This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2020.2979192, Journal of
Lightwave Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1

High-Speed Plasmonic-Silicon Modulator

Driven by Epsilon-near-zero Conductive Oxide
Bokun Zhou, Erwen Li, Yunfei Bo, and Alan X. Wang*

conventional silicon photonic devices by exploiting the strong

Abstract—Transparent conductive oxides (TCOs) such as Pockels effect [10]–[13] or the plasma dispersion effect [14],
indium-tin oxide (ITO) have attracted increasing interests [15] of the integrated active materials.
in integrated photonics and silicon photonics, owing to their In recent years, transparent conductive oxides (TCOs), such
large plasma dispersion and epsilon-near-zero (ENZ) effect. as indium-tin oxide (ITO), indium oxide (In2O3) and cadmium
The nonlinear change of refractive index induced by free oxide (CdO), have attracted increasing interests owing to the
carrier modulation leads to a large electro-optic modulation large plasma dispersion effect and epsilon-near-zero (ENZ)
with ultra-compact device footprint and unprecedented effect[16]–[18]. Due to the dramatic change of free carrier
energy efficiency. However, high-speed modulation result is concentration, TCOs can be electrically tuned from dielectric-
rare, mainly due to the lack of high-speed device design and like to metallic. During such transition, the absolute
fabrication quality. In this article, we characterize the permittivity will reach a minimum value with real part crosses
fundamental electro-optic modulation structure consisting zero, which is described as the ENZ effect[19]–[21]. When
of Au/ITO/oxide/p-Si capacitor, showing that the property ENZ occurs, light will be strongly confined in the ENZ layer
of ITO is greatly affected by the process condition. We also due to the continuity of electric displacement normal to the
report an 8-µm-long hybrid plasmonic-silicon modulator interface, inducing strong electro-absorption (EA) of light over
driven by an ENZ ITO capacitor, achieving 100fJ/bit a broad optical bandwidth. To date, TCO-based EA modulators,
energy efficiency, 3.5 GHz modulation bandwidth and 4.5 such as plasMOStor [22] and plasmonic metal-oxide-
Gb/s data rate. The electro-absorption modulator covers a semiconductor (MOS) waveguide modulator [23]-[24] have
broad optical bandwidth from 1515 to 1580 nm wavelength. been demonstrated to achieve both large optical bandwidth and
For future development of such modulators, we point out small device footprint, showing great potential for future
that by replacing ITO with higher mobility TCO materials, integrated optical interconnect systems. However, only a
we can achieve 40 GHz modulation bandwidth and 0.4 fJ/bit moderate [25] digital modulation rate of 2.5 Gb/s was reported
energy efficiency using a 3-µm-long device. due to the large capacitance of the metal-oxide-semiconductor
(MOS) capacitor and potential challenges of depositing high
Index Terms—Electro absorption modulator, epsilon- quality TCO materials.
near-zero material, plasmonic modulator, silicon photonics, In this work, we first characterize various Au/ITO/oxide/Si
transparent conductive oxide. MOS capacitors, showing that TCO process conditions, such as
low temperature baking and O2 plasma treatment, can greatly
I. INTRODUCTION affect properties of these MOS capacitors. Next, we design and
demonstrate a hybrid plasmonic-silicon modulator driven by
Plasmonics opens a new realm for ultra-compact, high-speed, ITO. Through optimizing the device layout and adjusting the
and energy-efficient photonic devices that can transform next gate layer thickness, the capacitance is reduced to 100 fF while
generation on-chip optical interconnect systems [1]–[3],
the series resistance remains small at 500Ω. The 8-µm-long EA
owning to its ultra-strong optical confinement even below the
modulator reaches 3.2 dB extinction ratio (ER) with only 2V
diffraction limit [4]. Since traditional plasmonic materials such
voltage swing and 100 fJ/bit energy efficiency. More
as gold and silver cannot provide the necessary electro-optic
importantly, it achieves a high modulation bandwidth of 3.5
effect for active control of light and suffer high optical loss,
GHz and digital modulation rates to 4.5 Gb/s. To further
various hybrid plasmonic-silicon photonic devices have been
improve the device performance, we prove through numerical
reported through the integration with active materials such as
simulation by integrating higher mobility TCOs. A 3-µm-long
graphene [5], [6], electro-optic polymer[7], [8], and phase
device can achieve a record-breaking performance with
change materials [9]. Such hybrid plasmonic-silicon photonic
extremely small voltage swing of 0.2V, unprecedented energy
devices take advantages of the matured silicon photonics
efficiency of 0.4 fJ/bit, and a high modulation bandwidth of 40
platform for easy coupling, low optical loss and large-scale
GHz.
integration. In the meanwhile, they can achieve higher electro-
optic modulation speed and higher energy efficiency than

This work is supported by the AFOSR MURI project FA9550-17-1-0071 Science, Oregon State University, Corvallis, OR 97331 USA (e-mail:
and NSF GOALI grant 1927271. [email protected]; [email protected]; [email protected];
The authors are with the School of Electrical Engineering and Computer *[email protected]).

