Professional Documents
Culture Documents
net/publication/257855073
CITATIONS READS
18 644
6 authors, including:
Alex Burenkov
Fraunhofer Institute for Integrated Systems and Device Technology IISB
133 PUBLICATIONS 664 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Dmitry Svintsov on 21 May 2014.
IX INTERNATIONAL CONFERENCE
“SILICON2012”,
ST. PETERSBURG, JULY 9–13, 2012
Abstract—The lack of an OFFstate has been the main obstacle to the application of graphenebased tran
sistors in digital circuits. Recently vertical graphene tunnel fieldeffect transistors with a low OFFstate cur
rent have been reported; however, they exhibited a relatively weak effect of gate voltage on channel conduc
tivity. We propose a novel lateral tunnel graphene transistor with the channel conductivity effectively con
trolled by the gate voltage and the subthreshold slope approaching the thermionic limit. The proposed
transistor has a semiconductor (dielectric) tunnel gap in the channel operated by gate and exhibits both high
ONstate current inherent to graphene channels and low OFFstate current inherent to semiconductor
channels.
DOI: 10.1134/S1063782613020218
279
280 SVINTSOV et al.
(a) Far from the tunnel contact and source and drain
electrodes, the local electric potential in the channel
Source Gate Drain does not depend on coordinate and equals ϕ0 (the
source is grounded). The value of ϕ0 can be obtained
Boron nitride Graphene from the gradual channel approximation [6] yield
ing the local relation between the charge density
Silicon and voltage:
−L/2 L/2 x κκ
(b) 0 ( V G – ϕ 0 ) = e [ n e ( ϕ 0 ) – n h ( ϕ 0 ) ], (2)
Source Drain
d
Oxide Gate
where κ is the permittivity of the gate dielectric
(κ = 5.04 for hexagonal boron nitride [7]), κ0 is the
Boron nitride Graphene electric constant, e > 0 is the elementary charge, ne and
Silicon nh are the electron and hole concentrations per unit
area.
x Assuming that the Fermi level in graphene at the
source and drain contacts is located exactly between
Fig. 1. Schematic view of the proposed FET structures:
(a) tunnel contact in the middle of the channel, (b) tunnel the conduction and valence bands (at the so called
contact near the drain. Dirac point), one can directly obtain the electron and
hole concentrations in the channel. As eϕ0 is the local
Fermi energy, at low temperatures (kT eϕ0) and a
work function of the transition of electron from positive gate potential one arrives at a simple expres
graphene to the barrier material should be low. The sion for the electron concentration:
barrier height is governed by the material of the tunnel
2 2
contact; it is rather high for dielectric materials. In [4] e ϕ0
a barrier height of 1.5 eV for holes and 4 eV for elec n e =
2 2
. (3)
trons was reported for the graphene–hexagonal boron π v F
nitride interface. To reduce the work function one can Hence, the local electric potential in the channel is
use a semiconductor material as the tunnel contact given by
(e.g. silicon); in this case the barrier height equals
approximately half the semiconductor band gap.
πv F κκ 0 ⎛ 4de V G ⎞
2 2 3
eϕ0 eϕ0
eVG
Φ0 Φ0
eϕ0 eϕ0
Φ0
eϕ0 Φ0
eϕ0
Gated section
Gated section
Fig. 3. Band diagrams for a graphene FET with a tunnel
Fig. 2. Band diagrams for a graphene FET with a tunnel contact near the drain at zero drain voltage (from the top)
contact in the middle of the channel (L d) at zero drain and positive drain voltage VD. A cross indicates the posi
voltage (from the top) and positive drain voltage VD. tion of the Dirac point with respect to the band diagram.
A cross indicates the position of the Dirac point with
respect to the band diagram.
In the above expression εc(x) is the position of the
tunnel contact. The latter assumption is valid if the conduction band bottom in the tunnel contact, mx is
resistance of the tunnel contact markedly exceeds the the effective tunneling mass, which is, in fact, a fitting
resistance of the graphene sheet. In both situations parameter obtained from comparison with experi
(Figs. 2 and 3) the tunnel barrier will be treated as mental results [11], [12], and m⊥ is the transverse
trapezoidal in form. effective mass (the latter masses, generally, can be dif
ferent).
The calculation of trapezoidalbarrier penetrability
Calculation of the Tunnel Current yields the following result:
Tunneling through the band gap of a semiconduc
4 2m ⎧
2 3/2
D ( ε, p ⊥ ) = exp – x ⎨ ⎛ Φ +
– ε⎞
tor was studied by Kane [8] and Keldysh [9] in the p⊥
middle of the 20th century, however, the generaliza 3eF ⎩ ⎝ 2m ⊥ ⎠
tion of these works accounting for the exact band
structure of silicon appeared quite recently [10]. 2 3/2
⎫
– ⎛ Φ – eV D +
– ε⎞
To our knowledge, a theoretical model for tunneling p⊥
⎝ ⎬
through the band gap in a graphene–silicon system 2m ⊥ ⎠ ⎭
has not yet been developed and it deserves separate (7)
study. In the following calculations we shall apply a at ε–
2
p ⊥ /2m ⊥ < Φ – eV D ,
simplified model sufficient for preliminary estima
2 3/2
tions of the FET characteristics. The model is based
D ( ε, p ⊥ ) = exp – x ⎛ Φ +
– ε⎞
4 2m p⊥
on quasiclassical description of tunneling: ⎝
3eF 2m ⊥ ⎠
x2
⎛ ⎞ at
2
Φ – eV D < ε – p ⊥ /2m ⊥ < Φ,
D ≈ exp ⎜ – 2 p x ( x ) dx⎟ ,
⎝ ∫ ⎠
(5)
where Φ is the effective barrier height, for structures
x1
with a long tunnel gap (L d) in the middle of the
where x1 and x2 are classical turning points, and the channel (Fig. 1a) it is efficiently controlled by the gate
momentum component px in the barrier region can be voltage:
obtained from energy and transverse (orthogonal to
the x axis) momentum conservation: Φ A = Φ 0 + e ( ϕ 0 – V G ). (8)
2 2 In structures with a short tunnel gap near the drain
px p⊥
ε = ε c ( x ) +
+
. (6) (Fig. 1b, band diagram in Fig. 3) the effective barrier
2m x 2m ⊥ height barely depends on the gate voltage: Φ = Φ0. In the
1
1
0
0 0.2 0.4 0.6 0.8 1.0 0
0 0.1 0.2 0.3 0.4 0.5
Gate voltage VG, V Drain voltage VD, V
Fig. 4. Current vs. gate voltage for a graphene FET with a Fig. 5. Current vs. drain voltage for graphene FET with a
tunnel contact in the middle of the channel. Gap width is tunnel contact in the middle of the channel. Gap length is
L = 5 nm, gate dielectric thickness is d = 1 nm, tempera L = 5 nm, gate dielectric thickness is d = 1 nm, tempera
ture is T = 300 K. ture is T = 300 K.
above expressions Φ0 is the work function of the tran tions in the source and drain contacts, respectively.
sition of electron from graphene to the tunnelcontact If the energy is counted from the Dirac point, these
material at zero gate voltage, counted off from the functions are given by:
Dirac point in graphene. The electric field in the gap is
denoted by F in Eq. (7). The expressions for F are dif 1
f S ( ε ) =
,
ferent for the two structures: ε – eϕ 0⎞
⎛
1 + exp
V ⎝ kT ⎠
F A = e D , (9) (12)
L 1
f D ( ε ) = .
