J Comput Electron (2013) 12:188–202

DOI 10.1007/s10825-013-0456-9

Gate-induced carrier density modulation in bulk graphene:

theories and electrostatic simulation using M ATLAB pdetool
Ming-Hao Liu

Published online: 25 April 2013

© Springer Science+Business Media New York 2013

Abstract This article aims at providing a self-contained in- 1 Introduction

troduction to theoretical modeling of gate-induced carrier
density in graphene sheets. For this, relevant theories are in- Electronic transport in graphene [1, 2], a one-atom-thick
troduced, namely, classical capacitance model (CCM), self- honeycomb carbon lattice, is one of its main issues among
consistent Poisson-Dirac method (PDM), and quantum ca- the increasing number of fundamental studies ever since
pacitance model (QCM). The usage of M ATLAB pdetool the first successful isolation of stable monolayer graphene
is also briefly introduced, pointing out the least knowledge flakes in 2004 [3]. What led to the explosive growth of
required for using this tool to solve the present electro- the graphene literature, however, was not only the discov-
static problem. Results based on the three approaches are ery of the mechanical exfoliation (Scotch-tape method) for
compared, showing that the quantum correction, which is graphene flake preparation, which made graphene easily ac-
not considered by the CCM but by the other two, plays a cessible to laboratories all over the world, but also the char-
role only when the metal gate is exceedingly close to the acterization of the electronic properties of graphene by elec-
graphene sheet, and that the exactly solvable QCM works trical gating, which provides a direct way to modulate the
equally well as the self-consistent PDM. Practical exam- carrier density, and hence the Fermi level, of graphene [3].
ples corresponding to realistic experimental conditions for Conductance (resistance) sweep using a single backgate is
generating graphene pnp junctions and superlattices, as well henceforth a standard electronic characterization tool for
as how a background potential linear in position can be graphene. Double-gated graphene opens even more possibil-
achieved in graphene, are shown to illustrate the applica- ities of graphene electronics and allows experimental studies
bility of the introduced methods. Furthermore, by treating of graphene pn and pnp junctions [4–8], as well as the inter-
metal contacts in the same way, the last example shows esting physics of Klein tunneling [9–13]. Gate-induced car-
that the PDM and the QCM are able to resolve the contact- rier density modulation, therefore, plays an essential role for
induced doping and screening potential, well agreeing with fundamental as well as advanced studies of graphene elec-
the previous first-principles studies. tronics.
Theory of the gate-induced carrier density modulation is
Keywords Graphene · Carrier density · Quantum mainly an electrostatic problem. How one should obtain the
capacitance · Self-consistent Poisson-Dirac method · gate-voltage dependence of the carrier density in graphene
Superlattice · pnp junction · Bloch-Zener oscillation · depends actually on how precise one wishes. For cheapest
Contact doping computation, the graphene sheet carrier density can be di-
rectly regarded as the induced surface charge density adja-
cent to graphene [3], which is treated as a conductor fixed at
zero potential. This corresponds to the classical capacitance
model (CCM) that is widely adopted in most experimen-
M.-H. Liu ()
tal works on graphene transport [2] and can be solved ex-
Institut für Theoretische Physik, Universität Regensburg,
93040 Regensburg, Germany actly. A more precise computation takes into account the re-
e-mail: [email protected] lation between the induced charge density on graphene and
J Comput Electron (2013) 12:188–202 189

the electric potential energy that those charge carriers gain, gate-induced carrier density modulation in graphene are re-
through the graphene density of states [14–16]. This requires viewed in Sect. 3, where the analytics based on the capac-
self-consistent iterative computation [8, 17, 18], which is a itance models and the numerics based on the PDM is also
bit more expensive, but actually corresponds to the quantum compared. Practical applications based on the theories re-
capacitance model (QCM) [19], within which exact solu- viewed in Sect. 3 are given in Sect. 4, and a summary of the
tions for single-gated pristine graphene [16] and even multi- present work is concluded in Sect. 5.
gated doped graphene [20] can be obtained. Further con-
siderations such as the Coulomb interaction of the induced
2 Usage of Matlab’s pdetool for electrostatics
charges on the graphene sheet are possible [15, 17], but these
would be out of the scope of the present discussion. The pdetool is a useful numerical tool built in M ATLAB and
Whereas a thorough and comprehensive review on the provides a convenient way to solve several classic partial
theory of gate-induced carrier density modulation of bulk differential equation (PDE) problems in two-dimension. For
graphene so far does not exist in the literature, part of this the electrostatics at our present interest, the Poisson equa-
article aims at providing this missing piece. The review in- tion,
cludes both the analytical and numerical aspects, as well as ρ
a brief introduction to the usage of M ATLAB’s pdetool, in −∇ · (εr ∇u) = , (1)
ε0
order for a self-contained context. Readers who happen to obtained from two of the Maxwell’s equations, ∇ × E = 0
be M ATLAB users would find this brief usage helpful, but and ∇ · D = ρ, which respectively lead to E = −∇u and
non-M ATLAB users may as well neglect it without encoun- ∇ · εr E = ρ/ε0 , is the central equation that the pdetool
tering further gaps. The analytics based on the CCM and solves for the electric potential u.1 In Eq. (1), the prod-
QCM and the numerics based on the self-consistent itera- uct of the dielectric constant (relative permittivity) εr and
tion method, namely, the Poisson-Dirac method (PDM), us- the free space permittivity ε0 gives the absolute permittivity
ing M ATLAB’s pdetool will be compared, showing that the ε = εr ε0 .
quantum correction plays usually a minor role, unless the A full introduction to the usage of the pdetool can be
metal gate is exceedingly close to the graphene sheet. In the found in the M ATLAB documentation [26] and need not be
case of single-gated pristine graphene, consistency between repeated here. To digest the full user’s guide of the tool,
the QCM and PDM is satisfactory even for capacitors with however, is not necessary for our present focus, which is es-
finite gates, and is exact for parallel-plate capacitors with sentially an electrostatic problem. This section is basically
infinitely extending gates. to elaborate those that are less clear in [26] but nevertheless
With a full understanding of the gate-voltage modula- important for our purpose of obtaining the gate-voltage de-
tion on the graphene carrier density, examples of its appli- pendence of the graphene carrier density, and to point out
cations aiming at providing realistic local energy band off- the least required knowledge for this purpose.
sets due to electric gating will be illustrated. This is par-
2.1 Overview of pdetool
ticularly important for an accurate theoretical modeling of
transport in graphene [21]. Examples include (i) graphene To solve a PDE problem using the pdetool, required neces-
pnp junctions, (ii) graphene superlattices, and (iii) genera- sary inputs can be exported from the graphical user interface
tion of background potential linear in position in graphene. (GUI) of the pdetool (initiated by executing “pdetool” from
Practically, the example (i) provides the study of the physics the command window) and are briefly described in the fol-
of Klein backscattering [12, 21, 22], while the combination lowing.
of the examples (ii) and (iii) is the underlying prerequisite of
(i) System geometry. The geometrical shapes of the build-
the Bloch-Zener oscillation in graphene [23]. Furthermore,
ing blocks, such as the oxide layers, metallic gates,
the introduced PDM and QCM are capable of treating the etc., which constitute the system where the PDE prob-
effects of metal contacts, for which example (iv) of contact- lem is defined, can be drawn in the “Draw Mode” of
induced doping and screening potential is also illustrated. the GUI. The resulted “decomposed geometry” allows
Taking palladium as a specific example of the metal con- us to proceed to the rest of the inputs, but there is no
tact, the results obtained by the PDM and QCM are shown need to “Export Decomposed Geometry, Set Formula,
to agree well with the previous first-principles studies [24, Labels. . .” from the “Draw menu” since not all of them
25]. will be needed by the PDE solvers.
This paper is organized as follows. In Sect. 2, we first pro-
vide a brief introduction to the usage of M ATLAB’s pdetool, 1 To be consistent with the pdetool, we name the electric potential as
pointing out the least knowledge required to apply the tool u, while reserve the variable V for the energy band offset (the “on-site
on the present specific electrostatic problem. Theories of the energy” in the language of tight-binding formulation).
190 J Comput Electron (2013) 12:188–202