0733-8724 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: OREGON STATE UNIV. Downloaded on May 03,2020 at 00:41:14 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2020.2979192, Journal of
Lightwave Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

II. CHARACTERIZATION OF AU/ITO/OXIDE/P-SI MOS

CAPACITOR
We first fabricated and characterized the fundamental
electro-optic modulation structure of the EA modulator
consisting of Au/ITO/oxide/p-Si MOS capacitor. The layered
structure of the MOS capacitor is shown in Fig. 1 (a). 10 nm of
SiO2 gate oxide is thermally oxidized from the bottom p-Si. On
top of that is the 15 nm ITO/Au gate. The electrical properties
of ITO depend on the deposition and post-processing conditions
[26]. The free carrier concentration can be tuned within the
range of 1×1019 to 1×1021 cm-3 by controlling the oxygen/argon
ratio of the gas flow during the RF sputtering, whereas the
mobility is in the range of 15 to 25 cm2V-1s-1, which are cross-
checked using ellipsometry and Hall measurement. After the
deposition, post annealing further increases the conductivity of
ITO due to increase of mobility caused by crystallization, which
has been systematically studied [27]. In addition, the ITO
conductivity can also be increased by O2 plasma treatment due
to the injection of oxygen vacancies at the ITO surface[28].
This effect is especially significant for thin ITO films. For
example, the carrier concentration of a 15 nm thick ITO film
can be increased from 8×1019 to 4×1020 cm-3 by five-minute O2
plasma treatment at 200 mTorr pressure. To be mentioned, the Fig. 1. (a) Cross sectional schematic of the Au/ITO/SiO2/Si capacitor. (b)
effect of O2 plasma is reversible to some extent. The carrier Leakage current density as a function of gate voltage for capacitors at different
conditions. (c) Capacitance density as a function of frequency for capacitors at
concentration can be reduced from 4×1020 to 1×1020 cm-3 by 5 different conditions.
minutes of 180°C low temperature baking in air, which is
common for some lithography process due to the reabsorption performances match well with the theoretical calculation, the
of oxygen in air. speed of which is only limited by the RC delay. However, for
We also compared MOS capacitors with two different ITO low conductivity TCO-Si MOS capacitor, both the leakage
layers: high conductivity ITO (green dot, observed after O2 current and capacitance density suggest that there is more
plasma) and low conductivity ITO (red triangle, observed after complex mechanism, which requires further investigation.
low temperature baking). Fig. 1(b) plots the leakage current
density at different gate voltages. The leakage current density III. DESIGN AND PRINCIPLE
of high-conductivity ITO capacitor shows a good match with
the theoretical calculation based on Fowler-Nordheim The 3D schematic of the plasmonic ENZ EA modulator is
tunneling [29]. Fig. 1(c) shows the AC capacitance density as a shown in Fig. 2(a). The modulator is based on an
function of frequency at 5V bias voltage. The capacitance Au/ITO/HfO2/p-Si MOS capacitor in its active region. As
density of high conductivity ITO at low frequency matches the illustrated in Fig. 2(b), the core of the p-type silicon rib
static capacitance density of 5×10-7 F/cm2, which was simulated waveguide is 250 nm in height and 450 nm in width, surrounded
by SilvacoTM using the quantum moment model [30]. At higher by 50 nm thick partially etched slab providing electrical
frequency, the measured capacitance density is limited by the conduction path. The waveguide is covered by 16 nm HfO2
RC delay, which is calculated to be around 1 MHz due to the working as the gate insulator. On the top of that is the gate
large capacitor area and high Si resistance. However, dramatic electrode consisting of 14 nm ITO and 100 nm Au film. Fig.
difference is observed for low-conductivity ITO. The leakage 2(c) shows the device layout in the active region consisting of
current density of low conductivity ITO capacitor is one order high-speed ground-signal-ground (GSG) electrodes, which will
of magnitude smaller. Even at low frequency, the capacitance be discussed later.
density is already more than one order of magnitude smaller We simulated the electro-absorption of the modulator. First,
than the static value. The capacitance density keeps decreasing carrier distribution in the accumulation layer is simulated by
as the frequency increases until reaching a flat region at which SilvacoTM based on quantum moment model. The quantum
the capacitance is 3 orders of magnitude smaller than the static moment model is proven to be more accurate in describing the
value at ~104 Hz. Such frequency response cannot be simply carrier distribution of accumulation layer in a MOS capacitor
explained by the RC delay, indicating the existence of carrier with an ultra-thin insulator, which is critical to predict the onset
transport processes for low conductivity ITO, which is not yet of the ENZ field confinement. A thorough discussion can be
fully understood. found in ref. [30]. Next, the carrier distribution is imported into
To conclude, the properties of TCO thin films highly depend Lumerical MODE for finite difference eigenmode simulation.
on the process conditions, such as low temperature baking and The optical properties of ITO and Si can be described by the
O2 plasma treatment. To minimize the effect from post-steps on well-established Drude model [31]:
2
𝜔𝜔𝑝𝑝 2 𝛾𝛾
𝜔𝜔𝑝𝑝
ITO properties, we optimized the process flow in our 𝜖𝜖𝑟𝑟 = 𝜖𝜖 ′ + 𝑖𝑖𝜖𝜖 ′′ = 𝜖𝜖∞ − + 𝑖𝑖 (1)
𝜔𝜔 +𝛾𝛾2
2 𝜔𝜔(𝜔𝜔2 +𝛾𝛾2 )
fabrication. For high conductivity TCO-Si MOS capacitor, the

