2.022
2.030
2.034
2.031
2.039
3.184
3.184
3.184
3.213
3.219
2.745
2.761
2.769
2.752
2.767
LA (Å) LB (Å)
3.631
3.641
3.646
3.653
3.665
2.965
2.973
2.977
2.983
2.993
rA (Å)
rB (Å) ath (Å)
0.630
0.630
0.630
0.648
0.651
8.400
8.409
8.418
8.432
8.468
Table III shows that the theoretical lattice parameter (ath), tetrahedral bond length (dAL),
octahedral bond length (dBL), tetrahedral edge (dAE), octahedral edge (dBE), unshared octahedral edge
(dBEU) increase with an increase in Nd-content. Table III also shows that tetrahedral radius (rA) and
octahedral radius (rB) increase as the Nd-content increases, while the oxygen parameter u remains
unchanged. The anion in the spinel structure, O2- ions, in this case, are not in general located at a
fixed position of the FCC sublattice. The anion is allowed to translate and this translation is
measured by a quantity named oxygen positional parameter or anion parameter. If we assume the
center of symmetry at (3/8, 3/8, 3/8) position that correspond to origin at A-site, the value of u ideal
is expected to be 0.375. Therefore, changes in the value of u can be interpreted as a relaxation of the
structure to accommodate the cations of different radius in the A and B sites [30, 47]. The jump
(hopping) lengths, LA and LB between the magnetic ions at A-site and B-site respectively, were
calculated (Table III). Since the Fe3+ radius (0.65Å) is smaller than that of the Nd3+ (0.98 Å), the
replacement of Fe3+ leads to an increase in rA and rB. Moreover, LA and LB increases with an
increase in Nd-content. The results showed that Nd3+ ions are present in both sites at different
concentrations, with the displacement of Fe3+ ions. Therefore, it was possible to observe changes in
structural parameters like bond lengths; shared and unshared edges, among others.
The magnetic properties of the particles depend on the exchange interactions between metal
ions. The bond angle and inter-ionic bond length between metal ions are the more important in the
overall magnitude of the magnetic exchange interaction. The magnitude of the magnetic exchange
10
interactions is proportional to the bond angles and inversely proportional to the inter-ionic bond
lengths. The inter-ionic bond lengths i.e., cation-cation distances and cation-anion distances (FIG.
S1(ESI)) were calculated using equations Eqs. (S12)-(S25)(ESI). The values obtained for inter-ionic
bond lengths were used for the evaluation of bond angles between the metal ions using equation
Eqs. (S21)-(S25). The values for bond angles are given in Table IV.
Table IV
Theoretical bond angles between metal ions based on the cation distribution. All angles are given in
degrees (º).
Comp (x)
0.00
0.02
0.04
0.06
0.10
θ1
122.14
122.28
122.35
121.99
122.10
θ2
138.45
138.63
138.72
138.28
138.41
θ3
94.33
94.14
94.05
94.52
94.38
θ4
126.06
126.03
126.02
126.08
126.06
θ5
70.91
71.10
71.19
70.74
70.87
The bond angles θ1, θ2, and θ5 are associated with the A-B and A-A exchange interactions,
while θ3, and θ4 are associated with the B-B exchange interactions (FIG. (S1)(ESI)). The observed
increase in θ1, θ2, and θ5, corresponding to x=0.00 to x=0.04 (Table IV) suggests the strengthening
of the A-B and A-A interactions, while θ3, and θ4 decrease indicates a weakening of the B-B
interaction Comparing the systems with x=0.06 and x=0.00, the B-B exchange interactions are
enhanced (θ3 and θ4), while A-B (θ1, θ2) and A-A (θ5) exchange interactions presents the lowest
values, suggesting weakening of A-B and A-A exchange interactions. For x=0.10 all bond angles
have values similar to x=0.00. This result suggests that, beyond x=0.06, we have a decreasing on
the B-B interaction and strengthening of the A-B and A-A interactions.
C. Optical Band Gap
The optical band-gaps of the FeO·Fe(2-x)NdxO3 nanoparticles were obtained by UV-Vis
spectrophotometry at the temperature of 25oC. The wavelength range of 350 nm to 850 nm using
DH-2000-BAL (Mikropack) equipped with deuterium and tungsten halogen lamps. Attached to a
DH-2000-BAL, an Ocean Optics® spectrometer USB4000 was used to measure the UV-Vis spectra
[42-44]. The band gap of the nanoparticles is related to the optical gap (Eg) and photon energy (hυ)
according to Eq. (12) [18, 48-50]:
(12)
11
where C is a constant, and α is the linear absorption coefficient. The linear absorption coefficient α
was calculated from the absorbance measurement A(hυ) as a function of the photon energy using
Eq. (13) [48-50]:
(13)
where Ascatt(hυ) is related to the Rayleigh-scattering contribution to the extinction measured data.
Ascatt(hυ) was estimated as having a λ-4 dependence, more details can be found elsewhere [42,44].
The value of n will be given according to the type of the electronic transition responsible for
the absorption: for allowed indirect transition is n = 1/2; n = 3/2 for forbidden indirect transition; for
allowed direct transition n = 2; n = 3 for forbidden direct transition [18,48-50]. The optical gap for
both direct and indirect allowed transitions were studied in this work.
Here, A(hυ) is the experimental absorbance measured for each sample. The absorbance was
measured using the UV-Vis spectrophotometry, and L = 1cm is the width of the cuvette cell. The
optical band-gap was obtained by extrapolating the linear region of the Tauc plot (plot of [αhυ] n
versus hυ) to a value of hυ= 0 [18, 48-50]. Tauc plot obtained for FF-REF and Nd +3 substituted
samples FF-ND1, FF-ND2, FF-ND3, FF-ND5 are shown in FIG. 2.
a)
b)
FIG. 2. (a) Tauc plots for [αhυ]2 and (b) [αhυ]1/2 versus the photon energy for determination of direct and
indirect optical gaps, respectively.
12
Table V
Bandgaps of magnetic nanoparticles studied.
Comp. (x)Nd
Sample
FF-REF
FF-ND1
FF-ND2
FF-ND3
FF-ND5
0.00
0.02
0.04
0.06
0.10
Ed (eV)
3.00 (0.01)
3.12 (0.01)
3.05 (0.01)
3.16 (0.01)
3.13 (0.01)
Ei (eV)
2.03 (0.01)
2.09 (0.01)
1.98 (0.01)
2.10 (0.01)
2.20 (0.01)
Fontijn et al reported [51-53] that, for magnetite electronic transitions, Fe2+ [t2g] → Fe3+ [e]
occurring in the tetrahedral sites, the bandgap is 3.11 eV, while Fe2+ [t2g] → Fe3+ [eg] transitions
occurring in the octahedral sites the bandgap is 1.94 eV. Their results were obtained through
magneto-optical polar Kerr measurements for bulk single-crystalline Fe3O4 and Mg2+ or Al2+
substituted Fe3O4. These values are consistent with those obtained here. One can see in Table V, that
for all samples with x ≠ 0.00, both Ed and Ei increase as x increases. We also point out that Ed
reaches the maximum value for x=0.06 (FF-ND3) rather than Ei that reaches the maximum value
for x=0.10.
J. Anghel and co-workers showed that exists a strong correlation between modifications in
the lattice parameters and the bandgap energy for Zn(1−x)MxO (M = Cr, Mn, Fe, Co, or Ni) samples
[54]. In their studies they observed that the unit cell volume (obtained by XRD) and bandgap
(obtained by spectrophotometry) reached the highest values of Fe3+ substitution [54]. For the series
of metals M that they studied, iron was the one with larger ionic radius, that leads to a higher lattice
parameter and hence, unit cell volume. A similar trend was also observed elsewhere [55,56]. From
XRD we found that the lattice parameter increases with increasing Nd-content and hence, unit cell
volume, bond lengths (dAL, dBL) and hopping lengths, LA and LB between the magnetic ions at A-site
and B-site. In this way, the increasing in the bandgap observed here, may be understood since the
bandgap is directly proportional to the interatomic separation, although it is also possible that new
electronic states may exist due to the presence of the dopant [55].
E. Transmission electronic microscopy (TEM)
Typical TEM micrographs of the samples investigated are shown in FIG. 3 and 4 for FF-
REF and FF-ND3 samples, respectively. Typical TEM micrographs for FF-ND1, FF-ND2 and FF-
ND5 samples is given in FIG. (S2), (S3) and (S4) (ESI), respectively. These results reveal a
polydisperse size and shape distribution, as expected for co-precipitation synthesis method, with a
broad number weighted-size distribution. The size distribution ranges from 5 to 30 nm and follows
the log-normal distribution. The mean-number weighted diameters are given in Table VI. The
13
polydispersity index (PDI = σ(DTEM )/
VI.
) also were evaluated and they are given in the Table
(a)
(b)
FIG. 3. (a) Typical TEM micrography and (b) Number-weighted diameter distribution of the FF-REF
sample. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(a)
(b)
FIG. 4 (a) Typical TEM micrography and (b) Number-weighted diameter distribution of FF-ND3 sample.
The solid line is the log-normal fitting of size distribution given by Eq. (1).
Table VI
Particles' mean diameter (D), distribution width (σ) and polydispersity and index (PDI) of the magnetic
fluids investigated.
Sample
FF-REF
FF-ND1
FF-ND2
FF-ND3
diameter D (σ) [nm]
12.81 (9.15)
12.50 (3.55)
11.67 (3.53)
7.99 (3.67)
14
PDI
0.71
0.28
0.30
0.46
FF-ND5
9.06 (2.83)
0.31
F. Small Angle X-ray Scattering (SAXS)
The direct obtained experimental scattering pattern by Small Angle X-ray Scaterring
(SAXS) can provide several characteristic properties of the sample irradiated by the X-ray beam [2 ,
9]. For polydisperse systems, the experimental scattering pattern corresponds to an average that is
performed over the particles present in solution [2, 9].
In the analysis of the scattered intensity data, it was assumed that I(q) was produced by a
system composed of globular nanoparticles. Therefore, I(q) has a contribution from the particles'
form-factor and the interaction between them (structure-factor). Assuming a model of polydisperse
hard spheres with radius r, the scattered intensity is given by Eq. (14) [12]:
(14)
In this model, the scattering elements are considered an ensemble of polydisperse non-
interacting spheres. The quantities r and V are the radius and volume for each sphere from the
ensemble. The fv(r) is the normalized-volume weighted-radius distribution function. Isph is the
normalized scattering intensity owing to a sphere of radius r and, Sc stands for a scaling factor [12].
The Gnom software was used to analyze the scattering intensity Is versus the scattering vector q
(modulus of q) [57,12]. This approach was used to determine the volume-weighted size-distribution
function fv(r) from the adjustment of the experimental scattering intensity data to Is(q), given by Eq.
(14). The solid lines in FIG. 5 and 6 for FF-REF and FF-ND3, represent the best fit of Eq. (14) to
the experimental data. For the remaining samples, namely FF-ND1, FF-ND2 and FF-ND5 the
experimental data and the best fit are shown in FIG. (S5),(S6) and (S7)(ESI), respectively. The
volume-weighted mean particle's diameter
was calculated according to Eq. (15) [12]:
(15)
15
(a)
FIG. 5: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to best
fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-REF.
(b)
(a)
(b)
FIG. 6: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to best
fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND3.
Table VII
Nanoparticles' Mean diameter
distribution function (σV=3 σD).
determined by SAXS and width of the log-normal volume
Sample
FF-REF
FF-ND1
FF-ND2
FF-ND3
FF-ND5
Comp. (x)Nd
0.00
0.02
0.04
0.06
0.10
(nm)
18.48
28.74
29.15
20.32
28.84
σV
1.83
1.86
1.83
2.07
1.77
The results given in Table VII for the volume-weighted average diameter measured by
SAXS are consistent with those determined by XRD.
16
G. Magnetization as a function of the applied field.
FIG. 7 (a) and (b) show the measured magnetization (M) versus the applied field (H) for
FeO·Fe(2-x)NdxO3 samples, at 6 K and 300 K, respectively. From these curves, the saturation
magnetization (Ms), coercivity field (Hc) and remanence magnetization (Mr) are obtained and are
given in Table VIII. The saturation magnetization increases in a steady manner upon increasing the
Nd3+ ion molar ratio from 0.00 up to 0.06, reaching the maximum value of 105.8±0.4 Am2/Kg at x=
0.06. It's worth to mention that the result for saturation magnetization value are higher than that of
the bulk material. In this case, suggesting the existence of a highly ordered spin configuration
distributed across the volume of the magnetic nanoparticles.
The M(H) curves for magnetic nanoparticles in superparamagnetic (SPM) regime (FIG. 7-
(b)) are fitted with a Langevin function (Eq. (16)), weighted with the particle-size distribution
function (PDF) [58,59]:
(16)
where M(H,T) is the magnetization of the magnetic nanoparticles at the temperature T (300 K) , Nsc
is a scaling constant, MS represents the saturation magnetization of the magnetic nanoparticles and
[59]. The quantity VV, stands for the volume-weighted average volume of the
nanoparticles given by
average radius and hence diameter, measured by SAXS [12].
[52]. Here we used,
from the volume-weighted
In the model expressed by Eq. (16), the magnetization M as a function of field H has
contributions from superparamagnetic (SPM) and paramagnetic (PM) particles [31]. The quantity c
is the paramagnetic contribution (linear with the magnetic field, H [36]). We found that the SPM
regime contributes with 95-97% for all samples, while only 3-5% comes from the PM regime [36].
The MNPs are expected to have a nonmagnetic layer around the magnetic core (core-shell
model) [60, 61]. The thickness of this magnetically inert shell was evaluated from the volume-
weighted size distribution function, according to Eq. (17) [60, 61]:
(17)
where δ is the thickness of the shell, MS is saturation magnetization of the nanoparticles
17
MSbulk stands for the saturation magnetization of the bulk material [60-62] and the reciprocal of the
volume-weighted average diameter (
) of the particles [60,61]. From Eq. (16) we found that
the thickness of the magnetically inert layer of the materials studied here, ranges from 0.5 Å to 7.8
Å.
The Neel's theory for collinear ferro/ferrimagnetism of two sub-lattices model, predicts the
net magnetic moment per formula unit (f.u.) according to Eq. (17):
(18)
where Mocta and Mtetra are the magnetic moments of B (octahedral) and A (tetrahedral) sites in µB
units (Bohr magneton) [36, 31]. Based on the site occupancy obtained from the XRD cation
distribution and magnetic moment of 5, 4 and 3.2 µB [63] for Fe3+ , Fe2+ and Nd3+ ions, respectively,
the nNeel values were calculated using Eq. (18) and are given in Table VIII. On the other hand, the
magnetic moment per formula unit in Bohr magneton unit nexp (experimental) can be calculated
from the saturation magnetization Ms according to Eq. (18):
(19)
where MW is the molecular weight. The calculated nNeel magneton number using the XRD data
agrees with the experimentally obtained magneton number from the M-H loops. The results are
comparable to those determined by Eq. (19). The calculated values of magneton number nNeel and
nexp are comparable. Furthermore, the values of nNeel and nexp show same trend, that is, increases with
x=0.00 to x =0.04, reaches the maximum value at x = 0.06 and decreases for x =0.10.
The Yafet-Kittel angles (αY-K) described by equation Eq. (20) can give us insight about the
type of magnetic ordering for FeO·Fe(2-x)NdxO3 samples [30, 36]. The quantity nexp is given by Eq.
(19) while Mtetra and Mocta are given by Eq. (18). The Yafet-Kittel angles (αY-K) obtained were 21.6
º, 16.3 º, 14.1 º, 0.0 º and 0.0 º for x=0.00, 0.02, 0.04, 0.06 and 0.10, respectively. From x=0.00 to
x=0.04 αY-K decreases with increasing Nd3+ substitution. The decrease in αY-K for this samples is
due to a noncollinear type of magnetic ordering, since the Yafet-Kittel angles are different from
zero. The existence of non-zero αY-K suggest a model of canted-spin magnetization, which should
have a triangular spin arrangement and is suitable on the B-site that leads to a reduction in the A -- B
exchange interaction and enhancement in the B-B exchange interaction [30]. For x=0.06, there is a
transition to the Néel-type magnetic ordering. This result suggests a reinforcement of a dominant A-
18
B super-exchange interaction and these results are corroborated by those found in the inter-ionic
bond angles of Table IV.
(20)
(a)
(b)
FIG. 7. Magnetization curves of FF-REF, FF-ND1, FF-ND2, FF-ND3 and FF-ND5 nanoparticles a)
at 6 K and b) at 300 K, the dots are experimental results and solid lines are their size-weighted
Langevin fits.
Table VIII
Magnetic parameters extracted from M(H) curves.
Sample
Comp.
(x)
Saturation
magnetization
(Ms) Am2/Kg
Remnant
magnetization
(Mr) Am2/Kg
Coercivity
(Hc) mT
Squareness
ratio (S)
FF-REF
FF-ND1
FF-ND2
FF-ND3
FF-ND5
0.00
0.02
0.04
0.06
0.10
81.9 (0.8)
86.1 (0.3)
87.7 (0.3)
105.8 (0.4)
91.0 (0.3)
30.4 (0.2)
28.3 (0.1)
38.6 (0.2)
24.3 (0.1)
32.8 (0.2)
156
156
121
157
156
0.37
0.33
0.44
0.23
0.36
Magnetic
anisotropy
energy
density
(Keff) 103
J/m3
33.1
35.1
27.9
44.8
40.2
Bohr
magneton
number
(µB) exp.
Neel's
Bohr
magneton
number
(µB) XRD.
3.40 (0.01) 4.00 (0.44)
3.60 (0.01) 3.96 (0.44)
3.69 (0.01) 3.92 (0.43)
4.49 (0.02) 4.08 (0.45)
3.97 (0.02) 3.95 (0.42)
For magnetite (Fe3O4), below the Verwey Temperature (TV), the magnetocrystalline
anisotropy is expected to be uniaxial [62, 64]. Therefore, the experimental coercivity field and
saturation magnetization are related to the effective anisotropy constant Keff through Eq. (21) [62,
64]:
19
To determine the squareness ratio (S), Eq. (22) was employed [65, 14]:
(21)
(22)
The evaluated values of the magnetocrystalline anisotropy constant (Keff) and S are given in
Table VIII. For all samples S < 0.5, which indicates uniaxial anisotropy contribution in the prepared
FeO·Fe(2-x)NdxO3 nanoparticles [60, 62-65]. The overall values of Keff obtained are of the same order
of magnitude of that from the magnetite bulk (1.1 -- 1.3 104 J/m3) [62-65] however, we have
obtained values 3 to 4 times higher.
H. Zero-field cooled and field-cooled (ZFC-FC)
FIG. 8(a)-(e) show the temperature dependence of the magnetization in low external applied
field, the zero-field cooled curve (ZFC) and field-cooled curve (FC) for FF-REF, FF-ND1, FF-ND2,
FF-ND3 and FF-ND5, respectively. In the ZFC-FC protocol the MNPs were frozen in the absence
of the magnetic field, fast enough that the random orientation of their easy axis is preserved [12,
13]. The system was superparamagnetic (SPM) at room temperature, and the magnetization curves
M(T) were collected in ZFC and FC modes. The overall shape of the ZFC-FC curves indicate weak
interaction between particles. In the ambit of the non-interacting particles' model, the blocking-
temperature distribution function f(TB) is expected to be broad [12]. One can use the Stoner-
Wohlfarth model to describe uniaxial, single-domain [8] and non-interacting particles to obtain the
blocking-temperature distribution function given by Eq. (23) [12]. FIG. 8(f)-(i) show f(TB) for FF-
REF, FF-ND1, FF-ND2, FF-ND3 and FF-ND5, respectively. A log-normal distribution function was
and standard deviation σT
used to fit these results and obtain the mean blocking temperature
[12]. For FF-REF, FF-ND1, FF-ND2, FF-ND3 and FF-ND5 samples, the average blocking
temperatures were 40.2 K, 28.8 K, 23.6, 25.1 and 41.7 K, respectively, at an applied field of 5 mT.
Note that FIG. 8(a)-(e) also show that the maximum of ZFC curve for all cases is close to the
temperature in which the ZFC-FC curves split. This effect is due to particle dipole-dipole
interaction.
(23)
20
(f)
(g)
(h)
(i)
(a)
(b)
(c)
(d)
21
(e)
(j)
FIG. 8. (a, b, c, d and e) ZFC-FC magnetization curves, where () blue circles represent the FC
susceptibility, whereas red circles () symbols represent the ZFC susceptibility for FF-REF, FF-ND1, FF-
ND2, FF-ND3, and FF-ND5, respectively. 8 (f, g, h, i and j) Temperature derivative [-d(MFC-MZFC)/dT] of
the difference between FC and ZFC magnetization curves for samples FF-ND, FF-RC and FF-REF,
respectively.
The log-normal width σT, obtained from the fitting to the experimental data -- FIG. 8(f)-(i) --
were used to evaluate the number of interacting particles (Ni) and the correlation volume (Λ3),
defined according to Eqs. (24) and (25) [12, 13]:
(24)
(25)
where σV , σT are the width of the volume and blocking temperature distribution functions. The
quantity
is the average volume defined as:
(26)
is the volume-weighted average diameter measured by SAXS. It's worth to mention,
where
that according with M. El-Hilo [58], the volume-weighted diameter distribution fv(D) should
converts to volume-weighted distribution fv(V) with σV= 3σD and average volume-weighted by
volume given by Eq.(26). The quantity ϕ is the volumetric fraction of particles in the solution [8,
12, 13], ~ 0.3% for all the samples investigated.
Table IX
Width of volume and blocking temperature distribution σV , σT. Number of interacting particles Ni.
The correlation volume (Λ3) and correlation length (Λ).
Sample
FF-REF
FF-ND1
σV
1.83
1.86
σT
0.37
0.76
Ni
5
3
22
Λ3 (10-23 m3)
5.4
10.1
Λ (nm)
176
216
FF-ND2
FF-ND3
FF-ND5
1.83
2.07
1.77
0.41
0.46
0.60
5
5
3
19.3
6.6
12.4
226
187
231
Table IX, shows estimation of the number of interacting particles (Ni) inside the correlation
volume (Λ3) and correlation length (Λ) [12]. The results did not show correlation between Ni and
the size of the particles. However, the correlation volume (Λ3) and correlation length (Λ) are
smaller for FF-REF and FF-ND3, which are those with smaller volume-weighted average diameter
measured by SAXS. The correlation volume (Λ3) and correlation length (Λ) also are larger for those
particles with larger volume-weighted average diameter (FF-ND1, FF-ND2 and FF-ND5). This is
due to the fact that for larger particles, the interparticle distance increases, as the mean-particle
diameter increases [12]. Note that, the correlation length (Λ) is out of the range accessible in our
SAXS experiments, that is, about ~60 nm.
V. Summary and Conclusions
Magnetic fluid based on magnetite were synthesized and XRD patterns of the samples
investigated revealed a single-phase cubic spinel structure of Fd3m space group. From the cation
distribution, the theoretical lattice parameter (ath), tetrahedral bond length (dAL), octahedral bond
length (dBL), tetrahedral edge (dAE), octahedral edge (dBE), unshared octahedral edge (dBEU) increase
with the increase in Nd-content. Since the Fe3+ (0.65Å) is smaller than Nd3+ (0.98 Å), the
replacement of Fe3+ leads to an increase in rA and rB. Moreover, LA and LB increase with higher Nd-
content. The results showed that Nd3+ ions are placed in both sites at different concentrations.
Therefore, it was possible to observe modifications in structural parameters like bond lengths;
shared and unshared edges. The increasing in the bandgap may be interpreted as a result of the
higher interatomic separation with the doping. TEM micrographs reveal a polydisperse size and
shape distribution of particles, as expected for co-precipitation method, with a broad number-
weighted size distribution. The results for volume-weighted average diameter measured by SAXS
are consistent with those determined by XRD. From the M-H Loops we found that the SPM regime
contributes with 95-97% for all samples, while only 3-5% contribution comes from the PM regime.
The saturation magnetization increases in a steady manner upon increasing the Nd3+ ion molar ratio
from 0.00 up to 0.06, reaching the maximum value of 105.8±0.4 Am2/Kg at x= 0.06. It's worth to
mention that the result for saturation magnetization value are higher than that of the bulk material.
In this case, suggesting the existence of a highly ordered spin configuration distributed across the
volume of the magnetic nanoparticles. The squareness values for all samples are less than 0.5,
which indicates uniaxial-anisotropy contribution in the FeO·Fe(2-x)NdxO3 nanoparticles. The overall
23
values of Keff obtained for all samples studied, have the same order of magnitude of that from the
bulk magnetite (1.1 -- 1.3 104 J/m3) however, we have obtained values 3 to 4 times higher. The
magnetic measurements indicate the existence of low interaction between the MNPs at the
concentrations investigated.
ACKNOWLEGMENTS
This research was financial supported by Brazil's agencies INCT/CNPq (Conselho Nacional
de Desenvolvimento Científico e Tecnológico; Grant Number: 465259/2014-6), INCT/FAPESP
(Fundação de Amparo à Pesquisa do Estado de São Paulo; Grant Number: 14/50983-3),
INCT/CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior; Grant Number:
88887.136373/2017-00), and INCT-Fcx (Instituto Nacional de Ciência e Tecnologia de Fluidos
Complexos).
References
[1] D. H. K. Reddy and Y. S. Yun,
Coord. Chem. Rev. 315, 90-111 (2016).
[2] K. Parekh et al,
Mater. Chem. Phys. 222, 217-226 (2019).
[3] L. Shen, et al,
Ceram. Int. 40, 1519-1524 (2014).
[4] I. Torres-Díaz and C. Rinaldi,
Soft Matter 10, 8584-8602 (2014).
[5] C. Scherer and A.M. Figueiredo Neto,
Braz. J. Phys. 35(3A), 718-727 (2005).
[6] A. G. Kolhatkar, A. C. Jamison, D. Litvinov, R. C. Willson, and T. R. Lee,
Int. J. Mol. Sci. 14, 15977-16009 (2013).
[7] K. Parekh, L.Almásy, H. S. Lee and R V Upadhyay,
Appl. Phys. A, 106, 223 (2012).
[8] G. F. Goya et al.,
J. Appl. Phys. 94 (5), 3520-3528 (2003).
[9] D. H. G. Espinosa, C. L. P. O Liveira And A. M. Figueiredo Neto,
J. Opt. Soc. Am. B 35(2), 346-356 (2018).
[10] E. A. Elfimova, A. O. Ivanov and P. J. Camp
J. Chem. Phys. 136, 194502 (2012).
24
[11] N. Daffé et al,
J. Magn. Magn. Mater. 460, 243-252 (2018).
[12] E. S. Gonçalves et al.,
Phys. Rev. E 91, 042317-1 -- 042317-7 (2015).
