of 25
Copyright WILEY-VCH Verlag GmbH & Co. KGaA, 69469 Weinheim, Germany, 2017.
Supporting Information
for
Adv. Mater.,
DOI: 10.1002/adma.201701044
Millivolt Modulation of Plasmonic Metasurface Optical
Response via Ionic Conductance
Krishnan Thyagarajan
, Ruzan Sokhoyan
, Leonardo Zornberg
,
and
Harry A. Atwater
*
1
Supporting Information
Article type: Communication
Millivolt Modulation of Plasmonic Metasurface Optical Response via Ionic
Conductance
Krish
nan Thyagarajan, Ruzan Sokhoyan
, Leonardo Zornber
g,
and Harry A. Atwater
*
Dr. K. Thyagarajan, Dr. R. Sokhoyan,
L. Zornberg, Prof. H. A. Atwater
Thomas J. Watson Laboratory of Applied Physics,
California Institute of Technology,
Pasadena, California 91125, USA
E
-
mail:
haa@caltech.edu
Dr. K. Thyagarajan, Prof. H. A. Atwater
Kavli Nanoscience Institute,
California I
nstitute of Technology,
Pasadena, California 91125, USA
Keywords:
Tunable metasurface, filament formation, ionic transport, memristor, indium tin
oxide (ITO)
2
S1.
Comparative analysis of previously reported tunable metasurfaces
In what follows we review the details of previously reported active metasurfaces and explain
challenges associated with obtaining tunable optical response in the visible wavelength range.
Field effect in semiconductors
In this scheme, each metasurface el
ement is effectively a metal
-
insulator
-
semiconductor
(MIS) capacitor with the metal serving as a gate and the semiconductor functioning as a field
effect channel
[1]
. When an electrical bias is applied between metal gate and semiconductor, a
charge depletio
n or accumulation layer is formed in the semiconductor at the interface of the
semiconductor and insulator. This results in modulation of the semiconductor dielectric
permittivity, altering the optical response of a metasurface.
From ellipsometry measure
ments it is known that the dielectric permittivity of doped
semiconductors can be typically described by a Drude
-
Lorentz model which is sum of a
Drude term and Lorentz oscillators. It is typically assumed that the Lorentz oscillators do not
depend on field
or voltage. On the other hand, the Drude term reads
ε
=
ε
-
ω
2
p
/(
ω
2
+
iωΓ
)
where
ω
p
is the plasma frequency which is related to the carrier density
N
and electron
effective mass
m
*
as
ω
p
2
=
Ne
2
/(
ε
0
m
*
).
Here
ε
0
is the dielectric permittivity of vacuum, e is the
electron charge, and
Γ
is the damping constant. Thus, the Drude term increases linearly with
carrier concentration. For a given change of semiconductor carrier density, the variation of
the complex dielec
tric permittivity will be higher at lower frequencies, or, equivalently, at
longer wavelengths. This constitutes one of the reasons why field effect modulation of the
optical response of a metasurface is challenging in the visible wavelength range.
When
studying tunable metasurfaces with GaAs or indium tin oxide (ITO) field
-
effect
channels, it has been shown that the epsilon
-
near
-
zero (ENZ) transition of a thin
semiconductor accumulation or depletion layer yields large optical modulation. At the ENZ
trans
ition, the real part of the dielectric permittivity of a thin semiconductor layer changes its
sign, undergoing a transition from an optically dielectric to an optically metallic phase. The
wavelength at which the bias
-
induced ENZ transition may occur is pr
imarily defined by the
semiconductor carrier concentration. For example, the ENZ transition of GaAs with carrier
concentration of 5.5×10
18
cm
-
3
is in the mid
-
infrared wavelength range. On the other hand,
the ENZ transition in the accumulation layer of ITO
with a carrier concentration of 3×10
20
3
cm
-
3
occurs at a wavelength of 1.5 μm. In principle, one could incorporate into a metasurface
a highly doped semiconductor such that ENZ wavelength of its accumulation layer lies in the
visible. However to date, field
-
effect tunable metasurfaces have not been demonstrated in the
visible.