0733-8724 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: OREGON STATE UNIV. Downloaded on May 03,2020 at 00:41:14 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2020.2979192, Journal of
Lightwave Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

accumulation of free carriers induces light absorption. The

changes of free carrier concentration distribution and refractive
index distribution inside ITO layer are shown in Fig. 3(b).
Normal to the HfO2/ITO interface, the major change occurs
within the 1nm-thick accumulation layer. Fig. 3(c) compares
the zoomed-in mode profile at the waveguide side-interface at
0V and -8V bias. Under the same linear scale, it is clearly shown

Fig. 2. (a) 3D schematic of the hybrid plasmonic-TCO-silicon EA modulator.

(b) Cross sectional and (c) top view schematic of the EA modulator. Core
waveguide is 450 nm in width and 250 nm in height, with slab of 50 nm in
height.

Here, 𝜖𝜖∞ is the high-frequency dielectric constant. 𝜔𝜔𝑝𝑝 is the

plasma frequency, which is proportional to the carrier
𝑁𝑁𝑐𝑐 𝑞𝑞 2
concentration 𝑁𝑁𝑐𝑐 by 𝜔𝜔𝑝𝑝 = , where 𝑞𝑞 is charge of electron.
𝜖𝜖0 𝑚𝑚 ∗
𝜖𝜖0 is permittivity of vacuum and 𝑚𝑚∗ is the effective mass of the
carrier. The plasma collision frequency 𝛾𝛾 is inversely
𝑚𝑚∗
proportional to the carrier mobility 𝜇𝜇 by γ = . The change of
𝑞𝑞𝑞𝑞
carrier concentration results in the change of both real and
imaginary part of the permittivity. Due to the small high-
frequency dielectric constant and large carrier concentration,
the real part permittivity of ITO can be reduced to zero and the
absolute value of the permittivity reaches the minimum, which
is the so-called ENZ state.
Fig. 3(a) shows the hybrid plasmonic-silicon waveguide
mode profile of the transverse-electric (TE) mode at 0V bias,
representing a low loss state (ON state). The mode profile is Fig. 3. (a) Simulated mode profile at 0V bias in the active region using
plotted in log scale to show the electric filed intensity both in Lumerical MODE. (b) Carrier concentration distribution and refractive index
the center and in the ITO layer. The propagation loss of the distribution in the ITO accumulation layer, normal to the HfO2/ITO interface,
as a function of applied bias from 0V to -9V. (c) Enlarged view at the HfO2/ITO
device is 1.6 dB/µm in simulation. Upon negative biasing the interface, showing field in accumulation layer at 0V and -8 V bias. Light is
MOS capacitor, electrons and holes accumulate at the strongly confined in the 1 nm accumulation layer. (d) Simulated transmission
ITO/HfO2 and Si/HfO2 interface, respectively. The and modulation strength for waveguide width of 450 nm

