theoretical performance evaluations of nmos double gate fets … kim... · 2009. 12. 2. ·...
TRANSCRIPT
Theoretical Performance Evaluations of
NMOS Double Gate FETs with
High Mobility Materials: Strained III-V, Ge and Si
A DISSERTATION SUBMITTED TO THE DEPARTMENT OF ELECTRICAL
ENGINEERING AND THE COMMITTEE ON GRADUATE STUDIES OF
STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
Donghyun Kim
Nov 2009
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/zg341nj2801
© 2010 by Donghyun Kim. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Krishna Saraswat, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Philip Wong
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Tejas Krishnamuhan
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
iv
Abstract
As Si CMOSFET technology scales down to nanometer scale, it becomes extremely
difficult to keep the same drive current due to limitations of channel mobility (µ), gate
oxide scaling and parasitics. Currently, strained-Si is the dominant technology for high
performance MOSFETs and increasing the strain provides a viable solution to scaling.
However, looking into future nanoscale MOSFETs, it becomes important to look at
novel channel materials such as strained SixGe(1-x) and strained III-Vs and new device
structures.
There are two fundamental problems with these candidates. First, the main advantage
of small transport mass to have high injection velocity may be undermined by low
density of states which reduces the inversion charge and hence reduces drive current.
Second, due to their small effective masses and small band gap combined with high
electric fields applied in nanometer-scale devices, large band to band tunneling (BTBT)
leakage current is a big concern.
To understand the physics and the performance limitation of nanoscale MOSFETs
with high mobility channel materials, we have developed first, a new BTBT model
which takes into account complete real and complex band structure, direct/phonon-
assisted tunneling, and quantum confinement effect, second, a quantum transport
simulator and third, a local pseudo-potential method (LPM) to calculate real and
complex band structures of strained semiconductors. While the BTBT models in
commercially available TCAD tools largely fail to predict accurate current even in
v
silicon, our model is well matched to experimental results of PIN diodes fabricated on
Si, Ge and GaAs, and heterostructure SiGe pMOSFETs.
We have investigated and benchmarked Double-Gate (DG) n-MOSFETs with
different channel materials (GaAs, InAs, InSb, Ge and Si). Our results show that with
gate oxide thickness of 0.7 nm, small density of states (DOS) of these materials does
not significantly limit the on-current (ION) and high µ materials still show higher ION
than Si. However, the high µ small bandgap materials like InAs, InSb and Ge, suffer
from excessive BTBT current and poor SCE, which limits their scalability. InP has
significantly higher ION and shorter intrinsic delay at a reduced IOFF,BTBT when
compared to Si.
Strained (uniaxial and biaxial) InxGa1-x As may have a very good tradeoff between the
excellent transport properties of InAs and the low leakage of GaAs. Full quantum
ballistic simulations with new BTBT models based on the bandstructures calculated by
LPM shows that at a l00nA/ µ m Ioff specification, 4% biaxial compressive strained
In0.75Ga0.25As (111) NMOS DGFET outperforms other InGaAs compositions because
of the excellent transport properties and reduced leakage current with strain
engineering.
vi
Acknowledgements
I wish to acknowledge the support of many people who have contributed to my
Ph.D. research and have made my life at Stanford very enjoyable.
First of all, I would like to express my heartfelt gratitude to my advisor
Professor Krishna Saraswat for meticulously guiding this dissertation, constant
encouragement and support and especially for his patience throughout the course of
my research. I am especially grateful for the time and the freedom he gave me to
choose a research topic and explore a large number of areas in novel device physics.
He has been undoubtedly one of the most kind and helpful people I have ever
interacted with at Stanford. I have been extremely fortunate to work with the finest
advisor that one could possibly hope for.
I would like to thank my associate advisor Professor Philip Wong for all his
help throughout my project especially with regards to nano-electronics and quantum
ballistic transport simulations. He has been an excellent role model and mentor. I have
learned a lot from his vast knowledge, his wide industry experience and strong
technical expertise.
My special thanks go to my mentor friend and once housemate, Prof. Tejas
Krishnamohan for all the wonderful discussions and guidance on a variety of topics
that forms large part of my dissertation. Tejas is one of the brightest scientists I ever
have interacted with. Throughout the hour-long discussions at office cubicle,
vii
restaurants, and at home I have learnt a lot. It was the most intellectual experience I
ever had.
I am also grateful to Professor David Miller for agreeing to serve as the chair
of my Ph.D. oral examination committee. Without his brilliant and insightful lectures
which formed the basis of my quantum mechanics and semiconductor physics, I would
not have been able to even start my Ph.D. project.
I would like to thank Dr. Lee Smith from Synopsys for being an excellent
mentor during my summer internships and for his support throughout my Ph. D. Initial
work on BTBT would not have been done without him letting me have an access to
the source code of TAURUSTM. I acknowledge the computational and technical help
and guidance offered by Professor Robert Dutton.
I am grateful to my undergraduate advisors – Professor Hyo Joon Eom,
Professor Kwyro Lee and Professor Kyunghoon Yang for initiating my interest in
research. I thank them for their guidance and support throughout my undergraduate
years.
I would also like to thank the administrative staffs - Diane, Gabrielle, Gail,
Irene, Miho and Natasha for being very helpful and supportive throughout my stay at
the Center for Integrated Systems.
My gratitude is expressed to the Defense Advanced Research Projects Agency
(DARPA), MARCO Materials, Structures and Devices (MSD) Center, Initiative for
Nanoscale Materials and Processes (INMP) and Samsung Scholarship Foundation for
their financial support.
viii
I owe a debt of appreciation to Professor Saraswat’s former and current group
members who have provided me with ideas, and have made my stay in CIS pleasant
and rewarding: Abhijit, Ali, Aneesh, Arunanshu, Chi On, Crystal, Duygu, Euijong,
Gunhan, Hoon, Hoyeol, Hyun-Yong, Jason, Jin-Hong, Ju Hyung, Jungyup, Kyung
Hoae, Pawan, Raja, Rohan, Rohit, Sarves, Shyam, Ting-Yen, WooShik and Yeul.
I am also grateful to my friends Abhi, Beomsoo, Bob, Byoung Du, Byoungil,
Christophe, Ed, Henry, Hopil, Hyejun, Hyunjoo, Insun, Jack, Karen, Kee-Bong,
Kyungryoon, Kyuwon, Jason, Jekwang, Jenny, Jeongha, Joon Seok, Joo Hyung,
Jungjoon, Kat, Katherine, Kyeongran, Kyungrok, Margie, Mathieu, Mehdi, Misung,
Moonjung, Paul, Rami, Rob, Saeroonter, Salman, Sangbum, Serene, Sergey,
Seunghwa, Soogine, Sora, Sooyoung, Sungwoo, Sunhae, Utkan, Wanki, Wonseok,
Yejin, Yeonjoo, Yeong Min, Youngjoon and Yoonyoung for their friendship and all
the fun that we have had together.
No words would ever do justice to express my deepest thanks and gratitude to
my most wonderful family – my parents and my sister. I will be eternally grateful to
them for being there for me at all times. Their continuous love, sacrifice, support and
encouragement have allowed me to pursue my ambitions.
This dissertation is dedicated to them.
ix
Table of Contents
Abstract .......................................................................................................................... iv
Acknowledgements ....................................................................................................... vi
Table of Contents .......................................................................................................... ix
List of Tables ............................................................................................................... xiv
List of Figures .............................................................................................................. xvi
Chapter 1 Introduction .................................................................................................... 1
1.1 Motivation ............................................................................................................ 1
1.2 Organization ......................................................................................................... 8
Chapter 2 Calculation of Bandstructure ......................................................................... 9
2.1 Introduction .......................................................................................................... 9
2.2 Principles of Pseudopotential ............................................................................. 10
2.3 Band structure calculation with plane wave basis .............................................. 12
2.4 Continuous q-space model potential .................................................................. 14
2.5 Strain dependence of the pseudopotentials ......................................................... 15
2.6 Kinetic-energy scaling ........................................................................................ 16
2.7 G-space smoothing function ............................................................................... 16
2.8 Spin orbit coupling ............................................................................................. 17
x
2.9 Atomic positions in strained materials ............................................................... 20
2.9.1 Strain tensor ................................................................................................. 20
2.9.2 Stress Tensor ............................................................................................... 22
2.9.3 Stiffness and Compliance Coefficients ....................................................... 24
2.9.4 Coordinate transformation ........................................................................... 24
2.9.5 Lattice vectors and reciprocal lattice vectors for a strained system ............ 26
2.9.6 Internal ionic displacement .......................................................................... 27
2.9.7 Uniaxial strain ............................................................................................. 28
2.9.8 Biaxial strain ................................................................................................ 29
2.10 Alloys material (virtual crystal approximation) ............................................... 32
2.11 Complex Bandstructure .................................................................................... 33
2.12 Summary ........................................................................................................... 34
Chapter 3 Modeling of Band to Band Tunneling ......................................................... 36
3.1 Introduction ........................................................................................................ 36
3.2 Direct-gap Band-to-Band Tunneling. ................................................................. 36
3.3 Phonon-Assisted Tunneling / Indirect-gap Tunneling ....................................... 46
3.4 2D Quantized System ......................................................................................... 59
3.5 Direct-gap Band-to-Band Tunneling (2D) ......................................................... 65
3.6 2-D Phonon-assisted Indirect-gap Band-to-Band Tunneling ............................. 71
3.7 Summary ............................................................................................................. 76
xi
Chapter 4 Phonon-Electron Interactions ...................................................................... 78
4.1 Introduction ........................................................................................................ 78
4.2 Electron-Phonon Interaction Element- Rigid Ion Model ................................... 79
4.3 Phonon Polarization Vectors .............................................................................. 80
4.4 Silicon. ................................................................................................................ 82
4.5 Germanium ......................................................................................................... 87
4.6 Summary ............................................................................................................. 91
Chapter 5 Application of band-to-band tunneling (BTBT) model ............................... 93
5.1 Introduction ........................................................................................................ 93
5.2 Comparison to Previous Models ........................................................................ 93
5.2.1 Direct Band .................................................................................................. 93
5.2.2 Indirect Band ............................................................................................... 95
5.2.3 Selection Rules on Phonon modes and Phonon-interaction Strength ......... 96
5.3 BTBT Generation Rates for Various Materials .................................................. 97
5.4 PIN Diodes ......................................................................................................... 99
5.5 Effect of Strain (Ge) ......................................................................................... 103
5.6 Quantum Confinement ..................................................................................... 106
5.7 Summary ........................................................................................................... 111
Chapter 6 Quantum-Ballistic Transport in 1-D Quantum Confined System ............. 113
6.1 Introduction ...................................................................................................... 113
xii
6.2 1-D Quantum Confined System ....................................................................... 114
6.3 1-D Schrodinger Equation with non-parabolic effective mass ......................... 116
6.4 Electron Wavefunction along the Channel ....................................................... 118
6.5 Non-parabolic Density of States ....................................................................... 120
6.6 Charge Density (with Non-parabolicity corrections) ....................................... 121
6.7 Quantum Ballistic Current (with Non-parabolicity corrections) ...................... 128
6.8 Band-to-Band Tunneling Current ..................................................................... 132
6.9 Summary ........................................................................................................... 139
Chapter 7 Properties and Trade-offs Of Compound Semiconductor MOSFETs ....... 141
7.1 Introduction ...................................................................................................... 141
7.2 Simulation Framework ..................................................................................... 145
7.2.1 Bandstructure calculation (Real and Complex) ......................................... 145
7.2.2 Band-to-Band Tunneling (Off-State Leakage) .......................................... 146
7.2.3 2D-Quantum Ballistic Current (On-State Drive Current) ......................... 149
7.2.4 Benchmark Test bed - Device Structure .................................................... 150
7.3 Power-Performance Tradeoffs in Binary III-V materials (GaAs, InAs, InP and
InSb) vs. Si and Ge. ................................................................................................ 151
7.3.1 Inversion charge and injection velocity ..................................................... 152
7.3.2 ION, IOFF,BTBT and Delay ............................................................................. 154
7.3.3 Effect of Scaling film thickness and VDD .................................................. 156
xiii
7.3.4 Power-Performance Tradeoff of Binary III-V Materials vs. Si and Ge .... 156
7.4 Power-Performance of Strained Ternary III-V Material (InxGa1-xAs) ............. 158
7.4.1 Strained InGaAs Band Structures .............................................................. 159
7.4.2 ION and IOFF,BTBT with Strain Engineering and Channel Orientation ......... 161
7.4.3 Power-Performance of Strained Ternary III-V Material (InGaAs) ........... 163
7.5 Strained III-V for p-MOSFETs ........................................................................ 165
7.5.1 Hole Mobility in Ternary III-V Materials (InGaAs vs. InGaSb) .............. 166
7.5.2 Hole Mobility Enhancement in III-Vs with Strain .................................... 167
7.6 Novel Device Structure and Parasitics ............................................................. 169
7.6.1 Quantum Well (QW) Strained Heterostructure III-V FETs ...................... 169
7.6.2 Parasitic Resistance ................................................................................... 171
7.6.3 Parasitic Capacitance ................................................................................. 172
7.7 Conclusion ........................................................................................................ 174
Chapter 8 Conclusion ................................................................................................. 176
References .................................................................................................................. 181
xiv
List of Tables
Chapter 4
Table 4.1 : Polarization vectors of major phonon modes in diamond lattice structure.
Since the diamond structure has a primitive cell formed by two-atom basis, each
phonon consists of two polarization vector. While longitudinal mode consists of one
LA and one LO polarizations, the transverse modes (TA and TO) are double
degenerate. .................................................................................................................... 81
Table 4.2 : The effective deformation potentials and phonon energies for intervalley
scattering between Γ25’ and ΔC ( 2 / (0,0,0.85)laπ ). Energy difference between two states is
1.12eV. Since there is degeneracy in Γ25’ states and transverse modes, the deformation
potentials were added to indicate the effective strength as follows: 2 2
,( ) ( )l
pp l
DK DK= ∑ ,
where p denotes the polarization of phonon and l is indices for degenerate valence bands. ... 83
Table 4.3 :The complex-k-space-intervalley-deformation potentials calculated for Band-to-
Band tunneling between Γ25’ and ΔC. Due to lack of theoretical work, previous BTBT models
(R,T,F,S) estimated the deformation potentials from dilation deformation potentials and
mobility deformation potentials. Experimental values are extracted from light absorption
coefficient study (Ge,Gp) and Esaki tunneling device (C). ................................................. 86
Table 4.4 :The effective deformation potentials and phonon energies for intervalley scattering
between Γ25’ and LC ( / (1,1,1)laπ ) (for LA phonon, between c2−Γ - c1L+ ). Energy difference
xv
between two states is 1.12eV. Since there is degeneracy in Γ25’ states and transverse modes,
the deformation potentials were added to indicate the effective strength as follows:
2 2
,( ) ( )l
pp l
DK DK= ∑ , where p denotes the polarization of phonon and l is indices for
degenerate valence bands. ................................................................................................ 89
Table 4.5 : The complex-k-space-intervalley-deformation potentials calculated for Band-to-
Band tunneling between Γ25’/ c2−Γ and c1L+ /ΔC. (100) at E=-4.12eV. ................................... 90
Chapter 5
Table 5.1 : Parameter F0 of Kane/Hurkx model AFσexp(-F0/F), extracted by our BTBT model
(by LEPM), Experiments and Kane(two band model for direct and effective mass
approximation (EPM) for indirect). The direction indicates tunnel direction ...................... 100
Chapter 7
Table 7.1 : Dominant leakage paths in nanoscale high mobility CMOS devices. High
mobility (low-bandgap) materials suffer from excessive BTBT leakage. ................. 144
xvi
List of Figures
Chapter 1
Fig. 1.1: Scaling the bulk MOSFET in accordance with Moore’s Law (source: Intel).. 1
Fig. 1.2: (a) Need to maintain an increase in the drive current (Ion) even at reduced
supply voltages (Vdd) in the future. (b) The increase in Ion with scaling is obtained at
the expense of an exponentially growing off-state leakage (Ioff) (source: Intel) ............ 2
Fig. 1.3: At the current rate, the exponentially growing static power dissipation will
cross the active power consumption (source: Intel). ...................................................... 3
Fig. 1.4: Carrier transport in ballistic limit. The current is proportional to the velocity
and the density of the carriers at the source, and the backscattering rate ...................... 5
Fig. 1.5: Leakage current components in the channel of highly scaled MOSFETs ....... 6
Fig. 1.6: Three major models to obtain I-V characteristics of highly scaled transistors 7
Chapter 2
Fig. 2.1: Pseudopotential approximation: The tightly bond inner electron shell can be
effectively replaced by the pseudopotential. ................................................................ 11
Fig. 2.2: of Ga and As in GaAs. This is the Fourier transform of the atomic
pseudopotential ............................................................................................................. 15
Fig. 2.3: Two- dimensional geometric deformation of an infinitesimal material
element. ...................................................................................................................... 21
xvii
Fig. 2.4: Components of stress in three dimensions ................................................... 23
Fig. 2.5: Crystal coordinate system ............................................................................ 25
Illustration of band-to-band tunneling paths. The electron can tunnel to indirect-gap
conduction band (L or ∆ valleys) through phonon interactions. Fig. 2.6: Location of an
atom relative to its four nearest neighbors in a crystal strained in the (111) direction.
For z=0 the location of the atom is deformed only by the homogeneous bulk strain,
while for z=1 all four bonds are of the same length. The vector u represents the
differential vector between two locations ..................................................................... 28
Chapter 3
Fig. 3.1:Crystal Potential and corresponding Wavefunctions ...................................... 37
Fig. 3.2: (a) 3D plot of complex bandstructure of germanium along the [111] direction.
The black lines depict the real bands and the red lines depict complex bands. (b) 2D
plot of complex bandstructure of germanium .............................................................. 39
Fig. 3.3: Illustration of electron tunneling process in the semiconductor with the strong
electric field. The electron in valence band tunnels in to the bandgap at the classical
turning point, , passes the branch point, and tunneling ends at . .................... 40
Fig. 3.4: Complex band structure of Ge obtained from local empirical pseudopotential
method (LEPM) along the [111] direction. Parabolic effective mass approximation (a)
fails to follow the complex dispersion in relaxed germanium. While the Kane’s two-
band model works well for relaxed materials (a), it breaks down in strained materials
xviii
(c). The proposed model well approximate the dispersion in both relaxed and strained
germanium (b, d). ......................................................................................................... 45
Fig. 3.5: Illustration of band-to-band tunneling paths. The electron can tunnel to
indirect-gap conduction band (L or ∆ valleys) through phonon interactions. ............. 46
Fig. 3.6: (a) Longitudinal and transverse waves in a 1D linear lattice. (b) Transverse
optical and transverse acoustical waves in a 1D diatomic lattice. (c) The ion is
displaced from equilibrium by an vector, . is the unit-length polarization vector of
the lattice vibration ....................................................................................................... 47
Fig. 3.7: Phonon dispersion relations in germanium. Generally at zone boundaries
such as 2 ⁄ 0.5,0.5,0.5 and 2 ⁄ 1,0,0 , the frequencies are
roughly constant ........................................................................................................... 50
Fig. 3.8: Complex band structure of germanium obtained from local empirical
pseudopotential method (LEPM) along the 110 direction. The valley located at k-
point 2 ⁄ 0.5,0.5,0.5 is projected to Γ point. The parabolic effective mass
approximation (a) completely fails to estimate the complex dispersion, since the
dispersion curve at Γ point deviates a lot from the parabolic curve. The suggested
model (b) successfully approximates the dispersion curve .......................................... 58
Fig. 3.9: Schematic view of the thin channel double-gate MOSFET (DGFET).
Thickness of the channel is . Current flows along x-direction and the channel is
quantized along the z-direction .................................................................................... 59
Fig. 3.10: Illustration of complex bandstructure of two-band two-subband system. The
left panel depicts the scenario where bands with same quantum number are connected
xix
and the right panel shows the scenario where bands with different quantum number are
connected ...................................................................................................................... 63
Fig. 3.11: Complex bandstructure of 14nm thick (001) germanium. The imaginary
values of the wavenumber k are drawn as red curves and the real values are shown in
black curves .................................................................................................................. 68
Fig. 3.12: Γ′ as a function of quantum confinement energy (Δ ) (eq. (3.83)) in
(001) germanium. Tunneling direction is along the (110) direction ........................... 70
Fig. 3.13: Γ , Γ , used in Fig. 3.12 and their fitting functions ..... 70
Fig. 3.14: L , ,′ as a function of quantum confinement energy (Δ ) (eq.
(3.104)) in (001) germanium. Tunneling direction is along the (110) direction .......... 75
Fig. 3.15: , Δ L and , Δ L used in Fig. 3.14 and their fitting
functions ....................................................................................................................... 76
Chapter 4
Fig. 4.1: Bandstructure of silicon calculated by EPM. Phonon-assisted transitions with
highest probability between band extrema are shown. In silicon, transitions between
Δ1-Γ25’ are most dominant. Due to selection rules imposed by symmetry of the two
points, TA and TO phonons are most effective in assisting the tunneling process ...... 82
Fig. 4.2: Comparison between the Lax’ Selection Rules and our work. Since most of
contribution in tunneling is coming at E=-4.67eV, the deformation potentials should
xx
be calculated between two points depicted in (b) not between two symmetry points in
(a). ................................................................................................................................. 85
Fig. 4.3: Bandstructure of germanium calculated by EPM. Phonon-assisted transitions
with highest probability between band extremum are shown. In germanium, transitions
between L1-Γ25’ are most dominant. Experimentally LA, TA and TO phonons are
most effective in assisting the tunneling process. Subindex c in Lc1 represents
conduction band while v represents valence band ....................................................... 88
Chapter 5
Fig. 5.1: (a) Kane’s 2-band model. (b) Full band model obtained by EPM. ................ 94
Fig. 5.2: ......................................................................................................................... 96
Fig. 5.3: (a) 3D plot of complex bandstructure of germanium along the [111] direction
(b) GaAs. ...................................................................................................................... 98
Fig. 5.4: BTBT generation rate vs. Electric Field ........................................................ 99
Fig. 5.5: Indirect tunneling process. Keldysh’s model does not capture full band-
information ................................................................................................................. 101
Fig. 5.6: I-V curves of GaAs PIN diodes. Solid lines with markers are actual I-V
characteristics of the diodes, solid lines are estimations from our BTBT model, and
dotted lines are currents calculated by Kane’s model. The length of I-region is varied
from 15nm to 55nm .................................................................................................... 102
xxi
Fig. 5.7: I-V curves of Ge PIN diodes. Solid lines with markers are actual I-V
characteristics of the diodes, solid lines are estimations from our BTBT model, and
dotted lines are currents calculated by Kane’s model ................................................ 103
Fig. 5.8: Calculated band shifts of germanium (001) as a function of in-plane biaxial
strain ( 0/ 1a a − ) ........................................................................................................ 104
Fig. 5.9: IOFF,MIN vs. biaxial strain. Compressive strain increases indirect BTBT while
it decreases direct BTBT. Optimum point at around x=0.6 ....................................... 105
Fig. 5.10: (a) A GaAs QW consists of 12 atom thick layer with 6 atom thick vacuum
layer. (b) Complex Bandstructure of the GaAs QW .................................................. 106
Fig. 5.11: Ultra-thin body and larger quantization increases the effective bandgap and
lowers the tunneling rate ............................................................................................ 107
Fig. 5.12: Iso-leakage (IOFF,MIN) Contours. These curves determine the maximum
allowable VDD to meet the ITRS off current requirement for a 15nm gate length, at a
given body thickness (constrained by technology). s-Si cannot be used for >0.5V
operation. Ge and InAs can be used for 0.9 operation when Tbody<5nm. For
Tbody<7nm s-Ge can be used for >1V operation. Si and GaAs comfortably meet the
off-current specification because of their large bandgap ........................................... 109
Fig. 5.13: Device structure and band diagrams of SiGe heterostructure p-MOSFET 110
Fig. 5.14: BTBT (GIDL) with varying TGe channel. Left plot is obtained from
experiments and right plot is drawn using simulation results .................................... 111
xxii
Chapter 6
Fig. 6.1: Schematic view of the thin channel double-gate MOSFET
(DGFET).Thickness of the channel is . Current flows along x-direction and the
channel is quantized along the z-direction ................................................................. 115
Fig. 6.2: Convergence of the first energy level. Effective mass is 0.069 and is
0.7eV . Channel thickness is 3nm. About in 5 iterations, it converges within 0.01eV
.................................................................................................................................... 118
Fig. 6.3: The error term Θ , becomes significantly large value when or are
large. (a) For 0.5, 4/ and 40, the error term accounts for more
than 10% of the correct integral. (b) For 2, 4/ and 40, the
error term accounts for more than 25% of the correct integral .................................. 124
Fig. 6.4: Relative error of the approximation in Eq. (6.34) vs. the integral in Eq. (6.24).
This approximation is accurate with an error below 0.3% ......................................... 126
Fig. 6.5: Relative error of the approximation of in ⁄ , Eq. (6.47) vs. the
integral in Eq.(6.44). This approximation is accurate in the range of error below 1.3%
.................................................................................................................................... 131
Fig. 6.6: Source to drain current ( DSI ) vs. gate voltage (Vg). When negatively large
Vg is applied, BTBT current generated at the edge between gate and drain grows
significantly ................................................................................................................ 133
Fig. 6.7: Illustration of the carrier flux components. Electron currents consist of the
one injected from source side, one from drain side and one generated by BTBT
xxiii
process. The BTB tunneling hole current is supplied to the channel cavity (around
). The accumulated holes can escape by thermionic emission from the cavity
................................................................................................................................... .134
Fig. 6.8: Sketchy of the band profile along the channel. Source Fermi level ( ),
Drain Fermi level ( ), Energy profile of the n-th state of valence band, hole quasi-
Fermi level ( ), hole barrier height ( ) are marked in the figure. Only the hole
with kinetic energy larger than can escape the cavity and reach the source side.. 136
Fig. 6.9: Flow chart of the carrier transport simulation of quantum-ballistic FETs ... 140
Chapter 7
Fig. 7.1: Dominant leakage paths in nanoscale high mobility CMOS devices. High
mobility (low-bandgap) materials suffer from excessive BTBT leakage .................. 142
Fig. 7.2: Tradeoffs between effective mass (m ), bandgap (EG), and dielectric constant
(κs) in semiconducting materials ................................................................................ 143
Fig. 7.3: (a) 3D plot of complex bandstructure of germanium along the [111] direction.
(b) 2D plot of complex bandstructure of InAs ........................................................... 146
Fig. 7.4: (a) Illustration of band-to-band tunneling paths. The electron can tunnel to
indirect-gap conduction band (L or ∆ valleys) through phonon interactions. (b) All
Possible (Direct and Indirect) BTBT Tunneling Paths, Γ-Γ, Γ-X and Γ-L, are captured
by the model ............................................................................................................... 148
xxiv
Fig. 7.5: Ultra-thin body and larger quantization increases the effective bandgap and
lowers the tunneling rate ............................................................................................ 149
Fig. 7.6: Test bed for benchmarking III-Vs as channel material. Gate Length LG=15nm,
equivalent gate oxide thickness (EOT) TOX=0.7nm, channel thickness TS=3-5nm and
VDD=0.3-0.7V ............................................................................................................. 151
Fig. 7.7: Inversion charge (Qi) at VGate=0.7V and Tox=0.7nm. Si channel stores 4 times
more electrons than InAs. As Ts scales, Qi becomes bigger due to reduced Vth and
nonparabolicity of band. ............................................................................................. 153
Fig. 7.8: Injection velocities (vinj) at VGate=0.7V and Tox=0.7nm. InAs and InSb have 6
times larger vinj than Si due to their small transport mass. As Ts scales, lowered
threshold voltage (vth) increases vinj, while nonparabolicity reduces vinj. .................. 154
Fig. 7.9: Drive currents (ION) for various (a) Ts and (b) VDD. LG=15nm, TOX=0.7nm.