[13] J. C. Denardin et al.,
Phys. Rev. B 65, 064422 (2002).
[14] R.R. Kanna, et al.,
J Mater Sci: Mater Electron 30, 4473 (2019).
[15] S. Aslam et al, Results in Physics 12 , 1334-1339 (2019).
[16] P. P. Naik, R. B. Tangsali, S.S. Meena and S. M. Yusuf,
Mater. Chem. Phys. 191, 215-224 (2017).
[17] S. Amiri and H. Shokrollani,
J. Magn. Magn. Mater. 345, 18-23 (2013).
[18] R. Jain et al.
J. Magn. Magn. Mater. 456, 179 -- 185 (2018).
[19] W. Huan et al,
J. Nanosci. Nanotechno. 14, 1 -- 9 (2014).
[20] A. A. Veligzhanin, D. I. Frey, A. V. Shulenina, A. Yu. Gruzinov, Ya. V. Zubavichus and M. V.
Avdeev, J. Mag. Mag. Mat. 459, 285-289 (2018).
[21] F. Döbrich, A. Michels and R. Birringer,
J. Mag. Mag. Mat. 316, e779 -- e782 (2007).
[22] Alvaro V. N. C. Teixeira et al.,
Phys. Rev. E 67, 021504-1 -- 021504-7 (2003).
[23] H. Harzali, A. Marzouki, F. Saida, A. Megriche and A. Mgaidi,
J. Magn. Magn. Mater. 460, 89-94 (2018).
[24] X. Huang et al,
25
J. Magn. Magn. Mater. 405, 36-41 (2016).
[25] Rueden, C. T.; Schindelin, J. & Hiner, M. C. et al,
BMC Bioinformatics 18:529, 1-26 (2017).
[26] J. F. D. F. Araujo, A. C. Bruno, and S. R. W. Louro,
Rev. Sci. Instrum 86; 105103 (2015).
[27] S. Arsalani et al,
J. Magn. Magn. Mater. 475, 458 -- 464 (2019).
[28] G. D. Gatta, I. Kantor, T. B. Ballaran, L. Dubrovinsky, and C. McCammon,
Phys. Chem. Min. 34, 627- 635 (2007).
[29] H. Harzali et al.,
J. Magn. Magn. Mater. 460, 89 -- 94 (2018).
[30] K.V. Zipare et al.,
Journal of Rare Earths 36, 86-94 (2018).
[31] A.S. Elkady et al.,
J. Magn. Magn. Mater. 385, 70 -- 76 (2015).
[32] Georg Will, Powder Diffraction, The Rietveld Method and the Two Stage Method to Determine
and Refine Crystal Structures from Powder Diffraction Data, Springer-Verlag Berlin Heidelberg,
(2006)
[33] L. Lutterotti,
Nucl. Instrum. Meth. B 268, 334-340 (2010).
[34] M. Dalal et al.,
Mater. Res. Bull. 76, 389 -- 401 (2016).
[35] Kumar et al.
International Nano Letters 3:8 (2013).
[36] J. Kurian, M. Jacob Mathew,
J. Magn. Magn. Mater. 428, 204 -- 212 (2017).
[37] D.V.Kurmude et al.,
J. Supercond. Nov. Magn. 27(2), 547-553 (2014).
[38] Patange et al.,
26
J. Appl. Phys. 109, 053909 (2011).
[39] C. E. Rodríguez Torres et al.,
Phys. Rev. B 89, 104411 (2014).
[40] E.F. Bertaut, Study of the nature of spinel ferrite, Weekly the Proceedings of
Sessions of the Academy of Sciences, C.R. Acad. Sci. 230, 213 -- 215 (1950).
[41] B.D. Cullity,
Elements of X-ray Diffraction, Addison-Wesley, Reading,
Massachusetts, USA, (1978).
[42] P. J. Brown, A. G. Fox, E. N. Maslen, M. A. O'Keefe and B. T. M. Willis. International Tables
for Crystallography Vol. C, ch. 6.1, 554-595 (2006).
[43] H. Kavas et al.,
J. Alloy and Compd. 479, 49 -- 55 (2009).
[44] R. D. Shannon,
Acta Cryst. A32, 751-767 (1976).
[45] A. Najafi Birgani et al,
J. Magn. Magn. Mater. 374, 179 -- 181 (2015).
[46] H. Furuhashi, M. Inagaki, S. Naka,
J. Inorg. Nucl. Chem. 35, 3009 -- 3014 (1973).
[47] B. B. V. S. Varaprasad, K. V. Ramesh and A. Srinivas,
J. Supercond. Nov. Magn. 30, 3523 -- 3535 (2017).
[48] Espinosa, Gonçalves, and Figueiredo Neto,
J. Appl. Phys. 121, 043103 (2017).
[49] M. Dongol et al.,
Optik 126, 1352 -- 1357 (2015).
[50] D. Espinosa, L. B. Carlsson, A. M. Figueiredo Neto, and S. Alves,
Phys. Rev. E 88, 032302 (2013).
[51] W. F. J. Fontijn, P. J. van der Zaag, L. F. Feiner, R. Metselaar, and M. A. C. Devillers,
J. Appl. Phys. 85, 5100 (1999).
[52] W. F. J. Fontijn, P. J. Van der Zaag, M. A. C. Devillers, V. A. M. Brabers, and R. Metselaar,
Phys. Rev. B 56, 5432 (1997).
27
[53] P. J. van der Zaag, W. F. J. Fontijn, P. Gaspard, R. M. Wolf, V. A. M. Brabers, R. J. M. van de
Veerdonk, and P. A. A. van der Heijden,
J. Appl. Phys. 79, 5936 (1996).
[54] J. Anghel, A. Thurber, D. A. Tenne, C. B. Hanna, and A. Punnoose,
J. Appl. Phys. 107, 09E314 (2010).
[55] Thurber et al.,
Phys. Rev. B 76, 165206 (2007).
[56] W. Luo, S. R. Nagel, T. F. Rosenbaum, and R. E. Rosensweig,
Phys. Rev. Lett. 67, 2721 (1991).
[57] Feigin L.A. & Svergun D.I., Structure Analysis by Small-Angle X-Ray and Neutron Scattering.
NY: Plenum Press (1987).
[58] M. El-Hilo,
J. Appl. Phys. 112, 103915 (2012).
[59] Tamion et al,
Appl. Phys. Lett. 95, 062503 (2009).
[60] D. Caruntu, G. Caruntu and C. J O'Connor,
J. Phys. D: Appl. Phys. 40, 5801 -- 5809 (2007).
[61] Chen J P, Sorensen C M, Klabunde K J, Hadjipanayis G C, Devlin E and Kostikas A,
Phys. Rev. B 54, 9288 (1996).
[62] Goya, G., Berquó, T. & Fonseca, F.,
J. Appl. Phys. 94, 3520 (2003).
[63] R. Przeniosto, I. Sosnowska and P. Fischer,
J. Mag. Mag. Mat., 140-144, 2153-2154 (1995).
[64] L. Tauxe, H. Neal Bertram and C. Seberino,
Geochem. Geophy. Geosy., 3(10) (2002), 1055
[65] M Graeser, K Bente and T M Buzug,
J. Phys. D: Appl. Phys. 48, 275001 (11pp) (2015)
28
Electronic supplementary information.
Magnetic, Structural and cation distribution studies on FeO·Fe(2-x)NdxO3
(x=0.00, 0.02, 0.04, 0.06 and 0.1) nanoparticles.
W. W. R. Araujo*, C. L. P. Oliveira, G. E. S. Brito, A. M. Figueiredo Neto
Institute of Physics, University of São Paulo, Rua do Matão 1371, 05508-090, São Paulo, SP,
Brazil.
Department of Physics, Pontifical Catholic University of Rio de Janeiro, RJ, Brazil.
J. F. D. F. Araujo,
The mean ionic radii in the tetrahedral (rA) and octahedral (rB) sites were calculated by using
Eqs. (S1) and (S2) [1-4]:
(S1)
(S2)
where, CAi ( i=Fe3+,Nd3+ ) are the concentration of ions in the tetrahedral site and CBi ( i=
Fe2+,Fe3+,Nd3+ ) the concentration of ions in the octahedral site. The values of the ionic radius for
rNd3+,rFe3+ and rFe3+were taken from the literature, 0.98 Å, 0.67 Å, and 0.49 Å, respectively.
The value of the oxygen positional parameter u can be determined with Eq. (S3) [1-4]:
where RO is the radius of oxygen ion (1.32 Å).
The theoretical lattice constant (ath) is calculated by using Eq. (S4):
(S3)
(S4)
The values of tetrahedral bond length (dAL), octahedral bond length (dBL), tetrahedral edge length
(dAE), shared (dBE ) and unshared (dBEU) octahedral edge length, the tetrahedral and octahedral jump
length (LA and LB ) are determined using Eqs. (S5)-(S11) [1-4]. The results are given in Table III.
(S5)
(S6)
(S7)
(S8)
(S9)
(S10)
(S11)
The configuration of ion pairs in the spinel structure, as well as their distances and angles has
been illustrated in FIG. (S1) [1-4].
FIG. S1. Configuration of the ion pairs in spinel ferrites with favorable distances and angles.
The bond lengths values for Me-Me (metal-metal) and Me-O (metal-oxygen) were calculed
theoreticaly using the following sets of equations, Eqs. (S12)-(S20) [3]:
Me-Me:
(S12)
(S13)
Me-O:
The values of bond angles were calculated using Eq. (S21)-(S25), as follows:
(S14)
(S15)
(S16)
(S17)
(S18)
(S19)
(S20)
(S21)
(S22)
(S23)
(S24)
(S25)
where b, c, d, e, f, p, q, r and s represent the parameters of ion pair distances associated with angles
θ1, θ2, θ3, θ4, and θ5 for spinel ferrites structures [1-4]. The parameter δ is the deviation of the
oxygen positional parameter u from ideal position uideal = (3/8, 3/8, 3/8) i.e. δ = u -- uideal [2].
Typical TEM micrographs for FF-ND1, FF-ND2 and FF-ND5 samples is given in FIG. (S2),
(S3) and (S4), respectively.
(a)
FIG. S2. (a) Typical TEM micrography and (b) Number-weighted diameter distribution for FF-ND1
samples. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(b)
(a)
FIG. S3 (a) Typical TEM micrography and (b) Number-weighted diameter distribution for FF-ND2
samples. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(b)
(a)
FIG. S4 (a) Typical TEM micrography and (b) Number-weighted diameter distribution for FF-ND5
samples. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(b)
FIG. S5: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to
best fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND1.
FIG. S6: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to
best fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND2.
(a)
(a)
(a)
(b)
(b)
(b)
FIG. S7: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to
best fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND5.
References
[1] K.V.Zipare, S.S.Bandgar and G.S.Shahane,Journal of Rare Earths, 36 (1), Pages 86-94 (2018)
[2] R. Sharma et al., Journal of Alloys and Compounds 704, 7-17 (2017)
[3] S. Chakrabarty et al., Journal of Alloys and Compounds 625, 216 -- 223 (2015)
[4] G. Kumar et al., Materials Research Bulletin 63, 216 -- 225 (2015)
Magnetic, Structural and cation distribution studies on FeO·Fe(2-x)NdxO3
(x=0.00, 0.02, 0.04, 0.06 and 0.1) nanoparticles.
W. W. R. Araujo*, C. L. P. Oliveira, G. E. S. Brito, A. M. Figueiredo Neto
Institute of Physics, University of São Paulo, Rua do Matão 1371, 05508-090, São Paulo, SP,
Brazil.
Department of Physics, Pontifical Catholic University of Rio de Janeiro, RJ, Brazil.
J. F. D. F. Araujo
Abstract
We synthesized and characterized the colloidal suspensions of FeO·Fe(2-x)NdxO3
nanoparticles with x=0.00, 0.02, 0.04, 0.06 and 0.1. The effect of the Fe3+ ion replacement by Nd3+
on the crystal structure is in-depth studied, through X-ray diffraction (XRD) and the obtained cation
distribution. The magnetic properties of the synthesized FeO·Fe(2-x)NdxO3 nanoparticles also were
investigated and corroborated by other physical methods. A remarkable saturation magnetization of
105 Am2/kg was achieved for x=0.06.
Keywords: Magnetic nanoparticles; Characterization; Synthesis; Magnetite; Iron oxide; ZFC-FC;
SAXS
PACS number(s): 82.70.Dd, 75.50.Mm, 82.20.Wt, 75.75.−c, 75.60.−d, 61.10.Eq
I. Introduction
There has been an enormous research effort recently in colloidal magnetic nanocrystals, so
that hundreds of papers are published per year in this field of scientific research [1]. These
nanocrystals can be obtained with controlled size, shape and composition [2, 3]. Colloidal magnetic
nanocrystals (CMNs) attract increasing interest both in fundamental sciences and in technological
applications [4]. Colloidal magnetic nanocrystals (CMNs) are suspensions of magnetic
nanoparticles (MNPs) dispersed in a liquid carrier [4, 5]. When the particle dimension is small
(typically ~10 nm), the particle presents a single magnetic domain, with a large magnetic moment
called superspin [6-8]. The surface of MNPs can be modified by several stabilizing agents or
functional groups. Once, the surface of the magnetic nanoparticles is modified, they become highly
functional materials. Some of the MNPs physical properties can be controlled by external magnetic
fields or magnetic-field gradients [9, 10], that is, they are stimuli-responsive systems. Some
applications of ferrofluids are separation media [1], heat-conduction media [4], gas fluidized beds
[1], sealants [2, 7, 11] and hydraulic car suspensions [7]. MNPs can also be used for drug delivery,
medical diagnosis, and cell destruction [4]. One of the peculiarities of a ferrofluid is how its
nanoscopic organization is affected by an applied external magnetic field [10]. In the field of
1
magneto-optical devices some applications can be found, for integrated optics [9, 11], optical fibers
[9, 11] and tunable beam splitter [11].
Magnetic nanoparticles are free to move when dispersed in a liquid carrier medium, and
different physical event may occur as a function of an applied magnetic field [12]. The blocking
temperature (TB) defines when the system of MNPs passes from blocked to superparamagnetic
(SPM) state [12, 13]. The Zero-field cooled (ZFC) and Field cooled (FC) technique is widely used
to study granular materials. ZFC-FC curves allows the determination of the blocking temperature of
the system TB [12, 13]. The Small Angle X-ray Scattering is used to study structural parameters of
fluid samples from the analysis of the experimental scattering pattern [2, 9]. For polydisperse
systems, the experimental scattering pattern corresponds to an average that is performed over the
particles in the solution. Thus, the experimentally reached values correspond to an average over the
entire ensemble of particles rather than a single particle [2, 9]. To tackle this problem, ZFC-FC
curves complemented by SAXS scattered intensities could supply several insights about the MNPs
investigated.
‒
‒
Rare-Earth doping significantly alters the nucleation and growth of nanoferrites, which
facilitate magnetic spin orientation [2, 14]. It is known that the magnetic behavior of spinel ferrite
compounds is mostly due to the interaction between the iron atoms [14]. Rare-Earth ions are more
favorable to enter in octahedral sites of the spinel structure; this causes 4f-3d interactions that
promotes structural distortion, lattice strain and changes in saturation magnetization [14]. The
literature reports the use of Neodymium (Nd) for doping copper nanoferrites [14]. Mixed
manganese neodymium copper (Mn -- Nd -- Cu) nanoferrites was produced by sonochemical method
[14]. Aslam and co-workers performed a co-doping of Nd3+ and Pr3+ on lithium nanoferrite and
reported the effects on the magnetic and structural properties of the system [15]. The effect of Nd+3
doping on Mn-Zn ferrite was reported in the literature [16, 17], and an enhancement of the
saturation magnetization, due to the new cation distribution imposed by Nd doping was reported.
Jain and co-workers studied the influence of rare earth ions on structural, magnetic and optical
properties of magnetite nanoparticles [18]. Jain also reported that there is a variation in the
saturation magnetization maximum value, directly proportional to the number of unpaired 4f
electrons in the dopant element [18]. Huan and co-workers found similar results for RE3+-doped
Fe3O4 samples (RE3+=Ln3+, Eu3+ and Dy3+) [19].
In the present study we report on the synthesis and characterization of colloidal suspensions
of FeO·Fe(2-x)NdxO3 nanoparticles with x=0.00, 0.02, 0.04, 0.06 and 0.1. The samples were
synthesized by co-precipitation method. The characterization was done by: X-ray diffraction
(XRD), Transmission Electron Microscopy (TEM), Small Angle X-ray Scattering (SAXS), Optical
bandgap and Zero-field and Field Cooling magnetization (ZFC-FC). Therefore, the effect of the Fe3+
2
ion replacement by Nd3+ on the crystal structure is in-depth studied, through X-ray diffraction
(XRD) and the obtained cation distribution. The magnetic properties of the synthesized FeO·Fe(2-
x)NdxO3 nanoparticles also were investigated and corroborated by other physical methods.
II. Material and methods
A. Materials reagents
The materials used to obtain the MNPs were: FeCl3·6H2O, ≥99%; FeCl2·4H2O, ≥99%;
NdCl3·6H2O, ≥99%; cis-9-Octadecenoic acid (Oleic acid), ≥99%; NaOH, ≥99%; and kerosene. All
primary materials were acquired from Sigma-Aldrich and used as received.
B. Synthesis procedure by co-precipitation method
The magnetic nanoparticles were synthesized by co-precipitation of an aqueous mixture of
FeCl3 (ferric chloride) and FeCl2 (ferrous chloride) salts and stabilized at pH~12 [11, 20-24].
Briefly, ferric chloride (4.00 mmol) and ferrous chloride (2.00 mmol) in a 2:1 molar ratio and 8 mL
of oleic acid are mixed in 25 mL of deionized water. The solution was heated up to 80 °C and
magnetically stirred for 30 min. Then, 30 mL of NaOH was added to the solution to precipitate the
particles at room conditions, under vigorously stirred for more 30 min. When the pH reaches ~ 11, a
severe reaction occurs and the solution becomes dark brownish. Thereafter, the resultant solution
was cooled to room temperature. At last, the MNPs were precipitated with a permanent magnet and
then washed ten times with deionized water to remove residual unreacted salts. The procedure
described above was used to produce magnetite nanoparticles with x=0.00 and that samples was
called as FF-REF.
The samples FF-ND1, FF-ND2, FF-ND3 and FF-ND5 are labeled according to molar
percentage amount of substitution of Fe3+ ions by Nd3+ ions. These percentages are 1 ,2, 3 and 5%
(x=0.02, 0.04, 0.06 and 0.10), respectively. FF-ND1, FF-ND2, FF-ND3 and FF-ND5 were prepared
with an aqueous solution composed of FeCl3·6H2O ((4 - i) mmol), NdCl3·6H2O (i=(0.04, 0.08, 0.12
and 0.20) mmol), FeCl2·4H2O (2.00 mmol) and 8 mL cis-9-octadecene. Then, 30 mL of NaOH was
added to the solution to precipitate the particles at room conditions, under vigorously stirring for 30
min. The remaining procedures were the same used to prepare the FF-REF sample.
Hereafter, the doped concentration (x=0.02, 0.04, 0.06 and 0.10) of Nd3+ in magnetite
(x=0.00, Fe3O4) will be denoted as FF-ND1, FF-ND2, FF-ND3 and FF-ND5, respectively. All
samples were sterically stabilized with a single oleic acid layer, chemisorbed on the particles'
surfaces. MNPs are dispersed in kerosene.
C. X-ray Diffraction
3
Powder X-ray Diffraction (XRD) patterns were obtained to investigate nanoparticles'
crystalline structure. XRD was done in a Brucker-AXS D8 series 2 diffractometer, set to a Bragg
Brentano Parafocussing Geometry. A Cu Ka source (λ = 1.5414 Å) generated X-rays at room
temperature. The difractometer was operated at 40 kV, 30 mA. The experimental pattern data were
registered in continuous scan mode, scattering angle 2θ from 15 to 80º, in steps of 0.02º.
D. Transmission Electron Microscopy
The overall form (morphology) and size distribution of the MNPs were examined by
transmission electron microscopy (TEM). TEM was performed on a JEOL 1010 (Japan) microscope
with an acceleration voltage of 80 kV. TEM micrographs were acquired by a Gatan Bioscan 782
CDD camera of 1K x 1K pixels. The colloidal suspension of MNPs were prepared by diluting the
original colloidal suspensions 100 times and maintained in an ultrasonic bath for 30 min. A drop of
the colloidal suspension was placed on a Formvar™ coated 200 mesh copper grid. The residual
excessed fluid was blotted and dried until the time that the grids was examined into microscope.
MNPs number-weighted-size distributions were obtained by measuring about ~500 MNPs with the
ImageJ freeware [25]. The number-weighted size-distribution data were fitted to a log-normal
distribution function given by [2]:
(1)
where Dm and s are fitting parameters. The number-weighted mean diameter,
deviation of particle size, , are written as [2]:
and standard
(2)
(3)
E. Small Angle X-ray Scattering
The Small Angle X-ray Scattering (SAXS) patterns of the solutions were investigated. The
setup used to data acquisition was a Xeuss (XenocsTM). The Xeuss is equipped with a microfocus x-
ray source Genix, with radiation of λ = 1.5414 Å (Cu). The system uses two scatterless slits for
beam collimation and it reaches the sample with a square cross section of 0.4 x 0.4 mm 2. The
4
primary and scattered beams remain in a vacuum (10-2 mbar) chamber to avoid scattering by the air.
Each sample were mounted in a cylindrical Mark-tubes of quartz glass capillary (Hilgenberg, 1.5
mm outer diameter) with principal axis in the vertical direction. The measured 2D scattered data
were recorded by a Pilatus (Dectris) 300 K 20 Hz 2D detector. The exposition time was 600 s and
all measurements were performed at room temperature (~ 22 °C).
The 1D scattering intensity versus scattering vector module, I(q) and q, defined by q = (4π
sinθ)/λ was obtained by averaging the data over a 20° slice in horizontal and vertical directions. The
data treatment, blank subtraction, and data normalization were performed with the software
SUPERSAXS (C.L.P. Oliveira and J. S. Pedersen, unpublished). The contribution of blank
(kerosene) was independently measured and subtracted from the sample data. To obtain the
scattered intensity data in absolute-scale units, cm-1, water was used as standard. The sample-
detector distance was 839 cm, which allows measurements with q in the range 0.01 < q < 0.35 Å−1
[2, 9, 12].
F. Zero-field cooling and field-cooling (ZFC-FC)
The magnetic properties of the MNPs were measured in a home-made Hall-effect
magnetometer [26, 27]. The magnetization curves were performed at room temperature and low
temperatures (6K) and under different applied magnetic fields from −2.0 to +2.0 T. The DC
magnetization as a function of temperature was measured in both zero-field cooled (ZFC) and field
cooled (FC) regimes. ZFC-FC protocol was performed in a temperature range from 5 to 300 K,
applying a 5.0 mT magnetic field [26, 27].
III. Results and Discussion
A. X-ray Diffraction (XRD)
XRD patterns of the samples investigated are shown in FIG. 1. These results reveal a single-
phase cubic spinel structure corresponding to the Fd3m space group [28]. The XRD pattern show
diffraction peaks corresponding to the (220), (311), (400), (422), (511), (440), (620) and (533)
crystallographic planes. These results agree with the XRD pattern of the Fe3O4.
The instrumental broadening (βhkl) was corrected, using Warren's relation
[29]. The average crystallite size (CS) of all the samples has been
estimated using full width at half maximum (FWHM) of each diffraction peak and the Debye-
Scherrer formula CS= 0.9 λ
βhkl cos(θ)
[ 28-31], where λ is the X-ray wavelength (Cu Kα, λ = 1.5414
Å), 2θ is the Bragg angle and βhkl is the FWHM of the diffraction peaks. The lattice parameter was
5
[28, 29]. The observed XRD patterns of all the
investigated using the equation
samples were analyzed by Rietveld method [32], using the MAUD 2.80 software [33] to get the
refinement parameters [28-32]. The peak shape was fitted with a pseudo-Voigt (pV) function in the
refinement procedure [30]. The cation distribution was evaluated by refining the changes in the
diffraction intensities, while locating the cation in an appropriate position. The background of each
pattern was fitted by a polynomial function of order 5 [30]. The densities of the samples were
estimated by XRD, using the relation
, where 8 refers to the number of atoms per
unit cell of the spinel structure, MW is the molecular weight of the sample. The quantities Na and
"a" are the Avogadro's number and the lattice parameter of the sample, respectively. The fitting
parameters, as well as the lattice parameter (aexp), unit cell volume V (Å3), crystallite size (CS) and
evaluated density ρXRD (g/cm3) are reported in Table I.
(a)
(b)
(c)
(d)
6
(e)
FIG. 1. XRD patterns of the magnetic nanoparticles investigated. (a) x=0.00, (b) x=0.02, (c) x=0.04, (d)
x=0.06 and (e) x=0.10. Solid lines are best fits of a pseudo-Voigt (pV) function [33].
In the XRD pattern of doped samples (FF-ND1, FF-ND2, FF-ND3 and FF-ND5) we did not
observe any peaks corresponding to NdFeO3. The ReFeO3 (RE3+=Rare-Earth) peak was observed
elsewhere due to rare earth doped ferrite [29, 30].
Table I
Average lattice parameter (aexp), unit cell volume V (Å3), crystallite size (CS), evaluated density ρXRD
(g/cm3) and results of Rietveld analysis for FeO·Fe(2-x)NdxO3 nanoparticles.
Sample
aexp (Å)
V (Å3)
CS (nm)
Comp.
(x)
0.00
FF-REF
0.02
FF-ND1
0.04
FF-ND2
0.06
FF-ND3
0.10
FF-ND5
8.39
(0.09)
8.41
(0.02)
8.42
(0.03)
8.44
(0.03)
8.46
(0.03)
590.59
(1.31)
594.82
(0.29)
596.94
(0.44)
601.21
(0.44)
605.49
(0.44)
23.9
(8.7)
18.3
(2.7)
15.6
(2.2)
16.2
(4.6)
14.9
(6.4)
ρXRD
(g/cm3)
5.18
(0.01)
5.24
(0.01)
5.27
(0.01)
5.39
(0.01)
5.56
(0.01)
Rwp(%) Rexp(%) GoF
2.23
2.26
2.60
1.58
1.83
1.29
1.73
1.24
1.81
1.26
2.06
1.32
1.19
1.23
1.48
The ionic radius of Nd3+ is 0.098 nm while the ionic radius of Fe3+ is 0.067 nm, which is
about 1.46 times smaller than the dopant radius. Therefore, replacement of Fe3+ by larger Nd3+ ions
causes an expansion of unit cell [30]. This causes an increase of the lattice parameter (aexp), as
shown in Table I, for FF-ND samples as compared with FF-REF sample. The robustness of fit
(GoF), the weighted profile factor (Rwp) and expected weighted profile factor (Rexp), assure the
reliability of the fits, since, low values of GoF were obtained [34,35].