0733-8724 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: OREGON STATE UNIV. Downloaded on May 03,2020 at 00:41:14 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2020.2979192, Journal of
Lightwave Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4

that at 0V bias, the electric field is relatively uniform in the ITO

layer, whereas at -8V bias, the electric field is strongly confined
in the accumulation layer at the HfO2/ITO interface owning to
the continuity of electric displacement. Fig. 3(d) plots the
extinction ratio (ER) and modulation strength as a function of
the applied bias. Clearly, there are two stages of operation. At
small bias, the extinction ratio increases almost linearly versus
the applied bias because the loss induced by free carrier
absorption is proportional to the concentration of free carriers.
At large bias, when the ITO accumulation layer reaches the
ENZ state, the absorption is further enhanced due to the optical
field confinement. The ER increases to 1.2 dB/µm at -8V bias.
According to the carrier distribution simulation, the peak carrier
concentration of the ITO accumulation layer reaches ENZ
carrier concentration of 6.5×1020cm-3 at -6.5V, indicating the
onset of ENZ confinement. Theoretically, by biasing the
modulator at the ENZ region, we can modulator the device with
a much smaller voltage swing. From the simulation, the
modulation strength peaks around -6.5V bias, reaching 0.28
dB/(V⋅µm). For example, this corresponds to a voltage swing
of Vpp=1.4 V for 3-dB ER in an 8µm-long EA modulator.
The bandwidth of the hybrid plasmonic-silicon EA
modulator is primarily limited by the RC delay, which can be
1
estimated by 𝑓𝑓 = , due to the accumulation mode operation.
2𝜋𝜋𝜋𝜋𝜋𝜋
In order to achieve high speed operation, we improved our Fig. 4. (a) Scanning electron micrograph (SEM) of the fabricated 8-µm-long
device design in two aspects compared with our previous EA modulator with false color. Enlarged (b) top view and (c) tilted view of
active region.
demonstration [32]. First, the total capacitance is reduced by
minimizing the gate electrode overlapping with the bottom
electrode is patterned by optical photolithography to form
silicon layer. Fig. 2(c) shows part of the device layout. Based
Ohmic contacts. The fabricated device is shown in Fig. 4(a). In
on the capacitance density of 8 fF/µm2 from the simulation, the
Fig. 4(b), the zoomed-in view shows the active region of the EA
total capacitance is around 100 fF. Second, because the ITO is
modulator. The partially etched slab is shown in the tiled view
covered by Au film, the series resistance mainly comes from
in Fig. 4(c).
the Si conduction path. To reduce the series resistance, the top
50 nm of the silicon layer is heavily doped to 1×1020 cm-3. The B. Testing of the Plasmonic ENZ EA modulator
series resistance can be reduced to <100Ω, which will give us a To test the EA modulator, light is coupled into and out of the
modulation bandwidth over 15GHz in design for 8µm-long EA silicon waveguide through the waveguide grating couplers. The
modulator. input/output fibers are mounted at a tilted angle of 10°. A
polarization controller is used at the input side to excite TE
IV. FABRICATION AND CHARACTERIZATION mode. The insertion loss (IL) of the hybrid plasmonic-silicon
A. Fabrication Processes of the EA modulator waveguide is estimated to be 2.7 dB/µm, which can be
The hybrid plasmonic-silicon EA modulator is fabricated on calculated from fitting the optical loss of devices with different
a commercial silicon-on-insulator (SOI) wafer, with 250 nm length. The measured IL is larger than the simulated value of
silicon layer and 3 µm buried oxide. First, the slab waveguide 1.6 dB/µm. We articulate that there are two major reasons for
and grating couplers are patterned by two-step electron-beam the discrepancy. First, the material property in the simulation
lithography (EBL) and reactive ion etching (RIE). At the top model differs from the actual thin film. For example, the
50nm-thick silicon layer, the active region and contact regions permittivity of Au used in the simulation is based on ideal bulk
are highly doped by ion implantation with 5 keV of B+ ions at a material. As comparison, the thermally evaporated Au thin film
flux of 6×1014 cm-2, equivalent to a dopant concentration of in the experiment is known to be low-density with rough grain
1×1020 cm-3. After ion implantation, the dopants are activated boundaries, which usually has three times larger imaginary part
by rapid thermal annealing (RTA) at 1000 °C for 10 s. Next, 16 of the optical permittivity [34]. Similarly, the optical property
nm-thick HfO2 layer is conformably deposited using atomic of TCO is highly dependent on process condition. The non-
layer deposition (ALD). After ALD, ITO/Au gate layer is ideality from the sputtering process such as non-uniformity,
patterned by EBL. Then, 14 nm of ITO and 100 nm of gold are grain boundaries may all contribute to additional optical loss of
RF sputtered [33] and thermally evaporated, respectively, the waveguide. Second, the waveguide surface roughness from
followed by a lift-off process. The ITO is sputter at 1% O2/Ar EBL and RIE processes plays a critical role because the optical
gas flow, yielding a carrier concentration of 3×1020 cm-3. The scattering loss can be significantly enhanced by the plasmonic
HfO2 gate oxide in the silicon contact region is removed using effect [35]. The localized plasmonic mode bounded at the
buffered hydrofluoric acid. After that, an Ni/Au coplanar GSG metal-dielectric interface can be strongly scattered by the
waveguide surface roughness, even at a few nanometer scale.