.................................................................................................................................... 155
Fig. 7.10: (a) IOFF,BTBT is minimum achievable leakage current defined as a
intersecting point between BTBT current and Drive current. (b) IOFF,BTBT for various
Ts. VDD=0.5V. (c) IOFF,BTBT for various VDD .............................................................. 155
Fig. 7.11: Intrinsic delay vs. IOFF,BTBT trade-off in various materials. Lg=15nm,
Tox=0.7nm, Ts=5nm. VDD is varied from 0.3V to 0.7V. ............................................. 157
Fig. 7.12: Band shifts in biaxially strained (a) GaAs (001) and (b) In0.75Ga0.25As (111)
calculated by local empirical pseudopotential method. For GaAs(001), 4% biaxial
compressive (BiC) strain is applied and for In0.75Ga0.25As (111), 4% biaxial tensile
xxv
(BiT) strain is applied. Solid lines represent relaxed material and dotted lines represent
strained material. ........................................................................................................ 159
Fig. 7.13 (a) Quantized bandgap as a function of strain (e||) in 5nm film. Compressive
strain increases bandgap. Especially (111) orientation is strongly modified (b) Γ-L
Separations as a function of strain (e||) in 5nm film. Tensile strain widens the
separation. With tensile strain, ΔEΓL in GaAs can be over 0.3 (eV). ......................... 160
Fig. 7.14: (a) 001 (b) 111. Tunnel mass as a function of x (InxGa1-xAs) in 5nm film.
On (111) Orientation, compressive strain increases mass, while on (001), it reduces
mass ............................................................................................................................ 161
Fig. 7.15: (a) ION as a function of strain (e||) in strained GaAs and In0.75Ga0.25As
NMOS DGFET. For comparison, IONs for 2% biaxially tensile-strained (BiT) Si and
unstrained Si are plotted as dotted lines. (b) IOFFBTBT as a function of strain in strained
GaAs and In0.75Ga0.25As. For comparison, IOFFBTBT for 2% biaxially tensile-strained Si
is drawn as dotted line. 4% (111) compressive strain reduces IOFF,BTBT by over 1000x
.................................................................................................................................... 163
Fig. 7.16: (a) ION vs. IOFFBTBT and (b) Delay vs. IOFFBTBT of the best performing n-
DGFETs with materials, biaxially tensile-strained GaAs (001), uniaxially
compressive-strained In0.25Ga0.75As (001) and biaxially compressive-strained
In0.75Ga0.25As (111). Strain levels are 0, 0.02 and 0.04. Values for biaxially tensile-
strained Si are given for comparison .......................................................................... 165
Fig. 7.17: Mobility for InXGa1-XSb and InXGa1-XAs. Sb’s have higher (~2X) hole
mobilities than As’s .................................................................................................... 167
xxvi
Fig. 7.18: Enhancement of mobility with respect to the unstrained case with biaxial
(left) uniaixal (right) strain. Enhancement with compression is always better than
tension. Upto ~4X / ~2X enhancement possible with 2% uniaxial /biaxial compression.
Enhancement with strain (especially for uniaxial) depends highly on the choice of
channel direction. ....................................................................................................... 168
Fig. 7.19: μh enhancement with biaxial stain; achievable by lattice mismatch between
channel and barrier layers during MBE growth. ........................................................ 169
Fig. 7.20: Conventional and QW double gate FET structures. .................................. 171
Fig. 7.21: Metal Source-Drain IIIV n-MOSFET. ....................................................... 172
Fig. 7.22: Delay in consideration of parasitic gate capacitance. Lg=15nm, Tox=0.7nm,
Ts=5nm. VDD =0.7V. .................................................................................................. 174
1
Chapter 1 Introduction
1.1 Motivation
The silcon-based Metal Oxide Semiconductor Field Effect Transistors (MOSFETs)
has been the most important building block of the Integrated Circuits (ICs), since they
were first demonstrated in 1960 [1]. Today's Integrated Circuits (ICs) use the
MOSFET as a basic switching element for digital logic applications and as an
amplifier for analog applications. Although the basic architecture of the device has
remained the same, the physical size has been continuously reduced by factor of two
every 2-3 years in accordance with Moore’s Law [2] (Fig. 2.1), resulting in faster and
more complex chips with lowered cost per transistor.
Figure 1.1 : Scaling the bulk MOSFET in accordance with Moore’s Law (source: Intel).
Scaling the digital switch.
2
This trend was the main driving force for electronics industry to put humongous
energy and budget into scaling device technologies. Currently, in the 32nm high-
performance processor technology, the physical gate length is scaled below 20nm,
with effective gate oxide thinner than 0.8 nm. This exponential scaling of the physical
feature size cannot continue forever, as beyond the 22nm node fundamental as well as
practical limitations start restricting the performance of the conventional Si MOSFETs.
As illustrated in Fig. 1.2 (a) and (b), after development of 90nm device technology,
enhanced drive currents were achieved with an exponential increase in the static, off-
state leakage of the device.
(a) (b)
Figure 1.2 : (a) Need to maintain an increase in the drive current (Ion) even at
reduced supply voltages (Vdd) in the future. (b) The increase in Ion with
scaling is obtained at the expense of an exponentially growing off-
state leakage (Ioff) (source: Intel).
3
Figure 1.3 depicts the evolution of static power dissipation (leakage) and active power
dissipation as the devices are scaled down. The static power density on the chip has
exponential increased at the much faster rate than the active power density, which
increases rather linearly with device scaling. The active power density arises due to
the dissipative flow of charges between the transistor gates and ground terminals
during switching of the device. The static power, also known as sub-threshold power,
or standby power, is dissipated even in the absence of any switching operation The
major source of the static power dissipation is the sub-threshold leakage in the
transistor because the MOSFET at below 50nm gate length scale does not behave as
an ideal switch and allows the leakage current. The static power dissipation was a
relatively insignificant component of the total chip power just a few generations back,
but it is now comparable in magnitude to the active power. Management and
Figure 1.3 : At the current rate, the exponentially growing static power
dissipation will cross the active power consumption (source: Intel).
4
suppression of static power is one of the major challenges to continued gate length
reduction for higher performance. Once the scaling of the conventional bulk Si
MOSFET starts slowing down, the insertion of performance boosters, like novel
materials and non-classical device structures, will become necessary to continue to
improve performance.
To achieve performance without sacrificing off state leakage power, strain technology
was adopted to boost the drive current. Mechanically stressed Silicon has higher
mobility than relaxed Silicon. However, for further scalability of the device, the
materials with even higher mobility than strained Si are needed. The materials such as
strained IIIV, strained SiGe or even more futuristic carbon nanotubes and graphene
have been actively investigated to study the possibility of substituting Si as a channel
material in future. While IIIV materials such as InAs and InSb exhibit 20 to 40 times
larger mobility than Silicon, high mobility generally comes along with large dielectric
constant and small bandgap which are the basic properties of semiconductor materials
which have weak bonding between atoms.
5
Figure 1.4 illustrates the carrier transport in ballistic limit. In transistors of which gate
length is smaller than 15nm, electrons move almost ballistic and back scattering is
negligible. In this limit, amount of current is multiplication of amount of charge in
source and injection velocity. High mobility materials show potential to improve the
drive current since they have high injection velocity due to its small mass compared to
Si. However, generally these materials suffer from the small density of states in source
side resulting in reduced number of carriers injected from source side. To answer the
question whether these high mobility materials outperform Si in terms of drive current
at the scale below 10nm gate length, more theoretical studies are needed.
Figure 1.4 : Carrier transport in ballistic limit. The current is proportional to the
velocity and the density of the carriers at the source, and the
backscattering rate [3].
Source
Drain
Channel
Injected (vinj)
Back scattered (r)Nsource
⎟⎠⎞
⎜⎝⎛
+−
×=rrvqNI inj
SourceSsat 1
1
Low m*transport High νinjLow r
Small mDOS Small Ns
Ballistic Limit
6
Figure 1.5 depicts possible paths of the leakage current in the channel of highly scaled
MOSFETs. Small mass and small bandgap of high mobility materials enhance source-
drain tunneling and band-to-band tunneling (BTBT) at the gate-drain edge. BTBT
leakage current is a major leakage component in III-V materials and this may hinder
scalability of transistor. To study applicability of III-V materials in highly scaled
MOSFETs, the investigation of off-state leakage current should be included in overall
benchmark.
Basic device performance can be evaluated by examining the Id-Vg characteristics
(Fig 1.6). The drain current has two main components, drive current and leakage
current (mostly BTBT current in III-Vs). Minimum off-state leakage is defined by the
intersection between leakage current line and the drive current line.
Figure 1.5 : Leakage current components in the channel of highly scaled
MOSFETs.
Source
Drain
Channel
Valence Band
Conduction Band S-D Tunneling
Band-to-Band Tunneling
7
To simulate the performance of transistor with novel materials, there are three major
steps. First, the energy band structure of the material should be calculated. Band
structure contains kinetic properties of electrons in semiconductors with periodic
lattice. To obtain energy band structures of semiconductors we have developed
empirical pseudopotential method (EPM). The energy band information in real
reciprocal space (k-space) obtained by the EPM method is fed into quantum ballistic
drive current model to calculate the drive current in the device. To calculate the BTBT
leakage in the device we information on real energy vs. wavenumber k is fed to
Figure 1.6 : Three major models to obtain I-V characteristics of highly scaled
transistors.
log (IDS)
Ioff,min
VGS
IBTBT
IDrive
-1 -0.5 0 0.5 1-12
-10
-8
-6
-4
-2
k ( 2π/A0)
E (e
V)
(111) (100)
Energy Bandstructure by Empirical
PseudopotentialMethod (EPM)
Quantum Ballistic Drive
Current Modeling
Complex E-k
Real E-k
Band-to-Band Tunneling Modeling
(BTBT)
8
quantum ballistic Drive Current Model. The complex band information is needed to
calculate BTBT leakage current.
1.2 Organization
The thesis is organized as follows:
Chapter 1: Introduction
Chapter 2: Calculation of Bandstructure
Chapter 3: Modeling of Band-to-Band Tunneling
Chapter 4: Phonon-Electron Interactions
Chapter 5: Application of BTBT model
Chapter 6: Quantum Ballistic Transport in 1-D Quantum Confined System
Chapter 7: Properties and Trade-offs Of Compound Semiconductor MOSFETs
Chapter 8: Conclusion
9
Chapter 2 Calculation of Bandstructure
2.1 Introduction
The complex movement of carriers in semiconductor can be simplified once we have
the bandstructure of the material. The bandstructure of a semiconductor describes
ranges of energy and crystal momentum (k) that an electron is “forbidden” or “allowed”
to have. It is due to the diffraction of the quantum mechanical electron waves in the
periodic crystal lattice. The bandstructure obtained in real 3-D crystal momentum (k)
space provides us with information on carrier’s density of states, quantum quantization
effective mass, transport effective mass. The band structure also can be expanded in
complex k space where k space is no longer real number rather complex number. Band
information in complex k-pace provides information on the decaying states coupling
two bands.
The pseudopotential method has been widely used to calculate the bandstructure of
semiconductors [4-9]. While ab-nitio methods are capable of calculating the
bandstructures of the wide range of materials and provide relatively accurate values in
entire k-space, they fail to predict observed bandgap and effective mass at the same
time. Local empirical pseudopotential method is a very powerful tool in that it requires
the least number of parameters and provides very accurate bandgaps (error <0.05eV)
and effective mass (error <5%) at the most important symmetry points (Γ, ∆, L).
10
2.2 Principles of Pseudopotential
The complete Hamiltonian for the periodic semiconductor structure is expressed as
follows [10]:
(2.1)
where
2e
(2.2)
and are the positions of the electrons and is the mass. The second term is:
2e
| | (2.3)
where are the positions, the atomic numbers and the masses of the nuclei.
The final term is:
e
,
(2.4)
In general, there are infinitely large numbers of electrons in a typical semiconductor,
and therefore it is almost impossible to track all these electrons. Furthermore, the
strong Coulomb potential of the nucleus terms tend to go to negative infinity as an
electron position gets close to the center of a nucleus, which requires very fine special
gridding in numerical calculation of this system.
However, there are several approximations [11] that allows one to achieve both
acceptably accurate bandstructure and to reduce the computation complexity.
11
i. Pseudopotential approximation (Fig.2.1): In semiconductor materials, the
electrons below the outer shell are tightly bound to the nucleus and the core
states remain almost unchanged by interaction from electrons in outer shell
(or valence bands). This permits us to replace the strong Coulomb potential
of the nucleus and the effects of the tightly bound core electron by an
effective ionic potential acting on the valence electrons.
ii. The independent electron approximation replace the complicated electron-
electron interactions with a time averaged potential.
With these approximations, this system can be treated as a one-electron problem and
the above Hamiltonian in Eq. (2.1) is simplified as:
2 (2.5)
Figure 2.1 : Pseudopotential approximation: The tightly bond inner electron shell can be effectively replaced by the pseudopotential.
0
V(r)
rqz eff
πε4−
Atomic Potential r
3s (valence electrons)
r
Ψ(r) 2s (non-valence)
Screened by core electrons (non-valence)
This Approximation reduces computational complexity
Pseudopotential
12
where is crystal potential that includes the interaction between the electrons and the
nuclei and the interaction between the electrons.
2.3 Band structure calculation with plane wave basis
The orthogonal plane wave (OPW) expansion [5] is the most general approach to
construct the Hamiltonian for periodic systems. Translational symmetry allows one to
build the complete set of plane waves:
,1
√Ωe (2.6)
where Ω is the volume of the lattice, G are the reciprocal lattice vectors which satisfy:
2 , where œ (2.7)
and are the Bravais lattice vectors.
Now we expand the Hamiltonian in Eq. (2.5) with the plane waves in eq. (2.6):
, , , 2| | , (2.8)
Here, we used the orthonormality property of the basis set:
, , , (2.9)
and is the potential given by:
1Ω
e e (2.10)
The crystal potential is expressed as a summation of potential contributions from
each atomic site, .
13
(2.11)
is assumed to be a local pseudopotential, dependent only upon . Although adding
non-local potentials is essential for photoluminescence study and provides better
fitting to wide range of electronic properties [4], it requires too many fitting
parameters. In this work since our interest is only around vicinity of a few symmetry
points (Γ, ∆, L) in valence band and conduction, treating only local terms is good
enough for accurate modeling of the range of interest and results in easier fitting
process. Using this Vc for eq. (2.10) gives:
1Ω
e d1Ω
e e d
ΩΩ
e1
Ωe d
(2.12)
where we write and Ω is the volume of single Bravais lattice point; in
diamond structure, that is:
Ω /4 (2.13)
where, 0A is the lattice constant.
The factor in eq.(2.12) :
1Ω
e d , (2.14)
is the Fourier transform of the atomic pseudopotential aV and is also called as
pseudopotential form factor. Therefore, the complete form for the matrix elements can
be written as:
14
, 2| | ,
ΩΩ
e , (2.15)
, can be obtained by self-consistent ab initio method or by fitting to empirical
data for the bulk constituents [12] (e.g., energy gaps, effective masses). In this work,
since the bulk constituents for the semiconductor materials in interest are well
measured and known, the local pseudopotential form factor will be empirically
fitted. This method is called the local empirical pseudopotential method [5] (LEPM).
2.4 Continuous q-space model potential
To derive the band-structures in relaxed IV or III-V semiconductor materials, only
information of on few numbers is required. In the standard derivation of the
EPM, the assumption is made that the atomic pseudopotentials are spherical with the
consequence that form factors, depend only on the magnitude of the reciprocal
lattice vectors , i.e., q where q | |. High frequency components in
the form factors are safely ignored and in general only 6 form factors are enough to
calculate acceptably accurate bandstructure. However for strained systems or
superlattice systems, the knowledge of the form factors on vicinity of non-reciprocal
lattice points is required. A variety of algebraic forms have been used in the past to fit
empirical pseudopotentials on continuous q-space [7-9, 12-14]. In this work, we take a
form suggested in [9], i.e.:
,1
Ω e 1 (2.16)
15
where are the fitting parameters. By using this form, where the value is
defined for q=0, fitting band-offsets and electron affinity is possible.
2.5 Strain dependence of the pseudopotentials
Under the hydrostatic strain, redistributed local charge affects the screened potentials
in (2.16) [15] . Since we solve this pseudo-potential non-self consistently, it is
necessary to introduce a new fitting parameter to accurately model the hydrostatic
deformation potentials [8, 15]. This approach is too add a strain dependent term is
analogous to the scaling law in the tight-binding method [16]. Our screened pseudo-
potential takes the form:
, , , , 0 1 Tr (2.17)
Figure 2.2 of Ga and As in GaAs. This is the Fourier transform of the atomic pseudopotential.
0 1 2 3 4-1
-0.8
-0.6
-0.4
-0.2
0
0.2
q (2π/A)
V(q)
(ryd
berg
)
MP of Ga and As in GaAs at 300k
GaAs
16
The local hydrostatic strain Tr for a given atom at is basically a hydrostatic
volume change by the strain and it can be expressed as:
TrΩ Ω 0
Ω 0 (2.18)
2.6 Kinetic-energy scaling
While using the strain dependence term and continuous G-space form factors, we can
obtain acceptably accurate (error <0.01eV) bandgaps, it is not easy to get accurate
effective masses. It is due to neglect of non-local many-body potential. Cohen in [6]
showed that when the self-energy term is expanded to Taylor series, the lowest order
term can be represented as the kinetic energy. Thus, in this work we introduce the
kinetic energy scaling constant, , and the Hamiltonian matrix elements in eq. (2.15)
are now written as [9]:
, 2| | ,
ΩΩ
e , (2.19)
By introducing the constant , simultaneous fitting of both bandgaps and bulk
effective masses is achieved.
2.7 G-space smoothing function
The atomic pseudopotential evanesces quickly with the length of the reciprocal-lattice
vector G and the long G vectors have only negligible effect on the band-structure [5].
This property lets us use only small number of plane waves and in this work we use a
17
cutoff energy of 5 Ry, which generates about 60 plane waves. However, for strained
lattices or superlattices, if the reciprocal-lattice vectors are cut off sharply, then plane
waves at high frequency may be arbitrarily added or removed; this may result in
sudden change in bandstructure. To ease this problem, it is beneficial to introduce a
smoothing function to the atomic pseudopotentials [7], i.e. :
12
tanh5
0.31 (2.20)
where q are in atomic unit. After adding the smoothing function and hydrostatic strain
dependent constant, the atomic pseudopotential in eq. (2.17) now becomes:
, ,1
Ω e 11 Tr
12
tanh5
0.31 (2.21)
2.8 Spin orbit coupling
Although it is screened by core electron, extra positive nuclear charge creates an
electric field pointing radially outward. According to the special relativity theory of
Einstein, in the reference frame of an electron in valence band, it appears as if the
nucleus is moving and the moving electron experiences a magnetic field. Therefore
there is a spin-orbit interaction between the magnetic field associated with the orbital
motion of electrons in valence bands and the internal magnetic moment of the electron.
Based on the special relativity theory, the magnetic field (to first order) experienced
by electron moving with momentum in an electrostatic field is given by:
18
(2.22)
where is the electron mass and c is the speed of light. Since the electrostatic field
is roughly spherically symmetric, it can be expressed in spherical coordinates:
14
(2.23)
where e is an electron charge, is the dielectric constant of the material and Z is the
effective atomic number which partially takes into account the effect of screening by
core electrons on nuclear charge. Using eq. (2.23). and the relation , the
magnetic field can be expressed with the angular momentum of the electron , i.e.:
4 4 (2.24)
The magnetic dipole moment of electron spin angular momentum is given as:
g2
(2.25)
,where the factor g is the gyromagnetic ratio of the electron. in eq. (2.25) can be
expressed using Pauli spin matrices.
2 (2.26)
Pauli spin matrices are defined as:
(2.27)
In conjunction with eq. (2.24) and eq. (2.25), the spin-orbit interaction energy can be
written as:
19
g2 4 4 2
g2
(2.28)
In most of semiconductor materials, the outermost p-core states provide the biggest
contribution on the spin-orbit interaction. Inner core states and d-core states may be
neglected [17]. The resulted spin-orbit interaction terms from outermost p-core states
expanded with reciprocal lattice vectors and the result is as follows [4, 17-18]:
, , ,Ω2Ω
e , (2.29)
where we define:
(2.30)
is an adjustable parameters to be fitted to the band splitting. Reiger calculated
and found the analytic approximation form [19]. i.e:
1 ⁄ /51 ⁄ (2.31)
In this work, we cut off G at 5 Ry which corresponds to 4 . For this range of G:
⁄ 0.15 and1 ⁄ /51 ⁄ 1 (2.32)
With this approximation and using the ratio of the spin-orbit splitting of the free
atoms, we only need to find one fitting parameters of to get the band splitting.
Finally, the spin-orbit interaction term used in this work is:
, , ,Ω2Ω
e , (2.33)
Thus, the complete matrix elements including the spin-orbit interactions are:
20
, , ,
2
2 | |2, ,
ΩcΩ e , ,
Ωc2Ω e , (2.34)
The possible spin states combinations (s’,s)
are , , , , , and , .
2.9 Atomic positions in strained materials
2.9.1 Strain tensor
Under applied stress, a solid is strained resulting in a change of volume and shape.
Consider a 2-dimensional deformation of an infinitesimal rectangular material element
with dimensions dx by dy (Fig. 2.3). A vector describes the displacement as a result
of the stretching and it is a function of position. After deformation, it becomes a
rhombus. The distance between a and b, now can be expressed with dx and .:
2 2
1 22 2
(2.35)
For very small displacement gradients (i.e. 1), the distance is approximately, :
(2.36)
21
The normal strain in the x-direction of the rectangular element is defined by
(2.37)
This strain quantity measures the extension per unit length of the vector. Similarly,
the normal strain in y and z directions is:
, (2.38)
When shear strain is applied, the angle between two orthogonal material lines is
changed. The line , is more tilted to y-direction by amount of . The angle
between the line and the line is:
Figure 2.3 Two- dimensional geometric deformation of an infinitesimal materialelement.
22
tan (2.39)
The angle adopting the same derivation is .
For the rigid body to be strained by shear strain should not be zero. If the
addition is zero, the rigid body just rotates without being strained. Thus, the net shear
strain could be defined as:
12
12 (2.40)
Using the same analogy to define other strain tensor components, the 3-dimensional
strain tensor is constructed as:
12
12
12
12
12
12
(2.41)
From the relation above, we can discern:
(2.42)
2.9.2 Stress Tensor
When a solid is strained by external forces and deviates from its equilibrium condition,
the strain leads to the generation of internal forces. The measure of this internal forces
acting on the surface of an imaginary body is called stress. The stress tensor is
defined by:
23
,, ,
(2.43)
Where the are the components of the force vector acting on a small area dA which
can be represented by a vector perpendicular to the area element, facing outwards
and with length equal to the area of the element.
Note that equilibrium requires that the moment resulted by the stress must be zero.
Taking the moment of the forces above about any suitable point, it can be shown that
(2.44)
Figure 2.4 Components of stress in three dimensions
24
2.9.3 Stiffness and Compliance Coefficients
According to the generalized Hooke’s law, it is possible to define the proportionality
constants between stress and strain [20]. :
, , , , , , … , , (2.45)
where are the compliance coefficients and are the stiffness coefficients. Note
that the strain used in eq. (2.45) is called engineering strain. The shear strain of
engineering strain is twice bigger than the strain notation we used in previous
equations. We can construct a 6x6 matrix with either the compliance coefficients or
the stiffness coefficients. When the crystal-axis coordinate system is used for a cubic
crystal, the stiffness coefficients reduce to the following matrix:
c c c 0 0 0c c c 0 0 0c c c 0 0 0
0 0 0 c 0 0 0 0 0 0 c 00 0 0 0 0 c
(2.46)
A similar matrix can be constructed for the compliance coefficients.
2.9.4 Coordinate transformation
To find the components of the strain tensor and elastic stiffness tensor for any device
coordinates, it is necessary to perform a coordinate transformation.
25
The rotational transformation from the crystal axes to the arbitrary system , is
expressed as [20]:
, 1,2,3 (2.47)
The transformation operator can be defined as:
(2.48)
and the operation between the initial coordinate X to the transformed coordinate X’
can be written as:
and (2.49)
Using the result in eq. (2.49), the transformed strain tensor can be expressed with
the initial strain tensor and vice versa:
(2.50)
Figure 2.5: Crystal coordinate system
(001) (111)
(010)
(110)(100)a
26
Thus,
and (2.51)
The elastic stiffness tensor is transformed as follows:
, , , , ,
(2.52)
Similarly the compliance tensor can be transformed as:
, , , , ,
(2.53)
2.9.5 Lattice vectors and reciprocal lattice vectors for a strained
system
The homogeneous bulk strain deforms the atomic locations. The strained lattice
vector can be written in terms of the strain tensor and the relaxed lattice vector ,
i.e:
(2.54)
,where is the unit tensor, of which diagonal elements are one and off-diagonal
elements are set to zero.
The deformation of the lattice vectors leads to change in reciprocal lattice vectors as
well. The strained reciprocal lattice vectors, are calculated as:
1 (2.55)
27
2.9.6 Internal ionic displacement
For orthorhombic strain, all the locations of atoms in a strained crystal can be
expressed as in eq. (2.51). However, in the presence of shear strain, they are deformed
in a different way. Shear strain results in different length of four bonds [21]. As
depicted in Fig. 2.6, if the atoms are relocated only by bulk strain, the resulted bonds
to three circumcenter atoms b, c and d have larger energy than the bond to atom a.
This creates an internal strain in order to minimize the overall bonding energy. The
effect of internal strain can be represented by single proportionality parameter z. The
internal strain relocates the central atom somewhere between the position (z=0)
determined by the bulk strain tensor and the location (z=1) where all four bonds are of
same length. The finally the location of a strained central atom is:
z (2.56)
where is an internal ionic displacement vector depicted in Fig. 2.6. Since z is a weak
function of the level of strain and [22], in this work we use a single constant for each
material.
28
2.9.7 Uniaxial strain
If one-directional pressure is applied along a non-crystal coordinate axis, the applied
pressure generates off-diagonal stress components. The easiest way not to deal with
the off diagonal stress is to transform the coordinate so that its axis is aligned with the
applied pressure. Within that coordinate the applied pressure only induces normal
stress.
Let’s say the pressure is applied along the x-axis of coordinate system A, and the
crystal axis coordinate system is C. Two coordinate systems are connected through
transformation U.
(2.57)
The stress tensor in coordinate system A is:
Figure 2.6: Location of an atom relative to its four nearest neighbors in a crystal strained in the (111) direction. For z=0 the location of the atom is deformed only by the homogeneous bulk strain, while for z=1 all four bonds are of the same length. The vector u represents the differential vector between two locations.
z=1
z=0
a
b
cd
u
29
A0 0
0 0 00 0 0
or 00000
(2.58)
The strain tensor in coordinate system A is then given by:
A A A2⁄ 2⁄
2⁄ 2⁄2⁄ 2⁄
(2.59)
where the elements of compliance matrix can be obtained by applying eq. (2.53) on
. The factor ½ is put to transform the engineering strain notation.
Finally, the uniaxial strain tensor in crystal axis coordinate system C can be found by
performing the coordinate transformation on A as explained in eq. (2.51).:
C A (2.60)
2.9.8 Biaxial strain
A very thin epitaxial film A, of which relaxed lattice constant is , grown on a
substrate with lattice constant is biaxially strained in order to conform to the
substrate. The in-plane strain condition can be expressed as:
(2.61)
In the substrate coordinate system, in-plane strain represents and . Other
elements of the strain tensor can be calculated using two conditions that under biaxial
30
strain only in-plane stress and exists and two in-plane strain elements are of
same value. We first equate the strain tensor to the stress tensor using the compliance
matrix :
222
0000
(2.62)
By equating the first strain tensor element and the second one, and can be
expressed in terms of , i.e:
and (2.63)
Since :
31 32 (2.64)
by plugging eq. (2.63) into eq. (2.64) we have :
(2.65)
Shear strain elements are
2 (2.66)
2 (2.67)
2 (2.68)
31
The most commonly used substrates for semiconductor wafers are [001], [110] and
[111]. Three transformation operators for those substrates take the forms:
1 0 00 1 00 0 1
(2.69)
01
√21
√2
01
√21
√21 0 0
(2.70)
1√6
1√2
1√3
1√6
1√2
1√3
23
01
√3
(2.71)
Using the eq. (2.65)-(2.68) and transforming the strain tensors, the elements for the
strain tensors in the crystal axis coordinate system for these three substrates can be
obtained as:
[001] substrate:
,
2 12
11,
0. (2.72)
[110] substrate:
2 44 12
11 12 44,
,
11 2 12
11 12 44,
0. (2.73)
32
[111] substrate:
4 44
11 2 12 4 44,
11 2 12
11 2 12 4 44. (2.74)
, , and are the stiffness coefficients in the crystal axis coordinate.
2.10 Alloys material (virtual crystal approximation)
Mixing two binary or single 4 column element semiconductors to form an alloy is a
very widely used technique to yield a ternary or binary material with a desirable
bandstructure for specific applications. To calculate the bandstructure of alloy material,
the simplest way would be to put particular atoms at each lattice site. However this
method requires large basis especially if the composition ratio of two mixed materials
is not one to one. The easier method would be assuming that the properties would vary
linearly between the two constituents. This method is known as the virtual crystal
approximation.
