IEEE SENSORS JOURNAL, VOL. 17, NO.

22, NOVEMBER 15, 2017 7239

Graphene Sensing Modulator: Toward Low-Noise,

Self-Powered Wireless Microsensors
Mehdi Hajizadegan, Student Member, IEEE, Maryam Sakhdari, Student Member, IEEE,
Liang Zhu, Qingsong Cui, Haiyu Huang, Student Member, IEEE, Mark M. C. Cheng,
Jonathan C. H. Hung, Member, IEEE, and Pai-Yen Chen, Senior Member, IEEE
Abstract— We present here new types of self-powered, low-
interference wireless sensors based on graphene circuits, which
can have dual functions: chemical sensing at the molecular level
and radio-frequency (RF) modulation. We demonstrate that a
fully passive, graphene-based harmonic (transponder) sensor can
display a chemically sensitive frequency multiplication effect,
which, when linked to a hybrid-fed small antenna, can realize an
ultrasensitive, low-profile, light-weight, and potentially flexible
RF sensor. We have designed two different types of circuits
comprising back-gate graphene field-effect transistors (GFETs)
and compared in detail their performance and implementation
complexity. We have also proposed a reliable readout method
based on the machine learning for extracting the mean value and
the fluctuation of chemical doping levels in GFETs. The proposed
graphene-based harmonic sensor may potentially benefit a wide Fig. 1. (a) Schematics of the conventional passive backscatter sensor (left)
range of sensing applications, including, but not limited to, power- and the nonlinear harmonic sensor (right), which allows for low-noise, low-
efficient, real-time monitoring of chemical/gas exposures and interference sensing by transmitting and receiving RF signals at orthogonal
biological agents, as well as emerging wearable and implantable frequencies. (b) Architectures for a harmonic sensor based on (left panel)
devices. the conventional CMOS circuits and (right panel) the new graphene-based
sensing-modulator circuit that integrates several functions into a single
Index Terms— Graphene sensors, wireless sensors, RFID sen- module.
sors, transponder sensors, healthcare Internet-of-Things.
tags could be made inexpensively and with a very small size,
I. I NTRODUCTION how to remove the non-negligible background noises, clutters,
and backscattered echoes remains challenging, as they may
O NE of the most important hardware advances in the
era of Internet of Things (IoT) is the large popu-
lation of wirelessly connected sensors. With tens of bil-
deteriorate the performance and reliability of wireless sensor
nodes. Recently, RFID harmonic transponder sensors have
lions of objects equipped with sensors wirelessly synced been proposed to overcome the above-mentioned drawbacks
to the internet, completely new applications that are cru- of conventional backscatter RFIDs. A harmonic transponder
cial, but were impossible before, can be conveniently sup- sensor, as a nonlinear RFID tag, receives the RF signal at
ported [1]–[4]. Comparing to other wireless technologies, the fundamental frequency f0 and retransmits a modulated
backscatter RF identification (RFID) is arguably the most RF signal at sub- or high-harmonic frequency (e.g. second har-
attractive candidate for realizing such a large quantity of wire- monic 2 f 0 ), such that the frequency-modulated backscattered
less sensors integrated into IoT, due to its potentially battery- signal could be free from severe background clutters, jamming,
free, ultralow-profile, and ultralow-cost features. Traditionally, multipath-scattering and background electromagnetic inter-
backscatter RFIDs containing RFID readers, tags, and software faces [13]–[19], as illustrated in Fig. 1(a). The harmonic sensor
have been widely used for wireless sensing applications, have been long used for tracking the location of electrically-
where a simple RFID tag scatters back a portion of a radio- small target [15]. Relatively new applications include wire-
frequency (RF) signal to the reader without any amplifica- lessly tracing conditions of miniature sensors, which are
tion or retransmission [5]–[12]. Although these types of RFID crucial for ubiquitous healthcare, such as temperature,
humidity, and micro-liter liquid volume, to name a
Manuscript received June 16, 2017; accepted July 30, 2017. Date of few [15]–[20].
publication August 11, 2017; date of current version October 24, 2017. With the rapid development of nanomaterials and nanofab-
The associate editor coordinating the review of this paper and approving
it for publication was Prof. Mehdi Javanmard. (Mehdi Hajizadegan and rication, the atomically-thin graphene, comprising a single
Maryam Sakhdari contributed equally to this work.) (Corresponding author: layer of carbon atoms arranged in a honeycomb lattice, has
Pai-Yen Chen.) had a revolutionary impact on electronic and photonic sensors
M. Hajizadegan, M. Sakhdari, L. Zhu, Q. Cui, M. M. C. Cheng,
and P.-Y. Chen are with the Department of Electrical and Computer due to its strong conductivity dependency on chemical and
Engineering, Wayne State University, Detroit, MI 48017 USA (e-mail: electrostatic gating effects, as well as extraordinary mechanical
pychen@wayne.edu). and optical characteristics [21]–[35]. In these graphene-based
H. Huang and J. C. H. Hung are with Maxim Integrated Inc., Dallas,
TX 75240 USA (e-mail: jonathan.hung@maximintegrated.com). sensors, high sensitivity is anticipated due to the fact that
Digital Object Identifier 10.1109/JSEN.2017.2737699 graphene has a relatively low density of states in the vicinity
1558-1748 © 2017 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.
7240
of its Dirac point, where a small change of carrier concentra-
tion (e.g. doping) may dramatically vary graphene’s electrical
conductance properties. Although the state-of-the-art graphene
device has demonstrated a single-molecule sensitivity, it still
lacks a promising system-level solution to turn this special
nanomaterial into a practical RFID or IoT sensor, since it is
difficult to integrate the graphene device with the conventional
silicon-based RF modulation and communication modules.
Graphene research has been accelerated exponentially in
high-frequency electronics because it has extremely high car-
rier mobility. The state-of-the-art graphene field-effect tran-
sistor (GFET) already exhibits a cutoff frequency ( f T ) up
to 300 GHz [36], which is much higher than conventional
silicon-based transistors. In addition, its unique symmetric
“V-shape” drain current-gate voltage characteristic, trig-
gered by the ambipolar charge transports in the gapless
graphene [37]–[42], makes GFET the simplest possible fre-
quency doubler that can operate at very high frequency
range (up to the tens of GHz [42]). In this work, we intend
to combine two unique properties of graphene, namely the
high chemical sensitivity and frequency modulation capability
in GFETs, to realize a nanomaterial-based harmonic transpon-
der sensor, as schematically illustrated in Fig. 1(b). Since the
electrical conductivity of graphene is highly sensitive to gas,
chemical, or biological dopants (e.g. bacterial or infectious
agents), a single back-gate or side-gate GFET may realize a
chemically-sensitive frequency multiplier, in which a shift in
Dirac point (or charge natural point Vcnp ) caused by chemi-
cal gating effects may modulate the second-harmonic output
signal and therefore tune the nonlinear conversion gain [43].
