first principle studies on graphene fet
DESCRIPTION
Discussed drift-diffusion theory in graphene FET from first principles approach.TRANSCRIPT
-
A first principles theoretical examination of graphene-based field effecttransistorsJames G. Champlain
Citation: J. Appl. Phys. 109, 084515 (2011); doi: 10.1063/1.3573517 View online: http://dx.doi.org/10.1063/1.3573517 View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v109/i8 Published by the AIP Publishing LLC.
Additional information on J. Appl. Phys.Journal Homepage: http://jap.aip.org/ Journal Information: http://jap.aip.org/about/about_the_journal Top downloads: http://jap.aip.org/features/most_downloaded Information for Authors: http://jap.aip.org/authors
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
A first principles theoretical examination of graphene-based fieldeffect transistors
James G. Champlaina)
Electronics Science and Technology Division, Naval Research Laboratory, Washington, D.C. 20375, USA
(Received 6 October 2010; accepted 5 March 2011; published online 22 April 2011)
This paper presents an in-depth theoretical examination of graphene-based field effect transistors,
looking at thermal statistics, electrostatics, and electrodynamics. Using a first principles approach,
the unique behavior observed in graphene-based field effect transistors, such as the V-shaped
transfer characteristic, limited channel pinch-off, and lack of off-state (under gate modulation), are
described. Unlike previous attempts, a description of both drift and diffusion currents in the device
is presented. The effect of external resistance on steady-state and high-frequency performance is
examined. Comparisons of the theoretical results to experimental results are made and show good
agreement. Finally, the theoretical work in this paper is used as a basis to discuss the possible
source of some observed behavior in practical graphene-based field effect transistors. VC 2011American Institute of Physics. [doi:10.1063/1.3573517]
I. INTRODUCTION
There has been much interest as of late in graphene and
its application in electronic devices, specifically field effect
transistors (FETs). Within the past decade there has been
increased work in the area of general thermal statistics and
fundamental transport for graphene.17 Also within the past
few years there has been initial work in the description and
modeling of device physics.812 Some of this work has incor-
rectly interpreted or applied the fundamental transport
physics to the conductivity of graphene sheets, particularly
when applied to gated structures, such as FETs, resulting in
the incorrect interpretation of device behavior, which will be
discussed.
This paper presents an in-depth theoretical examination
of the behavior of graphene-based FETs (gFETs) using pre-
vailing theory regarding fundamental graphene charge carrier
transport and first principles solid-state physics, including
thermal statistics, electrostatics, and electrodynamics. A de-
scription of current transport in a gFET is presented along
with standard FET parameters and characteristics, such as the
transfer characteristic, transconductance, drain characteristic,
and drain conductance, with discussions on several unique
phenomena observed in graphene (e.g., V-shaped transfer
characteristic, limited channel pinch-off, lack of an off-state).
The frequency performance and limits of these devices, in the
form of short-circuit current-gain cutoff frequency, are exam-
ined. The influence of external resistance and the observed
effects on gFET steady-state behavior (e.g., sublinear transfer
characteristic and transconductance suppression at high drain
currents) and high-frequency performance is explored. The
validity of the work presented in this paper is discussed, with
reference to experimental and other theoretical work. Specifi-
cally, comparisons to measured device data are made, show-
ing good agreement with theory, and the nature of the
conductivity of the graphene channel and its correct represen-
tation are discussed. Finally, the theoretical work in this paper
is used as a basis to discuss the possible sources of some
observed behavior in practical graphene-based FETs not
directly described by this work, such as some observed asym-
metric transfer characteristics and device performance.
II. GRAPHENE THERMAL STATISTICS
The thermal statistics of graphene have been previously
developed and presented.13,7,13 It is offered here as a foun-
dation and reference for the work within this paper.
As a result of graphenes two-dimensional (2D), hexag-
onal lattice structure, its band structure is unique among
semiconductors (Fig. 1). Some of graphenes distinctive fea-
tures are its lack of a bandgap (i.e., the so-called zero
bandgap) and linear energymomentum dispersion relation
about the zero momentum point,1,3,7,13
e eD 6hmFjkj: (1)Here, e is the energy of the particular momentum state withwave number k, eD is the Dirac energy of the graphene sheet(i.e., the zero-momentum energy, where conduction band
and valence band meet, eD ec ev), h is the reducedPlanks constant, and mF 106 m/sec is the Fermi velocityof carriers in graphene.2,3,7 The positive sign in Eq. (1) leads
to energies in the conduction band (i.e., e eD), and the neg-ative sign results in energies in the valence band (i.e.,
e eD). Arguably, graphene sits in a unique positionbetween a classical semiconductor (eg > 0) and a semimetal(ec < ev).
For such a band structure, it has been shown that the
conduction band and valence band density of states are linear
with energy:
DCe gsgv2p
e eDhvF 2
; (2a)a)Electronic mail: [email protected].
0021-8979/2011/109(8)/084515/19/$30.00 VC 2011 American Institute of Physics109, 084515-1
JOURNAL OF APPLIED PHYSICS 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
DV e gsgv2p
eD ehvF 2
; (2b)
with gs 2 and gv 2 being the spin and valley degenera-cies, respectively.3,7 The electron (sheet) concentration, n,and hole (sheet) concentration, p, are given by
n 1eD
DC e f e de NGF 1 gF ; (3a)
p eD1
DV e 1 f e de NGF 1 gF : (3b)
f e is the FermiDirac distribution, which dictates the occu-pational probability of a state depending on its displacement
in energy from the Fermi energy, eF. NG is the effective gra-phene (sheet) density of states given by
NG gsgv2p
kBT
hmF
2; (4)
with kB being Boltzmanns constant and T being the absolutetemperature of the system in Kelvin. F 1 6gF is the com-plete FermiDirac integral with index i 1:
F i 6gF 1
C i 1 10
ui
1 egFeu du; (5)
C is the gamma function. gF, introduced in Eq. (3), is arelative measure of the Fermi energy, eF, with respect to theDirac energy, eD, normalized to kBT:
gF eF eDkBT
: (6)
The complete FermiDirac integral is commonly approxi-
mated by
F i 6gF 1
C i 1 10
uie6gFeudu e6gF : (7)
This is because, in most cases, the Fermi energy lies well
outside the energetic range of states of concern (e.g., the
semiconductor is nondegenerate). In such instances, the
Fermi-Dirac distribution function can be replaced by
the simpler MaxwellBoltzmann distribution function. In the
case of graphene, the Fermi energy lies near to or within the
energetic states of concern. Therefore, except in specific
cases (e.g., very low temperature, very high carrier concen-
trations, very low carrier concentrations) the complete
FermiDirac integral as presented in Eq. (3) cannot be
replaced by approximations.
Figure 2 shows the electron concentration, n, and holeconcentration, p, dependence on Fermi energy, eF.
III. GRAPHENE FET ELECTROSTATICS
In practice, there have been two basic device layouts
employed in the fabrication of gFETs.1416 One is akin to a
classic silicon (Si) MOSFET layout, in which the gate metal
and oxide cover the entire length of the channel from the
source contact to the drain contact.14,16 The other is similar
to standard IIIV MESFET/HEMT layouts, in which the
gate metal sits between the source and drain contacts with an
amount of access region on either side of the gate. In the lat-
ter case, the gate oxide tends to cover the entire expanse
between the source and drain contacts.14,15 The specific lay-
out of the gFET is not critical to the work presented in this
paper.
Additionally, three methods have emerged as the domi-
nant means of growing or producing graphene for device
applications: silicon sublimation from silicon carbide (SiC),
graphene exfoliation from a graphite source, and graphene
chemical vapor deposition (CVD), commonly onto a metal
substrate.6,1422 The first method results in a graphene sheet
in intimate contact with a SiC substrate.6,14 The second and
third methods require minor subsequent processing but com-
monly result with a graphene sheet resting on a dielectric,
usually silicon dioxide (SiO2) on a Si substrate. In many
cases, the Si substrate is highly doped and used as a back
gate for the gFET.15,16,21,22 For the bulk of this paper, gra-
phene on SiC is considered. In general, this impacts only the
final numerical results, the physics governing current trans-
port and overall behavior of the gFET remain essentially
unchanged (See Appendix A for a treatment of graphene on
a dielectric substrate with back gate).
FIG. 2. (Color online) Plot of electron concentration (n) and hole concentra-tion (p) versus Fermi energy relative to Dirac energy (eF eD). The dashedlined corresponds to the electron concentration using the MaxwellBoltz-
mann distribution function, Eq. (7).
FIG. 1. (Color online) Unique band structure (e k diagram) of graphene,highlighting the linear energy-momentum dispersion of the material near
zero momentum.
084515-2 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
Figure 3 shows the physical structure of the gFET, under
the gate, used in this paper. As shown, the graphene sheet,
which constitutes the channel of the FET, lies within the
xz-plane. The source edge of the gate metal is located atx 0, the drain edge is located at x L. The width of thegate metal is W. tox is the thickness of the gate oxide. Thegate is treated as an equipotential with voltage VG. The volt-age at the drain edge (x L) of the graphene sheet (channel)is VD; at the source edge (x 0), VS.