B. XRD Cation Distribution
Several physical properties of a crystal can be accessed through the knowledge of the cation
distribution. Experimental Techniques such as: X-ray diffraction (XRD) pattern [30, 36-38], X-ray
magnetic circular dichroism (XMCD) [39] and X-ray absorption spectroscopy [39] can be used to
7
estimate the cation distribution for spinel ferrite materials. The cation distribution in the present work,
was obtained from X-ray diffraction pattern analysis. Experimental intensity ratios were compared
with the calculated intensity ratios, according to Bertaut et al.[40] method. In this method, pairs of
reflections are selected according to the expression [40].
(4)
and
where
are the experimental and calculated intensities for reflections (hkl),
respectively. We used the intensity ratios corresponding to the planes (220), (400), (440), which are
known to be sensitive to the cation distribution [30, 36-38]. In order to obtain the best-
simulated/evaluated structure, the R factor was defined, according to Eq. (5).
(5)
The determination of the structure is attained by varying the cation distribution in the
calculated intensity in such a way that the R factor will be minimized [30, 36-38].
The relative integrated intensity of the XRD lines can be calculated using Eq. (6):
where Ihkl corresponds to the relative integrated intensity. The quantity Fhkl is the structure factor,
while P is the multiplicity factor for the plane (hkl), and Lp is a Lorentz polarization factor (Eq. (7)),
and it will be a function of the Bragg diffraction angle. The multiplicity factor was obtained from
the literature [41]:
(6)
The structure factor of the spinel ferrite has 24 divalent and trivalent cations and 32 oxygen
anions [45]. The structural factors were calculated by using the equation proposed by Furuhashi et
al.[46], Eq. (8):
(7)
where
and
are related to crystal planes
and can be determined with Eq. (9) and (10):
8
(8)
and
(9)
(10)
For evaluation of the atomic scattering factor we used values reported in the International
Tables for X-ray Crystallography [42]. The temperature and absorption factors were neglected in
our evaluation because at room temperature these factors do not affect the relative XRD intensity
calculations [36]. In general, spinel structures have a high melting temperature. So, small thermo-
vibrational effect of spinel on XRD patterns is expected [43].
In the present evaluation all possible cation configurations were considered with 0.01
stoichiometric sensitivities that Nd3+ and Fe3+ ions can site in both tetrahedral and octahedral sites,
according with Eq. (11) [30]:
(Fe3+
(1-γ)Nd3+
γ]BO4
(1-δ)Nd3+
δ)A[ Fe2+Fe3+
(11)
and x are the molar stoichiometric amount of replacement of Fe3+ ions by Nd3+. The
where
closest correspondence with the actual sample structure was achieved by varying the cation
distribution of the calculated intensity, which will provide a minimum R factor. (Eq. (5)) [30]. The
cation distribution, the corresponding relative intensities of experimental and calculated XRD lines
are given in Table II.
Table II
Cation distribution of FeO·Fe(2-x)NdxO3 nanoparticles, experimental and calculated ratios between
peak intensities.
Comp. (x)
I220/I400
I220/I440
A-site
B-site
0.00
0.02
0.04
0.06
0.10
(Fe3+
(Fe3+
(Fe3+
(Fe3+
(Fe3+
1.00)
1.00)
1.00)
0.95Nd3+
0.94Nd3+
0.05)
0.06)
[Fe2+
[Fe2+
[Fe2+
[Fe2+
[Fe2+
1.00Fe3+
1.00Fe3+
1.00Fe3+
1.00Fe3+
1.00Fe3+
1.00]
0.98Nd3+
0.96Nd3+
0.99Nd3+
0.915Nd3+
0.02]
0.04]
0.01]
0.05]
Exp.
0.54
0.70
0.73
0.65
0.89
Calc
0.52
0.66
0.71
0.61
0.83
Exp.
1.15
1.50
1.38
1.62
2.04
Calc
1.03
1.33
1.29
1.57
2.03
The mean ionic radii in the tetrahedral (rA) and octahedral (rB) sites were calculated by
using Eqs. (S1) and (S2) from Electronic Supplementary Information (ESI) file. The values of the
ionic radius for rNd3+,rFe3+ and rFe3+were taken from the literature [44], 0.98 Å, 0.67 Å, and 0.49 Å,
9
respectively. The value of the oxygen positional parameter u can be determined with Eq. (S3)(ESI),
where RO is the radius of oxygen ion (1.32 Å) [25,31,39]. The theoretical lattice constant (ath) is
calculated by using Eq. (S4)(ESI) [30,36,45].
The structural parameters: tetrahedral bond length (dAL); octahedral bond length (dBL);
tetrahedral edge length (dAE); shared (dBE ) and unshared (dBEU) octahedral edge length; the
tetrahedral and octahedral jump length (LA and LB ) were calculated using Eqs. (S5)-(S11)(ESI).
The results are given in Table III.
Table III
Theoretical parameters based on the proposed cation distribution. See text for the symbols.
Comp
(x)
0.00
0.02
0.04
0.06
0.10
dBEU
(Å)
2.969
2.976
2.980
2.987
2.997
u (Å) dAL (Å) dBL (Å) dAE (Å) dBE (Å)
0.384
0.384
0.384
0.385
0.384
0.560
0.564
0.567
0.562
0.573
1.950
1.950
1.950
1.968
1.971
2.022
2.030
2.034
2.031
2.039
3.184
3.184
3.184
3.213
3.219
2.745
2.761
2.769
2.752
2.767
LA (Å) LB (Å)
3.631
3.641
3.646
3.653
3.665
2.965
2.973
2.977
2.983
2.993
rA (Å)
rB (Å) ath (Å)
0.630
0.630
0.630
0.648
0.651
8.400
8.409
8.418
8.432
8.468
Table III shows that the theoretical lattice parameter (ath), tetrahedral bond length (dAL),
octahedral bond length (dBL), tetrahedral edge (dAE), octahedral edge (dBE), unshared octahedral edge
(dBEU) increase with an increase in Nd-content. Table III also shows that tetrahedral radius (rA) and
octahedral radius (rB) increase as the Nd-content increases, while the oxygen parameter u remains
unchanged. The anion in the spinel structure, O2- ions, in this case, are not in general located at a
fixed position of the FCC sublattice. The anion is allowed to translate and this translation is
measured by a quantity named oxygen positional parameter or anion parameter. If we assume the
center of symmetry at (3/8, 3/8, 3/8) position that correspond to origin at A-site, the value of u ideal
is expected to be 0.375. Therefore, changes in the value of u can be interpreted as a relaxation of the
structure to accommodate the cations of different radius in the A and B sites [30, 47]. The jump
(hopping) lengths, LA and LB between the magnetic ions at A-site and B-site respectively, were
calculated (Table III). Since the Fe3+ radius (0.65Å) is smaller than that of the Nd3+ (0.98 Å), the
replacement of Fe3+ leads to an increase in rA and rB. Moreover, LA and LB increases with an
increase in Nd-content. The results showed that Nd3+ ions are present in both sites at different
concentrations, with the displacement of Fe3+ ions. Therefore, it was possible to observe changes in
structural parameters like bond lengths; shared and unshared edges, among others.
The magnetic properties of the particles depend on the exchange interactions between metal
ions. The bond angle and inter-ionic bond length between metal ions are the more important in the
overall magnitude of the magnetic exchange interaction. The magnitude of the magnetic exchange
10
interactions is proportional to the bond angles and inversely proportional to the inter-ionic bond
lengths. The inter-ionic bond lengths i.e., cation-cation distances and cation-anion distances (FIG.
S1(ESI)) were calculated using equations Eqs. (S12)-(S25)(ESI). The values obtained for inter-ionic
bond lengths were used for the evaluation of bond angles between the metal ions using equation
Eqs. (S21)-(S25). The values for bond angles are given in Table IV.
Table IV
Theoretical bond angles between metal ions based on the cation distribution. All angles are given in
degrees (º).
Comp (x)
0.00
0.02
0.04
0.06
0.10
θ1
122.14
122.28
122.35
121.99
122.10
θ2
138.45
138.63
138.72
138.28
138.41
θ3
94.33
94.14
94.05
94.52
94.38
θ4
126.06
126.03
126.02
126.08
126.06
θ5
70.91
71.10
71.19
70.74
70.87
The bond angles θ1, θ2, and θ5 are associated with the A-B and A-A exchange interactions,
while θ3, and θ4 are associated with the B-B exchange interactions (FIG. (S1)(ESI)). The observed
increase in θ1, θ2, and θ5, corresponding to x=0.00 to x=0.04 (Table IV) suggests the strengthening
of the A-B and A-A interactions, while θ3, and θ4 decrease indicates a weakening of the B-B
interaction Comparing the systems with x=0.06 and x=0.00, the B-B exchange interactions are
enhanced (θ3 and θ4), while A-B (θ1, θ2) and A-A (θ5) exchange interactions presents the lowest
values, suggesting weakening of A-B and A-A exchange interactions. For x=0.10 all bond angles
have values similar to x=0.00. This result suggests that, beyond x=0.06, we have a decreasing on
the B-B interaction and strengthening of the A-B and A-A interactions.
C. Optical Band Gap
The optical band-gaps of the FeO·Fe(2-x)NdxO3 nanoparticles were obtained by UV-Vis
spectrophotometry at the temperature of 25oC. The wavelength range of 350 nm to 850 nm using
DH-2000-BAL (Mikropack) equipped with deuterium and tungsten halogen lamps. Attached to a
DH-2000-BAL, an Ocean Optics® spectrometer USB4000 was used to measure the UV-Vis spectra
[42-44]. The band gap of the nanoparticles is related to the optical gap (Eg) and photon energy (hυ)
according to Eq. (12) [18, 48-50]:
(12)
11
where C is a constant, and α is the linear absorption coefficient. The linear absorption coefficient α
was calculated from the absorbance measurement A(hυ) as a function of the photon energy using
Eq. (13) [48-50]:
(13)
where Ascatt(hυ) is related to the Rayleigh-scattering contribution to the extinction measured data.
Ascatt(hυ) was estimated as having a λ-4 dependence, more details can be found elsewhere [42,44].
The value of n will be given according to the type of the electronic transition responsible for
the absorption: for allowed indirect transition is n = 1/2; n = 3/2 for forbidden indirect transition; for
allowed direct transition n = 2; n = 3 for forbidden direct transition [18,48-50]. The optical gap for
both direct and indirect allowed transitions were studied in this work.
Here, A(hυ) is the experimental absorbance measured for each sample. The absorbance was
measured using the UV-Vis spectrophotometry, and L = 1cm is the width of the cuvette cell. The
optical band-gap was obtained by extrapolating the linear region of the Tauc plot (plot of [αhυ] n
versus hυ) to a value of hυ= 0 [18, 48-50]. Tauc plot obtained for FF-REF and Nd +3 substituted
samples FF-ND1, FF-ND2, FF-ND3, FF-ND5 are shown in FIG. 2.
a)
b)
FIG. 2. (a) Tauc plots for [αhυ]2 and (b) [αhυ]1/2 versus the photon energy for determination of direct and
indirect optical gaps, respectively.
12
Table V
Bandgaps of magnetic nanoparticles studied.
Comp. (x)Nd
Sample
FF-REF
FF-ND1
FF-ND2
FF-ND3
FF-ND5
0.00
0.02
0.04
0.06
0.10
Ed (eV)
3.00 (0.01)
3.12 (0.01)
3.05 (0.01)
3.16 (0.01)
3.13 (0.01)
Ei (eV)
2.03 (0.01)
2.09 (0.01)
1.98 (0.01)
2.10 (0.01)
2.20 (0.01)
Fontijn et al reported [51-53] that, for magnetite electronic transitions, Fe2+ [t2g] → Fe3+ [e]
occurring in the tetrahedral sites, the bandgap is 3.11 eV, while Fe2+ [t2g] → Fe3+ [eg] transitions
occurring in the octahedral sites the bandgap is 1.94 eV. Their results were obtained through
magneto-optical polar Kerr measurements for bulk single-crystalline Fe3O4 and Mg2+ or Al2+
substituted Fe3O4. These values are consistent with those obtained here. One can see in Table V, that
for all samples with x ≠ 0.00, both Ed and Ei increase as x increases. We also point out that Ed
reaches the maximum value for x=0.06 (FF-ND3) rather than Ei that reaches the maximum value
for x=0.10.
J. Anghel and co-workers showed that exists a strong correlation between modifications in
the lattice parameters and the bandgap energy for Zn(1−x)MxO (M = Cr, Mn, Fe, Co, or Ni) samples
[54]. In their studies they observed that the unit cell volume (obtained by XRD) and bandgap
(obtained by spectrophotometry) reached the highest values of Fe3+ substitution [54]. For the series
of metals M that they studied, iron was the one with larger ionic radius, that leads to a higher lattice
parameter and hence, unit cell volume. A similar trend was also observed elsewhere [55,56]. From
XRD we found that the lattice parameter increases with increasing Nd-content and hence, unit cell
volume, bond lengths (dAL, dBL) and hopping lengths, LA and LB between the magnetic ions at A-site
and B-site. In this way, the increasing in the bandgap observed here, may be understood since the
bandgap is directly proportional to the interatomic separation, although it is also possible that new
electronic states may exist due to the presence of the dopant [55].
E. Transmission electronic microscopy (TEM)
Typical TEM micrographs of the samples investigated are shown in FIG. 3 and 4 for FF-
REF and FF-ND3 samples, respectively. Typical TEM micrographs for FF-ND1, FF-ND2 and FF-
ND5 samples is given in FIG. (S2), (S3) and (S4) (ESI), respectively. These results reveal a
polydisperse size and shape distribution, as expected for co-precipitation synthesis method, with a
broad number weighted-size distribution. The size distribution ranges from 5 to 30 nm and follows
the log-normal distribution. The mean-number weighted diameters are given in Table VI. The
13
polydispersity index (PDI = σ(DTEM )/
VI.
) also were evaluated and they are given in the Table
(a)
(b)
FIG. 3. (a) Typical TEM micrography and (b) Number-weighted diameter distribution of the FF-REF
sample. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(a)
(b)
FIG. 4 (a) Typical TEM micrography and (b) Number-weighted diameter distribution of FF-ND3 sample.
The solid line is the log-normal fitting of size distribution given by Eq. (1).
Table VI
Particles' mean diameter (D), distribution width (σ) and polydispersity and index (PDI) of the magnetic
fluids investigated.
Sample
FF-REF
FF-ND1
FF-ND2
FF-ND3
diameter D (σ) [nm]
12.81 (9.15)
12.50 (3.55)
11.67 (3.53)
7.99 (3.67)
14
PDI
0.71
0.28
0.30
0.46
FF-ND5
9.06 (2.83)
0.31
F. Small Angle X-ray Scattering (SAXS)
The direct obtained experimental scattering pattern by Small Angle X-ray Scaterring
(SAXS) can provide several characteristic properties of the sample irradiated by the X-ray beam [2 ,
9]. For polydisperse systems, the experimental scattering pattern corresponds to an average that is
performed over the particles present in solution [2, 9].
In the analysis of the scattered intensity data, it was assumed that I(q) was produced by a
system composed of globular nanoparticles. Therefore, I(q) has a contribution from the particles'
form-factor and the interaction between them (structure-factor). Assuming a model of polydisperse
hard spheres with radius r, the scattered intensity is given by Eq. (14) [12]:
(14)
In this model, the scattering elements are considered an ensemble of polydisperse non-
interacting spheres. The quantities r and V are the radius and volume for each sphere from the
ensemble. The fv(r) is the normalized-volume weighted-radius distribution function. Isph is the
normalized scattering intensity owing to a sphere of radius r and, Sc stands for a scaling factor [12].
The Gnom software was used to analyze the scattering intensity Is versus the scattering vector q
(modulus of q) [57,12]. This approach was used to determine the volume-weighted size-distribution
function fv(r) from the adjustment of the experimental scattering intensity data to Is(q), given by Eq.
(14). The solid lines in FIG. 5 and 6 for FF-REF and FF-ND3, represent the best fit of Eq. (14) to
the experimental data. For the remaining samples, namely FF-ND1, FF-ND2 and FF-ND5 the
experimental data and the best fit are shown in FIG. (S5),(S6) and (S7)(ESI), respectively. The
volume-weighted mean particle's diameter
was calculated according to Eq. (15) [12]:
(15)
15
(a)
FIG. 5: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to best
fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-REF.
(b)
(a)
(b)
FIG. 6: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to best
fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND3.
Table VII
Nanoparticles' Mean diameter
distribution function (σV=3 σD).
determined by SAXS and width of the log-normal volume
Sample
FF-REF
FF-ND1
FF-ND2
FF-ND3
FF-ND5
Comp. (x)Nd
0.00
0.02
0.04
0.06
0.10
(nm)
18.48
28.74
29.15
20.32
28.84
σV
1.83
1.86
1.83
2.07
1.77
The results given in Table VII for the volume-weighted average diameter measured by
SAXS are consistent with those determined by XRD.
16
G. Magnetization as a function of the applied field.
FIG. 7 (a) and (b) show the measured magnetization (M) versus the applied field (H) for
FeO·Fe(2-x)NdxO3 samples, at 6 K and 300 K, respectively. From these curves, the saturation
magnetization (Ms), coercivity field (Hc) and remanence magnetization (Mr) are obtained and are
given in Table VIII. The saturation magnetization increases in a steady manner upon increasing the
Nd3+ ion molar ratio from 0.00 up to 0.06, reaching the maximum value of 105.8±0.4 Am2/Kg at x=
0.06. It's worth to mention that the result for saturation magnetization value are higher than that of
the bulk material. In this case, suggesting the existence of a highly ordered spin configuration
distributed across the volume of the magnetic nanoparticles.
The M(H) curves for magnetic nanoparticles in superparamagnetic (SPM) regime (FIG. 7-
(b)) are fitted with a Langevin function (Eq. (16)), weighted with the particle-size distribution
function (PDF) [58,59]:
(16)
where M(H,T) is the magnetization of the magnetic nanoparticles at the temperature T (300 K) , Nsc
is a scaling constant, MS represents the saturation magnetization of the magnetic nanoparticles and
[59]. The quantity VV, stands for the volume-weighted average volume of the
nanoparticles given by
average radius and hence diameter, measured by SAXS [12].
[52]. Here we used,
from the volume-weighted
In the model expressed by Eq. (16), the magnetization M as a function of field H has
contributions from superparamagnetic (SPM) and paramagnetic (PM) particles [31]. The quantity c
is the paramagnetic contribution (linear with the magnetic field, H [36]). We found that the SPM
regime contributes with 95-97% for all samples, while only 3-5% comes from the PM regime [36].
The MNPs are expected to have a nonmagnetic layer around the magnetic core (core-shell
model) [60, 61]. The thickness of this magnetically inert shell was evaluated from the volume-
weighted size distribution function, according to Eq. (17) [60, 61]:
(17)
where δ is the thickness of the shell, MS is saturation magnetization of the nanoparticles
17
MSbulk stands for the saturation magnetization of the bulk material [60-62] and the reciprocal of the
volume-weighted average diameter (
) of the particles [60,61]. From Eq. (16) we found that
the thickness of the magnetically inert layer of the materials studied here, ranges from 0.5 Å to 7.8
Å.
The Neel's theory for collinear ferro/ferrimagnetism of two sub-lattices model, predicts the
net magnetic moment per formula unit (f.u.) according to Eq. (17):
(18)
where Mocta and Mtetra are the magnetic moments of B (octahedral) and A (tetrahedral) sites in µB
units (Bohr magneton) [36, 31]. Based on the site occupancy obtained from the XRD cation
distribution and magnetic moment of 5, 4 and 3.2 µB [63] for Fe3+ , Fe2+ and Nd3+ ions, respectively,
the nNeel values were calculated using Eq. (18) and are given in Table VIII. On the other hand, the
magnetic moment per formula unit in Bohr magneton unit nexp (experimental) can be calculated
from the saturation magnetization Ms according to Eq. (18):
(19)
where MW is the molecular weight. The calculated nNeel magneton number using the XRD data
agrees with the experimentally obtained magneton number from the M-H loops. The results are
comparable to those determined by Eq. (19). The calculated values of magneton number nNeel and
nexp are comparable. Furthermore, the values of nNeel and nexp show same trend, that is, increases with
x=0.00 to x =0.04, reaches the maximum value at x = 0.06 and decreases for x =0.10.
The Yafet-Kittel angles (αY-K) described by equation Eq. (20) can give us insight about the
type of magnetic ordering for FeO·Fe(2-x)NdxO3 samples [30, 36]. The quantity nexp is given by Eq.
(19) while Mtetra and Mocta are given by Eq. (18). The Yafet-Kittel angles (αY-K) obtained were 21.6
º, 16.3 º, 14.1 º, 0.0 º and 0.0 º for x=0.00, 0.02, 0.04, 0.06 and 0.10, respectively. From x=0.00 to
x=0.04 αY-K decreases with increasing Nd3+ substitution. The decrease in αY-K for this samples is
due to a noncollinear type of magnetic ordering, since the Yafet-Kittel angles are different from
zero. The existence of non-zero αY-K suggest a model of canted-spin magnetization, which should
have a triangular spin arrangement and is suitable on the B-site that leads to a reduction in the A -- B
exchange interaction and enhancement in the B-B exchange interaction [30]. For x=0.06, there is a
transition to the Néel-type magnetic ordering. This result suggests a reinforcement of a dominant A-
18
B super-exchange interaction and these results are corroborated by those found in the inter-ionic
bond angles of Table IV.
(20)
(a)
(b)
FIG. 7. Magnetization curves of FF-REF, FF-ND1, FF-ND2, FF-ND3 and FF-ND5 nanoparticles a)
at 6 K and b) at 300 K, the dots are experimental results and solid lines are their size-weighted
Langevin fits.
Table VIII
Magnetic parameters extracted from M(H) curves.
Sample
Comp.
(x)
Saturation
magnetization
(Ms) Am2/Kg
Remnant
magnetization
(Mr) Am2/Kg
Coercivity
(Hc) mT
Squareness
ratio (S)
FF-REF
FF-ND1
FF-ND2
FF-ND3
FF-ND5
0.00
0.02
0.04
0.06
0.10
81.9 (0.8)
86.1 (0.3)
87.7 (0.3)
105.8 (0.4)
91.0 (0.3)
30.4 (0.2)
28.3 (0.1)
38.6 (0.2)
24.3 (0.1)
32.8 (0.2)
156
156
121
157
156
0.37
0.33
0.44
0.23
0.36
Magnetic
anisotropy
energy
density
(Keff) 103
J/m3
33.1
35.1
27.9
44.8
40.2
Bohr
magneton
number
(µB) exp.
Neel's
Bohr
magneton
number
(µB) XRD.
3.40 (0.01) 4.00 (0.44)
3.60 (0.01) 3.96 (0.44)
3.69 (0.01) 3.92 (0.43)
4.49 (0.02) 4.08 (0.45)
3.97 (0.02) 3.95 (0.42)
For magnetite (Fe3O4), below the Verwey Temperature (TV), the magnetocrystalline
anisotropy is expected to be uniaxial [62, 64]. Therefore, the experimental coercivity field and
saturation magnetization are related to the effective anisotropy constant Keff through Eq. (21) [62,
64]:
19
To determine the squareness ratio (S), Eq. (22) was employed [65, 14]:
(21)
(22)
The evaluated values of the magnetocrystalline anisotropy constant (Keff) and S are given in
Table VIII. For all samples S < 0.5, which indicates uniaxial anisotropy contribution in the prepared
FeO·Fe(2-x)NdxO3 nanoparticles [60, 62-65]. The overall values of Keff obtained are of the same order
of magnitude of that from the magnetite bulk (1.1 -- 1.3 104 J/m3) [62-65] however, we have
obtained values 3 to 4 times higher.
H. Zero-field cooled and field-cooled (ZFC-FC)
FIG. 8(a)-(e) show the temperature dependence of the magnetization in low external applied
field, the zero-field cooled curve (ZFC) and field-cooled curve (FC) for FF-REF, FF-ND1, FF-ND2,
FF-ND3 and FF-ND5, respectively. In the ZFC-FC protocol the MNPs were frozen in the absence
of the magnetic field, fast enough that the random orientation of their easy axis is preserved [12,
13]. The system was superparamagnetic (SPM) at room temperature, and the magnetization curves
M(T) were collected in ZFC and FC modes. The overall shape of the ZFC-FC curves indicate weak
interaction between particles. In the ambit of the non-interacting particles' model, the blocking-
temperature distribution function f(TB) is expected to be broad [12]. One can use the Stoner-
Wohlfarth model to describe uniaxial, single-domain [8] and non-interacting particles to obtain the
blocking-temperature distribution function given by Eq. (23) [12]. FIG. 8(f)-(i) show f(TB) for FF-
REF, FF-ND1, FF-ND2, FF-ND3 and FF-ND5, respectively. A log-normal distribution function was
and standard deviation σT
used to fit these results and obtain the mean blocking temperature
[12]. For FF-REF, FF-ND1, FF-ND2, FF-ND3 and FF-ND5 samples, the average blocking
temperatures were 40.2 K, 28.8 K, 23.6, 25.1 and 41.7 K, respectively, at an applied field of 5 mT.
Note that FIG. 8(a)-(e) also show that the maximum of ZFC curve for all cases is close to the
temperature in which the ZFC-FC curves split. This effect is due to particle dipole-dipole
interaction.
(23)
20
(f)
(g)
(h)
(i)
(a)
(b)
(c)
(d)
21
(e)
(j)
FIG. 8. (a, b, c, d and e) ZFC-FC magnetization curves, where () blue circles represent the FC
susceptibility, whereas red circles () symbols represent the ZFC susceptibility for FF-REF, FF-ND1, FF-
ND2, FF-ND3, and FF-ND5, respectively. 8 (f, g, h, i and j) Temperature derivative [-d(MFC-MZFC)/dT] of
the difference between FC and ZFC magnetization curves for samples FF-ND, FF-RC and FF-REF,
respectively.
The log-normal width σT, obtained from the fitting to the experimental data -- FIG. 8(f)-(i) --
were used to evaluate the number of interacting particles (Ni) and the correlation volume (Λ3),
defined according to Eqs. (24) and (25) [12, 13]:
(24)
(25)
where σV , σT are the width of the volume and blocking temperature distribution functions. The
quantity
is the average volume defined as:
(26)
is the volume-weighted average diameter measured by SAXS. It's worth to mention,
where
that according with M. El-Hilo [58], the volume-weighted diameter distribution fv(D) should
converts to volume-weighted distribution fv(V) with σV= 3σD and average volume-weighted by
volume given by Eq.(26). The quantity ϕ is the volumetric fraction of particles in the solution [8,
12, 13], ~ 0.3% for all the samples investigated.