In this work, first, we determine the average lattice constant for the alloy. Since the
lattice constant of the alloy is not same as the ones for original materials, two
constituents should be strained to match it to the lattice of the alloy and then we
calculate the bandstructures of two strained constituents. The final bandstructure for
the alloy is obtained by mixing two strained bandstructures linearly in proportion from
components of each material.
33
This method has advantages over other commonly used methods where the form
factors are mixed. It allows using any pseudo-potential form factors without being
problematic for certain materials. In addition, it can accurately reproduce the values of
bowing parameters and hydrostatic deformation potentials measured in optical
experiments.
2.11 Complex Bandstructure
When a large electric-field is applied in semiconductor material, electron in valence
band can tunnel through the band gap and reach at conduction bands. This process is
called band-to –band tunneling (BTBT). During this process electron wavefunction
decays, in other words, it has imaginary wavenumber k. Thus, it is essential for precise
modeling of BTBT to obtain complex bandstructure of semiconductor material.
Normally the Hamiltonian is constructed and solved for eigen-energy for a given k. To
obtain complex band structure using the standard method, it is necessary to find a
specific complex k which ensures the Hamiltonian is still a Hermitian matrix so that
eigen-energy is still a real number. This process is very time consuming and
furthermore, it is quite difficult to find all of the correct complex k.
Instead, in this work, we transform the matrix into the equation finding for eigenvalue
k. Since the Hamiltonian matrix defined in eq. (2.34) is a Hermitian matrix, the eigen-
value of the matrix is also real:
, (2.75)
At the given real and energy , the Hamiltonian can be written as:
34
, , (2.76)
where . is a reciprocal lattice vector from which we
would like to find the complex k. is a complex number and is a unit
vector pointing the direction along which we would like to find a complex
bandstructure:
| | 1 (2.77)
We introduce a new vector , i.e.:
(2.78)
Using the new vector u, eq. (2.75) and eq. (2.76), the equation can be transformed into
the following form.
(2.79)
By diagonalizing a single matrix equation above, it is possible to obtain all of the
complex k solutions for fixed , and .
2.12 Summary
We have developed an empirical pseudo-potential method to calculate the
bandstructures of semiconductors in zinc-blende lattice structure. The atomic
dislocation induced by applied strain is carefully examined including the effect of
internal strain caused by shear strain based on the infinitesimal strain theory. Atomic
35
pseudo-potential is modeled in continuous reciprocal vector space as a function of
strain tensor and reciprocal vector. We included the spin-orbit coupling which causes
splitting in degenerate 3 valence bands. The concept of kinetic energy scaling is
adopted to take into account non-local many-body potential. For alloy materials,
virtual crystal approximation was used. Finally, the complex bandstructure was
obtained by modifying the Hamiltonian matrix such that crystal-momentum k-vector
is eigen-value.
36
Chapter 3 Modeling of Band to Band Tunneling
3.1 Introduction
High mobility materials are being investigated as channel materials in highly
scaled MOSFETs to enhance performance [23-34]. The materials such as GaAs, InAs,
Ge, strained Si and strained Ge have larger carrier mobility than silicon, but the
enhanced band to band tunneling (BTBT) because of their smaller bandgap or direct
band gap may limit their scalability [34]. Although it becomes important to predict the
band to band tunneling leakage of devices made with high-mobility materials, the
most commonly used BTBT models in commercial TCAD tools such as MINIMOSTM ,
MEDICITM, and SENTAURUSTM do not capture quantization effects and full band
information of nano-scale devices. In this work, we have developed a BTBT model,
which captures important quantum mechanical (QM) effects in highly scaled DGFETs.
3.2 Direct-gap Band-to-Band Tunneling.
Under slowly varying envelope function approximation, a wavefunction in lattice
structure can be written in the form:
Ψ , , (3.1)
where Ψ represents the actual electron (or hole) wave function in the band ‘b’ (either
conduction band or valence band). , is unit cell function and normalized in
37
space. , is an envelope function. Fig. 3.1 illustrates the unit cell function, the
envelope function and the crystal potential.
The Schrodinger equation for an electron in non-uniform external potential is:
2,
, ′ , ′
′,
(3.2)
,where xm is a transport effective mass, is an energy of the electron. The inter-
band coupling terms ′are defined as:
′ , ′ ′ , ′ (3.3)
In the above equation, is the force on the electron and it can be defined as,
. The quantity defined as eq. (3.4):
′ , ′ ′ , (3.4)
describes the amount of transition which occurs between the states per unit force.
These off-diagonal matrix elements connect between different bands and can be
Figure 3.1: Crystal Potential and corresponding Wavefunctions. Crystal Potential
Unit‐cell function
,
Envelope Function ,
38
interpreted as band-to-band tunneling (BTBT). If the inter-band terms in the eq. (3.2)
are ignored, the eigen-functions of the eq. (3.2) at the tunneling trails are (R.P.
Feynman, Revs. Modern Phys. 20, 367 (1948)) :
, 2
2 exp ′ ′ (3.5)
where is length of semiconductor along the tunneling direction and is an energy
of the band in the periodic lattice.
Note that ′ is a complex number inside the bandgap and the envelope wave function,
, becomes a decaying function toward the center of bandgap. Fig 3.2
illustrates the complex band structure of germanium along the [111] direction. The
black line depicts the real bands and the red line depicts the complex bands. The most
probable direct-gap tunneling occurs along the red line connecting Γ and Γ . When
electron at the top of the valence band starts tunneling toward a bandgap, ′ of the
tunneling wavefunction is given by values in the red line.
39
The tunneling probability per unit time between different bands is given by Fermi-
golden rule:
2′ (3.6)
and the perturbation potential connecting two bands b to b’, ′ is:
′ , ′ , ′ ⁄1
, ′ , ′(3.7)
where and are classical turning points as depicted in Fig. 3.3 and V is the
volume of the semiconductor. The density of state is given by inverse of energy
spacing between Wannier-Stark ladders (Δ :
(a) (b)
Figure 3.2: (a) 3D plot of complex bandstructure of germanium along the [111]direction. The black lines depict the real bands and the red lines depict complex bands. (b) 2D plot of complex bandstructure of germanium.
-0.2-0.15
-0.1-0.05
-0.200.20.40.60.8
‐3
‐3.5
‐4
‐4.5
‐5
‐5.5
00.8 0.400.10.2√ √
Energy (eV)
√√0 0.4 0.8 1‐0.2
Energy (eV)
‐3.5
‐4
‐4.5
‐5
‐5.5
40
1 Δ⁄2
(3.8)
,where and is a lattice constant.
Using equation (1.4), equation (1.5) is rewritten as:
′ (3.9)
By plugging eq. (3.5) into eq. (3.7), we have an expression:
exp ′ ′
.
(3.10)
After integrating over y and z dimensions first, the eq. (3.10) becomes:
Figure 3.3: Illustration of electron tunneling process in the semiconductor with thestrong electric field. The electron in valence band tunnels in to thebandgap at the classical turning point, , passes the branch point, and tunneling ends at .
E(k)
E(k)E(k)
Conduction BandValence Band
Energy
, ,
41
1
, ,
exp ′ ′
.
(3.11)
The terms , , represent the momentum conservation along the transverse
directions during the tunneling process. By using exchange of variables,
and , eq. (3.11) becomes as:
1
, ,
exp ′ ′
(3.12)
According to two band theory, has a pole around a branch point, . Around
the pole, is:
1
4
1
4
(3.13)
, where is a branch point in x-space. We rewrite the exponential term in eq. (3.12),
exp ′ ′ around the branch point ,
exp ′ ′ exp (3.14)
, where ⁄ and is:
′ ′,
′,
42
,,,
(3.15)
.With eq. (3.13) and eq. (3.14), perturbation energy element, becomes:
14 , , exp
(3.16)
The complex integration is performed in the vicinity of :
exp
exp
exp exp 3exp exp 3
, ∞
,
,
,
,
, ∞
2 6 exp43 exp
By using the complex integration result, off-diagonal perturbation energy elements is:
3 , , exp (3.17)
To compute the BTBT rate per unit volume, we integrate transition rate, in eq. (3.9)
over the entire k-space.
3 exp , , ,
2 3 exp 2 2 2
∞
∞
∞
∞
36 exp 2∞
∞
∞
∞
43
exp 2 1 ′ ′| ,
36′ ′
∞
∞
∞
∞
72
exp 2| ,
∂ ∂
(3.18)
, where is the imaginary part of complex number .To obtain the approximated
value of coefficient, / ∂ / ∂ , we adopt the Kane’s two band model.
1Γ Γ Γ⁄
Γ.
Γ.
4Γ
.
4 Γ 2 Γ 2 Γ
.
| ,
∂ ∂3 Γ
.Γ
.
16 Γ.
Γ.
(3.19)
where Γ is direct bandgap tunneling mass, Γ is relative density of states
(DOS) in y-direction and Γ is relative DOS in z-direction. These mass numbers
are to be determined by bandstructure calculation. Plugging eq. (3.19) into eq. (3.18)
results in total direct band-band tunneling rate per unit volume:
2 Γ.
Γ.
27 Γ.
Γ. exp 2 | , .
(3.20)
When electric field is relatively smaller than electric field in crystal momentum, the
path integration, in eq. (3.20) has pretty large number. Since the evaluation of
44
involves the path integral and is sensitive to changes in electric field during the
tunneling, we are allowed to compute only numerically. As pointed out by
Krieger [35], inaccurate evaluation of results in an error of several orders in
BTBT rate, due to the nature of exponential functions. In most of semiconductor
materials, where 4 valence bands are strongly interacting with each other, using the
Kane’s two band model inevitably leads to large amount of error. Furthermore, Kane’s
model is inherently unable to deal with strained materials.
Here we model as:
Γ′
1 Γ Γ Γ
Γ, 0 Γ
1 Γ Γ Γ Γ
Γ, Γ Γ
(3.21)
Γ Γ Γ Γ⁄ , Γ Γ Γ Γ , ΓΓ Γ
Γ Γ,
(3.22)
Γ′
(3.23)
where is the electron energy measured from the top of the valence band at point
x. Two fitting parameters Γ and Γ represent the BTBT tunneling mass for
valence band and conduction band respectively. Γ and Γ are extracted so that
Γ′ in eq. (3.21) matches with the complex band structure calculated by empirical
pseudo-potential method. We can see in Fig. 3.4 this suggested model can be precisely
fitted to bandstructure of semiconductors including strained materials. The parabolic
45
effective mass approximation (Fig. 3.4(a)) fails to follow the complex dispersion in
relaxed germanium. While the Kane’s two-band model works well for relaxed
materials (Fig. 3.4 (a)), it breaks down in strained materials (Fig. 3.4(c)). The
proposed model well approximates the dispersion in both relaxed and strained
germanium (Fig. 3.4( b, d)).
(a) (b)
(c) (d)
Figure 3.4: Complex band structure of Ge obtained from local empirical pseudopotential method (LEPM) along the [111] direction. Parabolic effective mass approximation (a) fails to follow the complex dispersion in relaxed germanium. While the Kane’s two-band model works well for relaxed materials (a), it breaks down in strained materials (c). The proposed model well approximate the dispersion in both relaxed and strained germanium (b, d).
0 0.04 0.080.020.06
‐3.4
‐4
‐4.6
‐5.2
Energy (eV)
√√
LEPM
Two‐Band
Effective mass
Relaxed Ge
0 0.04 0.080.020.06
‐3.4
‐4
‐4.6
‐5.2
Energy (eV)
√√
LEPM
This work
Relaxed Ge
0 0.04 0.080.020.06
‐3.4
‐4
‐4.6
‐5.2
Energy (eV)
LEPMTwo‐Band
Strained Ge on Si (001)
0 0.04 0.080.020.06
‐3.4
‐4
‐4.6
‐5.2
Energy (eV)
LEPM
This work
Strained Ge on Si (001)
46
3.3 Phonon-Assisted Tunneling / Indirect-gap Tunneling
In an indirect bandgap material, like silicon or aluminum arsenide, the smallest energy
separation for electrons in conduction band and holes in the valence band occurs for
very different crystal momentum. Band-to-Band tunneling is possible between these
states with the participation of a phonon to conserve momentum. Fig. 3.5 illustrates
the possible tunneling paths. Note that since this process requires both momentum and
energy to conserve, during the transition the electron either emits the phonon energy
or absorbs the phonon energy. Usually the energy of phonon is around tenths of meV.
The bandstructure of semiconductor is influenced by changes in lattice spacing. At
room temperature, vibrating atoms perturb lattice constant and bandstructure. For each
Figure 3.5: Illustration of band-to-band tunneling paths. The electron can tunnel to indirect-gap conduction band (L or ∆ valleys) through phonon interactions.
ConductionBand
Valence Band
Energy
Complex k‐+
-
-
Tunneling Process
Real k
Γc
Γv
-
-
L or ∆
Conduction Band
Real k
--
-
Phonon
Direct Tunneling (Γv Γc)
Indirect Tunneling : Phonon couples
Γv to Lc /∆c
47
wavelength of a vibration mode in a Zincblende crystal structure, there are six
branches, two transverse acoustic modes (TA), one longitudinal acoustic mode (LA),
two transverse optical modes (TO), and one longitudinal optical mode (LO) (Fig. 3.6
(a) and (b)).
Lattice displacement by phonon vibration of wave-vector can be expressed as:
12
12 2
exp (3.24)
where the top sign applies for phonon emission and the bottom for phonon absorption.
σ is phonon polarization vector (Fig. 3.6 (c)), is density of crystal, is phonon
frequency, is volume of crystal and is the number of the phonons in the mode.
The number of phonons is given by the Bose-Einstein factor as:
1exp ⁄ 1
. (3.25)
(a) (b) (c)
Figure 3.6: (a) Longitudinal and transverse waves in a 1D linear lattice. (b)Transverse optical and transverse acoustical waves in a 1D diatomiclattice. (c) The ion is displaced from equilibrium by an vector, . is the unit-length polarization vector of the lattice vibration.
Direction of Propagation
Displaced Atoms Atoms at equilibrium
Tran
sverse to Propagation Direction
Longitudinal mode
Transverse mode
Optical Mode
Acoustic Mode
Atom at equilibrium
Displaced Atom
48
According to the Rigid Ion Model [ref], the perturbation energy, δW caused by the
phonon mode is:
δW Nωβ12
12 2ρVωβ
exp (3.26)
where is the crystal potential.
We first establish the electron-phonon interaction matrix elements between two Bloch
states. Consider the Bloch functions in conduction band and valence band:
, exp i · and , exp i · . (3.27)
The electron-phonon interaction matrix elements H are:
H , exp · δW , exp · (3.28)
By inserting eq. (3.26) in eq. (3.28), we find:
H 12
12 2
, ,
exp ·
(3.29)
Since in the eq. (3.29) the unit-cell functions and the phonon perturbation terms are
varying fast in the unit cell, while the exponential term swings slowly, we calculate
the integration in the unit-cell first and then we have:
H12
12 2
ΩV
exp ·
, ,V
12
12
(3.30)
49
The delta functions in the matrix elements represent the conservation of momentum
and energy. The quantity
2 , ,V
(3.31)
describes the strength of interaction between electrons and phonons in the certain
phonon mode. These quantities have been measured for inner conduction bands or
inner valence bands but hardly measured for inter-bands. Several groups have
attempted to calculate by tight-binding or pseudo-potential method [36-38], but
most of the results are obtained between the symmetry points in real k-space. In the
next chapter we will elaborate on calculating proper the strength of interaction for
BTBT process and suggest values to be used in this BTBT model.
Now we work on the tunneling situation. Consider two tunneling states in different
bands defined as eq. (3.1) and eq. (3.5):
, exp i and ,
where is the momentum difference in the indirect bandgap. The electron-phonon
interaction matrix elements between these tunneling states are:
H , exp i δW , (3.32)
Since envelope functions and are slowly varying and stay nearly constant
in the unit cell, we can first perform the integrations within the each unit cell by
utilizing eq. (3.31).
50
H12
12 exp i
12
12
1exp i
(3.33)
Here we define as . Note that the integral part in eq. (3.33) have
significantly large amount only when is close to zero, or in other words, is close
to . This result naturally occurs to comply the momentum conservation rule.
Generally in indirect bandgap materials, valence band maxima is located at Γ point in
the Brillouin zone and conduction band minima at Δ, , or L points. Fig. 3.7 shows
phonon dispersion curves in germanium. For around these zone boundaries, can
be approximated to be constant and can be also assumed to be constant.
Figure 3.7: Phonon dispersion relations in germanium. Generally at zone boundaries such as 2 ⁄ 0.5,0.5,0.5 and 2 ⁄ 1,0,0 , the frequencies
are roughly constant.
The total generation rate of indirect-gap tunneling per unit time and per unit volume in
the semiconductor can be obtained by summing up transition rates for all the phonon
modes and carrier states in conduction band and valence band.
1012Hz8
6
4
2
TOLO
LA
TA
TO
LO
LA
TA
L (111) X (100)Γ
51
2, , ,
2H , 2
2, , , ,
2H , 2 2 2 2
(3.34)
In eq. (3.34), is conduction band wave-vector transverse to the tunneling direction
and is the one for valence band. represents the phonon modes such as
longitudinal acoustic mode (LA), longitudinal optical mode (LO), transverse acoustic
mode (TA) and transverse optical mode (TO). is the transition rate between two
states by phonon mode . is obtained by summing up all the transition rates over
the all the phonon states in the mode . Not all the phonon modes can participate in
tunneling process, since the selection rule, determined by symmetry of unit cell
functions and phonon modes, allows only specific types of phonon to couple between
bands. We first evaluate the first integral in eq. (3.34) :
H , 2
12
12 ,
1 ′ ′ ′ exp i ′
2 .(3.35)
Using the property of function:
exp i ′2
′ ,
we can rewrite eq. (3.35) as:
52
H , 2
12
12 , | | | | .
(3.36)
We can further evaluate the overlap integral in eq. (3.36) by substituting eq. (3.5) to
envelope functions:
| | | |
1
exp 2 ′ ′ .
(3.37)
The exponential term in eq. (3.37) has a maximum value at the branch point where the
first derivative of the exponent goes to zero, that is at , or in x-
space it is at . Existence of the maximum value and the second derivative of the
exponent let us use the method of steepest decent to compute the integral. After the
integration we have:
| | | |
|
exp 2
(3.38)
where
⁄ . (3.39)
53
Note that, since here we assume that the tunneling wavefunction in valence band
decays as x increases and the wavefunction in conduction band decays as x decreases,
both and are positive values. In eq. (3.39), has a correction term,
⁄ , since the phonon energy emission/absorption process effectively
increases/decreases the tunneling length.
Again, as we argued before, it is critical to accurately evaluate due to the nature of
exponential functions. However, the pre-factors in eq. (3.38) such as:
and
|
are safe to be obtained using an approximation. The parabolic band approximation let
us write:
2 Δ 2 Δ,
2.
(3.40)
In eq. (3.40), Δ is the indirect bandgap. Since the phonon energy is much
smaller than Δ , we ignored . We call the energy at the point , the
branch point energy . By taking the derivatives of eq. (3.40) the pre-factors in eq.
(3.38) are:
|
12 Δ
12 Δ
,
54
|
12 Δ
12
|
2 .Δ
. .
(3.41)
, where .
With the results of eq. (3.38) and eq. (3.41), we can rewrite eq. (3.37) as:
H , 2 Nωβ12
12 ,
. . .
2 . .Δ
. . exp 2ξ
(3.42)
The transition rate between two states by phonon mode , in eq. (3.34) is now
rewritten as:
2H , H , 2 exp 2ξ
(3.43)
where
. . .
2 . .Δ
. . Nωβ12
12 ,
(3.44)
By plugging eq. (3.43) into eq. (3.34), we are allowed to write:
55
∑ exp 2ξ
∑ exp 2ξ | , ∏ ⁄
| , , , ,
(3.45)
To evaluate the product term in eq. (3.45), we use the parabolic band approximation
again as described in eq. (3.40).
2 .
3
.Δ
.
,
2 .
3
.Δ
.
132 2 Δ
.
Δ.
2 . . (3.46)
Thus, the product term is:
/2
| ,, , ,
.
2 Δ .
(3.47)
Finally, by substituting eq. (3.47) into eq. (3.45), we have the generation rate of
indirect-gap tunneling per unit time per unit volume in the semiconductor, :
. .
2 . . .Δ
. ., , N , ,
12
12
exp 2 , | , ,
. .
2 . . .Δ
. ., , N , ,
12
12
exp 2 , | ,,
(3.48)
where denotes different conduction band valleys. In silicon, there are 6 conduction
band minima located at Δ points in the Brillouin zone. The efficiencies of tunneling in
56
these six valleys are different. Thus, to obtain the BTBT generation rate in silicon, it is
necessary to evaluate the tunneling rates in each of 6 valleys. In germanium or
aluminum arsenide, the conduction band minima are located at 4 L points in the
Brillouin zone and we need to repeat the process 4 times. In the materials like strained
SiGe, band minima of Γ, Δ, L valleys have similar energy levels. In such materials
both direct-gap tunneling and indirect-gap tunneling contribute significant amount to
tunneling current.
To calculate the path integral , in eq. (3.48), we need the precise model for E-k
dispersion relation in bandgap. For the model, we take the similar form as eq.(3.21),
(3.22) and (3.23).
Γ′
1 Γ Γ Γ
Γ, 0 Γ
1 Γ Γ Γ Γ
Γ, Γ Γ
(3.49)
Γ ΓΓ
Γ Γ, Γ Γ
Γ
Γ Γ
(3.50)
,′
1 , , , ,
,, α, ,
(3.51)
, Γ′
,′
, ⁄
(3.52)
57
In the model above, basically eq. (3.49) and eq. (3.50) are same as eq. (3.21) and
(3.22). Γ′ represents E-k dispersion in direct-gap and ,
′ represents E-k dispersion
in indirect-gap. Just like the direct-gap tunneling case, is the electron energy
measured from the top of the valence band at point x. Γ is the direct bandgap and
, is the indirect bandgap of the valley . Fitting parameters are Γ , Γ , ,
and , . Note that in all the semiconductors, the branch point of the indirect-gap,
, α, is located below the branch point of direct-gap, Γ . In all other non-
local BTBT generation models for TCADs, the E-k dispersion used in path integration,
, is modeled based on the parabolic effective mass approximation as in eq. (3.40).
Most of reported works on BTBT in germanium are done incorrectly based on the
same approximation as well [39-43]. As depicted in Fig.3.8 the parabolic effective
mass approximation completely breaks down especially in germanium where the
energy level of conduction band extremum at Γ point is close to the energy level of
bandedge of conduction band L valley.
58
(a) (b)
Figure 3.8: Complex band structure of germanium obtained from local empiricalpseudopotential method (LEPM) along the 110 direction. The valley located at k-point 2 ⁄ 0.5,0.5,0.5 is projected to Γ point.The parabolic effective mass approximation (a) completely fails toestimate the complex dispersion, since the dispersion curve at Γ pointdeviates a lot from the parabolic curve. The suggested model (b)successfully approximates the dispersion curve.
√√0.06 0.02 0.04 0.080
Energy (eV)
‐3.8
‐4
‐4.2
‐4.4
‐4.6
‐4.8
LEPMEffective mass
√
√0.06 0.02 0.04 0.080
Energy (eV)
‐3.8
‐4
‐4.2
‐4.4
‐4.6
‐4.8
LEPM
This work
59
3.4 2D Quantized System
In the thin channel, as seen in Fig. 3.9, states are quantized. When quantum
confinement is strong in the vertical direction (z), we can decouple the two-
dimensional Schrodinger equation in two one-dimensional equations in the vertical (z)
and longitudinal (x) directions. For certain valleys and orientations, vertical direction
and longitudinal direction are coupled due to nonalignment of the device coordinate
system with the principal axes of the constant energy conduction-band ellipsoids. If
the potential variations along the transport direction are slow enough, it is possible to
decouple two directions by adopting the Rahman’s decoupled effective-mass approach
[44]. In thin film channels with weak longitudinal voltage variations, wave function
can be decoupled using subband decomposition and it is,
Ψ , , , , ; , exp , , (3.53)
where n is the quantum number of quantized states and denotes bands (several
conduction bands valleys or a valence band). The wave-functions , ; are
Figure 3.9: Schematic view of the thin channel double-gate MOSFET (DGFET).Thickness of the channel is . Current flows along x-direction and the channel is quantized along the z-direction.
ChannelInsulator
Gate
Gate
xz
y
zx
z=0 z=tb
Eg
Insulator
Insulator
Energy
0
,
,
,
,
, ;
, ;
, ;
, ;
, ,
,
,
60
normalized in z direction and unit-cell functions , are normalized in the
space. 1D Schrodinger equation in the z confined direction provides eigen-functions
, ; and eigen-energies , .
21
,, ;
, , ; , , ; (3.54)
,where , is the bandedge energy level of band .
The wave functions , satisfy the following 1D Schrodinger equation in x
direction (or channel direction).
21 ,
′ ′,, , ′,
22 ′ ′ ′,
′, ,
0
(3.55)
where the coefficients, ′ , ′ and ′ , represent subband-to-subband
couplings and are given by,
′ , ′ ′ 1 , ′ , ′
, ; ′, ; ,
(3.56)
where
′ , , ′, , , (3.57)
, (3.58)
and
61
′ , ′ , ′ , ; ′, ; ,
′ , ′ , ′ , ; ′, ; . (3.59)
If channel thickness ( is much smaller than channel length ( ), , ; weakly
depends on x and the higher-order coupling terms, ′ , ′ , … can be safely
ignored. Since our semiconductor of interest is the slab where electron states are
confined, the solutions of Hamiltonian, sub-band Eigenfunctions , ; form
complete orthogonal set and the overlap integrals between them
( , ; , ; ) are zero as long as they stay in the same band ( ) .
Thus, the first order coupling terms ′ are zero for and the inner-band
couplings between two sub-bands are prohibited.
We can also prove that the inter-band couplings ( ) between two sub-bands with
different quantum number ( ) are prohibited. Consider the two-band system
with two sub-bands for each. The Hamiltonian for the system can be written as
Δ , 2⁄ ⁄ ⁄ ⁄
⁄ Δ , 2⁄ ⁄ ⁄⁄ ⁄ Δ , 2⁄ ⁄⁄ ⁄ ⁄ Δ , 2⁄
(1.60)
,where
62
, , , ′,
, , , ′, .
(3.61)
Δ , are the quantization energies of each sub-band and is the bandgap of the
semiconductor. From the orthogonal properties of , , we can set in eq.
(3.61) to be zero when and . The solutions of eq. (1.60) cannot be easily
put in analytical forms, but we cane roughly estimate them as
, Δ , 2
| |, Δ ,
| |, Δ ,
(3.62)
, Δ , 2
| |Δ , ,
| |Δ , ,
(3.63)
, Δ , 2| |
, Δ ,
| |Δ , ,
(3.64)
, Δ , 2| |
, Δ ,
| |Δ , ,
(3.65)
For the second sub-band in conduction band ( , ) to couple to the first sub-band in
valence band ( , ), there must be an intersection in energy between , and , at
the certain imaginary number k (Fig. 3.10).
63
Figure 3.10: Illustration of complex bandstructure of two-band two-subband system.
The left panel depicts the scenario where bands with same quantumnumber are connected and the right panel shows the scenario where bandswith different quantum number are connected.
By equating eq. (3.62) and (3.63), we have
Δ , Δ ,| |Δ , ,
| |, Δ ,
| |Δ , ,
| |, Δ ,
.
(3.66)
Using the eq. (3.61), eq. (3.69) can be rewritten as:
Δ , Δ ,| | , ,
Δ , ,
, ,
, Δ ,
, ,
Δ , ,
, ,
, Δ ,
(3.67)
Since Δ , Δ , and (k is a pure imaginary number) are both larger than
zero, to have a solution, the number inside the brackets in eq. (3.69) should be larger
than zero as well. Where the potential variation in z-direction is relatively small like in
Energy
Complex k
‐
Real k
Γc
Γv ΓvReal k
Intersection
(c,2) (v,2) (c,2) (v,1)
(c,1)
(c,2)
(v,1)
(v,2)
(c,1)
(c,2)
(v,1)
(v,2)
Γc
64
the thin-channel DGFETs or SOI FETs, , and , have similar symmetry and
generally the following inequality equation (3.68) holds:
, , > , , , , . (3.68)
Due to eq. (3.68) and the fact that Δ , Δ , , the number inside the brackets in
eq. (3.69) is smaller than zero and we can conclude that the there is no crossover in
energy between , and , at the certain imaginary number k and the inter-band
couplings ( ) between two sub-bands with different quantum number ( )
are prohibited.