In examining the electrostatics and electrodynamics of
the gFET, the gradual channel approximation (i.e., @Dy=@y @Dx=@x) is applied. This includes the assumption that allproperties of the gFET are uniform along the width of
the gate (i.e., z-direction).23,24 Therefore, Gausss law(r D q) becomes separable in all three dimensions and
dDxdx
0; (8a)
dDydy
q x; y ; (8b)
dDzdz
0: (8c)
Here D is the displacement field vector with Dx, Dy, and Dzbeing its components in the x-, y-, and z-directions, respec-tively. q x; y is the volume charge density at locationx; y. A band diagram and charge profile for a vertical slice(in the yz-plane) across the metal-oxide-graphene structureof Fig. 3, located at x, is presented in Fig. 4.
In the band diagram of Fig. 4, Vch Vch x is the (local)voltage in the graphene channel. As defined earlier,
eF qVch is the (local) Fermi energy within the channel(with q being the elementary electric charge) and eD eD x is the (local) Dirac energy. /mo vox vm and/so vox vsc are the metal-oxide and the semiconductor-oxide offsets, respectively, with vox, vm, and vsc being the ox-ide, metal, and semiconductor (graphene) electron affinities,
respectively. Eox Eox x is the (local) electric field withinthe oxide.
In the charge profile of Fig. 4, Qm Qm x is the chargesheet density in the metal, n n x is the electron concentra-tion within the graphene channel with qn being its associ-ated charge sheet density, p p x is the hole concentrationwithin the graphene channel with qp being its associatedcharge sheet density, and Qf is the fixed charge sheet densityrepresenting any fixed charge within the graphene sheet and/
or oxide, imaged to the graphene sheet. Qnet Qnet x q p n is simply the net mobile charge sheet densitywithin the graphene channel.
For the structure to be charge neutral (i.e., the electric
field goes to zero outside the structure), the total charge must
sum to zero:
Qm Qnet Qf 0: (9)To find a relationship between the mobile charges in the
channel and the applied voltages, a voltage (potential
energy) loop around the metal-oxide-graphene band diagram
is completed. Beginning at the gate metal and proceeding
around the loop,
q VG Vch /mo qEoxtox /so eD eF 0:(10)
The electric field in the oxide, Eox, can be found by integrat-ing Eq. (8b) across the gate metal-oxide interface:0
eoxEox
dDy toxtox
Qmd y tox dy: (11)
FIG. 3. Physical structure of the gFET examined in this paper. In addition
to the various spatial dimensions used in the analysis of the device, the
applied voltages (VG, VD, VS) and associated current densities (JT , Jn, Jp,see Sec. IV) are indicated. The band diagram and charge profile along the
section line a a0 is presented in Fig. 4.
FIG. 4. Band diagram (top) and charge profile (bottom) of the gFET metal-
oxide-graphene structure corresponding to the section line a a0 in Fig. 3.
084515-3 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
Here, d y tox is a Dirac delta function locating the chargein the metal, Qm, at the gate metal-oxide interface, and t
ox
and tox are positions just to either side of the interface aty tox. eox er; oxeo is the permittivity of the oxide layerwhere er; ox is the relative permittivity of the oxide and eo isthe free-space permittivity. The result of Eq. (11), substitut-
ing Qm Qnet Qf
from Eq. (9), is
Eox 1eox Qm 1
eoxQnet Qf
: (12)
Recalling gF eF eD =kBT from Eq. (6) and with somealgebraic manipulation, Eq. (10) becomes the charge-voltage
(QV) relationship for the gFET,
VG Vch 1Cox
Qnet kBTq
gF 1
CoxQf 1
q/mo /so 0:
(13)
Here, Cox eox=tox is the areal capacitance of the gate oxide.Equation (13) gives a unique relationship between the
voltages VG and Vch and the normalized, relative position ofthe Fermi energy with respect to the Dirac energy, gF; which,through Eq. (3), gives the electron, n, and hole, p, concentra-tions within the graphene channel. Because an assumption
on the hole and electron concentrations with relation to the
Fermi energy cannot be made (see Sec. II), the expressions
for n and p in Eq. (3), and associatively Qnet, cannot bereduced, leaving Eq. (13) transcendental.
The electron and hole concentrations and normalized,
relative Fermi energy (gF) are plotted as a function ofgate voltage, VG, with the channel grounded (VD VS Vch 0 x L 0), for various fixed charge densities(Qf), in Fig. 5. Taking a Ti/Au gate metal and Al2O3 gate ox-ide,14,22,25 the values used for the various device parameters
are listed in Table I.
A. Metal-oxide-graphene capacitive network
The metal-oxide-graphene structure as shown in Figs. 3
and 4 corresponds to an equivalent capacitive network,
which is a useful tool in analyzing the devices behavior.
Figure 6 shows the capacitive network with CT @Qm=@VG; ch 1=Cox 1=CQ
1being the total capacitance of
the structure, Cox @Qm=@VG; gr eox=tox being the capaci-tance of the gate oxide, and CQ @Qnet=@Vch; gr Cq F 0 gF F 0 gF being the quantum capacitance ofthe graphene sheet (with Cq q2NG=kBT); withVG; ch VG Vch as the voltage across the entire metal-ox-ide-graphene structure, VG; gr VG Vgr (with qVgr eD)as the voltage across the gate oxide, and Vch; gr Vch Vgr kBT=q gF as the voltage across the gra-phene sheet.
IV. GRAPHENE FET ELECTRODYNAMICS
The total electron current density (Jn) and total hole cur-rent density (Jp) are defined using the following equations:
Jn lnnreF; (14a)
Jp lppreF: (14b)
And the total current density is
JT Jn Jp: (15)
In Eq. (14), ln and lp are the electron and hole mobilitieswithin the graphene channel, respectively.
Under nonequilibrium conditions, a single Fermi energy
(eF), as in Eq. (14), cannot strictly be assumed. Instead, theelectron and hole concentrations and currents should follow
their respective and distinct quasi-Fermi energies (eF; n andeF; p). However, a general analysis with two quasi-Fermi
FIG. 5. (Color online) Plot of electron concentration (n), hole concentration(p), and normalized, relative Fermi energy (gF) versus applied gate voltage(VG) for fixed charge densities (Qf =q) of (a) 0, (b) 5 1012 cm2, and (c)5 1012 cm2.
TABLE I. Device parameters and dimensions used in this paper, unless oth-
erwise noted.
Parameter Value
er; ox er;Al2O3 a 10tox (A) 200
vTi (eV)b 4.33
vgraphene (eV)c 4.5
Qf =q (cm2) 0
vF (m/sec) 106
L (lm) 1ln (cm
2/Vsec) 5000
lp (cm2/Vsec) 5000
aReferences 26 and 27.bReference 28.cReferences 29 and 30.
084515-4 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
energies can quickly become intractable. Alternatively, if it
is assumed that the conduction band and valence band states
are in intimate communication (i.e., high generation/recom-
bination rate), which is highly possible in a zero bandgap
semiconductor where recombination lifetimes are very
short,3133 then the electron and hole quasi-Fermi energies
will deviate only slightly from each other. The assumption of
a single Fermi energy allows the analysis to become manage-
able while introducing only minimal error.
By using the definition that current density is propor-
tional to the gradient of the Fermi energy (J / reF), currentresulting from electric fields (i.e., drift current) and from
concentration gradients (i.e., diffusion current) are captured
within one expression.34 As will be shown in Sec. V, diffu-
sion current can comprise an appreciable portion of the total
current flowing through the channel of a gFET, and has been
commonly left unaddressed in previous discussions of cur-
rent transport.8,9,11
Within this paper, constant mobility is assumed. Present
theory takes transport in graphene to be limited by scattering
from impurities and phonons, with impurity scattering being
the dominant scattering mechanism.47 The scattering time
has been found to be inversely proportional to the impurity
concentration, which subsequently leads to a conductivity
proportional to the carrier concentration and inversely
proportional to the impurity concentration. Expressing the
conductivity as a product of the mobility and charge concen-
tration (r qln) results in the mobility being a constant fora specific impurity concentration. A continued discussion of
mobility and conductivity is presented in Sec. IX.