(ii) PDE coefficients. In the “PDE Mode” of the GUI, one • Rows 3 to 3 + N 2 − 1 contain the lengths for the strings
can designate different regions of materials by filling representing q. The lengths are stored in columnwise or-
in the respective dielectric constants and space charge der with respect to q. [See Eq. (2) below.]
densities. These are stored in certain PDE coefficients • Rows 3 + N 2 to 3 + N 2 + N − 1 contain the lengths for
matrices, which can be output from the GUI and will the strings representing g. [See Eq. (2) below.]
be required by the PDE solvers in programming. • Rows 3 + N 2 + N to 3 + N 2 + N + MN − 1 contain
(iii) Boundary conditions. In the “Boundary Mode” of the the lengths for the strings representing h. The lengths are
GUI, boundary conditions for each bounding edge can stored in columnwise order with respect to h. [See Eq. (3)
be assigned. The resulting boundary matrix b, which below.]
will be required by the PDE solvers in programming, • Rows 3 + N 2 + N + N M to 3 + N 2 + N + MN + M −
and the Decomposed Geometry, which will be required 1 contain the lengths for the strings representing r. [See
when visualizing the PDE geometry, can be exported Eq. (3) below.]
from the “Boundary menu”. An elaborated instruction • The following rows contain text expressions representing
about b will be given later. the actual boundary condition functions.
(iv) Mesh points. The mesh points are those spatial points Here, two types of boundary conditions2 are included,
at which the numerical solutions are desired. They can namely, the Neumann boundary
be created, refined, or jiggled in the “Mesh Mode” of  + qu = g,
n · (c∇u) (2)
the GUI. The resulting triangular mesh data, stored by
point, edge, and triangle matrices, can be exported by and the Dirichlet boundary
the GUI and will be used not only when calling for the hu = r. (3)
PDE solvers but also when visualizing the solution.
In Eq. (2), c contains the PDE coefficients (here the dielec-
With all these requirements completed, the PDE problem tric constants in different regions), and n is the normal of the
is then defined, and the solution can as well be obtained by boundary. So the boundary conditions for given gate volt-
clicking “Solve PDE” within the GUI, which is user-friendly ages would be the Dirichlet type, with h = 1 and r being
but cannot be “programmed”. When performing certain real the corresponding voltage. For the Neumann type boundary
calculations, however, especially when a systematic change condition, we normally consider q = 0, and g represents the
of variables is required, programming with, e.g., looping, is surface charge.
inevitable and the requirements of (ii)–(iv) will be the neces- In the following, let us be more specific about the format
sary inputs of the PDE solvers. For our purpose of simulat- of the boundary conditions matrix, considering the two types
ing the carrier density modulation due to gating, we would of boundaries with N = 1.
often need to change the gate voltages, which are described
by the boundary conditions. Thus although each of (ii)–(iv) 2.2.2 Dirichlet boundary
can be programmed by using relevant commands, in the fol-
lowing only the implementation of (iii) by commands will Following the general description of Sect. 2.2.1, the bound-
be described in detail. ary matrix b for a Dirichlet boundary is described as follows.
• Row 1 contains the dimension N of the system: 1.
2.2 Boundary conditions • Row 2 contains the number M of Dirichlet boundary con-
ditions: 1.
2.2.1 The boundary condition matrix: General description • Row 3 contains the length for the strings representing q,
which is 1 since q = 0, though not used.
• Row 4 contains the length for the strings representing g,
By searching “assemb” from the M ATLAB help or by look-
which is 1 since g = 0, though not used.
ing up in its documentation [26], we see that the boundary
• Row 5 contains the length for the strings representing h,
conditions are saved in a matrix called b, with the following
which is 1 since h = 1.
data format:
• Row 6 contains the length for the strings representing r.
• Row 1 contains the dimension N of the system. (Note: • Then come the text expressions of q, g, h, r.
normally N = 1. If one solves two coupled variables, then An example of a Dirichlet boundary with r = 3.5 would
N = 2, etc.; by examining the exported boundary condi- be:
tion matrix b, one would find that N = 0 for inner bound-
aries.)
• Row 2 contains the number M of Dirichlet boundary con- 2 Themixed type boundary conditions will not be encountered in the
ditions. present discussion.
J Comput Electron (2013) 12:188–202 191

b = [1 1 1 1 1 3 ’0’ ’0’ ’1’ ’3.5’]’; double(num2str(Vtg))’

The boundary condition may include the x and y position Note that the operator ’ at the end of these two examples
coordinates and their functions, and can be written even in is to take transpose of the converted text strings, since the
terms of the solution u (nonlinear solver required). For ex- boundary conditions are saved columnwise in b, and simi-
ample, larly in the previous examples for b.
For a real PDE problem, the number of columns of the b
b = [1 1 1 1 1 4 ’0’ ’0’ ’1’ ’x.^2’]’;
matrix depends on the total number of edges, including inner
For another example, and outer boundaries. The nth column records the boundary
b = [1 1 1 1 1 9 ’0’ ’0’ ’1’ ’sin(x).^u’]’; condition for the nth edge. Thus before exporting the ini-
tial b matrix from the GUI of pdetool, one has to check the
2.2.3 Neumann boundary boundary labels corresponding to, e.g., graphene or gates
(by showing the edge labels in the “Boundary Mode”).
Following the general description of Sect. 2.2.1, the bound-
ary matrix b for a Neumann boundary is described as fol- 2.3 Some important commands
lows.
2.3.1 Solving the PDE
• Row 1 contains the dimension N of the system: 1.
• Row 2 contains the number M of Dirichlet boundary con- A standard PDE solver is called assempde. An example
ditions: 0. for its usage is as follows.
• Row 3 contains the length for the strings representing q,
u = assempde(b,p,e,t,c,a,f);
which is 1 since q = 0. % b: matrix of boundary conditions
• Row 4 contains the length for the strings representing g. % p: points of the mesh grid
• Then come the text expressions of q, g. % e: edges
% t: triangles
An example of a Neumann boundary with surface charge % c,a,f: coefficients of the pde problem
g = 1.6 would be
When the solution is involved in the boundary conditions,
b = [1 0 1 3 ’0’ ’1.6’]’; the solution mode has to be switched to nonlinear. An exam-
ple for its usage is as follows.
Another example
[u,res] = pdenonlin(b,p,e,t,c,a,f, ...
b = [1 0 1 21 ’0’ ’-13.295*sign(u).*u.^2’]’;
’report’,’on’,’MaxIter’,1e5,’u0’,u0);
will actually be used when we apply the Poisson-Dirac iter- % u: the solution, res: not important here
ation method. % b,p,e,t,c,a,f: same as above
% ’report’:iteration process report
% ’MaxIter’: maximal number of iter. rounds
2.2.4 Text expression of the boundary conditions % ’u0’: initial guess of the solution

The boundary condition matrix b exported from the GUI 2.3.2 Interpolation
of the pdetool looks filled with purely integers. This is the
“number representation” of the text strings. For example, a To find the values at those points one desires, an important
number 48 within the b matrix actually means ‘0’: command called tri2grid should be used, which inter-
>> char(48) polates from the PDE triangular mesh to a given rectangular
ans = 0 grid. An example for usage is as follows.
Conversely, if we want to transform the strings into num- uxy = tri2grid(p,t,u,x,y);
bers, we can simply use the ‘double’ command: % u: the obtained solution
% p,t: same as above
>> double(’x.^2’) % x,y: rectangular grid points for interpol.
ans = 120 46 94 50
2.4 Remarks on units
Thus to enter a boundary condition of, e.g., −(x 2 + y 2 ), we
may fill in with:
In the pdetool, everything is displayed with dimensionless
double(’-(x.^2+y.^2)’)’ numbers. The actual units can be chosen as what we would
like. Deducing relevant coefficients for a specific set of cho-
To enter a Dirichlet boundary condition of a given number
sen units is therefore important before we use the pdetool to
assigned by a variable named Vtg, we may fill in with:
solve any actual problems.
192 J Comput Electron (2013) 12:188–202