Finally, after taking out the forbidden transitions, the eq. (3.55) is simplified as:
21 ,
, , ′ ′,′
, (3.69)
where
′ , ′ ′ , ; ′, ; . (3.70)
If the inter-band terms in eq. (3.69) are ignored, the eigen-functions of the eq. (3.69) at
the tunneling trails have a form:
, 2 2 exp ′ ′ (3.71)
65
3.5 Direct-gap Band-to-Band Tunneling (2D)
The coefficients ′ include both inner-band and inter-band coupling. For
intervalley coupling Tunneling probability between two states in different conduction
band can be calculated using eq. As seen in eq. (3.72) the zero-th order coefficients
contain the overlap integral along the vertical direction (z). This overlap integral in
fact, the momentum conservation between conduction band sub-band state and
valence band sub-band state.
, , ⁄
,,
(3.72)
where
, ; , ; . (3.73)
The constant in eq. (3.73) denotes the thickness of the channel. We can notice that
in eq. (3.72) is exactly like the corresponding result for 3D carriers in eq. (3.17),
except that a form factor is used instead of the delta function , . The
term , represents the momentum conservation which is applicable only to the
infinite space. In quantized 2D carriers, due to the Heisenberg’s uncertainty principle,
66
crystal momentum has an uncertainty in its value, and form factor appears as a
result.
Using the results from eq. (3.12) to eq. (3.17), we can find:
3 ,exp
(3.74)
where
Γ′ ′.
(3.75)
The point is the branch point between nth valence subband and nth conduction
band subband. The point is where the electron in n-th valence subband starts to
enter the bandgap and the point is where the electron ends tunneling by reaching
the n-th conduction subband.
The tunneling probability per unit time between different bands is given as:
, , 9 exp 2 , (3.76)
Total generation rate per unit area by direct-gap band-to-band tunneling is obtained by
summing up tunneling rates in eq. (3.76) over the entire k-space.
, , ,
, , ,
29 exp 2 2 2
67
118 exp 2
/2⁄ .
(3.77)
Taking the similar step as in eq. (3.19), we have:
,
.
4 1 2 G,
.
and /2
⁄2 . . .
DOS.
3 . .G,. . .
(3.78)
Thus the total generation rate per unit time and unit area is
,1
18, exp 2
2 . .,
.DOS
.
3 . .G,. .
2 . .
DOS.
3 . . .,
.
G,. exp 2
(3.79)
Total generation rate per unit volume per unit time can be expressed by multiplying
square of sub-band eigen-functions, , ; . The electron generation rate is:
, , ,
2 . .DOS
.
3 . . . , .
G,. exp 2 , ,
(3.80)
and the hole generation rate is:
68
, , ,
2 . .DOS
.
3 . . . , .
G,. exp 2 , ,
(3.81)
where
Γ′ ′.
(3.82)
Again, the precise modeling of the wavenumber Γ′ is critical in accurate
prediction of BTBT generation rate. Fig. 3.11 shows the complex bandstructure in the
quantized thin slab. Quantum confinement strongly modifies complex bandstructure.
Each state in conduction band and valence band with same quantum number connect
each other through forbidden bandgap.
Figure 3.11: Complex bandstructure of 14nm thick (001) germanium. The imaginaryvalues of the wavenumber k are drawn as red curves and the real valuesare shown in black curves.
Since it is difficult to track and store all the imaginary k connecting between Γ and Γ
inside the bandgap, instead we propose a fitting model (eq. (3.83)) for the imaginary
Energy (eV)
√√
(110)(110)
00.5 0.50.3 0.30.1 0.1
‐3
‐4
‐5
69
part of k by using two variables Γ , and Γ , . These variables are
functions of the quantum confinement energy (Δ , ) of the tunneling states (eq.
(3.84)).
Γ′
1 , ,
,
, 0 ,
1 , , ,
,
, , ,
(3.83)
where
, , , , Δ , , , (3.84)
, , ⁄ , (3.85)
, , , , (3.86)
and 1
1 Δ ,⁄ 1 Δ ,⁄.
(3.87)
where , and , are the nth eigen-energies of the Γ point states of the slice
at x, in conduction band and valence band, respectively.
Fig. 3.12 illustrates the imaginary wavenumber Γ′ generated by LEPM and by our
model. It shows that our model successfully approximate the data given by LEPM.
70
The fitting function for the variables Γ , and Γ , are shown in Fig.
3.13.
Figure 3.12: Γ′ as a function of quantum confinement energy (Δ ) (eq. (3.83)) in
(001) germanium. Tunneling direction is along the (110) direction.
(a) (b)
Figure 3.13: Γ , Γ , used in Fig. 3.12 and their fitting functions.
√√
Energy (eV)
(110)(110)
‐3.4
‐3.8
‐4.2
‐4.6
0 0.040.040.08
=0.54 eV0.350.170.05
0 eVLEPMThis model
4
6
8
4
6
0 0.2 0.4 0.6 0.8 10.02
0.06
0.10
0.14
.
0 0.2 0.4 0.6 0.8 1
0.1
0.2. .
71
3.6 2-D Phonon-assisted Indirect-gap Band-to-Band
Tunneling
Consider two tunneling states in different bands defined as eq. (3.53) and eq. (3.71),
first, the wavefunction for the nth subband in conduction band:
Ψ , , , , ; , exp · , , , (3.88)
and the wavefunction for the mth subband in valence band:
Ψ , , , , ; , exp · , , (3.89)
where is the momentum difference in the indirect bandgap.
Using the eq. (3.33), the phonon-electron interaction off-matrix element can be written
as
H , , ,
12
12
.
,.
, ; , , ; , exp i
(3.90)
where we define as and is the phonon wave vector. The letter
denotes the phonon mode.
To evaluate the generation rate we need to add the square of perturbation energy in eq.
(3.90) over all the phonon states in the phonon mode j. By following the steps given in
eq. (3.35) and eq. (3.36), we have:
72
H , , 2 ,
12
12 ,
, , /
(3.91)
where
1 , ; , ; (3.92)
Just like the quantity obtained for 2D Direct-gap tunneling in eq. (3.73), the
quantity describes the effective extent of the interaction in the z-direction.
We further evaluate eq. (3.91) using equations (3.37)-(3.42) :
H , , 2 ,
12
12 ,
exp 2 . . .
2 . . . .
(3.93)
where
, . (3.94)
The point is the branch point and the energy is an energy gap between
mth valence subband and nth conduction band. The wavefunction , starts to
intrude the bandgap at point and , at the point . The driving force
is defined as:
12 (3.95)
73
and .
Now we can evaluate the transition rate between two subband states by phonon mode
, using eq. (3.93):
2H , , H , 2
exp 2 (3.96)
and in eq. (3.96) is defined as:
. . .
2 . . . . ,
12
12 ,
(3.97)
where the quantity is the volume of the semiconductor and it equals to .
The total generation rate per unit time and unit area in semiconductor can be obtained
by summing up the transition rates for all the phonons modes and carrier states in
conduction and valence subbands:
, 2 exp 2, , ,
2 2 2
2
2, , ,
exp 2 | , 2⁄
, ,.
(3.98)
where denotes different valleys in conduction band.
74
The product term in eq. (3.98) was already evaluated in eq. (3.46). By inserting eq.
(3.46) into eq. (3.98), , becomes:
,
2
2 exp 2 | ,
.,
2 . . ., , ,
.
2 . . . .,
12
12 ,
.,
.
.
exp 2 | ,
., , ,
(3.99)
or it can be rewritten using 2D-DOS mass of the plane:
, 2 . . .
,12
12 , ,
.
. .
exp 2 | ,
, , ,
.
(3.100)
Total generation rate per unit volume per unit time can be expressed by multiplying
square of sub-band eigen-functions, , ; . The electron generation rate is:
, , , 2 . . .
,12
12 , ,
.
. .
exp 2 | , , ,, ,,
(3.101)
and the hole generation rate per unit volume per unit time is:
, , , 2 . . .
,12
12 , ,
.
. .
exp 2 | , , ,, ,,
(3.102)
75
Now we model the path integral | , for indirect-gap tunneling as follows:
, ,⁄
(3.103)
where is defined in eq. (3.83) and
, 1 Δ Δ , Δ
Δ ,
, . (3.104)
In the eq. (3.104), , is the energy level of the m-th subband in valley
measured from the valence band edge. Both Δ and Δ are variables
to fit eq. (3.104) to complex bandstructure of quantized semiconductor film.
Figure 3.14: L , ,′ as a function of quantum confinement energy (Δ ) (eq.
(3.104)) in (001) germanium. Tunneling direction is along the (110)direction.
00.050.1
Energy (eV)
‐4
‐4.4
‐3.6
0.05 0.1√√
(110)
LEPMThis model
=0.37 eV
0.26
0.17
0.09
0eV
76
(a) (b)
Figure 3.15: , Δ L and , Δ L used in Fig. 3.14 and their fitting functions.
3.7 Summary
We have developed the BTBT model which takes into account full bandstructure,
direct and phonon assisted indirect tunneling, quantum confinement effect, non-
uniform electric field (non-local) and strain induced enhanced/suppressed tunneling in
semiconductors. In this model, the tail of electron wavefunction penetrating into the
bandgap is modeled and used to evaluate the interband matrix elements between
conduction band states and valence band states. The final BTBT carrier generation rate
(GBTBT) was calculated by adding up the transition rates obtained by Fermi-Golden
rule for all the possible transitions.
0 0.1 0.2 0.3 0.4 0.50.08
0.1
0.12
0.09
0.11
,
, . .
0 0.1 0.2 0.3 0.4 0.5
0.9
1
1.1
1.2
,
, . .
77
78
Chapter 4 Phonon-Electron Interactions
4.1 Introduction
During indirect tunneling process, only phonons with certain symmetry are allowed
to couple the conduction band to the valence band. Selection rules concerns if the
phonon-interaction strength, Mj,phV-1vanishes or not. Previous works on indirect
tunneling by Kane [41] and Keldysh [45] does not distinguish between different
phonon modes. Although Fischetti [39] tried to take into account various phonon
modes, he failed to consider selection rules and arbitrarily assumed all four phonon
modes (transverse acoustic (TA), transverse optic (TO), longitudinal acoustic (LA)
and longitudinal optic (LO)) contribute to BTBT current with almost equal strength.
Lax [46] elaborated selection rules based on group theory among symmetric points in
the Brillouin zone and explained the experimental results on electron-phonon
interactions – mobility and electron photoluminescence [38-39, 47-48]. However, the
selection rules fail to explain the results obtained by Chynoweth on Si and Ge Esaki
tunneling devices. In previous studies only phonon matrix elements between the
symmetry points in real bandstructure were obtained. In our work, for the first time we
use the electron orbital wavefunction in decaying states (or inside bandgap) in
calculating the interaction strengths and show that treating the phonon-electron
interaction in complex band structure successfully explains experiments. By adopting
the rigid-ion model [49] and using the empirical pseudo-potential method (EPM), we
suggest proper phonon-interaction strength during band-to-band tunneling process.
79
4.2 Electron-Phonon Interaction Element- Rigid Ion Model
In the previous chapter, we have shown that the matrix element describing the
strength of phonon interaction connecting two states can be written as:
, , 2 , ,V
(4.1)
, where l is the phonon mode, lσ is the polarization vector described in the chapter
on BTBT modeling.
This equation can be expressed in reciprocal space as [49]:
, , 2′
, ′,′, ,
′ exp ′ (4.2)
, where jτ = ( )( )/ 8 1,1,1la± are basis vector, la is the lattice constant andC kG is the
expansion coefficient of the pseudo-plane-wave representation of ku .
Conventionally, the phonon interaction strength ph,l,βM is written in the form of
intervalley deformation potentials DK, which is defined as:
, ,2
, , ⁄ (4.3)
Equation (4.3) can be readily adapted for use with diamond crystal structures. We
have evaluated the equation by using 118 pseudo-plane-wave expansions including
spin orbit coupling.
80
4.3 Phonon Polarization Vectors
The phonon polarization vectors, lσ are given by Lax [50] using group theory. The
polarization vectors for major phonons connecting symmetry points in conduction
bands (X or L) and valence band (Γ) of diamond lattice structure are listed in table 4.1.
In diamond structure, phonon modes are either pure longitudinal or transverse. The
phase difference between two atoms in primitive cell, gives the fraction of the mode
which is “acoustic” and “optic”. Actual displacement of the particles can be obtained
by multiplying the phase term, ( )exp li τβi to the polarization vectors.
81
Direction (β)
Mode ∆ = (0,0,q) Λ = (q,q,q)
LA (0,0,1), (0,0,1) (1, 1, 1)/ 3 , 5 (1, 1, 1)exp4 3
i π⎛ ⎞⎜ ⎟⎝ ⎠
LO (0,0,1), (0,0,-1) (1, 1, 1)/ 3 , (1, 1, 1)exp
4 3iπ⎛ ⎞
⎜ ⎟⎝ ⎠
TA
(1, -1, 0)/ 2 , ( , - , 0)/ 2i i (1, -1, 0)/ 2 , (1, -1, 0)exp
4 2iπ⎛ ⎞
⎜ ⎟⎝ ⎠
(1, 1, 0)/ 2 , (- , - , 0)/ 2i i (1, 1, -2)/ 6 , (1, 1, -2)exp
4 6iπ⎛ ⎞
⎜ ⎟⎝ ⎠
TO
(1, -1, 0)/ 2 , ( - , , 0)/ 2i i (1, -1, 0)/ 2 , 5 (1, -1, 0)exp4 2
i π⎛ ⎞⎜ ⎟⎝ ⎠
(1, 1, 0)/ 2 , ( , , 0)/ 2i i (1, 1, -2)/ 6 , 5 (1, 1, -2)exp4 6
i π⎛ ⎞⎜ ⎟⎝ ⎠
Table 4.1 : Polarization vectors of major phonon modes in diamond lattice structure. Since the diamond structure has a primitive cell formed by two-atom basis, each phonon consists of two polarization vector. While longitudinal mode consists of one LA and one LO polarizations, the transverse modes (TA and TO) are double degenerate.
82
4.4 Silicon.
The transition with highest probability during band-to-band tunneling process in
silicon occurs between ΔC-Γ25’, where the bandgap is 1.12eV. The second lowest point
of conduction band is located at L1 point. Although the transition from L1 to Γ25’ is
also possible, the large bandgap between the two points of 1.8eV, significantly lowers
the tunneling probability compared to the ΔC-Γ25’ transitions.
Although experimental and theoretical studies to estimate the intervalley deformation
potentials exist for inner-band transitions [36-38, 51] (conduction to conduction or
valence to valence), photoluminescence and photon absorptions [52] , the strength of
Figure 4.1: Bandstructure of silicon calculated by EPM. Phonon-assistedtransitions with highest probability between band extrema are shown.In silicon, transitions between Δ1-Γ25’ are most dominant. Due toselection rules imposed by symmetry of the two points, TA and TOphonons are most effective in assisting the tunneling process.
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
Ener
gy (e
V)
1002aπ⎛ ⎞
⎜ ⎟⎝ ⎠
k 1113aπ⎛ ⎞
⎜ ⎟⎜ ⎟⎝ ⎠
k
X1
L1
Γ15
Γ25’
Δ1
Δ5 Δ2’
Λ3
Λ1
L3’
X4
Δ2’ Λ1
TA/TO
83
phonon-electron interaction during band-to-band tunneling process is not well-studied
[39].
According to Lax [46], the selection rules allow only TA, TO and LO phonons to be
effective in assisting the intervalley transition between Δ1-Γ25’.
Γ′ Δ Δ Δ TA, TO LO (4.4)
The deformation potentials for intervalley scattering between two symmetry points,
Δ1-Γ25’ are calculated and listed in table 4.2.
Silicon
Mode (001) Phonon Energya (meV) Deformation Potentials (eV/Å)
LA 46.3 0.0
LO 54.3 2.6
TA 18.4 5.6
TO 57.6 2.9
Table 4.2 : The effective deformation potentials and phonon energies for intervalley scattering between Γ25’ and ΔC ( 2 / (0,0,0.85)laπ ). Energy difference between two states is 1.12eV. Since there is degeneracy in Γ25’ states and transverse modes, the deformation potentials were added to indicate the effective strength as follows: 2 2
,( ) ( )l
pp l
DK DK= ∑ , where p denotes the polarization of phonon and l
is indices for degenerate valence bands. a : Chynoweth [53]
84
The deformation potentials calculated using EPM, reconfirm the Lax’ selection rules
in that LA phonon does not participate in the scattering events, but TA, TO and LO
phonons can couple the two states. The values we obtained are also in reasonable
agreement with the values used in previous BTBT works. Rivas[54] guessed TA
deformation potential to be 2.45 eV/Å and Tanaka [55] used 6 eV/Å for overall
effective deformation potential. Although the selectrion rules and theoretical
calculation of deformation potentials between the symmetry points predict strong
participation of LO phonon in scattering process, experimental results showed
otherwise.
Chynoweth [53] and Holonyak [56] conducted experiments on silicon Esaki junction
and observed that only TA and TO phonons significantly contribute to a tunneling
current, but contrary to the selection rules, contribution from LO phonon contribution
is weak. The reason for the weak response of LO mode phonon has not been clearly
studied yet. The nearly equal contributions from TA and TO phonons imply that the
phonon interaction strengths for these phonon modes are roughly same.
The limitation of the previous theoretical studies lies in the fact that they assumed
that in the moment of scattering the state of electron is still same as the one where it
started to tunnel through the classical turning point. As we have shown in the chapter
on BTBT modeling, the most fraction of transition occurs at the point where
( )c vimag k k+ is minimized. In silicon, the transition point is at -4.67 eV (or 0.61 eV
above the valence band edge).
85
As seen in Fig.4.2 we have calculated the deformation potentials using the unit-cell
function ku at the transition point where crystal momentum k is imaginary, instead of
evaluating the phonon interaction between the symmetry in real bandstructure. The
results are listed in table 4.3.
(a) (b)
Figure 4.2: Comparison between the Lax’ Selection Rules and our work. Sincemost of contribution in tunneling is coming at E=-4.67eV, thedeformation potentials should be calculated between two pointsdepicted in (b) not between two symmetry points in (a).
Conduction Band
Valence Band
Energy
Complex k‐+
Real k
Γc
Γv -
∆
Conduction Band
Real k
-
Lax, Selection Rules
ConductionBand
Valence Band
Energy
Complex k‐+
Real k
Γc
Γv
-∆
Conduction Band
Real k
--
Phonon
This work, E=‐4.67eV
86
The complex-k-space-intervalley-deformation potentials in table 4.3 deviate from the
values obtained in real k space. They clearly show that, while the contribution from
both LA and LO modes are negligible, TA and TO modes are most dominant, which is
in very good agreement with experimental results by Chynoweth [53] and Holonyak
[56].
Due to lack of theoretical and experimental data, previous BTBT calculations in
electronic devices were based on “educational guess [39]” mostly estimated from
either dilation deformation potentials or mobility deformation potentials. While Rivas
[54] took into account only TA and TO phonon modes in his tunneling Esaki diode
simulations, Fischetti [39] added all four phonon modes without considering the
selection rules and assuming that they have almost equal deformation potentials.
Silicon (Γ-∆)
Mode (meV) This work (eV/Å) Previous works (eV/Å)
LA (46.3) 0.28 weakC
LO (54.3) 0.19 weakC
TA (18.4) 6.93 2.45R, 6T, 2.74S, 5.9F, strongC
TO (57.6) 4.42 5.6R, 5.23S, 4.06Ge, 4.96Gp,6.0F, strongC
Table 4.3 : The complex-k-space-intervalley-deformation potentials calculated for Band-to-Band tunneling between Γ25’ and ΔC. Due to lack of theoretical work, previous BTBT models (R,T,F,S) estimated the deformation potentials from dilation deformation potentials and mobility deformation potentials. Experimental values are extracted from light absorption coefficient study (Ge,Gp) and Esaki tunneling device (C). R: Rivas [54], T: Tanaka [55], F: Fischetti [39], S: Schenk [57] Ge: Glembocki (electron-phonon) [36], Gp: Glembocki (hole-phonon) [36]
87
Glembocki [36] experimentally measured phonon-assisted photon absorption length
and extracted the deformation potential for TO mode. The deformation potential
obtained by Glembocki is for electron 4.06 and for hole 4.96 eV/Å, which are in
astonishingly good agreement with our potentials 4.42 eV/Å.
4.5 Germanium
Germanium has direct bandgap slightly larger (0.14eV larger) than indirect bandgap
(Lc1-Γ25’). Since, direct BTBT process is more than an order efficient, phonon-assisted
tunneling is only significant for a small electric field. The transition with highest
probability during phonon-assisted-band-to-band tunneling process in Germanium
occurs between Lc1-Γ25’, where the bandgap is 0.66eV. The phonon-assisted
transitions between ∆c1-Γ25’ are also possible, they are negligible compared to the
strong direct tunneling transitions. Although for relaxed germanium the transitions
between ∆c1-Γ25’ are not strong, in strained germanium (ex. pseudomorphically grown
germanium on silicon) ∆c1-Γ25’ forms lowest bandgap and the transitions through this
path becomes most dominant process.
88
According to Lax [46], the selection rules allow only LA, TA and LO phonons to be
effective in assisting the intervalley transition between Lc1-Γ25’.
L Γ L L TA LO
L Γ L LA (4.5)
TA and LO phonons couple valence band to conduction band by first order
perturbation. LA phonon couples two bands by second order perturbation. Since Γ
valley is close to L valley, the electron tunneling from v25+Γ is mixed to c2
−Γ and then
LA assisted tunneling takes place from c2−Γ to c1L+ [42-43].
Figure 4.3: Bandstructure of germanium calculated by EPM. Phonon-assistedtransitions with highest probability between band extremum areshown. In germanium, transitions between L1-Γ25’ are most dominant.Experimentally LA, TA and TO phonons are most effective inassisting the tunneling process. Subindex c in Lc1 representsconduction band while v represents valence band.
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1-7
-6.5
-6
-5.5
-5
-4.5
-4
-3.5
-3
-2.5
-2
LA/TA/TO
1002aπ⎛ ⎞
⎜ ⎟⎝ ⎠
k1113aπ⎛ ⎞
⎜ ⎟⎜ ⎟⎝ ⎠
k
L1
Λ3
Λ1L3’
Λ1 X1
X4
Energy (eV)
TA/TO Γ15
Γ25’
Δ1
Δ5
Δ2’
Δ2’
89
The deformation potentials for intervalley scattering between two symmetry points,
Lc1-Γ25’ (for LA phonon, between c2−Γ - c1L+ ) are calculated and listed in table 4.4.
Lax’ selection rules and the deformation potentials between symmetry points obtained
by EPM methods indicate that the contribution of LO phonons is comparable to LA or
TA phonons and the TO phonons do not participate in the tunneling process. However,
experiments on germanium Esaki diode performed by Chynoweth [53] show that,
while the transitions by LA and TA phonons are strongest, TO phonon interaction is
substantially large, the contribution from LO phonon is negligibly small.
Germanium
Mode (111) Phonon Energya (meV) Deformation Potentials (eV/Å)
LA 27.5 2.5
LO 31.6 4.5
TA 7.6 3.0
TO 36.0 0.0
Table 4.4 : The effective deformation potentials and phonon energies for intervalley scattering between Γ25’ and LC ( / (1,1,1)laπ ) (for LA phonon, between c2
−Γ - c1L+ ). Energy difference between two states is 1.12eV. Since there is degeneracy in Γ25’ states and transverse modes, the deformation potentials were added to indicate the effective strength as follows: 2 2
,( ) ( )l
pp l
DK DK= ∑ , where p denotes the
polarization of phonon and l is indices for degenerate valence bands. a : Chynoweth [53]
90
The discrepancy in TO and LO phonon interaction strength between the selection
rules and the experiments can be again explained by introducing complex-k space. As
we stated previously the most fraction of transition occurs at the point where
( )c vimag k k+ is minimized. In germanium, the transition point is around at -4.12 eV
Germanium
ModeC,W (meV) This work (eV/Å) Previous works (eV/Å)
LA (Γ-L) (27.5) 5.4 3.89K, 3.88G, 4.2Z, 4.5L, strongC
LO (Γ-L) (31.6) 0.22 very weakC
TA (Γ-L) (7.6) 1.6 1.42M, strongC
TO (Γ-L) (36.0) 1.8 mediumC
LA (Γ-∆) (25.6) 0.77 -
LO (Γ-∆) (31.7) 1.6 2.2L, 2.43K
TA (Γ-∆) (10.1) 0.15 -
TO (Γ-∆) (34.0) 0 -
Table 4.5 : The complex-k-space-intervalley-deformation potentials calculated for Band-to-Band tunneling between Γ25’/ c2
−Γ and c1L+ /ΔC. (100) at E=-4.12eV.
K: Krishnamurphy [51], ab initio tight-binding method (conduction band to conduction band scattering), G: Glembocki [37], rigid-ion model (conduction band to conduction band),
Z: Zhou [48], time-resolved photoluminescence method, L: Li [52], direct exciton optical absorption measurement, M: Kane[41], Macfarlane [47], optical absorption measurement,
C: Chynoweth [53], Esaki tunneling diode,
91
(or 0.54 eV above the valence band edge). The deformation potentials at the transition
point are listed in table 4.5.
The complex-k-space-intervalley-deformation potentials clearly shows prominent
LA(Γ-L) and TA(Γ-L) phonon interactions and vanishingly small contribution from
LO(Γ-L) phonon which is in very good agreement with Chynoweth [53]. Phonon
interaction strength of TO (Γ-L) phonon, TO( )LΓ−M is about ¼ of TA( )LΓ−M , which
successfully explain the reason for medium peak strength of TO (Γ-L) phonon [53].
The calculated deformation potentials for LA (Γ-L), TA (Γ-L) and LO (Γ-∆) are all
consistent with the values determined either by experiments[41, 47-48, 52] or
theory[37, 51].
4.6 Summary
The deformation potentials for phonon-assisted band-to-band tunneling have been
successfully calculated using rigid-ion model and empirical pseudo-potential method.
The previous theoretical calculations were all limited to inner-band scattering (ex.
conduction band to conduction band, valence band to valence band) or limited to
transitions between symmetric points in real k-space. By introducing complex-k space
in calculations of BTBT deformation potentials, we are able to explain the
discrepancies between previous theories to experimental results on tunneling diodes.
In relaxed silicon, only TA(Γ-∆) and TO(Γ-∆) phonons have significant contributions
92
on band-to-band tunneling and in relaxed germanium, LA(Γ-L), TA(Γ-L), TO(Γ-L)
phonons are participating tunneling process.
93
Chapter 5 Application of band-to-band
tunneling (BTBT) model
5.1 Introduction
It becomes important to predict the band-to-band tunneling leakage of devices made
with high-mobility materials. The most commonly used BTBT models in commercial
TCAD tools (MINIMOSTM, TAURUSTM, DESSISTM) do not capture quantization
effects, phonon-electron interactions and full band information of nano-scale devices.
The model developed in the last chapter captures important quantum mechanical (QM)
effects in highly scaled DGFETs. We will compare the previous models and our
model and discuss the applicability of this model for ultra thin body DGFETs with
high mobility channel.
5.2 Comparison to Previous Models
5.2.1 Direct Band
Mostly commonly used model is Kane’s model [40]. The model uses 2-band
approximation ( Fig. 5.1 (a) ) where it assumes only two bands exist (one conduction
band and one valence band) and interaction of bands is modeled using kÿp
approximation. This model works pretty well for relaxed small bandgap
semiconductors such as InSb. In InSb, the separation (∆so) between heavy hole (HH)
and split off bands is large enough compared to bandgap (Eg) – for InSb ∆so/Eg≈4.7.
Large separation ensures minimum interaction between these bands, which makes the
94
2-band based Kane’s model justifiable. However as we expands our interest of
material choice to large band-gap materials such as Si, InP and GaAs, this
approximation becomes inaccurate. Krieger [35, 59] expanded Kane’s model by using
Kane’s 4-band kÿp approximation, suggesting new analytical expression for effective
tunneling mass. This model also imposes limitation on applicability to semiconductors
such as Ge and Si, since Kane’s 4-band model assumes heavy hole band is isotropic.
In these semiconductors (Ge and Si), heavy hole band is severely warped, therefore
using analytic kÿp approximation leads to again inaccurate estimation of BTBT.
(a) (b)
Fig. 5.1: (a) Kane’s 2-band model. (b) Full band model obtained by EPM.