Due to the 2D nature of graphene, current only flows
within the plane of the graphene sheet (xz-plane), ignoringgate leakage. Additionally, since the gFET is uniform along
the gate width (z-direction), the gradients in Eq. (14) reduceto one-dimensional (1D) derivatives in the x-direction:
JT Jn Jp lnndeFdx
lppdeFdx
: (16)
Recalling that eF qVch, the total current density may bewritten as
JT Jn Jp qlnn dVchdx
qlpp
dVchdx
: (17)
Assuming no sources or sinks of charge within the channel
(r JT @q=@t 0), the total (steady-state) current den-sity must be constant. Therefore, multiplying Eq. (17) by dx,integrating from source (x 0) to drain (xL), and dividingby the gate length (L) results in
JT 1L
L0
Jndx 1L
L0
Jpdx Jn Jp; (18)
where Jn and Jp are the mean values of the total electron cur-rent and total hole current along the channel, respectively,
which upon evaluation are given by
Jn 1L
L0
Jndx 1LIngDgS; (19a)
Jp 1L
L0
Jpdx 1LIpgDgS; (19b)
with gS gF x 0 and gD gF x L as the normalized,relative Fermi energy at the source and drain edges of the
graphene channel, respectively, and
I n I n gF lnkBTNGF 2gF
CqCox
F 1gF2
gFF 2gF F 3gF; (20a)
I p Ip gF lpkBTNGF 2 gF
CqCox
F 1 gF 2
gFF 2 gF F 3 gF ; (20b)
with Cq q2NG=kBT (as defined in Sec. III A).Defining the the drain current (density), JD, as the cur-
rent that flows directed from drain to source, assuming
VD > VS, the drain current is then
JD JT Jn Jp
JD JD; n JD; p; (21)
with JD; n Jn 1=L I ngSgD
and JD; p Jp 1=L I pgSgD. The associated transconductance, gm, is given by
gm @JD@VGS
VDS
1Lqln nS nD qlp pS pD ; (22)
with VGS VG VS, VDS VD VS, nD n x L , nS n x 0 , pD p x L , and pS p x 0 . The drain con-ductance, gd, is
gd @JD@VDS
VGS
1L
qlnnD qlppD
: (23)
By means of the QV relationship in Eq. (13), to determine gFwith respect to VG and Vch and Eqs. (20) and (21), the draincurrent density in the gFET channel can be solved numeri-
cally as a function of applied biases VG, VD, and VS. Thetransconductance and drain conductance can be similarly
numerically solved.
Using the values in Table I (e.g., L 1 lm,ln lp 5; 000 cm2/Vsec) and Qf 0, the transfer
FIG. 6. Metal-oxide-graphene capacitive network.
084515-5 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
characteristic and transconductance and drain characteristic
of a gFET are shown in Figs. 7 and 8, respectively.
In Fig. 7(a), the characteristic V-shape of the gFET
transfer characteristic can clearly be seen. VDirac, the so-called Dirac voltage (i.e., the voltage associated with mini-
mum current, zero transconductance),20,35,36 given by
VDirac VGSjgm0 1
1 ln=lpVDS 1
CoxQf 1
q/mo /so ;
(24)
is not constant and shifts depending on bias conditions. Also,
as can be clearly seen, material related characteristics, such
as Qf, /mo, and /so, directly impact the value of VDirac. Thedirect dependence of VDirac on the fixed charge density (Qf),which is in essence the impurity concentration within the
structure, has been experimentally observed.5,10 Addition-
ally, Eq. (24) suggests that work function engineering (i.e.,
selection of gate metal based on its work function, /mo, orelectron affinity, vm) is a possible method for controlling theDirac voltage.
The transconductance (Fig. 7(b)) asymptotically
approaches a maximum value of gm;max lnCox=L VDSand a minimum value of gm;min lpCox=L
VDS. The
slope of the transconductance at the Dirac voltage is
@gm=@VGSjVGSVDirac ln lp
Cox=L. Though not a standard
figure of merit, @gm=@VGSjVDirac is useful as it is independentof drain-source bias (VDS) and gives insight into some physi-cal parameters (e.g., ln, lp) of the gFET. The derivationof the preceding equations (Eqs. (18)(24), gm;max, gm;min,@gm=@VGSjVDirac ) can be found in Appendix B.
In the drain characteristics of Fig. 8(a), a current satura-
tion is observed. This saturation is not a result of velocity
saturation within the graphene channel, which has not been
included in this work (see Sec. VIII), but the creation of a
highly resistive region within the graphene channel associ-
ated with the pinch-off (following Shockleys terminology)23
and transition of the majority charge carrier channel. Further
discussion is found in Sec. V.
In Fig. 9, the transfer characteristic is replotted, explic-
itly showing the relative electron and hole contributions to
the drain current. As has been previously theorized, at gate-
source voltages above the Dirac voltage (VGS > VDirac), thetotal current is dominated by electron current; below
the Dirac voltage (VGS < VDirac), the total current is domi-nated by hole current. Around VGS VDirac, the total currentis a combination of electron and hole currents with the
electron and hole current contributions being equal at
VGS VDirac.2,7,17,37
V. GRAPHENE FET ELECTROSTATIC ANDELECTRODYNAMIC PROFILES
As the gFET exhibits behavior uncommon in traditional
FETs, insight into this behavior can be gained by examining
the devices electrostatic and electrodynamic profiles along
FIG. 7. (Color online) (a) Plot of the transfer characteristics (JD versus VGS)for a gFET at three drain-source voltages (VDS 0:1; 1; 2:5V). (b) Plot ofthe associated transconductance (gm). The dashed line represents the asymp-totic behavior of the transconductance for the VDS 2:5 V curve, asdescribed by gm;max, gm;min, and @gm=@VGSjVGSVDirac .
FIG. 8. Plots of the drain characteristics (JD versus VDS) for the gFET: (a)VGS 0 to 4 V at 0.5 V steps, (b) VGS 0 to 4 V at 0.5 V steps. Theheavy line in both plots indicates the VGS 0 V curve. The dashed line in(a) corresponds to VGS 1:5 V, the gate-source bias used for Figs. 12and 13.
084515-6 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
the length of the channel. Although there is not a direct rela-
tion between the position within the channel and the various
device properties (as dervied in this paper), it is possible to
parametrically relate and plot the various properties with
respect to position within the channel as a function of the
local normalized, relative Fermi energy, gF.Given drain and source voltages (VD and VS, respec-
tively), the channel voltage (Vch) varies between these twovalues: Vch VD; VS . With the addition of gate voltage(VG), the normalized, relative Fermi energy at the drain andsource (gD and gS, respectively) may be found via Eq. (13)with gF in the channel varying between the two:gF gD; gS . The position along the channel, 0 x L,can be found, given gF, by
x gF LI n Ip gF
gS
I n I p gD
gS
: (25)
As previously defined, the Fermi energy in the channel is
directly related to the channel voltage, with the Dirac energy
related to the Fermi energy via gF: eF qVch andeD eF kBTgF.
The electron (n) and hole (p) concentrations, as func-tions of gF, are found using Eq. (3). The drain currentthroughout the channel is constant with respect to position
and given by Eqs. (20) and (21). The total electron and hole
components of the drain current are, respectively,
JD; n gF JDlnn
lnn lpp; (26a)
JD; p gF JDlpp
lnn lpp: (26b)
And the electron diffusion and drift current components are
JdiffD; n gF JD; n1
1 CQ=Cox ; (27a)
JdriftD; n gF JD; n JdiffD; n JD; n1
1 Cox=CQ ; (27b)
with n replaced by p for the hole components. Recall,CQ Cq F 0 gF F 0 gF is the quantum capacitance ofthe graphene sheet (see Sec. III A).
The electric field along the channel, Ech Ex, is
Ech gF JdriftD; nqlnn
JdriftD; p
qlpp: (28)
Note, the negative sign in Eq. (28) indicates that the field is
directed from drain to source, as expected for VD > VS. Thederivation of the preceding equations can be found in
Appendix C.
In Fig. 10, the band diagram, electron and hole concen-
trations, total electron and hole current densities, and electric
field in the channel versus position within the gFET channel
for a constant drain-source voltage (VDS 1V) and varyinggate-source voltage (VGS 0:5; 0; 0:33; 0:6; 1:25V; iv inFig. 9) are plotted. The evolution from an essentially p-chan-nel (Fig. 10, column i), to a mixture of p- and n-channels(Fig. 10, columns iiiv), to a n-channel (Fig. 10, column v) isclearly seen in the plots of band diagram, carrier concentra-
tion, and currents as the gate-source voltage swings from
less than the Dirac voltage (VGS < VDirac), to around theDirac voltage (VGS VDirac), to greater than the Dirac volt-age (VGS > VDirac).
The electric field within the channel (Ech) is always neg-ative (i.e., directed from drain to source). An interesting
behavior is how the peak electric field is located at the drain
edge for completely n-channel biases and located at thesource edge for completely p-channel biases, both of whichare expected, but the peak electric field is located at the loca-
tion where the Fermi energy crosses the Dirac energy
(eF x eD) for biases resulting in a mixed channel and fol-lows this point as it moves between drain and source. Such
behavior has also been proposed and experimentally
observed by Freitag et al. by monitoring the thermal emis-sions from a graphene channel under bias.38
Figure 11 shows an expanded view of the electron and
hole currents, with their drift and diffusion components, of
the current plots from Fig. 10 (row c). As can be seen in thefigure, although the majority of the current is comprised of
drift current, diffusion current comprises a nonnegligible
portion of the total current, across all biases.
In Fig. 12, the band diagram, electron and hole con-
centrations, total electron and hole current densities, and
electric field in the channel versus position within the gFET
channel for a constant gate-source voltage (VGS 1:5V, thedashed line in Fig. 8(a)) and varying drain-source voltage
(VDS 0:1; 2; 4V) are plotted. The saturation observed inFig. 8(a) is due to the pinch-off and transition of the majority
charge carrier channel, which eventually leads to a mixed
carrier channel. As a result of the pinch-off and transition of
the majority carrier channel, a high resistivity region within
the channel is formed. It is this high resistivity region that
leads to the observed saturation.