However, this sensing-modulator comprising a single GFET
may not be suitable to build a passive harmonic transponder
because it still requires a drain-to-source dc bias, namely a
power source like the piezoelectric transducer, micro-battery,
RF rectifier, or solar cell. In our recent seminar paper, we have
proposed a GFET-based circuit that represents a type of self-
powered harmonic sensor, which has a quad-ring mixer topol-
ogy formed by four GFETs [44], as shown in Fig. 2(a). Such
circuit combines functions of sensor, frequency modulator, and
energy harvester within a single module, as shown in Fig. 1(c).
Still, this complicated architecture brought challenges in fab-
rication and calibration, as it requires multiple transistors and
interconnection layers, as well as consistency of intrinsic Dirac
point for all four GFETs.
We propose here a new self-activated GFET-based harmonic
sensor comprising reduced number of GFETs and much sim-
plified interconnection scheme, as shown in Fig. 2(b). In this
single-balance RF circuit, the paired GFETs are inductively
coupled to a hybrid-fed antenna. Compared to the circuit
topology of the quad-ring mixer in [14], this design allows a
more compact device footprint and reduced fabrication com-
plexity, while improved sensing linearity and reliability. In this
study, we will compare the sensing performance between the
dual-ring (single-balanced) and quad-ring (double-balanced)
GFET-based sensing-modulator; here, the circuit simulations
were conducted based on the realistic physics-based device
compact model that has been verified with experimental
results. We also consider a practical scenario in which
IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017
Fig. 2. Circuit diagram for (a) the quad-ring (double-balanced) sensing-
modulator and (b) the dual-ring (single-balanced) sensing-modulator. Both
circuits are based on GFETs and inductively coupled to a hybrid-fed patch
antenna, of which the first port receives the input signal at the fundamental
frequency and the second port transmits the second-harmonic output to the
interrogator/reader.
Fig. 3. (a) Schematics and optical image of the fabricated GFET. (b) Drain
current-gate voltage characteristics for the GFET in (a); here, solid line and
dots represent theoretical and experimental results, respectively.
chemical agents could randomly bind to receptors on the
graphene surface, resulting in non-consistent drifts of Vcnp in
different GFETs and, therefore, fluctuations in the frequency
conversion efficiency of the GFET-based sensing-modulator.
To address this issue, we present a new decoding and error-
estimation process for analyzing chemical gating levels from
wirelessly sensed high harmonics, by means of the artificial
neural network (ANN), within the realm of machine learning.
We show that with the pattern recognition of the second-
and third-harmonic output signals, the mean value of Vcnp
and its variations Vcnp , and consequently the corresponding
changes in chemical dopant concentrations can be robustly and
accurately read. This pattern-recognition-based method could
potentially achieve more reliable harmonic sensing.
II. S ENSING AND F REQUENCY M ODULATION
C HARACTERISTICS OF G RAPHENE T RANSISTOR
Figure 3(a) shows the microscope image of the fabri-
cated back-gate GFET, with channel width W = 900 μm,
channel length L = 12 μm, and a 300 nm-thick SiO2
gate oxide; details of the fabrication method can be found
in [43]. Figure 3(b) reports the measured drain current–gate
voltage (IDS -VGS ) characteristic of this GFET arranged in
a single-stage common-source configuration, under different
drain-source bias voltages; here, solid lines and dots represent
theoretical and experimental data, respectively. It is clearly
evident that experimental results excellently agree with theo-
retical results calculated using the physics-based drift-diffusion
model, which will be described below. Here, the compact
physics-based GFET model is employed to numerically study
HAJIZADEGAN et al.: GRAPHENE SENSING MODULATOR: TOWARD LOW-NOISE, SELF-POWERED WIRELESS MICROSENSORS 7241

the GFET-based sensor circuits. In the drift-diffusion transport quantum capacitance Cq as a function of surface potential can
model, the drain-to-source current IDS is described by [45]: be explicitly written as [55]:
 VDS
W Cq = −∂ (Q s ) /∂ Vch
|Q
IDS = q W tot (x)| vdrift (x) = μeff |Q tot | d V . (1)   
L eff 0 2q 2 K B T q V ch
= ln 2 1 + cosh . (7)
where W is the channel width, the effective channel length π (h̄vF )2 KB T
V
L eff = L − 0 DS 1/νsat d V , accounting for the effect of
saturation velocity, νdrift is the drift velocity, which, in a Vcnp is the back-gate voltage at the Dirac point, where the
soft saturation model, is related to the channel electric field carrier sheet density becomes minimal for zero drain and
F = −d V /d x as [46]–[54]: source voltages. We note that Vcnp comprises the work function
difference between the gate and the graphene channel (similar
μF to the flatband voltage in conventional silicon transistors),
vdrift =  1/γ , (2)
1 + (μeff F/vsat )γ interfacial charges at the graphene-oxide interfaces, and the
charge neutrality point shift Vcnp due to intentional or
where vsat is the saturation velocity given by [46]: unintentional doping of graphene (e.g. chemical gating effects
 
c1 caused by gas, chemical or molecular agents). The channel
vsat = vF + c3 , (3) voltage V (x) in Eq. (6), as a function of the position in the
1 + c2 (Vch )2
channel, can be modeled using the gradual channel approx-
vF = 108 cm/s is the Fermi velocity in graphene, imation: V (x) = (x/L)VDS , which is zero at the source
c1 = (vsat,max − vsat,min )/vF , c2 = [q/(5K B T )]2 , c3 = end (x = 0) and equal to the drain-to-source voltage VDS at the
vsat,min /vF [47], [48], K B is the Boltzmann constant, T is drain end (x = L). The exact surface potential in graphene can
temperature, and Vch is the surface potential of graphene. be obtained by iteratively solving Eqs. (6) and (7). Once Vch is
The effective carrier mobility in Eq. (3) can be written calculated, the total charge density in the graphene channel can
as [49]–[51]: be obtained using Eq. (5), which, when substituted into Eq. (1)
 gives the drain current IDS (VDS , VGS ). For a moderate input
nμn + nμp + n pud (μn + μp )/2 m
μeff = , (4) RF voltage, one may numerically calculate the small-signal
n + n + n pud m + Vch
2
transconductance of GFET gm = d I DS /d VGS |VDS =const and
the drain conductance gds = 1/rds = d I DS /d V DS |VGS =const.