Table IX
Width of volume and blocking temperature distribution σV , σT. Number of interacting particles Ni.
The correlation volume (Λ3) and correlation length (Λ).
Sample
FF-REF
FF-ND1
σV
1.83
1.86
σT
0.37
0.76
Ni
5
3
22
Λ3 (10-23 m3)
5.4
10.1
Λ (nm)
176
216
FF-ND2
FF-ND3
FF-ND5
1.83
2.07
1.77
0.41
0.46
0.60
5
5
3
19.3
6.6
12.4
226
187
231
Table IX, shows estimation of the number of interacting particles (Ni) inside the correlation
volume (Λ3) and correlation length (Λ) [12]. The results did not show correlation between Ni and
the size of the particles. However, the correlation volume (Λ3) and correlation length (Λ) are
smaller for FF-REF and FF-ND3, which are those with smaller volume-weighted average diameter
measured by SAXS. The correlation volume (Λ3) and correlation length (Λ) also are larger for those
particles with larger volume-weighted average diameter (FF-ND1, FF-ND2 and FF-ND5). This is
due to the fact that for larger particles, the interparticle distance increases, as the mean-particle
diameter increases [12]. Note that, the correlation length (Λ) is out of the range accessible in our
SAXS experiments, that is, about ~60 nm.
V. Summary and Conclusions
Magnetic fluid based on magnetite were synthesized and XRD patterns of the samples
investigated revealed a single-phase cubic spinel structure of Fd3m space group. From the cation
distribution, the theoretical lattice parameter (ath), tetrahedral bond length (dAL), octahedral bond
length (dBL), tetrahedral edge (dAE), octahedral edge (dBE), unshared octahedral edge (dBEU) increase
with the increase in Nd-content. Since the Fe3+ (0.65Å) is smaller than Nd3+ (0.98 Å), the
replacement of Fe3+ leads to an increase in rA and rB. Moreover, LA and LB increase with higher Nd-
content. The results showed that Nd3+ ions are placed in both sites at different concentrations.
Therefore, it was possible to observe modifications in structural parameters like bond lengths;
shared and unshared edges. The increasing in the bandgap may be interpreted as a result of the
higher interatomic separation with the doping. TEM micrographs reveal a polydisperse size and
shape distribution of particles, as expected for co-precipitation method, with a broad number-
weighted size distribution. The results for volume-weighted average diameter measured by SAXS
are consistent with those determined by XRD. From the M-H Loops we found that the SPM regime
contributes with 95-97% for all samples, while only 3-5% contribution comes from the PM regime.
The saturation magnetization increases in a steady manner upon increasing the Nd3+ ion molar ratio
from 0.00 up to 0.06, reaching the maximum value of 105.8±0.4 Am2/Kg at x= 0.06. It's worth to
mention that the result for saturation magnetization value are higher than that of the bulk material.
In this case, suggesting the existence of a highly ordered spin configuration distributed across the
volume of the magnetic nanoparticles. The squareness values for all samples are less than 0.5,
which indicates uniaxial-anisotropy contribution in the FeO·Fe(2-x)NdxO3 nanoparticles. The overall
23
values of Keff obtained for all samples studied, have the same order of magnitude of that from the
bulk magnetite (1.1 -- 1.3 104 J/m3) however, we have obtained values 3 to 4 times higher. The
magnetic measurements indicate the existence of low interaction between the MNPs at the
concentrations investigated.
ACKNOWLEGMENTS
This research was financial supported by Brazil's agencies INCT/CNPq (Conselho Nacional
de Desenvolvimento Científico e Tecnológico; Grant Number: 465259/2014-6), INCT/FAPESP
(Fundação de Amparo à Pesquisa do Estado de São Paulo; Grant Number: 14/50983-3),
INCT/CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior; Grant Number:
88887.136373/2017-00), and INCT-Fcx (Instituto Nacional de Ciência e Tecnologia de Fluidos
Complexos).
References
[1] D. H. K. Reddy and Y. S. Yun,
Coord. Chem. Rev. 315, 90-111 (2016).
[2] K. Parekh et al,
Mater. Chem. Phys. 222, 217-226 (2019).
[3] L. Shen, et al,
Ceram. Int. 40, 1519-1524 (2014).
[4] I. Torres-Díaz and C. Rinaldi,
Soft Matter 10, 8584-8602 (2014).
[5] C. Scherer and A.M. Figueiredo Neto,
Braz. J. Phys. 35(3A), 718-727 (2005).
[6] A. G. Kolhatkar, A. C. Jamison, D. Litvinov, R. C. Willson, and T. R. Lee,
Int. J. Mol. Sci. 14, 15977-16009 (2013).
[7] K. Parekh, L.Almásy, H. S. Lee and R V Upadhyay,
Appl. Phys. A, 106, 223 (2012).
[8] G. F. Goya et al.,
J. Appl. Phys. 94 (5), 3520-3528 (2003).
[9] D. H. G. Espinosa, C. L. P. O Liveira And A. M. Figueiredo Neto,
J. Opt. Soc. Am. B 35(2), 346-356 (2018).
[10] E. A. Elfimova, A. O. Ivanov and P. J. Camp
J. Chem. Phys. 136, 194502 (2012).
24
[11] N. Daffé et al,
J. Magn. Magn. Mater. 460, 243-252 (2018).
[12] E. S. Gonçalves et al.,
Phys. Rev. E 91, 042317-1 -- 042317-7 (2015).
[13] J. C. Denardin et al.,
Phys. Rev. B 65, 064422 (2002).
[14] R.R. Kanna, et al.,
J Mater Sci: Mater Electron 30, 4473 (2019).
[15] S. Aslam et al, Results in Physics 12 , 1334-1339 (2019).
[16] P. P. Naik, R. B. Tangsali, S.S. Meena and S. M. Yusuf,
Mater. Chem. Phys. 191, 215-224 (2017).
[17] S. Amiri and H. Shokrollani,
J. Magn. Magn. Mater. 345, 18-23 (2013).
[18] R. Jain et al.
J. Magn. Magn. Mater. 456, 179 -- 185 (2018).
[19] W. Huan et al,
J. Nanosci. Nanotechno. 14, 1 -- 9 (2014).
[20] A. A. Veligzhanin, D. I. Frey, A. V. Shulenina, A. Yu. Gruzinov, Ya. V. Zubavichus and M. V.
Avdeev, J. Mag. Mag. Mat. 459, 285-289 (2018).
[21] F. Döbrich, A. Michels and R. Birringer,
J. Mag. Mag. Mat. 316, e779 -- e782 (2007).
[22] Alvaro V. N. C. Teixeira et al.,
Phys. Rev. E 67, 021504-1 -- 021504-7 (2003).
[23] H. Harzali, A. Marzouki, F. Saida, A. Megriche and A. Mgaidi,
J. Magn. Magn. Mater. 460, 89-94 (2018).
[24] X. Huang et al,
25
J. Magn. Magn. Mater. 405, 36-41 (2016).
[25] Rueden, C. T.; Schindelin, J. & Hiner, M. C. et al,
BMC Bioinformatics 18:529, 1-26 (2017).
[26] J. F. D. F. Araujo, A. C. Bruno, and S. R. W. Louro,
Rev. Sci. Instrum 86; 105103 (2015).
[27] S. Arsalani et al,
J. Magn. Magn. Mater. 475, 458 -- 464 (2019).
[28] G. D. Gatta, I. Kantor, T. B. Ballaran, L. Dubrovinsky, and C. McCammon,
Phys. Chem. Min. 34, 627- 635 (2007).
[29] H. Harzali et al.,
J. Magn. Magn. Mater. 460, 89 -- 94 (2018).
[30] K.V. Zipare et al.,
Journal of Rare Earths 36, 86-94 (2018).
[31] A.S. Elkady et al.,
J. Magn. Magn. Mater. 385, 70 -- 76 (2015).
[32] Georg Will, Powder Diffraction, The Rietveld Method and the Two Stage Method to Determine
and Refine Crystal Structures from Powder Diffraction Data, Springer-Verlag Berlin Heidelberg,
(2006)
[33] L. Lutterotti,
Nucl. Instrum. Meth. B 268, 334-340 (2010).
[34] M. Dalal et al.,
Mater. Res. Bull. 76, 389 -- 401 (2016).
[35] Kumar et al.
International Nano Letters 3:8 (2013).
[36] J. Kurian, M. Jacob Mathew,
J. Magn. Magn. Mater. 428, 204 -- 212 (2017).
[37] D.V.Kurmude et al.,
J. Supercond. Nov. Magn. 27(2), 547-553 (2014).
[38] Patange et al.,
26
J. Appl. Phys. 109, 053909 (2011).
[39] C. E. Rodríguez Torres et al.,
Phys. Rev. B 89, 104411 (2014).
[40] E.F. Bertaut, Study of the nature of spinel ferrite, Weekly the Proceedings of
Sessions of the Academy of Sciences, C.R. Acad. Sci. 230, 213 -- 215 (1950).
[41] B.D. Cullity,
Elements of X-ray Diffraction, Addison-Wesley, Reading,
Massachusetts, USA, (1978).
[42] P. J. Brown, A. G. Fox, E. N. Maslen, M. A. O'Keefe and B. T. M. Willis. International Tables
for Crystallography Vol. C, ch. 6.1, 554-595 (2006).
[43] H. Kavas et al.,
J. Alloy and Compd. 479, 49 -- 55 (2009).
[44] R. D. Shannon,
Acta Cryst. A32, 751-767 (1976).
[45] A. Najafi Birgani et al,
J. Magn. Magn. Mater. 374, 179 -- 181 (2015).
[46] H. Furuhashi, M. Inagaki, S. Naka,
J. Inorg. Nucl. Chem. 35, 3009 -- 3014 (1973).
[47] B. B. V. S. Varaprasad, K. V. Ramesh and A. Srinivas,
J. Supercond. Nov. Magn. 30, 3523 -- 3535 (2017).
[48] Espinosa, Gonçalves, and Figueiredo Neto,
J. Appl. Phys. 121, 043103 (2017).
[49] M. Dongol et al.,
Optik 126, 1352 -- 1357 (2015).
[50] D. Espinosa, L. B. Carlsson, A. M. Figueiredo Neto, and S. Alves,
Phys. Rev. E 88, 032302 (2013).
[51] W. F. J. Fontijn, P. J. van der Zaag, L. F. Feiner, R. Metselaar, and M. A. C. Devillers,
J. Appl. Phys. 85, 5100 (1999).
[52] W. F. J. Fontijn, P. J. Van der Zaag, M. A. C. Devillers, V. A. M. Brabers, and R. Metselaar,
Phys. Rev. B 56, 5432 (1997).
27
[53] P. J. van der Zaag, W. F. J. Fontijn, P. Gaspard, R. M. Wolf, V. A. M. Brabers, R. J. M. van de
Veerdonk, and P. A. A. van der Heijden,
J. Appl. Phys. 79, 5936 (1996).
[54] J. Anghel, A. Thurber, D. A. Tenne, C. B. Hanna, and A. Punnoose,
J. Appl. Phys. 107, 09E314 (2010).
[55] Thurber et al.,
Phys. Rev. B 76, 165206 (2007).
[56] W. Luo, S. R. Nagel, T. F. Rosenbaum, and R. E. Rosensweig,
Phys. Rev. Lett. 67, 2721 (1991).
[57] Feigin L.A. & Svergun D.I., Structure Analysis by Small-Angle X-Ray and Neutron Scattering.
NY: Plenum Press (1987).
[58] M. El-Hilo,
J. Appl. Phys. 112, 103915 (2012).
[59] Tamion et al,
Appl. Phys. Lett. 95, 062503 (2009).
[60] D. Caruntu, G. Caruntu and C. J O'Connor,
J. Phys. D: Appl. Phys. 40, 5801 -- 5809 (2007).
[61] Chen J P, Sorensen C M, Klabunde K J, Hadjipanayis G C, Devlin E and Kostikas A,
Phys. Rev. B 54, 9288 (1996).
[62] Goya, G., Berquó, T. & Fonseca, F.,
J. Appl. Phys. 94, 3520 (2003).
[63] R. Przeniosto, I. Sosnowska and P. Fischer,
J. Mag. Mag. Mat., 140-144, 2153-2154 (1995).
[64] L. Tauxe, H. Neal Bertram and C. Seberino,
Geochem. Geophy. Geosy., 3(10) (2002), 1055
[65] M Graeser, K Bente and T M Buzug,
J. Phys. D: Appl. Phys. 48, 275001 (11pp) (2015)
28
Electronic supplementary information.
Magnetic, Structural and cation distribution studies on FeO·Fe(2-x)NdxO3
(x=0.00, 0.02, 0.04, 0.06 and 0.1) nanoparticles.
W. W. R. Araujo*, C. L. P. Oliveira, G. E. S. Brito, A. M. Figueiredo Neto
Institute of Physics, University of São Paulo, Rua do Matão 1371, 05508-090, São Paulo, SP,
Brazil.
Department of Physics, Pontifical Catholic University of Rio de Janeiro, RJ, Brazil.
J. F. D. F. Araujo,
The mean ionic radii in the tetrahedral (rA) and octahedral (rB) sites were calculated by using
Eqs. (S1) and (S2) [1-4]:
(S1)
(S2)
where, CAi ( i=Fe3+,Nd3+ ) are the concentration of ions in the tetrahedral site and CBi ( i=
Fe2+,Fe3+,Nd3+ ) the concentration of ions in the octahedral site. The values of the ionic radius for
rNd3+,rFe3+ and rFe3+were taken from the literature, 0.98 Å, 0.67 Å, and 0.49 Å, respectively.
The value of the oxygen positional parameter u can be determined with Eq. (S3) [1-4]:
where RO is the radius of oxygen ion (1.32 Å).
The theoretical lattice constant (ath) is calculated by using Eq. (S4):
(S3)
(S4)
The values of tetrahedral bond length (dAL), octahedral bond length (dBL), tetrahedral edge length
(dAE), shared (dBE ) and unshared (dBEU) octahedral edge length, the tetrahedral and octahedral jump
length (LA and LB ) are determined using Eqs. (S5)-(S11) [1-4]. The results are given in Table III.
(S5)
(S6)
(S7)
(S8)
(S9)
(S10)
(S11)
The configuration of ion pairs in the spinel structure, as well as their distances and angles has
been illustrated in FIG. (S1) [1-4].
FIG. S1. Configuration of the ion pairs in spinel ferrites with favorable distances and angles.
The bond lengths values for Me-Me (metal-metal) and Me-O (metal-oxygen) were calculed
theoreticaly using the following sets of equations, Eqs. (S12)-(S20) [3]:
Me-Me:
(S12)
(S13)
Me-O:
The values of bond angles were calculated using Eq. (S21)-(S25), as follows:
(S14)
(S15)
(S16)
(S17)
(S18)
(S19)
(S20)
(S21)
(S22)
(S23)
(S24)
(S25)
where b, c, d, e, f, p, q, r and s represent the parameters of ion pair distances associated with angles
θ1, θ2, θ3, θ4, and θ5 for spinel ferrites structures [1-4]. The parameter δ is the deviation of the
oxygen positional parameter u from ideal position uideal = (3/8, 3/8, 3/8) i.e. δ = u -- uideal [2].
Typical TEM micrographs for FF-ND1, FF-ND2 and FF-ND5 samples is given in FIG. (S2),
(S3) and (S4), respectively.
(a)
FIG. S2. (a) Typical TEM micrography and (b) Number-weighted diameter distribution for FF-ND1
samples. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(b)
(a)
FIG. S3 (a) Typical TEM micrography and (b) Number-weighted diameter distribution for FF-ND2
samples. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(b)
(a)
FIG. S4 (a) Typical TEM micrography and (b) Number-weighted diameter distribution for FF-ND5
samples. The solid line is the log-normal fitting of size distribution given by Eq. (1).
(b)
FIG. S5: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to
best fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND1.
FIG. S6: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to
best fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND2.
(a)
(a)
(a)
(b)
(b)
(b)
FIG. S7: (a) X-ray scattering intensity versus the scattering vector q (modulus). Solid line corresponds to
best fit with Eq. (15); (b) Particle's radius numerical distribution function of sample FF-ND5.
References
[1] K.V.Zipare, S.S.Bandgar and G.S.Shahane,Journal of Rare Earths, 36 (1), Pages 86-94 (2018)
[2] R. Sharma et al., Journal of Alloys and Compounds 704, 7-17 (2017)
[3] S. Chakrabarty et al., Journal of Alloys and Compounds 625, 216 -- 223 (2015)
[4] G. Kumar et al., Materials Research Bulletin 63, 216 -- 225 (2015)
|
1910.03184 | 1 | 1910 | 2019-10-08T02:52:51 | Correlated fluctuations in spin orbit torque-coupled perpendicular nanomagnets | [
"physics.app-ph",
"cond-mat.mes-hall"
] | Low barrier nanomagnets have attracted a lot of research interest for their use as sources of high quality true random number generation. More recently, low barrier nanomagnets with tunable output have been shown to be a natural hardware platform for unconventional computing paradigms such as probabilistic spin logic. Efficient generation and tunability of high quality random bits is critical for these novel applications. However, current spintronic random number generators are based on superparamagnetic tunnel junctions (SMTJs) with tunability obtained through spin transfer torque (STT), which unavoidably leads to challenges in designing concatenated networks using these two terminal devices. The more recent development of utilizing spin orbit torque (SOT) allows for a three terminal device design, but can only tune in-plane magnetization freely, which is not very energy efficient due to the needs of overcoming a large demagnetization field. In this work, we experimentally demonstrate for the first time, a stochastic device with perpendicular magnetic anisotropy (PMA) that is completely tunable by SOT without the aid of any external magnetic field. Our measurements lead us to hypothesize that a tilted anisotropy might be responsible for the observed tunability. We carry out stochastic Landau-Lifshitz-Gilbert (sLLG) simulations to confirm our experimental observation. Finally, we build an electrically coupled network of two such stochastic nanomagnet based devices and demonstrate that finite correlation or anti-correlation can be established between their output fluctuations by a weak interconnection, despite having a large difference in their natural fluctuation time scale. Simulations based on a newly developed dynamical model for autonomous circuits composed of low barrier nanomagnets show close agreement with the experimental results. | physics.app-ph | physics | Correlated fluctuations in spin orbit torque-coupled
perpendicular nanomagnets
Punyashloka Debashis1, 2, Rafatul Faria2, Kerem Y. Camsari2, Supriyo Datta2, Zhihong Chen1, 2
1Birck Nanotechnology Center, 2Department of Electrical and Computer Engineering
Purdue University, West Lafayette, USA
Abstract - Low barrier nanomagnets have attracted a lot of research interest for their use as sources of high
quality true random number generation. More recently, low barrier nanomagnets with tunable output have been
shown to be a natural hardware platform for unconventional computing paradigms such as probabilistic spin logic.
Efficient generation and tunability of high quality random bits is critical for these novel applications. However,
current spintronic random number generators are based on superparamagnetic tunnel junctions (SMTJs) with
tunability obtained through spin transfer torque (STT), which unavoidably leads to challenges in designing
concatenated networks using these two terminal devices. The more recent development of utilizing spin orbit
torque (SOT) allows for a three terminal device design, but can only tune in-plane magnetization freely, which is
not very energy efficient due to the needs of overcoming a large demagnetization field. In this work, we
experimentally demonstrate for the first time, a stochastic device with perpendicular magnetic anisotropy (PMA)
that is completely tunable by SOT without the aid of any external magnetic field. Our measurements lead us to
hypothesize that a tilted anisotropy might be responsible for the observed tunability. We carry out stochastic
Landau-Lifshitz-Gilbert (sLLG) simulations to confirm our experimental observation. Finally, we build an electrically
coupled network of two such stochastic nanomagnet based devices and demonstrate that finite correlation or
anti-correlation can be established between their output fluctuations by a weak interconnection, despite having
a large difference in their natural fluctuation time scale. Simulations based on a newly developed dynamical model
for autonomous circuits composed of low barrier nanomagnets show close agreement with the experimental
results.
Introduction
When the energy barrier separating the two stable states ("UP" and "DOWN") of a nanomagnet is comparable or
smaller than the ambient thermal noise, its magnetization fluctuates randomly between the two states. One of
the simplest applications that harnesses this inherent stochasticity of a low barrier nanomagnet (LBNM) is true
random number generators (TRNG)1 -- 3. However, the full potential of LBNM based hardware can only be realized
when the probability of the LBNM magnetization being in the "UP" or "DN" state is tunable by an external input.
Such a hardware has been given the term 'p-bit', which stands for probabilistic bit4,5. Being essentially tunable
random number generators, p-bits have recently been shown as natural hardware accelerators for unconventional
computing tasks such as Ising computing6,7, Bayesian networks8,9, neural networks10,11 and invertible Boolean
logic4.
Several implementations of LBNM based TRNGs have been demonstrated in the last few years, while only a few
included the output tunability. One such device is based on an SMTJ with an in-plane LBNM as the free layer,
where the tunability of the output state is obtained through STT1. It is well known that the major reliability issue
in STT-MRAM is the result of the read and write operations sharing the same access path through the entire MTJ
stack. Furthermore, having a common read and write path does not allow for the isolation of the input and output
signals, and hence makes it difficult to concatenate these devices into a network. Therefore, a three-terminal
device with SOT based output tunability is much more desirable due to the separation of the write current path
from the read current path12. Such devices have been proposed for in-plane LBNMs13 -- 15. However, recent
simulation studies suggest that a dense array of in-plane LBNMs have significant magnetic dipolar interactions1.
Such interactions could lead to compromised randomness and unwanted correlations between SMTJs in a large
network. Moreover, SOT tunability of in-plane magnetization occurs through the so-called anti-damping
mechanism, which is energy inefficient since it must overcome a large demagnetization field16. Therefore, LBNMs
with perpendicular magnetic anisotropy (PMA) are ideal for high density, smaller pitch size arrays that are
essential for large network implementations. However, current material systems that exhibit SOT can only
generate spins with in-plane polarization, which is orthogonal to the magnetization of the low barrier PMA
magnet, hence, complete tunability of its output state is not possible through SOT alone17.
In this work, we demonstrate for the first time, an SOT tunable random number generator made of a PMA LBNM.
The SOT tunability is realized through a small tilt in the magnetic anisotropy axis, as is evidenced by our
experiments and supported by sLLG simulations. We then couple two such devices via electrical connections and
study the correlation in their output fluctuations. Our experiments show that a weak coupling strength, that is 10
times smaller than the critical current required for deterministic switching, is sufficient to establish correlations
between the outputs of the two devices. By changing the connection polarity, we show that the correlation can
be changed from positive to negative. Our studies also show that two LBNMs with different time scales of
fluctuation can get correlated efficiently. We perform simulations on this coupled 2 p-bit system using a dynamical
model of autonomous circuits with all the required parameters taken from experiments. The simulation results
show good matching with the experiments. This demonstration of a novel tunable TRNG and its behavior in an
electrically coupled network provides important insights towards realizing large p-bit networks for unconventional
computing tasks.
Designing low barrier nanomagnets with perpendicular magnetic anisotropy
The thermal stability factor of a nanomagnet is given by EB/kBT, where EB = KeffV/2 is the energetic barrier
separating the two stable magnetization states. Here, Keff is the effective anisotropy energy density and V is the
volume of the nanomagnet. When EB is comparable to the ambient thermal energy kBT, the magnetization
randomly fluctuates between the two stable states, thus realizing a "stochastic nanomagnet".
We first engineer Keff of our magnetic material (CoFeB) by varying the thickness (tCoFeB) of the deposited PMA films,
shown in Fig. 1 (a). The anisotropy of such a stack is given by18:
Keff = Ki/tCoFeB - MS
2/20 ……………………………………………………………………… (1)
2/20) [ref]. We
arising from the competition between the interface anisotropy (Ki) and the demagnetization (MS
then follow the method used by Hayashi et al.19 to characterize Keff of our stacks, as shown in Fig. 1 (b). For films
with out-of-plane anisotropy, the obtained anomalous Hall resistance (RAHE) in the presence of an in-plane
magnetic field is fitted with a second order curve to obtain HK (where HK is the effective anisotropy field, given
by Keff/MS). For films that have net in-plane anisotropy, HK is obtained through a linear fit of the RAHE vs. out-of-
plane field curve. We observe a clear decreasing trend of Keff×tCoFeB vs. tCoFeB for as deposited films as well as
samples annealed at 250 C for one hour, as shown in Fig. 1 (c). The annealed stack with tCoFeB=1.3 nm,
corresponding to the lowest Keff, is then chosen to fabricate the stochastic nanomagnet devices. The fabricated
devices consist of lithographically defined PMA nanomagnets with a diameter of 100 nm on top of tantalum (Ta)
Hall bars, as shown in Fig. 2 (a). The combination of low Keff through the thickness optimization and low volume
through the lithography patterned small diameter results in a small EB at room temperature. Consequently,
these uniquely designed stochastic nanomagnets fluctuate randomly between the "UP" and "DN" magnetic
states as depicted in the cartoon in the top right inset of Fig. 2 (a). This random fluctuation is electrically read
out through the anomalous Hall effect (AHE), giving the random telegraphic signal as the output, shown in Fig. 2
(b). The magnetization dwell time in the "UP" and the "DN" state forms a distribution that is well fitted by an
exponential envelope, which suggests that the fluctuation is a random Poisson process1. To further test the
quality of the randomness, we perform evaluations using the standard statistical test suite provided by the
National Institute of Standards and Technology (NIST)20. The generated bit stream by our device passed all 9
tests that were performed, showing cryptographic quality randomness (test results in the supplementary
information).
Figure 1: (a) PMA stack with varying CoFeB thickness. (b) Anomalous Hall resistance as a function of magnetic
field along the hard axis. For a PMA magnetic stack, the field is applied in the in-plane direction and the
measured data are fitted with a parabolic curve to extract the effective anisotropy field (HK). For an IMA
magnetic stack, the field in applied perpendicular to the plane and the resultant plot is fitted with a linear fit
to extract HK (c) Keff×tPMA as a function of CoFeB layer thickness before and after 250 C anneal.
Tunability of the random output through spin orbit torque
We demonstrate that the mean value of the random numbers can also be tuned by a DC current through the giant
spin Hall effect (GSHE) Ta Hall bar, as shown in Fig. 2. Fig. 2 (a) shows the measurement configuration, where a
constant DC current (IDC) through the GSHE underlayer is applied on top of a small AC read current (IC). As shown
in Fig. 2 (d), depending on the sign and magnitude of IDC, the magnetization direction favors the "UP" or "DN"
direction, resulting in the sigmoidal curve for the average. We call this device a probabilistic bit, i.e. a p-bit.
