Tesi

POLITECNICO DI TORINO
Corso di Laurea Magistrale in Ingegneria Elettronica
Tesi di Laurea Magistrale
Analysis and Simulation of

Emerging FET Devices:
FinFET, TFET
Relatori:
Prof. Gianluca Piccinini
Prof. Mariagrazia Graziano
Prof. Marco Vacca
Candidata:
Deborah Vergallo
Aprile 2018
Summary
In this thesis the electrostatic analysis of three-dimensional FinFETs was done

in detail. Taking all main issues of such devices in mind, new emerging devices like
Tunnel FET and sandwich tunnel barrier FET (STBFET) were investigated.
Firstly, different analytical models for long channel and short channel double-
gate (DG) FinFETs were simulated with MATLAB, focusing on their reliability.
In particular, TCAD Synopsys Sentaurus simulations were used to validate the
correctness among the analytical models that were implemented. Then, the most
reliable MATLAB model for DG FinFET was succesfully extended to triple-gate
(TG) and trapezoidal TG FinFETs. Finally, a capacitance analysis for TG FinFET
was carried out.
Secondly, several analytical models for Tunnel FET were implemented in MAT-
LAB. One among them was found as the most reliable and valid for different transis-
tor sizes, so analytical simulations in MATLAB have been implemented for different
structures of Tunnel FET, such as single-gate (SG), double-gate(DG) and gate-all-
around (GAA) TFETs. To verify the accuracy of such models, physical simulations
in TCAD Synopsys Sentaurus of Tunnel FET devices have been done, focusing on
cylindrical gate all around (GAA) structures. An electrostatic analysis, a capaci-
tance analysis and a temperature analysis was carried out in order to compare the
behaviour of both GAA FET and GAA TFET.
To enhance the behaviour of TFET in the ON-state, a study of material depen-

dence was done. A Si/Ge nanowire hetero-structure Tunnel FET was studied and
simulated in TCAD Synopsis Sentaurus. In fact, a Germanium source enables the
reduction of the barrier width and henance tunneling of electrons from source to
drain.
Another solution to let Tunnel FET be an applicable device is to modify the struc-
ture, such as in sandwich tunnel barrier FET (STBFET). In such device, the source
is sandwiched in the substrate and that allows to switch the tunneling from lateral
to vertical direction in the ON-state. STBFET presents a higher current and a
steeper slope, still mantaining a low leakage current.
I
Table of contents
Summary I
1 Introduction 1
1.1 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 FinFET 5
2.1 DG FinFET Analytic Models . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Taur Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Matlab Implementation . . . . . . . . . . . . . . . . . . . . . 10
2.3 Fasarakis Analytical Model . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 TG FinFET Analytic Models . . . . . . . . . . . . . . . . . . . . . . 20
3 Physical Simulations of FinFET 24

3.1 Double-Gate FinFET Simulations . . . . . . . . . . . . . . . . . . . . 24
3.2 Gate-All-Around FET Simulations . . . . . . . . . . . . . . . . . . . 29
3.2.1 Nanowires FET Simulations . . . . . . . . . . . . . . . . . . . 33
3.3 FinFET Main Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.1 Advantages and Drawbacks . . . . . . . . . . . . . . . . . . . 35
4 Tunnel Field Effect Transistors 38

4.1 Band-to-Band Tunneling in p-i-n Diodes . . . . . . . . . . . . . . . . 39
4.1.1 Physical Simulations of Esaki Diode . . . . . . . . . . . . . . . 42
4.2 Comparison between MOSFET and TFET . . . . . . . . . . . . . . . 43
4.2.1 Temperature Analysis . . . . . . . . . . . . . . . . . . . . . . 48
4.3 Compact Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4 Hao Lu Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4.1 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.5 Praveen Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.5.1 Matlab Implemantation . . . . . . . . . . . . . . . . . . . . . 60
4.6 Dash Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.6.1 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
II
5 Sentaurus Simulations 66
5.1 Single Gate Tunnel FET . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 Double Gate Tunnel FET . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3 Gate All Around Tunnel FET . . . . . . . . . . . . . . . . . . . . . . 81
5.3.1 NW TFET . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.4 GAA FET vs GAA TFET . . . . . . . . . . . . . . . . . . . . . . . . 85
6 Tunnel FET issues and solutions 87

6.1 Si/Ge Hetero-structure Tunnel Field Effect Transistor . . . . . . . . . 89
6.1.1 Physical simulation of Si/Ge heterostructure TFET . . . . . . 90
6.2 Sandwich Tunnel Barrier FET . . . . . . . . . . . . . . . . . . . . . . 93
6.2.1 Physical simulation of STBFET . . . . . . . . . . . . . . . . . 93
7 Conclusions 96
7.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Bibliography 98
III
List of tables
2.1 Input data used for implementation of Taur model [1] . . . . . . . . . 10

2.2 Current values varying gate length . . . . . . . . . . . . . . . . . . . 11
2.3 Input data used for implementation of Fasarakis model [6] . . . . . . 18
2.4 Parameters values varying gate length . . . . . . . . . . . . . . . . . . 19
2.5 Input data used for implementation of Fasarakis model of Tz-TG and
Re-TG FiNFET (taken from [5]) . . . . . . . . . . . . . . . . . . . . 21
2.6 Electrical parameters extracted from simulation . . . . . . . . . . . . 22
2.7 Input data used for implementation of Fasarakis model of Tz-TG
FiNFET (taken from [5]) . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.8 Parameters values of a Tz-TG FinFET simulated in Matlab . . . . . 23
3.1 Input data used for implementation of Fasarakis model [6] . . . . . . 24
3.2 Important parameters extrcted from TCAD Synopsis Sentaurus . . . 27
3.3 Input data used for implementation of GAA FinFET . . . . . . . . . 29
3.4 GAA FET results with gate dielectric variation . . . . . . . . . . . . 30
3.5 Input data used for implementation of GAA FinFET . . . . . . . . . 33
3.6 Parameters extracted from TCAD Synopsis Sentaurus for NW FET . 34
4.1 Input data used for physical simulations of MOSFET and Tunnel
FET [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 MOSFET vs TFET most important extracted parameters . . . . . . 48
4.3 MOSFET Temperature analysis . . . . . . . . . . . . . . . . . . . . . 50
4.4 TFET Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5 Input data used for implementation of DG Tunnel FET [1] . . . . . . 55
4.6 Most important extracted parameters from Praveen model . . . . . . 60
4.7 Important parameters extrcted from Model varying the geometric
structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.8 Important parameters extrcted from Model varing gate dielectric . . . 65
5.1 Input data used for physical simulation of SG Tunnel FET [1] . . . . 67
5.2 Important electrical parameters for Vds = 1V . . . . . . . . . . . . . 70
5.3 Temperature variation of electrical parameters for Vds = 1V . . . . . 73
5.4 Input data used for physical simulation of DG Tunnel FET [1] . . . . 75
5.5 SG vs DG TFET Extracted parameters from TCAD . . . . . . . . . 76
5.6 Gate dielectric variation . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.7 Temperature variation of electrical parameters for Vds = 1V . . . . . 80
5.9 Extracted parameters from physical simulation of GAA TFET . . . . 82
IV
5.11 NW TFET extracted parameters from physical simulation . . . . . . 84
5.13 Comparison between GAA TFET and GAA FET . . . . . . . . . . . 86
6.1 Input data used for physical simulation of heterostructure Tunnel
FET [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.2 Extracted parameters from physical simulation of Si/Ge heterostructure 92
6.4 Extracted parameters from the physical simulation of STBFET . . . 95
V
List of figures
2.1 Device Structure of SOI FinFET . . . . . . . . . . . . . . . . . . . . 5

2.2 Schematic diagram of a DG MOSFET (taken from [1]) . . . . . . . . 7
2.3 Id − Vgs characteristic of analytical model simulated in Matlab for
different gate length . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 3D Representation of FinFET (taken from [6]) . . . . . . . . . . . . . 13
2.5 Transcharacteristic in logarithmic representation varying gate length
(VDS = 1V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Schematic representation of (a) Rectangulal Re-TG FinFET, (b) Trape-
zoidal Tz-TG FinFET (taken from [5]) . . . . . . . . . . . . . . . . . 20
2.7 Comparison of Id − Vgs characteristic of Tz-TG FinFET and Re-TG
FinFET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.8 Simulated transfer-characteristic of a Tz-TG FinFET . . . . . . . . . 23
3.1 FinFET structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 L 200nm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 L 40nm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4 L 40nm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 TCAD Synopsis Sentaurus results of DG FinFET simulation varying
channel length L = 200nm - 40nm - 20nm . . . . . . . . . . . . . . . 28
3.6 Schematic representation of GAA FET for physical simulation . . . . 29
3.7 IV characteristic of GAA FET varying gate dielectric material, with
Vds=1V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.8 Comparative Temperature Analysis of GAA FinFET . . . . . . . . . 31
3.9 Comparative Temperature Analysis of drive current in GAA FinFET 31
3.10 Comparative Temperature Analysis of subthreshold current in GAA
FinFET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.11 Comparative Temperature Analysis of SS in GAA FinFET . . . . . . 32
3.12 NW FinFET Structure - Cross section of NW FET . . . . . . . . . . 33
3.13 Linear and logarithmic IV characteristic of nanowire FET with Vds=0.7V 34
4.1 Schematic representation of n-MOS like Tunnel FET . . . . . . . . . 38
4.2 Energy band diagram of tunnel diode (taken from [44]) . . . . . . . . 40
4.3 Triangular energy barrier approximation (taken from [44]) . . . . . . 41
4.4 Esaki Diode structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5 Esaki Diode IV characteristic . . . . . . . . . . . . . . . . . . . . . . 43
4.6 TFET and MOSFET structures . . . . . . . . . . . . . . . . . . . . . 44
4.7 Conventional MOSFET Band Diagram in OFF-state and ON-state . 44
4.8 TFET Band Diagram in OFF-state and ON-state . . . . . . . . . . . 45
VI
4.9 Id - Vgs Transcharacteristic for comparison of MOSFET and TFET
behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.10 MOSFET - Temperature analysis . . . . . . . . . . . . . . . . . . . . 49
4.11 SG TFET - Temperature analysis . . . . . . . . . . . . . . . . . . . . 49
4.12 InAs DG TFET implementation of analytical model [2] . . . . . . . . 56
4.13 ID − Vgs characteristic of TFET for the model implemented . . . . . 60
4.14 DG TFET SiO2 vs SGTFET Transcharacteristic Comparison . . . . 64
4.15 Transcharacteristic comparison among different structures . . . . . . 65
5.1 SG Tunnel FET Structure . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 Cross section of SG Tunnel FET used for TCAD simulation . . . . . 68
5.3 Energy band diagram of n-channel TFET in OFF-state (Vgs < Vth )
and ON-state (Vgs > Vth ) . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.4 Electric Field distribution in OFF-state and ON-state . . . . . . . . . 69
5.5 Drain current as a function of gate voltage in linear and logaritmic
scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.6 SG Tunnel FET Gate Capacitance varying drain voltage Vds . . . . . 71
5.7 SG TFET Temperature analysis . . . . . . . . . . . . . . . . . . . . . 72
5.8 SG Tunnel FET Transcharacteristic . . . . . . . . . . . . . . . . . . 73
5.9 DG Tunnel FET structure and cross section in Sentaurus . . . . . . . 75
5.10 Comparison of electrostatic analysis between a SG and a DG TFET . 76
5.11 Validation of MATLAB model with physical simulation . . . . . . . . 77
5.12 DG Tunnel FET varying gate dielectric, (a) Si3n4 and (b) HfO2 . . . 78
5.13 DG Tunnel FET Gate Dielectric Comparison with physical simulation 78
5.14 DG TFET Temperature analysis . . . . . . . . . . . . . . . . . . . . 79
5.15 GAA Tunnel FET Sentaurus Structure with HfO2 gate dielectric and
doping concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.16 GAA Tunnel FET characteristic varying gate dielectric . . . . . . . . 82
5.17 GAA Tunnel FET Sentaurus Structure with HfO2 gate dielectric and
doping concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.18 NW TFET IV characteristic for validation of Matlab model . . . . . 84
5.19 GAA Tunnel TFET and GAA FET . . . . . . . . . . . . . . . . . . . 85
5.20 Comparison of electrostatic analysis between cylindrical structures of
FET and TFET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.1 Band Diagram of a conventional TFET (left) and a Si/Ge heterostruc-
ture (right) in ON-state . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.2 Hetero-structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.3 Hetero-structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.4 Transcharacteristic of a Si/Ge heterostruture with Vds = 1V . . . . . 92
6.5 STBFET Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.6 Transcharacteristic of a STBFET . . . . . . . . . . . . . . . . . . . . 95
VII
Chapter 1
Introduction
Historically, device scaling in semiconductor industry has been driven by Moore’s

Law, ‘for a constant increase in performance and scaling of transistor 0 [41]. Gordon
Moore, one of Intel founders, in 1965 made an observation, which became a prophecy
for semiconductor companies:
‘The number of transistor on integrated circuits doubles approximately every
two years0 [42].
However, this trend seems to have an end around the 20nm technology node,
because of several transistor scaling limits known as short channel effects (SCEs).
Physical limits in CMOS technology are goig to became the most relevant issue in
Silicon process.
A first improvement in the last 20 years is the performance henances due to the
SOI instead of Bulk substrate.The key feature of the SOI structure (compare with
the bulk-Si) is the layer of silicon dioxide just below the surface [24].
Transistors are now made in a thin silicon layer sitting on top of a Si02 layer, that
allows to improve circuit speed and power consumption. Furthermore, SOI technol-
ogy brings improvements in temperature dependence and flexibility in the device
design, since substrate and device are physically separated by an insulator.
Finally, the main feature of SOI technology is the evident reduction of short channel
effets. A comparison in behaviour between Bulk and SOI devices could be useful to
better understand the trouble.
In a bulk device, the propagation of electric field is evident in depletion region

associated with the junctions.
The electric field contribution on channel performance can be reduced by increas-
ing doping concentration. Obviously, there is a limit in increase doping concentration
for proper device operation.
1
1 – Introduction
On the other hand, a FDSOI MOSFET is characterized by an electric field that

propagates through the buried oxide before going in the channel region. If a thin
burid oxide is realized over a ground plane, short channel effects are reduced.
The drawbacks behind that is the body effect and the increasing in junction capac-
itance, that can however be overcame by doubling the gate structure. In fact, by
the use of this approach, the electric field propagates to the bottom gate electrode
and cannot reach the channel region.
Furthermore, as MOSFET scales in dimensions, there is a need to preserve long-
channel behaviour as it possible. In fact, due to the reduction of channel length
dimension, the depletion width of the source and drain become comparable to the
channel length and punch-through between the drain and source will eventually oc-
cur [24].
This effect can be solved increasing channel doping, but as consequence there will
be an increase of the threshold voltage. To solve again that, the thickness of the
oxide can be reduced, but there is a limit related to that.
With this in mind, different scaling rules were used to optimize the device perfor-
mance, but even with that rules, the long-channel behaviour is difficult to reach
with the channel length scaling.
This is mainly due to the exponential increase of leakage current with reducing
the supply voltage. This behaviour is mainly related to a thermionic behaviour of
MOSFET.
Another short-channel effect is the CLM, which is defined as the reduction of
the inverted channel region length with increase in drain voltages, that results in an
increase in current with drain bias and a reduction of output resistance.
More in detail, as the drain voltage increases, the un-inverted region expands toward
the source due to the extension of the control over the current toward the source.
This results in an shortening of the channel region length. But, shortening the
channel decreases its resistance [24], causing an increase in current with increase in
drain bias. The effect is more evident as channel length reduces, the drain junction
becomes deeper and the thickness of the oxide increases.
In particular, if the sum of these depletion widths approaches the channel length, a
condition called punch-through is verified.
It results in an increase in leakage current between source and drain that is function
2
1 – Introduction
of the drain bias.