One of the major advantages of this approach is its technological maturity. Indeed,
contemporary low power integrated electronic circuit operation is based on field
-
effect
dynamics in
semiconductors. Another important advantage is high modulation frequency (or,
equivalently, small response time). For example, for a device with area of 40
×
40 μm, a
response time of 50 ns has been experimentally demonstrated.
A disadvantage of this appr
oach is that complex refractive index of a semiconductor is varied
only in an ultrathin layer. For example, for ITO with carrier concentration 3×10
20
cm
-
3
, the
thickness of the modulated layer is around 1 nm. This implies that to observe tunable optical
re
sponse of a metasurface, one needs extreme light concentration in this ultrathin material
layer which is typically achieved by using plasmonic structures. As a result, field effect
tunable metasurfaces exhibit strong absorption in plasmonic materials and s
emiconductor.
Graphene
g
ating
Tunable metasurfaces based on graphene gating have also been experimentally
demonstrated
[2, 3]
. The dielectric permittivity of graphene is proportional to the square root of
the carrier density in graphene and is inversely
proportional to the frequency squared. As it is
the case with field effect in semiconductors, here also modulation of dielectric permittivity of
graphene is larger at longer wavelengths. So far graphene based gated metasurfaces have
shown tunable response
in the mid
-
infrared wavelength range. In principle, operation
wavelength can be reduced by using high
-
k dielectrics to gate graphene. However, so far this
has not been experimentally demonstrated.
Similar to the case of field effect in semiconductors,
for observation of graphene based
tunable metasurfaces one needs to concentrate optical field in a graphene sheet. This extreme
field confinement in a graphene sheet is typically associated with increased optical losses. An
important advantage of graphene
metasurfaces is small response time. For example, response
time <50 ns has
been experimental
ly reported
[3]
.
4
Reorientation of Liquid Crystal Molecules
Active tuning of optical response of metasurfaces loaded with liquid crystal molecules has
been previousl
y demonstrated in the near
-
infrared waveleng
th range
[4]
. One of the major
drawbacks of liquid crystal loaded metasurfaces is anchoring of liquid crystal molecules to
the surfaces of the structures. Since the thickness of liquid crystal layer affected by a
nchoring
is comparable to the extension of the metasurface optical near fields, this will result in
reduced modulation upon application of external stimuli. When the metasurface feature size
is further reduced so that the operation wavelength of the metasu
rface is in the visible,
adverse effects of liquid crystal anchoring become even more severe. We believe this is the
major reason why liquid crystal loaded tunable metasurfaces have not been demonstrated in
the visible. The advantage of this approach is br
oadband refractive index modulation, and 1
ms response time.
Phase
c
hange
m
aterials
A tunable optical response of a metasurface has also been demonstrated via refractive index
modulation of vanadium dioxide (VO
2
) upon heating
[5]
. Heating induces a metal
-
insulator
transition in VO
2
. This has been shown to significantly alter the optical response of the
metasurface in the wavelength range from 1.5 μm to 5 μm
[5]
. Phase transition in GeSbTe
(GST) has also been used to demonstrate a tunable metasurface in the
near
-
infrared
wavelength range λ>1 μm
[6]
. In this case the phase change in GST was induced either via
heating the sample by placing it on a hot plate or via short high intensity laser pulses that
locally raise the temperature of a phase change material. It
is worth mentioning that refractive
index modulation of VO
2
and GST is relatively modest in the visible wavelength range as
compared to that in the near
-
infrared range. This is one of the reasons why tunable
metasurfaces incorporating phase change materia
ls show a tunable optical response in near
-
infrared wavelength range. Another reason is that operation in the visible implies small
resonator sizes resulting in higher localization of optical fields. This local heating could
interfere with phase change dyn
amics of GST or VO
2
thus hindering deterministic
performance of the device.
5
S
2
.
Reflectance from
planar structure
s
We calculate
reflectance
of
the planar stack shown in Figure S
1
a
. In our calculations we use
ellipsometrically measured dielectric perm
ittivity of ITO
,
while
for the dielectric permittivity
of alumina
we use
tabulated
Palik data
[7]
. First, we assume that
the
dielectric permittivity of
silver can also be described by using
tabulated
Palik data
[7]
.