Representative signals at three different IDC are presented in the panels to the right of Fig. 2 (d). The effect of DC
current can also be seen by plotting the dwell time in "UP" and "DN" states for various IDC, as shown in Fig. 2 (e)
(measurement done at 250 K). This modification in the dwell time directly results in the tunability of the average
magnetization. It is worth mentioning here that the small read current can also affect the state of the output,
especially for a LBNM having a thermal barrier close to zero, and hence has to be carefully mitigated by design.
This read disturb issue however is negligible in our case, where the energy barrier of the LBNM is ~18 kBT.
Figure 2: (a) Schematic of the measurement configuration with the fabricated device using the optimized
tPMA giving the lowest HK. The magnetic island has a diameter of ~100 nm. Cartoon representing the energy
diagram of the perpendicular magnetization is shown in the top right inset. The two states, i.e, "UP" and
"DN" are separated by a small energy barrier EB, so that thermal energy is sufficient to randomly fluctuate
the magnetization between the two states. (b) Measured anomalous hall resistance for a fixed small read
current (IC) and no DC current (IDC). The random telegraphic signals arise from the random fluctuations of
the perpendicular magnetization between "UP" and "DN" states. (c) Histogram of the dwell time in "UP"
and "DN" states. Both histograms are well fitted by an exponential envelope, showing that the
magnetization flipping can be represented by a random Poisson process. The average dwell time (UP and
DN) are calculated from the exponential fit. (d) Measurement with a DC charge current through the GSHE
underlayer to obtain tunability. A sigmoidal curve is obtained for the average RAHE vs. IDC, showing tunability
for a PMA LBNM without any external magnetic field. Each point on this curve is obtained by averaging the
random telegraphic output, representative data sets shown in the three panels on the right. (e) The dwell
times in "UP" and "DN" state changes as a function of IDC, which leads to the sigmoidal curve for average
magnetization state.
Physics of tuning low barrier PMA magnet through in-plane spins
Since the polarization direction of the generated spins due to the charge current through the GSHE underlayer lies
in the X-Y plane, it is surprising to see a tunability of the perpendicular magnetization by the DC current. An obvious
hypothesis that we first considered was: the Oersted field generated by IDC points along the Z-axis at the edges of
the hall bar, and could potentially favor one magnetization state over the other, leading to the observed tunability.
This is illustrated in Fig. 3 (a). To test this hypothesis, we measure the magnetization response as a function of an
applied magnetic field along the Z-direction. As expected, the average magnetization shows a sigmoidal behavior,
Figure 3: (a) Possible explanation of the obtained tunability. A lithographic misalignment could lead to the
magnet island being situated towards one edge of the Ta electrode, where the perpendicular component
of the Oersted field due to the charge current could lead to the observed tunability. (b) Device output as a
function of external magnetic field in the Z-direction, in the presence of DC current through the GSHE
underlayer. Offset field (Boffset) due to IDC is obtained from the horizontal shift in the output curves. The plot
on the right shows Boffset vs. IDC, which clearly displays a saturating behavior. Also, the slope in the linear
region is more than an order of magnitude larger than that expected from the Oersted field. (c) Another
possible explanation of the obtained tunability. A tilted anisotropy in the nanomagnet leads to a non-zero
mx component of the magnetization that can be tuned by the spin current through the GHSE underlayer.
Due to the tilted anisotropy field, tuning mx by the in plane spin currents leads to tuning mz. (d) Measured
anomalous Hall signal as a function of magnetic field in X, Y and Z direction. From the X-Z plot, we can
deduce the tilt angle from the ratio of saturation signal. From the X-Y plot, we notice that it is easier to
saturate the magnetization in plane in the X direction compared to the Y direction, suggesting that the tilt
of magnetization lies in the X-Z plane. (e) sLLG simulations of the above device with an applied DC charge
current for various magnetization tilt angles. The charge current flows in the Y direction in the GSHE
underlayer, producing spins with polarization along X direction that are responsible for the observed
tunability.
as this external field favors the "UP" direction for positive field values and "DN" direction for negative field values.
We then repeat the same measurement in the presence of a constant IDC. Any constant Oersted field in the Z
direction produced by this IDC would result in a horizontal shift or offset of the sigmoidal response, by an amount
equal to the average magnetic field exerted on the magnet along Z-direction due to IDC. We measure this shift,
"Boffset" for various IDC and plot Boffset vs. IDC in the right graph of Fig. 3 (b). There are two observations from this
graph that contradict the hypothesis of the Oersted field induced tunability. Firstly, Boffset is not a linear function
of IDC, which is different from the Oersted field linearly following the current, B = IDC/2W. It can be clearly seen
that Boffset saturates for IDC> 10 A. Secondly, in the region where Boffset is linear with IDC, the slope, Boffset/IDC =
4×10-1 mT/A, is much larger than the expected value of /2W = 3×10-3 mT/A for the case of current induced
Oersted field.
We therefore hypothesize that a slightly tilted magnetic anisotropy direction is responsible for the observed
tunability, as illustrated in Fig. 3 (c). Essentially, if the magnetization tilt is in the X-Z plane, it can lead to X-polarized
spins in the SOT underlayer favoring one state over other. A +X directed spin will favor the (X, Z) quadrant for the
magnetization, resulting in a positive mZ, which will register as an "UP" in the AHE measurement (since RAHE is
proportional to mZ). On the other hand, a -- X directed spin will favor the (-X,-Z) quadrant, resulting in a negative
mZ or "DN" direction for the magnetization. This is similar to the engineered tilted anisotropy work by You et al.21
applied on stable, large barrier magnets. In our case, since the magnetic stack is designed to have a very low
perpendicular anisotropy, any small in-plane anisotropy developed during the film deposition process can lead to
a significant tilt angle that is otherwise undetectable in magnets with strong perpendicular anisotropy. To test out
this hypothesis, we carry out AHE measurements as a function of external magnetic fields along Z, Y and X
directions on another device made of the same stack. Firstly, we carry out RAHE vs. BZ measurements as shown in
Fig. 3 (d) left plot. It can be seen that the saturation value of RAHE is noticeably larger than the remanent value.
The tilt angle, can be estimated by the relation = cos-1(RAHE, remanent/RAHE, saturation) as is done by You et al.21. From
the measured data, is estimated to be around 25 degrees. Next, we carryout RAHE measurements in the present
of BX (black curve) and BY (blue curve), shown in Fig. 3 (d) right plot. In these measurements, the RAHE saturates to
a zero value for large applied fields (B > Bsat) since the perpendicular component of the magnetization vanishes as
the magnetization is progressively forced to lie in the X-Y plane. By comparing Bsat for the X directed field and Y
directed field, it is seen that the magnetization can be forced along the X direction more easily than the Y direction,
as Bsat, X < Bsat, Y. This suggests that the tilt direction is in the Z-X plane, toward the X-axis.
Stochastic LLG simulations of the magnetization dynamics with tilted anisotropy
The feasibility of SOT based output tunability of a p-bit made of a low barrier PMA magnet with tilted anisotropy
is confirmed by numerically solving the stochastic Landau-Lifshitz-Gilbert (sLLG) equation with a monodomain
macro-spin assumption:
(1 + 𝛼2) 𝑑𝑚
𝑑𝑡
= −𝛾𝑚 × 𝐻⃗⃗ − 𝛼𝛾𝑚 × 𝑚 × 𝐻⃗⃗ −
1
𝑞𝑁𝑠
𝑚 × 𝑚 × 𝐼𝑠⃗⃗ +
𝛼
𝑞𝑁𝑠
𝑚 × 𝐼𝑠⃗⃗ …………………..…………………(2)
where, 𝐻⃗⃗ is the total effective field including the anisotropy field 𝐻𝑘⃗⃗⃗⃗⃗ along a direction 𝜃0 tilted with respect to the
𝑍 axis on the 𝑋 − 𝑍 plane and the three dimensional uncorrelated thermal noise field 𝐻𝑛⃗⃗⃗⃗⃗ having Gaussian
2〉 = 2𝛼𝑘𝑇/𝛾𝑀𝑠𝑉 , 𝐼𝑠⃗⃗⃗ is the spin current
distribution with mean of 〈𝐻𝑛〉 = 0 and standard deviation of 〈𝐻𝑛
polarized along the 𝑋 direction, 𝑁𝑠 = 𝑀𝑠𝑉 is the total magnetic moment with 𝑀𝑠 being the saturation
magnetization and 𝑉 being the volume of the magnet, 𝛼 is the damping coefficient, 𝛾 is the gyromagnetic ratio.
Magnet parameters used in the simulation are: 𝐻𝑘 = 400 Oe, 𝑀𝑠 = 600 emu/cc, diameter 𝐷 = 36 𝑛𝑚 ,
thickness 𝑡 = 1.3 𝑛𝑚, 𝛼 = 0.1. The magnetization fluctuates in time owing to its low thermal barrier. The
average magnetization component in the Z direction (<mZ>) is plotted as a function of DC spin current in Fig. 3 (e).
When the magnet's anisotropy does not have any tilt with respect to the Z direction, then the in-plane spins do
not affect preferred direction of mZ. Hence, the average of mZ stays around zero. However, for tilt angles larger
than 10 degrees, complete tunability of mZ can be obtained, as can be seen in Fig. 3 (e).
Note that at high currents the sLLG simulation suggests that the z component of the magnetization can completely
vanish as the magnetization is pulled into the ±X direction. We believe that these currents are very large and
experimentally not accessible in our system.
Electrically coupled network of two p-bits
Figure 4: (a) The circuit to implement directed connection between two p-bits. (b) Auto correlation of the
outputs of the driver (bottom) and the follower (top) for different connection configurations. Follower p-bit is
much slower than driver for no connection case, but starts to respond faster when positive or negative
connection is established between the two p-bits. (c) Time traces of the two p-bits. With positive connection
established between them, positive correlation starts appearing, which is also seen from by plotting the
histograms of the four possible states in (d). The parallel configurations (UU) are more frequent. This is closely
matched by PPSL simulations. (e) The "relatedness" between the driver and follower signals is quantified by
the cross correlation, which shows a positive peak. The correlation coefficient given by the height of the peak
and the time scale of the correlation, given by the FWHM of the peak are both closely matched by PPSL
simulations. (f) (g) and (h) are for the case of negative correlation.
Experimental results
In this section, we study two electrically coupled p-bits. The stochastic output of the first p-bit ("driver") is
amplified and provided as the input to the GSHE underlayer of the next p-bit ("follower"). The amplification is
done by SRS 830 lockin amplifiers, with an averaging time of 3 ms, which is much smaller than any other time scale
in the experiment. The outputs of the lockin amplifiers are then fed to comparators in order to digitize the signals.
We vary the connection configuration between the two devices to observe their behaviors. Fig. 4 (a) illustrates
the circuit setup. We study three configurations: no connection (Rweight = infinity), positive connection (Rweight = 400
K and amplifier gain is positive) and negative connection (Rweight = 400 K and amplifier gain is negative). Fig. 4
(b) shows the autocorrelation of the output signals of the two devices (𝐴𝑑𝑟𝑖𝑣𝑒𝑟 and 𝐴𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟 ) for the three
configurations, obtained by the following formula:
𝑇−∆𝑡
𝑇−∆𝑡
𝐴𝑑𝑟𝑖𝑣𝑒𝑟(∆𝑡) = ∑ (𝑋𝑡 − 𝑋)(𝑋𝑡+∆𝑡 − 𝑋)
; 𝐴𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(∆𝑡) = ∑ (𝑌𝑡 − 𝑌)(𝑌𝑡+∆𝑡 − 𝑌)
𝑡=0
𝑡=0
In the above formulae, 𝑋𝑡 and 𝑌𝑡 are the state of the driver and follower respectively at time t; 𝑋 and 𝑌 are the
respective mean values; T is the total measurement time. When unconnected, the two devices fluctuate at
different time scales, evidenced by the markedly different full width at half maximum (FWHM) of their
autocorrelation peaks. However, when either a positive or negative connection is established between the two
devices, the fluctuation time scale of the follower becomes closer to that of the driver, seen again by the FWHMs.
Fig. 4 (c) and (f) show representative sections of the time traces of the output signals of the driver and the follower
for positive and negative connection respectively, where the emergence of positive and negative relation can be
observed. To quantify the relatedness between the two signals, the histogram of the four possible configurations
are plotted in Fig. 4 (d) and (g). It is observed that the driver signal has some inherent bias towards the UP state,
possibly due to the presence of an unwanted magnetic field in the measurement chamber. Therefore, to
accurately quantify the relatedness between the two outputs, we calculate the cross correlation between the two
signals (𝐶𝑑𝑟𝑖𝑣𝑒𝑟,𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟), obtained by introducing a relative time shift (t) between the two output signals and
calculating the inner product as a function of this shift according to the following formula:
𝐶𝑑𝑟𝑖𝑣𝑒𝑟,𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(∆𝑡) = ∑ (𝑌𝑡 − 𝑌)(𝑋𝑡+∆𝑡 − 𝑋)
𝑇−∆𝑡
𝑡=0
This metric is less prone to the inherent bias as the correlations are calculated from signals after subtracting their
respective mean values. Also, this metric preserves the time dependence of the relatedness. Any misleading
relatedness observed in the histogram due to inherent biases in the two signals would not have time dependence,
and hence would not contribute to the peak structure on the cross correlation plots.
We plot the normalized cross correlation and autocorrelations obtained from the following normalization:
𝐶𝑑𝑟𝑖𝑣𝑒𝑟,𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(∆𝑡) =
𝐶𝑑𝑟𝑖𝑣𝑒𝑟,𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(∆𝑡)
1
2⁄
(𝐴𝑑𝑟𝑖𝑣𝑒𝑟(0) × 𝐴𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(0))
𝐴𝑑𝑟𝑖𝑣𝑒𝑟(∆𝑡) =
𝐴𝑑𝑟𝑖𝑣𝑒𝑟(∆𝑡)
𝐴𝑑𝑟𝑖𝑣𝑒𝑟(0)
; 𝐴𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(∆𝑡) =
𝐴𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(∆𝑡)
𝐴𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟(0)
The black curves in Fig. 4 (e) and (h) show 𝐶𝑑𝑟𝑖𝑣𝑒𝑟,𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟. The correlation coefficient is just the peak value of
𝐶𝑑𝑟𝑖𝑣𝑒𝑟,𝑓𝑜𝑙𝑙𝑜𝑤𝑒𝑟 . From the above analysis we obtain the following insights for the different connection
configurations.
No Connection: For the no connection case (Rweight = infinity), the outputs of the two devices are essentially two
independent random bit streams. An important finding from this experiment is that the two unconnected p-bits
here have markedly different time scales of fluctuation, as is seen by full width at half maximum (FWHM) of the
auto correlation plots for driver and follower signals. The driver fluctuates at a faster time scale, with an FWHM =
24 ms, whereas that for the follower p-bit is much slower with a FWHM of 648 ms.
Positive/Negative connection: Next, we use Rweight = 400 K and choose the connection polarity to implement a
positive correlation. The choice of Rweight and VDD together result in a current of 25 A input to the second device,
which is smaller than the critical current required for deterministic switching of the magnetization direction.
Considering a magnet with an energy barrier of EB ≈15 kBT, spin Hall angle of Tantalum, Ta =0.07 and the Hall bar
width of W =200 nm, the critical switching current without thermal assistance can be calculated to be Icritical ~300
A from the formula given by Liu et al.22. Therefore, the current required here for establishing a correlation
between the two p-bits is more than 10 times smaller than Icritical. The effect of a positive connection can be seen
in the time traces, where the follower's output signal weakly follows that of the driver, while showing intermittent
random flips. The cross-correlation peaks around t = 0 and dies off with a FWHM =162 ms, suggesting that the
follower p-bit responds to the input provided by the driver in the time scale of the driver. It is also interesting to
see that the follower, which was much slower than the driver in the unconnected case, starts to respond with a
speed close to that of the driver for the positive connection case. This is quantified by the follower FWHM
decreasing to 100 ms, as shown in the figure inset. Similarly, for the negative connection case, a negative peak in
the cross correlation can be seen around t = 0. The speed of the follower becomes closer to that of the driver,
as quantified by the reduction in FWHM of the follower to 75 ms.
Autonomous PSL simulations
Unlike inherently synchronous digital platforms, the hardware proposed in this article is completely autonomous
without any sequencers to enforce any specific update order. To model this autonomous hardware, we have used
a simple behavioral model as described by Sutton et. al23 that is benchmarked against coupled stochastic Landau-
Lifshitz-Gilbert (sLLG) equation for capturing low barrier nanomagnet physics. In this model, each p-bit in the
network flips with a probability of 𝑝 controlled by the input 𝐼𝑖 described by the following equation:
𝑚𝑖(𝑡 + ∆𝑡) = 𝑚𝑖(𝑡) × 𝑠𝑔𝑛(𝑒𝑥𝑝(−𝑝𝑖) − 𝑟[0,1])
𝑝𝑖 =
∆𝑡
𝜏𝑐𝑜𝑟𝑟,𝑖
𝑒𝑥𝑝(−𝑚𝑖(𝑡)𝐼𝑖(𝑡))
where, 𝑚𝑖 is the output state of the i-th p-bit, ∆𝑡 is the simulation time step, 𝑟[0,1] is a random number between
0 and 1, 𝜏𝑐𝑜𝑟𝑟,𝑖 is the correlation time of the p-bit under zero input.
The interconnection of the p-bits are described by the following synapse equation:
𝐼𝑖(𝑡) = ∑ 𝐽𝑖𝑗
𝑚𝑗(𝑡) + ℎ𝑖
𝑗
Where, 𝐽𝑖𝑗 is the dimensionless coupling term obtained from the experimental parameters using the following
mapping:
Where, 𝐼𝐷𝐶,0 is the tanh fitting parameter for the sigmoidal response of the follower.
𝐽𝑖𝑗 =
𝑉𝐷𝐷
𝑅𝑤𝑒𝑖𝑔ℎ𝑡,𝑖𝑗𝐼𝐷𝐶,0
Experimentally obtained parameters used in the PPSL simulation are: 𝐼𝐷𝐶,0 = 15 𝜇𝐴, 𝑉𝐷𝐷 = 10 𝑉, 𝑅𝑤𝑒𝑖𝑔ℎ𝑡,𝑖𝑗 =
400 𝐾Ω, 𝜏𝑐𝑜𝑟𝑟,1 = 24 𝑚𝑠 𝜏𝑐𝑜𝑟𝑟,2 = 648 𝑚𝑠.
ℎ𝑗 is the bias term provided to the j-th p-bit. The fractional occupation of the driver p-bit in the "UP" state that is
obtained from the experimental histograms gives ℎ=0.63 for the driver. For the follower, obtaining ℎ is not
straight forward. However, for our experiment, ℎ=0 fits the measurement data nicely, suggesting that the follower
did not have any significant bias. The results of the simulations are plotted in red along with the corresponding
experimental results in Fig. 4 (d), (e), (g) and (h).
There are two findings from the above experiments that are of critical importance for large networks of
interconnected p-bits:
1. A weak electrical interconnection, which is more than 10 times smaller current than that required for
deterministic switching, is sufficient to induce correlations between two p-bits. Weak interconnection
strength is crucial for low power consumption in a large network. Moreover, as correlations are present
even with weak interconnection strengths, it allows for electrical annealing7, where the interconnection
strength can be gradually turned up to further enhance the desired correlations and suppress the
undesired ones.
2. A large difference in the natural time scales of the two devices does not hamper the operation of such
circuits. Another important factor for a large p-bit network is its robustness against device to device
variations. Specifically, the natural fluctuation time scales of the p-bits depend exponentially on the
energy barrier of the individual nanomagnet, which can have a distribution due to process variability.
Therefore, for a network of p-bits with different energy barriers to work as desired, correlations need to
be established even with different individual fluctuation time scales. This important requirement has been
verified in our experiments, where correlations were successfully established between the two p-bits
despite their natural time scales being very different (24 ms and 648 ms for driver and follower
respectively).
Conclusion
In summary, we have demonstrated for the first time, a stochastic nanomagnet with perpendicular anisotropy,
tunable by an in-plane spin current. We hypothesize the possibility of a tilted anisotropy being responsible for the
observed tunability, which is supported by both experiments and sLLG simulations. We further demonstrate a
coupled network of two such stochastic devices, namely p-bits, and show that correlations between their
stochastic outputs can be manipulated through weak electrical interconnections, despite having difference in their
natural fluctuation time scale.
Acknowledgement
The authors would like to thank Prof. Joerg Appenzeller and Prof. Pramey Upadhyaya for many fruitful discussions
and their feedback on this work. P.D. would like to thank Vaibhav Ostwal, Terry Hung and Tingting Shen for their
helpful feedback during the conduction of this work.
This work was supported by the Center for Probabilistic Spin Logic for Low-Energy Boolean and Non-Boolean
Computing (CAPSL), one of the Nanoelectronic Computing Research (nCORE) Centers as task 2759.003 and
2759.004, a Semiconductor Research Corporation (SRC) program sponsored by the NSF through CCF 1739635.
References
1 D. Vodenicarevic, N. Locatelli, A. Mizrahi, J.S. Friedman, A.F. Vincent, M. Romera, A. Fukushima, K.
Yakushiji, H. Kubota, S. Yuasa, S. Tiwari, J. Grollier, and D. Querlioz, Phys. Rev. Appl. 8, 054045 (2017).
2 B. Parks, M. Bapna, J. Igbokwe, H. Almasi, W. Wang, and S.A. Majetich, AIP Adv. 8, 4 (2018).
3 P. Debashis and Z. Chen, in 2018 76th Device Res. Conf. (IEEE, 2018), pp. 1 -- 2.
4 K.Y. Camsari, R. Faria, B.M. Sutton, and S. Datta, Phys. Rev. X 7, 031014 (2017).
5 W.A. Borders, A.Z. Pervaiz, S. Fukami, K.Y. Camsari, H. Ohno, and S. Datta, Nature 573, 390 (2019).
6 P. Debashis, R. Faria, K.Y. Camsari, J. Appenzeller, S. Datta, and Z. Chen, Tech. Dig. - Int. Electron
Devices Meet. IEDM 34.3.1 (2017).
7 B. Sutton, K.Y. Camsari, B. Behin-Aein, and S. Datta, Sci. Rep. 7, 44370 (2017).
8 R. Faria, K.Y. Camsari, and S. Datta, AIP Adv. 8, 045101 (2018).
9 O. Hassan, K.Y. Camsari, and S. Datta, IEEE Des. Test (2019).
10 V. Ostwal, P. Debashis, R. Faria, Z. Chen, and J. Appenzeller, Sci. Rep. 8, 16689 (2018).
11 R. Zand, K.Y. Camsari, I. Ahmed, S.D. Pyle, C.H. Kim, S. Datta, and R.F. DeMara, (2017).
12 K. Garello, F. Yasin, S. Couet, L. Souriau, J. Swerts, S. Rao, S. Van Beek, W. Kim, E. Liu, S. Kundu, D.
Tsvetanova, K. Croes, N. Jossart, E. Grimaldi, M. Baumgartner, D. Crotti, A. Fumemont, P. Gambardella,
and G.S. Kar, in 2018 IEEE Symp. VLSI Circuits (IEEE, 2018), pp. 81 -- 82.
13 R. Faria, K.Y. Camsari, and S. Datta, IEEE Magn. Lett. 8, 1 (2017).
14 V. Ostwal and J. Appenzeller, IEEE Magn. Lett. 10, 1 (2019).
15 P. Debashis, R. Faria, K.Y. Camsari, and Z. Chen, IEEE Magn. Lett. 9, 1 (2018).
16 L. Liu, C.F. Pai, Y. Li, H.W. Tseng, D.C. Ralph, and R.A. Buhrman, Science (80-. ). 336, 555 (2012).
17 P. Debashis and Z. Chen, Sci. Rep. 8, 11405 (2018).
18 S. Ikeda, K. Miura, H. Yamamoto, K. Mizunuma, H.D. Gan, M. Endo, S. Kanai, J. Hayakawa, F.
Matsukura, and H. Ohno, Nat. Mater. 9, 721 (2010).
19 M. Hayashi, J. Kim, M. Yamanouchi, and H. Ohno, Phys. Rev. B - Condens. Matter Mater. Phys. 89, 1
(2014).
20 J. Soto, Proc. 22nd Natl. Inf. Syst. Secur. Conf. 10, 12 (1999).
21 L. You, O. Lee, D. Bhowmik, D. Labanowski, J. Hong, J. Bokor, and S. Salahuddin, Proc. Natl. Acad. Sci.
112, 10310 (2015).
22 L. Liu, O.J. Lee, T.J. Gudmundsen, D.C. Ralph, and R.A. Buhrman, Phys. Rev. Lett. 109, 096602 (2012).
23 B. Sutton, R. Faria, L.A. Ghantasala, K.Y. Camsari, and S. Datta, (2019).
Supplementary Information
This supplementary text contains two sections to characterize our p-bit as a random number generator:
(i)
(ii)
Dependence of the speed of random bit generation on the thermal stability factor of the
nanomagnet
Evaluation of the quality of the generated random bits according to the statistical test suite provided
by the National Institute of Standards and Technology (NIST)
Characterizing the device as a high quality random number generator
The telegraphic output obtained from our devices are digitized to obtain a random bit stream. This bit stream can
be used as a source of high quality random numbers, as required by many applications. We study speed of random
bit generation by our devices and provide strategies to improve it. Then we evaluate the quality of the generated
random bit stream using standard test protocols according to the statistical test suite (STS)SR1 provided by the
National Institute for Standards and Technology (NIST).
Speed of random bit generation
The obtained random bit stream is used to generate the histogram plot of the dwell time in the two metastable
states as shown in Fig. 2 (c). An exponential envelope fits well with the experimentally obtained dwell time
histogram, suggesting that the random fluctuations of nanomagnets can be well described by a Poisson process.
Average dwell time in the "UP" and "DN" states, i.e. UP and DN are obtained from the exponential fitting. For a
completely symmetrical energy landscape, these two dwell times should be identical. However, we obtain a
slightly skewed distribution due to the remanent magnetic field in the measurement setup. The speed of the
-1). This time scale
random bit generation can be obtained from the harmonic mean of UP and DN (− = UP
is determined by the energy barrier of the nanomagnet through:
-1 + DN
= ×exp(EB/kBT) …………………………………………………………………………………………………………………………………………… (2)
Hence, the speed of the random number generation can be increased by reducing the thermal stability factor,
EB/kBT. From the obtained , we determine EB/kBT ≈ 18 for the stochastic device presented here. Fig. S1 shows the
experimentally measured effect of changing this stability factor by changing temperature. The fluctuation time
scale, changes by more than 3 orders of magnitude. This stability factor can be reduced in practice by scaling
the volume of the nanomagnet further to increase the speed of random number generation.