FIG. 9. (Color online) Plot of the transfer characteristic for the gFET
(VDS 1V) explicitly showing the (mean value) electron current (JD; n) andhole current (JD; p) contributions to the drain current (JD). The triangles (~)with labels indicate the bias points used for Figs. 10 and 11.
084515-7 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
Specifically, for a given gate-source voltage, as the
drain-source voltage is increased, the majority carrier concen-
tration decreases (electrons in Fig. 12(b), columns i and ii),accompanied by a slight increase in minority carrier
concentration (holes), eventually leading to a situation similar
to pinch-off in classical FETs (Fig. 12(b), column ii; nD inFig. 13(a); and JD; n in Fig. 13(b)).
23 Associated with this
pinch-off of the majority carrier channel is a decrease in the
drain conductance (gd) to its minimum (Fig. 13(c)). As thedrain-source voltage is further increased, the majority and
FIG. 10. (Color online) Plots of the (a) band diagram (eD: solid line, eF: dashed line), (b) electron and hole concentrations (n: solid symbols, p: hollow sym-bols), (c) electron and hole current densities (JD; n: solid symbols, JD; p: hollow symbols), and (d) electric field in the channel (Ech) versus position within thechannel for a constant VDS 1 V and (i) VGS 0:5V, (ii) VGS 0V, (iii) VGS VDirac 0:33V, (iv) VGS 0:6V, and (v) VGS 1:25V, indicated in Fig. 9by the triangles (~). All axes are linear.
FIG. 11. (Color online) Plots of the electron and hole current densities
(JD; n: solid circles, JD; p: hollow circles) and their respective drift (squares)and diffusion current (triangles) components. The bias conditions corre-
spond to those for the first three columns in Fig. 10: (a) VGS 0:5V, (b)VGS 0V, and (c) VGS 0:33V, with VDS 1 V for all plots.
FIG. 12. (Color online) Plots of the (a) band diagram (eD: solid line, eF:dashed line), (b) electron and hole concentrations (n: solid symbols, p: hol-low symbols), (c) electron and hole current densities (JD; n: solid symbols,JD; p: hollow symbols), and (d) electric field in the channel (Ech) versus posi-tion within the channel for a constant VGS 1:5 V (corresponding to thedashed line in Fig. 8(a)) and (i) VDS 0:1V, (ii) VDS 2V, and (iii)VDS 4V. All axes are linear.
084515-8 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
minority carriers exchange roles and a new majority car-
rier channel is formed (Fig. 12(b), column iii; pD in Fig.13(a); and JD; p in Fig. 13(b)), alleviating the observed satura-tion (JD in Fig. 13(b)) and resulting in an associated increasein drain conductance, gd (Fig. 13(c)). A similar explanationfor this observed saturation for gFETs has been postulated
before by Meric et al.9
VI. INFLUENCE OF RESISTANCE
The general influence of external drain and source resist-
ance on the performance of the gFET will be examined in
this section. The specific source of the resistance is not con-
sidered here. A discussion of possible sources of resistance
is discussed in Sec. VIII.
Figure 14 shows a schematic of a gFET with external
drain and source resistances. In general, the addition of
external resistance affects the performance of gFETs in the
same manner as conventional FETs in other material sys-
tems, that is, it reduces the voltage drop across the intrinsic
portion of the device through simple voltage division, effec-
tively shifting the internal bias point of the device.
Using an iterative method to solve for the internal
biases, Fig. 15 shows the effect of various drain and source
resistances on the transfer characteristic of the gFET. As
expected, in both cases, the drain current is reduced for a
given drain-source (VDS) and gate-source (VGS) bias point.Sub-linear transfer characteristics have been observed exper-
imentally,15,22,39,40 and their basis as a resistively limited
current provides an uncomplicated explanation for the
observed behavior, without the need to invoke new physics
(e.g., alternative scattering methods).
The extrinsic transconductance (gm; ext) can be deter-mined using the classic, small-signal FET model (Fig. 16).
Evaluating at low frequency (f ! 0), the extrinsic transcon-ductance is
gm; ext gm1 gmRS gd RS RD : (29)
Plots of the extrinsic transconductance versus gate-source
voltage with different drain and source resistances are shown
in Fig. 17. As with the transfer characteristic, the transcon-
ductance is essentially reduced for a given bias point.
FIG. 13. (Color online) Plots of the (a) electron (nD) and hole (pD) concen-tration at the drain, (b) drain current (JD) and its (mean value) electron(JD; n) and hole (JD; p) components, and (c) drain conductance (gD) versusdrain-source voltage (VDS) for VGS 1:5 V.
FIG. 14. Schematic diagram of a gFET with extrinsic drain (RD) and source(RS) resistances. The portion bounded by the dashed line corresponds to theintrinsic gFET.
FIG. 15. (Color online) Plots of the gFET transfer characteristic
(VDS 2:5V) under the influence of various (a) drain and (b) source resis-tances. In both cases, the resistance is increased as 0.01 X mm, 0.1 X mm,and 1 X mm.
FIG. 16. Classic, small-signal model for a FET. The portion bounded by the
dashed line corresponds to the intrinsic FET. It should be noted that the vol-
tages in this figure are small-signal (ac) voltages, not bias-point (dc) voltages
(e.g.,VG t VG ~vG t ).
084515-9 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
Transconductance curves of this form have also been
observed experimentally,15,22,40 with the collapse of the ex-
trinsic transconductance at higher gate-source voltages being
a direct result of the high drain conductance (gd) seen atthese biases. As the drain conductance increases at higher
gate-source and/or drain-source biases, it effectively shorts
out the intrinsic device, leading toward the FET appearing
simply as the series combination of the drain and source
resistances, with the influence of the gate voltage on the
drain current removed.
VII. FREQUENCY PERFORMANCE
Using the same small-signal model for a FET (Fig. 16),
the intrinsic short-circuit current-gain cut-off frequency
(fs; int) is commonly given as fs; int gm=2p Cgs Cgd
gm=2pCg, where Cgs is the gate-source capacitance, Cgd isthe gate-drain capacitance, and Cg Cgs Cgd is the totalgate capacitance. In defining fs; int for the gFET, the moreappropriate form of
fs; int 12p
gmCg
2s jgmj
2pCg(30)
is used with the (intrinsic) transconductance (gm) as definedin Eq. (22). In the derivation of the common form of fs; int, itis assumed that the transconductance is always positive,
allowing the transconductance to be moved outside of the
square root in Eq. (30) without taking its absolute value.
Since graphene-based FETs have demonstrated both positive
and negative transconductance, this simplification cannot be
made, giving fs; int the form shown in Eq. (30).The corresponding extrinsic short-circuit current-gain
cut-off frequency (fs; ext) (assuming Cgd Cgs Cg) isgiven by
fs; ext jgmj2pCg 1 gd RS RD
fs; int1 gd RS RD ; (31)
with the drain conductance (gd) as defined in Eq. (23)The total (linear) gate capacitance (Cg), in Eqs. (30) and
(31), is defined as
Cg @Qgate@VG
VD;VS
@Qgate@VGS
VDS
; (32)
with Qgate L0Qmdx. Therefore,
Cg @Qgate@VGS
VDS
L0
@Qm@VG; ch
dx L0
CTdx; (33)
with CT 1=Cox 1=CQ 1
being the total (areal) capaci-
tance associated with the metal-oxide-graphene structure
(Sec. III A).
Performing the integration of Eq. (33) gives
Cg q2N2gJT
lnFn lpFpgDgS (34)
with
Fn Fn gF F 1 gF 2 gFF 2 gF F 3 gF ; (35a)Fp Fp gF F 1 gF 2
gFF 2 gF F 3 gF : (35b)
A plot of Cg versus gate-source voltage (VGS) is shownin Fig. 18. The minimum gate capacitance occurs at the
Dirac voltage, VDirac, and approaches its maximum value(Cg;max CoxL) for VGS VDirac and VGS VDirac.
Figure 19 shows a plot of fs; int and fs; ext versus drain cur-rent (JD) (RS RD 0:05 X mm for fs; ext). As can be seenin the plot, fs; int and fs; ext both track the transconductance.
FIG. 17. (Color online) Plots of extrinsic transconductance (gm; ext) versusgate-source voltage (VGS) under the influence of various (a) drain and (b)source resistances. In both cases, VDS 2:5V and the resistance is increasedas 0.01 X mm, 0.1 X mm, and 1 X mm.
FIG. 18. (Color online) Total gate capacitance (Cg) (normalized to its maxi-mum value, Cg;max CoxL) versus gate-source voltage (VGS) forVDS 0:1; 1; and 2:5 V.
084515-10 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
The maximum values of fs; int (for a given drain-sourcevoltage, VDS) are
f maxs;int
gm;min 2pCg;max
VGS VDiracgm;max 2pCg;max
VGS VDirac
8>>>>>:
1
2pLlp
VDSL
VGS VDirac
1
2pLln
VDSL
VGS VDirac
8>>>>>: (36)
fs; int has two maximum values as the electron and holemobilities, and subsequently jgm;minj lpCox=L
VDS and
jgm;maxj lnCox=L VDS, are not necessarily equal. Fig-ure 20 shows a plot of the maximum intrinsic short-circuit
current-gain cut-off frequency (f maxs; int) versus gate length (L).