As described in the previous chapter, we obtain the E-k dispersion in bandgap
(complex band structure) by empirical pseudopotential method (EPM), where entire
128 bands are included (Fig. 5.1 (b)). We found that using Kane’s 2-band model leads
Kane : 2-band Approximation
Conduction Band
One Valence Band
Forbidden Bandgap
Complex kReal k
Conduction Band
Three Valence Bands
HH
LHSO
Entire 128-bands obtained by EPM
Γc
Complex k
95
to 10 times smaller BTBT rate for 1MV/cm electric field and 300 times smaller BTBT
for 0.5MV/cm electric field.
5.2.2 Indirect Band
Most commonly used indirect tunneling model is Keldysh’s model [45]. In his
formulation, the author used band-edge effective mass and assumed this mass does not
change throughout the tunneling process. This assumption is justifiable for Si, where
transition energies between conduction band and valence band for Γ point and for Δ
points are pretty big (>3eV). In Si, since the energy transition an electron experiences
during tunneling process is relatively small compared to these transition energies,
using simple effective mass’ at band minima and maxima is a good approximation.
However, in Ge, proximately located Γc interacts strongly with Γv, bending the
dispersion curve steeper, thus substantially lowering tunneling mass [60].
96
Again as in direct tunneling case, precise modeling of band-dispersion is necessary.
We used EPM to numerically calculate the dispersion as discussed in the previous
chapters.
5.2.3 Selection Rules on Phonon modes and Phononinteraction
Strength
During indirect tunneling process, only phonons with certain symmetry are allowed
to couple the conduction band to the valence band. Selection rules concerns if the
phonon-interaction strength, Mj,phV-1 vanishes or not. Previous works on indirect
tunneling by Kane [41] and Keldysh [45] does not distinguish between different
phonon modes. Although Fischetti [39] tries to take into account various phonon
Figure 5.2: Indirect tunneling process. Keldysh’s model does not capture full band-information.
Valence Band
Energy
Complex k-+
Tunneling Process
Real k Γv
L or ∆
Conduction Band
Real k
--
-
Phononunknown-
Effective Mass Approx.
Γc band is ignored
Effective Mass Approx.
97
modes, he failed to consider selection rules and arbitrarily assumed all four phonon
modes (transverse acoustic (TA), transverse optic (TO), longitudinal acoustic (LA)
and longitudinal optic (LO)) contribute to BTBT current with almost equal strength. In
our work we use the electron orbital wavefunction in decaying states (or inside
bandgap) in calculating the interaction strengths, unlike all the previous studies where
phonon elements were calculated between the symmetry points in real bandstructure,
and show that treating the phonon-electron interaction in complex band structure
successfully explain experiments. As explained in the chapter on phonon By adopting
the rigid-ion model [49]and using the empirical pseudo-potential method, for the first
time we accurately treated the electron-phonon interactions during band-to-band
tunneling process.
5.3 BTBT Generation Rates for Various Materials
Since BTBT process requires the movement of electron in the bandgap, it is
important to predict the E-k relationship in the bandgap where k vector becomes
complex number. Local empirical pseudopotential method (LEPM) is used to obtain
complex bandstructure. We used simplex search method to fit a model potential to
experimental band parameters and exact effective mass at bandedge. Fig. 5.3 shows
the complex bandgap structures of Ge and GaAs.
98
(a) (b)
Fig. 5.3: (a) 3D plot of complex bandstructure of germanium along the [111] direction
(b) GaAs.
BTBT generation rate vs. electric field is calculated for various materials and
the results are shown in Fig. 5.4. Generally, the wider bandgap materials show smaller
BTBT current. Direct tunneling is dominant in Ge due to the proximity of the Γ-valley
to the L-valley. Si exhibits very low generation rate, due to its indirect bandgap.
Although the direct bandgap material, GaAs has a bigger bandgap than Si, the BTBT
rate of GaAs is two orders of magnitude higher than Si. Since InAs has smaller
bandgap and mass than Ge, BTBT generation rate for InAs is the highest among four
materials.
-0.2-0.15
-0.1 -0.05
-0.200.20.40.60.8
‐3
‐3.5
‐4
‐4.5
‐5
‐5.5
00.8 0.4
Energy (eV)
0.1 00.2
0
‐3
‐6
‐4
‐5
0.5 0
0.4
Energy (eV)
10 0.2
99
Fig. 5.4: BTBT generation rate vs. Electric Field
5.4 PIN Diodes
To verify our model, we compared our simulated current for PIN diode to fabricated
PIN diode’s IV data available on literature.
To validate our model, we collected existing experimental data on tunnel diodes and
compared it to the theoretically calculated values. Kane [40-41] and Hurkx [61]
showed that BTBTG can be expressed as 0exp( / )AF F Fσ − . A and 0F are constants
related to bandstructure and phonon-interactions of the materials, σ is 2 for direct
tunneling, 2.5 for phonon-assisted tunneling and F is an average electric field across
tunneling path. Although most of existing data on BTBT is very out dated and hard to
interpret, extracting F0 is possible. Table 1 shows the theoretically calculated 0F from
0 0.5 1 1.5 21010
1015
1020
1025
1030
Electric Field (MV/cm)
BTB
T G
ener
atio
n R
ate
(cm
-3s-1
)
SiGaAsGeInAs
100
our BTBT model explained in the chapter on BTBT, Kane’s two band model and
experimentally extracted values. Due to nonparabolicity of effective mass in bandgap,
0F predicted by our model is lower than the ones by Kane’s model and is closer to the
experimental values.
The experimental values of 0F in Si vary from 19 to 35 MV/cm. When applied bias
is low in a tunnel diode, the local E-field based model inevitably gives wrong results,
since it doesn’t account for the non-uniform E-field effect.
There are a few literature works on fabricated PIN structure where the accurate
information on doping profile is also available. Tyagi [62, 69] and Agarwal [70]
fabricated Si PIN diodes and measured the I-V characteristics in reverse bias region.
In Fig. 5.5, the I-V curves calculated using our BTBT model and Hurkx model [61]
are depicted together with experimental data.
Fo Mat.
Calc. (MV/cm) Exp. (MV/cm) Kane (MV/cm)
Si [100] 21.6 19~35[59, 61-64] 24
GaAs [111] 16.6 14.6[65] 17.5
Ge [111] Direct 5.3 5.7[66] 5.9
Indirect 4.6 4.6[67] 6.1
InAs[111] 1.3 1.3[68] 1.3
Table 5.1 : Parameter F0 of Kane/Hurkx model AFσexp(-F0/F), extracted by our BTBT model (by LEPM), Experiments and Kane(two band model fordirect and effective mass approximation (EPM) for indirect). Thedirection indicates tunnel direction.
101
Since Hurkx fitted his model to experimental data for high voltage bias, his model
matches with experiments pretty well at high bias region, but at low bias it largely
overestimates currents. At very low bias, defect assisted tunneling becomes more
dominant than BTBT [57], our model starts to underestimate the current in this region
of bias. In actual device, as tunneling distance increases, the trailing wavefunction
starts to scatter and lose quantum coherence. As we assume complete quantum
coherence throughout tunneling process, our model results in slight overestimation in
lightly doped diodes.
Gaul et al. [65] fabricated very abrupt junction GaAs PIN diode by MBE and their
data matches very well with our prediction. Fig 5.6 shows the I-V of PIN diode and
theoretical values given by our model and Kane’s BTBT model.
Figure 5.5 I-V curves of silicon PIN diodes. Solid lines with markers are actualI-V characteristics of the diodes, solid lines are estimations from ourBTBT model, and dotted lines are currents calculated by Hurkx’model.
0.5 1 1.5 2 2.5 3 3.5 4 4.5
10 -5
10 0
R e v e rs e B ia s (V )
J (A
cm-2
)
T h is w o rkH u rkxE xp .
Ty ag i N o1
Ty ag i N o2
A garw a l 33nm
102
Our calculated values closely match with experimental data while kane’s model
always underestimates by big amount due to inaccurate estimation of tunneling mass
using two-band model. The only deviation happens in device where I-region is 55nm
long. As in the case of Si, tunneling electrons in 55nm device are scattered and lose
quantum coherence during tunneling process.
Tyagi [66] and Tokuyama [71] fabricated abrupt junction PN diodes on germanium
wafer and measured the reverse bias breakdown currents. The estimations by our
model and Kane’s model are drawn in Fig. 5.7 and are compared with experimental I-
V characteristics. As in Si and GaAs, our model accurately predicts I-V characteristics
of Ge PN diode.
Figure 5.6 I-V curves of GaAs PIN diodes. Solid lines with markers are actual I-V characteristics of the diodes, solid lines are estimations from ourBTBT model, and dotted lines are currents calculated by Kane’smodel. The length of I-region is varied from 15nm to 55nm.
1 2 3 4 5 6
10-5
100
Reverse Bias (V)
J (A
cm-2
)
This workKaneExp.
15nm
20nm
55nm
103
5.5 Effect of Strain (Ge)
In Fig. 5.8, we show the energy shifts at symmetry points obtained from our LEPM
calculations in biaxially strained germanium (001). The strain is indicated by the ratio
0/ 1a a − . Since all three important symmetry points ( CΓ , 100Δ , CL ) form conduction
band minima at similar energy levels, strain can effectively deform and change the
electronic properties of germanium. The results show that compressive strain lifts up
CΓ valley and two-fold 001Δ valley, while it pushes down four-fold conduction band
100/010Δ valleys. When strong enough compressive strain is applied ( 0/ 1 0.02a a − < − ),
energy level of 100Δ valley comes down below CL valley leading to forming indirect
bandgap at 100Δ point. Tensile strain pushed CΓ further down and around at
Figure 5.7 I-V curves of Ge PIN diodes. Solid lines with markers are actual I-Vcharacteristics of the diodes, solid lines are estimations from ourBTBT model, and dotted lines are currents calculated by Kane’smodel.
1 1.5 2 2.5 3 3.5 4 4.5 510-6
10-4
10-2
100
102
Reverse Bias (V)
J (A
cm-2
)
This workKaneExp.
TokuyamaNo.12
TyagiNo.2
TyagiNo.1
104
0/ 1 0.025a a − = the strained germanium becomes direct bandgap materials. The
large separation between 1vΓ and 2vΓ is also observed due to breaking of symmetry
and degeneracy in diamond lattice structure. For both tensile strain and compressive
strain, the combined effect of the band shifts in conduction band and valence band
leads to shrinkage in bandgap
This change in bandgap and energy levels of symmetry points alters band-to-band
tunneling by large amount. In relaxed germanium where high electric field is applied,
major band-to- band tunneling component is direct tunneling through vΓ to CΓ . When
the small amount of biaxial-compressive strain is applied, band-to-band tunneling rate
goes down due to increased direct band-gap. Appling further strain reverses the trend
Figure 5.8 Calculated band shifts of germanium (001) as a function of in-plane
biaxial strain ( 0/ 1a a − ).
-0.04 -0.02 0 0.02 0.04-5
-4.8
-4.6
-4.4
-4.2
-4
-3.8
-3.6
-3.4
Strain (all/a0-1)
Ener
gy (e
V)
ΓC
ΓV2 ΓV1
Δ100
Δ001
L6,C
105
and brings up the band-to-band tunneling rate again, since the indirect tunneling
component starts to increase and become bigger than direct tunneling component. This
trend implies that current strain technology to boost a hole current in germanium
MOSFET would reach the barrier imposed by the enhanced leakage current.
Since our BTBT model is based on EPM calculations of complex bandstructures, it
can take into account any strain effects, while previous model which is solely based on
simple two band model fail to account for the effects of strains on BTBT. After
obtaining complex band structure information and phonon-electron matrix elements of
each strain level, we have simulated the I-V characteristics of strained germanium
double gate MOSFET device of which gate length is 15nm and oxide thickness is
0.9nm. As seen in figure 5.9, we found that 60% (out of max. 0.0410 strain) biaxially
compressive-strained Ge has the lowest band-to-band tunneling leakage.
Figure 5.9: IOFF,MIN vs. biaxial strain. Compressive strain increases indirect BTBTwhile it decreases direct BTBT. Optimum point at around x=0.6.
0 0.4 0.6 0.8 1 0.210
-9
10-8
10-7
xI O
FF,M
IN (A
/um
)Lg=50nm, Tox=0.9nm, Tb = 5nm
ss--Ge grown on SiGe grown on Si11--xxGeGexx
Direct TunnelDirect TunnelIndirect Indirect TunnelTunnel
Lowest BTBTLowest BTBT
s-Ge
106
5.6 Quantum Confinement
In highly scaled MOSFETs built on ultra thin bodies such as DGFETs or ultra-thin
silicon on insulator (UT-SOI), the bandstructure is also heavily altered by quantum
confinement effect. Figure 5.10 illustrates the complex bandstructure of GaAs (001)
quantum well (QW) which consists of 12 atomic layers (atm) of GaAs sandwiched by
two 6 atomic layers of vacuum. The bandstructure of QWs shows projections of bands
with out of plane k-component fall onto in-plane k-space. As each subband exhibits
different tunneling mass in QWs, for proper BTBT modeling it is required to calculate
the curvature of complex bandstructure of each subband.
Using the complex bandstructure of QWs, we have shown for the first time that
quantum confinement effect suppresses band to band tunneling. As seen in figure 5.11,
the tunneling wavefunctions in ultra thin body, experience exponentially bigger
(a) (b)
Figure 5.10: (a) A GaAs QW consists of 12 atom thick layer with 6 atom thickvacuum layer. (b) Complex Bandstructure of the GaAs QW.
GaAs (001)
Vacuum
Vacuum
12 atoms
6 atoms
6 atoms
Im(k) Γ Re(k) L Im(k)
ΓC
ΓV
LC
107
attenuations due to increased bandgap and tunneling mass which leads to reduced
band-to-band tunneling. To take into account the quantization effect caused by ultra-
thin body during device simulations, the wavefunctions and the energy levels of
quantized subband states were obtained by solving 1-D Schrödinger equation along z-
direction for both electrons and holes. By plugging the subband information into the 2-
D quantized BTBT model developed in the chapter on BTBT, it is possible to
calculate BTBT leakage currents in thin body devices [23-33, 72-74].
We have applied our model to calculate the BTBT leakage current in ultra thin film
(3~10nm) DGFETs with high mobility channel materials such as Ge, 4% biaxially
Figure 5.11:Ultra-thin body and larger quantization increases the effective bandgap and lowers the tunneling rate.
ΔETb
S
D
GBTBT
Small Tunneling Rate
ΨeΨh
Eg
Large Tunneling BarrierStrong Quantization
S
D
GBTBT
Large Tunneling Rate
ΨeΨh
Eg
Eg
Thick BodyDGFET
Thin Body DGFET
Tb
ΔE
Channel | OxideOxide |
108
compressive-strained Ge, Si, 4% biaxially compressive-strained Si, InAs, GaAs.
Through these simulations we found that reducing body thickness is very effective in
reducing the BTBT current. However it is not easy to manufacture thin body DGFETs
because of process variations, technological complexity and increased cost. Larger
VDD increases the drive current but results in a worse IOFF,MIN. To evaluate the
practicability of various materials based on BTBT leakage current at the 15nm
technology node, we plot the maximum VDD versus Tb curve (Fig. 5.12). The figure
shows the maximum allowable VDD at a given Tb for a constant IOFF,MIN of 0.1uA/um
(iso-leakage contours). The combination of VDD and Tb below the curve satisfies the
ITRS HP requirement for the 15nm node Off-current specification. The result shows
that 100% s-Si cannot be used unless it is operated under 0.5V, while Ge and InAs can
be operated at 0.9V as long as body is thinner than 5nm. Si and GaAs can be operated
over 1.5V due to their large bandgap. Although s-Ge has a smaller bandgap than Ge,
s-Ge shows a much lower BTBT leakage current than Ge due to its large quantization
and indirect bandgap.
109
The effect of quantum confinement on BTBT leakage current in MOSFET devices
predicted by theory is proved by experiments [32]. Strained-SiGe MOSFETs having
ultra-thin (UT) (TGe<3nm), very high germanium fraction (~80%) channel and Si cap
(TSi cap<3nm) have been successfully fabricated on relaxed Si substrates (Fig. 5.13). In
this device, big band-offset (>0.7eV) between germanium and silicon layer creates
quantum well for holes. The quantum confined holes in germanium layer are
quantized and the energy levels are shifted up.
Figure 5.12: Iso-leakage (IOFF,MIN) Contours. These curves determine the
maximum allowable VDD to meet the ITRS off current requirement for
a 15nm gate length, at a given body thickness (constrained by
technology). s-Si cannot be used for >0.5V operation. Ge and InAs
can be used for 0.9 operation when Tbody<5nm. For Tbody<7nm s-
Ge can be used for >1V operation. Si and GaAs comfortably meet the
off-current specification because of their large bandgap.
3 5 7 100
0.3
0.6
0.9
1.2
1.5
1.8
Tbody (nm)
VDD
(V)
SiGaAss-GeGeInAss-Si
Towards Decreasing
Ioff,MIN
110
Although predicting the BTBT current in complicated device structure is very
challenging work due to lack of accurate knowledge on dopant profile, it is still
possible to predict the trend in change of I-V characteristics as we vary design specs.
In this experiment, thickness of germanium layer is varied from 6nm to 2nm and we
observed over 10x BTBT reduction in 2nm strained Ge over 4nm strained Ge
(Fig.5.14). As illustrated in figure 5.14 this trend is also well predicted by simulations
using our BTBT model. Reducing the thickness of germanium layer imposes three-
fold effects on BTBT current. First, reducing the thickness lowers the BTBT current
linearly with respect to reduced cross-section area. Second, parallel electric field
applied across the germanium layer is increased, which results in higher BTBT current.
Third, increased quantum confinement effect widens up effective bandgap accordingly
the BTBT current is also reduced. Our simulations support the experimental results in
that despite of the enhanced parallel electric field, quantum confinement effect is
Figure 5.13: Device structure and band diagrams of SiGe heterostructure p-MOSFET.
Cross-sectionDevice geometry
Quantum Confinement
111
strong enough to bring down the BTBT current level when germanium layer is thinned
down below 2nm.
5.7 Summary
We have shown the limitations of previous BTBT tunneling models for TCAD
simulators. Since previous BTBT tunneling models are based on the two-band model
in relaxed bulk semiconductor system, it fails to predict BTBT current for highly
strained ultra thin film devices. Moreover the phonon-electron interactions were not
properly modeled in any of previous works. We have explained in this chapter the
difference in modeling direct tunneling (zero-phonon assisted) and indirect tunneling
(phonon-assisted). Depending on transition paths we have carefully chosen phonon
modes according to selection rules and deformation potentials calculated by applying
Figure 5.14:BTBT (GIDL) with varying TGe channel. Left plot is obtained from experiments and right plot is drawn using simulation results.
Experimental Id-Vg10x Reduction (reducing Tge)
Simulation Id-Vg (DGFET)
112
rigid-ion model on electron orbital in complex-k space. Our model was verified using
various experimental data on fabricated PIN structures on silicon, germanium and
GaAs wafers. We have shown for the first time that biaxial compressive-strain reduces
the direct tunneling component, but the increases in-direct tunneling component in
leakage current in germanium devices. With 2.4% biaxial compressive-strain applied
across germanium devices, the BTBT leakage current reaches the minimum point. By
applying our BTBT model in DGFET simulations we found that reducing the
thickness of channel layer is very effective in suppressing the BTBT leakage current.
This quantum confinement effect has been verified by comparing the simulation
results to the I-V characteristics of fabricated thin-film strained germanium MOSFET
devices.
.
113
Chapter 6 Quantum-Ballistic Transport in 1-D
Quantum Confined System
6.1 Introduction
According to the 2008 update of ITRS [75], MOSFETs with effective channel length
of 20 nm will enter large scale production in 2012, and MOSFETs with effective
channel length of 8.1 nm in 2022, at the end of the Roadmap. Such devices exhibits
several characteristics in nanometer scale: quantum confinement in the channel,
source-to-drain tunnel current, band-to-band tunnel current, and non-equilibrium
transport. Already in MOSFETs fabricated by the current technology node, the most
of fraction of electrons contributing to the drain current (in an n-MOSFET) go through
the channel without suffering from inelastic scattering. The ballistic fraction is
increasing with further scaling down, and it is predicted by some authors to be
predominant over the fraction of electrons experiencing inelastic scattering in devices
with channel length shorter than 20 nm [76]. Natori [77] has developed the first
analytical ballistic model for MOSFETs and Monte Carlo tools [76] have been
developed and used to treat carrier transports in nano-scale devices.
Nanoscale MOSFETs exhibits a significantly large quantum confinement effect in the
channel, due to the high vertical electric field and the thin channel thickness. A
quantum simulation is therefore necessary to take into account the quantized
subbands, the quantum gate capacitance, 2-D density of states and degeneracy lift in
114
conduction band of semiconductors [34]. Although various authors have shown works
on developing quantum transport simulations [78-79], they mostly ignore the non-
parabolicity of conduction bands. In very highly scaled devices, the quantum
quantized subband energies are located far above the band minima where the effective
masses obtained at the minima points are not valid anymore. In this chapter, we will
develop a quantum ballistic transport model in 1-D quantum confined double gate
FET (DGFETs) structures which takes in to account non-parabolicity of conduction
bands.
6.2 1-D Quantum Confined System
In the BTBT chapter, we have shown that if quantum confinement is strong only in
one direction (z direction) the states are written as a sum of 2-dimensional subbands,
the edges of which are obtained by solving the Schrodinger equation in the z-direction
for each mesh point along the x axis. For materials with degenerate minima in the
conduction band, the states can be computed with the directionally decoupled effective
mass approximation [44] for each minimum, taking into account mass anisotropy and
non-parabolicity. The same procedure is applied to the valence band maxima.
115
We recall the equations derived in the BTBT chapter. (Eq. 1.53 to 1.69). Carrier
wavefunctions are decoupled using subband decomposition and take the form:
Ψ , , ; exp , (6.1)
where , n and represent degenerate valleys, quantum number for subbands,
thickness of the channel, respectively. The wave-functions , ; are normalized
in z direction and unit-cell functions , are normalized in the space. 1D
Schrodinger equation in the z confined direction provides eigen-functions , ;
and eigen-energies , .
Figure 6.1: Schematic view of the thin channel double-gate MOSFET (DGFET).Thickness of the channel is . Current flows along x-direction and the channel is quantized along the z-direction.
ChannelInsulator
Gate
Gate
xz
y
xz=0 z=tb
Eg
Insulator
Insulator
Energy
0
,
,
,
,
, ;
, ;
, ;
, ;
, ,
,
,
116
21
,, ;
, , ; , , ; (6.2)
where , is the bandedge energy level of band . Under the assumptions that
subband wavefunctions , ; have very weak dependency on x and inter-subband
couplings are also very small due to the strong quantum confinement along z-direction,
the wave functions , satisfy the following 1D Schrodinger equation in x
direction (or channel direction).
21
′ ′
′ ,′
(6.3)
where coefficients ′ denote the inter-band couplings which represent the
BTBT.
6.3 1-D Schrodinger Equation with non-parabolic effective
mass
The electron levels in 1-D quantized Double Gate FETs with small effective mass like
in IIIV materials can be several tenths of an electron volt above the bulk-conduction
band edge. For that range of quantization, without considering non-parabolicity of
bandstructure, errors in eigen-energies obtained by solving eq. (6.2) could be
significantly large. The E-k dispersion relationship for the two-band Kane model [80]
can be expressed in the simple form:
12
(6.4)
117
where m* is the effective mass and a coefficient contains nonparabolicity
corrections. This simple relation is accurate enough in cases of our interest. In this
work we extract along the quantization direction (here z-direction) from the data
obtained by the empirical pseudo-potential method.
Although several different approaches have been used to take into account in solving
Schrodinger equations [81-82], we take a simpler approach utilizing energy dependent
effective mass and iterations. The E-k dispersion relationship in eq. (6.4) can be
interpreted as variant effective mass dependent on energy.
, 1 , ; (6.5)
At each iteration, solving the eq. (6.2) to get energy levels of quantized states, ,
are updated from eq. (6.5). To avoid oscillation and slow convergence we used a
damping method to prevent an abrupt change in , and the Hamiltonian. As
depicted in fig. (6.2), after about 5 iterations, energy levels converge within 0.01eV.
For effective mass of 0.069 , of 0.7eV and channel thickness of 3nm, the error
in using a parabolic band is over 0.15eV.
118
6.4 Electron Wavefunction along the Channel
Since in thin film DGFETs vertical quantum confinement is particularly strong and
wavefunctions along the vertical direction (z-direction) vary slowly along the transport
direction (x-direction), we have shown in the BTBT chapter that in this case the inter-
subband couplings can be ignored and the transport equation in eq. (6.3) can be
decoupled from other subbands. The only significant coupling is from inter-band
BTBT due to the breaking of orthogonality between unit-cell functions in different
bands by the applied electric field. Without BTBT coupling, the solution of eq. (6.3)
can be approximated as plane waves, i.e.:
Conduction band wavefunction from source-side
exp ,′ ′ (6.6)
Figure 6.2: Convergence of the first energy level. Effective mass is 0.069 and is 0.7eV . Channel thickness is 3nm. About in 5 iterations, it
converges within 0.01eV.
0 2 4 6 8 10 120.4
0.45
0.5
0.55
0.6
E(eV)
# of iteration
119
0 (6.7)
| | | | 2 | | (6.8)
Conduction band wavefunction from drain-side
exp ,′ ′ (6.9)
0 (6.10)
| | 2 | | (6.11)
where the non-parabolic imaginary part of ,′ was given in the BTBT chapter. The
term 2 | | represents the component traveling in opposite direction
to the injected flux. As a result, the magnitude of the wavefunction is 1 when all the
injected flux passes through the channel without any reflection and its 2 when all is
reflected back by the potential barrier. The term is added to ensure the flux
conservation. is given as:
| || | (6.12)
In tunneling problems, the necessity of dealing with complex k and negative kinetic
energy naturally arises. We followed the method by suggested by Buttiker [83-85] to
treat the velocity of tunneling and just took the absolute value of velocity.
120
6.5 Non-parabolic Density of States
To sum up total charges in the channel, non-parabolic density of states (DOS) is
needed to be obtained. Let us approximately write the dispersion relation between
and as:
1 Δ (6.13)
, where Δ is the quantization energy of the n-th state in the valley and it can be
written as:
Δ , ; (6.14)
One-dimensional (1D) DOS along the y direction is given by,
, 21
2 (6.15)
By taking a derivative of eq. (6.13) with respect to , we have
2 ΔExz
ΔExz 2 , ΔExz 0 (6.16)
, where
ΔE 1 Δ (6.17)
By plugging in the eq. (6.16) into (6.15), we obtain the explicit form of 1D DOS along
the y direction:
,Δ 2
2 ΔE
(6.18)
The 1D DOS along the x direction can be obtained by the following procedure:
121
, ,1
2| |
12
| | 2 | |
12
| | 2 | | (6.19)
,where represents the contact where the carriers are injected and represents the
contact where the carriers are absorbed. We assume here that the x-directional electron
density is limited by the electron flux from the contacts, thus it follows the electron
density at the contact. By utilizing the derivative obtained in Eq. (6.18) and introducing
the term | ; | | | 2 | | , the x-directional 1D DOS
can be accordingly rewritten as:
, ,Δ
2 2| | ΔE| ; | (6.20)
Note that for 1D DOS in x direction we do not multiply it with the directional factor 2
(Eq.(6.15)), since we already took into account the reflected wavefunctions when we
obtained them in eq. (6.6)-(6.11)
6.6 Charge Density (with Non-parabolicity corrections)
The source and drain contacts are assumed to be ideal reservoirs in thermal
equilibrium with the Fermi energies and . They supply carriers to the channel
and absorb carriers from the channel without reflections. The Fermi distribution
functions at the contacts are given as:
122
,1
1 exp
(6.21)
,where k is the Boltzmann constant and T is the temperature.
If the total energy of the subband E is , where nzE is the eigen-
energy of n-th quantized states in z-direction, the 3D DOS in the channel can be
expressed as follows:
2 , , ;
Δ
2 | |; ;
Δ 2ΔE
(6.22)
The total charge density injected from one-side of the contact is given as:
| |2
; ;
| |
Δ Δ 2
ΔE 1 exp
(6.23)
Eq. (6.23) requires double numerical integration and this is computationally very
intensive work to be done on every point. To relieve the burden, it is tempting to carry
out the integration with respect to first and find the analytical form. When is
sufficiently small, the integration can be rewritten as:
Δ Δ 2
ΔE 1 exp(6.24)
123
ΔE
1. 2 .
.
1
1 exp
ΔE
1. 2 . .