Fig. S1: Speed of random number generation can be changed by changing the thermal stability factor
EB/kBT. To show this, we measure the telegraphic output at different temperatures. This plot shows the
Arrhenius plot of the fluctuation time scale, of the magnet as a function of temperature. The time scale
can be changed by more than 3 orders of magnitude by changing the temperature by 50 K.
Quality of the generated random bit stream
We evaluate the quality of the random bits generated from our device using the STS provided by NIST. Various
aspects of the generated random sequence are tested for qualifying as cryptographic quality random numbers.
These tests compute the statistics of the input sequence such as mean value, standard deviation, entropy,
repeated structures, linear dependencies, autocorrelation, etc. and compare them against theoretical
expectations from a perfectly random sequence. Details of the tests can be found in the manual of the NIST STS
package. Prior to performing the tests, we do the following post-processing of the collected data in order to
reduce any present bias. After digitizing the data, we take the XOR of the bit stream with a copy of itself that is
shifted by one time step. This is shown in Fig. S2. The resulting p-values of the NIST test are given in table 1. A p-
value >0.01 means the test has been passed. The generated bit stream by our device passed all 9 tests that were
performed, showing cryptographic quality randomness. Some of the tests required > 106 bits and were not
performed due to the limited size of our data set (10,000 bits).
Fig. S2: Post processing of the output bit stream prior to performing NIST randomness test. The obtained
random bits are XORed with a time shifted version of itself to remove any bias.
out(cid:9)shi(cid:9)by(cid:9)(cid:9)1(cid:9)me(cid:9)step(cid:9)random(cid:9)bits(cid:9)from(cid:9)device(cid:9)(used(cid:9)for(cid:9)NIST(cid:9)test)(cid:9)
1
2
3
4
5
6
7
8
9
Test Name
Description
p-value
Frequency
Total number of 0's and 1's mismatch
0.3764
Block frequency
Number of 0's and 1's mismatch within a subset
0.6526
Runs
Longest run
Too many sequences of consecutive bits of one type
0.9693
Too many consecutive bits of one type
0.6600
Binary matrix rank
Linear dependence among fixed length substrings
0.4613
Discrete Fourier transform
Periodicity in bit stream
0.9884
Serial
Multiple low entropy sequences in a row
0.0344, 0.5958
Approximate Entropy
Bit sequence too unlikely
0.0601
Cumulative Sums (forward)
Running sum deviates too far from half the length
0.2143
10 Cumulative Sums (backward)
Same as previous, but in reverse direction
0.6776
Table 1: Results of the NIST STS tests on the pre-processed random bit stream. P-value>0.01 means that the
test has been passed according to the 1% confidence level.
References:
SR1: Soto, Juan. "Statistical testing of random number generators." Proceedings of the 22nd National
Information Systems Security Conference. Vol. 10. No. 99. Gaithersburg, MD: NIST, 1999.
|
1809.10558 | 1 | 1809 | 2018-09-26T16:15:36 | Left-handed Band in an Electromagnetic Metamaterial Inducedby Sub-wavelength Multiple Scattering | [
"physics.app-ph",
"physics.class-ph"
] | Due to the deep sub-wavelength unit cell in metamaterials, the quasi-static approximation is usually employed to describe the propagation. By making pairs of resonators, we highlight that multiple scattering also occurs at this scale and results in the existence of a dipolar resonance, which leads to a negative index of refraction when we consider several resonators. We experimentally verify the possibility of obtaining a negative index of refraction in periodic metamaterials in two different ways, and eventually demonstrates a superlensing effect in both cases. | physics.app-ph | physics | Left-handed Band in an Electromagnetic Metamaterial Induced
by Sub-wavelength Multiple Scattering
Simon Yves,1 Thomas Berthelot,2, 3 Mathias Fink,1 Geoffroy Lerosey,4 and Fabrice Lemoult1
1Institut Langevin, ESPCI Paris, PSL Research University,
CNRS UMR 7587, 1 rue Jussieu, 75005 Paris, France
2CEA Saclay, IRAMIS, NIMBE, LICSEN,
UMR 3685, F-91191, Gif sur Yvette, France
3KELENN Technology, Antony, France
4Greenerwave, ESPCI Paris Incubator PCup,
6 rue Jean Calvin, 75005 Paris, France
(Dated: July 10, 2019)
Abstract
Due to the deep sub-wavelength unit cell in metamaterials, the quasi-static approximation is
usually employed to describe the propagation. By making pairs of resonators, we highlight that
multiple scattering also occurs at this scale and results in the existence of a dipolar resonance, which
leads to a negative index of refraction when we consider several resonators. We experimentally
verify the possibility of obtaining a negative index of refraction in periodic metamaterials in two
different ways, and eventually demonstrates a superlensing effect in both cases.
8
1
0
2
p
e
S
6
2
]
h
p
-
p
p
a
.
s
c
i
s
y
h
p
[
1
v
8
5
5
0
1
.
9
0
8
1
:
v
i
X
r
a
1
The wave-matter interaction is the basis for many applications relying on the wave-like
nature of the system.
Indeed, from the waves scattered by an object we can extract a
specific property: for example the X-ray diffraction off the Bragg planes of a crystal allows
to recover its crystalline structure1, the infra-red spectroscopy gives information about the
chemical composition of a flask through molecular resonance excitations2, or even magnetic
resonance imaging builds medical images from mapping relaxation times of atoms3. All
of these understandings of the wave-matter interaction have recently lead to the advent
of a new class of artifical media: the metamaterials. In this case, the strategy is slightly
different; we engineer artificial objects that interact with waves in order to induce a new
macroscopic property4 -- 6. The motor of many advances in the field has been the quest for
left-handed media exhibiting negative refraction7,8, because of the inital proposal that a slab
made of such a medium should behave as a perfect lens9. Nevertheless, this does not limit
to this example, and many other ideas emerged as for example proposals of cloaking10 -- 12,
high-effective index13, or epsilon-near-zero media14,15.
The double negativity of a system has often been obtained by mixing two different kinds of
media, one bringing the negative effective permittivity16, and the other one being responsible
of the negative permeability17. This approach neglects all of the possible interactions at
the unit cell level between the two kinds of media. Recently, in acoustics, it has been
demonstrated that a single negative metamaterial based on a resonant unit cell can be
turned into a double negative one if the resonators are grouped by pairs18,19. Typically,
multiple scattering which occurs between the two-resonators creates a dipolar mode that
spectrally overlaps with the initial monopolar one resulting in a double negative property.
In a periodic medium, a very convenient way to build physically a bigger super-cell that
contains two resonators consists in either detuning one resonator out of two, thus ending on
a bi-disperse medium, or by slightly moving the position of this resonator and ending on a
bi-periodic medium. In this article, We experimentally explore these two different strategies
in the microwave domain.
In our study, the resonant unit cell consists in a simple quarter-wavelength metallic rod
on a ground plane. Seen from the top each of these resonators is very small compared to
the wavelength and their resonant nature comes from their length. Here, we choose a rod
diameter of 1 mm and a height of h0 =16 mm, therefore resonating around 4.68 GHz. As
a consequence, we can pack many of such elementary bricks within a wavelength and we
2
FIG. 1.
(a) The bi-periodic metamaterial consisting on a honey-comb lattice of identical metallic
wires on a ground plane, and (b) the corresponding experimental sample. (c) The bi-disperse one
based on a square lattice containing two wires of slightly different lengths, and (d) its experimen-
tal realization. (e) and (f) the numerical dispersion relations along the principal directions for
respectively the bi-periodic and the bi-disperse metamaterials (in red the negative branch).
therefore directly fall within the context of locally resonant metamaterials. It is now well
known that the finite-length wire medium can be described with an effective permittivity
which presents a Lorentzian profile near each resonance20. Basically with a metamaterial
approach we would assume that all the resonators see the same incident electric field and
their overall response is averaged in order to give a macroscopic effective property that only
depends on the number of resonators per wavelength. With this description, the spatial
organization of the wires does not influence the macroscopic behavior, but only their density
does.
Now, we want to induce a dipolar resonance by building a metamaterial that is still
based on pairs of the previous resonant unit cell. We can do it in the two different afore-
mentioned ways: the bi-periodic medium and the bi-dispersed one.
In two dimensions,
the most isotropic bi-periodic medium consists in a honey-comb arrangement of resonators
(Fig.1.a), and our choice of bi-dispersed sample consists of a square lattice with two detuned
resonators per unit cell (1.c, the shorter wires have now a height h1 =15.7 mm). For both
cases, we build a sample by metallizing a 3D-printed ABS-based plastic medium with the
same procedure as in the ref.21. In each case, nearest neighboring rods are separated by a
3
33.54(cid:31).(cid:30)XΓM344.53.5MΓKFrequency (GHz)Frequency (GHz)MƱKXƱMλ0λ0λ0λ0f0h0f0h0f1h1(a)(c)(e)(f)(b)(d) distance of 10 mm. The two fabricated samples are shown in figure 1.b and d and have the
shape of a rectangle paving an area of 160 mm x 60 mm.
For each lattice, the dispersion relation is computed numerically using Comsol Multi-
physics along the main directions of the crystals by simulating a single unit cell with Bloch
periodic boundary conditions (Fig. 1.e and f). As expected, both of them clearly reveal the
existence of a band with a negative gradient within the first Brillouin zone (red line). The
negativeness of this slope is a direct consequence of the existence of a dipolar resonance
overlapping with a monopolar one, ie. the two eigenmodes of the unit cell based on two
resonators. This first numerical result confirms that, albeit the deep sub-wavelength step
between the two resonators, a dipolar mode that exhibits a change of sign between the
them exists: the usual description of such a metamaterial that solely takes into account a
monopolar one is obviously not sufficient.
In order to verify that the presence of this band traduces a negative refraction phe-
nomenon, we perform a point source excitation at one interface of the slab samples. In both
cases, we measure the transmission, using a vector network analyser, between this source
and a near-field probe placed in the middle of the sample. This probe is mounted on a
motorized 2D-stage and will be used later to map the field on top of the slab. Typical
frequency spectra measured for both samples are represented in figure 2.a and b.
Both of them reveal the presence of resonance peaks, spanning frequencies from 3.5 GHz
to 4.3 GHz, falling within the previously calculated propagating bands. The linewidth of
the resonances seems to narrow as approaching the intrinsic resonance f0 of the wires as
in any finite-sized locally resonant metamaterial22 -- 24, but we cannot yet see if the spatial
organization of the rods has created any specific effect.
A first step to evidence the negative refraction comes from our two dimensional scan.
Indeed, we display in figure 2.c-f the intensity maps corresponding to different frequencies.
They can be classified in two very distinct groups. In the lower part of the resonance peaks
(c and e) the intensity is distributed over the entire slab, while for the peaks near the band
gap (d and f) the intensity seems to be confined along the horizontal axis facing the source
position. A ray tracing with a negative refraction occurring at each interface is tempting to
explain this effect, but canalization25 or anisotropic efficiency of the bandgap could also cause
such an intensity distribution. In any case, the intensity maps lack the phase information
that is required to clearly identify if those metamaterials could be described with a negative
4
FIG. 2. Point source excitation. (a) (respectively (b)) A plot of a typical measured spectrum
on top of wire within the metamaterial when excited by a monopole antenna near one interface for
the bi-periodic (respectively bi-disperse) sample. (c-f) Intensity maps on top of the medium for
one frequency within the first band (c and e) and one frequency in the second one (d and f) for
both samples.
index of refraction.
A convenient representation to extract the effective properties of a medium at a single
frequency is the reciprocal space. We therefore carry out two dimensional Fourier transforms
of the field maps and show the intensity distribution as a function of the wavevector k. Due
to the slab geometry this leads to anisotropic intensity distribution. We therefore perform
some symmetry operations on the resulting map. For the case of the honey-comb lattice
they consist in applying a C6 rotational symmetry and 2 mirrors (vertical and horizontal).
For the square lattice made of two different resonators these operations are a C4 rotational
symmetry combined with the two mirrors.
In both cases we also take into account the
translational symmetry with respect to the first Brillouin zone. Once these operations are
done on the maps initially represented in intensity in the real space (figure 2.c-f), we end
up on the intensity maps represented in the reciprocal space of figure 3.a-f. For convenience
we superimpose in white the first Brillouin zone as well as its periodic replica. We also
superimpose a line that corresponds to an automatic detection of the intensity maximum in
all propagating directions. For example, in figure 3.a it results in a rounded contour which
still shows a C6 symmetry but can fairly be approximated by a circle. The radius of this
5
014.44.243.83.63.43.2Frequency (GHz)Normalized Amplitudec(a)(b)01Normalized Amplitude4.44.243.83.63.43.2Frequency (GHz)e050100150y (mm)80400-40-80050100150x (mm)y (mm)050100150y (mm)60300-30-60x (mm)050100150y (mm)-600Electric Intensity (dB)(c)(d)(e)(f)df circle is bigger than the free space circle (white dotted line) meaning that the dispersion
relation at this frequency can be approximated by an effective index of refraction higher
than unity in norm. For figure 3.b the isofrequency contour looks more like an hexagon
with rounded edges than a circle meaning that some anisotropy effects exist, but again this
contour is outside of the freespace one (dashed line). Similar observations can be made for
the sub-panels c and d except that in this case the reminiscence of the C4 symmetry of
the crystal persists. At any rate, even if modeling the propagation in the metamaterial in
terms of a unique scalar, ie. the effective index refraction, is probably too restrictive due to
the anisotropy, this figure does not permit to conclude regarding the sign of this index of
refraction.
We applied the same procedure for several frequencies and plot all of the obtained isofre-
quency contours. We separate the results in two different graphs for each type of crystal
since they can be associated to the two desired bands with very different properties. For the
honey-comb lattice case, the lowest frequency (blue line in figure 3.e) corresponds to a circle
slightly bigger than the freespace one. Increasing the frequency (meaning changing the color
from blue to violet) we notice a rapid expansion of the circle's radius as expected from the
usual picture of the polariton. Another frequency increment leads to an anistropic behavior
and the propagation becomes forbidden in the ΓM directions while authorized in the ΓK
directions. This is an unambiguous evidence of the crystalline nature of the propagation
within such a metamaterial, albeit its deep sub-wavelength spatial scale. Moreover, even if
this is out of the scope of this article, our measurements reveal the existence of a Dirac cone
at the K-point26, which is far away from the free space light cone. Going to even higher
frequencies, we move to the second propagation band and to the sub-panel (f) of the figure.
This corresponds to the desired effect: after few frequencies still exhibiting some anisotropic
effects the isofrequency contours are circles centered on Γ with a radius that decreases with
frequency. The circle-shape indicates that the propagation can be well described with a sin-
gle scalar (the index of refraction) while the decrease in the radius with frequency indicates
the negativeness of the latter. Very similar conclusions can be made on the isofrequency
contours extracted from the measurement on the square lattice of rods (panel g and h) ex-
cept that the anisotropy is more pronounced since the shape of the contours is more similar
to rounded squares.
Let us now move to the last experiment which results from these observations. Since
6
FIG. 3. Experimental isofrequency contours. (a-d) Representation in the reciprocal space
of the field maps of Fig. 2 (see text for details). Superimposed in white are the first Brillouin zone
and its periodic replica (an hexagon for the bi-periodic crystal and a square for the bi-disperse
one), and in color the extracted isofrequency contour for this frequency. (e-g) The isofrequency
contours obtained when processing all the measured maps. (f) and (h) corresponds to the two
desired negative branches.
the seminal work of J. Pendry9, the quest for the negative index of refraction has been
motivated by the superlensing effect. Namely, he stated that a flat layer of a negative index
material not only works as a lens but also allows to beat the diffraction limit. This is
typically what we have tried to perform for both our slabs similarly to underwater acoustic
experiments27,28. The experiment consists now in emitting a sub-wavelength-scaled pattern
on one face of the slab and we map the field on the other interface. Typically, four point
sources linearly placed with a separation distance around λ0/3 emit simultaneously, at the
operating frequency of 4.2 GHz for the honey-comb sample, and of 4.27 GHz for the square
bi-disperse one. For the two experiments this corresponds to a frequency for which we have
obtained a circle-like isofrequency contour. The real part of the measured electric field at the
output interface (Fig. 4 b and d) strongly demonstrates that the flat lens allows to image our
initial object (black circle). Even the nearest neighbors separated by a distance of λ0/3 are
resolved. This effect here is actually a consequence of the high absolute value of the index
of refraction within the slabs as in solid immersion lenses29 for example. The superlensing
effect in those two examples therefore comes from the conversion of the evanescent waves
at the input into propagating waves within the slab, and they convert again to evanescent
ones at the output thus recovering small details of the initial source. There is actually no
7
(a)(d)(c)(b)-0.200.2.20.20.2-0.200.20.200.20.2(e)(f)Frequency (GHz)3.64.2Frequency (GHz)3.84.3(g)(h)kxkxkxkxkxkxkxkxkykykyky FIG. 4. Sub-wavelength imaging. For both crystals the source is now a simultaneous emission
of 4 point-antennas separated by a distance of λ0/3. We extracted the field oat the output interface
shown in white dashed-line on the measured intensity maps for the bi-periodic (a) and the bi-
disperse (c) samples. In both cases the reconstructed object (b) and (d) exhibits superlensing since
the initial profile (circles) is reconstructed below the diffraction limit.
evanescent amplification of the field within the metamaterial despite the description in terms
of a negative index.
As a summary, in this letter we have experimentally validated in microwaves the new
strategy proposed in18 to build a negative index metamaterial starting from a medium with
solely one negative effective property (here the permittivity). This is simply based upon the
idea that a bigger unit cell made of a pair of resonators should support a dipolar resonance.
Building such a meta-molecule is performed either by creating a bi-periodic medium (by
moving the position of one resonator out of two) or a bi-periodic one (by detuning the
same resonator), thus resulting in a periodic metamaterial whose unit cell contains two
resonators. This experimental study unambiguously demonstrates that in both cases the
multiple scattering induced dipolar resonance results in negative refraction. Thanks to a
complete mapping of the field within our samples we experimentally extracted the full band
diagram and strongly confirmed that a description in terms of a negative effective index of
refraction is relevant for a given frequency range. Eventually, we observed a superlensing
effect in both our samples for frequencies where the index of refraction is negative but with
8
(b)50100150-0.6-0.300.30.60y (mm)x (λ0)-0.6-0.300.30.6501001500y (mm)x (λ0)(c)(d)(a)-1-0.50.501Normalized Amplitude-1-0.50.501Normalized Amplitude an absolute value higher than one.
All of these results clearly show that the spatial organization of the resonators within a
metamaterial plays a crucial role, albeit the small separation distance between them with
respect to the free space wavelength. We cannot neglect anymore the correlation in positions
between the resonators19 and this effect is notoriously intensified in periodic media. There is
no doubt that more and more complex unit cells30 lead to new exotic macroscopic effective
properties in the here-opened context of crystalline metamaterials.
1 B. E. Warren, X-ray Diffraction (Courier Corporation, 1969).
2 B. Stuart, Infrared spectroscopy (Wiley Online Library, 2005).
3 P. C. Lauterbur et al., Nature 242, 190 (1973).
4 N. Engheta and R. W. Ziolkowski, Metamaterials: physics and engineering explorations (John
Wiley & Sons, 2006).
5 W. Cai and V. M. Shalaev, Optical metamaterials, Vol. 10 (Springer, 2010).
6 F. Capolino, Theory and phenomena of metamaterials (CRC press, 2009).
7 V. G. Veselago, Soviet physics uspekhi 10, 509 (1968).
8 D. R. Smith, J. B. Pendry, and M. C. K. Wiltshire, Science 305, 788 (2004).
9 J. B. Pendry, Phys. Rev. Lett. 85, 3966 (2000).
10 J. B. Pendry, D. Schurig, and D. R. Smith, Science 312, 1780 (2006).
11 U. Leonhardt, Science 312, 1777 (2006).
12 D. Schurig, J. J. Mock, B. J. Justice, S. A. Cummer, J. B. Pendry, A. F. Starr, and D. R.
Smith, Science 314, 977 (2006).
13 M. Choi, S. H. Lee, Y. Kim, S. B. Kang, J. Shin, M. H. Kwak, K.-Y. Kang, Y.-H. Lee, N. Park,
and B. Min, Nature 470, 369 (2011).
14 N. Engheta, Science 340, 286 (2013).
15 A. Alu, M. G. Silveirinha, A. Salandrino, and N. Engheta, Physical review B 75, 155410 (2007).
16 J. B. Pendry, A. J. Holden, W. J. Stewart, and I. Youngs, Phys. Rev. Lett. 76, 4773 (1996).
17 J. B. Pendry, A. J. Holden, D. J. Robbins, and W. Stewart, IEEE transactions on microwave
theory and techniques 47, 2075 (1999).
18 N. Kaina, F. Lemoult, M. Fink, and G. Lerosey, Nature 525, 77 (2015).
9
19 M. Lanoy, J. H. Page, G. Lerosey, F. Lemoult, A. Tourin, and V. Leroy, Phys. Rev. B 96,
220201 (2017).
20 C. R. Simovski, P. A. Belov, A. V. Atrashchenko, and Y. S. Kivshar, Advanced Materials 24,
4229 (2012).
21 N. Kaina, A. Causier, Y. Bourlier, M. Fink, T. Berthelot, and G. Lerosey, Scientific reports 7,
15105 (2017).
22 F. Lemoult, G. Lerosey, J. de Rosny, and M. Fink, Phys. Rev. Lett. 104, 203901 (2010).
23 F. Lemoult, N. Kaina, M. Fink, and G. Lerosey, Nature Physics 9, 55 (2013).
24 F. Lemoult, M. Fink, and G. Lerosey, Waves in Random and Complex Media 21, 591 (2011).
25 P. A. Belov, C. R. Simovski, and P. Ikonen, Phys. Rev. B 71, 193105 (2005).
26 A. K. Geim and K. S. Novoselov, in Nanoscience and Technology: A Collection of Reviews from
Nature Journals (World Scientific, 2010) pp. 11 -- 19.
27 J.-F. Robillard, J. Bucay, P. Deymier, A. Shelke, K. Muralidharan, B. Merheb, J. Vasseur,
A. Sukhovich, and J. Page, Physical Review B 83, 224301 (2011).
28 A. Sukhovich, B. Merheb, K. Muralidharan, J. Vasseur, Y. Pennec, P. Deymier, and J. Page,
Physical review letters 102, 154301 (2009).
29 S. M. Mansfield and G. S. Kino, Applied Physics Letters 57, 2615 (1990).
30 S. Yves, R. Fleury, T. Berthelot, M. Fink, F. Lemoult, and G. Lerosey, Nature communications
8, 16023 (2017).
10
|
1806.06661 | 2 | 1806 | 2018-07-02T09:16:03 | Performance analysis of heat and energy recovery ventilators using exergy analysis and nonequilibrium thermodynamics | [
"physics.app-ph"
] | The increased attention to energy savings has contributed to more widespread use of energy recovery systems for building ventilation. We investigate the efficiency of such systems under different outdoor conditions using exergy analysis and nonequilibrium thermodynamics. This analysis makes it possible to assess performance in terms of loss of work potential, to account for the different quality of energy and to localize and compare the different sources of loss in the system. It also enables the use of exergy efficiency as a single performance parameter, in contrast to the several indicators that are commonly used. These more common indicators are difficult to compare and relate to each other. Further, since there is no obvious optimal trade-off between them, it is challenging to combine them and develop a global performance indicator that allows for a sensible comparison of different technical solutions and different types of recovery devices. We illustrate the concepts by applying the analysis to a heat recovery ventilator (HRV) and to a structurally similar membrane energy recovery ventilator (MERV) that can exchange both heat and moisture. We show how the exergy efficiency can be used to identify the range of operating conditions for which the recovery ventilator is not beneficial as the energy cost is greater than the energy recovery. This is not trivial using traditional performance parameters, yet it is a natural outcome of exergy analysis. In addition, we identify the mechanism by which work potential is lost, which can help the eventual optimization of both the recovery process and the auxiliary systems present in ventilation systems. | physics.app-ph | physics |
Performance analysis of heat and energy recovery
ventilators using exergy analysis and nonequilibrium
thermodynamics
Magnus Aa. Gjennestada, Eskil Aursandb, Elisa Magnanellic,∗, Jon Pharoahd
aDepartment of Physics, NTNU - Norwegian University of Science and Technology
cDepartment of Chemistry, NTNU - Norwegian University of Science and Technology
NO-7491 Trondheim, Norway
dDepartment of Mechanical and Materials Engineering, Queen's University, K7L 3N6 Kingston, Canada
bDepartment of Energy and Process Engineering, NTNU - Norwegian University of Science and Technology
NO-7491 Trondheim, Norway
NO-7491 Trondheim, Norway
Abstract
The increased attention to energy savings has contributed to more widespread use of
energy recovery systems for building ventilation. We investigate the efficiency of such
systems under different outdoor conditions using exergy analysis and nonequilibrium
thermodynamics. This analysis makes it possible to assess performance in terms of
loss of work potential, to account for the different quality of energy and to localize and
compare the different sources of loss in the system. It also enables the use of exergy
efficiency as a single performance parameter, in contrast to the several indicators that
are commonly used. These more common indicators are difficult to compare and relate
to each other. Further, since there is no obvious optimal trade-off between them, it is
challenging to combine them and develop a global performance indicator that allows
for a sensible comparison of different technical solutions and different types of recovery
devices. We illustrate the concepts by applying the analysis to a heat recovery ventilator
(HRV) and to a structurally similar membrane energy recovery ventilator (MERV) that
can exchange both heat and moisture. We show how the exergy efficiency can be used
to identify the range of operating conditions for which the recovery ventilator is not
beneficial as the energy cost is greater than the energy recovery. This is not trivial using
traditional performance parameters, yet it is a natural outcome of exergy analysis. In
addition, we identify the mechanism by which work potential is lost, which can help the
eventual optimization of both the recovery process and the auxiliary systems present in
ventilation systems.
Keywords: exergy efficiency, energy efficiency, recovery ventilator, ventilation
systems, nonequilibrium thermodynamics
doi: 10.1016/j.enbuild.2018.04.013
© 2018. This manuscript version is made available under the CC-BY-NC-ND 4.0
license.
Preprint submitted to Energy and Buildings
July 3, 2018
1. Introduction
The increased attention to energy savings has led to buildings becoming more heav-
ily insulated [1, 2]. In tightly enveloped buildings, however, air quality becomes an
issue as pollutants may rapidly build up to dangerous levels [3–5]. In order to ensure
good air quality, it has therefore become important to install ventilation systems that
exchange indoor air with fresh outdoor air [6, 7].