In this case
, for the considered
wavelengt
h range, calculated reflectance varies between 75
%
and 87
%. However, measured
reflectance is as low as 30% for wavelengths between 650 and 800 nm
(see Figure 1)
. To
identify the reason behind this discrepancy, we fabricated a control sample of 80 nm Ag on Si
substrate and performed spectroscopic ellipsometry
measurement
every day for a week.
Indeed, we observed temporal evolution of the dielectric permittivity of
silver. Next, by using
Figure S1.
a) Schematic of the simulated structure; b)
Calculated reflectance from the planar
structure shown in
(a)
. The legend indicates the dielectric permittivity used to model silver. In
case of Ag
Palik the tabulated Pali
k data is used. Ag
-
day1 and Ag
-
day3 correspond to the
ellipsometrically m
easured silver data 1 day and
3 days after deposition, correspondingly.
the reflectance from the structure (see Figure
S
1b)
. However, this variation is not substantial
enough, to exp
lain the discrepancy between theory and experiment.
6
spectroscopically measured dielectric permittivity values of silver, we calculated the
reflectance from the planar heterostructure (Ag/Al
2
O
3
/ITO). We observe a minor variation
of
S
3
.
Transmission electr
on microscopy
(TEM)
images and
energy
-
dispersive X
-
ray
spectroscopy
(EDS)
of modulated films
Transmission electron microscopy (TEM) images of the flat film configurations were made
after the sample was exposed to repeated recycling at large voltages (2V).
A.
Bare silicon substrate:
To compare the ene
rgy dispersive spectroscopy (EDS
) signal across various regions of the
sample, the bare silicon substrate was measured.
Figure S
2
A: EDS signal from bare silicon substrate showing a clear peak at around
1.8 keV
B.
Indium tin oxide interface:
7
Figure
S
2
B:
EDS signal from the ITO layer. The
m
olybdenum is from the TEM grid.
C. Silver and silicon interface:
Figure S
2
C: The EDS signal from the interface of the silver and silicon clearly shows the
presence of t
wo elements. The small peak at higher energies is from the
m
olybdenum present
in the TEM sample grid.
D. Silver polycrystalline growth from the ITO alumina interface
:
8
Figure S
2
D1: TEM image of the interface of the delaminated ITO surface showing the
growth of the silver clusters from the interface downwards.
Figure S
2
D2: EDS analysis of the cluster and ITO interface shows the presence of indium, tin
and silver in the region,
indicating the existence of silver clusters.
S
4
.
TEM
images of
Si/Cr/Ag/Al
2
O
3
/ITO planar heterostructure
Epoxy
ITO
Delaminated region
Silver
growth
9
We have performed an additional TEM analysis of
Si/Cr/Ag/Al
2
O
3
/ITO planar
heterostructure
.
This was done to ensure the smoothness/continuous nature of the very thin
chromium layer. As shown below in the TEM images, the chromium layer is rather uniform
with an average thickness of around 1.2 nm, compared to the expected 1 nm. The silver
deposite
d using electron beam evaporation can clearly be seen to be polycrystalline.
Figure S
3
.
TEM
images of
Si/Cr/Ag/Al
2
O
3
/ITO planar heterostructu
re
(a)
-
(d). The
continuous layer of Cr can be seen across the sample. The scale bar in (d) corresponds to 1
nm. E
vidently the silver appears to be polycrystalline.
S
5
.
EDS of aged dolmen samples
We also undertook
EDS
analysis
of planar structures discussed in Section S
3
.
The presence
of the expected elements was thus confirmed
(Figure S4)
.
10
Figure
S
4
.
EDS
measurements on the above mentioned samples clearly showed the presence
of the expected elements
.
The molybdenum signal is from the TEM grids used during sample
preparation.
S
6
.
Energy dispersive spectroscopy
maps
of the dolmen structures
EDS
was underta
ken on the inverse dolmen samples with the chromium adhesion layer.
Chemical maps derived from energy dispersive X
-
ray spectroscopy detailing the elements
present in our samples were developed, and indicate the distribution of each element depicted
by the
color rendering.