The origin of punch-through is the lowering of the barrier near the source, com-
monly referred to as DIBL (drain-induced barrier lowering) [24]. When the drain
is close to the source, the drain bias can influence the barrier at the source end,
such that the channel carrier concentration at that location is no longer fixed [24].
For a long-channel device, a drain bias can change the effective channel length, but
the barrier at the source end remains constant [24]. For a short-channel device,
this same barrier is no longer fixed [24]. The lowering of the source barrier causes
an injection of extra carriers, thereby increasing the current substantially [24]. This
increase of current shows up in both, above-threshold and subthreshold regimes [24].
Furthermore, the fabrication of such short channel FETs increases in complexity.
Short-channel effects make the device operation more complicated and the perfor-
mance of the devices degrades. Several solutions are needed to minimize these
effects.
Recent solutions for future transistor scaling include a three-dimensional structure
such as FinFETs, in order to improve isolation, reduce the subthreshold swing and
parasitic capacitances, and increase on-current.
Multiple gate devices allow a reduction in parasitic transverse electric field in the
channel, since there is an increase of the mobility.
Obviously, there is a better channel control that results in higher on current, better
subthreshold slope and smaller DIBL.
However, FinFETs present a poor subthreshold slope and high leakage current when
the device gate length is scaled below 14nm, so the main issues related to short chan-
nel effects still remain. Several solutions have been studied to overcome such limits
and to become a future alternative to conventional MOSFETs.
One emerging device is the Tunnel FET. As for technological fabrication it is a

MOS-like structure, except for having opposite doping between source and drain.
The device can be seen as a gated p-i-n junction, whose operating principle is based
on
Band-to-Band Tunneling. If a bias is applied to the drain terminal, the conduction
band in the channel region is pushed below the valence band in the source region,
so that tunneling occurs at the source/channel junction.
3
1 – Introduction
Tunneling is a quantum mechanical effect where particles have a non zero prob-
ability to tunnel through an energy barrier due to their wave nature [30], and this
effect becomes domaninant as device scaled.
Tunnel FETs are based on this effect, since carriers are injected through a barrier
instead of over a thermal barrier, a in conventional MOSFET. [30].
The subthreshold slope of Tunnel FETs can go below the classic thermal limit
of 60mV/dec. Indeed, it is characterized by a very low off-current and a weak
temperature dependence. The main characteristics of Tunnel FET are an increase
in Ion to Ioff ratio and in a reduction of short channel effects.
Even if the drain potential can affect the tunneling barrier at very short channel
lengths, a reduction in VDD enabled by the steep subthreshold swing of a TFET can
likely lessen the effect [30]. Therefore, the TFET device structure potentially allows
for scaling to shorter channel lengths prolonging Moore’s Law, and these properties
make TFETs a candidate for ultra-low power logic applications [30]. However, an
important issue of Tunnel FETs is the very low current in the on-state. Indeed,
researchers are studying several solutions, such as different gate materials, metal
gate engineering, heterostructure or vertical structures.
1.1 Thesis Outline

This work is divided in two main parts. After a brief introduction on CMOS sce-
nario, the firs part is dedicated to the modelling of Double Gate and Triple Gate
FinFETs. Physical simulations in TCAD Synopsis Sentaurus will check the accu-
racy and the computational effciency of the implemented model.
The second part of the thesis is devoted to investigate on new emerging devices such
as Tunnel FET, as alternative to conventional FETs.
Modelling of different Tunnel FET structures and relative physical simulations were
done in order to validate the reliability of such analytical models.
Analysis and simulation of emerging devices include IV behaviour, a capacitance
analysis, a temperature analysis of DG,TG,GAA FinFET, and SG,DG,GAA TFETs.
Finally, physical simulations of heterostructure tunnel fet and STBFET occupy the
last part of this thesis, as possible solutions to TFET main issues.
4
Chapter 2
FinFET
Multi-Gate (MG) MOSFETs are nowadays the most promising devices for the short
channel effects reduction, since the device scaling brought them in evidence.
MG transistors allows to a better control of the channel, a better subthreshold slope
and smaller DIBL parameter, an higher on current since more inversion volumes are
created.
Among all multi-gate FETs, the simple structure an fabrication process of FinFET
lead to some advantages with respect to the planar-MOSFET. The FinFET is a
non-planar multi-gate transistor realized on a SOI substrate. Despite of MOSFET,
which have an horizontal channel, the FinFET presents a vertical channel, named
fin. It can be seen as an ultra thin MOSFET, where the conducting channel is
wrapped by a thin silicon fin. The channel length is the extension of the fin under
the gate.
Drain
Source Gate
Figure 2.1: Device Structure of SOI FinFET
It is a three-dimensional device which allows to a further scaling above the 20nm.
5
2 – FinFET
In fact, as the channel length scales down, short channel effects are reduced by
reducing the width of the fin. Furthermore, the use of three gate surrounding the
fin ensures excellent electrostatic control. Furthermore, DIBL effect is less than
planar MOSFET, since fins are wrapped into the gate.
Finally, since it is a three-dimensional structure, it allows to provide a greater
device width per wafer area. In this way, the density of packaging increases quicker
than for planar MOSFETs.
However, reducing fin widths it could be a problem for the device performance,
since a narrower fin means a high access resistane, reducing the current in the on-
state. A solution to increase such current is to use multiple fins in parallel.
2.1 DG FinFET Analytic Models

Several analytical models already present in leterature were implemented in MAT-
LAB. The aim was to realize an electrostatical analysis a DG FinFET and to inves-
tigate on advantages and limits of such three-dimensional FET.
The implemented models were used to analyze both long channel and short channel
FinFET, so that a scaling behaviour of FinFET could be studied. Next sections
will be dedicated to the theorical explanation of analytical models, than results on
simulations of such models will be done.
2.2 Taur Model

A Double Gate FinFET can be modeled as first approximation as a very thin body
DG MOSFET. With this in mind, a continuous analytic IV model is derived using
Pao-Sah integral with no charge sheet approximation.
In this way, all linear, saturation and subthrehold region of MOSFET can be
analyzed. The model is based on the Pao Sah’s gradual channel approach. If a cut
along the vertical direction of the Si film is done, Poisson’s equation can be derived
as follows:
6
2 – FinFET
Figure 2.2: Schematic diagram of a DG MOSFET (taken from [1])
d2 ψ q q(ψ−V )
= n i e kT (2.1)
dx2 εsi
where ψ is the electrostatic potential, q is the electronic charge, ni is the intrinsic

carrier density, ε is the silicon permittivity, kT /q is the thermal voltage, and V is
the electron quasi-Fermi potential.
In this model the n-MOSFET is evaluated with qψ/kT 1, so that the hole
density is negligible [1].
The major contribute of the current flow is from the source to the drain along the
y-direction, so also the gradient of the electron quasi-Fermi potential V is in the
y-direction. Hence, the electron quasi-Ferm potential V can be approximated to be
constant in the x-direction [1]. By integrating twice the above expression, surface
potential can be evaluated:
s
2kT tsi q 2 ni 2βx
ψ(x) = V − ln[ cos( )] (2.2)
q 2β 2εsi kT tsi
where β can be derived by solving the following equation as a function of V:
7
2 – FinFET
s
q(Vg − ∆φ − V ) 2 2εsi kT 2εsi tox
− ln[ 2
] = lnβ − ln[cosβ] + βtanβ (2.3)
2kT tsi q ni εox tsi
Where Vg is the voltage applied to front gate and back gate, ∆φ is the metal
work function, ts i and to x are respectively the silicon and oxide thickness, and εox
is the oxide permittivity.
The current can be derived integrating Ids dy and expressing dV /dy as (V /dβ)(dβ/dy).
So, Pao-Sah’s integral can be written as:
Z Vds Z βd
W W dV
Ids =µ Qi (V )dV = µ Qi (β) dβ (2.4)
L 0 L βs dβ
Where βs and βd the solutions of β equation, respectively for V=0 and V=Vds ,
while Qi derived from Gauss’s law as Qi = 2εsi (dψ/dx)x=tsi /2 . The charge equation
can be expressed as function of β using (2.15) : Qi = 2εsi (2kT /q)(2β/tsi )tanβ. The
derivative dV /dβ can be also expressed as function of β by differentiating (2.16).
The analytical integral can be now carried out:
Z βs
W 4εsi tox 2kT 2 2εsi tox d
Ids =µ ( ) [tanβ + βtan2 β + βtanβ (βtanβ)]dβ
L tsi q βd εox tsi dβ
2
(2.5)
W 4εsi 2kT 2 β εsi tox 2 2
=µ ( ) [βtanβ − + β tan β]
L tsi q 2 εox tsi
The range of β is 0 < β < π/2, while βs and βd are found for given Vgs and Vds
from the conditions:
fr (βs ) = (q/2kT )(Vg − V0 )

(2.6)
fr (βd ) = (q/2kT )(Vg − V0 − Vds )
Where fr (β) is the right hand side of (2.16) si equals to fr (β) = lnβ − ln[cosβ] +
2εsi tox
εox tsi
βtanβ, while V0 is found to be:
8
2 – FinFET
s
2kT 2 2εsi kT
V0 ≡ ∆φ + ln[ ] (2.7)
q tsi q 2 ni
The current Ids can be computed as Ids ∝ [gr (βs ) − gr (βd )], where the function
2
gr (β) derives from (2.18) so that: gr (β) = βtanβ − β2 + εεsi tox 2
ox tsi
β tan2 β.
Solving β function as function of the gate voltage, this analytical model can
extract the FET characteristic for all operating regions.
In the linear region above the threshold it is found to be fr (βs ), fr (βd ) 1, so
βs ,βd ∼ π/2. The current in the linear region can be expressed as follows:
W
Ids = µCox [(Vg − Vt )2 − (Vg − Vt − Vds )2 ]
L (2.8)
W Vds
= 2µCox (Vg − Vt − )Vds
L 2
In the saturation region, βs ∼ π/2, and fr (βd ) 1, so (βd ) 1. The current

can be derived as:
W 8εsi tox k 2 T 2 q(Vg − V0 − Vds )

Ids = µCox [(Vg − Vt )2 − 2
exp ] (2.9)
L q εox tsi kT
Finally, in the subthreshold region, it is found to be both βs ,βd π/2, so that
both fr ,fr ∼ β 2 /2. The current can be expressed as:
W q(Vg − ∆φ) qVds

Ids = µ kT ni tsi exp (1 − exp − ) (2.10)
L kT kT
”This continuous IV model is derived from analytic solutions of Poisson’s and
current continuity equation for long channel DG MOSFET” [1].
9
2 – FinFET
2.2.1 Matlab Implementation
Taur Model [1] was implemented in Matlab in order to simulate the double-
gate (DG) FinFET electrostatic characteristic. In fact, such model can be used to
describe the behaviour of a FinFET if the latter is seen as a MOSFET with a very
thin layer.
The device parameters extracted from Fasarakis model [4] are described in the table
below:
Quantity Name Value u.m. (S.I.)

Gate Length 200 - 20 nm
Oxide Thickness (tox ) 1 nm
Fin Width (Wf in ) 10 nm
Temperature (T) 300 K
Source/Drain doping concentration (Nsd ) 1020 cm−3
Channel doping concentration (Nch ) 1015 cm−3
Metal workfunction φm 4.1 eV
Table 2.1: Input data used for implementation of Taur model [1]
A simulation of DG FinFET were carried out in MATLAB varying the channel

length, hence the scaling behaviour of such FET could be put in evidence.
The following figure represents the trend in channel length scaling described from
the Taur model, with Vds = 1V.
10
2 – FinFET
I d -Vgs Transcharacteristic
10-5
log(I d ), A/um
10-10
L 200nm
L 40nm
10-15 L 20nm
0 0.5 1 1.5
Vgs, V
Figure 2.3: Id − Vgs characteristic of analytical model simulated in Matlab for dif-
ferent gate length
The table below includes the current values for on and off state. In this way,
the trend of the model implemented varying the gate length can be analyzed.
Parameters L = 200nm L = 40nm L = 20nm

Ioff, A/um 2.87e-16 1.44e-15 2.87e-15
Ion, A/um 1.08e-3 5.37e-3 1.07e-2
Table 2.2: Current values varying gate length
The leakage curent varies of one order of magnitude varying the gate length from
200nm to 20nm, while the drive current remains quite stable.
The model seems to be a quasi ideal model, since going from 200nm to 20nm of
channel length the characteristic doesn’t change. This is quite strange, since one of
main issues of FinFETs is the increasing of leakage current and subthreshold slope
when a short channel device is considered.
Here, the curves remains stable and increases in a quasi linear way, so that a real
difference between long channel and short channel behaviour can’t be appreciated.
Furthermore, the threshold voltage seems to be not influenced by channel length
variation.
11
2 – FinFET
In conclusion, a continuous analytical model for DG MOSFET were extended to

DG FinFET. The model has been derived from ”analytic solutions of Poisson’s and
current continuity equation fro long-channel DG MOSFETs” [1]. The results were
in agreement with those present in [4], hence include all operating regions of FET.
12
2 – FinFET
2.3 Fasarakis Analytical Model

This compact model has been developed for lightly-doped short-channel triple-gate
fin-shaped FET [6]. However, if the top gate oxide is much thicker than the gate
oxide on the two sides, the TG FinFET can be treated as a DG FinFET.
In this thesis, a DG FinFET is modeled first and compared with Taur model pre-
vious explained. Then , it is extended to the case of TG FinFET. Fasarakis model,
based on prevoius analytical modeling of drain current in DG MOSFETs, derives a
”fully analytical and compact drain current model valid in all regions of operation for
FinFETs” [6]. Several effect has been taken into account, so quantum-mechanical
(QMEs) and short-channel effects such as threshold-voltage shifts, DIBL and sub-
threshold slope degradation. Also the effects of series resistance, surface roughness
scattering, channel length modulation and saturation velocity in the sauration re-
gion were considered [6].
Figure 2.4: 3D Representation of FinFET (taken from [6])
This model is based on analytical expressions for the threshold voltage and SS
of lightly doped DG and TG MOSFETs [4]. Taking this as a starting point, it has
been derived a fully analytical and compact drain-current model valid in all regions
of operation for TG FinFETS.
In particular, it has been derived an unified expression for the inversion charge
13
2 – FinFET
and drain current that is valid in all operation regions.
The model can be extended to a DG FinFET by changing the effective channel

width definition.
Wf in
WT G = Hf in + WDG = 2 ∗ Hf in (2.11)
2
Where W is the effective channel width of a Triple Gate and of a Double Gate
FinFET, respectively. It has been considered here that each half of the top gate
width Wf in contributes to the side gate of width Hf in [6].
The drain-current equation of the modeled FinFET is similar to the one of DG

MOSFET expressed with the unified charge-based equation [6], but with modified
threshold voltage Vth and subthreshold swing coefficient ηT G and with increased
channel width W .
The TG FinFET drain current is derived as follows:
2W ox 2
(q 2 − qid )
Id = µo (2VT )2 [(qis − qid ) + is ] (2.12)
L tox 2
Where µ0 is the low-field electron mobility, W is the channel width, L is the
channel length, εox is the gate oxide permittivity, tox is the gate oxide thickness, and
Vth is the thermal voltage.
It has been noticed that the first term in the above equation brings a domi-
nant contribution in the subthreshold region, while the second charge term is more
dominant in the above-threshold region. In the above expression, qis and qid are
the normalized inversion sheet-charge densities calculated at the source and drain,
respectively, and that are calculated from the unified normalized inversion sheet-
charge density of DG MOSFET [4].
The characteristic of the device can be studied by evaluating the threshold volt-
age in an accurate way. In particular, starting from the strong inversion region, the
normalized sheet-charge density can be written as:
14
2 – FinFET
s
1 qtox n2i εsi
qix1 = LambertW [exp [(Vg + ∆Vt − Vf b − Vx ) + 2Vth ln( )]]
2Vth εox 2kT NA
(2.13)
The threshold voltage can be derived by equating the above equation with: qix2 =
LambertW [exp V1th (Vg − Vt0 − Vx )].
The threshold voltage for a DG FinFET can be expressed as follows:
A1,DG (Vbi + Vd ) + A2,DG Vbi Vt Qth NA

Vt h = Vf b − + ln( 2 ) (2.14)
1 − (A1,DG + A2,DG ) 1 − (A1,DG + A2,DG ) ni Wf in
Where Vb i is the built-in potential at the source/drain interface, and A1,DG ,A2,DG
are parameters which are function of the device natural length and channel length.
If such parameters are oppurtunal changed, the model can be succesfully extended
to the TG FinFET.
The minimum carrier sheet density obtained from the above analysis of the
sheet-charge density in the strong inversion, is given by:
2
2Vth Cox
Qth = ( )( ) (2.15)
q CSi
Where CSi = εsi /Wf in . This analytical expression leads to a calculation of the
characteristic of the device.
Several effects have been included in this model. The first one is the channel
length modulation. The CLM effect can be described as the pinchoff in the channel
that moves from drain toward the source when the Vds is increased beyond the
saturation voltage Vdsat = Vg − Vth . This effect can be seen as the reduction of the
physical gate length, such that the effective electrical gate length is given by
L0 = L − ∆L (2.16)
where ∆L is the gap between L and the channel pinchoff. So, considering the CLM
effect, the drain-current equation becomes [6]:
15
2 – FinFET
2 2
ox qis − qid 1 (qis − qid )
Id = 2W µo (2VT )2 [ + ] (2.17)
tox L 2 L − ∆L
The effects of series resisance and saturation velocity due to the horizontal drain
field and surfac roughness scattering due to the vertical gate filed were considered,
in order to compare the model with simulation and experimental results [6]. These
effects are included in the electron mobiliy expression like:
µo
µ= (2.18)
1 + Θ1 Vth qis
So, in this sense the final compact equation for drain current becomes:
2 2
ox qis − qid 1 (qis − qid )
Id = 2W µ (2VT )2 [ + ] (2.19)
tox L 2 L − ∆L
The last effects that have been included are the QMEs, since for short channel
FETs a quantum well is formed in the fin width between the two-side oxide layers.
The model of QMEs used in this model inserts the carrier-energy quantization caused
by the structural confinement, as a widening of the bandgap in thin films [6].
In the particular case of DG FinFETs, the minimum energy of the first subband
above the conduction band is [6]:
~2 π 2
EG1 = (2.20)
2me f f Wf2in
This correction results in an increase in the threshold voltage, such as:
(π~)2
∆VthQM = α (2.21)
2qmef f Wf2in
Where α is equal to 1 for DG and 2 for TG.

So, as first QME effect, this is the shift of Vth due to structural confinement, including
the second effect of bias-independent quantum degradation gate capacitance [22],
[23].
The second effect arises from the quantum-mechanical distribution of the inversion
charge, showing a peak inside the substrate at some distance away from the SiO2 −Si
interface. [6] This effect can be accounted for by considering two capacitance values
16
2 – FinFET
connected in series: the oxide capacitance formed by the physical oxide layer and
the capacitance developed within the average distance of ∆z inside the silicon from
the interface [22], [23].
The QME results in an henanced value of the gate oxide thickness given by:
ox
tQM
ox = tox + ∆z (2.22)
Si
It is obvious to understand that this modified value of tox results in a shift of the
QM
threshold voltage by ∆Vth,tox . This effect is calculated by replacing tox with tQMox .
Thus, including the QMEs in the drain-current equation, the classical threshold Vth
is replaced by VthQM = Vth + ∆VthQM + ∆Vth,tox
QM
and tox with tQM
ox .
17
2 – FinFET

The trancharacteristic of DG FinFET varying the channel length has been simulated.
The device parameters extracted from Fasarakis model [4] are described in the table
below:

Table 2.3: Input data used for implementation of Fasarakis model [6]
The channel length variation was useful to study the behaviour of the analytical
model and verify the reliability of the model with scaling devices.
10-5
log(I d ), A/um
10-10
L 200nm
L 40nm
L 20nm
10-15
0 0.5 1 1.5
Vgs, V
Figure 2.5: Transcharacteristic in logarithmic representation varying gate length

(VDS = 1V )
18
2 – FinFET
Figure (2.5) shows the results of the implemented model where gate length pa-
rameter was varied. The trend of the simulated model is in agreement with the
physical model described in [4]. In particular, the most important extracted values
are riassumed in the table below:
Parameters L = 200nm L = 40nm L = 20nm

Ioff, A/um 1.66e-15 2.71e-14 1.57e-11
Ion, A/um 9.31e-4 6.60e-3 1.25e-2
Vth, V 0.57 0.58 0.72
SS, mV/dec 59.53 61.6 70.63
Table 2.4: Parameters values varying gate length
In this model, an appreciable variation can be observed by varing the channel

length. In fact, the aim of this model validation was to check the accuracy of phys-
ical models, even if such models didn’t agree with results present in leterature. In
this case, there is a variation of 4 order of magnitude in leakage current if a long and
short channel DG FinFET is taken into account. It can be observed an increase of
25% of subthreshold slope with reducing the gate length from 200nm to 20nm, and
as consequence an increase of Vth of 30% is shown.
In conclusion, this analytical model for FinFET, which accounts short channel
effects such as QMEs, CLM, were implemented in MATLAB in order to check its
accuracy.
19
2 – FinFET
2.4 TG FinFET Analytic Models

Fasarakis compact model for a DG FinFET can be extended to a rectangular TG
FinFET and hence to a trapezoidal one [5], engineered by Intel. Equivalent non
planar device parameters and corner effects have been included.
For compact modeling, the non vertical sidewalls are assumed to be symmetric with
equal slopes of the lateral sides, such the real fabricated device by Intel [20].
The triangular shape of the fin can improve the electical characteristic of a trigate
FinFETs.
Figure 2.6: Schematic representation of (a) Rectangulal Re-TG FinFET, (b) Trape-
zoidal Tz-TG FinFET (taken from [5])
In the rectangular TG FinFET compact model, physical effects such as SCEs,

filed-dependent mobility, series resistance, CLM and QM effects were included [5].
Several considerations have to be done with a Re-TG FinFETs, such as the fact that
in non planar devices, significant error is introduced in Cox when tox/W f in > 0.1
[21]. The model implemented is the same as for DG FinFET, except for the effective
channel width. For the trapezoidal structure, the effective channel width is found
to be:
λ
Wf in = Wf in,top + (Wf in,bot − Wf in,top ) (2.23)
λ+1
20
2 – FinFET

The compact model for DG FinFET was extendend to Rectangula TG-FinFET.
Furthermore, a comparison of electrostatic behaviour between a rectangular and a
trapezoidal TG FinFET was carried out.
Device parameters are extracted from [5], and in the table below are riassumed the
most relevant parameters used for the simulation.

Gate Length 50 nm
(Wf in,bot ) 15 nm
(Wf in,bot ) 13 nm
Hf in 30 nm
Source/Drain doping concentration (Nsd ) 5 ∗ 1019 cm−3
Table 2.5: Input data used for implementation of Fasarakis model of Tz-TG and
Re-TG FiNFET (taken from [5])
In the figure below is shown the Id − Vg characteristic of the descripted device,

simulated with Vds = 1V .
I d -Vgs Transcharacteristic 10-4
Re-TG FinFET
10-4 Tz-TG FinFET 6
5
10-6
log(I d ), A/um
4
I d , A/um
10-8
3
-10
10 2
1
10-12
0
0 0.5 1 1.5
Vgs, V
Figure 2.7: Comparison of Id − Vgs characteristic of Tz-TG FinFET and Re-TG

FinFET
21
2 – FinFET
The characteristic in figure (2.7) shows the variation of electrical parameters

if a rectangular or a trapezoidal structure is considered for FinFET. As it can be
observed from the table below, the most evident variation between the two named
structure is in drive current. In fact, it can be noticed that the current in on state
increase by a factor of 1.5X, while all the other electrical parameters remain quite
stable.
Parameter Tz-TG Re-TG

IOF F , A/um 1.15e-13 3.27e-13
ION , A/um 4.21e-4 6.5e-4
Vth, V 0.57 0.54
SS, mV/dec 65.53 64.74
Table 2.6: Electrical parameters extracted from simulation
Furthermore, the compact model for TG FinFET was extended to Trapezoidal

TG FinFET. Since the model is the same of previous analysis of DG FiNFET, also
in this case all short channel effcts are taken into account. Furthermore, it were
considered the corner effect and the effect of the top gate on the natural length [5].
Device parameters used for MATLAB simulation are shown in the table below:

Gate Length 25 nm
(Wf in,bot ) 15 nm
(Wf in,bot ) 5 nm
Hf in 30 nm
Source/Drain doping concentration (Nsd ) 5 ∗ 1019 cm−3
Table 2.7: Input data used for implementation of Fasarakis model of Tz-TG FiNFET
(taken from [5])
22
2 – FinFET
I d -Vgs Transcharacteristic 10-4
5
10-4 4.5
3.5
log(I d ), A/um
-6
10
I d , A/um
3
2.5
2
-8
10 1.5
0.5
-10 0
10
0 0.2 0.4 0.6 0.8 1 1.2
Vgs, V
Figure 2.8: Simulated transfer-characteristic of a Tz-TG FinFET
In the figure above is shown the Id − Vg characteristic of the descripted device,

simulated with Vds = 1V . It can be observed that there is not so evident variation
in electrostatic behaviour with respect to the short channel DG FinFET simualted
in section (2.3.1).
Such result is in agreement with leterature, since a trapezoidal structure is done
because of a ease fabrication but it doesn’t lead to any real improvement in elec-
trostatic behaviour. However, reducing the gate length of the Tz-TG FinFET, the
current in off state increases of 3 order of magnitude for reducing the gate length of
an half.
Parameters Extracted Values

Ioff, A/um 1.52e-10
Ion, A/um 5.18e-4
Vth, V 0.492
SS, mV/dec 85.15
Table 2.8: Parameters values of a Tz-TG FinFET simulated in Matlab
23
Chapter 3
Physical Simulations of FinFET
3.1 Double-Gate FinFET Simulations

TCAD Synopsis Sentaurus simulations were carried out with the aim to validate
the previous implemented models. The structure which this work referred to is the
same for all the models. The device parameters are shown in tha table below:

Table 3.1: Input data used for implementation of Fasarakis model [6]
The structure extracted from the physical simulation is shown below and is quite
similare for both DG and TG FinFET, since the difference is in the thickness of top
gate oxide. The colours refer to the materials used for the simulations, so a Silicon
fin and a Silcon channel, Aluminum as gate metal and SiO2 for gate oxide.
24
3 – Physical Simulations of FinFET
Drain
Source
Hfin Gate
Figure 3.1: FinFET structure
The IV characteristic of a DG FinFET with channel length of 200nm was simu-

lated in TCAD Synopsis Sentaurus, with Vds=1V.
10-5
log(I d ), A/um
10-10
TCAD Sentaurus simulation

Fasarakis Model
Taur Model
10-15 Tsormpatzoglou Model
0 0.5 1 1.5
Vgs, V
Figure 3.2: L 200nm
The models based on DG MOSFET could be potentially good to describe a

long channel FinFET. However, the Tsormaptzoglou one seems to be far from the
physical simulation in the off state also for a long channel device. Instead, Taur
model, which is also based on DG MOSFET modelling, results in a quite good
approximation of the FinFET.
25
Other simulations had been done with the same approach, varying the gate
length in order to going in deep with scaling and investigate on FinFET limits.
The following figures represent the same simulation with gate length respectively
of 40nm and 20nm.
The same structure was simulated for all the models implemented. In particular
a comparison was done for long channel FinFET, and one for a short channel.
In fact, a difference in the behaviour was noticed if a long channel or a short
channel device was considered. This is due to the physics approximation of the
models.
In particular, since Taur and Tzompazoglou considered the FinFET as a MOS-
FET with a very thin box oxide, they are not so accurate as can be Fasarakis. In
the figure above, a DG FinFET with gate length of 200nm was compared among
different compact models and Sentaurus.
Taur and Fasarakis seems to be more accurate than Tzompazoglou, even if a
long channel is considered.
In the figure below, the same comparison was done for the case of a short channel
DG FinFET.
10-2
10-4
10-6
log(I d ), A/um
10-8
10-10
10-12
Fasarakis Model
Taur Model
10-14
Tsormpatzoglou Model
0 0.5 1 1.5
Vgs, V
Figure 3.3: L 40nm
26
100
10-5
log(I d ), A/um
10-10

Fasarakis Model
Taur Model
Tsormpatzoglou Model
10-15
0 0.5 1 1.5
Vgs, V
Figure 3.4: L 40nm
The last plots put in evidence the behaviour of the analytical models imple-
mented and the physical simulations, in the case of short channel FinFETs. Even
if Taur model seems to be the most accurate model, it is not considered as the best
one. In fact, it can be noticed that it doesn’t vary with gate length reduction, so it
is not so reliable. Also Tsormaptzoglou is not an accurate model, since it doesn’t
respect the trend of such physical simulations.
The table below shows the most important parameter extracted from the physical
simulation varying the channel length in a DG FinFET.
Parameters L=200nm L=40nm L=20nm

Ioff, nA/um 2e-5 1.57e-4 3.27e-6
Ion, uA/um 1.34e3 1.08e3 5.38e2
Vth, V 0.87 0.82 0.86
SS, mV/dec 67.74 61.4 60.5
Gm, S/um 2.15e-3 1.59e-3 8.41e-4
Table 3.2: Important parameters extrcted from TCAD Synopsis Sentaurus
27
Figure 3.5: TCAD Synopsis Sentaurus results of DG FinFET simulation varying

channel length L = 200nm - 40nm - 20nm
From the above results, it can be observed that the device simulated it is a quasi
ideal device, since the sbthreshold slope is near the thermal limit of 60mV/dec and
the current in the off state is lower than usual. In fact, in the paper which this part
of the work is referred to, the current in the on state is quite higher.
However, these simulation ar useful to understand and better validate the anlyt-
ical model implemented.
Finally, from these analysis, it can be observed that the Fasarakis model is the
most reliable model to describe the electrostatic characteristic, if an accurate and
reliable trend among the variation of channel length is considered. Since the aim
of these simulations was to check the most reliable and accurate analytical model,
what it did matter was not the extact value among all simulations compared, but
the respect in the trend of scaling transistor. The model that is the most reliable
from this point of view is the Fasarakis one, since an appreciable variation between
logn channel and short channel can be seen from the presented electrostatic analysis.
28
3.2 Gate-All-Around FET Simulations

The analysis of FinFET exploit in simulation of GAA FET, in order to investigate
on further scaling of device and improvement of electrostatic characteristic.
Device parameters are presented in the table below:

Gate Length 40 nm
(tsi ) 10 nm
Table 3.3: Input data used for implementation of GAA FinFET
The metal gate is Aluminum, while the channel is made of Silicon. A TCAD
Synopsis Sentaurus simulation was carried out to study the electrostac behaviour
and the temperature variation of a Silicon gate all around FET, whose structure is
shown below:
n+ Si Source
HfO2
Si Channel
Aluminum
n+ Si Drain
Figure 3.6: Schematic representation of GAA FET for physical simulation
29
In particular, a variation of gate dielectric material was done in order to study

the improvement of the current with dielectric permittivity, as shown in the following
figure:
Figure 3.7: IV characteristic of GAA FET varying gate dielectric material, with
Vds=1V
It can be demonstrated that varying the gate dielectric from SiO2 to HfO2 it
varies the permittivity by a factor of 6. This leads a decrease of the subhreshold
slope, paying with an increase of leakage current.
Parameters SiO2 HfO2

Ion, A/um 2.13e-5 2.28e-5
Ioff, A/um 9.23e-7 1.2e-6
Vth, V 0.164 0.25
SS, mV/dec 94.95 71.72
Table 3.4: GAA FET results with gate dielectric variation
30
Furthermore, a temperature analysis was carried out for the same structure. In
the following tables are described the most important electrical parameters influ-
enced by a variation of temperature.
Figure 3.8: Comparative Temperature Analysis of GAA FinFET
Figure 3.9: Comparative Temperature Analysis of drive current in GAA FinFET
In particular, as can be shown from figure (3.7), the current in the on state
31
does’nt vary in an evident way with temperature variations from 200K to 400K. On
the other hand, as can be seen in the figure below, the off current increases of about
two order of magnitude, while thesubthreshold slope increases in a dramatic way by
a factor 4.
Figure 3.10: Comparative Temperature Analysis of subthreshold current in GAA

FinFET
Figure 3.11: Comparative Temperature Analysis of SS in GAA FinFET
32
3.2.1 Nanowires FET Simulations
Finally, a further scaling of gate length was done. So, a nanowire fet was realized
and simulated in TCAD. Tungsten as gate metal and Hfo2 for gate dielectric.

Gate Length 15 nm
(tsi ) 8 nm
Table 3.5: Input data used for implementation of GAA FinFET
n+ Si Source
HfO2
Si Channel
Aluminum
n+ Si Drain
Figure 3.12: NW FinFET Structure - Cross section of NW FET
The characteristic shows a better behaviour of the nanowire FET, in particular

in the off state. In fact, the Ion over Ioff ratio increases reducing the channel length
and modifying the other materials. Also the subthreshold slope leads to the physical
thermal limit of such devices.
33
Figure 3.13: Linear and logarithmic IV characteristic of nanowire FET with

Vds=0.7V
Parameters Results
Ioff, A/um 9.55e-9
Ion, A/um 2.65e-5
SS, mV/dec 66.93
DIBL, mV/V 11e-3
Table 3.6: Parameters extracted from TCAD Synopsis Sentaurus for NW FET
The table above shows the most important parameters of an electrostatic analy-
sis. The result seems to be a very good one with respect to the theorical expectations.
Since the structure dimensions between the gaa and nw are completely different, no
comparison can be done, but a quantitative evaluation of behaviour in electrostatic
analysis can be done.
34
3.3 FinFET Main Issues
3.3.1 Advantages and Drawbacks
Altough FinFET is being a promising alternative to conventional MOSFET, since

several benefits are exploited from its fabrication.
In particular, a Double Gate FinFET present a reduced electric field from the gate
to the top of the fin, and this is due to the channel fabricated with an ultra-thin
layer of Silicon on an Insulator.
In the case of a Tri-Gate FinFET, the gate wraps around the channel on the three
sides which allows a better electrostatic control of the channel itself. Furthermore,
the main benefit derives from the higher packing density due to the vertical struc-
ture.
One important feature of FinFET is the fin thickness, which needs to be smaller
than or equal to the gate length [3]. The behaviour of FET is mainly due to the
lithography process, since the scaling of the fin thickness doesn’t depend on the
oxide thickness.
Another advantage is the low threshold voltage variability with drain voltage vari-
ations. As a consequence, the channel is well controlled and litghly doped.
Fabrication of FinFETs is the same as the one of the conventional MOSFET,

however several drawbacks have to be taken into account with device scaling in
semiconductor industry. In fact, challenges and complexity in fabrication play an
important role in the recent few years.
Limitations of MOSFET are mainly related to the channel length reduction and
to the thermal limits of the device.
Thermal limits refer to the poor SS that characterize conventional MOSFETs. Short
channel effects include velocity saturation effect, DIBL, impact ionization, hot car-
rier effect.
In fact, due to scaling of MOSFETs what can be noticed is the increase of the elec-
tric field, hence velocity of charge carriers increases. The problem is that when the
electric field goes to a high value, velocity saturates.
35
Since electrons velocity is high, they can impact on silicon atoms and pair of elec-
tron and holes are created and this is the impact ionization [10]. As a consequence,
this pair of electrons and holes enter into dielectric because of gaining high kinetic
energy [10], hence changing the capacitance of the system and making it less reliable
[10].
Secondly, as if for long channel devices the threshold voltage doesn’t vary with the
drain voltage, with the device scaling the threshold voltage decreases with the in-
creasing of Vds.
Going in deep with drawbacks related to the FinFET design, several effects have
to be taken into account in reducing the transistor dimensions.
Firstly, fin-width reduction let decrease of short channel effects, but there is an in-
crease of parasitic drain/source capacitance and as a consequence there is a current
reduction and transconductance reduction. Secondly, with the reduction of the fin
width device temperature increases since heat cannot easily flow through the device.
The effect is more pronounced in case of SOI technology, where buried insulating
layer causes severe self-heating effects due to low thermal conductivity of oxide layer
[3].
The well-known corner effect is defined as the increase in leakage current at the
corner of the fin with the Vgs increasing. In fact, because of the device sclaing, the
charge sharing occurs in the cornerregion of the two adjacent gates [3], hence there
is a premature inversion at the corners and a trnasversal electric field is verified in
being concentrated at the fin corners. As primary effect of the premature inversion
at the corners, subthreshold characteristics degrade so higher off state current can
be shown.
A solution to this problem is to fabricate a rounded profile of the fin. This allow a
leakage current reduction , but the extraction of prasatics increases in complexity.
In particular, with dimension reduction becames difficult to extend FinFET RC

parasitic models to be close to those extracted in simulations. A reduction of oxide
thickness or channel doping concentration could solve corner effects.
36
Finally, as FinFET thickness reduces, the quantum effect reduces the density of
available states at the band edge, hence carriers need more energy to occupy avail-
able states higher than the band edge, and be free to conduct device current.
Other challenges due to gate length reduction below 20nm are related to the
needed of double patterning to print correctly with current lithography equipment.
Secondly, electormigration becomes more of a concern as geometries shrink [3].While
double patterning will make immersion lithography practical at 20nm, a new ap-
proach will be needed at 10nm [3], known as SIT.
In conclusion, FinFETs have several benefits, such as a quite good control of SCEs.
However, fabrication is now complex with technological node below 20nm.
A new alternative is needed to overcome all the FinFET and MOSFET drawbacks.
37
Chapter 4
Tunnel Field Effect Transistors
”As the evice dimension further scales, the semiconductor devices are entering into
a tunneling epoch” [44]. Recently, the ”green-transistor” Tunnel Field Effect Tran-
sistor (TFET) has been proposed to overcome conventional MOSFET limits. It is a
MOS’ like tunneling transistor, and it can be seen as a gated Esaki tunneling diode
based on Band to band tunneling working principle. It is simple to realize since its
structure is the same of that of the conventional MOSFET, so the fabrication can
be realized in CMOS technology, but source and drain regions are doped of opposite
doping types.
Figure 4.1: Schematic representation of n-MOS like Tunnel FET
In the last few years TFET is under research, since could bring several advantages
with respect to the conventional MOSFET.
Since it can be seen as a gated p-i-n diode which works on the basis of reverse
bias [10], such reverse pin junction present an high barrier. The higher barrier lead
to a lower leakage current, so that such transistor could be the ideal device for low
power applications.
The most evident difference in structure between MOSFET and TFET is the
different doping between source and drain regions. Indeed, source and drain regions
are heavily doped with opposite types, depletion region forms at the junction of
38
4 – Tunnel Field Effect Transistors
intrinsic region and the n+ doped drain region [10].

While thermal injection is the mechanism for the source of carrier injection used in
MOSFET [25], BTBT is the mechanism used in Tunnel FET. A TFET is a three
terminal pin diode, where source terminal is the source of majority carrier, drain
terminal carries out the majority carrier, and finally gate terminal is used to control
the majority carriers that move from source to drain. In particular, the channel
needs of gate oxide since it prevents the current leakage from channel to gate [26].
The channel is an intristic region, and as consequence the threshold voltage fluc-
tuations are removed since there is no dopant atom distribution. As it will be
demonstrated in the next sections, the TFET present several issues. Since the tun-
neling region is smaller with respect to the silicon body [44], the current in on state
is relativel low. An high doping concentration in both source and drain regions
is needed in order to enhance the tunneling at the interface between source and
channel junctions.
Following sections will be dedicated to the theorical study of Band to Band Tun-
neling, then physical simulations of TFETs different structures will be carried out.
In particular, before the investigation on electrostatic behaviour of Tunnel FETs, a
study of Esaki Diode will be useful to better understand the behaviour of current
when a tunneling principle is considered. Finally, a comparison in electrostaci be-
haviour between MOFET and TFET will be done, to put in evidence the differences
between the two devices.
4.1 Band-to-Band Tunneling in p-i-n Diodes

The p-i-n diode working principle is discussed in this section to better understand
TFET. In fact, the latter can be seen as a gated p-i-n diode. A p-i-n diode is realized
as the union of two junctions heavily doped, where an intrinsic region is sandiwched
insde them. The intrinsic region lightly doped is useful to let the potential drop
mostly in this region [44].
”The Esaki tunnel diode was first presented by Dr. Esaki in Physical Review,
1958. By forming the heavily doped p-n junction, the negative resistance was found
in the forward I-V characteristics. This is the discovery of a new quantum mechanical
tunneling phenomenon. This tunneling effect is called Esaki-Tunneling” [44].
39
The energy band diagram a thermal equilibrium is sohwn in the figure below:
Figure 4.2: Energy band diagram of tunnel diode (taken from [44])
In off state, the barrier width is quite large, so that tunneling is suppressed.
In such p-i-n diode there are two main tunneling current contributions: the Zener
tunneling and the Esaki tunneling current. The first one is the current that flows
from the valence band to the conduction band, while the latter one results from
electrons flow from the conduction band to valence band. Furthermore, tunnel
diode current comprises three main components: the tunneling current, the excess
current and the thermal current.
The tunneling current from the conduction band to the valence band is given
by the ”number of electrons times the unoccupied states in the valance band times
the probability for tunneling from the conduction band to the valance band without
any energy change” [44].
Itunnel (E) = (fc (E)ρc (E))((1 − fv (E)ρv (E))T (E) (4.1)
Where fc and fv can be calculated with the Fermi- Dirac distribution functions
and are defined as the probabilities of a quantum state to occupy the cunduction
and valence band respectively [44]. Then, ρc and ρv are defined as the energy level
densities in the conduction and valence band respectively. E is the energy level, and
finally T(E) is the probability for tunneling from the conduction band to valence
band.
Furthermore, the Esaki and Zener tunneling current can be derived by the inte-
gral of the tunneling current over the range of overlapping energy states.
40
The transmission probability T(E) can be derived by applying the WKB approx-
imation on the Schrodinger time-dependent wave equation:
p(x) dx
R
T (E) ∼ exp−2 ~ (4.2)
Where p(x) is the absolute value of the momentum of the particle in the barrier
[44], ~ is the normalized Planck’s constant.
So, if the WKB approximation is applied, the energy barrieri for tunneling can
be approximate to a triangle, as shown in the figure below:
Figure 4.3: Triangular energy barrier approximation (taken from [44])
The probability of tunneling can be calculated as:

√ 3/2
4 2m∗ EB
T (E) = exp(− ) (4.3)
3 ~ eε
Naming the Esaki and Zener tunneling current respectively IE and IZ , the total
tunneling current can be derived as:
It = IE − IZ (4.4)
It can be demonstrated that if no bias is applied, the Zener tunneling current

equals to Esaki tunneling current [44]. If the bias is applied, he current increases as:
It = IP (V /VP )exp(1 − V /VP ) (4.5)
where IP and VP are respectively the peak current and peak voltage. Finally,
the complete current expression includes three main contirbutes: the esaki tunneling
curent, the excess current and the thermal current.
41
J = Jt + Jx + Jth
(4.6)
= JP (V /VP )exp(1 − V /VP ) + JV exp[A2 (V − VV )] + J0 (expqv/kT − 1)
4.1.1 Physical Simulations of Esaki Diode

In order to understand the behaviour of the Esaki Diode explained in the prevous
section, a p-i-n diode was simulated in TCAD Sentaurus.
Figure 4.4: Esaki Diode structure
An Esaki diode is a highly doped pn junction which is based on band to band

tunneling phenomenon.
The tunneling effect is defined as the non null probability of a quantum particle to
tunnel across the barrier of a junction even if it has energy less than the energy
barrier [29].
The probability increases with barrier energy decreasing , as shown in the above
expression :
P ∝ exp(−AEb W ) (4.7)
42
Where P is the probability of a particle to tunnel across the barrier, Eb is the

barrier energy of the junction, and W is the width of the barrier.
The IV characteristic that result from simulation is shown below:
Figure 4.5: Esaki Diode IV characteristic
The main feature of the IV characteristic of a tunnel diode is the NDR (Negative
diode region), while behind this region the Esaki diode acts as a normal diode. As
voltage increases the current also increases until it reaches the so called peak current
around 1e-15 A.
If voltage increases, the current starts to decrease, hence it is in the negative resis-
tance region.
Beyond valley point the tunnel diode acts as normal diode.
4.2 Comparison between MOSFET and TFET