0733-8724 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: OREGON STATE UNIV. Downloaded on May 03,2020 at 00:41:14 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2020.2979192, Journal of
Lightwave Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

A DC bias is applied across the capacitor. Fig. 5(a) shows the

transmission at 1.55 µm of an 8-µm-long modulator at different
DC biases (blue line) normalized to zero bias. The trend
matches well with our simulation. A dramatic increase of
absorption is observed at larger bias. From -6V to -8V, the
transmitted power reduces by 3.2 dB, indicating that the ITO in
accumulation layer has reached ENZ condition. The red line in
Fig. 5(a) shows the AC modulation ER at different DC bias
voltages, which is defined as difference of transmitted power at
each bias with 1V voltage swing. It shows that at -8V bias, the
AC ER is more than 2 times larger than that at 0V bias. In Fig.
5(b), the fabricated modulator exhibits a relatively uniform
optical modulation bandwidth from 1515 to 1580 nm, which is
mainly limited by the bandwidth of our grating couplers [36].
The intrinsic modulation bandwidth comes from the nature of
ENZ effect. Shorter wavelength requires larger carrier
concentration to trigger ENZ enhanced absorption. Our
simulation indicates a short wavelength limit to 1.45 µm for the
measured applied bias, which exceeds the bandwidth of the
grating coupler. The fringes on the spectrum comes from the
reflection between two grating couplers. The distance between
two grating couplers L is 500 µm and the group index ng is
2.64.The wavelength free spectral range (FSR) is calculated by
Δ𝜆𝜆𝐹𝐹𝐹𝐹𝐹𝐹 = 𝜆𝜆2 /𝑛𝑛𝑔𝑔 𝐿𝐿 ≈ 1.82 𝑛𝑛𝑛𝑛, which roughly matches with the
fringe period.

Fig. 6. (a) 1.5 GHz AC modulation of the fabricated device. (b) Eye diagram
of 4.5 Gb/s digital modulation. (c) Real part and imaginary part of the measured
and fitted S11 parameters. Inset: equivalent circuit model extracted from the
S11 parameter fitting.

Fig. 6(a) plots the modulation waveform of the modulator at

1.5 GHz. The upper curve shows the electrical input of the
device, whereas the lower curve is optical output to the Digital
Communication Analyzer (DCA) oscilloscopes, showing
electro-optical modulation. The rising time (10%-90%) is
measured to be ~0.1ns, which gives a bandwidth of ~3.5 GHz.
Fig. 5. (a) Experimental DC transmission and AC modulation strength for 8- We measured and fitted the S11 parameters of the modulator as
µm-long device at different bias. (b) Extinction ratio at -8V bias, showing shown in Fig. 6(c). The inset of Fig. 6(c) illustrates the extracted
broadband response from 1515 nm to 1580 nm. equivalent circuit. Here, Cp is the parasitic capacitance between

0733-8724 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: OREGON STATE UNIV. Downloaded on May 03,2020 at 00:41:14 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2020.2979192, Journal of
Lightwave Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

contact pads, and COX and RSi are the parasitic capacitance over means lower voltage swing is required. Table I also shows the
the buried oxide layer and the corresponding resistance. The simulated IL with the free carrier concentration of 3×1020cm-3,
modulation speed is determined by the MOS capacitance CMOS which clearly proves that high mobility TCOs also reduces the
and the device series resistance Rs, which are fitted to be 64 fF IL at the “ON” state.
and 642Ω. These two give us a RC bandwidth of 3.8GHz, which
matches well with the measured modulation bandwidth. The
MOS capacitance is slightly less than the designed value, which
may be attributed to dimension errors of the device fabrication
or the non-uniformity of the ITO film. However, the series
resistance is more than 6× larger than the estimated value of 100
Ω in our design, which is the limiting factor to achieve the
designed modulation bandwidth of 15 GHz for a 8-µm-long EA
modulator. We measured the sheet resistance of heavily doped
silicon slab to be ~1100 Ω/ , which larger than the design value
of ~200 Ω/ . We anticipate that the large silicon resistance is
mainly due to fabrication imperfection of the silicon
implantation and the annealing condition, which can be further
improved. Fig. 6(b) shows the eye diagram of 4.5 Gb/s data rate
with 2Vpp voltage swing. The ON and OFF states can be clearly
seen. Two main reasons are responsible for the limited open
width of the eye diagram. First, the optical output to the DCA
for eye diagram is close to the noise level of our instrument due
to high device loss, which would be improved by optimizing
taper design and fabrication process in the future work. Second,
the interfacial states in the MOS structure would compromise
the performance, which requires further investigation. The
energy consumption is estimated to be 64 fJ/bit using CVpp2/4.