Assuming that the physically maximum (group) velocity
achievable in the channel is the Fermi velocity, then the min-
imum transit time under the gate is sf ;min L=vF. The asso-ciated short-circuit current-gain frequency is
fs; vF 1
2psf ;min vF
2pL: (37)
For fs; vF values of 100 GHz and 1 THz, Eq. (37) suggests agate length no greater than L 1:6 lm and L 0:16 lm,respectively. Clearly, Eq. (37) represents an upper asymptote
for f maxs; int (Fig. 20), with actual achievable values fallingbelow fs; vF , as the saturation velocity within graphene mustbe below the Fermi velocity (vsat < vF).
VIII. THEORY VERSUS EXPERIMENT
Plots of the modeled transconductance versus gate-
source voltage and drain characteristics, along side the ex-
perimental results of Moon et al.,40 are shown in Fig. 21.Most of the values used for the model (Table II) were ei-
ther taken directly from Ref. 40 (e.g., L, W, tox) or by extrac-tion. For example, Eq. (24), with minor manipulation, may
be used to determined the electron to hole mobility ratio:
DVDirac 1= 1 ln=lp
DVDS. In conjunction with theexpression for @gm=@VGSjVDirac ln lp
Cox=L, the indi-
vidual electron and hole mobilities may be extracted.
FIG. 19. (Color online) fs; int (solid lines) and fs; ext (dashed lines) versusdrain current (JD) for VDS 0:1; 1; and 2:5 V. For fs; ext, RS RD 0:05 Xmm. The absolute value of the (intrinsic) transconductance ( gmj j) forVDS 2:5 V is plotted, illustrating the dependence of fs; int and fs; ext on gm.
FIG. 20. (Color online) f maxs; int versus gate length (L) forVDS 0:1; 1; and 2:5 V. The shaded region indicates frequencies over fs; vF .
FIG. 21. (Color online) (a) Plot of modeled transconductance (gm) versusgate-source voltage (VGS) (solid lines) and (b) modeled drain characteristic(solid lines) with the experimental results from Moon et al. (symbols withline, Ref. 40).
084515-11 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
As can be seen in the figures, the theoretical develop-
ment given in this work matches the experimental behavior
well. In the case presented here, the largest deviation from
the experiment is for biases resulting in p-type channels(e.g., the dashed lines in Fig. 21(a)). A possible source of
this deviation is the effect of type (i.e., n-type, p-type) transi-tion outside of the gate region. This phenomenon, in addition
to drift current saturation, has not been included in the theo-
retical work presented in this paper, however both are dis-
cussed in the following sections.
A. Graphene-type transition outside the gate region
It has been well established that the native doping or
type of graphene is heavily dependent upon not only the
method for producing the graphene but any possible process-
ing the graphene is exposed to, resulting in n-type graphene,p-type graphene, or any degree in between.6,10,41 Addition-ally, the choice of metal for contacting graphene has been
shown to effectively dope the graphene beneath, or in
other words, set its type.30,42
With regard to a gFET, this leads to the possibility of
the access region and/or contacts outside of the gate region
being of an incompatible type, as compared to the region
under the gate, resulting in an overall increased extrinsic re-
sistance dependent upon the relative type of the region under
the gate to the region outside the gate, which could appear as
a variable bias-dependent extrinsic resistance.
For example, taking the region outside of the gate to be
natively n-type (due to whatever growth or processing con-siderations), then for a gate, drain, and source bias resulting
in a n-channel underneath the gate, the gFET sees extrinsicresistances simply associated with the conduction through
the access region (Fig. 22, column i). As the drain bias takesthe channel into a mixed channel, one can envision the for-
mation of a low conductivity (high resistivity) region just to
the exterior of the gate (see Fig. 22, column ii). This lowconductivity region outside the gate region is identical to
that observed under the gate for bias schemes at or near
channel pinch-off (see Fig. 12). The resultant effect is a
higher observed external resistance for drain biases leading
to mixed channels (i.e., higher drain biases). Similarly, for
gate biases that take the entire channel into a p-channel re-gime, two low conductivity regions are formed to either side
of the gate, also resulting in a higher observed external re-
sistance as compared to biases resulting in an n-channel (Fig.22, column iii). Behavior of this form has been shownthrough simulation by Ancona.12 This example was for a
natively n-type channel, but the same could be said for anatively p-type channel.
This bias-dependent external resistance could explain
some of the asymmetry observed in the transfer characteris-
tics in some practical gFETs.14,22,39,43 And it is believed to
be the reason that the model over predicts the total current as
compared to that observed by Moon et al. at biases in thetransfer characteristic and drain characteristic associated
with mixed or p-channel operation (Fig. 21).14 Of course, asimilar argument is plausible for MOSFET-like gFETs
where the type beneath the contact (dictated by the contact
metal) differs from that in the channel.
This behavior would also suggest that care must be
taken in the choice of contact metal and/or native gra-
phene type with regard to the operation of the gFET. For
example, if one wishes the gFET to operate equally well
in both n-channel and p-channel biases then this wouldsuggest that an access region with an intrinsic-like typing,
and/or contacts that result in a similar intrinsic-typing, are
desirable. Although, such a typing would lead to a higher
overall external resistance compared to a strongly n-typeor p-type access region, the resistance presented to theintrinsic gFET would be essentially equivalent, regardless
of the channel being n-type, p-type, or mixed. On the otherhand, if the gFET is to be used primarily with one type of
channel (n or p), an access region and contact of similartype is desirable in order to reduce the overall external
resistance.
TABLE II. Device parameters and dimensions used for Fig. 21.
Parameter Value
vTi (eV) 4.33vgraphene (eV) 4.5vF (m/sec) 10
6
er; ox 3.9tox (A) 275
L (lm) 2.94W (lm) 4ln (cm
2/Vsec) 6000
lp (cm2/Vsec) 9000
RS (X mm) 0RD (X mm) 0.9Qf =q (cm
2) in Fig. 21(a) 3 1011Qf =q (cm
2) in Fig. 21(b) 1.2 1012
FIG. 22. (Color online) Illustrative plots of the (a) band diagrams (eD: solidline, eF: dashed line), (b) electron and hole concentrations (n: solid symbols,p: hollow symbols), and (c) conductivity (r) in the channel versus position,in the region underneath the gate and just outside the gate.
084515-12 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
B. Drift current saturation
Present thought is that drift current experiences a satura-
tion due to (relatively) high energy phonon scattering
encountered under high electric field strengths.9,11,44 To
date, though, no clear evidence of this current (or velocity)
saturation has been observed in graphene. In the work pre-
sented in this paper, in which drift and diffusion currents areconsidered, the electric field and any electric field dependent
phenomenon are not determinable until the total current in
the channel is solved (see Secs. IV and V). Without prior
knowledge or the simultaneous solution of the electric field,
drift current saturation can not be directly incorporated into
the solution of the total current in the channel.
Some papers have included drift current saturation, but,
in doing so, ignore diffusion current (i.e., assume drift cur-
rent comprises the total of the current in the channel).9,11 As
has been shown in this paper, diffusion current is a nonnegli-
ble portion of the total current in the channel. Given a spe-
cific current density within the channel, the assumption that
the total current is solely comprised of drift current leads to
an overestimation of the electric field within the channel.
This in turn leads to an early onset of drift current saturation.
In order to maintain a good fit, parameters used to describe
the drift current saturation must be altered. The process
becomes a mathematical fitting procedure, counter to the
original intent for including drift current saturation: the first
principles description of gFET behavior.
Clearly, future work must look to correctly incorporate
drift current saturation in the presence of both drift and diffu-
sion currents.
IX. CONDUCTIVITYAND MOBILITY
The conductivity of graphene, under the relaxation time
approximation, has been given by multiple sources as having
the general form of4,5,7,45
rn JE q
2s
ph2eF eD : (38)
Under specific conditions (e.g., T 0, n p), the conduc-tivity, due to impurity scattering, may be approximated as4,7
rn 2q2
h
p hvF 2u2o
n; (39)
with h being Planks constant; uo q4nimp=16e2r e
2o
p, the
strength of the scattering potential resulting from charged
impurities; and nimp, the impurity sheet density. Note,q nimp should not be confused with Qf, the fixed chargesheet density. Qf clearly includes charges that affect mobility(e.g., charges within or within proximity of the graphene
sheet) but also other fixed charges within the device structure
that do not have any real influence on mobility (e.g., charges
deep in the gate insulator).
By equating Eq. (39) to
rn 2q2
h
p hvF 2u2o
n qlnn; (40)
the mobility can be equated to
ln 2q
h
p hvF 2u2o
16e2r e
2ohv
2F
q31
nimp/ 1
nimp: (41)
The mobility in Eq. (41) is inversely proportional to the
impurity density, nimp, which has been experimentallyobserved;6 and is the premise for the assumption of constant
mobility, for a given impurity density, in the work presented
in this paper.