1 expΘ ,
ΔE Δ .. 1 Δ .
.
Δ Θ ,
,where is the complete Fermi-Dirac integral of j order:
1 exp (6.25)
and
(6.26)
The analytical approximations of the complete Fermi-Dirac integrals of the orders, 1.5
and 0.5 are suggested in [86] with error less than 0.7%. We extended this work to treat
the integral of order -0.5, 1 and 2. The detail of approximations for the Fermi-Dirac
integrals is covered in Appendix.
In Eq. (6.24), we also introduced new variables , and ΔE , which are defined as:
/ΔE (6.27)
ΔE (6.28)
and
124
ΔE 1 (6.29)
We introduced the error terms Θ ; resulted from the Taylor series expansion of
.
1
1 exp
.
1 expΘ ,
(6.30)
As seen in Fig. 6.3, this approximation becomes inaccurate and leaves significantly
large Θ ; when and become large and we cannot just ignore it. To find
the analytical form for Θ ; , the simplest approximation would be to consider the
asymptotic behavior of the integral.
(a) (b)
Figure 6.3: The error term Θ , becomes significantly large value when
or are large. (a) For 0.5, 4/ and 40, the error term accounts for more than 10% of the correct integral. (b) For
2, 4/ and 40, the error term accounts for more than 25% of the correct integral.
-40 -20 0 20 400
2
4
6
8
10
12
EF0/kT
Θ (
E F0, β
) /In
tegr
al (E
rror
%)
by=1/eV
by=2/eV
by=4/eV
bxEx=0.5
-40 -20 0 20 400
5
10
15
20
25
30
Θ (
E F0, β
) /In
tegr
al (E
rror
%)
EF0/kT
by=1/eV
by=2/eV
by=4/eV
bxEx=2
125
In the limit of ∞ , the asymptotic behavior of the integration of
. in Eq. (6.24) is:
lim.
11
1 exp
.
1
2 . . atan . ..
. Θ , (6.31)
For this limit, the analytical form of the error term can be found by subtracting the
integral with the one without non-parabolicity ( 0).
Θ , 2 . . atan . . 23
. Θ , 0
, (6.32)
For the limit of ∞, the asymptotic behavior is approximated as:
Θ , .. , (6.33)
Utilizing the Eq. (6.32) and (6.33), now we have the analytical approximated form of
Θ ⁄ , , which is
Θ ⁄ ,
.. , 0
2 atan23
0 , 0
(6.34)
126
The relative error of the expression in Eq. (6.34) is plotted in Fig. 6.4. The maximum
error for of 4/eV is below 1.3%. Normally, the semiconductor materials we are
interested in have lower than 4/eV. For example, the in the Γ valley of GaAs is
just 0.69/eV and that of InAs is about 2/eV.
Now, we go back to the DGFET case which forms two contacts to source and drain as
depicted in Fig. (6.1). The total electron charge density supplied from source and drain
sides can be obtained utilizing the previous results.
| |2
;(6.35)
(a) (b)
Figure 6.4: Relative error of the approximation in Eq. (6.34) vs. the integral in Eq. (6.24). This approximation is accurate with an error below 0.3%.
-40 -20 0 20 40-0.1
-0.05
0
0.05
0.1
0.15
E'F/kT
by=1/eV
by=2/eV
by=4/eV
bxEx=0.5
Error (%)
-40 -20 0 20 40-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
EF0/kT
by=1/eV
by=2/eV
by=4/eV
bxEx=2
Error (%)
127
| ; | Θ ,
| ; | Θ ,
ΔExzΔ
| |
,where the non-parabolicity correction term Θ ⁄ , is
Θ ⁄ , 1 Δ .. Δ Θ ,
(6.36)
and ΔE 1 Δ , E
, E
, ΔE 1 ,
and as already defined in Eq. (6.17),
(6.27) , (6.28) , (6.29) and (6.26). Θ , is defined in Eq. (6.34).
These results correspond to the charge density in ballistic DGFET formulated by
Natori ([77]). We can easily show that his formula is the special case of the quantum-
ballistic DGFET with parabolic bandstructure. By taking the conditions 0,
Δ 1, 0, ; 0 ; 0 1, we have:
| |2
; (6.37)
Using the relation
ln 1 exp (6.38)
Eq. (6.37) can be written as:
128
| |2
;
ln 1 exp 1 exp (6.39)
, which is same as what Natori derived.
6.7 Quantum Ballistic Current (with Non-parabolicity
corrections)
The quantum-ballistic current is given by the Landauer’s formula [87] where it is the
sum of the current component flowing in each sub-channel. Each current component is
expressed as the product of the unit charge, the number of carriers flowing through the
sub-channel per unit time and the transmission coefficient.
2;
,,
(6.40)
where W is the width (y) of the device, is the carrier group velocity, is the
transmission coefficient and is the thickness of the channel. Let’s take a look at the
flux from source to drain.
,
,| ; |
| | , ,∞
0
∞
∞
(6.41)
Using the definitions in Eq. (6.12), (6.15), (6.18), (6.19) and (6.20) to substitute terms in
Eq. (6.41) with, we rewrite Eq. (6.41) as
129
,
1
2 | |
Δ 2
ΔE
1
1 exp
∞
0
∞
∞
(6.42)
,where .
Eq. (6.42) requires double numerical integration and this is computationally very
intensive work to be done on every point. As we did for calculating charge density, we
will carry out the integration with respect to first and find the analytical form.
When is sufficiently small, the integration can be rewritten as:
Δ 2
ΔE
1
1 exp
ΔE1 exp
, ΔE ,
(6.43)
We introduced the error term ; resulted from the Taylor series expansion of
E
E
+Higher order terms.
To find the analytic solution for ; , we first take a look at the asymptotic
behavior of the integral in Eq. (6.43) in the limit of ∞.
limΔ 2
ΔE
1
1 exp
(6.44)
130
Δ 2
ΔE
ΔE 2 . ΔE ,
For this limit, the analytical form of the error term can be found by subtracting the
integral with the one without non-parabolicity ( 0 ).
, 2 . 2 . , 0 , (6.45)
In the limit of ∞, by expanding E
with Taylor series expansion,
we can easily find that:
,32
. , (6.46)
By properly connecting two limits of , at 0, we can write , as
⁄ ,
32
. , 0
2 232
. 0 , 0
(6.47)
Fig. 6.5 depicts the accuracy of this approximation. Even with 4/ , the error is
less than 1.3%.
131
After the integration with respect to , using the analytic form of , given in
Eq. (6.47), Eq. (6.41) can be rewritten as
2 | | ΔE
.. ⁄ ,
(6.48)
Plugging Eq. (6.48) into Eq. (6.40), we can derive the equation to evaluate the current
flowing from source to drain in the device.
2 ;
| | ΔExz
12
12 0.5 ⁄ ,
| | ΔExz
12
12 0.5 ⁄ ,
(6.49)
Figure 6.5: Relative error of the approximation of in ⁄ , Eq. (6.47)vs. the integral in Eq.(6.44). This approximation is accurate in the range of error below 1.3%.
-40 -30 -20 -10 0 10 20 30 40-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
EF0/kT
Error (%)
by=1/eV
by=2/eV
by=4/eV
132
ΔE 1 Δ , E
, and
as already defined in Eq. (6.17), (6.28) and (6.26). , is defined in
Eq. (6.47).
6.8 Band-to-Band Tunneling Current
In the last section, we have developed the current formula for the majority carriers (in
this case, electrons). In devices made of high mobility materials, due to their small
band gaps and effective mass, the current generated by band-to-band tunneling (BTBT)
becomes increasingly important.
As shown in Fig. 6.6, BTBT current increases as we try to turn off the by applying
negative gate bias (Vg). This behavior imposes a minimum value of the leakage
current. If the crossover between the quantum-ballistic current and the BTBT current
occurs at relatively large Vg, it is not possible to reduce the leakage current any
further by just lowering gate bias, even if we can engineer the device structure to
achieve steep sub-threshold slope. In MOSFETs, the BTBT currents are generated at
the gate-drain edge (Fig. 6.7). Low gate bias and high drain bias create large electric
field enough to pull out the electrons in valence band to conduction band and as a
result, electron-hole pairs are generated and flow to opposite direction until they are
collected by contacts. Thus, the stronger gate field to lift up the channel potential,
more BTBT current is generated.
133
The approach to treat the BTBT current is slightly different from the modeling of the
current by majority electron carriers. In the previous section, when an electron travels
from one contact to the other, we assumed that it flows through complete quantum
coherent transport and that no-scattering events are allowed, since the channel length
is shorter than the relaxation length. However, unless the drain Fermi level ( ) is
lower than the source Fermi level ( ) by more than the band-gap ( ),
the holes generated by BTBT cannot overcome the potential barrier created by valence
band at source-side, consequently they are unable to reach the source contact (See Fig.
6.7). Thus, generated holes are reflected back to the cavity created between two
potential barriers and accumulated in the channel. As trapped holes are reflected back
and forth, they start to scatter and eventually they are thermalized. Thermalization
Figure 6.6: Source to drain current ( DSI ) vs. gate voltage (Vg). When negatively large Vg is applied, BTBT current generated at the edge between gateand drain grows significantly.
-0.2 0 0.2 0.4 0.610
-10
10-8
10-6
10-4
10-2
Vg (V)
Ids (A/um) Lg=15nm
Quantum-Ballistic
BTBT
Total Current
134
allows a certain fraction of lucky holes get enough energy to overcome the barriers.
Since the potential barrier along the source side is lower than the barrier toward the
drain contact, most of thermionic emissions occur toward the source. This
accumulation process continues until the thermionic emissions are balanced by the
incoming holes supplied by BTBT process.
As the first step to model the hole transport in the channel, we assume that a hole
quasi-Fermi level ( ) can be defined in the hole cavity. This assumption that holes
roughly exhibit the Fermi-Dirac distribution is not irrational since in the off-state the
barrier height is large enough (>0.1eV or 5KT) that the most of holes in the channel
are trapped inside the cavity, while only a small fraction of holes can flow to reach
source side.
Figure 6.7: Illustration of the carrier flux components. Electron currents consistof the one injected from source side, one from drain side and onegenerated by BTBT process. The BTB tunneling hole current issupplied to the channel cavity (around ). The accumulated holes can escape by thermionic emission from the cavity.
IHole,Therm
IHole,BTBT
xEmax
Q-BallisticIe,Source
Q-BallisticIe, Drain
BTBT
Ie,BTBT
EFS
EFH
EFD
135
The procedure to find the hole BTBT current in equilibrium state can be roughly
divided into three steps. The first step involves calculation of BTBT current. The
second step is to evaluate the thermionic emission current and the final step is to find
the hole quasi-Fermi level such that it equates the thermionic current and the BTBT
current.
In the BTBT chapter, we derived the expressions for the BTBT tunneling electron-
hole pair generation rates in 2-D quantized system. The model is useful for the particle
based device simulator (ex. Monte Carlo or drift-diffusion), since it provides with the
carrier generation rate at each point. One way to find the BTBT current is to calculate
the generation rate through the path searching and then summing up the generation
rates of each point over the space. When the BTBT process rate is very large, the hole
quasi-Fermi level goes down below the valence band level. Low hole quasi-Fermi
level not only increases the increase the thermionic emission, but also reduces BTBT
rate. When the states in valence band are depleted (in other terms occupied with holes),
it limits the BTBT process. To take into account the depletion effect, we add the
Fermi-Dirac distribution to the generation rate, which is as follows:
,,,
1
1 exp
1
1 exp
(6.50)
, where is the Fermi level of the drain contact and is the hole quasi-Fermi
level in the channel (Fig. 6.7). and are the energy levels of states in valence
136
band and conduction band, respectively. Since the path finding algorithm searches for
the point of which energy level is same as the original points, and are
related such that:
, For Direct Tunneling
, For Phonon assisted Tunneling (6.51)
Now we evaluate the hole thermionic current injected from the channel to the source.
The energy levels governing the thermionic emission are illustrated in Fig.6.8. is
the barrier height that the hole in n-th state experiences.
We will use Eq. (6.49) to derive the expression for the thermionic emission from the
cavity. Let’s first take a look at in Eq. (6.47). BTBT current becomes significantly
large only when the drive current becomes few orders lowers than its maximum (Fig.
Figure 6.8: Sketchy of the band profile along the channel. Source Fermi level( ), Drain Fermi level ( ), Energy profile of the n-th state of valence band, hole quasi-Fermi level ( ), hole barrier height ( ) are marked in the figure. Only the hole with kinetic energy larger than
can escape the cavity and reach the source side.
xEmaxxS
EFS
EFH
EFD
Ev
Enz| EB |
137
6.6). Thus, we can assume that hole barrier , which is roughly same as electron
barrier height, is at least bigger than 0.1eV when BTBT current has any meaningful
number. This condition can be expressed as:
0.1 0.1 (6.52)
Note that the sign of has changed from the electron case. Under this condition,
can be approximated as
,32
34
exp , 0.1 (6.53)
With the same condition, the Fermi-Dirac Integral of -0.5 order, . in Eq. (6.49)
can be approximated as
. exp , 0.1 (6.54)
The | | for holes have the following values.
| | 1 0 (6.55)
Utilizing Eq. (6.53) , (6.54), and (6.55), Eq. (6.49) can be simplified:
2 ΔExz 1
34
exp (6.56)
, where
ΔE 1 E E (6.57)
and
/ΔE (6.58)
138
The integration over can be approximately carried out since the term exp
decreases very fast as increases, while the term ΔE 1 stays at
relatively same values. After the integration Eq. (6.56) becomes
2 ΔEB 1
34
exp(6.59)
where
ΔEB ΔE 1 E E EB (6.60)
Δ (6.61)
E EFH EB (6.62)
Physically, there should not be any current flowing between two thermal equilibrium
points when the Fermi levels of the points are same. To ensure the current obtained by
Eq. (6.59) goes to zero as EFH EFS , we slightly modify the Eq. (6.59), and that
becomes:
expEFS EFH 1 (6.63)
,where
2 ΔEB 1
34
expE EFS EB
(6.64)
Finally, by equating the thermionic emission current (Eq. (6.63)) and the BTBT current
(Eq. (6.50)), we can find the expression for the hole Fermi level EFH:
EFH EFS log 1 (6.65)
139
Once we have the hole Fermi level, hole density can be also obtained. The expression
for hole density is similar to the electron density in Eq. (6.35).
| | ;
Θ ,ΔExzΔ
| | (6.66)
,where the non-parabolicity correction term Θ ⁄ , is
Θ ⁄ , 1 Δ .. Δ Θ ,
(6.67)
and ΔE 1 Δ , E
, E
, ΔE 1 , and
. Θ , is defined in Eq. (6.34).
6.9 Summary
We have a developed quantum ballistic transport model to accurately model the carrier
transport in highly scaled DGFETs. This model takes into account the effects of non-
parabolicity of conduction bands on transport mass and 1-D density of states.
Asymptotic approximation helped to simplify model. The BTBT leakage current is
also included in the ballistic transport model. The flow chart of the entire device
simulation using this transport model is given in Fig. 6.9.
140
Figure 6.9: Flow chart of the carrier transport simulation of quantum-ballistic
FETs.
InitializeQe and Qh
Solve Poisson V
Solve 1-D Schrodinger Ea,n
Evaluate Qe , Qh , EFH , Ie and IBTBT
If Q and I areconverged
Set New Bias on Electrodes
No
Yes
141
Chapter 7 Properties and Trade-offs Of
Compound Semiconductor MOSFETs
7.1 Introduction
As we continue to aggressively scale transistors in accordance with Moore’s Law to
sub-20-nm dimensions, it becomes increasingly difficult to maintain the required
device performance. Currently, the increase in drive currents for faster switching
speeds at lower supply voltages is largely at the expense of an exponentially growing
leakage current, which leads to a large standby power dissipation. There is an
important need to explore novel channel materials and device structures that would
provide us with energy-efficient nanoscale MOSFETs. Due to their significant
transport advantage, high mobility materials are very actively being researched as
channel materials for future highly scaled CMOS [27, 31-32, 88-94]. However, most
high mobility materials also have a significantly smaller bandgap compared to Si,
leading to very high band-to-band tunneling (BTBT) leakage currents, which may
ultimately limit their scalability.
142
The different significant components of channel leakage are shown in Fig. 7.1.
(Assuming an ideal high-k gate dielectric technology, gate leakage has not been
addressed in the present study.) Conventionally, the exponentially rising diffusion (or
subthreshold) current due to lowered threshold voltages has been the dominant leakage
mechanism in Si MOSFETs. However, with continued device scaling into the
nanometer regime and the increasing E-fields in the channel, there is also an
exponential growth in the tunneling components (BTBT and direct source/drain (S–D)
tunneling) of leakage. Fig. 7.2 (a) and (b) shows the material parameters of low
effective mass (high mobility) semiconducting channel materials. A universal trend is
observed with respect to the bandgap and the dielectric constant of low effective mass
Figure 7.1 : Dominant leakage paths in nanoscale high mobility CMOS devices.
High mobility (low-bandgap) materials suffer from excessive BTBT
leakage.
IHole,Therm
IHole,BTBT
xEmax
Q-BallisticIe,Source
Q-BallisticIe, Drain
BTBT
Ie,BTBT
EFS
EFH
EFD
143
semiconductors. As we push toward lower effective mass (high mobility) channel
materials, the bandgap rapidly drops and the dielectric constant increases.
As seen in Table 7.1, III-V materials have significantly smaller effective mass and
higher electron mobility compared to Si and Ge. However, most high mobility
materials like Ge, InAs, and InSb also have a significantly lower bandgap compared to
Si. Due to the increasing E-fields in the channel and the smaller bandgap in these high
mobility materials, the BTBT leakage current can become excessive and can
ultimately limit the scalability of high mobility channel materials. Another point to
(a)
(b)
Figure 7.2 : Tradeoffs between effective mass (m ), bandgap (EG), and dielectric
constant (κs) in semiconducting materials.
144
note is that since most high mobility materials also typically have a higher permittivity
(κs), they also suffer from worse short cahnnel effects (SCEs).
At an initial glance, due to their extremely small transport mass leading to high
injection velocity (vinj), III-V materials appear to be very attractive candidates as
channel materials for highly scaled n-MOSFETs [26, 88, 95]. However, III-V
materials have many significant and fundamental issues such as small density of states
(DOS), bigger dielectric constant and large BTBT leakage current, which may prove
to be severe bottlenecks to their implementation. These tradeoffs need to be
systematically and thoroughly investigated from a theoretical standpoint in order to
help in providing useful guidelines, which can efficiently streamline the resources
invested by several research groups throughout the world in their experimental
development.
Table 7.1 : Dominant leakage paths in nanoscale high mobility CMOS devices.
High mobility (low-bandgap) materials suffer from excessive BTBT
leakage.
17.717.714.812.41611.8εr
0.170.170.360.361.420.661.12EG (eV)
77,00077,00040,00040,000920039001600μn(cm2/Vs)
0.0140.0140.0230.0230.0670.080.19meff*
InSbInSbInAsInAsGaAsGaAsGeGeSiSiMaterial/PMaterial/Propertyroperty
17.717.714.812.41611.8εr
0.170.170.360.361.420.661.12EG (eV)
77,00077,00040,00040,000920039001600μn(cm2/Vs)
0.0140.0140.0230.0230.0670.080.19meff*
InSbInSbInAsInAsGaAsGaAsGeGeSiSiMaterial/PMaterial/Propertyroperty
145
In this chapter, we will look at novel III-V channel materials for n- and p- MOSFETs,
and the application of various strain along different surface/channel orientations to
improve their performance, in terms of increased drive current and lower leakage. We
will also discuss novel device structures, and approaches to address the problem of
parasitics, which can enable energy-efficient nanoscale III-V n- and p- MOSFETs.
7.2 Simulation Framework
7.2.1 Bandstructure calculation (Real and Complex)
The transport of electric charge carriers (electrons and holes) in crystal
semiconductors is governed inherently by their bandstructures. Real part of
bandstructure provides information on how the carriers are distributed in energy-k
space (density of states and valleys) and how the carriers react to the external electric
fields (effective masses). The imaginary part of bandstructure helps us to understand
the tunneling process of carriers between different bands.
The Plane-wave based empirical pseudopotential method (EPM) [5] is one of the
useful ways to accurately estimate the details of the bandstructure. While the tight-
binding method or kÿp theory requires many fitting parameters, EPM only requires a
146
few fitting parameters and provides more accurate bandstructures, thus EPM is more
suitable when it is needed to study the entire first Brillouin zone and the effect of the
strain.
Fig. 7.3 shows the result of the bandstructure calculation done by EPM for Ge and
InAs described in the Chapter 2. Using this method, both complex part and real part of
bandstructure can be obtained.
7.2.2 BandtoBand Tunneling (OffState Leakage)
In semiconductors, most of states in valence bands are filled with electrons. Under a
strong applied external electric field, electrons in valence bands can tunnel through the
forbidden bandgap and when they reach the conduction bands, electron-hole pairs are
(a) (b)
Figure 7.3 : (a) 3D plot of complex bandstructure of germanium along the [111]
direction. (b) 2D plot of complex bandstructure of InAs.
147
created (Fig. 7.4) [96]. In in-direct band-gap materials such as Si, since the minimum
of the conduction band is located at ∆ point away from the top of valence band which
is located at Γv, the momentum conservation law prohibits the tunneling between two
bands (Fig. 7.4). Phonons created by thermal lattice vibration can provide the
momentum difference and allow two bands to be coupled. This tunneling process
incorporating phonons is called in-direct BTBT or phonon assisted BTBT, which is
the dominant BTBT process in Si [73].
We evaluated the tunneling rate interband matrix elements between quantized states
for both direct and indirect tunneling, as described in the Chapter 3. The final BTBT
carrier generation rate was calculated by adding up the transition rates for all the
possible transitions [72-73]. We included all the possible transition from valence band
to conduction bands such as ΓV-ΓC, ΓV-L and ΓV-X (Fig. 7.4 (b)). While counting, for
direct tunneling the momentum conservation selection rule was applied, while for the
indirect tunneling the selection rule was relieved.
148
The efficiency of BTBT process is altered by large amounts in few nm thick slabs
compared to bulk semiconductors [73]. An ultra-thin body has a larger effective band
gap and smaller DOS than bulk, due to strong quantization. With larger effective band
gap, the wave function decays faster inside the forbidden gap and lowers BTBT rate
(Fig. 7.5). To take into account the quantization effect caused by ultra-thin body, the
wavefunctions and the energy levels of quantized subband states were obtained by
solving 1-D Schrödinger equation along z-direction for both electrons and holes [97].
(a) (b)
Figure 7.4 : (a) Illustration of band-to-band tunneling paths. The electron can
tunnel to indirect-gap conduction band (L or ∆ valleys) through
phonon interactions. (b) All Possible (Direct and Indirect) BTBT
Tunneling Paths, Γ-Γ, Γ-X and Γ-L, are captured by the model.
149
7.2.3 2DQuantum Ballistic Current (OnState Drive Current)
For high mobility channel FETs with gate length shorter than 20nm, the majority
carriers injected from source travel through channel without significantly hindered by
scattering [98-99]. In [98] Wang and Lundstrom explained the validity of ballistic
transport and in [77] Natori formulated the ballistic transport in thin oxide MOSFETs
where states are quantized by strong vertical gate electric field. However, in highly
scaled DG MOSFETs where channel is only few nm thick, quantum confinement
Figure 7.5 : Ultra-thin body and larger quantization increases the effective
bandgap and lowers the tunneling rate.
150
effect is strong enough to lift quantized energy states far above the band-edge. In this
case, E-k dispersion significantly deviates from an ideal parabolic curve, thus adopting
Natori’s formula would result in over-estimations on quantized energy level and
under-estimations on the density of states. Moreover, in high mobility channel, its
small effective mass’ significantly enlarges the source-to-drain (S/D) tunneling which
makes sub-threshold slope lower and increases leakage current. As described in the
Chapter 6 , we extended applicability of Natori’s model so that the effects of non-
parabolicity and S/D tunneling are included.
7.2.4 Benchmark Test bed Device Structure
Since the purpose of this chapter is on exploring the suitability of high mobility III-Vs
as channel materials, we need to focus solely on carrier transport characteristics
(Ion/Ioff), while the structural variability and optimization should be excluded. The
structure, which suits best to the purpose and still closely resembles the structure
manufacturable in a next few generations, is a double gate FET structure (Fig. 7.6).
We assumed source and drain are perfect carrier absorbers, undoped channel and
abrupt doping profiles with a 1.3nm/dec. roll-off.
151
7.3 Power-Performance Tradeoffs in Binary III-V materials
(GaAs, InAs, InP and InSb) vs. Si and Ge.
Although their small transport mass leads to high injection velocity (vinj), III-V
materials have a low density of states (DOS) in the Γ-valley, tending to reduce the
inversion charge (Qinv) and hence reduce drive current [34, 99-102]. Furthermore, the
small direct band gaps of Ge and III-V materials inherently give rise to very large
Figure 7.6 : Test bed for benchmarking III-Vs as channel material. Gate Length
LG=15nm, equivalent gate oxide thickness (EOT) TOX=0.7nm,
channel thickness TS=3-5nm and VDD=0.3-0.7V.
152
band to band tunneling (BTBT) leakage current compared to Si [73-74]. They also
have a high permittivity and hence are more prone to short channel effects (SCE).
Quantum confinement in these ultra-thin nanoscale DGFETs plays a very important
role. In III-V materials, large quantum confinement makes electrons populate and
conduct in heavier L- or X-valleys. Furthermore, there is a significant increase in the
conductivity mass in quantized sub-bands due to large non-parabolicity of Γ-valley
[23, 34]. Quantum confinement effects also reduce BTBT leakage in an ultra thin
channel [72-74]. In this section, we will discuss Double Gate (DG) n-MOSFETs with
Ge and IIIV materials (GaAs, InP, InxGa(1-x)As, InAs and InSb) and compared to Si
[23-24, 33, 72].
7.3.1 Inversion charge and injection velocity
Generally, low DOS in III-V materials reduces the effective gate capacitance (CGeff),
resulting in lower inversion charge at the given gate voltage (Fig. 7.7). For example,
InAs can store only one third of electrons that Si can store. Fig. 7.8 shows injection
velocities (vinj) at VGate of 0.7V. Due to their small transport effective mass, III-V
materials and Ge have several times larger injection velocity than Si. Among all,
transport effective masses of InAs and InSb are only 0.023 m0 and 0.014 m0,
respectively and they exhibit 6 times larger vinj than Si. As we scale channel thickness
(TS), the Γ-valley energies in III-V materials rise up rapidly due to very low mz
resulting in most of the Qinv in thin films moving to the heavier L-valley. Once the L-
153
valleys are resided by electrons, III-V materials start to behave like Ge – large CGeff
and large Qinv. The L-valleys start to contribute for GaAs at TS=7nm and for InSb at
3nm. Scaling TS improves subthreshold slopes and DIBL effects, permitting larger
Qinv or higher Fermi level at the given gate voltages; thus carriers are injected at a
higher vinj from source to drain in thin body (Fig. 7.8). However, in small bandgap
materials such as InAs and InSb, due to strong non-parabolicity in Γ-valley, high
Fermi levels and lifted-up quantized energy states in thin body make electrons heavier
and injection velocity is reduced.
Fig. 7.7: Inversion charge (Qi) at VGate=0.7V and Tox=0.7nm. Si channel stores 4 times
more electrons than InAs. As Ts scales, Qi becomes bigger due to reduced Vth and
nonparabolicity of band.
S i G a A s I n P G e I n A s I n S b0
5
1 0
1 5
Qi (
1012
#/cm
2 )
1 0 n m5 n m3 n m
154
Fig. 7.8: Injection velocities (vinj) at VGate=0.7V and Tox=0.7nm. InAs and InSb have 6
times larger vinj than Si due to their small transport mass. As Ts scales, lowered
threshold voltage (vth) increases vinj, while nonparabolicity reduces vinj.
7.3.2 ION, IOFF,BTBT and Delay
Device simulation results for LG=15nm, TOX=0.7nm, TS=3-5nm and VDD=0.3-0.7V
are shown in Fig. 7.9 (ION) and Fig 7.10 (IOFF,BTBT). There have been concerns that III-
V materials would underperform than Si because of their low Qinv [23-24, 33, 99-100,
103]. However our study shows that despite of low Qinv, thanks to their large vinj, III-V
materials like InAs, InSb and InP can flow up to 80% larger drive current than Si.
Although L valleys of conduction bands in Ge has large density of states and small
transport effective mass, the drive current in Ge channel is lessened by worst SCE. In
3nm thick channel, SCE can be minimized, therefore Ge allows highest drive current.
S i G a A s I n P G e I n A s I n S b0
2
4
6
8
v inj
(107 cm
/s)
1 0 n m5 n m3 n m
155
(a) (b)
Fig. 7.9: Drive currents (ION) for various (a) Ts and (b) VDD. LG=15nm, TOX=0.7nm.
The minimum achievable standby leakage by BTBT (IOFF,BTBT) is at the intersection of
the BTBT leakage with the sub-threshold leakage. DGFET cannot have lower leakage
current than IOFF,BTBT by adjusting threshold voltage. Small bandgap materials such as
InAs, InSb and Ge have extremely large IOFF,BTBT higher than 0.1µA/µm. However, if
the technology allows to thin channel thickness to 3nm, it is possible to limit IOFF,BTBT
below 0.1µA/µm. Based on our study, larger than 1000x reduction is possible in
GaAs, InP and Ge by scaling Ts from 10nm to 3nm.
Si GaAs InP Ge InAs InSb0
1
2
3
4
5
6
ION (m
A/ μ
m)
10nm5nm3nm
Si GaAs InP Ge InAs InSb0
1
2
3
4
5
ION (m
A/ μ
m)
0.7V0.5V0.3V
(a) (b) (c)
Figure 7.10 : (a) IOFF,BTBT is minimum achievable leakage current defined as a
intersecting point between BTBT current and Drive current. (b)
IOFF,BTBT for various Ts. VDD=0.5V. (c) IOFF,BTBT for various VDD.
log (IDS)
IOFF,BTBT
VGS
IBTBT
IDrive
Si GaAs InP Ge InAs InSb
IOFF
,BT
BT
(A/ μ
m)
10- 1
310
-910
- 710
-510
- 310
-11
7nm5nm3nm
10nm
Si GaAs InP Ge InAs InSb
IOFF
,BT
BT
(A/ μ
m)
10-1
310
-910
-710
-510
-310
-11
0.7V0.5V0.3V
156
7.3.3 Effect of Scaling film thickness and VDD
Scaling TS enhances the device performance by eliminating the SCE effect and
reducing BTBT leakage. IOFF,BTBT in Ge, InAs, GaAs and InSb can be reduced by over
1000x by scaling Ts to 3nm (Fig. 7.10), due to enlarged bandgap by quantization of
sub-bands. As Ts scales, ION is increased by 2.5 times (Fig. 7.9), but concurrently
electrons become heavier, resulting in losing the advantage of III-Vs over Si. Scaling
VDD is desirable since it reduces BTBT leakage (Fig. 7.10) by relieving maximum
electric field applied in the device and it decreases switching energy. III-Vs and Ge
with VDD of 0.5V provide with higher drive current than Si does with VDD of 0.7V
(Fig. 7.9 (b)), which means a transistor made with III-Vs and Ge can consume only ½
of active energy consumed by a transistor with Si. However large threshold voltages
of III-Vs prevent VDD from scaling below 0.5V where III-Vs start to drive smaller
currents than Si. InP exhibits good performance (high ION, short delay and low
IOFF,BTBT) in any scaling scenarios.
7.3.4 PowerPerformance Tradeoff of Binary IIIV Materials vs.
Si and Ge
157
Relationships between delay and IOFF,BTBT for different materials are depicted in Fig.
7.11. Subthreshold leakage is fixed to be 0.1µA/µm and VDD is varied from 0.3V to
0.7V. The best material should show short delay and low IOFF,BTBT at the same time. Si
exhibits slowest switching and lowest IOFF,BTBT. Although delays for InAs and InSb are
small, they suffer extremely large IOFF,BTBT. InP exhibit fast switching time, large ION
at the reduced IOFF,BTBT.
Fig. 7.11: Intrinsic delay vs. IOFF,BTBT trade-off in various materials. Lg=15nm,
Tox=0.7nm, Ts=5nm. VDD is varied from 0.3V to 0.7V.
Our results show that with oxide thickness of 0.7 nm, small density of states (DOS) of
these materials does not significantly limit the on-current (ION) and high µ materials
still show higher ION than Si. However, the high µ small bandgap materials like InAs,
InSb and Ge, suffer from excessive BTBT current and poor SCE, which can limit their
scalability. Ge has highest ION due to its large DOS and small transport effective mass,
158
but it also suffers from large BTBT leakage. Scaling TS enhances the device
performance by eliminating the SCE effect and reducing BTBT leakage. Overall,
among all the unstrained IIIV materials In0.25Ga0.75As and InP exhibit the best
performance - high ION, low delay and low IOFF,BTBT compared to Si.
7.4 Power-Performance of Strained Ternary III-V Material
(InxGa1-xAs)
InxGa1-xAs is a very promising candidate for future NFETs because it allows for a very
good tradeoff between the excellent transport properties of InAs and the low leakage
of GaAs [23, 34, 102]. Just like in the case of Si or Ge, strain engineering can be used
to further enhance the performance of III-V materials in terms of both, increasing the
drive current and reducing the off-state leakage. By varying strain conditions and
orientations for the different compositions, the best performing strained InxGa(1-x)As
was identified. Unlike p-MOS [28], in the case of n-MOS uniaxial strain does not
significantly improve performance [25].
In this section, we will discuss power-performance tradeoffs in nanoscale DG NFETs
with channel materials composed of InxGa1-xAs under various biaxial strain conditions
with different orientations.
159
7.4.1 Strained InGaAs Band Structures
Local empirical pseudopotential method (LEPM) [8] is used to obtain the
bandstructure of strained InGaAs. Fig. 7.12 shows the band shifts of GaAs (001) and
In0.75Ga0.25As (111) by applying a biaxial strain.
(a) (b)
Fig. 7.12: Band shifts in biaxially strained (a) GaAs (001) and (b) In0.75Ga0.25As (111)
calculated by local empirical pseudopotential method. For GaAs(001), 4% biaxial
compressive (BiC) strain is applied and for In0.75Ga0.25As (111), 4% biaxial tensile
(BiT) strain is applied. Solid lines represent relaxed material and dotted lines represent
strained material.
Application of strain can strongly modify the bandgap (Eg) of material. Fig. 7.13 (a)
shows Eg of the different 5nm thin InxGa1-xAs as a function of biaxial
(compressive/tensile) strain and orientation. On quantization, due the small mass of
the electrons in the Γ-valley, the carriers start occupying the high L- and X- valleys.
1 0.8 0.4 0 0.4 0.8
-6
-5
-4
-3 eV
Relaxed
Strained
BiT 4%
LX1 0.8 0.4 0 0.4 0.8
-6
-5
-4
-3 eV
Relaxed
StrainedBiC 4%
LX
160
Hence, in III-V materials, apart from the Eg, the separation between the Γ- and L-
valley (ΔEΓL) is a very important parameter in determining the transport. Fig. 7.13 (b)
shows ΔEΓL under different strain/orientation conditions. Due to strong non-
parabolicity in the Γ-valley, transport effective mass is bigger in thin layers than in
bulk material.
(a) (b)
Fig. 7.13 (a) Quantized bandgap as a function of strain (e||) in 5nm film. Compressive
strain increases bandgap. Especially (111) orientation is strongly modified (b) Γ-L
Separations as a function of strain (e||) in 5nm film. Tensile strain widens the
separation. With tensile strain, ΔEΓL in GaAs can be over 0.3 (eV).
Tunnel mass (mTunnel) determines how well electron and hole penetrate into bandgap
and cause BTBT leakage. Tunnel mass is also strongly modified with application of
strain [25]. Unlike Eg and ΔEΓL, mTunnel shows a strong dependency on wafer
orientation (Fig. 7.14). In (111) wafer, compressive strain increases mTunnel, while in
(001) wafer, tensile strain increases mTunnel.
Ega
p (e
V)
-0.04 -0.02 0 0.02 0.040.4
0.8
1.2
1.6
2
2.4
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
Ts=5nm
Ega
p (e
V)
-0.04 -0.02 0 0.02 0.040.4
0.8
1.2
1.6
2
2.4
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
Ega
p (e
V)
-0.04 -0.02 0 0.02 0.040.4
0.8
1.2
1.6
2
2.4
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
Ts=5nm
ΔE Γ
L (e
V)
-0.04 -0.02 0 0.02 0.04
-0.4
0
0.4
0.8
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
Ts=5nm
ΔE Γ
L (e
V)
-0.04 -0.02 0 0.02 0.04
-0.4
0
0.4
0.8
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
Ts=5nm
161
7.4.2 ION and IOFF,BTBT with Strain Engineering and Channel
Orientation
Device simulation results for LG=15nm, TOX=0.7nm, TS=5nm, VDD=0.7V (IOFF=10-
7A/um) are shown in 7.15 (a) (ION) and 7.15 (b) (IOFF). In relaxed GaAs, ΔEΓL is not
large enough to prevent the population of electrons in L-valleys. In spite of the
occupation in L-valleys providing large channel charge, relaxed GaAs suffers from
low injection velocity (Fig. 7.8), which results in overall low drive current. This
implies that with 0.7nm-thick-gate oxide, having small mass and thus, fast injection
velocity is more effective for larger ION than having large mass and large DOS. Indeed,
with the application of tensile-biaxial strain, which increases ΔEΓL (Fig. 7.13(b)), the
drive current of GaAs can be boosted up to over 4mA/μm and the delay is shortened to
(a) (b)
Figure 7.14 : (a) 001 (b) 111. Tunnel mass as a function of x (InxGa1-xAs) in 5nm
film. On (111) Orientation, compressive strain increases mass, while
on (001), it reduces mass.
162
55fs [25]. The drive current of In0.75Ga0.25As is only slightly affected by strain, as
large ΔEΓL ensures that the majority of carriers reside in Γ-valley. Therefore,
regardless of strain levels, In0.75Ga0.25As exhibits 30% higher ION (4.3mA/μm) than 2%
biaxial tensile strained Si. Although both (111) and (001) orientation tensile strains
enhance the ION of GaAs, biaxial-tensile strained GaAs (111) suffers large IOFF,BTBT
due to reduction in bandgap and decrease in tunnel mass. Relaxed In0.75Ga0.25As have
small bandgap, leading to too high IOFF,BTBT (Fig. 7.15(b)). (111) compressive strain
effectively reduces leakage current in In0.75Ga0.25As by concurrently increasing
bandgap and tunneling mass. Due to the enlarged bandgap and tunneling mass, the
leakage current for 4% biaxially compressive-strained In0.75Ga0.25As (111) is below
the leakage specification of 0.1μmA/μm. However, (001) compressive strain does not
notably reduce the BTBT leakage current, since despite the increased bandgap, it
simultaneously reduces the tunneling mass, leading to no significant reduction in
BTBT.
163
7.4.3 PowerPerformance of Strained Ternary IIIV Material
(InGaAs)
InxGa1-xAs is a very promising candidate for future NFETs. Fig. 7.16 shows the best
channel materials and how these materials can be improved by strain engineering.
Based on our simulation results, GaAs (001), In0.25Ga0.75As (001) and In0.75Ga0.25As
(111) are selected as the best channel materials. Although GaAs (001) exhibits the
lowest IOFF,BTBT due to its large bandgap (>1.4eV), the drive current of GaAs (001) is
higher than strained Si. By applying biaxial tensile-strain, GaAs (001) can be
(a) (b)
Figure 7.15 : (a) ION as a function of strain (e||) in strained GaAs and In0.75Ga0.25As
NMOS DGFET. For comparison, IONs for 2% biaxially tensile-
strained (BiT) Si and unstrained Si are plotted as dotted lines. (b)
IOFFBTBT as a function of strain in strained GaAs and In0.75Ga0.25As. For
comparison, IOFFBTBT for 2% biaxially tensile-strained Si is drawn as
dotted line. 4% (111) compressive strain reduces IOFF,BTBT by over
1000x.
I ON
(mA
/ μm
)
-0.04 -0.02 0 0.02 0.04
2
3
4
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
Si
2% Strained Si (BiT)I O
N (m
A/ μ
m)
-0.04 -0.02 0 0.02 0.04
2
3
4
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
Si
2% Strained Si (BiT)
I OFF
,BT
BT
(A/ μ
m)
10-9
-0.04 -0.02 0 0.02 0.04
1
10-1
110
-710
-510
-3
2% Strained Si (BiT)
100nA/μm
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)I OFF
,BT
BT
(A/ μ
m)
10-9
-0.04 -0.02 0 0.02 0.04
1
10-1
110
-710
-510
-3
I OFF
,BT
BT
(A/ μ
m)
10-9
-0.04 -0.02 0 0.02 0.04
1
10-1
110
-710
-510
-3
2% Strained Si (BiT)
100nA/μm
GaAs (001)GaAs (111)In0.75Ga0.25As (001)
In0.75Ga0.25As (111)
164
improved to have ION as high as In0.25Ga0.75As, with marginal increase in IOFF,BTBT.
In0.25Ga0.75As (001) has both good ION and low IOFF,BTBT because of its large bandgap
(>1eV) and large ΔEΓ-L. 4% uniaxial compressive strain can further reduce the leakage
in In0.25Ga0.75As (001), without significant reduction in ION. In0.75Ga0.25As (111) has
excellent carrier transport properties, but it suffers large IOFF,BTBT. Biaxial compressive
strain can reduce the leakage in In0.75Ga0.25As (111) below 0.1μA/μm. Considering the
continuation in device scaling, In0.75Ga0.25As (111) would be the better material choice
over other InxGa(1-x)As compositions. As the gate length scales down close to 10nm,
further scaling in channel thickness would be necessary to control the short channel
effects. Larger quantization effect in thinner body will further increase the bandgap of
In0.75Ga0.25As (111), leading to even smaller leakage current [25]. By contrast, the
scalability of devices with the channel materials of GaAs (001) and In0.25Ga0.75As (111)
would be limited, since the strong quantization effect in Γ-valley will diminish ION due
to reduced ΔEΓ-L.
165
7.5 Strained III-V for p-MOSFETs
As seen in the previous sections, III-V compounds are one of the most promising
materials for future high-speed, low-power n-MOSFETs due to their high electron
mobility. Recently, high performance III-V n-FETs have been demonstrated. However,
for CMOS logic, there is a significant challenge of identifying high mobility III-V p-
FET candidates [104-106]. Strain in silicon has been successful in significantly
enhancing the p-MOS performance and is now employed ubiquitously in the industry.
(a) (b)
Figure 7.16 : (a) ION vs. IOFFBTBT and (b) Delay vs. IOFFBTBT of the best
performing n-DGFETs with materials, biaxially tensile-strained GaAs
(001), uniaxially compressive-strained In0.25Ga0.75As (001) and
biaxially compressive-strained In0.75Ga0.25As (111). Strain levels are 0,
0.02 and 0.04. Values for biaxially tensile-strained Si are given for
comparison.
166
Use of strain to reduce hole effective masses by splitting the heavy-hole (hh) and
light-hole (lh) valence bands was first demonstrated in p-channel InGaAs/ (Al)GaAs
[107-108]. More recently, the technique has been applied to strained InSb [109], GaSb
[110] and InGaSb [111] based channels for improving hole nobility (μh). Given the
many choices available for materials, stoichiometry, strain, channel orientation, a
modeling effort is necessary to evaluate different options and narrow down the choice
for experimentation.
7.5.1 Hole Mobility in Ternary IIIV Materials (InGaAs vs. InGaSb)
Unlike for electrons in III-V where polar scattering is the dominant scattering
mechanism (non polar optical phonons do not interact due to s-like spherical
symmetry of conduction band in III-V’s [12])) both deformation potential and polar
scattering mechanisms are important for hole mobility. Varying group III element
while keeping the group V element same (i.e. InAs → GaAs) μh does not change
appreciably while a large change is observed when the group V element is varied (i.e.
InP → InAs → InSb). This is a consequence of the fact that the valence bands of III-V
mostly derive from the p-orbitals of the anion [112]. Fig. 7.17 plots the low field μh of
InxGa1-xAs and InxGa1-xSb (Sheet Charge (NS) =1012/cm2). Sb’s have significantly
higher mobilities than As’s.
167
7.5.2 Hole Mobility Enhancement in IIIVs with Strain
Next we look at the effect of biaxial and uniaxial strain on μh enhancement in these
III-V materials. Fig. 7.18 shows the μh enhancement with respect to the unstrained
case (for uniaxial case the strain is always applied along the channel direction). In all
cases compression is better than tension for μh enhancement. In the case of biaxial
strain, the enhancement in (100) is isotropic and maximum enhancement with 2%
biaxial strain is ~2X. Much higher μh enhancement is achieved with same % of
uniaxial strain (~4X with 2% uniaxial compression). Also, in the case of uniaxial
strain, the enhancement is highly anisotropic. Also note that ~2 times lower stress is
Figure 7.17 : Mobility for InXGa1-XSb and InXGa1-XAs. Sb’s have higher (~2X)
hole mobilities than As’s [104].
168
required to produce the same amount of strain in InSb/GaSb as compared to Si, due to
their lower elasticity constants. Fig. 7.19 shows that a hole mobility greater than
1600cm2/Vs can be obtained by application of biaxial strain in InGaSb (achievable
through lattice mismatch between channel and barrier layers during MBE growth).
Fig. 7.18: Enhancement of mobility with respect to the unstrained case with biaxial
(left) uniaixal (right) strain. Enhancement with compression is always better than
tension. Upto ~4X / ~2X enhancement possible with 2% uniaxial /biaxial compression.
Enhancement with strain (especially for uniaxial) depends highly on the choice of
channel direction [104].
m/m
unst
rain
ed
BiaxialNs=1e12cm‐2, T=300K
Uniaxial
169
Fig. 7.19: μh enhancement with biaxial stain; achievable by lattice mismatch between
channel and barrier layers during MBE growth.
In summary, our comprehensive analysis of the hole mobility enhancement in ternary
III-V materials shows that Sb’s are significantly better than As’s. Further, application
of strain significantly enhances the mobility in (In/Ga/InXGa1-X)Sb’s. A hole mobility
greater than 1600cm2/Vs can be obtained by application of biaxial strain in InGaSb.
Further enhancement should be possible using uniaxial strain. Strained Sb’s are
extremely promising candidates for nanoscale III-V p-MOSFETs.
7.6 Novel Device Structure and Parasitics
7.6.1 Quantum Well (QW) Strained Heterostructure IIIV FETs
170
Strain in general results in reduction in the EG and hence enhanced off-state leakage
(IBTBT). Confinement on the other hand results in increased EG as shown in Fig. 7.5,
and hence reduces IBTBT. The heterostructure QW-FET of Fig. 7.20 proposes a unique
and novel device structure to combine strain and quantum mechanical confinement to
obtain desired transport properties with reduced off-state leakage [29-32]. In these
structures the transport can be confined to the center of the channel in a high mobility
material flanked by a high EG material. The capping layer can be chosen such that it
helps in providing a very good high-k gate dielectric interface and reduced mobility
degradation due to interface states. The mobility is further enhanced due to strain,
reduced electric field in the center of the double gate structure due to symmetry and
the channel being away from the dielectric interface. The bandgap of the center
channel can be increased due confinement by keeping it very thin. However, QW
FETs are not completely free from their tradeoffs. QW FETs are found to have
dramatic mobility degradation at QW thickness of less than ~4nm due to strong
quantum confinement effects [32, 113-114]. To take complete advantage of high
mobility III-V materials, heterostructure quantum-well FETs, which can
simultaneously achieve high drive currents and low off-state leakage should be
investigated, similar to the case of Ge [30, 32].
171
Fig. 7.20: Conventional and QW double gate FET structures.
7.6.2 Parasitic Resistance
Parasitic resistance in the S/D regions of the conventional MOSFET has been
identified as one of the primary problems of the non-scaling of drive currents in
transistor scaling. This problem can be significantly worse in III-V n-MOSFETs due
to their low value of the maximum achievable n-type dopant concentration. Replacing
the S/D regions of transistors with metals has been suggested as one of the techniques,
which might help address this problem (Fig. 7.21). Use of the metal S/D offers
additional advantages of low-temperature processing for S/D formation, elimination of
172
the parasitic bipolar action and inherent physical scalability of the gate lengths due to
the abrupt metal-semiconductor interface. However, for the ION of the metal S/D to be
better than diffused S/D the Schottky barrier to channel needs to be very small. The
free-electron wavefunction penetrates from the metal into the semiconductor, which
generates metal-induced gap-states (MIGS) and pins the Fermi level near the charge
neutrality level (ECNL) [115]. In the case of Ge we have found that the Fermi level at
metal-Ge Schottky barriers is pinned near the valence band of Ge for a variety of
metals, including, Ni, Co and Ti [116-117]. In the case of III-Vs such as InAs, the
Fermi level pins very close to conduction band [118]. This could provide a very small
barrier to the electrons in the channel in a IIIV n-MOSFETs. Due to the small barrier
height to electrons coupled with the high inversion electron mobility, Schottky S/D
IIIV n-MOS transistors are very attractive future device candidates.
Fig. 7.21: Metal Source-Drain IIIV n-MOSFET.
7.6.3 Parasitic Capacitance
173
It has been shown that scaling MOSFET below 32nm may increase parasitic gate
capacitance and it may significantly limit the performance [119-120]. Parasitic gate
capacitance can undermine the advantage in delay of III-V materials over silicon. Fig.
7.22 shows how much the delays worsen by adding parasitic gate capacitance. In this
analysis, transit time is time required for electron to pass through the channel and
defined as, Lg/vinj. Gate delay is ΔQ/Ieff. To take into account parasitic effect, we
assumed parasitic gate capacitance to be half of gate capacitance of Si and added the
parasitic gate capacitance to gate capacitance of all the materials as:
Delay = (ΔQChannel + ΔQParasitic) /Ieff
, where ΔQParasitic= 1/2 ΔQSi and Ieff =1/2×[ID(VDD,VDD/2) + ID(VDD/2,VDD)];
ID(VG,VD) [121].
Although parasitic gate capacitance reduces the gap between high mobility materials
and Si, III-V materials and Ge still outperform Si in delay.
174
Fig. 7.22: Delay in consideration of parasitic gate capacitance. Lg=15nm, Tox=0.7nm, Ts=5nm. VDD =0.7V.
7.7 Conclusion
MOSFETs utilizing high mobility III-V channel materials can take us to the sub-20
nm regime. Ternary III-V materials, such as strained InxGa(1-x)As could be suitable for
n-MOSFETs and strained InGaSb for p-MOSFETs. However, to take full advantage of
high mobility/small bandgap channel materials, novel device structures, such as
heterostructure quantum well (QW) FETs along with strain engineering will be needed,
in order to achieve high drive currents while maintaining low off-state leakage. For
III-V materials to become mainstream, important challenges, such as, good surface
passivation, low parasitic resistance/capacitance, and heterogeneous integration on Si
platform must be overcome.
175
176
Chapter 8 Conclusion
Due to several fundamental and manufacturing limits, it is becoming clear that new
materials will be needed to supplement or even replace the conventional bulk
MOSFET in future technology generations. In this thesis, we have developed models
to calculate the bandstructure, the quantum ballistic carrier transport and band-to-band
tunneling and have used these models to benchmark various high mobility materials.
In the chapter 2, an empirical pseudo-potential method is developed to calculate the
bandstructures of semiconductors in zinc-blende lattice structure. The atomic
dislocation induced by applied strain is carefully examined including the effect of
internal strain caused by shear strain based on the infinitesimal strain theory. Atomic
pseudo-potential is modeled in continuous reciprocal vector space as a function of
strain tensor and reciprocal vector. We included the spin-orbit coupling which causes
splitting in degenerate 3 valence bands. The concept of kinetic energy scaling is
adopted to take into account non-local many-body potential. The complex
bandstructure was obtained as well by modifying the Hamiltonian matrix such that
crystal-momentum k-vector is eigen-value.
In the chapter 3, new BTBT model is developed to takes into account full
bandstructure, direct and phonon assisted indirect tunneling, quantum confinement
effect, non-uniform electric field (non-local) and strain induced enhanced/suppressed
177
tunneling in semiconductors. In this model, the tail of electron wavefunction
penetrating into the bandgap is modeled and used to evaluate the interband matrix
elements between conduction band states and valence band states. The final BTBT
carrier generation rate (GBTBT) was calculated by adding up the transition rates
obtained by Fermi-Golden rule for all the possible transitions
In the chapter 4, the deformation potentials for phonon-assisted band-to-band
tunneling have been successfully calculated using rigid-ion model and empirical
pseudo-potential method. The previous theoretical calculations were all limited to
inner-band scattering (ex. conduction band to conduction band, valence band to
valence band) or limited to transitions between symmetric points in real k-space. By
introducing complex-k space in calculations of BTBT deformation potentials, we are
able to explain the discrepancies between previous theories to experimental results on
tunneling diodes. In relaxed silicon, only TA(Γ-∆) and TO(Γ-∆) phonons have
significant contributions on band-to-band tunneling and in relaxed germanium, LA(Γ-
L), TA(Γ-L), TO(Γ-L) phonons are participating tunneling process.
In the chapter 5, we have shown the limitations of previous BTBT tunneling models
for TCAD simulators. Since previous BTBT tunneling models are based on the two-
band model in relaxed bulk semiconductor system, it fails to predict BTBT current for
highly strained ultra thin film devices. Moreover the phonon-electron interactions
were not properly modeled in any of previous works. Depending on transition paths
178
we have carefully chosen phonon modes according to selection rules and deformation
potentials calculated by applying rigid-ion model on electron orbital in complex-k
space. Our model was verified using various experimental data on fabricated PIN
structures on silicon, germanium and GaAs wafers. We have shown for the first time
that biaxial compressive-strain reduces the direct tunneling component, but the
increases in-direct tunneling component in leakage current in germanium devices.
With 2.4% biaxial compressive-strain applied across germanium devices, the BTBT
leakage current reaches the minimum point. By applying our BTBT model in DGFET
simulations we found that reducing the thickness of channel layer is very effective in
suppressing the BTBT leakage current. This quantum confinement effect has been
verified by comparing the simulation results to the I-V characteristics of fabricated
thin-film strained germanium MOSFET devices.
.
In chapter 6, the quantum ballistic transport model is developed to accurately model
the carrier transport in highly scaled DGFETs. This model takes into account the
effects of non-parabolicity of conduction bands on transport mass and 1-D density of
states. Asymptotic approximation helped to simplify model. The BTBT leakage
current is also included in the ballistic transport model.
In the chapter 7, we have benchmarked strained III-V and strained SiGe to study the
scalability of DGFETs with these materials as a channel material.
179
Among the relaxed materials, Si exhibits the slowest switching and the lowest
IOFF,BTBT. Although delays for InAs and InSb are small, they suffer extremely large
IOFF,BTBT. InP exhibit fast switching time, large ION at the reduced IOFF,BTBT. Our results
show that with oxide thickness of 0.7 nm, small density of states (DOS) of these
materials does not significantly limit the on-current (ION) and high µ materials still
show higher ION than Si. However, the high µ small bandgap materials like InAs,
InSb and Ge, suffer from excessive BTBT current and poor SCE, which can limit their
scalability. Ge has highest ION due to its large DOS and small transport effective mass,
but it also suffers from large BTBT leakage. Scaling TS enhances the device
performance by eliminating the SCE effect and reducing BTBT leakage. Overall,
among all the unstrained IIIV materials In0.25Ga0.75As and InP exhibit the best
performance - high ION, low delay and low IOFF,BTBT compared to Si.
Strained InxGa1-xAs is a very promising candidate for future NFETs. Based on our
simulation results, GaAs (001), In0.25Ga0.75As (001) and In0.75Ga0.25As (111) are
selected as the best channel materials. Although GaAs (001) exhibits the lowest
IOFF,BTBT due to its large bandgap (>1.4eV), the drive current of GaAs (001) is higher
than strained Si. By applying biaxial tensile-strain, GaAs (001) can be improved to
have ION as high as In0.25Ga0.75As, with marginal increase in IOFF,BTBT. In0.25Ga0.75As
(001) has both good ION and low IOFF,BTBT because of its large bandgap (>1eV) and
large ΔEΓ-L. 4% uniaxial compressive strain can further reduce the leakage in
In0.25Ga0.75As (001), without significant reduction in ION. In0.75Ga0.25As (111) has
excellent carrier transport properties, but it suffers large IOFF,BTBT. Biaxial compressive
180
strain can reduce the leakage in In0.75Ga0.25As (111) below 0.1μA/μm. Considering the
continuation in device scaling, In0.75Ga0.25As (111) would be the better material choice
over other InxGa(1-x)As compositions. As the gate length scales down close to 10nm,
further scaling in channel thickness would be necessary to control the short channel
effects. Larger quantization effect in thinner body will further increase the bandgap of
In0.75Ga0.25As (111), leading to even smaller leakage current. By contrast, the
scalability of devices with the channel materials of GaAs (001) and In0.25Ga0.75As (111)
would be limited, since the strong quantization effect in Γ-valley will diminish ION due
to reduced ΔEΓ-L.