In addition to the physical units, the free space per-

mittivity ε0 is suppressed throughout the program. Recall
the boundary conditions for the displacement field D at a
conductor-dielectric boundary, which can be derived by ap-
plying the Gauss’s law: Dt = 0 and Dn = ρs , where Dt and
Dn represent the components tangential and normal to the
interface, respectively. The normal component of the dis-
placement field Dn therefore means the surface charge den-
sity:
ρs
−εr ∇u · n = . (4)
ε0
Comparing Eq. (4) to Eq. (2) (despite a minus sign due to
different convention of the sign of the surface normal) with
q = 0 and c representing εr (instead of ε), one can see that
the “Surface charge” actually means
ρs
g= , (5)
ε0
when filling the Neumann boundary condition in the
“Boundary Mode” of the GUI. Similarly, when filling the
PDE coefficients in the “PDE Mode”, the “Space charge Fig. 1 (a) Schematic of a single-gated graphene. (b) Equivalent circuit
plot of the classical capacitance model. (c) Equivalent circuit plot of the
density” actually means rho= ρ/ε0 , i.e., the right-hand side
quantum capacitance model
of Eq. (1).

charge (6) directly as those induced carriers in the graphene

layer, we have the carrier density
3 Theories of gate-induced carrier density modulation
ρs Cox
nC = = Vg , (7)
In this section, analytical theories of the gate-induced carrier −e e
density modulation, including the classical and quantum ca- which is a widely used formula for estimating the graphene
pacitance models, are briefly reviewed, a numerical scheme carrier density [2]. For uniform capacitors with Cox = ε/d,
of the self-consistent Poisson-Dirac iteration method is in- Eq. (7) numerically reads
troduced, and a numerical comparison between analytics εVg εr Vg
nC = = × 5.5263 × 1012 cm−2 , (8)
and numerics is provided. ed d
where Vg and d are in units of V and nm, respectively.
3.1 Classical capacitance model
3.1.2 Using pdetool
3.1.1 The model
The Dirichlet boundary conditions

0, at graphene boundary
We begin with the classical capacitance model, which con- u(x, z) = (9)
Vg , at gate boundary
siders a parallel-plate capacitor composed of an oxide di-
electric with permittivity ε = εr ε0 sandwiched by a metal- can be straightforwardly implemented in the pdetool.3 The
lic gate (at z = d) and a conducting graphene sheet (at standard PDE solver assempde introduced in Sect. 2.3
z = 0) as sketched in Fig. 1(a). Let the electric potential should be chosen. Working with units V and nm, the carrier
at the gate be u(x, z = d) = Vg and the graphene layer be density (7) is numerically given by
grounded: u(x, z = 0) = VG = 0. The surface charge den- 
∂u(x, z) 
sity at z = 0+ (the surface of the oxide dielectric in contact nC (x) = εr × 5.5263 × 1012 cm−2 . (10)
∂z  z=0
with the graphene layer) from Eq. (4) is given by

∂u  Vg − 0
ρs = −ε 
3 If a uniform capacitor (without x
= −ε = −Cox Vg , (6) dependence) is desired, one needs to
∂z z=0+ d −0 assign Neumann boundary conditions at the left and right sides of the
oxide boundaries with vanishing surface charge density g = 0, which
where Cox = ε/d is the classical capacitance (per unit area) forces the displacement field to be tangential (normal) to the side (top
of a uniform parallel-plate capacitor. Regarding this surface and bottom) boundaries.
J Comput Electron (2013) 12:188–202 193

Note that the interpolation command tri2grid intro- 3.2.1 Basic idea
duced in Sect. 2.3 may be useful in performing the numerical
derivative ∂u/∂z at z = 0. Consider a pristine graphene with Fermi level lying exactly
at the charge neutrality point, i.e., the Dirac point EF = 0.
3.1.3 Remark on the gate-induced Rashba spin splitting Application of the gate voltage Vg induces a certain amount
of additional charges on graphene, ρs = −en, which oc-
At this stage we may also estimate for graphene the gate- cupy the states in graphene according to its density of states
induced Rashba spin splitting, an intrinsic coupling between D(E) = 2|E|/π(vF )2 (within the Dirac model):
the spin and orbital degrees of freedom of charge carriers  E  
  1 E 2
in a two-dimensional conducting plane subject to a per- n(E) = D(E )dE = sgn(E) , (12)
0 π vF
pendicular electric field [27, 28]. In graphene, the Rashba
spin splitting has been shown by first principles to exhibit where vF ≈ 108 cm/s is the Fermi velocity of graphene.
a linear dependence on the electric field [29, 30]: R ≈ A positive (negative) electron number density n raises (low-
0.01|Ee | meV, where Ee is the electric field strength perpen- ers) the Fermi level from 0 to E. On the other hand, the
dicular to graphene given in units of V/nm. If the graphene electron at the Fermi level, which is responsible for trans-
carrier density stems from gating and is classically given by port in the linear response regime, gains an energy −eVG
n, the corresponding surface charge density |ρs | = e|n| = from the electric field, where −e is the electron charge4 and
ε|Ee |, in fact, has already revealed the displacement field VG is the electric potential at the graphene sheet obtained
on itself, allowing us to express the Rashba spin splitting in by solving the Poisson equation (1). This potential energy
terms of the carrier density, −eVG , which is equivalent to the “on-site energy” in the
n tight-binding transport formulation (see, for example, [21]),
R = × 1.8095 × 10−6 eV, (11) will raise the whole band structure, and thus lower the Fermi
εr
level by the same amount. We can therefore legitimately put
where n is in units of 1012 cm−2 , a typical order of the
graphene carrier density. E = −(−eVG ) = +eVG (13)
This estimation indicates that the Rashba spin splitting into Eq. (12), leading to
induced solely by electric gating typically lies in the order
of µeV, which may hinder the observation of those inter- ρs −en e (eVG )2
= = − sgn(VG ) . (14)
esting physics based on the Rashba spin-orbit coupling in ε0 ε0 ε0 π(vF )2
graphene, such as the interfacial spin and charge currents The surface charge density at the graphene layer is now ex-
[31, 32], or the spin-dependent Klein tunneling [33, 34]. pressed in terms of the solution u(x, z = 0) = VG , but is at
The position dependence of the Rashba coupling across a the same time the Neumann boundary condition that influ-
pn junction interface [35], on the other hand, can be accu- ences the numerical solution to the Poisson equation. This
rately taken into account by putting the x-dependence of n formally makes the solution process iterative.
(or even εr ) in Eq. (11). A stronger Rashba spin splitting
in graphene is therefore less possible by gating, but may be 3.2.2 Using pdetool
achieved by, for example, using a ferromagnetic substrate
with an intercalated gold monolayer [36]. The Dirichlet boundary condition (9) for the gate boundary
remains valid, while that for the graphene boundary has to
3.2 Self-consistent Poisson-Dirac iteration method be modified to the Neumann type:

g, at graphene boundary,
From Eq. (6) to Eq. (7), the assumption that “the in- u(x, z) = (15)
Vg , at gate boundary,
duced surface charge density at the dielectric surface is
the graphene carrier density” obviously has neglected a few where g = ρs /ε0 is given by Eq. (14). Working with units V
physical details, such as the graphene density of states that and nm together with vF = 108 cm/s, Eq. (14) becomes
govern the statistics of how the states in graphene should be  
ρs VG 2 V
filled by the carriers accordingly. In addition, filling the car- = −13.295 sgn(VG ) , (16)
ε0 V nm
riers into graphene causes the change of its Fermi level, im-
plying a potential energy shift that should further correspond which should be keyed as “-13.295*sign(u).*u. ˆ 2”
to the electric potential times the electron charge. These are in the boundary condition matrix, noting that the solution
what the classical capacitance model has neglected and what
the following Poisson-Dirac iteration method is going to 4 Throughoutthis paper, e = 1.60217733 × 10−19 C is the positive ele-
compensate. mentary charge.
194 J Comput Electron (2013) 12:188–202