where n, p are the electron and hole densities in a graphene
The mobile channel charge, depending on VGS and VDS , can
channel, μn,p is the carrier low-field mobility (here μn,p =
be modeled by the gate-to-source capacitance Cgs and the gate-
1300 cm2 /(V·s)), n pud is the residual charge due to electron-
to-drain capacitance Cgd : Cgs = −d Q ch /d V GS] |VDS =const , and
hole puddles (here n pud = 1.65 × 1012 cm−2 ), and m is the
Cds = −d Q ch /d V DS |VGS
 L=const, where the total net charge in
square of the reference channel potential, which is an empirical
fitting parameter (here m = 1 V2 ). The total mobile charge the channel Q ch = W 0 Q tot d x. The physics-driven GFET
density Q tot in graphene, involving the electron and hole sheet model was implemented in the circuit simulator [56].
densities n and p [cm−2 ] and the residual charge due to Consider an GFET biased at a certain DC drain voltage,
electron-hole puddles [50], [52], can be explicitly written as: an RF input signal (a monochromatic, sinusoidal wave) can
be efficiently converted to its second harmonic ( f 0 → 2 f 0 ),
Q tot = Q s + qn pud with magnitude much greater than the output fundamental

 2
tone and other harmonics [37], thanks to the V-shape current-
q V ch
Q s = q × n imp + voltage behaviors, as shown in Fig. 3(b). Such interesting
π h̄ 2 v2F property makes GFET an ideal candidate for the RF frequency
2 doubler [37]. Moreover, when chemical dopants (e.g. specific
n pud = (5) molecules or bacterial pathogens) selectively bind on the func-
π h̄ 2 v2F
tionalized graphene surface, the Dirac-point voltage (Vcnp ) is
where  is the inhomogeneity of the electrostatic potential shifted due to the chemical gating effect, as shown in Fig. 4(a).
(here  = 150 meV), n imp is the sum of the intrinsic field- Such effect alters the bias point and therefore changes the
independent and impurity carrier concentrations (here n imp = frequency conversion efficiency. Fig. 4(b) reports the output
7 × 1012 cm−2 ), h̄ is the reduced Planck constant, q is electric voltage at the fundamental frequency and the second har-
charge. The surface potential (Vch ) related to graphene’s Fermi monic for the practical GFET, of which the graphene channel
level E F , can be expressed as a function of the gate-to-source was fully exposed to the chemical solutions with different
voltage VGS as [46]–[54]: PH values; here we fixed the DC drain voltage to 0.25 V
Cox  
Vch (x) = E F (x)/q = VGS − Vcnp − V (x) , (6)
Cox + αCq
where α is the capacitance weighting factor [48] (α ≈ 1
when q |Vch |  kT , and α ≈ 0.5 when q |Vch |  kT ), the
electrostatic capacitance Cox is related to the permittivity εox
and thickness tox of gate oxide as Cox = εox /tox , and the
and the AC input voltage to 5 V.
It has been reported that the charge density in the graphene
channel depends on the type and the concentration of chemical
groups on the graphene surface. Therefore, the drift of Dirac
point may provide a quantitative measure of the acid-based
properties of the ionizable groups (e.g. as a function of
solution pH). The induced surface charge is mainly negative
7242
Fig. 4. (a) Schematics of frequency modulation in a single-state GFET.
(b) Measured output signals for the GFET in Fig. 3, with the graphene
channel fully exposed to solutions with different pH values. Insets show the
corresponding IDS -VGS characteristics at different PH values. (c) Shift of Vcnp
and magnitude of output signals at the first two harmonics, as a function of
PH value of buffer solution drop-casted onto the graphene channel; here, solid
lines and dots represent the theoretical and experimental results, respectively.
above pH 7 and is positive otherwise, typical of chemical
gating effects. A model for the surface potential of graphene as
a function of the electrolyte concentration can be derived from
the Grahame equation [57], which was used here to obtain the
relationship between the surface potential and the PH value.
In our experiment, a 0.5 μL pH buffer drop was carefully
placed on top of graphene channels. After approximately
10-20 seconds, the channel conductance became stabilized,
and buffer drops with different pH values were applied and
removed repeatedly for the same device. We mixed the stan-
dard pH 7 buffer solution (BX1632-1 buffer solution from
EMD Millipore) with the hydrogen chloride (HCL) solution
to prepare acid solutions with different pH values, which
were confirmed with the high accuracy pH meter. From the
insets of Fig. 4(b), it is clearly evident that the IDS -VGS
curve moves toward p-doping direction when pH of buffer
solution increases, which in turn varies the magnitude of
output fundamental tone and second harmonic, as summarized
in Fig. 4(c). We note that IDS -VGS curves in Fig. 4(b) are
not perfectly symmetric because ion-liquid buffers may lead
to the asymmetric electron-hole dispersion. Such effect was
also considered in our model by assuming different low-
field motilities for electrons and holes. It is worth mentioning
that in Fig. 4(c), the theoretical results obtained from the
compact model (solid lines) agree quite well with experimental
results (dots). For a thick gate oxide used here, the Dirac point
shift, as an indicator for the level of p/n-doping in graphene,
is about 1.38 V per PH. We should point out that when a thin-
ner high-K gate oxide is used (e.g. those prepared by atomic
layer deposition or insulating 2D materials [58]), the change
in the second harmonic magnitude may be more dramatic,
IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017
implying a higher sensitivity. In this proof-of-concept demon-
stration, structural and material parameters of GFET were
not yet optimized to achieve the maximum transconductance
and the nonlinear conversion gain. Nonetheless, our work has
demonstrated the feasibility of making a compact sensing-
modulator based on GFETs.
III. D ESIGN OF A S ELF -P OWERED , C HEMICALLY-O FFSET
F REQUENCY M ULTIPLIER
Figures 2(a) and 2(b) respectively present the circuit design
for two different types of fully-passive, graphene-based har-
monic sensors, which are based on quad-ring and dual-ring
self-biased GFETs. Such sensing-modulators can be cou-
pled either to the RF input/output ports of a hybrid-feed
antenna [59], or to two individual antennas resonating at fun-
damental frequency and second harmonic. In this fully-passive
circuit, the self-biased GFET plays a dual role: a receptor for
chemical sensing and a nonlinear RF component for frequency
modulation. In this scenario, Vcnp of individual GFETs would
determine magnitudes of high harmonic outputs. In contrast to
the quad-ring mixer circuit [Fig. 2(a)] [14], the single-balanced
design [Fig. 2(b)], proposed for the first time here, possesses
several advantages, as it requires fewer graphene transistors,
much simplified interconnection circuits, and, therefore, more
compact size and lower cost. More interestingly, it offers
improved linearity and sensitivity than the quad-ring design.