Many papers incorrectly apply the conductivity of
Eq. (39) in analyzing the behavior of graphene.4,8,11,20,35 The
conductivity of Eq. (39) was derived for a single carrier type
(e.g., electrons) in a single band of graphene (e.g., the con-
duction band, e > eD).4 However, in subsequent application,
the conductivity has been set proportional to the net carrierconcentration,4,8,11,20,35 which, using the terminology of this
paper, is
r lQnet ql n p : (42)This leads to two distinct conclusions for the conductivity:
(1) r 0 for Qnet 0 and (2) r < 0 (i.e., negative) for pre-dominately p-type conditions, which is nonphysical and hasnot been experimentally observed in graphene.1,2,7,10
In order to correct this, some authors have artificially
replaced the net carrier concentration by its absolute value
and introduced a residual carrier concentration (n), whichprovides for a nonzero conductivity when Qnet ! 0. Thisresults in a piecewise equation for the conductivity of
graphene:
r qln n pj j n < n
ql n pj j n pj j n > n
(; (43)
where n is an impurity concentration-dependent constantrelated to the Dirac voltage.4 Alternatively, some suggest
adding a minimum charge density to the net carrier concen-
tration (e.g., n pj j no).11 Expressions like Eq. (43) havebeen further extended by relating the net carrier concentra-
tion to a voltage within the structure (e.g., gate voltage,
channel voltage) via a capacitance (e.g., gate capacitance,
quantum capacitance), commonly under the restriction
jVch Vgrj jeF eDj=q kBT=q, with subsequent incor-rect application to biases outside this restriction:10,11,20,35
r Kq2
h
1
nimp
Coxq
VG n |{z}
n
n > n
Kq2
h
1
nimpn n < n
8>>>>>>>>>:
; (44a)
or
r 12lCQ Vch Vgr
r0; (44b)where K is a unitless constant associated with the strength ofthe scattering potential and r0 is a constant, in line with n,used to maintain a nonzero conductivity.11,35 Finally, in
084515-13 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
cases where behavior corresponding to different mobilities
for electrons and holes is observed, authors replace l by lnfor n p > 0 and lp for n p < 0.10
Although these equations have been shown to approxi-
mately fit the behavior of graphene fairly well, especially at
large gate voltages,11,20 they offer poor physical insight. In
fact, due to the proportionality to the net carrier concentra-
tion of Eq. (43), the conductivity leads to the nonphysical
interpretation that the electron and hole currents oppose
each other under a single driving force (e.g., electric field,
Fermi energy gradient), or, alternatively stated, the electron
and holes fluxes flow in the same direction under a singledriving force. Such phenomena has been attributed to ambi-
polar transport, which, stated simply, only describes the
motion of excess charge carriers under the condition oflocal (quasi-)neutrality (i.e., Qnet Qf 0 for all x and z).This is not the case here, as it has been shown theoretically
in this paper and experimentally elsewhere (Ref. 38) that
gFETs are not locally neutral under operation (i.e.,
Qnet Qf Qm 6 0).Instead, beginning with the original derivation of con-
ductivity for a single carrier type, expressed as a constant
mobility, the total current is a sum of the currents resulting
from each carrier type, as in Eq. (15). Rewriting Eq. (15),
using Eq. (14),
JT qlnn qlpp reF=q: (45)
Recognizing that reF=q can be interpreted as an electricfield associated with an applied bias, then the conductivity is
r qlnn qlpp; (46)
the classic definition of conductivity for a system of elec-
trons and holes.
In essence, this definition of conductivity is the basis for
the work presented in this paper. With the addition of ther-
mal statistics, electrostatics, and electrodynamics, the unique
behaviors of graphene have been described and character-
ized, without the need for artificial corrections.
For example, this can be seen in the drain conductance
derived earlier. If the drain conductance is simply defined as
the conductivity of the channel (referenced to the drain) mul-
tiplied by the appropriate geometric ratio, we have
Gd WLrjxL
W
LqlnnD qlppD
: (47)
The drain conductance per unit width is then
gd GdW 1
LqlnnD qlppD
; (48)
which is identical to the drain conductance of Eq. (23),
derived in Appendix B.
X. CONCLUSIONS
This paper has presented an in-depth theoretical exami-
nation of the graphene-based field effect transistor. Begin-
ning with the prevailing theory regarding the fundamental
charge carrier transport in graphene and using a first princi-
ples approach, the unique behaviors observed in graphene-
based FETs were described. Expressions describing the
current in the channel (JD), transconductance (gm), drain con-ductance (gd), and Dirac voltage (VDirac) were developed. Thebehavior under bias of the electron (n) and hole (p) carrierconcentrations, electron and hole currents (JD; n, J
driftD; n , J
diffD; n
and JD; p, JdriftD; p , J
diffD; p), and the electric field in the channel
(Ech), along the length of the channel, were described andexamined. The characteristic V-shaped transfer characteristic
observed in graphene-based FETs was shown to be due to gra-
phene channel being capable of supporting both an n- and p-channel, with biases around the Dirac voltage resulting in a
mixed carrier channel. In essence, the graphene channel is
never fully depleted of mobile charge carriers. This character-
istic of graphene also directly leads to the lack of an off-state
(under gate modulation) observed in these devices.
The frequency performance of these devices, in terms of
the short-circuit current-gain cutoff frequency (fs), was alsoexamined. The intrinsic and extrinsic fs was found to beheavily influenced by the transconductance (gm) of thegFET. The maximum intrinsic fs (f
maxs; int) was found to have an
absolute upper bound limited by the Fermi velocity (vF) ofgraphene (f maxs; int < fs; vF vF=2pL).
The influence of external resistance on the steady-
state and frequency performance of the graphene-based
FET was also examined. As in nongraphene-based FETs,
external resistance was found to reduce the current in the
channel and transconductance for a given set of external
biases. Consequently, fs under the influence of externalresistance (fs; ext) is reduced as compared to the intrinsiccase (fs; int).
The theoretical work developed in this paper was found
to agree well with observed experimental results. By means
of this work, various device and material parameters, such as
electron and hole mobilities and fixed charge density, can be
extracted from experimental results, aiding in device and
material analysis.
Finally, behaviors not directly predicted by the theory
presented here, such as carrier type transition outside of the
gate region, were discussed with concern to possible implica-
tions on device performance.
APPENDIX A: QV RELATIONSHIP FOR GRAPHENE ONA DIELECTRIC SUBSTRATE WITH BACK GATE
A band diagram and charge profile for graphene on a
dielectric substrate with back gate and top gate is shown in
Fig. 23.
All terms are as defined earlier for the case of graphene
on SiC (see Sec. III) with the addition of the subscript T forterms referring to the top gate and B for terms referring tothe bottom gate. The bottom gate, commonly highly doped
Si, is assumed to be an equipotential. Proceeding with the
same method as presented in Sec. III, two voltage (potential
energy) loops around the structure are completed. One from
the top gate to the graphene channel:
084515-14 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
q VG; T Vch /mo; T qEox; Ttox; T /so; T eD eF 0; (A1a)
and one from the graphene channel to the back gate:
eD eF /so;B qEox;Btox;B /mo;B q VG;B Vch
0: (A1b)A third voltage loop exists from the top gate to the back
gate, but is unnecessary for the analysis here.
In order to maintain charge neutrality, Qm; T QnetQf Qm;B 0. By Gausss law, Eox;B Qm;B=eox;B andEox; T Qnet Qf Qm;B
=eox; T . Solving Eq. (A1b) for
Qm;B, using the preceding relations, and plugging the resultinto Eq. (A1a) gives the QV relationship for graphene on adielectric substrate with back gate:
VG; T Cox;BCox; T
VG;B 1 Cox;BCox; T
Vch 1
Cox; TQnet
1 Cox;BCox; T
kBT
qgF
1
Cox; TQf 1
q/mo; T /so; T
1q
/mo;B /so;B Cox;B
Cox; T 0: (A2)
The various device characteristics for a gFET on a dielectric
substrate with back gate can now be solved, following the
same methods used in this paper.
APPENDIX B: FULL DERIVATION OF THE DRAINCURRENT, TRANSCONDUCTANCE, AND DRAINCONDUCTANCE
Beginning with Eq. (17), both sides are multiplied by
dx:
JTdx Jndx Jpdx qlnn dVch qlpp dVch : (B1)
As before, it is assumed there are no sources or sinks of
charge within the channel (r JT @q=@t 0), thereforethe total current density (JT) is constant. Integrating Eq. (B1)gives
JTx x0
Jndxx0
Jpdx
Vch x VS
qlnn dVch Vch x VS
qlpp dVch : (B2)
The electron (n) and hole (p) concentrations cannot bealgebraically related to the voltage in the channel (Vch).However, they are related by the normalized, relative
Fermi energy (gF) through the QV relationship of thegFET, established in Eq. (13). By differentiating the QVrelationship with respect to gF, with some algebraicmanipulation,
dVchdgF
1Cox
dQnetdgF
kBTq
; (B3)
recognizing that
dQnetdgF
ddgF
q p n
ddgF
q NGF 1 gF NGF 1 gF
qNG F 0 gF F 0 gF kBT
q
q
kBTqNG F 0 gF F 0 gF
dQnetdgF
kBTq
CQ; (B4)
and multiplying both sides of Eq. (B3) by dgF,
dVch kBTq
1 CQCox
dgF; (B5)
dVch in Eq. (B2) can be replaced:
JTx x0
Jndxx0
Jpdx
Vch x VS
qlnn dVch Vch x VS
qlpp dVch
JTx I ngF x gS
I pgF x gS
; (B6)
FIG. 23. Band diagram (top) and charge profile (bottom) of a gFET struc-
ture with top and back gates.