V. FUTURE DESIGN USING HIGH MOBILITY TCOS

The operation of the modulator replies on ENZ-induced field
enhancement. The enhancement factor is inversely proportional
to the absolute value of the permittivity, which approximately
equals to the imaginary part of permittivity and inversely
proportional to the carrier mobility of TCO, |𝜀𝜀𝐸𝐸𝐸𝐸𝐸𝐸 | ≈ 𝜀𝜀𝐸𝐸𝐸𝐸𝐸𝐸
′′
∝ Fig. 7. (a) Comparison of extinction ratio for hybrid plasmonic-silicon EA
𝜇𝜇 𝑇𝑇𝑇𝑇𝑇𝑇 . The larger the carrier mobility is, the smaller the
−1
modulators driven by different mobility TCOs. (b) Voltage swing required to
permittivity is at the ENZ condition. Thus, the field achieve 1 dB/µm modulation strength for TCOs with different mobility and the
corresponding energy efficiency.
enhancement of the plasmonic ENZ EA modulator can be
further improved by high mobility TCO materials. It has been Overall, we can make the plasmonic-silicon EA modulators
suggested that CdO with a high mobility of 300 cm2V-1s-1 can more compact and energy efficient using high mobility TCOs.
potentially reach an ER over 5dB/µm [37]. But the dependence For example, to achieve a 3dB ER of a 3-µm-long ENZ
of driving voltage and energy efficiency on mobility is not modulator, the required voltage swing reduces from 5.1V to
clearly stated. Here, we simulated and compared the 0.5V with mobility increased from 15 cm2V-1s-1 (ITO) to 100
performance of hybrid plasmonic-silicon EA modulators using cm2V-1s-1 (Ti:In2O3), or to 0.2V with mobility increased to 300
similar MOS capacitor structure with different TCOs. Table Ⅰ cm2V-1s-1 (CdO). Correspondingly, the energy consumption for
lists TCOs with different carrier mobility and the corresponding a 3-µm long device reduces from 260 fJ/bit to 2.5 fJ/bit and 0.4
absolute value of the ENZ permittivity. The results are shown fJ/bit, calculated by CVpp2/4. To clarify, the resistance majorly
in Fig. 7(a). As the mobility of TCO increases, two effects can comes from the conduction path from the electrode to the center
be observed. First, the maximum ER is increased, which means of the active region through Si slab, which is a fixed distance in
shorter device length can be achieved. Second, the onset of the current design. The shorter device length would only reduce the
ENZ-enhanced plasma absorption becomes steeper, which capacitance to be around 40fF, providing the same capacitance
TABLE Ⅰ Permittivity |ε| at ENZ for TCOs of different mobility density, while maintain the same resistance. With improved
condition of doping and annealing, the resistivity of Si slab
TCO Mobility ENZ Effective Reference Simulated
Materials (cm2V-1s-1) |ε| Mass IL
would reach the designed value of 1×10-3 Ω·cm, corresponding
(dB/µm) to series resistance less than 100Ω [25]. Moreover, the electrode
ITO 15~30 0.96 0.33-0.35 [26], [28] 1.66 could be put closer to the center of the active region, once the
In2O3 20~35 0.42 0.3-0.32 [28], [38] 0.89 device length shortened. Correspondingly, high modulation
Ti:In2O3 70~100 0.15 0.27-0.31 [39] 0.64 bandwidth is estimated to be over 40 GHz, calculated by 𝑓𝑓 =
CdO 280~300 0.05 0.22-0.24 [37] 0.46 1
.
2𝜋𝜋𝜋𝜋𝜋𝜋

0733-8724 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: OREGON STATE UNIV. Downloaded on May 03,2020 at 00:41:14 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2020.2979192, Journal of
Lightwave Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

VI. CONCLUSION [16] E. Li, Q. Gao, S. Liverman, and A. X. Wang, “One-volt silicon
photonic crystal nanocavity modulator with indium oxide gate,”
In summary, we characterized the fundamental electro-optic Opt. Lett., vol. 43, no. 18, pp. 4229–4432, 2018.
modulation structure consisting of Au/ITO/oxide/p-Si capacitor [17] E. Li, Q. Gao, R. T. Chen, and A. X. Wang, “Ultracompact Silicon-
and showed that the property of ITO is greatly affected by the Conductive Oxide Nanocavity Modulator with 0.02 Lambda-Cubic
Active Volume,” Nano Lett., vol. 18, no. 2, pp. 1075–1081, 2018.
process condition. We reported an 8-µm-long plasmonic-TCO- [18] E. Li, B. A. Nia, B. Zhou, and A. X. Wang, “Transparent conductive
silicon EA modulator with 3.5 GHz modulation bandwidth and oxide-gated silicon microring with extreme resonance wavelength
broadband response from 1515 nm to 1580 nm. By biasing the tunability,” Photonics Res., vol. 7, no. 4, p. 473, 2019.
[19] C. Argyropoulos, P. Y. Chen, G. D’Aguanno, N. Engheta, and A.
device at the ENZ region, 3.2 dB ER with 2V voltage swing is Alù, “Boosting optical nonlinearities in ε-near-zero plasmonic
observed. Eye diagram of 4.5 Gb/s digital modulation is channels,” Phys. Rev. B - Condens. Matter Mater. Phys., vol. 85, no.
measured. Furthermore, we demonstrated the dependence on 4, pp. 1–5, 2012.
TCO mobility for extinction ratio and energy consumption of [20] S. Campione, I. Brener, and F. Marquier, “Theory of epsilon-near-
zero modes in ultrathin films,” Phys. Rev. B - Condens. Matter
EA modulator, based on enhanced ENZ effect. By adopting Mater. Phys., vol. 91, no. 12, pp. 1–5, 2015.
high mobility TCOs, a record-breaking device performance of [21] A. P. Vasudev, J.-H. Kang, J. Park, X. Liu, and M. L. Brongersma,
energy consumption of 0.4 fJ/bits and high speed over 40 GHz “Electro-optical modulation of a silicon waveguide with an ‘epsilon-
is expected for a 3-µm-long EA modulator. near-zero’ material,” Opt. Express, vol. 21, no. 22, p. 26387, 2013.
[22] H. W. Lee et al., “Nanoscale conducting oxide PlasMOStor,” Nano
Lett., vol. 14, no. 11, pp. 6463–6468, 2014.
ACKNOWLEDGMENT [23] Q. Gao et al., “Ultra-compact and broadband electro-absorption
modulator using an epsilon-near-zero conductive oxide,” Photonics
The authors would acknowledge the support from the EM Res., vol. 6, no. 4, p. 277, 2018.
Facility and MASC center at Oregon State University for the [24] Z. Lu, W. Zhao, and K. Shi, “Ultracompact electroabsorption
device fabrication. modulators based on tunable epsilon-near-zero-slot waveguides,”
IEEE Photonics J., vol. 4, no. 3, pp. 735–740, 2012.
[25] M. G. Wood et al., “Gigahertz speed operation of epsilon-near-zero
REFERENCES silicon photonic modulators,” Optica, vol. 5, no. 3, p. 233, 2018.
[1] V. J. Sorger, R. F. Oulton, R. M. Ma, and X. Zhang, “Toward [26] Y. Gui et al., “Towards integrated metatronics: a holistic approach
integrated plasmonic circuits,” MRS Bull., vol. 37, no. 8, pp. 728– on precise optical and electrical properties of Indium Tin Oxide,”
738, 2012. Sci. Rep., vol. 9, no. 1, pp. 1–10, 2019.
[2] A. Emboras et al., “Electrically controlled plasmonic switches and [27] J. H. Park, C. Buurma, S. Sivananthan, R. Kodama, W. Gao, and T.
modulators,” IEEE J. Sel. Top. Quantum Electron., vol. 21, no. 4, A. Gessert, “The effect of post-annealing on Indium Tin Oxide thin
2015. films by magnetron sputtering method,” Appl. Surf. Sci., vol. 307,
[3] S. A. Maier, “Plasmonics: The promise of highly integrated optical pp. 388–392, 2014.
devices,” IEEE J. Sel. Top. Quantum Electron., vol. 12, no. 6, pp. [28] Z. Ma, Z. Li, K. Liu, C. Ye, and V. J. Sorger, “Indium-Tin-Oxide
1671–1677, 2006. for High-performance Electro-optic Modulation,” Nanophotonics,
[4] Ekmel Ozbay, “Plasmonics : Merging Photonics and Electronics at vol. 4, no. 1, pp. 198–213, 2015.
Nanoscale Dimensions,” Science (80-. )., vol. 189, no. 2006, pp. [29] J. C. Ranuárez, M. J. Deen, and C. H. Chen, “A review of gate
189–194, 2012. tunneling current in MOS devices,” Microelectron. Reliab., vol. 46,