We have conducted the circuit simulations for the sensing-
modulator circuits in Fig. 2, by assuming the following device
parameters: channel width W = 80 μm, channel length
L = 50 μm, relative dielectric constant of gate oxide εox /
ε0 = 9 (e.g. Al2 O3 ), thickness of gate oxide tox = 25 nm,
carrier mobility μ = 3000 cm2 /(V·s), which is close to what
reported in literature. For simplicity, we ignore the parasitic
capacitance and resistance. This is approximately valid at
moderately low operating frequencies.
Figures 5(a) and 5(b) respectively present the time-domain
and frequency-domain simulation results for the dual-ring
sensing-modulator, with Vcnp of GFETs varying from 0 V to
0.5 V; here the input is a monochromatic, sinusoidal signal
432 MHz [dashed line in Fig. 5(a)]. Figures 6(a) and 6(b) are
similar to Figs. 5(a) and 5(b), but for the quad-ring design.
It is seen that for both designs, the magnitude of the output
signal, particularly at the second harmonic, can be sensi-
tively modulated by the drift of Dirac point. In general, Vcnp
can be sensitively determined by concentrations of reactive
gases or molecules that lead to n-/p-type doping effects. From
Fig. 5(a), we find that the simple dual-ring design would act
like a full-wave rectifier, of which fundamental tone and other
odd harmonics are suppressed, with their magnitude much
lower than even harmonics. On the other hand, the quad-ring
circuit [Fig. 6(a)] acts like the half-wave rectifier, of which
even and odd harmonics are both significant. This is very
counterintuitive, as in the conventional (unipolar) silicon-based
frequency multiplier, the quad-ring circuit is comparatively
better than the dual-ring structure in terms of generating clean
odd-harmonic peaks.
Figures 5(c) and 6(c) respectively present contours of
the second-harmonic conversion gain (which is defined as the
HAJIZADEGAN et al.: GRAPHENE SENSING MODULATOR: TOWARD LOW-NOISE, SELF-POWERED WIRELESS MICROSENSORS
Fig. 5. Simulation results for the dual-ring sensing-modulator circuit
in Fig. 2(b). (a) Time-domain output signals for different Vcnp values of
GFETs. (b) First five harmonics of the output signal (spectral analysis) with
Vcnp = 0 V (blue lines) and Vcnp = 0.3 V (red lines); in (a) and (b),
the voltage of input fundamental tone is 1 V. (c) Contours of second-
harmonic conversion gain as a function of input voltage amplitude (Vin )
and GFET’s Vcnp . (d) Second-harmonic conversion gain against Vcnp for
different RF input voltages (error bar shows standard deviations obtained
by carrying out 104 simulations, with Vcnp of individual GFETs randomly
perturbed by ±10 %).
Fig. 6. Simulation results for the quad-ring sensing-modulator circuit
in Fig. 2(a). (a) Time-domain output signals for different Vcnp values of
GFETs. (b) First five harmonics of the output signal (spectral analysis) with
Vcnp = 0 V (blue lines) and Vcnp = 0.3 V (red lines); in (a) and (b),
the voltage of input fundamental tone is 1 V. (c) Contours of second-
harmonic conversion gain as a function of input voltage amplitude (Vin )
and GFETs’ Vcnp . (d) Second-harmonic conversion gain against Vcnp for
different RF input voltages (error bar shows standard deviations obtained
by carrying out 104 simulations, with Vcnp of individual GFETs randomly
perturbed by ±10 %).
voltage ratio between the output second harmonic and the
input fundamental tone) as a function of input voltage (Vin )
and GFETs’ Vcnp , for the GFET-based dual-ring and quad-
ring circuits. We note that if Vcnp of all graphene receptors
are matched to zero (i.e. pristine graphene), the conversion
gain is zero for any arbitrary input voltage, due to the
symmetric, V-shape current-voltage characteristics of GFETs.
From Figs. 5(c) and 6(c), it is evident that Vcnp of GFETs
can drastically affect the second-harmonic conversion gain,
regardless of amplitude of the input signal. We should note
that although the conversion gain increases with increasing
the magnitude of input signal, however, even for a small
7243
Fig. 7. Contours of sensitivity as a function of Vin and Vcnp for (a) the
dual-ring sensing-modulator and (b) the quad-ring sensing-modulator.
input signal, the second-harmonic conversion gain can be
modulated over a wide range by shifting Vcnp of GFETs.
Figures 7(a) and 7(b) respectively report the contour of the sen-
sitivity as functions of pH and Vin for the dual-ring and quad-
ring circuits; here, the sensitivity is defined as the derivative
of output second-harmonic amplitude with respect to the pH,
which is related to the Vcnp by the Grahame equation [57].
Again, it is seen that although the quad-ring circuit may offer
a very high sensitivity in the low Vcnp regime, the dual-ring
design, however, provides a better sensing linearity than the
quad-ring one. Moreover, the sensitivity is rather independent
of the amplitude of input signal, implying that robust sensing
is potentially possible. We note that a large input signal is
possible by using passive transformers, which are commonly
used in RFID tags and IoT sensors, or by adopting the near-
field communication (NFC) schemes.
Here, we also consider practical cases in which Vcnp of
individual GFETs could fluctuate. To elucidate the influence
of Vcnp fluctuation on robustness and reliability of the proposed
sensing-modulators, we have randomly perturbed Vcnp of
GFETs by ±10% in both quad-ring and dual-ring circuits, and
have conducted the simulation for 104 times for getting reliable
statistical results. Figures 5(d) and 6(d) present the second-
harmonic conversion gain against the average Vcnp for the
dual-ring and quad-ring sensing-modulators, under different
applied input voltages (Vin = 0.5, 1, 1.5 and 2 V). The
symbols and error bars in Figs. 5 (d) and 6(d) represent
the mean values and standard deviations of our numerical
experiments, clearly showing that the sensitivity of both types
of transponder sensors is quite robust to Vcnp fluctuations.
Comparing the conversion gain between the dual-ring and
quad-ring GFET circuits, we find that the dual-ring design
exhibits better performance in terms of linearity and sensitiv-
ity. When a quad-ring sensor is exposed to chemical agents,
there are two distinct operation regions. In the low Vcnp region
(Vcnp ≤ Vcnp,peak), a slight increment of Vcnp leads to a
dramatic increase in the magnitude of the second harmonic
output, whereas, in the high Vcnp region (Vcnp ≥ Vcnp,peak),
the conversion gain slowly decreases with increasing the Vcnp
value and eventually saturates. On the other hand, the dual-
ring GFET circuit shows a linear dependence between the
conversion gain and Vcnp , over a broad range of Vcnp , thereby
providing a better linearity than the quad-ring design. From
Figs. 5(d) and 6(d), it is worth mentioning that the maximum
second-harmonic conversion gain of the dual-ring circuit could
be greater than that of the quad-ring one, as a direct conse-
7244 IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017

via synapses to the second layer of neurons, and finally to the

output neuron layer. Mathematically, each neuron computes a
weighted sum of its n-input signals x j , for j = 1, 2, . . . , n,
where the neuron’s network function f (x) is defined as a
composition of other functions gi (x), which can further be
defined as a composition of other functions. A widely used
type of composition, which we have also used, is the nonlinear
weighted sum:

f (x) = K  ( ωi gi (x)), (8)
where K  is activation function and ωi is weight [62]–[68].