084515-15 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
with gS gFx 0 and
I ngg I n gF
gg
gg
lnkBTn 1CQCox
dgF; (B7a)
I pgg I p gF
gg
gg
lpkBTp 1CQCox
dgF: (B7b)
Evaluating the integrals of Eqs. (B7a) and (B7b) gives
In gF and I p gF :
In gF lnkBTNGF 2 gF
CqCox
F 1 gF 2
gFF 2 gF F 3 gF ; (B8a)
I p gF lpkBTNGF 2 gF
CqCox
F 1 gF 2
gFF 2 gF F 3 gF ; (B8b)
with Cq q2NG=kBTFinally, the total current density (at x) is found by divid-
ing Eq. (B6) by x,
JT 1x
x0
Jndx 1x
x0
Jpdx
1xI njgF x gS
1
xI pjgF x gS : (B9)
Solving Eq. (B9) at x L (gF L gD) leads to the resultspresented in Eq. (18),
JT 1L
L0
Jndx 1L
L0
Jpdx Jn Jp; (B10)
and Eq. (19),
Jn 1LI njgDgS ; Jp
1
LI pjgDgS : (B11)
Defining the drain current (density), JD, as the current thatflows directed from drain to source, assuming VD > VS, thedrain current is then
JD JT Jn Jp JD; n JD; p; (B12)
with simply JD; n Jn 1=L I njgSgD and JD; p Jp 1=L I pjgSgDThe transconductance is defined as
gm @JD@VGS
VDS
: (B13)
Using Eqs. (B11) and (B12), Eq. (B13) may be written as
gm @JD@VGS
@@VGS
1
LIngSgD 1LI pgSgD
gm 1L
@I n@VGS
gSgD
@I p@VGS
gSgD
!: (B14)
The derivatives in Eq. (B14) may be written as
@I n@VGS
@I n@gF
@gF@VGS
;@Ip@VGS
@I p@gF
@gF@VGS
: (B15)
Recall Eq. (B7),
I njgg gg
lnkBTn 1CQCox
dgF
In g In g
dI n; (B16a)
I pjgg gg
lpkBTp 1CQCox
dgF
Ip g Ip g
dI p: (B16b)
Therefore,
@I n@gF
lnkBTn 1CQCox
; (B17a)
@I p@gF
lpkBTp 1CQCox
: (B17b)
By differentiating the QV relationship from Eq. (13) withrespect to gF, following a similar procedure as in Eqs. (B3)and (B4) gives
dVGSdgF
kBTq
1 CQCox
: (B18)
Eq. (B15) then becomes
@I n@VGS
@I n@gF
@gF@VGS
@I n@gF
@VGS@gF
1
lnkBTn 1CQCox
kBT
q1 CQ
Cox
1@I n@VGS
qlnn; (B19a)
@Ip@VGS
@Ip@gF
@gF@VGS
@I p@gF
@VGS@gF
1
lpkBTp 1CQCox
kBT
q1 CQ
Cox
1@Ip@VGS
qlpp: (B19b)
084515-16 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
Finally, evaluating Eq. (B14) using Eq. (B19) results in the
transconductance being
gm 1LqlnnS qlnnD qlppS qlppD
1Lqln nS nD qlp pS pD ; (B20)
with nD n gD , nS n gS , pD p gD , and pS p gS The Dirac voltage is defined as VDirac VGSjgm0. Set-
ting the transconductance (gm, Eq. (B20)) equal to zeroresults in
ln nS nD lp pD pS : (B21)
From Fig. 9, it is clear that VDirac corresponds to a bias con-dition resulting in a mixed carrier channel, specifically with
an n-channel toward the source and a p-channel toward thedrain (Fig. 11(c)). Therefore, assuming gD gF x L 0and gS gF x 0 0, then nD pS 0 and Eq. (B21)becomes
lnnS lppD; (B22)
with
nS NG g2S
2
Coxq
VG VS 1
CoxQf 1
q/mo /so
; (B23a)
pD NG g2D
2
Coxq
VD VS VG VS 1Cox
Qf 1q
/mo /so
:
(B23b)
Recognizing that VG VS VGS VDirac (and VD VS VDS) in Eq. (B23) and applying Eq. (B23) to Eq. (B22),the Dirac voltage, with some algebraic manipulation, is
VDiracVGSjgm0 1
1 ln=lpVDS 1
CoxQf 1
q/mo /so :
(B24)
The maximum transconductance occurs for VGS VDirac,under which gD; gS 0 and pD; pS 0:
gm;max 1L
qln nS nD
qlnL
Coxq
VG VS 1
CoxQf 1
q/mo /so
Coxq
VG VD 1
CoxQf 1
q/mo /so
gm;max lnCox
LVD VS lnCox
LVDS: (B25)
The minimum transconductance occurs for VGS VDirac,under which gD; gS 0 and nD; nS 0:
gm;min 1Lqlp pS pD
qlpL
Coxq
VS VG 1
CoxQf 1
q/mo /so
Coxq
VD VG 1
CoxQf 1
q/mo /so
gm;min lpCoxL
VS VD lpCoxL
VDS: (B26)
The derivative of the transconductance with respect to the
gate-source voltage (@gm=@VGS) is given by
@gm@VGS
1L
qln@ nS nD
@VGS qlp
@ pS pD @VGS
: (B27)
Solving at the Dirac voltage (VGS VDirac), again gD 0,gS 0, and nD pS 0. With nS and pD given by Eq.(B23), the derivatives in Eq. (B27) can easily be solved
and
@gm@VGS
VGSVDirac
CoxL
ln lp
: (B28)
The drain conductance is defined as
gd @JD@VDS
VGS
@@VDS
1
LI ngSgD 1LI pgSgD
: (B29)
The derivative of any independent function by definition is
zero. Because gS is independent of the drain voltage,@I n gS =@VDS and @I p gS =@VDS are zero and
gd 1L @In gD
@VDS @I p gD
@VDS
1L @In gD
@gD
@gD@VDS
@Ip gD @gD
@gD@VDS
gd 1L @I n gD
@gD
@VDS@gD
1 @I p gD
@gD
@VDS@gD
1" #:
(B30)
Again, following a similar procedure as in Eqs. (B3) and
(B4) and evaluating Eq. (B17) at gF gD (Vch VD), thederivatives in Eq. (B30) can be replaced, and the drain con-
ductance becomes
gd 1L
qlnnD qlppD
: (B31)
APPENDIX C: DERIVATION OF POSITIONDEPENDENCYOF DEVICE PROPERTIES
The various device properties of the gFET (e.g., carrier
concentration, current density, electric field in the channel)
cannot be directly related to the position within the channel,
but can be related indirectly though the normalized, relative
Fermi energy, gF. The normalized, relative Fermi energy, gF,which varies between its value at the drain (gD) and source
084515-17 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
(gS), is determined via the QV relationship for the graphenechannel (Eq. (13)) from the applied biases: VG, gate voltage;VD, drain voltage; and VS, source voltage.
The position within the channel for the various device
properties can be found by rewriting Eq. (B6):
x gF x0Jndx
x0Jpdx
JT
InjgFgS I pjgFgSJT
x gF LI n I pgFgSIn I pgDgS
: (C1)
The Fermi energy in the channel is simply the channel volt-
age multiplied by a negative electron charge: eF qVch.The Dirac energy, in essence the band edge for both the
conduction band and valence band, is related to the
Fermi energy by the normalized, relative Fermi energy, gF:eD eF kBTgF.
The total current (density), JT, throughout the entirechannel is constant with position, ignoring gate leakage, and
is given by Eqs. (18)(20) or Eqs. (B8), (B10), and (B11),
knowing the drain and source normalized, relative Fermi
energies (gD and gS, respectively). The total current is equalto the sum of the total electron (Jn) and hole (Jp) current den-sities, Eq. (16). Taking the ratio of the total electron current
density to the total current density,
JnJT lnn deF=dx
lnn deF=dx lpp deF=dx lnn
lnn lpp; (C2)
and rewriting
Jn gF JTlnn
lnn lpp: (C3)
gives the total electron current density in the channel as a
function of gF, via n and p.By expanding the original definition of the total electron
current from Eq. (16), the electron drift and diffusion current
densities are found:
Jn lnndeFdx
lnnd
dxeD kBTgF
lnnd
dxqVgr lnkBTn dgFdx
qlnnEch lnkBTndgFdx
Jn Jdriftn Jdiffn ; (C4)
with qVgr eD and Ech Ex dVgr=dx.The derivative of the normalized, relative Fermi energy
(gF) with respect to x can be found by inverting the deriva-tive of x (Eq. (C1)) with respect to gF:
dgFdx
dxdgF
1
1JT
d
dgFI n IpgFgS
1: (C5)
As the normalized, relative Fermi energy at the source (gS) isfixed for a given bias condition, its derivative with respect to
gF is zero and Eq. (C5) reduces to
dgFdx
JT dIn gF dgF
dI p gF dgF
1: (C6)
The derivatives in Eq. (C6) have already been evaluated in
Eq. (B17). Therefore,
dgFdx
dxdgF
1 JT
lnn lpp
kBT 1 CQ=Cox : (C7)
The electron diffusion current density is then
Jdiffn gF lnkBTndgFdx
JT lnnlnn lpp1
1 CQ=CoxJdiffn gF Jn
1
1 CQ=Cox : (C8)
Rewriting Eq. (C4) gives the electron drift current density:
Jdriftn gF Jn Jdiffn Jn 1 1
1 CQ=Cox
Jdriftn gF Jn1
1 Cox=CQ : (C9)
Following a similar procedure, the hole current density, the
hole diffusion current density, and the hole drift current den-
sity are, respectively,
Jp gF Jdriftp Jdiffp JTlpp
lnn lpp(C10a)
Jdiffp gF Jp1
1 CQ=Cox (C10b)
Jdriftp gF Jp1
1 Cox=CQ (C10c)
The electric field in the channel is found by rewriting the
expression for the drift current from Eq. (C4):
Ech gF Jdriftnqlnn
Jdriftp
qlpp: (C11)
Finally, as the drain current (density) is defined as flowing
opposite to the total current density, JD JT , the variousproperties referenced to the drain current become
084515-18 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
-
JD; n gF Jn gF ; (C12a)
JdiffD; n gF Jdiffn gF ; (C12b)
JdriftD; n gF Jdriftn gF ; (C12c)
Ech gF JdriftD; nqlnn
; (C12d)
with p replacing n for the hole components.