As can be clearly seen in
F
igure S
5
,
there
is
sulphur and chlorine
contamination in the samples.
Figure
S
5
.
EDS
undertaken on the inverse dolmen samples showed the presence of
contaminants like sulphur and chlorine apart from the expected chemical elements.
Their
presence is hypothesized to improve the optical modulation with such a small applied bias.
11
Figure S6
. Energy dispersive X
-
ray spectroscopy undertaken on the fresh inverse dolmen
samples did not show the presence of contaminants like sulphur and chlorine. These
measurements were taken immediately after the sample fabrication. Their absence helps
explain w
hy the absolute magnitude of modulation is greater for an aged sample.
S
7
.
Simulating Reflectance and Transmittance of the Dolmen Metasurfaces
Figure S7
. Schematic of the unit cell of the simulated dolmen metasurface. (a) Top view; (b)
Cross
-
section of
the metasurface. Cross
-
section is taken normally to the surface of the
structure, along the dashed line shown in (a).
12
We use finite difference time domain (FDTD) method to simulate optical response of the
metasurface. In our simulation we illuminate struc
ture by plane wave and use periodic
boundary conditions with period equal to 400 nm. Schematic of the top view of the simulated
structure is shown in Figure S
7
(a). The three rectangles indicate the dolmen structure which
has been milled into silver via FIB. Dark grey frame of the rectangles is aimed to emphasize
5 nm thick alumina layer deposited via atomic layer deposition (even though alumina covers
whole dol
men structure). The dimensions
of the dolmen structures
shown in Figure S
7
(a)
indicate
the sizes of the rectangles milled in the silver film. This is schematically shown by
extending arrows also over alumina layer. Figure S
7
(b)
shows vertical cross
-
secti
on taken
along the dashed line shown in
F
igure S
7
(a). The considered cross
-
section is normal to the
metasurface interface.
As one can see from
Figure S
7
(b)
, the thickness of the silver film is
80 nm.
In our simulation we took into account that each hole
milled in the silver via FIB is
not a rectangular parallel piped but rather a truncated pyramid
with rectangular bases
.
Length
of the top base of each truncated pyramid is 30 nm longer than length of the bottom base.
Similarly, width of the top base of ea
ch truncated pyramid is 30 nm longer as compared to the
width of the bottom base of each truncated pyrami
d. As one can see from Figures 3
(d),
(f)
and (h)
, the simulated reflectance and transmittance spectra match optical measurement
results obtained from
fresh dolmen samples under no applied bias.
Finally, we note that silver
was modelled by using tabulated Palik data
,
[
7
]
while dielectric permittivity of ITO was
obtained from ellipsometric measurement of 110 nm thick ITO film deposited on Si substrate.
13
Figure
S8
: Simulated and measured reflectance, transmittance, and absorptance spect
ra of
fresh dolmen metasurface.
Yellow, purple, and red circles mark three wavelengths at which
optical field profiles are displayed (see below).
Next we discuss spatial
distribution of optical fields supported by a dolmen metasurface. We
plot optical fields at three different wavelengths which are marked by yellow,
purple, and red
circles in Figure
S8
. Schematic of the top view of the simulate
d structure is shown in Figur
e
S7
(a). Three rectangles indicate the dolmen structure which has been milled into silver via
FIB. Dark grey frame of the rectangles is aimed to emphasize 5 nm thick alumina layer
deposited via atomic layer deposition (even though alumina covers whole dol
men structure).
The dimensions of the dolmen structures shown in Figure
S7
(a) indicate the sizes of the
rectangles milled in the silver film. This is schematically shown by extending arrows a
lso
over alumina layer. Figure S7
(b) shows vertical cross
-
secti
on taken along the
vertical
dashed
line shown in Figure
S7
(a). The considered cross
-
section is normal to the metasurface
interface.