in the pdetool is by default named u. The nonlinear solver Equating (18) and (19), one obtains a quadratic equation for
pdenonlin introduced in Sect. 2.3 has to be chosen in this VG ,
case, where the solution is involved in the boundary condi-  
1 eVG 2 Cox Cox
tions, and the iteration will be automatically processed by sgn(eVG ) = Vg − VG . (20)
the pdetool. π vF e e
Once the solution u(x, z), and hence the electrostatic po- Solving Eq. (20) for VG and putting back to Eq. (18), the
tential at the graphene layer VG (x) = u(x, z = 0), is itera- graphene carrier density can be written as
tively obtained, the desired carrier density profile n(x) can
n = nC + n, (21)
then be expressed in terms of VG (x):
where nC given by Eq. (7) is the classical contribution, and
  VG (x) 2
−2 
nPD (x) = 7.3471 × 10 13
× sgn VG (x) cm ,  
V |nC |
n = sgn(nC )nQ 1 − 1 + 2 , (22)
(17) nQ
which follows from Eqs. (12) and (13). Note that we have with definition
added explicitly a subscript “PD” in Eq. (17) to distinguish  
π Cox vF 2
with the classical contribution, nC . nQ = , (23)
2 e2
3.3 Quantum capacitance model corresponds to the quantum correction.6
Furthermore, by comparing Eq. (21) with Eq. (18), one
The relation between the induced charge density on graph- can also write down the solution for the electric potential on
ene and the electric potential energy that those charge carri- graphene:
ers gain through the graphene density of states is taken into 
account by the PDM, with a price of iteration process paid. en sgn(n )n
C Q (1 − 1 + 2 |nnQC | )
VG = − =− , (24)
For single-gated graphene, there is an alternative that can Cox Cox /e
take this into account analytically: the quantum capacitance
which has a reasonable form of charge divided by capaci-
model [19], which we briefly review here for bulk graphene
tance. The “charge” in Eq. (24) contains only the quantum
following the work of [16].5
correction as expected, since the classical solution, which re-
gards graphene as a grounded conductor, does not contribute
3.3.1 The model
to the potential VG .
The single-gated graphene shown in Fig. 1(a) is treated by
3.3.2 Quantum correction for parallel-plate capacitors
the equivalent circuit plot as shown in Fig. 1(c), where an ad-
ditional capacitor CQ is inserted between the voltage point
For parallel-plate capacitors with uniform dielectrics, the ca-
VG and the ground, as contrary to the CCM, Fig. 1(b). As
pacitance is well known to be Cox = ε/d, such that Eq. (23)
Fig. 1(c) suggests Vg = VG + Vox , using Cox = |ρ|/Vox =
and hence the quantum correction (22) are solely determined
en/Vox we have
by the classical contribution nC = (Cox /e)Vg . In this case
en we can further express Eq. (23) as nQ = (εr /d)2 × 2.0784 ×
Vg = VG + ⇒

Cox
   1011 cm−2 , where d is in units of nm and vF = 108 cm/s is
Cox Cox again adopted. Together with Eq. (8), the quantum correc-
n= Vg + − VG . (18)
e e tion given by Eq. (22) can be written as
 
classical quantum  2   
εr |Vg |d
Following the same physics stated in Sect. 3.2.1, the car- n = sgn(Vg ) 1 − 1 + 53.178
d εr
rier density at the graphene layer, i.e., Eq. (14) divided by
−e/ε0 , is expressed in terms of the electric potential thereof × 2.0784 × 1011 cm−2 , (25)
as
  where Vg is in units of V and sgn(nC ) = sgn(Cox Vg /e) =
1 eVG 2 sgn(Vg ) has been substituted. We will soon see that this cor-
n = sgn(eVG ) . (19)
π vF rection derived within the analytical QCM for an infinitely

5 For the general case of multigated doped graphene, see [20]. The 6 Note that Eqs. (21)–(23) (with n > 0) were first derived in [16] and
C
derivation is similar, and the review here is restricted to the simple reviewed in [2], but a factor of 2 in the square root of the formula (1.15)
case of single-gated pristine graphene. in [2], corresponding to Eq. (22) here, is missing.
J Comput Electron (2013) 12:188–202 195

Fig. 3 The quantum correction to the gate-induced carrier density on

graphene in the case of parallel-plate capacitor, calculated by the an-
alytical quantum capacitance model and the numerical Poisson-Dirac
Fig. 2 (Upper panel) The carrier density as a function of gate volt- iteration method. (Inset) Schematic of the capacitor with d = 300 nm
age Vg , with and without the quantum correction n, considering ox- with the color shading representing the iterated electric potential solu-
ide thickness d = 5, 20, 40 nm with εr = 3.9 of the dielectric SiO2 tion u(x, z) obtained by the PDM at Vg = 50 V
assumed. (Lower panel) Two-dimensional color plot of n/nC as a
function of Vg and d
exclusion principle, and hence the name quantum capaci-
tance [19], which is not restricted to the material graphene.
extending parallel-plate capacitor will exactly correspond to The expression of CQ for graphene [16], however, is not
that by the numerical PDM. important for the present discussion. Instead, CQ leads to a
To give an overview of how much change the quantum quantum correction to the gate-induced carrier density n,
correction n causes as compared to the classical nC , we which is the main focus here.
plot in the upper panel of Fig. 2 the carrier density, with and In addition, recent experimental progress on the measure-
without n, as a function of Vg , considering oxide thickness ment of graphene quantum capacitance [37–39] suggests
d = 5, 20, 40 nm with εr = 3.9 of the assumed dielectric that the electron-hole puddles [40] induced by charged im-
SiO2 . Apparently, only when d is extremely thin can one purities may influence CQ at energies close to the charge
see a clear difference due to n (d  20 nm). With nonzero neutrality point. The corresponding carrier density fluctua-
Vg and large d, one can further approximate Eq. (25) as tion δn, which can be considered to develop a microscopic
 3/2
εr  model to account for the smoothing of the graphene quan-
n ≈ − sgn(Vg ) |Vg | × 1.5156 × 1012 cm−2 , tum capacitance at the charge neutrality point [41], is be-
d
yond the scope of the present discussion.
(26)
which shows a rapid decay of n with d to the power of 3.4 Analytics vs numerics
3/2.
In the opposite limit of vanishing Vg and thin d, however, The two analytical capacitance models and the numerical
the magnitude of n may become comparable with nC . In scheme of the Poisson-Dirac iteration method are compared
the lower panel of Fig. 2, the ratio n/nC is plotted as a in the following, considering a single-gated graphene with
function of Vg and d. As expected, in the region away from individually infinite and finite size of the gate.
Vg = 0 and d  20 nm, the ratio is close to zero, meaning We start with uniform parallel-plate capacitors as those
a minor role played by the quantum correction. Contrarily, considered in Sect. 3.3.2 with different spacings d =
the ratio grows significantly when approaching to the Vg and 100, 200, 300 nm. Schematic of the capacitor with d =
d axes,7 implying an important role played by the quantum 300 nm is sketched in the inset of Fig. 3, where the electric
correction. potential u(x, z) within the oxide layer is obtained by the
PDM at Vg = 50 V.8 The classical contribution nC is first
3.3.3 Remark on quantum capacitance
computed following Sect. 3.1.2 (which gives the same result
Note that the appearance of CQ stems from the finite den-
sity of states provided by the conducting layer for the elec- 8 Note that the spatial profile of the electric potential u(x, z), with the
trons to occupy following the quantum nature of the Pauli quantum correction on graphene taken into account, does not look too
much different compared to the classical solution u0 (x, z), where the
graphene layer is assumed to be grounded. The difference of them at
7 The ratio further diverges to n/nC → −100 % at Vg = 0, but at this z = 0, however, is crucial since the latter is always zero, i.e., u0 (x, z =
axis both nC and n vanish, and n/nC is strictly speaking undefined. 0) = 0.
196 J Comput Electron (2013) 12:188–202

actly solvable QCM, circumventing the self-consistent iter-

ation during the solution process that is needed in the PDM.

3.5 Beyond single-gated pristine graphene

The above discussion considers only single-gated graphene

in the absence of chemical doping. For double-gated graph-
ene with topgate and backgate at two sides of the graphene
sheet, the two gates can be regarded as independent, and
their contributions to the carrier density modulation can be
treated separately and superposed to yield the total carrier
density. When multiple gates are acting on the graphene
Fig. 4 Comparison between the position-dependent carrier densities sheet from the same side, however, such as using an embed-
calculated by the PDM, nPD (x), the QCM, nQC (x), as well as CCM, ded local gate in addition to a global backgate to create pnp
nC (x), at various topgate voltages. (Inset) Schematic of the graphene junctions with independent control of the globally and lo-
sheet in the presence of the topgate, suspended 20 nm above, with cross
cally gated regions [13] (see Sect. 4.2), or patterned topgates
section 50 nm ×50 nm. The spatial distribution of the electric potential
u(x, z) is obtained by the iterative PDM at Vtg = 1.5 V that may generate a graphene superlattice (see Sect. 4.3),
these gates should be simultaneously treated.
In fact, the CCM (Sect. 3.1.2) as well as the PDM
with Eq. (8)), and the quantum correction n is computed in (Sect. 3.2.2) are not restricted to the case of single-gated
two ways. For the analytical QCM, Eq. (25) is used to com- graphene. These two approaches work for any kind of gating
pute n. For the numerical PDM, the full carrier density geometry, provided that the boundary conditions [Eq. (9) for
nPD is computed following Sect. 3.2.2, and the correction is CCM and Eq. (15) for PDM] at the graphene layer are prop-
given by the difference nPD − nC . As shown in Fig. 3, the erly assigned. The applicability of the QCM of the presently
correspondence between the two approaches is exact. reviewed version, however, depends then on the gating ge-
Next we consider a finite-size suspended topgate (such ometry. When multiple gates are acting on graphene from
as those fabricated in [7, 8]) with various voltages Vtg = the same side but connected to each other to share the same
0.5, 1.0, 1.5 V and an extremely narrow spacing d = 20 nm; gate voltage (as the case of Sect. 4.3), there is effectively
see the inset of Fig. 4. The calculations are similar to those only one gate. In this case, Eq. (27) is still a good approxi-
for the infinite case described above. The only difference is mation to account for the oxide capacitance, and the QCM
the approximating form of the dielectric capacitance, can be directly applied. On the other hand, if the multiple
e · nC (x) gates can be separately controlled (as the case of Sect. 4.2),
Cox (x) ≡ , (27) Eq. (27) becomes insufficient due to the need of multiple
Vtg
self-partial capacitances, and the QCM (of the presently re-
from which nQ given by Eq. (23) and hence the quantum viewed version) cannot be applied. Generalization of the
correction, Eq. (22) from the analytical QCM, is obtained. model to take into account composite gating geometry has
In Fig. 4, we compare the carrier densities calculated by the been recently achieved [20], but is beyond the scope of the
PDM, nPD , the QCM, nQC , as well as the CCM, nC . Good present review.
agreement between PDM and QCM is again clearly seen, For the general case of multigated doped graphene sheets,
while the deviation of the CCM from them is observable the boundary condition for the PDM, Eq. (14), as well as
due to the rather thin spacing d = 20 nm between the gate the analytical expressions within the QCM, Eqs. (21)–(24),
and the graphene sheet. can be derived similarly to the above reviewed theories. The
From the above testing calculations (Figs. 3 and 4), we interested readers are referred to the recent work of [20].
may conclude that the QCM is equivalent to the PDM in
both cases of infinite (uniform) and finite (nonuniform) gate-
graphene capacitors. In particular, the correspondence be- 4 Applications
tween the two approaches in the former case is exact, while
the discrepancy in the latter is merely negligible, suggesting A successful simulation for electronic transport in bulk
that Eq. (27) is a good approximation for calculating the spa- graphene relies on not only sophisticated computation tech-
tially varying oxide capacitance Cox (x) that further deter- niques but also a realistic “potential profile” V that is exper-
mines the quantum correction n given by Eq. (22) through imentally relevant [21]. The term “potential” refers to the lo-
the definition (23) within the QCM. The classical solu- cal potential energy added to the system Hamiltonian when
tion nC (x) for the nonuniform case (following Sect. 3.1.2), modeling for graphene electronic transport. Thus the poten-
therefore, serves as the preliminary solution step for the ex- tial profile simply means the local energy band offset of the
J Comput Electron (2013) 12:188–202 197

graphene sheet subject to a spatially varying carrier density

due to electrical gating. This section is devoted to the appli-
cation of the carrier density calculation: the corresponding
potential profile, or the local energy band offset, which is
a simple computational task but nevertheless important for
graphene electronic transport calculations. A few concrete
examples will be illustrated, after a short review of the po-
tential profile is given.

4.1 Potential profile (local energy band offset)

In Sect. 3 we have introduced how to compute with a sat-

isfactory accuracy the graphene carrier density n, which is
√ quasi-Fermi level through Eq. (12) as EF =
related to the
sgn(n)vF π|n|. The energy band offset then reads V =
EF0 − EF , where EF0 is the global Fermi level. Choosing
EF0 = 0 (as is usually the case and will be adopted in the
rest of the calculations), the space-resolved band offset reads Fig. 5 A graphene pnp junction using a global backgate with various
[21] voltages Vbg and an embedded local gate fixed at Vlg = 4 V. (a) The it-
   erative solution u(x, z) in units of V to the electrostatic potential within
V (x) = − sgn n(x) vF π|n(x)| the oxide, subject to Vbg = 30 V. The induced graphene carrier densi-
 ties (b) and the corresponding potential profiles (c) based on the CCM
  |n(x)| and PDM show that the locally gated region |x| ≤ 500 nm is not af-
= −11.667 × sgn n(x) meV, (28)
1010 cm−2 fected by the backgate but only the local gate
which is termed on-site energy in the tight-binding formula-
tion for transport calculations.
the embedded local gate at the region above it. Here we will
Equation (28) interprets the carrier density profile n(x)
mainly compare the results from the CCM and those from
in terms of the potential profile V (x), and is valid for n(x)
the PDM, fixing the local gate voltage at Vlg = 4 V while
computed by either CCM, QCM, or PDM. It should be re-
varying the backgate voltage with Vbg = −60, −30, 0, 30 V.
marked, however, that the Poisson-Dirac iterative solution
The computed carrier densities are shown in Fig. 5(b),
to the electric potential at the graphene layer VG times −e
readily gives the desired energy band offset, and Eq. (28) is where the Poisson-Dirac solution agrees well with the simu-
not needed within this approach. Likewise in the QCM, VG lation presented in [13]. Since the local gate is embedded
is given by Eq. (24) and its product with −e also gives the only 10 nm below the graphene sheet, the quantum cor-
desired V (x). Thus within QCM and PDM, one does not rection excluded in the CCM becomes pronounced within
need to bother with Eq. (28) for obtaining the potential pro- the locally gated region, and can be estimated by Eq. (25)
file. Contrarily, the CCM always treat the graphene sheet as of the QCM. At the center of the locally gated region,
a grounded conductor and therefore needs the interpretation the electric field generated by the backgate is almost com-
(28). In other cases where PDM is partly used but the total pletely screened, and the classical contribution to the car-
carrier density is separately computed (such as Sect. 4.3), rier density can be estimated by nC (x = 0) = εVlg /ed =
one needs Eq. (28) as well. 8.6211 × 1012 cm−2 , leading to nQ = 3.612 × 1010 cm−2
and hence n = −7.0735 × 1011 cm−2 , which is pretty
4.2 Graphene pnp junctions close to nPD (x = 0) − nC (x = 0) = −7.0746 × 1011 cm−2
from the data of Fig. 5(b) for all Vbg .
We begin the illustrative examples with a graphene pnp junc-
The carrier density profiles n(x) of Fig. 5(b) are trans-
tion, using a global backgate and an embedded local gate.
lated into V (x) via Eq. (28), as shown in Fig. 5(c). The pos-
The gating geometry is sketched in Fig. 5(a), similar to those
itive Vlg charges the locally gated graphene with a positive
experimentally fabricated in [13]. In this case both of the
number of electrons, forming an n-type region with positive
global and local gates influence the graphene carrier density
from the same side, and therefore have to be treated at the quasi-Fermi level EF (x) > 0 that is equivalent to applying
same time. As remarked previously in Sect. 3.5, the single- a negative energy band offset V (x) < 0. Outside the locally
gate version of the QCM does not apply here,9 but neverthe- gated region, the carrier type of graphene is controlled by
less can be used to estimate the quantum correction due to the backgate with a similar principle. The most interesting