1S. Reich, J. Maultzsch, C. Thomsen, and P. Ordejon, Phys. Rev. B 66,035412 (2002).2K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson,
I. M. Grigorieva, S. V. Dubonos, and A. A. Firsov, Nature 438, 197(2005).3T. Fang, A. Konar, X. Huili, and D. Jena, Appl. Phys. Lett. 91, 092109(2007).4S. Adam, E. H. Hwang, V. M. Galitski, and S. Das Sarma, Proc. Natl.
Acad.Sci. U.S.A. 104, 18392 (2007).5S. Adam, E. H. Hwang, E. Rossi, and S. Das Sarma, Solid State Commun.
149, 1072 (2009).6J. L. Tedesco, B. L. VanMil, R. L. Myers-Ward, J. M. McCrate, S. A. Kitt,
P. M. Campbell, G. G. Jernigan, J. C. Culbertson, C. R. Eddy, and D. K.
Gaskill, Appl. Phys. Lett. 95, 122102 (2009).7A. H. C. Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K.
Geim, Rev. Mod. Phys. 81, 109 (2009).8K. L. Shepard, I. Meric, and P. Kim, in Proceedings of the 2008 IEEE/ACM International Conference on Computer-Aided Design (ICCAD)San Jose, CA, 1013 November 2008, pp. 406411 (IEEE, New York,
2008)9I. Meric, M. Y. Han, A. F. Young, B. Ozyilmaz, P. Kim, and K. L. She-
pard, Nature Nanotechnol. 3, 654 (2008).10J. H. Chen, C. Jang, M. Ishigami, S. Xiao, W. G. Cullen, E. D. Williams,
and M. S. Fuhrer, Solid State Commun. 149, 1080 (2009).11S. A. Thiele, J. A. Schaefer, and F. Schwierz, J. Appl. Phys. 107, 094505(2010).
12M. G. Ancona, IEEE Trans. Electron Devices 57, 681 (2010).13P. R. Wallace, Phys. Rev. 71, 622 (1947).14J. S. Moon, D. Curtis, M. Hu, D. Wong, C. McGuire, P. M. Campbell, G.
Jernigan, J. L. Tedesco, B. VanMil, R. Myers-Ward, C. R. Eddy, and D.
K. Gaskill, IEEE Electron Device Lett. 30, 650 (2009).15Y. M. Lin, C. Dimitrakopoulos, K. A. Jenkins, D. B. Farmer, H. Y. Chiu,
A. Grill, and P. Avouris, Science 327, 662 (2010).16C. Hummel, F. Schwierz, A. Hanisch, and J. Pezoldt, Phys. Status Solidi B
247, 903 (2010).17J. Kedzierski, H. Pei-Lan, P. Healey, P. Wyatt, and C. Keast, in Proceed-ings of the 66th Annual Device Research Conference (DRC) 2325 June2008, Santa Barbara, CA, pp. 2526 (IEEE, New York, 2008)
18L. Xuesong, C. Weiwei, A. Jinho, K. Seyoung, N. Junghyo, Y. Dongxing,
R. Piner, A. Velamakanni, J. Inhwa, E. Tutuc, S. K. Banerjee, L. Colombo,
and R. S. Ruoff, Science 324, 1312 (2009).19E. Sutter, P. Albrecht, and P. Sutter, Appl. Phys. Lett. 95, 133109 (2009).20K. Keun Soo, Z. Yue, J. Houk, L. Sang Yoon, K. Jong Min, K. S. Kim, A.
Jong-Hyun, P. Kim, C. Jae-Young, and H. Byung Hee, Nature 457, 706(2009).
21C. Helin, Y. Qingkai, R. Colby, D. Pandey, C. S. Park, L. Jie, D. Zemlya-
nov, I. Childres, V. Drachev, E. A. Stach, M. Hussain, L. Hao, S. S. Pei,
and Y. P. Chen, J. Appl. Phys. 107, 044310 (2010).22Y. M. Lin, H. Y. Chiu, K. A. Jenkins, D. B. Farmer, P. Avouris, and
A. Valdes-Garcia, IEEE Electron Device Lett. 31, 68 (2010).23W. Shockley, Proc. IRE 40, 1365 (1952).24R. R. Bockemuehl, IEEE Trans. Electron Devices 10, 31 (1963).25S. Kim, J. Nah, I. Jo, D. Shahrjerdi, L. Colombo, Y. Zhen, E. Tutuc, and
S. K. Banerjee, Appl. Phys. Lett. 94, 062107 (2009).26S. Skordas, F. Papadatos, S. Consiglio, E. T. Eisenbraun, A. E. Kaloyeros,
and E. P. Gusev, J. Mater. Res. 20, 1536 (2005).27C. Cibert, H. Hidalgo, C. Champeaux, P. Tristant, C. Tixier, J. Desmaison,
and A. Catherinot, Thin Solid Films 516, 1290 (2008).28D. E. Eastman, Phys. Rev. B 2, 1 (1970).29C. Oshima and A. Nagashima, J. Phys.: Condens. Matter 9, 1 (1997).30G. Giovannetti, P. A. Khomyakov, G. Brocks, V. M. Karpan, J. van den
Brink, and P. J. Kelly, Phys. Rev. Lett. 101, 026803 (2008).31F. Rana, Phys. Rev. B 76, 155431 (2007).32J. M. Dawlaty, S. Shivaraman, M. Chandrashekhar, F. Rana, and M. G.
Spencer, Appl. Phys. Lett. 92, 042116 (2008).33F. Rana, P. A. George, J. H. Strait, J. Dawlaty, S. Shivaraman, M. Chan-
drashekhar, and M. G. Spencer, Phys. Rev. B 79, 115447 (2009).34C. Kittel and H. Kroemer, Thermal Physics, 2nd ed., Chap. 13, pp. 379381 (W. H. Freeman and Company, New York, 1980)
35Y. W. Tan, Y. Zhang, K. Bolotin, Y. Zhao, S. Adam, E. H. Hwang, S. Das
Sarma, H. L. Stormer, and P. Kim, Phys. Rev. Lett. 99, 246803 (2007).36H. E. Romero, N. Shen, P. Joshi, H. R. Gutierrez, S. A. Tadigadapa, J. O.
Sofo, and P. C. Eklund, ACS Nano 2, 2037 (2008).37K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V.
Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004).38M. Freitag, H.-Y. Chiu, M. Steiner, V. Perebeinos, and P. Avouris, Nature
Nanotechnol. 5, 497 (2010).39J. Kedzierski, H. Pei-Lan, A. Reina, K. Jing, P. Healey, P. Wyatt, and C.
Keast, Electron Device Lett. 30, 745 (2009).40J. S. Moon, D. Curtis, S. Bui, M. Hu, D. K. Gaskill, J. L. Tedesco,
P. Asbeck, G. G. Jernigan, B. L. VanMil, R. L. Myers-Ward, C. R. Eddy,
P. M. Campbell, and X. Weng, Electron Device Lett. 31, 260 (2010).41K. Brenner and R. Murali, Appl. Phys. Lett. 96, 063104 (2010).42R. S. Sundaram, C. Gomez-Navarro, E. J. H. Lee, M. Burghard, and
K. Kern, Appl. Phys. Lett. 95, 223507 (2009).43B. Huard, N. Stander, J. A. Sulpizio, and D. Goldhaber-Gordon, Phys.
Rev. B 78, 121402 (2008).44V. Perebeinos and P. Avouris, Phys. Rev. B 81, 195442 (2010).45E. H. Hwang, S. Adam, and S. Das Sarma, Phys. Rev. Lett. 98, 186806/1(2007).
084515-19 James G. Champlain J. Appl. Phys. 109, 084515 (2011)
Downloaded 17 Sep 2013 to 130.39.223.34. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions