The Tunnel Field-Effect Transistor

Devin Verreck,1, 2, a) Anne S. Verhulst,1 and Guido Groeseneken1, 2

1)
imec, Kapeldreef 75, 3001 Leuven, Belgium
2)
Department of Electrical Engineering, KU Leuven, 3001 Leuven, Belgium

I. INTRODUCTION along with the two main types of tunneling in Section IV.
This is followed by an overview of different approaches
The tunnel field-effect transistor (TFET) is a semicon- to model the TFET operation in Section V. The main
ductor device aimed at low-power logic applications that performance challenges for TFET are then presented in
employs band-to-band tunneling (BTBT) as a carrier in- Section VI. In Section VII, different material options
jection mechanism to obtain a subthermionic subthresh- for TFET are considered. The subsequent sections are
old swing (SS). In particular, it relies on the energetic devoted to various implementation options that are be-
filtering of the tail of the electron Fermi-Dirac distribu- ing researched to improve the TFET performance, such
tion to go below the fundamental 60 mV/dec SS limit at as dopant pockets (Section VIII), specific gate configura-
room temperature of a metal-oxide-semiconductor FET tions (Section IX) and strain (Section X). Section XI dis-
(MOSFET). The goal is to combine a low leakage cur- cusses attention points upon using the TFET in a circuit.
rent with a low SS to allow the TFET to be more energy Finally, Section XII gives a brief update on experimental
efficient than a MOSFET. work in literature. Section XIII concludes the article and
Research in TFET has been driven by the fundamen- provides a future outlook.
tal power issues encountered by MOSFET as device scal-
ing continues along the path of Moore’s law. This law,
II. MOSFET POWER ISSUE
in its most common formulation today, states that the
number of electronic components per chip resulting in
a minimum component cost, doubles approximately ev- The inherent trade-off between ION and IOFF for de-
ery two years1,2 . Also, the rise of mobile applications creasing VDD lies at the heart of the power density is-
and the Internet of Things, which contain a plethora of sue of MOSFET-based logic. This can be understood
always-on sensor nodes, has increased the demand for by looking at the total dissipated power of a circuit of
devices with low supply voltage (VDD ) and low-leakage MOSFET-based logic gates, which consists of a static
operation3,4 . The TFET aims to fulfill this demand by and a dynamic component9 :
exploiting the quantum mechanical phenomenon of tun- 2
Ptot = Pstatic +Pdynamic = Ng IOFF VDD +αCtot VDD f (1)
neling. Although it is a leakage mechanism for MOS-
FET in today’s scaled architectures, tunneling enables with Ng the amount of gates, α the fraction of active
the TFET to go beyond the inherent ON-current (ION ) gates, Ctot the total load capacitance of all gates and f
- OFF-current (IOFF ) trade-off that hampers the low- the switching frequency. Based on Eq. (1), a key element
power performance of the MOSFET. Several challenges in the so-called Dennard scaling of MOSFET, proposed
still remain, however, before TFET can be implemented in 197410 , is the reduction of VDD as the physical tran-
as a low-power replacement for MOSFET. sistor dimensions are decreased with every new genera-
Several excellent overview works exist which summa- tion in order to maintain a constant power density (note
rize the TFET state-of-the-art or go into more detail on that although Ctot decreases, the VDD reduction also re-
specific TFET topics5–8 . This article aims to give a broad sults in an increase of f ). The threshold voltage (Vth ) is
overview of the TFET field, with a distinct focus on de- decreased accordingly to maintain sufficient ION , which
vice physics and architecture options. For more details, is proportional to (VDD - Vth )x . The Dennard scaling
the reader is invited to consult the references which are paradigm has enabled the continuation of Moore’s law
included in each section. The article is structured as fol- until the end of the 20th century. However, Dennard
lows. First, the power issue of the MOSFET, which the scaling eventually leads to an untenable increase in the
TFET seeks to solve, is outlined in Section II. Next, the static power component, as the OFF-state leakage is ex-
basic operation of the TFET is explained in Section III, ponentially dependent on Vth 5 :
−Vth
IOFF ∼ e nkT /q (2)

a) devin.verreck@imec.be where kT /q is the thermal voltage with k the Boltz-

This is the peer reviewed version of the following article: Verreck, mann constant, T the temperature and q the elemen-
D., Groeseneken, G. and Verhulst, A., The Tunnel Field-Effect tary
 charge,
 and where n is the body factor, equal to
Transistor, Wiley Encyclopedia of Electrical and Electronics Cd
1 + Cox with Cd and Cox respectively the depletion
Engineering, 1-24, 2016., which has been published in final form at
http://onlinelibrary.wiley.com/doi/10.1002/047134608X.W8333. and oxide capacitance of a planar MOSFET. Around
This article may be used for non-commercial purposes in accor- the year 2002, the path of Dennard scaling was there-
dance with Wiley Terms and Conditions for Self-Archiving. fore abandoned, with the scaling of VDD slowing down
 2

with respect to the scaling of the physical transistor di-

mensions. As a result, power density has been increasing
as more and more transistors are included on a chip, with
each transistor consuming roughly the same power as the
previous generation. Notably, a significant portion of this

)
DS
power is consumed in the OFF-state. The rising power V

ec
log(I
th

/d
density leads to issues with cooling and reliability.

V
m
The origin of the exponential dependence in Eq. (2)

60
is the Fermi-Dirac distribution of the charge carriers in
the source region. In the subthreshold regime, also called I

c
OFF

de
weak inversion, a potential energy barrier in the MOS-

V/
FET channel region prevents low energy carriers in the

m
50
source from flowing to the drain contact. However, the
high energy carriers in the exponential tail of the distri- V (V)
GS
bution can still diffuse over the barrier in a process called
thermionic emission. This leads to an exponential depen-
dence of the drain-source current IDS on the gate-source FIG. 1. Schematic transfer characteristics of a MOSFET with
voltage VGS 11 : an ideal SS and a steep slope device with a sub-60 mV/dec
VGS −Vth
SS, illustrating that the SS determines IOFF . Scaling of the
IDS ≈ ID0 e nkT /q for VGS < Vth (3) threshold voltage Vth increases the OFF-current IOFF expo-
nentially.
with ID0 the current at VGS = Vth .
The inverse of the slope of Eq. (3) determines the
trade-off between ION and IOFF . It is called the sub- junction’, aimed at investigating subband splitting and
threshold swing (SS) and is used as a metric for the many-body effects in quasi-2D systems15 . However, it
switching steepness of the device. SS is defined as the was not until 1987 that BTBT was suggested as the work-
increase in VGS that is required to increase IDS with one ing principle of a transistor in a DRAM trench transistor
order of magnitude5,10 , and can be expressed as: cell16 . Major research efforts started after 2004, when
  a carbon nanotube TFET was demonstrated with a sub-
VGS Cd kT 60 mV/dec SS at room temperature17 and the use of het-
SS = ≈ 1+ ln(10). (4)
log10 IDS Cox q erostructures was conceptually introduced18 . Since then,
significant research efforts have been invested worldwide
At room temperature (T = 300 K), the SS of a MOSFET in improving the TFET performance by optimization of
is therefore theoretically limited to about 60 mV/dec. In the device architecture and material system.
actual implementations, non-idealities result in SS val-
ues which can be significantly higher. Fig. 1 illustrates
graphically that the value of the SS determines the inter-
III. TFET BASICS
section with the VGS = 0 V axis, which corresponds to
IOFF . IOFF increases exponentially as Vth is decreased
and the curve shifts leftwards. As IOFF becomes unac- A. Basic structure and operating principle
ceptably high, it prevents further concurrent scaling of
VDD and Vth and hence leads to aforementioned power The basic TFET structure is similar to that of a MOS-
density issues in highly scaled technologies. The under- FET: it contains two contact regions and an intrinsic or
limit on the SS makes this a fundamental trade-off. lowly doped channel region, covered by a gate dielec-
Several new transistor concepts have been proposed to tric and a gate contact (see Fig. 2(a)). In contrast to a
break the ION -IOFF trade-off by having a SS lower than MOSFET, however, the TFET contact regions have an
the MOSFET limit. Examples include concepts which opposite doping polarity, resulting in a p-i-n profile. In
use negative capacitance12 , impact-ionization13 and me- an nTFET, the p-type region acts as the source region,
chanical switches14 . However, these concepts give rise to while in a pTFET, the n-type region acts as the source.
hysteretic behavior and typically require a high operating Many variations on the basic configuration are possible,
voltage (> 1 V) at one of the transistor contacts. with different gate overlaps or doping profiles. These
will be discussed in Sections VIII to X. First, the work-
The TFET is a new transistor concept, compatible ing principle will be explained for a basic p-i-n nTFET in
with CMOS technology, that has been proposed for being a semiclassical picture. In this discussion, it is assumed
capable of having a SS lower than the MOSFET limit. that the doping is such that the Fermi-level in the source
This is possible because the TFET relies on quantum is aligned with the valence band edge.
mechanical BTBT instead of thermionic emission as the The TFET operates by enabling and preventing BTBT
carrier injection mechanism. The basic TFET structure between the source and the channel region by modula-
was first proposed in 1978 as a ’surface channel tunnel tion of the electrostatic potential in these regions with
 3

(a)
Gate

-4 V = 0.5V
DS
10
Source Drain
p i n
ON-state

I (A/ m)
-6
10
ambipolar
(b)
0.7V -8
10 subthreshold

ec
0V 0.5V

/d
-10
10

V
m
60
-0.5 1.2 -12
V
onset
(c) eV 10
-0.5 0.0 0.5 1.0
1.0 V (V)
VDS = 0.5V GS

0.5
E (eV)

FIG. 3. Semiclassical (SC) simulated transfer characteris-

0.0 E tics of an In0.53 Ga0.47 As p-i-n TFET as shown in Fig. 2(a).
Fp qVDS
The TFET body thickness is 20 nm with a source doping of
-0.5 OFF V =0.2V ξn EFn
5x1019 cm−3 and an EOT of 0.6 nm.
GS Ec
-1.0 ON V =0.7V
GS
V =0.5V, V =0.7V
-1.5 Ev 1.0
DS GS

19 -3 19 -3 20 -3
1x10 cm 5x10 cm 1x10 cm
-20 0 20 40 60 80
x (nm) 0.5 f
D
f
D
f
D
E(eV)

FIG. 2. (a) Basic p-i-n TFET configuration. (b) 2D electro- 0.0 p p

static profile of a p-i-n In0.53 Ga0.47 As TFET in the ON-state. E

Fp E
c
E
Fp E
c
E
Fp E
c
(c) Energy band diagram along the cutline in (a) in the OFF
(dark lines) and ON-state (light lines). The black arrows il- -0.5

lustrate the shortening of the available tunnel paths. The

E E E
quasi-Fermi level for the holes in the source (EFp ) coincides -1.0
v v v
(a) (b) (c)
with the valence band edge, while at the drain side, there is
a degeneracy ξn . -20 0 20 40 -20 0 20 40 -20 0 20 40
x (nm) x (nm) x (nm)

FIG. 4. Energy band diagram of the p-i-n In0.53 Ga0.47 As

the gate contact (see Figs. 2(b) and (c)). The source
TFET of Fig. 2 with different degrees of source doping de-
and drain contacts are biased such that the p-i-n diode is generacy ξp , ranging from (a) no degeneracy to (b) medium
in reverse bias. In the OFF-state, the only current that degeneracy to (c) high source degeneracy. As the source de-
flows is the reverse leakage current of the p-i-n diode. generacy increases, less of the exponential tail of the Fermi-
This leakage current is typically caused by minority car- Dirac distribution fD is filtered by the source bandgap. For
rier diffusion currents and by defect-assisted processes, the ease of extraction, a cross-section parallel to the gate has
such as Shockley-Read-Hall (SRH) generation and trap- been used.
assisted-tunneling (TAT) (see Section VI C). As VGS is
increased, the source is depleted and the electric field at
the tunnel junction rises. This corresponds to an increas- TFET.
ing band bending at the source-channel junction. At a
given VGS , called the onset voltage (Vonset ), the conduc-
tion band in the channel crosses over with the valence B. Transfer characteristics
band in the source, such that tunneling transitions be-
tween these two bands become available. These transi- Fig. 3 plots an example of typical TFET transfer char-
tions can be direct, between the maximum of the valence acteristics, IDS as a function of VGS , showing the different
band and the minimum of the conduction band at the Γ- operating regimes. The tunneling current rises as VGS
point, or indirect, between the maximum of the valence is increased above Vonset , since the length of the avail-
band and the minimum of one of the conduction band able tunnel paths decreases and the electron tunneling
valleys. In the indirect case, the transition is assisted probability is exponentially dependent on the tunneling
by a phonon. The tunneling transitions, whether direct distance. At VGS equal to VDD , the transistor is in the
or indirect, form the current generating process of the ON-state, and IDS at this point is ION . The transition
 4

from the OFF to the ON-state is, in analogy to the MOS- -8

x10
FET, called the subthreshold swing (SS), although there
is no fixed-swing subthreshold regime as in a MOSFET 1.4
V = 2V
GS
(see Section VI B), and hence also no Vth . The I-V curve
in Fig. 3 shows a SS lower than the 60 mV/dec MOSFET
1.2
limit.

m)
1.9V
The low SS of the TFET originates from the energetic

I (A/
filtering effect of BTBT carrier injection. Fig. 4(a) illus- 0.8

trates how the bandgap of the source material cuts off a

1.8V
significant part of the exponential tail of the Fermi-Dirac
0.4
distribution. This band-pass filter action effectively cools 1.7V
the carrier distribution. The origin of the SS-limit in a 1.6V
1.5V
MOSFET, which was discussed in Section II, is thus re- 0.0
moved. The TFET is therefore able to obtain a SS lower 0.0 0.5 1.0 1.5

than 60 mV/dec at room temperature. When the Fermi- V (V)

DS

level in the source is no longer aligned with the valence

band in the case of a highly doped source, the SS can be FIG. 5. SC simulated output characteristics of a Si p-i-n
degraded if the source degeneracy is too large. Figs. 4(b) TFET as shown in Fig. 2(a) for varying VGS . The TFET
and (c) show that a part of the exponential tail is then no body thickness is 20 nm with a source doping of 5x1019 cm−3
longer filtered. If this part of the tail results in observ- and an EOT of 0.6 nm.
able current, it deteriorates the SS. On the other hand,
Figs. 4(b) and (c) also show that a higher source doping
increases the electric field at the tunnel junction, which Fig. 5 also shows that in contrast to the MOSFET,
leads to a higher ION . This SS-ION trade-off, along with the onset of the TFET output characteristics can be su-
other parameters influencing SS, is discussed further in perlinear. This occurs when the BTBT is inefficient for
Section VI B. small VDS , e.g. as a result of a large tunnel path caused
The TFET is an ambipolar device, meaning that an by a large effective oxide thickness (EOT), a low source
nTFET operates as a pTFET when a negative VGS is doping or as a result of very low carrier density available
applied. In this case, the depletion occurs in the n-type for BTBT19–21 (see also Section VI A). In the superlinear
drain region as the electric field rises at the channel-drain regime, the output current is small. This is unwanted,
junction. After cross-over of the valence band in the since it increases the settling time of a TFET inverter.
channel with the conduction band in the drain, a hole Further circuit implications of this superlinear onset will
tunneling current is injected into the channel and the be discussed in Section XI D.
current increases with decreasing VGS (see Fig. 3). The
ambipolar behavior can be both an advantage and a dis-
advantage, and will be further discussed in Section XI A. IV. POINT VERSUS LINE TUNNELING

After introducing the basic operating principle in the

C. Output characteristics previous section, a first major distinction in tunneling
types can now be made. Based on the alignment of the
The source doping, drain doping and VDS deter- electric field induced by the gate with the dominant tun-
mine the energetic window available for tunneling (see nel paths, two types of tunneling can be defined in a
Fig. 2(b)). VDS fixes the hole quasi-Fermi level EFp in TFET: point tunneling (Section IV A) and line tunnel-
the source relative to the electron quasi-Fermi level in ing (Section IV B), which can both be present in the same
the drain EFn . The total tunneling window, the ener- device (Section IV C).
getic distance between the valence band in the source
and the conduction band in the drain, is then the sum
of (EFp -EFn ) and any doping degeneracies in source ξp A. Point tunneling
and drain ξn (see Fig. 4). For a constant VGS and start-
ing at a VDS of 0 V, an increase in VDS enlarges the Point tunneling is the dominant type in a standard p-i-
tunneling window and increases IDS (see Fig. 5). The n/n-i-p TFET as depicted in Fig. 2(a) or Fig. 6(a), with
increase of IDS with VDS continues until the conduction the gate predominantly covering the channel. This con-
band edge in the drain falls below the conduction band figuration is also called a pointTFET. The term ’point
edge in the channel. Beyond this VDS , called VDS,SAT , tunneling’ originates from the assumption that the tun-
IDS saturates, since an increase in VDS no longer has an nel paths curve around a central point at the interface
impact on the tunneling window. Since the conduction between gate metal and gate dielectric, right above the
band edge in the channel is determined by VGS , VDS,SAT tunnel junction. In this approximate semiclassical view,
increases with increasing VGS , just like in a MOSFET. the tunnel paths are assumed to lie on circles formed
 5

(a) Gate (b) Gate

by the electric field lines between source and gate, start-
ing at the tunnel junction and ending at the interface
between gate-dielectric and channel20 . Note that the cir- S S
cular form of the field lines is based on the assumption of p i p i
infinitely high source doping. At onset, only the longest
tunnel paths are available, which then gradually shorten
FIG. 6. Source-channel region of a (a) point tunneling and (b)
as VGS is increased toward the ON-state. In Fig. 2(a), line tunneling TFET configuration. The arrows schematically
these circular lines have been replaced with a straight line indicate the tunneling paths.
indicating the shortest allowed tunnel path, with roughly
same start and end point as the circular lines.
The pointTFET performance is influenced by field- the band bending is stronger, resulting in shorter tunnel
induced quantum confinement (FIQC), which is present paths. Important to note is that ION is proportional to
in the triangular well formed by the conduction band the gate-source overlap.
edge in the channel and the dielectric barrier22 (see The lineTFET’s performance is more heavily impacted
Fig. 7, although for a lineTFET, the shape of the well by FIQC than the pointTFET, since the triangular well
is similar). In this well, subbands are formed. The on- underneath the gate is more pronounced. The stretching
set of BTBT is thereby shifted to a higher VGS , since effect due to the slower increase in the tunneling window,
the band edge determining the allowed tunnel path in- also seen in the pointTFET, can therefore have a nega-
side the well is defined by the first subband level, which tive impact on the SS. Additionally, due to the strong
is higher in energy than the bulk value of the band edge. FIQC, several distinct subband energy ladders can ap-
As VGS is increased and the triangular well deepens, the pear in the well, since the degree of confinement depends
first subband level shifts down at a slower pace than the on the effective masses of the different band structure
bulk band edge. Hence, the tunneling window increases valleys. E.g. in a Si nTFET, those conduction band val-
more slowly than it would in the absence of FIQC. This leys which have a heavy longitudinal effective mass in the
has a stretching effect on the transfer characteristics, and direction of confinement, are shifted in energy less than
hence negatively impacts SS. The degree of this confine- those for which the lower transverse mass determines the
ment is determined by the effective mass of the band confinement. This leads to two ladders of quantized en-
structure valley to which the tunneling transition is tak- ergy levels, as shown in Fig. 7(b). As a consequence, the
ing place and therefore depends on the material choice for tunneling currents to the different valleys see a relative
the channel. The impact of FIQC can be alleviated with shift, which is visible in the transfer characteristics as
the introduction of a doping pocket at the source-channel a kink in the SS23 . This relative shift also affects the
interface, which forces the tunnel paths more parallel to pTFET, in which the light hole band is shifted more by
the gate (see Section VIII A). the confinement than the heavy hole band. Tunneling
between conduction band and heavy hole band is not as
efficient as to the light hole band: because of symme-
B. Line tunneling try reasons, there is no direct coupling, so any tunnel-
ing transition must be assisted by a phonon. In the ab-
Line tunneling can be induced with a large overlap of sence of confinement, this inefficient tunneling current is
the gate over the source (see Fig. 6(b)). Such a configu- masked by the higher light hole-conduction band current.
ration is also called a lineTFET. In contrast to point tun- In the presence of FIQC, however, the relative shift re-
neling, the tunnel paths are equally long parallel straight sults in the heavy hole-conduction band current forming
lines perpendicular to the gate dielectric in the ideal case an undesired tail to the transfer characteristics24 .
of no parasitic paths. For an increasing VGS , the energy
bands bend toward the gate dielectric until cross-over
occurs between the conduction band edge at the gate C. Point and line tunneling combined
dielectric and the valence band edge in the bulk, such
that BTBT becomes possible (see Fig. 7). In the ideal- Point and line tunneling current components can be
ized case of a uniform field underneath the gate dielec- present together in the same configuration. This is cer-
tric, implying that also the drain voltage impact on the tainly true if the gate overlaps both the source and the
source region is completely neglected, the onset is more channel region, but can also be induced by the fring-
abrupt than for point tunneling. All tunnel paths un- ing field of the gate, even if no gate-channel overlap is
derneath the gate-source overlap become available at the present. Because the line tunneling is more impacted by
same amount of band bending, after which they shorten FIQC, Vonset for the point tunneling component is lower,
uniformly for increasing VGS . The tunneling is also lo- meaning the gradual onset of the point tunneling can
cated closer to the gate than in a pointTFET, while the degrade the abrupt line tunneling onset. On the other
electric field is in line with the tunnel paths. This means hand, however, a gate-channel overlap removes the po-
a smaller increase in VGS is required to achieve a given tential barrier between the source and the ungated chan-
amount of band bending. Additionally, in the ON-state, nel, which can impede carriers from flowing to the drain
 6

sen basis. Tunneling then arises as a consequence of the

V =0.5V
DS
1.5 wave-like character of the electron. Here, the most com-
E
c
mon examples of both approaches are discussed, without
1.5V
being exhaustive.
1.0 Ec V =1.2V
GS
E (eV)

2V
0.5 1.7V
A. Semiclassical
E E
v v

0.0
E
Fp
E
Fp
A common semiclassical approach to the solution of the
time-independent Schrödinger equation is the Wentzel-
-0.5 Kramers-Brillouin (WKB) approximation. The reason-
ing behind WKB starts from the one-electron wave func-
(a) In Ga As (b) Si tion ψ in zero electric field, which corresponds to a con-
-1.0 0.53 0.47

stant potential energy. Assuming the electric field to be

10 15 20 10 15 20
zero in the tunneling direction x, and disregarding the
z (nm) z (nm)
other directions for now, ψ takes the form of a plane
wave26 :
FIG. 7. Energy band diagrams of a lineTFET in the OFF
(dark lines) and ON-state (light lines) along a vertical cutline
through the center of the gate in the [100] crystal direction. ψ(x) = Aexp (±ikx x) (5)
(a) An In0.53 Ga0.47 As configuration for a VGS of 1.2 V and
1.7 V and (b) a Si configuration for a VGS of 1.5 V and 2 V. with i the imaginary number, x the tunneling direction,
Indicated schematically in brown and red are estimations of kx the wave number and A the amplitude. The plus
the first three quantized energy levels for the highest VGS , (minus) sign corresponds to a right (left) moving wave.
corresponding to different valley effective masses inside the The approximation then lies in assuming that the wave
potential well. The dashed line represents the quasi Fermi function in the presence of a small and smoothly varying
level for the holes in the source (EFp ).
non-zero field, can be described by introducing a position
dependence for kx . It can be shown that the phase φ(x)
of the wave function can be obtained from the integral of
and hence can decrease ION 25 . This introduces a trade- kx (x) over the given domain26 :
off between SS and ION and makes the alignment of the
gate to the source-channel junction an important poten-  Z 
tial source for device variability in a lineTFET. However, C
ψ(x) ≡ A(x)exp(iφ(x)) ≈ p exp ±i kx (x)dx
the unwanted lateral tunneling component and the asso- |kx (x)|
ciated variability can be removed with the introduction (6)
of a counterdoped pocket, as discussed in Section VIII B. with C a real constant. In a forbidden energy region,
like the bandgap, kx is imaginary (kx = iκx ), which re-
sults in an exponential decay of the wave function. This
V. MODELING corresponds to a tunneling process. By comparing the
probability density at each side of the tunneling barrier,
an expression for the transmission probability can be de-
To acquire more physical insight into TFET operation
rived. This expression typically ignores the prefactors of
and assess different architecture options, several mod-
the exponentials and therefore solely consists of a con-
els have been established. Since tunneling is a quantum
tour integral of the imaginary kx along the tunnel path
mechanical phenomenon, all of these models rely on the
through the forbidden region:
solution of some form of the Schrödinger equation. Even
so, the different solution approaches can be categorized  Z x2 
as either semiclassical (Section V A) or fully quantum TWKB = exp 2 κx (x)dx (7)
mechanical (Section V B). In semiclassical models, the x1
Schrödinger equation is not solved directly. Rather, the
tunneling probability is calculated based on the integral with x1 and x2 respectively the start and end point
of a position dependent imaginary wave vector along a of the tunnel path, also known as the classical turning
well-defined tunnel path. This implies the classical as- points26–28 . The tunnel path starts at the valence band
sumption that wave vector and position of the electron edge and ends at the conduction band edge for a particu-
are known at the same time, violating the Heisenberg lar set of perpendicular wave numbers ky and kz . For zero
uncertainty principle. In fully quantum mechanical ap- perpendicular momentum (ky = kz = 0 and κx = κx0 ),
proaches, on the other hand, the electron is described the application of Eq. (7) is illustrated in Fig. 8. For
entirely by its wave function, which is obtained from a non-zero ky and kz , the effective tunnel gap increases,
solution of the Schrödinger equation projected on a cho- making transmission less probable. This effect can be
 7

(a) (b)

Ec )&
!
*
)

Ev
κx0
E x1 x2 !"#

!"#$
κx0 *
+

E ''
$
x %&
('
$"#
5 6.,2 / ! 01,2 / 5
3 3
FIG. 8. (a) Real and imaginary 2-band k·p band structure of 4
3$! ,$-./
bulk InAs for zero perpendicular momentum (ky = kz = 0).
Indicated are the conduction (co) and light hole (lh) bands.
(b) Energy band diagram, superimposed with the imaginary FIG. 9. Real and imaginary 30-band k·p band structure of
band structure at a given energy E. The WKB transmission bulk In0.53 Ga0.47 As. Indicated are the conduction (co), light
probability is calculated from an integral of this imaginary hole (lh), heavy hole (hh) and split-off (so) bands.
dispersion.

a direct bandgap material can then be written as:

made explicit by rewriting Eq. (7) as:
!
3/2 1/2  
πEG mR E⊥
TKane (ky , kz ) = exp − exp −2 (9)
2qh̄F E
 Z x2 
TWKB (ky , kz ) = exp 2 κx0 (x)dx
x1 where EG is the effective bandgap at the tunnel junction
 Z x2
dx
 and mR is the reduced effective mass, defined as mR =
× exp −|k2y + k2z | me mlh
me +mlh , with me and mlh the effective masses of respec-
x1 κx0 (x)
tively the conduction band and the light hole band. In
(8) the second exponential factor, E⊥ = h̄2 (ky2 + kz2 )/(2mR )
  2
1/2
and E = 2qh̄F/ π(mR )1/2 EG . A factor π9 , present in
where the assumption has been made that ky2 + kz2 << the original Kane derivation, was shown later to be incor-
2 28
ktot . It is now clear that the second exponential factor rect and has been removed30 . Because of the assumption
of Eq. (8) reduces the transmission for non-zero perpen- of a constant electric field, the Kane model can deviate
dicular momentum27,28 . In the case of direct BTBT, κx0 significantly from the WKB approach in cases where the
can be extracted from the complex band structure of the field is strongly non-uniform30,31 .
material under study (see Fig. 8(a) and Fig. 9 for exam- From the transmission probabilities in Eq. (7) and
ple band structures). Although the WKB method can Eq. (9), the BTBT current can be calculated in two ways.
describe quantum phenomena like tunneling, it is still The first is based on the ballistic Landauer formalism
a semiclassical approach, because Eq. (8) requires that and entails a direct integration of the transmission prob-
both position and momentum (expressed by the wave abilities, weighted with the distribution functions in the
vector) are known at the same time. This is possible in a contacts. This is called the Tsu-Esaki formula32 , which
classical approach, but violates the quantum mechanical gives the BTBT current density as:
Heisenberg uncertainty principle. From the WKB trans-
mission probability in Eq. (7), the BTBT current can be
Z
2q dkx dky dkz
calculated. This is discussed later in this section. JDS = T (ky , kz )(fS (E) − fD (E))
h kx ,ky ,kz 2π 2π 2π
Another commonly used semiclassical model to cal- (10)
culate the transmission probability is Kane’s model29 . where T is the transmission probability and fS (E) and
Originally, the Kane formula for BTBT probability was fD (E) are the Fermi-Dirac distributions in the source and
derived for a uniform electric field in a perturbative ap- drain contacts respectively. The energy E is a function of
proach, using Fermi’s golden rule and assuming a 2-band kx , ky and kz . Eq. (10) illustrates that the BTBT current
k·p-model. An equivalent result can be obtained by is determined by both the transmission probability and
starting from the WKB method and assuming the electric the occupation probabilities at both sides of the tunnel
field F to be constant over the tunnel path length, while junction.
taking a two band k·p model to describe the complex A second way with stronger simplifications is to calcu-
wave vector dispersion. The transmission probability in late the BTBT current is to determine a generation rate
 8

per unit volume, which is then integrated over the full de- B. Quantum mechanical
vice volume. In this approach, device simulators search
for tunnel paths which connect points of sufficient poten- A fully quantum mechanical simulation approach en-
tial difference to allow for tunneling. Carriers are gener- tails the solution of the Schrödinger equation in some
ated by the BTBT process at the endpoints of the tun- form. In TFET modeling, the most commonly used is
nel paths. An expression for the generation rate can be the time independent one-electron form:
derived from Eq. (10) by assuming the distribution func-  2 
tions for the carriers to be step functions, corresponding −h̄ 2
Ĥψ(r) = ∇ + Ve (r) + Vc (r) ψ(r) = Eψ(r) (17)
to a temperature of 0 K. This effectively decouples the 2me
carrier distribution functions from the transition rates.
with me the free electron mass, E the total energy,
For the Kane model, the following generation rate for-
Ve (r) the external applied potential energy and Vc (r) the
mula is commonly used:
 D   crystal potential energy of the lattice. The solution of
F −B Eq. (17) generally occurs in two steps. First, the wave
GKane = A exp (11) function is decomposed onto a reduced basis. This is be-
F0 F
cause the complicated nature of Vc (r) prevents the direct
where F is the electric field, F0 is 1 V/cm, D is a param- solution of Eq. (17) for ψ. The choice of basis corre-
eter that is taken 2 for direct bandgap materials and A sponds to a particular band structure model. For TFET,
and B are parameters defined as: the most commonly used models are the k·p-based enve-
1/2 lope function method and the orbital-based tight-binding
gmR (qF0 )2
Adirect = (12) method. Once the Hamiltonian has been written in the
πh2 (EG )1/2
chosen basis, the second step is to construct a linear sys-
1/2 tem or an eigenvalue problem, the solutions of which can
π 2 mR (EG )3/2 be used to extract desired quantities like currents and
Bdirect = (13)
qh carrier densities. The two main approaches for this step
where g is a factor for the spin and valley degeneracies. are the wave function method and the non-equilibrium
Corrections that reintroduce the non-zero temperature Green’s function (NEGF) method.
distribution functions afterwards have been developed33 . A first band structure model is the envelope function
Extensions of Kane’s model have been made to describe method, which expands the wave function on the solu-
indirect phonon-assisted transitions as well27 . For indi- tions of the Schrödinger equation in bulk36 :
rect BTBT, D in Eq. (11) is 2.5 and A and B are modified X
to25 : ψ(r) = Fn (r)Un (r), (18)
n
g(mc mv )3/2 (1 + 2NPH )DP2 H (qF0 )5/2
Aindirect = 5/4 7/4
(14) where the Un (r) form a complete set of orthonormal ba-
221/4 h5/2 mR ρPH EG sis functions with the periodicity of the lattice and Fn (r)
1/2 3/2
are slowly varying envelope functions, which contain only
27/2 πmR EG Fourier components in the first Brillouin zone. The index
Bindirect = (15)
3qh n runs over all bands considered in the description. In-
where mv (mc ) is the valence (conduction) band den- serting the expansion of Eq. (18) into Eq. (17) and using
sity of states effective mass, NPH = 1/ [exp(PH /kT) − 1], the properties of the set Un (r), namely orthonormality
DPH and PH are respectively the occupation number, and completeness, the following system of equations re-
the deformation potential and the energy of the relevant sults (a detailed derivation can be found in the work of
phonons and ρ is the mass density. To obtain the BTBT Burt36 and Van de Put et al.34 ):
current, the generation rate of Eq. (11) is integrated over −h̄2 2 ih̄ X X
the device volume: ∇ Fn (r) − pnm · ∇Fm (r) + Hnm (r)Fm (r)
Z 2me me m m
IDS = q GdV (16) +Ve (r)Fn (r) = EFn (r)
(19)
with dV an elementary volume. The same approach can
be followed for the WKB approximation33 . where the external potential Ve is assumed to vary slowly
The disadvantage of semiclassical methods is that they on the scale of a unit cell, like the envelope functions.
neglect certain quantum phenomena which result from The pnm are known bulk k·p interband momentum ma-
the wave-like character of the electron, since they do not trix elements. They describe the coupling strength, and
directly solve the Schrödinger equation in the full de- hence BTBT, between bands n and m. Hnm are the
vice region. Important examples for TFET include field bulk Hamiltonian matrix elements, which correspond to
or size-induced confinement effects, which can effectively known bulk band edge energies. The main advantage of
increase the bandgap, and resonances and reflections in Eq. (19), compared to Eq. (17) is therefore that the crys-
regions of high field, which respectively increase or de- tal potential Vc (r) has been replaced with known material
crease the transmission probability34,35 . parameters.
 9

An alternative band structure description, called the transmission probabilities and currents. Similar to the
tight-binding method, consists of expanding the wave semiclassical case, the current can be calculated from the
function on Bloch sums of localized atomic orbital-like transmission probabilities directly with Eq. (10), or more
functions, instead of on extended bulk solutions37–39 : approximated by first converting to a generation rate and
using Eq. (16). To include the effect of the carrier con-
1
X X
ψ(r) = N − 2 Cn exp(ik · Ri )φn (r − Ri ) centration on the potential energy profile, the calculated
n i carrier densities can be used to calculate a new potential
=N − 21
X
Cn,i φn (r − Ri ) (20) energy profile with the Poisson equation, which in its
turn can be used to calculate a new carrier density. This
n,i
loop is then repeated until self-consistency is reached.
where φn is a Löwdin orbital with quantum number n, Instead of solving Eqs. (19) and (21) directly as in
located on the atom at position Ri . N is the num- the wave function method, an alternative is to calculate
ber of primitive unit cells in the crystal and serves as the non-equilibrium Green’s function (NEGF)38,47 . In an
a normalization constant. Cn,i = Cn exp(ik · Ri ) are NEGF approach, Eq. (17) is rewritten as:
the expansion coefficients, which serve a similar pur-
pose as the envelope functions Fn of Eq. (18). Inserting (EI − Ĥ − Σ̂S/D )G = I (22)
the expansion of Eq. (20) into the Schrödinger equation
with G the Green’s function, Ĥ the Hamiltonian in the
Eq.
P (17), multiplying from the left with the Bloch sum
∗ chosen basis and Σ̂S/D the self-energies of the source and
j exp(−ik · R )φ
j m (r − Rj ) and integrating over the full
drain contacts. The self-energy terms are added to in-
crystal, the following system of equations is obtained:
clude the open boundary conditions. Just like the wave
X Z X function, G contains all relevant device information. The
Cn,i−j φ∗m (r − Rj )Ĥφn (r − Ri )dr = Cn,i−j E advantage of the NEGF approach is that other interac-
i,j i,j tions can be included in a rather straightforward manner
(21) by adding the corresponding self-energy term. E.g. for
with Cn,i−j = Cn exp(ik · (Rj − Ri )). The integrals in electron-phonon interactions:
Eq. (21) can be replaced by parameters available in liter-
ature, similar to the pnm elements in the envelope func- (EI − Ĥ − Σ̂S/D − Σ̂el−ph )G = I (23)
tion approach.
For both the envelope function method and the tight- with Σ̂el−ph the self-energy for the electron-phonon
binding method, the inclusion of more basis functions interaction38 . In this way, rethermalization, phonon-
results in a more accurate band structure that can cap- assisted tunneling and other scattering processes can be
ture a larger part of the first Brillouin zone. If enough described. To avoid having to calculate the inverse of a
bands are included such that the full first Brillouin zone large matrix (EI − Ĥ − Σ̂S/D − Σ̂el−ph ), methods have
is captured, the model is called a full-zone model. For the been developed that calculate only the relevant entries in
envelope function method, this corresponds to a 30-band the Green’s function matrix, such as the contact block re-
basis40,41 , for the tight binding method to the sp3 d5 s∗ duction (CBR) method48 and the recursive Green’s func-
basis set42 . With more basis functions, also the coupling tion (RGF) method38 .
between the different bands is captured more accurately A separate quantum-mechanical framework based on
(compare e.g. Fig. 8 with Fig. 9). wave functions has also been established for phonon-
To solve Eq. (19) or Eq. (21) for an actual device assisted tunneling23 . It was shown that the current can
in which current enters and leaves through the source be determined from solving Eq. (19) for each band sep-
and drain contacts, open boundary conditions are re- arately. Two distinct sets of wave functions are then
quired. For TFET simulations, the quantum transmit- obtained for the conduction band and the valence band.
ting boundary method (QTBM) is often used43 . QTBM The transmission probability is calculated based on the
imposes that the electric field is zero in the contacts in overlap of the wave functions of the two bands, combined
the transport direction. The wave function is then known with the interaction probability with a phonon of the ap-
to take the form of a plane wave in that direction. It is propriate momentum.
this form which is imposed as a boundary condition on
the system.
With the appropriate boundary conditions, Eqs. (19) C. Semiclassical versus quantum mechanical
and (21) can be discretized and solved for the enve-
lope functions and tight binding coefficients respectively. Compared to semiclassical approaches, quantum me-
This direct solution approach is called the wave function chanical models are typically computationally more ex-
method. Available numerical techniques include finite pensive, but include the effects of the wave-like nature of
difference methods44 , finite element methods45 , spectral the electron, such as confinement and reflections in re-
methods46 and others. The envelope functions and the gions of high field or at a heterojunction. Quantum sim-
tight binding coefficients can be used to calculate im- ulations should therefore be preferred in cases where con-
portant device characteristics such as carrier densities, finement is expected to be strong, e.g. for TFETs with
 10

A. ION

-3
10 SC The TFET ION is typically lower than that of a MOS-
QM FET. This is due to the tunneling barrier, which all
V
DS
= 0.6V charge carriers have to overcome before they can drift
m)

-5 and diffuse toward the drain. This barrier is present

10
even in the ON-state of the device. In contrast, in a
I (A/

MOSFET, the carriers with an energy higher than the

source-channel barrier can drift and diffuse from source
-7
10 to drain unimpeded. The lack of sufficient drive current
negatively impacts the intrinsic delay of a TFET inverter
configuration, expressed as:
-9 (a) (b)
10 Cox VDD
0 1 2 3 0 1 2 3 τd = (24)
V
GS
(V) V
GS
(V) ION
From Eq. (24), it is clear that to keep the delay small in
FIG. 10. Transfer characteristics of a Si lineTFET simu- low-VDD operation, ION should be sufficiently high. In
lated with a semiclassical (SC) approach based on the WKB order to compete with MOSFET, it is generally accepted
method and a quantum mechanical (QM) approach based on
that ION should be in the range of several hundreds of
wave function overlap. (a) Unshifted and (b) shifted char-
acteristics such that the VGS at which IOFF is 1x10−9 A/µm µA/µm for VDD smaller than 0.5 V4 .
coincides. To identify the parameters that increase ION , the Kane
model can be used, although it is strictly speaking only
valid for uniform fields. The exponential factors in
Eqs. (11)-(15) show that the BTBT generation rate, and
a body thickness below the Bohr radius of the material, hence ION , is largest for a high electric field at the tunnel
or when the tunneling is oriented toward the quantum junction, a small bandgap and a small reduced effective
well underneath the gate dielectric. Other cases include mass. A high F and small EG correspond physically to
TFETs with a heterojunction49 or configurations that short tunnel paths, while the low mR signifies a small
rely on resonant tunneling50 . attenuation along the path, as it is correlated with the
imaginary dispersion in the bandgap. A short tunnel
To illustrate the differences that can exist be- path and a small attenuation result in a high probabil-
tween semiclassical and quantum mechanical approaches, ity for the charge carriers to tunnel into the channel and
Fig. 10 compares the simulated transfer characteristics of contribute to the current. A high F can be obtained
a Si lineTFET, whereby the last described quantum me- with sharp doping profiles and a large doping level at the
chanical approach for phonon-assisted tunneling is used. tunnel junction, by increasing the source doping and/or
Since the tunneling is oriented towards the gate, a strong with the introduction of a counterdoped pocket (see Sec-
impact of FIQC is expected. The effect of FIQC is visible tion VIII). Additionally, F can be improved by enhanc-
for the quantum simulation as a shift in Vonset compared ing control of the gate over the tunnel junction, e.g. by
to the semiclassical prediction and a kink due to different thinning the device body, by adding gates to obtain a
subband ladders (see also Section IV B). multi-gate (MuG) configuration or by modifying the de-
vice structure to a line tunneling set-up (see Sections IV
and IX). The small EG and reduced mR , on the other
hand, are the result of a proper material choice, with
III-V materials proving promising candidates, either in
VI. CHALLENGES a homostructure (one material) or in a heterostructure
(multiple materials) configuration (see Section VII).

Having a basic understanding of TFET and methods

to model the tunneling current, the main challenges for B. SS
TFET implementation are now discussed in more detail.
Although the operating principle of the TFET is promis- In addition to a high ION , a TFET should obtain a low,
ing for low power applications, several important chal- sub-60 mV/dec SS over a large current range of interest.
lenges remain. They can be summarized as reaching a The SS of a TFET at a specific VGS can be approximated
high ION (Section VI A), combined with a low SS over as:
several decades of current (Section VI B), while main-  −1
taining a low IOFF (Section VI C). These three device dVGS 1 dVR F + B dF
SS = = ln(10) +
characteristics are discussed with the appropriate metrics dlog10 IDS VR dVGS F2 dVGS
and the parameters which have an influence on them. (25)
 11

TABLE I. Examples of recent experimental values for I60 in

literature, rounded to order of magnitude. Target value for
-4
I60 is 10 µA/µm or larger. 10
I
nTFET I60 [µA/µm] pTFET I60 [µA/µm] ON

I (A/ m)
-6
Tomioka et al.53 10−3 Mayer et al.54 10−6 10 I
60
−3
Dewey et al. 55
10 Morita et al. 56
10−5
57 −4
Sarkar et al. 10 -8
10
SS
Huang et al. 58
<10−6 avg

Kim et al. 59
<10−6 -10
10 I
OFF

V V
-12
which is derived from an approximation of the tunnel 10
OFF DD

junction as a degenerately doped p-n junction with a 0.0 0.5

constant electric field51 . VR is then the reverse bias V

GS
(V)

of the tunnel junction (EFp -EFn ) and B is the exponen-

tial parameter of the Kane formalism (see Eq. (13) and FIG. 11. SC simulated transfer characteristics of an
Eq. (15)). Comparing Eq. (25) to Eq. (4) shows that in In0.53 Ga0.47 As p-i-n TFET as shown in Fig. 3, illustrating
contrast to the MOSFET, the SS of a TFET is indeed various metrics used to characterize TFET performance.
not limited by kT q ln(10). However, Eq. (25) also shows
that the TFET SS is not constant with VGS . It is there-
fore possible that the sub-60 mV/dec SS regime is only probability. This means the optimization of SS and ION
limited to very small levels of IDS . In this case, reducing are intertwined, with one generally benefiting from the
the supply voltage is still not feasible, as the majority of improvement of the other. This is not true, however,
the SS is above the thermionic limit. The minimum value for very high source doping levels. A large source dop-
of the SS as described in Eq. (25) at a particular VGS - ing improves ION , but can degrade the SS if the source
point, also called the minimum point-SS, is therefore not degeneracy becomes too high, as was mentioned in Sec-
a good metric to assess TFET performance. More rele- tion III. For a given doping level, the source degeneracy
vant is to look at the average SS over the full operating is determined by the density of states (DOS) in the va-
voltage range, as shown in Fig. 11: lence band (for an nTFET) or the conduction band (for a
pTFET) of the source material. A large DOS allows for a
VDD − VOFF high source doping while keeping the source degeneracy,
SSavg =   (26)
log IIOFF
ON and hence the SS degradation, small. As the conduction
band for most common bulk materials has a lower DOS
where VOFF is the VGS corresponding to the defined than the valence band, the pTFET has a more limiting
IOFF . SSavg still depends, however, on the choice of IOFF trade-off between ION and SS. This puts the pTFET at
and VDD , which can vary depending on the targeted ap- an inherent disadvantage compared to the nTFET. This
plication. Therefore, another metric called I60 has been imbalance can be alleviated using dopant pockets, dis-
proposed52 . I60 is defined as the point on the IDS -VGS cussed in Section VIII.
curve where the SS transitions from sub-60 mV/dec to The SS can be negatively impacted by parasitic current
super-60 mV/dec (also indicated in Fig. 11). Hence, mechanisms which occur in addition to the desired BTBT
it is the largest current for which the corresponding process, such as trap-assisted tunneling (TAT)60–62 . TAT
TFET outperforms a theoretically optimal MOSFET. is a process in which trap levels in the bandgap are used
I60 is independent of the choice of IOFF , VDD or the by the charge carriers as stepping stones between con-
gate workfunction, so it can be used to compare vari- duction and valence band (see Fig. 12(a)). These lev-
ous TFET configurations presented in literature. To be els are localized energy states resulting from defects in
competitive with MOSFET, I60 should be at least above the material, such as vacancies, impurities and dangling
10 µA/µm4,52 . As an example, Table I lists some experi- bonds in the bulk or at interfaces. The TAT transition
mental values obtained in literature. See also Section XII itself consists of at least one ballistic tunneling step and a
for more experimental results. These examples show that phonon-assisted thermal step to and from the trap level.
there is still a large gap with the SS goal. This means TAT has both a field and a temperature de-
From Eqs. (25) and (26), it is clear that SS can be mini- pendence. TAT degrades the SS, as it can turn on at
dVR
mized by increasing dV GS
and dF
dVGS , which represent the
max
lower VGS and add a thermal tail to the transfer charac-
gate control over the tunnel junction, and by increas- teristics. The impact of TAT can be reduced by improv-
ing the ratio ION /IOFF . Physically, these requirements ing the material quality, such that the trap density is low,
correspond to the capability of the gate to induce an or by enhancing the desired BTBT current, such that it
abrupt transition from an OFF-state with a low tunnel- better masks the current component caused by TAT. The
ing probability to an ON-state with a very high tunneling amount of TAT present in the transfer characteristics can
 12

be characterized by VTAT @IOFF , the increase in Vonset at

300 K at the specified IOFF compared to the Vonset value 1.0
at 77 K63 .
(a)
0.5

E (eV)
E
trap

C. IOFF E
0.0 c
E (b) E
Fp Fn
E
To obtain a low leakage power, IOFF should be suffi- trap
-0.5
ciently small. IOFF is composed of the reverse leakage E
v
current of a p-i-n diode, consisting mainly of SRH pro-
-1.0
cesses, minority carrier diffusion, and possibly some TAT -20 0 20 40
0 60 80
if the other two mechanisms are sufficiently small. Addi- x (nm)
tionally, parasitic TFET mechanisms such as ambipolar
current and direct source-drain tunneling can increase
FIG. 12. Energy band diagrams of the tunnel junction of
IOFF . For low-power applications, the ITRS roadmap
an In0.53 Ga0.47 As TFET showing (a) TAT and (b) SRH pro-
sets the target for IOFF at 10 pA/µm, with IIOFF ON
> 105 . cesses, which contribute to the TFET IOFF . The green lines
SRH, similar to TAT, is a trap-assisted process in represent localized trap energy levels, the red wavy lines are
which carriers are generated via localized trap levels thermal steps and the black solid line corresponds to a tun-
through phonon-assisted thermal steps when the electron neling step.
and hole densities are out of equilibrium (see Fig. 12(b)).
In contrast to TAT, SRH has a weak field dependence
and so does not impact the SS. However, if the trap den- IV (Section VII A) and compounds from group III and V
sity is sufficiently high, it can increase IOFF . Similar to of the periodic system of elements (Section VII B). Multi-
TAT, the remedy is to decrease the trap density. ple of these materials can be combined in a heterostruc-
The ambipolar current is determined by the efficiency ture (Section VII C). 2D materials are a new class of
of BTBT at the channel-drain junction. This means it materials, which is also gaining interest for TFET appli-
can be reduced by taking opposite measures as those sug- cations (Section VII D).
gested in Section VI A to increase ION , viz. a reduction
of F and EG and an increase in mR at the channel-drain
tunnel junction. F can be reduced by lowering the drain A. Group IV materials
doping, at the expense of increasing the resistance. EG
and mR depend on the material choice and cannot be Silicon is the most prevalent material in today’s MOS-
varied independently from the source in a homostruc- FETs for logic applications, and has therefore also been
ture, but they can be in a heterostructure (see Section extensively studied for TFET. Si has the distinct advan-
VII C). Another option is to introduce a gate-drain un- tage of being a well-known, abundant material that can
derlap region, which decreases the influence of the gate be obtained with a very low density of defects in the bulk
on the channel-drain junction64 (see Section IX B). Using of the material as well as at the gate dielectric interface.
a combination of these techniques, the ambipolar current Established fabrication infrastructure and processes ex-
can be kept to acceptable levels. ist in industry for a wide variety of process steps. This
Direct source-drain tunneling occurs when charge car- includes high quality oxide growth and high concentra-
riers can tunnel from source to drain, even when the de- tion doping. Si is also interesting for TFET because of
vice is in the OFF-state. This occurs when the tunnel its large conduction band DOS, which limits the source
paths in the OFF-state are too short or provide insuffi- doping degeneracy in pTFETs (the importance of which
cient attenuation to suppress the carrier tunneling proba- is discussed in Section VI B).
bility. Direct tunneling is particularly important in mate- However, Si is a relatively poor material for BTBT
rials with a small EG and mR and in configurations with and consequently, Si TFETs typically have a small ION .
a short channel. In vertical architectures, increasing the A first cause is the relatively large EG of 1.12 eV, which
channel length is a straightforward way to reduce the results in a low tunneling probability (see Section VI A).
source-drain tunneling. In planar configurations, how- Although beneficial for IOFF , the low tunneling rate is
ever, a longer channel length also increases the device detrimental for ION . Secondly, Si is an indirect bandgap
footprint. material, which reduces the overall efficiency of BTBT.
The smallest EG is measured between the valence band
maximum at the Γ-point and the conduction band val-
VII. MATERIAL CHOICE ley minimum along the ∆-direction40 . These two points
in the E-k diagram are not at the same k-value, which
To overcome the challenges outlined in the previous means they correspond to a different crystal momentum
section, different material options are being considered h̄k. A tunneling transition between these two points re-
for TFET. Of particular interest are materials from group quires a phonon to provide the necessary crystal momen-
 13

tum. This makes it less probable for a given energetic Straddled Staggered Broken
separation than a direct transition, in which the start Ec1 Ec1 Ec1
Ec2
and end points of the tunneling process are at the same Ec2
k-value, therefore requiring no phonon. In Si, the indi- Ev2
Ev1 Ev1 Ev1
rect transitions are dominant, since the bandgap at the Ev2 Ec2
Γ-point is much larger at 3.4 eV, making direct transi-
tions very improbable25 . Ev2
As an alternative to Si in group IV of the periodic
system of elements, Ge is more suited for BTBT. It has
a smaller indirect bandgap than Si of 0.66 eV. Unlike Si, FIG. 13. Types of band edge alignments at a heterostructure
tunnel junction. In a straddled alignment, Ec1 > Ec2 and Ev1
the direct bandgap at the Γ-point is only slightly larger at
< Ev2 . In a staggered alignment, Ec1 > Ec2 and Ev1 > Ev2 .
0.8 eV. Consequently, once the band bending is such that In a broken alignment, Ev1 > Ec2 . The green arrows indicate
direct transitions are allowed, they will dominate over the the effective bandgap at the tunnel junction. For the broken
indirect transitions and improve ION 25,65 . However, as configuration, the effective bandgap is negative.
discussed in Section VI C, a smaller bandgap facilitates
ambipolar tunneling at the channel-drain junction, which
is an important leakage mechanism in TFET. IOFF will tion VII C. Thirdly, a low electron effective mass means
therefore generally be larger than for Si implementations, a low DOS in the conduction band compared to Si. This
but it is expected that acceptable levels of IOFF can be exacerbates the ION -SS trade-off for pTFET for increas-
reached. ing source doping, discussed in Section VI B. Solutions
Since it belongs to the same group as Si, Ge can also exist in the form of dopant pockets, as discussed in Sec-
be used in an alloy with Si, forming Six Ge1−x , with x tion VIII.
the Si mole fraction. x determines the band structure,
including the bandgap and the relative importance of the
indirect and direct processes. A rather abrupt transition C. Heterostructures
occurs from Ge-like to Si-like behavior above x=0.225 .
The control over the band structure through x can be The material options discussed so far assume that the
used to find a compromise between improving ION , and active region of the device consists of only one material.
reducing IOFF . For such a configuration, an important issue in the ma-
terial choice is the link between ION and IOFF , with the
desired improvement of the former also inducing an un-
B. Group III-V materials wanted increase in the latter. A way to decouple the two
is to combine two materials in a heterostructure, such
Beyond group IV, a host of materials suited for TFET that the material in the source is different from the ma-
can be found in the group of III-V materials, which are terial in the channel and the drain.
compounds of elements from group III and group V. A heterostructure configuration has an extra degree of
They are actively researched as future channel materi- freedom: the band edge alignment at the tunnel junction.
als for MOSFET, because they can exhibit high elec- The alignment is determined to a first approximation by
tron mobilities compared to Si66 . III-V materials show a the electron affinity of the two constituent materials, and
wide variety of band structures, with bandgaps varying is classified as straddled, staggered or broken, as shown
from 0.17 eV for InSb up to 3.28 eV for GaN67 . The in Fig. 13. The effective bandgap which determines the
same variety can be found in the effective masses. The tunnel path lengths at the tunnel junction is determined
possibility of a small bandgap and small effective mass, by the alignment and not by the bandgaps of the individ-
combined with the direct nature of most III-V materi- ual materials. A hetero-TFET can therefore combine a
als, makes them interesting candidates for improving the very small effective bandgap at the source-channel junc-
TFET ION . Important examples which have been studied tion, enabling a high ION , with a large bandgap at the
for TFET applications are binary compounds like InAs, channel-drain junction, maintaining a low IOFF . Fig. 14
GaSb, InP and ternary compounds like Inx Ga1−x As and illustrates the performance improvement for a staggered
GaAsx Sb1−x 7 . heterojunction compared to a homojunction. The mate-
Nevertheless, III-V materials also present significant rial of the source can also be chosen to have a large DOS,
challenges. Firstly, they are much less known than Si, associated with a large bandgap, to limit the source dop-
so it is challenging to obtain high-quality materials and ing degeneracy. This can be done without increasing the
gate dielectrics with a low defect density66 . Secondly, effective bandgap at the tunnel junction, so long as the
the material properties that allow for a high ION also band alignment remains favorable68 .
raise IOFF undesirably. A small bandgap combined with In the special case of a broken band alignment, there
a low effective mass facilitates both ambipolar and di- is no forbidden region through which the carrier has to
rect source-drain tunneling. This issue can be alleviated tunnel. This enables a high transmission probability and
with the use of a heterostructure as discussed in Sec- is hence beneficial for ION . It is still unclear, however,
 14

-3
10 E
c
Homo

Hetero 0.5

E (eV)
-5 V = -1V
10 V = 0.3V GS
m)

DS

E
v -0.5V E
I (A/

Fn
-7
10 0.0
E
Fp

c
e

e
/d

/d
-9
V

V
10
m

m
0 20 40
0

0
6

6
x (nm)
-11 (a) (b)
10
0.6 0.9 0.6 0.9
V (V)
FIG. 15. Energy band diagram at the tunnel junction of a
V (V) GS
GS
GaSb/InAs heterostructure TFET with a broken band align-
ment in the OFF (dark lines) and ON-state (light lines). VGS
FIG. 14. QM simulated transfer characteristics for a is -1 V and -0.5 V respectively. Indicated is a leakage path in
In0.53 Ga0.47 As (bandgap of 0.73 eV) homostructure TFET the OFF-state. The red wavy line represents a thermal step.
and a GaAs0.5 Sb0.5 /In0.53 Ga0.47 As heterostructure TFET. The dashed line represents the quasi Fermi level for the holes
The heterostructure is lattice matched and has a staggered in the source (EFp ). The source doping is 5x1019 cm−3 .
band alignment with an effective bandgap of 0.29 eV. The
TFET is a double gate configuration with an EOT of 0.6 nm,
a body thickness of 10 nm and a source doping of 5x1019 cm−3 .
(a) Unshifted and (b) shifted characteristics such that the VGS
at which IOFF is 1x10−11 A/µm coincides. This shift can be
accomplished by a proper choice of gate workfunction. abling excellent gate control. In addition, it is expected
that low defect densities can be obtained with no dan-
gling bonds or roughness at the surface.
whether a broken gap alignment will allow for a suffi-
ciently low IOFF , since phonon-assisted leakage paths ex-
ist in the OFF-state69,70 . This is indicated in Fig. 15: The most studied 2D-material is graphene, but since it
after transitioning from source to channel, a rethermal- is a semi-metal, it does not have a bandgap and is there-
ization step can enable the electron to pass over the po- fore not suited for TFET as such. However, a bandgap
tential barrier in the channel, resulting in an increased can be introduced by symmetry-breaking operations such
IOFF . The importance of this leakage current is influ- as patterning the graphene into nanoribbons, or by stack-
enced by the quantization in the triangular wells at the ing two layers and applying an electric field73 . This al-
junction and the interaction strength with the available lows the bandgap to be tuned, either by controlling the
phonons. size of the ribbon or the strength of the electric field.
Depending on the choice of materials, heterostructures
can be either lattice-matched or lattice-mismatched. Transition metal dichalcogenides (TMD) form an al-
In the first case, both materials have the same lat- ternative to graphene in 2D TFET applications. TMDs
tice constant. The materials can be grown epitax- are a class of 2D materials with chemical formula MX2 ,
ially without any stress at the heterojunction. No- with M a transition metal and X a chalcogenide. In con-
table examples include In0.53 Ga0.47 As/InP (straddled) trast to graphene, TMDs do exhibit a variety of relatively
and GaAs0.5 Sb0.5 /In0.53 Ga0.47 As (staggered). In the sec- large bandgaps. They can be used as the channel mate-
ond case, the mismatch of lattice constants results in a rial of an atomically thin TFET74–76 (see Fig. 16(a)), or
non-uniform stress profile arond the tunnel junction, as layers of different TMDs can be stacked as a heterostruc-
in the case of GaSb/InAs (broken). The stress has an ture, bonded by Van der Waals interactions. Tunneling
impact on the local band structure (see Section X). The then occurs in the overlap region of the layers, perpen-
build-up of stress can also result in interfacial defects at dicular to the gate (see Fig. 16(b)). Such 2D heterostruc-
the junction if relaxation sets in71 . ture TFETs are predicted to combine high ION and low
IOFF in the same way as bulk heterostructures77 . TMDs
can also be combined with a bulk semiconductor, such
D. 2D materials as MoS2 on Ge57 (see Fig. 16(c)). Challenges that re-
main for 2D materials are the development of large area
2D materials are an alternative to the bulk semi- defect-free production, good ohmic contacts and tech-
conductors described in the previous sections, and niques to obtain high doping levels72,78 . As a result of
are being investigated for both MOSFET and TFET these challenges, only few experimental 2D TFETs with
applications72 . 2D materials are atomically thin, en- sub-60 mV/dec SS have been realized (see Fig. 30).
 15

(a) (a) Gate (b) Gate

Top gate
n
TMD
Source Drain S S Tpo
p n i p i
Bottom gate
Tpo
Top gate
(b)
TMD2 FIG. 17. Source-channel region of a pocketed (a) point tun-
Drain neling and (b) line tunneling TFET configuration.
TMD1
Source

Bottom gate shift of the BTBT onset to lower VGS and a reduction of
the stretching effect on the SS, discussed in Section IV A.
Gate An optimum exists for the pocket thickness (Tpo , see
(c)
TMD Fig. 17(a)). The strength of the built-in field, and conse-
Bulk Drain quently the improvement in ION , increases with Tpo (see
Source Fig. 20(a)). However, if the pocket is so thick that the
concentration of free carriers in the pocket becomes large
(close to the doping level in the source) and is larger than
FIG. 16. TMD TFET configurations. (a) homojunction TMD the concentration in the channel at the onset of tunneling,
TFET74–76 , (b) 2D heterojunction TMD TFET, also called
the abrupt onset is deteriorated by a potential barrier in
a Thin-TFET77 and (c) 3D-2D heterojunction bilayer TMD
TFET, also called ATLAS-TFET57 . The arrows indicate the the channel82 (see Fig. 20(b)). At the first moment of
direction of the tunneling paths. cross-over between valence and conduction band, a dip
is present in the energy bands at the tunnel junction,
followed by a bump in the channel. For such a config-
VIII. DOPANT POCKETS uration, a further increase in VGS only marginally goes
toward increasing the band bending at the tunnel junc-
tion, but is consumed largely by a further increase of the
To improve SS and/or ION , another design option is to carrier concentration in the pocket. The structure is then
use dopant pockets. A dopant pocket is a localized region effectively a p-n diode in series with an n-i-n MOSFET. A
of doping, which locally modifies the device electrostat- good estimate for the optimal Tpo is the depletion width
ics. In a TFET, they can be used both in point tunneling at the tunnel junction79,82 . Note that for configurations
(Section VIII A) and line tunneling (Section VIII B) con- with a large body thickness, a thick pocket can cause an
figurations. uncontrolled tunneling current through the body of the
device.
A pocket can also be used to mitigate the deteriorat-
A. Pocketed pointTFET ing effects of a large source doping degeneracy on the SS.
The doping in the source is kept low, except for a pocket
In a pocketed pointTFET, a counterdoped pocket is at the source side of the source-channel junction. This
added at the interface between source and channel to pocket retains the built-in electric field, while the lower
improve both ION and SS (see Fig. 17(a)). This forms a p- doping in the rest of the source reduces the degeneracy83 .
n-i-n doping profile in an nTFET, and a n-p-i-p profile for SS is consequently improved, while ION is maintained.
a pTFET. Compared to a p-i junction, the built-in field The source pocket can be combined with a counterdoped
at a p-n junction is larger, corresponding to a stronger pocket at the channel side, resulting in a p− p+ n+ in con-
built-in band bending and hence an onset at lower VGS figuration. This approach is particularly useful for III-V
and the potential for a higher ION 79–82 . As shown in the pTFET in which the source consists of a material with a
band diagrams of Fig. 18, for a properly designed pocket, low conduction band DOS, as discussed in Section VI B.
the enhanced band bending induces an abrupt transition
from long to very short tunnel paths at tunneling onset,
compared to more gradual shortening seen in a standard B. Pocketed lineTFET
p-i-n/n-i-p configuration. In the transfer characteristics
this results in an improvement of SS, as shown in Fig. 19. In a pocketed lineTFET, a pocket is introduced to mit-
An additional effect of the pocket is that it reduces the igate the impact of FIQC and to reduce the sensitiv-
impact of FIQC on the pointTFET performance. The ity to the gate-channel overlap (see Section IV B). The
strong built-in electric field at the tunnel junction forces pocket is located in the source region, underneath the
the tunnel paths in a direction more parallel to the gate gate-source overlap and adjacent to the source-channel
and induces tunneling before a strong band bending to- junction (see Fig. 17(b)). Similar to the pocketed point
wards the gate, causing FIQC, is induced. The result is a TFET, the doping type of the pocket is opposite to that
 16

V =0.5V V =0.5V
1.0 DS 1.0 DS

E E E E
c c c c

0.5 V =0.3V 0.5

E (eV)

E (eV)
GS
0.3V
E E
v v
E T E T =6nm
v po v po
0.0
E 0.8V E
Fp Fp 0.8V 0.0
E E
Fp Fp 4nm
-0.5 V =0.5V
GS

-0.5
0.3V
-1.0
(a) No pocket (b) Pocket 0.5V
(a) (b)
-1.0
-20 0 20 40 -20 0 20 40 -20 0 20 40 -20 0 20 40
x (nm) x (nm) x (nm) x (nm)

FIG. 18. Energy band diagrams of a In0.53 Ga0.47 As TFET FIG. 20. Energy band diagrams of an In0.53 Ga0.47 As TFET
in the OFF (dark lines) and ON-state (light lines). VGS is around onset of BTBT. The source and pocket doping is
respectively 0.3 V and 0.8 V. (a) A no-pocket configuration 5x1019 cm−3 . The EOT is 0.6 nm, with a body thickness of
and (b) a configuration with a 4 nm thick pocket at the source- 20 nm. (a) Impact of increasing Tpo (0 nm, 2 nm and 4 nm)
channel junction as in Fig. 17(a). The source and pocket for a constant VGS of 0.5 V, showing an increasing electric
doping is 5x1019 cm−3 . The EOT is 0.6 nm, with a body field at the tunnel junction. (b) Comparison of an optimal
thickness of 20 nm. The dashed line represents the quasi Tpo of 4 nm to a larger Tpo of 6 nm at onset, with the latter
Fermi level for the holes in the source (EFp ). showing a potential bump in the channel. VGS is respectively
0.5 V and 0.3 V. The dashed line represents the quasi Fermi
level for the holes in the source (EFp ).
Homo

Hetero

-3 Homo pocket
10 V = 0.3V
DS
Hetero pocket

-5
10
m)

V =0.5V
DS

1.0 E
I (A/

E
-7 c c
0.9V 0V
10
c

c
e

0.5 V =1.4V 0.5V

E(eV)
/d

/d

GS
-9
V

10
m

E E
0

v v
6

(a) (b)
0.0
-11
E E
10 Fp Fp
0.6 0.9 0.6 0.9

V (V) V (V)
GS
GS
-0.5

FIG. 19. QM simulated transfer characteristics of Fig. 14

(a) No pocket (b) Pocket
compared to the transfer characteristics of the same config- -1.0
urations with a counterdoped pocket of 3 nm. The TFET 10 15 20 10 15 20
is a double gate configuration with an EOT of 0.6 nm, a z (nm) z (nm)
body thickness of 10 nm and a source and pocket doping
of 5x1019 cm−3 . (a) Unshifted and (b) shifted characteristics FIG. 21. Energy band diagrams of a In0.53 Ga0.47 As lineTFET
such that the VGS at which IOFF is 1x10−11 A/µm coincides. in the OFF (dark lines) and ON-state (light lines) along a
This shift can be accomplished by a proper choice of gate vertical cutline through the center of the gate. The source
workfunction. and pocket doping is 5x1019 cm−3 . The EOT is 0.6 nm, with
a body thickness of 20 nm. (a) A no-pocket configuration for
a VGS of 0.9 V and 1.4 V and (b) a configuration with a 4 nm
of the source. thick pocket underneath the gate as in Fig. 17(b) for a VGS
The effect on the FIQC underneath the gate dielectric of 0 V and 0.5 V. Indicated in brown are estimations of the
is shown by the energy bands of Fig. 21: the triangu- first three quantized energy levels inside the potential well for
lar well is rounded off and becomes wider for increasing the highest VGS . The dashed line represents the quasi Fermi
Tpo . The rounding reduces the FIQC, such that the sub- level for the holes in the source (EFp ).
band quantization levels decrease in energy. The onset of
vertical BTBT, determined by the first quantized level,
 17

therefore shifts to a lower VGS . The resulting onset volt-

age Vonset can be calculated analytically as84 : Line 3nm

Line 2nm
W F − χe + Esub q(Npo + Ns )Tpo 1
V onset = − + ... Point
q Cox Cox -7
V = 0.5V
s 10 DS

m)
 2 
q(Npo + 2Ns )Tpo
Eg + Esub
× 2qs Ns + − q 2 Ns2 Tpo2
q 2s

I (A/
-9
(27)

c
c
10

V/de
V/de
provided the pocket is fully depleted, with WF the work

60m
60m
function of the gate, χe the semiconductor electron affin-
ity, Npo and Ns respectively the doping level of the pocket -11 (a) (b)
10
and the source and Esub the energy level of the first sub- 0.0 0.8 0.0 0.8
V (V)
band in the potential well underneath the gate, which GS
V
GS
(V)

can be solved from a 1D Schrödinger equation. Eq. (27)

shows that by adjusting the thickness and doping of the FIG. 22. QM simulated transfer characteristics of a Si pock-
pocket for a given WF of the gate, the FIQC-induced eted pointTFET with a Tpo of 4 nm and pocketed lineTFET
shift in Vonset seen in the no-pocket case can be re- with two different Tpo . The TFETs are double gate config-
moved. Note that for materials which are more prone urations with an EOT of 0.6 nm, a body thickness of 20 nm
to FIQC, i.e. those with a small effective mass and hence and a source doping of 1x1020 cm−3 . (a) Unshifted and (b)
a large Esub , the required Tpo or Npo to compensate for shifted characteristics such that the VGS at which IOFF is
1x10−11 A/µm coincides. This shift can be accomplished by a
the FIQC shift are larger.
proper choice of gate workfunction.
In materials like Si, which have two distinct subband
ladders in the well underneath the gate, the decreased
FIQC has the added benefit of reducing the relative shift
between the subband levels. The corresponding kink in
-3 Line V = 0.3V
the SS which was discussed in Section IV B is thereby 10
Point
DS

also reduced, as shown in Fig. 22, where two pocketed

lineTFETs are compared to a pocketed pointTFET con-
m)

figuration. For a thicker pocket, Vonset is reduced, while -5

10
the relative shift of the two current components is de-
I (A/

creased. The kink is not present in a III-V configura-

tion with an isotropic conduction band effective mass, as c

c
de

de
-7
shown in Fig. 23. 10
V/

V/
The presence of the counterdoped pocket also re-
m

m
60

60
duces the unwanted lateral point tunneling component
by blocking the shortest lateral tunneling paths closest to -9 (a) (b)
10
the gate, since the source-channel p-i junction is locally 0.4 0.8 0.4 0.8

replaced by a n-i junction. Combined with the earlier V (V) V (V)

GS GS

onset of vertical BTBT, the overall impact of the lateral

component on the transfer characteristics is thus dimin- FIG. 23. QM simulated transfer characteristics of an
ished, removing also the variability to the gate alignment. In0.53 Ga0.47 As pocketed pointTFET with a Tpo of 3 nm and
The comparison of transfer characteristics in Fig. 22 a pocketed lineTFET with a Tpo of 6 nm. The TFETs are
and Fig. 23 shows that although the pocketed lineT- double gate configurations with an EOT of 0.6 nm, a body
FET outperforms the pocketed pointTFET in a Si con- thickness of 10 nm for the pointTFET and 40 nm for the
lineTFET and a source and pocket doping of 5x1019 cm−3 .
figuration in terms of ION and SS, the difference be-
(a) Unshifted and (b) shifted characteristics such that the
comes smaller for materials with a smaller bandgap, like VGS at which IOFF is 1x10−11 A/µm coincides. This shift can
In0.53 Ga0.47 As82,85 . be accomplished by a proper choice of gate workfunction.

IX. GATE STACK CONFIGURATION the EOT. A second aspect is the degree in which a given
variation in gate voltage moves the equipotential lines in
As discussed in Section VI, the gate control is an essen- the device closer together and hence increases the electric
tial factor in the BTBT efficiency. Gate control is char- field locally. By tuning both aspects, the gate control at
acterized by two aspects. The first aspect is the portion a tunnel junction can be enhanced or decreased.
of the gate voltage observable in the semiconductor, after To improve TFET performance for conventional logic
the voltage drop over the oxide. This is directly related to applications, it is advantageous to enhance the gate con-
 18

(a) Gate the electric field now increases both due to a larger total
voltage drop between source and gate, and because the
tunneling moves closer to the source where the equipoten-
Source p i n Drain
tial lines are more closely spaced, and the electric field is
higher. The result is a beneficial impact on the SS. This
Gate
effect is enhanced further for a MuG or GAA structure
(b) as illustrated in Fig. 26(a).
Further scaling of the body thickness has an additional
Source p i n Drain beneficial impact on the gate control. As the gates move
closer together, the curvature of the equipotential lines
becomes more pronounced, and the electric field hence
Gate
increases. As explained in the previous paragraph, this
(c)
benefits the SS of a pointTFET as illustrated in Fig. 27
for a pocketed point TFET. For a double gate lineTFET,
Source p i n Drain
the body scaling does not benefit performance above a
certain minimum body thickness, as the gate depletes the
Gate source locally underneath the gate and tunneling starts
from close to the undepleted source region (see Fig. 21),
FIG. 24. (a) Single gate, (b) double gate and (c) gate-all- which is at source voltage. Hence, the two depleted
around TFET configuration. source regions underneath the two gates are disconnected
in the lineTFET.
Although body scaling can improve performance, size-
trol at the source-channel tunnel junction (Section IX A), induced quantum confinement (SIQC) effects become
and to decrease it at the channel-drain junction (Sec- more important as the body thickness is scaled to di-
tion IX B). The former improves the TFET SS and ION , mensions comparable to the exciton Bohr radius of the
while the latter reduces the ambipolar current and CGD . material91,92 . SIQC results in a global increase of the
In this section, gate configurations addressing both re- material bandgap, and hence an increase in tunnel path
quirements are discussed. lengths. As discussed in Section VI, this can be detri-
mental for ION and can consequently undo some of the
advantage that comes with improved gate control. In
A. Improving source-channel BTBT particular, a GAA structure is more affected by SIQC
than a MuG or SG configuration as it is confined in two
A first way to improve the gate control at the source- directions. Fig. 26(a) shows this can even result in a
channel junction is to decrease the EOT, which increases smaller ION for the GAA configuration than for the DG
the electrostatic potential drop in the semiconductor. configuration. Similar to the case of FIQC, materials
Like a MOSFET, the TFET performance therefore bene- with small effective masses are more affected by SIQC.
fits from the introduction of high-k materials as the gate Specific care must be taken when scaling the body
dielectric86–88 . thickness of lineTFET or pointTFET configurations, to
A second way to improve the gate control at the prevent excessive depletion of the source region82,93 . Un-
source-channel junction is to go from a single-gate (SG) derneath the gate-source overlap, a depletion region ex-
to a multi-gate (MuG) to a gate-all-around (GAA) tends into the source, with a maximal extent in the ON-
configuration89,90 (see Fig. 24), an approach which is also state (as explained in Section III A). If the body of the
applied to MOSFET. This requires a transition from a device is too thin, the depletion regions of different gates
planar to a protruding structure. A multi-gate configu- touch even before the device reaches the ON-state, result-
ration can be implemented as a finFET, where the gate ing in depletion of the source over the full body thickness.
covers three sides of the device body. Going further, the For a lineTFET, this means no more tunnel paths per-
TFET body can be realized as a horizontal or vertical pendicular to the gate are available. Only much longer
nanowire (NW), with the gate wrapped around. non-orthogonal tunnel paths are available and the cur-
The tighter gate control can be seen when comparing rent drops accordingly (see Fig. 27). Since the lineTFET
the electrostatic profile of a single gate to that of a dou- operation depends on the gate-source overlap, the body
ble gate configuration89 (see Fig. 25). When a second thickness can therefore not be scaled below a certain crit-
gate is added, the equipotential lines are curved more ical thickness. This thickness is determined by the sum
strongly. A consequence is that the electric field close to of the depletion lengths underneath the gate-source over-
the source region is enhanced and hence the tunnel path laps in the ON-state. The depletion lengths themselves
lengths are reduced, improving ION . The electric field is are determined by the doping level of the source and of
now increasing towards the source region. Therefore, a the pockets, if present. In a pointTFET, depletion over
given increase in VGS results in a larger increase in elec- the full body thickness increases the tunnel path lengths,
tric field than in the single-gate configuration, because resulting in a drop in current for high VGS . In this case,
 19

20 point: line:
(a) (b)
30nm 30nm
0.85 V
z (nm)

20nm 20nm

10 -7 15nm 15nm
10 10nm 10nm

V = 0.5V

I (A/ m)
-0.5 V DS

0 Tbody

-9
FIG. 25. Electrostatic potential profiles at the source- 10
channel junction for a (a) single gate and (b) double gate
In0.53 Ga0.47 As p-i-n TFET at a VGS of 1 V and a VDS of
0.5 V. Six contour lines have been highlighted in each config-
uration.
-11
10
-1.0 -0.5 0.0 0.5
V (V)
GS

FIG. 27. QM simulated transfer characteristics of Si pocketed

point and lineTFETs for varying body thickness. The con-
figurations are as shown in Fig. 17, but have a double gate.
The gate-source overlap for the 10 nm body pointTFET con-
figuration is removed. For the lineTFET, the length of the
gate-source overlap and the pocket are 20 nm. The EOT is
0.6 nm. Source and pocket doping are 1x1020 cm−3 .

B. Reducing channel-drain BTBT and CGD

The ambipolar current can be reduced by introduc-

ing a gate-drain underlap64 (see Fig. 28(a)). The un-
derlap decreases the gate control on the channel-drain
junction. Since the desired BTBT takes place locally
at the source-channel junction, the gate does not have
to cover the full channel as in a MOSFET to enforce a
particular band bending and hence tunnel path length
at the source-channel junction. The underlap increases
the distance between the gate and the channel drain-
FIG. 26. (a) QM simulated transfer characteristics at VDS junction and consequently reduces the BTBT efficiency
= 0.2 V of InAs (line with triangles) SG ultra thin body at the drain. The short gate has the added benefit of re-
(UTB), (dashed line) DG UTB, and (gray line) GAA NW de- ducing CGD , which is responsible for the Miller effect (see
vices. Body thickness or diameter is 6 nm. The NW current Section XI C). Shortening of the gate can proceed down
is normalized with its diameter. The OFF-to-ON gate volt- to the length at which it loses control over the source-
age swing ∆-VGS is indicated. (b) CB and VB edges along channel junction (typically lengths of 10 nm and below).
the transport direction x of the 6 nm p-i-n SG UTB device Care has to be taken as well that the resistance in the
at the onset gate voltage. Λ indicates the tunneling barrier
ungated channel region does not surpass the resistance
width, ψS is the electrostatic potential in the middle of the
gate, and Ef l and Ef r denote the source and drain Fermi
of the tunnel barrier, to avoid an observable reduction in
levels, respectively. (c) Same as (b) for the 6 nm DG UTB the desired TFET current.
device. (d) Same as (b) and (c) for the 6 nm GAA NW de- Instead of removing part of the gate, an alternative is
vice. Current leakage paths are indicated by dashed arrows to make the gate heterogeneous, for instance by intro-
in (b), (c), and (d) 2009
c IEEE. Reprinted, with permission, ducing two different metals, each with a different work
from IEEE Elec. Dev. Lett. vol. 30, no. 6, pp. 602604, 2009. function94 (see Fig. 28(b)). This is also called a dual ma-
terial gate TFET. Although both metals are biased at
the same VGS , the work function at each of the tunnel
junctions (source-channel and channel-drain) determines
the amount of local band bending and hence the local
full depletion of the source can be avoided by reducing tunnel barrier. The two work functions can be optimized
or removing the gate-source overlap for thin body thick- separately: at the source-channel junction for a minimal
nesses. tunnel barrier in the ON-state and at the channel-drain
 20

(a) Gate B. Strain modeling

S p i n D
A common approximate model to study the effect of
strain on the band structure is the deformation potential
(b) WF 1 WF 2 model98 . The strain due to the deformation is assumed
to be small and gradually varying, such that the effect on
S p i n D the energy band edges can be treated as if it were a small
applied electrostatic potential. The shift in the band
(c) Gate edges then varies linearly with the strain components:
X
S p i n ∆En = Ξij εij (28)
D
ij

FIG. 28. Gate stack configurations for a p-i-n TFET aimed at with En the band edge of band n, εij the elements of the
reducing ambipolar behavior. (a) Short gate, (b) dual metal strain tensor and Ξij the so-called deformation potential
work functions and (c) dual EOT dielectric. constants, available in literature. The deformation po-
tential model can be directly applied in semiclassical ef-
fective mass-based simulators as a correction on the band
junction for a maximal tunnel barrier in the OFF-state. edges.
Strain can be introduced into more complex band
The same can be achieved by reducing the EOT only at structure models as well. In the k·p-model, strain can
the source-channel junction, while retaining a large EOT be incorporated by the addition of a strain interaction
at the channel-drain interface95 (see Fig. 28(c)). This can term as proposed by Pikus and Bir99 , which includes the
be implemented with different physical oxide thicknesses, deformation potential constants of Eq. (28). In the tight-
or with a high-k dielectric at the source side and a low-k binding approach, the integrals of Eq. (21) are adjusted
dielectric at the drain. based on the strain-induced change in bond lengths and
angles100 . The introduction of strain in these models ac-
counts for both the shift in band edge energies and the
change in effective masses. Strained band structure mod-
X. STRAIN els can be used in fully quantum mechanical simulations,
but are also useful to extract band edge energies and
effective masses, which serve as inputs to semiclassical
Similar to a MOSFET, mechanical strain impacts models101 .
TFET performance. This is a result of the effect
strain has on the band structure (Section X A), an ef-
fect which can be incorporated in TFET models (Sec- C. Strain in TFET
tion X B). Strain can be externally applied as a per-
formance booster, and is also intrinsically present at a Since TFET operation depends heavily on the band
lattice-mismatched heterojunction (Section X C). structure, strain affects the TFET performance. Strain
can be induced intentionally like in a MOSFET, e.g.
by external stressors, to improve ION by reducing EG
and mR 102 (see Section VI A). Similarly, the shifting
of the band edges due to strain can induce a more fa-
A. Impact of strain on the band structure
vorable band alignment at a heterojunction103 (see Sec-
tion VII C). Specifically in indirect materials, the relative
Mechanical strain can have an important effect on shift of conduction band valleys can render the material
the band structure of a semiconductor, since it alters direct104 . This is of particular interest in Ge, where the
the bond lengths and changes the crystal symmetry96 . difference between the indirect and direct bandgaps is
Firstly, strain warps the curvature of the energy bands, small.
and hence influences the effective masses. This effect is Under uniaxial or biaxial strain conditions, however,
exploited in MOSFETs to increase carrier mobility, by the valence band DOS is reduced due to the splitting of
choosing a strain configuration such that the effective heavy and light hole bands. This is a result of a reduc-
masses in the transport direction are decreased97 . Sec- tion of crystal symmetry. The smaller valence band DOS
ondly, strain shifts the relative positions of valley edges, increases the source doping degeneracy and can hence re-
such that the bandgap changes. Thirdly, as a result of sult in an SS degradation (see Section VI B). The degra-
broken symmetries, strain lifts certain energy band de- dation can be prevented with a lowly doped source region
generacies, which modifies the DOS. Most notably, strain in combination with a highly doped pocket at the tunnel
can lift the degeneracy between the heavy and light hole junction, as discussed in Section VIII A. Fig. 29 illus-
valence band. trates using k·p-based quantum mechanical simulations
 21

Hetero p-n-i-n VDS = 0.3V applications, both the source-channel and channel-drain
Hetero p-p-n-i-n strained junction have to be optimized to obtain similar ION and
10-3 VDS = 0.3V SS. This stands in contrast to conventional logic applica-
tions, in which the channel-drain tunneling is suppressed
(Section IX B). In digital circuits, the ambipolarity can
10-5
I (A/Pm)

be used to create more complex logic gates with fewer

G components105 . These ambipolar gates have control in-
-7 puts that determine the bias point, and hence the polar-
10 S p n i n D
ity of the devices. In analog applications, an ambipolar

c
G amplifier for example, exhibits both positive and nega-

/de
G
-9 tive small-signal gain, depending on the bias point106 .

mV
10 S p- p+ n i n D This allows it to function as both a common-source and

60
G common-drain amplifier, without changing the physical
(a) (b)
10-11 0.6 0.9 0.6 0.9
implementation.
VGS(V) VGS(V)
B. Asymmetrical structure
FIG. 29. QM simulated transfer characteristics of the un-
strained pocketed GaAs0.5 Sb0.5 /In0.53 Ga0.47 As pointTFET
of Fig. 19 compared with a p− p+ n+ in version of this TFET Unlike a MOSFET, the TFET has an asymmetrical
(inset). The latter TFET is under a uniform tensile biaxial source and drain doping and therefore behaves signifi-
strain of 500 MPa and has a p-type pocket of 4 nm and and cantly different for a positive or negative VDS polarity107 .
an n-type pocket of 3 nm, both with a doping of 5x1019 cm−3 . As the p-i-n diode is reverse biased in normal TFET oper-
The source doping is 5x1018 cm−3 . The TFET is a double gate ation, the sign of VDS is fixed to positive for nTFET and
configuration with an EOT of 0.6 nm and a body thickness negative for pTFET. When properly designed, the dom-
of 10 nm. The strain was included in the quantum mechan- inant current is BTBT-based with leakage mechanisms
ical simulations using a k·p-based model. (a) Unshifted and as described in Section VI C. Reversing the sign of VDS
(b) shifted characteristics such that the VGS at which IOFF is
brings the p-i-n diode in forward bias. At small forward
1x10−11 A/µm coincides. This shift can be accomplished by a
proper choice of gate workfunction.
bias, the current is typically negligible compared to the
current at similar reverse bias, except for designs with
sufficiently high source doping such that (Esaki-) tunnel-
ing becomes significant108 . Even then, a drop in current
that with this approach uniform strain can be used to
is typically present upon increasing VDS as the tunneling
further improve performance103 .
decreases, before the forward bias diode diffusion current
Strain can also be inherently present in a TFET, e.g.
reaches sufficiently high values to become dominant.
at the heterojunction of two lattice mismatched materi-
als. The strain profile in this case is very non-uniform, The asymmetrical TFET conduction can be a concern
making it difficult to predict the exact impact on the for Static Random Access Memory (SRAM) circuits. In
TFET performance. standard 6-transistor (6T) SRAM cells, the two access
transistors are required to conduct current in two direc-
tions. The asymmetrical conduction leads to unaccept-
able read and write static noise margins and hence in-
XI. CIRCUIT CONSIDERATIONS
stability of the SRAM cell109 . Alternative SRAM cell
designs with increased number of transistors have been
Certain TFET features require additional care when proposed to circumvent this issue109,110 .
designing circuits. The most important are the am- Weak conduction in one direction also prevents the dis-
bipolar effect (Section XI A), the asymmetrical structure charging of transient noise signals that appear on a circuit
(Section XI B), the Miller capacitance (Section XI C) and node through capacitive coupling111 . These signals can
the superlinear onset in the output characteristics (Sec- at times become very large and significantly surpass VDD
tion XI D). and therefore be detrimental to device reliability. Also
the circuit timing suffers from these excessive noise sig-
nals. To alleviate this issue, the forward Esaki tunneling
A. Ambipolar effect current present at low VDS can be used to evacuate the
excess charge. This requires a sufficiently high degener-
The TFET ambipolarity can result in an increased acy in the source.
switching leakage in a TFET inverter, but can just as Several other logic circuits, like the basic inverter, are
well be used to design more compact digital and ana- mostly unaffected by the asymmetry. It is even possible
log circuits if the positive and negative switching of for specific basic circuits to exploit the unidirectional-
the gate voltage to turn on the tunneling current is ity to redesign the circuit with fewer transistors than in
exploited105,106 . For good ambipolar behavior in these conventional CMOS, e.g. the pass-gate multiplexor111 .
 22

C. Miller capacitance is divided accordingly between the contacts. Improving

the source tunneling efficiency is therefore also beneficial
in reducing the Miller effect.
A high CGD creates an enhanced Miller effect, which
results in over-and undershoots in the transient behavior
of an inverter. The peak voltages VP of the over- and
D. Superlinear onset
undershoots can be expressed as112 :
As discussed in Section III C, the TFET output char-
CM
VP = VDD (29) acteristics can have a superlinear onset19–21 . The small
CM + CL current at low VDS delays the last part of the charging or
discharging of the load capacitance in a TFET inverter.
with CL the load capacitance of the inverter and with CM This increases the inverter rise and fall times and hence
comprising the CGD of both nTFET and pTFET of the increases the total circuit delay. Additionally, recovery
inverter. It is possible to decrease VP by increasing CL , from transient noise signals is less efficient when the out-
but this has the downside of also increasing the inverter put characteristics are superlinear114 . As suggested in
fall time delay. Care must therefore be taken in the device Section III C, the superlinear onset can be reduced by
design to keep CGD low in the first place. improving the BTBT efficiency at small VDS .
The gate capacitance division in gate-source capaci-
tance CGS and gate-drain capacitance CGD is different
from the MOSFET-case, because charge injection from XII. EXPERIMENTAL STATUS
the source requires a tunneling event, while charge in-
jection from the drain is via thermal injection, similar After describing the device physics and architecture
to the MOSFET-case. Therefore, when the tunneling ef- options in previous sections, this section is intended to
ficiency at the source side is comparatively low, and in give an overview of experimental realizations. The cur-
the absence of a strong SRH process, the carriers in the rent status of experimental TFET transfer characteris-
channel are supplied by the drain and the full channel tics is summarized in Fig. 3017,53,55–58,81,115–129 . The
determines CGD . CGS is then determined by the deple- curves have been selected to have a point-SS lower than
tion of the highly-doped source region, hence resulting in 70 mV/dec, which means they are either close to or below
a decreasing capacitance with increasing VGS (nTFET). the ideal MOSFET limit of 60 mV/dec for a given volt-
The channel charge is determined by the position of the age range. A variety of experimental techniques is used
drain Fermi level and the conduction band in the chan- to fabricate these structures8 . Some configurations can
nel (nTFET). A channel material with a lower DOS in be seen to obtain a sub-60 mV/dec SS, demonstrating
the conduction band (nTFET) hence results in a smaller the steep switching potential of TFET. None of the dis-
charge build-up in the channel and the corresponding played curves, however, attains all requirements for SS,
CGD is also smaller112 . Despite the fact that Cg ≈ CGD ION , IOFF and I60 as specified in Section VI, and which
if the tunneling efficiency at the source is low, the CGD are required to be a viable replacement for MOSFET.
can hence be limited if the TFET is properly designed. The comparison between nTFET and pTFET shows that
Also a gate-drain underlap can help to reduce the Miller fewer experimental realizations of pTFET exist, with a
effect113 . performance generally inferior to nTFET. The reasons
In cases where the tunneling efficiency is high and the behind this discrepancy are discussed in Section VI B. It
device is fully in the ON-state, a significant amount of is clear from Fig. 30 that a gap still needs to be closed
charge in the channel is supplied by the source. Like a between the promising theoretical predictions and actual
MOSFET, the TFET then exhibits a linear region, in experiment. Note, however, that some configurations
which charge is contributed equally by source and drain, which are promising in simulations, confined III-V het-
and a saturation (pinch-off) regime, in which charge is erostructure NW, are challenging to fabricate, and have
uniquely supplied by the source. The gate capacitance not yet been experimentally realized.

XIII. CONCLUSIONS AND OUTLOOK promising candidate to circumvent the increasing power
issues in scaled MOSFET technologies. The main chal-
lenges for TFET are to obtain a low average SS, a high
In this article, the TFET was discussed from the per- ION and low IOFF , or stated differently: a high I60 , and
spective of device physics. The basic operating princi- this for both nTFET and pTFET. Accurate predictive
ple based on BTBT enables the TFET to obtain a sub- models are required to meet these challenges, with quan-
60 mV/dec SS at room temperature. This makes it a
 23

2
10
I
60

0
10
m)

-2
10
A/
I (

-4
10
6

c
0

e
m

/d
V

V
/d

m
-6
e

10

0
c

6
-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
V (V)
GS

pTFET nTFET In 0.7 Ga 0.3 As SG | 0.2V [127]

Si SG | -1V [126] Si/InAs NW | 1V [53] Si | 0.6V [58]

Si NW GAA | -0.5V [116] In 0.53 Ga 0.47 As/In 0.7 Ga 0.3 As CNT MuG | -0.5V [17]
Si line MuG | -0.05V [56] line SG | 0.05V [55] InP/GaAs NW | -0.75V [128]
sSiGe SG | -0.1V [125] MoS 2/Ge | 0.1V [57] Si NW GAA | 0.1V [117]
Si MuG | -0.1V [124] Si SG | 0.05V [81]
Si/Ge SG | 0.5V [121]

CNT MuG | -0.01V [120]

sGe DG | 1V [119]

FIG. 30. Experimental transfer characteristics in literature of nTFET and pTFET. Only curves with an SS lower than
70 mV/dec have been included. The legend includes the material system (s-prefix stands for strained), gate configuration and
VDS . The legend entries are ordered according to the intersection of the corresponding curves at a VGS of -0.2 V and 0.2 V.
Also indicated is the target value for I60 mentioned in Section VI B. This figure is an updated version of Fig. 2 of Lu et al 7 .

tum mechanical simulators complementing more and XIV. LIST OF ABBREVIATIONS

more the semiclassical models typically used for MOS-
FET. These models can help in choosing the optimal ma- BGN: BandGap Narrowing
terial system, with alternative group IV materials, III-V BTBT: Band-To-Band-Tunneling
materials and 2D materials proving interesting alterna- CMOS: Complementary Metal-Oxide-Semicon-
tives for Si. Combining materials in a heterostructure ductor
provides an additional degree of freedom in the device DG: Double Gate
design. Other performance improvements can be gained DOS: Density Of States
from dopant pockets or from an optimization of the gate EF: Envelope Function
stack configuration. The presence of strain can have a EOT: Effective Oxide Thickness
beneficial effect on the device performance. Finally, sev- FIQC: Field-Induced Quantum Confinement
eral specific characteristics of the TFET operation at a GAA: Gate-All-Around
device level also have consequences for its use in circuits, MOSFET: Metal-Oxide-Semiconductor Field-Effect
suggesting the need for co-optimization of device and cir- Transistor
cuit. The gap between theoretical predictions and exper- MuG: Multi-Gate
imental realizations today shows that many challenges NEGF: Non-Equilibrium Green’s Function
are still to be overcome, before the TFET’s appearance NW: NanoWire
in low-power products. QM: Quantum Mechanical
QTBM: Quantum Transmitting Boundary Method
SS: SemiClassical
SIQC: Size-Induced Quantum Confinement
 24

SG: Single Gate 1 E. Mollick, “Establishing moore’s law,” IEEE Annals of the
SRAM: Static Random Access Memory History of Computing, vol. 28, pp. 62–75, July 2006.
2 G. E. Moore, “Cramming more components onto integrated cir-
SRH: Shockley-Read-Hall
cuits,” vol. 38, pp. 114–117, Jan. 1965.
SS: Subthreshold Swing 3 R. Aitken, V. Chandra, J. Myers, B. Sandhu, L. Shifren, and
TAT: Trap-Assisted Tunneling G. Yeric, “Device and technology implications of the internet
TFET: Tunnel Field-Effect Transistor of things,” in VLSI Technology (VLSI-Technology): Digest of
TMD: Transition Metal Dichalcogenide Technical Papers, 2014 Symposium on, pp. 1–4, IEEE, 2014.
4 W. Dehaene and A. Verhulst, “New devices for internet of
WF: Work Function things: A circuit level perspective,” in Electron Devices Meet-
WKB: Wentzel-Kramers-Brillouin ing (IEDM), 2015 IEEE International, pp. 25.5.1–25.5.4, Dec
2015.
5 A. M. Ionescu and H. Riel, “Tunnel field-effect transistors as

XV. REFERENCES energy-efficient electronic switches.,” Nature, vol. 479, pp. 329–
37, Nov. 2011.
6 A. C. Seabaugh and Q. Zhang, “Low-Voltage Tunnel Transistors

for Beyond CMOS Logic,” Proc. IEEE, vol. 98, pp. 2095–2110,
Dec. 2010.
7 H. Lu and A. Seabaugh, “Tunnel field-effect transistors: State-

of-the-art,” IEEE J. Electron Dev. Soc., vol. 2, pp. 44–49, July

2014.
8 L. Zhang and M. Chan, Tunneling Field Effect Transistor Tech-

nology, ch. 2. Springer, 2016.

9 N. S. Kim, T. Austin, D. Baauw, T. Mudge, K. Flautner, J. S.

Hu, M. J. Irwin, M. Kandemir, and V. Narayanan, “Leakage

current: Moore’s law meets static power,” Computer, vol. 36,
pp. 68–75, Dec 2003.
10 R. Dennard, V. Rideout, E. Bassous, and A. LeBlanc, “Design

of ion-implanted mosfet’s with very small physical dimensions,”

Solid-State Circuits, IEEE Journal of, vol. 9, pp. 256–268, Oct
1974.
11 J. R. Brews, “Subthreshold behavior of uniformly and nonuni-

formly doped long-channel mosfet,” IEEE Transactions on

Electron Devices, vol. 26, pp. 1282–1291, Sep 1979.
12 S. Salahuddin, , and S. Datta, “Use of negative capacitance to

provide voltage amplification for low power nanoscale devices,”

Nano Letters, vol. 8, no. 2, pp. 405–410, 2008.
13 K. Gopalakrishnan, P. Griffin, and J. Plummer, “I-MOS: a

novel semiconductor device with a subthreshold slope lower than

kT/q,” in Electron Devices Meeting, 2002. IEDM ’02. Interna-
tional, pp. 289 – 292, 2002.
14 V. Pott, H. Kam, R. Nathanael, J. Jeon, E. Alon, and T. J. K.

Liu, “Mechanical computing redux: Relays for integrated circuit

applications,” Proceedings of the IEEE, vol. 98, pp. 2076–2094,
Dec 2010.
15 J. J. Quinn, G. Kawamoto, and B. D. McCombe, “Subband

spectroscopy by surface channel tunneling,” Surface Science,

vol. 73, pp. 190–196, May 1978.
16 S. Banerjee, W. Richardson, J. Coleman, and A. Chatterjee,

“A new three-terminal tunnel device,” Electron Device Letters,

IEEE, vol. 8, pp. 347–349, Aug 1987.
17 J. Appenzeller, Y.-M. Lin, J. Knoch, and P. Avouris, “Band-

to-band tunneling in carbon nanotube field-effect transistors,”

Phys. Rev. Lett., vol. 93, p. 196805, Nov 2004.
18 K. K. Bhuwalka, J. Schulze, and I. Eisele, “Scaling the vertical

tunnel FET with tunnel bandgap modulation and gate work-

function engineering,” IEEE Transactions on Electron Devices,
vol. 52, pp. 909–917, May 2005.
19 B. Rajamohanan, D. Mohata, A. Ali, and S. Datta, “Insight into

the output characteristics of iii-v tunneling field effect transis-

tors,” Applied Physics Letters, vol. 102, no. 9, 2013.
20 A. S. Verhulst, D. Leonelli, R. Rooyackers, and G. Groe-

seneken, “Drain voltage dependent analytical model of tunnel

field-effect transistors,” Journal of Applied Physics, vol. 110,
no. 2, p. 024510, 2011.
21 L. D. Michielis, L. Lattanzio, and A. M. Ionescu, “Understand-

ing the superlinear onset of tunnel-fet output characteristic,”

IEEE Electron Device Letters, vol. 33, pp. 1523–1525, Nov 2012.
22 Q. Smets, D. Verreck, A. S. Verhulst, R. Rooyackers, C. Merck-

ling, M. Van De Put, E. Simoen, W. Vandervorst, N. Collaert,

 25

V. Y. Thean, B. Sorée, G. Groeseneken, and M. M. Heyns, vol. 57, pp. 6493–6507, mar 1998.
“InGaAs tunnel diodes for the calibration of semi-classical and 43 C. S. Lent and D. J. Kirkner, “The quantum transmitting
quantum mechanical band-to-band tunneling models,” J. Appl. boundary method.,” Journal of Applied Physics, vol. 67, no. 10,
Phys., vol. 115, pp. 184503–184503–9, May 2014. p. 6353, 1990.
23 W. G. Vandenberghe, B. Sorée, W. Magnus, G. Groeseneken, 44 R. J. LeVeque, Finite difference methods for ordinary and

and M. V. Fischetti, “Impact of field-induced quantum confine- partial differential equations: steady-state and time-dependent
ment in tunneling field-effect devices,” Applied Physics Letters, problems, vol. 98. Siam, 2007.
vol. 98, no. 14, p. 143503, 2011. 45 L. R. Ram-Mohan, Finite element and boundary element appli-
24 A. S. Verhulst, D. Verreck, M. A. Pourghaderi, M. Van de Put, cations in quantum mechanics, vol. 5. Oxford University Press
B. Sore, G. Groeseneken, N. Collaert, and A. V.-Y. Thean, “Can on Demand, 2002.
p-channel tunnel field-effect transistors perform as good as n- 46 L. N. Trefethen, Spectral methods in MATLAB, vol. 10. Siam,

channel?,” Applied Physics Letters, vol. 105, no. 4, p. 043103, 2000.

2014. 47 S. Datta, Quantum Transport: Atom to Transistor. Cambridge
25 K.-H. Kao, A. Verhulst, W. Vandenberghe, B. Soree, G. Groe- University Press, 2005.
seneken, and K. de Meyer, “Direct and indirect band-to-band 48 H. Ryu, H. Park, and M. Shin, “Feasibility, accuracy, and per-

tunneling in germanium-based tfets,” Electron Devices, IEEE formance of contact block reduction method for multi-band sim-
Transactions on, vol. 59, pp. 292–301, Feb 2012. ulations of ballistic quantum transport,” Journal of Applied
26 D. Griffiths, Introduction to Quantum Mechanics. Prentice Hall, Physics, vol. 111, no. 6, pp. 1–10, 2012.
1995. 49 M. Van de Put, “Band-to-band tunneling in III-V semiconduc-
27 E. O. Kane, “Theory of tunneling,” Journal of Applied Physics, tor heterostructures,” in EUROCON, 2013 IEEE, pp. 2133 –
vol. 32, no. 1, pp. 83–91, 1961. 2139, 2013.
28 J. T. Teherani, “Band-to-band tunneling in silicon diodes and 50 P. Long, E. Wilson, J. Z. Huang, G. Klimeck, M. J. W. Rodwell,

tunnel transistors,” Master thesis, Master of Science in Electri- and M. Povolotskyi, “Design and simulation of GaSb/InAs 2D
cal Engineering and Computer Science, Massachusetts Institute transmission-enhanced tunneling FETs,” IEEE Electron Device
of Technology, 2010. Letters, vol. 37, pp. 107–110, Jan 2016.
29 E. Kane, “Zener tunneling in semiconductors,” Journal of 51 Q. Zhang, W. Zhao, and A. Seabaugh, “Low-subthreshold-

Physics and Chemistry of Solids, vol. 12, no. 2, pp. 181–188, swing tunnel transistors,” IEEE Electron Device Letters, vol. 27,
1960. pp. 297–300, April 2006.
30 W. Vandenberghe, B. Sorée, W. Magnus, and G. Groeseneken, 52 W. G. Vandenberghe, A. S. Verhulst, B. Sorée, W. Magnus,

“Zener tunneling in semiconductors under nonuniform electric G. Groeseneken, Q. Smets, M. Heyns, and M. V. Fischetti, “Fig-
fields,” Journal of Applied Physics, vol. 107, no. 5, p. 054520, ure of merit for and identification of sub-60 mv/decade devices,”
2010. Applied Physics Letters, vol. 102, no. 1, p. 013510, 2013.
31 K. Ganapathi and S. Salahuddin, “Zener tunneling: Congru- 53 K. Tomioka, M. Yoshimura, and T. Fukui, “Steep-slope tunnel

ence between semi-classical and quantum ballistic formalisms,” field-effect transistors using IIIV nanowire/Si heterojunction,”
Journal of Applied Physics, vol. 111, no. 12, p. 124506, 2012. VLSI Technology (VLSIT), 2012 Symposium on, no. 2010,
32 C. Duke, Tunneling in solids. Solid state physics: Supplement, pp. 47–48, 2012.
Academic Press, 1969. 54 F. Mayer, C. L. Royer, J. F. Damlencourt, K. Romanjek, F. An-
33 Synopsys, Sentaurus Device User Guide, 2015.06. drieu, C. Tabone, B. Previtali, and S. Deleonibus, “Impact of
34 M. L. Van de Put, W. G. Vandenberghe, W. Magnus, and SOI, Si1-xGexOI and GeOI substrates on CMOS compatible
B. Sorée, “An envelope function formalism for lattice-matched tunnel FET performance,” in Electron Devices Meeting, 2008.
heterostructures,” Physica B: Condensed Matter, vol. 470471, IEDM 2008. IEEE International, pp. 1–5, Dec 2008.
pp. 69 – 75, 2015. 55 G. Dewey, B. Chu-Kung, J. Boardman, J. M. Fastenau,
35 W. Vandenberghe, B. Sorée, W. Magnus, and M. V. Fischetti, J. Kavalieros, R. Kotlyar, W. K. Liu, D. Lubyshev, M. Metz,
“Generalized phonon-assisted Zener tunneling in indirect semi- N. Mukherjee, P. Oakey, R. Pillarisetty, M. Radosavljevic,
conductors with non-uniform electric fields: A rigorous ap- H. W. Then, and R. Chau, “Fabrication, characterization, and
proach,” Journal of Applied Physics, vol. 109, no. 12, p. 124503, physics of IIIV heterojunction tunneling field-effect transistors
2011. (H-TFET) for steep sub-threshold swing,” Int. Electron Dev.
36 M. G. Burt, “An exact formulation of the envelope function Meet., vol. 3, pp. 33.6.1–33.6.4, Dec. 2011.
method for the determination of electronic states in semicon- 56 Y. Morita, T. Mori, S. Migita, W. Mizubayashi, A. Tanabe,

ductor microstructures,” Semiconductor Science and Technol- K. Fukuda, T. Matsukawa, K. Endo, S. O’uchi, Y. X. Liu,
ogy, vol. 3, no. 8, p. 739, 1988. M. Masahara, and H. Ota, “Performance enhancement of tunnel
37 J. C. Slater and G. F. Koster, “Simplified LCAO method for the field-effect transistors by synthetic electric field effect,” IEEE
periodic potential problem,” Phys. Rev., vol. 94, pp. 1498–1524, Electron Device Letters, vol. 35, pp. 792–794, July 2014.
Jun 1954. 57 D. Sarkar, X. Xie, W. Liu, W. Cao, J. Kang, Y. Gong, S. Krae-
38 M. P. Anantram, M. S. Lundstrom, and D. E. Nikonov, “Mod- mer, P. M. Ajayan, and K. Banerjee, “A subthermionic tunnel
eling of nanoscale devices,” Proceedings of the IEEE, vol. 96, field-effect transistor with an atomically thin channel,” Nature,
pp. 1511–1550, Sept 2008. vol. 526, no. 7571, pp. 91–95, 2015.
39 M. Luisier, A. Schenk, W. Fichtner, and G. Klimeck, “Atom- 58 Q. Huang, R. Huang, C. Wu, H. Zhu, C. Chen, J. Wang, L. Guo,

istic simulation of nanowires in the s p 3 d 5 s* tight-binding R. Wang, L. Ye, and Y. Wang, “Comprehensive performance re-
formalism: From boundary conditions to strain calculations,” assessment of TFETs with a novel design by gate and source en-
Physical Review B, vol. 74, no. 20, p. 205323, 2006. gineering from device/circuit perspective,” in Electron Devices
40 M. Cardona and F. H. Pollak, “Energy-band structure of ger- Meeting (IEDM), 2014 IEEE International, pp. 13.3.1–13.3.4,
manium and silicon: The k.p method,” Phys. Rev., vol. 142, Dec 2014.
pp. 530–543, Feb 1966. 59 M. Kim, Y. Wakabayashi, R. Nakane, M. Yokoyama, M. Tak-
41 S. B. Radhia, N. Fraj, I. Saidi, and K. Boujdaria, “The eight- enaka, and S. Takagi, “High ion/ioff ge-source ultrathin body
level k.p model for the conduction and valence bands of InAs, strained-soi tunnel fets,” in Electron Devices Meeting (IEDM),
InP, InSb,” Semicond. Sci. Technol., vol. 22, no. 4, p. 427, 2007. 2014 IEEE International, pp. 13.2.1–13.2.4, Dec 2014.
42 J.-M. Jancu, R. Scholz, F. Beltram, and F. Bassani, “Empir- 60 M. G. Pala and D. Esseni, “Interface Traps in InAs Nanowire

ical spds* tight-binding calculation for cubic semiconductors: Tunnel-FETs and MOSFETs - Part I: Model Description and
General method and material parameters,” Physical Review B, Single Trap Analysis in Tunnel-FETs,” IEEE Transactions on
 26

Electron Devices, vol. 60, pp. 2795–2801, Sept 2013. 80 A. Tura, Z. Zhang, P. Liu, Y.-H. Xie, and J. Woo, “Vertical
61 S. Mookerjea, D. Mohata, T. Mayer, V. Narayanan, and silicon p-n-p-n tunnel nMOSFET with MBE-grown tunneling
S. Datta, “Temperature-dependent I - V characteristics of a junction,” Electron Devices, IEEE Transactions on, vol. 58,
vertical In0.53 Ga0.47 As tunnel FET,” IEEE Electron Device pp. 1907–1913, July 2011.
Letters, vol. 31, pp. 564–566, June 2010. 81 Q. Huang, R. Huang, Z. Zhan, Y. Qiu, W. Jiang, C. Wu, and
62 U. E. Avci, B. Chu-Kung, A. Agrawal, G. Dewey, V. Le, R. Rios, Y. Wang, “A novel Si tunnel FET with 36mV/dec subthreshold
D. H. Morris, S. Hasan, R. Kotlyar, J. Kavalieros, and I. A. slope based on junction depleted-modulation through striped
Young, “Study of TFET non-ideality effects for determination of gate configuration,” in Electron Devices Meeting (IEDM), 2012
geometry and defect density requirements for sub-60mV/dec Ge IEEE International, pp. 8.5.1–8.5.4, Dec 2012.
TFET,” in 2015 IEEE International Electron Devices Meeting 82 D. Verreck, A. Verhulst, K.-H. Kao, W. Vandenberghe,

(IEDM), pp. 34.5.1–34.5.4, Dec 2015. K. De Meyer, and G. Groeseneken, “Quantum mechanical per-
63 A. S. Verhulst, D. Verreck, Q. Smets, K. H. Kao, M. V. de Put, formance predictions of p-n-i-n versus pocketed line tunnel field-
R. Rooyackers, B. Sore, A. Vandooren, K. D. Meyer, G. Groe- effect transistors,” IEEE Trans. Electron Dev., vol. 60, no. 7,
seneken, M. M. Heyns, A. Mocuta, N. Collaert, and A. V. Y. pp. 2128–2134, 2013.
Thean, “Perspective of tunnel-FET for future low-power tech- 83 D. Verreck, A. S. Verhulst, B. Sore, N. Collaert, A. Mocuta,

nology nodes,” in Electron Devices Meeting (IEDM), 2014 A. Thean, and G. Groeseneken, “Improved source design for p-
IEEE International, pp. 30.2.1–30.2.4, Dec 2014. type tunnel field-effect transistors: Towards truly complemen-
64 A. S. Verhulst, W. G. Vandenberghe, K. Maex, and G. Groe- tary logic,” Applied Physics Letters, vol. 105, no. 24, 2014.
seneken, “Tunnel field-effect transistor without gate-drain over- 84 K.-H. Kao, A. Verhulst, W. Vandenberghe, and K. De Meyer,

lap,” Applied Physics Letters, vol. 91, no. 5, p. 053102, 2007. “Counterdoped pocket thickness optimization of gate-on-source-
65 J. V. Morgan and E. O. Kane, “Observation of direct tunneling only tunnel fets,” Electron Devices, IEEE Transactions on,
in germanium,” Phys. Rev. Lett., vol. 3, pp. 466–468, Nov 1959. vol. 60, pp. 6–12, Jan 2013.
66 J. A. Del Alamo, “Nanometre-scale electronics with III-V com- 85 D. Verreck, A. S. Verhulst, M. Van de Put, B. Sorée, W. Magnus,

pound semiconductors,” Nature, vol. 479, no. 7373, pp. 317–323, A. Mocuta, N. Collaert, A. Thean, and G. Groeseneken, “Full-
2011. zone spectral envelope function formalism for the optimization
67 Y. A. Goldberg, N. M. Shmidt, and Y. A. Vul, Handbook Series of line and point tunnel field-effect transistors,” Journal of Ap-
of Semiconductor Parameters, vol. 2, ch. 3,5. 1999. plied Physics, vol. 118, no. 13, p. 134502, 2015.
68 W. G. Vandenberghe, A. S. Verhulst, K.-H. Kao, K. D. Meyer, 86 K. Boucart and A. Ionescu, “Double-gate tunnel FET with

B. Sorée, W. Magnus, and G. Groeseneken, “A model deter- high-k gate dielectric,” Electron Devices, IEEE Transactions
mining optimal doping concentration and materials band gap on, vol. 54, pp. 1725–1733, July 2007.
of tunnel field-effect transistors,” Appl. Phys. Lett., vol. 100, 87 D. Leonelli, A. Vandooren, R. Rooyackers, S. De Gendt,

no. 19, p. 193509, 2012. M. Heyns, and G. Groeseneken, “Drive current enhancement in
69 S. Koswatta, “On the possibility of obtaining MOSFET-like p-tunnel fets by optimization of the process conditions,” Solid-
performance and sub-60-mV/dec swing in 1-D broken-gap tun- State Electronics, vol. 65, pp. 28–32, 2011.
nel transistors,” IEEE Trans. Electron Dev., vol. 57, no. 12, 88 R. Rooyackers, A. Vandooren, A. S. Verhulst, A. M. Walke,

pp. 3222–3230, 2010. E. Simoen, K. Devriendt, S. Lo-Corotondo, M. Demand,

70 J. Knoch and J. Appenzeller, “Modeling of high-performance G. Bryce, R. Loo, A. Hikavyy, T. Vandeweyer, C. Huyghebaert,
p-type IIIV heterojunction tunnel FETs,” IEEE Electron Dev. N. Collaert, and A. V. Y. Thean, “Ge-source vertical tunnel fets
Lett., vol. 31, no. 4, pp. 305–307, 2010. using a novel replacement-source integration scheme,” Electron
71 P. Yu and M. Cardona, Fundamentals of Semiconductors. Devices, IEEE Transactions on, vol. 61, no. 12, pp. 4032–4039,
No. 46-48, Springer, 2005. 2014.
72 D. Jena, “Tunneling transistors based on graphene and 2-d crys- 89 A. S. Verhulst, B. Sore, D. Leonelli, W. G. Vandenberghe, and

tals,” Proceedings of the IEEE, vol. 101, pp. 1585–1602, July G. Groeseneken, “Modeling the single-gate, double-gate, and
2013. gate-all-around tunnel field-effect transistor,” Journal of Ap-
73 Y. Zhang, T.-T. Tang, C. Girit, Z. Hao, M. C. Martin, A. Zettl, plied Physics, vol. 107, no. 2, p. 024518, 2010.
M. F. Crommie, Y. R. Shen, and F. Wang, “Direct observa- 90 M. Luisier and G. Klimeck, “Atomistic Full-Band Design Study

tion of a widely tunable bandgap in bilayer graphene.,” Nature, of InAs,” vol. 30, no. 6, pp. 602–604, 2009.
vol. 459, pp. 820–3, June 2009. 91 H. Carrillo-Nuñez, a. Ziegler, M. Luisier, and a. Schenk, “Mod-
74 H. Ilatikhameneh, Y. Tan, B. Novakovic, G. Klimeck, R. Rah- eling direct band-to-band tunneling: From bulk to quantum-
man, and J. Appenzeller, “Tunnel field-effect transistors in 2-d confined semiconductor devices,” Journal of Applied Physics,
transition metal dichalcogenide materials,” IEEE Journal on vol. 117, p. 234501, June 2015.
Exploratory Solid-State Computational Devices and Circuits, 92 D. Verreck, M. Van de Put, B. Sorée, A. S. Verhulst, W. Mag-

vol. 1, pp. 12–18, Dec 2015. nus, W. G. Vandenberghe, N. Collaert, A. Thean, and G. Groe-
75 Q. Zhang, G. Iannaccone, and G. Fiori, “Two-dimensional tun- seneken, “Quantum mechanical solver for confined heterostruc-
nel transistors based on thin film,” IEEE Electron Device Let- ture tunnel field-effect transistors,” Journal of Applied Physics,
ters, vol. 35, no. 1, pp. 129–131, 2014. vol. 115, no. 5, p. 053706, 2014.
76 R. K. Ghosh and S. Mahapatra, “Monolayer transition metal 93 K. Ganapathi, Y. Yoon, and S. Salahuddin, “Analysis of InAs

dichalcogenide channel-based tunnel transistor,” IEEE Journal vertical and lateral band-to-band tunneling transistors: Lever-
of the Electron Devices Society, vol. 1, pp. 175–180, Oct 2013. aging vertical tunneling for improved performance,” Applied
77 M. Li, D. Esseni, J. Nahas, D. Jena, and H. Xing, “Two- Physics Letters, vol. 97, no. 3, p. 033504, 2010.
dimensional heterojunction interlayer tunneling field effect tran- 94 S. Saurabh and M. Kumar, “Novel attributes of a dual material

sistors (Thin-TFETs),” Electron Devices Society, IEEE Journal gate nanoscale tunnel field-effect transistor,” Electron Devices,
of the, vol. 3, pp. 200–207, May 2015. IEEE Transactions on, vol. 58, pp. 404–410, Feb 2011.
78 F. Xia, V. Perebeinos, Y.-m. Lin, Y. Wu, and P. Avouris, “The 95 W. Y. Choi and W. Lee, “Hetero-gate-dielectric tunneling field-

origins and limits of metal-graphene junction resistance,” Na- effect transistors,” Electron Devices, IEEE Transactions on,
ture nanotechnology, vol. 6, no. 3, pp. 179–184, 2011. vol. 57, pp. 2317–2319, Sept. 2010.
79 V. Nagavarapu, R. Jhaveri, and J. Woo, “The tunnel source 96 Y. Sun, S. E. Thompson, and T. Nishida, “Physics of strain

(pnpn) n-mosfet: A novel high performance transistor,” Elec- effects in semiconductors and metal-oxide-semiconductor field-
tron Devices, IEEE Transactions on, vol. 55, pp. 1013–1019, effect transistors,” Journal of Applied Physics, vol. 101, no. 10,
April 2008. p. 104503, 2007.
 27

97 M. Chu, Y. Sun, U. Aghoram, and S. E. Thompson, “Strain: A 115 A. C. Ford, C. W. Yeung, S. Chuang, H. S. Kim, E. Plis, S. Kr-
Solution for Higher Carrier Mobility in Nanoscale MOSFETs,” ishna, C. Hu, and A. Javey, “Ultrathin body InAs tunneling
Annual Review of Materials Research, vol. 39, pp. 203–229, field-effect transistors on Si substrates,” Applied Physics Let-
2009. ters, vol. 98, no. 11, p. 113105, 2011.
98 J. Bardeen and W. Shockley, “Deformation potentials and mo- 116 R. Gandhi, Z. Chen, N. Singh, K. Banerjee, and S. Lee, “CMOS-

bilities in non-polar crystals,” Phys. Rev., vol. 80, pp. 72–80, compatible vertical-silicon-nanowire gate-all-around p-type tun-
Oct 1950. neling FETs with-mV/decade subthreshold swing,” IEEE Elec-
99 G. L. Bir, G. E. Pikus, P. Shelnitz, and D. Louvish, Symmetry tron Device Letters, vol. 32, no. 11, pp. 1504–1506, 2011.
and strain-induced effects in semiconductors, vol. 624. Wiley 117 R. Gandhi, Z. Chen, N. Singh, K. Banerjee, and S. Lee, “Ver-

New York, 1974. tical Si-nanowire n-type tunneling FETs with low subthreshold
100 W. A. Harrison, Electronic structure and the properties of swing (≤ 50 mV/decade ) at room temperature,” IEEE Electron
solids: the physics of the chemical bond. Courier Corporation, Device Lett., vol. 32, no. 4, pp. 437–439, 2011.
2012. 118 R. Li, Y. Lu, G. Zhou, Q. Liu, S. D. Chae, T. Vasen, W. S.
101 K.-H. Kao, A. S. Verhulst, M. Van de Put, W. G. Vandenberghe, Hwang, Q. Zhang, P. Fay, T. Kosel, M. Wistey, H. Xing, and
B. Sorée, W. Magnus, and K. De Meyer, “Tensile strained Ge A. Seabaugh, “AlGaSb/InAs tunnel field-effect transistor with
tunnel field-effect transistors: k.p material modeling and numer- on-current of 78 at 0.5 V,” IEEE electron device letters, vol. 33,
ical device simulation,” Journal of Applied Physics, vol. 115, no. 3, pp. 363–365, 2012.
no. 4, 2014. 119 T. Krishnamohan, D. Kim, S. Raghunathan, and K. Saraswat,
102 F. Conzatti, M. G. Pala, D. Esseni, E. Bano, and L. Selmi, “Double-gate strained-Ge heterostructure tunneling FET
“Strain-Induced Performance Improvements in InAs Nanowire (TFET) with record high drive currents and ¡ 60mV/dec sub-
Tunnel FETs,” IEEE Transactions on Electron Devices, vol. 59, threshold slope,” in 2008 IEEE International Electron Devices
pp. 2085–2092, Aug. 2012. Meeting, pp. 1–3, Dec 2008.
103 D. Verreck, A. S. Verhulst, M. L. V. de Put, B. Sore, N. Collaert, 120 Y. Lu, S. Bangsaruntip, X. Wang, L. Zhang, Y. Nishi, and

A. Mocuta, A. Thean, and G. Groeseneken, “Uniform strain in H. Dai, “DNA functionalization of carbon nanotubes for ul-
heterostructure tunnel field-effect transistors,” IEEE Electron trathin atomic layer deposition of high-k dielectrics for nan-
Device Letters, vol. 37, pp. 337–340, March 2016. otube transistors with 60 mV/decade switching,” Journal of
104 S. Wirths, A. Tiedemann, Z. Ikonic, P. Harrison, B. Holländer, the American Chemical Society, vol. 128, no. 11, pp. 3518–3519,
T. Stoica, G. Mussler, M. Myronov, J. Hartmann, 2006.
D. Grützmacher, D. Buca, and S. Mantl, “Band engineer- 121 S. H. Kim, H. Kam, C. Hu, and T. J. K. Liu, “Germanium-

ing and growth of tensile strained ge/(si) gesn heterostructures source tunnel field effect transistors with record high
for tunnel field effect transistors,” Applied physics letters, ION/IOFF,” in 2009 Symposium on VLSI Technology, pp. 178–
vol. 102, no. 19, p. 192103, 2013. 179, June 2009.
105 M. H. Ben-Jamaa, K. Mohanram, and G. D. Micheli, “An ef- 122 S. Mookerjea, D. Mohata, R. Krishnan, J. Singh, A. Val-

ficient gate library for ambipolar cntfet logic,” IEEE Trans- lett, A. Ali, T. Mayer, V. Narayanan, D. Schlom, A. Liu,
actions on Computer-Aided Design of Integrated Circuits and and S. Datta, “Experimental demonstration of 100nm channel
Systems, vol. 30, pp. 242–255, Feb 2011. length In0.53Ga0.47As-based vertical inter-band tunnel field ef-
106 X. Yang and K. Mohanram, “Ambipolar electronics,” Rice Uni- fect transistors (TFETs) for ultra low-power logic and SRAM
versity TREE1002, 2010. applications,” in 2009 IEEE International Electron Devices
107 U. Avci, D. Morris, and I. Young, “Tunnel field-effect transis- Meeting (IEDM), pp. 1–3, Dec 2009.
tors: Prospects and challenges,” Electron Devices Society, IEEE 123 G. Zhou, R. Li, T. Vasen, M. Qi, S. Chae, Y. Lu, Q. Zhang,

Journal of the, vol. 3, pp. 88–95, May 2015. H. Zhu, J. M. Kuo, T. Kosel, M. Wistey, P. Fay, A. Seabaugh,
108 L. Esaki, “New phenomenon in narrow germanium p- n junc- and H. Xing, “Novel gate-recessed vertical InAs/GaSb TFETs
tions,” Physical review, vol. 109, no. 2, p. 603, 1958. with record high ION of 180 uA/ um at VDS = 0.5 V,” in
109 D. Kim, Y. Lee, J. Cai, I. Lauer, L. Chang, S. J. Koester, Electron Devices Meeting (IEDM), 2012 IEEE International,
D. Sylvester, and D. Blaauw, “Low power circuit design based pp. 32.6.1–32.6.4, Dec 2012.
on heterojunction tunneling transistors (HETTs),” in Proceed- 124 D. Leonelli, A. Vandooren, R. Rooyackers, A. S. Verhulst,

ings of the 2009 ACM/IEEE international symposium on Low S. De Gendt, M. M. Heyns, and G. Groeseneken, “Performance
power electronics and design, pp. 219–224, ACM, 2009. enhancement in multi gate tunneling field effect transistors by
110 V. Saripalli, S. Datta, V. Narayanan, and J. P. Kulkarni, scaling the fin-width,” Japanese Journal of Applied Physics,
“Variation-tolerant ultra low-power heterojunction tunnel fet vol. 49, no. 4S, p. 04DC10, 2010.
sram design,” in Proceedings of the 2011 IEEE/ACM Interna- 125 A. Villalon, C. Le Royer, M. Cassé, D. Cooper, B. Prévitali,

tional Symposium on Nanoscale Architectures, pp. 45–52, IEEE C. Tabone, J.-M. Hartmann, P. Perreau, P. Rivallin, J.-F. Dam-
Computer Society, 2011. lencourt, et al., “Strained tunnel FETs with record ION: first
111 D. Morris, U. Avci, R. Rios, and I. Young, “Design of low voltage demonstration of ETSOI TFETs with SiGe channel and RSD,”
tunneling-fet logic circuits considering asymmetric conduction in VLSI technology (VLSIT), 2012 Symposium on, pp. 49–50,
characteristics,” Emerging and Selected Topics in Circuits and IEEE, 2012.
Systems, IEEE Journal on, vol. 4, pp. 380–388, Dec. 2014. 126 K. Jeon, W. Y. Loh, P. Patel, C. Y. Kang, J. Oh, A. Bowonder,
112 S. Mookerjea, R. Krishnan, S. Datta, and V. Narayanan, “On C. Park, C. S. Park, C. Smith, P. Majhi, H. H. Tseng, R. Jammy,
enhanced miller capacitance effect in interband tunnel transis- T. J. K. Liu, and C. Hu, “Si tunnel transistors with a novel
tors,” Electron Device Letters, IEEE, vol. 30, pp. 1102–1104, silicided source and 46mV/dec swing,” in 2010 Symposium on
Oct 2009. VLSI Technology, pp. 121–122, June 2010.
113 J. Zhuge, A. S. Verhulst, W. G. Vandenberghe, W. Dehaene, 127 A. Alian, J. Franco, A. Vandooren, Y. Mols, A. Verhulst, S. E.

R. Huang, Y. Wang, and G. Groeseneken, “Digital-circuit analy- Kazzi, R. Rooyackers, D. Verreck, Q. Smets, A. Mocuta, N. Col-
sis of short-gate tunnel fets for low-voltage applications,” Semi- laert, D. Lin, and A. Thean, “Record performance InGaAs
conductor Science and Technology, vol. 26, no. 8, p. 085001, homo-junction TFET with superior SS reliability over MOS-
2011. FET,” in 2015 IEEE International Electron Devices Meeting
114 N. Dagtekin and A. M. Ionescu, “Impact of super-linear onset, (IEDM), pp. 31.7.1–31.7.4, Dec 2015.
off-region due to uni-directional conductance and dominant Cgd 128 B. Ganjipour, J. Wallentin, M. T. Borgstrom, L. Samuelson,

on performance of TFET-based circuits,” IEEE Journal of the and C. Thelander, “Tunnel field-effect transistors based on InP-
Electron Devices Society, vol. 3, pp. 233–239, May 2015. GaAs heterostructure nanowires,” ACS nano, vol. 6, no. 4,
 28

pp. 3109–3113, 2012. “High current density InAsSb/GaSb tunnel field effect transis-
129 A.W. Dey, B. M. Borg, B. Ganjipour, M. Ek, K. A. Dick, tors,” in Device Research Conference (DRC), 2012 70th An-
E. Lind, P. Nilsson, C. Thelander, and L. E. Wernersson, nual, pp. 205–206, June 2012.