9 The multigate version of the QCM [20], which requires to compute gate and the backgate, can be shown to yield results well agreeing with
the self-partial capacitances Clg and Cbg due to respectively the local the PDM.
198 J Comput Electron (2013) 12:188–202

sketched in Fig. 6(a), the PDE problem is defined within

the region above the graphene sheet in order to solve the
electric potential due to the topgates with various voltages
Vtg = −5, −3, . . . , 5 V. The contribution from the backgate
is assumed to be uniform and can be treated independently.
The strategy here is to compute first the carrier density due
to topgates, and then include the backgate contribution to
yield the total carrier density that finally gives the energy
band offset profile from Eq. (28). In this case the approxi-
mation (27) for the QCM is rather acceptable, and we will
compare the results from all of the three approaches.
The computed carrier densities by PDM, QCM, and
CCM are shown in Fig. 6(b). The curves of nPD and nQC
almost coincide with each other. The relatively thick 40 nm
of Al2 O3 suppresses the quantum correction to a reasonably
small amount, such that here the CCM is not a bad approx-
imation, either. The discrepancy between nC and nPD (or
nQC ) is less pronounced at the regions between each ad-
jacent pair of topgates since the quantum correction |n|
roughly decreases with the 3/2-th power of the distance to
the gate, as mentioned in Eq. (26).
The carrier density modulation follows the patterned top-
gates with a periodicity of 100 nm, giving rise to a periodic
potential profile V (x) as shown in Fig. 6(c), where a back-
gate contribution with Vbg = 30 V is taken into account.
Since V (x) is related to n(x) through a square-root rela-
tion, Eq. (28), the shape of V (x) can be a bit different from
that of n(x), which is similar to a sine-like wave, especially
for those V (x) that alternate between positive and negative
values. Note that the backgate voltage chosen in Fig. 6(c)
results in a rather symmetric Vtg = −3 V curve since the
corresponding carrier density n(x) alternates symmetrically
between positive and negative at this combination of gate
voltages. In general, the alternation of n(x) is not necessar-
Fig. 6 Formation of a graphene superlattice using a series of patterned ily symmetric (about the charge neutrality point n = 0), and
topgates and a global backgate, which are treated separately. (a) Iter- the resulting V (x) profile can be of peculiar shapes.
ative solution u(x, z) with Vtg = 5 V, considering only the topgates.
(b) The corresponding carrier densities based on all of the three ap- 4.4 Linear potential
proaches. (c) The profiles of the energy band offset from the total car-
rier density composed of the patterned topgates and the uniform back-
gate contributions
In the previous example, we have demonstrated that fabri-
cating a series of patterned topgates may generate a peri-
odic potential, which, combined with a potential linear in
feature here is that the locally gated region can be controlled position as well as the periodically alternating mass poten-
independently due to the screening of the embedded local tial, forms the prerequisite of the Bloch-Zener oscillation
gate, as is evident in both Figs. 5(b) and 5(c). This indepen- in graphene [23]. In this demonstrating example, we point
dent control leads to the four quadrants of the conductance out a simple way to generate the linear potential: using a
map G(Vlg , Vbg ) with two boundaries perpendicular to each tilted backgate. As sketched in Fig. 7(a), where we consider
other [13], as contrary to those observed in top-gated devices a position-varying thickness of SiO2 with a slope of s = 0.05
[4–8, 11, 12]. (an increase of 50 nm per micron). In this case the quantum
correction does not play a role (see Sect. 3.3.2), and we show
4.3 Graphene superlattices in Fig. 7(b) only the carrier densities from the CCM and the
PDM, which coincide to each other.
Next we turn to the possibility of generating a graphene su- Since the classical capacitance model works well here,
perlattice by fabricating a series of patterned topgates. As with the oxide thickness d(x) = d0 +sx, where d0 = 300 nm
J Comput Electron (2013) 12:188–202 199

of Pd, W = 4.03 eV, so WG = 4.48 eV leads to EF =

−0.45 eV, according to the table summarized in [24]. This
means that the Fermi level of graphene coated by Pd(111)
is expected to be EF = 0.45 eV. Furthermore, the contact-
induced screening potential in graphene was calculated in
[25] using the density functional theory within the Thomas-
Fermi approximation, which is similar to the PDM intro-
duced here since the underlying equation that governs the
electrostatic potential is still the Poisson equation. In their
formulation, the metal contact is oriented at x ≤ 0, z ≥ 0
(graphene sheet also at z = 0), and the boundary condi-
tion for the surface (x = 0, z ≥ 0) of the contact is given
by Vc = (WM − WG )/e, where WM is the work function of
the corresponding clean metal. In the case of Pd and again
according to the table summarized in [24], WM = 5.67 eV,
leading to Vc = 1.19 V.
Fig. 7 (a) Electrostatic potential u(x, y) inside a trapezoidal oxide The numerically exact solution for the electric potential
layer. The tilted backgate with a slope of 0.05 generates a carrier den- in graphene, i.e., VG for x ≥ 0, was further fitted by a varia-
sity (b) that varies almost linearly with position. The corresponding tional solution [25]
potential profiles (c) also exhibit a linear behavior
VB
V (x) ≈ − 
is the thickness at the center, we can describe the carrier ( x/ ls + β22 + β1 − β2 )1/2 (x/ ls + β1−2 )1/4
density as n(x) = εVg /e(d 0 + sx) and hence the potential (30)
as V (x) = − sgn(Vg )vF πε|Vg |/e(d0 + sx). The slope
of the potential at x = 0, dV (x)/dx|x=0 , together with the with fitting parameters β1 = 0.915 and β2 = 0.128. The
intercept V (x = 0), allows us to approximate the potential scaling length ls in Eq. (30) is defined as ls = vF /πα|VB |
with a linear model, with α = e2 /4πε0 εr vF = 2.1877/εr , where vF = 108 cm/s
is assumed.10 For Pd, from their table (with β = π/2) one
V (x) ≈ V0 + Sx
 finds VB = 0.48 eV. Thus for vacuum with εr = 1, we have
εr |Vg | ls = 0.19952 nm. We will compare with this variational so-
V0 = − sgn(Vg ) × 0.27425 eV (29)
d0 lution (30).
 To apply the presently reviewed PDM and QCM to re-
s εr |Vg |
S = sgn(Vg ) 3/2
× 0.13713 eV nm−1 . solve the graphene electric potential VG , while keeping the
d0 results of [24, 25] unchanged, namely,
In Fig. 7(c), we plot the potential profiles obtained from nPD
(i) VG (x ≤ 0) = EF (x ≤ 0)/e = 0.45 V ≡ VG<
and from the linear model given by Eq. (29); the consistency
(ii) VG (x ≥ 0) decays nonlinearly with x
is almost perfect.
we may model the metal contact as a slightly floating gate
4.5 Contact-induced doping and screening potential located at z = zc , as schematically shown in the inset of
Fig. 8. By such modeling, both of the above stated con-
In the last application example, we turn to a practical is- clusions (i) and (ii) can be satisfied at one time by taking
sue for graphene electronics: contact-induced doping and its the same input of the boundary condition. Specifically, the
screening potential. Taking palladium as the specific exam- boundary conditions to be applied here are u(x = 0, z ≥
ple, we show that the solutions for the electric potential in zc ) = u(x ≤ 0, z = zc ) = Vc , with the same Vc = 1.19 V
graphene VG induced by the Pd contact solved by the PDM according to [25].
and by the QCM not only agree with each other but also are
To satisfy (i), we may apply the QCM for the parallel-
consistent with the previously reported results based on first
plate capacitor to deduce a proper floating height zc in order
principles [24, 25], which we first briefly review as follows.
to meet the proper shift of the Fermi level VG< = EF (x ≤
Previous first-principles study expected ideal Pd(111)
0)/e = 0.45 V. Thus the upper plate of the capacitor is the
contacts to dope graphene as n-type, with the shift of the
Fermi level given by EF = W − WG , where W is the work
function of the metal coated with graphene, and WG is the 10 In[25], α is given by 2.38/εr possibly because of the slightly differ-
work function of free-standing graphene [24]. In the case ent Fermi velocity vF .
200 J Comput Electron (2013) 12:188–202