[5] N. Youngblood, Y. Anugrah, R. Ma, S. J. Koester, and M. Li, no. 12, pp. 1939–1956, 2006.
“Multifunctional graphene optical modulator and photodetector [30] Q. Gao, E. Li, and A. X. Wang, “Comparative analysis of
integrated on silicon waveguides,” Nano Lett., vol. 14, no. 5, pp. transparent conductive oxide electro-absorption modulators
2741–2746, 2014. [Invited],” Opt. Mater. Express, vol. 8, no. 9, p. 2850, 2018.
[6] M. Liu et al., “A graphene-based broadband optical modulator,” [31] G. V. Naik, V. M. Shalaev, and A. Boltasseva, “Alternative
Nature, vol. 474, no. 7349, pp. 64–67, 2011. plasmonic materials: Beyond gold and silver,” Adv. Mater., vol. 25,
[7] C. Y. Lin et al., “Electro-optic polymer infiltrated silicon photonic no. 24, pp. 3264–3294, 2013.
crystal slot waveguide modulator with 23 dB slow light [32] Q. Gao, E. Li, B. Zhou, and A. X. Wang, “Hybrid silicon-
enhancement,” Appl. Phys. Lett., vol. 97, no. 9, pp. 2008–2011, conductive oxide- plasmonic electro-absorption modulator with 2-V
2010. swing voltage,” J. Nanophotonics, vol. 13, no. 3, p. 036005, 2019.
[8] F. Ren et al., “Surface-normal plasmonic modulator using sub- [33] G. V Naik, J. Kim, and A. Boltasseva, “Oxides and nitrides as
wavelength metal grating on electro-optic polymer thin film,” Opt. alternative plasmonic materials in the optical range,” Opt. Mater.
Commun., vol. 352, pp. 116–120, 2015. Express, vol. 1, no. 6, pp. 1090–1099, 2011.
[9] P. Markov, K. Appavoo, R. F. Haglund, and S. M. Weiss, “Hybrid [34] W. R. Veazey and C. D. Hodgman, Handbook of chemistry and
Si-VO_2-Au optical modulator based on near-field plasmonic physics, vol. 7. Chemical Rubber Publishing Company, 1919.
coupling,” Opt. Express, vol. 23, no. 5, p. 6878, 2015. [35] Q. Gao, S. Liverman, and A. X. Wang, “Design and
[10] A. Melikyan et al., “High-speed plasmonic phase modulators,” Nat. Characterization of High Efficiency Nanoantenna Couplers with
Photonics, vol. 8, no. 3, pp. 229–233, 2014. Plasmonic Integrated Circuit,” J. Light. Technol., vol. 35, no. 15, pp.
[11] R. Ding et al., “A low VπL modulator with GHz bandwidth based 3182–3188, 2017.
on an electro-optic polymer-clad silicon slot waveguide,” IEEE Int. [36] X. Xu, J. Covey, D. Kwong, R. T. Chen, H. Subbaraman, and A.
Conf. Gr. IV Photonics GFP, vol. 18, no. 15, pp. 201–203, 2010. Hosseini, “Complementary metal-oxide-semiconductor compatible
[12] J.-M. Brosi, C. Koos, L. C. Andreani, M. Waldow, J. Leuthold, and high efficiency subwavelength grating couplers for silicon
W. Freude, “High-speed low-voltage electro-optic modulator with a integrated photonics,” Appl. Phys. Lett., vol. 101, no. 3, 2012.
polymer-infiltrated silicon photonic crystal waveguide,” Opt. [37] S. Campione et al., “Submicrometer Epsilon-Near-Zero
Express, vol. 16, no. 6, p. 4177, 2008. Electroabsorption Modulators Enabled by High-Mobility Cadmium
[13] L. Alloatti et al., “42.7 Gbit/s electro-optic modulator in silicon Oxide,” IEEE Photonics J., vol. 9, no. 4, 2017.
technology,” vol. 19, no. 12, pp. 11841–11851, 2011. [38] F. Fuchs and F. Bechstedt, “Indium-oxide polymorphs from first
[14] J. A. Dionne, K. Diest, L. A. Sweatlock, and H. A. Atwater, principles: Quasiparticle electronic states,” Phys. Rev. B - Condens.
“PlasMOStor: A metal-oxide-si field effect plasmonic modulator,” Matter Mater. Phys., vol. 77, no. 15, pp. 1–10, 2008.
Nano Lett., vol. 9, no. 2, pp. 897–902, 2009. [39] A. Chaoumead, H. D. Park, B. H. Joo, D. J. Kwak, M. W. Park, and
[15] V. J. Sorger, N. D. Lanzillotti-Kimura, R. M. Ma, and X. Zhang, Y. M. Sung, “Structural and electrical properties of titanium-doped
“Ultra-compact silicon nanophotonic modulator with broadband indium oxide films deposited by RF sputtering,” Energy Procedia,
response,” Nanophotonics, vol. 1, no. 1, pp. 17–22, 2012. vol. 34, pp. 572–581, 2013.

0733-8724 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: OREGON STATE UNIV. Downloaded on May 03,2020 at 00:41:14 UTC from IEEE Xplore. Restrictions apply.