VD – ϕ0 ε – eϕ 0 + eV D⎞
F B = e
, (10) 1 + exp ⎛
L ⎝ kT ⎠
for the structures in Fig. 1a and Fig. 1b, respectively. Integration over the transverse momentum com
The authors of [13] proposed to substitute the ponent is performed from zero to the maximum
“light” mass of the conduction band for mx in Eq. (7) momentum at the given energy p⊥max = ε/vF; integra
as the tunneling transparency for light carriers is the tion over energy can be expanded to the range (–∞, +∞)
highest and, therefore, those carriers mainly contrib as the Fermi function (at high energies) and barrier
ute to current. In the following calculations we use the penetrability (at small energies) decrease exponen
band parameters mx = 0.19me, mt = 0.98 me, where tially. It should be noted that the states with negative
me is the free electron mass, Φ0 = Δ/2 = 0.56 eV, where energies are occupied by the valenceband electrons in
Δ = 1.12 eV is the band gap in silicon. graphene, which can contribute to the tunneling cur
Provided the potential distribution and the barrier rent too.
penetrability are known, one can apply the ballistic
formula to obtain the transistor current. To reduce the DISCUSSION OF THE RESULTS
number of equations in the intermediate calculations
we do not treat the tunnel and thermionic currents The calculated dependences of the current density
separately, instead, we set the penetrability to unity at (including both the tunneling and thermionic compo
2 nents) on the drain and gate voltages are presented in
ε – p ⊥ /2m ⊥ > Φ. In this notation the expression for the Figs. 4–7 for the two proposed structures. Prior to the
source–drain current density reads calculations it is reasonable to assume that the FET
with a long tunneling gap in the middle of the channel
∞ p ⊥max
is operated by the gate more efficiently than that with
2eg dε [ f ( ε ) – f ( ε ) ]
( 2π ) –∞ ∫
j =
2 S D ∫ D ( ε, p ⊥ ) dp ⊥ , (11) the gap near the drain.
Indeed, the current density in the FET with the
0
tunnel contact in the middle of the channel falls expo
where g = 4 is the electronic degeneracy factor in nentially at gate voltages lower than the work function
graphene, fS and fD are the Fermi distribution func of the transition of electron from graphene to the
0.6 0.4
0.4
0.2 0.2
0 0
0 1 2 3 4 0 0.2 0.4 0.6 0.8 1.0
Gate voltage VG, V Drain voltage VD, V
Fig. 6. Current vs. gate voltage for a graphene tunnel FET Fig. 7. Current vs. drain voltage for a graphene tunnel FET
with a tunnel contact near the drain. Gap length is L = with a tunnel contact near the drain. Gap length is L =
1 nm, gate dielectric thickness is d = 5 nm, temperature is 1 nm, gate dielectric thickness is d = 5 nm, temperature is
T = 300 K. T = 300 K.
semiconductor (eVG < Φ0). The subthreshold slope at graphene FET. It was already mentioned that a power
d = 1 nm and L = 5 nm (see Fig. 4) equals kT/e; there law dependence of the current on gate voltage was
fore one can conclude that the thermionic current observed when the gate affected only the density of
dominates over the tunnel current. Such ratio between states of tunneling electrons but not the barrier height.
the current components is also typical for Schottky In twodimensional systems with a linear spectrum,
barrier FETs with high barrier penetrability [14]. the tunneling density of states is proportional to the
Compared to the classical silicon MOSFETs on a bulk Fermi energy eϕ0, which, in turn, is a square function
substrate, the variation in the conductivity in our
structure is not only caused by variation in the barrier of the gate voltage.
height but also variation in the Fermi energy eϕ0 in the In structures with a short tunnel gap (L = 1 nm,
graphene contacts. The latter effect is especially pro d = 5 nm for the characteristics in Figs. 6 and 7), the
nounced for thin dielectrics. Substituting VG = 2 V, d = major component of current is the tunnel one, and
1 nm and κ = 5.06 in Eq. (4), we arrive at a Fermi high ONstate currents predicted for the previous
energy of 0.7 eV, which is even higher than the barrier structure cannot be achieved here. Despite the fact
at the graphene–silicon interface. Figuratively speak
that the current is mostly due to tunneling, the sub
ing, at such gate voltages the electrons can be freely
“poured” from source to drain. threshold slope of this FET is still limited by the value
of kT/e. It is known that the subthreshold slope of the
The dependences of the transistor current on the tunnel current itself can possess any value. However, at
drain voltage, exhibiting distinct saturation (Fig. 5), a high subthreshold slope of tunnel current, the tunnel
are similar to those for classical bulk MOSFETs. The current itself is less than the thermionic one, and vice
weak dependence of the saturation current on the versa. A rigorous proof concerning the ultimate sub
drain voltage is due to the tunnel current, as increasing threshold slope of this FET (as well as of any Schot
the drain voltage leads to increasing field strength in tkybarrier FET) is singled out in the appendix.