Our study utilizes a typical back-propagation neural net-
work (BPNN) [66]–[69], where the input data is first passed
through the network using a set of weights; then the output
of each neuron is the weighted sum of its inputs filtered
by a sigmoid function, where the neuron’s network functions
(input-output relationships) are given as follows:
 Ni
xj = wij x i + θj
i=1
  1
yj = σ x j =   , i = 1, 2, . . . , Ni (9)
1 + exp −x j
Nj
Fig. 8. (a) Schematics of ANN model for harmonic sensing. Verification of xk = wjk x j + θk
the ANN-predicted Vcnp and Vcnp with the exact values of GFETs for (b) the j=1
training data and (c) the testing data. We have conducted 103 simulations, 1
of which 70 % (30 %) data was collected for training (testing). yk = σ (x k ) = , j = 1, 2, . . . , Nj , (10)
1 + exp (−x k )
quence of the major difference between full-wave and half- where the indices i , j , and k for the neurons correspond to
wave rectifications. the input, hidden, and output layers, respectively, x i(j,k) and
yi(j,k) show the input and output of the i-th (j-th, k-th) neuron,
IV. A RTIFICIAL N EURAL N ETWORK (ANN) and wij (wjk ) represents the weight from the i-th ( j -th) neuron
FOR DATA A NALYSIS of the input (hidden) layer to the j-th (k-th) neuron of the
The fluctuation of Vcnp in individual GFETs may break the hidden (output) layer. Also, θ is the threshold parameter for
symmetricity of dual-ring and quad-ring circuits, leading to each neuron. Initially, the weights are randomly given. Then,
an overall decrease in generated even harmonics, while an the final outputs and training data are compared, and the
increase in odd harmonics. In this section, we will exploit the training error function is calculated as follows:
artificial neural network (ANN), as a typical machine learn- Ns 1  Ns  Nk  s 2
ing approach, to determine the Dirac point of the reference E= Es = t − yks . (11)
s=1 2 s=1 k=1 k
GFET (Vcnp,0 ) and fluctuations (Vcnp) occurring in the rest
GFETs; here, the Dirac point of the i -th GFET (except for where Ns is the number of training samples, tks and yks
the reference one) is given by Vcnp,i = Vcnp,0 + Vcnp,i, represent the desired output and k-th computed output for the

N−1 s-th inputting sample, respectively. The renewed weights are
and Vcnp = 1/(N − 1) |Vcnp,i|. We note that both then updated by the following:
i=1
positive and negative Vcnp,i lead to the same result, due to the ∂E
wij(jk)(n + 1) = wij(jk) (n) − η
ambipolar transport characteristic in GFETs. It is noteworthy ∂wij(jk)(n)
that the GFET sensing-modulator can produce more than one + α (wij(jk)(n) − wij(jk) (n − 1)). (12)
high harmonics [Figs. 5(b) and 6(b)], thus providing redundant ⎧  Ns
⎨ ∂E
information for signal post-processing and data analysis in the = δ s (n)yjs (n)
∂wjk (n) s=1 k (13)
harmonic sensing scheme. Detecting multiple high harmonics, ⎩ s
δ (n) = (yks (n) − tks (n))x ks (n)(1 − x ks (n))
although inevitably increasing complexities in antenna and ⎧ k  Ns
reader designs, may improve the robustness and accuracy of ⎪ ∂E =
⎨ δ s (n)yis (n)
sensing results. Here, we consider both second- and third- ∂wij (n) s=1 j
 Nk (14)
⎪
⎩δ s = x s (n)(1 − x s (n))
harmonic output signals as input variables in the ANN model,
j j j wjk (n)δ s (n)
k
as shown in Fig. 8(a). The ANN is implemented based on k=1

a large collection of neural units (artificial neurons), loosely
mimicking the way a biological brain solves problems with
large clusters of biological neurons connected by axons [61].
A typical ANN consists of at least three layers, as shown
in Fig. 8(a): the first layer has input neurons, which send data
where η ∈ (0, 1) and α ∈ (0, 1) are the learning rate
and momentum factor, respectively; here, a learning rate of
η = 0.2, momentum factor of α = 0.1, and threshold of
θ = 0 were used. For generating the training data for the ANN,
we have simulated 700 dual-ring sensing-modulators, of which
HAJIZADEGAN et al.: GRAPHENE SENSING MODULATOR: TOWARD LOW-NOISE, SELF-POWERED WIRELESS MICROSENSORS
the Dirac point of the reference GFET was randomly assigned
and that of the other GFET was given by Vcnp  = Vcnp,0 +
Vcnp , where Vcnp = τ Vcnp,0 and τ is a random value
between 0 and 1. Figure 8(b) reports the training results, where
x- and y-axes represent the exact Vcnp,0 and Vcnp values, and
those predicted by the ANN, respectively.
Once the ANN was trained, we simulated 300 sensors with
randomly generated Vcnp,0 and Vcnp , with their output second
and third harmonic recorded as input variables of ANN.
Figure 8(c) presents exact Vcnp,0 and Vcnp in comparison
with what predicted by the trained ANN for all 300 sensors
under test. We find that results predicted by the ANN agree
quite well with the exact Vcnp,0 and Vcnp obtained from
circuit simulations. As a result, the ANN model that consider
multiple high-harmonic outputs may help one to better analyze
sensed parameters, thanks to the increased number of degrees
of freedom (harmonics). We note that the performance of ANN
may be further enhanced by analyzing more high harmonics
in the RF spectrum.