Temperature and humidity are parameters that play a vital role in the indoor en-
vironment and they should be maintained within certain limits to ensure comfort and
health [8]. This can be a challenge in both cold climates, where outdoor air coming
in through the ventilation system can be much colder and drier than is optimal, and in
warm climates where the outdoor air may be too moist and warm. Auxiliary heating,
air-conditioning, humidification and dehumidification systems are then necessary to
compensate for differences between the conditions of the incoming air and the desired
indoor conditions.
In recent years, higher demand for indoor comfort, together with an increase in
time spent indoors, has caused the energy use of buildings to rise to the same level as
the transport and industry sectors [9–11]. This increase has resulted in the installation
of heat recovery systems, which are nowadays often required by regulations and codes,
in order to reduce building energy consumption [12–14].
Heat and energy recovery ventilators (HRV/ERV) are devices that transfer heat, or
both heat and moisture, from one air stream and deliver it to another [15]. Numerous
types of recovery ventilators are currently available, having different characteristics
and performance [16].
A widespread solution for heat recovery is the flat plate heat exchanger, which
is often used due to low capital costs and ease of operation [16]. In this device, the
exhaust and fresh air streams exchange heat across the wall that separates them. In
order to obtain large heat exchange surfaces, many parallel channels are usually stacked
on top of each other [17]. A major limitation of this technology is that moisture cannot
be transferred between the two air streams. For this reason, Zhang and Jiang [18]
proposed an alternative solution to conventional flat plate heat exchangers that makes
use of a water-permeable membrane as separating wall between the two air streams. In
this way, the exhaust and the fresh air can exchange moisture in addition to heat.
The performance of a HRV/ERV is influenced by several factors. One is the ability
to transfer heat from one air stream to the other. As long as the air moisture stays in
vapor phase, only sensible heat is exchanged between the streams. However, if the
cooled stream reaches the saturation temperature then water condenses [19] and the
latent heat of condensation is released and is available to be transferred to the heated
stream. Another factor is moisture recovery, which can be achieved in membrane-
based energy recovery ventilators (MERV). A third factor is the additional fan power
required to drive the air flow through the recovery system. While recovering energy
from the exhausted air stream, HRVs/ERVs increase the electrical energy required by
∗Corresponding author.
Email address: [email protected] (Elisa Magnanelli)
2
ventilation fans [20] and thus a net energy saving is not guaranteed.
Many works can be found in the literature that assess the performance of recovery
ventilators [18, 21–29]. Three performance indicators are traditionally considered. The
first parameter, s, accounts for the recovery of sensible heat and it is often defined
as [30]
s =
in − T ex
T ex
in − T0
T ex
out
,
(1)
where T ex
recovery ventilator, respectively, and T0 is the temperature of the outdoor ambient.
out are the temperatures of the exhaust air at the inlet and outlet of the
in and T ex
If water is condensed on the wall separating the air streams, latent heat is released,
and a second performance parameter can be defined as [26]
h =
in − Hex
Hex
in − H0
Hex
out
,
(2)
in and Hex
where Hex
out are enthalpy flows entering and leaving the system with the exhaust
air stream, and H0 is the enthalpy flow that enters the ventilator with the fresh air
stream.
In systems where water is also exchanged, a third performance indicator is used to
describe the recovery of moisture [30]
w,in − xex
xex
w,in − xw,0
xex
w,out
,
w =
(3)
where xex
and outlet, and xw,0 is the water mole fractions in the outdoor air.
w,out are the water mole fractions of the exhaust air at the system inlet
w,in and xex
While useful by themselves, these indicators are difficult to compare and relate to
each other. Since there is no obvious optimal trade-off between them it is challenging
to develop a global performance indicator that allows for a sensible comparison of
different technical solutions and different types of recovery devices. Moreover, these
parameters do not take into consideration the additional need for fan power. Indeed,
a fundamental challenge with recovery ventilators is that they involve various energy
forms (i.e. thermal energy, chemical potential energy, and electric energy), which differ
from each other both from a thermodynamic and an economic point of view (e.g. a
certain amount of heat at 294 K has less energy quality and it is in principle cheaper to
produce than the same amount of electricity).
In this work, we investigate the efficiency of ventilation recovery systems using
exergy analysis and nonequilibrium thermodynamics. This analysis makes it possible
to assess the performance of HRV/ERV in terms of loss of work potential (i.e. exergy
destruction) or, equivalently, in terms of entropy produced by the process [31]. The
method allows for the definition of one single parameter to quantify the performance
of the process, the exergy efficiency [32], which can account for the quality of different
forms of energy. An easier and more meaningful comparison of different technical
alternatives is therefore possible. In addition, it is possible to localize and quantitatively
compare the different sources of lost work potential in the system, which may aid in
design and optimization of the global ventilation system. This approach is in line
3
Figure 1: Schematic representation of the recovery ventilator. The exhaust air from the indoors flows parallel
to the wall that separates the two sides of the system (z-direction). The fresh air from the outdoor ambient
flows in opposite direction (counter-current configuration). Fluxes of heat, J(cid:48)
q, and water, Jw, are exchanged
across the separating wall (x-direction).
with a part of the strategy of the American Society of Heating, Refrigerating and Air-
Conditioning Engineers (ASHRAE), which has established a technical group named
Exergy Analysis for Sustainable Buildings, to promote the use of this concept in the
assessment of energy use of buildings [33].
In our analysis, we include the losses due to pressure drops in the ventilation re-
covery system, and thus we account for the additional power consumed by fans. These
may contribute significantly to the overall losses [34] but they are often neglected in
performance studies and are not reflected in the performance indicators that are used.
We will consider examples where we calculate and compare the exergy efficiency
and the heat and moisture effectiveness of a flat plate HRV and those of a structurally
similar MERV. We shall see how ambient conditions influence the efficiency of the
two systems in a non-trivial way, that cannot be predicted by the most commonly used
heat and moisture effectiveness. We will consider ambient temperatures and relative
humidities characteristic of both cold and warm climates.
After presenting the recovery systems in Section 2, the thermo-fluid dynamic model
developed to describe the recovery ventilators is introduced in Section 3. In Section 4,
we present the analyzed cases and the most relevant data. The solution procedure and
model validation are illustrated in Section 5, before results are presented and discussed
in Section 7. Conclusions are summarized in Section 8.
2. System
Figure 1 shows a schematic representation of a recovery ventilator. The exhaust air
from the indoor environment flows from left to right in the positive z-direction, while
the fresh stream from the outdoor ambient flows parallel to it, but in opposite direction.
Since the two streams have different temperatures and composition, driving forces exist
for heat and mass transport between them.
In the thermo-fluid dynamic description of the system, we make the following as-
sumptions.
4
• Each air stream can be described as consisting of a bulk region and of two con-
vective boundary layers, one at each side of the walls separating the air streams.
In order to enhance heat and mass transfer (and to maintain the structural in-
tegrity of the channel), spacers are present in the air channels. These favor tur-
bulence [35]. The bulk of the air streams are therefore considered well mixed in
the x- and y-direction. In the convective boundary layers, however, temperature
and concentration gradients are present in the x-direction [36].
• Diffusive fluxes along the z-direction are neglected, since they are small in com-
parison to the advective flow. The diffusive contributions can typically be ne-
glected in systems where the Peclet number, Pe, is much larger then one [37]. In
the considered system, we have Pe ∼ 200.
• Humid air is treated as an ideal mixture of ideal gases, where the only compo-
nents are water and dry air. This assumption is justified by the fact that, in the
considered range of temperature and relative humidity, the compressibility factor
of humid air at ambient pressure is close to one [38].
• The dry air is assumed to be 79% mole fraction of nitrogen and 21% oxygen.
Other components are present, but they constitute less than 1% of the gas and
are therefore neglected [39].
We consider and compare two structurally similar recovery ventilator solutions.
They have the same geometry, but differ in the structural and physical properties of the
wall that separates the air streams:
Heat recovery ventilator (HRV): The wall separating the air streams is a metal plate.
Therefore, only sensible heat can be transferred from one air stream to the other.
Membrane energy recovery ventilator (MERV): The wall separating the two streams
is a water-permeable membrane. Both sensible heat and moisture can be ex-
changed between the two streams. We assume that water is the only component
that can permeate the membrane.
The recovery performances of these systems depends greatly on the amount of ex-
change surface between the two streams. Therefore, in practical applications, N pairs
of exhaust/fresh air channels are stacked to obtain a suitable configuration for the sys-
tem.
3. Theoretical formulation
In Section 3.1, we present the equations used to describe the spatial variation of the
system thermodynamic variables. Section 3.2 deals with heat and mass transport across
the system (x-direction). The description of transport makes use of the nonequilibrium
thermodynamic framework [40]. In Section 3.3, we present the different contributions
to the loss of work potential (i.e. exergy destruction), and define the exergy efficiency
for the system.
5
3.1. Balance equations
The thermodynamic driving forces between the two air streams cause transport
of heat and mass from one side to the other. Humid air flows in the ducts, and the
water mole fraction varies in space. Therefore, two mass balances are necessary to
describe the system, one for the water content of air (subscript w) and one for the dry
air (subscript a). At steady state, the mass balance equations for the exhaust air can be
written as:
dFex
w
dz
dFex
a
dz
= −W Jw
= −W Ja
=
0
(4)
(5)
w and Fex
where Fex
a are the water and the dry air molar flow rates, W is the width of the
duct (in the y-direction), and Jw and Ja are the molar fluxes across the wall separating
the streams. Since the separating wall is impermeable to dry air, Ja equals zero and Fex
a
is constant. The total molar flow rate is given by the sum of the flow rates of the two
components, Fex = Fex
Since we assume that the diffusive fluxes in the z-direction are negligible in compar-
ison to the bulk flow velocity, the component flow rates can be related to the component
mole fraction as:
w + Fex
a .
(6)
where xex
is the mole fraction of the component i, cex is the total molar concentration
i
of gas, vex is the molar velocity, and Aex is the duct cross sectional area. We use the
subscript i to indicate the different components, i ∈ [w, a].
i Fex = xex
i cexvexAex
= xex
Fex
i
The pressure drop in a duct can be calculated by Darcy–Weisbach equation [36]:
dpex
dz
= − f ex ρex vex vex
2Dex
h
(7)
(9)
where f ex is the friction factor, ρex is the mass density, pex is the pressure, and Dex
the hydraulic diameter of the duct.
h is
The energy balance of the exhaust air stream can be written as:
d (Fexhex)
dz
= −W Jq
(8)
where hex is the molar enthalpy of the gas, and Jq is the total heat flux from the exhaust
to the fresh air. Applying the chain rule and substituting Eqs. 4 and 5 into Eq. 8, the
energy balance can be rewritten in terms of the molar enthalpy as:
(cid:16)
(cid:17)
Fex dhex
dz
= −W
Jq − hexJ
where J = Jw + Ja is the total molar flux through the separating wall. Eq. 9 can be
reformulated as an ordinary differential equation (ODE) for the temperature. For an
ideal gas, the total enthalpy differential can be written as:
dhex = cex
p dT ex +
hex
i dxex
i
(10)
(cid:88)
i
6
where cex
enthalpy. Eq. 9 can then be rewritten as:
p is the molar heat capacity, T ex is the temperature, and hex
i
is the partial molar
cex
p
dT ex
dz
= − W
Fex
Jq − hexJ
hex
i
dxex
i
dz
(11)
(cid:16)
(cid:17) −(cid:88)
i
where Mex is the molar mass of the exhaust air.
All the thermodynamic variables introduced above refer to the exhaust air, and they
are referred to with the superscript ex. We indicate the corresponding variables on the
fresh air side with a similar notation and the superscript fr. The balance equations for
the fresh air are:
dFfr
w
dz
dFfr
a
dz
dpfr
dz
dT fr
dz
cfr
p
= W Jw
= W Ja = 0
= − f fr ρfr(cid:12)(cid:12)(cid:12)vfr(cid:12)(cid:12)(cid:12) vfr
(cid:16)
2Dfr
h
Jq − hfrJ
(cid:17) −(cid:88)
i
=
W
Ffr
(12)
(13)
(14)
(15)
hfr
i
dxfr
i
dz
3.2. Transport between the air streams
Transport between the air streams can be described by use of flux-force relations
from nonequilibrium thermodynamics [40]:
(cid:33)
(cid:33)
(cid:32) 1
(cid:32) 1
T
T
(cid:18) µw
(cid:19)
T
−∆
∆
+ hex
w ∆
(cid:48)ex
= rqJ
q
= rwJw
(16)
(17)
q = Jq − Jwhex
(cid:48)ex
where µw is the chemical potential of water, rq and rw are the heat and mass transport
coefficients, and J
w is the measurable heat flux on the exhaust air side.
We use the symbol ∆ to indicate the difference in thermodynamic variables between
the two sides. The left-hand sides of Eqs. 16-17 represent the driving forces for heat
and mass transport. Coupling between fluxes has been neglected because they are most
often small in comparison to the main transport contributions.
Two main phenomena contribute to the transport coefficients. The first contribution
is due to the resistance to transport offered by the wall between the two streams. The
second contribution is given by the resistance to transport of the convective boundary
layers that form in the fluid on each side of the separating wall. The calculation of the
transport coefficients is further discussed in Appendix A.
3.3. Exergy efficiency and loss of useful work
For a recovery ventilator, a task exergy efficiency can be defined as [31]:
η = 1 − Ed
Eu
7
(18)
where Eu is the exergy of the exhaust air that enters the recovery ventilator (i.e. max-
imum useful work that can be extracted from the exhaust air stream), and Ed is the
exergy that is destroyed in the process (i.e. useful work that is lost).
Two main phenomena contribute to the overall exergy destruction:
Ed = Ed,irr + Ed,amb
(19)
where Ed,irr is the exergy destroyed due to irreversibilities inside the system, and Ed,amb
is the exergy that is lost by discharging the exhaust air in the outdoor ambient.
The exergy that is destroyed due to irreversibilities is directly related to the total
entropy that is produced during the process, Σirr, by the Gouy–Stodola theorem [41]:
Ed,irr = T0Σirr
(20)
where T0 is the temperature of the ambient. The total entropy production of the process
is the integral of the local entropy production of a system cross section, σ, over the
system length, L [40]:
σ dz
(21)
(cid:90) L
Σirr =
0
The entropy production can be written as sum of the products of all thermodynamic
forces in the system and their conjugated fluxes [40]. Thus, for the present case, the
local entropy production is:
σ = J
(cid:48)ex
q ∆
(cid:33)
(cid:32) 1
(cid:32)
T
(cid:32)
−∆
(cid:19)
(cid:18) µw
(cid:33)
T
+ Jw
− 1
T ex
dpex
dz
+ hex
(cid:33)(cid:33)
(cid:32) 1
(cid:32)
w ∆
− 1
T fr
T
dpfr
dz
+Aexvex
+ Afrvfr
(cid:33)
(22)
dp
dz
T
where the first term on the right-hand side represents the product between the mea-
surable heat flux and the thermal driving force, and the second is the product between
the mass flux and the chemical driving force. The last two terms represent the entropy
production due to friction, where the flux is the stream velocity and the viscous flow
driving force is
(cid:16)− 1
(cid:17)
.
The exergy destroyed due to irreversibilities can also be calculated from the exergy
balance on the system, as presented in Appendix B. The alternative formulation given
by Eq. B.2 does not require detailed knowledge of the thermodynamic variables across
the system, but relies exclusively on inlet and outlet values of the variables. However,
it does not localize where entropy is produced in the system and by what phenomena.
This information can be used in designing, improving and optimizing recovery devices.
The exergy that is lost to the ambient corresponds to the exergy that is destroyed by
discharging the exhaust air stream in the ambient (i.e. the exergy of the exhausted air
stream when it leaves the recovery ventilator, Eex
out [31]:
Ed,amb = Eex
out
(cid:16)
= Fex
out
out − hex
hex
out,0 − T0
= Ed,amb,ph + Ed,amb,ch
(cid:17)(cid:17)
(cid:16)
out − sex
sex
out,0
8
(cid:88)
i
(23)
(cid:33)
(cid:32) xex
i,out
xi,0
+ Fex
outRT0
xex
i,out ln
out,0 and sex
where hex
out,0 are the enthalpy and the entropy of the gas evaluated at T0 and p0.
The subscript out indicates the thermodynamic variable at the outlet of the considered
air stream. Here, Ed,amb,ph and Ed,amb,ch are the physical and the chemical contributions
to the exergy lost to ambient.
The exergy of the exhaust air that enters the recovery ventilator is defined as [31]:
Eu = Eex
out
(cid:16)
(cid:16)
(cid:17)(cid:17)
= Fex
in
in − hex
hex
in,0 − T0
in − sex
sex
in,0
+ Fex
in RT0
xex
i,in ln
(cid:33)
(cid:32) xex
i,in
xi,0
(24)
(cid:88)
i
in,0 and sex
in is the molar entropy of the inlet exhaust air, hex
where sex
in,0 are the enthalpy and
entropy of the gas evaluated at the temperature and pressure of the ambient (i.e. T0 and
p0), and xi,0 is the reference composition of the ambient. The subscript in indicates the
thermodynamic variable at the inlet of the considered air stream. Similarly to Eq. 23,
the first term on the right-hand side of Eq. 24 represents the physical contribution to the
exergy of the exhaust air stream, Eu,ph, while the second term represents the chemical
contribution, Eu,ch [31].
By substituting Eq. 20, Eq. 23 and Eq. 24 into Eq. 18, it is possible to reformulate
the expression for η as:
(cid:16)
in − Eex
Eex
out
Eex
in
η =
(cid:17) − Ed,irr
(25)
out
(cid:17)
(cid:16)
in − Eex
Eex
This reformulation allows us to better understand the meaning of η. The term at the
numerator of Eq. 25 represents the net useful work recovered by the recovery ventilator,
which is given by the difference between the exergy that is recovered from the exhaust
, and the exergy that is dissipated due to irreversibilities, Ed,irr.
air stream,
The denominator in Eq. 25 represents the maximum useful work that can be ex-
tracted from the exhausted air stream before it is discharged in the ambient. Thus, η
represents the fraction of the exhaust air exergy that is recovered. In the ideal reversible
case where the exergy of the exhaust air is completely recovered with no pressure drop
in the system, η would be equal to one. The closer η is to one, the larger is the fraction
of the exergy available in the exhaust air which is recovered.
in − Eex
Eex
Moreover, this parameter allows us to identify those operating conditions for which
, is smaller than the exergy that is destroyed by the
the recovered exergy,
ventilator due to irreversibilities, Ed,irr. For such conditions, η is negative. From a
thermodynamic point of view, a negative η indicates that there exists a better alternative
to provide the exergy that is recovered by the recovery ventilator, and it would be better
to let the air streams bypass the recovery unit.
out
(cid:16)
(cid:17)
4. Case specification
4.1. Relevant data
The most relevant data used in the calculations are summarized in Table 1. We
in , of 40 %, and indoor temperature of 294 K,
adopt a target indoor relative humidity, RHex
9
Table 1: Operating conditions and structural and physical parameters assumed in calculations.
294
40
0.6
Value
105
Symbol
p0
Indoor environment
T indoor
RHindoor
vex
in
Outdoor ambient
T0
263-303
RH0
10-90
Geometrical parameters
L
0.185
Wduct
0.185
4.0 · 10−3
duct, Hfr
Hex
N
HRV
δ
λwall
Pw
MERV
δ
λwall
Pw
5.0 · 10−4
2.0 · 102
1.02 · 10−4
1.0 · 10−13
0.13
57
duct
0
Unit
Pa
K
%
m s−1
K
%
m
m
m
-
m
W m−1 K−1
kmol Pa−1 s−1 m−1
m
W m−1 K−1
kmol Pa−1 s−1 m−1
Ref.
[14]
[14]
[24]
[24]
[24]
[24]
[24]
[24]
[42]
which lie in the range of indoor humidities and temperatures recommended by the Nor-
wegian building code [14]. We consider a range of outdoor temperatures and humidity,
typical of both cold and warm climates.
The geometrical configuration of the two systems is the same, but they differ in
structural and physical characteristics of the wall that separates the air streams. The
geometrical parameters and the wall properties in Table 1 are used to calculate the
transport coefficients in Table 2. Usually, the air channels contain spacers that improve
the structural integrity channels and enhance heat and mass transfer. However, the pres-
ence of spacers makes the friction factor and, therefore, the pressure drops larger (Eq. 7
and Eq. 14). We use the relation for friction factor in ducts with spacers determined by
the experimental work in Ref. [35].
In order to obtain a larger exchange area between the two streams, several air chan-
nels are alternately stacked on top of each other to obtain a convenient configuration of
the system. Because of symmetry, the complex stacked system can be described by a
single pair of exhaust/fresh air channels, where the height of the channel is half of the
one of a single channel, H = Hduct/2, and the width is double that of a single channel
multiplied by the number of channel pairs, W = 2WductN.
10
4.2. Boundary conditions
In order to solve the set of eight ODEs, we need to impose eight boundary condi-
tions:
w
= T indoor
T ex
in
= Aex vex
in p0
Fex
in
RT indoor
= Fex
in xindoor
Fex
w,in
pex
out = p0
T fr
= T0
in
Ffr
= Fex
a,in
a,in
= Ffr
Ffr
w,in
a,in
pfr
out = p0
xw,0
1 − xw,0
(26)
(27)
(28)
(29)
(30)
(31)
(32)
(33)
The boundary condition values are calculated from the data reported in Table 1. In
addition, we consider that the dry air stream extracted from the indoor environment
needs to be the same as the dry air stream supplied from the outdoor ambient.
5. Solution procedure
With the given assumptions, the system is properly described by a set of eight
ordinary differential equations (Eqs. 4-5, Eq. 11, Eq. 7, and Eqs. 12-14). Since the
boundary conditions are specified at both ends of the system (Section 4.2), the problem
is a two-point boundary value problem. The problem is solved by use of the MATLAB
bvp4c-solver, which is a collocation solver that implements the three-stage Lobatto
IIIa formula and bases the mesh selection and error control on the residual of the solu-
tion [43].
6. Model validation
6.1. Consistency check
In order to check the consistency of the thermodynamic framework presented in
Section 3, the exergy destruction due to irreversibility has been calculated for all inves-
tigated cases according to both Eq. 20 and Eq. B.2. This allows us to check that neither
the first law or second law is violated anywhere inside the system. The relative error
between the results given by the two equations are order of the numerical accuracy of
the calculations (10−5).
11
Table 2: Calculated transport coefficients for the HRV and for the MERV, at z = L/2. Calculations are carried
out at T0 = 263 K and RH0 = 60 %.
rh,conv
q
rwall
q
rc,conv
q
rh,conv
w
rwall
w
rc,conv
w
HRV
1.4 · 10−7
2.7 · 10−11
1.5 · 10−7
-
-
-
MERV
1.4 · 10−7
8.8 · 10−9
1.5 · 10−7
2.6 · 106
4.5 · 109
2.6 · 106
Unit
m2 s J−1 K−1
m2 s J−1 K−1
m2 s J−1 K−1
m2 s J K−1 kmol−2
m2 s J K−1 kmol−2
m2 s J K−1 kmol−2
6.2. Comparison with empirical correlations
Heat and moisture exchange effectiveness, s and w, are two of the parameters
that are most commonly used in the literature to characterize recovery performance.
Most published research uses heat and moisture effectiveness as a criterion to verify
the accuracy of simulations as well as of experiments [30]. Widely used empirical
correlations exist to calculate the heat and moisture effectiveness based on the design
parameters of the HRV/MERV (i.e. the number of transfer unit for transport of heat
and moisture, NTUs and NTUw) [44, 45]. In order to verify the accuracy of the model,
the computed heat and moisture effectiveness were compared to those predicted by the
-NTU relations that presented in References [44] and [45]. The results agree within
1.6 % (with the exception of a narrow range of outdoor conditions where s diverges),
indicating that the model gives a good description of a recovery ventilator.
6.3. Sensitivity analysis
A sensitivity analysis has been carried out on four parameters, to show how the
exergy efficiency varies as a function of the selected parameters. The result of the
sensitivity analysis (reported in Appendix C) shows that uncertainty in the input pa-
rameters does not significantly affect the results and it does not reveal any unexpected
relationships between inputs and outputs.
7. Results
7.1. Transport between streams
Two main phenomena contribute to the overall heat and mass transport coefficients.
One contribution is due to the heat transport resistance of the wall material, while a sec-
ond contribution depends on transport across the convective boundary layers adjacent
to the separating wall. Table 2 presents the different contributions to the total heat
and mass transport coefficients in the HRV and MERV as calculated according to the
equations presented in Appendix A. The convective layers are responsible for most of
the heat transport resistance. In the HRV, the convective resistances to heat transport
are four orders larger than that due to conduction across the separating wall. This jus-
tifies the common practice of neglecting the wall conductive resistance [30] for this
case. Membrane materials for MERV typically have a lower thermal conductivity than
12
Figure 2: Sensible heat exchange effectiveness, s, for the HRV (solid lines) and for the MERV (dashed
lines), for RH0 = 40 % (light blue lines), RH0 = 60 % (blue lines), and RH0 = 80 % (black lines). At
T0 = T indoor, s is not defined.
the metallic materials in HRV (three orders of magnitude lower, in the present case).
The resistance to conductive heat transport is small in comparison to the convective
resistance even in the case of the MERV, contributing only 3.5 % to the total resistance.
In contrast to the case of heat transport, the main resistance to mass transport is
caused by diffusion through the membrane material. In the present case, the convective
resistance is only 0.1 % of the total resistance to mass transport. Convective transport
may however play a more important role under different operating conditions as other
studies have reported convective contributions to water mass transport as high as 30 %
for different flow regimes [28].
7.2. Heat and moisture effectiveness and exergy efficiency
Figure 2 shows the sensible heat exchange effectiveness of the HRV (solid lines)
and of the MERV (dashed lines) calculated according to Eq. 1. The sensible effective-
ness is very similar for the HRV and the MERV for all considered outdoor tempera-
tures and relative humidities. The slightly higher s for the HRV is due to the lower
resistance to heat transport offered by the metal wall that separates the streams. The
sensible effectiveness does not have a strong dependence on the outdoor conditions,
and increases only slightly for higher outdoor temperatures. Since the denominator in
Eq. 1 is zero when the outdoor and indoor temperatures are equal (at T0 = T indoor), s
cannot be defined for this case. Because of friction, a small increase of the exhausted
stream temperature occurs even when indoor and outdoor temperatures are equal. This
phenomenon makes s diverge for outdoor temperatures close to the indoor tempera-
ture.