5 Summary

In conclusion, theories of the gate-induced carrier density

modulation in bulk graphene have been reviewed. The clas-
sical capacitance model, the widely adopted tool for carrier
density estimation, does not include the quantum correc-
tion but nevertheless plays usually the dominant role in the
gate modulation, unless the metal is rather close to graphene
(Fig. 2). The quantum correction stems from the finite ca-
pacity of the graphene sheet for the electrons to reside, and
can be treated by the self-consistent Poisson-Dirac itera-
tion method, as well as the exactly solvable quantum ca-
Fig. 8 Contact-induced doping and screening potential in the
pacitance model [16, 20]. By inspecting the numerical ex-
graphene sheet. The ideal Pd(111) contact is modeled by a “slightly
floating gate” with potential Vc = (WM − WG )/e = 1.19 V; quantities amples of single-gated graphene, these two approaches are
WM and WG are described in the text. (Inset) Schematic of the contac- shown to agree with each other. In particular, the correspon-
t-graphene system, with the electric potential u(x, z) obtained by the dence is exact for the case of infinite parallel-plate capaci-
PDM
tors (Fig. 3). For the case with finite gates, the agreement be-
tween QCM and PDM remains good (Fig. 4), implying that
Pd contact with voltage Vc , while the lower plate is graphene Eq. (27) is a good approximation for numerically calculating
with voltage VG< . Equation (24) therefore reads the spatially varying capacitance. This further suggests that
  the classical solution nC (x) serves as the preliminary solu-
nQ |nC | tion step for the exactly solvable QCM, circumventing the
VG =
<
1+2 − 1 Vc , (31)
nC nQ self-consistent iteration during the solution process that is
needed in the PDM. Along with the brief introduction to the
where nC = Cc Vc /e, with Cc = ε0 /zc the classical capaci- usage of the M ATLAB pdetool, the former part of this work
tance of the contact-vacuum-graphene sandwich, has been (Sects. 2–3) provides a self-contained instruction to calcu-
substituted. Thus using nC /nQ = (2Vc /π)(e/vF )2 (e/ lating the carrier density of pristine graphene sheets subject
ε0 )zd in Eq. (31), one may solve for zc to obtain to complicated gating geometry. For the generalized theory
  for multigated doped graphene, the readers are referred to
vF 2 ε0 Vc − VG<
zc = π . (32) [20].
e e (VG< )2
To demonstrate the applicability of the introduced CCM,
For the present case of Pd and following VG< = 0.45 V of PDM, and QCM, the latter part of this work (Sect. 4) is
[24] and Vc = 1.19 V of [25], this effective height given by devoted to illustration of practical examples for calculating
Eq. (32) amounts to zc ≈ 0.27 nm. gate-induced carrier density in graphene sheets, including
Setting zc = 0.27 nm for the contact and treating it as a the graphene pnp junction using an embedded local gate in
“gate” with fixed potential Vc = 1.19 V, the electric poten- addition to a global backgate (Fig. 5), graphene superlattice
tial in the graphene sheet VG is calculated by using the self- potential by a series of patterned topgates (Fig. 6), quasi-
consistent PDM and the analytical QCM, as shown in Fig. 8. linear potential by using a tilted backgate (Fig. 7), and finally
The two approaches again coincide with each other. In ad- the contact-induced doping and screening potential (Fig. 8).
dition, VG (x ≤ 0) ≈ 0.45 V is clearly observed, while the The first three examples correspond to the experimental con-
nonlinearly decaying VG (x ≥ 0) agrees well with the varia- ditions that provide a flexible platform to test the physics of
tional solution of [25], thus satisfying both (i) and (ii). Klein backscattering [12, 13, 21, 22] and the Bloch-Zener
Note that despite the consistency with the previous theory oscillation [23] in graphene, while the last example shows
shown here, experiments for transport measurements usu- that the effects of metal contacts can be treated equally well
ally do not have single-crystal contacts grown along (111), by the PDM and QCM, as compared to the previous first-
and the contact/graphene interface is certainly dirty. The principles studies [24, 25]. In either case, once the realis-
charge transfer between the metal contact and graphene due tic potential profile V (x) is obtained, satisfactory electronic
to their different work functions is, therefore, greatly re- transport calculation for the relevant structure following [21]
duced, leading to a much lower Vc . For example, a recent ex- can then be guaranteed.
periment observing the ballistic interferences in ultra-clean
Acknowledgements The author thanks T. Fang and D. Jena for their
suspended graphene uses Pd as contacts, and theoretical
illuminating suggestions, F.-X. Schrettenbrunner, J. Eroms, P. Rick-
modeling with Vc of the order of 0.01 V is found to better fit haus, and R. Maurand for sharing their experimental viewpoints, and
the transport measurement [42]. V. Krueckl and K. Richter for valuable discussions. Financial supports
J Comput Electron (2013) 12:188–202 201