It could be useful to first compare MOSFET and Tunnel FET behaviour, to better
understand the most evident differences among the two devices.
43
Figure 4.6: TFET and MOSFET structures
In the above figure it can be noticed that in the conventional MOSFET both
source and drain regions are doped in the same way, while in the Tunnel FET the
type of doping of source and drain is different.
From the band diagrams in off and on state that are shown below, it is easy
to see that the first main difference between the two FETs is the working principle
which are based on.
Figure 4.7: Conventional MOSFET Band Diagram in OFF-state and ON-state
44
Figure 4.8: TFET Band Diagram in OFF-state and ON-state
In fact, electrons in MOSFET flow from source to drain through a thermionic

emission physic, while Tunnel FET is based on band to band tunneling. So that,
when the conduction band of the intrinsic channel goes below the valence band of
the p-doped source, the electrons flow through the barrier toward the drain.
The main difference between a conventional MOSFET and the emerging TFET is
in the working principle, since the latter is based on the principle of band to band
tunneling phenomenon.
Before explaining the band to band tunneling principle, it could be useful to recall
the conventional MOSFET principle.
In the OFF state, the conduction in MOSFET is limited by the source side p-n
junction barrier which prevents the thermionic emission of carriers. In the ON state,
the source barrier is negligible enabling thermionic emission of carrieris [31].
On the other hand, for TFETs in OFF state the transmission probability is low
due to the wide source to channel tunnel junction barrier (low electric field) [31],
and as consequence a very low leakage current can be shown.
In the ON state the tunnel barrier get narrower, hence it henables carriers to tunnel
through it and go into the channel.
The TFET works by modulating the width of a tunneling barrier trhough the
45
gate, that is a different operation. In fact, in MOSFET the gate is used to modu-
lat the height of the barrier and carrier must surmount that height via thermionic
emission. The device is normally off, hence a wide barrier potential is present be-
tween the channel and the source, so that no BTBT occur and the leakage current
is present but in a very small contribution.
The main effect in the working principle difference can be seen with physical simu-
lation.
When a voltage is applied on the gate such that it is higher than the threshold
voltage, the barrier potential between the source and the channel gets narrower. In
this case, a tunneling current occurs and the device is the on state.
The transmission probability in a Tunnel FET is derived by the use of Wentzel
Kramers Brillouin (WKB) approximation:
p
4λ 2mEg3
Twkb ≈[ ] (4.8)
3qh[Eq + ∆φ]
where m is the effective mass, Eg is the band gap energy of the channel mate-
rial, λ is the screening tunnel length. The screening tunnel length is defined as the
extention of the transistion region at the source-channel interface [26], [28] .
While in MOSFET, the subthreshold swing is limited by the tail of the Fermi-Dirac
distribution of electrons in the n+ source region, in the TFET the Fermi tail is
cut-off by the band gap in the source-region.
Similar to a tunnel diode equation, the drain current is expressed as follows:

Z EV
I=A [FS (E) − FD (E)]T (E)NS ND dE (4.9)
EC
Where T(E) is the tunneling probability, Fs(E) and Fd(E) are the source and
drain side Fermi-Dirac distributin and Ns and Nd are the corresponding density of
States [31]. In particular, the WKB approximation is used for tunneling probability.
A MOSFET and TFET devices were realized in TCAD Synopsis Sentaurus. The
device parameters are shown in the table below:
46

Gate Length 40 nm
Silicon Thickness (tsi ) 10 nm
Table 4.1: Input data used for physical simulations of MOSFET and Tunnel FET
[1]
The IV characteristic has been simulated by the use of TCAD Sentaurus, in order
to check the theorical advantages of Tunnel FET over the conventional MOSFET.
Figure 4.9: Id - Vgs Transcharacteristic for comparison of MOSFET and TFET

behaviour
As can be seen from the plot and the table above, Tunnel FET shows a steeper
characteristic with respect to the SOI MOSFET, but also a very low Ioff current.
The main issue of this emerging FET is the low Ion and high threshold voltage,
47
Parameter MOSFET TFET

IOF F , A/um 1.81e-9 1.82e-18
ION , A/um 1.02e-3 1.55e-9
Gm , S/um 7.11e-4 6.35e-9
SS, mV/dec 111.85 86.51
Vth , V 0.45 1.34
Table 4.2: MOSFET vs TFET most important extracted parameters
which pin it not to be used for low power applications.
4.2.1 Temperature Analysis
Also a temperature analysis has been carried out. In fact, since TFET is based
on tunneling effect, it does not depend on temperature in on state. But, in the
subthreshold region the dominant effect is the recombination, so the current in off
state varies and it is strongly influenced by the temperature.
On the other hand, MOSFET characteristic degrades both in on and off state with
temperature.
48
Figure 4.10: MOSFET - Temperature analysis
Figure 4.11: SG TFET - Temperature analysis
49
In particular, varying the temperature from 200K to 400K, what can be demon-
strated is that in MOSFET case the off current increases of about 3 order of magni-
tude, SS increases by a factor of 3 (from 76,67 to 207,97), threshold voltge decreases
and Ion decreases more than one order of magnitude.
On the other hand, TFET shows a stable on current due to tunnel effect, an increase
of off current of about 8 order of magnitude but it still remains very low, and an
increase of SS of a factor 5, but until the room temperature the subthreshold slope
can overcome the thermal limit of 60mV/dec.
In Tunnel FET, the temperature dependence comes from the energy bandgap
term in the expression of tunneling current [17]:
3/2
|E|2 BEg
Ids = A p exp(− ) (4.10)
Eg |E|
Where E is the electric field along the channel, Eg is the bandgap, A and B are
Kane’s parameters.
In particular, the bandgap expression depends on temperature as follows:
αT 2
Eg (T ) = Eg (300) − (4.11)
T +β
Where α = 4.73e-4 eV/K and β = 636 K, while Eg (300) = 1.08 eV in the Silicon
case.
From the above expression, it can be noticed that with temperature raising,
bandgap reduces and as direct consequence drive current increases.
What can be noticed from a physical analysis of Tunnel FET is that, as in Esaki
Temperature Ioff, A/um Ion, A/um SS, mV/dec Vth, V

250 6.5e-7 1.12e-3 76.67 0.15
300 2.61e-6 1.02e-3 110.05 0.129
350 6.74e-6 9.39e-4 161.78 0.1
400 1.3e-5 8.71e-4 207.97 0.075
Table 4.3: MOSFET Temperature analysis
50
Diode case, both recombination and tunnel effect are combined in subthreshold re-
gion.
In particular, for low voltages applied to the gate,what can be shown is that recom-
bination and tunneling are combined in a sum. While, over the threshold voltage
tunneling is the dominant effect in the device.
For that reason, it is evident that at high gate voltages the temperature influences
weakly the device, since BTBT is weakly depedent temperature. It is more domi-
nant in the off state, since the SRH contribution dominates at low electric field and
has a strong temperature dependence [17].
On the other hand, for a MOSFET there are several parameters that are temperature-
dependent, such as threshold voltage, carrier mobility, saturation velocity and par-
asitic series resistance [17].
What can be noticed is that MOSFET drain current is more dependent on temper-
ature fluctuations in comparison to Tunnel FET. This difference can be noticed in a
particular way in th subthreshold region, since in the case of a MOSFET, the sub-
threshold current strongly depends on temperature through the square of intrinsic
carrier concentration term [18].
In fact, the intrinsic carrier concentration ni has a temperature dependence as:
Eg
ni = Na exp(− ) (4.12)
2kT
Furthermore, the drain current in a MOSFET is directly proportional to mobil-
ity and threshold voltage which has temperature-dependent terms [17].
Temperature Ioff, A/um SS, mV/dec

200 6.15e-24 20.71
250 9.7e-21 48.10
300 1.85e-18 60.94
350 2.96e-16 70.38
400 3.71e-14 115.04
Table 4.4: TFET Temperature
51
In fact, if an increase in temperature is present, the threshold voltage shows a de-

creasing, so it can be seen is that the drain current increase with temperature. On
the other hand, the temperature-dependent mobility term results in reduction of
drain current with increase in temperature due to lattice scattering which degrades
channel mobility [17].
52
4.3 Compact Models

To build TFET circuits and hence to make quite reliable performance predictions,
closed form models for the drain current that contain the basic physics expression
of the tunneling have been developed.
Next sections are dedicated to the description of quite a few compact models for the
TFET that have been implemented in Matlab.
4.4 Hao Lu Model

This is an analytical model, based on the Kane-Sze formula, that describes the be-
haviour in IV characteristic of TFETs.
The generalized formulation widely configurable since is not specific to a particular
TFET geometry [2].
The model is applied to the representation of a homojunction InAs double-gate n-
TFET and a broken gap AlGaSb/InAs SG TFET. The model is extended into other
operating regions by incorporating the ambipolar current and negative differential
resistance [2].
4.4.1 Modeling
The device is normally in off-state, so that the minimum conduction band of the
channel is over the maximum valence band of the source. In this case the device
is in off state, suppressing band-to-band tunneling. If a bias is applied to the gate,
the conduction band of the channel is shifted below the valence band of the source
[2] . Electrons in the valence band tunnel into empty states in the channel and the
transistor is ON [2].
Tunnel FET can be seen as a gated p-i-n junction. With this is mind, the main
expression for tunneling current in such model derives from a generalization for
three-terminals Zener tunneling in p-n junctions.
”The drain current is evaluated by integrating the product of charge flux and
53
the tunneling probability in the tunneling window, where the tunneling probability
is calculated by applying the WKB approximation” [2]:
Idt = af EVtw e−b/E (4.13)
Where Vtw is the tunneling window, which relates to the energy band crossed by
charges, E is the maximum electric field in the reverse biased junction, and a and b
are geometric coefficients:
s p
W Tch q 3
2m∗r 4 2m∗r Eg3
a= b= (4.14)
8π 2 ~2 Eg 3q~
Where m∗r is the reduced effective mass, W is the channel width, tch is the chan-
nel thickness, Eg is the semiconductor bandgap.
”The tunneling process is evaluated as a particle tunneling through a triangular

barrier, with a slope given by the electron charge times the electric field” [2]. ”The
detailed dependence of maximum electric field is taken to be linearly dependent on
Vgs and Vds” [2] by dropping the second and higher order terms:
E = E0 (1 + γ1 VDS + γ2 VGS ) (4.15)
Where E0 is the electric field at the source/channel junction if no bias is applied

to both gate and drain terminals, while γ1 and γ2 are model coefficients. From the
above formula it can be observed the electric field dependence from gate and drain
voltages. In fact, as gate bias voltage increases, the electric field rises at the source-
channel junction ”by both enlarging the voltage drop and narrowing the tunneling
barrier region” [2]. On the other hand, increasing the drain bias has a lesser effect
due to the screening of the drain field by the gate electrode.
The tunneling window expression can control in a continuous way both sub-
threshold and above threshold regions, as follows:
Vtw = U ln[1 + e(Vgs −Vth )/U ] (4.16)
54
In subthreshold region, the tunneling window has an exponenetial dependence

from gate voltage, while in the above-threshold region, ”it tends to a linear depen-
dence with the gate bias”.
The model is extended to the other regions, with several fitting parameters,
hence it accounts for bias dependent subthreshold swing, saturation, the superlinear
current onset, ambipolar conduction, and negative differential resistance. In this
work, only the current for both positive Vgs and Vds is considered.
Double Gate homojunction Tunnel FET were implemented in this model. Both
devices shares the same channel material that is InAs, so both energy gap Eg and
effective mass are the same. All fitting parameters are taken from Hao Lu model [2]
and are shown in the table below:

Gate Length 20 nm
Vth 0.15 V
γ0 0.64 -
γ1 0.01 m−1
γ2 1.89 m−1
Electric field E 0.507 MV/cm
λ 0.19 V
Table 4.5: Input data used for implementation of DG Tunnel FET [1]
55
I DS-VGS Transcharacteristic 10-5

4.5
4
10-5
3.5
3
log 10(I d ) , A/um
10-6
I d , A/um
2.5
2
10-7
1.5
1
10-8
Vds 0.1V 0.5
Vds 0.5V
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Vgs, V
Figure 4.12: InAs DG TFET implementation of analytical model [2]
The results above are in according to the refering paper. However, several issues
have been noticed in implementing such model. In fact, there is no possiblity to
exploit structures different from that described in the paper.
In particular, the model seems to be not dependent on doping concentrations, ge-
ometries or gate dielectric variations.
56
4.5 Praveen Model

Another model on the same WKB approximation was implemented in MATLAB,
with some differences. In fact this model allows to a gate engineering of the structure.
Gate engineering refers to changing the architecture of the gate terminal of the
TFET [15] by means of variation in screening length. The screening length or
natural length, is defined as the spatial extent of the electric field, or the length over
which an electric charge has an influence before being screened out by the opposite
charges around it [19].
It depends upon gate geometry:
p εsi
λSG = ( tsi tox ) (4.17)
εox
p εsi
λDG = ( tsi tox ) (4.18)
2εox
s
2tox
2εsi t2si (1 + tsi
) + εox t2si
λGAA = (4.19)
16εox
Based on Kane’s model, tunneling current general expression can be evaluated
as
Z
4q
I= T (E)[fS (E) − fD (E)]dE (4.20)
h
Hence:
p √ ∗ 3
W tch 2m∗ q 3 EV 4 2m Eg
− 3q~E
I= p exp (4.21)
4pi2 ~2 Eg
where the tunneling probability can be approximated as a triangle using the
WKB approximation:
q
3/2
4 2m∗r Eg
Ttunnel ≈ exp(− ) (4.22)
3q~ξ
where ξ is the uniform electrical field if triangle potential barrier is assumed.
It should be noted the BTBT occurs only if Ttunnel is high enough and there
57
are enough electrons at the starting side under Ev and enough empty states at the
ending side above Ec [15].
With this in mind, ”the current expression in the TFET model is an experi-
mentally well-estabilished equation for band-to-band, Zener tunneling in planar p-n
junctions” [2]. ”The two-teminal Zener tunneling behavior is then generalized to
three terminals by introducing physics-based expressions for the bias dependent tun-
neling window Vtw and a dimenionless factor f, which accounts for the superlinear
current onset in the output characteristic” [15].
Idt = af EVtw e−b/E (4.23)
s p
W Tch q 3 2m∗r 4 2m∗r Eg3
a= b= (4.24)
8π 2 ~2 Eg 3q~
The factor f is given by
1 − e−Vdse /Γ
f= Vthds −Vdse (4.25)
1+e Γ
Where Γ is the saturation shape parameter.
Vds 2 √ 2
r
Vds
Vdse = Vdsmin [ + ∆2 + ( ) − ∆ + 1] (4.26)
2Vdsmin 2Vdsmin
Vthds = Λtanh(Vgs ) (4.27)
The tunneling window is given by
Vtw = ln(1 + e(Vgs −Vth )/U ) (4.28)
Tunnel FETs garnered interest by its virtues of reduced SCEs, SS and low power
consumption [15]. A drawback is the very low Ion current.
One of the possible solutions to enhance Ion includes structural engineering: a DG
or GAA structure could be a solution of Ion with no impact on Ioff current. GAA
TFETs have shown a great deal in performance improvement , compared to SG and
58
DG.
Tunnel FET is an emerging device in afield of continuing research. Obviously, MAT-
LAB analysis allow to limited explorations. Thus, detailed simulations using ad-
vanced TCAD tools have to been carried out.
59
4.5.1 Matlab Implemantation

Same structure as before was used to validate this model, since the approximation
used is the same. The same conclusions have been done for the Praveen model, since
both analytical models used WKB approximation to derive the current expression
of the tunnel fet. So, the model seems to lead an accurate result, but it doesn’t take
care of several parameter. The model can’t be considered as a universal model for
different structures, materials and sizes.
I ds-Vgs Transcharacteristic
10-6
10-8
log(I d ) , A/um
10-10
GAA
10-12 DG
SG
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Vgs, V
Figure 4.13: ID − Vgs characteristic of TFET for the model implemented
Parameter SG DG GAA
IOF F , A/um 1.38e-13 1.95e-13 3.01e-13
ION , A/um 3.03e-6 4.29e-6 6.62e-6
Table 4.6: Most important extracted parameters from Praveen model
60
4.6 Dash Model

The current is evaluated using initial and final tunneling length of band-to-band pro-
cess. The tunneling process of charge carriers has been realized analytically using
initial and final tunneling point. Both the tunneling points play important roles for
the estimation of the DC parameters such as SS, ON current and transconductance
[7].
4.6.1 Modeling
First, the potential profile in the channel is studied, in order to derive the surface
potential. Here the effect of electron space charge has been neglected due to the
light doping nature of intrinsic channel [7].
With this in mind, the surface potential is seen as the potential profile along the
x-axis with specified boundary conditions at the gate-channel interface, and it plays
a crucial role in the analysis of the tunneling path and drain current [7].
Vs (x) = CO eαx + C1 e−αx − (VF B − Vgs ) (4.29)
where:
r
εox
α= (4.30)
εsi tsi tox
1
C0 = [−Vbi (1 + e−α` ) + (VF B − Vgs )(1 − e−α` ) + Vds ] (4.31)
2sinh(αL)
1
C0 = − [−Vbi (1 + eα` ) + (VF B − Vgs )(1 − eα` ) + Vds ] (4.32)
2sinh(αL)
At the same time, the eletric field along the x-axis plays an important role for
the evaluation of the tunneling volume. It is derived by differentiating the surface
potential, as follows:
61
∂∅(x,y)
Ex = − = −C0 αeαx + C1 αe−αx (4.33)
∂x
If the applied gate voltage is zero, there is a wide potential barrier between the
source and channel region there is no BTBT of charge carriers, hence the device is
OFF.
When positive gate voltage is applied, the potential barrier between source and
channel region gets narrower gradually until the gate voltage exceeds the threshold
voltage so that the potential barrier becomes narrow enough to allow tunneling of
charge carriers.
In the ON state the conduction band of channel goes below with respect to the
valence band of source, so that it enables the charge carriers to tunnel from source
to drain. However the charge carriers move to the drain end by the process of drift
diffusion mechanism [7].
Tunneling path is defined as the distance between L1 and L2 along the channel
length. It is responsible for BTBT among carriers.
In particular, L1 is defined as the initial tunneling length from the source which
indicates the start of BTBT tunneling process and can be evaluated as [7]:
√
1 Z+ Z 2 − 4C0 C1
L1 = ln( ) (4.34)
α 2C0
Where
Eg
Z = Vbi + + (VF B − Vgs ) (4.35)
q
While, the final tunneling length L2 indicates the end of the tunneling process
and can be evaluated as the value in the channel region at which surface potential
is maximum [7], so:
r
1 C1
L2 = ln( ) (4.36)
α C0
The drain current in the BTBT process can be evaluated by the evaluation of
band to band generation rate, which is defined as follows:
62
GR (x,y) = AK Eavg Ex e−BK /Eavg (4.37)
where AK and BK are the Kane’s tunneling-dependent parameters, while the

average electric field is expressed as
Eg
Eavg = (4.38)
q`path
Where `path is the length of tunneling path varying from L1 to L2.
So, the tunneling current can be determined by integrating the band to band
generation rate over the TFET volume:
Z Z
ID = q GR (x,y)dxdy (4.39)
By sobsituiting the Generation rate expression in the integral above, it can been
obtained:
Z tsi Z L2
ID = q AK Eavg Ex e−BK /Eavg dxdy (4.40)
0 L1
Again, sobstituting the electrical field expression, the drain current final expres-
sion can be get:
Z tsi Z L2
Eg
ID = q AK (−C0 αeαx + C1 αe−αx )e−BK qx/Eavg dxdy (4.41)
0 L1 qx
63

Since the explained model seems to be the most reliable analytical model to describe
the Tunnel FET behaviour, also TCAD Sentaurus simulations were done to compare
the results.
In particular, the same structure was used, by varying the gate engineering.
The analytical model results in agreement with physical simulations, except for
the subthreshold region.
The reason is that the Tunnel FET in subthreshold region is dominated by SRH
more than tunneling, but the analytical model takes into account the tunneling ef-
fect only. It results in a quite difference of off region.
It could be useful to consider an analytical model that consider separately sub-

threshold and above-threshold region, in order to integrated such model in the Dash
Matlab model that has been implemented in this work.
I d - V gs Transcharacteristic
10-8
10-10
10-12
log(I d ), A/um
10-14
10-16
SG TFET
DG TFET
10-18
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Vgs, V
Figure 4.14: DG TFET SiO2 vs SGTFET Transcharacteristic Comparison
64
Parameters SG TFET DG TFET GAA TFET

Ioff, nA/um 2.73e-18 7.7e-17 8.94e-18
Ion, uA/um 5.12e-10 6.98e-9 4.66e-9
Table 4.7: Important parameters extrcted from Model varying the geometric struc-
ture
Figure 4.15: Transcharacteristic comparison among different structures
Parameters SiO2 Si3N4 HfO2

Ioff, nA/um 7.e-17 1.91e-17 4.08e-17
Ion, uA/um 6.98e-9 2.17e-8 1.79e-7
Table 4.8: Important parameters extrcted from Model varing gate dielectric
65
Chapter 5
Sentaurus Simulations
Several simulations of Tunnel FETs structures had been carried out in TCAD Syn-
opsis Sentaurus in order to check and validate the analytical models implemented
in MATLAB.
A further aim of such simulations was to verify the possibility of new emerging de-
vices like Tunnel FET to overcome the main issues of FinFET. In fact, from previous
simulations of different FinFET structures, it had been emerged several problems
and limits in scaling such devices below 20nm.
With this in mind, different structures of Tunel FET were investigated going in deep
with scaling and analyzing new problems emerged from this kind of analysis.
This chapter is divided into several parts as follows:
• Simulation of single-gate (SG) Tunnel FET investigating on electrostatic anal-

ysis, temperature analysis and capacitance analysis
• Simulation of double-gate (DG) Tunnel FET varying the gate dielectric ma-
terial
• Simulation of gate-all-around (GAA) nanowire Tunnel FETs with high-k di-

electric
• Comparison of such physical simulations with analytical model implemented

in the previous part in MATLAB
66
5 – Sentaurus Simulations
5.1 Single Gate Tunnel FET
Firstly, a Single Gate TFET was simulated in TCAD Synopsis Sentaurus. As can
be shown in the following picture, a SG TFET is realized in CMOS’like structure.
The main feature of such device is the different doping type between the source and
drain terminal.
Figure 5.1: SG Tunnel FET Structure
The device parameters are: channel length Lg = 40nm, source and drain length
Lsd = 20nm, oxide thickness tox = 2nm using SiO2 material, source doping con-
centration Ns = 1e20 cm-3, drain doping concentration Nd = 5e19 cm-3, channel
doping concentration Nchannel = 1e15 cm-3, silicon layer thickness of tsi = 10nm.

Gate Length 40 nm
Source doping concentration (Ns ) 1020 cm−3
Drain doping concentration (Nd ) 5 ∗ 1019 cm−3
Table 5.1: Input data used for physical simulation of SG Tunnel FET [1]
67
A cross section of the device is shown in the figure below:
Figure 5.2: Cross section of SG Tunnel FET used for TCAD simulation
The above cross section is useful to better understand the difference in doping
concentration of such devices. In fact, the source terminal is p-doped, while the
drain terminal is n-doped, with an intrinsic channel. As it was explained in the
previous sections, the device is normally off. Then, it is in on state when a drain
bias is applied so that the concuction band of the intrinsic channel goes below the
valence band of the source p region.
Band diagrams of OFF and ON state respectively from Sentaurus inspection are
shown below.
The figure below shoes the variation of electric field alog the channel varying the
gate voltage.
It can be observed that both in OFF and ON state, the electric field takes its
maximum value along the x-axis at the source/channel interface.
In fact, thanks to the high electric field, the electron charges tunnel from source to
68
Figure 5.3: Energy band diagram of n-channel TFET in OFF-state (Vgs < Vth ) and
ON-state (Vgs > Vth )
Figure 5.4: Electric Field distribution in OFF-state and ON-state
drain with an higher probability.

The transfer characteristic of a SG TFET is simulated in TCAD Synopsis Sen-
taurus, with Vds = 1V, and varying the gate voltage from 0V to 2 V.
For small Vgs, the current at the drain terminal lightly decreases until the named
”peak valley”. Then, increasing the gate voltage beyond the threshold voltage, the
current increases in an exponential way with a fixed drain bias.
69
I DS-VGS Transcharacteristic 10-10

10-8 7
6
10-10
5
log(I d ) , A/um
10-12
4
I d , A/um
3
10-14
2
-16
10
1
10-18 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Vgs, V
Figure 5.5: Drain current as a function of gate voltage in linear and logaritmic scale
More detailed results are shown in the following table:
Parameters Results
Iof f ,nA/µm 1.85e-9
Ion ,µA/µm 1.55e-3
SS,mV /dec 49.55
Vth ,V 1.46
Table 5.2: Important electrical parameters for Vds = 1V
From this first simulation, it can be observed how the leakage current in Tunnel
FET is effectively lower than in MOSFET. However, the threshold voltage is quite
high and as a consequence the tunneling current it doesn’t go further the nA order
of magnitude. This represents the main issue of such emerging device, since it can’t
be seen as an applicable switch.
A capacitance analysis was then exploited in TCAD Sentaurus, in order to in-

vestigate on possible differences in capacitance behaviour among FETs.
70
The figure below represent the gate capacitance variation varying the drain voltage.
Figure 5.6: SG Tunnel FET Gate Capacitance varying drain voltage Vds
The intrinsic capacitance Cgg increases with the increase of the gate voltage in
on state. The increased capacitive effect is due to the combined enhancement of both
drain capacitance (Cgd) and source capacitance (Cgs) [7]. From the graph above,
it can be observed that the main contribution to the intrinsic gate capacitance is
given by the drain capacitance. In fact, as the drain voltage V − ds increases, the
gate capacitance drecreases. But, low capacitances limit the cut-off frequency of the
device.
A temperature analysis of a DG TFET has been done, in order to evaluate the
temperature dependence in the device.
Since TFETs are based on tunneling phenomenon, they are weakly dependent
on temperature.
In particular, in the case of indirect tunneling there is a dependence on phonon-
electron interactions for the change in momentum from the Γ -valley to the X-valley
71
minimum. The phonon occupation obeys to Bose-Einstein statistics, and the num-
ber of phonons is sensitive to temperature [30]. But, indirect tunneling can occur
even with a zero phonon occupancy due to phonon emission by the tunneling par-
ticle, hence tunneling does not depend on temperature in an exponential way.
TCAD Synopsis Sentaurus was used to carry out simulations of FET devices
varying the temperature.
Figure 5.7: SG TFET Temperature analysis
72
Temperature, K SS, mV/dec Ioff, nA/um

200 20,71 6,15e-15
250 44,10 9,70e-12
300 49,55 1,85e-9
350 70,35 2,96e-7
400 115,04 3,71e-6
Table 5.3: Temperature variation of electrical parameters for Vds = 1V
In order to check the reliability of Matlab models, a transcharacteristic of such

device was simulated in TCAD Sentaurus, with Vd = 1V and Vg = 2V. The same
behaviour of a Tunnel Diode was exploited.
10-8
10-10
10-12
log(I d ), A/um
10-14
10-16
Sentaurus
Matlab
10-18
0 0.5 1 1.5 2
Vgs, V
Figure 5.8: SG Tunnel FET Transcharacteristic
The model agrees with the physical simulation.It presents a difference in the sub-
threshold slope, even if the value of leakage current is the same. This is probably
due to the accuracy of the model: such model presents an accurate analysis of the
device, describing the tunneling current. The current is calculated by integrating
the generation rate in the device volume. Such generation rate is evaluated as the
electron probability to tunnel from the minimum tunneling distance to the maxi-
mum tunneling distance. The current is derived from this tunneling path, so the
73
subthreshold region where the recombination is the major contribute to the current
is omitted. This can be explain this disagreement in the curve slope.
The Ion/Ioff ratio has been raised of about 4 order of magnitude with respect to
the MOSFET case. The subthreshold swing goes below the conventional 60mV/dec
due to the working principle which TFET is based on.
However, the main issue of a Tunnel FET realized in Silicon material is the low
ON current.
74
5.2 Double Gate Tunnel FET

The study of Tunnel FET was exploited by simulating a double gate structure. The
device parameters are the same of the Single Gate structure, with the only difference
of doubling the gate.
p+ Source Front Gate

L
Tox
Tsi
Tox n+ Drain
Back Gate
Figure 5.9: DG Tunnel FET structure and cross section in Sentaurus
The device parameters are:

Gate Length 40 nm
Table 5.4: Input data used for physical simulation of DG Tunnel FET [1]
The same analysis as before has been carried out for a double gate structure. So,
75
several comparison were done in order to investigate the behaviour of the Tunnel
FET varying the gate engineering and the gate dielectric materials.
Figure 5.10: Comparison of electrostatic analysis between a SG and a DG TFET
From figure(5.10) the difference in the I-V characteristic can be noticed between
a single-gate and double-gate structure. The table below recaps the improvement
in current and subthreshold slope by varying the gate engineering.
Parameters SG TFET DG TFET

