This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON ELECTRON DEVICES 1

A Compact Current–Voltage Model for

2-D-Semiconductor-Based Lateral
Homo-/Hetero-Junction
Tunnel-FETs
Arnab Pal , Graduate Student Member, IEEE, Wei Cao , Member, IEEE,
and Kaustav Banerjee , Fellow, IEEE

Abstract — A fully analytical surface potential and dimensional scaling, which is not feasible in nanoscale
current–voltage model is presented for the first time MOSFETs because of their Boltzmann-limited minimum sub-
for both lateral homojunction (HMJ) and heterojunction threshold swing (SS) of 60 mV/decade at room temperature,
(HTJ) tunneling-field-effect transistors (TFETs) based on
2-D semiconducting channel materials. The dynamic gate- and constitutes a major concern for their scalability into
modulated electrostatic potential at the source/channel the sub-10 nm regime [1], [2]. Among the several sub-
tunneling junction is suitably captured by solving a kT/q switches proposed to overcome this fundamental bar-
quasi-2-D Poisson’s equation in both source and chan- rier, tunneling field-effect transistors (TFETs) have emerged
nel. Subsequently, the band-to-band tunneling current is as the most promising devices [3], [4]. However, TFETs
accurately derived starting from the Landauer’s equation
by integrating over all possible carrier energies (or wave- fabricated from conventional 3-D materials like Si, Ge, and
vectors) over which tunneling is possible. The model III–V compounds, exhibit either low ON-current, or a steep SS
employs Fermi–Dirac statistics in both the degenerate (i.e., SS < 60 mV/decade) only at very low current values,
source and drain to compute the surface potential and because of their nonoptimal electrostatics and presence of
net current, which yields more physical results than the interface traps [5]–[7]. These nonidealities that are detrimental
commonly employed Boltzmann statistics. Its use in Lan-
dauer’s approach for evaluating the net ON-current leads to TFET performance can be significantly alleviated by using
to an analytical model of the TFET, which physically guar- 2-D-materials, which, owing to their ultrathin body, relatively
antees zero drain current at zero drain–source bias. Input large bandgap, and pristine interfaces [8]–[10], provide excel-
and output characteristics for both HMJ and HTJ TFETs are lent electrostatics and lead to low OFF-current and steep SS,
computed and compared against rigorous nonequilibrium as theoretically suggested in [11]–[13]. Moreover, the strong
Green’s function (NEGF) simulations for different device
parameters to prove the veracity of the model, and the suppression of the density of states (DOS) into the bandgap
match has been found to be excellent up to ultrashort of any 2-D material, commonly known as the band-tail,
channel length of 5 nm. results in very steep SS for various 2-D–2-D and 3-D–2-D
Index Terms — 2-D materials, 2-D semiconductors, band- source/channel heterostructures [14], as was experimentally
to-band tunneling, compact modeling, heterojunction (HTJ), demonstrated in [15] using a degenerately doped 3-D (Ge)
low-power, nonequilibrium Green’s function (NEGF), quan- source and 2-D (MoS2 ) channel vertical heterojunction (HTJ)
tum device, steep-slope transistor, tunneling field-effect TFET, where a minimum SS of 3.9 mV/decade and an average
transistor (TFET), van der Waals heterostructures. SS of ∼31 mV/decade over four orders of the drain current
was observed for the first time albeit with very low ON-current,
I. I NTRODUCTION
which limits its applicability in practical circuits. Among the

S TEEP-SLOPE (or sub-kT/q) transistors are desirable for

simultaneous supply-voltage or power scaling along with
several novel solutions proposed to overcome the typical low
ON -currents in TFETs [16]–[18], 2-D semiconductor-based
Manuscript received June 3, 2020; revised July 8, 2020; accepted lateral HTJ TFETs introduced in [16] are the most attractive
July 16, 2020. This work was supported in part by the Japan Science because their tunneling barrier heights can be made small
and Technology Agency (JST) Core Research for Evolutional Science by appropriately selecting the band overlap of source/channel
and Technology (CREST) Program under Grant SB180064, in part by
the Army Research Office (ARO) under Grant W911NF1810366, and materials and due to their atomic-scale tunneling barrier
in part by Intel Corporation. The review of this article was arranged by widths, which ultimately determines the ON-current. First
Editor E. Gnani. (Corresponding author: Kaustav Banerjee.) principle transport studies in [16] have shown that the ON-
The authors are with the Nanoelectronics Research Laboratory,
Department of Electrical and Computer Engineering, University of current of 2-D lateral HTJ TFETs can be substantially higher
California, Santa Barbara, CA 93106 USA (e-mail: arnab@ucsb.edu; than that for the 2-D lateral homojunction (HMJ) TFET or the
weicao@ucsb.edu; kaustav@ece.ucsb.edu). 2-D-channel based vertical HTJ TFETs. Moreover, 2-D lateral-
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org. HTJ TFETs can achieve the smallest SSmin values compared
Digital Object Identifier 10.1109/TED.2020.3011350 to any other 2-D-channel based TFETs [14]. Hence, the wide

0018-9383 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2 IEEE TRANSACTIONS ON ELECTRON DEVICES

gamut of 2-D materials available provides additional moti-

vation to pursue 2-D source-channel lateral-HTJ TFETs for
future low-power/energy-efficient electronics. Also, recent
experimental breakthroughs on the fabrication of lateral HTJs
of 2-D semiconductors [19] and doping of 2-D materials [20]
has made fabrication of 2-D lateral-HTJ TFETs feasible.
In general, 2-D TFETs, owing to their ultralow leakage
and superior performance, could be ideal for building next-
generation ultralow-power electronics such as brain-inspired
neuromorphic circuits [21]. 2-D-TFETs can also be employed
for building a revolutionary new class of bio/gas sensors [8]
due to their combined advantages of high sensitivity arising
from their steep SS [22] and the atomically thin and pristine
channel of the 2-D semiconductors with sizable bandgaps [8].
Hence, it is opportune to examine the prospects of 2-D Fig. 1. (a) Cross-sectional view of the DG 2-D n-TFET. The source is
lateral-HTJ TFETs for exploring various circuits, including strongly p-doped, the channel intrinsic or slightly n-doped, and the drain
strongly n-type doped. (b) Cross section of the device under operation
low-power sensors, neuromorphic computing circuits, and at a certain bias showing the fringing electric field lines from the gate
other low-power integrated circuits. However, to study the terminating in the source, the fully depleted channel, and the profile of the
behavior of these devices in circuits, we need to develop their source depletion region. The channel length of the device is L, L1 refers
to the source depletion length, t2D is the thickness of the 2-D channel,
compact models for accurate and fast simulations and carry while tox is the gate oxide thickness. The source, channel, and drain
out various design optimizations. There have been studies on dopings are denoted by Ns , Nch , and Nd , respectively. The source is
the modeling of both HMJ and HTJ TFETs fabricated on bulk grounded, and a voltage of VGS and VDS are applied to the gate and
drain terminal, respectively. The direction of transport is along the x-axis,
semiconductors [23]–[33], but only a few based on 2-D mate- y-axis represents the confined direction along the two gates, and z-axis
rials [34]–[36]. Moreover, most of these models [24], [28], is along the direction of the transistor width.
[29], [32], [34], [35] are numerical, or do not model the
source and drain Fermi degeneracy [23]–[27], [31]–[33], [36], in Fig. 1(b) where the source depletion region, the fringing
or the drain bias dependence [24], [29], [32], [36] and electric-field lines, and the device dimensions are shown.
cannot physically model zero drain current at zero drain– As shown in Fig. 1(a), the source, channel and drain are
source bias [18], [23]–[27], [29], [31], [32]. Essentially, made of 2-D semiconductor material. The lateral source-
the incorporation of Kane’s band-to-band tunneling model channel junction is either an HMJ or an HTJ, while the
in [18], [24]–[27], [29], [31], [32], which predicts a nonzero channel-drain junction is an HMJ. In this work, we discuss the
tunneling carrier generation rate even at zero drain–source bias behavior of an n-TFET, and therefore, the source is heavily
(due to presence of a nonzero electric field at source-channel doped p-type and the drain is heavily doped n-type [4]; while
junction), and an implicit assumption of fully occupied valence the channel is assumed to be intrinsic, or slightly n-doped.
and fully empty conduction band [37] is what makes physi- The requirement on the degenerate source and drain doping
cally modeling zero drain current at zero drain bias unfeasible. necessitate the consideration of Fermi–Dirac (FD) statistics to
Nevertheless, the inclusion of all the above-mentioned effects correctly account for the carrier distribution, over the more
is critical in designing an accurate compact model, which will widely used Boltzmann distribution [see Fig. 2(a)]. Applica-
be useful for designing various circuits and predicting the tion of a positive gate-bias lowers the energy bands in the
performance of the 2-D TFETs in circuit operation. At present, channel, which under sufficient applied bias causes the channel
a physically consistent, fully analytical model, which accounts conduction band to overlap with that of the source valence
for both the drain and gate bias dependence of drain current for band, allowing an appreciable number of carriers to tunnel,
both lateral- HMJ and HTJ TFETs, fabricated on 2-D materials thereby constituting an ON-current [see Fig. 2(b)]. The use of
is lacking, and that is what this article intends to address. an HTJ reduces the gate bias required to achieve this overlap
This article, therefore, starts by describing the device struc- of bands, thereby delivering more current at a particular gate
ture and the basic operation in detail (Section II), devel- bias. For designing an effective source-channel HTJ, both the
oping its analytical surface potential model by solving the conduction and valence bands in the source should be higher
pseudo-2-D Poisson’s equation in both source and channel than the corresponding ones in the channel, as indicated in
(Section III) to extract the energy band diagram, which is then Fig. 2(b). However, too big an offset (broken gap – top sketch
applied to the analytically developed band-to-band tunneling in [see Fig. 2(c)] will cause tunneling to take place even at
current model (Section IV) to yield the drain–source current. negative gate-biases, and not form a good TFET. Therefore,
Finally, conclusions are drawn in Section V. in this work, we only consider staggered HTJs [bottom sketch
in Fig. 2(c)].
II. D EVICE S TRUCTURE
A generic 2-D-semiconductor-channel double-gate (DG) III. S URFACE P OTENTIAL M ODEL
n-type TFET device under consideration is shown in Fig. 1(a), In this section we derive the surface potential model for
and the device cross section under a certain bias is shown the device. For a moderate gate–source voltage applied, it is

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PAL et al.: COMPACT CURRENT–VOLTAGE MODEL FOR 2-D-SEMICONDUCTOR 3

where d1 and d2 are constants of integration, L 1 is the source

depletion length (to be evaluated), VFBS = [ϕ m − (χs +
E GS /2)]/q is the flat-band voltage of the source with respect
to gate (χs and E GS are the electron affinity and bandgap of
the source, respectively, and ϕm is the metal work-function),
Cox = εox /tox (εox is the permittivity of the gate insulator)
√ the insulator capacitance per unit area, and β1 =
is
(4Cox )/(πεs t2D ).
Similarly, with the source intrinsic Fermi level as the
Fig. 2. (a) Comparison of FD and Boltzmann distribution as a function reference, assuming full depletion of the channel, the 2-D
of energy, showing the inaccuracy of the latter in estimating carrier distri-
bution (top) at energies E < EF , the simulated source depletion length of
Poisson’s equation in the channel can be written as
a lateral 2-D-TFET as a function of the applied VGS (bottom). (b) Sketch d 2 ϕCH (x, y) d 2 ϕCH (x, y) q Nch
of the energy bands [conduction band (EC ), valence band (EV )], source + = (3)
(EFS ) and drain (EFD ) Fermi levels showing an enhanced source deple- dx 2 dy 2 εch
tion length (L1 ) and the OFF- and ON-state of the device under the
application of a certain VGS . The electron distribution is shown in the where ϕCH (x, y) is the potential in the channel, εch is the
p+ -doped source showing how the FD tail filtering gives rise to sub- permittivity of channel semiconductor material, and Nch is the
60 SS. Also shown are the forward (flowing from drain to source) and small n-type channel doping. In case the channel is p-doped,
reverse (flowing from source–drain) tunneling currents, the magnitude of
which are determined by the position of the source and drain Fermi levels, we must use a doping of −Nch .
respectively. (c) Schematic of the broken and staggered band-alignments The solution of (3) under the same boundary conditions as
at a HTJ showing band-offsets of ΔEC and ΔEV . assumed in (1), but with a gate oxide thickness of tox , yields
the following expression for the channel surface potential (ϕch )
reasonable to assume that the entire ultrathin channel and
evaluated at y = ±t2D /2:
only a small part of the source (Fig. 2(a), because of its high
doping) is fully depleted. Under such a condition, the cross- ϕch = c1 sinh(βx) + c2 sinh[β(L − x)]
sectional view of the device is illustrated in Fig. 1(b), where + VGS − VFBCH + q Nch t2D /(2Cox ) (4)
we show the source depletion region (Region I), the fully
depleted channel (Region II), and the drain (Region III). The where c1 and c2 are constants of integration, VFBCH = [ϕm −
extreme thinness of the 2-D body causes a negligible voltage (χch + E GCH /2) + E C + E G /2]/q is the flat-band voltage
to drop across it, and therefore, the source depletion profile of the channel with respect to gate (χch is the electron affinity
can be assumed to be uniform, up to a length of L 1 . Also, of the channel material, E GCH is the bandgap and E C is the
note that although under the application of a strong positive conduction band offset of the channel with respect to source),
gate bias, the part of the channel closer to the drain can be E G = E GCH − E GS is the√ difference of the channel and
accumulated with electrons [23], we do not consider this in our source bandgaps, and β = (2Cox )/(εch t2D ). Note that for
model because 2-D TFETs are meant for low power operation, simplicity in our model, we have considered the conduction
and therefore, the channel can be assumed to be depleted for band offset to be equal to the difference of the electron
all practical gate biases. affinities of the channel and source (χ s ), i.e., E C = χch −χs .
The valence band offset (E V ) of the channel with respect to
A. Solution for the Surface Potential the source is therefore, given by: E V = E G + E C .
The surface potential solution is obtained by solving the The complete solution of the surface potential in the device
pseudo-2-D Poisson’s equation in both the source and the requires the applications of suitable boundary conditions,
channel and using suitable boundary conditions in Fig. 1(b). which, when applied to (2) and (4) yield the values for the
The Poisson’s equation modeling the potential in the depleted constants of integration and the source depletion length. With
source can be written as the intrinsic Fermi level of the source as the reference, we can
d 2 ϕS (x, y) d 2 ϕS (x, y) q Ns write the following boundary condition (BCsource ) at the end
+ =− (1)
dx 2 dy 2 εs of the source depletion region (x = −L 1 ):
where ϕS (x, y) is the source potential, q is the electronic ϕs|x=−L 1 = −[E GS /(2q) + (E V − E FS )/q]. (5)
charge, εs is the permittivity of the source material, and Ns
is the source doping. The solution of (1) is achieved using Similarly, at the channel-drain junction (x = L), neglecting
the parabolic approximation method [32] with the intrinsic drain depletion (because ON-current is independent of drain
Fermi level of the source as the reference. Gauss’s law is depletion length), we can write the following BC (BCdrain ):
applied to the top and bottom surfaces, while equating the E GS (E FD − E C ) E C
vertical electric field to zero at y = 0 (because of symmetry). ϕch|x=L = + + VDS − . (6)
2q q q
In addition, a gate-insulator effective thickness of πtox /2 is
The other two boundary conditions are obtained by equating
assumed (because of fringing electric field) [18], which leads
the surface potential and the lateral displacement field between
to the following solution of the surface potential in the source
ϕs and ϕch at the source-channel junction (x = 0).
(ϕs ), evaluated at y = ±t2D /2:
Since the source and drain doping in TFET are generally
ϕs = d1 sinh(β1 x) + d2 sinh[β1 (x + L 1 )] degenerate to allow for greater ON-current and low contact
+ VGS − VFBS − qπ Ns t2D /(4Cox ) (2) resistance, it necessitates the use of FD statistics [Fig. 2(a)] to

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4 IEEE TRANSACTIONS ON ELECTRON DEVICES

evaluate the correct boundary conditions. For example, with a

p-type source doping of Ns , we can write
 EV
Ns = DOS2D [1 − f (E)]d E (7)
−∛

where DOS2D = g1 m ∗ /(π2 ) is the 2-D DOS, g1 is the valley

√
degeneracy [38], m ∗ = m xvs m zv s is the DOS effective mass
of holes (m xvs and m zvs are effective masses of holes in source
valence band along x- and z-direction, respectively),  is
the reduced Planck’s constant, f (E) is the Fermi occupation Fig. 3. Comparison of our surface potential model against NEGF
probability for electrons, and E V is the maxima of the valence simulations. (a) Lateral HTJ DG-2-D-TFET has been simulated with
Nch = 5 × 103 cm−2 , Nd = 2.5 × 1013 cm−2 , VDS = 0.3 V, L = 5
band. nm, t2D = 1 nm, tox (SiO2 dielectric) = 1 nm, ϕm = 4.5 eV and variable
Using (7) we can find out (E V − E FS ) to compute (5). Ns , and VGS . WTe2 is the source material, and MoS2 forms the channel
Similarly, by applying Fermi statistics to the n + drain, we can and drain materials. (b) Lateral HMJ 2-D-TFET simulation with MoS2 as
the source, channel, and drain material. All other structural parameters
obtain (E FD − E C ) to compute (6). Finally, applying all the are same as above, except L = 25 nm and Ns = Nd = 5 × 1013 cm−2 ,
BCs and after suitable mathematical substitutions, we arrive ϕm = 4.96 eV, and two different VDS .
at the following expression from which L 1 can be computed
analytically:
⎧ ⎫
⎪ ε s d 1 β1   ⎪
⎨ c1 − 1 − cosh2 (β1 L 1 ) ⎬
tanh(β L) ε ch β
⎩ +(VFBS − VFBCH ) + qt2D (Nch + Ns π/2) ⎪
⎪ ⎭
 2C
ox
q Ns t2D π
= ϕs|x=−L 1 + −VGS +VFBS cosh(β1 L 1 ). (8)
4Cox
Note that c1 is obtained by evaluating (4) at x = L and
equating it to (6), and d1 is obtained by evaluating (2) at
x = −L 1 and equating it to (5). Once L 1 is obtained from (8), Fig. 4. (a) Simplified tunneling barrier where the energy bands, Fermi
all other constants of integration in (2) and (4) are obtained level in the p-(EFp ) and n-(EFn ) regions, average tunneling electric field
to yield the complete solution of the surface potential in the (ξ) across the depletion region, and the net energy overlap of the bands
(qV R ) are shown. ξ is calculated by the ratio of the electrostatic band-
entire device. bending to the total tunneling distance of the carriers. (b) Diagram
To compare the validity of our surface potential model illustrating the reduced k-space (shaded) of electron wavevectors in
we compare our model results against those obtained from the source that contribute to the tunneling current. The limits of the k
wave-vector are obtained as noted in (9) and (10). The maximum value
first principle 2-D numerical NEGF simulations [16]. We plot of kx is kmax (9), which denotes the maximum electron wave-vector
the obtained energy bands for both HMJ and HTJ TFETs corresponding to an energy of qVR , while the maximum value of k⊥ ,
as a function of lateral distance for different values of the the electron wave-vector along z-direction, is kmax /η. Also, since only
positive electron wave-vectors along the tunneling direction contribute to
VGS , and list all device specifications, along with the model tunneling, kx has only positive values. mr⊥ is the reduced perpendicular
comparisons in Fig. 3. The parameters for the energy bands effective mass and has been defined in (10).
and the associated effective masses for both WTe2 (source) and
MoS2 (channel) used for the simulations are: χs = 3.77 eV,
χch = 4.5 eV, E GS = 1.33 eV, E GCH = 1.6 eV, for
E C = 0.73 eV and E V = 1.0 eV. The effective mass (surface potential dictated by gate electrode) except at the
of electrons and holes in WTe2 along x- (transport) and source and drain junctions, however, for shorter channel length
z-directions are 0.32m 0 and 0.42m 0 in both the conduction transistors (∼5-nm channel length) the short channel effects
(m xcs = m zcs = m cs ) and valence bands (m xvs = m zvs = degrade the gate control over the channel potential as seen
m vs ), respectively, where m 0 is the mass of an electron. For from Fig. 3(a). Although the effect of the drain depletion
MoS2 , the effective mass of electrons in the conduction band region becomes more apparent near the drain in these short
along x- (m xcch ) and z- (m zcch ) are 0.5788m 0 and 0.5664m 0, channel transistors, however, as shown later in Fig. 7, its
respectively, while for the holes in the valence band they are effect on the ON-current is still negligible because the source-
equal at (m xvch = m zvch = m vch ) 0.66m 0. channel tunneling electric field is still well accounted for. Also,
From Fig. 3 it can be observed that the match of our results the negative gate voltages in the simulations are because of the
against NEGF simulations is very good for both HMJ and choice of the gate metal work-function. The position of the
HTJ TFETs for the simulated values of VGS and VDS , channel Fermi level in the source and drain regions relative to E V and
length, and source doping. Note that the doping concentra- E C , respectively, highlights the importance of incorporating
tions have been normalized to a monolayer (∼0.5-nm thick) FD statistics in our model. The small mismatch between the
2-D-body. Because of the thin body and gate-dielectric, model and simulated energy bands near the drain is because
the electrostatic control over the channel potential is excellent of the presence of the intrinsic contact resistance present in
for long channel transistors leading to almost flat energy bands any NEGF simulation.

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PAL et al.: COMPACT CURRENT–VOLTAGE MODEL FOR 2-D-SEMICONDUCTOR 5

After establishing our model and confirming its validity where, kmax2
= 2m vs q VR /2 denotes the maximum tunneling
against NEGF simulations, we move on to Section IV where electron wave-vector, and 1/m r⊥ = (1/m zvs + 1/m zcch ) is the
we model the drain current. reduced perpendicular effective mass. Therefore, the limits of
k1x is kmax , and that of k⊥ is kmax /η (defined in Fig. 4), which
IV. D RAIN C URRENT M ODEL corresponds to an ellipsoidal region of the momentum space
where the tunneling takes place [40]. However, as we will
In this section we derive the analytical tunneling current
show later, since we are only considering electron tunneling
model for 2-D lateral-TFETs starting from Landauer’s equa-
from source to channel, k1x has only positive values, and
tion [39]. Both the forward current flowing from the drain to
therefore, the lower limit of k1x in (9) is 0. Also, since we are
the source and the reverse current flowing from the source to
not considering any direct source to drain tunneling, the ranges
the drain are computed to evaluate the net ON-current, which
of electron wave-vectors derived in (9, 10) are for electrons
is given by the difference of these two. Moreover, as already
tunneling from source to channel only.
stated, the doping concentration in both source and drain are
Applying Wentzel–Kramer–Brillouin (WKB) approxima-
high in TFETs, thereby making it imperative that we take the
tion [41], we can calculate the effect of E ⊥ on the tunneling
FD distribution of the carriers into account. This introduces a
probability of carriers T (E ⊥ ), as
dependence of doping into the model and helps in achieving    
zero drain current at zero drain–source bias when used in 3/2 
T (E ⊥ ) = exp −4 2m r∗ E GS /(3qξ ) exp −E ⊥ / Ē (11)
Landauer’s model for evaluating the current. This is because

the source and the drain Fermi levels align at zero drain–source where Ē = qξ/ 8m r∗ E GS and m r∗ = (1/m xvs + 1/m xcch )−1
bias leading to symmetric source and drain carrier distribution, is the reduced effective mass, which effectively accounts for
and hence, equal forward and reverse current. the electron-hole duality during band-to-band tunneling. Note
that E GS in (11) is the net energy barrier that the tunneling
A. Ranges of Electron Wave-Vector for Tunneling electrons have to surmount, and corresponds to the bandgap
and Tunneling Probability of the material where the tunneling commences from, and is
therefore, of the source.
Fig. 4 shows a simplistic diagram of an electron tunneling
from the valence band of a p+ -material to the conduction
band of an n+ -material, across a generic bandgap of E G . Also B. Extraction of the Tunneling Electric
shown is the difference in the curvatures of the respective Field and the Band Overlap
energy bands arising because of the difference in the effective To evaluate the tunneling probability (11) and the maxi-
mass of the carriers. Note that although this represents a mum allowable electron wave-vectors (9), (10), we need to
generic p-n-junction tunneling diode, it must be realized that extract the tunneling electric field (ξ ) and the band energy
the physics of tunneling across the energy bands in this device overlap (qVR ). These two parameters are obtained from the
is similar to that of an n-TFET which has a p+ source and an solution of the surface potential obtained in Section III.
n-channel, where the carrier concentration of the channel is Since the reference of the surface potential solution is taken
varied by the application of VGS . The applied VGS , therefore, to be that of the intrinsic Fermi level of the source and
modulates the energy overlap (qVR ) and the tunneling electric is at a potential of ϕs (x = −L 1 ), therefore, the tunneling
field (ξ ) in the device, hence, affecting the ON-current. electrons will reach the channel conduction band where the
Since the energy of an electron (E) is related to its potential would be E GS /q more than ϕs (x = −L 1 ). This
momentum wave-vector (k) and its effective mass (m ∗ ) as ensures that the total energy separation between the bands
E = 2 k 2 /2m ∗ , hence, any change of m ∗ during tunneling becomes equal to the source band gap energy E GS . This
(isoenergetic process) from one band to another changes its net energy difference is due to both the electrostatic energy
k vector. This is manifested during tunneling from the source difference and the energy difference due to the band offset.
valence band to the channel conduction band of the TFET, In an HTJ therefore, the electrostatic energy difference must
whose difference in curvatures, along with the conservation of be E GS −E C . Denoting the x-coordinate in the channel where
the perpendicular electron wave-vector (along the z-axis) while the tunneling terminates by L 2 , L 2 is obtained by equating (4)
tunneling, causes a change in the lateral wave-vector (along to (E GS − E C )/q + ϕs (x = −L 1 ). For an HMJ, however,
x-axis) of the electron. Assuming the perpendicular electron E C = 0, and results in larger L 2 than that in an HTJ.
wave-vector to be k⊥ , the lateral wave-vector in the source Once L 2 is obtained, an average electrostatic tunneling
valence band to be k1x and that in the channel conduction electric field is extracted, which is assumed to remain constant
band to be k2x , we obtain the following limits of the wave- across the entire tunneling distance for simplicity. For HTJs,
vectors [23], [40] over which tunneling occurs. this tunneling electric field is given by
For k1x
  E GS − E C
ξ= . (12)
m v k 2
mv k2 q(L 1 + L 2 )
− kmax
2 − s ⊥ ≤ k1x ≤ kmax 2 − s ⊥. (9)
m r⊥ m r⊥ For HMJs E C = 0 in (12). This electric field is then used
for all subsequent drain current calculations.
For k⊥
  The screening length (λ), which is a measure of the decay
−kmax m r⊥ /m vs ≤ k ⊥ ≤ kmax m r⊥ /m vs (10) of the surface potential [42], in the center of the channel of

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6 IEEE TRANSACTIONS ON ELECTRON DEVICES

Fig. 5. (a) Band diagram of a TFET during ON-condition showing

an energy overlap of qVR (green shaded region) between the channel
conduction and source valence bands, and the point where the channel Fig. 6. (a) Diagram illustrating the approximation of (16) into two separate
conduction bands cross the source valence bands (x = L2 ). The overlap equations under different biases of operation. When VGS < VGSwitch the
of the energy bands indicated by a red shaded region (around 5λ from the factor A is lower than log(4), and can be approximated as an exponential
drain) does not contribute to the tunneling current because of its large function, while for higher gate biases it can be approximated as a linear
tunneling distance from the source energy bands, thereby resulting in function. The gate voltage around which this switch happens is denoted
low tunneling probability. The center surface potential (ϕch ) therefore, is in (17). (b) Comparison of our analytical drain current model (solid line)
obtained by averaging out the channel surface potential from x = (L −5λ) against numerical simulations (symbol) for both WTe2 -MoS2 DG-HTJ and
to x = L2 . (b) BZ of a general hexagonal lattice showing symmetry MoS2 -MoS2 DG-HMJ with Ns = 5 × 1013 cm−2 , Nd = 2.5 × 1013 cm−2 ,
points K , M , and Γ (on xz -plane), and vertically on top of them (along Nch = 0, VDS = 0.3 V, L = 25 nm, t2D = 1 nm, tox (SiO2 dielectric) = 1 nm,
ky -direction) are the H , L, and A points (in a coordinate system cor- and variable VGS . Metal with a work-function of 4.5 eV is the gate
responding to Fig. 1(b)). (c) Material choices (top); WTe2 (purple) and electrode.
MoS2 (green); schematic of WTe2 -MoS2 HTJ and (bottom); MoS2 -MoS2
HMJ.
Fermi voltage (E FS /q) to the drain Fermi voltage (E FD /q) by:
any DG field-effect transistor
√ (FET) can be reduced to that (E FD /q) = (E FS /q) + VDS to account for applied VDS . Please
of at its surface, given by tox t2D (εch /2εox ), when the body note that (14) implicitly assumes physics of 2-D materials
thickness is very small. Since the net potential range over because of the integration over 2-D DOS, and therefore, is not
which the tunneling takes place (VR ) is defined by the useful applicable for conventional bulk semiconductors. Moreover,
overlap of energy bands [see Fig. 5(a)], it was found from monolayer 2-D materials have similar lattice constants and
extensive simulations that it is best approximated (for greater are generally direct bandgap semiconductors with the valence
drain current saturation) by subtracting the potential at x = L 2 band maxima/conduction band minima occurring around the
from the channel potential averaged out between x = L 2 to K -point [see Fig. 5(b)]. This implies that the tunneling current
x = L − 5λ. Therefore, VR is model for both lateral-2-D based homo- and hetero-junction
 L−5λ TFETs [see Fig. 5(c)] do not need consideration of phonon-
1
VR = ϕch d x − [ϕch ]x=L 2 (13) assisted tunneling physics. For vertical 2-D-TFETs, however,
L − 5λ − L 2 L 2
the K -point in the Brillouin zone (BZ) of one layer can overlap
where ϕch is the channel surface potential and is obtained with either the K or K  point of the second layer, based on
from (4). A A or AB stacking, respectively, and hence, tunneling can
Note that when L 2 > (L − 5λ), we average the channel either be direct or through the assistance of phonons.
potential over the entire channel length. Once VR and ξ are Substituting T (E ⊥ ) from (11) into (14) and segregating
obtained, the tunneling probability and the ranges of electron terms, we can write the following:
   
3/2  k⊥ max 2 
wave-vector over which tunneling takes place (9), (10) are
evaluated to yield the total drain current. q 4 2m r∗ E GS 2 k ⊥
J = exp − exp − dk⊥
2π 2 m vs 3qξ  k⊥ min 2m vs E
 k1 max
C. Analytical Model of Drain Current k1x
×   dk1x (15)
−E 2 k 2 2 k 2
The 1-D current density (A/m) can be written as [39] 0 1+exp E V
kT
FS
− 2m v ⊥kT − 2m v 1xkT
 s s

d 2k where k⊥min and k⊥max are the minimum and maximum ranges
J = 2q v(E)T (E ⊥ )( fsv − f dc ) (14)
(2π)2 of (10), and k1max is the maximum value of (9) for a certain k⊥ .
where v(E) = (1/)(d E/dk) is the velocity of electrons, Since k1x is the electron wave-vector along tunneling direction
f sv = 1/[1+exp{−(E 1x + E ⊥ −(E V − E FS ))/kT }] is the Fermi in the source, hence, the integration in (15) is only carried out
occupation of electrons in the source valence band (E 1x and for positive values of k1x .
E ⊥ are energies corresponding to wave-vectors k1x and k⊥ , The finite integration over k1x in (15) is evaluated as
  
respectively), f dc is the Fermi occupation of electrons in the 1 1 + exp ak12 max + b
drain conduction band. For brevity, we show the derivation of log (16)
2a 1 + exp(b)
the forward drain–source tunneling current (limited by fsv )
in this section. The derivation of the reverse source–drain where a = 2 /(2m vs kT ) and b = 2 k⊥
2
/(2m vs kT ) −
current is similar, except for a small modification of the source (E V − E FS )/kT .

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PAL et al.: COMPACT CURRENT–VOLTAGE MODEL FOR 2-D-SEMICONDUCTOR 7

To obtain an analytical model of tunneling current, (16) link these two piecewise models seamlessly. This is enabled
must be simplified, as both k1max (9) and b (16) are functions of by using a tan-hyperbolic smoothing function (S)
k⊥ , which will subsequently be integrated over in (15). As will
S = 0.5 + 0.5 tanh[28(VGS − VGSwitch )] (20)
be shown later, the drain current for the term log[1+exp(b)] in
(16) is negligibly small compared to the other terms primarily which switches from a minimum value of 0 to a maximum
due to the absence of the maximum tunneling wave-vector value of 1 around VGS = VGSwitch . The factor of 28 determines
(k1max ) and can be safely ignored for all practical purposes. the steepness of this switch and has been found to give the best
Also, plotted in Fig. 6(a) is the maximum value of the function match with numerical simulation results. The final expression
(ak1max
2
+ b) = A as a function of VGS for a certain device of the forward tunneling (J f ) current flowing from the drain
configuration, and under certain bias. As can be observed from to the source can therefore, be expressed, using (18)–(20) to
Fig. 6(a), as VGS is swept from a negative to a positive bias, yield
A changes from a negative value to a positive value around
J f = J1 [1 − S] + J2 S. (21)
a certain VGS . When exp(A) <∼ 4, the logarithmic function
in (16) can be approximated as an exponential function [see Similarly, the reverse current (J r ) flowing from the source to
Fig. 6(a)], while for higher values, it is best approximated the drain can be calculated by the simple substitution of the
as a linear function [see Fig. 6(a)]. The gate bias where this source Fermi level with the drain Fermi level in (15)–(17) to
demarcation lies is denoted by VGSwitch and is obtained by compute (18)–(20) and subsequently, (21). Therefore, the net
2
equating exp(ak1max + b) to 4 in (16) to obtain tunneling current density (J ) flowing from the drain to the
  source of the TFET is given as
E GS − E C
VGSwitch = ϕs|x=−L 1 +
q J = J f − Jr . (22)
m vs q Nch t2D
+ [1.6kT + E FD ] − To ascertain the validity of our approximations and sim-
qm xcch 2Cox plifications made in the derivation of the analytical model,
1 we compare and show the excellent match of our analytical
+ [c2 cosh{β(L − x)}
β(L − L 2 ) model results against those obtained by numerically integrating
− c1 cosh{βx}] LL 2 + VFBCH . (17) (15) in Fig. 6(b), where the schematic of the TFETs is shown
in Fig. 5(c). As observed from Fig. 6(b), our model results
When VGS < V GSwitch , the logarithmic function in (16) compare exceedingly well against the numerical simulations,
can be approximated solely by an exponential function of thereby confirming the validity of all the simplifications,
(ak1max
2
+ b) to yield the following expression for current: the neglect of the log[1+exp(b)] term in (16), and the smooth-
   
qkT  ness of the piecewise drain current model. Also observed in
3/2
  2πm vs  4 2m r∗ E GS
J1 =  2 exp − the figure is the higher ON-current of the HTJ TFET over
42 π 2 η −1
+ 1 3qξ 
kT E HMJ TFET for the entire range of the gate bias, bringing out
  the importance of using a suitable source-channel junction for
m xcch q V R (E V − E FS )
× exp − improving the electrical characteristics.
m vs kT kT
   k⊥ max A constant leakage current of 10−20 A/μm is assumed to be
 2 
η −1 1   flowing in the device from the drain to source, except at zero
× erf k⊥  2 + / 2m vs . (18) VDS where the net current in the device should be zero. This is
kT E k⊥ min accomplished by modeling the net OFF-current per unit width
When VGS > V GSwitch , the logarithmic function in (16) is (W ) of the device as a function of VDS as
best approximated as a linear function, and after suitable math-
IOFF = 10−14 tanh[30VDS ] (23)
ematical manipulations, we obtain the following expression for
the current density: where the tan hyperbolic function ensures continuity of the
  3/2
 IDS − VDS characteristics.
qkT 4 2m r∗ E GS The results obtained from our model have been compared
J2 = exp −
2π 2 3qξ  against those obtained from NEGF simulations [16] in Fig. 7,
⎡  √  k⊥ max ⎤ where the transfer characteristics (IDS − VGS ) of both HMJ
m xcch q VR (E V −E FS ) 2m vs Eπ k
⎢ m vs kT − kT erf √ ⊥ ⎥ (MoS2 -MoS2 ) and HTJ (WTe2 -MoS2 ) DG-2-D-TFETs (note
⎢ 2 2m vs E
k⊥ min ⎥
⎢ ⎡   ⎤ ⎥ that DIBL, important for short channel technology nodes, has
×⎢ ⎥.
√ k ⊥ max
⎢ 2 η2 −1 ⎢ π k⊥ ⎥
erf √ ⎥
already been accounted for in the surface potential model)
⎢− ( ) 4(2 /2m vs E )3/2 ⎥
⎣ 2m vs kT ⎣ k m E 
 k⊥
2 2

2m vs E
⎦ ⎦ have been plotted and compared against. The reasonably good
− 2 exp − 2m E
⊥ vs
match of the results against NEGF simulations for a variety
vs k⊥ min
of device parameters, such as channel length, gate dielectric
(19)
constant, and gate and drain biases, proves the veracity and
Therefore, we have a piecewise model for the drain current, robustness of the model, which has been derived from scratch
where J1 (18) is valid until VGS < VGSwitch , and J2 (19) is valid without employing any fitting parameters.
when VGS > VGSwitch . To make a smooth model for the drain Fig. 8 shows the output characteristics (IDS − VDS ) for
current valid over the entire range of gate biases, we need to various VGS and device parameters. At zero VDS the net drain

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8 IEEE TRANSACTIONS ON ELECTRON DEVICES

compared to a MOSFET because of the presence of a tunnel

barrier in the former limits its gate–source capacitance (CGS );
however, the larger channel charge at the drain side of the
TFET (n-n+ junction) compared to the MOSFET leads to
increased gate–drain capacitance (CGD ). Although the accurate
modeling of these capacitances relies on modeling the channel
charge from the surface potential model, a simplistic modeling
allows a 30–70 partition of the gate capacitance to CGS and
CGD , respectively, [16], irrespective of the applied voltage, and
the same has been incorporated into our compact model.

V. C ONCLUSION
A rigorous compact modeling framework for 2-D channel
TFETs has been developed that is valid for both lateral
homo- and hetero-source-channel junctions down to ultrashort
channel length of 5 nm. The framework incorporates fully
analytical modeling of the surface potential in the source and
channel by accurately considering the junction electrostatics,
the fringing gate electric field lines, and the carrier distribution
Fig. 7. IDS –VGS comparison of our analytical model (solid lines)
against NEGF simulations (symbols) for both (a) WSe2 -MoS2 DG-HTJ
profile dictated by FD statistics. Subsequently, the average
and (b) MoS2 -MoS2 DG-HMJ device at varying gate oxide permittivity, tunneling electric field and the net potential range for the
channel length, and VDS . The devices have a source and drain doping tunneling carriers are derived from the developed surface
concentrations of 5 × 1013 cm−2 and 2.5 × 1013 cm−2 , respectively with potential model to evaluate the transmission probability of
intrinsic channel. The body (t2D ) and gate oxide thicknesses (tox ) are
1 nm each. Metals with work-function of 4.56 eV and 4.5 eV are used for the tunneling carriers by employing the WKB approximation.
HTJ and HMJ simulations, respectively. Finally, the drain current model was derived by analytically
solving the 2-D Landauer’s equation by considering both the
forward and reverse currents and by appropriately integrating
over the allowed electron wave-vectors. The good agreement
of our model with NEGF simulations, without the use of
any fitting parameters, for both 2-D lateral HMJ and HTJ
TFETs, at various channel lengths and structural parameters,
provides necessary validation. Our developed compact model
can be utilized to study the performance of these various lateral
2-D TFETs and circuits derived from them, including novel
neuromorphic circuits.

ACKNOWLEDGMENT
Fig. 8. IDS –VDS comparison against NEGF simulations for WTe2 -MoS2 The authors would like to thank Vivek De of Intel Labs,
DG-HTJ device with various gate dielectrics and VGS . Device parameters
for simulation are same as in Fig. 6(b) with a channel length of L = 20 nm
Intel Corporation, Hillsboro, OR, USA, for useful discussions
and ϕm = 4.96 eV. and feedback. The authors also acknowledge Tanmay Chavan
of the Nanoelectronics Research Laboratory at UCSB for his
meticulous proof reading of this article.
current is zero because of the equal contribution of the forward
and the reverse current. However, as VDS continues to increase, R EFERENCES
the contribution of the reverse current diminishes because of [1] T. Sakurai, “Perspectives of low-power VLSI’s,” IEICE Trans. Electron.,
the decrease in the carrier occupation at the drain, thereby vol. E87-C, no. 4, pp. 429–436, 2004.
[2] W. Cao, J. Kang, D. Sarkar, W. Liu, and K. Banerjee, “2D semiconductor
increasing the net drain current, until it eventually saturates at FETs—Projections and design for sub-10 nm VLSI,” IEEE Trans.
higher VDS . TFETs, therefore, show more pronounced drain Electron Devices, vol. 62, no. 11, pp. 3459–3469, Nov. 2015, doi:
current saturation with VDS compared to MOSFETs, and the 10.1109/TED.2015.2443039.
[3] T. Baba, “Proposal for surface tunnel transistors,” Jpn. J. Appl. Phys.,
same is observed clearly from our figure. Most importantly, vol. 31, no. 4B, pp. L455–L457, Apr. 1992, doi: 10.1143/JJAP.31.L455.
Figs. 7 and 8 show the excellent match of our model results [4] Y. Khatami and K. Banerjee, “Steep subthreshold slope n- and p-
compared to the NEGF simulations for a variety of device type tunnel-FET devices for low-power and energy-efficient digital
circuits,” IEEE Trans. Electron Devices, vol. 56, no. 11, pp. 2752–2761,
parameters. Nov. 2009, doi: 10.1109/TED.2009.2030831.
The development of any compact model is incomplete with- [5] R. Gandhi, Z. Chen, N. Singh, K. Banerjee, and S. Lee,
out the inclusion of the capacitance in the device, which allows “Vertical Si-nanowire n-type tunneling FETs with low subthresh-
old swing (≤ 50 mV/decade) at room temperature,” IEEE Elec-
the modeling of transient characteristics. The low efficiency tron Device Lett., vol. 32, no. 4, pp. 437–439, Apr. 2011, doi:
of charge transport at the source-channel junction in a TFET 10.1109/LED.2011.2106757.

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
PAL et al.: COMPACT CURRENT–VOLTAGE MODEL FOR 2-D-SEMICONDUCTOR 9

[6] K. Tomioka, M. Yoshimura, and T. Fukui, “Steep-slope tun- [25] M. Gholizadeh and S. E. Hosseini, “A 2-D analytical model for double-
nel field-effect transistors using III–V nanowire/Si heterojunction,” gate tunnel FETs,” IEEE Trans. Electron Devices, vol. 61, no. 5,
in Proc. IEEE Symp. VLSI Technol., Jun. 2012, pp. 47–48, doi: pp. 1494–1500, May 2014, doi: 10.1109/TED.2014.2313037.
10.1109/VLSIT.2012.6242454. [26] S. Kumar et al., “2-D analytical drain current model of double-gate
[7] C. Convertino, C. B. Zota, H. Schmid, A. M. Ionescu, and heterojunction TFETs with a SiO2 /HfO2 stacked gate-oxide structure,”
K. E. Moselund, “III–V heterostructure tunnel field-effect transistor,” IEEE Trans. Electron Devices, vol. 65, no. 1, pp. 331–338, Jan. 2018,
J. Phys., Condens. Matter, vol. 30, no. 26, Jul. 2018, Art. no. 264005, doi: 10.1109/TED.2017.2773560.
doi: 10.1088/1361-648X/aac5b4. [27] Y. Guan, Z. Li, W. Zhang, and Y. Zhang, “An accurate analytical
[8] P. Ajayan, P. Kim, and K. Banerjee, “Two-dimensional van der Waals current model of double-gate heterojunction tunneling FET,” IEEE
materials,” Phys. Today, vol. 69, no. 9, pp. 38–44, Sep. 2016, doi: Trans. Electron Devices, vol. 64, no. 3, pp. 938–944, Mar. 2017, doi:
10.1063/PT.3.3297. 10.1109/ted.2017.2654248.
[9] B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, and A. Kis, [28] Y.-K. Lin, S. Khandelwal, J. P. Duarte, H.-L. Chang, S. Salahuddin,
“Single-layer MoS2 transistors,” Nature Nanotechnol., vol. 6, no. 3, and C. Hu, “A predictive tunnel FET compact model with atomistic
pp. 147–150, Mar. 2011, doi: 10.1038/nnano.2010.279. simulation validation,” IEEE Trans. Electron Devices, vol. 64, no. 2,
[10] W. Cao et al., “2-D layered materials for next-generation electronics: pp. 599–605, Feb. 2017, doi: 10.1109/TED.2016.2639547.
Opportunities and challenges,” IEEE Trans. Electron Devices, vol. 65, [29] R. B. Salazar, H. Ilatikhameneh, R. Rahman, G. Klimeck, and
no. 10, pp. 4109–4121, Oct. 2018, doi: 10.1109/TED.2018.2867441. J. Appenzeller, “A predictive analytic model for high-performance tun-
[11] Y. Khatami, J. Kang, and K. Banerjee, “Graphene nanoribbon based neling field-effect transistors approaching non-equilibrium Green’s func-
negative resistance device for ultra-low voltage digital logic applica- tion simulations,” J. Appl. Phys., vol. 118, no. 16, pp. 164305–164311,
tions,” Appl. Phys. Lett., vol. 102, no. 4, Jan. 2013, Art. no. 043114, 2015, doi: 10.1063/1.4934682.
doi: 10.1063/1.4788684. [30] Y. Taur, J. Wu, and J. Min, “An analytic model for heterojunction
[12] W. Cao, D. Sarkar, Y. Khatami, J. Kang, and K. Banerjee, “Subthreshold- tunnel FETs with exponential barrier,” IEEE Trans. Electron Devices,
swing physics of tunnel field-effect transistors,” AIP Adv., vol. 4, no. 6, vol. 62, no. 5, pp. 1399–1404, May 2015, doi: 10.1109/TED.2015.
Jun. 2014, Art. no. 067141, doi: 10.1063/1.4881979. 2407695.
[13] Y. Taur, J. Wu, and J. Min, “Dimensionality dependence of TFET [31] J. U. Mehta, W. A. Borders, H. Liu, R. Pandey, S. Datta, and L. Lunardi,
performance down to 0.1 V supply voltage,” IEEE Trans. Electron “III–V tunnel FET model with closed-form analytical solution,” IEEE
Devices, vol. 63, no. 2, pp. 877–880, Feb. 2016, doi: 10.1109/TED.2015. Trans. Electron Devices, vol. 63, no. 5, pp. 2163–2168, May 2016, doi:
2508282. 10.1109/TED.2015.2471808.
[14] H. Zhang, W. Cao, J. Kang, and K. Banerjee, “Effect of band-tails on [32] M. G. Bardon, H. P. Neves, R. Puers, and C. Van Hoof, “Pseudo-
the subthreshold performance of 2D tunnel-FETs,” in IEDM Tech. Dig., two-dimensional model for double-gate tunnel FETs considering
Dec. 2016, pp. 30.3.1–30.3.4, doi: 10.1109/IEDM.2016.7838512. the junctions depletion regions,” IEEE Trans. Electron Devices,
[15] D. Sarkar et al., “A subthermionic tunnel field-effect transistor with vol. 57, no. 4, pp. 827–834, Apr. 2010, doi: 10.1109/TED.2010.
an atomically thin channel,” Nature, vol. 526, no. 7571, pp. 91–95, 2040661.
Oct. 2015, doi: 10.1038/nature15387. [33] Y. Guan et al., “An accurate analytical model for tunnel FET output char-
[16] W. Cao, J. Jiang, J. Kang, D. Sarkar, W. Liu, and K. Banerjee, acteristics,” IEEE Electron Device Lett., vol. 40, no. 6, pp. 1001–1004,
“Designing band-to-band tunneling field-effect transistors with 2D semi- Jun. 2019, doi: 10.1109/LED.2019.2914014.
conductors for next-generation low-power VLSI,” in IEDM Tech. Dig., [34] M. O. Li, D. Esseni, J. J. Nahas, D. Jena, and H. G. Xing, “Two-
Dec. 2015, pp. 12.3.1–12.3.4, doi: 10.1109/IEDM.2015.7409682. dimensional heterojunction interlayer tunneling field effect transistors
[17] A. Szabo, C. Klinkert, D. Campi, C. Stieger, N. Marzari, and M. Luisier, (thin-TFETs),” IEEE J. Electron Devices Soc., vol. 3, no. 3, pp. 200–207,
“Ab initio simulation of band-to-band tunneling FETs with single- and May 2015, doi: 10.1109/JEDS.2015.2390643.
few-layer 2-D materials as channels,” IEEE Trans. Electron Devices, [35] J. Min and P. M. Asbeck, “Compact modeling of distributed effects in
vol. 65, no. 10, pp. 4180–4187, Oct. 2018, doi: 10.1109/ted.2018. 2-D vertical tunnel FETs and their impact on DC and RF performances,”
2840436. IEEE J. Explor. Solid-State Comput. Devices Circuits, vol. 3, pp. 18–26,
[18] Y. Dong, L. Zhang, X. Li, X. Lin, and M. Chan, “A compact model Dec. 2017, doi: 10.1109/JXCDC.2017.2670606.
for double-gate heterojunction tunnel FETs,” IEEE Trans. Electron [36] Y. Zhang, Z. Li, and Y. Guan, “Analytical drain current model of
Devices, vol. 63, no. 11, pp. 4506–4513, Nov. 2016, doi: 10.1109/TED. double-gate monolayer transition metal dichalcogenide TFET,” IEEE
2016.2604001. Trans. Electron Devices, vol. 66, no. 8, pp. 3652–3658, Aug. 2019, doi:
[19] P. K. Sahoo, S. Memaran, Y. Xin, L. Balicas, and H. R. Gutiérrez, “One- 10.1109/TED.2019.2922421.
pot growth of two-dimensional lateral heterostructures via sequential [37] D. Esseni, M. Pala, P. Palestri, C. Alper, and T. Rollo, “A review of
edge-epitaxy,” Nature, vol. 553, no. 7686, pp. 63–67, Jan. 2018, doi: selected topics in physics based modeling for tunnel field-effect transis-
10.1038/nature25155. tors,” Semicond. Sci. Technol., vol. 32, no. 8, pp. 083005-1–083005-27,
[20] P. Luo et al., “Doping engineering and functionalization of two- 2017, doi: 10.1088/1361-6641/aa6fca.
dimensional metal chalcogenides,” Nanosc. Horizons, vol. 4, no. 1, [38] W. Cao, J. Kang, W. Liu, and K. Banerjee, “A compact current–voltage
pp. 26–51, 2019, doi: 10.1039/c8nh00150b. model for 2D semiconductor based field-effect transistors considering
[21] K. Banerjee, “2D materials for smart life,” in Proc. IEEE 2nd Electron interface traps, mobility degradation, and inefficient doping effect,” IEEE
Devices Technol. Manuf. Conf. (EDTM), Kobe, Japan, Mar. 2018, Trans. Electron Devices, vol. 61, no. 12, pp. 4282–4290, Dec. 2014, doi:
pp. 4–6, doi: 10.1109/EDTM.2018.8421451. 10.1109/TED.2014.2365028.
[22] D. Sarkar and K. Banerjee, “Proposal for tunnel-field-effect-transistor [39] R. Landauer, “Spatial variation of currents and fields due to localized
as ultra-sensitive and label-free biosensors,” Appl. Phys. Lett., vol. 100, scatterers in metallic conduction,” IBM J. Res. Develop., vol. 1, no. 3,
no. 14, Apr. 2012, Art. no. 143108, doi: 10.1063/1.3698093. pp. 223–231, Jul. 1957, doi: 10.1147/rd.13.0223.
[23] A. Pal and A. K. Dutta, “Analytical drain current modeling of double- [40] N. Ma and D. Jena, “Interband tunneling in two-dimensional
gate tunnel field-effect transistors,” IEEE Trans. Electron Devices, crystal semiconductors,” Appl. Phys. Lett., vol. 102, no. 13,
vol. 63, no. 8, pp. 3213–3221, Aug. 2016, doi: 10.1109/TED.2016. pp. 132102–132105, 2013, doi: 10.1063/1.4799498.
2581842. [41] L. D. Landau and E. M. Lifshitz, Quantum Mechanics. Reading, MA,
[24] L. Liu, D. Mohata, and S. Datta, “Scaling length theory of double- USA: Addison-Wesley, 1958, p. 174.
gate interband tunnel field-effect transistors,” IEEE Trans. Electron [42] R.-H. Yan, A. Ourmazd, and K. F. Lee, “Scaling the Si MOSFET: From
Devices, vol. 59, no. 4, pp. 902–908, Apr. 2012, doi: 10.1109/TED.2012. bulk to SOI to bulk,” IEEE Trans. Electron Devices, vol. 39, no. 7,
2183875. pp. 1704–1710, Jul. 1992, doi: 10.1109/16.141237.

Authorized licensed use limited to: National Institute of Technology Patna. Downloaded on August 27,2020 at 10:44:07 UTC from IEEE Xplore. Restrictions apply.