Figure S9
shows spatial distribution of optical fields along the metasurfa
ce cross
-
section
shown in Fig
ure
S7
(
b
)
. At λ =
596 nm and λ = 760 nm, we can see some hotspots on the top
corners of the silver structures. One can see that although the hotspots are at similar locations
and are of similar magnitudes at λ = 596 nm and λ = 760 nm, the latter demonstrates a larger
volume
over which the field is high (including the ITO region). We also see larger optical
fields at the bottom of the structure at λ = 760 nm as compared to that at λ = 596 nm and λ =
670 nm. This could explain enhanced transmission observed at λ = 760 nm.
Next
, we plot field profiles plotted
(Figure S10)
along the cross
-
section of the structure taken
normally to the surface of the image, along the
horizontal
dashed line shown in Figure S7(a)
.
14
At λ= 596 nm, we observe significant electric field enhancement at Ag
interface but there is
no major
E
-
field build up inside the structure. Note that at λ= 596 nm the reflectance of the
incident plane wave is high. At λ=670 nm and λ = 760 nm, one can see a finite E
-
field build
up in the ITO region of the structure, thus l
eading to larger losses due to the absorptive nature
of ITO. It is also the region in which silver nucleation is taking place. Hence, a significant
optical modulation is observed at wavelengths of λ=670 nm and λ = 760 nm. At λ = 760 nm
the value of optical
is higher at the bottom of dolmen structure. Since the field has now
reached the bottom of dolmen structure, we observe relatively high transmission at λ = 760
nm as compared to the case of λ = 670 nm. Finally, other simulations done by us (not shown
here
) indicate that there is a finite coupling between optical fields supported by individual
rods of a single inverse dolmen structure.
Fig
ure S9
: Field profiles plotted along the metasurface cross
-
section shown in Figure 3b.
15
Figure
S
10
. Field profiles p
lotted along the cross
-
section of the metasurface unit cell. The
cross
-
section is taken normally to the surface of the structure, along the
horizontal
dashed
line shown in Figure S7
(
a
)
.
S
8
. Reflectance from aged dolmen structures over larger voltage
range
Figure S
11
plots reflectance and transmittance from aged dolmen samples as a function of
applied bias at wavelength of
= 625 nm.
For biases up to 6 mV, reflectance shows
monotonic increase while transmittance shows monotonic decrease.
Reflectance
and
transmittance spectra modulation in this bias range is shown in Figure 3 of the main
m
anuscript. As one can see from F
igure S
11
, upon increase of applied voltage beyond 6 mV,
reflectance shows abrupt decrease, while transmittance shows abrupt increase.
The data
depicted in Figure
S
11
is an averag
e over nine consecutive optical measurements performed
on the device. The average absolute reflectance modulation upon application of 100 mV is
70.8% while the record absolute change of reflectance obtained from
a single measurement is
78%.
16
Figure S
11
. Measured (a) reflectance and (b) transmittance from the aged dolmen structures
at
= 625 nm, showing a modulation of up to
70.8 %
in reflectance and 25.3 % in
transmittance.
S
9
. Influence of
continuity of
the bottom electrode: Cr adhesion layer
We have also fabricated a set of inverse dolmen metasurface samples that do not contain thin
metal film between bottom alumina and substrate. As it has been described in the main text of
the manuscript, the inverse d
olmen structure was written in 80 nm thin Ag film via
FIB. To
17
Figure S
12
. Measured reflectance (a) and (b) and transmittance (c) and (d) spectra with and
without the chromium adhesion layer respectively showing the dramatic influence of the
presence of c
hromium as an adhesion layer to permit the ultralow power millivolt level
optical modulation.
o
btain the samples without thin metal layer between bottom alumina and substrate, the FIB
milling was interrupted only when whole bottom Ag and Cr layers were re
moved and
substrate was reached. Interestingly, optical measurements performed on this set of samples
showed negligible o
ptical modulation (see Figure S
12). Figures S12
(
a
)
and S
12
(
c
)
show
optical measurement results for samples with bottom electrode laye
r (also discussed in the
main text
of
the manuscript) while Figures S12
(
b
)
and S
12
(
d
)
indicate optical measurements
performed on
the samples that have been milled till reaching the substrate.
The multiple
curves show the measured reflect
ance
when
biasing
silver positively with respect to ITO
. The