V. C ONCLUSIONS
In summary, we have proposed a passive, compact and
ultrasensitive wireless harmonic sensor based on a simple ring
circuit formed by graphene transistors. We have theoretically
shown that such circuit can act as a chemically-modulated
frequency doubler with its second-harmonic conversion gain
sensitively responsible for the chemical gating effect, which
is not possible with conventional solid-state circuits. We have
first experimentally demonstrated the possibility to achieve
sensing and frequency modulation using a properly biased
single-state GFET. Then, we have developed a physics-based
compact model to describe characteristics of GFETs, and have
applied it to simulate new types of dual-ring (single-balanced)
and quad-ring (double-balanced) sensing-modulators consti-
tuted by back-gate GFETs. We have compared the sensing
performance of these two circuits and found that the dual-ring
topology may not only ease the implementation complexity
and cost, but also enable better sensing linearity. Finally, we
have proposed a new wireless readout scheme based on the
artificial neural network for accurately determining the drift of
Dirac point (resulting from binding of gas, molecular or chem-
ical agents onto GFETs) and the range of Dirac-point fluc-
tuation in GFETs. The proposed graphene-based harmonic
sensors, receiving the fundamental tone and retransmitting the
high harmonic, show great potential to realize the low-noise,
low-interference chemical sensing in scattering-rich and noisy
measurement environments.
R EFERENCES
[1] S. Amendola, R. Lodato, S. Manzari, C. Occhiuzzi, and G. Marrocco,
“RFID technology for IoT-based personal healthcare in smart spaces,”
IEEE Internet Things J., vol. 1, no. 2, pp. 144–152, Apr. 2014.
[2] L. Y. Chen et al., “Continuous wireless pressure monitoring and mapping
with ultra-small passive sensors for health monitoring and critical care,”
Nature Commun., vol. 5, p. 5028, Oct. 2014.
[3] R. S. H. Istepanian, S. Hu, N. Y. Philip, and A. Sungoor, “The potential
of Internet of m-health Things ‘m-IoT’ for non-invasive glucose level
sensing,” in Proc. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBS),
Aug. 2011, pp. 5264–5266.
[4] H. Huang et al., “RFID tag helix antenna sensors for wireless drug
dosage monitoring,” IEEE J. Translational Eng. Health Med., vol. 2,
pp. 1–8, 2014.
7245
[5] R. Want, “An introduction to RFID technology,” IEEE Pervas. Comput.,
vol. 5, no. 1, pp. 25–33, Jan. 2006.
[6] F. Klaus, RFID Handbook: Fundamentals and Applications in Contact-
less Smart Cards and Identification, 2nd ed. Hoboken, NJ, USA: Wiley,
2003.
[7] D. Kim, M. A. Ingram, and W. W. Smith, “Measurements of small-scale
fading and path loss for long range RF tags,” IEEE Trans. Antennas
Propag., vol. 51, no. 8, pp. 1740–1749, Aug. 2003.
[8] P. V. Nikitin and K. V. S. Rao, “Theory and measurement of backscat-
tering from RFID tags,” IEEE Antennas Propag. Mag., vol. 48, no. 6,
pp. 212–218, Dec. 2006.
[9] J. Landt, “The history of RFID,” IEEE Potentials, vol. 24, no. 4,
pp. 8–11, Oct. 2005.
[10] C. He, Z. J. Wang, and V. C. M. Leung, “Unitary query for the M×L×N
MIMO backscatter RFID channel,” IEEE Trans. Wireless Commun.,
vol. 14, no. 5, pp. 2613–2625, May 2015.
[11] C. He, Z. J. Wang, C. Miao, and V. C. M. Leung, “Block-level unitary
query: Enabling orthogonal-like space-time code with query diversity
for MIMO backscatter RFID,” IEEE Trans. Wireless Commun., vol. 15,
no. 3, pp. 1937–1949, Mar. 2016.
[12] M. M. Islam, J. Song, K. Rasilainen, and V. Viikari, “Optimization of
RFID sensor with frequency modulation,” IEEE Sensors J., vol. 16,
no. 15, pp. 5993–6002, Aug. 2016.
[13] M. Farhat, M. M. C. Cheng, K. Q. Le, and P. Y. Chen, “Nanoantenna
harmonic sensor: Theoretical analysis of contactless detection of mole-
cules with light,” Nanotechnology, vol. 26, p. 415201, Oct. 2015.
[14] H. Huang, M. Sakhdari, M. Hajizadegan, A. Shahini, D. Akinwande, and
P. Y. Chen, “Toward transparent and self-activated graphene harmonic
transponder sensors,” Appl. Phys. Lett., vol. 108, no. 17, p. 173503,
2016.
[15] J. R. Riley, A. D. Smith, D. R. Reynolds, and A. S. Edwards, “Tracking
bees with harmonic radar,” Nature, vol. 379, p. 29, Jan. 1996.
[16] E. A. Capaldi et al., “Ontogeny of orientation flight in the hon-
eybee revealed by harmonic radar,” Nature, vol. 403, pp. 537–540,
Feb. 2000.
[17] K. G. Ong and C. A. Grimes, “Tracking the harmonic response of
magnetically-soft sensors for wireless temperature, stress, and corrosive
monitoring,” Sens. Actuators A, Phys., vol. 101, pp. 49–61, Sep. 2002.
[18] A. Lazaro, R. Villarino, and D. Girbau, “A passive harmonic tag for
humidity sensing,” Int. J. Antennas Propag., vol. 2014, Jul. 2014,
Art. no. 670345.
[19] H. Huang, P. Y. Chen, C. H. Hung, R. Gharpurey, and D. Akinwande,
“A zero power harmonic transponder sensor for ubiquitous wireless μL
liquid-volume monitoring,” Sci. Rep., vol. 6, Jan. 2016, Art. no. 18795.
[20] H. Huang, P. Y. Chen, C. H. Hung, R. Gharpurey, and D. Akinwande,
“Frequency hopped wireless passive sensing system with harmonic
transponder antenna sensor,” in IEEE MTT-S Int. Microw. Symp. Dig.,
May 2015, pp. 1–4.
[21] A. K. Geim and K. S. Novoselov, “The rise of graphene,” Nature Mater.,
vol. 6, no. 3, pp. 183–191, 2007.
[22] K. S. Novoselov et al., “Electric field effect in atomically thin carbon
films,” Science, vol. 306, no. 5696, pp. 666–669, 2004.
[23] D.-W. Shin et al., “A facile route to recover intrinsic graphene over large
scale,” ACS Na, vol. 6, no. 9, pp. 7781–7788, 2012.
[24] H. Wang, Y. Wu, C. Cong, J. Shang, and T. Yu, “Hysteresis of electronic
transport in graphene transistors,” ACS Na, vol. 4, no. 12, pp. 7221–7228,
2010.
[25] M. Lafkioti et al., “Graphene on a hydrophobic substrate: Doping
reduction and hysteresis suppression under ambient conditions,” Nano
Lett., vol. 10, no. 4, pp. 1149–1153, 2010.
[26] P. L. Levesque et al., “Probing charge transfer at surfaces using graphene
transistors,” Nano Lett., vol. 11, no. 1, pp. 132–137, 2011.