13
Figure 3: Moisture exchange effectiveness, w for the MERV (dashed line) as functions of the outdoor
temperature, for RH0 = 40 % (light blue line), RH0 = 60 % (blue line), and RH0 = 80 % (black line). Since
HRV does not exchange any water, w is everywhere zero for the HRV. At xw,0 = xindoor
, w is not defined.
w
Figure 3 represents the moisture transfer effectiveness, w, for the MERV (dashed
lines) as a function of the outdoor temperature, for RH0 = 40 % (light blue line), RH0 =
60 % (blue line), and RH0 = 80 % (black line). In the HRV, the air streams do not
exchange any moisture, and this parameter is always zero. For the MERV, w decreases
with decreasing outdoor temperature. Similar behaviour has been observed by others
(e.g. the experimental work in Reference [28]). When the water mole fractions of the
indoor and outdoor air are equal w is not defined.
Figure 4 illustrates the exergy efficiency, as defined by Eq. 18, of the HRV (solid
lines) and the MERV (dashed lines). The behavior of η as a function of the ambient
temperature is very complex, however it is similar for different outdoor relative hu-
midities. In cold climates, for RH0 = 60 % (blue line), and RH0 = 80 % (black line),
η of the HRV increases as the outdoor temperature increases, until it reaches the value
for which indoor and outdoor water mole fraction are the same, xindoor
= xw,0. For a
further increase of T0, η drops. When T indoor = T0, the exergy efficiency of the HRV
is negative, since useful work is lost, due to pressure drops, without generating any
useful output. For T0 higher than the indoor temperature, η increases with increasing
ambient temperature. When xindoor
= xw,0, the air streams cannot exchange moisture
in any of the configurations, and the efficiencies of the HRV and MERV are approx-
imately the same. For RH0 = 40 % (light blue lines), xindoor
and xw,0 are the same
when T indoor = T0. Under such conditions, no useful work can be recovered from the
exhausted stream, and the exergy efficiency is not defined.
w
For the MERV, η is roughly constant both in cold and warm climates, and it drops
w
w
14
Figure 4: Exergy efficiency, ex, as defined by Eq. 18 for the HRV (solid lines) and for the MERV (light blue
lines) as functions of the outdoor temperature, for RH0 = 40 % (red lines), RH0 = 60 % (blue lines), and
RH0 = 80 % (black lines).
from one value to the other in the temperature interval between the conditions for which
xindoor
= xw,0 and T indoor = T0. The comparison between the two systems shows that
w
the exergy efficiency of the MERV is never smaller than that of the HRV. In cold cli-
mates, the difference between the two efficiencies increases with decreasing tempera-
ture. For RH0 = 60 % (blue dashed line) and RH0 = 80 % (black dashed line), the ex
of the MERV is never negative. However, for temperatures and water fraction of the
outside air equal to those of the indoor space, the MERV exergy efficiency is negative
(light blue dashed line), indicating that no net benefit is obtained from the recovery
ventilator. Instead, it is detrimental to the performance of the ventilation system as
electric energy is expended to compensate for pressure drop. Under such conditions it
would be better to let the air streams bypass the recovery unit.
Figure 5 shows how the exergy efficiency of the HRV (solid line) and of the MERV
(dashed line) varies as a function of the exhaust air volume flow rate. The air flow rate
is varied within a range that allows for the air velocity to stay under 5 m s−1, in order to
avoid unnecessary noise generation. The exergy efficiency is largest when the flow rate
approaches zero, and it decreases for increasing flow rates. There is thus a trade-off
between the amount of air that can be treated by a ventilator of a given size and the
resulting exergy efficiency. A larger HRV/MERV will retain a high efficiency for larger
flow rates, but would require larger investment costs. In practice, the air volume flow
rate is determined by the size of the building and by the rate at which the indoor air
needs to be replaced. The latter is often dictated by local regulations, e.g. [14]. With the
flow rate given, one may choose an acceptable ex and increase the size of HRV/MERV,
15
Figure 5: Exergy efficiency, η, for the HRV (solid line) and for the MERV (dashed line) as functions of the
volume flow rate. Calculations are carried out at T0 = 263 K and RH0 = 60 %.
or change other design parameters, until the desired efficiency is achieved.
This is one example of a way in which the exergy efficiency may guide the design
and sizing of such devices. Nowadays, graphs picturing heat exchange effectiveness as
a function of volume flow rate are used to dimension recovery ventilators. By switching
to the use of a graphs like the one in Fig. 5, it is possible to dimension the recovery
ventilators taking into account not only the effect of heat recovery, but also the effect
of moisture recovery and pressure drops.
Figure 5 shows that for high flow rates the exergy efficiency of the HRV is higher
than that of the MERV. This is due to the fact that the mass transport resistance is large
in comparison to that of heat transport. Therefore, the transport of heat is more facile
than the mass transport. Accordingly, for high air velocities, the water recovery is very
low, and heat recovery represents the dominant contribution to the exergy efficiency.
The resistance to convective heat transport decreases with increasing flow rate and,
hence the wall heat transport resistance becomes relatively more important. Since the
wall thermal resistance of the MERV is two orders of magnitude larger than that of the
HRV, the MERV heat recovery and exergy efficiency are smaller than those of the HRV.
Figures 6 and 7 show the exergy efficiency as function of both outdoor temperature
and relative humidity. These figures are useful e.g. to identify the range of outdoor
conditions for which the recovery ventilator is responsible for increasing the energy
need of the building (area enclosed by the light blue lines where η is zero). Under such
conditions, it is more beneficial to bypass the HRV/MERV. The comparison between
the two pictures shows that the range of unfavorable outdoor conditions is much smaller
for the MERV than for the HRV. Moreover, the MERV retains a exergy efficiency larger
16
Figure 6: Exergy efficiency, η, for the HRV as function of the outdoor ambient temperature and of the outdoor
relative humidity.
Figure 7: Exergy efficiency, η, for the MERV as function of the outdoor ambient temperature and of the
outdoor relative humidity.
17
Figure 8: Physical useful work lost to the ambient, Wamb,ph (purple lines), chemical useful work lost to the
ambient, Wamb,ch (orange lines), and useful work lost to irreversibilities, Wirr (yellow lines), for the HRV
(solid lines) and for the MERV (dashed lines) as functions of the outdoor temperature. The solid and dashed
purple lines (Wamb,ph) coincide at the scale of the representation. Calculations are carried out at RH0 = 60 %.
than 60 % in a larger portion of the outdoor condition space.
In the present case, pressure losses are low, and thus the operational range for
which η is negative is small. However, under different operating conditions and config-
urations (e.g. higher air velocities or larger friction coefficients inside the ducts), such
losses could have larger influence on the overall performance, and enlarge the range of
operating conditions for which the use of a HRV/MERV is not beneficial.
If the indoor ambient conditions are to be maintained constant, the exergy of the
building needs to stay constant over time. This means that the exergy which is not
recovered by the recovery ventilator, needs to be supplied to the indoor ambient by
other auxiliary systems (i.e. heaters/air coolers, humidifiers/dehumidifiers, and fans),
whose energy consumption one needs to pay for. In an ideal case, η would be equal to
one and none of the auxiliary systems would be necessary. For a fixed set of operating
conditions, the higher η is, the lower is the exergy load that the auxiliary systems need
to supply for. When η is negative, the exergy load that the auxiliary systems need to
supply is larger than that they would need to provide if the recovery ventilator was not
used.
The comparison between s, w and the exergy efficiency η shows that these three
parameters have very different behaviors, and that the combination of s, w into a single
performance indicator is not straightforward. Performance analysis based on s and
w can therefore create difficulties in the comparison of different solutions, especially
when alternatives where only heat is recovered need to be compared with solutions
where both heat and moisture are recovered.
18
Figure 8 shows for RH0 = 60 % the contributions to exergy destruction due to
physical (purple lines) and chemical (orange lines) work potential that is lost to the
ambient, and due to the work potential lost due to irreversibilities in the system (yellow
lines), for the HRV (solid lines) and for the MERV (dashed lines). Similar results are
obtained for the other considered outdoor relative humidities (not reported here).
The physical lost work (or physical exergy destruction, Ed,amb,ph) is approximately
the same for the two systems (solid and dashed purple lines coincides at the plot ac-
curacy), since the outlet temperature of the exhaust air is about the same in the HRV
and MERV. Moreover, Ed,amb,ph is quite small in comparison to the other work losses,
especially at very low and at very high outdoor temperatures.
On the other hand, Ed,amb,ch is much larger in the HRV than in the MERV, since
none of the moisture in the exhaust air is recovered in the HRV and all the chemical
potential work of the inlet exhaust air is lost.
The loss of work potential due to irreversibilities (Ed,irr) in the MERV is always
larger than in the case of the HRV. Indeed, Ed,irr strongly depends on the irreversible
heat and mass transfer and on the pressure drops in the system (see Eq. 22). Since the
contributions due to heat transfer and pressure drops are approximately the same in the
two systems, the difference between Ed,irr is mainly due to the additional contribution
of moisture transport in the MERV. Figure 8 shows that Ed,irr goes through a minimum.
For the HRV, minimum Ed,irr is obtained for T indoor = T0. Under these conditions, no
heat is exchanged between the air streams, and the only contribution to Ed,irr is due to
the pressure drops. In the MERV, the minimum in Ed,irr is located between the outdoor
conditions for which the sensible heat transfer is zero (T indoor = T0), and the condition
for which the moisture transfer is zero (xindoor
= xw,0).
The overall exergy destruction in the system gives an indication of the energy that
one needs to supply to the indoor ambient to maintain temperature and humidity levels
that are compatible with comfortable indoor conditions. In order to maintain comfort,
the ventilation system is normally coupled with auxiliary systems that compensate for
the work lost during ventilation. In particular, auxiliary systems add or remove sensible
heat from the indoor air (e.g. radiators or air conditioners), others control the moisture
level (e.g. dehumidifiers or humidifiers), while fans compensate for the pressure drops
in the ventilation system. The knowledge on the different contributions to the lost work
allows us to locate and compare the ideal amount of energy needed by the different
auxiliary systems to maintain the desired temperature and humidity levels. Since dif-
ferent auxiliary systems might be supplied by different energy sources, this approach
makes it easier to include considerations on the primary energy use of buildings.
w
Figure 8 shows that, for the HRV, most of the energy that needs to be supplied to
the system is used for moisture control, both in cold and warm climates. With the use
of the MERV in cold climates, the energy used for such purpose is reduced to a level
similar to that needed for temperature control. This is not the case for warm climates.
For ambient temperatures close to that of the indoor temperature, the use of the HRV is
not beneficial, since it increases the fan power, without bringing any recovery of heat or
moisture. On the other hand, due to moisture exchange, the use of the MERV remains
beneficial for ambient temperatures close to the indoor temperature, when the indoor
and outdoor water content is different.
19
8. Conclusions
We have used exergy analysis and nonequilibrium thermodynamics to characterize
the performance of heat and energy recovery ventilators. We considered and compared
the performances of a heat recovery ventilator (HRV) and those of a structurally similar
membrane energy recovery ventilator (MERV). This kind of analysis made it possible
to quantify, assess and compare the performances of HRV/MERV in terms of a single
parameter, the exergy efficiency. This parameter describes the loss of work potential
and may account for all different sources of work loss and for differences in energy
quality.
Traditional effectiveness parameters defined by Eqs. 1-3 are most commonly used
in the literature to characterize recovery ventilators performance. While useful by
themselves, these effectiveness parameters are difficult to compare and relate to each
other. Since there is no obvious optimal trade-off between them, it is challenging to
combine them in a way that allows for a sensible comparison of different technical
solutions such as the HRV and MERV.
We showed that the exergy efficiency can be used to identify the range of operating
conditions for which the recovery ventilator is responsible for increasing the energy
need of the building, and for which it is more beneficial to bypass the HRV/MERV.
When the exergy efficiency is negative, the recovered heat and moisture are not enough
to compensate for the fan power necessary to drive the air flow in the recovery unit.
Such phenomenon cannot trivially be predicted by traditional performance parameters.
Moreover, by using graphs that represent exergy efficiency as a function of air
volume flow rate, it is possible to dimension the recovery ventilators taking into account
not only the effect of heat recovery, but also the effect of moisture recovery and pressure
drops.
The exergy analysis additionally enabled the localization and comparison of the
various components of loss of useful work in the systems. This means that the analysis
can be used not only to assess the efficiency of the process, but also to draw considera-
tions on the design of the auxiliary devices used in the process.
Acknowledgments
The Norwegian University for Science and Technology, the Research Council of
Norway (project number 262644) and the VISTA program, a basic research program
in collaboration with The Norwegian Academy of Science and Letters and Statoil, are
acknowledged for financial support.
Appendix A. Transport coefficients
The transport coefficients have two main contributions. A first contribution is due
to the resistance to transport offered by the wall separating the two streams, rwall
. A
second contribution is given by the resistance offered by the convective boundary layers
that are present at the two sides of the separating wall, rh,conv
and rc,conv
. Similarly to
resistances in series, the overall transport coefficients can be written as:
i
i
i
ri = rh,conv
i
+ rwall
i
+ rc,conv
i
(A.1)
20
Appendix A.1. Heat transport coefficients
The wall heat transport coefficient can be derived from experimental data on the
thermal conductivity of the wall material. Thermal conductivity experiments are typ-
ically carried out at zero mass flux. According to Fourier's law, the measurable heat
flux is:
q = −λwall ∆T
J(cid:48)
δ
(A.2)
where λwall is the thermal conductivity of the wall material, and δ is the wall thickness.
By comparing Eq. A.2 and Eq. 16, we can relate the wall heat transport coefficient to
the experimental thermal conductivity of the wall material:
rwall
q =
δ
λwallT 2
(A.3)
The heat transport resistance due to the convective boundary layer can be calculated
from the convective heat transfer coefficient, hconv, which is defined as:
q = −hconv∆T
J(cid:48)
(A.4)
where ∆T is the difference in temperature between the membrane surface and the gas
bulk. The coefficient depends on the characteristics of the flow, and it is related to the
Nusselt number by the following relation:
hconv = Nu λ
Dh
(A.5)
where λ is the thermal conductivity of the fluid. Empiric relations exist that relate the
Nusselt number to the characteristics of the flow.
In order to maintain the channel
geometry and to enhance heat and mass transfer, spacers are usually present in the
air channels. In this work, we use the correlations for Nusselt number that have been
established for flow in ducts with spacers [35]:
Nu = jRePr1/3
(A.6)
where j is the Colburn heat transfer factor, Re is the Reynolds number, and Pr is the
Prandtl number. The Colburn heat transport factor is determined experimentally as a
function of the Reynolds number [35]:
j =
C0
Rem
(A.7)
where C0 and m are two fitted parameters.
By comparing Eq. A.5 with Eq. 16, we can relate the convective heat transport
resistance to the convective heat transport coefficient:
rconv
q
=
1
hconvT 2
21
(A.8)
Appendix A.2. Mass transport coefficients
Mass transport through semi-permeable membranes is usually characterized by per-
meability parameters. The permeability of a material to a component is determined ex-
perimentally at isothermal conditions (i.e. ∆T =0). In the present case, the membrane
permeability to water, Pw, is defined as:
Jw = −Pw
∆pw
δ
(A.9)
where ∆pw is the water partial pressure difference between the two sides of the mem-
brane. By comparing Eq. A.9 with Eq. 17, we can relate the membrane mass transport
resistance to the permeability:
rwall
w =
δ
T Pw
∆µw
∆pw
(A.10)
The convective resistance to water transport can be calculated from the convective
mass transport coefficient, kconv, which is defined as [36]:
Jw = −kconv∆cw
(A.11)
where cw is the water molar concentration. For air-water vapor mixtures, the con-
vective mass transport coefficient can be expressed with good accuracy by the Lewis
relation [36]:
kconv =
hconv
cpc
(A.12)
where c and cp are the molar concentration and the molar heat capacity of humid air.
The comparison of Eq. A.11 with Eq. 17 gives:
rconv
w =
1
Tkconv
∆µw
∆cw
(A.13)
Appendix B. Exergy destruction calculations from the entropy balance on the
system
In a steady state, the exergy destruction of a process due to internal irreversibilities
can also be calculated from the exergy balance on the system [31]:
(cid:16)
(cid:104)(cid:16)
(cid:17) −(cid:16)
(cid:17) −(cid:16)
Ed,irr =
=
Eex
in
Hex
in
+ Efr
in
+ Hfr
in
out + Efr
Eex
out
out + Hfr
Hex
out
(cid:17)
(cid:17)(cid:105) − T0
(cid:104)(cid:16)
(cid:17) −(cid:16)
(cid:17)(cid:105)
(B.1)
S ex
in
+ S fr
in
out + S fr
S ex
out
in and Efr
where Eex
in are the exergy streams entering the recovery ventilator with the ex-
haust air stream and the fresh air stream respectively, and Eex
out are the exergy
streams leaving the system with the exhaust air stream and the fresh air stream respec-
tively. Here, H and S are enthalpy and entropy flow respectively.
out and Efr
22
Figure C.1: Exergy efficiency, η, of the MERV as a function of the outdoor temperature, for variation of the
considered input parameter of 20 % (solid lines) and −20 % (dashed lines). The considered input parameters
are duct height (red lines), membrane permeability (blue lines), inlet air velocity (green lines), and convective
heat transport coefficient (light blue lines). The black line represent the results for unvaried input parameters.
Calculations are carried out at T0 = 263 K and RH0 = 60 %.
By considering that, due to energy conservation, the sum of the enthalpy streams
entering the system needs to be equal to the sum of those leaving the system, then Eq.
B.1 reduces to:
(cid:17)(cid:105)
+ Ffr
insfr
in
(B.2)
Ed,irr = T0
= T0
out + S fr
S ex
out
Fex
outsex
out + Ffr
S ex
in
outsfr
out
(cid:17) −(cid:16)
(cid:17)(cid:105)
(cid:17) −(cid:16)
+ S fr
in
Fex
in sex
in
(cid:104)(cid:16)
(cid:104)(cid:16)
With this formulation, Ed,irr can be calculated from the knowledge of flow rates, tem-
perature, composition, and pressure at the inlet and outlet of the system only. There-
fore, we do not need any additional information than the one needed to calculate the
effectiveness parameters according to Eqs. 1-3.
Appendix C. Sensitivity analysis
The sensitivity analysis is carried out with a dual purpose. First, the analysis al-
lows us to test the sensitivity of the results to uncertainties in the input parameters.
Second, it enables us to obtain a deeper understanding of the relationships between
input parameters and results and to possibly reveal any errors in the model, if unex-
pected behaviors should appear. The input parameters considered in the analysis are
duct height, Hex
duct, membrane permeability, Pw, and inlet velocity of the ex-
duct
hausted air, vex
in , and convective heat transport coefficient, hconv, which are varied from
= Hex
23
−20 % to 20 % of their original value reported in Table 1 or calculated according to
Eq. A.5.
Figure C.1 reports the results of the sensitivity analysis for the exergy efficiency.
The variation of the input parameters has only a modest impact on η. The maximum
variation of η at 263 K is −5.2 % (obtained for −20 % variation of hconv), while it is
−9.3 % at 303 K (obtained for 20 % variation of the inlet air velocity).
References
[1] D. Chwieduk, Towards sustainable-energy buildings, Applied Energy 76 (1)
(2003) 211–217.
[2] S. B. Sadineni, S. Madala, R. F. Boehm, Passive building energy savings: A
review of building envelope components, Renewable and Sustainable Energy Re-
views 15 (8) (2011) 3617–3631.
[3] P. J. Walsh, C. S. Dudney, E. D. Copenhaver, Indoor air quality, CRC Press, 1983.
[4] A. P. Jones, Indoor air quality and health, Atmospheric Environment 33 (28)
(1999) 4535–4564.
[5] K. Sakai, D. Norbäck, Y. Mi, E. Shibata, M. Kamijima, T. Yamada, Y. Takeuchi, A
comparison of indoor air pollutants in Japan and Sweden: formaldehyde, nitrogen
dioxide, and chlorinated volatile organic compounds, Environmental Research
94 (1) (2004) 75–85.
[6] HVAC systems and equipment, ASHRAE Handbook (1996).
[7] S. C. Sugarman, HVAC fundamentals, CRC Press, 2005.
[8] P. O. Fanger, Thermal comfort. Analysis and applications in environmental engi-
neering, Danish Technical Press, Copenhagen, Denmark, 1970.
[9] M. Fehrm, W. Reiners, M. Ungemach, Exhaust air heat recovery in buildings,
International Journal of Refrigeration 25 (4) (2002) 439–449.
[10] A. M. Omer, Energy, environment and sustainable development, Renewable and
Sustainable Energy Reviews 12 (9) (2008) 2265–2300.
[11] L. Pérez-Lombard, J. Ortiz, C. Pout, A review on buildings energy consumption
information, Energy and Buildings 40 (3) (2008) 394–398.
[12] K. McCormick, L. Neij, Experience of Policy Instruments for Energy Efficiency
in Buildings in the Nordic Countries, Lund University, 2009.
[13] 62.2, Ventilation and Acceptable Indoor Air Quality in Low-rise Residential
Buildings, ASHRAE Standard (2013).
[14] Byggteknisk forskrift (TEK 10), Direktoratet for Byggkvalitet (2016).
24
[15] W. A. Shurcliff, Air-to-air heat-exchangers for houses, Annual Review of Energy
13 (1) (1988) 1–22.
[16] A. Mardiana-Idayu, S. Riffat, Review on heat recovery technologies for building
applications, Renewable and Sustainable Energy Reviews 16 (2) (2012) 1241–
1255.
[17] S. Kakac, H. Liu, A. Pramuanjaroenkij, Heat exchangers: selection, rating, and
thermal design, CRC Press, 2012.
[18] L. Zhang, Y. Jiang, Heat and mass transfer in a membrane-based energy recovery
ventilator, Journal of Membrane Science 163 (1) (1999) 29–38.
[19] G. Vasilyev, I. A. Tabunshchikov, M. Brodach, V. Leskov, N. Mitrofanova,
N. Timofeev, V. Gornov, G. Esaulov, Modeling moisture condensation in humid
air flow in the course of cooling and heat recovery, Energy and Buildings 112
(2016) 93–100.
[20] A. Dodoo, L. Gustavsson, R. Sathre, Primary energy implications of ventilation
heat recovery in residential buildings, Energy and Buildings 43 (7) (2011) 1566–
1572.
[21] Y. Zhang, Y. Jiang, L. Z. Zhang, Y. Deng, Z. Jin, Analysis of thermal performance
and energy savings of membrane based heat recovery ventilator, Energy 25 (6)
(2000) 515–527.
[22] L. Zhang, J. Niu, Performance comparisons of desiccant wheels for air dehu-
midification and enthalpy recovery, Applied Thermal Engineering 22 (12) (2002)
1347–1367.
[23] M. A. Mandegari, H. Pahlavanzadeh, Introduction of a new definition for effec-
tiveness of desiccant wheels, Energy 34 (6) (2009) 797–803.
[24] L.-Z. Zhang, Heat and mass transfer in a quasi-counter flow membrane-based
total heat exchanger, International Journal of Heat and Mass Transfer 53 (23)
(2010) 5478–5486.
[25] J. Liu, W. Li, J. Liu, B. Wang, Efficiency of energy recovery ventilator with vari-
ous weathers and its energy saving performance in a residential apartment, Energy
and Buildings 42 (1) (2010) 43–49.
[26] J. Min, M. Su, Performance analysis of a membrane-based energy recovery ven-
tilator: Effects of membrane spacing and thickness on the ventilator performance,
Applied Thermal Engineering 30 (8) (2010) 991–997.
[27] M. Nasif, R. Al-Waked, G. Morrison, M. Behnia, Membrane heat exchanger in
hvac energy recovery systems, systems energy analysis, Energy and Buildings
42 (10) (2010) 1833–1840.
25
[28] J. Min, M. Su, Performance analysis of a membrane-based energy recovery ven-
tilator: Effects of outdoor air state, Applied Thermal Engineering 31 (17) (2011)
4036–4043.
[29] S. Dugaria, L. Moro, D. Del Col, Modelling heat and mass transfer in a
membrane-based air-to-air enthalpy exchanger, in: Journal of Physics: Confer-
ence Series, Vol. 655, IOP Publishing, 2015, p. 012035.
[30] P. Liu, M. J. Alonso, H. M. Mathisen, C. Simonson, Energy transfer and energy
saving potentials of air-to-air membrane energy exchanger for ventilation in cold
climates, Energy and Buildings 135 (2017) 95–108.
[31] T. J. Kotas, The exergy method of thermal plant analysis, Elsevier, London, 2013.
[32] N. Lior, N. Zhang, Energy, exergy, and second law performance criteria, Energy
32 (4) (2007) 281–296.
[33] Exergy Analysis for Sustainable Buildings, ASHRAE Technical Committee 7.4.
URL https://tc0704.ashraetcs.org/
[34] K.-C. Noh, J. Hwang, The effect of ventilation rate and filter performance on
indoor particle concentration and fan power consumption in a residential housing
unit, Indoor and Built Environment.
[35] J. Woods, E. Kozubal, Heat transfer and pressure drop in spacer-filled channels
for membrane energy recovery ventilators, Applied Thermal Engineering 50 (1)
(2013) 868–876.
[36] Y. A. Cengel, A. J. Ghajar, H. Ma, Heat and Mass Transfer: Fundamentals &
Applications, McGraw-Hill, New York, 2011.
[37] S. Patankar, Numerical heat transfer and fluid flow, CRC press, 1980.
[38] R. S. Davis, Equation for the determination of the density of moist air (1981/91),
Metrologia 29 (1) (1992) 67.
[39] P. Brimblecombe, Air composition and chemistry, Cambridge University Press,
1996.
[40] S. Kjelstrup, D. Bedeaux, Non-equilibrium thermodynamics of heterogeneous
systems, Vol. 16, World Scientific, Singapore, 2008.
[41] G. Gouy, Sur l'énergie utilisable, Journal de Physique 8 (1889) 501–518.
[42] R. N. Huizing, F. K. Ko, Selective water vapour transport membranes compris-
ing a nanofibrous layer and methods for making the same, US Patent 8,936,668
(Jan. 20 2015).
[43] L. F. Shampine, J. Kierzenka, M. W. Reichelt, Solving boundary value problems
for ordinary differential equations in MATLAB with bvp4c, Tutorial Notes (2000)
437–448.
26
[44] W. M. Kays, A. L. London, Compact heat exchangers, McGraw-Hill, New York,
NY, 1984.
[45] L. Zhang, J. Niu, Effectiveness correlations for heat and moisture transfer pro-
cesses in an enthalpy exchanger with membrane cores, Transactions-American
Society of Mechanical Engineers Journal of Heat Transfer 124 (5) (2002) 922–
929.
27
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.