Ioff, A/um 1.82e-18 5.58e-17
Ion, A/um 1.55e-9 1.84e-8
SS, mV/dec 76.11 69.6
Vth, V 1.34 1.33
Gm, S/um 6.35e-9 6.88e-8
Table 5.5: SG vs DG TFET Extracted parameters from TCAD
For such comparison, a SiO2 gate dielectric material is used, but also the same
gate metal material and structure parameters. From the results above, it can be
observed that gate coupling leads to an increase of drain current of one order of
76
magnitude. Also the off current increases with respect to the case of a SG TFET,
but it still remains a very low leakage current.
The results in TCAD were compared with MALTAB simulations, using the same
model as for SG TFET. Since the model is the same, the same accuracy can be
noticed, bu also the same problem of accuracy in the subthreshold region.
10-8
10-10
log(I d ), A/um
10-12
10-14
10-16
Sentaurus
Matlab
10-18
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Vgs, V
Figure 5.11: Validation of MATLAB model with physical simulation
A Double-gate structure presents several advantages with respect to a single-gate

structure. However, the tunneling current still remains too low for every possible
applications. Several solutions can be studied in order to overcome such limit.
One among them could be change the gate dielectric material, since the tunneling
probability depends from this factor in an exponential way.
With this in mind, an anlysis in both MATLAB and TCAD were carried out,
where the gate dielectric were varied. In the figures below, the characteristic of a
DG TFET with respectively Si3N4 and Hfo2 is represented.
It can be observed that the difference in the behaviour between the model and the
physiscal simulation get less evident with increasing the permittivity of the dielectric.
Furthermore, the evident dependency of current on gate dielectricis shown below.
From figure (5.13) it can be seen the increase of on current with increasing of
permittivity, but also a steeper slope of the characteristic.
77
I d -Vgs Transcharacteristic I d -Vgs Transcharacteristic

10-6
10-8
10-8
10-10 10-10
log(I d ), A/um
log(I d ), A/um
10-12 10-12
10-14 10-14
10-16 10-16
Sentaurus Sentaurus
Matlab Matlab
-18 -18
10 10
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Vgs, V Vgs, V
Figure 5.12: DG Tunnel FET varying gate dielectric, (a) Si3n4 and (b) HfO2
Figure 5.13: DG Tunnel FET Gate Dielectric Comparison with physical simulation
The most evident results stays in 38% in SS reduction, a 3 order of magnitude

in increase of Ion current and in a not so evident Vth threshold voltage variation.
The above table shows the results of the comparative study of the current depen-
dency on gate dielectric variation. It can be seen that the most relevant variation
in device parameters is the growing of on current of about 3 order off magnitude,
and the reduction of the SS of about 38%.
78
Parameters SiO2 Si3N4 HfO2

Ioff, A/um 5.48e-17 9.03e-16 1.16e-14
Ion, A/um 1.84e-8 1.3e-7 7.36e-7
SS, mV/dec 69.6 52.2 36.72
Vth, V 1.33 1.37 1.36
Gm, S/um 6.88e-8 4.26e-7 1.5e-6
Table 5.6: Gate dielectric variation
A temperature analysis of a DG TFET has been done, with an high k dielectric,

in order to evaluate the temperature dependence in the device.
Figure 5.14: DG TFET Temperature analysis
79
Temperature, K SS,mV/dec Ioff, nA/um

200 36.15 8,9e-15
250 33.61 9,89e-15
300 34,72 1,09e-14
350 51,01 1,29e-14
400 47,75 1,25e-13
Table 5.7: Temperature variation of electrical parameters for Vds = 1V
From the results above, it can be noticed that the variation in off current is less
evident than in sg tfet case, even if the recombination remains the major contribution
in that region. As before, the on current doesn’t vary with the temperature as in the
case of a conventional MOSFET. Except for the valley of negative conductance, the
current characteristic seems to be not influenced from the temperature variation.
80
5.3 Gate All Around Tunnel FET

A cylindrical structure for Tunnel FET was realized in TCAD Sentaurus, based on
the same parameters as for the case of single gate and double gate structure.
p++ Si Source
HfO2
p+ Si Channel
Aluminum
n++ Si Drain
Figure 5.15: GAA Tunnel FET Sentaurus Structure with HfO2 gate dielectric and
doping concentration

Gate Length 40 nm
The same analysis as before was carried out. In particular, the structure of a
gate-all-around was simulated, varying the dielectric from SiO2 to HfO2. In this way,
it was further demonstrate the improvement in tunneling current and subthreshold
slope if a high k dielectric is used.
In fact, as can be noticed, the current again increase by an order of magnitude,
and a steeper slope characteristic can be appreciated. However, an issue of such
81
I d - V gs Transcharacteristic
10-8
10-10
10-12
log(I d ), A/um
10-14
10-16
SiO 2
HfO2
-18
10
0 0.5 1 1.5
Vgs, V
Figure 5.16: GAA Tunnel FET characteristic varying gate dielectric
Parameters SiO2 HfO2

Ion, A/um 3.19e-10 1.77e-9
Ioff, A/um 4.48e-21 1.26e-18
Vth, V 1.11 1
SS, mV/dec 51.9 30,61
Table 5.9: Extracted parameters from physical simulation of GAA TFET
device is the on current that is even lower with respect t the double gate structure.
Even if a gate wrapping the channel could better control it, for a tunnel FET it
seems to have no effect on the tunneling current.
82
5.3.1 NW TFET
The analysis can be further carried out, by further scaling the device. SO, a nanowire
tunnel fet was simulated in TCAD Sentaurus, to investigate on deep scaling of such
emerging devices.
The device parameters are:

Gate Length 40 nm
p++ Si Source
HfO2
p+ Si Channel
Aluminum
n++ Si Drain
Figure 5.17: GAA Tunnel FET Sentaurus Structure with HfO2 gate dielectric and
doping concentration
83
10-10
10-12
log(I d ), A/um
10-14
10-16
10-18
Sentaurus
Matlab
10-20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Vgs, V
Figure 5.18: NW TFET IV characteristic for validation of Matlab model
Parameters Results with HfO2, Vds = 1V

Ion, A/um 3.39e-10
Ioff, A/um 1.26e-20
Vth, V 0.8
SS, mV/dec 35.54
Table 5.11: NW TFET extracted parameters from physical simulation
The case of a nanowire tfet represents the best result for the validity of the
analytical model. As shown in the figure above, the model is in agreement with the
TCAD simulation.
However, even if a deep scaling has been carried out, the poor tunneling current
remains the main issue in this device. Probably, it can be improved by study the
trend of variation of the diameter in a proportional way to the channel length. The
Ion over Ioff ratio is optimum since it reaches about 101 0, but the low current didn’t
allow the device to be applicable in a circuit.
What is strange is the fact that a double gate structure presents a better be-
haviour rather than a gate all around structure. Probably the nanowire tfet is not
the way to further improve the current.
84
5.4 GAA FET vs GAA TFET
The cylindrical structure allows to compare FET and TFET electrostatic behaviour.
A simulation study was carried out in order to evaluate the advantages of a
device with respect to the other one. The structure is the same as before, the only
difference stays in the doping types between the tunnel fet and the fet. So, a high
k dielectric is used for the gate, the channel length is of 40nm and the diameter of
10nm, with a temperature of 300K.
p++ Si Source n+ Si Source
HfO2 HfO2
p+ Si Channel Si Channel
Aluminum Aluminum
n++ Si Drain n+ Si Drain
Figure 5.19: GAA Tunnel TFET and GAA FET

Gate Length 40 nm
85
Figure 5.20: Comparison of electrostatic analysis between cylindrical structures of

FET and TFET
Parameters GAA FET GAA TFET

Ion, A/um 2.22e-5 1.72e-9
Ioff, A/um 1.32e-6 2.92e-18
Gm, S/um 4.11e-5 3.58e-9
SS, mV/dec 98.38 56.91
Table 5.13: Comparison between GAA TFET and GAA FET
The table above shows in evidence the main issues of both devices. In particular,
TFET has a far better Ion over Ioff ratio, since it is of the order of 109 compared to
10 of the conventional FET. This results in a steeper slope, in fact the TFET leads
to go below the FET thermal limit of 60mV/dec. However, the TFET shows a low
current in tunneling compared to a FET, and this is an important limit which has
to be overcame.
86
Chapter 6
Tunnel FET issues and solutions
Tunnel Field Effect Transistors show several advantages with respect to FinFETs
and could be presented as promising device which is able to overcome main conven-
tional MOSFET issues.
In fact, from previous studies it was noticed a better Ion over Ioff and a steeper
slope in the Id Vg characteristic.
Also a different behaviour on temperature variations was investigated. With this in
mind, TFETs could be the best candidate for low power applications.
On the other hand, what is evident from the simulations in the prevoius chapters is
that TFET on current is too low (order of nA) to be considered for any technological
applications, even if the low Ioff is good for low power dissipation.
Low Ion current is due to carrier tunneling between energy states of different sim-
metry, which causes a large tunneling resistance [37].
Several issues have been emerged from the analysis of Tunnel FET in chapter
5. The most evident and relevant is the low value of currnt in tunneling region. In
fact, such limit doesn’t let the device to be applicable as a switch and exploit the
advantage of low leakage current and steeper slope.
Several solutions have to be studied in order to overcome such limit. It could
be useful to investigate on the expression of tunneling current. In this way, indeed,
the main contributions to tunnel could emerge and could be improved. According
to the triangular Wentzel-Kramer-Brillouin (WKB) approximation:
p
4λ 2m∗ sqrtEg3
TW KB ≈ exp(− ) (6.1)
3q~(Eg + ∆φ)
where m∗ is the effective mass, Eg i the band gap, λ is the screening tunneling
length, and ∆φ is the potential difference between the source valence band and the
87
6 – Tunnel FET issues and solutions
channel conduction bands.

In particular the screening tunneling length varies with the device geometry.
From the formula above, it can be easily noticed that in order to increase tun-
neling probability, the effective mass, the band gap and the screening lenght should
be reduced.
With this in mind, reserarches have been investigated on several solution to

improve current in the on state. In this section, simulation in TCAD Sentaurus
have been carried out in particular for heterostructure TFET and STBFET.
88
6.1 Si/Ge Hetero-structure Tunnel Field Effect

Transistor
A heterostructure is a junction between two different semiconductor material [38].
The most evident feature of semiconductor materials are band gap, effective mass
and electron affinity, so that the interface properties of such junction could change
if two different material are used. In particular, the study of energy band diagrams
related to the change in physical properties is of great interest. In heterostructure
such as Si/Ge, the electron affinity model is used to relate the electron affinities and
band gap for the energy band diagram study, such that:
∆EC = χ1 − χ2 ∆EV = (Eg,1 + χ1 ) − (Eg,2 + χ2 ) (6.2)
where χ is the electron affinity and Eg is the band gap energy of the materia,
and 1 and 2 reder to different semiconductor material.
Ec
Ev
Figure 6.1: Band Diagram of a conventional TFET (left) and a Si/Ge heterostruc-
ture (right) in ON-state
Hetero-structure TFETs are realized by the use of lower band gap materials for
source region, which enhance drive current behaviour.
In fact, if a material with a lower band gap and effective mass with respect to Silicon
89
is used in the source region, the tunneling probability will increase, according to
(6.1). So materials like Germanium should replace Silicon in the source region so
that the tunneling barrier widht can be reduced.
Indeed Germanium has band gap of 0.66 eV compared to 1.12eV of Silicon, and
effective mass of 0.06 compared to 0.02 of silicon. On the other hand, a reducing
the screening length means increase the modulation of the channel bands by the
gate, and thus smaller barrier for tunneling [35]. Since the screening length depend
on device geometry, it is found to be the smallest in the case of a gate-all-around
structure.
6.1.1 Physical simulation of Si/Ge heterostructure TFET

A hetero-structure Si/Ge nanowire TFET have been realized and simulated in
TCAD Synopsis Sentaurus. 3D Physical simulations have been carried out to vali-
date the theorical study, so that an imporvement in on current should be observed.

Gate Length 30 nm
Oxide Thickness (tox ) 1.5 nm
Drain doping concentration (Nd ) 1018 cm−3
Table 6.1: Input data used for physical simulation of heterostructure Tunnel FET
[1]
The heterostructure NW TFET is realized with a gate-to-source overlap of 5nm,

an HfO2 gate dielectric of 1.5nm, the diameter of 8nm. A p++ Germanium material
was used for source region, with acceptor active concentration of Na = 1e19. Silicon
was used for both channel and drain region. The channel was p+ doped with
Na 1e18, while drain was n-doped region with donor concentration of 1e19. As
for [35], ”A dynamic nonlocal band-to-band (BTB) tunneling model is utilized in
90
conjunction with Shockley-Reed-Hall (SRH) recombination, drift-diffusion physics,

and Fermi statistics” [35].
p++ Ge Source
HfO2
p++ GeSi
p+ Si Channel
Aluminum
n++ Si Drain
Figure 6.2: Hetero-structure
Figure 6.3: Hetero-structure
The figure below represents the simulated heterostructure at Vd = 1V, while the
table recaps the electrical parameters extracted from the simulation.
91
Figure 6.4: Transcharacteristic of a Si/Ge heterostruture with Vds = 1V
It can be observed that the current in on state improves of

3 order o magnitude with respect to homojunction nanowire tfet, while mantaining
a steep slope and a low leakage current. This is a great result for the consideration
of TFET as applicable device, and could be a starting point for an enanched tfet.
Parameters Values
Ioff, A/um 1.27e-14
Ion, A/um 1.39e-6
Vth, V 0.86
SS, mV/dec 46.05
Table 6.2: Extracted parameters from physical simulation of Si/Ge heterostructure
92
6.2 Sandwich Tunnel Barrier FET

Many other considerations must be done and other solutions have to be taken into
account.
Different materials which present lower bandgap, such as SiGe or InAs, or the use
of high-k dielectric and vertical TFET were studied in the prevous section, in order
to enhance on current performances.
In fact, both tunneling area and tunneling probability influence the current in on
state. So, the conventional planar TFET even if it is an heterostructure device, can
not provide an high drain current because of a small tunneling area [37].
A Sandwich tunnel barrier FET structure is realized by creating a source region

p-doped which is sandwiched in between two drain regions, and which is under the
gate and channel region [37].
This allows to enhance Ion current performances mantaining a low Ioff current, that
result in an highr Ion over Ioff ration with respect to MOSFET or planar TFETs.
This is mainly due to the difference in gate dependency: in particular, tunneling
current reduces with gate area instead of reducing with gate width.
While, the off current is dependent on the spacers thickness.
6.2.1 Physical simulation of STBFET

The STBFET structure is simulated by the use of TCAD Synopsis Sentaurus, in
order to investigate this modified TFET behaviour.
The channel region is an epitaxially grown Si layer which is 2nm thick. The source
region is under the gate and channel region [36] and is p-doped with Boron dop-
ing of 2e20 cm-3. The drain regions are on both sides of the high-k spacers and
they are n-doped with Phosphorus doping of 5e19 cm-3. The gate is realized with
Aluminum, with workfunction of 4.1eV, while HfO2 material is used for the gate
dielectric. The spacers are realized using high-k HfO2 material, in order to reduce
the channel resistance.
The channel length is of 30nm, while the gate dielectric is 0.4nm thick.
A simulation was carried out, with Vds at 1V.
93
Device parameters are riassumed in the table below:

Gate Length 30 nm
Oxide Thickness (tox ) 0.04 nm
Source doping concentration (Ns ) 2 ∗ 1020 cm−3
When the device is off, there is no channel inversion under the gate and spacer
region [36]. With the increas of gate bias, an inversion layer forms in the channel
region, so that the device can be seen as a p+ n+ reverse biased diode with a tunnel
distance equivalent to the epilayer thickness [36]. An higher tunneling current is
obtained as a consequence of the eduction in tunnel distance and the use of high-k
spacers.
Figure 6.5: STBFET Structure
In STBFET, the band diagram behaviour looks different with respect to TFET.
In fact, the band bending at the tunnel junction is negligibly small [36]. With the
94
increase of Vds, a large band bending in the soacer-chanel region can be seen, with
negligible drop across source-channel region [36].
The voltage dropp across the tunnel junction does not vary with the drain bias,
there is negligible modulation of the tunnel distance [36].
The transcharacteristic of a n-channel STBFET is shown in the figure below.
Figure 6.6: Transcharacteristic of a STBFET
The simulated Ion is of the order of uA / um when compared with nA/um

in a conventional silicon TFET. The off current is quite higher with respect to
conventional TFET, which allows an higher SS.
Parameters Values
Ioff, A/um 7.22e-13
Ion, A/um 2.48e-6
Vth, V 0.3
SS, mV/dec 28.43
Table 6.4: Extracted parameters from the physical simulation of STBFET
95
Chapter 7
Conclusions
The work of thesis was divided in two main parts. After a brief introduction on
FinFET working principle, the main structures and their issues, the firs part was
dedicated to the modelling of Double Gate and Triple Gate FinFETs.
It had been implemented several models in order to study the scaling behaviour of
such Multiple-Gate Transistors. Validation of the model through simulation proves
the accuracy and the computational effciency of the resulting model [31].
The second part of the thesis is devoted to investigate on new emerging devices
as alternative to conventional FETs and that can overcome their main issues, as
TFETs.
Again, after a study of the working principle of TFET, a modelling of different
Tunnel FET structures and relative simulations were done to validate the reliability
of the analytical models.
The characterization of such devices includes IV behaviour, a capacitance analysis,
a temperature analysis of DG,TG,GAA FinFET, and SG,DG,GAA TFETs. In
particulare a comparative study of GAA FinFET and TFET behaviour was done in
order to underline the promising benefits of Tunnel FET.
The description of the obtained results occupies the last part of the work together
with the discussion of the main theoretical insight gained with the conducted study
[31].
96
7 – Conclusions
7.1 Future Work

Future work behind the presented thesis could focus on analytical modelling of
TFET in subthreshold region, going in deep with scaling of nanowire structures, for
example by the use of NEGF .
Then, a stability analysis can be performed to compare the two emerging devices
analyzed, so that a simulation at circuit level can be done.
In fact, it can be interesting to investigate on the TFET behaviour in a NAND gate
or in an analog differential amplifier circuit.
Finally, other emerging structures such as heterostructures, strained-Ge can be stud-
ied and analyzed, in order to overcome Tunnel FET main issues.
97
Bibliography
[1] Taur, Y., Liang, X., Wang, W., & Lu, H. (2004). A continuous, analytic drain-
current model for DG MOSFETs. IEEE Electron Device Letters, 25(2), 107-109.
[2] Lu, H., Esseni, D., & Seabaugh, A. (2015). Universal analytic model for tunnel
FET circuit simulation. Solid-State Electronics, 108, 110-117.
[3] Bhole, M., Kurude, A., & Pawar, S. (2013). FinFET-benefits, drawbacks and
challenges. Int. J. of Engineering, Sciences and Research Technology, 2, 3219-
3222.
[4] Fasarakis, N., Tsormpatzoglou, A., Tassis, D. H., Dimitriadis, C. A., Papathana-
siou, K., Jomaah, J., & Ghibaudo, G. (2011). Analytical unified threshold voltage
model of short-channel FinFETs and implementation. Solid-State Electronics,
64(1), 34-41.
[5] Fasarakis, N., Karatsori, T. A., Tsormpatzoglou, A., Tassis, D. H., Papathana-
siou, K., Bucher, M., ... & Dimitriadis, C. A. (2014). Compact modeling of
nanoscale trapezoidal FinFETs. IEEE Transactions on Electron Devices, 61(2),
324-332.
[6] Fasarakis, N., Tsormpatzoglou, A., Tassis, D. H., Pappas, I., Papathanasiou,
K., Bucher, M., ... & Dimitriadis, C. A. (2012). Compact model of drain current
in short-channel triple-gate FinFETs. IEEE Transactions on Electron Devices,
59(7), 1891-1898.
[7] Kumari, P., Dash, S., & Mishra, G. P. (2015). Impact of technology scaling
on analog and RF performance of SOI-TFET. Advances in Natural Sciences:
Nanoscience and Nanotechnology, 6(4), 045005.
[8] Kuzmicz, W. (2017, June). The future of CMOS: More Moore or the next big
thing?. In Mixed Design of Integrated Circuits and Systems, 2017 MIXDES-24th
International Conference (pp. 21-26). IEEE.
[9] Saini, G., & Rana, A. K. (2011). Physical scaling limits of FinFET structure: A
simulation study. International Journal of VLSI design & communication Sys-
tems (VLSICS), 2(1).
[10] Katiyar, J. (2016). Atlas Based Simulation of Double Gate Tunnel Field Effect
98
Bibliography
Transistor (Doctoral dissertation).

[11] [Online]. Available at: http://public.itrs.net/
[12] https://blog.lamresearch.com/tech-brief-finfet-fundamentals/
[13] S. Donati, Microelectronic Devices, DET, Politecnico di Torino, 2017.
[14] G. Piccinini, Integrated System Technology, DET, Politecnico di Torino, 2017.
[15] Praveen, C. S., Ravindran, A., & Varghese, A. (2015). Analysis of GAA Tun-
nel FET using MATLAB. In IJCA Proceedings on International Conference on
Emerging Trends in Technology and Applied Sciences (pp. 30-35).
[16] Rana, P. (2017). Electrostatic Analysis of Gate All Around (GAA) Nanowire
over FinFET (Doctoral dissertation, Arizona State University).
[17] Narang, R., Saxena, M., Gupta, R. S., & Gupta, M. (2013). Impact of tem-
perature variations on the device and circuit performance of tunnel FET: A
simulation study. IEEE transactions on Nanotechnology, 12(6), 951-957.
[18] Arora, N. D. (2012). MOSFET models for VLSI circuit simulation: theory and
practice. Springer Science & Business Media.
[19] Knoch, J., Mantl, S., & Appenzeller, J. (2007). Impact of the dimensionality
on the performance of tunneling FETs: Bulk versus one-dimensional devices.
Solid-State Electronics, 51(4), 572-578.
[20] [Online]. Available: http://www.eetimes.com/electronics-news/ 4373195/Intel-
FinFETs-shape-revealed
[21] Ruiz, F. J. G., Tienda-Luna, I. M., Godoy, A., Donetti, L., & Gamiz, F. (2009).
Equivalent Oxide Thickness of Trigate SOI MOSFETs With High-κ Insulators.
IEEE Transactions on Electron Devices, 56(11), 2711-2719.
[22] Mukhopadhyay, B., Biswas, A., Basu, P. K., Eneman, G., Verheyen, P., Simoen,
E., & Claeys, C. (2008). Modelling of threshold voltage and subthreshold slope
of strained-Si MOSFETs including quantum effects. Semiconductor Science and
Technology, 23(9), 095017.
[23] Bhattacherjee, S., & Biswas, A. (2007). Modeling of threshold voltage and sub-
threshold slope of nanoscale DG MOSFETs. Semiconductor Science and Tech-
nology, 23(1), 015010.
[24] Tsiara A. (2015). Simulation of nanoscale triple gate FinFETs, with TCAD
tools - A comparative study (MSc thesis, Aristotle University of Thessaloniki).
[25] Tamak, P., & Mehra, R. (2017). Review on Tunnel Field Effect Transistors
99
Bibliography
(TFET).
[26] Ionescu, A. M., & Riel, H. (2011). Tunnel field-effect transistors as energy-
efficient electronic switches. nature, 479(7373), 329.
[27] Calhoun, B. H., & Chandrakasan, A. (2004, August). Characterizing and mod-
eling minimum energy operation for subthreshold circuits. In Proceedings of the
2004 international symposium on Low power electronics and design (pp. 90-95).
ACM.
[28] Nirschl, T., Henzler, S., Pacha, C., Wang, P. F., Hansch, W., Georgakos, G.,
& Schmitt-Landsiedel, D. (2004, August). The tunneling field effect transistor
(TFET) used in a single-event-upset (SEU) insensitive 6 transistor SRAM cell in
ultra-low voltage applications. In Nanotechnology, 2004. 4th IEEE Conference
on (pp. 402-404). IEEE.
[29] [Online] Available: http://ecetutorials.com/analog-electronics/tunnel-diode/
[30] Teherani, J. T. (2010). Band-to-band tunneling in silicon diodes and tunnel
transistors (Doctoral dissertation, Massachusetts Institute of Technology).
[31] Mookerjea, S. A. (2010). Band-to-band tunneling field effect transistor for low
power logic and memory applications: Design, fabrication and characterization
(Doctoral dissertation, The Pennsylvania State University).
[32] Antidormi, A. (2016). Modelling and Simulation of Silicon Nanowire-Based
Electron Devices for Computation and Sensing (Doctoral dissertation, Politec-
nico di Torino).
[33] Chakraverty, M. (2014). A Compact Model of Silicon-Based Nanowire
Field Effect Transistor for Circuit Simulation and Design. arXiv preprint
arXiv:1407.2358.
[34] Wu, W., & Chan, M. (2007). Analysis of geometry-dependent parasitics in
multifin double-gate FinFETs. IEEE Transactions on Electron Devices, 54(4),
692-698.
[35] Hanna, A. N., & Hussain, M. M. (2015). Si/Ge hetero-structure nanotube tun-
nel field effect transistor. Journal of Applied Physics, 117(1), 014310.
[36] Asra, R., Shrivastava, M., Murali, K. V., Pandey, R. K., Gossner, H., & Rao, V.
R. (2011). A tunnel FET for VDD scaling below 0.6 V with a CMOS-comparable
performance. IEEE Transactions on Electron Devices, 58(7), 1855-1863.
[37] Wang, Y., Zhang, W. H., Yu, C. H., & Cao, F. (2016). Sandwich double gate
100
Bibliography
vertical tunneling field-effect transistor. Superlattices and Microstructures, 93,

138-143.
[38] Tang, C. K. (2009). Modelling and characterization of electrical properties of
ZnO-based heterostructure for solar cell applications (Master’s thesis).
[39] Thompson, S. E., & Parthasarathy, S. (2006). Moore’s law: the future of Si
microelectronics. Materials today, 9(6), 20-25.
[40] Neisser, M., & Wurm, S. (2015). ITRS lithography roadmap: 2015 challenges.
Advanced Optical Technologies, 4(4), 235-240.
[41] Alper, C. (2017). Modeling and optimization of Tunnel-FET architectures ex-
ploiting carrier gas dimensionality.
[42] [Online] Available: https : //en.wikipedia.org/wiki/T ransistorc ount
[43] Bhattacharya, D., & Jha, N. K. (2014). FinFETs: From devices to architec-
tures. Advances in Electronics, 2014.
[44] Wang, P. F. (2003). Complementary tunneling-FETs (CTFET) in CMOS tech-
nology (Doctoral dissertation, Technische Universitat Munchen).
[45] Manual, M. A. T. L. A. B. (1995). the MathWorks. Inc., Natick, MA.
[46] Sentaurus, T. C. A. D. (2009). Manuals, Synopsys Inc. Mountain View, CA,
94043.
101

Tesi

Uploaded by

Copyright:

Available Formats

Tesi

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tesi

Uploaded by

Copyright:

Available Formats

POLITECNICO DI TORINO

Corso di Laurea Magistrale in Ingegneria Elettronica

Tesi di Laurea Magistrale

Analysis and Simulation of

In this thesis the electrostatic analysis of three-dimensional FinFETs was done

To enhance the behaviour of TFET in the ON-state, a study of material depen-

3 Physical Simulations of FinFET 24

4 Tunnel Field Effect Transistors 38

6 Tunnel FET issues and solutions 87

2.1 Input data used for implementation of Taur model [1] . . . . . . . . . 10

2.1 Device Structure of SOI FinFET . . . . . . . . . . . . . . . . . . . . 5

Historically, device scaling in semiconductor industry has been driven by Moore’s

In a bulk device, the propagation of electric field is evident in depletion region

On the other hand, a FDSOI MOSFET is characterized by an electric field that

of the drain bias.

One emerging device is the Tunnel FET. As for technological fabrication it is a

1.1 Thesis Outline

Figure 2.1: Device Structure of SOI FinFET

It is a three-dimensional device which allows to a further scaling above the 20nm.

2.1 DG FinFET Analytic Models

2.2 Taur Model

Figure 2.2: Schematic diagram of a DG MOSFET (taken from [1])

where ψ is the electrostatic potential, q is the electronic charge, ni is the intrinsic

where β can be derived by solving the following equation as a function of V:

fr (βs ) = (q/2kT )(Vg − V0 )

In the saturation region, βs ∼ π/2, and fr (βd )  1, so (βd )  1. The current

W 8εsi tox k 2 T 2 q(Vg − V0 − Vds )

W q(Vg − ∆φ) qVds

2.2.1 Matlab Implementation

Quantity Name Value u.m. (S.I.)

A simulation of DG FinFET were carried out in MATLAB varying the channel

Parameters L = 200nm L = 40nm L = 20nm

In conclusion, a continuous analytical model for DG MOSFET were extended to

2.3 Fasarakis Analytical Model

Figure 2.4: 3D Representation of FinFET (taken from [6])

and drain current that is valid in all operation regions.

The model can be extended to a DG FinFET by changing the effective channel

The drain-current equation of the modeled FinFET is similar to the one of DG

The TG FinFET drain current is derived as follows:

A1,DG (Vbi + Vd ) + A2,DG Vbi Vt Qth NA

This correction results in an increase in the threshold voltage, such as:

Where α is equal to 1 for DG and 2 for TG.

2.3.1 Matlab Implementation

Quantity Name Value u.m. (S.I.)

Figure 2.5: Transcharacteristic in logarithmic representation varying gate length

Parameters L = 200nm L = 40nm L = 20nm

In this model, an appreciable variation can be observed by varing the channel

2.4 TG FinFET Analytic Models

In the rectangular TG FinFET compact model, physical effects such as SCEs,

2.4.1 Matlab Implementation

Quantity Name Value u.m. (S.I.)

In the figure below is shown the Id − Vg characteristic of the descripted device,

Figure 2.7: Comparison of Id − Vgs characteristic of Tz-TG FinFET and Re-TG

The characteristic in figure (2.7) shows the variation of electrical parameters

Parameter Tz-TG Re-TG

Furthermore, the compact model for TG FinFET was extended to Trapezoidal

Quantity Name Value u.m. (S.I.)

I d -Vgs Transcharacteristic 10-4

Figure 2.8: Simulated transfer-characteristic of a Tz-TG FinFET

In the saturation region, βs ∼ π/2, and fr (βd ) 1, so (βd ) 1. The current