[27] P. G. Collins, K. Bradley, M. Ishigami, and A. Zettl, “Extreme oxy-
gen sensitivity of electronic properties of carbon nanotubes,” Science,
vol. 287, pp. 1801–1804, Mar. 2000.
[28] J. Kong et al., “Nanotube molecular wires as chemical sensors,” Science,
vol. 287, no. 5453, pp. 622–625, 2000.
[29] M. S. Mannoor et al., “Graphene-based wireless bacteria detection on
tooth enamel,” Nature Commun., vol. 3, p. 763, Mar. 2012.
[30] F. Schedin et al., “Detection of individual gas molecules adsorbed on
graphene,” Nature Mater., vol. 6, pp. 652–655, Jul. 2007.
[31] C. A. Merchant et al., “DNA translocation through graphene nanopores,”
Nano Lett., vol. 10, pp. 2915–2921, Jul. 2010.
[32] P. Y. Chen and J. Jung, “PT symmetry and singularity-enhanced sensing
based on photoexcited graphene metasurfaces,” Phys. Rev. Appl., vol. 5,
p. 064018, Jun. 2016.
7246
[33] P. Y. Chen and A. Alù, “Atomically thin surface cloak using graphene
monolayers,” ACS Na, vol. 5, pp. 5855–5863, Jun. 2011.
[34] P.-Y. Chen and A. Alu, “Terahertz metamaterial devices based on
graphene nanostructures,” IEEE Trans. THz Sci. Technol., vol. 3, no. 6,
pp. 748–756, Nov. 2013.
[35] P. Y. Chen, H. Huang, D. Akinwande, and A. Alù, “Graphene-based plas-
monic platform for reconfigurable terahertz nanodevices,” ACS Photon.,
vol. 1, no. 8, pp. 647–654, 2014.
[36] C. Yu et al., “Quasi-free-standing bilayer epitaxial graphene field-effect
transistors on 4H-SiC (0001) substrates,” Appl. Phys. Lett., vol. 108,
no. 1, p. 013102, 2016.
[37] H. Wang, A. Hsu, K. K. Kim, J. Kong, and T. Palacios, “Gigahertz
ambipolar frequency multiplier based on CVD graphene,” in IEDM
Tech. Dig., vol. 1, Dec. 2010, pp. 572–575.
[38] M. E. Ramon et al., “Three-gigahertz graphene frequency doubler
on quartz operating beyond the transit frequency,” IEEE Trans.
Nanotechnol., vol. 11, no. 5, pp. 877–883, Sep. 2012.
[39] H. Wang, D. Nezich, J. Kong, and T. Palacios, “Graphene frequency
multipliers,” IEEE Electron Device Lett., vol. 30, no. 5, pp. 547–549,
May 2009.
[40] K. N. Parrish and D. Akinwande, “Even-odd symmetry and the con-
version efficiency of ideal and practical graphene transistor frequency
multipliers,” Appl. Phys. Lett., vol. 99, no. 22, p. 223512, 2011.
[41] Z. Wang, Z. Zhang, H. Xu, L. Ding, S. Wang, and L. M. Peng, “A high-
performance top-gate graphene field-effect transistor based frequency
doubler,” Appl. Phys. Lett., vol. 96, no. 17, pp. 17–19, 2010.
[42] Y. Wu et al., “High-frequency, scaled graphene transistors on diamond-
like carbon,” Nature, vol. 472, pp. 74–78, Apr. 2011.
[43] H. Huang et al., “Chemical-sensitive graphene modulator with a memory
effect for Internet-of-Things applications,” Microsyst. Nanoeng., vol. 2,
p. 15049, May 2016.
[44] A. Shahini, M. Hajizadegan, M. Sakhdari, M. M. C. Cheng,
P. Y. Chen, and H. H. Huang, “Self-powered and transparent all-graphene
biosensor,” in Proc. IEEE SENSORS, Oct. 2016, pp. 1–3.
[45] S. M. Sze, and K. K. Ng, Physics of Semiconductor Devices. Hoboken,
NJ, USA: Wiley, 2006.
[46] K. N. Parrish and D. Akinwande, “An exactly solvable model for the
graphene transistor in the quantum capacitance limit,” Appl. Phys. Lett.,
vol. 101, no. 5, p. 053501, 2012.
[47] J. Tian, A. Katsounaros, D. Smith, and Y. Hao, “Graphene field-effect
transistor model with improved carrier mobility analysis,” IEEE Trans.
Electron Devices, vol. 62, no. 10, pp. 3433–3440, Oct. 2015.
[48] G. M. Landauer, D. Jímenez, and J. L. González, “An accurate and
Verilog-A compatible compact model for graphene field-effect transis-
tors,” IEEE Trans. Nanotechnol., vol. 13, no. 5, pp. 895–904, Sep. 2014.
[49] S. Wang, P. K. Ang, Z. Wang, A. L. L. Tang, J. T. Thong, and K. P. Loh,
“High mobility, printable, and solution-processed graphene electronics,”
Nano Lett., vol. 10, no. 1, pp. 92–98, 2009.
[50] W. Zhu, V. Perebeinos, M. Freitag, and P. Avouris, “Carrier scattering,
mobilities, and electrostatic potential in monolayer, bilayer, and trilayer
graphene,” Phys. Rev. B, vol. 80, p. 235402, Dec. 2009.
[51] V. E. Dorgan, M. H. Bae, and E. Pop, “Mobility and saturation velocity
in graphene on SiO2 ,” Appl. Phys. Lett., vol. 97, no. 8, p. 082112, 2010.
[52] J. Martin et al., “Observation of electron-hole puddles in graphene
using a scanning single-electron transistor,” Nature Phys., vol. 4, no. 2,
pp. 144–148, 2008.
[53] S. A. Thiele, J. A. Schaefer, and F. Schwierz, “Modeling of
graphene metal-oxide-semiconductor field-effect transistors with gap-
less large-area graphene channels,” J. Appl. Phys., vol. 107, no. 9,
pp. 094505-1–094505-8, 2010.
[54] D. Jimenez and O. Moldovan, “Explicit drain-current model of graphene
field-effect transistors targeting analog and radio-frequency applica-
tions,” IEEE Trans. Electron Devices, vol. 58, no. 11, pp. 4049–4052,
Nov. 2011
[55] T. Fang, A. Konar, H. Xing, and D. Jena, “Carrier statistics and quantum
capacitance of graphene sheets and ribbons,” Appl. Phys. Lett., vol. 91,
no. 9, p. 092109, 2007.