from Alexander von Humboldt Foundation (former part of the work) 17. Shylau, A.A., Kłos, J.W., Zozoulenko, I.V.: Capacitance of
and Deutsche Forschungsgemeinschaft within SFB 689 (present) are graphene nanoribbons. Phys. Rev. B 80, 205402 (2009). Available
gratefully acknowledged. online: http://link.aps.org/doi/10.1103/PhysRevB.80.205402
18. Andrijauskas, T., Shylau, A.A., Zozoulenko, I.V.: Thomas-Fermi
and Poisson modeling of gate electrostatics in graphene nanorib-
bon. Liet. Fiz. Zh. 52(1), 63–69 (2012)
References 19. Luryi, S.: Quantum capacitance devices. Appl. Phys. Lett.
52(6), 501–503 (1988). Available online: http://link.aip.org/
1. Castro Neto, A.H., Guinea, F., Peres, N.M.R., Novoselov, K.S., link/?APL/52/501/1
Geim, A.K.: The electronic properties of graphene. Rev. Mod. 20. Liu, M.-H.: Theory of carrier density in multigated doped
Phys. 81, 109 (2009) graphene sheets with quantum correction. Phys. Rev. B
2. Das Sarma, S., Adam, S., Hwang, E.H., Rossi, E.: Elec- 87, 125427 (2013). Available online: http://link.aps.org/
tronic transport in two-dimensional graphene. Rev. Mod. doi/10.1103/PhysRevB.87.125427
Phys. 83, 407–470 (2011). Available online: http://link.aps.org/ 21. Liu, M.-H., Richter, K.: Efficient quantum transport sim-
doi/10.1103/RevModPhys.83.407 ulation for bulk graphene heterojunctions. Phys. Rev. B
3. Novoselov, K.S., Geim, A.K., Morozov, S.V., Jiang, D., Zhang, 86, 115455 (2012). Available online: http://link.aps.org/doi/
Y., Dubonos, S.V., Grigorieva, I.V., Firsov, A.A.: Electric field ef- 10.1103/PhysRevB.86.115455
fect in atomically thin carbon films. Science 306(5696), 666–669 22. Shytov, A.V., Rudner, M.S., Levitov, L.S.: Klein backscat-
(2004) tering and Fabry-Pérot interference in graphene heterojunc-
4. Huard, B., Sulpizio, J.A., Stander, N., Todd, K., Yang, tions. Phys. Rev. Lett. 101, 156804 (2008). Available online:
B., Goldhaber-Gordon, D.: Transport measurements across a http://link.aps.org/doi/10.1103/PhysRevLett.101.156804
tunable potential barrier in graphene. Phys. Rev. Lett. 98,
23. Krueckl, V., Richter, K.: Bloch-Zener oscillations in graphene and
236803 (2007). Available online: http://link.aps.org/doi/10.1103/
topological insulators. Phys. Rev. B 85, 115433 (2012). Available
PhysRevLett.98.236803
online: http://link.aps.org/doi/10.1103/PhysRevB.85.115433
5. Williams, J.R., DiCarlo, L., Marcus, C.M.: Quantum hall effect in
24. Khomyakov, P.A., Giovannetti, G., Rusu, P.C., Brocks, G.,
a gate-controlled p–n junction of graphene. Science 317(5838),
van den Brink, J., Kelly, P.J.: First-principles study of the inter-
638–641 (2007). Available online: http://www.sciencemag.org/
action and charge transfer between graphene and metals. Phys.
content/317/5838/638.abstract
Rev. B 79, 195425 (2009). Available online: http://link.aps.
6. Özyilmaz, B., Jarillo-Herrero, P., Efetov, D., Abanin, D.A.,
org/doi/10.1103/PhysRevB.79.195425
Levitov, L.S., Kim, P.: Electronic transport and quantum
25. Khomyakov, P.A., Starikov, A.A., Brocks, G., Kelly, P.J., Non-
hall effect in bipolar graphene p–n–p junctions. Phys. Rev.
linear screening of charges induced in graphene by metal
Lett. 99, 166804 (2007). Available online: http://link.aps.org/
contacts. Phys. Rev. B 82, 115437 (2010). Available online:
doi/10.1103/PhysRevLett.99.166804
http://link.aps.org/doi/10.1103/PhysRevB.82.115437
7. Liu, G., Velasco, J.J., Bao, W., Lau, C.N.: Fabrication of
graphene p–n–p junctions with contactless top gates. Appl. 26. Partial differential equation toolboxTM User’s guide, Matlab
Phys. Lett. 92(20), 203103 (2008). Available online: http://dx.doi. 2012a ed., The MathWorks, Inc. (2012)
org/doi/10.1063/1.2928234 27. Rashba, E.I.: Properties of semiconductors with an extremum
8. Gorbachev, R.V., Mayorov, A.S., Savchenko, A.K., Horsell, D.W., loop. I. Cyclotron and combinational resonance in a magnetic field
Guinea, F.: Conductance of p–n–p graphene structures with “air- perpendicular to the plane of the loop. Sov. Phys., Solid State 2,
bridge” top gates. Nano Lett. 8(7), 1995–1999 (2008). Available 1109 (1960)
online: http://pubs.acs.org/doi/abs/10.1021/nl801059v 28. Bychkov, Y.A., Rashba, E.I.: Properties of a 2d electron-gas with
9. Cheianov, V.V., Fal’ko, V.I.: Selective transmission of Dirac lifted spectral degeneracy. JETP Lett. 39, 78 (1984)
electrons and ballistic magnetoresistance of n–p junctions in 29. Gmitra, M., Konschuh, S., Ertler, C., Ambrosch-Draxl, C., Fabian,
graphene. Phys. Rev. B 74(4), 041403 (2006) J.: Band-structure topologies of graphene: spin-orbit coupling ef-
10. Katsnelson, M.I., Novoselov, K.S., Geim, A.K.: Chiral tunnelling fects from first principles. Phys. Rev. B 80, 235431 (2009). Avail-
and the Klein paradox in graphene. Nat. Phys. 2(9), 620 (2006) able online: http://link.aps.org/doi/10.1103/PhysRevB.80.235431
11. Stander, N., Huard, B., Goldhaber-Gordon, D.: Evidence 30. Abdelouahed, S., Ernst, A., Henk, J., Maznichenko, I.V., Mertig,
for Klein tunneling in graphene p–n junctions. Phys. Rev. I.: Spin-split electronic states in graphene: effects due to lattice
Lett. 102, 026807 (2009). Available online: http://link.aps.org/ deformation, Rashba effect, and adatoms by first principles. Phys.
doi/10.1103/PhysRevLett.102.026807 Rev. B 82, 125424 (2010). Available online: http://link.aps.org/
12. Young, A.F., Kim, P.: Quantum interference and Klein tunnelling doi/10.1103/PhysRevB.82.125424
in graphene heterojunctions. Nat. Phys. 5(3), 222–226 (2009) 31. Yamakage, A., Imura, K.-I., Cayssol, J., Kuramoto, Y.: Interfa-
13. Nam, S.-G., Ki, D.-K., Park, J.W., Kim, Y., Kim, J.S., Lee, H.-J.: cial charge and spin transport in 2 topological insulators. Phys.
Ballistic transport of graphene pnp junctions with embedded local Rev. B 83, 125401 (2011). Available online: http://link.aps.org/
gates. Nanotechnology 22(41), 415203 (2011) doi/10.1103/PhysRevB.83.125401
14. Guo, J., Yoon, Y., Ouyang, Y.: Gate electrostatics and quan- 32. Tian, H.Y., Yang, Y.H., Wang, J.: Interfacial charge current in
tum capacitance of graphene nanoribbons. Nano Lett. 7(7), a magnetised/normal graphene junction. Eur. Phys. J. B 85(8)
1935–1940 (2007). Available online: http://pubs.acs.org/doi/abs/ (2012)
10.1021/nl0706190 33. Yamakage, A., Imura, K.I., Cayssol, J., Kuramoto, Y.: Spin-orbit
15. Fernández-Rossier, J., Palacios, J.J., Brey, L.: Electronic struc- effects in a graphene bipolar pn junction. EPL 87(4) (2009)
ture of gated graphene and graphene ribbons. Phys. Rev. 34. Liu, M.-H., Bundesmann, J., Richter, K.: Spin-dependent Klein
B 75, 205441 (2007). Available online: http://link.aps.org/ tunneling in graphene: role of Rashba spin-orbit coupling. Phys.
doi/10.1103/PhysRevB.75.205441 Rev. B 85, 085406 (2012). Available online: http://link.aps.org/
16. Fang, T., Konar, A., Xing, H., Jena, D., Carrier statistics doi/10.1103/PhysRevB.85.085406
and quantum capacitance of graphene sheets and ribbons. 35. Rataj, M., Barnaś, J.: Graphene p–n junctions with nonuniform
Appl. Phys. Lett. 91(9), 092109 (2007). Available online: Rashba spin-orbit coupling. Appl. Phys. Lett. 99(16), 162107
http://link.aip.org/link/?APL/91/092109/1 (2011). Available online: http://dx.doi.org/10.1063/1.3641873
202 J Comput Electron (2013) 12:188–202

36. Sánchez-Barriga, J., Varykhalov, A., Scholz, M.R., Rader, O., sity of states and zero landau level probed through capacitance of
Marchenko, D., Rybkin, A., Shikin, A.M., Vescovo, E.: Chemical graphene. Phys. Rev. Lett. 105, 136801 (2010). Available online:
vapour deposition of graphene on Ni(111) and Co(0001) and inter- http://link.aps.org/doi/10.1103/PhysRevLett.105.136801
calation with Au to study Dirac-Cone formation and Rashba split-
40. Martin, J., Akerman, N., Ulbricht, G., Lohmann, T., Smet, J.H.,
ting. In: Diamond and Related Materials, 19, no. 7–9, pp. 734–
Von Klitzing, K., Yacoby, A.: Observation of electron-hole pud-
741, 20th European Conference on Diamond, Diamond-Like Ma-
terials, Carbon Nanotubes and Nitrides, Athens, Greece, 6–10 Sep. dles in graphene using a scanning single-electron transistor. Nat.
2009 (2009) Phys. 4(2), 144–148 (2008)
37. Xia, J., Chen, F., Li, J., Tao, N.: Measurement of the quantum 41. Xu, H., Zhang, Z., Peng, L.-M.: Measurements and mi-
capacitance of graphene. Nat. Nanotechnol. 4(8), 505–509 (2009) croscopic model of quantum capacitance in graphene.
38. Dröscher, S., Roulleau, P., Molitor, F., Studerus, P., Stampfer, C., Appl. Phys. Lett. 98(13), 133122 (2011). Available online:
Ensslin, K., Ihn, T.: Quantum capacitance and density of states http://link.aip.org/link/?APL/98/133122/1
of graphene. Appl. Phys. Lett. 96(15), 152104 (2010). Available
online: http://link.aip.org/link/?APL/96/152104/1 42. Rickhaus, P., Maurand, R., Liu, M.-H., Weiss, M., Richter, K.,
39. Ponomarenko, L.A., Yang, R., Gorbachev, R.V., Blake, P., May- Schönenberger, C.: Ballistic interferences in suspended graphene.
orov, A.S., Novoselov, K.S., Katsnelson, M.I., Geim, A.K.: Den- March (2013, unpublished)