MOSFETs utilizing high mobility III-V channel materials can take us to the sub-20
nm regime. Ternary III-V materials, such as strained InxGa(1-x)As could be suitable for
n-MOSFETs and strained InGaSb for p-MOSFETs. However, to take full advantage of
high mobility/small bandgap channel materials, novel device structures, such as
heterostructure quantum well (QW) FETs along with strain engineering will be needed,
in order to achieve high drive currents while maintaining low off-state leakage. For
III-V materials to become mainstream, important challenges, such as, good surface
passivation, low parasitic resistance/capacitance, and heterogeneous integration on Si
platform must be overcome.
181
References [1] D. Kahng, and M. M. Atalla, “Silicon-silicon dioxide field induced surface
devices,” in IRE-AIEEE Solid-State Device Research Conference, Carmegie
Institute of Technology, Pittsburgh, PA, 1960.
[2] G. E. Moore, “Cramming more components onto integrated circuits (Reprinted
from Electronics, pg 114-117, April 19, 1965),” Proceedings of the Ieee, vol.
86, no. 1, pp. 82-85, Jan, 1998.
[3] K. Natori, “Ballistic metal-oxide-semiconductor field effect transistor,”
Journal of Applied Physics, vol. 76, no. 8, pp. 4879, 15 Oct. 1994, 1994.
[4] J. R. Chelikowsky, and M. L. Cohen, “Nonlocal pseudopotential calculations
for the electronic structure of eleven diamond and zinc-blende semiconductors,”
Physcial Review B, vol. 14, no. 2, pp. 556, 1976.
[5] M. L. Cohen, and T. K. Bergstresser, “Band Structures and Pseudopotential
Form Factors for Fourteen Semiconductors of the Diamond and Zinc-blende
Structures,” Physical Review, vol. 141, no. 2, pp. 789, January 1966, 1966.
[6] Author ed.^eds., “The fitting of pseudopotentials to experimental data and their
subsequetn application,” Solid State Physics, Advances in Research and
Applications, New York: Academic Press, 1970, p.^pp. Pages.
[7] P. Friedel, M. S. Hybertsen, and M. Schluter, “Local empirical pseudopotential
approach to the optical properties of Si/Ge superlattices,” Physcial Review B,
vol. 39, no. 11, pp. 7974, 1989.
182
[8] K. Kim, P. R. C. Kent, and A. Zunger, “Atomistic description of the electronic
structure of InxGa1-xAs alloys and InAs/GaAs superlattices,” Physcial Review
B, vol. 66, pp. 045208, 2002.
[9] L.-W. Wang, S.-H. Wei, T. Mattila et al., “Multiband coupling and electronic
structure of (InAs)n/(GaSb)n superlattices,” Physical Review B, vol. 60, no. 8,
pp. 5590, 15 August 1999, 1999.
[10] P. Harrison, Quantum Wells, Wires and Dots, Second ed.: Wiley, 2005.
[11] R. M. Martin, Electronic Structure: Basic Theory and Practical Methods,
Cambridge: Cambridge University Press, 2004.
[12] M. J. Shaw, “Microscoptic theory of scattering in imperfect strained
antimonide-based heterostructures,” Physical Review B, vol. 61, no. 8, pp.
5431, 15 February 2000, 2000.
[13] J.-B. Xia, “Theoretical analysis of electronic structures of short-period
superlattices (GaAs)m/(AlAs)n and corresponding alloys
Aln/(m+n)Gam/(m+n)As,” Physcial Review B, vol. 38, no. 12, pp. 8358, 1988.
[14] I. Morrison, M. Jaros, and K. B. Wong, “Strain-induced confinement in
Si0.75Ge0.25 (Si/Si0.5Ge0.5) (001) superlattice systems,” Phsycial Review B,
vol. 35, no. 18, pp. 9693, 1987.
[15] L.-W. Wang, J. Kim, and A. Zunger, “Electronic structures of [110]-faceted
self-assembled pyramidal InAs/GaAs quantum dots,” Physcial Review B, vol.
59, no. 8, pp. 5679, 1999.
183
[16] G. Grosso, and C. Piermarocchi, “Tight-binding model and interactions scaling
laws for silicon and germanium,” Physcial Review B, vol. 51, no. 23, pp.
16772, 1995.
[17] J. P. Walter, and M. L. Cohen, “Calulated and Measured Reflectivity of ZnTe
and ZnSe,” Physcial Review B, vol. 1, no. 6, pp. 2661, 1970.
[18] G. Weisz, “Band Structure and Fermi Surface of White Tin,” Physical Review,
vol. 149, no. 2, pp. 504, 1966.
[19] M. M. Rieger, and P. Vogl, “Electronic-band parameters in strained Si1-xGex
alloys on Si1-yGey substrates,” Physcial Review B, vol. 48, no. 19, pp. 14276,
1993.
[20] W. J. Wortman, and R. A. Evans, “Young's Modulus, Shear Modulus, and
Poisson's Ratio in Silicon and Germanium,” Journal of Applied Physics, vol.
36, no. 1, pp. 153, 1965.
[21] L. Kleinman, “Deformation Potentials in Silicon. I. Uniaxial Strain,” Physical
Review, vol. 128, no. 6, pp. 2614, 1962.
[22] O. H. Nielsen, and R. M. Martin, “Stressees in semiconductors: Ab initio
calculations on Si, Ge, and GaAs,” Physcial Review B, vol. 32, no. 6, pp. 3792,
1985.
[23] D. Kim, T. Krishnamohan, and K. C. Saraswat, "Performance evaluation of III-
V double-gate n-MOSFETs." pp. 67-8.
184
[24] D. Kim, T. Krishnamohan, and K. C. Saraswat, “Performance evaluation of
15nm gate length double-gate n-MOSFETs with high mobility channels: IIIV,
Ge and Si ” ECS Transactions, vol. 16, no. 10, pp. 47-55, 2008.
[25] D. Kim, T. Krishnamohan, and K. C. Saraswat, “Performance evaluation of
uniaxial- and biaxial-strained In(x)Ga(1-x)As NMOS DGFETs,” International
Conference on Simulation of Semiconductor Processes and Devices (SISPAD),
pp. 101-104, 2008.
[26] D.-H. Kim, J. d. Alamo, J.-H. Lee et al., “Logic suitability of 50-nm
In0.7Ga0.3AsHEMTs for Beyond-CMOS applications,” IEEE Transactions on
Electron Devices, vol. 54, no. 10, pp. 2606-13, Oct. 2007, 2007.
[27] T. Krishnamohan, C. Jungemann, and K. C. Saraswat, “A novel, very high
performance, sub-20 nm depletion-mode double-gate (DMDG),” Technical
Digest - International Electron Devices Meeting, pp. 687-90, 2003.
[28] T. Krishnamohan, D. Kim, T. V. Dinh et al., "Comparison of (001), (110) and
(111) uniaxial- And biaxial- strained-Ge and strained-Si PMOS DGFETs for
all channel orientations: Mobility enhancement, drive current, delay and off-
state leakage."
[29] T. Krishnamohan, D. Kim, C. Jungemann et al., “Strained-Si, relaxed-Ge or
strained-(Si)Ge for future nanoscale p-MOSFETs?,” Digest of Technical
Papers - 2006 Symposium on VLSI Technology, pp. 144-5, 2006.
[30] T. Krishnamohan, D. Kim, C. Nguyen et al., “High-mobility low band-to-
band-tunneling strained-germanium double-gate heterostructure FETs:
185
Simulations,” IEEE Transactions on Electron Devices, vol. 53, no. 5, pp.
1000-9, May 2006, 2006.
[31] T. Krishnamohan, Z. Krivokapic, K. Uchida et al., “Low defect ultrathin fully
strained Ge MOSFET on relaxed Si with high mobility and low Band-To-
Band-Tunneling (BTBT) ” Digest of Technical Papers - Symposium on VLSI
Technology, pp. 82-3, 2005.
[32] T. Krishnamohan, Z. Krivokapic, K. Uchida et al., “High-mobility ultrathin
strained Ge MOSFETs on bulk and SOI with low band-to-band tunneling
leakage: Experiments ” IEEE Transactions on Electron Devices, vol. 53, no. 5,
pp. 990-9, May 2006, 2006.
[33] T. Krishnamohan, and K. C. Saraswat, “High mobility Ge and III-V materials
and novel device structures for high performance nanoscale MOSFETS,”
ESSDERC 2008, pp. 38-46, 2008.
[34] A. Pethe, T. Krishnamohan, D. Kim et al., “Investigation of the performance
limits of III-V double-gate n-MOSFETs,” Technical Digest - International
Electron Devices Meeting, pp. 605-608, 2005.
[35] J. B. Krieger, “Theory of Electron Tunneling in Semiconductors with
Degenerate Band Structure,” Annals of Physics, vol. 36, no. 1, pp. 1-60, 1966.
[36] O. J. Glembocki, and F. H. Pollak, “Calculation of the Gamma-Delta-Electron-
Phonon and Hole-Phonon Scattering Matrix-Elements in Silicon,” Physical
Review Letters, vol. 48, no. 6, pp. 413-416, 1982.
186
[37] O. J. Glembocki, and F. H. Pollak, “Calculation of the Gamma-L Electron-
Phonon and Hole-Phonon Scattering Matrix-Elements in Germanium,”
Physical Review B, vol. 25, no. 12, pp. 7863-7866, 1982.
[38] O. J. Glembocki, and F. H. Pollak, “Intervalley Electron-Phonon and Hole-
Phonon Scattering Matrix-Elements in Germanium,” Physica B & C, vol. 117,
no. Mar, pp. 546-548, 1983.
[39] M. V. Fischetti, T. P. O'Regan, S. Narayanan et al., “Theoretical study of some
physical aspects of electronic transport in nMOSFETs at the 10-nm gate-
length,” Ieee Transactions on Electron Devices, vol. 54, no. 9, pp. 2116-2136,
Sep, 2007.
[40] E. O. Kane, “Zener Tunneling in Semiconductors,” Journal of Physics and
Chemistry of Solids, vol. 12, no. 2, pp. 181-188, 1959.
[41] E. O. Kane, “Theory of Tunneling,” Journal of Applied Physics, vol. 32, no. 1,
pp. 83-91, 1961.
[42] J. J. Tiemann, and Fritzsch.H, “Theory of Indirect Interband Tunneling in
Semiconductors,” Physical Review, vol. 1, no. 6A, pp. 1910-&, 1965.
[43] J. J. Tiemann, and H. Fritzsche, “Temperature Dependence of Indirect
Interband Tunneling in Germanium,” Physical Review, vol. 132, no. 6, pp.
2506-&, 1963.
[44] A. Rahman, M. S. Lundstrom, and A. W. Ghosh, “Generalized effective-mass
approach for n-type metal-oxide-semiconductor field-effect transistors on
187
arbitrarily oriented wafers,” Journal of Applied Physics, vol. 97, no. 5, pp. -,
Mar 1, 2005.
[45] L. Keldysh, “Influence of the Lattice Vibrations of a Crystal on the Production
of Electron-Hole Pairs in a Strong Electrical Field,” Soviet Physics JETP -
USSR, vol. 7, no. 4, pp. 665-668, 1958.
[46] M. Lax, and J. J. Hopfield, “Selection Rules Connecting Different Points in
Brillouin Zone,” Physical Review, vol. 1, no. 1, pp. 115-123, 1961.
[47] G. G. Macfarlane, T. P. Mclean, J. E. Quarrington et al., “Fine Structure in the
Absorption-Edge Spectrum of Ge,” Physical Review, vol. 108, no. 6, pp. 1377-
1383, 1957.
[48] X. Q. Zhou, H. M. Vandriel, and G. Mak, “Femtosecond Kinetics of
Photoexcited Carriers in Germanium,” Physical Review B, vol. 50, no. 8, pp.
5226-5230, Aug 15, 1994.
[49] M. Schluter, G. Martinez, and M. L. Cohen, “Pressure and Temperature-
Dependence of Electronic-Energy Levels in Pbse and Pbte,” Physical Review B,
vol. 12, no. 2, pp. 650-658, 1975.
[50] M. J. Lax, Symmetry principles in solid state and molecular physics, New
York: Wiley, 1974.
[51] S. Krishnamurthy, and M. Cardona, “Self-Consistent Calculation of Intervalley
Deformation Potentials in Gaas and Ge,” Journal of Applied Physics, vol. 74,
no. 3, pp. 2117-2119, Aug 1, 1993.
188
[52] G. H. Li, A. R. Goni, K. Syassen et al., “Intervalley Scattering Potentials of Ge
from Direct Exciton Absorption under Pressure,” Physical Review B, vol. 49,
no. 12, pp. 8017-8023, Mar 15, 1994.
[53] A. G. Chynoweth, D. E. Thomas, and R. A. Logan, “Phonon-Assisted
Tunneling in Silicon and Germanium Esaki Junctions,” Physical Review, vol.
125, no. 3, pp. 877-881, 1962.
[54] C. Rivas, R. Lake, G. Klimeck et al., “Full-band simulation of indirect phonon
assisted tunneling in a silicon tunnel diode with delta-doped contacts,” Applied
Physics Letters, vol. 78, no. 6, pp. 814-816, Feb 5, 2001.
[55] S. Tanaka, “Theory of the Drain Leakage Current in Silicon Mosfets,” Solid-
State Electronics, vol. 38, no. 3, pp. 683-691, Mar, 1995.
[56] N. Holonyak, I. A. Lesk, R. N. Hall et al., “Direct Observation of Phonons
during Tunneling in Narrow Junction Diodes,” Physical Review Letters, vol. 3,
no. 4, pp. 167-168, 1959.
[57] A. Schenk, “Rigorous Theory and Simplified Model of the Band-to-Band
Tunneling in Silicon,” Solid-State Electronics, vol. 36, no. 1, pp. 19-34, Jan,
1993.
[58] W. Weber, “Adiabatic Bond Charge Model for Phonons in Diamond, Si, Ge,
and Alpha-Sn,” Physical Review B, vol. 15, no. 10, pp. 4789-4803, 1977.
[59] P. M. Solomon, J. Jopling, D. J. Frank et al., “Universal tunneling behavior in
technologically relevant P/N junction diodes,” Journal of Applied Physics, vol.
95, no. 10, pp. 5800-5812, May 15, 2004.
189
[60] M. I. Nathan, “Current Voltage Characteristics of Germanium Tunnel Diodes,”
Journal of Applied Physics, vol. 33, no. 4, pp. 1460-1469, 1962.
[61] G. A. M. Hurkx, “On the Modeling of Tunnelling Currents in Reverse-Biased
P-N-Junctions,” Solid-State Electronics, vol. 32, no. 8, pp. 665-668, Aug, 1989.
[62] M. S. Tyagi, “Zener and Avalanche Breakdown in Silicon Alloyed P-N
Junctions .I. Analysis of Reverse Characteristics,” Solid-State Electronics, vol.
11, no. 1, pp. 99-115, 1968.
[63] R. A. Logan, and A. G. Chynoweth, “Effect of Degenerate Semiconductor
Band Structure on Current-Voltage Characteristics of Silicon Tunnel Diodes,”
Physical Review, vol. 131, no. 1, pp. 89-95, 1963.
[64] R. H. Haitz, “Mechanisms Contributing to Noise Pulse Rate of Avalanche
Diodes,” Journal of Applied Physics, vol. 36, no. 10, pp. 3123-3131, 1965.
[65] L. Gaul, S. Huber, J. Freyer et al., “Determination of Tunnel-Generation Rate
from Gaas Pin-Structures,” Solid-State Electronics, vol. 34, no. 7, pp. 723-726,
Jul, 1991.
[66] M. S. Tyagi, “Determination of Effective Mass and Pair Production Energy for
Electrons in Germanium from Zener Diode Characteristics,” Japanese Journal
of Applied Physics, vol. 12, no. 1, pp. 106-108, 1973.
[67] P. N. Butcher, K. F. Hulme, and J. R. Morgan, “Dependence of Peak Current
Density on Acceptor Concentration in Germanium Tunnel Diodes,” Solid-State
Electronics, vol. 5, no. Sep-Oct, pp. 358-360, 1962.
190
[68] H. P. Kleinknecht, “Indium Arsenide Tunnel Diodes,” Solid-State Electronics,
vol. 2, no. 2-3, pp. 133-140, 1961.
[69] M. S. Tyagi, “Zener and Avalanche Breakdown in Silicon Alloyed P-N
Junctions .2. Effect of Temperature on Reverse Characteristics and Criteria for
Distinguishing between 2 Breakdown Mechanisms,” Solid-State Electronics,
vol. 11, no. 1, pp. 117-&, 1968.
[70] P. Agarwal, M. J. Goossens, V. Zieren et al., “Impact ionization in thin silicon
diodes,” Ieee Electron Device Letters, vol. 25, no. 12, pp. 807-809, Dec, 2004.
[71] T. Tokuyama, “Zener Breakdown in Alloyed Germanium P+-N Junctions,”
Solid-State Electronics, vol. 5, no. May-J, pp. 161-&, 1962.
[72] D. Kim, “Theoretical Performance Evaluations of NMOS Double Gate FETs
with High Mobility Materials : Strained III-V, Ge and Si,” Ph. D. DIssertation,
Electrical Engineering, Stanford University, Stanford, 2009.
[73] D. Kim, T. Krishnamohan, Y. Nishi et al., "Band to band tunneling limited off
state current in ultra-thin body double gate FETs with high mobility materials:
III-V, Ge and strained Si/Ge." pp. 389-92.
[74] D. Kim, T. Krishnamohan, L. Smith et al., “Band to band tunneling study in
high mobility materials: III-V, Si, Ge and strained SiGe,” 2007 65th DRC
Device Research Conference, pp. 57-8, 2007.
[75] "International Technology Roadmap for Semiconduction," http://public.itrs.net.
191
[76] S. E. Laux, “A simulation study of the switching times of 22-and 17-nm gate-
length SOI nFETs on high mobility substrates and Si,” Ieee Transactions on
Electron Devices, vol. 54, no. 9, pp. 2304-2320, Sep, 2007.
[77] K. Natori, “Ballistic metal-oxide-semiconductor field effect transistor,”
Journal of Applied Physics, vol. 76, no. 8, pp. 4897, 1994.
[78] S. Scaldaferri, G. Curatola, and G. Iannaccone, “Direct solution of the
boltzmann transport equation and poisson-schrodinger equation for nanoscale
MOSFETs,” Ieee Transactions on Electron Devices, vol. 54, no. 11, pp. 2901-
2909, Nov, 2007.
[79] L. Worschech, A. Schliemann, H. Hsin et al., “Ballistic transport in nanoscale
field effect transistors revealed by four-terminal DC characterization,”
Superlattices and Microstructures, vol. 34, no. 3-6, pp. 271-275, Sep-Dec,
2003.
[80] G. Bastard, “Theoretical investigations of superlattice band structure in the
envelope-function approximation,” Physcial Review B, vol. 25, no. 12, pp.
7584, 1982.
[81] D. F. Nelson, R. C. Miller, and D. A. Kleinman, “Band nonparabolicity effects
in semiconductor quantum wells,” Physcial Review B, vol. 35, no. 14, pp. 7770,
1987.
[82] A. Godoy, Z. Y. a. U. Ravaioli, and F. Gamiz, “Effects of nonparabolic bands
in quantum wires,” Journal of Applied Physics, vol. 98, pp. 013702, 2005.
192
[83] M. Buttiker, “Larmor precession and the traversal time for tunneling,” Physical
Review B, vol. 27, no. 10, pp. 6178, 1983.
[84] M. Buttiker, and R. Landauer, “Traversal Time for Tunneling,” IBM Journal of
Reserach and Development, vol. 30, no. 5, pp. 451, September, 1986.
[85] M. Buttiker, and R. Landauer, “Traversal Time for Tunneling,” Physical
Review Letters, vol. 49, no. 23, pp. 1739, 1982.
[86] X. Aymerich-Humet, F. Serra-Mestres, and J. Millan, “An Analytical
Approximation for the Fermi-Dirac Integral F3/2(eta),” Solid State Electronics,
vol. 24, no. 10, pp. 981, 1981.
[87] R. Landuer, “Electrical Resistance of Disordered One-dimensional Lattices,”
Philosophical Magazine, vol. 21, no. 172, pp. 863, 1970.
[88] R. Chau, “Benchmarking nanotechnology for high performance and low-power
logic transistor applications,” IEEE Transactions on Nanotechnology, vol. 4,
no. 2, pp. 153-8, Mar. 2005, 2005.
[89] S. Datta, G. Dewey, M. Doczy et al., “High mobility Si/SiGe strained channel
MOS transistors with HfO2/TiN gate stack,” Technical Digest - International
Electron Devices Meeting pp. 653-6, 2003.
[90] T. Tezuka, N. Sugiyama, T. Mizuno et al., “Ultrathin body SiGe-on-insulator
pMOSFETs with high-mobility SiGe surface channels,” IEEE Transactions on
Electron Devices, vol. 50, no. 5, pp. 1328-33, 2003.
193
[91] H. Shang, K.-L. Lee, P. Kozlowski et al., “Epitaxial silicon and germanium on
buried insulator heterostructures and devices,” Applied Physics Letters, vol. 83,
no. 26, pp. 5443-5, 29 Dec 2003, 2003.
[92] H. Shang, J. O. Cho, X. Wang et al., “Channel design and mobility
enhancement in strained germanium buried channel MOSFETs ” Digest of
Technical Papers - Symposium on VLSI Technology, pp. 204-205, 2004.
[93] A. Ritenour, S. Yu, M. L. Lee et al., “Epitaxial strained germanium p-
MOSFETs with HfO<sub>2</sub> gate dielectric and TaN gate electrode ”
Technical Digest - International Electron Devices Meeting, pp. 433-6, 2003.
[94] M. Lee, E. Fitzgerald, M. Bulsara et al., “Strained Si, SiGe and Ge channels
for high mobility metal oxide semiconductor field effect transistors,” Journal
of Applied Physics, vol. 97, no. 1, pp. 11101-1-27, 2005.
[95] S. Datta, T. Ashley, J. Brask et al., “85nm gate length enhancement and
depletion mode InSb quantum well transistors for ultra high speed and very
low power digital logic applications ” Technical Digest - International
Electron Devices Meeting, pp. 763-6, Dec. 2005, 2005.
[96] E. O. Kane, “Zener Tunneling in Semiconductors,” Journal of Physics and
Chemistry of Solids, vol. 12, no. 2, pp. 181-8, 1959.
[97] A. Rahman, M. S. Lundstrom, and A. W. Ghosh, “Generalized effective-mass
approach for n -type metal-oxide-semiconductor field-effect transistors on
arbitrarily oriented wafers,” Journal of Applied Physics, vol. 97, no. 5, pp. 1-12,
2005.
194
[98] J. Wang, and M. Lundstrom, “Ballistic Transport in High Electron Mobility
Transistors,” IEEE Transactions on Electron Devices, vol. 5, no. 7, pp. 1604-9,
2003.
[99] S. E. Laux, “A simulation study of the switching times of 22- and 17-nm gate-
length SOI nFETs on high mobility substrates and Si,” IEEE Transactions on
Electron Devices, vol. 54, no. 9, pp. 2304-20, 2007.
[100] M. V. Fischetti, and S. E. LAUX, “Monte Carlo simulation of transport in
technologically significant semiconductors of the diamond and zinc-blende
structures--II: Submicrometer MOSFET's ” IEEE Transactions on Electron
Devices, vol. 38, no. 3, pp. 650-60, Mar. 1991, 1991.
[101] A. Asenov, K. Kalna, I. Thayne et al., “Simulation of implant free III-V
MOSFETs for high performance low power Nano-CMOS applications ”
Microelectronic Engineering, vol. 84, no. 9-10, pp. 2398-403, Sept. - Oct.
2007, 2007.
[102] M. D. Michielis, D. Esseni, and F. Driussi, “Analytical models for the insight
into the use of alternative channel materials in ballistic nano-MOSFETs,”
IEEE Transactions on Electron Devices, vol. 54, no. 1, pp. 115-23, 2007.
[103] M. V. Fischetti, T. P. O'Regan, S. Narayanan et al., “Theoretical study of some
physical aspects of electronic transport in nMOSFETs at the 10-nm gate-
length,” IEEE Transactions on Electron Devices, vol. 54, no. 9, pp. 2116-36,
2007.
195
[104] A. Nainani, D. Kim, T. Krishnamohan et al., "Hole Mobility and its
Enhancement with Strain for Technologically Relevant III-V Semiconductors."
[105] P. P. Ruden, M. Shur, D. K. Arch et al., “Quantum-well p-channel
AlGaAs/InGaAs/GaAs heterostructure insulated-gate field-effect transistors,”
IEEE Transactions on Electron Devices, vol. 36, no. 11, pp. 2371-9, 1989.
[106] A. Nainani, S. Raghunathan, D. Witte et al., "Engineering of Strained III-V
Heterostructures for High Hole Mobility."
[107] G. C. Osbourn, “Electron and hole effective masses for two-dimensional
transport in strained-layer superlattices,” Superlattices and Microstructures,
vol. 1, no. 3, pp. 223-6, 1985.
[108] M. Jaffe, J. E. Oh, J. Pamulapati et al., “In-plane hole effective masses in
InxGa1-xAs/Al0.15Ga0.85As modulation-doped heterostructures ” Applied
Physics Letters, vol. 54, no. 23, pp. 2345-6, 1989.
[109] M. Radosavljevic, T. Ashley, A. Andreev et al., “High-performance 40nm gate
length InSb p-channel compressively strained quantum well field effect
transistors for low-power (VCC=0.5V) logic applications,” Technical Digest -
2008 IEEE International Electron Devices (IEDM), pp. 727-30, 2008.
[110] B. R. Bennett, M. G. Ancona, J. B. Boos et al., “Strained GaSb/AlAsSb
quantum wells for p-channel field-effect transistors,” Journal of Crystal
Growth, vol. 311, no. 1, pp. 47-53, 15 Dec 2008, 2008.
196
[111] B. R. Bennett, M. G. Ancona, J. B. Boos et al., “Mobility enhancement in
strained p-InGaSb quantum wells,” Applied Physics Letters, vol. 91, no. 4, pp.
042104, 2007.
[112] G. Bastard, Wave Mechanics applied to semiconductor heterostructures, New
York: Halsted Press, Wiley, 1988.
[113] F. A. T. M, and A. V. K., “Quantum engineering of nanoelectronic devices: the
role of quantum confinement on mobility degradation,” Microelectronics
Journal, vol. 32, no. 8, pp. 679-86, Aug. 2001, 2001.
[114] K. Uchida, and S.-i. Takagi, “Carrier scattering induced by thickness
fluctuation of silicon-on-insulator film in ultrathin-body metal–oxide–
semiconductor field-effect transistors,” Applied Physics Letters, vol. 82, no. 17,
pp. 2916, 2003.
[115] J. Tersoff, “Schottky Barriers and the Continuum of Gap States,” Physical
Review Letters, vol. 52, no. 6, pp. 465-8, 1984.
[116] A. Pethe, and K. C. Saraswat, “High-mobility, low parasitic resistance Si/Ge/Si
heterostructure channel Schottky source/drain PMOSFETs,” 65th Device
Research Conference, pp. 55-6, 2007.
[117] A. Dimoulas, P. Tsipas, A. Sotiropoulos et al., “Fermi-level pinning and
charge neutrality level in germanium,” Applied Physics Letters, vol. 89, no. 25,
pp. 252110-1-3, 18 Dec. 2006, 2006.
197
[118] H. H. Wieder, “Surface and interface barriers of InxGa1-xAs binary and
ternary alloys,” Journal of Vacuum Science and Technology B, vol. 21, no. 4,
pp. 1915-9, 2003.
[119] A. Khakifirooz, and D. A. Antoniadis, “MOSFET performance scaling. Part I.
Historical trends,” IEEE Transactions on Electron Devices, vol. 55, no. 6, pp.
1391-400, 2008.
[120] A. Khakifirooz, and D. A. Antoniadis, “MOSFET performance scaling-Part II:
future directions ” IEEE Transactions on Electron Devices, vol. 55, no. 6, pp.
1401-8, 2008.
[121] D. A. Antoniadis, I. Aberg, C. Ni Chleirigh et al., “Continuous MOSFET
performance increase with device scaling: The role of strain and channel
material innovations,” IBM Journal of Research and Development, vol. 50, no.
4-5, pp. 363-76, 2006.