[56] Advanced Design System. [Online]. Available: http://www.keysight.
com/en/pc-1297113/advanced-Design-System
[57] I. Heller, S. Chatoor, J. Männik, M. A. G. Zevenbergen, C. Dekker,
and S. G. Lemay, “Influence of electrolyte composition on liquid-
gated carbon nanotube and graphene transistors,” J. Amer. Chem. Soc.,
vol. 132, no. 48, pp. 17149–17156, 2010.
[58] Q. Cui et al., “ltrathin and atomically flat transition-metal oxide:
Promising building blocks for metal–insulator electronics,” ACS Appl.
Mater. Interfaces, vol. 8, no. 50, pp. 34552–34558, 2016.
IEEE SENSORS JOURNAL, VOL. 17, NO. 22, NOVEMBER 15, 2017
[59] K. Wei, Z. Zhang, W. Chen, and Z. Feng, “A novel hybrid-fed patch
antenna with pattern diversity,” IEEE Antennas Wireless Propag. Lett.,
vol. 9, pp. 562–565, 2010.
[60] D. M. Pozar, Microwave Engineering. Hoboken, NJ, USA: Wiley, 2009.
[61] A. K. Jain, J. Mao, and K. M. Mohiuddin, “Artificial neural networks:
A tutorial,” Computer, vol. 29, no. 3, pp. 31–44, Mar. 1996.
[62] N. Jones, “The learning machines,” Nature, vol. 505, no. 7482, p. 146,
2014.
[63] D. H. Ballard, G. E. Hinton, and T. J. Sejnowski, “Parallel visual
computation,” Nature, vol. 306, pp. 21–26, Nov. 1983.
[64] P. R. Gray, and R. G. Meyer, Analysis and Design of Analog Integrated
Circuits. Hoboken, NJ, USA: Wiley, 1990.
[65] M. Hagan, H. B. Demuth, M. H. Beale, O. D. Jess, and M. T. Hagan,
Neural Network Design. Seattle, WA, USA: Amazon, 2014.
[66] R. Hecht-Nielsen, “Theory of the backpropagation neural network,”
Neural Netw., vol. 1, pp. 445–448, Jan. 1988.
[67] C. Christodoulou and M. Georgiopoulos, Applications of Neural Net-
works in Electromagnetics. Norwood, MA, USA: Artech House, 2000.
[68] A. Patnaik and R. K. Mishra, “ANN techniques in microwave engineer-
ing,” IEEE Microw. Mag., vol. 1, no. 1, pp. 55–60, Mar. 2000.
[69] P. Y. Chen, C. H. Chen, and H. Wang, “A neural network: Family
competition genetic algorithm and its applications in electromagnetic
optimization,” Appl. Comput. Intell. Soft Comput., p. 4, Jan. 2009.
Mehdi Hajizadegan (S’17) received the M.Sc.
degree in electronics from Tarbiat Modares Univer-
sity, Tehran, Iran, in 2013. He is currently pursu-
ing the Ph.D. degree in electrical engineering with
Wayne State University, Detroit. Since 2015, he
has been a Graduate Research Assistant with the
Department of Electrical and Computer Engineering,
Wayne State University. His research mainly focuses
on graphene and other carbon-based biosensors,
PT-symmetry-based wearable biosensors and leaky
wave antennas, passive and active electromagnetic
metamaterials and plasmonics, and their applications for energy harvesting.
Maryam Sakhdari (S’17) received the B.Sc. degree
in electronic from the Iran University of Science and
Technology, Tehran, Iran, in 2010, and the M.Sc.
degree in electronic from Tarbiat Modares Univer-
sity, Tehran, in 2013. She is currently pursuing
the Ph.D. degree with the Department of Electrical
and Computer Engineering, Wayne State Univer-
sity, Detroit. Her Ph.D. thesis mainly focuses on
low-noise harmonic transponder sensors, ultrasensi-
tive PT-symmetric wireless sensors, electromagnetic
metamaterials and plasmonics, and their RF, THz,
infrared, and optical applications.
Liang Zhu is currently pursuing the Ph.D. degree with the Department of
Electrical and Computer Engineering, Wayne State University, Detroit.
Qingsong Cui received the Ph.D. degree in mea-
surement technology and instruments from Beihang
University, China, in 2013. He is currently a Post-
Doctoral Researcher with Wayne State University.
His research interests include RF engineering, wire-
less sensors, and nanophotonics.
HAJIZADEGAN et al.: GRAPHENE SENSING MODULATOR: TOWARD LOW-NOISE, SELF-POWERED WIRELESS MICROSENSORS 7247

Haiyu Huang (S’07) received the M.Sc. degree Jonathan C. H. Hung (M’15) received the
in electrical engineering from Columbia University, B.S. degree from National Chiao-Tung University,
New York, NY, USA, in 2008, and the Ph.D. degree Hsinchu, Taiwan, in 2001, the M.S. degree in
in electrical and computer engineering from the electronic engineering from National Taiwan Uni-
University of Texas at Austin, Austin, TX, USA, versity, Taipei, Taiwan, in 2003, and the Ph.D.
in 2014. He was a Research Assistant with the degree in electrical and computer engineering from
Methodist Hospital Research Institute. Since 2014, the University of Texas at Austin, in 2015. From
he has been a Member of Technical Staff with 2003 to 2009, he was a Design Engineer with Taiwan
Maxim Integrated Inc. His research interests include Semiconductor Manufacturing Company, Hsinchu,
wireless medical telemetry and novel nanoelectronic in embedded DRAM and RF circuit design. Since
devices for healthcare applications. 2014, he has been a Principal Member of Technical
Staff with Maxim Integrated Inc., leading a team engaged in the developments
of low-power RFIC. His research interest includes mixed-signal, analog, and
RF integrated circuits.

Pai-Yen Chen (S’09–M’13–SM’17) received the

Ph.D. degree from the University of Texas at Austin
in 2013. He was a Research Scientist with Intellec-
tual Ventures’ Metamaterial Commercialization Cen-
Mark M. C. Cheng is currently an Associate Pro- ter during 2013–2014, and a Member of Research
fessor with the Department of Electrical and Com- Staff with the National Nano Device Laboratory,
puter Engineering and the Department of Biomedical Taiwan, during 2006–2009. He is currently an Assis-
Engineering, Wayne State University. tant Professor with the Department of Electrical
and Computer Engineering, Wayne State University.
He has published approximately 70 peer-reviewed
papers (four journal covers), 65 conference proceed-
ings, six book chapters, one co-edited book, and nine U.S. patents. He has
been involved in multidisciplinary research on applied electromagnetics,
wireless telemetry sensors, integrated micro-/nano-sensor systems, and nano-
optoelectronics, including plasmonics and nanophotonics.