the gap and, as a consequence, in the tunneling pene
trability as well. The ONstate current of the proposed The values of the current density obtained are just
FET is rather high compared to that of silicon FETs rough estimates. There are several phenomena omit
which is the consequence of the outstanding injection ted in this consideration, which can either increase or
properties of the graphene contacts. The silicon insert decrease the estimated value. First, the spatial charge
does not restrict the maximum FET current as the elec in the tunnel gap can restrict the growth of current at
tron motion in silicon at L = 5 nm is purely ballistic. high gate voltages. Second, the reflection of electrons
The characteristics of a graphene tunnel FET with from the graphene–silicon interface can substantially
a short tunnel gap near the drain exhibit a completely reduce barrier penetrability. Third, the local Fermi
different behavior (Figs. 6 and 7). Some peculiarities energy in the vicinity of the graphene edge may
of the characteristics (nonexponential decrease in cur exceed eϕ0. This fact, omitted in the preceding cal
rent at low drain voltages and lack of saturation) are culations would lead to a higher estimate of current
akin to those reported in [3] for vertical tunnel density.
g'' ( ε 0 ) REFERENCES
∫
≈ exp – g ( ε 0 ) –
2
( ε – ε 0 ) dε (A.2)
2 1. A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. No
voselov, and A. K. Geim, Rev. Mod. Phys. 81, 109
= A ( V G ) exp [ – g ( ε 0 ) ], (2009).
where the function A weakly (nonexponentially) 2. E. McCann and V. I. Fal’ko, Phys. Rev. Lett. 96,
086805 (2006).
depends on the gate voltage.
3. L. Britnell et al., Science 335, 947 (2012).
The saddle point ε0 is determined from the condi 4. N. Kharche and S. K. Nayak, Nano Lett. 11, 5274
tion of the maximum g(ε): (2011).
x2 5. K. Kim et al., Nature 479, 7373 (2011).
1 2d 6. M. Shur, Physics of Semiconductor Devices (Pentice
+
kT dε ∫ 2m [ U ( x, V G ) – ε ] dx = 0. (A.3) Hall, Englewood Clifs, NJ, 1990).
x1 ε = ε0 7. R. Geick, C. H. Perry, and G. Rupprecht, Phys. Rev.
146, 543 (1966).
The subthreshold slope, according to the defini 8. E. O. Kane, J. Appl. Phys. 32, 83 (1961).
tion, is 9. L. V. Keldysh, Sov. Phys. JETP 6, 33 (1958).
d ln j sub dg ( ε 0 ) 10. A. Schenk, Solid State Electron. 36, 1 (1993).
≈ –
11. L. F. Mao, J. L. Wei, Ch. H. Tan, and M. Zh. Xu, Solid
dV G dV G
State Commun. 114, 383 (2000).
⎧ x2 ⎫ 12. J. Shannon and K. Nieuwestee, Appl. Phys. Lett. 62,
2 d ⎪ ⎪ 1815 (1993).
dV G ⎪ ∫
≈ – ⎨ 2m [ U ( x, V G ) – ε ] dx ⎬
⎪ (A.4) 13. S. Xiong, T. King, and J. Bokor, IEEE Trans. Electron.
Dev. 52, 8 (2005).
⎩ x1 ⎭ 14. R. A. Vega, IEEE Trans. Electron. Dev. 53, 7 (2006).
x2 15. Q. Zhang, T. Fang, H. Xing, A. Seabaugh, and D. Jena,
= – 2
d { 2m [ U ( x, V ) – ε ] }
dU dx. IEEE Electron. Dev. Lett. 54, 10 (2008).
dU ∫ G
dV G 16. A. C. Seabaugh and Q. Zhang, Proc. IEEE 98, 12
x1 (2010).